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="$remotehome/.backup/${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 $username@$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 $username@$computer \"nice -n 19 ionice -c3 tar --one-file-system --use-compress-program $compressor -cf - -C $filesystem . $lastcmd \
 116                 --exclude-from=$remotehome/.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 $username@$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                 # Fix permissions
 139                 chmod 600 *tgz                      #only for us
 140                 chmod 644 *idx 2>/dev/null          #everyone can read
 141
 142                 # Let the remote computer know that we ran a successful backup (for nagios)
 143                 ssh $username@$computer touch $filesystem/.lastbackup
 144         else
 145                 # Something wrong happened.
 146                 rm $TARFILE
 147                 diemsg "tar failed or $TARFILE empty. $backuplevel backup of $computer:$filesystem failed and deleted"
 148         fi
 149 }
 150
 151 # Non-privileged backups currently have a few limitations compared to regular ones:
 152 #
 153 #  - No .idx files.
 154 #  - No .snar-style incremental backups, only date-based.
 155 #  - No touching of .lastbackup for Nagios.
 156 #  - Failed tar operations may go undetected, since we only see the error code
 157 #    the last process in the pipe chain, which is gpg.
 158 nonprivbackup()
 159 {
 160         infomsg "$computer:$filesystem $backuplevel non-privileged backup"
 161
 162         if [ "$backuplevel" = "daglig" ]; then
 163                 lastd=`cat ../.date`
 164                 lastcmd="'$lastd'"
 165         else
 166                 lastcmd=""
 167         fi
 168
 169         TARFILE=$DATE.tmp
 170         TARCMD="ssh $username@$computer \"sudo /usr/local/sbin/output-encrypted-backup $filesystem $lastcmd\""
 171         infomsg "Running $TARCMD"
 172         eval $TARCMD > $TARFILE
 173
 174         if [ $? -eq 0 ] && [ -s $TARFILE ]; then
 175                 # File is >0 in size and the remote side did not report any fatal errors; we assume it worked.
 176
 177                 # Move tar file in place
 178                 mv $TARFILE $DATE.tgz.gpg
 179
 180                 # Update timestamp
 181                 echo $DATEs > ../.date && touch -t $DATE ../.date
 182
 183                 # Fix permissions
 184                 chmod 600 *tgz.gpg                  #only for us
 185         else
 186                 # Something wrong happened.
 187                 rm $TARFILE
 188                 diemsg "tar failed or $TARFILE empty. $backuplevel backup of $computer:$filesystem failed and deleted"
 189         fi
 190 }
 191
 192 # Check that the target filesystem is mounted (actually check that it's not
 193 # the root filesystem)
 194 if [ ! -d "$storagedir/$computer" ]; then
 195         diemsg "Target filesystem ($storagedir/$computer) does not exist."
 196 fi
 197 rootfilesystem=`df -P /`
 198 targetfilesystem=`df -P "$storagedir/$computer"`
 199 if [ "$rootfilesystem" == "$targetfilesystem" ]; then
 200         diemsg "Target filesystem ($storagedir/$computer) was mounted on /."
 201 fi
 202
 203 # Find the user to do the backup as.
 204 if [ -f $confdir/username.$computer ] ; then
 205         username=`cat $confdir/username.$computer`
 206 else
 207         username=root
 208 fi
 209
 210 infomsg "Backing up $computer"
 211
 212 # Try to SSH to the computer without entering a password.
 213 if ! ssh -n -o NumberOfPasswordPrompts=0 $username@$computer /bin/true; then
 214         diemsg "Could not use passwordless SSH."
 215 fi
 216
 217 # Find the home directory of the backup user
 218 remotehome=`ssh $username@$computer "echo ~"`
 219 if [ -z "$remotehome" ]; then
 220         diemsg "Could not expand ~ for user $username"
 221 fi
 222
 223 # Check if pigz is available
 224 if ssh -n $username@$computer "pigz -V 2>/dev/null"; then
 225         compressor=pigz
 226 else
 227         infomsg "pigz missing; falling back to gzip."
 228         compressor=gzip
 229 fi
 230
 231 # Check dump bit in fstab
 232 filesystems=`ssh -n $username@$computer "cat /etc/fstab" \
 233         | grep -v nfs \
 234         | grep -v "^#" \
 235         | grep -v "^$" \
 236         | awk '{ if ( $(NF-1) != "0" ) print $2}' `
 237
 238 # Clean up our dir at this client
 239 if ! ssh $username@$computer "rm -r $remotehome/.backup ; mkdir -m 700 $remotehome/.backup"; then
 240         diemsg "Could not create backup staging area at $computer:$remotehome/.backup"
 241 fi
 242
 243 # Try to copy $exclude to $computer
 244 if ! scp $exclude $username@$computer:$remotehome/.backup/exclude > /dev/null; then
 245         diemsg "Could not copy exclude.txt to $computer"
 246 fi
 247
 248 # Try to copy preeexec and postexec if they exist
 249 if [ -f $confdir/preexec.$computer ]; then
 250         if ! scp $confdir/preexec.$computer $username@$computer:$remotehome/.backup/preexec >&2; then
 251                 diemsg "Could not copy preexec.$computer to $computer:$remotehome/.backup/preexec"
 252         fi
 253 fi
 254 if [ -f $confdir/postexec.$computer ]; then
 255         if ! scp $confdir/postexec.$computer $username@$computer:$remotehome/.backup/postexec >&2; then
 256                 diemsg "Could not copy postexec.$computer to $computer:$remotehome/.backup/postexec"
 257         fi
 258 fi
 259
 260 # Try to run preexec if it exists
 261 if ! ssh $username@$computer "[ ! -f $remotehome/.backup/preexec ] || /bin/bash -x $remotehome/.backup/preexec" >&2; then
 262         diemsg "Could not run $computer:$remotehome/.backup/preexec"
 263 fi
 264
 265 for filesystem in $filesystems; do
 266         # Remove / and $ (for Windows) for path names
 267         sfilesystem=`echo $filesystem | tr '\/\$' '__'`
 268
 269         # Prepare storage area
 270         mkdir -m 755 -p $storagedir/$computer/$sfilesystem/full 2>/dev/null
 271         mkdir -m 755 -p $storagedir/$computer/$sfilesystem/daglig 2>/dev/null
 272
 273         # Default backuplevel
 274         backuplevel=daglig
 275
 276         if [ ! -s $storagedir/$computer/$sfilesystem/.date ]; then
 277                 # Take the first full backup of this filesystem on this computer
 278                 backuplevel=full
 279         fi
 280
 281         # Check if we want a full backup
 282         if [ -f $confdir/fastfullbackupdag.$computer.$sfilesystem ]; then
 283                 fullbackup_min_for_this_machine=$daysbetweenfullforfixed
 284                 if [ "$DAY_OF_MONTH" = "`cat $confdir/fastfullbackupdag.$computer.$sfilesystem`" ]; then
 285                         backuplevel=full
 286                 fi
 287         elif [ -f $confdir/fastfullbackupdag.$computer ]; then
 288                 fullbackup_min_for_this_machine=$daysbetweenfullforfixed
 289                 if [ "$DAY_OF_MONTH" = "`cat $confdir/fastfullbackupdag.$computer`" ]; then
 290                         backuplevel=full
 291                 fi
 292         else
 293                 fullbackup_min_for_this_machine=$daysbetweenfull
 294         fi
 295
 296         if [ -z "`find $storagedir/$computer/$sfilesystem/full/ -name \*tgz\* -mtime -$fullbackup_min_for_this_machine`" ]; then
 297                 backuplevel=full
 298         fi
 299
 300         # We want to be here
 301         cd $storagedir/$computer/$sfilesystem/$backuplevel || diemsg "$storagedir/$computer/$sfilesystem/$backuplevel does not exist"
 302
 303         # Perform the actual backup
 304         if [ "$username" = "root" ]; then
 305                 backup
 306         else
 307                 nonprivbackup
 308         fi
 309
 310         # Check if this box has a custom number of full backups
 311         if [ -f $confdir/maksfulle.$computer.$sfilesystem ]; then
 312                 mf=$((`cat $confdir/maksfulle.$computer`+1))
 313         elif [ -f $confdir/maksfulle.$computer ] ; then
 314                 mf=$((`cat $confdir/maksfulle.$computer`+1))
 315         else
 316                 mf=$(($maxnumfull+1))
 317         fi
 318
 319         # Delete old full backups
 320         for full in `ls -1t $storagedir/$computer/$sfilesystem/full/*tgz* | tail -n +$mf`; do
 321                 prefix=`echo $full | sed "s/\.[^.]*$//"`
 322                 infomsg "$computer:$filesystem Deleting full backup $prefix"
 323                 rm $prefix*
 324         done
 325
 326         # Delete incremental backups older than the oldest full backup
 327         oldf=`ls -t1 $storagedir/$computer/$sfilesystem/full/*tgz* | tail -1`
 328         find \
 329                 $storagedir/$computer/$sfilesystem/daglig \
 330                 -type f \
 331                 \! -newer $oldf \
 332                 -printf "`date`: $computer: Deleting old incremental backup: %p\n" \
 333                 -exec rm {} \; >&2
 334 done
 335
 336 # Try to run postexec if it exist
 337 if ! ssh $username@$computer "[ ! -f $remotehome/.backup/postexec ] || /bin/bash -x $remotehome/.backup/postexec" >&2; then
 338         diemsg "Could not run $computer:$remotehome/.backup/postexec"
 339 fi
 340
 341 # Use warn so that we will get an email even on success.
 342 warnmsg "Backup completed successfully."
 343
 344 # Remove lockfile
 345 rm $LOCKFILE