Re: [PATCH] First casuality of hlist poisoning in 2.5.70

[Trond Myklebust <trond.myklebust@fys.uio.no>]

>>>>> " " == Linus Torvalds <torvalds@transmeta.com> writes:

     > But I suspect that neither dentry nor target should really ever
     > be unhashed by the time we call d_move(). That's reinforced by
     > the fact that it looks like a unhashed dentry in d_move() would
     > have been a silent bug previously - staying unhashed if it just
     > shared the bucket.

It is not a bug.

Looking more carefully at the Oops, it seems that this problem is
occurring inside an nfs_rename() in which the target name belongs to
an open file - and thus needs to be sillyrenamed first.

In that case, we certainly do not want to rehash the dentry in order
to do the d_move() since that would give rise to a race: we want to do
a real rename into the same dentry after we're done with the
sillyrename.

I can agree that the patch was flawed, but I still believe that we do
need to allow d_move to work with unhashed dentries.

Cheers,
  Trond