X-Git-Url: https://git.sesse.net/?a=blobdiff_plain;f=backup.sh;h=307a14f23aae338f15b76d4c795aa0823ccdd180;hb=040ea36adf3804e541e9973d88c2168da11e8eae;hp=8fb85398db817ef0911bbf83d99c952e3f35b539;hpb=c19b4e77c62749f203010e310e41e942dddbb02f;p=backup.sh

diff --git a/backup.sh b/backup.sh
index 8fb8539..307a14f 100755
--- a/backup.sh
+++ b/backup.sh
@@ -1,185 +1,297 @@
 #!/bin/bash
-
-#dirs
-confdir=/home/backup/conf/           #configuration files
-webdir=/home/backup/public_html      #web-pages
-storagedir=/backup                   #mountpoint of huge disc
-###
-
+# 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:
+#
+# KISS. And by that I mean really simple. Only regular Unix commands, please
+# (but GNU extensions are used and moreutils has to be installed).
+#
+# 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
+maksantallfulle=3                       # Number of full backups
+dagermellomfulle=30                     # Days between full backups
+dagermellomfulleforfast=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
 
-PATH=/local/bin:$PATH:/store/bin
-export PATH
+if [ -z "$computer" ]; then
+	echo "Usage: $0 COMPUTER"
+	exit 1
+fi
+
+# Die more or less gracefully (at least notify user)
+die() {
+	echo `date`": Something nasty happened. Ending, and not deleting lockfile $LOCKFILE."
+	exit 1
+}
+
+# Trap C-c and kill
+trap die SIGINT SIGTERM
 
-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 "Exclude pattern:"
-cat $exclude
-echo
-echo
+echo `date`": Backup run starting" >&2
 
 umask 027
 
-# The computers we want to back up
-unixcomputers=`cat $confdir/computers.unix`
-ntcomputers=`cat $confdir/computers.nt`
-###
-
-#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
-dagermellomfulle=30        #antall dager før det er på tide med ny full
-logw=40
-###
-
-#lager datovariabeler
-DATE=`date "+%Y%m%d%H%M"`            #format: touch
-DATEs=`date "+%Y-%m-%d %H:%M"`       #format: tar
-###
-
-#selve backupen
-# krever at noen variabler er satt
-# krever at vi er i rett katalog
+
 backup()
 {
 
- printf "%-${logw}s %s\n" "$computer:$filesystem" "$backuplevel backup"
+ echo -n `date` >&2
+ printf " $computer: $computer:$filesystem $backuplevel backup\n" >&2
 
- if [ "$backuplevel" = "daglig" ] || [ "$backuplevel" = "incremental" ]
+ SNARFILE="${storagedir}/${computer}/${sfilesystem}/.incremental.snar"
+ incrementalsnar="/root/.backup/${sfilesystem}.snar"
+ if [ "$backuplevel" = "daglig" ]
  then
-  forrige=`cat ../.date`
-  forrigecmd=--newer=\"$forrige\"
+   # 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
+       echo `date`" $computer: Could not copy .incremental.snar to $computer"
+       die
+     fi
+     lastcmd="--listed-incremental=$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
-  forrigecmd=""
+    lastcmd="--listed-incremental=$incrementalsnar"
  fi
 
- #Prøver å ta backup med tar
- if `ssh $computer "cd $filesystem ;\
-     tar cfl - . $forrigecmd \
-      --exclude-from=/tmp/.backup/exclude \
-        | gzip" \
-   |  dd of=$DATE.tmp 2>/dev/null` 
- then
-  #perhaps it did work
-  mv $DATE.tmp $DATE.tgz
-  #make a filelist and update the datefile
-  tar tvfz $DATE.tgz > $DATE.idx 2>&1 &&
-    echo $DATEs > ../.date            &&
-    touch -t $DATE ../.date
+ # expand the exclude-path for use with tar
+ exf=`ssh root@$computer "ls ~/.backup/exclude"`
+
+ # 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 . $lastcmd \
+	--exclude-from=$exf\" | pee \"cat > $TARFILE\" \"tar tzvf -\""
+ echo `date`" $computer: Running $TARCMD" >&2
+ eval $TARCMD > $DATE.idx
+
+ if [ $? -eq 0 ] && [ -s $TARFILE ]; then
+  # File is >0 in size and neither cat or tar tzvf failed; we assume it worked.
+
+  if ! scp root@$computer:$incrementalsnar $SNARFILE.tmp; then
+    echo `date`" $computer: Could not copy .incremental.snar from $computer"
+    die
+  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 sortet filelist
-  grep -v ^d $DATE.idx | sort -n -r +2 > $DATE.sdx
+  # Make a sorted file list as well
+  sort -k6 < $DATE.idx > $DATE.sdx
 
-  #fix perm
-  chmod 600 *tgz
-  chmod 644 *sdx *idx
+  # 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
-  #it did not work
-  rm $DATE.tmp
-  echo "$backuplevel backup of $computer:$filesystem failed"
+  # Something wrong happened.
+  rm $TARFILE
+  echo `date`" $computer: $TARFILE empty. $backuplevel backup of $computer:$filesystem failed and deleted"
+  die
  fi
 
 }
 
-#Løper gjennom listen av unixmaskiner som vi skal ta backup av
-for computer in $unixcomputers
-do
- 
- #tømmer variabelen for sikkerhets skyld
- filesystems=""
+ # Check that the target filesystem is mounted (actually check that it's not
+ # the root filesystem)
+ if [ ! -d "$storagedir/$computer" ]; then
+ 	echo `date`" $computer: Target filesystem ($storagedir/$computer) does not exist. Aborting"
+	die
+ fi
+ rootfilesystem=`df -P /`
+ targetfilesystem=`df -P "$storagedir/$computer"`
+ if [ "$rootfilesystem" == "$targetfilesystem" ]; then
+ 	echo `date`" $computer: Target filesystem ($storagedir/$computer) was mounted on /. Aborting"
+	die
+ fi
+
+ echo `date`" $computer: Backing up $computer" >&2
 
- #Sjekker nest siste felt i fstab. Om det er 0 tar vi ikke backup
- filesystems=`ssh -n $computer "cat /etc/fstab" \
+ # Try to SSH to the computer without entering a password.
+ if ! `ssh -n -o NumberOfPasswordPrompts=0 root@$computer /bin/true`; then
+  echo `date`" $computer: Could not use passwordless SSH."
+  die
+ fi
+
+ # Check if pigz is available
+ if ssh -n root@$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 "^#" \
+  | grep -v "^$" \
   | awk '{ if ( $(NF-1) != "0" ) print $2}' `
 
- #clean up our dir at this client
- ssh $computer "rm -rf /tmp/.backup ; mkdir -m 700 /tmp/.backup"
+ # Clean up our dir at this client
+ if ! ssh root@$computer "rm -r ~/.backup ; mkdir -m 700 ~/.backup"; then
+  echo `date`" $computer: Could not create backup staging area at $computer:~/.backup"
+  die
+ fi
+
+ # Try to copy $exclude to $computer
+ if ! scp $exclude root@$computer:~/.backup/exclude > /dev/null; then
+  echo `date`" $computer: Could not copy exclude.txt to $computer"
+  die
+ fi
 
- #try to copy $exclude to $computer
- scp $exclude $computer:/tmp/.backup/exclude ||
-    echo "Vi klarte ikke kopiere over exclude.txt til $computer :("
+ # Try to copy preeexec and postexec if they exist
+ if [ -f $confdir/preexec.$computer ]; then
+   if ! scp $confdir/preexec.$computer  root@$computer:~/.backup/preexec >&2; then
+     echo `date`" $computer: Could not copy preexec.$computer to $computer:~/.backup/preexec"
+     die
+   fi
+ fi
+ if [ -f $confdir/postexec.$computer ]; then
+   if ! scp $confdir/postexec.$computer root@$computer:~/.backup/postexec >&2; then
+     echo `date`" $computer: Could not copy postexec.$computer to $computer:~/.backup/postexec"
+     die
+   fi
+ fi
 
- #try to copy preeexec and postexec if they exist
- [ -f $confdir/preexec.$computer ] &&
-    scp $confdir/preexec.$computer  $computer:/tmp/.backup/preexec
- [ -f $confdir/postexec.$computer ] &&
-    scp $confdir/postexec.$computer $computer:/tmp/.backup/postexec
+ # Try to run preexec if it exists
+ if ! ssh root@$computer "[ ! -f ~/.backup/preexec ] || /bin/bash -x ~/.backup/preexec" >&2; then
+ 	echo `date`" $computer: Could not run $computer:~/.backup/preexec"
+	die
+ fi
 
- #try to run preexec if it exist
- ssh $computer "[ -f /tmp/.backup/preexec ] && 
-                   /bin/bash -x /tmp/.backup/preexec"
 
  for filesystem in $filesystems
  do
-  #lager en variant uten tegnet "/" eller $ (gjelder NT)
+  # Remove / and $ (for Windows) for path names
   sfilesystem=`echo $filesystem | tr '\/\$' '__'`
 
-  #lager det som trengs av kataloger
-  mkdir -m 755 -p $storagedir/$computer/$sfilesystem/{full,daglig} 2>/dev/null
-
-  #
-  echo $filesystem > $storagedir/$computer/$sfilesystem.name
-  chmod 644 $storagedir/$computer/$sfilesystem.name
+  # 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
 
-  #set default backuplevel
+  # Default backuplevel
   backuplevel=daglig
 
   if [ ! -f $storagedir/$computer/$sfilesystem/.date ]
   then
-   #take the first full backup of this filesystem on this computer
+   # Take the first full backup of this filesystem on this computer
    backuplevel=full
    echo $DATEs > $storagedir/$computer/$sfilesystem/.date
   fi
 
-  #sjekker om det er på tide med en full
-  if 
- [ -z `find $storagedir/$computer/$sfilesystem/full/ -name \*tgz -mtime -$dagermellomfulle` ]
-  then
+  # Check if we want a full backup
+  if [ -f $confdir/fastfullbackupdag.$computer.$sfilesystem ]; then
+    fullbackup_min_for_this_machine=$dagermellomfulleforfast
+    if [ "$DAY_OF_MONTH" = "`cat $confdir/fastfullbackupdag.$computer.$sfilesystem`" ]; then
+      backuplevel=full
+    fi
+  elif [ -f $confdir/fastfullbackupdag.$computer ]; then
+    fullbackup_min_for_this_machine=$dagermellomfulleforfast
+    if [ "$DAY_OF_MONTH" = "`cat $confdir/fastfullbackupdag.$computer`" ]; then
+      backuplevel=full
+    fi
+  else
+    fullbackup_min_for_this_machine=$dagermellomfulle 
+  fi
+
+  if [ -z "`find $storagedir/$computer/$sfilesystem/full/ -name \*tgz -mtime -$fullbackup_min_for_this_machine`" ]; then
    backuplevel=full
   fi
- 
-  #gå ned i rett katalog
-  cd $storagedir/$computer/$sfilesystem/$backuplevel
 
-  #perform the actual backup
+  # We want to be here
+  cd $storagedir/$computer/$sfilesystem/$backuplevel || die
+
+  # Perform the actual backup
   backup
 
-  #delete complete backups
-  mf=$(($maksantallfulle+1))
-  for full in `ls -1t $storagedir/$computer/$sfilesystem/full/*tgz | tail +$mf`
+  # Check if this box has a custom number of full backups
+  if [ -f $confdir/maksfulle.$computer ] ; then
+    mf=$((`cat $confdir/maksfulle.$computer`+1))
+  else
+    mf=$(($maksantallfulle+1))
+  fi
+
+  # Delete old full backups
+  for full in `ls -1t $storagedir/$computer/$sfilesystem/full/*tgz | tail -n +$mf`
   do
-   prefix=`echo $full|cut -d . -f 1`
-   echo "$computer:$filesystem sletter full $prefix (for mange)"
+   prefix=`echo $full | sed "s/\.[^.]*$//"`
+   echo `date`": $computer:$filesystem Deleting full backup $prefix" >&2
    rm $prefix*
   done
 
-  #delete incremental backups older than the oldest complete backup
+  # 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 $oldf \
-     -printf "$computer:$filesystem sletter daglig %f (for gammel)\n" \
-     -exec rm {} \;
-
+     -printf "`date`: $computer: Deleting old incremental backup: %p\n" \
+     -exec rm {} \; >&2
  done
 
- #try to run postexec if it exist
- ssh $computer "[ -f /tmp/.backup/postexec ] &&
-                   /bin/bash -x /tmp/.backup/postexec"
-
-done
-
-#Finne de backupene som har tryna og 
-#lage en webside med de
-/home/backup/finn_ufullstendige.sh $storagedir $webdir/ufullstendige.html
-
-#Lage søkbar indeks over listefilene
-/home/backup/indekser.sh $storagedir
+ # Try to run postexec if it exist
+ if ! ssh root@$computer "[ ! -f ~/.backup/postexec ] || /bin/bash -x ~/.backup/postexec" >&2; then
+ 	echo `date`" $computer: Could not run $computer:~/.backup/postexec"
+	die
+ fi
 
-date
+# Remove lockfile
+rm $LOCKFILE