Re: [RFC PATCH v4 07/14] fork: Support CLONE_INTO_ROOTNS in clone3()

From: Jarkko Sakkinen

Date: Fri Nov 20 2026 - 05:33:48 EST


On Fri, Nov 20, 2026 at 12:43:05AM +0200, Jarkko Sakkinen wrote:
> Multithreaded userspace daemons cannot call setns(CLONE_NEWUSER) or
> safely mutate process-wide filesystem state in place before spawning a
> child into a target namespace set.
>
> Add CLONE_INTO_ROOTNS and struct clone_args.rootns (CLONE_ARGS_SIZE_VER3)
> to spawn a child process directly into a rootns whose root mount has
> been installed. Transition the rootns from ROOTNS_INIT through
> ROOTNS_FORKING to ROOTNS_RUNNING for the first process so it becomes
> PID 1, serialize concurrent entrants on waitq until bootstrap resolves,
> and admit subsequent callers into the running PID namespace until
> teardown begins.
>
> Derive child credentials from current, require CAP_SYS_ADMIN and
> CAP_SYS_CHROOT in the target user namespace and an ancestor active PID
> namespace, preserve caller umask, reset dumpability across user namespace
> transitions, commit time namespace vvar mappings after copy_process()
> succeeds, and look up the live mount namespace root. Restrict companion
> flags to CLONE_VFORK, CLONE_PIDFD, CLONE_PIDFD_AUTOKILL, CLONE_AUTOREAP,
> CLONE_CLEAR_SIGHAND, CLONE_INTO_CGROUP, and CLONE_NNP.
>
> Co-developed-by: David Howells <dhowells@xxxxxxxxxx>
> Signed-off-by: David Howells <dhowells@xxxxxxxxxx>
> Signed-off-by: Jarkko Sakkinen <jarkko@xxxxxxxxxx>

Being fd-bound object the gist is that the first clone3() binds
cgroups, which could e.g., different process than who build the
container.

CLONE_INTO_CGROUPS is of course available as an optional flag when
using CLONE_INTO_ROOTNS.

Br, Jarkko