ietf-mxcomp
[Top] [All Lists]

Re: Backward compatibility with deployed SPF records (and choice of domain)

2004-06-25 22:27:40

MOstly agreeing with Roy here, but I'm raising a couple of questions that could use extra pairs of eyes and feedback...



1. Placement of TXT records.

    The co-chairs find that the working group has not come to consensus on
    the use of a record prefix vs [unprefixed] TXT records


I think I would agree with Roy that allowing MARID-compliant MTAs to check the unprefixed name is desirable.

In other words, if MARID-Brand LMAP works substantially similar to SPF-Brand, let's go ahead and use their data.

One concern here is that we would want the semantics to be similar enough to not upset or surprise anyone who isn't expecting their data to be used a slightly different way, but I think we are able to do that. (That certainly was the intent of the SenderID merged proposal)

In a straight race, I would probably root for the non-prefixed records to win, simply because the underscore might scare and confuse people (it was removed from early versions of SPF for mostly that reason). I do understand the issue that the prefix is intended to address, but I don't think the research we have shows it is anything to worry about. I think most users will not have TXT records of any other type, and of those that do, the records will mostly be small enough to coexist peacefully. I don't have a strong preference in this area, except that backward compatibility would allow MARID to harness and redirect the momentum SPF has built up. (Of course we will be going through the process of getting our own RR so we will eventually have an escape plan)

Does anyone have a strong objection to using unprefixed txt records?


1b. [the working group] has not
    reached consensus on how to address domains with wildcard MX records.



Actually I think this one is pretty easy to fix... if there is a wildcard A or wildcard MX record already there, the site should also create a wildcard TXT record.

(I really wish we could go back in time and tell people they need to set up MX records for all deliverable domains and hosts and not fallback to A... any chance we could declare the fallback-to-A practice deprecated? It may be officially outside our scope, but it really makes the job of publishing LMAP records a lot harder.)


Let me add this sort of related point...

1c. The group has not agreed on an alternate placement for TXT records other than the label exactly matching the (host part of the) identity being checked. I.e. a MARID record at example.com does not cover user(_at_)www(_dot_)example(_dot_)com(_dot_)

That means: the domain owner should create LMAP records for ALL hosts in his domain that have A or MX records. (Note that a wildcard TXT record does not cover names that have A or MX but no TXT, only names with no defined data at all)

Is this something we need to work on more, or should we leave it like that? We talked about looking for the "closest SOA name" and looking there, and applying the TXT record at that location to other subdomains that have no TXT of their own.

Another idea might be to have records set up like:
example.com.  IN TXT  "v=xxx1 a mx ptr ?all"   ; only for mail @example.com
mail1.example.com.  IN TXT "v=xxx1 a -all"     ; for @mail1.example.com
_all.example.com. IN TXT "v=xxx1 -all" ; for @*.example.com if no TXT

In this case I would totally support use of the underscore because this is a little bit of an advanced feature anyway, but it doesn't have to be an underscore to work. Then again this could be a performance problem... if most domains don't have TXT records we would have to do a second query to find the SOA and a third to find the higher-level (either example.com or _all.example.com) so I don't want to push too hard for climbing up the chain if there isn't a lot of support for it.

Any feedback as to whether we should pursue this any further or just leave it as the domain itself only and no searching at higher levels?



Given MARID records are similar, but not identical, to SPF records,
and that it is at least possible that there will shortly be a large
number of SPF checkers in real world use, I'd like to propose the
following strategy:

MARID should define syntax versions 1 and 2 (identified by v=spf1 and
v=spf2) as identical.  MARID-compliant implementations must interpret
both of these records identically, and must use the highest understood
version record if multiple versions are found (this is the same rule
as in SPF).


I think the records will be similar enough that we can honor the original publisher's intentions. However, there's nothing really *wrong* with defining a new prefix. MTAs will need to be revised a bit anyway to be MARID-compliant, and a TXT lookup will still get both records.

What would folks in the group think about defining our own prefix like "v=lmap1"? I think this is only really needed in the off chance that the semantics are different enough for people to notice. I expect that people will continue to publish "v=spf1" for a while until the new MARID-compatible MTAs are widespread, then they could publish either. Someone would really only want to publish both if they want one policy for SPF Classic and a different policy for MARID-Brand LMAP.

(I called it LMAP because that seems to be a better name for the protocol, and it leaves the door open for MARID-the-group to work on and publish other stuff. The protocol is important but it's not the only important thing the group has worked on :)




--
Greg Connor <gconnor(_at_)nekodojo(_dot_)org>