[K12OSN] restore backup with chroot + grub-install

[Robert Arkiletian]

Les Mikesell wrote:

>If you have booted an install disk in rescue mode it will be
>/mnt/sysinstall.  Do the the 'chroot /mnt/sysinstall' then
>at the next prompt '/sbin/grub-install /dev/sda' and
>exit twice to reboot.
>  
>

Thanks for reminding me to exit twice.  I would have forgot.

>>Would I still need to edit fstab and grub.conf if I chroot to the newly 
>>mounted and untared drive and do a grub-install from there?
>>    
>>
>
>Yes - you either have to add the labels yourself with e2label to
>match the old ones or change the references to use device names.
>It doesn't matter which you do, but the labels won't be added
>automatically and the system won't boot unless the references
>match something.  Don't panic if it doesn't boot the first time.
>  
>

I prefer to go without labels. So I will edit fstab and grub.conf. It 
looks easy enough.

>You can always reboot with the CD and fix things.  Note also that
>you have a swap partition at /dev/sda2.  You'll also have to fdisk
>this partition, setting its type to 82 (linux swap) and then
>'mkswap /dev/sda2'.  You can move this as long as you edit
>/etc/fstab to match.
>
>  
>

I  was planning on adding sda2 as a swap partition when I fdisk the new 
drive but I did not know I had to mkswap. Thanks Les. I assume the RH9 
will call swapon when it boots.

>
>I have some machines with an even worse problem with the RH9 CD - it
>won't even see the drives.  A fedora core1/2 or corresponding k12ltsp
>install CD will work in rescue mode, but as long as RH9 sees the drives
>it should be ok for the final step to make the disk bootable.  I think
>RH9 was still able to boot from a floppy so you might want to make
>one.  Later systems just don't fit.
>
>  
>

Yeah RH9 does see the drive, it will even work in production with my 
class but under heavy load it corrupted the whole filesystem last 
summer. So I stick to the newer aic79xx driver from Adaptec which has 
been stable.

>If you are going to make an exact image, you can trade system time
>for human trouble and just do a raw copy.  Plug the new drive in
>jumpered for a drive select above your existing 2, boot from CD and do:
>  dd if=/dev/sda of=/dev/sdc
>  
>

Actually Les, I did that TOO. You see my backup drive is a 120GB ide 
drive plugged into a mobile rack. My server has two 36GB scsi drives, 
one for / and one for /home. I used fdisk to make the first partition on 
the backup drive exactly match /dev/sda1 (the / partition).  Then

dd if=/dev/sda1 of=/dev/hdb1

then to copy the boot record

dd if=/dev/sda of=/dev/hdb bs=446 count=1

Actually I mistakenly did bs=512 and learned the hard way that the 
partition table was also copied. So I have a stupid swap partition hdb2 
on the backup drive mbr.
I cannot dd over /dev/sdb1 (the home partition) to another partition 
(maybe dd if=/dev/sdb1 of=/dev/hdb3) since the cylinders will not match. 
So I decided to copy /home using tar to hdb3. Plus, I don't have to 
worry about boot records and such for /home. Then I thought, with good 
backup mentality, why not also backup / using tar for redundancy, just 
in case you have problems with dd. If  I had another drive I would also 
dd over /home as well.

>to copy your first drive to the new (third) drive.  This will take all
>the boot, partition, and label information along.   Then do the
>same for /dev/sdb to copy the second drive.  You can put the disks
>in some other machine and copy over the network too.  Boot the source
>disk with gnoppix/knoppix to make sure the image doesn't change - the
>target can be running normally as long as the drives aren't mounted.
>The copy command in this situation would be:
> dd if=/dev/sda |ssh root at target_machine dd of=/dev/disk_name
>Be careful to get the remote disk name right, of course...
>  
>

Cool, I love seeing how you can use dd over a network.  In a perfect 
world I would have two identical scsi drives as the backup drives. So in 
time of need it would be quick and easy to restore. But two 36GB scsi 
drives are 5 times the cost of a 120GB ide drive.  Not to mention 
finding a mobile scsi rack.

>These copies will take a while but they are a good way to clone
>machines - just boot into single user mode and change the host name and
>IP number if you aren't replacing the source drives on an existing
>machine.
>
>I'll still plug backuppc as the way to handle backups though.  Install
>it once and it will be able to give you the tar images you are starting
>from any time you need them - current or back a few days.  The only
>change from your restore plan would be that you'd have to look up the
>command line to get the tar image you need and issue it via ssh from
>your gnoppix boot, piping to tar to extract locally on your new drive.
>  
>

I really appreciate this list and your expertise Les. Thank you. I have 
learned so much since I started K12LTSP at my school last May.

-- 


Robert Arkiletian
C++ GUI tutorial http://fltk.org/links.php?V219