Alessandro Vesely <[email protected]> wrote:

>On 09/Feb/12 22:08, Scott Kitterman wrote:
>> Alessandro Vesely <[email protected]> wrote:
>>>On 09/Feb/12 06:54, Scott Kitterman wrote:
>>>> 
>>>> I think this is a MUST NOT.  All it will lead to is pain.
>>>
>>> A corner case: an ISP configure SPF according to their own mail 
>>> traffic, e.g. "v=spf1 mx:ISP.example ~all".  The ISP don't trust 
>>> their customers enough to point WHOIS records directly to their 
>>> abuse-mailboxes, but does count-and-forward according to 8.10. An
>>> ISP's customer sends spam with MAIL FROM:<[email protected]>.
>> 
>> There's no reliable way to distinguish that scenario from regular
>> spoofing, so it's pointless to use to drive design.
>
>Except that we'd be providing a recipe for building non-reportable
>forgeries.
>
>> ISP can police his own network without external help, so it's not a
>> useful distinction in any case.
>
>The amount of support that ISPs may want to give to abuse reporting is
>unknown at this time.  Requiring them to proactively policing abusive
>use of domain names in outgoing mail is probably not an option, also
>because that might interfere with contractual habits (more than
>explicit complaints from third parties.)

This would only be relevant for their own domain, some other domain would be a 
different case. What I suggest isn't nearly so broad as you infer. 

>Advice that abuse-mailboxes shouldn't be spam-filtered is spread
>throughout several papers.  That the domain part of abuse-mailboxes
>should not be SPF-protected would be a new, somewhat obscure
>requirement, that is not going to improve SPF adoption.

I don't agree. I don't see such a requirement. 

>>> We are already saying that the domain in MAIL FROM is likely a 
>>> reasonable candidate if SPF gave "pass".  Doesn't that already
>>> mean you should not use it if SPF gave any other result?
>> 
>> It's a little more complicated than that and I think the suggested
>> text navigates the complexity correctly.
>
>We mention how to derive a destination address from
>
>* WHOIS,
>* rDNS, and
>* "pass" for SPF/DKIM domain.
>
>Any other destination is dicey, and SHOULD NOT be used.  IOW, an
>address derived from WHOIS, rDNS, or DKIM does not need to be doubly
>checked against SPF none/neutral.  And let me stress this once more:
>we are talking _only_ of addresses derived in one of those three ways.
>
>Yes, it is more complicated.  For example, if I get DKIM pass and SPF
>fail for the same domain, I should hypothesize a replay attack which
>might be better reported as an authentication failure.  But that's
>well beyond what this I-D is going to specify/state.

I think the DKIM pass with the same domain is a reasonable case.  I can imagine 
a rouge ESP scenario where notifying the domain owner is a good idea.

Thanks.

I'll go back and modify my answer to Murray.

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

Reply via email to