prole/prole-net/init-prole-dns.sh

#!/usr/bin/env bash
set -euo pipefail

# init-prole-dns.sh
# Sync prole.org host records into Samba AD DNS using samba-tool.
#
# Host source:
#   1) Postgres (preferred when available):
#      SELECT host_name, ip_address FROM prole.hosts;
#   2) Fallback file:
#      /etc/prole-dns-sync/hosts.txt  (one hostname per line; FQDN or short)
#
# Commands:
#   start    - run sync now + install hourly cron job
#   stop     - remove cron job
#   status   - show last run time, last exit, and journal messages
#   sync     - run one sync pass
#   flush    - clear local state (forces "unknown" status); does NOT delete DNS records
#   list|show|details - show configured hostnames and detected DC info
#
# Options:
#   -db=HOST:PORT/DBNAME   default: localhost:5432/knoe-db
#
# Environment overrides (optional):
#   AD_DNS_ZONE        default: prole.org
#   AD_DNS_SERVER      default: auto-detect (hostname -f)
#   AD_DNS_SERVER_IP   default: auto-detect via getent hosts
#   PUBLIC_DNS_SERVER  default: first NS of prole.org from system resolv OR "ns1.name.com"
#   DRY_RUN            if set to 1, prints actions without changing AD DNS

PROG="init_prole_dns"
STATE_DIR="/var/lib/${PROG}"
STATE_FILE="${STATE_DIR}/state.env"
CRON_FILE="/etc/cron.d/${PROG}"
CONF_DIR="/etc/prole-dns-sync"
HOSTS_FILE="${CONF_DIR}/hosts.txt"
LOG_TAG="${PROG}"

AD_DNS_ZONE="${AD_DNS_ZONE:-prole.org}"
DRY_RUN="${DRY_RUN:-0}"

DEFAULT_DB_ENDPOINT="localhost:5432/knoe-db"
DB_ENDPOINT="$DEFAULT_DB_ENDPOINT"

mkdir -p "$STATE_DIR" "$CONF_DIR"
chmod 700 "$STATE_DIR"

log() {
  local msg="$*"
  logger -t "$LOG_TAG" -- "$msg" 2>/dev/null || true
  echo "$msg"
}

die() {
  log "ERROR: $*"
  exit 1
}

need_root() {
  [[ "${EUID:-$(id -u)}" -eq 0 ]] || die "Must run as root."
}

have_cmd() {
  command -v "$1" >/dev/null 2>&1
}

detect_dc_fqdn() {
  local fqdn
  fqdn="$(hostname -f 2>/dev/null || true)"
  [[ -n "$fqdn" ]] || die "Unable to detect DC FQDN (hostname -f failed)."
  echo "$fqdn"
}

detect_dc_ip() {
  local dc_fqdn="$1"
  local ip
  ip="$(getent ahostsv4 "$dc_fqdn" 2>/dev/null | awk '{print $1; exit}' || true)"
  [[ -n "$ip" ]] || die "Unable to detect DC IP for $dc_fqdn (getent ahostsv4 failed)."
  echo "$ip"
}

detect_public_dns_server() {
  local ns=""
  if have_cmd dig; then
    ns="$(dig +short NS "${AD_DNS_ZONE}" 2>/dev/null | head -n1 | sed 's/\.$//' || true)"
  fi
  [[ -n "$ns" ]] || ns="ns1.name.com"
  echo "$ns"
}

# Parse DB endpoint of form host:port/dbname
parse_db_endpoint() {
  local ep="$1"
  local hostport="${ep%%/*}"
  local dbname="${ep#*/}"
  local host="${hostport%%:*}"
  local port="${hostport#*:}"

  if [[ "$hostport" == "$host" ]]; then
    # no port specified
    port="5432"
  fi
  [[ -n "$host" && -n "$port" && -n "$dbname" ]] || return 1

  echo "$host|$port|$dbname"
}

psql_can_connect() {
  have_cmd psql || return 1
  local parsed
  parsed="$(parse_db_endpoint "$DB_ENDPOINT")" || return 1
  local host="${parsed%%|*}"; parsed="${parsed#*|}"
  local port="${parsed%%|*}"; local db="${parsed##*|}"

  # Allow interactive password prompting if needed:
  # We DO NOT add -w (never prompt). We add a short statement timeout.
  # If auth requires a password and none is cached, this command will prompt.
  psql -h "$host" -p "$port" -d "$db" -v ON_ERROR_STOP=1 -At \
    -c "SET statement_timeout='5s'; SELECT 1;" >/dev/null 2>&1
}

# Functions to check if a string is an IP address
is_ipv4() {
  [[ $1 =~ ^[0-9]{1,3}(\.[0-9]{1,3}){3}$ ]]
}

is_ipv6() {
  [[ $1 == *:* ]]
}

is_ip() {
  is_ipv4 "$1" || is_ipv6 "$1"
}

read_hosts_file() {
  [[ -f "$HOSTS_FILE" ]] || die "Missing $HOSTS_FILE and DB not available. Create it with: hostname [TAB/SPACE] ip_or_target"
  grep -vE '^\s*($|#)' "$HOSTS_FILE" | awk '{print $1 "|" $2}' || true
}

read_hosts_db() {
  local parsed
  parsed="$(parse_db_endpoint "$DB_ENDPOINT")" || die "Invalid -db endpoint '$DB_ENDPOINT' (expected host:port/dbname)"
  local host="${parsed%%|*}"; parsed="${parsed#*|}"
  local port="${parsed%%|*}"; local db="${parsed##*|}"

  # This will prompt for password if required by libpq/pg_hba.
  # Output format: host_name|ip_address
  psql -h "$host" -p "$port" -d "$db" -v ON_ERROR_STOP=1 -At -F '|' \
    -c "SELECT host_name, ip_address FROM prole.hosts ORDER BY host_name;"
}

normalize_host() {
  local raw="$1"
  raw="${raw%%#*}"
  # Trim whitespace
  raw="$(echo "$raw" | xargs)"
  [[ -n "$raw" ]] || return 1

  local fqdn="$raw"
  # If it's a short name, append the zone
  if [[ "$raw" != *.* ]]; then
    fqdn="${raw}.${AD_DNS_ZONE}"
  fi

  # Validate it's within our zone
  if [[ "$fqdn" != *".${AD_DNS_ZONE}" && "$fqdn" != "${AD_DNS_ZONE}" ]]; then
    # We might be reading a line that has two columns, but normalize_host is called with only the first.
    # If the user provided a FQDN outside the zone, it's an error for an A/CNAME record name in this script's context.
    die "Host '$raw' is not within zone '${AD_DNS_ZONE}' (got fqdn '$fqdn')."
  fi

  local rel="${fqdn%.$AD_DNS_ZONE}"
  rel="${rel%.}"

  echo "$rel|$fqdn"
}

dig_records() {
  local dns_server="$1"
  local fqdn="$2"
  local a_records aaaa_records

  a_records="$(dig +short @"$dns_server" A "$fqdn" 2>/dev/null | tr '\n' ' ' | xargs || true)"
  aaaa_records="$(dig +short @"$dns_server" AAAA "$fqdn" 2>/dev/null | tr '\n' ' ' | xargs || true)"
  echo "$a_records|$aaaa_records"
}

samba_query_rr() {
  local dc="$1" zone="$2" name="$3" rrtype="$4"
  samba-tool dns query "$dc" "$zone" "$name" "$rrtype" --use-kerberos=required 2>/dev/null \
    | awk '
      BEGIN{inrec=0}
      /Records:/ {inrec=1; next}
      inrec && NF>=1 { print $0 }
    ' \
    | tr '\n' ' ' | xargs || true
}

samba_add_rr() {
  local dc="$1" zone="$2" name="$3" rrtype="$4" value="$5"
  if [[ "$DRY_RUN" == "1" ]]; then
    log "[DRY_RUN] samba-tool dns add $dc $zone $name $rrtype $value --use-kerberos=required"
    return 0
  fi
  local out
  if out=$(samba-tool dns add "$dc" "$zone" "$name" "$rrtype" "$value" --use-kerberos=required 2>&1); then
    return 0
  else
    if [[ "$out" == *"WERR_DNS_ERROR_RECORD_ALREADY_EXISTS"* ]]; then
      log "INFO: Record $name $rrtype $value exists"
      return 0
    fi
    echo "$out" >&2
    return 1
  fi
}

samba_update_rr() {
  local dc="$1" zone="$2" name="$3" rrtype="$4" old="$5" new="$6"
  if [[ "$DRY_RUN" == "1" ]]; then
    log "[DRY_RUN] samba-tool dns update $dc $zone $name $rrtype $old $new --use-kerberos=required"
    return 0
  fi
  local out
  if out=$(samba-tool dns update "$dc" "$zone" "$name" "$rrtype" "$old" "$new" --use-kerberos=required 2>&1); then
    return 0
  else
    if [[ "$out" == *"WERR_DNS_ERROR_RECORD_ALREADY_EXISTS"* ]]; then
      log "INFO: Record $name $rrtype $new exists"
      return 0
    fi
    echo "$out" >&2
    return 1
  fi
}

sync_rr_set() {
  local dc="$1" zone="$2" name_rel="$3" fqdn="$4" rrtype="$5" desired_values="$6"
  [[ -n "$desired_values" ]] || return 0

  local existing
  existing="$(samba_query_rr "$dc" "$zone" "$name_rel" "$rrtype")"

  for val in $desired_values; do
    if [[ " $existing " == *" $val "* ]]; then
      log "OK: $fqdn $rrtype already has $val"
    else
      local first_old
      first_old="$(echo "$existing" | awk '{print $1}' || true)"
      if [[ -n "$first_old" ]]; then
        log "UPDATE: $fqdn $rrtype $first_old -> $val"
        samba_update_rr "$dc" "$zone" "$name_rel" "$rrtype" "$first_old" "$val" || samba_add_rr "$dc" "$zone" "$name_rel" "$rrtype" "$val"
      else
        log "ADD: $fqdn $rrtype $val"
        samba_add_rr "$dc" "$zone" "$name_rel" "$rrtype" "$val"
      fi
    fi
  done
}

write_state() {
  local key="$1" val="$2"
  if [[ -f "$STATE_FILE" ]] && grep -qE "^${key}=" "$STATE_FILE"; then
    sed -i "s|^${key}=.*|${key}=${val}|g" "$STATE_FILE"
  else
    echo "${key}=${val}" >> "$STATE_FILE"
  fi
  chmod 600 "$STATE_FILE"
}

load_state() {
  [[ -f "$STATE_FILE" ]] && source "$STATE_FILE" || true
}

install_cron() {
  local script_path
  script_path="$(readlink -f "$0")"
  cat > "$CRON_FILE" <<EOF
# Managed by $PROG. Runs hourly.
SHELL=/bin/bash
PATH=/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin
0 * * * * root $script_path -db=$DB_ENDPOINT sync >/dev/null 2>&1
EOF
  chmod 644 "$CRON_FILE"
  log "Installed hourly cron job at $CRON_FILE"
}

remove_cron() {
  if [[ -f "$CRON_FILE" ]]; then
    rm -f "$CRON_FILE"
    log "Removed cron job $CRON_FILE"
  else
    log "Cron job not present ($CRON_FILE)"
  fi
}

choose_host_source() {
  # Echo "db" if we can connect; otherwise "file"
  if have_cmd psql; then
    # We want to allow interactive prompting; connectivity test here should not block forever.
    # We'll do a *non-interactive* quick check first: if it fails, fall back to file.
    # (If you want forced DB prompting, run: sudo PSQL_FORCE_PROMPT=1 ... sync)
    if psql_can_connect; then
      echo "db"
      return 0
    fi
  fi
  echo "file"
}

cmd_list() {
  local dc_fqdn dc_ip public_dns source
  dc_fqdn="${AD_DNS_SERVER:-$(detect_dc_fqdn)}"
  dc_ip="${AD_DNS_SERVER_IP:-$(detect_dc_ip "$dc_fqdn")}"
  public_dns="${PUBLIC_DNS_SERVER:-$(detect_public_dns_server)}"
  source="$(choose_host_source)"

  echo "AD zone        : $AD_DNS_ZONE"
  echo "AD DC (server) : $dc_fqdn"
  echo "AD DC IP       : $dc_ip"
  echo "Public resolver: $public_dns"
  echo "DB endpoint    : $DB_ENDPOINT"
  echo "Hosts file     : $HOSTS_FILE"
  echo "Host source    : $source"
  echo

  echo "Hostnames:"
  if [[ "$source" == "db" ]]; then
    read_hosts_db | while IFS='|' read -r host ip; do
      [[ -n "${host// }" ]] || continue
      local norm rel fqdn
      norm="$(normalize_host "$host")" || continue
      rel="${norm%%|*}"
      fqdn="${norm##*|}"
      if [[ -n "${ip// }" ]]; then
        if is_ip "$ip"; then
          echo "  - $fqdn  (name='$rel', ip=$ip)"
        else
          echo "  - $fqdn  (name='$rel', cname=$ip)"
        fi
      else
        echo "  - $fqdn  (name='$rel', ip=<lookup>)"
      fi
    done
  else
    read_hosts_file | while IFS='|' read -r host ip; do
      [[ -n "${host// }" ]] || continue
      local norm rel fqdn
      norm="$(normalize_host "$host")" || continue
      rel="${norm%%|*}"
      fqdn="${norm##*|}"
      if [[ -n "${ip// }" ]]; then
        if is_ip "$ip"; then
          echo "  - $fqdn  (name='$rel', ip=$ip)"
        else
          echo "  - $fqdn  (name='$rel', cname=$ip)"
        fi
      else
        echo "  - $fqdn  (name='$rel', ip=<lookup>)"
      fi
    done
  fi
}

cmd_start() {
  need_root
  cmd_sync
  install_cron
}

cmd_stop() {
  need_root
  remove_cron
}

cmd_flush() {
  need_root
  rm -f "$STATE_FILE"
  log "Cleared local state ($STATE_FILE). DNS records were NOT changed."
}

cmd_status() {
  load_state
  echo "State file: $STATE_FILE"
  echo "  last_run_epoch : ${LAST_RUN_EPOCH:-<unknown>}"
  if [[ -n "${LAST_RUN_EPOCH:-}" ]]; then
    date -d "@${LAST_RUN_EPOCH}" +"  last_run_time  : %F %T %Z"
  else
    echo "  last_run_time  : <unknown>"
  fi
  echo "  last_exit      : ${LAST_EXIT:-<unknown>}"
  echo "  last_source    : ${LAST_SOURCE:-<unknown>}"
  echo "  last_public_ns : ${LAST_PUBLIC_DNS:-<unknown>}"
  echo "  last_dc        : ${LAST_DC:-<unknown>}"
  echo "  db_endpoint    : ${DB_ENDPOINT}"
  echo
  echo "Recent logs:"
  echo "----------------------------------------"
  journalctl -t "$LOG_TAG" -n 50 --no-pager 2>/dev/null || \
    echo "(journalctl tag '$LOG_TAG' not available; try: sudo journalctl -n 50 | grep $LOG_TAG)"
}

cmd_sync() {
  need_root
  have_cmd samba-tool || die "samba-tool not found (run this on the Samba AD DC)."
  have_cmd dig || die "dig not found. Install bind9-dnsutils or dnsutils."

  local dc_fqdn dc_ip public_dns source
  dc_fqdn="${AD_DNS_SERVER:-$(detect_dc_fqdn)}"
  dc_ip="${AD_DNS_SERVER_IP:-$(detect_dc_ip "$dc_fqdn")}"
  public_dns="${PUBLIC_DNS_SERVER:-$(detect_public_dns_server)}"
  source="$(choose_host_source)"

  write_state "LAST_RUN_EPOCH" "$(date +%s)"
  write_state "LAST_PUBLIC_DNS" "$public_dns"
  write_state "LAST_DC" "$dc_fqdn"
  write_state "LAST_SOURCE" "$source"

  log "Starting sync: zone=${AD_DNS_ZONE}, dc=${dc_fqdn} (${dc_ip}), public_dns=${public_dns}, source=${source}, db=${DB_ENDPOINT}, dry_run=${DRY_RUN}"

  # Verify Kerberos ticket if running as root via sudo (common case)
  if [[ -z "${KRB5CCNAME:-}" ]]; then
    if [[ -n "${SUDO_UID:-}" ]]; then
      local user_candidate="FILE:/tmp/krb5cc_${SUDO_UID}"
      if [[ -f "${user_candidate#FILE:}" ]]; then
        # Try to use it, but if it's not readable by root (Operation not permitted),
        # we might need to tell the user to kinit as root or fix permissions.
        if [[ -r "${user_candidate#FILE:}" ]]; then
          export KRB5CCNAME="$user_candidate"
          log "INFO: Automatically using Kerberos cache from sudoer: $KRB5CCNAME"
        else
          log "WARN: Found Kerberos cache at ${user_candidate}, but it is not readable by root. Try: sudo kinit administrator@${AD_DNS_ZONE^^}"
        fi
      fi
    fi
  fi

  # If KRB5CCNAME is set but not readable, it will fail later.
  if [[ -n "${KRB5CCNAME:-}" && ! -r "${KRB5CCNAME#FILE:}" ]]; then
     log "WARN: Kerberos cache ${KRB5CCNAME} is not readable by current user ($(id -u)). samba-tool will likely fail."
     log "HINT: Try running 'kinit' as root, or use 'sudo kinit administrator@${AD_DNS_ZONE^^}'"
  fi

  # If KRB5CCNAME not set, check if default cache exists (for root kinit case)
  if [[ -z "${KRB5CCNAME:-}" ]]; then
    local default_cache="/tmp/krb5cc_$(id -u)"
    if [[ ! -f "$default_cache" ]]; then
      log "HINT: No Kerberos credentials found. If samba-tool fails, run: sudo -E KRB5CCNAME=/tmp/krb5cc_\$(id -u) $0 sync"
    fi
  fi

  local exit_code=0

  # Helper to process a single host entry
  process_host() {
    local host="$1" ip="$2"
    [[ -n "${host// }" ]] || return 0
    local norm rel fqdn
    norm="$(normalize_host "$host")" || return 0
    rel="${norm%%|*}"
    fqdn="${norm##*|}"

    # Skip zone apex
    if [[ "$fqdn" == "$AD_DNS_ZONE" ]]; then
      log "INFO: Skipping zone apex $fqdn"
      return 0
    fi

    local a_vals="" aaaa_vals="" cname_vals=""
    if [[ -n "${ip// }" ]]; then
      if is_ip "$ip"; then
        # If a single IP is provided, classify as v4/v6
        if is_ipv6 "$ip"; then
          aaaa_vals="$ip"
        else
          a_vals="$ip"
        fi
      else
        # It's a hostname, use CNAME
        cname_vals="$ip"
      fi
    else
      # No IP in entry: resolve from public DNS
      local recs
      recs="$(dig_records "$public_dns" "$fqdn")"
      a_vals="${recs%%|*}"
      aaaa_vals="${recs##*|}"
      if [[ -z "$a_vals" && -z "$aaaa_vals" ]]; then
        log "WARN: No A/AAAA in public DNS for $fqdn (via $public_dns); skipping"
        return 0
      fi
    fi

    local ret=0
    if [[ -n "$a_vals" ]]; then
      if ! sync_rr_set "$dc_fqdn" "$AD_DNS_ZONE" "$rel" "$fqdn" "A" "$a_vals"; then ret=1; fi
    fi
    if [[ -n "$aaaa_vals" ]]; then
      if ! sync_rr_set "$dc_fqdn" "$AD_DNS_ZONE" "$rel" "$fqdn" "AAAA" "$aaaa_vals"; then ret=1; fi
    fi
    if [[ -n "$cname_vals" ]]; then
      # Samba-tool CNAME values should ideally be FQDN with trailing dot for safety, 
      # but let's see how it handles it. The issue says "samba-tool dns add .. cname"
      if ! sync_rr_set "$dc_fqdn" "$AD_DNS_ZONE" "$rel" "$fqdn" "CNAME" "$cname_vals"; then ret=1; fi
    fi
    return $ret
  }

  if [[ "$source" == "db" ]]; then
    while IFS='|' read -r host ip; do
      if ! process_host "$host" "$ip"; then exit_code=1; fi
    done < <(read_hosts_db)
  else
    while IFS='|' read -r host ip; do
      if ! process_host "$host" "$ip"; then exit_code=1; fi
    done < <(read_hosts_file)
  fi

  write_state "LAST_EXIT" "$exit_code"
  if [[ "$exit_code" -eq 0 ]]; then
    log "Sync complete: success"
  else
    log "Sync complete: errors occurred"
  fi
  return "$exit_code"
}

usage() {
  cat <<EOF
Usage: $0 [-db=host:port/dbname] <command>

Commands:
  start     Run sync now and install hourly cron job
  stop      Remove hourly cron job
  status    Show last run time, exit status, and recent logs
  sync      Run one sync pass now
  flush     Clear local state file (does not change DNS records)
  list      Show host list + DC info (alias: show, details)
  show      Alias for list
  details   Alias for list

DB option:
  -db=HOST:PORT/DBNAME   default: ${DEFAULT_DB_ENDPOINT}

Host list sources:
  1) If psql exists and DB is reachable:
     SELECT host_name, ip_address FROM prole.hosts;
     (psql may prompt for password interactively)
  2) Else: $HOSTS_FILE

Examples:
  sudo $0 sync
  sudo kinit administrator@PROLE.ORG && sudo $0 sync
  sudo -E $0 sync (preserves Kerberos ticket environment)
  sudo DRY_RUN=1 $0 sync
  sudo $0 start
  sudo $0 status
EOF
}

main() {
  # Parse -db=... and keep remaining args
  local args=()
  for a in "$@"; do
    case "$a" in
      -db=*) DB_ENDPOINT="${a#-db=}" ;;
      *) args+=("$a") ;;
    esac
  done

  local cmd="${args[0]:-}"
  case "$cmd" in
    start)   cmd_start ;;
    stop)    cmd_stop ;;
    status)  cmd_status ;;
    sync)    cmd_sync ;;
    flush)   cmd_flush ;;
    list|show|details) cmd_list ;;
    ""|-h|--help|help) usage ;;
    *) die "Unknown command: $cmd (try --help)" ;;
  esac
}

main "$@"