Archive for January 2nd, 2016

Is no crypto always better than bad crypto?

SSL (Secure Sockets Layer, the code that forms the basis of the https:// in a URL) can use any number of different encryption methods (protocols) and key strengths. While all of the protocols / strengths were presumed to be secure at the time they were designed, faster computers have made “cracking” some of the older protocols practical, or at least potentially practical. Additionally, concerns have been raised that some of the underlying math may have been intentionally weakened by the proponents (for example, NIST and the NSA) of those protocols. Perhaps an underlying flaw in the protocol has been discovered. Due to this, web browsers have been removing support for these older, insecure protocols.

Additionally, even if a protocol is still considered secure, a browser may start enforcing additional requirements for the SSL certificate used with that protocol. “Under the covers” this is a rather different situation, but for the purpose of this discussion I will lump them together, since the average user doesn’t care about the technical differences, only that a service that they used to be able to access no longer works.

In theory, this is a good idea – nobody wants their financial details “sniffed” on the way between you and your bank. However, the browser authors have decided that all usage of those older protocols is bad and should be prohibited. They make no distinction between a conversation between you and your bank vs. a conversation between you and another site (which could be a web server, UPS – battery backup, a water heater, or even a light bulb!) in your house or company. Instead, they force you to disable all encryption and communicate “in the clear”.

To add to the complexity, each browser does things in a different way. And the way a given browser handles a particular situation can change depending on the version of the browser. That isn’t too bad for Internet Explorer, which doesn’t change that often. Two other browsers that I use (Mozilla Firefox and Google Chrome) seem to release new versions almost weekly. In addition, the behavior of a browser may change depending on what operating system it is running under. Browsers also behave differently depending on when the host at the other end of the connection obtained its security certificate. A certificate issued on December 31st, 2015 at 23:59:59 is treated differently than one issued one second later on January 1st, 2016 at 00:00:00.

In the following discussion, the terms “site” and “device” are generally interchangeable. I sometimes use the term “device” to refer to the system the browser is attempting to connect to. “Site” might be a more accurate term, but for many users a “site” implies a sophisticated system such as an online store, while an intelligent light bulb is more a “device” than a “site”.

In a perfect world, people could just deal with the browser blocking issue by installing new software and / or certificates on all of the devices they administer. Sure, that would be a lot of work (here at home, I have several dozen devices with SSL certificates and in my day job, I have many hundreds of devices) and possibly expense (the companies that sell the certificates don’t always allow users to request updated certificates for free, and updated software to handle the new protocol may not be free – for example, Cisco requires a paid support contract to download updated software). However, it is not that “easy” – any given device may not have new software available, or the new software still doesn’t handle some of the latest protocols.

This leads to an unfortunate game of “whack-a-mole”, where a browser will change its behavior, a company will implement new software to deal with that new behavior, but by the time the software has gone through testing and is released, the browser has changed its behavior again and the updated software is useless. A number of vendors have just given up supporting their older products because of this – they have finite resources and they choose to allocate them to new products.

The browser authors seem to feel that this is just fine and that users should either turn encryption off or throw away the device and buy a new one. Since the “device” is often a management function embedded in an expensive piece of hardware, that simply isn’t practical. A home user may not feel that replacing a working device is necessary and a business likely won’t replace a device until the end of its depreciation cycle (often 3 or 5 years).

This strikes me as a very poor way for browsers to deal with the situation. Instead of a binary good / bad decision which the user cannot override, it seems to me that a more nuanced approach would be beneficial. If browsers allowed continued usage of these “obsolete” protocols in certain limited cases, I think the situation would be better.

First, I agree with the current browser behavior when dealing with “Extended Validation” sites. These are sites that display a (usually) green indication with the verified company name in the browser’s address bar. In order to purchase an EV certificate, the site needs to prove that they are who they say they are. For example, your bank almost certainly uses an EV certificate. Users should expect that sites with EV certificates are using secure methods to protect connections. If a site with an EV certificate is using an obsolete protocol, something is definitely wrong at that site and the connection should not be allowed.

Second, the current behavior is OK when dealing with well-known sites (for example, amazon.com). This is a little more difficult for browsers to deal with, as they would need to keep a list of sites as well as deciding on criteria for including a site on that list. However, there already is a “master list” of sites which is shared between various browsers – it is called the HSTS Preload list. It could be used for this purpose.

Now we get to the heart of the matter – how to deal with non-EV, non-well-known sites. Instead of refusing to allow access to a site which uses an insecure protocol, a browser could:

  • Display a warning box the first time a site is accessed via an insecure protocol and let the user choose whether or not to proceed.
  • Re-display the warning after a reasonable period of time (perhaps 30 days) and ask the user to re-confirm that they want to use the insecure protocol to access the site.
  • On each page, indicate that the page is using an insecure protocol. This could be done by displaying the URL in the address bar on a red background or similar. Google Chrome does something similar with its red strikethrough on top of the https:// in the address bar. Unfortunately, in most cases Chrome will simply refuse to access a site it deems insecure.
  • NOT require dismissing a warning each time the user accesses the site.
  • NOT require a non-standard way of specifying the site URL in the address bar, bookmarks, etc.

Security experts will probably be thinking “But… That’s insecure!” It certainly is, but is it less secure than using no encryption at all (which is what the browsers are currently forcing users to do)? I don’t think so. In many cases, both the user and the site they are connecting to are on the same network, perhaps isolated from the larger Internet. For example, most devices here are only accessible from the local network – they are firewalled from the outside world.

Technical note: I am only talking about insecure protocols in this post. There is a different issue of bugs (problems) in some particular implementation of SSL – for example, OpenSSL. However, those problems can usually be fixed on the server side by updating to a newer SSL implementation version and generally do not remove protocols as part of fixing the bug. My post is focused on servers that are too old and / or cannot be updated for some reason, which is a completely different issue from server implementation bugs.

What do you think? I’d like to see comments from end users and security experts – feel free to try to shoot holes in my argument. I’d love to see comments from browser authors, too.