You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
This repository has been archived by the owner on Mar 22, 2021. It is now read-only.
Right now we're using the defaults for Docker (of the host) - uid 0, default seccomp policy, etc.
There are cases where we want to be both more and less restrictive. First, I think we should push a lot of users towards OpenShift v3's RunAsRange SCC by default. Though this would break make install.
Second, I'd like to support recursive containerization - we should provide containers with access to bubblewrap. This is somewhat tricky to implement securely with Docker today - we have to have CAP_SYS_ADMIN etc. in the bounding set of the container. The easiest thing I can think of right now is to only support this with trusted base images, and explicitly turn off every other setuid binary (mount everything nosuid except for a special volume mount with bwrap?) Alternatively, we could run Docker (or any other container framework) with userns enabled (recursively) - segregate these hosts from the rest of the cluster which doesn't need it.
The text was updated successfully, but these errors were encountered:
Right now we're using the defaults for Docker (of the host) - uid 0, default seccomp policy, etc.
There are cases where we want to be both more and less restrictive. First, I think we should push a lot of users towards OpenShift v3's
RunAsRange
SCC by default. Though this would breakmake install
.Second, I'd like to support recursive containerization - we should provide containers with access to bubblewrap. This is somewhat tricky to implement securely with Docker today - we have to have
CAP_SYS_ADMIN
etc. in the bounding set of the container. The easiest thing I can think of right now is to only support this with trusted base images, and explicitly turn off every other setuid binary (mount everything nosuid except for a special volume mount with bwrap?) Alternatively, we could run Docker (or any other container framework) with userns enabled (recursively) - segregate these hosts from the rest of the cluster which doesn't need it.The text was updated successfully, but these errors were encountered: