On May 28, 2014, at 4:05 PM, Brandon Long <[email protected]> wrote:

> 
> 
> 
> On Wed, May 28, 2014 at 3:46 PM, Barry Leiba <[email protected]> wrote:
> > We could attempt to define a dkim canonicalization that would pass through a
> > mailing list.
> 
> This was beaten pretty severely during the DKIM work, and we couldn't
> come up with anything that was workable.
> 
> > It should include the subject, but have rules for stripping "standard"
> > subject prefixes.  It should obviously include other headers, but not List-*
> > headers (since the RFC for those says the mailing list should strip the
> > existing ones).
> >
> > The body is challenging.  We could have it specify a length... that does
> > allow for some weirdness with html positioning.  We could require no-html
> > part.
> >
> > We could optionally require footers to be added as separate parts, and have
> > the canonicalization be for "first text part".
> 
> Anything that requires mailing list software to change won't work.  If
> mailing list software is changed, the right answer is for the mailing
> list to re-sign the message.  That doesn't help the DMARC situation
> now, but DMARC could be given other options once that happens.
> 
> If the mailing list signs and verifies, that's OAR... and then the challenge 
> becomes knowing when to trust the OAR.

Exactly. TPA-Label provides this answer.

> Also, it seems to me that mailing lists without changes and DMARC are 
> basically incompatible, so I'm unclear of any solution that will allow them 
> to work unchanged.

Again, TPA-Label provides this answer.
By the way, the TPA-Label draft has been updated to include 
Original-Authentication-Results draft reference which also states this concern 
as well. 

> We discussed using l= at (um...) length, and no one liked that
> approach.  There were too many holes, yes: allowing arbitrary amounts
> of data to be added to the message text, having it fail anyway if text
> isn't added at the end (such as for multipart messages), and so on.
> 
> Heuristics involved in tweaking the subject are problematic.
> 
> Some of this could perhaps work if we had a canonicalization that was
> *specific* to mailing list posts... but then how would the signing
> domain know that the message it was signing was going to a mailing
> list?
> 
> Theoretically, a client could easily know its a mailing list in most cases, 
> but yes, where the signing often occurs, its unlikely to know.  The easiest 
> thing to do would be to sign it both ways, and then the receiver might only 
> trust the mailing list canonicalization if it went through a mailing list... 
> and of course that can be forged as well.

That is what spam-traps and DMARC feedback provides.  When a problem is 
detected and reported to the DMARC domain, the response can be to withdraw 
authorization and report problems to third-party service providers.  Fast and 
easy.  Depending on the urgency, it would be nicer to report problems to the 
service provider and give then an opportunity to fix issues before causing an 
unnecessary disruption. The point is to ensure everyone has an incentive to 
cooperate.

There are many cases that are never originally signed by the DMARC domain.  
Such as an accounting package that sends out invoices on behalf of some company 
that wants their email address in the From header since this is what their 
customers will recognize. 

> I really think this isn't a useful approach, but perhaps someone might
> come up with the necessary "aha" to make it work.
> 
> I think it depends on what the goal is.  If the goal is "unforgeable messages 
> through a mailing list to pass DMARC", then yes, this is probably not going 
> to work.  The only solution there is for the mailing lists to actually not 
> munge messages.
> 
> I might argue that "if the mailing list is going to munge the message, then 
> it needs to munge From as well"... but there seems to be a fairly high 
> resistance to that.

This would be an understatement which also ignores other valid uses.  People 
might remember someone said something profound without recalling related 
details. Not being able to search through their message stream would be fairly 
frustrating. Not all mailing lists offer usable reference header fields for 
such use (perhaps to ensure user privacy).  References: 
<CABa8R6v6tk7ZdgtTkBQGT6F7A17VP=zccpmcbojfze7sy99...@mail.gmail.com> might not 
be something that can be used by most MUAs to review what someone said in the 
past, for example

> I suspect that many parties that implement DMARC are "cheating" by allowing 
> things that look like mailing list or forwarded mail through even if they 
> fail auth and the domain is p=REJECT.  Providing a higher bar for when to 
> "cheat" may be useful, then.
> 
> Now, my anti-spam colleagues will say that any hole will eventually be 
> exploited... but given that I don't see how we can make this work with no 
> holes, and if we can't change mailing lists... and any sort of whitelisting 
> is adding a hole... I feel like I'm in a circular argument.

The hurdle that seems to be in everyone's mind is how to go about facilitating 
feedback that is not a lot of work.

Some have used specialize DNS servers to cut corners.  
http://www.corpit.ru/mjt/rbldnsd.html is an example related to dealing with 
email abuse.  There are off-the-shelf DNS servers able to return 30K resource 
records without an issue.  Add a few scripts and voila. 

A specialized server however may assist in the administration of providing 
DMARC related authorizations for third-party domains.  This could be 
mailing-lists, office and facility services using a separate Sender header, 
"acquaintance" initiated message services, etc.  It seems the same specialized 
server could be tasked with processing DMARC feedback looking for confirmation 
in outbound logs.  The server itself might even look for which log domains 
match hashed labels.  This could provide a heads-up something is outside the 
federation.  Once the initial corpus has been accommodated, keeping the system 
maintained should become progressively easier.

Or of course we could say it is too hard which places this problem back into 
making guesses.  When that happens, vast amounts of abuse will suddenly appear. 
 Malefactors quickly learn where the gray areas exist.  The goal is to ensure 
there is as little guesswork as possible while keeping the DMARC domain in 
control.

It seems unlikely there will be many outside service providers willing to make 
decisions about which DMARC related messages are legitimate or not.  Only the 
DMARC domain can realistically be expected to make those decisions.  This then 
means the real task is to then communicate these decisions to all of the DMARC 
domain's potential receivers.

Regards,
Douglas Otis




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

Reply via email to