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         # We try to run tar on the remote computer
 100         #    c create archive
 101         #    C change to directory first
 102         #    . where to start taring (see C)
 103         #    $lastcmd only files newer than this
 104         #    --one-file-system don't traverse file systems
 105         #    --use-compress-program compress it using gzip or pigz
 106         #    --exclude-from file to get exclusion pattern from
 107         #    Pipe the stuff over ssh to ourselves, run pee to cat the contents to a
 108         #    file and build a file list, respectivly.
 109         TARFILE=$DATE.tmp
 110         TARCMD="ssh root@$computer \"nice -n 19 ionice -c3 tar --one-file-system --use-compress-program $compressor -cf - -C $filesystem . $lastcmd \
 111                 --exclude-from=/root/.backup/exclude\" | pee \"cat > $TARFILE\" \"tar tzvf -\""
 112         echo `date`" $computer: Running $TARCMD" >&2
 113         eval $TARCMD > $DATE.idx
 114
 115         if [ $? -eq 0 ] && [ -s $TARFILE ]; then
 116                 # File is >0 in size and neither cat or tar tzvf failed; we assume it worked.
 117
 118                 if ! scp root@$computer:$incrementalsnar $SNARFILE.tmp; then
 119                         echo `date`" $computer: Could not copy .incremental.snar from $computer"
 120                         die
 121                 fi
 122
 123                 # Move tar file in place
 124                 mv $TARFILE $DATE.tgz
 125                 mv $SNARFILE.tmp $SNARFILE
 126
 127                 # Update timestamp
 128                 echo $DATEs > ../.date && touch -t $DATE ../.date
 129
 130                 # Make a sorted file list as well
 131                 sort -k6 < $DATE.idx > $DATE.sdx
 132
 133                 # Fix permissions
 134                 chmod 600 *tgz                      #only for us
 135                 chmod 644 *sdx *idx 2>/dev/null     #everyone can read
 136
 137                 # Let the remote computer know that we ran a successful backup (for nagios)
 138                 ssh root@$computer touch $filesystem/.lastbackup
 139         else
 140                 # Something wrong happened.
 141                 rm $TARFILE
 142                 echo `date`" $computer: $TARFILE empty. $backuplevel backup of $computer:$filesystem failed and deleted"
 143                 die
 144         fi
 145 }
 146
 147 # Check that the target filesystem is mounted (actually check that it's not
 148 # the root filesystem)
 149 if [ ! -d "$storagedir/$computer" ]; then
 150         echo `date`" $computer: Target filesystem ($storagedir/$computer) does not exist. Aborting"
 151         die
 152 fi
 153 rootfilesystem=`df -P /`
 154 targetfilesystem=`df -P "$storagedir/$computer"`
 155 if [ "$rootfilesystem" == "$targetfilesystem" ]; then
 156         echo `date`" $computer: Target filesystem ($storagedir/$computer) was mounted on /. Aborting"
 157         die
 158 fi
 159
 160 echo `date`" $computer: Backing up $computer" >&2
 161
 162 # Try to SSH to the computer without entering a password.
 163 if ! `ssh -n -o NumberOfPasswordPrompts=0 root@$computer /bin/true`; then
 164         echo `date`" $computer: Could not use passwordless SSH."
 165         die
 166 fi
 167
 168 # Check if pigz is available
 169 if ssh -n root@$computer "pigz -V 2>/dev/null"; then
 170         compressor=pigz
 171 else
 172         echo `date`" $computer: pigz missing; falling back to gzip."
 173         compressor=gzip
 174 fi
 175
 176 # Check dump bit in fstab
 177 filesystems=`ssh -n root@$computer "cat /etc/fstab" \
 178         | grep -v nfs \
 179         | grep -v "^#" \
 180         | grep -v "^$" \
 181         | awk '{ if ( $(NF-1) != "0" ) print $2}' `
 182
 183 # Clean up our dir at this client
 184 if ! ssh root@$computer "rm -r /root/.backup ; mkdir -m 700 /root/.backup"; then
 185         echo `date`" $computer: Could not create backup staging area at $computer:/root/.backup"
 186         die
 187 fi
 188
 189 # Try to copy $exclude to $computer
 190 if ! scp $exclude root@$computer:/root/.backup/exclude > /dev/null; then
 191         echo `date`" $computer: Could not copy exclude.txt to $computer"
 192         die
 193 fi
 194
 195 # Try to copy preeexec and postexec if they exist
 196 if [ -f $confdir/preexec.$computer ]; then
 197         if ! scp $confdir/preexec.$computer  root@$computer:/root/.backup/preexec >&2; then
 198                 echo `date`" $computer: Could not copy preexec.$computer to $computer:/root/.backup/preexec"
 199                 die
 200         fi
 201 fi
 202 if [ -f $confdir/postexec.$computer ]; then
 203         if ! scp $confdir/postexec.$computer root@$computer:/root/.backup/postexec >&2; then
 204                 echo `date`" $computer: Could not copy postexec.$computer to $computer:/root/.backup/postexec"
 205                 die
 206         fi
 207 fi
 208
 209 # Try to run preexec if it exists
 210 if ! ssh root@$computer "[ ! -f /root/.backup/preexec ] || /bin/bash -x /root/.backup/preexec" >&2; then
 211         echo `date`" $computer: Could not run $computer:/root/.backup/preexec"
 212         die
 213 fi
 214
 215 for filesystem in $filesystems
 216 do
 217         # Remove / and $ (for Windows) for path names
 218         sfilesystem=`echo $filesystem | tr '\/\$' '__'`
 219
 220         # Prepare storage area
 221         mkdir -m 755 -p $storagedir/$computer/$sfilesystem/full 2>/dev/null
 222         mkdir -m 755 -p $storagedir/$computer/$sfilesystem/daglig 2>/dev/null
 223
 224         # Default backuplevel
 225         backuplevel=daglig
 226
 227         if [ ! -f $storagedir/$computer/$sfilesystem/.date ]
 228         then
 229                 # Take the first full backup of this filesystem on this computer
 230                 backuplevel=full
 231                 echo $DATEs > $storagedir/$computer/$sfilesystem/.date
 232         fi
 233
 234         # Check if we want a full backup
 235         if [ -f $confdir/fastfullbackupdag.$computer.$sfilesystem ]; then
 236                 fullbackup_min_for_this_machine=$daysbetweenfullforfixed
 237                 if [ "$DAY_OF_MONTH" = "`cat $confdir/fastfullbackupdag.$computer.$sfilesystem`" ]; then
 238                         backuplevel=full
 239                 fi
 240         elif [ -f $confdir/fastfullbackupdag.$computer ]; then
 241                 fullbackup_min_for_this_machine=$daysbetweenfullforfixed
 242                 if [ "$DAY_OF_MONTH" = "`cat $confdir/fastfullbackupdag.$computer`" ]; then
 243                         backuplevel=full
 244                 fi
 245         else
 246                 fullbackup_min_for_this_machine=$daysbetweenfull
 247         fi
 248
 249         if [ -z "`find $storagedir/$computer/$sfilesystem/full/ -name \*tgz -mtime -$fullbackup_min_for_this_machine`" ]; then
 250                 backuplevel=full
 251         fi
 252
 253         # We want to be here
 254         cd $storagedir/$computer/$sfilesystem/$backuplevel || die
 255
 256         # Perform the actual backup
 257         backup
 258
 259         # Check if this box has a custom number of full backups
 260         if [ -f $confdir/maksfulle.$computer ] ; then
 261                 mf=$((`cat $confdir/maksfulle.$computer`+1))
 262         else
 263                 mf=$(($maxnumfull+1))
 264         fi
 265
 266         # Delete old full backups
 267         for full in `ls -1t $storagedir/$computer/$sfilesystem/full/*tgz | tail -n +$mf`
 268         do
 269                 prefix=`echo $full | sed "s/\.[^.]*$//"`
 270                 echo `date`": $computer:$filesystem Deleting full backup $prefix" >&2
 271                 rm $prefix*
 272         done
 273
 274         # Delete incremental backups older than the oldest full backup
 275         oldf=`ls -t1 $storagedir/$computer/$sfilesystem/full/*tgz | tail -1`
 276         find \
 277                 $storagedir/$computer/$sfilesystem/daglig \
 278                 -type f \
 279                 \! -newer $oldf \
 280                 -printf "`date`: $computer: Deleting old incremental backup: %p\n" \
 281                 -exec rm {} \; >&2
 282 done
 283
 284 # Try to run postexec if it exist
 285 if ! ssh root@$computer "[ ! -f /root/.backup/postexec ] || /bin/bash -x /root/.backup/postexec" >&2; then
 286         echo `date`" $computer: Could not run $computer:/root/.backup/postexec"
 287         die
 288 fi
 289
 290 # Remove lockfile
 291 rm $LOCKFILE