On Sun, Jul 15, 2001 at 12:09:04AM -0700, Jason R Thorpe wrote:
> On Sun, Jul 15, 2001 at 12:17:27AM +0700, Robert Elz wrote:
> 
>  > Surely the cleaner could just be fixed to look see if there's a boot block
>  > that refers to blocks in the filesystem it is cleaning, and if so, update
>  > the boot block whenever it moves the blocks it refers to (ie: /boot, though
>  > the cleaner doesn't need to know that).
>  > 
>  > The other option (of somehow marking the /boot blocks as not being moveable)
>  > seems sub-optimal to me.
>  > 
>  > Keeping on requiring boot out of an ffs is permanently relegating lfs to
>  > 2nd class status.
> 
> If you can fit pathname resolution into the i386 primary boot, then you
> can just look up /boot.  This is what the Alpha and pmax boot programs
> do.

Yes, improving the i386 boot blocks so that the "primary" bootblock
does this would also greatly simplify the setup of having `/' as a
RAIDframe mirrored (RAID 1) set.

There is enough room in the first 8K of a ffs or lfs partition for this,
assuming such a partition is the first partition in the NetBSD `slice'
(DOS partition) of the disk.  This would normally be the case, of course.
You only have to skip the 2nd sector (where the disklabel is stored :);
leaving you with a 0.5K chunk and a 7K chunk...

Making `/boot' a "normal" file (rather than one with its location hard-coded
into the primary boot block) would be cool in so many ways...

(Now, if only I had enough PC clue to fix this myself...)