-/******************************************************************************
- * *
- * Copyright 2014 Intel Corporation *
- * *
- * Licensed under the Apache License, Version 2.0 (the "License"); *
- * you may not use this file except in compliance with the License. *
- * You may obtain a copy of the License at *
- * *
- * http://www.apache.org/licenses/LICENSE-2.0 *
- * *
- * Unless required by applicable law or agreed to in writing, software *
- * distributed under the License is distributed on an "AS IS" BASIS, *
- * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. *
- * See the License for the specific language governing permissions and *
- * limitations under the License. *
- * *
- ******************************************************************************
- * *
- * Developers and authors: *
- * Shay Gueron (1, 2), and Vlad Krasnov (1) *
- * (1) Intel Corporation, Israel Development Center *
- * (2) University of Haifa *
- * Reference: *
- * S.Gueron and V.Krasnov, "Fast Prime Field Elliptic Curve Cryptography with *
- * 256 Bit Primes" *
- * *
- ******************************************************************************/
+/*
+ * Copyright 2014-2018 The OpenSSL Project Authors. All Rights Reserved.
+ * Copyright (c) 2014, Intel Corporation. All Rights Reserved.
+ * Copyright (c) 2015, CloudFlare, Inc.
+ *
+ * Licensed under the Apache License 2.0 (the "License"). You may not use
+ * this file except in compliance with the License. You can obtain a copy
+ * in the file LICENSE in the source distribution or at
+ * https://www.openssl.org/source/license.html
+ *
+ * Originally written by Shay Gueron (1, 2), and Vlad Krasnov (1, 3)
+ * (1) Intel Corporation, Israel Development Center, Haifa, Israel
+ * (2) University of Haifa, Israel
+ * (3) CloudFlare, Inc.
+ *
+ * Reference:
+ * S.Gueron and V.Krasnov, "Fast Prime Field Elliptic Curve Cryptography with
+ * 256 Bit Primes"
+ */
#include <string.h>
-#include "cryptlib.h"
-#include "internal/bn_int.h"
-#include "ec_lcl.h"
+#include "internal/cryptlib.h"
+#include "crypto/bn.h"
+#include "ec_local.h"
+#include "internal/refcount.h"
#if BN_BITS2 != 64
# define TOBN(hi,lo) lo,hi
typedef P256_POINT_AFFINE PRECOMP256_ROW[64];
/* structure for precomputed multiples of the generator */
-typedef struct ec_pre_comp_st {
+struct nistz256_pre_comp_st {
const EC_GROUP *group; /* Parent EC_GROUP object */
size_t w; /* Window size */
/*
*/
PRECOMP256_ROW *precomp;
void *precomp_storage;
- int references;
-} EC_PRE_COMP;
+ CRYPTO_REF_COUNT references;
+ CRYPTO_RWLOCK *lock;
+};
/* Functions implemented in assembly */
+/*
+ * Most of below mentioned functions *preserve* the property of inputs
+ * being fully reduced, i.e. being in [0, modulus) range. Simply put if
+ * inputs are fully reduced, then output is too. Note that reverse is
+ * not true, in sense that given partially reduced inputs output can be
+ * either, not unlikely reduced. And "most" in first sentence refers to
+ * the fact that given the calculations flow one can tolerate that
+ * addition, 1st function below, produces partially reduced result *if*
+ * multiplications by 2 and 3, which customarily use addition, fully
+ * reduce it. This effectively gives two options: a) addition produces
+ * fully reduced result [as long as inputs are, just like remaining
+ * functions]; b) addition is allowed to produce partially reduced
+ * result, but multiplications by 2 and 3 perform additional reduction
+ * step. Choice between the two can be platform-specific, but it was a)
+ * in all cases so far...
+ */
+/* Modular add: res = a+b mod P */
+void ecp_nistz256_add(BN_ULONG res[P256_LIMBS],
+ const BN_ULONG a[P256_LIMBS],
+ const BN_ULONG b[P256_LIMBS]);
/* Modular mul by 2: res = 2*a mod P */
void ecp_nistz256_mul_by_2(BN_ULONG res[P256_LIMBS],
const BN_ULONG a[P256_LIMBS]);
-/* Modular div by 2: res = a/2 mod P */
-void ecp_nistz256_div_by_2(BN_ULONG res[P256_LIMBS],
- const BN_ULONG a[P256_LIMBS]);
/* Modular mul by 3: res = 3*a mod P */
void ecp_nistz256_mul_by_3(BN_ULONG res[P256_LIMBS],
const BN_ULONG a[P256_LIMBS]);
-/* Modular add: res = a+b mod P */
-void ecp_nistz256_add(BN_ULONG res[P256_LIMBS],
- const BN_ULONG a[P256_LIMBS],
- const BN_ULONG b[P256_LIMBS]);
+
+/* Modular div by 2: res = a/2 mod P */
+void ecp_nistz256_div_by_2(BN_ULONG res[P256_LIMBS],
+ const BN_ULONG a[P256_LIMBS]);
/* Modular sub: res = a-b mod P */
void ecp_nistz256_sub(BN_ULONG res[P256_LIMBS],
const BN_ULONG a[P256_LIMBS],
TOBN(0xffffffff, 0xffffffff), TOBN(0x00000000, 0xfffffffe)
};
-static void *ecp_nistz256_pre_comp_dup(void *);
-static void ecp_nistz256_pre_comp_free(void *);
-static void ecp_nistz256_pre_comp_clear_free(void *);
-static EC_PRE_COMP *ecp_nistz256_pre_comp_new(const EC_GROUP *group);
+static NISTZ256_PRE_COMP *ecp_nistz256_pre_comp_new(const EC_GROUP *group);
/* Precomputed tables for the default generator */
extern const PRECOMP256_ROW ecp_nistz256_precomputed[37];
{
in |= (0 - in);
in = ~in;
- in &= BN_MASK2;
in >>= BN_BITS2 - 1;
return in;
}
return is_zero(res);
}
-static BN_ULONG is_one(const BN_ULONG a[P256_LIMBS])
+static BN_ULONG is_one(const BIGNUM *z)
{
- BN_ULONG res;
-
- res = a[0] ^ ONE[0];
- res |= a[1] ^ ONE[1];
- res |= a[2] ^ ONE[2];
- res |= a[3] ^ ONE[3];
- if (P256_LIMBS == 8) {
- res |= a[4] ^ ONE[4];
- res |= a[5] ^ ONE[5];
- res |= a[6] ^ ONE[6];
+ BN_ULONG res = 0;
+ BN_ULONG *a = bn_get_words(z);
+
+ if (bn_get_top(z) == (P256_LIMBS - P256_LIMBS / 8)) {
+ res = a[0] ^ ONE[0];
+ res |= a[1] ^ ONE[1];
+ res |= a[2] ^ ONE[2];
+ res |= a[3] ^ ONE[3];
+ if (P256_LIMBS == 8) {
+ res |= a[4] ^ ONE[4];
+ res |= a[5] ^ ONE[5];
+ res |= a[6] ^ ONE[6];
+ /*
+ * no check for a[7] (being zero) on 32-bit platforms,
+ * because value of "one" takes only 7 limbs.
+ */
+ }
+ res = is_zero(res);
}
- return is_zero(res);
+ return res;
}
+/*
+ * For reference, this macro is used only when new ecp_nistz256 assembly
+ * module is being developed. For example, configure with
+ * -DECP_NISTZ256_REFERENCE_IMPLEMENTATION and implement only functions
+ * performing simplest arithmetic operations on 256-bit vectors. Then
+ * work on implementation of higher-level functions performing point
+ * operations. Then remove ECP_NISTZ256_REFERENCE_IMPLEMENTATION
+ * and never define it again. (The correct macro denoting presence of
+ * ecp_nistz256 module is ECP_NISTZ256_ASM.)
+ */
#ifndef ECP_NISTZ256_REFERENCE_IMPLEMENTATION
void ecp_nistz256_point_double(P256_POINT *r, const P256_POINT *a);
void ecp_nistz256_point_add(P256_POINT *r,
const BN_ULONG *in2_y = b->Y;
const BN_ULONG *in2_z = b->Z;
- /* We encode infinity as (0,0), which is not on the curve,
- * so it is OK. */
- in1infty = (in1_x[0] | in1_x[1] | in1_x[2] | in1_x[3] |
- in1_y[0] | in1_y[1] | in1_y[2] | in1_y[3]);
+ /*
+ * Infinity in encoded as (,,0)
+ */
+ in1infty = (in1_z[0] | in1_z[1] | in1_z[2] | in1_z[3]);
if (P256_LIMBS == 8)
- in1infty |= (in1_x[4] | in1_x[5] | in1_x[6] | in1_x[7] |
- in1_y[4] | in1_y[5] | in1_y[6] | in1_y[7]);
+ in1infty |= (in1_z[4] | in1_z[5] | in1_z[6] | in1_z[7]);
- in2infty = (in2_x[0] | in2_x[1] | in2_x[2] | in2_x[3] |
- in2_y[0] | in2_y[1] | in2_y[2] | in2_y[3]);
+ in2infty = (in2_z[0] | in2_z[1] | in2_z[2] | in2_z[3]);
if (P256_LIMBS == 8)
- in2infty |= (in2_x[4] | in2_x[5] | in2_x[6] | in2_x[7] |
- in2_y[4] | in2_y[5] | in2_y[6] | in2_y[7]);
+ in2infty |= (in2_z[4] | in2_z[5] | in2_z[6] | in2_z[7]);
in1infty = is_zero(in1infty);
in2infty = is_zero(in2infty);
const BN_ULONG *in2_y = b->Y;
/*
- * In affine representation we encode infty as (0,0), which is not on the
- * curve, so it is OK
+ * Infinity in encoded as (,,0)
*/
- in1infty = (in1_x[0] | in1_x[1] | in1_x[2] | in1_x[3] |
- in1_y[0] | in1_y[1] | in1_y[2] | in1_y[3]);
+ in1infty = (in1_z[0] | in1_z[1] | in1_z[2] | in1_z[3]);
if (P256_LIMBS == 8)
- in1infty |= (in1_x[4] | in1_x[5] | in1_x[6] | in1_x[7] |
- in1_y[4] | in1_y[5] | in1_y[6] | in1_y[7]);
+ in1infty |= (in1_z[4] | in1_z[5] | in1_z[6] | in1_z[7]);
+ /*
+ * In affine representation we encode infinity as (0,0), which is
+ * not on the curve, so it is OK
+ */
in2infty = (in2_x[0] | in2_x[1] | in2_x[2] | in2_x[3] |
in2_y[0] | in2_y[1] | in2_y[2] | in2_y[3]);
if (P256_LIMBS == 8)
* ecp_nistz256_bignum_to_field_elem copies the contents of |in| to |out| and
* returns one if it fits. Otherwise it returns zero.
*/
-static int ecp_nistz256_bignum_to_field_elem(BN_ULONG out[P256_LIMBS],
- const BIGNUM *in)
+__owur static int ecp_nistz256_bignum_to_field_elem(BN_ULONG out[P256_LIMBS],
+ const BIGNUM *in)
{
return bn_copy_words(out, in, P256_LIMBS);
}
/* r = sum(scalar[i]*point[i]) */
-static void ecp_nistz256_windowed_mul(const EC_GROUP *group,
- P256_POINT *r,
- const BIGNUM **scalar,
- const EC_POINT **point,
- size_t num, BN_CTX *ctx)
+__owur static int ecp_nistz256_windowed_mul(const EC_GROUP *group,
+ P256_POINT *r,
+ const BIGNUM **scalar,
+ const EC_POINT **point,
+ size_t num, BN_CTX *ctx)
{
size_t i;
- int j;
+ int j, ret = 0;
unsigned int idx;
unsigned char (*p_str)[33] = NULL;
const unsigned int window_size = 5;
}
/*
- * row[0] is implicitly (0,0,0) (the point at infinity), therefore it
- * is not stored. All other values are actually stored with an offset
- * of -1 in table.
+ * row[0] is implicitly (0,0,0) (the point at infinity), therefore it
+ * is not stored. All other values are actually stored with an offset
+ * of -1 in table.
*/
ecp_nistz256_scatter_w5 (row, &temp[0], 1);
ecp_nistz256_point_add(r, r, &temp[0]);
}
+ ret = 1;
err:
- if (table_storage)
- OPENSSL_free(table_storage);
- if (p_str)
- OPENSSL_free(p_str);
- if (scalars)
- OPENSSL_free(scalars);
+ OPENSSL_free(table_storage);
+ OPENSSL_free(p_str);
+ OPENSSL_free(scalars);
+ return ret;
}
/* Coordinates of G, for which we have precomputed tables */
-const static BN_ULONG def_xG[P256_LIMBS] = {
+static const BN_ULONG def_xG[P256_LIMBS] = {
TOBN(0x79e730d4, 0x18a9143c), TOBN(0x75ba95fc, 0x5fedb601),
TOBN(0x79fb732b, 0x77622510), TOBN(0x18905f76, 0xa53755c6)
};
-const static BN_ULONG def_yG[P256_LIMBS] = {
+static const BN_ULONG def_yG[P256_LIMBS] = {
TOBN(0xddf25357, 0xce95560a), TOBN(0x8b4ab8e4, 0xba19e45c),
TOBN(0xd2e88688, 0xdd21f325), TOBN(0x8571ff18, 0x25885d85)
};
{
return (bn_get_top(generator->X) == P256_LIMBS) &&
(bn_get_top(generator->Y) == P256_LIMBS) &&
- (bn_get_top(generator->Z) == (P256_LIMBS - P256_LIMBS / 8)) &&
is_equal(bn_get_words(generator->X), def_xG) &&
is_equal(bn_get_words(generator->Y), def_yG) &&
- is_one(bn_get_words(generator->Z));
+ is_one(generator->Z);
}
-static int ecp_nistz256_mult_precompute(EC_GROUP *group, BN_CTX *ctx)
+__owur static int ecp_nistz256_mult_precompute(EC_GROUP *group, BN_CTX *ctx)
{
/*
* We precompute a table for a Booth encoded exponent (wNAF) based
* implicit value of infinity at index zero. We use window of size 7, and
* therefore require ceil(256/7) = 37 tables.
*/
- BIGNUM *order;
+ const BIGNUM *order;
EC_POINT *P = NULL, *T = NULL;
const EC_POINT *generator;
- EC_PRE_COMP *pre_comp;
+ NISTZ256_PRE_COMP *pre_comp;
+ BN_CTX *new_ctx = NULL;
int i, j, k, ret = 0;
size_t w;
PRECOMP256_ROW *preComputedTable = NULL;
unsigned char *precomp_storage = NULL;
- /* if there is an old EC_PRE_COMP object, throw it away */
- EC_EX_DATA_free_data(&group->extra_data, ecp_nistz256_pre_comp_dup,
- ecp_nistz256_pre_comp_free,
- ecp_nistz256_pre_comp_clear_free);
-
+ /* if there is an old NISTZ256_PRE_COMP object, throw it away */
+ EC_pre_comp_free(group);
generator = EC_GROUP_get0_generator(group);
if (generator == NULL) {
ECerr(EC_F_ECP_NISTZ256_MULT_PRECOMPUTE, EC_R_UNDEFINED_GENERATOR);
return 0;
if (ctx == NULL) {
- ctx = BN_CTX_new();
+ ctx = new_ctx = BN_CTX_new_ex(group->libctx);
if (ctx == NULL)
goto err;
}
BN_CTX_start(ctx);
- order = BN_CTX_get(ctx);
+ order = EC_GROUP_get0_order(group);
if (order == NULL)
goto err;
- if (!EC_GROUP_get_order(group, order, ctx))
- goto err;
-
if (BN_is_zero(order)) {
ECerr(EC_F_ECP_NISTZ256_MULT_PRECOMPUTE, EC_R_UNKNOWN_ORDER);
goto err;
P = EC_POINT_new(group);
T = EC_POINT_new(group);
+ if (P == NULL || T == NULL)
+ goto err;
/*
* The zero entry is implicitly infinity, and we skip it, storing other
* values with -1 offset.
*/
- EC_POINT_copy(T, generator);
+ if (!EC_POINT_copy(T, generator))
+ goto err;
for (k = 0; k < 64; k++) {
- EC_POINT_copy(P, T);
+ if (!EC_POINT_copy(P, T))
+ goto err;
for (j = 0; j < 37; j++) {
P256_POINT_AFFINE temp;
/*
- * It would be faster to use ec_GFp_simple_points_make_affine and
+ * It would be faster to use EC_POINTs_make_affine and
* make multiple points affine at the same time.
*/
- ec_GFp_simple_make_affine(group, P, ctx);
- ecp_nistz256_bignum_to_field_elem(temp.X, P->X);
- ecp_nistz256_bignum_to_field_elem(temp.Y, P->Y);
+ if (!EC_POINT_make_affine(group, P, ctx))
+ goto err;
+ if (!ecp_nistz256_bignum_to_field_elem(temp.X, P->X) ||
+ !ecp_nistz256_bignum_to_field_elem(temp.Y, P->Y)) {
+ ECerr(EC_F_ECP_NISTZ256_MULT_PRECOMPUTE,
+ EC_R_COORDINATES_OUT_OF_RANGE);
+ goto err;
+ }
ecp_nistz256_scatter_w7(preComputedTable[j], &temp, k);
- for (i = 0; i < 7; i++)
- ec_GFp_simple_dbl(group, P, P, ctx);
+ for (i = 0; i < 7; i++) {
+ if (!EC_POINT_dbl(group, P, P, ctx))
+ goto err;
+ }
}
- ec_GFp_simple_add(group, T, T, generator, ctx);
+ if (!EC_POINT_add(group, T, T, generator, ctx))
+ goto err;
}
pre_comp->group = group;
pre_comp->w = w;
pre_comp->precomp = preComputedTable;
pre_comp->precomp_storage = precomp_storage;
-
precomp_storage = NULL;
-
- if (!EC_EX_DATA_set_data(&group->extra_data, pre_comp,
- ecp_nistz256_pre_comp_dup,
- ecp_nistz256_pre_comp_free,
- ecp_nistz256_pre_comp_clear_free)) {
- goto err;
- }
-
+ SETPRECOMP(group, nistz256, pre_comp);
pre_comp = NULL;
-
ret = 1;
err:
- if (ctx != NULL)
- BN_CTX_end(ctx);
- ecp_nistz256_pre_comp_free(pre_comp);
- if (precomp_storage)
- OPENSSL_free(precomp_storage);
+ BN_CTX_end(ctx);
+ BN_CTX_free(new_ctx);
+
+ EC_nistz256_pre_comp_free(pre_comp);
+ OPENSSL_free(precomp_storage);
EC_POINT_free(P);
EC_POINT_free(T);
return ret;
*/
#if defined(ECP_NISTZ256_AVX2)
# if !(defined(__x86_64) || defined(__x86_64__) || \
- defined(_M_AMD64) || defined(_MX64)) || \
+ defined(_M_AMD64) || defined(_M_X64)) || \
!(defined(__GNUC__) || defined(_MSC_VER)) /* this is for ALIGN32 */
# undef ECP_NISTZ256_AVX2
# else
# endif
#endif
-static int ecp_nistz256_set_from_affine(EC_POINT *out, const EC_GROUP *group,
- const P256_POINT_AFFINE *in,
- BN_CTX *ctx)
+__owur static int ecp_nistz256_set_from_affine(EC_POINT *out, const EC_GROUP *group,
+ const P256_POINT_AFFINE *in,
+ BN_CTX *ctx)
{
- BIGNUM *x, *y;
- BN_ULONG d_x[P256_LIMBS], d_y[P256_LIMBS];
int ret = 0;
- x = BN_new();
- if (!x)
- return 0;
- y = BN_new();
- if (!y) {
- BN_free(x);
- return 0;
- }
- memcpy(d_x, in->X, sizeof(d_x));
- bn_set_static_words(x, d_x, P256_LIMBS);
-
- memcpy(d_y, in->Y, sizeof(d_y));
- bn_set_static_words(y, d_y, P256_LIMBS);
-
- ret = EC_POINT_set_affine_coordinates_GFp(group, out, x, y, ctx);
-
- if (x)
- BN_free(x);
- if (y)
- BN_free(y);
+ if ((ret = bn_set_words(out->X, in->X, P256_LIMBS))
+ && (ret = bn_set_words(out->Y, in->Y, P256_LIMBS))
+ && (ret = bn_set_words(out->Z, ONE, P256_LIMBS)))
+ out->Z_is_one = 1;
return ret;
}
/* r = scalar*G + sum(scalars[i]*points[i]) */
-static int ecp_nistz256_points_mul(const EC_GROUP *group,
- EC_POINT *r,
- const BIGNUM *scalar,
- size_t num,
- const EC_POINT *points[],
- const BIGNUM *scalars[], BN_CTX *ctx)
+__owur static int ecp_nistz256_points_mul(const EC_GROUP *group,
+ EC_POINT *r,
+ const BIGNUM *scalar,
+ size_t num,
+ const EC_POINT *points[],
+ const BIGNUM *scalars[], BN_CTX *ctx)
{
int i = 0, ret = 0, no_precomp_for_generator = 0, p_is_infinity = 0;
- size_t j;
unsigned char p_str[33] = { 0 };
const PRECOMP256_ROW *preComputedTable = NULL;
- const EC_PRE_COMP *pre_comp = NULL;
+ const NISTZ256_PRE_COMP *pre_comp = NULL;
const EC_POINT *generator = NULL;
+ const BIGNUM **new_scalars = NULL;
+ const EC_POINT **new_points = NULL;
unsigned int idx = 0;
const unsigned int window_size = 7;
const unsigned int mask = (1 << (window_size + 1)) - 1;
return 0;
}
- if (group->meth != r->meth) {
- ECerr(EC_F_ECP_NISTZ256_POINTS_MUL, EC_R_INCOMPATIBLE_OBJECTS);
- return 0;
- }
- if ((scalar == NULL) && (num == 0))
- return EC_POINT_set_to_infinity(group, r);
-
- for (j = 0; j < num; j++) {
- if (group->meth != points[j]->meth) {
- ECerr(EC_F_ECP_NISTZ256_POINTS_MUL, EC_R_INCOMPATIBLE_OBJECTS);
- return 0;
- }
- }
-
- /* Need 256 bits for space for all coordinates. */
- bn_wexpand(r->X, P256_LIMBS);
- bn_wexpand(r->Y, P256_LIMBS);
- bn_wexpand(r->Z, P256_LIMBS);
- bn_set_top(r->X, P256_LIMBS);
- bn_set_top(r->Y, P256_LIMBS);
- bn_set_top(r->Z, P256_LIMBS);
+ BN_CTX_start(ctx);
if (scalar) {
generator = EC_GROUP_get0_generator(group);
}
/* look if we can use precomputed multiples of generator */
- pre_comp =
- EC_EX_DATA_get_data(group->extra_data, ecp_nistz256_pre_comp_dup,
- ecp_nistz256_pre_comp_free,
- ecp_nistz256_pre_comp_clear_free);
+ pre_comp = group->pre_comp.nistz256;
if (pre_comp) {
/*
if (pre_comp_generator == NULL)
goto err;
+ ecp_nistz256_gather_w7(&p.a, pre_comp->precomp[0], 1);
if (!ecp_nistz256_set_from_affine(pre_comp_generator,
- group, pre_comp->precomp[0],
- ctx))
+ group, &p.a, ctx)) {
+ EC_POINT_free(pre_comp_generator);
goto err;
+ }
if (0 == EC_POINT_cmp(group, generator, pre_comp_generator, ctx))
preComputedTable = (const PRECOMP256_ROW *)pre_comp->precomp;
} else
#endif
{
+ BN_ULONG infty;
+
/* First window */
wvalue = (p_str[0] << 1) & mask;
idx += window_size;
ecp_nistz256_neg(p.p.Z, p.p.Y);
copy_conditional(p.p.Y, p.p.Z, wvalue & 1);
- memcpy(p.p.Z, ONE, sizeof(ONE));
+ /*
+ * Since affine infinity is encoded as (0,0) and
+ * Jacobian ias (,,0), we need to harmonize them
+ * by assigning "one" or zero to Z.
+ */
+ infty = (p.p.X[0] | p.p.X[1] | p.p.X[2] | p.p.X[3] |
+ p.p.Y[0] | p.p.Y[1] | p.p.Y[2] | p.p.Y[3]);
+ if (P256_LIMBS == 8)
+ infty |= (p.p.X[4] | p.p.X[5] | p.p.X[6] | p.p.X[7] |
+ p.p.Y[4] | p.p.Y[5] | p.p.Y[6] | p.p.Y[7]);
+
+ infty = 0 - is_zero(infty);
+ infty = ~infty;
+
+ p.p.Z[0] = ONE[0] & infty;
+ p.p.Z[1] = ONE[1] & infty;
+ p.p.Z[2] = ONE[2] & infty;
+ p.p.Z[3] = ONE[3] & infty;
+ if (P256_LIMBS == 8) {
+ p.p.Z[4] = ONE[4] & infty;
+ p.p.Z[5] = ONE[5] & infty;
+ p.p.Z[6] = ONE[6] & infty;
+ p.p.Z[7] = ONE[7] & infty;
+ }
for (i = 1; i < 37; i++) {
unsigned int off = (idx - 1) / 8;
* Without a precomputed table for the generator, it has to be
* handled like a normal point.
*/
- const BIGNUM **new_scalars;
- const EC_POINT **new_points;
-
new_scalars = OPENSSL_malloc((num + 1) * sizeof(BIGNUM *));
- if (!new_scalars) {
+ if (new_scalars == NULL) {
ECerr(EC_F_ECP_NISTZ256_POINTS_MUL, ERR_R_MALLOC_FAILURE);
- return 0;
+ goto err;
}
new_points = OPENSSL_malloc((num + 1) * sizeof(EC_POINT *));
- if (!new_points) {
- OPENSSL_free(new_scalars);
+ if (new_points == NULL) {
ECerr(EC_F_ECP_NISTZ256_POINTS_MUL, ERR_R_MALLOC_FAILURE);
- return 0;
+ goto err;
}
memcpy(new_scalars, scalars, num * sizeof(BIGNUM *));
if (p_is_infinity)
out = &p.p;
- ecp_nistz256_windowed_mul(group, out, scalars, points, num, ctx);
+ if (!ecp_nistz256_windowed_mul(group, out, scalars, points, num, ctx))
+ goto err;
if (!p_is_infinity)
ecp_nistz256_point_add(&p.p, &p.p, out);
}
- if (no_precomp_for_generator) {
- OPENSSL_free(points);
- OPENSSL_free(scalars);
- }
-
- bn_set_data(r->X, p.p.X, sizeof(p.p.X));
- bn_set_data(r->Y, p.p.Y, sizeof(p.p.Y));
- bn_set_data(r->Z, p.p.Z, sizeof(p.p.Z));
/* Not constant-time, but we're only operating on the public output. */
- bn_correct_top(r->X);
- bn_correct_top(r->Y);
- bn_correct_top(r->Z);
- r->Z_is_one = is_one(p.p.Z);
+ if (!bn_set_words(r->X, p.p.X, P256_LIMBS) ||
+ !bn_set_words(r->Y, p.p.Y, P256_LIMBS) ||
+ !bn_set_words(r->Z, p.p.Z, P256_LIMBS)) {
+ goto err;
+ }
+ r->Z_is_one = is_one(r->Z) & 1;
ret = 1;
- err:
+err:
+ BN_CTX_end(ctx);
+ OPENSSL_free(new_points);
+ OPENSSL_free(new_scalars);
return ret;
}
-static int ecp_nistz256_get_affine(const EC_GROUP *group,
- const EC_POINT *point,
- BIGNUM *x, BIGNUM *y, BN_CTX *ctx)
+__owur static int ecp_nistz256_get_affine(const EC_GROUP *group,
+ const EC_POINT *point,
+ BIGNUM *x, BIGNUM *y, BN_CTX *ctx)
{
BN_ULONG z_inv2[P256_LIMBS];
BN_ULONG z_inv3[P256_LIMBS];
BN_ULONG x_aff[P256_LIMBS];
BN_ULONG y_aff[P256_LIMBS];
BN_ULONG point_x[P256_LIMBS], point_y[P256_LIMBS], point_z[P256_LIMBS];
+ BN_ULONG x_ret[P256_LIMBS], y_ret[P256_LIMBS];
if (EC_POINT_is_at_infinity(group, point)) {
ECerr(EC_F_ECP_NISTZ256_GET_AFFINE, EC_R_POINT_AT_INFINITY);
ecp_nistz256_mul_mont(x_aff, z_inv2, point_x);
if (x != NULL) {
- bn_wexpand(x, P256_LIMBS);
- bn_set_top(x, P256_LIMBS);
- ecp_nistz256_from_mont(bn_get_words(x), x_aff);
- bn_correct_top(x);
+ ecp_nistz256_from_mont(x_ret, x_aff);
+ if (!bn_set_words(x, x_ret, P256_LIMBS))
+ return 0;
}
if (y != NULL) {
ecp_nistz256_mul_mont(z_inv3, z_inv3, z_inv2);
ecp_nistz256_mul_mont(y_aff, z_inv3, point_y);
- bn_wexpand(y, P256_LIMBS);
- bn_set_top(y, P256_LIMBS);
- ecp_nistz256_from_mont(bn_get_words(y), y_aff);
- bn_correct_top(y);
+ ecp_nistz256_from_mont(y_ret, y_aff);
+ if (!bn_set_words(y, y_ret, P256_LIMBS))
+ return 0;
}
return 1;
}
-static EC_PRE_COMP *ecp_nistz256_pre_comp_new(const EC_GROUP *group)
+static NISTZ256_PRE_COMP *ecp_nistz256_pre_comp_new(const EC_GROUP *group)
{
- EC_PRE_COMP *ret = NULL;
+ NISTZ256_PRE_COMP *ret = NULL;
if (!group)
return NULL;
- ret = (EC_PRE_COMP *)OPENSSL_malloc(sizeof(EC_PRE_COMP));
+ ret = OPENSSL_zalloc(sizeof(*ret));
- if (!ret) {
+ if (ret == NULL) {
ECerr(EC_F_ECP_NISTZ256_PRE_COMP_NEW, ERR_R_MALLOC_FAILURE);
return ret;
}
ret->group = group;
ret->w = 6; /* default */
- ret->precomp = NULL;
- ret->precomp_storage = NULL;
ret->references = 1;
- return ret;
-}
-
-static void *ecp_nistz256_pre_comp_dup(void *src_)
-{
- EC_PRE_COMP *src = src_;
-
- /* no need to actually copy, these objects never change! */
- CRYPTO_add(&src->references, 1, CRYPTO_LOCK_EC_PRE_COMP);
- return src_;
+ ret->lock = CRYPTO_THREAD_lock_new();
+ if (ret->lock == NULL) {
+ ECerr(EC_F_ECP_NISTZ256_PRE_COMP_NEW, ERR_R_MALLOC_FAILURE);
+ OPENSSL_free(ret);
+ return NULL;
+ }
+ return ret;
}
-static void ecp_nistz256_pre_comp_free(void *pre_)
+NISTZ256_PRE_COMP *EC_nistz256_pre_comp_dup(NISTZ256_PRE_COMP *p)
{
int i;
- EC_PRE_COMP *pre = pre_;
-
- if (!pre)
- return;
-
- i = CRYPTO_add(&pre->references, -1, CRYPTO_LOCK_EC_PRE_COMP);
- if (i > 0)
- return;
-
- if (pre->precomp_storage)
- OPENSSL_free(pre->precomp_storage);
-
- OPENSSL_free(pre);
+ if (p != NULL)
+ CRYPTO_UP_REF(&p->references, &i, p->lock);
+ return p;
}
-static void ecp_nistz256_pre_comp_clear_free(void *pre_)
+void EC_nistz256_pre_comp_free(NISTZ256_PRE_COMP *pre)
{
int i;
- EC_PRE_COMP *pre = pre_;
- if (!pre)
+ if (pre == NULL)
return;
- i = CRYPTO_add(&pre->references, -1, CRYPTO_LOCK_EC_PRE_COMP);
+ CRYPTO_DOWN_REF(&pre->references, &i, pre->lock);
+ REF_PRINT_COUNT("EC_nistz256", pre);
if (i > 0)
return;
+ REF_ASSERT_ISNT(i < 0);
- if (pre->precomp_storage) {
- OPENSSL_cleanse(pre->precomp,
- 32 * sizeof(unsigned char) * (1 << pre->w) * 2 * 37);
- OPENSSL_free(pre->precomp_storage);
- }
- OPENSSL_cleanse(pre, sizeof *pre);
+ OPENSSL_free(pre->precomp_storage);
+ CRYPTO_THREAD_lock_free(pre->lock);
OPENSSL_free(pre);
}
+
static int ecp_nistz256_window_have_precompute_mult(const EC_GROUP *group)
{
/* There is a hard-coded table for the default generator. */
const EC_POINT *generator = EC_GROUP_get0_generator(group);
+
if (generator != NULL && ecp_nistz256_is_affine_G(generator)) {
/* There is a hard-coded table for the default generator. */
return 1;
}
- return EC_EX_DATA_get_data(group->extra_data, ecp_nistz256_pre_comp_dup,
- ecp_nistz256_pre_comp_free,
- ecp_nistz256_pre_comp_clear_free) != NULL;
+ return HAVEPRECOMP(group, nistz256);
+}
+
+#if defined(__x86_64) || defined(__x86_64__) || \
+ defined(_M_AMD64) || defined(_M_X64) || \
+ defined(__powerpc64__) || defined(_ARCH_PP64) || \
+ defined(__aarch64__)
+/*
+ * Montgomery mul modulo Order(P): res = a*b*2^-256 mod Order(P)
+ */
+void ecp_nistz256_ord_mul_mont(BN_ULONG res[P256_LIMBS],
+ const BN_ULONG a[P256_LIMBS],
+ const BN_ULONG b[P256_LIMBS]);
+void ecp_nistz256_ord_sqr_mont(BN_ULONG res[P256_LIMBS],
+ const BN_ULONG a[P256_LIMBS],
+ BN_ULONG rep);
+
+static int ecp_nistz256_inv_mod_ord(const EC_GROUP *group, BIGNUM *r,
+ const BIGNUM *x, BN_CTX *ctx)
+{
+ /* RR = 2^512 mod ord(p256) */
+ static const BN_ULONG RR[P256_LIMBS] = {
+ TOBN(0x83244c95,0xbe79eea2), TOBN(0x4699799c,0x49bd6fa6),
+ TOBN(0x2845b239,0x2b6bec59), TOBN(0x66e12d94,0xf3d95620)
+ };
+ /* The constant 1 (unlike ONE that is one in Montgomery representation) */
+ static const BN_ULONG one[P256_LIMBS] = {
+ TOBN(0,1), TOBN(0,0), TOBN(0,0), TOBN(0,0)
+ };
+ /*
+ * We don't use entry 0 in the table, so we omit it and address
+ * with -1 offset.
+ */
+ BN_ULONG table[15][P256_LIMBS];
+ BN_ULONG out[P256_LIMBS], t[P256_LIMBS];
+ int i, ret = 0;
+ enum {
+ i_1 = 0, i_10, i_11, i_101, i_111, i_1010, i_1111,
+ i_10101, i_101010, i_101111, i_x6, i_x8, i_x16, i_x32
+ };
+
+ /*
+ * Catch allocation failure early.
+ */
+ if (bn_wexpand(r, P256_LIMBS) == NULL) {
+ ECerr(EC_F_ECP_NISTZ256_INV_MOD_ORD, ERR_R_BN_LIB);
+ goto err;
+ }
+
+ if ((BN_num_bits(x) > 256) || BN_is_negative(x)) {
+ BIGNUM *tmp;
+
+ if ((tmp = BN_CTX_get(ctx)) == NULL
+ || !BN_nnmod(tmp, x, group->order, ctx)) {
+ ECerr(EC_F_ECP_NISTZ256_INV_MOD_ORD, ERR_R_BN_LIB);
+ goto err;
+ }
+ x = tmp;
+ }
+
+ if (!ecp_nistz256_bignum_to_field_elem(t, x)) {
+ ECerr(EC_F_ECP_NISTZ256_INV_MOD_ORD, EC_R_COORDINATES_OUT_OF_RANGE);
+ goto err;
+ }
+
+ ecp_nistz256_ord_mul_mont(table[0], t, RR);
+#if 0
+ /*
+ * Original sparse-then-fixed-window algorithm, retained for reference.
+ */
+ for (i = 2; i < 16; i += 2) {
+ ecp_nistz256_ord_sqr_mont(table[i-1], table[i/2-1], 1);
+ ecp_nistz256_ord_mul_mont(table[i], table[i-1], table[0]);
+ }
+
+ /*
+ * The top 128bit of the exponent are highly redudndant, so we
+ * perform an optimized flow
+ */
+ ecp_nistz256_ord_sqr_mont(t, table[15-1], 4); /* f0 */
+ ecp_nistz256_ord_mul_mont(t, t, table[15-1]); /* ff */
+
+ ecp_nistz256_ord_sqr_mont(out, t, 8); /* ff00 */
+ ecp_nistz256_ord_mul_mont(out, out, t); /* ffff */
+
+ ecp_nistz256_ord_sqr_mont(t, out, 16); /* ffff0000 */
+ ecp_nistz256_ord_mul_mont(t, t, out); /* ffffffff */
+
+ ecp_nistz256_ord_sqr_mont(out, t, 64); /* ffffffff0000000000000000 */
+ ecp_nistz256_ord_mul_mont(out, out, t); /* ffffffff00000000ffffffff */
+
+ ecp_nistz256_ord_sqr_mont(out, out, 32); /* ffffffff00000000ffffffff00000000 */
+ ecp_nistz256_ord_mul_mont(out, out, t); /* ffffffff00000000ffffffffffffffff */
+
+ /*
+ * The bottom 128 bit of the exponent are processed with fixed 4-bit window
+ */
+ for(i = 0; i < 32; i++) {
+ /* expLo - the low 128 bits of the exponent we use (ord(p256) - 2),
+ * split into nibbles */
+ static const unsigned char expLo[32] = {
+ 0xb,0xc,0xe,0x6,0xf,0xa,0xa,0xd,0xa,0x7,0x1,0x7,0x9,0xe,0x8,0x4,
+ 0xf,0x3,0xb,0x9,0xc,0xa,0xc,0x2,0xf,0xc,0x6,0x3,0x2,0x5,0x4,0xf
+ };
+
+ ecp_nistz256_ord_sqr_mont(out, out, 4);
+ /* The exponent is public, no need in constant-time access */
+ ecp_nistz256_ord_mul_mont(out, out, table[expLo[i]-1]);
+ }
+#else
+ /*
+ * https://briansmith.org/ecc-inversion-addition-chains-01#p256_scalar_inversion
+ *
+ * Even though this code path spares 12 squarings, 4.5%, and 13
+ * multiplications, 25%, on grand scale sign operation is not that
+ * much faster, not more that 2%...
+ */
+
+ /* pre-calculate powers */
+ ecp_nistz256_ord_sqr_mont(table[i_10], table[i_1], 1);
+
+ ecp_nistz256_ord_mul_mont(table[i_11], table[i_1], table[i_10]);
+
+ ecp_nistz256_ord_mul_mont(table[i_101], table[i_11], table[i_10]);
+
+ ecp_nistz256_ord_mul_mont(table[i_111], table[i_101], table[i_10]);
+
+ ecp_nistz256_ord_sqr_mont(table[i_1010], table[i_101], 1);
+
+ ecp_nistz256_ord_mul_mont(table[i_1111], table[i_1010], table[i_101]);
+
+ ecp_nistz256_ord_sqr_mont(table[i_10101], table[i_1010], 1);
+ ecp_nistz256_ord_mul_mont(table[i_10101], table[i_10101], table[i_1]);
+
+ ecp_nistz256_ord_sqr_mont(table[i_101010], table[i_10101], 1);
+
+ ecp_nistz256_ord_mul_mont(table[i_101111], table[i_101010], table[i_101]);
+
+ ecp_nistz256_ord_mul_mont(table[i_x6], table[i_101010], table[i_10101]);
+
+ ecp_nistz256_ord_sqr_mont(table[i_x8], table[i_x6], 2);
+ ecp_nistz256_ord_mul_mont(table[i_x8], table[i_x8], table[i_11]);
+
+ ecp_nistz256_ord_sqr_mont(table[i_x16], table[i_x8], 8);
+ ecp_nistz256_ord_mul_mont(table[i_x16], table[i_x16], table[i_x8]);
+
+ ecp_nistz256_ord_sqr_mont(table[i_x32], table[i_x16], 16);
+ ecp_nistz256_ord_mul_mont(table[i_x32], table[i_x32], table[i_x16]);
+
+ /* calculations */
+ ecp_nistz256_ord_sqr_mont(out, table[i_x32], 64);
+ ecp_nistz256_ord_mul_mont(out, out, table[i_x32]);
+
+ for (i = 0; i < 27; i++) {
+ static const struct { unsigned char p, i; } chain[27] = {
+ { 32, i_x32 }, { 6, i_101111 }, { 5, i_111 },
+ { 4, i_11 }, { 5, i_1111 }, { 5, i_10101 },
+ { 4, i_101 }, { 3, i_101 }, { 3, i_101 },
+ { 5, i_111 }, { 9, i_101111 }, { 6, i_1111 },
+ { 2, i_1 }, { 5, i_1 }, { 6, i_1111 },
+ { 5, i_111 }, { 4, i_111 }, { 5, i_111 },
+ { 5, i_101 }, { 3, i_11 }, { 10, i_101111 },
+ { 2, i_11 }, { 5, i_11 }, { 5, i_11 },
+ { 3, i_1 }, { 7, i_10101 }, { 6, i_1111 }
+ };
+
+ ecp_nistz256_ord_sqr_mont(out, out, chain[i].p);
+ ecp_nistz256_ord_mul_mont(out, out, table[chain[i].i]);
+ }
+#endif
+ ecp_nistz256_ord_mul_mont(out, out, one);
+
+ /*
+ * Can't fail, but check return code to be consistent anyway.
+ */
+ if (!bn_set_words(r, out, P256_LIMBS))
+ goto err;
+
+ ret = 1;
+err:
+ return ret;
}
+#else
+# define ecp_nistz256_inv_mod_ord NULL
+#endif
const EC_METHOD *EC_GFp_nistz256_method(void)
{
ec_GFp_mont_group_set_curve,
ec_GFp_simple_group_get_curve,
ec_GFp_simple_group_get_degree,
+ ec_group_simple_order_bits,
ec_GFp_simple_group_check_discriminant,
ec_GFp_simple_point_init,
ec_GFp_simple_point_finish,
ec_GFp_mont_field_mul,
ec_GFp_mont_field_sqr,
0, /* field_div */
+ ec_GFp_mont_field_inv,
ec_GFp_mont_field_encode,
ec_GFp_mont_field_decode,
- ec_GFp_mont_field_set_to_one
+ ec_GFp_mont_field_set_to_one,
+ ec_key_simple_priv2oct,
+ ec_key_simple_oct2priv,
+ 0, /* set private */
+ ec_key_simple_generate_key,
+ ec_key_simple_check_key,
+ ec_key_simple_generate_public_key,
+ 0, /* keycopy */
+ 0, /* keyfinish */
+ ecdh_simple_compute_key,
+ ecdsa_simple_sign_setup,
+ ecdsa_simple_sign_sig,
+ ecdsa_simple_verify_sig,
+ ecp_nistz256_inv_mod_ord, /* can be #define-d NULL */
+ 0, /* blind_coordinates */
+ 0, /* ladder_pre */
+ 0, /* ladder_step */
+ 0 /* ladder_post */
};
return &ret;