On Wed, Aug 07, 2024 at 10:54:08AM GMT, Shakeel Butt wrote:
> On Wed, Aug 07, 2024 at 06:46:00PM GMT, Matthew Wilcox wrote:
> > On Wed, Aug 07, 2024 at 10:19:11AM -0700, Shakeel Butt wrote:
> > > On Wed, Aug 07, 2024 at 04:28:53PM GMT, Matthew Wilcox wrote:
> > > > On Tue, Jul 30, 2024 at 01:39:06PM -0700, Andrii Nakryiko wrote:
> > > > > + union {
> > > > > + struct {
> > > > > + struct address_space *mapping;
> > > > > + struct page *page;
> > > >
> > > > NAK. All the page-based interfaces are deprecated. Only we can't mark
> > > > them as deprecated because our tooling is a pile of crap.
> > > >
> > > > > + void *page_addr;
> > > > > + u64 file_off;
> > > >
> > > > loff_t pos.
> > > >
> > > > > + r->page = find_get_page(r->mapping, pg_off);
> > > >
> > > > r->folio = read_mapping_folio(r->mapping, r->pos / PAGE_SIZE, ...)
> > > >
> > > > OK, for network filesystems, you're going to need to retain the struct
> > > > file that's used to access them. So maybe this becomes
> > > > read_mapping_folio(r->file->f_mapping, r->pos, r->file)
> > >
> > > This code path can be called from non-sleepable context. What would be
> > > the appropriate way to get the folio in that case?
> >
> > There isn't. If there's no folio, or the folio isn't uptodate, we need
> > to sleep to wait for I/O. We can't busy-wait for I/O.
> >
>
> Failure is fine if there is no folio or the folio is not uptodate. I
> assume we can do:
>
> folio = __filemap_get_folio(r->mapping, pg_off, 0, 0);
> if (!folio || !folio_test_uptodate(folio))
> return -EFAULT;
And a folio_put(folio) if we return due to !folio_test_uptodate().
>
> Is this appropriate here?
