On Sun, 2006-09-10 at 20:58 -0400, Hector Santos wrote:
> ----- Original Message -----
> From: "Douglas Otis" <[EMAIL PROTECTED]>

> >> Why would you want to send a signed message to a mailing
> >> list server that is
> >> a) not DKIM-SSP compliant and
> >> b) known to alter the integrity of the message?
> >
> > Although it may be practical for a domain to ensure all outbound mail
> > is appropriately signed, it is _not_ practical for a domain to alter
> > the operations of thousands of mailing-list services without outright
> > banning their use.
> 
> So I ask again, why would you *blindly* send signed mail to a mailing list
> or to just any receiver without having any clue about how they will handle
> it?   What do you expect from promusciously sending your precious signed
> mail into the wild wild west?  What benefit are you looking for and what you
> are expecting?

An administrator may make an assertion all email-addresses within the
domain are initially signed, AND only compliant services are used.
Making the first assertion, the administrator needs to ensure users
utilize specific outbound services.  This added requirement should not
limit where messages are eventually sent.  

Making the second assertion will impact an ability to participate in a
mailing-list.  This second assertion might be desired to curb spoofing
by limiting acceptance of invalid signatures as an alternative to
reliance upon retained email-address annotations.  Retained
email-address annotation also prevents look-alike attacks however, and
provides far better protection.  This second assertion might be seen as
just a stop-gap measure.

Indivdual users may decide to participate in a mailing-list where a
signature is likely damaged.  The eventual recipients are the
mailing-list subscribers, and the situation where signatures are damaged
will not change anytime soon.  When the first assertion is made,
email-addresses with invalid signatures could be limited to sources
known to damage signatures, if limited at all. 

You are suggesting that by signing a message, it should only be accepted
when the signature verifies.  This will greatly discourage deployment of
DKIM, let alone publishing policy that all messages are initially
signed.  Consider that DKIM annotations based upon retained
email-addresses offers superior anti-spoofing protection without a need
to block messages with invalid signatures or to selective sign messages.

> > This provides a practical means to deal with signature failures in
> > the most restrictive fashion possible without expecting the world to
> > suddenly become DKIM compliant.
> 
> Huh? You mean you expect failures when you submit signed mail to a
> mailing list?  I don't get it.

Absolutely.  How do you expect initial signatures to survive the typical
formating and flattening, in addition to subject line changes and
unsubscribe and ad information normally appended or prefixed to the
message body?  Any effort to accommodate message modifications only
invites abuse that could now come from anywhere, rather than known
services.

> >> All this does is promote the "Cry Wolf" syndrome, harming your
> >> own domain while also putting others at risk of receiving more
> >> DKIM junk disguised as your domain.
> >
> > The only alternative option when dealing with you or Thomas would
> > be to not publish any assertion that all email-addresses are
> > initially signed.
> 
> You mean NEUTRAL operations? Legacy Operations?  If I understand you,
> you're right.  Legacy Operations is still the #1 requirement.  But
> once you step up, you better cross the tees, dot the eyes, correctly.
> Rest assured, abuse will not be tolerated. Certainly it turned out to
> be worst than every before.

A bad actor can sign using DKIM and publish any highly restrictive
policy they wish.  Navigating policy aimed at blocking messages will
prove most difficult for legitimate users and administrators.  The bad
actor will keep turning the knob until their messages are accepted.  

> > Rather than permitting a practical transitional process, perhaps too
> > many share your mindset and wish to force everything into becoming
> > instantly DKIM compliant.  That approach is almost certain to backfire
> > and may cause policy not to be published, or cause DKIM not to be
> > deployed. : (
> 
> You're right.  I expect people with high value domains will avoid DKIM
> all together without a dual SIGNER/RECEIVER negotiation concept. They
> will quickly realize there is a high risk of exploitation otherwise.

Exactly how would DKIM be exploited?  An invalid signature should offer
no advantage.

> But worst, if anything, the receivers, if they see an increase
> bombardment of subjectively determined bad messages with DKIM
> fingerprints, they might just use it as a quick and dirty public key
> detector for filtering the mail.

While this approach might be employed in an attempt to curb replay
abuse, this will also need to accommodate the number of subscribers in
various mailing-lists.  List that identify mail-lists should be easy to
compose.  Perhaps bulk senders might be forced into using unqiue
signatures for each subscriber. : (

> I already see this now with the junk DomainKeys mail. I can easily see
> adding a rule to quick DNS lookup just to check if there is a public
> key for mail with DKIM and DKEYS signatures.  No need to bother with
> the signature/message hashing verification overhead.

This would leave your domain exposed to abuse once that strategy becomes
apparent to the bad actors.

> If you want to do this right, when you have to be selective on what
> you sign, know who you sending it to and have some clue or expectation
> about how its going to be handled. Otherwise, whats the point?  It
> would be a "Cry Wolf" waste of time to process DKIM mail without any
> payoff to it.

While that strategy might be considered plausible, it seems unlikely to
find concensus.  SMTP is a store and forward system.  There is no way to
know where the finial destination of the message might be, and what
might happen to the message in transit.  DKIM should be robust and able
to deal with damaged signatures.  The use of retained email-address
annotation ensures protection is not lost as a result of signature
damage.

> In short, half-based solutions don't usually work very well and we
> have enough of that already.

There should be a strategy that makes sense and that is easy to
administer.

-Doug







_______________________________________________
NOTE WELL: This list operates according to 
http://mipassoc.org/dkim/ietf-list-rules.html

Reply via email to