xref: /freebsd/crypto/libecc/src/examples/basic/nn_miller_rabin.c (revision f0865ec9906d5a18fa2a3b61381f22ce16e606ad)
1 /*
2  *  Copyright (C) 2017 - This file is part of libecc project
3  *
4  *  Authors:
5  *      Ryad BENADJILA <ryadbenadjila@gmail.com>
6  *      Arnaud EBALARD <arnaud.ebalard@ssi.gouv.fr>
7  *      Jean-Pierre FLORI <jean-pierre.flori@ssi.gouv.fr>
8  *
9  *  Contributors:
10  *      Nicolas VIVET <nicolas.vivet@ssi.gouv.fr>
11  *      Karim KHALFALLAH <karim.khalfallah@ssi.gouv.fr>
12  *
13  *  This software is licensed under a dual BSD and GPL v2 license.
14  *  See LICENSE file at the root folder of the project.
15  */
16 /* We include the NN layer API header */
17 #include <libecc/libarith.h>
18 
19 ATTRIBUTE_WARN_UNUSED_RET int miller_rabin(nn_src_t n, const unsigned int t, int *res);
20 
21 /* Miller-Rabin primality test.
22  * See "Handbook of Applied Cryptography", alorithm 4.24:
23  *
24  *   Algorithm: Miller-Rabin probabilistic primality test
25  *   MILLER-RABIN(n,t)
26  *   INPUT: an odd integer n ≥ 3 and security parameter t ≥ 1.
27  *   OUTPUT: an answer “prime” or “composite” to the question: “Is n prime?”
28  *     1. Write n − 1 = 2**s x r such that r is odd.
29  *     2. For i from 1 to t do the following:
30  *       2.1 Choose a random integer a, 2 ≤ a ≤ n − 2.
31  *       2.2 Compute y = a**r mod n using Algorithm 2.143.
32  *       2.3 If y != 1 and y != n − 1 then do the following:
33  *         j←1.
34  *         While j ≤ s − 1 and y != n − 1 do the following:
35  *           Compute y←y2 mod n.
36  *           If y = 1 then return(“composite”).
37  *           j←j + 1.
38  *           If y != n − 1 then return (“composite”).
39  *     3. Return(“maybe prime”).
40  *
41  * The Miller-Rabin test can give false positives when
42  * answering "maybe prime", but is always right when answering
43  * "composite".
44  */
miller_rabin(nn_src_t n,const unsigned int t,int * res)45 int miller_rabin(nn_src_t n, const unsigned int t, int *res)
46 {
47 	int ret, iszero, cmp, isodd, cmp1, cmp2;
48 	unsigned int i;
49 	bitcnt_t k;
50 	/* Temporary NN variables */
51 	nn s, q, r, d, a, y, j, one, two, tmp;
52 	s.magic = q.magic = r.magic = d.magic = a.magic = y.magic = j.magic = WORD(0);
53 	one.magic = two.magic = tmp.magic = WORD(0);
54 
55 	ret = nn_check_initialized(n); EG(ret, err);
56 	MUST_HAVE((res != NULL), ret, err);
57 	(*res) = 0;
58 
59 	/* Initialize our local NN variables */
60 	ret = nn_init(&s, 0); EG(ret, err);
61 	ret = nn_init(&q, 0); EG(ret, err);
62 	ret = nn_init(&r, 0); EG(ret, err);
63 	ret = nn_init(&d, 0); EG(ret, err);
64 	ret = nn_init(&a, 0); EG(ret, err);
65 	ret = nn_init(&y, 0); EG(ret, err);
66 	ret = nn_init(&j, 0); EG(ret, err);
67 	ret = nn_init(&one, 0); EG(ret, err);
68 	ret = nn_init(&two, 0); EG(ret, err);
69 	ret = nn_init(&tmp, 0); EG(ret, err);
70 
71 	/* Security parameter t must be >= 1 */
72 	MUST_HAVE((t >= 1), ret, err);
73 
74 	/* one = 1 */
75 	ret = nn_one(&one); EG(ret, err);
76 	/* two = 2 */
77 	ret = nn_set_word_value(&two, WORD(2)); EG(ret, err);
78 
79 	/* If n = 0, this is not a prime */
80 	ret = nn_iszero(n, &iszero); EG(ret, err);
81 	if (iszero) {
82 		ret = 0;
83 		(*res) = 0;
84 		goto err;
85 	}
86 	/* If n = 1, this is not a prime */
87 	ret = nn_cmp(n, &one, &cmp); EG(ret, err);
88 	if (cmp == 0) {
89 		ret = 0;
90 		(*res) = 0;
91 		goto err;
92 	}
93 	/* If n = 2, this is a prime number */
94 	ret = nn_cmp(n, &two, &cmp); EG(ret, err);
95 	if (cmp == 0) {
96 		ret = 0;
97 		(*res) = 1;
98 		goto err;
99 	}
100 	/* If n = 3, this is a prime number */
101 	ret = nn_copy(&tmp, n); EG(ret, err);
102 	ret = nn_dec(&tmp, &tmp); EG(ret, err);
103 	ret = nn_cmp(&tmp, &two, &cmp); EG(ret, err);
104 	if (cmp == 0) {
105 		ret = 0;
106 		(*res) = 1;
107 		goto err;
108 	}
109 
110 	/* If n >= 4 is even, this is not a prime */
111 	ret = nn_isodd(n, &isodd); EG(ret, err);
112 	if (!isodd) {
113 		ret = 0;
114 		(*res) = 0;
115 		goto err;
116 	}
117 
118 	/* n − 1 = 2^s x r, repeatedly try to divide n-1 by 2 */
119 	/* s = 0 and r = n-1 */
120 	ret = nn_zero(&s); EG(ret, err);
121 	ret = nn_copy(&r, n); EG(ret, err);
122 	ret = nn_dec(&r, &r); EG(ret, err);
123 	while (1) {
124 		ret = nn_divrem(&q, &d, &r, &two); EG(ret, err);
125 		ret = nn_inc(&s, &s); EG(ret, err);
126 		ret = nn_copy(&r, &q); EG(ret, err);
127 		/* If r is odd, we have finished our division */
128 		ret = nn_isodd(&r, &isodd); EG(ret, err);
129 		if (isodd) {
130 			break;
131 		}
132 	}
133 	/* 2. For i from 1 to t do the following: */
134 	for (i = 1; i <= t; i++) {
135 		bitcnt_t blen;
136 		/* 2.1 Choose a random integer a, 2 ≤ a ≤ n − 2 */
137 		ret = nn_copy(&tmp, n); EG(ret, err);
138 		ret = nn_dec(&tmp, &tmp); EG(ret, err);
139 		ret = nn_zero(&a); EG(ret, err);
140 		ret = nn_cmp(&a, &two, &cmp); EG(ret, err);
141 		while (cmp < 0) {
142 			ret = nn_get_random_mod(&a, &tmp); EG(ret, err);
143 			ret = nn_cmp(&a, &two, &cmp); EG(ret, err);
144 		}
145 		/* A very loose (and NOT robust) implementation of
146 		 * modular exponentiation with square and multiply
147 		 * to compute y = a**r (n)
148 		 * WARNING: NOT to be used in production code!
149 		 */
150 		ret = nn_one(&y); EG(ret, err);
151 		ret = nn_bitlen(&r, &blen); EG(ret, err);
152 		for (k = 0; k < blen; k++) {
153 			u8 bit;
154 			ret = nn_getbit(&r, k, &bit); EG(ret, err);
155 			if (bit) {
156 				/* Warning: the multiplication is not modular, we
157 				 * have to take care of our size here!
158 				 */
159 				MUST_HAVE((NN_MAX_BIT_LEN >=
160 					  (WORD_BITS * (y.wlen + a.wlen))), ret, err);
161 				ret = nn_mul(&y, &y, &a); EG(ret, err);
162 				ret = nn_mod(&y, &y, n); EG(ret, err);
163 			}
164 			MUST_HAVE((NN_MAX_BIT_LEN >= (2 * WORD_BITS * a.wlen)), ret, err);
165 			ret = nn_sqr(&a, &a); EG(ret, err);
166 			ret = nn_mod(&a, &a, n); EG(ret, err);
167 		}
168 		/* 2.3 If y != 1 and y != n − 1 then do the following
169 		 * Note: tmp still contains n - 1 here.
170 		 */
171 		ret = nn_cmp(&y, &one, &cmp1); EG(ret, err);
172 		ret = nn_cmp(&y, &tmp, &cmp2); EG(ret, err);
173 		if ((cmp1 != 0) && (cmp2 != 0)) {
174 			/* j←1. */
175 			ret = nn_one(&j); EG(ret, err);
176 			/*  While j ≤ s − 1 and y != n − 1 do the following: */
177 			ret = nn_cmp(&j, &s, &cmp1); EG(ret, err);
178 			ret = nn_cmp(&y, &tmp, &cmp2); EG(ret, err);
179 			while ((cmp1 < 0) && (cmp2 != 0)) {
180 				/* Compute y←y2 mod n. */
181 				MUST_HAVE((NN_MAX_BIT_LEN >=
182 					  (2 * WORD_BITS * y.wlen)), ret, err);
183 				ret = nn_sqr(&y, &y); EG(ret, err);
184 				ret = nn_mod(&y, &y, n); EG(ret, err);
185 				/* If y = 1 then return(“composite”). */
186 				ret = nn_cmp(&y, &one, &cmp); EG(ret, err);
187 				if (cmp == 0) {
188 					ret = 0;
189 					(*res) = 0;
190 					goto err;
191 				}
192 				/* j←j + 1. */
193 				ret = nn_inc(&j, &j); EG(ret, err);
194 				ret = nn_cmp(&j, &s, &cmp1); EG(ret, err);
195 				ret = nn_cmp(&y, &tmp, &cmp2); EG(ret, err);
196 			}
197 			/* If y != n − 1 then return (“composite”). */
198 			ret = nn_cmp(&y, &tmp, &cmp); EG(ret, err);
199 			if (cmp != 0) {
200 				ret = 0;
201 				(*res) = 0;
202 				goto err;
203 			}
204 		}
205 		/* 3. Return(“maybe prime”). */
206 		ret = 0;
207 		(*res) = 1;
208 	}
209 
210  err:
211 	nn_uninit(&s);
212 	nn_uninit(&q);
213 	nn_uninit(&r);
214 	nn_uninit(&d);
215 	nn_uninit(&a);
216 	nn_uninit(&y);
217 	nn_uninit(&j);
218 	nn_uninit(&one);
219 	nn_uninit(&two);
220 	nn_uninit(&tmp);
221 
222 	return ret;
223 }
224