OpenBSD readonly root

Theory

In a lot of situations, you'll find yourself not wanting to check filesystems, thereby
facilitating hard power-downs with ~no data loss. I've come across this case a few times,
mostly in the end-user or volatile-power routing environments. Most often, I've done this
on Soekris net4801 devices.
We'll show an example of how to do it on a net4801, which is the same procedure for a net5501,
and we'll be using OpenBSD 4.6 for the example.

To perform such a magic trick, you need to do a bit more planning. It's not as easy to run some
services in a configuration like this, since they need areas to write to. Even worse is when they
need to write data that you actually want to keep. In this case, I like to avoid the RO root
scenario, because it adds some awkward complexity that involves a regular mount-writable/sync/remount-ro
cycle. In general though, most services you can get away with just providing them ramdisk scratch
space.

Practice

Installation

Perform a PXE-boot standard OpenBSD 4.6 i386 install, the following differences apply:

• Press ^P at the 5-second BIOS countdown, then "boot f0" to boot from the first ethernet port.
• Interrupt the loader after it gives you a prompt (type anything, it stops the counter).
• Type the following, in this order:

 stty com0 19200
 set tty com0
 boot /bsd.rd

You should now be booting a kernel and be tossed into the OpenBSD installer.. look elsewhere
for instructions on how to properly setup PXE booting for OpenBSD, or maybe I'll document it
someday.

• When installing, you should partition your CF card with 1 giant / partition, and *no swap*
• It'll ask you later on if you want to change the console to com0, tell it yes and choose 19200 to match the soekris BIOS settings.
  
• Once installed, reboot into your installation and do basic post-install configuration, including setting up network interfaces, adding packages, configuring sudo, etc.. the idea is to have a fully installed system that just needs configuration.

Login and 'sudo ksh' if you haven't already: you'll need to perform a number of steps,
which I'll detail below..

 # cp /etc/rc /etc/rc-dist
 # ftp -o /etc/rc http://wtf.hijacked.us/random/OpenBSD-ro-rc/4.6-rc
 # cp /dev/MAKEDEV /etc/MAKEDEV
 # touch /var/run/ntpd.drift
 # touch /var/run/dhcpd.leases
 # touch /var/run/dhclient.leases.vr0
 # touch /var/run/dhclient.leases.vr1
 # touch /var/run/dhclient.leases.vr2
 # touch /var/run/dhclient.leases.vr3
 # ln -s /var/run/dhclient.leases.vr0 /var/db/dhclient.leases.vr0
 # ln -s /var/run/dhclient.leases.vr1 /var/db/dhclient.leases.vr1
 # ln -s /var/run/dhclient.leases.vr2 /var/db/dhclient.leases.vr2
 # ln -s /var/run/dhclient.leases.vr3 /var/db/dhclient.leases.vr3
 # ln -s /var/run/dhcpd.leases /var/db/dhcpd.leases
 # ln -s /var/run/ntpd.drift /var/db/ntpd.drift
 # pkill -9 cron
 # mv /var/cron /var/cron-dist
 # rm -f /var/cron-dist/tabs/.sock
 # cp -Rp /var/cron-dist /var/log/cron-running
 # ln -s /var/log/cron-running /var/cron
 # mv /var/named /var/named-dist
 # mkdir /var/named
 # cat /etc/fstab | sed '1,$s/\/ ffs rw/\/ ffs ro/g' >/etc/fstab
 # mv /home /cfhome
 # mkdir /home
 # cat /etc/syslog.conf | grep -v -e xferlog -e lpd-errs >/etc/syslog.conf
 # cat /etc/newsyslog.conf | grep -v -e xferlog -e lpd-errs >/etc/newsyslog.conf
 # reboot

There.. if you did everything correctly and didn't do anything too funny with your system
before setting up read-only-root, things should come up with *no errors during boot*.. good luck.

Jontow 16:10, 3 December 2009 (EST)