[REVIEW][PATCH 3/5] pidns: Don't have unshare(CLONE_NEWPID) imply CLONE_THREAD

[Serge E. Hallyn]

Quoting Eric W. Biederman (ebiederm at xmission.com):
> "Serge E. Hallyn" <serge at hallyn.com> writes:
> 
> > Quoting Eric W. Biederman (ebiederm at xmission.com):
> >> 
> >> I goofed when I made unshare(CLONE_NEWPID) only work in a
> >> single-threaded process.  There is no need for that requirement and in
> >> fact I analyzied things right for setns.  The hard requirement
> >> is for tasks that share a VM to all be in the pid namespace and
> >> we properly prevent that in do_fork.
> >
> > I don't understand though - copy_process does have the right test:
> >
> >    1176          * If the new process will be in a different pid namespace
> >    1177          * don't allow the creation of threads.
> >    1178          */
> >    1179         if ((clone_flags & (CLONE_VM|CLONE_NEWPID)) &&
> >    1180             (task_active_pid_ns(current) != current->nsproxy->pid_ns))
> >    1181                 return ERR_PTR(-EINVAL);
> >
> > but why is it ok for sys_unshare not to do that?  Note that
> > in order for check_unshare_flags() to bail on &current->mm->mm_users > 1
> > you do have to set CLONE_VM (for inverse interpretation).
> >
> > So it seems to me this isn't safe as is, and we need to at least
> > set CLONE_VM if CLONE_PID is set.
> 
> Partly this is the difference in the meaning of the flags between
> unshare and clone.
> 
> Basically in unshare all othat gets changed is
> current->nsproxy->pid_ns_for_children (the rename is in the net tree).

D'oh, right.  Thanks!

Acked-by: Serge Hallyn <serge.hallyn at canonical.com>


> So because unshare of the pid namespace does not actually effect the
> current processes, just the pid namespace the children of the current
> thread will be in this is safe.
> 
> And frankly having the checks be obviously different is a good thing
> because it means that people will ask why in the world this is so and
> realize the difference in meaning.
> 
> Eric