BGP graceful restart support.

Also significant core protocol state changes needed for that,
global graceful restart recovery state and kernel proto support
for recovery.

4 files changed, 54 insertions, 8 deletions

diff --git a/sysdep/unix/krt.Y b/sysdep/unix/krt.Y
index 469c136d..630cda38 100644
--- a/sysdep/unix/krt.Y
+++ b/sysdep/unix/krt.Y
@@ -17,7 +17,7 @@ CF_DEFINES
 
 CF_DECLS
 
-CF_KEYWORDS(KERNEL, PERSIST, SCAN, TIME, LEARN, DEVICE, ROUTES, KRT_SOURCE, KRT_METRIC)
+CF_KEYWORDS(KERNEL, PERSIST, SCAN, TIME, LEARN, DEVICE, ROUTES, GRACEFUL, RESTART, KRT_SOURCE, KRT_METRIC)
 
 CF_GRAMMAR
 
@@ -46,6 +46,7 @@ kern_item:
 #endif
    }
  | DEVICE ROUTES bool { THIS_KRT->devroutes = $3; }
+ | GRACEFUL RESTART bool { THIS_KRT->graceful_restart = $3; }
  ;
 
 /* Kernel interface protocol */
diff --git a/sysdep/unix/krt.c b/sysdep/unix/krt.c
index 6fdef619..bff3001f 100644
--- a/sysdep/unix/krt.c
+++ b/sysdep/unix/krt.c
@@ -653,6 +653,13 @@ krt_got_route(struct krt_proto *p, rte *e)
       return;
     }
 
+  if (!p->ready)
+    {
+      /* We wait for the initial feed to have correct KRF_INSTALLED flag */
+      verdict = KRF_IGNORE;
+      goto sentenced;
+    }
+
   old = net->routes;
   if ((net->n.flags & KRF_INSTALLED) && rte_is_valid(old))
     {
@@ -779,7 +786,9 @@ krt_prune(struct krt_proto *p)
   if (KRT_CF->learn)
     krt_learn_prune(p);
 #endif
-  p->initialized = 1;
+
+  if (p->ready)
+    p->initialized = 1;
 }
 
 void
@@ -852,7 +861,7 @@ krt_scan_timer_start(struct krt_proto *p)
 
   krt_scan_count++;
 
-  tm_start(krt_scan_timer, 0);
+  tm_start(krt_scan_timer, 1);
 }
 
 static void
@@ -867,6 +876,12 @@ krt_scan_timer_stop(struct krt_proto *p)
   }
 }
 
+static void
+krt_scan_timer_kick(struct krt_proto *p UNUSED)
+{
+  tm_start(krt_scan_timer, 0);
+}
+
 #else
 
 static void
@@ -885,7 +900,7 @@ static void
 krt_scan_timer_start(struct krt_proto *p)
 {
   p->scan_timer = tm_new_set(p->p.pool, krt_scan, p, 0, KRT_CF->scan_time);
-  tm_start(p->scan_timer, 0);
+  tm_start(p->scan_timer, 1);
 }
 
 static void
@@ -894,6 +909,12 @@ krt_scan_timer_stop(struct krt_proto *p)
   tm_stop(p->scan_timer);
 }
 
+static void
+krt_scan_timer_kick(struct krt_proto *p UNUSED)
+{
+  tm_start(p->scan_timer, 0);
+}
+
 #endif
 
 
@@ -970,6 +991,16 @@ krt_notify(struct proto *P, struct rtable *table UNUSED, net *net,
     krt_replace_rte(p, net, new, old, eattrs);
 }
 
+static void
+krt_feed_done(struct proto *P)
+{
+  struct krt_proto *p = (struct krt_proto *) P;
+
+  p->ready = 1;
+  krt_scan_timer_kick(p);
+}
+
+
 static int
 krt_rte_same(rte *a, rte *b)
 {
@@ -992,6 +1023,7 @@ krt_init(struct proto_config *c)
   p->p.accept_ra_types = RA_OPTIMAL;
   p->p.import_control = krt_import_control;
   p->p.rt_notify = krt_notify;
+  p->p.feed_done = krt_feed_done;
   p->p.make_tmp_attrs = krt_make_tmp_attrs;
   p->p.store_tmp_attrs = krt_store_tmp_attrs;
   p->p.rte_same = krt_rte_same;
@@ -1015,6 +1047,9 @@ krt_start(struct proto *P)
 
   krt_scan_timer_start(p);
 
+  if (P->gr_recovery && KRT_CF->graceful_restart)
+    P->gr_wait = 1;
+
   return PS_UP;
 }
 
@@ -1029,6 +1064,9 @@ krt_shutdown(struct proto *P)
   if (p->initialized && !KRT_CF->persist)
     krt_flush_routes(p);
 
+  p->ready = 0;
+  p->initialized = 0;
+
   krt_sys_shutdown(p);
 
   rem_node(&p->krt_node);
@@ -1045,7 +1083,7 @@ krt_reconfigure(struct proto *p, struct proto_config *new)
   if (!krt_sys_reconfigure((struct krt_proto *) p, n, o))
     return 0;
 
-  /* persist needn't be the same */
+  /* persist, graceful restart need not be the same */
   return o->scan_time == n->scan_time && o->learn == n->learn && o->devroutes == n->devroutes;
 }
 
diff --git a/sysdep/unix/krt.h b/sysdep/unix/krt.h
index 99983ccd..2cd23165 100644
--- a/sysdep/unix/krt.h
+++ b/sysdep/unix/krt.h
@@ -48,6 +48,7 @@ struct krt_config {
   int scan_time;		/* How often we re-scan routes */
   int learn;			/* Learn routes from other sources */
   int devroutes;		/* Allow export of device routes */
+  int graceful_restart;		/* Regard graceful restart recovery */
 };
 
 struct krt_proto {
@@ -63,7 +64,8 @@ struct krt_proto {
 #endif
 
   node krt_node;		/* Node in krt_proto_list */
-  int initialized;		/* First scan has already been finished */
+  byte ready;			/* Initial feed has been finished */
+  byte initialized;		/* First scan has been finished */
 };
 
 extern pool *krt_pool;
diff --git a/sysdep/unix/main.c b/sysdep/unix/main.c
index e9217bc9..31094c52 100644
--- a/sysdep/unix/main.c
+++ b/sysdep/unix/main.c
@@ -602,7 +602,7 @@ signal_init(void)
  *	Parsing of command-line arguments
  */
 
-static char *opt_list = "c:dD:ps:P:u:g:f";
+static char *opt_list = "c:dD:ps:P:u:g:fR";
 static int parse_and_exit;
 char *bird_name;
 static char *use_user;
@@ -612,7 +612,7 @@ static int run_in_foreground = 0;
 static void
 usage(void)
 {
-  fprintf(stderr, "Usage: %s [-c <config-file>] [-d] [-D <debug-file>] [-p] [-s <control-socket>] [-P <pid-file>] [-u <user>] [-g <group>] [-f]\n", bird_name);
+  fprintf(stderr, "Usage: %s [-c <config-file>] [-d] [-D <debug-file>] [-p] [-s <control-socket>] [-P <pid-file>] [-u <user>] [-g <group>] [-f] [-R]\n", bird_name);
   exit(1);
 }
 
@@ -723,6 +723,9 @@ parse_args(int argc, char **argv)
       case 'f':
 	run_in_foreground = 1;
 	break;
+      case 'R':
+	graceful_restart_recovery();
+	break;
       default:
 	usage();
       }
@@ -805,6 +808,8 @@ main(int argc, char **argv)
 
   config_commit(conf, RECONFIG_HARD, 0);
 
+  graceful_restart_init();
+
 #ifdef LOCAL_DEBUG
   async_dump_flag = 1;
 #endif