Re: [RFC PATCH 2/3] mm/filemap: Avoid modifying iocb->ki_flags for AIO in filemap_get_pages()

[Chi Zhiling]

On 2025/7/23 22:33, Matthew Wilcox wrote:
> On Wed, Jul 23, 2025 at 06:18:24PM +0800, Chi Zhiling wrote:
> > From: Chi Zhiling <chizhiling@xxxxxxxxxx>
> >
> > Setting IOCB_NOWAIT in filemap_get_pages() for AIO is only used to
> > indicate not to block in the filemap_update_page(), with no other purpose.
> > Moreover, in filemap_read(), IOCB_NOWAIT will be set again for AIO.
> >
> > Therefore, adding a parameter to the filemap_update_page function to
> > explicitly indicate not to block serves the same purpose as indicating
> > through iocb->ki_flags, thus avoiding modifications to iocb->ki_flags.
> >
> > This patch does not change the original logic and is preparation for the
> > next patch.
>
> Passing multiple booleans to a function is an antipattern.
> Particularly in this case, since we could just pass iocb->ki_flags
> to the function.
>
> But I think there's a less complicated way to do what you want.
> Just don't call filemap_update_page() if there are uptodate folios
> in the batch:
>
> +++ b/mm/filemap.c
> @@ -2616,9 +2616,10 @@ static int filemap_get_pages(struct kiocb *iocb, size_t count,
>                          goto err;
>          }
>          if (!folio_test_uptodate(folio)) {
> -               if ((iocb->ki_flags & IOCB_WAITQ) &&
> -                   folio_batch_count(fbatch) > 1)
> -                       iocb->ki_flags |= IOCB_NOWAIT;
> +               if (folio_batch_count(fbatch) > 1) {
> +                       err = -EAGAIN;
> +                       goto err;
> +               }


Yes, this is a completely better way.


Would you mind if I add
"Signed-off-by: Matthew Wilcox (Oracle) <willy@xxxxxxxxxxxxx>"
in the next version?

>                  err = filemap_update_page(iocb, mapping, count, folio,
>                                            need_uptodate);
>                  if (err)

Thanks,