H. Peter Anvin writes:
> > H. Peter Anvin writes:
> > > Will do. For what it's worth, I discussed a few options for
> > > supporting direct mounts (based on Richard's suggestion) with Linus
> > > today, and I might actually be able to get a VFS change into 2.3
> > > that would let us support direct mounts; we're evaluating a few
> > > methods (including one option of supporting sandwich mounts that
> > > hadn't been considered before.) Please don't get your hopes too far
> > > up, but there is at least a possibility that direct mounts might be
> > > supportable after all.
> >
> > Could you explain what you have in mind for these sandwich mounts? How
> > would they work?
>
> Basically, I argued with Linus for letting me add another pointer to
> the dentry. Instead of d_mounts and d_covers we would have a d_top,
> d_bottom and d_down pointers. This means you still don't need any
> tests: d_top will always take you to the topmost dentry (like the
> current d_mounts) and d_bottom will take you to the bottommost entry
> (like the current d_covers). The d_down pointers would form a
> linked list of entries to be popped off by umount.
Brilliant. I wish I'd thought of it ;-) That's much better than
hacking around with fakes.
Regards,
Richard....
