Re: [PATCH 1/6] user namespace: make signal.c respect user namespaces (v4)

["Serge E. Hallyn" <serge.hallyn@canonical.com>]

Quoting Andrew Morton (akpm@linux-foundation.org):
> On Fri,  4 Nov 2011 22:24:37 +0000
> Serge Hallyn <serge@hallyn.com> wrote:
> 
> > +static inline void fixup_uid(struct siginfo *info, struct task_struct *t)
> > +{
> > +#ifdef CONFIG_USER_NS
> > +	if (current_user_ns() == task_cred_xxx(t, user_ns))
> > +#endif
> > +		return;
> > +
> > +	if (SI_FROMKERNEL(info))
> > +		return;
> > +
> > +	info->si_uid = user_ns_map_uid(task_cred_xxx(t, user_ns),
> > +					current_cred(), info->si_uid);
> > +}
> 
> err, this function is a no-op if CONFIG_USER_NS=n.  If that was
> intentional then why on earth do this in such a weird fashion?  If

(I didn't write it, but)  I think the assumption was that the compiler
would optimize it all away if !CONFIG_USER_NS, and in this way the
call site didn't need to be obscured with the #ifdef.  Is there a way
that's better?  Would you prefer something like:

+#ifdef CONFIG_USER_NS
+static inline void fixup_uid(struct siginfo *info, struct task_struct *t)
+{
+	if (current_user_ns() == task_cred_xxx(t, user_ns))
+		return;
+
+	if (SI_FROMKERNEL(info))
+		return;
+
+	info->si_uid = user_ns_map_uid(task_cred_xxx(t, user_ns),
+					current_cred(), info->si_uid);
+}
+#else
+static inline void fixup_uid(struct siginfo *info, struct task_struct *t)
+{
+		return;
+}
+#endif

?  It's less sneaky at least.

> unintentional then it makes me wonder how well tested all this was with
> CONFIG_USER_NS=n?
> 
> I vaguely remember that I've forgotten how all this stuff works.  Some
> additional review input would be nice (cough-oleg-cough).
>