Re: [PATCH v4 3/9] rust: file: add Rust abstraction for `struct file`

[Trevor Gross <tmgross@umich.edu>]

On Fri, Feb 2, 2024 at 5:56 AM Alice Ryhl <aliceryhl@google.com> wrote:
>
> From: Wedson Almeida Filho <wedsonaf@gmail.com>
>
> This abstraction makes it possible to manipulate the open files for a
> process. The new `File` struct wraps the C `struct file`. When accessing
> it using the smart pointer `ARef<File>`, the pointer will own a
> reference count to the file. When accessing it as `&File`, then the
> reference does not own a refcount, but the borrow checker will ensure
> that the reference count does not hit zero while the `&File` is live.
>
> Since this is intended to manipulate the open files of a process, we
> introduce an `fget` constructor that corresponds to the C `fget`
> method. In future patches, it will become possible to create a new fd in
> a process and bind it to a `File`. Rust Binder will use these to send
> fds from one process to another.
>
> We also provide a method for accessing the file's flags. Rust Binder
> will use this to access the flags of the Binder fd to check whether the
> non-blocking flag is set, which affects what the Binder ioctl does.
>
> This introduces a struct for the EBADF error type, rather than just
> using the Error type directly. This has two advantages:
> * `File::from_fd` returns a `Result<ARef<File>, BadFdError>`, which the
>   compiler will represent as a single pointer, with null being an error.
>   This is possible because the compiler understands that `BadFdError`
>   has only one possible value, and it also understands that the
>   `ARef<File>` smart pointer is guaranteed non-null.
> * Additionally, we promise to users of the method that the method can
>   only fail with EBADF, which means that they can rely on this promise
>   without having to inspect its implementation.
> That said, there are also two disadvantages:
> * Defining additional error types involves boilerplate.
> * The question mark operator will only utilize the `From` trait once,
>   which prevents you from using the question mark operator on
>   `BadFdError` in methods that return some third error type that the
>   kernel `Error` is convertible into. (However, it works fine in methods
>   that return `Error`.)
>
> Signed-off-by: Wedson Almeida Filho <wedsonaf@gmail.com>
> Co-developed-by: Daniel Xu <dxu@dxuuu.xyz>
> Signed-off-by: Daniel Xu <dxu@dxuuu.xyz>
> Co-developed-by: Alice Ryhl <aliceryhl@google.com>
> Signed-off-by: Alice Ryhl <aliceryhl@google.com>
> ---
>  fs/file.c                       |   7 +
>  rust/bindings/bindings_helper.h |   2 +
>  rust/helpers.c                  |   7 +
>  rust/kernel/file.rs             | 249 ++++++++++++++++++++++++++++++++
>  rust/kernel/lib.rs              |   1 +
>  5 files changed, 266 insertions(+)
>  create mode 100644 rust/kernel/file.rs
>
> diff --git a/fs/file.c b/fs/file.c
> index 3b683b9101d8..f2eab5fcb87f 100644
> --- a/fs/file.c
> +++ b/fs/file.c
> @@ -1115,18 +1115,25 @@ EXPORT_SYMBOL(task_lookup_next_fdget_rcu);
>  /*
>   * Lightweight file lookup - no refcnt increment if fd table isn't shared.
>   *
>   * You can use this instead of fget if you satisfy all of the following
>   * conditions:
>   * 1) You must call fput_light before exiting the syscall and returning control
>   *    to userspace (i.e. you cannot remember the returned struct file * after
>   *    returning to userspace).
>   * 2) You must not call filp_close on the returned struct file * in between
>   *    calls to fget_light and fput_light.
>   * 3) You must not clone the current task in between the calls to fget_light
>   *    and fput_light.
>   *
>   * The fput_needed flag returned by fget_light should be passed to the
>   * corresponding fput_light.
> + *
> + * (As an exception to rule 2, you can call filp_close between fget_light and
> + * fput_light provided that you capture a real refcount with get_file before
> + * the call to filp_close, and ensure that this real refcount is fput *after*
> + * the fput_light call.)
> + *
> + * See also the documentation in rust/kernel/file.rs.
>   */
>  static unsigned long __fget_light(unsigned int fd, fmode_t mask)
>  {

Should this be split to its own patch so it can be applied separately if needed?

> [...]
> +    /// Also known as `O_NDELAY`.
> +    ///
> +    /// This is effectively the same flag as [`O_NONBLOCK`] on all architectures
> +    /// except SPARC64.
> +    pub const O_NDELAY: u32 = bindings::O_NDELAY;

This is O_NDELAY, should the AKA say O_NONBLOCK?
> [...]
> +/// Wraps the kernel's `struct file`.

It is probably better to say what it does for the summary, and mention
what it wraps later.

> +/// # Refcounting
> +///
> +/// Instances of this type are reference-counted. The reference count is incremented by the
> +/// `fget`/`get_file` functions and decremented by `fput`. The Rust type `ARef<File>` represents a
> +/// pointer that owns a reference count on the file.
> +///
> +/// Whenever a process opens a file descriptor (fd), it stores a pointer to the file in its `struct
> +/// files_struct`. This pointer owns a reference count to the file, ensuring the file isn't
> +/// prematurely deleted while the file descriptor is open. In Rust terminology, the pointers in
> +/// `struct files_struct` are `ARef<File>` pointers.
> +///
> +/// ## Light refcounts
> +///
> +/// Whenever a process has an fd to a file, it may use something called a "light refcount" as a
> +/// performance optimization. Light refcounts are acquired by calling `fdget` and released with
> +/// `fdput`. The idea behind light refcounts is that if the fd is not closed between the calls to
> +/// `fdget` and `fdput`, then the refcount cannot hit zero during that time, as the `struct
> +/// files_struct` holds a reference until the fd is closed. This means that it's safe to access the
> +/// file even if `fdget` does not increment the refcount.
> +///
> +/// The requirement that the fd is not closed during a light refcount applies globally across all
> +/// threads - not just on the thread using the light refcount. For this reason, light refcounts are
> +/// only used when the `struct files_struct` is not shared with other threads, since this ensures
> +/// that other unrelated threads cannot suddenly start using the fd and close it. Therefore,
> +/// calling `fdget` on a shared `struct files_struct` creates a normal refcount instead of a light
> +/// refcount.
> +///
> +/// Light reference counts must be released with `fdput` before the system call returns to
> +/// userspace. This means that if you wait until the current system call returns to userspace, then
> +/// all light refcounts that existed at the time have gone away.
> +///
> +/// ## Rust references
> +///
> +/// The reference type `&File` is similar to light refcounts:
> +///
> +/// * `&File` references don't own a reference count. They can only exist as long as the reference
> +///   count stays positive, and can only be created when there is some mechanism in place to ensure
> +///   this.
> +///
> +/// * The Rust borrow-checker normally ensures this by enforcing that the `ARef<File>` from which
> +///   a `&File` is created outlives the `&File`.
> +///
> +/// * Using the unsafe [`File::from_ptr`] means that it is up to the caller to ensure that the
> +///   `&File` only exists while the reference count is positive.
> +///
> +/// * You can think of `fdget` as using an fd to look up an `ARef<File>` in the `struct
> +///   files_struct` and create an `&File` from it. The "fd cannot be closed" rule is like the Rust
> +///   rule "the `ARef<File>` must outlive the `&File`".
> +///
> +/// # Invariants
> +///
> +/// * Instances of this type are refcounted using the `f_count` field.
> +/// * If an fd with active light refcounts is closed, then it must be the case that the file
> +///   refcount is positive until all light refcounts of the fd have been dropped.
> +/// * A light refcount must be dropped before returning to userspace.
> +#[repr(transparent)]
> +pub struct File(Opaque<bindings::file>);
> +
> +// SAFETY:
> +// - `File::dec_ref` can be called from any thread.
> +// - It is okay to send ownership of `File` across thread boundaries.

Shouldn't this be lowecase `file` because it is referring to the
underlying C object?

> +unsafe impl Send for File {}
> [...]
> +    /// Returns the flags associated with the file.
> +    ///
> +    /// The flags are a combination of the constants in [`flags`].
> +    pub fn flags(&self) -> u32 {

A typedef used here and in the constants module could be useful

    type FileFlags = u32;

> +        // This `read_volatile` is intended to correspond to a READ_ONCE call.
> +        //
> +        // SAFETY: The file is valid because the shared reference guarantees a nonzero refcount.
> +        //
> +        // TODO: Replace with `read_once` when available on the Rust side.

Shouldn't the TODO become a `FIXME(read_once): ...` since it is going
into the codebase?

> +        unsafe { core::ptr::addr_of!((*self.as_ptr()).f_flags).read_volatile() }
> +    }
> +}

Some suggestions but nothing blocking

Reviewed-by: Trevor Gross <tmgross@umich.edu>