tun_rs/platform/linux/

device.rs

use crate::platform::linux::offload::{
    gso_none_checksum, gso_split, handle_gro, VirtioNetHdr, VIRTIO_NET_HDR_F_NEEDS_CSUM,
    VIRTIO_NET_HDR_GSO_NONE, VIRTIO_NET_HDR_GSO_TCPV4, VIRTIO_NET_HDR_GSO_TCPV6,
    VIRTIO_NET_HDR_GSO_UDP_L4, VIRTIO_NET_HDR_LEN,
};
use crate::platform::unix::device::{ctl, ctl_v6};
use crate::platform::{ExpandBuffer, GROTable};
use crate::{
    builder::{DeviceConfig, Layer},
    platform::linux::sys::*,
    platform::{
        unix::{ipaddr_to_sockaddr, sockaddr_union, Fd, Tun},
        ETHER_ADDR_LEN,
    },
    ToIpv4Address, ToIpv4Netmask, ToIpv6Address, ToIpv6Netmask,
};
use ipnet::IpNet;
use libc::{
    self, c_char, c_short, ifreq, in6_ifreq, ARPHRD_ETHER, IFF_MULTI_QUEUE, IFF_NO_PI, IFF_RUNNING,
    IFF_TAP, IFF_TUN, IFF_UP, IFNAMSIZ, O_RDWR,
};
use std::net::Ipv6Addr;
use std::sync::{Arc, RwLock};
use std::{
    ffi::CString,
    io, mem,
    net::{IpAddr, Ipv4Addr},
    os::unix::io::{AsRawFd, RawFd},
    ptr,
};

const OVERWRITE_SIZE: usize = mem::size_of::<libc::__c_anonymous_ifr_ifru>();

/// A TUN device using the TUN/TAP Linux driver.
pub struct DeviceImpl {
    pub(crate) tun: Tun,
    pub(crate) vnet_hdr: bool,
    pub(crate) udp_gso: bool,
    flags: c_short,
    pub(crate) op_lock: Arc<RwLock<()>>,
}

impl DeviceImpl {
    /// Create a new `Device` for the given `Configuration`.
    pub(crate) fn new(config: DeviceConfig) -> std::io::Result<Self> {
        let dev_name = match config.dev_name.as_ref() {
            Some(tun_name) => {
                let tun_name = CString::new(tun_name.clone())?;

                if tun_name.as_bytes_with_nul().len() > IFNAMSIZ {
                    return Err(std::io::Error::new(
                        std::io::ErrorKind::InvalidInput,
                        "device name too long",
                    ));
                }

                Some(tun_name)
            }

            None => None,
        };

        // Create the device node if it is missing.
        // Silently ignore errors, let opening the device report an error.
        // This way, we don't fail if someone races us to create the device node.
        if let Ok(false) = std::fs::exists("/dev/net/tun") {
            std::fs::create_dir_all("/dev/net").ok();
            unsafe {
                libc::mknod(
                    c"/dev/net/tun".as_ptr(),
                    0o666 | libc::S_IFCHR,
                    libc::makedev(10, 200),
                );
            }
        }

        unsafe {
            let mut req: ifreq = mem::zeroed();

            if let Some(dev_name) = dev_name.as_ref() {
                ptr::copy_nonoverlapping(
                    dev_name.as_ptr() as *const c_char,
                    req.ifr_name.as_mut_ptr(),
                    dev_name.as_bytes_with_nul().len(),
                );
            }
            let multi_queue = config.multi_queue.unwrap_or(false);
            let device_type: c_short = config.layer.unwrap_or(Layer::L3).into();
            let iff_no_pi = IFF_NO_PI as c_short;
            let iff_vnet_hdr = libc::IFF_VNET_HDR as c_short;
            let iff_multi_queue = IFF_MULTI_QUEUE as c_short;
            let packet_information = config.packet_information.unwrap_or(false);
            let offload = config.offload.unwrap_or(false);
            req.ifr_ifru.ifru_flags = device_type
                | if packet_information { 0 } else { iff_no_pi }
                | if multi_queue { iff_multi_queue } else { 0 }
                | if offload { iff_vnet_hdr } else { 0 };

            let fd = libc::open(
                c"/dev/net/tun".as_ptr() as *const _,
                O_RDWR | libc::O_CLOEXEC,
                0,
            );
            let tun_fd = Fd::new(fd)?;
            if let Err(err) = tunsetiff(tun_fd.inner, &mut req as *mut _ as *mut _) {
                return Err(io::Error::from(err));
            }
            let (vnet_hdr, udp_gso) = if offload && libc::IFF_VNET_HDR != 0 {
                // tunTCPOffloads were added in Linux v2.6. We require their support if IFF_VNET_HDR is set.
                let tun_tcp_offloads = libc::TUN_F_CSUM | libc::TUN_F_TSO4 | libc::TUN_F_TSO6;
                let tun_udp_offloads = libc::TUN_F_USO4 | libc::TUN_F_USO6;
                if let Err(err) = tunsetoffload(tun_fd.inner, tun_tcp_offloads as _) {
                    log::warn!("unsupported offload: {err:?}");
                    (false, false)
                } else {
                    // tunUDPOffloads were added in Linux v6.2. We do not return an
                    // error if they are unsupported at runtime.
                    let rs =
                        tunsetoffload(tun_fd.inner, (tun_tcp_offloads | tun_udp_offloads) as _);
                    (true, rs.is_ok())
                }
            } else {
                // The TUN_F_* offload mask is device-wide state, not
                // per-fd. When attaching to a persistent TUN that a
                // previous process configured with offload=true, the
                // mask remains set across detaches: the kernel will
                // still deliver GSO aggregates to this new fd even
                // though IFF_VNET_HDR is not set, and the offload-
                // unaware caller will read them as oversized single
                // packets (ping survives, TCP bulk transfer fails).
                // Explicitly reset the mask. No-op on a freshly
                // created device (mask is already zero). Failure is
                // logged but not propagated, mirroring the
                // best-effort treatment of UDP offload above.
                if let Err(err) = tunsetoffload(tun_fd.inner, 0 as _) {
                    log::warn!("failed to clear TUN offload mask: {err:?}");
                }
                (false, false)
            };

            let device = DeviceImpl {
                tun: Tun::new(tun_fd),
                vnet_hdr,
                udp_gso,
                flags: req.ifr_ifru.ifru_flags,
                op_lock: Arc::new(RwLock::new(())),
            };
            Ok(device)
        }
    }
    unsafe fn set_tcp_offloads(&self) -> io::Result<()> {
        let tun_tcp_offloads = libc::TUN_F_CSUM | libc::TUN_F_TSO4 | libc::TUN_F_TSO6;
        tunsetoffload(self.as_raw_fd(), tun_tcp_offloads as _)
            .map(|_| ())
            .map_err(|e| e.into())
    }
    unsafe fn set_tcp_udp_offloads(&self) -> io::Result<()> {
        let tun_tcp_offloads = libc::TUN_F_CSUM | libc::TUN_F_TSO4 | libc::TUN_F_TSO6;
        let tun_udp_offloads = libc::TUN_F_USO4 | libc::TUN_F_USO6;
        tunsetoffload(self.as_raw_fd(), (tun_tcp_offloads | tun_udp_offloads) as _)
            .map(|_| ())
            .map_err(|e| e.into())
    }
    pub(crate) fn from_tun(tun: Tun) -> io::Result<Self> {
        Ok(Self {
            tun,
            vnet_hdr: false,
            udp_gso: false,
            flags: 0,
            op_lock: Arc::new(RwLock::new(())),
        })
    }

    /// # Prerequisites
    /// - The `IFF_MULTI_QUEUE` flag must be enabled.
    /// - The system must support network interface multi-queue functionality.
    ///
    /// # Description
    /// When multi-queue is enabled, create a new queue by duplicating an existing one.
    pub(crate) fn try_clone(&self) -> io::Result<DeviceImpl> {
        let flags = self.flags;
        if flags & (IFF_MULTI_QUEUE as c_short) != IFF_MULTI_QUEUE as c_short {
            return Err(io::Error::new(
                io::ErrorKind::Unsupported,
                "iff_multi_queue not enabled",
            ));
        }
        unsafe {
            let mut req = self.request()?;
            req.ifr_ifru.ifru_flags = flags;
            let fd = libc::open(
                c"/dev/net/tun".as_ptr() as *const _,
                O_RDWR | libc::O_CLOEXEC,
            );
            let tun_fd = Fd::new(fd)?;
            if let Err(err) = tunsetiff(tun_fd.inner, &mut req as *mut _ as *mut _) {
                return Err(io::Error::from(err));
            }
            let dev = DeviceImpl {
                tun: Tun::new(tun_fd),
                vnet_hdr: self.vnet_hdr,
                udp_gso: self.udp_gso,
                flags,
                op_lock: self.op_lock.clone(),
            };
            if dev.vnet_hdr {
                if dev.udp_gso {
                    dev.set_tcp_udp_offloads()?
                } else {
                    dev.set_tcp_offloads()?;
                }
            }

            Ok(dev)
        }
    }
    /// Returns whether UDP Generic Segmentation Offload (GSO) is enabled.
    ///
    /// This is determined by the `udp_gso` flag in the device.
    pub fn udp_gso(&self) -> bool {
        self.udp_gso
    }
    /// Returns whether TCP Generic Segmentation Offload (GSO) is enabled.
    ///
    /// In this implementation, this is represented by the `vnet_hdr` flag.
    pub fn tcp_gso(&self) -> bool {
        self.vnet_hdr
    }
    /// Sets the transmit queue length for the network interface.
    ///
    /// This method constructs an interface request (`ifreq`) structure,
    /// assigns the desired transmit queue length to the `ifru_metric` field,
    /// and calls the `change_tx_queue_len` function using the control file descriptor.
    /// If the underlying operation fails, an I/O error is returned.
    pub fn set_tx_queue_len(&self, tx_queue_len: u32) -> io::Result<()> {
        let _guard = self.op_lock.write().unwrap();
        unsafe {
            let mut ifreq = self.request()?;
            ifreq.ifr_ifru.ifru_metric = tx_queue_len as _;
            if let Err(err) = change_tx_queue_len(ctl()?.as_raw_fd(), &ifreq) {
                return Err(io::Error::from(err));
            }
        }
        Ok(())
    }
    /// Retrieves the current transmit queue length for the network interface.
    ///
    /// This function constructs an interface request structure and calls `tx_queue_len`
    /// to populate it with the current transmit queue length. The value is then returned.
    pub fn tx_queue_len(&self) -> io::Result<u32> {
        let _guard = self.op_lock.read().unwrap();
        unsafe {
            let mut ifreq = self.request()?;
            if let Err(err) = tx_queue_len(ctl()?.as_raw_fd(), &mut ifreq) {
                return Err(io::Error::from(err));
            }
            Ok(ifreq.ifr_ifru.ifru_metric as _)
        }
    }
    /// Make the device persistent.
    ///
    /// By default, TUN/TAP devices are destroyed when the process exits.
    /// Calling this method makes the device persist after the program terminates,
    /// allowing it to be reused by other processes.
    ///
    /// # Example
    ///
    /// ```no_run
    /// # #[cfg(all(target_os = "linux", not(target_env = "ohos")))]
    /// # {
    /// use tun_rs::DeviceBuilder;
    ///
    /// let dev = DeviceBuilder::new()
    ///     .name("persistent-tun")
    ///     .ipv4("10.0.0.1", 24, None)
    ///     .build_sync()?;
    ///
    /// // Make the device persistent so it survives after program exit
    /// dev.persist()?;
    /// println!("Device will persist after program exits");
    /// # }
    /// # Ok::<(), std::io::Error>(())
    /// ```
    pub fn persist(&self) -> io::Result<()> {
        let _guard = self.op_lock.write().unwrap();
        unsafe {
            if let Err(err) = tunsetpersist(self.as_raw_fd(), &1) {
                Err(io::Error::from(err))
            } else {
                Ok(())
            }
        }
    }

    /// Set the owner (UID) of the device.
    ///
    /// This allows non-root users to access the TUN/TAP device.
    ///
    /// # Example
    ///
    /// ```no_run
    /// # #[cfg(all(target_os = "linux", not(target_env = "ohos")))]
    /// # {
    /// use tun_rs::DeviceBuilder;
    ///
    /// let dev = DeviceBuilder::new()
    ///     .ipv4("10.0.0.1", 24, None)
    ///     .build_sync()?;
    ///
    /// // Set ownership to UID 1000 (typical first user on Linux)
    /// dev.user(1000)?;
    /// println!("Device ownership set to UID 1000");
    /// # }
    /// # Ok::<(), std::io::Error>(())
    /// ```
    pub fn user(&self, value: i32) -> io::Result<()> {
        let _guard = self.op_lock.write().unwrap();
        unsafe {
            if let Err(err) = tunsetowner(self.as_raw_fd(), &value) {
                Err(io::Error::from(err))
            } else {
                Ok(())
            }
        }
    }

    /// Set the group (GID) of the device.
    ///
    /// This allows members of a specific group to access the TUN/TAP device.
    ///
    /// # Example
    ///
    /// ```no_run
    /// # #[cfg(all(target_os = "linux", not(target_env = "ohos")))]
    /// # {
    /// use tun_rs::DeviceBuilder;
    ///
    /// let dev = DeviceBuilder::new()
    ///     .ipv4("10.0.0.1", 24, None)
    ///     .build_sync()?;
    ///
    /// // Set group ownership to GID 1000
    /// dev.group(1000)?;
    /// println!("Device group ownership set to GID 1000");
    /// # }
    /// # Ok::<(), std::io::Error>(())
    /// ```
    pub fn group(&self, value: i32) -> io::Result<()> {
        let _guard = self.op_lock.write().unwrap();
        unsafe {
            if let Err(err) = tunsetgroup(self.as_raw_fd(), &value) {
                Err(io::Error::from(err))
            } else {
                Ok(())
            }
        }
    }
    /// Sends multiple packets in a batch with GRO (Generic Receive Offload) coalescing.
    ///
    /// This method allows efficient transmission of multiple packets by batching them together
    /// and applying GRO optimizations. When offload is enabled, packets may be coalesced
    /// to reduce system call overhead and improve throughput.
    ///
    /// # Arguments
    ///
    /// * `gro_table` - A mutable reference to a [`GROTable`] that manages packet coalescing state.
    ///   This table can be reused across multiple calls to amortize allocation overhead.
    /// * `bufs` - A mutable slice of buffers containing the packets to send. Each buffer must
    ///   implement the [`ExpandBuffer`] trait.
    /// * `offset` - The byte offset within each buffer where the packet data begins.
    ///   Must be >= `VIRTIO_NET_HDR_LEN` to accommodate the virtio network header when offload is enabled.
    ///
    /// # Returns
    ///
    /// Returns the total number of bytes successfully sent, or an I/O error.
    ///
    /// # Example
    ///
    /// ```no_run
    /// # #[cfg(all(target_os = "linux", not(target_env = "ohos")))]
    /// # {
    /// use tun_rs::{DeviceBuilder, GROTable, VIRTIO_NET_HDR_LEN};
    ///
    /// let dev = DeviceBuilder::new()
    ///     .ipv4("10.0.0.1", 24, None)
    ///     .with(|builder| {
    ///         builder.offload(true) // Enable offload for GRO
    ///     })
    ///     .build_sync()?;
    ///
    /// let mut gro_table = GROTable::default();
    /// let offset = VIRTIO_NET_HDR_LEN;
    ///
    /// // Prepare packets to send
    /// let mut packet1 = vec![0u8; offset + 100]; // VIRTIO_NET_HDR + packet data
    /// let mut packet2 = vec![0u8; offset + 200];
    /// // Fill in packet data at offset...
    ///
    /// let mut bufs = vec![packet1, packet2];
    ///
    /// // Send all packets in one batch
    /// let bytes_sent = dev.send_multiple(&mut gro_table, &mut bufs, offset)?;
    /// println!("Sent {} bytes across {} packets", bytes_sent, bufs.len());
    /// # }
    /// # Ok::<(), std::io::Error>(())
    /// ```
    ///
    /// # Platform
    ///
    /// This method is only available on Linux.
    ///
    /// # Performance Notes
    ///
    /// - Use `IDEAL_BATCH_SIZE` for optimal batch size (typically 128 packets)
    /// - Reuse the same `GROTable` instance across calls to avoid allocations
    /// - Enable offload via `.offload(true)` in `DeviceBuilder` for best performance
    pub fn send_multiple<B: ExpandBuffer>(
        &self,
        gro_table: &mut GROTable,
        bufs: &mut [B],
        offset: usize,
    ) -> io::Result<usize> {
        self.send_multiple0(gro_table, bufs, offset, |tun, buf| tun.send(buf))
    }
    pub(crate) fn send_multiple0<B: ExpandBuffer, W: FnMut(&Tun, &[u8]) -> io::Result<usize>>(
        &self,
        gro_table: &mut GROTable,
        bufs: &mut [B],
        mut offset: usize,
        mut write_f: W,
    ) -> io::Result<usize> {
        gro_table.reset();
        if bufs.is_empty() {
            return Ok(0);
        }
        if bufs.len() > u16::MAX as usize {
            return Err(io::Error::new(
                io::ErrorKind::InvalidInput,
                "too many packet buffers",
            ));
        }
        if self.vnet_hdr {
            handle_gro(
                bufs,
                offset,
                &mut gro_table.tcp_gro_table,
                &mut gro_table.udp_gro_table,
                self.udp_gso,
                &mut gro_table.to_write,
            )?;
            offset -= VIRTIO_NET_HDR_LEN;
        } else {
            for i in 0..bufs.len() {
                gro_table.to_write.push(i);
            }
        }

        let mut total = 0;
        let mut err = Ok(());
        for buf_idx in &gro_table.to_write {
            let Some(buf) = bufs[*buf_idx].as_ref().get(offset..) else {
                return Err(io::Error::new(
                    io::ErrorKind::InvalidInput,
                    "invalid offset",
                ));
            };
            match write_f(&self.tun, buf) {
                Ok(n) => {
                    total += n;
                }
                Err(e) => {
                    if let Some(code) = e.raw_os_error() {
                        if libc::EBADFD == code {
                            return Err(e);
                        }
                    }
                    err = Err(e)
                }
            }
        }
        err?;
        Ok(total)
    }
    /// Receives multiple packets in a batch with GSO (Generic Segmentation Offload) splitting.
    ///
    /// When offload is enabled, this method can receive large GSO packets from the TUN device
    /// and automatically split them into MTU-sized segments, significantly improving receive
    /// performance for high-bandwidth traffic.
    ///
    /// # Arguments
    ///
    /// * `original_buffer` - A mutable buffer to store the raw received data, including the
    ///   virtio network header and the potentially large GSO packet. Recommended size is
    ///   `VIRTIO_NET_HDR_LEN + 65535` bytes.
    /// * `bufs` - A mutable slice of buffers to store the segmented packets. Each buffer will
    ///   receive one MTU-sized packet after GSO splitting.
    /// * `sizes` - A mutable slice to store the actual size of each packet in `bufs`.
    ///   Must have the same length as `bufs`.
    /// * `offset` - The byte offset within each output buffer where packet data should be written.
    ///   This allows for pre-allocated header space.
    ///
    /// # Returns
    ///
    /// Returns the number of packets successfully received and split, or an I/O error.
    ///
    /// # Example
    ///
    /// ```no_run
    /// # #[cfg(all(target_os = "linux", not(target_env = "ohos")))]
    /// # {
    /// use tun_rs::{DeviceBuilder, IDEAL_BATCH_SIZE, VIRTIO_NET_HDR_LEN};
    ///
    /// let dev = DeviceBuilder::new()
    ///     .ipv4("10.0.0.1", 24, None)
    ///     .with(|builder| {
    ///         builder.offload(true) // Enable offload for GSO
    ///     })
    ///     .build_sync()?;
    ///
    /// // Buffer for the raw received packet (with virtio header)
    /// let mut original_buffer = vec![0u8; VIRTIO_NET_HDR_LEN + 65535];
    ///
    /// // Output buffers for segmented packets
    /// let mut bufs = vec![vec![0u8; 1500]; IDEAL_BATCH_SIZE];
    /// let mut sizes = vec![0usize; IDEAL_BATCH_SIZE];
    /// let offset = 0;
    ///
    /// loop {
    ///     // Receive and segment packets
    ///     let num_packets = dev.recv_multiple(&mut original_buffer, &mut bufs, &mut sizes, offset)?;
    ///
    ///     // Process each segmented packet
    ///     for i in 0..num_packets {
    ///         let packet = &bufs[i][offset..offset + sizes[i]];
    ///         println!("Received packet {}: {} bytes", i, sizes[i]);
    ///         // Process packet...
    ///     }
    /// }
    /// # }
    /// # Ok::<(), std::io::Error>(())
    /// ```
    ///
    /// # Platform
    ///
    /// This method is only available on Linux.
    ///
    /// # Performance Notes
    ///
    /// - Use `IDEAL_BATCH_SIZE` (128) for the number of output buffers
    /// - A single `recv_multiple` call may return multiple MTU-sized packets from one large GSO packet
    /// - The performance benefit is most noticeable with TCP traffic using large send/receive windows
    pub fn recv_multiple<B: AsRef<[u8]> + AsMut<[u8]>>(
        &self,
        original_buffer: &mut [u8],
        bufs: &mut [B],
        sizes: &mut [usize],
        offset: usize,
    ) -> io::Result<usize> {
        self.recv_multiple0(original_buffer, bufs, sizes, offset, |tun, buf| {
            tun.recv(buf)
        })
    }
    pub(crate) fn recv_multiple0<
        B: AsRef<[u8]> + AsMut<[u8]>,
        R: Fn(&Tun, &mut [u8]) -> io::Result<usize>,
    >(
        &self,
        original_buffer: &mut [u8],
        bufs: &mut [B],
        sizes: &mut [usize],
        offset: usize,
        read_f: R,
    ) -> io::Result<usize> {
        if bufs.is_empty() || bufs.len() != sizes.len() {
            return Err(io::Error::other("bufs error"));
        }
        if self.vnet_hdr {
            let len = read_f(&self.tun, original_buffer)?;
            if len <= VIRTIO_NET_HDR_LEN {
                Err(io::Error::other(format!(
                    "length of packet ({len}) <= VIRTIO_NET_HDR_LEN ({VIRTIO_NET_HDR_LEN})",
                )))?
            }
            let hdr = VirtioNetHdr::decode(&original_buffer[..VIRTIO_NET_HDR_LEN])?;
            self.handle_virtio_read(
                hdr,
                &mut original_buffer[VIRTIO_NET_HDR_LEN..len],
                bufs,
                sizes,
                offset,
            )
        } else {
            let Some(buf) = bufs[0].as_mut().get_mut(offset..) else {
                return Err(io::Error::new(
                    io::ErrorKind::InvalidInput,
                    "invalid offset",
                ));
            };
            let len = read_f(&self.tun, buf)?;
            sizes[0] = len;
            Ok(1)
        }
    }
    /// https://github.com/WireGuard/wireguard-go/blob/12269c2761734b15625017d8565745096325392f/tun/tun_linux.go#L375
    /// handleVirtioRead splits in into bufs, leaving offset bytes at the front of
    /// each buffer. It mutates sizes to reflect the size of each element of bufs,
    /// and returns the number of packets read.
    pub(crate) fn handle_virtio_read<B: AsRef<[u8]> + AsMut<[u8]>>(
        &self,
        mut hdr: VirtioNetHdr,
        input: &mut [u8],
        bufs: &mut [B],
        sizes: &mut [usize],
        offset: usize,
    ) -> io::Result<usize> {
        if sizes.len() < bufs.len() {
            return Err(io::Error::other("sizes must be at least as long as bufs"));
        }
        if bufs.len() > u16::MAX as usize {
            return Err(io::Error::new(
                io::ErrorKind::InvalidInput,
                "too many packet buffers",
            ));
        }
        for buf in bufs.iter() {
            if offset > buf.as_ref().len() {
                return Err(io::Error::new(
                    io::ErrorKind::InvalidInput,
                    "invalid offset",
                ));
            }
        }
        let len = input.len();
        if hdr.gso_type == VIRTIO_NET_HDR_GSO_NONE {
            if hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM != 0 {
                // This means CHECKSUM_PARTIAL in skb context. We are responsible
                // for computing the checksum starting at hdr.csumStart and placing
                // at hdr.csumOffset.
                gso_none_checksum(input, hdr.csum_start, hdr.csum_offset);
            }
            if bufs[0].as_ref()[offset..].len() < len {
                Err(io::Error::other(format!(
                    "read len {len} overflows bufs element len {}",
                    bufs[0].as_ref().len()
                )))?
            }
            sizes[0] = len;
            bufs[0].as_mut()[offset..offset + len].copy_from_slice(input);
            return Ok(1);
        }
        if hdr.gso_size == 0 {
            return Err(io::Error::new(
                io::ErrorKind::InvalidInput,
                "virtioNetHdr.gsoSize must be non-zero",
            ));
        }
        if hdr.gso_type != VIRTIO_NET_HDR_GSO_TCPV4
            && hdr.gso_type != VIRTIO_NET_HDR_GSO_TCPV6
            && hdr.gso_type != VIRTIO_NET_HDR_GSO_UDP_L4
        {
            Err(io::Error::other(format!(
                "unsupported virtio GSO type: {}",
                hdr.gso_type
            )))?
        }
        let ip_version = input[0] >> 4;
        match ip_version {
            4 => {
                if hdr.gso_type != VIRTIO_NET_HDR_GSO_TCPV4
                    && hdr.gso_type != VIRTIO_NET_HDR_GSO_UDP_L4
                {
                    Err(io::Error::other(format!(
                        "ip header version: 4, GSO type: {}",
                        hdr.gso_type
                    )))?
                }
            }
            6 => {
                if hdr.gso_type != VIRTIO_NET_HDR_GSO_TCPV6
                    && hdr.gso_type != VIRTIO_NET_HDR_GSO_UDP_L4
                {
                    Err(io::Error::other(format!(
                        "ip header version: 6, GSO type: {}",
                        hdr.gso_type
                    )))?
                }
            }
            ip_version => Err(io::Error::other(format!(
                "invalid ip header version: {ip_version}"
            )))?,
        }
        // Don't trust hdr.hdrLen from the kernel as it can be equal to the length
        // of the entire first packet when the kernel is handling it as part of a
        // FORWARD path. Instead, parse the transport header length and add it onto
        // csumStart, which is synonymous for IP header length.
        if hdr.gso_type == VIRTIO_NET_HDR_GSO_UDP_L4 {
            hdr.hdr_len = hdr.csum_start + 8
        } else {
            if len <= hdr.csum_start as usize + 12 {
                Err(io::Error::other("packet is too short"))?
            }

            let tcp_h_len = ((input[hdr.csum_start as usize + 12] as u16) >> 4) * 4;
            if !(20..=60).contains(&tcp_h_len) {
                // A TCP header must be between 20 and 60 bytes in length.
                Err(io::Error::other(format!(
                    "tcp header len is invalid: {tcp_h_len}"
                )))?
            }
            hdr.hdr_len = hdr.csum_start + tcp_h_len
        }
        if len < hdr.hdr_len as usize {
            Err(io::Error::other(format!(
                "length of packet ({len}) < virtioNetHdr.hdr_len ({})",
                hdr.hdr_len
            )))?
        }
        if hdr.hdr_len < hdr.csum_start {
            Err(io::Error::other(format!(
                "virtioNetHdr.hdrLen ({}) < virtioNetHdr.csumStart ({})",
                hdr.hdr_len, hdr.csum_start
            )))?
        }
        let c_sum_at = (hdr.csum_start + hdr.csum_offset) as usize;
        if c_sum_at + 1 >= len {
            Err(io::Error::other(format!(
                "end of checksum offset ({}) exceeds packet length ({len})",
                c_sum_at + 1,
            )))?
        }
        gso_split(input, hdr, bufs, sizes, offset, ip_version == 6)
    }
    pub fn remove_address_v6_impl(&self, addr: Ipv6Addr, prefix: u8) -> io::Result<()> {
        unsafe {
            let if_index = self.if_index_impl()?;
            let ctl = ctl_v6()?;
            let mut ifrv6: in6_ifreq = mem::zeroed();
            ifrv6.ifr6_ifindex = if_index as i32;
            ifrv6.ifr6_prefixlen = prefix as _;
            ifrv6.ifr6_addr = sockaddr_union::from(std::net::SocketAddr::new(addr.into(), 0))
                .addr6
                .sin6_addr;
            if let Err(err) = siocdifaddr_in6(ctl.as_raw_fd(), &ifrv6) {
                return Err(io::Error::from(err));
            }
        }
        Ok(())
    }
}

impl DeviceImpl {
    /// Prepare a new request.
    unsafe fn request(&self) -> io::Result<ifreq> {
        request(&self.name_impl()?)
    }
    fn set_address_v4(&self, addr: Ipv4Addr) -> io::Result<()> {
        unsafe {
            let mut req = self.request()?;
            ipaddr_to_sockaddr(addr, 0, &mut req.ifr_ifru.ifru_addr, OVERWRITE_SIZE);
            if let Err(err) = siocsifaddr(ctl()?.as_raw_fd(), &req) {
                return Err(io::Error::from(err));
            }
        }
        Ok(())
    }
    fn set_netmask(&self, value: Ipv4Addr) -> io::Result<()> {
        unsafe {
            let mut req = self.request()?;
            ipaddr_to_sockaddr(value, 0, &mut req.ifr_ifru.ifru_netmask, OVERWRITE_SIZE);
            if let Err(err) = siocsifnetmask(ctl()?.as_raw_fd(), &req) {
                return Err(io::Error::from(err));
            }
            Ok(())
        }
    }

    fn set_destination(&self, value: Ipv4Addr) -> io::Result<()> {
        unsafe {
            let mut req = self.request()?;
            ipaddr_to_sockaddr(value, 0, &mut req.ifr_ifru.ifru_dstaddr, OVERWRITE_SIZE);
            if let Err(err) = siocsifdstaddr(ctl()?.as_raw_fd(), &req) {
                return Err(io::Error::from(err));
            }
            Ok(())
        }
    }

    /// Retrieves the name of the network interface.
    pub(crate) fn name_impl(&self) -> io::Result<String> {
        unsafe { name(self.as_raw_fd()) }
    }

    fn ifru_flags(&self) -> io::Result<i16> {
        unsafe {
            let ctl = ctl()?;
            let mut req = self.request()?;

            if let Err(err) = siocgifflags(ctl.as_raw_fd(), &mut req) {
                return Err(io::Error::from(err));
            }
            Ok(req.ifr_ifru.ifru_flags)
        }
    }

    fn remove_all_address_v4(&self) -> io::Result<()> {
        let interface = netconfig_rs::Interface::try_from_index(self.if_index_impl()?)
            .map_err(io::Error::from)?;
        let list = interface.addresses().map_err(io::Error::from)?;
        for x in list {
            if x.addr().is_ipv4() {
                interface.remove_address(x).map_err(io::Error::from)?;
            }
        }
        Ok(())
    }
}

//Public User Interface
impl DeviceImpl {
    /// Retrieves the name of the network interface.
    pub fn name(&self) -> io::Result<String> {
        let _guard = self.op_lock.read().unwrap();
        self.name_impl()
    }
    pub fn remove_address_v6(&self, addr: Ipv6Addr, prefix: u8) -> io::Result<()> {
        let _guard = self.op_lock.write().unwrap();
        self.remove_address_v6_impl(addr, prefix)
    }
    /// Sets a new name for the network interface.
    ///
    /// This function converts the provided name into a C-compatible string,
    /// checks that its length does not exceed the maximum allowed (IFNAMSIZ),
    /// and then copies it into an interface request structure. It then uses a system call
    /// (via `siocsifname`) to apply the new name.
    ///
    /// # Example
    ///
    /// ```no_run
    /// # #[cfg(all(target_os = "linux", not(target_env = "ohos")))]
    /// # {
    /// use tun_rs::DeviceBuilder;
    ///
    /// let dev = DeviceBuilder::new()
    ///     .name("tun0")
    ///     .ipv4("10.0.0.1", 24, None)
    ///     .build_sync()?;
    ///
    /// // Rename the device
    /// dev.set_name("vpn-tun")?;
    /// println!("Device renamed to vpn-tun");
    /// # }
    /// # Ok::<(), std::io::Error>(())
    /// ```
    pub fn set_name(&self, value: &str) -> io::Result<()> {
        let _guard = self.op_lock.write().unwrap();
        unsafe {
            let tun_name = CString::new(value)?;

            if tun_name.as_bytes_with_nul().len() > IFNAMSIZ {
                return Err(io::Error::new(io::ErrorKind::InvalidInput, "name too long"));
            }

            let mut req = self.request()?;
            ptr::copy_nonoverlapping(
                tun_name.as_ptr() as *const c_char,
                req.ifr_ifru.ifru_newname.as_mut_ptr(),
                value.len(),
            );

            if let Err(err) = siocsifname(ctl()?.as_raw_fd(), &req) {
                return Err(io::Error::from(err));
            }

            Ok(())
        }
    }
    /// Checks whether the network interface is currently running.
    ///
    /// The interface is considered running if both the IFF_UP and IFF_RUNNING flags are set.
    pub fn is_running(&self) -> io::Result<bool> {
        let _guard = self.op_lock.read().unwrap();
        let flags = self.ifru_flags()?;
        Ok(flags & (IFF_UP | IFF_RUNNING) as c_short == (IFF_UP | IFF_RUNNING) as c_short)
    }
    /// Enables or disables the network interface.
    ///
    /// If `value` is true, the interface is enabled by setting the IFF_UP and IFF_RUNNING flags.
    /// If false, the IFF_UP flag is cleared. The change is applied using a system call.
    pub fn enabled(&self, value: bool) -> io::Result<()> {
        let _guard = self.op_lock.write().unwrap();
        unsafe {
            let ctl = ctl()?;
            let mut req = self.request()?;

            if let Err(err) = siocgifflags(ctl.as_raw_fd(), &mut req) {
                return Err(io::Error::from(err));
            }

            if value {
                req.ifr_ifru.ifru_flags |= (IFF_UP | IFF_RUNNING) as c_short;
            } else {
                req.ifr_ifru.ifru_flags &= !(IFF_UP as c_short);
            }

            if let Err(err) = siocsifflags(ctl.as_raw_fd(), &req) {
                return Err(io::Error::from(err));
            }

            Ok(())
        }
    }
    /// Retrieves the broadcast address of the network interface.
    ///
    /// This function populates an interface request with the broadcast address via a system call,
    /// converts it into a sockaddr structure, and then extracts the IP address.
    ///
    /// # Example
    ///
    /// ```no_run
    /// # #[cfg(all(target_os = "linux", not(target_env = "ohos")))]
    /// # {
    /// use tun_rs::DeviceBuilder;
    ///
    /// let dev = DeviceBuilder::new()
    ///     .ipv4("10.0.0.1", 24, None)
    ///     .build_sync()?;
    ///
    /// // Get the broadcast address
    /// let broadcast = dev.broadcast()?;
    /// println!("Broadcast address: {}", broadcast);
    /// # }
    /// # Ok::<(), std::io::Error>(())
    /// ```
    pub fn broadcast(&self) -> io::Result<IpAddr> {
        let _guard = self.op_lock.read().unwrap();
        unsafe {
            let mut req = self.request()?;
            if let Err(err) = siocgifbrdaddr(ctl()?.as_raw_fd(), &mut req) {
                return Err(io::Error::from(err));
            }
            let sa = sockaddr_union::from(req.ifr_ifru.ifru_broadaddr);
            Ok(std::net::SocketAddr::try_from(sa)?.ip())
        }
    }
    /// Sets the broadcast address of the network interface.
    ///
    /// This function converts the given IP address into a sockaddr structure (with a specified overwrite size)
    /// and then applies it to the interface via a system call.
    pub fn set_broadcast(&self, value: IpAddr) -> io::Result<()> {
        let _guard = self.op_lock.write().unwrap();
        unsafe {
            let mut req = self.request()?;
            ipaddr_to_sockaddr(value, 0, &mut req.ifr_ifru.ifru_broadaddr, OVERWRITE_SIZE);
            if let Err(err) = siocsifbrdaddr(ctl()?.as_raw_fd(), &req) {
                return Err(io::Error::from(err));
            }
            Ok(())
        }
    }
    /// Sets the IPv4 network address, netmask, and an optional destination address.
    /// Remove all previous set IPv4 addresses and set the specified address.
    ///
    /// # Example
    ///
    /// ```no_run
    /// # #[cfg(all(target_os = "linux", not(target_env = "ohos")))]
    /// # {
    /// use tun_rs::DeviceBuilder;
    ///
    /// let dev = DeviceBuilder::new()
    ///     .ipv4("10.0.0.1", 24, None)
    ///     .build_sync()?;
    ///
    /// // Change the primary IPv4 address
    /// dev.set_network_address("10.1.0.1", 24, None)?;
    /// println!("Updated device address to 10.1.0.1/24");
    /// # }
    /// # Ok::<(), std::io::Error>(())
    /// ```
    pub fn set_network_address<IPv4: ToIpv4Address, Netmask: ToIpv4Netmask>(
        &self,
        address: IPv4,
        netmask: Netmask,
        destination: Option<IPv4>,
    ) -> io::Result<()> {
        let _guard = self.op_lock.write().unwrap();
        self.remove_all_address_v4()?;
        self.set_address_v4(address.ipv4()?)?;
        self.set_netmask(netmask.netmask()?)?;
        if let Some(destination) = destination {
            self.set_destination(destination.ipv4()?)?;
        }
        Ok(())
    }
    /// Add IPv4 network address and netmask to the interface.
    ///
    /// This allows multiple IPv4 addresses on a single TUN/TAP device.
    ///
    /// # Example
    ///
    /// ```no_run
    /// # #[cfg(all(target_os = "linux", not(target_env = "ohos")))]
    /// # {
    /// use tun_rs::DeviceBuilder;
    ///
    /// let dev = DeviceBuilder::new()
    ///     .ipv4("10.0.0.1", 24, None)
    ///     .build_sync()?;
    ///
    /// // Add additional IPv4 addresses
    /// dev.add_address_v4("10.0.1.1", 24)?;
    /// dev.add_address_v4("10.0.2.1", 24)?;
    /// println!("Added multiple IPv4 addresses");
    /// # }
    /// # Ok::<(), std::io::Error>(())
    /// ```
    pub fn add_address_v4<IPv4: ToIpv4Address, Netmask: ToIpv4Netmask>(
        &self,
        address: IPv4,
        netmask: Netmask,
    ) -> io::Result<()> {
        let _guard = self.op_lock.write().unwrap();
        let interface = netconfig_rs::Interface::try_from_index(self.if_index_impl()?)
            .map_err(io::Error::from)?;
        interface
            .add_address(IpNet::new_assert(address.ipv4()?.into(), netmask.prefix()?))
            .map_err(io::Error::from)
    }
    /// Removes an IP address from the interface.
    ///
    /// For IPv4 addresses, it iterates over the current addresses and if a match is found,
    /// resets the address to `0.0.0.0` (unspecified).
    /// For IPv6 addresses, it retrieves the interface addresses by name and removes the matching address,
    /// taking into account its prefix length.
    ///
    /// # Example
    ///
    /// ```no_run
    /// # #[cfg(all(target_os = "linux", not(target_env = "ohos")))]
    /// # {
    /// use std::net::IpAddr;
    /// use tun_rs::DeviceBuilder;
    ///
    /// let dev = DeviceBuilder::new()
    ///     .ipv4("10.0.0.1", 24, None)
    ///     .build_sync()?;
    ///
    /// // Add an additional address
    /// dev.add_address_v4("10.0.1.1", 24)?;
    ///
    /// // Later, remove it
    /// dev.remove_address("10.0.1.1".parse::<IpAddr>().unwrap())?;
    /// println!("Removed address 10.0.1.1");
    /// # }
    /// # Ok::<(), std::io::Error>(())
    /// ```
    pub fn remove_address(&self, addr: IpAddr) -> io::Result<()> {
        let _guard = self.op_lock.write().unwrap();
        match addr {
            IpAddr::V4(_) => {
                let interface = netconfig_rs::Interface::try_from_index(self.if_index_impl()?)
                    .map_err(io::Error::from)?;
                let list = interface.addresses().map_err(io::Error::from)?;
                for x in list {
                    if x.addr() == addr {
                        interface.remove_address(x).map_err(io::Error::from)?;
                    }
                }
            }
            IpAddr::V6(addr_v6) => {
                let addrs = crate::platform::get_if_addrs_by_name(self.name_impl()?)?;
                for x in addrs {
                    if let Some(ip_addr) = x.address.ip_addr() {
                        if ip_addr == addr {
                            if let Some(netmask) = x.address.netmask() {
                                let prefix = ipnet::ip_mask_to_prefix(netmask).unwrap_or(0);
                                self.remove_address_v6_impl(addr_v6, prefix)?
                            }
                        }
                    }
                }
            }
        }
        Ok(())
    }
    /// Adds an IPv6 address to the interface.
    ///
    /// This function creates an `in6_ifreq` structure, fills in the interface index,
    /// prefix length, and IPv6 address (converted into a sockaddr structure),
    /// and then applies it using a system call.
    ///
    /// # Example
    ///
    /// ```no_run
    /// # #[cfg(all(target_os = "linux", not(target_env = "ohos")))]
    /// # {
    /// use tun_rs::DeviceBuilder;
    ///
    /// let dev = DeviceBuilder::new()
    ///     .ipv4("10.0.0.1", 24, None)
    ///     .build_sync()?;
    ///
    /// // Add IPv6 addresses
    /// dev.add_address_v6("fd00::1", 64)?;
    /// dev.add_address_v6("fd00::2", 64)?;
    /// println!("Added IPv6 addresses");
    /// # }
    /// # Ok::<(), std::io::Error>(())
    /// ```
    pub fn add_address_v6<IPv6: ToIpv6Address, Netmask: ToIpv6Netmask>(
        &self,
        addr: IPv6,
        netmask: Netmask,
    ) -> io::Result<()> {
        let _guard = self.op_lock.write().unwrap();
        unsafe {
            let if_index = self.if_index_impl()?;
            let ctl = ctl_v6()?;
            let mut ifrv6: in6_ifreq = mem::zeroed();
            ifrv6.ifr6_ifindex = if_index as i32;
            ifrv6.ifr6_prefixlen = netmask.prefix()? as u32;
            ifrv6.ifr6_addr =
                sockaddr_union::from(std::net::SocketAddr::new(addr.ipv6()?.into(), 0))
                    .addr6
                    .sin6_addr;
            if let Err(err) = siocsifaddr_in6(ctl.as_raw_fd(), &ifrv6) {
                return Err(io::Error::from(err));
            }
        }
        Ok(())
    }
    /// Retrieves the current MTU (Maximum Transmission Unit) for the interface.
    ///
    /// This function constructs an interface request and uses a system call (via `siocgifmtu`)
    /// to obtain the MTU. The result is then converted to a u16.
    pub fn mtu(&self) -> io::Result<u16> {
        let _guard = self.op_lock.read().unwrap();
        unsafe {
            let mut req = self.request()?;

            if let Err(err) = siocgifmtu(ctl()?.as_raw_fd(), &mut req) {
                return Err(io::Error::from(err));
            }

            req.ifr_ifru
                .ifru_mtu
                .try_into()
                .map_err(|e| io::Error::new(io::ErrorKind::InvalidData, format!("{e:?}")))
        }
    }
    /// Sets the MTU (Maximum Transmission Unit) for the interface.
    ///
    /// This function creates an interface request, sets the `ifru_mtu` field to the new value,
    /// and then applies it via a system call.
    ///
    /// # Example
    ///
    /// ```no_run
    /// # #[cfg(all(target_os = "linux", not(target_env = "ohos")))]
    /// # {
    /// use tun_rs::DeviceBuilder;
    ///
    /// let dev = DeviceBuilder::new()
    ///     .ipv4("10.0.0.1", 24, None)
    ///     .mtu(1400)
    ///     .build_sync()?;
    ///
    /// // Change MTU to accommodate larger packets
    /// dev.set_mtu(9000)?; // Jumbo frames
    /// println!("MTU set to 9000 bytes");
    /// # }
    /// # Ok::<(), std::io::Error>(())
    /// ```
    pub fn set_mtu(&self, value: u16) -> io::Result<()> {
        let _guard = self.op_lock.write().unwrap();
        unsafe {
            let mut req = self.request()?;
            req.ifr_ifru.ifru_mtu = value as i32;

            if let Err(err) = siocsifmtu(ctl()?.as_raw_fd(), &req) {
                return Err(io::Error::from(err));
            }
            Ok(())
        }
    }
    /// Sets the MAC (hardware) address for the interface.
    ///
    /// This function constructs an interface request and copies the provided MAC address
    /// into the hardware address field. It then applies the change via a system call.
    /// This operation is typically supported only for TAP devices.
    pub fn set_mac_address(&self, eth_addr: [u8; ETHER_ADDR_LEN as usize]) -> io::Result<()> {
        let _guard = self.op_lock.write().unwrap();
        unsafe {
            let mut req = self.request()?;
            req.ifr_ifru.ifru_hwaddr.sa_family = ARPHRD_ETHER;
            req.ifr_ifru.ifru_hwaddr.sa_data[0..ETHER_ADDR_LEN as usize]
                .copy_from_slice(eth_addr.map(|c| c as _).as_slice());
            if let Err(err) = siocsifhwaddr(ctl()?.as_raw_fd(), &req) {
                return Err(io::Error::from(err));
            }
            Ok(())
        }
    }
    /// Retrieves the MAC (hardware) address of the interface.
    ///
    /// This function queries the MAC address by the interface name using a helper function.
    /// An error is returned if the MAC address cannot be found.
    pub fn mac_address(&self) -> io::Result<[u8; ETHER_ADDR_LEN as usize]> {
        let _guard = self.op_lock.read().unwrap();
        unsafe {
            let mut req = self.request()?;

            siocgifhwaddr(ctl()?.as_raw_fd(), &mut req).map_err(io::Error::from)?;

            let hw = &req.ifr_ifru.ifru_hwaddr.sa_data;

            let mut mac = [0u8; ETHER_ADDR_LEN as usize];
            for (i, b) in hw.iter().take(6).enumerate() {
                mac[i] = *b as u8;
            }

            Ok(mac)
        }
    }
}

unsafe fn name(fd: RawFd) -> io::Result<String> {
    let mut req: ifreq = mem::zeroed();
    if let Err(err) = tungetiff(fd, &mut req as *mut _ as *mut _) {
        return Err(io::Error::from(err));
    }
    let c_str = std::ffi::CStr::from_ptr(req.ifr_name.as_ptr() as *const c_char);
    let tun_name = c_str.to_string_lossy().into_owned();
    Ok(tun_name)
}

unsafe fn request(name: &str) -> io::Result<ifreq> {
    let mut req: ifreq = mem::zeroed();
    ptr::copy_nonoverlapping(
        name.as_ptr() as *const c_char,
        req.ifr_name.as_mut_ptr(),
        name.len(),
    );
    Ok(req)
}

impl From<Layer> for c_short {
    fn from(layer: Layer) -> Self {
        match layer {
            Layer::L2 => IFF_TAP as c_short,
            Layer::L3 => IFF_TUN as c_short,
        }
    }
}