Dave Crocker writes:
 > On 6/12/2014 11:58 AM, Stephen J. Turnbull wrote:
 > > Dave Crocker writes:
 > > 
 > >  > The scenario being discussed is for a recipient who gets both signatures
 > >  > when they are valid, but who does not know about DKIM-Delegate.
 > > 
 > > I didn't understand that from previous posts.  At least Hector seems
 > > to be concerned (though not exclusively so) with the case I presented.
 > > I suspect John as well.  And I think that case is important.
 > 
 > Now it's my turn to not understand.  Really, what scenario is involved here?

Any situation where the token signature is the only valid signature
from the Author Domain.

 > >  > It ought to prefer the 'stronger' one, but the point being raised
 > >  > is that this is not an issue that has been at issue until now.
 > >  > (Or, at least, not much of an issue until now.)
 > > 
 > > If they're both valid, isn't this "no blood, no foul"?
 > 
 > The concern is a a write-down attack, where the weaker signature is
 > effectively an exploit.

We're talking about verification by the destination MTA (no mention of
mediators anywhere.  If there *is* an attack involving two valid
signatures from the Author Domain, I don't see how either DKIM or
DKIM-delegate can do anything about it.

 > > Is there a concern is that having seen a token signature, it will
 > > ignore the valid signature, and treat the message as high-risk?
 > 
 > Sounds like a generic issue with receiver use of DKIM.  A worthy
 > question, but not inherent to -Delegate.

That's part of what I meant by "quality-of-implementation issue".
Sorry it wasn't clear.

 > >  > The concern is that the weaker signature (that I call a token,
 > >  > given how little of the message it is likely to cover) is more
 > >  > easily re-used for a replay attack.
 > > 
 > > I don't understand what attack you have in mind, if that attack
 > > involves two valid signatures from the Author Domain, the
 > > content-covering signature has a "good" selection of headers,
 > > and doesn't use the l= tag.
 > 
 > The premise is that the weaker nature of the token signature (needed to
 > get it to survive through the mailing list) will make it more subject to
 > some form of replay attack that includes bad content.
 > 
 > I've no idea how serious the threat is, but it's certainly a
 > mathematically legitimate concern.

Yes, I understand that, but I don't see mathematical legitimacy
without an invalid/missing content-coveting signature in the picture
(unless users at the Author Domain have been suborned by the abusers).
So I don't understand why your analysis is limited to the two-valid-
signatures case, which doesn't have an "interesting" failure mode as
far as I can see.

Mathematically, a crappy implementation could indeed choose to reject
because of the presence of a token signature, despite the presence of
a valid content-covering signature.  But the answer is "Well, if that
hurts, don't do it."

_______________________________________________
dmarc mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/dmarc

Reply via email to