There has been a debate ensuing on Debian Planet since last week about Firefox 3’s new behavior for what it views as invalid SSL certificates. Having upgraded to Ubuntu 8.04 back in February, I’ve been using Firefox 3 since it hit rc1, so I can definitely relate to the problems that people are having. I completely agree with the sentiment of those who view the new behavior as a necessary evil. Unsigned SSL Certificates are a potentially huge security risk. Unfortunately, they’re common as spit and most people just click right past them because they’re getting in the way of the user doing what they want.
Firefox’s new approach is pretty heavy handed. So much so, in fact, that it appears you can’t work around it without some non-trivial changes to Gekco. This probably wouldn’t be so bad, except that most users have absolutely no idea what to do when confronted with this:
I know that my wife didn’t when the wireless network of the hotel we stayed at following our wedding redirected us to a site with an Invalid SSL Certificate. Hell, it threw me for a loop the first time I saw it. Other people have, of course, reported similar experiences.
In reality, I blame the insane cost of SSL Certificates. Partially, this is due to the standard for SSL security in web browsers is an all-or-nothing deal. You’re either signed by a Certificate Authority (CA) in the browsers certificate file, or you’re not. Because of this, CA’s have no incentive to change the way that they offer Certificates, you pay through the nose for a ‘valid’ one, or your don’t and use a self-signed ‘invalid’ one. The absolute cheapest you can get a Web-enabled certificate from Thawte, is $150/year, and in that case they only identify the domain, not the user. Want your company identified for better security? That’ll be an extra $100/year. Not that most users will notice. Want the fancy Green Address bar (at least in newer browsers)? Be prepared to spend a whopping $800/year.
Actually, I fully support this sort of pricing model (though I think that $150/year for a domain-only SSL certificate is ridiculous), but we need better mechanisms to communicate how much the key should be trusted. The Extended Validation Certificate (EV) is a huge step forward in this, but it’s still not very fine-grained, especially when many sites who need, or require like Microsoft Office SharePoint Server, encryption simply can’t justify that sort of expenditure for a signed SSL certificate.
Admittedly, organizations can create their own CA’s for internal use, and sign certificates all they want. This becomes impractical at some point, however, because you need to make sure that every user in your organization has the CA certificate installed. Washington State University has a CA certificate, that I suspect is installed in almost every departmental computer on campus, but most organizations simply don’t use it. This is likely due, in part, to the number of off-campus users, and the freedom which we provide users to bring their own hardware. My Eee PC spends quite a bit of time on the WSU network, but I don’t have the WSU CA certificate. Still, I would prefer that a lot of these self-signed sites were using the WSU certificate, as then I could install that cert and have them just work. As it stands, I have no reason to really even consider that course of action.
What we really need, is for the web to be tied into a true Web of Trust. I choose the Root CAs I want to honor, but signing their key with my own, and I can assign trust to other user’s signatures, so that I can opt to trust them simply because someone I trust trusts them. Since most Trust applications allow you to specify differing levels of trust, this is practically built into the encryption scheme. And I can explicitly set my trust on the Firefox key, so that I accept keys that Firefox trusts, and amazingly, my situation doesn’t really change much.
Of course, the above paragraph is a pipe-dream. The majority of encryption software is too difficult for the average user to use, and most users simply don’t care to learn. But as I’m a huge advocate for large-scale public-key encryption, I’m going to keep dreaming. In the meantime, we need a trusted Root CA who sells discounted certificates so that non-commercial entities who want (or need, which isn’t always the same thing), can have valid one’s without inconveniencing their users significantly.
There is the other side of this, that perhaps Firefox is trying to annoy users, to force web developers to do what they feel is right. Microsoft did the same thing with the UAC in Vista, after all. However, if this is the case, Mozilla has made an enormous mistake. For Windows Vista, redesigning the application just a little bit, can get rid of those annoying UAC boxes, and actually result in a net-increase in application security. Requiring signed certificates makes the web more secure, without a doubt, but the cost involved for many organizations seems prohibitive, especially for Open Source projects that feel that they’re doing their users a favor by encrypting logins to web-based systems.
I’m glad the Mozilla is trying to do something, but I agree with those who feel that they’ve gone too far. I’d be happy if, on the first alert screen, there was a button that allowed me to trivially accept the key on a temporary basis, while still requiring the full process to add the key permanently. And Ideally, I wouldn’t have to click on the “Or you can add an exception…” link to see the actualy options.