Merge branch 'master' into mq-filter-stack

19 files changed, 990 insertions, 120 deletions

diff --git a/proto/bgp/attrs.c b/proto/bgp/attrs.c
index 7c6f2ee9..69c4b172 100644
--- a/proto/bgp/attrs.c
+++ b/proto/bgp/attrs.c
@@ -1302,7 +1302,7 @@ bgp_withdraw_bucket(struct bgp_channel *c, struct bgp_bucket *b)
 #define PXH_FN(n,i,h)		h
 
 #define PXH_REHASH		bgp_pxh_rehash
-#define PXH_PARAMS		/8, *2, 2, 2, 8, 20
+#define PXH_PARAMS		/8, *2, 2, 2, 8, 24
 
 
 HASH_DEFINE_REHASH_FN(PXH, struct bgp_prefix)
@@ -1730,7 +1730,7 @@ bgp_rte_better(rte *new, rte *old)
     return 0;
 
   /* RFC 4271 9.1.2.2. g) Compare peer IP adresses */
-  return (ipa_compare(new_bgp->cf->remote_ip, old_bgp->cf->remote_ip) < 0);
+  return ipa_compare(new_bgp->remote_ip, old_bgp->remote_ip) < 0;
 }
 
 
diff --git a/proto/bgp/bgp.c b/proto/bgp/bgp.c
index 8dedde9f..b68575a5 100644
--- a/proto/bgp/bgp.c
+++ b/proto/bgp/bgp.c
@@ -129,6 +129,9 @@ static list bgp_sockets;		/* Global list of listening sockets */
 
 static void bgp_connect(struct bgp_proto *p);
 static void bgp_active(struct bgp_proto *p);
+static void bgp_setup_conn(struct bgp_proto *p, struct bgp_conn *conn);
+static void bgp_setup_sk(struct bgp_conn *conn, sock *s);
+static void bgp_send_open(struct bgp_conn *conn);
 static void bgp_update_bfd(struct bgp_proto *p, int use_bfd);
 
 static int bgp_incoming_connection(sock *sk, uint dummy UNUSED);
@@ -149,7 +152,7 @@ bgp_open(struct bgp_proto *p)
   struct bgp_socket *bs = NULL;
   struct iface *ifa = p->cf->strict_bind ? p->cf->iface : NULL;
   ip_addr addr = p->cf->strict_bind ? p->cf->local_ip :
-    (ipa_is_ip4(p->cf->remote_ip) ? IPA_NONE4 : IPA_NONE6);
+    (p->ipv4 ? IPA_NONE4 : IPA_NONE6);
   uint port = p->cf->local_port;
 
   /* FIXME: Add some global init? */
@@ -272,8 +275,17 @@ bgp_startup(struct bgp_proto *p)
   BGP_TRACE(D_EVENTS, "Started");
   p->start_state = BSS_CONNECT;
 
-  if (!p->cf->passive)
+  if (!p->passive)
     bgp_active(p);
+
+  if (p->postponed_sk)
+  {
+    /* Apply postponed incoming connection */
+    bgp_setup_conn(p, &p->incoming_conn);
+    bgp_setup_sk(&p->incoming_conn, p->postponed_sk);
+    bgp_send_open(&p->incoming_conn);
+    p->postponed_sk = NULL;
+  }
 }
 
 static void
@@ -387,7 +399,7 @@ bgp_close_conn(struct bgp_conn *conn)
 void
 bgp_update_startup_delay(struct bgp_proto *p)
 {
-  struct bgp_config *cf = p->cf;
+  const struct bgp_config *cf = p->cf;
 
   DBG("BGP: Updating startup delay\n");
 
@@ -410,7 +422,7 @@ bgp_update_startup_delay(struct bgp_proto *p)
 }
 
 static void
-bgp_graceful_close_conn(struct bgp_conn *conn, uint subcode, byte *data, uint len)
+bgp_graceful_close_conn(struct bgp_conn *conn, int subcode, byte *data, uint len)
 {
   switch (conn->state)
   {
@@ -426,7 +438,13 @@ bgp_graceful_close_conn(struct bgp_conn *conn, uint subcode, byte *data, uint le
   case BS_OPENSENT:
   case BS_OPENCONFIRM:
   case BS_ESTABLISHED:
-    bgp_error(conn, 6, subcode, data, len);
+    if (subcode < 0)
+    {
+      bgp_conn_enter_close_state(conn);
+      bgp_schedule_packet(conn, NULL, PKT_SCHEDULE_CLOSE);
+    }
+    else
+      bgp_error(conn, 6, subcode, data, len);
     return;
 
   default:
@@ -456,7 +474,7 @@ bgp_decision(void *vp)
   if ((p->p.proto_state == PS_START) &&
       (p->outgoing_conn.state == BS_IDLE) &&
       (p->incoming_conn.state != BS_OPENCONFIRM) &&
-      !p->cf->passive)
+      !p->passive)
     bgp_active(p);
 
   if ((p->p.proto_state == PS_STOP) &&
@@ -465,8 +483,31 @@ bgp_decision(void *vp)
     bgp_down(p);
 }
 
+static struct bgp_proto *
+bgp_spawn(struct bgp_proto *pp, ip_addr remote_ip)
+{
+  struct symbol *sym;
+  char fmt[SYM_MAX_LEN];
+
+  bsprintf(fmt, "%s%%0%dd", pp->cf->dynamic_name, pp->cf->dynamic_name_digits);
+
+  /* This is hack, we would like to share config, but we need to copy it now */
+  new_config = config;
+  cfg_mem = config->mem;
+  conf_this_scope = config->root_scope;
+  sym = cf_default_name(fmt, &(pp->dynamic_name_counter));
+  proto_clone_config(sym, pp->p.cf);
+  new_config = NULL;
+  cfg_mem = NULL;
+
+  /* Just pass remote_ip to bgp_init() */
+  ((struct bgp_config *) sym->proto)->remote_ip = remote_ip;
+
+  return (void *) proto_spawn(sym->proto, 0);
+}
+
 void
-bgp_stop(struct bgp_proto *p, uint subcode, byte *data, uint len)
+bgp_stop(struct bgp_proto *p, int subcode, byte *data, uint len)
 {
   proto_notify_state(&p->p, PS_STOP);
   bgp_graceful_close_conn(&p->outgoing_conn, subcode, data, len);
@@ -491,6 +532,7 @@ bgp_conn_enter_openconfirm_state(struct bgp_conn *conn)
 }
 
 static const struct bgp_af_caps dummy_af_caps = { };
+static const struct bgp_af_caps basic_af_caps = { .ready = 1 };
 
 void
 bgp_conn_enter_established_state(struct bgp_conn *conn)
@@ -503,8 +545,12 @@ bgp_conn_enter_established_state(struct bgp_conn *conn)
   BGP_TRACE(D_EVENTS, "BGP session established");
 
   /* For multi-hop BGP sessions */
-  if (ipa_zero(p->source_addr))
-    p->source_addr = conn->sk->saddr;
+  if (ipa_zero(p->local_ip))
+    p->local_ip = conn->sk->saddr;
+
+  /* For promiscuous sessions */
+  if (!p->remote_as)
+    p->remote_as = conn->received_as;
 
   /* In case of LLv6 is not valid during BGP start */
   if (ipa_zero(p->link_addr) && p->neigh && p->neigh->iface && p->neigh->iface->llv6)
@@ -541,6 +587,13 @@ bgp_conn_enter_established_state(struct bgp_conn *conn)
     const struct bgp_af_caps *loc = bgp_find_af_caps(local, c->afi);
     const struct bgp_af_caps *rem = bgp_find_af_caps(peer,  c->afi);
 
+    /* Use default if capabilities were not announced */
+    if (!local->length && (c->afi == BGP_AF_IPV4))
+      loc = &basic_af_caps;
+
+    if (!peer->length && (c->afi == BGP_AF_IPV4))
+      rem = &basic_af_caps;
+
     /* Ignore AFIs that were not announced in multiprotocol capability */
     if (!loc || !loc->ready)
       loc = &dummy_af_caps;
@@ -880,6 +933,7 @@ bgp_send_open(struct bgp_conn *conn)
   conn->sk->rx_hook = bgp_rx;
   conn->sk->tx_hook = bgp_tx;
   tm_stop(conn->connect_timer);
+  bgp_prepare_capabilities(conn);
   bgp_schedule_packet(conn, NULL, PKT_OPEN);
   bgp_conn_set_state(conn, BS_OPENSENT);
   bgp_start_timer(conn->hold_timer, conn->bgp->cf->initial_hold_time);
@@ -1039,8 +1093,8 @@ bgp_connect(struct bgp_proto *p)	/* Enter Connect state and start establishing c
   DBG("BGP: Connecting\n");
   sock *s = sk_new(p->p.pool);
   s->type = SK_TCP_ACTIVE;
-  s->saddr = p->source_addr;
-  s->daddr = p->cf->remote_ip;
+  s->saddr = p->local_ip;
+  s->daddr = p->remote_ip;
   s->dport = p->cf->remote_port;
   s->iface = p->neigh ? p->neigh->iface : NULL;
   s->vrf = p->p.vrf;
@@ -1075,6 +1129,9 @@ err:
   return;
 }
 
+static inline int bgp_is_dynamic(struct bgp_proto *p)
+{ return ipa_zero(p->remote_ip); }
+
 /**
  * bgp_find_proto - find existing proto for incoming connection
  * @sk: TCP socket
@@ -1083,6 +1140,7 @@ err:
 static struct bgp_proto *
 bgp_find_proto(sock *sk)
 {
+  struct bgp_proto *best = NULL;
   struct bgp_proto *p;
 
   /* sk->iface is valid only if src or dst address is link-local */
@@ -1090,13 +1148,20 @@ bgp_find_proto(sock *sk)
 
   WALK_LIST(p, proto_list)
     if ((p->p.proto == &proto_bgp) &&
-	(p->sock == sk->data) &&
-	ipa_equal(p->cf->remote_ip, sk->daddr) &&
+	(ipa_equal(p->remote_ip, sk->daddr) || bgp_is_dynamic(p)) &&
+	(!p->cf->remote_range || ipa_in_netX(sk->daddr, p->cf->remote_range)) &&
+	(p->p.vrf == sk->vrf) &&
+	(p->cf->local_port == sk->sport) &&
 	(!link || (p->cf->iface == sk->iface)) &&
 	(ipa_zero(p->cf->local_ip) || ipa_equal(p->cf->local_ip, sk->saddr)))
-      return p;
+    {
+      best = p;
 
-  return NULL;
+      if (!bgp_is_dynamic(p))
+	break;
+    }
+
+  return best;
 }
 
 /**
@@ -1175,6 +1240,16 @@ bgp_incoming_connection(sock *sk, uint dummy UNUSED)
     sk_reallocate(sk);
   }
 
+  /* For dynamic BGP, spawn new instance and postpone the socket */
+  if (bgp_is_dynamic(p))
+  {
+    p = bgp_spawn(p, sk->daddr);
+    p->postponed_sk = sk;
+    rmove(sk, p->p.pool);
+    return 0;
+  }
+
+  rmove(sk, p->p.pool);
   bgp_setup_conn(p, &p->incoming_conn);
   bgp_setup_sk(&p->incoming_conn, sk);
   bgp_send_open(&p->incoming_conn);
@@ -1201,11 +1276,11 @@ bgp_start_neighbor(struct bgp_proto *p)
 {
   /* Called only for single-hop BGP sessions */
 
-  if (ipa_zero(p->source_addr))
-    p->source_addr = p->neigh->ifa->ip;
+  if (ipa_zero(p->local_ip))
+    p->local_ip = p->neigh->ifa->ip;
 
-  if (ipa_is_link_local(p->source_addr))
-    p->link_addr = p->source_addr;
+  if (ipa_is_link_local(p->local_ip))
+    p->link_addr = p->local_ip;
   else if (p->neigh->iface->llv6)
     p->link_addr = p->neigh->iface->llv6->ip;
 
@@ -1293,8 +1368,8 @@ bgp_bfd_notify(struct bfd_request *req)
 static void
 bgp_update_bfd(struct bgp_proto *p, int use_bfd)
 {
-  if (use_bfd && !p->bfd_req)
-    p->bfd_req = bfd_request_session(p->p.pool, p->cf->remote_ip, p->source_addr,
+  if (use_bfd && !p->bfd_req && !bgp_is_dynamic(p))
+    p->bfd_req = bfd_request_session(p->p.pool, p->remote_ip, p->local_ip,
 				     p->cf->multihop ? NULL : p->neigh->iface,
 				     bgp_bfd_notify, p);
 
@@ -1375,7 +1450,7 @@ static void
 bgp_start_locked(struct object_lock *lock)
 {
   struct bgp_proto *p = lock->data;
-  struct bgp_config *cf = p->cf;
+  const struct bgp_config *cf = p->cf;
 
   if (p->p.proto_state != PS_START)
   {
@@ -1385,17 +1460,17 @@ bgp_start_locked(struct object_lock *lock)
 
   DBG("BGP: Got lock\n");
 
-  if (cf->multihop)
+  if (cf->multihop || bgp_is_dynamic(p))
   {
     /* Multi-hop sessions do not use neighbor entries */
     bgp_initiate(p);
     return;
   }
 
-  neighbor *n = neigh_find(&p->p, cf->remote_ip, cf->iface, NEF_STICKY);
+  neighbor *n = neigh_find(&p->p, p->remote_ip, cf->iface, NEF_STICKY);
   if (!n)
   {
-    log(L_ERR "%s: Invalid remote address %I%J", p->p.name, cf->remote_ip, cf->iface);
+    log(L_ERR "%s: Invalid remote address %I%J", p->p.name, p->remote_ip, cf->iface);
     /* As we do not start yet, we can just disable protocol */
     p->p.disabled = 1;
     bgp_store_error(p, NULL, BE_MISC, BEM_INVALID_NEXT_HOP);
@@ -1406,7 +1481,7 @@ bgp_start_locked(struct object_lock *lock)
   p->neigh = n;
 
   if (n->scope <= 0)
-    BGP_TRACE(D_EVENTS, "Waiting for %I%J to become my neighbor", cf->remote_ip, cf->iface);
+    BGP_TRACE(D_EVENTS, "Waiting for %I%J to become my neighbor", p->remote_ip, cf->iface);
   else if (p->cf->check_link && !(n->iface->flags & IF_LINK_UP))
     BGP_TRACE(D_EVENTS, "Waiting for link on %s", n->iface->name);
   else
@@ -1417,14 +1492,29 @@ static int
 bgp_start(struct proto *P)
 {
   struct bgp_proto *p = (struct bgp_proto *) P;
-  struct object_lock *lock;
+  const struct bgp_config *cf = p->cf;
+
+  p->local_ip = cf->local_ip;
+  p->local_as = cf->local_as;
+  p->remote_as = cf->remote_as;
+  p->public_as = cf->local_as;
+
+  /* For dynamic BGP childs, remote_ip is already set */
+  if (ipa_nonzero(cf->remote_ip))
+    p->remote_ip = cf->remote_ip;
+
+  /* Confederation ID is used for truly external peers */
+  if (p->cf->confederation && !p->is_interior)
+    p->public_as = cf->confederation;
+
+  p->passive = cf->passive || bgp_is_dynamic(p);
 
-  DBG("BGP: Startup.\n");
   p->start_state = BSS_PREPARE;
   p->outgoing_conn.state = BS_IDLE;
   p->incoming_conn.state = BS_IDLE;
   p->neigh = NULL;
   p->bfd_req = NULL;
+  p->postponed_sk = NULL;
   p->gr_ready = 0;
   p->gr_active_num = 0;
 
@@ -1437,7 +1527,6 @@ bgp_start(struct proto *P)
     p->rr_cluster_id = p->cf->rr_cluster_id ? p->cf->rr_cluster_id : p->local_id;
 
   p->remote_id = 0;
-  p->source_addr = p->cf->local_ip;
   p->link_addr = IPA_NONE;
 
   /* Lock all channels when in GR recovery mode */
@@ -1452,9 +1541,9 @@ bgp_start(struct proto *P)
    * Before attempting to create the connection, we need to lock the port,
    * so that we are the only instance attempting to talk with that neighbor.
    */
-
+  struct object_lock *lock;
   lock = p->lock = olock_new(P->pool);
-  lock->addr = p->cf->remote_ip;
+  lock->addr = p->remote_ip;
   lock->port = p->cf->remote_port;
   lock->iface = p->cf->iface;
   lock->vrf = p->cf->iface ? NULL : p->p.vrf;
@@ -1472,7 +1561,7 @@ static int
 bgp_shutdown(struct proto *P)
 {
   struct bgp_proto *p = (struct bgp_proto *) P;
-  uint subcode = 0;
+  int subcode = 0;
 
   char *message = NULL;
   byte *data = NULL;
@@ -1493,6 +1582,7 @@ bgp_shutdown(struct proto *P)
 
   case PDC_CMD_DISABLE:
   case PDC_CMD_SHUTDOWN:
+  shutdown:
     subcode = 2; // Errcode 6, 2 - administrative shutdown
     message = P->message;
     break;
@@ -1502,6 +1592,14 @@ bgp_shutdown(struct proto *P)
     message = P->message;
     break;
 
+  case PDC_CMD_GR_DOWN:
+    if ((p->cf->gr_mode != BGP_GR_ABLE) &&
+	(p->cf->llgr_mode != BGP_LLGR_ABLE))
+      goto shutdown;
+
+    subcode = -1; // Do not send NOTIFICATION, just close the connection
+    break;
+
   case PDC_RX_LIMIT_HIT:
   case PDC_IN_LIMIT_HIT:
     subcode = 1; // Errcode 6, 1 - max number of prefixes reached
@@ -1528,7 +1626,7 @@ bgp_shutdown(struct proto *P)
   if (message)
   {
     uint msg_len = strlen(message);
-    msg_len = MIN(msg_len, 128);
+    msg_len = MIN(msg_len, 255);
 
     /* Buffer will be freed automatically by protocol shutdown */
     data = mb_alloc(p->p.pool, msg_len + 1);
@@ -1562,17 +1660,21 @@ bgp_init(struct proto_config *CF)
   P->rte_modify = bgp_rte_modify_stale;
 
   p->cf = cf;
-  p->local_as = cf->local_as;
-  p->remote_as = cf->remote_as;
-  p->public_as = cf->local_as;
   p->is_internal = (cf->local_as == cf->remote_as);
   p->is_interior = p->is_internal || cf->confederation_member;
   p->rs_client = cf->rs_client;
   p->rr_client = cf->rr_client;
 
-  /* Confederation ID is used for truly external peers */
-  if (cf->confederation && !p->is_interior)
-    p->public_as = cf->confederation;
+  p->ipv4 = ipa_nonzero(cf->remote_ip) ?
+    ipa_is_ip4(cf->remote_ip) :
+    (cf->remote_range && (cf->remote_range->type == NET_IP4));
+
+  p->remote_ip = cf->remote_ip;
+  p->remote_as = cf->remote_as;
+
+  /* Hack: We use cf->remote_ip just to pass remote_ip from bgp_spawn() */
+  if (cf->c.parent)
+    cf->remote_ip = IPA_NONE;
 
   /* Add all channels */
   struct bgp_channel_config *cc;
@@ -1604,7 +1706,7 @@ bgp_channel_start(struct channel *C)
 {
   struct bgp_proto *p = (void *) C->proto;
   struct bgp_channel *c = (void *) C;
-  ip_addr src = p->source_addr;
+  ip_addr src = p->local_ip;
 
   if (c->igp_table_ip4)
     rt_lock_table(c->igp_table_ip4);
@@ -1745,14 +1847,19 @@ void
 bgp_postconfig(struct proto_config *CF)
 {
   struct bgp_config *cf = (void *) CF;
-  int internal = (cf->local_as == cf->remote_as);
-  int interior = internal || cf->confederation_member;
 
   /* Do not check templates at all */
   if (cf->c.class == SYM_TEMPLATE)
     return;
 
 
+  /* Handle undefined remote_as, zero should mean unspecified external */
+  if (!cf->remote_as && (cf->peer_type == BGP_PT_INTERNAL))
+    cf->remote_as = cf->local_as;
+
+  int internal = (cf->local_as == cf->remote_as);
+  int interior = internal || cf->confederation_member;
+
   /* EBGP direct by default, IBGP multihop by default */
   if (cf->multihop < 0)
     cf->multihop = internal ? 64 : 0;
@@ -1769,11 +1876,20 @@ bgp_postconfig(struct proto_config *CF)
   if (!cf->local_as)
     cf_error("Local AS number must be set");
 
-  if (ipa_zero(cf->remote_ip))
+  if (ipa_zero(cf->remote_ip) && !cf->remote_range)
     cf_error("Neighbor must be configured");
 
-  if (!cf->remote_as)
-    cf_error("Remote AS number must be set");
+  if (ipa_zero(cf->local_ip) && cf->strict_bind)
+    cf_error("Local address must be configured for strict bind");
+
+  if (!cf->remote_as && !cf->peer_type)
+    cf_error("Remote AS number (or peer type) must be set");
+
+  if ((cf->peer_type == BGP_PT_INTERNAL) && !internal)
+    cf_error("IBGP cannot have different ASNs");
+
+  if ((cf->peer_type == BGP_PT_EXTERNAL) &&  internal)
+    cf_error("EBGP cannot have the same ASNs");
 
   if (!cf->iface && (ipa_is_link_local(cf->local_ip) ||
 		     ipa_is_link_local(cf->remote_ip)))
@@ -1885,8 +2001,8 @@ static int
 bgp_reconfigure(struct proto *P, struct proto_config *CF)
 {
   struct bgp_proto *p = (void *) P;
-  struct bgp_config *new = (void *) CF;
-  struct bgp_config *old = p->cf;
+  const struct bgp_config *new = (void *) CF;
+  const struct bgp_config *old = p->cf;
 
   if (proto_get_router_id(CF) != p->local_id)
     return 0;
@@ -1896,7 +2012,12 @@ bgp_reconfigure(struct proto *P, struct proto_config *CF)
 		     // password item is last and must be checked separately
 		     OFFSETOF(struct bgp_config, password) - sizeof(struct proto_config))
     && ((!old->password && !new->password)
-	|| (old->password && new->password && !strcmp(old->password, new->password)));
+	|| (old->password && new->password && !strcmp(old->password, new->password)))
+    && ((!old->remote_range && !new->remote_range)
+	|| (old->remote_range && new->remote_range && net_equal(old->remote_range, new->remote_range)))
+    && ((!old->dynamic_name && !new->dynamic_name)
+	|| (old->dynamic_name && new->dynamic_name && !strcmp(old->dynamic_name, new->dynamic_name)))
+    && (old->dynamic_name_digits == new->dynamic_name_digits);
 
   /* FIXME: Move channel reconfiguration to generic protocol code ? */
   struct channel *C, *C2;
@@ -1926,6 +2047,9 @@ bgp_reconfigure(struct proto *P, struct proto_config *CF)
   if (same)
     p->cf = new;
 
+  /* Reset name counter */
+  p->dynamic_name_counter = 0;
+
   return same;
 }
 
@@ -2056,7 +2180,7 @@ bgp_state_dsc(struct bgp_proto *p)
     return "Down";
 
   int state = MAX(p->incoming_conn.state, p->outgoing_conn.state);
-  if ((state == BS_IDLE) && (p->start_state >= BSS_CONNECT) && p->cf->passive)
+  if ((state == BS_IDLE) && (p->start_state >= BSS_CONNECT) && p->passive)
     return "Passive";
 
   return bgp_state_names[state];
@@ -2232,8 +2356,14 @@ bgp_show_proto_info(struct proto *P)
   struct bgp_proto *p = (struct bgp_proto *) P;
 
   cli_msg(-1006, "  BGP state:          %s", bgp_state_dsc(p));
-  cli_msg(-1006, "    Neighbor address: %I%J", p->cf->remote_ip, p->cf->iface);
+
+  if (bgp_is_dynamic(p) && p->cf->remote_range)
+    cli_msg(-1006, "    Neighbor range:   %N", p->cf->remote_range);
+  else
+    cli_msg(-1006, "    Neighbor address: %I%J", p->remote_ip, p->cf->iface);
+
   cli_msg(-1006, "    Neighbor AS:      %u", p->remote_as);
+  cli_msg(-1006, "    Local AS:         %u", p->cf->local_as);
 
   if (p->gr_active_num)
     cli_msg(-1006, "    Neighbor graceful restart active");
@@ -2269,7 +2399,7 @@ bgp_show_proto_info(struct proto *P)
 	    p->rr_client ? " route-reflector" : "",
 	    p->rs_client ? " route-server" : "",
 	    p->as4_session ? " AS4" : "");
-    cli_msg(-1006, "    Source address:   %I", p->source_addr);
+    cli_msg(-1006, "    Source address:   %I", p->local_ip);
     cli_msg(-1006, "    Hold timer:       %t/%u",
 	    tm_remains(p->conn->hold_timer), p->conn->hold_time);
     cli_msg(-1006, "    Keepalive timer:  %t/%u",
diff --git a/proto/bgp/bgp.h b/proto/bgp/bgp.h
index b604c7aa..075e1bb9 100644
--- a/proto/bgp/bgp.h
+++ b/proto/bgp/bgp.h
@@ -83,6 +83,7 @@ struct bgp_config {
   struct iface *iface;			/* Interface for link-local addresses */
   u16 local_port;			/* Local listening port */
   u16 remote_port; 			/* Neighbor destination port */
+  int peer_type;			/* Internal or external BGP (BGP_PT_*, optional) */
   int multihop;				/* Number of hops if multihop */
   int strict_bind;			/* Bind listening socket to local address */
   int ttl_security;			/* Enable TTL security [RFC 5082] */
@@ -123,6 +124,9 @@ struct bgp_config {
   u32 disable_after_cease;		/* Disable it when cease is received, bitfield */
 
   char *password;			/* Password used for MD5 authentication */
+  net_addr *remote_range;		/* Allowed neighbor range for dynamic BGP */
+  char *dynamic_name;			/* Name pattern for dynamic BGP */
+  int dynamic_name_digits;		/* Minimum number of digits for dynamic names */
   int check_link;			/* Use iface link state for liveness detection */
   int bfd;				/* Use BFD for liveness detection */
 };
@@ -136,6 +140,7 @@ struct bgp_channel_config {
   ip_addr next_hop_addr;		/* Local address for NEXT_HOP attribute */
   u8 next_hop_self;			/* Always set next hop to local IP address (NH_*) */
   u8 next_hop_keep;			/* Do not modify next hop attribute (NH_*) */
+  u8 mandatory;				/* Channel is mandatory in capability negotiation */
   u8 missing_lladdr;			/* What we will do when we don' know link-local addr, see MLL_* */
   u8 gw_mode;				/* How we compute route gateway from next_hop attr, see GW_* */
   u8 secondary;				/* Accept also non-best routes (i.e. RA_ACCEPTED) */
@@ -151,6 +156,9 @@ struct bgp_channel_config {
   struct rtable_config *igp_table_ip6;	/* Table for recursive IPv6 next hop lookups */
 };
 
+#define BGP_PT_INTERNAL		1
+#define BGP_PT_EXTERNAL		2
+
 #define NH_NO			0
 #define NH_ALL			1
 #define NH_IBGP			2
@@ -213,8 +221,11 @@ struct bgp_caps {
   u16 gr_time;				/* Graceful restart time in seconds */
 
   u8 llgr_aware;			/* Long-lived GR capability, RFC draft */
+  u8 any_ext_next_hop;			/* Bitwise OR of per-AF ext_next_hop */
+  u8 any_add_path;			/* Bitwise OR of per-AF add_path */
 
   u16 af_count;				/* Number of af_data items */
+  u16 length;				/* Length of capabilities in OPEN msg */
 
   struct bgp_af_caps af_data[0];	/* Per-AF capability data */
 };
@@ -235,6 +246,7 @@ struct bgp_conn {
   u8 state;				/* State of connection state machine */
   u8 as4_session;			/* Session uses 4B AS numbers in AS_PATH (both sides support it) */
   u8 ext_messages;			/* Session uses extended message length */
+  u32 received_as;			/* ASN received in OPEN message */
 
   struct bgp_caps *local_caps;
   struct bgp_caps *remote_caps;
@@ -254,18 +266,21 @@ struct bgp_conn {
 
 struct bgp_proto {
   struct proto p;
-  struct bgp_config *cf;		/* Shortcut to BGP configuration */
+  const struct bgp_config *cf;		/* Shortcut to BGP configuration */
+  ip_addr local_ip, remote_ip;
   u32 local_as, remote_as;
   u32 public_as;			/* Externally visible ASN (local_as or confederation id) */
   u32 local_id;				/* BGP identifier of this router */
   u32 remote_id;			/* BGP identifier of the neighbor */
   u32 rr_cluster_id;			/* Route reflector cluster ID */
-  int start_state;			/* Substates that partitions BS_START */
+  u8 start_state;			/* Substates that partitions BS_START */
   u8 is_internal;			/* Internal BGP session (local_as == remote_as) */
   u8 is_interior;			/* Internal or intra-confederation BGP session */
   u8 as4_session;			/* Session uses 4B AS numbers in AS_PATH (both sides support it) */
   u8 rr_client;				/* Whether neighbor is RR client of me */
   u8 rs_client;				/* Whether neighbor is RS client of me */
+  u8 ipv4;				/* Use IPv4 connection, i.e. remote_ip is IPv4 */
+  u8 passive;				/* Do not initiate outgoing connection */
   u8 route_refresh;			/* Route refresh allowed to send [RFC 2918] */
   u8 enhanced_refresh;			/* Enhanced refresh is negotiated [RFC 7313] */
   u8 gr_ready;				/* Neighbor could do graceful restart */
@@ -282,11 +297,12 @@ struct bgp_proto {
   struct neighbor *neigh;		/* Neighbor entry corresponding to remote ip, NULL if multihop */
   struct bgp_socket *sock;		/* Shared listening socket */
   struct bfd_request *bfd_req;		/* BFD request, if BFD is used */
-  ip_addr source_addr;			/* Local address used as an advertised next hop */
-  ip_addr link_addr;			/* Link-local version of source_addr */
+  struct birdsock *postponed_sk;	/* Postponed incoming socket for dynamic BGP */
+  ip_addr link_addr;			/* Link-local version of local_ip */
   event *event;				/* Event for respawning and shutting process */
   timer *startup_timer;			/* Timer used to delay protocol startup due to previous errors (startup_delay) */
   timer *gr_timer;			/* Timer waiting for reestablishment after graceful restart */
+  int dynamic_name_counter;		/* Counter for dynamic BGP names */
   uint startup_delay;			/* Delay (in seconds) of protocol startup due to previous errors */
   btime last_proto_error;		/* Time of last error that leads to protocol stop */
   u8 last_error_class; 			/* Error class of last error */
@@ -472,7 +488,7 @@ void bgp_graceful_restart_done(struct bgp_channel *c);
 void bgp_refresh_begin(struct bgp_channel *c);
 void bgp_refresh_end(struct bgp_channel *c);
 void bgp_store_error(struct bgp_proto *p, struct bgp_conn *c, u8 class, u32 code);
-void bgp_stop(struct bgp_proto *p, uint subcode, byte *data, uint len);
+void bgp_stop(struct bgp_proto *p, int subcode, byte *data, uint len);
 
 struct rte_source *bgp_find_source(struct bgp_proto *p, u32 path_id);
 struct rte_source *bgp_get_source(struct bgp_proto *p, u32 path_id);
@@ -549,6 +565,7 @@ void bgp_get_route_info(struct rte *, byte *buf);
 /* packets.c */
 
 void bgp_dump_state_change(struct bgp_conn *conn, uint old, uint new);
+void bgp_prepare_capabilities(struct bgp_conn *conn);
 const struct bgp_af_desc *bgp_get_af_desc(u32 afi);
 const struct bgp_af_caps *bgp_find_af_caps(struct bgp_caps *caps, u32 afi);
 void bgp_schedule_packet(struct bgp_conn *conn, struct bgp_channel *c, int type);
diff --git a/proto/bgp/config.Y b/proto/bgp/config.Y
index ac8d024a..bbc7d9a4 100644
--- a/proto/bgp/config.Y
+++ b/proto/bgp/config.Y
@@ -29,7 +29,8 @@ CF_KEYWORDS(BGP, LOCAL, NEIGHBOR, AS, HOLD, TIME, CONNECT, RETRY, KEEPALIVE,
 	SECURITY, DETERMINISTIC, SECONDARY, ALLOW, BFD, ADD, PATHS, RX, TX,
 	GRACEFUL, RESTART, AWARE, CHECK, LINK, PORT, EXTENDED, MESSAGES, SETKEY,
 	STRICT, BIND, CONFEDERATION, MEMBER, MULTICAST, FLOW4, FLOW6, LONG,
-	LIVED, STALE, IMPORT, IBGP, EBGP)
+	LIVED, STALE, IMPORT, IBGP, EBGP, MANDATORY, INTERNAL, EXTERNAL,
+	DYNAMIC, RANGE, NAME, DIGITS)
 
 %type <i> bgp_nh
 %type <i32> bgp_afi
@@ -68,6 +69,7 @@ bgp_proto_start: proto_start BGP {
      BGP_CFG->llgr_mode = -1;
      BGP_CFG->llgr_time = 3600;
      BGP_CFG->setkey = 1;
+     BGP_CFG->dynamic_name = "dynbgp";
      BGP_CFG->check_link = -1;
    }
  ;
@@ -82,6 +84,8 @@ bgp_nbr_opts:
    /* empty */
  | bgp_nbr_opts PORT expr { BGP_CFG->remote_port = $3; if (($3<1) || ($3>65535)) cf_error("Invalid port number"); }
  | bgp_nbr_opts AS expr { BGP_CFG->remote_as = $3; }
+ | bgp_nbr_opts INTERNAL { BGP_CFG->peer_type = BGP_PT_INTERNAL; }
+ | bgp_nbr_opts EXTERNAL { BGP_CFG->peer_type = BGP_PT_EXTERNAL; }
  ;
 
 bgp_cease_mask:
@@ -118,11 +122,18 @@ bgp_proto:
    }
  | bgp_proto NEIGHBOR bgp_nbr_opts ';'
  | bgp_proto NEIGHBOR ipa ipa_scope bgp_nbr_opts ';' {
-     if (ipa_nonzero(BGP_CFG->remote_ip))
+     if (ipa_nonzero(BGP_CFG->remote_ip) || BGP_CFG->remote_range)
        cf_error("Only one neighbor per BGP instance is allowed");
      BGP_CFG->remote_ip = $3;
      if ($4) BGP_CFG->iface = $4;
    }
+ | bgp_proto NEIGHBOR RANGE net_ip bgp_nbr_opts ';' {
+     if (ipa_nonzero(BGP_CFG->remote_ip) || BGP_CFG->remote_range)
+       cf_error("Only one neighbor per BGP instance is allowed");
+     net_addr *n = cfg_alloc($4.length);
+     net_copy(n, &($4));
+     BGP_CFG->remote_range = n;
+   }
  | bgp_proto INTERFACE TEXT ';' { BGP_CFG->iface = if_get_by_name($3); }
  | bgp_proto RR CLUSTER ID idval ';' { BGP_CFG->rr_cluster_id = $5; }
  | bgp_proto RR CLIENT bool ';' { BGP_CFG->rr_client = $4; }
@@ -134,6 +145,12 @@ bgp_proto:
  | bgp_proto DIRECT ';' { BGP_CFG->multihop = 0; }
  | bgp_proto MULTIHOP ';' { BGP_CFG->multihop = 64; }
  | bgp_proto MULTIHOP expr ';' { BGP_CFG->multihop = $3; if (($3<1) || ($3>255)) cf_error("Multihop must be in range 1-255"); }
+ | bgp_proto DYNAMIC NAME text ';' {
+     if (strchr($4, '%')) cf_error("Forbidden character '%%' in dynamic name");
+     if (strlen($4) > (SYM_MAX_LEN - 16)) cf_error("Dynamic name too long");
+     BGP_CFG->dynamic_name = $4;
+   }
+ | bgp_proto DYNAMIC NAME DIGITS expr ';' { BGP_CFG->dynamic_name_digits = $5; if ($5>10) cf_error("Dynamic name digits must be at most 10"); }
  | bgp_proto STRICT BIND bool ';' { BGP_CFG->strict_bind = $4; }
  | bgp_proto PATH METRIC bool ';' { BGP_CFG->compare_path_lengths = $4; }
  | bgp_proto MED METRIC bool ';' { BGP_CFG->med_metric = $4; }
@@ -223,6 +240,7 @@ bgp_channel_item:
  | NEXT HOP ADDRESS ipa { BGP_CC->next_hop_addr = $4; }
  | NEXT HOP SELF bgp_nh { BGP_CC->next_hop_self = $4; }
  | NEXT HOP KEEP bgp_nh { BGP_CC->next_hop_keep = $4; }
+ | MANDATORY bool { BGP_CC->mandatory = $2; }
  | MISSING LLADDR SELF { BGP_CC->missing_lladdr = MLL_SELF; }
  | MISSING LLADDR DROP { BGP_CC->missing_lladdr = MLL_DROP; }
  | MISSING LLADDR IGNORE { BGP_CC->missing_lladdr = MLL_IGNORE; }
diff --git a/proto/bgp/packets.c b/proto/bgp/packets.c
index 2b5cc440..daa88630 100644
--- a/proto/bgp/packets.c
+++ b/proto/bgp/packets.c
@@ -100,7 +100,7 @@ init_mrt_bgp_data(struct bgp_conn *conn, struct mrt_bgp_data *d)
   d->peer_as = p->remote_as;
   d->local_as = p->local_as;
   d->index = (p->neigh && p->neigh->iface) ? p->neigh->iface->index : 0;
-  d->af = ipa_is_ip4(p->cf->remote_ip) ? BGP_AFI_IPV4 : BGP_AFI_IPV6;
+  d->af = ipa_is_ip4(p->remote_ip) ? BGP_AFI_IPV4 : BGP_AFI_IPV6;
   d->peer_ip = conn->sk ? conn->sk->daddr : IPA_NONE;
   d->local_ip = conn->sk ? conn->sk->saddr : IPA_NONE;
   d->as4 = p_ok ? p->as4_session : 0;
@@ -208,19 +208,22 @@ bgp_af_caps_cmp(const void *X, const void *Y)
 }
 
 
-static byte *
-bgp_write_capabilities(struct bgp_conn *conn, byte *buf)
+void
+bgp_prepare_capabilities(struct bgp_conn *conn)
 {
   struct bgp_proto *p = conn->bgp;
   struct bgp_channel *c;
   struct bgp_caps *caps;
   struct bgp_af_caps *ac;
-  uint any_ext_next_hop = 0;
-  uint any_add_path = 0;
-  byte *data;
 
-  /* Prepare bgp_caps structure */
+  if (!p->cf->capabilities)
+  {
+    /* Just prepare empty local_caps */
+    conn->local_caps = mb_allocz(p->p.pool, sizeof(struct bgp_caps));
+    return;
+  }
 
+  /* Prepare bgp_caps structure */
   int n = list_length(&p->p.channels);
   caps = mb_allocz(p->p.pool, sizeof(struct bgp_caps) + n * sizeof(struct bgp_af_caps));
   conn->local_caps = caps;
@@ -251,10 +254,10 @@ bgp_write_capabilities(struct bgp_conn *conn, byte *buf)
     ac->ready = 1;
 
     ac->ext_next_hop = bgp_channel_is_ipv4(c) && c->cf->ext_next_hop;
-    any_ext_next_hop |= ac->ext_next_hop;
+    caps->any_ext_next_hop |= ac->ext_next_hop;
 
     ac->add_path = c->cf->add_path;
-    any_add_path |= ac->add_path;
+    caps->any_add_path |= ac->add_path;
 
     if (c->cf->gr_able)
     {
@@ -276,7 +279,16 @@ bgp_write_capabilities(struct bgp_conn *conn, byte *buf)
 
   /* Sort capability fields by AFI/SAFI */
   qsort(caps->af_data, caps->af_count, sizeof(struct bgp_af_caps), bgp_af_caps_cmp);
+}
 
+static byte *
+bgp_write_capabilities(struct bgp_conn *conn, byte *buf)
+{
+  struct bgp_proto *p = conn->bgp;
+  struct bgp_caps *caps = conn->local_caps;
+  struct bgp_af_caps *ac;
+  byte *buf_head = buf;
+  byte *data;
 
   /* Create capability list in buffer */
 
@@ -301,7 +313,7 @@ bgp_write_capabilities(struct bgp_conn *conn, byte *buf)
     *buf++ = 0;			/* Capability data length */
   }
 
-  if (any_ext_next_hop)
+  if (caps->any_ext_next_hop)
   {
     *buf++ = 5;			/* Capability 5: Support for extended next hop */
     *buf++ = 0;			/* Capability data length, will be fixed later */
@@ -353,7 +365,7 @@ bgp_write_capabilities(struct bgp_conn *conn, byte *buf)
     buf += 4;
   }
 
-  if (any_add_path)
+  if (caps->any_add_path)
   {
     *buf++ = 69;		/* Capability 69: Support for ADD-PATH */
     *buf++ = 0;			/* Capability data length, will be fixed later */
@@ -394,6 +406,8 @@ bgp_write_capabilities(struct bgp_conn *conn, byte *buf)
     data[-1] = buf - data;
   }
 
+  caps->length = buf - buf_head;
+
   return buf;
 }
 
@@ -405,6 +419,8 @@ bgp_read_capabilities(struct bgp_conn *conn, struct bgp_caps *caps, byte *pos, i
   int i, cl;
   u32 af;
 
+  caps->length += len;
+
   while (len > 0)
   {
     if (len < 2 || len < (2 + pos[1]))
@@ -569,6 +585,42 @@ err:
 }
 
 static int
+bgp_check_capabilities(struct bgp_conn *conn)
+{
+  struct bgp_proto *p = conn->bgp;
+  struct bgp_caps *local = conn->local_caps;
+  struct bgp_caps *remote = conn->remote_caps;
+  struct bgp_channel *c;
+  int count = 0;
+
+  /* This is partially overlapping with bgp_conn_enter_established_state(),
+     but we need to run this just after we receive OPEN message */
+
+  WALK_LIST(c, p->p.channels)
+  {
+    const struct bgp_af_caps *loc = bgp_find_af_caps(local,  c->afi);
+    const struct bgp_af_caps *rem = bgp_find_af_caps(remote, c->afi);
+
+    /* Find out whether this channel will be active */
+    int active = loc && loc->ready &&
+      ((rem && rem->ready) || (!remote->length && (c->afi == BGP_AF_IPV4)));
+
+    /* Mandatory must be active */
+    if (c->cf->mandatory && !active)
+      return 0;
+
+    if (active)
+      count++;
+  }
+
+  /* We need at least one channel active */
+  if (!count)
+    return 0;
+
+  return 1;
+}
+
+static int
 bgp_read_options(struct bgp_conn *conn, byte *pos, int len)
 {
   struct bgp_proto *p = conn->bgp;
@@ -635,9 +687,6 @@ bgp_create_open(struct bgp_conn *conn, byte *buf)
   }
   else
   {
-    /* Prepare empty local_caps */
-    conn->local_caps = mb_allocz(p->p.pool, sizeof(struct bgp_caps));
-
     buf[9] = 0;			/* No optional parameters */
     return buf + 10;
   }
@@ -678,6 +727,10 @@ bgp_rx_open(struct bgp_conn *conn, byte *pkt, uint len)
   if (!id || (p->is_internal && id == p->local_id))
   { bgp_error(conn, 2, 3, pkt+24, -4); return; }
 
+  /* RFC 5492 4 - check for required capabilities */
+  if (p->cf->capabilities && !bgp_check_capabilities(conn))
+  { bgp_error(conn, 2, 7, NULL, 0); return; }
+
   struct bgp_caps *caps = conn->remote_caps;
 
   if (caps->as4_support)
@@ -687,13 +740,18 @@ bgp_rx_open(struct bgp_conn *conn, byte *pkt, uint len)
     if ((as4 != asn) && (asn != AS_TRANS))
       log(L_WARN "%s: Peer advertised inconsistent AS numbers", p->p.name);
 
-    if (as4 != p->remote_as)
+    /* When remote ASN is unspecified, it must be external one */
+    if (p->remote_as ? (as4 != p->remote_as) : (as4 == p->local_as))
     { as4 = htonl(as4); bgp_error(conn, 2, 2, (byte *) &as4, 4); return; }
+
+    conn->received_as = as4;
   }
   else
   {
-    if (asn != p->remote_as)
+    if (p->remote_as ? (asn != p->remote_as) : (asn == p->local_as))
     { bgp_error(conn, 2, 2, pkt+20, 2); return; }
+
+    conn->received_as = asn;
   }
 
   /* Check the other connection */
@@ -962,7 +1020,7 @@ bgp_update_next_hop_ip(struct bgp_export_state *s, eattr *a, ea_list **to)
     WITHDRAW(NO_NEXT_HOP);
 
   ip_addr *nh = (void *) a->u.ptr->data;
-  ip_addr peer = s->proto->cf->remote_ip;
+  ip_addr peer = s->proto->remote_ip;
   uint len = a->u.ptr->length;
 
   /* Forbid zero next hop */
@@ -2280,7 +2338,7 @@ bgp_decode_nlri(struct bgp_parse_state *s, u32 afi, byte *nlri, uint len, ea_lis
 
     a->source = RTS_BGP;
     a->scope = SCOPE_UNIVERSE;
-    a->from = s->proto->cf->remote_ip;
+    a->from = s->proto->remote_ip;
     a->eattrs = ea;
 
     c->desc->decode_next_hop(s, nh, nh_len, a);
@@ -2634,6 +2692,12 @@ bgp_fire_tx(struct bgp_conn *conn)
     end = bgp_create_notification(conn, pkt);
     return bgp_send(conn, PKT_NOTIFICATION, end - buf);
   }
+  else if (s & (1 << PKT_OPEN))
+  {
+    conn->packets_to_send &= ~(1 << PKT_OPEN);
+    end = bgp_create_open(conn, pkt);
+    return bgp_send(conn, PKT_OPEN, end - buf);
+  }
   else if (s & (1 << PKT_KEEPALIVE))
   {
     conn->packets_to_send &= ~(1 << PKT_KEEPALIVE);
@@ -2641,12 +2705,6 @@ bgp_fire_tx(struct bgp_conn *conn)
     bgp_start_timer(conn->keepalive_timer, conn->keepalive_time);
     return bgp_send(conn, PKT_KEEPALIVE, BGP_HEADER_LENGTH);
   }
-  else if (s & (1 << PKT_OPEN))
-  {
-    conn->packets_to_send &= ~(1 << PKT_OPEN);
-    end = bgp_create_open(conn, pkt);
-    return bgp_send(conn, PKT_OPEN, end - buf);
-  }
   else while (conn->channels_to_send)
   {
     c = bgp_get_channel_to_send(p, conn);
@@ -2731,15 +2789,18 @@ bgp_schedule_packet(struct bgp_conn *conn, struct bgp_channel *c, int type)
   if ((conn->sk->tpos == conn->sk->tbuf) && !ev_active(conn->tx_ev))
     ev_schedule(conn->tx_ev);
 }
-
 void
 bgp_kick_tx(void *vconn)
 {
   struct bgp_conn *conn = vconn;
 
   DBG("BGP: kicking TX\n");
-  while (bgp_fire_tx(conn) > 0)
+  uint max = 1024;
+  while (--max && (bgp_fire_tx(conn) > 0))
     ;
+
+  if (!max && !ev_active(conn->tx_ev))
+    ev_schedule(conn->tx_ev);
 }
 
 void
@@ -2748,8 +2809,12 @@ bgp_tx(sock *sk)
   struct bgp_conn *conn = sk->data;
 
   DBG("BGP: TX hook\n");
-  while (bgp_fire_tx(conn) > 0)
+  uint max = 1024;
+  while (--max && (bgp_fire_tx(conn) > 0))
     ;
+
+  if (!max && !ev_active(conn->tx_ev))
+    ev_schedule(conn->tx_ev);
 }
 
 
@@ -2835,7 +2900,7 @@ bgp_handle_message(struct bgp_proto *p, byte *data, uint len, byte **bp)
     return 1;
 
   /* Handle proper message */
-  if ((msg_len > 128) && (msg_len + 1 > len))
+  if ((msg_len > 255) && (msg_len + 1 > len))
     return 0;
 
   /* Some elementary cleanup */
diff --git a/proto/mrt/mrt.c b/proto/mrt/mrt.c
index e4f1acea..7a396a84 100644
--- a/proto/mrt/mrt.c
+++ b/proto/mrt/mrt.c
@@ -361,7 +361,7 @@ mrt_peer_table_dump(struct mrt_table_dump_state *s)
     if ((P->proto == &proto_bgp) && (P->proto_state != PS_DOWN))
     {
       struct bgp_proto *p = (void *) P;
-      mrt_peer_table_entry(s, p->remote_id, p->remote_as, p->cf->remote_ip);
+      mrt_peer_table_entry(s, p->remote_id, p->remote_as, p->remote_ip);
     }
 #endif
 
@@ -429,7 +429,7 @@ mrt_rib_table_entry(struct mrt_table_dump_state *s, rte *r)
   {
     struct bgp_proto *p = (void *) r->attrs->src->proto;
     struct mrt_peer_entry *n =
-      HASH_FIND(s->peer_hash, PEER, p->remote_id, p->remote_as, p->cf->remote_ip);
+      HASH_FIND(s->peer_hash, PEER, p->remote_id, p->remote_as, p->remote_ip);
 
     peer = n ? n->index : 0;
   }
diff --git a/proto/ospf/config.Y b/proto/ospf/config.Y
index 2e9ed0ac..2ec8c0b6 100644
--- a/proto/ospf/config.Y
+++ b/proto/ospf/config.Y
@@ -200,6 +200,7 @@ CF_KEYWORDS(RX, BUFFER, LARGE, NORMAL, STUBNET, HIDDEN, SUMMARY, TAG, EXTERNAL)
 CF_KEYWORDS(WAIT, DELAY, LSADB, ECMP, LIMIT, WEIGHT, NSSA, TRANSLATOR, STABILITY)
 CF_KEYWORDS(GLOBAL, LSID, ROUTER, SELF, INSTANCE, REAL, NETMASK, TX, PRIORITY, LENGTH)
 CF_KEYWORDS(MERGE, LSA, SUPPRESSION, MULTICAST, RFC5838, VPN, PE)
+CF_KEYWORDS(GRACEFUL, RESTART, AWARE, TIME)
 
 %type <ld> lsadb_args
 %type <i> ospf_variant ospf_af_mc nbma_eligible
@@ -226,6 +227,8 @@ ospf_proto_start: proto_start ospf_variant
   OSPF_CFG->tick = OSPF_DEFAULT_TICK;
   OSPF_CFG->ospf2 = $2;
   OSPF_CFG->af_ext = !$2;
+  OSPF_CFG->gr_mode = OSPF_GR_AWARE;
+  OSPF_CFG->gr_time = OSPF_DEFAULT_GR_TIME;
 };
 
 ospf_proto:
@@ -258,6 +261,9 @@ ospf_proto_item:
  | RFC5838 bool { OSPF_CFG->af_ext = $2; if (!ospf_cfg_is_v3()) cf_error("RFC5838 option requires OSPFv3"); }
  | VPN PE bool { OSPF_CFG->vpn_pe = $3; }
  | STUB ROUTER bool { OSPF_CFG->stub_router = $3; }
+ | GRACEFUL RESTART bool { OSPF_CFG->gr_mode = $3; }
+ | GRACEFUL RESTART AWARE { OSPF_CFG->gr_mode = OSPF_GR_AWARE; }
+ | GRACEFUL RESTART TIME expr { OSPF_CFG->gr_time = $4; if (($4 < 1) || ($4 > 1800)) cf_error("Graceful restart time must be in range 1-1800"); }
  | ECMP bool { OSPF_CFG->ecmp = $2 ? OSPF_DEFAULT_ECMP_LIMIT : 0; }
  | ECMP bool LIMIT expr { OSPF_CFG->ecmp = $2 ? $4 : 0; }
  | MERGE EXTERNAL bool { OSPF_CFG->merge_external = $3; }
diff --git a/proto/ospf/dbdes.c b/proto/ospf/dbdes.c
index a1559782..b39595d9 100644
--- a/proto/ospf/dbdes.c
+++ b/proto/ospf/dbdes.c
@@ -215,7 +215,7 @@ ospf_send_dbdes(struct ospf_proto *p, struct ospf_neighbor *n)
 
   ASSERT((n->state == NEIGHBOR_EXSTART) || (n->state == NEIGHBOR_EXCHANGE));
 
-  if (n->ifa->oa->rt == NULL)
+  if (!n->ifa->oa->rt && !p->gr_recovery)
     return;
 
   ospf_prepare_dbdes(p, n);
@@ -279,6 +279,10 @@ ospf_process_dbdes(struct ospf_proto *p, struct ospf_packet *pkt, struct ospf_ne
     if (LSA_SCOPE(lsa_type) == LSA_SCOPE_RES)
       DROP1("LSA with invalid scope");
 
+    /* RFC 3623 2.2 (2) special case - check for my router-LSA (GR recovery) */
+    if ((lsa_type == LSA_T_RT) && (lsa.rt == p->router_id))
+      n->got_my_rt_lsa = 1;
+
     en = ospf_hash_find(p->gr, lsa_domain, lsa.id, lsa.rt, lsa_type);
     if (!en || (lsa_comp(&lsa, &(en->lsa)) == CMP_NEWER))
     {
diff --git a/proto/ospf/iface.c b/proto/ospf/iface.c
index 388c91c8..f5c69199 100644
--- a/proto/ospf/iface.c
+++ b/proto/ospf/iface.c
@@ -772,6 +772,14 @@ ospf_iface_reconfigure(struct ospf_iface *ifa, struct ospf_iface_patt *new)
   ifa->cf = new;
   ifa->marked = 0;
 
+  /* Cancel GR peers if GR is disabled */
+  if (!p->gr_mode && p->gr_count)
+  {
+    struct ospf_neighbor *n, *nx;
+    WALK_LIST_DELSAFE(n, nx, ifa->neigh_list)
+      if (n->gr_active)
+	ospf_neigh_cancel_graceful_restart(n);
+  }
 
   /* HELLO TIMER */
   if (ifa->helloint != new->helloint)
diff --git a/proto/ospf/lsalib.c b/proto/ospf/lsalib.c
index 7ddf64e3..7767700f 100644
--- a/proto/ospf/lsalib.c
+++ b/proto/ospf/lsalib.c
@@ -12,6 +12,9 @@
 
 #include "lib/fletcher16.h"
 
+#define HDRLEN sizeof(struct ospf_lsa_header)
+
+
 #ifndef CPU_BIG_ENDIAN
 void
 lsa_hton_hdr(struct ospf_lsa_header *h, struct ospf_lsa_header *n)
@@ -61,7 +64,6 @@ lsa_ntoh_body(void *n, void *h, u16 len)
 #endif /* little endian */
 
 
-
 int
 lsa_flooding_allowed(u32 type, u32 domain, struct ospf_iface *ifa)
 {
@@ -147,11 +149,13 @@ static const u16 lsa_v2_types[] = {
 
 /* Maps OSPFv2 opaque types to OSPFv3 function codes */
 static const u16 opaque_lsa_types[] = {
+  [LSA_OT_GR] = LSA_T_GR,
   [LSA_OT_RI] = LSA_T_RI_,
 };
 
 /* Maps (subset of) OSPFv3 function codes to OSPFv2 opaque types */
 static const u8 opaque_lsa_types_inv[] = {
+  [LSA_T_GR] = LSA_OT_GR,
   [LSA_T_RI_] = LSA_OT_RI,
 };
 
@@ -168,7 +172,13 @@ lsa_get_type_domain_(u32 type, u32 id, struct ospf_iface *ifa, u32 *otype, u32 *
     uint code;
     if (LSA_FUNCTION(type) == LSA_T_OPAQUE_)
       if (code = LOOKUP(opaque_lsa_types, id >> 24))
+      {
 	type = code | LSA_UBIT | LSA_SCOPE(type);
+
+	/* Hack for Grace-LSA: It does not use U-bit for link-scoped LSAs */
+	if (type == (LSA_T_GR | LSA_UBIT))
+	  type = LSA_T_GR;
+      }
   }
   else
   {
@@ -196,6 +206,13 @@ lsa_get_type_domain_(u32 type, u32 id, struct ospf_iface *ifa, u32 *otype, u32 *
   }
 }
 
+int
+lsa_is_opaque(u32 type)
+{
+  u32 fn = LSA_FUNCTION(type);
+  return LOOKUP(opaque_lsa_types_inv, fn) || (fn == LSA_T_OPAQUE_);
+}
+
 u32
 lsa_get_opaque_type(u32 type)
 {
@@ -267,6 +284,51 @@ lsa_comp(struct ospf_lsa_header *l1, struct ospf_lsa_header *l2)
 }
 
 
+#define LSA_TLV_LENGTH(tlv) \
+  (sizeof(struct ospf_tlv) + BIRD_ALIGN((tlv)->length, 4))
+
+#define LSA_NEXT_TLV(tlv) \
+  ((struct ospf_tlv *) ((byte *) (tlv) + LSA_TLV_LENGTH(tlv)))
+
+#define LSA_WALK_TLVS(tlv,buf,len)					\
+  for(struct ospf_tlv *tlv = (void *) (buf);				\
+      (byte *) tlv < (byte *) (buf) + (len);				\
+      tlv = LSA_NEXT_TLV(tlv))
+
+struct ospf_tlv *
+lsa_get_tlv(struct top_hash_entry *en, uint type)
+{
+  LSA_WALK_TLVS(tlv, en->lsa_body, en->lsa.length - HDRLEN)
+    if (tlv->type == type)
+      return tlv;
+
+  return NULL;
+}
+
+int
+lsa_validate_tlvs(byte *buf, uint len)
+{
+  byte *pos = buf;
+  byte *end = buf + len;
+
+  while (pos < end)
+  {
+    if ((pos + sizeof(struct ospf_tlv)) > end)
+      return 0;
+
+    struct ospf_tlv *tlv = (void *) pos;
+    uint len = LSA_TLV_LENGTH(tlv);
+
+    if ((pos + len) > end)
+      return 0;
+
+    pos += len;
+  }
+
+  return 1;
+}
+
+
 static inline int
 lsa_walk_rt2(struct ospf_lsa_rt_walk *rt)
 {
@@ -408,7 +470,6 @@ lsa_parse_ext(struct top_hash_entry *en, int ospf2, int af, struct ospf_lsa_ext_
   }
 }
 
-#define HDRLEN sizeof(struct ospf_lsa_header)
 
 static int
 lsa_validate_rt2(struct ospf_lsa_header *lsa, struct ospf_lsa_rt *body)
@@ -604,6 +665,12 @@ lsa_validate_prefix(struct ospf_lsa_header *lsa, struct ospf_lsa_prefix *body)
 }
 
 static int
+lsa_validate_gr(struct ospf_lsa_header *lsa, void *body)
+{
+  return lsa_validate_tlvs(body, lsa->length - HDRLEN);
+}
+
+static int
 lsa_validate_ri(struct ospf_lsa_header *lsa UNUSED, struct ospf_lsa_net *body UNUSED)
 {
   /*
@@ -643,6 +710,8 @@ lsa_validate(struct ospf_lsa_header *lsa, u32 lsa_type, int ospf2, void *body)
     case LSA_T_EXT:
     case LSA_T_NSSA:
       return lsa_validate_ext2(lsa, body);
+    case LSA_T_GR:
+      return lsa_validate_gr(lsa, body);
     case LSA_T_RI_LINK:
     case LSA_T_RI_AREA:
     case LSA_T_RI_AS:
@@ -674,6 +743,8 @@ lsa_validate(struct ospf_lsa_header *lsa, u32 lsa_type, int ospf2, void *body)
       return lsa_validate_link(lsa, body);
     case LSA_T_PREFIX:
       return lsa_validate_prefix(lsa, body);
+    case LSA_T_GR:
+      return lsa_validate_gr(lsa, body);
     case LSA_T_RI_LINK:
     case LSA_T_RI_AREA:
     case LSA_T_RI_AS:
diff --git a/proto/ospf/lsalib.h b/proto/ospf/lsalib.h
index af8901ce..eca138d7 100644
--- a/proto/ospf/lsalib.h
+++ b/proto/ospf/lsalib.h
@@ -44,10 +44,7 @@ static inline void lsa_get_type_domain(struct ospf_lsa_header *lsa, struct ospf_
 static inline u32 lsa_get_etype(struct ospf_lsa_header *h, struct ospf_proto *p)
 { return ospf_is_v2(p) ? (h->type_raw & LSA_T_V2_MASK) : h->type_raw; }
 
-/* Assuming OSPFv2 - All U-bit LSAs are mapped to Opaque LSAs */
-static inline int lsa_is_opaque(u32 type)
-{ return !!(type & LSA_UBIT); }
-
+int lsa_is_opaque(u32 type);
 u32 lsa_get_opaque_type(u32 type);
 int lsa_flooding_allowed(u32 type, u32 domain, struct ospf_iface *ifa);
 int lsa_is_acceptable(u32 type, struct ospf_neighbor *n, struct ospf_proto *p);
@@ -58,6 +55,16 @@ u16 lsa_verify_checksum(const void *lsa_n, int lsa_len);
 #define CMP_SAME 0
 #define CMP_OLDER -1
 int lsa_comp(struct ospf_lsa_header *l1, struct ospf_lsa_header *l2);
+
+struct ospf_tlv * lsa_get_tlv(struct top_hash_entry *en, uint type);
+
+static inline u32
+lsa_get_tlv_u32(struct top_hash_entry *en, uint type)
+{
+  struct ospf_tlv *tlv = lsa_get_tlv(en, type);
+  return (tlv && (tlv->length == 4)) ? tlv->data[0] : 0;
+}
+
 void lsa_walk_rt_init(struct ospf_proto *po, struct top_hash_entry *act, struct ospf_lsa_rt_walk *rt);
 int lsa_walk_rt(struct ospf_lsa_rt_walk *rt);
 void lsa_parse_sum_net(struct top_hash_entry *en, int ospf2, int af, net_addr *net, u8 *pxopts, u32 *metric);
diff --git a/proto/ospf/lsupd.c b/proto/ospf/lsupd.c
index 7318b751..fafe4872 100644
--- a/proto/ospf/lsupd.c
+++ b/proto/ospf/lsupd.c
@@ -185,6 +185,13 @@ static int ospf_flood_lsupd(struct ospf_proto *p, struct top_hash_entry **lsa_li
 static void
 ospf_enqueue_lsa(struct ospf_proto *p, struct top_hash_entry *en, struct ospf_iface *ifa)
 {
+  /* Exception for local Grace-LSA, they are flooded synchronously */
+  if ((en->lsa_type == LSA_T_GR) && (en->lsa.rt == p->router_id))
+  {
+    ospf_flood_lsupd(p, &en, 1, 1, ifa);
+    return;
+  }
+
   if (ifa->flood_queue_used == ifa->flood_queue_size)
   {
     /* If we already have full queue, we send some packets */
@@ -591,8 +598,9 @@ ospf_receive_lsupd(struct ospf_packet *pkt, struct ospf_iface *ifa,
       }
 
       /* 13. (5f) - handle self-originated LSAs, see also 13.4. */
-      if ((lsa.rt == p->router_id) ||
-	  (ospf_is_v2(p) && (lsa_type == LSA_T_NET) && ospf_addr_is_local(p, ifa->oa, ipa_from_u32(lsa.id))))
+      if (!p->gr_recovery &&
+	  ((lsa.rt == p->router_id) ||
+	   (ospf_is_v2(p) && (lsa_type == LSA_T_NET) && ospf_addr_is_local(p, ifa->oa, ipa_from_u32(lsa.id)))))
       {
 	OSPF_TRACE(D_EVENTS, "Received unexpected self-originated LSA");
 	ospf_advance_lsa(p, en, &lsa, lsa_type, lsa_domain, body);
@@ -629,6 +637,14 @@ ospf_receive_lsupd(struct ospf_packet *pkt, struct ospf_iface *ifa,
       if (lsa_type == LSA_T_LINK)
 	ospf_notify_net_lsa(ifa);
 
+      /* RFC 3623 3.1 - entering graceful restart helper mode */
+      if (lsa_type == LSA_T_GR)
+	ospf_neigh_notify_grace_lsa(n, en);
+
+      /* Link received pre-restart router LSA */
+      if (p->gr_recovery && (lsa_type == LSA_T_RT) && (lsa.rt == p->router_id))
+	ifa->oa->rt = en;
+
       /* 13. (5b) - flood new LSA */
       int flood_back = ospf_flood_lsa(p, en, n);
 
diff --git a/proto/ospf/neighbor.c b/proto/ospf/neighbor.c
index c143b130..50ef6a49 100644
--- a/proto/ospf/neighbor.c
+++ b/proto/ospf/neighbor.c
@@ -28,6 +28,8 @@ static void dbdes_timer_hook(timer *t);
 static void lsrq_timer_hook(timer *t);
 static void lsrt_timer_hook(timer *t);
 static void ackd_timer_hook(timer *t);
+static void ospf_neigh_stop_graceful_restart_(struct ospf_neighbor *n);
+static void graceful_restart_timeout(timer *t);
 
 
 static void
@@ -163,7 +165,7 @@ ospf_neigh_chstate(struct ospf_neighbor *n, u8 state)
   if (old_state == NEIGHBOR_FULL)
     ifa->fadj--;
 
-  if (ifa->fadj != old_fadj)
+  if ((ifa->fadj != old_fadj) && !n->gr_active)
   {
     /* RFC 2328 12.4 Event 4 - neighbor enters/leaves Full state */
     ospf_notify_rt_lsa(ifa->oa);
@@ -182,6 +184,7 @@ ospf_neigh_chstate(struct ospf_neighbor *n, u8 state)
 
     n->dds++;
     n->myimms = DBDES_IMMS;
+    n->got_my_rt_lsa = 0;
 
     tm_start(n->dbdes_timer, 0);
     tm_start(n->ackd_timer, ifa->rxmtint S / 2);
@@ -191,9 +194,9 @@ ospf_neigh_chstate(struct ospf_neighbor *n, u8 state)
     n->myimms &= ~DBDES_I;
 
   /* Generate NeighborChange event if needed, see RFC 2328 9.2 */
-  if ((state == NEIGHBOR_2WAY) && (old_state < NEIGHBOR_2WAY))
+  if ((state == NEIGHBOR_2WAY) && (old_state < NEIGHBOR_2WAY) && !n->gr_active)
     ospf_iface_sm(ifa, ISM_NEICH);
-  if ((state < NEIGHBOR_2WAY) && (old_state >= NEIGHBOR_2WAY))
+  if ((state < NEIGHBOR_2WAY) && (old_state >= NEIGHBOR_2WAY) && !n->gr_active)
     ospf_iface_sm(ifa, ISM_NEICH);
 }
 
@@ -291,6 +294,17 @@ ospf_neigh_sm(struct ospf_neighbor *n, int event)
   case INM_KILLNBR:
   case INM_LLDOWN:
   case INM_INACTTIM:
+    if (n->gr_active && (event == INM_INACTTIM))
+    {
+      /* Just down the neighbor, but do not remove it */
+      reset_lists(p, n);
+      ospf_neigh_chstate(n, NEIGHBOR_DOWN);
+      break;
+    }
+
+    if (n->gr_active)
+      ospf_neigh_stop_graceful_restart_(n);
+
     /* No need for reset_lists() */
     ospf_neigh_chstate(n, NEIGHBOR_DOWN);
     ospf_neigh_down(n);
@@ -356,6 +370,180 @@ can_do_adj(struct ospf_neighbor *n)
   return i;
 }
 
+static void
+ospf_neigh_start_graceful_restart(struct ospf_neighbor *n, uint gr_time)
+{
+  struct ospf_proto *p = n->ifa->oa->po;
+
+  OSPF_TRACE(D_EVENTS, "Neighbor %R on %s started graceful restart",
+	     n->rid, n->ifa->ifname);
+
+  n->gr_active = 1;
+  p->gr_count++;
+
+  n->gr_timer = tm_new_init(n->pool, graceful_restart_timeout, n, 0, 0);
+  tm_start(n->gr_timer, gr_time S);
+}
+
+static void
+ospf_neigh_stop_graceful_restart_(struct ospf_neighbor *n)
+{
+  struct ospf_proto *p = n->ifa->oa->po;
+  struct ospf_iface *ifa = n->ifa;
+
+  n->gr_active = 0;
+  p->gr_count--;
+
+  rfree(n->gr_timer);
+  n->gr_timer = NULL;
+
+  ospf_notify_rt_lsa(ifa->oa);
+  ospf_notify_net_lsa(ifa);
+
+  if (ifa->type == OSPF_IT_VLINK)
+    ospf_notify_rt_lsa(ifa->voa);
+
+  ospf_iface_sm(ifa, ISM_NEICH);
+}
+
+static void
+ospf_neigh_stop_graceful_restart(struct ospf_neighbor *n)
+{
+  struct ospf_proto *p = n->ifa->oa->po;
+
+  OSPF_TRACE(D_EVENTS, "Neighbor %R on %s finished graceful restart",
+	     n->rid, n->ifa->ifname);
+
+  ospf_neigh_stop_graceful_restart_(n);
+}
+
+void
+ospf_neigh_cancel_graceful_restart(struct ospf_neighbor *n)
+{
+  struct ospf_proto *p = n->ifa->oa->po;
+
+  OSPF_TRACE(D_EVENTS, "Graceful restart canceled for nbr %R on %s",
+	     n->rid, n->ifa->ifname);
+
+  ospf_neigh_stop_graceful_restart_(n);
+
+  if (n->state == NEIGHBOR_DOWN)
+    ospf_neigh_down(n);
+}
+
+static void
+graceful_restart_timeout(timer *t)
+{
+  struct ospf_neighbor *n = t->data;
+  struct ospf_proto *p = n->ifa->oa->po;
+
+  OSPF_TRACE(D_EVENTS, "Graceful restart timer expired for nbr %R on %s",
+	     n->rid, n->ifa->ifname);
+
+  ospf_neigh_stop_graceful_restart_(n);
+
+  if (n->state == NEIGHBOR_DOWN)
+    ospf_neigh_down(n);
+}
+
+static inline int
+changes_in_lsrtl(struct ospf_neighbor *n)
+{
+  /* This could be improved, see RFC 3623 3.1 (2) */
+
+  struct top_hash_entry *en;
+  WALK_SLIST(en, n->lsrtl)
+    if (LSA_FUNCTION(en->lsa_type) <= LSA_FUNCTION(LSA_T_NSSA))
+      return 1;
+
+  return 0;
+}
+
+void
+ospf_neigh_notify_grace_lsa(struct ospf_neighbor *n, struct top_hash_entry *en)
+{
+  struct ospf_iface *ifa = n->ifa;
+  struct ospf_proto *p = ifa->oa->po;
+
+  /* In OSPFv2, neighbors are identified by either IP or Router ID, based on network type */
+  uint t = ifa->type;
+  if (ospf_is_v2(p) && ((t == OSPF_IT_BCAST) || (t == OSPF_IT_NBMA) || (t == OSPF_IT_PTMP)))
+  {
+    struct ospf_tlv *tlv = lsa_get_tlv(en, LSA_GR_ADDRESS);
+    if (!tlv || tlv->length != 4)
+      return;
+
+    ip_addr addr = ipa_from_u32(tlv->data[0]);
+    if (!ipa_equal(n->ip, addr))
+      n = find_neigh_by_ip(ifa, addr);
+  }
+  else
+  {
+    if (n->rid != en->lsa.rt)
+      n = find_neigh(ifa, en->lsa.rt);
+  }
+
+  if (!n)
+    return;
+
+  if (en->lsa.age < LSA_MAXAGE)
+  {
+    u32 period = lsa_get_tlv_u32(en, LSA_GR_PERIOD);
+
+    /* Exception for updating grace period */
+    if (n->gr_active)
+    {
+      tm_start(n->gr_timer, (period S) - (en->lsa.age S));
+      return;
+    }
+
+    /* RFC 3623 3.1 (1) - full adjacency */
+    if (n->state != NEIGHBOR_FULL)
+      return;
+
+    /* RFC 3623 3.1 (2) - no changes in LSADB */
+    if (changes_in_lsrtl(n))
+      return;
+
+    /* RFC 3623 3.1 (3) - grace period not expired */
+    if (en->lsa.age >= period)
+      return;
+
+    /* RFC 3623 3.1 (4) - helper mode allowed */
+    if (!p->gr_mode)
+      return;
+
+    /* RFC 3623 3.1 (5) - no local graceful restart */
+    if (p->p.gr_recovery)
+      return;
+
+    ospf_neigh_start_graceful_restart(n, period - en->lsa.age);
+  }
+  else /* Grace-LSA is flushed */
+  {
+    if (n->gr_active)
+      ospf_neigh_stop_graceful_restart(n);
+  }
+}
+
+void
+ospf_neigh_lsadb_changed_(struct ospf_proto *p, struct top_hash_entry *en)
+{
+  struct ospf_iface *ifa;
+  struct ospf_neighbor *n, *nx;
+
+  if (LSA_FUNCTION(en->lsa_type) > LSA_FUNCTION(LSA_T_NSSA))
+    return;
+
+  /* RFC 3623 3.2 (3) - cancel graceful restart when LSdb changed */
+  WALK_LIST(ifa, p->iface_list)
+    if (lsa_flooding_allowed(en->lsa_type, en->domain, ifa))
+      WALK_LIST_DELSAFE(n, nx, ifa->neigh_list)
+	if (n->gr_active)
+	  ospf_neigh_cancel_graceful_restart(n);
+}
+
+
 
 static inline u32 neigh_get_id(struct ospf_proto *p, struct ospf_neighbor *n)
 { return ospf_is_v2(p) ? ipa_to_u32(n->ip) : n->rid; }
diff --git a/proto/ospf/ospf.c b/proto/ospf/ospf.c
index f26f0160..b6d5570c 100644
--- a/proto/ospf/ospf.c
+++ b/proto/ospf/ospf.c
@@ -92,7 +92,9 @@
  * - RFC 2328 - main OSPFv2 standard
  * - RFC 5340 - main OSPFv3 standard
  * - RFC 3101 - OSPFv2 NSSA areas
+ * - RFC 3623 - OSPFv2 Graceful Restart
  * - RFC 4576 - OSPFv2 VPN loop prevention
+ * - RFC 5187 - OSPFv3 Graceful Restart
  * - RFC 5250 - OSPFv2 Opaque LSAs
  * - RFC 5709 - OSPFv2 HMAC-SHA Cryptographic Authentication
  * - RFC 5838 - OSPFv3 Support of Address Families
@@ -207,7 +209,6 @@ ospf_area_remove(struct ospf_area *oa)
   mb_free(oa);
 }
 
-
 struct ospf_area *
 ospf_find_area(struct ospf_proto *p, u32 aid)
 {
@@ -228,6 +229,37 @@ ospf_find_vlink(struct ospf_proto *p, u32 voa, u32 vid)
   return NULL;
 }
 
+static void
+ospf_start_gr_recovery(struct ospf_proto *p)
+{
+  OSPF_TRACE(D_EVENTS, "Graceful restart started");
+
+  p->gr_recovery = 1;
+  p->gr_timeout = current_time() + (p->gr_time S);
+  channel_graceful_restart_lock(p->p.main_channel);
+  p->p.main_channel->gr_wait = 1;
+
+  /* NOTE: We should get end of grace period from non-volatile storage */
+}
+
+void
+ospf_stop_gr_recovery(struct ospf_proto *p)
+{
+  p->gr_recovery = 0;
+  p->gr_timeout = 0;
+  channel_graceful_restart_unlock(p->p.main_channel);
+
+  /* Reorigination of router/network LSAs is already scheduled */
+  ospf_mark_lsadb(p);
+
+  /*
+   * NOTE: We should move channel_graceful_restart_unlock() to the end of
+   * ospf_disp() in order to have local LSA reorigination / LSAdb cleanup /
+   * routing table recomputation before official end of GR. It does not matter
+   * when we are single-threaded.
+   */
+}
+
 static int
 ospf_start(struct proto *P)
 {
@@ -246,6 +278,8 @@ ospf_start(struct proto *P)
   p->asbr = c->asbr;
   p->vpn_pe = c->vpn_pe;
   p->ecmp = c->ecmp;
+  p->gr_mode = c->gr_mode;
+  p->gr_time = c->gr_time;
   p->tick = c->tick;
   p->disp_timer = tm_new_init(P->pool, ospf_disp, p, p->tick S, 0);
   tm_start(p->disp_timer, 100 MS);
@@ -267,6 +301,10 @@ ospf_start(struct proto *P)
   p->log_pkt_tbf = (struct tbf){ .rate = 1, .burst = 5 };
   p->log_lsa_tbf = (struct tbf){ .rate = 4, .burst = 20 };
 
+  /* Lock the channel when in GR recovery mode */
+  if (p->p.gr_recovery && (p->gr_mode == OSPF_GR_ABLE))
+    ospf_start_gr_recovery(p);
+
   WALK_LIST(ac, c->area_list)
     ospf_area_add(p, ac);
 
@@ -398,6 +436,9 @@ ospf_disp(timer * timer)
 {
   struct ospf_proto *p = timer->data;
 
+  if (p->gr_recovery)
+    ospf_update_gr_recovery(p);
+
   /* Originate or flush local topology LSAs */
   ospf_update_topology(p);
 
@@ -475,9 +516,18 @@ ospf_shutdown(struct proto *P)
 
   OSPF_TRACE(D_EVENTS, "Shutdown requested");
 
-  /* And send to all my neighbors 1WAY */
-  WALK_LIST(ifa, p->iface_list)
-    ospf_iface_shutdown(ifa);
+  if ((P->down_code == PDC_CMD_GR_DOWN) && (p->gr_mode == OSPF_GR_ABLE))
+  {
+    /* Originate Grace LSAs */
+    WALK_LIST(ifa, p->iface_list)
+      ospf_originate_gr_lsa(p, ifa);
+  }
+  else
+  {
+    /* Send to all my neighbors 1WAY */
+    WALK_LIST(ifa, p->iface_list)
+      ospf_iface_shutdown(ifa);
+  }
 
   /* Cleanup locked rta entries */
   FIB_WALK(&p->rtf, ort, nf)
@@ -664,6 +714,8 @@ ospf_reconfigure(struct proto *P, struct proto_config *CF)
   p->merge_external = new->merge_external;
   p->asbr = new->asbr;
   p->ecmp = new->ecmp;
+  p->gr_mode = new->gr_mode;
+  p->gr_time = new->gr_time;
   p->tick = new->tick;
   p->disp_timer->recurrent = p->tick S;
   tm_start(p->disp_timer, 10 MS);
diff --git a/proto/ospf/ospf.h b/proto/ospf/ospf.h
index 82ae4df4..beecd2b6 100644
--- a/proto/ospf/ospf.h
+++ b/proto/ospf/ospf.h
@@ -75,6 +75,7 @@
 #define OSPF_DEFAULT_TICK 1
 #define OSPF_DEFAULT_STUB_COST 1000
 #define OSPF_DEFAULT_ECMP_LIMIT 16
+#define OSPF_DEFAULT_GR_TIME 120
 #define OSPF_DEFAULT_TRANSINT 40
 
 #define OSPF_MIN_PKT_SIZE 256
@@ -82,6 +83,9 @@
 
 #define OSPF_VLINK_ID_OFFSET 0x80000000
 
+#define OSPF_GR_ABLE		1
+#define OSPF_GR_AWARE		2
+
 struct ospf_config
 {
   struct proto_config c;
@@ -97,7 +101,9 @@ struct ospf_config
   u8 abr;
   u8 asbr;
   u8 vpn_pe;
-  int ecmp;
+  u8 gr_mode;			/* Graceful restart mode (OSPF_GR_*) */
+  uint gr_time;			/* Graceful restart interval */
+  uint ecmp;
   list area_list;		/* list of area configs (struct ospf_area_config) */
   list vlink_list;		/* list of configured vlinks (struct ospf_iface_patt) */
 };
@@ -216,6 +222,9 @@ struct ospf_proto
   list area_list;		/* List of OSPF areas (struct ospf_area) */
   int areano;			/* Number of area I belong to */
   int padj;			/* Number of neighbors in Exchange or Loading state */
+  int gr_count;			/* Number of neighbors in graceful restart state */
+  int gr_recovery;		/* Graceful restart recovery is active */
+  btime gr_timeout;		/* The end time of grace restart recovery */
   struct fib rtf;		/* Routing table */
   struct idm idm;		/* OSPFv3 LSA ID map */
   u8 ospf2;			/* OSPF v2 or v3 */
@@ -228,6 +237,8 @@ struct ospf_proto
   u8 asbr;			/* May i originate any ext/NSSA lsa? */
   u8 vpn_pe;			/* Should we do VPN PE specific behavior (RFC 4577)? */
   u8 ecmp;			/* Maximal number of nexthops in ECMP route, or 0 */
+  u8 gr_mode;			/* Graceful restart mode (OSPF_GR_*) */
+  uint gr_time;			/* Graceful restart interval */
   u64 csn64;			/* Last used cryptographic sequence number */
   struct ospf_area *backbone;	/* If exists */
   event *flood_event;		/* Event for flooding LS updates */
@@ -346,6 +357,8 @@ struct ospf_neighbor
   pool *pool;
   struct ospf_iface *ifa;
   u8 state;
+  u8 gr_active;			/* We act as GR helper for the neighbor */
+  u8 got_my_rt_lsa;		/* Received my Rt-LSA in DBDES exchanged */
   timer *inactim;		/* Inactivity timer */
   u8 imms;			/* I, M, Master/slave received */
   u8 myimms;			/* I, M Master/slave */
@@ -388,6 +401,7 @@ struct ospf_neighbor
 #define ACKL_DIRECT 0
 #define ACKL_DELAY 1
   timer *ackd_timer;		/* Delayed ack timer */
+  timer *gr_timer;		/* Graceful restart timer, non-NULL only if gr_active */
   struct bfd_request *bfd_req;	/* BFD request, if BFD is used */
   void *ldd_buffer;		/* Last database description packet */
   u32 ldd_bsize;		/* Buffer size for ldd_buffer */
@@ -555,6 +569,7 @@ struct ospf_auth3
 #define LSA_T_NSSA		0x2007
 #define LSA_T_LINK		0x0008
 #define LSA_T_PREFIX		0x2009
+#define LSA_T_GR		0x000B
 #define LSA_T_RI_		0x000C
 #define LSA_T_RI_LINK		0x800C
 #define LSA_T_RI_AREA		0xA00C
@@ -569,6 +584,7 @@ struct ospf_auth3
 
 /* OSPFv2 Opaque LSA Types */
 /* https://www.iana.org/assignments/ospf-opaque-types/ospf-opaque-types.xhtml#ospf-opaque-types-2 */
+#define LSA_OT_GR		0x03
 #define LSA_OT_RI		0x04
 
 #define LSA_FUNCTION_MASK	0x1FFF
@@ -613,6 +629,12 @@ struct ospf_auth3
 #define LSA_EXT3_FBIT	0x02000000
 #define LSA_EXT3_TBIT	0x01000000
 
+/* OSPF Grace LSA (GR) TLVs */
+/* https://www.iana.org/assignments/ospfv2-parameters/ospfv2-parameters.xhtml#ospfv2-parameters-13 */
+#define LSA_GR_PERIOD		1
+#define LSA_GR_REASON		2
+#define LSA_GR_ADDRESS		3
+
 /* OSPF Router Information (RI) TLVs */
 /* https://www.iana.org/assignments/ospf-parameters/ospf-parameters.xhtml#ri-tlv */
 #define LSA_RI_RIC		1
@@ -959,6 +981,8 @@ static inline int oa_is_ext(struct ospf_area *oa)
 static inline int oa_is_nssa(struct ospf_area *oa)
 { return oa->options & OPT_N; }
 
+void ospf_stop_gr_recovery(struct ospf_proto *p);
+
 void ospf_sh_neigh(struct proto *P, char *iff);
 void ospf_sh(struct proto *P);
 void ospf_sh_iface(struct proto *P, char *iff);
@@ -990,12 +1014,18 @@ static inline struct nbma_node * find_nbma_node(struct ospf_iface *ifa, ip_addr
 /* neighbor.c */
 struct ospf_neighbor *ospf_neighbor_new(struct ospf_iface *ifa);
 void ospf_neigh_sm(struct ospf_neighbor *n, int event);
+void ospf_neigh_cancel_graceful_restart(struct ospf_neighbor *n);
+void ospf_neigh_notify_grace_lsa(struct ospf_neighbor *n, struct top_hash_entry *en);
+void ospf_neigh_lsadb_changed_(struct ospf_proto *p, struct top_hash_entry *en);
 void ospf_dr_election(struct ospf_iface *ifa);
 struct ospf_neighbor *find_neigh(struct ospf_iface *ifa, u32 rid);
 struct ospf_neighbor *find_neigh_by_ip(struct ospf_iface *ifa, ip_addr ip);
 void ospf_neigh_update_bfd(struct ospf_neighbor *n, int use_bfd);
 void ospf_sh_neigh_info(struct ospf_neighbor *n);
 
+static inline void ospf_neigh_lsadb_changed(struct ospf_proto *p, struct top_hash_entry *en)
+{ if (p->gr_count) ospf_neigh_lsadb_changed_(p, en); }
+
 /* packet.c */
 void ospf_pkt_fill_hdr(struct ospf_iface *ifa, void *buf, u8 h_type);
 int ospf_rx_hook(sock * sk, uint size);
diff --git a/proto/ospf/rt.c b/proto/ospf/rt.c
index 6ddd6c9f..126ef201 100644
--- a/proto/ospf/rt.c
+++ b/proto/ospf/rt.c
@@ -10,7 +10,7 @@
 
 #include "ospf.h"
 
-static void add_cand(struct ospf_area *oa, struct top_hash_entry *en, struct top_hash_entry *par, u32 dist, int i, uint lif, uint nif);
+static void add_cand(struct ospf_area *oa, struct top_hash_entry *en, struct top_hash_entry *par, u32 dist, int i, uint data, uint lif, uint nif);
 static void rt_sync(struct ospf_proto *p);
 
 
@@ -392,6 +392,40 @@ px_pos_to_ifa(struct ospf_area *oa, int pos)
   return NULL;
 }
 
+static inline struct ospf_iface *
+rt_find_iface2(struct ospf_area *oa, uint data)
+{
+  ip_addr addr = ipa_from_u32(data);
+
+  /* We should handle it differently for unnumbered PTP links */
+  struct ospf_iface *ifa;
+  WALK_LIST(ifa, oa->po->iface_list)
+    if ((ifa->oa == oa) && ifa->addr && (ipa_equal(ifa->addr->ip, addr)))
+      return ifa;
+
+  return NULL;
+}
+
+static inline struct ospf_iface *
+rt_find_iface3(struct ospf_area *oa, uint lif)
+{
+  struct ospf_iface *ifa;
+  WALK_LIST(ifa, oa->po->iface_list)
+    if ((ifa->oa == oa) && (ifa->iface_id == lif))
+      return ifa;
+
+  return NULL;
+}
+
+static struct ospf_iface *
+rt_find_iface(struct ospf_area *oa, int pos, uint data, uint lif)
+{
+  if (0)
+    return rt_pos_to_ifa(oa, pos);
+  else
+    return ospf_is_v2(oa->po) ? rt_find_iface2(oa, data) : rt_find_iface3(oa, lif);
+}
+
 
 static void
 add_network(struct ospf_area *oa, net_addr *net, int metric, struct top_hash_entry *en, int pos)
@@ -503,7 +537,7 @@ spfa_process_rt(struct ospf_proto *p, struct ospf_area *oa, struct top_hash_entr
       break;
     }
 
-    add_cand(oa, tmp, act, act->dist + rtl.metric, i, rtl.lif, rtl.nif);
+    add_cand(oa, tmp, act, act->dist + rtl.metric, i, rtl.data, rtl.lif, rtl.nif);
   }
 }
 
@@ -526,7 +560,7 @@ spfa_process_net(struct ospf_proto *p, struct ospf_area *oa, struct top_hash_ent
   for (i = 0; i < cnt; i++)
   {
     tmp = ospf_hash_find_rt(p->gr, oa->areaid, ln->routers[i]);
-    add_cand(oa, tmp, act, act->dist, -1, 0, 0);
+    add_cand(oa, tmp, act, act->dist, -1, 0, 0, 0);
   }
 }
 
@@ -1708,7 +1742,7 @@ link_lsa_lladdr(struct ospf_proto *p, struct top_hash_entry *en)
 
 static struct nexthop *
 calc_next_hop(struct ospf_area *oa, struct top_hash_entry *en,
-	      struct top_hash_entry *par, int pos, uint lif, uint nif)
+	      struct top_hash_entry *par, int pos, uint data, uint lif, uint nif)
 {
   struct ospf_proto *p = oa->po;
   struct nexthop *pn = par->nhs;
@@ -1735,7 +1769,7 @@ calc_next_hop(struct ospf_area *oa, struct top_hash_entry *en,
   /* The first case - local network */
   if ((en->lsa_type == LSA_T_NET) && (par == oa->rt))
   {
-    ifa = rt_pos_to_ifa(oa, pos);
+    ifa = rt_find_iface(oa, pos, data, lif);
     if (!ifa)
       return NULL;
 
@@ -1748,7 +1782,7 @@ calc_next_hop(struct ospf_area *oa, struct top_hash_entry *en,
   /* The second case - ptp or ptmp neighbor */
   if ((en->lsa_type == LSA_T_RT) && (par == oa->rt))
   {
-    ifa = rt_pos_to_ifa(oa, pos);
+    ifa = rt_find_iface(oa, pos, data, lif);
     if (!ifa)
       return NULL;
 
@@ -1838,7 +1872,7 @@ calc_next_hop(struct ospf_area *oa, struct top_hash_entry *en,
 /* Add LSA into list of candidates in Dijkstra's algorithm */
 static void
 add_cand(struct ospf_area *oa, struct top_hash_entry *en, struct top_hash_entry *par,
-	 u32 dist, int pos, uint lif, uint nif)
+	 u32 dist, int pos, uint data, uint lif, uint nif)
 {
   struct ospf_proto *p = oa->po;
   node *prev, *n;
@@ -1871,7 +1905,7 @@ add_cand(struct ospf_area *oa, struct top_hash_entry *en, struct top_hash_entry
   if (!link_back(oa, en, par, lif, nif))
     return;
 
-  struct nexthop *nhs = calc_next_hop(oa, en, par, pos, lif, nif);
+  struct nexthop *nhs = calc_next_hop(oa, en, par, pos, data, lif, nif);
   if (!nhs)
   {
     log(L_WARN "%s: Cannot find next hop for LSA (Type: %04x, Id: %R, Rt: %R)",
@@ -2086,3 +2120,133 @@ again2:
     if (en->mode == LSA_M_STALE)
       ospf_flush_lsa(p, en);
 }
+
+
+/* RFC 3623 2.2 - checking for graceful restart termination conditions */
+void
+ospf_update_gr_recovery(struct ospf_proto *p)
+{
+  struct top_hash_entry *rt, *net, *nbr;
+  struct ospf_lsa_rt_walk rtl;
+  struct ospf_neighbor *n;
+  struct ospf_iface *ifa;
+  struct ospf_area *oa;
+  const char *err_dsc = NULL;
+  uint i, j, missing = 0, err_val = 0;
+
+  /*
+   * We check here for three cases:
+   * RFC 3623 2.2 (1) - success when all adjacencies are established
+   * RFC 3623 2.2 (2) - failure when inconsistent LSA was received
+   * RFC 3623 2.2 (3) - grace period timeout
+   *
+   * It is handled by processing pre-restart local router-LSA and adjacent
+   * network-LSAs, checking neighbor association for referenced routers (1)
+   * and checking back links from their router-LSAs (2).
+   *
+   * TODO: Use timer for grace period timeout. We avoided that as function
+   * ospf_stop_gr_recovery() called from ospf_disp() makes ending of graceful
+   * restart uninterrupted by other events.
+   */
+
+  #define CONTINUE { missing++; continue; }
+
+  if (current_time() > p->gr_timeout)
+    goto timeout;
+
+  WALK_LIST(oa, p->area_list)
+  {
+    /* Get the router-LSA */
+    rt = oa->rt;
+    if (!rt || (rt->lsa.age == LSA_MAXAGE))
+      CONTINUE;
+
+    for (lsa_walk_rt_init(p, rt, &rtl), i = 0; lsa_walk_rt(&rtl); i++)
+    {
+      if (rtl.type == LSART_STUB)
+	continue;
+
+      ifa = rt_find_iface(oa, i, rtl.data, rtl.lif);
+      if (!ifa)
+	DROP("inconsistent interface", ospf_is_v2(p) ? rtl.data : rtl.lif);
+
+      switch (rtl.type)
+      {
+      case LSART_NET:
+	/* Find the network-LSA */
+	net = ospf_hash_find_net(p->gr, oa->areaid, rtl.id, rtl.nif);
+	if (!net)
+	  CONTINUE;
+
+	if (!link_back(oa, net, rt, rtl.lif, rtl.nif))
+	  DROP("Inconsistent network-LSA", net->lsa.id);
+
+	if (ifa->state == OSPF_IS_DR)
+	{
+	  /* Find all neighbors from the network-LSA */
+	  struct ospf_lsa_net *net_body = net->lsa_body;
+	  uint cnt = lsa_net_count(&net->lsa);
+	  for (j = 0; j < cnt; i++)
+	  {
+	    n = find_neigh(ifa, net_body->routers[j]);
+	    if (!n || (n->state != NEIGHBOR_FULL))
+	      CONTINUE;
+
+	    if (!n->got_my_rt_lsa)
+	      DROP("not received my router-LSA", n->rid);
+
+	    nbr = ospf_hash_find_rt(p->gr, oa->areaid, n->rid);
+	    if (!link_back(oa, nbr, net, 0, 0))
+	      DROP("inconsistent router-LSA", n->rid);
+	  }
+	}
+	else
+	{
+	  /* Find the DR (by IP for OSPFv2) */
+	  n = ospf_is_v2(p) ?
+	    find_neigh_by_ip(ifa, ipa_from_u32(rtl.id)) :
+	    find_neigh(ifa, rtl.id);
+	  if (!n || (n->state != NEIGHBOR_FULL))
+	    CONTINUE;
+
+	  if (!n->got_my_rt_lsa)
+	    DROP("not received my router-LSA", n->rid);
+	}
+	break;
+
+      case LSART_VLNK:
+      case LSART_PTP:
+	/* Find the PtP peer */
+	n = find_neigh(ifa, rtl.id);
+	if (!n || (n->state != NEIGHBOR_FULL))
+	  CONTINUE;
+
+	if (!n->got_my_rt_lsa)
+	  DROP("not received my router-LSA", n->rid);
+
+	nbr = ospf_hash_find_rt(p->gr, oa->areaid, rtl.id);
+	if (!link_back(oa, nbr, rt, rtl.lif, rtl.nif))
+	  DROP("inconsistent router-LSA", rtl.id);
+      }
+    }
+  }
+
+  #undef CONTINUE
+
+  if (missing)
+    return;
+
+  OSPF_TRACE(D_EVENTS, "Graceful restart finished");
+  ospf_stop_gr_recovery(p);
+  return;
+
+drop:
+  log(L_INFO "%s: Graceful restart ended - %s (%R)", p->p.name, err_dsc, err_val);
+  ospf_stop_gr_recovery(p);
+  return;
+
+timeout:
+  log(L_INFO "%s: Graceful restart ended - grace period expired", p->p.name);
+  ospf_stop_gr_recovery(p);
+  return;
+}
diff --git a/proto/ospf/rt.h b/proto/ospf/rt.h
index 589d2bc5..094e125b 100644
--- a/proto/ospf/rt.h
+++ b/proto/ospf/rt.h
@@ -130,6 +130,7 @@ static inline int rt_is_nssa(ort *nf)
 
 void ospf_rt_spf(struct ospf_proto *p);
 void ospf_rt_initort(struct fib_node *fn);
+void ospf_update_gr_recovery(struct ospf_proto *p);
 
 
 #endif /* _BIRD_OSPF_RT_H_ */
diff --git a/proto/ospf/topology.c b/proto/ospf/topology.c
index 7d5deca0..efd03b54 100644
--- a/proto/ospf/topology.c
+++ b/proto/ospf/topology.c
@@ -83,7 +83,10 @@ ospf_install_lsa(struct ospf_proto *p, struct ospf_lsa_header *lsa, u32 type, u3
 	     en->lsa_type, en->lsa.id, en->lsa.rt, en->lsa.sn, en->lsa.age);
 
   if (change)
+  {
+    ospf_neigh_lsadb_changed(p, en);
     ospf_schedule_rtcalc(p);
+  }
 
   return en;
 }
@@ -243,6 +246,7 @@ ospf_do_originate_lsa(struct ospf_proto *p, struct top_hash_entry *en, void *lsa
   en->lsa.age = 0;
   en->init_age = 0;
   en->inst_time = current_time();
+  en->dirty = 0;
   lsa_generate_checksum(&en->lsa, en->lsa_body);
 
   OSPF_TRACE(D_EVENTS, "Originating LSA: Type: %04x, Id: %R, Rt: %R, Seq: %08x",
@@ -251,7 +255,10 @@ ospf_do_originate_lsa(struct ospf_proto *p, struct top_hash_entry *en, void *lsa
   ospf_flood_lsa(p, en, NULL);
 
   if (en->mode == LSA_M_BASIC)
+  {
+    ospf_neigh_lsadb_changed(p, en);
     ospf_schedule_rtcalc(p);
+  }
 
   return 1;
 }
@@ -321,7 +328,8 @@ ospf_originate_lsa(struct ospf_proto *p, struct ospf_new_lsa *lsa)
   if ((en->lsa.age < LSA_MAXAGE) &&
       (lsa_length == en->lsa.length) &&
       !memcmp(lsa_body, en->lsa_body, lsa_blen) &&
-      (!ospf_is_v2(p) || (lsa->opts == lsa_get_options(&en->lsa))))
+      (!ospf_is_v2(p) || (lsa->opts == lsa_get_options(&en->lsa))) &&
+      !en->dirty)
     goto drop;
 
   lsa_body = lsab_flush(p);
@@ -433,7 +441,10 @@ ospf_flush_lsa(struct ospf_proto *p, struct top_hash_entry *en)
   ospf_flood_lsa(p, en, NULL);
 
   if (en->mode == LSA_M_BASIC)
+  {
+    ospf_neigh_lsadb_changed(p, en);
     ospf_schedule_rtcalc(p);
+  }
 
   en->mode = LSA_M_BASIC;
 }
@@ -509,6 +520,12 @@ ospf_update_lsadb(struct ospf_proto *p)
       continue;
     }
 
+    if (en->dirty)
+    {
+      ospf_flush_lsa(p, en);
+      continue;
+    }
+
     if ((en->lsa.rt == p->router_id) && (real_age >= LSREFRESHTIME))
     {
       ospf_refresh_lsa(p, en);
@@ -525,6 +542,16 @@ ospf_update_lsadb(struct ospf_proto *p)
   }
 }
 
+void
+ospf_mark_lsadb(struct ospf_proto *p)
+{
+  struct top_hash_entry *en;
+
+  /* Mark all local LSAs as dirty */
+  WALK_SLIST(en, p->lsal)
+    if (en->lsa.rt == p->router_id)
+      en->dirty = 1;
+}
 
 static u32
 ort_to_lsaid(struct ospf_proto *p, ort *nf)
@@ -1424,6 +1451,7 @@ prepare_prefix_rt_lsa_body(struct ospf_proto *p, struct ospf_area *oa)
   struct ospf_config *cf = (struct ospf_config *) (p->p.cf);
   struct ospf_iface *ifa;
   struct ospf_lsa_prefix *lp;
+  uint max = ospf_is_ip4(p) ? IP4_MAX_PREFIX_LENGTH : IP6_MAX_PREFIX_LENGTH;
   int host_addr = 0;
   int net_lsa;
   int i = 0;
@@ -1457,7 +1485,7 @@ prepare_prefix_rt_lsa_body(struct ospf_proto *p, struct ospf_area *oa)
 	  (a->scope <= SCOPE_LINK))
 	continue;
 
-      if (((a->prefix.pxlen < IP6_MAX_PREFIX_LENGTH) && net_lsa) ||
+      if (((a->prefix.pxlen < max) && net_lsa) ||
 	  configured_stubnet(oa, a))
 	continue;
 
@@ -1465,8 +1493,13 @@ prepare_prefix_rt_lsa_body(struct ospf_proto *p, struct ospf_area *oa)
 	  (ifa->state == OSPF_IS_LOOP) ||
 	  (ifa->type == OSPF_IT_PTMP))
       {
-	net_addr_ip6 net = NET_ADDR_IP6(a->ip, IP6_MAX_PREFIX_LENGTH);
-	lsab_put_prefix(p, (net_addr *) &net, 0);
+	net_addr net;
+	if (a->prefix.type == NET_IP4)
+	  net_fill_ip4(&net, ipa_to_ip4(a->ip), IP4_MAX_PREFIX_LENGTH);
+	else
+	  net_fill_ip6(&net, ipa_to_ip6(a->ip), IP6_MAX_PREFIX_LENGTH);
+
+	lsab_put_prefix(p, &net, 0);
 	host_addr = 1;
       }
       else
@@ -1482,7 +1515,7 @@ prepare_prefix_rt_lsa_body(struct ospf_proto *p, struct ospf_area *oa)
     if (!sn->hidden)
     {
       lsab_put_prefix(p, &sn->prefix, sn->cost);
-      if (sn->prefix.pxlen == IP6_MAX_PREFIX_LENGTH)
+      if (sn->prefix.pxlen == max)
 	host_addr = 1;
       i++;
     }
@@ -1670,6 +1703,59 @@ ospf_originate_prefix_net_lsa(struct ospf_proto *p, struct ospf_iface *ifa)
 
 
 /*
+ *	Grace LSA handling
+ *	Type = LSA_T_GR, opaque type = LSA_OT_GR
+ */
+
+static inline void
+ospf_add_gr_period_tlv(struct ospf_proto *p, uint period)
+{
+  struct ospf_tlv *tlv = lsab_allocz(p, sizeof(struct ospf_tlv) + sizeof(u32));
+  tlv->type = LSA_GR_PERIOD;
+  tlv->length = 4;
+  tlv->data[0] = period;
+}
+
+static inline void
+ospf_add_gr_reason_tlv(struct ospf_proto *p, uint reason)
+{
+  struct ospf_tlv *tlv = lsab_allocz(p, sizeof(struct ospf_tlv) + sizeof(u32));
+  tlv->type = LSA_GR_REASON;
+  tlv->length = 1;
+  tlv->data[0] = reason << 24;
+}
+
+static inline void
+ospf_add_gr_address_tlv(struct ospf_proto *p, ip4_addr addr)
+{
+  struct ospf_tlv *tlv = lsab_allocz(p, sizeof(struct ospf_tlv) + sizeof(u32));
+  tlv->type = LSA_GR_ADDRESS;
+  tlv->length = 4;
+  tlv->data[0] = ip4_to_u32(addr);
+}
+
+void
+ospf_originate_gr_lsa(struct ospf_proto *p, struct ospf_iface *ifa)
+{
+  struct ospf_new_lsa lsa = {
+    .type = LSA_T_GR,
+    .dom  = ifa->iface_id,
+    .id   = ospf_is_v2(p) ? 0 : ifa->iface_id,
+    .ifa  = ifa
+  };
+
+  ospf_add_gr_period_tlv(p, p->gr_time);
+  ospf_add_gr_reason_tlv(p, 0);
+
+  uint t = ifa->type;
+  if (ospf_is_v2(p) && ((t == OSPF_IT_BCAST) || (t == OSPF_IT_NBMA) || (t == OSPF_IT_PTMP)))
+    ospf_add_gr_address_tlv(p, ipa_to_ip4(ifa->addr->ip));
+
+  ospf_originate_lsa(p, &lsa);
+}
+
+
+/*
  *	Router Information LSA handling
  *	Type = LSA_T_RI_AREA, opaque type = LSA_OT_RI
  */
@@ -1712,6 +1798,10 @@ ospf_update_topology(struct ospf_proto *p)
   struct ospf_area *oa;
   struct ospf_iface *ifa;
 
+  /* No LSA reorigination during GR recovery */
+  if (p->gr_recovery)
+    return;
+
   WALK_LIST(oa, p->area_list)
   {
     if (oa->update_rt_lsa)
diff --git a/proto/ospf/topology.h b/proto/ospf/topology.h
index fd70239d..ffae436a 100644
--- a/proto/ospf/topology.h
+++ b/proto/ospf/topology.h
@@ -33,6 +33,7 @@ struct top_hash_entry
   u32 lb_id;			/* Interface ID of link back iface (for bcast or NBMA networks) */
   u32 dist;			/* Distance from the root */
   int ret_count;		/* Number of retransmission lists referencing the entry */
+  u8 dirty;			/* Will be flushed during next LSAdb update unless reoriginated*/
   u8 color;
 #define OUTSPF 0
 #define CANDIDATE 1
@@ -180,6 +181,7 @@ struct top_hash_entry * ospf_originate_lsa(struct ospf_proto *p, struct ospf_new
 void ospf_advance_lsa(struct ospf_proto *p, struct top_hash_entry *en, struct ospf_lsa_header *lsa, u32 type, u32 domain, void *body);
 void ospf_flush_lsa(struct ospf_proto *p, struct top_hash_entry *en);
 void ospf_update_lsadb(struct ospf_proto *p);
+void ospf_mark_lsadb(struct ospf_proto *p);
 
 static inline void ospf_flush2_lsa(struct ospf_proto *p, struct top_hash_entry **en)
 { if (*en) { ospf_flush_lsa(p, *en); *en = NULL; } }
@@ -187,6 +189,7 @@ static inline void ospf_flush2_lsa(struct ospf_proto *p, struct top_hash_entry *
 void ospf_originate_sum_net_lsa(struct ospf_proto *p, struct ospf_area *oa, ort *nf, int metric);
 void ospf_originate_sum_rt_lsa(struct ospf_proto *p, struct ospf_area *oa, u32 drid, int metric, u32 options);
 void ospf_originate_ext_lsa(struct ospf_proto *p, struct ospf_area *oa, ort *nf, u8 mode, u32 metric, u32 ebit, ip_addr fwaddr, u32 tag, int pbit, int dn);
+void ospf_originate_gr_lsa(struct ospf_proto *p, struct ospf_iface *ifa);
 
 void ospf_rt_notify(struct proto *P, struct channel *ch, net *n, rte *new, rte *old);
 void ospf_update_topology(struct ospf_proto *p);