On Mon, 8 Sep 2003, Dave Warren wrote:

> If your mail server supports SSL, then the password will
> be sent encrypted. If not, then it sends plain text since
> that is all your mail server is able to handle.

 Badly programmed application resulting in a unencrypted 
version of the passwords is huge bug no matter whether 
receiving party supports encryption or not.
 The application would have to simply refuse to connect to a
party which is not verified and authenticated to be the
reseller. And last time I checked SMTP doesn't provide for
any such authentication. SSL could support this, but *ONLY*
with a (pre-specified) certificate, and it doesn't look like
OpenSRS is storing a certificate.

 *And* it is not a secure systems design to rely on MX
records to deliver something sensitive. Since domain
registration and handling is *above* MX resolution in the
food chain, maning that domain registration and handling
*controls* what an MX record will resolve to, and therefore
must not be controlled by what an MX record will resolve to.
 And OpenSRS has allowed an MX record to determine how a
domain registration and handling will happen(*1) - this is
wrong and upside down !
 (*1) I.e. if anyone manages to steer the traffic that
depends on MX records to themselves, !they gain control over
whole existence of *all* reseller's customers' domains!

 The *ONLY* automated, secure solution is to use data-level 
encryption, not transport-level encryption (unless OpenSRS 
holds a pre-stored certificate *required* for a connection 
to be established).

 Such wide-spread data-level secure solution exists, and is 
called PGP. Each customer should have the option to supply 
the public PGP key to use for sensitive (if not for all) 
email communication with OpenSRS.


> Personally, I don't think the password itself should be
> sent out, but rather a token which can be used to change
> the password, but that's just me.

  I can't believe that OpenSRS actually stores the password
in a form that can be changed back to plaintext. That should
be barred by law (isn't it?). Hash is sufficient ! They have
no business knowing what our password is. Their business is
to know whether supplied password matches the one that was
previously set up. (This is what hashes are for.) And, even
better, only the password hash would be transmitted (would
be calculated by JavaScript on JavaScript-enabled clients.  
But this is not necessary. I have enough trust in OpenSRS
that if they don't store their customers' plaintext
passwords, they won't go to the lengths to leak it to
anybody with a network monitoring tool).


  Mark.

Reply via email to