[Cc'ed into the RedHat mailing list as this might be useful to others
listening there... just beware that if you reply back into that list
then you'll need to be a subscriber for it to be successful.]
On Wed Jul 28 1999 at 00:53, Rusty Brooks wrote:
> Ok, upgraded today (via ftp) and have some BIG problems. Seems like
> there's always bugs in the upgrade routines.
>
> First off, /dev/console had the wrong major/minor number. The symptoms
> were that I coudln't see the results of any startup scripts! I changed
> this to 4,0 which I think is correct. i believe it was 6,1 before!
>
> OK, so I figured that would fix everything. Nope.
From my own considerable experience with building and rebuilding
linux boxes (hundreds by now), I'd like to offer some advice about
upgrading. What I say here definitely applies to redhat, but very
likely also to most other distributions that have an "upgrade"
option during the install.
My advice is simple: DON'T upgrade. NEVER upgrade. Re-install
instead. Why do I say this? Let me explain...
When setting up a linux box, create your partitions so that there is
at least a /home partition on its own partition, and that it is as big
as possible. This filesystem will act as your "backup" target.
Move everything in /usr/src to /home/src and create a symbolic link
from /usr/src to the new location.
Do the same with /usr/local - move it to /home/local and make a
symlink.
Have a cron job each night make a complete copy (cp -a or even a tar
zcf) of the entire contents of /etc into /home/etc/. Do the same with
/root into /home/root. On redhat boxes, copy (or move)
/tmp/install.log into /root before you do this. (Make sure the
permissions are set securely on these directories in /home, especially
on a multi-user box).
If you have any customising, like /opt and so on, save copies of them
onto the /home partition too. The principle with most linux
distributions is that you leave the main filesystems (like /usr /sbin
/bin and so on) alone, and only let "rpm-managed" files to go into
there. All "custom-built" software should be built in /usr/src/ and
installed (as much as possible) into /usr/local/.
Then back up /home - this partition now contains *everything* that is
needed to recover the box should it need ever need rebuilding. Don't
bother about backing up places like /bin /sbin /var (with exceptions
like /var/yp and /var/named - back these up too if you need them!) -
it's always possible to recover all this with your linux installation
disk. Why back it up when you already have it?
Here we are wanting to talk about upgrading rather than backup
recovering, but the principles are the same. But - as I say - don't
upgrade, re-install. Choose the install option from the installation
program and *reformat* every partition that the installation will go
into --- except the one mounted on /home ! Wipe the slate completely
clean. This way you end up with a fresh filesystem that is almost
guaranteed to work as expected out of the box.
BTW, specify that partition to the install program to mount it as
something *other* than /home (like /oldhome) so that none of your old
stuff (/home/httpd/ /home/ftp/ /home/samba/ and so on) gets mashed
over during the process. After you first bootup,
What happens with an upgrade is that every package that is already
installed is replaced by the new version. That's not so bad, but
other "extra" software packages you might have also installed get left
behind, along with all sorts of bits and pieces.
And as is the case with RH6.0, the jump from RH5.x is a big one -- a
brand new C compiler with a brand new set of runtime libraries that
are incompatable with the old binaries. Not to mention a brand new
kernel (2.0.x to 2.2.x).
Using the upgrade option when installing RH6 onto a RH5 system just
doesn't work. Believe me, it doesn't, no matter what they say. Been
there done that. Don't do it.
One thing I was noticing with the upgrades I had initially been trying
is that the /var/lib/rpm/ database would often (usually) become
corrupted after the upgrade. (I note with interest that there have
been some bugfix packages for the rpm utility recently issued for both
RH5 and RH6 :)
> Kerneld wasn't part of the new rc.sysinit. So I put it in like in the old
> one. Kerneld does load, but no modules are automatically loaded! I can
Don't do this. The 2.2.x kernels DO NOT NEED kerneld! This daemon is
now a standard part of the kernel is isn't needed.
Undo what you did.
> load them with insmod OK. I get some error from modprobe on startup but I
> can't find it at this time and I obviously can't reboot at the moment.
You didn't read the upgrade docs, didn't you? :)
> X doesn't work at all. I am using AcceleratedX so -probeonly doesn't
> actually do anything but start X. I get a blank screen that I Can't exit
> out of. I get a messagesaying that it can't set the default font path
> (all the dirs do exist) and that it can't find the default font, fixed.
> It won't start without it.
Two reasons that this may be happening. The version of Accel-X that
are using could be binary incompatable with the RH6 shared libraries.
It's not actually segfaulting (not that you've mentioned), so it's
less likely the cause.
RedHat 6.0 now has /usr/X11R6/bin/startx exec xinit with the --auth
flag to make it more secure. It also creates the X authority for you,
a new one each time it is run.
Try rm'ing ~/.Xauthority and running startx again. If this doesn't
work, don't bother again until you've done a re-install.
> Rebooting (reboot or halt) tries to unmount but I get errors about
> "umount2" which I can find no reference to in the rc files! As a
> result, I think my partitions are not unmounted because they are
> always marked as dirty. The message is
> "Unmounting file sysetms umount 2: Function not implemented"
Not sure what's going on here, but mount is one of the packages that
needed upgrading if you wanted to use a 2.2 kernel on a RH5.2 system.
Makes me suspicious that something's gone wrong here.
Hey, I have a very strong suspicion that you're booting up with your
old 2.0.36 kernel, which wouldn't know about the new functionality
that is expected of it by your new utilities! This might also explain
some of the weirdness with modprobe.
> Help! I have a barely working system at the moment!
There's very likely to be more grues lurking about the place than the
ones you've mentioned here.
Save off your valuable data and config files, then install it again -
and this time do a complete reformat of the partititions where the
main filesystems will be installed.
After this, all should be sweet :)
Cheers
Tony
-=*#*=-=*#*=-=*#*=-=*#*=-=*#*=-=*#*=-=*#*=-=*#*=-=*#*=-=*#*=-=*#*=-
Tony Nugent <[EMAIL PROTECTED]> <[EMAIL PROTECTED]>
Computer Systems Officer Faculty of Science
University of Southern Queensland, Toowoomba Oueensland Australia
-=*#*=-=*#*=-=*#*=-=*#*=-=*#*=-=*#*=-=*#*=-=*#*=-=*#*=-=*#*=-=*#*=-
