Bug 1942744

Created attachment 1766059 [details]
collector-scc.yaml

Created attachment 1766059 [details]
collector-scc.yaml

Description of problem:

openshift-api pods need to write privileges for the root filesystem. 
If you create an SCC with readOnlyRootFilesystem: true and openshift-api pick that one startup will fail.


Version-Release number of selected component (if applicable):

OpenShift 4.7.3

How reproducible:

$  oc get pods -o "custom-columns=NAME:.metadata.name,SCC:.metadata.annotations.openshift\.io/scc,SERVICEACCOUNT:.spec.serviceAccountName"
NAME                        SCC             SERVICEACCOUNT
apiserver-598d4c595-59fdh   node-exporter   openshift-apiserver-sa
apiserver-598d4c595-rgvp4   node-exporter   openshift-apiserver-sa
apiserver-598d4c595-vkf8b   node-exporter   openshift-apiserver-sa
$ oc create -f  - <<EOF
> apiVersion: security.openshift.io/v1
> kind: SecurityContextConstraints
> metadata:
>   name: collector
>   labels:
>     app.kubernetes.io/instance: stackrox-secured-cluster-services
>     app.kubernetes.io/managed-by: Helm
>     app.kubernetes.io/name: stackrox
>     app.kubernetes.io/part-of: stackrox-secured-cluster-services
>     app.kubernetes.io/version: 3.0.56.1
>     helm.sh/chart: sensor-56.1.0
>     auto-upgrade.stackrox.io/component: "sensor"
>   annotations:
>     email: support
>     meta.helm.sh/release-name: stackrox-secured-cluster-services
>     meta.helm.sh/release-namespace: stackrox
>     owner: stackrox
>     kubernetes.io/description: This SCC is based on privileged, hostaccess, and hostmount-anyuid
> users:
>   - system:serviceaccount:stackrox:collector
> allowHostDirVolumePlugin: true
> allowPrivilegedContainer: true
> fsGroup:
>   type: RunAsAny
> groups: []
> priority: 0
> readOnlyRootFilesystem: true
> runAsUser:
>   type: RunAsAny
> seLinuxContext:
>   type: RunAsAny
> seccompProfiles:
>   - '*'
> supplementalGroups:
>   type: RunAsAny
> allowHostIPC: false
> allowHostNetwork: false
> allowHostPID: falsef
> allowHostPorts: false
> allowPrivilegeEscalation: true
> allowedCapabilities: []
> defaultAddCapabilities: []
> requiredDropCapabilities: []
> volumes:
>   - configMap
>   - downwardAPI
>   - emptyDir
>   - hostPath
>   - secret
> EOF
securitycontextconstraints.security.openshift.io/collector created
$ oc delete pod apiserver-598d4c595-59fdh
pod "apiserver-598d4c595-59fdh" deleted
$ oc get pods
NAME                        READY   STATUS             RESTARTS   AGE
apiserver-598d4c595-rgvp4   2/2     Running            0          5h27m
apiserver-598d4c595-vkf8b   2/2     Running            0          5h33m
apiserver-598d4c595-z2rvf   0/2     CrashLoopBackOff   4          119s
$  oc get pods -o "custom-columns=NAME:.metadata.name,SCC:.metadata.annotations.openshift\.io/scc,SERVICEACCOUNT:.spec.serviceAccountName"
NAME                        SCC             SERVICEACCOUNT
apiserver-598d4c595-rgvp4   node-exporter   openshift-apiserver-sa
apiserver-598d4c595-vkf8b   node-exporter   openshift-apiserver-sa
apiserver-598d4c595-z2rvf   collector       openshift-apiserver-sa

$ oc logs apiserver-598d4c595-z2rvf -c openshift-apiserver
Copying system trust bundle
cp: cannot remove '/etc/pki/ca-trust/extracted/pem/tls-ca-bundle.pem': Read-only file system



Steps to Reproduce:
1. Create SCC, attached (collector-scc.yaml)
2. Delete one openshift-api pod and wait for the new one. 
3.

Actual results:

Startup fail because of wrong SCC

Copying system trust bundle
cp: cannot remove '/etc/pki/ca-trust/extracted/pem/tls-ca-bundle.pem': Read-only file system

Expected results:

Choose a propper SCC

Additional info:

 * We had exactly the same problem with the OpenShift authentication deployment: 
   Bug 1824800 - openshift authentication operator is in a CrashLoopBackOff