X-Git-Url: https://git.sesse.net/?a=blobdiff_plain;f=backup.sh;h=f054e1aae68fa9d93e875e3db4cb04a9077a9914;hb=966d195df94a5d99952b18e32eb4fc23f52b3a70;hp=c3444fb1f18dea5a36fa1a9c560697bdfa5c90a6;hpb=eaf1676e55df105381c52cad4acfc9f3f5777b3a;p=backup.sh

diff --git a/backup.sh b/backup.sh
index c3444fb..f054e1a 100755
--- a/backup.sh
+++ b/backup.sh
@@ -1,168 +1,282 @@
 #!/bin/bash
+# backup.sh
+# 
+# Backup up Unix-like computers
+# 
+# itk@samfundet.no
+#
+# This script was first checked into RCS in 2000, and has since backed up a few
+# servers almost every day, whenever not broken.
+# 
+# It's a bit peculiar, but quite effective. The following is worth keeping in
+# mind when hacking:
+#
+# Because of the way we run this script from cron, informational logging goes
+# to stderr and warnings and errors to stdout. This way, we get mail whenever
+# stuff fails, and other output is redirected to the log file.
+# 
+# The script is halfway Norwegian and halfway English. Newer modifications are
+# in English, we should probably stick to that.
+
+# The computer to backup.
+computer=$1
+
+# Configuration
+LOCKFILE=/home/backup/backuprun.lockfile.$computer
+confdir=/home/backup/conf/              # Configuration files
+storagedir=/backup                      # Where we keep backups
+maxnumfull=3                            # Number of full backups
+daysbetweenfull=30                      # Days between full backups
+daysbetweenfullforfixed=32              # Days between full backups for machines with fixed
+                                        # full-backup date (in case the fixed day gets missed)
+DAY_OF_MONTH=`date "+%d" | sed s/^0//`  # from 1 to 31
+DATE=`date "+%Y%m%d%H%M"`               #format: touch
+DATEs=`date "+%Y-%m-%d %H:%M"`          #format: tar
+
+# Exclude-pattern
+exclude=$confdir/exclude
+[ ! -f $exclude ] && printf "tmp\ncore\n" > $exclude
+
+if [ -z "$computer" ]; then
+	echo "Usage: $0 COMPUTER"
+	exit 1
+fi
+
+# Die more or less gracefully (at least notify user)
+die() {
+	echo `date`": Something nasty happened."
+	rm $LOCKFILE
+	exit 1
+}
+diemsg() {
+	echo `date`": $computer: $@"
+	die
+}
 
-PATH=/local/bin:$PATH
-export PATH
-exclude=/home/backup/exclude.txt
+# Trap C-c and kill
+trap die SIGINT SIGTERM
 
-echo -n "Starter... "
-date
+# Don't start if we're already running
+if [ -e $LOCKFILE ]; then
+	echo `date`": $LOCKFILE exists for pid `cat $LOCKFILE`, exiting."
+	echo `date`": $LOCKFILE exists for pid `cat $LOCKFILE`, exiting." >&2
+	exit 1
+fi;
+touch $LOCKFILE
+echo $$ > $LOCKFILE
 
-#Skriver litt til loggefilen
-echo
-echo "Slik ser regexpen som forteller hva vi ikke skal ta backup av:"
-cat $exclude
-echo
-echo
+echo `date`": Backup run starting" >&2
 
-#newgrp itk
 umask 027
 
-#liste over maskinene vi tar backup av. (space mellom maskinnavn)
-computers="cirkus backup josef"
-#for debug
-#computers="josef"
-###
-
-#disse bør kunne varieres fra fs til fs?
-#disse bør flyttes ut i egen fil som kan editeres fra web kanskje?
-maksantallfulle=2          #hvor mange fulle vi tar vare på
-#maksantalldaglige=30       #hvor mange daglige vi tar vare på
-dagermellomfulle=30        #antall dager før det er på tide med ny full
-###
-
-#lager datovariabeler
-DATE=`date "+%Y%m%d%H%M"`            #passer for bruk med touch
-DATEs=`date "+%Y-%m-%d %H:%M"`       #passer for bruk med tar
-###
-
-#selve backupen
-# krever at noen variabler er satt
-# krever at vi er i rett katalog
+
 backup()
 {
-
- echo "$computer:$filesystem tar $backuplevel backup"
-
- if [ "$backuplevel" = "daglig" ]
- then
-  forrige=`cat ../.dato`
-  forrigecmd=--newer=\"$forrige\"
- else
-  forrigecmd=""
- fi
-
- #Prøver å kopiere over $exclude til $computer
- scp $exclude $computer:/tmp/exclude.txt ||
-    echo "Vi klarte ikke kopiere over exclude.txt til $computer :("
-
- if `ssh $computer "cd $filesystem ;\
-     tar cfl - . $forrigecmd \
-      --exclude-from=/tmp/exclude.txt \
-        | gzip" \
-   |  dd of=$DATE.tmp 2>/dev/null` 
- then
-  #Det gikk kanskje bra
-  mv $DATE.tmp $DATE.tgz
-  #lage listefil
-  tar tvfz $DATE.tgz > $DATE.idx 2>&1
-  grep -v ^d $DATE.idx | sort -n -r +2 > $DATE.sdx
-
-  #oppdatere datofil
-  echo $DATEs > ../.dato
-  touch -t $DATE ../.dato
-
-  #fikse rettigheter
-  chmod 600 *tgz
-  chmod 644 *sdx *idx
-
- else
-  #Det gikk ikke bra
-  rm $DATE.tmp
-  echo "$backuplevel backup av $computer:$filesystem gikk ikke bra"
- fi
-}
-
-for computer in $computers
-do
- 
- #tømmer variabelen for sikkerhets skyld
- filesystems=""
-
- #Sjekker nest siste felt i fstab. Om det er 0 tar vi ikke backup
- filesystems=`ssh -n $computer "cat /etc/fstab" \
-  | grep -v nfs \
-  | grep -v ^# \
-  | awk '{ if ( $(NF-1) != "0" ) print $2}' `
-
- #for debugging
- #filesystems="/"
-
- for filesystem in $filesystems
- do
-  #lager en variant uten tegnet "/"
-  sfilesystem=`echo $filesystem | tr '/' '_'`
-
-  #
-  echo $filesystem > /backup/$computer/$sfilesystem.navn
-  chmod 644 /backup/$computer/$sfilesystem.navn
-
-  #standard backuplevel. full og daglig er eneste muligheter
-  backuplevel=daglig
-  #backuplevel=full
-
-  #lager det som trengs av kataloger
-  mkdir -m 755 -p /backup/$computer/$sfilesystem/{full,daglig} 2>/dev/null
-
-  #Bør bare skje ved første gangs kjøring for dette fs
-  if [ ! -f /backup/$computer/$sfilesystem/.dato ]
-  then
-   backuplevel=full
-   echo $DATEs > /backup/$computer/$sfilesystem/.dato
-   echo "$computer:$filesystem datofil laget"
-   echo "$computer:$filesystem trenger full pga ingen fra foer"
-  fi
-
-  #sjekker om det er på tide med en full
-  if 
- [ -z `find /backup/$computer/$sfilesystem/full/ -name \*tgz -mtime -$dagermellomfulle` ]
-  then
-   echo "$computer:$filesystem trenger full pga lenge siden sist"
-   backuplevel=full
-  fi
+	echo -n `date` >&2
+	printf " $computer: $computer:$filesystem $backuplevel backup\n" >&2
+
+	SNARFILE="${storagedir}/${computer}/${sfilesystem}/.incremental.snar"
+	incrementalsnar="/root/.backup/${sfilesystem}.snar"
+	if [ "$backuplevel" = "daglig" ]
+	then
+		# If incremental backup, we need to copy the incremental status to $computer.
+		# If it does not exist, whine a bit and then run date-based instead.
+		if [ -s "$SNARFILE" ]; then
+			if ! scp $SNARFILE root@$computer:$incrementalsnar; then
+				diemsg "Could not copy .incremental.snar to $computer"
+			fi
+			lastcmd="--listed-incremental=$incrementalsnar --no-check-device"
+		else
+			echo `date`" $computer: Missing incremental.snar for $filesystem, doing date-based backup instead until next full backup" >&2
+			incrementalsnar=""
+			lastd=`cat ../.date`
+			lastcmd="--newer='$lastd'"
+		fi
+	else
+		lastcmd="--listed-incremental=$incrementalsnar"
+	fi
+
+	# We try to run tar on the remote computer
+	#    c create archive
+	#    C change to directory first
+	#    . where to start taring (see C)
+	#    $lastcmd only files newer than this
+	#    --one-file-system don't traverse file systems
+	#    --use-compress-program compress it using gzip or pigz
+	#    --exclude-from file to get exclusion pattern from
+	#    Pipe the stuff over ssh to ourselves, run pee to cat the contents to a
+	#    file and build a file list, respectivly.  
+	TARFILE=$DATE.tmp
+	TARCMD="ssh root@$computer \"nice -n 19 ionice -c3 tar --one-file-system --use-compress-program $compressor -cf - -C $filesystem . $lastcmd \
+		--exclude-from=/root/.backup/exclude\" | pee \"cat > $TARFILE\" \"tar tzvf -\""
+	echo `date`" $computer: Running $TARCMD" >&2
+	eval $TARCMD > $DATE.idx
+
+	if [ $? -eq 0 ] && [ -s $TARFILE ]; then
+		# File is >0 in size and neither cat or tar tzvf failed; we assume it worked.
  
-  #gå ned i rett katalog
-  cd /backup/$computer/$sfilesystem/$backuplevel
-
-  #selve backupen
-  backup
-
-  #Sjekker om det er på tide å slette noen fulle backuper
-  mf=$(($maksantallfulle+1))
-  for full in `ls -1t /backup/$computer/$sfilesystem/full/*tgz | tail +$mf`
-  do
-   prefix=`echo $full|cut -d . -f 1`
-   echo "$computer:$filesystem sletter full $prefix (for mange)"
-   rm $prefix*
-  done
-
-  #Sjekker om det er på tide å slette noen daglige backuper
-  eldste=`ls -t1 /backup/$computer/$sfilesystem/full/*tgz | tail -1`
-  find \
-     /backup/$computer/$sfilesystem/daglig \
-     \! -newer $eldste \
-     -printf "$computer:$filesystem sletter daglig %f (for gammel)\n" \
-     -exec rm {} \;
-
- done
+		if [ "$incrementalsnar" ]; then
+			if ! scp root@$computer:$incrementalsnar $SNARFILE.tmp; then
+				diemsg "Could not copy .incremental.snar from $computer"
+			fi
+		fi
+
+		# Move tar file in place
+		mv $TARFILE $DATE.tgz
+		mv $SNARFILE.tmp $SNARFILE
+
+		# Update timestamp
+		echo $DATEs > ../.date && touch -t $DATE ../.date
+
+		# Make a sorted file list as well
+		sort -k6 < $DATE.idx > $DATE.sdx
+
+		# Fix permissions
+		chmod 600 *tgz                      #only for us
+		chmod 644 *sdx *idx 2>/dev/null     #everyone can read
+
+		# Let the remote computer know that we ran a successful backup (for nagios)
+		ssh root@$computer touch $filesystem/.lastbackup
+	else
+		# Something wrong happened.
+		rm $TARFILE
+		diemsg "tar failed or $TARFILE empty. $backuplevel backup of $computer:$filesystem failed and deleted"
+	fi
+}
 
+# Check that the target filesystem is mounted (actually check that it's not
+# the root filesystem)
+if [ ! -d "$storagedir/$computer" ]; then
+	diemsg "Target filesystem ($storagedir/$computer) does not exist. Aborting"
+fi
+rootfilesystem=`df -P /`
+targetfilesystem=`df -P "$storagedir/$computer"`
+if [ "$rootfilesystem" == "$targetfilesystem" ]; then
+	diemsg "Target filesystem ($storagedir/$computer) was mounted on /. Aborting"
+fi
+
+echo `date`" $computer: Backing up $computer" >&2
+
+# Try to SSH to the computer without entering a password.
+if ! `ssh -n -o NumberOfPasswordPrompts=0 root@$computer /bin/true`; then
+	diemsg "Could not use passwordless SSH."
+fi
+
+# Check if pigz is available
+if ssh -n root@$computer "pigz -V 2>/dev/null"; then
+	compressor=pigz
+else
+	echo `date`" $computer: pigz missing; falling back to gzip."
+	compressor=gzip
+fi
+
+# Check dump bit in fstab
+filesystems=`ssh -n root@$computer "cat /etc/fstab" \
+	| grep -v nfs \
+	| grep -v "^#" \
+	| grep -v "^$" \
+	| awk '{ if ( $(NF-1) != "0" ) print $2}' `
+
+# Clean up our dir at this client
+if ! ssh root@$computer "rm -r /root/.backup ; mkdir -m 700 /root/.backup"; then
+	diemsg "Could not create backup staging area at $computer:/root/.backup"
+fi
+
+# Try to copy $exclude to $computer
+if ! scp $exclude root@$computer:/root/.backup/exclude > /dev/null; then
+	diemsg "Could not copy exclude.txt to $computer"
+fi
+
+# Try to copy preeexec and postexec if they exist
+if [ -f $confdir/preexec.$computer ]; then
+	if ! scp $confdir/preexec.$computer  root@$computer:/root/.backup/preexec >&2; then
+		diemsg "Could not copy preexec.$computer to $computer:/root/.backup/preexec"
+	fi
+fi
+if [ -f $confdir/postexec.$computer ]; then
+	if ! scp $confdir/postexec.$computer root@$computer:/root/.backup/postexec >&2; then
+		diemsg "Could not copy postexec.$computer to $computer:/root/.backup/postexec"
+	fi
+fi
+
+# Try to run preexec if it exists
+if ! ssh root@$computer "[ ! -f /root/.backup/preexec ] || /bin/bash -x /root/.backup/preexec" >&2; then
+	diemsg "Could not run $computer:/root/.backup/preexec"
+fi
+
+for filesystem in $filesystems; do
+	# Remove / and $ (for Windows) for path names
+	sfilesystem=`echo $filesystem | tr '\/\$' '__'`
+
+	# Prepare storage area
+	mkdir -m 755 -p $storagedir/$computer/$sfilesystem/full 2>/dev/null
+	mkdir -m 755 -p $storagedir/$computer/$sfilesystem/daglig 2>/dev/null
+
+	# Default backuplevel
+	backuplevel=daglig
+
+	if [ ! -s $storagedir/$computer/$sfilesystem/.date ]
+	then
+		# Take the first full backup of this filesystem on this computer
+		backuplevel=full
+		echo $DATEs > $storagedir/$computer/$sfilesystem/.date
+	fi
+
+	# Check if we want a full backup
+	if [ -f $confdir/fastfullbackupdag.$computer.$sfilesystem ]; then
+		fullbackup_min_for_this_machine=$daysbetweenfullforfixed
+		if [ "$DAY_OF_MONTH" = "`cat $confdir/fastfullbackupdag.$computer.$sfilesystem`" ]; then
+			backuplevel=full
+		fi
+	elif [ -f $confdir/fastfullbackupdag.$computer ]; then
+		fullbackup_min_for_this_machine=$daysbetweenfullforfixed
+		if [ "$DAY_OF_MONTH" = "`cat $confdir/fastfullbackupdag.$computer`" ]; then
+			backuplevel=full
+		fi
+	else
+		fullbackup_min_for_this_machine=$daysbetweenfull 
+	fi
+
+	if [ -z "`find $storagedir/$computer/$sfilesystem/full/ -name \*tgz -mtime -$fullbackup_min_for_this_machine`" ]; then
+		backuplevel=full
+	fi
+
+	# We want to be here
+	cd $storagedir/$computer/$sfilesystem/$backuplevelmsg || die "$storagedir/$computer/$sfilesystem/$backuplevel does not exist"
+
+	# Perform the actual backup
+	backup
+
+	# Check if this box has a custom number of full backups
+	if [ -f $confdir/maksfulle.$computer.$sfilesystem ]; then
+		mf=$((`cat $confdir/maksfulle.$computer`+1))
+	elif [ -f $confdir/maksfulle.$computer ] ; then
+		mf=$((`cat $confdir/maksfulle.$computer`+1))
+	else
+		mf=$(($maxnumfull+1))
+	fi
+
+	# Delete old full backups
+	for full in `ls -1t $storagedir/$computer/$sfilesystem/full/*tgz | tail -n +$mf`; do
+		prefix=`echo $full | sed "s/\.[^.]*$//"`
+		echo `date`": $computer:$filesystem Deleting full backup $prefix" >&2
+		rm $prefix*
+	done
+
+	# Delete incremental backups older than the oldest full backup
+	oldf=`ls -t1 $storagedir/$computer/$sfilesystem/full/*tgz | tail -1`
+	find \
+		$storagedir/$computer/$sfilesystem/daglig \
+		-type f \
+		\! -newer $oldf \
+		-printf "`date`: $computer: Deleting old incremental backup: %p\n" \
+		-exec rm {} \; >&2
 done
 
-#Finne de backupene som har tryna og 
-#lage en webside med de
-/home/backup/finn_ufullstendige.sh
-
-#Lage søkbar indeks over listefilene
-/home/backup/indekser.sh
+# Try to run postexec if it exist
+if ! ssh root@$computer "[ ! -f /root/.backup/postexec ] || /bin/bash -x /root/.backup/postexec" >&2; then
+	diemsg "Could not run $computer:/root/.backup/postexec"
+fi
 
-echo -n "Ferdig... "
-date
+# Remove lockfile
+rm $LOCKFILE