spf-discuss
[Top] [All Lists]

Re: Re: [IETF] Allocation of the new RR type for SPF

2004-11-24 06:40:53
In <6076(_at_)rama(_dot_)pamho(_dot_)net> "Roger Moser" 
<Roger(_dot_)Moser(_at_)rama(_dot_)pamho(_dot_)net> writes:

Wayne wrote:

I suspect that the IETF won't want to see SPF RRs used for things
other than SPF records for the same reasons they don't want to see TXT
RRs used:  it might fill up the 512 byte UDP packet

Why should for example the sender policy of yahoo.com be too large to fit in
an UDP packet. (Currently it fits).

I'm probably not a good person to speak for what the IETF/DNS folks
will want or do, as I really don't think all the stuff they do is
rational.

My guess is that they will say something like "What if someone comes
up with a new Sender Policy type that uses XML with large chunks of
VB?  That would cause all SPF RR lookups to become too large!"  Or,
the slight variation: "What if 1000 people come up with new Sender
Policies?"




and there is no way to select only one sender policy type.

You will get all policy types in one UPD packet and then you can select the
desired policy. (Currently SPF does it like this: It gets all TXT records
and selects the "v=spf1" TXT record.)

The need for selection is not because it is hard to find the policy
type that you are interested in.  The need for selection is so that
you can make sure that everything always fits in a 512 UDP packet.


The folks promoting DK, SES, IIM, etc. will need to ask the DNS folks
to allocate special records for each of those uses.

You mean the receiver should then do 4 or more DNS queries to retrieve the
sender policy?

I expect the IETF/DNS folks to say "yes", and then immediately turn
around and complain about the load that this places on DNS servers.  I
get the impression that many of them don't want DNS used for anything
new and that they should be able to pick and choose which exceptions
to this "no new DNS usage" can be adopted.  


As I mentioned above, there is no good method to migrate from one RR type
to another and it is a nasty catch-22.

Okay, then let us declare the TXT RR as the sender policy record. It already
is that anyway and it will remain that for quite some years.

If you want to have some fun, post that to the namedroppers list.  I'm
not sure where in the world you live, but I'm sure you will be able to
hear the screams of horror anyway.


I see two possible solutions to this catch-22:

1) Create a large number of new RRs that act like TXT records, maybe
   100 or even 1000 of them.  New DNS RR usage could pick any one they
   think is unlikely to be heavily used and would need to make sure
   their new usage has a Unix-like "magic number" or token that makes
   their new usage distinct.

2) Use a EDNS0 type extention to DNS so that DNS lookups can select
   which TXT records they want.  The selection method shouldn't be
   anything fancy, just something to select on, say, the first 5-10
   characters of the TXT RR.

The major problem I see with both of these solutions is that the
IETF/DNS folks would then have zero control over new DNS usage.


-wayne