RDS: restore the exponential back-off scheme

Lower IP and exponential back-off scheme was added to save the
SM queries because of races but it doesn't do what its intended.
The exponential back-off scheme does a good job of backing off
for races. The code just falls back to the original scheme.

Orabug: 22347191

Tested-by: Michael Nowak <[email protected]>
Tested-by: Rafael Alejandro Peralez <[email protected]>
Tested-by: Liwen Huang <[email protected]>
Tested-by: Hong Liu <[email protected]>
Reviewed-by: Mukesh Kacker <[email protected]>
Signed-off-by: Santosh Shilimkar <[email protected]>

Author: SantoshShilimkar

net/rds/connection.c

@@ -750,7 +750,6 @@ void rds_conn_drop(struct rds_connection *conn, int reason)
 		conn->c_reconnect_warn = 1;
 		conn->c_reconnect_drops = 0;
 		conn->c_reconnect_err = 0;
-		conn->c_reconnect_racing = 0;
 		printk(KERN_INFO "RDS/IB: connection "
 			"<%u.%u.%u.%u,%u.%u.%u.%u,%d> dropped due to '%s'\n",
 			NIPQUAD(conn->c_laddr),

net/rds/ib_cm.c

@@ -915,8 +915,6 @@ int rds_ib_cm_handle_connect(struct rdma_cm_id *cm_id,
 		} else if (rds_conn_state(conn) == RDS_CONN_CONNECTING) {
 			unsigned long now = get_seconds();
 
-			conn->c_reconnect_racing++;
-
 			/*
 			 * after 15 seconds, give up on existing connection
 			 * attempts and make them try again.  At this point

net/rds/rdma_transport.c

@@ -258,10 +258,6 @@ int rds_rdma_cm_event_handler(struct rdma_cm_id *cm_id,
 	case RDMA_CM_EVENT_REJECTED:
 		err = (int *)event->param.conn.private_data;
 
-		if (conn && event->status == RDS_REJ_CONSUMER_DEFINED &&
-		    *err <= 1)
-			conn->c_reconnect_racing++;
-
 		if (conn) {
 			if (event->status == RDS_REJ_CONSUMER_DEFINED &&
 			    (*err) == 0) {

net/rds/rds.h

@@ -293,7 +293,6 @@ struct rds_connection {
 
 	unsigned int		c_rdsinfo_pending;
 
-	unsigned int		c_reconnect_racing;
 	unsigned int		c_route_resolved;
 
 	enum rds_conn_drop_src	c_drop_source;

net/rds/threads.c

@@ -150,12 +150,8 @@ void rds_queue_reconnect(struct rds_connection *conn)
 		rand % conn->c_reconnect_jiffies, conn->c_reconnect_jiffies,
 		conn, &conn->c_laddr, &conn->c_faddr, conn->c_tos);
 
-	if (conn->c_laddr >= conn->c_faddr)
 		queue_delayed_work(conn->c_wq, &conn->c_conn_w,
 				   rand % conn->c_reconnect_jiffies);
-	else
-		queue_delayed_work(conn->c_wq, &conn->c_conn_w,
-				   msecs_to_jiffies(100));
 
 	conn->c_reconnect_jiffies = min(conn->c_reconnect_jiffies * 2,
 					rds_sysctl_reconnect_max_jiffies);
@@ -297,7 +293,6 @@ void rds_reconnect_timeout(struct work_struct *work)
 			NIPQUAD(conn->c_laddr),	NIPQUAD(conn->c_faddr),
 			conn->c_tos);
 		rds_conn_drop(conn, DR_RECONNECT_TIMEOUT);
-		conn->c_reconnect_racing = 0;
 	}
 }
 
@@ -315,25 +310,7 @@ void rds_shutdown_worker(struct work_struct *work)
 				conn->c_tos,
 				conn_drop_reason_str(conn->c_drop_source));
 
-	/* if racing is detected, lower IP backs off and let the higher IP
-	 * drives the reconnect (one-sided reconnect)
-	 */
-	if (conn->c_laddr < conn->c_faddr && conn->c_reconnect_racing) {
-		rds_rtd(RDS_RTD_CM,
-			"calling rds_conn_shutdown, conn %p:0 <%u.%u.%u.%u,%u.%u.%u.%u,%d>\n",
-			conn, NIPQUAD(conn->c_laddr), NIPQUAD(conn->c_faddr),
-			conn->c_tos);
-		rds_conn_shutdown(conn, 0);
-		queue_delayed_work(conn->c_wq, &conn->c_reconn_w,
-				   msecs_to_jiffies(5000));
-	} else {
-		rds_rtd(RDS_RTD_CM,
-			"calling rds_conn_shutdown, conn %p:1 <%u.%u.%u.%u,%u.%u.%u.%u,%d>\n",
-			conn, NIPQUAD(conn->c_laddr), NIPQUAD(conn->c_faddr),
-			conn->c_tos);
-		rds_conn_shutdown(conn, 1);
-	}
-
+	rds_conn_shutdown(conn, 1);
 }
 
 void rds_threads_exit(void)