Security mechanisms can be defined in the following ways “something you know”, “something you have” and “something you are”.

Passwords are something you know
PKI cards/tokens are something you have
Biometrics is something you are

The problem I have with biometrics is you can’t change the tokens, and this can be bad for a number of reasons. For example, some new cars come with a biometric reader so they can claim they are harder to steal, but as one proud new owner found out it just makes criminals hurt you more, so now he doesn’t have a car and he has one less finger, that’s right, they stole his car and cut off his finger as well!

My preference lies with something you have, that is PKI hardware, which in most cases also requires a PIN, which is something you know, which adds up to 2 factor authentication. The beauty of this system is that the PIN and the card by themselves are useless, having the card by itself is useless because if you get the PIN wrong 3 times the cards will lock themselves to prevent brute force attacks, and of course the PIN by itself is pointless.

And so begins my epic tale of getting PKI hardware to work with Linux, and the difficulty I encountered highlighting one of the many reasons PKI hasn’t taken off in a big way.

This week I met up with a nice gentlemen, who happened to be the distributor for Gemplus products in Australia/New Zealand, and was kind enough to give me a few of their products for evaluation purposes. I believe others have also managed to get evaluation kit from Aladdin as well, check the main mailing list archive for details on that.

In any case this was my first look at any kind of PKI based hardware, and as per usual for Linux driver support and integration between applications leave a lot to be desired, but the lack of coherent documentation was an even bigger headache.

Read on for more Continue reading →

One FUD issue some people keep regurgitating to keep us from being included in browsers is they worry about us issuing certificates for the likes of paypal.com, most people pushing this line tend to neglect to mention that issuing a certificate on it’s own is mostly useless, unless you can attack the host file on a users computer or the DNS name system, in which case there is bigger problems then falsely issued SSL certificates, especially since most phishing attacks (which is the assumption likely to abuse this) don’t even resort to using SSL.

Currently we require people to have code signing access before issuing IDN/punycode domain/email certificates, and it has been suggested that we have a similar requirement for anyone requesting certificates for high profile sites.

One way to determine popularity is by sites like alexa.com which give out rankings.

I guess the question is how popular must a site be if we want to enforce this, and over what time period?

Another concern is with large organisations as a lot of departments inside these organisations run their own sub-domain and the TLD is handled usually by the main IT department, and this could be cause for concern if someone registers the TLD and starts getting certificates for either the entire organisation or for sub-domains they shouldn’t be allowed to control, this is usually controlled by an organisations IT policy, but this call also lead to someone intercepting traffic by setting up a reverse proxy, and there is questions hanging over this as it will potentially effect legit users one way or another.

Many people have cited the reason for excluding us is based on our perceived ability to protect our root certificate and in fact most consider it worst then a critical browser exploit, but the more I think about this, the more I’m convinced this is just wrong, so I went to the trouble of trying to break the situation down logically, and here’s my risk analysis of the situation:

A browser exploit can effect all users of a particular browser (mozilla says 50mill so I’ll run with estimates based on that).

Browser exploits are pretty clear cut to calculate and would have the potential base of 50 million users to exploit.

A bad certificate on the other hand, the numbers aren’t so clear and you have to do some educated guessing as to what the risk would be closer to.

Without any more specific details of region break downs I’ll have to assume that the 50 million users are evenly distributed more or less on eastern and western Europe, North America, some parts of Central and Latin America and the Asia Pacific regions.

We also have to assume that most banks are either very geographically specific, or at most have a website on a per country basis and they operate different sites in different countries.

To exploit DNS effectively you either have to control a root name server or be able to exploit individual name servers of ISPs in a concurrent fashion. The banking industry and large merchants already pay large sums of money to be notified of DNS based attacks, so the risk here is going to be mitigated some what compared to normal merchant sites, and if we’re talking about normal merchants the threat is considerably lower due to lack of continuous contact that people would have, compared with their banks, and of course replication of the entire shopping cart since you need to make product selection before purchasing.

Ok, so if we evenly distribute the number of firefox copies over 6 areas and assume a penetration rate about equal we end up with about 8 to 10 million users in each location, the above numbers are spread over multiple countries so we’ll assume for the time being that at most, there are approx 3 million users in any given country.

Further to that the potential number of users likely to be effected by a DNS based attack is in the 100’s of 1000’s at most (I’m being generous, more then likely it will be MUCH less) for a banking website used nationally. To attack companies like Amazon.com or ebay.com you’d have to replicate the entire shopping cart system, of which there are easier attacks currently being deployed.

So a browser exploit is likely to effect: 50,000,000
A root certificate breach is likely to effect 100,000 or less, and that’s based on the assumption of a successful DNS breach on a mass scale, where a browser exploit may only need the user to visit a web page.

So the difference between a browser exploit having a detrimental effect or an SSL root cert exploited is somewhere in the vicinity 500x greater, although this easily could be 5000x or more depending on what figures you based your breakdown on, how proactive the bank is preventing other forms of attack so on and so forth.

Just one final note, if the domain is hijacked or even just DNS spoofed you don’t need have a root cert escape into the wild there are plenty of CAs already in the browser root stores that will issue control of domain certificates including Verisign via Thawte 123, Geotrust and Godaddy to name but a few, and this is part of the reason banks employee the services to prevent DNS based attacks, although the real reason is the fact people just don’t take enough care and verify they are connected by SSL before sending sensitive information.

So no matter how the above risk is twisted with FUD, the facts are that an SSL root key loose in the wild is highly over rated due to other factors mitigating risks.