Received-SPF: pass (google.com: domain of linux-kernel+bounces-95561-linux.lists.archive=gmail.com@vger.kernel.org designates 2604:1380:40f1:3f00::1 as permitted sender) client-ip=2604:1380:40f1:3f00::1;
From: Uladzislau Rezki <urezki@gmail.com>
Date: Thu, 7 Mar 2024 13:57:10 +0100
To: Joel Fernandes <joel@joelfernandes.org>
Cc: "Uladzislau Rezki (Sony)" <urezki@gmail.com>,
	"Paul E . McKenney" <paulmck@kernel.org>, RCU <rcu@vger.kernel.org>,
	Neeraj upadhyay <Neeraj.Upadhyay@amd.com>,
	Boqun Feng <boqun.feng@gmail.com>, Hillf Danton <hdanton@sina.com>,
	LKML <linux-kernel@vger.kernel.org>,
	Oleksiy Avramchenko <oleksiy.avramchenko@sony.com>,
	Frederic Weisbecker <frederic@kernel.org>
Subject: Re: [PATCH 1/2] rcu: Do not release a wait-head from a GP kthread
Message-ID: <Zem5pkNiAy-EZdo9@pc636>
References: <20240305195720.42687-1-urezki@gmail.com>
 <a1faf101-c689-4530-a9a5-c7f95b8825d6@joelfernandes.org>
Precedence: bulk
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <a1faf101-c689-4530-a9a5-c7f95b8825d6@joelfernandes.org>

On Wed, Mar 06, 2024 at 05:31:31PM -0500, Joel Fernandes wrote:
> 
> 
> On 3/5/2024 2:57 PM, Uladzislau Rezki (Sony) wrote:
> > Fix a below race by not releasing a wait-head from the
> > GP-kthread as it can lead for reusing it whereas a worker
> > can still access it thus execute newly added callbacks too
> > early.
> > 
> > CPU 0                              CPU 1
> > -----                              -----
> > 
> > // wait_tail == HEAD1
> > rcu_sr_normal_gp_cleanup() {
> >     // has passed SR_MAX_USERS_WAKE_FROM_GP
> >     wait_tail->next = next;
> >     // done_tail = HEAD1
> >     smp_store_release(&rcu_state.srs_done_tail, wait_tail);
> >     queue_work() {
> >         test_and_set_bit(WORK_STRUCT_PENDING_BIT, work_data_bits(work)
> >         __queue_work()
> >     }
> > }
> > 
> >                                set_work_pool_and_clear_pending()
> >                                rcu_sr_normal_gp_cleanup_work() {
> > // new GP, wait_tail == HEAD2
> > rcu_sr_normal_gp_cleanup() {
> >     // executes all completion, but stop at HEAD1
> >     wait_tail->next = HEAD1;
> >     // done_tail = HEAD2
> >     smp_store_release(&rcu_state.srs_done_tail, wait_tail);
> >     queue_work() {
> >         test_and_set_bit(WORK_STRUCT_PENDING_BIT, work_data_bits(work)
> >         __queue_work()
> >     }
> > }
> >                                  // done = HEAD2
> >                                  done = smp_load_acquire(&rcu_state.srs_done_tail);
> >                                  // head = HEAD1
> >                                  head = done->next;
> >                                  done->next = NULL;
> >                                  llist_for_each_safe() {
> >                                  // completes all callbacks, release HEAD1
> >                                  }
> >                                }
> >                                // Process second queue
> >                                set_work_pool_and_clear_pending()
> >                                rcu_sr_normal_gp_cleanup_work() {
> >                                // done = HEAD2
> >                                done = smp_load_acquire(&rcu_state.srs_done_tail);
> > 
> > // new GP, wait_tail == HEAD3
> > rcu_sr_normal_gp_cleanup() {
> >     // Finds HEAD2 with ->next == NULL at the end
> >     rcu_sr_put_wait_head(HEAD2)
> >     ...
> > 
> > // A few more GPs later
> > rcu_sr_normal_gp_init() {
> >      HEAD2 = rcu_sr_get_wait_head();
> >      llist_add(HEAD2, &rcu_state.srs_next);
> >                                // head == rcu_state.srs_next
> >                                head = done->next;
> >                                done->next = NULL;
> >                                llist_for_each_safe() {
> >                                 // EXECUTE CALLBACKS TOO EARLY!!!
> >                                 }
> >                                }
> > 
> > Reported-by: Frederic Weisbecker <frederic@kernel.org>
> > Fixes: 05a10b921000 ("rcu: Support direct wake-up of synchronize_rcu() users")
> > Signed-off-by: Uladzislau Rezki (Sony) <urezki@gmail.com>
> > ---
> >  kernel/rcu/tree.c | 22 ++++++++--------------
> >  1 file changed, 8 insertions(+), 14 deletions(-)
> > 
> > diff --git a/kernel/rcu/tree.c b/kernel/rcu/tree.c
> > index 31f3a61f9c38..475647620b12 100644
> > --- a/kernel/rcu/tree.c
> > +++ b/kernel/rcu/tree.c
> > @@ -1656,21 +1656,11 @@ static void rcu_sr_normal_gp_cleanup(void)
> >  	WARN_ON_ONCE(!rcu_sr_is_wait_head(wait_tail));
> >  
> >  	/*
> > -	 * Process (a) and (d) cases. See an illustration. Apart of
> > -	 * that it handles the scenario when all clients are done,
> > -	 * wait-head is released if last. The worker is not kicked.
> > +	 * Process (a) and (d) cases. See an illustration.
> >  	 */
> >  	llist_for_each_safe(rcu, next, wait_tail->next) {
> > -		if (rcu_sr_is_wait_head(rcu)) {
> > -			if (!rcu->next) {
> > -				rcu_sr_put_wait_head(rcu);
> > -				wait_tail->next = NULL;
> > -			} else {
> > -				wait_tail->next = rcu;
> > -			}
> > -
> > +		if (rcu_sr_is_wait_head(rcu))
> >  			break;
> > -		}
> >  
> >  		rcu_sr_normal_complete(rcu);
> >  		// It can be last, update a next on this step.
> > @@ -1684,8 +1674,12 @@ static void rcu_sr_normal_gp_cleanup(void)
> >  	smp_store_release(&rcu_state.srs_done_tail, wait_tail);
> >  	ASSERT_EXCLUSIVE_WRITER(rcu_state.srs_done_tail);
> >  
> > -	if (wait_tail->next)
> > -		queue_work(system_highpri_wq, &rcu_state.srs_cleanup_work);
> > +	/*
> > +	 * We schedule a work in order to perform a final processing
> > +	 * of outstanding users(if still left) and releasing wait-heads
> > +	 * added by rcu_sr_normal_gp_init() call.
> > +	 */
> > +	queue_work(system_highpri_wq, &rcu_state.srs_cleanup_work);
> >  }
> 
> Ah, nice. So instead of allocating/freeing in GP thread and freeing in worker,
> you allocate heads only in GP thread and free them only in worker, thus
> essentially fixing the UAF that Frederick found.
> 
> AFAICS, this fixes the issue.
> 
> Reviewed-by: Joel Fernandes (Google) <joel@joelfernandes.org>
> 
Thank you for the review-by!

> There might a way to prevent queuing new work as fast-path optimization, incase
> the CBs per GP will always be < SR_MAX_USERS_WAKE_FROM_GP but I could not find a
> workqueue API that helps there, and work_busy() has comments saying not to use that.
> 
This is not really critical but yes, we can think of it.

--
Uladzislau Rezki