Dotfiles/setup/reset-arch.sh

380 lines
19 KiB
Bash
Executable File

#!/usr/bin/env bash
# ╔══════════════════════════════════════════════════════════════════════════════╗
# ║ setup/reset-arch.sh — Wipe and reinstall the OS while keeping user data ║
# ║ ║
# ║ PURPOSE: ║
# ║ Performs a clean OS reinstall on an existing btrfs-on-LUKS system ║
# ║ WITHOUT destroying user home directories or passwords. Useful when ║
# ║ the system is broken, corrupted, or needs a clean slate. ║
# ║ ║
# ║ WHAT THIS DOES: ║
# ║ 1. Detects LUKS encryption; unlocks via FIDO2 token and/or passphrase ║
# ║ 2. Saves user credentials and system config from the old @ subvolume ║
# ║ 3. Clears app configs (~/.config) from @home, preserving auth keys ║
# ║ (Yubico/ is kept so FIDO2 login works after the reset) ║
# ║ 4. Deletes and recreates the @ (root) btrfs subvolume ║
# ║ 5. Reinstalls the base system via pacstrap ║
# ║ 6. Restores credentials, PAM, fstab, mkinitcpio, GRUB config ║
# ║ 7. Regenerates initramfs and GRUB menu from chroot ║
# ║ ║
# ║ REQUIREMENTS: ║
# ║ - Run from an Arch Linux live environment (archiso USB) ║
# ║ - Target disk uses btrfs with @ and @home subvolumes ║
# ║ - Optional: LUKS2 encryption on the root partition ║
# ║ ║
# ║ WARNING: The root subvolume (@ i.e. /etc, /usr, /var) is DESTROYED. ║
# ║ User home data (@home) is preserved. Run with extreme caution. ║
# ╚══════════════════════════════════════════════════════════════════════════════╝
# reset-arch.sh — Reset the root btrfs subvolume while preserving user home data.
#
# What this does:
# 1. Detects LUKS encryption; unlocks via FIDO2 token and/or passphrase
# 2. Saves user credentials and system config from the old @ subvolume
# 3. Clears app configs (~/.config) from @home, preserving auth keys (Yubico/)
# 4. Deletes and recreates the @ (root) btrfs subvolume
# 5. Reinstalls the base system via pacstrap
# 6. Restores credentials, PAM, fstab, mkinitcpio, GRUB config
# 7. Regenerates initramfs and GRUB menu from chroot so the system boots cleanly
set -euo pipefail
# -e: abort immediately on any error — wrong moves here could corrupt the system
# -u: error on unset variable references
# -o pipefail: catch failures in pipes
# ── Temporary working directory ───────────────────────────────────────────────
# All temporary files (saved configs, mount points) go here.
# Cleaned up automatically on EXIT, INT, or error via trap.
TMPDIR=$(mktemp -d /tmp/arch-reset.XXXXXX)
trap 'rm -rf "$TMPDIR"' EXIT
# Simple "pause and wait for user" helper
pause() { read -rp "Press ENTER to continue..."; }
# ── Banner ────────────────────────────────────────────────────────────────────
echo "======================================="
echo " M-Archy System Reset"
echo "======================================="
echo "This will:"
echo " • Delete and recreate the root (@) btrfs subvolume"
echo " • Reinstall base system packages from scratch"
echo " • Clear all user ~/.config directories (auth keys preserved)"
echo " • Preserve home directories, passwords, and FIDO2 login keys"
echo ""
# ── Required tools in live environment ──────────────────────────────────────
# WHY: The archiso live environment has minimal packages. We need these specific
# tools that may not be pre-installed on the ISO:
# - cryptsetup: to detect and unlock LUKS partitions
# - btrfs-progs: to manage btrfs subvolumes
# - jq: used by some sub-scripts
# - libfido2: for FIDO2/token-based LUKS unlock
# -yd: sync database, download-only skip (don't install dependencies blindly)
pacman -Syd --noconfirm cryptsetup btrfs-progs jq libfido2
# ── Drive selection ──────────────────────────────────────────────────────────
# List all block devices so the user can see what's available
lsblk
echo ""
read -rp "Enter drive to reset (e.g., /dev/sda): " DRIVE
# Assume standard partition layout from arch-autoinstall.sh:
# partition 1 = EFI/boot
# partition 2 = root (possibly LUKS)
# NOTE: This assumes a non-NVMe drive. NVMe would need ${DRIVE}p2 etc.
ROOT_PART="${DRIVE}2"
EFI_PART="${DRIVE}1"
echo ""
echo "WARNING: The root subvolume on $ROOT_PART will be DELETED and reinstalled."
echo " User home directories will be preserved."
echo " App configs (~/.config) will be wiped (Yubico auth keys excepted)."
echo ""
# Require explicit "YES" (uppercase) to prevent accidental data loss
read -rp "Type YES to continue: " _CONFIRM
[[ "$_CONFIRM" == "YES" ]] || { echo "Aborted."; exit 1; }
# ── LUKS detection and unlock ────────────────────────────────────────────────
# Default assumption: root partition is NOT encrypted — MAPPER_DEV points to it directly.
# If LUKS is detected, MAPPER_DEV is updated to /dev/mapper/cryptroot after unlock.
MAPPER_DEV="$ROOT_PART"
# cryptsetup isLuks: exits 0 if the partition has a LUKS header, non-zero if not.
if cryptsetup isLuks "$ROOT_PART" 2>/dev/null; then
echo ""
echo "Partition $ROOT_PART is LUKS2-encrypted."
echo "Select unlock method:"
echo " 1) Try enrolled token (FIDO2/TPM2) first, fall back to passphrase [recommended]"
echo " 2) Passphrase only"
echo " 3) Enrolled token only (FIDO2/TPM2)"
read -rp "Choice [1]: " _UNLOCK
_UNLOCK="${_UNLOCK:-1}" # Default to option 1 if user just presses Enter
case "$_UNLOCK" in
1)
# Best of both worlds: try hardware token first, fall back to passphrase.
# This works whether or not the user has their FIDO2 key with them.
echo "Insert FIDO2 key if using one, then press ENTER..."
pause
# --token-only: use enrolled tokens (FIDO2, TPM2) without prompting passphrase
# If token unlock fails (key not present / not enrolled), fall back to password
if ! cryptsetup open --token-only "$ROOT_PART" cryptroot 2>/dev/null; then
echo "Token unlock failed — enter passphrase..."
cryptsetup open "$ROOT_PART" cryptroot
fi
;;
2)
# Traditional passphrase-only unlock — no hardware key needed
cryptsetup open "$ROOT_PART" cryptroot
;;
3)
# Token-only mode — will fail if key isn't present/enrolled
echo "Insert FIDO2 key and press ENTER..."
pause
cryptsetup open --token-only "$ROOT_PART" cryptroot
;;
*)
echo "Invalid choice, using passphrase..."
cryptsetup open "$ROOT_PART" cryptroot
;;
esac
# After unlock, the decrypted device is available at /dev/mapper/cryptroot
MAPPER_DEV="/dev/mapper/cryptroot"
echo "Partition unlocked."
fi
# ── Detect installed kernel from EFI partition ───────────────────────────────
# WHY: We need to know which kernel to reinstall (linux, linux-lts, linux-zen, etc.).
# HOW: Mount the EFI partition temporarily and look for vmlinuz-* kernel images.
# The filename suffix is the package name (e.g. vmlinuz-linux-lts → linux-lts).
TMPBOOT=$(mktemp -d /tmp/arch-reset-boot.XXXXXX)
mount "$EFI_PART" "$TMPBOOT"
KERNEL_PKG="linux" # Default to mainline kernel
for _img in "$TMPBOOT"/vmlinuz-*; do
# Loop over glob; first match wins (strip the vmlinuz- prefix)
[[ -f "$_img" ]] && KERNEL_PKG=$(basename "$_img" | sed 's/^vmlinuz-//') && break
done
umount "$TMPBOOT"; rmdir "$TMPBOOT"
echo "Detected kernel: $KERNEL_PKG"
# ── Detect GPU from live hardware ────────────────────────────────────────────
# WHY: Different GPU vendors need different Xorg/DRM driver packages.
# lspci reads the PCI device list; grep filters for display-related devices
# (VGA=traditional GPU, 3D=compute GPU/integrated graphics).
# GPU_PKGS will be passed to pacstrap as an extra package.
GPU_INFO=$(lspci 2>/dev/null | grep -E "VGA|3D" || true)
GPU_PKGS=""
if echo "$GPU_INFO" | grep -qi nvidia; then GPU_PKGS="nvidia-open" # NVIDIA open-source driver
elif echo "$GPU_INFO" | grep -qi amd; then GPU_PKGS="xf86-video-amdgpu" # AMD open-source driver
elif echo "$GPU_INFO" | grep -qi intel; then GPU_PKGS="xf86-video-intel" # Intel driver
fi
# If no GPU detected, GPU_PKGS remains empty and the kernel's DRM will handle it
# ── Mount btrfs top-level ────────────────────────────────────────────────────
# WHY: We need to access the raw btrfs filesystem (not via a subvolume) so we
# can manipulate subvolumes directly.
# HOW: subvolid=5 is always the btrfs top-level (root of all subvolumes).
BTRFS_MNT="$TMPDIR/btrfs"
mkdir -p "$BTRFS_MNT"
mount -o subvolid=5 "$MAPPER_DEV" "$BTRFS_MNT"
# ── Save critical configuration from @ ──────────────────────────────────────
# WHY: We're about to destroy the @ subvolume, which contains /etc.
# We need to preserve critical system files so the reinstalled system
# has the same users, passwords, locales, hostname, and boot config.
echo "Saving system configuration..."
SAVED="$TMPDIR/saved"
# _save: helper to copy a file/dir from the old @ subvolume to our save area.
# Skips silently if the source doesn't exist (not all systems have all configs).
_save() {
local src="$BTRFS_MNT/@/etc/$1"
local dst="$SAVED/etc/$1"
[[ -e "$src" ]] || return 0 # Skip if source doesn't exist
mkdir -p "$(dirname "$dst")"
cp -a "$src" "$dst" # -a: archive mode (preserves permissions, timestamps, symlinks)
}
# User and group databases — contain usernames, UIDs, group memberships
_save passwd
_save shadow # Hashed passwords (mode 000 — only root can read)
_save group
_save gshadow # Group passwords (mode 000)
# Privilege escalation config
_save sudoers
_save sudoers.d
# PAM (Pluggable Authentication Modules) — contains FIDO2 PAM config
# WHY: Must preserve this so FIDO2 login works after reset
_save pam.d
# System identity and locale
_save hostname
_save locale.conf
_save locale.gen
_save vconsole.conf # Console keymap and font settings
# Boot configuration
_save fstab # Mount point definitions (UUIDs etc.)
_save mkinitcpio.conf # Initramfs hooks config (encryption hooks must be preserved)
_save mkinitcpio.conf.d
_save default/grub # GRUB cmdline (has cryptdevice= or rd.luks.name= parameters)
# Network connection profiles
_save NetworkManager # Contains /etc/NetworkManager/system-connections/ WiFi passwords
# Save timezone symlink target as plain text because symlinks can't cross roots.
# The symlink /etc/localtime → /usr/share/zoneinfo/Europe/Vienna would be broken
# if we copied the symlink itself into a different root.
{ readlink "$BTRFS_MNT/@/etc/localtime" 2>/dev/null || echo "/usr/share/zoneinfo/UTC"; } \
> "$SAVED/timezone"
# ── Clear ~/.config in @home (preserve auth-critical subdirs) ────────────────
# WHY: After a system reset, old app configs may be incompatible with the
# freshly installed versions. Clearing them gives apps a clean start.
# EXCEPTION: Yubico/ and pam-u2f/ contain FIDO2/U2F PAM key files.
# Deleting these would break FIDO2 login for all users after the reset.
echo "Clearing user app configs..."
PRESERVED_CONFIG_DIRS=("Yubico" "pam-u2f")
# Iterate over all user home directories in the @home subvolume
for _homedir in "$BTRFS_MNT/@home"/*/; do
[[ -d "$_homedir" ]] || continue # Skip if glob matched nothing
_user=$(basename "$_homedir")
_cfgdir="$_homedir/.config"
[[ -d "$_cfgdir" ]] || continue # Skip users with no .config
echo " Clearing ~/.config for: $_user"
# Delete all top-level files in .config (not directories — handled separately)
# -mindepth 1: don't delete .config itself
# -maxdepth 1: only direct children
# ! -type d: only files, not directories
find "$_cfgdir" -mindepth 1 -maxdepth 1 ! -type d -delete
# Delete subdirectories, except preserved ones
# Using process substitution + read with NUL delimiter for paths with spaces
while IFS= read -r -d '' _subdir; do
_dname=$(basename "$_subdir")
_skip=false
# Check if this directory is in the preserved list
for _keep in "${PRESERVED_CONFIG_DIRS[@]}"; do
[[ "$_dname" == "$_keep" ]] && _skip=true && break
done
$_skip || rm -rf "$_subdir" # Delete if not in preserved list
done < <(find "$_cfgdir" -mindepth 1 -maxdepth 1 -type d -print0)
done
# ── Delete @ and recreate fresh ──────────────────────────────────────────────
# WHY: btrfs subvolume delete removes the entire @ filesystem tree instantly.
# Creating a new empty @ gives us a clean slate for pacstrap.
echo "Deleting root subvolume @..."
btrfs subvolume delete "$BTRFS_MNT/@"
echo "Creating fresh root subvolume @..."
btrfs subvolume create "$BTRFS_MNT/@"
# ── Mount for installation ───────────────────────────────────────────────────
# Mount the new empty @ subvolume and the existing @home subvolume for pacstrap.
umount "$BTRFS_MNT"
mount -o subvol=@ "$MAPPER_DEV" /mnt
mkdir -p /mnt/home
mount -o subvol=@home "$MAPPER_DEV" /mnt/home
mkdir -p /mnt/boot
mount "$EFI_PART" /mnt/boot
# ── Pacstrap base system ─────────────────────────────────────────────────────
# WHY: pacstrap installs packages into the new root at /mnt.
# We install only the minimal base needed to boot and manage the system.
# The full package set is reinstalled via tui-install.sh after first boot.
echo "Reinstalling base system (this will take a while)..."
# shellcheck disable=SC2086 — GPU_PKGS is intentionally word-split here
pacstrap /mnt \
base base-devel "$KERNEL_PKG" linux-firmware vim zsh git networkmanager grub efibootmgr \
btrfs-progs cryptsetup libfido2 pam-u2f sudo less jq $GPU_PKGS
# ── Restore saved configuration ──────────────────────────────────────────────
# WHY: The fresh @ has default /etc files from pacstrap. We overwrite them
# with the saved versions to restore users, passwords, hostname, etc.
echo "Restoring system configuration..."
# _restore: copy from save area into the new /mnt/etc
_restore() {
local src="$SAVED/etc/$1"
local dst="/mnt/etc/$1"
[[ -e "$src" ]] || return 0 # Skip if we didn't save this file
mkdir -p "$(dirname "$dst")"
cp -a "$src" "$dst"
}
# Auth files get explicit permission modes with `install` rather than `cp -a`
# because cp -a preserves the original permissions which may be wrong after
# copying across filesystem boundaries.
# passwd/group: world-readable (644), shadow/gshadow: inaccessible to non-root (000)
[[ -f "$SAVED/etc/passwd" ]] && install -m 644 "$SAVED/etc/passwd" /mnt/etc/passwd
[[ -f "$SAVED/etc/group" ]] && install -m 644 "$SAVED/etc/group" /mnt/etc/group
[[ -f "$SAVED/etc/shadow" ]] && install -m 000 "$SAVED/etc/shadow" /mnt/etc/shadow
[[ -f "$SAVED/etc/gshadow" ]] && install -m 000 "$SAVED/etc/gshadow" /mnt/etc/gshadow
# Restore everything else using the _restore helper
_restore sudoers
_restore sudoers.d
_restore pam.d # FIDO2 PAM rules restored here
_restore hostname
_restore locale.conf
_restore locale.gen
_restore vconsole.conf
_restore fstab # CRITICAL: without fstab the system won't mount properly on boot
_restore mkinitcpio.conf # CRITICAL: must have correct encryption hooks
_restore mkinitcpio.conf.d
_restore default/grub # CRITICAL: must have correct cryptdevice= kernel parameters
_restore NetworkManager
# Restore the timezone symlink using the saved target path
TZ_TARGET=$(cat "$SAVED/timezone")
ln -sf "$TZ_TARGET" /mnt/etc/localtime
# ── Chroot: regenerate initramfs, GRUB menu, services ───────────────────────
# WHY: mkinitcpio and grub-mkconfig must run from inside the new root because
# they reference files and kernel modules specific to that root.
# Running them from outside would produce incorrect initramfs/GRUB configs.
echo "Finalizing inside chroot..."
arch-chroot /mnt /bin/bash <<'CHROOT_EOF'
set -euo pipefail
# Regenerate locale files (locale.gen was restored, re-run locale-gen to apply)
locale-gen
# Sync hardware clock from system time
hwclock --systohc
# Enable NetworkManager so the system has network on first boot
systemctl enable NetworkManager
# Regenerate initramfs for ALL installed kernels (-P = all presets)
# WHY: The restored mkinitcpio.conf has the correct hooks (encrypt, btrfs, etc.)
# but the initramfs itself was in the old @ and is now gone.
mkinitcpio -P
# Regenerate GRUB boot menu
# WHY: GRUB needs a fresh grub.cfg pointing to the new kernel images.
# The restored /etc/default/grub already has the correct cmdline.
grub-mkconfig -o /boot/grub/grub.cfg
# Fix home directory ownership using the restored UID/GID from /etc/passwd
# WHY: @home was preserved but the new @ has a fresh uid/gid namespace until
# /etc/passwd is restored. Now that passwd is in place, we re-apply
# ownership to ensure home dirs are accessible by the right users.
# IFS=: splits /etc/passwd fields by colon; fields: name:pass:uid:gid:gecos:home:shell
while IFS=: read -r _uname _ _uid _gid _ _home _; do
# Only fix user accounts (UID >= 1000), not system accounts
(( _uid >= 1000 )) && [[ -d "$_home" ]] && chown -R "${_uid}:${_gid}" "$_home" || true
done < /etc/passwd
CHROOT_EOF
echo ""
echo "======================================="
echo " Reset complete!"
echo "======================================="
echo " umount -R /mnt && reboot"