[Linux-cachefs] [PATCH 14/14] mm, netfs, fscache: Stop read optimisation when folio removed from pagecache

[David Howells]

Matthew Wilcox <willy at infradead.org> wrote:

> On Thu, Apr 07, 2022 at 12:05:05AM +0100, David Howells wrote:
> > Fix this by adding an extra address_space operation, ->removing folio(),
> > and flag, AS_NOTIFY_REMOVING_FOLIO.  The operation is called if the flag is
> > set when a folio is removed from the pagecache.  The flag should be set if
> > a non-NULL cookie is obtained from fscache and cleared in ->evict_inode()
> > before truncate_inode_pages_final() is called.
> 
> What's wrong with ->freepage?

It's too late.  The optimisation must be cancelled before there's a chance
that a new page can be allocated and attached to the pagecache - but
->freepage() is called after the folio has been removed.  Doing it in
->freepage() would allow ->readahead(), ->readpage() or ->write_begin() to
jump in and start a new read (which gets skipped because the optimisation is
still in play).

Another possibility could be that the FSCACHE_COOKIE_HAVE_DATA and
FSCACHE_COOKIE_NO_DATA_TO_READ flags could be moved from cookie->flags to
mapping->flags and the VM could do the twiddling itself (no aop required) -
except that fscache can't currently then find them (maybe use an aop for
that?).

David