void ge_neg(ge_p3* r, const ge_p3 *p)
{
fe_neg(r->X, p->X);
fe_copy(r->Y, p->Y);
fe_copy(r->Z, p->Z);
fe_neg(r->T, p->T);
}
void ge_montx_to_p2(ge_p2* p, const fe u, const unsigned char ed_sign_bit)
{
fe x, y, A, v, v2, iv, nx;
fe_frombytes(A, A_bytes);
/* given u, recover edwards y */
/* given u, recover v */
/* given u and v, recover edwards x */
fe_montx_to_edy(y, u); /* y = (u - 1) / (u + 1) */
fe_mont_rhs(v2, u); /* v^2 = u(u^2 + Au + 1) */
fe_sqrt(v, v2); /* v = sqrt(v^2) */
fe_mul(x, u, A); /* x = u * sqrt(-(A+2)) */
fe_invert(iv, v); /* 1/v */
fe_mul(x, x, iv); /* x = (u/v) * sqrt(-(A+2)) */
fe_neg(nx, x); /* negate x to match sign bit */
fe_cmov(x, nx, fe_isnegative(x) ^ ed_sign_bit);
fe_copy(p->X, x);
fe_copy(p->Y, y);
fe_1(p->Z);
/* POSTCONDITION: check that p->X and p->Y satisfy the Ed curve equation */
/* -x^2 + y^2 = 1 + dx^2y^2 */
#ifndef NDEBUG
{
fe one, d, x2, y2, x2y2, dx2y2;
unsigned char dbytes[32] = {
0xa3, 0x78, 0x59, 0x13, 0xca, 0x4d, 0xeb, 0x75,
0xab, 0xd8, 0x41, 0x41, 0x4d, 0x0a, 0x70, 0x00,
0x98, 0xe8, 0x79, 0x77, 0x79, 0x40, 0xc7, 0x8c,
0x73, 0xfe, 0x6f, 0x2b, 0xee, 0x6c, 0x03, 0x52
};
fe_frombytes(d, dbytes);
fe_1(one);
fe_sq(x2, p->X); /* x^2 */
fe_sq(y2, p->Y); /* y^2 */
fe_mul(dx2y2, x2, y2); /* x^2y^2 */
fe_mul(dx2y2, dx2y2, d); /* dx^2y^2 */
fe_add(dx2y2, dx2y2, one); /* dx^2y^2 + 1 */
fe_neg(x2y2, x2); /* -x^2 */
fe_add(x2y2, x2y2, y2); /* -x^2 + y^2 */
assert(fe_isequal(x2y2, dx2y2));
}
#endif
}
extern void ge_p3_to_cached(ge_cached *r,const ge_p3 *p)
{
fe_add(r->YplusX,p->Y,p->X);
fe_sub(r->YminusX,p->Y,p->X);
fe_copy(r->Z,p->Z);
fe_mul(r->T2d,p->T,d2);
}
/*
out = (1 / in) % m;
... using Fermat's Little Theorem
44 mul, 262 sq
*/
static void _inv(fe out, const fe in) {
fe o, x2, x4, x8, x16, x32, t[16];
long long i;
_sq(o, in);
for (i = 0; i < 1 - 1; ++i) _sq(o, o);
_mul(x2, o, in);
_sq(o, x2);
for (i = 0; i < 2 - 1; ++i) _sq(o, o);
_mul(x4, o, x2);
_sq(o, x4);
for (i = 0; i < 4 - 1; ++i) _sq(o, o);
_mul(x8, o, x4);
_sq(o, x8);
for (i = 0; i < 8 - 1; ++i) _sq(o, o);
_mul(x16, o, x8);
_sq(o, x16);
for (i = 0; i < 16 - 1; ++i) _sq(o, o);
_mul(x32, o, x16);
_sq(o, x32);
for (i = 0; i < 32 - 1; ++i) _sq(o, o);
for (i = 0; i < 32; ++i) _sq(o, o);
_mul(o, o, x32);
for (i = 0; i < 32; ++i) _sq(o, o);
_mul(o, o, x32);
fe_0(t[0]);
fe_copy(t[1], in);
_sq(t[2], t[1]);
fe_copy(t[3], x2);
for (i = 4; i < 15; ++i) {
if ((i & 1) == 0) _sq(t[i], t[i / 2]);
else _mul(t[i], t[i - 1], in);
}
fe_copy(t[15], x4);
for (i = 0; i < 32; ++i) {
_sq(o, o); _sq(o, o); _sq(o, o); _sq(o, o);
_mul(o, o, t[m2[i]]);
}
fe_copy(out, o);
cleanup(o); cleanup(t);
cleanup(x2); cleanup(x4);
cleanup(x8); cleanup(x16); cleanup(x32);
}
/*
Test if the public key can be uncommpressed and negate it (-X,Y,Z,-T)
return 0 on success
*/
int ge_frombytes_negate_vartime(ge_p3 *p,const unsigned char *s)
{
byte parity;
byte x[F25519_SIZE];
byte y[F25519_SIZE];
byte a[F25519_SIZE];
byte b[F25519_SIZE];
byte c[F25519_SIZE];
int ret = 0;
/* unpack the key s */
parity = s[31] >> 7;
fe_copy(y, s);
y[31] &= 127;
fe_mul__distinct(c, y, y);
fe_mul__distinct(b, c, ed25519_d);
fe_add(a, b, f25519_one);
fe_inv__distinct(b, a);
fe_sub(a, c, f25519_one);
fe_mul__distinct(c, a, b);
fe_sqrt(a, c);
fe_neg(b, a);
fe_select(x, a, b, (a[0] ^ parity) & 1);
/* test that x^2 is equal to c */
fe_mul__distinct(a, x, x);
fe_normalize(a);
fe_normalize(c);
ret |= ConstantCompare(a, c, F25519_SIZE);
/* project the key s onto p */
fe_copy(p->X, x);
fe_copy(p->Y, y);
fe_load(p->Z, 1);
fe_mul__distinct(p->T, x, y);
/* negate, the point becomes (-X,Y,Z,-T) */
fe_neg(p->X,p->X);
fe_neg(p->T,p->T);
return ret;
}
int ge_compress_key(byte* out, const byte* xIn, const byte* yIn,
word32 keySz)
{
byte tmp[F25519_SIZE];
byte parity;
byte pt[32];
int i;
fe_copy(tmp, xIn);
parity = (tmp[0] & 1) << 7;
fe_copy(pt, yIn);
pt[31] |= parity;
for(i = 0; i < 32; i++) {
out[32-i-1] = pt[i];
}
(void)keySz;
return 0;
}
// fe_inv calculates |out| = |in|^{-1}
//
// Based on Fermat's Little Theorem:
// a^p = a (mod p)
// a^{p-1} = 1 (mod p)
// a^{p-2} = a^{-1} (mod p)
static void fe_inv(fe out, const fe in) {
fe ftmp, ftmp2;
// each e_I will hold |in|^{2^I - 1}
fe e2, e4, e8, e16, e32, e64;
fe_sqr(ftmp, in); // 2^1
fe_mul(ftmp, in, ftmp); // 2^2 - 2^0
fe_copy(e2, ftmp);
fe_sqr(ftmp, ftmp); // 2^3 - 2^1
fe_sqr(ftmp, ftmp); // 2^4 - 2^2
fe_mul(ftmp, ftmp, e2); // 2^4 - 2^0
fe_copy(e4, ftmp);
fe_sqr(ftmp, ftmp); // 2^5 - 2^1
fe_sqr(ftmp, ftmp); // 2^6 - 2^2
fe_sqr(ftmp, ftmp); // 2^7 - 2^3
fe_sqr(ftmp, ftmp); // 2^8 - 2^4
fe_mul(ftmp, ftmp, e4); // 2^8 - 2^0
fe_copy(e8, ftmp);
for (size_t i = 0; i < 8; i++) {
fe_sqr(ftmp, ftmp);
} // 2^16 - 2^8
fe_mul(ftmp, ftmp, e8); // 2^16 - 2^0
fe_copy(e16, ftmp);
for (size_t i = 0; i < 16; i++) {
fe_sqr(ftmp, ftmp);
} // 2^32 - 2^16
fe_mul(ftmp, ftmp, e16); // 2^32 - 2^0
fe_copy(e32, ftmp);
for (size_t i = 0; i < 32; i++) {
fe_sqr(ftmp, ftmp);
} // 2^64 - 2^32
fe_copy(e64, ftmp);
fe_mul(ftmp, ftmp, in); // 2^64 - 2^32 + 2^0
for (size_t i = 0; i < 192; i++) {
fe_sqr(ftmp, ftmp);
} // 2^256 - 2^224 + 2^192
fe_mul(ftmp2, e64, e32); // 2^64 - 2^0
for (size_t i = 0; i < 16; i++) {
fe_sqr(ftmp2, ftmp2);
} // 2^80 - 2^16
fe_mul(ftmp2, ftmp2, e16); // 2^80 - 2^0
for (size_t i = 0; i < 8; i++) {
fe_sqr(ftmp2, ftmp2);
} // 2^88 - 2^8
fe_mul(ftmp2, ftmp2, e8); // 2^88 - 2^0
for (size_t i = 0; i < 4; i++) {
fe_sqr(ftmp2, ftmp2);
} // 2^92 - 2^4
fe_mul(ftmp2, ftmp2, e4); // 2^92 - 2^0
fe_sqr(ftmp2, ftmp2); // 2^93 - 2^1
fe_sqr(ftmp2, ftmp2); // 2^94 - 2^2
fe_mul(ftmp2, ftmp2, e2); // 2^94 - 2^0
fe_sqr(ftmp2, ftmp2); // 2^95 - 2^1
fe_sqr(ftmp2, ftmp2); // 2^96 - 2^2
fe_mul(ftmp2, ftmp2, in); // 2^96 - 3
fe_mul(out, ftmp2, ftmp); // 2^256 - 2^224 + 2^192 + 2^96 - 3
}
int ge_double_scalarmult_vartime(ge_p2* R, const unsigned char *h,
const ge_p3 *inA,const unsigned char *sig)
{
ge_p3 p, A;
int ret = 0;
XMEMCPY(&A, inA, sizeof(ge_p3));
/* find SB */
ed25519_smult(&p, &ed25519_base, sig);
/* find H(R,A,M) * -A */
ed25519_smult(&A, &A, h);
/* SB + -H(R,A,M)A */
ed25519_add(&A, &p, &A);
fe_copy(R->X, A.X);
fe_copy(R->Y, A.Y);
fe_copy(R->Z, A.Z);
return ret;
}
static int curve25519(unsigned char* q, unsigned char* n, unsigned char* p)
{
unsigned char e[32];
unsigned int i;
fe x1;
fe x2;
fe z2;
fe x3;
fe z3;
fe tmp0;
fe tmp1;
int pos;
unsigned int swap;
unsigned int b;
for (i = 0;i < 32;++i) e[i] = n[i];
e[0] &= 248;
e[31] &= 127;
e[31] |= 64;
fe_frombytes(x1,p);
fe_1(x2);
fe_0(z2);
fe_copy(x3,x1);
fe_1(z3);
swap = 0;
for (pos = 254;pos >= 0;--pos) {
b = e[pos / 8] >> (pos & 7);
b &= 1;
swap ^= b;
fe_cswap(x2,x3,swap);
fe_cswap(z2,z3,swap);
swap = b;
#include <cyassl/ctaocrypt/ecc25519_montgomery.h>
}
fe_cswap(x2,x3,swap);
fe_cswap(z2,z3,swap);
fe_invert(z2,z2);
fe_mul(x2,x2,z2);
fe_tobytes(q,x2);
return 0;
}
/* pack the point h into array s */
void ge_tobytes(unsigned char *s,const ge_p2 *h)
{
byte x[F25519_SIZE];
byte y[F25519_SIZE];
byte z1[F25519_SIZE];
byte parity;
fe_inv__distinct(z1, h->Z);
fe_mul__distinct(x, h->X, z1);
fe_mul__distinct(y, h->Y, z1);
fe_normalize(x);
fe_normalize(y);
parity = (x[0] & 1) << 7;
fe_copy(s, y);
fe_normalize(s);
s[31] |= parity;
}
void fe_sqrt(fe out, const fe a)
{
fe exp, b, b2, bi, i;
#ifndef NDEBUG
fe legendre, zero, one;
#endif
fe_frombytes(i, i_bytes);
fe_pow22523(exp, a); /* b = a^(q-5)/8 */
/* PRECONDITION: legendre symbol == 1 (square) or 0 (a == zero) */
#ifndef NDEBUG
fe_sq(legendre, exp); /* in^((q-5)/4) */
fe_sq(legendre, legendre); /* in^((q-5)/2) */
fe_mul(legendre, legendre, a); /* in^((q-3)/2) */
fe_mul(legendre, legendre, a); /* in^((q-1)/2) */
fe_0(zero);
fe_1(one);
assert(fe_isequal(legendre, zero) || fe_isequal(legendre, one));
#endif
fe_mul(b, a, exp); /* b = a * a^(q-5)/8 */
fe_sq(b2, b); /* b^2 = a * a^(q-1)/4 */
/* note b^4 == a^2, so b^2 == a or -a
* if b^2 != a, multiply it by sqrt(-1) */
fe_mul(bi, b, i);
fe_cmov(b, bi, 1 ^ fe_isequal(b2, a));
fe_copy(out, b);
/* PRECONDITION: out^2 == a */
#ifndef NDEBUG
fe_sq(b2, out);
assert(fe_isequal(a, b2));
#endif
}
extern void ge_p3_to_p2(ge_p2 *r,const ge_p3 *p)
{
fe_copy(r->X,p->X);
fe_copy(r->Y,p->Y);
fe_copy(r->Z,p->Z);
}
// point_add calcuates (x1, y1, z1) + (x2, y2, z2)
//
// The method is taken from:
// http://hyperelliptic.org/EFD/g1p/auto-shortw-jacobian-3.html#addition-add-2007-bl,
// adapted for mixed addition (z2 = 1, or z2 = 0 for the point at infinity).
//
// Coq transcription and correctness proof:
// <https://github.com/mit-plv/fiat-crypto/blob/79f8b5f39ed609339f0233098dee1a3c4e6b3080/src/Curves/Weierstrass/Jacobian.v#L135>
// <https://github.com/mit-plv/fiat-crypto/blob/79f8b5f39ed609339f0233098dee1a3c4e6b3080/src/Curves/Weierstrass/Jacobian.v#L205>
//
// This function includes a branch for checking whether the two input points
// are equal, (while not equal to the point at infinity). This case never
// happens during single point multiplication, so there is no timing leak for
// ECDH or ECDSA signing.
static void point_add(fe x3, fe y3, fe z3, const fe x1,
const fe y1, const fe z1, const int mixed,
const fe x2, const fe y2, const fe z2) {
fe x_out, y_out, z_out;
limb_t z1nz = fe_nz(z1);
limb_t z2nz = fe_nz(z2);
// z1z1 = z1z1 = z1**2
fe z1z1; fe_sqr(z1z1, z1);
fe u1, s1, two_z1z2;
if (!mixed) {
// z2z2 = z2**2
fe z2z2; fe_sqr(z2z2, z2);
// u1 = x1*z2z2
fe_mul(u1, x1, z2z2);
// two_z1z2 = (z1 + z2)**2 - (z1z1 + z2z2) = 2z1z2
fe_add(two_z1z2, z1, z2);
fe_sqr(two_z1z2, two_z1z2);
fe_sub(two_z1z2, two_z1z2, z1z1);
fe_sub(two_z1z2, two_z1z2, z2z2);
// s1 = y1 * z2**3
fe_mul(s1, z2, z2z2);
fe_mul(s1, s1, y1);
} else {
// We'll assume z2 = 1 (special case z2 = 0 is handled later).
// u1 = x1*z2z2
fe_copy(u1, x1);
// two_z1z2 = 2z1z2
fe_add(two_z1z2, z1, z1);
// s1 = y1 * z2**3
fe_copy(s1, y1);
}
// u2 = x2*z1z1
fe u2; fe_mul(u2, x2, z1z1);
// h = u2 - u1
fe h; fe_sub(h, u2, u1);
limb_t xneq = fe_nz(h);
// z_out = two_z1z2 * h
fe_mul(z_out, h, two_z1z2);
// z1z1z1 = z1 * z1z1
fe z1z1z1; fe_mul(z1z1z1, z1, z1z1);
// s2 = y2 * z1**3
fe s2; fe_mul(s2, y2, z1z1z1);
// r = (s2 - s1)*2
fe r;
fe_sub(r, s2, s1);
fe_add(r, r, r);
limb_t yneq = fe_nz(r);
if (!xneq && !yneq && z1nz && z2nz) {
point_double(x3, y3, z3, x1, y1, z1);
return;
}
// I = (2h)**2
fe i;
fe_add(i, h, h);
fe_sqr(i, i);
// J = h * I
fe j; fe_mul(j, h, i);
// V = U1 * I
fe v; fe_mul(v, u1, i);
// x_out = r**2 - J - 2V
fe_sqr(x_out, r);
fe_sub(x_out, x_out, j);
fe_sub(x_out, x_out, v);
fe_sub(x_out, x_out, v);
// y_out = r(V-x_out) - 2 * s1 * J
fe_sub(y_out, v, x_out);
fe_mul(y_out, y_out, r);
fe s1j;
fe_mul(s1j, s1, j);
fe_sub(y_out, y_out, s1j);
fe_sub(y_out, y_out, s1j);
fe_cmovznz(x_out, z1nz, x2, x_out);
fe_cmovznz(x3, z2nz, x1, x_out);
fe_cmovznz(y_out, z1nz, y2, y_out);
fe_cmovznz(y3, z2nz, y1, y_out);
fe_cmovznz(z_out, z1nz, z2, z_out);
fe_cmovznz(z3, z2nz, z1, z_out);
}
// Interleaved point multiplication using precomputed point multiples: The
// small point multiples 0*P, 1*P, ..., 17*P are in p_pre_comp, the scalar
// in p_scalar, if non-NULL. If g_scalar is non-NULL, we also add this multiple
// of the generator, using certain (large) precomputed multiples in g_pre_comp.
// Output point (X, Y, Z) is stored in x_out, y_out, z_out.
static void batch_mul(fe x_out, fe y_out, fe z_out,
const uint8_t *p_scalar, const uint8_t *g_scalar,
const fe p_pre_comp[17][3]) {
// set nq to the point at infinity
fe nq[3] = {{0},{0},{0}}, ftmp, tmp[3];
uint64_t bits;
uint8_t sign, digit;
// Loop over both scalars msb-to-lsb, interleaving additions of multiples
// of the generator (two in each of the last 32 rounds) and additions of p
// (every 5th round).
int skip = 1; // save two point operations in the first round
size_t i = p_scalar != NULL ? 255 : 31;
for (;;) {
// double
if (!skip) {
point_double(nq[0], nq[1], nq[2], nq[0], nq[1], nq[2]);
}
// add multiples of the generator
if (g_scalar != NULL && i <= 31) {
// first, look 32 bits upwards
bits = get_bit(g_scalar, i + 224) << 3;
bits |= get_bit(g_scalar, i + 160) << 2;
bits |= get_bit(g_scalar, i + 96) << 1;
bits |= get_bit(g_scalar, i + 32);
// select the point to add, in constant time
select_point(bits, 16, g_pre_comp[1], tmp);
if (!skip) {
point_add(nq[0], nq[1], nq[2], nq[0], nq[1], nq[2], 1 /* mixed */,
tmp[0], tmp[1], tmp[2]);
} else {
fe_copy(nq[0], tmp[0]);
fe_copy(nq[1], tmp[1]);
fe_copy(nq[2], tmp[2]);
skip = 0;
}
// second, look at the current position
bits = get_bit(g_scalar, i + 192) << 3;
bits |= get_bit(g_scalar, i + 128) << 2;
bits |= get_bit(g_scalar, i + 64) << 1;
bits |= get_bit(g_scalar, i);
// select the point to add, in constant time
select_point(bits, 16, g_pre_comp[0], tmp);
point_add(nq[0], nq[1], nq[2], nq[0], nq[1], nq[2], 1 /* mixed */, tmp[0],
tmp[1], tmp[2]);
}
// do other additions every 5 doublings
if (p_scalar != NULL && i % 5 == 0) {
bits = get_bit(p_scalar, i + 4) << 5;
bits |= get_bit(p_scalar, i + 3) << 4;
bits |= get_bit(p_scalar, i + 2) << 3;
bits |= get_bit(p_scalar, i + 1) << 2;
bits |= get_bit(p_scalar, i) << 1;
bits |= get_bit(p_scalar, i - 1);
ec_GFp_nistp_recode_scalar_bits(&sign, &digit, bits);
// select the point to add or subtract, in constant time.
select_point(digit, 17, p_pre_comp, tmp);
fe_opp(ftmp, tmp[1]); // (X, -Y, Z) is the negative point.
fe_cmovznz(tmp[1], sign, tmp[1], ftmp);
if (!skip) {
point_add(nq[0], nq[1], nq[2], nq[0], nq[1], nq[2], 0 /* mixed */,
tmp[0], tmp[1], tmp[2]);
} else {
fe_copy(nq[0], tmp[0]);
fe_copy(nq[1], tmp[1]);
fe_copy(nq[2], tmp[2]);
skip = 0;
}
}
if (i == 0) {
break;
}
--i;
}
fe_copy(x_out, nq[0]);
fe_copy(y_out, nq[1]);
fe_copy(z_out, nq[2]);
}
static void ec_GFp_nistp256_point_mul_public(const EC_GROUP *group,
EC_RAW_POINT *r,
const EC_SCALAR *g_scalar,
const EC_RAW_POINT *p,
const EC_SCALAR *p_scalar) {
#define P256_WSIZE_PUBLIC 4
// Precompute multiples of |p|. p_pre_comp[i] is (2*i+1) * |p|.
fe p_pre_comp[1 << (P256_WSIZE_PUBLIC-1)][3];
fe_from_generic(p_pre_comp[0][0], &p->X);
fe_from_generic(p_pre_comp[0][1], &p->Y);
fe_from_generic(p_pre_comp[0][2], &p->Z);
fe p2[3];
point_double(p2[0], p2[1], p2[2], p_pre_comp[0][0], p_pre_comp[0][1],
p_pre_comp[0][2]);
for (size_t i = 1; i < OPENSSL_ARRAY_SIZE(p_pre_comp); i++) {
point_add(p_pre_comp[i][0], p_pre_comp[i][1], p_pre_comp[i][2],
p_pre_comp[i - 1][0], p_pre_comp[i - 1][1], p_pre_comp[i - 1][2],
0 /* not mixed */, p2[0], p2[1], p2[2]);
}
// Set up the coefficients for |p_scalar|.
int8_t p_wNAF[257];
ec_compute_wNAF(group, p_wNAF, p_scalar, 256, P256_WSIZE_PUBLIC);
// Set |ret| to the point at infinity.
int skip = 1; // Save some point operations.
fe ret[3] = {{0},{0},{0}};
for (int i = 256; i >= 0; i--) {
if (!skip) {
point_double(ret[0], ret[1], ret[2], ret[0], ret[1], ret[2]);
}
// For the |g_scalar|, we use the precomputed table without the
// constant-time lookup.
if (i <= 31) {
// First, look 32 bits upwards.
uint64_t bits = get_bit(g_scalar->bytes, i + 224) << 3;
bits |= get_bit(g_scalar->bytes, i + 160) << 2;
bits |= get_bit(g_scalar->bytes, i + 96) << 1;
bits |= get_bit(g_scalar->bytes, i + 32);
point_add(ret[0], ret[1], ret[2], ret[0], ret[1], ret[2], 1 /* mixed */,
g_pre_comp[1][bits][0], g_pre_comp[1][bits][1],
g_pre_comp[1][bits][2]);
skip = 0;
// Second, look at the current position.
bits = get_bit(g_scalar->bytes, i + 192) << 3;
bits |= get_bit(g_scalar->bytes, i + 128) << 2;
bits |= get_bit(g_scalar->bytes, i + 64) << 1;
bits |= get_bit(g_scalar->bytes, i);
point_add(ret[0], ret[1], ret[2], ret[0], ret[1], ret[2], 1 /* mixed */,
g_pre_comp[0][bits][0], g_pre_comp[0][bits][1],
g_pre_comp[0][bits][2]);
}
int digit = p_wNAF[i];
if (digit != 0) {
assert(digit & 1);
int idx = digit < 0 ? (-digit) >> 1 : digit >> 1;
fe *y = &p_pre_comp[idx][1], tmp;
if (digit < 0) {
fe_opp(tmp, p_pre_comp[idx][1]);
y = &tmp;
}
if (!skip) {
point_add(ret[0], ret[1], ret[2], ret[0], ret[1], ret[2],
0 /* not mixed */, p_pre_comp[idx][0], *y, p_pre_comp[idx][2]);
} else {
fe_copy(ret[0], p_pre_comp[idx][0]);
fe_copy(ret[1], *y);
fe_copy(ret[2], p_pre_comp[idx][2]);
skip = 0;
}
}
}
void ed25519_key_exchange(unsigned char *shared_secret, const unsigned char *public_key, const unsigned char *private_key) {
unsigned char e[32];
unsigned int i;
fe x1;
fe x2;
fe z2;
fe x3;
fe z3;
fe tmp0;
fe tmp1;
int pos;
unsigned int swap;
unsigned int b;
/* copy the private key and make sure it's valid */
for (i = 0; i < 32; ++i) {
e[i] = private_key[i];
}
e[0] &= 248;
e[31] &= 63;
e[31] |= 64;
/* unpack the public key and convert edwards to montgomery */
/* due to CodesInChaos: montgomeryX = (edwardsY + 1)*inverse(1 - edwardsY) mod p */
fe_frombytes(x1, public_key);
fe_1(tmp1);
fe_add(tmp0, x1, tmp1);
fe_sub(tmp1, tmp1, x1);
fe_invert(tmp1, tmp1);
fe_mul(x1, tmp0, tmp1);
fe_1(x2);
fe_0(z2);
fe_copy(x3, x1);
fe_1(z3);
swap = 0;
for (pos = 254; pos >= 0; --pos) {
b = e[pos / 8] >> (pos & 7);
b &= 1;
swap ^= b;
fe_cswap(x2, x3, swap);
fe_cswap(z2, z3, swap);
swap = b;
/* from montgomery.h */
fe_sub(tmp0, x3, z3);
fe_sub(tmp1, x2, z2);
fe_add(x2, x2, z2);
fe_add(z2, x3, z3);
fe_mul(z3, tmp0, x2);
fe_mul(z2, z2, tmp1);
fe_sq(tmp0, tmp1);
fe_sq(tmp1, x2);
fe_add(x3, z3, z2);
fe_sub(z2, z3, z2);
fe_mul(x2, tmp1, tmp0);
fe_sub(tmp1, tmp1, tmp0);
fe_sq(z2, z2);
fe_mul121666(z3, tmp1);
fe_sq(x3, x3);
fe_add(tmp0, tmp0, z3);
fe_mul(z3, x1, z2);
fe_mul(z2, tmp1, tmp0);
}
fe_cswap(x2, x3, swap);
fe_cswap(z2, z3, swap);
fe_invert(z2, z2);
fe_mul(x2, x2, z2);
fe_tobytes(shared_secret, x2);
}
static int
crypto_scalarmult_curve25519_ref10(unsigned char *q,
const unsigned char *n,
const unsigned char *p)
{
unsigned char e[32];
unsigned int i;
fe x1;
fe x2;
fe z2;
fe x3;
fe z3;
fe tmp0;
fe tmp1;
int pos;
unsigned int swap;
unsigned int b;
for (i = 0;i < 32;++i) e[i] = n[i];
e[0] &= 248;
e[31] &= 127;
e[31] |= 64;
fe_frombytes(x1,p);
fe_1(x2);
fe_0(z2);
fe_copy(x3,x1);
fe_1(z3);
swap = 0;
for (pos = 254;pos >= 0;--pos) {
b = e[pos / 8] >> (pos & 7);
b &= 1;
swap ^= b;
fe_cswap(x2,x3,swap);
fe_cswap(z2,z3,swap);
swap = b;
fe_sub(tmp0,x3,z3);
fe_sub(tmp1,x2,z2);
fe_add(x2,x2,z2);
fe_add(z2,x3,z3);
fe_mul(z3,tmp0,x2);
fe_mul(z2,z2,tmp1);
fe_sq(tmp0,tmp1);
fe_sq(tmp1,x2);
fe_add(x3,z3,z2);
fe_sub(z2,z3,z2);
fe_mul(x2,tmp1,tmp0);
fe_sub(tmp1,tmp1,tmp0);
fe_sq(z2,z2);
fe_mul121666(z3,tmp1);
fe_sq(x3,x3);
fe_add(tmp0,tmp0,z3);
fe_mul(z3,x1,z2);
fe_mul(z2,tmp1,tmp0);
}
fe_cswap(x2,x3,swap);
fe_cswap(z2,z3,swap);
fe_invert(z2,z2);
fe_mul(x2,x2,z2);
fe_tobytes(q,x2);
return 0;
}
