On 19.7.2017 05:15, Mark Andrews wrote:
> In message <[email protected]>, 
> =?UTF-8?B?UGV0ciDFoHBhxI1law==?= writes:
>> On 11.7.2017 13:23, Ted Lemon wrote:
>>> On Jul 11, 2017, at 3:17 AM, Petr paek <[email protected]
>>> <mailto:[email protected]>> wrote:
>>>> I feel that implications from switch to non-RR format are
>> underestimated
>>>> and following e-mail attempts to explain why I believe it is a bad
>> idea.
>>>> Please accept my apology for such long e-mail.
>>>
>>> Petr, with all due respect, I did not see a counter-proposal here, and
>>> your comments seem to reflect a misunderstanding of what
>>> session-signaling is.
>>>
>>> In fact, the opposite of what you said is true: if this were done as a
>>> normal query with EDNS0-like encapsulation, _then_ we would see
>>> problems, because session signaling messages would look more like DNS
>>> queries, and less like control messages.  This is not a desirable
>> quality.
>>>
>>> It's true that, for example, the DNS packet compression format would
>>> have to deal with this specially, but that would also be true if this
>>> were done EDNS0-style.   It's true that packet dumpers would have to
>>> deal with this specially, but that's also true if it's done EDNS0-style.
>>>   Etc.
>>
>> Let me clarify that I'm not insisting specifically on EDNS0. From my
>> perspective anything which does not deviate from current wire format is
>> okay.
>>
>>
>>> It may be that there is a good point in your argument somewhere, but at
>>> the moment, I don't see one.   E.g., in your python example, yes, if
>>> this were an RR, not being able to plop it into your RR-handling switch
>>> would suck.   But it's not an RR, doesn't have the semantics of an RR,
>>> and if you plop it into your RR-handling switch, you're probably getting
>>> the semantics wrong.
>>
>> This might be key to the the misunderstanding:
>>
>> I'm not talking about semantics at all. What I object to is the proposed
>> wire format, not the semantics. The fact that bytes are stored in format
>> compatible with RR (RFC 1035 section 4.1.3) is not related to its
>> semantics.
>>
>>> So if you want to make this case, I think you need to be more specific
>>> about why this is a problem: when I think about how to implement this
>>> (which I have done, because I'm using it for dnssd), what you are
>>> advocating seems harder, not easier, than what is currently being
>> proposed.
>> I'm trying to explain that deviance from RR format (RFC 1035 section
>> 4.1.3) will force parties in the DNS ecosystem to implement support for
>> the new wire format even though they are not interested in session
>> signaling at all.
> 
> Hogwash.  Servers which are unaware of the opcode will return NOTIMP
> or FORMERR depending on how forward thinking the developer was.
> 
> If you have a packet dumper YOU ALREADY HAVE TO ACCOUNT FOR THIS.
> There are ZERO guarentees that a packet addresses to/from port 53
> is a DNS packet.  If your packet dumper can't handle these then
> it is already broken.
> 
> If you are just proxying DNS messages this should be transparent. If
> the proxy is parsing the message then it returns NOTIMP or FORMERR.
> 
>> For a lot of cases it would be sufficient if SS data could be treated as
>> one of many "unknown RR types" (RFC 3597). For example tools for
>> statistics would work, the capture formats would not need special
>> extensions for SS, etc.
>>
>> We can discuss this in depth during dnsop session today.
WG decided to go forward with TLV format. I still think it is a mistake
but I'm going to respect WG decicion and thus I'm giving up. Further
discussion on this topic is pointless.

-- 
Petr Špaček  @  CZ.NIC

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

Reply via email to