# v-i---a Debian installer using vmdb2 **WARNING: Running v-i is like waking up after an alien invasion, in a post-apocalyptic world, with everything you knew or owned gone forever. When you run v-i, it *will* wipe away everything on that computer. All volume groups will be deleted, all storage drives emptied. Any existing partitions will be lost. Forget any data you used to have, and operating systems you used to have installed. If you don't know what you're doing, leave.** **v-i** installs enough of Debian installed so that the system can be configured with a configuration management system once it's up an running. **v-i** isn't meant to install a full desktop stack or service suite: it's meant to provide a base on which such can be installed. However, **v-i** can probably do that, but the [author][] thinks such setup and configuration is better done post-install. **v-i** installs a very basic Debian onto a PC. It's entirely non-interactive and unhelpful. The author wrote it so that repeated installations would be less of a chore than using the official Debian installer. (Actually, the author thought it'd be a quick, easy hack, and was too stubborn to give up, when it turned out to be a bit tricky.) **v-i** uses **vmdb2** to install onto bare metal hardware. [**vmdb2**][] is a program to create a disk image virtual machines with Debian, by the same author. It "installs Debian" to a file representing a hard drive. It's basically [debootstrap][], except the target is a disk image instead of a directory. It's used to create [Debian images for Raspberry Pis][]. To use **v-i** to install Debian on a PC: * Boot the target machine off a live system that has **v-i** installed. - the author uses a USB stick with an image built with the [`build-installer.sh`][] script - the author logs into the installer system via SSH * Create a v-i target specification file. See below for an example. * Run the command: `v-i --verbose exolobe5.yaml` * See `installer.log` for what happened during the installation, if anything failed. Example target specification file: ```yaml hostname: exolobe5 drive: /dev/nvme0n1 ``` A number of fields are allowed in the specification file: * `hostname`---the hostname of the installed system. This is so that when the installed system boots, and gets a network address using DHCP, it can provide a name. The author's home network setup automatically adds that hostname to the internal DNS. This avoids a manual DNS configuration step, and the author is lazy. ([dnsmasq](https://dnsmasq.org/) is lovely.) * `drive`---the main drive to install to. This is where the EFI and `/boot` partitions are created, and where GRUB gets installed. The rest of the drive will be a physical volume for LVM2. * `extra_drives`---a list of any additional physical volumes for LVM2. These will not be partitioned, and will be used entirely as physical volumes. * `luks`---the password for full disk encryption for all LVM2 physical volumes. If not set, LUKS is not used. This is a single, fixed password that is in cleartext. _You are expected to change it_ after the system is installed and booted. If you'd rather use, say, a hardware token's challenge/response feature or TPM for LUKS, that's better done on a running system. * `extra_playbooks`---additional Ansible playbooks to use on the installed system. **v-i** comes with a "standard playbook" (in [`std.yml`][]) that it uses unconditionally, to set up a "standard system" that the author likes. You can provide additional playbooks, for additional configuration at installation time. * `ansible_vars`---variables to set for Ansible playbooks. - the `user_pub` variable contains an SSH public key that gets installed into the `root` user `authorized_keys` file on the installed system by the standard playbook - the `user_ca_pubkey` variable contains public key for an SSH CA whose user certificates are to be trusted With all this configuration in a file, which you can keep in git, you can install a base system repeatedly to a specific computer, and do it the same way every time. If that's not something you do, then you may want to use the official Debian installer instead. (Caveat: **v-i** does nothing to configure your BIOS/UEFI. It can't. You have to manually configure it the way you want it to be. For example, one of the author's machines needs to have its boot order adjusted after every operating system installation. It's quite tedious.) [**vmdb2**]: https://vmdb2.liw.fi/ [debootstrap]: https://wiki.debian.org/Debootstrap [Debian images for Raspberry Pis]: https://raspi.debian.net/ [`build-installer.sh`]: build-installer.sh [`v-i`]: v-i [`std.yml`]: std.yml [author]: https://liw.fi/ [Debian installer]: https://www.debian.org/devel/debian-installer/ [preseed files]: https://wiki.debian.org/DebianInstaller/Preseed [udeb]: https://en.wikipedia.org/wiki/Deb_(file_format) ## Motivation The official [Debian installer][] is often referred to as _d-i_. It works quite well, for almost any hardware Debian can run on, and supports a lot of languages, and it's flexible enough to be acceptable for nearly every use case. Millions upon millions of people are satisfied users of it. It is a great achievement of Debian, and the people of the `debian-boot` team. However, the **v-i** author felt it was lacking for their needs: * d-i is not entirely easy to understand and modify. It requires building special [udeb][] packages for any software that's to be part of the installer environment, which makes it harder to make changes to the installer without co-operation from maintainers of those packages. The architecture of d-i is also a little non-linear. d-i also needs to support a very wide variety of hardware and use cases, which has made it large and complex. **v-i** is happy with normal deb packages, and is a thin Python wrapper script around **vmdb2**, making it reasonably easy to understand and change. Well, easy for its author. The price for this is less flexibility and less ease of use. * d-i is primarily meant to be used interactively, but it does support [preseed files][] for automating an installation. Preseeding means providing answers, in a file, to questions a package being installed may ask during its installation. This is fine, if a little cumbersome, but only helps to answer questions the packages ask when installed. **v-i** lets you have the full power of Ansible during initial installation. If **v-i** isn't suitable for your uses, that's OK. The author is happy with his toy. ## Architecture **vmdb2** is given a sequence of _steps_ to execute: create this partition, make that file system, install those packages, etc. **vmdb2** runs the steps against a disk image or physical hard drive, with a chroot of the file systems, to do things like installing a package in the system being installed. **v-i** defines a fairly minimal _standard install_, whose goal is to get the target system into a state where it boots from its own, internal storage, and where the rest of the system configuration can be finished using your configuration management system of choice. The standard system looks like this: * UEFI boot, with an EFI partition (500 MiB) * a cleartext `/boot` partition (500 MiB) * LVM2 for the rest of the target drive, and all extra drives - one VG - no RAID at this time * optional LUKS for all physical volumes, using the same passphrase * a `root` LV (20 GiB) * the rest of the VG not allocated * a basic Debian installation - `root` password is locked - network setup on `eth0` using systemd-networkd - SSH host key and host certificate installed if defined - log in to `root` over SSH using a key or user certificate While **vmdb2** can, and does, run Ansible to configure the system being installed, in practice some things work better if most configuration is done to a running system; if nothing else, some Ansible modules don't work well in a chroot. The goal of **v-i** is to get a system into that state as quickly and easily as possible. For example, the Ansible module to set a hostname on a system with systemd requires systemd to be running. That's awkward while the system is still being installed in a chroot. Thus, **v-i** does the following: * delete any trace of LVM2 from all drives, wipe all SSDs, and generally reset the system to as close to a blank state as possible - there is no question "are you sure?" to give the user a chance to repent: as soon as you run **v-i**, you've lost all your data * create a partition table ("label") on the target drive * create EFI and boot partitions, needed to boot with UEFI and LUKS * create a physical volume for LVM2, and a logical volume for the root file system - add any additional drives as physical volumes to the volume group - optionally use LUKS for full disk encryption for each physical volume * install the Debian base system - run `debootstrap`, install a boot loader, and create fstab and `crypttab` files * run the standard Ansible playbook (see [`std.yml`][]) - set hostname - set keyboard layout - configure networking (using systemd-networkd) - install an SSH server - add a chosen SSH public key to the root user's authorized keys file * run any additional playbooks **v-i** uses the **vmdb2** caching feature, where the results of `debootstrap` and some other steps get stored in a compressed tar archive. On subsequent runs, if the cache file exists, it's unpacked, instead of running the commands. This speeds things up a bit: running **v-i** without the cache file takes the author about 5 minutes; with the cache file it takes about 1.5 minutes. This matters if there is a need to do many installations. It also matters if you're developing an installer and need to run it tens of times a day. ## Hacking The main files of **v-i** are: * [`v-i`][]---the actual installer, a Python script * [`std.yml`][]---the Ansible playbook to configure a standard install Also, to build an image to boot off for running the installer: * `build-installer.sh`---build a disk image where **v-i** can be run - put image on a USB drive, boot off that drive, run installer - note that you can use any live image with **vmdb2** installed; the image built with this is just the easiest for the author * `installer.vmdb`---the **vmdb2** specification file for creating the installer image * `installer.yml`---the Ansible playbook for creating the installer image See the tutorial about ways to add your SSH public key to the image so that you can log into the installer via SSH. You probably mostly only need to modify `v-i` and `std.yml`. The rest is to get you and your target machine into a state where you can run the installer. ## FAQ This section is prescient: the author hasn't been asked any questions yet, but expects the following to be asked. ### What version of Debian does v-i install? **v-i** installs Debian 11 (bullseye). ### What about other releases of Debian? The Debian 11 (bullseye) release is the earliest release the author has gotten to work with **v-i**, and is the only release the author is installing on bare metal systems. Later versions of Debian may work, we will see. ### Is only UEFI supported? Yes. ### Why is BIOS (without UEFI) not supported? All of the author's PCs have UEFI, and the author doesn't care to do the work to add support for BIOS. ### What about multi-boot? **v-i** doesn't support installing more than one operating system on one computer. The author has no need for this. ### What about installing something else than Debian? The author only cares about Debian, but in principle, fairly little of **vmdb2** and **v-i** are specific to Debian. It should be possible to add support for other operating systems to be installed, at least ones based on Linux. If you're interested, you need to change or replace at least the following steps in **vmdb2** code, and then change the [`v-i`][] script to generate a specification using those steps: * `debootstrap`---install base operating system into a directory - after this step, all the files in a base installation should be in the specified directory tree, except the boot loader and kernel - could probably be replaced with providing **v-i** with a pre-built cache tar archive * `apt`---install packages - whatever package manager the system has probably works - you can probably run the package manager from a **vmdb2** `chroot` step * `grub`---install boot loader - this chooses the appropriate Debian package automatically - might possibly be doable as a `shell` step - this is likely the trickiest bit: booting is _intricate_ * `cryptsetup`---format a drive for full disk encryption - this just runs the `cryptsetup` program and tells the fstab step to create a crypttab file - might just work * `vgcreate` and `lvcreate`---create LVM2 - these just run the relevant LVM2 commands - might just work ### What about other kinds of computers than PCs? The author only uses 64-bit PC computers (`amd64` arhitecture in Debian; also known as x86-64). **v-i** may well work for other kinds of computers, as long as they can boot off an installer image ("live image"), and use GRUB for booting. The author would be interested to hear if that is the case. ### Why is the LUKS password in cleartext? It would be ideal if **v-i** (or **vmdb2**) got the LUKS password for full disk encryption in a secure way from a secure source, but that turned out to be tricky to do. The author felt it was too tricky to do well in the installer environment, while it's pretty easy to do in a running system. Thus, the cleartext password _in the installer_ is a compromise. You're expected to change the password after the installation is done. It would be possible to ask the person doing the installation to enter the password manually, but this would mean the installation would not be fully automated. The author didn't want that. ### Do I have to use Ansible? No. Use whatever you like once you've installed a system with **v-i** and booted it. **v-i** itself uses Ansible, because that was easy for the author to use. ### I'd like to use v-i, but I need changes If you can make the changes yourself, go ahead: this is free and open source software, have at it. If you don't have the skill or time to make changes yourself, you'll need to find someone else to make them. This might require paying them. The author is, unfortunately, probably not willing to spend their free time to make changes that don't benefit them directly, for free. Sorry. They _are_ willing to review and merge changes that would make the software better. (Also, the author is willing to be paid for such work, for corporate customers. Unfortunately, invoicing private people is too complicated.) ### Why write v-i, when X, Y, and Z already exist? The author likes writing software, and dislikes evaluating software. # Legalese Copyright 2018-2022 Lars Wirzenius This program is free software: you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation, either version 3 of the License, or (at your option) any later version. This program is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details. You should have received a copy of the GNU General Public License along with this program. If not, see .