Current problematic cases with immutable loginuid

Current problematic cases with immutable loginuid

[Andreas Hasenack]

Hi,

I was reading up on setting loginuid immutable, and was wondering what
are the current known problematic cases.

In general, anything that requires switching a set loginuid to another
value will be blocked:
- sshd started on another port by the logged in user to debug
something, and that debug requires logging in as a different user than
the one who started it up
- container that starts up within the user's session, instead of via
dockerd/containerd, systemd, or some other already-running daemon. I
read a lengthy bug in Redhat's bugzilla about a bad interaction with
systemd's nspawn, where apparently the container is started directly,
and thus inheriting the user's loginuid, instead of being started via
a request to systemd (the daemon)

The manpage mentions "certain kinds of containers", and I assume it's
a reference to nspawn's case above.

Are there other prominent problematic situations that people have
encountered while setting loginuid immutable?

--
Linux-audit mailing list
Linux-audit@redhat.com
https://listman.redhat.com/mailman/listinfo/linux-audit

Re: Current problematic cases with immutable loginuid

[Richard Guy Briggs]

On 2021-06-07 12:09, Andreas Hasenack wrote:
> Hi,
> 
> I was reading up on setting loginuid immutable, and was wondering what
> are the current known problematic cases.
> 
> In general, anything that requires switching a set loginuid to another
> value will be blocked:
> - sshd started on another port by the logged in user to debug
> something, and that debug requires logging in as a different user than
> the one who started it up

I could see that being an issue since that user may not have
CAP_AUDIT_CONTROL and be starting it as an unprivileged user on a port
number larger than 1024.

> - container that starts up within the user's session, instead of via
> dockerd/containerd, systemd, or some other already-running daemon. I
> read a lengthy bug in Redhat's bugzilla about a bad interaction with
> systemd's nspawn, where apparently the container is started directly,
> and thus inheriting the user's loginuid, instead of being started via
> a request to systemd (the daemon)

My understanding was that any container with systemd running as PID=1 in
a PID namespace was causing this problem.  This is a known and
intentional limitation.  Auditd should be able to run in the initial
user namespace not under a namespaced systemd.  There should be no issue
here.

> The manpage mentions "certain kinds of containers", and I assume it's
> a reference to nspawn's case above.
> 
> Are there other prominent problematic situations that people have
> encountered while setting loginuid immutable?

- RGB

--
Richard Guy Briggs <rgb@redhat.com>
Sr. S/W Engineer, Kernel Security, Base Operating Systems
Remote, Ottawa, Red Hat Canada
IRC: rgb, SunRaycer
Voice: +1.647.777.2635, Internal: (81) 32635

--
Linux-audit mailing list
Linux-audit@redhat.com
https://listman.redhat.com/mailman/listinfo/linux-audit