memif: add ip mode

In IP mode memif interface is L3 point-to-point interfaces and
we don't pass l2 header. There is no l2 header rewrite operation and
received packets are sent straight to ip4-input / ip6-input nodes.

Change-Id: I4177f3fce3004da7ecf14d235006ae053fcf3f09
Signed-off-by: Damjan Marion <damarion@cisco.com>

4 files changed, 132 insertions, 35 deletions

diff --git a/src/plugins/memif/cli.c b/src/plugins/memif/cli.c
index 34c77def79a..8af761646a8 100644
--- a/src/plugins/memif/cli.c
+++ b/src/plugins/memif/cli.c
@@ -62,6 +62,8 @@ memif_create_command_fn (vlib_main_t * vm, unformat_input_t * input,
 	args.is_master = 1;
       else if (unformat (line_input, "slave"))
 	args.is_master = 0;
+      else if (unformat (line_input, "mode ip"))
+	args.mode = MEMIF_INTERFACE_MODE_IP;
       else if (unformat (line_input, "hw-addr %U",
 			 unformat_ethernet_address, args.hw_addr))
 	args.hw_addr_set = 1;
@@ -107,7 +109,8 @@ VLIB_CLI_COMMAND (memif_create_command, static) = {
   .path = "create memif",
   .short_help = "create memif [id <id>] [socket <path>] "
                 "[ring-size <size>] [buffer-size <size>] [hw-addr <mac-address>] "
-		"<master|slave> [rx-queues <number>] [tx-queues <number>]",
+		"<master|slave> [rx-queues <number>] [tx-queues <number>]"
+		"[mode ip]",
   .function = memif_create_command_fn,
 };
 /* *INDENT-ON* */
diff --git a/src/plugins/memif/memif.c b/src/plugins/memif/memif.c
index d21a30eb061..ec67023b673 100644
--- a/src/plugins/memif/memif.c
+++ b/src/plugins/memif/memif.c
@@ -482,7 +482,10 @@ memif_delete_if (vlib_main_t * vm, memif_if_t * mif)
   clib_error_free (err);
 
   /* remove the interface */
-  ethernet_delete_interface (vnm, mif->hw_if_index);
+  if (mif->mode == MEMIF_INTERFACE_MODE_IP)
+    vnet_delete_hw_interface (vnm, mif->hw_if_index);
+  else
+    ethernet_delete_interface (vnm, mif->hw_if_index);
   mif->hw_if_index = ~0;
 
   /* free interface data structures */
@@ -519,6 +522,14 @@ memif_delete_if (vlib_main_t * vm, memif_if_t * mif)
   return 0;
 }
 
+/* *INDENT-OFF* */
+VNET_HW_INTERFACE_CLASS (memif_ip_hw_if_class, static) =
+{
+  .name = "memif-ip",
+  .flags = VNET_HW_INTERFACE_CLASS_FLAG_P2P,
+};
+/* *INDENT-ON* */
+
 int
 memif_create_if (vlib_main_t * vm, memif_create_if_args_t * args)
 {
@@ -614,28 +625,43 @@ memif_create_if (vlib_main_t * vm, memif_create_if_args_t * args)
   mif->sw_if_index = mif->hw_if_index = mif->per_interface_next_index = ~0;
   mif->conn_unix_file_index = ~0;
   mif->conn_fd = -1;
+  mif->mode = args->mode;
   if (args->secret)
     mif->secret = vec_dup (args->secret);
 
   if (tm->n_vlib_mains > 1)
     clib_spinlock_init (&mif->lockp);
 
-  if (!args->hw_addr_set)
+
+  if (mif->mode == MEMIF_INTERFACE_MODE_ETHERNET)
     {
-      f64 now = vlib_time_now (vm);
-      u32 rnd;
-      rnd = (u32) (now * 1e6);
-      rnd = random_u32 (&rnd);
-
-      memcpy (args->hw_addr + 2, &rnd, sizeof (rnd));
-      args->hw_addr[0] = 2;
-      args->hw_addr[1] = 0xfe;
-    }
 
-  error = ethernet_register_interface (vnm, memif_device_class.index,
-				       mif->dev_instance, args->hw_addr,
-				       &mif->hw_if_index,
-				       memif_eth_flag_change);
+      if (!args->hw_addr_set)
+	{
+	  f64 now = vlib_time_now (vm);
+	  u32 rnd;
+	  rnd = (u32) (now * 1e6);
+	  rnd = random_u32 (&rnd);
+
+	  memcpy (args->hw_addr + 2, &rnd, sizeof (rnd));
+	  args->hw_addr[0] = 2;
+	  args->hw_addr[1] = 0xfe;
+	}
+      error = ethernet_register_interface (vnm, memif_device_class.index,
+					   mif->dev_instance, args->hw_addr,
+					   &mif->hw_if_index,
+					   memif_eth_flag_change);
+    }
+  else if (mif->mode == MEMIF_INTERFACE_MODE_IP)
+    {
+      mif->hw_if_index =
+	vnet_register_interface (vnm, memif_device_class.index,
+				 mif->dev_instance,
+				 memif_ip_hw_if_class.index,
+				 mif->dev_instance);
+    }
+  else
+    error = clib_error_return (0, "unsupported interface mode");
 
   if (error)
     {
@@ -725,7 +751,10 @@ memif_create_if (vlib_main_t * vm, memif_create_if_args_t * args)
 error:
   if (mif->hw_if_index != ~0)
     {
-      ethernet_delete_interface (vnm, mif->hw_if_index);
+      if (mif->mode == MEMIF_INTERFACE_MODE_IP)
+	vnet_delete_hw_interface (vnm, mif->hw_if_index);
+      else
+	ethernet_delete_interface (vnm, mif->hw_if_index);
       mif->hw_if_index = ~0;
     }
   memif_delete_if (vm, mif);
diff --git a/src/plugins/memif/node.c b/src/plugins/memif/node.c
index c6403fef7f3..4acc714902e 100644
--- a/src/plugins/memif/node.c
+++ b/src/plugins/memif/node.c
@@ -30,7 +30,8 @@
 #include <memif/memif.h>
 #include <memif/private.h>
 
-#define foreach_memif_input_error
+#define foreach_memif_input_error \
+  _(NOT_IP, "not ip packet")
 
 typedef enum
 {
@@ -200,16 +201,33 @@ memif_copy_buffer_from_rx_ring (vlib_main_t * vm, memif_if_t * mif,
   return (total_bytes);
 }
 
+
+static_always_inline u32
+memif_next_from_ip_hdr (vlib_node_runtime_t * node, vlib_buffer_t * b)
+{
+  u8 *ptr = vlib_buffer_get_current (b);
+  u8 v = *ptr & 0xf0;
+
+  if (PREDICT_TRUE (v == 0x40))
+    return VNET_DEVICE_INPUT_NEXT_IP4_NCS_INPUT;
+  else if (PREDICT_TRUE (v == 0x60))
+    return VNET_DEVICE_INPUT_NEXT_IP6_INPUT;
+
+  b->error = node->errors[MEMIF_INPUT_ERROR_NOT_IP];
+  return VNET_DEVICE_INPUT_NEXT_DROP;
+}
+
 static_always_inline uword
 memif_device_input_inline (vlib_main_t * vm, vlib_node_runtime_t * node,
 			   vlib_frame_t * frame, memif_if_t * mif,
-			   memif_ring_type_t type, u16 qid)
+			   memif_ring_type_t type, u16 qid,
+			   memif_interface_mode_t mode)
 {
   vnet_main_t *vnm = vnet_get_main ();
   memif_ring_t *ring;
   memif_queue_t *mq;
   u16 head;
-  u32 next_index = VNET_DEVICE_INPUT_NEXT_ETHERNET_INPUT;
+  u32 next_index;
   uword n_trace = vlib_get_trace_count (vm, node);
   memif_main_t *nm = &memif_main;
   u32 n_rx_packets = 0;
@@ -227,8 +245,14 @@ memif_device_input_inline (vlib_main_t * vm, vlib_node_runtime_t * node,
   ring_size = 1 << mq->log2_ring_size;
   mask = ring_size - 1;
 
-  if (mif->per_interface_next_index != ~0)
-    next_index = mif->per_interface_next_index;
+  if (mode == MEMIF_INTERFACE_MODE_IP)
+    {
+      next_index = VNET_DEVICE_INPUT_NEXT_IP6_INPUT;
+    }
+  else
+    {
+      next_index = VNET_DEVICE_INPUT_NEXT_ETHERNET_INPUT;
+    }
 
   n_free_bufs = vec_len (nm->rx_buffers[thread_index]);
   if (PREDICT_FALSE (n_free_bufs < ring_size))
@@ -308,6 +332,24 @@ memif_device_input_inline (vlib_main_t * vm, vlib_node_runtime_t * node,
 	  to_next += 2;
 	  n_left_to_next -= 2;
 
+
+	  if (mode == MEMIF_INTERFACE_MODE_IP)
+	    {
+	      next0 = memif_next_from_ip_hdr (node, first_b0);
+	      next1 = memif_next_from_ip_hdr (node, first_b1);
+	    }
+	  else if (mode == MEMIF_INTERFACE_MODE_ETHERNET)
+	    {
+	      if (PREDICT_FALSE (mif->per_interface_next_index != ~0))
+		next0 = next1 = mif->per_interface_next_index;
+	      else
+		/* redirect if feature path
+		 * enabled */
+		vnet_feature_start_device_input_x2 (mif->sw_if_index,
+						    &next0, &next1,
+						    first_b0, first_b1);
+	    }
+
 	  /* trace */
 	  VLIB_BUFFER_TRACE_TRAJECTORY_INIT (first_b0);
 	  VLIB_BUFFER_TRACE_TRAJECTORY_INIT (first_b1);
@@ -343,11 +385,6 @@ memif_device_input_inline (vlib_main_t * vm, vlib_node_runtime_t * node,
 		}
 	    }
 
-	  /* redirect if feature path enabled */
-	  vnet_feature_start_device_input_x2 (mif->sw_if_index,
-					      &next0, &next1, first_b0,
-					      first_b1);
-
 	  /* enqueue */
 	  vlib_validate_buffer_enqueue_x2 (vm, node, next_index, to_next,
 					   n_left_to_next, first_bi0,
@@ -368,6 +405,21 @@ memif_device_input_inline (vlib_main_t * vm, vlib_node_runtime_t * node,
 						     &first_bi0, &bi0,
 						     &num_slots);
 
+	  if (mode == MEMIF_INTERFACE_MODE_IP)
+	    {
+	      next0 = memif_next_from_ip_hdr (node, first_b0);
+	    }
+	  else if (mode == MEMIF_INTERFACE_MODE_ETHERNET)
+	    {
+	      if (PREDICT_FALSE (mif->per_interface_next_index != ~0))
+		next0 = mif->per_interface_next_index;
+	      else
+		/* redirect if feature path
+		 * enabled */
+		vnet_feature_start_device_input_x1 (mif->sw_if_index,
+						    &next0, first_b0);
+	    }
+
 	  /* trace */
 	  VLIB_BUFFER_TRACE_TRAJECTORY_INIT (first_b0);
 
@@ -391,10 +443,6 @@ memif_device_input_inline (vlib_main_t * vm, vlib_node_runtime_t * node,
 	  to_next += 1;
 	  n_left_to_next--;
 
-	  /* redirect if feature path enabled */
-	  vnet_feature_start_device_input_x1 (mif->sw_if_index, &next0,
-					      first_b0);
-
 	  /* enqueue */
 	  vlib_validate_buffer_enqueue_x1 (vm, node, next_index, to_next,
 					   n_left_to_next, first_bi0, next0);
@@ -433,11 +481,27 @@ memif_input_fn (vlib_main_t * vm, vlib_node_runtime_t * node,
 	(mif->flags & MEMIF_IF_FLAG_CONNECTED))
       {
 	if (mif->flags & MEMIF_IF_FLAG_IS_SLAVE)
-	  n_rx += memif_device_input_inline (vm, node, frame, mif,
-					     MEMIF_RING_M2S, dq->queue_id);
+	  {
+	    if (mif->mode == MEMIF_INTERFACE_MODE_IP)
+	      n_rx += memif_device_input_inline (vm, node, frame, mif,
+						 MEMIF_RING_M2S, dq->queue_id,
+						 MEMIF_INTERFACE_MODE_IP);
+	    else
+	      n_rx += memif_device_input_inline (vm, node, frame, mif,
+						 MEMIF_RING_M2S, dq->queue_id,
+						 MEMIF_INTERFACE_MODE_ETHERNET);
+	  }
 	else
-	  n_rx += memif_device_input_inline (vm, node, frame, mif,
-					     MEMIF_RING_S2M, dq->queue_id);
+	  {
+	    if (mif->mode == MEMIF_INTERFACE_MODE_IP)
+	      n_rx += memif_device_input_inline (vm, node, frame, mif,
+						 MEMIF_RING_S2M, dq->queue_id,
+						 MEMIF_INTERFACE_MODE_IP);
+	    else
+	      n_rx += memif_device_input_inline (vm, node, frame, mif,
+						 MEMIF_RING_S2M, dq->queue_id,
+						 MEMIF_INTERFACE_MODE_ETHERNET);
+	  }
       }
   }
 
diff --git a/src/plugins/memif/socket.c b/src/plugins/memif/socket.c
index d1a96db361e..a3c8d544b7d 100644
--- a/src/plugins/memif/socket.c
+++ b/src/plugins/memif/socket.c
@@ -132,6 +132,7 @@ memif_msg_enq_init (memif_if_t * mif)
   e->fd = -1;
   i->version = MEMIF_VERSION;
   i->id = mif->id;
+  i->mode = mif->mode;
   s = format (0, "VPP %s%c", VPP_BUILD_VER, 0);
   strncpy ((char *) i->name, (char *) s, sizeof (i->name));
   if (mif->secret)