/src/openssl/crypto/bn/bn_exp.c
Line | Count | Source (jump to first uncovered line) |
1 | | /* crypto/bn/bn_exp.c */ |
2 | | /* Copyright (C) 1995-1998 Eric Young (eay@cryptsoft.com) |
3 | | * All rights reserved. |
4 | | * |
5 | | * This package is an SSL implementation written |
6 | | * by Eric Young (eay@cryptsoft.com). |
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 (tjh@cryptsoft.com). |
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 (eay@cryptsoft.com)" |
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 (tjh@cryptsoft.com)" |
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 | | * openssl-core@openssl.org. |
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 | | * (eay@cryptsoft.com). This product includes software written by Tim |
108 | | * Hudson (tjh@cryptsoft.com). |
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 | 0 | { |
145 | 0 | int i, bits, ret = 0; |
146 | 0 | BIGNUM *v, *rr; |
147 | |
|
148 | 0 | if (BN_get_flags(p, BN_FLG_CONSTTIME) != 0 |
149 | 0 | || BN_get_flags(a, BN_FLG_CONSTTIME) != 0) { |
150 | | /* BN_FLG_CONSTTIME only supported by BN_mod_exp_mont() */ |
151 | 0 | BNerr(BN_F_BN_EXP, ERR_R_SHOULD_NOT_HAVE_BEEN_CALLED); |
152 | 0 | return -1; |
153 | 0 | } |
154 | | |
155 | 0 | BN_CTX_start(ctx); |
156 | 0 | if ((r == a) || (r == p)) |
157 | 0 | rr = BN_CTX_get(ctx); |
158 | 0 | else |
159 | 0 | rr = r; |
160 | 0 | v = BN_CTX_get(ctx); |
161 | 0 | if (rr == NULL || v == NULL) |
162 | 0 | goto err; |
163 | | |
164 | 0 | if (BN_copy(v, a) == NULL) |
165 | 0 | goto err; |
166 | 0 | bits = BN_num_bits(p); |
167 | |
|
168 | 0 | if (BN_is_odd(p)) { |
169 | 0 | if (BN_copy(rr, a) == NULL) |
170 | 0 | goto err; |
171 | 0 | } else { |
172 | 0 | if (!BN_one(rr)) |
173 | 0 | goto err; |
174 | 0 | } |
175 | | |
176 | 0 | for (i = 1; i < bits; i++) { |
177 | 0 | if (!BN_sqr(v, v, ctx)) |
178 | 0 | goto err; |
179 | 0 | if (BN_is_bit_set(p, i)) { |
180 | 0 | if (!BN_mul(rr, rr, v, ctx)) |
181 | 0 | goto err; |
182 | 0 | } |
183 | 0 | } |
184 | 0 | if (r != rr && BN_copy(r, rr) == NULL) |
185 | 0 | goto err; |
186 | | |
187 | 0 | ret = 1; |
188 | 0 | err: |
189 | 0 | BN_CTX_end(ctx); |
190 | 0 | bn_check_top(r); |
191 | 0 | return (ret); |
192 | 0 | } |
193 | | |
194 | | int BN_mod_exp(BIGNUM *r, const BIGNUM *a, const BIGNUM *p, const BIGNUM *m, |
195 | | BN_CTX *ctx) |
196 | 0 | { |
197 | 0 | int ret; |
198 | |
|
199 | 0 | bn_check_top(a); |
200 | 0 | bn_check_top(p); |
201 | 0 | bn_check_top(m); |
202 | | |
203 | | /*- |
204 | | * For even modulus m = 2^k*m_odd, it might make sense to compute |
205 | | * a^p mod m_odd and a^p mod 2^k separately (with Montgomery |
206 | | * exponentiation for the odd part), using appropriate exponent |
207 | | * reductions, and combine the results using the CRT. |
208 | | * |
209 | | * For now, we use Montgomery only if the modulus is odd; otherwise, |
210 | | * exponentiation using the reciprocal-based quick remaindering |
211 | | * algorithm is used. |
212 | | * |
213 | | * (Timing obtained with expspeed.c [computations a^p mod m |
214 | | * where a, p, m are of the same length: 256, 512, 1024, 2048, |
215 | | * 4096, 8192 bits], compared to the running time of the |
216 | | * standard algorithm: |
217 | | * |
218 | | * BN_mod_exp_mont 33 .. 40 % [AMD K6-2, Linux, debug configuration] |
219 | | * 55 .. 77 % [UltraSparc processor, but |
220 | | * debug-solaris-sparcv8-gcc conf.] |
221 | | * |
222 | | * BN_mod_exp_recp 50 .. 70 % [AMD K6-2, Linux, debug configuration] |
223 | | * 62 .. 118 % [UltraSparc, debug-solaris-sparcv8-gcc] |
224 | | * |
225 | | * On the Sparc, BN_mod_exp_recp was faster than BN_mod_exp_mont |
226 | | * at 2048 and more bits, but at 512 and 1024 bits, it was |
227 | | * slower even than the standard algorithm! |
228 | | * |
229 | | * "Real" timings [linux-elf, solaris-sparcv9-gcc configurations] |
230 | | * should be obtained when the new Montgomery reduction code |
231 | | * has been integrated into OpenSSL.) |
232 | | */ |
233 | |
|
234 | 0 | #define MONT_MUL_MOD |
235 | 0 | #define MONT_EXP_WORD |
236 | 0 | #define RECP_MUL_MOD |
237 | |
|
238 | 0 | #ifdef MONT_MUL_MOD |
239 | | /* |
240 | | * I have finally been able to take out this pre-condition of the top bit |
241 | | * being set. It was caused by an error in BN_div with negatives. There |
242 | | * was also another problem when for a^b%m a >= m. eay 07-May-97 |
243 | | */ |
244 | | /* if ((m->d[m->top-1]&BN_TBIT) && BN_is_odd(m)) */ |
245 | |
|
246 | 0 | if (BN_is_odd(m)) { |
247 | 0 | # ifdef MONT_EXP_WORD |
248 | 0 | if (a->top == 1 && !a->neg |
249 | 0 | && (BN_get_flags(p, BN_FLG_CONSTTIME) == 0) |
250 | 0 | && (BN_get_flags(a, BN_FLG_CONSTTIME) == 0) |
251 | 0 | && (BN_get_flags(m, BN_FLG_CONSTTIME) == 0)) { |
252 | 0 | BN_ULONG A = a->d[0]; |
253 | 0 | ret = BN_mod_exp_mont_word(r, A, p, m, ctx, NULL); |
254 | 0 | } else |
255 | 0 | # endif |
256 | 0 | ret = BN_mod_exp_mont(r, a, p, m, ctx, NULL); |
257 | 0 | } else |
258 | 0 | #endif |
259 | 0 | #ifdef RECP_MUL_MOD |
260 | 0 | { |
261 | 0 | ret = BN_mod_exp_recp(r, a, p, m, ctx); |
262 | 0 | } |
263 | | #else |
264 | | { |
265 | | ret = BN_mod_exp_simple(r, a, p, m, ctx); |
266 | | } |
267 | | #endif |
268 | |
|
269 | 0 | bn_check_top(r); |
270 | 0 | return (ret); |
271 | 0 | } |
272 | | |
273 | | int BN_mod_exp_recp(BIGNUM *r, const BIGNUM *a, const BIGNUM *p, |
274 | | const BIGNUM *m, BN_CTX *ctx) |
275 | 0 | { |
276 | 0 | int i, j, bits, ret = 0, wstart, wend, window, wvalue; |
277 | 0 | int start = 1; |
278 | 0 | BIGNUM *aa; |
279 | | /* Table of variables obtained from 'ctx' */ |
280 | 0 | BIGNUM *val[TABLE_SIZE]; |
281 | 0 | BN_RECP_CTX recp; |
282 | |
|
283 | 0 | if (BN_get_flags(p, BN_FLG_CONSTTIME) != 0 |
284 | 0 | || BN_get_flags(a, BN_FLG_CONSTTIME) != 0 |
285 | 0 | || BN_get_flags(m, BN_FLG_CONSTTIME) != 0) { |
286 | | /* BN_FLG_CONSTTIME only supported by BN_mod_exp_mont() */ |
287 | 0 | BNerr(BN_F_BN_MOD_EXP_RECP, ERR_R_SHOULD_NOT_HAVE_BEEN_CALLED); |
288 | 0 | return -1; |
289 | 0 | } |
290 | | |
291 | 0 | bits = BN_num_bits(p); |
292 | 0 | if (bits == 0) { |
293 | | /* x**0 mod 1 is still zero. */ |
294 | 0 | if (BN_is_one(m)) { |
295 | 0 | ret = 1; |
296 | 0 | BN_zero(r); |
297 | 0 | } else { |
298 | 0 | ret = BN_one(r); |
299 | 0 | } |
300 | 0 | return ret; |
301 | 0 | } |
302 | | |
303 | 0 | BN_CTX_start(ctx); |
304 | 0 | aa = BN_CTX_get(ctx); |
305 | 0 | val[0] = BN_CTX_get(ctx); |
306 | 0 | if (!aa || !val[0]) |
307 | 0 | goto err; |
308 | | |
309 | 0 | BN_RECP_CTX_init(&recp); |
310 | 0 | if (m->neg) { |
311 | | /* ignore sign of 'm' */ |
312 | 0 | if (!BN_copy(aa, m)) |
313 | 0 | goto err; |
314 | 0 | aa->neg = 0; |
315 | 0 | if (BN_RECP_CTX_set(&recp, aa, ctx) <= 0) |
316 | 0 | goto err; |
317 | 0 | } else { |
318 | 0 | if (BN_RECP_CTX_set(&recp, m, ctx) <= 0) |
319 | 0 | goto err; |
320 | 0 | } |
321 | | |
322 | 0 | if (!BN_nnmod(val[0], a, m, ctx)) |
323 | 0 | goto err; /* 1 */ |
324 | 0 | if (BN_is_zero(val[0])) { |
325 | 0 | BN_zero(r); |
326 | 0 | ret = 1; |
327 | 0 | goto err; |
328 | 0 | } |
329 | | |
330 | 0 | window = BN_window_bits_for_exponent_size(bits); |
331 | 0 | if (window > 1) { |
332 | 0 | if (!BN_mod_mul_reciprocal(aa, val[0], val[0], &recp, ctx)) |
333 | 0 | goto err; /* 2 */ |
334 | 0 | j = 1 << (window - 1); |
335 | 0 | for (i = 1; i < j; i++) { |
336 | 0 | if (((val[i] = BN_CTX_get(ctx)) == NULL) || |
337 | 0 | !BN_mod_mul_reciprocal(val[i], val[i - 1], aa, &recp, ctx)) |
338 | 0 | goto err; |
339 | 0 | } |
340 | 0 | } |
341 | | |
342 | 0 | start = 1; /* This is used to avoid multiplication etc |
343 | | * when there is only the value '1' in the |
344 | | * buffer. */ |
345 | 0 | wvalue = 0; /* The 'value' of the window */ |
346 | 0 | wstart = bits - 1; /* The top bit of the window */ |
347 | 0 | wend = 0; /* The bottom bit of the window */ |
348 | |
|
349 | 0 | if (!BN_one(r)) |
350 | 0 | goto err; |
351 | | |
352 | 0 | for (;;) { |
353 | 0 | if (BN_is_bit_set(p, wstart) == 0) { |
354 | 0 | if (!start) |
355 | 0 | if (!BN_mod_mul_reciprocal(r, r, r, &recp, ctx)) |
356 | 0 | goto err; |
357 | 0 | if (wstart == 0) |
358 | 0 | break; |
359 | 0 | wstart--; |
360 | 0 | continue; |
361 | 0 | } |
362 | | /* |
363 | | * We now have wstart on a 'set' bit, we now need to work out how bit |
364 | | * a window to do. To do this we need to scan forward until the last |
365 | | * set bit before the end of the window |
366 | | */ |
367 | 0 | j = wstart; |
368 | 0 | wvalue = 1; |
369 | 0 | wend = 0; |
370 | 0 | for (i = 1; i < window; i++) { |
371 | 0 | if (wstart - i < 0) |
372 | 0 | break; |
373 | 0 | if (BN_is_bit_set(p, wstart - i)) { |
374 | 0 | wvalue <<= (i - wend); |
375 | 0 | wvalue |= 1; |
376 | 0 | wend = i; |
377 | 0 | } |
378 | 0 | } |
379 | | |
380 | | /* wend is the size of the current window */ |
381 | 0 | j = wend + 1; |
382 | | /* add the 'bytes above' */ |
383 | 0 | if (!start) |
384 | 0 | for (i = 0; i < j; i++) { |
385 | 0 | if (!BN_mod_mul_reciprocal(r, r, r, &recp, ctx)) |
386 | 0 | goto err; |
387 | 0 | } |
388 | | |
389 | | /* wvalue will be an odd number < 2^window */ |
390 | 0 | if (!BN_mod_mul_reciprocal(r, r, val[wvalue >> 1], &recp, ctx)) |
391 | 0 | goto err; |
392 | | |
393 | | /* move the 'window' down further */ |
394 | 0 | wstart -= wend + 1; |
395 | 0 | wvalue = 0; |
396 | 0 | start = 0; |
397 | 0 | if (wstart < 0) |
398 | 0 | break; |
399 | 0 | } |
400 | 0 | ret = 1; |
401 | 0 | err: |
402 | 0 | BN_CTX_end(ctx); |
403 | 0 | BN_RECP_CTX_free(&recp); |
404 | 0 | bn_check_top(r); |
405 | 0 | return (ret); |
406 | 0 | } |
407 | | |
408 | | int BN_mod_exp_mont(BIGNUM *rr, const BIGNUM *a, const BIGNUM *p, |
409 | | const BIGNUM *m, BN_CTX *ctx, BN_MONT_CTX *in_mont) |
410 | 0 | { |
411 | 0 | int i, j, bits, ret = 0, wstart, wend, window, wvalue; |
412 | 0 | int start = 1; |
413 | 0 | BIGNUM *d, *r; |
414 | 0 | const BIGNUM *aa; |
415 | | /* Table of variables obtained from 'ctx' */ |
416 | 0 | BIGNUM *val[TABLE_SIZE]; |
417 | 0 | BN_MONT_CTX *mont = NULL; |
418 | |
|
419 | 0 | if (BN_get_flags(p, BN_FLG_CONSTTIME) != 0 |
420 | 0 | || BN_get_flags(a, BN_FLG_CONSTTIME) != 0 |
421 | 0 | || BN_get_flags(m, BN_FLG_CONSTTIME) != 0) { |
422 | 0 | return BN_mod_exp_mont_consttime(rr, a, p, m, ctx, in_mont); |
423 | 0 | } |
424 | | |
425 | 0 | bn_check_top(a); |
426 | 0 | bn_check_top(p); |
427 | 0 | bn_check_top(m); |
428 | |
|
429 | 0 | if (!BN_is_odd(m)) { |
430 | 0 | BNerr(BN_F_BN_MOD_EXP_MONT, BN_R_CALLED_WITH_EVEN_MODULUS); |
431 | 0 | return (0); |
432 | 0 | } |
433 | 0 | bits = BN_num_bits(p); |
434 | 0 | if (bits == 0) { |
435 | | /* x**0 mod 1 is still zero. */ |
436 | 0 | if (BN_is_one(m)) { |
437 | 0 | ret = 1; |
438 | 0 | BN_zero(rr); |
439 | 0 | } else { |
440 | 0 | ret = BN_one(rr); |
441 | 0 | } |
442 | 0 | return ret; |
443 | 0 | } |
444 | | |
445 | 0 | BN_CTX_start(ctx); |
446 | 0 | d = BN_CTX_get(ctx); |
447 | 0 | r = BN_CTX_get(ctx); |
448 | 0 | val[0] = BN_CTX_get(ctx); |
449 | 0 | if (!d || !r || !val[0]) |
450 | 0 | goto err; |
451 | | |
452 | | /* |
453 | | * If this is not done, things will break in the montgomery part |
454 | | */ |
455 | | |
456 | 0 | if (in_mont != NULL) |
457 | 0 | mont = in_mont; |
458 | 0 | else { |
459 | 0 | if ((mont = BN_MONT_CTX_new()) == NULL) |
460 | 0 | goto err; |
461 | 0 | if (!BN_MONT_CTX_set(mont, m, ctx)) |
462 | 0 | goto err; |
463 | 0 | } |
464 | | |
465 | 0 | if (a->neg || BN_ucmp(a, m) >= 0) { |
466 | 0 | if (!BN_nnmod(val[0], a, m, ctx)) |
467 | 0 | goto err; |
468 | 0 | aa = val[0]; |
469 | 0 | } else |
470 | 0 | aa = a; |
471 | 0 | if (BN_is_zero(aa)) { |
472 | 0 | BN_zero(rr); |
473 | 0 | ret = 1; |
474 | 0 | goto err; |
475 | 0 | } |
476 | 0 | if (!BN_to_montgomery(val[0], aa, mont, ctx)) |
477 | 0 | goto err; /* 1 */ |
478 | | |
479 | 0 | window = BN_window_bits_for_exponent_size(bits); |
480 | 0 | if (window > 1) { |
481 | 0 | if (!BN_mod_mul_montgomery(d, val[0], val[0], mont, ctx)) |
482 | 0 | goto err; /* 2 */ |
483 | 0 | j = 1 << (window - 1); |
484 | 0 | for (i = 1; i < j; i++) { |
485 | 0 | if (((val[i] = BN_CTX_get(ctx)) == NULL) || |
486 | 0 | !BN_mod_mul_montgomery(val[i], val[i - 1], d, mont, ctx)) |
487 | 0 | goto err; |
488 | 0 | } |
489 | 0 | } |
490 | | |
491 | 0 | start = 1; /* This is used to avoid multiplication etc |
492 | | * when there is only the value '1' in the |
493 | | * buffer. */ |
494 | 0 | wvalue = 0; /* The 'value' of the window */ |
495 | 0 | wstart = bits - 1; /* The top bit of the window */ |
496 | 0 | wend = 0; /* The bottom bit of the window */ |
497 | |
|
498 | 0 | #if 1 /* by Shay Gueron's suggestion */ |
499 | 0 | j = m->top; /* borrow j */ |
500 | 0 | if (m->d[j - 1] & (((BN_ULONG)1) << (BN_BITS2 - 1))) { |
501 | 0 | if (bn_wexpand(r, j) == NULL) |
502 | 0 | goto err; |
503 | | /* 2^(top*BN_BITS2) - m */ |
504 | 0 | r->d[0] = (0 - m->d[0]) & BN_MASK2; |
505 | 0 | for (i = 1; i < j; i++) |
506 | 0 | r->d[i] = (~m->d[i]) & BN_MASK2; |
507 | 0 | r->top = j; |
508 | | /* |
509 | | * Upper words will be zero if the corresponding words of 'm' were |
510 | | * 0xfff[...], so decrement r->top accordingly. |
511 | | */ |
512 | 0 | bn_correct_top(r); |
513 | 0 | } else |
514 | 0 | #endif |
515 | 0 | if (!BN_to_montgomery(r, BN_value_one(), mont, ctx)) |
516 | 0 | goto err; |
517 | 0 | for (;;) { |
518 | 0 | if (BN_is_bit_set(p, wstart) == 0) { |
519 | 0 | if (!start) { |
520 | 0 | if (!BN_mod_mul_montgomery(r, r, r, mont, ctx)) |
521 | 0 | goto err; |
522 | 0 | } |
523 | 0 | if (wstart == 0) |
524 | 0 | break; |
525 | 0 | wstart--; |
526 | 0 | continue; |
527 | 0 | } |
528 | | /* |
529 | | * We now have wstart on a 'set' bit, we now need to work out how bit |
530 | | * a window to do. To do this we need to scan forward until the last |
531 | | * set bit before the end of the window |
532 | | */ |
533 | 0 | j = wstart; |
534 | 0 | wvalue = 1; |
535 | 0 | wend = 0; |
536 | 0 | for (i = 1; i < window; i++) { |
537 | 0 | if (wstart - i < 0) |
538 | 0 | break; |
539 | 0 | if (BN_is_bit_set(p, wstart - i)) { |
540 | 0 | wvalue <<= (i - wend); |
541 | 0 | wvalue |= 1; |
542 | 0 | wend = i; |
543 | 0 | } |
544 | 0 | } |
545 | | |
546 | | /* wend is the size of the current window */ |
547 | 0 | j = wend + 1; |
548 | | /* add the 'bytes above' */ |
549 | 0 | if (!start) |
550 | 0 | for (i = 0; i < j; i++) { |
551 | 0 | if (!BN_mod_mul_montgomery(r, r, r, mont, ctx)) |
552 | 0 | goto err; |
553 | 0 | } |
554 | | |
555 | | /* wvalue will be an odd number < 2^window */ |
556 | 0 | if (!BN_mod_mul_montgomery(r, r, val[wvalue >> 1], mont, ctx)) |
557 | 0 | goto err; |
558 | | |
559 | | /* move the 'window' down further */ |
560 | 0 | wstart -= wend + 1; |
561 | 0 | wvalue = 0; |
562 | 0 | start = 0; |
563 | 0 | if (wstart < 0) |
564 | 0 | break; |
565 | 0 | } |
566 | | #if defined(SPARC_T4_MONT) |
567 | | if (OPENSSL_sparcv9cap_P[0] & (SPARCV9_VIS3 | SPARCV9_PREFER_FPU)) { |
568 | | j = mont->N.top; /* borrow j */ |
569 | | val[0]->d[0] = 1; /* borrow val[0] */ |
570 | | for (i = 1; i < j; i++) |
571 | | val[0]->d[i] = 0; |
572 | | val[0]->top = j; |
573 | | if (!BN_mod_mul_montgomery(rr, r, val[0], mont, ctx)) |
574 | | goto err; |
575 | | } else |
576 | | #endif |
577 | 0 | if (!BN_from_montgomery(rr, r, mont, ctx)) |
578 | 0 | goto err; |
579 | 0 | ret = 1; |
580 | 0 | err: |
581 | 0 | if ((in_mont == NULL) && (mont != NULL)) |
582 | 0 | BN_MONT_CTX_free(mont); |
583 | 0 | BN_CTX_end(ctx); |
584 | 0 | bn_check_top(rr); |
585 | 0 | return (ret); |
586 | 0 | } |
587 | | |
588 | | #if defined(SPARC_T4_MONT) |
589 | | static BN_ULONG bn_get_bits(const BIGNUM *a, int bitpos) |
590 | | { |
591 | | BN_ULONG ret = 0; |
592 | | int wordpos; |
593 | | |
594 | | wordpos = bitpos / BN_BITS2; |
595 | | bitpos %= BN_BITS2; |
596 | | if (wordpos >= 0 && wordpos < a->top) { |
597 | | ret = a->d[wordpos] & BN_MASK2; |
598 | | if (bitpos) { |
599 | | ret >>= bitpos; |
600 | | if (++wordpos < a->top) |
601 | | ret |= a->d[wordpos] << (BN_BITS2 - bitpos); |
602 | | } |
603 | | } |
604 | | |
605 | | return ret & BN_MASK2; |
606 | | } |
607 | | #endif |
608 | | |
609 | | /* |
610 | | * BN_mod_exp_mont_consttime() stores the precomputed powers in a specific |
611 | | * layout so that accessing any of these table values shows the same access |
612 | | * pattern as far as cache lines are concerned. The following functions are |
613 | | * used to transfer a BIGNUM from/to that table. |
614 | | */ |
615 | | |
616 | | static int MOD_EXP_CTIME_COPY_TO_PREBUF(const BIGNUM *b, int top, |
617 | | unsigned char *buf, int idx, |
618 | | int window) |
619 | 0 | { |
620 | 0 | int i, j; |
621 | 0 | int width = 1 << window; |
622 | 0 | BN_ULONG *table = (BN_ULONG *)buf; |
623 | |
|
624 | 0 | if (top > b->top) |
625 | 0 | top = b->top; /* this works because 'buf' is explicitly |
626 | | * zeroed */ |
627 | 0 | for (i = 0, j = idx; i < top; i++, j += width) { |
628 | 0 | table[j] = b->d[i]; |
629 | 0 | } |
630 | |
|
631 | 0 | return 1; |
632 | 0 | } |
633 | | |
634 | | static int MOD_EXP_CTIME_COPY_FROM_PREBUF(BIGNUM *b, int top, |
635 | | unsigned char *buf, int idx, |
636 | | int window) |
637 | 0 | { |
638 | 0 | int i, j; |
639 | 0 | int width = 1 << window; |
640 | 0 | volatile BN_ULONG *table = (volatile BN_ULONG *)buf; |
641 | |
|
642 | 0 | if (bn_wexpand(b, top) == NULL) |
643 | 0 | return 0; |
644 | | |
645 | 0 | if (window <= 3) { |
646 | 0 | for (i = 0; i < top; i++, table += width) { |
647 | 0 | BN_ULONG acc = 0; |
648 | |
|
649 | 0 | for (j = 0; j < width; j++) { |
650 | 0 | acc |= table[j] & |
651 | 0 | ((BN_ULONG)0 - (constant_time_eq_int(j,idx)&1)); |
652 | 0 | } |
653 | |
|
654 | 0 | b->d[i] = acc; |
655 | 0 | } |
656 | 0 | } else { |
657 | 0 | int xstride = 1 << (window - 2); |
658 | 0 | BN_ULONG y0, y1, y2, y3; |
659 | |
|
660 | 0 | i = idx >> (window - 2); /* equivalent of idx / xstride */ |
661 | 0 | idx &= xstride - 1; /* equivalent of idx % xstride */ |
662 | |
|
663 | 0 | y0 = (BN_ULONG)0 - (constant_time_eq_int(i,0)&1); |
664 | 0 | y1 = (BN_ULONG)0 - (constant_time_eq_int(i,1)&1); |
665 | 0 | y2 = (BN_ULONG)0 - (constant_time_eq_int(i,2)&1); |
666 | 0 | y3 = (BN_ULONG)0 - (constant_time_eq_int(i,3)&1); |
667 | |
|
668 | 0 | for (i = 0; i < top; i++, table += width) { |
669 | 0 | BN_ULONG acc = 0; |
670 | |
|
671 | 0 | for (j = 0; j < xstride; j++) { |
672 | 0 | acc |= ( (table[j + 0 * xstride] & y0) | |
673 | 0 | (table[j + 1 * xstride] & y1) | |
674 | 0 | (table[j + 2 * xstride] & y2) | |
675 | 0 | (table[j + 3 * xstride] & y3) ) |
676 | 0 | & ((BN_ULONG)0 - (constant_time_eq_int(j,idx)&1)); |
677 | 0 | } |
678 | |
|
679 | 0 | b->d[i] = acc; |
680 | 0 | } |
681 | 0 | } |
682 | |
|
683 | 0 | b->top = top; |
684 | 0 | bn_correct_top(b); |
685 | 0 | return 1; |
686 | 0 | } |
687 | | |
688 | | /* |
689 | | * Given a pointer value, compute the next address that is a cache line |
690 | | * multiple. |
691 | | */ |
692 | | #define MOD_EXP_CTIME_ALIGN(x_) \ |
693 | 0 | ((unsigned char*)(x_) + (MOD_EXP_CTIME_MIN_CACHE_LINE_WIDTH - (((size_t)(x_)) & (MOD_EXP_CTIME_MIN_CACHE_LINE_MASK)))) |
694 | | |
695 | | /* |
696 | | * This variant of BN_mod_exp_mont() uses fixed windows and the special |
697 | | * precomputation memory layout to limit data-dependency to a minimum to |
698 | | * protect secret exponents (cf. the hyper-threading timing attacks pointed |
699 | | * out by Colin Percival, |
700 | | * http://www.daemonology.net/hyperthreading-considered-harmful/) |
701 | | */ |
702 | | int BN_mod_exp_mont_consttime(BIGNUM *rr, const BIGNUM *a, const BIGNUM *p, |
703 | | const BIGNUM *m, BN_CTX *ctx, |
704 | | BN_MONT_CTX *in_mont) |
705 | 0 | { |
706 | 0 | int i, bits, ret = 0, window, wvalue; |
707 | 0 | int top; |
708 | 0 | BN_MONT_CTX *mont = NULL; |
709 | |
|
710 | 0 | int numPowers; |
711 | 0 | unsigned char *powerbufFree = NULL; |
712 | 0 | int powerbufLen = 0; |
713 | 0 | unsigned char *powerbuf = NULL; |
714 | 0 | BIGNUM tmp, am; |
715 | | #if defined(SPARC_T4_MONT) |
716 | | unsigned int t4 = 0; |
717 | | #endif |
718 | |
|
719 | 0 | bn_check_top(a); |
720 | 0 | bn_check_top(p); |
721 | 0 | bn_check_top(m); |
722 | |
|
723 | 0 | if (!BN_is_odd(m)) { |
724 | 0 | BNerr(BN_F_BN_MOD_EXP_MONT_CONSTTIME, BN_R_CALLED_WITH_EVEN_MODULUS); |
725 | 0 | return (0); |
726 | 0 | } |
727 | | |
728 | 0 | top = m->top; |
729 | |
|
730 | 0 | bits = BN_num_bits(p); |
731 | 0 | if (bits == 0) { |
732 | | /* x**0 mod 1 is still zero. */ |
733 | 0 | if (BN_is_one(m)) { |
734 | 0 | ret = 1; |
735 | 0 | BN_zero(rr); |
736 | 0 | } else { |
737 | 0 | ret = BN_one(rr); |
738 | 0 | } |
739 | 0 | return ret; |
740 | 0 | } |
741 | | |
742 | 0 | BN_CTX_start(ctx); |
743 | | |
744 | | /* |
745 | | * Allocate a montgomery context if it was not supplied by the caller. If |
746 | | * this is not done, things will break in the montgomery part. |
747 | | */ |
748 | 0 | if (in_mont != NULL) |
749 | 0 | mont = in_mont; |
750 | 0 | else { |
751 | 0 | if ((mont = BN_MONT_CTX_new()) == NULL) |
752 | 0 | goto err; |
753 | 0 | if (!BN_MONT_CTX_set(mont, m, ctx)) |
754 | 0 | goto err; |
755 | 0 | } |
756 | | |
757 | 0 | #ifdef RSAZ_ENABLED |
758 | | /* |
759 | | * If the size of the operands allow it, perform the optimized |
760 | | * RSAZ exponentiation. For further information see |
761 | | * crypto/bn/rsaz_exp.c and accompanying assembly modules. |
762 | | */ |
763 | 0 | if ((16 == a->top) && (16 == p->top) && (BN_num_bits(m) == 1024) |
764 | 0 | && rsaz_avx2_eligible()) { |
765 | 0 | if (NULL == bn_wexpand(rr, 16)) |
766 | 0 | goto err; |
767 | 0 | RSAZ_1024_mod_exp_avx2(rr->d, a->d, p->d, m->d, mont->RR.d, |
768 | 0 | mont->n0[0]); |
769 | 0 | rr->top = 16; |
770 | 0 | rr->neg = 0; |
771 | 0 | bn_correct_top(rr); |
772 | 0 | ret = 1; |
773 | 0 | goto err; |
774 | 0 | } else if ((8 == a->top) && (8 == p->top) && (BN_num_bits(m) == 512)) { |
775 | 0 | if (NULL == bn_wexpand(rr, 8)) |
776 | 0 | goto err; |
777 | 0 | RSAZ_512_mod_exp(rr->d, a->d, p->d, m->d, mont->n0[0], mont->RR.d); |
778 | 0 | rr->top = 8; |
779 | 0 | rr->neg = 0; |
780 | 0 | bn_correct_top(rr); |
781 | 0 | ret = 1; |
782 | 0 | goto err; |
783 | 0 | } |
784 | 0 | #endif |
785 | | |
786 | | /* Get the window size to use with size of p. */ |
787 | 0 | window = BN_window_bits_for_ctime_exponent_size(bits); |
788 | | #if defined(SPARC_T4_MONT) |
789 | | if (window >= 5 && (top & 15) == 0 && top <= 64 && |
790 | | (OPENSSL_sparcv9cap_P[1] & (CFR_MONTMUL | CFR_MONTSQR)) == |
791 | | (CFR_MONTMUL | CFR_MONTSQR) && (t4 = OPENSSL_sparcv9cap_P[0])) |
792 | | window = 5; |
793 | | else |
794 | | #endif |
795 | 0 | #if defined(OPENSSL_BN_ASM_MONT5) |
796 | 0 | if (window >= 5) { |
797 | 0 | window = 5; /* ~5% improvement for RSA2048 sign, and even |
798 | | * for RSA4096 */ |
799 | | /* reserve space for mont->N.d[] copy */ |
800 | 0 | powerbufLen += top * sizeof(mont->N.d[0]); |
801 | 0 | } |
802 | 0 | #endif |
803 | 0 | (void)0; |
804 | | |
805 | | /* |
806 | | * Allocate a buffer large enough to hold all of the pre-computed powers |
807 | | * of am, am itself and tmp. |
808 | | */ |
809 | 0 | numPowers = 1 << window; |
810 | 0 | powerbufLen += sizeof(m->d[0]) * (top * numPowers + |
811 | 0 | ((2 * top) > |
812 | 0 | numPowers ? (2 * top) : numPowers)); |
813 | 0 | #ifdef alloca |
814 | 0 | if (powerbufLen < 3072) |
815 | 0 | powerbufFree = |
816 | 0 | alloca(powerbufLen + MOD_EXP_CTIME_MIN_CACHE_LINE_WIDTH); |
817 | 0 | else |
818 | 0 | #endif |
819 | 0 | if ((powerbufFree = |
820 | 0 | (unsigned char *)OPENSSL_malloc(powerbufLen + |
821 | 0 | MOD_EXP_CTIME_MIN_CACHE_LINE_WIDTH)) |
822 | 0 | == NULL) |
823 | 0 | goto err; |
824 | | |
825 | 0 | powerbuf = MOD_EXP_CTIME_ALIGN(powerbufFree); |
826 | 0 | memset(powerbuf, 0, powerbufLen); |
827 | |
|
828 | 0 | #ifdef alloca |
829 | 0 | if (powerbufLen < 3072) |
830 | 0 | powerbufFree = NULL; |
831 | 0 | #endif |
832 | | |
833 | | /* lay down tmp and am right after powers table */ |
834 | 0 | tmp.d = (BN_ULONG *)(powerbuf + sizeof(m->d[0]) * top * numPowers); |
835 | 0 | am.d = tmp.d + top; |
836 | 0 | tmp.top = am.top = 0; |
837 | 0 | tmp.dmax = am.dmax = top; |
838 | 0 | tmp.neg = am.neg = 0; |
839 | 0 | tmp.flags = am.flags = BN_FLG_STATIC_DATA; |
840 | | |
841 | | /* prepare a^0 in Montgomery domain */ |
842 | 0 | #if 1 /* by Shay Gueron's suggestion */ |
843 | 0 | if (m->d[top - 1] & (((BN_ULONG)1) << (BN_BITS2 - 1))) { |
844 | | /* 2^(top*BN_BITS2) - m */ |
845 | 0 | tmp.d[0] = (0 - m->d[0]) & BN_MASK2; |
846 | 0 | for (i = 1; i < top; i++) |
847 | 0 | tmp.d[i] = (~m->d[i]) & BN_MASK2; |
848 | 0 | tmp.top = top; |
849 | 0 | } else |
850 | 0 | #endif |
851 | 0 | if (!BN_to_montgomery(&tmp, BN_value_one(), mont, ctx)) |
852 | 0 | goto err; |
853 | | |
854 | | /* prepare a^1 in Montgomery domain */ |
855 | 0 | if (a->neg || BN_ucmp(a, m) >= 0) { |
856 | 0 | if (!BN_mod(&am, a, m, ctx)) |
857 | 0 | goto err; |
858 | 0 | if (!BN_to_montgomery(&am, &am, mont, ctx)) |
859 | 0 | goto err; |
860 | 0 | } else if (!BN_to_montgomery(&am, a, mont, ctx)) |
861 | 0 | goto err; |
862 | | |
863 | | #if defined(SPARC_T4_MONT) |
864 | | if (t4) { |
865 | | typedef int (*bn_pwr5_mont_f) (BN_ULONG *tp, const BN_ULONG *np, |
866 | | const BN_ULONG *n0, const void *table, |
867 | | int power, int bits); |
868 | | int bn_pwr5_mont_t4_8(BN_ULONG *tp, const BN_ULONG *np, |
869 | | const BN_ULONG *n0, const void *table, |
870 | | int power, int bits); |
871 | | int bn_pwr5_mont_t4_16(BN_ULONG *tp, const BN_ULONG *np, |
872 | | const BN_ULONG *n0, const void *table, |
873 | | int power, int bits); |
874 | | int bn_pwr5_mont_t4_24(BN_ULONG *tp, const BN_ULONG *np, |
875 | | const BN_ULONG *n0, const void *table, |
876 | | int power, int bits); |
877 | | int bn_pwr5_mont_t4_32(BN_ULONG *tp, const BN_ULONG *np, |
878 | | const BN_ULONG *n0, const void *table, |
879 | | int power, int bits); |
880 | | static const bn_pwr5_mont_f pwr5_funcs[4] = { |
881 | | bn_pwr5_mont_t4_8, bn_pwr5_mont_t4_16, |
882 | | bn_pwr5_mont_t4_24, bn_pwr5_mont_t4_32 |
883 | | }; |
884 | | bn_pwr5_mont_f pwr5_worker = pwr5_funcs[top / 16 - 1]; |
885 | | |
886 | | typedef int (*bn_mul_mont_f) (BN_ULONG *rp, const BN_ULONG *ap, |
887 | | const void *bp, const BN_ULONG *np, |
888 | | const BN_ULONG *n0); |
889 | | int bn_mul_mont_t4_8(BN_ULONG *rp, const BN_ULONG *ap, const void *bp, |
890 | | const BN_ULONG *np, const BN_ULONG *n0); |
891 | | int bn_mul_mont_t4_16(BN_ULONG *rp, const BN_ULONG *ap, |
892 | | const void *bp, const BN_ULONG *np, |
893 | | const BN_ULONG *n0); |
894 | | int bn_mul_mont_t4_24(BN_ULONG *rp, const BN_ULONG *ap, |
895 | | const void *bp, const BN_ULONG *np, |
896 | | const BN_ULONG *n0); |
897 | | int bn_mul_mont_t4_32(BN_ULONG *rp, const BN_ULONG *ap, |
898 | | const void *bp, const BN_ULONG *np, |
899 | | const BN_ULONG *n0); |
900 | | static const bn_mul_mont_f mul_funcs[4] = { |
901 | | bn_mul_mont_t4_8, bn_mul_mont_t4_16, |
902 | | bn_mul_mont_t4_24, bn_mul_mont_t4_32 |
903 | | }; |
904 | | bn_mul_mont_f mul_worker = mul_funcs[top / 16 - 1]; |
905 | | |
906 | | void bn_mul_mont_vis3(BN_ULONG *rp, const BN_ULONG *ap, |
907 | | const void *bp, const BN_ULONG *np, |
908 | | const BN_ULONG *n0, int num); |
909 | | void bn_mul_mont_t4(BN_ULONG *rp, const BN_ULONG *ap, |
910 | | const void *bp, const BN_ULONG *np, |
911 | | const BN_ULONG *n0, int num); |
912 | | void bn_mul_mont_gather5_t4(BN_ULONG *rp, const BN_ULONG *ap, |
913 | | const void *table, const BN_ULONG *np, |
914 | | const BN_ULONG *n0, int num, int power); |
915 | | void bn_flip_n_scatter5_t4(const BN_ULONG *inp, size_t num, |
916 | | void *table, size_t power); |
917 | | void bn_gather5_t4(BN_ULONG *out, size_t num, |
918 | | void *table, size_t power); |
919 | | void bn_flip_t4(BN_ULONG *dst, BN_ULONG *src, size_t num); |
920 | | |
921 | | BN_ULONG *np = mont->N.d, *n0 = mont->n0; |
922 | | int stride = 5 * (6 - (top / 16 - 1)); /* multiple of 5, but less |
923 | | * than 32 */ |
924 | | |
925 | | /* |
926 | | * BN_to_montgomery can contaminate words above .top [in |
927 | | * BN_DEBUG[_DEBUG] build]... |
928 | | */ |
929 | | for (i = am.top; i < top; i++) |
930 | | am.d[i] = 0; |
931 | | for (i = tmp.top; i < top; i++) |
932 | | tmp.d[i] = 0; |
933 | | |
934 | | bn_flip_n_scatter5_t4(tmp.d, top, powerbuf, 0); |
935 | | bn_flip_n_scatter5_t4(am.d, top, powerbuf, 1); |
936 | | if (!(*mul_worker) (tmp.d, am.d, am.d, np, n0) && |
937 | | !(*mul_worker) (tmp.d, am.d, am.d, np, n0)) |
938 | | bn_mul_mont_vis3(tmp.d, am.d, am.d, np, n0, top); |
939 | | bn_flip_n_scatter5_t4(tmp.d, top, powerbuf, 2); |
940 | | |
941 | | for (i = 3; i < 32; i++) { |
942 | | /* Calculate a^i = a^(i-1) * a */ |
943 | | if (!(*mul_worker) (tmp.d, tmp.d, am.d, np, n0) && |
944 | | !(*mul_worker) (tmp.d, tmp.d, am.d, np, n0)) |
945 | | bn_mul_mont_vis3(tmp.d, tmp.d, am.d, np, n0, top); |
946 | | bn_flip_n_scatter5_t4(tmp.d, top, powerbuf, i); |
947 | | } |
948 | | |
949 | | /* switch to 64-bit domain */ |
950 | | np = alloca(top * sizeof(BN_ULONG)); |
951 | | top /= 2; |
952 | | bn_flip_t4(np, mont->N.d, top); |
953 | | |
954 | | bits--; |
955 | | for (wvalue = 0, i = bits % 5; i >= 0; i--, bits--) |
956 | | wvalue = (wvalue << 1) + BN_is_bit_set(p, bits); |
957 | | bn_gather5_t4(tmp.d, top, powerbuf, wvalue); |
958 | | |
959 | | /* |
960 | | * Scan the exponent one window at a time starting from the most |
961 | | * significant bits. |
962 | | */ |
963 | | while (bits >= 0) { |
964 | | if (bits < stride) |
965 | | stride = bits + 1; |
966 | | bits -= stride; |
967 | | wvalue = bn_get_bits(p, bits + 1); |
968 | | |
969 | | if ((*pwr5_worker) (tmp.d, np, n0, powerbuf, wvalue, stride)) |
970 | | continue; |
971 | | /* retry once and fall back */ |
972 | | if ((*pwr5_worker) (tmp.d, np, n0, powerbuf, wvalue, stride)) |
973 | | continue; |
974 | | |
975 | | bits += stride - 5; |
976 | | wvalue >>= stride - 5; |
977 | | wvalue &= 31; |
978 | | bn_mul_mont_t4(tmp.d, tmp.d, tmp.d, np, n0, top); |
979 | | bn_mul_mont_t4(tmp.d, tmp.d, tmp.d, np, n0, top); |
980 | | bn_mul_mont_t4(tmp.d, tmp.d, tmp.d, np, n0, top); |
981 | | bn_mul_mont_t4(tmp.d, tmp.d, tmp.d, np, n0, top); |
982 | | bn_mul_mont_t4(tmp.d, tmp.d, tmp.d, np, n0, top); |
983 | | bn_mul_mont_gather5_t4(tmp.d, tmp.d, powerbuf, np, n0, top, |
984 | | wvalue); |
985 | | } |
986 | | |
987 | | bn_flip_t4(tmp.d, tmp.d, top); |
988 | | top *= 2; |
989 | | /* back to 32-bit domain */ |
990 | | tmp.top = top; |
991 | | bn_correct_top(&tmp); |
992 | | OPENSSL_cleanse(np, top * sizeof(BN_ULONG)); |
993 | | } else |
994 | | #endif |
995 | 0 | #if defined(OPENSSL_BN_ASM_MONT5) |
996 | 0 | if (window == 5 && top > 1) { |
997 | | /* |
998 | | * This optimization uses ideas from http://eprint.iacr.org/2011/239, |
999 | | * specifically optimization of cache-timing attack countermeasures |
1000 | | * and pre-computation optimization. |
1001 | | */ |
1002 | | |
1003 | | /* |
1004 | | * Dedicated window==4 case improves 512-bit RSA sign by ~15%, but as |
1005 | | * 512-bit RSA is hardly relevant, we omit it to spare size... |
1006 | | */ |
1007 | 0 | void bn_mul_mont_gather5(BN_ULONG *rp, const BN_ULONG *ap, |
1008 | 0 | const void *table, const BN_ULONG *np, |
1009 | 0 | const BN_ULONG *n0, int num, int power); |
1010 | 0 | void bn_scatter5(const BN_ULONG *inp, size_t num, |
1011 | 0 | void *table, size_t power); |
1012 | 0 | void bn_gather5(BN_ULONG *out, size_t num, void *table, size_t power); |
1013 | 0 | void bn_power5(BN_ULONG *rp, const BN_ULONG *ap, |
1014 | 0 | const void *table, const BN_ULONG *np, |
1015 | 0 | const BN_ULONG *n0, int num, int power); |
1016 | 0 | int bn_get_bits5(const BN_ULONG *ap, int off); |
1017 | 0 | int bn_from_montgomery(BN_ULONG *rp, const BN_ULONG *ap, |
1018 | 0 | const BN_ULONG *not_used, const BN_ULONG *np, |
1019 | 0 | const BN_ULONG *n0, int num); |
1020 | |
|
1021 | 0 | BN_ULONG *n0 = mont->n0, *np; |
1022 | | |
1023 | | /* |
1024 | | * BN_to_montgomery can contaminate words above .top [in |
1025 | | * BN_DEBUG[_DEBUG] build]... |
1026 | | */ |
1027 | 0 | for (i = am.top; i < top; i++) |
1028 | 0 | am.d[i] = 0; |
1029 | 0 | for (i = tmp.top; i < top; i++) |
1030 | 0 | tmp.d[i] = 0; |
1031 | | |
1032 | | /* |
1033 | | * copy mont->N.d[] to improve cache locality |
1034 | | */ |
1035 | 0 | for (np = am.d + top, i = 0; i < top; i++) |
1036 | 0 | np[i] = mont->N.d[i]; |
1037 | |
|
1038 | 0 | bn_scatter5(tmp.d, top, powerbuf, 0); |
1039 | 0 | bn_scatter5(am.d, am.top, powerbuf, 1); |
1040 | 0 | bn_mul_mont(tmp.d, am.d, am.d, np, n0, top); |
1041 | 0 | bn_scatter5(tmp.d, top, powerbuf, 2); |
1042 | |
|
1043 | | # if 0 |
1044 | | for (i = 3; i < 32; i++) { |
1045 | | /* Calculate a^i = a^(i-1) * a */ |
1046 | | bn_mul_mont_gather5(tmp.d, am.d, powerbuf, np, n0, top, i - 1); |
1047 | | bn_scatter5(tmp.d, top, powerbuf, i); |
1048 | | } |
1049 | | # else |
1050 | | /* same as above, but uses squaring for 1/2 of operations */ |
1051 | 0 | for (i = 4; i < 32; i *= 2) { |
1052 | 0 | bn_mul_mont(tmp.d, tmp.d, tmp.d, np, n0, top); |
1053 | 0 | bn_scatter5(tmp.d, top, powerbuf, i); |
1054 | 0 | } |
1055 | 0 | for (i = 3; i < 8; i += 2) { |
1056 | 0 | int j; |
1057 | 0 | bn_mul_mont_gather5(tmp.d, am.d, powerbuf, np, n0, top, i - 1); |
1058 | 0 | bn_scatter5(tmp.d, top, powerbuf, i); |
1059 | 0 | for (j = 2 * i; j < 32; j *= 2) { |
1060 | 0 | bn_mul_mont(tmp.d, tmp.d, tmp.d, np, n0, top); |
1061 | 0 | bn_scatter5(tmp.d, top, powerbuf, j); |
1062 | 0 | } |
1063 | 0 | } |
1064 | 0 | for (; i < 16; i += 2) { |
1065 | 0 | bn_mul_mont_gather5(tmp.d, am.d, powerbuf, np, n0, top, i - 1); |
1066 | 0 | bn_scatter5(tmp.d, top, powerbuf, i); |
1067 | 0 | bn_mul_mont(tmp.d, tmp.d, tmp.d, np, n0, top); |
1068 | 0 | bn_scatter5(tmp.d, top, powerbuf, 2 * i); |
1069 | 0 | } |
1070 | 0 | for (; i < 32; i += 2) { |
1071 | 0 | bn_mul_mont_gather5(tmp.d, am.d, powerbuf, np, n0, top, i - 1); |
1072 | 0 | bn_scatter5(tmp.d, top, powerbuf, i); |
1073 | 0 | } |
1074 | 0 | # endif |
1075 | 0 | bits--; |
1076 | 0 | for (wvalue = 0, i = bits % 5; i >= 0; i--, bits--) |
1077 | 0 | wvalue = (wvalue << 1) + BN_is_bit_set(p, bits); |
1078 | 0 | bn_gather5(tmp.d, top, powerbuf, wvalue); |
1079 | | |
1080 | | /* |
1081 | | * Scan the exponent one window at a time starting from the most |
1082 | | * significant bits. |
1083 | | */ |
1084 | 0 | if (top & 7) |
1085 | 0 | while (bits >= 0) { |
1086 | 0 | for (wvalue = 0, i = 0; i < 5; i++, bits--) |
1087 | 0 | wvalue = (wvalue << 1) + BN_is_bit_set(p, bits); |
1088 | |
|
1089 | 0 | bn_mul_mont(tmp.d, tmp.d, tmp.d, np, n0, top); |
1090 | 0 | bn_mul_mont(tmp.d, tmp.d, tmp.d, np, n0, top); |
1091 | 0 | bn_mul_mont(tmp.d, tmp.d, tmp.d, np, n0, top); |
1092 | 0 | bn_mul_mont(tmp.d, tmp.d, tmp.d, np, n0, top); |
1093 | 0 | bn_mul_mont(tmp.d, tmp.d, tmp.d, np, n0, top); |
1094 | 0 | bn_mul_mont_gather5(tmp.d, tmp.d, powerbuf, np, n0, top, |
1095 | 0 | wvalue); |
1096 | 0 | } else { |
1097 | 0 | while (bits >= 0) { |
1098 | 0 | wvalue = bn_get_bits5(p->d, bits - 4); |
1099 | 0 | bits -= 5; |
1100 | 0 | bn_power5(tmp.d, tmp.d, powerbuf, np, n0, top, wvalue); |
1101 | 0 | } |
1102 | 0 | } |
1103 | |
|
1104 | 0 | ret = bn_from_montgomery(tmp.d, tmp.d, NULL, np, n0, top); |
1105 | 0 | tmp.top = top; |
1106 | 0 | bn_correct_top(&tmp); |
1107 | 0 | if (ret) { |
1108 | 0 | if (!BN_copy(rr, &tmp)) |
1109 | 0 | ret = 0; |
1110 | 0 | goto err; /* non-zero ret means it's not error */ |
1111 | 0 | } |
1112 | 0 | } else |
1113 | 0 | #endif |
1114 | 0 | { |
1115 | 0 | if (!MOD_EXP_CTIME_COPY_TO_PREBUF(&tmp, top, powerbuf, 0, window)) |
1116 | 0 | goto err; |
1117 | 0 | if (!MOD_EXP_CTIME_COPY_TO_PREBUF(&am, top, powerbuf, 1, window)) |
1118 | 0 | goto err; |
1119 | | |
1120 | | /* |
1121 | | * If the window size is greater than 1, then calculate |
1122 | | * val[i=2..2^winsize-1]. Powers are computed as a*a^(i-1) (even |
1123 | | * powers could instead be computed as (a^(i/2))^2 to use the slight |
1124 | | * performance advantage of sqr over mul). |
1125 | | */ |
1126 | 0 | if (window > 1) { |
1127 | 0 | if (!BN_mod_mul_montgomery(&tmp, &am, &am, mont, ctx)) |
1128 | 0 | goto err; |
1129 | 0 | if (!MOD_EXP_CTIME_COPY_TO_PREBUF(&tmp, top, powerbuf, 2, |
1130 | 0 | window)) |
1131 | 0 | goto err; |
1132 | 0 | for (i = 3; i < numPowers; i++) { |
1133 | | /* Calculate a^i = a^(i-1) * a */ |
1134 | 0 | if (!BN_mod_mul_montgomery(&tmp, &am, &tmp, mont, ctx)) |
1135 | 0 | goto err; |
1136 | 0 | if (!MOD_EXP_CTIME_COPY_TO_PREBUF(&tmp, top, powerbuf, i, |
1137 | 0 | window)) |
1138 | 0 | goto err; |
1139 | 0 | } |
1140 | 0 | } |
1141 | | |
1142 | 0 | bits--; |
1143 | 0 | for (wvalue = 0, i = bits % window; i >= 0; i--, bits--) |
1144 | 0 | wvalue = (wvalue << 1) + BN_is_bit_set(p, bits); |
1145 | 0 | if (!MOD_EXP_CTIME_COPY_FROM_PREBUF(&tmp, top, powerbuf, wvalue, |
1146 | 0 | window)) |
1147 | 0 | goto err; |
1148 | | |
1149 | | /* |
1150 | | * Scan the exponent one window at a time starting from the most |
1151 | | * significant bits. |
1152 | | */ |
1153 | 0 | while (bits >= 0) { |
1154 | 0 | wvalue = 0; /* The 'value' of the window */ |
1155 | | |
1156 | | /* Scan the window, squaring the result as we go */ |
1157 | 0 | for (i = 0; i < window; i++, bits--) { |
1158 | 0 | if (!BN_mod_mul_montgomery(&tmp, &tmp, &tmp, mont, ctx)) |
1159 | 0 | goto err; |
1160 | 0 | wvalue = (wvalue << 1) + BN_is_bit_set(p, bits); |
1161 | 0 | } |
1162 | | |
1163 | | /* |
1164 | | * Fetch the appropriate pre-computed value from the pre-buf |
1165 | | */ |
1166 | 0 | if (!MOD_EXP_CTIME_COPY_FROM_PREBUF(&am, top, powerbuf, wvalue, |
1167 | 0 | window)) |
1168 | 0 | goto err; |
1169 | | |
1170 | | /* Multiply the result into the intermediate result */ |
1171 | 0 | if (!BN_mod_mul_montgomery(&tmp, &tmp, &am, mont, ctx)) |
1172 | 0 | goto err; |
1173 | 0 | } |
1174 | 0 | } |
1175 | | |
1176 | | /* Convert the final result from montgomery to standard format */ |
1177 | | #if defined(SPARC_T4_MONT) |
1178 | | if (OPENSSL_sparcv9cap_P[0] & (SPARCV9_VIS3 | SPARCV9_PREFER_FPU)) { |
1179 | | am.d[0] = 1; /* borrow am */ |
1180 | | for (i = 1; i < top; i++) |
1181 | | am.d[i] = 0; |
1182 | | if (!BN_mod_mul_montgomery(rr, &tmp, &am, mont, ctx)) |
1183 | | goto err; |
1184 | | } else |
1185 | | #endif |
1186 | 0 | if (!BN_from_montgomery(rr, &tmp, mont, ctx)) |
1187 | 0 | goto err; |
1188 | 0 | ret = 1; |
1189 | 0 | err: |
1190 | 0 | if ((in_mont == NULL) && (mont != NULL)) |
1191 | 0 | BN_MONT_CTX_free(mont); |
1192 | 0 | if (powerbuf != NULL) { |
1193 | 0 | OPENSSL_cleanse(powerbuf, powerbufLen); |
1194 | 0 | if (powerbufFree) |
1195 | 0 | OPENSSL_free(powerbufFree); |
1196 | 0 | } |
1197 | 0 | BN_CTX_end(ctx); |
1198 | 0 | return (ret); |
1199 | 0 | } |
1200 | | |
1201 | | int BN_mod_exp_mont_word(BIGNUM *rr, BN_ULONG a, const BIGNUM *p, |
1202 | | const BIGNUM *m, BN_CTX *ctx, BN_MONT_CTX *in_mont) |
1203 | 0 | { |
1204 | 0 | BN_MONT_CTX *mont = NULL; |
1205 | 0 | int b, bits, ret = 0; |
1206 | 0 | int r_is_one; |
1207 | 0 | BN_ULONG w, next_w; |
1208 | 0 | BIGNUM *d, *r, *t; |
1209 | 0 | BIGNUM *swap_tmp; |
1210 | 0 | #define BN_MOD_MUL_WORD(r, w, m) \ |
1211 | 0 | (BN_mul_word(r, (w)) && \ |
1212 | 0 | (/* BN_ucmp(r, (m)) < 0 ? 1 :*/ \ |
1213 | 0 | (BN_mod(t, r, m, ctx) && (swap_tmp = r, r = t, t = swap_tmp, 1)))) |
1214 | | /* |
1215 | | * BN_MOD_MUL_WORD is only used with 'w' large, so the BN_ucmp test is |
1216 | | * probably more overhead than always using BN_mod (which uses BN_copy if |
1217 | | * a similar test returns true). |
1218 | | */ |
1219 | | /* |
1220 | | * We can use BN_mod and do not need BN_nnmod because our accumulator is |
1221 | | * never negative (the result of BN_mod does not depend on the sign of |
1222 | | * the modulus). |
1223 | | */ |
1224 | 0 | #define BN_TO_MONTGOMERY_WORD(r, w, mont) \ |
1225 | 0 | (BN_set_word(r, (w)) && BN_to_montgomery(r, r, (mont), ctx)) |
1226 | |
|
1227 | 0 | if (BN_get_flags(p, BN_FLG_CONSTTIME) != 0 |
1228 | 0 | || BN_get_flags(m, BN_FLG_CONSTTIME) != 0) { |
1229 | | /* BN_FLG_CONSTTIME only supported by BN_mod_exp_mont() */ |
1230 | 0 | BNerr(BN_F_BN_MOD_EXP_MONT_WORD, ERR_R_SHOULD_NOT_HAVE_BEEN_CALLED); |
1231 | 0 | return -1; |
1232 | 0 | } |
1233 | | |
1234 | 0 | bn_check_top(p); |
1235 | 0 | bn_check_top(m); |
1236 | |
|
1237 | 0 | if (!BN_is_odd(m)) { |
1238 | 0 | BNerr(BN_F_BN_MOD_EXP_MONT_WORD, BN_R_CALLED_WITH_EVEN_MODULUS); |
1239 | 0 | return (0); |
1240 | 0 | } |
1241 | 0 | if (m->top == 1) |
1242 | 0 | a %= m->d[0]; /* make sure that 'a' is reduced */ |
1243 | |
|
1244 | 0 | bits = BN_num_bits(p); |
1245 | 0 | if (bits == 0) { |
1246 | | /* x**0 mod 1 is still zero. */ |
1247 | 0 | if (BN_is_one(m)) { |
1248 | 0 | ret = 1; |
1249 | 0 | BN_zero(rr); |
1250 | 0 | } else { |
1251 | 0 | ret = BN_one(rr); |
1252 | 0 | } |
1253 | 0 | return ret; |
1254 | 0 | } |
1255 | 0 | if (a == 0) { |
1256 | 0 | BN_zero(rr); |
1257 | 0 | ret = 1; |
1258 | 0 | return ret; |
1259 | 0 | } |
1260 | | |
1261 | 0 | BN_CTX_start(ctx); |
1262 | 0 | d = BN_CTX_get(ctx); |
1263 | 0 | r = BN_CTX_get(ctx); |
1264 | 0 | t = BN_CTX_get(ctx); |
1265 | 0 | if (d == NULL || r == NULL || t == NULL) |
1266 | 0 | goto err; |
1267 | | |
1268 | 0 | if (in_mont != NULL) |
1269 | 0 | mont = in_mont; |
1270 | 0 | else { |
1271 | 0 | if ((mont = BN_MONT_CTX_new()) == NULL) |
1272 | 0 | goto err; |
1273 | 0 | if (!BN_MONT_CTX_set(mont, m, ctx)) |
1274 | 0 | goto err; |
1275 | 0 | } |
1276 | | |
1277 | 0 | r_is_one = 1; /* except for Montgomery factor */ |
1278 | | |
1279 | | /* bits-1 >= 0 */ |
1280 | | |
1281 | | /* The result is accumulated in the product r*w. */ |
1282 | 0 | w = a; /* bit 'bits-1' of 'p' is always set */ |
1283 | 0 | for (b = bits - 2; b >= 0; b--) { |
1284 | | /* First, square r*w. */ |
1285 | 0 | next_w = w * w; |
1286 | 0 | if ((next_w / w) != w) { /* overflow */ |
1287 | 0 | if (r_is_one) { |
1288 | 0 | if (!BN_TO_MONTGOMERY_WORD(r, w, mont)) |
1289 | 0 | goto err; |
1290 | 0 | r_is_one = 0; |
1291 | 0 | } else { |
1292 | 0 | if (!BN_MOD_MUL_WORD(r, w, m)) |
1293 | 0 | goto err; |
1294 | 0 | } |
1295 | 0 | next_w = 1; |
1296 | 0 | } |
1297 | 0 | w = next_w; |
1298 | 0 | if (!r_is_one) { |
1299 | 0 | if (!BN_mod_mul_montgomery(r, r, r, mont, ctx)) |
1300 | 0 | goto err; |
1301 | 0 | } |
1302 | | |
1303 | | /* Second, multiply r*w by 'a' if exponent bit is set. */ |
1304 | 0 | if (BN_is_bit_set(p, b)) { |
1305 | 0 | next_w = w * a; |
1306 | 0 | if ((next_w / a) != w) { /* overflow */ |
1307 | 0 | if (r_is_one) { |
1308 | 0 | if (!BN_TO_MONTGOMERY_WORD(r, w, mont)) |
1309 | 0 | goto err; |
1310 | 0 | r_is_one = 0; |
1311 | 0 | } else { |
1312 | 0 | if (!BN_MOD_MUL_WORD(r, w, m)) |
1313 | 0 | goto err; |
1314 | 0 | } |
1315 | 0 | next_w = a; |
1316 | 0 | } |
1317 | 0 | w = next_w; |
1318 | 0 | } |
1319 | 0 | } |
1320 | | |
1321 | | /* Finally, set r:=r*w. */ |
1322 | 0 | if (w != 1) { |
1323 | 0 | if (r_is_one) { |
1324 | 0 | if (!BN_TO_MONTGOMERY_WORD(r, w, mont)) |
1325 | 0 | goto err; |
1326 | 0 | r_is_one = 0; |
1327 | 0 | } else { |
1328 | 0 | if (!BN_MOD_MUL_WORD(r, w, m)) |
1329 | 0 | goto err; |
1330 | 0 | } |
1331 | 0 | } |
1332 | | |
1333 | 0 | if (r_is_one) { /* can happen only if a == 1 */ |
1334 | 0 | if (!BN_one(rr)) |
1335 | 0 | goto err; |
1336 | 0 | } else { |
1337 | 0 | if (!BN_from_montgomery(rr, r, mont, ctx)) |
1338 | 0 | goto err; |
1339 | 0 | } |
1340 | 0 | ret = 1; |
1341 | 0 | err: |
1342 | 0 | if ((in_mont == NULL) && (mont != NULL)) |
1343 | 0 | BN_MONT_CTX_free(mont); |
1344 | 0 | BN_CTX_end(ctx); |
1345 | 0 | bn_check_top(rr); |
1346 | 0 | return (ret); |
1347 | 0 | } |
1348 | | |
1349 | | /* The old fallback, simple version :-) */ |
1350 | | int BN_mod_exp_simple(BIGNUM *r, const BIGNUM *a, const BIGNUM *p, |
1351 | | const BIGNUM *m, BN_CTX *ctx) |
1352 | 0 | { |
1353 | 0 | int i, j, bits, ret = 0, wstart, wend, window, wvalue; |
1354 | 0 | int start = 1; |
1355 | 0 | BIGNUM *d; |
1356 | | /* Table of variables obtained from 'ctx' */ |
1357 | 0 | BIGNUM *val[TABLE_SIZE]; |
1358 | |
|
1359 | 0 | if (BN_get_flags(p, BN_FLG_CONSTTIME) != 0 |
1360 | 0 | || BN_get_flags(a, BN_FLG_CONSTTIME) != 0 |
1361 | 0 | || BN_get_flags(m, BN_FLG_CONSTTIME) != 0) { |
1362 | | /* BN_FLG_CONSTTIME only supported by BN_mod_exp_mont() */ |
1363 | 0 | BNerr(BN_F_BN_MOD_EXP_SIMPLE, ERR_R_SHOULD_NOT_HAVE_BEEN_CALLED); |
1364 | 0 | return -1; |
1365 | 0 | } |
1366 | | |
1367 | 0 | bits = BN_num_bits(p); |
1368 | 0 | if (bits == 0) { |
1369 | | /* x**0 mod 1 is still zero. */ |
1370 | 0 | if (BN_is_one(m)) { |
1371 | 0 | ret = 1; |
1372 | 0 | BN_zero(r); |
1373 | 0 | } else { |
1374 | 0 | ret = BN_one(r); |
1375 | 0 | } |
1376 | 0 | return ret; |
1377 | 0 | } |
1378 | | |
1379 | 0 | BN_CTX_start(ctx); |
1380 | 0 | d = BN_CTX_get(ctx); |
1381 | 0 | val[0] = BN_CTX_get(ctx); |
1382 | 0 | if (!d || !val[0]) |
1383 | 0 | goto err; |
1384 | | |
1385 | 0 | if (!BN_nnmod(val[0], a, m, ctx)) |
1386 | 0 | goto err; /* 1 */ |
1387 | 0 | if (BN_is_zero(val[0])) { |
1388 | 0 | BN_zero(r); |
1389 | 0 | ret = 1; |
1390 | 0 | goto err; |
1391 | 0 | } |
1392 | | |
1393 | 0 | window = BN_window_bits_for_exponent_size(bits); |
1394 | 0 | if (window > 1) { |
1395 | 0 | if (!BN_mod_mul(d, val[0], val[0], m, ctx)) |
1396 | 0 | goto err; /* 2 */ |
1397 | 0 | j = 1 << (window - 1); |
1398 | 0 | for (i = 1; i < j; i++) { |
1399 | 0 | if (((val[i] = BN_CTX_get(ctx)) == NULL) || |
1400 | 0 | !BN_mod_mul(val[i], val[i - 1], d, m, ctx)) |
1401 | 0 | goto err; |
1402 | 0 | } |
1403 | 0 | } |
1404 | | |
1405 | 0 | start = 1; /* This is used to avoid multiplication etc |
1406 | | * when there is only the value '1' in the |
1407 | | * buffer. */ |
1408 | 0 | wvalue = 0; /* The 'value' of the window */ |
1409 | 0 | wstart = bits - 1; /* The top bit of the window */ |
1410 | 0 | wend = 0; /* The bottom bit of the window */ |
1411 | |
|
1412 | 0 | if (!BN_one(r)) |
1413 | 0 | goto err; |
1414 | | |
1415 | 0 | for (;;) { |
1416 | 0 | if (BN_is_bit_set(p, wstart) == 0) { |
1417 | 0 | if (!start) |
1418 | 0 | if (!BN_mod_mul(r, r, r, m, ctx)) |
1419 | 0 | goto err; |
1420 | 0 | if (wstart == 0) |
1421 | 0 | break; |
1422 | 0 | wstart--; |
1423 | 0 | continue; |
1424 | 0 | } |
1425 | | /* |
1426 | | * We now have wstart on a 'set' bit, we now need to work out how bit |
1427 | | * a window to do. To do this we need to scan forward until the last |
1428 | | * set bit before the end of the window |
1429 | | */ |
1430 | 0 | j = wstart; |
1431 | 0 | wvalue = 1; |
1432 | 0 | wend = 0; |
1433 | 0 | for (i = 1; i < window; i++) { |
1434 | 0 | if (wstart - i < 0) |
1435 | 0 | break; |
1436 | 0 | if (BN_is_bit_set(p, wstart - i)) { |
1437 | 0 | wvalue <<= (i - wend); |
1438 | 0 | wvalue |= 1; |
1439 | 0 | wend = i; |
1440 | 0 | } |
1441 | 0 | } |
1442 | | |
1443 | | /* wend is the size of the current window */ |
1444 | 0 | j = wend + 1; |
1445 | | /* add the 'bytes above' */ |
1446 | 0 | if (!start) |
1447 | 0 | for (i = 0; i < j; i++) { |
1448 | 0 | if (!BN_mod_mul(r, r, r, m, ctx)) |
1449 | 0 | goto err; |
1450 | 0 | } |
1451 | | |
1452 | | /* wvalue will be an odd number < 2^window */ |
1453 | 0 | if (!BN_mod_mul(r, r, val[wvalue >> 1], m, ctx)) |
1454 | 0 | goto err; |
1455 | | |
1456 | | /* move the 'window' down further */ |
1457 | 0 | wstart -= wend + 1; |
1458 | 0 | wvalue = 0; |
1459 | 0 | start = 0; |
1460 | 0 | if (wstart < 0) |
1461 | 0 | break; |
1462 | 0 | } |
1463 | 0 | ret = 1; |
1464 | 0 | err: |
1465 | 0 | BN_CTX_end(ctx); |
1466 | 0 | bn_check_top(r); |
1467 | 0 | return (ret); |
1468 | 0 | } |