[v3,2/9] rust: sync: add `Arc::{from_raw, into_raw}`

  From: Wedson Almeida Filho <walmeida@microsoft.com>

These methods can be used to turn an `Arc` into a raw pointer and back,
in a way that preserves the metadata for fat pointers.

This is done using the unstable ptr_metadata feature [1]. However, it
could also be done using the unstable pointer_byte_offsets feature [2],
which is likely to have a shorter path to stabilization than
ptr_metadata.

Link: https://github.com/rust-lang/rust/issues/81513 [1]
Link: https://github.com/rust-lang/rust/issues/96283 [2]
Signed-off-by: Wedson Almeida Filho <walmeida@microsoft.com>
Co-developed-by: Alice Ryhl <aliceryhl@google.com>
Signed-off-by: Alice Ryhl <aliceryhl@google.com>
Reviewed-by: Gary Guo <gary@garyguo.net>
---
v2 -> v3:
 * Use `transmute_copy` to convert metadata.
 * Use `add` instead of `wrapping_add`.
 * Dropped Reviewed-by from Martin and Benno due to changes.

 rust/kernel/lib.rs      |  1 +
 rust/kernel/sync/arc.rs | 49 +++++++++++++++++++++++++++++++++++++++--
 2 files changed, 48 insertions(+), 2 deletions(-)
  

On 7/11/23 06:32, Alice Ryhl wrote:
> From: Wedson Almeida Filho <walmeida@microsoft.com>
> 
> These methods can be used to turn an `Arc` into a raw pointer and back,
> in a way that preserves the metadata for fat pointers.
> 
> This is done using the unstable ptr_metadata feature [1]. However, it
> could also be done using the unstable pointer_byte_offsets feature [2],
> which is likely to have a shorter path to stabilization than
> ptr_metadata.
> 
> Link: https://github.com/rust-lang/rust/issues/81513 [1]
> Link: https://github.com/rust-lang/rust/issues/96283 [2]
> Signed-off-by: Wedson Almeida Filho <walmeida@microsoft.com>
> Co-developed-by: Alice Ryhl <aliceryhl@google.com>
> Signed-off-by: Alice Ryhl <aliceryhl@google.com>
> Reviewed-by: Gary Guo <gary@garyguo.net>
> ---
> [...]

Reviewed-by: Martin Rodriguez Reboredo <yakoyoku@gmail.com>
  

> From: Wedson Almeida Filho <walmeida@microsoft.com>
> 
> These methods can be used to turn an `Arc` into a raw pointer and back,
> in a way that preserves the metadata for fat pointers.
> 
> This is done using the unstable ptr_metadata feature [1]. However, it
> could also be done using the unstable pointer_byte_offsets feature [2],
> which is likely to have a shorter path to stabilization than
> ptr_metadata.
> 
> Link: https://github.com/rust-lang/rust/issues/81513 [1]
> Link: https://github.com/rust-lang/rust/issues/96283 [2]
> Signed-off-by: Wedson Almeida Filho <walmeida@microsoft.com>
> Co-developed-by: Alice Ryhl <aliceryhl@google.com>
> Signed-off-by: Alice Ryhl <aliceryhl@google.com>
> Reviewed-by: Gary Guo <gary@garyguo.net>

LGTM, two small documentation nits below, with them fixed you can add:

Reviewed-by: Benno Lossin <benno.lossin@proton.me>

> ---
> v2 -> v3:
>  * Use `transmute_copy` to convert metadata.
>  * Use `add` instead of `wrapping_add`.
>  * Dropped Reviewed-by from Martin and Benno due to changes.
> 
>  rust/kernel/lib.rs      |  1 +
>  rust/kernel/sync/arc.rs | 49 +++++++++++++++++++++++++++++++++++++++--
>  2 files changed, 48 insertions(+), 2 deletions(-)
> 
> diff --git a/rust/kernel/lib.rs b/rust/kernel/lib.rs
> index 788914fd935f..6963d11092c4 100644
> --- a/rust/kernel/lib.rs
> +++ b/rust/kernel/lib.rs
> @@ -17,6 +17,7 @@
>  #![feature(const_refs_to_cell)]
>  #![feature(dispatch_from_dyn)]
>  #![feature(new_uninit)]
> +#![feature(ptr_metadata)]
>  #![feature(receiver_trait)]
>  #![feature(unsize)]
> 
> diff --git a/rust/kernel/sync/arc.rs b/rust/kernel/sync/arc.rs
> index a89843cacaad..4046b534e53b 100644
> --- a/rust/kernel/sync/arc.rs
> +++ b/rust/kernel/sync/arc.rs
> @@ -24,13 +24,13 @@
>  };
>  use alloc::boxed::Box;
>  use core::{
> -    alloc::AllocError,
> +    alloc::{AllocError, Layout},
>      fmt,
>      marker::{PhantomData, Unsize},
>      mem::{ManuallyDrop, MaybeUninit},
>      ops::{Deref, DerefMut},
>      pin::Pin,
> -    ptr::NonNull,
> +    ptr::{NonNull, Pointee},
>  };
>  use macros::pin_data;
> 
> @@ -212,6 +212,51 @@ unsafe fn from_inner(inner: NonNull<ArcInner<T>>) -> Self {
>          }
>      }
> 
> +    /// Convert the [`Arc`] into a raw pointer.
> +    ///
> +    /// The raw pointer has ownership of the refcount that this Arc object owned.
> +    pub fn into_raw(self) -> *const T {
> +        let ptr = self.ptr.as_ptr();
> +        core::mem::forget(self);
> +        // SAFETY: The pointer is valid.
> +        unsafe { core::ptr::addr_of!((*ptr).data) }
> +    }
> +
> +    /// Recreates an [`Arc`] instance previously deconstructed via [`Arc::into_raw`].
> +    ///
> +    /// This code relies on the `repr(C)` layout of structs as described in
> +    /// <https://doc.rust-lang.org/reference/type-layout.html#reprc-structs>.

Why is this in the documentation? I feel like it should be a normal code
comment at the very start of the function.

> +    ///
> +    /// # Safety
> +    ///
> +    /// `ptr` must have been returned by a previous call to [`Arc::into_raw`]. Additionally, it
> +    /// can only be called once for each previous call to [`Arc::into_raw`].

"it can only" -> "it must only"

--
Cheers,
Benno

> +    pub unsafe fn from_raw(ptr: *const T) -> Self {
> +        let refcount_layout = Layout::new::<bindings::refcount_t>();
> +        // SAFETY: The caller guarantees that the pointer is valid.
> +        let val_layout = Layout::for_value(unsafe { &*ptr });
> +        // SAFETY: We're computing the layout of a real struct that existed when compiling this
> +        // binary, so its layout is not so large that it can trigger arithmetic overflow.
> +        let val_offset = unsafe { refcount_layout.extend(val_layout).unwrap_unchecked().1 };
> +
> +        let metadata: <T as Pointee>::Metadata = core::ptr::metadata(ptr);
> +        // SAFETY: The metadata of `T` and `ArcInner<T>` is the same because `ArcInner` is a struct
> +        // with `T` as its last field.
> +        //
> +        // This is documented at:
> +        // <https://doc.rust-lang.org/std/ptr/trait.Pointee.html>.
> +        let metadata: <ArcInner<T> as Pointee>::Metadata =
> +            unsafe { core::mem::transmute_copy(&metadata) };
> +        // SAFETY: The pointer is in-bounds of an allocation both before and after offsetting the
> +        // pointer, since it originates from a previous call to `Arc::into_raw` and is still valid.
> +        let ptr = unsafe { (ptr as *mut u8).sub(val_offset) as *mut () };
> +        let ptr = core::ptr::from_raw_parts_mut(ptr, metadata);
> +
> +        // SAFETY: By the safety requirements we know that `ptr` came from `Arc::into_raw`, so the
> +        // reference count held then will be owned by the new `Arc` object.
> +        unsafe { Self::from_inner(NonNull::new_unchecked(ptr)) }
> +    }
> +
>      /// Returns an [`ArcBorrow`] from the given [`Arc`].
>      ///
>      /// This is useful when the argument of a function call is an [`ArcBorrow`] (e.g., in a method
> --
> 2.41.0.255.g8b1d071c50-goog
>
  

Benno Lossin <benno.lossin@proton.me> writes:
>> +    /// This code relies on the `repr(C)` layout of structs as described in
>> +    /// <https://doc.rust-lang.org/reference/type-layout.html#reprc-structs>.
> 
> Why is this in the documentation? I feel like it should be a normal code
> comment at the very start of the function.

In fact, I think we can drop this comment entirely. The motivation
behind using `Layout::extend` for computing `val_offset` is that its
correctness does not rely on how the repr(C) layout algorithm works.

(As opposed to how the previous implementation's correctness *does*
depend on knowing the repr(C) layout algorithm:
    Layout::new::<ArcInner<()>>().align_to(align).unwrap().pad_to_align().size()
)

>> +    ///
>> +    /// # Safety
>> +    ///
>> +    /// `ptr` must have been returned by a previous call to [`Arc::into_raw`]. Additionally, it
>> +    /// can only be called once for each previous call to [`Arc::into_raw`].
> 
> "it can only" -> "it must only"

Sounds good. I'll change it to use "must" in the next version.

Alice