bin/backup-script

   1 #!/bin/bash
   2 #
   3 # backup-script system for cloning systems using rsync
   4 # Copyright (c)2008-2015 Alexander Barton <alex@barton.de>
   5 #
   6 # This program is free software; you can redistribute it and/or modify
   7 # it under the terms of the GNU General Public License as published by
   8 # the Free Software Foundation; either version 2 of the License, or
   9 # (at your option) any later version.
  10 # Please read the file COPYING, README and AUTHORS for more information.
  11 #
  12
  13 NAME=$(basename "$0")
  14 PIDFILE="/var/run/$NAME.pid"
  15
  16 DRYRUN=0
  17 VERBOSE=0
  18
  19 export LC_ALL=C
  20
  21 declare -i count_all=0
  22 declare -i count_started=0
  23 declare -i count_ok=0
  24 declare -i count_ok_vanished=0
  25
  26 destinations=""
  27
  28 # Default settings, can be overwritten in backup-script.conf:
  29 conf_d="/etc/backup-script.d"
  30 pre_exec=""
  31 post_exec=""
  32 default_source_root="/"
  33 default_target=""
  34 default_user="root"
  35 default_ssh_args_add=""
  36 default_rsync_args_add=""
  37 default_exclude_args_add=""
  38 default_compress=1
  39 default_ping=1
  40 default_local=0
  41 default_generations=0
  42 default_job_pre_exec=""
  43 default_job_post_exec=""
  44
  45 Usage() {
  46         echo "Usage: $NAME [<options>] [<system> [<system> [...]]]"
  47         echo
  48         echo "  -p, --progress    Show progress, see rsync(1)."
  49         echo "  -n, --dry-run     Test run only, don't copy any data."
  50         echo
  51         echo "When no <system> is given, all defined systems are used."
  52         echo
  53         exit 1
  54 }
  55
  56 CleanUp() {
  57         if [ -n "$post_exec" ]; then
  58                 echo "Executing \"$post_exec\" ..."
  59                 sh -c $post_exec
  60                 if [ $? -ne 0 ]; then
  61                         echo "Warning: post-exec command failed!"
  62                 fi
  63                 echo
  64         fi
  65         rm -f "$PIDFILE"
  66 }
  67
  68 GotSignal() {
  69         echo
  70         echo "--> Got break signal, cleaning up & aborting ..."
  71         echo
  72         CleanUp
  73         echo -n "Aborted: "; date
  74         echo
  75         exit 9
  76 }
  77
  78 ExecJob() {
  79         local what="$1"
  80         local cmd="$2"
  81
  82         echo "Running job ${what}-exec command ..."
  83         [ "$local" -eq 0 ] \
  84                 && cmd="$ssh_cmd ${user}@${system} $cmd"
  85         echo -n "Start date (${what}-exec): "; date
  86         echo "$cmd"
  87         if [ "$DRYRUN" -eq 0 ]; then
  88                 $SHELL -c "$cmd"; local ret=$?
  89         else
  90                 echo " *** Trial run, not executing ${what}-exec command!"
  91                 ret=0
  92         fi
  93         [ $ret -eq 0 ] \
  94                 && echo "The ${what}-exec command completed with status 0, OK." \
  95                 || echo "The ${what}-exec command completed with ERRORS, code $ret!"
  96         return $ret
  97 }
  98
  99 GetFS() {
 100         local dir="$1"
 101
 102         while [ -n "$dir" ]; do
 103                 findmnt -fn -o FSTYPE --raw "$dir" 2>/dev/null; local r=$?
 104                 if [ $r -eq 0 ]; then
 105                         return 0
 106                 elif [ $r -eq 127 ]; then
 107                         echo "UNKNOWN"
 108                         return 1
 109                 fi
 110                 dir=$(dirname "$dir") || return 1
 111         done
 112 }
 113
 114 CreateSubvolume() {
 115         local volume="$1"
 116         local fs
 117         local dir
 118
 119         dir=$(dirname "$volume")
 120         fs=$(GetFS "$dir")
 121         case "$fs" in
 122           "btrfs")
 123                 btrfs subvolume create "$volume"  >/dev/null || return 1
 124                 ;;
 125           *)
 126                 echo "CreateSubvolume: Incompatible FS type \"$fs\" on \"$dir\"!"
 127                 return 9
 128         esac
 129         return 0
 130 }
 131
 132 CloneSubvolume() {
 133         local source="$1"
 134         local volume="$2"
 135         local dir
 136         local fs
 137
 138         dir=$(dirname "source")
 139         fs=$(GetFS "$source")
 140         case "$fs" in
 141           "btrfs")
 142                 btrfs subvolume snapshot "$source" "$volume"  >/dev/null || return 1
 143                 ;;
 144           *)
 145                 echo "CloneSubvolume: Incompatible FS type \"$fs\" on \"$source\"!"
 146                 return 9
 147         esac
 148         return 0
 149 }
 150
 151 RenameSubvolume() {
 152         local source="$1"
 153         local target="$2"
 154         local fs
 155
 156         fs=$(GetFS "$source")
 157         case "$fs" in
 158           "btrfs")
 159                 mv "$source" "$target" || return 1
 160                 ;;
 161           *)
 162                 echo "RenameSubvolume: Incompatible FS type \"$fs\" on \"$source\"!"
 163                 return 9
 164         esac
 165         return 0
 166 }
 167
 168 DeleteSubvolume() {
 169         local volume="$1"
 170         local fs
 171
 172         fs=$(GetFS "$source")
 173         case "$fs" in
 174           "btrfs")
 175                 btrfs subvolume delete "$volume" >/dev/null || return 1
 176                 ;;
 177           *)
 178                 echo "DeleteSubvolume: Incompatible FS type \"$fs\" on \"$volume\"!"
 179                 return 9
 180         esac
 181         return 0
 182 }
 183
 184 while [ $# -gt 0 ]; do
 185         case "$1" in
 186           "-n"|"--dry-run")
 187                 DRYRUN=1; shift
 188                 ;;
 189           "-p"|"--progress")
 190                 VERBOSE=1; shift
 191                 ;;
 192           "-"*)
 193                 Usage
 194                 ;;
 195           *)
 196                 break
 197         esac
 198 done
 199
 200 trap GotSignal SIGINT
 201
 202 echo -n "Started: "; date
 203
 204 for conf in "/etc/backup-script.conf" "${conf_d}/backup-script.conf"; do
 205         if [ -r "$conf" ]; then
 206                 echo "Reading configuration: \"$conf\" ..."
 207                 source "$conf"
 208         fi
 209 done
 210 echo
 211
 212 if [ $# -ge 1 ]; then
 213         for s in "$@"; do
 214                 if [ ! -r "${conf_d}/$s" ]; then
 215                         echo "$NAME: Can' read \"${conf_d}/$s\"!"
 216                         exit 1
 217                 fi
 218                 sys="$sys ${conf_d}/$s"
 219         done
 220 else
 221         sys="${conf_d}/"*
 222 fi
 223
 224 # check and create PID file
 225 if [ -e "$PIDFILE" ]; then
 226         echo "Lockfile \"$PIDFILE\" already exists."
 227         echo "Is an other instance still running?"
 228         echo
 229         echo -n "Aborted: "; date
 230         echo
 231         exit 3
 232 fi
 233 touch "$PIDFILE" 2>/dev/null
 234 if [ $? -ne 0 ]; then
 235         echo "Warning: can't create PID file \"$PIDFILE\"!"
 236         echo
 237 else
 238         echo "$$" >>"$PIDFILE"
 239 fi
 240
 241 if [ -n "$pre_exec" ]; then
 242         echo "Executing \"$pre_exec\" ..."
 243         sh -c $pre_exec
 244         if [ $? -ne 0 ]; then
 245                 echo "Error: pre-exec command failed!"; echo
 246                 CleanUp
 247                 echo "Aborting backup."; echo
 248                 exit 2
 249         fi
 250         sleep 2
 251         echo
 252 fi
 253
 254 for f in $sys; do
 255         [ -r "$f" -a -f "$f" ] || continue
 256
 257         fname=$(basename "$f")
 258         case "$fname" in
 259                 "backup-script.conf"|*.sh)
 260                         continue
 261                         ;;
 262         esac
 263
 264         # Set global defaults
 265         system="$fname"
 266         user="$default_user"
 267         source_root="$default_source_root"
 268         target="$default_target"
 269         ssh_args_add="$default_ssh_args_add"
 270         rsync_args_add="$default_rsync_args_add"
 271         exclude_args_add="$default_exclude_args_add"
 272         compress="$default_compress"
 273         ping="$default_ping"
 274         local="$default_local"
 275         generations="$default_generations"
 276         job_pre_exec="$default_job_pre_exec"
 277         job_post_exec="$default_job_post_exec"
 278
 279         # Read in system configuration file
 280         source "$f"
 281
 282         # Validate configuration
 283         [ "$system" = "localhost" -o "$system" = "127.0.0.1" ] && local=1
 284
 285         [ "$system" = "$fname" ] \
 286                 && systxt="\"$system\"" \
 287                 || systxt="\"$fname\" [\"$system\"]"
 288         [ "$local" -eq 0 ] \
 289                 && echo "Working on $systxt ..." \
 290                 || echo "Working on $systxt (local system) ..."
 291
 292         count_all=$count_all+1
 293
 294         # Check target directory
 295         if [ -z "$target" ]; then
 296                 echo "No target directory specified for \"$system\"!? Skipped!"
 297                 echo; continue
 298         fi
 299         if [ ! -d "$target" ]; then
 300                 echo "Target \"$target\" is not a directory!? \"$system\" skipped!"
 301                 echo; continue
 302         fi
 303
 304         sys_target="$target/$fname"
 305         sys_root="$sys_target"
 306         if [ "$DRYRUN" -eq 0 ]; then
 307                 mkdir -p "$sys_target" >/dev/null 2>&1
 308                 if [ $? -ne 0 ]; then
 309                         echo "Can't create \"$sys_target\"!? \"$system\" skipped!"
 310                         echo; continue
 311                 fi
 312         fi
 313
 314         if [ "$local" -eq 0 -a "$ping" -ne 0 ]; then
 315                 # Check if system is alive
 316                 ping -c 1 "$system" >/dev/null 2>&1
 317                 if [ $? -ne 0 ]; then
 318                         echo "Host \"$system\" seems not to be alive!? Skipped."
 319                         echo; continue
 320                 fi
 321                 echo "OK, host \"$system\" seems to be alive."
 322         fi
 323
 324         if [ $generations -gt 0 ]; then
 325                 # Make sure no old backup is stored in system directory
 326                 if [ -e "$sys_target/.stamp" ]; then
 327                         # There seems to be a genearation-less backup in the
 328                         # target directory!
 329                         echo "Target directory \"$sys_target\" seems to be unclean!? \"$system\" skipped!"
 330                         echo; continue
 331                 fi
 332
 333                 # Search directory of last generation, if any
 334                 last=$(ls -1d "$sys_target"/[0-9]* 2>/dev/null | sort -r | head -n1)
 335                 if [ -n "$last" ]; then
 336                         if [ ! -d "$last" ]; then
 337                                 echo "Last snapshot \"$last\" seems not to be a directory!? \"$system\" skipped!"
 338                                 echo; continue
 339                         fi
 340                 fi
 341                 sys_target="$sys_target/$(date +%Y%m%d-%H%M%S)"
 342
 343                 if [ -n "$last" -a ! -e "$last/.stamp" ]; then
 344                         # Old backup directory without "stamp file", continue
 345                         echo "Found incomplete snapshot in \"$last\", reusing and renaming it ..."
 346                         RenameSubvolume "$last" "$sys_target"
 347                         if [ $? -ne 0 ]; then
 348                                 echo "Failed to rename last snapshot \"$last\" to \"$sys_target\"!? \"$system\" skipped!"
 349                                 echo; continue
 350                         fi
 351                 elif [ -n "$last" ]; then
 352                         # Old backup directory found, create new snapshot
 353                         echo "Found last snapshot in \"$last\"."
 354                         if [ "$DRYRUN" -eq 0 ]; then
 355                                 CloneSubvolume "$last" "$sys_target"; r=$?
 356                                 if [ $r -ne 0 ]; then
 357                                         echo "Can't create snapshot \"$sys_target\" of \"$last\", code $r!? \"$system\" skipped!"
 358                                         echo; continue
 359                                 fi
 360                                 echo "Created new snapshot in \"$sys_target\"."
 361                         else
 362                                 echo " *** Trial run, not creating new snapshot in \"$sys_target\"!"
 363                         fi
 364                 else
 365                         # No old backup found, create new subvolume
 366                         if [ "$DRYRUN" -eq 0 ]; then
 367                                 CreateSubvolume "$sys_target"; r=$?
 368                                 if [ $r -ne 0 ]; then
 369                                         echo "Can't create subvolume \"$sys_target\", code $r!? \"$system\" skipped!"
 370                                         echo; continue
 371                                 fi
 372                                 echo "Created new subvolume in \"$sys_target\"."
 373                         else
 374                                 echo " *** Trial run, not creating new subvolume \"$sys_target\"!"
 375                         fi
 376                 fi
 377         fi
 378
 379         ssh_cmd="ssh"
 380         [ -n "$ssh_args_add" ] && ssh_cmd="$ssh_cmd $ssh_args_add"
 381
 382         # execute job "pre-exec" command, if any
 383         if [ -n "$job_pre_exec" ]; then
 384                 ExecJob pre "$job_pre_exec" ; ret=$?
 385                 if [ $ret -ne 0 ]; then
 386                         [ $ret -ne 99 ] && count_started=$count_started+1
 387                         echo "Pre-exec command failed, \"$system\" skipped!"
 388                         echo; continue
 389                 fi
 390         fi
 391
 392         # prepare (remote) command ...
 393         cmd="rsync --archive"
 394         [ "$compress" -ne 0 ] && cmd="$cmd --compress"
 395         [ "$local" -eq 0 ] && cmd="$cmd --rsh=\"$ssh_cmd\""
 396         cmd="$cmd --delete --delete-excluded --sparse"
 397         [ "$VERBOSE" -gt 0 ] && cmd="$cmd --progress"
 398         if [ "$source_root" = "$default_source_root" ]; then
 399                 cmd="$cmd --exclude=/dev --exclude=/proc --exclude=/sys"
 400                 cmd="$cmd --exclude=/run --exclude=/tmp --exclude=/var/tmp"
 401                 cmd="$cmd --exclude=/media --exclude=/mnt --exclude=/net"
 402                 cmd="$cmd --exclude=/var/cache/apt --exclude=/var/log"
 403         fi
 404         [ -n "$exclude_args_add" ] && cmd="$cmd $exclude_args_add"
 405         [ -n "$rsync_args_add" ] && cmd="$cmd $rsync_args_add"
 406
 407         [ "$local" -eq 0 ] \
 408                 && cmd="$cmd ${user}@${system}:$source_root $sys_target/" \
 409                 || cmd="$cmd $source_root $sys_target/"
 410
 411         echo "Backing up to \"$sys_target\" ..."
 412         echo -n "Start date: "; date
 413         echo "$cmd"
 414         count_started=$count_started+1
 415         ok=0
 416
 417         if [ "$DRYRUN" -eq 0 ]; then
 418                 rm -f "$sys_target/.stamp"
 419                 $SHELL -c "$cmd"; ret=$?
 420                 echo "code=$ret" >"$sys_target/.stamp"
 421         else
 422                 echo " *** Trial run, not executing synchronization command!"
 423                 ret=0
 424         fi
 425
 426         if [ $ret -eq 20 ]; then
 427                 echo "Backup of \"$system\" interrupted. Aborting ..."
 428                 CleanUp
 429                 exit 1
 430         fi
 431
 432         echo -n "End date: "; date
 433         if [ $ret -eq 0 -o $ret -eq 24 ]; then
 434                 [ $ret -eq 24 ] && count_ok_vanished=$count_ok_vanished+1
 435
 436                 echo "System \"$system\" completed with status $ret, OK."
 437                 [ "$DRYRUN" -gt 0 ] || count_ok=$count_ok+1
 438                 ok=1
 439         else
 440                 echo "System \"$system\" completed with ERRORS, code $ret!"
 441         fi
 442
 443         # execute job "post-exec" command, if any
 444         if [ -n "$job_post_exec" ]; then
 445                 ExecJob post "$job_post_exec"
 446         fi
 447
 448         if [ $generations -gt 0 ]; then
 449                 # Update "latest" symlink
 450                 if [ "$DRYRUN" -eq 0 ]; then
 451                         rm -f "$sys_root/latest"
 452                         ln -s "$sys_target" "$sys_root/latest"
 453                 fi
 454                 # Clean up old generations
 455                 declare -i gen_count=$generations+2
 456                 to_delete=$(ls -1t "$sys_root" 2>/dev/null | tail -n+$gen_count | sort)
 457                 if [ -n "$to_delete" -a $ok -eq 1 ]; then
 458                         [ "$DRYRUN" -eq 0 ] \
 459                                 && echo "Deleting old backup generations (keep $generations) ..." \
 460                                 || echo " *** Trial run, not deleting old generations:"
 461                         for delete in $to_delete; do
 462                                 dir="$sys_root/$delete"
 463                                 if [ ! -e "$dir/.stamp" ]; then
 464                                         echo "Not deleting \"$dir\", not a backup directory!?"
 465                                         continue
 466                                 fi
 467                                 last=$(stat "$dir/.stamp" 2>/dev/null | grep "^Modify: " \
 468                                  | cut -d':' -f2- | cut -d. -f1)
 469                                 echo "Removing backup from" $last "..."
 470                                 if [ "$DRYRUN" -eq 0 ]; then
 471                                         DeleteSubvolume "$dir"
 472                                         [ $? -eq 0 ] || \
 473                                           echo "Failed to delete \"$dir\"!"
 474                                 fi
 475                         done
 476                         echo -n "Clean up finished: "; date
 477                 elif [ -n "$to_delete" ]; then
 478                         echo "There have been errors, not cleaning up old generations!"
 479                 else
 480                         echo "Nothing to clean up."
 481                 fi
 482         fi
 483
 484         destinations="$destinations $target"
 485         echo
 486 done
 487
 488 sync
 489
 490 paths=$( echo $destinations | sed -e 's/ /\n/g' | sort | uniq )
 491 if [ "$DRYRUN" -eq 0 -a -n "$paths" ]; then
 492         df -h $paths
 493         echo
 494 fi
 495
 496 CleanUp
 497
 498 echo -n "Done: "; date
 499 echo
 500 [ $count_all -eq 1 ] && s="" || s="s"
 501 echo " - $count_all job$s defined,"
 502 [ $count_started -eq 1 ] && s="" || s="s"
 503 echo " - $count_started job$s started,"
 504 echo " - $count_ok done without errors."
 505 echo
 506
 507 if [ $count_started -ne $count_ok ]; then
 508         echo "----->  THERE HAVE BEEN ERRORS!  <-----"
 509         echo
 510 fi
 511
 512 # -eof-