sandman
v0.5.3
Published
Sandbox for executing (somewhat) untrusted code
Downloads
18
Maintainers
Readme
Sandman
A container for running untrusted(-ish) Node.js programs.
If you're in any kind of multi-tenant environment where you're automatically running Node.js written by other people, or using modules that you haven't seen ahead of time, just running them is quite dangerous - even accidentally. A malicious (or poorly written) module would have access to your environment variables, any files that Node has access to, and can start programs and connect to outside servers.
If you have a multi-tenant environment purely for hosting Node.js to be executed (a la Heroku), containerization and virtual machines are a much better solution. But if untrusted Node.js is a small side effect of your application, Sandman provides a comparatively simple and low overhead way to execute code.
Usage
Basic Usage
entrypoint.js
var untrusted = require('untrusted');
module.exports = function (arg1, arg2, callback) {
var myValue = untrusted(arg1, arg2);
callback(null, myValue);
};
runner.js
var Sandman = require('sandman');
Sandman.run('./entrypoint.js', '/path/to/safe/root', [arg1, arg2], function (err, myValue) {
if(err) throw err;
console.log(myValue); // outputs the result of `untrusted(arg1, arg2)`
});
When you run Sandman on your entrypoint file, it will call the resulting module.exports
with your arguments and a callback. Call the callback when you've run your untrusted modules.
Advanced Usage
Constructor
You can have more control over the Sandman instance by using the constructor:
var sandman = new Sandman("./entrypoint.js", "/some/safe/root");
The Sandman instance exposes the following properties and methods:
interface
a ChildProcess connected to the Sandman client. This can be used to send messages and sockets to your entrypoint file.run
- Start the entrypoint file. Called with an array of arguments and a callback._onMessage
- Listens formessage
from the client. Change it before callingrun
._onError
- Listens forerror
from the client. Change it before callingrun
._onExit
- Listens forexit
from the client. Change it before callingrun
.
Client
Within the entrypoint file, you can access the client by doing:
var sandmanClient = require('sandman')
The Sandman Client exposes the following methods and properties:
arguments
array of arguments called onSandman#run
. (these are applied to the entrypoint file'smodule.exports
function)callback
Callback to be called when the entrypoint file is done. (this is the last argument supplied to the entrypoint file'smodule.exports
function)root
the secure root that Sandman was called infilename
the entrypoint file name. (this is also available on the global scope in__filename
)interface
An alias forprocess
, this is used to communicate with the Sandman constructor viaClient#interface.send
sendError
A sugary method for sending errors to the Sandman constructor (used internally whencallback
's first parameter is an error)
Be warned - do not pass the client to any untrusted code - it probably has everything you need to break out of the jailed environment in a hot second.
About
Security
This isn't truly secure, and there likely isn't a way to make Node.js secure at the application level (See this discussion). Instead, this module provides a way to execute Node.js code that should prevent against most kinds of bad stuff, intentional or unintentional.
Features
- Operates in a new process, and every module in its own context
- No
fs
access outside the definedroot
, except files within anode_modules
directory - Cannot change the current working directory to be outside of
root
- Cannot change uid or gid
- Timeout to kill runaway processes
- Cannot
require
files outside ofroot
, except fiels within anode_modules
directory - No access to dangerous core modules:
- child_process
- cluster
- http
- https
- net
- tls
- dgram
- repl
The key difference between Sandman and most other sandboxing libraries is that the entire dependency chain is contained. So, for example, requiring fs-extra
, which in turn requires fs
will not get you outside the sandbox.
Limitations
The most obvious limitation is the attack surface area - there are almost certainly ways to exploit this kind of sandboxing, so don't rely on it for anything super important.
Most specifically, for reasons of practicality and interop with npm, require
and fs
calls are not limited to root
when the requested file is inside of a node_modules
directory. This should be fine in most cases, as published modules shouldn't contain anything sensitive, but it certainly opens up a potential attack vector.
The other limitation is one of speed and memory. Each Sandman instance creates a new process, and each dependency is put in a new context. This is much lower overhead than OS level sandboxing, but you obviously can't have tons of these (i.e. thousands) on a single machine.