interface: improve MTU handling

 - per hw-interface-class handlers
 - ethernet set_mtu callback
 - driver can now refuse MTU change

Type: improvement
Change-Id: I3d37c9129930ebec7bb70caf4263025413873048
Signed-off-by: Damjan Marion <damarion@cisco.com>

8 files changed, 98 insertions, 47 deletions

diff --git a/src/plugins/af_xdp/device.c b/src/plugins/af_xdp/device.c
index 4aa36d95dd5..a5e0b739300 100644
--- a/src/plugins/af_xdp/device.c
+++ b/src/plugins/af_xdp/device.c
@@ -67,6 +67,15 @@ af_xdp_mac_change (vnet_hw_interface_t * hw, const u8 * old, const u8 * new)
   return 0;
 }
 
+static clib_error_t *
+af_xdp_set_mtu (vnet_main_t *vnm, vnet_hw_interface_t *hw, u32 mtu)
+{
+  af_xdp_main_t *am = &af_xdp_main;
+  af_xdp_device_t *ad = vec_elt_at_index (am->devices, hw->dev_instance);
+  af_xdp_log (VLIB_LOG_LEVEL_ERR, ad, "set mtu not supported yet");
+  return vnet_error (VNET_ERR_UNSUPPORTED, 0);
+}
+
 static u32
 af_xdp_flag_change (vnet_main_t * vnm, vnet_hw_interface_t * hw, u32 flags)
 {
@@ -82,9 +91,6 @@ af_xdp_flag_change (vnet_main_t * vnm, vnet_hw_interface_t * hw, u32 flags)
       af_xdp_log (VLIB_LOG_LEVEL_ERR, ad,
 		  "set promiscuous not supported yet");
       return ~0;
-    case ETHERNET_INTERFACE_FLAG_MTU:
-      af_xdp_log (VLIB_LOG_LEVEL_ERR, ad, "set mtu not supported yet");
-      return ~0;
     }
 
   af_xdp_log (VLIB_LOG_LEVEL_ERR, ad, "unknown flag %x requested", flags);
@@ -651,6 +657,7 @@ af_xdp_create_if (vlib_main_t * vm, af_xdp_create_if_args_t * args)
   eir.dev_instance = ad->dev_instance;
   eir.address = ad->hwaddr;
   eir.cb.flag_change = af_xdp_flag_change;
+  eir.cb.set_mtu = af_xdp_set_mtu;
   ad->hw_if_index = vnet_eth_register_interface (vnm, &eir);
 
   sw = vnet_get_hw_sw_interface (vnm, ad->hw_if_index);
diff --git a/src/plugins/dpdk/device/init.c b/src/plugins/dpdk/device/init.c
index 5e06e48d37e..4faf6290dba 100644
--- a/src/plugins/dpdk/device/init.c
+++ b/src/plugins/dpdk/device/init.c
@@ -21,6 +21,7 @@
 #include <vlib/unix/unix.h>
 #include <vlib/log.h>
 
+#include <vnet/vnet.h>
 #include <vnet/ethernet/ethernet.h>
 #include <vnet/interface/rx_queue_funcs.h>
 #include <dpdk/buffer.h>
@@ -68,6 +69,41 @@ const struct
   { ETH_LINK_SPEED_1G, "GigabitEthernet" },
 };
 
+static clib_error_t *
+dpdk_set_mtu (vnet_main_t *vnm, vnet_hw_interface_t *hi, u32 mtu)
+{
+  dpdk_main_t *dm = &dpdk_main;
+  dpdk_device_t *xd = vec_elt_at_index (dm->devices, hi->dev_instance);
+  int rv;
+
+  rv = rte_eth_dev_set_mtu (xd->port_id, mtu);
+
+  if (rv < 0)
+    {
+      dpdk_log_err ("[%u] rte_eth_dev_set_mtu failed (mtu %u, rv %d)",
+		    xd->port_id, mtu, rv);
+      switch (rv)
+	{
+	case -ENOTSUP:
+	  return vnet_error (VNET_ERR_UNSUPPORTED,
+			     "dpdk driver doesn't support MTU change");
+	case -EBUSY:
+	  return vnet_error (VNET_ERR_BUSY, "port is running");
+	case -EINVAL:
+	  return vnet_error (VNET_ERR_INVALID_VALUE, "invalid MTU");
+	default:
+	  return vnet_error (VNET_ERR_BUG,
+			     "unexpected return value %d returned from "
+			     "rte_eth_dev_set_mtu(...)",
+			     rv);
+	}
+    }
+  else
+    dpdk_log_debug ("[%u] mtu set to %u", xd->port_id, mtu);
+
+  return 0;
+}
+
 static u32
 dpdk_flag_change (vnet_main_t * vnm, vnet_hw_interface_t * hi, u32 flags)
 {
@@ -84,15 +120,6 @@ dpdk_flag_change (vnet_main_t * vnm, vnet_hw_interface_t * hi, u32 flags)
     case ETHERNET_INTERFACE_FLAG_ACCEPT_ALL:
       dpdk_device_flag_set (xd, DPDK_DEVICE_FLAG_PROMISC, 1);
       break;
-    case ETHERNET_INTERFACE_FLAG_MTU:
-      if (xd->flags & DPDK_DEVICE_FLAG_ADMIN_UP)
-	rte_eth_dev_stop (xd->port_id);
-      rte_eth_dev_set_mtu (xd->port_id, hi->max_packet_bytes);
-      if (xd->flags & DPDK_DEVICE_FLAG_ADMIN_UP)
-	rte_eth_dev_start (xd->port_id);
-      dpdk_log_debug ("[%u] mtu changed to %u", xd->port_id,
-		      hi->max_packet_bytes);
-      return 0;
     default:
       return ~0;
     }
@@ -370,6 +397,7 @@ dpdk_lib_init (dpdk_main_t * dm)
       eir.dev_instance = xd->device_index;
       eir.address = addr;
       eir.cb.flag_change = dpdk_flag_change;
+      eir.cb.set_mtu = dpdk_set_mtu;
       xd->hw_if_index = vnet_eth_register_interface (vnm, &eir);
       hi = vnet_get_hw_interface (vnm, xd->hw_if_index);
       hi->numa_node = xd->cpu_socket = (i8) rte_eth_dev_socket_id (port_id);
diff --git a/src/plugins/rdma/device.c b/src/plugins/rdma/device.c
index cad5fdb541f..167a23213cc 100644
--- a/src/plugins/rdma/device.c
+++ b/src/plugins/rdma/device.c
@@ -183,11 +183,10 @@ rdma_mac_change (vnet_hw_interface_t * hw, const u8 * old, const u8 * new)
   return 0;
 }
 
-static u32
-rdma_dev_change_mtu (rdma_device_t * rd)
+static clib_error_t *
+rdma_set_mtu (vnet_main_t *vnm, vnet_hw_interface_t *hw, u32 mtu)
 {
-  rdma_log__ (VLIB_LOG_LEVEL_ERR, rd, "MTU change not supported");
-  return ~0;
+  return vnet_error (VNET_ERR_UNSUPPORTED, 0);
 }
 
 static u32
@@ -202,8 +201,6 @@ rdma_flag_change (vnet_main_t * vnm, vnet_hw_interface_t * hw, u32 flags)
       return rdma_dev_set_ucast (rd);
     case ETHERNET_INTERFACE_FLAG_ACCEPT_ALL:
       return rdma_dev_set_promisc (rd);
-    case ETHERNET_INTERFACE_FLAG_MTU:
-      return rdma_dev_change_mtu (rd);
     }
 
   rdma_log__ (VLIB_LOG_LEVEL_ERR, rd, "unknown flag %x requested", flags);
@@ -361,6 +358,7 @@ rdma_register_interface (vnet_main_t * vnm, rdma_device_t * rd)
   eir.dev_instance = rd->dev_instance;
   eir.address = rd->hwaddr.bytes;
   eir.cb.flag_change = rdma_flag_change;
+  eir.cb.set_mtu = rdma_set_mtu;
   rd->hw_if_index = vnet_eth_register_interface (vnm, &eir);
   /* Indicate ability to support L3 DMAC filtering and
    * initialize interface to L3 non-promisc mode */
diff --git a/src/vnet/devices/af_packet/af_packet.c b/src/vnet/devices/af_packet/af_packet.c
index cf32e92c94c..cf4f91e2fcd 100644
--- a/src/vnet/devices/af_packet/af_packet.c
+++ b/src/vnet/devices/af_packet/af_packet.c
@@ -56,31 +56,26 @@ unsigned int if_nametoindex (const char *ifname);
 
 typedef struct tpacket_req tpacket_req_t;
 
-static u32
-af_packet_eth_flag_change (vnet_main_t * vnm, vnet_hw_interface_t * hi,
-			   u32 flags)
+static clib_error_t *
+af_packet_eth_set_mtu (vnet_main_t *vnm, vnet_hw_interface_t *hi, u32 mtu)
 {
-  clib_error_t *error;
+  clib_error_t *error, *rv;
   af_packet_main_t *apm = &af_packet_main;
-  af_packet_if_t *apif =
-    pool_elt_at_index (apm->interfaces, hi->dev_instance);
+  af_packet_if_t *apif = pool_elt_at_index (apm->interfaces, hi->dev_instance);
 
-  if (flags == ETHERNET_INTERFACE_FLAG_MTU)
-    {
-      error =
-	vnet_netlink_set_link_mtu (apif->host_if_index, hi->max_packet_bytes);
+  error = vnet_netlink_set_link_mtu (apif->host_if_index, mtu);
 
-      if (error)
-	{
-	  vlib_log_err (apm->log_class, "netlink failed to change MTU: %U",
-			format_clib_error, error);
-	  clib_error_free (error);
-	  return VNET_API_ERROR_SYSCALL_ERROR_1;
-	}
-      else
-	apif->host_mtu = hi->max_packet_bytes;
+  if (error)
+    {
+      vlib_log_err (apm->log_class, "netlink failed to change MTU: %U",
+		    format_clib_error, error);
+      rv = vnet_error (VNET_ERR_SYSCALL_ERROR_1, "netlink error: %U",
+		       format_clib_error, error);
+      clib_error_free (error);
+      return rv;
     }
-
+  else
+    apif->host_mtu = mtu;
   return 0;
 }
 
@@ -403,7 +398,7 @@ af_packet_create_if (af_packet_create_if_arg_t *arg)
       eir.dev_class_index = af_packet_device_class.index;
       eir.dev_instance = if_index;
       eir.address = hw_addr;
-      eir.cb.flag_change = af_packet_eth_flag_change;
+      eir.cb.set_mtu = af_packet_eth_set_mtu;
       apif->hw_if_index = vnet_eth_register_interface (vnm, &eir);
     }
   else
diff --git a/src/vnet/ethernet/ethernet.h b/src/vnet/ethernet/ethernet.h
index 9621429e4ee..b6adeb6f44d 100644
--- a/src/vnet/ethernet/ethernet.h
+++ b/src/vnet/ethernet/ethernet.h
@@ -130,7 +130,11 @@ typedef u32 (ethernet_flag_change_function_t)
 
 typedef struct
 {
+  /* ethernet interface flags change */
   ethernet_flag_change_function_t *flag_change;
+
+  /* set MTU callback */
+  vnet_interface_set_mtu_function_t *set_mtu;
 } vnet_eth_if_callbacks_t;
 
 #define ETHERNET_MIN_PACKET_BYTES  64
@@ -166,9 +170,6 @@ typedef struct ethernet_interface
   /* Set interface to accept all packets (promiscuous mode). */
 #define ETHERNET_INTERFACE_FLAG_ACCEPT_ALL 1
 
-  /* Change MTU on interface from hw interface structure */
-#define ETHERNET_INTERFACE_FLAG_MTU        2
-
   /* Callback, e.g. to turn on/off promiscuous mode */
   vnet_eth_if_callbacks_t cb;
 
diff --git a/src/vnet/ethernet/interface.c b/src/vnet/ethernet/interface.c
index b1513a79774..bac882228cb 100644
--- a/src/vnet/ethernet/interface.c
+++ b/src/vnet/ethernet/interface.c
@@ -310,6 +310,18 @@ ethernet_mac_change (vnet_hw_interface_t * hi,
   return (NULL);
 }
 
+static clib_error_t *
+ethernet_set_mtu (vnet_main_t *vnm, vnet_hw_interface_t *hi, u32 mtu)
+{
+  ethernet_interface_t *ei =
+    pool_elt_at_index (ethernet_main.interfaces, hi->hw_instance);
+
+  if (ei->cb.set_mtu)
+    return ei->cb.set_mtu (vnm, hi, mtu);
+
+  return 0;
+}
+
 /* *INDENT-OFF* */
 VNET_HW_INTERFACE_CLASS (ethernet_hw_interface_class) = {
   .name = "Ethernet",
@@ -321,6 +333,7 @@ VNET_HW_INTERFACE_CLASS (ethernet_hw_interface_class) = {
   .build_rewrite = ethernet_build_rewrite,
   .update_adjacency = ethernet_update_adjacency,
   .mac_addr_change_function = ethernet_mac_change,
+  .set_mtu = ethernet_set_mtu,
 };
 /* *INDENT-ON* */
 
@@ -367,8 +380,8 @@ vnet_eth_register_interface (vnet_main_t *vnm,
 
   hi->min_packet_bytes = hi->min_supported_packet_bytes =
     ETHERNET_MIN_PACKET_BYTES;
-  hi->max_packet_bytes = hi->max_supported_packet_bytes =
-    ETHERNET_MAX_PACKET_BYTES;
+  hi->max_supported_packet_bytes = ETHERNET_MAX_PACKET_BYTES;
+  hi->max_packet_bytes = em->default_mtu;
 
   /* Default ethernet MTU, 9000 unless set by ethernet_config see below */
   vnet_sw_interface_set_mtu (vnm, hi->sw_if_index, em->default_mtu);
@@ -460,8 +473,6 @@ ethernet_set_flags (vnet_main_t * vnm, u32 hw_if_index, u32 flags)
 	  /* fall through */
 	case ETHERNET_INTERFACE_FLAG_ACCEPT_ALL:
 	  ei->flags &= ~ETHERNET_INTERFACE_FLAG_STATUS_L3;
-	  /* fall through */
-	case ETHERNET_INTERFACE_FLAG_MTU:
 	  return ei->cb.flag_change (vnm, hi, opn_flags);
 	default:
 	  return ~0;
diff --git a/src/vnet/interface.c b/src/vnet/interface.c
index 05a1c7c34a9..a1493c66c23 100644
--- a/src/vnet/interface.c
+++ b/src/vnet/interface.c
@@ -772,6 +772,9 @@ clib_error_t *
 vnet_hw_interface_set_mtu (vnet_main_t *vnm, u32 hw_if_index, u32 mtu)
 {
   vnet_hw_interface_t *hi = vnet_get_hw_interface (vnm, hw_if_index);
+  vnet_hw_interface_class_t *hw_if_class =
+    vnet_get_hw_interface_class (vnm, hi->hw_class_index);
+  clib_error_t *err = 0;
 
   if (hi->max_packet_bytes != mtu)
     {
@@ -781,8 +784,10 @@ vnet_hw_interface_set_mtu (vnet_main_t *vnm, u32 hw_if_index, u32 mtu)
 			   "requested mtu must be in the %u to %u range",
 			   hi->min_supported_packet_bytes,
 			   hi->max_supported_packet_bytes);
+      if (hw_if_class->set_mtu)
+	if ((err = hw_if_class->set_mtu (vnm, hi, mtu)))
+	  return err;
       hi->max_packet_bytes = mtu;
-      ethernet_set_flags (vnm, hw_if_index, ETHERNET_INTERFACE_FLAG_MTU);
       vnet_hw_interface_walk_sw (vnm, hw_if_index, sw_interface_walk_callback,
 				 &mtu);
     }
diff --git a/src/vnet/interface.h b/src/vnet/interface.h
index 801298a0f05..2eb50aed5e8 100644
--- a/src/vnet/interface.h
+++ b/src/vnet/interface.h
@@ -70,6 +70,10 @@ typedef clib_error_t *(vnet_subif_add_del_function_t)
   (struct vnet_main_t * vnm, u32 if_index,
    struct vnet_sw_interface_t * template, int is_add);
 
+/* Interface set mtu callback. */
+typedef clib_error_t *(vnet_interface_set_mtu_function_t) (
+  struct vnet_main_t *vnm, struct vnet_hw_interface_t *hi, u32 mtu);
+
 /* Interface set mac address callback. */
 typedef clib_error_t *(vnet_interface_set_mac_address_function_t)
   (struct vnet_hw_interface_t * hi,
@@ -428,6 +432,8 @@ typedef struct _vnet_hw_interface_class
 
   /* Function to add/delete additional MAC addresses */
   vnet_interface_add_del_mac_address_function_t *mac_addr_add_del_function;
+  /* Function to set mtu. */
+  vnet_interface_set_mtu_function_t *set_mtu;
 
   /* Format function to display interface name. */
   format_function_t *format_interface_name;