Re: [RFC] cgroups: implement device whitelist lsm (v2)

["Serge E. Hallyn" <serue@us.ibm.com>]

Quoting Paul Menage (menage@google.com):
> On Wed, Mar 12, 2008 at 8:27 PM, Serge E. Hallyn <serue@us.ibm.com> wrote:
> >
> >  While composing this with the ns_cgroup may seem logical, it is not
> >  the right thing to do, because updates to /cg/cg1/devcg.deny are
> >  not reflected in /cg/cg1/cg2/devcg.allow.
> 
> Maybe you should follow up the tree to ensure that all parent groups
> have access to the device too? Or alternatively, cache the results of
> this lookup whenever permissions for a device change?

Yes, I considered that.  Alternatively additions to a parent cgroup's
.deny could be propagated to all its descendents (but not additions to
the .allow).

I've noted this as something to add to the next version.

> >  A task may only be moved to another devcgroup if it is moving to
> >  a direct descendent of its current devcgroup.
> 
> What's the rationale for that?

To prevent it escaping to laxer device permissions, which of course only
makes sense if we do what you recommend above :)

> >  CAP_NS_OVERRIDE is defined as the capability needed to cross namespaces.
> >  A task needs both CAP_NS_OVERRIDE and CAP_SYS_ADMIN to create a new
> >  devcgroup, update a devcgroup's access, or move a task to a new
> >  devcgroup.
> 
> But this isn't necessarily crossing namespaces. It could be used for
> device control in the same namespace (e.g. allowing a job to access a
> raw disk for its data storage rather than going through the
> filesystem).

Yeah it should be renamed.  I want to use the same cap which we would
use for user namespaces though.  CAP_NS_CONT(ainer)?  Even though there
really is no such thing as a 'container'.  But that would tie together
any such privileges for cgroups and namespaces.

thanks,
-serge