2016-05-17 18:51:04 +00:00
|
|
|
/*
|
2019-02-26 14:05:09 +00:00
|
|
|
* Copyright 1995-2019 The OpenSSL Project Authors. All Rights Reserved.
|
2000-06-08 09:39:28 +00:00
|
|
|
*
|
2016-05-17 18:51:04 +00:00
|
|
|
* Licensed under the OpenSSL license (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
|
2000-06-08 09:39:28 +00:00
|
|
|
*/
|
|
|
|
|
2015-05-14 14:56:48 +00:00
|
|
|
#include "internal/cryptlib.h"
|
2016-01-25 19:38:38 +00:00
|
|
|
#include "internal/constant_time_locl.h"
|
1998-12-21 10:52:47 +00:00
|
|
|
#include "bn_lcl.h"
|
2000-06-07 21:29:25 +00:00
|
|
|
|
2011-08-12 16:44:32 +00:00
|
|
|
#include <stdlib.h>
|
|
|
|
#ifdef _WIN32
|
|
|
|
# include <malloc.h>
|
|
|
|
# ifndef alloca
|
|
|
|
# define alloca _alloca
|
|
|
|
# endif
|
|
|
|
#elif defined(__GNUC__)
|
|
|
|
# ifndef alloca
|
|
|
|
# define alloca(s) __builtin_alloca((s))
|
|
|
|
# endif
|
2013-06-30 21:09:09 +00:00
|
|
|
#elif defined(__sun)
|
|
|
|
# include <alloca.h>
|
2011-08-12 16:44:32 +00:00
|
|
|
#endif
|
|
|
|
|
2015-06-22 23:55:46 +00:00
|
|
|
#include "rsaz_exp.h"
|
2013-07-05 19:39:47 +00:00
|
|
|
|
2013-07-05 18:52:58 +00:00
|
|
|
#undef SPARC_T4_MONT
|
2013-06-01 07:58:07 +00:00
|
|
|
#if defined(OPENSSL_BN_ASM_MONT) && (defined(__sparc__) || defined(__sparc))
|
2012-11-17 10:34:11 +00:00
|
|
|
# include "sparc_arch.h"
|
|
|
|
extern unsigned int OPENSSL_sparcv9cap_P[];
|
2013-07-05 18:52:58 +00:00
|
|
|
# define SPARC_T4_MONT
|
2012-11-17 10:34:11 +00:00
|
|
|
#endif
|
|
|
|
|
2005-05-16 01:43:31 +00:00
|
|
|
/* maximum precomputation table size for *variable* sliding windows */
|
2015-01-22 03:40:55 +00:00
|
|
|
#define TABLE_SIZE 32
|
1998-12-21 11:00:56 +00:00
|
|
|
|
1998-12-21 10:56:39 +00:00
|
|
|
/* this one works - simple but works */
|
2000-11-06 21:15:54 +00:00
|
|
|
int BN_exp(BIGNUM *r, const BIGNUM *a, const BIGNUM *p, BN_CTX *ctx)
|
2015-01-22 03:40:55 +00:00
|
|
|
{
|
|
|
|
int i, bits, ret = 0;
|
|
|
|
BIGNUM *v, *rr;
|
|
|
|
|
2017-09-27 10:13:47 +00:00
|
|
|
if (BN_get_flags(p, BN_FLG_CONSTTIME) != 0
|
|
|
|
|| BN_get_flags(a, BN_FLG_CONSTTIME) != 0) {
|
2015-01-22 03:40:55 +00:00
|
|
|
/* BN_FLG_CONSTTIME only supported by BN_mod_exp_mont() */
|
|
|
|
BNerr(BN_F_BN_EXP, ERR_R_SHOULD_NOT_HAVE_BEEN_CALLED);
|
2016-08-14 19:52:13 +00:00
|
|
|
return 0;
|
2015-01-22 03:40:55 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
BN_CTX_start(ctx);
|
2017-06-22 10:52:29 +00:00
|
|
|
rr = ((r == a) || (r == p)) ? BN_CTX_get(ctx) : r;
|
2015-01-22 03:40:55 +00:00
|
|
|
v = BN_CTX_get(ctx);
|
|
|
|
if (rr == NULL || v == NULL)
|
|
|
|
goto err;
|
|
|
|
|
|
|
|
if (BN_copy(v, a) == NULL)
|
|
|
|
goto err;
|
|
|
|
bits = BN_num_bits(p);
|
|
|
|
|
|
|
|
if (BN_is_odd(p)) {
|
|
|
|
if (BN_copy(rr, a) == NULL)
|
|
|
|
goto err;
|
|
|
|
} else {
|
|
|
|
if (!BN_one(rr))
|
|
|
|
goto err;
|
|
|
|
}
|
|
|
|
|
|
|
|
for (i = 1; i < bits; i++) {
|
|
|
|
if (!BN_sqr(v, v, ctx))
|
|
|
|
goto err;
|
|
|
|
if (BN_is_bit_set(p, i)) {
|
|
|
|
if (!BN_mul(rr, rr, v, ctx))
|
|
|
|
goto err;
|
|
|
|
}
|
|
|
|
}
|
2016-11-15 23:54:28 +00:00
|
|
|
if (r != rr && BN_copy(r, rr) == NULL)
|
|
|
|
goto err;
|
|
|
|
|
2015-03-11 15:31:16 +00:00
|
|
|
ret = 1;
|
|
|
|
err:
|
2015-01-22 03:40:55 +00:00
|
|
|
BN_CTX_end(ctx);
|
|
|
|
bn_check_top(r);
|
2017-06-22 10:52:29 +00:00
|
|
|
return ret;
|
2015-01-22 03:40:55 +00:00
|
|
|
}
|
2000-06-07 21:29:25 +00:00
|
|
|
|
2000-11-06 21:15:54 +00:00
|
|
|
int BN_mod_exp(BIGNUM *r, const BIGNUM *a, const BIGNUM *p, const BIGNUM *m,
|
2015-01-22 03:40:55 +00:00
|
|
|
BN_CTX *ctx)
|
|
|
|
{
|
|
|
|
int ret;
|
|
|
|
|
|
|
|
bn_check_top(a);
|
|
|
|
bn_check_top(p);
|
|
|
|
bn_check_top(m);
|
|
|
|
|
2015-01-05 11:30:03 +00:00
|
|
|
/*-
|
2016-06-28 20:51:01 +00:00
|
|
|
* For even modulus m = 2^k*m_odd, it might make sense to compute
|
2015-01-05 11:30:03 +00:00
|
|
|
* a^p mod m_odd and a^p mod 2^k separately (with Montgomery
|
|
|
|
* exponentiation for the odd part), using appropriate exponent
|
|
|
|
* reductions, and combine the results using the CRT.
|
|
|
|
*
|
|
|
|
* For now, we use Montgomery only if the modulus is odd; otherwise,
|
|
|
|
* exponentiation using the reciprocal-based quick remaindering
|
|
|
|
* algorithm is used.
|
|
|
|
*
|
|
|
|
* (Timing obtained with expspeed.c [computations a^p mod m
|
|
|
|
* where a, p, m are of the same length: 256, 512, 1024, 2048,
|
|
|
|
* 4096, 8192 bits], compared to the running time of the
|
|
|
|
* standard algorithm:
|
|
|
|
*
|
|
|
|
* BN_mod_exp_mont 33 .. 40 % [AMD K6-2, Linux, debug configuration]
|
|
|
|
* 55 .. 77 % [UltraSparc processor, but
|
|
|
|
* debug-solaris-sparcv8-gcc conf.]
|
|
|
|
*
|
|
|
|
* BN_mod_exp_recp 50 .. 70 % [AMD K6-2, Linux, debug configuration]
|
|
|
|
* 62 .. 118 % [UltraSparc, debug-solaris-sparcv8-gcc]
|
|
|
|
*
|
|
|
|
* On the Sparc, BN_mod_exp_recp was faster than BN_mod_exp_mont
|
|
|
|
* at 2048 and more bits, but at 512 and 1024 bits, it was
|
|
|
|
* slower even than the standard algorithm!
|
|
|
|
*
|
|
|
|
* "Real" timings [linux-elf, solaris-sparcv9-gcc configurations]
|
|
|
|
* should be obtained when the new Montgomery reduction code
|
|
|
|
* has been integrated into OpenSSL.)
|
|
|
|
*/
|
2000-11-26 16:42:38 +00:00
|
|
|
|
|
|
|
#define MONT_MUL_MOD
|
2000-11-30 09:45:26 +00:00
|
|
|
#define MONT_EXP_WORD
|
2000-11-26 16:42:38 +00:00
|
|
|
#define RECP_MUL_MOD
|
|
|
|
|
1998-12-21 10:52:47 +00:00
|
|
|
#ifdef MONT_MUL_MOD
|
2015-01-22 03:40:55 +00:00
|
|
|
if (BN_is_odd(m)) {
|
|
|
|
# ifdef MONT_EXP_WORD
|
|
|
|
if (a->top == 1 && !a->neg
|
2017-09-27 10:13:47 +00:00
|
|
|
&& (BN_get_flags(p, BN_FLG_CONSTTIME) == 0)
|
|
|
|
&& (BN_get_flags(a, BN_FLG_CONSTTIME) == 0)
|
|
|
|
&& (BN_get_flags(m, BN_FLG_CONSTTIME) == 0)) {
|
2015-01-22 03:40:55 +00:00
|
|
|
BN_ULONG A = a->d[0];
|
|
|
|
ret = BN_mod_exp_mont_word(r, A, p, m, ctx, NULL);
|
|
|
|
} else
|
|
|
|
# endif
|
|
|
|
ret = BN_mod_exp_mont(r, a, p, m, ctx, NULL);
|
|
|
|
} else
|
1998-12-21 10:52:47 +00:00
|
|
|
#endif
|
|
|
|
#ifdef RECP_MUL_MOD
|
2015-01-22 03:40:55 +00:00
|
|
|
{
|
|
|
|
ret = BN_mod_exp_recp(r, a, p, m, ctx);
|
|
|
|
}
|
1998-12-21 10:52:47 +00:00
|
|
|
#else
|
2015-01-22 03:40:55 +00:00
|
|
|
{
|
|
|
|
ret = BN_mod_exp_simple(r, a, p, m, ctx);
|
|
|
|
}
|
1998-12-21 10:52:47 +00:00
|
|
|
#endif
|
|
|
|
|
2015-01-22 03:40:55 +00:00
|
|
|
bn_check_top(r);
|
2017-10-17 14:04:09 +00:00
|
|
|
return ret;
|
2015-01-22 03:40:55 +00:00
|
|
|
}
|
2000-06-07 21:29:25 +00:00
|
|
|
|
1999-06-04 22:23:10 +00:00
|
|
|
int BN_mod_exp_recp(BIGNUM *r, const BIGNUM *a, const BIGNUM *p,
|
2015-01-22 03:40:55 +00:00
|
|
|
const BIGNUM *m, BN_CTX *ctx)
|
|
|
|
{
|
|
|
|
int i, j, bits, ret = 0, wstart, wend, window, wvalue;
|
|
|
|
int start = 1;
|
|
|
|
BIGNUM *aa;
|
|
|
|
/* Table of variables obtained from 'ctx' */
|
|
|
|
BIGNUM *val[TABLE_SIZE];
|
|
|
|
BN_RECP_CTX recp;
|
|
|
|
|
2017-09-27 10:13:47 +00:00
|
|
|
if (BN_get_flags(p, BN_FLG_CONSTTIME) != 0
|
|
|
|
|| BN_get_flags(a, BN_FLG_CONSTTIME) != 0
|
|
|
|
|| BN_get_flags(m, BN_FLG_CONSTTIME) != 0) {
|
2015-01-22 03:40:55 +00:00
|
|
|
/* BN_FLG_CONSTTIME only supported by BN_mod_exp_mont() */
|
|
|
|
BNerr(BN_F_BN_MOD_EXP_RECP, ERR_R_SHOULD_NOT_HAVE_BEEN_CALLED);
|
2016-08-14 19:52:13 +00:00
|
|
|
return 0;
|
2015-01-22 03:40:55 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
bits = BN_num_bits(p);
|
|
|
|
if (bits == 0) {
|
2018-05-24 15:12:52 +00:00
|
|
|
/* x**0 mod 1, or x**0 mod -1 is still zero. */
|
|
|
|
if (BN_abs_is_word(m, 1)) {
|
2015-12-14 15:38:15 +00:00
|
|
|
ret = 1;
|
|
|
|
BN_zero(r);
|
|
|
|
} else {
|
|
|
|
ret = BN_one(r);
|
|
|
|
}
|
2015-01-22 03:40:55 +00:00
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
|
|
|
BN_CTX_start(ctx);
|
|
|
|
aa = BN_CTX_get(ctx);
|
|
|
|
val[0] = BN_CTX_get(ctx);
|
2017-06-22 10:52:29 +00:00
|
|
|
if (val[0] == NULL)
|
2015-01-22 03:40:55 +00:00
|
|
|
goto err;
|
|
|
|
|
|
|
|
BN_RECP_CTX_init(&recp);
|
|
|
|
if (m->neg) {
|
|
|
|
/* ignore sign of 'm' */
|
|
|
|
if (!BN_copy(aa, m))
|
|
|
|
goto err;
|
|
|
|
aa->neg = 0;
|
|
|
|
if (BN_RECP_CTX_set(&recp, aa, ctx) <= 0)
|
|
|
|
goto err;
|
|
|
|
} else {
|
|
|
|
if (BN_RECP_CTX_set(&recp, m, ctx) <= 0)
|
|
|
|
goto err;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (!BN_nnmod(val[0], a, m, ctx))
|
|
|
|
goto err; /* 1 */
|
|
|
|
if (BN_is_zero(val[0])) {
|
|
|
|
BN_zero(r);
|
|
|
|
ret = 1;
|
|
|
|
goto err;
|
|
|
|
}
|
|
|
|
|
|
|
|
window = BN_window_bits_for_exponent_size(bits);
|
|
|
|
if (window > 1) {
|
|
|
|
if (!BN_mod_mul_reciprocal(aa, val[0], val[0], &recp, ctx))
|
|
|
|
goto err; /* 2 */
|
|
|
|
j = 1 << (window - 1);
|
|
|
|
for (i = 1; i < j; i++) {
|
|
|
|
if (((val[i] = BN_CTX_get(ctx)) == NULL) ||
|
|
|
|
!BN_mod_mul_reciprocal(val[i], val[i - 1], aa, &recp, ctx))
|
|
|
|
goto err;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
start = 1; /* This is used to avoid multiplication etc
|
|
|
|
* when there is only the value '1' in the
|
|
|
|
* buffer. */
|
|
|
|
wvalue = 0; /* The 'value' of the window */
|
|
|
|
wstart = bits - 1; /* The top bit of the window */
|
|
|
|
wend = 0; /* The bottom bit of the window */
|
|
|
|
|
|
|
|
if (!BN_one(r))
|
|
|
|
goto err;
|
|
|
|
|
|
|
|
for (;;) {
|
|
|
|
if (BN_is_bit_set(p, wstart) == 0) {
|
|
|
|
if (!start)
|
|
|
|
if (!BN_mod_mul_reciprocal(r, r, r, &recp, ctx))
|
|
|
|
goto err;
|
|
|
|
if (wstart == 0)
|
|
|
|
break;
|
|
|
|
wstart--;
|
|
|
|
continue;
|
|
|
|
}
|
|
|
|
/*
|
|
|
|
* We now have wstart on a 'set' bit, we now need to work out how bit
|
|
|
|
* a window to do. To do this we need to scan forward until the last
|
|
|
|
* set bit before the end of the window
|
|
|
|
*/
|
|
|
|
j = wstart;
|
|
|
|
wvalue = 1;
|
|
|
|
wend = 0;
|
|
|
|
for (i = 1; i < window; i++) {
|
|
|
|
if (wstart - i < 0)
|
|
|
|
break;
|
|
|
|
if (BN_is_bit_set(p, wstart - i)) {
|
|
|
|
wvalue <<= (i - wend);
|
|
|
|
wvalue |= 1;
|
|
|
|
wend = i;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
/* wend is the size of the current window */
|
|
|
|
j = wend + 1;
|
|
|
|
/* add the 'bytes above' */
|
|
|
|
if (!start)
|
|
|
|
for (i = 0; i < j; i++) {
|
|
|
|
if (!BN_mod_mul_reciprocal(r, r, r, &recp, ctx))
|
|
|
|
goto err;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* wvalue will be an odd number < 2^window */
|
|
|
|
if (!BN_mod_mul_reciprocal(r, r, val[wvalue >> 1], &recp, ctx))
|
|
|
|
goto err;
|
|
|
|
|
|
|
|
/* move the 'window' down further */
|
|
|
|
wstart -= wend + 1;
|
|
|
|
wvalue = 0;
|
|
|
|
start = 0;
|
|
|
|
if (wstart < 0)
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
ret = 1;
|
|
|
|
err:
|
|
|
|
BN_CTX_end(ctx);
|
|
|
|
BN_RECP_CTX_free(&recp);
|
|
|
|
bn_check_top(r);
|
2017-10-17 14:04:09 +00:00
|
|
|
return ret;
|
2015-01-22 03:40:55 +00:00
|
|
|
}
|
2000-06-07 21:29:25 +00:00
|
|
|
|
2000-11-06 21:15:54 +00:00
|
|
|
int BN_mod_exp_mont(BIGNUM *rr, const BIGNUM *a, const BIGNUM *p,
|
2015-01-22 03:40:55 +00:00
|
|
|
const BIGNUM *m, BN_CTX *ctx, BN_MONT_CTX *in_mont)
|
|
|
|
{
|
|
|
|
int i, j, bits, ret = 0, wstart, wend, window, wvalue;
|
|
|
|
int start = 1;
|
|
|
|
BIGNUM *d, *r;
|
|
|
|
const BIGNUM *aa;
|
|
|
|
/* Table of variables obtained from 'ctx' */
|
|
|
|
BIGNUM *val[TABLE_SIZE];
|
|
|
|
BN_MONT_CTX *mont = NULL;
|
|
|
|
|
2017-09-27 10:13:47 +00:00
|
|
|
if (BN_get_flags(p, BN_FLG_CONSTTIME) != 0
|
|
|
|
|| BN_get_flags(a, BN_FLG_CONSTTIME) != 0
|
|
|
|
|| BN_get_flags(m, BN_FLG_CONSTTIME) != 0) {
|
2015-01-22 03:40:55 +00:00
|
|
|
return BN_mod_exp_mont_consttime(rr, a, p, m, ctx, in_mont);
|
|
|
|
}
|
|
|
|
|
|
|
|
bn_check_top(a);
|
|
|
|
bn_check_top(p);
|
|
|
|
bn_check_top(m);
|
|
|
|
|
|
|
|
if (!BN_is_odd(m)) {
|
|
|
|
BNerr(BN_F_BN_MOD_EXP_MONT, BN_R_CALLED_WITH_EVEN_MODULUS);
|
2017-10-17 14:04:09 +00:00
|
|
|
return 0;
|
2015-01-22 03:40:55 +00:00
|
|
|
}
|
|
|
|
bits = BN_num_bits(p);
|
|
|
|
if (bits == 0) {
|
2018-05-24 15:12:52 +00:00
|
|
|
/* x**0 mod 1, or x**0 mod -1 is still zero. */
|
|
|
|
if (BN_abs_is_word(m, 1)) {
|
2015-12-14 15:38:15 +00:00
|
|
|
ret = 1;
|
|
|
|
BN_zero(rr);
|
|
|
|
} else {
|
|
|
|
ret = BN_one(rr);
|
|
|
|
}
|
2015-01-22 03:40:55 +00:00
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
|
|
|
BN_CTX_start(ctx);
|
|
|
|
d = BN_CTX_get(ctx);
|
|
|
|
r = BN_CTX_get(ctx);
|
|
|
|
val[0] = BN_CTX_get(ctx);
|
2017-06-22 10:52:29 +00:00
|
|
|
if (val[0] == NULL)
|
2015-01-22 03:40:55 +00:00
|
|
|
goto err;
|
|
|
|
|
|
|
|
/*
|
|
|
|
* If this is not done, things will break in the montgomery part
|
|
|
|
*/
|
|
|
|
|
|
|
|
if (in_mont != NULL)
|
|
|
|
mont = in_mont;
|
|
|
|
else {
|
|
|
|
if ((mont = BN_MONT_CTX_new()) == NULL)
|
|
|
|
goto err;
|
|
|
|
if (!BN_MONT_CTX_set(mont, m, ctx))
|
|
|
|
goto err;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (a->neg || BN_ucmp(a, m) >= 0) {
|
|
|
|
if (!BN_nnmod(val[0], a, m, ctx))
|
|
|
|
goto err;
|
|
|
|
aa = val[0];
|
|
|
|
} else
|
|
|
|
aa = a;
|
2018-07-06 13:13:15 +00:00
|
|
|
if (!bn_to_mont_fixed_top(val[0], aa, mont, ctx))
|
2015-01-22 03:40:55 +00:00
|
|
|
goto err; /* 1 */
|
|
|
|
|
|
|
|
window = BN_window_bits_for_exponent_size(bits);
|
|
|
|
if (window > 1) {
|
2018-07-06 13:13:15 +00:00
|
|
|
if (!bn_mul_mont_fixed_top(d, val[0], val[0], mont, ctx))
|
2015-01-22 03:40:55 +00:00
|
|
|
goto err; /* 2 */
|
|
|
|
j = 1 << (window - 1);
|
|
|
|
for (i = 1; i < j; i++) {
|
|
|
|
if (((val[i] = BN_CTX_get(ctx)) == NULL) ||
|
2018-07-06 13:13:15 +00:00
|
|
|
!bn_mul_mont_fixed_top(val[i], val[i - 1], d, mont, ctx))
|
2015-01-22 03:40:55 +00:00
|
|
|
goto err;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
start = 1; /* This is used to avoid multiplication etc
|
|
|
|
* when there is only the value '1' in the
|
|
|
|
* buffer. */
|
|
|
|
wvalue = 0; /* The 'value' of the window */
|
|
|
|
wstart = bits - 1; /* The top bit of the window */
|
|
|
|
wend = 0; /* The bottom bit of the window */
|
|
|
|
|
|
|
|
#if 1 /* by Shay Gueron's suggestion */
|
|
|
|
j = m->top; /* borrow j */
|
|
|
|
if (m->d[j - 1] & (((BN_ULONG)1) << (BN_BITS2 - 1))) {
|
|
|
|
if (bn_wexpand(r, j) == NULL)
|
|
|
|
goto err;
|
|
|
|
/* 2^(top*BN_BITS2) - m */
|
|
|
|
r->d[0] = (0 - m->d[0]) & BN_MASK2;
|
|
|
|
for (i = 1; i < j; i++)
|
|
|
|
r->d[i] = (~m->d[i]) & BN_MASK2;
|
|
|
|
r->top = j;
|
2018-07-06 13:13:15 +00:00
|
|
|
r->flags |= BN_FLG_FIXED_TOP;
|
2015-01-22 03:40:55 +00:00
|
|
|
} else
|
2013-06-18 08:39:38 +00:00
|
|
|
#endif
|
2018-07-06 13:13:15 +00:00
|
|
|
if (!bn_to_mont_fixed_top(r, BN_value_one(), mont, ctx))
|
2015-01-22 03:40:55 +00:00
|
|
|
goto err;
|
|
|
|
for (;;) {
|
|
|
|
if (BN_is_bit_set(p, wstart) == 0) {
|
|
|
|
if (!start) {
|
2018-07-06 13:13:15 +00:00
|
|
|
if (!bn_mul_mont_fixed_top(r, r, r, mont, ctx))
|
2015-01-22 03:40:55 +00:00
|
|
|
goto err;
|
|
|
|
}
|
|
|
|
if (wstart == 0)
|
|
|
|
break;
|
|
|
|
wstart--;
|
|
|
|
continue;
|
|
|
|
}
|
|
|
|
/*
|
|
|
|
* We now have wstart on a 'set' bit, we now need to work out how bit
|
|
|
|
* a window to do. To do this we need to scan forward until the last
|
|
|
|
* set bit before the end of the window
|
|
|
|
*/
|
|
|
|
j = wstart;
|
|
|
|
wvalue = 1;
|
|
|
|
wend = 0;
|
|
|
|
for (i = 1; i < window; i++) {
|
|
|
|
if (wstart - i < 0)
|
|
|
|
break;
|
|
|
|
if (BN_is_bit_set(p, wstart - i)) {
|
|
|
|
wvalue <<= (i - wend);
|
|
|
|
wvalue |= 1;
|
|
|
|
wend = i;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
/* wend is the size of the current window */
|
|
|
|
j = wend + 1;
|
|
|
|
/* add the 'bytes above' */
|
|
|
|
if (!start)
|
|
|
|
for (i = 0; i < j; i++) {
|
2018-07-06 13:13:15 +00:00
|
|
|
if (!bn_mul_mont_fixed_top(r, r, r, mont, ctx))
|
2015-01-22 03:40:55 +00:00
|
|
|
goto err;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* wvalue will be an odd number < 2^window */
|
2018-07-06 13:13:15 +00:00
|
|
|
if (!bn_mul_mont_fixed_top(r, r, val[wvalue >> 1], mont, ctx))
|
2015-01-22 03:40:55 +00:00
|
|
|
goto err;
|
|
|
|
|
|
|
|
/* move the 'window' down further */
|
|
|
|
wstart -= wend + 1;
|
|
|
|
wvalue = 0;
|
|
|
|
start = 0;
|
|
|
|
if (wstart < 0)
|
|
|
|
break;
|
|
|
|
}
|
2018-07-06 13:13:15 +00:00
|
|
|
/*
|
|
|
|
* Done with zero-padded intermediate BIGNUMs. Final BN_from_montgomery
|
|
|
|
* removes padding [if any] and makes return value suitable for public
|
|
|
|
* API consumer.
|
|
|
|
*/
|
2013-07-05 18:52:58 +00:00
|
|
|
#if defined(SPARC_T4_MONT)
|
2015-01-22 03:40:55 +00:00
|
|
|
if (OPENSSL_sparcv9cap_P[0] & (SPARCV9_VIS3 | SPARCV9_PREFER_FPU)) {
|
|
|
|
j = mont->N.top; /* borrow j */
|
|
|
|
val[0]->d[0] = 1; /* borrow val[0] */
|
|
|
|
for (i = 1; i < j; i++)
|
|
|
|
val[0]->d[i] = 0;
|
|
|
|
val[0]->top = j;
|
|
|
|
if (!BN_mod_mul_montgomery(rr, r, val[0], mont, ctx))
|
|
|
|
goto err;
|
|
|
|
} else
|
2013-06-18 08:39:38 +00:00
|
|
|
#endif
|
2015-01-22 03:40:55 +00:00
|
|
|
if (!BN_from_montgomery(rr, r, mont, ctx))
|
|
|
|
goto err;
|
|
|
|
ret = 1;
|
|
|
|
err:
|
2015-05-01 01:37:06 +00:00
|
|
|
if (in_mont == NULL)
|
2015-01-22 03:40:55 +00:00
|
|
|
BN_MONT_CTX_free(mont);
|
|
|
|
BN_CTX_end(ctx);
|
|
|
|
bn_check_top(rr);
|
2017-10-17 14:04:09 +00:00
|
|
|
return ret;
|
2015-01-22 03:40:55 +00:00
|
|
|
}
|
2000-06-07 21:29:25 +00:00
|
|
|
|
2013-06-18 08:39:38 +00:00
|
|
|
static BN_ULONG bn_get_bits(const BIGNUM *a, int bitpos)
|
2015-01-22 03:40:55 +00:00
|
|
|
{
|
|
|
|
BN_ULONG ret = 0;
|
|
|
|
int wordpos;
|
|
|
|
|
|
|
|
wordpos = bitpos / BN_BITS2;
|
|
|
|
bitpos %= BN_BITS2;
|
|
|
|
if (wordpos >= 0 && wordpos < a->top) {
|
|
|
|
ret = a->d[wordpos] & BN_MASK2;
|
|
|
|
if (bitpos) {
|
|
|
|
ret >>= bitpos;
|
|
|
|
if (++wordpos < a->top)
|
|
|
|
ret |= a->d[wordpos] << (BN_BITS2 - bitpos);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
return ret & BN_MASK2;
|
2013-06-18 08:39:38 +00:00
|
|
|
}
|
2005-05-16 01:43:31 +00:00
|
|
|
|
2015-01-22 03:40:55 +00:00
|
|
|
/*
|
|
|
|
* BN_mod_exp_mont_consttime() stores the precomputed powers in a specific
|
|
|
|
* layout so that accessing any of these table values shows the same access
|
|
|
|
* pattern as far as cache lines are concerned. The following functions are
|
|
|
|
* used to transfer a BIGNUM from/to that table.
|
|
|
|
*/
|
2005-05-16 01:43:31 +00:00
|
|
|
|
2015-01-22 03:40:55 +00:00
|
|
|
static int MOD_EXP_CTIME_COPY_TO_PREBUF(const BIGNUM *b, int top,
|
|
|
|
unsigned char *buf, int idx,
|
2016-01-25 19:38:38 +00:00
|
|
|
int window)
|
2015-01-22 03:40:55 +00:00
|
|
|
{
|
2016-01-25 19:38:38 +00:00
|
|
|
int i, j;
|
|
|
|
int width = 1 << window;
|
|
|
|
BN_ULONG *table = (BN_ULONG *)buf;
|
2005-05-16 01:43:31 +00:00
|
|
|
|
2015-01-22 03:40:55 +00:00
|
|
|
if (top > b->top)
|
|
|
|
top = b->top; /* this works because 'buf' is explicitly
|
|
|
|
* zeroed */
|
2016-01-25 19:38:38 +00:00
|
|
|
for (i = 0, j = idx; i < top; i++, j += width) {
|
|
|
|
table[j] = b->d[i];
|
2015-01-22 03:40:55 +00:00
|
|
|
}
|
2005-05-16 01:43:31 +00:00
|
|
|
|
2015-01-22 03:40:55 +00:00
|
|
|
return 1;
|
|
|
|
}
|
2005-05-16 01:43:31 +00:00
|
|
|
|
2015-01-22 03:40:55 +00:00
|
|
|
static int MOD_EXP_CTIME_COPY_FROM_PREBUF(BIGNUM *b, int top,
|
|
|
|
unsigned char *buf, int idx,
|
2016-01-25 19:38:38 +00:00
|
|
|
int window)
|
2015-01-22 03:40:55 +00:00
|
|
|
{
|
2016-01-25 19:38:38 +00:00
|
|
|
int i, j;
|
|
|
|
int width = 1 << window;
|
2016-05-26 19:40:09 +00:00
|
|
|
/*
|
|
|
|
* We declare table 'volatile' in order to discourage compiler
|
|
|
|
* from reordering loads from the table. Concern is that if
|
|
|
|
* reordered in specific manner loads might give away the
|
|
|
|
* information we are trying to conceal. Some would argue that
|
|
|
|
* compiler can reorder them anyway, but it can as well be
|
|
|
|
* argued that doing so would be violation of standard...
|
|
|
|
*/
|
2016-01-25 19:38:38 +00:00
|
|
|
volatile BN_ULONG *table = (volatile BN_ULONG *)buf;
|
2005-05-16 01:43:31 +00:00
|
|
|
|
2015-01-22 03:40:55 +00:00
|
|
|
if (bn_wexpand(b, top) == NULL)
|
|
|
|
return 0;
|
2005-05-16 01:43:31 +00:00
|
|
|
|
2016-01-25 19:38:38 +00:00
|
|
|
if (window <= 3) {
|
|
|
|
for (i = 0; i < top; i++, table += width) {
|
|
|
|
BN_ULONG acc = 0;
|
|
|
|
|
|
|
|
for (j = 0; j < width; j++) {
|
|
|
|
acc |= table[j] &
|
|
|
|
((BN_ULONG)0 - (constant_time_eq_int(j,idx)&1));
|
|
|
|
}
|
|
|
|
|
|
|
|
b->d[i] = acc;
|
|
|
|
}
|
|
|
|
} else {
|
|
|
|
int xstride = 1 << (window - 2);
|
|
|
|
BN_ULONG y0, y1, y2, y3;
|
|
|
|
|
|
|
|
i = idx >> (window - 2); /* equivalent of idx / xstride */
|
|
|
|
idx &= xstride - 1; /* equivalent of idx % xstride */
|
|
|
|
|
|
|
|
y0 = (BN_ULONG)0 - (constant_time_eq_int(i,0)&1);
|
|
|
|
y1 = (BN_ULONG)0 - (constant_time_eq_int(i,1)&1);
|
|
|
|
y2 = (BN_ULONG)0 - (constant_time_eq_int(i,2)&1);
|
|
|
|
y3 = (BN_ULONG)0 - (constant_time_eq_int(i,3)&1);
|
|
|
|
|
|
|
|
for (i = 0; i < top; i++, table += width) {
|
|
|
|
BN_ULONG acc = 0;
|
|
|
|
|
|
|
|
for (j = 0; j < xstride; j++) {
|
|
|
|
acc |= ( (table[j + 0 * xstride] & y0) |
|
|
|
|
(table[j + 1 * xstride] & y1) |
|
|
|
|
(table[j + 2 * xstride] & y2) |
|
|
|
|
(table[j + 3 * xstride] & y3) )
|
|
|
|
& ((BN_ULONG)0 - (constant_time_eq_int(j,idx)&1));
|
|
|
|
}
|
|
|
|
|
|
|
|
b->d[i] = acc;
|
|
|
|
}
|
2015-01-22 03:40:55 +00:00
|
|
|
}
|
2005-05-16 01:43:31 +00:00
|
|
|
|
2015-01-22 03:40:55 +00:00
|
|
|
b->top = top;
|
2018-07-06 13:13:15 +00:00
|
|
|
b->flags |= BN_FLG_FIXED_TOP;
|
2015-01-22 03:40:55 +00:00
|
|
|
return 1;
|
|
|
|
}
|
2005-05-16 01:43:31 +00:00
|
|
|
|
2015-01-22 03:40:55 +00:00
|
|
|
/*
|
|
|
|
* Given a pointer value, compute the next address that is a cache line
|
|
|
|
* multiple.
|
|
|
|
*/
|
2005-05-16 01:43:31 +00:00
|
|
|
#define MOD_EXP_CTIME_ALIGN(x_) \
|
2015-01-22 03:40:55 +00:00
|
|
|
((unsigned char*)(x_) + (MOD_EXP_CTIME_MIN_CACHE_LINE_WIDTH - (((size_t)(x_)) & (MOD_EXP_CTIME_MIN_CACHE_LINE_MASK))))
|
|
|
|
|
|
|
|
/*
|
|
|
|
* This variant of BN_mod_exp_mont() uses fixed windows and the special
|
|
|
|
* precomputation memory layout to limit data-dependency to a minimum to
|
|
|
|
* protect secret exponents (cf. the hyper-threading timing attacks pointed
|
|
|
|
* out by Colin Percival,
|
2015-12-19 14:38:17 +00:00
|
|
|
* http://www.daemonology.net/hyperthreading-considered-harmful/)
|
2005-05-16 01:43:31 +00:00
|
|
|
*/
|
|
|
|
int BN_mod_exp_mont_consttime(BIGNUM *rr, const BIGNUM *a, const BIGNUM *p,
|
2015-01-22 03:40:55 +00:00
|
|
|
const BIGNUM *m, BN_CTX *ctx,
|
|
|
|
BN_MONT_CTX *in_mont)
|
|
|
|
{
|
bn/bn_exp.c: mitigation of the One-and-Done side-channel attack.
The One&Done attack, which is described in a paper to appear in the
USENIX Security'18 conference, uses EM emanations to recover the values
of the bits that are obtained using BN_is_bit_set while constructing
the value of the window in BN_mod_exp_consttime. The EM signal changes
slightly depending on the value of the bit, and since the lookup of a
bit is surrounded by highly regular execution (constant-time Montgomery
multiplications) the attack is able to isolate the (very brief) part of
the signal that changes depending on the bit. Although the change is
slight, the attack recovers it successfully >90% of the time on several
phones and IoT devices (all with ARM processors with clock rates around
1GHz), so after only one RSA decryption more than 90% of the bits in
d_p and d_q are recovered correctly, which enables rapid recovery of
the full RSA key using an algorithm (also described in the paper) that
modifies the branch-and-prune approach for a situation in which the
exponents' bits are recovered with errors, i.e. where we do not know
a priori which bits are correctly recovered.
The mitigation for the attack is relatively simple - all the bits of
the window are obtained at once, along with other bits so that an
entire integer's worth of bits are obtained together using masking and
shifts, without unnecessarily considering each bit in isolation. This
improves performance somewhat (one call to bn_get_bits is faster than
several calls to BN_is_bit_set), so the attacker now gets one signal
snippet per window (rather than one per bit) in which the signal is
affected by all bits in the integer (rather than just the one bit).
Reviewed-by: Andy Polyakov <appro@openssl.org>
Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/6276)
2018-05-16 17:59:36 +00:00
|
|
|
int i, bits, ret = 0, window, wvalue, wmask, window0;
|
2015-01-22 03:40:55 +00:00
|
|
|
int top;
|
|
|
|
BN_MONT_CTX *mont = NULL;
|
|
|
|
|
|
|
|
int numPowers;
|
|
|
|
unsigned char *powerbufFree = NULL;
|
|
|
|
int powerbufLen = 0;
|
|
|
|
unsigned char *powerbuf = NULL;
|
|
|
|
BIGNUM tmp, am;
|
2013-07-05 18:52:58 +00:00
|
|
|
#if defined(SPARC_T4_MONT)
|
2015-01-22 03:40:55 +00:00
|
|
|
unsigned int t4 = 0;
|
2012-11-17 10:34:11 +00:00
|
|
|
#endif
|
2005-05-16 01:43:31 +00:00
|
|
|
|
2015-01-22 03:40:55 +00:00
|
|
|
bn_check_top(a);
|
|
|
|
bn_check_top(p);
|
|
|
|
bn_check_top(m);
|
|
|
|
|
2015-08-31 13:51:27 +00:00
|
|
|
if (!BN_is_odd(m)) {
|
2015-01-22 03:40:55 +00:00
|
|
|
BNerr(BN_F_BN_MOD_EXP_MONT_CONSTTIME, BN_R_CALLED_WITH_EVEN_MODULUS);
|
2017-10-17 14:04:09 +00:00
|
|
|
return 0;
|
2015-01-22 03:40:55 +00:00
|
|
|
}
|
2015-08-31 13:51:27 +00:00
|
|
|
|
|
|
|
top = m->top;
|
|
|
|
|
Don't leak the exponent bit width in BN_mod_exp_mont_consttime.
The exponent here is one of d, dmp1, or dmq1 for RSA. This value and its
bit length are both secret. The only public upper bound is the bit width
of the corresponding modulus (RSA n, p, and q, respectively).
Although BN_num_bits is constant-time (sort of; see bn_correct_top notes
in preceding patch), this does not fix the root problem, which is that
the windows are based on the minimal bit width, not the upper bound. We
could use BN_num_bits(m), but BN_mod_exp_mont_consttime is public API
and may be called with larger exponents. Instead, use all top*BN_BITS2
bits in the BIGNUM. This is still sensitive to the long-standing
bn_correct_top leak, but we need to fix that regardless.
This may cause us to do a handful of extra multiplications for RSA keys
which are just above a whole number of words, but that is not a standard
RSA key size.
Reviewed-by: Paul Dale <paul.dale@oracle.com>
Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/5154)
2018-01-23 18:57:10 +00:00
|
|
|
/*
|
|
|
|
* Use all bits stored in |p|, rather than |BN_num_bits|, so we do not leak
|
|
|
|
* whether the top bits are zero.
|
|
|
|
*/
|
|
|
|
bits = p->top * BN_BITS2;
|
2015-01-22 03:40:55 +00:00
|
|
|
if (bits == 0) {
|
2018-05-24 15:12:52 +00:00
|
|
|
/* x**0 mod 1, or x**0 mod -1 is still zero. */
|
|
|
|
if (BN_abs_is_word(m, 1)) {
|
2015-12-14 15:38:15 +00:00
|
|
|
ret = 1;
|
|
|
|
BN_zero(rr);
|
|
|
|
} else {
|
|
|
|
ret = BN_one(rr);
|
|
|
|
}
|
2015-01-22 03:40:55 +00:00
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
|
|
|
BN_CTX_start(ctx);
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Allocate a montgomery context if it was not supplied by the caller. If
|
|
|
|
* this is not done, things will break in the montgomery part.
|
|
|
|
*/
|
|
|
|
if (in_mont != NULL)
|
|
|
|
mont = in_mont;
|
|
|
|
else {
|
|
|
|
if ((mont = BN_MONT_CTX_new()) == NULL)
|
|
|
|
goto err;
|
|
|
|
if (!BN_MONT_CTX_set(mont, m, ctx))
|
|
|
|
goto err;
|
|
|
|
}
|
2005-05-16 01:43:31 +00:00
|
|
|
|
2018-09-11 20:49:28 +00:00
|
|
|
if (a->neg || BN_ucmp(a, m) >= 0) {
|
|
|
|
BIGNUM *reduced = BN_CTX_get(ctx);
|
|
|
|
if (reduced == NULL
|
|
|
|
|| !BN_nnmod(reduced, a, m, ctx)) {
|
|
|
|
goto err;
|
|
|
|
}
|
|
|
|
a = reduced;
|
|
|
|
}
|
|
|
|
|
2013-07-05 19:39:47 +00:00
|
|
|
#ifdef RSAZ_ENABLED
|
2018-09-11 20:49:28 +00:00
|
|
|
/*
|
|
|
|
* If the size of the operands allow it, perform the optimized
|
|
|
|
* RSAZ exponentiation. For further information see
|
|
|
|
* crypto/bn/rsaz_exp.c and accompanying assembly modules.
|
|
|
|
*/
|
|
|
|
if ((16 == a->top) && (16 == p->top) && (BN_num_bits(m) == 1024)
|
|
|
|
&& rsaz_avx2_eligible()) {
|
|
|
|
if (NULL == bn_wexpand(rr, 16))
|
2015-01-22 03:40:55 +00:00
|
|
|
goto err;
|
2018-09-11 20:49:28 +00:00
|
|
|
RSAZ_1024_mod_exp_avx2(rr->d, a->d, p->d, m->d, mont->RR.d,
|
|
|
|
mont->n0[0]);
|
|
|
|
rr->top = 16;
|
|
|
|
rr->neg = 0;
|
|
|
|
bn_correct_top(rr);
|
|
|
|
ret = 1;
|
|
|
|
goto err;
|
|
|
|
} else if ((8 == a->top) && (8 == p->top) && (BN_num_bits(m) == 512)) {
|
|
|
|
if (NULL == bn_wexpand(rr, 8))
|
2015-01-22 03:40:55 +00:00
|
|
|
goto err;
|
2018-09-11 20:49:28 +00:00
|
|
|
RSAZ_512_mod_exp(rr->d, a->d, p->d, m->d, mont->n0[0], mont->RR.d);
|
|
|
|
rr->top = 8;
|
|
|
|
rr->neg = 0;
|
|
|
|
bn_correct_top(rr);
|
|
|
|
ret = 1;
|
|
|
|
goto err;
|
2015-01-22 03:40:55 +00:00
|
|
|
}
|
2013-07-05 19:39:47 +00:00
|
|
|
#endif
|
|
|
|
|
2015-01-22 03:40:55 +00:00
|
|
|
/* Get the window size to use with size of p. */
|
|
|
|
window = BN_window_bits_for_ctime_exponent_size(bits);
|
2013-07-05 18:52:58 +00:00
|
|
|
#if defined(SPARC_T4_MONT)
|
2015-01-22 03:40:55 +00:00
|
|
|
if (window >= 5 && (top & 15) == 0 && top <= 64 &&
|
|
|
|
(OPENSSL_sparcv9cap_P[1] & (CFR_MONTMUL | CFR_MONTSQR)) ==
|
|
|
|
(CFR_MONTMUL | CFR_MONTSQR) && (t4 = OPENSSL_sparcv9cap_P[0]))
|
|
|
|
window = 5;
|
|
|
|
else
|
2012-11-17 10:34:11 +00:00
|
|
|
#endif
|
2011-08-12 16:44:32 +00:00
|
|
|
#if defined(OPENSSL_BN_ASM_MONT5)
|
2015-01-22 03:40:55 +00:00
|
|
|
if (window >= 5) {
|
|
|
|
window = 5; /* ~5% improvement for RSA2048 sign, and even
|
|
|
|
* for RSA4096 */
|
2016-01-25 22:41:01 +00:00
|
|
|
/* reserve space for mont->N.d[] copy */
|
|
|
|
powerbufLen += top * sizeof(mont->N.d[0]);
|
2015-01-22 03:40:55 +00:00
|
|
|
}
|
2011-08-12 16:44:32 +00:00
|
|
|
#endif
|
2015-01-22 03:40:55 +00:00
|
|
|
(void)0;
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Allocate a buffer large enough to hold all of the pre-computed powers
|
|
|
|
* of am, am itself and tmp.
|
|
|
|
*/
|
|
|
|
numPowers = 1 << window;
|
|
|
|
powerbufLen += sizeof(m->d[0]) * (top * numPowers +
|
|
|
|
((2 * top) >
|
|
|
|
numPowers ? (2 * top) : numPowers));
|
2011-08-27 19:38:55 +00:00
|
|
|
#ifdef alloca
|
2015-01-22 03:40:55 +00:00
|
|
|
if (powerbufLen < 3072)
|
|
|
|
powerbufFree =
|
|
|
|
alloca(powerbufLen + MOD_EXP_CTIME_MIN_CACHE_LINE_WIDTH);
|
|
|
|
else
|
2011-08-27 19:38:55 +00:00
|
|
|
#endif
|
2015-01-22 03:40:55 +00:00
|
|
|
if ((powerbufFree =
|
2015-04-28 19:28:14 +00:00
|
|
|
OPENSSL_malloc(powerbufLen + MOD_EXP_CTIME_MIN_CACHE_LINE_WIDTH))
|
2015-01-22 03:40:55 +00:00
|
|
|
== NULL)
|
|
|
|
goto err;
|
|
|
|
|
|
|
|
powerbuf = MOD_EXP_CTIME_ALIGN(powerbufFree);
|
|
|
|
memset(powerbuf, 0, powerbufLen);
|
2005-05-16 01:43:31 +00:00
|
|
|
|
2011-08-27 19:38:55 +00:00
|
|
|
#ifdef alloca
|
2015-01-22 03:40:55 +00:00
|
|
|
if (powerbufLen < 3072)
|
|
|
|
powerbufFree = NULL;
|
2011-08-27 19:38:55 +00:00
|
|
|
#endif
|
2011-08-12 16:44:32 +00:00
|
|
|
|
2015-01-22 03:40:55 +00:00
|
|
|
/* lay down tmp and am right after powers table */
|
|
|
|
tmp.d = (BN_ULONG *)(powerbuf + sizeof(m->d[0]) * top * numPowers);
|
|
|
|
am.d = tmp.d + top;
|
|
|
|
tmp.top = am.top = 0;
|
|
|
|
tmp.dmax = am.dmax = top;
|
|
|
|
tmp.neg = am.neg = 0;
|
|
|
|
tmp.flags = am.flags = BN_FLG_STATIC_DATA;
|
|
|
|
|
|
|
|
/* prepare a^0 in Montgomery domain */
|
|
|
|
#if 1 /* by Shay Gueron's suggestion */
|
|
|
|
if (m->d[top - 1] & (((BN_ULONG)1) << (BN_BITS2 - 1))) {
|
|
|
|
/* 2^(top*BN_BITS2) - m */
|
|
|
|
tmp.d[0] = (0 - m->d[0]) & BN_MASK2;
|
|
|
|
for (i = 1; i < top; i++)
|
|
|
|
tmp.d[i] = (~m->d[i]) & BN_MASK2;
|
|
|
|
tmp.top = top;
|
|
|
|
} else
|
2011-10-17 17:41:49 +00:00
|
|
|
#endif
|
2018-07-06 13:13:15 +00:00
|
|
|
if (!bn_to_mont_fixed_top(&tmp, BN_value_one(), mont, ctx))
|
2015-01-22 03:40:55 +00:00
|
|
|
goto err;
|
|
|
|
|
|
|
|
/* prepare a^1 in Montgomery domain */
|
2018-09-11 20:49:28 +00:00
|
|
|
if (!bn_to_mont_fixed_top(&am, a, mont, ctx))
|
2015-01-22 03:40:55 +00:00
|
|
|
goto err;
|
2011-08-12 16:44:32 +00:00
|
|
|
|
2013-07-05 18:52:58 +00:00
|
|
|
#if defined(SPARC_T4_MONT)
|
2015-01-22 03:40:55 +00:00
|
|
|
if (t4) {
|
|
|
|
typedef int (*bn_pwr5_mont_f) (BN_ULONG *tp, const BN_ULONG *np,
|
|
|
|
const BN_ULONG *n0, const void *table,
|
|
|
|
int power, int bits);
|
|
|
|
int bn_pwr5_mont_t4_8(BN_ULONG *tp, const BN_ULONG *np,
|
|
|
|
const BN_ULONG *n0, const void *table,
|
|
|
|
int power, int bits);
|
|
|
|
int bn_pwr5_mont_t4_16(BN_ULONG *tp, const BN_ULONG *np,
|
|
|
|
const BN_ULONG *n0, const void *table,
|
|
|
|
int power, int bits);
|
|
|
|
int bn_pwr5_mont_t4_24(BN_ULONG *tp, const BN_ULONG *np,
|
|
|
|
const BN_ULONG *n0, const void *table,
|
|
|
|
int power, int bits);
|
|
|
|
int bn_pwr5_mont_t4_32(BN_ULONG *tp, const BN_ULONG *np,
|
|
|
|
const BN_ULONG *n0, const void *table,
|
|
|
|
int power, int bits);
|
|
|
|
static const bn_pwr5_mont_f pwr5_funcs[4] = {
|
|
|
|
bn_pwr5_mont_t4_8, bn_pwr5_mont_t4_16,
|
|
|
|
bn_pwr5_mont_t4_24, bn_pwr5_mont_t4_32
|
|
|
|
};
|
|
|
|
bn_pwr5_mont_f pwr5_worker = pwr5_funcs[top / 16 - 1];
|
|
|
|
|
|
|
|
typedef int (*bn_mul_mont_f) (BN_ULONG *rp, const BN_ULONG *ap,
|
|
|
|
const void *bp, const BN_ULONG *np,
|
|
|
|
const BN_ULONG *n0);
|
|
|
|
int bn_mul_mont_t4_8(BN_ULONG *rp, const BN_ULONG *ap, const void *bp,
|
|
|
|
const BN_ULONG *np, const BN_ULONG *n0);
|
|
|
|
int bn_mul_mont_t4_16(BN_ULONG *rp, const BN_ULONG *ap,
|
|
|
|
const void *bp, const BN_ULONG *np,
|
|
|
|
const BN_ULONG *n0);
|
|
|
|
int bn_mul_mont_t4_24(BN_ULONG *rp, const BN_ULONG *ap,
|
|
|
|
const void *bp, const BN_ULONG *np,
|
|
|
|
const BN_ULONG *n0);
|
|
|
|
int bn_mul_mont_t4_32(BN_ULONG *rp, const BN_ULONG *ap,
|
|
|
|
const void *bp, const BN_ULONG *np,
|
|
|
|
const BN_ULONG *n0);
|
|
|
|
static const bn_mul_mont_f mul_funcs[4] = {
|
|
|
|
bn_mul_mont_t4_8, bn_mul_mont_t4_16,
|
|
|
|
bn_mul_mont_t4_24, bn_mul_mont_t4_32
|
|
|
|
};
|
|
|
|
bn_mul_mont_f mul_worker = mul_funcs[top / 16 - 1];
|
|
|
|
|
|
|
|
void bn_mul_mont_vis3(BN_ULONG *rp, const BN_ULONG *ap,
|
|
|
|
const void *bp, const BN_ULONG *np,
|
|
|
|
const BN_ULONG *n0, int num);
|
|
|
|
void bn_mul_mont_t4(BN_ULONG *rp, const BN_ULONG *ap,
|
|
|
|
const void *bp, const BN_ULONG *np,
|
|
|
|
const BN_ULONG *n0, int num);
|
|
|
|
void bn_mul_mont_gather5_t4(BN_ULONG *rp, const BN_ULONG *ap,
|
|
|
|
const void *table, const BN_ULONG *np,
|
|
|
|
const BN_ULONG *n0, int num, int power);
|
|
|
|
void bn_flip_n_scatter5_t4(const BN_ULONG *inp, size_t num,
|
|
|
|
void *table, size_t power);
|
|
|
|
void bn_gather5_t4(BN_ULONG *out, size_t num,
|
|
|
|
void *table, size_t power);
|
|
|
|
void bn_flip_t4(BN_ULONG *dst, BN_ULONG *src, size_t num);
|
|
|
|
|
|
|
|
BN_ULONG *np = mont->N.d, *n0 = mont->n0;
|
|
|
|
int stride = 5 * (6 - (top / 16 - 1)); /* multiple of 5, but less
|
|
|
|
* than 32 */
|
|
|
|
|
|
|
|
/*
|
|
|
|
* BN_to_montgomery can contaminate words above .top [in
|
|
|
|
* BN_DEBUG[_DEBUG] build]...
|
|
|
|
*/
|
|
|
|
for (i = am.top; i < top; i++)
|
|
|
|
am.d[i] = 0;
|
|
|
|
for (i = tmp.top; i < top; i++)
|
|
|
|
tmp.d[i] = 0;
|
|
|
|
|
|
|
|
bn_flip_n_scatter5_t4(tmp.d, top, powerbuf, 0);
|
|
|
|
bn_flip_n_scatter5_t4(am.d, top, powerbuf, 1);
|
|
|
|
if (!(*mul_worker) (tmp.d, am.d, am.d, np, n0) &&
|
|
|
|
!(*mul_worker) (tmp.d, am.d, am.d, np, n0))
|
|
|
|
bn_mul_mont_vis3(tmp.d, am.d, am.d, np, n0, top);
|
|
|
|
bn_flip_n_scatter5_t4(tmp.d, top, powerbuf, 2);
|
|
|
|
|
|
|
|
for (i = 3; i < 32; i++) {
|
|
|
|
/* Calculate a^i = a^(i-1) * a */
|
|
|
|
if (!(*mul_worker) (tmp.d, tmp.d, am.d, np, n0) &&
|
|
|
|
!(*mul_worker) (tmp.d, tmp.d, am.d, np, n0))
|
|
|
|
bn_mul_mont_vis3(tmp.d, tmp.d, am.d, np, n0, top);
|
|
|
|
bn_flip_n_scatter5_t4(tmp.d, top, powerbuf, i);
|
|
|
|
}
|
|
|
|
|
|
|
|
/* switch to 64-bit domain */
|
|
|
|
np = alloca(top * sizeof(BN_ULONG));
|
|
|
|
top /= 2;
|
|
|
|
bn_flip_t4(np, mont->N.d, top);
|
|
|
|
|
2018-06-13 12:00:04 +00:00
|
|
|
/*
|
|
|
|
* The exponent may not have a whole number of fixed-size windows.
|
|
|
|
* To simplify the main loop, the initial window has between 1 and
|
|
|
|
* full-window-size bits such that what remains is always a whole
|
|
|
|
* number of windows
|
|
|
|
*/
|
|
|
|
window0 = (bits - 1) % 5 + 1;
|
|
|
|
wmask = (1 << window0) - 1;
|
|
|
|
bits -= window0;
|
|
|
|
wvalue = bn_get_bits(p, bits) & wmask;
|
2015-01-22 03:40:55 +00:00
|
|
|
bn_gather5_t4(tmp.d, top, powerbuf, wvalue);
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Scan the exponent one window at a time starting from the most
|
|
|
|
* significant bits.
|
|
|
|
*/
|
2018-06-13 12:00:04 +00:00
|
|
|
while (bits > 0) {
|
2015-01-22 03:40:55 +00:00
|
|
|
if (bits < stride)
|
2018-06-13 12:00:04 +00:00
|
|
|
stride = bits;
|
2015-01-22 03:40:55 +00:00
|
|
|
bits -= stride;
|
2018-06-13 12:00:04 +00:00
|
|
|
wvalue = bn_get_bits(p, bits);
|
2015-01-22 03:40:55 +00:00
|
|
|
|
|
|
|
if ((*pwr5_worker) (tmp.d, np, n0, powerbuf, wvalue, stride))
|
|
|
|
continue;
|
|
|
|
/* retry once and fall back */
|
|
|
|
if ((*pwr5_worker) (tmp.d, np, n0, powerbuf, wvalue, stride))
|
|
|
|
continue;
|
|
|
|
|
|
|
|
bits += stride - 5;
|
|
|
|
wvalue >>= stride - 5;
|
|
|
|
wvalue &= 31;
|
|
|
|
bn_mul_mont_t4(tmp.d, tmp.d, tmp.d, np, n0, top);
|
|
|
|
bn_mul_mont_t4(tmp.d, tmp.d, tmp.d, np, n0, top);
|
|
|
|
bn_mul_mont_t4(tmp.d, tmp.d, tmp.d, np, n0, top);
|
|
|
|
bn_mul_mont_t4(tmp.d, tmp.d, tmp.d, np, n0, top);
|
|
|
|
bn_mul_mont_t4(tmp.d, tmp.d, tmp.d, np, n0, top);
|
|
|
|
bn_mul_mont_gather5_t4(tmp.d, tmp.d, powerbuf, np, n0, top,
|
|
|
|
wvalue);
|
|
|
|
}
|
|
|
|
|
|
|
|
bn_flip_t4(tmp.d, tmp.d, top);
|
|
|
|
top *= 2;
|
|
|
|
/* back to 32-bit domain */
|
|
|
|
tmp.top = top;
|
|
|
|
bn_correct_top(&tmp);
|
|
|
|
OPENSSL_cleanse(np, top * sizeof(BN_ULONG));
|
|
|
|
} else
|
2012-11-17 10:34:11 +00:00
|
|
|
#endif
|
2011-08-12 16:44:32 +00:00
|
|
|
#if defined(OPENSSL_BN_ASM_MONT5)
|
2015-01-22 03:40:55 +00:00
|
|
|
if (window == 5 && top > 1) {
|
|
|
|
/*
|
|
|
|
* This optimization uses ideas from http://eprint.iacr.org/2011/239,
|
|
|
|
* specifically optimization of cache-timing attack countermeasures
|
|
|
|
* and pre-computation optimization.
|
|
|
|
*/
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Dedicated window==4 case improves 512-bit RSA sign by ~15%, but as
|
|
|
|
* 512-bit RSA is hardly relevant, we omit it to spare size...
|
|
|
|
*/
|
|
|
|
void bn_mul_mont_gather5(BN_ULONG *rp, const BN_ULONG *ap,
|
|
|
|
const void *table, const BN_ULONG *np,
|
|
|
|
const BN_ULONG *n0, int num, int power);
|
|
|
|
void bn_scatter5(const BN_ULONG *inp, size_t num,
|
|
|
|
void *table, size_t power);
|
|
|
|
void bn_gather5(BN_ULONG *out, size_t num, void *table, size_t power);
|
|
|
|
void bn_power5(BN_ULONG *rp, const BN_ULONG *ap,
|
|
|
|
const void *table, const BN_ULONG *np,
|
|
|
|
const BN_ULONG *n0, int num, int power);
|
|
|
|
int bn_get_bits5(const BN_ULONG *ap, int off);
|
|
|
|
int bn_from_montgomery(BN_ULONG *rp, const BN_ULONG *ap,
|
|
|
|
const BN_ULONG *not_used, const BN_ULONG *np,
|
|
|
|
const BN_ULONG *n0, int num);
|
|
|
|
|
2016-01-25 22:41:01 +00:00
|
|
|
BN_ULONG *n0 = mont->n0, *np;
|
2015-01-22 03:40:55 +00:00
|
|
|
|
|
|
|
/*
|
|
|
|
* BN_to_montgomery can contaminate words above .top [in
|
|
|
|
* BN_DEBUG[_DEBUG] build]...
|
|
|
|
*/
|
|
|
|
for (i = am.top; i < top; i++)
|
|
|
|
am.d[i] = 0;
|
|
|
|
for (i = tmp.top; i < top; i++)
|
|
|
|
tmp.d[i] = 0;
|
|
|
|
|
2016-01-25 22:41:01 +00:00
|
|
|
/*
|
|
|
|
* copy mont->N.d[] to improve cache locality
|
|
|
|
*/
|
|
|
|
for (np = am.d + top, i = 0; i < top; i++)
|
|
|
|
np[i] = mont->N.d[i];
|
2015-01-22 03:40:55 +00:00
|
|
|
|
|
|
|
bn_scatter5(tmp.d, top, powerbuf, 0);
|
|
|
|
bn_scatter5(am.d, am.top, powerbuf, 1);
|
|
|
|
bn_mul_mont(tmp.d, am.d, am.d, np, n0, top);
|
|
|
|
bn_scatter5(tmp.d, top, powerbuf, 2);
|
|
|
|
|
|
|
|
# if 0
|
|
|
|
for (i = 3; i < 32; i++) {
|
|
|
|
/* Calculate a^i = a^(i-1) * a */
|
2016-01-25 22:41:01 +00:00
|
|
|
bn_mul_mont_gather5(tmp.d, am.d, powerbuf, np, n0, top, i - 1);
|
2015-01-22 03:40:55 +00:00
|
|
|
bn_scatter5(tmp.d, top, powerbuf, i);
|
|
|
|
}
|
|
|
|
# else
|
|
|
|
/* same as above, but uses squaring for 1/2 of operations */
|
|
|
|
for (i = 4; i < 32; i *= 2) {
|
|
|
|
bn_mul_mont(tmp.d, tmp.d, tmp.d, np, n0, top);
|
|
|
|
bn_scatter5(tmp.d, top, powerbuf, i);
|
|
|
|
}
|
|
|
|
for (i = 3; i < 8; i += 2) {
|
|
|
|
int j;
|
2016-01-25 22:41:01 +00:00
|
|
|
bn_mul_mont_gather5(tmp.d, am.d, powerbuf, np, n0, top, i - 1);
|
2015-01-22 03:40:55 +00:00
|
|
|
bn_scatter5(tmp.d, top, powerbuf, i);
|
|
|
|
for (j = 2 * i; j < 32; j *= 2) {
|
|
|
|
bn_mul_mont(tmp.d, tmp.d, tmp.d, np, n0, top);
|
|
|
|
bn_scatter5(tmp.d, top, powerbuf, j);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
for (; i < 16; i += 2) {
|
2016-01-25 22:41:01 +00:00
|
|
|
bn_mul_mont_gather5(tmp.d, am.d, powerbuf, np, n0, top, i - 1);
|
2015-01-22 03:40:55 +00:00
|
|
|
bn_scatter5(tmp.d, top, powerbuf, i);
|
|
|
|
bn_mul_mont(tmp.d, tmp.d, tmp.d, np, n0, top);
|
|
|
|
bn_scatter5(tmp.d, top, powerbuf, 2 * i);
|
|
|
|
}
|
|
|
|
for (; i < 32; i += 2) {
|
2016-01-25 22:41:01 +00:00
|
|
|
bn_mul_mont_gather5(tmp.d, am.d, powerbuf, np, n0, top, i - 1);
|
2015-01-22 03:40:55 +00:00
|
|
|
bn_scatter5(tmp.d, top, powerbuf, i);
|
|
|
|
}
|
|
|
|
# endif
|
2018-06-13 12:00:04 +00:00
|
|
|
/*
|
|
|
|
* The exponent may not have a whole number of fixed-size windows.
|
|
|
|
* To simplify the main loop, the initial window has between 1 and
|
|
|
|
* full-window-size bits such that what remains is always a whole
|
|
|
|
* number of windows
|
|
|
|
*/
|
|
|
|
window0 = (bits - 1) % 5 + 1;
|
|
|
|
wmask = (1 << window0) - 1;
|
|
|
|
bits -= window0;
|
|
|
|
wvalue = bn_get_bits(p, bits) & wmask;
|
2015-01-22 03:40:55 +00:00
|
|
|
bn_gather5(tmp.d, top, powerbuf, wvalue);
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Scan the exponent one window at a time starting from the most
|
|
|
|
* significant bits.
|
|
|
|
*/
|
2018-06-13 12:00:04 +00:00
|
|
|
if (top & 7) {
|
|
|
|
while (bits > 0) {
|
2015-01-22 03:40:55 +00:00
|
|
|
bn_mul_mont(tmp.d, tmp.d, tmp.d, np, n0, top);
|
|
|
|
bn_mul_mont(tmp.d, tmp.d, tmp.d, np, n0, top);
|
|
|
|
bn_mul_mont(tmp.d, tmp.d, tmp.d, np, n0, top);
|
|
|
|
bn_mul_mont(tmp.d, tmp.d, tmp.d, np, n0, top);
|
|
|
|
bn_mul_mont(tmp.d, tmp.d, tmp.d, np, n0, top);
|
|
|
|
bn_mul_mont_gather5(tmp.d, tmp.d, powerbuf, np, n0, top,
|
2018-06-13 12:00:04 +00:00
|
|
|
bn_get_bits5(p->d, bits -= 5));
|
|
|
|
}
|
2015-01-22 03:40:55 +00:00
|
|
|
} else {
|
2018-06-13 12:00:04 +00:00
|
|
|
while (bits > 0) {
|
|
|
|
bn_power5(tmp.d, tmp.d, powerbuf, np, n0, top,
|
|
|
|
bn_get_bits5(p->d, bits -= 5));
|
2015-01-22 03:40:55 +00:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2016-01-25 22:41:01 +00:00
|
|
|
ret = bn_from_montgomery(tmp.d, tmp.d, NULL, np, n0, top);
|
2015-01-22 03:40:55 +00:00
|
|
|
tmp.top = top;
|
|
|
|
bn_correct_top(&tmp);
|
|
|
|
if (ret) {
|
|
|
|
if (!BN_copy(rr, &tmp))
|
|
|
|
ret = 0;
|
|
|
|
goto err; /* non-zero ret means it's not error */
|
|
|
|
}
|
|
|
|
} else
|
2011-08-12 16:44:32 +00:00
|
|
|
#endif
|
2015-01-22 03:40:55 +00:00
|
|
|
{
|
2016-01-25 19:38:38 +00:00
|
|
|
if (!MOD_EXP_CTIME_COPY_TO_PREBUF(&tmp, top, powerbuf, 0, window))
|
2015-01-22 03:40:55 +00:00
|
|
|
goto err;
|
2016-01-25 19:38:38 +00:00
|
|
|
if (!MOD_EXP_CTIME_COPY_TO_PREBUF(&am, top, powerbuf, 1, window))
|
2015-01-22 03:40:55 +00:00
|
|
|
goto err;
|
|
|
|
|
|
|
|
/*
|
|
|
|
* If the window size is greater than 1, then calculate
|
|
|
|
* val[i=2..2^winsize-1]. Powers are computed as a*a^(i-1) (even
|
|
|
|
* powers could instead be computed as (a^(i/2))^2 to use the slight
|
|
|
|
* performance advantage of sqr over mul).
|
|
|
|
*/
|
|
|
|
if (window > 1) {
|
2018-07-06 13:13:15 +00:00
|
|
|
if (!bn_mul_mont_fixed_top(&tmp, &am, &am, mont, ctx))
|
2015-01-22 03:40:55 +00:00
|
|
|
goto err;
|
2016-01-25 19:38:38 +00:00
|
|
|
if (!MOD_EXP_CTIME_COPY_TO_PREBUF(&tmp, top, powerbuf, 2,
|
|
|
|
window))
|
2015-01-22 03:40:55 +00:00
|
|
|
goto err;
|
|
|
|
for (i = 3; i < numPowers; i++) {
|
|
|
|
/* Calculate a^i = a^(i-1) * a */
|
2018-07-06 13:13:15 +00:00
|
|
|
if (!bn_mul_mont_fixed_top(&tmp, &am, &tmp, mont, ctx))
|
2015-01-22 03:40:55 +00:00
|
|
|
goto err;
|
2016-01-25 19:38:38 +00:00
|
|
|
if (!MOD_EXP_CTIME_COPY_TO_PREBUF(&tmp, top, powerbuf, i,
|
|
|
|
window))
|
2015-01-22 03:40:55 +00:00
|
|
|
goto err;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2018-06-13 12:00:04 +00:00
|
|
|
/*
|
bn/bn_exp.c: mitigation of the One-and-Done side-channel attack.
The One&Done attack, which is described in a paper to appear in the
USENIX Security'18 conference, uses EM emanations to recover the values
of the bits that are obtained using BN_is_bit_set while constructing
the value of the window in BN_mod_exp_consttime. The EM signal changes
slightly depending on the value of the bit, and since the lookup of a
bit is surrounded by highly regular execution (constant-time Montgomery
multiplications) the attack is able to isolate the (very brief) part of
the signal that changes depending on the bit. Although the change is
slight, the attack recovers it successfully >90% of the time on several
phones and IoT devices (all with ARM processors with clock rates around
1GHz), so after only one RSA decryption more than 90% of the bits in
d_p and d_q are recovered correctly, which enables rapid recovery of
the full RSA key using an algorithm (also described in the paper) that
modifies the branch-and-prune approach for a situation in which the
exponents' bits are recovered with errors, i.e. where we do not know
a priori which bits are correctly recovered.
The mitigation for the attack is relatively simple - all the bits of
the window are obtained at once, along with other bits so that an
entire integer's worth of bits are obtained together using masking and
shifts, without unnecessarily considering each bit in isolation. This
improves performance somewhat (one call to bn_get_bits is faster than
several calls to BN_is_bit_set), so the attacker now gets one signal
snippet per window (rather than one per bit) in which the signal is
affected by all bits in the integer (rather than just the one bit).
Reviewed-by: Andy Polyakov <appro@openssl.org>
Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/6276)
2018-05-16 17:59:36 +00:00
|
|
|
* The exponent may not have a whole number of fixed-size windows.
|
|
|
|
* To simplify the main loop, the initial window has between 1 and
|
|
|
|
* full-window-size bits such that what remains is always a whole
|
|
|
|
* number of windows
|
2018-06-13 12:00:04 +00:00
|
|
|
*/
|
bn/bn_exp.c: mitigation of the One-and-Done side-channel attack.
The One&Done attack, which is described in a paper to appear in the
USENIX Security'18 conference, uses EM emanations to recover the values
of the bits that are obtained using BN_is_bit_set while constructing
the value of the window in BN_mod_exp_consttime. The EM signal changes
slightly depending on the value of the bit, and since the lookup of a
bit is surrounded by highly regular execution (constant-time Montgomery
multiplications) the attack is able to isolate the (very brief) part of
the signal that changes depending on the bit. Although the change is
slight, the attack recovers it successfully >90% of the time on several
phones and IoT devices (all with ARM processors with clock rates around
1GHz), so after only one RSA decryption more than 90% of the bits in
d_p and d_q are recovered correctly, which enables rapid recovery of
the full RSA key using an algorithm (also described in the paper) that
modifies the branch-and-prune approach for a situation in which the
exponents' bits are recovered with errors, i.e. where we do not know
a priori which bits are correctly recovered.
The mitigation for the attack is relatively simple - all the bits of
the window are obtained at once, along with other bits so that an
entire integer's worth of bits are obtained together using masking and
shifts, without unnecessarily considering each bit in isolation. This
improves performance somewhat (one call to bn_get_bits is faster than
several calls to BN_is_bit_set), so the attacker now gets one signal
snippet per window (rather than one per bit) in which the signal is
affected by all bits in the integer (rather than just the one bit).
Reviewed-by: Andy Polyakov <appro@openssl.org>
Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/6276)
2018-05-16 17:59:36 +00:00
|
|
|
window0 = (bits - 1) % window + 1;
|
|
|
|
wmask = (1 << window0) - 1;
|
|
|
|
bits -= window0;
|
|
|
|
wvalue = bn_get_bits(p, bits) & wmask;
|
2016-01-25 19:38:38 +00:00
|
|
|
if (!MOD_EXP_CTIME_COPY_FROM_PREBUF(&tmp, top, powerbuf, wvalue,
|
|
|
|
window))
|
2015-01-22 03:40:55 +00:00
|
|
|
goto err;
|
|
|
|
|
bn/bn_exp.c: mitigation of the One-and-Done side-channel attack.
The One&Done attack, which is described in a paper to appear in the
USENIX Security'18 conference, uses EM emanations to recover the values
of the bits that are obtained using BN_is_bit_set while constructing
the value of the window in BN_mod_exp_consttime. The EM signal changes
slightly depending on the value of the bit, and since the lookup of a
bit is surrounded by highly regular execution (constant-time Montgomery
multiplications) the attack is able to isolate the (very brief) part of
the signal that changes depending on the bit. Although the change is
slight, the attack recovers it successfully >90% of the time on several
phones and IoT devices (all with ARM processors with clock rates around
1GHz), so after only one RSA decryption more than 90% of the bits in
d_p and d_q are recovered correctly, which enables rapid recovery of
the full RSA key using an algorithm (also described in the paper) that
modifies the branch-and-prune approach for a situation in which the
exponents' bits are recovered with errors, i.e. where we do not know
a priori which bits are correctly recovered.
The mitigation for the attack is relatively simple - all the bits of
the window are obtained at once, along with other bits so that an
entire integer's worth of bits are obtained together using masking and
shifts, without unnecessarily considering each bit in isolation. This
improves performance somewhat (one call to bn_get_bits is faster than
several calls to BN_is_bit_set), so the attacker now gets one signal
snippet per window (rather than one per bit) in which the signal is
affected by all bits in the integer (rather than just the one bit).
Reviewed-by: Andy Polyakov <appro@openssl.org>
Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/6276)
2018-05-16 17:59:36 +00:00
|
|
|
wmask = (1 << window) - 1;
|
2015-01-22 03:40:55 +00:00
|
|
|
/*
|
|
|
|
* Scan the exponent one window at a time starting from the most
|
|
|
|
* significant bits.
|
|
|
|
*/
|
bn/bn_exp.c: mitigation of the One-and-Done side-channel attack.
The One&Done attack, which is described in a paper to appear in the
USENIX Security'18 conference, uses EM emanations to recover the values
of the bits that are obtained using BN_is_bit_set while constructing
the value of the window in BN_mod_exp_consttime. The EM signal changes
slightly depending on the value of the bit, and since the lookup of a
bit is surrounded by highly regular execution (constant-time Montgomery
multiplications) the attack is able to isolate the (very brief) part of
the signal that changes depending on the bit. Although the change is
slight, the attack recovers it successfully >90% of the time on several
phones and IoT devices (all with ARM processors with clock rates around
1GHz), so after only one RSA decryption more than 90% of the bits in
d_p and d_q are recovered correctly, which enables rapid recovery of
the full RSA key using an algorithm (also described in the paper) that
modifies the branch-and-prune approach for a situation in which the
exponents' bits are recovered with errors, i.e. where we do not know
a priori which bits are correctly recovered.
The mitigation for the attack is relatively simple - all the bits of
the window are obtained at once, along with other bits so that an
entire integer's worth of bits are obtained together using masking and
shifts, without unnecessarily considering each bit in isolation. This
improves performance somewhat (one call to bn_get_bits is faster than
several calls to BN_is_bit_set), so the attacker now gets one signal
snippet per window (rather than one per bit) in which the signal is
affected by all bits in the integer (rather than just the one bit).
Reviewed-by: Andy Polyakov <appro@openssl.org>
Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/6276)
2018-05-16 17:59:36 +00:00
|
|
|
while (bits > 0) {
|
2015-01-22 03:40:55 +00:00
|
|
|
|
bn/bn_exp.c: mitigation of the One-and-Done side-channel attack.
The One&Done attack, which is described in a paper to appear in the
USENIX Security'18 conference, uses EM emanations to recover the values
of the bits that are obtained using BN_is_bit_set while constructing
the value of the window in BN_mod_exp_consttime. The EM signal changes
slightly depending on the value of the bit, and since the lookup of a
bit is surrounded by highly regular execution (constant-time Montgomery
multiplications) the attack is able to isolate the (very brief) part of
the signal that changes depending on the bit. Although the change is
slight, the attack recovers it successfully >90% of the time on several
phones and IoT devices (all with ARM processors with clock rates around
1GHz), so after only one RSA decryption more than 90% of the bits in
d_p and d_q are recovered correctly, which enables rapid recovery of
the full RSA key using an algorithm (also described in the paper) that
modifies the branch-and-prune approach for a situation in which the
exponents' bits are recovered with errors, i.e. where we do not know
a priori which bits are correctly recovered.
The mitigation for the attack is relatively simple - all the bits of
the window are obtained at once, along with other bits so that an
entire integer's worth of bits are obtained together using masking and
shifts, without unnecessarily considering each bit in isolation. This
improves performance somewhat (one call to bn_get_bits is faster than
several calls to BN_is_bit_set), so the attacker now gets one signal
snippet per window (rather than one per bit) in which the signal is
affected by all bits in the integer (rather than just the one bit).
Reviewed-by: Andy Polyakov <appro@openssl.org>
Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/6276)
2018-05-16 17:59:36 +00:00
|
|
|
/* Square the result window-size times */
|
|
|
|
for (i = 0; i < window; i++)
|
2018-07-06 13:13:15 +00:00
|
|
|
if (!bn_mul_mont_fixed_top(&tmp, &tmp, &tmp, mont, ctx))
|
2015-01-22 03:40:55 +00:00
|
|
|
goto err;
|
|
|
|
|
2018-06-13 12:00:04 +00:00
|
|
|
/*
|
bn/bn_exp.c: mitigation of the One-and-Done side-channel attack.
The One&Done attack, which is described in a paper to appear in the
USENIX Security'18 conference, uses EM emanations to recover the values
of the bits that are obtained using BN_is_bit_set while constructing
the value of the window in BN_mod_exp_consttime. The EM signal changes
slightly depending on the value of the bit, and since the lookup of a
bit is surrounded by highly regular execution (constant-time Montgomery
multiplications) the attack is able to isolate the (very brief) part of
the signal that changes depending on the bit. Although the change is
slight, the attack recovers it successfully >90% of the time on several
phones and IoT devices (all with ARM processors with clock rates around
1GHz), so after only one RSA decryption more than 90% of the bits in
d_p and d_q are recovered correctly, which enables rapid recovery of
the full RSA key using an algorithm (also described in the paper) that
modifies the branch-and-prune approach for a situation in which the
exponents' bits are recovered with errors, i.e. where we do not know
a priori which bits are correctly recovered.
The mitigation for the attack is relatively simple - all the bits of
the window are obtained at once, along with other bits so that an
entire integer's worth of bits are obtained together using masking and
shifts, without unnecessarily considering each bit in isolation. This
improves performance somewhat (one call to bn_get_bits is faster than
several calls to BN_is_bit_set), so the attacker now gets one signal
snippet per window (rather than one per bit) in which the signal is
affected by all bits in the integer (rather than just the one bit).
Reviewed-by: Andy Polyakov <appro@openssl.org>
Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/6276)
2018-05-16 17:59:36 +00:00
|
|
|
* Get a window's worth of bits from the exponent
|
|
|
|
* This avoids calling BN_is_bit_set for each bit, which
|
|
|
|
* is not only slower but also makes each bit vulnerable to
|
|
|
|
* EM (and likely other) side-channel attacks like One&Done
|
|
|
|
* (for details see "One&Done: A Single-Decryption EM-Based
|
2018-10-19 14:05:49 +00:00
|
|
|
* Attack on OpenSSL's Constant-Time Blinded RSA" by M. Alam,
|
bn/bn_exp.c: mitigation of the One-and-Done side-channel attack.
The One&Done attack, which is described in a paper to appear in the
USENIX Security'18 conference, uses EM emanations to recover the values
of the bits that are obtained using BN_is_bit_set while constructing
the value of the window in BN_mod_exp_consttime. The EM signal changes
slightly depending on the value of the bit, and since the lookup of a
bit is surrounded by highly regular execution (constant-time Montgomery
multiplications) the attack is able to isolate the (very brief) part of
the signal that changes depending on the bit. Although the change is
slight, the attack recovers it successfully >90% of the time on several
phones and IoT devices (all with ARM processors with clock rates around
1GHz), so after only one RSA decryption more than 90% of the bits in
d_p and d_q are recovered correctly, which enables rapid recovery of
the full RSA key using an algorithm (also described in the paper) that
modifies the branch-and-prune approach for a situation in which the
exponents' bits are recovered with errors, i.e. where we do not know
a priori which bits are correctly recovered.
The mitigation for the attack is relatively simple - all the bits of
the window are obtained at once, along with other bits so that an
entire integer's worth of bits are obtained together using masking and
shifts, without unnecessarily considering each bit in isolation. This
improves performance somewhat (one call to bn_get_bits is faster than
several calls to BN_is_bit_set), so the attacker now gets one signal
snippet per window (rather than one per bit) in which the signal is
affected by all bits in the integer (rather than just the one bit).
Reviewed-by: Andy Polyakov <appro@openssl.org>
Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/6276)
2018-05-16 17:59:36 +00:00
|
|
|
* H. Khan, M. Dey, N. Sinha, R. Callan, A. Zajic, and
|
|
|
|
* M. Prvulovic, in USENIX Security'18)
|
|
|
|
*/
|
|
|
|
bits -= window;
|
|
|
|
wvalue = bn_get_bits(p, bits) & wmask;
|
2015-01-22 03:40:55 +00:00
|
|
|
/*
|
|
|
|
* Fetch the appropriate pre-computed value from the pre-buf
|
|
|
|
*/
|
2016-01-25 19:38:38 +00:00
|
|
|
if (!MOD_EXP_CTIME_COPY_FROM_PREBUF(&am, top, powerbuf, wvalue,
|
|
|
|
window))
|
2015-01-22 03:40:55 +00:00
|
|
|
goto err;
|
|
|
|
|
|
|
|
/* Multiply the result into the intermediate result */
|
2018-07-06 13:13:15 +00:00
|
|
|
if (!bn_mul_mont_fixed_top(&tmp, &tmp, &am, mont, ctx))
|
2015-01-22 03:40:55 +00:00
|
|
|
goto err;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2018-07-06 13:13:15 +00:00
|
|
|
/*
|
|
|
|
* Done with zero-padded intermediate BIGNUMs. Final BN_from_montgomery
|
|
|
|
* removes padding [if any] and makes return value suitable for public
|
|
|
|
* API consumer.
|
|
|
|
*/
|
2013-07-05 18:52:58 +00:00
|
|
|
#if defined(SPARC_T4_MONT)
|
2015-01-22 03:40:55 +00:00
|
|
|
if (OPENSSL_sparcv9cap_P[0] & (SPARCV9_VIS3 | SPARCV9_PREFER_FPU)) {
|
|
|
|
am.d[0] = 1; /* borrow am */
|
|
|
|
for (i = 1; i < top; i++)
|
|
|
|
am.d[i] = 0;
|
|
|
|
if (!BN_mod_mul_montgomery(rr, &tmp, &am, mont, ctx))
|
|
|
|
goto err;
|
|
|
|
} else
|
2013-06-18 08:39:38 +00:00
|
|
|
#endif
|
2015-01-22 03:40:55 +00:00
|
|
|
if (!BN_from_montgomery(rr, &tmp, mont, ctx))
|
|
|
|
goto err;
|
|
|
|
ret = 1;
|
|
|
|
err:
|
2015-05-01 01:37:06 +00:00
|
|
|
if (in_mont == NULL)
|
2015-01-22 03:40:55 +00:00
|
|
|
BN_MONT_CTX_free(mont);
|
|
|
|
if (powerbuf != NULL) {
|
|
|
|
OPENSSL_cleanse(powerbuf, powerbufLen);
|
2015-05-01 14:02:07 +00:00
|
|
|
OPENSSL_free(powerbufFree);
|
2015-01-22 03:40:55 +00:00
|
|
|
}
|
|
|
|
BN_CTX_end(ctx);
|
2017-10-17 14:04:09 +00:00
|
|
|
return ret;
|
2015-01-22 03:40:55 +00:00
|
|
|
}
|
2005-05-16 01:43:31 +00:00
|
|
|
|
2000-06-07 21:29:25 +00:00
|
|
|
int BN_mod_exp_mont_word(BIGNUM *rr, BN_ULONG a, const BIGNUM *p,
|
|
|
|
const BIGNUM *m, BN_CTX *ctx, BN_MONT_CTX *in_mont)
|
2015-01-22 03:40:55 +00:00
|
|
|
{
|
|
|
|
BN_MONT_CTX *mont = NULL;
|
|
|
|
int b, bits, ret = 0;
|
|
|
|
int r_is_one;
|
|
|
|
BN_ULONG w, next_w;
|
2017-06-22 10:52:29 +00:00
|
|
|
BIGNUM *r, *t;
|
2015-01-22 03:40:55 +00:00
|
|
|
BIGNUM *swap_tmp;
|
2000-06-08 09:39:28 +00:00
|
|
|
#define BN_MOD_MUL_WORD(r, w, m) \
|
2015-01-22 03:40:55 +00:00
|
|
|
(BN_mul_word(r, (w)) && \
|
|
|
|
(/* BN_ucmp(r, (m)) < 0 ? 1 :*/ \
|
|
|
|
(BN_mod(t, r, m, ctx) && (swap_tmp = r, r = t, t = swap_tmp, 1))))
|
|
|
|
/*
|
|
|
|
* BN_MOD_MUL_WORD is only used with 'w' large, so the BN_ucmp test is
|
|
|
|
* probably more overhead than always using BN_mod (which uses BN_copy if
|
|
|
|
* a similar test returns true).
|
|
|
|
*/
|
|
|
|
/*
|
|
|
|
* We can use BN_mod and do not need BN_nnmod because our accumulator is
|
|
|
|
* never negative (the result of BN_mod does not depend on the sign of
|
|
|
|
* the modulus).
|
|
|
|
*/
|
2000-06-23 05:51:37 +00:00
|
|
|
#define BN_TO_MONTGOMERY_WORD(r, w, mont) \
|
2015-01-22 03:40:55 +00:00
|
|
|
(BN_set_word(r, (w)) && BN_to_montgomery(r, r, (mont), ctx))
|
|
|
|
|
2017-09-27 10:13:47 +00:00
|
|
|
if (BN_get_flags(p, BN_FLG_CONSTTIME) != 0
|
|
|
|
|| BN_get_flags(m, BN_FLG_CONSTTIME) != 0) {
|
2015-01-22 03:40:55 +00:00
|
|
|
/* BN_FLG_CONSTTIME only supported by BN_mod_exp_mont() */
|
|
|
|
BNerr(BN_F_BN_MOD_EXP_MONT_WORD, ERR_R_SHOULD_NOT_HAVE_BEEN_CALLED);
|
2016-08-14 19:52:13 +00:00
|
|
|
return 0;
|
2015-01-22 03:40:55 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
bn_check_top(p);
|
|
|
|
bn_check_top(m);
|
|
|
|
|
|
|
|
if (!BN_is_odd(m)) {
|
|
|
|
BNerr(BN_F_BN_MOD_EXP_MONT_WORD, BN_R_CALLED_WITH_EVEN_MODULUS);
|
2017-10-17 14:04:09 +00:00
|
|
|
return 0;
|
2015-01-22 03:40:55 +00:00
|
|
|
}
|
|
|
|
if (m->top == 1)
|
|
|
|
a %= m->d[0]; /* make sure that 'a' is reduced */
|
|
|
|
|
|
|
|
bits = BN_num_bits(p);
|
|
|
|
if (bits == 0) {
|
2018-05-24 15:12:52 +00:00
|
|
|
/* x**0 mod 1, or x**0 mod -1 is still zero. */
|
|
|
|
if (BN_abs_is_word(m, 1)) {
|
2015-01-22 03:40:55 +00:00
|
|
|
ret = 1;
|
|
|
|
BN_zero(rr);
|
2015-12-14 15:38:15 +00:00
|
|
|
} else {
|
2015-01-22 03:40:55 +00:00
|
|
|
ret = BN_one(rr);
|
2015-12-14 15:38:15 +00:00
|
|
|
}
|
2015-01-22 03:40:55 +00:00
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
if (a == 0) {
|
|
|
|
BN_zero(rr);
|
|
|
|
ret = 1;
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
|
|
|
BN_CTX_start(ctx);
|
|
|
|
r = BN_CTX_get(ctx);
|
|
|
|
t = BN_CTX_get(ctx);
|
2017-06-22 10:52:29 +00:00
|
|
|
if (t == NULL)
|
2015-01-22 03:40:55 +00:00
|
|
|
goto err;
|
|
|
|
|
|
|
|
if (in_mont != NULL)
|
|
|
|
mont = in_mont;
|
|
|
|
else {
|
|
|
|
if ((mont = BN_MONT_CTX_new()) == NULL)
|
|
|
|
goto err;
|
|
|
|
if (!BN_MONT_CTX_set(mont, m, ctx))
|
|
|
|
goto err;
|
|
|
|
}
|
|
|
|
|
|
|
|
r_is_one = 1; /* except for Montgomery factor */
|
|
|
|
|
|
|
|
/* bits-1 >= 0 */
|
|
|
|
|
|
|
|
/* The result is accumulated in the product r*w. */
|
|
|
|
w = a; /* bit 'bits-1' of 'p' is always set */
|
|
|
|
for (b = bits - 2; b >= 0; b--) {
|
|
|
|
/* First, square r*w. */
|
|
|
|
next_w = w * w;
|
|
|
|
if ((next_w / w) != w) { /* overflow */
|
|
|
|
if (r_is_one) {
|
|
|
|
if (!BN_TO_MONTGOMERY_WORD(r, w, mont))
|
|
|
|
goto err;
|
|
|
|
r_is_one = 0;
|
|
|
|
} else {
|
|
|
|
if (!BN_MOD_MUL_WORD(r, w, m))
|
|
|
|
goto err;
|
|
|
|
}
|
|
|
|
next_w = 1;
|
|
|
|
}
|
|
|
|
w = next_w;
|
|
|
|
if (!r_is_one) {
|
|
|
|
if (!BN_mod_mul_montgomery(r, r, r, mont, ctx))
|
|
|
|
goto err;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Second, multiply r*w by 'a' if exponent bit is set. */
|
|
|
|
if (BN_is_bit_set(p, b)) {
|
|
|
|
next_w = w * a;
|
|
|
|
if ((next_w / a) != w) { /* overflow */
|
|
|
|
if (r_is_one) {
|
|
|
|
if (!BN_TO_MONTGOMERY_WORD(r, w, mont))
|
|
|
|
goto err;
|
|
|
|
r_is_one = 0;
|
|
|
|
} else {
|
|
|
|
if (!BN_MOD_MUL_WORD(r, w, m))
|
|
|
|
goto err;
|
|
|
|
}
|
|
|
|
next_w = a;
|
|
|
|
}
|
|
|
|
w = next_w;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Finally, set r:=r*w. */
|
|
|
|
if (w != 1) {
|
|
|
|
if (r_is_one) {
|
|
|
|
if (!BN_TO_MONTGOMERY_WORD(r, w, mont))
|
|
|
|
goto err;
|
|
|
|
r_is_one = 0;
|
|
|
|
} else {
|
|
|
|
if (!BN_MOD_MUL_WORD(r, w, m))
|
|
|
|
goto err;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
if (r_is_one) { /* can happen only if a == 1 */
|
|
|
|
if (!BN_one(rr))
|
|
|
|
goto err;
|
|
|
|
} else {
|
|
|
|
if (!BN_from_montgomery(rr, r, mont, ctx))
|
|
|
|
goto err;
|
|
|
|
}
|
|
|
|
ret = 1;
|
|
|
|
err:
|
2015-05-01 01:37:06 +00:00
|
|
|
if (in_mont == NULL)
|
2015-01-22 03:40:55 +00:00
|
|
|
BN_MONT_CTX_free(mont);
|
|
|
|
BN_CTX_end(ctx);
|
|
|
|
bn_check_top(rr);
|
2017-10-17 14:04:09 +00:00
|
|
|
return ret;
|
2015-01-22 03:40:55 +00:00
|
|
|
}
|
1998-12-21 10:52:47 +00:00
|
|
|
|
|
|
|
/* The old fallback, simple version :-) */
|
2003-12-02 03:28:24 +00:00
|
|
|
int BN_mod_exp_simple(BIGNUM *r, const BIGNUM *a, const BIGNUM *p,
|
2015-01-22 03:40:55 +00:00
|
|
|
const BIGNUM *m, BN_CTX *ctx)
|
|
|
|
{
|
|
|
|
int i, j, bits, ret = 0, wstart, wend, window, wvalue;
|
|
|
|
int start = 1;
|
|
|
|
BIGNUM *d;
|
|
|
|
/* Table of variables obtained from 'ctx' */
|
|
|
|
BIGNUM *val[TABLE_SIZE];
|
|
|
|
|
2017-09-27 10:13:47 +00:00
|
|
|
if (BN_get_flags(p, BN_FLG_CONSTTIME) != 0
|
|
|
|
|| BN_get_flags(a, BN_FLG_CONSTTIME) != 0
|
|
|
|
|| BN_get_flags(m, BN_FLG_CONSTTIME) != 0) {
|
2015-01-22 03:40:55 +00:00
|
|
|
/* BN_FLG_CONSTTIME only supported by BN_mod_exp_mont() */
|
|
|
|
BNerr(BN_F_BN_MOD_EXP_SIMPLE, ERR_R_SHOULD_NOT_HAVE_BEEN_CALLED);
|
2016-08-14 19:52:13 +00:00
|
|
|
return 0;
|
2015-01-22 03:40:55 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
bits = BN_num_bits(p);
|
2017-11-05 19:35:47 +00:00
|
|
|
if (bits == 0) {
|
2018-05-24 15:12:52 +00:00
|
|
|
/* x**0 mod 1, or x**0 mod -1 is still zero. */
|
|
|
|
if (BN_abs_is_word(m, 1)) {
|
2015-12-14 15:38:15 +00:00
|
|
|
ret = 1;
|
|
|
|
BN_zero(r);
|
|
|
|
} else {
|
|
|
|
ret = BN_one(r);
|
|
|
|
}
|
2015-01-22 03:40:55 +00:00
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
|
|
|
BN_CTX_start(ctx);
|
|
|
|
d = BN_CTX_get(ctx);
|
|
|
|
val[0] = BN_CTX_get(ctx);
|
2017-06-22 10:52:29 +00:00
|
|
|
if (val[0] == NULL)
|
2015-01-22 03:40:55 +00:00
|
|
|
goto err;
|
|
|
|
|
|
|
|
if (!BN_nnmod(val[0], a, m, ctx))
|
|
|
|
goto err; /* 1 */
|
|
|
|
if (BN_is_zero(val[0])) {
|
|
|
|
BN_zero(r);
|
|
|
|
ret = 1;
|
|
|
|
goto err;
|
|
|
|
}
|
|
|
|
|
|
|
|
window = BN_window_bits_for_exponent_size(bits);
|
|
|
|
if (window > 1) {
|
|
|
|
if (!BN_mod_mul(d, val[0], val[0], m, ctx))
|
|
|
|
goto err; /* 2 */
|
|
|
|
j = 1 << (window - 1);
|
|
|
|
for (i = 1; i < j; i++) {
|
|
|
|
if (((val[i] = BN_CTX_get(ctx)) == NULL) ||
|
|
|
|
!BN_mod_mul(val[i], val[i - 1], d, m, ctx))
|
|
|
|
goto err;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
start = 1; /* This is used to avoid multiplication etc
|
|
|
|
* when there is only the value '1' in the
|
|
|
|
* buffer. */
|
|
|
|
wvalue = 0; /* The 'value' of the window */
|
|
|
|
wstart = bits - 1; /* The top bit of the window */
|
|
|
|
wend = 0; /* The bottom bit of the window */
|
|
|
|
|
|
|
|
if (!BN_one(r))
|
|
|
|
goto err;
|
|
|
|
|
|
|
|
for (;;) {
|
|
|
|
if (BN_is_bit_set(p, wstart) == 0) {
|
|
|
|
if (!start)
|
|
|
|
if (!BN_mod_mul(r, r, r, m, ctx))
|
|
|
|
goto err;
|
|
|
|
if (wstart == 0)
|
|
|
|
break;
|
|
|
|
wstart--;
|
|
|
|
continue;
|
|
|
|
}
|
|
|
|
/*
|
|
|
|
* We now have wstart on a 'set' bit, we now need to work out how bit
|
|
|
|
* a window to do. To do this we need to scan forward until the last
|
|
|
|
* set bit before the end of the window
|
|
|
|
*/
|
|
|
|
j = wstart;
|
|
|
|
wvalue = 1;
|
|
|
|
wend = 0;
|
|
|
|
for (i = 1; i < window; i++) {
|
|
|
|
if (wstart - i < 0)
|
|
|
|
break;
|
|
|
|
if (BN_is_bit_set(p, wstart - i)) {
|
|
|
|
wvalue <<= (i - wend);
|
|
|
|
wvalue |= 1;
|
|
|
|
wend = i;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
/* wend is the size of the current window */
|
|
|
|
j = wend + 1;
|
|
|
|
/* add the 'bytes above' */
|
|
|
|
if (!start)
|
|
|
|
for (i = 0; i < j; i++) {
|
|
|
|
if (!BN_mod_mul(r, r, r, m, ctx))
|
|
|
|
goto err;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* wvalue will be an odd number < 2^window */
|
|
|
|
if (!BN_mod_mul(r, r, val[wvalue >> 1], m, ctx))
|
|
|
|
goto err;
|
|
|
|
|
|
|
|
/* move the 'window' down further */
|
|
|
|
wstart -= wend + 1;
|
|
|
|
wvalue = 0;
|
|
|
|
start = 0;
|
|
|
|
if (wstart < 0)
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
ret = 1;
|
|
|
|
err:
|
|
|
|
BN_CTX_end(ctx);
|
|
|
|
bn_check_top(r);
|
2017-10-17 14:04:09 +00:00
|
|
|
return ret;
|
2015-01-22 03:40:55 +00:00
|
|
|
}
|