check_btrfs: check for errors/corruption on devices

2023-10-02 15:54:37 +02:00 · 2023-10-02 15:54:37 +02:00 · 76668f21b4
commit 76668f21b4
parent 05fcddb210
1 changed files with 131 additions and 0 deletions
--- a/nagios/check_btrfs.sh
+++ b/nagios/check_btrfs.sh
@ -0,0 +1,131 @@
+#!/bin/sh
+
+# Little check for device errors on Btrfs volumes
+#
+# GPL v3+ (copyright chl-dev@bugness.org)
+
+PROGPATH=$( echo $0 | sed -e 's,[\\/][^\\/][^\\/]*$,,' )
+REVISION="0.1"
+
+# Stop at the first non-catched error
+set -e
+
+# Include check_range()
+# Not needed at the moment
+#. $PROGPATH/utils.sh
+STATE_OK=0
+STATE_WARNING=1
+STATE_CRITICAL=2
+STATE_UNKNOWN=3
+STATE_DEPENDENT=4
+
+# Default values
+WARNING_STATS=0
+CRITICAL_STATS=0
+
+# Output
+OUTPUT_EXIT_STATUS=$STATE_OK
+OUTPUT_DETAIL_OK=""
+OUTPUT_DETAIL_WARNING=""
+OUTPUT_DETAIL_CRITICAL=""
+OUTPUT_PERFDATA=""
+NB_DEVICES=0
+
+#
+# Help function
+#
+usage() {
+	cat <<EOF
+Usage :
+  $0 -V volume -V...
+
+Example :
+  ./check_btrfs.sh -V /home
+EOF
+}
+
+check_range_syntax() {
+	return 0
+}
+#	check_range 0 "$1" >/dev/null 2>&1
+#	if [ "$?" -eq "2" ]; then
+#		return 1
+
+# Some early checks
+if ! which btrfs 2>&1 >/dev/null; then
+	echo "UNKNOWN 'btrfs' command not found."
+	exit $STATE_UNKNOWN
+fi
+
+#
+# Loop on parameters + tests
+#
+while getopts hw:c:V: f; do
+	case "$f" in
+		'h')
+			usage
+			exit
+			;;
+
+		'V')
+			VOLUME="$OPTARG"
+
+			# Quickly check if we have enough permission to launch btrfs commands
+			if ! btrfs device stats "$VOLUME" >/dev/null 2>&1 ; then
+				echo "UNKNOWN: unable to launch 'btrfs device stats $VOLUME' command (permissions ?)."
+				exit $STATE_UNKNOWN
+			fi
+
+			# For information, add the number of devices to the perfdata
+			NB_DEVICES="$(( $NB_DEVICES + $( btrfs fi show --raw "$VOLUME" | sed -n 's/^[[:space:]]*Total devices \([0-9]\+\) .*/\1/p' ) ))"
+
+			# Check the device stats
+			# (the kinda weird <<EOF at the end is to avoid entering a sub-shell, so we can access
+			# our precious vars' content outside of the loop, even with prosix-strict dash)
+			while read ITEM VALUE; do
+				# Add the value to the perfdata
+				OUTPUT_PERFDATA="$( printf "%s\n'%s'=%d;%d;%d;0;" "$OUTPUT_PERFDATA" "$ITEM" "$VALUE" "$WARNING_STATS" "$CRITICAL_STATS" )"
+
+				# Check if the value is within ok/warn/critical ranges
+				if [ "$VALUE" -gt "$CRITICAL_STATS" ]; then
+					OUTPUT_EXIT_STATUS=$STATE_CRITICAL
+					OUTPUT_DETAIL_CRITICAL="$OUTPUT_DETAIL_CRITICAL $ITEM:$VALUE"
+				elif [ "$VALUE" -gt "$WARNING_STATS" ] && [ "$OUTPUT_EXIT_STATUS" != "$STATE_CRITICAL" ]; then
+					OUTPUT_EXIT_STATUS=$STATE_WARNING
+					OUTPUT_DETAIL_WARNING="$OUTPUT_DETAIL_WARNING $ITEM:$VALUE"
+				fi
+
+			done <<EOF
+$( btrfs device stats "$VOLUME" )
+EOF
+			;;
+
+		\?)
+			usage
+			exit 1
+			;;
+	esac
+done
+
+# Insert nb devices data into output
+OUTPUT_PERFDATA="$( printf "%s\n%s" "$OUTPUT_PERFDATA" "'nb_devices_total'=$NB_DEVICES;;;1;" )"
+OUTPUT_DETAIL_OK="$OUTPUT_DETAIL_OK - $NB_DEVICES $( [ "$NB_DEVICES" -eq 1 ] && echo "device" || echo "devices" )"
+
+case "$OUTPUT_EXIT_STATUS" in
+	'0')
+		printf "OK%s" "$OUTPUT_DETAIL_OK"
+		;;
+	'1')
+		printf "WARNING%s" "$OUTPUT_DETAIL_WARNING"
+		;;
+	'2')
+		printf "CRITICAL%s" "$OUTPUT_DETAIL_CRITICAL"
+		;;
+	*)
+		printf "UNKNOWN"
+		;;
+esac
+
+# We sort the perfdata because of a weird bug in some icinga/pnp4nagios interactions
+printf "|%s\n" "$( printf "%s" "$OUTPUT_PERFDATA" | sort | grep -v "^$" | tr "\n" " " )"
+exit $OUTPUT_EXIT_STATUS