X-Git-Url: https://git.sesse.net/?p=backup.sh;a=blobdiff_plain;f=backup.sh;h=67a9e1ecc18f0b7c4bf5b6a949ec314a952ac01a;hp=5bbf1378b09c83cf736881e59926fd752b5d1f02;hb=5cc0592bad59b6f6d978b768924849be33a2145c;hpb=744c43e41f7db3dbf8cc9a2b8a6bee8b6fe84ee4

diff --git a/backup.sh b/backup.sh
index 5bbf137..67a9e1e 100755
--- a/backup.sh
+++ b/backup.sh
@@ -1,12 +1,58 @@
 #!/bin/sh
-
-# Locking
+# backup.sh
+# 
+# Backup up Unix-like computers
+# 
+# itk@samfundet.no
+#
+# This script was first checked into RCS in 2000, and has since backed up a few
+# servers almost every day, whenever not broken.
+# 
+# It's a bit peculiar, but quite effective. The following is worth keeping in
+# mind when hacking:
+#
+# KISS. And by that I mean really simple. Only regular Unix commands, please
+# (but GNU extensions are used).
+#
+# Because of the way we run this script from cron, informational logging goes
+# to stderr and warnings and errors to stdout. This way, we get mail whenever
+# stuff fails, and other output is redirected to the log file.
+# 
+# The computers are backed up in parallell (but the filesystems on each
+# computer in serial). This is implemented highly simplistic, with forking
+# subshells. The script has no concept about threads, and stuff might go wrong,
+# which usually requires manual cleanups.
+#
+# The script is halfway Norwegian and halfway English. Newer modifications are
+# in English, we should probably stick to that.
+# 
+# Known issues:
+#
+# o $R has no proper meaning because of the forking subshell stuff. We need a
+#   more robust way of returning non-zero on errors, which we probably don't
+#   right now.
+#
+
+# Configuration
 LOCKFILE=/home/backup/backuprun.lockfile
+confdir=/home/backup/conf/              # Configuration files
+storagedir=/backup                      # Where we keep backups
+maksantallfulle=3                       # Number of full backups
+dagermellomfulle=30                     # Days between full backups
+dagermellomfulleforfast=32              # Days between full backups for machines with fixed
+                                        # full-backup date (in case the fixed day gets missed)
+DAY_OF_MONTH=`date "+%d" | sed s/^0//`  # from 1 to 31
+DATE=`date "+%Y%m%d%H%M"`               #format: touch
+DATEs=`date "+%Y-%m-%d %H:%M"`          #format: tar
+
+# Exclude-pattern
+exclude=$confdir/exclude
+[ ! -f $exclude ] && printf "tmp\ncore\n" > $exclude
 
 # Initially, we consider everything to be just fine.
 R=0
 
-# Die gracefully (ie. remove lockfile)
+# Die more or less gracefully (at least notify user)
 die() {
 	echo `date`": Something nasty happened, and since I fork a lot:"
         echo `date`": I CANNOT CLEAN UP THE MESS MYSELF."
@@ -19,20 +65,12 @@ trap die SIGINT SIGTERM
 
 # Don't start if we're already running
 if [ -e $LOCKFILE ]; then
-	echo `date`": $LOCKFILE exists, exiting."
+	echo `date`": $LOCKFILE exists for pid `cat $LOCKFILE`, exiting."
+	echo `date`": $LOCKFILE exists for pid `cat $LOCKFILE`, exiting." >&2
 	exit 1
 fi;
 touch $LOCKFILE
-
-#dirs
-confdir=/home/backup/conf/           #configuration files
-storagedir=/backup                   #mountpoint of huge disc
-###
-
-#exclude-pattern
-exclude=$confdir/exclude
-[ ! -f $exclude ] && printf "tmp\ncore\n" > $exclude
-###
+echo $$ > $LOCKFILE
 
 #syntax of remotestat:
 #hostname:/directory/
@@ -43,11 +81,7 @@ remotestatf=$confdir/remotestat
 PATH=/local/bin:$PATH:/store/bin
 export PATH
 
-#start the logfile
-echo `date`": Backup run starting"
-echo "Exclude pattern:"
-cat $exclude
-echo "End exclude pattern"
+echo `date`": Backup run starting" >&2
 
 umask 027
 
@@ -61,16 +95,6 @@ if [ $1 ] ; then
   unixcomputers=$1
 fi;
 
-#disse bør kunne varieres fra fs til fs?
-maksantallfulle=3          #hvor mange fulle vi tar vare på
-dagermellomfulle=30        #antall dager før det er på tide med ny full
-logw=40
-###
-
-#lager datovariabeler
-DATE=`date "+%Y%m%d%H%M"`            #format: touch
-DATEs=`date "+%Y-%m-%d %H:%M"`       #format: tar
-###
 
 #selve backupen
 # krever at noen variabler er satt
@@ -78,7 +102,8 @@ DATEs=`date "+%Y-%m-%d %H:%M"`       #format: tar
 backup()
 {
 
- printf "%-${logw}s %s\n" "$computer:$filesystem" "$backuplevel backup"
+ echo -n `date` >&2
+ printf " $computer: $computer:$filesystem $backuplevel backup\n" >&2
 
  if [ "$backuplevel" = "daglig" ] || [ "$backuplevel" = "incremental" ]
  then
@@ -88,52 +113,45 @@ backup()
   lastcmd=""
  fi
 
- #if this client has a special tar
- #we need to find a better solution to this conf-issue
- if [ -f $confdir/tar.$computer ] ; then
-   tar=`cat $confdir/tar.$computer`
- else 
-   tar=tar
- fi
-
- #expand the exclude-path for use with tar
- exf=`ssh $computer "ls ~/.backup/exclude"`
+ # expand the exclude-path for use with tar
+ exf=`ssh root@$computer "ls ~/.backup/exclude"`
 
- #We try to run tar on the remote computer
+ # We try to run tar on the remote computer
  #    z gzip it
  #    c create archive
  #    C change to directory first
- #    - output to stdout (we pipe to gzip, then to dd)
  #    . where to start taring (see C)
  #    $lastcmd only files newer than this
+ #    --one-file-system don't traverse file systems
  #    --exclude-from file to get exclusion pattern from
- #    pipe to gzip, which in turn pipes over the ssh-stream
- #    ..to dd, to output to a file. We surpress messages from dd.
- #    And at last, redirect stderr to stdout, to get output logged.
+ #    Pipe the stuff over ssh to ourselves, run pee to cat the contents to a
+ #    file and build a file list, respectivly.  
  TARFILE=$DATE.tmp
- TARCMD="ssh $computer \"$tar --one-file-system -zcf - -C $filesystem . $lastcmd \
-	--exclude-from=$exf\" > $TARFILE"
- echo "cmdline: $TARCMD"
+ TARCMD="ssh root@$computer \"nice -n 19 ionice -c3 tar --one-file-system -zcf - -C $filesystem . $lastcmd \
+	--exclude-from=$exf\" | pee \"cat > $TARFILE\" \"tar tzvf -\""
+ echo `date`" $computer: Running $TARCMD" >&2
+ eval $TARCMD > $DATE.idx
 
- if eval $TARCMD; then
+ if [ -s $TARFILE ]; then
+  # File is >0 in size, we assume it worked.
+
+  # Move tar file in place
   mv $TARFILE $DATE.tgz
-  #make a filelist.
-  #update the datefile if the filelist is ok.
-  tar tvfz $DATE.tgz > $DATE.idx 2>&1 &&
-    echo $DATEs > ../.date            &&
-    touch -t $DATE ../.date
 
-  #make a sortet filelist
-  grep -v ^d $DATE.idx | sort -n -r +2 > $DATE.sdx
+  # Update timestamp
+  echo $DATEs > ../.date && touch -t $DATE ../.date
+
+  # Make a sorted filelist as well
+  sort -k6 < $DATE.idx > $DATE.sdx
 
-  #fix perm
+  # Fix permissions
   chmod 600 *tgz                      #only for us
   chmod 644 *sdx *idx 2>/dev/null     #everyone can read
 
  else
-  #it did not work
+  # Something wrong happened.
   rm $TARFILE
-  echo `date`": $TARFILE empty. $backuplevel backup of $computer:$filesystem failed and deleted"
+  echo `date`" $computer: $TARFILE empty. $backuplevel backup of $computer:$filesystem failed and deleted"
 
   # We don't want to return 0
   R=1
@@ -141,147 +159,168 @@ backup()
 
 }
 
-#Løper gjennom listen av unixmaskiner som vi skal ta backup av
 for computer in $unixcomputers
 do
 ( 
- echo `date`": Backing up $computer"
+ # Check that the target filesystem is mounted (actually check that it's not
+ # the root filesystem)
+ rootfilesystem=`df -P /`
+ targetfilesystem=`df -P "$storagedir/$computer"`
+ if [ "$rootfilesystem" == "$targetfilesystem" ]; then
+ 	echo `date`" $computer: Target filesystem ($storagedir/$computer) was mounted on /. Aborting"
+	continue
+ fi
+
+ echo `date`" $computer: Backing up $computer" >&2
 
  # Try to SSH to the computer without entering a password.
- if `ssh -n -o NumberOfPasswordPrompts=0 $computer /bin/true`; then
-  echo "Passwordless SSH to $computer works."
- else
-  echo "Could not use passwordless SSH to $computer."
+ if ! `ssh -n -o NumberOfPasswordPrompts=0 root@$computer /bin/true`; then
+  echo `date`" $computer: Could not use passwordless SSH."
 
   # We don't want to return 0
   R=1
-  break;
+  # Do next computer
+  continue;
  fi
  
- #tømmer variabelen for sikkerhets skyld
  filesystems=""
 
- #Sjekker nest siste felt i fstab. Om det er 0 tar vi ikke backup
- filesystems=`ssh -n $computer "cat /etc/fstab" \
+ # Check dump bit in fstab
+ filesystems=`ssh -n root@$computer "cat /etc/fstab" \
   | grep -v nfs \
   | grep -v "^#" \
   | grep -v "^$" \
   | awk '{ if ( $(NF-1) != "0" ) print $2}' `
 
- echo "Filesystems to backup on $computer: $filesystems"
-
- #clean up our dir at this client
- if ! ssh $computer "rm -r ~/.backup ; mkdir -m 700 ~/.backup"; then
-  echo "Could not create backup staging area at $computer:~/.backup"
-  # We don't want to return 0
+ # Clean up our dir at this client
+ if ! ssh root@$computer "rm -r ~/.backup ; mkdir -m 700 ~/.backup"; then
+  echo `date`" $computer: Could not create backup staging area at $computer:~/.backup - skipping backup of $computer"
   R=1
-  break;
+  continue;
  fi
 
- #try to copy $exclude to $computer
- if ! scp $exclude $computer:~/.backup/exclude > /dev/null; then
-  echo "Could not copy exclude.txt to $computer"
-  # We don't want to return 0
+ # Try to copy $exclude to $computer
+ if ! scp $exclude root@$computer:~/.backup/exclude > /dev/null; then
+  echo `date`" $computer: Could not copy exclude.txt to $computer - skipping backup of $computer"
   R=1
-  break;
+  continue;
+ fi
+
+ # Try to copy preeexec and postexec if they exist
+ if [ -f $confdir/preexec.$computer ]; then
+   if ! scp $confdir/preexec.$computer  root@$computer:~/.backup/preexec >&2; then
+     echo `date`" $computer: Could not copy preexec.$computer to $computer:~/.backup/preexec - skipping backup of $computer"
+     R=1
+     continue
+   fi
+ fi
+ if [ -f $confdir/postexec.$computer ]; then
+   if ! scp $confdir/postexec.$computer root@$computer:~/.backup/postexec >&2; then
+     echo `date`" $computer: Could not copy postexec.$computer to $computer:~/.backup/postexec - skipping backup of $computer"
+     R=1
+     continue
+   fi
+ fi
+
+ # Try to run preexec if it exist
+ if ! ssh root@$computer "[ ! -f ~/.backup/preexec ] || /bin/bash -x ~/.backup/preexec" >&2; then
+ 	echo `date`" $computer: Could not run $computer:~/.backup/preexec - skipping backup of $computer"
+	R=1
+	continue
  fi
 
- #try to copy preeexec and postexec if they exist
-# TODO: Gah, clean this mess!
- [ -f $confdir/preexec.$computer ] && (
-    scp $confdir/preexec.$computer  $computer:~/.backup/preexec ||
-     ( echo "Could not copy preexec.$computer to $computer:~/.backup/preexec";
-       R=1
-       break
-     )
-    )
- [ -f $confdir/postexec.$computer ] && (
-    scp $confdir/postexec.$computer $computer:~/.backup/postexec ||
-     ( echo "Could not copy postexec.$computer to $computer:~/.backup/postexec"
-       break
-       R=1
-     )
-    )
-
- #try to run preexec if it exist
- ssh $computer "[ -f ~/.backup/preexec ] && /bin/bash -x ~/.backup/preexec"
 
  for filesystem in $filesystems
  do
-  #lager en variant uten tegnet "/" eller $ (gjelder NT)
+  # Remove / and $ (for Windows) for path names
   sfilesystem=`echo $filesystem | tr '\/\$' '__'`
 
-  #lager det som trengs av kataloger
+  # Prepare storage area
   mkdir -m 755 -p $storagedir/$computer/$sfilesystem/{full,daglig} 2>/dev/null
-
   echo $filesystem > ${storagedir}/${computer}/.${sfilesystem}.name 
   chmod 644 ${storagedir}/${computer}/.${sfilesystem}.name
 
-  #set default backuplevel
+  # Default backuplevel
   backuplevel=daglig
 
   if [ ! -f $storagedir/$computer/$sfilesystem/.date ]
   then
-   #take the first full backup of this filesystem on this computer
+   # Take the first full backup of this filesystem on this computer
    backuplevel=full
    echo $DATEs > $storagedir/$computer/$sfilesystem/.date
   fi
 
-  #sjekker om det er på tide med en full
-  if [ -z "`find $storagedir/$computer/$sfilesystem/full/ -name \*tgz -mtime -$dagermellomfulle`" ]; then
+  # Check if we want a full backup - Debug statement that causes noise in cron mail:
+  # printf " $computer $filesystem: $confdir/fastfullbackupdag.$computer.$sfilesystem\n"
+  if [ -f $confdir/fastfullbackupdag.$computer.$sfilesystem ]; then
+    fullbackup_min_for_this_machine=$dagermellomfulleforfast
+    if [ "$DAY_OF_MONTH" = "`cat $confdir/fastfullbackupdag.$computer.$sfilesystem`" ]; then
+      backuplevel=full
+    fi
+  elif [ -f $confdir/fastfullbackupdag.$computer ]; then
+    fullbackup_min_for_this_machine=$dagermellomfulleforfast
+    if [ "$DAY_OF_MONTH" = "`cat $confdir/fastfullbackupdag.$computer`" ]; then
+      backuplevel=full
+    fi
+  else
+    fullbackup_min_for_this_machine=$dagermellomfulle 
+  fi
+
+  if [ -z "`find $storagedir/$computer/$sfilesystem/full/ -name \*tgz -mtime -$fullbackup_min_for_this_machine`" ]; then
    backuplevel=full
   fi
- 
-  #gå ned i rett katalog, eller dø 
-  # TODO bør sende mail om dette skjer!
+
+  # We want to be here
   cd $storagedir/$computer/$sfilesystem/$backuplevel || die
 
-  #perform the actual backup
+  # Perform the actual backup
   backup
 
-  # Sjekk om det skal være et annet antall fulle backuper av en boks
+  # Check if this box has a custom number of full backups
   if [ -f $confdir/maksfulle.$computer ] ; then
     mf=$((`cat $confdir/maksfulle.$computer`+1))
   else
     mf=$(($maksantallfulle+1))
   fi
 
-  #delete complete backups
-  for full in `ls -1t $storagedir/$computer/$sfilesystem/full/*tgz | tail +$mf`
+  # Delete old full backups
+  for full in `ls -1t $storagedir/$computer/$sfilesystem/full/*tgz | tail -n +$mf`
   do
    prefix=`echo $full | sed "s/\.[^.]*$//"`
-   echo "$computer:$filesystem sletter full $prefix (for mange)"
+   echo `date`": $computer:$filesystem Deleting full backup $prefix" >&2
    rm $prefix*
   done
 
-  #delete incremental backups older than the oldest complete backup
+  # Delete incremental backups older than the oldest full backup
   oldf=`ls -t1 $storagedir/$computer/$sfilesystem/full/*tgz | tail -1`
   find \
      $storagedir/$computer/$sfilesystem/daglig \
      -type f \
      \! -newer $oldf \
-     -exec rm {} \;
+     -printf "`date`: $computer: Deleting old incremental backup: %p\n" \
+     -exec rm {} \; >&2
  done
 
- #try to run postexec if it exist
- ssh $computer "[ -f ~/.backup/postexec ] && /bin/bash -x ~/.backup/postexec"
+ # Try to run postexec if it exist
+ if ! ssh root@$computer "[ ! -f ~/.backup/postexec ] || /bin/bash -x ~/.backup/postexec" >&2; then
+ 	echo `date`" $computer: Could not run $computer:~/.backup/postexec"
+	R=1
+ fi
 
-)  
-done &
+) &
+done
 
 wait
 
-# print diskusage to logfile
-echo "Current disk usage:"
-df -k $storagedir
-
 # Remove lockfile
 rm $LOCKFILE
 
 # Did anything go wrong?
+# (IMPORTANT NOTE: The R-business does not work as expected, since this script
+# forks itself).
 if [ $R != 0 ]; then
 	echo `date`": Backup run ended with errors, check logs."
 	exit 1
 else
-	echo `date`": Backup run ended"
+	echo `date`": Backup run ended" >&2
 fi