refactor: I/O safety for unistd.rs

[SteveLauC]

What does this PR do

Add I/O safety to module unistd.

File descriptor duplication

The safe interface for dup2() proposed in this PR looks like this:

pub fn dup2<Fd: AsFd>(oldfd: Fd, newfd: &mut OwnedFd) -> Result<()>

We take newfd by mutable reference so that we can ensure we have exclusive access. This interface stops users from specifying newfd with arbitrary fd values, unsafe fn dup2_raw() comes as a remedy:

pub unsafe fn dup2_raw<Fd1: AsFd, Fd2: IntoRawFd>(
    oldfd: Fd1,
    newfd: Fd2,
) -> Result<OwnedFd>

dup2_raw() is unsafe because when specifying a fd that is open, one has to ensure the returned OwnedFd is the ONLY owner of this fd, or a double close will happen.

Using the above dup2() interface for stdin/stdout/stderr redirection is troublesome, one has to construct an OwnedFd from stdin/stdout/stderr, then mem::forget() it. We provide 3 helper functions to make this easier:

pub fn dup2_stdin<Fd: AsFd>(fd: Fd) -> Result<()>
pub fn dup2_stdout<Fd: AsFd>(fd: Fd) -> Result<()>
pub fn dup2_stderr<Fd: AsFd>(fd: Fd) -> Result<()>

Since the only flag that can be used with dup3() is O_CLOEXEC, which does not make much sense to stdin/stdout/stderr, I didn't provide the dup3 versions of these utilities.

Checklist:

• I have read CONTRIBUTING.md
• I have written necessary tests and rustdoc comments
• A change log has been added if this PR modifies nix's API

[SteveLauC]

This can be removed when we add I/O safety to module fanotify.

[SteveLauC]

This Read implementation (and Write) feels a little bit weird to me, I kinda think we should not implement std Read/Write traits for Nix wrapper types, the I/O syscalls should be used instead.

But this implementation would definitely make our interface more rusty, so I am not quite against it, though I think we should be consistent, i.e., if one wrapper type has this trait implemented, we should do this to all the wrapper types.

[SteveLauC]

Not sure if hard link is supported by RedoxFS, but linkat() is for hard link rather than symlink