Implemented io-uring Op<Statx> and applied to read_uring and fs::try_exists

[asder8215]

Motivation

This builds up on vrtgs PR in their attempt of implementing Statx for io-uring and apply it to fs::try_exists. I used libc instead of linux-raw-sys for the Statx implementation, tried to preserve more of how the original code (prior to vrtgs' changes) looked in certain files. There are some inlining done for the metadata functions for Statx because I thought the body of the code in these functions were pretty small and didn't seem to involve generics (let me know if I should remove that). Some other review comments mentioned in review of vrtgs' PR is addressed such as safety comments, inlining for the try_exists_uring, avoiding using CString literals, etc.

To note, this is my first time contributing to Tokio, so let me know if there's anything that I'm doing incorrectly or if I missed something as per procedure here. Let me know if I should also squash some commits as well.

ccing @Daksh14 since I got to working on this PR per him sharing vrtgs' PR with me.

[asder8215]

Any reason why libc statx constants are not found for Linux 7.0?

[Darksonn]

The compilation error is probably related to whether the build is using gnu or musl libc. See rust-lang/rust#154981 with similar issue.

[martin-g]

Does something guarantee that the File won't be dropped during the io uring call ?
It this happens the kernel will try to use a closed fd or worse - a reused fd.
IMO it should use an OwnedFd here.

[asder8215]

I might need a bit more clarity to understand what you're seeing.

Op::file_metadata doesn't own File, it takes a reference to it. Wouldn't Rust's borrow checker prohibit this file.as_raw_fd() from being invalid since &File has to live for as long as the caller's File?

Moreover, I would assume that the statx syscall populates the buffer with the file metadata once, and doesn't need access to the fd/path after invocation (correct me if I'm wrong though)?