nixos-install-scripts/hosters/ovh-dedicated/ovh-dedicated-wipe-and-install-nixos.sh
2020-12-27 22:45:07 +01:00

324 lines
14 KiB
Bash

#!/usr/bin/env bash
# Installs NixOS on an OVH server, wiping the server.
#
# This is for a specific server configuration; adjust where needed.
# Originally written for an OVH STOR-1 server.
#
# Prerequisites:
# * Create a LUKS key file at /root/benacofs-luks-key
# e.g. by copying it up.
# * Update the script to put in your SSH pubkey, adjust hostname, NixOS version etc.
#
# Usage:
# ssh root@YOUR_SERVERS_IP bash -s < ovh-dedicated-wipe-and-install-nixos.sh
#
# When the script is done, make sure to boot the server from HD, not rescue mode again.
# Explanations:
#
# * Following largely https://nixos.org/nixos/manual/index.html#sec-installing-from-other-distro.
# * **Important:** We boot in UEFI mode, thus requiring an ESP.
# Booting in LEGACY mode (non-UEFI boot, without ESP) would require that:
# * `/boot` is on the same device as GRUB
# * NVMe devices aren not used for booting (those require EFI boot)
# We also did not manage to boot our OVH server in LEGACY mode on our SuperMicro mainboard, even when we installed `/` (including `/boot`) directly to a simple RAID1ed GPT partition. The screen just stayed black.
# * We set a custom `configuration.nix` so that we can connect to the machine afterwards.
# * This server has 1 SSD and 4 HDDs.
# We'll ignore the SSD, putting the OS on the HDDs as well, so that everything is on RAID1.
# We wipe the SSD though, so that if it had some boot partitions on it, they don't interfere.
# Storage scheme: `partitions -> RAID -> LUKS -> LVM -> ext4`.
# * A root user with empty password is created, so that you can just login
# as root and press enter when using the OVH KVM.
# Of course that empty-password login isn't exposed to the Internet.
# Change the password afterwards to avoid anyone with physical access
# being able to login without any authentication.
# * The script reboots at the end.
set -eu
set -o pipefail
set -x
# Inspect existing disks
lsblk
# Undo existing setups to allow running the script multiple times to iterate on it.
# We allow these operations to fail for the case the script runs the first time.
set +e
umount /mnt/boot/ESP*
umount /mnt
vgchange -an
cryptsetup luksClose data0-unencrypted
cryptsetup luksClose data1-unencrypted
set -e
# Stop all mdadm arrays that the boot may have activated.
mdadm --stop --scan
# Create partition tables (--script to not ask)
parted --script /dev/sda mklabel gpt
parted --script /dev/sdb mklabel gpt
parted --script /dev/sdc mklabel gpt
parted --script /dev/sdd mklabel gpt
parted --script /dev/nvme0n1 mklabel gpt
# Create partitions (--script to not ask)
#
# Create EFI system partition (ESP) and main partition for each boot device.
# We make it 550 M as recommended by the author of gdisk (https://www.rodsbooks.com/linux-uefi/);
# using 550 ensures it's greater than 512 MiB, no matter if Mi or M were used.
# For the non-boot devices, we still make space for an ESP partition
# (in case the disks get repurposed for that at some point) but mark
# it as `off` and label it `*-unused` to avoid confusion.
#
# Note we use "MB" instead of "MiB" because otherwise `--align optimal` has no effect;
# as per documentation https://www.gnu.org/software/parted/manual/html_node/unit.html#unit:
# > Note that as of parted-2.4, when you specify start and/or end values using IEC
# > binary units like "MiB", "GiB", "TiB", etc., parted treats those values as exact
#
# Note: When using `mkpart` on GPT, as per
# https://www.gnu.org/software/parted/manual/html_node/mkpart.html#mkpart
# the first argument to `mkpart` is not a `part-type`, but the GPT partition name:
# ... part-type is one of 'primary', 'extended' or 'logical', and may be specified only with 'msdos' or 'dvh' partition tables.
# A name must be specified for a 'gpt' partition table.
# GPT partition names are limited to 36 UTF-16 chars, see https://en.wikipedia.org/wiki/GUID_Partition_Table#Partition_entries_(LBA_2-33).
parted --script --align optimal /dev/sda -- mklabel gpt mkpart 'ESP-partition0' fat32 1MB 551MB set 1 esp on mkpart 'OS-partition0' 551MB 500GB mkpart 'data-partition0' 500GB '100%'
parted --script --align optimal /dev/sdb -- mklabel gpt mkpart 'ESP-partition1' fat32 1MB 551MB set 1 esp on mkpart 'OS-partition1' 551MB 500GB mkpart 'data-partition1' 500GB '100%'
parted --script --align optimal /dev/sdc -- mklabel gpt mkpart 'ESP-partition2-unused' fat32 1MB 551MB set 1 esp off mkpart 'data-partition2' 551MB '100%'
parted --script --align optimal /dev/sdd -- mklabel gpt mkpart 'ESP-partition3-unused' fat32 1MB 551MB set 1 esp off mkpart 'data-partition3' 551MB '100%'
# Relaod partitions
partprobe
# Wait for all devices to exist
udevadm settle --timeout=5 --exit-if-exists=/dev/sda1
udevadm settle --timeout=5 --exit-if-exists=/dev/sda2
udevadm settle --timeout=5 --exit-if-exists=/dev/sda3
udevadm settle --timeout=5 --exit-if-exists=/dev/sdb1
udevadm settle --timeout=5 --exit-if-exists=/dev/sdb2
udevadm settle --timeout=5 --exit-if-exists=/dev/sdb3
udevadm settle --timeout=5 --exit-if-exists=/dev/sdc1
udevadm settle --timeout=5 --exit-if-exists=/dev/sdc2
udevadm settle --timeout=5 --exit-if-exists=/dev/sdd1
udevadm settle --timeout=5 --exit-if-exists=/dev/sdd2
# Wipe any previous RAID signatures
mdadm --zero-superblock /dev/sda2
mdadm --zero-superblock /dev/sda3
mdadm --zero-superblock /dev/sdb2
mdadm --zero-superblock /dev/sdb3
mdadm --zero-superblock /dev/sdc2
mdadm --zero-superblock /dev/sdd2
# Create RAIDs
# Note that during creating and boot-time assembly, mdadm cares about the
# host name, and the existence and contents of `mdadm.conf`!
# This also affects the names appearing in /dev/md/ being different
# before and after reboot in general (but we take extra care here
# to pass explicit names, and set HOMEHOST for the rebooting system further
# down, so that the names appear the same).
# Almost all details of this are explained in
# https://bugzilla.redhat.com/show_bug.cgi?id=606481#c14
# and the followup comments by Doug Ledford.
mdadm --create --run --verbose /dev/md/root0 --level=1 --raid-devices=2 --homehost=benaco-cdn --name=root0 /dev/sda2 /dev/sdb2
mdadm --create --run --verbose /dev/md/data0-encrypted --level=1 --raid-devices=2 --homehost=benaco-cdn --name=data0-encrypted /dev/sda3 /dev/sdb3
mdadm --create --run --verbose /dev/md/data1-encrypted --level=1 --raid-devices=2 --homehost=benaco-cdn --name=data1-encrypted /dev/sdc2 /dev/sdd2
# Assembling the RAID can result in auto-activation of previously-existing LVM
# groups, preventing the RAID block device wiping below with
# `Device or resource busy`. So disable all VGs first.
vgchange -an
# Wipe filesystem signatures that might be on the RAID from some
# possibly existing older use of the disks (RAID creation does not do that).
# See https://serverfault.com/questions/911370/why-does-mdadm-zero-superblock-preserve-file-system-information
wipefs -a /dev/md/root0
wipefs -a /dev/md/data0-encrypted
wipefs -a /dev/md/data1-encrypted
# Disable RAID recovery. We don't want this to slow down machine provisioning
# in the rescue mode. It can run in normal operation after reboot.
echo 0 > /proc/sys/dev/raid/speed_limit_max
# LUKS encryption (--batch-mode to not ask)
cryptsetup --batch-mode luksFormat /dev/md/data0-encrypted /root/benacofs-luks-key
cryptsetup --batch-mode luksFormat /dev/md/data1-encrypted /root/benacofs-luks-key
# Decrypt
cryptsetup luksOpen /dev/md/data0-encrypted data0-unencrypted --key-file /root/benacofs-luks-key
cryptsetup luksOpen /dev/md/data1-encrypted data1-unencrypted --key-file /root/benacofs-luks-key
# LVM
# PVs
pvcreate /dev/mapper/data0-unencrypted
pvcreate /dev/mapper/data1-unencrypted
# VGs
vgcreate vg0 /dev/mapper/data0-unencrypted /dev/mapper/data1-unencrypted
# LVs
lvcreate --extents 95%FREE -n benacofs vg0 # 5% slack space
# Filesystems (-F to not ask on preexisting FS)
mkfs.fat -F 32 -n esp0 /dev/disk/by-partlabel/ESP-partition0
mkfs.fat -F 32 -n esp1 /dev/disk/by-partlabel/ESP-partition1
mkfs.ext4 -F -L root /dev/md/root0
mkfs.ext4 -F -L benacofs /dev/mapper/vg0-benacofs
# Creating file systems changes their UUIDs.
# Trigger udev so that the entries in /dev/disk/by-uuid get refreshed.
# `nixos-generate-config` depends on those being up-to-date.
# See https://github.com/NixOS/nixpkgs/issues/62444
udevadm trigger
# Wait for FS labels to appear
udevadm settle --timeout=5 --exit-if-exists=/dev/disk/by-label/root
udevadm settle --timeout=5 --exit-if-exists=/dev/disk/by-label/benacofs
# NixOS pre-installation mounts
# Mount target root partition
mount /dev/disk/by-label/root /mnt
# Mount efivars unless already mounted
# (OVH rescue doesn't have them by default and the NixOS installer needs this)
mount | grep efivars || mount -t efivarfs efivarfs /sys/firmware/efi/efivars
# Mount our ESP partitions
mkdir -p /mnt/boot/ESP0
mkdir -p /mnt/boot/ESP1
mount /dev/disk/by-label/esp0 /mnt/boot/ESP0
mount /dev/disk/by-label/esp1 /mnt/boot/ESP1
# Installing nix
# Allow installing nix as root, see
# https://github.com/NixOS/nix/issues/936#issuecomment-475795730
mkdir -p /etc/nix
echo "build-users-group =" > /etc/nix/nix.conf
curl -L https://nixos.org/nix/install | sh
set +u +x # sourcing this may refer to unset variables that we have no control over
. $HOME/.nix-profile/etc/profile.d/nix.sh
set -u -x
nix-channel --add https://nixos.org/channels/nixos-19.03 nixpkgs
nix-channel --update
# Getting NixOS installation tools
nix-env -iE "_: with import <nixpkgs/nixos> { configuration = {}; }; with config.system.build; [ nixos-generate-config nixos-install nixos-enter manual.manpages ]"
nixos-generate-config --root /mnt
# On the OVH rescue mode, the default Internet interface is called `eth0`.
# Find what its name will be under NixOS, which uses stable interface names.
# See https://major.io/2015/08/21/understanding-systemds-predictable-network-device-names/#comment-545626
INTERFACE=$(udevadm info -e | grep -A 11 ^P.*eth0 | grep -o -E 'ID_NET_NAME_ONBOARD=\w+' | cut -d= -f2)
echo "Determined INTERFACE as $INTERFACE"
IP_V4=$(ip route get 8.8.8.8 | head -1 | cut -d' ' -f8)
echo "Determined IP_V4 as $IP_V4"
# From https://stackoverflow.com/questions/1204629/how-do-i-get-the-default-gateway-in-linux-given-the-destination/15973156#15973156
read _ _ DEFAULT_GATEWAY _ < <(ip route list match 0/0); echo "$DEFAULT_GATEWAY"
echo "Determined DEFAULT_GATEWAY as $DEFAULT_GATEWAY"
# Generate `configuration.nix`. Note that we splice in shell variables.
cat > /mnt/etc/nixos/configuration.nix <<EOF
{ config, pkgs, ... }:
{
imports =
[ # Include the results of the hardware scan.
./hardware-configuration.nix
];
# Use GRUB2 as the EFI boot loader.
# We don't use systemd-boot because then
# * we can't use boot.loader.grub.mirroredBoots to mirror the ESP over multiple disks
# * we can't put /boot on the same partition as /
# (boot.loader.efi.efiSysMountPoint = "/boot/EFI" apparently does not have
# the desired outcome then, just puts all of /boot under /boot/EFI instead)
boot.loader.systemd-boot.enable = false;
boot.loader.grub = {
enable = true;
efiSupport = true;
mirroredBoots = [
{ devices = [ "nodev" ]; path = "/boot/ESP0"; }
{ devices = [ "nodev" ]; path = "/boot/ESP1"; }
];
};
boot.loader.efi.canTouchEfiVariables = true;
# Don't put NixOS kernels, initrds etc. on the ESP, because
# the ESP is not RAID1ed.
# Mount the ESP at /boot/efi instead of the default /boot so that
# boot is just on the / partition.
boot.loader.efi.efiSysMountPoint = "/boot/EFI";
# OVH has an issue where on newer kernels, it can take up to 20 minutes
# for the default gateway to not be 'linkdown' when booting. See #812.
# We observed that with the 4.15 kernels so far, including OVH's own
# Ubuntu 18.04.
# Our workaround so far is to use the 4.9 kernel.
boot.kernelPackages = pkgs.linuxPackages_4_9;
networking.hostName = "benaco-cdn-na1";
# The mdadm RAID1s were created with 'mdadm --create ... --homehost=benaco-cdn',
# but the hostname for each CDN machine is different, and mdadm's HOMEHOST
# setting defaults to '<system>' (using the system hostname).
# This results mdadm considering such disks as "foreign" as opposed to
# "local", and showing them as e.g. '/dev/md/benaco-cdn:data0'
# instead of '/dev/md/data0'.
# This is mdadm's protection against accidentally putting a RAID disk
# into the wrong machine and corrupting data by accidental sync, see
# https://bugzilla.redhat.com/show_bug.cgi?id=606481#c14 and onward.
# We set the HOMEHOST manually go get the short '/dev/md' names,
# and so that things look and are configured the same on all such CDN
# machines irrespective of host names.
# We do not worry about plugging disks into the wrong machine because
# we will never exchange disks between CDN machines.
environment.etc."mdadm.conf".text = ''
HOMEHOST benaco-cdn
'';
# The RAIDs are assembled in stage1, so we need to make the config
# available there.
boot.initrd.mdadmConf = config.environment.etc."mdadm.conf".text;
# Network (OVH uses static IP assignments, no DHCP)
networking.useDHCP = false;
networking.interfaces."$INTERFACE".ipv4.addresses = [
{
address = "$IP_V4";
prefixLength = 24;
}
];
networking.defaultGateway = "$DEFAULT_GATEWAY";
networking.nameservers = [ "8.8.8.8" ];
# Initial empty root password for easy login:
users.users.root.initialHashedPassword = "";
services.openssh.permitRootLogin = "prohibit-password";
users.users.root.openssh.authorizedKeys.keys = [
# Replace this by your pubkey!
"ssh-rsa AAAAAAAAAAA..."
];
services.openssh.enable = true;
# This value determines the NixOS release with which your system is to be
# compatible, in order to avoid breaking some software such as database
# servers. You should change this only after NixOS release notes say you
# should.
system.stateVersion = "19.03"; # Did you read the comment?
}
EOF
# Install NixOS
PATH="$PATH" NIX_PATH="$NIX_PATH" `which nixos-install` --no-root-passwd --root /mnt --max-jobs 40
reboot