Re: [PATCH 1/5] NFS: Protect against 'eof page pollution'

[Trond Myklebust <trondmy@kernel.org>]

On Sat, 2026-01-24 at 21:01 +0000, Lorenzo Stoakes wrote:
> On Fri, Sep 05, 2025 at 03:35:16PM -0400, Trond Myklebust wrote:
> > From: Trond Myklebust <trond.myklebust@hammerspace.com>
> > 
> > This commit fixes the failing xfstest 'generic/363'.
> > 
> > When the user mmaps() an area that extends beyond the end of file,
> > and
> > proceeds to write data into the folio that straddles that eof,
> > we're
> > required to discard that folio data if the user calls some function
> > that
> > extends the file length.
> > 
> > Signed-off-by: Trond Myklebust <trond.myklebust@hammerspace.com>
> 
> Hi Trond,
> 
> Sorry to dig up an old patch but I wanted to ask about this change:
> 
> > ---
> >  fs/nfs/file.c      | 33 +++++++++++++++++++++++++++++++++
> >  fs/nfs/inode.c     |  9 +++++++--
> >  fs/nfs/internal.h  |  2 ++
> >  fs/nfs/nfs42proc.c | 14 +++++++++++---
> >  fs/nfs/nfstrace.h  |  1 +
> >  5 files changed, 54 insertions(+), 5 deletions(-)
> > 
> > diff --git a/fs/nfs/file.c b/fs/nfs/file.c
> > index 86e36c630f09..af2fdbfcbbf6 100644
> > --- a/fs/nfs/file.c
> > +++ b/fs/nfs/file.c
> > @@ -28,6 +28,7 @@
> >  #include <linux/mm.h>
> >  #include <linux/pagemap.h>
> >  #include <linux/gfp.h>
> > +#include <linux/rmap.h>
> >  #include <linux/swap.h>
> >  #include <linux/compaction.h>
> > 
> > @@ -280,6 +281,37 @@ nfs_file_fsync(struct file *file, loff_t
> > start, loff_t end, int datasync)
> >  }
> >  EXPORT_SYMBOL_GPL(nfs_file_fsync);
> > 
> > +void nfs_truncate_last_folio(struct address_space *mapping, loff_t
> > from,
> > +      loff_t to)
> 
> So this seems to be a slightly adjusted version of
> pagecache_isize_extend(),
> what was it about that that didn't work for you?
> 
> It seems the main differences are - block size alignment (surely you
> still need
> that though?) switching folio_test_dirty() for folio_test_uptodate()
> (I'm not
> sure about this change though?) and adding the trace line.

   1. NFS is not a block protocol. Reads and writes are byte aligned.
   2. The test for bsize >= PAGE_SIZE is nonsense for a byte aligned
      filesystem.
   3. NFS does care about using folio_mkclean() to fix races between an
      application that is writing to the folio, and any zeroing of the
      data that may result from the file truncation.
   4. The existing folio dirty state isn't of interest here, since NFS
      won't extend existing writes to cover the part of the folio that
      lies after the old eof.
   5. The folio uptodate state is of interest, since any future
      pagecache read needs to see zeroed bytes starting at the old eof
      and extending either to the offset at the end of the folio, or to
      the new eof (whichever of the two is smaller).


-- 
Trond Myklebust Linux NFS client maintainer, Hammerspace
trondmy@kernel.org, trond.myklebust@hammerspace.com