Add basic structure for netlink route socket

kernel/src/net/mod.rs

@@ -5,6 +5,7 @@ pub mod socket;
 
 pub fn init() {
     iface::init();
+    socket::netlink::init();
     socket::vsock::init();
 }
 

kernel/src/net/socket/mod.rs

@@ -15,6 +15,7 @@ use crate::{
 };
 
 pub mod ip;
+pub mod netlink;
 pub mod options;
 pub mod unix;
 mod util;

kernel/src/net/socket/netlink/addr/mod.rs

@@ -0,0 +1,71 @@
+// SPDX-License-Identifier: MPL-2.0
+
+mod multicast;
+
+pub use multicast::{GroupIdSet, MAX_GROUPS};
+
+use crate::{net::socket::SocketAddr, prelude::*};
+
+/// The socket address of a netlink socket.
+///
+/// The address contains the port number for unicast
+/// and the group IDs for multicast.
+#[derive(Debug, Clone, Copy, PartialEq, Eq)]
+pub struct NetlinkSocketAddr {
+    port: PortNum,
+    groups: GroupIdSet,
+}
+
+impl NetlinkSocketAddr {
+    /// Creates a new netlink address.
+    pub const fn new(port: PortNum, groups: GroupIdSet) -> Self {
+        Self { port, groups }
+    }
+
+    /// Creates a new unspecified address.
+    ///
+    /// Both the port ID and group numbers are left unspecified.
+    ///
+    /// Note that an unspecified address can also represent the kernel socket address.
+    pub const fn new_unspecified() -> Self {
+        Self {
+            port: UNSPECIFIED_PORT,
+            groups: GroupIdSet::new_empty(),
+        }
+    }
+
+    /// Returns the port number.
+    pub const fn port(&self) -> PortNum {
+        self.port
+    }
+
+    /// Returns the group ID set.
+    pub const fn groups(&self) -> GroupIdSet {
+        self.groups
+    }
+}
+
+impl TryFrom<SocketAddr> for NetlinkSocketAddr {
+    type Error = Error;
+
+    fn try_from(value: SocketAddr) -> Result<Self> {
+        match value {
+            SocketAddr::Netlink(addr) => Ok(addr),
+            _ => return_errno_with_message!(
+                Errno::EAFNOSUPPORT,
+                "the address is in an unsupported address family"
+            ),
+        }
+    }
+}
+
+impl From<NetlinkSocketAddr> for SocketAddr {
+    fn from(value: NetlinkSocketAddr) -> Self {
+        SocketAddr::Netlink(value)
+    }
+}
+
+pub type NetlinkProtocolId = u32;
+pub type PortNum = u32;
+
+pub const UNSPECIFIED_PORT: PortNum = 0;

kernel/src/net/socket/netlink/addr/multicast.rs

@@ -0,0 +1,91 @@
+// SPDX-License-Identifier: MPL-2.0
+
+use crate::prelude::*;
+
+/// A set of group IDs.
+#[derive(Debug, Clone, Copy, PartialEq, Eq)]
+pub struct GroupIdSet(u32);
+
+impl GroupIdSet {
+    /// Creates a new empty `GroupIdSet`.
+    pub const fn new_empty() -> Self {
+        Self(0)
+    }
+
+    /// Creates a new `GroupIdSet` with multiple groups.
+    ///
+    /// Each 1 bit in `groups` represent a group.
+    pub const fn new(groups: u32) -> Self {
+        Self(groups)
+    }
+
+    /// Creates an iterator over all group IDs.
+    pub const fn ids_iter(&self) -> GroupIdIter {
+        GroupIdIter::new(self)
+    }
+
+    /// Adds a new group.
+    ///
+    /// If the group already exists, this method will return an error.
+    pub fn add_group(&mut self, group_id: GroupId) -> Result<()> {
+        if group_id >= 32 {
+            return_errno_with_message!(Errno::EINVAL, "the group ID is invalid");
+        }
+
+        let mask = 1u32 << group_id;
+        if self.0 & mask != 0 {
+            return_errno_with_message!(Errno::EINVAL, "the group ID already exists");
+        }
+        self.0 |= mask;
+
+        Ok(())
+    }
+
+    /// Sets new groups.
+    pub fn set_groups(&mut self, new_groups: u32) {
+        self.0 = new_groups;
+    }
+
+    /// Clears all groups.
+    pub fn clear(&mut self) {
+        self.0 = 0;
+    }
+
+    /// Checks if the set of group IDs is empty.
+    pub fn is_empty(&self) -> bool {
+        self.0 == 0
+    }
+
+    /// Returns the group IDs as a u32.
+    pub fn as_u32(&self) -> u32 {
+        self.0
+    }
+}
+
+/// Iterator over a set of group IDs.
+pub struct GroupIdIter {
+    groups: u32,
+}
+
+impl GroupIdIter {
+    const fn new(groups: &GroupIdSet) -> Self {
+        Self { groups: groups.0 }
+    }
+}
+
+impl Iterator for GroupIdIter {
+    type Item = GroupId;
+
+    fn next(&mut self) -> Option<Self::Item> {
+        if self.groups > 0 {
+            let group_id = self.groups.trailing_zeros();
+            self.groups &= self.groups - 1;
+            return Some(group_id);
+        }
+
+        None
+    }
+}
+
+pub const MAX_GROUPS: u32 = 32;
+pub type GroupId = u32;

kernel/src/net/socket/netlink/mod.rs

@@ -0,0 +1,51 @@
+// SPDX-License-Identifier: MPL-2.0
+
+//! This module defines netlink sockets.
+//!
+//! Netlink provides a standardized, socket-based interface,
+//! typically used for communication between user space and kernel space.
+//! It can also be used for interaction between two user processes.
+//!
+//! Each netlink socket corresponds to
+//! a netlink protocol identified by a protocol ID (u32).
+//! Protocols are generally defined to serve specific functions.
+//! For instance, the NETLINK_ROUTE protocol is employed
+//! to retrieve or modify network device settings.
+//! Only sockets associated with the same protocol can communicate with each other.
+//! Some protocols are pre-defined by the kernel and serve fixed purposes,
+//! but users can also establish custom protocols by specifying new protocol IDs.
+//!
+//! Before initiating communication,
+//! a netlink socket must be bound to an address,
+//! which consists of a port number and a multicast group number.
+//!
+//! The port number is used for unicast communication,
+//! whereas the multicast group number is meant for multicast communication.
+//!
+//! In terms of unicast communication within each protocol,
+//! a port number can only be bound to one socket.
+//! However, the same port number can be utilized across different protocols.
+//! Typically, the port number corresponds to the process ID of the running process.
+//!
+//! Multicast communication allows a message
+//! to be sent to one or multiple multicast groups simultaneously.
+//! Each protocol can support up to 32 multicast groups,
+//! and a socket can belong to zero or multiple multicast groups.
+//!
+//! Netlink communication is akin to UDP in that
+//! it does not require a connection to be established before sending messages.
+//! The destination address must be specified when dispatching a message.
+//!
+
+mod addr;
+mod message;
+mod route;
+mod table;
+
+pub use addr::{GroupIdSet, NetlinkSocketAddr};
+pub use route::NetlinkRouteSocket;
+pub use table::{is_valid_protocol, StandardNetlinkProtocol};
+
+pub(in crate::net) fn init() {
+    table::init();
+}

kernel/src/net/socket/netlink/route/bound.rs

@@ -0,0 +1,139 @@
+// SPDX-License-Identifier: MPL-2.0
+
+use core::ops::Sub;
+
+use super::message::RtnlMessage;
+use crate::{
+    events::IoEvents,
+    net::socket::{
+        netlink::{
+            message::ProtocolSegment, route::kernel::get_netlink_route_kernel, table::BoundHandle,
+            NetlinkSocketAddr,
+        },
+        SendRecvFlags,
+    },
+    prelude::*,
+    util::{MultiRead, MultiWrite},
+};
+
+pub(super) struct BoundNetlinkRoute {
+    handle: BoundHandle,
+    receive_queue: Mutex<VecDeque<RtnlMessage>>,
+}
+
+impl BoundNetlinkRoute {
+    pub(super) const fn new(handle: BoundHandle) -> Self {
+        Self {
+            handle,
+            receive_queue: Mutex::new(VecDeque::new()),
+        }
+    }
+
+    pub(super) const fn addr(&self) -> NetlinkSocketAddr {
+        self.handle.addr()
+    }
+
+    pub(super) fn try_send(
+        &self,
+        reader: &mut dyn MultiRead,
+        remote: Option<&NetlinkSocketAddr>,
+        flags: SendRecvFlags,
+    ) -> Result<usize> {
+        // TODO: Deal with flags
+        if !flags.is_all_supported() {
+            warn!("unsupported flags: {:?}", flags);
+        }
+
+        if let Some(remote) = remote {
+            // TODO: Further check whether other socket address can be supported.
+            if *remote != NetlinkSocketAddr::new_unspecified() {
+                return_errno_with_message!(
+                    Errno::ECONNREFUSED,
+                    "sending netlink route messages to user space is not supported"
+                );
+            }
+        } else {
+            // TODO: We should use the connected remote address, if any.
+        }
+
+        let mut nlmsg = {
+            let sum_lens = reader.sum_lens();
+
+            match RtnlMessage::read_from(reader) {
+                Ok(nlmsg) => nlmsg,
+                Err(e) if e.error() == Errno::EFAULT => {
+                    // EFAULT indicates an error occurred while copying data from user space,
+                    // and this error should be returned back to user space.
+                    return Err(e);
+                }
+                Err(e) => {
+                    // Errors other than EFAULT indicate a failure in parsing the netlink message.
+                    // These errors should be silently ignored.
+                    warn!("failed to send netlink message: {:?}", e);
+                    return Ok(sum_lens);
+                }
+            }
+        };
+
+        let local_port = self.addr().port();
+        for segment in nlmsg.segments_mut() {
+            // The header's PID should be the sender's port ID.
+            // However, the sender can also leave it unspecified.
+            // In such cases, we will manually set the PID to the sender's port ID.
+            let header = segment.header_mut();
+            if header.pid == 0 {
+                header.pid = local_port;
+            }
+        }
+
+        get_netlink_route_kernel().request(&nlmsg, |response| {
+            self.receive_queue.lock().push_back(response);
+        });
+
+        Ok(nlmsg.total_len())
+    }
+
+    pub(super) fn try_receive(
+        &self,
+        writer: &mut dyn MultiWrite,
+        flags: SendRecvFlags,
+    ) -> Result<(usize, NetlinkSocketAddr)> {
+        // TODO: Deal with other flags. Only MSG_PEEK is handled here.
+        if !flags.sub(SendRecvFlags::MSG_PEEK).is_all_supported() {
+            warn!("unsupported flags: {:?}", flags);
+        }
+
+        let mut receive_queue = self.receive_queue.lock();
+
+        let Some(response) = receive_queue.front() else {
+            return_errno_with_message!(Errno::EAGAIN, "nothing to receive");
+        };
+
+        let len = {
+            let max_len = writer.sum_lens();
+            response.total_len().min(max_len)
+        };
+
+        response.write_to(writer)?;
+
+        if !flags.contains(SendRecvFlags::MSG_PEEK) {
+            receive_queue.pop_front().unwrap();
+        }
+
+        // TODO: The message can only come from kernel socket currently.
+        let remote = NetlinkSocketAddr::new_unspecified();
+
+        Ok((len, remote))
+    }
+
+    pub(super) fn check_io_events(&self) -> IoEvents {
+        let mut events = IoEvents::OUT;
+
+        let receive_queue = self.receive_queue.lock();
+        if !receive_queue.is_empty() {
+            events |= IoEvents::IN;
+        }
+
+        events
+    }
+}

kernel/src/net/socket/netlink/route/mod.rs

@@ -0,0 +1,207 @@
+// SPDX-License-Identifier: MPL-2.0
+
+//! Netlink Route Socket.
+
+use core::sync::atomic::{AtomicBool, Ordering};
+
+use bound::BoundNetlinkRoute;
+use takeable::Takeable;
+use unbound::UnboundNetlinkRoute;
+
+use super::NetlinkSocketAddr;
+use crate::{
+    events::IoEvents,
+    net::socket::{
+        options::SocketOption, private::SocketPrivate, MessageHeader, SendRecvFlags, Socket,
+        SocketAddr,
+    },
+    prelude::*,
+    process::signal::{PollHandle, Pollable, Pollee},
+    util::{MultiRead, MultiWrite},
+};
+
+mod bound;
+mod kernel;
+mod message;
+mod unbound;
+
+pub struct NetlinkRouteSocket {
+    is_nonblocking: AtomicBool,
+    pollee: Pollee,
+    inner: RwMutex<Takeable<Inner>>,
+}
+
+enum Inner {
+    Unbound(UnboundNetlinkRoute),
+    Bound(BoundNetlinkRoute),
+}
+
+impl NetlinkRouteSocket {
+    pub fn new(is_nonblocking: bool) -> Self {
+        Self {
+            is_nonblocking: AtomicBool::new(is_nonblocking),
+            pollee: Pollee::new(),
+            inner: RwMutex::new(Takeable::new(Inner::Unbound(UnboundNetlinkRoute::new()))),
+        }
+    }
+
+    fn try_receive(
+        &self,
+        writer: &mut dyn MultiWrite,
+        flags: SendRecvFlags,
+    ) -> Result<(usize, NetlinkSocketAddr)> {
+        let inner = self.inner.read();
+
+        let bound = match inner.as_ref() {
+            Inner::Unbound(_) => {
+                return_errno_with_message!(Errno::EAGAIN, "the socket is not bound")
+            }
+            Inner::Bound(bound_netlink_route) => bound_netlink_route,
+        };
+
+        let received = bound.try_receive(writer, flags)?;
+        self.pollee.invalidate();
+
+        Ok(received)
+    }
+
+    fn try_send(
+        &self,
+        reader: &mut dyn MultiRead,
+        remote: Option<&NetlinkSocketAddr>,
+        flags: SendRecvFlags,
+    ) -> Result<usize> {
+        let inner = self.inner.read();
+
+        let bound = match inner.as_ref() {
+            Inner::Unbound(_) => todo!(),
+            Inner::Bound(bound) => bound,
+        };
+
+        let sent_bytes = bound.try_send(reader, remote, flags)?;
+        self.pollee.notify(IoEvents::OUT | IoEvents::IN);
+
+        Ok(sent_bytes)
+    }
+
+    fn check_io_events(&self) -> IoEvents {
+        let inner = self.inner.read();
+        match inner.as_ref() {
+            Inner::Unbound(unbound) => unbound.check_io_events(),
+            Inner::Bound(bound) => bound.check_io_events(),
+        }
+    }
+}
+
+impl Socket for NetlinkRouteSocket {
+    fn bind(&self, socket_addr: SocketAddr) -> Result<()> {
+        let SocketAddr::Netlink(netlink_addr) = socket_addr else {
+            return_errno_with_message!(
+                Errno::EAFNOSUPPORT,
+                "the provided address is not netlink address"
+            );
+        };
+
+        let mut inner = self.inner.write();
+        inner.borrow_result(|owned_inner| match owned_inner.bind(&netlink_addr) {
+            Ok(bound_inner) => (bound_inner, Ok(())),
+            Err((err, err_inner)) => (err_inner, Err(err)),
+        })
+    }
+
+    fn addr(&self) -> Result<SocketAddr> {
+        let netlink_addr = match self.inner.read().as_ref() {
+            Inner::Unbound(_) => NetlinkSocketAddr::new_unspecified(),
+            Inner::Bound(bound) => bound.addr(),
+        };
+
+        Ok(SocketAddr::Netlink(netlink_addr))
+    }
+
+    fn sendmsg(
+        &self,
+        reader: &mut dyn MultiRead,
+        message_header: MessageHeader,
+        flags: SendRecvFlags,
+    ) -> Result<usize> {
+        let MessageHeader {
+            addr,
+            control_message,
+        } = message_header;
+
+        let remote = match addr {
+            None => None,
+            Some(addr) => Some(addr.try_into()?),
+        };
+
+        if control_message.is_some() {
+            // TODO: Support sending control message
+            warn!("sending control message is not supported");
+        }
+
+        // TODO: Make sure our blocking behavior matches that of Linux
+        self.try_send(reader, remote.as_ref(), flags)
+    }
+
+    fn recvmsg(
+        &self,
+        writers: &mut dyn MultiWrite,
+        flags: SendRecvFlags,
+    ) -> Result<(usize, MessageHeader)> {
+        let (received_len, addr) =
+            self.block_on(IoEvents::IN, || self.try_receive(writers, flags))?;
+
+        // TODO: Receive control message
+
+        let message_header = {
+            let addr = SocketAddr::Netlink(addr);
+            MessageHeader::new(Some(addr), None)
+        };
+
+        Ok((received_len, message_header))
+    }
+
+    fn set_option(&self, _option: &dyn SocketOption) -> Result<()> {
+        // TODO: This dummy option is added to pass the libnl test
+        Ok(())
+    }
+}
+
+impl SocketPrivate for NetlinkRouteSocket {
+    fn is_nonblocking(&self) -> bool {
+        self.is_nonblocking.load(Ordering::Relaxed)
+    }
+
+    fn set_nonblocking(&self, nonblocking: bool) {
+        self.is_nonblocking.store(nonblocking, Ordering::Relaxed);
+    }
+}
+
+impl Pollable for NetlinkRouteSocket {
+    fn poll(&self, mask: IoEvents, poller: Option<&mut PollHandle>) -> IoEvents {
+        self.pollee
+            .poll_with(mask, poller, || self.check_io_events())
+    }
+}
+
+impl Inner {
+    fn bind(self, addr: &NetlinkSocketAddr) -> core::result::Result<Self, (Error, Self)> {
+        let unbound = match self {
+            Inner::Unbound(unbound) => unbound,
+            Inner::Bound(bound) => {
+                // FIXME: We need to further check the Linux behavior
+                // whether we should return error if the socket is bound.
+                // The socket may call `bind` syscall to join new multicast groups.
+                return Err((
+                    Error::with_message(Errno::EINVAL, "the socket is already bound"),
+                    Self::Bound(bound),
+                ));
+            }
+        };
+
+        match unbound.bind(addr) {
+            Ok(bound) => Ok(Self::Bound(bound)),
+            Err((err, unbound)) => Err((err, Self::Unbound(unbound))),
+        }
+    }
+}

kernel/src/net/socket/netlink/route/unbound.rs

@@ -0,0 +1,35 @@
+// SPDX-License-Identifier: MPL-2.0
+
+use super::bound::BoundNetlinkRoute;
+use crate::{
+    events::IoEvents,
+    net::socket::netlink::{
+        table::NETLINK_SOCKET_TABLE, NetlinkSocketAddr, StandardNetlinkProtocol,
+    },
+    prelude::*,
+};
+
+pub(super) struct UnboundNetlinkRoute {
+    _private: (),
+}
+
+impl UnboundNetlinkRoute {
+    pub(super) const fn new() -> Self {
+        Self { _private: () }
+    }
+
+    pub(super) fn bind(
+        self,
+        addr: &NetlinkSocketAddr,
+    ) -> core::result::Result<BoundNetlinkRoute, (Error, Self)> {
+        let bound_handle = NETLINK_SOCKET_TABLE
+            .bind(StandardNetlinkProtocol::ROUTE as _, addr)
+            .map_err(|err| (err, self))?;
+
+        Ok(BoundNetlinkRoute::new(bound_handle))
+    }
+
+    pub(super) fn check_io_events(&self) -> IoEvents {
+        IoEvents::OUT
+    }
+}

kernel/src/net/socket/netlink/table/mod.rs

@@ -0,0 +1,228 @@
+// SPDX-License-Identifier: MPL-2.0
+
+use multicast::MulticastGroup;
+
+use super::addr::{GroupIdSet, NetlinkProtocolId, NetlinkSocketAddr, PortNum, MAX_GROUPS};
+use crate::{net::socket::netlink::addr::UNSPECIFIED_PORT, prelude::*, util::random::getrandom};
+
+mod multicast;
+
+pub(super) static NETLINK_SOCKET_TABLE: NetlinkSocketTable = NetlinkSocketTable::new();
+
+/// All bound netlink sockets.
+pub(super) struct NetlinkSocketTable {
+    protocols: [Mutex<Option<ProtocolSocketTable>>; MAX_ALLOWED_PROTOCOL_ID as usize],
+}
+
+impl NetlinkSocketTable {
+    pub(super) const fn new() -> Self {
+        Self {
+            protocols: [const { Mutex::new(None) }; MAX_ALLOWED_PROTOCOL_ID as usize],
+        }
+    }
+
+    /// Adds a new netlink protocol.
+    fn add_new_protocol(&self, protocol_id: NetlinkProtocolId) {
+        if protocol_id >= MAX_ALLOWED_PROTOCOL_ID {
+            return;
+        }
+
+        let mut protocol = self.protocols[protocol_id as usize].lock();
+        if protocol.is_some() {
+            return;
+        }
+
+        let new_protocol = ProtocolSocketTable::new(protocol_id);
+        *protocol = Some(new_protocol);
+    }
+
+    pub(super) fn bind(
+        &self,
+        protocol: NetlinkProtocolId,
+        addr: &NetlinkSocketAddr,
+    ) -> Result<BoundHandle> {
+        if protocol >= MAX_ALLOWED_PROTOCOL_ID {
+            return_errno_with_message!(Errno::EINVAL, "the netlink protocol does not exist");
+        }
+
+        let mut protocol = self.protocols[protocol as usize].lock();
+
+        let Some(protocol_sockets) = protocol.as_mut() else {
+            return_errno_with_message!(Errno::EINVAL, "the netlink protocol does not exist")
+        };
+
+        protocol_sockets.bind(addr)
+    }
+}
+
+/// Bound socket table of a single netlink protocol.
+///
+/// Each table can have bound sockets for unicast
+/// and at most 32 groups for multicast.
+struct ProtocolSocketTable {
+    id: NetlinkProtocolId,
+    // TODO: This table should maintain the port number-to-socket relationship
+    // to support both unicast and multicast effectively.
+    unicast_sockets: BTreeSet<PortNum>,
+    multicast_groups: Box<[MulticastGroup]>,
+}
+
+impl ProtocolSocketTable {
+    /// Creates a new table.
+    fn new(id: NetlinkProtocolId) -> Self {
+        let multicast_groups = (0u32..MAX_GROUPS).map(|_| MulticastGroup::new()).collect();
+        Self {
+            id,
+            unicast_sockets: BTreeSet::new(),
+            multicast_groups,
+        }
+    }
+
+    /// Binds a socket to the table.
+    /// Returns the bound handle.
+    ///
+    /// The socket will be bound to a port specified by `addr.port()`.
+    /// If `addr.port()` is zero, the kernel will assign a port,
+    /// typically corresponding to the process ID of the current process.
+    /// If the assigned port is already in use,
+    /// this function will try to allocate a random unused port.
+    ///
+    /// Additionally, this socket can join one or more multicast groups,
+    /// as specified in `addr.groups()`.
+    fn bind(&mut self, addr: &NetlinkSocketAddr) -> Result<BoundHandle> {
+        let port = if addr.port() != UNSPECIFIED_PORT {
+            addr.port()
+        } else {
+            let mut random_port = current!().pid();
+            while random_port == UNSPECIFIED_PORT || self.unicast_sockets.contains(&random_port) {
+                getrandom(random_port.as_bytes_mut()).unwrap();
+            }
+            random_port
+        };
+
+        if self.unicast_sockets.contains(&port) {
+            return_errno_with_message!(Errno::EADDRINUSE, "the netlink port is already in use");
+        }
+
+        self.unicast_sockets.insert(port);
+
+        for group_id in addr.groups().ids_iter() {
+            let group = &mut self.multicast_groups[group_id as usize];
+            group.add_member(port);
+        }
+
+        Ok(BoundHandle::new(self.id, port, addr.groups()))
+    }
+}
+
+/// A bound netlink socket address.
+///
+/// When dropping a `BoundHandle`,
+/// the port will be automatically released.
+#[derive(Debug)]
+pub(super) struct BoundHandle {
+    protocol: NetlinkProtocolId,
+    port: PortNum,
+    groups: GroupIdSet,
+}
+
+impl BoundHandle {
+    fn new(protocol: NetlinkProtocolId, port: PortNum, groups: GroupIdSet) -> Self {
+        debug_assert_ne!(port, UNSPECIFIED_PORT);
+
+        Self {
+            protocol,
+            port,
+            groups,
+        }
+    }
+
+    pub(super) const fn addr(&self) -> NetlinkSocketAddr {
+        NetlinkSocketAddr::new(self.port, self.groups)
+    }
+}
+
+impl Drop for BoundHandle {
+    fn drop(&mut self) {
+        let mut protocol_sockets = NETLINK_SOCKET_TABLE.protocols[self.protocol as usize].lock();
+
+        let Some(protocol_sockets) = protocol_sockets.as_mut() else {
+            return;
+        };
+
+        protocol_sockets.unicast_sockets.remove(&self.port);
+
+        for group_id in self.groups.ids_iter() {
+            let group = &mut protocol_sockets.multicast_groups[group_id as usize];
+            group.remove_member(self.port);
+        }
+    }
+}
+
+pub(super) fn init() {
+    for protocol in 0..MAX_ALLOWED_PROTOCOL_ID {
+        if is_standard_protocol(protocol) {
+            NETLINK_SOCKET_TABLE.add_new_protocol(protocol);
+        }
+    }
+}
+
+/// Returns whether the `protocol` is valid.
+pub fn is_valid_protocol(protocol: NetlinkProtocolId) -> bool {
+    protocol < MAX_ALLOWED_PROTOCOL_ID
+}
+
+/// Returns whether the `protocol` is reserved for system use.
+fn is_standard_protocol(protocol: NetlinkProtocolId) -> bool {
+    StandardNetlinkProtocol::try_from(protocol).is_ok()
+}
+
+/// Netlink protocols that are assigned for specific usage.
+///
+/// Reference: <https://elixir.bootlin.com/linux/v6.0.9/source/include/uapi/linux/netlink.h#L9>.
+#[allow(non_camel_case_types)]
+#[repr(u32)]
+#[derive(Debug, Clone, Copy, TryFromInt)]
+pub enum StandardNetlinkProtocol {
+    /// Routing/device hook
+    ROUTE = 0,
+    /// Unused number
+    UNUSED = 1,
+    /// Reserved for user mode socket protocols
+    USERSOCK = 2,
+    /// Unused number, formerly ip_queue
+    FIREWALL = 3,
+    /// Socket monitoring
+    SOCK_DIAG = 4,
+    /// Netfilter/iptables ULOG
+    NFLOG = 5,
+    /// IPsec
+    XFRM = 6,
+    /// SELinux event notifications
+    SELINUX = 7,
+    /// Open-iSCSI
+    ISCSI = 8,
+    /// Auditing
+    AUDIT = 9,
+    FIB_LOOKUP = 10,
+    CONNECTOR = 11,
+    /// Netfilter subsystem
+    NETFILTER = 12,
+    IP6_FW = 13,
+    /// DECnet routing messages
+    DNRTMSG = 14,
+    /// Kernel messages to userspace
+    KOBJECT_UEVENT = 15,
+    GENERIC = 16,
+    /// Leave room for NETLINK_DM (DM Events)
+    /// SCSI Transports
+    SCSITRANSPORT = 18,
+    ECRYPTFS = 19,
+    RDMA = 20,
+    /// Crypto layer
+    CRYPTO = 21,
+    /// SMC monitoring
+    SMC = 22,
+}
+
+const MAX_ALLOWED_PROTOCOL_ID: NetlinkProtocolId = 32;

kernel/src/net/socket/netlink/table/multicast.rs

@@ -0,0 +1,38 @@
+// SPDX-License-Identifier: MPL-2.0
+
+use crate::{net::socket::netlink::addr::PortNum, prelude::*};
+
+/// A netlink multicast group.
+///
+/// A group can contain multiple sockets,
+/// each identified by its bound port number.
+pub struct MulticastGroup {
+    members: BTreeSet<PortNum>,
+}
+
+impl MulticastGroup {
+    /// Creates a new multicast group.
+    pub const fn new() -> Self {
+        Self {
+            members: BTreeSet::new(),
+        }
+    }
+
+    /// Returns whether the group contains a member.
+    #[expect(unused)]
+    pub fn contains_member(&self, port_num: PortNum) -> bool {
+        self.members.contains(&port_num)
+    }
+
+    /// Adds a new member to the multicast group.
+    pub fn add_member(&mut self, port_num: PortNum) {
+        debug_assert!(!self.members.contains(&port_num));
+        self.members.insert(port_num);
+    }
+
+    /// Removes a member from the multicast group.
+    pub fn remove_member(&mut self, port_num: PortNum) {
+        debug_assert!(self.members.contains(&port_num));
+        self.members.remove(&port_num);
+    }
+}

kernel/src/net/socket/util/socket_addr.rs

@@ -3,7 +3,7 @@
 use aster_bigtcp::wire::{Ipv4Address, PortNum};
 
 use crate::{
-    net::socket::{unix::UnixSocketAddr, vsock::addr::VsockSocketAddr},
+    net::socket::{netlink::NetlinkSocketAddr, unix::UnixSocketAddr, vsock::addr::VsockSocketAddr},
     prelude::*,
 };
 
@@ -11,5 +11,6 @@ use crate::{
 pub enum SocketAddr {
     Unix(UnixSocketAddr),
     IPv4(Ipv4Address, PortNum),
+    Netlink(NetlinkSocketAddr),
     Vsock(VsockSocketAddr),
 }

kernel/src/net/socket/vsock/mod.rs

@@ -15,7 +15,7 @@ pub use stream::VsockStreamSocket;
 // init static driver
 pub static VSOCK_GLOBAL: Once<Arc<VsockSpace>> = Once::new();
 
-pub fn init() {
+pub(in crate::net) fn init() {
     if let Some(driver) = get_device(DEVICE_NAME) {
         VSOCK_GLOBAL.call_once(|| Arc::new(VsockSpace::new(driver)));
         register_recv_callback(DEVICE_NAME, || {

kernel/src/syscall/socket.rs

@@ -5,6 +5,7 @@ use crate::{
     fs::{file_handle::FileLike, file_table::FdFlags},
     net::socket::{
         ip::{datagram::DatagramSocket, stream::StreamSocket},
+        netlink::{is_valid_protocol, NetlinkRouteSocket, StandardNetlinkProtocol},
         unix::UnixStreamSocket,
         vsock::VsockStreamSocket,
     },
@@ -16,29 +17,62 @@ pub fn sys_socket(domain: i32, type_: i32, protocol: i32, ctx: &Context) -> Resu
     let domain = CSocketAddrFamily::try_from(domain)?;
     let sock_type = SockType::try_from(type_ & SOCK_TYPE_MASK)?;
     let sock_flags = SockFlags::from_bits_truncate(type_ & !SOCK_TYPE_MASK);
-    let protocol = Protocol::try_from(protocol)?;
     debug!(
-        "domain = {:?}, sock_type = {:?}, sock_flags = {:?}, protocol = {:?}",
-        domain, sock_type, sock_flags, protocol
+        "domain = {:?}, sock_type = {:?}, sock_flags = {:?}",
+        domain, sock_type, sock_flags
     );
-    let nonblocking = sock_flags.contains(SockFlags::SOCK_NONBLOCK);
-    let file_like = match (domain, sock_type, protocol) {
+    let is_nonblocking = sock_flags.contains(SockFlags::SOCK_NONBLOCK);
+    let file_like = match (domain, sock_type) {
         // FIXME: SOCK_SEQPACKET is added to run fcntl_test, not supported yet.
-        (CSocketAddrFamily::AF_UNIX, SockType::SOCK_STREAM | SockType::SOCK_SEQPACKET, _) => {
-            UnixStreamSocket::new(nonblocking) as Arc<dyn FileLike>
+        (CSocketAddrFamily::AF_UNIX, SockType::SOCK_STREAM | SockType::SOCK_SEQPACKET) => {
+            UnixStreamSocket::new(is_nonblocking) as Arc<dyn FileLike>
         }
-        (
-            CSocketAddrFamily::AF_INET,
-            SockType::SOCK_STREAM,
-            Protocol::IPPROTO_IP | Protocol::IPPROTO_TCP,
-        ) => StreamSocket::new(nonblocking) as Arc<dyn FileLike>,
-        (
-            CSocketAddrFamily::AF_INET,
-            SockType::SOCK_DGRAM,
-            Protocol::IPPROTO_IP | Protocol::IPPROTO_UDP,
-        ) => DatagramSocket::new(nonblocking) as Arc<dyn FileLike>,
-        (CSocketAddrFamily::AF_VSOCK, SockType::SOCK_STREAM, _) => {
-            Arc::new(VsockStreamSocket::new(nonblocking)) as Arc<dyn FileLike>
+        (CSocketAddrFamily::AF_INET, SockType::SOCK_STREAM) => {
+            let protocol = Protocol::try_from(protocol)?;
+            debug!("protocol = {:?}", protocol);
+            match protocol {
+                Protocol::IPPROTO_IP | Protocol::IPPROTO_TCP => {
+                    StreamSocket::new(is_nonblocking) as Arc<dyn FileLike>
+                }
+                _ => return_errno_with_message!(Errno::EAFNOSUPPORT, "unsupported protocol"),
+            }
+        }
+        (CSocketAddrFamily::AF_INET, SockType::SOCK_DGRAM) => {
+            let protocol = Protocol::try_from(protocol)?;
+            debug!("protocol = {:?}", protocol);
+            match protocol {
+                Protocol::IPPROTO_IP | Protocol::IPPROTO_UDP => {
+                    DatagramSocket::new(is_nonblocking) as Arc<dyn FileLike>
+                }
+                _ => return_errno_with_message!(Errno::EAFNOSUPPORT, "unsupported protocol"),
+            }
+        }
+        (CSocketAddrFamily::AF_NETLINK, SockType::SOCK_RAW | SockType::SOCK_DGRAM) => {
+            let netlink_family = StandardNetlinkProtocol::try_from(protocol as u32);
+            debug!("netlink family = {:?}", netlink_family);
+            match netlink_family {
+                Ok(StandardNetlinkProtocol::ROUTE) => {
+                    Arc::new(NetlinkRouteSocket::new(is_nonblocking))
+                }
+                Ok(_) => {
+                    return_errno_with_message!(
+                        Errno::EAFNOSUPPORT,
+                        "some standard netlink families are not supported yet"
+                    );
+                }
+                Err(_) => {
+                    if is_valid_protocol(protocol as u32) {
+                        return_errno_with_message!(
+                            Errno::EAFNOSUPPORT,
+                            "user-provided netlink family is not supported"
+                        )
+                    }
+                    return_errno_with_message!(Errno::EAFNOSUPPORT, "invalid netlink family");
+                }
+            }
+        }
+        (CSocketAddrFamily::AF_VSOCK, SockType::SOCK_STREAM) => {
+            Arc::new(VsockStreamSocket::new(is_nonblocking)) as Arc<dyn FileLike>
         }
         _ => return_errno_with_message!(Errno::EAFNOSUPPORT, "unsupported domain"),
     };

kernel/src/util/net/addr/family.rs

@@ -4,7 +4,7 @@ use core::cmp::min;
 
 use ostd::task::Task;
 
-use super::{ip::CSocketAddrInet, unix, vsock::CSocketAddrVm};
+use super::{ip::CSocketAddrInet, netlink::CSocketAddrNetlink, unix, vsock::CSocketAddrVm};
 use crate::{current_userspace, net::socket::SocketAddr, prelude::*};
 
 /// Address family.
@@ -162,6 +162,13 @@ pub fn read_socket_addr_from_user(addr: Vaddr, addr_len: usize) -> Result<Socket
             let addr = unix::from_c_bytes(&storage.as_bytes()[..addr_len])?;
             SocketAddr::Unix(addr)
         }
+        Ok(CSocketAddrFamily::AF_NETLINK) => {
+            if addr_len < size_of::<CSocketAddrNetlink>() {
+                return_errno_with_message!(Errno::EINVAL, "the socket address length is too small");
+            }
+            let addr = CSocketAddrNetlink::from_bytes(storage.as_bytes());
+            SocketAddr::Netlink(addr.into())
+        }
         Ok(CSocketAddrFamily::AF_VSOCK) => {
             if addr_len < size_of::<CSocketAddrVm>() {
                 return_errno_with_message!(Errno::EINVAL, "the socket address length is too small");
@@ -238,36 +245,45 @@ pub fn write_socket_addr_with_max_len(
         );
     }
 
-    let current_task = Task::current().unwrap();
-    let user_space = CurrentUserSpace::new(&current_task);
-
     let actual_len = match socket_addr {
-        SocketAddr::IPv4(addr, port) => {
-            let socket_addr = CSocketAddrInet::from((*addr, *port));
-            let actual_len = size_of::<CSocketAddrInet>();
-            let written_len = min(actual_len, max_len as _);
-            user_space.write_bytes(
+        SocketAddr::IPv4(addr, port) => write_c_socket_address_util::<CSocketAddrInet, _>(
+            (*addr, *port),
             dest,
-                &mut VmReader::from(&socket_addr.as_bytes()[..written_len]),
-            )?;
-            actual_len
-        }
+            max_len as usize,
+        )?,
         SocketAddr::Unix(addr) => unix::into_c_bytes_and(addr, |bytes| {
             let written_len = min(bytes.len(), max_len as _);
-            user_space.write_bytes(dest, &mut VmReader::from(&bytes[..written_len]))?;
+            current_userspace!().write_bytes(dest, &mut VmReader::from(&bytes[..written_len]))?;
             Ok::<usize, Error>(bytes.len())
         })?,
+        SocketAddr::Netlink(addr) => {
+            write_c_socket_address_util::<CSocketAddrNetlink, _>(*addr, dest, max_len as usize)?
+        }
         SocketAddr::Vsock(addr) => {
-            let socket_addr = CSocketAddrVm::from(*addr);
-            let actual_len = size_of::<CSocketAddrVm>();
-            let written_len = min(actual_len, max_len as _);
-            user_space.write_bytes(
-                dest,
-                &mut VmReader::from(&socket_addr.as_bytes()[..written_len]),
-            )?;
-            actual_len
+            write_c_socket_address_util::<CSocketAddrVm, _>(*addr, dest, max_len as usize)?
         }
     };
 
     Ok(actual_len as i32)
 }
+
+// Utility function to write a C socket address to user space.
+fn write_c_socket_address_util<TCSockAddr: Pod, TSockAddr>(
+    addr: TSockAddr,
+    dest: Vaddr,
+    max_len: usize,
+) -> Result<usize>
+where
+    TCSockAddr: From<TSockAddr>,
+{
+    let c_socket_addr = TCSockAddr::from(addr);
+    let actual_len = size_of::<TCSockAddr>();
+    let written_len = min(actual_len, max_len);
+
+    current_userspace!().write_bytes(
+        dest,
+        &mut VmReader::from(&c_socket_addr.as_bytes()[..written_len]),
+    )?;
+
+    Ok(actual_len)
+}

kernel/src/util/net/addr/ip.rs

@@ -38,6 +38,7 @@ impl From<(Ipv4Address, PortNum)> for CSocketAddrInet {
 
 impl From<CSocketAddrInet> for (Ipv4Address, PortNum) {
     fn from(value: CSocketAddrInet) -> Self {
+        debug_assert_eq!(value.sin_family, CSocketAddrFamily::AF_INET as u16);
         (value.sin_addr.into(), value.sin_port.into())
     }
 }

kernel/src/util/net/addr/mod.rs

@@ -7,5 +7,6 @@ pub use family::{
 
 mod family;
 mod ip;
+mod netlink;
 mod unix;
 mod vsock;

kernel/src/util/net/addr/netlink.rs

@@ -0,0 +1,41 @@
+// SPDX-License-Identifier: MPL-2.0
+
+use super::CSocketAddrFamily;
+use crate::{
+    net::socket::netlink::{GroupIdSet, NetlinkSocketAddr},
+    prelude::*,
+};
+
+/// Netlink socket address.
+#[repr(C)]
+#[derive(Debug, Clone, Copy, Pod)]
+pub struct CSocketAddrNetlink {
+    /// Address family (AF_NETLINK).
+    nl_family: u16,
+    /// Pad bytes (always zero).
+    nl_pad: u16,
+    /// Port ID.
+    nl_pid: u32,
+    /// Multicast groups mask.
+    nl_groups: u32,
+}
+
+impl From<NetlinkSocketAddr> for CSocketAddrNetlink {
+    fn from(value: NetlinkSocketAddr) -> Self {
+        Self {
+            nl_family: CSocketAddrFamily::AF_NETLINK as _,
+            nl_pad: 0,
+            nl_pid: value.port(),
+            nl_groups: value.groups().as_u32(),
+        }
+    }
+}
+
+impl From<CSocketAddrNetlink> for NetlinkSocketAddr {
+    fn from(value: CSocketAddrNetlink) -> Self {
+        debug_assert_eq!(value.nl_family, CSocketAddrFamily::AF_NETLINK as u16);
+        let port = value.nl_pid;
+        let groups = GroupIdSet::new(value.nl_groups);
+        NetlinkSocketAddr::new(port, groups)
+    }
+}

kernel/src/util/net/addr/vsock.rs

@@ -33,6 +33,7 @@ impl From<VsockSocketAddr> for CSocketAddrVm {
 
 impl From<CSocketAddrVm> for VsockSocketAddr {
     fn from(value: CSocketAddrVm) -> Self {
+        debug_assert_eq!(value.svm_family, CSocketAddrFamily::AF_VSOCK as u16);
         Self {
             cid: value.svm_cid,
             port: value.svm_port,