releases/4.19.51/nfsd-allow-fh_want_write-to-be-called-twice.patch

   1 From ce9fca48a0fa1140bc35a07fced42d8580b39d8b Mon Sep 17 00:00:00 2001
   2 From: "J. Bruce Fields" <bfields@redhat.com>
   3 Date: Fri, 12 Apr 2019 16:37:30 -0400
   4 Subject: nfsd: allow fh_want_write to be called twice
   5
   6 [ Upstream commit 0b8f62625dc309651d0efcb6a6247c933acd8b45 ]
   7
   8 A fuzzer recently triggered lockdep warnings about potential sb_writers
   9 deadlocks caused by fh_want_write().
  10
  11 Looks like we aren't careful to pair each fh_want_write() with an
  12 fh_drop_write().
  13
  14 It's not normally a problem since fh_put() will call fh_drop_write() for
  15 us.  And was OK for NFSv3 where we'd do one operation that might call
  16 fh_want_write(), and then put the filehandle.
  17
  18 But an NFSv4 protocol fuzzer can do weird things like call unlink twice
  19 in a compound, and then we get into trouble.
  20
  21 I'm a little worried about this approach of just leaving everything to
  22 fh_put().  But I think there are probably a lot of
  23 fh_want_write()/fh_drop_write() imbalances so for now I think we need it
  24 to be more forgiving.
  25
  26 Signed-off-by: J. Bruce Fields <bfields@redhat.com>
  27 Signed-off-by: Sasha Levin <sashal@kernel.org>
  28 ---
  29  fs/nfsd/vfs.h | 5 ++++-
  30  1 file changed, 4 insertions(+), 1 deletion(-)
  31
  32 diff --git a/fs/nfsd/vfs.h b/fs/nfsd/vfs.h
  33 index a7e107309f76..db351247892d 100644
  34 --- a/fs/nfsd/vfs.h
  35 +++ b/fs/nfsd/vfs.h
  36 @@ -120,8 +120,11 @@ void               nfsd_put_raparams(struct file *file, struct raparms *ra);
  37
  38  static inline int fh_want_write(struct svc_fh *fh)
  39  {
  40 -       int ret = mnt_want_write(fh->fh_export->ex_path.mnt);
  41 +       int ret;
  42
  43 +       if (fh->fh_want_write)
  44 +               return 0;
  45 +       ret = mnt_want_write(fh->fh_export->ex_path.mnt);
  46         if (!ret)
  47                 fh->fh_want_write = true;
  48         return ret;
  49 --
  50 2.20.1
  51