check_md

#!/bin/bash

# Check MD RAIDS using /sys filesystem.
# See https://www.kernel.org/doc/Documentation/md.txt

CAT="/bin/cat"
BASENAME="/usr/bin/basename"
SED="/bin/sed"

STATUS_OK=0
STATUS_WARNING=1
STATUS_CRITICAL=2
STATUS_UNKNOWN=3

STATUS=$STATUS_OK
MSG=""

function add_message() {
  [ -n "$MSG" ] && MSG="$MSG :: "
  MSG="$MSG$1"
}
function raise_status() {
  [ $1 -gt $STATUS ] && STATUS=$1
}

cd /sys/block 2>&1 || exit $STATUS_UNKNOWN
for MD in md*; do
  if [ ! -e "$MD" ]; then
    # md* did not expand, i.e. no arrays defined
	add_message "No MD arrays defined!"
	raise_status $STATUS_UNKNOWN
	break;
  fi
  MD_LEVEL="$($CAT $MD/md/level)"
  if [ -z "$MD_LEVEL" ]; then
    # the device is present in the kernel, but not actuall active
    continue
  fi
  MD_STATUS="$($CAT $MD/md/array_state)"
  MD_SYNC_STATUS="$($CAT $MD/md/sync_action)"
  MD_DEGRADED="$($CAT $MD/md/degraded)"
  MD_SUMMARY=""
  if [ "$MD_DEGRADED" != "0" ]; then
    MD_STATUS="degraded=$MD_DEGRADED"
	raise_status $STATUS_CRITICAL
	MD_SUMMARY="CRITICAL"
  fi
  if [ "$MD_STATUS" != "clean" -a "$MD_STATUS" != active -a "$MD_STATUS" != active-idle -a "$MD_STATUS" != "write-pending" ]; then
    raise_status $STATUS_CRITICAL
	[ -z "MD_SUMMARY" ] && MD_SUMMARY="CRITICAL"
  fi
  if [ "$MD_SYNC_STATUS" != "idle" ]; then
    MD_STATUS="$MD_STATUS,$MD_SYNC_STATUS"
	raise_status $STATUS_WARNING
	[ -z "$MD_SUMMARY" ] && MD_SUMMARY="WARNING"
  fi

  for DEV in $MD/md/dev-*; do
    DEV_NAME="$($BASENAME "$DEV")"
	DEV_STATE="$($CAT $DEV/state)"
	MD_STATUS="$MD_STATUS,$DEV_NAME=$DEV_STATE"
	DEV_STATE="$(echo "$DEV_STATE" | $SED 's/,*write_*mostly//g')"
	if [ "$DEV_STATE" != "in_sync" -a "$DEV_STATE" != "spare" ]; then
	  raise_status $STATUS_CRITICAL
	  MD_SUMMARY="CRITICAL"
	fi
  done


  [ -z "$MD_SUMMARY" ] && MD_SUMMARY="OK"
  add_message "$MD $MD_SUMMARY ($MD_LEVEL,$MD_STATUS)"
done

echo "$MSG" 
exit $STATUS