rds: Change heartbeat params from module params to sysctl

Orabug: 30418039

Signed-off-by: Håkon Bugge <[email protected]>
Reviewed-by: Ka-Cheong Poon <[email protected]>
Tested-by: Michael Nowak <[email protected]>

---

Since Exadata is not (yet) using the heartbeat mechanism, now is a
good time to change the params to sysctl.

v1->v2:
	* Added commentary to the sysctl parameters, as suggested by
          Ka-Cheong
	* s/proc_dointvec_minmax/proc_douintvec_minmax/ as detected by
          Ka-Cheong
	* Added Ka-Cheong's r-b and Mike's t-b
	* Changed the default heartbeat interval to 60 seconds, as per
          discussions with Exadata (Mike Nowak)
	* Enabled heart-beating by default, as suggested by Mike Nowak

Signed-off-by: Somasundaram Krishnasamy <[email protected]>

Author: Hakon-Bugge

net/rds/rds.h

@@ -1317,6 +1317,8 @@ extern unsigned int  rds_sysctl_trace_level;
 extern unsigned int  rds_sysctl_shutdown_trace_start_time;
 extern unsigned int  rds_sysctl_shutdown_trace_end_time;
 extern unsigned int  rds_sysctl_passive_connect_delay_percent;
+extern unsigned int  rds_sysctl_conn_hb_timeout;
+extern unsigned int  rds_sysctl_conn_hb_interval;
 
 /* threads.c */
 int rds_threads_init(void);

net/rds/sysctl.c

@@ -60,6 +60,23 @@ unsigned int rds_sysctl_passive_connect_delay_min_percent = 1;
 unsigned int rds_sysctl_passive_connect_delay_max_percent = 1000;
 unsigned int rds_sysctl_passive_connect_delay_percent = 100;
 
+/* Heartbeat timeout in seconds. That is the time from a heartbeat ping
+ * is sent, before the heartbeat mechanism drops the connection, unless
+ * a heartbeat pong has been received.
+ */
+unsigned int rds_sysctl_min_conn_hb_timeout;
+unsigned int rds_sysctl_max_conn_hb_timeout = 60;
+unsigned int rds_sysctl_conn_hb_timeout = 10;
+
+/* Heartbeat interval in seconds. The time from a successful heartbeat
+ * pong has been received until a new heartbeat ping is sent out is
+ * pseudo randomly chosen in the interval 50% to 150% of
+ * rds_sysctl_min_conn_hb_interval.
+ */
+unsigned int rds_sysctl_min_conn_hb_interval = 10;
+unsigned int rds_sysctl_max_conn_hb_interval = 600;
+unsigned int rds_sysctl_conn_hb_interval = 60;
+
 /*
  * We have official values, but must maintain the sysctl interface for existing
  * software that expects to find these values here.
@@ -153,6 +170,24 @@ static struct ctl_table rds_sysctl_rds_table[] = {
 		.extra1		= &rds_sysctl_passive_connect_delay_min_percent,
 		.extra2		= &rds_sysctl_passive_connect_delay_max_percent,
 	},
+	{
+		.procname       = "conn_heartbeat_timeout_secs",
+		.data           = &rds_sysctl_conn_hb_timeout,
+		.maxlen         = sizeof(rds_sysctl_conn_hb_timeout),
+		.mode           = 0644,
+		.proc_handler   = proc_douintvec_minmax,
+		.extra1		= &rds_sysctl_min_conn_hb_timeout,
+		.extra2		= &rds_sysctl_max_conn_hb_timeout,
+	},
+	{
+		.procname       = "conn_heartbeat_interval_secs",
+		.data           = &rds_sysctl_conn_hb_interval,
+		.maxlen         = sizeof(rds_sysctl_conn_hb_interval),
+		.mode           = 0644,
+		.proc_handler   = proc_douintvec_minmax,
+		.extra1		= &rds_sysctl_min_conn_hb_interval,
+		.extra2		= &rds_sysctl_max_conn_hb_interval,
+	},
 	{ }
 };
 

net/rds/threads.c

@@ -36,14 +36,6 @@
 
 #include "rds.h"
 
-static unsigned int rds_conn_hb_timeout = 0;
-module_param(rds_conn_hb_timeout, int, 0444);
-MODULE_PARM_DESC(rds_conn_hb_timeout, " Connection heartbeat timeout (seconds)");
-static unsigned int rds_conn_hb_interval = 10;
-module_param(rds_conn_hb_interval, int, 0444);
-MODULE_PARM_DESC(rds_conn_hb_interval, " Connection heartbeat interval (seconds)");
-
-
 /*
  * All of connection management is simplified by serializing it through
  * work queues that execute in a connection managing thread.
@@ -344,7 +336,7 @@ void rds_hb_worker(struct work_struct *work)
 	int ret;
 	struct rds_connection *conn = cp->cp_conn;
 
-	if (!rds_conn_hb_timeout || conn->c_loopback ||
+	if (!rds_sysctl_conn_hb_timeout || conn->c_loopback ||
 	    conn->c_trans->t_type == RDS_TRANS_TCP)
 		return;
 
@@ -354,9 +346,9 @@ void rds_hb_worker(struct work_struct *work)
 			if (!cp->cp_hb_start) {
 				cp->cp_hb_state = HB_PONG_RCVD;
 				/* Pseudo random from 50% to 150% of interval */
-				delay = msecs_to_jiffies(rds_conn_hb_interval * 1000 / 2) +
-					msecs_to_jiffies(prandom_u32() % rds_conn_hb_interval * 1000);
-			} else if (now - cp->cp_hb_start > rds_conn_hb_timeout) {
+				delay = msecs_to_jiffies(rds_sysctl_conn_hb_interval * 1000 / 2) +
+					msecs_to_jiffies(prandom_u32() % rds_sysctl_conn_hb_interval * 1000);
+			} else if (now - cp->cp_hb_start > rds_sysctl_conn_hb_timeout) {
 				rds_rtd_ptr(RDS_RTD_CM,
 					    "RDS/IB: connection <%pI6c,%pI6c,%d> timed out (0x%lx,0x%lx)..discon and recon\n",
 					    &conn->c_laddr, &conn->c_faddr,