virtio: fast TAP interfaces with vhost-net backend

Change-Id: Ided667356d5c6fb9648eb34685aabd6b16a598b7
Signed-off-by: Damjan Marion <damarion@cisco.com>
Signed-off-by: Steven Luong <sluong@cisco.com>

14 files changed, 2271 insertions, 2 deletions

diff --git a/src/vat/api_format.c b/src/vat/api_format.c
index cfdce0bec2e..2a56423265e 100644
--- a/src/vat/api_format.c
+++ b/src/vat/api_format.c
@@ -1694,6 +1694,74 @@ static void vl_api_tap_delete_reply_t_handler_json
   vam->result_ready = 1;
 }
 
+static void
+vl_api_tap_create_v2_reply_t_handler (vl_api_tap_create_v2_reply_t * mp)
+{
+  vat_main_t *vam = &vat_main;
+  i32 retval = ntohl (mp->retval);
+  if (vam->async_mode)
+    {
+      vam->async_errors += (retval < 0);
+    }
+  else
+    {
+      vam->retval = retval;
+      vam->sw_if_index = ntohl (mp->sw_if_index);
+      vam->result_ready = 1;
+    }
+
+}
+
+static void vl_api_tap_create_v2_reply_t_handler_json
+  (vl_api_tap_create_v2_reply_t * mp)
+{
+  vat_main_t *vam = &vat_main;
+  vat_json_node_t node;
+
+  vat_json_init_object (&node);
+  vat_json_object_add_int (&node, "retval", ntohl (mp->retval));
+  vat_json_object_add_uint (&node, "sw_if_index", ntohl (mp->sw_if_index));
+
+  vat_json_print (vam->ofp, &node);
+  vat_json_free (&node);
+
+  vam->retval = ntohl (mp->retval);
+  vam->result_ready = 1;
+
+}
+
+static void
+vl_api_tap_delete_v2_reply_t_handler (vl_api_tap_delete_v2_reply_t * mp)
+{
+  vat_main_t *vam = &vat_main;
+  i32 retval = ntohl (mp->retval);
+  if (vam->async_mode)
+    {
+      vam->async_errors += (retval < 0);
+    }
+  else
+    {
+      vam->retval = retval;
+      vam->result_ready = 1;
+    }
+}
+
+static void vl_api_tap_delete_v2_reply_t_handler_json
+  (vl_api_tap_delete_v2_reply_t * mp)
+{
+  vat_main_t *vam = &vat_main;
+  vat_json_node_t node;
+
+  vat_json_init_object (&node);
+  vat_json_object_add_int (&node, "retval", ntohl (mp->retval));
+
+  vat_json_print (vam->ofp, &node);
+  vat_json_free (&node);
+
+  vam->retval = ntohl (mp->retval);
+  vam->result_ready = 1;
+}
+
 static void vl_api_mpls_tunnel_add_del_reply_t_handler
   (vl_api_mpls_tunnel_add_del_reply_t * mp)
 {
@@ -5378,6 +5446,9 @@ _(TAP_CONNECT_REPLY, tap_connect_reply)					\
 _(TAP_MODIFY_REPLY, tap_modify_reply)					\
 _(TAP_DELETE_REPLY, tap_delete_reply)					\
 _(SW_INTERFACE_TAP_DETAILS, sw_interface_tap_details)                   \
+_(TAP_CREATE_V2_REPLY, tap_create_v2_reply)				\
+_(TAP_DELETE_V2_REPLY, tap_delete_v2_reply)				\
+_(SW_INTERFACE_TAP_V2_DETAILS, sw_interface_tap_v2_details)             \
 _(IP_ADD_DEL_ROUTE_REPLY, ip_add_del_route_reply)			\
 _(IP_TABLE_ADD_DEL_REPLY, ip_table_add_del_reply)			\
 _(IP_MROUTE_ADD_DEL_REPLY, ip_mroute_add_del_reply)			\
@@ -7733,6 +7804,142 @@ api_tap_delete (vat_main_t * vam)
 }
 
 static int
+api_tap_create_v2 (vat_main_t * vam)
+{
+  unformat_input_t *i = vam->input;
+  vl_api_tap_create_v2_t *mp;
+  u8 mac_address[6];
+  u8 random_mac = 1;
+  u8 name_set = 0;
+  u8 *tap_name;
+  u8 *net_ns;
+  u8 net_ns_set = 0;
+  int ret;
+  int rx_ring_sz = 0, tx_ring_sz = 0;
+
+  memset (mac_address, 0, sizeof (mac_address));
+
+  /* Parse args required to build the message */
+  while (unformat_check_input (i) != UNFORMAT_END_OF_INPUT)
+    {
+      if (unformat (i, "hw-addr %U", unformat_ethernet_address, mac_address))
+	{
+	  random_mac = 0;
+	}
+      else if (unformat (i, "name %s", &tap_name))
+	name_set = 1;
+      else if (unformat (i, "host-ns %s", &net_ns))
+	net_ns_set = 1;
+      else if (unformat (i, "rx-ring-size %d", &rx_ring_sz))
+	;
+      else if (unformat (i, "tx-ring-size %d", &tx_ring_sz))
+	;
+      else
+	break;
+    }
+
+  if (name_set == 0)
+    {
+      errmsg ("missing tap name. ");
+      return -99;
+    }
+  if (vec_len (tap_name) > 63)
+    {
+      errmsg ("tap name too long. ");
+      return -99;
+    }
+  if (vec_len (net_ns) > 63)
+    {
+      errmsg ("host name space too long. ");
+      return -99;
+    }
+  if (!is_pow2 (rx_ring_sz))
+    {
+      errmsg ("rx ring size must be power of 2. ");
+      return -99;
+    }
+  if (rx_ring_sz > 32768)
+    {
+      errmsg ("rx ring size must be 32768 or lower. ");
+      return -99;
+    }
+  if (!is_pow2 (tx_ring_sz))
+    {
+      errmsg ("tx ring size must be power of 2. ");
+      return -99;
+    }
+  if (tx_ring_sz > 32768)
+    {
+      errmsg ("tx ring size must be 32768 or lower. ");
+      return -99;
+    }
+
+  vec_add1 (tap_name, 0);
+  vec_add1 (net_ns, 0);
+
+  /* Construct the API message */
+  M (TAP_CREATE_V2, mp);
+
+  mp->use_random_mac = random_mac;
+  clib_memcpy (mp->mac_address, mac_address, 6);
+  clib_memcpy (mp->tap_name, tap_name, vec_len (tap_name));
+  mp->net_ns_set = net_ns_set;
+  mp->rx_ring_sz = rx_ring_sz;
+  mp->tx_ring_sz = tx_ring_sz;
+  if (net_ns)
+    clib_memcpy (mp->net_ns, net_ns, vec_len (net_ns));
+
+  vec_free (tap_name);
+  vec_free (net_ns);
+
+  /* send it... */
+  S (mp);
+
+  /* Wait for a reply... */
+  W (ret);
+  return ret;
+}
+
+static int
+api_tap_delete_v2 (vat_main_t * vam)
+{
+  unformat_input_t *i = vam->input;
+  vl_api_tap_delete_v2_t *mp;
+  u32 sw_if_index = ~0;
+  u8 sw_if_index_set = 0;
+  int ret;
+
+  /* Parse args required to build the message */
+  while (unformat_check_input (i) != UNFORMAT_END_OF_INPUT)
+    {
+      if (unformat (i, "%U", api_unformat_sw_if_index, vam, &sw_if_index))
+	sw_if_index_set = 1;
+      else if (unformat (i, "sw_if_index %d", &sw_if_index))
+	sw_if_index_set = 1;
+      else
+	break;
+    }
+
+  if (sw_if_index_set == 0)
+    {
+      errmsg ("missing vpp interface name. ");
+      return -99;
+    }
+
+  /* Construct the API message */
+  M (TAP_DELETE_V2, mp);
+
+  mp->sw_if_index = ntohl (sw_if_index);
+
+  /* send it... */
+  S (mp);
+
+  /* Wait for a reply... */
+  W (ret);
+  return ret;
+}
+
+static int
 api_ip_table_add_del (vat_main_t * vam)
 {
   unformat_input_t *i = vam->input;
@@ -12199,6 +12406,53 @@ api_sw_interface_tap_dump (vat_main_t * vam)
   return ret;
 }
 
+static void vl_api_sw_interface_tap_v2_details_t_handler
+  (vl_api_sw_interface_tap_v2_details_t * mp)
+{
+  vat_main_t *vam = &vat_main;
+
+  print (vam->ofp, "%-16s %d",
+	 mp->dev_name, clib_net_to_host_u32 (mp->sw_if_index));
+}
+
+static void vl_api_sw_interface_tap_v2_details_t_handler_json
+  (vl_api_sw_interface_tap_v2_details_t * mp)
+{
+  vat_main_t *vam = &vat_main;
+  vat_json_node_t *node = NULL;
+
+  if (VAT_JSON_ARRAY != vam->json_tree.type)
+    {
+      ASSERT (VAT_JSON_NONE == vam->json_tree.type);
+      vat_json_init_array (&vam->json_tree);
+    }
+  node = vat_json_array_add (&vam->json_tree);
+
+  vat_json_init_object (node);
+  vat_json_object_add_uint (node, "sw_if_index", ntohl (mp->sw_if_index));
+  vat_json_object_add_string_copy (node, "dev_name", mp->dev_name);
+}
+
+static int
+api_sw_interface_tap_v2_dump (vat_main_t * vam)
+{
+  vl_api_sw_interface_tap_v2_dump_t *mp;
+  vl_api_control_ping_t *mp_ping;
+  int ret;
+
+  print (vam->ofp, "\n%-16s %s", "dev_name", "sw_if_index");
+  /* Get list of tap interfaces */
+  M (SW_INTERFACE_TAP_V2_DUMP, mp);
+  S (mp);
+
+  /* Use a control ping for synchronization */
+  MPING (CONTROL_PING, mp_ping);
+  S (mp_ping);
+
+  W (ret);
+  return ret;
+}
+
 static uword unformat_vxlan_decap_next
   (unformat_input_t * input, va_list * args)
 {
@@ -22375,6 +22629,11 @@ _(tap_modify,                                                           \
 _(tap_delete,                                                           \
   "<vpp-if-name> | sw_if_index <id>")                                   \
 _(sw_interface_tap_dump, "")                                            \
+_(tap_create_v2,                                                        \
+  "name <name> [hw-addr <mac-addr>] [host-ns <name>] [rx-ring-size <num> [tx-ring-size <num>]") \
+_(tap_delete_v2,                                                        \
+  "<vpp-if-name> | sw_if_index <id>")                                   \
+_(sw_interface_tap_v2_dump, "")                                         \
 _(ip_table_add_del,                                                     \
   "table-id <n> [ipv6]\n")                                              \
 _(ip_add_del_route,                                                     \
diff --git a/src/vnet.am b/src/vnet.am
index 72e67dcb298..00a6b231c7b 100644
--- a/src/vnet.am
+++ b/src/vnet.am
@@ -868,13 +868,23 @@ API_FILES += vnet/pg/pg.api
 ########################################
 
 libvnet_la_SOURCES +=       \
+  vnet/devices/virtio/cli.c		\
+  vnet/devices/virtio/device.c		\
+  vnet/devices/virtio/node.c		\
+  vnet/devices/virtio/tap.c		\
+  vnet/devices/virtio/tapv2_api.c  	\
   vnet/devices/virtio/vhost-user.c	\
-  vnet/devices/virtio/vhost_user_api.c
+  vnet/devices/virtio/vhost_user_api.c	\
+  vnet/devices/virtio/virtio.c
+
 
 nobase_include_HEADERS +=     \
+  vnet/devices/virtio/virtio.h		\
+  vnet/devices/virtio/tapv2.api.h \
   vnet/devices/virtio/vhost-user.h	\
   vnet/devices/virtio/vhost_user.api.h
 
+API_FILES += vnet/devices/virtio/tapv2.api
 API_FILES += vnet/devices/virtio/vhost_user.api
 
 ########################################
diff --git a/src/vnet/api_errno.h b/src/vnet/api_errno.h
index e4ba8ddfb95..8c0cc0aba9b 100644
--- a/src/vnet/api_errno.h
+++ b/src/vnet/api_errno.h
@@ -132,7 +132,9 @@ _(NAME_SERVER_NO_ADDRESSES, -139, "No addresses available")		\
 _(NAME_SERVER_NEXT_SERVER, -140, "Retry with new server")		\
 _(APP_CONNECT_FILTERED, -141, "Connect was filtered")			\
 _(ACL_IN_USE_INBOUND, -142, "Inbound ACL in use")			\
-_(ACL_IN_USE_OUTBOUND, -143, "Outbound ACL in use")
+_(ACL_IN_USE_OUTBOUND, -143, "Outbound ACL in use")                     \
+_(NAMESPACE_CREATE, -144, "Failed to create netlink namespace")         \
+_(VIRTIO_INIT, -145, "Failed to init virtio ring")
 
 typedef enum
 {
diff --git a/src/vnet/devices/virtio/cli.c b/src/vnet/devices/virtio/cli.c
new file mode 100644
index 00000000000..0c1b75f8c4e
--- /dev/null
+++ b/src/vnet/devices/virtio/cli.c
@@ -0,0 +1,300 @@
+/*
+ *------------------------------------------------------------------
+ * Copyright (c) 2016 Cisco and/or its affiliates.
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at:
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ *------------------------------------------------------------------
+ */
+#include <stdint.h>
+#include <net/if.h>
+#include <sys/ioctl.h>
+#include <inttypes.h>
+
+#include <vlib/vlib.h>
+#include <vlib/unix/unix.h>
+#include <vnet/ethernet/ethernet.h>
+#include <linux/virtio_net.h>
+#include <linux/vhost.h>
+#include <vnet/devices/virtio/virtio.h>
+#include <vnet/devices/virtio/tap.h>
+
+static clib_error_t *
+tap_create_command_fn (vlib_main_t * vm, unformat_input_t * input,
+		       vlib_cli_command_t * cmd)
+{
+  unformat_input_t _line_input, *line_input = &_line_input;
+  int rv;
+  tap_create_if_args_t args = { 0 };
+
+  /* Get a line of input. */
+  if (!unformat_user (input, unformat_line_input, line_input))
+    return clib_error_return (0, "Missing name <interface>");
+
+  while (unformat_check_input (line_input) != UNFORMAT_END_OF_INPUT)
+    {
+      if (unformat (line_input, "name %s", &args.name))
+	;
+      else if (unformat (line_input, "host-ns %s", &args.net_ns))
+	;
+      else if (unformat (line_input, "rx-ring-size %d", &args.rx_ring_sz))
+	;
+      else if (unformat (line_input, "tx-ring-size %d", &args.tx_ring_sz))
+	;
+      else if (unformat (line_input, "hw-addr %U",
+			 unformat_ethernet_address, args.hw_addr))
+	args.hw_addr_set = 1;
+      else
+	return clib_error_return (0, "unknown input `%U'",
+				  format_unformat_error, input);
+    }
+  unformat_free (line_input);
+
+  rv = tap_create_if (vm, &args);
+
+  vec_free (args.name);
+
+  if (rv == VNET_API_ERROR_SYSCALL_ERROR_1)
+    return clib_error_return_unix (0, "open '/dev/vhost-net'");
+  else if (rv == VNET_API_ERROR_SYSCALL_ERROR_2)
+    return clib_error_return_unix (0, "open '/dev/net/tun'");
+  else if (rv == VNET_API_ERROR_UNSUPPORTED)
+    return clib_error_return (0, "vhost-net backend doesn't support needed"
+			      " features");
+  else if (rv == VNET_API_ERROR_NAMESPACE_CREATE)
+    return clib_error_return (0, "failed to create netlink namespace");
+  else if (rv == VNET_API_ERROR_VIRTIO_INIT)
+    return clib_error_return (0, "failed to init virtio ring");
+  else if (rv == VNET_API_ERROR_INVALID_REGISTRATION)
+    return clib_error_return (0, "failed to register interface");
+  else if (rv != 0)
+    return clib_error_return (0, "error on creating tap interface");
+
+  return 0;
+}
+
+/* *INDENT-OFF* */
+VLIB_CLI_COMMAND (tap_create_command, static) = {
+  .path = "create tap",
+  .short_help = "create tap {name <if-name>} [hw-addr <mac-address>]"
+    "[rx-ring-size <size>] [tx-ring-size <size>] [host-ns <netns>]",
+  .function = tap_create_command_fn,
+};
+/* *INDENT-ON* */
+
+static clib_error_t *
+tap_delete_command_fn (vlib_main_t * vm, unformat_input_t * input,
+		       vlib_cli_command_t * cmd)
+{
+  unformat_input_t _line_input, *line_input = &_line_input;
+  u32 sw_if_index = ~0;
+  vnet_main_t *vnm = vnet_get_main ();
+  int rv;
+
+  /* Get a line of input. */
+  if (!unformat_user (input, unformat_line_input, line_input))
+    return clib_error_return (0, "Missing <interface>");
+
+  while (unformat_check_input (line_input) != UNFORMAT_END_OF_INPUT)
+    {
+      if (unformat (line_input, "sw_if_index %d", &sw_if_index))
+	;
+      else if (unformat (line_input, "%U", unformat_vnet_sw_interface,
+			 vnm, &sw_if_index))
+	;
+      else
+	return clib_error_return (0, "unknown input `%U'",
+				  format_unformat_error, input);
+    }
+  unformat_free (line_input);
+
+  if (sw_if_index == ~0)
+    return clib_error_return (0,
+			      "please specify interface name or sw_if_index");
+
+  rv = tap_delete_if (vm, sw_if_index);
+  if (rv == VNET_API_ERROR_INVALID_SW_IF_INDEX)
+    return clib_error_return (0, "not a tap interface");
+  else if (rv != 0)
+    return clib_error_return (0, "error on deleting tap interface");
+
+  return 0;
+}
+
+/* *INDENT-OFF* */
+VLIB_CLI_COMMAND (tap_delete__command, static) =
+{
+  .path = "delete tap",
+  .short_help = "delete tap {<interface> | sw_if_index <sw_idx>}",
+  .function = tap_delete_command_fn,
+};
+/* *INDENT-ON* */
+
+static clib_error_t *
+tap_show_command_fn (vlib_main_t * vm, unformat_input_t * input,
+		     vlib_cli_command_t * cmd)
+{
+  virtio_main_t *mm = &virtio_main;
+  virtio_if_t *vif;
+  vnet_main_t *vnm = vnet_get_main ();
+  int show_descr = 0;
+  clib_error_t *error = 0;
+  u32 hw_if_index, *hw_if_indices = 0;
+  virtio_vring_t *vring;
+  int i, j;
+  struct feat_struct
+  {
+    u8 bit;
+    char *str;
+  };
+  struct feat_struct *feat_entry;
+
+  static struct feat_struct feat_array[] = {
+#define _(s,b) { .str = #s, .bit = b, },
+    foreach_virtio_net_features
+#undef _
+    {.str = NULL}
+  };
+
+  struct feat_struct *flag_entry;
+  static struct feat_struct flags_array[] = {
+#define _(b,e,s) { .bit = b, .str = s, },
+    foreach_virtio_if_flag
+#undef _
+    {.str = NULL}
+  };
+
+  while (unformat_check_input (input) != UNFORMAT_END_OF_INPUT)
+    {
+      if (unformat
+	  (input, "%U", unformat_vnet_hw_interface, vnm, &hw_if_index))
+	vec_add1 (hw_if_indices, hw_if_index);
+      else if (unformat (input, "descriptors"))
+	show_descr = 1;
+      else
+	{
+	  error = clib_error_return (0, "unknown input `%U'",
+				     format_unformat_error, input);
+	  goto done;
+	}
+    }
+
+  if (vec_len (hw_if_indices) == 0)
+    {
+      /* *INDENT-OFF* */
+      pool_foreach (vif, mm->interfaces,
+	  vec_add1 (hw_if_indices, vif->hw_if_index);
+      );
+      /* *INDENT-ON* */
+    }
+
+  for (hw_if_index = 0; hw_if_index < vec_len (hw_if_indices); hw_if_index++)
+    {
+      vnet_hw_interface_t *hi =
+	vnet_get_hw_interface (vnm, hw_if_indices[hw_if_index]);
+      vif = pool_elt_at_index (mm->interfaces, hi->dev_instance);
+      vlib_cli_output (vm, "interface %U", format_vnet_sw_if_index_name,
+		       vnm, vif->sw_if_index);
+      if (vif->name)
+	vlib_cli_output (vm, "  name \"%s\"", vif->name);
+      if (vif->net_ns)
+	vlib_cli_output (vm, "  host-ns \"%s\"", vif->net_ns);
+      vlib_cli_output (vm, "  flags 0x%x", vif->flags);
+      flag_entry = (struct feat_struct *) &flags_array;
+      while (flag_entry->str)
+	{
+	  if (vif->flags & (1ULL << flag_entry->bit))
+	    vlib_cli_output (vm, "    %s (%d)", flag_entry->str,
+			     flag_entry->bit);
+	  flag_entry++;
+	}
+      vlib_cli_output (vm, "  fd %d", vif->fd);
+      vlib_cli_output (vm, "  tap-fd %d", vif->tap_fd);
+      vlib_cli_output (vm, "  features 0x%lx", vif->features);
+      feat_entry = (struct feat_struct *) &feat_array;
+      while (feat_entry->str)
+	{
+	  if (vif->features & (1ULL << feat_entry->bit))
+	    vlib_cli_output (vm, "    %s (%d)", feat_entry->str,
+			     feat_entry->bit);
+	  feat_entry++;
+	}
+      vlib_cli_output (vm, "  remote-features 0x%lx", vif->remote_features);
+      feat_entry = (struct feat_struct *) &feat_array;
+      while (feat_entry->str)
+	{
+	  if (vif->remote_features & (1ULL << feat_entry->bit))
+	    vlib_cli_output (vm, "    %s (%d)", feat_entry->str,
+			     feat_entry->bit);
+	  feat_entry++;
+	}
+      vec_foreach_index (i, vif->vrings)
+      {
+	// RX = 0, TX = 1
+	vring = vec_elt_at_index (vif->vrings, i);
+	vlib_cli_output (vm, "  Virtqueue (%s)", (i & 1) ? "TX" : "RX");
+	vlib_cli_output (vm, "    qsz %d, last_used_idx %d, desc_in_use %d",
+			 vring->size, vring->last_used_idx,
+			 vring->desc_in_use);
+	vlib_cli_output (vm,
+			 "    avail.flags 0x%x avail.idx %d used.flags 0x%x used.idx %d",
+			 vring->avail->flags, vring->avail->idx,
+			 vring->used->flags, vring->used->idx);
+	vlib_cli_output (vm, "    kickfd %d, callfd %d", vring->kick_fd,
+			 vring->call_fd);
+	if (show_descr)
+	  {
+	    vlib_cli_output (vm, "\n  descriptor table:\n");
+	    vlib_cli_output (vm,
+			     "   id          addr         len  flags  next      user_addr\n");
+	    vlib_cli_output (vm,
+			     "  ===== ================== ===== ====== ===== ==================\n");
+	    vring = vif->vrings;
+	    for (j = 0; j < vring->size; j++)
+	      {
+		struct vring_desc *desc = &vring->desc[j];
+		vlib_cli_output (vm,
+				 "  %-5d 0x%016lx %-5d 0x%04x %-5d 0x%016lx\n",
+				 j, desc->addr,
+				 desc->len,
+				 desc->flags, desc->next, desc->addr);
+	      }
+	  }
+      }
+    }
+done:
+  vec_free (hw_if_indices);
+  return error;
+}
+
+/* *INDENT-OFF* */
+VLIB_CLI_COMMAND (tap_show_command, static) = {
+  .path = "show tap",
+  .short_help = "show tap {<interface>] [descriptors]",
+  .function = tap_show_command_fn,
+};
+/* *INDENT-ON* */
+
+clib_error_t *
+tap_cli_init (vlib_main_t * vm)
+{
+  return 0;
+}
+
+VLIB_INIT_FUNCTION (tap_cli_init);
+
+/*
+ * fd.io coding-style-patch-verification: ON
+ *
+ * Local Variables:
+ * eval: (c-set-style "gnu")
+ * End:
+ */
diff --git a/src/vnet/devices/virtio/device.c b/src/vnet/devices/virtio/device.c
new file mode 100644
index 00000000000..275a3c74990
--- /dev/null
+++ b/src/vnet/devices/virtio/device.c
@@ -0,0 +1,328 @@
+/*
+ *------------------------------------------------------------------
+ * Copyright (c) 2016 Cisco and/or its affiliates.
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at:
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ *------------------------------------------------------------------
+ */
+
+#include <sys/types.h>
+#include <sys/stat.h>
+#include <fcntl.h>
+#include <linux/virtio_net.h>
+#include <linux/vhost.h>
+
+#include <vlib/vlib.h>
+#include <vlib/unix/unix.h>
+#include <vnet/ethernet/ethernet.h>
+#include <vnet/devices/virtio/virtio.h>
+
+#define foreach_virtio_tx_func_error	       \
+_(NO_FREE_SLOTS, "no free tx slots")           \
+_(TRUNC_PACKET, "packet > buffer size -- truncated in tx ring") \
+_(PENDING_MSGS, "pending msgs in tx ring") \
+_(NO_TX_QUEUES, "no tx queues")
+
+typedef enum
+{
+#define _(f,s) TAP_TX_ERROR_##f,
+  foreach_virtio_tx_func_error
+#undef _
+    TAP_TX_N_ERROR,
+} virtio_tx_func_error_t;
+
+static char *virtio_tx_func_error_strings[] = {
+#define _(n,s) s,
+  foreach_virtio_tx_func_error
+#undef _
+};
+
+u8 *
+format_virtio_device_name (u8 * s, va_list * args)
+{
+  u32 dev_instance = va_arg (*args, u32);
+  virtio_main_t *mm = &virtio_main;
+  virtio_if_t *vif = pool_elt_at_index (mm->interfaces, dev_instance);
+
+  if (vif->type == VIRTIO_IF_TYPE_TAP)
+    {
+      s = format (s, "tap-%s", vif->name);
+    }
+  else
+    s = format (s, "virtio%lu", vif->dev_instance);
+
+  return s;
+}
+
+static u8 *
+format_virtio_device (u8 * s, va_list * args)
+{
+  u32 dev_instance = va_arg (*args, u32);
+  int verbose = va_arg (*args, int);
+  u32 indent = format_get_indent (s);
+
+  s = format (s, "VIRTIO interface");
+  if (verbose)
+    {
+      s = format (s, "\n%U instance %u", format_white_space, indent + 2,
+		  dev_instance);
+    }
+  return s;
+}
+
+static u8 *
+format_virtio_tx_trace (u8 * s, va_list * args)
+{
+  s = format (s, "Unimplemented...");
+  return s;
+}
+
+static_always_inline void
+virtio_free_used_desc (vlib_main_t * vm, virtio_vring_t * vring)
+{
+  u16 used = vring->desc_in_use;
+  u16 sz = vring->size;
+  u16 mask = sz - 1;
+  u16 last = vring->last_used_idx;
+  u16 n_left = vring->used->idx - last;
+
+  if (n_left == 0)
+    return;
+
+  while (n_left)
+    {
+      struct vring_used_elem *e = &vring->used->ring[last & mask];
+      u16 slot = e->id;
+      struct vring_desc *d = &vring->desc[slot];
+
+      if (PREDICT_FALSE (d->flags & VRING_DESC_F_INDIRECT))
+	{
+	  d = uword_to_pointer (d->addr, struct vring_desc *);
+	  vec_free (d);
+	}
+
+      vlib_buffer_free (vm, &vring->buffers[slot], 1);
+      used--;
+      last++;
+      n_left--;
+    }
+  vring->desc_in_use = used;
+  vring->last_used_idx = last;
+}
+
+static_always_inline u16
+add_buffer_to_slot (vlib_main_t * vm, virtio_vring_t * vring, u32 bi,
+		    u16 avail, u16 next, u16 mask)
+{
+  u16 n_added = 0;
+  const int hdr_sz = sizeof (struct virtio_net_hdr_v1);
+  struct vring_desc *d;
+  d = &vring->desc[next];
+  vlib_buffer_t *b = vlib_get_buffer (vm, bi);
+
+  if (PREDICT_TRUE ((b->flags & VLIB_BUFFER_NEXT_PRESENT) == 0))
+    {
+      d->addr = pointer_to_uword (vlib_buffer_get_current (b)) - hdr_sz;
+      d->len = b->current_length + hdr_sz;
+      d->flags = 0;
+    }
+  else
+    {
+      struct vring_desc *id, *descs = 0;
+
+      /* first buffer in chain */
+      vec_add2_aligned (descs, id, 1, CLIB_CACHE_LINE_BYTES);
+      id->addr = pointer_to_uword (vlib_buffer_get_current (b)) - hdr_sz;
+      id->len = b->current_length + hdr_sz;
+
+      while (b->flags & VLIB_BUFFER_NEXT_PRESENT)
+	{
+	  id->flags = VRING_DESC_F_NEXT;
+	  id->next = vec_len (descs);
+	  vec_add2_aligned (descs, id, 1, CLIB_CACHE_LINE_BYTES);
+	  b = vlib_get_buffer (vm, b->next_buffer);
+	  id->addr = pointer_to_uword (vlib_buffer_get_current (b));
+	  id->len = b->current_length;
+	}
+
+      d->addr = pointer_to_uword (descs);
+      d->len = vec_len (descs) * sizeof (struct vring_desc);
+      d->flags = VRING_DESC_F_INDIRECT;
+    }
+  vring->buffers[next] = bi;
+  vring->avail->ring[avail & mask] = next;
+  n_added++;
+  return n_added;
+}
+
+
+static_always_inline uword
+virtio_interface_tx_inline (vlib_main_t * vm, vlib_node_runtime_t * node,
+			    vlib_frame_t * frame, virtio_if_t * vif)
+{
+  u8 qid = 0;
+  u16 n_left = frame->n_vectors;
+  virtio_vring_t *vring = vec_elt_at_index (vif->vrings, (qid << 1) + 1);
+  u16 used, next, avail;
+  u16 sz = vring->size;
+  u16 mask = sz - 1;
+  u32 *buffers = vlib_frame_args (frame);
+
+  /* free consumed buffers */
+  virtio_free_used_desc (vm, vring);
+
+  used = vring->desc_in_use;
+  next = vring->desc_next;
+  avail = vring->avail->idx;
+
+  while (n_left && used < sz)
+    {
+      u16 n_added;
+      n_added = add_buffer_to_slot (vm, vring, buffers[0], avail, next, mask);
+      avail += n_added;
+      next = (next + n_added) & mask;
+      used += n_added;
+      buffers++;
+      n_left--;
+    }
+
+  if (n_left != frame->n_vectors)
+    {
+      CLIB_MEMORY_STORE_BARRIER ();
+      vring->avail->idx = avail;
+      vring->desc_next = next;
+      vring->desc_in_use = used;
+      if ((vring->used->flags & VIRTIO_RING_FLAG_MASK_INT) == 0)
+	{
+	  u64 x = 1;
+	  CLIB_UNUSED (int r) = write (vring->kick_fd, &x, sizeof (x));
+	}
+    }
+
+
+  if (n_left)
+    {
+      vlib_error_count (vm, node->node_index, TAP_TX_ERROR_NO_FREE_SLOTS,
+			n_left);
+      vlib_buffer_free (vm, buffers, n_left);
+    }
+
+  return frame->n_vectors - n_left;
+}
+
+static uword
+virtio_interface_tx (vlib_main_t * vm,
+		     vlib_node_runtime_t * node, vlib_frame_t * frame)
+{
+  virtio_main_t *nm = &virtio_main;
+  vnet_interface_output_runtime_t *rund = (void *) node->runtime_data;
+  virtio_if_t *vif = pool_elt_at_index (nm->interfaces, rund->dev_instance);
+
+  return virtio_interface_tx_inline (vm, node, frame, vif);
+}
+
+static void
+virtio_set_interface_next_node (vnet_main_t * vnm, u32 hw_if_index,
+				u32 node_index)
+{
+  virtio_main_t *apm = &virtio_main;
+  vnet_hw_interface_t *hw = vnet_get_hw_interface (vnm, hw_if_index);
+  virtio_if_t *vif = pool_elt_at_index (apm->interfaces, hw->dev_instance);
+
+  /* Shut off redirection */
+  if (node_index == ~0)
+    {
+      vif->per_interface_next_index = node_index;
+      return;
+    }
+
+  vif->per_interface_next_index =
+    vlib_node_add_next (vlib_get_main (), virtio_input_node.index,
+			node_index);
+}
+
+static void
+virtio_clear_hw_interface_counters (u32 instance)
+{
+  /* Nothing for now */
+}
+
+static clib_error_t *
+virtio_interface_rx_mode_change (vnet_main_t * vnm, u32 hw_if_index, u32 qid,
+				 vnet_hw_interface_rx_mode mode)
+{
+  virtio_main_t *mm = &virtio_main;
+  vnet_hw_interface_t *hw = vnet_get_hw_interface (vnm, hw_if_index);
+  virtio_if_t *vif = pool_elt_at_index (mm->interfaces, hw->dev_instance);
+  virtio_vring_t *vring = vec_elt_at_index (vif->vrings, qid);
+
+  if (mode == VNET_HW_INTERFACE_RX_MODE_POLLING)
+    vring->avail->flags |= VIRTIO_RING_FLAG_MASK_INT;
+  else
+    vring->avail->flags &= ~VIRTIO_RING_FLAG_MASK_INT;
+
+  return 0;
+}
+
+static clib_error_t *
+virtio_interface_admin_up_down (vnet_main_t * vnm, u32 hw_if_index, u32 flags)
+{
+  virtio_main_t *mm = &virtio_main;
+  vnet_hw_interface_t *hw = vnet_get_hw_interface (vnm, hw_if_index);
+  virtio_if_t *vif = pool_elt_at_index (mm->interfaces, hw->dev_instance);
+  static clib_error_t *error = 0;
+
+  if (flags & VNET_SW_INTERFACE_FLAG_ADMIN_UP)
+    vif->flags |= VIRTIO_IF_FLAG_ADMIN_UP;
+  else
+    vif->flags &= ~VIRTIO_IF_FLAG_ADMIN_UP;
+
+  return error;
+  return 0;
+}
+
+static clib_error_t *
+virtio_subif_add_del_function (vnet_main_t * vnm,
+			       u32 hw_if_index,
+			       struct vnet_sw_interface_t *st, int is_add)
+{
+  /* Nothing for now */
+  return 0;
+}
+
+/* *INDENT-OFF* */
+VNET_DEVICE_CLASS (virtio_device_class) = {
+  .name = "virtio",
+  .tx_function = virtio_interface_tx,
+  .format_device_name = format_virtio_device_name,
+  .format_device = format_virtio_device,
+  .format_tx_trace = format_virtio_tx_trace,
+  .tx_function_n_errors = TAP_TX_N_ERROR,
+  .tx_function_error_strings = virtio_tx_func_error_strings,
+  .rx_redirect_to_node = virtio_set_interface_next_node,
+  .clear_counters = virtio_clear_hw_interface_counters,
+  .admin_up_down_function = virtio_interface_admin_up_down,
+  .subif_add_del_function = virtio_subif_add_del_function,
+  .rx_mode_change_function = virtio_interface_rx_mode_change,
+};
+
+VLIB_DEVICE_TX_FUNCTION_MULTIARCH(virtio_device_class,
+				  virtio_interface_tx)
+/* *INDENT-ON* */
+
+/*
+ * fd.io coding-style-patch-verification: ON
+ *
+ * Local Variables:
+ * eval: (c-set-style "gnu")
+ * End:
+ */
diff --git a/src/vnet/devices/virtio/node.c b/src/vnet/devices/virtio/node.c
new file mode 100644
index 00000000000..f746ada7326
--- /dev/null
+++ b/src/vnet/devices/virtio/node.c
@@ -0,0 +1,302 @@
+/*
+ *------------------------------------------------------------------
+ * Copyright (c) 2016 Cisco and/or its affiliates.
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at:
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ *------------------------------------------------------------------
+ */
+
+#include <sys/types.h>
+#include <sys/stat.h>
+#include <fcntl.h>
+#include <net/if.h>
+#include <linux/if_tun.h>
+#include <sys/ioctl.h>
+#include <linux/virtio_net.h>
+#include <linux/vhost.h>
+#include <sys/eventfd.h>
+
+#include <vlib/vlib.h>
+#include <vlib/unix/unix.h>
+#include <vnet/ethernet/ethernet.h>
+#include <vnet/devices/devices.h>
+#include <vnet/feature/feature.h>
+#include <vnet/devices/virtio/virtio.h>
+
+
+#define foreach_virtio_input_error \
+  _(UNKNOWN, "unknown")
+
+typedef enum
+{
+#define _(f,s) TAP_INPUT_ERROR_##f,
+  foreach_virtio_input_error
+#undef _
+    TAP_INPUT_N_ERROR,
+} virtio_input_error_t;
+
+static char *virtio_input_error_strings[] = {
+#define _(n,s) s,
+  foreach_virtio_input_error
+#undef _
+};
+
+typedef struct
+{
+  u32 next_index;
+  u32 hw_if_index;
+  u16 ring;
+  u16 len;
+  struct virtio_net_hdr_v1 hdr;
+} virtio_input_trace_t;
+
+static u8 *
+format_virtio_input_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 *);
+  virtio_input_trace_t *t = va_arg (*args, virtio_input_trace_t *);
+  u32 indent = format_get_indent (s);
+
+  s = format (s, "virtio: hw_if_index %d next-index %d vring %u len %u",
+	      t->hw_if_index, t->next_index, t->ring, t->len);
+  s = format (s, "\n%Uhdr: flags 0x%02x gso_type 0x%02x hdr_len %u "
+	      "gso_size %u csum_start %u csum_offset %u num_buffers %u",
+	      format_white_space, indent + 2,
+	      t->hdr.flags, t->hdr.gso_type, t->hdr.hdr_len, t->hdr.gso_size,
+	      t->hdr.csum_start, t->hdr.csum_offset, t->hdr.num_buffers);
+  return s;
+}
+
+static_always_inline void
+virtio_refill_vring (vlib_main_t * vm, virtio_vring_t * vring)
+{
+  const int hdr_sz = sizeof (struct virtio_net_hdr_v1);
+  u16 used, next, avail, n_slots, n_alloc;
+  u16 sz = vring->size;
+  u16 mask = sz - 1;
+  int i;
+
+  used = vring->desc_in_use;
+
+  if (sz - used < sz / 8)
+    return;
+
+  n_slots = sz - used;
+  next = vring->desc_next;
+  avail = vring->avail->idx;
+  n_alloc = vlib_buffer_alloc (vm, &vring->buffers[next], n_slots);
+
+  if (PREDICT_FALSE (n_alloc < n_slots))
+    n_slots = n_alloc;
+
+  i = next + n_slots - sz;
+  if (PREDICT_FALSE (i > 0))
+    clib_memcpy (vring->buffers, &vring->buffers[sz], i * sizeof (u32));
+
+  while (n_slots)
+    {
+      struct vring_desc *d = &vring->desc[next];;
+      vlib_buffer_t *b = vlib_get_buffer (vm, vring->buffers[next]);
+      d->addr = pointer_to_uword (vlib_buffer_get_current (b)) - hdr_sz;
+      d->len = VLIB_BUFFER_DATA_SIZE + hdr_sz;
+      d->flags = VRING_DESC_F_WRITE;
+      vring->avail->ring[avail & mask] = next;
+      avail++;
+      next = (next + 1) & mask;
+      n_slots--;
+      used++;
+    }
+  CLIB_MEMORY_STORE_BARRIER ();
+  vring->avail->idx = avail;
+  vring->desc_next = next;
+  vring->desc_in_use = used;
+
+  if ((vring->used->flags & VIRTIO_RING_FLAG_MASK_INT) == 0)
+    {
+      u64 b = 1;
+      CLIB_UNUSED (int r) = write (vring->kick_fd, &b, sizeof (b));
+    }
+}
+
+static_always_inline uword
+virtio_device_input_inline (vlib_main_t * vm, vlib_node_runtime_t * node,
+			    vlib_frame_t * frame, virtio_if_t * vif, u16 qid)
+{
+  vnet_main_t *vnm = vnet_get_main ();
+  u32 thread_index = vlib_get_thread_index ();
+  uword n_trace = vlib_get_trace_count (vm, node);
+  virtio_vring_t *vring = vec_elt_at_index (vif->vrings, 0);
+  u32 next_index = VNET_DEVICE_INPUT_NEXT_ETHERNET_INPUT;
+  const int hdr_sz = sizeof (struct virtio_net_hdr_v1);
+  u32 *to_next = 0;
+  u32 n_rx_packets = 0;
+  u32 n_rx_bytes = 0;
+  u16 mask = vring->size - 1;
+  u16 last = vring->last_used_idx;
+  u16 n_left = vring->used->idx - last;
+
+  if (n_left == 0)
+    goto refill;
+
+  while (n_left)
+    {
+      u32 n_left_to_next;
+      u32 next0 = next_index;
+      vlib_get_next_frame (vm, node, next_index, to_next, n_left_to_next);
+
+      while (n_left && n_left_to_next)
+	{
+	  u16 num_buffers;
+	  struct vring_used_elem *e = &vring->used->ring[last & mask];
+	  struct virtio_net_hdr_v1 *hdr;
+	  u16 slot = e->id;
+	  u16 len = e->len - hdr_sz;
+	  u32 bi0 = vring->buffers[slot];
+	  vlib_buffer_t *b0 = vlib_get_buffer (vm, bi0);
+	  hdr = vlib_buffer_get_current (b0) - hdr_sz;
+	  num_buffers = hdr->num_buffers;
+
+	  b0->current_data = 0;
+	  b0->current_length = len;
+	  b0->total_length_not_including_first_buffer = 0;
+	  b0->flags = VLIB_BUFFER_TOTAL_LENGTH_VALID;
+	  vnet_buffer (b0)->sw_if_index[VLIB_RX] = vif->sw_if_index;
+	  vnet_buffer (b0)->sw_if_index[VLIB_TX] = (u32) ~ 0;
+
+	  /* if multisegment packet */
+	  if (PREDICT_FALSE (num_buffers > 1))
+	    {
+	      vlib_buffer_t *pb, *cb;
+	      pb = b0;
+	      while (num_buffers > 1)
+		{
+		  last++;
+		  e = &vring->used->ring[last & mask];
+		  u32 cbi = vring->buffers[e->id];
+		  cb = vlib_get_buffer (vm, cbi);
+
+		  /* current buffer */
+		  cb->current_data = -hdr_sz;
+		  cb->current_length = e->len;
+
+		  /* previous buffer */
+		  pb->next_buffer = cbi;
+		  pb->flags |= VLIB_BUFFER_NEXT_PRESENT;
+
+		  /* first buffer */
+		  b0->total_length_not_including_first_buffer += e->len;
+
+		  pb = cb;
+		  vring->desc_in_use--;
+		  num_buffers--;
+		  n_left--;
+		}
+	    }
+
+	  if (PREDICT_FALSE (vif->per_interface_next_index != ~0))
+	    next0 = vif->per_interface_next_index;
+	  else
+	    /* redirect if feature path enabled */
+	    vnet_feature_start_device_input_x1 (vif->sw_if_index, &next0, b0);
+	  /* trace */
+	  VLIB_BUFFER_TRACE_TRAJECTORY_INIT (b0);
+
+	  if (PREDICT_FALSE (n_trace > 0))
+	    {
+	      virtio_input_trace_t *tr;
+	      vlib_trace_buffer (vm, node, next0, b0,
+				 /* follow_chain */ 0);
+	      vlib_set_trace_count (vm, node, --n_trace);
+	      tr = vlib_add_trace (vm, node, b0, sizeof (*tr));
+	      tr->next_index = next0;
+	      tr->hw_if_index = vif->hw_if_index;
+	      tr->len = len;
+	      clib_memcpy (&tr->hdr, hdr, hdr_sz);
+	    }
+
+	  /* enqueue buffer */
+	  to_next[0] = bi0;
+	  vring->desc_in_use--;
+	  to_next += 1;
+	  n_left_to_next--;
+	  n_left--;
+	  last++;
+
+	  /* enqueue */
+	  vlib_validate_buffer_enqueue_x1 (vm, node, next_index, to_next,
+					   n_left_to_next, bi0, next0);
+
+	  /* next packet */
+	  n_rx_packets++;
+	  n_rx_bytes += len;
+	}
+      vlib_put_next_frame (vm, node, next_index, n_left_to_next);
+    }
+  vring->last_used_idx = last;
+
+  vlib_increment_combined_counter (vnm->interface_main.combined_sw_if_counters
+				   + VNET_INTERFACE_COUNTER_RX, thread_index,
+				   vif->hw_if_index, n_rx_packets,
+				   n_rx_bytes);
+
+refill:
+  virtio_refill_vring (vm, vring);
+
+  return n_rx_packets;
+}
+
+static uword
+virtio_input_fn (vlib_main_t * vm, vlib_node_runtime_t * node,
+		 vlib_frame_t * frame)
+{
+  u32 n_rx = 0;
+  virtio_main_t *nm = &virtio_main;
+  vnet_device_input_runtime_t *rt = (void *) node->runtime_data;
+  vnet_device_and_queue_t *dq;
+
+  foreach_device_and_queue (dq, rt->devices_and_queues)
+  {
+    virtio_if_t *mif;
+    mif = vec_elt_at_index (nm->interfaces, dq->dev_instance);
+    if (mif->flags & VIRTIO_IF_FLAG_ADMIN_UP)
+      {
+	n_rx += virtio_device_input_inline (vm, node, frame, mif,
+					    dq->queue_id);
+      }
+  }
+
+  return n_rx;
+}
+
+/* *INDENT-OFF* */
+VLIB_REGISTER_NODE (virtio_input_node) = {
+  .function = virtio_input_fn,
+  .name = "virtio-input",
+  .sibling_of = "device-input",
+  .format_trace = format_virtio_input_trace,
+  .type = VLIB_NODE_TYPE_INPUT,
+  .state = VLIB_NODE_STATE_INTERRUPT,
+  .n_errors = TAP_INPUT_N_ERROR,
+  .error_strings = virtio_input_error_strings,
+};
+
+VLIB_NODE_FUNCTION_MULTIARCH (virtio_input_node, virtio_input_fn)
+/* *INDENT-ON* */
+
+/*
+ * fd.io coding-style-patch-verification: ON
+ *
+ * Local Variables:
+ * eval: (c-set-style "gnu")
+ * End:
+ */
diff --git a/src/vnet/devices/virtio/tap.c b/src/vnet/devices/virtio/tap.c
new file mode 100644
index 00000000000..658ba6bfc68
--- /dev/null
+++ b/src/vnet/devices/virtio/tap.c
@@ -0,0 +1,361 @@
+/*
+ *------------------------------------------------------------------
+ * Copyright (c) 2017 Cisco and/or its affiliates.
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at:
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ *------------------------------------------------------------------
+ */
+
+#include <sys/types.h>
+#include <sys/stat.h>
+#include <fcntl.h>
+#include <net/if.h>
+#include <linux/if_tun.h>
+#include <sys/ioctl.h>
+#include <linux/virtio_net.h>
+#include <linux/vhost.h>
+#include <sys/eventfd.h>
+
+#include <linux/netlink.h>
+#include <linux/rtnetlink.h>
+
+#include <vlib/vlib.h>
+#include <vlib/unix/unix.h>
+#include <vnet/ethernet/ethernet.h>
+#include <vnet/devices/virtio/virtio.h>
+#include <vnet/devices/virtio/tap.h>
+
+#define _IOCTL(fd,a,...) \
+  if (ioctl (fd, a, __VA_ARGS__) < 0) \
+    { \
+      err = clib_error_return_unix (0, "ioctl(" #a ")"); \
+      goto error; \
+    }
+
+static u32
+virtio_eth_flag_change (vnet_main_t * vnm, vnet_hw_interface_t * hi,
+			u32 flags)
+{
+  /* nothing for now */
+  return 0;
+}
+
+
+clib_error_t *
+clib_netlink_set_if_attr (int ifindex, unsigned short rta_type, void *data,
+			  int data_len)
+{
+  clib_error_t *err = 0;
+  int sock;
+  struct sockaddr_nl ra = { 0 };
+  struct
+  {
+    struct nlmsghdr nh;
+    struct ifinfomsg ifmsg;
+    char attrbuf[512];
+  } req;
+  struct rtattr *rta;
+
+  memset (&req, 0, sizeof (req));
+  if ((sock = socket (AF_NETLINK, SOCK_RAW, NETLINK_ROUTE)) == -1)
+    {
+      err = clib_error_return_unix (0, "socket(AF_NETLINK)");
+      goto error;
+    }
+
+  ra.nl_family = AF_NETLINK;
+  ra.nl_pid = getpid ();
+
+  if ((bind (sock, (struct sockaddr *) &ra, sizeof (ra))) == -1)
+    return clib_error_return_unix (0, "bind");
+
+  req.nh.nlmsg_len = NLMSG_LENGTH (sizeof (struct ifinfomsg));
+  req.nh.nlmsg_flags = NLM_F_REQUEST;
+  req.nh.nlmsg_type = RTM_SETLINK;
+  req.ifmsg.ifi_family = AF_UNSPEC;
+  req.ifmsg.ifi_index = ifindex;
+  req.ifmsg.ifi_change = 0xffffffff;
+  rta = (struct rtattr *) (((char *) &req) + NLMSG_ALIGN (req.nh.nlmsg_len));
+  rta->rta_type = rta_type;
+  rta->rta_len = RTA_LENGTH (data_len);
+  req.nh.nlmsg_len = NLMSG_ALIGN (req.nh.nlmsg_len) + RTA_LENGTH (data_len);
+  memcpy (RTA_DATA (rta), data, data_len);
+
+  if ((send (sock, &req, req.nh.nlmsg_len, 0)) == -1)
+    err = clib_error_return_unix (0, "send");
+
+error:
+  return err;
+}
+
+clib_error_t *
+clib_netlink_set_if_mtu (int ifindex, int mtu)
+{
+  clib_error_t *err;
+
+  err = clib_netlink_set_if_attr (ifindex, IFLA_MTU, &mtu, sizeof (int));
+  return err;
+}
+
+clib_error_t *
+clib_netlink_set_if_namespace (int ifindex, char *net_ns)
+{
+  clib_error_t *err;
+  int ns_fd;
+  u8 *s;
+  s = format (0, "/var/run/netns/%s%c", net_ns, 0);
+  ns_fd = open ((char *) s, O_RDONLY);
+  vec_free (s);
+  if (ns_fd == -1)
+    return clib_error_return (0, "namespace '%s' doesn't exist", net_ns);
+
+  err =
+    clib_netlink_set_if_attr (ifindex, IFLA_NET_NS_FD, &ns_fd, sizeof (int));
+  close (ns_fd);
+  return err;
+}
+
+int
+tap_create_if (vlib_main_t * vm, tap_create_if_args_t * args)
+{
+  vnet_main_t *vnm = vnet_get_main ();
+  virtio_main_t *vim = &virtio_main;
+  vnet_sw_interface_t *sw;
+  vnet_hw_interface_t *hw;
+  int i;
+  clib_error_t *err = 0;
+  struct ifreq ifr;
+  size_t hdrsz;
+  struct vhost_memory *vhost_mem = 0;
+  virtio_if_t *vif = 0;
+  int rv = 0;
+
+  memset (&ifr, 0, sizeof (ifr));
+  pool_get (vim->interfaces, vif);
+  vif->dev_instance = vif - vim->interfaces;
+  vif->tap_fd = -1;
+
+  if ((vif->fd = open ("/dev/vhost-net", O_RDWR | O_NONBLOCK)) < 0)
+    {
+      rv = VNET_API_ERROR_SYSCALL_ERROR_1;
+      goto error;
+    }
+
+  _IOCTL (vif->fd, VHOST_GET_FEATURES, &vif->remote_features);
+
+  if ((vif->remote_features & (1ULL << VIRTIO_NET_F_MRG_RXBUF)) == 0)
+    {
+      rv = VNET_API_ERROR_UNSUPPORTED;
+      goto error;
+    }
+
+  if ((vif->remote_features & (1ULL << VIRTIO_RING_F_INDIRECT_DESC)) == 0)
+    {
+      rv = VNET_API_ERROR_UNSUPPORTED;
+      goto error;
+    }
+
+  if ((vif->remote_features & (1ULL << VIRTIO_F_VERSION_1)) == 0)
+    {
+      rv = VNET_API_ERROR_UNSUPPORTED;
+      goto error;
+    }
+
+  vif->features |= 1ULL << VIRTIO_NET_F_MRG_RXBUF;
+  vif->features |= 1ULL << VIRTIO_F_VERSION_1;
+  vif->features |= 1ULL << VIRTIO_RING_F_INDIRECT_DESC;
+
+  _IOCTL (vif->fd, VHOST_SET_FEATURES, &vif->features);
+
+  if ((vif->tap_fd = open ("/dev/net/tun", O_RDWR | O_NONBLOCK)) < 0)
+    {
+      rv = VNET_API_ERROR_SYSCALL_ERROR_2;
+      goto error;
+    }
+
+  ifr.ifr_flags = IFF_TAP | IFF_NO_PI | IFF_ONE_QUEUE | IFF_VNET_HDR;
+  strncpy (ifr.ifr_ifrn.ifrn_name, (char *) args->name, IF_NAMESIZE);
+  _IOCTL (vif->tap_fd, TUNSETIFF, (void *) &ifr);
+
+  vif->ifindex = if_nametoindex ((char *) args->name);
+
+  unsigned int offload = 0;
+  hdrsz = sizeof (struct virtio_net_hdr_v1);
+  _IOCTL (vif->tap_fd, TUNSETOFFLOAD, offload);
+  _IOCTL (vif->tap_fd, TUNSETVNETHDRSZ, &hdrsz);
+  _IOCTL (vif->fd, VHOST_SET_OWNER, 0);
+
+  if (args->net_ns)
+    {
+      err = clib_netlink_set_if_namespace (vif->ifindex,
+					   (char *) args->net_ns);
+      if (err)
+	{
+	  rv = VNET_API_ERROR_NAMESPACE_CREATE;
+	  goto error;
+	}
+    }
+
+  /* Set vhost memory table */
+  i = sizeof (struct vhost_memory) + sizeof (struct vhost_memory_region);
+  vhost_mem = clib_mem_alloc (i);
+  memset (vhost_mem, 0, i);
+  vhost_mem->nregions = 1;
+  vhost_mem->regions[0].memory_size = (1ULL << 47) - 4096;
+  _IOCTL (vif->fd, VHOST_SET_MEM_TABLE, vhost_mem);
+
+  if ((err = virtio_vring_init (vm, vif, 0, args->rx_ring_sz)))
+    {
+      rv = VNET_API_ERROR_VIRTIO_INIT;
+      goto error;
+    }
+
+  if ((err = virtio_vring_init (vm, vif, 1, args->tx_ring_sz)))
+    {
+      rv = VNET_API_ERROR_VIRTIO_INIT;
+      goto error;
+    }
+
+  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;
+    }
+  vif->name = args->name;
+  args->name = 0;
+  vif->net_ns = args->net_ns;
+  args->net_ns = 0;
+  err = ethernet_register_interface (vnm, virtio_device_class.index,
+				     vif->dev_instance, args->hw_addr,
+				     &vif->hw_if_index,
+				     virtio_eth_flag_change);
+  if (err)
+    rv = VNET_API_ERROR_INVALID_REGISTRATION;
+
+  sw = vnet_get_hw_sw_interface (vnm, vif->hw_if_index);
+  vif->sw_if_index = sw->sw_if_index;
+  args->sw_if_index = vif->sw_if_index;
+  hw = vnet_get_hw_interface (vnm, vif->hw_if_index);
+  hw->flags |= VNET_HW_INTERFACE_FLAG_SUPPORTS_INT_MODE;
+  vnet_hw_interface_set_input_node (vnm, vif->hw_if_index,
+				    virtio_input_node.index);
+  vnet_hw_interface_assign_rx_thread (vnm, vif->hw_if_index, 0, ~0);
+  vnet_hw_interface_set_rx_mode (vnm, vif->hw_if_index, 0,
+				 VNET_HW_INTERFACE_RX_MODE_DEFAULT);
+  vif->per_interface_next_index = ~0;
+  vif->type = VIRTIO_IF_TYPE_TAP;
+  vif->flags |= VIRTIO_IF_FLAG_ADMIN_UP;
+  vnet_hw_interface_set_flags (vnm, vif->hw_if_index,
+			       VNET_HW_INTERFACE_FLAG_LINK_UP);
+  goto done;
+
+error:
+  if (vif->tap_fd != -1)
+    close (vif->tap_fd);
+  if (vif->fd != -1)
+    close (vif->fd);
+  vec_foreach_index (i, vif->vrings) virtio_vring_free (vif, i);
+  memset (vif, 0, sizeof (virtio_if_t));
+  pool_put (vim->interfaces, vif);
+
+done:
+  if (vhost_mem)
+    clib_mem_free (vhost_mem);
+
+  return rv;
+}
+
+int
+tap_delete_if (vlib_main_t * vm, u32 sw_if_index)
+{
+  vnet_main_t *vnm = vnet_get_main ();
+  virtio_main_t *mm = &virtio_main;
+  int i;
+  virtio_if_t *vif;
+  vnet_hw_interface_t *hw;
+
+  hw = vnet_get_sup_hw_interface (vnm, sw_if_index);
+  if (hw == NULL || virtio_device_class.index != hw->dev_class_index)
+    return VNET_API_ERROR_INVALID_SW_IF_INDEX;
+
+  vif = pool_elt_at_index (mm->interfaces, hw->dev_instance);
+
+  /* bring down the interface */
+  vnet_hw_interface_set_flags (vnm, vif->hw_if_index, 0);
+  vnet_sw_interface_set_flags (vnm, vif->sw_if_index, 0);
+
+  ethernet_delete_interface (vnm, vif->hw_if_index);
+  vif->hw_if_index = ~0;
+
+  if (vif->tap_fd != -1)
+    close (vif->tap_fd);
+  if (vif->fd != -1)
+    close (vif->fd);
+
+  vec_foreach_index (i, vif->vrings) virtio_vring_free (vif, i);
+  vec_free (vif->vrings);
+
+  memset (vif, 0, sizeof (*vif));
+  pool_put (mm->interfaces, vif);
+
+  return 0;
+}
+
+int
+tap_dump_ifs (tap_interface_details_t ** out_tapids)
+{
+  vnet_main_t *vnm = vnet_get_main ();
+  virtio_main_t *mm = &virtio_main;
+  virtio_if_t *vif;
+  vnet_hw_interface_t *hi;
+  tap_interface_details_t *r_tapids = NULL;
+  tap_interface_details_t *tapid = NULL;
+
+  /* *INDENT-OFF* */
+  pool_foreach (vif, mm->interfaces,
+    vec_add2(r_tapids, tapid, 1);
+    memset (tapid, 0, sizeof (*tapid));
+    tapid->sw_if_index = vif->sw_if_index;
+    hi = vnet_get_hw_interface (vnm, vif->hw_if_index);
+    clib_memcpy(tapid->dev_name, hi->name,
+		MIN (ARRAY_LEN (tapid->dev_name) - 1,
+		     strlen ((const char *) hi->name)));
+  );
+  /* *INDENT-ON* */
+
+  *out_tapids = r_tapids;
+
+  return 0;
+}
+
+static clib_error_t *
+tap_init (vlib_main_t * vm)
+{
+
+  return 0;
+}
+
+VLIB_INIT_FUNCTION (tap_init);
+
+/*
+ * fd.io coding-style-patch-verification: ON
+ *
+ * Local Variables:
+ * eval: (c-set-style "gnu")
+ * End:
+ */
diff --git a/src/vnet/devices/virtio/tap.h b/src/vnet/devices/virtio/tap.h
new file mode 100644
index 00000000000..58dcb5bda51
--- /dev/null
+++ b/src/vnet/devices/virtio/tap.h
@@ -0,0 +1,56 @@
+/*
+ *------------------------------------------------------------------
+ * Copyright (c) 2017 Cisco and/or its affiliates.
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at:
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ *------------------------------------------------------------------
+ */
+
+#ifndef _VNET_DEVICES_VIRTIO_TAP_H_
+#define _VNET_DEVICES_VIRTIO_TAP_H_
+
+#ifndef MIN
+#define MIN(x,y) (((x)<(y))?(x):(y))
+#endif
+
+typedef struct
+{
+  u8 *name;
+  u8 *net_ns;
+  u8 hw_addr_set;
+  u8 hw_addr[6];
+  u16 rx_ring_sz;
+  u16 tx_ring_sz;
+  /* return */
+  u32 sw_if_index;
+} tap_create_if_args_t;
+
+/** TAP interface details struct */
+typedef struct
+{
+  u32 sw_if_index;
+  u8 dev_name[64];
+} tap_interface_details_t;
+
+int tap_create_if (vlib_main_t * vm, tap_create_if_args_t * args);
+int tap_delete_if (vlib_main_t * vm, u32 sw_if_index);
+int tap_dump_ifs (tap_interface_details_t ** out_tapids);
+
+#endif /* _VNET_DEVICES_VIRTIO_TAP_H_ */
+
+/*
+ * fd.io coding-style-patch-verification: ON
+ *
+ * Local Variables:
+ * eval: (c-set-style "gnu")
+ * End:
+ */
diff --git a/src/vnet/devices/virtio/tapv2.api b/src/vnet/devices/virtio/tapv2.api
new file mode 100644
index 00000000000..e1592cf7d61
--- /dev/null
+++ b/src/vnet/devices/virtio/tapv2.api
@@ -0,0 +1,94 @@
+/*
+ * Copyright (c) 2017 Cisco and/or its affiliates.
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at:
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+/** \file
+
+    This file defines vpe control-plane API messages for
+    the Linux kernel TAP device driver
+*/
+
+vl_api_version 1.0.0
+
+/** \brief Initialize a new tap interface with the given paramters
+    @param client_index - opaque cookie to identify the sender
+    @param context - sender context, to match reply w/ request
+    @param use_random_mac - let the system generate a unique mac address
+    @param tap_name - name to associate with the new interface
+    @param mac_address - mac addr to assign to the interface if use_radom not set
+    @param net_ns_set - net_ns is entered
+    @param net_ns - netlink name space
+    @param tx_ring_sz - the number of entries of TX ring
+    @param rx_ring_sz - the number of entries of RX ring
+*/
+define tap_create_v2
+{
+  u32 client_index;
+  u32 context;
+  u8 use_random_mac;
+  u8 tap_name[64];
+  u8 mac_address[6];
+  u8 net_ns_set;
+  u8 net_ns[64];
+  u16 tx_ring_sz; /* optional, default is 256 entries, must be power of 2 */
+  u16 rx_ring_sz; /* optional, default is 256 entries, must be power of 2 */
+};
+
+/** \brief Reply for tap create reply
+    @param context - returned sender context, to match reply w/ request
+    @param retval - return code
+    @param sw_if_index - software index allocated for the new tap interface
+*/
+define tap_create_v2_reply
+{
+  u32 context;
+  i32 retval;
+  u32 sw_if_index;
+};
+
+/** \brief Delete tap interface
+    @param client_index - opaque cookie to identify the sender
+    @param context - sender context, to match reply w/ request
+    @param sw_if_index - interface index of existing tap interface
+*/
+autoreply define tap_delete_v2
+{
+  u32 client_index;
+  u32 context;
+  u32 sw_if_index;
+};
+
+/** \brief Dump tap interfaces request */
+define sw_interface_tap_v2_dump
+{
+  u32 client_index;
+  u32 context;
+};
+
+/** \brief Reply for tap dump request
+    @param sw_if_index - software index of tap interface
+    @param dev_name - Linux tap device name
+*/
+define sw_interface_tap_v2_details
+{
+  u32 context;
+  u32 sw_if_index;
+  u8 dev_name[64];
+};
+
+/*
+ * Local Variables:
+ * eval: (c-set-style "gnu")
+ * End:
+ */
diff --git a/src/vnet/devices/virtio/tapv2_api.c b/src/vnet/devices/virtio/tapv2_api.c
new file mode 100644
index 00000000000..1c559e6c28e
--- /dev/null
+++ b/src/vnet/devices/virtio/tapv2_api.c
@@ -0,0 +1,222 @@
+/*
+ *------------------------------------------------------------------
+ * tap_api.c - vnet tap device driver API support
+ *
+ * Copyright (c) 2017 Cisco and/or its affiliates.
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at:
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ *------------------------------------------------------------------
+ */
+
+#include <vnet/vnet.h>
+#include <vlibmemory/api.h>
+
+#include <vnet/interface.h>
+#include <vnet/api_errno.h>
+#include <vnet/ethernet/ethernet.h>
+#include <vnet/ip/ip.h>
+
+#include <vnet/vnet_msg_enum.h>
+
+#define vl_typedefs		/* define message structures */
+#include <vnet/vnet_all_api_h.h>
+#undef vl_typedefs
+
+#define vl_endianfun		/* define message structures */
+#include <vnet/vnet_all_api_h.h>
+#undef vl_endianfun
+
+/* instantiate all the print functions we know about */
+#define vl_print(handle, ...) vlib_cli_output (handle, __VA_ARGS__)
+#define vl_printfun
+#include <vnet/vnet_all_api_h.h>
+#undef vl_printfun
+
+#include <vlibapi/api_helper_macros.h>
+#include <vnet/devices/virtio/tap.h>
+
+#define foreach_tapv2_api_msg                     \
+_(TAP_CREATE_V2, tap_create_v2)                   \
+_(TAP_DELETE_V2, tap_delete_v2)                   \
+_(SW_INTERFACE_TAP_V2_DUMP, sw_interface_tap_v2_dump)
+
+static void
+vl_api_tap_create_v2_t_handler (vl_api_tap_create_v2_t * mp)
+{
+  vlib_main_t *vm = vlib_get_main ();
+  int rv;
+  vl_api_tap_create_v2_reply_t *rmp;
+  unix_shared_memory_queue_t *q;
+  tap_create_if_args_t _a, *ap = &_a;
+
+  memset (ap, 0, sizeof (*ap));
+
+  ap->name = mp->tap_name;
+  if (!mp->use_random_mac)
+    {
+      clib_memcpy (ap->hw_addr, mp->mac_address, 6);
+      ap->hw_addr_set = 1;
+    }
+  ap->rx_ring_sz = mp->rx_ring_sz;
+  ap->tx_ring_sz = mp->tx_ring_sz;
+  ap->sw_if_index = (u32) ~ 0;
+  if (mp->net_ns_set)
+    ap->net_ns = mp->net_ns;
+
+  rv = tap_create_if (vm, ap);
+
+  q = vl_api_client_index_to_input_queue (mp->client_index);
+  if (!q)
+    return;
+
+  rmp = vl_msg_api_alloc (sizeof (*rmp));
+  rmp->_vl_msg_id = ntohs (VL_API_TAP_CREATE_V2_REPLY);
+  rmp->context = mp->context;
+  rmp->retval = ntohl (rv);
+  rmp->sw_if_index = ntohl (ap->sw_if_index);
+
+  vl_msg_api_send_shmem (q, (u8 *) & rmp);
+}
+
+static void
+tap_send_sw_interface_event_deleted (vpe_api_main_t * am,
+				     unix_shared_memory_queue_t * q,
+				     u32 sw_if_index)
+{
+  vl_api_sw_interface_event_t *mp;
+
+  mp = vl_msg_api_alloc (sizeof (*mp));
+  memset (mp, 0, sizeof (*mp));
+  mp->_vl_msg_id = ntohs (VL_API_SW_INTERFACE_EVENT);
+  mp->sw_if_index = ntohl (sw_if_index);
+
+  mp->admin_up_down = 0;
+  mp->link_up_down = 0;
+  mp->deleted = 1;
+  vl_msg_api_send_shmem (q, (u8 *) & mp);
+}
+
+static void
+vl_api_tap_delete_v2_t_handler (vl_api_tap_delete_v2_t * mp)
+{
+  vlib_main_t *vm = vlib_get_main ();
+  int rv;
+  vpe_api_main_t *vam = &vpe_api_main;
+  vl_api_tap_delete_v2_reply_t *rmp;
+  unix_shared_memory_queue_t *q;
+  u32 sw_if_index = ntohl (mp->sw_if_index);
+
+  rv = tap_delete_if (vm, sw_if_index);
+
+  q = vl_api_client_index_to_input_queue (mp->client_index);
+  if (!q)
+    return;
+
+  rmp = vl_msg_api_alloc (sizeof (*rmp));
+  rmp->_vl_msg_id = ntohs (VL_API_TAP_DELETE_V2_REPLY);
+  rmp->context = mp->context;
+  rmp->retval = ntohl (rv);
+
+  vl_msg_api_send_shmem (q, (u8 *) & rmp);
+
+  if (!rv)
+    tap_send_sw_interface_event_deleted (vam, q, sw_if_index);
+}
+
+static void
+tap_send_sw_interface_details (vpe_api_main_t * am,
+			       unix_shared_memory_queue_t * q,
+			       tap_interface_details_t * tap_if, u32 context)
+{
+  vl_api_sw_interface_tap_v2_details_t *mp;
+  mp = vl_msg_api_alloc (sizeof (*mp));
+  memset (mp, 0, sizeof (*mp));
+  mp->_vl_msg_id = ntohs (VL_API_SW_INTERFACE_TAP_V2_DETAILS);
+  mp->sw_if_index = ntohl (tap_if->sw_if_index);
+  clib_memcpy (mp->dev_name, tap_if->dev_name,
+	       MIN (ARRAY_LEN (mp->dev_name) - 1,
+		    strlen ((const char *) tap_if->dev_name)));
+  mp->context = context;
+
+  vl_msg_api_send_shmem (q, (u8 *) & mp);
+}
+
+static void
+vl_api_sw_interface_tap_v2_dump_t_handler (vl_api_sw_interface_tap_v2_dump_t *
+					   mp)
+{
+  int rv;
+  vpe_api_main_t *am = &vpe_api_main;
+  unix_shared_memory_queue_t *q;
+  tap_interface_details_t *tapifs = NULL;
+  tap_interface_details_t *tap_if = NULL;
+
+  q = vl_api_client_index_to_input_queue (mp->client_index);
+  if (q == 0)
+    return;
+
+  rv = tap_dump_ifs (&tapifs);
+  if (rv)
+    return;
+
+  vec_foreach (tap_if, tapifs)
+  {
+    tap_send_sw_interface_details (am, q, tap_if, mp->context);
+  }
+
+  vec_free (tapifs);
+}
+
+#define vl_msg_name_crc_list
+#include <vnet/vnet_all_api_h.h>
+#undef vl_msg_name_crc_list
+
+static void
+tap_setup_message_id_table (api_main_t * am)
+{
+#define _(id,n,crc) vl_msg_api_add_msg_name_crc (am, #n "_" #crc, id);
+  foreach_vl_msg_name_crc_tapv2;
+#undef _
+}
+
+static clib_error_t *
+tapv2_api_hookup (vlib_main_t * vm)
+{
+  api_main_t *am = &api_main;
+
+#define _(N,n)                                                  \
+    vl_msg_api_set_handlers(VL_API_##N, #n,                     \
+                           vl_api_##n##_t_handler,              \
+                           vl_noop_handler,                     \
+                           vl_api_##n##_t_endian,               \
+                           vl_api_##n##_t_print,                \
+                           sizeof(vl_api_##n##_t), 1);
+  foreach_tapv2_api_msg;
+#undef _
+
+  /*
+   * Set up the (msg_name, crc, message-id) table
+   */
+  tap_setup_message_id_table (am);
+
+  return 0;
+}
+
+VLIB_API_INIT_FUNCTION (tapv2_api_hookup);
+
+/*
+ * fd.io coding-style-patch-verification: ON
+ *
+ * Local Variables:
+ * eval: (c-set-style "gnu")
+ * End:
+ */
diff --git a/src/vnet/devices/virtio/virtio.c b/src/vnet/devices/virtio/virtio.c
new file mode 100644
index 00000000000..63ca6011a9e
--- /dev/null
+++ b/src/vnet/devices/virtio/virtio.c
@@ -0,0 +1,159 @@
+/*
+ *------------------------------------------------------------------
+ * Copyright (c) 2017 Cisco and/or its affiliates.
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at:
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ *------------------------------------------------------------------
+ */
+
+#include <sys/types.h>
+#include <sys/stat.h>
+#include <fcntl.h>
+#include <net/if.h>
+#include <linux/if_tun.h>
+#include <sys/ioctl.h>
+#include <linux/virtio_net.h>
+#include <linux/vhost.h>
+#include <sys/eventfd.h>
+
+#include <vlib/vlib.h>
+#include <vlib/unix/unix.h>
+#include <vnet/ethernet/ethernet.h>
+#include <vnet/devices/virtio/virtio.h>
+
+virtio_main_t virtio_main;
+
+#define _IOCTL(fd,a,...) \
+  if (ioctl (fd, a, __VA_ARGS__) < 0) \
+    { \
+      err = clib_error_return_unix (0, "ioctl(" #a ")"); \
+      goto error; \
+    }
+
+static clib_error_t *
+call_read_ready (clib_file_t * uf)
+{
+  virtio_main_t *nm = &virtio_main;
+  vnet_main_t *vnm = vnet_get_main ();
+  u16 qid = uf->private_data & 0xFFFF;
+  virtio_if_t *vif =
+    vec_elt_at_index (nm->interfaces, uf->private_data >> 16);
+  u64 b;
+
+  CLIB_UNUSED (ssize_t size) = read (uf->file_descriptor, &b, sizeof (b));
+  if ((qid & 1) == 0)
+    vnet_device_input_set_interrupt_pending (vnm, vif->hw_if_index, qid);
+
+  return 0;
+}
+
+
+clib_error_t *
+virtio_vring_init (vlib_main_t * vm, virtio_if_t * vif, u16 idx, u16 sz)
+{
+  clib_error_t *err = 0;
+  virtio_vring_t *vring;
+  struct vhost_vring_state state;
+  struct vhost_vring_addr addr;
+  struct vhost_vring_file file;
+  clib_file_t t = { 0 };
+  int i;
+
+  if (!is_pow2 (sz))
+    return clib_error_return (0, "ring size must be power of 2");
+
+  if (sz > 32768)
+    return clib_error_return (0, "ring size must be 32768 or lower");
+
+  if (sz == 0)
+    sz = 256;
+
+  vec_validate_aligned (vif->vrings, idx, CLIB_CACHE_LINE_BYTES);
+  vring = vec_elt_at_index (vif->vrings, idx);
+
+  i = sizeof (struct vring_desc) * sz;
+  i = round_pow2 (i, CLIB_CACHE_LINE_BYTES);
+  vring->desc = clib_mem_alloc_aligned (i, CLIB_CACHE_LINE_BYTES);
+  memset (vring->desc, 0, i);
+
+  i = sizeof (struct vring_avail) + sz * sizeof (vring->avail->ring[0]);
+  i = round_pow2 (i, CLIB_CACHE_LINE_BYTES);
+  vring->avail = clib_mem_alloc_aligned (i, CLIB_CACHE_LINE_BYTES);
+  memset (vring->avail, 0, i);
+  // tell kernel that we don't need interrupt
+  vring->avail->flags = VIRTIO_RING_FLAG_MASK_INT;
+
+  i = sizeof (struct vring_used) + sz * sizeof (struct vring_used_elem);
+  i = round_pow2 (i, CLIB_CACHE_LINE_BYTES);
+  vring->used = clib_mem_alloc_aligned (i, CLIB_CACHE_LINE_BYTES);
+  memset (vring->used, 0, i);
+
+  ASSERT (vring->buffers == 0);
+  vec_validate_aligned (vring->buffers, sz * 2, CLIB_CACHE_LINE_BYTES);
+
+  vring->size = sz;
+  vring->call_fd = eventfd (0, EFD_NONBLOCK | EFD_CLOEXEC);
+  vring->kick_fd = eventfd (0, EFD_CLOEXEC);
+
+  t.read_function = call_read_ready;
+  t.file_descriptor = vring->call_fd;
+  t.private_data = vif->dev_instance << 16 | idx;
+  vring->call_file_index = clib_file_add (&file_main, &t);
+
+  state.index = idx;
+  state.num = sz;
+  _IOCTL (vif->fd, VHOST_SET_VRING_NUM, &state);
+
+  addr.index = idx;
+  addr.flags = 0;
+  addr.desc_user_addr = pointer_to_uword (vring->desc);
+  addr.avail_user_addr = pointer_to_uword (vring->avail);
+  addr.used_user_addr = pointer_to_uword (vring->used);
+  _IOCTL (vif->fd, VHOST_SET_VRING_ADDR, &addr);
+
+  file.index = idx;
+  file.fd = vring->kick_fd;
+  _IOCTL (vif->fd, VHOST_SET_VRING_KICK, &file);
+  file.fd = vring->call_fd;
+  _IOCTL (vif->fd, VHOST_SET_VRING_CALL, &file);
+  file.fd = vif->tap_fd;
+  _IOCTL (vif->fd, VHOST_NET_SET_BACKEND, &file);
+
+error:
+  return err;
+}
+
+clib_error_t *
+virtio_vring_free (virtio_if_t * vif, u32 idx)
+{
+  //TODO free buffers and indirect descriptor allocs
+  virtio_vring_t *vring = vec_elt_at_index (vif->vrings, idx);
+  if (vring->desc)
+    clib_mem_free (vring->desc);
+  if (vring->avail)
+    clib_mem_free (vring->avail);
+  if (vring->used)
+    clib_mem_free (vring->used);
+  clib_file_del_by_index (&file_main, vring->call_file_index);
+  close (vring->kick_fd);
+  close (vring->call_fd);
+  vec_free (vring->buffers);
+  return 0;
+}
+
+/*
+ * fd.io coding-style-patch-verification: ON
+ *
+ * Local Variables:
+ * eval: (c-set-style "gnu")
+ * End:
+ */
diff --git a/src/vnet/devices/virtio/virtio.h b/src/vnet/devices/virtio/virtio.h
new file mode 100644
index 00000000000..7dcd90ac353
--- /dev/null
+++ b/src/vnet/devices/virtio/virtio.h
@@ -0,0 +1,131 @@
+/*
+ *------------------------------------------------------------------
+ * Copyright (c) 2017 Cisco and/or its affiliates.
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at:
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ *------------------------------------------------------------------
+ */
+
+#ifndef _VNET_DEVICES_VIRTIO_VIRTIO_H_
+#define _VNET_DEVICES_VIRTIO_VIRTIO_H_
+
+#define foreach_virtio_net_features      \
+  _ (VIRTIO_NET_F_CSUM, 0)	/* Host handles pkts w/ partial csum */ \
+  _ (VIRTIO_NET_F_GUEST_CSUM, 1) /* Guest handles pkts w/ partial csum */ \
+  _ (VIRTIO_NET_F_CTRL_GUEST_OFFLOADS, 2) /* Dynamic offload configuration. */ \
+  _ (VIRTIO_NET_F_MTU, 3)       /* Initial MTU advice. */ \
+  _ (VIRTIO_NET_F_MAC, 5)	/* Host has given MAC address. */ \
+  _ (VIRTIO_NET_F_GSO, 6)	/* Host handles pkts w/ any GSO. */ \
+  _ (VIRTIO_NET_F_GUEST_TSO4, 7)	/* Guest can handle TSOv4 in. */ \
+  _ (VIRTIO_NET_F_GUEST_TSO6, 8)	/* Guest can handle TSOv6 in. */ \
+  _ (VIRTIO_NET_F_GUEST_ECN, 9)	/* Guest can handle TSO[6] w/ ECN in. */ \
+  _ (VIRTIO_NET_F_GUEST_UFO, 10)	/* Guest can handle UFO in. */ \
+  _ (VIRTIO_NET_F_HOST_TSO4, 11)	/* Host can handle TSOv4 in. */ \
+  _ (VIRTIO_NET_F_HOST_TSO6, 12)	/* Host can handle TSOv6 in. */ \
+  _ (VIRTIO_NET_F_HOST_ECN, 13)	/* Host can handle TSO[6] w/ ECN in. */ \
+  _ (VIRTIO_NET_F_HOST_UFO, 14)	/* Host can handle UFO in. */ \
+  _ (VIRTIO_NET_F_MRG_RXBUF, 15)	/* Host can merge receive buffers. */ \
+  _ (VIRTIO_NET_F_STATUS, 16)	/* virtio_net_config.status available */ \
+  _ (VIRTIO_NET_F_CTRL_VQ, 17)	/* Control channel available */ \
+  _ (VIRTIO_NET_F_CTRL_RX, 18)	/* Control channel RX mode support */ \
+  _ (VIRTIO_NET_F_CTRL_VLAN, 19)	/* Control channel VLAN filtering */ \
+  _ (VIRTIO_NET_F_CTRL_RX_EXTRA, 20)	/* Extra RX mode control support */ \
+  _ (VIRTIO_NET_F_GUEST_ANNOUNCE, 21)	/* Guest can announce device on the network */ \
+  _ (VIRTIO_NET_F_MQ, 22)	        /* Device supports Receive Flow Steering */ \
+  _ (VIRTIO_NET_F_CTRL_MAC_ADDR, 23)	/* Set MAC address */ \
+  _ (VIRTIO_F_NOTIFY_ON_EMPTY, 24) \
+  _ (VHOST_F_LOG_ALL, 26)      /* Log all write descriptors */ \
+  _ (VIRTIO_F_ANY_LAYOUT, 27)  /* Can the device handle any descripor layout */ \
+  _ (VIRTIO_RING_F_INDIRECT_DESC, 28)   /* Support indirect buffer descriptors */ \
+  _ (VIRTIO_RING_F_EVENT_IDX, 29)       /* The Guest publishes the used index for which it expects an interrupt \
+ * at the end of the avail ring. Host should ignore the avail->flags field. */ \
+/* The Host publishes the avail index for which it expects a kick \
+ * at the end of the used ring. Guest should ignore the used->flags field. */ \
+  _ (VHOST_USER_F_PROTOCOL_FEATURES, 30) \
+  _ (VIRTIO_F_VERSION_1, 32)
+
+#define foreach_virtio_if_flag		\
+  _(0, ADMIN_UP, "admin-up")		\
+  _(1, DELETING, "deleting")
+
+typedef enum
+{
+#define _(a, b, c) VIRTIO_IF_FLAG_##b = (1 << a),
+  foreach_virtio_if_flag
+#undef _
+} virtio_if_flag_t;
+
+typedef enum
+{
+  VIRTIO_IF_TYPE_TAP,
+  VIRTIO_IF_N_TYPES,
+} virtio_if_type_t;
+
+
+typedef struct
+{
+  struct vring_desc *desc;
+  struct vring_used *used;
+  struct vring_avail *avail;
+  u16 desc_in_use;
+  u16 desc_next;
+  int kick_fd;
+  int call_fd;
+  u16 size;
+#define VIRTIO_RING_FLAG_MASK_INT 1
+  u32 flags;
+  u32 call_file_index;
+  u32 *buffers;
+  u16 last_used_idx;
+} virtio_vring_t;
+
+typedef struct
+{
+  u32 flags;
+  u32 dev_instance;
+  u32 hw_if_index;
+  u32 sw_if_index;
+  u32 per_interface_next_index;
+  int fd;
+  int tap_fd;
+  virtio_vring_t *vrings;
+
+  u64 features, remote_features;
+
+  virtio_if_type_t type;
+  u8 *name;
+  u8 *net_ns;
+  int ifindex;
+} virtio_if_t;
+
+typedef struct
+{
+  virtio_if_t *interfaces;
+} virtio_main_t;
+
+extern virtio_main_t virtio_main;
+extern vnet_device_class_t virtio_device_class;
+extern vlib_node_registration_t virtio_input_node;
+
+clib_error_t *virtio_vring_init (vlib_main_t * vm, virtio_if_t * vif, u16 idx,
+				 u16 sz);
+clib_error_t *virtio_vring_free (virtio_if_t * vif, u32 idx);
+
+#endif /* _VNET_DEVICES_VIRTIO_VIRTIO_H_ */
+
+/*
+ * fd.io coding-style-patch-verification: ON
+ *
+ * Local Variables:
+ * eval: (c-set-style "gnu")
+ * End:
+ */
diff --git a/src/vnet/vnet_all_api_h.h b/src/vnet/vnet_all_api_h.h
index 6c7c63683f8..2e7419b77ce 100644
--- a/src/vnet/vnet_all_api_h.h
+++ b/src/vnet/vnet_all_api_h.h
@@ -32,6 +32,7 @@
 #include <vnet/devices/af_packet/af_packet.api.h>
 #include <vnet/devices/netmap/netmap.api.h>
 #include <vnet/devices/virtio/vhost_user.api.h>
+#include <vnet/devices/virtio/tapv2.api.h>
 #include <vnet/gre/gre.api.h>
 #include <vnet/interface.api.h>
 #include <vnet/map/map.api.h>
diff --git a/src/vpp/api/custom_dump.c b/src/vpp/api/custom_dump.c
index f27acda24dd..428e1636569 100644
--- a/src/vpp/api/custom_dump.c
+++ b/src/vpp/api/custom_dump.c
@@ -547,6 +547,47 @@ static void *vl_api_sw_interface_tap_dump_t_print
   FINISH;
 }
 
+static void *vl_api_tap_create_v2_t_print
+  (vl_api_tap_create_v2_t * mp, void *handle)
+{
+  u8 *s;
+  u8 null_mac[6];
+
+  memset (null_mac, 0, sizeof (null_mac));
+
+  s = format (0, "SCRIPT: tap_create_v2 ");
+  s = format (s, "name %s ", mp->tap_name);
+  if (memcmp (mp->mac_address, null_mac, 6))
+    s = format (s, "hw-addr %U ", format_ethernet_address, mp->mac_address);
+  if (mp->net_ns_set)
+    s = format (s, "host-ns %s ", mp->net_ns);
+  if (mp->tx_ring_sz)
+    s = format (s, "tx-ring-size %d ", mp->tx_ring_sz);
+  if (mp->rx_ring_sz)
+    s = format (s, "rx-ring-size %d ", mp->rx_ring_sz);
+  FINISH;
+}
+
+static void *vl_api_tap_delete_v2_t_print
+  (vl_api_tap_delete_v2_t * mp, void *handle)
+{
+  u8 *s;
+
+  s = format (0, "SCRIPT: tap_delete_v2 ");
+  s = format (s, "sw_if_index %d ", ntohl (mp->sw_if_index));
+
+  FINISH;
+}
+
+static void *vl_api_sw_interface_tap_v2_dump_t_print
+  (vl_api_sw_interface_tap_v2_dump_t * mp, void *handle)
+{
+  u8 *s;
+
+  s = format (0, "SCRIPT: sw_interface_tap_v2_dump ");
+
+  FINISH;
+}
 
 static void *vl_api_ip_add_del_route_t_print
   (vl_api_ip_add_del_route_t * mp, void *handle)
@@ -3269,6 +3310,9 @@ _(TAP_CONNECT, tap_connect)                                             \
 _(TAP_MODIFY, tap_modify)                                               \
 _(TAP_DELETE, tap_delete)                                               \
 _(SW_INTERFACE_TAP_DUMP, sw_interface_tap_dump)                         \
+_(TAP_CREATE_V2, tap_create_v2)                                         \
+_(TAP_DELETE_V2, tap_delete_v2)                                         \
+_(SW_INTERFACE_TAP_V2_DUMP, sw_interface_tap_v2_dump)                   \
 _(IP_ADD_DEL_ROUTE, ip_add_del_route)                                   \
 _(PROXY_ARP_ADD_DEL, proxy_arp_add_del)                                 \
 _(PROXY_ARP_INTFC_ENABLE_DISABLE, proxy_arp_intfc_enable_disable)       \