Re: Potential bug in win32 binary version?

[Simon Josefsson]

"Daniel Armyr" <address@hidden> writes:

> Hi.
> I attempted to secure an existing program win32  by adding gsasl as the
> authentication process. Everything works fine, but we suddenly realized that
> the challenge strings are allways identical. Is this a known bug?
>
> Here is a sample code snippet so show the problem:
>
> char  password[40];
> Gsasl *ctx;
> Gsasl_session *session;
> char *p;
>
> gsasl_init (&ctx);
> gsasl_server_start (ctx,"CRAM-MD5",&session); gsasl_property_set (session,
> GSASL_AUTHID, "admin" ); gsasl_property_set (session, GSASL_PASSWORD,
> password ); gsasl_step64( session, "", &p );
>
> In this case, p is allways exactly the same string every time. Anyone see
> the problem?

Hi!  Thanks for the report.  Did you link gsasl with libgcrypt?
Otherwise, gsasl will read the nonce from /dev/random which probably
doesn't exist on win32.  You should have seen a warning from ./configure
when you built gsasl.

This seems like a serious problem, and we should make sure it can't
happen.  The code in the CRAM-MD5 server doesn't check the return value
from the randomness-function.  Fixing that would be the first step.  A
self-test to test whether two challenges are different should be added
too.

Thanks,
/Simon