From: eastcoastcoder@...
Date: 2006-02-09T02:43:22+09:00
Subject: Sandboxing eval'd code

I'm working on a web app with complicated and ever changing business
rules.

On any given day, a new rule may need to be introduced - quickly.
There's not enough commonality between them to factor out and just make
configuration changes.  So the developers are always adding and
modifying, which of course makes reliability and testing difficult.

I thought of the following solution: New business rules can be coded in
Ruby and serialized into the database, along with configuration options
as to when they apply.  (The rules look at a bunch of things and return
either continue or abort - so making hooks for them is easy.  They
should never modify *anything* - just return a value.)
The app will load them and run them in a sandbox, catching all
exceptions and also timeing them out if they fail to return.
Developers can add/modify these rules easily, without touching the core
app, and, should one have a bug, although it may give the wrong result,
it won't bring the rest of the system down.

One concern I have is that I know that eval'd code can modify class
definitions, and access other objects via ObjectSpace.  Is there anyway
to eval code so that it can't change Classes and the like?  A true
Sandbox.

I'm not familiar enough with $SAFE to know what it can do (although
I've heard it is not reliable).  http://tryruby.hobix.com/ probably
does this, but source doesn't seem available.
http://approximity.com/cgi-bin/rubybuch_wiki/wpage.rb?nd=214 looks
relevant, but I can't figure it out.

In general, comments about this greatly appreciated.