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 maksantallfulle=3                       # Number of full backups
  32 dagermellomfulle=30                     # Days between full backups
  33 dagermellomfulleforfast=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. Ending, and not deleting lockfile $LOCKFILE."
  51         exit 1
  52 }
  53
  54 # Trap C-c and kill
  55 trap die SIGINT SIGTERM
  56
  57 # Don't start if we're already running
  58 if [ -e $LOCKFILE ]; then
  59         echo `date`": $LOCKFILE exists for pid `cat $LOCKFILE`, exiting."
  60         echo `date`": $LOCKFILE exists for pid `cat $LOCKFILE`, exiting." >&2
  61         exit 1
  62 fi;
  63 touch $LOCKFILE
  64 echo $$ > $LOCKFILE
  65
  66 echo `date`": Backup run starting" >&2
  67
  68 umask 027
  69
  70
  71 backup()
  72 {
  73         echo -n `date` >&2
  74         printf " $computer: $computer:$filesystem $backuplevel backup\n" >&2
  75
  76         SNARFILE="${storagedir}/${computer}/${sfilesystem}/.incremental.snar"
  77         incrementalsnar="/root/.backup/${sfilesystem}.snar"
  78         if [ "$backuplevel" = "daglig" ]
  79         then
  80                 # If incremental backup, we need to copy the incremental status to $computer.
  81                 # If it does not exist, whine a bit and then run date-based instead.
  82                 if [ -s "$SNARFILE" ]; then
  83                         if ! scp $SNARFILE root@$computer:$incrementalsnar; then
  84                                 echo `date`" $computer: Could not copy .incremental.snar to $computer"
  85                                 die
  86                         fi
  87                         lastcmd="--listed-incremental=$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=$incrementalsnar"
  96         fi
  97
  98         # expand the exclude-path for use with tar
  99         exf=`ssh root@$computer "ls ~/.backup/exclude"`
 100
 101         # We try to run tar on the remote computer
 102         #    c create archive
 103         #    C change to directory first
 104         #    . where to start taring (see C)
 105         #    $lastcmd only files newer than this
 106         #    --one-file-system don't traverse file systems
 107         #    --use-compress-program compress it using gzip or pigz
 108         #    --exclude-from file to get exclusion pattern from
 109         #    Pipe the stuff over ssh to ourselves, run pee to cat the contents to a
 110         #    file and build a file list, respectivly.
 111         TARFILE=$DATE.tmp
 112         TARCMD="ssh root@$computer \"nice -n 19 ionice -c3 tar --one-file-system --use-compress-program $compressor -cf - -C $filesystem . $lastcmd \
 113                 --exclude-from=$exf\" | pee \"cat > $TARFILE\" \"tar tzvf -\""
 114         echo `date`" $computer: Running $TARCMD" >&2
 115         eval $TARCMD > $DATE.idx
 116
 117         if [ $? -eq 0 ] && [ -s $TARFILE ]; then
 118                 # File is >0 in size and neither cat or tar tzvf failed; we assume it worked.
 119
 120                 if ! scp root@$computer:$incrementalsnar $SNARFILE.tmp; then
 121                         echo `date`" $computer: Could not copy .incremental.snar from $computer"
 122                         die
 123                 fi
 124
 125                 # Move tar file in place
 126                 mv $TARFILE $DATE.tgz
 127                 mv $SNARFILE.tmp $SNARFILE
 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                 echo `date`" $computer: $TARFILE empty. $backuplevel backup of $computer:$filesystem failed and deleted"
 145                 die
 146         fi
 147 }
 148
 149 # Check that the target filesystem is mounted (actually check that it's not
 150 # the root filesystem)
 151 if [ ! -d "$storagedir/$computer" ]; then
 152         echo `date`" $computer: Target filesystem ($storagedir/$computer) does not exist. Aborting"
 153         die
 154 fi
 155 rootfilesystem=`df -P /`
 156 targetfilesystem=`df -P "$storagedir/$computer"`
 157 if [ "$rootfilesystem" == "$targetfilesystem" ]; then
 158         echo `date`" $computer: Target filesystem ($storagedir/$computer) was mounted on /. Aborting"
 159         die
 160 fi
 161
 162 echo `date`" $computer: Backing up $computer" >&2
 163
 164 # Try to SSH to the computer without entering a password.
 165 if ! `ssh -n -o NumberOfPasswordPrompts=0 root@$computer /bin/true`; then
 166         echo `date`" $computer: Could not use passwordless SSH."
 167         die
 168 fi
 169
 170 # Check if pigz is available
 171 if ssh -n root@$computer "pigz -V 2>/dev/null"; then
 172         compressor=pigz
 173 else
 174         echo `date`" $computer: pigz missing; falling back to gzip."
 175         compressor=gzip
 176 fi
 177
 178 # Check dump bit in fstab
 179 filesystems=`ssh -n root@$computer "cat /etc/fstab" \
 180         | grep -v nfs \
 181         | grep -v "^#" \
 182         | grep -v "^$" \
 183         | awk '{ if ( $(NF-1) != "0" ) print $2}' `
 184
 185 # Clean up our dir at this client
 186 if ! ssh root@$computer "rm -r ~/.backup ; mkdir -m 700 ~/.backup"; then
 187         echo `date`" $computer: Could not create backup staging area at $computer:~/.backup"
 188         die
 189 fi
 190
 191 # Try to copy $exclude to $computer
 192 if ! scp $exclude root@$computer:~/.backup/exclude > /dev/null; then
 193         echo `date`" $computer: Could not copy exclude.txt to $computer"
 194         die
 195 fi
 196
 197 # Try to copy preeexec and postexec if they exist
 198 if [ -f $confdir/preexec.$computer ]; then
 199         if ! scp $confdir/preexec.$computer  root@$computer:~/.backup/preexec >&2; then
 200                 echo `date`" $computer: Could not copy preexec.$computer to $computer:~/.backup/preexec"
 201                 die
 202         fi
 203 fi
 204 if [ -f $confdir/postexec.$computer ]; then
 205         if ! scp $confdir/postexec.$computer root@$computer:~/.backup/postexec >&2; then
 206                 echo `date`" $computer: Could not copy postexec.$computer to $computer:~/.backup/postexec"
 207                 die
 208         fi
 209 fi
 210
 211 # Try to run preexec if it exists
 212 if ! ssh root@$computer "[ ! -f ~/.backup/preexec ] || /bin/bash -x ~/.backup/preexec" >&2; then
 213         echo `date`" $computer: Could not run $computer:~/.backup/preexec"
 214         die
 215 fi
 216
 217 for filesystem in $filesystems
 218 do
 219         # Remove / and $ (for Windows) for path names
 220         sfilesystem=`echo $filesystem | tr '\/\$' '__'`
 221
 222         # Prepare storage area
 223         mkdir -m 755 -p $storagedir/$computer/$sfilesystem/full 2>/dev/null
 224         mkdir -m 755 -p $storagedir/$computer/$sfilesystem/daglig 2>/dev/null
 225
 226         # Default backuplevel
 227         backuplevel=daglig
 228
 229         if [ ! -f $storagedir/$computer/$sfilesystem/.date ]
 230         then
 231                 # Take the first full backup of this filesystem on this computer
 232                 backuplevel=full
 233                 echo $DATEs > $storagedir/$computer/$sfilesystem/.date
 234         fi
 235
 236         # Check if we want a full backup
 237         if [ -f $confdir/fastfullbackupdag.$computer.$sfilesystem ]; then
 238                 fullbackup_min_for_this_machine=$dagermellomfulleforfast
 239                 if [ "$DAY_OF_MONTH" = "`cat $confdir/fastfullbackupdag.$computer.$sfilesystem`" ]; then
 240                         backuplevel=full
 241                 fi
 242         elif [ -f $confdir/fastfullbackupdag.$computer ]; then
 243                 fullbackup_min_for_this_machine=$dagermellomfulleforfast
 244                 if [ "$DAY_OF_MONTH" = "`cat $confdir/fastfullbackupdag.$computer`" ]; then
 245                         backuplevel=full
 246                 fi
 247         else
 248                 fullbackup_min_for_this_machine=$dagermellomfulle
 249         fi
 250
 251         if [ -z "`find $storagedir/$computer/$sfilesystem/full/ -name \*tgz -mtime -$fullbackup_min_for_this_machine`" ]; then
 252                 backuplevel=full
 253         fi
 254
 255         # We want to be here
 256         cd $storagedir/$computer/$sfilesystem/$backuplevel || die
 257
 258         # Perform the actual backup
 259         backup
 260
 261         # Check if this box has a custom number of full backups
 262         if [ -f $confdir/maksfulle.$computer ] ; then
 263                 mf=$((`cat $confdir/maksfulle.$computer`+1))
 264         else
 265                 mf=$(($maksantallfulle+1))
 266         fi
 267
 268         # Delete old full backups
 269         for full in `ls -1t $storagedir/$computer/$sfilesystem/full/*tgz | tail -n +$mf`
 270         do
 271                 prefix=`echo $full | sed "s/\.[^.]*$//"`
 272                 echo `date`": $computer:$filesystem Deleting full backup $prefix" >&2
 273                 rm $prefix*
 274         done
 275
 276         # Delete incremental backups older than the oldest full backup
 277         oldf=`ls -t1 $storagedir/$computer/$sfilesystem/full/*tgz | tail -1`
 278         find \
 279                 $storagedir/$computer/$sfilesystem/daglig \
 280                 -type f \
 281                 \! -newer $oldf \
 282                 -printf "`date`: $computer: Deleting old incremental backup: %p\n" \
 283                 -exec rm {} \; >&2
 284 done
 285
 286 # Try to run postexec if it exist
 287 if ! ssh root@$computer "[ ! -f ~/.backup/postexec ] || /bin/bash -x ~/.backup/postexec" >&2; then
 288         echo `date`" $computer: Could not run $computer:~/.backup/postexec"
 289         die
 290 fi
 291
 292 # Remove lockfile
 293 rm $LOCKFILE