Re: [PATCH v3] fs: Replace user_access_{begin/end} by scoped user access

[David Laight]

On Wed, 18 Mar 2026 08:53:52 -0700
Linus Torvalds <torvalds@xxxxxxxxxxxxxxxxxxxx> wrote:

> On Wed, 18 Mar 2026 at 08:49, Linus Torvalds
> <torvalds@xxxxxxxxxxxxxxxxxxxx> wrote:
> >
> >     #define dirent_size(dirent, len) offsetof(typeof(dirent), d_name[len])  
> 
> That 'typeof(dirent)' needs to be 'typeof(*(dirent))' to be convenient.
> 
> It was correct in the patch I attached, but I'll just point it out anyway.
> 
> And we actually have a helper macro for that: struct_offset(). Which
> wasn't what I used in that attached patch, but *should* have been.
> 
> IOW, the macro should look something like
> 
>     #define dirent_size(dirent, len) struct_offset(dirent, d_name[len])
> 
> instead. That looks fairly clean, no?

And unnecessary - the struct_offset() isn't that much longer.
Actually, from a readability point of view even struct_offset() almost
makes things worse.
If you are being paranoid it is another definition to find and check.

There isn't that much difference in the lengths:
	dirent_size(dirent, len)
	struct_offset(dirent, d_name[len])
	offsetof(typeof(*dirent), d_name[len])
and the last one really does tell you what it being calculated.

I do remember one compiler (and I thought it was gcc, but it might only
have been msvc) that required that offsetof() always generate a constant.
ISTR the C language requires that - which makes all the above invalid. 

> 
>                 Linus