Revision 5 as of 2006-03-13 13:13:27

Clear message

TableOfContents

While a lot of effort has gone into making it difficult or impossible to crash the Python interpreter in normal usage, there are lots fairly easy ways to crash the interpreter. The BDFL [http://mail.python.org/pipermail/python-dev/2006-January/059621.html pronounced] recently on the python-dev mailing list:

    I'm not saying it's uncrashable. I'm saying that if you crash it, it's a
    bug unless proven harebrained.

I thought it might be worthwhile to document some ways the interpreter can be crashed -- although most of these are very unlikely to crop up in real code.

There is also [http://svn.python.org/view/python/trunk/Lib/test/crashers/ a directory in SVN repository] that demonstrates known ways to crash Python.

Bogus Input

Through Python 2.4 you could crash the interpreter by redirecting stdin from a directory:

    % python2.4 -c 'import sys ; print sys.version'
    2.4.1 (#3, Jul 28 2005, 22:08:40) 
    [GCC 3.3 20030304 (Apple Computer, Inc. build 1671)]
    % python2.4 < .
    Bus error

Starting with 2.5 the interpreter notices and aborts:

    % python2.5 -c 'import sys ; print sys.version'
    2.5a0 (41847M, Dec 29 2005, 22:21:03) 
    [GCC 3.3 20030304 (Apple Computer, Inc. build 1671)]
    % python2.5 < .
    Fatal Python error: <stdin> is a directory
    Abort trap

Some modules also do not perform rigorous checking of data they operate on. The marshal module can cause it to crash the interpreter when given certain strings:

    $ python
    Python 2.4.2 (#2, Sep 30 2005, 21:19:01) 
    [GCC 4.0.2 20050808 (prerelease) (Ubuntu 4.0.1-4ubuntu8)] on linux2
    Type "help", "copyright", "credits" or "license" for more information.
    >>> import os, marshal
    >>> while 1:
    ...     try:
    ...         marshal.loads(os.urandom(16))
    ...     except:
    ...         pass
    ... 
    Segmentation fault

Exhausting Resources

There are a number of areas where resource exhaustion can crash the interpreter. Here's one fairly easy to demonstrate way to do it though:

    % python
    Python 2.5a0 (41847M, Dec 29 2005, 22:21:03) 
    [GCC 3.3 20030304 (Apple Computer, Inc. build 1671)] on darwin
    Type "help", "copyright", "credits" or "license" for more information.
    >>> import sys
    >>> sys.setrecursionlimit(1<<30)
    >>> f = lambda f:f(f)
    >>> f(f)
    Segmentation fault

A slightly subtler example involves getting the interpreter to exhaust some resource internally while performing a single operation:

    $ python
    Python 2.4.2 (#2, Sep 30 2005, 21:19:01) 
    [GCC 4.0.2 20050808 (prerelease) (Ubuntu 4.0.1-4ubuntu8)] on linux2
    Type "help", "copyright", "credits" or "license" for more information.
    >>> f = lambda: None
    >>> for i in xrange(1000000):
    ...     f = f.__call__
    ... 
    >>> del f
    Segmentation fault

GC/weakref interaction

Interaction between these two systems has historically been a sticky point for CPython. There is still at least one problem in Python 2.4.2:

    $ python
    Python 2.4.2 (#2, Sep 30 2005, 21:19:01) 
    [GCC 4.0.2 20050808 (prerelease) (Ubuntu 4.0.1-4ubuntu8)] on linux2
    Type "help", "copyright", "credits" or "license" for more information.
    >>> import weakref
    >>> ref = None
    >>> class Target:
    ...   def __del__(self):
    ...       global ref
    ...       ref = weakref.ref(self)
    ... 
    >>> def g():
    ...   w = Target()
    ...   w = None
    ...   print ref()
    ... 
    >>> g()
    Segmentation fault

Dangerous Modules

Some modules are designed to allow programmers access to the guts of things. Naturally, they also give programmers the opportunity to shoot themselves in the foot. Here are a few.

Unable to edit the page? See the FrontPage for instructions.