On Fri, 24 Dec 2004, Ognyan Kulev wrote:
> On Thu, Dec 23, 2004 at 06:52:00PM -0600, Frederick Bruckman wrote:
>> In article <41CAEA77.9070009@fmi.uni-sofia.bg>,
>> 	ogi@fmi.uni-sofia.bg (Ognyan Kulev) writes:
>>>
>>> The problems started when I started transferring kern-GENERIC.  I made the
>>> mistake to split and transfer netbsd-GENERIC.gz.  Of course, untarring failed.
>>> I became aware of my mistake and I have transferred kern-GENERIC.tgz.  But
>>> untarring failed again!  After more time of trying, I understood that
>>> kern-GENERIC.tgz in /targetroot/usr/INSTALL is not truncated before transferring
>>> the split tar.
>>
>> You need the INSTALL floppies for that...

> Yes, I booted with boot1.fs and boot2.fs, and later selected floppy
> installation media.  But before kern-GENERIC.a{a,b,c} are loaded from
> floppy, /targetroot/usr/INSTALL/kern-GENERIC.tgz is not truncated, so when
> loading set is canceled and is tried again later during install, the tgz is
> corrupted -- this is the (first) problem.
>
>>> So this is the first problem -- I think that before getting split parts and
>>> concatenating them, the tgz should be truncated.  Why it isn't so?

Ah, I misunderstood. I guess that came with the change so that the 
number of sets did not have to be hard-coded in the installer.

Starting from scratch doesn't "newfs" the target file-system? In that 
case, you could break out to the shell (either from the utility menu, 
or simply exit the installer), mount the targetroot manually, if 
necessary, and clean out .../usr/INSTALL manually.

If the installer doesn't have an exit option, hit "Ctrl-C". Now you 
have "disklabel", "mount", "newfs", "dd".  If all else fails, for 
example, you can re-initialize and re-certify the drive with the

   dd if=/dev/zero of=/dev/rwd0d bs=16k

(Warning: that destroys any other partitions, too). I actually prefer 
to do that always on new installations, because it not only makes the 
drive more reliable -- bad blocks are marked -- but sometimes, it 
makes it perform better, too, and it can't hurt.  Following that, 
"Ctrl-D" exits the single user shell and reloads "sysinst" without 
reloading the kernel (which is just a little faster than rebooting 
all over from floppy).

One of my first NetBSD/i386 networks involved a "fast" 486 laptop 
build host connected to the old 386 file server only over PPP. NFS, 
too.  Good times, heh.

The INSTALL kernels do suppport SLIP, by the way, if the all-floppy 
install proves too cumbersome.


Frederick