Bringing up 4.3BSD, a half-step at a time

[Roy Smith]

	Did you ever do something that you thought was really clever and
just couldn't wait to tell the world about it.  Well, if you have, you'll
know why I'm writing this.  Maybe I'm not the first person to think of
this, maybe it's even obvious, but I thought it was pretty neat.  Anyway...

	We're currently running 4.2BSD on a Vax with a single RA-81 disk.
When I got the 4.3 tapes, I started to realize it might not be so trivial
to get it running.  How to change to a new version of the OS while
minimizing downtime on a production machine?  The installation guide
mentions some of what you have to do, but it's not that easy if you only
have a single disk; you can't just load up 4.3 on ra1a while keeping a
working copy of 4.2 on ra0a.  It's easy to have kernels for both systems on
your root partition so you can boot one or the other, but you'll have the
wrong /etc and /bin directories if you just do it that way.

	Things are complicated because the disk partitions we use match
neither those distributed on the 4.2 or 4.3 tapes.  This means, for
example, that the stand-alone boot and copy programs won't work with
anything other than ra(0,0).  We've got a 200 Mbyte user partition (ra0h),
2 100 Mbyte partitions (ra0d is /usr, ra0f is random overflow), 2 8 Mbyte
partitions (ra0a is root and ra0e is /tmp) and a 16 Mbyte swap partition
(ra0b) (all sizes approximate).

	First, I punted on having lots of /tmp space and made ra0e into
/4.3root.  /4.3usr got put on ra0f.  Then I followed the directions and
merged our 4.2 setup and configuration files in the 4.3 file systems.  At
this point, I still couldn't easily boot 4.3 because its root is not on an
"a" partition.  To switch back and forth between 4.2 and 4.3 would have
been a commotion because you have to get the standalone copy program up,
then get the mini-root up, and use that to copy the system you want to boot
to ra0a, making a level 0 dump of what's there now so you can go back to it
if there is trouble.

	The trick is that the root file system resides on partition "a",
aka minor device 0.  Nobody says, however, that minor device 0 has to start
at cylinder 0.  So, I built two versions of 4.3, one with the partition
table in the uda driver matching the one we use for 4.2, and the other with
the entries for the "a" and "e" partitions switched around (this was done
while 4.2 is still running, after some minor problems with the make file).
I put a copy of each kernel in both ra0a and ra0e, using the same names on
both file systems.  In the /etc/fstab in the 4.2 root partition, I mount
ra0d as /usr, ra0f as /4.3usr, and ra0e as /4.3root.  In the /etc/fstab on
the 4.3 root partition, I mount ra0f as /usr, ra0d as /4.2usr, and ra0e as
/4.2root.  There are various symlinks in ra0f pointing places in ra0d; for
example, /4.3usr/local is a symlink to /4.2usr/local.

	The end result is as follows.  If you boot ra(0,0)vmunix, you get
the 4.2 kernel with the normal 4.2 root and /usr file systems mounted where
they should be, and /4.3root and /4.3usr available on-line for easy access.
If you boot ra(0,0)/4.3vmunix.swapped, you get the 4.3 kernel running with
the 4.3 root and /usr partitions, with the 4.2 counterparts on-line and
ounted in convenient places.  In this latter case, the file you booted and
the file known as "/4.3vmunix.swapped" are two distinct, but identical,
files.  If you boot ra(0,0)4.3vmunix.normal, you get the 4.3 kernel with
the 4.2 root and /usr partitions mounted; I'm not at all sure why you would
want to do this, but you can if you want.

	Of course, this is far too confusing for normal use.  Once we are
sure we have 4.3 running the way we want it, I can just copy ra0a to ra0e
(which as far as a normal partition table is concerned means copying ra0e
to ra0a) and reboot ra(0,0)4.3vmunix.normal.  At my leisure, I can delete
the redundant stuff from /4.2usr and mount that file system as /usr/local,
and we're all done.

	The bottom line is that with a single disk drive, I'm able to flip
back and forth between the 4.2 and 4.3 in literally just as long as it
takes to do a normal reboot.  Just one more reason why "if I can't get
source, I don't want it".  Unfortunately, if Berkeley ever gets around to
writing the partition tables on the disks "like it should be", this scheme
will break, but you can't have everything, can you?
-- 
Roy Smith, {allegra,cmcl2,philabs}!phri!roy
System Administrator, Public Health Research Institute
455 First Avenue, New York, NY 10016

"you can't spell deoxyribonucleic without unix!"