4 ms·
Before I critique it, I should point out for those at home that Hanno's position here is absolutely the conventional position, lots of important smart people wi
by tialaramex 7y ago
Before I critique it, I should point out for those at home that Hanno's position here is absolutely the conventional position, lots of important smart people will tell you Hanno is right about this, not just a few kooks. If it's important to you to have popular opinions about such topics - rather than thinking about it on your own - this is the "correct" opinion to have.
Now, having explained that, I'm sure it's quite wrong after spending many years thinking about this (more than 10 years but probably less than 20) and I'll try to briefly explain why.
The DNS is a key source of truth, and so we need to actually secure the DNS hierarchy. Because of the belief Hanno espouses, there have been a lot of attempts to dodge this work over the years but it's the nature of the problem that dodging it can't work, it just moves the problem and then we're back again from a different angle.
DNSSEC is narrowly tailored to solve only the problem we actually had, securing the source of truth, and nothing else. It isn't a privacy technology, or an anti-censorship technology, or a delicious new flavour of ice cream, it just secures the source of truth.
The thing about security is that the Happy Path looks exactly the same whether you have security or not. You've all been here with HTTP. Remember when most sites you visited either didn't use HTTPS or only used it for a "login" step and then kicked you back to HTTP? How often did a site you visited actually have a breach because it used HTTP? Probably never. The happy path, where nobody was trying to steal your credentials or whatever was nearly always the path you were on anyway, HTTP was dangerous but you didn't really feel it. In hindsight that seems crazy but people put up with it for years.
For DNS we're still in that dark period, and unfortunately lots of people are wrongly confident that it's for the best.
- akerl_ 7y agoI think is a pretty rough mischaracterization to say “lots of people are wrongly confident that it’s for the best [that we stay in that ‘dark period’]”. Speaking as one of those “people”, I’m confident that DNSSEC doesn’t meaningfully improve the security of DNS. The root of trust for DNSSEC is TLD operators, who are in many cases direct subsidiaries of governments and in other cases are effectively controlled by governments, and it uses woefully outdated crypto to provide the chain of trust. And the failure more for when anything goes wrong is that your site drops off the internet at the DNS resolution step, which is pretty gnarly. I’d be happy for a proposal that aimed to improve the threat model for network connections, but DNSSEC does not do that.
- tialaramex 7y ago> Speaking as one of those “people”, I’m confident that DNSSEC doesn’t meaningfully improve the security of DNS Yes, as I said, you are indeed confident you're right. That won't help by the way, but I don't expect that to have any impact on you because you have confidence and facts are a poor substitute. It does actually matter which government. My sister's web site undoubtedly could be interfered with by the government of the country she lives in as it controls the registry her domain name is in, but if they have got it in for her then she has bigger problems than a web site. On the other hand, the status quo today without DNSSEC is that another government she's barely aware exists has the resources to intercept her traffic, because there is no cryptographic obstacle to them doing so at the DNS level. Not to mention plenty of non-government entities, we're hardly talking about Apollo programme levels of resources. Now, in truth today what you see a lot of is lower hanging fruit. Back when I worked for the last startup we spent a bunch of time looking at cases where DNS for sensitive domain mail servers gets changed and then for a few hours or days all the email to some domain is going via a different country. Sometimes the hijackers would get themselves a TLS certificate, sometimes they wouldn't bother (presumably the targets didn't do name checks or didn't make TLS mandatory). DNSSEC wouldn't help here because the bad guys literally just broke into a registrar DNS account which was poorly secured, their new answers would have been signed. But of course even though bad guys won't pick your locks when there's a fire door left open that's not a reason to use crappier locks, it's a reason to shut the fire door. The existence of lower hanging fruit is not a defence, if anything it means you're likely to underestimate the risks you face and be unprepared. The fail closed behaviour is desirable. Soft-fail is, to use Adam Langley's preferred metaphor, "a seat-belt that snaps when you crash". If there's a way to have a config that isn't secure but seems to work, that's exactly the config bad guys will rely on to attack you. This was an understandable mistake when Netscape made it in the 1990s, less understandable when Microsoft did it at the turn of the century, and unforgivable today.
- akerl_ 7y agoIt’s difficult to have a discussion when in both your comments here, you’ve proposed that people who disagree with you aren’t discussing in good faith (in the first case by attempting to cast it as the politically correct opinion, and in the latter case by suggesting I’m not interested in facts). The status quo today is that DNS interference is being mitigated by the CA system, which is no hero but does have the benefit that rogue CAs (or CAs which have been forced to behave badly by other actors) can be delisted. The push to increase TLS deployment, detect TLS cert shenanigans, and move additional technologies behind TLS (DNS-over-HTTPS, for example) all work to move the needle on the risks in the current threat model for the vast majority of internet-based workflows. These changes are what’s actually making things better for your example of your sister’s website. Regarding fail-closed, I didn’t say that security controls should fail open. I said that unlike other security controls, DNSSEC fails by dropping the site off the internet. As an end user, you don’t even know it’s a DNSSEC problem, you just know the site has disappeared.
- tptacek 7y agoHere I think it's worth mentioning that Hanno is not, as you're sort of implying, some Internet rando, but rather one of the more important Internet security researchers working today.
- tialaramex 7y agoIs that what's implied? I certainly didn't intend it, and I hope Hanno knows that. We've never quite met in person but our circles cross here and there when it comes to the Web PKI. Anyway, Hanno writes mostly in German and for German language publications but https://www.feistyduck.com/bulletproof-tls-newsletter/ https://www.feistyduck.com/bulletproof-tls-newsletter/ is in English and is worth a read. Perhaps the most important thing about Hanno in this context is his independence, he's not making a buck from any of the players as far as I know. I just happen to think he (and lots of other important Internet security professionals) is wrong about DNSSEC.
- tptacek 7y agoHanno has presented in the US multiple times, most memorably at Black Hat, when he and Sean Devlin used a GCM bug to host their talk slides on GCHQ's website. He's also a published researcher (for instance, he's one of the ROBOT authors). And virtually none of the people who disagree with you about DNSSEC, and I agree with you that there are many of them, are "making a buck from any of the players" in the debate; if that's something you were trying to imply about me, again, you're wrong.