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         sleep 3
  76         exit 9
  77 }
  78
  79 ExecJob() {
  80         local what="$1"
  81         local cmd="$2"
  82
  83         echo "Running job ${what}-exec command ..."
  84         [ "$local" -eq 0 ] \
  85                 && cmd="$ssh_cmd ${user}@${system} $cmd"
  86         echo -n "Start date (${what}-exec): "; date
  87         echo "$cmd"
  88         if [ "$DRYRUN" -eq 0 ]; then
  89                 $SHELL -c "$cmd"; local ret=$?
  90         else
  91                 echo " *** Trial run, not executing ${what}-exec command!"
  92                 ret=0
  93         fi
  94         [ $ret -eq 0 ] \
  95                 && echo "The ${what}-exec command completed with status 0, OK." \
  96                 || echo "The ${what}-exec command completed with ERRORS, code $ret!"
  97         return $ret
  98 }
  99
 100 GetFS() {
 101         local dir="$1"
 102
 103         while [ -n "$dir" ]; do
 104                 findmnt -fn -o FSTYPE --raw "$dir" 2>/dev/null; local r=$?
 105                 if [ $r -eq 0 ]; then
 106                         return 0
 107                 elif [ $r -eq 127 ]; then
 108                         echo "UNKNOWN"
 109                         return 1
 110                 fi
 111                 dir=$(dirname "$dir") || return 1
 112         done
 113 }
 114
 115 CreateSubvolume() {
 116         local volume="$1"
 117         local fs
 118         local dir
 119
 120         dir=$(dirname "$volume")
 121         fs=$(GetFS "$dir")
 122         case "$fs" in
 123           "btrfs")
 124                 btrfs subvolume create "$volume"  >/dev/null || return 1
 125                 ;;
 126           "zfs")
 127                 zfs create "$(echo "$volume" | cut -c2-)" || return 1
 128                 ;;
 129           *)
 130                 echo "CreateSubvolume: Incompatible FS type \"$fs\" on \"$dir\"!"
 131                 return 9
 132         esac
 133         return 0
 134 }
 135
 136 CloneSubvolume() {
 137         local source="$1"
 138         local volume="$2"
 139         local snapshot="$3"
 140         local dir
 141         local fs
 142         local link_name
 143
 144         dir=$(dirname "source")
 145         fs=$(GetFS "$source")
 146         case "$fs" in
 147           "btrfs")
 148                 btrfs subvolume snapshot "$source" "$snapshot"  >/dev/null || return 1
 149                 ;;
 150           "zfs")
 151                 zfs snapshot "$snapshot" || return 1
 152                 link_name="$(echo "$snapshot" | cut -d@ -f2-)"
 153                 ln -s \
 154                         "$volume/.zfs/snapshot/$link_name" \
 155                         "$(dirname "$volume")/$link_name"
 156                 ;;
 157           *)
 158                 echo "CloneSubvolume: Incompatible FS type \"$fs\" on \"$source\"!"
 159                 return 9
 160         esac
 161         return 0
 162 }
 163
 164 RenameSubvolume() {
 165         local source="$1"
 166         local target="$2"
 167         local fs
 168
 169         fs=$(GetFS "$source")
 170         case "$fs" in
 171           "btrfs")
 172                 mv "$source" "$target" || return 1
 173                 ;;
 174           "zfs")
 175                 zfs rename \
 176                   "$(echo "$source" | cut -c2-)" \
 177                   "$(echo "$target" | cut -c2-)" \
 178                         || return 1
 179                 ;;
 180           *)
 181                 echo "RenameSubvolume: Incompatible FS type \"$fs\" on \"$source\"!"
 182                 return 9
 183         esac
 184         return 0
 185 }
 186
 187 DeleteSubvolume() {
 188         local volume="$1"
 189         local fs
 190         local id
 191         local snapshot
 192
 193         fs=$(GetFS "$volume")
 194         case "$fs" in
 195           "btrfs")
 196                 btrfs subvolume delete "$volume" >/dev/null || return 1
 197                 ;;
 198           "zfs")
 199                 id="$(basename "$volume")"
 200                 if [ -h "$volume" ]; then
 201                         snapshot="$(dirname "$volume")/current@$id"
 202                 else
 203                         snapshot="$volume"
 204                 fi
 205                 zfs destroy -r "$(echo "$snapshot" | cut -c2-)" >/dev/null || return 1
 206                 [ -h "$volume" ] && rm "$volume"
 207                 ;;
 208           *)
 209                 echo "DeleteSubvolume: Incompatible FS type \"$fs\" on \"$volume\"!"
 210                 return 9
 211         esac
 212         return 0
 213 }
 214
 215 Initialize_Last_SysTarget_Snapshot() {
 216         sys_target="$1"
 217         unset last
 218         unset snapshot
 219
 220         fs=$(GetFS "$sys_target")
 221         case "$fs" in
 222           "btrfs")
 223                 # Search directory of last generation, if any
 224                 last=$(ls -1d "$sys_target"/[0-9]* 2>/dev/null | sort -r | head -n1)
 225                 if [ -n "$last" ]; then
 226                         if [ ! -d "$last" ]; then
 227                                 echo "Last snapshot \"$last\" seems not to be a directory!? \"$system\" skipped!"
 228                                 echo
 229                                 return 1
 230                         fi
 231                 fi
 232                 sys_target="$sys_target/$(date +%Y%m%d-%H%M%S)"
 233                 snapshot="$sys_target"
 234                 ;;
 235           "zfs")
 236                 # On ZFS, the last generation is always named "current"
 237                 [ -e "$sys_target/current" ] \
 238                         && last="$sys_target/current" \
 239                         || last=""
 240                 snapshot="$(echo "$sys_target/current" | cut -c2-)@$(date +%Y%m%d-%H%M%S)"
 241                 sys_target="$sys_target/current"
 242                 ;;
 243           *)
 244                 echo "Initialize_Last_SysTarget_Snapshot: Incompatible FS type \"$fs\" on \"$sys_target\"!"
 245                 return 1
 246         esac
 247         return 0
 248 }
 249
 250 while [ $# -gt 0 ]; do
 251         case "$1" in
 252           "-n"|"--dry-run")
 253                 DRYRUN=1; shift
 254                 ;;
 255           "-p"|"--progress")
 256                 VERBOSE=1; shift
 257                 ;;
 258           "-"*)
 259                 Usage
 260                 ;;
 261           *)
 262                 break
 263         esac
 264 done
 265
 266 trap GotSignal SIGINT
 267
 268 echo -n "Started: "; date
 269
 270 for conf in "/etc/backup-script.conf" "${conf_d}/backup-script.conf"; do
 271         if [ -r "$conf" ]; then
 272                 echo "Reading configuration: \"$conf\" ..."
 273                 source "$conf"
 274         fi
 275 done
 276 echo
 277
 278 if [ $# -ge 1 ]; then
 279         for s in "$@"; do
 280                 if [ ! -r "${conf_d}/$s" ]; then
 281                         echo "$NAME: Can' read \"${conf_d}/$s\"!"
 282                         exit 1
 283                 fi
 284                 sys="$sys ${conf_d}/$s"
 285         done
 286 else
 287         sys="${conf_d}/"*
 288 fi
 289
 290 # check and create PID file
 291 if [ -e "$PIDFILE" ]; then
 292         echo "Lockfile \"$PIDFILE\" already exists."
 293         echo "Is an other instance still running?"
 294         echo
 295         echo -n "Aborted: "; date
 296         echo
 297         exit 3
 298 fi
 299 touch "$PIDFILE" 2>/dev/null
 300 if [ $? -ne 0 ]; then
 301         echo "Warning: can't create PID file \"$PIDFILE\"!"
 302         echo
 303 else
 304         echo "$$" >>"$PIDFILE"
 305 fi
 306
 307 if [ -n "$pre_exec" ]; then
 308         echo "Executing \"$pre_exec\" ..."
 309         sh -c $pre_exec
 310         if [ $? -ne 0 ]; then
 311                 echo "Error: pre-exec command failed!"; echo
 312                 CleanUp
 313                 echo "Aborting backup."; echo
 314                 exit 2
 315         fi
 316         sleep 2
 317         echo
 318 fi
 319
 320 for f in $sys; do
 321         [ -r "$f" -a -f "$f" ] || continue
 322
 323         fname=$(basename "$f")
 324         case "$fname" in
 325                 "backup-script.conf"|*.sh)
 326                         continue
 327                         ;;
 328         esac
 329
 330         # Set global defaults
 331         system="$fname"
 332         user="$default_user"
 333         source_root="$default_source_root"
 334         target="$default_target"
 335         ssh_args_add="$default_ssh_args_add"
 336         rsync_args_add="$default_rsync_args_add"
 337         exclude_args_add="$default_exclude_args_add"
 338         compress="$default_compress"
 339         ping="$default_ping"
 340         local="$default_local"
 341         generations="$default_generations"
 342         job_pre_exec="$default_job_pre_exec"
 343         job_post_exec="$default_job_post_exec"
 344
 345         # Compatibility with backup-pull(1) script: Save global values ...
 346         pre_exec_saved="$pre_exec"
 347         post_exec_saved="$post_exec"
 348
 349         # Compatibility with backup-pull(1) script: Set defaults
 350         unset host
 351         unset source
 352         unset pre_exec
 353         unset post_exec
 354
 355         # Read in system configuration file
 356         source "$f"
 357
 358         # Compatibility with backup-pull(1) script: Fix up configuration
 359         [ "$system" = "$fname" -a -n "$host" ] \
 360                 && system="$host"
 361         [ "$source_root" = "$default_source_root" -a -n "$source" ] \
 362                 && source_root="$source"
 363         [ -z "$job_pre_exec" -a -n "$pre_exec" ] \
 364                 && job_pre_exec="$pre_exec"
 365         [ -z "$job_post_exec" -a -n "$post_exec" ] \
 366                 && job_post_exec="$post_exec"
 367
 368         # Compatibility with backup-pull(1) script: Restore global values ...
 369         pre_exec="$pre_exec_saved"
 370         post_exec="$post_exec_saved"
 371
 372         # Validate configuration
 373         [ "$system" = "localhost" -o "$system" = "127.0.0.1" ] && local=1
 374
 375         [ "$system" = "$fname" ] \
 376                 && systxt="\"$system\"" \
 377                 || systxt="\"$fname\" [\"$system\"]"
 378         [ "$local" -eq 0 ] \
 379                 && echo "Working on $systxt ..." \
 380                 || echo "Working on $systxt (local system) ..."
 381
 382         count_all=$count_all+1
 383
 384         # Check target directory
 385         if [ -z "$target" ]; then
 386                 echo "No target directory specified for \"$system\"!? Skipped!"
 387                 echo; continue
 388         fi
 389         if [ ! -d "$target" ]; then
 390                 echo "Target \"$target\" is not a directory!? \"$system\" skipped!"
 391                 echo; continue
 392         fi
 393
 394         sys_target="$target/$fname"
 395         sys_root="$sys_target"
 396         if [ "$DRYRUN" -eq 0 -a ! -e "$sys_target" ]; then
 397                 if [ $generations -gt 0 ]; then
 398                         CreateSubvolume "$sys_target"
 399                 else
 400                         mkdir -p "$sys_target"
 401                 fi
 402                 if [ $? -ne 0 ]; then
 403                         echo "Can't create \"$sys_target\"!? \"$system\" skipped!"
 404                         echo; continue
 405                 fi
 406         fi
 407
 408         if [ "$local" -eq 0 -a "$ping" -ne 0 ]; then
 409                 # Check if system is alive
 410                 ping -c 1 "$system" >/dev/null 2>&1
 411                 if [ $? -ne 0 ]; then
 412                         echo "Host \"$system\" seems not to be alive!? Skipped."
 413                         echo; continue
 414                 fi
 415                 echo "OK, host \"$system\" seems to be alive."
 416         fi
 417
 418         if [ $generations -gt 0 ]; then
 419                 # Make sure no old backup is stored in system directory
 420                 if [ -e "$sys_target/.stamp" ]; then
 421                         # There seems to be a genearation-less backup in the
 422                         # target directory!
 423                         echo "Target directory \"$sys_target\" seems to be unclean!? \"$system\" skipped!"
 424                         echo; continue
 425                 fi
 426
 427                 Initialize_Last_SysTarget_Snapshot "$sys_target" || continue
 428
 429                 if [ -n "$last" -a ! -e "$last/.stamp" ]; then
 430                         # Old backup directory without "stamp file", continue
 431                         echo "Found incomplete snapshot in \"$last\", reusing and renaming it ..."
 432                         RenameSubvolume "$last" "$sys_target"
 433                         if [ $? -ne 0 ]; then
 434                                 echo "Failed to rename last snapshot \"$last\" to \"$sys_target\"!? \"$system\" skipped!"
 435                                 echo; continue
 436                         fi
 437                 elif [ -n "$last" ]; then
 438                         # Old backup directory found, create new snapshot
 439                         echo "Found last snapshot in \"$last\"."
 440                         if [ "$DRYRUN" -eq 0 ]; then
 441                                 CloneSubvolume "$last" "$sys_target" "$snapshot"; r=$?
 442                                 if [ $r -ne 0 ]; then
 443                                         echo "Can't create snapshot \"$snapshot\" of \"$last\", code $r!? \"$system\" skipped!"
 444                                         echo; continue
 445                                 fi
 446                                 echo "Created new snapshot in \"$snapshot\"."
 447                         else
 448                                 echo " *** Trial run, not creating new snapshot in \"$sys_target\"!"
 449                         fi
 450                 else
 451                         # No old backup found, create new subvolume
 452                         if [ "$DRYRUN" -eq 0 ]; then
 453                                 CreateSubvolume "$sys_target"; r=$?
 454                                 if [ $r -ne 0 ]; then
 455                                         echo "Can't create subvolume \"$sys_target\", code $r!? \"$system\" skipped!"
 456                                         echo; continue
 457                                 fi
 458                                 echo "Created new subvolume in \"$sys_target\"."
 459                         else
 460                                 echo " *** Trial run, not creating new subvolume \"$sys_target\"!"
 461                         fi
 462                 fi
 463         fi
 464
 465         ssh_cmd="ssh"
 466         [ -n "$ssh_args_add" ] && ssh_cmd="$ssh_cmd $ssh_args_add"
 467
 468         # execute job "pre-exec" command, if any
 469         if [ -n "$job_pre_exec" ]; then
 470                 ExecJob pre "$job_pre_exec" ; ret=$?
 471                 if [ $ret -ne 0 ]; then
 472                         [ $ret -ne 99 ] && count_started=$count_started+1
 473                         echo "Pre-exec command failed, \"$system\" skipped!"
 474                         echo; continue
 475                 fi
 476         fi
 477
 478         # prepare (remote) command ...
 479         cmd="rsync --archive"
 480         [ "$compress" -ne 0 ] && cmd="$cmd --compress"
 481         [ "$local" -eq 0 ] && cmd="$cmd --rsh=\"$ssh_cmd\""
 482         cmd="$cmd --delete --delete-excluded --sparse"
 483         [ "$VERBOSE" -gt 0 ] && cmd="$cmd --progress"
 484         if [ "$source_root" = "$default_source_root" ]; then
 485                 cmd="$cmd --exclude=/dev --exclude=/proc --exclude=/sys"
 486                 cmd="$cmd --exclude=/run --exclude=/tmp --exclude=/var/tmp"
 487                 cmd="$cmd --exclude=/media --exclude=/mnt --exclude=/net"
 488                 cmd="$cmd --exclude=/var/cache/apt --exclude=/var/log"
 489         fi
 490         [ -n "$exclude_args_add" ] && cmd="$cmd $exclude_args_add"
 491         [ -n "$rsync_args_add" ] && cmd="$cmd $rsync_args_add"
 492
 493         [ "$local" -eq 0 ] \
 494                 && cmd="$cmd ${user}@${system}:$source_root $sys_target/" \
 495                 || cmd="$cmd $source_root $sys_target/"
 496
 497         echo "Backing up to \"$sys_target\" ..."
 498         echo -n "Start date: "; date
 499         echo "$cmd"
 500         count_started=$count_started+1
 501         ok=0
 502
 503         if [ "$DRYRUN" -eq 0 ]; then
 504                 rm -f "$sys_target/.stamp"
 505                 $SHELL -c "$cmd"; ret=$?
 506                 echo "code=$ret" >"$sys_target/.stamp"
 507         else
 508                 echo " *** Trial run, not executing synchronization command!"
 509                 ret=0
 510         fi
 511
 512         if [ $ret -eq 20 ]; then
 513                 echo "Backup of \"$system\" interrupted. Aborting ..."
 514                 CleanUp
 515                 exit 1
 516         fi
 517
 518         echo -n "End date: "; date
 519         if [ $ret -eq 0 -o $ret -eq 24 ]; then
 520                 [ $ret -eq 24 ] && count_ok_vanished=$count_ok_vanished+1
 521
 522                 echo "System \"$system\" completed with status $ret, OK."
 523                 [ "$DRYRUN" -gt 0 ] || count_ok=$count_ok+1
 524                 ok=1
 525         else
 526                 echo "System \"$system\" completed with ERRORS, code $ret!"
 527         fi
 528
 529         # execute job "post-exec" command, if any
 530         if [ -n "$job_post_exec" ]; then
 531                 ExecJob post "$job_post_exec"
 532         fi
 533
 534         if [ $generations -gt 0 ]; then
 535                 # Update "latest" symlink
 536                 if [ "$DRYRUN" -eq 0 ]; then
 537                         rm -f "$sys_root/latest"
 538                         ln -s "$sys_target" "$sys_root/latest"
 539                 fi
 540                 # Clean up old generations
 541                 declare -i gen_count=$generations+2
 542                 to_delete=$(ls -1t "$sys_root" 2>/dev/null | tail -n+$gen_count | sort)
 543                 if [ -n "$to_delete" -a $ok -eq 1 ]; then
 544                         [ "$DRYRUN" -eq 0 ] \
 545                                 && echo "Deleting old backup generations (keep $generations) ..." \
 546                                 || echo " *** Trial run, not deleting old generations:"
 547                         for delete in $to_delete; do
 548                                 dir="$sys_root/$delete"
 549                                 if [ ! -e "$dir/.stamp" ]; then
 550                                         echo "Not deleting \"$dir\", not a backup directory!?"
 551                                         continue
 552                                 fi
 553                                 last=$(stat "$dir/.stamp" 2>/dev/null | grep "^Modify: " \
 554                                  | cut -d':' -f2- | cut -d. -f1)
 555                                 echo "Removing backup from" $last "..."
 556                                 if [ "$DRYRUN" -eq 0 ]; then
 557                                         DeleteSubvolume "$dir"
 558                                         [ $? -eq 0 ] || \
 559                                           echo "Failed to delete \"$dir\"!"
 560                                 fi
 561                         done
 562                         echo -n "Clean up finished: "; date
 563                 elif [ -n "$to_delete" ]; then
 564                         echo "There have been errors, not cleaning up old generations!"
 565                 else
 566                         echo "Nothing to clean up."
 567                 fi
 568         fi
 569
 570         destinations="$destinations $target"
 571         echo
 572 done
 573
 574 sync
 575
 576 paths=$( echo $destinations | sed -e 's/ /\n/g' | sort | uniq )
 577 if [ "$DRYRUN" -eq 0 -a -n "$paths" ]; then
 578         df -h $paths
 579         echo
 580 fi
 581
 582 CleanUp
 583
 584 echo -n "Done: "; date
 585 echo
 586 [ $count_all -eq 1 ] && s="" || s="s"
 587 echo " - $count_all job$s defined,"
 588 [ $count_started -eq 1 ] && s="" || s="s"
 589 echo " - $count_started job$s started,"
 590 echo " - $count_ok done without errors."
 591 echo
 592
 593 if [ $count_started -ne $count_ok ]; then
 594         echo "----->  THERE HAVE BEEN ERRORS!  <-----"
 595         echo
 596 fi
 597
 598 # -eof-