A rack middleware to prevent timing attacks.
In a timing attack an attacker times how long a query takes to discover sensitive information about the application. In the past, timing attacks have been known to reveal,
- usernames
- credentials (or hashes thereof)
- cryptographic keys
- session cookies (e.g. CVE-2013-0263)
- mapping authorization or firewall rules
- etc.
An authoratative description of timing attacks on web applications may be found in "Exposing Private Information by Timing Web Applications" by Bortz, Boneh and Nandy. See https://crypto.stanford.edu/~dabo/papers/webtiming.pdf.
This gem implements a defense described in the paper mentioned above. It delays requests so that they take a multiple of 100ms to complete. The value of 100ms is configurable; increasing it causes the site to respond slower, while decreasing it makes the application more vulnerable. This parameter must be tuned for each application.
(Note that random delays have been shown not to prevent these attacks, which is why they aren't implemented here)
-
Update your gemfile. For now, use
gem 'rack-timesec', :git => 'git://github.com/bchurchill/rack-timesec.git' -
Add the middleware in the application.rb file. The location in the middleware stack is important. You want TimeSec to run before any sensitive operations, such as working with the session cookie, or authentication/authorization. For security, it's best not to use Rack::Runtime anywhere, because this would help a timing attacker. However, there are some dependencies on Rack::Runtime in parts of rails. If you use Rack::Runtime anyway, then Rack::TimeSec must follow it, or attackers will be given timing information for free (the protection will be useless, but it will apear to work superficially). In general, it's best to setup Rack::TimeSec to run sooner rather than later. Therefore, my recommendation is to use the following in application.rb, which places Rack::TimeSec immediately after Rack::Runtime.
config.middleware.insert_after Rack::Runtime, Rack::TimeSec -
Configure (described below)
-
That's it! AFAIK, there are no dependencies. In principle this should work with just about any rack application, although it's only been tested with rails 4. It should also work with older versions of ruby, but this hasn't been tested either.
Rails::TimeSec will work right out of the box. There are two configurable options, both of which are important:
The first is the 'interval' setting. This defines the maximum interval (in seconds) that TimeSec will delay a request. Setting this value higher often provides better security, while setting it lower provides better performance. The default setting is 0.1 seconds (100ms). This seems to work well for many sites. To set this value to 0.2 seconds (200ms), you would use:
config.middleware.insert_after Rack::Runtime, Rack::TimeSec, :interval => 0.2
To test, you should do some sensitive operations that take different amounts of time on the server and ensure they look the same to a client. For example, try logging in with a valid user and an invalid password, and then with an invalid user and an invalid password. If one of these operations takes less time than the other (e.g. one takes 100ms while the other takes 200ms), then it's necessary to adjust your interval.
You should also perform this test when the server is at minimum load, typical load, and maximum load. The load on the server affects the response times, and so your interval should work in all cases. Additionally, it may be necessary to use different values in different environments. For example, in moving from development to production the database backend might have a very different latency.
Note that a greater interval doesn't always provide better security. For example, if you have a sensitive operation that takes 120ms or 170ms on the server, then using an interval of 100 would entirely mask this difference. However, an interval of 150 would cause the client to see these operations taking 150 or 300ms (respectively), and this could be a vulnerability.
If you're very paranoid, you may set the interval so large that all requests will complete in the same interval. An example would be setting it to one second.
For performance reasons, especially in a development environment where assets are served statically, we want to exclude certain URLs which are not sensitive to timing attacks. For example, one can exclude all paths starting with /assets/ as follows:
config.middleware.insert_after Rack::Runtime, Rack::TimeSec, :except => [/^\/assets\//]
Please read http://tenderlovemaking.com/2011/03/03/rack-api-is-awkward.html about how my method could miscalculate the time that a response takes. I'm also not sure about chuncked responses and if they'll be handled properly. If someone wants to look into this, that would be superb!
If you have a security bug report, please let me know confidentially. It's best to email me at [email protected]. You may download my GPG key from https://www.berkeleychurchill.com/contact.php. Your contribution will be publicly acknowledged once it's fixed (if you like).
Non-security bugs can be posted on the github tracker. If you're in doubt, consider it a security bug.
If you would like to contribute, please look into the TODO.md file for thoughts on what to do (other contributions are welcome too). Please let me know if you start working on something so effort isn't duplicated. Pull requests are welcomed.
Berkeley Churchill