Re: [RFC PATCH v4 00/14] Namespace for namespace roots

From: Jarkko Sakkinen

Date: Fri Nov 20 2026 - 05:10:59 EST


On Fri, Nov 20, 2026 at 07:58:10AM +0200, Jarkko Sakkinen wrote:
> I've now re-sliced these into a bit more sane pieces:
>
> $ git log --oneline master..
> d8e24a6bc14e (HEAD -> rootns, origin/rootns) MAINTAINERS: Add entry for ROOT NAMESPACE
> f2569e56a267 selftests/rootns: Add root namespace test suite
> bea05e758f8e docs: Document root namespace uAPI
> dd05a49aff77 arch: Wire up rootns() system call
> 2dee7a5fddec bpf: Register rootns LSM hooks
> 73c51e02b830 security: Add rootns LSM hooks
> 3cfd6576a037 mount: Support MOVE_MOUNT_T_MNTNS in move_mount()
> 9ed7387038ea ipc/mqueue: Implement ->set_rootns() fs_context operation
> 624f0ed9c1c7 proc: Implement ->set_rootns() fs_context operation
> 1085e6ae2eb3 fs: Support FSCONFIG_SET_ROOTNS in fsconfig()
> d2898dbdc39b fork: Support CLONE_INTO_ROOTNS in clone3()
> 42dbccb5890e rootns: Add ROOTNS_KILL ioctl
> 14cbda3e5bca rootns: Add ROOTNS_WAIT ioctl
> b2cc82539b75 rootns: Add rootns() system call
> d29215aa7f0f rootns: Track init lifecycle via copy_process() and pid_namespace
> 0f9ce7febb49 rootns: Add core infrastructure for root namespaces
> 7a0e8f969fc7 mount: Enforce mount_too_revealing() for foreign mntns
> 8c61aee3afbf mount: Add mnt_ns_get_root()
> 07d239d0287c userns: Factor out install_user_ns_cred()
> bf74588e706d nsfs: Add nsproxy_open_namespace()
> 645d31a1c516 nsproxy: Pass nsproxy to create_new_namespaces()

A "small" fact correction: rootns itself is not actually a namespace. It
was at some point during the course of doing something over alonger time
span :-)

I'll adjust the language and documentation.

It's like SGX enclave i.e., an isolated and opaque object with *nothing*
at Epoch. Similarly as enclaves there's a build phase for building the
memory environment, and a live phase that follows.

Those are the glasses that I've looked at the topic all the time; not
so much with namespace glasses, as it's not my main beef and actually
doing this has very little to do with namespaces and so much more on how
you manage and control the threads and memory.

This is more like cross-cutting feature with no obvious subsystem
ownership more than anything else.

Br, Jarkko