Re: [PATCH] rust: mm: Mark VmaNew as transparent

[Alice Ryhl]

On Thu, Aug 21, 2025 at 1:29 AM Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx> wrote:
>
> On Tue, 12 Aug 2025 15:26:56 +0200 Baptiste Lepers <baptiste.lepers@xxxxxxxxx> wrote:
>
> > Unsafe code in VmaNew's methods assumes that the type has the same
> > layout as the inner `bindings::vm_area_struct`. This is not guaranteed by
> > the default struct representation in Rust, but requires specifying the
> > `transparent` representation.
> >
> > ...
> >
> > +++ b/rust/kernel/mm/virt.rs
> > @@ -209,6 +209,7 @@ pub fn vm_insert_page(&self, address: usize, page: &Page) -> Result {
> >  ///
> >  /// For the duration of 'a, the referenced vma must be undergoing initialization in an
> >  /// `f_ops->mmap()` hook.
> > +#[repr(transparent)]
> >  pub struct VmaNew {
> >      vma: VmaRef,
> >  }
>
> Alice suggests that I add a cc:stable to this.  But I see nothing in
> the changelog which explains why we're proposing a backport.
>
> So please send us a description of the userspace-visible runtime
> impact of this flaw and I'll paste it into the changelog, thanks.

I don't think it has any userspace-visible runtime impact. But I've
seen many things get backported when they are incorrect even if it
works in practice, so that is why I suggested to backport it anyway.

The annotation makes it so that VmaNew is guaranteed to have the same
layout and ABI as struct vm_area_struct, which is required for
correctness. Without the annotation, rustc doesn't *guarantee* that
the layout/ABI is identical, but in this case, they are identical in
practice even if the annotation is missing.

Alice