Live install via nfs fails filling the squashfs

[David Tangye]

Binary package hint: ubiquity-casper

I have set up a PC as a netboot (bootpd), tftp and nfs server. I have tried to install the unpacked Hardy desktop CD iso onto 2 PCs, and both fail late in the installation process. I have monitored the processes and kept almost all logs. Key info from them follows.

The PCs boot into the live desktop ok, using the casper kernels. When I start the install via the desktop icon df -h reports

Filesystem Size Used Avail Use% Mounted on
tmpfs 363M 16M 347M 5% /lib/modules/2.6.24-16-generic/volatile
tmpfs 363M 16M 347M 5% /lib/modules/2.6.24-16-generic/volatile
varrun 363M 100K 363M 1% /var/run
varlock 363M 0 363M 0% /var/lock
udev 363M 52K 363M 1% /dev
devshm 363M 12K 363M 1% /dev/shm
tmpfs 363M 16K 363M 1% /tmp
gvfs-fuse-daemon 363M 70M 294M 20% /home/ubuntu/.gvfs

After the packages are installed the gvfs-fuse-daemon filesystem suddenly fills, eg

gvfs-fuse-daemon 363M 363M 0 100% /home/ubuntu/.gvfs

At this time the relevant parts of ps -ef are:
ubuntu 10214 10212 3 21:52 ? 00:00:43 /usr/bin/python /usr/lib/ubiquity/bin/ubiquity gtk_ui
root 29652 10214 0 22:04 ? 00:00:00 /usr/bin/perl -w /usr/bin/debconf-communicate -fnoninteractive ubiquity
root 29654 10214 0 22:04 ? 00:00:00 log-output -t ubiquity --pass-stdout /usr/share/ubiquity/install.py
root 29655 29654 21 22:04 ? 00:01:33 /usr/bin/python /usr/share/ubiquity/install.py
root 30323 29655 27 22:10 ? 00:00:16 log-output -t ubiquity --pass-stdout /usr/share/ubiquity/apt-setup
root 30324 30323 0 22:10 ? 00:00:00 /bin/sh -e /usr/share/ubiquity/apt-setup
root 30341 30324 0 22:10 ? 00:00:00 /bin/sh /usr/lib/ubiquity/apt-setup/apt-setup /target
root 30367 30341 0 22:10 ? 00:00:00 /bin/sh /usr/lib/ubiquity/apt-setup/generators/50cdrom /target/tmp/filewYJNvx
root 30369 30367 4 22:10 ? 00:00:02 apt-cdrom add -o Dir::Etc::SourceList=/tmp/filecmbRrA

I think that the last process is looping and filling the squashfs. It might be something to do with it trying to register a real CDROM as a source for apt, when the iso image is actually coming via this line in the netboot file.

LABEL live
 kernel ubuntu-8.04-desktop/casper/vmlinuz
 append initrd=ubuntu-8.04-desktop/casper/initrd.gz boot=casper netboot=nfs nfsroot=192.168.0.42:/mnt/u01/tftpboot/i386/ubuntu-8.04-desktop --
ie, its getting its packages via nfs, not from a real CD device.

/var/log/syslog, messages & user.log are >80mB each, so apt-cdrom must be looping and writing the following error in those logs, repeatedly :
Jul 4 22:13:47 ubuntu ubiquity: Name: ''
Jul 4 22:13:47 ubuntu ubiquity: That is not a valid name, try again.
Jul 4 22:13:47 ubuntu ubiquity: Please provide a name for this Disc, such as 'Debian 2.1r1 Disk 1':

This repeated error loop writing to these logs is filling up the squashfs.

[Colin Watson (cjwatson)]

Looking at the code paths involved, I think you must have forgotten to copy over the (hidden) .disk directory from the unpacked CD. Doing that should avoid this bug.

Nevertheless, it's a legitimate bug and I'll fix it shortly. Thanks!

[Launchpad Janitor (janitor)]

This bug was fixed in the package apt-setup - 1:0.37ubuntu3

---------------
apt-setup (1:0.37ubuntu3) intrepid; urgency=low

  * Bail out cleanly from cdrom generator if /cdrom/.disk/info doesn't exist
    or has zero size (LP: #245519).

 -- Colin Watson <email address hidden> Sun, 27 Jul 2008 00:05:58 +0100