Re: factoring out the change_root() and making exporting change_{root,dir}() as interface

[Antti Kantee <pooka%cs.hut.fi@localhost>]

On Thu Jul 30 2009 at 15:03:09 +0300, Antti Kantee wrote:
> On Wed Jul 29 2009 at 00:36:37 +0200, Christoph Badura wrote:
> > I'd like to factor out the common code of the *chroot syscalls like in
> > the following patch.  I'd also like to export change_dir() and the new
> > change_root() as an interface from vfs_syscalls.h
> > 
> > I want to use change_dir() and change_root() in a different kernel module.
> > 
> > I've been the code in 5.99.4 kernel for a while.  But this is the diff 
> > ported
> > to 5.99.15 for discussion purposes.
> > 
> > Comments? Objections?
> 
> Little comment: I use something similar in rump for wrapping the root
> vnode for ukfs calls.  Since the root vnode is always the same and
> referenced for a given ukfs mount, there is no need to do namei() with
> every invocation.  If you'd expose it on a vnode level, I could get rid
> of the hairbrained rump_rcvp_lwpset() implementation (which, admittedly,
> is an "interface" in the style of "let's throw some funk there so that
> I can work on the interesting bits").
> 
> but, either way, no objection.

Per some private discussion with Christoph, seems like I misread some
other comments in this thread about paths and apparently change_root()
takes only a vnode.  So edit my reply to: "yes, change_root() change
would be nice as is".