[vox-tech] Conversion

[Matt Holland]

On Thursday, May 22, 2003, at 01:39 PM, Mike Simons wrote:

> On Thu, May 22, 2003 at 12:59:21PM -0700, Matt Holland wrote:

<snip>

>>>  This is easy ... as Gabe mentioned it can be done remotely.
>>
>> Is it really?  I've been having problems doing this myself, as I try
> to
>> migrate a Gentoo system to Debian stable.
>
>   Yes, I've installed Debian on many dozens of machines that had some
> other linux system before hand at least 5 remotely...  I'm sorry he
> didn't ask for steps so I didn't bring them up.
>
>> The main problem is that I
>> can't get the system to boot, and I think this is because I can't get
>> mkinitrd to work.
>
> If you have physical access:
>   Step one - Format your old root partition,
>   Step two - Proceed through the install process
>   Step three - ...
>   Step four - Working system (or profit!).

I'm trying to do this while keeping the old system around; hence, I 
added a second drive (hdb) and made two partitions: hdb1 for root, hdb2 
for swap.  Formatted /dev/hdb1, mounted it at /mnt/debinst, and 
debootstrap'd into /mnt/debinst... chroot'd to /mnt/debinst, set up 
apt, and things pretty much worked as expected.  I'm probably leaving 
some steps out, but I followed the instructions in section 3.7 of the 
Woody installation manual (Installing Debian GNU/Linux from a 
Unix/Linux System).

>> I have noticed some curious things when I'm in the chroot environment.
>
>> For one thing, /etc/mtab claims that /proc is the only mounted file
>> system:
> [...]
>> My /etc/fstab looks like:
>>
>> # file system   mount point     type    options         dump    pass
>> /dev/hdb1       /               ext2    defaults        1       1
>> /dev/hdb2       none            swap    sw              0       0
>> proc            /proc           proc    defaults        0       0
>> /dev/fd0        /mnt/floppy     auto    noauto,rw,sync,user,exec 0 0
>
>   This is not the system he described, he said home was another
> partition.

Right, this is my system... I'm horning in on an established thread.

>> So... it seems to me that maybe something is wrong in the chroot?  It
>> seems that if this is the way things should be, it would be impossible
>
>> to run mkinitrd from inside a chroot, in which case it would seem like
>
>> for your initial boot you'd have to use a kernel that doesn't use
>> initrd,
>
>   Debian does not (at least upto woody) use initrd for the base
> system...
> I don't think it even uses initrd for the install process... it loads
> a compressed ramdisk filesystem as root.  Unless I'm mistaken this is
> not initrd, it was available _long_ before initrd became available
> (which is a little root like filesystem does some setup and then
> either disappears or trades place with the "new" root filesystem).

Debootstrap doesn't install a kernel, but in order to boot the system 
after setting it up in the chroot, you need to install a kernel, and 
all of the appropriate images that I see in "apt-cache search 
kernel-image" seem to use initrd.  I've thought of compiling my own 
kernel without initrd, but "make menuconfig" doesn't work in the chroot 
(some complaint about missing ncurses, even though libncurses is 
installed and other curses interfaces [e.g., setting up apt] work 
fine), and I'm too impatient to survive an old-fashioned "make config".

>
>> Ah, one more thing.  I'm using Grub as my bootloader, as I'm keeping
>> the Gentoo system around for the time being, and it's set up with
> Grub.
>>  My Grub entry for Debian looks like:
>>
>> title=Debian
>> root (hd1,0)
>> kernel /boot/vmlinuz-2.4.18-k6 root=/dev/hdb1
>                                              ^^
>   Perhaps someone more familiar with Grub can comment, but it appears
> you
> are passing the wrong partition to the kernel... if the fstab above is
> correct.

How so?  (hd1,0) is Grub-speak for /dev/hdb1, and /dev/hdb1 *is* the 
root device.

> So anyway there are a bunch of different ways to load Debian remotely
> if that is what you are after... what I have done most often is to
> use a local machine to do the base install and configuration on
> (including sshd), then once I'm sure things are setup for the target
> machine, I tar up the filesystem image, convert the remote machine's
> swap
> into ext3 and dump the tar there.  Configure lilo to boot the new
> system,
> and do that... from there I nuke the old root filesystem and dump the
> tar again, reconfigure lilo to boot that one by default... and reboot
> again.  Take back the swap partition and go from there...

Maybe I'll do something like that to make the initrd image... then 
perhaps I can find out if that's really my problem.

>   The installing into a chroot environment sounds interesting, but
> I've not had a reason to fiddle with it so I can't help with pointers.

Seems like it should work, though, since it's documented in the install 
guide.

Thanks,
Matt

--
Matt Holland
Population Biology Graduate Group
University of California
Davis, CA 95616