Fix: sbd watchdog rebooting upon restart of pacemaker-remote

Author: wenningerk

src/sbd-cluster.c

@@ -58,7 +58,15 @@ static crm_cluster_t cluster;
 static gboolean sbd_remote_check(gpointer user_data);
 static long unsigned int find_pacemaker_remote(void);
 static void sbd_membership_destroy(gpointer user_data);
+static bool wait_for_pacemaker_remote_lost = false;
 
+static void signal_exitreq(void)
+{
+    union sigval signal_value;
+    pid_t ppid = getppid();
+
+    sigqueue(ppid, SIG_EXITREQ, signal_value);
+}
 
 #if SUPPORT_PLUGIN
 static void
@@ -675,6 +683,10 @@ sbd_remote_check(gpointer user_data)
         set_servant_health(pcmk_health_online, LOG_INFO,
                            "Connected to Pacemaker Remote %lu", (long unsigned int)remoted_pid);
     } else {
+        if (wait_for_pacemaker_remote_lost) {
+            signal_exitreq();
+            return true;
+        }
         set_servant_health(pcmk_health_unclean, LOG_WARNING,
                            "Connection to Pacemaker Remote %lu lost", (long unsigned int)remoted_pid);
     }
@@ -742,6 +754,16 @@ cluster_shutdown(int nsig)
     clean_up(0);
 }
 
+static void
+trigger_wait_for_pacemaker_remote_lost(int nsig)
+{
+    /* if we've never seen pacemaker_remoted request exit immeditely */
+    if ((remoted_pid <= 0) || !remote_node) {
+        signal_exitreq();
+    }
+    wait_for_pacemaker_remote_lost = true;
+}
+
 int
 servant_cluster(const char *diskname, int mode, const void* argp)
 {
@@ -761,6 +783,7 @@ servant_cluster(const char *diskname, int mode, const void* argp)
 
     mainloop_add_signal(SIGTERM, cluster_shutdown);
     mainloop_add_signal(SIGINT, cluster_shutdown);
+	mainloop_add_signal(SIGUSR2, trigger_wait_for_pacemaker_remote_lost);
 
     g_main_loop_run(mainloop);
     g_main_loop_unref(mainloop);

src/sbd-inquisitor.c

@@ -248,14 +248,14 @@ void servants_start(void)
 	}
 }
 
-void servants_kill(void)
+void servants_kill(int sig)
 {
 	struct servants_list_item *s;
 	union sigval svalue;
 
 	for (s = servants_leader; s; s = s->next) {
 		if (s->pid != 0)
-			sigqueue(s->pid, SIGKILL, svalue);
+			sigqueue(s->pid, sig, svalue);
 	}
 }
 
@@ -536,7 +536,7 @@ void inquisitor_child(void)
 		clock_gettime(CLOCK_MONOTONIC, &t_now);
 
 		if (sig == SIG_EXITREQ || sig == SIGTERM) {
-			servants_kill();
+			servants_kill(SIGKILL);
 			watchdog_close(true);
 			exiting = 1;
 		} else if (sig == SIGCHLD) {
@@ -610,6 +610,8 @@ void inquisitor_child(void)
 			if (exiting)
 				continue;
 			servants_start();
+		} else if (sig == SIGUSR2) {
+			servants_kill(SIGUSR2);
 		}
 
 		if (exiting) {
@@ -718,7 +720,7 @@ void inquisitor_child(void)
                      */
                     cl_log(LOG_DEBUG, "Decoupling");
                     if (inquisitor_decouple() < 0) {
-                        servants_kill();
+                        servants_kill(SIGKILL);
                         exiting = 1;
                         continue;
                     } else {
@@ -734,7 +736,7 @@ void inquisitor_child(void)
 				/* We're still being watched by our
 				 * parent. We don't fence, but exit. */
 				cl_log(LOG_ERR, "SBD: Not enough votes to proceed. Aborting start-up.");
-				servants_kill();
+				servants_kill(SIGKILL);
 				exiting = 1;
 				continue;
 			}

src/sbd_remote.service.in

@@ -11,7 +11,7 @@ Type=forking
 PIDFile=@runstatedir@/sbd.pid
 EnvironmentFile=-@CONFIGDIR@/sbd
 ExecStart=@sbindir@/sbd $SBD_OPTS -p @runstatedir@/sbd.pid watch
-ExecStop=@bindir@/kill -TERM $MAINPID
+ExecStop=@bindir@/kill -USR2 $MAINPID
 
 # Could this benefit from exit codes for restart?
 # Does this need to be set to msgwait * 1.2?