[Python-Dev] Sandboxing Python

[Victor Stinner]

On 01/03/2012 22:59, Victor Stinner wrote:
>> I challenge anymore to break pysandbox! I would be happy if anyone
>> breaks it because it would make it more stronger.

Results, one week later. Nobody found a vulnerability giving access to 
the filesystem or to the sandbox.

Armin Rigo complained that CPython has known "crasher" bugs. Except of 
the compiler recursion, I fixed those bugs in CPython 3.3.

Serhiy Storchaka found a bug in the pysandbox timeout: long operations 
implemented in C hangs the sandbox, the timeout contrain is not applied. 
Guido proposed to abort the process (use the default SIGALRM action). I 
proposed to add an option to use a subprocess. Both solutions are not 
exclusive.

Armin Rigo also noticed that PyPy sandbox design is more robust than 
pysandbox design, I agree with him even if I think a CPython sandbox is 
useful and users ask for such protection.

I have no idea how many developers tried to break the pysandbox security.

Victor