On 2010-06-14 at 12:09 +0100, Ian Eiloart wrote:
> --On 11 June 2010 22:37:09 -0700 Phil Pennock <exim-users@???>
> wrote:
> > DKIM contains two ways of forming the message body, for signing
> > purposes, and optionally lets you only sign the first 'l' bytes of the
> > body. So you could theoretically use l=0 but at this time I can't
> > conceive of a scenario where that would be wise.
>
> It might allow the signature to survive the addition of a mailing list sig.
So for that, you'd sign l=<message-len>, rather than l=0.
It would fail for mailing-list sigs inside MIME parts, or where a new
container MIME part was added to support the sig. Or any sort of MIME,
where the original MIME termination would fail.
I'd still be worried about a spammer finding a way to abuse a lax MIME
parser and add a new HTML part which uses an iframe and manages to
obscure the original content.
I think that if you run a mailing-list manager which modifies content at
all, whether it's a message footer or Subject: manipulation, then you
should be looking to strip DKIM-Signature: from mails as part of
processing the mails. There's no need to embed any replacement
signatures or know anything more than "this is a checksum header, we're
breaking the checksum, strip the header out". It would probably be more
polite to rename it to Old-DKIM-Signature: rather than remove it. And
processing DomainKey-Signature: in the same way would be good.
There's no obligation on MLM maintainers or admins to use DKIM
themselves, just to strip it.
Senders who advertise policies that mails from their domain will always
be signed will find that their domain can't be reliably used for sending
to a mailing-list. That puts the cost and burden squarely back on the
people using DKIM, instead of it being an externality. This is fair,
and I say this as a DKIM proponent.
-Phil
