Eric Radman : a Journal

OpenBSD VPS Installation

Some virtual private server providers support OpenBSD as an install target. This is a wonderful start, but only a start because the image applied and resided will probably not a useful partition scheme and mount options that makes sense for a given security and operational strategy.

Fortunately, a hosting provider such as Vultr can deliver an OpenBSD instance that can reinstall itself using and autoinstall answers file and disklabel fetched over HTTP. 

# /etc/fstab
/dev/sd0a / ffs rw 1 1
/dev/sd0b none swap sw
/dev/sd0d /usr ffs rw,nodev 1 2
/dev/sd0e /usr/local ffs rw,wxallowed,nodev 1 2
/dev/sd0f /tmp ffs rw,nodev,nosuid 1 2
/dev/sd0g /home ffs rw,nodev,nosuid 1 2
/dev/sd0h /var ffs rw,nodev,nosuid 1 2

Using a pre-built cloud image as a trampoline also provides the opportunity to exclude some base filesets or include other capabilities using siteXX.tgz.

Declaring Cloud Resrouces

Infrastructure configuration systems such as terraform are not strictly necessary if there is a well-documented REST interface, but such a tool can save the very repetitive work of using the API

Vultr offers a resource called a startup script which allows a shell script to be run when the instance first boots. Define the contents of the auto install script using a heredoc 

# .tf
locals {
  auto_install_script = <<-EOF
    # ...
  EOF
}

resource "vultr_startup_script" "auto_reinstall" {
    name = "auto_reinstall"
    script = base64encode(local.auto_install_script)
}

resource "vultr_instance" "svc1" {
    plan = "vc2-1c-1gb"
    region = "ord"
    label = "svc1"
    os_id = "2464"  # OpenBSD 7.6
    hostname = "svc1.eradman.com"
    backups = "disabled"
    ddos_protection = false
    activation_email = false
    ssh_key_ids = ["06f97132-e980-4c13-b9e7-d2e8f228f194"]  # root
    script_id = vultr_startup_script.auto_reinstall.id
}

Where the OS ID for the OpenBSD installation can be found in their web UI or using their API 

$ curl -s "https://api.vultr.com/v2/os" | flattenjs | grep -B1 'OpenBSD 7.6'
{os,39,id} 2464
{os,39,name} OpenBSD 7.6 x64

Spin up the new infrastructure using terraform apply.

Changing the os_id will result in a destroy/create operation, which will assign a new IP. To work around this, change the OS in the Vultr web user interface, then change the terraform plan to match.

OpenBSD Install Templates

There are two configuration files we will want to host on our install mirror. The first is the disklabel containing our partition layout 

/           1G
swap        512M
/usr        4G
/usr/local  5G
/tmp        2G-4G     20%
/home       2G-4G     20%
/var        5G-*      40%

The second is a script that will emit a completed auto_install(8) configuration 

#!/bin/sh -eu

cat <<EOF
System hostname = $(hostname)
Password for root = ${root_password}
Network interfaces = vio0
IPv4 address for vio0 = autoconf
Do you expect to run the X Window System = no
Setup a user = ${admin_user}
Password for user ${admin_user} = ${admin_password}
Public ssh key for user = ${pub_key}
Which disk is the root disk = sd0
What timezone are you in = US/Eastern
Unable to connect using https. Use http instead = yes
Location of sets = http
Server = ${mirror}
Set name(s) = -all bsd* base* etc* man*
URL to autopartitioning template for disklabel = http://${mirror}/install/$(hostname -s).disklabel
EOF

Building a Self-Installer at First Boot

The trick to automating an install without PXE is to spin a new ramdisk to boot from that contains the auto_install.conf answer file 

# auto_install_script
cd /tmp
mkdir -p mnt

for mirror in "relay1.sidecomment.io:8080" "svc1.sidecomment.io:8080"
do
    ftp -o auto_install.sh "http://$mirror/install/$(hostname -s).auto_install"
    ftp -o bsd.rd.gz "http://$mirror/pub/OpenBSD/7.6/amd64/bsd.rd"
    [ -f bsd.rd.gz ] && [ -f auto_install.sh ] && break
done

gzip -d bsd.rd.gz
rdsetroot -x bsd.rd disk.fs
vnconfig vnd0 disk.fs
mount /dev/vnd0a mnt

root_password='$2b$08$v2y8L...5DYQllk.8ji'
admin_user='admin'
admin_password='$2b$08$0MMjh...2ZN2VOZSQWC'
pub_key='ssh-ed25519 AAAAC3NzaC1...g3Aqre admin@localhost'

export root_password admin_user admin_password pub_key mirror
sh -eu auto_install.sh > mnt/auto_install.conf

umount mnt
vnconfig -u vnd0
rdsetroot bsd.rd disk.fs
gzip -c9n bsd.rd > /bsd.rd.new
chmod 700 /bsd.rd.new
echo "boot bsd.rd.new" > /etc/boot.conf
shutdown -r now

The steps this configuration script takes are

  1. Fetch auto_install(8) template and the latest release of bsd.rd
  2. Unpack bsd.rd and mount it with vnconfig(8)
  3. Set password hashes generated by encrypt(1) and public key from ssh-keygen(1)
  4. Evaluate the autoinstall template and install to the ramdisk root
  5. Update the boot image rdsetroot(8), compress it with gzip(1)
  6. Copy the new .rd image to /, set it as the default boot file
  7. Reboot

The Vultr REST API and support for a startup script makes automating builds seamless, but if the hosting service that does not have a similar feature, copy the auto-install script to the target host and start it by hand.

Local Terraform Build

Upstream builds are typcially not rebuilt when a new release of OpenBSD. If the upstream provider does not work, the Vultr provider can be built locally 

cd /tmp
RELEASE="2.21.0"
DST="$HOME/.terraform.d/plugins/local/vultr/vultr/$RELEASE/openbsd_amd64"
mkdir -p $DST
ftp https://github.com/vultr/terraform-provider-vultr/archive/refs/tags/v$RELEASE.tar.gz
tar -zxf v$RELEASE.tar.gz
cd terraform-provider-vultr-$RELEASE
go build
mv terraform-provider-vultr $DST/terraform-provider-vultr_v$RELEASE

Update providers.tf to source this module locally 

terraform {
  required_providers {
    vultr = {
      source = "local/vultr/vultr"
      version = "2.21.0"
    }
  }
}