HTTPS is broken and Firefox 3 makes it worse

[Note: this is mainly a long rant. If you’re just curious about my possible solutions, scroll to the last few paragraphs.]

I’ve been using the Firefox 3 betas (and now release candidates) for quite a few months now, and overall, I think it’s a great improvement over Firefox 2. My main issues in the past with FF2 have always been with performance and memory usage, and FF3 seems to go very far in addressing both of these (though there’s always still room to improve).

However, there’s one thing I just can’t get over: the SSL error pages. Before I describe FF3’s added annoyance here, let me start by saying one thing: the HTTPS model is broken and much less useful than it could be.

Why?

Because the HTTPS/SSL model tries to simultaneously combine two things: encryption and authentication. When you visit a HTTPS site, you get an end-to-end encrypted connection between your web browser and the web server, obviously. But you also get something else, in theory: assurance that the web server on the other end is actually owned and operated by who it says it is. This is done through a hierarchical certificate signing scheme, wherein the browser knows a few “trusted” root certificate authorities (CAs), and any web server SSL certificate that wants to be considered “trusted” (as far as the browser is concerned) needs to ultimately be signed by one of those root CAs. (In practice, this doesn’t mean each and every web server SSL cert needs to be signed by a root CA, just that every trusted SSL cert needs to be signed by another signing certificate that was signed by a root CA, or by another signing cert that is signed by a root CA somewhere in its ancestry.)

So where’s the flaw? Getting your SSL cert signed by a trusted cert authority costs money. This is fine for banks, but not so fine for some random guy who just wants secure communications with his home server, or webmail, or just a small self-funded website where an encrypted connection makes sense. A semi-solution is self-signed certificates. You can create your own SSL certificate, and your web server will use it, and people who connect to your web server on the HTTPS port will get encryption. But, as far as the browser is concerned, your cert isn’t signed by anyone on its trusted list.

So, the website visitor is usually presented with a confusing dialog box warning that the website they’re visiting shouldn’t be trusted. Visitors who understand all this stuff just roll their eyes, click the “allow” button and go on with their day. People who don’t really understand all this, but just want to visit the site, blindly click “allow” and go on with their day. Some panicky users may freak out, click “cancel,” and not visit the site.

The people in the middle there are the ones in danger. Blindly accepting dialog boxes just to get them out of the way can get you in trouble if the dialog box is actually warning you of something important. In the context of visiting Gentoo’s HTTPS bug tracker website, the warning actually is unimportant: Gentoo is just using a self-signed cert to avoid paying for a “real” cert, and just wants HTTPS for the encryption, not for the authentication. However, in the context of Bank of America’s online banking service (just to pick one), the warning is very important: it likely means that someone has intercepted your communications and is trying to masquerade as BoA’s server so they can steal your bank account credentials. Or it could also mean you typed the URL incorrectly, and someone malicious has registered the typo-hostname in the hopes of snaring someone with careless fingers.

So there’s a problem, definitely. The Firefox 3 developers have chosen to attack this problem by attempting to push people into the panicky-user camp, though they misguidedly believe that they’re actually getting users to try to read the messages and make an “informed” decision (despite the fact that they haven’t given – and really can’t give – enough information to inform most users). The new SSL error pages in Firefox 3 put up a scary, confusing, uninformative message right inside the browser window. The message, at first glance, looks reminiscent of one of the several “connection error” messages, so the first reaction is to look up at the address bar to see if you typed the address correctly, and then wonder why the site is down. Then you go and read the error message. Basically, it says a lot of confusing things – including a semi-raw error enumeration code that is sure to confuse the user even more – with a very short explanation of the real problem: “The certificate is not trusted because it is self signed.” What are the chances that the average user will know what this means? Slim to none.

Then there’s some text about adding an exception, and instead of a button, a URL link (in a small font) that you can click to add an exception. Well, sorta. Clicking the link merely modifies the error page to display another semi-scary message, and then shows two buttons: one with a stupid caption: “Get me out of here!” (don’t get me started on this one), and the other saying “Add Exception…” Ok, fine, so I click “Add Exception…” Now I get a dialog box, and my eyes automatically seek the bottom of the box, assuming the “confirm” button will be there, waiting for me to click. But it’s not. It’s there, but it’s disabled. Huh? So I look up higher, and I see that only two buttons can be clicked: a “Cancel” button (well, duh, obviously that’s not it), and a “Get Certificate” button. Ok, well, I guess I’ll try that one. Now the text in the middle of the box changes to tell me “Certificate is not trusted, because it hasn’t been verified by a recognized authority.” What? I know what all this stuff means, and I have to read it twice to get it. Your average user doesn’t stand a chance. Finally, though, the “confirm” button is active, and they’ve helpfully (wow, they actually did something remotely helpful here!) pre-checked the “Permanently store this exception” check box for me.

So, in total, I have to make four clicks, plus read a bunch of confusing terminology, to get to a website with a self-signed SSL certificate.

This sucks. The UI is absolutely terrible, and the average web user is going to have no idea what’s going on. According to the Firefox developers, one of the goals of these new error pages is to cut down on “dialog box whack-a-mole” where users just blindly click to get the dialog out of the way. I fail to see how this is going to help. Some users will now get even more confused, and not visit the site. For the extremely rare case where a malicious site is masquerading as the site they actually want, this is good. But for the much more common case of an innocent site that just wants SSL for the encryption, this is bad. And for most of the users in the “blindly dismiss” bunch, they’ll just get used to blindly dismissing this new page+dialog in record time.

What’s the real solution? Ideally: scrap the current system. There’s no reason why connection encryption needs to be so tightly coupled with authentication/identification. Design a new system, possibly with a new protocol scheme. One scheme should be used for “strong” security, where both the identity of the site and cryptographic strength of the connection are checked. The other scheme will just check the encryption. Users will need to be educated that you never try to connect to your banking site using the less-secure scheme. Sure, user education is always a problem, but it’s a problem we still face with the current non-solutions in place.

Of course, in the real world, we can’t just scrap a protocol that has been in use for over a decade, and expect everyone (web servers, web browsers, the cert-signing industry) to change overnight – or at all.

So I’d advocate dropping all the panic about self-signed certs and, in addition, handling HTTPS sites by keeping more state about them in the browser. For starters, at the simplest level, let’s stop putting up scary messages when we hit a self-signed site. We’re already coloring the address bar differently for HTTPS sites; why not have a different (slightly scary?) color for HTTPS sites that use self-signed certs? There could even be an easily-disableable bubble popup (that doesn’t take focus!) that points out the “problem” and is clickable for more information. Or something like that. At minimum, if you visit your bank site all the time and don’t see this warning and color change, you’ll think twice the time you do.

But to make this stronger, the browser can keep state about the HTTPS sites we’ve visited. Odds are, the first time you visit your banking site, it’ll probably be all correct and proper. So the browser notes the cert’s fingerprint, and the fact that it’s signed by a “trusted” authority, and checks it every time you visit the site. If the cert changes, and is now self-signed, the browser can raise a larger red flag: “Hey, this site that you visit all the time that usually has a trusted cert? Well, the cert is now self-signed and this might be someone trying to trick you.”

What about “phishing” attempts that use common misspellings? Well, if the browser knows that I’ve visited https://bankofamerica.com/ in the past, and it had a valid, trusted cert, and now I’ve visited https://bankofamarica.com/, and it not only has a self-signed cert, but has remarkably similar spelling to another site I visit that has a trusted cert, the browser can raise another big red flag. Finding similar spellings is nothing new: there are algorithms to do this that are old and very well-established.

How about phishing attempts that hide the true website URL by using an inline username/password string that looks familiar to the user? Well, we already cover this: Firefox pops up a dialog asking if you really want to connect to the site, presenting the real hostname, and showing the username passed in the URL. So, this new scheme doesn’t harm this case’s solution.

End rant.