I had another look at what it would take to reimplement shiftfs as a true bind mount. It turns out we do have struct path threaded in almost enough places to make it work. There really is only one API that needs updating and that's notify_change(), so the following patch fixes that and pulls do_truncate() along as well. The updates are mostly smooth and pretty obvious because the path was actually already present, except for in overlayfs where trying to sort out what the path should be is somewhat of a nightmare. If the overlayfs people could take a look and make sure I got it right, I'd be grateful. I think this is the only needed change, but I've only just got a functional implementation of a uid/gid shifting bind mount, so there might be other places that need rethreading as I find deficiencies in the current implementation. I'll send them along as additional patches if I find them James
