summaryrefslogtreecommitdiff
path: root/doc/security.rdoc
blob: 06ebb830353b2624c72f9d32c3bc67863546f733 (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
= 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.

== <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.

== +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:

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

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...")