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 # Because of the way we run this script from cron, informational logging goes
  15 # to stderr and warnings and errors to stdout. This way, we get mail whenever
  16 # stuff fails, and other output is redirected to the log file.
  17 #
  18 # The script is halfway Norwegian and halfway English. Newer modifications are
  19 # in English, we should probably stick to that.
  20
  21 # The computer to backup.
  22 computer=$1
  23
  24 # Configuration
  25 LOCKFILE=/home/backup/backuprun.lockfile.$computer
  26 confdir=/home/backup/conf/              # Configuration files
  27 storagedir=/backup                      # Where we keep backups
  28 maxnumfull=3                            # Number of full backups
  29 daysbetweenfull=30                      # Days between full backups
  30 daysbetweenfullforfixed=32              # Days between full backups for machines with fixed
  31                                         # full-backup date (in case the fixed day gets missed)
  32 DAY_OF_MONTH=`date "+%d" | sed s/^0//`  # from 1 to 31
  33 DATE=`date "+%Y%m%d%H%M"`               #format: touch
  34 DATEs=`date "+%Y-%m-%d %H:%M"`          #format: tar
  35
  36 # Exclude-pattern
  37 exclude=$confdir/exclude
  38 [ ! -f $exclude ] && printf "tmp\ncore\n" > $exclude
  39
  40 if [ -z "$computer" ]; then
  41         echo "Usage: $0 COMPUTER"
  42         exit 1
  43 fi
  44
  45 # Die more or less gracefully (at least notify user)
  46 die() {
  47         echo `date`": Something nasty happened."
  48         rm $LOCKFILE
  49         exit 1
  50 }
  51 diemsg() {
  52         echo `date`": $computer: $@"
  53         die
  54 }
  55
  56 # Trap C-c and kill
  57 trap die SIGINT SIGTERM
  58
  59 # Don't start if we're already running
  60 if [ -e $LOCKFILE ]; then
  61         echo `date`": $LOCKFILE exists for pid `cat $LOCKFILE`, exiting."
  62         echo `date`": $LOCKFILE exists for pid `cat $LOCKFILE`, exiting." >&2
  63         exit 1
  64 fi;
  65 touch $LOCKFILE
  66 echo $$ > $LOCKFILE
  67
  68 echo `date`": Backup run starting" >&2
  69
  70 umask 027
  71
  72
  73 backup()
  74 {
  75         echo -n `date` >&2
  76         printf " $computer: $computer:$filesystem $backuplevel backup\n" >&2
  77
  78         SNARFILE="${storagedir}/${computer}/${sfilesystem}/.incremental.snar"
  79         incrementalsnar="/root/.backup/${sfilesystem}.snar"
  80         if [ "$backuplevel" = "daglig" ]
  81         then
  82                 # If incremental backup, we need to copy the incremental status to $computer.
  83                 # If it does not exist, whine a bit and then run date-based instead.
  84                 if [ -s "$SNARFILE" ]; then
  85                         if ! scp $SNARFILE root@$computer:$incrementalsnar; then
  86                                 diemsg "Could not copy .incremental.snar to $computer"
  87                         fi
  88                         lastcmd="--listed-incremental=$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                         diemsg "Could not copy .incremental.snar from $computer"
 120                 fi
 121
 122                 # Move tar file in place
 123                 mv $TARFILE $DATE.tgz
 124                 mv $SNARFILE.tmp $SNARFILE
 125
 126                 # Update timestamp
 127                 echo $DATEs > ../.date && touch -t $DATE ../.date
 128
 129                 # Make a sorted file list as well
 130                 sort -k6 < $DATE.idx > $DATE.sdx
 131
 132                 # Fix permissions
 133                 chmod 600 *tgz                      #only for us
 134                 chmod 644 *sdx *idx 2>/dev/null     #everyone can read
 135
 136                 # Let the remote computer know that we ran a successful backup (for nagios)
 137                 ssh root@$computer touch $filesystem/.lastbackup
 138         else
 139                 # Something wrong happened.
 140                 rm $TARFILE
 141                 diemsg "tar failed or $TARFILE empty. $backuplevel backup of $computer:$filesystem failed and deleted"
 142         fi
 143 }
 144
 145 # Check that the target filesystem is mounted (actually check that it's not
 146 # the root filesystem)
 147 if [ ! -d "$storagedir/$computer" ]; then
 148         diemsg "Target filesystem ($storagedir/$computer) does not exist. Aborting"
 149 fi
 150 rootfilesystem=`df -P /`
 151 targetfilesystem=`df -P "$storagedir/$computer"`
 152 if [ "$rootfilesystem" == "$targetfilesystem" ]; then
 153         diemsg "Target filesystem ($storagedir/$computer) was mounted on /. Aborting"
 154 fi
 155
 156 echo `date`" $computer: Backing up $computer" >&2
 157
 158 # Try to SSH to the computer without entering a password.
 159 if ! `ssh -n -o NumberOfPasswordPrompts=0 root@$computer /bin/true`; then
 160         diemsg "Could not use passwordless SSH."
 161 fi
 162
 163 # Check if pigz is available
 164 if ssh -n root@$computer "pigz -V 2>/dev/null"; then
 165         compressor=pigz
 166 else
 167         echo `date`" $computer: pigz missing; falling back to gzip."
 168         compressor=gzip
 169 fi
 170
 171 # Check dump bit in fstab
 172 filesystems=`ssh -n root@$computer "cat /etc/fstab" \
 173         | grep -v nfs \
 174         | grep -v "^#" \
 175         | grep -v "^$" \
 176         | awk '{ if ( $(NF-1) != "0" ) print $2}' `
 177
 178 # Clean up our dir at this client
 179 if ! ssh root@$computer "rm -r /root/.backup ; mkdir -m 700 /root/.backup"; then
 180         diemsg "Could not create backup staging area at $computer:/root/.backup"
 181 fi
 182
 183 # Try to copy $exclude to $computer
 184 if ! scp $exclude root@$computer:/root/.backup/exclude > /dev/null; then
 185         diemsg "Could not copy exclude.txt to $computer"
 186 fi
 187
 188 # Try to copy preeexec and postexec if they exist
 189 if [ -f $confdir/preexec.$computer ]; then
 190         if ! scp $confdir/preexec.$computer  root@$computer:/root/.backup/preexec >&2; then
 191                 diemsg "Could not copy preexec.$computer to $computer:/root/.backup/preexec"
 192         fi
 193 fi
 194 if [ -f $confdir/postexec.$computer ]; then
 195         if ! scp $confdir/postexec.$computer root@$computer:/root/.backup/postexec >&2; then
 196                 diemsg "Could not copy postexec.$computer to $computer:/root/.backup/postexec"
 197         fi
 198 fi
 199
 200 # Try to run preexec if it exists
 201 if ! ssh root@$computer "[ ! -f /root/.backup/preexec ] || /bin/bash -x /root/.backup/preexec" >&2; then
 202         diemsg "Could not run $computer:/root/.backup/preexec"
 203 fi
 204
 205 for filesystem in $filesystems
 206 do
 207         # Remove / and $ (for Windows) for path names
 208         sfilesystem=`echo $filesystem | tr '\/\$' '__'`
 209
 210         # Prepare storage area
 211         mkdir -m 755 -p $storagedir/$computer/$sfilesystem/full 2>/dev/null
 212         mkdir -m 755 -p $storagedir/$computer/$sfilesystem/daglig 2>/dev/null
 213
 214         # Default backuplevel
 215         backuplevel=daglig
 216
 217         if [ ! -f $storagedir/$computer/$sfilesystem/.date ]
 218         then
 219                 # Take the first full backup of this filesystem on this computer
 220                 backuplevel=full
 221                 echo $DATEs > $storagedir/$computer/$sfilesystem/.date
 222         fi
 223
 224         # Check if we want a full backup
 225         if [ -f $confdir/fastfullbackupdag.$computer.$sfilesystem ]; then
 226                 fullbackup_min_for_this_machine=$daysbetweenfullforfixed
 227                 if [ "$DAY_OF_MONTH" = "`cat $confdir/fastfullbackupdag.$computer.$sfilesystem`" ]; then
 228                         backuplevel=full
 229                 fi
 230         elif [ -f $confdir/fastfullbackupdag.$computer ]; then
 231                 fullbackup_min_for_this_machine=$daysbetweenfullforfixed
 232                 if [ "$DAY_OF_MONTH" = "`cat $confdir/fastfullbackupdag.$computer`" ]; then
 233                         backuplevel=full
 234                 fi
 235         else
 236                 fullbackup_min_for_this_machine=$daysbetweenfull
 237         fi
 238
 239         if [ -z "`find $storagedir/$computer/$sfilesystem/full/ -name \*tgz -mtime -$fullbackup_min_for_this_machine`" ]; then
 240                 backuplevel=full
 241         fi
 242
 243         # We want to be here
 244         cd $storagedir/$computer/$sfilesystem/$backuplevelmsg || die "$storagedir/$computer/$sfilesystem/$backuplevel does not exist"
 245
 246         # Perform the actual backup
 247         backup
 248
 249         # Check if this box has a custom number of full backups
 250         if [ -f $confdir/maksfulle.$computer ] ; then
 251                 mf=$((`cat $confdir/maksfulle.$computer`+1))
 252         else
 253                 mf=$(($maxnumfull+1))
 254         fi
 255
 256         # Delete old full backups
 257         for full in `ls -1t $storagedir/$computer/$sfilesystem/full/*tgz | tail -n +$mf`
 258         do
 259                 prefix=`echo $full | sed "s/\.[^.]*$//"`
 260                 echo `date`": $computer:$filesystem Deleting full backup $prefix" >&2
 261                 rm $prefix*
 262         done
 263
 264         # Delete incremental backups older than the oldest full backup
 265         oldf=`ls -t1 $storagedir/$computer/$sfilesystem/full/*tgz | tail -1`
 266         find \
 267                 $storagedir/$computer/$sfilesystem/daglig \
 268                 -type f \
 269                 \! -newer $oldf \
 270                 -printf "`date`: $computer: Deleting old incremental backup: %p\n" \
 271                 -exec rm {} \; >&2
 272 done
 273
 274 # Try to run postexec if it exist
 275 if ! ssh root@$computer "[ ! -f /root/.backup/postexec ] || /bin/bash -x /root/.backup/postexec" >&2; then
 276         diemsg "Could not run $computer:/root/.backup/postexec"
 277 fi
 278
 279 # Remove lockfile
 280 rm $LOCKFILE