On Friday, May 23, 2025 8:58:49 PM CEST Paul Wouters wrote:
> [ speaking only as individual DNS enthousiast ]
High five! ^^

> So this is when the application "knows" it is online, but does not know
> if its system-overriding nameserver is reachable or not?
> 
> i.e. you connect to startbucks, it gives you DNS 1.2.3.4, but your
> application wants to use 8.8.8.8 and so it has to probe for something
> and you propose this something to be some hardcoded vendor neutral name
> that hopefully doesn't cause queries beyond the actual resolver
> targetted, due to the zone being locall served by that remote DNS
> resolver?

I think this is something worth drilling into, with a set scenario (e.g. local 
network failure), and set configuration styles.

Assume for example such public network that uses an internal DNS server. Let's 
not include captive portals just yet, but something along those lines.

So you join the network, get some parameters from DHCP, and that includes a 
local DNS server but the gateway doesn't function for whatever reason. You 
could ask the local DNS server about names it is locally authoritative for, 
and maybe it can respond to some of them (maybe including 
probe.resolver.arpa). But what gives? It responded, not the SOA or anything 
else more conclusive on the path.

Meanwhile if the network connectivity does work properly, and perhaps the 
local DNS server does not have this hardcoded in an RPZ or such. So it decides 
to forward that query to whatever it is configured to relay to. Where would 
that query end up? Should other entities on the path be configured to respond 
to this query like the local resolver would've done otherwise? What does that 
say about connectivity? What if it's not just Starbucks or Flixbus or whatever 
that's down, what if it's their upstream ISP being under e.g. DDoS attack? 
What meaning does their ability to serve an ISP-local request serve?

Don't get me wrong, I do like the idea of a vendor-neutral name -- even if 
that currently means ambiguity on where those requests would be handled. I'd 
imagine solving that to be the purpose of this here WG.

> I see the advantage of this. Response time should be fairly quick
> because no further resolving is required on the remote resolver part.

So this, completely agreed. But the purpose of the servers on the path should 
be well-defined. And it should ultimately lead to a full path towards the 
server that's being actually probed against. If that is 8.8.8.8, then Google 
should respond with whatever CDN technology they have. And we could have a 
discussion about whether a CDN says much about connectivity from IXP to other 
nations, but I'll admit that to be somewhat far-fetched. If it is Cloudflare or 
Quad9 or anyone else to be probed against meanwhile, same thing. They're being 
asked the question, that's what the application expects a response from.
 
> But I am bit puzzled by why a DNS probe is even needed when doing
> DoH or DoT. Since you already build a TLS connection, isn't that in
> itself already a proof of reachability - without sending a DNS query?
> 
> Doing this for port 53 seems odd. If unencrypted, you might as well use
> the local network's DNS server since you are giving them all the data
> anyway and all the opportunities to falsify the responses.
> 
> Paul


-- 
Met vriendelijke groet,
Michael De Roover

Mail: [email protected]
Web: michael.de.roover.eu.org


_______________________________________________
DNSOP mailing list -- [email protected]
To unsubscribe send an email to [email protected]

Reply via email to