Win32-OpenSSH/fe25519.c
dkulwin ddace27b97 VC2015 doesn't like it when you take a negative of on unsigned value
Negative of an unsigned value should just be the two's complement.  Add
code to change code with negative unsigned values to two's compliment
values if compiling under visual studio.
2015-11-04 19:34:12 -06:00

346 lines
8.2 KiB
C

/* $OpenBSD: fe25519.c,v 1.3 2013/12/09 11:03:45 markus Exp $ */
/*
* Public Domain, Authors: Daniel J. Bernstein, Niels Duif, Tanja Lange,
* Peter Schwabe, Bo-Yin Yang.
* Copied from supercop-20130419/crypto_sign/ed25519/ref/fe25519.c
*/
#include "includes.h"
#define WINDOWSIZE 1 /* Should be 1,2, or 4 */
#define WINDOWMASK ((1<<WINDOWSIZE)-1)
#include "fe25519.h"
static crypto_uint32 equal(crypto_uint32 a,crypto_uint32 b) /* 16-bit inputs */
{
crypto_uint32 x = a ^ b; /* 0: yes; 1..65535: no */
x -= 1; /* 4294967295: yes; 0..65534: no */
x >>= 31; /* 1: yes; 0: no */
return x;
}
static crypto_uint32 ge(crypto_uint32 a,crypto_uint32 b) /* 16-bit inputs */
{
unsigned int x = a;
x -= (unsigned int) b; /* 0..65535: yes; 4294901761..4294967295: no */
x >>= 31; /* 0: yes; 1: no */
x ^= 1; /* 1: yes; 0: no */
return x;
}
static crypto_uint32 times19(crypto_uint32 a)
{
return (a << 4) + (a << 1) + a;
}
static crypto_uint32 times38(crypto_uint32 a)
{
return (a << 5) + (a << 2) + (a << 1);
}
static void reduce_add_sub(fe25519 *r)
{
crypto_uint32 t;
int i,rep;
for(rep=0;rep<4;rep++)
{
t = r->v[31] >> 7;
r->v[31] &= 127;
t = times19(t);
r->v[0] += t;
for(i=0;i<31;i++)
{
t = r->v[i] >> 8;
r->v[i+1] += t;
r->v[i] &= 255;
}
}
}
static void reduce_mul(fe25519 *r)
{
crypto_uint32 t;
int i,rep;
for(rep=0;rep<2;rep++)
{
t = r->v[31] >> 7;
r->v[31] &= 127;
t = times19(t);
r->v[0] += t;
for(i=0;i<31;i++)
{
t = r->v[i] >> 8;
r->v[i+1] += t;
r->v[i] &= 255;
}
}
}
/* reduction modulo 2^255-19 */
void fe25519_freeze(fe25519 *r)
{
int i;
crypto_uint32 m = equal(r->v[31],127);
for(i=30;i>0;i--)
m &= equal(r->v[i],255);
m &= ge(r->v[0],237);
#ifndef _WIN32
m = -m;
#else
m = (~m + 1u);
#endif
r->v[31] -= m&127;
for(i=30;i>0;i--)
r->v[i] -= m&255;
r->v[0] -= m&237;
}
void fe25519_unpack(fe25519 *r, const unsigned char x[32])
{
int i;
for(i=0;i<32;i++) r->v[i] = x[i];
r->v[31] &= 127;
}
/* Assumes input x being reduced below 2^255 */
void fe25519_pack(unsigned char r[32], const fe25519 *x)
{
int i;
fe25519 y = *x;
fe25519_freeze(&y);
for(i=0;i<32;i++)
r[i] = y.v[i];
}
int fe25519_iszero(const fe25519 *x)
{
int i;
int r;
fe25519 t = *x;
fe25519_freeze(&t);
r = equal(t.v[0],0);
for(i=1;i<32;i++)
r &= equal(t.v[i],0);
return r;
}
int fe25519_iseq_vartime(const fe25519 *x, const fe25519 *y)
{
int i;
fe25519 t1 = *x;
fe25519 t2 = *y;
fe25519_freeze(&t1);
fe25519_freeze(&t2);
for(i=0;i<32;i++)
if(t1.v[i] != t2.v[i]) return 0;
return 1;
}
void fe25519_cmov(fe25519 *r, const fe25519 *x, unsigned char b)
{
int i;
crypto_uint32 mask = b;
#ifndef _WIN32
mask = -mask;
#else
mask = (~mask + 1u);
#endif
for(i=0;i<32;i++) r->v[i] ^= mask & (x->v[i] ^ r->v[i]);
}
unsigned char fe25519_getparity(const fe25519 *x)
{
fe25519 t = *x;
fe25519_freeze(&t);
return t.v[0] & 1;
}
void fe25519_setone(fe25519 *r)
{
int i;
r->v[0] = 1;
for(i=1;i<32;i++) r->v[i]=0;
}
void fe25519_setzero(fe25519 *r)
{
int i;
for(i=0;i<32;i++) r->v[i]=0;
}
void fe25519_neg(fe25519 *r, const fe25519 *x)
{
fe25519 t;
int i;
for(i=0;i<32;i++) t.v[i]=x->v[i];
fe25519_setzero(r);
fe25519_sub(r, r, &t);
}
void fe25519_add(fe25519 *r, const fe25519 *x, const fe25519 *y)
{
int i;
for(i=0;i<32;i++) r->v[i] = x->v[i] + y->v[i];
reduce_add_sub(r);
}
void fe25519_sub(fe25519 *r, const fe25519 *x, const fe25519 *y)
{
int i;
crypto_uint32 t[32];
t[0] = x->v[0] + 0x1da;
t[31] = x->v[31] + 0xfe;
for(i=1;i<31;i++) t[i] = x->v[i] + 0x1fe;
for(i=0;i<32;i++) r->v[i] = t[i] - y->v[i];
reduce_add_sub(r);
}
void fe25519_mul(fe25519 *r, const fe25519 *x, const fe25519 *y)
{
int i,j;
crypto_uint32 t[63];
for(i=0;i<63;i++)t[i] = 0;
for(i=0;i<32;i++)
for(j=0;j<32;j++)
t[i+j] += x->v[i] * y->v[j];
for(i=32;i<63;i++)
r->v[i-32] = t[i-32] + times38(t[i]);
r->v[31] = t[31]; /* result now in r[0]...r[31] */
reduce_mul(r);
}
void fe25519_square(fe25519 *r, const fe25519 *x)
{
fe25519_mul(r, x, x);
}
void fe25519_invert(fe25519 *r, const fe25519 *x)
{
fe25519 z2;
fe25519 z9;
fe25519 z11;
fe25519 z2_5_0;
fe25519 z2_10_0;
fe25519 z2_20_0;
fe25519 z2_50_0;
fe25519 z2_100_0;
fe25519 t0;
fe25519 t1;
int i;
/* 2 */ fe25519_square(&z2,x);
/* 4 */ fe25519_square(&t1,&z2);
/* 8 */ fe25519_square(&t0,&t1);
/* 9 */ fe25519_mul(&z9,&t0,x);
/* 11 */ fe25519_mul(&z11,&z9,&z2);
/* 22 */ fe25519_square(&t0,&z11);
/* 2^5 - 2^0 = 31 */ fe25519_mul(&z2_5_0,&t0,&z9);
/* 2^6 - 2^1 */ fe25519_square(&t0,&z2_5_0);
/* 2^7 - 2^2 */ fe25519_square(&t1,&t0);
/* 2^8 - 2^3 */ fe25519_square(&t0,&t1);
/* 2^9 - 2^4 */ fe25519_square(&t1,&t0);
/* 2^10 - 2^5 */ fe25519_square(&t0,&t1);
/* 2^10 - 2^0 */ fe25519_mul(&z2_10_0,&t0,&z2_5_0);
/* 2^11 - 2^1 */ fe25519_square(&t0,&z2_10_0);
/* 2^12 - 2^2 */ fe25519_square(&t1,&t0);
/* 2^20 - 2^10 */ for (i = 2;i < 10;i += 2) { fe25519_square(&t0,&t1); fe25519_square(&t1,&t0); }
/* 2^20 - 2^0 */ fe25519_mul(&z2_20_0,&t1,&z2_10_0);
/* 2^21 - 2^1 */ fe25519_square(&t0,&z2_20_0);
/* 2^22 - 2^2 */ fe25519_square(&t1,&t0);
/* 2^40 - 2^20 */ for (i = 2;i < 20;i += 2) { fe25519_square(&t0,&t1); fe25519_square(&t1,&t0); }
/* 2^40 - 2^0 */ fe25519_mul(&t0,&t1,&z2_20_0);
/* 2^41 - 2^1 */ fe25519_square(&t1,&t0);
/* 2^42 - 2^2 */ fe25519_square(&t0,&t1);
/* 2^50 - 2^10 */ for (i = 2;i < 10;i += 2) { fe25519_square(&t1,&t0); fe25519_square(&t0,&t1); }
/* 2^50 - 2^0 */ fe25519_mul(&z2_50_0,&t0,&z2_10_0);
/* 2^51 - 2^1 */ fe25519_square(&t0,&z2_50_0);
/* 2^52 - 2^2 */ fe25519_square(&t1,&t0);
/* 2^100 - 2^50 */ for (i = 2;i < 50;i += 2) { fe25519_square(&t0,&t1); fe25519_square(&t1,&t0); }
/* 2^100 - 2^0 */ fe25519_mul(&z2_100_0,&t1,&z2_50_0);
/* 2^101 - 2^1 */ fe25519_square(&t1,&z2_100_0);
/* 2^102 - 2^2 */ fe25519_square(&t0,&t1);
/* 2^200 - 2^100 */ for (i = 2;i < 100;i += 2) { fe25519_square(&t1,&t0); fe25519_square(&t0,&t1); }
/* 2^200 - 2^0 */ fe25519_mul(&t1,&t0,&z2_100_0);
/* 2^201 - 2^1 */ fe25519_square(&t0,&t1);
/* 2^202 - 2^2 */ fe25519_square(&t1,&t0);
/* 2^250 - 2^50 */ for (i = 2;i < 50;i += 2) { fe25519_square(&t0,&t1); fe25519_square(&t1,&t0); }
/* 2^250 - 2^0 */ fe25519_mul(&t0,&t1,&z2_50_0);
/* 2^251 - 2^1 */ fe25519_square(&t1,&t0);
/* 2^252 - 2^2 */ fe25519_square(&t0,&t1);
/* 2^253 - 2^3 */ fe25519_square(&t1,&t0);
/* 2^254 - 2^4 */ fe25519_square(&t0,&t1);
/* 2^255 - 2^5 */ fe25519_square(&t1,&t0);
/* 2^255 - 21 */ fe25519_mul(r,&t1,&z11);
}
void fe25519_pow2523(fe25519 *r, const fe25519 *x)
{
fe25519 z2;
fe25519 z9;
fe25519 z11;
fe25519 z2_5_0;
fe25519 z2_10_0;
fe25519 z2_20_0;
fe25519 z2_50_0;
fe25519 z2_100_0;
fe25519 t;
int i;
/* 2 */ fe25519_square(&z2,x);
/* 4 */ fe25519_square(&t,&z2);
/* 8 */ fe25519_square(&t,&t);
/* 9 */ fe25519_mul(&z9,&t,x);
/* 11 */ fe25519_mul(&z11,&z9,&z2);
/* 22 */ fe25519_square(&t,&z11);
/* 2^5 - 2^0 = 31 */ fe25519_mul(&z2_5_0,&t,&z9);
/* 2^6 - 2^1 */ fe25519_square(&t,&z2_5_0);
/* 2^10 - 2^5 */ for (i = 1;i < 5;i++) { fe25519_square(&t,&t); }
/* 2^10 - 2^0 */ fe25519_mul(&z2_10_0,&t,&z2_5_0);
/* 2^11 - 2^1 */ fe25519_square(&t,&z2_10_0);
/* 2^20 - 2^10 */ for (i = 1;i < 10;i++) { fe25519_square(&t,&t); }
/* 2^20 - 2^0 */ fe25519_mul(&z2_20_0,&t,&z2_10_0);
/* 2^21 - 2^1 */ fe25519_square(&t,&z2_20_0);
/* 2^40 - 2^20 */ for (i = 1;i < 20;i++) { fe25519_square(&t,&t); }
/* 2^40 - 2^0 */ fe25519_mul(&t,&t,&z2_20_0);
/* 2^41 - 2^1 */ fe25519_square(&t,&t);
/* 2^50 - 2^10 */ for (i = 1;i < 10;i++) { fe25519_square(&t,&t); }
/* 2^50 - 2^0 */ fe25519_mul(&z2_50_0,&t,&z2_10_0);
/* 2^51 - 2^1 */ fe25519_square(&t,&z2_50_0);
/* 2^100 - 2^50 */ for (i = 1;i < 50;i++) { fe25519_square(&t,&t); }
/* 2^100 - 2^0 */ fe25519_mul(&z2_100_0,&t,&z2_50_0);
/* 2^101 - 2^1 */ fe25519_square(&t,&z2_100_0);
/* 2^200 - 2^100 */ for (i = 1;i < 100;i++) { fe25519_square(&t,&t); }
/* 2^200 - 2^0 */ fe25519_mul(&t,&t,&z2_100_0);
/* 2^201 - 2^1 */ fe25519_square(&t,&t);
/* 2^250 - 2^50 */ for (i = 1;i < 50;i++) { fe25519_square(&t,&t); }
/* 2^250 - 2^0 */ fe25519_mul(&t,&t,&z2_50_0);
/* 2^251 - 2^1 */ fe25519_square(&t,&t);
/* 2^252 - 2^2 */ fe25519_square(&t,&t);
/* 2^252 - 3 */ fe25519_mul(r,&t,x);
}