How do I secure scripts run using javax.scripting?

Question

I am using javax.scripting to add support for running arbitrary user-uploaded JavaScripts on the server-side. Obviously I want to secure those scripts!

Rhino, on it's own, has a framework for securing scripts at runtime. The documentation for javax.scripting, however, doesn't mention security, permissions or restricting classes available to the script. So is this just a huge hole in the javax.scripting API that it doesn't offer a framework to secure scripts it executes?

I don't want to use Rhino directly because I originally tried that but had some problems exposing Java instances to the running script. The javax.scripting framework made it (which uses Rhino under the hood) made this trivial and also simplified running scripts in a multi-threaded server.

I would like to white-list Java classes that can be accessed/instantiated within the running script. Can anyone point me to an example or documentation on how to achieve this?

Answer 1

FYI, this is now possible in the new Java 8 implementation of javax.scripting which uses a new engine called Nashorn. See Secure Nashorn JS Execution

Answer 2

http://codeutopia.net/blog/2009/01/02/sandboxing-rhino-in-java/ describes a way to sandbox rhino, and javax.scripting uses Rhino as the JS script engine so you should be able to use the above, though the package names might differ.

I’ve been working on a Java app which needed Rhino for scripting. The app would need to run untrusted JavaScript code from 3rd parties, so I had to find a way to block access to all Java methods, except the ones I wanted. This would not be a problem if there was an easy way to disable LiveConnect - the feature of Rhino which provides java access to scripts - but there is no such thing.

However, after a lot of digging around, I finally found a way to do this without too much hacking. In fact, it can be done by just extending a few of the Rhino classes, and using the setters provided to override some of the default ones.

Answer 3

It turns out that javax.scripting does not offer a security framework. After some searching I found a document in Google's cache that suggested trying to use Java's doPrivilegedAction framework but after some experimentation, I was unable to get this to prevent the scripts from opening sockets or accessing the filesystem.

After I asked this question I discovered it was previously asked here on StackOverflow: How can you run Javascript using Rhino for Java in a sandbox? On that page, it falsely indicates that the Rhino included in the JDK6 has security worked out already. As I indicated, I was able to open sockets and other harmful actions from the script.

In the end I abandoned javax.scripting and embedded Rhino directly. By building a custom ContextFactory that is also a ClassShutter I was able to achieve two results easily:

1. Restricts script execution time to a maximum time limit
2. Restricts class access to those I have white-listed, which is basically java.lang.* and a select few classes in my server's hierarchy.

CodeUtopia (which I can't link to because, as a new user, I'm not allowed to link to multiple pages in a single post; but it's linked in the other StackOverflow post) was valuable in describing the ClassShutter architecture and Rhino's own ContextFactory API page describes how to build a custom ContextFactory.