devices: add support for af-packet v2

Type: feature

Signed-off-by: Mohsin Kazmi <sykazmi@cisco.com>
Change-Id: I2ccaf1d512dcb72e414be8c69cbb538ebbe0e933

7 files changed, 645 insertions, 186 deletions

diff --git a/src/vnet/devices/af_packet/af_packet.api b/src/vnet/devices/af_packet/af_packet.api
index 4a5cfb0fc3a..a12da37796a 100644
--- a/src/vnet/devices/af_packet/af_packet.api
+++ b/src/vnet/devices/af_packet/af_packet.api
@@ -94,6 +94,7 @@ enum af_packet_mode {
 enum af_packet_flags {
   AF_PACKET_API_FLAG_QDISC_BYPASS = 1, /* enable the qdisc bypass */
   AF_PACKET_API_FLAG_CKSUM_GSO = 2, /* enable checksum/gso */
+  AF_PACKET_API_FLAG_VERSION_2 = 8 [backwards_compatible], /* af packet v2, default is v3 */
 };
 
 /** \brief Create host-interface
diff --git a/src/vnet/devices/af_packet/af_packet.c b/src/vnet/devices/af_packet/af_packet.c
index ec65bf6d493..010bc1c266c 100644
--- a/src/vnet/devices/af_packet/af_packet.c
+++ b/src/vnet/devices/af_packet/af_packet.c
@@ -48,6 +48,10 @@ VNET_HW_INTERFACE_CLASS (af_packet_ip_device_hw_interface_class, static) = {
 #define AF_PACKET_DEFAULT_TX_FRAME_SIZE	      (2048 * 33) // GSO packet of 64KB
 #define AF_PACKET_TX_BLOCK_NR		1
 
+#define AF_PACKET_DEFAULT_RX_FRAMES_PER_BLOCK_V2 1024
+#define AF_PACKET_DEFAULT_RX_FRAME_SIZE_V2	 (2048 * 33) // GSO packet of 64KB
+#define AF_PACKET_RX_BLOCK_NR_V2		 1
+
 #define AF_PACKET_DEFAULT_RX_FRAMES_PER_BLOCK 32
 #define AF_PACKET_DEFAULT_RX_FRAME_SIZE	      2048
 #define AF_PACKET_RX_BLOCK_NR		      160
@@ -189,23 +193,16 @@ af_packet_set_tx_queues (vlib_main_t *vm, af_packet_if_t *apif)
 }
 
 static int
-create_packet_v3_sock (int host_if_index, tpacket_req3_t *rx_req,
-		       tpacket_req3_t *tx_req, int *fd, af_packet_ring_t *ring,
-		       u32 fanout_id, af_packet_if_flags_t *flags)
+create_packet_sock (int host_if_index, tpacket_req_u_t *rx_req,
+		    tpacket_req_u_t *tx_req, int *fd, af_packet_ring_t *ring,
+		    u32 fanout_id, af_packet_if_flags_t *flags, int ver)
 {
   af_packet_main_t *apm = &af_packet_main;
   struct sockaddr_ll sll;
   socklen_t req_sz = sizeof (tpacket_req3_t);
   int ret;
-  int ver = TPACKET_V3;
   u32 ring_sz = 0;
 
-  if (rx_req)
-    ring_sz += rx_req->tp_block_size * rx_req->tp_block_nr;
-
-  if (tx_req)
-    ring_sz += tx_req->tp_block_size * tx_req->tp_block_nr;
-
   if ((*fd = socket (AF_PACKET, SOCK_RAW, htons (ETH_P_ALL))) < 0)
     {
       vlib_log_err (apm->log_class,
@@ -297,7 +294,13 @@ create_packet_v3_sock (int host_if_index, tpacket_req3_t *rx_req,
 	      goto error;
 	    }
 	}
-
+      if (ver == TPACKET_V2)
+	{
+	  req_sz = sizeof (tpacket_req_t);
+	  ring_sz += rx_req->req.tp_block_size * rx_req->req.tp_block_nr;
+	}
+      else
+	ring_sz += rx_req->req3.tp_block_size * rx_req->req3.tp_block_nr;
       if (setsockopt (*fd, SOL_PACKET, PACKET_RX_RING, rx_req, req_sz) < 0)
 	{
 	  vlib_log_err (apm->log_class,
@@ -309,15 +312,23 @@ create_packet_v3_sock (int host_if_index, tpacket_req3_t *rx_req,
     }
 
   if (tx_req)
-    if (setsockopt (*fd, SOL_PACKET, PACKET_TX_RING, tx_req, req_sz) < 0)
-      {
-	vlib_log_err (apm->log_class,
-		      "Failed to set packet tx ring options: %s (errno %d)",
-		      strerror (errno), errno);
-	ret = VNET_API_ERROR_SYSCALL_ERROR_1;
-	goto error;
-      }
-
+    {
+      if (ver == TPACKET_V2)
+	{
+	  req_sz = sizeof (tpacket_req_t);
+	  ring_sz += tx_req->req.tp_block_size * tx_req->req.tp_block_nr;
+	}
+      else
+	ring_sz += tx_req->req3.tp_block_size * tx_req->req3.tp_block_nr;
+      if (setsockopt (*fd, SOL_PACKET, PACKET_TX_RING, tx_req, req_sz) < 0)
+	{
+	  vlib_log_err (apm->log_class,
+			"Failed to set packet tx ring options: %s (errno %d)",
+			strerror (errno), errno);
+	  ret = VNET_API_ERROR_SYSCALL_ERROR_1;
+	  goto error;
+	}
+    }
   ring->ring_start_addr = mmap (NULL, ring_sz, PROT_READ | PROT_WRITE,
 				MAP_SHARED | MAP_LOCKED, *fd, 0);
   if (ring->ring_start_addr == MAP_FAILED)
@@ -347,8 +358,8 @@ af_packet_queue_init (vlib_main_t *vm, af_packet_if_t *apif,
 		      u8 queue_id)
 {
   af_packet_main_t *apm = &af_packet_main;
-  tpacket_req3_t *rx_req = 0;
-  tpacket_req3_t *tx_req = 0;
+  tpacket_req_u_t *rx_req = 0;
+  tpacket_req_u_t *tx_req = 0;
   int ret, fd = -1;
   af_packet_ring_t ring = { 0 };
   u8 *ring_addr = 0;
@@ -360,22 +371,32 @@ af_packet_queue_init (vlib_main_t *vm, af_packet_if_t *apif,
     {
       rx_frames_per_block = arg->rx_frames_per_block ?
 				    arg->rx_frames_per_block :
-				    AF_PACKET_DEFAULT_RX_FRAMES_PER_BLOCK;
-
-      rx_frame_size = arg->rx_frame_size ? arg->rx_frame_size :
-						 AF_PACKET_DEFAULT_RX_FRAME_SIZE;
+				    ((apif->version == TPACKET_V3) ?
+				       AF_PACKET_DEFAULT_RX_FRAMES_PER_BLOCK :
+				       AF_PACKET_DEFAULT_RX_FRAMES_PER_BLOCK_V2);
+
+      rx_frame_size =
+	arg->rx_frame_size ?
+		arg->rx_frame_size :
+		((apif->version == TPACKET_V3) ? AF_PACKET_DEFAULT_RX_FRAME_SIZE :
+						 AF_PACKET_DEFAULT_RX_FRAME_SIZE_V2);
       vec_validate (rx_queue->rx_req, 0);
-      rx_queue->rx_req->tp_block_size = rx_frame_size * rx_frames_per_block;
-      rx_queue->rx_req->tp_frame_size = rx_frame_size;
-      rx_queue->rx_req->tp_block_nr = AF_PACKET_RX_BLOCK_NR;
-      rx_queue->rx_req->tp_frame_nr =
-	AF_PACKET_RX_BLOCK_NR * rx_frames_per_block;
-      rx_queue->rx_req->tp_retire_blk_tov = 1; // 1 ms block timout
-      rx_queue->rx_req->tp_feature_req_word = 0;
-      rx_queue->rx_req->tp_sizeof_priv = 0;
+      rx_queue->rx_req->req.tp_block_size =
+	rx_frame_size * rx_frames_per_block;
+      rx_queue->rx_req->req.tp_frame_size = rx_frame_size;
+      rx_queue->rx_req->req.tp_block_nr = (apif->version == TPACKET_V3) ?
+						  AF_PACKET_RX_BLOCK_NR :
+						  AF_PACKET_RX_BLOCK_NR_V2;
+      rx_queue->rx_req->req.tp_frame_nr =
+	rx_queue->rx_req->req.tp_block_nr * rx_frames_per_block;
+      if (apif->version == TPACKET_V3)
+	{
+	  rx_queue->rx_req->req3.tp_retire_blk_tov = 1; // 1 ms block timout
+	  rx_queue->rx_req->req3.tp_feature_req_word = 0;
+	  rx_queue->rx_req->req3.tp_sizeof_priv = 0;
+	}
       rx_req = rx_queue->rx_req;
     }
-
   if (tx_queue)
     {
       tx_frames_per_block = arg->tx_frames_per_block ?
@@ -385,21 +406,26 @@ af_packet_queue_init (vlib_main_t *vm, af_packet_if_t *apif,
 						 AF_PACKET_DEFAULT_TX_FRAME_SIZE;
 
       vec_validate (tx_queue->tx_req, 0);
-      tx_queue->tx_req->tp_block_size = tx_frame_size * tx_frames_per_block;
-      tx_queue->tx_req->tp_frame_size = tx_frame_size;
-      tx_queue->tx_req->tp_block_nr = AF_PACKET_TX_BLOCK_NR;
-      tx_queue->tx_req->tp_frame_nr =
+      tx_queue->tx_req->req.tp_block_size =
+	tx_frame_size * tx_frames_per_block;
+      tx_queue->tx_req->req.tp_frame_size = tx_frame_size;
+      tx_queue->tx_req->req.tp_block_nr = AF_PACKET_TX_BLOCK_NR;
+      tx_queue->tx_req->req.tp_frame_nr =
 	AF_PACKET_TX_BLOCK_NR * tx_frames_per_block;
-      tx_queue->tx_req->tp_retire_blk_tov = 0;
-      tx_queue->tx_req->tp_sizeof_priv = 0;
-      tx_queue->tx_req->tp_feature_req_word = 0;
+      if (apif->version == TPACKET_V3)
+	{
+	  tx_queue->tx_req->req3.tp_retire_blk_tov = 0;
+	  tx_queue->tx_req->req3.tp_sizeof_priv = 0;
+	  tx_queue->tx_req->req3.tp_feature_req_word = 0;
+	}
       tx_req = tx_queue->tx_req;
     }
 
   if (rx_queue || tx_queue)
     {
-      ret = create_packet_v3_sock (apif->host_if_index, rx_req, tx_req, &fd,
-				   &ring, apif->dev_instance, &arg->flags);
+      ret =
+	create_packet_sock (apif->host_if_index, rx_req, tx_req, &fd, &ring,
+			    apif->dev_instance, &arg->flags, apif->version);
 
       if (ret != 0)
 	goto error;
@@ -411,28 +437,28 @@ af_packet_queue_init (vlib_main_t *vm, af_packet_if_t *apif,
   if (rx_queue)
     {
       rx_queue->fd = fd;
-      vec_validate (rx_queue->rx_ring, rx_queue->rx_req->tp_block_nr - 1);
+      vec_validate (rx_queue->rx_ring, rx_queue->rx_req->req.tp_block_nr - 1);
       vec_foreach_index (i, rx_queue->rx_ring)
 	{
 	  rx_queue->rx_ring[i] =
-	    ring_addr + i * rx_queue->rx_req->tp_block_size;
+	    ring_addr + i * rx_queue->rx_req->req.tp_block_size;
 	}
 
       rx_queue->next_rx_block = 0;
       rx_queue->queue_id = queue_id;
       rx_queue->is_rx_pending = 0;
-      ring_addr = ring_addr + rx_queue->rx_req->tp_block_size *
-				rx_queue->rx_req->tp_block_nr;
+      ring_addr = ring_addr + rx_queue->rx_req->req.tp_block_size *
+				rx_queue->rx_req->req.tp_block_nr;
     }
 
   if (tx_queue)
     {
       tx_queue->fd = fd;
-      vec_validate (tx_queue->tx_ring, tx_queue->tx_req->tp_block_nr - 1);
+      vec_validate (tx_queue->tx_ring, tx_queue->tx_req->req.tp_block_nr - 1);
       vec_foreach_index (i, tx_queue->tx_ring)
 	{
 	  tx_queue->tx_ring[i] =
-	    ring_addr + i * tx_queue->tx_req->tp_block_size;
+	    ring_addr + i * tx_queue->tx_req->req.tp_block_size;
 	}
 
       tx_queue->next_tx_frame = 0;
@@ -604,6 +630,11 @@ af_packet_create_if (af_packet_create_if_arg_t *arg)
   apif->per_interface_next_index = ~0;
   apif->mode = arg->mode;
 
+  if (arg->is_v2)
+    apif->version = TPACKET_V2;
+  else
+    apif->version = TPACKET_V3;
+
   ret = af_packet_device_init (vm, apif, arg);
   if (ret != 0)
     goto error;
diff --git a/src/vnet/devices/af_packet/af_packet.h b/src/vnet/devices/af_packet/af_packet.h
index 940acbb1372..7aa5e6d5b9a 100644
--- a/src/vnet/devices/af_packet/af_packet.h
+++ b/src/vnet/devices/af_packet/af_packet.h
@@ -23,9 +23,17 @@
 #include <vlib/log.h>
 
 typedef struct tpacket_block_desc block_desc_t;
+typedef struct tpacket_req tpacket_req_t;
 typedef struct tpacket_req3 tpacket_req3_t;
+typedef struct tpacket2_hdr tpacket2_hdr_t;
 typedef struct tpacket3_hdr tpacket3_hdr_t;
 
+typedef union _tpacket_req_u
+{
+  tpacket_req_t req;
+  tpacket_req3_t req3;
+} tpacket_req_u_t;
+
 typedef enum
 {
   AF_PACKET_IF_MODE_ETHERNET = 1,
@@ -37,6 +45,7 @@ typedef enum
   AF_PACKET_IF_FLAGS_QDISC_BYPASS = 1,
   AF_PACKET_IF_FLAGS_CKSUM_GSO = 2,
   AF_PACKET_IF_FLAGS_FANOUT = 4,
+  AF_PACKET_IF_FLAGS_VERSION_2 = 8,
 } af_packet_if_flags_t;
 
 typedef struct
@@ -58,8 +67,8 @@ typedef struct
   int fd;
   union
   {
-    tpacket_req3_t *rx_req;
-    tpacket_req3_t *tx_req;
+    tpacket_req_u_t *rx_req;
+    tpacket_req_u_t *tx_req;
   };
 
   union
@@ -71,6 +80,7 @@ typedef struct
   union
   {
     u32 next_rx_block;
+    u32 next_rx_frame;
     u32 next_tx_frame;
   };
 
@@ -95,7 +105,7 @@ typedef struct
   af_packet_if_mode_t mode;
   u8 is_admin_up;
   u8 is_cksum_gso_enabled;
-
+  u8 version;
   af_packet_queue_t *rx_queues;
   af_packet_queue_t *tx_queues;
 
@@ -139,6 +149,7 @@ typedef struct
   u32 tx_frames_per_block;
   u8 num_rxqs;
   u8 num_txqs;
+  u8 is_v2;
   af_packet_if_mode_t mode;
   af_packet_if_flags_t flags;
 
diff --git a/src/vnet/devices/af_packet/af_packet_api.c b/src/vnet/devices/af_packet/af_packet_api.c
index 21f2c381809..6ca79f9705e 100644
--- a/src/vnet/devices/af_packet/af_packet_api.c
+++ b/src/vnet/devices/af_packet/af_packet_api.c
@@ -135,6 +135,10 @@ vl_api_af_packet_create_v3_t_handler (vl_api_af_packet_create_v3_t *mp)
     ((int) AF_PACKET_API_FLAG_CKSUM_GSO == (int) AF_PACKET_IF_FLAGS_CKSUM_GSO),
     "af-packet checksum/gso offload api flag mismatch");
 
+  STATIC_ASSERT (
+    ((int) AF_PACKET_API_FLAG_VERSION_2 == (int) AF_PACKET_IF_FLAGS_VERSION_2),
+    "af-packet version 2 api flag mismatch");
+
   // Default flags
   arg->flags = clib_net_to_host_u32 (mp->flags);
 
@@ -144,6 +148,7 @@ vl_api_af_packet_create_v3_t_handler (vl_api_af_packet_create_v3_t *mp)
   if (clib_net_to_host_u16 (mp->num_tx_queues) > 1)
     arg->num_txqs = clib_net_to_host_u16 (mp->num_tx_queues);
 
+  arg->is_v2 = (arg->flags & AF_PACKET_API_FLAG_VERSION_2) ? 1 : 0;
   rv = af_packet_create_if (arg);
 
 error:
diff --git a/src/vnet/devices/af_packet/cli.c b/src/vnet/devices/af_packet/cli.c
index e730659bfcd..d20ef61a09d 100644
--- a/src/vnet/devices/af_packet/cli.c
+++ b/src/vnet/devices/af_packet/cli.c
@@ -89,6 +89,8 @@ af_packet_create_command_fn (vlib_main_t * vm, unformat_input_t * input,
 	arg->flags &= ~AF_PACKET_IF_FLAGS_CKSUM_GSO;
       else if (unformat (line_input, "mode ip"))
 	arg->mode = AF_PACKET_IF_MODE_IP;
+      else if (unformat (line_input, "v2"))
+	arg->is_v2 = 1;
       else if (unformat (line_input, "hw-addr %U", unformat_ethernet_address,
 			 hwaddr))
 	arg->hw_addr = hwaddr;
@@ -160,7 +162,7 @@ done:
 ?*/
 VLIB_CLI_COMMAND (af_packet_create_command, static) = {
   .path = "create host-interface",
-  .short_help = "create host-interface name <ifname> [num-rx-queues <n>] "
+  .short_help = "create host-interface [v2] name <ifname> [num-rx-queues <n>] "
 		"[num-tx-queues <n>] [hw-addr <mac-addr>] [mode ip] "
 		"[qdisc-bypass-disable] [cksum-gso-disable]",
   .function = af_packet_create_command_fn,
diff --git a/src/vnet/devices/af_packet/device.c b/src/vnet/devices/af_packet/device.c
index 1d14c9b8dcf..2e9b7a4ed83 100644
--- a/src/vnet/devices/af_packet/device.c
+++ b/src/vnet/devices/af_packet/device.c
@@ -61,7 +61,12 @@ typedef struct
   u32 buffer_index;
   u32 hw_if_index;
   u16 queue_id;
-  tpacket3_hdr_t tph;
+  u8 is_v2;
+  union
+  {
+    tpacket2_hdr_t tph2;
+    tpacket3_hdr_t tph3;
+  };
   vnet_virtio_net_hdr_t vnet_hdr;
   vlib_buffer_t buffer;
 } af_packet_tx_trace_t;
@@ -91,7 +96,8 @@ format_af_packet_device (u8 * s, va_list * args)
   af_packet_queue_t *rx_queue = 0;
   af_packet_queue_t *tx_queue = 0;
 
-  s = format (s, "Linux PACKET socket interface");
+  s = format (s, "Linux PACKET socket interface %s",
+	      (apif->version == TPACKET_V2) ? "v2" : "v3");
   s = format (s, "\n%UFEATURES:", format_white_space, indent);
   if (apif->is_qdisc_bypass_enabled)
     s = format (s, "\n%Uqdisc-bpass-enabled", format_white_space, indent + 2);
@@ -102,17 +108,20 @@ format_af_packet_device (u8 * s, va_list * args)
 
   vec_foreach (rx_queue, apif->rx_queues)
     {
-      u32 rx_block_size = rx_queue->rx_req->tp_block_size;
-      u32 rx_frame_size = rx_queue->rx_req->tp_frame_size;
-      u32 rx_frame_nr = rx_queue->rx_req->tp_frame_nr;
-      u32 rx_block_nr = rx_queue->rx_req->tp_block_nr;
+      u32 rx_block_size = rx_queue->rx_req->req.tp_block_size;
+      u32 rx_frame_size = rx_queue->rx_req->req.tp_frame_size;
+      u32 rx_frame_nr = rx_queue->rx_req->req.tp_frame_nr;
+      u32 rx_block_nr = rx_queue->rx_req->req.tp_block_nr;
 
       s = format (s, "\n%URX Queue %u:", format_white_space, indent,
 		  rx_queue->queue_id);
       s = format (s, "\n%Ublock size:%d nr:%d  frame size:%d nr:%d",
 		  format_white_space, indent + 2, rx_block_size, rx_block_nr,
 		  rx_frame_size, rx_frame_nr);
-      s = format (s, " next block:%d", rx_queue->next_rx_block);
+      if (apif->version == TPACKET_V2)
+	s = format (s, " next frame:%d", rx_queue->next_rx_frame);
+      else
+	s = format (s, " next block:%d", rx_queue->next_rx_block);
       if (rx_queue->is_rx_pending)
 	{
 	  s = format (
@@ -125,15 +134,16 @@ format_af_packet_device (u8 * s, va_list * args)
   vec_foreach (tx_queue, apif->tx_queues)
     {
       clib_spinlock_lock (&tx_queue->lockp);
-      u32 tx_block_sz = tx_queue->tx_req->tp_block_size;
-      u32 tx_frame_sz = tx_queue->tx_req->tp_frame_size;
-      u32 tx_frame_nr = tx_queue->tx_req->tp_frame_nr;
-      u32 tx_block_nr = tx_queue->tx_req->tp_block_nr;
+      u32 tx_block_sz = tx_queue->tx_req->req.tp_block_size;
+      u32 tx_frame_sz = tx_queue->tx_req->req.tp_frame_size;
+      u32 tx_frame_nr = tx_queue->tx_req->req.tp_frame_nr;
+      u32 tx_block_nr = tx_queue->tx_req->req.tp_block_nr;
       int block = 0;
       int n_send_req = 0, n_avail = 0, n_sending = 0, n_tot = 0, n_wrong = 0;
       u8 *tx_block_start = tx_queue->tx_ring[block];
       u32 tx_frame = tx_queue->next_tx_frame;
-      tpacket3_hdr_t *tph;
+      tpacket3_hdr_t *tph3;
+      tpacket2_hdr_t *tph2;
 
       s = format (s, "\n%UTX Queue %u:", format_white_space, indent,
 		  tx_queue->queue_id);
@@ -141,22 +151,40 @@ format_af_packet_device (u8 * s, va_list * args)
 		  format_white_space, indent + 2, tx_block_sz, tx_block_nr,
 		  tx_frame_sz, tx_frame_nr);
       s = format (s, " next frame:%d", tx_queue->next_tx_frame);
-
-      do
-	{
-	  tph = (tpacket3_hdr_t *) (tx_block_start + tx_frame * tx_frame_sz);
-	  tx_frame = (tx_frame + 1) % tx_frame_nr;
-	  if (tph->tp_status == 0)
-	    n_avail++;
-	  else if (tph->tp_status & TP_STATUS_SEND_REQUEST)
-	    n_send_req++;
-	  else if (tph->tp_status & TP_STATUS_SENDING)
-	    n_sending++;
-	  else
-	    n_wrong++;
-	  n_tot++;
-	}
-      while (tx_frame != tx_queue->next_tx_frame);
+      if (apif->version & TPACKET_V3)
+	do
+	  {
+	    tph3 =
+	      (tpacket3_hdr_t *) (tx_block_start + tx_frame * tx_frame_sz);
+	    tx_frame = (tx_frame + 1) % tx_frame_nr;
+	    if (tph3->tp_status == 0)
+	      n_avail++;
+	    else if (tph3->tp_status & TP_STATUS_SEND_REQUEST)
+	      n_send_req++;
+	    else if (tph3->tp_status & TP_STATUS_SENDING)
+	      n_sending++;
+	    else
+	      n_wrong++;
+	    n_tot++;
+	  }
+	while (tx_frame != tx_queue->next_tx_frame);
+      else
+	do
+	  {
+	    tph2 =
+	      (tpacket2_hdr_t *) (tx_block_start + tx_frame * tx_frame_sz);
+	    tx_frame = (tx_frame + 1) % tx_frame_nr;
+	    if (tph2->tp_status == 0)
+	      n_avail++;
+	    else if (tph2->tp_status & TP_STATUS_SEND_REQUEST)
+	      n_send_req++;
+	    else if (tph2->tp_status & TP_STATUS_SENDING)
+	      n_sending++;
+	    else
+	      n_wrong++;
+	    n_tot++;
+	  }
+	while (tx_frame != tx_queue->next_tx_frame);
       s =
 	format (s, "\n%Uavailable:%d request:%d sending:%d wrong:%d total:%d",
 		format_white_space, indent + 2, n_avail, n_send_req, n_sending,
@@ -177,24 +205,46 @@ format_af_packet_tx_trace (u8 *s, va_list *va)
   s = format (s, "af_packet: hw_if_index %u tx-queue %u", t->hw_if_index,
 	      t->queue_id);
 
-  s =
-    format (s,
-	    "\n%Utpacket3_hdr:\n%Ustatus 0x%x len %u snaplen %u mac %u net %u"
-	    "\n%Usec 0x%x nsec 0x%x vlan %U"
+  if (t->is_v2)
+    {
+      s = format (
+	s,
+	"\n%Utpacket2_hdr:\n%Ustatus 0x%x len %u snaplen %u mac %u net %u"
+	"\n%Usec 0x%x nsec 0x%x vlan %U"
 #ifdef TP_STATUS_VLAN_TPID_VALID
-	    " vlan_tpid %u"
+	" vlan_tpid %u"
 #endif
-	    ,
-	    format_white_space, indent + 2, format_white_space, indent + 4,
-	    t->tph.tp_status, t->tph.tp_len, t->tph.tp_snaplen, t->tph.tp_mac,
-	    t->tph.tp_net, format_white_space, indent + 4, t->tph.tp_sec,
-	    t->tph.tp_nsec, format_ethernet_vlan_tci, t->tph.hv1.tp_vlan_tci
+	,
+	format_white_space, indent + 2, format_white_space, indent + 4,
+	t->tph2.tp_status, t->tph2.tp_len, t->tph2.tp_snaplen, t->tph2.tp_mac,
+	t->tph2.tp_net, format_white_space, indent + 4, t->tph2.tp_sec,
+	t->tph2.tp_nsec, format_ethernet_vlan_tci, t->tph2.tp_vlan_tci
 #ifdef TP_STATUS_VLAN_TPID_VALID
-	    ,
-	    t->tph.hv1.tp_vlan_tpid
+	,
+	t->tph2.tp_vlan_tpid
 #endif
-    );
-
+      );
+    }
+  else
+    {
+      s = format (
+	s,
+	"\n%Utpacket3_hdr:\n%Ustatus 0x%x len %u snaplen %u mac %u net %u"
+	"\n%Usec 0x%x nsec 0x%x vlan %U"
+#ifdef TP_STATUS_VLAN_TPID_VALID
+	" vlan_tpid %u"
+#endif
+	,
+	format_white_space, indent + 2, format_white_space, indent + 4,
+	t->tph3.tp_status, t->tph3.tp_len, t->tph3.tp_snaplen, t->tph3.tp_mac,
+	t->tph3.tp_net, format_white_space, indent + 4, t->tph3.tp_sec,
+	t->tph3.tp_nsec, format_ethernet_vlan_tci, t->tph3.hv1.tp_vlan_tci
+#ifdef TP_STATUS_VLAN_TPID_VALID
+	,
+	t->tph3.hv1.tp_vlan_tpid
+#endif
+      );
+    }
   s = format (s,
 	      "\n%Uvnet-hdr:\n%Uflags 0x%02x gso_type 0x%02x hdr_len %u"
 	      "\n%Ugso_size %u csum_start %u csum_offset %u",
@@ -214,17 +264,23 @@ format_af_packet_tx_trace (u8 *s, va_list *va)
 
 static void
 af_packet_tx_trace (vlib_main_t *vm, vlib_node_runtime_t *node,
-		    vlib_buffer_t *b0, u32 bi, tpacket3_hdr_t *tph,
+		    vlib_buffer_t *b0, u32 bi, void *tph,
 		    vnet_virtio_net_hdr_t *vnet_hdr, u32 hw_if_index,
-		    u16 queue_id)
+		    u16 queue_id, u8 is_v2)
 {
   af_packet_tx_trace_t *t;
   t = vlib_add_trace (vm, node, b0, sizeof (t[0]));
   t->hw_if_index = hw_if_index;
   t->queue_id = queue_id;
   t->buffer_index = bi;
+  t->is_v2 = is_v2;
 
-  clib_memcpy_fast (&t->tph, tph, sizeof (*tph));
+  if (is_v2)
+    clib_memcpy_fast (&t->tph2, (tpacket2_hdr_t *) tph,
+		      sizeof (tpacket2_hdr_t));
+  else
+    clib_memcpy_fast (&t->tph3, (tpacket3_hdr_t *) tph,
+		      sizeof (tpacket3_hdr_t));
   clib_memcpy_fast (&t->vnet_hdr, vnet_hdr, sizeof (*vnet_hdr));
   clib_memcpy_fast (&t->buffer, b0, sizeof (*b0) - sizeof (b0->pre_data));
   clib_memcpy_fast (t->buffer.pre_data, vlib_buffer_get_current (b0),
@@ -337,95 +393,177 @@ VNET_DEVICE_CLASS_TX_FN (af_packet_device_class) (vlib_main_t * vm,
   af_packet_queue_t *tx_queue = vec_elt_at_index (apif->tx_queues, queue_id);
   u32 block = 0, frame_size = 0, frame_num = 0, tx_frame = 0;
   u8 *block_start = 0;
-  tpacket3_hdr_t *tph = 0;
+  tpacket3_hdr_t *tph3 = 0;
+  tpacket2_hdr_t *tph2 = 0;
   u32 frame_not_ready = 0;
   u8 is_cksum_gso_enabled = (apif->is_cksum_gso_enabled == 1) ? 1 : 0;
+  u32 tpacket_align = 0;
+  u8 is_v2 = (apif->version == TPACKET_V2) ? 1 : 0;
 
   if (tf->shared_queue)
     clib_spinlock_lock (&tx_queue->lockp);
 
-  frame_size = tx_queue->tx_req->tp_frame_size;
-  frame_num = tx_queue->tx_req->tp_frame_nr;
+  frame_size = tx_queue->tx_req->req.tp_frame_size;
+  frame_num = tx_queue->tx_req->req.tp_frame_nr;
   block_start = tx_queue->tx_ring[block];
   tx_frame = tx_queue->next_tx_frame;
-
-  while (n_left)
+  if (is_v2)
     {
-      u32 len;
-      vnet_virtio_net_hdr_t *vnet_hdr = 0;
-      u32 offset = 0;
-      vlib_buffer_t *b0 = 0, *b0_first = 0;
-      u32 bi, bi_first;
-
-      bi = bi_first = buffers[0];
-      n_left--;
-      buffers++;
-
-      tph = (tpacket3_hdr_t *) (block_start + tx_frame * frame_size);
-      if (PREDICT_FALSE (tph->tp_status &
-			 (TP_STATUS_SEND_REQUEST | TP_STATUS_SENDING)))
+      tpacket_align = TPACKET_ALIGN (sizeof (tpacket2_hdr_t));
+      while (n_left)
 	{
-	  frame_not_ready++;
-	  goto next;
-	}
+	  u32 len;
+	  vnet_virtio_net_hdr_t *vnet_hdr = 0;
+	  u32 offset = 0;
+	  vlib_buffer_t *b0 = 0, *b0_first = 0;
+	  u32 bi, bi_first;
+
+	  bi = bi_first = buffers[0];
+	  n_left--;
+	  buffers++;
+
+	  tph2 = (tpacket2_hdr_t *) (block_start + tx_frame * frame_size);
+	  if (PREDICT_FALSE (tph2->tp_status &
+			     (TP_STATUS_SEND_REQUEST | TP_STATUS_SENDING)))
+	    {
+	      frame_not_ready++;
+	      goto nextv2;
+	    }
 
-      b0_first = b0 = vlib_get_buffer (vm, bi);
+	  b0_first = b0 = vlib_get_buffer (vm, bi);
 
-      if (PREDICT_TRUE (is_cksum_gso_enabled))
-	{
-	  vnet_hdr =
-	    (vnet_virtio_net_hdr_t *) ((u8 *) tph + TPACKET_ALIGN (sizeof (
-						      tpacket3_hdr_t)));
+	  if (PREDICT_TRUE (is_cksum_gso_enabled))
+	    {
+	      vnet_hdr =
+		(vnet_virtio_net_hdr_t *) ((u8 *) tph2 + tpacket_align);
 
-	  clib_memset_u8 (vnet_hdr, 0, sizeof (vnet_virtio_net_hdr_t));
-	  offset = sizeof (vnet_virtio_net_hdr_t);
+	      clib_memset_u8 (vnet_hdr, 0, sizeof (vnet_virtio_net_hdr_t));
+	      offset = sizeof (vnet_virtio_net_hdr_t);
 
-	  if (b0->flags & VNET_BUFFER_F_GSO)
-	    fill_gso_offload (b0, vnet_hdr);
-	  else if (b0->flags & VNET_BUFFER_F_OFFLOAD)
-	    fill_cksum_offload (b0, vnet_hdr);
-	}
-
-      len = b0->current_length;
-      clib_memcpy_fast ((u8 *) tph + TPACKET_ALIGN (sizeof (tpacket3_hdr_t)) +
-			  offset,
-			vlib_buffer_get_current (b0), len);
-      offset += len;
+	      if (b0->flags & VNET_BUFFER_F_GSO)
+		fill_gso_offload (b0, vnet_hdr);
+	      else if (b0->flags & VNET_BUFFER_F_OFFLOAD)
+		fill_cksum_offload (b0, vnet_hdr);
+	    }
 
-      while (b0->flags & VLIB_BUFFER_NEXT_PRESENT)
-	{
-	  b0 = vlib_get_buffer (vm, b0->next_buffer);
 	  len = b0->current_length;
-	  clib_memcpy_fast ((u8 *) tph +
-			      TPACKET_ALIGN (sizeof (tpacket3_hdr_t)) + offset,
+	  clib_memcpy_fast ((u8 *) tph2 + tpacket_align + offset,
 			    vlib_buffer_get_current (b0), len);
 	  offset += len;
-	}
 
-      tph->tp_len = tph->tp_snaplen = offset;
-      tph->tp_status = TP_STATUS_SEND_REQUEST;
-      n_sent++;
+	  while (b0->flags & VLIB_BUFFER_NEXT_PRESENT)
+	    {
+	      b0 = vlib_get_buffer (vm, b0->next_buffer);
+	      len = b0->current_length;
+	      clib_memcpy_fast ((u8 *) tph2 + tpacket_align + offset,
+				vlib_buffer_get_current (b0), len);
+	      offset += len;
+	    }
+
+	  tph2->tp_len = tph2->tp_snaplen = offset;
+	  tph2->tp_status = TP_STATUS_SEND_REQUEST;
+	  n_sent++;
 
-      if (PREDICT_FALSE (b0_first->flags & VLIB_BUFFER_IS_TRACED))
+	  if (PREDICT_FALSE (b0_first->flags & VLIB_BUFFER_IS_TRACED))
+	    {
+	      if (PREDICT_TRUE (is_cksum_gso_enabled))
+		af_packet_tx_trace (vm, node, b0_first, bi_first, tph2,
+				    vnet_hdr, apif->hw_if_index, queue_id, 1);
+	      else
+		{
+		  vnet_virtio_net_hdr_t vnet_hdr2 = {};
+		  af_packet_tx_trace (vm, node, b0_first, bi_first, tph2,
+				      &vnet_hdr2, apif->hw_if_index, queue_id,
+				      1);
+		}
+	    }
+	  tx_frame = (tx_frame + 1) % frame_num;
+
+	nextv2:
+	  /* check if we've exhausted the ring */
+	  if (PREDICT_FALSE (frame_not_ready + n_sent == frame_num))
+	    break;
+	}
+    }
+  else
+    {
+      tpacket_align = TPACKET_ALIGN (sizeof (tpacket3_hdr_t));
+
+      while (n_left)
 	{
+	  u32 len;
+	  vnet_virtio_net_hdr_t *vnet_hdr = 0;
+	  u32 offset = 0;
+	  vlib_buffer_t *b0 = 0, *b0_first = 0;
+	  u32 bi, bi_first;
+
+	  bi = bi_first = buffers[0];
+	  n_left--;
+	  buffers++;
+
+	  tph3 = (tpacket3_hdr_t *) (block_start + tx_frame * frame_size);
+	  if (PREDICT_FALSE (tph3->tp_status &
+			     (TP_STATUS_SEND_REQUEST | TP_STATUS_SENDING)))
+	    {
+	      frame_not_ready++;
+	      goto nextv3;
+	    }
+
+	  b0_first = b0 = vlib_get_buffer (vm, bi);
+
 	  if (PREDICT_TRUE (is_cksum_gso_enabled))
-	    af_packet_tx_trace (vm, node, b0_first, bi_first, tph, vnet_hdr,
-				apif->hw_if_index, queue_id);
-	  else
 	    {
-	      vnet_virtio_net_hdr_t vnet_hdr2 = {};
-	      af_packet_tx_trace (vm, node, b0_first, bi_first, tph,
-				  &vnet_hdr2, apif->hw_if_index, queue_id);
+	      vnet_hdr =
+		(vnet_virtio_net_hdr_t *) ((u8 *) tph3 + tpacket_align);
+
+	      clib_memset_u8 (vnet_hdr, 0, sizeof (vnet_virtio_net_hdr_t));
+	      offset = sizeof (vnet_virtio_net_hdr_t);
+
+	      if (b0->flags & VNET_BUFFER_F_GSO)
+		fill_gso_offload (b0, vnet_hdr);
+	      else if (b0->flags & VNET_BUFFER_F_OFFLOAD)
+		fill_cksum_offload (b0, vnet_hdr);
 	    }
-	}
-      tx_frame = (tx_frame + 1) % frame_num;
 
-    next:
-      /* check if we've exhausted the ring */
-      if (PREDICT_FALSE (frame_not_ready + n_sent == frame_num))
-	break;
-    }
+	  len = b0->current_length;
+	  clib_memcpy_fast ((u8 *) tph3 + tpacket_align + offset,
+			    vlib_buffer_get_current (b0), len);
+	  offset += len;
+
+	  while (b0->flags & VLIB_BUFFER_NEXT_PRESENT)
+	    {
+	      b0 = vlib_get_buffer (vm, b0->next_buffer);
+	      len = b0->current_length;
+	      clib_memcpy_fast ((u8 *) tph3 + tpacket_align + offset,
+				vlib_buffer_get_current (b0), len);
+	      offset += len;
+	    }
+
+	  tph3->tp_len = tph3->tp_snaplen = offset;
+	  tph3->tp_status = TP_STATUS_SEND_REQUEST;
+	  n_sent++;
 
+	  if (PREDICT_FALSE (b0_first->flags & VLIB_BUFFER_IS_TRACED))
+	    {
+	      if (PREDICT_TRUE (is_cksum_gso_enabled))
+		af_packet_tx_trace (vm, node, b0_first, bi_first, tph3,
+				    vnet_hdr, apif->hw_if_index, queue_id, 0);
+	      else
+		{
+		  vnet_virtio_net_hdr_t vnet_hdr2 = {};
+		  af_packet_tx_trace (vm, node, b0_first, bi_first, tph3,
+				      &vnet_hdr2, apif->hw_if_index, queue_id,
+				      0);
+		}
+	    }
+	  tx_frame = (tx_frame + 1) % frame_num;
+
+	nextv3:
+	  /* check if we've exhausted the ring */
+	  if (PREDICT_FALSE (frame_not_ready + n_sent == frame_num))
+	    break;
+	}
+    }
   CLIB_MEMORY_BARRIER ();
 
   if (PREDICT_TRUE (n_sent || tx_queue->is_tx_pending))
diff --git a/src/vnet/devices/af_packet/node.c b/src/vnet/devices/af_packet/node.c
index 8c72afb2456..724924f5f4b 100644
--- a/src/vnet/devices/af_packet/node.c
+++ b/src/vnet/devices/af_packet/node.c
@@ -57,8 +57,13 @@ typedef struct
   u32 pkt_num;
   void *block_start;
   block_desc_t bd;
-  tpacket3_hdr_t tph;
+  union
+  {
+    tpacket3_hdr_t tph3;
+    tpacket2_hdr_t tph2;
+  };
   vnet_virtio_net_hdr_t vnet_hdr;
+  u8 is_v3;
 } af_packet_input_trace_t;
 
 static u8 *
@@ -72,27 +77,51 @@ format_af_packet_input_trace (u8 * s, va_list * args)
   s = format (s, "af_packet: hw_if_index %d rx-queue %u next-index %d",
 	      t->hw_if_index, t->queue_id, t->next_index);
 
-  s = format (
-    s, "\n%Ublock %u:\n%Uaddress %p version %u seq_num %lu pkt_num %u",
-    format_white_space, indent + 2, t->block, format_white_space, indent + 4,
-    t->block_start, t->bd.version, t->bd.hdr.bh1.seq_num, t->pkt_num);
-  s =
-    format (s,
-	    "\n%Utpacket3_hdr:\n%Ustatus 0x%x len %u snaplen %u mac %u net %u"
-	    "\n%Usec 0x%x nsec 0x%x vlan %U"
+  if (t->is_v3)
+    {
+      s = format (
+	s, "\n%Ublock %u:\n%Uaddress %p version %u seq_num %lu pkt_num %u",
+	format_white_space, indent + 2, t->block, format_white_space,
+	indent + 4, t->block_start, t->bd.version, t->bd.hdr.bh1.seq_num,
+	t->pkt_num);
+      s = format (
+	s,
+	"\n%Utpacket3_hdr:\n%Ustatus 0x%x len %u snaplen %u mac %u net %u"
+	"\n%Usec 0x%x nsec 0x%x vlan %U"
 #ifdef TP_STATUS_VLAN_TPID_VALID
-	    " vlan_tpid %u"
+	" vlan_tpid %u"
 #endif
-	    ,
-	    format_white_space, indent + 2, format_white_space, indent + 4,
-	    t->tph.tp_status, t->tph.tp_len, t->tph.tp_snaplen, t->tph.tp_mac,
-	    t->tph.tp_net, format_white_space, indent + 4, t->tph.tp_sec,
-	    t->tph.tp_nsec, format_ethernet_vlan_tci, t->tph.hv1.tp_vlan_tci
+	,
+	format_white_space, indent + 2, format_white_space, indent + 4,
+	t->tph3.tp_status, t->tph3.tp_len, t->tph3.tp_snaplen, t->tph3.tp_mac,
+	t->tph3.tp_net, format_white_space, indent + 4, t->tph3.tp_sec,
+	t->tph3.tp_nsec, format_ethernet_vlan_tci, t->tph3.hv1.tp_vlan_tci
 #ifdef TP_STATUS_VLAN_TPID_VALID
-	    ,
-	    t->tph.hv1.tp_vlan_tpid
+	,
+	t->tph3.hv1.tp_vlan_tpid
 #endif
-    );
+      );
+    }
+  else
+    {
+      s = format (
+	s,
+	"\n%Utpacket2_hdr:\n%Ustatus 0x%x len %u snaplen %u mac %u net %u"
+	"\n%Usec 0x%x nsec 0x%x vlan %U"
+#ifdef TP_STATUS_VLAN_TPID_VALID
+	" vlan_tpid %u"
+#endif
+	,
+	format_white_space, indent + 2, format_white_space, indent + 4,
+	t->tph2.tp_status, t->tph2.tp_len, t->tph2.tp_snaplen, t->tph2.tp_mac,
+	t->tph2.tp_net, format_white_space, indent + 4, t->tph2.tp_sec,
+	t->tph2.tp_nsec, format_ethernet_vlan_tci, t->tph2.tp_vlan_tci
+#ifdef TP_STATUS_VLAN_TPID_VALID
+	,
+	t->tph2.tp_vlan_tpid
+#endif
+      );
+    }
 
   s = format (s,
 	      "\n%Uvnet-hdr:\n%Uflags 0x%02x gso_type 0x%02x hdr_len %u"
@@ -222,9 +251,9 @@ fill_cksum_offload (vlib_buffer_t *b, u8 *l4_hdr_sz, u8 is_ip)
 }
 
 always_inline uword
-af_packet_device_input_fn (vlib_main_t *vm, vlib_node_runtime_t *node,
-			   vlib_frame_t *frame, af_packet_if_t *apif,
-			   u16 queue_id, u8 is_cksum_gso_enabled)
+af_packet_v3_device_input_fn (vlib_main_t *vm, vlib_node_runtime_t *node,
+			      vlib_frame_t *frame, af_packet_if_t *apif,
+			      u16 queue_id, u8 is_cksum_gso_enabled)
 {
   af_packet_main_t *apm = &af_packet_main;
   af_packet_queue_t *rx_queue = vec_elt_at_index (apif->rx_queues, queue_id);
@@ -237,12 +266,12 @@ af_packet_device_input_fn (vlib_main_t *vm, vlib_node_runtime_t *node,
   u32 total = 0;
   u32 *to_next = 0;
   u32 block = rx_queue->next_rx_block;
-  u32 block_nr = rx_queue->rx_req->tp_block_nr;
+  u32 block_nr = rx_queue->rx_req->req3.tp_block_nr;
   u8 *block_start = 0;
   uword n_trace = vlib_get_trace_count (vm, node);
   u32 thread_index = vm->thread_index;
   u32 n_buffer_bytes = vlib_buffer_get_default_data_size (vm);
-  u32 min_bufs = rx_queue->rx_req->tp_frame_size / n_buffer_bytes;
+  u32 min_bufs = rx_queue->rx_req->req3.tp_frame_size / n_buffer_bytes;
   u32 num_pkts = 0;
   u32 rx_frame_offset = 0;
   block_desc_t *bd = 0;
@@ -458,6 +487,7 @@ af_packet_device_input_fn (vlib_main_t *vm, vlib_node_runtime_t *node,
 		  af_packet_input_trace_t *tr;
 		  vlib_set_trace_count (vm, node, --n_trace);
 		  tr = vlib_add_trace (vm, node, first_b0, sizeof (*tr));
+		  tr->is_v3 = 1;
 		  tr->next_index = next0;
 		  tr->hw_if_index = apif->hw_if_index;
 		  tr->queue_id = queue_id;
@@ -465,7 +495,7 @@ af_packet_device_input_fn (vlib_main_t *vm, vlib_node_runtime_t *node,
 		  tr->block_start = bd;
 		  tr->pkt_num = bd->hdr.bh1.num_pkts - num_pkts;
 		  clib_memcpy_fast (&tr->bd, bd, sizeof (block_desc_t));
-		  clib_memcpy_fast (&tr->tph, tph, sizeof (tpacket3_hdr_t));
+		  clib_memcpy_fast (&tr->tph3, tph, sizeof (tpacket3_hdr_t));
 		  if (is_cksum_gso_enabled)
 		    clib_memcpy_fast (&tr->vnet_hdr, vnet_hdr,
 				      sizeof (vnet_virtio_net_hdr_t));
@@ -528,6 +558,247 @@ done:
   return n_rx_packets;
 }
 
+always_inline uword
+af_packet_v2_device_input_fn (vlib_main_t *vm, vlib_node_runtime_t *node,
+			      vlib_frame_t *frame, af_packet_if_t *apif,
+			      u16 queue_id, u8 is_cksum_gso_enabled)
+{
+  af_packet_main_t *apm = &af_packet_main;
+  af_packet_queue_t *rx_queue = vec_elt_at_index (apif->rx_queues, queue_id);
+  tpacket2_hdr_t *tph;
+  u32 next_index;
+  u32 block = 0;
+  u32 rx_frame;
+  u32 n_free_bufs;
+  u32 n_rx_packets = 0;
+  u32 n_rx_bytes = 0;
+  u32 *to_next = 0;
+  u32 frame_size = rx_queue->rx_req->req.tp_frame_size;
+  u32 frame_num = rx_queue->rx_req->req.tp_frame_nr;
+  u8 *block_start = rx_queue->rx_ring[block];
+  uword n_trace = vlib_get_trace_count (vm, node);
+  u32 thread_index = vm->thread_index;
+  u32 n_buffer_bytes = vlib_buffer_get_default_data_size (vm);
+  u32 min_bufs = rx_queue->rx_req->req.tp_frame_size / n_buffer_bytes;
+  u8 is_ip = (apif->mode == AF_PACKET_IF_MODE_IP);
+  vlib_buffer_t bt = {};
+
+  if (is_ip)
+    {
+      next_index = VNET_DEVICE_INPUT_NEXT_IP4_INPUT;
+    }
+  else
+    {
+      next_index = VNET_DEVICE_INPUT_NEXT_ETHERNET_INPUT;
+      if (PREDICT_FALSE (apif->per_interface_next_index != ~0))
+	next_index = apif->per_interface_next_index;
+
+      /* redirect if feature path enabled */
+      vnet_feature_start_device_input_x1 (apif->sw_if_index, &next_index, &bt);
+    }
+
+  n_free_bufs = vec_len (apm->rx_buffers[thread_index]);
+  if (PREDICT_FALSE (n_free_bufs < VLIB_FRAME_SIZE))
+    {
+      vec_validate (apm->rx_buffers[thread_index],
+		    VLIB_FRAME_SIZE + n_free_bufs - 1);
+      n_free_bufs += vlib_buffer_alloc (
+	vm, &apm->rx_buffers[thread_index][n_free_bufs], VLIB_FRAME_SIZE);
+      vec_set_len (apm->rx_buffers[thread_index], n_free_bufs);
+    }
+
+  rx_frame = rx_queue->next_rx_frame;
+  tph = (tpacket2_hdr_t *) (block_start + rx_frame * frame_size);
+  while ((tph->tp_status & TP_STATUS_USER) && (n_free_bufs > min_bufs))
+    {
+      vlib_buffer_t *b0 = 0, *first_b0 = 0, *prev_b0 = 0;
+      u32 next0 = next_index;
+
+      u32 n_left_to_next;
+      vlib_get_next_frame (vm, node, next_index, to_next, n_left_to_next);
+      while ((tph->tp_status & TP_STATUS_USER) && (n_free_bufs > min_bufs) &&
+	     n_left_to_next)
+	{
+	  vnet_virtio_net_hdr_t *vnet_hdr = 0;
+	  u32 data_len = tph->tp_snaplen;
+	  u32 offset = 0;
+	  u32 bi0 = 0, first_bi0 = 0;
+	  u8 l4_hdr_sz = 0;
+
+	  if (is_cksum_gso_enabled)
+	    vnet_hdr =
+	      (vnet_virtio_net_hdr_t *) ((u8 *) tph + tph->tp_mac -
+					 sizeof (vnet_virtio_net_hdr_t));
+	  while (data_len)
+	    {
+	      /* grab free buffer */
+	      u32 last_empty_buffer =
+		vec_len (apm->rx_buffers[thread_index]) - 1;
+	      bi0 = apm->rx_buffers[thread_index][last_empty_buffer];
+	      b0 = vlib_get_buffer (vm, bi0);
+	      vec_set_len (apm->rx_buffers[thread_index], last_empty_buffer);
+	      n_free_bufs--;
+
+	      /* copy data */
+	      u32 bytes_to_copy =
+		data_len > n_buffer_bytes ? n_buffer_bytes : data_len;
+	      u32 vlan_len = 0;
+	      u32 bytes_copied = 0;
+	      b0->current_data = 0;
+	      /* Kernel removes VLAN headers, so reconstruct VLAN */
+	      if (PREDICT_FALSE (tph->tp_status & TP_STATUS_VLAN_VALID))
+		{
+		  if (PREDICT_TRUE (offset == 0))
+		    {
+		      clib_memcpy_fast (vlib_buffer_get_current (b0),
+					(u8 *) tph + tph->tp_mac,
+					sizeof (ethernet_header_t));
+		      ethernet_header_t *eth = vlib_buffer_get_current (b0);
+		      ethernet_vlan_header_t *vlan =
+			(ethernet_vlan_header_t *) (eth + 1);
+		      vlan->priority_cfi_and_id =
+			clib_host_to_net_u16 (tph->tp_vlan_tci);
+		      vlan->type = eth->type;
+		      eth->type = clib_host_to_net_u16 (ETHERNET_TYPE_VLAN);
+		      vlan_len = sizeof (ethernet_vlan_header_t);
+		      bytes_copied = sizeof (ethernet_header_t);
+		    }
+		}
+	      clib_memcpy_fast (((u8 *) vlib_buffer_get_current (b0)) +
+				  bytes_copied + vlan_len,
+				(u8 *) tph + tph->tp_mac + offset +
+				  bytes_copied,
+				(bytes_to_copy - bytes_copied));
+
+	      /* fill buffer header */
+	      b0->current_length = bytes_to_copy + vlan_len;
+
+	      if (offset == 0)
+		{
+		  b0->total_length_not_including_first_buffer = 0;
+		  b0->flags = VLIB_BUFFER_TOTAL_LENGTH_VALID;
+		  vnet_buffer (b0)->sw_if_index[VLIB_RX] = apif->sw_if_index;
+		  vnet_buffer (b0)->sw_if_index[VLIB_TX] = (u32) ~0;
+		  first_bi0 = bi0;
+		  first_b0 = vlib_get_buffer (vm, first_bi0);
+
+		  if (is_cksum_gso_enabled)
+		    {
+		      if (vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM)
+			fill_cksum_offload (first_b0, &l4_hdr_sz, is_ip);
+		      if (vnet_hdr->gso_type & (VIRTIO_NET_HDR_GSO_TCPV4 |
+						VIRTIO_NET_HDR_GSO_TCPV6))
+			fill_gso_offload (first_b0, vnet_hdr->gso_size,
+					  l4_hdr_sz);
+		    }
+		}
+	      else
+		buffer_add_to_chain (b0, first_b0, prev_b0, bi0);
+
+	      prev_b0 = b0;
+	      offset += bytes_to_copy;
+	      data_len -= bytes_to_copy;
+	    }
+	  n_rx_packets++;
+	  n_rx_bytes += tph->tp_snaplen;
+	  to_next[0] = first_bi0;
+	  to_next += 1;
+	  n_left_to_next--;
+
+	  /* drop partial packets */
+	  if (PREDICT_FALSE (tph->tp_len != tph->tp_snaplen))
+	    {
+	      next0 = VNET_DEVICE_INPUT_NEXT_DROP;
+	      first_b0->error =
+		node->errors[AF_PACKET_INPUT_ERROR_PARTIAL_PKT];
+	    }
+	  else
+	    {
+	      if (PREDICT_FALSE (is_ip))
+		{
+		  switch (first_b0->data[0] & 0xf0)
+		    {
+		    case 0x40:
+		      next0 = VNET_DEVICE_INPUT_NEXT_IP4_INPUT;
+		      break;
+		    case 0x60:
+		      next0 = VNET_DEVICE_INPUT_NEXT_IP6_INPUT;
+		      break;
+		    default:
+		      next0 = VNET_DEVICE_INPUT_NEXT_DROP;
+		      break;
+		    }
+		  if (PREDICT_FALSE (apif->per_interface_next_index != ~0))
+		    next0 = apif->per_interface_next_index;
+		}
+	      else
+		{
+		  /* copy feature arc data from template */
+		  first_b0->current_config_index = bt.current_config_index;
+		  vnet_buffer (first_b0)->feature_arc_index =
+		    vnet_buffer (&bt)->feature_arc_index;
+		}
+	    }
+
+	  /* trace */
+	  if (PREDICT_FALSE (n_trace > 0 &&
+			     vlib_trace_buffer (vm, node, next0, first_b0,
+						/* follow_chain */ 0)))
+	    {
+	      af_packet_input_trace_t *tr;
+	      vlib_set_trace_count (vm, node, --n_trace);
+	      tr = vlib_add_trace (vm, node, first_b0, sizeof (*tr));
+	      tr->is_v3 = 0;
+	      tr->next_index = next0;
+	      tr->hw_if_index = apif->hw_if_index;
+	      tr->queue_id = queue_id;
+	      clib_memcpy_fast (&tr->tph2, tph, sizeof (struct tpacket2_hdr));
+	      if (is_cksum_gso_enabled)
+		clib_memcpy_fast (&tr->vnet_hdr, vnet_hdr,
+				  sizeof (vnet_virtio_net_hdr_t));
+	      else
+		clib_memset_u8 (&tr->vnet_hdr, 0,
+				sizeof (vnet_virtio_net_hdr_t));
+	    }
+
+	  /* enque and take next packet */
+	  vlib_validate_buffer_enqueue_x1 (vm, node, next_index, to_next,
+					   n_left_to_next, first_bi0, next0);
+
+	  /* next packet */
+	  tph->tp_status = TP_STATUS_KERNEL;
+	  rx_frame = (rx_frame + 1) % frame_num;
+	  tph = (struct tpacket2_hdr *) (block_start + rx_frame * frame_size);
+	}
+
+      vlib_put_next_frame (vm, node, next_index, n_left_to_next);
+    }
+
+  rx_queue->next_rx_frame = rx_frame;
+
+  vlib_increment_combined_counter (
+    vnet_get_main ()->interface_main.combined_sw_if_counters +
+      VNET_INTERFACE_COUNTER_RX,
+    vlib_get_thread_index (), apif->hw_if_index, n_rx_packets, n_rx_bytes);
+
+  vnet_device_increment_rx_packets (thread_index, n_rx_packets);
+  return n_rx_packets;
+}
+
+always_inline uword
+af_packet_device_input_fn (vlib_main_t *vm, vlib_node_runtime_t *node,
+			   vlib_frame_t *frame, af_packet_if_t *apif,
+			   u16 queue_id, u8 is_cksum_gso_enabled)
+
+{
+  if (apif->version == TPACKET_V3)
+    return af_packet_v3_device_input_fn (vm, node, frame, apif, queue_id,
+					 is_cksum_gso_enabled);
+  else
+    return af_packet_v2_device_input_fn (vm, node, frame, apif, queue_id,
+					 is_cksum_gso_enabled);
+}
+
 VLIB_NODE_FN (af_packet_input_node) (vlib_main_t * vm,
 				     vlib_node_runtime_t * node,
 				     vlib_frame_t * frame)