Put more debug screening in BN_div() and correct a comment.

[openssl.git] / crypto / bn / bn_div.c

diff --git a/crypto/bn/bn_div.c b/crypto/bn/bn_div.c
index 64b84ac1a76d06f294cae875b44f9578f3bc6fa8..0fef7ced207fb80b60a2bf7c051c36e32fe4c46b 100644 (file)
--- a/crypto/bn/bn_div.c
+++ b/crypto/bn/bn_div.c
@@ -127,9 +127,10 @@ int BN_div(BIGNUM *dv, BIGNUM *rem, const BIGNUM *m, const BIGNUM *d,
 
 #else
 
-#if !defined(NO_ASM) && !defined(NO_INLINE_ASM) && !defined(PEDANTIC) && !defined(BN_DIV3W)
+#if !defined(OPENSSL_NO_ASM) && !defined(OPENSSL_NO_INLINE_ASM) \
+    && !defined(PEDANTIC) && !defined(BN_DIV3W)
 # if defined(__GNUC__) && __GNUC__>=2
-#  if defined(__i386)
+#  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)
@@ -149,9 +150,23 @@ int BN_div(BIGNUM *dv, BIGNUM *rem, const BIGNUM *m, const BIGNUM *d,
            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>
+    */
+#  define bn_div_words(n0,n1,d0)               \
+       ({  asm volatile (                      \
+               "divq   %4"                     \
+               : "=a"(q), "=d"(rem)            \
+               : "a"(n1), "d"(n0), "g"(d0)     \
+               : "cc");                        \
+           q;                                  \
+       })
+#  define REMAINDER_IS_ALREADY_CALCULATED
 #  endif /* __<cpu> */
 # endif /* __GNUC__ */
-#endif /* NO_ASM */
+#endif /* OPENSSL_NO_ASM */
 
 
 /* BN_div computes  dv := num / divisor,  rounding towards zero, and sets up
@@ -170,6 +185,8 @@ int BN_div(BIGNUM *dv, BIGNUM *rm, const BIGNUM *num, const BIGNUM *divisor,
        BN_ULONG d0,d1;
        int num_n,div_n;
 
+       bn_check_top(dv);
+       bn_check_top(rm);
        bn_check_top(num);
        bn_check_top(divisor);
 
@@ -199,10 +216,10 @@ int BN_div(BIGNUM *dv, BIGNUM *rm, const BIGNUM *num, const BIGNUM *divisor,
 
        /* First we normalise the numbers */
        norm_shift=BN_BITS2-((BN_num_bits(divisor))%BN_BITS2);
-       BN_lshift(sdiv,divisor,norm_shift);
+       if (!(BN_lshift(sdiv,divisor,norm_shift))) goto err;
        sdiv->neg=0;
        norm_shift+=BN_BITS2;
-       BN_lshift(snum,num,norm_shift);
+       if (!(BN_lshift(snum,num,norm_shift))) goto err;
        snum->neg=0;
        div_n=sdiv->top;
        num_n=snum->top;
@@ -212,12 +229,13 @@ int BN_div(BIGNUM *dv, BIGNUM *rm, const BIGNUM *num, const BIGNUM *divisor,
         * This is the part that corresponds to the current
         * 'area' being divided */
        BN_init(&wnum);
+       wnum.flags = BN_FLG_STATIC_DATA; /* prevent accidental "expands" */
        wnum.d=  &(snum->d[loop]);
        wnum.top= div_n;
-       wnum.dmax= snum->dmax+1; /* a bit of a lie */
+       wnum.dmax= snum->dmax - loop; /* so we don't step out of bounds */
 
        /* Get the top 2 words of sdiv */
-       /* i=sdiv->top; */
+       /* div_n=sdiv->top; */
        d0=sdiv->d[div_n-1];
        d1=(div_n == 1)?0:sdiv->d[div_n-2];
 
@@ -233,6 +251,7 @@ int BN_div(BIGNUM *dv, BIGNUM *rm, const BIGNUM *num, const BIGNUM *divisor,
        /* space for temp */
        if (!bn_wexpand(tmp,(div_n+1))) goto err;
 
+       bn_correct_top(&wnum);
        if (BN_ucmp(&wnum,sdiv) >= 0)
                {
                if (!BN_usub(&wnum,&wnum,sdiv)) goto err;
@@ -248,7 +267,8 @@ int BN_div(BIGNUM *dv, BIGNUM *rm, const BIGNUM *num, const BIGNUM *divisor,
        for (i=0; i<loop-1; i++)
                {
                BN_ULONG q,l0;
-#ifdef BN_DIV3W
+#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;
@@ -266,6 +286,11 @@ int BN_div(BIGNUM *dv, BIGNUM *rm, const BIGNUM *num, const BIGNUM *divisor,
                        q=(BN_ULONG)(((((BN_ULLONG)n0)<<BN_BITS2)|n1)/d0);
 #else
                        q=bn_div_words(n0,n1,d0);
+#ifdef BN_DEBUG_LEVITTE
+                       fprintf(stderr,"DEBUG: bn_div_words(0x%08X,0x%08X,0x%08\
+X) -> 0x%08X\n",
+                               n0, n1, d0, q);
+#endif
 #endif
 
 #ifndef REMAINDER_IS_ALREADY_CALCULATED
@@ -290,11 +315,18 @@ int BN_div(BIGNUM *dv, BIGNUM *rm, const BIGNUM *num, const BIGNUM *divisor,
                        BN_ULONG t2l,t2h,ql,qh;
 
                        q=bn_div_words(n0,n1,d0);
+#ifdef BN_DEBUG_LEVITTE
+                       fprintf(stderr,"DEBUG: bn_div_words(0x%08X,0x%08X,0x%08\
+X) -> 0x%08X\n",
+                               n0, n1, d0, q);
+#endif
 #ifndef REMAINDER_IS_ALREADY_CALCULATED
                        rem=(n1-q*d0)&BN_MASK2;
 #endif
 
-#ifdef BN_UMULT_HIGH
+#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
@@ -318,14 +350,19 @@ int BN_div(BIGNUM *dv, BIGNUM *rm, const BIGNUM *num, const BIGNUM *divisor,
 #endif /* !BN_DIV3W */
 
                l0=bn_mul_words(tmp->d,sdiv->d,div_n,q);
-               wnum.d--; wnum.top++;
+               wnum.d--; wnum.top++; wnum.dmax++;
                tmp->d[div_n]=l0;
+               /* XXX: Couldn't we replace this with;
+                *    tmp->top = div_n;
+                *    bn_fix_top(tmp);
+                */
                for (j=div_n+1; j>0; j--)
                        if (tmp->d[j-1]) break;
                tmp->top=j;
 
                j=wnum.top;
-               BN_sub(&wnum,&wnum,tmp);
+               bn_correct_top(&wnum);
+               if (!BN_sub(&wnum,&wnum,tmp)) goto err;
 
                snum->top=snum->top+wnum.top-j;
 
@@ -333,7 +370,7 @@ int BN_div(BIGNUM *dv, BIGNUM *rm, const BIGNUM *num, const BIGNUM *divisor,
                        {
                        q--;
                        j=wnum.top;
-                       BN_add(&wnum,&wnum,sdiv);
+                       if (!BN_add(&wnum,&wnum,sdiv)) goto err;
                        snum->top+=wnum.top-j;
                        }
                *(resp--)=q;
@@ -341,13 +378,20 @@ int BN_div(BIGNUM *dv, BIGNUM *rm, const BIGNUM *num, const BIGNUM *divisor,
                }
        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_correct_top(snum);
                BN_rshift(rm,snum,norm_shift);
                if (!BN_is_zero(rm))
-                       rm->neg = num->neg;
+                       rm->neg = neg;
+               bn_check_top(rm);
                }
        BN_CTX_end(ctx);
        return(1);
 err:
+       bn_check_top(rm);
        BN_CTX_end(ctx);
        return(0);
        }