#!/bin/bash

########################################################################
# check_backup v1.0
#
# Nagios plugin by John E.P. Hynes, HyTronix [03/22/2015] john@hytronix.com
#
# Arguments can be in any order: d<days> h<hours> m<minutes> s<seconds>
# x<multiplier> f<filename> u<ssh user/host> and i<path to ssh key>.
#
# All except filename are optional.  (Well, except that time defaulting
# to "0" will always fail...) and if you specify u<ssh user/host>, you
# must also specify i<path to ssh key>.
#
# Example: ./check_backup h3 m45 x2 f/tmp/timefile
#
# u<ssh user/host>, if provided, means to retrieve the contents of 
# <filename> from a remote host via ssh.  It must be in the search path.
#
# This string should contain user@host, so the complete declaration would
# be:
#
# ujohn@mybox (for user "john" at host "mybox")
#
# i<path to ssh key> tells ssh to log in using the ssh key provided.
# It would also allow you to just use a local key
# for which there was no local user.  Make the remote user privilege-less
# for security.  (So long as they can read <filename>...)
#
# I suppose ssh agent setup would work too; if you know about this you
# don't need me to tell you how to do it.
#
# ssh example:
#
# ./check_backup h3 m45 x2 f/tmp/timefile ujohn@mybox i~/.ssh/id_rsa
#
# Returns (from above example): 
# OK if timestamp in /tmp/timefile is no older
# than 3 hours and 45 minutes ago, and the exit code in 
# /tmp/timefile is 0.  WARNING if time exceeded, FAIL if exceeded by
# multiplier - in this case, FAIL at 7 hours 30 minutes overdue.
#
# Always FAIL if exit code is not 0.
#
# If x<multiplier> is omitted, goes right to FAIL on time exceeded.
#
# Your backup scripts must produce this file.  As implied, it's format
# is:
#
# ------8<------ CUT
# <timestamp> <exit code>
# ------8<------ CUT
#
# <timestamp> is seconds since the epoch, 
# <exit code> is the backup process exit code (make this 0 if successful).
#
# For example, your script could, when complete, do a:
#
# echo -n `date +%s | tr -d '\n'> /tmp/test`; \
# echo -n " " >> /tmp/test; echo "0" >> /tmp/test
#
# ...or somesuch.  I'm sure you get the idea.
#
# Other than supporting getting timefile data via ssh, this script has
# no dependencies other than bash itself.
#
# Questions, comments, and suggestions appreciated.
#
########################################################################
# Nagios exit codes

OK="0"
WARNING="1"
CRITICAL="2"
UNKNOWN="3"
########################################################################
# Pre-set time variables to zero (except multiplier)

DAYS="0"
HOURS="0"
MINUTES="0"
SECONDS="0"
MULTIPLIER="1"
SSH_USER=""
SSH_KEY=""
########################################################################
# Pre-set time and flie variables to false/unset

DAYS_SET=false
HOURS_SET=false
MINUTES_SET=false
SECONDS_SET=false
FILENAME_SET=false
MULTIPLIER_SET=false
SSH_USER_SET=false
SSH_KEY_SET=false
########################################################################
# File descriptor for timefile

FD="3"
########################################################################
# Function arg_set - tests if a variable is set already, sets it if not.

arg_set() {
	if ! eval \$$1_SET; then
		eval $1="${argument:1}"
		eval $1_SET=true
	else
		echo "$1 already set."
		exit $UNKNOWN 
	fi
}
########################################################################
# Function check_args - checks for time and file statements from caller

check_args() {
	case $1 in
		"d")
			arg_set DAYS
			;;
		"h")
			arg_set HOURS
			;;
		"m")
			arg_set MINUTES
			;;
		"s")
			arg_set SECONDS
			;;
		"f")
			arg_set FILENAME
			;;
		"x")
			arg_set MULTIPLIER
			;;
		"u")
			arg_set SSH_USER
			;;
		"i")
			arg_set SSH_KEY
			;;
		*)
			echo "Invalid argument: $argument"
			exit $UNKNOWN 
			;;
	esac
}
########################################################################
# Function validate_input - check that time declarations are numeric

validate_input() {
	NUM_REGEX='^[0-9]+$'

	if ! [[ $1 =~ $NUM_REGEX ]]; then
        	echo "Invalid numeric input $1."
                exit $UNKNOWN 
	fi
}
########################################################################
# Function convert_time_to_seconds - so time differential can be calc'ed

convert_time_to_seconds() {
	HOURS=`expr $HOURS + $DAYS \* 24`
	MINUTES=`expr $MINUTES + $HOURS \* 60`
	SECONDS=`expr $SECONDS + $MINUTES \* 60`
}
########################################################################
# Function read_bu_data - local or via ssh

read_bu_data() {
	if [ -z "$SSH_USER" ]; then
		if [ ! -f "$FILENAME" ]; then
        		echo "Bad filename: $FILENAME."
        		exit $UNKNOWN;
		fi

		eval "exec $FD<$FILENAME"
        	eval "read -u $FD STAMPTIME EXITCODE"
        	eval "exec $FD<&-"
	else
		TIME_CODE=$(ssh $SSH_USER -i $SSH_KEY "cat $FILENAME")
		HOLDIFS=$IFS
		IFS=" "
		read -ra INFOLINE <<< "$TIME_CODE"
		STAMPTIME="${INFOLINE[0]}"
		EXITCODE="${INFOLINE[1]}"
		IFS=$HOLDIFS
	fi
}
########################################################################
# Main script

for argument in "$@"
	do
		check_args ${argument:0:1}
	done

for CHECK_VAR_NUM in $DAYS $HOURS $MINUTES $SECONDS $MULTIPLIER
	do
		validate_input $CHECK_VAR_NUM
	done

convert_time_to_seconds

read_bu_data

if [ -z "$STAMPTIME" ] || [ -z "$EXITCODE" ]; then
	echo "<timefile> not in proper format."
	exit $UNKNOWN
else
	validate_input $STAMPTIME
	validate_input $EXITCODE
fi

printf -v TIMENOW '%(%s)T' -1

DIFFERENCE=`expr $TIMENOW - $STAMPTIME`
MARGIN=`expr $SECONDS - $DIFFERENCE`
WARNZONE=`expr $SECONDS \* $MULTIPLIER`

if [ "$DIFFERENCE" -lt "$SECONDS" ] && [ "$EXITCODE" -eq "0" ]; then
	echo "OK: Backup completed within time range. | Time differential (in seconds): $MARGIN" 
	exit $OK
else
	if [ "$DIFFERENCE" -gt "$SECONDS" ] && 
	   [ "$DIFFERENCE" -lt "$WARNZONE" ]; then
		if [ "$EXITCODE" -eq "0" ]; then
			echo "WARNING: Last backup too long ago - Investigate. | Time differential (in seconds): $MARGIN"
			exit $WARNING
		fi
	fi	

	if [ "$DIFFERENCE" -gt "$SECONDS" ]; then
		echo "FAIL: Last backup too long ago! | Time differential (in seconds): $MARGIN"
	fi
	
	if [ "$EXITCODE" -ne "0" ]; then
		echo "FAIL: Backup exited with code $EXITCODE"
	fi
	exit $CRITICAL
fi
########################################################################