Wandering Thoughts archives

2020-03-13

Sensible heuristics for when to use DNS over HTTPS can't work for us

If Firefox is using DNS over HTTPS in general, it has various heuristics for whether or not to use it to resolve any particular name; for instance, right now it doesn't use DNS over HTTPS for any domain in your DNS suffixes (this happens even if you explicitly turned on DNS over HTTPS, which disables checking for the canary domain). Presumably other browsers will also have their own set of heuristics when they implement DNS over HTTPS, and at some point the set of heuristics that various browsers use may even be documented.

(This DNS over HTTPS bypass is intended to deal with two cases; where the name you're looking up doesn't exist in public DNS, and where the name has a different IP address.)

Almost six months ago I wrote a cautiously optimistic entry about Firefox, DNS over HTTPS, and us, where I hoped that Firefox's heuristics would be able to deal with our split horizon DNS setup where some names resolve to different IPs internally than they do externally. Unfortunately, I've changed my mind (based on experience and experimentation since then); I now believe that no sensible set of heuristics can cover all of our cases, and so anyone using DNS over HTTPS (with external resolvers) will sooner or later be unable to connect to some of the websites run by people in the department.

The fundamental issue that sinks the entire thing here is that people sometimes want to host websites on their machines here but give them names not under our university domains (for various good reasons, such a regular yearly conference that just happens to be hosted here this time). We do know what these domains are, because we have to set up the split DNS for them, but it's definitely not appropriate to add them to our DNS suffixes and they have different public DNS servers than our internal resolvers.

(In some cases they have public DNS servers that aren't even associated with us, and we carefully shadow bits of their domain internally to make it work. We prefer to host their DNS, though.)

I can't think of any sensible heuristic that could detect this situation, especially if you don't want to leak information about the person's DNS lookups to the local resolver. You could detect connection failures and try a non DNS over HTTPS name lookup, but that leaks data in various circumstances and even if it works there's a long connection delay the first time around.

So I think we're going to always have the 'disable DNS over HTTPS' canary domain signal in our local DNS servers, and we'll hope that someday this signal is respected even for Firefox users who have explicitly turned on DNS over HTTPS (unless they explicitly picked a new option of 'always use DoH even if the local network signals that it shouldn't be used and might give you broken results'). This makes me a little bit sad; I'd like heuristics that magically worked, so we could let people use DNS over HTTPS and hide things from us that we don't want to know anyway.

DNSOverHTTPSHeuristicsAndUs written at 01:22:57; Add Comment


Page tools: See As Normal.
Search:
Login: Password:
Atom Syndication: Recent Pages, Recent Comments.

This dinky wiki is brought to you by the Insane Hackers Guild, Python sub-branch.