Re: [RFC PATCH v2 00/13] Namespace for namespace roots

From: Jarkko Sakkinen

Date: Wed Nov 18 2026 - 10:47:05 EST


On Wed, Nov 18, 2026 at 05:39:47PM +0200, Jarkko Sakkinen wrote:
> On Wed, Nov 18, 2026 at 05:28:10PM +0200, Jarkko Sakkinen wrote:
> > On Wed, Nov 18, 2026 at 02:47:34PM +0000, Josef Bacik wrote:
> > > On Wed, Nov 18, 2026 at 01:13:39AM +0200, Jarkko Sakkinen wrote:
> > > > OVERVIEW
> > > > ========
> > > >
> > > > A root namespace (rootns) is a kernel object backed by a file descriptor
> > > > that binds together:
> > > >
> > > > 1. A full or partial set of detached namespaces.
> > > > 2. A dedicated PID namespace for spawning processes into, signaling
> > > > members, and polling or waiting on PID 1 exit.
> > > > 3. An initially empty mount namespace (CLONE_EMPTY_MNTNS) that must be
> > > > populated before going live, without inheriting host mount-tree
> > > > ancestors.
> > >
> > > This is what cells is doing https://lwn.net/Articles/1102593/, you may
> > > want to take a look at that work. Christian, Lennart, Alex, myself
> > > and others are working towards this in a more wholistic way. There's a
> > > lot more that goes into this outside of just the object itself. I've got
> > > my version deployed inside Anthropic for securing containers and I've
> > > tripped over a lot of pre-work that needs to come because we can tackle
> > > this. Alex has my rough outline for kuid64 for example, which is
> > > necessary for making sure we don't overlap UID's on the system and
> > > create holes this way. The cells work/anything that wants to have clear
> > > security boundaries is going to require that as a pre-requisite. If
> > > you'd like I can include you in our planning so we can get something
> > > that will solve all of our problems. Thanks,
> > >
> > > Josef
> >
> > No it's not doing the same, and this why I sent the patch set.
> >
> > I don't see cells fighting with rootns and given that rootns has a small
> > surface and it leverages already battle-tested code, instead of
> > reinventing the wheel, it also has benefits. It neither doubles the
> > number of container frameworks in kernel.
> >
> > What I'm not claiming is that this is better. I just wanted to fact
> > correct "this is what cells is doing". It is not true. It could be
> > better solution but still that statement is untrue.
>
> For better or worse, cells does a lot more; just looking at the patch set
> size.
>
> For rootns one can point out with a finger its possible problems BUT here's
> the thing: it does not haven unknown unknowns. Even if cells had less
> of known problems; we have zero idea what will come up if that ever
> lands. The possible consequence have less predictability.
>
> Also, as per integration, they actually are not fighting features, as
> they are in different layers. There's no same thing at all.
>

Yeah, and there is no obvious causality either that would result a
zero sum game. Rootns is something OCI runtime, any of them, can
quickly adapt to the existing framework. Cells takes longer, but
could be very worth of price. It's called "defense in depth".

Br, Jarkko