Update check_nvme.sh

check_nvme.sh

@@ -1,7 +1,15 @@
 #!/bin/bash
 #
 # Simple Nagios check for nvme using nvme-cli
-# Author: Sam McLeod https://smcleod.net
+# Original Author: Sam McLeod https://smcleod.net
+# https://github.com/sammcj/nagios/blob/master/check_nvme.sh
+# Maintainer : Pierre D. - https://dutiko.com/
+#
+# v3 : change detection method
+# v2.3 : check if script is runned as root/sudo, exit with unknown error if not
+# v2.2 : add check to detect if nvme disks are detected, exit with unknown error if not
+# v2.1 : add checks to detect if nvme-cli is present, exit with unknown error if not
+# v1 : Original
 #
 # Requirements:
 # nvme-cli - git clone https://github.com/linux-nvme/nvme-cli
@@ -13,26 +21,30 @@ DISKS=$(lsblk -e 11,253 -dn -o NAME | grep nvme)
 CRIT=false
 MESSAGE=""
 
+command -v sudo nvme >/dev/null 2>&1 || { echo >&2 "UNKNOWN: nvme-cli not found ; please install it" ; exit 3; }
+if [ -z "$DISKS" ] ; then echo "UNKNOWN: no nvme disks found"; exit 3; fi
+
 for DISK in $DISKS ; do
   # Check for critical_warning
-  $(sudo nvme smart-log /dev/$DISK | awk 'FNR == 2 && $3 != 0 {exit 1}')
-  if [ $? == 1 ]; then
+  CRITICAL_WARNING=$(sudo nvme smart-log /dev/$DISK | grep "critical_warning" | awk '{print $3}')
+  if [ $CRITICAL_WARNING -ne 0 ]; then
     CRIT=true
-    MESSAGE="$MESSAGE $DISK has critical warning "
+    MESSAGE="$MESSAGE $DISK has $CRITICAL_WARNING critical warning "
   fi
 
   # Check media_errors
-  $(sudo nvme smart-log /dev/$DISK | awk 'FNR == 15 && $3 != 0 {exit 1}')
-  if [ $? == 1 ]; then
+  MEDIA_ERRORS=$(sudo nvme smart-log /dev/$DISK | grep "media_errors" | awk '{print $3}')
+  if [ $MEDIA_ERRORS -ne 0 ]; then
     CRIT=true
-    MESSAGE="$MESSAGE $DISK has media errors "
+    MESSAGE="$MESSAGE $DISK has $MEDIA_ERRORS media errors "
   fi
 
   # Check num_err_log_entries
   $(sudo nvme smart-log /dev/$DISK | awk 'FNR == 16 && $3 != 0 {exit 1}')
-  if [ $? == 1 ]; then
+  NUM_ERR_LOG_ENTRIES=$(sudo nvme smart-log /dev/$DISK | grep ""num_err_log_entries | awk '{print $3}')
+  if [ $NUM_ERR_LOG_ENTRIES -ne 0 ]; then
     CRIT=true
-    MESSAGE="$MESSAGE $DISK has errors logged "
+    MESSAGE="$MESSAGE $DISK has $NUM_ERR_LOG_ENTRIES errors logged "
   fi
 done