On Fri, Jul 22, 2011 at 08:42:35PM -0700, Linus Torvalds wrote:

quoted

... why is that a problem? ?Locking there is mere ->i_lock and getting
a refcount is atomic_inc(). ?Grabbing a reference might be Not Nice from
the cacheline bouncing POV, but...

I agree in theory, but it's not something we've done before. Some of
the posix-acl code is pretty disgusting, I didn't even want to go
there.

And the case that tends to really *matter* is the "no acls" case
anyway, and that's the case that is guaranteed to have no nasty races
or odd issues with having to allocate/de-allocate any acl structures.
But yes, this could be looked at.

Heh...  In addition to ocfs2 leak: 9p leaks nicely if v9fs_acl_mode() is
called with !S_ISDIR(mode).  In that case acl reference is simply lost.
So yes, it's worth looking at.