linux-kernel - [PATCH 07/11] rcu: set special.b.deferred_qs before wake

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-Id: <20191031100806.1326-8-laijs@linux.alibaba.com>
Date:   Thu, 31 Oct 2019 10:08:02 +0000
From:   Lai Jiangshan <laijs@...ux.alibaba.com>
To:     linux-kernel@...r.kernel.org
Cc:     Lai Jiangshan <laijs@...ux.alibaba.com>,
        "Paul E. McKenney" <paulmck@...nel.org>,
        Josh Triplett <josh@...htriplett.org>,
        Steven Rostedt <rostedt@...dmis.org>,
        Mathieu Desnoyers <mathieu.desnoyers@...icios.com>,
        Lai Jiangshan <jiangshanlai@...il.com>,
        Joel Fernandes <joel@...lfernandes.org>, rcu@...r.kernel.org
Subject: [PATCH 07/11] rcu: set special.b.deferred_qs before wake_up()

The patch 10f39bb1b2c1
(rcu: protect __rcu_read_unlock() against scheduler-using irq handlers)
unveiled a kind of deadlock and resolved the deadlock problem by
avoiding the condition when ->rcu_read_lock_nesting is zero &&
->rcu_read_unlock_special is non-zero. To achieve it, the commit
used negative values for ->rcu_read_lock_nesting.

But now we have deferred_qs mechanism, we can defer qs rather
than persevere in reporting qs and deadlock. All we need is
setting special.b.deferred_qs before scheduler locks
such as wake_up() and leave the qs deferred and return.

After this change, rcu_read_unlock_special() is safe to be
called in any context, including nested in __rcu_read_unlock()
in interrupt.

This change is important to change ->rcu_read_lock_nesting
back to non-negative and further simplify the rcu_read_unlock().

Signed-off-by: Lai Jiangshan <laijs@...ux.alibaba.com>
---
 kernel/rcu/tree_plugin.h | 15 ++++++++++++---
 1 file changed, 12 insertions(+), 3 deletions(-)

diff --git a/kernel/rcu/tree_plugin.h b/kernel/rcu/tree_plugin.h
index e612c77dc446..dbded2b8c792 100644
--- a/kernel/rcu/tree_plugin.h
+++ b/kernel/rcu/tree_plugin.h
@@ -591,6 +591,7 @@ static void rcu_read_unlock_special(struct task_struct *t)
 	irqs_were_disabled = irqs_disabled_flags(flags);
 	if (preempt_bh_were_disabled || irqs_were_disabled) {
 		bool exp;
+		bool deferred_qs = t->rcu_read_unlock_special.b.deferred_qs;
 		struct rcu_data *rdp = this_cpu_ptr(&rcu_data);
 		struct rcu_node *rnp = rdp->mynode;
 
@@ -599,9 +600,18 @@ static void rcu_read_unlock_special(struct task_struct *t)
 		      (rdp->grpmask & rnp->expmask) ||
 		      tick_nohz_full_cpu(rdp->cpu);
 		// Need to defer quiescent state until everything is enabled.
+		// In some cases when in_interrupt() returns false,
+		// raise_softirq_irqoff() has to call wake_up(),
+		// and the !deferred_qs says that scheduler locks
+		// cannot be held, so the wakeup will be safe now.
+		// But this wake_up() may have RCU critical section nested
+		// in the scheduler locks and its rcu_read_unlock() would
+		// call rcu_read_unlock_special() and then wake_up()
+		// recursively and deadlock if deferred_qs is still false.
+		// To avoid it, deferred_qs has to be set beforehand.
+		t->rcu_read_unlock_special.b.deferred_qs = true;
 		if (irqs_were_disabled && use_softirq &&
-		    (in_interrupt() ||
-		     (exp && !t->rcu_read_unlock_special.b.deferred_qs))) {
+		    (in_interrupt() || (exp && !deferred_qs))) {
 			// Using softirq, safe to awaken, and we get
 			// no help from enabling irqs, unlike bh/preempt.
 			raise_softirq_irqoff(RCU_SOFTIRQ);
@@ -620,7 +630,6 @@ static void rcu_read_unlock_special(struct task_struct *t)
 				irq_work_queue_on(&rdp->defer_qs_iw, rdp->cpu);
 			}
 		}
-		t->rcu_read_unlock_special.b.deferred_qs = true;
 		local_irq_restore(flags);
 		return;
 	}
-- 
2.20.1