Re: [PATCH] rust: drm: fix unsound initialization in drm::Device::new

[Alice Ryhl <aliceryhl@google.com>]

On Tue, Apr 28, 2026 at 09:20:21PM +0900, Eliot Courtney wrote:
> If pinned initialization of drm::Device::Data fails, it calls
> drm::Device::release via drm_dev_put. This materializes a reference to
> &drm::Device, but it's not fully constructed yet, because initializing
> `data` failed. It should not be dropped either. Instead, if pinned
> initialization fails, make sure drm::Device::release isn't called.
> 
> Fixes: 2e9fdbe5ec7a ("rust: drm: device: drop_in_place() the drm::Device in release()")
> Signed-off-by: Eliot Courtney <ecourtney@nvidia.com>

For the concerns about duplicating vtables, could the temporary vtable
be a stack variable?

>  rust/kernel/drm/device.rs | 22 ++++++++++++++++------
>  1 file changed, 16 insertions(+), 6 deletions(-)
> 
> diff --git a/rust/kernel/drm/device.rs b/rust/kernel/drm/device.rs
> index adbafe8db54d..78ea0eb12535 100644
> --- a/rust/kernel/drm/device.rs
> +++ b/rust/kernel/drm/device.rs
> @@ -111,6 +111,13 @@ impl<T: drm::Driver> Device<T> {
>          fops: &Self::GEM_FOPS,
>      };
>  
> +    // Use a vtable without a `release` callback until `data` is initialized, so init failure
> +    // can release the DRM device without dropping uninitialized fields.
> +    const ALLOC_VTABLE: bindings::drm_driver = bindings::drm_driver {
> +        release: None,
> +        ..Self::VTABLE
> +    };
> +
>      const GEM_FOPS: bindings::file_operations = drm::gem::create_fops();
>  
>      /// Create a new `drm::Device` for a `drm::Driver`.
> @@ -120,12 +127,12 @@ pub fn new(dev: &device::Device, data: impl PinInit<T::Data, Error>) -> Result<A
>          let layout = Kmalloc::aligned_layout(Layout::new::<Self>());
>  
>          // SAFETY:
> -        // - `VTABLE`, as a `const` is pinned to the read-only section of the compilation,
> +        // - `ALLOC_VTABLE`, as a `const` is pinned to the read-only section of the compilation,
>          // - `dev` is valid by its type invarants,
>          let raw_drm: *mut Self = unsafe {
>              bindings::__drm_dev_alloc(
>                  dev.as_raw(),
> -                &Self::VTABLE,
> +                &Self::ALLOC_VTABLE,
>                  layout.size(),
>                  mem::offset_of!(Self, dev),
>              )
> @@ -133,6 +140,10 @@ pub fn new(dev: &device::Device, data: impl PinInit<T::Data, Error>) -> Result<A
>          .cast();
>          let raw_drm = NonNull::new(from_err_ptr(raw_drm)?).ok_or(ENOMEM)?;
>  
> +        // SAFETY: `raw_drm` is a valid pointer to `Self`, given that `__drm_dev_alloc` was
> +        // successful.
> +        let drm_dev = unsafe { Self::into_drm_device(raw_drm) };
> +
>          // SAFETY: `raw_drm` is a valid pointer to `Self`.
>          let raw_data = unsafe { ptr::addr_of_mut!((*raw_drm.as_ptr()).data) };
>  
> @@ -140,15 +151,14 @@ pub fn new(dev: &device::Device, data: impl PinInit<T::Data, Error>) -> Result<A
>          // - `raw_data` is a valid pointer to uninitialized memory.
>          // - `raw_data` will not move until it is dropped.
>          unsafe { data.__pinned_init(raw_data) }.inspect_err(|_| {
> -            // SAFETY: `raw_drm` is a valid pointer to `Self`, given that `__drm_dev_alloc` was
> -            // successful.
> -            let drm_dev = unsafe { Self::into_drm_device(raw_drm) };
> -
>              // SAFETY: `__drm_dev_alloc()` was successful, hence `drm_dev` must be valid and the
>              // refcount must be non-zero.
>              unsafe { bindings::drm_dev_put(drm_dev) };
>          })?;
>  
> +        // SAFETY: `drm_dev` is still private to this function.
> +        unsafe { (*drm_dev).driver = &Self::VTABLE };

It would be bad if this ended up being a reference to a local variable.
Please use `&const { Self::VTABLE }` so that it doesn't compile if this
occurs.

Alice