Re: [RFC PATCH v5 14/25] rootns: Add ROOTNS_WAIT ioctl
From: Jarkko Sakkinen
Date: Fri Nov 20 2026 - 19:18:46 EST
On Sat, Nov 21, 2026 at 02:17:20AM +0200, Jarkko Sakkinen wrote:
> On Sat, Nov 21, 2026 at 02:08:33AM +0200, Jarkko Sakkinen wrote:
> > wait4() and waitid() only allow a parent or subreaper to collect exit
> > status, so an external manager holding a rootns file descriptor cannot
> > wait on a rootns init process forked by another process or observe exit
> > status multiple times across observers.
> >
> > Add the ROOTNS_WAIT ioctl and struct rootns_wait_args to wait freezably
> > and interruptibly on a rootns file descriptor until its init thread group
> > exits and report the recorded wait status. Return the init TGID
> > translated into the caller's active PID namespace, 0 for WNOHANG while
> > running, or -ECHILD if init was never started or is not visible to the
> > caller.
> >
> > Co-developed-by: David Howells <dhowells@xxxxxxxxxx>
> > Signed-off-by: David Howells <dhowells@xxxxxxxxxx>
> > Signed-off-by: Jarkko Sakkinen <jarkko@xxxxxxxxxx>
> > ---
> > include/uapi/linux/rootns.h | 26 +++++++++++++++++++++
> > kernel/rootns.c | 46 +++++++++++++++++++++++++++++++++++++
> > 2 files changed, 72 insertions(+)
> >
> > diff --git a/include/uapi/linux/rootns.h b/include/uapi/linux/rootns.h
> > index bd643bdfb43c..3fd215185398 100644
> > --- a/include/uapi/linux/rootns.h
> > +++ b/include/uapi/linux/rootns.h
> > @@ -8,6 +8,7 @@
> > #define _UAPI_LINUX_ROOTNS_H
> >
> > #include <linux/ioctl.h>
> > +#include <linux/types.h>
> >
> > /**
> > * enum rootns_flags - Flags for rootns()
> > @@ -24,6 +25,16 @@ enum rootns_flags {
> > ROOTNS_ALL_FLAGS = 0x00000003,
> > };
> >
> > +/**
> > + * struct rootns_wait_args - Arguments for %ROOTNS_WAIT
> > + * @flags: Wait options (0 or %WNOHANG).
> > + * @status: Output wait status of rootns init in waitpid(2) format.
> > + */
> > +struct rootns_wait_args {
> > + __u32 flags;
> > + __s32 status;
> > +};
> > +
> > #define ROOTNS_IOCTL_MAGIC 0xB9
> >
> > /**
> > @@ -41,4 +52,19 @@ enum rootns_flags {
> > */
> > #define ROOTNS_GET_NSFD _IO(ROOTNS_IOCTL_MAGIC, 0x01)
> >
> > +/**
> > + * DOC: ROOTNS_WAIT
> > + *
> > + * ``ioctl(rootnsfd, ROOTNS_WAIT, &args)`` waits interruptibly (unless
> > + * ``WNOHANG`` is set in ``args.flags``) until ``rootnsfd`` reaches
> > + * ``ROOTNS_DEAD``, then writes the recorded exit status of ``rootns`` init
> > + * into ``args.status``.
> > + *
> > + * Returns the caller-visible TGID of ``rootns`` init on success, ``0`` if
> > + * ``WNOHANG`` is set and the ``rootns`` has not terminated, or ``-1`` with
> > + * ``errno`` set to ``ECHILD`` if init was never started or is not visible in
> > + * the caller's active PID namespace.
> > + */
> > +#define ROOTNS_WAIT _IOWR(ROOTNS_IOCTL_MAGIC, 0x02, struct rootns_wait_args)
> > +
> > #endif /* _UAPI_LINUX_ROOTNS_H */
> > diff --git a/kernel/rootns.c b/kernel/rootns.c
> > index 5d59fc9a591d..cb74ddc8363c 100644
> > --- a/kernel/rootns.c
> > +++ b/kernel/rootns.c
> > @@ -117,6 +117,47 @@ static long rootns_get_nsfd(struct rootns *rootns, unsigned long nstype)
> > return nsproxy_open_namespace(rootns->ns, rootns->user_ns, nstype);
> > }
> >
> > +static long rootns_wait(struct rootns *rootns, void __user *uarg)
> > +{
> > + struct pid_namespace *pid_ns = task_active_pid_ns(current);
> > + struct rootns_wait_args args;
> > + long ret;
> > +
> > + if (!uarg)
> > + return -EINVAL;
> > + if (copy_from_user(&args, uarg, sizeof(args)))
> > + return -EFAULT;
> > + if (args.flags & ~WNOHANG)
> > + return -EINVAL;
> > + if (!pidns_is_ancestor(rootns->ns->pid_ns_for_children, pid_ns))
> > + return -ECHILD;
> > + if (READ_ONCE(rootns->state) == ROOTNS_INIT)
> > + return -ECHILD;
> > +
> > + if (!(args.flags & WNOHANG)) {
> > + ret = wait_event_freezable(rootns->waitq,
> > + READ_ONCE(rootns->state) == ROOTNS_DEAD ||
> > + READ_ONCE(rootns->state) == ROOTNS_INIT);
> > + if (ret)
> > + return ret;
> > + }
> > +
> > + scoped_guard(spinlock, &rootns->lock) {
> > + if (rootns->state == ROOTNS_INIT)
> > + return -ECHILD;
> > + if (rootns->state != ROOTNS_DEAD)
> > + return 0;
> > + ret = pid_nr_ns(rootns->init_pid, pid_ns);
> > + if (ret <= 0)
> > + return -ECHILD;
> > + args.status = rootns->exit_code;
> > + }
> > +
> > + if (copy_to_user(uarg, &args, sizeof(args)))
> > + return -EFAULT;
> > + return ret;
> > +}
> > +
> > static long rootns_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
> > {
> > struct rootns *rootns = file->private_data;
> > @@ -124,6 +165,8 @@ static long rootns_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
> > switch (cmd) {
> > case ROOTNS_GET_NSFD:
> > return rootns_get_nsfd(rootns, arg);
> > + case ROOTNS_WAIT:
> > + return rootns_wait(rootns, (void __user *)arg);
> > default:
> > return -ENOIOCTLCMD;
> > }
> > @@ -134,6 +177,8 @@ static long rootns_compat_ioctl(struct file *file, unsigned int cmd,
> > unsigned long arg)
> > {
> > switch (cmd) {
> > + case ROOTNS_WAIT:
> > + return rootns_ioctl(file, cmd, (unsigned long)compat_ptr(arg));
> > case ROOTNS_GET_NSFD:
> > return rootns_ioctl(file, cmd, arg);
> > default:
> > @@ -328,3 +373,4 @@ SYSCALL_DEFINE1(rootns, unsigned int, flags)
> >
> > return fd;
> > }
> > +
> > --
> > 2.47.3
> >
>
> Ought to be dropped but seemed too irrelevant detail for the time
> being.
I mean replaced with a syscall path.
Br, Jarkko