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
  18 # immediately, the received subvolume doesn't get a Received UUID:
  19 # field, and we won't mount it. Need to figure out a solution that will
  20 # fix this.
  21
  22
  23 [[ $EUID == 0 ]] || exec sudo -E "${BASH_SOURCE[0]}" "$@"
  24
  25 source /usr/local/lib/err
  26
  27 usage() {
  28   cat <<'EOF'
  29 btrbk-run [OPTIONS]
  30 usually -t TARGET_HOST or -s SOURCE_HOST
  31
  32 Note, at source location, intentionally not executable, run and read
  33 install-my-scripts.
  34
  35 EOF
  36   echo "top of script file:"
  37   sed -n '1,/^[# ]*end command line/{p;b};q' "$0"
  38   exit $1
  39 }
  40
  41
  42
  43 pre="${0##*/}: "
  44 m() { if $verbose; then printf "$pre%s\n"  "$*"; fi;  "$@"; }
  45 e() { printf "$pre%s\n"  "$*"; }
  46 die() { printf "$pre%s\n" "$*" >&2; echo "exiting with status 1" >&2; exit 1; }
  47 mexit() { echo "$pre: exiting with status $1"; exit $1; }
  48
  49 # latest $MAIL_HOST
  50 if [[ -e /b/bash_unpublished/source-state ]]; then
  51   source /b/bash_unpublished/source-state
  52 fi
  53
  54 # note q is owned by root:1000
  55
  56 mountpoints=()
  57
  58 rsync_mountpoint=/q
  59
  60 ret=0
  61 # default options
  62 conf_only=false
  63 dry_run=false # mostly for testing
  64 rate_limit=no
  65 verbose=true; verbose_arg=-v
  66 progress_arg="--progress"
  67 incremental_strict=false
  68 pull_reexec=false
  69
  70 default_args_file=/etc/btrbk-run.conf
  71 if [[ -s $default_args_file ]]; then
  72   set -- $(< $default_args_file) "$@"
  73   # i havent used this feature yet, so warn about it
  74   echo "$0: warning: default btrbk-run options set in $default_args_file (sleeping 5 seconds):"
  75   cat $default_args_file
  76   sleep 5
  77 fi
  78
  79 early=false # quit early, just btrbk, no extra remounting etc.
  80 cron=false
  81 orig_args=("$@")
  82 temp=$(getopt -l cron,pull-reexec,help ceil:m:npqs:t:vh "$@") || usage 1
  83 eval set -- "$temp"
  84 while true; do
  85   case $1 in
  86     --cron)
  87       cron=true
  88       pre=
  89       shift
  90       ;;
  91     # only creates the config file, does not run btrbk
  92     -c) conf_only=true; shift ;;
  93     -e) early=true; shift ;;
  94     -i) incremental_strict=true; shift ;;
  95     # bytes per second, suffix k m g
  96     -l) rate_limit=$2; shift 2 ;;
  97     # Comma separated mountpoints to backup. This has defaults set below.
  98     -m) IFS=, mountpoints=($2); unset IFS; shift 2 ;;
  99     -n) dry_run=true; dry_run_arg=-n; shift ;;
 100     -p) progress_arg="--progress"; shift ;;
 101     --pull-reexec) pull_reexec=true; shift ;;
 102     -q) verbose=false; verbose_arg=; progress_arg=; shift ;;
 103     # source host to receive a backup from
 104     -s)
 105       source=$2
 106       bbksource=$source
 107       if [[ $source == *:* ]]; then
 108         bbksource="[$source]"
 109       fi
 110       shift 2
 111       ;;
 112     # target hosts to send to. empty is valid for just doing local
 113     # snapshot. we have default hosts we will populate.
 114     -t) IFS=, targets=($2); unset IFS; shift 2 ;;
 115     -v) verbose=true; verbose_arg=-v; shift ;;
 116     -h|--help) usage ;;
 117     --) shift; break ;;
 118     *) die "Internal error!" ;;
 119   esac
 120 done
 121
 122 # only tested commands are resume and archive
 123 cmd_arg=${1:-run}
 124
 125 case $cmd_arg in
 126   run|resume|archive) : ;;
 127   *) die "untested command arg" ;;
 128 esac
 129
 130 if (( $# > 1 )); then
 131   die: "only 1 nonoption arg is supported"
 132 fi
 133
 134 if [[ -v targets && $source ]]; then
 135   # note, this doesnt need to be the case, but
 136   # we would need to think about it.
 137   die "error: -t and -s are mutually exclusive"
 138 fi
 139
 140 if $verbose; then
 141   printf "options: conf_only=%s\ndry_run=%s\nrate_limit=%s\nverbose=%s\ncmd_arg=%s" "$conf_only" "$dry_run" "$rate_limit" "$verbose" "$cmd_arg"
 142 fi
 143 ### end options parsing
 144
 145 # remove path from earlier version of btrbk
 146 rm -f /usr/sbin/btrbk
 147 # note, this still works as intended if there is no /usr/bin/btrbk
 148 if [[ /a/opt/btrbk/btrbk -nt /usr/bin/btrbk ]]; then
 149   if [[ -e /b/distro-functions/src/package-manager-abstractions ]]; then
 150     . /b/distro-functions/src/package-manager-abstractions
 151     pi asciidoctor
 152   fi
 153   cd /a/opt/btrbk
 154   m make install
 155 fi
 156
 157 # TODO: i wonder if there should be an option to send to the default
 158 # targets, plus any given on the command line.
 159
 160
 161 amy=false
 162 # set default targets
 163 if [[ ! -v targets && ! $source ]]; then
 164   if [[ $HOSTNAME != "$MAIL_HOST" ]] && $cron ; then
 165     echo "MAIL_HOST=$MAIL_HOST, nothing to do"
 166     mexit 0
 167   else
 168     amy=true
 169   fi
 170
 171   at_work=false
 172
 173   targets=(frodo.b8.nz)
 174   case $HOSTNAME in
 175     kw)
 176       at_work=true
 177       ;;&
 178     x2|x3|sy|bo)
 179       if ping -q -c1 -w1 hal.office.fsf.org \
 180           && ip n show 192.168.0.26 | grep . &>/dev/null; then
 181         at_work=true
 182       fi
 183       ;;&
 184     kw|x2|x3|sy|bo)
 185       if $at_work; then
 186         if ping -q -c1 -w1 iank.vpn.office.fsf.org &>/dev/null; then
 187           home=iank.vpn.office.fsf.org
 188         else
 189           home=i.b8.nz
 190         fi
 191       else
 192         home=b8.nz
 193       fi
 194       ;;&
 195     kw)
 196       targets+=($home x3)
 197       ;;
 198     x2|x3|sy|bo)
 199       targets+=($home)
 200       if $at_work; then
 201         targets+=(kw.office.fsf.org x2.b8.nz)
 202       else
 203         targets+=(kw.b8.nz)
 204       fi
 205       ;;
 206     kd)
 207       targets+=(x2.b8.nz kw.b8.nz)
 208       # temporarily disabled while doing maint
 209       # if ping -q -c1 -w1 bo.b8.nz &>/dev/null; then
 210       #   targets+=(bo.b8.nz)
 211       # else
 212       #   targets+=(bow.b8.nz)
 213       # fi
 214       ;;
 215     frodo)
 216       # no targets
 217       targets=()
 218       ;;
 219     *)
 220       die "error: no default targets for this host, use -t"
 221       ;;
 222   esac
 223 fi
 224
 225 if [[ -v targets ]]; then
 226   echo "targets: ${targets[*]}"
 227 fi
 228
 229 if [[ $source ]]; then
 230   echo "source: $source"
 231 fi
 232
 233 if [[ ${mountpoints[0]} ]]; then
 234   for mp in ${mountpoints[@]}; do
 235     if [[ -e /nocow/btrfs-stale/$mp ]]; then
 236       die "error: $mp is stale, mount-latest-subvol first"
 237     fi
 238   done
 239 else
 240   # set default mountpoints
 241   if [[ ${targets[0]} == tp ]]; then
 242     prospective_mps=(/a)
 243   else
 244     case $HOSTNAME in
 245       frodo)
 246         prospective_mps=(/i)
 247         ;;
 248       *)
 249         prospective_mps=(/a /q)
 250         if [[ $source ]]; then
 251           source_state="$(ssh $source cat /a/bin/bash_unpublished/source-state)"
 252           eval "$source_state"
 253           source_host="$(ssh $source cat /etc/hostname)"
 254           if [[ $source_host == "$MAIL_HOST" ]]; then
 255             prospective_mps+=(/o)
 256           fi
 257         else
 258           if [[ $HOSTNAME == "$MAIL_HOST" ]]; then
 259             # HOST2 is really the mail host if it exists
 260             if [[ $HOST2 && $HOST2 != "$HOSTNAME" ]]; then
 261               echo "skipping /o because HOST2 is not us"
 262             else
 263               prospective_mps+=(/o)
 264             fi
 265           fi
 266         fi
 267         ;;
 268     esac
 269   fi
 270   for mp in ${prospective_mps[@]}; do # default mountpoints to sync
 271     if [[ -e /nocow/btrfs-stale/$mp ]]; then
 272       e "warning: $mp stale, not adding to default mountpoints"
 273       continue
 274     fi
 275     if awk '{print $2}' /etc/fstab | grep -xF $mp &>/dev/null; then
 276       mountpoints+=($mp)
 277     fi
 278   done
 279 fi
 280
 281 echo "mountpoints: ${mountpoints[*]}"
 282
 283 ##### end command line parsing ########
 284
 285 if [[ $source ]]; then
 286   if [[ $(ssh $source ps --no-headers -o comm 1) == systemd ]]; then
 287     status=$(ssh $source systemctl is-active btrbk.service) || : # normally returns 3
 288     case $status in
 289       inactive|failed) : ;;
 290       *)
 291         echo "$0: error: cron btrbk is running on source. exiting out of caution"
 292         mexit 1
 293     esac
 294   fi
 295 fi
 296
 297 # pull_reexec stops us from getting into an infinite loop if there is some
 298 # kind of weird problem
 299 pulla=false
 300 for m in "${mountpoints[@]}"; do
 301   if [[ $m == /a ]]; then
 302     pulla=true
 303     break
 304   fi
 305 done
 306 if ! $pull_reexec && [[ $source ]] && $pulla ; then
 307   tmpf=$(mktemp)
 308   m rsync -ra $source:/usr/local/bin/{mount-latest-subvol,check-subvol-stale} /usr/local/bin
 309   m rsync -ra $source:/usr/local/lib/err /usr/local/lib
 310   m scp $source:/a/bin/distro-setup/btrbk-run $tmpf
 311   if ! diff -q $tmpf ${BASH_SOURCE[0]}; then
 312     e "found different version on host $source. reexecing"
 313     install -T $tmpf /usr/local/bin/btrbk-run
 314     m /usr/local/bin/btrbk-run --pull-reexec "${orig_args[@]}"
 315     mexit 0
 316   fi
 317 fi
 318
 319
 320 if ! which btrbk &>/dev/null; then
 321   die "error: no btrbk binary found"
 322 fi
 323 # if our mountpoints are from stale snapshots,
 324 # it doesn't make sense to do a backup.
 325 check-subvol-stale ${mountpoints[@]} || die "found stale mountpoints in ${mountpoints[*]}"
 326
 327 # for an initial run, btrbk requires the dir to exist.
 328 mkdir -p /mnt/root/btrbk
 329 local_zone=$(date +%z)
 330
 331 if [[ $source ]]; then
 332   if ! zone=$(ssh root@$source date +%z); then
 333     die failed to ssh to root@$source
 334   fi
 335   if [[ $zone != "$local_zone" ]]; then
 336     die "error: dont confuse yourself with multiple time zones. $h has different timezone than localhost"
 337   fi
 338
 339 else
 340
 341   sshable=()
 342   sshfail=()
 343   min_idle_ms=$((1000 * 60 * 15))
 344   for h in ${targets[@]}; do
 345     if remote_info=( $(timeout -s 9 6 ssh root@$h "mkdir -p /mnt/root/btrbk && date +%z && df --output=size,pcent / | tail -n1") ); then
 346       zone=${remote_info[0]}
 347       root_size=${remote_info[1]}
 348       percent_used=${remote_info[2]%%%}
 349
 350       if (( ${#remote_info[@]} != 3 )); then
 351         die "error: didnt get 3 fields in test ssh to target $h. investigate"
 352       fi
 353
 354
 355       # we may be booted into a bootstrap fs or something
 356       min_root_kb=$(( 1024 * 1024 * 200 )) # 200 gb
 357       if (( root_size < min_root_kb )); then
 358         continue
 359       fi
 360
 361       if (( percent_used >= 98 )); then
 362         die "error: filesystem on target $h is $percent_used % full"
 363       fi
 364
 365       # This is a separate ssh because xprintidle can fail and thats ok.
 366       if $cron && idle_ms=$(timeout -s 9 6 ssh $h DISPLAY=:0 xprintidle); then
 367         if (( idle_ms < min_idle_ms )); then
 368
 369           # Ignore this host. i sometimes use a non-main machine for
 370           # testing or web browsing, knowing that everything will be wiped
 371           # by the next backup, but I dont want it to happen as Im using
 372           # it from cronjob.
 373           e "warning: $h: active X session in the last 15 minutes, skipping for now"
 374           continue
 375         fi
 376       fi
 377       sshable+=($h)
 378       if [[ $zone != "$local_zone" ]]; then
 379         die "error: dont confuse yourself with multiple time zones. $h has different timezone than localhost"
 380       fi
 381     else
 382       sshfail+=($h)
 383     fi
 384   done
 385   if [[ ! ${sshable[*]} ]] || { ! $cron && [[ ${sshfail[*]} ]]; }; then
 386     die "failed to ssh to hosts: ${sshfail[*]}"
 387   else
 388     if [[ ${sshfail[*]} ]]; then
 389       ret=1
 390       e "error: failed to ssh to ${sshfail[*]} but continuing with other hosts"
 391     fi
 392     targets=(${sshable[@]})
 393   fi
 394 fi
 395
 396
 397 cat >/etc/btrbk.conf <<EOF
 398 ssh_identity /q/root/h
 399 #ssh_identity /root/.ssh/home
 400
 401 # Just a guess that local7 is a good facility to pick.
 402 # It's a bit odd that the transaction log has to be logged to
 403 # a file or syslog, while other output is sent to std out.
 404 # The man does not mention a way for them to be together, but
 405 # I dunno if setting a log level like warn might also output
 406 # transaction info.
 407 transaction_syslog local7
 408
 409 # trying this out
 410 stream_compress zstd
 411
 412 # so we only run one at a time
 413 lockfile                   /var/lock/btrbk.lock
 414
 415 # default format of short does not accomidate hourly preservation setting
 416 timestamp_format long-iso
 417
 418 # only make a snapshot if things have changed
 419 snapshot_create onchange
 420 # I could make this different from target_preserve,
 421 # if one disk had less space.
 422 # for now, keeping them equal.
 423 snapshot_preserve 18h 14d 8w 24m
 424 snapshot_preserve_min 2h
 425 snapshot_dir btrbk
 426
 427 # so, total backups = ~58
 428 target_preserve 18h 14d 8w 24m
 429 target_preserve_min 2h
 430
 431 # if something fails and it's not obvious, try doing
 432 # btrbk -l debug -v dryrun
 433
 434 rate_limit $rate_limit
 435 EOF
 436
 437 if $incremental_strict; then
 438   cat >>/etc/btrbk.conf <<EOF
 439 incremental strict
 440 EOF
 441 fi
 442
 443 vol=/mnt/root
 444 for m in ${mountpoints[@]}; do
 445   sub=${m#/}
 446   if [[ $source ]]; then
 447     cat >>/etc/btrbk.conf <<EOF
 448 volume ssh://$bbksource$vol
 449 subvolume $sub
 450 target send-receive $vol/btrbk
 451 EOF
 452   fi
 453   if (( ${#targets[@]} )); then
 454     cat >>/etc/btrbk.conf <<EOF
 455 volume $vol
 456 subvolume $sub
 457 EOF
 458     for tg in ${targets[@]}; do
 459       # handle ipv6
 460       if [[ $tg == *:* ]]; then
 461         tg="[$tg]"
 462       fi
 463       cat >>/etc/btrbk.conf <<EOF
 464 target send-receive ssh://$tg$vol/btrbk
 465 EOF
 466     done
 467   fi
 468 done
 469
 470 # if $amy; then
 471 #   # to manually backup amy,
 472 #   # bbk -e -s amy -m root_ubuntubionic
 473 #   cat >>/etc/btrbk.conf <<'EOF'
 474 # volume ssh://amy/mnt/root
 475 # subvolume root_ubuntubionic
 476 # target send-receive /mnt/root/btrbk
 477 # EOF
 478 # fi
 479
 480 # todo: umount first to ensure we don't have any errors
 481 # todo: do some kill fuser stuff to make umount more reliable
 482
 483
 484 if $conf_only; then
 485   mexit 0
 486 fi
 487
 488
 489
 490 if $dry_run; then
 491   m btrbk -v -n $cmd_arg
 492   mexit 0
 493 elif [[ $cmd_arg == archive ]]; then
 494   if [[ $source ]]; then
 495     m btrbk $verbose_arg $progress_arg $cmd_arg ssh://$source$vol $vol
 496   else
 497     for tg in ${targets[@]}; do
 498       m btrbk $verbose_arg $progress_arg $cmd_arg $vol ssh://$tg$vol
 499     done
 500   fi
 501   mexit 0
 502 fi
 503 # -q and just using the syslog option seemed nice,
 504 # but it doesn't show when a send has a parent and when it doesn't.
 505 m btrbk $verbose_arg $progress_arg $cmd_arg
 506
 507 # todo: tp not valid anymore.
 508 # if we have it, sync to systems which don't
 509 if mountpoint $rsync_mountpoint >/dev/null; then
 510   for tg in ${targets[@]}; do
 511     case $tg in
 512       tp)
 513         dirs=(/p/c/machine_specific/tp)
 514         for x in /p/c/machine_specific/*.hosts; do
 515           if grep -qxF $tg $x; then
 516             dirs+=(${x%.hosts})
 517           fi
 518         done
 519         m rsync -aSAXPH --specials --devices --delete --relative ${dirs[@]} root@$tg:/
 520         ;;
 521     esac
 522   done
 523 fi
 524
 525 if [[ $source ]]; then
 526   m mount-latest-subvol
 527 else
 528   m /a/exe/mount-latest-remote ${targets[@]}
 529 fi
 530
 531 if [[ $ret == 0 ]]; then
 532   for tg in ${targets[@]}; do
 533     :
 534     #ssh root@$tg /a/exe/mail-backup-clean
 535   done
 536 fi
 537
 538 mexit $ret
 539
 540 # todo: move variable data we don't care about backing up
 541 # to /nocow and symlink it.
 542
 543
 544 # background on btrbk timezones. with short/long, timestamps use local time.
 545 # for long, if your local time moves backwards, by moving timezones or
 546 # for an hour when daylight savings changes it, you will temporarily get
 547 # a more aggressive retention policy for the overlapping period, and
 548 # vice versa for the opposite timezone move. The alternative is using
 549 # long-iso, which puts timezone info into the timestamp, which means
 550 # that instead of shifting time, you shift the start of day/week/month
 551 # which is used for retention to your new local time, which means for
 552 # example, if you moved forward by 8 hours, the daily/weekly/monthly
 553 # retention will be 8 hours more aggressive since midnight is at a new
 554 # time, unless you fake the timzeone using the TZ env variable.
 555 # However, in the short term, there will be no inconsistencies.
 556 # I don't see any problem with shifting when the day starts for
 557 # retention, so I'm using long-iso.
 558
 559 # note to create a long-iso timestamp: date +%Y%m%dT%H%M%S%z