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" = "daglig" ]; 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         # 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 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 # Get basic configuration.
 211 storagedir=$( findconf storagedir $computer )
 212 username=$( findconf username $computer )
 213 exclude=$( findconf excludefile $computer )
 214
 215 # Check that the target filesystem is mounted (actually check that it's not
 216 # the root filesystem)
 217 if [ ! -d "$storagedir/$computer" ]; then
 218         diemsg "Target filesystem ($storagedir/$computer) does not exist."
 219 fi
 220 rootfilesystem=`df -P /`
 221 targetfilesystem=`df -P "$storagedir/$computer"`
 222 if [ "$rootfilesystem" == "$targetfilesystem" ]; then
 223         diemsg "Target filesystem ($storagedir/$computer) was mounted on /."
 224 fi
 225
 226 infomsg "Backing up $computer"
 227
 228 # Try to SSH to the computer without entering a password.
 229 if ! ssh -n -o NumberOfPasswordPrompts=0 $username@$computer /bin/true; then
 230         diemsg "Could not use passwordless SSH."
 231 fi
 232
 233 # Find the home directory of the backup user
 234 remotehome=`ssh -n $username@$computer "echo ~"`
 235 if [ -z "$remotehome" ]; then
 236         diemsg "Could not expand ~ for user $username"
 237 fi
 238
 239 # Check if pigz is available
 240 if ssh -n $username@$computer "pigz -V 2>/dev/null"; then
 241         compressor=pigz
 242 else
 243         infomsg "pigz missing; falling back to gzip."
 244         compressor=gzip
 245 fi
 246
 247 # Check if nice and ionice are available
 248 if ssh -n $username@$computer "nice -n 19 ionice -c3 /bin/true 2>/dev/null"; then
 249         nice="nice -n 19 ionice -c3"
 250 else
 251         infomsg "nice and/or ionice missing; will run at normal priority."
 252         nice=""
 253 fi
 254
 255 # Check dump bit in fstab
 256 filesystems=`ssh -n $username@$computer "cat /etc/fstab" \
 257         | grep -v nfs \
 258         | grep -v "^#" \
 259         | grep -v "^$" \
 260         | awk '{ if ( $(NF-1) != "0" ) print $2}' `
 261
 262 # Clean up our dir at this client
 263 if ! ssh -n $username@$computer "rm -r $remotehome/.backup ; mkdir -m 700 $remotehome/.backup"; then
 264         diemsg "Could not create backup staging area at $computer:$remotehome/.backup"
 265 fi
 266
 267 # Try to copy $exclude to $computer
 268 if ! scp $exclude $username@$computer:$remotehome/.backup/exclude >&2; then
 269         diemsg "Could not copy exclude.txt to $computer"
 270 fi
 271
 272 # Try to copy preeexec and postexec if they exist
 273 if [ -f $confdir/preexec.$computer ]; then
 274         if ! scp $confdir/preexec.$computer $username@$computer:$remotehome/.backup/preexec >&2; then
 275                 diemsg "Could not copy preexec.$computer to $computer:$remotehome/.backup/preexec"
 276         fi
 277 fi
 278 if [ -f $confdir/postexec.$computer ]; then
 279         if ! scp $confdir/postexec.$computer $username@$computer:$remotehome/.backup/postexec >&2; then
 280                 diemsg "Could not copy postexec.$computer to $computer:$remotehome/.backup/postexec"
 281         fi
 282 fi
 283
 284 # Try to run preexec if it exists
 285 if ! ssh -n $username@$computer "[ ! -f $remotehome/.backup/preexec ] || /bin/bash -x $remotehome/.backup/preexec" >&2; then
 286         diemsg "Could not run $computer:$remotehome/.backup/preexec"
 287 fi
 288
 289 for filesystem in $filesystems; do
 290         # Remove / and $ (for Windows) for path names
 291         sfilesystem=`echo $filesystem | tr '\/\$' '__'`
 292
 293         # Prepare storage area
 294         mkdir -m 755 -p $storagedir/$computer/$sfilesystem/full 2>/dev/null
 295         mkdir -m 755 -p $storagedir/$computer/$sfilesystem/daglig 2>/dev/null
 296
 297         # Default backuplevel
 298         backuplevel=daglig
 299
 300         if [ ! -s $storagedir/$computer/$sfilesystem/.date ]; then
 301                 # Take the first full backup of this filesystem on this computer
 302                 backuplevel=full
 303         fi
 304
 305         # Check if we want a full backup
 306         fixedfullbackupday=$( findconf fixedfullbackupday $computer $filesystem )
 307         if [ "$fixedfullbackupday" ]; then
 308                 fullbackup_min_for_this_machine=$( findconf daysbetweenfullforfixed $computer $filesystem )
 309                 if [ "$DAY_OF_MONTH" = "$fixedfullbackupday" ]; then
 310                         backuplevel=full
 311                 fi
 312         else
 313                 fullbackup_min_for_this_machine=$( findconf daysbetweenfull $computer $filesystem )
 314         fi
 315
 316         if [ -z "`find $storagedir/$computer/$sfilesystem/full/ -name \*tgz\* -mtime -$fullbackup_min_for_this_machine`" ]; then
 317                 backuplevel=full
 318         fi
 319
 320         # We want to be here
 321         cd $storagedir/$computer/$sfilesystem/$backuplevel || diemsg "$storagedir/$computer/$sfilesystem/$backuplevel does not exist"
 322
 323         # Perform the actual backup
 324         if [ "$( findconf nonpriv $computer )" ]; then
 325                 nonprivbackup
 326         else
 327                 backup
 328         fi
 329
 330         # Delete old full backups
 331         mf=$( findconf maxnumfull $computer $filesystem )
 332         mf=$(( mf + 1 ))
 333         for full in `ls -1t $storagedir/$computer/$sfilesystem/full/*tgz* | tail -n +$mf`; do
 334                 prefix=`echo $full | sed "s/\.[^.]*$//"`
 335                 infomsg "$computer:$filesystem Deleting full backup $prefix"
 336                 rm $prefix*
 337         done
 338
 339         # Delete incremental backups older than the oldest full backup
 340         oldf=`ls -t1 $storagedir/$computer/$sfilesystem/full/*tgz* | tail -1`
 341         find \
 342                 $storagedir/$computer/$sfilesystem/daglig \
 343                 -type f \
 344                 \! -newer $oldf \
 345                 -printf "`date`: $computer: Deleting old incremental backup: %p\n" \
 346                 -exec rm {} \; >&2
 347 done
 348
 349 # Try to run postexec if it exist
 350 if ! ssh -n $username@$computer "[ ! -f $remotehome/.backup/postexec ] || /bin/bash -x $remotehome/.backup/postexec" >&2; then
 351         diemsg "Could not run $computer:$remotehome/.backup/postexec"
 352 fi
 353
 354 # Use warn so that we will get an email even on success.
 355 warnmsg "Backup completed successfully."
 356
 357 # Remove lockfile
 358 rm $LOCKFILE