Debian Lenny was released on 14th February, 2009, so the following weekend I started to upgrade the various etch installations I have. There are two under VMWare on laptops, two native desktop and one native laptop instances.
I followed the Lenny
release notes, which worked flawlessly. I ignored the advice to
use aptitude and used apt-get instead. However,
once I rebooted using the new 2.6.26 kernel, it did not see the
hard drive (/dev/sda). This was due to the old version of VMWare I
was using. I upgraded from VMWare 1.0.3 to 1.0.6 and the emulated
hard drive was recognised by the new kernel.
The upgrade to lenny was very similar for my two desktop machines,
gold and chrome. Gold is a home-built machine, based on an ABIT
motherboard, from 2003. However, chrome is Dell Dimension p100t
from circa 1997, with a 486 processor, which led to a problem later. I
tackled gold first, as it is my main machine. Lilo was the boot
loader on both machines and, given that the upgrade notes indicated
there might be problems with the use of Lilo (the default settings
for initramfs will create an image too large for Lilo to
load), I took the opportunity to switch to grub prior to the
upgrade.
First, I ensured that /etc/apt/sources.list pointed to etch
and not stable, then made sure etch was completely up-to-date:
apt-get update apt-get upgrade
Next, installed grub:
apt-get install grub
Then created the default grub directory /boot/grub and ran
update-grub to create the initial menu.lst file.
On one system, the default groot had to be changed as the
Debian root partition was the second partition on the drive. In
addition, a Windows 2000 boot stanza was added:
# groot = (hd0,1) title Windows 2000 rootnoverify (hd0,0) makeactive chainloader +1
To ensure that the grub menu file was updated automatically as
kernels were added or removed, the following lines were placed in
/etc/kernel-img.conf:
postinst_hook = update-grub postrm_hook = update-grub do_bootloader = no
Grub was installed on the MBR by:
grub-install /dev/hda
Once this was tested, the old Lilo could be removed:
apt-get remove lilo
The final piece of preparation was to clean out any old package
files from the package repository under /var:
apt-get clean && apt-get autoclean
The /etc/apt/source.list file was changed to point at lenny:
deb ftp://ftp.uk.debian.org/debian/ lenny main contrib non-free deb-src ftp://ftp.uk.debian.org/debian/ lenny main contrib non-free deb http://security.debian.org/ lenny/updates main
OK, ready to go, so updated the package index files:
apt-get update
Next, as advised, attempted to upgrade apt itself first:
[root@gold:~]# apt-get install apt Reading package lists... Done Building dependency tree... 62% Building dependency tree... Done Some packages could not be installed. This may mean that you have requested an impossible situation or if you are using the unstable distribution that some required packages have not yet been created or been moved out of Incoming. Since you only requested a single operation it is extremely likely that the package is simply not installable and a bug report against that package should be filed. The following information may help to resolve the situation: The following packages have unmet dependencies: apt: Depends: libc6 (>= 2.7-1) but 2.3.6.ds1-13etch9+b1 is to be installed Depends: libstdc++6 (>= 4.2.1) but it is not going to be installed E: Broken packages [root@gold:~]#
Bugger. This didn't happen with the first lenny install. I decided to try with aptitude instead, with more success. Aptitude was upgraded, and pulled in the packages mentioned in the upgrade notes (and a few more beside).
The next stage was to upgrade those packages that did not require new packages to be installed:
aptitude search "?false" # convert aptitude's installed list for use with apt aptitude upgrade
This too, ran successfully, except that you need to ensure any NFS mounts are unmounted before this step. If you don't (as in my case), the upgrade process will complain and wait for you to umount the offending drives.
OK, the final phase:
aptitude dist-upgrade
Aptitude listed a lot of packages but I hit "Y" and let it run.
Returning a little while later, I was concerned to see that the disk
space in /usr was nearing the limit (2GB). That wasn't
good. Perhaps it had to do with the mass of texlive packages that
were being installed that I didn't want? As I feared, towards the
end of the dist-upgrade process the /usr partition became
full. I noted a couple of failures to install packages but, as by
this time, most activity was in /etc the damage didn't look
too bad.
Once the dist-upgrade had finished, there was one additional problem to deal with. The exim4 upgrade had borked:
Stopping MTA for restart:DEBCONFsomethingDEBCONF found in exim
configuration.
This is most probably caused by you upgrading to exim4 4.67-3 or
later without accepting the suggested conffile changes. Please read
/usr/share/doc/exim4-config/NEWS.Debian.gz for 4.67-2 and 4.67-4
2009-02-24 15:37:10 Exim configuration error in line 28 of
/var/lib/exim4/config.autogenerated.tmp:
malformed macro definition
Invalid new configfile /var/lib/exim4/config.autogenerated.tmp, not
installing /var/lib/exim4/config.autogenerated.tmp to
/var/lib/exim4/config.autogenerated
invoke-rc.d: initscript exim4, action "restart" failed.
Well, I wasn't going to deal with that now, so I removed exim4 and all its works by:
apt-get autoremove exim4 && apt-get purge exim4
The first part of recovery was to remove all the packages I did not
need, that's everything related to tetex and
texlive. That got me back to just under 1GB
free. Secondly, I wanted to reinstall everything remaining to ensure
it was installed correctly:
aptitude -y reinstall `dpkg -l|grep ^ii|awk '{print $2}'`
I let this run overnight. The log file revealed a number of packages that could not be re-installed due to the fact there were old packages left over from previous releases. These I could safely remove from the system, except:
linux-image-2.6.24-etchnhalf.1-686 # need a working kernel! msttcorefonts xfonts-artwiz
At this point I could reinstall exim4 and re-configure it, although leaving out the specials for local network mail delivery.
I then installed the lenny 2.6.26 kernel
linux-image-2.6.26-1-686 and rebooted to check that
everything was OK. The only thing I did not like was the new
default option to print an elapsed time against the kernel
messages. This can be removed by setting the following default
parameters in the /boot/grub/menu.lst file:
# kopt=root=/dev/hda2 ro printk.time=0
The last part of the upgrade was to add the new syslog function:
apt-get install rsyslog
Now that one desktop was upgraded, I turned my attention to chrome.
Here, I had the same problem as described above when attempting to
update apt, i.e broken dependencies with libc6. This time, I tried
a different tack, sticking with apt-get and trying to
install the packages it was complaining about:
[root@chrome:~]# apt-get install libc6 libstdc++6 Reading package lists... Building dependency tree... Done The following extra packages will be installed: binutils cpp cpp-4.1 cpp-4.3 gcc gcc-4.1 gcc-4.1-base gcc-4.3 gcc-4.3-base itcl3.1 libgcc1 libgmp3c2 libgomp1 libmpfr1ldbl libmudflap0 libncurses5 libselinux1 libslang2 locales tcl8.3 tzdata util-linux util-linux-locales Suggested packages: binutils-doc cpp-doc gcc-4.1-locales gcc-4.3-locales gcc-multilib automake1.9 gcc-doc gcc-4.1-multilib gcc-4.1-doc gcc-4.3-multilib libmudflap0-4.3-dev gcc-4.3-doc libgcc1-dbg libgomp1-dbg libmudflap0-dbg glibc-doc libc6-i686 tclreadline dosfstools Recommended packages: libgpm2 The following packages will be REMOVED: g++ g++-2.95 g++-4.1 itcl3.1-dev libc6-dev libmudflap0-dev libssp0 libstdc++2.10-dev libstdc++6-4.1-dev libtool tcl8.3-dev The following NEW packages will be installed: cpp-4.3 gcc-4.3 gcc-4.3-base libgmp3c2 libgomp1 libmpfr1ldbl The following packages will be upgraded: binutils cpp cpp-4.1 gcc gcc-4.1 gcc-4.1-base itcl3.1 libc6 libgcc1 libmudflap0 libncurses5 libselinux1 libslang2 libstdc++6 locales tcl8.3 tzdata util-linux util-linux-locales 19 upgraded, 6 newly installed, 11 to remove and 352 not upgraded. Need to get 25.1MB/26.0MB of archives. After unpacking 18.1MB disk space will be freed. Do you want to continue [Y/n]?
That update worked OK, so I tried to upgrade apt again:
[root@chrome:~]# apt-get install apt Reading package lists... Building dependency tree... Done The following extra packages will be installed: apt-utils aptitude dpkg dpkg-dev libapt-pkg-perl libcwidget3 libdb4.6 libept0 liblocale-gettext-perl libncursesw5 libtext-charwidth-perl libtext-iconv-perl libtimedate-perl libxapian15 lzma perl perl-base perl-modules Suggested packages: apt-doc python-apt debtags debian-keyring libcwidget-dev xapian-tools perl-doc libterm-readline-gnu-perl libterm-readline-perl-perl Recommended packages: aptitude-doc-en aptitude-doc libparse-debianchangelog-perl build-essential libgpm2 The following NEW packages will be installed: libcwidget3 libdb4.6 libept0 libtimedate-perl libxapian15 lzma The following packages will be upgraded: apt apt-utils aptitude dpkg dpkg-dev libapt-pkg-perl liblocale-gettext-perl libncursesw5 libtext-charwidth-perl libtext-iconv-perl perl perl-base perl-modules 13 upgraded, 6 newly installed, 0 to remove and 339 not upgraded. Need to get 20.0MB of archives. After unpacking 16.2MB of additional disk space will be used. Do you want to continue [Y/n]?
OK, much better. Now the upgrade was back on track. Everything else
proceeded as normal, and no massive texlive install
occured. Of course, that may have be due to cruft left over on
gold. However, even chrome had some stuff installed I didn't want,
e.g. X11, sane and the avahi daemon.
Following the completion of the upgrade, I cleaned-up the packages
using apt-get autoremove:
The upgrade had automatically installed the lenny kernel, 2.6.26, so I confidently rebooted into it. Ulp, no IDE devices:
[ 21.065487] Uniform Multi-Platform E-IDE driver [ 21.065749] ide: Assuming 33 MHz system bus speed for PIO modes; override with idebus=xx [ 22.320593] PIIXa: IDE controller (0x8086:0x122e rev 0x02) at PCI slot 0000:00:01.0 [ 22.321073] PIIXa: bad irq (0): will probe later [ 22.321259] PIIXa: IDE port disabled [ 22.321401] PIIXa: IDE port disabled
More worryingly, when I attempted to boot using the latest etch kernel, linux-image-2.6.18-6-486, installed as part of the final etch upgrade, that produced the same problem. Was their some co-incidental hardware problem? No, because moving back to linux-image-2.6.18-5-486 booted OK. This problem has already been reported.
My last upgrade was to the old Toshiba Satellite 4060 XCDT laptop, steel. This did not manifest the same problem as the two server machines when trying to upgrade apt as the first step. My theory is that both gold and chrome had been previously upgraded from sarge to etch, as the two machine with no apt issue had been fresh etch installs.
Apart from having to fix up the exim4 configuration after the upgrade, there were two weird issues to deal with on steel, post-installation.
The first was generated by apache2, a message of the form:
apache2: Could not reliably determine the server's fully qualified domain name, using 127.0.0.1 for ServerName
After a bit of digging around, this turned out to be a FAQ on the apache web site. The issue can be solved by either:
127.0.0.1 in the /etc/hosts fileServerName F.Q.D.N directive in the
/etc/apache2/httpd.conf file, which is used to hold local
settings.The second problem was intermittent. Sometimes, on boot, steel would hang, for about three minutes, just after printing the fsck messages:
Checking all file systems. /dev/hda8: clean, 10667/3662848 files, 381431/7323624 blocks /dev/hda9: clean, 4475/4833280 files, 4400887/9649032 blocks /dev/hda6: clean, 13/122880 files, 26056/489948 blocks /dev/hda2: clean, 47825/488640 files, 254580/975948 blocks /dev/hda5: clean, 79/366528 files, 29122/732957 blocks /dev/hda7: clean, 5485/244320 files, 271041/487966 blocks
There was nothing I could find about this via Google. In an attempt
to gather further information, I changed the default in the
/etc/default/rcS script to VERBOSE=yes. Of
course, the problem does not now occur. I have no idea what the
problem is, but at least I have a fix.
I noticed there were other anomalies with steel; sometimes the sound
card would not be detected by alsa, and the fan would come on, never
to stop. I messed around with toshiba_acpi module, which
allowed me to turn off the fan with:
echo "force_on: 0" >/proc/acpi/toshiba/fan"
However, the fan would always come back on, even when the CPU was
idling. I finally turned off acpi on the lenny kernel, 2.6.26, as
seemed to be the default with the old etch kernel 2.6.18, by adding
acpi=off to the kernel boot line. Instant sane behaviour;
none of the problems recurred, including the hang on fsck with
VERBOSE=no in /etc/default/rcS.
