bignum: fix boundary condition in montgomery logic
authorGeoff Thorpe <geoff@openssl.org>
Wed, 30 Apr 2014 15:39:24 +0000 (11:39 -0400)
committerGeoff Thorpe <geoff@openssl.org>
Wed, 30 Apr 2014 15:49:31 +0000 (11:49 -0400)
It's not clear whether this inconsistency could lead to an actual
computation error, but it involved a BIGNUM being passed around the
montgomery logic in an inconsistent state. This was found using flags
-DBN_DEBUG -DBN_DEBUG_RAND, and working backwards from this assertion
in 'ectest';

ectest: bn_mul.c:960: BN_mul: Assertion `(_bnum2->top == 0) ||
(_bnum2->d[_bnum2->top - 1] != 0)' failed

Signed-off-by: Geoff Thorpe <geoff@openssl.org>
crypto/bn/bn_exp.c

index baeda37..ea2bd0a 100644 (file)
@@ -494,6 +494,9 @@ int BN_mod_exp_mont(BIGNUM *rr, const BIGNUM *a, const BIGNUM *p,
                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;
+               /* Upper words will be zero if the corresponding words of 'm'
+                * were 0xfff[...], so decrement r->top accordingly. */
+               bn_correct_top(r);
                }
        else
 #endif