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" ]; 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                                 diemsg "Could not copy .incremental.snar to $computer"
  86                         fi
  87                         lastcmd="--listed-incremental=$incrementalsnar --no-check-device"
  88                 else
  89                         echo `date`" $computer: Missing incremental.snar for $filesystem, doing date-based backup instead until next full backup" >&2
  90                         incrementalsnar=""
  91                         lastd=`cat ../.date`
  92                         lastcmd="--newer='$lastd'"
  93                 fi
  94         else
  95                 lastcmd="--listed-incremental=$incrementalsnar"
  96         fi
  97
  98         # We try to run tar on the remote computer
  99         #    c create archive
 100         #    C change to directory first
 101         #    . where to start taring (see C)
 102         #    $lastcmd only files newer than this
 103         #    --one-file-system don't traverse file systems
 104         #    --use-compress-program compress it using gzip or pigz
 105         #    --exclude-from file to get exclusion pattern from
 106         #    Pipe the stuff over ssh to ourselves, run pee to cat the contents to a
 107         #    file and build a file list, respectivly.
 108         TARFILE=$DATE.tmp
 109         TARCMD="ssh root@$computer \"nice -n 19 ionice -c3 tar --one-file-system --use-compress-program $compressor -cf - -C $filesystem . $lastcmd \
 110                 --exclude-from=/root/.backup/exclude\" | pee \"cat > $TARFILE\" \"tar tzvf -\""
 111         echo `date`" $computer: Running $TARCMD" >&2
 112         eval $TARCMD > $DATE.idx
 113
 114         if [ $? -eq 0 ] && [ -s $TARFILE ]; then
 115                 # File is >0 in size and neither cat or tar tzvf failed; we assume it worked.
 116
 117                 if [ "$incrementalsnar" ]; then
 118                         if ! scp root@$computer:$incrementalsnar $SNARFILE.tmp; then
 119                                 diemsg "Could not copy .incremental.snar from $computer"
 120                         fi
 121                 fi
 122
 123                 # Move tar file in place
 124                 mv $TARFILE $DATE.tgz
 125                 if [ "$incrementalsnar" ]; then
 126                         mv $SNARFILE.tmp $SNARFILE
 127                 fi
 128
 129                 # Update timestamp
 130                 echo $DATEs > ../.date && touch -t $DATE ../.date
 131
 132                 # Make a sorted file list as well
 133                 sort -k6 < $DATE.idx > $DATE.sdx
 134
 135                 # Fix permissions
 136                 chmod 600 *tgz                      #only for us
 137                 chmod 644 *sdx *idx 2>/dev/null     #everyone can read
 138
 139                 # Let the remote computer know that we ran a successful backup (for nagios)
 140                 ssh root@$computer touch $filesystem/.lastbackup
 141         else
 142                 # Something wrong happened.
 143                 rm $TARFILE
 144                 diemsg "tar failed or $TARFILE empty. $backuplevel backup of $computer:$filesystem failed and deleted"
 145         fi
 146 }
 147
 148 # Check that the target filesystem is mounted (actually check that it's not
 149 # the root filesystem)
 150 if [ ! -d "$storagedir/$computer" ]; then
 151         diemsg "Target filesystem ($storagedir/$computer) does not exist. Aborting"
 152 fi
 153 rootfilesystem=`df -P /`
 154 targetfilesystem=`df -P "$storagedir/$computer"`
 155 if [ "$rootfilesystem" == "$targetfilesystem" ]; then
 156         diemsg "Target filesystem ($storagedir/$computer) was mounted on /. Aborting"
 157 fi
 158
 159 echo `date`" $computer: Backing up $computer" >&2
 160
 161 # Try to SSH to the computer without entering a password.
 162 if ! `ssh -n -o NumberOfPasswordPrompts=0 root@$computer /bin/true`; then
 163         diemsg "Could not use passwordless SSH."
 164 fi
 165
 166 # Check if pigz is available
 167 if ssh -n root@$computer "pigz -V 2>/dev/null"; then
 168         compressor=pigz
 169 else
 170         echo `date`" $computer: pigz missing; falling back to gzip."
 171         compressor=gzip
 172 fi
 173
 174 # Check dump bit in fstab
 175 filesystems=`ssh -n root@$computer "cat /etc/fstab" \
 176         | grep -v nfs \
 177         | grep -v "^#" \
 178         | grep -v "^$" \
 179         | awk '{ if ( $(NF-1) != "0" ) print $2}' `
 180
 181 # Clean up our dir at this client
 182 if ! ssh root@$computer "rm -r /root/.backup ; mkdir -m 700 /root/.backup"; then
 183         diemsg "Could not create backup staging area at $computer:/root/.backup"
 184 fi
 185
 186 # Try to copy $exclude to $computer
 187 if ! scp $exclude root@$computer:/root/.backup/exclude > /dev/null; then
 188         diemsg "Could not copy exclude.txt to $computer"
 189 fi
 190
 191 # Try to copy preeexec and postexec if they exist
 192 if [ -f $confdir/preexec.$computer ]; then
 193         if ! scp $confdir/preexec.$computer  root@$computer:/root/.backup/preexec >&2; then
 194                 diemsg "Could not copy preexec.$computer to $computer:/root/.backup/preexec"
 195         fi
 196 fi
 197 if [ -f $confdir/postexec.$computer ]; then
 198         if ! scp $confdir/postexec.$computer root@$computer:/root/.backup/postexec >&2; then
 199                 diemsg "Could not copy postexec.$computer to $computer:/root/.backup/postexec"
 200         fi
 201 fi
 202
 203 # Try to run preexec if it exists
 204 if ! ssh root@$computer "[ ! -f /root/.backup/preexec ] || /bin/bash -x /root/.backup/preexec" >&2; then
 205         diemsg "Could not run $computer:/root/.backup/preexec"
 206 fi
 207
 208 for filesystem in $filesystems; do
 209         # Remove / and $ (for Windows) for path names
 210         sfilesystem=`echo $filesystem | tr '\/\$' '__'`
 211
 212         # Prepare storage area
 213         mkdir -m 755 -p $storagedir/$computer/$sfilesystem/full 2>/dev/null
 214         mkdir -m 755 -p $storagedir/$computer/$sfilesystem/daglig 2>/dev/null
 215
 216         # Default backuplevel
 217         backuplevel=daglig
 218
 219         if [ ! -s $storagedir/$computer/$sfilesystem/.date ]; then
 220                 # Take the first full backup of this filesystem on this computer
 221                 backuplevel=full
 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/$sfilesystem/$backuplevel || diemsg "$storagedir/$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.$sfilesystem ]; then
 251                 mf=$((`cat $confdir/maksfulle.$computer`+1))
 252         elif [ -f $confdir/maksfulle.$computer ] ; then
 253                 mf=$((`cat $confdir/maksfulle.$computer`+1))
 254         else
 255                 mf=$(($maxnumfull+1))
 256         fi
 257
 258         # Delete old full backups
 259         for full in `ls -1t $storagedir/$computer/$sfilesystem/full/*tgz | tail -n +$mf`; do
 260                 prefix=`echo $full | sed "s/\.[^.]*$//"`
 261                 echo `date`": $computer:$filesystem Deleting full backup $prefix" >&2
 262                 rm $prefix*
 263         done
 264
 265         # Delete incremental backups older than the oldest full backup
 266         oldf=`ls -t1 $storagedir/$computer/$sfilesystem/full/*tgz | tail -1`
 267         find \
 268                 $storagedir/$computer/$sfilesystem/daglig \
 269                 -type f \
 270                 \! -newer $oldf \
 271                 -printf "`date`: $computer: Deleting old incremental backup: %p\n" \
 272                 -exec rm {} \; >&2
 273 done
 274
 275 # Try to run postexec if it exist
 276 if ! ssh root@$computer "[ ! -f /root/.backup/postexec ] || /bin/bash -x /root/.backup/postexec" >&2; then
 277         diemsg "Could not run $computer:/root/.backup/postexec"
 278 fi
 279
 280 echo `date`": $computer: Backup completed successfully."
 281 echo `date`": $computer: Backup completed successfully." >&2
 282
 283 # Remove lockfile
 284 rm $LOCKFILE