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 SCRIPTPATH="$( cd "$(dirname "$0")" ; pwd -P )"  # Absolute directory the script lives in
  22 confdir=$SCRIPTPATH/conf/                        # Directory configuration files
  23 [ -r $confdir/env ] && . $confdir/env
  24
  25 # Default configuration; used unless $confdir/env has set something different.
  26 storagedir=${storagedir:-/backup}                # Where we keep backups
  27 exclude=${exclude:-$confdir/exclude}             # List of exclude patterns
  28 maxnumfull=${maxnumfull:-3}                      # Number of full backups
  29 daysbetweenfull=${daysbetweenfull:-30}           # Days between full backups
  30
  31 # Days between full backups for machines with fixed full-backup date
  32 # (in case the fixed day gets missed).
  33 daysbetweenfullforfixed=${daysbetweenfullforfixed:-32}
  34
  35 # End of configuration.
  36
  37 computer=$1                                      # The computer to backup.
  38 LOCKFILE=$SCRIPTPATH/backuprun.lockfile.$computer
  39 DAY_OF_MONTH=`date "+%d" | sed s/^0//`           # from 1 to 31
  40 DATE=`date "+%Y%m%d%H%M"`                        # format: touch
  41 DATEs=`date "+%Y-%m-%d %H:%M"`                   # format: tar
  42
  43 if [ -z "$computer" ]; then
  44         echo "Usage: $0 COMPUTER"
  45         exit 1
  46 fi
  47
  48 [ ! -f $exclude ] && printf "tmp\ncore\n" > $exclude
  49
  50 # Die more or less gracefully (at least notify user)
  51 die() {
  52         warnmsg "Aborting."
  53         rm $LOCKFILE
  54         exit 1
  55 }
  56 diemsg() {
  57         warnmsg "$@"
  58         die
  59 }
  60 warnmsg() {
  61         echo `date`": $computer: $@"
  62         echo `date`": $computer: $@" >&2
  63 }
  64 infomsg() {
  65         echo `date`": $computer: $@" >&2
  66 }
  67
  68 # Trap C-c and kill
  69 trap die SIGINT SIGTERM
  70
  71 # Don't start if we're already running. Don't use diemsg, as that removes the lock file!
  72 if [ -e $LOCKFILE ]; then
  73         warnmsg "$LOCKFILE exists for pid `cat $LOCKFILE`."
  74         exit 1
  75 fi;
  76 touch $LOCKFILE
  77 echo $$ > $LOCKFILE
  78
  79 infomsg "Backup run starting"
  80
  81 umask 027
  82
  83 backup()
  84 {
  85         infomsg "$computer:$filesystem $backuplevel backup"
  86
  87         localsnar="${storagedir}/${computer}/${sfilesystem}/.incremental.snar"
  88         remotesnar="$remotehome/.backup/${sfilesystem}.snar"
  89         if [ "$backuplevel" = "daglig" ]; then
  90                 # If incremental backup, we need to copy the incremental status to $computer.
  91                 # If it does not exist, whine a bit and then run date-based instead.
  92                 if [ -s "$localsnar" ]; then
  93                         if ! scp $localsnar $username@$computer:$remotesnar; then
  94                                 diemsg "Could not copy .incremental.snar to $computer"
  95                         fi
  96                         lastcmd="--listed-incremental=$remotesnar --no-check-device"
  97                 else
  98                         infomsg "Missing incremental.snar for $filesystem, doing date-based backup instead until next full backup"
  99                         remotesnar=""
 100                         lastd=`cat ../.date`
 101                         lastcmd="--newer='$lastd'"
 102                 fi
 103         else
 104                 # Note: Since we rm-ed .backup previously, and we don't scp over
 105                 # anything in this branch, this file will not exist, and we
 106                 # will be doing a full backup (but we want the output of it).
 107                 lastcmd="--listed-incremental=$remotesnar"
 108         fi
 109
 110         # We try to run tar on the remote computer
 111         #    c create archive
 112         #    C change to directory first
 113         #    . where to start taring (see C)
 114         #    $lastcmd only files newer than this
 115         #    --one-file-system don't traverse file systems
 116         #    --use-compress-program compress it using gzip or pigz
 117         #    --exclude-from file to get exclusion pattern from
 118         #    Pipe the stuff over ssh to ourselves, run pee to cat the contents to a
 119         #    file and build a file list, respectivly.
 120         TARFILE=$DATE.tmp
 121         TARCMD="ssh $username@$computer \"nice -n 19 ionice -c3 tar --one-file-system --use-compress-program $compressor -cf - -C $filesystem . $lastcmd \
 122                 --exclude-from=$remotehome/.backup/exclude\" | pee \"cat > $TARFILE\" \"tar tzvf -\""
 123         infomsg "Running $TARCMD"
 124         eval $TARCMD > $DATE.idx
 125
 126         if [ $? -eq 0 ] && [ -s $TARFILE ]; then
 127                 # File is >0 in size and neither cat or tar tzvf failed; we assume it worked.
 128
 129                 if [ "$remotesnar" ]; then
 130                         if ! scp $username@$computer:$remotesnar $localsnar.tmp; then
 131                                 diemsg "Could not copy .incremental.snar from $computer"
 132                         fi
 133                 fi
 134
 135                 # Move tar file in place
 136                 mv $TARFILE $DATE.tgz
 137                 if [ "$remotesnar" ]; then
 138                         mv $localsnar.tmp $localsnar
 139                 fi
 140
 141                 # Update timestamp
 142                 echo $DATEs > ../.date && touch -t $DATE ../.date
 143
 144                 # Fix permissions
 145                 chmod 600 *tgz                      #only for us
 146                 chmod 644 *idx 2>/dev/null          #everyone can read
 147
 148                 # Let the remote computer know that we ran a successful backup (for nagios)
 149                 ssh $username@$computer touch $filesystem/.lastbackup
 150         else
 151                 # Something wrong happened.
 152                 rm $TARFILE
 153                 diemsg "tar failed or $TARFILE empty. $backuplevel backup of $computer:$filesystem failed and deleted"
 154         fi
 155 }
 156
 157 # Non-privileged backups currently have a few limitations compared to regular ones:
 158 #
 159 #  - No .idx files.
 160 #  - No .snar-style incremental backups, only date-based.
 161 #  - No touching of .lastbackup for Nagios.
 162 #  - Failed tar operations may go undetected, since we only see the error code
 163 #    the last process in the pipe chain, which is gpg.
 164 nonprivbackup()
 165 {
 166         infomsg "$computer:$filesystem $backuplevel non-privileged backup"
 167
 168         if [ "$backuplevel" = "daglig" ]; then
 169                 lastd=`cat ../.date`
 170                 lastcmd="'$lastd'"
 171         else
 172                 lastcmd=""
 173         fi
 174
 175         TARFILE=$DATE.tmp
 176         TARCMD="ssh $username@$computer \"sudo /usr/local/sbin/output-encrypted-backup $filesystem $lastcmd\""
 177         infomsg "Running $TARCMD"
 178         eval $TARCMD > $TARFILE
 179
 180         if [ $? -eq 0 ] && [ -s $TARFILE ]; then
 181                 # File is >0 in size and the remote side did not report any fatal errors; we assume it worked.
 182
 183                 # Move tar file in place
 184                 mv $TARFILE $DATE.tgz.gpg
 185
 186                 # Update timestamp
 187                 echo $DATEs > ../.date && touch -t $DATE ../.date
 188
 189                 # Fix permissions
 190                 chmod 600 *tgz.gpg                  #only for us
 191         else
 192                 # Something wrong happened.
 193                 rm $TARFILE
 194                 diemsg "tar failed or $TARFILE empty. $backuplevel backup of $computer:$filesystem failed and deleted"
 195         fi
 196 }
 197
 198 # Check that the target filesystem is mounted (actually check that it's not
 199 # the root filesystem)
 200 if [ ! -d "$storagedir/$computer" ]; then
 201         diemsg "Target filesystem ($storagedir/$computer) does not exist."
 202 fi
 203 rootfilesystem=`df -P /`
 204 targetfilesystem=`df -P "$storagedir/$computer"`
 205 if [ "$rootfilesystem" == "$targetfilesystem" ]; then
 206         diemsg "Target filesystem ($storagedir/$computer) was mounted on /."
 207 fi
 208
 209 # Find the user to do the backup as.
 210 if [ -f $confdir/username.$computer ] ; then
 211         username=`cat $confdir/username.$computer`
 212 else
 213         username=root
 214 fi
 215
 216 infomsg "Backing up $computer"
 217
 218 # Try to SSH to the computer without entering a password.
 219 if ! ssh -n -o NumberOfPasswordPrompts=0 $username@$computer /bin/true; then
 220         diemsg "Could not use passwordless SSH."
 221 fi
 222
 223 # Find the home directory of the backup user
 224 remotehome=`ssh $username@$computer "echo ~"`
 225 if [ -z "$remotehome" ]; then
 226         diemsg "Could not expand ~ for user $username"
 227 fi
 228
 229 # Check if pigz is available
 230 if ssh -n $username@$computer "pigz -V 2>/dev/null"; then
 231         compressor=pigz
 232 else
 233         infomsg "pigz missing; falling back to gzip."
 234         compressor=gzip
 235 fi
 236
 237 # Check dump bit in fstab
 238 filesystems=`ssh -n $username@$computer "cat /etc/fstab" \
 239         | grep -v nfs \
 240         | grep -v "^#" \
 241         | grep -v "^$" \
 242         | awk '{ if ( $(NF-1) != "0" ) print $2}' `
 243
 244 # Clean up our dir at this client
 245 if ! ssh $username@$computer "rm -r $remotehome/.backup ; mkdir -m 700 $remotehome/.backup"; then
 246         diemsg "Could not create backup staging area at $computer:$remotehome/.backup"
 247 fi
 248
 249 # Try to copy $exclude to $computer
 250 if ! scp $exclude $username@$computer:$remotehome/.backup/exclude > /dev/null; then
 251         diemsg "Could not copy exclude.txt to $computer"
 252 fi
 253
 254 # Try to copy preeexec and postexec if they exist
 255 if [ -f $confdir/preexec.$computer ]; then
 256         if ! scp $confdir/preexec.$computer $username@$computer:$remotehome/.backup/preexec >&2; then
 257                 diemsg "Could not copy preexec.$computer to $computer:$remotehome/.backup/preexec"
 258         fi
 259 fi
 260 if [ -f $confdir/postexec.$computer ]; then
 261         if ! scp $confdir/postexec.$computer $username@$computer:$remotehome/.backup/postexec >&2; then
 262                 diemsg "Could not copy postexec.$computer to $computer:$remotehome/.backup/postexec"
 263         fi
 264 fi
 265
 266 # Try to run preexec if it exists
 267 if ! ssh $username@$computer "[ ! -f $remotehome/.backup/preexec ] || /bin/bash -x $remotehome/.backup/preexec" >&2; then
 268         diemsg "Could not run $computer:$remotehome/.backup/preexec"
 269 fi
 270
 271 for filesystem in $filesystems; do
 272         # Remove / and $ (for Windows) for path names
 273         sfilesystem=`echo $filesystem | tr '\/\$' '__'`
 274
 275         # Prepare storage area
 276         mkdir -m 755 -p $storagedir/$computer/$sfilesystem/full 2>/dev/null
 277         mkdir -m 755 -p $storagedir/$computer/$sfilesystem/daglig 2>/dev/null
 278
 279         # Default backuplevel
 280         backuplevel=daglig
 281
 282         if [ ! -s $storagedir/$computer/$sfilesystem/.date ]; then
 283                 # Take the first full backup of this filesystem on this computer
 284                 backuplevel=full
 285         fi
 286
 287         # Check if we want a full backup
 288         if [ -f $confdir/fastfullbackupdag.$computer.$sfilesystem ]; then
 289                 fullbackup_min_for_this_machine=$daysbetweenfullforfixed
 290                 if [ "$DAY_OF_MONTH" = "`cat $confdir/fastfullbackupdag.$computer.$sfilesystem`" ]; then
 291                         backuplevel=full
 292                 fi
 293         elif [ -f $confdir/fastfullbackupdag.$computer ]; then
 294                 fullbackup_min_for_this_machine=$daysbetweenfullforfixed
 295                 if [ "$DAY_OF_MONTH" = "`cat $confdir/fastfullbackupdag.$computer`" ]; then
 296                         backuplevel=full
 297                 fi
 298         else
 299                 fullbackup_min_for_this_machine=$daysbetweenfull
 300         fi
 301
 302         if [ -z "`find $storagedir/$computer/$sfilesystem/full/ -name \*tgz\* -mtime -$fullbackup_min_for_this_machine`" ]; then
 303                 backuplevel=full
 304         fi
 305
 306         # We want to be here
 307         cd $storagedir/$computer/$sfilesystem/$backuplevel || diemsg "$storagedir/$computer/$sfilesystem/$backuplevel does not exist"
 308
 309         # Perform the actual backup
 310         if [ "$username" = "root" ]; then
 311                 backup
 312         else
 313                 nonprivbackup
 314         fi
 315
 316         # Check if this box has a custom number of full backups
 317         if [ -f $confdir/maksfulle.$computer.$sfilesystem ]; then
 318                 mf=$((`cat $confdir/maksfulle.$computer`+1))
 319         elif [ -f $confdir/maksfulle.$computer ] ; then
 320                 mf=$((`cat $confdir/maksfulle.$computer`+1))
 321         else
 322                 mf=$(($maxnumfull+1))
 323         fi
 324
 325         # Delete old full backups
 326         for full in `ls -1t $storagedir/$computer/$sfilesystem/full/*tgz* | tail -n +$mf`; do
 327                 prefix=`echo $full | sed "s/\.[^.]*$//"`
 328                 infomsg "$computer:$filesystem Deleting full backup $prefix"
 329                 rm $prefix*
 330         done
 331
 332         # Delete incremental backups older than the oldest full backup
 333         oldf=`ls -t1 $storagedir/$computer/$sfilesystem/full/*tgz* | tail -1`
 334         find \
 335                 $storagedir/$computer/$sfilesystem/daglig \
 336                 -type f \
 337                 \! -newer $oldf \
 338                 -printf "`date`: $computer: Deleting old incremental backup: %p\n" \
 339                 -exec rm {} \; >&2
 340 done
 341
 342 # Try to run postexec if it exist
 343 if ! ssh $username@$computer "[ ! -f $remotehome/.backup/postexec ] || /bin/bash -x $remotehome/.backup/postexec" >&2; then
 344         diemsg "Could not run $computer:$remotehome/.backup/postexec"
 345 fi
 346
 347 # Use warn so that we will get an email even on success.
 348 warnmsg "Backup completed successfully."
 349
 350 # Remove lockfile
 351 rm $LOCKFILE