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         warnmsg "Aborting."
  48         rm $LOCKFILE
  49         exit 1
  50 }
  51 diemsg() {
  52         warnmsg "$@"
  53         die
  54 }
  55 warnmsg() {
  56         echo `date`": $computer: $@"
  57         echo `date`": $computer: $@" >&2
  58 }
  59 infomsg() {
  60         echo `date`": $computer: $@" >&2
  61 }
  62
  63 # Trap C-c and kill
  64 trap die SIGINT SIGTERM
  65
  66 # Don't start if we're already running
  67 if [ -e $LOCKFILE ]; then
  68         diemsg "$LOCKFILE exists for pid `cat $LOCKFILE`."
  69 fi;
  70 touch $LOCKFILE
  71 echo $$ > $LOCKFILE
  72
  73 infomsg "Backup run starting"
  74
  75 umask 027
  76
  77 backup()
  78 {
  79         infomsg "$computer:$filesystem $backuplevel backup"
  80
  81         localsnar="${storagedir}/${computer}/${sfilesystem}/.incremental.snar"
  82         remotesnar="/root/.backup/${sfilesystem}.snar"
  83         if [ "$backuplevel" = "daglig" ]; then
  84                 # If incremental backup, we need to copy the incremental status to $computer.
  85                 # If it does not exist, whine a bit and then run date-based instead.
  86                 if [ -s "$localsnar" ]; then
  87                         if ! scp $localsnar root@$computer:$remotesnar; then
  88                                 diemsg "Could not copy .incremental.snar to $computer"
  89                         fi
  90                         lastcmd="--listed-incremental=$remotesnar --no-check-device"
  91                 else
  92                         infomsg "Missing incremental.snar for $filesystem, doing date-based backup instead until next full backup"
  93                         remotesnar=""
  94                         lastd=`cat ../.date`
  95                         lastcmd="--newer='$lastd'"
  96                 fi
  97         else
  98                 # Note: Since we rm-ed .backup previously, and we don't scp over
  99                 # anything in this branch, this file will not exist, and we
 100                 # will be doing a full backup (but we want the output of it).
 101                 lastcmd="--listed-incremental=$remotesnar"
 102         fi
 103
 104         # We try to run tar on the remote computer
 105         #    c create archive
 106         #    C change to directory first
 107         #    . where to start taring (see C)
 108         #    $lastcmd only files newer than this
 109         #    --one-file-system don't traverse file systems
 110         #    --use-compress-program compress it using gzip or pigz
 111         #    --exclude-from file to get exclusion pattern from
 112         #    Pipe the stuff over ssh to ourselves, run pee to cat the contents to a
 113         #    file and build a file list, respectivly.
 114         TARFILE=$DATE.tmp
 115         TARCMD="ssh root@$computer \"nice -n 19 ionice -c3 tar --one-file-system --use-compress-program $compressor -cf - -C $filesystem . $lastcmd \
 116                 --exclude-from=/root/.backup/exclude\" | pee \"cat > $TARFILE\" \"tar tzvf -\""
 117         infomsg "Running $TARCMD"
 118         eval $TARCMD > $DATE.idx
 119
 120         if [ $? -eq 0 ] && [ -s $TARFILE ]; then
 121                 # File is >0 in size and neither cat or tar tzvf failed; we assume it worked.
 122
 123                 if [ "$remotesnar" ]; then
 124                         if ! scp root@$computer:$remotesnar $localsnar.tmp; then
 125                                 diemsg "Could not copy .incremental.snar from $computer"
 126                         fi
 127                 fi
 128
 129                 # Move tar file in place
 130                 mv $TARFILE $DATE.tgz
 131                 if [ "$remotesnar" ]; then
 132                         mv $localsnar.tmp $localsnar
 133                 fi
 134
 135                 # Update timestamp
 136                 echo $DATEs > ../.date && touch -t $DATE ../.date
 137
 138                 # Make a sorted file list as well
 139                 sort -k6 < $DATE.idx > $DATE.sdx
 140
 141                 # Fix permissions
 142                 chmod 600 *tgz                      #only for us
 143                 chmod 644 *sdx *idx 2>/dev/null     #everyone can read
 144
 145                 # Let the remote computer know that we ran a successful backup (for nagios)
 146                 ssh root@$computer touch $filesystem/.lastbackup
 147         else
 148                 # Something wrong happened.
 149                 rm $TARFILE
 150                 diemsg "tar failed or $TARFILE empty. $backuplevel backup of $computer:$filesystem failed and deleted"
 151         fi
 152 }
 153
 154 # Check that the target filesystem is mounted (actually check that it's not
 155 # the root filesystem)
 156 if [ ! -d "$storagedir/$computer" ]; then
 157         diemsg "Target filesystem ($storagedir/$computer) does not exist."
 158 fi
 159 rootfilesystem=`df -P /`
 160 targetfilesystem=`df -P "$storagedir/$computer"`
 161 if [ "$rootfilesystem" == "$targetfilesystem" ]; then
 162         diemsg "Target filesystem ($storagedir/$computer) was mounted on /."
 163 fi
 164
 165 infomsg "Backing up $computer"
 166
 167 # Try to SSH to the computer without entering a password.
 168 if ! ssh -n -o NumberOfPasswordPrompts=0 root@$computer /bin/true; then
 169         diemsg "Could not use passwordless SSH."
 170 fi
 171
 172 # Check if pigz is available
 173 if ssh -n root@$computer "pigz -V 2>/dev/null"; then
 174         compressor=pigz
 175 else
 176         infomsg "pigz missing; falling back to gzip."
 177         compressor=gzip
 178 fi
 179
 180 # Check dump bit in fstab
 181 filesystems=`ssh -n root@$computer "cat /etc/fstab" \
 182         | grep -v nfs \
 183         | grep -v "^#" \
 184         | grep -v "^$" \
 185         | awk '{ if ( $(NF-1) != "0" ) print $2}' `
 186
 187 # Clean up our dir at this client
 188 if ! ssh root@$computer "rm -r /root/.backup ; mkdir -m 700 /root/.backup"; then
 189         diemsg "Could not create backup staging area at $computer:/root/.backup"
 190 fi
 191
 192 # Try to copy $exclude to $computer
 193 if ! scp $exclude root@$computer:/root/.backup/exclude > /dev/null; then
 194         diemsg "Could not copy exclude.txt to $computer"
 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@$computer:/root/.backup/preexec >&2; then
 200                 diemsg "Could not copy preexec.$computer to $computer:/root/.backup/preexec"
 201         fi
 202 fi
 203 if [ -f $confdir/postexec.$computer ]; then
 204         if ! scp $confdir/postexec.$computer root@$computer:/root/.backup/postexec >&2; then
 205                 diemsg "Could not copy postexec.$computer to $computer:/root/.backup/postexec"
 206         fi
 207 fi
 208
 209 # Try to run preexec if it exists
 210 if ! ssh root@$computer "[ ! -f /root/.backup/preexec ] || /bin/bash -x /root/.backup/preexec" >&2; then
 211         diemsg "Could not run $computer:/root/.backup/preexec"
 212 fi
 213
 214 for filesystem in $filesystems; do
 215         # Remove / and $ (for Windows) for path names
 216         sfilesystem=`echo $filesystem | tr '\/\$' '__'`
 217
 218         # Prepare storage area
 219         mkdir -m 755 -p $storagedir/$computer/$sfilesystem/full 2>/dev/null
 220         mkdir -m 755 -p $storagedir/$computer/$sfilesystem/daglig 2>/dev/null
 221
 222         # Default backuplevel
 223         backuplevel=daglig
 224
 225         if [ ! -s $storagedir/$computer/$sfilesystem/.date ]; then
 226                 # Take the first full backup of this filesystem on this computer
 227                 backuplevel=full
 228         fi
 229
 230         # Check if we want a full backup
 231         if [ -f $confdir/fastfullbackupdag.$computer.$sfilesystem ]; then
 232                 fullbackup_min_for_this_machine=$daysbetweenfullforfixed
 233                 if [ "$DAY_OF_MONTH" = "`cat $confdir/fastfullbackupdag.$computer.$sfilesystem`" ]; then
 234                         backuplevel=full
 235                 fi
 236         elif [ -f $confdir/fastfullbackupdag.$computer ]; then
 237                 fullbackup_min_for_this_machine=$daysbetweenfullforfixed
 238                 if [ "$DAY_OF_MONTH" = "`cat $confdir/fastfullbackupdag.$computer`" ]; then
 239                         backuplevel=full
 240                 fi
 241         else
 242                 fullbackup_min_for_this_machine=$daysbetweenfull
 243         fi
 244
 245         if [ -z "`find $storagedir/$computer/$sfilesystem/full/ -name \*tgz -mtime -$fullbackup_min_for_this_machine`" ]; then
 246                 backuplevel=full
 247         fi
 248
 249         # We want to be here
 250         cd $storagedir/$computer/$sfilesystem/$backuplevel || diemsg "$storagedir/$computer/$sfilesystem/$backuplevel does not exist"
 251
 252         # Perform the actual backup
 253         backup
 254
 255         # Check if this box has a custom number of full backups
 256         if [ -f $confdir/maksfulle.$computer.$sfilesystem ]; then
 257                 mf=$((`cat $confdir/maksfulle.$computer`+1))
 258         elif [ -f $confdir/maksfulle.$computer ] ; then
 259                 mf=$((`cat $confdir/maksfulle.$computer`+1))
 260         else
 261                 mf=$(($maxnumfull+1))
 262         fi
 263
 264         # Delete old full backups
 265         for full in `ls -1t $storagedir/$computer/$sfilesystem/full/*tgz | tail -n +$mf`; do
 266                 prefix=`echo $full | sed "s/\.[^.]*$//"`
 267                 infomsg "$computer:$filesystem Deleting full backup $prefix"
 268                 rm $prefix*
 269         done
 270
 271         # Delete incremental backups older than the oldest full backup
 272         oldf=`ls -t1 $storagedir/$computer/$sfilesystem/full/*tgz | tail -1`
 273         find \
 274                 $storagedir/$computer/$sfilesystem/daglig \
 275                 -type f \
 276                 \! -newer $oldf \
 277                 -printf "`date`: $computer: Deleting old incremental backup: %p\n" \
 278                 -exec rm {} \; >&2
 279 done
 280
 281 # Try to run postexec if it exist
 282 if ! ssh root@$computer "[ ! -f /root/.backup/postexec ] || /bin/bash -x /root/.backup/postexec" >&2; then
 283         diemsg "Could not run $computer:/root/.backup/postexec"
 284 fi
 285
 286 # Use warn so that we will get an email even on success.
 287 warnmsg "$computer: Backup completed successfully."
 288
 289 # Remove lockfile
 290 rm $LOCKFILE