File: | out/../deps/openssl/openssl/crypto/bn/bn_exp.c |
Warning: | line 233, column 5 Value stored to 'wvalue' is never read |
Press '?' to see keyboard shortcuts
Keyboard shortcuts:
1 | /* |
2 | * Copyright 1995-2022 The OpenSSL Project Authors. All Rights Reserved. |
3 | * |
4 | * Licensed under the Apache License 2.0 (the "License"). You may not use |
5 | * this file except in compliance with the License. You can obtain a copy |
6 | * in the file LICENSE in the source distribution or at |
7 | * https://www.openssl.org/source/license.html |
8 | */ |
9 | |
10 | #include "internal/cryptlib.h" |
11 | #include "internal/constant_time.h" |
12 | #include "bn_local.h" |
13 | |
14 | #include <stdlib.h> |
15 | #ifdef _WIN32 |
16 | # include <malloc.h> |
17 | # ifndef alloca |
18 | # define alloca _alloca |
19 | # endif |
20 | #elif defined(__GNUC__4) |
21 | # ifndef alloca |
22 | # define alloca(s)__builtin_alloca (s) __builtin_alloca((s)) |
23 | # endif |
24 | #elif defined(__sun) |
25 | # include <alloca.h> |
26 | #endif |
27 | |
28 | #include "rsaz_exp.h" |
29 | |
30 | #undef SPARC_T4_MONT |
31 | #if defined(OPENSSL_BN_ASM_MONT1) && (defined(__sparc__) || defined(__sparc)) |
32 | # include "crypto/sparc_arch.h" |
33 | # define SPARC_T4_MONT |
34 | #endif |
35 | |
36 | /* maximum precomputation table size for *variable* sliding windows */ |
37 | #define TABLE_SIZE32 32 |
38 | |
39 | /* this one works - simple but works */ |
40 | int BN_exp(BIGNUM *r, const BIGNUM *a, const BIGNUM *p, BN_CTX *ctx) |
41 | { |
42 | int i, bits, ret = 0; |
43 | BIGNUM *v, *rr; |
44 | |
45 | if (BN_get_flags(p, BN_FLG_CONSTTIME0x04) != 0 |
46 | || BN_get_flags(a, BN_FLG_CONSTTIME0x04) != 0) { |
47 | /* BN_FLG_CONSTTIME only supported by BN_mod_exp_mont() */ |
48 | ERR_raise(ERR_LIB_BN, ERR_R_SHOULD_NOT_HAVE_BEEN_CALLED)(ERR_new(), ERR_set_debug("../deps/openssl/openssl/crypto/bn/bn_exp.c" ,48,__func__), ERR_set_error)((3),((257|((0x1 << 18L)|( 0x2 << 18L)))),((void*)0)); |
49 | return 0; |
50 | } |
51 | |
52 | BN_CTX_start(ctx); |
53 | rr = ((r == a) || (r == p)) ? BN_CTX_get(ctx) : r; |
54 | v = BN_CTX_get(ctx); |
55 | if (rr == NULL((void*)0) || v == NULL((void*)0)) |
56 | goto err; |
57 | |
58 | if (BN_copy(v, a) == NULL((void*)0)) |
59 | goto err; |
60 | bits = BN_num_bits(p); |
61 | |
62 | if (BN_is_odd(p)) { |
63 | if (BN_copy(rr, a) == NULL((void*)0)) |
64 | goto err; |
65 | } else { |
66 | if (!BN_one(rr)(BN_set_word((rr),1))) |
67 | goto err; |
68 | } |
69 | |
70 | for (i = 1; i < bits; i++) { |
71 | if (!BN_sqr(v, v, ctx)) |
72 | goto err; |
73 | if (BN_is_bit_set(p, i)) { |
74 | if (!BN_mul(rr, rr, v, ctx)) |
75 | goto err; |
76 | } |
77 | } |
78 | if (r != rr && BN_copy(r, rr) == NULL((void*)0)) |
79 | goto err; |
80 | |
81 | ret = 1; |
82 | err: |
83 | BN_CTX_end(ctx); |
84 | bn_check_top(r); |
85 | return ret; |
86 | } |
87 | |
88 | int BN_mod_exp(BIGNUM *r, const BIGNUM *a, const BIGNUM *p, const BIGNUM *m, |
89 | BN_CTX *ctx) |
90 | { |
91 | int ret; |
92 | |
93 | bn_check_top(a); |
94 | bn_check_top(p); |
95 | bn_check_top(m); |
96 | |
97 | /*- |
98 | * For even modulus m = 2^k*m_odd, it might make sense to compute |
99 | * a^p mod m_odd and a^p mod 2^k separately (with Montgomery |
100 | * exponentiation for the odd part), using appropriate exponent |
101 | * reductions, and combine the results using the CRT. |
102 | * |
103 | * For now, we use Montgomery only if the modulus is odd; otherwise, |
104 | * exponentiation using the reciprocal-based quick remaindering |
105 | * algorithm is used. |
106 | * |
107 | * (Timing obtained with expspeed.c [computations a^p mod m |
108 | * where a, p, m are of the same length: 256, 512, 1024, 2048, |
109 | * 4096, 8192 bits], compared to the running time of the |
110 | * standard algorithm: |
111 | * |
112 | * BN_mod_exp_mont 33 .. 40 % [AMD K6-2, Linux, debug configuration] |
113 | * 55 .. 77 % [UltraSparc processor, but |
114 | * debug-solaris-sparcv8-gcc conf.] |
115 | * |
116 | * BN_mod_exp_recp 50 .. 70 % [AMD K6-2, Linux, debug configuration] |
117 | * 62 .. 118 % [UltraSparc, debug-solaris-sparcv8-gcc] |
118 | * |
119 | * On the Sparc, BN_mod_exp_recp was faster than BN_mod_exp_mont |
120 | * at 2048 and more bits, but at 512 and 1024 bits, it was |
121 | * slower even than the standard algorithm! |
122 | * |
123 | * "Real" timings [linux-elf, solaris-sparcv9-gcc configurations] |
124 | * should be obtained when the new Montgomery reduction code |
125 | * has been integrated into OpenSSL.) |
126 | */ |
127 | |
128 | #define MONT_MUL_MOD |
129 | #define MONT_EXP_WORD |
130 | #define RECP_MUL_MOD |
131 | |
132 | #ifdef MONT_MUL_MOD |
133 | if (BN_is_odd(m)) { |
134 | # ifdef MONT_EXP_WORD |
135 | if (a->top == 1 && !a->neg |
136 | && (BN_get_flags(p, BN_FLG_CONSTTIME0x04) == 0) |
137 | && (BN_get_flags(a, BN_FLG_CONSTTIME0x04) == 0) |
138 | && (BN_get_flags(m, BN_FLG_CONSTTIME0x04) == 0)) { |
139 | BN_ULONGunsigned long A = a->d[0]; |
140 | ret = BN_mod_exp_mont_word(r, A, p, m, ctx, NULL((void*)0)); |
141 | } else |
142 | # endif |
143 | ret = BN_mod_exp_mont(r, a, p, m, ctx, NULL((void*)0)); |
144 | } else |
145 | #endif |
146 | #ifdef RECP_MUL_MOD |
147 | { |
148 | ret = BN_mod_exp_recp(r, a, p, m, ctx); |
149 | } |
150 | #else |
151 | { |
152 | ret = BN_mod_exp_simple(r, a, p, m, ctx); |
153 | } |
154 | #endif |
155 | |
156 | bn_check_top(r); |
157 | return ret; |
158 | } |
159 | |
160 | int BN_mod_exp_recp(BIGNUM *r, const BIGNUM *a, const BIGNUM *p, |
161 | const BIGNUM *m, BN_CTX *ctx) |
162 | { |
163 | int i, j, bits, ret = 0, wstart, wend, window, wvalue; |
164 | int start = 1; |
165 | BIGNUM *aa; |
166 | /* Table of variables obtained from 'ctx' */ |
167 | BIGNUM *val[TABLE_SIZE32]; |
168 | BN_RECP_CTX recp; |
169 | |
170 | if (BN_get_flags(p, BN_FLG_CONSTTIME0x04) != 0 |
171 | || BN_get_flags(a, BN_FLG_CONSTTIME0x04) != 0 |
172 | || BN_get_flags(m, BN_FLG_CONSTTIME0x04) != 0) { |
173 | /* BN_FLG_CONSTTIME only supported by BN_mod_exp_mont() */ |
174 | ERR_raise(ERR_LIB_BN, ERR_R_SHOULD_NOT_HAVE_BEEN_CALLED)(ERR_new(), ERR_set_debug("../deps/openssl/openssl/crypto/bn/bn_exp.c" ,174,__func__), ERR_set_error)((3),((257|((0x1 << 18L)| (0x2 << 18L)))),((void*)0)); |
175 | return 0; |
176 | } |
177 | |
178 | bits = BN_num_bits(p); |
179 | if (bits == 0) { |
180 | /* x**0 mod 1, or x**0 mod -1 is still zero. */ |
181 | if (BN_abs_is_word(m, 1)) { |
182 | ret = 1; |
183 | BN_zero(r)BN_zero_ex(r); |
184 | } else { |
185 | ret = BN_one(r)(BN_set_word((r),1)); |
186 | } |
187 | return ret; |
188 | } |
189 | |
190 | BN_RECP_CTX_init(&recp); |
191 | |
192 | BN_CTX_start(ctx); |
193 | aa = BN_CTX_get(ctx); |
194 | val[0] = BN_CTX_get(ctx); |
195 | if (val[0] == NULL((void*)0)) |
196 | goto err; |
197 | |
198 | if (m->neg) { |
199 | /* ignore sign of 'm' */ |
200 | if (!BN_copy(aa, m)) |
201 | goto err; |
202 | aa->neg = 0; |
203 | if (BN_RECP_CTX_set(&recp, aa, ctx) <= 0) |
204 | goto err; |
205 | } else { |
206 | if (BN_RECP_CTX_set(&recp, m, ctx) <= 0) |
207 | goto err; |
208 | } |
209 | |
210 | if (!BN_nnmod(val[0], a, m, ctx)) |
211 | goto err; /* 1 */ |
212 | if (BN_is_zero(val[0])) { |
213 | BN_zero(r)BN_zero_ex(r); |
214 | ret = 1; |
215 | goto err; |
216 | } |
217 | |
218 | window = BN_window_bits_for_exponent_size(bits)((bits) > 671 ? 6 : (bits) > 239 ? 5 : (bits) > 79 ? 4 : (bits) > 23 ? 3 : 1); |
219 | if (window > 1) { |
220 | if (!BN_mod_mul_reciprocal(aa, val[0], val[0], &recp, ctx)) |
221 | goto err; /* 2 */ |
222 | j = 1 << (window - 1); |
223 | for (i = 1; i < j; i++) { |
224 | if (((val[i] = BN_CTX_get(ctx)) == NULL((void*)0)) || |
225 | !BN_mod_mul_reciprocal(val[i], val[i - 1], aa, &recp, ctx)) |
226 | goto err; |
227 | } |
228 | } |
229 | |
230 | start = 1; /* This is used to avoid multiplication etc |
231 | * when there is only the value '1' in the |
232 | * buffer. */ |
233 | wvalue = 0; /* The 'value' of the window */ |
Value stored to 'wvalue' is never read | |
234 | wstart = bits - 1; /* The top bit of the window */ |
235 | wend = 0; /* The bottom bit of the window */ |
236 | |
237 | if (!BN_one(r)(BN_set_word((r),1))) |
238 | goto err; |
239 | |
240 | for (;;) { |
241 | if (BN_is_bit_set(p, wstart) == 0) { |
242 | if (!start) |
243 | if (!BN_mod_mul_reciprocal(r, r, r, &recp, ctx)) |
244 | goto err; |
245 | if (wstart == 0) |
246 | break; |
247 | wstart--; |
248 | continue; |
249 | } |
250 | /* |
251 | * We now have wstart on a 'set' bit, we now need to work out how bit |
252 | * a window to do. To do this we need to scan forward until the last |
253 | * set bit before the end of the window |
254 | */ |
255 | wvalue = 1; |
256 | wend = 0; |
257 | for (i = 1; i < window; i++) { |
258 | if (wstart - i < 0) |
259 | break; |
260 | if (BN_is_bit_set(p, wstart - i)) { |
261 | wvalue <<= (i - wend); |
262 | wvalue |= 1; |
263 | wend = i; |
264 | } |
265 | } |
266 | |
267 | /* wend is the size of the current window */ |
268 | j = wend + 1; |
269 | /* add the 'bytes above' */ |
270 | if (!start) |
271 | for (i = 0; i < j; i++) { |
272 | if (!BN_mod_mul_reciprocal(r, r, r, &recp, ctx)) |
273 | goto err; |
274 | } |
275 | |
276 | /* wvalue will be an odd number < 2^window */ |
277 | if (!BN_mod_mul_reciprocal(r, r, val[wvalue >> 1], &recp, ctx)) |
278 | goto err; |
279 | |
280 | /* move the 'window' down further */ |
281 | wstart -= wend + 1; |
282 | wvalue = 0; |
283 | start = 0; |
284 | if (wstart < 0) |
285 | break; |
286 | } |
287 | ret = 1; |
288 | err: |
289 | BN_CTX_end(ctx); |
290 | BN_RECP_CTX_free(&recp); |
291 | bn_check_top(r); |
292 | return ret; |
293 | } |
294 | |
295 | int BN_mod_exp_mont(BIGNUM *rr, const BIGNUM *a, const BIGNUM *p, |
296 | const BIGNUM *m, BN_CTX *ctx, BN_MONT_CTX *in_mont) |
297 | { |
298 | int i, j, bits, ret = 0, wstart, wend, window, wvalue; |
299 | int start = 1; |
300 | BIGNUM *d, *r; |
301 | const BIGNUM *aa; |
302 | /* Table of variables obtained from 'ctx' */ |
303 | BIGNUM *val[TABLE_SIZE32]; |
304 | BN_MONT_CTX *mont = NULL((void*)0); |
305 | |
306 | if (BN_get_flags(p, BN_FLG_CONSTTIME0x04) != 0 |
307 | || BN_get_flags(a, BN_FLG_CONSTTIME0x04) != 0 |
308 | || BN_get_flags(m, BN_FLG_CONSTTIME0x04) != 0) { |
309 | return BN_mod_exp_mont_consttime(rr, a, p, m, ctx, in_mont); |
310 | } |
311 | |
312 | bn_check_top(a); |
313 | bn_check_top(p); |
314 | bn_check_top(m); |
315 | |
316 | if (!BN_is_odd(m)) { |
317 | ERR_raise(ERR_LIB_BN, BN_R_CALLED_WITH_EVEN_MODULUS)(ERR_new(), ERR_set_debug("../deps/openssl/openssl/crypto/bn/bn_exp.c" ,317,__func__), ERR_set_error)((3),(102),((void*)0)); |
318 | return 0; |
319 | } |
320 | bits = BN_num_bits(p); |
321 | if (bits == 0) { |
322 | /* x**0 mod 1, or x**0 mod -1 is still zero. */ |
323 | if (BN_abs_is_word(m, 1)) { |
324 | ret = 1; |
325 | BN_zero(rr)BN_zero_ex(rr); |
326 | } else { |
327 | ret = BN_one(rr)(BN_set_word((rr),1)); |
328 | } |
329 | return ret; |
330 | } |
331 | |
332 | BN_CTX_start(ctx); |
333 | d = BN_CTX_get(ctx); |
334 | r = BN_CTX_get(ctx); |
335 | val[0] = BN_CTX_get(ctx); |
336 | if (val[0] == NULL((void*)0)) |
337 | goto err; |
338 | |
339 | /* |
340 | * If this is not done, things will break in the montgomery part |
341 | */ |
342 | |
343 | if (in_mont != NULL((void*)0)) |
344 | mont = in_mont; |
345 | else { |
346 | if ((mont = BN_MONT_CTX_new()) == NULL((void*)0)) |
347 | goto err; |
348 | if (!BN_MONT_CTX_set(mont, m, ctx)) |
349 | goto err; |
350 | } |
351 | |
352 | if (a->neg || BN_ucmp(a, m) >= 0) { |
353 | if (!BN_nnmod(val[0], a, m, ctx)) |
354 | goto err; |
355 | aa = val[0]; |
356 | } else |
357 | aa = a; |
358 | if (!bn_to_mont_fixed_top(val[0], aa, mont, ctx)) |
359 | goto err; /* 1 */ |
360 | |
361 | window = BN_window_bits_for_exponent_size(bits)((bits) > 671 ? 6 : (bits) > 239 ? 5 : (bits) > 79 ? 4 : (bits) > 23 ? 3 : 1); |
362 | if (window > 1) { |
363 | if (!bn_mul_mont_fixed_top(d, val[0], val[0], mont, ctx)) |
364 | goto err; /* 2 */ |
365 | j = 1 << (window - 1); |
366 | for (i = 1; i < j; i++) { |
367 | if (((val[i] = BN_CTX_get(ctx)) == NULL((void*)0)) || |
368 | !bn_mul_mont_fixed_top(val[i], val[i - 1], d, mont, ctx)) |
369 | goto err; |
370 | } |
371 | } |
372 | |
373 | start = 1; /* This is used to avoid multiplication etc |
374 | * when there is only the value '1' in the |
375 | * buffer. */ |
376 | wvalue = 0; /* The 'value' of the window */ |
377 | wstart = bits - 1; /* The top bit of the window */ |
378 | wend = 0; /* The bottom bit of the window */ |
379 | |
380 | #if 1 /* by Shay Gueron's suggestion */ |
381 | j = m->top; /* borrow j */ |
382 | if (m->d[j - 1] & (((BN_ULONGunsigned long)1) << (BN_BITS2(8 * 8) - 1))) { |
383 | if (bn_wexpand(r, j) == NULL((void*)0)) |
384 | goto err; |
385 | /* 2^(top*BN_BITS2) - m */ |
386 | r->d[0] = (0 - m->d[0]) & BN_MASK2(0xffffffffffffffffL); |
387 | for (i = 1; i < j; i++) |
388 | r->d[i] = (~m->d[i]) & BN_MASK2(0xffffffffffffffffL); |
389 | r->top = j; |
390 | r->flags |= BN_FLG_FIXED_TOP0; |
391 | } else |
392 | #endif |
393 | if (!bn_to_mont_fixed_top(r, BN_value_one(), mont, ctx)) |
394 | goto err; |
395 | for (;;) { |
396 | if (BN_is_bit_set(p, wstart) == 0) { |
397 | if (!start) { |
398 | if (!bn_mul_mont_fixed_top(r, r, r, mont, ctx)) |
399 | goto err; |
400 | } |
401 | if (wstart == 0) |
402 | break; |
403 | wstart--; |
404 | continue; |
405 | } |
406 | /* |
407 | * We now have wstart on a 'set' bit, we now need to work out how bit |
408 | * a window to do. To do this we need to scan forward until the last |
409 | * set bit before the end of the window |
410 | */ |
411 | wvalue = 1; |
412 | wend = 0; |
413 | for (i = 1; i < window; i++) { |
414 | if (wstart - i < 0) |
415 | break; |
416 | if (BN_is_bit_set(p, wstart - i)) { |
417 | wvalue <<= (i - wend); |
418 | wvalue |= 1; |
419 | wend = i; |
420 | } |
421 | } |
422 | |
423 | /* wend is the size of the current window */ |
424 | j = wend + 1; |
425 | /* add the 'bytes above' */ |
426 | if (!start) |
427 | for (i = 0; i < j; i++) { |
428 | if (!bn_mul_mont_fixed_top(r, r, r, mont, ctx)) |
429 | goto err; |
430 | } |
431 | |
432 | /* wvalue will be an odd number < 2^window */ |
433 | if (!bn_mul_mont_fixed_top(r, r, val[wvalue >> 1], mont, ctx)) |
434 | goto err; |
435 | |
436 | /* move the 'window' down further */ |
437 | wstart -= wend + 1; |
438 | wvalue = 0; |
439 | start = 0; |
440 | if (wstart < 0) |
441 | break; |
442 | } |
443 | /* |
444 | * Done with zero-padded intermediate BIGNUMs. Final BN_from_montgomery |
445 | * removes padding [if any] and makes return value suitable for public |
446 | * API consumer. |
447 | */ |
448 | #if defined(SPARC_T4_MONT) |
449 | if (OPENSSL_sparcv9cap_P[0] & (SPARCV9_VIS3 | SPARCV9_PREFER_FPU)) { |
450 | j = mont->N.top; /* borrow j */ |
451 | val[0]->d[0] = 1; /* borrow val[0] */ |
452 | for (i = 1; i < j; i++) |
453 | val[0]->d[i] = 0; |
454 | val[0]->top = j; |
455 | if (!BN_mod_mul_montgomery(rr, r, val[0], mont, ctx)) |
456 | goto err; |
457 | } else |
458 | #endif |
459 | if (!BN_from_montgomery(rr, r, mont, ctx)) |
460 | goto err; |
461 | ret = 1; |
462 | err: |
463 | if (in_mont == NULL((void*)0)) |
464 | BN_MONT_CTX_free(mont); |
465 | BN_CTX_end(ctx); |
466 | bn_check_top(rr); |
467 | return ret; |
468 | } |
469 | |
470 | static BN_ULONGunsigned long bn_get_bits(const BIGNUM *a, int bitpos) |
471 | { |
472 | BN_ULONGunsigned long ret = 0; |
473 | int wordpos; |
474 | |
475 | wordpos = bitpos / BN_BITS2(8 * 8); |
476 | bitpos %= BN_BITS2(8 * 8); |
477 | if (wordpos >= 0 && wordpos < a->top) { |
478 | ret = a->d[wordpos] & BN_MASK2(0xffffffffffffffffL); |
479 | if (bitpos) { |
480 | ret >>= bitpos; |
481 | if (++wordpos < a->top) |
482 | ret |= a->d[wordpos] << (BN_BITS2(8 * 8) - bitpos); |
483 | } |
484 | } |
485 | |
486 | return ret & BN_MASK2(0xffffffffffffffffL); |
487 | } |
488 | |
489 | /* |
490 | * BN_mod_exp_mont_consttime() stores the precomputed powers in a specific |
491 | * layout so that accessing any of these table values shows the same access |
492 | * pattern as far as cache lines are concerned. The following functions are |
493 | * used to transfer a BIGNUM from/to that table. |
494 | */ |
495 | |
496 | static int MOD_EXP_CTIME_COPY_TO_PREBUF(const BIGNUM *b, int top, |
497 | unsigned char *buf, int idx, |
498 | int window) |
499 | { |
500 | int i, j; |
501 | int width = 1 << window; |
502 | BN_ULONGunsigned long *table = (BN_ULONGunsigned long *)buf; |
503 | |
504 | if (top > b->top) |
505 | top = b->top; /* this works because 'buf' is explicitly |
506 | * zeroed */ |
507 | for (i = 0, j = idx; i < top; i++, j += width) { |
508 | table[j] = b->d[i]; |
509 | } |
510 | |
511 | return 1; |
512 | } |
513 | |
514 | static int MOD_EXP_CTIME_COPY_FROM_PREBUF(BIGNUM *b, int top, |
515 | unsigned char *buf, int idx, |
516 | int window) |
517 | { |
518 | int i, j; |
519 | int width = 1 << window; |
520 | /* |
521 | * We declare table 'volatile' in order to discourage compiler |
522 | * from reordering loads from the table. Concern is that if |
523 | * reordered in specific manner loads might give away the |
524 | * information we are trying to conceal. Some would argue that |
525 | * compiler can reorder them anyway, but it can as well be |
526 | * argued that doing so would be violation of standard... |
527 | */ |
528 | volatile BN_ULONGunsigned long *table = (volatile BN_ULONGunsigned long *)buf; |
529 | |
530 | if (bn_wexpand(b, top) == NULL((void*)0)) |
531 | return 0; |
532 | |
533 | if (window <= 3) { |
534 | for (i = 0; i < top; i++, table += width) { |
535 | BN_ULONGunsigned long acc = 0; |
536 | |
537 | for (j = 0; j < width; j++) { |
538 | acc |= table[j] & |
539 | ((BN_ULONGunsigned long)0 - (constant_time_eq_int(j,idx)&1)); |
540 | } |
541 | |
542 | b->d[i] = acc; |
543 | } |
544 | } else { |
545 | int xstride = 1 << (window - 2); |
546 | BN_ULONGunsigned long y0, y1, y2, y3; |
547 | |
548 | i = idx >> (window - 2); /* equivalent of idx / xstride */ |
549 | idx &= xstride - 1; /* equivalent of idx % xstride */ |
550 | |
551 | y0 = (BN_ULONGunsigned long)0 - (constant_time_eq_int(i,0)&1); |
552 | y1 = (BN_ULONGunsigned long)0 - (constant_time_eq_int(i,1)&1); |
553 | y2 = (BN_ULONGunsigned long)0 - (constant_time_eq_int(i,2)&1); |
554 | y3 = (BN_ULONGunsigned long)0 - (constant_time_eq_int(i,3)&1); |
555 | |
556 | for (i = 0; i < top; i++, table += width) { |
557 | BN_ULONGunsigned long acc = 0; |
558 | |
559 | for (j = 0; j < xstride; j++) { |
560 | acc |= ( (table[j + 0 * xstride] & y0) | |
561 | (table[j + 1 * xstride] & y1) | |
562 | (table[j + 2 * xstride] & y2) | |
563 | (table[j + 3 * xstride] & y3) ) |
564 | & ((BN_ULONGunsigned long)0 - (constant_time_eq_int(j,idx)&1)); |
565 | } |
566 | |
567 | b->d[i] = acc; |
568 | } |
569 | } |
570 | |
571 | b->top = top; |
572 | b->flags |= BN_FLG_FIXED_TOP0; |
573 | return 1; |
574 | } |
575 | |
576 | /* |
577 | * Given a pointer value, compute the next address that is a cache line |
578 | * multiple. |
579 | */ |
580 | #define MOD_EXP_CTIME_ALIGN(x_)((unsigned char*)(x_) + (( 64 ) - (((size_t)(x_)) & ((( 64 ) - 1))))) \ |
581 | ((unsigned char*)(x_) + (MOD_EXP_CTIME_MIN_CACHE_LINE_WIDTH( 64 ) - (((size_t)(x_)) & (MOD_EXP_CTIME_MIN_CACHE_LINE_MASK(( 64 ) - 1))))) |
582 | |
583 | /* |
584 | * This variant of BN_mod_exp_mont() uses fixed windows and the special |
585 | * precomputation memory layout to limit data-dependency to a minimum to |
586 | * protect secret exponents (cf. the hyper-threading timing attacks pointed |
587 | * out by Colin Percival, |
588 | * http://www.daemonology.net/hyperthreading-considered-harmful/) |
589 | */ |
590 | int BN_mod_exp_mont_consttime(BIGNUM *rr, const BIGNUM *a, const BIGNUM *p, |
591 | const BIGNUM *m, BN_CTX *ctx, |
592 | BN_MONT_CTX *in_mont) |
593 | { |
594 | int i, bits, ret = 0, window, wvalue, wmask, window0; |
595 | int top; |
596 | BN_MONT_CTX *mont = NULL((void*)0); |
597 | |
598 | int numPowers; |
599 | unsigned char *powerbufFree = NULL((void*)0); |
600 | int powerbufLen = 0; |
601 | unsigned char *powerbuf = NULL((void*)0); |
602 | BIGNUM tmp, am; |
603 | #if defined(SPARC_T4_MONT) |
604 | unsigned int t4 = 0; |
605 | #endif |
606 | |
607 | bn_check_top(a); |
608 | bn_check_top(p); |
609 | bn_check_top(m); |
610 | |
611 | if (!BN_is_odd(m)) { |
612 | ERR_raise(ERR_LIB_BN, BN_R_CALLED_WITH_EVEN_MODULUS)(ERR_new(), ERR_set_debug("../deps/openssl/openssl/crypto/bn/bn_exp.c" ,612,__func__), ERR_set_error)((3),(102),((void*)0)); |
613 | return 0; |
614 | } |
615 | |
616 | top = m->top; |
617 | |
618 | /* |
619 | * Use all bits stored in |p|, rather than |BN_num_bits|, so we do not leak |
620 | * whether the top bits are zero. |
621 | */ |
622 | bits = p->top * BN_BITS2(8 * 8); |
623 | if (bits == 0) { |
624 | /* x**0 mod 1, or x**0 mod -1 is still zero. */ |
625 | if (BN_abs_is_word(m, 1)) { |
626 | ret = 1; |
627 | BN_zero(rr)BN_zero_ex(rr); |
628 | } else { |
629 | ret = BN_one(rr)(BN_set_word((rr),1)); |
630 | } |
631 | return ret; |
632 | } |
633 | |
634 | BN_CTX_start(ctx); |
635 | |
636 | /* |
637 | * Allocate a montgomery context if it was not supplied by the caller. If |
638 | * this is not done, things will break in the montgomery part. |
639 | */ |
640 | if (in_mont != NULL((void*)0)) |
641 | mont = in_mont; |
642 | else { |
643 | if ((mont = BN_MONT_CTX_new()) == NULL((void*)0)) |
644 | goto err; |
645 | if (!BN_MONT_CTX_set(mont, m, ctx)) |
646 | goto err; |
647 | } |
648 | |
649 | if (a->neg || BN_ucmp(a, m) >= 0) { |
650 | BIGNUM *reduced = BN_CTX_get(ctx); |
651 | if (reduced == NULL((void*)0) |
652 | || !BN_nnmod(reduced, a, m, ctx)) { |
653 | goto err; |
654 | } |
655 | a = reduced; |
656 | } |
657 | |
658 | #ifdef RSAZ_ENABLED |
659 | /* |
660 | * If the size of the operands allow it, perform the optimized |
661 | * RSAZ exponentiation. For further information see |
662 | * crypto/bn/rsaz_exp.c and accompanying assembly modules. |
663 | */ |
664 | if ((16 == a->top) && (16 == p->top) && (BN_num_bits(m) == 1024) |
665 | && rsaz_avx2_eligible()) { |
666 | if (NULL((void*)0) == bn_wexpand(rr, 16)) |
667 | goto err; |
668 | RSAZ_1024_mod_exp_avx2(rr->d, a->d, p->d, m->d, mont->RR.d, |
669 | mont->n0[0]); |
670 | rr->top = 16; |
671 | rr->neg = 0; |
672 | bn_correct_top(rr); |
673 | ret = 1; |
674 | goto err; |
675 | } else if ((8 == a->top) && (8 == p->top) && (BN_num_bits(m) == 512)) { |
676 | if (NULL((void*)0) == bn_wexpand(rr, 8)) |
677 | goto err; |
678 | RSAZ_512_mod_exp(rr->d, a->d, p->d, m->d, mont->n0[0], mont->RR.d); |
679 | rr->top = 8; |
680 | rr->neg = 0; |
681 | bn_correct_top(rr); |
682 | ret = 1; |
683 | goto err; |
684 | } |
685 | #endif |
686 | |
687 | /* Get the window size to use with size of p. */ |
688 | window = BN_window_bits_for_ctime_exponent_size(bits)((bits) > 937 ? 6 : (bits) > 306 ? 5 : (bits) > 89 ? 4 : (bits) > 22 ? 3 : 1); |
689 | #if defined(SPARC_T4_MONT) |
690 | if (window >= 5 && (top & 15) == 0 && top <= 64 && |
691 | (OPENSSL_sparcv9cap_P[1] & (CFR_MONTMUL | CFR_MONTSQR)) == |
692 | (CFR_MONTMUL | CFR_MONTSQR) && (t4 = OPENSSL_sparcv9cap_P[0])) |
693 | window = 5; |
694 | else |
695 | #endif |
696 | #if defined(OPENSSL_BN_ASM_MONT51) |
697 | if (window >= 5) { |
698 | window = 5; /* ~5% improvement for RSA2048 sign, and even |
699 | * for RSA4096 */ |
700 | /* reserve space for mont->N.d[] copy */ |
701 | powerbufLen += top * sizeof(mont->N.d[0]); |
702 | } |
703 | #endif |
704 | (void)0; |
705 | |
706 | /* |
707 | * Allocate a buffer large enough to hold all of the pre-computed powers |
708 | * of am, am itself and tmp. |
709 | */ |
710 | numPowers = 1 << window; |
711 | powerbufLen += sizeof(m->d[0]) * (top * numPowers + |
712 | ((2 * top) > |
713 | numPowers ? (2 * top) : numPowers)); |
714 | #ifdef alloca |
715 | if (powerbufLen < 3072) |
716 | powerbufFree = |
717 | alloca(powerbufLen + MOD_EXP_CTIME_MIN_CACHE_LINE_WIDTH)__builtin_alloca (powerbufLen + ( 64 )); |
718 | else |
719 | #endif |
720 | if ((powerbufFree = |
721 | OPENSSL_malloc(powerbufLen + MOD_EXP_CTIME_MIN_CACHE_LINE_WIDTH)CRYPTO_malloc(powerbufLen + ( 64 ), "../deps/openssl/openssl/crypto/bn/bn_exp.c" , 721)) |
722 | == NULL((void*)0)) |
723 | goto err; |
724 | |
725 | powerbuf = MOD_EXP_CTIME_ALIGN(powerbufFree)((unsigned char*)(powerbufFree) + (( 64 ) - (((size_t)(powerbufFree )) & ((( 64 ) - 1))))); |
726 | memset(powerbuf, 0, powerbufLen); |
727 | |
728 | #ifdef alloca |
729 | if (powerbufLen < 3072) |
730 | powerbufFree = NULL((void*)0); |
731 | #endif |
732 | |
733 | /* lay down tmp and am right after powers table */ |
734 | tmp.d = (BN_ULONGunsigned long *)(powerbuf + sizeof(m->d[0]) * top * numPowers); |
735 | am.d = tmp.d + top; |
736 | tmp.top = am.top = 0; |
737 | tmp.dmax = am.dmax = top; |
738 | tmp.neg = am.neg = 0; |
739 | tmp.flags = am.flags = BN_FLG_STATIC_DATA0x02; |
740 | |
741 | /* prepare a^0 in Montgomery domain */ |
742 | #if 1 /* by Shay Gueron's suggestion */ |
743 | if (m->d[top - 1] & (((BN_ULONGunsigned long)1) << (BN_BITS2(8 * 8) - 1))) { |
744 | /* 2^(top*BN_BITS2) - m */ |
745 | tmp.d[0] = (0 - m->d[0]) & BN_MASK2(0xffffffffffffffffL); |
746 | for (i = 1; i < top; i++) |
747 | tmp.d[i] = (~m->d[i]) & BN_MASK2(0xffffffffffffffffL); |
748 | tmp.top = top; |
749 | } else |
750 | #endif |
751 | if (!bn_to_mont_fixed_top(&tmp, BN_value_one(), mont, ctx)) |
752 | goto err; |
753 | |
754 | /* prepare a^1 in Montgomery domain */ |
755 | if (!bn_to_mont_fixed_top(&am, a, mont, ctx)) |
756 | goto err; |
757 | |
758 | #if defined(SPARC_T4_MONT) |
759 | if (t4) { |
760 | typedef int (*bn_pwr5_mont_f) (BN_ULONGunsigned long *tp, const BN_ULONGunsigned long *np, |
761 | const BN_ULONGunsigned long *n0, const void *table, |
762 | int power, int bits); |
763 | int bn_pwr5_mont_t4_8(BN_ULONGunsigned long *tp, const BN_ULONGunsigned long *np, |
764 | const BN_ULONGunsigned long *n0, const void *table, |
765 | int power, int bits); |
766 | int bn_pwr5_mont_t4_16(BN_ULONGunsigned long *tp, const BN_ULONGunsigned long *np, |
767 | const BN_ULONGunsigned long *n0, const void *table, |
768 | int power, int bits); |
769 | int bn_pwr5_mont_t4_24(BN_ULONGunsigned long *tp, const BN_ULONGunsigned long *np, |
770 | const BN_ULONGunsigned long *n0, const void *table, |
771 | int power, int bits); |
772 | int bn_pwr5_mont_t4_32(BN_ULONGunsigned long *tp, const BN_ULONGunsigned long *np, |
773 | const BN_ULONGunsigned long *n0, const void *table, |
774 | int power, int bits); |
775 | static const bn_pwr5_mont_f pwr5_funcs[4] = { |
776 | bn_pwr5_mont_t4_8, bn_pwr5_mont_t4_16, |
777 | bn_pwr5_mont_t4_24, bn_pwr5_mont_t4_32 |
778 | }; |
779 | bn_pwr5_mont_f pwr5_worker = pwr5_funcs[top / 16 - 1]; |
780 | |
781 | typedef int (*bn_mul_mont_f) (BN_ULONGunsigned long *rp, const BN_ULONGunsigned long *ap, |
782 | const void *bp, const BN_ULONGunsigned long *np, |
783 | const BN_ULONGunsigned long *n0); |
784 | int bn_mul_mont_t4_8(BN_ULONGunsigned long *rp, const BN_ULONGunsigned long *ap, const void *bp, |
785 | const BN_ULONGunsigned long *np, const BN_ULONGunsigned long *n0); |
786 | int bn_mul_mont_t4_16(BN_ULONGunsigned long *rp, const BN_ULONGunsigned long *ap, |
787 | const void *bp, const BN_ULONGunsigned long *np, |
788 | const BN_ULONGunsigned long *n0); |
789 | int bn_mul_mont_t4_24(BN_ULONGunsigned long *rp, const BN_ULONGunsigned long *ap, |
790 | const void *bp, const BN_ULONGunsigned long *np, |
791 | const BN_ULONGunsigned long *n0); |
792 | int bn_mul_mont_t4_32(BN_ULONGunsigned long *rp, const BN_ULONGunsigned long *ap, |
793 | const void *bp, const BN_ULONGunsigned long *np, |
794 | const BN_ULONGunsigned long *n0); |
795 | static const bn_mul_mont_f mul_funcs[4] = { |
796 | bn_mul_mont_t4_8, bn_mul_mont_t4_16, |
797 | bn_mul_mont_t4_24, bn_mul_mont_t4_32 |
798 | }; |
799 | bn_mul_mont_f mul_worker = mul_funcs[top / 16 - 1]; |
800 | |
801 | void bn_mul_mont_vis3(BN_ULONGunsigned long *rp, const BN_ULONGunsigned long *ap, |
802 | const void *bp, const BN_ULONGunsigned long *np, |
803 | const BN_ULONGunsigned long *n0, int num); |
804 | void bn_mul_mont_t4(BN_ULONGunsigned long *rp, const BN_ULONGunsigned long *ap, |
805 | const void *bp, const BN_ULONGunsigned long *np, |
806 | const BN_ULONGunsigned long *n0, int num); |
807 | void bn_mul_mont_gather5_t4(BN_ULONGunsigned long *rp, const BN_ULONGunsigned long *ap, |
808 | const void *table, const BN_ULONGunsigned long *np, |
809 | const BN_ULONGunsigned long *n0, int num, int power); |
810 | void bn_flip_n_scatter5_t4(const BN_ULONGunsigned long *inp, size_t num, |
811 | void *table, size_t power); |
812 | void bn_gather5_t4(BN_ULONGunsigned long *out, size_t num, |
813 | void *table, size_t power); |
814 | void bn_flip_t4(BN_ULONGunsigned long *dst, BN_ULONGunsigned long *src, size_t num); |
815 | |
816 | BN_ULONGunsigned long *np = mont->N.d, *n0 = mont->n0; |
817 | int stride = 5 * (6 - (top / 16 - 1)); /* multiple of 5, but less |
818 | * than 32 */ |
819 | |
820 | /* |
821 | * BN_to_montgomery can contaminate words above .top [in |
822 | * BN_DEBUG build... |
823 | */ |
824 | for (i = am.top; i < top; i++) |
825 | am.d[i] = 0; |
826 | for (i = tmp.top; i < top; i++) |
827 | tmp.d[i] = 0; |
828 | |
829 | bn_flip_n_scatter5_t4(tmp.d, top, powerbuf, 0); |
830 | bn_flip_n_scatter5_t4(am.d, top, powerbuf, 1); |
831 | if (!(*mul_worker) (tmp.d, am.d, am.d, np, n0) && |
832 | !(*mul_worker) (tmp.d, am.d, am.d, np, n0)) |
833 | bn_mul_mont_vis3(tmp.d, am.d, am.d, np, n0, top); |
834 | bn_flip_n_scatter5_t4(tmp.d, top, powerbuf, 2); |
835 | |
836 | for (i = 3; i < 32; i++) { |
837 | /* Calculate a^i = a^(i-1) * a */ |
838 | if (!(*mul_worker) (tmp.d, tmp.d, am.d, np, n0) && |
839 | !(*mul_worker) (tmp.d, tmp.d, am.d, np, n0)) |
840 | bn_mul_mont_vis3(tmp.d, tmp.d, am.d, np, n0, top); |
841 | bn_flip_n_scatter5_t4(tmp.d, top, powerbuf, i); |
842 | } |
843 | |
844 | /* switch to 64-bit domain */ |
845 | np = alloca(top * sizeof(BN_ULONG))__builtin_alloca (top * sizeof(unsigned long)); |
846 | top /= 2; |
847 | bn_flip_t4(np, mont->N.d, top); |
848 | |
849 | /* |
850 | * The exponent may not have a whole number of fixed-size windows. |
851 | * To simplify the main loop, the initial window has between 1 and |
852 | * full-window-size bits such that what remains is always a whole |
853 | * number of windows |
854 | */ |
855 | window0 = (bits - 1) % 5 + 1; |
856 | wmask = (1 << window0) - 1; |
857 | bits -= window0; |
858 | wvalue = bn_get_bits(p, bits) & wmask; |
859 | bn_gather5_t4(tmp.d, top, powerbuf, wvalue); |
860 | |
861 | /* |
862 | * Scan the exponent one window at a time starting from the most |
863 | * significant bits. |
864 | */ |
865 | while (bits > 0) { |
866 | if (bits < stride) |
867 | stride = bits; |
868 | bits -= stride; |
869 | wvalue = bn_get_bits(p, bits); |
870 | |
871 | if ((*pwr5_worker) (tmp.d, np, n0, powerbuf, wvalue, stride)) |
872 | continue; |
873 | /* retry once and fall back */ |
874 | if ((*pwr5_worker) (tmp.d, np, n0, powerbuf, wvalue, stride)) |
875 | continue; |
876 | |
877 | bits += stride - 5; |
878 | wvalue >>= stride - 5; |
879 | wvalue &= 31; |
880 | bn_mul_mont_t4(tmp.d, tmp.d, tmp.d, np, n0, top); |
881 | bn_mul_mont_t4(tmp.d, tmp.d, tmp.d, np, n0, top); |
882 | bn_mul_mont_t4(tmp.d, tmp.d, tmp.d, np, n0, top); |
883 | bn_mul_mont_t4(tmp.d, tmp.d, tmp.d, np, n0, top); |
884 | bn_mul_mont_t4(tmp.d, tmp.d, tmp.d, np, n0, top); |
885 | bn_mul_mont_gather5_t4(tmp.d, tmp.d, powerbuf, np, n0, top, |
886 | wvalue); |
887 | } |
888 | |
889 | bn_flip_t4(tmp.d, tmp.d, top); |
890 | top *= 2; |
891 | /* back to 32-bit domain */ |
892 | tmp.top = top; |
893 | bn_correct_top(&tmp); |
894 | OPENSSL_cleanse(np, top * sizeof(BN_ULONGunsigned long)); |
895 | } else |
896 | #endif |
897 | #if defined(OPENSSL_BN_ASM_MONT51) |
898 | if (window == 5 && top > 1) { |
899 | /* |
900 | * This optimization uses ideas from https://eprint.iacr.org/2011/239, |
901 | * specifically optimization of cache-timing attack countermeasures, |
902 | * pre-computation optimization, and Almost Montgomery Multiplication. |
903 | * |
904 | * The paper discusses a 4-bit window to optimize 512-bit modular |
905 | * exponentiation, used in RSA-1024 with CRT, but RSA-1024 is no longer |
906 | * important. |
907 | * |
908 | * |bn_mul_mont_gather5| and |bn_power5| implement the "almost" |
909 | * reduction variant, so the values here may not be fully reduced. |
910 | * They are bounded by R (i.e. they fit in |top| words), not |m|. |
911 | * Additionally, we pass these "almost" reduced inputs into |
912 | * |bn_mul_mont|, which implements the normal reduction variant. |
913 | * Given those inputs, |bn_mul_mont| may not give reduced |
914 | * output, but it will still produce "almost" reduced output. |
915 | */ |
916 | void bn_mul_mont_gather5(BN_ULONGunsigned long *rp, const BN_ULONGunsigned long *ap, |
917 | const void *table, const BN_ULONGunsigned long *np, |
918 | const BN_ULONGunsigned long *n0, int num, int power); |
919 | void bn_scatter5(const BN_ULONGunsigned long *inp, size_t num, |
920 | void *table, size_t power); |
921 | void bn_gather5(BN_ULONGunsigned long *out, size_t num, void *table, size_t power); |
922 | void bn_power5(BN_ULONGunsigned long *rp, const BN_ULONGunsigned long *ap, |
923 | const void *table, const BN_ULONGunsigned long *np, |
924 | const BN_ULONGunsigned long *n0, int num, int power); |
925 | int bn_get_bits5(const BN_ULONGunsigned long *ap, int off); |
926 | |
927 | BN_ULONGunsigned long *n0 = mont->n0, *np; |
928 | |
929 | /* |
930 | * BN_to_montgomery can contaminate words above .top [in |
931 | * BN_DEBUG build... |
932 | */ |
933 | for (i = am.top; i < top; i++) |
934 | am.d[i] = 0; |
935 | for (i = tmp.top; i < top; i++) |
936 | tmp.d[i] = 0; |
937 | |
938 | /* |
939 | * copy mont->N.d[] to improve cache locality |
940 | */ |
941 | for (np = am.d + top, i = 0; i < top; i++) |
942 | np[i] = mont->N.d[i]; |
943 | |
944 | bn_scatter5(tmp.d, top, powerbuf, 0); |
945 | bn_scatter5(am.d, am.top, powerbuf, 1); |
946 | bn_mul_mont(tmp.d, am.d, am.d, np, n0, top); |
947 | bn_scatter5(tmp.d, top, powerbuf, 2); |
948 | |
949 | # if 0 |
950 | for (i = 3; i < 32; i++) { |
951 | /* Calculate a^i = a^(i-1) * a */ |
952 | bn_mul_mont_gather5(tmp.d, am.d, powerbuf, np, n0, top, i - 1); |
953 | bn_scatter5(tmp.d, top, powerbuf, i); |
954 | } |
955 | # else |
956 | /* same as above, but uses squaring for 1/2 of operations */ |
957 | for (i = 4; i < 32; i *= 2) { |
958 | bn_mul_mont(tmp.d, tmp.d, tmp.d, np, n0, top); |
959 | bn_scatter5(tmp.d, top, powerbuf, i); |
960 | } |
961 | for (i = 3; i < 8; i += 2) { |
962 | int j; |
963 | bn_mul_mont_gather5(tmp.d, am.d, powerbuf, np, n0, top, i - 1); |
964 | bn_scatter5(tmp.d, top, powerbuf, i); |
965 | for (j = 2 * i; j < 32; j *= 2) { |
966 | bn_mul_mont(tmp.d, tmp.d, tmp.d, np, n0, top); |
967 | bn_scatter5(tmp.d, top, powerbuf, j); |
968 | } |
969 | } |
970 | for (; i < 16; i += 2) { |
971 | bn_mul_mont_gather5(tmp.d, am.d, powerbuf, np, n0, top, i - 1); |
972 | bn_scatter5(tmp.d, top, powerbuf, i); |
973 | bn_mul_mont(tmp.d, tmp.d, tmp.d, np, n0, top); |
974 | bn_scatter5(tmp.d, top, powerbuf, 2 * i); |
975 | } |
976 | for (; i < 32; i += 2) { |
977 | bn_mul_mont_gather5(tmp.d, am.d, powerbuf, np, n0, top, i - 1); |
978 | bn_scatter5(tmp.d, top, powerbuf, i); |
979 | } |
980 | # endif |
981 | /* |
982 | * The exponent may not have a whole number of fixed-size windows. |
983 | * To simplify the main loop, the initial window has between 1 and |
984 | * full-window-size bits such that what remains is always a whole |
985 | * number of windows |
986 | */ |
987 | window0 = (bits - 1) % 5 + 1; |
988 | wmask = (1 << window0) - 1; |
989 | bits -= window0; |
990 | wvalue = bn_get_bits(p, bits) & wmask; |
991 | bn_gather5(tmp.d, top, powerbuf, wvalue); |
992 | |
993 | /* |
994 | * Scan the exponent one window at a time starting from the most |
995 | * significant bits. |
996 | */ |
997 | if (top & 7) { |
998 | while (bits > 0) { |
999 | bn_mul_mont(tmp.d, tmp.d, tmp.d, np, n0, top); |
1000 | bn_mul_mont(tmp.d, tmp.d, tmp.d, np, n0, top); |
1001 | bn_mul_mont(tmp.d, tmp.d, tmp.d, np, n0, top); |
1002 | bn_mul_mont(tmp.d, tmp.d, tmp.d, np, n0, top); |
1003 | bn_mul_mont(tmp.d, tmp.d, tmp.d, np, n0, top); |
1004 | bn_mul_mont_gather5(tmp.d, tmp.d, powerbuf, np, n0, top, |
1005 | bn_get_bits5(p->d, bits -= 5)); |
1006 | } |
1007 | } else { |
1008 | while (bits > 0) { |
1009 | bn_power5(tmp.d, tmp.d, powerbuf, np, n0, top, |
1010 | bn_get_bits5(p->d, bits -= 5)); |
1011 | } |
1012 | } |
1013 | |
1014 | tmp.top = top; |
1015 | /* |
1016 | * The result is now in |tmp| in Montgomery form, but it may not be |
1017 | * fully reduced. This is within bounds for |BN_from_montgomery| |
1018 | * (tmp < R <= m*R) so it will, when converting from Montgomery form, |
1019 | * produce a fully reduced result. |
1020 | * |
1021 | * This differs from Figure 2 of the paper, which uses AMM(h, 1) to |
1022 | * convert from Montgomery form with unreduced output, followed by an |
1023 | * extra reduction step. In the paper's terminology, we replace |
1024 | * steps 9 and 10 with MM(h, 1). |
1025 | */ |
1026 | } else |
1027 | #endif |
1028 | { |
1029 | if (!MOD_EXP_CTIME_COPY_TO_PREBUF(&tmp, top, powerbuf, 0, window)) |
1030 | goto err; |
1031 | if (!MOD_EXP_CTIME_COPY_TO_PREBUF(&am, top, powerbuf, 1, window)) |
1032 | goto err; |
1033 | |
1034 | /* |
1035 | * If the window size is greater than 1, then calculate |
1036 | * val[i=2..2^winsize-1]. Powers are computed as a*a^(i-1) (even |
1037 | * powers could instead be computed as (a^(i/2))^2 to use the slight |
1038 | * performance advantage of sqr over mul). |
1039 | */ |
1040 | if (window > 1) { |
1041 | if (!bn_mul_mont_fixed_top(&tmp, &am, &am, mont, ctx)) |
1042 | goto err; |
1043 | if (!MOD_EXP_CTIME_COPY_TO_PREBUF(&tmp, top, powerbuf, 2, |
1044 | window)) |
1045 | goto err; |
1046 | for (i = 3; i < numPowers; i++) { |
1047 | /* Calculate a^i = a^(i-1) * a */ |
1048 | if (!bn_mul_mont_fixed_top(&tmp, &am, &tmp, mont, ctx)) |
1049 | goto err; |
1050 | if (!MOD_EXP_CTIME_COPY_TO_PREBUF(&tmp, top, powerbuf, i, |
1051 | window)) |
1052 | goto err; |
1053 | } |
1054 | } |
1055 | |
1056 | /* |
1057 | * The exponent may not have a whole number of fixed-size windows. |
1058 | * To simplify the main loop, the initial window has between 1 and |
1059 | * full-window-size bits such that what remains is always a whole |
1060 | * number of windows |
1061 | */ |
1062 | window0 = (bits - 1) % window + 1; |
1063 | wmask = (1 << window0) - 1; |
1064 | bits -= window0; |
1065 | wvalue = bn_get_bits(p, bits) & wmask; |
1066 | if (!MOD_EXP_CTIME_COPY_FROM_PREBUF(&tmp, top, powerbuf, wvalue, |
1067 | window)) |
1068 | goto err; |
1069 | |
1070 | wmask = (1 << window) - 1; |
1071 | /* |
1072 | * Scan the exponent one window at a time starting from the most |
1073 | * significant bits. |
1074 | */ |
1075 | while (bits > 0) { |
1076 | |
1077 | /* Square the result window-size times */ |
1078 | for (i = 0; i < window; i++) |
1079 | if (!bn_mul_mont_fixed_top(&tmp, &tmp, &tmp, mont, ctx)) |
1080 | goto err; |
1081 | |
1082 | /* |
1083 | * Get a window's worth of bits from the exponent |
1084 | * This avoids calling BN_is_bit_set for each bit, which |
1085 | * is not only slower but also makes each bit vulnerable to |
1086 | * EM (and likely other) side-channel attacks like One&Done |
1087 | * (for details see "One&Done: A Single-Decryption EM-Based |
1088 | * Attack on OpenSSL's Constant-Time Blinded RSA" by M. Alam, |
1089 | * H. Khan, M. Dey, N. Sinha, R. Callan, A. Zajic, and |
1090 | * M. Prvulovic, in USENIX Security'18) |
1091 | */ |
1092 | bits -= window; |
1093 | wvalue = bn_get_bits(p, bits) & wmask; |
1094 | /* |
1095 | * Fetch the appropriate pre-computed value from the pre-buf |
1096 | */ |
1097 | if (!MOD_EXP_CTIME_COPY_FROM_PREBUF(&am, top, powerbuf, wvalue, |
1098 | window)) |
1099 | goto err; |
1100 | |
1101 | /* Multiply the result into the intermediate result */ |
1102 | if (!bn_mul_mont_fixed_top(&tmp, &tmp, &am, mont, ctx)) |
1103 | goto err; |
1104 | } |
1105 | } |
1106 | |
1107 | /* |
1108 | * Done with zero-padded intermediate BIGNUMs. Final BN_from_montgomery |
1109 | * removes padding [if any] and makes return value suitable for public |
1110 | * API consumer. |
1111 | */ |
1112 | #if defined(SPARC_T4_MONT) |
1113 | if (OPENSSL_sparcv9cap_P[0] & (SPARCV9_VIS3 | SPARCV9_PREFER_FPU)) { |
1114 | am.d[0] = 1; /* borrow am */ |
1115 | for (i = 1; i < top; i++) |
1116 | am.d[i] = 0; |
1117 | if (!BN_mod_mul_montgomery(rr, &tmp, &am, mont, ctx)) |
1118 | goto err; |
1119 | } else |
1120 | #endif |
1121 | if (!BN_from_montgomery(rr, &tmp, mont, ctx)) |
1122 | goto err; |
1123 | ret = 1; |
1124 | err: |
1125 | if (in_mont == NULL((void*)0)) |
1126 | BN_MONT_CTX_free(mont); |
1127 | if (powerbuf != NULL((void*)0)) { |
1128 | OPENSSL_cleanse(powerbuf, powerbufLen); |
1129 | OPENSSL_free(powerbufFree)CRYPTO_free(powerbufFree, "../deps/openssl/openssl/crypto/bn/bn_exp.c" , 1129); |
1130 | } |
1131 | BN_CTX_end(ctx); |
1132 | return ret; |
1133 | } |
1134 | |
1135 | int BN_mod_exp_mont_word(BIGNUM *rr, BN_ULONGunsigned long a, const BIGNUM *p, |
1136 | const BIGNUM *m, BN_CTX *ctx, BN_MONT_CTX *in_mont) |
1137 | { |
1138 | BN_MONT_CTX *mont = NULL((void*)0); |
1139 | int b, bits, ret = 0; |
1140 | int r_is_one; |
1141 | BN_ULONGunsigned long w, next_w; |
1142 | BIGNUM *r, *t; |
1143 | BIGNUM *swap_tmp; |
1144 | #define BN_MOD_MUL_WORD(r, w, m)(BN_mul_word(r, (w)) && ( (BN_div(((void*)0),(t),(r), (m),(ctx)) && (swap_tmp = r, r = t, t = swap_tmp, 1)) )) \ |
1145 | (BN_mul_word(r, (w)) && \ |
1146 | (/* BN_ucmp(r, (m)) < 0 ? 1 :*/ \ |
1147 | (BN_mod(t, r, m, ctx)BN_div(((void*)0),(t),(r),(m),(ctx)) && (swap_tmp = r, r = t, t = swap_tmp, 1)))) |
1148 | /* |
1149 | * BN_MOD_MUL_WORD is only used with 'w' large, so the BN_ucmp test is |
1150 | * probably more overhead than always using BN_mod (which uses BN_copy if |
1151 | * a similar test returns true). |
1152 | */ |
1153 | /* |
1154 | * We can use BN_mod and do not need BN_nnmod because our accumulator is |
1155 | * never negative (the result of BN_mod does not depend on the sign of |
1156 | * the modulus). |
1157 | */ |
1158 | #define BN_TO_MONTGOMERY_WORD(r, w, mont)(BN_set_word(r, (w)) && BN_to_montgomery(r, r, (mont) , ctx)) \ |
1159 | (BN_set_word(r, (w)) && BN_to_montgomery(r, r, (mont), ctx)) |
1160 | |
1161 | if (BN_get_flags(p, BN_FLG_CONSTTIME0x04) != 0 |
1162 | || BN_get_flags(m, BN_FLG_CONSTTIME0x04) != 0) { |
1163 | /* BN_FLG_CONSTTIME only supported by BN_mod_exp_mont() */ |
1164 | ERR_raise(ERR_LIB_BN, ERR_R_SHOULD_NOT_HAVE_BEEN_CALLED)(ERR_new(), ERR_set_debug("../deps/openssl/openssl/crypto/bn/bn_exp.c" ,1164,__func__), ERR_set_error)((3),((257|((0x1 << 18L) |(0x2 << 18L)))),((void*)0)); |
1165 | return 0; |
1166 | } |
1167 | |
1168 | bn_check_top(p); |
1169 | bn_check_top(m); |
1170 | |
1171 | if (!BN_is_odd(m)) { |
1172 | ERR_raise(ERR_LIB_BN, BN_R_CALLED_WITH_EVEN_MODULUS)(ERR_new(), ERR_set_debug("../deps/openssl/openssl/crypto/bn/bn_exp.c" ,1172,__func__), ERR_set_error)((3),(102),((void*)0)); |
1173 | return 0; |
1174 | } |
1175 | if (m->top == 1) |
1176 | a %= m->d[0]; /* make sure that 'a' is reduced */ |
1177 | |
1178 | bits = BN_num_bits(p); |
1179 | if (bits == 0) { |
1180 | /* x**0 mod 1, or x**0 mod -1 is still zero. */ |
1181 | if (BN_abs_is_word(m, 1)) { |
1182 | ret = 1; |
1183 | BN_zero(rr)BN_zero_ex(rr); |
1184 | } else { |
1185 | ret = BN_one(rr)(BN_set_word((rr),1)); |
1186 | } |
1187 | return ret; |
1188 | } |
1189 | if (a == 0) { |
1190 | BN_zero(rr)BN_zero_ex(rr); |
1191 | ret = 1; |
1192 | return ret; |
1193 | } |
1194 | |
1195 | BN_CTX_start(ctx); |
1196 | r = BN_CTX_get(ctx); |
1197 | t = BN_CTX_get(ctx); |
1198 | if (t == NULL((void*)0)) |
1199 | goto err; |
1200 | |
1201 | if (in_mont != NULL((void*)0)) |
1202 | mont = in_mont; |
1203 | else { |
1204 | if ((mont = BN_MONT_CTX_new()) == NULL((void*)0)) |
1205 | goto err; |
1206 | if (!BN_MONT_CTX_set(mont, m, ctx)) |
1207 | goto err; |
1208 | } |
1209 | |
1210 | r_is_one = 1; /* except for Montgomery factor */ |
1211 | |
1212 | /* bits-1 >= 0 */ |
1213 | |
1214 | /* The result is accumulated in the product r*w. */ |
1215 | w = a; /* bit 'bits-1' of 'p' is always set */ |
1216 | for (b = bits - 2; b >= 0; b--) { |
1217 | /* First, square r*w. */ |
1218 | next_w = w * w; |
1219 | if ((next_w / w) != w) { /* overflow */ |
1220 | if (r_is_one) { |
1221 | if (!BN_TO_MONTGOMERY_WORD(r, w, mont)(BN_set_word(r, (w)) && BN_to_montgomery(r, r, (mont) , ctx))) |
1222 | goto err; |
1223 | r_is_one = 0; |
1224 | } else { |
1225 | if (!BN_MOD_MUL_WORD(r, w, m)(BN_mul_word(r, (w)) && ( (BN_div(((void*)0),(t),(r), (m),(ctx)) && (swap_tmp = r, r = t, t = swap_tmp, 1)) ))) |
1226 | goto err; |
1227 | } |
1228 | next_w = 1; |
1229 | } |
1230 | w = next_w; |
1231 | if (!r_is_one) { |
1232 | if (!BN_mod_mul_montgomery(r, r, r, mont, ctx)) |
1233 | goto err; |
1234 | } |
1235 | |
1236 | /* Second, multiply r*w by 'a' if exponent bit is set. */ |
1237 | if (BN_is_bit_set(p, b)) { |
1238 | next_w = w * a; |
1239 | if ((next_w / a) != w) { /* overflow */ |
1240 | if (r_is_one) { |
1241 | if (!BN_TO_MONTGOMERY_WORD(r, w, mont)(BN_set_word(r, (w)) && BN_to_montgomery(r, r, (mont) , ctx))) |
1242 | goto err; |
1243 | r_is_one = 0; |
1244 | } else { |
1245 | if (!BN_MOD_MUL_WORD(r, w, m)(BN_mul_word(r, (w)) && ( (BN_div(((void*)0),(t),(r), (m),(ctx)) && (swap_tmp = r, r = t, t = swap_tmp, 1)) ))) |
1246 | goto err; |
1247 | } |
1248 | next_w = a; |
1249 | } |
1250 | w = next_w; |
1251 | } |
1252 | } |
1253 | |
1254 | /* Finally, set r:=r*w. */ |
1255 | if (w != 1) { |
1256 | if (r_is_one) { |
1257 | if (!BN_TO_MONTGOMERY_WORD(r, w, mont)(BN_set_word(r, (w)) && BN_to_montgomery(r, r, (mont) , ctx))) |
1258 | goto err; |
1259 | r_is_one = 0; |
1260 | } else { |
1261 | if (!BN_MOD_MUL_WORD(r, w, m)(BN_mul_word(r, (w)) && ( (BN_div(((void*)0),(t),(r), (m),(ctx)) && (swap_tmp = r, r = t, t = swap_tmp, 1)) ))) |
1262 | goto err; |
1263 | } |
1264 | } |
1265 | |
1266 | if (r_is_one) { /* can happen only if a == 1 */ |
1267 | if (!BN_one(rr)(BN_set_word((rr),1))) |
1268 | goto err; |
1269 | } else { |
1270 | if (!BN_from_montgomery(rr, r, mont, ctx)) |
1271 | goto err; |
1272 | } |
1273 | ret = 1; |
1274 | err: |
1275 | if (in_mont == NULL((void*)0)) |
1276 | BN_MONT_CTX_free(mont); |
1277 | BN_CTX_end(ctx); |
1278 | bn_check_top(rr); |
1279 | return ret; |
1280 | } |
1281 | |
1282 | /* The old fallback, simple version :-) */ |
1283 | int BN_mod_exp_simple(BIGNUM *r, const BIGNUM *a, const BIGNUM *p, |
1284 | const BIGNUM *m, BN_CTX *ctx) |
1285 | { |
1286 | int i, j, bits, ret = 0, wstart, wend, window, wvalue; |
1287 | int start = 1; |
1288 | BIGNUM *d; |
1289 | /* Table of variables obtained from 'ctx' */ |
1290 | BIGNUM *val[TABLE_SIZE32]; |
1291 | |
1292 | if (BN_get_flags(p, BN_FLG_CONSTTIME0x04) != 0 |
1293 | || BN_get_flags(a, BN_FLG_CONSTTIME0x04) != 0 |
1294 | || BN_get_flags(m, BN_FLG_CONSTTIME0x04) != 0) { |
1295 | /* BN_FLG_CONSTTIME only supported by BN_mod_exp_mont() */ |
1296 | ERR_raise(ERR_LIB_BN, ERR_R_SHOULD_NOT_HAVE_BEEN_CALLED)(ERR_new(), ERR_set_debug("../deps/openssl/openssl/crypto/bn/bn_exp.c" ,1296,__func__), ERR_set_error)((3),((257|((0x1 << 18L) |(0x2 << 18L)))),((void*)0)); |
1297 | return 0; |
1298 | } |
1299 | |
1300 | bits = BN_num_bits(p); |
1301 | if (bits == 0) { |
1302 | /* x**0 mod 1, or x**0 mod -1 is still zero. */ |
1303 | if (BN_abs_is_word(m, 1)) { |
1304 | ret = 1; |
1305 | BN_zero(r)BN_zero_ex(r); |
1306 | } else { |
1307 | ret = BN_one(r)(BN_set_word((r),1)); |
1308 | } |
1309 | return ret; |
1310 | } |
1311 | |
1312 | BN_CTX_start(ctx); |
1313 | d = BN_CTX_get(ctx); |
1314 | val[0] = BN_CTX_get(ctx); |
1315 | if (val[0] == NULL((void*)0)) |
1316 | goto err; |
1317 | |
1318 | if (!BN_nnmod(val[0], a, m, ctx)) |
1319 | goto err; /* 1 */ |
1320 | if (BN_is_zero(val[0])) { |
1321 | BN_zero(r)BN_zero_ex(r); |
1322 | ret = 1; |
1323 | goto err; |
1324 | } |
1325 | |
1326 | window = BN_window_bits_for_exponent_size(bits)((bits) > 671 ? 6 : (bits) > 239 ? 5 : (bits) > 79 ? 4 : (bits) > 23 ? 3 : 1); |
1327 | if (window > 1) { |
1328 | if (!BN_mod_mul(d, val[0], val[0], m, ctx)) |
1329 | goto err; /* 2 */ |
1330 | j = 1 << (window - 1); |
1331 | for (i = 1; i < j; i++) { |
1332 | if (((val[i] = BN_CTX_get(ctx)) == NULL((void*)0)) || |
1333 | !BN_mod_mul(val[i], val[i - 1], d, m, ctx)) |
1334 | goto err; |
1335 | } |
1336 | } |
1337 | |
1338 | start = 1; /* This is used to avoid multiplication etc |
1339 | * when there is only the value '1' in the |
1340 | * buffer. */ |
1341 | wvalue = 0; /* The 'value' of the window */ |
1342 | wstart = bits - 1; /* The top bit of the window */ |
1343 | wend = 0; /* The bottom bit of the window */ |
1344 | |
1345 | if (!BN_one(r)(BN_set_word((r),1))) |
1346 | goto err; |
1347 | |
1348 | for (;;) { |
1349 | if (BN_is_bit_set(p, wstart) == 0) { |
1350 | if (!start) |
1351 | if (!BN_mod_mul(r, r, r, m, ctx)) |
1352 | goto err; |
1353 | if (wstart == 0) |
1354 | break; |
1355 | wstart--; |
1356 | continue; |
1357 | } |
1358 | /* |
1359 | * We now have wstart on a 'set' bit, we now need to work out how bit |
1360 | * a window to do. To do this we need to scan forward until the last |
1361 | * set bit before the end of the window |
1362 | */ |
1363 | wvalue = 1; |
1364 | wend = 0; |
1365 | for (i = 1; i < window; i++) { |
1366 | if (wstart - i < 0) |
1367 | break; |
1368 | if (BN_is_bit_set(p, wstart - i)) { |
1369 | wvalue <<= (i - wend); |
1370 | wvalue |= 1; |
1371 | wend = i; |
1372 | } |
1373 | } |
1374 | |
1375 | /* wend is the size of the current window */ |
1376 | j = wend + 1; |
1377 | /* add the 'bytes above' */ |
1378 | if (!start) |
1379 | for (i = 0; i < j; i++) { |
1380 | if (!BN_mod_mul(r, r, r, m, ctx)) |
1381 | goto err; |
1382 | } |
1383 | |
1384 | /* wvalue will be an odd number < 2^window */ |
1385 | if (!BN_mod_mul(r, r, val[wvalue >> 1], m, ctx)) |
1386 | goto err; |
1387 | |
1388 | /* move the 'window' down further */ |
1389 | wstart -= wend + 1; |
1390 | wvalue = 0; |
1391 | start = 0; |
1392 | if (wstart < 0) |
1393 | break; |
1394 | } |
1395 | ret = 1; |
1396 | err: |
1397 | BN_CTX_end(ctx); |
1398 | bn_check_top(r); |
1399 | return ret; |
1400 | } |
1401 | |
1402 | /* |
1403 | * This is a variant of modular exponentiation optimization that does |
1404 | * parallel 2-primes exponentiation using 256-bit (AVX512VL) AVX512_IFMA ISA |
1405 | * in 52-bit binary redundant representation. |
1406 | * If such instructions are not available, or input data size is not supported, |
1407 | * it falls back to two BN_mod_exp_mont_consttime() calls. |
1408 | */ |
1409 | int BN_mod_exp_mont_consttime_x2(BIGNUM *rr1, const BIGNUM *a1, const BIGNUM *p1, |
1410 | const BIGNUM *m1, BN_MONT_CTX *in_mont1, |
1411 | BIGNUM *rr2, const BIGNUM *a2, const BIGNUM *p2, |
1412 | const BIGNUM *m2, BN_MONT_CTX *in_mont2, |
1413 | BN_CTX *ctx) |
1414 | { |
1415 | int ret = 0; |
1416 | |
1417 | #ifdef RSAZ_ENABLED |
1418 | BN_MONT_CTX *mont1 = NULL((void*)0); |
1419 | BN_MONT_CTX *mont2 = NULL((void*)0); |
1420 | |
1421 | if (ossl_rsaz_avx512ifma_eligible() && |
1422 | ((a1->top == 16) && (p1->top == 16) && (BN_num_bits(m1) == 1024) && |
1423 | (a2->top == 16) && (p2->top == 16) && (BN_num_bits(m2) == 1024))) { |
1424 | |
1425 | if (bn_wexpand(rr1, 16) == NULL((void*)0)) |
1426 | goto err; |
1427 | if (bn_wexpand(rr2, 16) == NULL((void*)0)) |
1428 | goto err; |
1429 | |
1430 | /* Ensure that montgomery contexts are initialized */ |
1431 | if (in_mont1 != NULL((void*)0)) { |
1432 | mont1 = in_mont1; |
1433 | } else { |
1434 | if ((mont1 = BN_MONT_CTX_new()) == NULL((void*)0)) |
1435 | goto err; |
1436 | if (!BN_MONT_CTX_set(mont1, m1, ctx)) |
1437 | goto err; |
1438 | } |
1439 | if (in_mont2 != NULL((void*)0)) { |
1440 | mont2 = in_mont2; |
1441 | } else { |
1442 | if ((mont2 = BN_MONT_CTX_new()) == NULL((void*)0)) |
1443 | goto err; |
1444 | if (!BN_MONT_CTX_set(mont2, m2, ctx)) |
1445 | goto err; |
1446 | } |
1447 | |
1448 | ret = ossl_rsaz_mod_exp_avx512_x2(rr1->d, a1->d, p1->d, m1->d, |
1449 | mont1->RR.d, mont1->n0[0], |
1450 | rr2->d, a2->d, p2->d, m2->d, |
1451 | mont2->RR.d, mont2->n0[0], |
1452 | 1024 /* factor bit size */); |
1453 | |
1454 | rr1->top = 16; |
1455 | rr1->neg = 0; |
1456 | bn_correct_top(rr1); |
1457 | bn_check_top(rr1); |
1458 | |
1459 | rr2->top = 16; |
1460 | rr2->neg = 0; |
1461 | bn_correct_top(rr2); |
1462 | bn_check_top(rr2); |
1463 | |
1464 | goto err; |
1465 | } |
1466 | #endif |
1467 | |
1468 | /* rr1 = a1^p1 mod m1 */ |
1469 | ret = BN_mod_exp_mont_consttime(rr1, a1, p1, m1, ctx, in_mont1); |
1470 | /* rr2 = a2^p2 mod m2 */ |
1471 | ret &= BN_mod_exp_mont_consttime(rr2, a2, p2, m2, ctx, in_mont2); |
1472 | |
1473 | #ifdef RSAZ_ENABLED |
1474 | err: |
1475 | if (in_mont2 == NULL((void*)0)) |
1476 | BN_MONT_CTX_free(mont2); |
1477 | if (in_mont1 == NULL((void*)0)) |
1478 | BN_MONT_CTX_free(mont1); |
1479 | #endif |
1480 | |
1481 | return ret; |
1482 | } |