#!/bin/bash -x

# $Id: backup.sh 7 2006-04-15 23:27:00Z root $
#
# Introduction:
# ----------------------------- --------- ---- - -
# This is a simple back-it-up script by Marcus Fritzsch
# http://fritschy.de
#
# This code is lisenced under the GNU GPL Version 2 or any later
# version at your option.
#
# History:
# ----------------------------- --------- ---- - -
#  * 2006-04-10: first version, author: Marcus Fritzsch
#  * 2006-04-11: more configuration variables and error checks
#  * 2006-04-13: - more lethal errors, and misc cleanups
#                - introduced $BACKUP_BASE as base directory
#                  to not abuse / (still possible though!)
#                - gpg now reads through a pipe from tar
#  * 2006-04-16: - added $CONFIGURED variable
#                - added space checking on target and warn/fail
#
# Prerequisites:
# ----------------------------- --------- ---- - -
#  * the BACKUP_SVC variable vontains user@host info of where to store
#    the backup files. Setup a backup service user on some secure host
#    and authorize the user's key that runs this script on that host.
#    Note: you may set a _complex_ password after this procedure to
#          avoid abuse of this user account
#
#  * The backup will be temporarily be written to /tmp/backup_$DATE
#    feel free to change this according to your needs - but make sure
#    there is enough free space for creating backup gzipped-tarballs
#    and also enough space for the encryption step (twice the tarball
#    plus 20% for an ascii armored encrypted file - i guess)
#
#  * This script may be run by a cron job every day or every two days
#    As you like too.
#    When there are problems with the locking, i.e. there is a lock
#    left in the lock-directory the script will wait up to 10h for
#    the lock to be removed (and/or remove it on it's own!)
#    If this script should run more than once a day you should be aware
#    of this!
#
#    my cron-job:
#    17 3 * * * test -x /root/backup.sh && /root/backup.sh &> /dev/null
#
#  * Additionally to the $BACKUP_WHAT variable a dpkg --get-selections
#    will be backed up
#
# Notes:
# ----------------------------- --------- ---- - -
#  * The check for the storage-host to be available is currently done
#    by doing a test-ssh-connection with nc(1) and filtering out some
#    part of the connect string the ssh server sends - feel free to
#    adopt some more appropriate method.
#
# Notes On Encryption:
# ----------------------------- --------- ---- - -
#  * Currently gnupg and a public key of some user is used to encrypt
#    the backup files. If there is another easy way of asynchronous
#    encryption - this should work as well!
#
#  * Make sure to have the public key of the user installed you want
#    the files to be encrypted for _and_ make also sure you trust this
#    user ultimately. To set the trust status of a user:
#      gpg --edit-key $KEYID
#
# Configuration:
# ----------------------------- --------- ---- - -
ENCRYPTKEY=gpg-keyid-here         # if undefined/empty: no encryption!
                                  # this is a gpg keyid or other _uniq_
                                  # identifier

LOCK=/var/lock/backup.sh          # lock file to use

POSTMASTER=postmaster@localhost   # where shall emails go on error

BACKUP_SVC=user@host              # backup-user@storage-host

BACKUP_NAME=backup.sh             # a date-string will pe appended!
                                  # the backup will be put into a
                                  # directory named that way
BACKUP_WHAT="/etc /root /var/wiki /var/www /var/spool /boot /home \
             /var/lib/filetraq /svn"

                                  # which directories should be saved

LOCK_RELEASE_WAIT=4               # time to wait before releasing a foreign
                                  # lock in hours

COPY_WAIT=16                      # time to wait for storage-host to become
                                  # available and copy files over
                                  # If this time is exceeded, fail() is
                                  # called

TAR_OPTS="cjpslPf"                # options passed to tar on compression

BACKUP_BASE="/var/lib/backuptmp"  # where the backup is temporarily stored
                                  # possibly anothe fs to store all that
                                  # stuff ;)


CONFIGURED=                       # set this to some value to enable the
                                  # script



if [ -z "$CONFIGURED" ]; then
	echo "Please run this script _after_ having configured it" >&2
	exit 1
fi



# this script is tested only on bash 2.05/3.0
if [ -z "$BASH_VERSINFO" ]; then
	echo "This script is only tested on BaSH" >&2
	echo ""
	exit 1
fi

shopt -s expand_aliases

ELOG=/tmp/backup.err
exec 2> $ELOG

# make sure, backup temp storage being available
test -d $BACKUP_BASE || mkdir $BACKUP_BASE
chmod 700 $BACKUP_BASE

alias log="/usr/bin/logger -t backup.sh[$$] --"

START=$(date +%s)
DATE=$(date +%Y%m%d-%H%M)
BACKUP_NAME=${BACKUP_NAME}_$DATE
test "$ENCRYPTKEY" && REMOVE_CMD="rm -f" || REMOVE_CMD="/usr/bin/shred -f -n 7 -u -z"

# error status, ! -z if there were errors
ERRS=

err () {
	# before exiting this will be tested to be ! -z
	ERRS="yes"
}

mail () {
	log "some errors happened - giving up and mailing home"
	/usr/bin/mail -s "[BACKUP] backup.sh $DATE failed" -N root@localhost \
		< $ELOG
}

unlock () {
	rm -f $LOCK || err
}

cleanup () {
	cd $BACKUP_BASE/backup_$$_$DATE && $REMOVE_CMD * || err
	cd $BACKUP_BASE/ || err
	rmdir $BACKUP_BASE/backup_$$_$DATE || err
	unlock
	trap - SIGINT SIGHUP SIGTERM
}

fail () {
	cleanup
	mail
	exit 1
}

signal () {
	echo "Got a signal - exiting" >&2
	fail
}



# lock checking is mainly to prevent the HD to become full too fast when
# some backup runs overlap
C=0
while [ $C -lt $((LOCK_RELEASE_WAIT * 10)) ]; do
	C=$((C+1))
	test -f $LOCK || break
	log "backup lock in /var/lock present - waiting"
	sleep 360 # 6min, 10times an hour
done

if [ $C -ge $((LOCK_RELEASE_WAIT * 10)) ]; then
	err
	log "LOCK seems to be from a dead backup instance... removing"
	rm -f $LOCK
fi



# regiter some signal handlers
trap signal SIGINT SIGHUP SIGTERM



# make sure that $COPY_WAIT + $LOCK_RELEASE_WAIT < 24 to not overlap with
# possible later runs of this script
# XXX: when the cron job runs every 2 or 3 days - should i allow a longer
#      wait time?!?
if [ $(($LOCK_RELEASE_WAIT + $COPY_WAIT)) -ge 24 ]; then
	echo "MISCONFIGURATION" >&2
	echo "Script parameters \$LOCK_RELEASE_WAIT and \$COPY_WAIT can make" >&2
	echo "it wait longer than 24h - this might lead to overlapping with" >&2
	echo "other backup runs" >&2
	fail
fi



date +%s > $LOCK

log "starting backup on $DATE"

mkdir $BACKUP_BASE/backup_$$_$DATE || fail
chmod 700 $BACKUP_BASE/backup_$$_$DATE || err
cd $BACKUP_BASE/backup_$$_$DATE || fail

# the data collection starts here
log "taking dpkg selections list"
dpkg --get-selections > dpkg-get-selections.txt

for i in $BACKUP_WHAT; do
	TARB=${i#/}
	TARB=${TARB//\//_}.tar.bz2

	if [ "$ENCRYPTKEY" ]; then
		log "collecting and encrypting data for $i"
		(tar $TAR_OPTS - $i | gpg -r $ENCRYPTKEY -e > $TARB.gpg) || fail
	else
		log "collecting data for $i"
		tar $TAR_OPTS $TARB $i || fail
	fi
done

DATE2=$(date +%Y%m%d-%H%M)

log "finished on $DATE2, runtime: $(expr $(date +%s) - $START) seconds"



C=0
COPIED=
while [ $C -lt $((COPY_WAIT * 10)) ]; do
	C=$((C+1))

	# wait until storage-hosts becomes available
	SSH="$(echo ""|nc -w1 nexus 22|head -n1|cut -d- -f1)"
	if [ "$SSH" = "SSH" ]; then
		# wait some seconds, to not disturb a bootup in progress!
		sleep 60

		cd $BACKUP_BASE/
		RFREE=$(ssh $BACKUP_SVC df --portability -k .|awk 'NR==2{print $4}')
		BSIZE=$(du -ks $BACKUP_BASE/|awk '{print $1}')

		# not enough space... roughly ;)
		if [ $(($RFREE - $BSIZE / 4)) -lt $BSIZE ]; then
			log "copying of backup not possible, not enough space on target"
			fail
		fi

		# just warn but continue
		if [ $(($RFREE - $BSIZE * 3)) -lt $BSIZE ]; then
			log "TARGET SPACE IS GOING LOW"
			err
		fi

		log "start copying backup data to storage host"
		scp -r $BACKUP_BASE/backup_$$_$DATE $BACKUP_SVC:$BACKUP_NAME || fail
		COPIED=yes
		break
	fi
	sleep 360 # 6min, 10times an hour
done

# could not copy - storage host not available
test -z "$COPIED" && fail

log "copying finished, cleaning up"
cleanup
log "everything done - exiting"



test "$ERRS" && mail
rm -f $ELOG



exit 0