Re: Is it possible to chroot and then install NetBSD to the boot drive

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

Hi,

On Mar 20, 2013, at 6:58 AM, David Lord wrote:

> On 19 Mar 2013 at 8:51, Al Zick wrote:
>
> > Hi,
> >
> > On Mar 17, 2013, at 6:57 PM, Greg Troxel wrote:
> >
> > > Al <al%familysafeinternet.com@localhost> writes:
> > >
> > > > What I would like to be able to do is do an install on either  
> > > > another
> > > > drive or on an NFS partition. Then chroot to it and then delete the
> > > > files that are on /, /var, and /usr and then do a clean install of
> > > > NetBSD. Is this possible? If it is, could someone give me a how  
> > > > to on
> > > > doing something like this?
> > >
> > > chrooting will be awkward, because once you chroot then you will be
> > > running the binaries from the new install, which only works if your
> > > host
> > > system is compatible (NetBSD, same or later, more or less).  And if
> > > you
> > > remove files, you won't be able to run new commands.
> > >
> > > However, you don't need to chroot.  If you take a new drive, and
> > > fdisk/disklabel it, and mount it, and then unpack the sets, and then
> > > install bootblocks, it should work.  You didn't mention, but if you
> > > mean
> > > i386, then bootblocks consists of installboot for bootxx_ffsv1
> > > (probably) and also /boot, plus mbr boot records.  If the host is
> > > NetBSD
> > > you can just use installboot and cp, pointing to /usr/mdec in the  
> > > new
> > > system.
> > >
> > > I have done something similar a number of times, basically moving a
> > > system from old disks to new (bigger, less aged and in theory more
> > > reliable) disks.  So instead of unpacking sets, I have done dump/
> > > restore
> > > or rsync from old to new (under /mnt) and then installed bootblocks.
> >
> > I have a system that I sent new drives out to the data center not
> > that long ago. I did a new install of NetBSD on them, but then I
> > decided to update to NetBSD 6.0. In the process, I broke some things.
> > Now the system is not very usable. I can get another hard drive and
> > install NetBSD on it and then send them that drive, but someone at
> > the data center will need to swap out the drive again. I don't see
> > how I could netboot this system and it is not real easy to tell it to
> > boot from another drive or partition, so I thought that there might
> > be some way to setup a root filesystem much like you do when you
> > netboot. Temporarily mount this partition to run make dev files, then
> > switch over to it, and unmount /home, /var, /usr,  and finely /. Run
> > newfs on the old /, /var, and /usr. Mount / as /mnt, then /var  as /
> > mnt/var and /usr as /mnt/usr. Then cd to /mnt and untar a working
> > install that I did on a test box that has the exact same config as
> > this system, so that everything will be right in /etc. Is there any
> > way to do this?
> >
> > Thanks,
> > Al
>
> Assuming you still have remote access to the system  it may
> not require the above and you can try method below on your
> test setup.
>
> I had an unusable system when I had tried update from 5.x
> to 6.0-beta and then tried to backout to 5.y. Possibly I
> had a bad 6.0beta cdrom. I then found that backing out is
> not supported.

This is exactly what I am looking to do is go from a system with  
NetBSD 6 back to NetBSD 5.2. There is no real way to down grade a  
NetBSD install that I know of.

> I still had ssh access and eventually I copied a new kernel
> and tried to use a script to install sets. This is best done
> single user but would require network access to be enabled
> and I've not tried that as my pcs are all accessible.
> Stopping as many services as practical may be advisable.
>
> (1) copy sets to some directory
>
> (2) copy new kernel to /
>
> (3) copy script to some directory
>
> (4) reboot
>
> (5) unpack sets
> #!/bin/sh
> cd /
> for i in base comp games man misc modules tests text ; do
>     pax -rvzpoe -f /sets/${i}.tgz
> done
> #
>
> (6) reboot
>
> The etc.tgz set is deliberately missing from above
> If feeling brave you could then use /etcmanage to tidy up
> but I prefer to use an occasional sysinstall from a cd.
>
> Without using /etcmanage I've updated from 4.x through to
> 6.1RC1 but have other systems with more recent cd installs
> and along with warnings in /var/log/messages or
> /var/run/dmesg can fix problems in /etc later.
>
> With 4.x and possibly 5.0 I needed to run "fsck_ffs -c 4"
> for all the filesystems and change entries in fstab to
> enable wapbl.

You did give me an idea though. I tested it on a local machine and I  
untarred the kernel, base.tgz, and etc.tgz on an nfs. I then updated  
the etc files, then "cd /mnt/dev' ran '/bin/sh MAKEDEV all'. After  
that I ran 'find . -exec touch -u -t todays date {}\;'. I then tarred  
it. I then went on to the old install and ran 'find . -exec touch -u - 
t 20091212000000 {} \;'. and untarred that file in /. After that I  
did 'find / -mtime 2 -delete' I then ran pwd_mkdb -p /etc/ 
master.passwd and run '/bin/sh MAKEDEV all' again this time in /dev.  
I then rebooted into NetBSD 5.1.2 and was able to ssh in. It is  
building NetBSD 5.2. now.

I have not tried it on a live system though.

Best Regards,
Al