Open ghost opened 3 years ago
Otherwise, security is seriously affected
What specifically?
Can you summarize? One shouldn't need to watch a 24 minute video to get an accurate bug report. No container runs with extended capabilities so I'm not sure what the attack vector is here.
Summarize what? Read the first comment carefully.
No container runs with extended capabilities
NET_RAW capability and readOnlyRootFilesystem false = already extended capabilities
And there is nothing we can do thanks to the images. That's the problem.
You haven't explained what the attack vector is.
Also, how is the net raw cap enabled? Is that a default? We don't enable it explicitly AFAICT.
@saghul I think once we move s6-overlay to use user other than the root then most of the things would be fixed. As of now we can't start the container without root user as S6-overlay uses it for managing configuration. We can try version 2.X.X as they have introduced User directive(run S6-overlay without root user) in this version.
@Ais8Ooz8 we get it that this question is more about the best practices for running containers. But there are services like jibri where we require root access for /dev/shm
hence it wouldn't be easy to make them work with non-root user. It would be great if you can create a PR and help the community.
@saghul I think once we move s6-overlay to use user other than the root then most of the things would be fixed. As of now we can't start the container without root user as S6-overlay uses it for managing configuration. We can try version 2.X.X as they have introduced User directive(run S6-overlay without root user) in this version.
Indeed experimenting with 2.X.X should be the way to go here.
@Ais8Ooz8 we get it that this question is more about the best practices for running containers. But there are services like jibri where we require root access for
/dev/shm
hence it wouldn't be easy to make them work with non-root user. It would be great if you can create a PR and help the community.
IIRC there is a command line flag to prevent Chromium from using /dev/shm, but we can see how to cross that river when we get there.
For example, in Kubernetes it would be normal to run containers with such directives:
Various DevSecOps tools check these directives, for example https://docs.kics.io/queries/kubernetes-queries
But jitsi images prevent such important settings from being applied
Perhaps this needs revision? Otherwise, security is seriously affected
How can we avoid using https://github.com/just-containers/s6-overlay? It seems we need to keep something like this in our mind https://github.com/GoogleContainerTools/distroless and try to move in this direction