On Wed, Jun 11, 2014 at 3:22 PM, John Levine <[email protected]> wrote:

> Someone sends off a message to a mailing list with the two DKIM
> signatures and DKIM-Delegate.  Someone else, perhaps a list
> subscriber, notes that the weaker signature doesn't cover the body, so
> he replaces the body with nose enlargement spam and blasts it out.
> Recipient MTAs which haven't been updated since 2014 and don't know
> about DKIM-Delegate see the weak but valid signature and since the
> signer has a generally good reputation, delivers it.  Ugh.
>

Right, that's the "considerable latitude" that's already mentioned in the
draft.  The "x=" is the current protection against that sort of abuse.
It's not much, but it's what we've thought of so far.  The alternative is
to use an "l=" that is the length of the original rather than 0, which
constrains the abuse considerably but might reduce the likelihood of this
working.


> This hasn't been a problem before.  Although you've always been
> allowed to use weak signatures, there's been no advantage to doing so,
> so nobody did.  Now you do, but with new semantics that you shouldn't
> pay much (any?) attention to that signature unless it's paired with
> the forwarder's.  One could make an argument that it's not technically
> a semantic change to DKIM (indeed, Dave just did), but in practical
> terms, it is likely to interact poorly with existing unupgraded
> software, so I'd want a version bump so that the old software ignores
> the special purpose signature.
>

I see your point, though it seems strange to do a version bump when that's
really the only change to the bits on the wire, and the only real change
then is how the signatures are interpreted; syntax vs. semantics.


> Bonus question: why put the author domain and target domain fields in
> a new header rather than just addding ddd=<author> and
> ddt=<forwarders> to the signature header?
>

Originally it was done as a separate tag, but that meant changing DKIM in
some way even if it means registering a new tag that some/most will just
ignore.  Using a separate header field keeps DKIM itself pristine, and just
builds on top of it.  Would reverting to a tag method obviate the need for
a version bump?

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

Reply via email to