Simplify ec_GFp_nistp256_points_mul logic. Passing in an array of scalars was removed some time ago, but a few remnants of it remain. Change-Id: Ia51dcf1f85116ec663e657cc8dbef7f23ffa2edb Reviewed-on: https://boringssl-review.googlesource.com/13055 Commit-Queue: David Benjamin <davidben@google.com> Reviewed-by: Adam Langley <agl@google.com>
diff --git a/crypto/ec/p256-64.c b/crypto/ec/p256-64.c index 38fee45..0f32c2e 100644 --- a/crypto/ec/p256-64.c +++ b/crypto/ec/p256-64.c
@@ -1427,14 +1427,12 @@ } /* Interleaved point multiplication using precomputed point multiples: The - * small point multiples 0*P, 1*P, ..., 17*P are in pre_comp[], the scalars - * in scalars[]. If g_scalar is non-NULL, we also add this multiple of the - * generator, using certain (large) precomputed multiples in g_pre_comp. + * 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(felem x_out, felem y_out, felem z_out, - const felem_bytearray scalars[], - const size_t num_points, const u8 *g_scalar, - const smallfelem pre_comp[][17][3]) { +static void batch_mul(felem x_out, felem y_out, felem z_out, const u8 *p_scalar, + const u8 *g_scalar, const smallfelem p_pre_comp[17][3]) { felem nq[3], ftmp; smallfelem tmp[3]; u64 bits; @@ -1443,12 +1441,12 @@ /* set nq to the point at infinity */ OPENSSL_memset(nq, 0, 3 * sizeof(felem)); - /* Loop over all scalars msb-to-lsb, interleaving additions of multiples - * of the generator (two in each of the last 32 rounds) and additions of - * other points multiples (every 5th round). */ + /* 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 = num_points != 0 ? 255 : 31; + size_t i = p_scalar != NULL ? 255 : 31; for (;;) { /* double */ if (!skip) { @@ -1487,34 +1485,30 @@ } /* do other additions every 5 doublings */ - if (num_points != 0 && i % 5 == 0) { - /* loop over all scalars */ - size_t num; - for (num = 0; num < num_points; ++num) { - bits = get_bit(scalars[num], i + 4) << 5; - bits |= get_bit(scalars[num], i + 3) << 4; - bits |= get_bit(scalars[num], i + 2) << 3; - bits |= get_bit(scalars[num], i + 1) << 2; - bits |= get_bit(scalars[num], i) << 1; - bits |= get_bit(scalars[num], i - 1); - ec_GFp_nistp_recode_scalar_bits(&sign, &digit, bits); + 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, pre_comp[num], tmp); - smallfelem_neg(ftmp, tmp[1]); /* (X, -Y, Z) is the negative - * point */ - copy_small_conditional(ftmp, tmp[1], (((limb)sign) - 1)); - felem_contract(tmp[1], ftmp); + /* select the point to add or subtract, in constant time. */ + select_point(digit, 17, p_pre_comp, tmp); + smallfelem_neg(ftmp, tmp[1]); /* (X, -Y, Z) is the negative + * point */ + copy_small_conditional(ftmp, tmp[1], (((limb)sign) - 1)); + felem_contract(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 { - smallfelem_expand(nq[0], tmp[0]); - smallfelem_expand(nq[1], tmp[1]); - smallfelem_expand(nq[2], tmp[2]); - skip = 0; - } + 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 { + smallfelem_expand(nq[0], tmp[0]); + smallfelem_expand(nq[1], tmp[1]); + smallfelem_expand(nq[2], tmp[2]); + skip = 0; } } @@ -1581,31 +1575,17 @@ return 1; } -static int ec_GFp_nistp256_points_mul(const EC_GROUP *group, - EC_POINT *r, - const BIGNUM *g_scalar, - const EC_POINT *p_, - const BIGNUM *p_scalar_, - BN_CTX *ctx) { - /* TODO: This function used to take |points| and |scalars| as arrays of - * |num| elements. The code below should be simplified to work in terms of |p| - * and |p_scalar|. */ - size_t num = p_ != NULL ? 1 : 0; - const EC_POINT **points = p_ != NULL ? &p_ : NULL; - BIGNUM const *const *scalars = p_ != NULL ? &p_scalar_ : NULL; - +static int ec_GFp_nistp256_points_mul(const EC_GROUP *group, EC_POINT *r, + const BIGNUM *g_scalar, const EC_POINT *p, + const BIGNUM *p_scalar, BN_CTX *ctx) { int ret = 0; BN_CTX *new_ctx = NULL; BIGNUM *x, *y, *z, *tmp_scalar; - felem_bytearray g_secret; - felem_bytearray *secrets = NULL; - smallfelem(*pre_comp)[17][3] = NULL; + felem_bytearray g_secret, p_secret; + smallfelem p_pre_comp[17][3]; felem_bytearray tmp; - size_t num_points = num; smallfelem x_in, y_in, z_in; felem x_out, y_out, z_out; - const EC_POINT *p = NULL; - const BIGNUM *p_scalar = NULL; if (ctx == NULL) { ctx = new_ctx = BN_CTX_new(); @@ -1622,65 +1602,44 @@ goto err; } - if (num_points > 0) { - secrets = OPENSSL_malloc(num_points * sizeof(felem_bytearray)); - pre_comp = OPENSSL_malloc(num_points * sizeof(smallfelem[17][3])); - if (secrets == NULL || pre_comp == NULL) { - OPENSSL_PUT_ERROR(EC, ERR_R_MALLOC_FAILURE); + if (p != NULL && p_scalar != NULL) { + /* We treat NULL scalars as 0, and NULL points as points at infinity, i.e., + * they contribute nothing to the linear combination. */ + OPENSSL_memset(&p_secret, 0, sizeof(p_secret)); + OPENSSL_memset(&p_pre_comp, 0, sizeof(p_pre_comp)); + size_t num_bytes; + /* Reduce g_scalar to 0 <= g_scalar < 2^256. */ + if (BN_num_bits(p_scalar) > 256 || BN_is_negative(p_scalar)) { + /* This is an unusual input, and we don't guarantee constant-timeness. */ + if (!BN_nnmod(tmp_scalar, p_scalar, &group->order, ctx)) { + OPENSSL_PUT_ERROR(EC, ERR_R_BN_LIB); + goto err; + } + num_bytes = BN_bn2bin(tmp_scalar, tmp); + } else { + num_bytes = BN_bn2bin(p_scalar, tmp); + } + flip_endian(p_secret, tmp, num_bytes); + /* Precompute multiples. */ + if (!BN_to_felem(x_out, &p->X) || + !BN_to_felem(y_out, &p->Y) || + !BN_to_felem(z_out, &p->Z)) { goto err; } - - /* we treat NULL scalars as 0, and NULL points as points at infinity, - * i.e., they contribute nothing to the linear combination. */ - OPENSSL_memset(secrets, 0, num_points * sizeof(felem_bytearray)); - OPENSSL_memset(pre_comp, 0, num_points * 17 * 3 * sizeof(smallfelem)); - for (size_t i = 0; i < num_points; ++i) { - if (i == num) { - /* we didn't have a valid precomputation, so we pick the generator. */ - p = EC_GROUP_get0_generator(group); - p_scalar = g_scalar; + felem_shrink(p_pre_comp[1][0], x_out); + felem_shrink(p_pre_comp[1][1], y_out); + felem_shrink(p_pre_comp[1][2], z_out); + for (size_t j = 2; j <= 16; ++j) { + if (j & 1) { + point_add_small(p_pre_comp[j][0], p_pre_comp[j][1], + p_pre_comp[j][2], p_pre_comp[1][0], + p_pre_comp[1][1], p_pre_comp[1][2], + p_pre_comp[j - 1][0], p_pre_comp[j - 1][1], + p_pre_comp[j - 1][2]); } else { - /* the i^th point */ - p = points[i]; - p_scalar = scalars[i]; - } - if (p_scalar != NULL && p != NULL) { - size_t num_bytes; - /* reduce g_scalar to 0 <= g_scalar < 2^256 */ - if (BN_num_bits(p_scalar) > 256 || BN_is_negative(p_scalar)) { - /* this is an unusual input, and we don't guarantee - * constant-timeness. */ - if (!BN_nnmod(tmp_scalar, p_scalar, &group->order, ctx)) { - OPENSSL_PUT_ERROR(EC, ERR_R_BN_LIB); - goto err; - } - num_bytes = BN_bn2bin(tmp_scalar, tmp); - } else { - num_bytes = BN_bn2bin(p_scalar, tmp); - } - flip_endian(secrets[i], tmp, num_bytes); - /* precompute multiples */ - if (!BN_to_felem(x_out, &p->X) || - !BN_to_felem(y_out, &p->Y) || - !BN_to_felem(z_out, &p->Z)) { - goto err; - } - felem_shrink(pre_comp[i][1][0], x_out); - felem_shrink(pre_comp[i][1][1], y_out); - felem_shrink(pre_comp[i][1][2], z_out); - for (size_t j = 2; j <= 16; ++j) { - if (j & 1) { - point_add_small(pre_comp[i][j][0], pre_comp[i][j][1], - pre_comp[i][j][2], pre_comp[i][1][0], - pre_comp[i][1][1], pre_comp[i][1][2], - pre_comp[i][j - 1][0], pre_comp[i][j - 1][1], - pre_comp[i][j - 1][2]); - } else { - point_double_small(pre_comp[i][j][0], pre_comp[i][j][1], - pre_comp[i][j][2], pre_comp[i][j / 2][0], - pre_comp[i][j / 2][1], pre_comp[i][j / 2][2]); - } - } + point_double_small(p_pre_comp[j][0], p_pre_comp[j][1], + p_pre_comp[j][2], p_pre_comp[j / 2][0], + p_pre_comp[j / 2][1], p_pre_comp[j / 2][2]); } } } @@ -1703,9 +1662,10 @@ } flip_endian(g_secret, tmp, num_bytes); } - batch_mul(x_out, y_out, z_out, (const felem_bytearray(*))secrets, - num_points, g_scalar != NULL ? g_secret : NULL, - (const smallfelem(*)[17][3])pre_comp); + batch_mul(x_out, y_out, z_out, + (p != NULL && p_scalar != NULL) ? p_secret : NULL, + g_scalar != NULL ? g_secret : NULL, + (const smallfelem(*)[3]) &p_pre_comp); /* reduce the output to its unique minimal representation */ felem_contract(x_in, x_out); @@ -1722,8 +1682,6 @@ err: BN_CTX_end(ctx); BN_CTX_free(new_ctx); - OPENSSL_free(secrets); - OPENSSL_free(pre_comp); return ret; }