> > I don't like assigning 'priv' memebers twice, so I'd like to keep it as
> > is.
> 
> But this will give better understanding of the steps the code performs, no?

IMO the current version is readable :)

> (Because this function basically contains two steps at once. I assume it's
>  done this way due to absence of vrealloc(), right?)
> 
> But we have kvrealloc(). Can it be used here?

This has a huge disadvantage: realloc() requests new memory before the
old one is discarded. If you want to make a huge buffer bigger, this can
lead to OOM. The old contents do not matter, so the old buffer can go
away in favor of the new one.

> > > Can it be wrapped by DEFINE_SHOW_ATTRIBUTE()?
> > 
> > I don't see a way. Do you?
> 
> Me neither. I mixed this up with (not upstreamed yet) DEFINE_STORE_ATTRIBUTE.

Still no cigar. That one is for rw files. Mine is writable only, so I use
e.g. no_llseek.

Probably, the above macro should be named, DEFINE_SHOWSTORE_ATTRIBUTE.
But I see this macro is stalled since late 2020 anyhow.