netlink_sys/

socket.rs

// SPDX-License-Identifier: MIT

use std::{
    io::{Error, Result},
    mem,
    os::unix::io::{AsRawFd, FromRawFd, RawFd},
};

use crate::SocketAddr;

/// A netlink socket.
///
/// # Example
///
/// In this example we:
///
/// 1. open a new socket
/// 2. send a message to the kernel
/// 3. read the reponse
///
/// ```rust
/// use netlink_sys::{protocols::NETLINK_ROUTE, Socket, SocketAddr};
/// use std::process;
///
/// // open a new socket for the NETLINK_ROUTE subsystem (see "man 7 rtnetlink")
/// let mut socket = Socket::new(NETLINK_ROUTE).unwrap();
/// // address of the remote peer we'll send a message to. This particular address is for the kernel
/// let kernel_addr = SocketAddr::new(0, 0);
/// // this is a valid message for listing the network links on the system
/// let pkt = vec![
///     0x14, 0x00, 0x00, 0x00, 0x12, 0x00, 0x01, 0x03, 0xfd, 0xfe, 0x38, 0x5c, 0x00, 0x00, 0x00,
///     0x00, 0x00, 0x00, 0x00, 0x00,
/// ];
/// // send the message to the kernel
/// let n_sent = socket.send_to(&pkt[..], &kernel_addr, 0).unwrap();
/// assert_eq!(n_sent, pkt.len());
/// // buffer for receiving the response
/// let mut buf = vec![0; 4096];
/// loop {
///     // receive a datagram
///     let (n_received, sender_addr) = socket.recv_from(&mut &mut buf[..], 0).unwrap();
///     assert_eq!(sender_addr, kernel_addr);
///     println!("received datagram {:?}", &buf[..n_received]);
///     if buf[4] == 2 && buf[5] == 0 {
///         println!("the kernel responded with an error");
///         return;
///     }
///     if buf[4] == 3 && buf[5] == 0 {
///         println!("end of dump");
///         return;
///     }
/// }
/// ```
#[derive(Clone, Debug)]
pub struct Socket(RawFd);

impl AsRawFd for Socket {
    fn as_raw_fd(&self) -> RawFd {
        self.0
    }
}

impl FromRawFd for Socket {
    unsafe fn from_raw_fd(fd: RawFd) -> Self {
        Socket(fd)
    }
}

impl Drop for Socket {
    fn drop(&mut self) {
        unsafe { libc::close(self.0) };
    }
}

impl Socket {
    /// Open a new socket for the given netlink subsystem. `protocol` must be
    /// one of the [`netlink_sys::protocols`][protos] constants.
    ///
    /// [protos]: crate::protocols
    pub fn new(protocol: isize) -> Result<Self> {
        let res = unsafe {
            libc::socket(
                libc::PF_NETLINK,
                libc::SOCK_DGRAM | libc::SOCK_CLOEXEC,
                protocol as libc::c_int,
            )
        };
        if res < 0 {
            return Err(Error::last_os_error());
        }
        Ok(Socket(res))
    }

    /// Bind the socket to the given address
    pub fn bind(&mut self, addr: &SocketAddr) -> Result<()> {
        let (addr_ptr, addr_len) = addr.as_raw();
        let res = unsafe { libc::bind(self.0, addr_ptr, addr_len) };
        if res < 0 {
            return Err(Error::last_os_error());
        }
        Ok(())
    }

    /// Bind the socket to an address assigned by the kernel, and return that
    /// address.
    pub fn bind_auto(&mut self) -> Result<SocketAddr> {
        let mut addr = SocketAddr::new(0, 0);
        self.bind(&addr)?;
        self.get_address(&mut addr)?;
        Ok(addr)
    }

    /// Get the socket address
    pub fn get_address(&self, addr: &mut SocketAddr) -> Result<()> {
        let (addr_ptr, mut addr_len) = addr.as_raw_mut();
        let addr_len_copy = addr_len;
        let addr_len_ptr = &mut addr_len as *mut libc::socklen_t;
        let res = unsafe { libc::getsockname(self.0, addr_ptr, addr_len_ptr) };
        if res < 0 {
            return Err(Error::last_os_error());
        }
        assert_eq!(addr_len, addr_len_copy);
        Ok(())
    }

    // when building with --features smol we don't need this
    #[allow(dead_code)]
    /// Make this socket non-blocking
    pub fn set_non_blocking(&self, non_blocking: bool) -> Result<()> {
        let mut non_blocking = non_blocking as libc::c_int;
        let res =
            unsafe { libc::ioctl(self.0, libc::FIONBIO, &mut non_blocking) };
        if res < 0 {
            return Err(Error::last_os_error());
        }
        Ok(())
    }

    /// Connect the socket to the given address. Netlink is a connection-less
    /// protocol, so a socket can communicate with multiple peers with the
    /// [`Socket::send_to`] and [`Socket::recv_from`] methods. However, if the
    /// socket only needs to communicate with one peer, it is convenient not
    /// to have to bother with the peer address. This is what `connect` is
    /// for. After calling `connect`, [`Socket::send`] and [`Socket::recv`]
    /// respectively send and receive datagrams to and from `remote_addr`.
    ///
    /// # Examples
    ///
    /// In this example we:
    ///
    /// 1. open a socket
    /// 2. connect it to the kernel with [`Socket::connect`]
    /// 3. send a request to the kernel with [`Socket::send`]
    /// 4. read the response (which can span over several messages)
    /// [`Socket::recv`]
    ///
    /// ```rust
    /// use netlink_sys::{protocols::NETLINK_ROUTE, Socket, SocketAddr};
    /// use std::process;
    ///
    /// let mut socket = Socket::new(NETLINK_ROUTE).unwrap();
    /// let _ = socket.bind_auto().unwrap();
    /// let kernel_addr = SocketAddr::new(0, 0);
    /// socket.connect(&kernel_addr).unwrap();
    /// // This is a valid message for listing the network links on the system
    /// let msg = vec![
    ///     0x14, 0x00, 0x00, 0x00, 0x12, 0x00, 0x01, 0x03, 0xfd, 0xfe, 0x38, 0x5c, 0x00, 0x00, 0x00,
    ///     0x00, 0x00, 0x00, 0x00, 0x00,
    /// ];
    /// let n_sent = socket.send(&msg[..], 0).unwrap();
    /// assert_eq!(n_sent, msg.len());
    /// // buffer for receiving the response
    /// let mut buf = vec![0; 4096];
    /// loop {
    ///     let mut n_received = socket.recv(&mut &mut buf[..], 0).unwrap();
    ///     println!("received {:?}", &buf[..n_received]);
    ///     if buf[4] == 2 && buf[5] == 0 {
    ///         println!("the kernel responded with an error");
    ///         return;
    ///     }
    ///     if buf[4] == 3 && buf[5] == 0 {
    ///         println!("end of dump");
    ///         return;
    ///     }
    /// }
    /// ```
    pub fn connect(&self, remote_addr: &SocketAddr) -> Result<()> {
        // FIXME:
        //
        // Event though for SOCK_DGRAM sockets there's no IO, if our socket is
        // non-blocking, connect() might return EINPROGRESS. In theory,
        // the right way to treat EINPROGRESS would be to ignore the
        // error, and let the user poll the socket to check when it becomes
        // writable, indicating that the connection succeeded. The code already
        // exists in mio for TcpStream:
        //
        // > pub fn connect(stream: net::TcpStream, addr: &SocketAddr) ->
        // > io::Result<TcpStream> {
        // > set_non_block(stream.as_raw_fd())?;
        // > match stream.connect(addr) {
        // > Ok(..) => {}
        // > Err(ref e) if e.raw_os_error() == Some(libc::EINPROGRESS) => {}
        // > Err(e) => return Err(e),
        // > }
        // > Ok(TcpStream {  inner: stream })
        // > }
        //
        // In practice, since the connection does not require any IO for
        // SOCK_DGRAM sockets, it almost never returns EINPROGRESS and
        // so for now, we just return whatever libc::connect returns. If
        // it returns EINPROGRESS, the caller will have to handle the error
        // themself
        //
        // Refs:
        //
        // - https://stackoverflow.com/a/14046386/1836144
        // - https://lists.isc.org/pipermail/bind-users/2009-August/077527.html
        let (addr, addr_len) = remote_addr.as_raw();
        let res = unsafe { libc::connect(self.0, addr, addr_len) };
        if res < 0 {
            return Err(Error::last_os_error());
        }
        Ok(())
    }

    // Most of the comments in this method come from a discussion on rust users
    // forum. [thread]: https://users.rust-lang.org/t/help-understanding-libc-call/17308/9
    //
    /// Read a datagram from the socket and return the number of bytes that have
    /// been read and the address of the sender. The data being read is
    /// copied into `buf`. If `buf` is too small, the datagram is truncated. The
    /// supported flags are the `MSG_*` described in `man 2 recvmsg`
    ///
    /// # Warning
    ///
    /// In datagram oriented protocols, `recv` and `recvfrom` receive normally
    /// only ONE datagram, but this seems not to be always true for netlink
    /// sockets: with some protocols like `NETLINK_AUDIT`, multiple netlink
    /// packets can be read with a single call.
    pub fn recv_from<B>(
        &self,
        buf: &mut B,
        flags: libc::c_int,
    ) -> Result<(usize, SocketAddr)>
    where
        B: bytes::BufMut,
    {
        // Create an empty storage for the address. Note that Rust standard
        // library create a sockaddr_storage so that it works for any
        // address family, but here, we already know that we'll have a
        // Netlink address, so we can create the appropriate storage.
        let mut addr = unsafe { mem::zeroed::<libc::sockaddr_nl>() };

        // recvfrom takes a *sockaddr as parameter so that it can accept any
        // kind of address storage, so we need to create such a pointer
        // for the sockaddr_nl we just initialized.
        //
        //                     Create a raw pointer to        Cast our raw
        // pointer to a                     our storage. We cannot
        // generic pointer to *sockaddr                     pass it to
        // recvfrom yet.       that recvfrom can use
        // ^                              ^
        // |                              |
        // +--------------+---------------+    +---------+--------+
        //                 /                                \  /
        // \
        let addr_ptr =
            &mut addr as *mut libc::sockaddr_nl as *mut libc::sockaddr;

        // Why do we need to pass the address length? We're passing a generic
        // *sockaddr to recvfrom. Somehow recvfrom needs to make sure
        // that the address of the received packet would fit into the
        // actual type that is behind *sockaddr: it could be a sockaddr_nl but
        // also a sockaddr_in, a sockaddr_in6, or even the generic
        // sockaddr_storage that can store any address.
        let mut addrlen = mem::size_of_val(&addr);
        // recvfrom does not take the address length by value (see [thread]), so
        // we need to create a pointer to it.
        let addrlen_ptr = &mut addrlen as *mut usize as *mut libc::socklen_t;

        let chunk = buf.chunk_mut();
        //                        Cast the *mut u8 into *mut void.
        //                 This is equivalent to casting a *char into *void
        //                                   See [thread]
        //                                         ^
        //             Create a *mut u8            |
        //                    ^                    |
        //                    |                    |
        //             +------+-------+   +--------+-------+
        //            /                \ /                  \
        let buf_ptr = chunk.as_mut_ptr() as *mut libc::c_void;
        let buf_len = chunk.len() as libc::size_t;

        let res = unsafe {
            libc::recvfrom(
                self.0,
                buf_ptr,
                buf_len,
                flags,
                addr_ptr,
                addrlen_ptr,
            )
        };
        if res < 0 {
            return Err(Error::last_os_error());
        } else {
            // with `MSG_TRUNC` `res` might exceed `buf_len`
            let written = std::cmp::min(buf_len, res as usize);
            unsafe {
                buf.advance_mut(written);
            }
        }
        Ok((res as usize, SocketAddr(addr)))
    }

    /// For a connected socket, `recv` reads a datagram from the socket. The
    /// sender is the remote peer the socket is connected to (see
    /// [`Socket::connect`]). See also [`Socket::recv_from`]
    pub fn recv<B>(&self, buf: &mut B, flags: libc::c_int) -> Result<usize>
    where
        B: bytes::BufMut,
    {
        let chunk = buf.chunk_mut();
        let buf_ptr = chunk.as_mut_ptr() as *mut libc::c_void;
        let buf_len = chunk.len() as libc::size_t;

        let res = unsafe { libc::recv(self.0, buf_ptr, buf_len, flags) };
        if res < 0 {
            return Err(Error::last_os_error());
        } else {
            // with `MSG_TRUNC` `res` might exceed `buf_len`
            let written = std::cmp::min(buf_len, res as usize);
            unsafe {
                buf.advance_mut(written);
            }
        }
        Ok(res as usize)
    }

    /// Receive a full message. Unlike [`Socket::recv_from`], which truncates
    /// messages that exceed the length of the buffer passed as argument,
    /// this method always reads a whole message, no matter its size.
    pub fn recv_from_full(&self) -> Result<(Vec<u8>, SocketAddr)> {
        // Peek
        let mut buf: Vec<u8> = Vec::new();
        let (peek_len, _) =
            self.recv_from(&mut buf, libc::MSG_PEEK | libc::MSG_TRUNC)?;

        // Receive
        buf.clear();
        buf.reserve(peek_len);
        let (rlen, addr) = self.recv_from(&mut buf, 0)?;
        assert_eq!(rlen, peek_len);
        Ok((buf, addr))
    }

    /// Send the given buffer `buf` to the remote peer with address `addr`. The
    /// supported flags are the `MSG_*` values documented in `man 2 send`.
    pub fn send_to(
        &self,
        buf: &[u8],
        addr: &SocketAddr,
        flags: libc::c_int,
    ) -> Result<usize> {
        let (addr_ptr, addr_len) = addr.as_raw();
        let buf_ptr = buf.as_ptr() as *const libc::c_void;
        let buf_len = buf.len() as libc::size_t;

        let res = unsafe {
            libc::sendto(self.0, buf_ptr, buf_len, flags, addr_ptr, addr_len)
        };
        if res < 0 {
            return Err(Error::last_os_error());
        }
        Ok(res as usize)
    }

    /// For a connected socket, `send` sends the given buffer `buf` to the
    /// remote peer the socket is connected to. See also [`Socket::connect`]
    /// and [`Socket::send_to`].
    pub fn send(&self, buf: &[u8], flags: libc::c_int) -> Result<usize> {
        let buf_ptr = buf.as_ptr() as *const libc::c_void;
        let buf_len = buf.len() as libc::size_t;

        let res = unsafe { libc::send(self.0, buf_ptr, buf_len, flags) };
        if res < 0 {
            return Err(Error::last_os_error());
        }
        Ok(res as usize)
    }

    pub fn set_pktinfo(&mut self, value: bool) -> Result<()> {
        let value: libc::c_int = value.into();
        setsockopt(self.0, libc::SOL_NETLINK, libc::NETLINK_PKTINFO, value)
    }

    pub fn get_pktinfo(&self) -> Result<bool> {
        let res = getsockopt::<libc::c_int>(
            self.0,
            libc::SOL_NETLINK,
            libc::NETLINK_PKTINFO,
        )?;
        Ok(res == 1)
    }

    pub fn add_membership(&mut self, group: u32) -> Result<()> {
        setsockopt(
            self.0,
            libc::SOL_NETLINK,
            libc::NETLINK_ADD_MEMBERSHIP,
            group,
        )
    }

    pub fn drop_membership(&mut self, group: u32) -> Result<()> {
        setsockopt(
            self.0,
            libc::SOL_NETLINK,
            libc::NETLINK_DROP_MEMBERSHIP,
            group,
        )
    }

    // pub fn list_membership(&self) -> Vec<u32> {
    //     unimplemented!();
    //     // getsockopt won't be enough here, because we may need to perform 2
    // calls, and because the     // length of the list returned by
    // libc::getsockopt is returned by mutating the length     // argument,
    // which our implementation of getsockopt forbids. }

    /// `NETLINK_BROADCAST_ERROR` (since Linux 2.6.30). When not set,
    /// `netlink_broadcast()` only reports `ESRCH` errors and silently
    /// ignore `NOBUFS` errors.
    pub fn set_broadcast_error(&mut self, value: bool) -> Result<()> {
        let value: libc::c_int = value.into();
        setsockopt(
            self.0,
            libc::SOL_NETLINK,
            libc::NETLINK_BROADCAST_ERROR,
            value,
        )
    }

    pub fn get_broadcast_error(&self) -> Result<bool> {
        let res = getsockopt::<libc::c_int>(
            self.0,
            libc::SOL_NETLINK,
            libc::NETLINK_BROADCAST_ERROR,
        )?;
        Ok(res == 1)
    }

    /// `NETLINK_NO_ENOBUFS` (since Linux 2.6.30). This flag can be used by
    /// unicast and broadcast listeners to avoid receiving `ENOBUFS` errors.
    pub fn set_no_enobufs(&mut self, value: bool) -> Result<()> {
        let value: libc::c_int = value.into();
        setsockopt(self.0, libc::SOL_NETLINK, libc::NETLINK_NO_ENOBUFS, value)
    }

    pub fn get_no_enobufs(&self) -> Result<bool> {
        let res = getsockopt::<libc::c_int>(
            self.0,
            libc::SOL_NETLINK,
            libc::NETLINK_NO_ENOBUFS,
        )?;
        Ok(res == 1)
    }

    /// `NETLINK_LISTEN_ALL_NSID` (since Linux 4.2). When set, this socket will
    /// receive netlink notifications from  all  network  namespaces that
    /// have an nsid assigned into the network namespace where the socket
    /// has been opened. The nsid is sent to user space via an ancillary
    /// data.
    pub fn set_listen_all_namespaces(&mut self, value: bool) -> Result<()> {
        let value: libc::c_int = value.into();
        setsockopt(
            self.0,
            libc::SOL_NETLINK,
            libc::NETLINK_LISTEN_ALL_NSID,
            value,
        )
    }

    pub fn get_listen_all_namespaces(&self) -> Result<bool> {
        let res = getsockopt::<libc::c_int>(
            self.0,
            libc::SOL_NETLINK,
            libc::NETLINK_LISTEN_ALL_NSID,
        )?;
        Ok(res == 1)
    }

    /// `NETLINK_CAP_ACK` (since Linux 4.2). The kernel may fail to allocate the
    /// necessary room for the acknowledgment message back to user space.
    /// This option trims off the payload of the original netlink message.
    /// The netlink message header is still included, so the user can
    /// guess from the sequence  number which message triggered the
    /// acknowledgment.
    pub fn set_cap_ack(&mut self, value: bool) -> Result<()> {
        let value: libc::c_int = value.into();
        setsockopt(self.0, libc::SOL_NETLINK, libc::NETLINK_CAP_ACK, value)
    }

    pub fn get_cap_ack(&self) -> Result<bool> {
        let res = getsockopt::<libc::c_int>(
            self.0,
            libc::SOL_NETLINK,
            libc::NETLINK_CAP_ACK,
        )?;
        Ok(res == 1)
    }

    /// `NETLINK_EXT_ACK`
    /// Extended ACK controls reporting of additional error/warning TLVs in
    /// NLMSG_ERROR and NLMSG_DONE messages.
    pub fn set_ext_ack(&mut self, value: bool) -> Result<()> {
        let value: libc::c_int = value.into();
        setsockopt(self.0, libc::SOL_NETLINK, libc::NETLINK_EXT_ACK, value)
    }

    pub fn get_ext_ack(&self) -> Result<bool> {
        let res = getsockopt::<libc::c_int>(
            self.0,
            libc::SOL_NETLINK,
            libc::NETLINK_EXT_ACK,
        )?;
        Ok(res == 1)
    }

    /// Sets socket receive buffer in bytes.
    /// The kernel doubles this value (to allow space for bookkeeping overhead),
    /// and this doubled value is returned by [get_rx_buf_sz].(see socket(7)
    /// The default value is set by the proc/sys/net/core/rmem_default file, and
    /// the maximum allowed value is set by the /proc/sys/net/core/rmem_max
    /// file. The minimum (doubled) value for this option is 256.
    pub fn set_rx_buf_sz<T>(&self, size: T) -> Result<()> {
        setsockopt(self.0, libc::SOL_SOCKET, libc::SO_RCVBUF, size)
    }

    /// Gets socket receive buffer in bytes
    pub fn get_rx_buf_sz(&self) -> Result<usize> {
        let res = getsockopt::<libc::c_int>(
            self.0,
            libc::SOL_SOCKET,
            libc::SO_RCVBUF,
        )?;
        Ok(res as usize)
    }

    /// Set strict input checking(`NETLINK_GET_STRICT_CHK`) in netlink route
    /// protocol. By default, `NETLINK_GET_STRICT_CHK` is not enabled.
    pub fn set_netlink_get_strict_chk(&self, value: bool) -> Result<()> {
        let value: u32 = value.into();
        setsockopt(
            self.0,
            libc::SOL_NETLINK,
            libc::NETLINK_GET_STRICT_CHK,
            value,
        )
    }
}

/// Wrapper around `getsockopt`:
///
/// ```no_rust
/// int getsockopt(int socket, int level, int option_name, void *restrict option_value, socklen_t *restrict option_len);
/// ```
pub(crate) fn getsockopt<T: Copy>(
    fd: RawFd,
    level: libc::c_int,
    option: libc::c_int,
) -> Result<T> {
    // Create storage for the options we're fetching
    let mut slot: T = unsafe { mem::zeroed() };

    // Create a mutable raw pointer to the storage so that getsockopt can fill
    // the value
    let slot_ptr = &mut slot as *mut T as *mut libc::c_void;

    // Let getsockopt know how big our storage is
    let mut slot_len = mem::size_of::<T>() as libc::socklen_t;

    // getsockopt takes a mutable pointer to the length, because for some
    // options like NETLINK_LIST_MEMBERSHIP where the option value is a list
    // with arbitrary length, getsockopt uses this parameter to signal how
    // big the storage needs to be.
    let slot_len_ptr = &mut slot_len as *mut libc::socklen_t;

    let res =
        unsafe { libc::getsockopt(fd, level, option, slot_ptr, slot_len_ptr) };
    if res < 0 {
        return Err(Error::last_os_error());
    }

    // Ignore the options that require the legnth to be set by getsockopt.
    // We'll deal with them individually.
    assert_eq!(slot_len as usize, mem::size_of::<T>());

    Ok(slot)
}

// adapted from rust standard library
fn setsockopt<T>(
    fd: RawFd,
    level: libc::c_int,
    option: libc::c_int,
    payload: T,
) -> Result<()> {
    let payload = &payload as *const T as *const libc::c_void;
    let payload_len = mem::size_of::<T>() as libc::socklen_t;

    let res =
        unsafe { libc::setsockopt(fd, level, option, payload, payload_len) };
    if res < 0 {
        return Err(Error::last_os_error());
    }
    Ok(())
}

#[cfg(test)]
mod test {
    use super::*;
    use crate::protocols::NETLINK_ROUTE;

    #[test]
    fn new() {
        Socket::new(NETLINK_ROUTE).unwrap();
    }

    #[test]
    fn connect() {
        let sock = Socket::new(NETLINK_ROUTE).unwrap();
        sock.connect(&SocketAddr::new(0, 0)).unwrap();
    }

    #[test]
    fn bind() {
        let mut sock = Socket::new(NETLINK_ROUTE).unwrap();
        sock.bind(&SocketAddr::new(4321, 0)).unwrap();
    }

    #[test]
    fn bind_auto() {
        let mut sock = Socket::new(NETLINK_ROUTE).unwrap();
        let addr = sock.bind_auto().unwrap();
        // make sure that the address we got from the kernel is there
        assert!(addr.port_number() != 0);
    }

    #[test]
    fn set_non_blocking() {
        let sock = Socket::new(NETLINK_ROUTE).unwrap();
        sock.set_non_blocking(true).unwrap();
        sock.set_non_blocking(false).unwrap();
    }

    #[test]
    fn options() {
        let mut sock = Socket::new(NETLINK_ROUTE).unwrap();

        sock.set_cap_ack(true).unwrap();
        assert!(sock.get_cap_ack().unwrap());
        sock.set_cap_ack(false).unwrap();
        assert!(!sock.get_cap_ack().unwrap());

        sock.set_no_enobufs(true).unwrap();
        assert!(sock.get_no_enobufs().unwrap());
        sock.set_no_enobufs(false).unwrap();
        assert!(!sock.get_no_enobufs().unwrap());

        sock.set_broadcast_error(true).unwrap();
        assert!(sock.get_broadcast_error().unwrap());
        sock.set_broadcast_error(false).unwrap();
        assert!(!sock.get_broadcast_error().unwrap());

        // FIXME: these require root permissions
        // sock.set_listen_all_namespaces(true).unwrap();
        // assert!(sock.get_listen_all_namespaces().unwrap());
        // sock.set_listen_all_namespaces(false).unwrap();
        // assert!(!sock.get_listen_all_namespaces().unwrap());
    }
}