Installing with d-i: summary and notes
Hey guys
As a follow up to Tollef Fog Heen's debian-installer status -- 2002-10-14
I decided to try out the new install system and send some feedback. There
is probably some sorting to do, some things are really minor details and
my own opinions on stuff. Some other reflect more general issues.
First I can't say that I've been successful, as I wasn't able to complete
a full installation process and had to switch back to regular netinst
burned on a CD-RW [0] [1] (bootbf2.4.iso).
Anyway, grabbed the floppy img net-1440.img [3], and booted up. A single
floppy, so easy to get started (instead of the usual 4-5 floppies CD set).
Auto-detect and config of my network hardware (3c59x) went smoothly.
Then the first step that caused trouble was '1 - Finish setting up the
debian-installer'. This steps asks for hostname of the debian-installer
repository, used people.debian.org/~tfheen/d-i, then the path, instead of
default debian/ I used /. (I guess I could just have entered
people.debian.org and /~tfheen/d-i/ instead)
Anyway the main problem is that once the params are entered, in case they
are wrong there is no way to enter them again. Use Alt-F1 or use Step 5 -
Open a shell to issue the reboot command, do network conf again, then run
through step 1 and try better params.
You get spammed by messages like:
kmod: failed to exec /sbin/modprobe -s - net-pf-1, errno = 2
that's noisy but harmless. see [4]
Once Step 1 is complete, Step 2 - Partition the hard drive is not working
atm. You won't see your HD devices until devfs has done his probing job.
Run step 7 first. I got a bunch of unresolved messages from the pcmcia
modules, harmless.
Need to setup partitioning manually. Open a shell, find the drives you
want to partition in /dev (with devfs, it can be something like
/dev/ide/host0/bus0/target0/lun0/disc). At this point you realize that
some basic command history and tab completion in ash would be awesome. Run
fdisk, then mkfs.ext2 (or mkfs.ext mke2fs). mkswap is not available to
create a swap partition, would have to mkswap and swapon later when the
base system is running.
Next is to install the base system. I don't remember the names of the
steps, but you have choice between 3, Mounting target partitions,
installing base system, and installing lilo. Those didn't seem to work
quite well. I went directly to 'installing base system', it spawned fdisk
(exited without doing anything) then prompted for partition to mount as /
(I use a single / partition, no /var /boot /home partitioning).
Installation of the base system asks wether to install Woody, Sarge or
Sid. If you try sid, the download will fail at some point with an error
about libpcap. Just use Sarge.
Once base is installed, LILO setup. The direct step in the text UI can be
used, or directly issuing the command chroot /target liloconfig in the
shell. The strange thing is that I didn't end up with the same
/target/etc/lilo.conf wether I used one solution or the other.
If rebooting now, none of the two LILO setup solutions would work. Kernel
panics with VFS: Unable to mount root filesystem. Tollef said to add an
initrd=/initrd.img to the lilo.conf. You can do this by echo
initrd=/initrd.img >> lilo.conf (in /target/etc) or use nano to edit the
file.
nano didn't start because of missing libslang.so.1. did export
LD_LIBRARY_PATH=/target/lib and it worked.
then, chroot /target lilo to update with the modified lilo.conf
Last step is to reboot and start on the base install. This is were I got
stuck. No more VFS panic though, but mounting root complaining that it
doesn't understand some options of the filesystem. Then I have a kernel
line 'Checking root filesystem'. fsck complains about filesystem having
'unsupported features'. I end up with the root filesystem mounted as
readonly, asking me to run fsck manually and reboot.
Tollef told me that this is the sign of a badly unmounted partition. But I
tried a lot of combinations of e2fsck etc., and afaik I got a clean
partition. Got the same error in all my attempts.
That was about when I ran short of time and dropped the idea of doing a
full complete install. Doesn't mean that someone with a bit more skills
than me can't do it.
Some general thoughts:
The new architecture looks good. Probing hardware, devfs usage, loading
modules on an as-needed basis. Very small start footprint. All the
retriever options (net, cdrom etc.).
The install steps are still a bit incomplete. As I used
people.debian.org/~tfheen/d-i each time I had to re-get the d-i base
stuff, it was painfully slow between each try.
FB based UI. Using a good FrameBuffer UI is the next step obviously. The
fb stuff that's in [1] is quite good already. Something even better can
probably be done.
i18n? Not relevant to me in particular, but you'll have to deal with i18n
of the installer at some point.
cheers
TTimo
0. http://www.debian.org/CD/netinst/
1. http://people.debian.org/~dwhedon/boot-floppies/
3. http://people.debian.org/~tfheen/d-i/
4. http://bugs.debian.org/163748
Reply to: