rcheck: A periodic remote monitoring tool.

[rcheck] / check.d / disks

#! /bin/bash
###
### Check the health of attached physical disks.

set -e

## Build a list of actual disk devices according to their buses.
disks=
for p in /sys/class/block/*; do
  bus=none devtype=none idtype=none
  while read code assg; do
    case "$assg" in
      DEVNAME=*) name=${assg#*=} ;;
      ID_BUS=*) bus=${assg#*=} ;;
      DEVTYPE=*) devtype=${assg#*=} ;;
      ID_TYPE=*) idtype=${assg#*=} ;;
    esac
  done <<EOF
$(udevadm info --query=all --path=$p)
EOF
  case "$bus,$devtype,$idtype" in
    ata,disk,disk | scsi,disk,disk) disks=${disks+$disks }$name ;;
  esac
done

## Now go through each disk.
for disk in $disks; do
  set +e; smartctl -qsilent $disk; rc=$?; set -e
  if (( $rc & 2 )); then continue; fi

  if (( $rc & 8 )); then echo "W: SMART reports disk $disk failing"; fi

  smartctl -A $disk |
  while read id attr flag value worst thresh type upd when raw; do
    case "$id" in *[!0-9]*) continue ;; ?*) ;; *) continue ;; esac
    while :; do
      case "$value" in 0*?) value=${value#0} ;; *) break ;; esac;
    done
    while :; do
      case "$thresh" in 0*?) thresh=${thresh#0} ;; *) break ;; esac;
    done
    case "$attr,$raw,$when" in
      Current_Pending_Sector,*[!0]*,*)
	echo "W: disk $disk has $raw pending sector(s)"
	;;
      Offline_Uncorrectable,*[!0]*,*)
	echo "W: disk $disk has $raw offline-uncorrectable sector(s)"
	;;
      *,*,FAILING_NOW)
	echo "W: disk $disk attribute $attr failing (value = $raw)"
	;;
      *)
	if (( $value < $thresh )); then
	  echo "I: disk $disk attribute $attr below thresh (value = $raw)"
	fi
	;;
    esac
  done
done