Re: [PATCH] rcu: Safe access to rcu_node structure's->exp_tasks

["Paul E. McKenney" <paulmck@kernel.org>]

On Mon, Jan 09, 2023 at 02:21:01PM +0100, Frederic Weisbecker wrote:
> On Thu, Jan 05, 2023 at 07:41:46PM -0800, Paul E. McKenney wrote:
> > On Sat, Dec 24, 2022 at 01:25:53PM +0800, Zqiang wrote:
> > > For kernels built with CONFIG_PREEMPT_RCU=y, the following scenario
> > > can result system oops.
> > > 
> > >            CPU1                                           CPU2
> > > rcu_preempt_deferred_qs_irqrestore                rcu_print_task_exp_stall
> > >   if (special.b.blocked)                            READ_ONCE(rnp->exp_tasks) != NULL
> > >     raw_spin_lock_rcu_node
> > >     np = rcu_next_node_entry(t, rnp)
> > >     if (&t->rcu_node_entry == rnp->exp_tasks)
> > >       WRITE_ONCE(rnp->exp_tasks, np)
> > >       ....
> > >       raw_spin_unlock_irqrestore_rcu_node
> > >                                                     raw_spin_lock_irqsave_rcu_node
> > >                                                     t = list_entry(rnp->exp_tasks->prev,
> > >                                                         struct task_struct, rcu_node_entry)
> > >                                                     (if rnp->exp_tasks is NULL
> > >                                                        will trigger oops)
> > > 
> > > This problem is that CPU2 accesses rcu_node structure's->exp_tasks
> > > without holding the rcu_node structure's ->lock and CPU2 did not
> > > observe CPU1's change to rcu_node structure's->exp_tasks in time,
> > > if rcu_node structure's->exp_tasks is set null pointer by CPU1, after
> > > that CPU2 accesses members of rcu_node structure's->exp_tasks will
> > > trigger oops.
> > > 
> > > This commit therefore allows rcu_node structure's->exp_tasks to be
> > > accessed while holding rcu_node structure's ->lock.
> > > 
> > > Signed-off-by: Zqiang <qiang1.zhang@intel.com>
> > 
> > Apologies for the delay and thank you for the reminder!
> > 
> > Please check the wordsmithed version below, which I have queued.
> > 
> > 						Thanx, Paul
> > 
> > ------------------------------------------------------------------------
> > 
> > commit 389b0eafd72829fd63548f7ff4e8d6ac90fa1f98
> > Author: Zqiang <qiang1.zhang@intel.com>
> > Date:   Sat Dec 24 13:25:53 2022 +0800
> > 
> >     rcu: Protect rcu_print_task_exp_stall() ->exp_tasks access
> >     
> >     For kernels built with CONFIG_PREEMPT_RCU=y, the following scenario can
> >     result in a NULL-pointer dereference:
> >     
> >                CPU1                                           CPU2
> >     rcu_preempt_deferred_qs_irqrestore                rcu_print_task_exp_stall
> >       if (special.b.blocked)                            READ_ONCE(rnp->exp_tasks) != NULL
> >         raw_spin_lock_rcu_node
> >         np = rcu_next_node_entry(t, rnp)
> >         if (&t->rcu_node_entry == rnp->exp_tasks)
> >           WRITE_ONCE(rnp->exp_tasks, np)
> >           ....
> >           raw_spin_unlock_irqrestore_rcu_node
> >                                                         raw_spin_lock_irqsave_rcu_node
> >                                                         t = list_entry(rnp->exp_tasks->prev,
> >                                                             struct task_struct, rcu_node_entry)
> >                                                         (if rnp->exp_tasks is NULL, this
> >                                                            will dereference a NULL pointer)
> >     
> >     The problem is that CPU2 accesses the rcu_node structure's->exp_tasks
> >     field without holding the rcu_node structure's ->lock and CPU2 did
> >     not observe CPU1's change to rcu_node structure's ->exp_tasks in time.
> >     Therefore, if CPU1 sets rcu_node structure's->exp_tasks pointer to NULL,
> >     then CPU2 might dereference that NULL pointer.
> >     
> >     This commit therefore holds the rcu_node structure's ->lock while
> >     accessing that structure's->exp_tasks field.
> >     
> >     Signed-off-by: Zqiang <qiang1.zhang@intel.com>
> >     Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
> > 
> > diff --git a/kernel/rcu/tree_exp.h b/kernel/rcu/tree_exp.h
> > index 7cc4856da0817..902e7c8709c7e 100644
> > --- a/kernel/rcu/tree_exp.h
> > +++ b/kernel/rcu/tree_exp.h
> > @@ -803,9 +803,11 @@ static int rcu_print_task_exp_stall(struct rcu_node *rnp)
> >  	int ndetected = 0;
> >  	struct task_struct *t;
> >  
> > -	if (!READ_ONCE(rnp->exp_tasks))
> > -		return 0;
> >  	raw_spin_lock_irqsave_rcu_node(rnp, flags);
> > +	if (!READ_ONCE(rnp->exp_tasks)) {
> 
> Does it have to be READ_ONCE then?

Good point, that should not be necessary.  I will drop the READ_ONCE on
my next rebase.  (Unless someone tells me there is something subtle that
I am missing.)

							Thanx, Paul

> Thanks.
> 
> > +		raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
> > +		return 0;
> > +	}
> >  	t = list_entry(rnp->exp_tasks->prev,
> >  		       struct task_struct, rcu_node_entry);
> >  	list_for_each_entry_continue(t, &rnp->blkd_tasks, rcu_node_entry) {