Received-SPF: pass (google.com: domain of linux-ext4-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
MIME-Version: 1.0
References: <20221216150626.670312-1-agruenba@redhat.com> <20221216150626.670312-3-agruenba@redhat.com>
 <Y6XDAG25Qumt/iyM@infradead.org>
In-Reply-To: <Y6XDAG25Qumt/iyM@infradead.org>
From:   =?UTF-8?Q?Andreas_Gr=C3=BCnbacher?= <andreas.gruenbacher@gmail.com>
Date:   Fri, 23 Dec 2022 21:54:34 +0100
Message-ID: <CAHpGcMJAnyn_7hHvsPL5GAiwbJs_DX04+Tt0P+6jfi_kb7jGUg@mail.gmail.com>
Subject: Re: [RFC v3 2/7] iomap: Add iomap_folio_done helper
To:     Christoph Hellwig <hch@infradead.org>
Cc:     Andreas Gruenbacher <agruenba@redhat.com>,
        "Darrick J . Wong" <djwong@kernel.org>,
        Alexander Viro <viro@zeniv.linux.org.uk>,
        Matthew Wilcox <willy@infradead.org>,
        linux-xfs@vger.kernel.org, linux-fsdevel@vger.kernel.org,
        linux-ext4@vger.kernel.org, cluster-devel@redhat.com
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

Am Fr., 23. Dez. 2022 um 16:12 Uhr schrieb Christoph Hellwig
<hch@infradead.org>:
> On Fri, Dec 16, 2022 at 04:06:21PM +0100, Andreas Gruenbacher wrote:
> > +static void iomap_folio_done(struct iomap_iter *iter, loff_t pos, size_t ret,
> > +             struct folio *folio)
> > +{
> > +     const struct iomap_page_ops *page_ops = iter->iomap.page_ops;
> > +
> > +     if (folio)
> > +             folio_unlock(folio);
> > +     if (page_ops && page_ops->page_done)
> > +             page_ops->page_done(iter->inode, pos, ret, &folio->page);
> > +     if (folio)
> > +             folio_put(folio);
> > +}
>
> How is the folio dereference going to work if folio is NULL?

'&folio->page' is effectively a type cast, not a dereference. I
realize iomap_folio_done() as introduced here is not pretty, but it's
only an intermediary step and the ugliness goes away later in this
series.

> That being said, I really wonder if the current API is the right way to
> go.  Can't we just have a ->get_folio method with the same signature as
> __filemap_get_folio, and then do the __filemap_get_folio from the file
> system and avoid the page/folio == NULL clean path entirely?  Then on
> the done side move the unlock and put into the done method as well.

Yes, this is what happens later in this series (as you've seen by now).

> >       if (!folio) {
> >               status = (iter->flags & IOMAP_NOWAIT) ? -EAGAIN : -ENOMEM;
> > -             goto out_no_page;
> > +             iomap_folio_done(iter, pos, 0, NULL);
> > +             return status;
> >       }
> >
> >       /*
> > @@ -656,13 +670,9 @@ static int iomap_write_begin(struct iomap_iter *iter, loff_t pos,
> >       return 0;
> >
> >  out_unlock:
> > -     folio_unlock(folio);
> > -     folio_put(folio);
> > +     iomap_folio_done(iter, pos, 0, folio);
> >       iomap_write_failed(iter->inode, pos, len);
> >
> > -out_no_page:
> > -     if (page_ops && page_ops->page_done)
> > -             page_ops->page_done(iter->inode, pos, 0, NULL);
> >       return status;
>
> But for the current version I don't really understand why the error
> unwinding changes here.

Currently, we have this order of operations in iomap_write_begin():

  folio_unlock() // folio_put() // iomap_write_failed() // ->page_done()

and this order in iomap_write_end():

  folio_unlock() // ->page_done() // folio_put() // iomap_write_failed()

The unwinding in iomap_write_begin() works because this is the trivial
case in which nothing happens to the page. We might just as well use
the same order of operations there as in iomap_write_end() though, and
when you switch to that, this is what you get.

Thank you for the review.

Andreas