git.sesse.net Git - backup.sh/blob - backup.sh

   1 #!/bin/bash
   2 # backup.sh
   3 #
   4 # Backup up Unix-like computers
   5 #
   6 # itk@samfundet.no
   7 #
   8 # This script was first checked into RCS in 2000, and has since backed up a few
   9 # servers almost every day, whenever not broken.
  10 #
  11 # It's a bit peculiar, but quite effective. The following is worth keeping in
  12 # mind when hacking:
  13 #
  14 # KISS. And by that I mean really simple. Only regular Unix commands, please
  15 # (but GNU extensions are used and moreutils has to be installed).
  16 #
  17 # Because of the way we run this script from cron, informational logging goes
  18 # to stderr and warnings and errors to stdout. This way, we get mail whenever
  19 # stuff fails, and other output is redirected to the log file.
  20 #
  21 # The script is halfway Norwegian and halfway English. Newer modifications are
  22 # in English, we should probably stick to that.
  23
  24 # The computer to backup.
  25 computer=$1
  26
  27 # Configuration
  28 LOCKFILE=/home/backup/backuprun.lockfile.$computer
  29 confdir=/home/backup/conf/              # Configuration files
  30 storagedir=/backup                      # Where we keep backups
  31 maxnumfull=3                            # Number of full backups
  32 daysbetweenfull=30                      # Days between full backups
  33 daysbetweenfullforfixed=32              # Days between full backups for machines with fixed
  34                                         # full-backup date (in case the fixed day gets missed)
  35 DAY_OF_MONTH=`date "+%d" | sed s/^0//`  # from 1 to 31
  36 DATE=`date "+%Y%m%d%H%M"`               #format: touch
  37 DATEs=`date "+%Y-%m-%d %H:%M"`          #format: tar
  38
  39 # Exclude-pattern
  40 exclude=$confdir/exclude
  41 [ ! -f $exclude ] && printf "tmp\ncore\n" > $exclude
  42
  43 if [ -z "$computer" ]; then
  44         echo "Usage: $0 COMPUTER"
  45         exit 1
  46 fi
  47
  48 # Die more or less gracefully (at least notify user)
  49 die() {
  50         echo `date`": Something nasty happened."
  51         rm $LOCKFILE
  52         exit 1
  53 }
  54
  55 # Trap C-c and kill
  56 trap die SIGINT SIGTERM
  57
  58 # Don't start if we're already running
  59 if [ -e $LOCKFILE ]; then
  60         echo `date`": $LOCKFILE exists for pid `cat $LOCKFILE`, exiting."
  61         echo `date`": $LOCKFILE exists for pid `cat $LOCKFILE`, exiting." >&2
  62         exit 1
  63 fi;
  64 touch $LOCKFILE
  65 echo $$ > $LOCKFILE
  66
  67 echo `date`": Backup run starting" >&2
  68
  69 umask 027
  70
  71
  72 backup()
  73 {
  74         echo -n `date` >&2
  75         printf " $computer: $computer:$filesystem $backuplevel backup\n" >&2
  76
  77         SNARFILE="${storagedir}/${computer}/${sfilesystem}/.incremental.snar"
  78         incrementalsnar="/root/.backup/${sfilesystem}.snar"
  79         if [ "$backuplevel" = "daglig" ]
  80         then
  81                 # If incremental backup, we need to copy the incremental status to $computer.
  82                 # If it does not exist, whine a bit and then run date-based instead.
  83                 if [ -s "$SNARFILE" ]; then
  84                         if ! scp $SNARFILE root@$computer:$incrementalsnar; then
  85                                 echo `date`" $computer: Could not copy .incremental.snar to $computer"
  86                                 die
  87                         fi
  88                         lastcmd="--listed-incremental=$incrementalsnar --no-check-device"
  89                 else
  90                         echo `date`" $computer: Missing incremental.snar for $filesystem, doing date-based backup instead until next full backup" >&2
  91                         incrementalsnar=""
  92                         lastd=`cat ../.date`
  93                         lastcmd="--newer='$lastd'"
  94                 fi
  95         else
  96                 lastcmd="--listed-incremental=$incrementalsnar"
  97         fi
  98
  99         # expand the exclude-path for use with tar
 100         exf=`ssh root@$computer "ls ~/.backup/exclude"`
 101
 102         # We try to run tar on the remote computer
 103         #    c create archive
 104         #    C change to directory first
 105         #    . where to start taring (see C)
 106         #    $lastcmd only files newer than this
 107         #    --one-file-system don't traverse file systems
 108         #    --use-compress-program compress it using gzip or pigz
 109         #    --exclude-from file to get exclusion pattern from
 110         #    Pipe the stuff over ssh to ourselves, run pee to cat the contents to a
 111         #    file and build a file list, respectivly.
 112         TARFILE=$DATE.tmp
 113         TARCMD="ssh root@$computer \"nice -n 19 ionice -c3 tar --one-file-system --use-compress-program $compressor -cf - -C $filesystem . $lastcmd \
 114                 --exclude-from=$exf\" | pee \"cat > $TARFILE\" \"tar tzvf -\""
 115         echo `date`" $computer: Running $TARCMD" >&2
 116         eval $TARCMD > $DATE.idx
 117
 118         if [ $? -eq 0 ] && [ -s $TARFILE ]; then
 119                 # File is >0 in size and neither cat or tar tzvf failed; we assume it worked.
 120
 121                 if ! scp root@$computer:$incrementalsnar $SNARFILE.tmp; then
 122                         echo `date`" $computer: Could not copy .incremental.snar from $computer"
 123                         die
 124                 fi
 125
 126                 # Move tar file in place
 127                 mv $TARFILE $DATE.tgz
 128                 mv $SNARFILE.tmp $SNARFILE
 129
 130                 # Update timestamp
 131                 echo $DATEs > ../.date && touch -t $DATE ../.date
 132
 133                 # Make a sorted file list as well
 134                 sort -k6 < $DATE.idx > $DATE.sdx
 135
 136                 # Fix permissions
 137                 chmod 600 *tgz                      #only for us
 138                 chmod 644 *sdx *idx 2>/dev/null     #everyone can read
 139
 140                 # Let the remote computer know that we ran a successful backup (for nagios)
 141                 ssh root@$computer touch $filesystem/.lastbackup
 142         else
 143                 # Something wrong happened.
 144                 rm $TARFILE
 145                 echo `date`" $computer: $TARFILE empty. $backuplevel backup of $computer:$filesystem failed and deleted"
 146                 die
 147         fi
 148 }
 149
 150 # Check that the target filesystem is mounted (actually check that it's not
 151 # the root filesystem)
 152 if [ ! -d "$storagedir/$computer" ]; then
 153         echo `date`" $computer: Target filesystem ($storagedir/$computer) does not exist. Aborting"
 154         die
 155 fi
 156 rootfilesystem=`df -P /`
 157 targetfilesystem=`df -P "$storagedir/$computer"`
 158 if [ "$rootfilesystem" == "$targetfilesystem" ]; then
 159         echo `date`" $computer: Target filesystem ($storagedir/$computer) was mounted on /. Aborting"
 160         die
 161 fi
 162
 163 echo `date`" $computer: Backing up $computer" >&2
 164
 165 # Try to SSH to the computer without entering a password.
 166 if ! `ssh -n -o NumberOfPasswordPrompts=0 root@$computer /bin/true`; then
 167         echo `date`" $computer: Could not use passwordless SSH."
 168         die
 169 fi
 170
 171 # Check if pigz is available
 172 if ssh -n root@$computer "pigz -V 2>/dev/null"; then
 173         compressor=pigz
 174 else
 175         echo `date`" $computer: pigz missing; falling back to gzip."
 176         compressor=gzip
 177 fi
 178
 179 # Check dump bit in fstab
 180 filesystems=`ssh -n root@$computer "cat /etc/fstab" \
 181         | grep -v nfs \
 182         | grep -v "^#" \
 183         | grep -v "^$" \
 184         | awk '{ if ( $(NF-1) != "0" ) print $2}' `
 185
 186 # Clean up our dir at this client
 187 if ! ssh root@$computer "rm -r ~/.backup ; mkdir -m 700 ~/.backup"; then
 188         echo `date`" $computer: Could not create backup staging area at $computer:~/.backup"
 189         die
 190 fi
 191
 192 # Try to copy $exclude to $computer
 193 if ! scp $exclude root@$computer:~/.backup/exclude > /dev/null; then
 194         echo `date`" $computer: Could not copy exclude.txt to $computer"
 195         die
 196 fi
 197
 198 # Try to copy preeexec and postexec if they exist
 199 if [ -f $confdir/preexec.$computer ]; then
 200         if ! scp $confdir/preexec.$computer  root@$computer:~/.backup/preexec >&2; then
 201                 echo `date`" $computer: Could not copy preexec.$computer to $computer:~/.backup/preexec"
 202                 die
 203         fi
 204 fi
 205 if [ -f $confdir/postexec.$computer ]; then
 206         if ! scp $confdir/postexec.$computer root@$computer:~/.backup/postexec >&2; then
 207                 echo `date`" $computer: Could not copy postexec.$computer to $computer:~/.backup/postexec"
 208                 die
 209         fi
 210 fi
 211
 212 # Try to run preexec if it exists
 213 if ! ssh root@$computer "[ ! -f ~/.backup/preexec ] || /bin/bash -x ~/.backup/preexec" >&2; then
 214         echo `date`" $computer: Could not run $computer:~/.backup/preexec"
 215         die
 216 fi
 217
 218 for filesystem in $filesystems
 219 do
 220         # Remove / and $ (for Windows) for path names
 221         sfilesystem=`echo $filesystem | tr '\/\$' '__'`
 222
 223         # Prepare storage area
 224         mkdir -m 755 -p $storagedir/$computer/$sfilesystem/full 2>/dev/null
 225         mkdir -m 755 -p $storagedir/$computer/$sfilesystem/daglig 2>/dev/null
 226
 227         # Default backuplevel
 228         backuplevel=daglig
 229
 230         if [ ! -f $storagedir/$computer/$sfilesystem/.date ]
 231         then
 232                 # Take the first full backup of this filesystem on this computer
 233                 backuplevel=full
 234                 echo $DATEs > $storagedir/$computer/$sfilesystem/.date
 235         fi
 236
 237         # Check if we want a full backup
 238         if [ -f $confdir/fastfullbackupdag.$computer.$sfilesystem ]; then
 239                 fullbackup_min_for_this_machine=$daysbetweenfullforfixed
 240                 if [ "$DAY_OF_MONTH" = "`cat $confdir/fastfullbackupdag.$computer.$sfilesystem`" ]; then
 241                         backuplevel=full
 242                 fi
 243         elif [ -f $confdir/fastfullbackupdag.$computer ]; then
 244                 fullbackup_min_for_this_machine=$daysbetweenfullforfixed
 245                 if [ "$DAY_OF_MONTH" = "`cat $confdir/fastfullbackupdag.$computer`" ]; then
 246                         backuplevel=full
 247                 fi
 248         else
 249                 fullbackup_min_for_this_machine=$daysbetweenfull
 250         fi
 251
 252         if [ -z "`find $storagedir/$computer/$sfilesystem/full/ -name \*tgz -mtime -$fullbackup_min_for_this_machine`" ]; then
 253                 backuplevel=full
 254         fi
 255
 256         # We want to be here
 257         cd $storagedir/$computer/$sfilesystem/$backuplevel || die
 258
 259         # Perform the actual backup
 260         backup
 261
 262         # Check if this box has a custom number of full backups
 263         if [ -f $confdir/maksfulle.$computer ] ; then
 264                 mf=$((`cat $confdir/maksfulle.$computer`+1))
 265         else
 266                 mf=$(($maxnumfull+1))
 267         fi
 268
 269         # Delete old full backups
 270         for full in `ls -1t $storagedir/$computer/$sfilesystem/full/*tgz | tail -n +$mf`
 271         do
 272                 prefix=`echo $full | sed "s/\.[^.]*$//"`
 273                 echo `date`": $computer:$filesystem Deleting full backup $prefix" >&2
 274                 rm $prefix*
 275         done
 276
 277         # Delete incremental backups older than the oldest full backup
 278         oldf=`ls -t1 $storagedir/$computer/$sfilesystem/full/*tgz | tail -1`
 279         find \
 280                 $storagedir/$computer/$sfilesystem/daglig \
 281                 -type f \
 282                 \! -newer $oldf \
 283                 -printf "`date`: $computer: Deleting old incremental backup: %p\n" \
 284                 -exec rm {} \; >&2
 285 done
 286
 287 # Try to run postexec if it exist
 288 if ! ssh root@$computer "[ ! -f ~/.backup/postexec ] || /bin/bash -x ~/.backup/postexec" >&2; then
 289         echo `date`" $computer: Could not run $computer:~/.backup/postexec"
 290         die
 291 fi
 292
 293 # Remove lockfile
 294 rm $LOCKFILE