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