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