ruby--ruby/doc/security.rdoc

= Ruby Security

The Ruby programming language is large and complex and there are many security
pitfalls often encountered by newcomers and experienced Rubyists alike.

This document aims to discuss many of these pitfalls and provide more secure
alternatives where applicable.

Please check the full list of publicly known CVEs and how to correctly report a
security vulnerability, at: http://www.ruby-lang.org/en/security/
Japanese version is here: http://www.ruby-lang.org/ja/security/

Security vulnerabilities should be reported via an email to
mailto:security@ruby-lang.org ({the PGP public
key}[http://www.ruby-lang.org/security.asc]), which is a private mailing list.
Reported problems will be published after fixes.

== <code>$SAFE</code>

Ruby provides a mechanism to restrict what operations can be performed by Ruby
code in the form of the <code>$SAFE</code> variable.

However, <code>$SAFE</code> does not provide a secure environment for executing
untrusted code even at its maximum level of +4+. <code>$SAFE</code> is
inherently flawed as a security mechanism, as it relies on every unsafe
operation performed by any C method to be guarded by a <code>$SAFE</code>
check. If this check is ever missed, the entire security of the system is
compromised. <code>$SAFE</code> also does not offer any protection against
denial of service attacks.

If you need to execute untrusted code, you should use an operating system level
sandboxing mechanism. On Linux, ptrace or LXC can be used to sandbox
potentially malicious code. Other similar mechanisms exist on every major
operating system.

== +Marshal.load+

Ruby's +Marshal+ module provides methods for serializing and deserializing Ruby
object trees to and from a binary data format.

Never use +Marshal.load+ to deserialize untrusted or user supplied data.
Because +Marshal+ can deserialize to almost any Ruby object and has full
control over instance variables, it is possible to craft a malicious payload
that executes code shortly after deserialization.

If you need to deserialize untrusted data, you should use JSON as it is only
capable of returning 'primitive' types such as strings, arrays, hashes, numbers
and nil. If you need to deserialize other classes, you should handle this
manually. Never deserialize to a user specified class.

== YAML

YAML is a popular human readable data serialization format used by many Ruby
programs for configuration and database persistance of Ruby object trees.

Similar to +Marshal+, it is able to deserialize into arbitrary Ruby classes.
For example, the following YAML data will create an +ERB+ object when
deserialized:

  !ruby/object:ERB
  src: puts `uname`

Because of this, many of the security considerations applying to Marshal are
also applicable to YAML. Do not use YAML to deserialize untrusted data.

== Symbols

Symbols are often seen as syntax sugar for simple strings, but they play a much
more crucial role. The MRI Ruby implementation uses Symbols internally for
method, variable and constant names. The reason for this is that symbols are
simply integers with names attached to them, so they are faster to look up in
hashtables.

Once a symbol is created, the memory used by it is never freed. If you convert
user input to symbols with +to_sym+ or +intern+, it is possible for an attacker
to mount a denial of service attack against your application by flooding it
with unique strings. Because each string is kept in memory until the Ruby
process exits, this will cause memory consumption to grow and grow until Ruby
runs out of memory and crashes.

Be careful with passing user input to methods such as +send+,
+instance_variable_get+ or +_set+, +const_get+ or +_set+, etc. as these methods
will convert string parameters to symbols internally and pose the same DoS
potential as direct conversion through +to_sym+/+intern+.

The workaround to this is simple - don't convert user input to symbols. You
should attempt to leave user input in string form instead.

== Regular expressions

Ruby's regular expression syntax has some minor differences when compared to
other languages. In Ruby, the <code>^</code> and <code>$</code> anchors do not
refer to the beginning and end of the string, rather the beginning and end of a
*line*.

This means that if you're using a regular expression like
<code>/^[a-z]+$/</code> to restrict a string to only letters, an attacker can
bypass this check by passing a string containing a letter, then a newline, then
any string of their choosing.

If you want to match the beginning and end of the entire string in Ruby, use
the anchors +\A+ and +\z+.

== +eval+

Never pass untrusted or user controlled input to +eval+.

Unless you are implementing a REPL like +irb+ or +pry+, +eval+ is almost
certainly not what you want. Do not attempt to filter user input before passing
it to +eval+ - this approach is fraught with danger and will most likely open
your application up to a serious remote code execution vulnerability.

== +send+

'Global functions' in Ruby (+puts+, +exit+, etc.) are actually private instance
methods on +Object+. This means it is possible to invoke these methods with
+send+, even if the call to +send+ has an explicit receiver.

For example, the following code snippet writes "Hello world" to the terminal:

  1.send(:puts, "Hello world")

You should never call +send+ with user supplied input as the first parameter.
Doing so can introduce a denial of service vulnerability:

  foo.send(params[:bar]) # params[:bar] is "exit!"

If an attacker can control the first two arguments to +send+, remote code
execution is possible:

  # params is { :a => "eval", :b => "...ruby code to be executed..." }
  foo.send(params[:a], params[:b])

When dispatching a method call based on user input, carefully verify that the
method name. If possible, check it against a whitelist of safe method names.

Note that the use of +public_send+ is also dangerous, as +send+ itself is
public:

  1.public_send("send", "eval", "...ruby code to be executed...")

== DRb

As DRb allows remote clients to invoke arbitrary methods, it is not suitable to
expose to untrusted clients.

When using DRb, try to avoid exposing it over the network if possible. If this
isn't possible and you need to expose DRb to the world, you *must* configure an
appropriate security policy with <code>DRb::ACL</code>.