Network filtering means blocking traffic at the network level. Trying to use DNS for this leaves you trusting the client - and there are decades of precedent for clients bypassing that for various reasons, such as this post shows.
The solution is to start doing network filtering: if you block packets to unapproved servers, you can actually stop this. You’ll need to run your own proxy, of course, but that’s always been the only way to actually accomplish that goal.
That isn’t really possible, though: CDNs mean that blocking by IP just doesn’t work. The most effective method I’ve found is transparently redirecting all traffic on port 53 to a DNS server I control. DoH means that I might as well setup a transparent HTTPS proxy.
"That isn’t really possible, though: CDNs mean that blocking by IP just doesn’t work. The most effective method I’ve found is transparently redirecting all traffic on port 53 to a DNS server I control. DoH means that I might as well setup a transparent HTTPS proxy."
This is a very good point and I am dealing with this myself on my home networks.
Like any household/family we have some number of dubious/untrusted devices that still need Internet access.
By establishing my own recursive resolver I can act as a chokepoint (and monitoring point) for their behavior online. It's a very elegant solution, actually, and I have created a nice integration between my datacenter-hosted resolver and nextdns.io as the adblocking upstream DNS.
DoH breaks all of this.
I have no interest in diving down the "MITM my own network by inserting custom certs into embedded devices that may or may not use them".
Since we're talking about it, though, it occurs to me that you could quickly do a DoH lookup to every single new IP connected to, outbound, from your network - and then block all IPs that answer your DoH query. You're basically pre-testing all new SSL connections to see if they are to a DoH resolver that you (presumably) don't want to talk to ...
This solves the CDN problem ... does it solve the problem entirely ? I have only just thought of this moments ago ...
> By establishing my own recursive resolver I can act as a chokepoint (and monitoring point) for their behavior online. It's a very elegant solution, actually, and I have created a nice integration between my datacenter-hosted resolver and nextdns.io as the adblocking upstream DNS.
This only works for the subset of devices which use the local DNS. If they use any of the well-known techniques to avoid that filtering it's completely ineffective.
> Since we're talking about it, though, it occurs to me that you could quickly do a DoH lookup to every single new IP that initiates a new connection, outbound, from your network - and then block all IPs that answer your DoH query.
It doesn't solve the CDN problem: CDNs will route traffic based on the hostname and blocking them will have a degree of collateral damage which most people can't work with. Setting up your own HTTPS proxy avoids this.
"This only works for the subset of devices which use the local DNS. If they use any of the well-known techniques to avoid that filtering it's completely ineffective."
If you also block all port 53 after allowing your own resolver ... you may have some headaches with devices that refuse to use the DHCP provided resolvers but you know they aren't going to other resolvers.
That kind of control is what DoH breaks and I'd love to find an elegant (non-MITM proxy) solution for it ...
There isn’t an effective solution for a device which ignores local network policy other than returning it so the manufacturer pays the cost of designing a bad system.
With the side effect of your local vendor refusing to do further business with you, "the problem customer", with your "unreasonable demands" and technobable.
DNS over HTTPS has been a thing since before IETF standardized it, the technique was just in the form of a non-standardized API running on some benign domain.
Thinking about it, if you’re willing to give up TLS 1.3, you could probably just break all https connections with encrypted SNI and then filter based on the SNI information.
Just to nitpick: TLS 1.3 still uses plaintext SNI by default. You need to explicitly put public keys in DNS to enable the encrypted SNI extension.
And in the context of pihole and such, avoiding that means editing the DNS response to remove those public keys. Which takes us full circle back to "do I control DNS for this gadget, or not".
You’ll note that I mentioned a proxy. That’s why: you need to force all traffic through a proxy you control or you’re just hoping that a client doesn’t use hard-coded IPs or an outside API of some source. If your network allows the client to send traffic to port 443 anywhere, any blocking is on the honor system.
My question here is how are you going to install your root certificate on a $300 smart TV? Or, if that is not required because the TV does not verify DoH certificates, how bad is that for security (which we already know is awful on these devices)?
This comes back to the core decision: do you care about controlling the network enough to block access? If the device can't be managed / is no longer supported the safest choice is not to allow it online at all. Different people will have different risk tolerances – it might make sense to put, say, a remote-control power switch on an IoT no-man's land network but if it has access to personal information or cameras/microphones it's not unreasonable to say it should just be blocked unless you're actively using those features.
You have two choices: trust the device or don’t let it on the network. Voluntary measures like local DNS only work to the extent that the device maker wants them to.
The solution is to start doing network filtering: if you block packets to unapproved servers, you can actually stop this. You’ll need to run your own proxy, of course, but that’s always been the only way to actually accomplish that goal.