iankelling.org Git - distro-setup/blob - btrbk-run

   1 #!/bin/bash
   2 # Copyright (C) 2016 Ian Kelling
   3
   4 # Licensed under the Apache License, Version 2.0 (the "License");
   5 # you may not use this file except in compliance with the License.
   6 # You may obtain a copy of the License at
   7
   8 #     http://www.apache.org/licenses/LICENSE-2.0
   9
  10 # Unless required by applicable law or agreed to in writing, software
  11 # distributed under the License is distributed on an "AS IS" BASIS,
  12 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
  13 # See the License for the specific language governing permissions and
  14 # limitations under the License.
  15
  16
  17 # todo: if we cancel in the middle of a btrfs send, then run again immediately, the received subvolume doesn't get a Received UUID: field, and we won't mount it. Need to figure out a solution that will fix this.
  18
  19 err-bash-trace() {
  20   local -i argc_index=0 frame i start=${1:-0} max_indent=8 indent
  21   local source
  22   local extdebug=false
  23   if [[ $(shopt -p extdebug) == *-s* ]]; then
  24     extdebug=true
  25   fi
  26   for ((frame=0; frame < ${#FUNCNAME[@]}-1; frame++)); do
  27     argc=${BASH_ARGC[frame]}
  28     argc_index+=$argc
  29     ((frame < start)) && continue
  30     if (( ${#BASH_SOURCE[@]} > 1 )); then
  31       source="${BASH_SOURCE[frame+1]}:${BASH_LINENO[frame]}:"
  32     fi
  33     indent=$((frame-start + 1))
  34     indent=$((indent < max_indent ? indent : max_indent))
  35     printf "%${indent}s↳%sin \`%s" '' "$source" "${FUNCNAME[frame]}"
  36     if $extdebug; then
  37       for ((i=argc_index-1; i >= argc_index-argc; i--)); do
  38         printf " %s" "${BASH_ARGV[i]}"
  39       done
  40     fi
  41     echo \'
  42   done
  43   return 0
  44 }
  45 err-catch() {
  46   set -E; shopt -s extdebug
  47   _err-trap() {
  48     err=$?
  49     exec >&2
  50     set +x
  51     echo "${BASH_SOURCE[1]}:${BASH_LINENO[0]}: \`$BASH_COMMAND' returned $err"
  52     err-bash-trace 2
  53     set -e # err trap does not work within an error trap
  54     "${_errcatch_cleanup[@]:-:}" # note :-: is to be compatible with set -u
  55     echo "$0: exiting with code $err"
  56     exit $err
  57   }
  58   trap _err-trap ERR
  59   set -o pipefail
  60 }
  61 err-catch
  62
  63
  64 [[ $EUID == 0 ]] || exec sudo -E "$BASH_SOURCE" "$@"
  65
  66
  67 usage() {
  68   cat <<'EOF'
  69 btrbk-run [OPTIONS]
  70 usually -t TARGET_HOST or -s SOURCE_HOST
  71
  72 Note, at source location, intentionally not executable, run and read
  73 install-my-scripts.
  74
  75 EOF
  76   echo "top of script file:"
  77   sed -n '1,/^[# ]*end command line/{p;b};q' "$0"
  78   exit $1
  79 }
  80
  81 rsync-dirs() {
  82   local host=$1
  83   local path=$2
  84   m rsync $dry_run_arg -ahi --relative --delete "$path" "root@$host:/"
  85 }
  86
  87
  88 pre="${0##*/}:"
  89 m() { if $verbose; then printf "$pre %s\n"  "$*"; fi;  "$@"; }
  90 die() { printf "$pre %s\n" "$*" >&2;  exit 1; }
  91
  92 # latest $MAIL_HOST
  93 if [[ -e /b/bash_unpublished/source-state ]]; then
  94   source /b/bash_unpublished/source-state
  95 fi
  96
  97 # note q is owned by root:1000
  98
  99 mountpoints=()
 100
 101 rsync_mountpoint=/q
 102
 103 # default options
 104 conf_only=false
 105 dry_run=false # mostly for testing
 106 rate_limit=no
 107 verbose=true; verbose_arg=-v
 108 progress_arg="--progress"
 109 pull_reexec=false
 110
 111 default_args_file=/etc/btrbk-run.conf
 112 if [[ -s $default_args_file ]]; then
 113   set -- $(< $default_args_file) "$@"
 114   # i havent used this feature yet, so warn about it
 115   echo "$0: warning: default btrbk-run options set in $default_args_file (sleeping 5 seconds):"
 116   cat $default_args_file
 117   sleep 5
 118 fi
 119
 120 cron=false
 121 orig_args=("$@")
 122 temp=$(getopt -l cron,pull-reexec,help cl:m:npqs:t:vh "$@") || usage 1
 123 eval set -- "$temp"
 124 while true; do
 125   case $1 in
 126     --cron)
 127       cron=true
 128       pre=
 129       shift
 130       ;;
 131     # only creates the config file, does not run btrbk
 132     -c) conf_only=true; shift ;;
 133     # bytes per second, suffix k m g
 134     -l) rate_limit=$2; shift 2 ;;
 135     # Comma separated mountpoints to backup. This has defaults set below.
 136     -m) IFS=, mountpoints=($2); unset IFS; shift 2 ;;
 137     -n) dry_run=true; dry_run_arg=-n; shift ;;
 138     -p) progress_arg="--progress"; shift ;;
 139     --pull-reexec) pull_reexec=true; shift ;;
 140     -q) verbose=false; verbose_arg=; progress_arg=; shift ;;
 141     # source host to receive a backup from
 142     -s) source=$2; shift 2 ;;
 143     # target hosts to send to. empty is valid for just doing local
 144     # snapshot. we have default hosts we will populate.
 145     -t) IFS=, targets=($2); unset IFS; shift 2 ;;
 146     -v) verbose=true; verbose_arg=-v; shift ;;
 147     -h|--help) usage ;;
 148     --) shift; break ;;
 149     *) die "Internal error!" ;;
 150   esac
 151 done
 152
 153 # only tested commands are resume and archive
 154 cmd_arg=${1:-run}
 155
 156 case $cmd_arg in
 157   run|resume|archive) : ;;
 158   *) die "untested command arg" ;;
 159 esac
 160
 161 if (( $# > 1 )); then
 162   die: "only 1 nonoption arg is supported"
 163 fi
 164
 165 if [[ -v targets && $source ]]; then
 166   die "error: -t and -s are mutually exclusive"
 167 fi
 168
 169 if $verbose; then
 170   printf "$pre options: conf_only=%s\ndry_run=%s\nrate_limit=%s\nverbose=%s\ncmd_arg=%s" "$conf_only" "$dry_run" "$rate_limit" "$verbose" "$cmd_arg"
 171 fi
 172 ### end options parsing
 173
 174 # TODO: i wonder if there should be an option to send to the default
 175 # targets, plus any given on the command line.
 176
 177 # set default targets
 178 if [[ ! -v targets && ! $source ]]; then
 179   if [[ $HOSTNAME != "$MAIL_HOST" ]] && $cron ; then
 180     echo "MAIL_HOST=$MAIL_HOST, nothing to do"
 181     exit 0
 182   fi
 183   case $HOSTNAME in
 184     kw|x2)
 185       if ping -q -c1 -w1 iank.vpn.office.fsf.org &>/dev/null; then
 186         home=iank.vpn.office.fsf.org
 187       else
 188         home=b8.nz
 189       fi
 190       ;;&
 191     kw)
 192       targets=($home x2)
 193       ;;
 194     x2)
 195       targets=($home kw)
 196       ;;
 197     tp)
 198       targets=(frodo kd)
 199       # might not be connected to the vpn
 200       if timeout -s 9 10 ssh kw :; then
 201         targets+=(kw)
 202       fi
 203       ;;
 204     kd)
 205       targets=(frodo tp)
 206       # might not be connected to the vpn
 207       if timeout -s 9 10 ssh kw :; then
 208         targets+=(kw)
 209       fi
 210       ;;
 211     *)
 212       die "error: no default targets for this host, use -t"
 213       ;;
 214   esac
 215 fi
 216
 217 if [[ -v targets ]]; then
 218   echo "targets: ${targets[*]}"
 219 fi
 220
 221 if [[ $source ]]; then
 222   echo "source: $source"
 223 fi
 224
 225
 226 if [[ $mountpoints ]]; then
 227   for mp in ${mountpoints[@]}; do # default mountpoints to sync
 228     if [[ -e /nocow/btrfs-stale/$mp ]]; then
 229       die "error: $mp is stale, mount-latest-subvol first"
 230     fi
 231   done
 232 else
 233   # set default mountpoints
 234   case $HOSTNAME in
 235     # no remote backups atm. note, if we do enable this, configuration below will need some changes.
 236     #  frodo)
 237     #    prospective_mps=(/i)
 238     #    ;;
 239     *)
 240       prospective_mps=(/a /q)
 241       if [[ $source ]]; then
 242         source_state="$(ssh $source cat /a/bin/bash_unpublished/source-state)"
 243         eval "$source_state"
 244         source_host="$(ssh $source cat /etc/hostname)"
 245         if [[ $source_host == "$MAIL_HOST" ]]; then
 246           prospective_mps+=(/o)
 247         fi
 248       else
 249         if [[ $HOSTNAME == "$MAIL_HOST" ]]; then
 250           prospective_mps+=(/o)
 251         fi
 252       fi
 253       ;;
 254   esac
 255   for mp in ${prospective_mps[@]}; do # default mountpoints to sync
 256     if [[ -e /nocow/btrfs-stale/$mp ]]; then
 257       echo "$pre warning: $mp stale, not adding to default mountpoints"
 258       continue
 259     fi
 260     if awk '{print $2}' /etc/fstab | grep -xF $mp &>/dev/null; then
 261       mountpoints+=($mp)
 262     fi
 263   done
 264 fi
 265
 266 echo "mountpoints: ${mountpoints[*]}"
 267
 268 ##### end command line parsing ########
 269
 270 if [[ $source ]]; then
 271   if [[ $(ssh $source systemctl is-active btrbk.service) != inactive ]]; then
 272     echo "$0: error: cron btrbk is running on source. exiting out of caution"
 273     exit 1
 274   fi
 275 fi
 276
 277 # pull_reexec stops us from getting into an infinite loop if there is some
 278 # kind of weird problem
 279 pulla=false
 280 for m in "${mountpoints[@]}"; do
 281   if [[ $m == /a ]]; then
 282     pulla=true
 283     break
 284   fi
 285 done
 286 if ! $pull_reexec && [[ $source ]] && $pulla ; then
 287   tmpf=$(mktemp)
 288   scp $source:/a/bin/distro-setup/btrbk-run $tmpf
 289   if ! diff -q $tmpf $BASH_SOURCE; then
 290     echo "$pre found newer version on host $source. reexecing"
 291     install -T $tmpf /usr/local/bin/btrbk-run
 292     m /usr/local/bin/btrbk-run --pull-reexec "${orig_args[@]}"
 293     exit
 294   fi
 295 fi
 296
 297
 298
 299
 300
 301 if ! which btrbk &>/dev/null; then
 302   die "error: no btrbk binary found"
 303 fi
 304 # if our mountpoints are from stale snapshots,
 305 # it doesn't make sense to do a backup.
 306 check-subvol-stale ${mountpoints[@]} || die "found stale mountpoints in ${mountpoints[*]}"
 307
 308 # for an initial run, btrbk requires the dir to exist.
 309 mkdir -p /mnt/root/btrbk
 310 local_zone=$(date +%z)
 311
 312 if [[ $source ]]; then
 313   if ! zone=$(ssh root@$source date +%z); then
 314     die failed to ssh to root@$source
 315   fi
 316   if [[ $zone != $local_zone ]]; then
 317     die "error: dont confuse yourself with multiple time zones. $h has different timezone than localhost"
 318   fi
 319
 320 else
 321
 322   sshable=()
 323   sshfail=()
 324   min_idle_ms=$((1000 * 60 * 15))
 325   for h in ${targets[@]}; do
 326     if zone=$(ssh root@$h "mkdir -p /mnt/root/btrbk && date +%z"); then
 327       if $cron && DISPLAY=:0 xprintidle; then
 328         # This is a separate ssh because xprintidle can fail and thats ok.
 329         # Ignore this host. i sometimes use a non-main machine for testing or web browsing, knowing that
 330         # everything will be wiped by the next backup, but I dont want it to happen as Im using
 331         # it from cronjob.
 332         continue
 333       fi
 334       sshable+=($h)
 335       if [[ $zone != $local_zone ]]; then
 336         die "error: dont confuse yourself with multiple time zones. $h has different timezone than localhost"
 337       fi
 338     else
 339       sshfail+=($h)
 340     fi
 341   done
 342   if [[ ! $sshable ]] || { ! $cron && [[ $sshfail ]]; }; then
 343     die "failed to ssh to hosts: ${sshfail[*]}"
 344   else
 345     if [[ $sshfail ]]; then
 346       ret=1
 347       echo "$pre error: failed to ssh to ${sshfail[*]} but continuing with other hosts"
 348     fi
 349     targets=(${sshable[@]})
 350   fi
 351 fi
 352
 353
 354 cat >/etc/btrbk.conf <<EOF
 355 ssh_identity /root/.ssh/home
 356 # Just a guess that local7 is a good facility to pick.
 357 # It's a bit odd that the transaction log has to be logged to
 358 # a file or syslog, while other output is sent to std out.
 359 # The man does not mention a way for them to be together, but
 360 # I dunno if setting a log level like warn might also output
 361 # transaction info.
 362 transaction_syslog local7
 363
 364 # note, i had this because man said 20% speedup, but ran into
 365 # this issue, https://github.com/digint/btrbk/issues/275
 366 #stream_buffer 512m
 367
 368 # so we only run one at a time
 369 lockfile                   /var/lock/btrbk.lock
 370
 371 # default format of short does not accomidate hourly preservation setting
 372 timestamp_format long-iso
 373
 374 # only make a snapshot if things have changed
 375 snapshot_create onchange
 376 # I could make this different from target_preserve,
 377 # if one disk had less space.
 378 # for now, keeping them equal.
 379 snapshot_preserve 36h 14d 8w 24m
 380 snapshot_preserve_min 4h
 381 snapshot_dir btrbk
 382
 383 # so, total backups = ~89
 384 target_preserve 36h 14d 8w 24m
 385 target_preserve_min 4h
 386
 387 # if something fails and it's not obvious, try doing
 388 # btrbk -l debug -v dryrun
 389
 390 rate_limit $rate_limit
 391 EOF
 392
 393
 394
 395
 396
 397 vol=/mnt/root
 398 for m in ${mountpoints[@]}; do
 399   sub=${m##*/}
 400   if [[ $source ]]; then
 401     cat >>/etc/btrbk.conf <<EOF
 402 volume ssh://$source$vol
 403 subvolume $sub
 404 target send-receive $vol/btrbk
 405 EOF
 406   else
 407     cat >>/etc/btrbk.conf <<EOF
 408 volume $vol
 409 subvolume $sub
 410 EOF
 411     for tg in ${targets[@]}; do
 412       cat >>/etc/btrbk.conf <<EOF
 413 target send-receive ssh://$tg$vol/btrbk
 414 EOF
 415     done
 416   fi
 417 done
 418
 419
 420 # todo: umount first to ensure we don't have any errors
 421 # todo: do some kill fuser stuff to make umount more reliable
 422
 423
 424 if $conf_only; then
 425   exit
 426 fi
 427
 428
 429
 430 if $dry_run; then
 431   m btrbk -v -n $cmd_arg
 432   exit 0
 433 elif [[ $cmd_arg == archive ]]; then
 434   if [[ $source ]]; then
 435     m btrbk $verbose_arg $progress_arg $cmd_arg ssh://$source$vol $vol
 436   else
 437     for tg in ${targets[@]}; do
 438       m btrbk $verbose_arg $progress_arg $cmd_arg $vol ssh://$tg$vol
 439     done
 440   fi
 441   exit 0
 442 fi
 443 # -q and just using the syslog option seemed nice,
 444 # but it doesn't show when a send has a parent and when it doesn't.
 445 m btrbk $verbose_arg $progress_arg $cmd_arg
 446
 447 # if we have it, sync to systems which don't
 448 if mountpoint $rsync_mountpoint >/dev/null; then
 449   for tg in ${targets[@]}; do
 450     case $tg in
 451       li|lk)
 452         for x in /p/c/machine_specific/*.hosts; do
 453           if grep -qxF $tg $x; then
 454             dir=${x%.hosts}
 455             rsync-dirs $tg $dir
 456           fi
 457         done
 458         ;;
 459     esac
 460   done
 461 fi
 462
 463 if [[ $source ]]; then
 464   m mount-latest-subvol $verbose_arg
 465 else
 466   m /a/exe/mount-latest-remote ${targets[@]}
 467 fi
 468
 469 exit $ret
 470
 471 # todo: move variable data we don't care about backing up
 472 # to /nocow and symlink it.
 473
 474
 475 # background on btrbk timezones. with short/long, timestamps use local time.
 476 # for long, if your local time moves backwards, by moving timezones or
 477 # for an hour when daylight savings changes it, you will temporarily get
 478 # a more aggressive retention policy for the overlapping period, and
 479 # vice versa for the opposite timezone move. The alternative is using
 480 # long-iso, which puts timezone info into the timestamp, which means
 481 # that instead of shifting time, you shift the start of day/week/month
 482 # which is used for retention to your new local time, which means for
 483 # example, if you moved forward by 8 hours, the daily/weekly/monthly
 484 # retention will be 8 hours more aggressive since midnight is at a new
 485 # time, unless you fake the timzeone using the TZ env variable.
 486 # However, in the short term, there will be no inconsistencies.
 487 # I don't see any problem with shifting when the day starts for
 488 # retention, so I'm using long-iso.
 489
 490 # note to create a long-iso timestamp: date +%Y%m%dT%H%M%S%z