Re: [PATCH 5/8] IB/hfi1: make hfi1_write_iter() deal with ITER_UBUF iov_iter

[Matthew Wilcox <willy@xxxxxxxxxxxxx>]

On Tue, Mar 28, 2023 at 11:43:34AM -0700, Linus Torvalds wrote:
>         -       size_t count;
>         -       union {
>         -               const struct iovec *iov;
>         -               const struct kvec *kvec;
>         -               const struct bio_vec *bvec;
>         -               struct xarray *xarray;
>         -               struct pipe_inode_info *pipe;
>         -               void __user *ubuf;
>         +
>         +       /*
>         +        * This has the same layout as 'struct iovec'!
>         +        * In particular, the ITER_UBUF form can create
>         +        * a single-entry 'struct iovec' by casting the
>         +        * address of the 'ubuf' member to that.
>         +        */
>         +       struct {
>         +               union {
>         +                       const struct iovec *iov;
>         +                       const struct kvec *kvec;
>         +                       const struct bio_vec *bvec;
>         +                       struct xarray *xarray;
>         +                       struct pipe_inode_info *pipe;
>         +                       void __user *ubuf;
>         +               };
>         +               size_t count;
>                 };
>                 union {
>                         unsigned long nr_segs;
> 
> and if you accept the above, then you can do
> 
>    #define iter_ubuf_to_iov(iter) ((const struct iovec *)&(iter)->ubuf)
> 
> which I will admit is not *pretty*, but it's kind of clever, I think.

I think it'll annoy gcc, and particularly the randstruct plugin.
How about:

	union {
		struct iovec ubuf;
		struct {
			const struct iovec *iov;
			size_t count; /* Also valid for subsequent types */
		};
		const struct kvec *kvec;
		const struct bio_vec *bvec;
		struct xarray *xarray;
		struct pipe_inode_info *pipe;
	}