#!/bin/bash
# Written by Rudy Broersma, Duocast BV <r.broersma@duocast.nl> 
#
# This script needs work:
# We assume sda for now on LSI
# We assume Samsung SSDs or Intel SSDs
# We assume 'Wear_Leveling_Count' or 'Media_Wearout_Indicator' (can be different for different manufacturers)
#
# According to http://www.anandtech.com/show/8239/update-on-samsung-850-pro-endurance-vnand-die-size
# a drive should continue to work fine even when the value reaches 0, because:
# "according to JEDEC spec the P/E cycle rating is with one-year data retention, 
# meaning that there can still be plenty of life left."
#
# I purposely set the values to 90 (warning) and 80 (critical) to get very early warnings
# to see if this check works okay. We can later set it to reasonable values such as 1 
# and 10
#
# TODO: smartctl does not (fully) support NVME. Checking NVME drives is thus currently
#       not supported.
# TODO: We assume the first block device is the correct block device for LSI
# TODO: We only support one 3Ware controller as of now
#
VERSION="1.0"

STORCLI="/usr/bin/storcli"
TWARE="/usr/bin/tw_cli"
BC="/usr/bin/bc"
SMARTCTL="/usr/sbin/smartctl"


MESSAGE=""
EXITCODE=0  
PREFIX="SSD OK"
TESTMODE=0
DEBUG=0
HAS_SSD=0
BRAND="SAMSUNG\|INTEL"

# Parse commandline parameters
for i in "$@"
do
case $i in
    -c=*|--card=*)
    CARD="${i#*=}"
    shift # past argument=value
    ;;
    -d=*|--device=*)
    DEVICE="${i#*=}"
    shift # past argument=value
    ;;
    -b=*|--brand=*)
    BRAND="${i#*=}"
    BRAND="${BRAND^^}"
    shift
    ;;
    -h|--help)
    echo -e "check_ssd - version $VERSION.\n\nThis tool can be used to check the life expectancy of a SSD drive by checking wear leveling indicators using smartmontools."
    echo -e "You can specify the following parameters:\n\n"
    echo -e "-c=,  --card=              Instead of autodetecting using lspci, set the card type. We accept \"lsi\", \"3ware\" and \"auto\" for now. Auto is autodetect"
    echo -e "-d=,  --device=            The blockdevice we should use for calling smartmontools. Can be any blockdevice in /dev, but is driver specific"
    echo -e "-b=,  --brand=             The brand of SSD to search for. We accept \"samsung\" and \"intel\""
    echo -e ""
    echo -e "-d,   --debug              Enable debug output"
    echo -e "-t,   --test               Only test if there are SSDs present in the system. exits with 0 when found, 1 if not"
    echo -e "-h,   --help               Show what you are reading now!"
    echo -e "\n\n"
    exit;
    ;;
    -t|--test)
    TESTMODE=1
    ;;
    -d|--debug)
    DEBUG=1
    ;;
    --default)
    DEFAULT=YES
    shift # past argument with no value
    ;;
    *)
            # unknown option
    ;;
esac
done
# End parsing parameters

if [ -z ${CARD+X} ]; then 
  # Check for local devices
  LOCAL_SSD=`/bin/lsblk -d -o name,rota | /usr/bin/awk '$2 == 0' | grep -v nvme | wc -l`
  if [ $LOCAL_SSD -gt 0 ];
  then
    CONTROLLER="auto"
  else 
    CONTROLLER=`lspci | grep RAID`
  fi
else
  CONTROLLER=${CARD,,} #,, means lowercase
fi


function checkTooling {
  case ${CONTROLLER,,} in
    *lsi*)
      if [ ! -f $STORCLI ]; then echo "UNKNOWN: storcli tool not found"; exit 4; fi
      ;;
    *3ware*)
      if [ ! -f $TWARE ]; then echo "UNKNOWN: tw_cli tool not found"; exit 4; fi
      ;;
  esac

  if [ ! -f $SMARTCTL ]; then echo "UNKNOWN: smartctl not found"; exit 4; fi
  if [ ! -f $BC ]; then echo "UNKNOWN: bc not found"; exit 4; fi
}

function getDIDlist {
  case ${CONTROLLER,,} in
    *lsi*)
      FRSTBLKDEV=`/bin/lsblk -d -o name,rota | /usr/bin/tail -n +2 | awk '{ print "/dev/"$1 } ' | head --lines 1`
      if [ $DEBUG -eq 1 ]; then echo "LSI/AVAGO controller detected"; fi;
      DIDLIST=`$STORCLI /c0 show  | grep SSD | awk '{ print $2 }'`
      DRIVER="megaraid"
      DEVICE=$FRSTBLKDEV
    ;;
    *3ware*)
      if [ $DEBUG -eq 1 ]; then echo "3ware controller detected"; fi;
      TWARE_CID=`$TWARE show | grep ^c | awk '{ print $1 }' | head --lines 1`
      DIDLIST=`$TWARE /$TWARE_CID show | grep "$BRAND" | awk '{ print $1 }' | sed "s/p//g"`
      DRIVER="3ware"
      DEVICE="/dev/twl0" # This seems to be the proper device regardless of which unit (and thus device name) the drive belongs too
    ;;
    *auto*)
      if [ $DEBUG -eq 1 ]; then echo "auto controller detected"; fi;
      DIDLIST=`/bin/lsblk -d -o name,rota | /usr/bin/awk '$2 == 0 { print "/dev/"$1 }' | grep -v nvme`
      DRIVER="auto"
    ;;
    *)
      echo "UNKNOWN: Unknown controller or no controller found"
      exit 4
    ;;
  esac
}


function hasSSDs {
  if [ "$CONTROLLER" = "auto" ]; then
    if [ `echo $DIDLIST | sed '/^\s*$/d' | wc -l` -ne 0 ]; then
      HAS_SSD=1
    fi
  else
    if [ `echo $DIDLIST | sed '/^\s*$/d' | wc -l` -ne 0 ]; then
      HAS_SSD=1
    fi
  fi
}

function checkSSDs {
  for d in `echo $DIDLIST`
  do
    if [ "$DRIVER" = "auto" ]; then
      VALUE=`echo \`$SMARTCTL -A -d auto $d | grep "^177\|^233" | awk '{ print $4 }'\` + 0 | bc`  
    else
      VALUE=`echo \`$SMARTCTL -A -d $DRIVER,$d $DEVICE | grep "^177\|^233" | awk '{ print $4 }'\` + 0 | bc`  
    fi
    case $VALUE in
    100)
      MESSAGE+="Drive $d WLC/MWI $VALUE. "
      ;;
    [4-9][0-9])
      MESSAGE+="Drive $d WLC/MWI $VALUE. "
      ;;
    [2-3][0-9])
      MESSAGE+="Drive $d MEDIUM WLC/MWI ($VALUE). "
      if [ "$EXITCODE" -eq "0" ]; then     
        EXITCODE=1
        PREFIX="SSD WARNING"
      fi
      ;;
    [0-1]*)
      MESSAGE+="Drive $d CRITICAL WLC/MWI ($VALUE). "  
      EXITCODE=2  
      PREFIX="SSD CRITICAL"
      ;;
    *)
      MESSAGE="Unknown error occured."
      EXITCODE=4
      PREFIX="SSD UNKNOWN"
    ;;
    esac
  done
}


checkTooling # must run as first
getDIDlist # Must run second
hasSSDs

if [ $TESTMODE -eq 1 ];
then
  if [ $HAS_SSD -eq 0 ];
  then
    if [ $DEBUG -eq 1 ]; then echo "No SSDs found in the system on controller $CONTROLLER"; fi;
    exit 1
  else
    if [ $DEBUG -eq 1 ]; then echo "SSDs are found in the system on controller $CONTROLLER"; fi;
    exit 0
  fi
else
  if [ $HAS_SSD -ne 0 ];
  then
    checkSSDs
  else
    MESSAGE="No SSDs found in this system on controller $CONTROLLER"
    EXITCODE=4
    PREFIX="UNKNOWN"
  fi

  MESSAGE="$PREFIX: ($DRIVER) $MESSAGE"
  echo -e $MESSAGE
  exit $EXITCODE  

fi