Conditional partitioning in debian installer
I needed to create an automatic, unattended installer that would
repartition the whole disk, but keep an existing home partition if
it is already existing. The idea is to replace the operating
system, but keep the user's home files. If the home partition does
not exist, however, we need to create it.
The starting point for this is
in Holger's
website, which does everything except create the partition if
it's missing.
The technique I used is to
create a udeb that would scan the partitions using
/usr/lib/partconf/find-partitions and then load the
right preseeding for
partman-auto, the automatic
partitioner.
Here you find the
udeb source. I'll comment the most important bits:
First, the two
partman-auto recipes that we are
going to preseed. The format
is documented in the debian-installer documentation.
This will create
/home if missing:
d-i partman-auto/expert_recipe string \
condpart :: \
2048 2048 2048 ext3 \
$primary $bootable \
method format format \
use_filesystem filesystem ext3 \
mountpoint / \
. \
1024 1024 1024 linux-swap \
method swap format \
. \
1024 4096 10000000 ext3 \
method format format \
use_filesystem filesystem ext3 \
mountpoint /home \
.
And this will keep
/home if it exists:
d-i partman-auto/expert_recipe string \
condpart :: \
2048 2048 2048 ext3 \
$primary $bootable \
method format format \
use_filesystem filesystem ext3 \
mountpoint / \
. \
1024 1024 1024 linux-swap \
method swap format \
. \
1024 4096 10000000 ext3 \
method keep \
use_filesystem filesystem ext3 \
mountpoint /home \
.
Then the
postinst file of the package, where the
logic is. Otavio and Daniel on
#debian-boot have
taught me that udebs are, simply put, just big postinsts (installer
wise, apart from shipping binaries). Postinst is where the logic is
done, or applications are started.
set -e
FINDPARTS="/usr/lib/partconf/find-partitions"
notice ()
logger -t condpart "$@"
error ()
logger -t condpart "error: $@"
. /usr/share/debconf/confmodule
notice "Checking if we need to recreate /home"
if [ ! -x $FINDPARTS ]
then
error "$FINDPARTS not found"
exit 1
fi
if $FINDPARTS grep -q part6
then
notice "/home exists: we keep it"
debconf-set-selections /usr/lib/condpart/part-keep.preseed
else
notice "/home does not exist: we create it"
debconf-set-selections /usr/lib/condpart/part-create.preseed
fi
notice "Finished special handling of /home"
exit 0
Finally,
debian/control. The basic bits about
creating a d-i udeb are to put it in
Section:
debian-installer and to use
XC-Package-Type:
udeb. Then there is
XB-Installer-Menu-Item,
that defines
the order of execution of postinsts. The d-i
documentation
lists the numbers that are currently assigned.
Source: condpart
Section: debian-installer
Priority: optional
Maintainer: Enrico Zini <enrico@debian.org>
Build-Depends: debhelper (>= 4.0.0), cdbs
Standards-Version: 3.7.2.1
Package: condpart
Architecture: any
XC-Package-Type: udeb
XB-Installer-Menu-Item: 38
Description: Conditional partitioning
Tells partman-auto to create /home if it's missing, or if it exists, to keep
it.
I am using 38 rather than 3800 because I'm targeting etch d-i.
After etch, "numbers have been multiplied by 100 from what they
were before to give us more space".
This is it.
debian/rules is just boilerplate,
except installing the preseed files in the right place. CDBS is
useful to make this self-evident:
include /usr/share/cdbs/1/rules/debhelper.mk
install/condpart::
install -o root -g root -m 0644 *.preseed debian/$(cdbs_curpkg)/usr/lib/condpart/
To build the installer CD with this d-i, I'm using
simple-cdd;
here are the
profile files I'm using at the moment, with the main preseeding for
fully unattended install (it is designed to trash your hard
disk if you boot it, so only test it in an emulated
environment) and the inclusion of the custom udeb.
There's only one week left for D-I translation updates. I
wholeheartedly thank Otavio who agreed for such a long string freeze
as it allowed me to do my best to get translations completed. And this
is a huge communication effort, considering how easily the translators
vanish...
46 languages are now qualified for the release, without discussion.
Since my last blog entry, welcome back to Greek, Esperanto, Hungarian, Norwegian Nynorsk, Wolof.
There are still uncertainties for 14 languages:
Amharic, Bengali, Bosnian, Catalan, Danish, Estonian, Kurdish,
Latvian, Macedonian, Northern Sami, Slovenian, Albanian, Tagalog,
Ukrainian
Again, several of these have *very few missing strings*.
In short, if your language is among these 14, there are 6.5 days left
(the end of the translation update round is on Saturday 20th midnight
UTC). RUSH!
What will happen in 6.5 days?
I will examine the situation of each of those languages and discuss it
with the D-I release managers. On a case by case basis, we will decide
together to keep them or not.
Etch has 58 supported languages (English included). Lenny currently
has 47 that are sure. Will Lenny be the first Debian release ever with
*less* supported languages in D-I than the former release? It would
give some groung to my claim that Debian translators are currently
escaping somewhere else (guess where?).
Otavio has updated kernel-wedge in SVN to 2.6.26, so I updated ARM and MIPS
today. The udebs built after minor modifications and I can also build
installer images after having fixed some bugs in the build process. I
tested a 2.6.26 based image on the QNAP TS-409 (armel) as well as in QEMU
(mipsel).
2.6.26 will bring major improvements on ARM. Orion support is much better
and our 2.6.26 has a number of patches from Marvell that help with
performance on Orion. Our 2.6.26 also has Riku Voipio's LED driver for
Thecus N2100 which has been accepted for the 2.6.27 mainline kernel.
I haven't actually uploaded the udebs yet because I'm waiting for the
go-ahead as well as for the new version of the 2.6.26 kernel package to
build. But at least we're getting closer to having 2.6.26 as the kernel
for lenny.
Another point release for Etch has been done; now it's the time for the CD
team to roll out new images after the next mirror pulse. The official
announcements (prepared by Alexander Reichle-Schmehl, thanks!) will follow
shortly afterwards. FTP master of the day was Joerg Jaspert, who
did his first point release since Woody, as he told us on IRC. We
appreciate your work and you spending your time that shortly before
going to Argentina.
This point release includes the etchnhalf update introducing a new kernel
image (based on 2.6.24) and some driver updates. Additionally the
infamous openssl hole will be fixed for good, even for new installs.
Again I want to present you a list of people who contributed to this
release. It cannot be complete as I got the information out of the
It’s nice to be able to count on friends across vendors. The two new Parted Team members, Joel Andres Granados Moreno and Bryn Reeves from Red Hat, has been working with me to solve the s390 error that was blocking Debian to move to latest parted release (1.8.8 up to now).
They’ve had a hard time to identify the real problem and we helped each other for those past two weeks. That’s awesome to be able to share the work load and be able to get help to solve problems. Even more when this is done for a very important software like Parted.
I do believe this is the way to go for Free Software projects… maybe others could learn from that?
Cheers
I have never seem so many people at fisl. Seriously! All the rooms were crowded, all the time same as the expo and the user-group area. Our official counter were stopped at 6667 participants, but that is just because we ve taken the registration database off-line in order to process new registrations on-site. By the end of fisl all the numbers will be summed up, but I heard rumors of 7 thousand people! Awesome!
I spent this first day mostly in the 
And, since I was not available in the morning, when the huge crowd was arriving, I took a picture of a picture taken by Marlon of this initial moments:
(It is not really clear in this picture, but beyond the red gates there were as many people as inside)
If you're interested in working upon your CV/Resume, as 
I ended changing Adrew’s style to a class and writing a cuple of Makefiles to automate all this. All well done 
I suggest that anyone interested on it to take a look on the