At each new release, some sysadmins seem to get a bit skittish about upgrading critical production machines. In some environments, you have to be 100% sure you won't break something or at the very least be sure to have a fall-back plan should everything go pear-shaped. While you may have a patching policy in place, upgrading the entire OS in one fell swoop can be risky business if you are not completely prepared. Too much apprehension, and you don't get to take advantage of 'the new hotness' features and possibly result in running releases with either known issues or security problems.

There is a solution however, even on a tight budget. That solution (as the title suggests) is disk cloning. If you tried this before you may be saying, "Ghost is crap: it doesn't like *BSD disks" or "must be nice to have exact duplicate disks lying around."

Those approaches are good but we can do better.

Assuming you have another disk at least as large as the current one, we can clone the currently running system using only the built in tools OpenBSD provides.

The first step depends on how paranoid you are. If you can turn off the machine, then I would suggest putting the "other drive" in the machine this way. If you cannot turn the machine off and it have a free USB port, you can use a USB to IDE/SATA adapter (slow but effective). I'm not sure if a decent USB to SCSI adapter exists, but I do know of an LVD to IDE adapter that can be chained to a USB to IDE/SATA converter. I'm not all that keen on popping drives onto an active SCSI bus but if you can find a safe way of adding another raw drive device to the machine, that's all you really need.

Make sure you've got a couple disk device nodes (ie. wd0 and wd1, or sd0 and sd1 or wd0 and sd1 etc.).

Also, in this case I'm going to clone the entire disk (label and all). You can modify this method to upgrade to larger disks (i.e. grow slices) or convert a larger sparsely used drive to a compact flash card or other medium. In order to go through all the basic steps I'll just do a straight clone.

In this case I'll say my root drive is sd0 and my destination is sd1.

Cloning a drive:

1. We will first need save a copy of the source label. This is really easy. Just run the disklabel(8) command and store the output somewhere.

   disklabel /dev/sd0 > /tmp/label
   

2. The next step is to initialize the second disk, we can do this by using the fdisk(8) command and tell it to re-initialize the MBR (Master Book Record). If you already have a configured partition map then feel free to move on, but we're assuming here the disc is brand new and either formatted to another system or blank from the manufacturer.

   fdisk -yi /dev/sd1
   

3. Now that we have a copy of the source label and we've primed the destination we will need to copy that disklabel over to the new drive:

   disklabel -Rr /dev/sd1 /tmp/label
   

   NOTE: If the drives are of differing sizes you'll need to preserve the 'c' slice from the destination drive. Just copy the 'c' slice from disk label from the second drive and replace the 'c' slice in the /tmp/label and you'll be fine. This is also the point where you can change slice sizes though you'll want to use the disklabel editor if you are not comfortable editing the label manually. You can do that via disklabel -E sd1.
   
   
4. Now we initialize the individual slices on our drive and copy the data over. You will need to create new file systems on each of the slices. I will only show how to do the first using dump(8) and restore(8), but doing the rest is just a matter of rote.

   newfs /dev/sd1a
   mount /dev/sd1a /mnt
   (dump -0f - / )|(cd /mnt/; restore -rf -)
   umount /mnt;
   

5. Once all the slices are initialized and the data copied over, we just need to make the drive bootable, which is also deceptively easy. All we need to do is copy the bootstrap code (biosboot(8)) over to the drive and install it on to the disk using installboot(8):

   mount /dev/sd1a /mnt
   cp /usr/mdec/boot /mnt/boot
   /usr/mdec/installboot -v /mnt/boot /usr/mdec/biosboot /dev/sd1
   

That's all there is to it. You now have a clone of your disk. You can reboot it and run through the upgrade if you wish. You can even install the upgrade before rebooting by copying the release from the CD and extracting it to the new disk (you'll have to mount the slices of course).

With this method you can easily and safely upgrade your production machines to any particular release. The safety comes from being able to fall back to the previously known good drive configuration just in case something from a new release breaks your configuration and you've just got to get it working again.

Another use for this method is to do a "poor man's" RAID mirroring. You can leave the drive attached to your system and periodically, via cron(8), mirror the disk state to the other drive. Again, if you screw up or one of the drives die, you have a quick fix already raring to go. My suggestion here is to connect the drive every so often, mirror the data and when done disconnect the power and data cables (that way PSU or drive frying situation doesn't lay waste to your backup). This is a very inexpensive disaster recovery method, at least.

In case you had to do this over and over or periodically, here is a nice little script I threw together which will clone two disks (slice for slice) using everything mentioned previously.

#!/bin/sh
source=/dev/sd0
dest=/dev/sd1

# Initialize disk
disklabel $source > /tmp/label
fdisk -yi $dest
disklabel -Rr $dest /tmp/label

# Obtain slices to clone
slices=`grep ^\ /tmp/label  | cut -c 3 | grep -v c`
rm /tmp/label

# Clone each slice
for i in $slices; do
       newfs $dest$i
       mount $dest$i /mnt
       path=`df -h | grep $source$i | awk '{print $NF}'`
       (dump -0f - $path)|(cd /mnt; restore -rf -)
       umount /mnt;
done

# Install boot record.
mount ${dest}a /mnt
cp /usr/mdec/boot /mnt/boot
/usr/mdec/installboot -v /mnt/boot /usr/mdec/biosboot $dest

I hope that this article not only opened your eyes a bit as to how your system really works, even if only allowing you to hedge your bets or swoop in and save the day (like when that disk starts making a racket). Even if you don't wish to use this or prefer another method (please share), just going through this procedure once will make you more familiar with how your machine boots, is laid out and, best of all, it reduces the mysteries behind the installer (see, it is easier than you think).

While cloning disks is good I've always felt that upgrade path for OpenBSD is a little rough.

Uh?

Some other unix-like operating systems have come up with tools that perform the all necessary operations (apt, yum). Yes, I know BSD is not the same but that does not mean it can't become easier.

Here is how to do it the easy way:

- Boot on the CD-ROM

- Select 'U' for Upgrade

- Once the upgrade has been performed, reboot your computer

- Make sure your root .profile is configured correctly, with a valid PKG_PATH

- As root, enter:
pkg_info | awk '{print $1}' | xargs pkg_add -u -vv -i

That's it, your system is now up-to-date to the latest version (system+packages). Your configuration is intact, your personal directories are there and the machine should be up and running.

Is that so complicated?

Please note that the only exception to the above is if you have a wifi adapter on your machine (like I do) in which case you may need to download an upgraded firmware.

I have done this on 3+ personal machines for at least 3 or 4 past versions of OpenBSD and I have never had any problem. These are not production machines, but OpenBSD has never let me down in that respect.

The same applies also to patch branch. I think it would be very nice to be able to have a binary upgrade option available to follow the stable.

Agreed. That's about the only thing I would say is missing from OpenBSD. There were some sites that offered binary upgrades though.

Please don't forget /etc and friends (mergemaster, in packages, is the easy way). Sometimes bugs are fixed, e.g. in /etc/rc or /etc/netstart (as happened between 4.2 and 4.3). Occasionally there are other steps too: -current users should see http://www.openbsd.org/faq/current.html - when upgrading between versions there's a guide produced for each version e.g. http://www.openbsd.org/faq/upgrade42.html.

Even if you make a clean installation rather than upgrading, it's a good idea to read these guides - they will highlight some important changes (e.g. the recent flag day for carp(4) with IP or ARP balancing).

> - Make sure your root .profile is configured correctly, with a valid PKG_PATH
> - As root, enter:
> pkg_info | awk '{print $1}' | xargs pkg_add -u -vv -i

"pkg_add -ui" defaults to updating everything you have installed - no need for the pkg_info part.

I made a few changes to the script to get it to work:

#!/bin/sh
sourcedisk=wd0
targetdisk=wd1
source=/dev/$sourcedisk
dest=/dev/$targetdisk

# Initialize disk
disklabel $sourcedisk > /tmp/label
fdisk -yi $targetdisk
disklabel -Rr $targetdisk /tmp/label

# Obtain slices to clone
slices=`grep "^ " /tmp/label  | cut -c 3 | grep -v c`
rm /tmp/label

# Clone each slice
for i in $slices; do
       newfs $dest$i
       mount $dest$i /mnt
       path=`df -h | grep $source$i | awk '{print $NF}'`
       (dump -0f - $path)|(cd /mnt; restore -rf -)
       umount /mnt;
done

# Install boot record.
mount ${dest}a /mnt
cp /usr/mdec/boot /mnt/boot
/usr/mdec/installboot -v /mnt/boot /usr/mdec/biosboot $targetdisk

I added two variables to hold just the disknames and used those for the commands that also take this abbreviated form (disklabel, installboot, fdisk)

I also changes ^\ to "^ " which is more readable and better shows that there should be two spaces behinde the backslash, now there is only one, which simply hangs the grep command waiting for input

Hoping this helps others trying to get this script to work

And last but not least, THANKS A LOT for this post/script. I literally had a post-it on my desk today telling me to "FINALLY FIX THAT BACKUP TO SECOND DISK" ;-)

A number of years ago I had the same problem. At the time, I needed to clone FreeBSD, NetBSD, OpenBSD and Linux hard drives. And I didn't have a floppy drive or cdrom to use as a "staging area" to hold the boot sectors. Back then bootloaders weren't nearly as nice and advanced as they are now. Accomplishing this task was a pain in the butt.

So I wrote it up into a little HOWTO. You can still view it here Migrating Your System to Another Hard Drive. The procedure still works, although the document shows its age. It just uses tar, mkdir, and fdisk.