Re: [PATCH 1/3] xfs: Remove xfs_filemap_map_pages() wrapper

[Dave Chinner <david@fromorbit.com>]

On Thu, Feb 09, 2023 at 02:44:20AM +0000, Matthew Wilcox wrote:
> On Thu, Feb 09, 2023 at 08:53:11AM +1100, Dave Chinner wrote:
> > > If XFS really needs it,
> > > it can trylock the semaphore and return 0 if it fails, falling back to
> > > the ->fault path.  But I don't think XFS actually needs it.
> > >
> > > The ->map_pages path trylocks the folio, checks the folio->mapping,
> > > checks uptodate, then checks beyond EOF (not relevant to hole punch).
> > > Then it takes the page table lock and puts the page(s) into the page
> > > tables, unlocks the folio and moves on to the next folio.
> > > 
> > > The hole-punch path, like the truncate path, takes the folio lock,
> > > unmaps the folio (which will take the page table lock) and removes
> > > it from the page cache.
> > > 
> > > So what's the race?
> > 
> > Hole punch is a multi-folio operation, so while we are operating on
> > invalidating one folio, another folio in the range we've already
> > invalidated could be instantiated and mapped, leaving mapped
> > up-to-date pages over a range we *require* the page cache to empty.
> 
> Nope.  ->map_pages is defined to _not_ instantiate new pages.
> If there are uptodate pages in the page cache, they can be mapped, but
> missing pages will be skipped, and left to ->fault to bring in.

Sure, but *at the time this change was made* other operations could
instantiate pages whilst an invalidate was running, and then
->map_pages could also find them and map them whilst that
invalidation was still running. i.e. the race conditions that
existed before the mapping->invalidate_lock was introduced (ie. we
couldn't intercept read page faults instantiating pages in the page
cache at all) didn't require ->map_pages to instantiate the page for
it to be able to expose incorrect data to userspace when page faults
raced with an ongoing invalidation operation.

While this may not be able to happen now if everything is using the
mapping->invalidate_lock correctly (because read faults are now
intercepted before they can instatiate new page cache pages), it
doesn't mean it wasn't possible in the past.....

-Dave.
-- 
Dave Chinner
david@fromorbit.com