HSTS for new TLDs | Hacker News Reader

HSTS for new TLDs

imperialviolet.org

85 pointstomkwok42 comments

This seems like a great idea for the new .trust gTLD [1] (formerly known as .secure), as they already want sites under it to be "secure".

[1] https://www.nccgroupdomainservices.com/

I wish there was an alternative to the CA system we have now. I'm part of a private forum with a self-signed cert and while initially there was some hesitation to it, most members have added it to their browser exceptions.

There is a proposed solution for this scenario: https://en.wikipedia.org/wiki/DNS-based_Authentication_of_Na...

It will allow for pinning certificates (self-signed too) to records in in DNS with DNSSEC.

That's interesting. I hope it sees more widespread adoption by browser vendors and the recent security disclosures will hopefully help to tip implementation in that direction.

If you're using a resolver which supports DNSSEC, and you're using Firefox with the DNSSEC-Validator addon from https://www.dnssec-validator.cz/ (which supports DANE) and you visit https://grepular.com/, you will see a nice little green icon in the address bar to show you that DNSSEC was used, and another green icon to show you that the SSL cert was validated using DANE.

DNSSEC still relays on CAs. They were just renamed to "Trust Anchors". The trust anchor for .com is Verisign.

> DNSSEC still relays on CAs.

No, DNSSEC has nothing to do with CAs. Each DNS authority defines its own keys used to sign its records.

> They were just renamed to "Trust Anchors".

You are thinking about DANE [1], which is what the a protocol on top of DNSSEC. Using DANE you authorize X.509 certificates and/or CAs for certain domains. This allows you to restrict your domain to a specific well-known CA (such as Verisign), but it also allows you to authorize your own CA or even a specific certificate directly. It even works per-service, so you do not need to use the same CA/certificates for all your services.

If you were suggesting that DANE does not solve the traditional CA issue you are wrong.

[1] http://tools.ietf.org/html/rfc6698

> The trust anchor for .com is Verisign.

Err.. no? I don't even understand what are you trying to say here. The "com" domain does not seem to have any TLSA records...

>> The trust anchor for .com is Verisign.

>Err.. no? I don't even understand what are you trying to say here. The "com" domain does not seem to have any TLSA records...

Verisign runs the "com." zone, so I guess they would have to admin any DNSSEC/DANE/TLSA stuff.

They only have to admin the DNSSEC stuff (publish your domain name keys along with your nameservers, as you set them up through your regular domain provider).

The DANE part needs no further support than you being able to use DNSSEC for your domain's DNS. Obviously there is always some entity controling each TLD, and that entity can screw up your domain if it acts improperly.

That last property is what makes that entity a "trust anchor", and what DANE critics specifically don't like or want to get rid of.

You're not following. DNSSEC is secured by a chain of keys leading to a root; the entity that controls the root controls the chain. If you use DNSSEC to authenticate your certificate, you're giving control over your certificate to whoever runs the roots.

So you will have to trust Verisign not publish a different, compromised, set of domain name keys along with your nameservers.

You have to trust them not to do that anyway right now, as they're a CA. You also have to trust any of the >100 other CAs that your system trusts.

Nobody is suggesting that the naive X.509 scenario we had a year ago is secure. They are, however, skeptical of the idea that we should invest millions of dollars into an architectural change to the core of the Internet to land... right back where we are now.

> No, DNSSEC has nothing to do with CAs

You're interpreting "CA" too literally. DNSSEC doesn't rely on X509 certificate authorities but in effect it relies on an equivalent, in that Verisign is a central authority certifying ownership of all .com domains.

A single trusted root is better than the hundreds of roots in the CA system. I don't trust Verisign very far, but I certainly trust them more than Verisign AND Diginotar AND honest achmed's used cars and certificates AND ...

And shouldn't it be possible to implement certificate pinning for whole tlds? Then we'd only have to trust root for unknown tlds.

First, you don't get "a single trusted root". If DNSSEC/DANE had been deployed 3 years ago, Qhadaffi's Libyan government would have effectively been the CA for BIT.LY.

Second, and more importantly: the smaller number of trust anchors you end up with in DNSSEC are controlled by world governments.

It seems absurd to me that the Internet's response to the "global passive adversary" of NSA would be to hand the entire PKI system over to the USG formally. That's what DNSSEC does.

> ...Qhadaffi's Libyan government would have effectively been the CA for BIT.LY.

They already are, in practice. Many reputable CAs will issue certificates to anyone who can forge an MX record for a domain. With or without DNSSEC, TLD operators are capable of forging those records.

>It seems absurd to me that the Internet's response to the "global passive adversary" of NSA would be to hand the entire PKI system over to the USG formally. That's what DNSSEC does.

No, DNSSEC builds authentication into a system that is and has always been centrally controlled. And just like with the X.509 CA system, you can use pinning or Convergence or anything else you want to supplement that.

You shouldn't start a sentence with "no" when it confirms the sentence it's responding to.

> No, DNSSEC has nothing to do with CAs. Each DNS authority defines its own keys used to sign its records.

which in turn must be signed by the zone operator (e.g.: Verisign for .com) who publishes them in DNS. So we still have Central Authorities - in the sense that there is still some overlord controlling everything.

Moxie Marlinspike created an alternative to the CA system called Convergence[1]. His DEFCON video[2] does a good job describing the problem and his proposed solution. There's a Firefox plugin for it, but Chrome has come out and said they will likely never support it[3].

[1] http://convergence.io/

[2] https://www.youtube.com/watch?v=pDmj_xe7EIQ (you can skip the first 5 minutes; if you want to jump straight to his proposed solution, it's around 35 minutes in)

[3] http://www.theregister.co.uk/2011/09/08/google_chrome_reject...

There are alternatives - as long as you're willing to give up on regular DNS domains and use .dns or .bit domains:

https://github.com/okTurtles/dnschain

http://www.freespeechme.org/

The university I work for uses Self-signed certs for everything but ssl. Apparently setting up properly signed certs is too expensive. And it's more effective to have the IT staff spend an inordinate amount of time convincing end-users that the warnings about invalid certs from various applications are nothing to worry about (I'm looking at you Andoid and OSX).

Are you being sarcastic or do you really think they are nothing to worry about?

If people just accept any cert they see anyone could just mitm them with a self signed cert.

I'm being sarcastic. Also some departments use different self-signed certs than the university at large. Not a great situation.

Couldn't they distribute the signatures to the faculty and students in some semi-secure way?

I'm surprised no one here is talking about Namecoin [1]. It has become obvious that the PKI system is failing, and the main issue is centralization.

I don't know if Namecoin itself will be the solution, or some other blockchain-based method that achieves decentralization such as the more generic Ethereum [2], but as engineers we should be supportive of such systems, as they provide better security and true -not just delegated- domain name ownership.

[1] http://namecoin.info/

[2] https://www.ethereum.org/

>It has become obvious that the PKI system is failing, and the main issue is centralization.

Actually, the problem is exactly the opposite. Any CA can sign a certificate for any domain.

What compounds the problem with the CA system is that CA trust is sticky. When a new CA gets added to trust stores, they are effectively trusted for life. That's not how trust actually works.

What we need is a system that is distributed and that allows for trust to be granted and - most importantly - revoked at will without causing collateral damage (i.e. for innocent sites that end up using a bad CA).

Moxie Marlinspike's convergence[0] was a great idea, but it's never really taken off. TACK[1], also by Marlinspike, looks like nice feature to help transition away from the CA system when we come up with something better.

[0] http://convergence.io/ [1] http://tack.io/

Good idea, especially for some tlds aimed at industries which rely on strong security like .bank. Will send the link to a couple of ngtld applicants.

I'm confused on the implementation of something like this. HSTS is an HTTP header, so it is up to the HTTP server, not the DNS server. How is HSTS enforced at the domain layer?

Chrome and Firefox both have a preloaded list of sites that should always use HSTS. https://blog.mozilla.org/security/2012/11/01/preloading-hsts...

Right but site owners need to request inclusion on that list only after they've enabled HSTS on the server. I've done this myself and I must say I'd be pretty angry if browsers forced an entire TLD to require HSTS. The result would be SSL warnings everywhere.

That's why he's suggesting that the new TLDs do this, to start early, so that everyone's expectations are set properly.

A preload does not require the header to be set. It would obviously be the smart thing to do, but it's not required.

HSTS is only a header.

A preload does not require the header to be set.

did you read evilpie's link?

"Only if a host responds with a valid HSTS header with an appropriately large max-age value (currently greater than or equal to 10886400, which is eighteen weeks) do we include it in our list. ... We limit the list to hosts that send a large max-age under the assumption that these sites will not revert to non-HSTS status."

but you obviously can't check every domain in the TLD. A preload will still function, that's the whole point: make sure a MITM can't cut off the HSTS header.

"I can't speak for Firefox and Safari but I think it's safe to assume that Firefox"

Maybe one of these Firefoxes should say Chrome

It is a bit confusing, but I'm not sure that this is actually an error. It sounds like he may be in a position to directly make that change to Chrome, but is trying to clarify that his thoughts on Firefox and Safari are his, and he can't speak on their behalf. When you read it that way, it makes sense.

What's more, Firefox previously adopted a related thing that agl invented (HSTS preloads in general), so he has personal experience of their being amenable to this sort of thing.

The Chrome SSL team works very closely with that of Firefox.

not to mention Firefox pretty much uses the HSTS preload list that Chrome does.

Reply on news.ycombinator.com