netdev - [PATCH 2/6] net/rds: Re-factor and avoid superfluous queuing of reconnect work

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: <20250227042638.82553-3-allison.henderson@oracle.com>
Date: Wed, 26 Feb 2025 21:26:34 -0700
From: allison.henderson@...cle.com
To: netdev@...r.kernel.org
Subject: [PATCH 2/6] net/rds: Re-factor and avoid superfluous queuing of reconnect work

From: Håkon Bugge <haakon.bugge@...cle.com>

rds_conn_path_connect_if_down() queues reconnect work with a zero
delay, not checking if it is a passive or active connector. Re-factor
by calling rds_queue_reconnect(). The zero re-connect delay on the
passive side may cause a connect race.

Helper functions are added to conditionally queue reconnect and
properly clear the RDS_RECONNECT_PENDING bit.

The clearing of said bit is moved to the end of the
rds_connect_worker() function and by that also fixing a bug when
rds_tcp is used, where the bit was never cleared.

Signed-off-by: Håkon Bugge <haakon.bugge@...cle.com>
Signed-off-by: Gerd Rausch <gerd.rausch@...cle.com>
Signed-off-by: Allison Henderson <allison.henderson@...cle.com>
---
 net/rds/connection.c |  5 ++---
 net/rds/rds.h        | 15 +++++++++++++++
 net/rds/threads.c    |  9 +++++----
 3 files changed, 22 insertions(+), 7 deletions(-)

diff --git a/net/rds/connection.c b/net/rds/connection.c
index 1d80586fdda2..73de221bd7c2 100644
--- a/net/rds/connection.c
+++ b/net/rds/connection.c
@@ -913,9 +913,8 @@ void rds_conn_path_connect_if_down(struct rds_conn_path *cp)
 		rcu_read_unlock();
 		return;
 	}
-	if (rds_conn_path_state(cp) == RDS_CONN_DOWN &&
-	    !test_and_set_bit(RDS_RECONNECT_PENDING, &cp->cp_flags))
-		queue_delayed_work(rds_wq, &cp->cp_conn_w, 0);
+	if (rds_conn_path_state(cp) == RDS_CONN_DOWN)
+		rds_queue_reconnect(cp);
 	rcu_read_unlock();
 }
 EXPORT_SYMBOL_GPL(rds_conn_path_connect_if_down);
diff --git a/net/rds/rds.h b/net/rds/rds.h
index c9a22d0e887b..1fb27e1a2e46 100644
--- a/net/rds/rds.h
+++ b/net/rds/rds.h
@@ -794,6 +794,21 @@ void __rds_conn_path_error(struct rds_conn_path *cp, const char *, ...);
 	__rds_conn_path_error(cp, KERN_WARNING "RDS: " fmt)
 
 extern struct workqueue_struct *rds_wq;
+static inline void rds_cond_queue_reconnect_work(struct rds_conn_path *cp, unsigned long delay)
+{
+	if (!test_and_set_bit(RDS_RECONNECT_PENDING, &cp->cp_flags))
+		queue_delayed_work(rds_wq, &cp->cp_conn_w, delay);
+}
+
+static inline void rds_clear_reconnect_pending_work_bit(struct rds_conn_path *cp)
+{
+	/* clear_bit() does not imply a memory barrier */
+	smp_mb__before_atomic();
+	clear_bit(RDS_RECONNECT_PENDING, &cp->cp_flags);
+	/* clear_bit() does not imply a memory barrier */
+	smp_mb__after_atomic();
+}
+
 static inline void rds_cond_queue_send_work(struct rds_conn_path *cp, unsigned long delay)
 {
 	if (!test_and_set_bit(RDS_SEND_WORK_QUEUED, &cp->cp_flags))
diff --git a/net/rds/threads.c b/net/rds/threads.c
index eedae5653051..634e9f431fd6 100644
--- a/net/rds/threads.c
+++ b/net/rds/threads.c
@@ -153,8 +153,8 @@ void rds_queue_reconnect(struct rds_conn_path *cp)
 		 conn, &conn->c_laddr, &conn->c_faddr);
 	rcu_read_lock();
 	if (!rds_destroy_pending(cp->cp_conn))
-		queue_delayed_work(rds_wq, &cp->cp_conn_w,
-				   rand % cp->cp_reconnect_jiffies);
+		rds_cond_queue_reconnect_work(cp,
+					      rand % cp->cp_reconnect_jiffies);
 	rcu_read_unlock();
 
 	cp->cp_reconnect_jiffies = min(cp->cp_reconnect_jiffies * 2,
@@ -171,8 +171,7 @@ void rds_connect_worker(struct work_struct *work)
 
 	if (cp->cp_index > 0 &&
 	    rds_addr_cmp(&cp->cp_conn->c_laddr, &cp->cp_conn->c_faddr) >= 0)
-		return;
-	clear_bit(RDS_RECONNECT_PENDING, &cp->cp_flags);
+		goto out;
 	ret = rds_conn_path_transition(cp, RDS_CONN_DOWN, RDS_CONN_CONNECTING);
 	if (ret) {
 		ret = conn->c_trans->conn_path_connect(cp);
@@ -188,6 +187,8 @@ void rds_connect_worker(struct work_struct *work)
 				rds_conn_path_error(cp, "connect failed\n");
 		}
 	}
+out:
+	rds_clear_reconnect_pending_work_bit(cp);
 }
 
 void rds_send_worker(struct work_struct *work)
-- 
2.43.0