On Tue, Oct 27, 2020 at 11:45 AM Christoph Hellwig <hch@xxxxxxxxxxxxx> wrote:
>
> > index f277d023ebcd..b55e6ef4d677 100644
> > --- a/fs/kernfs/file.c
> > +++ b/fs/kernfs/file.c
> > @@ -124,7 +124,7 @@ static void *kernfs_seq_start(struct seq_file *sf, loff_t *ppos)
> > * The same behavior and code as single_open(). Returns
> > * !NULL if pos is at the beginning; otherwise, NULL.
> > */
> > - return NULL + !*ppos;
> > + return (void *)(uintptr_t)!*ppos;
>
> Yikes. This is just horrible, why bnot the completely obvious:
>
> if (ops->seq_start) {
> ...
> return next;
> }
>
> if (*ppos)
> return NULL;
> return ppos; /* random cookie */
I was trying to not change the behavior, but I guess we can do better
than either the original version mine. Not sure I'd call your version
'obvious' either though, at least it was immediately clear to me that
returning an unrelated pointer here is the right thing to do (it works,
since it is guaranteed to be neither NULL nor an error pointer
and it is never dereferenced, but it's still odd).
I'd rather define something like
#define SEQ_OPEN_SINGLE (void *)1ul
and return that here. I'll send a patch doing that, let me know what
you think.
Arnd
