[semiPATCH] another vxfs fix

With the brelse() fix, a failed vxfs mount is OK.
But after a successful vxfs mount the umount still yields
  VFS: Busy inodes after unmount. Self-destruct in 5 seconds.  Have a nice day
(And today's [yesterday's] patch by Christoph makes no difference.)

The reason is that the routine vxfs_fake_inode() does
new_inode(), but this inode is never returned.
I fixed this but am not sure against which base source
the patch should be described.
A minimal version would be to change the six [1] calls of
	vxfs_put_inode()
in vxfs_super.c into calls of
	iput()

(In my source I turned vxfs_fake_inode() into vxfs_get_fake_inode()
and added vxfs_put_fake_inode() that just does iput().)

Andries

[1] 2.4.7 has three calls, Christoph's patch adds three more.