This morning a container escape vulnerability in runc was announced. We wanted to provide some guidance to Kubernetes users to ensure everyone is safe and secure.
What Is Runc?
Very briefly, runc is the low-level tool which does the heavy lifting of spawning a Linux container. Other tools like Docker, Containerd, and CRI-O sit on top of runc to deal with things like data formatting and serialization, but runc is at the heart of all of these systems.
Kubernetes in turn sits on top of those tools, and so while no part of Kubernetes itself is vulnerable, most Kubernetes installations are using runc under the hood.
What Is The Vulnerability?
While full details are still embargoed to give people time to patch, the rough version is that when running a process as root (UID 0) inside a container, that process can exploit a bug in runc to gain root privileges on the host running the container. This then allows them unlimited access to the server as well as any other containers on that server.
If the process inside the container is either trusted (something you know is not hostile) or is not running as UID 0, then the vulnerability does not apply. It can also be prevented by SELinux, if an appropriate policy has been applied. RedHat Enterprise Linux, CentOS, and Fedora all include appropriate SELinux permissions with their packages and so are believed to be unaffected.
The most common source of risk is attacker-controller container images, such as unvetted images from public repositories.
What Should I Do?
As with all security issues, the two main options are to mitigate the vulnerability or upgrade your version of runc to one that includes the fix.
As the exploit requires UID 0 within the container, a direct mitigation is to ensure all your containers are running as a non-0 user. This can be set within the container image, or via your pod specification:
apiVersion: v1 kind: Pod metadata: name: run-as-uid-1000 spec: securityContext: runAsUser: 1000 # ...
This can also be enforced globally using a PodSecurityPolicy:
apiVersion: policy/v1beta1 kind: PodSecurityPolicy metadata: name: non-root spec: privileged: false allowPrivilegeEscalation: false runAsUser: # Require the container to run without root privileges. rule: 'MustRunAsNonRoot'
Setting a policy like this is highly encouraged given the overall risks of running as UID 0 inside a container.
Another potential mitigation is to ensure all your container images are vetted and trusted. This can be accomplished by building all your images yourself, or by vetting the contents of an image and then pinning to the image version hash (
image: external/[email protected]:7832659873hacdef).
Upgrading runc can generally be accomplished by upgrading the package
runc for your distribution or by upgrading your OS image if using immutable images. This is a list of known safe versions for various distributions and platforms:
- Ubuntu –
- Debian –
- RedHat Enterprise Linux –
docker 1.13.1-91.git07f3374.el7(if SELinux is disabled)
- Amazon Linux –
- CoreOS –
- Kops Debian – in progress
- Docker –
Some platforms have also posted more specific instructions:
Google Container Engine (GKE)
Google has issued a security bulletin with more detailed information but in short, if you are using the default GKE node image then you are safe. If you are using an Ubuntu or CoreOS node image then you will need to mitigate or upgrade to an image with a fixed version of runc.
Amazon Elastic Container Service for Kubernetes (EKS)
Amazon has also issued a security bulletin with more detailed information. All EKS users should mitigate the issue or upgrade to a new node image.
We don’t have specific confirmation that Docker for Mac and Docker for Windows are vulnerable, however it seems likely. Docker has released a fix in version 18.09.2 and it is recommended you upgrade to it. This also applies to other deploy systems using Docker under the hood.
If you are unable to upgrade Docker, the Rancher team has provided backports of the fix for many older versions at github.com/rancher/runc-cve.
Getting More Information
If you have any further questions about how this vulnerability impacts Kubernetes, please join us at discuss.kubernetes.io.
Originally posted on Kubernetes – Production-Grade Container Orchestration