Re: building userland with a netbooted system

[Al <al%familysafeinternet.com@localhost>]

Hi,

On Sun, 17 Feb 2013, Mouse wrote:
> > I tried the latest 5.1 source and it did get a little bit farther (I
> > think).
>
> As I understand it, 5.1 is released and thus is frozen, so "latest" is
> pretty meaningless as applied to it.

I used cvs checkout -r netbsd-5-1 -P src, the old source I was trying to 
use, I used cvs to get in the summer or fall of 2011.

I guess then I don't understand. So, maybe what I did was not that 
significant then. I use satellite internet and most times it takes a lot 
of effort to download anything. Since cvs seems to fail so much over the 
satellite, I use cvs on a system that is in another location, tar it, and 
then download it. There was an nfsd bug and when I updated the source with 
cvs and then compiled and instaledl it, that system became stable. In 
fact, I don't know if it crashed again. I want to make sure that I have 
source that is newer than this, because I use nfs all the time.

In this message someone talked about raising some limits in m68k or maybe 
just port-amiga. Here is a link: 
http://mail-index.netbsd.org/port-m68k/2012/06/13/msg000340.html (I can't 
find the email where someone said they committed these changes. I am not 
sure how to find out if it was added to NetBSD 5.1 or just to NetBSD 6 or 
if it was even put in at all. It would be nice to have the source with 
these changes in them too, because I would like to use the same source on 
all my systems. Although, I have heard that NetBSD 6 has -m68060 bug in 
gcc creating compiling issues.

> > Still does not complete the build. Here is a link:
> > http://datazap.net/sites/14/build.log
>
> Quoting from that,
>
> > /usr/src/../tools/bin/powerpc--netbsd-ar cq libedit_pic.a [...]
> > nbmake: don't know how to make nbmake:. Stop
>
> I have no clue what's wrong here.
>
> I dug out my macppc machine (a 7300 with a G3 CPU upgrade).  It has, as
> I thought, 4.0.1 on it.  I have 5.2 cross-building from amd64; onjce
> it's built I'll put it on the drive (there's plenty of spare space) and
> try a native build.
>
> (Back to quoting the email.)
>
> > Just wondering if I am doing something wrong when I setup the
> > netboot.  I untar everything with tar -xzpf.  In the online docs it
> > says to untar the kernel in one step and then a few steps later to
> > untar base.tgz and etc.tgz with tar -xvpzf which I would think is
> > effectively the same thing or is it?
>
> The presence or absence of v doesn't matter except for how noisy it is.
>
> > When I did it, I untarred all the tars files in the order they would
> > they would be in with 'ls'.  Could it matter what order I untar them
> > in?
>
> Maybe.  I'm not sure; I always unpack base first, then etc, then do the
> rest in whatever order is convenient.

I redid my diskless install and I created a 1 GB swap file. Unpacking the 
kernel, base and then etc first.

> > I think that I can netboot a live server and then I can transfer the
> > netbooted install to the hard drive with 'dump | restore',
>
> Yes and no.  Your concept is right but your details are wrong.
>
> You can transfer your netbooted system to a disk and have it work
> (you'll also need to set up whatever is needed for disk booting, which
> IIRC isn't much for macppc).  But you can't do it with dump and
> restore, because dump dumps UFS filesystems, not NFS.

You are absolutely right, there is no way that dump is going to work with 
NFS.

> There are (at least) two possibilities here.  One is, you can do it
> with tar (or moral equivalent, eg pax) instead of dump and restore.
> The other is, if the filesystem on your NFS server doesn't have
> anything but the macppc system on it, you can run dump there and pipe
> the dump canister over the net to your macppc machine to restore it
> onto the real disk.
>
> Personally, I'd go with tar.  But there are two reasons for that, one
> of which does not apply to you and the other of which might: (1) I have
> my own tar, which trust to get such things right more than I trust
> dump/restore; (2) my NFS server has multiple clients on the same
> filesystem.
>
> Actually, I'd suggest doing that transfer now, before you do the build.
> That way, your build can be done on real disk, which is likely to be
> faster and might even be non-crashy.

This is true. It would be much faster to install to it's hard drive (the 
NFS server is a 68k).

I should have explained this better. I want to take the diskless install 
after I compile everything and tar it up, so that I can upload it to 
another nfsd server in another location. Then I will netboot another G3 
Blue & White server there, with the idea that it can be live while I pax 
the install to one of it's hard drives (the drive I boot off of now).

If did the transfer now to the Blue & White that I have here, I could then 
test to see if it would compile. If it did, I can tar it,t saving the file 
on another system that is here locally with ssh. Then I can untar that on 
my local nfsd server to see if everthing will be ok first, before 
uploading to another system. I should be able to just untar the same file 
on the drive, I would think. This would probably be the better solution by 
far.

Unless you know of a better way?

Thanks,
Al