On Thursday, February 24, 2022 3:36:18 AM EST Alessandro Vesely wrote:
> On Wed 23/Feb/2022 19:55:35 +0100 Scott Kitterman wrote:
> > On Wednesday, February 23, 2022 1:25:50 PM EST John Levine wrote:
> >> It appears that Scott Kitterman  <[email protected]> said:
> >>>Leaving that aside, then I think it's:
> >>>
> >>>1.  Lookup DMARC record for the 5322.From domain.  If found, that's the
> >>>policy.
> >>>
> >>>2.  Walk up from up to 5 levels in the DNS hierarchy and the last
> >>>
> >>> (shortest) non-PSD DMARC record you find is the org.
> >>>
> >>>I think that matches the RFC 7489 discovery approach, just not using PSL.
> >>>Is that right?
> 
> It looks right, except for non-flagged PSDs.
> 
> However, what if you find no DMARC record at step 1?  For compatibility with
> RFC 7489, I guess you imply that the policy is taken from the org domain.
> >> I prefer the first (longest) but could live with the last if people think
> >> that will in practice be less surprising.  I do worry about foo.us.com vs
> >> bar.us.com.
> > 
> > I think it will (less surprising).  Currently if you have a.b.bar.us.com
> > and "a" and "b" need different policies you can just publish them.  If
> > you stop at the first one, then b.bar.us.com would be identified as the
> > org domain for a.b.bar.us.com vice bar.us.com.
> > 
> > This could be problematic.  The RFC 7489 definition of relaxed alignment
> > says that the From org domain and the DKIM/SPF org domain have to be the
> > same to be aligned.
> > 
> > A specific example:
> > 
> > A business (example.com) uses subdomains for particular business units and
> > allows them to run their own email infrastructure if needed.  One
> > particular business unit (bu.example.com) uses a large number of third
> > party providers with a dedicated mail from for each to avoid restrictions
> > associated with SPF DNS lookup limits.  It sends mail from (among others)
> > ext1.bu.example.com and ext2.bu.example.com with from and signing domain
> > of bu.example.com.  Because this business unit has decided to accept the
> > risks associated with a DMARC reject policy and the company as a whole
> > has not, it publishes its own DMARC record at _dmarc.bu.example.com.  Of
> > course, .com is in the PSL, so the org domain is example.com under RFC
> > 7489 rules.
> > 
> > In this example, under RFC 7489 rules, ext1.bu.example.com,
> > ext2.bu.example.com, and bu.example.com all meet the requirements for
> > relaxed alignment because the org domain for all three is example.com.
> > 
> > Using the proposed tree walk approach, it matters if you take the first
> > DMARC record or the last one.
> > 
> > ext1: no dmarc
> > bu: dmarc, p=reject, sp=reject
> > example: dmarc, p=none, sp=none
> > com: no dmarc
> 
> I like this example.  Let's leave the alignment question for a moment, and
> consider policy.  Mail from [email protected], according to the plan
> above, goes with p=none.  That agrees with the PSL.
> 
> Can we be more accurate than the PSL?  Yes, we can, since we read the record
> at bu.example.com.
> 
> > If you walk up the tree and take the first DMARC record you hit, then the
> > org domain for ext1.bu.example.com is bu.example.com and the org domain
> > for bu.example.com is example.com.  They don't match, so there is no
> > alignment.
> > 
> > If you take the last non-PSD DMARC record then the org domain for
> > ext1.bu.example.com is example.com and the org domain for bu.example.com
> > is
> > example.com and they meet the requirements for relaxed alignment.
> 
> Allowing bu.example.com to publish org=y would make both identifiers aligned
> to bu.example.com.  However, we could also devise flags so as to override
> the policy but keep the alignment.
> 
> I don't know why you dismiss possible enhancements as if psd=y were already
> standard.  Of course, users of legacy software won't deploy new
> enhancements. That is, org=y would initially not be honored by the majority
> of receiving servers.  That's normal.  OTOH, a change which implies issuing
> more DNS queries just to get a slightly worse quality than the PSL —due to
> missing psd=y flags— is not really so attractive.

I think using psd=n, so we don't need to define yet another new tag, as 
discussed in another branch of this thread, accomplishes what you are 
suggesting.

Scott K


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

Reply via email to