If you don't pay much attention to how TLS works, you should know that NSA (presumably) does not have a magic ability to inject new certs into your root cert repository. If you remove every CA cert from your browser and selectively allow certs, they can't MITM that. The CAs aren't baked into TLS! They're a software configuration detail. And when MITM certs appear on the wire, for them to be honored, they have to somehow chain to a specific CA.
What things like pinning and TACK do is give us the opportunity to discover MITM certificates and start tracing them. If that capability becomes widespread enough, it can potentially foreclose on dragnet TLS MITM attacks, because there will be too much of a risk that deploying a dragnet MITM net will result in the death penalty for the implicated CA.
TACK (and the related efforts) are hugely more important than I think most people think they are. If you want to advocate for something in the wake of the NSA debacle, I think TACK is a great choice.
Unfortunately, ditching the root CAs is way harder than it should be, and flat out impossible in a lot of environments. Compulsory trust isn't trust.
The question I have then is, what happen if look at TACK as an IDS. What is its false positive rate? Can it be lowered? Maybe it should inform the website owner as a way to inform both side of the communication that something eerie is happening.
Let's say that the NSA would like to track bitcoin transactions through MtGox. I don't know how easy it would be for them to plug a backdoor into a server in Japan, and let's assume that the NSA can't break the RC4 crypto their web server is configured to use ..
Since MtGox uses Google Analytics, and possibly pull other scripts from Google's CDN, they could either eavesdrop on whatever data comes back from them by default -- or insist that changes are made to ... pick up more.
Analytics, however, will remain something I'm not overly fond of. For many sites it's unnecessary. For others it's something they could nearly just as easily license and deploy to their own servers. Pulling scripts in from Google Analytics, Statcounter and others -- and especially into privacy concerned apps -- is downright irresponsible.
As I noted here: https://2x.io/read/would-the-nsa-infiltrate-cdns-to-circumve..., even Norway's tax returns site (which hosts info I'd rather not have in any foreign company's hands) use external analytic scripts. They and 90% of the rest of the internet.
No wonder the NSA claim they can circumvent most HTTPS encryption.
I'm curious who build that? Can they not count the filed docs.
As part of my Matasano internship, I've been working on a patch to NSS (Mozilla's crypto library, used by both Firefox and Chrome for TLS) that will give the necessary hooks needed for browsers that use NSS to implement TACK. I'll hopefully be submitting that for review soon (you can see it on bugzilla[1] and fresher in-progress GitHub[2]).
At this point, it mostly makes sense for me to finish that (though I'll have a lot less time when I go back to school, so there might be some work that needs to get done when the review comments come back that I can't get to for awhile, so someone to power that to the finish might be helpful).
But once that's in, we'll need to work on getting patches into the Firefox and Chrome that actually integrate TACK. There's also some other work to do, especially for some other longterm ideas we have (eg. securing communication between mail servers with TACK).
If you're interested in helping, either email Trevor (email is on his website[3]), or email me (danielj@matasano for the next week, personal email in my profile).
1: https://bugzilla.mozilla.org/show_bug.cgi?id=905848
https://en.wikipedia.org/wiki/Jim_Barksdale
http://www.nytimes.com/2013/07/04/us/monitoring-of-snail-mai...
http://www.mozilla.org/security/announce/2013/mfsa2013-53.ht...
http://www.wired.com/threatlevel/2013/09/freedom-hosting-fbi...
https://www.schneier.com/essay-448.html
https://www.schneier.com/blog/archives/2013/06/blowback_from...
"I generally do not connect to web sites from my own machine, aside from a few sites I have some special relationship with. I fetch web pages from other sites by sending mail to a program (see git://git.gnu.org/womb/hacks.git) that fetches them, much like wget, and then mails them back to me. Then I look at them using a web browser, unless it is easy to see the text in the HTML page directly. I usually try lynx first, then a graphical browser if the page needs it.
I also browse from other people's computers, with their permission. Since I don't identify myself to the sites I visit, this browsing can't be connected with me.
One consequence of this method is that most of the surveillance methods used on the Internet can't see me.
Another consequence is that I never pay for anything on the Web. Anything on the net that requires payment, I don't do.
I would not mind paying for a copy of an e-book or music recording on the Internet if I could do so anonymously, and it were ethical in other ways (no DRM or EULA). But that option almost never exists. I keep looking for ways to make it happen."
Also, how does TACK relate to certificate pinning? The introduction has a few differences (e.g. tack expiration), but the two techniques seem pretty similar, am I missing something?
https://lists.riseup.net/www/arc/tack
Regarding certificate pinning: TACK is a different "style" of pinning, it lets you pin to a self-managed signing key instead of certificate keys. So it gives you more control and security than CA pinning, if you're willing to assume more key management responsibilities (storing a TACK private key, and using it to sign new TLS keys).
I think there's room for both approaches: I'd like to see CA pinning be made easy enough that most sites could deploy a fairly safe, simple pin to popular CAs, while high-security sites make use of TACK pins.
But we'll see how things go.
I will read the spec for more information (e.g. what prevents the MITM from becoming my CA?) but it sounds very interesting, thanks again.
The "pin activation" algorithm is intended to make it difficult for a MITM to foist bad pins on clients, and to limit the lifetime of such pins.
It's not a big deal to rebuild nginx/apache but I would think that getting TACK to ship default with OpenSSL would go a long way in getting more adoption.
It will hopefully appear in OpenSSL 1.0.2. We'll be submitting Apache patches for it shortly. Daniel Jackoway (in this thread somewhere) is working on the NSS (client-side) corollary.
TACK is certainly a must-have and I'm really looking forward it being a native part of browsers, but there will still be likely a lot of (high-profile) websites that won't play along :-/
Now that you mentioned it, are those safe from say a government agency having access to the companies that serve as certificate authorities? Or is it all a house of cards, as it is now?
Which is why nothing serious happens to Comodo and Verisign when they get hacked: anybody knowledgeable is already aware that a certificate means very little on its own, and we can't stop the internet anyway, so hey, life goes on.
An attacker who is able to sit in the middle with their own trusted cert is simply going to not bother passing the stapled OCSP response along. The victims browser will do an OCSP lookup (or not...) for the trusted cert the attacker provided, and that will succeed.
I'm not sure how OCSP caching is implemented in the browsers that implement OCSP in the first place (is there a spec for caching?), but as it's a method for checking the revocation status of certs, it can't be cached long.
openssl test reports
OCSP Response Status: successful (0x0)
Cert Status: good
This Update: Sep 8 xx:xx:xx 2013 GMT
Next Update: Sep 10 xx:xx:xx 2013 GMT
Validity
Not Before: Aug 22 xx:xx:xx 2013 GMT
Not After : Oct 2 xx:xx:xx 2013 GMT
So that's quite a long cache. I thought it would fail after the 10th but apparently it is using the CA valid date range?
In the current state of the world, we're all dependent on CA signatures for each connection we make to a website. TACK is a layer of indirection away from CA certificates, such that we'd only be dependent on CA signatures the very first time we contacted a website. It doesn't introduce any new authorities or change the default UX at all.
After the Comodo breaches a few years ago, I put together a talk about these types of attacks, where the fundamental problems lie, and why approaches like DANE are similarly ineffective:
http://youtu.be/8N4sb-SEpcg?t=4m47s