ospfd, ospf6d: introduce the "graceful-restart hello-delay" command

This command makes unplanned GR more reliable by manipulating the
sending of Grace-LSAs and Hello packets for a certain amount of time,
increasing the chance that the neighboring routers are aware of
the ongoing graceful restart before resuming normal OSPF operation.

Signed-off-by: Renato Westphal <renato@opensourcerouting.org>

doc/user/ospf6d.rst

@@ -287,6 +287,19 @@ OSPF6 interface
 
    Sets interface's Router Dead Interval. Default value is 40.
 
+.. clicmd:: ipv6 ospf6 graceful-restart hello-delay HELLODELAYINTERVAL
+
+   Set the length of time during which Grace-LSAs are sent at 1-second intervals
+   while coming back up after an unplanned outage. During this time, no hello
+   packets are sent.
+
+   A higher hello delay will increase the chance that all neighbors are notified
+   about the ongoing graceful restart before receiving a hello packet (which is
+   crucial for the graceful restart to succeed). The hello delay shouldn't be set
+   too high, however, otherwise the adjacencies might time out. As a best practice,
+   it's recommended to set the hello delay and hello interval with the same values.
+   The default value is 10 seconds.
+
 .. clicmd:: ipv6 ospf6 retransmit-interval RETRANSMITINTERVAL
 
    Sets interface's Rxmt Interval. Default value is 5.

doc/user/ospfd.rst

@@ -635,6 +635,19 @@ Interfaces
    :clicmd:`ip ospf dead-interval minimal hello-multiplier (2-20)` is also
    specified for the interface.
 
+.. clicmd:: ip ospf graceful-restart hello-delay (1-1800)
+
+   Set the length of time during which Grace-LSAs are sent at 1-second intervals
+   while coming back up after an unplanned outage. During this time, no hello
+   packets are sent.
+
+   A higher hello delay will increase the chance that all neighbors are notified
+   about the ongoing graceful restart before receiving a hello packet (which is
+   crucial for the graceful restart to succeed). The hello delay shouldn't be set
+   too high, however, otherwise the adjacencies might time out. As a best practice,
+   it's recommended to set the hello delay and hello interval with the same values.
+   The default value is 10 seconds.
+
 .. clicmd:: ip ospf network (broadcast|non-broadcast|point-to-multipoint|point-to-point [dmvpn])
 
    When configuring a point-to-point network on an interface and the interface

lib/libospf.h

@@ -55,6 +55,7 @@ extern "C" {
 #define OSPF_ROUTER_DEAD_INTERVAL_DEFAULT  40
 #define OSPF_ROUTER_DEAD_INTERVAL_MINIMAL   1
 #define OSPF_HELLO_INTERVAL_DEFAULT        10
+#define OSPF_HELLO_DELAY_DEFAULT           10
 #define OSPF_ROUTER_PRIORITY_DEFAULT        1
 #define OSPF_RETRANSMIT_INTERVAL_DEFAULT    5
 #define OSPF_TRANSMIT_DELAY_DEFAULT         1

ospf6d/ospf6_gr.c

@@ -190,6 +190,14 @@ static void ospf6_gr_restart_exit(struct ospf6 *ospf6, const char *reason)
 		OSPF6_INTRA_PREFIX_LSA_SCHEDULE_STUB(area);
 
 		for (ALL_LIST_ELEMENTS_RO(area->if_list, anode, oi)) {
+			/* Disable hello delay. */
+			if (oi->gr.hello_delay.t_grace_send) {
+				oi->gr.hello_delay.elapsed_seconds = 0;
+				EVENT_OFF(oi->gr.hello_delay.t_grace_send);
+				event_add_event(master, ospf6_hello_send, oi, 0,
+						&oi->thread_send_hello);
+			}
+
 			/* Reoriginate Link-LSA. */
 			if (oi->type != OSPF_IFTYPE_VIRTUALLINK)
 				OSPF6_LINK_LSA_EXECUTE(oi);
@@ -520,6 +528,21 @@ static void ospf6_gr_grace_period_expired(struct event *thread)
 	ospf6_gr_restart_exit(ospf6, "grace period has expired");
 }
 
+/* Send extra Grace-LSA out the interface (unplanned outages only). */
+void ospf6_gr_iface_send_grace_lsa(struct event *thread)
+{
+	struct ospf6_interface *oi = EVENT_ARG(thread);
+
+	ospf6_gr_lsa_originate(oi, oi->area->ospf6->gr_info.reason);
+
+	if (++oi->gr.hello_delay.elapsed_seconds < oi->gr.hello_delay.interval)
+		event_add_timer(master, ospf6_gr_iface_send_grace_lsa, oi, 1,
+				&oi->gr.hello_delay.t_grace_send);
+	else
+		event_add_event(master, ospf6_hello_send, oi, 0,
+				&oi->thread_send_hello);
+}
+
 /*
  * Record in non-volatile memory that the given OSPF instance is attempting to
  * perform a graceful restart.
@@ -677,6 +700,11 @@ void ospf6_gr_unplanned_start_interface(struct ospf6_interface *oi)
 
 	/* Send Grace-LSA. */
 	ospf6_gr_lsa_originate(oi, oi->area->ospf6->gr_info.reason);
+
+	/* Start GR hello-delay interval. */
+	oi->gr.hello_delay.elapsed_seconds = 0;
+	event_add_timer(master, ospf6_gr_iface_send_grace_lsa, oi, 1,
+			&oi->gr.hello_delay.t_grace_send);
 }
 
 /* Prepare to start a Graceful Restart. */

ospf6d/ospf6_gr.h

@@ -155,6 +155,7 @@ extern int config_write_ospf6_gr(struct vty *vty, struct ospf6 *ospf6);
 extern int config_write_ospf6_gr_helper(struct vty *vty, struct ospf6 *ospf6);
 extern int config_write_ospf6_debug_gr_helper(struct vty *vty);
 
+extern void ospf6_gr_iface_send_grace_lsa(struct event *thread);
 extern void ospf6_gr_restart_enter(struct ospf6 *ospf6,
 				   enum ospf6_gr_restart_reason reason,
 				   int timestamp);

ospf6d/ospf6_interface.c

@@ -35,6 +35,7 @@
 #include "ospf6_proto.h"
 #include "lib/keychain.h"
 #include "ospf6_auth_trailer.h"
+#include "ospf6d/ospf6_interface_clippy.c"
 
 DEFINE_MTYPE_STATIC(OSPF6D, OSPF6_IF, "OSPF6 interface");
 DEFINE_MTYPE(OSPF6D, OSPF6_AUTH_KEYCHAIN, "OSPF6 auth keychain");
@@ -202,6 +203,7 @@ struct ospf6_interface *ospf6_interface_create(struct interface *ifp)
 	oi->priority = OSPF6_INTERFACE_PRIORITY;
 
 	oi->hello_interval = OSPF_HELLO_INTERVAL_DEFAULT;
+	oi->gr.hello_delay.interval = OSPF_HELLO_DELAY_DEFAULT;
 	oi->dead_interval = OSPF_ROUTER_DEAD_INTERVAL_DEFAULT;
 	oi->rxmt_interval = OSPF_RETRANSMIT_INTERVAL_DEFAULT;
 	oi->type = ospf6_default_iftype(ifp);
@@ -324,6 +326,9 @@ void ospf6_interface_disable(struct ospf6_interface *oi)
 	EVENT_OFF(oi->thread_intra_prefix_lsa);
 	EVENT_OFF(oi->thread_as_extern_lsa);
 	EVENT_OFF(oi->thread_wait_timer);
+
+	oi->gr.hello_delay.elapsed_seconds = 0;
+	EVENT_OFF(oi->gr.hello_delay.t_grace_send);
 }
 
 static struct in6_addr *
@@ -1191,6 +1196,9 @@ static int ospf6_interface_show(struct vty *vty, struct interface *ifp,
 				json_arr, json_object_new_string(lsa->name));
 		json_object_object_add(json_obj, "pendingLsaLsAck", json_arr);
 
+		if (oi->gr.hello_delay.interval != 0)
+			json_object_int_add(json_obj, "grHelloDelaySecs",
+					    oi->gr.hello_delay.interval);
 	} else {
 		timerclear(&res);
 		if (event_is_scheduled(oi->thread_send_lsupdate))
@@ -1216,6 +1224,10 @@ static int ospf6_interface_show(struct vty *vty, struct interface *ifp,
 								   : "off"));
 		for (ALL_LSDB(oi->lsack_list, lsa, lsanext))
 			vty_out(vty, "      %s\n", lsa->name);
+
+		if (oi->gr.hello_delay.interval != 0)
+			vty_out(vty, "  Graceful Restart hello delay: %us\n",
+				oi->gr.hello_delay.interval);
 	}
 
 	/* BFD specific. */
@@ -2168,6 +2180,50 @@ ALIAS (ipv6_ospf6_deadinterval,
        "Interval time after which a neighbor is declared down\n"
        SECONDS_STR)
 
+DEFPY(ipv6_ospf6_gr_hdelay, ipv6_ospf6_gr_hdelay_cmd,
+      "ipv6 ospf6 graceful-restart hello-delay (1-1800)",
+      IP6_STR
+      OSPF6_STR
+      "Graceful Restart parameters\n"
+      "Delay the sending of the first hello packets.\n"
+      "Delay in seconds\n")
+{
+	VTY_DECLVAR_CONTEXT(interface, ifp);
+	struct ospf6_interface *oi;
+
+	oi = ifp->info;
+	if (oi == NULL)
+		oi = ospf6_interface_create(ifp);
+
+	/* Note: new or updated value won't affect ongoing graceful restart. */
+	oi->gr.hello_delay.interval = hello_delay;
+
+	return CMD_SUCCESS;
+}
+
+DEFPY(no_ipv6_ospf6_gr_hdelay, no_ipv6_ospf6_gr_hdelay_cmd,
+      "no ipv6 ospf6 graceful-restart hello-delay [(1-1800)]",
+      NO_STR
+      IP6_STR
+      OSPF6_STR
+      "Graceful Restart parameters\n"
+      "Delay the sending of the first hello packets.\n"
+      "Delay in seconds\n")
+{
+	VTY_DECLVAR_CONTEXT(interface, ifp);
+	struct ospf6_interface *oi;
+
+	oi = ifp->info;
+	if (oi == NULL)
+		oi = ospf6_interface_create(ifp);
+
+	oi->gr.hello_delay.interval = OSPF_HELLO_DELAY_DEFAULT;
+	oi->gr.hello_delay.elapsed_seconds = 0;
+	EVENT_OFF(oi->gr.hello_delay.t_grace_send);
+
+	return CMD_SUCCESS;
+}
+
 /* interface variable set command */
 DEFUN (ipv6_ospf6_transmitdelay,
        ipv6_ospf6_transmitdelay_cmd,
@@ -2635,6 +2691,11 @@ static int config_write_ospf6_interface(struct vty *vty, struct vrf *vrf)
 		else if (oi->type_cfg && oi->type == OSPF_IFTYPE_BROADCAST)
 			vty_out(vty, " ipv6 ospf6 network broadcast\n");
 
+		if (oi->gr.hello_delay.interval != OSPF_HELLO_DELAY_DEFAULT)
+			vty_out(vty,
+				" ipv6 ospf6 graceful-restart hello-delay %u\n",
+				oi->gr.hello_delay.interval);
+
 		ospf6_bfd_write_config(vty, oi);
 
 		ospf6_auth_write_config(vty, &oi->at_data);
@@ -2733,12 +2794,14 @@ void ospf6_interface_init(void)
 
 	install_element(INTERFACE_NODE, &ipv6_ospf6_deadinterval_cmd);
 	install_element(INTERFACE_NODE, &ipv6_ospf6_hellointerval_cmd);
+	install_element(INTERFACE_NODE, &ipv6_ospf6_gr_hdelay_cmd);
 	install_element(INTERFACE_NODE, &ipv6_ospf6_priority_cmd);
 	install_element(INTERFACE_NODE, &ipv6_ospf6_retransmitinterval_cmd);
 	install_element(INTERFACE_NODE, &ipv6_ospf6_transmitdelay_cmd);
 	install_element(INTERFACE_NODE, &ipv6_ospf6_instance_cmd);
 	install_element(INTERFACE_NODE, &no_ipv6_ospf6_deadinterval_cmd);
 	install_element(INTERFACE_NODE, &no_ipv6_ospf6_hellointerval_cmd);
+	install_element(INTERFACE_NODE, &no_ipv6_ospf6_gr_hdelay_cmd);
 	install_element(INTERFACE_NODE, &no_ipv6_ospf6_priority_cmd);
 	install_element(INTERFACE_NODE, &no_ipv6_ospf6_retransmitinterval_cmd);
 	install_element(INTERFACE_NODE, &no_ipv6_ospf6_transmitdelay_cmd);

ospf6d/ospf6_interface.h

@@ -74,6 +74,15 @@ struct ospf6_interface {
 	uint16_t dead_interval;
 	uint32_t rxmt_interval;
 
+	/* Graceful-Restart data. */
+	struct {
+		struct {
+			uint16_t interval;
+			uint16_t elapsed_seconds;
+			struct event *t_grace_send;
+		} hello_delay;
+	} gr;
+
 	uint32_t state_change;
 
 	/* Cost */

ospf6d/ospf6_message.c

@@ -2244,6 +2244,10 @@ void ospf6_hello_send(struct event *thread)
 
 	oi = (struct ospf6_interface *)EVENT_ARG(thread);
 
+	/* Check if the GR hello-delay is active. */
+	if (oi->gr.hello_delay.t_grace_send)
+		return;
+
 	if (oi->state <= OSPF6_INTERFACE_DOWN) {
 		if (IS_OSPF6_DEBUG_MESSAGE(OSPF6_MESSAGE_TYPE_HELLO, SEND_HDR))
 			zlog_debug("Unable to send Hello on down interface %s",

ospf6d/subdir.am

@@ -78,6 +78,7 @@ clippy_scan += \
 	ospf6d/ospf6_top.c \
 	ospf6d/ospf6_area.c \
 	ospf6d/ospf6_asbr.c \
+	ospf6d/ospf6_interface.c \
 	ospf6d/ospf6_lsa.c \
 	ospf6d/ospf6_gr_helper.c \
 	ospf6d/ospf6_gr.c \

ospfd/ospf_gr.c

@@ -237,13 +237,22 @@ static void ospf_gr_restart_exit(struct ospf *ospf, const char *reason)
 		 */
 		ospf_router_lsa_update_area(area);
 
-		/*
-		 * 2) The router should reoriginate network-LSAs on all segments
-		 *    where it is the Designated Router.
-		 */
-		for (ALL_LIST_ELEMENTS_RO(area->oiflist, anode, oi))
+		for (ALL_LIST_ELEMENTS_RO(area->oiflist, anode, oi)) {
+			/* Disable hello delay. */
+			if (oi->gr.hello_delay.t_grace_send) {
+				oi->gr.hello_delay.elapsed_seconds = 0;
+				EVENT_OFF(oi->gr.hello_delay.t_grace_send);
+				OSPF_ISM_TIMER_MSEC_ON(oi->t_hello,
+						       ospf_hello_timer, 1);
+			}
+
+			/*
+			 * 2) The router should reoriginate network-LSAs on all
+			 * segments where it is the Designated Router.
+			 */
 			if (oi->state == ISM_DR)
 				ospf_network_lsa_update(oi);
+		}
 	}
 
 	/*
@@ -561,6 +570,21 @@ static char *ospf_gr_nvm_filepath(struct ospf *ospf)
 	return filepath;
 }
 
+/* Send extra Grace-LSA out the interface (unplanned outages only). */
+void ospf_gr_iface_send_grace_lsa(struct event *thread)
+{
+	struct ospf_interface *oi = EVENT_ARG(thread);
+	struct ospf_if_params *params = IF_DEF_PARAMS(oi->ifp);
+
+	ospf_gr_lsa_originate(oi, oi->ospf->gr_info.reason, false);
+
+	if (++oi->gr.hello_delay.elapsed_seconds < params->v_gr_hello_delay)
+		event_add_timer(master, ospf_gr_iface_send_grace_lsa, oi, 1,
+				&oi->gr.hello_delay.t_grace_send);
+	else
+		OSPF_ISM_TIMER_MSEC_ON(oi->t_hello, ospf_hello_timer, 1);
+}
+
 /*
  * Record in non-volatile memory that the given OSPF instance is attempting to
  * perform a graceful restart.
@@ -714,6 +738,11 @@ void ospf_gr_unplanned_start_interface(struct ospf_interface *oi)
 {
 	/* Send Grace-LSA. */
 	ospf_gr_lsa_originate(oi, oi->ospf->gr_info.reason, false);
+
+	/* Start GR hello-delay interval. */
+	oi->gr.hello_delay.elapsed_seconds = 0;
+	event_add_timer(master, ospf_gr_iface_send_grace_lsa, oi, 1,
+			&oi->gr.hello_delay.t_grace_send);
 }
 
 /* Prepare to start a Graceful Restart. */

ospfd/ospf_gr.h

@@ -166,6 +166,7 @@ extern void ospf_gr_helper_supported_gracetime_set(struct ospf *ospf,
 						   uint32_t interval);
 extern void ospf_gr_helper_set_supported_planned_only_restart(struct ospf *ospf,
 							     bool planned_only);
+extern void ospf_gr_iface_send_grace_lsa(struct event *thread);
 extern void ospf_gr_restart_enter(struct ospf *ospf,
 				  enum ospf_gr_restart_reason reason,
 				  int timestamp);

ospfd/ospf_interface.c

@@ -527,6 +527,7 @@ static struct ospf_if_params *ospf_new_if_params(void)
 	UNSET_IF_PARAM(oip, passive_interface);
 	UNSET_IF_PARAM(oip, v_hello);
 	UNSET_IF_PARAM(oip, fast_hello);
+	UNSET_IF_PARAM(oip, v_gr_hello_delay);
 	UNSET_IF_PARAM(oip, v_wait);
 	UNSET_IF_PARAM(oip, priority);
 	UNSET_IF_PARAM(oip, type);
@@ -676,6 +677,9 @@ int ospf_if_new_hook(struct interface *ifp)
 	SET_IF_PARAM(IF_DEF_PARAMS(ifp), fast_hello);
 	IF_DEF_PARAMS(ifp)->fast_hello = OSPF_FAST_HELLO_DEFAULT;
 
+	SET_IF_PARAM(IF_DEF_PARAMS(ifp), v_gr_hello_delay);
+	IF_DEF_PARAMS(ifp)->v_gr_hello_delay = OSPF_HELLO_DELAY_DEFAULT;
+
 	SET_IF_PARAM(IF_DEF_PARAMS(ifp), v_wait);
 	IF_DEF_PARAMS(ifp)->v_wait = OSPF_ROUTER_DEAD_INTERVAL_DEFAULT;
 

ospfd/ospf_interface.h

@@ -72,6 +72,9 @@ struct ospf_if_params {
 	DECLARE_IF_PARAM(uint32_t, v_wait);  /* Router Dead Interval */
 	bool is_v_wait_set;                  /* Check for Dead Interval set */
 
+	/* GR Hello Delay Interval */
+	DECLARE_IF_PARAM(uint16_t, v_gr_hello_delay);
+
 	/* MTU mismatch check (see RFC2328, chap 10.6) */
 	DECLARE_IF_PARAM(uint8_t, mtu_ignore);
 
@@ -214,6 +217,14 @@ struct ospf_interface {
 	/* List of configured NBMA neighbor. */
 	struct list *nbr_nbma;
 
+	/* Graceful-Restart data. */
+	struct {
+		struct {
+			uint16_t elapsed_seconds;
+			struct event *t_grace_send;
+		} hello_delay;
+	} gr;
+
 	/* self-originated LSAs. */
 	struct ospf_lsa *network_lsa_self; /* network-LSA. */
 	struct list *opaque_lsa_self;      /* Type-9 Opaque-LSAs */