Re: introduce struct fderr, convert overlayfs uses to that

[Al Viro <viro@zeniv.linux.org.uk>]

On Fri, Oct 04, 2024 at 12:47:09PM +0200, Amir Goldstein wrote:

> I had already posted an alternative code for overlayfs, but in case this
> is going to be used anyway in overlayfs or in another code, see some
> comments below...

As far as I can see, the current #overlayfs-next kills the case for
struct fderr; we might eventually get a valid use for it, but for the
time being I'm going to strip the overlayfs-related parts of that branch
(obviously), fix the braino you've spotted in fdput() and archive the
branch in case it's ever needed.

> > +#define fd_empty(f)    _Generic((f), \
> > +                               struct fd: unlikely(!(f).word), \
> > +                               struct fderr: IS_ERR_VALUE((f).word))
> 
> 
> I suggest adding a fd_is_err(f) helper to rhyme with IS_ERR().

Umm...  Dropping fd_empty() for that one, you mean?

> > +#define fdput(f)       (void) (_Generic((f), \
> > +                               struct fderr: IS_ERR_VALUE((f).word),   \
> 
> Should that be !IS_ERR_VALUE((f).word)?

It should, thanks for spotting that braino.
 
> or better yet
> 
> #define fd_is_err(f) _Generic((f), \
>                                 struct fd: false, \
>                                 struct fderr: IS_ERR_VALUE((f).word))

I think that's a bad idea; too likely to spill into struct fd users,
with "it's never false here" being a nasty surprise.