Closed dongsupark closed 6 years ago
I think the problem is related to how we implement seccomp in the rktlet.
In https://github.com/kubernetes-incubator/rktlet/commit/8088ab5394eb26cf77f56c5cd5adad1eebf505a3 I implemented the forwarding of seccomp options, but I only did it through annotations since that's what's implemented in v1.7.x
. However, later (in v1.8.0
) they moved it to a field in SecurityContext
, which is what should support seccomp docker/default on the container is using. I'm not sure why the other seccomp tests pass.
Even though the cri-tools readme says "We are currently working toward an alpha version of CRI validation tests to be used in conjunction with Kubernetes 1.7.", it seems cri-tools v0.2 bumps the CRI interface to v1.8.
There are many seccomp tests that fail. One of the simple examples is the test "should support seccomp docker/default on the container".
That means basically that
/proc/self/status
showsSeccomp: 0
(not in seccomp mode). As rkt already supports the default seccomp profiledocker/default
, I'm not sure I understand.On the other hand, not all seccomp tests fail. For example, the test
runtime should support setting hostname with docker/default seccomp profile and SYS_ADMIN
succeeds. Strange.