LibrePortal/scripts/dev/lp-install-matrix

#!/bin/bash
# Install LibrePortal across the combinations of relocatable roots and check
# each one landed where it was asked to.
#
#   sudo scripts/dev/lp-install-matrix up          # loopback disks first
#   sudo scripts/dev/lp-install-matrix 1|2|3|4|all
#   sudo scripts/dev/lp-install-matrix verify N    # re-check without reinstalling
#
# The three roots are independently relocatable, and the interesting failures
# are the ones where only ONE of them moves: paths get baked into root-owned
# helpers, a systemd unit and the CLI wrapper at install time, so a root that is
# resolved at runtime instead of baked works on a default install and silently
# points at the wrong disk on a relocated one. Testing only "everything default"
# or only "everything moved" misses exactly that.
#
#   1  system /            apps /             baseline
#   2  system /            apps disk2         app data on its own disk
#   3  system disk1        apps /             control plane on its own disk
#   4  system disk1        apps disk2         all three roots moved (backups too)
set -u

REPO="$(cd "$(dirname "$0")/../.." && pwd)"
D1="${LP_TESTDISK_MNT:-/mnt/lptest}1"
D2="${LP_TESTDISK_MNT:-/mnt/lptest}2"
LOG_DIR=/var/tmp/lp-matrix
mkdir -p "$LOG_DIR"

[[ $EUID -eq 0 ]] || { echo "lp-install-matrix: run with sudo" >&2; exit 1; }

case_dirs() {                 # $1 = case -> sets SYS/CON/BAK
    case "$1" in
        1) SYS=/libreportal-system;        CON=/libreportal-containers;        BAK=/libreportal-backups ;;
        2) SYS=/libreportal-system;        CON="$D2/libreportal-containers";   BAK=/libreportal-backups ;;
        3) SYS="$D1/libreportal-system";   CON=/libreportal-containers;        BAK=/libreportal-backups ;;
        4) SYS="$D1/libreportal-system";   CON="$D2/libreportal-containers";   BAK="$D2/libreportal-backups" ;;
        *) echo "unknown case '$1'" >&2; return 2 ;;
    esac
}

fail=0
chk() { if [[ "$2" == "$3" ]]; then printf '  ok   %-42s %s\n' "$1" "$2"
        else printf '  FAIL %-42s got %s want %s\n' "$1" "$2" "$3"; fail=1; fi; }
chk_true() { if eval "$2" >/dev/null 2>&1; then printf '  ok   %s\n' "$1"
             else printf '  FAIL %s\n' "$1"; fail=1; fi; }

# Which filesystem a path actually sits on. The point of the whole exercise:
# a root that resolved to the right STRING but the wrong DEVICE is the bug.
devof() { stat -c '%d' "$1" 2>/dev/null || echo "-"; }

verify() {
    local c="$1"; case_dirs "$c" || return 2
    fail=0
    echo "── case $c ─────────────────────────────────────────────"
    printf '  system=%s\n  apps=%s\n  backups=%s\n\n' "$SYS" "$CON" "$BAK"

    chk_true "system root exists"      "[[ -d '$SYS' ]]"
    chk_true "containers root exists"  "[[ -d '$CON' ]]"
    chk_true "backups root exists"     "[[ -d '$BAK' ]]"

    # On the intended DEVICE, not merely at the intended path.
    chk "system on expected fs"     "$(devof "$SYS")" "$(devof "$(dirname "$SYS")")"
    chk "containers on expected fs" "$(devof "$CON")" "$(devof "$(dirname "$CON")")"

    chk "system owner"     "$(stat -c '%U' "$SYS" 2>/dev/null)"  "libreportal"
    chk "containers owner" "$(stat -c '%U' "$CON" 2>/dev/null)"  "dockerinstall"

    # Baked, not resolved: the whole trust model rests on these being fixed at
    # install so the manager cannot redirect a root operation.
    local w=/usr/local/lib/libreportal/libreportal
    chk "wrapper LP_SYSTEM_DIR"     "$(grep -oP '(?<=^LP_SYSTEM_DIR=")[^"]+'     "$w" 2>/dev/null | head -1)" "$SYS"
    chk "wrapper LP_CONTAINERS_DIR" "$(grep -oP '(?<=^LP_CONTAINERS_DIR=")[^"]+' "$w" 2>/dev/null | head -1)" "$CON"

    local o=/usr/local/lib/libreportal/libreportal-ownership
    chk "ownership helper SYSTEM_DIR"     "$(grep -oP '(?<=^SYSTEM_DIR=")[^"]+'     "$o" 2>/dev/null | head -1)" "$SYS"
    chk "ownership helper CONTAINERS_DIR" "$(grep -oP '(?<=^CONTAINERS_DIR=")[^"]+' "$o" 2>/dev/null | head -1)" "$CON"

    chk_true "no unbaked __PLACEHOLDER__ left" "! grep -q '__[A-Z_]*__' $o"
    chk_true "systemd unit installed"          "[[ -f /etc/systemd/system/libreportal.service ]]"
    chk_true "CLI on PATH"                     "[[ -x /usr/local/bin/libreportal ]]"
    chk_true "scoped sudoers present"          "[[ -f /etc/sudoers.d/libreportal ]]"

    # The WebUI is the thing a person actually opens.
    local port
    port=$(su -c 'docker ps --filter name=libreportal-service --format "{{.Ports}}"' dockerinstall 2>/dev/null \
           | grep -oE '0\.0\.0\.0:[0-9]+' | head -1 | cut -d: -f2)
    if [[ -n "$port" ]]; then
        printf '  ok   %-42s http://127.0.0.1:%s\n' "WebUI container up" "$port"
        local code; code=$(curl -s -o /dev/null -w '%{http_code}' --max-time 10 "http://127.0.0.1:$port/" 2>/dev/null)
        chk "WebUI responds" "$code" "200"
        echo "$port" > "$LOG_DIR/case$c.port"
    else
        echo "  FAIL WebUI container not running"; fail=1
    fi

    echo ""
    if (( fail )); then echo "  case $c: FAILED"; else echo "  case $c: passed"; fi
    return $fail
}

install_case() {
    local c="$1"; case_dirs "$c" || return 2
    echo "== case $c: uninstalling =="
    ( cd "$REPO" && bash init.sh --unattended --skip-rootless uninstall ) \
        > "$LOG_DIR/case$c-uninstall.log" 2>&1
    echo "   uninstall rc=$? (log: $LOG_DIR/case$c-uninstall.log)"

    echo "== case $c: installing =="
    printf '   system=%s\n   apps=%s\n   backups=%s\n' "$SYS" "$CON" "$BAK"
    ( cd "$REPO" && bash init.sh --random-password --local --unattended \
        --skip-os-update --skip-prereqs \
        --system-dir="$SYS" --containers-dir="$CON" --backups-dir="$BAK" init ) \
        > "$LOG_DIR/case$c-install.log" 2>&1
    echo "   install rc=$? (log: $LOG_DIR/case$c-install.log)"
    echo ""
    verify "$c"
}

case "${1:-}" in
    up)      bash "$REPO/scripts/dev/lp-testdisk" up 2 30G ;;
    verify)  verify "${2:?case number}" ;;
    1|2|3|4) install_case "$1" ;;
    all)     rc=0; for c in 1 2 3 4; do install_case "$c" || rc=1; echo; done; exit $rc ;;
    *) sed -n '2,20p' "$0"; exit 2 ;;
esac