20.04 autoinstall

pacman366

I wasn’t able to get 20.04 to pxe boot using the traditional nfs methods outlined in these forums. I did get it booting with the new 20.04 approach. I copied ubuntu-20.04-live-server-amd64.iso to /var/www, mounted the iso, copied the kernel and initrd from the iso to /tftpboot/os/ubuntu/Server20.04/ and create the following menu entry using these parameters

kernel tftp://${fog-ip}/os/ubuntu/Server20.04/vmlinuz
initrd tftp://${fog-ip}/os/ubuntu/Server20.04/initrd
imgargs vmlinuz initrd=initrd root=/dev/ram0 ramdisk_size=1800000 ip=dhcp url=http://172.16.13.3/ubuntu-20.04-live-server-amd64.iso ro
boot

One note when using this method, it only works on systems with >= 4G ram because it loads the whole iso into ram. Otherwise it works pretty well for me. Desktop and Mint variants of 20.04 also boot using this method.

londonfog

@pacman366 this is pretty similar to how I’ve gotten it to “work” Have you tried setting up the autoinstall method? I just started to attempt this today, and no luck so far.

george1421

@pacman366 said in 20.04 autoinstall:

url=http://172.16.13.3/ubuntu-20.04-live-server-amd64.iso

I haven’t had a chance to work on this yet. But I did help someone get a HP server update iso to pxe boot. It was a problem but we got it to work via pxe to nfs. The point is that url= reminded me of that experience. In that case we had to provide the uri in a format similar to this url=nfs://${fog-ip}/images/os/ubuntu/20.04/ it kind of replaced the url=http:// protocol with the nfs uri url=nfs:// I don’t know if that is what ubuntu did here or not.

pacman366

@londonfog I’ve also gotten autoinstall to work with the following imgargs

imgargs vmlinuz initrd=initrd root=/dev/ram0 ramdisk_size=1800000 ip=dhcp url=http://172.16.13.3/ubuntu-20.04.1-live-server-amd64.iso net.ifnames=0 autoinstall ds=nocloud-net;s=http://172.16.13.3/2004/ ro
boot

172.16.13.3 is my fog servers ip address, emphasis on this string

autoinstall ds=nocloud-net;s=http://172.16.13.3/2004/

That part of the sting tells subiquity were to find the user-data and the meta-data files. http://172.16.13.3/2004/user-data has the yaml containing all the autoinstall parameters, http://172.16.13.3/2004/meta-data is just a blank file. Both files need to exist in order for subiquity to perform an autoinstall. subiquity is also extremely strict about the formatting of the yaml within the user-data file. I found this out when attempting to copy/paste another user-data example from the internet only to crash the live installer. I had to hand write all the parameters in user-data in order for it all to work. this is what my user-data file looks like.

#cloud-config
autoinstall:
  interactive-sections:
    - storage
  apt:
    geoip: true
    preserve_sources_list: false
    primary:
    - arches: [amd64, i386]
      uri: http://us.archive.ubuntu.com/ubuntu
    - arches: [default]
      uri: http://ports.ubuntu.com/ubuntu-ports
  identity: {hostname: 69changeme69, password:  PASSWORD HASH,
    realname: companyadmin, username: companyadmin}
  keyboard: {layout: us, toggle: null, variant: ''}
  locale: C
  network:
    ethernets:
      eth0: {dhcp4: true, dhcp-identifier: mac}
    version: 2
  ssh:
    allow-pw: true
    authorized-keys: []
    install-server: true
  packages:
    - ubuntu-desktop
    - landscape-client
    - openjdk-8-jdk
    - libpwquality-tools
    - wpasupplicant
    - python
    - python-dbus
    - python-argparse
  late-commands:
    - | cat <<EOT >> /target/lib/systemd/system/postinstall.service
      [Unit]
      Wants=network-online.target
      [Service]
      Type=oneshot
      ExecStart=/opt/postinstall.sh
      RemainAfterExit=true
      [Install]
      WantedBy=multi-user.target
      EOT
    - systemctl daemon-reload
    - wget http://172.16.13.3/2004/postinstall.sh -P /target/opt/
    - chmod +x /target/opt/postinstall.sh
    - echo 'companyadmin ALL=(ALL) NOPASSWD:ALL' > /target/etc/sudoers.d/companyadmin
  version: 1

This installs mostly automatic. I have mine set to prompt for partitioning. Hope this helps

londonfog

@pacman366 This is great, thank you! I had some other projects to work on, so I hadn’t touched this in a while. Where did you get your vmlinuz and initrd files from? Did you extract from the iso?

fogman4

Hi .

I’m wondering is there anyone succeeded in fully automated iPXE ubuntu 20.04 install via fog.

I tried many parameters with no success.

It seems url= is only for http and ftp am i wrong ?

By the way i managed to boot Fedora workstations , Debian via a regular pxe with grub chainloaded in EFI Secure boot with the shim.signed.

Is there a way to chainload grub2 in fog ? It could be the easiest.

george1421

@fogman4 Very close to what you are asking https://forums.fogproject.org/topic/15129/preseeded-unattended-netboot-uefi-debian-installation Since ubuntu is based on debian the logic should be pretty close.

It seems url= is only for http and ftp am i wrong ?

You are correct

Is there a way to chainload grub2 in fog ? It could be the easiest.

Yes you can, but to what end?

To leave secure boot enabled? If yes, that will work until you attempt to run a kernel that has not been signed then booting will stop.

fogman4

@george1421 Thank you very much for those suggestions. I’ll dig and share a way to do it.

My purpose is to do a Fully automated ubuntu 20.04.2 Desktop (UEFI) installation via FOG’s iPXE.

I got more than 100 Workstations and need to make it viable with an Ansible post-installation.

In the first time i wanted to do it with preseed but autoinstall seems to be the new shit so i’ll try it out.

To do Secure boot with Ubuntu 20.04 you’re right we need the signed kernels.

I need to RTM and do some tests.

Don’t you think chainloading signed grub via iPXE in FOG could be a more efficient and easier way to do tasks centralized over all grub’s options ?

For example those files (chainloaded) could boot via GRUB2 a secure boot installation for example :

grubx64.efi.signed

grubnetx64.efi.signed

Regards.

george1421

@fogman4 The problem is that neither iPXE or FOS Linux are signed. That is where the problem comes in.

I can PXE boot a signed kernel using GRUB. with the shimx64.efi and grubx64.efi signed. Then grub loading the kernel. But that won’t help with doing anything with FOG.

faboulous

Hi I run into the same issue trying to boot a linuxmint 20 (and also ubuntu 20). After some digging, it appears that the Unable to find a live file system on the network show up because do_nfsmount() from the casper script chipped inside the initrd isn’t able to mount correctly the nfs share.

I extracted (using unmkinitramfs) the nfsmount binary to test it standalone on a working nfs share using the same option provided by this casper script (nfsmount -o nolock -o ro ${NFSOPTS} ${NFSROOT} ${mountpoint}) which work well.

I also tried to use the initrd from a linuxmint 19 to load the linuxmint 20 iso and… it worked but without any big surprise I ran into issue while I tried to install the os.

The differences between mint19 and mint20 are some check performed after nfsmount call:

do_nfsmount from mint 20

do_nfsmount() {
    rc=1
    modprobe "${MP_QUIET}" nfs
    if [ -z "${NFSOPTS}" ]; then
        NFSOPTS=""
    else
        NFSOPTS=",${NFSOPTS}"
    fi

    [ "$quiet" != "y" ] && log_begin_msg "Trying nfsmount -o nolock -o ro ${NFSOPTS} ${NFSROOT} ${mountpoint}"
    # FIXME: This while loop is an ugly HACK round an nfs bug
    i=0
    while [ "$i" -lt 60 ]; do
        if nfsmount -o nolock -o ro${NFSOPTS} "${NFSROOT}" "${mountpoint}"; then
            if is_casper_path $mountpoint && matches_uuid $mountpoint; then
                rc=0
            else
                umount $mountpoint
            fi
            break
        fi
        sleep 1
        i="$(($i + 1))"
    done
    return ${rc}
}

do_nfsmount from mint 19

do_nfsmount() {
    rc=1
    modprobe "${MP_QUIET}" nfs
    if [ -z "${NFSOPTS}" ]; then
        NFSOPTS=""
    else
        NFSOPTS=",${NFSOPTS}"
    fi

    [ "$quiet" != "y" ] && log_begin_msg "Trying nfsmount -o nolock -o ro ${NFSOPTS} ${NFSROOT} ${mountpoint}"
    # FIXME: This while loop is an ugly HACK round an nfs bug
    i=0
    while [ "$i" -lt 60 ]; do
        nfsmount -o nolock -o ro${NFSOPTS} "${NFSROOT}" "${mountpoint}" && rc=0 && break
        sleep 1
        i="$(($i + 1))"
    done
    return ${rc}
}

I tried to add some debug to identify what is going wrong but I couldn’t manage to “repack” the extracted initrd into a clean working one ending up on a kernel panic (I spend quite some time trying out, initrd archive structure seems to have recently evolve to chip intel and amd firmware inside). The only changed I made was adding some echo into the casper file. If someone manage to do so, it would be interesting to check what is_casper_path and matches_uuid does and return.

As I wasn’t able to produce a custom initrd, I tried to configure a http server to serve the iso on similar way as suggested on the comment bellow by @pacman366. But ubuntu/mint desktop iso are quite big (2GB) and 4GB ram isn’t enough to be able to extract the iso. I also tried to boot from http following these instructions: https://www.plop.at/en/ploplinux/live/networkboot-linux.html#pxel61 without any success. I’m not sure how | in the url work (no request received on the nginx web server).

I’d really like to be able to fix this, any idea/help would be appreciated

Sebastian Roth

@faboulous said in 20.04 autoinstall:

If someone manage to do so, it would be interesting to check what is_casper_path and matches_uuid does and return.

I don’t have a setup to dive in this right now and so I just started looking at the code. But maybe that’s of any help to you too.

~~Looking at the definition of is_caspar_path I get the impression that there is no possible way for this to return 0 (true in bash logic) to actually proceed to the next check.~~

Edit: Ok, forget what I said here. This is using bash globbing to expand filenames when a file exists matching the glob used.

Anyhow, I figured out why it would not return successful from the do_nfsmount() call. It checks UUIDs that are stored within the ISO in path .../.disk/casper-uuid*. Now when we prepare things and copy contents from the ISO we are missing that hidden folder (starting with a dot)! To fix that run the following command on your FOG server with the ISO mounted in /mnt/loop:

cp -R /mnt/loop/.disk /images/os/mint/20

I’m able to PXE boot into the Mint20 XFCE Live system with that fix on a VirtualBox VM (hosted on Debian 10) with only 768 MB of RAM set for the VM.

@george1421 We might think about switching to rsync -a /mnt/loop/ /images/os/... in your great tutorials on PXE booting installers to prevent that from happening. What do you think?

@londonfog I just tried Ubuntu Server 20.04 as well and it is working fine too - even using kernel and initrd from the ISO and not download the netboot ones! The important part is that you need to copy the .disk folder over as mentioned above.

20.04 autoinstall

312

11.7k

17.1k

154.7k