ietf-dkim
[Top] [All Lists]

[ietf-dkim] Mixing path registration together with ADSP

2009-11-04 19:50:05
Mail From path registration was aimed at reducing a "backscatter"  
spoofing.  This problem had grown to where now most adopt strategies  
that keep abusive messages from being issued by their servers in the  
form of DSNs and where MTAs now normally verify valid recipients prior  
to acceptance.  The levels of abuse often required valid recipient  
checks to conserve server resources.  Secondly, most automated sources  
of DSNs are now better at removing much of the body to keep abuse  
out.  IMHO,  a "report" mime subtype for auto responses could also  
reduce false positives.

Nevertheless, some now hope to base email acceptance upon Mail From  
path registration or upon valid Author Domain Signatures.  Some expect  
either technique can be used to white-list their domain, and to limit  
the issuance of DSNs.  Unfortunately, determining whether a message is  
within a registered Mail From path can expend many transactions, and  
may fail for forwarded messages.  Some have proclaimed DKIM can amend  
Mail From path registration to include valid Author Domain Signatures  
as alternative acceptance criteria.  While that might sound good,   
neither Mail From or PRA path registration prevent deceptive use of  
 From email addresses.  The general goal behind ADSP was to prevent  
 From email address deception.  So the other question that might be  
asked, should ADSP practices be amended to include Mail From path  
registration as an optional acceptance criteria for mailing-lists or  
other third-party signers?

Those wanting to cross-the-beams, or to splice together the different  
kinds of Mail Tubes are likely to expose users to fraudulent messages  
that are indistinguishable from those of genuine mail.  For SMTP to  
survive, it needs to deal with abuse early within the process, which  
entails acceptance at connections likely based upon consistent use of  
hostnames.  Receivers need to pay attention to hostnames reported by  
any particular IP address, and not dismiss it because some path  
registration had been found.  This should not be hard, since normally  
only a few hostnames used by each over long periods.  Exceptions can  
be handled separately.  Sources that appear genuine then need to be  
monitored individually, where this too is not hard when limited to  
trusted sources.  DNS based self references are easily manipulated,  
and often DNS servers are also compromised systems.  As such, finding  
a matching address record in the forward direction says little unless  
published by a trusted domain.  The question of trust becomes more  
difficult as a slew of new TLDs are added.  It would be nice to have a  
protocol that offered an easier way to determine answers to question  
of who can be trusted without needing a massive database or a  
centralized service.  So which are the trustworthy domains?

DKIM/ADSP is really about mitigating ongoing fraud.  Connecting the  
Mail From Tube together with the From Tube defeats the From Tube's  
otherwise good protection.  The TPA-Label scheme offers a safe way to  
combine Tubes via an authorization strategy.  By not depending upon  
path registration, a receiver is more likely to monitor for connection  
abuse that can be determined without any additional transactions.  A  
trusted name list can be developed based upon those domains being  
authorized-by-name.  Third-party handlers would receive "votes of  
trust" from domains who are attempting to protect against fraud in  
disguise.  Lately, I have received several mailing-list like messages  
designed to appear as if from a mailing list or a social network with  
whom I have subscribed.  It is not hard to establish reasonable  
thresholds of genuine traffic that would be needed to keep this type  
of assessment from being easily gamed.

Unless your domain is a likely phishing target or perhaps a Fortune  
500 company, ADSP is unlikely to offer the type of handling that may  
be desired.  However, the web today has become a massive voting  
scheme, which is a fairly democratic trend.  Admittedly, many have  
learned how to game the system, but the gaming can be mitigated  
through greater participation.  Delegating domains can be detrimental,  
and exchanging keys and synchronously setting up DNS references is  
error prone and expensive, which together reduces the needed  
participation.  The TPA-Label scheme should make it easy for domains  
to vote for the services being trusted by-name.

Trust-by-name, the only scheme that scales. :^)

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

<Prev in Thread] Current Thread [Next in Thread>
  • [ietf-dkim] Mixing path registration together with ADSP, Douglas Otis <=