net: Give the user the ability to set MTU

Add a new "mtu" parameter to the NetConfig structure and therefore to
the --net option. This allows Cloud Hypervisor's users to define the
Maximum Transmission Unit (MTU) they want to use for the network
interface that they create.

In details, there are two main aspects. On the one hand, the TAP
interface is created with the proper MTU if it is provided. And on the
other hand the guest is made aware of the MTU through the VIRTIO
configuration. That means the MTU is properly set on both the TAP on the
host and the network interface in the guest.

Signed-off-by: Sebastien Boeuf <sebastien.boeuf@intel.com>

net_util/src/lib.rs

@@ -131,17 +131,19 @@ pub fn build_net_config_space(
     config: &mut VirtioNetConfig,
     mac: MacAddr,
     num_queues: usize,
+    mtu: Option<u16>,
     avail_features: &mut u64,
 ) {
     config.mac.copy_from_slice(mac.get_bytes());
     *avail_features |= 1 << VIRTIO_NET_F_MAC;
 
-    build_net_config_space_with_mq(config, num_queues, avail_features);
+    build_net_config_space_with_mq(config, num_queues, mtu, avail_features);
 }
 
 pub fn build_net_config_space_with_mq(
     config: &mut VirtioNetConfig,
     num_queues: usize,
+    mtu: Option<u16>,
     avail_features: &mut u64,
 ) {
     let num_queue_pairs = (num_queues / 2) as u16;
@@ -151,6 +153,9 @@ pub fn build_net_config_space_with_mq(
         config.max_virtqueue_pairs = num_queue_pairs;
         *avail_features |= 1 << VIRTIO_NET_F_MQ;
     }
+    if let Some(mtu) = mtu {
+        config.mtu = mtu;
+    }
 }
 
 pub fn virtio_features_to_tap_offload(features: u64) -> c_uint {

net_util/src/open_tap.rs

@@ -30,6 +30,8 @@ pub enum Error {
     TapGetMac(TapError),
     #[error("Setting vnet header size failed: {0}")]
     TapSetVnetHdrSize(TapError),
+    #[error("Setting MTU failed: {0}")]
+    TapSetMtu(TapError),
     #[error("Enabling tap interface failed: {0}")]
     TapEnable(TapError),
 }
@@ -63,6 +65,7 @@ pub fn open_tap(
     ip_addr: Option<Ipv4Addr>,
     netmask: Option<Ipv4Addr>,
     host_mac: &mut Option<MacAddr>,
+    mtu: Option<u16>,
     num_rx_q: usize,
     flags: Option<i32>,
 ) -> Result<Vec<Tap>> {
@@ -95,6 +98,9 @@ pub fn open_tap(
             } else {
                 *host_mac = Some(tap.get_mac_addr().map_err(Error::TapGetMac)?)
             }
+            if let Some(mtu) = mtu {
+                tap.set_mtu(mtu as i32).map_err(Error::TapSetMtu)?;
+            }
             tap.enable().map_err(Error::TapEnable)?;
 
             tap.set_vnet_hdr_size(vnet_hdr_size)

net_util/src/tap.rs

@@ -297,6 +297,37 @@ impl Tap {
         Ok(())
     }
 
+    pub fn mtu(&self) -> Result<i32> {
+        let sock = create_unix_socket().map_err(Error::NetUtil)?;
+
+        let ifreq = self.get_ifreq();
+
+        // ioctl is safe. Called with a valid sock fd, and we check the return.
+        let ret = unsafe { ioctl_with_ref(&sock, net_gen::sockios::SIOCGIFMTU as c_ulong, &ifreq) };
+        if ret < 0 {
+            return Err(Error::IoctlError(IoError::last_os_error()));
+        }
+
+        let mtu = unsafe { ifreq.ifr_ifru.ifru_mtu };
+
+        Ok(mtu)
+    }
+
+    pub fn set_mtu(&self, mtu: i32) -> Result<()> {
+        let sock = create_unix_socket().map_err(Error::NetUtil)?;
+
+        let mut ifreq = self.get_ifreq();
+        ifreq.ifr_ifru.ifru_mtu = mtu;
+
+        // ioctl is safe. Called with a valid sock fd, and we check the return.
+        let ret = unsafe { ioctl_with_ref(&sock, net_gen::sockios::SIOCSIFMTU as c_ulong, &ifreq) };
+        if ret < 0 {
+            return Err(Error::IoctlError(IoError::last_os_error()));
+        }
+
+        Ok(())
+    }
+
     /// Set the offload flags for the tap interface.
     pub fn set_offload(&self, flags: c_uint) -> Result<()> {
         // ioctl is safe. Called with a valid tap fd, and we check the return.

test_infra/src/lib.rs

@@ -796,6 +796,13 @@ impl Guest {
         )
     }
 
+    pub fn default_net_string_w_mtu(&self, mtu: u16) -> String {
+        format!(
+            "tap=,mac={},ip={},mask=255.255.255.0,mtu={}",
+            self.network.guest_mac, self.network.host_ip, mtu
+        )
+    }
+
     pub fn ssh_command(&self, command: &str) -> Result<String, SshCommandError> {
         ssh_command_ip(
             command,

tests/integration.rs

@@ -2411,9 +2411,9 @@ mod common_parallel {
             .args(["--memory", "size=512M"])
             .args(["--kernel", direct_kernel_boot_path().to_str().unwrap()])
             .args(["--cmdline", DIRECT_KERNEL_BOOT_CMDLINE])
+            .args(["--net", guest.default_net_string_w_mtu(3000).as_str()])
             .capture_output()
-            .default_disks()
-            .default_net();
+            .default_disks();
 
         let mut child = cmd.spawn().unwrap();
 
@@ -2434,6 +2434,13 @@ mod common_parallel {
                     .trim(),
                 "success"
             );
+            assert_eq!(
+                guest
+                    .ssh_command(format!("cat /sys/class/net/{}/mtu", iface).as_str())
+                    .unwrap()
+                    .trim(),
+                "3000"
+            );
         });
 
         let _ = child.kill();
@@ -6098,6 +6105,7 @@ mod common_parallel {
             Some(std::net::Ipv4Addr::from_str(&guest.network.host_ip).unwrap()),
             None,
             &mut None,
+            None,
             num_queue_pairs,
             Some(libc::O_RDWR | libc::O_NONBLOCK),
         )

vhost_user_net/src/lib.rs

@@ -133,6 +133,7 @@ impl VhostUserNetBackend {
             Some(ip_addr),
             Some(netmask),
             &mut Some(host_mac),
+            None,
             num_queues / 2,
             None,
         )

virtio-devices/src/net.rs

@@ -49,6 +49,9 @@ use vmm_sys_util::eventfd::EventFd;
 // Event available on the control queue.
 const CTRL_QUEUE_EVENT: u16 = EPOLL_HELPER_EVENT_LAST + 1;
 
+// Following the VIRTIO specification, the MTU should be at least 1280.
+pub const MIN_MTU: u16 = 1280;
+
 pub struct NetCtrlEpollHandler {
     pub mem: GuestMemoryAtomic<GuestMemoryMmap>,
     pub kill_evt: EventFd,
@@ -428,7 +431,7 @@ impl VersionMapped for NetState {}
 impl Net {
     /// Create a new virtio network device with the given TAP interface.
     #[allow(clippy::too_many_arguments)]
-    pub fn new_with_tap(
+    fn new_with_tap(
         id: String,
         taps: Vec<Tap>,
         guest_mac: Option<MacAddr>,
@@ -439,6 +442,11 @@ impl Net {
         rate_limiter_config: Option<RateLimiterConfig>,
         exit_evt: EventFd,
     ) -> Result<Self> {
+        let mut mtu = None;
+        if !taps.is_empty() {
+            mtu = Some(taps[0].mtu().map_err(Error::TapError)? as u16);
+        }
+
         let mut avail_features = 1 << VIRTIO_NET_F_CSUM
             | 1 << VIRTIO_NET_F_CTRL_GUEST_OFFLOADS
             | 1 << VIRTIO_NET_F_GUEST_CSUM
@@ -453,6 +461,9 @@ impl Net {
             | 1 << VIRTIO_RING_F_EVENT_IDX
             | 1 << VIRTIO_F_VERSION_1;
 
+        if mtu.is_some() {
+            avail_features |= 1u64 << VIRTIO_NET_F_MTU;
+        }
         if iommu {
             avail_features |= 1u64 << VIRTIO_F_IOMMU_PLATFORM;
         }
@@ -462,9 +473,9 @@ impl Net {
 
         let mut config = VirtioNetConfig::default();
         if let Some(mac) = guest_mac {
-            build_net_config_space(&mut config, mac, num_queues, &mut avail_features);
+            build_net_config_space(&mut config, mac, num_queues, mtu, &mut avail_features);
         } else {
-            build_net_config_space_with_mq(&mut config, num_queues, &mut avail_features);
+            build_net_config_space_with_mq(&mut config, num_queues, mtu, &mut avail_features);
         }
 
         Ok(Net {
@@ -497,6 +508,7 @@ impl Net {
         netmask: Option<Ipv4Addr>,
         guest_mac: Option<MacAddr>,
         host_mac: &mut Option<MacAddr>,
+        mtu: Option<u16>,
         iommu: bool,
         num_queues: usize,
         queue_size: u16,
@@ -504,8 +516,16 @@ impl Net {
         rate_limiter_config: Option<RateLimiterConfig>,
         exit_evt: EventFd,
     ) -> Result<Self> {
-        let taps = open_tap(if_name, ip_addr, netmask, host_mac, num_queues / 2, None)
-            .map_err(Error::OpenTap)?;
+        let taps = open_tap(
+            if_name,
+            ip_addr,
+            netmask,
+            host_mac,
+            mtu,
+            num_queues / 2,
+            None,
+        )
+        .map_err(Error::OpenTap)?;
 
         Self::new_with_tap(
             id,

virtio-devices/src/vhost_user/net.rs

@@ -24,7 +24,7 @@ use virtio_bindings::bindings::virtio_net::{
     VIRTIO_NET_F_CSUM, VIRTIO_NET_F_CTRL_VQ, VIRTIO_NET_F_GUEST_CSUM, VIRTIO_NET_F_GUEST_ECN,
     VIRTIO_NET_F_GUEST_TSO4, VIRTIO_NET_F_GUEST_TSO6, VIRTIO_NET_F_GUEST_UFO,
     VIRTIO_NET_F_HOST_ECN, VIRTIO_NET_F_HOST_TSO4, VIRTIO_NET_F_HOST_TSO6, VIRTIO_NET_F_HOST_UFO,
-    VIRTIO_NET_F_MAC, VIRTIO_NET_F_MRG_RXBUF,
+    VIRTIO_NET_F_MAC, VIRTIO_NET_F_MRG_RXBUF, VIRTIO_NET_F_MTU,
 };
 use virtio_bindings::bindings::virtio_ring::VIRTIO_RING_F_EVENT_IDX;
 use virtio_queue::{Queue, QueueT};
@@ -70,6 +70,7 @@ impl Net {
     pub fn new(
         id: String,
         mac_addr: MacAddr,
+        mtu: Option<u16>,
         vu_cfg: VhostUserConfig,
         server: bool,
         seccomp_action: SeccompAction,
@@ -126,8 +127,12 @@ impl Net {
             | 1 << VIRTIO_F_VERSION_1
             | VhostUserVirtioFeatures::PROTOCOL_FEATURES.bits();
 
+        if mtu.is_some() {
+            avail_features |= 1u64 << VIRTIO_NET_F_MTU;
+        }
+
         let mut config = VirtioNetConfig::default();
-        build_net_config_space(&mut config, mac_addr, num_queues, &mut avail_features);
+        build_net_config_space(&mut config, mac_addr, num_queues, mtu, &mut avail_features);
 
         let mut vu =
             VhostUserHandle::connect_vhost_user(server, &vu_cfg.socket, num_queues as u64, false)?;

vmm/src/api/openapi/cloud-hypervisor.yaml

@@ -819,6 +819,11 @@ components:
           default: "255.255.255.0"
         mac:
           type: string
+        host_mac:
+          type: string
+        mtu:
+          type: integer
+          default: 1280
         iommu:
           type: boolean
           default: false

vmm/src/config.rs

@@ -180,6 +180,8 @@ pub enum ValidationError {
     IommuNotSupported,
     /// Duplicated device path (device added twice)
     DuplicateDevicePath(String),
+    /// Provided MTU is lower than what the VIRTIO specification expects
+    InvalidMtu(u16),
 }
 
 type ValidationResult<T> = std::result::Result<T, ValidationError>;
@@ -280,6 +282,13 @@ impl fmt::Display for ValidationError {
                 write!(f, "Device does not support being placed behind IOMMU")
             }
             DuplicateDevicePath(p) => write!(f, "Duplicated device path: {}", p),
+            &InvalidMtu(mtu) => {
+                write!(
+                    f,
+                    "Provided MTU {} is lower than 1280 (expected by VIRTIO specification)",
+                    mtu
+                )
+            }
         }
     }
 }
@@ -1238,6 +1247,8 @@ pub struct NetConfig {
     pub mac: MacAddr,
     #[serde(default)]
     pub host_mac: Option<MacAddr>,
+    #[serde(default = "default_netconfig_mtu")]
+    pub mtu: u16,
     #[serde(default)]
     pub iommu: bool,
     #[serde(default = "default_netconfig_num_queues")]
@@ -1275,6 +1286,10 @@ fn default_netconfig_mac() -> MacAddr {
     MacAddr::local_random()
 }
 
+fn default_netconfig_mtu() -> u16 {
+    virtio_devices::net::MIN_MTU
+}
+
 fn default_netconfig_num_queues() -> usize {
     DEFAULT_NUM_QUEUES_VUNET
 }
@@ -1291,6 +1306,7 @@ impl Default for NetConfig {
             mask: default_netconfig_mask(),
             mac: default_netconfig_mac(),
             host_mac: None,
+            mtu: default_netconfig_mtu(),
             iommu: false,
             num_queues: default_netconfig_num_queues(),
             queue_size: default_netconfig_queue_size(),
@@ -1322,6 +1338,7 @@ impl NetConfig {
             .add("mask")
             .add("mac")
             .add("host_mac")
+            .add("mtu")
             .add("iommu")
             .add("queue_size")
             .add("num_queues")
@@ -1353,6 +1370,10 @@ impl NetConfig {
             .map_err(Error::ParseNetwork)?
             .unwrap_or_else(default_netconfig_mac);
         let host_mac = parser.convert("host_mac").map_err(Error::ParseNetwork)?;
+        let mtu = parser
+            .convert("mtu")
+            .map_err(Error::ParseNetwork)?
+            .unwrap_or_else(default_netconfig_mtu);
         let iommu = parser
             .convert::<Toggle>("iommu")
             .map_err(Error::ParseNetwork)?
@@ -1442,6 +1463,7 @@ impl NetConfig {
             mask,
             mac,
             host_mac,
+            mtu,
             iommu,
             num_queues,
             queue_size,
@@ -1493,6 +1515,10 @@ impl NetConfig {
             }
         }
 
+        if self.mtu < virtio_devices::net::MIN_MTU {
+            return Err(ValidationError::InvalidMtu(self.mtu));
+        }
+
         Ok(())
     }
 }

vmm/src/device_manager.rs

@@ -2241,6 +2241,7 @@ impl DeviceManager {
                 match virtio_devices::vhost_user::Net::new(
                     id.clone(),
                     net_cfg.mac,
+                    Some(net_cfg.mtu),
                     vu_cfg,
                     server,
                     self.seccomp_action.clone(),
@@ -2271,6 +2272,7 @@ impl DeviceManager {
                         None,
                         Some(net_cfg.mac),
                         &mut net_cfg.host_mac,
+                        None,
                         self.force_iommu | net_cfg.iommu,
                         net_cfg.num_queues,
                         net_cfg.queue_size,
@@ -2307,6 +2309,7 @@ impl DeviceManager {
                         Some(net_cfg.mask),
                         Some(net_cfg.mac),
                         &mut net_cfg.host_mac,
+                        Some(net_cfg.mtu),
                         self.force_iommu | net_cfg.iommu,
                         net_cfg.num_queues,
                         net_cfg.queue_size,

vmm/src/seccomp_filters.rs

@@ -66,6 +66,8 @@ const SIOCGIFFLAGS: u64 = 0x8913;
 const SIOCGIFHWADDR: u64 = 0x8927;
 const SIOCSIFFLAGS: u64 = 0x8914;
 const SIOCSIFADDR: u64 = 0x8916;
+const SIOCGIFMTU: u64 = 0x8921;
+const SIOCSIFMTU: u64 = 0x8922;
 const SIOCSIFHWADDR: u64 = 0x8924;
 const SIOCSIFNETMASK: u64 = 0x891c;
 
@@ -252,9 +254,11 @@ fn create_vmm_ioctl_seccomp_rule_common(
         and![Cond::new(1, ArgLen::Dword, Eq, FIONBIO)?],
         and![Cond::new(1, ArgLen::Dword, Eq, SIOCGIFFLAGS)?],
         and![Cond::new(1, ArgLen::Dword, Eq, SIOCGIFHWADDR)?],
+        and![Cond::new(1, ArgLen::Dword, Eq, SIOCGIFMTU)?],
         and![Cond::new(1, ArgLen::Dword, Eq, SIOCSIFADDR)?],
         and![Cond::new(1, ArgLen::Dword, Eq, SIOCSIFFLAGS)?],
         and![Cond::new(1, ArgLen::Dword, Eq, SIOCSIFHWADDR)?],
+        and![Cond::new(1, ArgLen::Dword, Eq, SIOCSIFMTU)?],
         and![Cond::new(1, ArgLen::Dword, Eq, SIOCSIFNETMASK)?],
         and![Cond::new(1, ArgLen::Dword, Eq, TCSETS)?],
         and![Cond::new(1, ArgLen::Dword, Eq, TCGETS)?],