check.d/{disks,ntp}: Only perform the checks if relevant programs exist.
[rcheck] / check.d / disks
1 #! /bin/bash
2 ###
3 ### Check the health of attached physical disks.
4
5 set -e
6 if [ ! -x /usr/sbin/smartctl ]; then exit 0; fi
7
8 ## Build a list of actual disk devices according to their buses.
9 disks=
10 for p in /sys/class/block/*; do
11 bus=none devtype=none idtype=none
12 while read code assg; do
13 case "$assg" in
14 DEVNAME=*) name=${assg#*=} ;;
15 ID_BUS=*) bus=${assg#*=} ;;
16 DEVTYPE=*) devtype=${assg#*=} ;;
17 ID_TYPE=*) idtype=${assg#*=} ;;
18 esac
19 done <<EOF
20 $(udevadm info --query=all --path=$p)
21 EOF
22 case "$bus,$devtype,$idtype" in
23 ata,disk,disk | scsi,disk,disk) disks=${disks+$disks }$name ;;
24 esac
25 done
26
27 ## Now go through each disk.
28 for disk in $disks; do
29 set +e; smartctl -qsilent $disk; rc=$?; set -e
30 if (( $rc & 2 )); then continue; fi
31
32 if (( $rc & 8 )); then echo "W: SMART reports disk $disk failing"; fi
33
34 smartctl -A $disk |
35 while read id attr flag value worst thresh type upd when raw; do
36 case "$id" in *[!0-9]*) continue ;; ?*) ;; *) continue ;; esac
37 while :; do
38 case "$value" in 0*?) value=${value#0} ;; *) break ;; esac;
39 done
40 while :; do
41 case "$thresh" in 0*?) thresh=${thresh#0} ;; *) break ;; esac;
42 done
43 case "$attr,$raw,$when" in
44 Current_Pending_Sector,*[!0]*,*)
45 echo "W: disk $disk has $raw pending sector(s)"
46 ;;
47 Offline_Uncorrectable,*[!0]*,*)
48 echo "W: disk $disk has $raw offline-uncorrectable sector(s)"
49 ;;
50 *,*,FAILING_NOW)
51 echo "W: disk $disk attribute $attr failing (value = $raw)"
52 ;;
53 *)
54 if (( $value < $thresh )); then
55 echo "I: disk $disk attribute $attr below thresh (value = $raw)"
56 fi
57 ;;
58 esac
59 done
60 done