Re: [PATCH 09/22] rust: pin-init: move impl `Zeroable` for `Opaque` and `Option<KBox<T>>` into the kernel crate

[Benno Lossin]

On Wed Mar 5, 2025 at 1:11 PM CET, Alice Ryhl wrote:
> On Wed, Mar 5, 2025 at 1:05 PM Benno Lossin <benno.lossin@xxxxxxxxx> wrote:
>>
>> On Wed Mar 5, 2025 at 12:26 PM CET, Andreas Hindborg wrote:
>> > "Benno Lossin" <benno.lossin@xxxxxxxxx> writes:
>> >
>> >> In order to make pin-init a standalone crate, move kernel-specific code
>> >> directly into the kernel crate. Since `Opaque<T>` and `KBox<T>` are part
>> >> of the kernel, move their `Zeroable` implementation into the kernel
>> >> crate.
>> >>
>> >> Signed-off-by: Benno Lossin <benno.lossin@xxxxxxxxx>
>> >> ---
>> >>  rust/kernel/alloc/kbox.rs | 8 +++++++-
>> >>  rust/kernel/types.rs      | 5 ++++-
>> >>  rust/pin-init/src/lib.rs  | 8 +-------
>> >>  3 files changed, 12 insertions(+), 9 deletions(-)
>> >>
>> >> diff --git a/rust/kernel/alloc/kbox.rs b/rust/kernel/alloc/kbox.rs
>> >> index 39a3ea7542da..9861433559dc 100644
>> >> --- a/rust/kernel/alloc/kbox.rs
>> >> +++ b/rust/kernel/alloc/kbox.rs
>> >> @@ -15,7 +15,7 @@
>> >>  use core::ptr::NonNull;
>> >>  use core::result::Result;
>> >>
>> >> -use crate::init::{InPlaceWrite, Init, PinInit};
>> >> +use crate::init::{InPlaceWrite, Init, PinInit, Zeroable};
>> >>  use crate::init_ext::InPlaceInit;
>> >>  use crate::types::ForeignOwnable;
>> >>
>> >> @@ -100,6 +100,12 @@
>> >>  /// ```
>> >>  pub type KVBox<T> = Box<T, super::allocator::KVmalloc>;
>> >>
>> >> +// SAFETY: All zeros is equivalent to `None` (option layout optimization guarantee).
>> >> +//
>> >> +// In this case we are allowed to use `T: ?Sized`, since all zeros is the `None` variant and there
>> >> +// is no problem with a VTABLE pointer being null.
>> >> +unsafe impl<T: ?Sized, A: Allocator> Zeroable for Option<Box<T, A>> {}
>> >
>> > Could you elaborate the statement related to vtable pointers? How does
>> > that come into play for `Option<Box<_>>`? Is it for fat pointers to
>> > trait objects?
>>
>> Yes it is for fat pointers, if you have a `x: *mut dyn Trait`, then you
>> aren't allowed to write all zeroes to `x`, because the VTABLE pointer
>> (that is part of the fat pointer) is not allowed to be null.
>>
>> Now for `Option<Box<_>>`, this doesn't matter, as there if the normal
>> pointer part of the fat pointer is all zeroes, then the VTABLE pointer
>> part is considered padding bytes, as it's the `None` variant.
>
> The standard library only guarantees that all zeros is valid for
> Option<Box<T,A>> when T:Sized and A=Global.
> https://doc.rust-lang.org/stable/std/option/index.html#representation

Oh! That's a problem then... I'll remove that then (and I can also get
rid of the `ZeroableOption` trait).

We should also backport (& fix it in mainline), I'll submit a patch
shortly.

---
Cheers,
Benno