Skip to main content

Sandboxing Library for Python

Project description

This project is a total work in progress, no documentation yet.

It is being developed as part of the Baserock project.

The goal of this library is to provide the sandboxing functionality that is already present in the build tools Morph and YBD. We want this new library to be usable without depending on linux-user-chroot, so that it can be used on Mac OS X, and hopefully other platforms too.

A longer term goal is to become a useful, generic, cross-platform tool for running commands in an environment that is partially isolated from the host system in some way.

The library is implemented in Python currently. This is mostly because it is an adaptation of existing Python code, not because of any desire to exclude other languages. Maybe we could rewrite it as a C library with Python bindings.

SANDBOXLIB DOES NOT GUARANTEE YOU ANY KIND OF SECURITY. Its main purpose is for isolating software builds from the host system, to ensure that builds are not contacting the network, or reading or writing files outside the build environment.

Current backends

  • chroot: any POSIX OS, requires ‘root’ priviliges
  • linux-user-chroot: Linux-only, does not require ‘root’, requires linux-user-chroot to be installed and setuid root

Relationship to other projects

Sandboxing

libsandbox / pysandbox

The libsandbox library is a Linux-specific implementation of process sandboxing, which supports intercepting syscalls, calling setrlimit(), and dropping certain privileges.

PRoot

The PRoot tool provids features similar to linux-user-chroot, plus some extra code to allow running programs for a different architecture using virtualisation. The PRoot tool is discontinued.

Sandstorm.io

Sandstorm.io aims to be a platform for running web applications on shared infrastructure, with individual users in mind.

It uses the ‘namespaces’ feature of Linux. See https://github.com/sandstorm-io/sandstorm for more information.

Sandstorm.io is for a specific use case of web application sandboxing, so it doesn’t make sense for sandboxlib to wrap it. Use it directly if it suits your purpose!

seccomp

The Linux kernel provides the seccomp syscall, which can be used in two ways.

The SECCOMP_SET_MODE_STRICT operation creates a very restrictive but secure sandbox. Most programs wouldn’t work in this sandbox, but it does have some uses. It is used by Google Chrome, among other things.

The SECCOMP_SET_MODE_FILTER operation allows blacklisting certain system calls. This can be done in such a way that most existing programs work, but certain obvious security holes in a sandbox are closed (for example, the kexec() system call).

xdg-app (GNOME Application Sandboxing)

The xdg-app project started from a desire in the GNOME desktop project to allow running 3rd-party applications with some isolation from the host system. Mobile platforms like Android and iOS have been doing this for some time already.

It implements sandboxing mainly using the ‘namespaces’ feature of Linux. Find out more about the project and how the sandboxing is implemented.

xdg-app is for a specific use case of desktop application sandboxing, so it doesn’t make sense for sandboxlib to wrap it. Use it directly if it suits your purpose!

Containerisation

There is a lot of overlap between the topics of ‘containerisation’ and ‘sandboxing’. Many tools that work with ‘containers’ expect that containers are long-lived things, where the ‘sandboxlib’ library treats a sandbox as a much more lightweight, temporary thing.

App Container spec

I have been using the App Container spec as a reference during development. The scope of ‘sandboxlib’ is different to that of the App Container spec: ‘sandboxlib’ only deals with a single, isolated sandbox (which may or may not be a ‘container’), where the App Container spec is focused on multiple containers. However, ‘sandboxlib’ would be a useful building block for implementing a complete App Container runtime, and simple App Container images (.acis) should be runnable with the run-sandbox tool directly.

Clear Containers

Intel are producing a Linux distribution named Clear Linux, as part of a project to develop what they call Clear Containers. The idea is to make virtualisation with QEMU fast enough and convenient enough to compete with current containerisation software. All current containerisation systems use kernel namespacing, which provide a much weaker security barrier than full virtualisation.

The implementation depends on Linux’s KVM feature, plus patched versions of QEMU and Linux.

Docker

Docker allows managing multiple prebuilt container systems. While it can support multiple platform-specific backends for running containers, I am only aware of Linux-specific backends at the time of writing.

schroot

The use case for the schroot tool is ‘I want to define a contained environment once, and use it many times.’ The ‘sandboxlib’ library is more about dynamically creating sandboxes. If schroot suits your needs, just use it directly without any abstraction layer.

Python-specific Sandboxing

The ‘sandboxlib’ library is for sandboxing any program, at the operating system level.

If you want to do language-level sandboxing (i.e. run untrusted Python code within a larger Python program), there are some ways to do it.

The concensus seems to be that Python language-level sandboxing is pretty much impossible with the default ‘cpython’ Python runtime:

However, other Python runtimes do support language-level sandboxing. PyPy is one:

Build tools

Bazel

The Bazel build tool contains a Linux-specific sandbox implementation.

Morph

The Morph build tool (from Baserock) is the original source of the ‘sandboxlib’ linux_user_chroot backend. Hopefully Morph will adopt the ‘sandboxlib’ library in future.

License

License is GPLv2 but other licensing can be considered on request

Most of the copyright is currently Codethink but don’t let that put you off. There’s no intent to keep this as a Codethink-only project, nor will there be any attempt to get folks to sign a contributor agreement. Contributors retain their own copyright.

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Filename, size & hash SHA256 hash help File type Python version Upload date
sandboxlib-0.3.1.tar.gz (21.8 kB) Copy SHA256 hash SHA256 Source None

Supported by

Elastic Elastic Search Pingdom Pingdom Monitoring Google Google BigQuery Sentry Sentry Error logging AWS AWS Cloud computing DataDog DataDog Monitoring Fastly Fastly CDN SignalFx SignalFx Supporter DigiCert DigiCert EV certificate StatusPage StatusPage Status page