Re: AVC denied for docker while trying to set labels for tmpfs mounts

[Daniel Walsh <dwalsh@redhat.com>]

On 7/21/21 18:17, Sujithra P wrote:
> Thanks Paul!
>
> Is there any specific centos/RH mailing list that I can ask? Not sure
> whether it is a problem with kernel/docker/kubelet.
> semodule -R seems to fix the problem, but not sure what is causing the
> loaded policy to get corrupt.
> Any insight on how to figure this out would be very much appreciated.
>
> Thanks
> Sujithra.
I am guessing that one of the containers is loading policy.  You should 
be able to see something in the auditlog, about a policy load.
> On Wed, Jul 21, 2021 at 2:01 PM Paul Moore <paul@paul-moore.com> wrote:
>> On Wed, Jul 21, 2021 at 2:46 PM Sujithra P <sujithrap@gmail.com> wrote:
>>> Hi SELinux Experts,
>>>
>>> The following issue is described in the below post as well.
>>> https://github.com/containers/container-selinux/issues/141
>>>
>>> Occasionally running into the following selinux denials for docker
>>>
>>> type=AVC msg=audit(1626732057.636:4583): avc:  denied  { associate }
>>> for  pid=57450 comm="dockerd" name="/" dev="tmpfs" ino=150014
>>> scontext=system_u:object_r:container_file_t:s0:c263,c914
>>> tcontext=system_u:object_r:lib_t:s0 tclass=filesystem permissive=0
>>>
>>> type=AVC msg=audit(1626812823.170:9434): avc:  denied  { associate }
>>> for  pid=20027 comm="dockerd" name="/" dev="tmpfs" ino=198147
>>> scontext=system_u:object_r:container_file_t:s0:c578,c672
>>> tcontext=system_u:object_r:locale_t:s0 tclass=filesystem permissive=0
>>>
>>>
>>>   level=error msg="Handler for POST
>>> /v1.40/containers/a3a875e7896384e3bff53b8317e91ed4301a13957f42187eb227f28e09bd877c/start
>>> returned error: error setting label on mount source
>>> '/var/lib/kubelet/pods/f7cee5b2-bcd9-4aa1-9d67-c75b677ba2a1/volumes/kubernetes.io~secret/secret':
>>> failed to set file label on
>>> /var/lib/kubelet/pods/f7cee5b2-bcd9-4aa1-9d67-c75b677ba2a1/volumes/kubernetes.io~secret/secret:
>>> permission denied"
>>>
>>>
>>> Docker is not able to set labels for these tmpfs mounts because they
>>> end up having wrong labels when they are created (sometimes
>>> "locale_t", sometimes "lib_t" which of course is not the
>>> default/correct context for tmpfs fs).
>>> Apparently semodule -R and deleting these tmps files or reboot of the
>>> node fixes the problem.
>>> Not sure what is causing the tmpfs mounts to get wrong labels in the
>>> first place.
>>>
>>> Everything seems to be fine to begin with, but as the system keeps
>>> scheduling pods on the node, this behavior is observed sometimes (not
>>> consistent always).
>>>
>>>
>>> OS Details:
>>>
>>> NAME="CentOS Linux"
>>> VERSION="8 (Core)"
>>> ID="centos"
>>> ID_LIKE="rhel fedora"
>>> VERSION_ID="8"
>>> PLATFORM_ID="platform:el8"
>>> PRETTY_NAME="CentOS Linux 8 (Core)"
>>>
>>> Docker Version:
>>> Client: Docker Engine - Community
>>> Version: 19.03.13
>>> API version: 1.40
>>> Go version: go1.13.15
>>> Git commit: 4484c46d9d
>>> Built: Wed Sep 16 17:02:36 2020
>>> OS/Arch: linux/amd64
>>> Experimental: false
>>>
>>> Kubernetes Version*
>>> v1.20.8-gke.1500
>>>
>>>
>>> Any help on how to debug this issue  would be greatly appreciated.
>> This sounds like it might be a problem with CentOS and/or your Docker
>> install, have you tried talking with the RH/CentOS folks about this
>> problem?  We focus mostly on upstream issues here and it isn't clear
>> to me at this moment that this is an upstream issue.
>>
>> --
>> paul moore
>> www.paul-moore.com