X-Git-Url: https://git.openssl.org/?p=openssl.git;a=blobdiff_plain;f=crypto%2Fbn%2Fbn_mont.c;h=e3955fe47523c769fea8fdddf558af09d9727e3c;hp=e435df61f86130f3cfe4b527b91f893c1c33639f;hb=9f040d6decca7930e978784c917f731e5c45e8f0;hpb=7dfb0b774e6592dcbfe47015168a0ac8b44e2a17 diff --git a/crypto/bn/bn_mont.c b/crypto/bn/bn_mont.c index e435df61f8..e3955fe475 100644 --- a/crypto/bn/bn_mont.c +++ b/crypto/bn/bn_mont.c @@ -5,21 +5,21 @@ * This package is an SSL implementation written * by Eric Young (eay@cryptsoft.com). * The implementation was written so as to conform with Netscapes SSL. - * + * * This library is free for commercial and non-commercial use as long as * the following conditions are aheared to. The following conditions * apply to all code found in this distribution, be it the RC4, RSA, * lhash, DES, etc., code; not just the SSL code. The SSL documentation * included with this distribution is covered by the same copyright terms * except that the holder is Tim Hudson (tjh@cryptsoft.com). - * + * * Copyright remains Eric Young's, and as such any Copyright notices in * the code are not to be removed. * If this package is used in a product, Eric Young should be given attribution * as the author of the parts of the library used. * This can be in the form of a textual message at program startup or * in documentation (online or textual) provided with the package. - * + * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: @@ -34,10 +34,10 @@ * Eric Young (eay@cryptsoft.com)" * The word 'cryptographic' can be left out if the rouines from the library * being used are not cryptographic related :-). - * 4. If you include any Windows specific code (or a derivative thereof) from + * 4. If you include any Windows specific code (or a derivative thereof) from * the apps directory (application code) you must include an acknowledgement: * "This product includes software written by Tim Hudson (tjh@cryptsoft.com)" - * + * * THIS SOFTWARE IS PROVIDED BY ERIC YOUNG ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE @@ -49,258 +49,485 @@ * 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. - * + * * The licence and distribution terms for any publically available version or * derivative of this code cannot be changed. i.e. this code cannot simply be * copied and put under another distribution licence * [including the GNU Public Licence.] */ +/* ==================================================================== + * Copyright (c) 1998-2006 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). + * + */ -#include -#include "cryptlib.h" +/* + * Details about Montgomery multiplication algorithms can be found at + * http://security.ece.orst.edu/publications.html, e.g. + * http://security.ece.orst.edu/koc/papers/j37acmon.pdf and + * sections 3.8 and 4.2 in http://security.ece.orst.edu/koc/papers/r01rsasw.pdf + */ + +#include "internal/cryptlib.h" #include "bn_lcl.h" -int BN_mod_mul_montgomery(r,a,b,mont,ctx) -BIGNUM *r,*a,*b; -BN_MONT_CTX *mont; -BN_CTX *ctx; - { - BIGNUM *tmp; - - tmp=ctx->bn[ctx->tos++]; - - if (a == b) - { - if (!BN_sqr(tmp,a,ctx)) goto err; - } - else - { - if (!BN_mul(tmp,a,b)) goto err; - } - /* reduce from aRR to aR */ - if (!BN_from_montgomery(r,tmp,mont,ctx)) goto err; - ctx->tos--; - return(1); -err: - return(0); - } - -#define MONT_WORD +#define MONT_WORD /* use the faster word-based algorithm */ #ifdef MONT_WORD -int BN_from_montgomery(ret,a,mont,ctx) -BIGNUM *ret; -BIGNUM *a; -BN_MONT_CTX *mont; -BN_CTX *ctx; - { - BIGNUM *n,*t1,*r; - BN_ULONG *ap,*np,*rp,n0,v; - int al,nl,max,i,x,ri; - int retn=0; - - t1=ctx->bn[ctx->tos]; - r=ctx->bn[ctx->tos+1]; - - if (!BN_copy(r,a)) goto err; - n=mont->N; - - ap=a->d; - /* mont->ri is the size of mont->N in bits/words */ - al=ri=mont->ri/BN_BITS2; - - nl=n->top; - if ((al == 0) || (nl == 0)) { r->top=0; return(1); } - - max=(nl+al+1); /* allow for overflow (no?) XXX */ - if (bn_wexpand(r,max) == NULL) goto err; - if (bn_wexpand(ret,max) == NULL) goto err; - - r->neg=a->neg^n->neg; - np=n->d; - rp=r->d; - - /* clear the top words of T */ -#if 1 - for (i=r->top; id[i]=0; -#else - memset(&(r->d[r->top]),0,(max-r->top)*sizeof(BN_ULONG)); +static int BN_from_montgomery_word(BIGNUM *ret, BIGNUM *r, BN_MONT_CTX *mont); #endif - r->top=max; - n0=mont->n0; - - for (i=0; i nl) - { - x2=nl; - x1=0; - } - else - { - x2=i+4; - x1=nl-x2; - } - v=bn_mul_add_words(&(rp[x1]),&(np[x1]),x2,(rp[x1]*n0)&BN_MASK2); -#else - v=bn_mul_add_words(rp,np,nl,(rp[0]*n0)&BN_MASK2); -#endif - - if (((rp[nl]+=v)&BN_MASK2) < v) - { - for (x=(nl+1); (((++rp[x])&BN_MASK2) == 0); x++) - ; - } - rp++; - } - while (r->d[r->top-1] == 0) - r->top--; - - /* mont->ri will be a multiple of the word size */ -#if 0 - BN_rshift(ret,r,mont->ri); -#else - ap=r->d; - rp=ret->d; - x=ri; - al=r->top-x; - for (i=0; itop=al; +int BN_mod_mul_montgomery(BIGNUM *r, const BIGNUM *a, const BIGNUM *b, + BN_MONT_CTX *mont, BN_CTX *ctx) +{ + BIGNUM *tmp; + int ret = 0; +#if defined(OPENSSL_BN_ASM_MONT) && defined(MONT_WORD) + int num = mont->N.top; + + if (num > 1 && a->top == num && b->top == num) { + if (bn_wexpand(r, num) == NULL) + return (0); + if (bn_mul_mont(r->d, a->d, b->d, mont->N.d, mont->n0, num)) { + r->neg = a->neg ^ b->neg; + r->top = num; + bn_correct_top(r); + return (1); + } + } #endif - if (BN_ucmp(ret,mont->N) >= 0) - { - bn_qsub(ret,ret,mont->N); /* XXX */ - } - retn=1; -err: - return(retn); - } + BN_CTX_start(ctx); + tmp = BN_CTX_get(ctx); + if (tmp == NULL) + goto err; + + bn_check_top(tmp); + if (a == b) { + if (!BN_sqr(tmp, a, ctx)) + goto err; + } else { + if (!BN_mul(tmp, a, b, ctx)) + goto err; + } + /* reduce from aRR to aR */ +#ifdef MONT_WORD + if (!BN_from_montgomery_word(r, tmp, mont)) + goto err; #else -int BN_from_montgomery(r,a,mont,ctx) -BIGNUM *r; -BIGNUM *a; -BN_MONT_CTX *mont; -BN_CTX *ctx; - { - BIGNUM *t1,*t2; - - t1=ctx->bn[ctx->tos]; - t2=ctx->bn[ctx->tos+1]; - - if (!BN_copy(t1,a)) goto err; - /* can cheat */ - BN_mask_bits(t1,mont->ri); - - if (!BN_mul(t2,t1,mont->Ni)) goto err; - BN_mask_bits(t2,mont->ri); - - if (!BN_mul(t1,t2,mont->N)) goto err; - if (!BN_add(t2,a,t1)) goto err; - BN_rshift(r,t2,mont->ri); - - if (BN_ucmp(r,mont->N) >= 0) - bn_qsub(r,r,mont->N); - - return(1); -err: - return(0); - } + if (!BN_from_montgomery(r, tmp, mont, ctx)) + goto err; #endif + bn_check_top(r); + ret = 1; + err: + BN_CTX_end(ctx); + return (ret); +} -BN_MONT_CTX *BN_MONT_CTX_new() - { - BN_MONT_CTX *ret; - - if ((ret=(BN_MONT_CTX *)Malloc(sizeof(BN_MONT_CTX))) == NULL) - return(NULL); - ret->ri=0; - ret->RR=BN_new(); - ret->N=BN_new(); - ret->Ni=NULL; - if ((ret->RR == NULL) || (ret->N == NULL)) - { - BN_MONT_CTX_free(ret); - return(NULL); - } - return(ret); - } - -void BN_MONT_CTX_free(mont) -BN_MONT_CTX *mont; - { - if (mont->RR != NULL) BN_free(mont->RR); - if (mont->N != NULL) BN_free(mont->N); - if (mont->Ni != NULL) BN_free(mont->Ni); - Free(mont); - } - -int BN_MONT_CTX_set(mont,mod,ctx) -BN_MONT_CTX *mont; -BIGNUM *mod; -BN_CTX *ctx; - { - BIGNUM *Ri=NULL,*R=NULL; - - if (mont->RR == NULL) mont->RR=BN_new(); - if (mont->N == NULL) mont->N=BN_new(); - - R=mont->RR; /* grab RR as a temp */ - BN_copy(mont->N,mod); /* Set N */ +#ifdef MONT_WORD +static int BN_from_montgomery_word(BIGNUM *ret, BIGNUM *r, BN_MONT_CTX *mont) +{ + BIGNUM *n; + BN_ULONG *ap, *np, *rp, n0, v, carry; + int nl, max, i; + + n = &(mont->N); + nl = n->top; + if (nl == 0) { + ret->top = 0; + return (1); + } + + max = (2 * nl); /* carry is stored separately */ + if (bn_wexpand(r, max) == NULL) + return (0); + + r->neg ^= n->neg; + np = n->d; + rp = r->d; + + /* clear the top words of T */ + i = max - r->top; + if (i) + memset(&rp[r->top], 0, sizeof(*rp) * i); + + r->top = max; + n0 = mont->n0[0]; + + for (carry = 0, i = 0; i < nl; i++, rp++) { + v = bn_mul_add_words(rp, np, nl, (rp[0] * n0) & BN_MASK2); + v = (v + carry + rp[nl]) & BN_MASK2; + carry |= (v != rp[nl]); + carry &= (v <= rp[nl]); + rp[nl] = v; + } + + if (bn_wexpand(ret, nl) == NULL) + return (0); + ret->top = nl; + ret->neg = r->neg; + + rp = ret->d; + ap = &(r->d[nl]); + +# define BRANCH_FREE 1 +# if BRANCH_FREE + { + BN_ULONG *nrp; + size_t m; + + v = bn_sub_words(rp, ap, np, nl) - carry; + /* + * if subtraction result is real, then trick unconditional memcpy + * below to perform in-place "refresh" instead of actual copy. + */ + m = (0 - (size_t)v); + nrp = + (BN_ULONG *)(((PTR_SIZE_INT) rp & ~m) | ((PTR_SIZE_INT) ap & m)); + + for (i = 0, nl -= 4; i < nl; i += 4) { + BN_ULONG t1, t2, t3, t4; + + t1 = nrp[i + 0]; + t2 = nrp[i + 1]; + t3 = nrp[i + 2]; + ap[i + 0] = 0; + t4 = nrp[i + 3]; + ap[i + 1] = 0; + rp[i + 0] = t1; + ap[i + 2] = 0; + rp[i + 1] = t2; + ap[i + 3] = 0; + rp[i + 2] = t3; + rp[i + 3] = t4; + } + for (nl += 4; i < nl; i++) + rp[i] = nrp[i], ap[i] = 0; + } +# else + if (bn_sub_words(rp, ap, np, nl) - carry) + memcpy(rp, ap, nl * sizeof(BN_ULONG)); +# endif + bn_correct_top(r); + bn_correct_top(ret); + bn_check_top(ret); + + return (1); +} +#endif /* MONT_WORD */ +int BN_from_montgomery(BIGNUM *ret, const BIGNUM *a, BN_MONT_CTX *mont, + BN_CTX *ctx) +{ + int retn = 0; #ifdef MONT_WORD + BIGNUM *t; + + BN_CTX_start(ctx); + if ((t = BN_CTX_get(ctx)) && BN_copy(t, a)) + retn = BN_from_montgomery_word(ret, t, mont); + BN_CTX_end(ctx); +#else /* !MONT_WORD */ + BIGNUM *t1, *t2; + + BN_CTX_start(ctx); + t1 = BN_CTX_get(ctx); + t2 = BN_CTX_get(ctx); + if (t1 == NULL || t2 == NULL) + goto err; + + if (!BN_copy(t1, a)) + goto err; + BN_mask_bits(t1, mont->ri); + + if (!BN_mul(t2, t1, &mont->Ni, ctx)) + goto err; + BN_mask_bits(t2, mont->ri); + + if (!BN_mul(t1, t2, &mont->N, ctx)) + goto err; + if (!BN_add(t2, a, t1)) + goto err; + if (!BN_rshift(ret, t2, mont->ri)) + goto err; + + if (BN_ucmp(ret, &(mont->N)) >= 0) { + if (!BN_usub(ret, ret, &(mont->N))) + goto err; + } + retn = 1; + bn_check_top(ret); + err: + BN_CTX_end(ctx); +#endif /* MONT_WORD */ + return (retn); +} + +BN_MONT_CTX *BN_MONT_CTX_new(void) { - BIGNUM tmod; - BN_ULONG buf[2]; - /* int z; */ - - mont->ri=(BN_num_bits(mod)+(BN_BITS2-1))/BN_BITS2*BN_BITS2; - BN_lshift(R,BN_value_one(),BN_BITS2); /* R */ - /* I was bad, this modification of a passed variable was - * breaking the multithreaded stuff :-( - * z=mod->top; - * mod->top=1; */ - - buf[0]=mod->d[0]; - buf[1]=0; - tmod.d=buf; - tmod.top=1; - tmod.max=mod->max; - tmod.neg=mod->neg; - - if ((Ri=BN_mod_inverse(R,&tmod,ctx)) == NULL) goto err; /* Ri */ - BN_lshift(Ri,Ri,BN_BITS2); /* R*Ri */ - bn_qsub(Ri,Ri,BN_value_one()); /* R*Ri - 1 */ - BN_div(Ri,NULL,Ri,&tmod,ctx); - mont->n0=Ri->d[0]; - BN_free(Ri); - /* mod->top=z; */ + BN_MONT_CTX *ret; + + if ((ret = OPENSSL_malloc(sizeof(*ret))) == NULL) + return (NULL); + + BN_MONT_CTX_init(ret); + ret->flags = BN_FLG_MALLOCED; + return (ret); } -#else - mont->ri=BN_num_bits(mod); - BN_lshift(R,BN_value_one(),mont->ri); /* R */ - if ((Ri=BN_mod_inverse(R,mod,ctx)) == NULL) goto err; /* Ri */ - BN_lshift(Ri,Ri,mont->ri); /* R*Ri */ - bn_qsub(Ri,Ri,BN_value_one()); /* R*Ri - 1 */ - BN_div(Ri,NULL,Ri,mod,ctx); - if (mont->Ni != NULL) BN_free(mont->Ni); - mont->Ni=Ri; /* Ni=(R*Ri-1)/N */ + +void BN_MONT_CTX_init(BN_MONT_CTX *ctx) +{ + ctx->ri = 0; + BN_init(&(ctx->RR)); + BN_init(&(ctx->N)); + BN_init(&(ctx->Ni)); + ctx->n0[0] = ctx->n0[1] = 0; + ctx->flags = 0; +} + +void BN_MONT_CTX_free(BN_MONT_CTX *mont) +{ + if (mont == NULL) + return; + + BN_free(&(mont->RR)); + BN_free(&(mont->N)); + BN_free(&(mont->Ni)); + if (mont->flags & BN_FLG_MALLOCED) + OPENSSL_free(mont); +} + +int BN_MONT_CTX_set(BN_MONT_CTX *mont, const BIGNUM *mod, BN_CTX *ctx) +{ + int ret = 0; + BIGNUM *Ri, *R; + + BN_CTX_start(ctx); + if ((Ri = BN_CTX_get(ctx)) == NULL) + goto err; + R = &(mont->RR); /* grab RR as a temp */ + if (!BN_copy(&(mont->N), mod)) + goto err; /* Set N */ + mont->N.neg = 0; + +#ifdef MONT_WORD + { + BIGNUM tmod; + BN_ULONG buf[2]; + + BN_init(&tmod); + tmod.d = buf; + tmod.dmax = 2; + tmod.neg = 0; + + mont->ri = (BN_num_bits(mod) + (BN_BITS2 - 1)) / BN_BITS2 * BN_BITS2; + +# if defined(OPENSSL_BN_ASM_MONT) && (BN_BITS2<=32) + /* + * Only certain BN_BITS2<=32 platforms actually make use of n0[1], + * and we could use the #else case (with a shorter R value) for the + * others. However, currently only the assembler files do know which + * is which. + */ + + BN_zero(R); + if (!(BN_set_bit(R, 2 * BN_BITS2))) + goto err; + + tmod.top = 0; + if ((buf[0] = mod->d[0])) + tmod.top = 1; + if ((buf[1] = mod->top > 1 ? mod->d[1] : 0)) + tmod.top = 2; + + if ((BN_mod_inverse(Ri, R, &tmod, ctx)) == NULL) + goto err; + if (!BN_lshift(Ri, Ri, 2 * BN_BITS2)) + goto err; /* R*Ri */ + if (!BN_is_zero(Ri)) { + if (!BN_sub_word(Ri, 1)) + goto err; + } else { /* if N mod word size == 1 */ + + if (bn_expand(Ri, (int)sizeof(BN_ULONG) * 2) == NULL) + goto err; + /* Ri-- (mod double word size) */ + Ri->neg = 0; + Ri->d[0] = BN_MASK2; + Ri->d[1] = BN_MASK2; + Ri->top = 2; + } + if (!BN_div(Ri, NULL, Ri, &tmod, ctx)) + goto err; + /* + * Ni = (R*Ri-1)/N, keep only couple of least significant words: + */ + mont->n0[0] = (Ri->top > 0) ? Ri->d[0] : 0; + mont->n0[1] = (Ri->top > 1) ? Ri->d[1] : 0; +# else + BN_zero(R); + if (!(BN_set_bit(R, BN_BITS2))) + goto err; /* R */ + + buf[0] = mod->d[0]; /* tmod = N mod word size */ + buf[1] = 0; + tmod.top = buf[0] != 0 ? 1 : 0; + /* Ri = R^-1 mod N */ + if ((BN_mod_inverse(Ri, R, &tmod, ctx)) == NULL) + goto err; + if (!BN_lshift(Ri, Ri, BN_BITS2)) + goto err; /* R*Ri */ + if (!BN_is_zero(Ri)) { + if (!BN_sub_word(Ri, 1)) + goto err; + } else { /* if N mod word size == 1 */ + + if (!BN_set_word(Ri, BN_MASK2)) + goto err; /* Ri-- (mod word size) */ + } + if (!BN_div(Ri, NULL, Ri, &tmod, ctx)) + goto err; + /* + * Ni = (R*Ri-1)/N, keep only least significant word: + */ + mont->n0[0] = (Ri->top > 0) ? Ri->d[0] : 0; + mont->n0[1] = 0; +# endif + } +#else /* !MONT_WORD */ + { /* bignum version */ + mont->ri = BN_num_bits(&mont->N); + BN_zero(R); + if (!BN_set_bit(R, mont->ri)) + goto err; /* R = 2^ri */ + /* Ri = R^-1 mod N */ + if ((BN_mod_inverse(Ri, R, &mont->N, ctx)) == NULL) + goto err; + if (!BN_lshift(Ri, Ri, mont->ri)) + goto err; /* R*Ri */ + if (!BN_sub_word(Ri, 1)) + goto err; + /* + * Ni = (R*Ri-1) / N + */ + if (!BN_div(&(mont->Ni), NULL, Ri, &mont->N, ctx)) + goto err; + } #endif - /* setup RR for conversions */ - BN_lshift(mont->RR,BN_value_one(),mont->ri*2); - BN_mod(mont->RR,mont->RR,mont->N,ctx); + /* setup RR for conversions */ + BN_zero(&(mont->RR)); + if (!BN_set_bit(&(mont->RR), mont->ri * 2)) + goto err; + if (!BN_mod(&(mont->RR), &(mont->RR), &(mont->N), ctx)) + goto err; + + ret = 1; + err: + BN_CTX_end(ctx); + return ret; +} - return(1); -err: - return(0); - } +BN_MONT_CTX *BN_MONT_CTX_copy(BN_MONT_CTX *to, BN_MONT_CTX *from) +{ + if (to == from) + return (to); + + if (!BN_copy(&(to->RR), &(from->RR))) + return NULL; + if (!BN_copy(&(to->N), &(from->N))) + return NULL; + if (!BN_copy(&(to->Ni), &(from->Ni))) + return NULL; + to->ri = from->ri; + to->n0[0] = from->n0[0]; + to->n0[1] = from->n0[1]; + return (to); +} +BN_MONT_CTX *BN_MONT_CTX_set_locked(BN_MONT_CTX **pmont, int lock, + const BIGNUM *mod, BN_CTX *ctx) +{ + BN_MONT_CTX *ret; + + CRYPTO_r_lock(lock); + ret = *pmont; + CRYPTO_r_unlock(lock); + if (ret) + return ret; + + /* + * We don't want to serialise globally while doing our lazy-init math in + * BN_MONT_CTX_set. That punishes threads that are doing independent + * things. Instead, punish the case where more than one thread tries to + * lazy-init the same 'pmont', by having each do the lazy-init math work + * independently and only use the one from the thread that wins the race + * (the losers throw away the work they've done). + */ + ret = BN_MONT_CTX_new(); + if (!ret) + return NULL; + if (!BN_MONT_CTX_set(ret, mod, ctx)) { + BN_MONT_CTX_free(ret); + return NULL; + } + + /* The locked compare-and-set, after the local work is done. */ + CRYPTO_w_lock(lock); + if (*pmont) { + BN_MONT_CTX_free(ret); + ret = *pmont; + } else + *pmont = ret; + CRYPTO_w_unlock(lock); + return ret; +}