1 /*
2 * Copyright 1995-2024 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 /*
11 * Details about Montgomery multiplication algorithms can be found at
12 * http://security.ece.orst.edu/publications.html, e.g.
13 * http://security.ece.orst.edu/koc/papers/j37acmon.pdf and
14 * sections 3.8 and 4.2 in http://security.ece.orst.edu/koc/papers/r01rsasw.pdf
15 */
16
17 #include "internal/cryptlib.h"
18 #include "bn_local.h"
19
20 #define MONT_WORD /* use the faster word-based algorithm */
21
22 #ifdef MONT_WORD
23 static int bn_from_montgomery_word(BIGNUM *ret, BIGNUM *r, BN_MONT_CTX *mont);
24 #endif
25
BN_mod_mul_montgomery(BIGNUM * r,const BIGNUM * a,const BIGNUM * b,BN_MONT_CTX * mont,BN_CTX * ctx)26 int BN_mod_mul_montgomery(BIGNUM *r, const BIGNUM *a, const BIGNUM *b,
27 BN_MONT_CTX *mont, BN_CTX *ctx)
28 {
29 int ret = bn_mul_mont_fixed_top(r, a, b, mont, ctx);
30
31 bn_correct_top(r);
32 bn_check_top(r);
33
34 return ret;
35 }
36
bn_mul_mont_fixed_top(BIGNUM * r,const BIGNUM * a,const BIGNUM * b,BN_MONT_CTX * mont,BN_CTX * ctx)37 int bn_mul_mont_fixed_top(BIGNUM *r, const BIGNUM *a, const BIGNUM *b,
38 BN_MONT_CTX *mont, BN_CTX *ctx)
39 {
40 BIGNUM *tmp;
41 int ret = 0;
42 int num = mont->N.top;
43
44 #if defined(OPENSSL_BN_ASM_MONT) && defined(MONT_WORD)
45 if (num > 1 && num <= BN_SOFT_LIMIT && a->top == num && b->top == num) {
46 if (bn_wexpand(r, num) == NULL)
47 return 0;
48 if (bn_mul_mont(r->d, a->d, b->d, mont->N.d, mont->n0, num)) {
49 r->neg = a->neg ^ b->neg;
50 r->top = num;
51 r->flags |= BN_FLG_FIXED_TOP;
52 return 1;
53 }
54 }
55 #endif
56
57 if ((a->top + b->top) > 2 * num)
58 return 0;
59
60 BN_CTX_start(ctx);
61 tmp = BN_CTX_get(ctx);
62 if (tmp == NULL)
63 goto err;
64
65 bn_check_top(tmp);
66 if (a == b) {
67 if (!bn_sqr_fixed_top(tmp, a, ctx))
68 goto err;
69 } else {
70 if (!bn_mul_fixed_top(tmp, a, b, ctx))
71 goto err;
72 }
73 /* reduce from aRR to aR */
74 #ifdef MONT_WORD
75 if (!bn_from_montgomery_word(r, tmp, mont))
76 goto err;
77 #else
78 if (!BN_from_montgomery(r, tmp, mont, ctx))
79 goto err;
80 #endif
81 ret = 1;
82 err:
83 BN_CTX_end(ctx);
84 return ret;
85 }
86
87 #ifdef MONT_WORD
bn_from_montgomery_word(BIGNUM * ret,BIGNUM * r,BN_MONT_CTX * mont)88 static int bn_from_montgomery_word(BIGNUM *ret, BIGNUM *r, BN_MONT_CTX *mont)
89 {
90 BIGNUM *n;
91 BN_ULONG *ap, *np, *rp, n0, v, carry;
92 int nl, max, i;
93 unsigned int rtop;
94
95 n = &(mont->N);
96 nl = n->top;
97 if (nl == 0) {
98 ret->top = 0;
99 return 1;
100 }
101
102 max = (2 * nl); /* carry is stored separately */
103 if (bn_wexpand(r, max) == NULL)
104 return 0;
105
106 r->neg ^= n->neg;
107 np = n->d;
108 rp = r->d;
109
110 /* clear the top words of T */
111 for (rtop = r->top, i = 0; i < max; i++) {
112 v = (BN_ULONG)0 - ((i - rtop) >> (8 * sizeof(rtop) - 1));
113 rp[i] &= v;
114 }
115
116 r->top = max;
117 r->flags |= BN_FLG_FIXED_TOP;
118 n0 = mont->n0[0];
119
120 /*
121 * Add multiples of |n| to |r| until R = 2^(nl * BN_BITS2) divides it. On
122 * input, we had |r| < |n| * R, so now |r| < 2 * |n| * R. Note that |r|
123 * includes |carry| which is stored separately.
124 */
125 for (carry = 0, i = 0; i < nl; i++, rp++) {
126 v = bn_mul_add_words(rp, np, nl, (rp[0] * n0) & BN_MASK2);
127 v = (v + carry + rp[nl]) & BN_MASK2;
128 carry |= (v != rp[nl]);
129 carry &= (v <= rp[nl]);
130 rp[nl] = v;
131 }
132
133 if (bn_wexpand(ret, nl) == NULL)
134 return 0;
135 ret->top = nl;
136 ret->flags |= BN_FLG_FIXED_TOP;
137 ret->neg = r->neg;
138
139 rp = ret->d;
140
141 /*
142 * Shift |nl| words to divide by R. We have |ap| < 2 * |n|. Note that |ap|
143 * includes |carry| which is stored separately.
144 */
145 ap = &(r->d[nl]);
146
147 carry -= bn_sub_words(rp, ap, np, nl);
148 /*
149 * |carry| is -1 if |ap| - |np| underflowed or zero if it did not. Note
150 * |carry| cannot be 1. That would imply the subtraction did not fit in
151 * |nl| words, and we know at most one subtraction is needed.
152 */
153 for (i = 0; i < nl; i++) {
154 rp[i] = (carry & ap[i]) | (~carry & rp[i]);
155 ap[i] = 0;
156 }
157
158 return 1;
159 }
160 #endif /* MONT_WORD */
161
BN_from_montgomery(BIGNUM * ret,const BIGNUM * a,BN_MONT_CTX * mont,BN_CTX * ctx)162 int BN_from_montgomery(BIGNUM *ret, const BIGNUM *a, BN_MONT_CTX *mont,
163 BN_CTX *ctx)
164 {
165 int retn;
166
167 retn = bn_from_mont_fixed_top(ret, a, mont, ctx);
168 bn_correct_top(ret);
169 bn_check_top(ret);
170
171 return retn;
172 }
173
bn_from_mont_fixed_top(BIGNUM * ret,const BIGNUM * a,BN_MONT_CTX * mont,BN_CTX * ctx)174 int bn_from_mont_fixed_top(BIGNUM *ret, const BIGNUM *a, BN_MONT_CTX *mont,
175 BN_CTX *ctx)
176 {
177 int retn = 0;
178 #ifdef MONT_WORD
179 BIGNUM *t;
180
181 BN_CTX_start(ctx);
182 if ((t = BN_CTX_get(ctx)) && BN_copy(t, a)) {
183 retn = bn_from_montgomery_word(ret, t, mont);
184 }
185 BN_CTX_end(ctx);
186 #else /* !MONT_WORD */
187 BIGNUM *t1, *t2;
188
189 BN_CTX_start(ctx);
190 t1 = BN_CTX_get(ctx);
191 t2 = BN_CTX_get(ctx);
192 if (t2 == NULL)
193 goto err;
194
195 if (!BN_copy(t1, a))
196 goto err;
197 BN_mask_bits(t1, mont->ri);
198
199 if (!BN_mul(t2, t1, &mont->Ni, ctx))
200 goto err;
201 BN_mask_bits(t2, mont->ri);
202
203 if (!BN_mul(t1, t2, &mont->N, ctx))
204 goto err;
205 if (!BN_add(t2, a, t1))
206 goto err;
207 if (!BN_rshift(ret, t2, mont->ri))
208 goto err;
209
210 if (BN_ucmp(ret, &(mont->N)) >= 0) {
211 if (!BN_usub(ret, ret, &(mont->N)))
212 goto err;
213 }
214 retn = 1;
215 bn_check_top(ret);
216 err:
217 BN_CTX_end(ctx);
218 #endif /* MONT_WORD */
219 return retn;
220 }
221
bn_to_mont_fixed_top(BIGNUM * r,const BIGNUM * a,BN_MONT_CTX * mont,BN_CTX * ctx)222 int bn_to_mont_fixed_top(BIGNUM *r, const BIGNUM *a, BN_MONT_CTX *mont,
223 BN_CTX *ctx)
224 {
225 return bn_mul_mont_fixed_top(r, a, &(mont->RR), mont, ctx);
226 }
227
BN_MONT_CTX_new(void)228 BN_MONT_CTX *BN_MONT_CTX_new(void)
229 {
230 BN_MONT_CTX *ret;
231
232 if ((ret = OPENSSL_malloc(sizeof(*ret))) == NULL)
233 return NULL;
234
235 BN_MONT_CTX_init(ret);
236 ret->flags = BN_FLG_MALLOCED;
237 return ret;
238 }
239
BN_MONT_CTX_init(BN_MONT_CTX * ctx)240 void BN_MONT_CTX_init(BN_MONT_CTX *ctx)
241 {
242 ctx->ri = 0;
243 bn_init(&ctx->RR);
244 bn_init(&ctx->N);
245 bn_init(&ctx->Ni);
246 ctx->n0[0] = ctx->n0[1] = 0;
247 ctx->flags = 0;
248 }
249
BN_MONT_CTX_free(BN_MONT_CTX * mont)250 void BN_MONT_CTX_free(BN_MONT_CTX *mont)
251 {
252 if (mont == NULL)
253 return;
254 BN_clear_free(&mont->RR);
255 BN_clear_free(&mont->N);
256 BN_clear_free(&mont->Ni);
257 if (mont->flags & BN_FLG_MALLOCED)
258 OPENSSL_free(mont);
259 }
260
BN_MONT_CTX_set(BN_MONT_CTX * mont,const BIGNUM * mod,BN_CTX * ctx)261 int BN_MONT_CTX_set(BN_MONT_CTX *mont, const BIGNUM *mod, BN_CTX *ctx)
262 {
263 int i, ret = 0;
264 BIGNUM *Ri, *R;
265
266 if (BN_is_zero(mod))
267 return 0;
268
269 BN_CTX_start(ctx);
270 if ((Ri = BN_CTX_get(ctx)) == NULL)
271 goto err;
272 R = &(mont->RR); /* grab RR as a temp */
273 if (!BN_copy(&(mont->N), mod))
274 goto err; /* Set N */
275 if (BN_get_flags(mod, BN_FLG_CONSTTIME) != 0)
276 BN_set_flags(&(mont->N), BN_FLG_CONSTTIME);
277 mont->N.neg = 0;
278
279 #ifdef MONT_WORD
280 {
281 BIGNUM tmod;
282 BN_ULONG buf[2];
283
284 bn_init(&tmod);
285 tmod.d = buf;
286 tmod.dmax = 2;
287 tmod.neg = 0;
288
289 if (BN_get_flags(mod, BN_FLG_CONSTTIME) != 0)
290 BN_set_flags(&tmod, BN_FLG_CONSTTIME);
291
292 mont->ri = (BN_num_bits(mod) + (BN_BITS2 - 1)) / BN_BITS2 * BN_BITS2;
293
294 # if defined(OPENSSL_BN_ASM_MONT) && (BN_BITS2<=32)
295 /*
296 * Only certain BN_BITS2<=32 platforms actually make use of n0[1],
297 * and we could use the #else case (with a shorter R value) for the
298 * others. However, currently only the assembler files do know which
299 * is which.
300 */
301
302 BN_zero(R);
303 if (!(BN_set_bit(R, 2 * BN_BITS2)))
304 goto err;
305
306 tmod.top = 0;
307 if ((buf[0] = mod->d[0]))
308 tmod.top = 1;
309 if ((buf[1] = mod->top > 1 ? mod->d[1] : 0))
310 tmod.top = 2;
311
312 if (BN_is_one(&tmod))
313 BN_zero(Ri);
314 else if ((BN_mod_inverse(Ri, R, &tmod, ctx)) == NULL)
315 goto err;
316 if (!BN_lshift(Ri, Ri, 2 * BN_BITS2))
317 goto err; /* R*Ri */
318 if (!BN_is_zero(Ri)) {
319 if (!BN_sub_word(Ri, 1))
320 goto err;
321 } else { /* if N mod word size == 1 */
322
323 if (bn_expand(Ri, (int)sizeof(BN_ULONG) * 2) == NULL)
324 goto err;
325 /* Ri-- (mod double word size) */
326 Ri->neg = 0;
327 Ri->d[0] = BN_MASK2;
328 Ri->d[1] = BN_MASK2;
329 Ri->top = 2;
330 }
331 if (!BN_div(Ri, NULL, Ri, &tmod, ctx))
332 goto err;
333 /*
334 * Ni = (R*Ri-1)/N, keep only couple of least significant words:
335 */
336 mont->n0[0] = (Ri->top > 0) ? Ri->d[0] : 0;
337 mont->n0[1] = (Ri->top > 1) ? Ri->d[1] : 0;
338 # else
339 BN_zero(R);
340 if (!(BN_set_bit(R, BN_BITS2)))
341 goto err; /* R */
342
343 buf[0] = mod->d[0]; /* tmod = N mod word size */
344 buf[1] = 0;
345 tmod.top = buf[0] != 0 ? 1 : 0;
346 /* Ri = R^-1 mod N */
347 if (BN_is_one(&tmod))
348 BN_zero(Ri);
349 else if ((BN_mod_inverse(Ri, R, &tmod, ctx)) == NULL)
350 goto err;
351 if (!BN_lshift(Ri, Ri, BN_BITS2))
352 goto err; /* R*Ri */
353 if (!BN_is_zero(Ri)) {
354 if (!BN_sub_word(Ri, 1))
355 goto err;
356 } else { /* if N mod word size == 1 */
357
358 if (!BN_set_word(Ri, BN_MASK2))
359 goto err; /* Ri-- (mod word size) */
360 }
361 if (!BN_div(Ri, NULL, Ri, &tmod, ctx))
362 goto err;
363 /*
364 * Ni = (R*Ri-1)/N, keep only least significant word:
365 */
366 mont->n0[0] = (Ri->top > 0) ? Ri->d[0] : 0;
367 mont->n0[1] = 0;
368 # endif
369 }
370 #else /* !MONT_WORD */
371 { /* bignum version */
372 mont->ri = BN_num_bits(&mont->N);
373 BN_zero(R);
374 if (!BN_set_bit(R, mont->ri))
375 goto err; /* R = 2^ri */
376 /* Ri = R^-1 mod N */
377 if ((BN_mod_inverse(Ri, R, &mont->N, ctx)) == NULL)
378 goto err;
379 if (!BN_lshift(Ri, Ri, mont->ri))
380 goto err; /* R*Ri */
381 if (!BN_sub_word(Ri, 1))
382 goto err;
383 /*
384 * Ni = (R*Ri-1) / N
385 */
386 if (!BN_div(&(mont->Ni), NULL, Ri, &mont->N, ctx))
387 goto err;
388 }
389 #endif
390
391 /* setup RR for conversions */
392 BN_zero(&(mont->RR));
393 if (!BN_set_bit(&(mont->RR), mont->ri * 2))
394 goto err;
395 if (!BN_mod(&(mont->RR), &(mont->RR), &(mont->N), ctx))
396 goto err;
397
398 for (i = mont->RR.top, ret = mont->N.top; i < ret; i++)
399 mont->RR.d[i] = 0;
400 mont->RR.top = ret;
401 mont->RR.flags |= BN_FLG_FIXED_TOP;
402
403 ret = 1;
404 err:
405 BN_CTX_end(ctx);
406 return ret;
407 }
408
BN_MONT_CTX_copy(BN_MONT_CTX * to,BN_MONT_CTX * from)409 BN_MONT_CTX *BN_MONT_CTX_copy(BN_MONT_CTX *to, BN_MONT_CTX *from)
410 {
411 if (to == from)
412 return to;
413
414 if (!BN_copy(&(to->RR), &(from->RR)))
415 return NULL;
416 if (!BN_copy(&(to->N), &(from->N)))
417 return NULL;
418 if (!BN_copy(&(to->Ni), &(from->Ni)))
419 return NULL;
420 to->ri = from->ri;
421 to->n0[0] = from->n0[0];
422 to->n0[1] = from->n0[1];
423 return to;
424 }
425
BN_MONT_CTX_set_locked(BN_MONT_CTX ** pmont,CRYPTO_RWLOCK * lock,const BIGNUM * mod,BN_CTX * ctx)426 BN_MONT_CTX *BN_MONT_CTX_set_locked(BN_MONT_CTX **pmont, CRYPTO_RWLOCK *lock,
427 const BIGNUM *mod, BN_CTX *ctx)
428 {
429 BN_MONT_CTX *ret;
430
431 if (!CRYPTO_THREAD_read_lock(lock))
432 return NULL;
433 ret = *pmont;
434 CRYPTO_THREAD_unlock(lock);
435 if (ret)
436 return ret;
437
438 /*
439 * We don't want to serialize globally while doing our lazy-init math in
440 * BN_MONT_CTX_set. That punishes threads that are doing independent
441 * things. Instead, punish the case where more than one thread tries to
442 * lazy-init the same 'pmont', by having each do the lazy-init math work
443 * independently and only use the one from the thread that wins the race
444 * (the losers throw away the work they've done).
445 */
446 ret = BN_MONT_CTX_new();
447 if (ret == NULL)
448 return NULL;
449 if (!BN_MONT_CTX_set(ret, mod, ctx)) {
450 BN_MONT_CTX_free(ret);
451 return NULL;
452 }
453
454 /* The locked compare-and-set, after the local work is done. */
455 if (!CRYPTO_THREAD_write_lock(lock)) {
456 BN_MONT_CTX_free(ret);
457 return NULL;
458 }
459
460 if (*pmont) {
461 BN_MONT_CTX_free(ret);
462 ret = *pmont;
463 } else
464 *pmont = ret;
465 CRYPTO_THREAD_unlock(lock);
466 return ret;
467 }
468
ossl_bn_mont_ctx_set(BN_MONT_CTX * ctx,const BIGNUM * modulus,int ri,const unsigned char * rr,size_t rrlen,uint32_t nlo,uint32_t nhi)469 int ossl_bn_mont_ctx_set(BN_MONT_CTX *ctx, const BIGNUM *modulus, int ri, const unsigned char *rr,
470 size_t rrlen, uint32_t nlo, uint32_t nhi)
471 {
472 if (BN_copy(&ctx->N, modulus) == NULL)
473 return 0;
474 if (BN_bin2bn(rr, rrlen, &ctx->RR) == NULL)
475 return 0;
476 ctx->ri = ri;
477 #if (BN_BITS2 <= 32) && defined(OPENSSL_BN_ASM_MONT)
478 ctx->n0[0] = nlo;
479 ctx->n0[1] = nhi;
480 #elif BN_BITS2 <= 32
481 ctx->n0[0] = nlo;
482 ctx->n0[1] = 0;
483 #else
484 ctx->n0[0] = ((BN_ULONG)nhi << 32)| nlo;
485 ctx->n0[1] = 0;
486 #endif
487
488 return 1;
489 }
490
ossl_bn_mont_ctx_eq(const BN_MONT_CTX * m1,const BN_MONT_CTX * m2)491 int ossl_bn_mont_ctx_eq(const BN_MONT_CTX *m1, const BN_MONT_CTX *m2)
492 {
493 if (m1->ri != m2->ri)
494 return 0;
495 if (BN_cmp(&m1->RR, &m2->RR) != 0)
496 return 0;
497 if (m1->flags != m2->flags)
498 return 0;
499 #ifdef MONT_WORD
500 if (m1->n0[0] != m2->n0[0])
501 return 0;
502 if (m1->n0[1] != m2->n0[1])
503 return 0;
504 #else
505 if (BN_cmp(&m1->Ni, &m2->Ni) != 0)
506 return 0;
507 #endif
508 return 1;
509 }
510