Rails is a relatively secure web framework out of the box provided you follow some standard (basic) security practices such as sanitizing user input and the like. A good security strategy will pursue "defense in depth", meaning that the application (and its deployment) will be design and executed with security in mind at many levels, not just at the outer "perimeter" level. In this way the fallout from a break-in can hopefully be mitigated by limiting just how far an intruder can penetrate and how much damage can be done.
Use trusted deployment solutions
In picking a "trusted" deployment solution you want to evaluate your options using common-sense criteria like:
- level of testedness
- history (previous security issues, response times, handling of vulnerabilities)
- widespread deployment (popularity and a wider installed base may also bring with it more extensive testing and a more active developer community)
At the moment, by these criteria the most "trusted" and reliable tools are probably nginx (front-end HTTP web server, load balancer and proxy), Mongrel (application server) and monit (process monitoring). There are upcoming contenders out there (especially in the application server arena) but if you’re serious about stability you’ll wait for these things to pan out and for the time-being will stick with the most tried-and-tested tools. These things do occasionally change; nginx, for example, has only recently displaced Apache as the front-end preferred by people "in the know".
Prefer stable versions
Don’t give in to the hype-machine-fueled excitement by "living on the edge" (the Rails trunk). Instead prefer stable released versions. By all means run pre-release software in staging and test environments, but do not turn your production server into a guinea pig.
Run your migrations as a separate user
Your Rails application should connect to the database as a relatively unprivileged user which can do things like create, read, update and delete records (the basic CRUD actions which are the staples of REST). This user should not have the privileges necessary to do the kinds of things that you can do in migrations like creating and dropping tables, or even dropping the entire database.
Create a separate database user with the privileges required for migrations and run them using that user. In this way if someone ever manages to pull off an SQL injection attack on your server then at least they won’t be able to drop entire tables.
Run your applications as different users
This is basic compartmentalization which you should do if you can. For example, if you only have a couple of applications on your server then nginx is lightweight enough that it’s probably realistic to run two separate master processes, each running as a different user, one for each application.
In this way if one app gets compromised then the scope for damage will be limited as the compromised user won’t be able to meddle with the other application.
Run your applications as an unprivileged user
Again, basic common sense: there is no need for your application servers to be running as root so you should be running them as normal, unprivileged users. In this way an attacker won’t own the entire machine as soon as they break in to your app.
Use sensible filesystem permissions
Bog standard security practice again. If you run multiple applications set your file permissions so that they can’t write each other’s files; better still, if you can, make it so that they can’t read each other’s files.
Obviously sensitive files which contain passwords like your
config/database.yml file should have the most restrictive permissions possible, but if you can you should make the entire home folder (or other folder that contains your application) readable only by the application user.
Set up automated log-file rotation
Too many requests can swell your log files to the point where they amount to a DOS. Use an automated log-rotation system and don’t run your server "close to the metal" in terms of free disk space to ensure that this kind of issue never becomes a problem.
Design with efficiency in mind
Again, with a view to avoiding a Denial of Service your application should make use of caching and efficient query design to minimize the risk of a DOS due to a malicious attack (or even a non-malicious one, such as being linked-to from Slashdot or Digg).
Use public-key authentication
If you need to connect to your server via SSH (and you probably do if you use Capistrano or Vlad) then use public-key authentication rather than passphrase-based authentication; if you adequately protect your private keys (by controlling access to your local machine and using appropriate file permissions) then this system is more secure.
Use sudo to limit deployment privileges
There are somethings that Capistrano needs to do with superuser privileges and it uses sudo to gain those privileges. Read up in the
sudoers man pages on how to configure your
/etc/sudoers file to make sure that only pre-approved commands are run using sudo and no others. In this way you limit the damage that could be done if somebody ever got hold of the private key needed to deploy your application.
Keep sensitive information out of your repository
Even if you’re the only person with access to your version control system it’s still good practice to keep sensitive information like passwords out of the repository, just in case it’s ever breached or some other lapse allows others to gain access.
config/database.yml file is an obvious example. Another is your Capistrano
config/deploy.rb; this is complex enough and evolves enough over time that you should track it in your repository, but obviously you shouldn’t store any plaintext passwords in it. If you really insist on not having to enter plaintext passwords (for example, at the sudo prompt) then store them in a local, protected
~/.caprc file rather than in the
Test your code, and specifically write tests that try to circumvent the protections that you put in place in your application to control access. In other words, pay special attention to confirming that your login and logout methods work as you think they do, and that you can’t access resources to which you shouldn’t have access.
Keep it simple
In the interest of keeping out the bugs you should always favor simple designs wherever possible (luckily, the whole "Web 2.0" movement encourages this trend so you won’t stand out if you do it). So, rather than implementing a full-blown user-and-groups permissions system with Access Control Lists and all sorts of other sophisticated paraphernalia, think about whether a simpler system (perhaps just with the public/private distinction) might be just as usable.
Don’t leak information
Avoid leaking internal implementation details wherever possible. In general these sorts of details are of zero interest to your users anyway, so in doing so you’ll actually be improving the interface for them. As an example, if you write a weblog application use human-readable permalinks like
/blog/italy-photos rather than
Obviously the word-based permalinks are nicer for users (and nicer for Google too), but they are also more secure than the id-based URL. The latter exposes an internal implementation detail of your application, and an attacker can guess where your next post will be before you’ve even published it. This in itself is not a problem but if you’ve let your security lapse in other areas (for example, not restricting access to unpublished drafts) then your minor information leak (of an implementation detail) starts leaking real information (unpublished content).
Don’t use predictable sequences
The classic example here is the "confirmation email": a user signs up, you send a confirmation email with a link in it that they must click to prove that they control the email account. Your confirmation link should be essentially random to prevent an attacker from guessing it. Cryptographic digests are the way to go here.
You can also make use of Rails nested resources to provide another layer of security. Given a secret token
e44c7e1472adcd12bbc84108df48bdfd52f1257d for confirmation
3214, then you can make sure that the token is only accepted within the context of the resource (http://example.com/confirmations/3214/e44c7e1472adcd12bbc84108df48bdfd52f1257d). In this way even if an attacker brute forces through the entire possible token space (unlikely), their efforts are useless unless they also manage to guess the confirmation id.
Salt your digests
The most cryptographically-strong digest in the world is worthless if it’s not properly used. For example, if you want to produce a "secret" token and send it to a user (again, let’s go with the "confirmation link" example mentioned above) you’ll want to include an component that’s unique to that user (their email address), a random or varying component (such as a timestamp and a random number; this ensures that tokens are always unique), and a secret salt (something that is never transmitted outside your application). In this way even if an attacker knows some of the information (for example, the victim’s email address) and can guess some of the rest (for example, the time stamp), they won’t know the random component or the salt, and so they have no remedy but to resort to a brute force attack.
Code review and code auditing
If you can, work with a colleague to review one another’s code. A pair of fresh eyes can often spot potential problems. If that’s not an option for you, periodically go back and look at your own old code; the elapsed time can sometimes be enough to help you see things anew.
attr_accessible in models
Get in the habit of using
attr_accessible in all models. A whitelist (
attr_accessible) is always going to be more secure than a blacklist (
Remember that even if you think you are protecting a model by passing
attr_protected, an attacker could still wreak havoc with your model because mass-assignment will allow them to call any method in the model, not just attributes corresponding to table rows in the database; all methods, even methods in
private sections are vulnerable.
class User < ActiveRecord::Base has_many :comments attr_protected :is_superuser end
Did you know that given a model like the above Rails provides a
comment_ids method thanks to the
has_many association? This is vulnerable to mass-assignment because of the use of
attr_protected rather than
attr_accessible. An attacker can thus reassign comment ownership at will, and if they assign ownership of a comment to themselves then they are free to edit the rest of it as they please.
Even if you know about the existence of the
comment_ids method (which isn’t visible!), how can you be sure you know about all the methods that Rails might be adding behind the scenes? How can you be sure you’ll know about the methods that Rails 3.0 adds? Rails 4.0? And so on. Clearly,
attr_accessible is the only sane, secure choice (I personally think that
attr_protected is a trap that gives a false sense of security and should never have been added to Rails).
And as I mentioned above, it’s not only attributes that correspond to columns in the database that are vulnerable. Given a method like this in the
class User < ActiveRecord::Base private def really_delete_all_records= confirm delete_all if confirm end end
attr_protected won’t stop an attacker here from assigning to the
really_delete_all_records attribute, even though it is a private method. Witness:
# goodbye records! User.find(:first).update_attributes(:really_delete_all_records => 1)
If the input is coming from a form, passing in
really_delete_all_records is trivial.
There are many, many techniques that can be applied, so I’ll update this page with those in the future. Some of the things I’d like to cover include: