scripts/trigger-rcmd

   1 #!/bin/sh
   2 #
   3 # trigger-rcmp: Trigger remote commands
   4 # Copyright (c)2014-2020 Alexander Barton (alex@barton.de)
   5 #
   6 # This program is free software; you can redistribute it and/or modify
   7 # it under the terms of the GNU General Public License as published by
   8 # the Free Software Foundation; either version 2 of the License, or
   9 # (at your option) any later version.
  10 # Please read the file COPYING, README and AUTHORS for more information.
  11 #
  12
  13 BASENAME=$(basename "$0")
  14 NAME="$BASENAME"
  15 RCMD_D="/usr/local/etc/rcmd.d"
  16
  17 PATH="$PATH:/usr/local/sbin:/usr/sbin"
  18
  19 do_help() {
  20         echo "Usage: $0 [--no-wake|-n] <rcmd> [<rcmd> [...]]" >&2
  21         exit 2
  22 }
  23
  24 do_rcmd() {
  25         SYSTEM="localhost"
  26         IFACE="eth0"
  27         CMD="uptime"
  28         TIMEOUT="300"
  29         SSH_CHECKS="6"
  30         WAKE_DELAY="10s"
  31         BOOT_DELAY="1m"
  32         SHUTDOWN_TIME="+10"
  33
  34         # Read in job/configuration file, but ignore shellcheck(1) warnings:
  35         # shellcheck disable=1090
  36         if ! . "$cnf"; then
  37                 echo "$NAME: Failed to read \"$cnf\"!"
  38                 return 2
  39         fi
  40
  41         echo "$NAME: Checking system \"$SYSTEM\" ..."
  42         if ! fping -c1 -q "$SYSTEM" 2>/dev/null; then
  43                 if [ -n "$NO_WAKE" ]; then
  44                         echo "$NAME: \"$SYSTEM\" seems to be down, skipping job."
  45                         return 9
  46                 fi
  47
  48                 echo "$NAME: \"$SYSTEM\" seems to be down, wake it up ..."
  49                 if ! etherwake -b -i "$IFACE" "$(echo "$SYSTEM" | cut -d'.' -f1)"; then
  50                         echo "$NAME: Failed to wake \"$SYSTEM\"!" >&2
  51                         return 3
  52                 fi
  53                 echo "$NAME: Waiting for \"$SYSTEM\" to respond ..."
  54                 for i in $(seq 1 $TIMEOUT); do
  55                         fping -c1 -q "$SYSTEM" 2>/dev/null && break
  56                         sleep 1s
  57                 done
  58                 if ! fping -c1 -q "$SYSTEM" 2>/dev/null; then
  59                         echo "$NAME: Failed to wake \"$SYSTEM\"!" >&2
  60                         return 4
  61                 fi
  62                 sleep "$WAKE_DELAY"
  63                 rsys_was_running=0
  64         else
  65                 echo "$NAME: \"$SYSTEM\" is already alive, ok."
  66                 rsys_was_running=1
  67         fi
  68
  69         for i in $(seq 1 $SSH_CHECKS); do
  70                 echo "$NAME: Checking SSH connection [publickey] ($i/$SSH_CHECKS) ..."
  71                 if ssh -o PreferredAuthentications=publickey -q "$SYSTEM" true >/dev/null 2>&1; then
  72                         echo "$NAME: Ok, SSH seems to be available on \"$SYSTEM\"."
  73                         break
  74                 fi
  75                 echo "$NAME: SSH on \"$SYSTEM\" not ready. Waiting ..."
  76                 sleep 10s
  77         done
  78
  79         if [ "$rsys_was_running" -eq 0 ]; then
  80                 echo "$NAME: \"$SYSTEM\" just started up. Delaying actions for $BOOT_DELAY ..."
  81                 sleep "$BOOT_DELAY"
  82         fi
  83
  84         echo "$NAME: Calling \"$CMD\" on \"$SYSTEM\" ..."
  85         echo
  86
  87         # shellcheck disable=2029
  88         ssh -o PreferredAuthentications=publickey -q "$SYSTEM" "$CMD" </dev/null ; r=$?
  89
  90         echo
  91         echo "$NAME: Remote command ended with return code $r."
  92         sleep 2
  93
  94         if [ "$rsys_was_running" -eq 0 ]; then
  95                 rsys_users=$(ssh -q "$SYSTEM" w | tail -n +3 | wc -l)
  96                 if [ "$rsys_users" = "0" ]; then
  97                         echo "$NAME: Power off \"$SYSTEM\" again ..."
  98                         # shellcheck disable=2029
  99                         if ! ssh -o PreferredAuthentications=publickey -q "$SYSTEM" "sync; shutdown -hP $SHUTDOWN_TIME" >/dev/null 2>&1; then
 100                                 echo "$NAME: Failed to power off \"$SYSTEM\"!" >&2
 101                                 return 5
 102                         fi
 103                 else
 104                         echo "$NAME: Not shutting down \"$SYSTEM\", \c"
 105                         [ "$rsys_users" -eq 1 ] \
 106                                 && echo "There is 1 user logged in!" \
 107                                 || echo "There are $rsys_users users logged in!"
 108                 fi
 109         else
 110                 echo "$NAME: \"$SYSTEM\" was already alive, not shutting down."
 111         fi
 112         [ $r -eq 0 ] && return 0 || return 1
 113 }
 114
 115 # Defaults
 116 NO_WAKE=
 117
 118 # Check required tools
 119 for t in fping ssh etherwake lockfile-create; do
 120         if ! command -v "$t" >/dev/null 2>&1; then
 121                 echo "$NAME: Required tool \"$t\" missing, aborting!"
 122                 exit 3
 123         fi
 124 done
 125
 126 while [ $# -gt 0 ]; do
 127         case "$1" in
 128           --no-wake|-n)
 129                 NO_WAKE=1; ;;
 130           --help|-*)
 131                 do_help; ;;
 132           *)
 133                 break
 134         esac
 135         shift
 136 done
 137
 138 [ $# -gt 0 ] || do_help
 139
 140 result=0
 141 while [ $# -gt 0 ]; do
 142         cnf="$RCMD_D/$1"
 143         if [ ! -r "$cnf" ]; then
 144                 echo "$NAME: Can't read \"$cnf\"!"
 145                 r=1
 146         else
 147                 echo "$NAME: Working on \"$1\" - $(date "+%Y-%m-%d %H:%M:%S")"
 148                 NAME="$BASENAME($1)"
 149                 lck="/run/lock/$NAME-$1.lock"
 150                 if lockfile-create --quiet --use-pid --retry 2 --lock-name "$lck"; then
 151                         do_rcmd "$1" ; r=$?
 152                         echo "$NAME: Done ($r) - $(date "+%Y-%m-%d %H:%M:%S")"
 153                         rm -f "$lck"
 154                 else
 155                         echo "$NAME: Failed to acquire lock file, skipping job! ($lck)"
 156                         r=1
 157                 fi
 158                 NAME=$BASENAME
 159         fi
 160         if [ $r -ne 0 ]; then
 161                 # Exit code indicates an error. But only flag this as error
 162                 # when target should have been woken up and the indicated error
 163                 # is not "system down" (9).
 164                 [ -z "$NO_WAKE" ] || [ $r -ne 9 ] && result=1
 165         fi
 166         shift
 167 done
 168
 169 echo "$NAME: All done, exit code $result - $(date "+%Y-%m-%d %H:%M:%S")"
 170 exit $result