X-Git-Url: https://git.sesse.net/?a=blobdiff_plain;f=backup.sh;h=dfe8d45564859a4cc317814404d78e8cd44c00a3;hb=8dde456890eb37f4efdc7e47fb80a6c191ef4d02;hp=18968c8202ab309afc8a7f5f0567b8bffe477dfa;hpb=e86926f14d8822fcdd0d633eb7de4244e33d6cc1;p=backup.sh diff --git a/backup.sh b/backup.sh index 18968c8..dfe8d45 100755 --- a/backup.sh +++ b/backup.sh @@ -1,4 +1,4 @@ -#!/bin/sh +#!/bin/bash # backup.sh # # Backup up Unix-like computers @@ -11,308 +11,348 @@ # 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 $R has no proper meaning because of the forking subshell stuff. We need a -# more robust way of returning non-zero on errors, which we probably don't -# right now. -# o The code around line 211 and downwards is messy. -# -# Configuration -LOCKFILE=/home/backup/backuprun.lockfile -confdir=/home/backup/conf/ # Configuration files -storagedir=/backup # Where we keep backups -maksantallfulle=3 # Number of full backups -dagermellomfulle=30 # Days between full backups -DATE=`date "+%Y%m%d%H%M"` #format: touch -DATEs=`date "+%Y-%m-%d %H:%M"` #format: tar - -# Exclude-pattern -exclude=$confdir/exclude -[ ! -f $exclude ] && printf "tmp\ncore\n" > $exclude +computer=$1 # The computer to backup. +SCRIPTPATH="$( cd "$(dirname "$0")" ; pwd -P )" # Absolute directory the script lives in +confdir=$SCRIPTPATH/conf # Directory for configuration files +conffile=$confdir/backup.config +LOCKFILE=$SCRIPTPATH/backuprun.lockfile.$computer +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 +TOP_PID=$$ + +if [ -z "$computer" ]; then + echo "Usage: $0 COMPUTER" + exit 1 +fi -# Initially, we consider everything to be just fine. -R=0 +[ ! -f $exclude ] && printf "tmp\ncore\n" > $exclude # Die more or less gracefully (at least notify user) die() { - echo `date`": Something nasty happened, and since I fork a lot:" - echo `date`": I CANNOT CLEAN UP THE MESS MYSELF." - echo `date`": You need to get rid of lost process named stuff like $0, tar and ssh." - exit 255 + warnmsg "Aborting." + rm $LOCKFILE + exit 1 +} +diemsg() { + warnmsg "$@" + die +} +warnmsg() { + echo `date`": $computer: $@" + echo `date`": $computer: $@" >&2 +} +infomsg() { + echo `date`": $computer: $@" >&2 +} +findconf() { # getconf is already taken + key=$1 + computer=$2 + filesystem=$( echo $3 | tr / . ) + + if [ "$computer" ] && [ "$filesystem" ] && git config -f $conffile $computer$filesystem.$key > /dev/null; then + git config -f $conffile $computer$filesystem.$key + elif [ "$computer" ] && git config -f $conffile $computer.$key > /dev/null; then + git config -f $conffile $computer.$key + elif git config -f $conffile defaults.$key > /dev/null; then + git config -f $conffile defaults.$key + else + warnmsg "No configuration found for $key in $conffile, aborting." + kill -TERM $TOP_PID + fi } # Trap C-c and kill trap die SIGINT SIGTERM -# Don't start if we're already running +# Don't start if we're already running. Don't use diemsg, as that removes the lock file! if [ -e $LOCKFILE ]; then - echo `date`": $LOCKFILE exists, exiting." - echo `date`": $LOCKFILE exists, exiting." >&2 + warnmsg "$LOCKFILE exists for pid `cat $LOCKFILE`." exit 1 fi; touch $LOCKFILE -echo "$$" > $LOCKFILE +echo $$ > $LOCKFILE -#syntax of remotestat: -#hostname:/directory/ -remotestatf=$confdir/remotestat -[ -f $remotestatf ] && remotestat=`cat $remotestatf` -### +infomsg "Backup run starting" -PATH=/local/bin:$PATH:/store/bin -export PATH +umask 027 -echo `date`": Backup run starting" >&2 +backup() +{ + infomsg "$computer:$filesystem $backuplevel backup" + + localsnar="${storagedir}/${computer}/${sfilesystem}/.incremental.snar" + remotesnar="$remotehome/.backup/${sfilesystem}.snar" + if [ "$backuplevel" = "daily" ]; then + # If incremental backup, we need to copy the incremental status to $computer. + # If it does not exist, whine a bit and then run date-based instead. + if [ -s "$localsnar" ]; then + if ! scp $localsnar $username@$computer:$remotesnar >&2; then + diemsg "Could not copy .incremental.snar to $computer" + fi + lastcmd="--listed-incremental=$remotesnar --no-check-device" + else + infomsg "Missing incremental.snar for $filesystem, doing date-based backup instead until next full backup" + remotesnar="" + lastd=`cat ../.date` + lastcmd="--newer='$lastd'" + fi + else + # Note: Since we rm-ed .backup previously, and we don't scp over + # anything in this branch, this file will not exist, and we + # will be doing a full backup (but we want the output of it). + lastcmd="--listed-incremental=$remotesnar" + fi + + # The file is usually stored to disk with a simple cat > $TARFILE, + # but in some cases, like backing up to a remote and untrusted + # file system, it could be useful to replace it with e.g. + # a GPG command line. + storageprogram=$( findconf storageprogram $computer $sfilesystem ) + + # We try to run tar on the remote computer + # c create archive + # C change to directory first + # . where to start taring (see C) + # $lastcmd only files newer than this + # --one-file-system don't traverse file systems + # --use-compress-program compress it using gzip or pigz + # --exclude-from file to get exclusion pattern from + # Pipe the stuff over ssh to ourselves, run pee to cat the contents to a + # file and build a file list, respectivly. + TARFILE=$DATE.tmp + TARCMD="ssh -n $username@$computer \"$nice tar --one-file-system --use-compress-program $compressor -cf - -C $filesystem . $lastcmd \ + --exclude-from=$remotehome/.backup/exclude\" | pee \"$storageprogram > $TARFILE\" \"tar tzvf -\"" + infomsg "Running $TARCMD" + eval $TARCMD > $DATE.idx + + if [ $? -eq 0 ] && [ -s $TARFILE ]; then + # File is >0 in size and neither cat or tar tzvf failed; we assume it worked. + + if [ "$remotesnar" ]; then + if ! scp $username@$computer:$remotesnar $localsnar.tmp >&2; then + diemsg "Could not copy .incremental.snar from $computer" + fi + fi + + # Move tar file in place + mv $TARFILE $DATE.tgz + if [ "$remotesnar" ]; then + mv $localsnar.tmp $localsnar + fi + + # Update timestamp + echo $DATEs > ../.date && touch -t $DATE ../.date + + # Fix permissions + chmod 600 *tgz #only for us + chmod 644 *idx 2>/dev/null #everyone can read + + # Let the remote computer know that we ran a successful backup (for nagios) + ssh -n $username@$computer touch $filesystem/.lastbackup + else + # Something wrong happened. + rm $TARFILE + diemsg "tar failed or $TARFILE empty. $backuplevel backup of $computer:$filesystem failed and deleted" + fi +} -umask 027 +# Non-privileged backups currently have a few limitations compared to regular ones: +# +# - No .idx files. +# - No .snar-style incremental backups, only date-based. +# - No touching of .lastbackup for Nagios. +# - Failed tar operations may go undetected, since we only see the error code +# the last process in the pipe chain, which is gpg. +# - No storageprogram support (see above), because the file is typically already +# encrypted. +nonprivbackup() +{ + infomsg "$computer:$filesystem $backuplevel non-privileged backup" + + if [ "$backuplevel" = "daily" ]; then + lastd=`cat ../.date` + lastcmd="'$lastd'" + else + lastcmd="" + fi + + TARFILE=$DATE.tmp + TARCMD="ssh -n $username@$computer \"sudo /usr/local/sbin/output-encrypted-backup $filesystem $lastcmd\"" + infomsg "Running $TARCMD" + eval $TARCMD > $TARFILE + + if [ $? -eq 0 ] && [ -s $TARFILE ]; then + # File is >0 in size and the remote side did not report any fatal errors; we assume it worked. + + # Move tar file in place + mv $TARFILE $DATE.tgz.gpg + + # Update timestamp + echo $DATEs > ../.date && touch -t $DATE ../.date + + # Fix permissions + chmod 600 *tgz.gpg #only for us + else + # Something wrong happened. + rm $TARFILE + diemsg "tar failed or $TARFILE empty. $backuplevel backup of $computer:$filesystem failed and deleted" + fi +} -# The computers we want to back up -unixcomputers=`cat $confdir/computers.unix \ - | grep -v "^#" \ - | grep -v "^$" ` +# Get basic configuration. +storagedir=$( findconf storagedir $computer ) +username=$( findconf username $computer ) +exclude=$( findconf excludefile $computer ) -# Backup only one computer, from command line? -if [ $1 ] ; then - unixcomputers=$1 -fi; +# Check that the target filesystem is mounted (actually check that it's not +# the root filesystem) +if [ ! -d "$storagedir/$computer" ]; then + diemsg "Target filesystem ($storagedir/$computer) does not exist." +fi +rootfilesystem=`df -P /` +targetfilesystem=`df -P "$storagedir/$computer"` +if [ "$rootfilesystem" == "$targetfilesystem" ]; then + diemsg "Target filesystem ($storagedir/$computer) was mounted on /." +fi + +infomsg "Backing up $computer" +# Try to SSH to the computer without entering a password. +if ! ssh -n -o NumberOfPasswordPrompts=0 $username@$computer /bin/true; then + diemsg "Could not use passwordless SSH." +fi -#selve backupen -# krever at noen variabler er satt -# krever at vi er i rett katalog -backup() -{ +# Find the home directory of the backup user +remotehome=`ssh -n $username@$computer "echo ~"` +if [ -z "$remotehome" ]; then + diemsg "Could not expand ~ for user $username" +fi - echo -n `date` >&2 - printf " $computer: $computer:$filesystem $backuplevel backup" >&2 - - if [ "$backuplevel" = "daglig" ] || [ "$backuplevel" = "incremental" ] - then - lastd=`cat ../.date` - lastcmd="--newer='$lastd'" - else - 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 root@$computer "ls ~/.backup/exclude"` - - # We try to run tar on the remote computer - # z gzip it - # c create archive - # C change to directory first - # . where to start taring (see C) - # $lastcmd only files newer than this - # --one-file-system don't traverse file systems - # --exclude-from file to get exclusion pattern from - # Pipe the stuff over ssh to ourselves, run pee to cat the contents to a - # file and build a file list, respectivly. - TARFILE=$DATE.tmp - TARCMD="ssh root@$computer \"$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 [ -s $TARFILE ]; then - # File is >0 in size, we assume it worked. - - # Move tar file in place - mv $TARFILE $DATE.tgz - - # Update timestamp - echo $DATEs > ../.date && touch -t $DATE ../.date - - # Make a sorted filelist as well - sort -k6 < $DATE.idx > $DATE.sdx - - # Fix permissions - chmod 600 *tgz #only for us - chmod 644 *sdx *idx 2>/dev/null #everyone can read - - else - # Something wrong happened. - rm $TARFILE - echo `date`" $computer: $TARFILE empty. $backuplevel backup of $computer:$filesystem failed and deleted" - - # We don't want to return 0 - R=1 - fi +# Check if pigz is available +if ssh -n $username@$computer "pigz -V 2>/dev/null"; then + compressor=pigz +else + infomsg "pigz missing; falling back to gzip." + compressor=gzip +fi -} +# Check if nice and ionice are available +if ssh -n $username@$computer "nice -n 19 ionice -c3 /bin/true 2>/dev/null"; then + nice="nice -n 19 ionice -c3" +else + infomsg "nice and/or ionice missing; will run at normal priority." + nice="" +fi -for computer in $unixcomputers -do -( - # Check that the target filesystem is mounted (actually check that it's not - # the root filesystem) - rootfilesystem=`df -P /` - targetfilesystem=`df -P "$storagedir/$computer"` - if [ "$rootfilesystem" == "$targetfilesystem" ]; then - echo `date`" $computer: Target filesystem ($storagedir/$computer) was mounted on /. Aborting" - continue - fi - - echo `date`" $computer: Backing up $computer" >&2 - - # Try to SSH to the computer without entering a password. - if ! `ssh -n -o NumberOfPasswordPrompts=0 root@$computer /bin/true`; then - echo `date`" $computer: Could not use passwordless SSH." - - # We don't want to return 0 - R=1 - # Do next computer - continue; - fi - - filesystems="" - - # Check dump bit in fstab - filesystems=`ssh -n root@$computer "cat /etc/fstab" \ - | grep -v nfs \ - | grep -v "^#" \ - | grep -v "^$" \ - | awk '{ if ( $(NF-1) != "0" ) print $2}' ` - - # Clean up our dir at this client - if ! ssh root@$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 - continue; - fi - - # 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 - continue; - fi - - # Try to copy preeexec and postexec if they exist -# TODO: Gah, clean this mess! - [ -f $confdir/preexec.$computer ] && ( - scp $confdir/preexec.$computer root@$computer:~/.backup/preexec >&2 || - ( echo `date`" $computer: Could not copy preexec.$computer to $computer:~/.backup/preexec - skipping backup of $computer" - R=1 - continue - ) - ) - [ -f $confdir/postexec.$computer ] && ( - scp $confdir/postexec.$computer root@$computer:~/.backup/postexec >&2 || - ( echo `date`" $computer: Could not copy postexec.$computer to $computer:~/.backup/postexec - skipping backup of $computer" - R=1 - continue - ) - ) - - # Try to run preexec if it exist - if ! ssh root@$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 - - - for filesystem in $filesystems - do - # Remove / and $ (for Windows) for path names - sfilesystem=`echo $filesystem | tr '\/\$' '__'` - - # 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 - - # Default backuplevel - backuplevel=daglig - - if [ ! -f $storagedir/$computer/$sfilesystem/.date ] - then - # Take the first full backup of this filesystem on this computer - backuplevel=full - echo $DATEs > $storagedir/$computer/$sfilesystem/.date - fi - - # Check if we want a full backup - if [ -z "`find $storagedir/$computer/$sfilesystem/full/ -name \*tgz -mtime -$dagermellomfulle`" ]; then - backuplevel=full - fi - - # We want to be here - cd $storagedir/$computer/$sfilesystem/$backuplevel || die - - # Perform the actual backup - backup - - # 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 old full backups - for full in `ls -1t $storagedir/$computer/$sfilesystem/full/*tgz | tail -n +$mf` - do - prefix=`echo $full | sed "s/\.[^.]*$//"` - echo `date`": $computer:$filesystem Deleting full backup $prefix" >&2 - rm $prefix* - done - - # 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 \ - -printf "`date`: $computer: Deleting old incremental backup: %p\n" \ - -exec rm {} \; >&2 - done - - # Try to run postexec if it exist - if ! ssh root@$computer "[ ! -f ~/.backup/postexec ] || /bin/bash -x ~/.backup/postexec" >&2; then - echo `date`" $computer: Could not run $computer:~/.backup/postexec" - R=1 - fi - -) & +# Check dump bit in fstab +filesystems=`ssh -n $username@$computer "cat /etc/fstab" \ + | grep -v nfs \ + | grep -v "^#" \ + | grep -v "^$" \ + | awk '{ if ( $(NF-1) != "0" ) print $2}' ` + +# Clean up our dir at this client +if ! ssh -n $username@$computer "rm -r $remotehome/.backup ; mkdir -m 700 $remotehome/.backup"; then + diemsg "Could not create backup staging area at $computer:$remotehome/.backup" +fi + +# Try to copy $exclude to $computer +if ! scp $exclude $username@$computer:$remotehome/.backup/exclude >&2; then + diemsg "Could not copy exclude.txt to $computer" +fi + +# Try to copy preeexec and postexec if they exist +if [ -f $confdir/preexec.$computer ]; then + if ! scp $confdir/preexec.$computer $username@$computer:$remotehome/.backup/preexec >&2; then + diemsg "Could not copy preexec.$computer to $computer:$remotehome/.backup/preexec" + fi +fi +if [ -f $confdir/postexec.$computer ]; then + if ! scp $confdir/postexec.$computer $username@$computer:$remotehome/.backup/postexec >&2; then + diemsg "Could not copy postexec.$computer to $computer:$remotehome/.backup/postexec" + fi +fi + +# Try to run preexec if it exists +if ! ssh -n $username@$computer "[ ! -f $remotehome/.backup/preexec ] || /bin/bash -x $remotehome/.backup/preexec" >&2; then + diemsg "Could not run $computer:$remotehome/.backup/preexec" +fi + +for filesystem in $filesystems; do + # Remove / and $ (for Windows) for path names + sfilesystem=`echo $filesystem | tr '\/\$' '__'` + + # Prepare storage area + mkdir -m 755 -p $storagedir/$computer/$sfilesystem/full 2>/dev/null + mkdir -m 755 -p $storagedir/$computer/$sfilesystem/daily 2>/dev/null + + # Default backuplevel + backuplevel=daily + + if [ ! -s $storagedir/$computer/$sfilesystem/.date ]; then + # Take the first full backup of this filesystem on this computer + backuplevel=full + fi + + # Check if we want a full backup + fixedfullbackupday=$( findconf fixedfullbackupday $computer $filesystem ) + if [ "$fixedfullbackupday" ]; then + fullbackup_min_for_this_machine=$( findconf daysbetweenfullforfixed $computer $filesystem ) + if [ "$DAY_OF_MONTH" = "$fixedfullbackupday" ]; then + backuplevel=full + fi + else + fullbackup_min_for_this_machine=$( findconf daysbetweenfull $computer $filesystem ) + fi + + if [ -z "`find $storagedir/$computer/$sfilesystem/full/ -name \*tgz\* -mtime -$fullbackup_min_for_this_machine`" ]; then + backuplevel=full + fi + + # We want to be here + cd $storagedir/$computer/$sfilesystem/$backuplevel || diemsg "$storagedir/$computer/$sfilesystem/$backuplevel does not exist" + + # Perform the actual backup + if [ "$( findconf nonpriv $computer )" ]; then + nonprivbackup + else + backup + fi + + # Delete old full backups + mf=$( findconf maxnumfull $computer $filesystem ) + mf=$(( mf + 1 )) + for full in `ls -1t $storagedir/$computer/$sfilesystem/full/*tgz* | tail -n +$mf`; do + prefix=`echo $full | sed "s/\.[^.]*$//"` + infomsg "$computer:$filesystem Deleting full backup $prefix" + rm $prefix* + done + + # Delete incremental backups older than the oldest full backup + oldf=`ls -t1 $storagedir/$computer/$sfilesystem/full/*tgz* | tail -1` + find \ + $storagedir/$computer/$sfilesystem/daily \ + -type f \ + \! -newer $oldf \ + -printf "`date`: $computer: Deleting old incremental backup: %p\n" \ + -exec rm {} \; >&2 done -wait +# Try to run postexec if it exist +if ! ssh -n $username@$computer "[ ! -f $remotehome/.backup/postexec ] || /bin/bash -x $remotehome/.backup/postexec" >&2; then + diemsg "Could not run $computer:$remotehome/.backup/postexec" +fi + +# Use warn so that we will get an email even on success. +warnmsg "Backup completed successfully." # 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" >&2 -fi