Devices already use pinned certificates, but DoH gives the device manufacturer a convenient and unblockabke way to resolve names against a trusted server. Device manufacturers could be jerks already, but I'd argue that DoH lowers the barrier-to-entry for being a jerk.
Consumers don't, by and large, give a damn about whether they actually own their devices or not. This is yet another reason why those of us who care about ownership and control of our own networks can't have the cool new "toys". We're back to feudalism.
Call it the opposite of defense-in-depth. Device manufacturers who aren't thorough might allow a class of "attacks" by way of using DNS. Having ubiquitous DoH takes away that "attack" vector.
I'm not arguing device manufacturers need DoH to be jerks. Having it available just enables the less sophisticated ones to be jerky more easily. DNS-over-TLS does the same thing, it's just marginally less frustrating because it can be blocked at the network layer, at least.
I understand that allowing owners of devices to do what they want on their own networks runs counter to the whole "protect the public from bad network operators" / "protect the public from bad people" narrative. I'm in the minority. It's just damned frustrating. It's sucking all the fun out of computing and networking. It's irrational and self-serving, but dammit it makes me angry.
You see, one thing is one or two vendors misusing protocols to remove users control from their devices. Another thing is an official protocol that does the same thing. We find have anywhere to run anymore.
As bi dollar corporations and organizations are more and more "protecting" ourselves they need to remove control from ourselves.
Well, with regular DNS, you can force-route 8.8.8.8 to pi-hole and regain control relatively easily.
That's not possible with DoH due to TLS.
Yeah static IP routing is an option - for both nameservers and nefarious servers. That is, we may just have to resort to using IP blocks.
And now that it's being marketed heavily, it just gave them another reason to do it that they didn't have before ("Mozilla is doing it too!")
No. Run your own DoH server, point your DoH client to your own DoH server, and your DoH server can then provide any responses you want. (And it in turn can use DoH or do standard recursive resolution.)
End-to-end encryption means that malicious parties can't MITM your traffic or otherwise make you talk to the wrong destination server. That doesn't prevent you from intentionally running your own local servers and pointing clients to them.
If you have a client device on your network that you don't trust and can't change the configuration of, it has far more secure methods for protecting itself from you than just using DoH.
Except when the client ignores what you tell it. Per Paul Vixie:
google, this is bogus as hell. my dhcp server gives you dns servers to
use. please don't make me route and answer 8.8.8.8 just to watch youtube.
> [71] 2019-02-13 16:39:40.548137 [#68 vtnet0 4095] \
> [24.104.150.186].56915 [8.8.8.8].53 \
> dns QUERY,NOERROR,7357,rd \
> 1 lh3.googleusercontent.com,IN,A 0 0 0
> [71] 2019-02-13 16:39:40.548210 [#69 vtnet0 4095] \
> [24.104.150.186].56915 [8.8.8.8].53 \
> dns QUERY,NOERROR,49247,rd \
> 1 lh3.googleusercontent.com,IN,AAAA 0 0 0
(no, this device i've paid for, will NOT be allowed to send you any
information, other than what i personally approve, which will never
include DNS traffic. if you don't like that deal, buy it back from me
and i'll find some other video appliance that doesn't twist my arm.)
Of course with DoH you can't just do a UDP redirect. With DoT, which uses tcp/953, you can at least block access.
But with DoH, you have just lost control of your network. (Unless block 443 and force everything to go through a (Squid) proxy?)
If you can't trust your device, you can't trust it without DoH either. As I said upthread, devices don't even need to use DNS. They can open up encrypted C&C channels, using whatever protocol they'd like, tunneled over whatever they'd like. They can bake rendezvous IP addresses in, they can cryptographically verify endpoints to break MITM proxies, they can even sneak data out in ICMP packets. DoH has literally nothing to do with it.
What's insidious about this argument is that it preys on peoples legitimate concerns about uncontrolled devices on their network. We're all concerned about that! But that doesn't mean we should use plaintext DNS, or Paul Vixie's preferred "encrypted DNS with a killswitch for network operators".
Might as well argue against using windmills to grind wheat.
Sometimes you don't choose the software, the software chooses you:
* https://www.zdnet.com/article/first-ever-malware-strain-spot...
* https://www.zdnet.com/article/psixbot-malware-upgraded-with-...
I'm happy to implement DoT on work's network to help encourage encrypted DNS. My problem is DoH can now be used by all sorts garbage and we can't tell it from legitimate HTTPS--which is by design.
All this for the laudable goal of protecting Internet surfers' privacy. However I'm curious to know how well this will work against the Great Firewall of China, and how many other regimes will now implement Great Firewalls of their own.
Malware can ship with an arbitrary number of DNS-over-HTTPS servers accessible anyway, no matter whether Google supports it. The cat was out of the bag as soon as you allowed encryption on your network in general.
Finally, there are system architectures that are inherently less susceptible to malware, which is what we should really be focusing on. If your document viewer doesn't need to access the Internet, why does it have access to the Internet? Why does your general-web-browsing browser have access to your intranet and worse, your network drives and USB sticks? The reason we're stuck with what we have now is that operating systems research largely died in the 90s, but we're now aware that e.g. microkernel capability systems can be implemented in ways that are both fast and secure.
[1] https://www.cs.kau.se/philwint/scramblesuit/wpes2013.pdf
Expect “enterprise” network equipment using an approach like this to provide a “DOH firewall” in the near future.
Net privacy gained: negative.
The goal for DoH is to improve privacy for individual humans, and it does that.
No reason you can't get a real certificate, to ensure that your device talks to the server it intends to.
> If you're locked down... well, it's game over.
Devices today can already use pinned certificates and other similar measures to ensure that they only talk to the server they want to talk to. DoH doesn't make that any more or less effective, just simpler to get right.
You'd have to get a certificate for whatever domain the device wants to talk to - a domain you very likely have no control over.
No sane CA will give you a cert for that. The whole point of the PKI is to keep MITMs from pretending to be domains that they actually don't control - which is exactly what we'd be doing here.
If 90% of users leave it as shipped set to (say) Google, 9% switch to their preferred service and only a handful like you actually type in a DNS name, that freedom to do is preserved. If it doesn't check the certificate from the DoH server when you override the configuration, you are immediately exposed to a MITM so it needs to check. However you'll have no trouble getting a cert for your own FQDN, and that's what the poster you replied to is thinking of.
I believe there are ideas for a network-level autodiscovery of DoH servers, analogous to regular DNS. That would solve the problem, but I can't see how it would get buy-in from the DoH crowd, as it would tie DNS to the network again - and it would still have to be implemented by every single device.
Of course all of this is moot if the vendor actively wants to prevent you from intercepting DNS.
What DOH client would that be? Where would I configure it? How do I configure a systemwide DOH setting?
Right now none of those questions can be given a simple answer. DOH is a rogue technology.
And don't forget that you can change the resolver today, as you can install uBO, but what about tomorrow?
It won't if it defaults to centralized on-profit business like Cloudflare. They can tap for govts or sell aggregated data to ISPs someday. They aready own a lot of unencrypted traffic from TLS-termination.
Wouldn't it be better to use public/private key signing and onion-like infrastructure for DNS resolving and still keep it distributed?
Isn't the real problem that you don't control the device in the first place? Perhaps the better solution is to only run devices and software that you can actually trust, if the security of those devices is important to you, or if you care about the device's behavior.
(Edit: removed bit about root CA certs because I misunderstood)
I've completely blocked any outbound traffic to 8.8.8.8 and 8.8.4.4 from my network.
DNS yes but IP will be observed and censored eventually.
DoH is not effective against gov censuring that's just propaganda.
So why push DoH so hard? Because it moves DNS control to developers
In July of this year my ISP (Entel Chile) started blocking all ICMP and DNS to non-Entel resolvers. They will not respond to tickets requesting an explanation or confirmation - DNS and ICMP just went dark. In response, I set up DoH on my Pi-Hole with the instructions above.
I don't trust Cloudflare, so a couple weeks ago I started modifying the tools from doh-proxy [2] to run in a container [3]. I have this running in three locations under my control, in networks that I trust (or at least that I trust more than Cloudflare), and in one location I run a proxy that uses Nginx to load-balance requests across the multiple locations.
I made two videos that show how you can set this up, first about the proxy [4] and how to run a stub DNS-to-DoH forwarder in Docker on your local machine if you don't control the network DNS server, and then about how to set up your own private DoH resolvers in Kubernetes (using a $5 VPS and running k3s) [5].
These solutions, while more technical than what my mom can put together, will not only use systems that you control, but depending on how you set it up, will encrypt all DNS traffic from your machine. Encrypting browser traffic is good for the masses, but it only solves a small part of the problem.
Even if you only use the proxy, you can forward to public DoH resolvers that include parental filters, adblocking, and other benefits. If you don't have a Pi-Hole or can't run one, you can get the same benefits with a service like NextDNS [6], who have a generous free tier.
The next step is to configure the resolver to support auth, so that it's not all just hanging out on the Internet like it is at the moment.
For those who are asking about malware and if DoH makes it easier - consider that this is an HTTPS request. All any malware has to do (and already does) is bypass your system's DNS configuration and make the HTTPS request for DNS resolution of C2 systems directly. Whether or not you have DoH active on your system for other requests is irrelevant.
[1]: https://docs.pi-hole.net/guides/dns-over-https/
[2]: https://github.com/facebookexperimental/doh-proxy
[3]: https://hub.docker.com/r/monachus/doh-tools
[4]: https://youtu.be/1RTHCieZqls
[5]: https://youtu.be/Z-_SpFWloQo
[6]: https://nextdns.io
Edit: Link formatting
DNS over HTTPS feels to me like we're edging towards end-to-end encryption for DNS. This seems like a good thing, but even though it will protect against ISP and state level observations of DNS, will it not reduce ones control over DNS locally?
My threat model remains web site operators and the malware and tracking inserted at that point. Additionally devices on the network that may observe locally and communicate externally like TV boxes and games consoles.
To manage those threats I use Pi-Hole. But once we start end-to-end encrypting, and especially when we start verifying those ends, how do I locally intercept and manage DNS for my privacy and security?
Is this a genuine concern about DNS over HTTPS? Or is the plan to enable systems like Pi-Hole to be a trusted and configurable resolver too so that consumers retain the ability to control their own systems like this?