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 >&2; 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         # The file is usually stored to disk with a simple cat > $TARFILE,
 111         # but in some cases, like backing up to a remote and untrusted
 112         # file system, it could be useful to replace it with e.g.
 113         # a GPG command line.
 114         if [ -f $confdir/storageprogram.$computer ]; then
 115                 storageprogram=`cat $confdir/storageprogram.$computer`
 116         else
 117                 storageprogram=cat
 118         fi
 119
 120         # We try to run tar on the remote computer
 121         #    c create archive
 122         #    C change to directory first
 123         #    . where to start taring (see C)
 124         #    $lastcmd only files newer than this
 125         #    --one-file-system don't traverse file systems
 126         #    --use-compress-program compress it using gzip or pigz
 127         #    --exclude-from file to get exclusion pattern from
 128         #    Pipe the stuff over ssh to ourselves, run pee to cat the contents to a
 129         #    file and build a file list, respectivly.
 130         TARFILE=$DATE.tmp
 131         TARCMD="ssh -n $username@$computer \"nice -n 19 ionice -c3 tar --one-file-system --use-compress-program $compressor -cf - -C $filesystem . $lastcmd \
 132                 --exclude-from=$remotehome/.backup/exclude\" | pee \"$storageprogram > $TARFILE\" \"tar tzvf -\""
 133         infomsg "Running $TARCMD"
 134         eval $TARCMD > $DATE.idx
 135
 136         if [ $? -eq 0 ] && [ -s $TARFILE ]; then
 137                 # File is >0 in size and neither cat or tar tzvf failed; we assume it worked.
 138
 139                 if [ "$remotesnar" ]; then
 140                         if ! scp $username@$computer:$remotesnar $localsnar.tmp >&2; then
 141                                 diemsg "Could not copy .incremental.snar from $computer"
 142                         fi
 143                 fi
 144
 145                 # Move tar file in place
 146                 mv $TARFILE $DATE.tgz
 147                 if [ "$remotesnar" ]; then
 148                         mv $localsnar.tmp $localsnar
 149                 fi
 150
 151                 # Update timestamp
 152                 echo $DATEs > ../.date && touch -t $DATE ../.date
 153
 154                 # Fix permissions
 155                 chmod 600 *tgz                      #only for us
 156                 chmod 644 *idx 2>/dev/null          #everyone can read
 157
 158                 # Let the remote computer know that we ran a successful backup (for nagios)
 159                 ssh -n $username@$computer touch $filesystem/.lastbackup
 160         else
 161                 # Something wrong happened.
 162                 rm $TARFILE
 163                 diemsg "tar failed or $TARFILE empty. $backuplevel backup of $computer:$filesystem failed and deleted"
 164         fi
 165 }
 166
 167 # Non-privileged backups currently have a few limitations compared to regular ones:
 168 #
 169 #  - No .idx files.
 170 #  - No .snar-style incremental backups, only date-based.
 171 #  - No touching of .lastbackup for Nagios.
 172 #  - Failed tar operations may go undetected, since we only see the error code
 173 #    the last process in the pipe chain, which is gpg.
 174 #  - No storageprogram support (see above), because the file is typically already
 175 #    encrypted.
 176 nonprivbackup()
 177 {
 178         infomsg "$computer:$filesystem $backuplevel non-privileged backup"
 179
 180         if [ "$backuplevel" = "daglig" ]; then
 181                 lastd=`cat ../.date`
 182                 lastcmd="'$lastd'"
 183         else
 184                 lastcmd=""
 185         fi
 186
 187         TARFILE=$DATE.tmp
 188         TARCMD="ssh -n $username@$computer \"sudo /usr/local/sbin/output-encrypted-backup $filesystem $lastcmd\""
 189         infomsg "Running $TARCMD"
 190         eval $TARCMD > $TARFILE
 191
 192         if [ $? -eq 0 ] && [ -s $TARFILE ]; then
 193                 # File is >0 in size and the remote side did not report any fatal errors; we assume it worked.
 194
 195                 # Move tar file in place
 196                 mv $TARFILE $DATE.tgz.gpg
 197
 198                 # Update timestamp
 199                 echo $DATEs > ../.date && touch -t $DATE ../.date
 200
 201                 # Fix permissions
 202                 chmod 600 *tgz.gpg                  #only for us
 203         else
 204                 # Something wrong happened.
 205                 rm $TARFILE
 206                 diemsg "tar failed or $TARFILE empty. $backuplevel backup of $computer:$filesystem failed and deleted"
 207         fi
 208 }
 209
 210 # Check that the target filesystem is mounted (actually check that it's not
 211 # the root filesystem)
 212 if [ ! -d "$storagedir/$computer" ]; then
 213         diemsg "Target filesystem ($storagedir/$computer) does not exist."
 214 fi
 215 rootfilesystem=`df -P /`
 216 targetfilesystem=`df -P "$storagedir/$computer"`
 217 if [ "$rootfilesystem" == "$targetfilesystem" ]; then
 218         diemsg "Target filesystem ($storagedir/$computer) was mounted on /."
 219 fi
 220
 221 # Find the user to do the backup as.
 222 if [ -f $confdir/username.$computer ] ; then
 223         username=`cat $confdir/username.$computer`
 224 else
 225         username=root
 226 fi
 227
 228 infomsg "Backing up $computer"
 229
 230 # Try to SSH to the computer without entering a password.
 231 if ! ssh -n -o NumberOfPasswordPrompts=0 $username@$computer /bin/true; then
 232         diemsg "Could not use passwordless SSH."
 233 fi
 234
 235 # Find the home directory of the backup user
 236 remotehome=`ssh -n $username@$computer "echo ~"`
 237 if [ -z "$remotehome" ]; then
 238         diemsg "Could not expand ~ for user $username"
 239 fi
 240
 241 # Check if pigz is available
 242 if ssh -n $username@$computer "pigz -V 2>/dev/null"; then
 243         compressor=pigz
 244 else
 245         infomsg "pigz missing; falling back to gzip."
 246         compressor=gzip
 247 fi
 248
 249 # Check dump bit in fstab
 250 filesystems=`ssh -n $username@$computer "cat /etc/fstab" \
 251         | grep -v nfs \
 252         | grep -v "^#" \
 253         | grep -v "^$" \
 254         | awk '{ if ( $(NF-1) != "0" ) print $2}' `
 255
 256 # Clean up our dir at this client
 257 if ! ssh -n $username@$computer "rm -r $remotehome/.backup ; mkdir -m 700 $remotehome/.backup"; then
 258         diemsg "Could not create backup staging area at $computer:$remotehome/.backup"
 259 fi
 260
 261 # Try to copy $exclude to $computer
 262 if ! scp $exclude $username@$computer:$remotehome/.backup/exclude >&2; then
 263         diemsg "Could not copy exclude.txt to $computer"
 264 fi
 265
 266 # Try to copy preeexec and postexec if they exist
 267 if [ -f $confdir/preexec.$computer ]; then
 268         if ! scp $confdir/preexec.$computer $username@$computer:$remotehome/.backup/preexec >&2; then
 269                 diemsg "Could not copy preexec.$computer to $computer:$remotehome/.backup/preexec"
 270         fi
 271 fi
 272 if [ -f $confdir/postexec.$computer ]; then
 273         if ! scp $confdir/postexec.$computer $username@$computer:$remotehome/.backup/postexec >&2; then
 274                 diemsg "Could not copy postexec.$computer to $computer:$remotehome/.backup/postexec"
 275         fi
 276 fi
 277
 278 # Try to run preexec if it exists
 279 if ! ssh -n $username@$computer "[ ! -f $remotehome/.backup/preexec ] || /bin/bash -x $remotehome/.backup/preexec" >&2; then
 280         diemsg "Could not run $computer:$remotehome/.backup/preexec"
 281 fi
 282
 283 for filesystem in $filesystems; do
 284         # Remove / and $ (for Windows) for path names
 285         sfilesystem=`echo $filesystem | tr '\/\$' '__'`
 286
 287         # Prepare storage area
 288         mkdir -m 755 -p $storagedir/$computer/$sfilesystem/full 2>/dev/null
 289         mkdir -m 755 -p $storagedir/$computer/$sfilesystem/daglig 2>/dev/null
 290
 291         # Default backuplevel
 292         backuplevel=daglig
 293
 294         if [ ! -s $storagedir/$computer/$sfilesystem/.date ]; then
 295                 # Take the first full backup of this filesystem on this computer
 296                 backuplevel=full
 297         fi
 298
 299         # Check if we want a full backup
 300         if [ -f $confdir/fastfullbackupdag.$computer.$sfilesystem ]; then
 301                 fullbackup_min_for_this_machine=$daysbetweenfullforfixed
 302                 if [ "$DAY_OF_MONTH" = "`cat $confdir/fastfullbackupdag.$computer.$sfilesystem`" ]; then
 303                         backuplevel=full
 304                 fi
 305         elif [ -f $confdir/fastfullbackupdag.$computer ]; then
 306                 fullbackup_min_for_this_machine=$daysbetweenfullforfixed
 307                 if [ "$DAY_OF_MONTH" = "`cat $confdir/fastfullbackupdag.$computer`" ]; then
 308                         backuplevel=full
 309                 fi
 310         else
 311                 fullbackup_min_for_this_machine=$daysbetweenfull
 312         fi
 313
 314         if [ -z "`find $storagedir/$computer/$sfilesystem/full/ -name \*tgz\* -mtime -$fullbackup_min_for_this_machine`" ]; then
 315                 backuplevel=full
 316         fi
 317
 318         # We want to be here
 319         cd $storagedir/$computer/$sfilesystem/$backuplevel || diemsg "$storagedir/$computer/$sfilesystem/$backuplevel does not exist"
 320
 321         # Perform the actual backup
 322         if [ "$username" = "root" ]; then
 323                 backup
 324         else
 325                 nonprivbackup
 326         fi
 327
 328         # Check if this box has a custom number of full backups
 329         if [ -f $confdir/maksfulle.$computer.$sfilesystem ]; then
 330                 mf=$((`cat $confdir/maksfulle.$computer`+1))
 331         elif [ -f $confdir/maksfulle.$computer ] ; then
 332                 mf=$((`cat $confdir/maksfulle.$computer`+1))
 333         else
 334                 mf=$(($maxnumfull+1))
 335         fi
 336
 337         # Delete old full backups
 338         for full in `ls -1t $storagedir/$computer/$sfilesystem/full/*tgz* | tail -n +$mf`; do
 339                 prefix=`echo $full | sed "s/\.[^.]*$//"`
 340                 infomsg "$computer:$filesystem Deleting full backup $prefix"
 341                 rm $prefix*
 342         done
 343
 344         # Delete incremental backups older than the oldest full backup
 345         oldf=`ls -t1 $storagedir/$computer/$sfilesystem/full/*tgz* | tail -1`
 346         find \
 347                 $storagedir/$computer/$sfilesystem/daglig \
 348                 -type f \
 349                 \! -newer $oldf \
 350                 -printf "`date`: $computer: Deleting old incremental backup: %p\n" \
 351                 -exec rm {} \; >&2
 352 done
 353
 354 # Try to run postexec if it exist
 355 if ! ssh -n $username@$computer "[ ! -f $remotehome/.backup/postexec ] || /bin/bash -x $remotehome/.backup/postexec" >&2; then
 356         diemsg "Could not run $computer:$remotehome/.backup/postexec"
 357 fi
 358
 359 # Use warn so that we will get an email even on success.
 360 warnmsg "Backup completed successfully."
 361
 362 # Remove lockfile
 363 rm $LOCKFILE