Warn opers about unresponsive servers

doc/reference.conf

@@ -1315,6 +1315,13 @@ general {
 	 */
 	ping_cookie = no;
 
+	/* ping warn time: how long to wait after pinging a server before starting
+	 * to complain it is unresponsive. Note that the ping check interval is 30 
+	 * seconds, so the first complaint will come at the next check after this 
+	 * time has passed.
+	 */
+	ping_warn_time = 15 seconds;
+
 	/* connect timeout: sets how long we should wait for a connection
 	 * request to succeed
 	 */

include/client.h

@@ -419,6 +419,7 @@ struct ListClient
 #define FLAGS_EXEMPTSHIDE	0x04000000
 #define FLAGS_EXEMPTJUPE	0x08000000
 #define FLAGS_IDENTIFIED	0x10000000	/* owns their current nick */
+#define FLAGS_PINGWARN		0x20000000	/* whether we've warned about this client being unresponsive */
 
 
 /* flags for local clients, this needs stuff moved from above to here at some point */

include/s_conf.h

@@ -204,6 +204,7 @@ struct config_file_entry
 	int operspy_admin_only;
 	int pace_wait;
 	int pace_wait_simple;
+	int ping_warn_time;
 	int short_motd;
 	int no_oper_flood;
 	int hide_server;

ircd/client.c

@@ -420,6 +420,27 @@ check_pings_list(rb_dlink_list * list)
 				client_p->localClient->lasttime = rb_current_time() - ping;
 				sendto_one(client_p, "PING :%s", me.name);
 			}
+			else if (ConfigFileEntry.ping_warn_time > 0 && (IsServer(client_p) || IsHandshake(client_p)) &&
+					(rb_current_time() - client_p->localClient->lasttime) >= (ping + ConfigFileEntry.ping_warn_time))
+			{
+				/*
+				 * if we haven't heard from a server in a while,
+				 * warn opers that something could be wrong...
+				 *
+				 * we'll do this about every 30 seconds until
+				 * the server either becomes responsive or
+				 * pings out. whichever comes first.
+				 */
+				client_p->flags |= FLAGS_PINGWARN;
+				sendto_realops_snomask(SNO_GENERAL, L_NETWIDE,
+					     "Warning: No response from %s for %ld seconds",
+					     client_p->name,
+					     (rb_current_time() - client_p->localClient->lasttime - ping));
+				ilog(L_SERVER,
+					     "Warning: No response from %s for %ld seconds",
+					     log_client_name(client_p, HIDE_IP),
+					     (rb_current_time() - client_p->localClient->lasttime - ping));
+			}
 		}
 		/* ping_timeout: */
 

ircd/newconf.c

@@ -2751,6 +2751,7 @@ static struct ConfEntry conf_general_table[] =
 	{ "pace_wait",		CF_TIME,  NULL, 0, &ConfigFileEntry.pace_wait		},
 	{ "pace_wait_simple",	CF_TIME,  NULL, 0, &ConfigFileEntry.pace_wait_simple	},
 	{ "ping_cookie",	CF_YESNO, NULL, 0, &ConfigFileEntry.ping_cookie		},
+	{ "ping_warn_time",	CF_TIME,  NULL, 0, &ConfigFileEntry.ping_warn_time	},
 	{ "reject_after_count",	CF_INT,   NULL, 0, &ConfigFileEntry.reject_after_count	},
 	{ "reject_ban_time",	CF_TIME,  NULL, 0, &ConfigFileEntry.reject_ban_time	},
 	{ "reject_duration",	CF_TIME,  NULL, 0, &ConfigFileEntry.reject_duration	},

ircd/packet.c

@@ -272,6 +272,23 @@ read_packet(rb_fde_t * F, void *data)
 			client_p->localClient->lasttime = rb_current_time();
 		client_p->flags &= ~FLAGS_PINGSENT;
 
+		if (client_p->flags & FLAGS_PINGWARN)
+		{
+			/*
+			 * if we warned about this server being unresponsive
+			 * before, let's let everyone know there's no need
+			 * to panic
+			 */
+			client_p->flags &= ~FLAGS_PINGWARN;
+			sendto_realops_snomask(SNO_GENERAL, L_NETWIDE,
+				"Received response from previously unresponsive link %s",
+				client_p->name);
+			ilog(L_SERVER,
+				"Received response from previously unresponsive link %s",
+				log_client_name(client_p, HIDE_IP));
+		}
+
+
 		/*
 		 * Before we even think of parsing what we just read, stick
 		 * it on the end of the receive queue and do it when its

modules/m_info.c

@@ -421,6 +421,11 @@ static struct InfoStruct info_table[] = {
 		"Require ping cookies to connect",
 		INFO_INTBOOL(&ConfigFileEntry.ping_cookie),
 	},
+	{
+		"ping_warn_time",
+		"Amount of time between warnings about unresponsive servers",
+		INFO_DECIMAL(&ConfigFileEntry.ping_warn_time),
+	},
 	{
 		"reject_after_count",
 		"Client rejection threshold setting",