Source code

Revision control

Copy as Markdown

Other Tools

// Copyright 2015 The Rust Project Developers.
//
// Licensed under the Apache License, Version 2.0 <LICENSE-APACHE or
// <LICENSE-MIT or https://opensource.org/licenses/MIT>, at your
// option. This file may not be copied, modified, or distributed
// except according to those terms.
use std::fmt;
use std::io::{self, Read, Write};
#[cfg(not(target_os = "redox"))]
use std::io::{IoSlice, IoSliceMut};
use std::mem::MaybeUninit;
#[cfg(not(target_os = "nto"))]
use std::net::Ipv6Addr;
use std::net::{self, Ipv4Addr, Shutdown};
#[cfg(unix)]
use std::os::unix::io::{FromRawFd, IntoRawFd};
#[cfg(windows)]
use std::os::windows::io::{FromRawSocket, IntoRawSocket};
use std::time::Duration;
use crate::sys::{self, c_int, getsockopt, setsockopt, Bool};
use crate::{Domain, Protocol, SockAddr, TcpKeepalive, Type};
#[cfg(not(target_os = "redox"))]
use crate::{MaybeUninitSlice, RecvFlags};
/// Owned wrapper around a system socket.
///
/// This type simply wraps an instance of a file descriptor (`c_int`) on Unix
/// and an instance of `SOCKET` on Windows. This is the main type exported by
/// this crate and is intended to mirror the raw semantics of sockets on
/// platforms as closely as possible. Almost all methods correspond to
/// precisely one libc or OS API call which is essentially just a "Rustic
/// translation" of what's below.
///
/// ## Converting to and from other types
///
/// This type can be freely converted into the network primitives provided by
/// the standard library, such as [`TcpStream`] or [`UdpSocket`], using the
/// [`From`] trait, see the example below.
///
/// [`TcpStream`]: std::net::TcpStream
/// [`UdpSocket`]: std::net::UdpSocket
///
/// # Notes
///
/// Some methods that set options on `Socket` require two system calls to set
/// there options without overwriting previously set options. We do this by
/// first getting the current settings, applying the desired changes and than
/// updating the settings. This means that the operation is **not** atomic. This
/// can lead to a data race when two threads are changing options in parallel.
///
/// # Examples
/// ```no_run
/// # fn main() -> std::io::Result<()> {
/// use std::net::{SocketAddr, TcpListener};
/// use socket2::{Socket, Domain, Type};
///
/// // create a TCP listener bound to two addresses
/// let socket = Socket::new(Domain::IPV4, Type::STREAM, None)?;
///
/// let address: SocketAddr = "[::1]:12345".parse().unwrap();
/// let address = address.into();
/// socket.bind(&address)?;
/// socket.bind(&address)?;
/// socket.listen(128)?;
///
/// let listener: TcpListener = socket.into();
/// // ...
/// # drop(listener);
/// # Ok(()) }
/// ```
pub struct Socket {
inner: Inner,
}
/// Store a `TcpStream` internally to take advantage of its niche optimizations on Unix platforms.
pub(crate) type Inner = std::net::TcpStream;
impl Socket {
/// # Safety
///
/// The caller must ensure `raw` is a valid file descriptor/socket. NOTE:
/// this should really be marked `unsafe`, but this being an internal
/// function, often passed as mapping function, it's makes it very
/// inconvenient to mark it as `unsafe`.
pub(crate) fn from_raw(raw: sys::Socket) -> Socket {
Socket {
inner: unsafe {
// SAFETY: the caller must ensure that `raw` is a valid file
// descriptor, but when it isn't it could return I/O errors, or
// potentially close a fd it doesn't own. All of that isn't
// memory unsafe, so it's not desired but never memory unsafe or
// causes UB.
//
// However there is one exception. We use `TcpStream` to
// represent the `Socket` internally (see `Inner` type),
// `TcpStream` has a layout optimisation that doesn't allow for
// negative file descriptors (as those are always invalid).
// Violating this assumption (fd never negative) causes UB,
// something we don't want. So check for that we have this
// `assert!`.
#[cfg(unix)]
assert!(raw >= 0, "tried to create a `Socket` with an invalid fd");
sys::socket_from_raw(raw)
},
}
}
pub(crate) fn as_raw(&self) -> sys::Socket {
sys::socket_as_raw(&self.inner)
}
pub(crate) fn into_raw(self) -> sys::Socket {
sys::socket_into_raw(self.inner)
}
/// Creates a new socket and sets common flags.
///
/// This function corresponds to `socket(2)` on Unix and `WSASocketW` on
/// Windows.
///
/// On Unix-like systems, the close-on-exec flag is set on the new socket.
/// Additionally, on Apple platforms `SOCK_NOSIGPIPE` is set. On Windows,
/// the socket is made non-inheritable.
///
/// [`Socket::new_raw`] can be used if you don't want these flags to be set.
pub fn new(domain: Domain, ty: Type, protocol: Option<Protocol>) -> io::Result<Socket> {
let ty = set_common_type(ty);
Socket::new_raw(domain, ty, protocol).and_then(set_common_flags)
}
/// Creates a new socket ready to be configured.
///
/// This function corresponds to `socket(2)` on Unix and `WSASocketW` on
/// Windows and simply creates a new socket, no other configuration is done.
pub fn new_raw(domain: Domain, ty: Type, protocol: Option<Protocol>) -> io::Result<Socket> {
let protocol = protocol.map(|p| p.0).unwrap_or(0);
sys::socket(domain.0, ty.0, protocol).map(Socket::from_raw)
}
/// Creates a pair of sockets which are connected to each other.
///
/// This function corresponds to `socketpair(2)`.
///
/// This function sets the same flags as in done for [`Socket::new`],
/// [`Socket::pair_raw`] can be used if you don't want to set those flags.
#[cfg(any(doc, all(feature = "all", unix)))]
#[cfg_attr(docsrs, doc(cfg(all(feature = "all", unix))))]
pub fn pair(
domain: Domain,
ty: Type,
protocol: Option<Protocol>,
) -> io::Result<(Socket, Socket)> {
let ty = set_common_type(ty);
let (a, b) = Socket::pair_raw(domain, ty, protocol)?;
let a = set_common_flags(a)?;
let b = set_common_flags(b)?;
Ok((a, b))
}
/// Creates a pair of sockets which are connected to each other.
///
/// This function corresponds to `socketpair(2)`.
#[cfg(any(doc, all(feature = "all", unix)))]
#[cfg_attr(docsrs, doc(cfg(all(feature = "all", unix))))]
pub fn pair_raw(
domain: Domain,
ty: Type,
protocol: Option<Protocol>,
) -> io::Result<(Socket, Socket)> {
let protocol = protocol.map(|p| p.0).unwrap_or(0);
sys::socketpair(domain.0, ty.0, protocol)
.map(|[a, b]| (Socket::from_raw(a), Socket::from_raw(b)))
}
/// Binds this socket to the specified address.
///
/// This function directly corresponds to the `bind(2)` function on Windows
/// and Unix.
pub fn bind(&self, address: &SockAddr) -> io::Result<()> {
sys::bind(self.as_raw(), address)
}
/// Initiate a connection on this socket to the specified address.
///
/// This function directly corresponds to the `connect(2)` function on
/// Windows and Unix.
///
/// An error will be returned if `listen` or `connect` has already been
/// called on this builder.
///
/// # Notes
///
/// When using a non-blocking connect (by setting the socket into
/// non-blocking mode before calling this function), socket option can't be
/// set *while connecting*. This will cause errors on Windows. Socket
/// options can be safely set before and after connecting the socket.
pub fn connect(&self, address: &SockAddr) -> io::Result<()> {
sys::connect(self.as_raw(), address)
}
/// Initiate a connection on this socket to the specified address, only
/// only waiting for a certain period of time for the connection to be
/// established.
///
/// Unlike many other methods on `Socket`, this does *not* correspond to a
/// single C function. It sets the socket to nonblocking mode, connects via
/// connect(2), and then waits for the connection to complete with poll(2)
/// on Unix and select on Windows. When the connection is complete, the
/// socket is set back to blocking mode. On Unix, this will loop over
/// `EINTR` errors.
///
/// # Warnings
///
/// The non-blocking state of the socket is overridden by this function -
/// it will be returned in blocking mode on success, and in an indeterminate
/// state on failure.
///
/// If the connection request times out, it may still be processing in the
/// background - a second call to `connect` or `connect_timeout` may fail.
pub fn connect_timeout(&self, addr: &SockAddr, timeout: Duration) -> io::Result<()> {
self.set_nonblocking(true)?;
let res = self.connect(addr);
self.set_nonblocking(false)?;
match res {
Ok(()) => return Ok(()),
Err(ref e) if e.kind() == io::ErrorKind::WouldBlock => {}
#[cfg(unix)]
Err(ref e) if e.raw_os_error() == Some(libc::EINPROGRESS) => {}
Err(e) => return Err(e),
}
sys::poll_connect(self, timeout)
}
/// Mark a socket as ready to accept incoming connection requests using
/// [`Socket::accept()`].
///
/// This function directly corresponds to the `listen(2)` function on
/// Windows and Unix.
///
/// An error will be returned if `listen` or `connect` has already been
/// called on this builder.
pub fn listen(&self, backlog: c_int) -> io::Result<()> {
sys::listen(self.as_raw(), backlog)
}
/// Accept a new incoming connection from this listener.
///
/// This function uses `accept4(2)` on platforms that support it and
/// `accept(2)` platforms that do not.
///
/// This function sets the same flags as in done for [`Socket::new`],
/// [`Socket::accept_raw`] can be used if you don't want to set those flags.
pub fn accept(&self) -> io::Result<(Socket, SockAddr)> {
// Use `accept4` on platforms that support it.
#[cfg(any(
target_os = "android",
target_os = "dragonfly",
target_os = "freebsd",
target_os = "fuchsia",
target_os = "illumos",
target_os = "linux",
target_os = "netbsd",
target_os = "openbsd",
))]
return self._accept4(libc::SOCK_CLOEXEC);
// Fall back to `accept` on platforms that do not support `accept4`.
#[cfg(not(any(
target_os = "android",
target_os = "dragonfly",
target_os = "freebsd",
target_os = "fuchsia",
target_os = "illumos",
target_os = "linux",
target_os = "netbsd",
target_os = "openbsd",
)))]
{
let (socket, addr) = self.accept_raw()?;
let socket = set_common_flags(socket)?;
// `set_common_flags` does not disable inheritance on Windows because `Socket::new`
// unlike `accept` is able to create the socket with inheritance disabled.
#[cfg(windows)]
socket._set_no_inherit(true)?;
Ok((socket, addr))
}
}
/// Accept a new incoming connection from this listener.
///
/// This function directly corresponds to the `accept(2)` function on
/// Windows and Unix.
pub fn accept_raw(&self) -> io::Result<(Socket, SockAddr)> {
sys::accept(self.as_raw()).map(|(inner, addr)| (Socket::from_raw(inner), addr))
}
/// Returns the socket address of the local half of this socket.
///
/// # Notes
///
/// Depending on the OS this may return an error if the socket is not
/// [bound].
///
/// [bound]: Socket::bind
pub fn local_addr(&self) -> io::Result<SockAddr> {
sys::getsockname(self.as_raw())
}
/// Returns the socket address of the remote peer of this socket.
///
/// # Notes
///
/// This returns an error if the socket is not [`connect`ed].
///
/// [`connect`ed]: Socket::connect
pub fn peer_addr(&self) -> io::Result<SockAddr> {
sys::getpeername(self.as_raw())
}
/// Returns the [`Type`] of this socket by checking the `SO_TYPE` option on
/// this socket.
pub fn r#type(&self) -> io::Result<Type> {
unsafe { getsockopt::<c_int>(self.as_raw(), sys::SOL_SOCKET, sys::SO_TYPE).map(Type) }
}
/// Creates a new independently owned handle to the underlying socket.
///
/// # Notes
///
/// On Unix this uses `F_DUPFD_CLOEXEC` and thus sets the `FD_CLOEXEC` on
/// the returned socket.
///
/// On Windows this uses `WSA_FLAG_NO_HANDLE_INHERIT` setting inheriting to
/// false.
///
/// On Windows this can **not** be used function cannot be used on a
/// QOS-enabled socket, see
pub fn try_clone(&self) -> io::Result<Socket> {
sys::try_clone(self.as_raw()).map(Socket::from_raw)
}
/// Moves this TCP stream into or out of nonblocking mode.
///
/// # Notes
///
/// On Unix this corresponds to calling `fcntl` (un)setting `O_NONBLOCK`.
///
/// On Windows this corresponds to calling `ioctlsocket` (un)setting
/// `FIONBIO`.
pub fn set_nonblocking(&self, nonblocking: bool) -> io::Result<()> {
sys::set_nonblocking(self.as_raw(), nonblocking)
}
/// Shuts down the read, write, or both halves of this connection.
///
/// This function will cause all pending and future I/O on the specified
/// portions to return immediately with an appropriate value.
pub fn shutdown(&self, how: Shutdown) -> io::Result<()> {
sys::shutdown(self.as_raw(), how)
}
/// Receives data on the socket from the remote address to which it is
/// connected.
///
/// The [`connect`] method will connect this socket to a remote address.
/// This method might fail if the socket is not connected.
///
/// [`connect`]: Socket::connect
///
/// # Safety
///
/// Normally casting a `&mut [u8]` to `&mut [MaybeUninit<u8>]` would be
/// unsound, as that allows us to write uninitialised bytes to the buffer.
/// However this implementation promises to not write uninitialised bytes to
/// the `buf`fer and passes it directly to `recv(2)` system call. This
/// promise ensures that this function can be called using a `buf`fer of
/// type `&mut [u8]`.
///
/// Note that the [`io::Read::read`] implementation calls this function with
/// a `buf`fer of type `&mut [u8]`, allowing initialised buffers to be used
/// without using `unsafe`.
pub fn recv(&self, buf: &mut [MaybeUninit<u8>]) -> io::Result<usize> {
self.recv_with_flags(buf, 0)
}
/// Receives out-of-band (OOB) data on the socket from the remote address to
/// which it is connected by setting the `MSG_OOB` flag for this call.
///
/// For more information, see [`recv`], [`out_of_band_inline`].
///
/// [`recv`]: Socket::recv
/// [`out_of_band_inline`]: Socket::out_of_band_inline
pub fn recv_out_of_band(&self, buf: &mut [MaybeUninit<u8>]) -> io::Result<usize> {
self.recv_with_flags(buf, sys::MSG_OOB)
}
/// Identical to [`recv`] but allows for specification of arbitrary flags to
/// the underlying `recv` call.
///
/// [`recv`]: Socket::recv
pub fn recv_with_flags(
&self,
buf: &mut [MaybeUninit<u8>],
flags: sys::c_int,
) -> io::Result<usize> {
sys::recv(self.as_raw(), buf, flags)
}
/// Receives data on the socket from the remote address to which it is
/// connected. Unlike [`recv`] this allows passing multiple buffers.
///
/// The [`connect`] method will connect this socket to a remote address.
/// This method might fail if the socket is not connected.
///
/// In addition to the number of bytes read, this function returns the flags
/// for the received message. See [`RecvFlags`] for more information about
/// the returned flags.
///
/// [`recv`]: Socket::recv
/// [`connect`]: Socket::connect
///
/// # Safety
///
/// Normally casting a `IoSliceMut` to `MaybeUninitSlice` would be unsound,
/// as that allows us to write uninitialised bytes to the buffer. However
/// this implementation promises to not write uninitialised bytes to the
/// `bufs` and passes it directly to `recvmsg(2)` system call. This promise
/// ensures that this function can be called using `bufs` of type `&mut
/// [IoSliceMut]`.
///
/// Note that the [`io::Read::read_vectored`] implementation calls this
/// function with `buf`s of type `&mut [IoSliceMut]`, allowing initialised
/// buffers to be used without using `unsafe`.
#[cfg(not(target_os = "redox"))]
#[cfg_attr(docsrs, doc(cfg(not(target_os = "redox"))))]
pub fn recv_vectored(
&self,
bufs: &mut [MaybeUninitSlice<'_>],
) -> io::Result<(usize, RecvFlags)> {
self.recv_vectored_with_flags(bufs, 0)
}
/// Identical to [`recv_vectored`] but allows for specification of arbitrary
/// flags to the underlying `recvmsg`/`WSARecv` call.
///
/// [`recv_vectored`]: Socket::recv_vectored
///
/// # Safety
///
/// `recv_from_vectored` makes the same safety guarantees regarding `bufs`
/// as [`recv_vectored`].
///
/// [`recv_vectored`]: Socket::recv_vectored
#[cfg(not(target_os = "redox"))]
#[cfg_attr(docsrs, doc(cfg(not(target_os = "redox"))))]
pub fn recv_vectored_with_flags(
&self,
bufs: &mut [MaybeUninitSlice<'_>],
flags: c_int,
) -> io::Result<(usize, RecvFlags)> {
sys::recv_vectored(self.as_raw(), bufs, flags)
}
/// Receives data on the socket from the remote adress to which it is
/// connected, without removing that data from the queue. On success,
/// returns the number of bytes peeked.
///
/// Successive calls return the same data. This is accomplished by passing
/// `MSG_PEEK` as a flag to the underlying `recv` system call.
///
/// # Safety
///
/// `peek` makes the same safety guarantees regarding the `buf`fer as
/// [`recv`].
///
/// [`recv`]: Socket::recv
pub fn peek(&self, buf: &mut [MaybeUninit<u8>]) -> io::Result<usize> {
self.recv_with_flags(buf, sys::MSG_PEEK)
}
/// Receives data from the socket. On success, returns the number of bytes
/// read and the address from whence the data came.
///
/// # Safety
///
/// `recv_from` makes the same safety guarantees regarding the `buf`fer as
/// [`recv`].
///
/// [`recv`]: Socket::recv
pub fn recv_from(&self, buf: &mut [MaybeUninit<u8>]) -> io::Result<(usize, SockAddr)> {
self.recv_from_with_flags(buf, 0)
}
/// Identical to [`recv_from`] but allows for specification of arbitrary
/// flags to the underlying `recvfrom` call.
///
/// [`recv_from`]: Socket::recv_from
pub fn recv_from_with_flags(
&self,
buf: &mut [MaybeUninit<u8>],
flags: c_int,
) -> io::Result<(usize, SockAddr)> {
sys::recv_from(self.as_raw(), buf, flags)
}
/// Receives data from the socket. Returns the amount of bytes read, the
/// [`RecvFlags`] and the remote address from the data is coming. Unlike
/// [`recv_from`] this allows passing multiple buffers.
///
/// [`recv_from`]: Socket::recv_from
///
/// # Safety
///
/// `recv_from_vectored` makes the same safety guarantees regarding `bufs`
/// as [`recv_vectored`].
///
/// [`recv_vectored`]: Socket::recv_vectored
#[cfg(not(target_os = "redox"))]
#[cfg_attr(docsrs, doc(cfg(not(target_os = "redox"))))]
pub fn recv_from_vectored(
&self,
bufs: &mut [MaybeUninitSlice<'_>],
) -> io::Result<(usize, RecvFlags, SockAddr)> {
self.recv_from_vectored_with_flags(bufs, 0)
}
/// Identical to [`recv_from_vectored`] but allows for specification of
/// arbitrary flags to the underlying `recvmsg`/`WSARecvFrom` call.
///
/// [`recv_from_vectored`]: Socket::recv_from_vectored
///
/// # Safety
///
/// `recv_from_vectored` makes the same safety guarantees regarding `bufs`
/// as [`recv_vectored`].
///
/// [`recv_vectored`]: Socket::recv_vectored
#[cfg(not(target_os = "redox"))]
#[cfg_attr(docsrs, doc(cfg(not(target_os = "redox"))))]
pub fn recv_from_vectored_with_flags(
&self,
bufs: &mut [MaybeUninitSlice<'_>],
flags: c_int,
) -> io::Result<(usize, RecvFlags, SockAddr)> {
sys::recv_from_vectored(self.as_raw(), bufs, flags)
}
/// Receives data from the socket, without removing it from the queue.
///
/// Successive calls return the same data. This is accomplished by passing
/// `MSG_PEEK` as a flag to the underlying `recvfrom` system call.
///
/// On success, returns the number of bytes peeked and the address from
/// whence the data came.
///
/// # Safety
///
/// `peek_from` makes the same safety guarantees regarding the `buf`fer as
/// [`recv`].
///
/// # Note: Datagram Sockets
/// For datagram sockets, the behavior of this method when `buf` is smaller than
/// the datagram at the head of the receive queue differs between Windows and
/// Unix-like platforms (Linux, macOS, BSDs, etc: colloquially termed "*nix").
///
/// On *nix platforms, the datagram is truncated to the length of `buf`.
///
/// On Windows, an error corresponding to `WSAEMSGSIZE` will be returned.
///
/// For consistency between platforms, be sure to provide a sufficiently large buffer to avoid
/// truncation; the exact size required depends on the underlying protocol.
///
/// If you just want to know the sender of the data, try [`peek_sender`].
///
/// [`recv`]: Socket::recv
/// [`peek_sender`]: Socket::peek_sender
pub fn peek_from(&self, buf: &mut [MaybeUninit<u8>]) -> io::Result<(usize, SockAddr)> {
self.recv_from_with_flags(buf, sys::MSG_PEEK)
}
/// Retrieve the sender for the data at the head of the receive queue.
///
/// This is equivalent to calling [`peek_from`] with a zero-sized buffer,
/// but suppresses the `WSAEMSGSIZE` error on Windows.
///
/// [`peek_from`]: Socket::peek_from
pub fn peek_sender(&self) -> io::Result<SockAddr> {
sys::peek_sender(self.as_raw())
}
/// Sends data on the socket to a connected peer.
///
/// This is typically used on TCP sockets or datagram sockets which have
/// been connected.
///
/// On success returns the number of bytes that were sent.
pub fn send(&self, buf: &[u8]) -> io::Result<usize> {
self.send_with_flags(buf, 0)
}
/// Identical to [`send`] but allows for specification of arbitrary flags to the underlying
/// `send` call.
///
/// [`send`]: #method.send
pub fn send_with_flags(&self, buf: &[u8], flags: c_int) -> io::Result<usize> {
sys::send(self.as_raw(), buf, flags)
}
/// Send data to the connected peer. Returns the amount of bytes written.
#[cfg(not(target_os = "redox"))]
#[cfg_attr(docsrs, doc(cfg(not(target_os = "redox"))))]
pub fn send_vectored(&self, bufs: &[IoSlice<'_>]) -> io::Result<usize> {
self.send_vectored_with_flags(bufs, 0)
}
/// Identical to [`send_vectored`] but allows for specification of arbitrary
/// flags to the underlying `sendmsg`/`WSASend` call.
///
/// [`send_vectored`]: Socket::send_vectored
#[cfg(not(target_os = "redox"))]
#[cfg_attr(docsrs, doc(cfg(not(target_os = "redox"))))]
pub fn send_vectored_with_flags(
&self,
bufs: &[IoSlice<'_>],
flags: c_int,
) -> io::Result<usize> {
sys::send_vectored(self.as_raw(), bufs, flags)
}
/// Sends out-of-band (OOB) data on the socket to connected peer
/// by setting the `MSG_OOB` flag for this call.
///
/// For more information, see [`send`], [`out_of_band_inline`].
///
/// [`send`]: #method.send
/// [`out_of_band_inline`]: #method.out_of_band_inline
pub fn send_out_of_band(&self, buf: &[u8]) -> io::Result<usize> {
self.send_with_flags(buf, sys::MSG_OOB)
}
/// Sends data on the socket to the given address. On success, returns the
/// number of bytes written.
///
/// This is typically used on UDP or datagram-oriented sockets.
pub fn send_to(&self, buf: &[u8], addr: &SockAddr) -> io::Result<usize> {
self.send_to_with_flags(buf, addr, 0)
}
/// Identical to [`send_to`] but allows for specification of arbitrary flags
/// to the underlying `sendto` call.
///
/// [`send_to`]: Socket::send_to
pub fn send_to_with_flags(
&self,
buf: &[u8],
addr: &SockAddr,
flags: c_int,
) -> io::Result<usize> {
sys::send_to(self.as_raw(), buf, addr, flags)
}
/// Send data to a peer listening on `addr`. Returns the amount of bytes
/// written.
#[cfg(not(target_os = "redox"))]
#[cfg_attr(docsrs, doc(cfg(not(target_os = "redox"))))]
pub fn send_to_vectored(&self, bufs: &[IoSlice<'_>], addr: &SockAddr) -> io::Result<usize> {
self.send_to_vectored_with_flags(bufs, addr, 0)
}
/// Identical to [`send_to_vectored`] but allows for specification of
/// arbitrary flags to the underlying `sendmsg`/`WSASendTo` call.
///
/// [`send_to_vectored`]: Socket::send_to_vectored
#[cfg(not(target_os = "redox"))]
#[cfg_attr(docsrs, doc(cfg(not(target_os = "redox"))))]
pub fn send_to_vectored_with_flags(
&self,
bufs: &[IoSlice<'_>],
addr: &SockAddr,
flags: c_int,
) -> io::Result<usize> {
sys::send_to_vectored(self.as_raw(), bufs, addr, flags)
}
}
/// Set `SOCK_CLOEXEC` and `NO_HANDLE_INHERIT` on the `ty`pe on platforms that
/// support it.
#[inline(always)]
fn set_common_type(ty: Type) -> Type {
// On platforms that support it set `SOCK_CLOEXEC`.
#[cfg(any(
target_os = "android",
target_os = "dragonfly",
target_os = "freebsd",
target_os = "fuchsia",
target_os = "illumos",
target_os = "linux",
target_os = "netbsd",
target_os = "openbsd",
))]
let ty = ty._cloexec();
// On windows set `NO_HANDLE_INHERIT`.
#[cfg(windows)]
let ty = ty._no_inherit();
ty
}
/// Set `FD_CLOEXEC` and `NOSIGPIPE` on the `socket` for platforms that need it.
#[inline(always)]
#[allow(clippy::unnecessary_wraps)]
fn set_common_flags(socket: Socket) -> io::Result<Socket> {
// On platforms that don't have `SOCK_CLOEXEC` use `FD_CLOEXEC`.
#[cfg(all(
unix,
not(any(
target_os = "android",
target_os = "dragonfly",
target_os = "freebsd",
target_os = "fuchsia",
target_os = "illumos",
target_os = "linux",
target_os = "netbsd",
target_os = "openbsd",
))
))]
socket._set_cloexec(true)?;
// On Apple platforms set `NOSIGPIPE`.
#[cfg(target_vendor = "apple")]
socket._set_nosigpipe(true)?;
Ok(socket)
}
/// A local interface specified by its index or an address assigned to it.
///
/// `Index(0)` and `Address(Ipv4Addr::UNSPECIFIED)` are equivalent and indicate
/// that an appropriate interface should be selected by the system.
#[cfg(not(any(
target_os = "haiku",
target_os = "illumos",
target_os = "netbsd",
target_os = "redox",
target_os = "solaris",
)))]
#[derive(Debug)]
pub enum InterfaceIndexOrAddress {
/// An interface index.
Index(u32),
/// An address assigned to an interface.
Address(Ipv4Addr),
}
/// Socket options get/set using `SOL_SOCKET`.
///
/// Additional documentation can be found in documentation of the OS.
impl Socket {
/// Get the value of the `SO_BROADCAST` option for this socket.
///
/// For more information about this option, see [`set_broadcast`].
///
/// [`set_broadcast`]: Socket::set_broadcast
pub fn broadcast(&self) -> io::Result<bool> {
unsafe {
getsockopt::<c_int>(self.as_raw(), sys::SOL_SOCKET, sys::SO_BROADCAST)
.map(|broadcast| broadcast != 0)
}
}
/// Set the value of the `SO_BROADCAST` option for this socket.
///
/// When enabled, this socket is allowed to send packets to a broadcast
/// address.
pub fn set_broadcast(&self, broadcast: bool) -> io::Result<()> {
unsafe {
setsockopt(
self.as_raw(),
sys::SOL_SOCKET,
sys::SO_BROADCAST,
broadcast as c_int,
)
}
}
/// Get the value of the `SO_ERROR` option on this socket.
///
/// This will retrieve the stored error in the underlying socket, clearing
/// the field in the process. This can be useful for checking errors between
/// calls.
pub fn take_error(&self) -> io::Result<Option<io::Error>> {
match unsafe { getsockopt::<c_int>(self.as_raw(), sys::SOL_SOCKET, sys::SO_ERROR) } {
Ok(0) => Ok(None),
Ok(errno) => Ok(Some(io::Error::from_raw_os_error(errno))),
Err(err) => Err(err),
}
}
/// Get the value of the `SO_KEEPALIVE` option on this socket.
///
/// For more information about this option, see [`set_keepalive`].
///
/// [`set_keepalive`]: Socket::set_keepalive
pub fn keepalive(&self) -> io::Result<bool> {
unsafe {
getsockopt::<Bool>(self.as_raw(), sys::SOL_SOCKET, sys::SO_KEEPALIVE)
.map(|keepalive| keepalive != 0)
}
}
/// Set value for the `SO_KEEPALIVE` option on this socket.
///
/// Enable sending of keep-alive messages on connection-oriented sockets.
pub fn set_keepalive(&self, keepalive: bool) -> io::Result<()> {
unsafe {
setsockopt(
self.as_raw(),
sys::SOL_SOCKET,
sys::SO_KEEPALIVE,
keepalive as c_int,
)
}
}
/// Get the value of the `SO_LINGER` option on this socket.
///
/// For more information about this option, see [`set_linger`].
///
/// [`set_linger`]: Socket::set_linger
pub fn linger(&self) -> io::Result<Option<Duration>> {
unsafe {
getsockopt::<sys::linger>(self.as_raw(), sys::SOL_SOCKET, sys::SO_LINGER)
.map(from_linger)
}
}
/// Set value for the `SO_LINGER` option on this socket.
///
/// If `linger` is not `None`, a close(2) or shutdown(2) will not return
/// until all queued messages for the socket have been successfully sent or
/// the linger timeout has been reached. Otherwise, the call returns
/// immediately and the closing is done in the background. When the socket
/// is closed as part of exit(2), it always lingers in the background.
///
/// # Notes
///
/// On most OSs the duration only has a precision of seconds and will be
/// silently truncated.
///
/// On Apple platforms (e.g. macOS, iOS, etc) this uses `SO_LINGER_SEC`.
pub fn set_linger(&self, linger: Option<Duration>) -> io::Result<()> {
let linger = into_linger(linger);
unsafe { setsockopt(self.as_raw(), sys::SOL_SOCKET, sys::SO_LINGER, linger) }
}
/// Get value for the `SO_OOBINLINE` option on this socket.
///
/// For more information about this option, see [`set_out_of_band_inline`].
///
/// [`set_out_of_band_inline`]: Socket::set_out_of_band_inline
#[cfg(not(target_os = "redox"))]
#[cfg_attr(docsrs, doc(cfg(not(target_os = "redox"))))]
pub fn out_of_band_inline(&self) -> io::Result<bool> {
unsafe {
getsockopt::<c_int>(self.as_raw(), sys::SOL_SOCKET, sys::SO_OOBINLINE)
.map(|oob_inline| oob_inline != 0)
}
}
/// Set value for the `SO_OOBINLINE` option on this socket.
///
/// If this option is enabled, out-of-band data is directly placed into the
/// receive data stream. Otherwise, out-of-band data is passed only when the
/// `MSG_OOB` flag is set during receiving. As per RFC6093, TCP sockets
/// using the Urgent mechanism are encouraged to set this flag.
#[cfg(not(target_os = "redox"))]
#[cfg_attr(docsrs, doc(cfg(not(target_os = "redox"))))]
pub fn set_out_of_band_inline(&self, oob_inline: bool) -> io::Result<()> {
unsafe {
setsockopt(
self.as_raw(),
sys::SOL_SOCKET,
sys::SO_OOBINLINE,
oob_inline as c_int,
)
}
}
/// Get value for the `SO_RCVBUF` option on this socket.
///
/// For more information about this option, see [`set_recv_buffer_size`].
///
/// [`set_recv_buffer_size`]: Socket::set_recv_buffer_size
pub fn recv_buffer_size(&self) -> io::Result<usize> {
unsafe {
getsockopt::<c_int>(self.as_raw(), sys::SOL_SOCKET, sys::SO_RCVBUF)
.map(|size| size as usize)
}
}
/// Set value for the `SO_RCVBUF` option on this socket.
///
/// Changes the size of the operating system's receive buffer associated
/// with the socket.
pub fn set_recv_buffer_size(&self, size: usize) -> io::Result<()> {
unsafe {
setsockopt(
self.as_raw(),
sys::SOL_SOCKET,
sys::SO_RCVBUF,
size as c_int,
)
}
}
/// Get value for the `SO_RCVTIMEO` option on this socket.
///
/// If the returned timeout is `None`, then `read` and `recv` calls will
/// block indefinitely.
pub fn read_timeout(&self) -> io::Result<Option<Duration>> {
sys::timeout_opt(self.as_raw(), sys::SOL_SOCKET, sys::SO_RCVTIMEO)
}
/// Set value for the `SO_RCVTIMEO` option on this socket.
///
/// If `timeout` is `None`, then `read` and `recv` calls will block
/// indefinitely.
pub fn set_read_timeout(&self, duration: Option<Duration>) -> io::Result<()> {
sys::set_timeout_opt(self.as_raw(), sys::SOL_SOCKET, sys::SO_RCVTIMEO, duration)
}
/// Get the value of the `SO_REUSEADDR` option on this socket.
///
/// For more information about this option, see [`set_reuse_address`].
///
/// [`set_reuse_address`]: Socket::set_reuse_address
pub fn reuse_address(&self) -> io::Result<bool> {
unsafe {
getsockopt::<c_int>(self.as_raw(), sys::SOL_SOCKET, sys::SO_REUSEADDR)
.map(|reuse| reuse != 0)
}
}
/// Set value for the `SO_REUSEADDR` option on this socket.
///
/// This indicates that futher calls to `bind` may allow reuse of local
/// addresses. For IPv4 sockets this means that a socket may bind even when
/// there's a socket already listening on this port.
pub fn set_reuse_address(&self, reuse: bool) -> io::Result<()> {
unsafe {
setsockopt(
self.as_raw(),
sys::SOL_SOCKET,
sys::SO_REUSEADDR,
reuse as c_int,
)
}
}
/// Get the value of the `SO_SNDBUF` option on this socket.
///
/// For more information about this option, see [`set_send_buffer_size`].
///
/// [`set_send_buffer_size`]: Socket::set_send_buffer_size
pub fn send_buffer_size(&self) -> io::Result<usize> {
unsafe {
getsockopt::<c_int>(self.as_raw(), sys::SOL_SOCKET, sys::SO_SNDBUF)
.map(|size| size as usize)
}
}
/// Set value for the `SO_SNDBUF` option on this socket.
///
/// Changes the size of the operating system's send buffer associated with
/// the socket.
pub fn set_send_buffer_size(&self, size: usize) -> io::Result<()> {
unsafe {
setsockopt(
self.as_raw(),
sys::SOL_SOCKET,
sys::SO_SNDBUF,
size as c_int,
)
}
}
/// Get value for the `SO_SNDTIMEO` option on this socket.
///
/// If the returned timeout is `None`, then `write` and `send` calls will
/// block indefinitely.
pub fn write_timeout(&self) -> io::Result<Option<Duration>> {
sys::timeout_opt(self.as_raw(), sys::SOL_SOCKET, sys::SO_SNDTIMEO)
}
/// Set value for the `SO_SNDTIMEO` option on this socket.
///
/// If `timeout` is `None`, then `write` and `send` calls will block
/// indefinitely.
pub fn set_write_timeout(&self, duration: Option<Duration>) -> io::Result<()> {
sys::set_timeout_opt(self.as_raw(), sys::SOL_SOCKET, sys::SO_SNDTIMEO, duration)
}
}
fn from_linger(linger: sys::linger) -> Option<Duration> {
if linger.l_onoff == 0 {
None
} else {
Some(Duration::from_secs(linger.l_linger as u64))
}
}
fn into_linger(duration: Option<Duration>) -> sys::linger {
match duration {
Some(duration) => sys::linger {
l_onoff: 1,
l_linger: duration.as_secs() as _,
},
None => sys::linger {
l_onoff: 0,
l_linger: 0,
},
}
}
/// Socket options for IPv4 sockets, get/set using `IPPROTO_IP`.
///
/// Additional documentation can be found in documentation of the OS.
impl Socket {
/// Get the value of the `IP_HDRINCL` option on this socket.
///
/// For more information about this option, see [`set_header_included`].
///
/// [`set_header_included`]: Socket::set_header_included
#[cfg(all(feature = "all", not(target_os = "redox")))]
#[cfg_attr(docsrs, doc(all(feature = "all", not(target_os = "redox"))))]
pub fn header_included(&self) -> io::Result<bool> {
unsafe {
getsockopt::<c_int>(self.as_raw(), sys::IPPROTO_IP, sys::IP_HDRINCL)
.map(|included| included != 0)
}
}
/// Set the value of the `IP_HDRINCL` option on this socket.
///
/// If enabled, the user supplies an IP header in front of the user data.
/// Valid only for [`SOCK_RAW`] sockets; see [raw(7)] for more information.
/// When this flag is enabled, the values set by `IP_OPTIONS`, [`IP_TTL`],
/// and [`IP_TOS`] are ignored.
///
/// [`SOCK_RAW`]: Type::RAW
/// [`IP_TTL`]: Socket::set_ttl
/// [`IP_TOS`]: Socket::set_tos
#[cfg(all(feature = "all", not(target_os = "redox")))]
#[cfg_attr(docsrs, doc(all(feature = "all", not(target_os = "redox"))))]
pub fn set_header_included(&self, included: bool) -> io::Result<()> {
unsafe {
setsockopt(
self.as_raw(),
sys::IPPROTO_IP,
sys::IP_HDRINCL,
included as c_int,
)
}
}
/// Get the value of the `IP_TRANSPARENT` option on this socket.
///
/// For more information about this option, see [`set_ip_transparent`].
///
/// [`set_ip_transparent`]: Socket::set_ip_transparent
#[cfg(any(doc, all(feature = "all", target_os = "linux")))]
#[cfg_attr(docsrs, doc(cfg(all(feature = "all", target_os = "linux"))))]
pub fn ip_transparent(&self) -> io::Result<bool> {
unsafe {
getsockopt::<c_int>(self.as_raw(), sys::IPPROTO_IP, libc::IP_TRANSPARENT)
.map(|transparent| transparent != 0)
}
}
/// Set the value of the `IP_TRANSPARENT` option on this socket.
///
/// Setting this boolean option enables transparent proxying
/// on this socket. This socket option allows the calling
/// application to bind to a nonlocal IP address and operate
/// both as a client and a server with the foreign address as
/// the local endpoint. NOTE: this requires that routing be
/// set up in a way that packets going to the foreign address
/// are routed through the TProxy box (i.e., the system
/// hosting the application that employs the IP_TRANSPARENT
/// socket option). Enabling this socket option requires
/// superuser privileges (the `CAP_NET_ADMIN` capability).
///
/// TProxy redirection with the iptables TPROXY target also
/// requires that this option be set on the redirected socket.
#[cfg(any(doc, all(feature = "all", target_os = "linux")))]
#[cfg_attr(docsrs, doc(cfg(all(feature = "all", target_os = "linux"))))]
pub fn set_ip_transparent(&self, transparent: bool) -> io::Result<()> {
unsafe {
setsockopt(
self.as_raw(),
sys::IPPROTO_IP,
libc::IP_TRANSPARENT,
transparent as c_int,
)
}
}
/// Join a multicast group using `IP_ADD_MEMBERSHIP` option on this socket.
///
/// This function specifies a new multicast group for this socket to join.
/// The address must be a valid multicast address, and `interface` is the
/// address of the local interface with which the system should join the
/// multicast group. If it's [`Ipv4Addr::UNSPECIFIED`] (`INADDR_ANY`) then
/// an appropriate interface is chosen by the system.
pub fn join_multicast_v4(&self, multiaddr: &Ipv4Addr, interface: &Ipv4Addr) -> io::Result<()> {
let mreq = sys::IpMreq {
imr_multiaddr: sys::to_in_addr(multiaddr),
imr_interface: sys::to_in_addr(interface),
};
unsafe { setsockopt(self.as_raw(), sys::IPPROTO_IP, sys::IP_ADD_MEMBERSHIP, mreq) }
}
/// Leave a multicast group using `IP_DROP_MEMBERSHIP` option on this socket.
///
/// For more information about this option, see [`join_multicast_v4`].
///
/// [`join_multicast_v4`]: Socket::join_multicast_v4
pub fn leave_multicast_v4(&self, multiaddr: &Ipv4Addr, interface: &Ipv4Addr) -> io::Result<()> {
let mreq = sys::IpMreq {
imr_multiaddr: sys::to_in_addr(multiaddr),
imr_interface: sys::to_in_addr(interface),
};
unsafe {
setsockopt(
self.as_raw(),
sys::IPPROTO_IP,
sys::IP_DROP_MEMBERSHIP,
mreq,
)
}
}
/// Join a multicast group using `IP_ADD_MEMBERSHIP` option on this socket.
///
/// This function specifies a new multicast group for this socket to join.
/// The address must be a valid multicast address, and `interface` specifies
/// the local interface with which the system should join the multicast
/// group. See [`InterfaceIndexOrAddress`].
#[cfg(not(any(
target_os = "haiku",
target_os = "illumos",
target_os = "netbsd",
target_os = "openbsd",
target_os = "redox",
target_os = "solaris",
target_os = "nto",
)))]
pub fn join_multicast_v4_n(
&self,
multiaddr: &Ipv4Addr,
interface: &InterfaceIndexOrAddress,
) -> io::Result<()> {
let mreqn = sys::to_mreqn(multiaddr, interface);
unsafe {
setsockopt(
self.as_raw(),
sys::IPPROTO_IP,
sys::IP_ADD_MEMBERSHIP,
mreqn,
)
}
}
/// Leave a multicast group using `IP_DROP_MEMBERSHIP` option on this socket.
///
/// For more information about this option, see [`join_multicast_v4_n`].
///
/// [`join_multicast_v4_n`]: Socket::join_multicast_v4_n
#[cfg(not(any(
target_os = "haiku",
target_os = "illumos",
target_os = "netbsd",
target_os = "openbsd",
target_os = "redox",
target_os = "solaris",
target_os = "nto",
)))]
pub fn leave_multicast_v4_n(
&self,
multiaddr: &Ipv4Addr,
interface: &InterfaceIndexOrAddress,
) -> io::Result<()> {
let mreqn = sys::to_mreqn(multiaddr, interface);
unsafe {
setsockopt(
self.as_raw(),
sys::IPPROTO_IP,
sys::IP_DROP_MEMBERSHIP,
mreqn,
)
}
}
/// Join a multicast SSM channel using `IP_ADD_SOURCE_MEMBERSHIP` option on this socket.
///
/// This function specifies a new multicast channel for this socket to join.
/// The group must be a valid SSM group address, the source must be the address of the sender
/// and `interface` is the address of the local interface with which the system should join the
/// multicast group. If it's [`Ipv4Addr::UNSPECIFIED`] (`INADDR_ANY`) then
/// an appropriate interface is chosen by the system.
#[cfg(not(any(
target_os = "dragonfly",
target_os = "haiku",
target_os = "netbsd",
target_os = "openbsd",
target_os = "redox",
target_os = "fuchsia",
target_os = "nto",
)))]
pub fn join_ssm_v4(
&self,
source: &Ipv4Addr,
group: &Ipv4Addr,
interface: &Ipv4Addr,
) -> io::Result<()> {
let mreqs = sys::IpMreqSource {
imr_multiaddr: sys::to_in_addr(group),
imr_interface: sys::to_in_addr(interface),
imr_sourceaddr: sys::to_in_addr(source),
};
unsafe {
setsockopt(
self.as_raw(),
sys::IPPROTO_IP,
sys::IP_ADD_SOURCE_MEMBERSHIP,
mreqs,
)
}
}
/// Leave a multicast group using `IP_DROP_SOURCE_MEMBERSHIP` option on this socket.
///
/// For more information about this option, see [`join_ssm_v4`].
///
/// [`join_ssm_v4`]: Socket::join_ssm_v4
#[cfg(not(any(
target_os = "dragonfly",
target_os = "haiku",
target_os = "netbsd",
target_os = "openbsd",
target_os = "redox",
target_os = "fuchsia",
target_os = "nto",
)))]
pub fn leave_ssm_v4(
&self,
source: &Ipv4Addr,
group: &Ipv4Addr,
interface: &Ipv4Addr,
) -> io::Result<()> {
let mreqs = sys::IpMreqSource {
imr_multiaddr: sys::to_in_addr(group),
imr_interface: sys::to_in_addr(interface),
imr_sourceaddr: sys::to_in_addr(source),
};
unsafe {
setsockopt(
self.as_raw(),
sys::IPPROTO_IP,
sys::IP_DROP_SOURCE_MEMBERSHIP,
mreqs,
)
}
}
/// Get the value of the `IP_MULTICAST_IF` option for this socket.
///
/// For more information about this option, see [`set_multicast_if_v4`].
///
/// [`set_multicast_if_v4`]: Socket::set_multicast_if_v4
pub fn multicast_if_v4(&self) -> io::Result<Ipv4Addr> {
unsafe {
getsockopt(self.as_raw(), sys::IPPROTO_IP, sys::IP_MULTICAST_IF).map(sys::from_in_addr)
}
}
/// Set the value of the `IP_MULTICAST_IF` option for this socket.
///
/// Specifies the interface to use for routing multicast packets.
pub fn set_multicast_if_v4(&self, interface: &Ipv4Addr) -> io::Result<()> {
let interface = sys::to_in_addr(interface);
unsafe {
setsockopt(
self.as_raw(),
sys::IPPROTO_IP,
sys::IP_MULTICAST_IF,
interface,
)
}
}
/// Get the value of the `IP_MULTICAST_LOOP` option for this socket.
///
/// For more information about this option, see [`set_multicast_loop_v4`].
///
/// [`set_multicast_loop_v4`]: Socket::set_multicast_loop_v4
pub fn multicast_loop_v4(&self) -> io::Result<bool> {
unsafe {
getsockopt::<c_int>(self.as_raw(), sys::IPPROTO_IP, sys::IP_MULTICAST_LOOP)
.map(|loop_v4| loop_v4 != 0)
}
}
/// Set the value of the `IP_MULTICAST_LOOP` option for this socket.
///
/// If enabled, multicast packets will be looped back to the local socket.
/// Note that this may not have any affect on IPv6 sockets.
pub fn set_multicast_loop_v4(&self, loop_v4: bool) -> io::Result<()> {
unsafe {
setsockopt(
self.as_raw(),
sys::IPPROTO_IP,
sys::IP_MULTICAST_LOOP,
loop_v4 as c_int,
)
}
}
/// Get the value of the `IP_MULTICAST_TTL` option for this socket.
///
/// For more information about this option, see [`set_multicast_ttl_v4`].
///
/// [`set_multicast_ttl_v4`]: Socket::set_multicast_ttl_v4
pub fn multicast_ttl_v4(&self) -> io::Result<u32> {
unsafe {
getsockopt::<c_int>(self.as_raw(), sys::IPPROTO_IP, sys::IP_MULTICAST_TTL)
.map(|ttl| ttl as u32)
}
}
/// Set the value of the `IP_MULTICAST_TTL` option for this socket.
///
/// Indicates the time-to-live value of outgoing multicast packets for
/// this socket. The default value is 1 which means that multicast packets
/// don't leave the local network unless explicitly requested.
///
/// Note that this may not have any affect on IPv6 sockets.
pub fn set_multicast_ttl_v4(&self, ttl: u32) -> io::Result<()> {
unsafe {
setsockopt(
self.as_raw(),
sys::IPPROTO_IP,
sys::IP_MULTICAST_TTL,
ttl as c_int,
)
}
}
/// Get the value of the `IP_TTL` option for this socket.
///
/// For more information about this option, see [`set_ttl`].
///
/// [`set_ttl`]: Socket::set_ttl
pub fn ttl(&self) -> io::Result<u32> {
unsafe {
getsockopt::<c_int>(self.as_raw(), sys::IPPROTO_IP, sys::IP_TTL).map(|ttl| ttl as u32)
}
}
/// Set the value of the `IP_TTL` option for this socket.
///
/// This value sets the time-to-live field that is used in every packet sent
/// from this socket.
pub fn set_ttl(&self, ttl: u32) -> io::Result<()> {
unsafe { setsockopt(self.as_raw(), sys::IPPROTO_IP, sys::IP_TTL, ttl as c_int) }
}
/// Set the value of the `IP_TOS` option for this socket.
///
/// This value sets the type-of-service field that is used in every packet
/// sent from this socket.
///
/// documents that not all versions of windows support `IP_TOS`.
#[cfg(not(any(
target_os = "fuchsia",
target_os = "redox",
target_os = "solaris",
target_os = "illumos",
)))]
pub fn set_tos(&self, tos: u32) -> io::Result<()> {
unsafe { setsockopt(self.as_raw(), sys::IPPROTO_IP, sys::IP_TOS, tos as c_int) }
}
/// Get the value of the `IP_TOS` option for this socket.
///
/// For more information about this option, see [`set_tos`].
///
/// documents that not all versions of windows support `IP_TOS`.
///
/// [`set_tos`]: Socket::set_tos
#[cfg(not(any(
target_os = "fuchsia",
target_os = "redox",
target_os = "solaris",
target_os = "illumos",
)))]
pub fn tos(&self) -> io::Result<u32> {
unsafe {
getsockopt::<c_int>(self.as_raw(), sys::IPPROTO_IP, sys::IP_TOS).map(|tos| tos as u32)
}
}
/// Set the value of the `IP_RECVTOS` option for this socket.
///
/// If enabled, the IP_TOS ancillary message is passed with
/// incoming packets. It contains a byte which specifies the
/// Type of Service/Precedence field of the packet header.
#[cfg(not(any(
target_os = "dragonfly",
target_os = "fuchsia",
target_os = "illumos",
target_os = "netbsd",
target_os = "openbsd",
target_os = "redox",
target_os = "solaris",
target_os = "windows",
target_os = "nto",
)))]
pub fn set_recv_tos(&self, recv_tos: bool) -> io::Result<()> {
let recv_tos = if recv_tos { 1 } else { 0 };
unsafe {
setsockopt(
self.as_raw(),
sys::IPPROTO_IP,
sys::IP_RECVTOS,
recv_tos as c_int,
)
}
}
/// Get the value of the `IP_RECVTOS` option for this socket.
///
/// For more information about this option, see [`set_recv_tos`].
///
/// [`set_recv_tos`]: Socket::set_recv_tos
#[cfg(not(any(
target_os = "dragonfly",
target_os = "fuchsia",
target_os = "illumos",
target_os = "netbsd",
target_os = "openbsd",
target_os = "redox",
target_os = "solaris",
target_os = "windows",
target_os = "nto",
)))]
pub fn recv_tos(&self) -> io::Result<bool> {
unsafe {
getsockopt::<c_int>(self.as_raw(), sys::IPPROTO_IP, sys::IP_RECVTOS)
.map(|recv_tos| recv_tos > 0)
}
}
}
/// Socket options for IPv6 sockets, get/set using `IPPROTO_IPV6`.
///
/// Additional documentation can be found in documentation of the OS.
impl Socket {
/// Join a multicast group using `IPV6_ADD_MEMBERSHIP` option on this socket.
///
/// Some OSs use `IPV6_JOIN_GROUP` for this option.
///
/// This function specifies a new multicast group for this socket to join.
/// The address must be a valid multicast address, and `interface` is the
/// index of the interface to join/leave (or 0 to indicate any interface).
#[cfg(not(target_os = "nto"))]
pub fn join_multicast_v6(&self, multiaddr: &Ipv6Addr, interface: u32) -> io::Result<()> {
let mreq = sys::Ipv6Mreq {
ipv6mr_multiaddr: sys::to_in6_addr(multiaddr),
// NOTE: some OSs use `c_int`, others use `c_uint`.
ipv6mr_interface: interface as _,
};
unsafe {
setsockopt(
self.as_raw(),
sys::IPPROTO_IPV6,
sys::IPV6_ADD_MEMBERSHIP,
mreq,
)
}
}
/// Leave a multicast group using `IPV6_DROP_MEMBERSHIP` option on this socket.
///
/// Some OSs use `IPV6_LEAVE_GROUP` for this option.
///
/// For more information about this option, see [`join_multicast_v6`].
///
/// [`join_multicast_v6`]: Socket::join_multicast_v6
#[cfg(not(target_os = "nto"))]
pub fn leave_multicast_v6(&self, multiaddr: &Ipv6Addr, interface: u32) -> io::Result<()> {
let mreq = sys::Ipv6Mreq {
ipv6mr_multiaddr: sys::to_in6_addr(multiaddr),
// NOTE: some OSs use `c_int`, others use `c_uint`.
ipv6mr_interface: interface as _,
};
unsafe {
setsockopt(
self.as_raw(),
sys::IPPROTO_IPV6,
sys::IPV6_DROP_MEMBERSHIP,
mreq,
)
}
}
/// Get the value of the `IPV6_MULTICAST_HOPS` option for this socket
///
/// For more information about this option, see [`set_multicast_hops_v6`].
///
/// [`set_multicast_hops_v6`]: Socket::set_multicast_hops_v6
pub fn multicast_hops_v6(&self) -> io::Result<u32> {
unsafe {
getsockopt::<c_int>(self.as_raw(), sys::IPPROTO_IPV6, sys::IPV6_MULTICAST_HOPS)
.map(|hops| hops as u32)
}
}
/// Set the value of the `IPV6_MULTICAST_HOPS` option for this socket
///
/// Indicates the number of "routers" multicast packets will transit for
/// this socket. The default value is 1 which means that multicast packets
/// don't leave the local network unless explicitly requested.
pub fn set_multicast_hops_v6(&self, hops: u32) -> io::Result<()> {
unsafe {
setsockopt(
self.as_raw(),
sys::IPPROTO_IPV6,
sys::IPV6_MULTICAST_HOPS,
hops as c_int,
)
}
}
/// Get the value of the `IPV6_MULTICAST_IF` option for this socket.
///
/// For more information about this option, see [`set_multicast_if_v6`].
///
/// [`set_multicast_if_v6`]: Socket::set_multicast_if_v6
pub fn multicast_if_v6(&self) -> io::Result<u32> {
unsafe {
getsockopt::<c_int>(self.as_raw(), sys::IPPROTO_IPV6, sys::IPV6_MULTICAST_IF)
.map(|interface| interface as u32)
}
}
/// Set the value of the `IPV6_MULTICAST_IF` option for this socket.
///
/// Specifies the interface to use for routing multicast packets. Unlike
/// ipv4, this is generally required in ipv6 contexts where network routing
/// prefixes may overlap.
pub fn set_multicast_if_v6(&self, interface: u32) -> io::Result<()> {
unsafe {
setsockopt(
self.as_raw(),
sys::IPPROTO_IPV6,
sys::IPV6_MULTICAST_IF,
interface as c_int,
)
}
}
/// Get the value of the `IPV6_MULTICAST_LOOP` option for this socket.
///
/// For more information about this option, see [`set_multicast_loop_v6`].
///
/// [`set_multicast_loop_v6`]: Socket::set_multicast_loop_v6
pub fn multicast_loop_v6(&self) -> io::Result<bool> {
unsafe {
getsockopt::<c_int>(self.as_raw(), sys::IPPROTO_IPV6, sys::IPV6_MULTICAST_LOOP)
.map(|loop_v6| loop_v6 != 0)
}
}
/// Set the value of the `IPV6_MULTICAST_LOOP` option for this socket.
///
/// Controls whether this socket sees the multicast packets it sends itself.
/// Note that this may not have any affect on IPv4 sockets.
pub fn set_multicast_loop_v6(&self, loop_v6: bool) -> io::Result<()> {
unsafe {
setsockopt(
self.as_raw(),
sys::IPPROTO_IPV6,
sys::IPV6_MULTICAST_LOOP,
loop_v6 as c_int,
)
}
}
/// Get the value of the `IPV6_UNICAST_HOPS` option for this socket.
///
/// Specifies the hop limit for ipv6 unicast packets
pub fn unicast_hops_v6(&self) -> io::Result<u32> {
unsafe {
getsockopt::<c_int>(self.as_raw(), sys::IPPROTO_IPV6, sys::IPV6_UNICAST_HOPS)
.map(|hops| hops as u32)
}
}
/// Set the value for the `IPV6_UNICAST_HOPS` option on this socket.
///
/// Specifies the hop limit for ipv6 unicast packets
pub fn set_unicast_hops_v6(&self, hops: u32) -> io::Result<()> {
unsafe {
setsockopt(
self.as_raw(),
sys::IPPROTO_IPV6,
sys::IPV6_UNICAST_HOPS,
hops as c_int,
)
}
}
/// Get the value of the `IPV6_V6ONLY` option for this socket.
///
/// For more information about this option, see [`set_only_v6`].
///
/// [`set_only_v6`]: Socket::set_only_v6
pub fn only_v6(&self) -> io::Result<bool> {
unsafe {
getsockopt::<c_int>(self.as_raw(), sys::IPPROTO_IPV6, sys::IPV6_V6ONLY)
.map(|only_v6| only_v6 != 0)
}
}
/// Set the value for the `IPV6_V6ONLY` option on this socket.
///
/// If this is set to `true` then the socket is restricted to sending and
/// receiving IPv6 packets only. In this case two IPv4 and IPv6 applications
/// can bind the same port at the same time.
///
/// If this is set to `false` then the socket can be used to send and
/// receive packets from an IPv4-mapped IPv6 address.
pub fn set_only_v6(&self, only_v6: bool) -> io::Result<()> {
unsafe {
setsockopt(
self.as_raw(),
sys::IPPROTO_IPV6,
sys::IPV6_V6ONLY,
only_v6 as c_int,
)
}
}
}
/// Socket options for TCP sockets, get/set using `IPPROTO_TCP`.
///
/// Additional documentation can be found in documentation of the OS.
impl Socket {
/// Get the value of the `TCP_KEEPIDLE` option on this socket.
///
/// This returns the value of `TCP_KEEPALIVE` on macOS and iOS and `TCP_KEEPIDLE` on all other
/// supported Unix operating systems.
#[cfg(any(
doc,
all(
feature = "all",
not(any(windows, target_os = "haiku", target_os = "openbsd"))
)
))]
#[cfg_attr(
docsrs,
doc(cfg(all(
feature = "all",
not(any(windows, target_os = "haiku", target_os = "openbsd"))
)))
)]
pub fn keepalive_time(&self) -> io::Result<Duration> {
sys::keepalive_time(self.as_raw())
}
/// Get the value of the `TCP_KEEPINTVL` option on this socket.
///
/// For more information about this option, see [`set_tcp_keepalive`].
///
/// [`set_tcp_keepalive`]: Socket::set_tcp_keepalive
#[cfg(all(
feature = "all",
any(
doc,
target_os = "android",
target_os = "dragonfly",
target_os = "freebsd",
target_os = "fuchsia",
target_os = "illumos",
target_os = "linux",
target_os = "netbsd",
target_vendor = "apple",
)
))]
#[cfg_attr(
docsrs,
doc(cfg(all(
feature = "all",
any(
target_os = "android",
target_os = "dragonfly",
target_os = "freebsd",
target_os = "fuchsia",
target_os = "illumos",
target_os = "linux",
target_os = "netbsd",
target_vendor = "apple",
)
)))
)]
pub fn keepalive_interval(&self) -> io::Result<Duration> {
unsafe {
getsockopt::<c_int>(self.as_raw(), sys::IPPROTO_TCP, sys::TCP_KEEPINTVL)
.map(|secs| Duration::from_secs(secs as u64))
}
}
/// Get the value of the `TCP_KEEPCNT` option on this socket.
///
/// For more information about this option, see [`set_tcp_keepalive`].
///
/// [`set_tcp_keepalive`]: Socket::set_tcp_keepalive
#[cfg(all(
feature = "all",
any(
doc,
target_os = "android",
target_os = "dragonfly",
target_os = "freebsd",
target_os = "fuchsia",
target_os = "illumos",
target_os = "linux",
target_os = "netbsd",
target_vendor = "apple",
)
))]
#[cfg_attr(
docsrs,
doc(cfg(all(
feature = "all",
any(
target_os = "android",
target_os = "dragonfly",
target_os = "freebsd",
target_os = "fuchsia",
target_os = "illumos",
target_os = "linux",
target_os = "netbsd",
target_vendor = "apple",
)
)))
)]
pub fn keepalive_retries(&self) -> io::Result<u32> {
unsafe {
getsockopt::<c_int>(self.as_raw(), sys::IPPROTO_TCP, sys::TCP_KEEPCNT)
.map(|retries| retries as u32)
}
}
/// Set parameters configuring TCP keepalive probes for this socket.
///
/// The supported parameters depend on the operating system, and are
/// configured using the [`TcpKeepalive`] struct. At a minimum, all systems
/// support configuring the [keepalive time]: the time after which the OS
/// will start sending keepalive messages on an idle connection.
///
/// [keepalive time]: TcpKeepalive::with_time
///
/// # Notes
///
/// * This will enable `SO_KEEPALIVE` on this socket, if it is not already
/// enabled.
/// * On some platforms, such as Windows, any keepalive parameters *not*
/// configured by the `TcpKeepalive` struct passed to this function may be
/// overwritten with their default values. Therefore, this function should
/// either only be called once per socket, or the same parameters should
/// be passed every time it is called.
///
/// # Examples
///
/// ```
/// use std::time::Duration;
///
/// use socket2::{Socket, TcpKeepalive, Domain, Type};
///
/// # fn main() -> std::io::Result<()> {
/// let socket = Socket::new(Domain::IPV4, Type::STREAM, None)?;
/// let keepalive = TcpKeepalive::new()
/// .with_time(Duration::from_secs(4));
/// // Depending on the target operating system, we may also be able to
/// // configure the keepalive probe interval and/or the number of
/// // retries here as well.
///
/// socket.set_tcp_keepalive(&keepalive)?;
/// # Ok(()) }
/// ```
///
pub fn set_tcp_keepalive(&self, params: &TcpKeepalive) -> io::Result<()> {
self.set_keepalive(true)?;
sys::set_tcp_keepalive(self.as_raw(), params)
}
/// Get the value of the `TCP_NODELAY` option on this socket.
///
/// For more information about this option, see [`set_nodelay`].
///
/// [`set_nodelay`]: Socket::set_nodelay
pub fn nodelay(&self) -> io::Result<bool> {
unsafe {
getsockopt::<Bool>(self.as_raw(), sys::IPPROTO_TCP, sys::TCP_NODELAY)
.map(|nodelay| nodelay != 0)
}
}
/// Set the value of the `TCP_NODELAY` option on this socket.
///
/// If set, this option disables the Nagle algorithm. This means that
/// segments are always sent as soon as possible, even if there is only a
/// small amount of data. When not set, data is buffered until there is a
/// sufficient amount to send out, thereby avoiding the frequent sending of
/// small packets.
pub fn set_nodelay(&self, nodelay: bool) -> io::Result<()> {
unsafe {
setsockopt(
self.as_raw(),
sys::IPPROTO_TCP,
sys::TCP_NODELAY,
nodelay as c_int,
)
}
}
}
impl Read for Socket {
fn read(&mut self, buf: &mut [u8]) -> io::Result<usize> {
// Safety: the `recv` implementation promises not to write uninitialised
// bytes to the `buf`fer, so this casting is safe.
let buf = unsafe { &mut *(buf as *mut [u8] as *mut [MaybeUninit<u8>]) };
self.recv(buf)
}
#[cfg(not(target_os = "redox"))]
fn read_vectored(&mut self, bufs: &mut [IoSliceMut<'_>]) -> io::Result<usize> {
// Safety: both `IoSliceMut` and `MaybeUninitSlice` promise to have the
// same layout, that of `iovec`/`WSABUF`. Furthermore `recv_vectored`
// promises to not write unitialised bytes to the `bufs` and pass it
// directly to the `recvmsg` system call, so this is safe.
let bufs = unsafe { &mut *(bufs as *mut [IoSliceMut<'_>] as *mut [MaybeUninitSlice<'_>]) };
self.recv_vectored(bufs).map(|(n, _)| n)
}
}
impl<'a> Read for &'a Socket {
fn read(&mut self, buf: &mut [u8]) -> io::Result<usize> {
// Safety: see other `Read::read` impl.
let buf = unsafe { &mut *(buf as *mut [u8] as *mut [MaybeUninit<u8>]) };
self.recv(buf)
}
#[cfg(not(target_os = "redox"))]
fn read_vectored(&mut self, bufs: &mut [IoSliceMut<'_>]) -> io::Result<usize> {
// Safety: see other `Read::read` impl.
let bufs = unsafe { &mut *(bufs as *mut [IoSliceMut<'_>] as *mut [MaybeUninitSlice<'_>]) };
self.recv_vectored(bufs).map(|(n, _)| n)
}
}
impl Write for Socket {
fn write(&mut self, buf: &[u8]) -> io::Result<usize> {
self.send(buf)
}
#[cfg(not(target_os = "redox"))]
fn write_vectored(&mut self, bufs: &[IoSlice<'_>]) -> io::Result<usize> {
self.send_vectored(bufs)
}
fn flush(&mut self) -> io::Result<()> {
Ok(())
}
}
impl<'a> Write for &'a Socket {
fn write(&mut self, buf: &[u8]) -> io::Result<usize> {
self.send(buf)
}
#[cfg(not(target_os = "redox"))]
fn write_vectored(&mut self, bufs: &[IoSlice<'_>]) -> io::Result<usize> {
self.send_vectored(bufs)
}
fn flush(&mut self) -> io::Result<()> {
Ok(())
}
}
impl fmt::Debug for Socket {
fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
f.debug_struct("Socket")
.field("raw", &self.as_raw())
.field("local_addr", &self.local_addr().ok())
.field("peer_addr", &self.peer_addr().ok())
.finish()
}
}
from!(net::TcpStream, Socket);
from!(net::TcpListener, Socket);
from!(net::UdpSocket, Socket);
from!(Socket, net::TcpStream);
from!(Socket, net::TcpListener);
from!(Socket, net::UdpSocket);