Optimize GRE Tunnel and add support for ERSPAN encap

Change GRE tunnel to use the interface type where the same encap
node is used as output node for all GRE tunnels, instead of having
dedicated output and tx node for each tunnel. This allows for more
efficient tunnel creation and deletion at scale tested at 1000's
of GRE tunnels.

Add support for ERSPAN encap as another tunnel type, in addition
to the existing L3 and TEB types. The GRE ERSPAN encap supported
is type 2 thus GRE encap need to include sequence number and GRE-
ERSPAN tunnel can be created with user secified ERSPAN session ID.
The GRE tunnel lookup hash key is updated to inclue tunnel type
and session ID, in addition to SIP/DIP and FIB index.
Thus, GRE-ERSPAN tunnel can be created, with the appropriate
session ID, to be used as output interface for SPAN config to
send mirrored packets.

Change interface naming so that all GRE tunnels, irrespective of
tunnel type, uses "greN" where N is the instance number. Removed
interface reuse on tunnel creation and deletion to enable unfied
tunnel interface name.

Add support of user specified instance on GRE tunnel creation.
Thus, N in the "greN" interface name can optionally be specified
by user via CLI/API.

Optimize GRE tunnel encap DPO stacking to bypass load-balance DPO
node since packet output on GRE tunnel always belong to the same
flow after 5-tupple hash.

Change-Id: Ifa83915744a1a88045c998604777cc3583f4da52
Signed-off-by: John Lo <loj@cisco.com>

1 files changed, 190 insertions, 81 deletions

diff --git a/src/vnet/gre/gre.c b/src/vnet/gre/gre.c
index c049b87b6a7..a3702920334 100644
--- a/src/vnet/gre/gre.c
+++ b/src/vnet/gre/gre.c
@@ -62,7 +62,7 @@ format_gre_tx_trace (u8 * s, va_list * args)
   gre_tx_trace_t *t = va_arg (*args, gre_tx_trace_t *);
 
   s = format (s, "GRE: tunnel %d len %d src %U dst %U",
-	      t->tunnel_id, clib_net_to_host_u16 (t->length),
+	      t->tunnel_id, t->length,
 	      format_ip46_address, &t->src, IP46_TYPE_ANY,
 	      format_ip46_address, &t->dst, IP46_TYPE_ANY);
   return s;
@@ -100,7 +100,7 @@ format_gre_header_with_length (u8 * s, va_list * args)
 
   s = format (s, "GRE %U", format_gre_protocol, p);
 
-  if (max_header_bytes != 0 && header_bytes > max_header_bytes)
+  if (max_header_bytes != 0 && header_bytes < max_header_bytes)
     {
       gre_protocol_info_t *pi = gre_get_protocol_info (gm, p);
       vlib_node_t *node = vlib_get_node (gm->vlib_main, pi->node_index);
@@ -208,6 +208,7 @@ gre_build_rewrite (vnet_main_t * vnm,
   gre_main_t *gm = &gre_main;
   ip4_and_gre_header_t *h4;
   ip6_and_gre_header_t *h6;
+  gre_header_t *gre;
   u8 *rewrite = NULL;
   gre_tunnel_t *t;
   u32 ti;
@@ -227,9 +228,7 @@ gre_build_rewrite (vnet_main_t * vnm,
     {
       vec_validate (rewrite, sizeof (*h4) - 1);
       h4 = (ip4_and_gre_header_t *) rewrite;
-      h4->gre.protocol =
-	clib_host_to_net_u16 (gre_proto_from_vnet_link (link_type));
-
+      gre = &h4->gre;
       h4->ip4.ip_version_and_header_length = 0x45;
       h4->ip4.ttl = 254;
       h4->ip4.protocol = IP_PROTOCOL_GRE;
@@ -242,9 +241,7 @@ gre_build_rewrite (vnet_main_t * vnm,
     {
       vec_validate (rewrite, sizeof (*h6) - 1);
       h6 = (ip6_and_gre_header_t *) rewrite;
-      h6->gre.protocol =
-	clib_host_to_net_u16 (gre_proto_from_vnet_link (link_type));
-
+      gre = &h6->gre;
       h6->ip6.ip_version_traffic_class_and_flow_label =
 	clib_host_to_net_u32 (6 << 28);
       h6->ip6.hop_limit = 255;
@@ -256,6 +253,15 @@ gre_build_rewrite (vnet_main_t * vnm,
       h6->ip6.dst_address.as_u64[1] = t->tunnel_dst.fp_addr.ip6.as_u64[1];
     }
 
+  if (PREDICT_FALSE (t->type == GRE_TUNNEL_TYPE_ERSPAN))
+    {
+      gre->protocol = clib_host_to_net_u16 (GRE_PROTOCOL_erspan);
+      gre->flags_and_version = clib_host_to_net_u16 (GRE_FLAGS_SEQUENCE);
+    }
+  else
+    gre->protocol =
+      clib_host_to_net_u16 (gre_proto_from_vnet_link (link_type));
+
   return (rewrite);
 }
 
@@ -302,33 +308,43 @@ gre_update_adj (vnet_main_t * vnm, u32 sw_if_index, adj_index_t ai)
   t = pool_elt_at_index (gm->tunnels, ti);
   is_ipv6 = t->tunnel_dst.fp_proto == FIB_PROTOCOL_IP6 ? 1 : 0;
 
-  adj_nbr_midchain_update_rewrite (ai, !is_ipv6 ? gre4_fixup : gre6_fixup,
-				   NULL,
-				   (VNET_LINK_ETHERNET ==
-				    adj_get_link_type (ai) ?
-				    ADJ_FLAG_MIDCHAIN_NO_COUNT :
-				    ADJ_FLAG_NONE), gre_build_rewrite (vnm,
-								       sw_if_index,
-								       adj_get_link_type
-								       (ai),
-								       NULL));
+  adj_nbr_midchain_update_rewrite
+    (ai, !is_ipv6 ? gre4_fixup : gre6_fixup, NULL,
+     (VNET_LINK_ETHERNET == adj_get_link_type (ai) ?
+      ADJ_FLAG_MIDCHAIN_NO_COUNT : ADJ_FLAG_NONE),
+     gre_build_rewrite (vnm, sw_if_index, adj_get_link_type (ai), NULL));
 
   gre_tunnel_stack (ai);
 }
 
+
+typedef enum
+{
+  GRE_ENCAP_NEXT_DROP,
+  GRE_ENCAP_NEXT_L2_MIDCHAIN,
+  GRE_ENCAP_N_NEXT,
+} gre_encap_next_t;
+
+#define NEXT_IDX (GRE_ENCAP_NEXT_L2_MIDCHAIN)
+
 /**
- * @brief TX function. Only called L2. L3 traffic uses the adj-midchains
+ * @brief TX function. Only called for L2 payload including TEB or ERSPAN.
+ *        L3 traffic uses the adj-midchains.
  */
 static uword
-gre_interface_tx_inline (vlib_main_t * vm,
-			 vlib_node_runtime_t * node, vlib_frame_t * frame)
+gre_interface_tx (vlib_main_t * vm,
+		  vlib_node_runtime_t * node, vlib_frame_t * frame)
 {
   gre_main_t *gm = &gre_main;
+  vnet_main_t *vnm = gm->vnet_main;
   u32 next_index;
   u32 *from, *to_next, n_left_from, n_left_to_next;
-  vnet_interface_output_runtime_t *rd = (void *) node->runtime_data;
-  const gre_tunnel_t *gt = pool_elt_at_index (gm->tunnels, rd->dev_instance);
-  u8 is_ipv6 = gt->tunnel_dst.fp_proto == FIB_PROTOCOL_IP6 ? 1 : 0;
+  u32 sw_if_index0 = 0;
+  u32 sw_if_index1 = 0;
+  adj_index_t adj_index0 = ADJ_INDEX_INVALID;
+  adj_index_t adj_index1 = ADJ_INDEX_INVALID;
+  gre_tunnel_t *gt0 = NULL;
+  gre_tunnel_t *gt1 = NULL;
 
   /* Vector of buffer / pkt indices we're supposed to process */
   from = vlib_frame_vector_args (frame);
@@ -344,77 +360,193 @@ gre_interface_tx_inline (vlib_main_t * vm,
       /* set up to enqueue to our disposition with index = next_index */
       vlib_get_next_frame (vm, node, next_index, to_next, n_left_to_next);
 
-      /*
-       * FIXME DUAL LOOP
-       */
+      while (n_left_from >= 4 && n_left_to_next >= 2)
+	{
+	  u32 bi0 = from[0];
+	  u32 bi1 = from[1];
+	  vlib_buffer_t *b0 = vlib_get_buffer (vm, bi0);
+	  vlib_buffer_t *b1 = vlib_get_buffer (vm, bi1);
+
+	  to_next[0] = bi0;
+	  to_next[1] = bi1;
+	  from += 2;
+	  to_next += 2;
+	  n_left_to_next -= 2;
+	  n_left_from -= 2;
+
+	  if (sw_if_index0 != vnet_buffer (b0)->sw_if_index[VLIB_TX])
+	    {
+	      sw_if_index0 = vnet_buffer (b0)->sw_if_index[VLIB_TX];
+	      vnet_hw_interface_t *hi0 =
+		vnet_get_sup_hw_interface (vnm, sw_if_index0);
+	      gt0 = &gm->tunnels[hi0->dev_instance];
+	      adj_index0 = gt0->l2_adj_index;
+	    }
+
+	  if (sw_if_index0 == vnet_buffer (b1)->sw_if_index[VLIB_TX])
+	    {
+	      sw_if_index1 = sw_if_index0;
+	      gt1 = gt0;
+	      adj_index1 = adj_index0;
+	    }
+	  else if (sw_if_index1 != vnet_buffer (b1)->sw_if_index[VLIB_TX])
+	    {
+	      sw_if_index1 = vnet_buffer (b1)->sw_if_index[VLIB_TX];
+	      vnet_hw_interface_t *hi1 =
+		vnet_get_sup_hw_interface (vnm, sw_if_index1);
+	      gt1 = &gm->tunnels[hi1->dev_instance];
+	      adj_index1 = gt1->l2_adj_index;
+	    }
+
+	  vnet_buffer (b0)->ip.adj_index[VLIB_TX] = adj_index0;
+	  vnet_buffer (b1)->ip.adj_index[VLIB_TX] = adj_index1;
+
+	  if (PREDICT_FALSE (gt0->type == GRE_TUNNEL_TYPE_ERSPAN))
+	    {
+	      /* Encap GRE seq# and ERSPAN type II header */
+	      vlib_buffer_advance (b0, -sizeof (erspan_t2_t));
+	      erspan_t2_t *h0 = vlib_buffer_get_current (b0);
+	      u32 seq_num = clib_smp_atomic_add (&gt0->gre_sn->seq_num, 1);
+	      u64 ver1 = clib_host_to_net_u64 (0x1000000000000000);
+	      h0->seq_num = clib_host_to_net_u32 (seq_num);
+	      h0->t2_u64 = ver1;	/* all 0's except ver=1 */
+	      h0->t2.cos_en_t_session =
+		clib_host_to_net_u16 (gt0->session_id);
+	    }
+	  if (PREDICT_FALSE (gt1->type == GRE_TUNNEL_TYPE_ERSPAN))
+	    {
+	      /* Encap GRE seq# and ERSPAN type II header */
+	      vlib_buffer_advance (b1, -sizeof (erspan_t2_t));
+	      erspan_t2_t *h1 = vlib_buffer_get_current (b1);
+	      u32 seq_num = clib_smp_atomic_add (&gt1->gre_sn->seq_num, 1);
+	      u64 ver1 = clib_host_to_net_u64 (0x1000000000000000);
+	      h1->seq_num = clib_host_to_net_u32 (seq_num);
+	      h1->t2_u64 = ver1;	/* all 0's except ver=1 */
+	      h1->t2.cos_en_t_session =
+		clib_host_to_net_u16 (gt1->session_id);
+	    }
+
+	  if (PREDICT_FALSE (b0->flags & VLIB_BUFFER_IS_TRACED))
+	    {
+	      gre_tx_trace_t *tr0 = vlib_add_trace (vm, node,
+						    b0, sizeof (*tr0));
+	      tr0->tunnel_id = gt0 - gm->tunnels;
+	      tr0->src = gt0->tunnel_src;
+	      tr0->dst = gt0->tunnel_dst.fp_addr;
+	      tr0->length = vlib_buffer_length_in_chain (vm, b0);
+	    }
+	  if (PREDICT_FALSE (b1->flags & VLIB_BUFFER_IS_TRACED))
+	    {
+	      gre_tx_trace_t *tr1 = vlib_add_trace (vm, node,
+						    b1, sizeof (*tr1));
+	      tr1->tunnel_id = gt1 - gm->tunnels;
+	      tr1->src = gt1->tunnel_src;
+	      tr1->dst = gt1->tunnel_dst.fp_addr;
+	      tr1->length = vlib_buffer_length_in_chain (vm, b1);
+	    }
+
+	  vlib_validate_buffer_enqueue_x2 (vm, node, next_index,
+					   to_next, n_left_to_next,
+					   bi0, bi1, NEXT_IDX, NEXT_IDX);
+	}
 
       while (n_left_from > 0 && n_left_to_next > 0)
 	{
-	  vlib_buffer_t *b0;
-	  u32 bi0;
+	  u32 bi0 = from[0];
+	  vlib_buffer_t *b0 = vlib_get_buffer (vm, bi0);
 
-	  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);
+	  if (sw_if_index0 != vnet_buffer (b0)->sw_if_index[VLIB_TX])
+	    {
+	      sw_if_index0 = vnet_buffer (b0)->sw_if_index[VLIB_TX];
+	      vnet_hw_interface_t *hi0 =
+		vnet_get_sup_hw_interface (vnm, sw_if_index0);
+	      gt0 = &gm->tunnels[hi0->dev_instance];
+	      adj_index0 = gt0->l2_adj_index;
+	    }
 
-	  vnet_buffer (b0)->ip.adj_index[VLIB_TX] = gt->l2_adj_index;
+	  vnet_buffer (b0)->ip.adj_index[VLIB_TX] = adj_index0;
+
+	  if (PREDICT_FALSE (gt0->type == GRE_TUNNEL_TYPE_ERSPAN))
+	    {
+	      /* Encap GRE seq# and ERSPAN type II header */
+	      vlib_buffer_advance (b0, -sizeof (erspan_t2_t));
+	      erspan_t2_t *h0 = vlib_buffer_get_current (b0);
+	      u32 seq_num = clib_smp_atomic_add (&gt0->gre_sn->seq_num, 1);
+	      u64 ver1 = clib_host_to_net_u64 (0x1000000000000000);
+	      h0->seq_num = clib_host_to_net_u32 (seq_num);
+	      h0->t2_u64 = ver1;	/* all 0's except ver=1 */
+	      h0->t2.cos_en_t_session =
+		clib_host_to_net_u16 (gt0->session_id);
+	    }
 
 	  if (PREDICT_FALSE (b0->flags & VLIB_BUFFER_IS_TRACED))
 	    {
 	      gre_tx_trace_t *tr = vlib_add_trace (vm, node,
 						   b0, sizeof (*tr));
-	      tr->tunnel_id = gt - gm->tunnels;
-	      tr->src = gt->tunnel_src;
-	      tr->dst = gt->tunnel_src;
+	      tr->tunnel_id = gt0 - gm->tunnels;
+	      tr->src = gt0->tunnel_src;
+	      tr->dst = gt0->tunnel_dst.fp_addr;
 	      tr->length = vlib_buffer_length_in_chain (vm, b0);
 	    }
 
 	  vlib_validate_buffer_enqueue_x1 (vm, node, next_index,
 					   to_next, n_left_to_next,
-					   bi0, gt->l2_tx_arc);
+					   bi0, NEXT_IDX);
 	}
 
       vlib_put_next_frame (vm, node, next_index, n_left_to_next);
     }
 
-  vlib_node_increment_counter (vm, !is_ipv6 ? gre4_input_node.index :
-			       gre6_input_node.index,
+  vlib_node_increment_counter (vm, node->node_index,
 			       GRE_ERROR_PKTS_ENCAP, frame->n_vectors);
 
   return frame->n_vectors;
 }
 
-static uword
-gre_interface_tx (vlib_main_t * vm,
-		  vlib_node_runtime_t * node, vlib_frame_t * frame)
-{
-  return (gre_interface_tx_inline (vm, node, frame));
-}
+static char *gre_error_strings[] = {
+#define gre_error(n,s) s,
+#include "error.def"
+#undef gre_error
+};
 
-static uword
-gre_teb_interface_tx (vlib_main_t * vm,
-		      vlib_node_runtime_t * node, vlib_frame_t * frame)
+/* *INDENT-OFF* */
+VLIB_REGISTER_NODE (gre_encap_node) =
 {
-  return (gre_interface_tx_inline (vm, node, frame));
-}
+  .function = gre_interface_tx,
+  .name = "gre-encap",
+  .vector_size = sizeof (u32),
+  .format_trace = format_gre_tx_trace,
+  .type = VLIB_NODE_TYPE_INTERNAL,
+  .n_errors = GRE_N_ERROR,
+  .error_strings = gre_error_strings,
+  .n_next_nodes = GRE_ENCAP_N_NEXT,
+  .next_nodes = {
+    [GRE_ENCAP_NEXT_DROP] = "error-drop",
+    [GRE_ENCAP_NEXT_L2_MIDCHAIN] = "adj-l2-midchain",
+  },
+};
+
+VLIB_NODE_FUNCTION_MULTIARCH (gre_encap_node, gre_interface_tx)
+/* *INDENT-ON* */
 
 static u8 *
 format_gre_tunnel_name (u8 * s, va_list * args)
 {
   u32 dev_instance = va_arg (*args, u32);
-  return format (s, "gre%d", dev_instance);
-}
+  gre_main_t *gm = &gre_main;
+  gre_tunnel_t *t;
 
-static u8 *
-format_gre_tunnel_teb_name (u8 * s, va_list * args)
-{
-  u32 dev_instance = va_arg (*args, u32);
-  return format (s, "teb-gre%d", dev_instance);
+  if (dev_instance >= vec_len (gm->tunnels))
+    return format (s, "<improperly-referenced>");
+
+  t = pool_elt_at_index (gm->tunnels, dev_instance);
+  return format (s, "gre%d", t->user_instance);
 }
 
 static u8 *
@@ -433,36 +565,11 @@ VNET_DEVICE_CLASS (gre_device_class) = {
   .format_device_name = format_gre_tunnel_name,
   .format_device = format_gre_device,
   .format_tx_trace = format_gre_tx_trace,
-  .tx_function = gre_interface_tx,
   .admin_up_down_function = gre_interface_admin_up_down,
 #ifdef SOON
   .clear counter = 0;
 #endif
 };
-/* *INDENT-ON* */
-
-
-/* *INDENT-OFF* */
-VLIB_DEVICE_TX_FUNCTION_MULTIARCH (gre_device_class,
-                                   gre_interface_tx)
-
-VNET_DEVICE_CLASS (gre_device_teb_class) = {
-  .name = "GRE TEB tunnel device",
-  .format_device_name = format_gre_tunnel_teb_name,
-  .format_device = format_gre_device,
-  .format_tx_trace = format_gre_tx_trace,
-  .tx_function = gre_teb_interface_tx,
-  .admin_up_down_function = gre_interface_admin_up_down,
-#ifdef SOON
-  .clear counter = 0;
-#endif
-};
-
-/* *INDENT-ON* */
-
-/* *INDENT-OFF* */
-VLIB_DEVICE_TX_FUNCTION_MULTIARCH (gre_device_teb_class,
-                                   gre_teb_interface_tx)
 
 VNET_HW_INTERFACE_CLASS (gre_hw_interface_class) = {
   .name = "GRE",
@@ -523,6 +630,8 @@ gre_init (vlib_main_t * vm)
     hash_create_mem (0, sizeof (gre_tunnel_key4_t), sizeof (uword));
   gm->tunnel_by_key6 =
     hash_create_mem (0, sizeof (gre_tunnel_key6_t), sizeof (uword));
+  gm->seq_num_by_key =
+    hash_create_mem (0, sizeof (gre_sn_key_t), sizeof (uword));
 
 #define _(n,s) add_protocol (gm, GRE_PROTOCOL_##s, #s);
   foreach_gre_protocol