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 computer=$1                                      # The computer to backup.
  22 SCRIPTPATH="$( cd "$(dirname "$0")" ; pwd -P )"  # Absolute directory the script lives in
  23 confdir=$SCRIPTPATH/conf                         # Directory for configuration files
  24 conffile=$confdir/backup.config
  25 LOCKFILE=$SCRIPTPATH/backuprun.lockfile.$computer
  26 DAY_OF_MONTH=`date "+%d" | sed s/^0//`           # from 1 to 31
  27 DATE=`date "+%Y%m%d%H%M"`                        # format: touch
  28 DATEs=`date "+%Y-%m-%d %H:%M"`                   # format: tar
  29 TOP_PID=$$
  30
  31 if [ -z "$computer" ]; then
  32         echo "Usage: $0 COMPUTER"
  33         exit 1
  34 fi
  35
  36 [ ! -f $exclude ] && printf "tmp\ncore\n" > $exclude
  37
  38 # Die more or less gracefully (at least notify user)
  39 die() {
  40         warnmsg "Aborting."
  41         rm $LOCKFILE
  42         exit 1
  43 }
  44 diemsg() {
  45         warnmsg "$@"
  46         die
  47 }
  48 warnmsg() {
  49         echo `date`": $computer: $@"
  50         echo `date`": $computer: $@" >&2
  51 }
  52 infomsg() {
  53         echo `date`": $computer: $@" >&2
  54 }
  55 findconf() {  # getconf is already taken
  56         key=$1
  57         computer=$2
  58         filesystem=$( echo $3 | tr / . )
  59
  60         if [ "$computer" ] && [ "$filesystem" ] && git config -f $conffile $computer$filesystem.$key > /dev/null; then
  61                 git config -f $conffile $computer$filesystem.$key
  62         elif [ "$computer" ] && git config -f $conffile $computer.$key > /dev/null; then
  63                 git config -f $conffile $computer.$key
  64         elif git config -f $conffile defaults.$key > /dev/null; then
  65                 git config -f $conffile defaults.$key
  66         else
  67                 warnmsg "No configuration found for $key in $conffile, aborting."
  68                 kill -TERM $TOP_PID
  69         fi
  70 }
  71
  72 # Trap C-c and kill
  73 trap die SIGINT SIGTERM
  74
  75 # Don't start if we're already running. Don't use diemsg, as that removes the lock file!
  76 if [ -e $LOCKFILE ]; then
  77         warnmsg "$LOCKFILE exists for pid `cat $LOCKFILE`."
  78         exit 1
  79 fi;
  80 touch $LOCKFILE
  81 echo $$ > $LOCKFILE
  82
  83 infomsg "Backup run starting"
  84
  85 umask 027
  86
  87 backup()
  88 {
  89         infomsg "$computer:$filesystem $backuplevel backup"
  90
  91         localsnar="${storagedir}/${computer}/${sfilesystem}/.incremental.snar"
  92         remotesnar="$remotehome/.backup/${sfilesystem}.snar"
  93         if [ "$backuplevel" = "daily" ]; then
  94                 # If incremental backup, we need to copy the incremental status to $computer.
  95                 # If it does not exist, whine a bit and then run date-based instead.
  96                 if [ -s "$localsnar" ]; then
  97                         if ! scp $localsnar $username@$computer:$remotesnar >&2; then
  98                                 diemsg "Could not copy .incremental.snar to $computer"
  99                         fi
 100                         lastcmd="--listed-incremental=$remotesnar --no-check-device"
 101                 else
 102                         infomsg "Missing incremental.snar for $filesystem, doing date-based backup instead until next full backup"
 103                         remotesnar=""
 104                         lastd=`cat ../.date`
 105                         lastcmd="--newer='$lastd'"
 106                 fi
 107         else
 108                 # Note: Since we rm-ed .backup previously, and we don't scp over
 109                 # anything in this branch, this file will not exist, and we
 110                 # will be doing a full backup (but we want the output of it).
 111                 lastcmd="--listed-incremental=$remotesnar"
 112         fi
 113
 114         # The file is usually stored to disk with a simple cat > $TARFILE,
 115         # but in some cases, like backing up to a remote and untrusted
 116         # file system, it could be useful to replace it with e.g.
 117         # a GPG command line.
 118         storageprogram=$( findconf storageprogram $computer $sfilesystem )
 119
 120         # Find compressor and decompressor for this filesystem.
 121         fscompressor=$( findconf compressor $computer $filesystem )
 122         if [ -z "$fscompressor" ]; then
 123                 fscompressor=$compressor
 124         fi
 125         decompressor=$( findconf decompressor $computer $filesystem )
 126
 127         # We try to run tar on the remote computer
 128         #    c create archive
 129         #    C change to directory first
 130         #    . where to start taring (see C)
 131         #    $lastcmd only files newer than this
 132         #    --one-file-system don't traverse file systems
 133         #    --use-compress-program compress it using gzip or pigz
 134         #    --exclude-from file to get exclusion pattern from
 135         #    Pipe the stuff over ssh to ourselves, run pee to cat the contents to a
 136         #    file and build a file list, respectivly.
 137         TARFILE=$DATE.tmp
 138         TARCMD="ssh -n $username@$computer \"$nice tar --one-file-system --use-compress-program $fscompressor -cf - -C $filesystem . $lastcmd \
 139                 --exclude-from=$remotehome/.backup/exclude\" | pee \"$storageprogram > $TARFILE\" \"$decompressor | tar tvf -\""
 140         infomsg "Running $TARCMD"
 141         eval $TARCMD > $DATE.idx
 142
 143         if [ $? -eq 0 ] && [ -s $TARFILE ]; then
 144                 # File is >0 in size and neither cat or tar tzvf failed; we assume it worked.
 145
 146                 if [ "$remotesnar" ]; then
 147                         if ! scp $username@$computer:$remotesnar $localsnar.tmp >&2; then
 148                                 diemsg "Could not copy .incremental.snar from $computer"
 149                         fi
 150                 fi
 151
 152                 # Move tar file in place
 153                 mv $TARFILE $DATE.tgz
 154                 if [ "$remotesnar" ]; then
 155                         mv $localsnar.tmp $localsnar
 156                 fi
 157
 158                 # Update timestamp
 159                 echo $DATEs > ../.date && touch -t $DATE ../.date
 160
 161                 # Fix permissions
 162                 chmod 600 *tgz                      #only for us
 163                 chmod 644 *idx 2>/dev/null          #everyone can read
 164
 165                 # Let the remote computer know that we ran a successful backup (for nagios)
 166                 ssh -n $username@$computer touch $filesystem/.lastbackup
 167         else
 168                 # Something wrong happened.
 169                 rm $TARFILE
 170                 diemsg "tar failed or $TARFILE empty. $backuplevel backup of $computer:$filesystem failed and deleted"
 171         fi
 172 }
 173
 174 # Non-privileged backups currently have a few limitations compared to regular ones:
 175 #
 176 #  - No .idx files.
 177 #  - No .snar-style incremental backups, only date-based.
 178 #  - No touching of .lastbackup for Nagios.
 179 #  - Failed tar operations may go undetected, since we only see the error code
 180 #    the last process in the pipe chain, which is gpg.
 181 #  - No storageprogram support (see above), because the file is typically already
 182 #    encrypted.
 183 nonprivbackup()
 184 {
 185         infomsg "$computer:$filesystem $backuplevel non-privileged backup"
 186
 187         if [ "$backuplevel" = "daily" ]; then
 188                 lastd=`cat ../.date`
 189                 lastcmd="'$lastd'"
 190         else
 191                 lastcmd=""
 192         fi
 193
 194         TARFILE=$DATE.tmp
 195         TARCMD="ssh -n $username@$computer \"sudo /usr/local/sbin/output-encrypted-backup $filesystem $lastcmd\""
 196         infomsg "Running $TARCMD"
 197         eval $TARCMD > $TARFILE
 198
 199         if [ $? -eq 0 ] && [ -s $TARFILE ]; then
 200                 # File is >0 in size and the remote side did not report any fatal errors; we assume it worked.
 201
 202                 # Move tar file in place
 203                 mv $TARFILE $DATE.tgz.gpg
 204
 205                 # Update timestamp
 206                 echo $DATEs > ../.date && touch -t $DATE ../.date
 207
 208                 # Fix permissions
 209                 chmod 600 *tgz.gpg                  #only for us
 210         else
 211                 # Something wrong happened.
 212                 rm $TARFILE
 213                 diemsg "tar failed or $TARFILE empty. $backuplevel backup of $computer:$filesystem failed and deleted"
 214         fi
 215 }
 216
 217 # Get basic configuration.
 218 storagedir=$( findconf storagedir $computer )
 219 username=$( findconf username $computer )
 220 exclude=$( findconf excludefile $computer )
 221
 222 # Check that the target filesystem is mounted (actually check that it's not
 223 # the root filesystem)
 224 if [ ! -d "$storagedir/$computer" ]; then
 225         diemsg "Target filesystem ($storagedir/$computer) does not exist."
 226 fi
 227 rootfilesystem=`df -P /`
 228 targetfilesystem=`df -P "$storagedir/$computer"`
 229 if [ "$rootfilesystem" == "$targetfilesystem" ]; then
 230         diemsg "Target filesystem ($storagedir/$computer) was mounted on /."
 231 fi
 232
 233 infomsg "Backing up $computer"
 234
 235 # Try to SSH to the computer without entering a password.
 236 if ! ssh -n -o NumberOfPasswordPrompts=0 $username@$computer /bin/true; then
 237         diemsg "Could not use passwordless SSH."
 238 fi
 239
 240 # Find the home directory of the backup user
 241 remotehome=`ssh -n $username@$computer "echo ~"`
 242 if [ -z "$remotehome" ]; then
 243         diemsg "Could not expand ~ for user $username"
 244 fi
 245
 246 # Check if pigz is available. Note that this may be overridden per-filesystem later.
 247 compressor=$( findconf compressor $computer )
 248 if [ -z "$compressor" ]; then
 249         if ssh -n $username@$computer "pigz -V 2>/dev/null"; then
 250                 compressor=pigz
 251         else
 252                 infomsg "pigz missing; falling back to gzip."
 253                 compressor=gzip
 254         fi
 255 fi
 256
 257 # Check if nice and ionice are available
 258 if ssh -n $username@$computer "nice -n 19 ionice -c3 /bin/true 2>/dev/null"; then
 259         nice="nice -n 19 ionice -c3"
 260 else
 261         infomsg "nice and/or ionice missing; will run at normal priority."
 262         nice=""
 263 fi
 264
 265 # Check dump bit in fstab
 266 filesystems=`ssh -n $username@$computer "cat /etc/fstab" \
 267         | grep -v nfs \
 268         | grep -v "^#" \
 269         | grep -v "^$" \
 270         | awk '{ if ( $(NF-1) != "0" ) print $2}' `
 271
 272 # Clean up our dir at this client
 273 if ! ssh -n $username@$computer "rm -r $remotehome/.backup ; mkdir -m 700 $remotehome/.backup"; then
 274         diemsg "Could not create backup staging area at $computer:$remotehome/.backup"
 275 fi
 276
 277 # Try to copy $exclude to $computer
 278 if ! scp $exclude $username@$computer:$remotehome/.backup/exclude >&2; then
 279         diemsg "Could not copy exclude.txt to $computer"
 280 fi
 281
 282 # Try to copy preeexec and postexec if they exist
 283 if [ -f $confdir/preexec.$computer ]; then
 284         if ! scp $confdir/preexec.$computer $username@$computer:$remotehome/.backup/preexec >&2; then
 285                 diemsg "Could not copy preexec.$computer to $computer:$remotehome/.backup/preexec"
 286         fi
 287 fi
 288 if [ -f $confdir/postexec.$computer ]; then
 289         if ! scp $confdir/postexec.$computer $username@$computer:$remotehome/.backup/postexec >&2; then
 290                 diemsg "Could not copy postexec.$computer to $computer:$remotehome/.backup/postexec"
 291         fi
 292 fi
 293
 294 # Try to run preexec if it exists
 295 if ! ssh -n $username@$computer "[ ! -f $remotehome/.backup/preexec ] || /bin/bash -x $remotehome/.backup/preexec" >&2; then
 296         diemsg "Could not run $computer:$remotehome/.backup/preexec"
 297 fi
 298
 299 for filesystem in $filesystems; do
 300         # Remove / and $ (for Windows) for path names
 301         sfilesystem=`echo $filesystem | tr '\/\$' '__'`
 302
 303         # Prepare storage area
 304         mkdir -m 755 -p $storagedir/$computer/$sfilesystem/full 2>/dev/null
 305         mkdir -m 755 -p $storagedir/$computer/$sfilesystem/daily 2>/dev/null
 306
 307         # Default backuplevel
 308         backuplevel=daily
 309
 310         if [ ! -s $storagedir/$computer/$sfilesystem/.date ]; then
 311                 # Take the first full backup of this filesystem on this computer
 312                 backuplevel=full
 313         fi
 314
 315         # Check if we want a full backup
 316         fixedfullbackupday=$( findconf fixedfullbackupday $computer $filesystem )
 317         if [ "$fixedfullbackupday" ]; then
 318                 fullbackup_min_for_this_machine=$( findconf daysbetweenfullforfixed $computer $filesystem )
 319                 if [ "$DAY_OF_MONTH" = "$fixedfullbackupday" ]; then
 320                         backuplevel=full
 321                 fi
 322         else
 323                 fullbackup_min_for_this_machine=$( findconf daysbetweenfull $computer $filesystem )
 324         fi
 325
 326         if [ -z "`find $storagedir/$computer/$sfilesystem/full/ -name \*tgz\* -mtime -$fullbackup_min_for_this_machine`" ]; then
 327                 backuplevel=full
 328         fi
 329
 330         # We want to be here
 331         cd $storagedir/$computer/$sfilesystem/$backuplevel || diemsg "$storagedir/$computer/$sfilesystem/$backuplevel does not exist"
 332
 333         # Perform the actual backup
 334         if [ "$( findconf nonpriv $computer )" ]; then
 335                 nonprivbackup
 336         else
 337                 backup
 338         fi
 339
 340         # Delete old full backups
 341         mf=$( findconf maxnumfull $computer $filesystem )
 342         mf=$(( mf + 1 ))
 343         for full in `ls -1t $storagedir/$computer/$sfilesystem/full/*tgz* | tail -n +$mf`; do
 344                 prefix=`echo $full | sed "s/\.[^.]*$//"`
 345                 infomsg "$computer:$filesystem Deleting full backup $prefix"
 346                 rm $prefix*
 347         done
 348
 349         # Delete incremental backups older than the oldest full backup
 350         oldf=`ls -t1 $storagedir/$computer/$sfilesystem/full/*tgz* | tail -1`
 351         find \
 352                 $storagedir/$computer/$sfilesystem/daily \
 353                 -type f \
 354                 \! -newer $oldf \
 355                 -printf "`date`: $computer: Deleting old incremental backup: %p\n" \
 356                 -exec rm {} \; >&2
 357 done
 358
 359 # Try to run postexec if it exist
 360 if ! ssh -n $username@$computer "[ ! -f $remotehome/.backup/postexec ] || /bin/bash -x $remotehome/.backup/postexec" >&2; then
 361         diemsg "Could not run $computer:$remotehome/.backup/postexec"
 362 fi
 363
 364 # Use warn so that we will get an email even on success.
 365 warnmsg "Backup completed successfully."
 366
 367 # Remove lockfile
 368 rm $LOCKFILE