[Devel] Re: The issues for agreeing on a virtualization/namespaces implementation.

[Serge E. Hallyn]

Quoting Eric W. Biederman (ebiederm at xmission.com):
> Alexey Kuznetsov <kuznet at ms2.inr.ac.ru> writes:
> 
> > Hello!
> >
> >> >2) What is the syscall interface to create these namespaces?
> >> >   - Do we add clone flags?  
> >> >     (Plan 9 style)
> >> 
> >> Like that approach .. flexible .. particular when one has well specified 
> >> namespaces.
> >> 
> >> >   - Do we add a syscall (similar to setsid) per namespace?
> >> >     (Traditional unix style)?
> >> 
> >> Where does that approach end .. what's wrong with doing it at clone() time ?
> >
> > That most of those namespaces need a special setup rather than a plain copy?
> >
> > F.e. what are you going to do with NETWORK namespace? The only valid thing
> > to do is to prepare a new context and to configure its content (addresses,
> > routing tables, iptables...) later. So that, in this case it is natural
> > to inherit the context through clone() and to create new context
> > with a separate syscall.
> 
> With a NETWORK namespace what I implemented was that you get a empty
> namespace with a loopback interface.
> 
> But setting up the namespace from the inside is clearly the sane thing
> todo.

What I tried to do in a proof of concept long ago was to have
CLONE_NETNS mean that you get access to all the network devices, but
then you could drop/add them.  Conceptually I prefer that to getting an
empty namespace, but I'm not sure whether there's any practical use
where you'd want that...

-serge