Re: [PATCH v2] uaccess: decouple INLINE_COPY_FROM_USER and CONFIG_RUST

[Andrew Morton]

On Tue, 18 Nov 2025 10:39:56 +0000 Alice Ryhl <aliceryhl@xxxxxxxxxx> wrote:

> On Fri, Oct 24, 2025 at 11:47:53AM -0400, Yury Norov (NVIDIA) wrote:
> > Commit 1f9a8286bc0c ("uaccess: always export _copy_[from|to]_user with
> > CONFIG_RUST") exports _copy_{from,to}_user() unconditionally, if RUST
> > is enabled. This pollutes exported symbols namespace, and spreads RUST
> > ifdefery in core files.
> > 
> > It's better to declare a corresponding helper under the rust/helpers,
> > similarly to how non-underscored copy_{from,to}_user() is handled.
> > 
> > Reviewed-by: Alice Ryhl <aliceryhl@xxxxxxxxxx>
> > Tested-by: Alice Ryhl <aliceryhl@xxxxxxxxxx>
> > Signed-off-by: Yury Norov (NVIDIA) <yury.norov@xxxxxxxxx>
> 
> It looks like this is not quite correct. The header file still has this
> declaration unconditionally:
> 
> extern __must_check unsigned long
> _copy_from_user(void *, const void __user *, unsigned long);
> 
> extern __must_check unsigned long
> _copy_to_user(void __user *, const void *, unsigned long);
> 
> this causes:
> 
> ERROR: modpost: "_copy_from_user" [samples/rust/rust_misc_device.ko] undefined!
> ERROR: modpost: "_copy_to_user" [samples/rust/rust_misc_device.ko] undefined!

I wonder why this wasn't caught in linux-next testing.

Please check the rust Kconfig defaults, see if there's something we can
do to get more compilation test coverage?

> This is because when Rust sees both a helper and a non-helper of the
> same function, it prefers to call the non-helper version. In this case,
> it saw a declaration of the function in the header file, and so it tried
> to call that instead of the helper.
> 
> To fix this, we need to wrap the above declarations in:
> 
> 	#ifndef INLINE_COPY_FROM_USER

It's in mainline now - do you have time to send along a patch?