Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Add x-only ecmult_const version with x specified as n/d #1118

Merged
merged 2 commits into from
Apr 10, 2023
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
21 changes: 21 additions & 0 deletions src/ecmult_const.h
Original file line number Diff line number Diff line change
Expand Up @@ -18,4 +18,25 @@
*/
static void secp256k1_ecmult_const(secp256k1_gej *r, const secp256k1_ge *a, const secp256k1_scalar *q, int bits);

/**
* Same as secp256k1_ecmult_const, but takes in an x coordinate of the base point
* only, specified as fraction n/d (numerator/denominator). Only the x coordinate of the result is
* returned.
*
* If known_on_curve is 0, a verification is performed that n/d is a valid X
* coordinate, and 0 is returned if not. Otherwise, 1 is returned.
*
* d being NULL is interpreted as d=1. If non-NULL, d must not be zero. q must not be zero.
*
* Constant time in the value of q, but not any other inputs.
real-or-random marked this conversation as resolved.
Show resolved Hide resolved
*/
static int secp256k1_ecmult_const_xonly(
secp256k1_fe *r,
const secp256k1_fe *n,
const secp256k1_fe *d,
const secp256k1_scalar *q,
int bits,
int known_on_curve
);

#endif /* SECP256K1_ECMULT_CONST_H */
135 changes: 135 additions & 0 deletions src/ecmult_const_impl.h
Original file line number Diff line number Diff line change
Expand Up @@ -228,4 +228,139 @@ static void secp256k1_ecmult_const(secp256k1_gej *r, const secp256k1_ge *a, cons
secp256k1_fe_mul(&r->z, &r->z, &Z);
}

static int secp256k1_ecmult_const_xonly(secp256k1_fe* r, const secp256k1_fe *n, const secp256k1_fe *d, const secp256k1_scalar *q, int bits, int known_on_curve) {

/* This algorithm is a generalization of Peter Dettman's technique for
* avoiding the square root in a random-basepoint x-only multiplication
* on a Weierstrass curve:
* https://mailarchive.ietf.org/arch/msg/cfrg/7DyYY6gg32wDgHAhgSb6XxMDlJA/
*
*
* === Background: the effective affine technique ===
*
* Let phi_u be the isomorphism that maps (x, y) on secp256k1 curve y^2 = x^3 + 7 to
* x' = u^2*x, y' = u^3*y on curve y'^2 = x'^3 + u^6*7. This new curve has the same order as
sipa marked this conversation as resolved.
Show resolved Hide resolved
* the original (it is isomorphic), but moreover, has the same addition/doubling formulas, as
* the curve b=7 coefficient does not appear in those formulas (or at least does not appear in
* the formulas implemented in this codebase, both affine and Jacobian). See also Example 9.5.2
* in https://www.math.auckland.ac.nz/~sgal018/crypto-book/ch9.pdf.
*
* This means any linear combination of secp256k1 points can be computed by applying phi_u
* (with non-zero u) on all input points (including the generator, if used), computing the
* linear combination on the isomorphic curve (using the same group laws), and then applying
* phi_u^{-1} to get back to secp256k1.
*
* Switching to Jacobian coordinates, note that phi_u applied to (X, Y, Z) is simply
* (X, Y, Z/u). Thus, if we want to compute (X1, Y1, Z) + (X2, Y2, Z), with identical Z
* coordinates, we can use phi_Z to transform it to (X1, Y1, 1) + (X2, Y2, 1) on an isomorphic
* curve where the affine addition formula can be used instead.
* If (X3, Y3, Z3) = (X1, Y1) + (X2, Y2) on that curve, then our answer on secp256k1 is
* (X3, Y3, Z3*Z).
*
* This is the effective affine technique: if we have a linear combination of group elements
* to compute, and all those group elements have the same Z coordinate, we can simply pretend
* that all those Z coordinates are 1, perform the computation that way, and then multiply the
* original Z coordinate back in.
*
* The technique works on any a=0 short Weierstrass curve. It is possible to generalize it to
* other curves too, but there the isomorphic curves will have different 'a' coefficients,
* which typically does affect the group laws.
*
*
* === Avoiding the square root for x-only point multiplication ===
*
* In this function, we want to compute the X coordinate of q*(n/d, y), for
* y = sqrt((n/d)^3 + 7). Its negation would also be a valid Y coordinate, but by convention
* we pick whatever sqrt returns (which we assume to be a deterministic function).
*
* Let g = y^2*d^3 = n^3 + 7*d^3. This also means y = sqrt(g/d^3).
* Further let v = sqrt(d*g), which must exist as d*g = y^2*d^4 = (y*d^2)^2.
*
* The input point (n/d, y) also has Jacobian coordinates:
*
* (n/d, y, 1)
* = (n/d * v^2, y * v^3, v)
* = (n/d * d*g, y * sqrt(d^3*g^3), v)
* = (n/d * d*g, sqrt(y^2 * d^3*g^3), v)
* = (n*g, sqrt(g/d^3 * d^3*g^3), v)
* = (n*g, sqrt(g^4), v)
* = (n*g, g^2, v)
*
* It is easy to verify that both (n*g, g^2, v) and its negation (n*g, -g^2, v) have affine X
* coordinate n/d, and this holds even when the square root function doesn't have a
* determinstic sign. We choose the (n*g, g^2, v) version.
*
* Now switch to the effective affine curve using phi_v, where the input point has coordinates
* (n*g, g^2). Compute (X, Y, Z) = q * (n*g, g^2) there.
*
* Back on secp256k1, that means q * (n*g, g^2, v) = (X, Y, v*Z). This last point has affine X
* coordinate X / (v^2*Z^2) = X / (d*g*Z^2). Determining the affine Y coordinate would involve
* a square root, but as long as we only care about the resulting X coordinate, no square root
* is needed anywhere in this computation.
*/

secp256k1_fe g, i;
secp256k1_ge p;
secp256k1_gej rj;

/* Compute g = (n^3 + B*d^3). */
secp256k1_fe_sqr(&g, n);
secp256k1_fe_mul(&g, &g, n);
if (d) {
secp256k1_fe b;
#ifdef VERIFY
VERIFY_CHECK(!secp256k1_fe_normalizes_to_zero(d));
#endif
secp256k1_fe_sqr(&b, d);
VERIFY_CHECK(SECP256K1_B <= 8); /* magnitude of b will be <= 8 after the next call */
secp256k1_fe_mul_int(&b, SECP256K1_B);
sipa marked this conversation as resolved.
Show resolved Hide resolved
secp256k1_fe_mul(&b, &b, d);
secp256k1_fe_add(&g, &b);
if (!known_on_curve) {
/* We need to determine whether (n/d)^3 + 7 is square.
*
* is_square((n/d)^3 + 7)
* <=> is_square(((n/d)^3 + 7) * d^4)
* <=> is_square((n^3 + 7*d^3) * d)
* <=> is_square(g * d)
*/
secp256k1_fe c;
secp256k1_fe_mul(&c, &g, d);
if (!secp256k1_fe_is_square_var(&c)) return 0;
}
} else {
secp256k1_fe_add_int(&g, SECP256K1_B);
if (!known_on_curve) {
/* g at this point equals x^3 + 7. Test if it is square. */
if (!secp256k1_fe_is_square_var(&g)) return 0;
}
}

/* Compute base point P = (n*g, g^2), the effective affine version of (n*g, g^2, v), which has
* corresponding affine X coordinate n/d. */
secp256k1_fe_mul(&p.x, &g, n);
secp256k1_fe_sqr(&p.y, &g);
p.infinity = 0;

/* Perform x-only EC multiplication of P with q. */
#ifdef VERIFY
VERIFY_CHECK(!secp256k1_scalar_is_zero(q));
#endif
secp256k1_ecmult_const(&rj, &p, q, bits);
#ifdef VERIFY
VERIFY_CHECK(!secp256k1_gej_is_infinity(&rj));
#endif

/* The resulting (X, Y, Z) point on the effective-affine isomorphic curve corresponds to
* (X, Y, Z*v) on the secp256k1 curve. The affine version of that has X coordinate
* (X / (Z^2*d*g)). */
secp256k1_fe_sqr(&i, &rj.z);
secp256k1_fe_mul(&i, &i, &g);
if (d) secp256k1_fe_mul(&i, &i, d);
secp256k1_fe_inv(&i, &i);
secp256k1_fe_mul(r, &rj.x, &i);

return 1;
}

#endif /* SECP256K1_ECMULT_CONST_IMPL_H */
63 changes: 63 additions & 0 deletions src/tests.c
Original file line number Diff line number Diff line change
Expand Up @@ -4452,6 +4452,68 @@ static void ecmult_const_mult_zero_one(void) {
ge_equals_ge(&res2, &point);
}

static void ecmult_const_mult_xonly(void) {
int i;

/* Test correspondence between secp256k1_ecmult_const and secp256k1_ecmult_const_xonly. */
for (i = 0; i < 2*COUNT; ++i) {
secp256k1_ge base;
secp256k1_gej basej, resj;
secp256k1_fe n, d, resx, v;
secp256k1_scalar q;
int res;
/* Random base point. */
random_group_element_test(&base);
/* Random scalar to multiply it with. */
random_scalar_order_test(&q);
/* If i is odd, n=d*base.x for random non-zero d */
if (i & 1) {
do {
random_field_element_test(&d);
} while (secp256k1_fe_normalizes_to_zero_var(&d));
secp256k1_fe_mul(&n, &base.x, &d);
} else {
n = base.x;
}
/* Perform x-only multiplication. */
res = secp256k1_ecmult_const_xonly(&resx, &n, (i & 1) ? &d : NULL, &q, 256, i & 2);
CHECK(res);
/* Perform normal multiplication. */
secp256k1_gej_set_ge(&basej, &base);
secp256k1_ecmult(&resj, &basej, &q, NULL);
/* Check that resj's X coordinate corresponds with resx. */
secp256k1_fe_sqr(&v, &resj.z);
secp256k1_fe_mul(&v, &v, &resx);
CHECK(check_fe_equal(&v, &resj.x));
}

/* Test that secp256k1_ecmult_const_xonly correctly rejects X coordinates not on curve. */
for (i = 0; i < 2*COUNT; ++i) {
secp256k1_fe x, n, d, c, r;
int res;
secp256k1_scalar q;
random_scalar_order_test(&q);
/* Generate random X coordinate not on the curve. */
do {
random_field_element_test(&x);
secp256k1_fe_sqr(&c, &x);
secp256k1_fe_mul(&c, &c, &x);
secp256k1_fe_add(&c, &secp256k1_fe_const_b);
} while (secp256k1_fe_is_square_var(&c));
/* If i is odd, n=d*x for random non-zero d. */
if (i & 1) {
do {
random_field_element_test(&d);
} while (secp256k1_fe_normalizes_to_zero_var(&d));
secp256k1_fe_mul(&n, &x, &d);
} else {
n = x;
}
res = secp256k1_ecmult_const_xonly(&r, &n, (i & 1) ? &d : NULL, &q, 256, 0);
CHECK(res == 0);
}
}

static void ecmult_const_chain_multiply(void) {
/* Check known result (randomly generated test problem from sage) */
const secp256k1_scalar scalar = SECP256K1_SCALAR_CONST(
Expand Down Expand Up @@ -4483,6 +4545,7 @@ static void run_ecmult_const_tests(void) {
ecmult_const_random_mult();
ecmult_const_commutativity();
ecmult_const_chain_multiply();
ecmult_const_mult_xonly();
}

typedef struct {
Expand Down
48 changes: 44 additions & 4 deletions src/tests_exhaustive.c
Original file line number Diff line number Diff line change
Expand Up @@ -59,6 +59,19 @@ static void random_fe(secp256k1_fe *x) {
}
} while(1);
}

static void random_fe_non_zero(secp256k1_fe *nz) {
int tries = 10;
while (--tries >= 0) {
random_fe(nz);
secp256k1_fe_normalize(nz);
if (!secp256k1_fe_is_zero(nz)) {
break;
}
}
/* Infinitesimal probability of spurious failure here */
CHECK(tries >= 0);
real-or-random marked this conversation as resolved.
Show resolved Hide resolved
}
/** END stolen from tests.c */

static uint32_t num_cores = 1;
Expand Down Expand Up @@ -174,10 +187,37 @@ static void test_exhaustive_ecmult(const secp256k1_ge *group, const secp256k1_ge
secp256k1_ecmult(&tmp, &groupj[r_log], &na, &ng);
ge_equals_gej(&group[(i * r_log + j) % EXHAUSTIVE_TEST_ORDER], &tmp);

if (i > 0) {
secp256k1_ecmult_const(&tmp, &group[i], &ng, 256);
ge_equals_gej(&group[(i * j) % EXHAUSTIVE_TEST_ORDER], &tmp);
}
}
}
}

for (j = 0; j < EXHAUSTIVE_TEST_ORDER; j++) {
for (i = 1; i < EXHAUSTIVE_TEST_ORDER; i++) {
int ret;
secp256k1_gej tmp;
secp256k1_fe xn, xd, tmpf;
secp256k1_scalar ng;

if (skip_section(&iter)) continue;

secp256k1_scalar_set_int(&ng, j);

/* Test secp256k1_ecmult_const. */
secp256k1_ecmult_const(&tmp, &group[i], &ng, 256);
ge_equals_gej(&group[(i * j) % EXHAUSTIVE_TEST_ORDER], &tmp);

if (j != 0) {
/* Test secp256k1_ecmult_const_xonly with all curve X coordinates, and xd=NULL. */
ret = secp256k1_ecmult_const_xonly(&tmpf, &group[i].x, NULL, &ng, 256, 0);
CHECK(ret);
CHECK(secp256k1_fe_equal_var(&tmpf, &group[(i * j) % EXHAUSTIVE_TEST_ORDER].x));

/* Test secp256k1_ecmult_const_xonly with all curve X coordinates, with random xd. */
random_fe_non_zero(&xd);
secp256k1_fe_mul(&xn, &xd, &group[i].x);
ret = secp256k1_ecmult_const_xonly(&tmpf, &xn, &xd, &ng, 256, 0);
CHECK(ret);
CHECK(secp256k1_fe_equal_var(&tmpf, &group[(i * j) % EXHAUSTIVE_TEST_ORDER].x));
}
}
}
Expand Down