>We've been looking at weak dkim from the angle of reducing what's covered
>by the signature... but unfortunately, that takes out the major parts of
>the message that we care about.
>
>I'm wondering if there is a different alternative.

When we were doing DKIM, we went around and around and around trying
to figure out what sort of modifications we could allow and still
consider messages to be "the same", with negligible success.

All we came up with was relaxed canon and l=NN.  The point of relaxed
was to allow for the sort of message tidying up that sendmail used to
do before people understood the distinction between mail submission
and mail relay.  It is my impression that these days it's rare to have
a message that passes relaxed that wouldn't also pass strict.  The
point of l= was to allow mailing lists to add footers, not so much
because we thought that would solve a lot of problems but because it
was the only thing we could think of that might solve one problem and
it was cheap to specify and implement.  Again, I don't get the
impression that it's used very much, and messages with l=N would
usually pass anyway.

Google has some great algorithms people, so maybe they can come up
with some essence of message fingerprint that works, but I'd be
surprised.  Anything that a mailing list might do to add innocuous
text to a message, a spammer can do the same but adding spam text.  I
don't see any way around that.  We can probably come up with a hack
for subject lines just because subject lines are so short, but I'd be
pretty surprised if anything more general were workable.  (Proposed
subject hack: add a new code to DKIM-Signature that says to use the
copied subject from z= when verifying the signature, but only if the
contents of that subject is a substring of the current subject.
De-MIME-ing is a quality of implementation issue.)

For my double signing hack, I'm coming at it from the other direction,
what problem are we trying to solve.  When AOL and Yahoo made their
unfortunate DMARC decisions, they had a very concrete problem to
solve.  They'd allowed crooks to steal address books, so Yahoo (or
AOL) users were getting spam with Yahoo addresses of people they knew,
sent from random botnets, not from Yahoo.  I believe that double
signing still solves 99.9% of that problem while allowing mailing
lists to do most of what they ordinarily do.

While it is true that a malicious re-signing mediator could still
delete the contents of a weakly signed message and replace it with
spam, this greatly decreases the attack surface.  The malicious party
has to have a recent real message in hand from the real author, and
the real author's mail system has fine grained control over who they
allow to re-sign.  Start with a heuristic about what domains are
likely to be mailing lists (something I know gmail has from the
comments in the aggregate reports), put weak signatures on mail sent
to them.  If a particular mediator's spamminess spikes, you can stop
putting weak signatures on mail to the bad actor without having to
break mail to everyone else.  

For incoming mail, the double signature allows you to tell the
difference between direct mail and mediated mail, with the double
signature not meaning that the mail is "real", but rather that it came
via someone with at least a weak relationship with the author so it's
likely non-horrible enough to be worth running through the normal spam
filters.

This does require heuristics and tuning, but so does any spam filter,
and what this does seems similar to what any spam filter does anyway.

R's,
John

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

Reply via email to