1*2a598d0bSHerbert Xu /* mpi-mod.c - Modular reduction
2*2a598d0bSHerbert Xu * Copyright (C) 1998, 1999, 2001, 2002, 2003,
3*2a598d0bSHerbert Xu * 2007 Free Software Foundation, Inc.
4*2a598d0bSHerbert Xu *
5*2a598d0bSHerbert Xu * This file is part of Libgcrypt.
6*2a598d0bSHerbert Xu */
7*2a598d0bSHerbert Xu
8*2a598d0bSHerbert Xu
9*2a598d0bSHerbert Xu #include "mpi-internal.h"
10*2a598d0bSHerbert Xu #include "longlong.h"
11*2a598d0bSHerbert Xu
12*2a598d0bSHerbert Xu /* Context used with Barrett reduction. */
13*2a598d0bSHerbert Xu struct barrett_ctx_s {
14*2a598d0bSHerbert Xu MPI m; /* The modulus - may not be modified. */
15*2a598d0bSHerbert Xu int m_copied; /* If true, M needs to be released. */
16*2a598d0bSHerbert Xu int k;
17*2a598d0bSHerbert Xu MPI y;
18*2a598d0bSHerbert Xu MPI r1; /* Helper MPI. */
19*2a598d0bSHerbert Xu MPI r2; /* Helper MPI. */
20*2a598d0bSHerbert Xu MPI r3; /* Helper MPI allocated on demand. */
21*2a598d0bSHerbert Xu };
22*2a598d0bSHerbert Xu
23*2a598d0bSHerbert Xu
24*2a598d0bSHerbert Xu
mpi_mod(MPI rem,MPI dividend,MPI divisor)25*2a598d0bSHerbert Xu void mpi_mod(MPI rem, MPI dividend, MPI divisor)
26*2a598d0bSHerbert Xu {
27*2a598d0bSHerbert Xu mpi_fdiv_r(rem, dividend, divisor);
28*2a598d0bSHerbert Xu }
29*2a598d0bSHerbert Xu
30*2a598d0bSHerbert Xu /* This function returns a new context for Barrett based operations on
31*2a598d0bSHerbert Xu * the modulus M. This context needs to be released using
32*2a598d0bSHerbert Xu * _gcry_mpi_barrett_free. If COPY is true M will be transferred to
33*2a598d0bSHerbert Xu * the context and the user may change M. If COPY is false, M may not
34*2a598d0bSHerbert Xu * be changed until gcry_mpi_barrett_free has been called.
35*2a598d0bSHerbert Xu */
mpi_barrett_init(MPI m,int copy)36*2a598d0bSHerbert Xu mpi_barrett_t mpi_barrett_init(MPI m, int copy)
37*2a598d0bSHerbert Xu {
38*2a598d0bSHerbert Xu mpi_barrett_t ctx;
39*2a598d0bSHerbert Xu MPI tmp;
40*2a598d0bSHerbert Xu
41*2a598d0bSHerbert Xu mpi_normalize(m);
42*2a598d0bSHerbert Xu ctx = kcalloc(1, sizeof(*ctx), GFP_KERNEL);
43*2a598d0bSHerbert Xu if (!ctx)
44*2a598d0bSHerbert Xu return NULL;
45*2a598d0bSHerbert Xu
46*2a598d0bSHerbert Xu if (copy) {
47*2a598d0bSHerbert Xu ctx->m = mpi_copy(m);
48*2a598d0bSHerbert Xu ctx->m_copied = 1;
49*2a598d0bSHerbert Xu } else
50*2a598d0bSHerbert Xu ctx->m = m;
51*2a598d0bSHerbert Xu
52*2a598d0bSHerbert Xu ctx->k = mpi_get_nlimbs(m);
53*2a598d0bSHerbert Xu tmp = mpi_alloc(ctx->k + 1);
54*2a598d0bSHerbert Xu
55*2a598d0bSHerbert Xu /* Barrett precalculation: y = floor(b^(2k) / m). */
56*2a598d0bSHerbert Xu mpi_set_ui(tmp, 1);
57*2a598d0bSHerbert Xu mpi_lshift_limbs(tmp, 2 * ctx->k);
58*2a598d0bSHerbert Xu mpi_fdiv_q(tmp, tmp, m);
59*2a598d0bSHerbert Xu
60*2a598d0bSHerbert Xu ctx->y = tmp;
61*2a598d0bSHerbert Xu ctx->r1 = mpi_alloc(2 * ctx->k + 1);
62*2a598d0bSHerbert Xu ctx->r2 = mpi_alloc(2 * ctx->k + 1);
63*2a598d0bSHerbert Xu
64*2a598d0bSHerbert Xu return ctx;
65*2a598d0bSHerbert Xu }
66*2a598d0bSHerbert Xu
mpi_barrett_free(mpi_barrett_t ctx)67*2a598d0bSHerbert Xu void mpi_barrett_free(mpi_barrett_t ctx)
68*2a598d0bSHerbert Xu {
69*2a598d0bSHerbert Xu if (ctx) {
70*2a598d0bSHerbert Xu mpi_free(ctx->y);
71*2a598d0bSHerbert Xu mpi_free(ctx->r1);
72*2a598d0bSHerbert Xu mpi_free(ctx->r2);
73*2a598d0bSHerbert Xu if (ctx->r3)
74*2a598d0bSHerbert Xu mpi_free(ctx->r3);
75*2a598d0bSHerbert Xu if (ctx->m_copied)
76*2a598d0bSHerbert Xu mpi_free(ctx->m);
77*2a598d0bSHerbert Xu kfree(ctx);
78*2a598d0bSHerbert Xu }
79*2a598d0bSHerbert Xu }
80*2a598d0bSHerbert Xu
81*2a598d0bSHerbert Xu
82*2a598d0bSHerbert Xu /* R = X mod M
83*2a598d0bSHerbert Xu *
84*2a598d0bSHerbert Xu * Using Barrett reduction. Before using this function
85*2a598d0bSHerbert Xu * _gcry_mpi_barrett_init must have been called to do the
86*2a598d0bSHerbert Xu * precalculations. CTX is the context created by this precalculation
87*2a598d0bSHerbert Xu * and also conveys M. If the Barret reduction could no be done a
88*2a598d0bSHerbert Xu * straightforward reduction method is used.
89*2a598d0bSHerbert Xu *
90*2a598d0bSHerbert Xu * We assume that these conditions are met:
91*2a598d0bSHerbert Xu * Input: x =(x_2k-1 ...x_0)_b
92*2a598d0bSHerbert Xu * m =(m_k-1 ....m_0)_b with m_k-1 != 0
93*2a598d0bSHerbert Xu * Output: r = x mod m
94*2a598d0bSHerbert Xu */
mpi_mod_barrett(MPI r,MPI x,mpi_barrett_t ctx)95*2a598d0bSHerbert Xu void mpi_mod_barrett(MPI r, MPI x, mpi_barrett_t ctx)
96*2a598d0bSHerbert Xu {
97*2a598d0bSHerbert Xu MPI m = ctx->m;
98*2a598d0bSHerbert Xu int k = ctx->k;
99*2a598d0bSHerbert Xu MPI y = ctx->y;
100*2a598d0bSHerbert Xu MPI r1 = ctx->r1;
101*2a598d0bSHerbert Xu MPI r2 = ctx->r2;
102*2a598d0bSHerbert Xu int sign;
103*2a598d0bSHerbert Xu
104*2a598d0bSHerbert Xu mpi_normalize(x);
105*2a598d0bSHerbert Xu if (mpi_get_nlimbs(x) > 2*k) {
106*2a598d0bSHerbert Xu mpi_mod(r, x, m);
107*2a598d0bSHerbert Xu return;
108*2a598d0bSHerbert Xu }
109*2a598d0bSHerbert Xu
110*2a598d0bSHerbert Xu sign = x->sign;
111*2a598d0bSHerbert Xu x->sign = 0;
112*2a598d0bSHerbert Xu
113*2a598d0bSHerbert Xu /* 1. q1 = floor( x / b^k-1)
114*2a598d0bSHerbert Xu * q2 = q1 * y
115*2a598d0bSHerbert Xu * q3 = floor( q2 / b^k+1 )
116*2a598d0bSHerbert Xu * Actually, we don't need qx, we can work direct on r2
117*2a598d0bSHerbert Xu */
118*2a598d0bSHerbert Xu mpi_set(r2, x);
119*2a598d0bSHerbert Xu mpi_rshift_limbs(r2, k-1);
120*2a598d0bSHerbert Xu mpi_mul(r2, r2, y);
121*2a598d0bSHerbert Xu mpi_rshift_limbs(r2, k+1);
122*2a598d0bSHerbert Xu
123*2a598d0bSHerbert Xu /* 2. r1 = x mod b^k+1
124*2a598d0bSHerbert Xu * r2 = q3 * m mod b^k+1
125*2a598d0bSHerbert Xu * r = r1 - r2
126*2a598d0bSHerbert Xu * 3. if r < 0 then r = r + b^k+1
127*2a598d0bSHerbert Xu */
128*2a598d0bSHerbert Xu mpi_set(r1, x);
129*2a598d0bSHerbert Xu if (r1->nlimbs > k+1) /* Quick modulo operation. */
130*2a598d0bSHerbert Xu r1->nlimbs = k+1;
131*2a598d0bSHerbert Xu mpi_mul(r2, r2, m);
132*2a598d0bSHerbert Xu if (r2->nlimbs > k+1) /* Quick modulo operation. */
133*2a598d0bSHerbert Xu r2->nlimbs = k+1;
134*2a598d0bSHerbert Xu mpi_sub(r, r1, r2);
135*2a598d0bSHerbert Xu
136*2a598d0bSHerbert Xu if (mpi_has_sign(r)) {
137*2a598d0bSHerbert Xu if (!ctx->r3) {
138*2a598d0bSHerbert Xu ctx->r3 = mpi_alloc(k + 2);
139*2a598d0bSHerbert Xu mpi_set_ui(ctx->r3, 1);
140*2a598d0bSHerbert Xu mpi_lshift_limbs(ctx->r3, k + 1);
141*2a598d0bSHerbert Xu }
142*2a598d0bSHerbert Xu mpi_add(r, r, ctx->r3);
143*2a598d0bSHerbert Xu }
144*2a598d0bSHerbert Xu
145*2a598d0bSHerbert Xu /* 4. while r >= m do r = r - m */
146*2a598d0bSHerbert Xu while (mpi_cmp(r, m) >= 0)
147*2a598d0bSHerbert Xu mpi_sub(r, r, m);
148*2a598d0bSHerbert Xu
149*2a598d0bSHerbert Xu x->sign = sign;
150*2a598d0bSHerbert Xu }
151*2a598d0bSHerbert Xu
152*2a598d0bSHerbert Xu
mpi_mul_barrett(MPI w,MPI u,MPI v,mpi_barrett_t ctx)153*2a598d0bSHerbert Xu void mpi_mul_barrett(MPI w, MPI u, MPI v, mpi_barrett_t ctx)
154*2a598d0bSHerbert Xu {
155*2a598d0bSHerbert Xu mpi_mul(w, u, v);
156*2a598d0bSHerbert Xu mpi_mod_barrett(w, w, ctx);
157*2a598d0bSHerbert Xu }
158