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