tap: add support to configure tap interface host MTU size

This patch adds support to configure host mtu size using
api, cli or startup.conf.

Type: feature

Change-Id: I8ab087d82dbe7dedc498825c1a3ea3fcb2cce030
Signed-off-by: Mohsin Kazmi <sykazmi@cisco.com>

11 files changed, 110 insertions, 3 deletions

diff --git a/MAINTAINERS b/MAINTAINERS
index cd8bf49afbc..eeaca9b03f0 100644
--- a/MAINTAINERS
+++ b/MAINTAINERS
@@ -78,6 +78,15 @@ I:	devices
 M:	Damjan Marion <damarion@cisco.com>
 F:	src/vnet/devices/
 
+VNET TAP Drivers
+I:      tap
+M:      Damjan Marion <damarion@cisco.com>
+M:      Steven Luong <sluong@cisco.com>
+M:      Mohsin Kazmi <sykazmi@cisco.com>
+F:      src/vnet/devices/tap/
+F:      src/vnet/devices/virtio/node.c
+F:      src/vnet/devices/virtio/device.c
+
 VNET Feature Arcs
 I:	feature
 M:	Dave Barach <dave@barachs.net>
diff --git a/src/vat/api_format.c b/src/vat/api_format.c
index d9a9d223804..f107246774c 100644
--- a/src/vat/api_format.c
+++ b/src/vat/api_format.c
@@ -7429,6 +7429,7 @@ api_tap_create_v2 (vat_main_t * vam)
 {
   unformat_input_t *i = vam->input;
   vl_api_tap_create_v2_t *mp;
+#define TAP_FLAG_GSO (1 << 0)
   u8 mac_address[6];
   u8 random_mac = 1;
   u32 id = ~0;
@@ -7445,6 +7446,9 @@ api_tap_create_v2 (vat_main_t * vam)
   ip6_address_t host_ip6_gw;
   u8 host_ip6_gw_set = 0;
   u32 host_ip6_prefix_len = 0;
+  u8 host_mtu_set = 0;
+  u32 host_mtu_size = 0;
+  u32 tap_flags = 0;
   int ret;
   u32 rx_ring_sz = 0, tx_ring_sz = 0;
 
@@ -7484,6 +7488,12 @@ api_tap_create_v2 (vat_main_t * vam)
 	;
       else if (unformat (i, "tx-ring-size %d", &tx_ring_sz))
 	;
+      else if (unformat (i, "host-mtu-size %d", &host_mtu_size))
+	host_mtu_set = 1;
+      else if (unformat (i, "no-gso"))
+	tap_flags &= ~TAP_FLAG_GSO;
+      else if (unformat (i, "gso"))
+	tap_flags |= TAP_FLAG_GSO;
       else
 	break;
     }
@@ -7533,6 +7543,11 @@ api_tap_create_v2 (vat_main_t * vam)
       errmsg ("tx ring size must be 32768 or lower. ");
       return -99;
     }
+  if (host_mtu_set && (host_mtu_size < 64 || host_mtu_size > 65355))
+    {
+      errmsg ("host MTU size must be in between 64 and 65355. ");
+      return -99;
+    }
 
   /* Construct the API message */
   M (TAP_CREATE_V2, mp);
@@ -7546,6 +7561,9 @@ api_tap_create_v2 (vat_main_t * vam)
   mp->host_ip6_addr_set = host_ip6_prefix_len != 0;
   mp->rx_ring_sz = ntohs (rx_ring_sz);
   mp->tx_ring_sz = ntohs (tx_ring_sz);
+  mp->host_mtu_set = host_mtu_set;
+  mp->host_mtu_size = ntohl (host_mtu_size);
+  mp->tap_flags = ntohl (tap_flags);
 
   if (random_mac == 0)
     clib_memcpy (mp->mac_address, mac_address, 6);
@@ -21843,7 +21861,7 @@ _(l2_flags,                                                             \
 _(bridge_flags,                                                         \
   "bd_id <bridge-domain-id> [learn] [forward] [uu-flood] [flood] [arp-term] [disable]\n") \
 _(tap_create_v2,                                                        \
-  "id <num> [hw-addr <mac-addr>] [host-ns <name>] [rx-ring-size <num> [tx-ring-size <num>]") \
+  "id <num> [hw-addr <mac-addr>] [host-ns <name>] [rx-ring-size <num> [tx-ring-size <num>] [host-mtu-size <mtu>] [gso | no-gso]") \
 _(tap_delete_v2,                                                        \
   "<vpp-if-name> | sw_if_index <id>")                                   \
 _(sw_interface_tap_v2_dump, "")                                         \
diff --git a/src/vnet/devices/netlink.h b/src/vnet/devices/netlink.h
index 0d3e0d23ac6..4c765685091 100644
--- a/src/vnet/devices/netlink.h
+++ b/src/vnet/devices/netlink.h
@@ -22,6 +22,7 @@ clib_error_t *vnet_netlink_set_link_netns (int ifindex, int netns_fd,
 clib_error_t *vnet_netlink_set_link_master (int ifindex, char *master_ifname);
 clib_error_t *vnet_netlink_set_link_addr (int ifindex, u8 * addr);
 clib_error_t *vnet_netlink_set_link_state (int ifindex, int up);
+clib_error_t *vnet_netlink_set_link_mtu (int ifindex, int mtu);
 clib_error_t *vnet_netlink_add_ip4_addr (int ifindex, void *addr,
 					 int pfx_len);
 clib_error_t *vnet_netlink_add_ip6_addr (int ifindex, void *addr,
diff --git a/src/vnet/devices/tap/cli.c b/src/vnet/devices/tap/cli.c
index 084fb908dc9..c74d24a12a9 100644
--- a/src/vnet/devices/tap/cli.c
+++ b/src/vnet/devices/tap/cli.c
@@ -76,6 +76,10 @@ tap_create_command_fn (vlib_main_t * vm, unformat_input_t * input,
 	    ;
 	  else if (unformat (line_input, "tx-ring-size %d", &args.tx_ring_sz))
 	    ;
+	  else
+	    if (unformat
+		(line_input, "host-mtu-size %d", &args.host_mtu_size))
+	    args.host_mtu_set = 1;
 	  else if (unformat (line_input, "no-gso"))
 	    args.tap_flags &= ~TAP_FLAG_GSO;
 	  else if (unformat (line_input, "gso"))
@@ -114,7 +118,8 @@ VLIB_CLI_COMMAND (tap_create_command, static) = {
     "[rx-ring-size <size>] [tx-ring-size <size>] [host-ns <netns>] "
     "[host-bridge <bridge-name>] [host-ip4-addr <ip4addr/mask>] "
     "[host-ip6-addr <ip6-addr>] [host-ip4-gw <ip4-addr>] "
-    "[host-ip6-gw <ip6-addr>] [host-if-name <name>] [no-gso|gso]",
+    "[host-ip6-gw <ip6-addr>] [host-mac-addr <host-mac-address>] "
+    "[host-if-name <name>] [host-mtu-size <size>] [no-gso|gso]",
   .function = tap_create_command_fn,
 };
 /* *INDENT-ON* */
diff --git a/src/vnet/devices/tap/tap.c b/src/vnet/devices/tap/tap.c
index 35f1f2a3451..8dc798a3cb3 100644
--- a/src/vnet/devices/tap/tap.c
+++ b/src/vnet/devices/tap/tap.c
@@ -361,6 +361,29 @@ tap_create_if (vlib_main_t * vm, tap_create_if_args_t * args)
 	}
     }
 
+  if (args->host_mtu_set)
+    {
+      args->error =
+	vnet_netlink_set_link_mtu (vif->ifindex, args->host_mtu_size);
+      if (args->error)
+	{
+	  args->rv = VNET_API_ERROR_NETLINK_ERROR;
+	  goto error;
+	}
+    }
+  else if (tm->host_mtu_size != 0)
+    {
+      args->error =
+	vnet_netlink_set_link_mtu (vif->ifindex, tm->host_mtu_size);
+      if (args->error)
+	{
+	  args->rv = VNET_API_ERROR_NETLINK_ERROR;
+	  goto error;
+	}
+      args->host_mtu_set = 1;
+      args->host_mtu_size = tm->host_mtu_size;
+    }
+
   /* Set vhost memory table */
   i = sizeof (struct vhost_memory) + sizeof (struct vhost_memory_region);
   vhost_mem = clib_mem_alloc (i);
@@ -396,6 +419,7 @@ tap_create_if (vlib_main_t * vm, tap_create_if_args_t * args)
   args->host_namespace = 0;
   vif->host_bridge = args->host_bridge;
   args->host_bridge = 0;
+  vif->host_mtu_size = args->host_mtu_size;
   clib_memcpy (vif->host_mac_addr, args->host_mac_addr, 6);
   vif->host_ip4_prefix_len = args->host_ip4_prefix_len;
   vif->host_ip6_prefix_len = args->host_ip6_prefix_len;
@@ -627,6 +651,7 @@ tap_dump_ifs (tap_interface_details_t ** out_tapids)
     if (vif->host_ip6_prefix_len)
       clib_memcpy(tapid->host_ip6_addr, &vif->host_ip6_addr, 16);
     tapid->host_ip6_prefix_len = vif->host_ip6_prefix_len;
+    tapid->host_mtu_size = vif->host_mtu_size;
   );
   /* *INDENT-ON* */
 
@@ -636,6 +661,26 @@ tap_dump_ifs (tap_interface_details_t ** out_tapids)
 }
 
 static clib_error_t *
+tap_mtu_config (vlib_main_t * vm, unformat_input_t * input)
+{
+  tap_main_t *tm = &tap_main;
+
+  while (unformat_check_input (input) != UNFORMAT_END_OF_INPUT)
+    {
+      if (unformat (input, "host-mtu %d", &tm->host_mtu_size))
+	;
+      else
+	return clib_error_return (0, "unknown input `%U'",
+				  format_unformat_error, input);
+    }
+
+  return 0;
+}
+
+/* tap { host-mtu <size> } configuration. */
+VLIB_CONFIG_FUNCTION (tap_mtu_config, "tap");
+
+static clib_error_t *
 tap_init (vlib_main_t * vm)
 {
   tap_main_t *tm = &tap_main;
@@ -644,6 +689,8 @@ tap_init (vlib_main_t * vm)
   tm->log_default = vlib_log_register_class ("tap", 0);
   vlib_log_debug (tm->log_default, "initialized");
 
+  tm->host_mtu_size = 0;
+
   return error;
 }
 
diff --git a/src/vnet/devices/tap/tap.h b/src/vnet/devices/tap/tap.h
index 745f9fca304..45ff1d9910e 100644
--- a/src/vnet/devices/tap/tap.h
+++ b/src/vnet/devices/tap/tap.h
@@ -43,6 +43,8 @@ typedef struct
   u8 host_ip6_prefix_len;
   ip6_address_t host_ip6_gw;
   u8 host_ip6_gw_set;
+  u8 host_mtu_set;
+  u32 host_mtu_size;
   /* return */
   u32 sw_if_index;
   int rv;
@@ -66,6 +68,7 @@ typedef struct
   u8 host_ip4_prefix_len;
   u8 host_ip6_addr[16];
   u8 host_ip6_prefix_len;
+  u32 host_mtu_size;
 } tap_interface_details_t;
 
 typedef struct
@@ -75,6 +78,9 @@ typedef struct
 
   /* bit-map of in-use IDs */
   uword *tap_ids;
+
+  /* host mtu size, configurable through startup.conf */
+  int host_mtu_size;
 } tap_main_t;
 
 void tap_create_if (vlib_main_t * vm, tap_create_if_args_t * args);
diff --git a/src/vnet/devices/tap/tapv2.api b/src/vnet/devices/tap/tapv2.api
index fb90483efbe..2d4d5c3497d 100644
--- a/src/vnet/devices/tap/tapv2.api
+++ b/src/vnet/devices/tap/tapv2.api
@@ -19,7 +19,7 @@
     the Linux kernel TAP device driver
 */
 
-option version = "2.0.0";
+option version = "2.1.0";
 
 /** \brief Initialize a new tap interface with the given paramters
     @param client_index - opaque cookie to identify the sender
@@ -47,6 +47,8 @@ option version = "2.0.0";
     @param host_ip4_gw - host IPv4 default gateway
     @param host_ip6_gw_set - host IPv6 default gateway should be set
     @param host_ip6_gw - host IPv6 default gateway
+    @param host_mtu_set - host MTU should be set
+    @param host_mtu_size - host MTU size
     @param tap_flags - flags for the TAP interface creation
 */
 define tap_create_v2
@@ -76,6 +78,8 @@ define tap_create_v2
   u8 host_ip4_gw[4];
   u8 host_ip6_gw_set;
   u8 host_ip6_gw[16];
+  u8 host_mtu_set;
+  u32 host_mtu_size;
   u8 tag[64];
   u32 tap_flags;
 };
@@ -125,6 +129,7 @@ define sw_interface_tap_v2_dump
     @param host_ip4_prefix_len - host IPv4 ip address prefix length; 0 if unset
     @param host_ip6_addr - host IPv6 ip address
     @param host_ip6_prefix_len - host IPv6 ip address prefix length; 0 if unset
+    @param host_mtu_size - host mtu size
 */
 define sw_interface_tap_v2_details
 {
@@ -142,6 +147,7 @@ define sw_interface_tap_v2_details
   u8 host_ip4_prefix_len;
   u8 host_ip6_addr[16];
   u8 host_ip6_prefix_len;
+  u32 host_mtu_size;
   u32 tap_flags;
 };
 
diff --git a/src/vnet/devices/tap/tapv2_api.c b/src/vnet/devices/tap/tapv2_api.c
index 40ff22eaeaa..2471d0084e1 100644
--- a/src/vnet/devices/tap/tapv2_api.c
+++ b/src/vnet/devices/tap/tapv2_api.c
@@ -109,6 +109,12 @@ vl_api_tap_create_v2_t_handler (vl_api_tap_create_v2_t * mp)
       ap->host_ip6_gw_set = 1;
     }
 
+  if (mp->host_mtu_set)
+    {
+      ap->host_mtu_size = ntohl (mp->host_mtu_size);
+      ap->host_mtu_set = 1;
+    }
+
   ap->tap_flags = ntohl (mp->tap_flags);
 
   tap_create_if (vm, ap);
@@ -190,6 +196,7 @@ tap_send_sw_interface_details (vpe_api_main_t * am,
   clib_memcpy (mp->host_bridge, tap_if->host_bridge,
 	       MIN (ARRAY_LEN (mp->host_bridge) - 1,
 		    strlen ((const char *) tap_if->host_bridge)));
+  mp->host_mtu_size = htonl (tap_if->host_mtu_size);
   if (tap_if->host_ip4_prefix_len)
     clib_memcpy (&mp->host_ip4_addr, &tap_if->host_ip4_addr, 4);
   mp->host_ip4_prefix_len = tap_if->host_ip4_prefix_len;
diff --git a/src/vnet/devices/virtio/virtio.c b/src/vnet/devices/virtio/virtio.c
index a3fd05dbb55..9e2067bd521 100644
--- a/src/vnet/devices/virtio/virtio.c
+++ b/src/vnet/devices/virtio/virtio.c
@@ -317,6 +317,9 @@ virtio_show (vlib_main_t * vm, u32 * hw_if_indices, u8 show_descr, u32 type)
 	    vlib_cli_output (vm, "  name \"%s\"", vif->host_if_name);
 	  if (vif->net_ns)
 	    vlib_cli_output (vm, "  host-ns \"%s\"", vif->net_ns);
+	  if (vif->host_mtu_size)
+	    vlib_cli_output (vm, "  host-mtu-size \"%d\"",
+			     vif->host_mtu_size);
 	  vlib_cli_output (vm, "  fd %d", vif->fd);
 	  vlib_cli_output (vm, "  tap-fd %d", vif->tap_fd);
 	  vlib_cli_output (vm, "  gso-enabled %d", vif->gso_enabled);
diff --git a/src/vnet/devices/virtio/virtio.h b/src/vnet/devices/virtio/virtio.h
index 55b6271b337..52b57e9d95f 100644
--- a/src/vnet/devices/virtio/virtio.h
+++ b/src/vnet/devices/virtio/virtio.h
@@ -178,6 +178,7 @@ typedef struct
   u8 host_ip4_prefix_len;
   ip6_address_t host_ip6_addr;
   u8 host_ip6_prefix_len;
+  u32 host_mtu_size;
   u32 tap_file_index;
   int gso_enabled;
   int ifindex;
diff --git a/src/vpp/api/custom_dump.c b/src/vpp/api/custom_dump.c
index 94a868e0353..524ad368c89 100644
--- a/src/vpp/api/custom_dump.c
+++ b/src/vpp/api/custom_dump.c
@@ -579,6 +579,10 @@ static void *vl_api_tap_create_v2_t_print
     s = format (s, "tx-ring-size %u ", ntohs (mp->tx_ring_sz));
   if (mp->rx_ring_sz)
     s = format (s, "rx-ring-size %u ", ntohs (mp->rx_ring_sz));
+  if (mp->host_mtu_set)
+    s = format (s, "host-mtu-size %u ", ntohl (mp->host_mtu_size));
+  if (ntohl (mp->tap_flags) & 0x1)
+    s = format (s, "gso-enabled");
   FINISH;
 }