X-Git-Url: https://git.openssl.org/?p=openssl.git;a=blobdiff_plain;f=crypto%2Fec%2Fec_mult.c;h=aea2afd5804d5f08e3277a9ab94a974ece04efd7;hp=3c283e5ed6b5be4534a7ecac6ecaa0bf91e49740;hb=c2f2db9b6fb75ca2d672bb50f4f1f5a23991a6c3;hpb=9b398ef297dd1b74527dd0afee9f59cd3f5bc33d diff --git a/crypto/ec/ec_mult.c b/crypto/ec/ec_mult.c index 3c283e5ed6..aea2afd580 100644 --- a/crypto/ec/ec_mult.c +++ b/crypto/ec/ec_mult.c @@ -1,77 +1,36 @@ /* - * Originally written by Bodo Moeller and Nils Larsch for the OpenSSL project. - */ -/* ==================================================================== - * Copyright (c) 1998-2007 The OpenSSL Project. All rights reserved. - * - * Redistribution and use in source and binary forms, with or without - * modification, are permitted provided that the following conditions - * are met: - * - * 1. Redistributions of source code must retain the above copyright - * notice, this list of conditions and the following disclaimer. - * - * 2. Redistributions in binary form must reproduce the above copyright - * notice, this list of conditions and the following disclaimer in - * the documentation and/or other materials provided with the - * distribution. - * - * 3. All advertising materials mentioning features or use of this - * software must display the following acknowledgment: - * "This product includes software developed by the OpenSSL Project - * for use in the OpenSSL Toolkit. (http://www.openssl.org/)" - * - * 4. The names "OpenSSL Toolkit" and "OpenSSL Project" must not be used to - * endorse or promote products derived from this software without - * prior written permission. For written permission, please contact - * openssl-core@openssl.org. - * - * 5. Products derived from this software may not be called "OpenSSL" - * nor may "OpenSSL" appear in their names without prior written - * permission of the OpenSSL Project. - * - * 6. Redistributions of any form whatsoever must retain the following - * acknowledgment: - * "This product includes software developed by the OpenSSL Project - * for use in the OpenSSL Toolkit (http://www.openssl.org/)" - * - * THIS SOFTWARE IS PROVIDED BY THE OpenSSL PROJECT ``AS IS'' AND ANY - * EXPRESSED OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE - * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR - * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE OpenSSL PROJECT OR - * ITS CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, - * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT - * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; - * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) - * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, - * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) - * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED - * OF THE POSSIBILITY OF SUCH DAMAGE. - * ==================================================================== - * - * This product includes cryptographic software written by Eric Young - * (eay@cryptsoft.com). This product includes software written by Tim - * Hudson (tjh@cryptsoft.com). + * Copyright 2001-2020 The OpenSSL Project Authors. All Rights Reserved. + * Copyright (c) 2002, Oracle and/or its affiliates. All rights reserved * + * 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 */ -/* ==================================================================== - * Copyright 2002 Sun Microsystems, Inc. ALL RIGHTS RESERVED. - * Portions of this software developed by SUN MICROSYSTEMS, INC., - * and contributed to the OpenSSL project. + +/* + * ECDSA low level APIs are deprecated for public use, but still ok for + * internal use. */ +#include "internal/deprecated.h" #include #include #include "internal/cryptlib.h" -#include "internal/bn_int.h" -#include "ec_lcl.h" +#include "crypto/bn.h" +#include "ec_local.h" +#include "internal/refcount.h" /* * This file implements the wNAF-based interleaving multi-exponentiation method - * (); - * for multiplication with precomputation, we use wNAF splitting - * (). + * Formerly at: + * http://www.informatik.tu-darmstadt.de/TI/Mitarbeiter/moeller.html#multiexp + * You might now find it here: + * http://link.springer.com/chapter/10.1007%2F3-540-45537-X_13 + * http://www.bmoeller.de/pdf/TI-01-08.multiexp.pdf + * For multiplication with precomputation, we use wNAF splitting, formerly at: + * http://www.informatik.tu-darmstadt.de/TI/Mitarbeiter/moeller.html#fastexp */ /* structure for precomputed multiples of the generator */ @@ -85,7 +44,7 @@ struct ec_pre_comp_st { * generator: 'num' pointers to EC_POINT * objects followed by a NULL */ size_t num; /* numblocks * 2^(w-1) */ - int references; + CRYPTO_REF_COUNT references; CRYPTO_RWLOCK *lock; }; @@ -120,7 +79,7 @@ EC_PRE_COMP *EC_ec_pre_comp_dup(EC_PRE_COMP *pre) { int i; if (pre != NULL) - CRYPTO_atomic_add(&pre->references, 1, &i, pre->lock); + CRYPTO_UP_REF(&pre->references, &i, pre->lock); return pre; } @@ -131,7 +90,7 @@ void EC_ec_pre_comp_free(EC_PRE_COMP *pre) if (pre == NULL) return; - CRYPTO_atomic_add(&pre->references, -1, &i, pre->lock); + CRYPTO_DOWN_REF(&pre->references, &i, pre->lock); REF_PRINT_COUNT("EC_ec", pre); if (i > 0) return; @@ -148,6 +107,285 @@ void EC_ec_pre_comp_free(EC_PRE_COMP *pre) OPENSSL_free(pre); } +#define EC_POINT_BN_set_flags(P, flags) do { \ + BN_set_flags((P)->X, (flags)); \ + BN_set_flags((P)->Y, (flags)); \ + BN_set_flags((P)->Z, (flags)); \ +} while(0) + +/*- + * This functions computes a single point multiplication over the EC group, + * using, at a high level, a Montgomery ladder with conditional swaps, with + * various timing attack defenses. + * + * It performs either a fixed point multiplication + * (scalar * generator) + * when point is NULL, or a variable point multiplication + * (scalar * point) + * when point is not NULL. + * + * `scalar` cannot be NULL and should be in the range [0,n) otherwise all + * constant time bets are off (where n is the cardinality of the EC group). + * + * This function expects `group->order` and `group->cardinality` to be well + * defined and non-zero: it fails with an error code otherwise. + * + * NB: This says nothing about the constant-timeness of the ladder step + * implementation (i.e., the default implementation is based on EC_POINT_add and + * EC_POINT_dbl, which of course are not constant time themselves) or the + * underlying multiprecision arithmetic. + * + * The product is stored in `r`. + * + * This is an internal function: callers are in charge of ensuring that the + * input parameters `group`, `r`, `scalar` and `ctx` are not NULL. + * + * Returns 1 on success, 0 otherwise. + */ +int ec_scalar_mul_ladder(const EC_GROUP *group, EC_POINT *r, + const BIGNUM *scalar, const EC_POINT *point, + BN_CTX *ctx) +{ + int i, cardinality_bits, group_top, kbit, pbit, Z_is_one; + EC_POINT *p = NULL; + EC_POINT *s = NULL; + BIGNUM *k = NULL; + BIGNUM *lambda = NULL; + BIGNUM *cardinality = NULL; + int ret = 0; + + /* early exit if the input point is the point at infinity */ + if (point != NULL && EC_POINT_is_at_infinity(group, point)) + return EC_POINT_set_to_infinity(group, r); + + if (BN_is_zero(group->order)) { + ECerr(EC_F_EC_SCALAR_MUL_LADDER, EC_R_UNKNOWN_ORDER); + return 0; + } + if (BN_is_zero(group->cofactor)) { + ECerr(EC_F_EC_SCALAR_MUL_LADDER, EC_R_UNKNOWN_COFACTOR); + return 0; + } + + BN_CTX_start(ctx); + + if (((p = EC_POINT_new(group)) == NULL) + || ((s = EC_POINT_new(group)) == NULL)) { + ECerr(EC_F_EC_SCALAR_MUL_LADDER, ERR_R_MALLOC_FAILURE); + goto err; + } + + if (point == NULL) { + if (!EC_POINT_copy(p, group->generator)) { + ECerr(EC_F_EC_SCALAR_MUL_LADDER, ERR_R_EC_LIB); + goto err; + } + } else { + if (!EC_POINT_copy(p, point)) { + ECerr(EC_F_EC_SCALAR_MUL_LADDER, ERR_R_EC_LIB); + goto err; + } + } + + EC_POINT_BN_set_flags(p, BN_FLG_CONSTTIME); + EC_POINT_BN_set_flags(r, BN_FLG_CONSTTIME); + EC_POINT_BN_set_flags(s, BN_FLG_CONSTTIME); + + cardinality = BN_CTX_get(ctx); + lambda = BN_CTX_get(ctx); + k = BN_CTX_get(ctx); + if (k == NULL) { + ECerr(EC_F_EC_SCALAR_MUL_LADDER, ERR_R_MALLOC_FAILURE); + goto err; + } + + if (!BN_mul(cardinality, group->order, group->cofactor, ctx)) { + ECerr(EC_F_EC_SCALAR_MUL_LADDER, ERR_R_BN_LIB); + goto err; + } + + /* + * Group cardinalities are often on a word boundary. + * So when we pad the scalar, some timing diff might + * pop if it needs to be expanded due to carries. + * So expand ahead of time. + */ + cardinality_bits = BN_num_bits(cardinality); + group_top = bn_get_top(cardinality); + if ((bn_wexpand(k, group_top + 2) == NULL) + || (bn_wexpand(lambda, group_top + 2) == NULL)) { + ECerr(EC_F_EC_SCALAR_MUL_LADDER, ERR_R_BN_LIB); + goto err; + } + + if (!BN_copy(k, scalar)) { + ECerr(EC_F_EC_SCALAR_MUL_LADDER, ERR_R_BN_LIB); + goto err; + } + + BN_set_flags(k, BN_FLG_CONSTTIME); + + if ((BN_num_bits(k) > cardinality_bits) || (BN_is_negative(k))) { + /*- + * this is an unusual input, and we don't guarantee + * constant-timeness + */ + if (!BN_nnmod(k, k, cardinality, ctx)) { + ECerr(EC_F_EC_SCALAR_MUL_LADDER, ERR_R_BN_LIB); + goto err; + } + } + + if (!BN_add(lambda, k, cardinality)) { + ECerr(EC_F_EC_SCALAR_MUL_LADDER, ERR_R_BN_LIB); + goto err; + } + BN_set_flags(lambda, BN_FLG_CONSTTIME); + if (!BN_add(k, lambda, cardinality)) { + ECerr(EC_F_EC_SCALAR_MUL_LADDER, ERR_R_BN_LIB); + goto err; + } + /* + * lambda := scalar + cardinality + * k := scalar + 2*cardinality + */ + kbit = BN_is_bit_set(lambda, cardinality_bits); + BN_consttime_swap(kbit, k, lambda, group_top + 2); + + group_top = bn_get_top(group->field); + if ((bn_wexpand(s->X, group_top) == NULL) + || (bn_wexpand(s->Y, group_top) == NULL) + || (bn_wexpand(s->Z, group_top) == NULL) + || (bn_wexpand(r->X, group_top) == NULL) + || (bn_wexpand(r->Y, group_top) == NULL) + || (bn_wexpand(r->Z, group_top) == NULL) + || (bn_wexpand(p->X, group_top) == NULL) + || (bn_wexpand(p->Y, group_top) == NULL) + || (bn_wexpand(p->Z, group_top) == NULL)) { + ECerr(EC_F_EC_SCALAR_MUL_LADDER, ERR_R_BN_LIB); + goto err; + } + + /* ensure input point is in affine coords for ladder step efficiency */ + if (!p->Z_is_one && (group->meth->make_affine == NULL + || !group->meth->make_affine(group, p, ctx))) { + ECerr(EC_F_EC_SCALAR_MUL_LADDER, ERR_R_EC_LIB); + goto err; + } + + /* Initialize the Montgomery ladder */ + if (!ec_point_ladder_pre(group, r, s, p, ctx)) { + ECerr(EC_F_EC_SCALAR_MUL_LADDER, EC_R_LADDER_PRE_FAILURE); + goto err; + } + + /* top bit is a 1, in a fixed pos */ + pbit = 1; + +#define EC_POINT_CSWAP(c, a, b, w, t) do { \ + BN_consttime_swap(c, (a)->X, (b)->X, w); \ + BN_consttime_swap(c, (a)->Y, (b)->Y, w); \ + BN_consttime_swap(c, (a)->Z, (b)->Z, w); \ + t = ((a)->Z_is_one ^ (b)->Z_is_one) & (c); \ + (a)->Z_is_one ^= (t); \ + (b)->Z_is_one ^= (t); \ +} while(0) + + /*- + * The ladder step, with branches, is + * + * k[i] == 0: S = add(R, S), R = dbl(R) + * k[i] == 1: R = add(S, R), S = dbl(S) + * + * Swapping R, S conditionally on k[i] leaves you with state + * + * k[i] == 0: T, U = R, S + * k[i] == 1: T, U = S, R + * + * Then perform the ECC ops. + * + * U = add(T, U) + * T = dbl(T) + * + * Which leaves you with state + * + * k[i] == 0: U = add(R, S), T = dbl(R) + * k[i] == 1: U = add(S, R), T = dbl(S) + * + * Swapping T, U conditionally on k[i] leaves you with state + * + * k[i] == 0: R, S = T, U + * k[i] == 1: R, S = U, T + * + * Which leaves you with state + * + * k[i] == 0: S = add(R, S), R = dbl(R) + * k[i] == 1: R = add(S, R), S = dbl(S) + * + * So we get the same logic, but instead of a branch it's a + * conditional swap, followed by ECC ops, then another conditional swap. + * + * Optimization: The end of iteration i and start of i-1 looks like + * + * ... + * CSWAP(k[i], R, S) + * ECC + * CSWAP(k[i], R, S) + * (next iteration) + * CSWAP(k[i-1], R, S) + * ECC + * CSWAP(k[i-1], R, S) + * ... + * + * So instead of two contiguous swaps, you can merge the condition + * bits and do a single swap. + * + * k[i] k[i-1] Outcome + * 0 0 No Swap + * 0 1 Swap + * 1 0 Swap + * 1 1 No Swap + * + * This is XOR. pbit tracks the previous bit of k. + */ + + for (i = cardinality_bits - 1; i >= 0; i--) { + kbit = BN_is_bit_set(k, i) ^ pbit; + EC_POINT_CSWAP(kbit, r, s, group_top, Z_is_one); + + /* Perform a single step of the Montgomery ladder */ + if (!ec_point_ladder_step(group, r, s, p, ctx)) { + ECerr(EC_F_EC_SCALAR_MUL_LADDER, EC_R_LADDER_STEP_FAILURE); + goto err; + } + /* + * pbit logic merges this cswap with that of the + * next iteration + */ + pbit ^= kbit; + } + /* one final cswap to move the right value into r */ + EC_POINT_CSWAP(pbit, r, s, group_top, Z_is_one); +#undef EC_POINT_CSWAP + + /* Finalize ladder (and recover full point coordinates) */ + if (!ec_point_ladder_post(group, r, s, p, ctx)) { + ECerr(EC_F_EC_SCALAR_MUL_LADDER, EC_R_LADDER_POST_FAILURE); + goto err; + } + + ret = 1; + + err: + EC_POINT_free(p); + EC_POINT_clear_free(s); + BN_CTX_end(ctx); + + return ret; +} + +#undef EC_POINT_BN_set_flags + /* * TODO: table should be optimised for the wNAF-based implementation, * sometimes smaller windows will give better performance (thus the @@ -173,7 +411,6 @@ int ec_wNAF_mul(const EC_GROUP *group, EC_POINT *r, const BIGNUM *scalar, size_t num, const EC_POINT *points[], const BIGNUM *scalars[], BN_CTX *ctx) { - BN_CTX *new_ctx = NULL; const EC_POINT *generator = NULL; EC_POINT *tmp = NULL; size_t totalnum; @@ -198,26 +435,33 @@ int ec_wNAF_mul(const EC_GROUP *group, EC_POINT *r, const BIGNUM *scalar, * precomputation is not available */ int ret = 0; - if (group->meth != r->meth) { - ECerr(EC_F_EC_WNAF_MUL, EC_R_INCOMPATIBLE_OBJECTS); - return 0; - } - - if ((scalar == NULL) && (num == 0)) { - return EC_POINT_set_to_infinity(group, r); - } - - for (i = 0; i < num; i++) { - if (group->meth != points[i]->meth) { - ECerr(EC_F_EC_WNAF_MUL, EC_R_INCOMPATIBLE_OBJECTS); - return 0; + if (!BN_is_zero(group->order) && !BN_is_zero(group->cofactor)) { + /*- + * Handle the common cases where the scalar is secret, enforcing a + * scalar multiplication implementation based on a Montgomery ladder, + * with various timing attack defenses. + */ + if ((scalar != group->order) && (scalar != NULL) && (num == 0)) { + /*- + * In this case we want to compute scalar * GeneratorPoint: this + * codepath is reached most prominently by (ephemeral) key + * generation of EC cryptosystems (i.e. ECDSA keygen and sign setup, + * ECDH keygen/first half), where the scalar is always secret. This + * is why we ignore if BN_FLG_CONSTTIME is actually set and we + * always call the ladder version. + */ + return ec_scalar_mul_ladder(group, r, scalar, NULL, ctx); + } + if ((scalar == NULL) && (num == 1) && (scalars[0] != group->order)) { + /*- + * In this case we want to compute scalar * VariablePoint: this + * codepath is reached most prominently by the second half of ECDH, + * where the secret scalar is multiplied by the peer's public point. + * To protect the secret scalar, we ignore if BN_FLG_CONSTTIME is + * actually set and we always call the ladder version. + */ + return ec_scalar_mul_ladder(group, r, scalars[0], points[0], ctx); } - } - - if (ctx == NULL) { - ctx = new_ctx = BN_CTX_new(); - if (ctx == NULL) - goto err; } if (scalar != NULL) { @@ -265,11 +509,11 @@ int ec_wNAF_mul(const EC_GROUP *group, EC_POINT *r, const BIGNUM *scalar, totalnum = num + numblocks; - wsize = OPENSSL_malloc(totalnum * sizeof wsize[0]); - wNAF_len = OPENSSL_malloc(totalnum * sizeof wNAF_len[0]); - wNAF = OPENSSL_malloc((totalnum + 1) * sizeof wNAF[0]); /* includes space - * for pivot */ - val_sub = OPENSSL_malloc(totalnum * sizeof val_sub[0]); + wsize = OPENSSL_malloc(totalnum * sizeof(wsize[0])); + wNAF_len = OPENSSL_malloc(totalnum * sizeof(wNAF_len[0])); + /* include space for pivot */ + wNAF = OPENSSL_malloc((totalnum + 1) * sizeof(wNAF[0])); + val_sub = OPENSSL_malloc(totalnum * sizeof(val_sub[0])); /* Ensure wNAF is initialised in case we end up going to err */ if (wNAF != NULL) @@ -359,6 +603,7 @@ int ec_wNAF_mul(const EC_GROUP *group, EC_POINT *r, const BIGNUM *scalar, numblocks = (tmp_len + blocksize - 1) / blocksize; if (numblocks > pre_comp->numblocks) { ECerr(EC_F_EC_WNAF_MUL, ERR_R_INTERNAL_ERROR); + OPENSSL_free(tmp_wNAF); goto err; } totalnum = num + numblocks; @@ -373,6 +618,7 @@ int ec_wNAF_mul(const EC_GROUP *group, EC_POINT *r, const BIGNUM *scalar, wNAF_len[i] = blocksize; if (tmp_len < blocksize) { ECerr(EC_F_EC_WNAF_MUL, ERR_R_INTERNAL_ERROR); + OPENSSL_free(tmp_wNAF); goto err; } tmp_len -= blocksize; @@ -413,7 +659,7 @@ int ec_wNAF_mul(const EC_GROUP *group, EC_POINT *r, const BIGNUM *scalar, * 'val_sub[i]' is a pointer to the subarray for the i-th point, or to a * subarray of 'pre_comp->points' if we already have precomputation. */ - val = OPENSSL_malloc((num_val + 1) * sizeof val[0]); + val = OPENSSL_malloc((num_val + 1) * sizeof(val[0])); if (val == NULL) { ECerr(EC_F_EC_WNAF_MUL, ERR_R_MALLOC_FAILURE); goto err; @@ -466,7 +712,8 @@ int ec_wNAF_mul(const EC_GROUP *group, EC_POINT *r, const BIGNUM *scalar, } } - if (!EC_POINTs_make_affine(group, num_val, val, ctx)) + if (group->meth->points_make_affine == NULL + || !group->meth->points_make_affine(group, num_val, val, ctx)) goto err; r_is_at_infinity = 1; @@ -501,6 +748,20 @@ int ec_wNAF_mul(const EC_GROUP *group, EC_POINT *r, const BIGNUM *scalar, if (r_is_at_infinity) { if (!EC_POINT_copy(r, val_sub[i][digit >> 1])) goto err; + + /*- + * Apply coordinate blinding for EC_POINT. + * + * The underlying EC_METHOD can optionally implement this function: + * ec_point_blind_coordinates() returns 0 in case of errors or 1 on + * success or if coordinate blinding is not implemented for this + * group. + */ + if (!ec_point_blind_coordinates(group, r, ctx)) { + ECerr(EC_F_EC_WNAF_MUL, EC_R_POINT_COORDINATES_BLIND_FAILURE); + goto err; + } + r_is_at_infinity = 0; } else { if (!EC_POINT_add @@ -524,7 +785,6 @@ int ec_wNAF_mul(const EC_GROUP *group, EC_POINT *r, const BIGNUM *scalar, ret = 1; err: - BN_CTX_free(new_ctx); EC_POINT_free(tmp); OPENSSL_free(wsize); OPENSSL_free(wNAF_len); @@ -570,12 +830,14 @@ int ec_wNAF_precompute_mult(EC_GROUP *group, BN_CTX *ctx) { const EC_POINT *generator; EC_POINT *tmp_point = NULL, *base = NULL, **var; - BN_CTX *new_ctx = NULL; const BIGNUM *order; size_t i, bits, w, pre_points_per_block, blocksize, numblocks, num; EC_POINT **points = NULL; EC_PRE_COMP *pre_comp; int ret = 0; +#ifndef FIPS_MODULE + BN_CTX *new_ctx = NULL; +#endif /* if there is an old EC_PRE_COMP object, throw it away */ EC_pre_comp_free(group); @@ -588,11 +850,12 @@ int ec_wNAF_precompute_mult(EC_GROUP *group, BN_CTX *ctx) goto err; } - if (ctx == NULL) { +#ifndef FIPS_MODULE + if (ctx == NULL) ctx = new_ctx = BN_CTX_new(); - if (ctx == NULL) - goto err; - } +#endif + if (ctx == NULL) + goto err; BN_CTX_start(ctx); @@ -688,7 +951,8 @@ int ec_wNAF_precompute_mult(EC_GROUP *group, BN_CTX *ctx) } } - if (!EC_POINTs_make_affine(group, num, points, ctx)) + if (group->meth->points_make_affine == NULL + || !group->meth->points_make_affine(group, num, points, ctx)) goto err; pre_comp->group = group; @@ -703,9 +967,10 @@ int ec_wNAF_precompute_mult(EC_GROUP *group, BN_CTX *ctx) ret = 1; err: - if (ctx != NULL) - BN_CTX_end(ctx); + BN_CTX_end(ctx); +#ifndef FIPS_MODULE BN_CTX_free(new_ctx); +#endif EC_ec_pre_comp_free(pre_comp); if (points) { EC_POINT **p;