On 27/Aug/11 05:26, John Levine wrote:
> In sec 5, it says you can identify the consumer domain by DKIM d=,
> SMTP bounce address domain, or MTA rDNS.  I'm wondering whether it
> is a problem that they all funnel into _report.<domain> without
> any way to tell the difference.  If it mattered, I'd suggest adding
> a tag in the record saying which sort of name the record applies
> to, with multiple records if need be.

+1, how about Discovery-Origin?

However, I'm skeptical about MTA rDNS.  It is not the same level of
domain that one would expect in DKIM d= and the domain-part of an
address.  If we include host-X.example.com as a possible discovery
domain, we should also allow helo-names.  IME, admins won't bother
defining _report.host-X.example.com for each host, so I'd drop them.

> For feedback generators, I have no idea how I'm supposed to pick a
> domain name to look up.  If, say, I wanted feedback from Gmail, would
> that be google.com? gmail.com? googlemail.com? The name of one of
> their MX hosts?  Any or all of the bazillion mail domains they host?
> Something else?

Good question.  Where did you get "Gmail" from? :-)  Perhaps, the
following could be added, e.g. as a fourth paragraph, in section 5:

  A feedback consumer who wishes to receive feedback from a
  generator, may also query the domains it targets.  For example, an
  MTA sending mail to [email protected] may want to query
  _report.example.org in order to ascertain under what conditions it
  can have generated feedback sent back to it.

> In 5.1, I'd delete the rt= and ri= tags since I don't see how they
> improve interoperability.  If someone is inviting me to send them
> abuse reports, I'm going to send them abuse reports, and the number of
> reports will be limited by the amount of mail they send.  They can
> always ignore the ones they don't understand.

Yeah, this conundrum transverses various reporting docs.  I don't
think we can afford to specify a different behavior for each report type.

> For the r= tag, I don't understand why it's supposed to be able both
> to accept ARF abuse reports and to respond to a subscription
> verification request.  Presumably the response is automated, since no
> human is going to be looking through the stream of ARF for
> subscriptions.  Is there some yet to be defined protocol here?
> 
> Also it might be better to make r= a URI rather than an email address,
> so it can be a mailto: or an http: allowing POST.
> 
> Why is there a gp=r policy but no rp=r policy?  I accept random mail
> to my abuse address, but for feedback senders I know, I treat their
> ARFs differently (basically, I believe all of the latter and treat
> them as unconditional unsubscribes.)

I concur on the former point, but would prefer that r=, if given, be
an email address, for ease of use.  Perhaps, we could allow rp=r and
specify that r= is /Optional, but MUST be defined unless rp=r/.  The
URI in ru= could provide for communicating a generator-specific email
target address.  If r is not given, only acknowledged generators can
send feedback.

> In 5.4, I would suggest specifically limiting the kind of URIs that
> the fields can include, unless someone is prepared to explain what
> to do if you see ru=nntp:news.example.com.

How about fax:+12024562461?  Currently, URIs are meant to be used
interactively, so users can decide on their own.

Perhaps, we might reserve some kind of URIs, e.g. http queries, for
automated to-be-defined purposes.

> Sec 9, security, the obvious problem is that a malicious sender can
> publish "rp=o [email protected]" and indirectly mailbomb people.

Should that be equivalent to a redirection?  I mean, should

   _report.example.org. IN TXT "rp=o [email protected]"

be considered equivalent to, say,

   _report.example.org. IN CNAME _report.somewhere.com?

> The mention of subscription verification suggests there's supposed to
> be some sort of handshake before turning on the cannon, but if it's
> there, I've missed it.

Verifying that somewhere.com wants feedback for example.org's spam
looks like one of those automated to-be-defined purposes...

jm2c
_______________________________________________
marf mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/marf

Reply via email to