a bunch of fixes and improvements

[automated-distro-installer] / README

PXE install w multi-boot, btrfs & Libreboot support

Some things are specific to my home network, and uses files with secrets
that are not in this repo. I use this for bare metal and vms, and two
scripts which can run post boot so I use them on vps distributed image
as well.

Features people may find useful: installs encrypted trisquel, debian,
ubuntu, arch, and parabola (archlike install is likely broken, I've only
done pxe boots recently), in a multi-boot setup using multiple
subvolumes of a single btrfs filesystem.  Utilizes multiple disks, with
scripts to automatically decrypt on intentional reboots, but not after
shutdown or power loss.

Normal install mode for fai is using pxe, but on a libreboot system,
there is no pxe. The pxe in a normal computer is nonfree
firmware. Alternatives to normal pxe that I've tried:

* libreboot + seabios + ipxe

* Use a live cd to call pxe-kexec, this is described later in this file.

* Use the fai autodiscover iso. This is more automated, so nicer.

* Use an install method above to setup a gnu/linux disk partition that
  coordinates with libreboot grub to acts like a pxe boot using
  kexec. The boot process takes a bit longer than normal pxe. This is
  the bootstrap partition in my scripts.

Things I haven't tried:

* The bios chip has enough room for an initrd. This could be setup to
  work like the partition I use to kexec, but it would be faster, and
  not require installing to disk.

The partititioning and filesystem script is at
fai/config/hooks/partition.DEFAULT. Disks are grouped as ssd or hdd and
raided in raid 1 or raid 0 per configuration. The base partitions are
divided into boot, swap, and root, (only boot is unencrypted). There are
scripts to resize those partitions post-provision and while the system
is running.

People who use fai may find these things as useful examples: it uses
dnsmasq (on a openwrt machine) for dhcp instead of the isc
dhcp. fai-wrapper is a small script to use basic fai classes outside of
fai. It does not use the fai partitioning tool, but the script is
inspired from it and works outside of fai. It supports running a fai
server on debian within android via Maru.

It also automates configuration of an openwrt router after manual
initial installation.

After provisionining is done, I sync files using btrfs, or unison for
vps, then automate further setup using a different set of scripts,
https://iankelling.org/git/?p=distro-setup;a=tree.

My network is a wndr3700v2 router with openwrt on it and a few pcs/laptops.

Since fai requires a debian server as the fai server, there are also
scripts to automate a debian install using pxe and preseeding, which can
be done from any distro.

Some of the scripts have dependencies for some simple obvious utility
scripts from https://iankelling.org/git, and of course there are some
hostnames that are specific to my network.


# Per-host/install configuration

Before doing a fai install, you will need to populate a class file.  I
use one called 51-multi-boot, which you can see example of in
fai/config/class/50-host-classes.



Before doing a fai install, you will need to populate /q/root/luks and
/q/root/shadow, see their references. You might also want to copy
existing /etc/ssh/*host* to
/p/c/machine_specific/HOST/filesystem/etc/ssh.

host-* luks keyfiles generated like:
head -c 2048 /dev/urandom | od | s dd of=/q/root/luks/host-demohost

Configuration of which luks key to use is in
fai/config/hooks/partition.DEFAULT

Configuration of which (if any) shadow file to use is in
fai/config/distro-install-common/end
and which shadow file / luks file(s) to copy into the new machine depends
on fai-redep arguments.

# Scripts (meant to be used directly):


# Setup the environment for the install

# create tiny autodiscover cd
fai-redep && sudo fai-cd -g $PWD/grub.cfg.autodiscover -f -A $BASEFILE_DIR/autodiscover.iso
# create normal fai cd (replace TARGET_HOSTNAME)
fai-redep -t TARGET_HOSTNAME && sudo fai-cd -M -g $PWD/grub.cfg.netinst-noreboot -f $BASEFILE_DIR/netinst.iso
# note, may need to set hostname, depending on config,
# and some other things for environment not on your lan
# for example see fai/config/class/LINODE.var. See linode notes below.

mymk-basefile # Create basefiles for various distros
archlike-pxe # Setup pxe boot server from an archlike base image
fai-redep # Deploy fai configuration to host "faiserver"
faiserver-uninstall # uninstall fai-server
faiserver-setup # install fai-server on the current machine
myfai-chboot # setup fai tftp and nfs. useful for doing pxe-kexec
pxe-server # disable/enable pxe dhcp, tfp, and nfs. calls myfai-chboot
wrt-setup  # setup my router in general: dhcp, dns, etc.


# Script to do a distro install

faiserver-revm # using pxe & preseed, create a vm which is a fai server
dsfull # install & post-install a new fai distro
arch-init-remote # install arch after it's been booted into it's setup env
live-kexec # Kexec this or a remote machine using host faiserver. also
             useful to run as curl live-kexec|bash


# Test scripts

arch-revm # test arch install on a fresh vm
fai-revm  # test fai install on a fresh vm


# Scripts to call after a distro install for various reasons

chboot # Set grub to boot into a different distro (installed earlier)
install-chboot # reinstall chboot to /boot subvols, for chboot updates.
eboot # reboot without automatic disk decryption
fai-wrapper # use fai classes outside of fai. sourced, not called.
faiserver-disable # Disable the fai nfs server exports
fresize # resize swap or boot partitions in a host


# Replacing a raid 10 disk

# i expect better results with newer kernel and btrfs progs than the default stretch
fai-server buster

pxe-server -S HOST fai

# btrfs replace or delete. prefer replace. to setup partitions on replacement drive:
scp fai-wrapper HOST:
ssh root@HOST
. fai-wrapper
export SPECIAL_DISK=/dev/REPLACEMENT_DEV
/var/lib/fai/config/hooks/partition.DEFAULT


ssh root@HOST
for x in /target/* /target; do umount $x; done
cat >p
PASSWORD HERE(ctrl-d ctrl-d)
cd /dev/disk/by-id/
for d in ata*part1; do cryptsetup luksOpen -d /root/p $d crypt_dev_$d; done
x=(/dev/mapper/*part1); mount -o subvol=root_trisquelflidas $x /mnt
# btrfs fi show /mnt
# btrfs replace start -f /dev/mapper/OLD_DEV /dev/mapper/NEW_DEV /mnt
# btrfs replace status /mnt
# nohup btrfs dev delete /dev/sde1 /mnt
mount -o subvol=boot_trisquelflidas /dev/sda3 /mnt/boot
# also replace or delete disk for boot
for x in dev proc sys; do mount -o bind /$x /mnt/$x; done
chroot /mnt /bin/bash
# replace disk in fstab
# replace disk in /etc/crypttab
update-grub
update-initramfs -u
mount /a
/a/exe/keyscript-on
exit
reboot


# Expected output in fai logs

For flidas, when installing systemd, this error happens, and it's
a superflous upstream bug based on reading the post install script:

addgroup: The group `systemd-journal' already exists as a system group. Exiting.
Operation failed: No such file or directory


# linode notes

* create 2 disks, installer (3000 mb, raw), boot (remaining, raw)
* create 2 profiles w direct boot, no helpers:
  * installer (sda=boot, sdb=installer, boot dev=sdb)
  * boot (sda=boot)
* Boot into rescue mode, ssh in with lish,
  curl url_to_some_fai_cd_created_image | dd of=/dev/sda
  poweroff
* boot into installer.
* Lish shows console, at the end of install, it gives prompt because
  logs failed to save remotely, check the logs, then reboot into boot
  profile if all is well. If that doesn't happen, turn off lassie in
  settings.



# TODO
Change arch to archlike and to support arch and parabola


# License

The license for the project is GPLv2 or later, mostly because fai is and
I periodically merge the upstream example config, which contains small
scripts. Also, there is a modified encrypt.upstream, which is from the
cryptsetup package in arch, which is under the same license.