On Wed, Sep 7, 2016 at 4:57 PM, Viktor Dukhovni <[email protected]>
wrote:

> If 404 is a soft fail, and maxage=0 is revocation, what should a sending
> MTA
> do when it encounters a 404 (or indeed an HTTPS connection timeout,
> handshake
> failure, ...)?  Retry the HTTPS policy lookup on every delivery?  Employ
> some
> sort of hold-down timer between soft-failed lookups?  How should the
> hold-down
> be chosen?
>

My point was that a 404 could be treated the same as a DNS lookup failure.
Admittedly, it need not be so; 404s are of course signed implicitly if
delivered over HTTPS. But because of the operational concerns I outlined, I
had thought it made more sense to treat them however we treat DNS
resolution failures.

(In particular, I believe we specify that if a new, valid policy cannot be
found, the cached policy should be used. So in the case of a 404 we would
temp-fail and so forth.)


> My preference is to not have to hammer away at broken or misconfigured the
> HTTPS servers with every message delivery, that can have a big effect on
> SMTP
> delivery latency and thereby throughput.
>

The only time it results in that is also if the cached policy is invalid.
That is to say, the possibilities are:

Sender has cached policy: Y/N
Cached policy validates the MX: Y/N

If the above is Y/Y, the sender can merely deliver. If the above is Y/N,
the sender should not deliver. If the above is N/_, the sender can deliver,
same as for any other domain missing a policy. Regardless of the result of
a new policy fetch, that applies (so long as the cached policy is not
expired).

No?


> Returning to John Levine's suggestion of SRV records, what should the
> reference
> identifier (and SNI value) be in that case?  If it is the SRC record
> target host,
> should any sub-domain (at any depth) of the nexthop domain be admissible
> if that
> becomes the reference identifier?  Or must the target host present a
> certificate
> that exactly matches the original service (nexthop) domain?
>
> I think some language that nails this down is needed in the spec.  Please
> discuss.
>

Well, only if we go with SRV records. Which (as you describe) seems to
expand the complexity slightly. I think this can (of course) be done
without any security misses, but I'm not sure it's preferable to the
hardcoded hostname.

That said, I don't want to poo-poo anyone's ideas of a nice zonefile. ;)
Maybe we should bite the bullet and use SRV, but I'd like to hear some
strong consensus for that before I try to write the language for it.


>
> --
>         Viktor.
>
> _______________________________________________
> Uta mailing list
> [email protected]
> https://www.ietf.org/mailman/listinfo/uta
>
_______________________________________________
Uta mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/uta

Reply via email to