[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [oss-security] Mitigating malicious packages in gnu/linux
My name is Aditya, and I represent the NYU Secure Systems Lab, where we
actively research and develop solutions for the kind of problems discussed
in this thread.
In particular, we believe that a combination of The Update Framework (TUF)
and in-toto will help to address the problem of malicious packages. TUF is
the technology that ensures that packages have not been tampered between
the repository and end-users, whereas in-toto is the technology that
ensures that packages have been built correctly by the CI/CD using source
code signed by developers, securing the software supply chain end to end. A
critical property of both technologies is that they are fundamentally
designed to be resilient against a compromise of some signing keys in the
system. By combining both TUF and in-toto [X], it is possible for a Linux
distribution to guarantee that packages have not been tampered with
anywhere between developers and end-users, despite being built and
distributed by machines.
TUF is the de-facto standard for signing container images, used by Docker,
IBM, and Microsoft. It is also being used by Google to update everything on
their next-gen Fuchsia OS. A version of TUF called Uptane has been
standardized for use by North American ground vehicles [Y]. in-toto is
being used to verify reproducible builds for Debian packages.
We would be happy to help your community integrate both technologies, if so
desired. Let us know if you have questions!
On Wed, Nov 20, 2019 at 7:45 AM Solar Designer <solar AT openwall.com> wrote:
> On Tue, Nov 19, 2019 at 01:33:48PM +0200, Georgi Guninski wrote:
> > As end user and contributor of gnu/linux, I am concerned about malicious
> > packages (either hostile developers or hacked developers or another
> > and have two questions:
> > * What do linux vendors to avoid malicious packages?
> Back when Openwall GNU/*/Linux was being actively developed, I used to
> review each contributor's changes before making them public. I also
> (re-)verified authenticity of third-party source tarballs instead of
> blindly trusting whatever the contributor could have uploaded to us.
> (I'd do the same now, but without active development there's simply
> nothing to review lately.)
> Of course, this approach doesn't scale as-is (with just one person to
> review and publish everything) to larger distros, but some kind of peer
> review can and should be present.
> > * As end user what can I do to mitigate malicious packages?
> Try to install only what's needed, or not a lot more than what's needed.
> (Can't be done perfectly with larger distros and their dependency hell.)
> Contrary to traditional best practices, update only what and when needs
> to be updated. (Of course, you take responsibility to watch for any
> relevant security updates, or accept the risk if you neglect to do that.
> You also miss silent security fixes, but on the other hand you similarly
> miss newly introduced vulnerabilities.)
> Use a long-term support distro, preferably starting half-way into its
> lifetime when updates are already infrequent. (Similar risk of missing
> silent security fixes in new upstream versions, but also avoiding new
> Setup packet filters with blocking and logging of unexpected outbound
> packets, including to console so that you'd notice.
> Setup custom anomaly detection and actually watch it - e.g., for new
> programs running that haven't ever run before, etc.
> Use multiple pseudo-user accounts (doesn't protect against issues in
> packages' pre/post-install scripts, etc.), containers, VMs - but even
> then you have the risk of getting the same malicious package in multiple
> VMs, which e.g. on Qubes OS could happen through updating a template VM.