Re: [PATCH] sys_getppid: add missing rcu_dereference

["Paul E. McKenney" <paulmck@linux.vnet.ibm.com>]

On Wed, Dec 07, 2011 at 04:24:20PM -0800, Mandeep Singh Baines wrote:
> Kees Cook (keescook@chromium.org) wrote:
> > On Tue, Dec 6, 2011 at 6:45 PM, Mandeep Singh Baines <msb@chromium.org> wrote:
> > > In order to safely dereference current->real_parent inside an
> > > rcu_read_lock, we need an rcu_dereference.
> > >
> > > Signed-off-by: Mandeep Singh Baines <msb@chromium.org>
> > > Cc: Thomas Gleixner <tglx@linutronix.de>
> > > Cc: Pavel Emelyanov <xemul@openvz.org>
> > > Cc: Oleg Nesterov <oleg@redhat.com>
> > > Cc: Andrew Morton <akpm@linux-foundation.org>
> > > Cc: Kees Cook <keescook@chromium.org>
> > > ---
> > >  kernel/timer.c |    2 +-
> > >  1 files changed, 1 insertions(+), 1 deletions(-)
> > >
> > > diff --git a/kernel/timer.c b/kernel/timer.c
> > > index dbaa624..9c3c62b 100644
> > > --- a/kernel/timer.c
> > > +++ b/kernel/timer.c
> > > @@ -1368,7 +1368,7 @@ SYSCALL_DEFINE0(getppid)
> > >        int pid;
> > >
> > >        rcu_read_lock();
> > > -       pid = task_tgid_vnr(current->real_parent);
> > > +       pid = task_tgid_vnr(rcu_dereference(current->real_parent));
> > >        rcu_read_unlock();
> > >
> > >        return pid;
> > 
> > Should parent and real_parent also be marked in sched.h with __rcu so
> > sparse can find other missing rcu_dereference()s? And if not, why?
> > (tasklist lock?)
> > 
> 
> Good idea. I was thinking the same thing. There's gotta be some way of
> finding these bugs via lockdep or sparse.

Indeed there is!  There is a CONFIG_SPARSE_RCU_POINTER kernel parameter
that tells sparse to check for proper use of RCU-protected pointers.
For this to work, the RCU-protected pointer in question must be marked
with "__rcu".  Sparse will then complain if the pointer is accessed
without using one of the rcu_dereference() family of functions.

> One idea I had was to create a ->real_parent accessor and insert an
> rcu_dereference_check there. That way lockdep could catch these bugs.
> 
> static inline struct task_struct *task_real_parent(struct task_struct *task)
> {
>         return rcu_dereference_check(task->real_parent,
> 			             lockdep_is_held(&tasklist_lock));
> }

The above is useful for common code that might be called from both
RCU readers (where rcu_read_lock() is in effect) and updaters
(which hold tasklist_lock).  But although a task_real_parent()-style
function can be extremely useful, it will not catch cases where
rcu_dereference() was not used but should have been.  For that,
as noted above, we have CONFIG_SPARSE_RCU_POINTER. __rcu, and sparse.

							Thanx, Paul

> > I think I see at least are few other users (security/apparmor/audit.c,
> > security/tomoyo/common.h, kernel/sched.c) that need rcu_dereference()
> > when accessing real_parent, there are probably more.
> > 
> > -Kees
> > 
> > -- 
> > Kees Cook
> > ChromeOS Security
>