Implemented automatic restart after error with all the timers needed.

3 files changed, 74 insertions, 16 deletions

diff --git a/proto/bgp/bgp.c b/proto/bgp/bgp.c
index 11ad6264..1be652f3 100644
--- a/proto/bgp/bgp.c
+++ b/proto/bgp/bgp.c
@@ -26,6 +26,7 @@ static list bgp_list;			/* List of active BGP instances */
 static char *bgp_state_names[] = { "Idle", "Connect", "Active", "OpenSent", "OpenConfirm", "Established" };
 
 static void bgp_connect(struct bgp_proto *p);
+static void bgp_initiate(struct bgp_proto *p);
 
 void
 bgp_close(struct bgp_proto *p)
@@ -40,7 +41,6 @@ bgp_close(struct bgp_proto *p)
       rfree(bgp_linpool);
       bgp_linpool = NULL;
     }
-  /* FIXME: Automatic restart after errors? */
 }
 
 void
@@ -60,6 +60,7 @@ void
 bgp_close_conn(struct bgp_conn *conn)
 {
   struct bgp_proto *p = conn->bgp;
+  struct bgp_config *cf = p->cf;
 
   DBG("BGP: Closing connection\n");
   conn->packets_to_send = 0;
@@ -72,14 +73,29 @@ bgp_close_conn(struct bgp_conn *conn)
   sk_close(conn->sk);
   conn->sk = NULL;
   conn->state = BS_IDLE;
+  if (conn->error_flag > 1)
+    {
+      if (cf->disable_after_error)
+	p->p.disabled = 1;
+      if (p->last_connect && (bird_clock_t)(p->last_connect + cf->error_amnesia_time) < now)
+	p->startup_delay = 0;
+      if (!p->startup_delay)
+	p->startup_delay = cf->error_delay_time_min;
+      else
+	{
+	  p->startup_delay *= 2;
+	  if (p->startup_delay > cf->error_delay_time_max)
+	    p->startup_delay = cf->error_delay_time_max;
+	}
+    }
   if (conn->primary)
     {
       bgp_close(p);
       p->conn = NULL;
-      if (conn->error_flag)		/* FIXME: Enable automatically? */
-	p->p.disabled = 1;
       proto_notify_state(&p->p, PS_DOWN);
     }
+  else if (conn->error_flag > 1)
+    bgp_initiate(p);
 }
 
 static int
@@ -175,19 +191,12 @@ bgp_keepalive_timeout(timer *t)
 }
 
 static void
-bgp_setup_sk(struct bgp_proto *p, struct bgp_conn *conn, sock *s)
+bgp_setup_conn(struct bgp_proto *p, struct bgp_conn *conn)
 {
   timer *t;
 
-  s->data = conn;
-  s->ttl = p->cf->multihop ? : 1;
-  s->rbsize = BGP_RX_BUFFER_SIZE;
-  s->tbsize = BGP_TX_BUFFER_SIZE;
-  s->err_hook = bgp_sock_err;
-  s->tos = IP_PREC_INTERNET_CONTROL;
-
+  conn->sk = NULL;
   conn->bgp = p;
-  conn->sk = s;
   conn->packets_to_send = 0;
   conn->error_flag = 0;
   conn->primary = 0;
@@ -204,12 +213,25 @@ bgp_setup_sk(struct bgp_proto *p, struct bgp_conn *conn, sock *s)
 }
 
 static void
+bgp_setup_sk(struct bgp_proto *p, struct bgp_conn *conn, sock *s)
+{
+  s->data = conn;
+  s->ttl = p->cf->multihop ? : 1;
+  s->rbsize = BGP_RX_BUFFER_SIZE;
+  s->tbsize = BGP_TX_BUFFER_SIZE;
+  s->err_hook = bgp_sock_err;
+  s->tos = IP_PREC_INTERNET_CONTROL;
+  conn->sk = s;
+}
+
+static void
 bgp_connect(struct bgp_proto *p)	/* Enter Connect state and start establishing connection */
 {
   sock *s;
   struct bgp_conn *conn = &p->outgoing_conn;
 
   DBG("BGP: Connecting\n");
+  p->last_connect = now;
   s = sk_new(p->p.pool);
   s->type = SK_TCP_ACTIVE;
   if (ipa_nonzero(p->cf->source_addr))
@@ -218,6 +240,7 @@ bgp_connect(struct bgp_proto *p)	/* Enter Connect state and start establishing c
     s->saddr = p->local_addr;
   s->daddr = p->cf->remote_ip;
   s->dport = BGP_PORT;
+  bgp_setup_conn(p, conn);
   bgp_setup_sk(p, conn, s);
   s->tx_hook = bgp_connected;
   conn->state = BS_CONNECT;
@@ -230,6 +253,24 @@ bgp_connect(struct bgp_proto *p)	/* Enter Connect state and start establishing c
   bgp_start_timer(conn->connect_retry_timer, p->cf->connect_retry_time);
 }
 
+static void
+bgp_initiate(struct bgp_proto *p)
+{
+  unsigned delay;
+
+  delay = p->cf->start_delay_time;
+  if (p->startup_delay > delay)
+    delay = p->startup_delay;
+  if (delay)
+    {
+      DBG("BGP: Connect delayed by %d seconds\n", delay);
+      bgp_setup_conn(p, &p->outgoing_conn);
+      bgp_start_timer(p->outgoing_conn.connect_retry_timer, delay);
+    }
+  else
+    bgp_connect(p);
+}
+
 static int
 bgp_incoming_connection(sock *sk, int dummy)
 {
@@ -247,6 +288,7 @@ bgp_incoming_connection(sock *sk, int dummy)
 	      DBG("BGP: But one incoming connection already exists, how is that possible?\n");
 	      break;
 	    }
+	  bgp_setup_conn(p, &p->incoming_conn);
 	  bgp_setup_sk(p, &p->incoming_conn, sk);
 	  bgp_send_open(&p->incoming_conn);
 	  return 0;
@@ -285,7 +327,7 @@ bgp_start_neighbor(struct bgp_proto *p)
   if (!bgp_linpool)
     bgp_linpool = lp_new(&root_pool, 4080);
   add_tail(&bgp_list, &p->bgp_node);
-  bgp_connect(p);
+  bgp_initiate(p);
 }
 
 static void
@@ -305,7 +347,6 @@ bgp_neigh_notify(neighbor *n)
       bgp_graceful_close_conn(&p->outgoing_conn);
       bgp_graceful_close_conn(&p->incoming_conn);
       proto_notify_state(&p->p, PS_DOWN);
-      /* FIXME: Remember to delay protocol startup here! */
     }
 }
 
@@ -340,6 +381,7 @@ bgp_start(struct proto *P)
   DBG("BGP: Startup.\n");
   p->outgoing_conn.state = BS_IDLE;
   p->incoming_conn.state = BS_IDLE;
+  p->startup_delay = 0;
 
   /*
    *  Before attempting to create the connection, we need to lock the
@@ -415,7 +457,7 @@ bgp_error(struct bgp_conn *c, unsigned code, unsigned subcode, byte *data, int l
   if (c->error_flag)
     return;
   bgp_log_error(c->bgp, "Error", code, subcode, data, (len > 0) ? len : -len);
-  c->error_flag = 1;
+  c->error_flag = 1 + (code != 6);
   c->notify_code = code;
   c->notify_subcode = subcode;
   c->notify_data = data;
diff --git a/proto/bgp/bgp.h b/proto/bgp/bgp.h
index 85399c1b..5c29488a 100644
--- a/proto/bgp/bgp.h
+++ b/proto/bgp/bgp.h
@@ -28,6 +28,11 @@ struct bgp_config {
   unsigned connect_retry_time;
   unsigned hold_time, initial_hold_time;
   unsigned keepalive_time;
+  unsigned start_delay_time;		/* Minimum delay between connects */
+  unsigned error_amnesia_time;		/* Errors are forgotten after */
+  unsigned error_delay_time_min;	/* Time to wait after an error is detected */
+  unsigned error_delay_time_max;
+  unsigned disable_after_error;		/* Disable the protocol when error is detected */
 };
 
 struct bgp_conn {
@@ -65,6 +70,8 @@ struct bgp_proto {
   struct fib prefix_fib;		/* Prefixes to be sent */
   list bucket_queue;			/* Queue of buckets to send */
   struct bgp_bucket *withdraw_bucket;	/* Withdrawn routes */
+  unsigned startup_delay;		/* Time to delay protocol startup by due to errors */
+  bird_clock_t last_connect;		/* Time of last connect attempt */
 };
 
 struct bgp_prefix {
diff --git a/proto/bgp/config.Y b/proto/bgp/config.Y
index 36356c50..f6701062 100644
--- a/proto/bgp/config.Y
+++ b/proto/bgp/config.Y
@@ -16,6 +16,7 @@ CF_DECLS
 
 CF_KEYWORDS(BGP, LOCAL, NEIGHBOR, AS, HOLD, TIME, CONNECT, RETRY, KEEPALIVE,
 	MULTIHOP, STARTUP, VIA, NEXT, HOP, SELF, DEFAULT, PATH, METRIC,
+	ERROR, START, DELAY, FORGET, WAIT, DISABLE, AFTER,
 	BGP_PATH, BGP_LOCAL_PREF, BGP_MED, BGP_ORIGIN, BGP_NEXT_HOP,
 	BGP_ATOMIC_AGGR, BGP_AGGREGATOR, BGP_COMMUNITY, SOURCE, ADDRESS)
 
@@ -31,7 +32,11 @@ bgp_proto_start: proto_start BGP {
      BGP_CFG->initial_hold_time = 240;
      BGP_CFG->default_med = ~0;		/* RFC 1771 doesn't specify this, draft-09 says ~0 */
      BGP_CFG->compare_path_lengths = 1;
-  }
+     BGP_CFG->start_delay_time = 5;
+     BGP_CFG->error_amnesia_time = 300;
+     BGP_CFG->error_delay_time_min = 60;
+     BGP_CFG->error_delay_time_max = 300;
+ }
  ;
 
 bgp_proto:
@@ -56,6 +61,10 @@ bgp_proto:
  | bgp_proto DEFAULT BGP_MED NUM ';' { BGP_CFG->default_med = $4; }
  | bgp_proto DEFAULT BGP_LOCAL_PREF NUM ';' { BGP_CFG->default_local_pref = $4; }
  | bgp_proto SOURCE ADDRESS IPA ';' { BGP_CFG->source_addr = $4; }
+ | bgp_proto START DELAY TIME NUM ';' { BGP_CFG->start_delay_time = $5; }
+ | bgp_proto ERROR FORGET TIME NUM ';' { BGP_CFG->error_amnesia_time = $5; } 
+ | bgp_proto ERROR WAIT TIME NUM ',' NUM ';' { BGP_CFG->error_delay_time_min = $5; BGP_CFG->error_delay_time_max = $7; }
+ | bgp_proto DISABLE AFTER ERROR ';' { BGP_CFG->disable_after_error = 1; }
  ;
 
 CF_ADDTO(dynamic_attr, BGP_PATH