Merge branch 'int-new' into nexthop-merged

5 files changed, 542 insertions, 115 deletions

diff --git a/proto/bgp/attrs.c b/proto/bgp/attrs.c
index 1b124a17..f2a8e8b5 100644
--- a/proto/bgp/attrs.c
+++ b/proto/bgp/attrs.c
@@ -41,24 +41,6 @@
  * specifies that such updates should be ignored, but that is generally
  * a bad idea.
  *
- * Error checking of optional transitive attributes is done according to
- * draft-ietf-idr-optional-transitive-03, but errors are handled always
- * as withdraws.
- *
- * Unexpected AS_CONFED_* segments in AS_PATH are logged and removed,
- * but unknown segments cause a session drop with Malformed AS_PATH
- * error (see validate_path()). The behavior in such case is not
- * explicitly specified by RFC 4271. RFC 5065 specifies that
- * inconsistent AS_CONFED_* segments should cause a session drop, but
- * implementations that pass invalid AS_CONFED_* segments are
- * widespread.
- *
- * Error handling of AS4_* attributes is done as specified by
- * draft-ietf-idr-rfc4893bis-03. There are several possible
- * inconsistencies between AGGREGATOR and AS4_AGGREGATOR that are not
- * handled by that draft, these are logged and ignored (see
- * bgp_reconstruct_4b_attrs()).
- *
  * BGP attribute table has several hooks:
  *
  * export - Hook that validates and normalizes attribute during export phase.
@@ -281,11 +263,19 @@ bgp_encode_as_path(struct bgp_write_state *s, eattr *a, byte *buf, uint size)
 static void
 bgp_decode_as_path(struct bgp_parse_state *s, uint code UNUSED, uint flags, byte *data, uint len, ea_list **to)
 {
+  struct bgp_proto *p = s->proto;
+  int as_length = s->as4_session ? 4 : 2;
+  int as_confed = p->cf->confederation && p->is_interior;
   char err[128];
 
-  if (!as_path_valid(data, len, (s->as4_session ? 4 : 2), err, sizeof(err)))
+  if (!as_path_valid(data, len, as_length, as_confed, err, sizeof(err)))
     WITHDRAW("Malformed AS_PATH attribute - %s", err);
 
+  /* In some circumstances check for initial AS_CONFED_SEQUENCE; RFC 5065 5.0 */
+  if (p->is_interior && !p->is_internal &&
+      ((len < 2) || (data[0] != AS_PATH_CONFED_SEQUENCE)))
+    WITHDRAW("Malformed AS_PATH attribute - %s", "missing initial AS_CONFED_SEQUENCE");
+
   if (!s->as4_session)
   {
     /* Prepare 32-bit AS_PATH (from 16-bit one) in a temporary buffer */
@@ -603,11 +593,20 @@ bgp_decode_as4_path(struct bgp_parse_state *s, uint code UNUSED, uint flags, byt
   if (len < 6)
     DISCARD(BAD_LENGTH, "AS4_PATH", len);
 
-  if (!as_path_valid(data, len, 4, err, sizeof(err)))
+  if (!as_path_valid(data, len, 4, 1, err, sizeof(err)))
     DISCARD("Malformed AS4_PATH attribute - %s", err);
 
-  /* XXXX remove CONFED segments */
-  bgp_set_attr_data(to, s->pool, BA_AS4_PATH, flags, data, len);
+  struct adata *a = lp_alloc_adata(s->pool, len);
+  memcpy(a->data, data, len);
+
+  /* AS_CONFED* segments are invalid in AS4_PATH; RFC 6793 6 */
+  if (as_path_contains_confed(a))
+  {
+    REPORT("Discarding AS_CONFED* segment from AS4_PATH attribute");
+    a = as_path_strip_confed(s->pool, a);
+  }
+
+  bgp_set_attr_ptr(to, s->pool, BA_AS4_PATH, flags, a);
 }
 
 static void
@@ -1042,7 +1041,7 @@ bgp_decode_attrs(struct bgp_parse_state *s, byte *data, uint len)
   if (bgp_as_path_loopy(p, attrs, p->local_as))
     goto withdraw;
 
-  /* Reject routes with our Confederation ID in AS_PATH attribute; RFC 5065 4 */
+  /* Reject routes with our Confederation ID in AS_PATH attribute; RFC 5065 4.0 */
   if ((p->public_as != p->local_as) && bgp_as_path_loopy(p, attrs, p->public_as))
     goto withdraw;
 
@@ -1221,8 +1220,17 @@ bgp_init_prefix_table(struct bgp_channel *c)
 {
   HASH_INIT(c->prefix_hash, c->pool, 8);
 
-  c->prefix_slab = sl_new(c->pool, sizeof(struct bgp_prefix) +
-			  net_addr_length[c->c.net_type]);
+  uint alen = net_addr_length[c->c.net_type];
+  c->prefix_slab = alen ? sl_new(c->pool, sizeof(struct bgp_prefix) + alen) : NULL;
+}
+
+void
+bgp_free_prefix_table(struct bgp_channel *c)
+{
+  HASH_FREE(c->prefix_hash);
+
+  rfree(c->prefix_slab);
+  c->prefix_slab = NULL;
 }
 
 static struct bgp_prefix *
@@ -1237,7 +1245,11 @@ bgp_get_prefix(struct bgp_channel *c, net_addr *net, u32 path_id)
     return px;
   }
 
-  px = sl_alloc(c->prefix_slab);
+  if (c->prefix_slab)
+    px = sl_alloc(c->prefix_slab);
+  else
+    px = mb_alloc(c->pool, sizeof(struct bgp_prefix) + net->length);
+
   px->buck_node.next = NULL;
   px->buck_node.prev = NULL;
   px->hash = hash;
@@ -1254,7 +1266,11 @@ bgp_free_prefix(struct bgp_channel *c, struct bgp_prefix *px)
 {
   rem_node(&px->buck_node);
   HASH_REMOVE2(c->prefix_hash, PXH, c->pool, px);
-  sl_free(c->prefix_slab, px);
+
+  if (c->prefix_slab)
+    sl_free(c->prefix_slab, px);
+  else
+    mb_free(px);
 }
 
 
@@ -1278,6 +1294,8 @@ bgp_import_control(struct proto *P, rte **new, ea_list **attrs UNUSED, struct li
   if (src == NULL)
     return 0;
 
+  // XXXX: Check next hop AF
+
   /* IBGP route reflection, RFC 4456 */
   if (p->is_internal && src->is_internal && (p->local_as == src->local_as))
   {
@@ -1314,82 +1332,86 @@ bgp_import_control(struct proto *P, rte **new, ea_list **attrs UNUSED, struct li
   return 0;
 }
 
-static const adata null_adata;	/* adata of length 0 */
 
-static inline void
-bgp_path_prepend(ea_list **attrs, struct linpool *pool, int seg, u32 as, int strip)
-{
-  eattr *a = bgp_find_attr(*attrs, BA_AS_PATH);
-  adata *d = as_path_prepend2(pool, a ? a->u.ptr : &null_adata, seg, as, strip);
-  bgp_set_attr_ptr(attrs, pool, BA_AS_PATH, 0, d);
-}
-
-static inline void
-bgp_cluster_list_prepend(ea_list **attrs, struct linpool *pool, u32 id)
-{
-  eattr *a = bgp_find_attr(*attrs, BA_CLUSTER_LIST);
-  adata *d = int_set_add(pool, a ? a->u.ptr : NULL, id);
-  bgp_set_attr_ptr(attrs, pool, BA_CLUSTER_LIST, 0, d);
-}
+static adata null_adata;	/* adata of length 0 */
 
 static ea_list *
-bgp_update_attrs(struct bgp_proto *p, struct bgp_channel *c, rte *e, ea_list *attrs, struct linpool *pool)
+bgp_update_attrs(struct bgp_proto *p, struct bgp_channel *c, rte *e, ea_list *attrs0, struct linpool *pool)
 {
   struct proto *SRC = e->attrs->src->proto;
   struct bgp_proto *src = (SRC->proto == &proto_bgp) ? (void *) SRC : NULL;
   struct bgp_export_state s = { .proto = p, .channel =c, .pool = pool, .src = src, .route = e };
+  ea_list *attrs = attrs0;
   eattr *a;
+  adata *ad;
 
   /* ORIGIN attribute - mandatory, attach if missing */
-  if (! bgp_find_attr(attrs, BA_ORIGIN))
+  if (! bgp_find_attr(attrs0, BA_ORIGIN))
     bgp_set_attr_u32(&attrs, pool, BA_ORIGIN, 0, src ? ORIGIN_INCOMPLETE : ORIGIN_IGP);
 
+  /* AS_PATH attribute - mandatory */
+  a = bgp_find_attr(attrs0, BA_AS_PATH);
+  ad = a ? a->u.ptr : &null_adata;
+
+  /* AS_PATH attribute - strip AS_CONFED* segments outside confederation */
+  if ((!p->cf->confederation || !p->is_interior) && as_path_contains_confed(ad))
+    ad = as_path_strip_confed(pool, ad);
+
   /* AS_PATH attribute - keep or prepend ASN */
   if (p->is_internal ||
       (p->rs_client && src && src->rs_client))
   {
     /* IBGP or route server -> just ensure there is one */
-    if (! bgp_find_attr(attrs, BA_AS_PATH))
-      bgp_set_attr_ptr(&attrs, pool, BA_AS_PATH, 0, lp_alloc_adata(pool, 0));
+    if (!a)
+      bgp_set_attr_ptr(&attrs, pool, BA_AS_PATH, 0, &null_adata);
   }
   else if (p->is_interior)
   {
-    /* Confederation -> prepend ASN as CONFED_SEQUENCE, keep CONFED_* segments */
-    bgp_path_prepend(&attrs, pool, AS_PATH_CONFED_SEQUENCE, p->public_as, 0);
+    /* Confederation -> prepend ASN as AS_CONFED_SEQUENCE */
+    ad = as_path_prepend2(pool, ad, AS_PATH_CONFED_SEQUENCE, p->public_as);
+    bgp_set_attr_ptr(&attrs, pool, BA_AS_PATH, 0, ad);
   }
   else /* Regular EBGP (no RS, no confederation) */
   {
-    /* Regular EBGP -> prepend ASN as regular segment, strip CONFED_* segments */
-    bgp_path_prepend(&attrs, pool, AS_PATH_SEQUENCE, p->public_as, 1);
+    /* Regular EBGP -> prepend ASN as regular sequence */
+    ad = as_path_prepend2(pool, ad, AS_PATH_SEQUENCE, p->public_as);
+    bgp_set_attr_ptr(&attrs, pool, BA_AS_PATH, 0, ad);
 
     /* MULTI_EXIT_DESC attribute - accept only if set in export filter */
-    a = bgp_find_attr(attrs, BA_MULTI_EXIT_DISC);
+    a = bgp_find_attr(attrs0, BA_MULTI_EXIT_DISC);
     if (a && !(a->type & EAF_FRESH))
       bgp_unset_attr(&attrs, pool, BA_MULTI_EXIT_DISC);
   }
 
   /* NEXT_HOP attribute - delegated to AF-specific hook */
-  a = bgp_find_attr(attrs, BA_NEXT_HOP);
+  a = bgp_find_attr(attrs0, BA_NEXT_HOP);
   bgp_update_next_hop(&s, a, &attrs);
 
   /* LOCAL_PREF attribute - required for IBGP, attach if missing */
-  if (p->is_interior && ! bgp_find_attr(attrs, BA_LOCAL_PREF))
+  if (p->is_interior && ! bgp_find_attr(attrs0, BA_LOCAL_PREF))
     bgp_set_attr_u32(&attrs, pool, BA_LOCAL_PREF, 0, p->cf->default_local_pref);
 
   /* IBGP route reflection, RFC 4456 */
   if (src && src->is_internal && p->is_internal && (src->local_as == p->local_as))
   {
     /* ORIGINATOR_ID attribute - attach if not already set */
-    if (! bgp_find_attr(attrs, BA_ORIGINATOR_ID))
+    if (! bgp_find_attr(attrs0, BA_ORIGINATOR_ID))
       bgp_set_attr_u32(&attrs, pool, BA_ORIGINATOR_ID, 0, src->remote_id);
 
     /* CLUSTER_LIST attribute - prepend cluster ID */
+    a = bgp_find_attr(attrs0, BA_CLUSTER_LIST);
+    ad = a ? a->u.ptr : NULL;
+
+    /* Prepend src cluster ID */
     if (src->rr_cluster_id)
-      bgp_cluster_list_prepend(&attrs, pool, src->rr_cluster_id);
+      ad = int_set_prepend(pool, ad, src->rr_cluster_id);
 
-    /* Handle different src and dst cluster ID - prepend both ones */
+    /* Prepend dst cluster ID if src and dst clusters are different */
     if (p->rr_cluster_id && (src->rr_cluster_id != p->rr_cluster_id))
-      bgp_cluster_list_prepend(&attrs, pool, p->rr_cluster_id);
+      ad = int_set_prepend(pool, ad, p->rr_cluster_id);
+
+    /* Should be at least one prepended cluster ID */
+    bgp_set_attr_ptr(&attrs, pool, BA_CLUSTER_LIST, 0, ad);
   }
 
   /* AS4_* transition attributes, RFC 6793 4.2.2 */
@@ -1410,6 +1432,12 @@ bgp_update_attrs(struct bgp_proto *p, struct bgp_channel *c, rte *e, ea_list *at
     }
   }
 
+  /*
+   * Presence of mandatory attributes ORIGIN and AS_PATH is ensured by above
+   * conditions. Presence and validity of quasi-mandatory NEXT_HOP attribute
+   * should be checked in AF-specific hooks.
+   */
+
   /* Apply per-attribute export hooks for validatation and normalization */
   return bgp_export_attrs(&s, attrs);
 }
@@ -1452,10 +1480,12 @@ bgp_get_neighbor(rte *r)
   eattr *e = ea_find(r->attrs->eattrs, EA_CODE(EAP_BGP, BA_AS_PATH));
   u32 as;
 
-  if (e && as_path_get_first(e->u.ptr, &as))
+  if (e && as_path_get_first_regular(e->u.ptr, &as))
     return as;
-  else
-    return ((struct bgp_proto *) r->attrs->src->proto)->remote_as;
+
+  /* If AS_PATH is not defined, we treat rte as locally originated */
+  struct bgp_proto *p = (void *) r->attrs->src->proto;
+  return p->cf->confederation ?: p->local_as;
 }
 
 static inline int
@@ -1653,7 +1683,7 @@ bgp_rte_mergable(rte *pri, rte *sec)
   }
 
   /* RFC 4271 9.1.2.2. d) Prefer external peers */
-  if (pri_bgp->is_internal != sec_bgp->is_internal)
+  if (pri_bgp->is_interior != sec_bgp->is_interior)
     return 0;
 
   /* RFC 4271 9.1.2.2. e) Compare IGP metrics */
@@ -1843,6 +1873,7 @@ bgp_process_as4_attrs(ea_list **attrs, struct linpool *pool)
   /* Handle AS_PATH attribute */
   if (p2 && p4)
   {
+    /* Both as_path_getlen() and as_path_cut() take AS_CONFED* as zero length */
     int p2_len = as_path_getlen(p2->u.ptr);
     int p4_len = as_path_getlen(p4->u.ptr);
 
diff --git a/proto/bgp/bgp.c b/proto/bgp/bgp.c
index 2ca153ab..5e95e6b4 100644
--- a/proto/bgp/bgp.c
+++ b/proto/bgp/bgp.c
@@ -521,12 +521,17 @@ bgp_conn_enter_established_state(struct bgp_conn *conn)
     if (peer->gr_aware)
       c->load_state = BFS_LOADING;
 
+    c->ext_next_hop = c->cf->ext_next_hop && (bgp_channel_is_ipv6(c) || rem->ext_next_hop);
     c->add_path_rx = (loc->add_path & BGP_ADD_PATH_RX) && (rem->add_path & BGP_ADD_PATH_TX);
     c->add_path_tx = (loc->add_path & BGP_ADD_PATH_TX) && (rem->add_path & BGP_ADD_PATH_RX);
 
-    // XXXX reset back to non-ANY?
+    /* Update RA mode */
     if (c->add_path_tx)
       c->c.ra_mode = RA_ANY;
+    else if (c->cf->secondary)
+      c->c.ra_mode = RA_ACCEPTED;
+    else
+      c->c.ra_mode = RA_OPTIMAL;
   }
 
   p->afi_map = mb_alloc(p->p.pool, num * sizeof(u32));
@@ -554,6 +559,10 @@ bgp_conn_leave_established_state(struct bgp_proto *p)
   BGP_TRACE(D_EVENTS, "BGP session closed");
   p->conn = NULL;
 
+  // XXXX free these tables to avoid memory leak during graceful restart
+  // bgp_free_prefix_table(p);
+  // bgp_free_bucket_table(p);
+
   if (p->p.proto_state == PS_UP)
     bgp_stop(p, 0);
 }
@@ -1411,8 +1420,6 @@ bgp_channel_init(struct channel *C, struct channel_config *CF)
   struct bgp_channel *c = (void *) C;
   struct bgp_channel_config *cf = (void *) CF;
 
-  C->ra_mode = cf->secondary ? RA_ACCEPTED : RA_OPTIMAL;
-
   c->cf = cf;
   c->afi = cf->afi;
   c->desc = bgp_get_af_desc(c->afi);
@@ -1757,10 +1764,131 @@ bgp_get_status(struct proto *P, byte *buf)
 }
 
 static void
+bgp_show_afis(int code, char *s, u32 *afis, uint count)
+{
+  buffer b;
+  LOG_BUFFER_INIT(b);
+
+  buffer_puts(&b, s);
+
+  for (u32 *af = afis; af < (afis + count); af++)
+  {
+    const struct bgp_af_desc *desc = bgp_get_af_desc(*af);
+    if (desc)
+      buffer_print(&b, " %s", desc->name);
+    else
+      buffer_print(&b, " <%u/%u>", BGP_AFI(*af), BGP_SAFI(*af));
+  }
+
+  if (b.pos == b.end)
+    strcpy(b.end - 32, " ... <too long>");
+
+  cli_msg(code, b.start);
+}
+
+static void
+bgp_show_capabilities(struct bgp_proto *p UNUSED, struct bgp_caps *caps)
+{
+  struct bgp_af_caps *ac;
+  uint any_mp_bgp = 0;
+  uint any_gr_able = 0;
+  uint any_add_path = 0;
+  uint any_ext_next_hop = 0;
+  u32 *afl1 = alloca(caps->af_count * sizeof(u32));
+  u32 *afl2 = alloca(caps->af_count * sizeof(u32));
+  uint afn1, afn2;
+
+  WALK_AF_CAPS(caps, ac)
+  {
+    any_mp_bgp |= ac->ready;
+    any_gr_able |= ac->gr_able;
+    any_add_path |= ac->add_path;
+    any_ext_next_hop |= ac->ext_next_hop;
+  }
+
+  if (any_mp_bgp)
+  {
+    cli_msg(-1006, "      Multiprotocol");
+
+    afn1 = 0;
+    WALK_AF_CAPS(caps, ac)
+      if (ac->ready)
+	afl1[afn1++] = ac->afi;
+
+    bgp_show_afis(-1006, "        AF announced:", afl1, afn1);
+  }
+
+  if (caps->route_refresh)
+    cli_msg(-1006, "      Route refresh");
+
+  if (any_ext_next_hop)
+  {
+    cli_msg(-1006, "      Extended next hop");
+
+    afn1 = 0;
+    WALK_AF_CAPS(caps, ac)
+      if (ac->ext_next_hop)
+	afl1[afn1++] = ac->afi;
+
+    bgp_show_afis(-1006, "        IPv6 nexthop:", afl1, afn1);
+  }
+
+  if (caps->ext_messages)
+    cli_msg(-1006, "      Extended message");
+
+  if (caps->gr_aware)
+    cli_msg(-1006, "      Graceful restart");
+
+  if (any_gr_able)
+  {
+    /* Continues from gr_aware */
+    cli_msg(-1006, "        Restart time: %u", caps->gr_time);
+    if (caps->gr_flags & BGP_GRF_RESTART)
+      cli_msg(-1006, "        Restart recovery");
+
+    afn1 = afn2 = 0;
+    WALK_AF_CAPS(caps, ac)
+    {
+      if (ac->gr_able)
+	afl1[afn1++] = ac->afi;
+
+      if (ac->gr_af_flags & BGP_GRF_FORWARDING)
+	afl2[afn2++] = ac->afi;
+    }
+
+    bgp_show_afis(-1006, "        AF supported:", afl1, afn1);
+    bgp_show_afis(-1006, "        AF preserved:", afl2, afn2);
+  }
+
+  if (caps->as4_support)
+    cli_msg(-1006, "      4-octet AS numbers");
+
+  if (any_add_path)
+  {
+    cli_msg(-1006, "      ADD-PATH");
+
+    afn1 = afn2 = 0;
+    WALK_AF_CAPS(caps, ac)
+    {
+      if (ac->add_path & BGP_ADD_PATH_RX)
+	afl1[afn1++] = ac->afi;
+
+      if (ac->add_path & BGP_ADD_PATH_TX)
+	afl2[afn2++] = ac->afi;
+    }
+
+    bgp_show_afis(-1006, "        RX:", afl1, afn1);
+    bgp_show_afis(-1006, "        TX:", afl2, afn2);
+  }
+
+  if (caps->enhanced_refresh)
+    cli_msg(-1006, "      Enhanced refresh");
+}
+
+static void
 bgp_show_proto_info(struct proto *P)
 {
   struct bgp_proto *p = (struct bgp_proto *) P;
-  struct bgp_conn *c = p->conn;
 
   cli_msg(-1006, "  BGP state:          %s", bgp_state_dsc(p));
   cli_msg(-1006, "    Neighbor address: %I%J", p->cf->remote_ip, p->cf->iface);
@@ -1789,15 +1917,11 @@ bgp_show_proto_info(struct proto *P)
   else if (P->proto_state == PS_UP)
   {
     cli_msg(-1006, "    Neighbor ID:      %R", p->remote_id);
+    cli_msg(-1006, "    Local capabilities");
+    bgp_show_capabilities(p, p->conn->local_caps);
+    cli_msg(-1006, "    Neighbor capabilities");
+    bgp_show_capabilities(p, p->conn->remote_caps);
 /* XXXX
-      cli_msg(-1006, "    Neighbor caps:   %s%s%s%s%s%s%s",
-	      c->peer_refresh_support ? " refresh" : "",
-	      c->peer_enhanced_refresh_support ? " enhanced-refresh" : "",
-	      c->peer_gr_able ? " restart-able" : (c->peer_gr_aware ? " restart-aware" : ""),
-	      c->peer_as4_support ? " AS4" : "",
-	      (c->peer_add_path & ADD_PATH_RX) ? " add-path-rx" : "",
-	      (c->peer_add_path & ADD_PATH_TX) ? " add-path-tx" : "",
-	      c->peer_ext_messages_support ? " ext-messages" : "");
       cli_msg(-1006, "    Session:          %s%s%s%s%s%s%s%s",
 	      p->is_internal ? "internal" : "external",
 	      p->cf->multihop ? " multihop" : "",
@@ -1810,9 +1934,9 @@ bgp_show_proto_info(struct proto *P)
 */
     cli_msg(-1006, "    Source address:   %I", p->source_addr);
     cli_msg(-1006, "    Hold timer:       %d/%d",
-	    tm_remains(c->hold_timer), c->hold_time);
+	    tm_remains(p->conn->hold_timer), p->conn->hold_time);
     cli_msg(-1006, "    Keepalive timer:  %d/%d",
-	    tm_remains(c->keepalive_timer), c->keepalive_time);
+	    tm_remains(p->conn->keepalive_timer), p->conn->keepalive_time);
   }
 
   if ((p->last_error_class != BE_NONE) &&
@@ -1846,7 +1970,7 @@ struct protocol proto_bgp = {
   .template = 		"bgp%d",
   .attr_class = 	EAP_BGP,
   .preference = 	DEF_PREF_BGP,
-  .channel_mask =	NB_IP,
+  .channel_mask =	NB_IP | NB_FLOW4 | NB_FLOW6,
   .proto_size =		sizeof(struct bgp_proto),
   .config_size =	sizeof(struct bgp_config),
   .postconfig =		bgp_postconfig,
diff --git a/proto/bgp/bgp.h b/proto/bgp/bgp.h
index 4ecb86a0..5d2539d5 100644
--- a/proto/bgp/bgp.h
+++ b/proto/bgp/bgp.h
@@ -31,6 +31,7 @@ struct eattr;
 
 #define BGP_SAFI_UNICAST	1
 #define BGP_SAFI_MULTICAST	2
+#define BGP_SAFI_FLOW		133
 
 /* Internal AF codes */
 
@@ -42,6 +43,8 @@ struct eattr;
 #define BGP_AF_IPV6		BGP_AF( BGP_AFI_IPV6, BGP_SAFI_UNICAST )
 #define BGP_AF_IPV4_MC		BGP_AF( BGP_AFI_IPV4, BGP_SAFI_MULTICAST )
 #define BGP_AF_IPV6_MC		BGP_AF( BGP_AFI_IPV6, BGP_SAFI_MULTICAST )
+#define BGP_AF_FLOW4		BGP_AF( BGP_AFI_IPV4, BGP_SAFI_FLOW )
+#define BGP_AF_FLOW6		BGP_AF( BGP_AFI_IPV6, BGP_SAFI_FLOW )
 
 
 struct bgp_write_state;
@@ -70,7 +73,7 @@ struct bgp_config {
   u16 local_port;			/* Local listening port */
   u16 remote_port; 			/* Neighbor destination port */
   int multihop;				/* Number of hops if multihop */
-  int strict_bind;			/* Bind listening socket to local address XXXX */
+  int strict_bind;			/* Bind listening socket to local address */
   int ttl_security;			/* Enable TTL security [RFC 5082] */
   int compare_path_lengths;		/* Use path lengths when selecting best route */
   int med_metric;			/* Compare MULTI_EXIT_DISC even between routes from differen ASes */
@@ -120,6 +123,7 @@ struct bgp_channel_config {
   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) */
   u8 gr_able;				/* Allow full graceful restart for the channel */
+  u8 ext_next_hop;			/* Allow both IPv4 and IPv6 next hops */
   u8 add_path;				/* Use ADD-PATH extension [RFC 7911] */
 
   struct rtable_config *igp_table;	/* Table used for recursive next hop lookups */
@@ -151,6 +155,7 @@ struct bgp_af_caps {
   u8 ready;				/* Multiprotocol capability, RFC 4760 */
   u8 gr_able;				/* Graceful restart support, RFC 4724 */
   u8 gr_af_flags;			/* Graceful restart per-AF flags */
+  u8 ext_next_hop;			/* Extended IPv6 next hop,   RFC 5549 */
   u8 add_path;				/* Multiple paths support,   RFC 7911 */
 };
 
@@ -171,6 +176,10 @@ struct bgp_caps {
   struct bgp_af_caps af_data[0];	/* Per-AF capability data */
 };
 
+#define WALK_AF_CAPS(caps,ac) \
+  for (ac = caps->af_data; ac < &caps->af_data[caps->af_count]; ac++)
+
+
 struct bgp_socket {
   node n;				/* Node in global bgp_sockets */
   sock *sk;				/* Real listening socket */
@@ -267,6 +276,8 @@ struct bgp_channel {
   u8 gr_ready;				/* Neighbor could do GR on this AF */
   u8 gr_active;				/* Neighbor is doing GR and keeping fwd state */
 
+  u8 ext_next_hop;			/* Session allows both IPv4 and IPv6 next hops */
+
   u8 add_path_rx;			/* Session expects receive of ADD-PATH extended NLRI */
   u8 add_path_tx;			/* Session expects transmit of ADD-PATH extended NLRI */
 
@@ -446,18 +457,6 @@ bgp_unset_attr(ea_list **to, struct linpool *pool, uint code)
 { eattr *e = bgp_set_attr(to, pool, code, 0, 0); e->type = EAF_TYPE_UNDEF; }
 
 
-
-
-/* Hack: although BA_NEXT_HOP attribute has type EAF_TYPE_IP_ADDRESS, in IPv6
- * we store two addesses in it - a global address and a link local address.
- */
-#ifdef XXX
-#define NEXT_HOP_LENGTH (2*sizeof(ip_addr))
-static inline void set_next_hop(byte *b, ip_addr addr) { ((ip_addr *) b)[0] = addr; ((ip_addr *) b)[1] = IPA_NONE; }
-#define NEXT_HOP_LENGTH sizeof(ip_addr)
-static inline void set_next_hop(byte *b, ip_addr addr) { ((ip_addr *) b)[0] = addr; }
-#endif
-
 int bgp_encode_attrs(struct bgp_write_state *s, ea_list *attrs, byte *buf, byte *end);
 ea_list * bgp_decode_attrs(struct bgp_parse_state *s, byte *data, uint len);
 
@@ -510,26 +509,22 @@ void bgp_update_next_hop(struct bgp_export_state *s, eattr *a, ea_list **to);
 #define BAF_PARTIAL		0x20
 #define BAF_EXT_LEN		0x10
 
-#define BA_ORIGIN		0x01	/* [RFC1771] */		/* WM */
+#define BA_ORIGIN		0x01	/* RFC 4271 */		/* WM */
 #define BA_AS_PATH		0x02				/* WM */
 #define BA_NEXT_HOP		0x03				/* WM */
 #define BA_MULTI_EXIT_DISC	0x04				/* ON */
 #define BA_LOCAL_PREF		0x05				/* WD */
 #define BA_ATOMIC_AGGR		0x06				/* WD */
 #define BA_AGGREGATOR		0x07				/* OT */
-#define BA_COMMUNITY		0x08	/* [RFC1997] */		/* OT */
-#define BA_ORIGINATOR_ID	0x09	/* [RFC1966] */		/* ON */
-#define BA_CLUSTER_LIST		0x0a				/* ON */
-/* We don't support these: */
-#define BA_DPA			0x0b	/* ??? */
-#define BA_ADVERTISER		0x0c	/* [RFC1863] */
-#define BA_RCID_PATH		0x0d
-#define BA_MP_REACH_NLRI	0x0e	/* [RFC2283] */
-#define BA_MP_UNREACH_NLRI	0x0f
+#define BA_COMMUNITY		0x08	/* RFC 1997 */		/* OT */
+#define BA_ORIGINATOR_ID	0x09	/* RFC 4456 */		/* ON */
+#define BA_CLUSTER_LIST		0x0a	/* RFC 4456 */		/* ON */
+#define BA_MP_REACH_NLRI	0x0e	/* RFC 4760 */
+#define BA_MP_UNREACH_NLRI	0x0f	/* RFC 4760 */
 #define BA_EXT_COMMUNITY	0x10	/* RFC 4360 */
 #define BA_AS4_PATH             0x11	/* RFC 6793 */
 #define BA_AS4_AGGREGATOR       0x12	/* RFC 6793 */
-#define BA_LARGE_COMMUNITY	0x20	/* [draft-ietf-idr-large-community] */
+#define BA_LARGE_COMMUNITY	0x20	/* RFC 8092 */
 
 /* BGP connection states */
 
diff --git a/proto/bgp/config.Y b/proto/bgp/config.Y
index 10a338d8..8c63b331 100644
--- a/proto/bgp/config.Y
+++ b/proto/bgp/config.Y
@@ -28,7 +28,7 @@ CF_KEYWORDS(BGP, LOCAL, NEIGHBOR, AS, HOLD, TIME, CONNECT, RETRY, KEEPALIVE,
 	BGP_CLUSTER_LIST, IGP, TABLE, GATEWAY, DIRECT, RECURSIVE, MED, TTL,
 	SECURITY, DETERMINISTIC, SECONDARY, ALLOW, BFD, ADD, PATHS, RX, TX,
 	GRACEFUL, RESTART, AWARE, CHECK, LINK, PORT, EXTENDED, MESSAGES, SETKEY,
-	STRICT, BIND, CONFEDERATION, MEMBER, MULTICAST)
+	STRICT, BIND, CONFEDERATION, MEMBER, MULTICAST, FLOW4, FLOW6)
 
 %type <i32> bgp_afi
 
@@ -139,6 +139,8 @@ bgp_afi:
  | IPV6 { $$ = BGP_AF_IPV6; }
  | IPV4 MULTICAST { $$ = BGP_AF_IPV4_MC; }
  | IPV6 MULTICAST { $$ = BGP_AF_IPV6_MC; }
+ | FLOW4 { $$ = BGP_AF_FLOW4; }
+ | FLOW6 { $$ = BGP_AF_FLOW6; }
  ;
 
 bgp_channel_start: bgp_afi
@@ -150,6 +152,7 @@ bgp_channel_start: bgp_afi
 
   this_channel = channel_config_new(&channel_bgp, desc->net, this_proto);
   BGP_CC->c.name = desc->name;
+  BGP_CC->c.ra_mode = RA_UNDEF;
   BGP_CC->afi = $1;
   BGP_CC->gr_able = 0xff;	/* undefined */
 };
@@ -166,6 +169,7 @@ bgp_channel_item:
  | GATEWAY RECURSIVE { BGP_CC->gw_mode = GW_RECURSIVE; }
  | SECONDARY bool { BGP_CC->secondary = $2; }
  | GRACEFUL RESTART bool { BGP_CC->gr_able = $3; }
+ | EXTENDED NEXT HOP bool { BGP_CC->ext_next_hop = $4; }
  | ADD PATHS RX { BGP_CC->add_path = BGP_ADD_PATH_RX; }
  | ADD PATHS TX { BGP_CC->add_path = BGP_ADD_PATH_TX; }
  | ADD PATHS bool { BGP_CC->add_path = $3 ? BGP_ADD_PATH_FULL : 0; }
diff --git a/proto/bgp/packets.c b/proto/bgp/packets.c
index 9380f999..2106e0d1 100644
--- a/proto/bgp/packets.c
+++ b/proto/bgp/packets.c
@@ -20,6 +20,7 @@
 #include "nest/mrtdump.h"
 #include "conf/conf.h"
 #include "lib/unaligned.h"
+#include "lib/flowspec.h"
 #include "lib/socket.h"
 
 #include "nest/cli.h"
@@ -184,9 +185,6 @@ bgp_create_notification(struct bgp_conn *conn, byte *buf)
 
 /* Capability negotiation as per RFC 5492 */
 
-#define WALK_AF_CAPS(caps,ac) \
-  for (ac = caps->af_data; ac < &caps->af_data[caps->af_count]; ac++)
-
 const struct bgp_af_caps *
 bgp_find_af_caps(struct bgp_caps *caps, u32 afi)
 {
@@ -230,6 +228,7 @@ bgp_write_capabilities(struct bgp_conn *conn, byte *buf)
   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;
 
@@ -261,6 +260,9 @@ bgp_write_capabilities(struct bgp_conn *conn, byte *buf)
     ac->afi = c->afi;
     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;
+
     ac->add_path = c->cf->add_path;
     any_add_path |= ac->add_path;
 
@@ -279,6 +281,12 @@ bgp_write_capabilities(struct bgp_conn *conn, byte *buf)
 
   /* Create capability list in buffer */
 
+  /*
+   * Note that max length is ~ 20+14*af_count. With max 6 channels that is
+   * 104. Option limit is 253 and buffer size is 4096, so we cannot overflow
+   * unless we add new capabilities or more AFs.
+   */
+
   WALK_AF_CAPS(caps, ac)
     if (ac->ready)
     {
@@ -294,6 +302,23 @@ bgp_write_capabilities(struct bgp_conn *conn, byte *buf)
     *buf++ = 0;			/* Capability data length */
   }
 
+  if (any_ext_next_hop)
+  {
+    *buf++ = 5;			/* Capability 5: Support for extended next hop */
+    *buf++ = 0;			/* Capability data length, will be fixed later */
+    data = buf;
+
+    WALK_AF_CAPS(caps, ac)
+      if (ac->ext_next_hop)
+      {
+	put_af4(buf, ac->afi);
+	put_u16(buf+4, BGP_AFI_IPV6);
+	buf += 6;
+      }
+
+    data[-1] = buf - data;
+  }
+
   if (caps->ext_messages)
   {
     *buf++ = 6;			/* Capability 6: Support for extended messages */
@@ -352,8 +377,6 @@ bgp_write_capabilities(struct bgp_conn *conn, byte *buf)
     *buf++ = 0;			/* Capability data length */
   }
 
-  /* FIXME: Should not XXXX 255 */
-
   return buf;
 }
 
@@ -392,6 +415,23 @@ bgp_read_capabilities(struct bgp_conn *conn, struct bgp_caps *caps, byte *pos, i
       caps->route_refresh = 1;
       break;
 
+    case  5: /* Extended next hop encoding capability, RFC 5549 */
+      if (cl % 6)
+	goto err;
+
+      for (i = 0; i < cl; i += 6)
+      {
+	/* Specified only for IPv4 prefixes with IPv6 next hops */
+	if ((get_u16(pos+2+i+0) != BGP_AFI_IPV4) ||
+	    (get_u16(pos+2+i+4) != BGP_AFI_IPV6))
+	  continue;
+
+	af = get_af4(pos+2+i);
+	ac = bgp_get_af_caps(caps, af);
+	ac->ext_next_hop = 1;
+      }
+      break;
+
     case  6: /* Extended message length capability, RFC draft */
       if (cl != 0)
 	goto err;
@@ -673,9 +713,13 @@ bgp_rx_open(struct bgp_conn *conn, byte *pkt, uint len)
 #define REPORT(msg, args...) \
   ({ log(L_REMOTE "%s: " msg, s->proto->p.name, ## args); })
 
+#define DISCARD(msg, args...) \
+  ({ REPORT(msg, ## args); return; })
+
 #define WITHDRAW(msg, args...) \
   ({ REPORT(msg, ## args); s->err_withdraw = 1; return; })
 
+#define BAD_AFI		"Unexpected AF <%u/%u> in UPDATE"
 #define BAD_NEXT_HOP	"Invalid NEXT_HOP attribute"
 #define NO_NEXT_HOP	"Missing NEXT_HOP attribute"
 
@@ -792,6 +836,32 @@ bgp_update_next_hop_ip(struct bgp_export_state *s, eattr *a, ea_list **to)
     WITHDRAW(BAD_NEXT_HOP);
 }
 
+static uint
+bgp_encode_next_hop_none(struct bgp_write_state *s UNUSED, eattr *a UNUSED, byte *buf UNUSED, uint size UNUSED)
+{
+  return 0;
+}
+
+static void
+bgp_decode_next_hop_none(struct bgp_parse_state *s UNUSED, byte *data UNUSED, uint len UNUSED, rta *a UNUSED)
+{
+  /*
+   * Although we expect no next hop and RFC 7606 7.11 states that attribute
+   * MP_REACH_NLRI with unexpected next hop length is considered malformed,
+   * FlowSpec RFC 5575 4 states that next hop shall be ignored on receipt.
+   */
+
+  return;
+}
+
+static void
+bgp_update_next_hop_none(struct bgp_export_state *s, eattr *a, ea_list **to)
+{
+  /* NEXT_HOP shall not pass */
+  if (a)
+    bgp_unset_attr(to, s->pool, BA_NEXT_HOP);
+}
+
 
 /*
  *	UPDATE
@@ -1065,6 +1135,190 @@ bgp_decode_next_hop_ip6(struct bgp_parse_state *s, byte *data, uint len, rta *a)
 }
 
 
+static uint
+bgp_encode_nlri_flow4(struct bgp_write_state *s, struct bgp_bucket *buck, byte *buf, uint size)
+{
+  byte *pos = buf;
+
+  while (!EMPTY_LIST(buck->prefixes) && (size >= 4))
+  {
+    struct bgp_prefix *px = HEAD(buck->prefixes);
+    struct net_addr_flow4 *net = (void *) px->net;
+    uint flen = net->length - sizeof(net_addr_flow4);
+
+    /* Encode path ID */
+    if (s->add_path)
+    {
+      put_u32(pos, px->path_id);
+      ADVANCE(pos, size, 4);
+    }
+
+    if (flen > size)
+      break;
+
+    /* Copy whole flow data including length */
+    memcpy(pos, net->data, flen);
+    ADVANCE(pos, size, flen);
+
+    bgp_free_prefix(s->channel, px);
+  }
+
+  return pos - buf;
+}
+
+static void
+bgp_decode_nlri_flow4(struct bgp_parse_state *s, byte *pos, uint len, rta *a)
+{
+  while (len)
+  {
+    u32 path_id = 0;
+
+    /* Decode path ID */
+    if (s->add_path)
+    {
+      if (len < 4)
+	bgp_parse_error(s, 1);
+
+      path_id = get_u32(pos);
+      ADVANCE(pos, len, 4);
+    }
+
+    if (len < 2)
+      bgp_parse_error(s, 1);
+
+    /* Decode flow length */
+    uint hlen = flow_hdr_length(pos);
+    uint dlen = flow_read_length(pos);
+    uint flen = hlen + dlen;
+    byte *data = pos + hlen;
+
+    if (len < flen)
+      bgp_parse_error(s, 1);
+
+    /* Validate flow data */
+    enum flow_validated_state r = flow4_validate(data, dlen);
+    if (r != FLOW_ST_VALID)
+    {
+      log(L_REMOTE "%s: Invalid flow route: %s", s->proto->p.name, flow_validated_state_str(r));
+      bgp_parse_error(s, 1);
+    }
+
+    if (data[0] != FLOW_TYPE_DST_PREFIX)
+    {
+      log(L_REMOTE "%s: No dst prefix at first pos", s->proto->p.name);
+      bgp_parse_error(s, 1);
+    }
+
+    /* Decode dst prefix */
+    ip4_addr px = IP4_NONE;
+    uint pxlen = data[1];
+
+    // FIXME: Use some generic function
+    memcpy(&px, data, BYTES(pxlen));
+    px = ip4_and(px, ip4_mkmask(pxlen));
+
+    /* Prepare the flow */
+    net_addr *n = alloca(sizeof(struct net_addr_flow4) + flen);
+    net_fill_flow4(n, px, pxlen, pos, flen);
+    ADVANCE(pos, len, flen);
+
+    bgp_rte_update(s, n, path_id, a);
+  }
+}
+
+
+static uint
+bgp_encode_nlri_flow6(struct bgp_write_state *s, struct bgp_bucket *buck, byte *buf, uint size)
+{
+  byte *pos = buf;
+
+  while (!EMPTY_LIST(buck->prefixes) && (size >= 4))
+  {
+    struct bgp_prefix *px = HEAD(buck->prefixes);
+    struct net_addr_flow6 *net = (void *) px->net;
+    uint flen = net->length - sizeof(net_addr_flow6);
+
+    /* Encode path ID */
+    if (s->add_path)
+    {
+      put_u32(pos, px->path_id);
+      ADVANCE(pos, size, 4);
+    }
+
+    if (flen > size)
+      break;
+
+    /* Copy whole flow data including length */
+    memcpy(pos, net->data, flen);
+    ADVANCE(pos, size, flen);
+
+    bgp_free_prefix(s->channel, px);
+  }
+
+  return pos - buf;
+}
+
+static void
+bgp_decode_nlri_flow6(struct bgp_parse_state *s, byte *pos, uint len, rta *a)
+{
+  while (len)
+  {
+    u32 path_id = 0;
+
+    /* Decode path ID */
+    if (s->add_path)
+    {
+      if (len < 4)
+	bgp_parse_error(s, 1);
+
+      path_id = get_u32(pos);
+      ADVANCE(pos, len, 4);
+    }
+
+    if (len < 2)
+      bgp_parse_error(s, 1);
+
+    /* Decode flow length */
+    uint hlen = flow_hdr_length(pos);
+    uint dlen = flow_read_length(pos);
+    uint flen = hlen + dlen;
+    byte *data = pos + hlen;
+
+    if (len < flen)
+      bgp_parse_error(s, 1);
+
+    /* Validate flow data */
+    enum flow_validated_state r = flow6_validate(data, dlen);
+    if (r != FLOW_ST_VALID)
+    {
+      log(L_REMOTE "%s: Invalid flow route: %s", s->proto->p.name, flow_validated_state_str(r));
+      bgp_parse_error(s, 1);
+    }
+
+    if (data[0] != FLOW_TYPE_DST_PREFIX)
+    {
+      log(L_REMOTE "%s: No dst prefix at first pos", s->proto->p.name);
+      bgp_parse_error(s, 1);
+    }
+
+    /* Decode dst prefix */
+    ip6_addr px = IP6_NONE;
+    uint pxlen = data[1];
+
+    // FIXME: Use some generic function
+    memcpy(&px, data, BYTES(pxlen));
+    px = ip6_and(px, ip6_mkmask(pxlen));
+
+    /* Prepare the flow */
+    net_addr *n = alloca(sizeof(struct net_addr_flow6) + flen);
+    net_fill_flow6(n, px, pxlen, pos, flen);
+    ADVANCE(pos, len, flen);
+
+    bgp_rte_update(s, n, path_id, a);
+  }
+}
+
+
 static const struct bgp_af_desc bgp_af_table[] = {
   {
     .afi = BGP_AF_IPV4,
@@ -1087,6 +1341,16 @@ static const struct bgp_af_desc bgp_af_table[] = {
     .update_next_hop = bgp_update_next_hop_ip,
   },
   {
+    .afi = BGP_AF_FLOW4,
+    .net = NET_FLOW4,
+    .name = "flow4",
+    .encode_nlri = bgp_encode_nlri_flow4,
+    .decode_nlri = bgp_decode_nlri_flow4,
+    .encode_next_hop = bgp_encode_next_hop_none,
+    .decode_next_hop = bgp_decode_next_hop_none,
+    .update_next_hop = bgp_update_next_hop_none,
+  },
+  {
     .afi = BGP_AF_IPV6,
     .net = NET_IP6,
     .name = "ipv6",
@@ -1106,6 +1370,16 @@ static const struct bgp_af_desc bgp_af_table[] = {
     .decode_next_hop = bgp_decode_next_hop_ip6,
     .update_next_hop = bgp_update_next_hop_ip,
   },
+  {
+    .afi = BGP_AF_FLOW6,
+    .net = NET_FLOW6,
+    .name = "flow6",
+    .encode_nlri = bgp_encode_nlri_flow6,
+    .decode_nlri = bgp_decode_nlri_flow6,
+    .encode_next_hop = bgp_encode_next_hop_none,
+    .decode_next_hop = bgp_decode_next_hop_none,
+    .update_next_hop = bgp_update_next_hop_none,
+  },
 };
 
 const struct bgp_af_desc *
@@ -1387,15 +1661,15 @@ bgp_create_end_mark(struct bgp_channel *c, byte *buf)
 }
 
 static inline void
-bgp_rx_end_mark(struct bgp_proto *p, u32 afi)
+bgp_rx_end_mark(struct bgp_parse_state *s, u32 afi)
 {
+  struct bgp_proto *p = s->proto;
   struct bgp_channel *c = bgp_get_channel(p, afi);
 
   BGP_TRACE(D_PACKETS, "Got END-OF-RIB");
 
-  /* XXXX handle unknown AF in MP_*_NLRI */
   if (!c)
-    return;
+    DISCARD(BAD_AFI, BGP_AFI(afi), BGP_SAFI(afi));
 
   if (c->load_state == BFS_LOADING)
     c->load_state = BFS_NONE;
@@ -1413,9 +1687,8 @@ bgp_decode_nlri(struct bgp_parse_state *s, u32 afi, byte *nlri, uint len, ea_lis
   struct bgp_channel *c = bgp_get_channel(s->proto, afi);
   rta *a = NULL;
 
-  /* XXXX handle unknown AF in MP_*_NLRI */
   if (!c)
-    return;
+    DISCARD(BAD_AFI, BGP_AFI(afi), BGP_SAFI(afi));
 
   s->channel = c;
   s->add_path = c->add_path_rx;
@@ -1523,12 +1796,12 @@ bgp_rx_update(struct bgp_conn *conn, byte *pkt, uint len)
 
   /* Check for End-of-RIB marker */
   if (!s.attr_len && !s.ip_unreach_len && !s.ip_reach_len)
-  { bgp_rx_end_mark(p, BGP_AF_IPV4); goto done; }
+  { bgp_rx_end_mark(&s, BGP_AF_IPV4); goto done; }
 
   /* Check for MP End-of-RIB marker */
   if ((s.attr_len < 8) && !s.ip_unreach_len && !s.ip_reach_len &&
-      !s.mp_reach_len && !s.mp_unreach_len && s.mp_unreach_af) /* XXXX  See RFC 7606 5.2 */
-  { bgp_rx_end_mark(p, s.mp_unreach_af); goto done; }
+      !s.mp_reach_len && !s.mp_unreach_len && s.mp_unreach_af)
+  { bgp_rx_end_mark(&s, s.mp_unreach_af); goto done; }
 
   if (s.ip_unreach_len)
     bgp_decode_nlri(&s, BGP_AF_IPV4, s.ip_unreach_nlri, s.ip_unreach_len, NULL, NULL, 0);