Re: [PATCH v6 01/27] mm: Introduce struct folio

[Matthew Wilcox]

On Tue, Apr 06, 2021 at 04:05:50PM +0100, Christoph Hellwig wrote:
> On Tue, Apr 06, 2021 at 03:55:11PM +0100, Matthew Wilcox wrote:
> > Assuming we're getting rid of them all though, we have to include:
> > 
> > $ git grep 'page->mapping' fs |wc -l
> > 358
> > $ git grep 'page->index' fs |wc -l
> > 355
> 
> Are they all going to stay?  Or are we going to clean up some of that
> mess.  A lot of ->index should be page_offet, and on the mapping side
> the page_mapping and page_file_mapping mess is also waiting to be
> sorted..

About a third of ->index can be folio_offset(), based on a crude:

$ git grep 'page->index.*PAGE_' |wc -l
101

and I absolutely don't mind cleaning that up as part of the folio work,
but that still leaves 200-250 instances that would need to be changed
later.

I don't want to change the page->mapping to calls to folio_mapping().
That's a lot of extra work for a page which the filesystem knows belongs
to it.  folio_mapping() only needs to be used for pages which might not
belong to a filesystem.

page_file_mapping() absolutely needs to go away.  The way to do that
is to change swap-over-nfs to use direct IO, and then NFS can use
folio->mapping like all other filesystems.  f2fs is just terminally
confused and shouldn't be using page_file_mapping at all.  I'll fix
that as part of the folio work.