Crate rustix

rustix provides efficient memory-safe and I/O-safe wrappers to POSIX-like, Unix-like, Linux, and Winsock syscall-like APIs, with configurable backends.

With rustix, you can write code like this:

# #[cfg(feature = "net")]
# fn read(sock: std::net::TcpStream, buf: &mut [u8]) -> std::io::Result<()> {
# use rustix::net::RecvFlags;
let (nread, _received) = rustix::net::recv(&sock, buf, RecvFlags::PEEK)?;
# let _ = nread;
# Ok(())
# }

instead of like this:

# #[cfg(feature = "net")]
# fn read(sock: std::net::TcpStream, buf: &mut [u8]) -> std::io::Result<()> {
# #[cfg(unix)]
# use std::os::unix::io::AsRawFd;
# #[cfg(target_os = "wasi")]
# use std::os::wasi::io::AsRawFd;
# #[cfg(windows)]
# use windows_sys::Win32::Networking::WinSock as libc;
# #[cfg(windows)]
# use std::os::windows::io::AsRawSocket;
# const MSG_PEEK: i32 = libc::MSG_PEEK;
let nread = unsafe {
    #[cfg(any(unix, target_os = "wasi"))]
    let raw = sock.as_raw_fd();
    #[cfg(windows)]
    let raw = sock.as_raw_socket();
    match libc::recv(
        raw as _,
        buf.as_mut_ptr().cast(),
        buf.len().try_into().unwrap_or(i32::MAX as _),
        MSG_PEEK,
    ) {
        -1 => return Err(std::io::Error::last_os_error()),
        nread => nread as usize,
    }
};
# let _ = nread;
# Ok(())
# }

rustix's APIs perform the following tasks:

• Error values are translated to Results.
• Buffers are passed as Rust slices.
• Out-parameters are presented as return values.
• Path arguments use Arg, so they accept any string type.
• File descriptors are passed and returned via AsFd and OwnedFd instead of bare integers, ensuring I/O safety.
• Constants use enums and bitflags types, and enable support for externally defined flags.
• Multiplexed functions (eg. fcntl, ioctl, etc.) are de-multiplexed.
• Variadic functions (eg. openat, etc.) are presented as non-variadic.
• Functions that return strings automatically allocate sufficient memory and retry the syscall as needed to determine the needed length.
• Functions and types which need l prefixes or 64 suffixes to enable large-file support (LFS) are used automatically. File sizes and offsets are always presented as u64 and i64.
• Behaviors that depend on the sizes of C types like long are hidden.
• In some places, more human-friendly and less historical-accident names are used (and documentation aliases are used so that the original names can still be searched for).
• Provide y2038 compatibility, on platforms which support this.
• Correct selected platform bugs, such as behavioral differences when running under seccomp.
• Use timespec for timestamps and timeouts instead of timeval and c_int milliseconds.

Things they don't do include:

• Detecting whether functions are supported at runtime, except in specific cases where new interfaces need to be detected to support y2038 and LFS.
• Hiding significant differences between platforms.
• Restricting ambient authorities.
• Imposing sandboxing features such as filesystem path or network address sandboxing.

See cap-std, system-interface, and io-streams for libraries which do hide significant differences between platforms, and cap-std which does perform sandboxing and restricts ambient authorities.