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