123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423 |
- /*
- * Copyright 1995-2016 The OpenSSL Project Authors. All Rights Reserved.
- *
- * 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
- */
- #include <openssl/bn.h>
- #include "internal/cryptlib.h"
- #include "bn_lcl.h"
- /* The old slow way */
- #if 0
- int BN_div(BIGNUM *dv, BIGNUM *rem, const BIGNUM *m, const BIGNUM *d,
- BN_CTX *ctx)
- {
- int i, nm, nd;
- int ret = 0;
- BIGNUM *D;
- bn_check_top(m);
- bn_check_top(d);
- if (BN_is_zero(d)) {
- BNerr(BN_F_BN_DIV, BN_R_DIV_BY_ZERO);
- return (0);
- }
- if (BN_ucmp(m, d) < 0) {
- if (rem != NULL) {
- if (BN_copy(rem, m) == NULL)
- return (0);
- }
- if (dv != NULL)
- BN_zero(dv);
- return (1);
- }
- BN_CTX_start(ctx);
- D = BN_CTX_get(ctx);
- if (dv == NULL)
- dv = BN_CTX_get(ctx);
- if (rem == NULL)
- rem = BN_CTX_get(ctx);
- if (D == NULL || dv == NULL || rem == NULL)
- goto end;
- nd = BN_num_bits(d);
- nm = BN_num_bits(m);
- if (BN_copy(D, d) == NULL)
- goto end;
- if (BN_copy(rem, m) == NULL)
- goto end;
- /*
- * The next 2 are needed so we can do a dv->d[0]|=1 later since
- * BN_lshift1 will only work once there is a value :-)
- */
- BN_zero(dv);
- if (bn_wexpand(dv, 1) == NULL)
- goto end;
- dv->top = 1;
- if (!BN_lshift(D, D, nm - nd))
- goto end;
- for (i = nm - nd; i >= 0; i--) {
- if (!BN_lshift1(dv, dv))
- goto end;
- if (BN_ucmp(rem, D) >= 0) {
- dv->d[0] |= 1;
- if (!BN_usub(rem, rem, D))
- goto end;
- }
- /* CAN IMPROVE (and have now :=) */
- if (!BN_rshift1(D, D))
- goto end;
- }
- rem->neg = BN_is_zero(rem) ? 0 : m->neg;
- dv->neg = m->neg ^ d->neg;
- ret = 1;
- end:
- BN_CTX_end(ctx);
- return (ret);
- }
- #else
- # if !defined(OPENSSL_NO_ASM) && !defined(OPENSSL_NO_INLINE_ASM) \
- && !defined(PEDANTIC) && !defined(BN_DIV3W)
- # if defined(__GNUC__) && __GNUC__>=2
- # if defined(__i386) || defined (__i386__)
- /*-
- * There were two reasons for implementing this template:
- * - GNU C generates a call to a function (__udivdi3 to be exact)
- * in reply to ((((BN_ULLONG)n0)<<BN_BITS2)|n1)/d0 (I fail to
- * understand why...);
- * - divl doesn't only calculate quotient, but also leaves
- * remainder in %edx which we can definitely use here:-)
- *
- * <appro@fy.chalmers.se>
- */
- # undef bn_div_words
- # define bn_div_words(n0,n1,d0) \
- ({ asm volatile ( \
- "divl %4" \
- : "=a"(q), "=d"(rem) \
- : "a"(n1), "d"(n0), "r"(d0) \
- : "cc"); \
- q; \
- })
- # define REMAINDER_IS_ALREADY_CALCULATED
- # elif defined(__x86_64) && defined(SIXTY_FOUR_BIT_LONG)
- /*
- * Same story here, but it's 128-bit by 64-bit division. Wow!
- * <appro@fy.chalmers.se>
- */
- # undef bn_div_words
- # define bn_div_words(n0,n1,d0) \
- ({ asm volatile ( \
- "divq %4" \
- : "=a"(q), "=d"(rem) \
- : "a"(n1), "d"(n0), "r"(d0) \
- : "cc"); \
- q; \
- })
- # define REMAINDER_IS_ALREADY_CALCULATED
- # endif /* __<cpu> */
- # endif /* __GNUC__ */
- # endif /* OPENSSL_NO_ASM */
- /*-
- * BN_div computes dv := num / divisor, rounding towards
- * zero, and sets up rm such that dv*divisor + rm = num holds.
- * Thus:
- * dv->neg == num->neg ^ divisor->neg (unless the result is zero)
- * rm->neg == num->neg (unless the remainder is zero)
- * If 'dv' or 'rm' is NULL, the respective value is not returned.
- */
- int BN_div(BIGNUM *dv, BIGNUM *rm, const BIGNUM *num, const BIGNUM *divisor,
- BN_CTX *ctx)
- {
- int norm_shift, i, loop;
- BIGNUM *tmp, wnum, *snum, *sdiv, *res;
- BN_ULONG *resp, *wnump;
- BN_ULONG d0, d1;
- int num_n, div_n;
- int no_branch = 0;
- /*
- * Invalid zero-padding would have particularly bad consequences so don't
- * just rely on bn_check_top() here (bn_check_top() works only for
- * BN_DEBUG builds)
- */
- if ((num->top > 0 && num->d[num->top - 1] == 0) ||
- (divisor->top > 0 && divisor->d[divisor->top - 1] == 0)) {
- BNerr(BN_F_BN_DIV, BN_R_NOT_INITIALIZED);
- return 0;
- }
- bn_check_top(num);
- bn_check_top(divisor);
- if ((BN_get_flags(num, BN_FLG_CONSTTIME) != 0)
- || (BN_get_flags(divisor, BN_FLG_CONSTTIME) != 0)) {
- no_branch = 1;
- }
- bn_check_top(dv);
- bn_check_top(rm);
- /*- bn_check_top(num); *//*
- * 'num' has been checked already
- */
- /*- bn_check_top(divisor); *//*
- * 'divisor' has been checked already
- */
- if (BN_is_zero(divisor)) {
- BNerr(BN_F_BN_DIV, BN_R_DIV_BY_ZERO);
- return (0);
- }
- if (!no_branch && BN_ucmp(num, divisor) < 0) {
- if (rm != NULL) {
- if (BN_copy(rm, num) == NULL)
- return (0);
- }
- if (dv != NULL)
- BN_zero(dv);
- return (1);
- }
- BN_CTX_start(ctx);
- tmp = BN_CTX_get(ctx);
- snum = BN_CTX_get(ctx);
- sdiv = BN_CTX_get(ctx);
- if (dv == NULL)
- res = BN_CTX_get(ctx);
- else
- res = dv;
- if (sdiv == NULL || res == NULL || tmp == NULL || snum == NULL)
- goto err;
- /* First we normalise the numbers */
- norm_shift = BN_BITS2 - ((BN_num_bits(divisor)) % BN_BITS2);
- if (!(BN_lshift(sdiv, divisor, norm_shift)))
- goto err;
- sdiv->neg = 0;
- norm_shift += BN_BITS2;
- if (!(BN_lshift(snum, num, norm_shift)))
- goto err;
- snum->neg = 0;
- if (no_branch) {
- /*
- * Since we don't know whether snum is larger than sdiv, we pad snum
- * with enough zeroes without changing its value.
- */
- if (snum->top <= sdiv->top + 1) {
- if (bn_wexpand(snum, sdiv->top + 2) == NULL)
- goto err;
- for (i = snum->top; i < sdiv->top + 2; i++)
- snum->d[i] = 0;
- snum->top = sdiv->top + 2;
- } else {
- if (bn_wexpand(snum, snum->top + 1) == NULL)
- goto err;
- snum->d[snum->top] = 0;
- snum->top++;
- }
- }
- div_n = sdiv->top;
- num_n = snum->top;
- loop = num_n - div_n;
- /*
- * Lets setup a 'window' into snum This is the part that corresponds to
- * the current 'area' being divided
- */
- wnum.neg = 0;
- wnum.d = &(snum->d[loop]);
- wnum.top = div_n;
- /*
- * only needed when BN_ucmp messes up the values between top and max
- */
- wnum.dmax = snum->dmax - loop; /* so we don't step out of bounds */
- /* Get the top 2 words of sdiv */
- /* div_n=sdiv->top; */
- d0 = sdiv->d[div_n - 1];
- d1 = (div_n == 1) ? 0 : sdiv->d[div_n - 2];
- /* pointer to the 'top' of snum */
- wnump = &(snum->d[num_n - 1]);
- /* Setup to 'res' */
- if (!bn_wexpand(res, (loop + 1)))
- goto err;
- res->neg = (num->neg ^ divisor->neg);
- res->top = loop - no_branch;
- resp = &(res->d[loop - 1]);
- /* space for temp */
- if (!bn_wexpand(tmp, (div_n + 1)))
- goto err;
- if (!no_branch) {
- if (BN_ucmp(&wnum, sdiv) >= 0) {
- /*
- * If BN_DEBUG_RAND is defined BN_ucmp changes (via bn_pollute)
- * the const bignum arguments => clean the values between top and
- * max again
- */
- bn_clear_top2max(&wnum);
- bn_sub_words(wnum.d, wnum.d, sdiv->d, div_n);
- *resp = 1;
- } else
- res->top--;
- }
- /* Increase the resp pointer so that we never create an invalid pointer. */
- resp++;
- /*
- * if res->top == 0 then clear the neg value otherwise decrease the resp
- * pointer
- */
- if (res->top == 0)
- res->neg = 0;
- else
- resp--;
- for (i = 0; i < loop - 1; i++, wnump--) {
- BN_ULONG q, l0;
- /*
- * the first part of the loop uses the top two words of snum and sdiv
- * to calculate a BN_ULONG q such that | wnum - sdiv * q | < sdiv
- */
- # if defined(BN_DIV3W) && !defined(OPENSSL_NO_ASM)
- BN_ULONG bn_div_3_words(BN_ULONG *, BN_ULONG, BN_ULONG);
- q = bn_div_3_words(wnump, d1, d0);
- # else
- BN_ULONG n0, n1, rem = 0;
- n0 = wnump[0];
- n1 = wnump[-1];
- if (n0 == d0)
- q = BN_MASK2;
- else { /* n0 < d0 */
- # ifdef BN_LLONG
- BN_ULLONG t2;
- # if defined(BN_LLONG) && defined(BN_DIV2W) && !defined(bn_div_words)
- q = (BN_ULONG)(((((BN_ULLONG) n0) << BN_BITS2) | n1) / d0);
- # else
- q = bn_div_words(n0, n1, d0);
- # endif
- # ifndef REMAINDER_IS_ALREADY_CALCULATED
- /*
- * rem doesn't have to be BN_ULLONG. The least we
- * know it's less that d0, isn't it?
- */
- rem = (n1 - q * d0) & BN_MASK2;
- # endif
- t2 = (BN_ULLONG) d1 *q;
- for (;;) {
- if (t2 <= ((((BN_ULLONG) rem) << BN_BITS2) | wnump[-2]))
- break;
- q--;
- rem += d0;
- if (rem < d0)
- break; /* don't let rem overflow */
- t2 -= d1;
- }
- # else /* !BN_LLONG */
- BN_ULONG t2l, t2h;
- q = bn_div_words(n0, n1, d0);
- # ifndef REMAINDER_IS_ALREADY_CALCULATED
- rem = (n1 - q * d0) & BN_MASK2;
- # endif
- # if defined(BN_UMULT_LOHI)
- BN_UMULT_LOHI(t2l, t2h, d1, q);
- # elif defined(BN_UMULT_HIGH)
- t2l = d1 * q;
- t2h = BN_UMULT_HIGH(d1, q);
- # else
- {
- BN_ULONG ql, qh;
- t2l = LBITS(d1);
- t2h = HBITS(d1);
- ql = LBITS(q);
- qh = HBITS(q);
- mul64(t2l, t2h, ql, qh); /* t2=(BN_ULLONG)d1*q; */
- }
- # endif
- for (;;) {
- if ((t2h < rem) || ((t2h == rem) && (t2l <= wnump[-2])))
- break;
- q--;
- rem += d0;
- if (rem < d0)
- break; /* don't let rem overflow */
- if (t2l < d1)
- t2h--;
- t2l -= d1;
- }
- # endif /* !BN_LLONG */
- }
- # endif /* !BN_DIV3W */
- l0 = bn_mul_words(tmp->d, sdiv->d, div_n, q);
- tmp->d[div_n] = l0;
- wnum.d--;
- /*
- * ingore top values of the bignums just sub the two BN_ULONG arrays
- * with bn_sub_words
- */
- if (bn_sub_words(wnum.d, wnum.d, tmp->d, div_n + 1)) {
- /*
- * Note: As we have considered only the leading two BN_ULONGs in
- * the calculation of q, sdiv * q might be greater than wnum (but
- * then (q-1) * sdiv is less or equal than wnum)
- */
- q--;
- if (bn_add_words(wnum.d, wnum.d, sdiv->d, div_n))
- /*
- * we can't have an overflow here (assuming that q != 0, but
- * if q == 0 then tmp is zero anyway)
- */
- (*wnump)++;
- }
- /* store part of the result */
- resp--;
- *resp = q;
- }
- bn_correct_top(snum);
- if (rm != NULL) {
- /*
- * Keep a copy of the neg flag in num because if rm==num BN_rshift()
- * will overwrite it.
- */
- int neg = num->neg;
- BN_rshift(rm, snum, norm_shift);
- if (!BN_is_zero(rm))
- rm->neg = neg;
- bn_check_top(rm);
- }
- if (no_branch)
- bn_correct_top(res);
- BN_CTX_end(ctx);
- return (1);
- err:
- bn_check_top(rm);
- BN_CTX_end(ctx);
- return (0);
- }
- #endif
|