X-Git-Url: https://git.openssl.org/?p=openssl.git;a=blobdiff_plain;f=crypto%2Fbn%2Fbn_exp.c;h=8454d42f84609d64025687697f23cf2e149f61a7;hp=c056a5083f87a5465e6ed90b553cabbd6de16ddc;hb=093388711298d91223f20bd75d44f614335b3e75;hpb=78414a6a897db42c9bcf06aa21c705811ab33921 diff --git a/crypto/bn/bn_exp.c b/crypto/bn/bn_exp.c index c056a5083f..8454d42f84 100644 --- a/crypto/bn/bn_exp.c +++ b/crypto/bn/bn_exp.c @@ -55,112 +55,168 @@ * copied and put under another distribution licence * [including the GNU Public Licence.] */ +/* ==================================================================== + * Copyright (c) 1998-2005 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). + * + */ + +#define OPENSSL_FIPSAPI -#include #include "cryptlib.h" #include "bn_lcl.h" -/* slow but works */ -int BN_mod_mul(ret, a, b, m, ctx) -BIGNUM *ret; -BIGNUM *a; -BIGNUM *b; -BIGNUM *m; -BN_CTX *ctx; - { - BIGNUM *t; - int r=0; +#include +#ifdef _WIN32 +# include +# ifndef alloca +# define alloca _alloca +# endif +#elif defined(__GNUC__) +# ifndef alloca +# define alloca(s) __builtin_alloca((s)) +# endif +#endif - t=ctx->bn[ctx->tos++]; - if (a == b) - { if (!BN_sqr(t,a,ctx)) goto err; } - else - { if (!BN_mul(t,a,b)) goto err; } - if (!BN_mod(ret,t,m,ctx)) goto err; - r=1; -err: - ctx->tos--; - return(r); - } +/* maximum precomputation table size for *variable* sliding windows */ +#define TABLE_SIZE 32 -#if 0 /* this one works - simple but works */ -int BN_mod_exp(r,a,p,m,ctx) -BIGNUM *r,*a,*p,*m; -BN_CTX *ctx; +int BN_exp(BIGNUM *r, const BIGNUM *a, const BIGNUM *p, BN_CTX *ctx) { int i,bits,ret=0; - BIGNUM *v,*tmp; - - v=ctx->bn[ctx->tos++]; - tmp=ctx->bn[ctx->tos++]; - - if (BN_copy(v,a) == NULL) goto err; - bits=BN_num_bits(p); - - if (BN_is_odd(p)) - { if (BN_copy(r,a) == NULL) goto err; } - else { if (BN_one(r)) goto err; } + BIGNUM *v,*rr; - for (i=1; itos-=2; - return(ret); - } - -#endif - -/* this one works - simple but works */ -int BN_exp(r,a,p,ctx) -BIGNUM *r,*a,*p; -BN_CTX *ctx; - { - int i,bits,ret=0; - BIGNUM *v,*tmp; - v=ctx->bn[ctx->tos++]; - tmp=ctx->bn[ctx->tos++]; + BN_CTX_start(ctx); + if ((r == a) || (r == p)) + rr = BN_CTX_get(ctx); + else + rr = r; + 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(r,a) == NULL) goto err; } - else { if (BN_one(r)) goto err; } + { if (BN_copy(rr,a) == NULL) goto err; } + else { if (!BN_one(rr)) goto err; } for (i=1; itos-=2; + if (r != rr) BN_copy(r,rr); + BN_CTX_end(ctx); + bn_check_top(r); return(ret); } -int BN_mod_exp(r,a,p,m,ctx) -BIGNUM *r; -BIGNUM *a; -BIGNUM *p; -BIGNUM *m; -BN_CTX *ctx; + +int BN_mod_exp(BIGNUM *r, const BIGNUM *a, const BIGNUM *p, const BIGNUM *m, + BN_CTX *ctx) { int ret; + bn_check_top(a); + bn_check_top(p); + bn_check_top(m); + + /* For even modulus m = 2^k*m_odd, it might make sense to compute + * 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.) + */ + +#define MONT_MUL_MOD +#define MONT_EXP_WORD +#define RECP_MUL_MOD + #ifdef MONT_MUL_MOD /* I have finally been able to take out this pre-condition of * the top bit being set. It was caused by an error in BN_div @@ -169,7 +225,17 @@ BN_CTX *ctx; /* if ((m->d[m->top-1]&BN_TBIT) && BN_is_odd(m)) */ if (BN_is_odd(m)) - { ret=BN_mod_exp_mont(r,a,p,m,ctx,NULL); } + { +# ifdef MONT_EXP_WORD + if (a->top == 1 && !a->neg && (BN_get_flags(p, BN_FLG_CONSTTIME) == 0)) + { + 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 #endif #ifdef RECP_MUL_MOD @@ -178,58 +244,77 @@ BN_CTX *ctx; { ret=BN_mod_exp_simple(r,a,p,m,ctx); } #endif + bn_check_top(r); return(ret); } -/* #ifdef RECP_MUL_MOD */ -int BN_mod_exp_recp(r,a,p,m,ctx) -BIGNUM *r; -BIGNUM *a; -BIGNUM *p; -BIGNUM *m; -BN_CTX *ctx; + +int BN_mod_exp_recp(BIGNUM *r, const BIGNUM *a, const BIGNUM *p, + const BIGNUM *m, BN_CTX *ctx) { - int nb,i,j,bits,ret=0,wstart,wend,window,wvalue; + int i,j,bits,ret=0,wstart,wend,window,wvalue; int start=1; - BIGNUM *d,*aa; - BIGNUM *val[16]; + BIGNUM *aa; + /* Table of variables obtained from 'ctx' */ + BIGNUM *val[TABLE_SIZE]; + BN_RECP_CTX recp; + + if (BN_get_flags(p, BN_FLG_CONSTTIME) != 0) + { + /* BN_FLG_CONSTTIME only supported by BN_mod_exp_mont() */ + BNerr(BN_F_BN_MOD_EXP_RECP,ERR_R_SHOULD_NOT_HAVE_BEEN_CALLED); + return -1; + } - d=ctx->bn[ctx->tos++]; - aa=ctx->bn[ctx->tos++]; bits=BN_num_bits(p); if (bits == 0) { - BN_one(r); - return(1); + ret = BN_one(r); + return ret; } - nb=BN_reciprocal(d,m,ctx); - if (nb == -1) goto err; - val[0]=BN_new(); - if (!BN_mod(val[0],a,m,ctx)) goto err; /* 1 */ - if (!BN_mod_mul_reciprocal(aa,val[0],val[0],m,d,nb,ctx)) - goto err; /* 2 */ + BN_CTX_start(ctx); + aa = BN_CTX_get(ctx); + val[0] = BN_CTX_get(ctx); + if(!aa || !val[0]) goto err; - if (bits <= 17) /* This is probably 3 or 0x10001, so just do singles */ - window=1; - else if (bits >= 256) - window=5; /* max size of window */ - else if (bits >= 128) - window=4; + 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 - window=3; + { + if (BN_RECP_CTX_set(&recp,m,ctx) <= 0) goto err; + } - j=1<<(window-1); - for (i=1; i 1) + { + if (!BN_mod_mul_reciprocal(aa,val[0],val[0],&recp,ctx)) + goto err; /* 2 */ + j=1<<(window-1); + for (i=1; i>1],m,d,nb,ctx)) + if (!BN_mod_mul_reciprocal(r,r,val[wvalue>>1],&recp,ctx)) goto err; /* move the 'window' down further */ @@ -290,84 +375,91 @@ BN_CTX *ctx; } ret=1; err: - ctx->tos-=2; - for (i=0; i<16; i++) - if (val[i] != NULL) BN_clear_free(val[i]); + BN_CTX_end(ctx); + BN_RECP_CTX_free(&recp); + bn_check_top(r); return(ret); } -/* #endif */ - -/* #ifdef MONT_MUL_MOD */ -int BN_mod_exp_mont(r,a,p,m,ctx,in_mont) -BIGNUM *r; -BIGNUM *a; -BIGNUM *p; -BIGNUM *m; -BN_CTX *ctx; -BN_MONT_CTX *in_mont; + + +int BN_mod_exp_mont(BIGNUM *rr, const BIGNUM *a, const BIGNUM *p, + const BIGNUM *m, BN_CTX *ctx, BN_MONT_CTX *in_mont) { -#define TABLE_SIZE 16 int i,j,bits,ret=0,wstart,wend,window,wvalue; int start=1; - BIGNUM *d,*aa; + BIGNUM *d,*r; + const BIGNUM *aa; + /* Table of variables obtained from 'ctx' */ BIGNUM *val[TABLE_SIZE]; BN_MONT_CTX *mont=NULL; - if (!(m->d[0] & 1)) + if (BN_get_flags(p, BN_FLG_CONSTTIME) != 0) + { + 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); return(0); } - d=ctx->bn[ctx->tos++]; bits=BN_num_bits(p); if (bits == 0) { - BN_one(r); - return(1); + ret = BN_one(rr); + return ret; } + BN_CTX_start(ctx); + d = BN_CTX_get(ctx); + r = BN_CTX_get(ctx); + val[0] = BN_CTX_get(ctx); + if (!d || !r || !val[0]) goto err; + /* If this is not done, things will break in the montgomery * part */ -#if 1 if (in_mont != NULL) mont=in_mont; else -#endif { if ((mont=BN_MONT_CTX_new()) == NULL) goto err; if (!BN_MONT_CTX_set(mont,m,ctx)) goto err; } - val[0]=BN_new(); - if (BN_ucmp(a,m) >= 0) + if (a->neg || BN_ucmp(a,m) >= 0) { - BN_mod(val[0],a,m,ctx); - aa=val[0]; + if (!BN_nnmod(val[0],a,m,ctx)) + goto err; + aa= val[0]; } else aa=a; + if (BN_is_zero(aa)) + { + BN_zero(rr); + ret = 1; + goto err; + } if (!BN_to_montgomery(val[0],aa,mont,ctx)) goto err; /* 1 */ - if (!BN_mod_mul_montgomery(d,val[0],val[0],mont,ctx)) goto err; /* 2 */ - - if (bits <= 20) /* This is probably 3 or 0x10001, so just do singles */ - window=1; - else if (bits > 250) - window=5; /* max size of window */ - else if (bits >= 120) - window=4; - else - window=3; - j=1<<(window-1); - for (i=1; i 1) { - val[i]=BN_new(); - if (!BN_mod_mul_montgomery(val[i],val[i-1],d,mont,ctx)) - goto err; + if (!BN_mod_mul_montgomery(d,val[0],val[0],mont,ctx)) goto err; /* 2 */ + j=1<<(window-1); + for (i=1; i b->top) + top = b->top; /* this works because 'buf' is explicitly zeroed */ + for (i = 0, j=idx; i < top * sizeof b->d[0]; i++, j+=width) + { + buf[j] = ((unsigned char*)b->d)[i]; + } + + return 1; + } + +static int MOD_EXP_CTIME_COPY_FROM_PREBUF(BIGNUM *b, int top, unsigned char *buf, int idx, int width) + { + size_t i, j; + + if (bn_wexpand(b, top) == NULL) + return 0; + + for (i=0, j=idx; i < top * sizeof b->d[0]; i++, j+=width) + { + ((unsigned char*)b->d)[i] = buf[j]; + } + + b->top = top; + bn_correct_top(b); + return 1; + } + +/* Given a pointer value, compute the next address that is a cache line multiple. */ +#define MOD_EXP_CTIME_ALIGN(x_) \ + ((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, + * http://www.daemonology.net/hyperthreading-considered-harmful/) + */ +int BN_mod_exp_mont_consttime(BIGNUM *rr, const BIGNUM *a, const BIGNUM *p, + const BIGNUM *m, BN_CTX *ctx, BN_MONT_CTX *in_mont) + { + int i,bits,ret=0,window,wvalue; + int top; + BN_MONT_CTX *mont=NULL; + + int numPowers; + unsigned char *powerbufFree=NULL; + int powerbufLen = 0; + unsigned char *powerbuf=NULL; + BIGNUM tmp, am; + + bn_check_top(a); + bn_check_top(p); + bn_check_top(m); + + top = m->top; + + if (!(m->d[0] & 1)) + { + BNerr(BN_F_BN_MOD_EXP_MONT_CONSTTIME,BN_R_CALLED_WITH_EVEN_MODULUS); + return(0); + } + bits=BN_num_bits(p); + if (bits == 0) + { + ret = BN_one(rr); + 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; + } + + /* Get the window size to use with size of p. */ + window = BN_window_bits_for_ctime_exponent_size(bits); +#if defined(OPENSSL_BN_ASM_MONT5) + if (window==6 && bits<=1024) window=5; /* ~5% improvement of 2048-bit RSA sign */ +#endif + + /* 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)); +#ifdef alloca + if (powerbufLen < 3072) + powerbufFree = alloca(powerbufLen+MOD_EXP_CTIME_MIN_CACHE_LINE_WIDTH); + else +#endif + if ((powerbufFree=(unsigned char*)OPENSSL_malloc(powerbufLen+MOD_EXP_CTIME_MIN_CACHE_LINE_WIDTH)) == NULL) + goto err; + + powerbuf = MOD_EXP_CTIME_ALIGN(powerbufFree); + memset(powerbuf, 0, powerbufLen); + +#ifdef alloca + if (powerbufLen < 3072) + powerbufFree = NULL; +#endif + + /* 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 + if (!BN_to_montgomery(&tmp,BN_value_one(),mont,ctx)) goto err; +#else + tmp.d[0] = (0-m->d[0])&BN_MASK2; /* 2^(top*BN_BITS2) - m */ + for (i=1;id[i])&BN_MASK2; + tmp.top = top; +#endif + + /* prepare a^1 in Montgomery domain */ + if (a->neg || BN_ucmp(a,m) >= 0) + { + if (!BN_mod(&am,a,m,ctx)) goto err; + if (!BN_to_montgomery(&am,&am,mont,ctx)) goto err; + } + else if (!BN_to_montgomery(&am,a,mont,ctx)) goto err; + +#if defined(OPENSSL_BN_ASM_MONT5) + /* 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... */ + if (window==5) + { + 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); + + BN_ULONG *np=mont->N.d, *n0=mont->n0; + + /* BN_to_montgomery can contaminate words above .top + * [in BN_DEBUG[_DEBUG] build]... */ + for (i=am.top; i=0; i--,bits--) + wvalue = (wvalue<<1)+BN_is_bit_set(p,bits); + bn_gather5(tmp.d,top,powerbuf,wvalue); + + /* Scan the exponent one window at a time starting from the most + * significant bits. + */ + while (bits >= 0) + { + for (wvalue=0, i=0; i<5; i++,bits--) + wvalue = (wvalue<<1)+BN_is_bit_set(p,bits); + + 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,wvalue); + } + + tmp.top=top; + bn_correct_top(&tmp); + } + else +#endif + { + if (!MOD_EXP_CTIME_COPY_TO_PREBUF(&tmp, top, powerbuf, 0, numPowers)) goto err; + if (!MOD_EXP_CTIME_COPY_TO_PREBUF(&am, top, powerbuf, 1, numPowers)) 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) + { + if (!BN_mod_mul_montgomery(&tmp,&am,&am,mont,ctx)) goto err; + if (!MOD_EXP_CTIME_COPY_TO_PREBUF(&tmp, top, powerbuf, 2, numPowers)) goto err; + for (i=3; i=0; i--,bits--) + wvalue = (wvalue<<1)+BN_is_bit_set(p,bits); + if (!MOD_EXP_CTIME_COPY_FROM_PREBUF(&tmp,top,powerbuf,wvalue,numPowers)) goto err; + + /* Scan the exponent one window at a time starting from the most + * significant bits. + */ + while (bits >= 0) + { + wvalue=0; /* The 'value' of the window */ + + /* Scan the window, squaring the result as we go */ + for (i=0; itos--; - for (i=0; itop == 1) + a %= m->d[0]; /* make sure that 'a' is reduced */ + + bits = BN_num_bits(p); + if (bits == 0) + { + ret = BN_one(rr); + return ret; + } + if (a == 0) + { + BN_zero(rr); + ret = 1; + return ret; + } + + BN_CTX_start(ctx); + d = BN_CTX_get(ctx); + r = BN_CTX_get(ctx); + t = BN_CTX_get(ctx); + if (d == NULL || r == NULL || t == NULL) 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: + if ((in_mont == NULL) && (mont != NULL)) BN_MONT_CTX_free(mont); + BN_CTX_end(ctx); + bn_check_top(rr); + return(ret); + } + /* The old fallback, simple version :-) */ -int BN_mod_exp_simple(r,a,p,m,ctx) -BIGNUM *r; -BIGNUM *a; -BIGNUM *p; -BIGNUM *m; -BN_CTX *ctx; +int BN_mod_exp_simple(BIGNUM *r, const BIGNUM *a, const BIGNUM *p, + const BIGNUM *m, BN_CTX *ctx) { int i,j,bits,ret=0,wstart,wend,window,wvalue; int start=1; BIGNUM *d; - BIGNUM *val[16]; + /* Table of variables obtained from 'ctx' */ + BIGNUM *val[TABLE_SIZE]; + + if (BN_get_flags(p, BN_FLG_CONSTTIME) != 0) + { + /* BN_FLG_CONSTTIME only supported by BN_mod_exp_mont() */ + BNerr(BN_F_BN_MOD_EXP_SIMPLE,ERR_R_SHOULD_NOT_HAVE_BEEN_CALLED); + return -1; + } - d=ctx->bn[ctx->tos++]; bits=BN_num_bits(p); if (bits == 0) { - BN_one(r); - return(1); + ret = BN_one(r); + return ret; } - val[0]=BN_new(); - if (!BN_mod(val[0],a,m,ctx)) goto err; /* 1 */ - if (!BN_mod_mul(d,val[0],val[0],m,ctx)) - goto err; /* 2 */ + BN_CTX_start(ctx); + d = BN_CTX_get(ctx); + val[0] = BN_CTX_get(ctx); + if(!d || !val[0]) goto err; - if (bits <= 17) /* This is probably 3 or 0x10001, so just do singles */ - window=1; - else if (bits >= 256) - window=5; /* max size of window */ - else if (bits >= 128) - window=4; - else - window=3; + if (!BN_nnmod(val[0],a,m,ctx)) goto err; /* 1 */ + if (BN_is_zero(val[0])) + { + BN_zero(r); + ret = 1; + goto err; + } - j=1<<(window-1); - for (i=1; i 1) { - val[i]=BN_new(); - if (!BN_mod_mul(val[i],val[i-1],d,m,ctx)) - goto err; + if (!BN_mod_mul(d,val[0],val[0],m,ctx)) + goto err; /* 2 */ + j=1<<(window-1); + for (i=1; itos--; - for (i=0; i<16; i++) - if (val[i] != NULL) BN_clear_free(val[i]); + BN_CTX_end(ctx); + bn_check_top(r); return(ret); } -