Basic support for LISP-GPE encapsulated NSH packets

Change-Id: I97fedb0f70dd18ed9bbe985407cc5fe714e8a2e2
Signed-off-by: Florin Coras <fcoras@cisco.com>

6 files changed, 543 insertions, 12 deletions

diff --git a/src/vnet/lisp-gpe/interface.c b/src/vnet/lisp-gpe/interface.c
index 3288b241..d12dc362 100644
--- a/src/vnet/lisp-gpe/interface.c
+++ b/src/vnet/lisp-gpe/interface.c
@@ -201,7 +201,7 @@ VNET_HW_INTERFACE_CLASS (lisp_gpe_hw_class) = {
 
 typedef struct
 {
-  u32 lb_index;
+  u32 dpo_index;
 } l2_lisp_gpe_tx_trace_t;
 
 static u8 *
@@ -211,7 +211,7 @@ format_l2_lisp_gpe_tx_trace (u8 * s, va_list * args)
   CLIB_UNUSED (vlib_node_t * node) = va_arg (*args, vlib_node_t *);
   l2_lisp_gpe_tx_trace_t *t = va_arg (*args, l2_lisp_gpe_tx_trace_t *);
 
-  s = format (s, "L2-LISP-GPE-TX: load-balance %d", t->lb_index);
+  s = format (s, "L2-LISP-GPE-TX: load-balance %d", t->dpo_index);
   return s;
 }
 
@@ -278,7 +278,7 @@ l2_lisp_gpe_interface_tx (vlib_main_t * vm, vlib_node_runtime_t * node,
 	    {
 	      l2_lisp_gpe_tx_trace_t *tr = vlib_add_trace (vm, node, b0,
 							   sizeof (*tr));
-	      tr->lb_index = lbi0;
+	      tr->dpo_index = lbi0;
 	    }
 	  vlib_validate_buffer_enqueue_x1 (vm, node, next_index, to_next,
 					   n_left_to_next, bi0, l2_arc_to_lb);
@@ -306,6 +306,110 @@ VNET_DEVICE_CLASS (l2_lisp_gpe_device_class,static) = {
 };
 /* *INDENT-ON* */
 
+typedef struct
+{
+  u32 dpo_index;
+} nsh_lisp_gpe_tx_trace_t;
+
+u8 *
+format_nsh_lisp_gpe_tx_trace (u8 * s, va_list * args)
+{
+  CLIB_UNUSED (vlib_main_t * vm) = va_arg (*args, vlib_main_t *);
+  CLIB_UNUSED (vlib_node_t * node) = va_arg (*args, vlib_node_t *);
+  nsh_lisp_gpe_tx_trace_t *t = va_arg (*args, nsh_lisp_gpe_tx_trace_t *);
+
+  s = format (s, "NSH-GPE-TX: tunnel %d", t->dpo_index);
+  return s;
+}
+
+/**
+ * @brief LISP-GPE interface TX for NSH overlays.
+ * @node nsh_lisp_gpe_interface_tx
+ *
+ * The NSH LISP-GPE interface TX function.
+ *
+ * @param[in]   vm        vlib_main_t corresponding to the current thread.
+ * @param[in]   node      vlib_node_runtime_t data for this node.
+ * @param[in]   frame     vlib_frame_t whose contents should be dispatched.
+ *
+ * @return number of vectors in frame.
+ */
+static uword
+nsh_lisp_gpe_interface_tx (vlib_main_t * vm, vlib_node_runtime_t * node,
+			   vlib_frame_t * from_frame)
+{
+  u32 n_left_from, next_index, *from, *to_next;
+  lisp_gpe_main_t *lgm = &lisp_gpe_main;
+
+  from = vlib_frame_vector_args (from_frame);
+  n_left_from = from_frame->n_vectors;
+
+  next_index = node->cached_next_index;
+
+  while (n_left_from > 0)
+    {
+      u32 n_left_to_next;
+
+      vlib_get_next_frame (vm, node, next_index, to_next, n_left_to_next);
+
+      while (n_left_from > 0 && n_left_to_next > 0)
+	{
+	  vlib_buffer_t *b0;
+	  u32 bi0;
+	  u32 *nsh0, next0;
+	  const dpo_id_t *dpo0;
+
+	  bi0 = from[0];
+	  to_next[0] = bi0;
+	  from += 1;
+	  to_next += 1;
+	  n_left_from -= 1;
+	  n_left_to_next -= 1;
+
+	  b0 = vlib_get_buffer (vm, bi0);
+	  nsh0 = vlib_buffer_get_current (b0);
+
+	  vnet_buffer (b0)->lisp.overlay_afi = LISP_AFI_LCAF;
+
+	  /* lookup SPI + SI (second word of the NSH header).
+	   * NB: Load balancing was done by the control plane */
+	  dpo0 = lisp_nsh_fib_lookup (lgm, nsh0[1]);
+
+	  next0 = dpo0->dpoi_next_node;
+	  vnet_buffer (b0)->ip.adj_index[VLIB_TX] = dpo0->dpoi_index;
+
+	  if (PREDICT_FALSE (b0->flags & VLIB_BUFFER_IS_TRACED))
+	    {
+	      nsh_lisp_gpe_tx_trace_t *tr = vlib_add_trace (vm, node, b0,
+							    sizeof (*tr));
+	      tr->dpo_index = dpo0->dpoi_index;
+	    }
+	  vlib_validate_buffer_enqueue_x1 (vm, node, next_index, to_next,
+					   n_left_to_next, bi0, next0);
+	}
+
+      vlib_put_next_frame (vm, node, next_index, n_left_to_next);
+    }
+
+  return from_frame->n_vectors;
+}
+
+static u8 *
+format_nsh_lisp_gpe_name (u8 * s, va_list * args)
+{
+  u32 dev_instance = va_arg (*args, u32);
+  return format (s, "nsh_lisp_gpe%d", dev_instance);
+}
+
+/* *INDENT-OFF* */
+VNET_DEVICE_CLASS (nsh_lisp_gpe_device_class,static) = {
+  .name = "NSH_LISP_GPE",
+  .format_device_name = format_nsh_lisp_gpe_name,
+  .format_tx_trace = format_nsh_lisp_gpe_tx_trace,
+  .tx_function = nsh_lisp_gpe_interface_tx,
+};
+/* *INDENT-ON* */
+
 static vnet_hw_interface_t *
 lisp_gpe_create_iface (lisp_gpe_main_t * lgm, u32 vni, u32 dp_table,
 		       vnet_device_class_t * dev_class,
@@ -615,6 +719,72 @@ lisp_gpe_del_l2_iface (lisp_gpe_main_t * lgm, u32 vni, u32 bd_id)
   lisp_gpe_remove_iface (lgm, hip[0], bd_index, &lgm->l2_ifaces);
 }
 
+/**
+ * @brief Add LISP-GPE NSH interface.
+ *
+ * Creates LISP-GPE interface, sets it in L3 mode.
+ *
+ * @param[in]   lgm     Reference to @ref lisp_gpe_main_t.
+ * @param[in]   a       Parameters to create interface.
+ *
+ * @return sw_if_index.
+ */
+u32
+lisp_gpe_add_nsh_iface (lisp_gpe_main_t * lgm)
+{
+  vnet_main_t *vnm = lgm->vnet_main;
+  tunnel_lookup_t *nsh_ifaces = &lgm->nsh_ifaces;
+  vnet_hw_interface_t *hi;
+  uword *hip, *si;
+
+  hip = hash_get (nsh_ifaces->hw_if_index_by_dp_table, 0);
+
+  if (hip)
+    {
+      clib_warning ("NSH interface 0 already exists");
+      return ~0;
+    }
+
+  si = hash_get (nsh_ifaces->sw_if_index_by_vni, 0);
+  if (si)
+    {
+      clib_warning ("NSH interface already exists");
+      return ~0;
+    }
+
+  /* create lisp iface and populate tunnel tables */
+  hi = lisp_gpe_create_iface (lgm, 0, 0,
+			      &nsh_lisp_gpe_device_class, &lgm->nsh_ifaces);
+
+  /* enable interface */
+  vnet_sw_interface_set_flags (vnm, hi->sw_if_index,
+			       VNET_SW_INTERFACE_FLAG_ADMIN_UP);
+  vnet_hw_interface_set_flags (vnm, hi->hw_if_index,
+			       VNET_HW_INTERFACE_FLAG_LINK_UP);
+
+  return (hi->sw_if_index);
+}
+
+/**
+ * @brief Del LISP-GPE NSH interface.
+ *
+ */
+void
+lisp_gpe_del_nsh_iface (lisp_gpe_main_t * lgm)
+{
+  tunnel_lookup_t *nsh_ifaces = &lgm->nsh_ifaces;
+  uword *hip;
+
+  hip = hash_get (nsh_ifaces->hw_if_index_by_dp_table, 0);
+
+  if (hip == 0)
+    {
+      clib_warning ("The NSH 0 interface doesn't exist");
+      return;
+    }
+  lisp_gpe_remove_iface (lgm, hip[0], 0, &lgm->nsh_ifaces);
+}
+
 static clib_error_t *
 lisp_gpe_add_del_iface_command_fn (vlib_main_t * vm, unformat_input_t * input,
 				   vlib_cli_command_t * cmd)
@@ -623,6 +793,7 @@ lisp_gpe_add_del_iface_command_fn (vlib_main_t * vm, unformat_input_t * input,
   u8 is_add = 1;
   u32 table_id, vni, bd_id;
   u8 vni_is_set = 0, vrf_is_set = 0, bd_index_is_set = 0;
+  u8 nsh_iface = 0;
 
   if (vnet_lisp_gpe_enable_disable_status () == 0)
     {
@@ -651,6 +822,10 @@ lisp_gpe_add_del_iface_command_fn (vlib_main_t * vm, unformat_input_t * input,
 	{
 	  bd_index_is_set = 1;
 	}
+      else if (unformat (line_input, "nsh"))
+	{
+	  nsh_iface = 1;
+	}
       else
 	{
 	  return clib_error_return (0, "parse error: '%U'",
@@ -689,6 +864,21 @@ lisp_gpe_add_del_iface_command_fn (vlib_main_t * vm, unformat_input_t * input,
 	lisp_gpe_tenant_l3_iface_unlock (vni);
     }
 
+  if (nsh_iface)
+    {
+      if (is_add)
+	{
+	  if (~0 == lisp_gpe_add_nsh_iface (&lisp_gpe_main))
+	    {
+	      return clib_error_return (0, "NSH interface not created");
+	    }
+	  else
+	    {
+	      lisp_gpe_del_nsh_iface (&lisp_gpe_main);
+	    }
+	}
+    }
+
   return (NULL);
 }
 
diff --git a/src/vnet/lisp-gpe/lisp_gpe.c b/src/vnet/lisp-gpe/lisp_gpe.c
index e78d45c9..e76c03f0 100644
--- a/src/vnet/lisp-gpe/lisp_gpe.c
+++ b/src/vnet/lisp-gpe/lisp_gpe.c
@@ -151,6 +151,7 @@ lisp_gpe_add_del_fwd_entry_command_fn (vlib_main_t * vm,
   gid_address_copy (&a->lcl_eid, leid);
   gid_address_copy (&a->rmt_eid, reid);
   a->locator_pairs = pairs;
+  a->action = action;
 
   rv = vnet_lisp_gpe_add_del_fwd_entry (a, 0);
   if (0 != rv)
@@ -291,7 +292,6 @@ format_vnet_lisp_gpe_status (u8 * s, va_list * args)
   return format (s, "%s", lgm->is_en ? "enabled" : "disabled");
 }
 
-
 /** LISP-GPE init function. */
 clib_error_t *
 lisp_gpe_init (vlib_main_t * vm)
diff --git a/src/vnet/lisp-gpe/lisp_gpe.h b/src/vnet/lisp-gpe/lisp_gpe.h
index 3288c99f..e92df385 100644
--- a/src/vnet/lisp-gpe/lisp_gpe.h
+++ b/src/vnet/lisp-gpe/lisp_gpe.h
@@ -119,6 +119,15 @@ typedef struct lisp_gpe_main
   /** Load-balance for a miss in the table */
   dpo_id_t l2_lb_cp_lkup;
 
+  /* NSH data structures
+   * ================== */
+
+    BVT (clib_bihash) nsh_fib;
+
+  tunnel_lookup_t nsh_ifaces;
+
+  const dpo_id_t *nsh_cp_lkup;
+
   /** convenience */
   vlib_main_t *vlib_main;
   vnet_main_t *vnet_main;
diff --git a/src/vnet/lisp-gpe/lisp_gpe_adjacency.c b/src/vnet/lisp-gpe/lisp_gpe_adjacency.c
index 8c96a25c..1dbf8677 100644
--- a/src/vnet/lisp-gpe/lisp_gpe_adjacency.c
+++ b/src/vnet/lisp-gpe/lisp_gpe_adjacency.c
@@ -211,6 +211,8 @@ lisp_gpe_adj_proto_from_vnet_link_type (vnet_link_t linkt)
       return (LISP_GPE_NEXT_PROTO_IP6);
     case VNET_LINK_ETHERNET:
       return (LISP_GPE_NEXT_PROTO_ETHERNET);
+    case VNET_LINK_NSH:
+      return (LISP_GPE_NEXT_PROTO_NSH);
     default:
       ASSERT (0);
     }
@@ -254,14 +256,14 @@ lisp_gpe_update_adjacency (vnet_main_t * vnm, u32 sw_if_index, adj_index_t ai)
   ladj = pool_elt_at_index (lisp_adj_pool, lai);
   lgt = lisp_gpe_tunnel_get (ladj->tunnel_index);
   linkt = adj_get_link_type (ai);
-
   adj_nbr_midchain_update_rewrite
     (ai, lisp_gpe_fixup,
      (VNET_LINK_ETHERNET == linkt ?
       ADJ_MIDCHAIN_FLAG_NO_COUNT :
       ADJ_MIDCHAIN_FLAG_NONE),
-     lisp_gpe_tunnel_build_rewrite
-     (lgt, ladj, lisp_gpe_adj_proto_from_vnet_link_type (linkt)));
+     lisp_gpe_tunnel_build_rewrite (lgt, ladj,
+				    lisp_gpe_adj_proto_from_vnet_link_type
+				    (linkt)));
 
   lisp_gpe_adj_stack_one (ladj, ai);
 }
diff --git a/src/vnet/lisp-gpe/lisp_gpe_fwd_entry.c b/src/vnet/lisp-gpe/lisp_gpe_fwd_entry.c
index 7ad8679e..e51b585e 100644
--- a/src/vnet/lisp-gpe/lisp_gpe_fwd_entry.c
+++ b/src/vnet/lisp-gpe/lisp_gpe_fwd_entry.c
@@ -340,10 +340,14 @@ gid_to_dp_address (gid_address_t * g, dp_address_t * d)
       d->type = FID_ADDR_IP_PREF;
       break;
     case GID_ADDR_MAC:
-    default:
       mac_copy (&d->mac, &gid_address_mac (g));
       d->type = FID_ADDR_MAC;
       break;
+    case GID_ADDR_NSH:
+    default:
+      d->nsh = gid_address_nsh (g).spi << 8 | gid_address_nsh (g).si;
+      d->type = FID_ADDR_NSH;
+      break;
     }
 }
 
@@ -671,7 +675,7 @@ del_l2_fwd_entry (lisp_gpe_main_t * lgm,
 }
 
 /**
- * @brief Construct and insert the forwarding information used by a L2 entry
+ * @brief Construct and insert the forwarding information used by an L2 entry
  */
 static void
 lisp_gpe_l2_update_fwding (lisp_gpe_fwd_entry_t * lfe)
@@ -688,7 +692,16 @@ lisp_gpe_l2_update_fwding (lisp_gpe_fwd_entry_t * lfe)
     }
   else
     {
-      dpo_copy (&dpo, &lgm->l2_lb_cp_lkup);
+      switch (lfe->action)
+	{
+	case SEND_MAP_REQUEST:
+	  dpo_copy (&dpo, &lgm->l2_lb_cp_lkup);
+	  break;
+	case NO_ACTION:
+	case FORWARD_NATIVE:
+	case DROP:
+	  dpo_copy (&dpo, drop_dpo_get (DPO_PROTO_ETHERNET));
+	}
     }
 
   /* add entry to l2 lisp fib */
@@ -785,6 +798,276 @@ add_l2_fwd_entry (lisp_gpe_main_t * lgm,
 }
 
 /**
+ * @brief Lookup NSH SD FIB entry
+ *
+ * Does an SPI+SI lookup in the NSH LISP FIB.
+ *
+ * @param[in]   lgm             Reference to @ref lisp_gpe_main_t.
+ * @param[in]   spi_si          SPI + SI.
+ *
+ * @return next node index.
+ */
+const dpo_id_t *
+lisp_nsh_fib_lookup (lisp_gpe_main_t * lgm, u32 spi_si)
+{
+  int rv;
+  BVT (clib_bihash_kv) kv, value;
+
+  memset (&kv, 0, sizeof (kv));
+  kv.key[0] = spi_si;
+  rv = BV (clib_bihash_search_inline_2) (&lgm->nsh_fib, &kv, &value);
+
+  if (rv != 0)
+    {
+      return lgm->nsh_cp_lkup;
+    }
+  else
+    {
+      lisp_gpe_fwd_entry_t *lfe;
+      lfe = pool_elt_at_index (lgm->lisp_fwd_entry_pool, value.value);
+      return &lfe->nsh.choice;
+    }
+}
+
+/**
+ * @brief Add/del NSH FIB entry
+ *
+ * Inserts value in NSH FIB keyed by SPI+SI. If entry is
+ * overwritten the associated value is returned.
+ *
+ * @param[in]   lgm             Reference to @ref lisp_gpe_main_t.
+ * @param[in]   spi_si          SPI + SI.
+ * @param[in]   dpo             Load balanced mapped to SPI + SI
+ *
+ * @return ~0 or value of overwritten entry.
+ */
+static u32
+lisp_nsh_fib_add_del_entry (u32 spi_si, u32 lfei, u8 is_add)
+{
+  lisp_gpe_main_t *lgm = &lisp_gpe_main;
+  BVT (clib_bihash_kv) kv, value;
+  u32 old_val = ~0;
+
+  memset (&kv, 0, sizeof (kv));
+  kv.key[0] = spi_si;
+  kv.value = 0ULL;
+
+  if (BV (clib_bihash_search) (&lgm->nsh_fib, &kv, &value) == 0)
+    old_val = value.value;
+
+  if (!is_add)
+    BV (clib_bihash_add_del) (&lgm->nsh_fib, &kv, 0 /* is_add */ );
+  else
+    {
+      kv.value = lfei;
+      BV (clib_bihash_add_del) (&lgm->nsh_fib, &kv, 1 /* is_add */ );
+    }
+  return old_val;
+}
+
+#define NSH_FIB_DEFAULT_HASH_NUM_BUCKETS (64 * 1024)
+#define NSH_FIB_DEFAULT_HASH_MEMORY_SIZE (32<<20)
+
+static void
+nsh_fib_init (lisp_gpe_main_t * lgm)
+{
+  BV (clib_bihash_init) (&lgm->nsh_fib, "nsh fib",
+			 1 << max_log2 (NSH_FIB_DEFAULT_HASH_NUM_BUCKETS),
+			 NSH_FIB_DEFAULT_HASH_MEMORY_SIZE);
+
+  /*
+   * the result from a 'miss' in a NSH Table
+   */
+  lgm->nsh_cp_lkup = lisp_cp_dpo_get (DPO_PROTO_NSH);
+}
+
+static void
+del_nsh_fwd_entry_i (lisp_gpe_main_t * lgm, lisp_gpe_fwd_entry_t * lfe)
+{
+  lisp_fwd_path_t *path;
+
+  if (LISP_GPE_FWD_ENTRY_TYPE_NEGATIVE != lfe->type)
+    {
+      vec_foreach (path, lfe->paths)
+      {
+	lisp_gpe_adjacency_unlock (path->lisp_adj);
+      }
+      fib_path_list_child_remove (lfe->nsh.path_list_index,
+				  lfe->nsh.child_index);
+      dpo_reset (&lfe->nsh.choice);
+    }
+
+  lisp_nsh_fib_add_del_entry (fid_addr_nsh (&lfe->key->rmt), (u32) ~ 0, 0);
+
+  hash_unset_mem (lgm->lisp_gpe_fwd_entries, lfe->key);
+  clib_mem_free (lfe->key);
+  pool_put (lgm->lisp_fwd_entry_pool, lfe);
+}
+
+/**
+ * @brief Delete LISP NSH forwarding entry.
+ *
+ * Coordinates the removal of forwarding entries for NSH LISP overlay:
+ *
+ * @param[in]   lgm     Reference to @ref lisp_gpe_main_t.
+ * @param[in]   a       Parameters for building the forwarding entry.
+ *
+ * @return 0 on success.
+ */
+static int
+del_nsh_fwd_entry (lisp_gpe_main_t * lgm,
+		   vnet_lisp_gpe_add_del_fwd_entry_args_t * a)
+{
+  lisp_gpe_fwd_entry_key_t key;
+  lisp_gpe_fwd_entry_t *lfe;
+
+  lfe = find_fwd_entry (lgm, a, &key);
+
+  if (NULL == lfe)
+    return VNET_API_ERROR_INVALID_VALUE;
+
+  del_nsh_fwd_entry_i (lgm, lfe);
+
+  return (0);
+}
+
+/**
+ * @brief Construct and insert the forwarding information used by an NSH entry
+ */
+static void
+lisp_gpe_nsh_update_fwding (lisp_gpe_fwd_entry_t * lfe)
+{
+  lisp_gpe_main_t *lgm = vnet_lisp_gpe_get_main ();
+  dpo_id_t dpo = DPO_INVALID;
+  vnet_hw_interface_t *hi;
+  uword *hip;
+
+  if (LISP_GPE_FWD_ENTRY_TYPE_NEGATIVE != lfe->type)
+    {
+      fib_path_list_contribute_forwarding (lfe->nsh.path_list_index,
+					   FIB_FORW_CHAIN_TYPE_NSH,
+					   &lfe->nsh.dpo);
+
+      /*
+       * LISP encap is always the same for this SPI+SI so we do that hash now
+       * and stack on the choice.
+       */
+      if (DPO_LOAD_BALANCE == lfe->nsh.dpo.dpoi_type)
+	{
+	  const dpo_id_t *tmp;
+	  const load_balance_t *lb;
+	  int hash;
+
+	  lb = load_balance_get (lfe->nsh.dpo.dpoi_index);
+	  hash = fid_addr_nsh (&lfe->key->rmt) % lb->lb_n_buckets;
+	  tmp =
+	    load_balance_get_bucket_i (lb, hash & lb->lb_n_buckets_minus_1);
+
+	  dpo_copy (&dpo, tmp);
+	}
+    }
+  else
+    {
+      switch (lfe->action)
+	{
+	case SEND_MAP_REQUEST:
+	  dpo_copy (&dpo, lgm->nsh_cp_lkup);
+	  break;
+	case NO_ACTION:
+	case FORWARD_NATIVE:
+	case DROP:
+	  dpo_copy (&dpo, drop_dpo_get (DPO_PROTO_NSH));
+	}
+    }
+
+  /* We have only one nsh-lisp interface (no NSH virtualization) */
+  hip = hash_get (lgm->nsh_ifaces.hw_if_index_by_dp_table, 0);
+  hi = vnet_get_hw_interface (lgm->vnet_main, hip[0]);
+
+  dpo_stack_from_node (hi->tx_node_index, &lfe->nsh.choice, &dpo);
+
+  /* add entry to nsh lisp fib */
+  lisp_nsh_fib_add_del_entry (fid_addr_nsh (&lfe->key->rmt),
+			      lfe - lgm->lisp_fwd_entry_pool, 1);
+
+  dpo_reset (&dpo);
+}
+
+/**
+ * @brief Add LISP NSH forwarding entry.
+ *
+ * Coordinates the creation of forwarding entries for L2 LISP overlay:
+ * creates lisp-gpe tunnel and injects new entry in Source/Dest L2 FIB.
+ *
+ * @param[in]   lgm     Reference to @ref lisp_gpe_main_t.
+ * @param[in]   a       Parameters for building the forwarding entry.
+ *
+ * @return 0 on success.
+ */
+static int
+add_nsh_fwd_entry (lisp_gpe_main_t * lgm,
+		   vnet_lisp_gpe_add_del_fwd_entry_args_t * a)
+{
+  lisp_gpe_fwd_entry_key_t key;
+  lisp_gpe_fwd_entry_t *lfe;
+
+  lfe = find_fwd_entry (lgm, a, &key);
+
+  if (NULL != lfe)
+    /* don't support updates */
+    return VNET_API_ERROR_INVALID_VALUE;
+
+  pool_get (lgm->lisp_fwd_entry_pool, lfe);
+  memset (lfe, 0, sizeof (*lfe));
+  lfe->key = clib_mem_alloc (sizeof (key));
+  memcpy (lfe->key, &key, sizeof (key));
+
+  hash_set_mem (lgm->lisp_gpe_fwd_entries, lfe->key,
+		lfe - lgm->lisp_fwd_entry_pool);
+
+  lfe->type = (a->is_negative ?
+	       LISP_GPE_FWD_ENTRY_TYPE_NEGATIVE :
+	       LISP_GPE_FWD_ENTRY_TYPE_NORMAL);
+  lfe->tenant = 0;
+
+  if (LISP_GPE_FWD_ENTRY_TYPE_NEGATIVE != lfe->type)
+    {
+      fib_route_path_t *rpaths;
+
+      /*
+       * Make the sorted array of LISP paths with their resp. adjacency
+       */
+      lisp_gpe_fwd_entry_mk_paths (lfe, a);
+
+      /*
+       * From the LISP paths, construct a FIB path list that will
+       * contribute a load-balance.
+       */
+      rpaths = lisp_gpe_mk_fib_paths (lfe->paths);
+
+      lfe->nsh.path_list_index =
+	fib_path_list_create (FIB_PATH_LIST_FLAG_NONE, rpaths);
+
+      /*
+       * become a child of the path-list so we receive updates when
+       * its forwarding state changes. this includes an implicit lock.
+       */
+      lfe->nsh.child_index =
+	fib_path_list_child_add (lfe->nsh.path_list_index,
+				 FIB_NODE_TYPE_LISP_GPE_FWD_ENTRY,
+				 lfe - lgm->lisp_fwd_entry_pool);
+    }
+  else
+    {
+      lfe->action = a->action;
+    }
+
+  lisp_gpe_nsh_update_fwding (lfe);
+
+  return 0;
+}
+
+/**
  * @brief conver from the embedded fib_node_t struct to the LSIP entry
  */
 static lisp_gpe_fwd_entry_t *
@@ -802,7 +1085,12 @@ static fib_node_back_walk_rc_t
 lisp_gpe_fib_node_back_walk (fib_node_t * node,
 			     fib_node_back_walk_ctx_t * ctx)
 {
-  lisp_gpe_l2_update_fwding (lisp_gpe_fwd_entry_from_fib_node (node));
+  lisp_gpe_fwd_entry_t *lfe = lisp_gpe_fwd_entry_from_fib_node (node);
+
+  if (fid_addr_type (&lfe->key->rmt) == FID_ADDR_MAC)
+    lisp_gpe_l2_update_fwding (lfe);
+  else if (fid_addr_type (&lfe->key->rmt) == FID_ADDR_NSH)
+    lisp_gpe_nsh_update_fwding (lfe);
 
   return (FIB_NODE_BACK_WALK_CONTINUE);
 }
@@ -877,6 +1165,11 @@ vnet_lisp_gpe_add_del_fwd_entry (vnet_lisp_gpe_add_del_fwd_entry_args_t * a,
 	return add_l2_fwd_entry (lgm, a);
       else
 	return del_l2_fwd_entry (lgm, a);
+    case GID_ADDR_NSH:
+      if (a->is_add)
+	return add_nsh_fwd_entry (lgm, a);
+      else
+	return del_nsh_fwd_entry (lgm, a);
     default:
       clib_warning ("Forwarding entries for type %d not supported!", type);
       return -1;
@@ -903,6 +1196,9 @@ vnet_lisp_gpe_fwd_entry_flush (void)
       case FID_ADDR_IP_PREF:
 	del_ip_fwd_entry_i (lgm, lfe);
 	break;
+      case FID_ADDR_NSH:
+        del_nsh_fwd_entry_i (lgm, lfe);
+        break;
       }
   }));
   /* *INDENT-ON* */
@@ -967,6 +1263,10 @@ format_lisp_gpe_fwd_entry (u8 * s, va_list ap)
 	  s = format (s, " fib-path-list:%d\n", lfe->l2.path_list_index);
 	  s = format (s, " dpo:%U\n", format_dpo_id, &lfe->l2.dpo, 0);
 	  break;
+	case FID_ADDR_NSH:
+	  s = format (s, " fib-path-list:%d\n", lfe->nsh.path_list_index);
+	  s = format (s, " dpo:%U\n", format_dpo_id, &lfe->nsh.dpo, 0);
+	  break;
 	case FID_ADDR_IP_PREF:
 	  break;
 	}
@@ -1036,6 +1336,7 @@ lisp_gpe_fwd_entry_init (vlib_main_t * vm)
     return (error);
 
   l2_fib_init (lgm);
+  nsh_fib_init (lgm);
 
   fib_node_register_type (FIB_NODE_TYPE_LISP_GPE_FWD_ENTRY, &lisp_fwd_vft);
 
diff --git a/src/vnet/lisp-gpe/lisp_gpe_fwd_entry.h b/src/vnet/lisp-gpe/lisp_gpe_fwd_entry.h
index f7923671..d58895a3 100644
--- a/src/vnet/lisp-gpe/lisp_gpe_fwd_entry.h
+++ b/src/vnet/lisp-gpe/lisp_gpe_fwd_entry.h
@@ -81,7 +81,7 @@ typedef struct lisp_gpe_fwd_entry_t_
   fib_node_t node;
 
   /**
-   * The Entry's key: {lEID,r-EID,vni}
+   * The Entry's key: {lEID,rEID,vni}
    */
   lisp_gpe_fwd_entry_key_t *key;
 
@@ -150,6 +150,33 @@ typedef struct lisp_gpe_fwd_entry_t_
        */
       dpo_id_t dpo;
     } l2;
+
+    /**
+     * Fields relevant to an NSH entry
+     */
+    struct
+    {
+      /**
+       * The path-list created for the forwarding
+       */
+      fib_node_index_t path_list_index;
+
+      /**
+       * Child index of this entry on the path-list
+       */
+      u32 child_index;
+
+      /**
+       * The DPO contributed by NSH
+       */
+      dpo_id_t dpo;
+
+      /**
+       * The DPO used for forwarding. Obtained after stacking tx node
+       * onto lb choice
+       */
+      dpo_id_t choice;
+    } nsh;
   };
 
   union
@@ -177,6 +204,8 @@ extern void vnet_lisp_gpe_fwd_entry_flush (void);
 extern u32 lisp_l2_fib_lookup (lisp_gpe_main_t * lgm,
 			       u16 bd_index, u8 src_mac[8], u8 dst_mac[8]);
 
+extern const dpo_id_t *lisp_nsh_fib_lookup (lisp_gpe_main_t * lgm,
+					    u32 spi_si);
 #endif
 
 /*