Re: [PATCH 2/4] cgroup: add bpf hook for attach

[Christian Brauner <brauner@kernel.org>]

On Fri, Feb 20, 2026 at 05:16:13AM -1000, Tejun Heo wrote:
> Hello,
> 
> On Fri, Feb 20, 2026 at 01:38:30AM +0100, Christian Brauner wrote:
> > Add a hook to manage attaching tasks to cgroup. I'm in the process of
> > adding various "universal truth" bpf programs to systemd that will make
> > use of this.
> > 
> > This has been a long-standing request (cf. [1] and [2]). It will allow us to
> > enforce cgroup migrations and ensure that services can never escape their
> > cgroups. This is just one of many use-cases.
> 
> >From cgroup POV, this looks fine to me but I'm curious whether something
> dumber would also work. With CLONE_INTO_CGROUP, cgroup migration isn't
> necessary at all. Would something dumber like a mount option disabling
> cgroup migrations completely work too or would that be too restrictive?

It would be too restrictive. I've played with various policies. For
example, a small set of tasks (like PID 1 or the session manager) are
allowed to move processes between cgroups (detectable via e.g., xattrs).
No other task is allowd. But that's already too restrictive because it
fscks over delegated subcgroups were tasks need to be moved around
(container managers etc.). IOW, any policy must be quite modular and
dynamic so a simple mount option wouldn't cover it.

As a sidenote, there would be other mount options that would be useful
but that currently aren't that easy to support/implement because of the
way cgroupfs (for historical reasons ofc) is architected where it shares
a single superblock.

I have a series (from quite some time ago) that makes cgroupfs truly
multi-instance. It would effectively behave just like tmpfs does. A new
mount gets you a new superblock. But once you have that you can e.g.,
simplify cgroup namespaces as well. I've done that work originally to
support idmapped mounts with cgroupfs but I can't find that branch
anymore.