Re: [PATCH] mm/filemap: do not allocate cache pages beyond end of file at read

[Konstantin Khlebnikov]

On 28/10/2019 15.57, Kirill A. Shutemov wrote:
> On Mon, Oct 28, 2019 at 01:47:16PM +0100, Linus Torvalds wrote:
> > On Mon, Oct 28, 2019 at 1:42 PM Kirill A. Shutemov <kirill@xxxxxxxxxxxxx> wrote:
> > > I've tried something of this sort back in 2013:
> > >
> > > http://lore.kernel.org/r/1377099441-2224-1-git-send-email-kirill.shutemov@xxxxxxxxxxxxxxx
> > >
> > > and I've got push back.
> > >
> > > Apparently, some filesystems may not have valid i_size before >readpage().
> > > Not sure if it's still the case...
> >
> > Well, I agree that there might be some network filesystem that might
> > have inode sizes that are stale, but if that's the case then I don't
> > think your previous patch works either.
> >
> > It too will avoid the readpage() if the read position is beyond i_size.
> >
> > No?
>
> Yes. That's the reason the patch was rejected back then.
>
> My point is that we need to make sure that this patch not break anything.

I think all network filesystems which synchronize metadata lazily should be
marked. For example as "SB_VOLATILE". And vfs could handle them specially.

For this case generic_file_buffered_read() could call for them readpages
for single page (rather than readpage) to let filesystem revalidate
metadata and drop unneeded page without inserting it into inode and lru.