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=$(getopt -l cron,pull-reexec,help ceil:m:npqrs:t:vh "$@") || 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     -r) archive=false; shift ;;
 103     -q) verbose=false; verbose_arg=; progress_arg=; shift ;;
 104     # source host to receive a backup from
 105     -s)
 106       source=$2
 107       bbksource=$source
 108       if [[ $source == *:* ]]; then
 109         bbksource="[$source]"
 110       fi
 111       shift 2
 112       ;;
 113     # target hosts to send to. empty is valid for just doing local
 114     # snapshot. we have default hosts we will populate.
 115     -t) IFS=, targets=($2); unset IFS; shift 2 ;;
 116     -v) verbose=true; verbose_arg=-v; shift ;;
 117     -h|--help) usage ;;
 118     --) shift; break ;;
 119     *) die "Internal error!" ;;
 120   esac
 121 done
 122
 123 # only tested commands are resume and archive
 124 cmd_arg=${1:-run}
 125
 126 case $cmd_arg in
 127   run|resume|archive) : ;;
 128   *) die "untested command arg" ;;
 129 esac
 130
 131 if (( $# > 1 )); then
 132   die: "only 1 nonoption arg is supported"
 133 fi
 134
 135 if [[ -v targets && $source ]]; then
 136   # note, this doesnt need to be the case, but
 137   # we would need to think about it.
 138   die "error: -t and -s are mutually exclusive"
 139 fi
 140
 141 if $verbose; then
 142   printf "options: conf_only=%s\ndry_run=%s\nrate_limit=%s\nverbose=%s\ncmd_arg=%s" "$conf_only" "$dry_run" "$rate_limit" "$verbose" "$cmd_arg"
 143 fi
 144 ### end options parsing
 145
 146 # remove path from earlier version of btrbk
 147 rm -f /usr/sbin/btrbk
 148 # note, this still works as intended if there is no /usr/bin/btrbk
 149 if [[ /a/opt/btrbk/btrbk -nt /usr/bin/btrbk ]]; then
 150   if [[ -e /b/distro-functions/src/package-manager-abstractions ]]; then
 151     . /b/distro-functions/src/package-manager-abstractions
 152     pi asciidoctor
 153   fi
 154   cd /a/opt/btrbk
 155   m make install
 156 fi
 157
 158 # TODO: i wonder if there should be an option to send to the default
 159 # targets, plus any given on the command line.
 160
 161
 162 amy=false
 163 # set default targets
 164 if [[ ! -v targets && ! $source ]]; then
 165   if [[ $HOSTNAME != "$MAIL_HOST" ]] && $cron ; then
 166     echo "MAIL_HOST=$MAIL_HOST, nothing to do"
 167     mexit 0
 168   else
 169     amy=true
 170   fi
 171
 172   at_work=false
 173
 174   targets=(frodo.b8.nz)
 175   case $HOSTNAME in
 176     x2|kw)
 177       at_work=true
 178       ;;&
 179     x2|x3|sy|bo)
 180       if ping -q -c1 -w1 hal.office.fsf.org \
 181           && ip n show 192.168.0.26 | grep . &>/dev/null; then
 182         at_work=true
 183       fi
 184       ;;&
 185     kw|x2|x3|sy|bo)
 186       if $at_work; then
 187         if ping -q -c1 -w1 iank.vpn.office.fsf.org &>/dev/null; then
 188           home=iank.vpn.office.fsf.org
 189         else
 190           home=i.b8.nz
 191         fi
 192       else
 193         home=b8.nz
 194       fi
 195       ;;&
 196     kw)
 197       targets+=($home x2.office.fsf.org)
 198       ;;
 199     x2|x3|sy|bo)
 200       targets+=($home)
 201       if $at_work; then
 202         targets+=(x2.office.fsf.org x2.b8.nz)
 203       else
 204         targets+=(x2wg.b8.nz)
 205       fi
 206       ;;
 207     kd)
 208       targets+=(x2wg.b8.nz x3.b8.nz)
 209       if ping -q -c1 -w1 sy.b8.nz &>/dev/null; then
 210         targets+=(sy.b8.nz)
 211       else
 212         targets+=(syw.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=()
 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         # note: put q last just in case its specific retention options were to
 268         # affect other config sections. I havent tested if that is the case.
 269         prospective_mps+=(/a /ar /qr /q)
 270         ;;
 271     esac
 272   fi
 273   for mp in ${prospective_mps[@]}; do # default mountpoints to sync
 274     if [[ -e /nocow/btrfs-stale/$mp ]]; then
 275       e "warning: $mp stale, not adding to default mountpoints"
 276       continue
 277     fi
 278     if awk '{print $2}' /etc/fstab | grep -xF $mp &>/dev/null; then
 279       mountpoints+=($mp)
 280     fi
 281   done
 282 fi
 283
 284 echo "mountpoints: ${mountpoints[*]}"
 285
 286 ##### end command line parsing ########
 287
 288 if [[ $source ]]; then
 289   if [[ $(ssh $source ps --no-headers -o comm 1) == systemd ]]; then
 290     status=$(ssh $source systemctl is-active btrbk.service) || : # normally returns 3
 291     case $status in
 292       inactive|failed) : ;;
 293       *)
 294         echo "$0: error: cron btrbk is running on source. exiting out of caution"
 295         mexit 1
 296     esac
 297   fi
 298 fi
 299
 300 # pull_reexec stops us from getting into an infinite loop if there is some
 301 # kind of weird problem
 302 pulla=false
 303 for m in "${mountpoints[@]}"; do
 304   if [[ $m == /a ]]; then
 305     pulla=true
 306     break
 307   fi
 308 done
 309 if ! $pull_reexec && [[ $source ]] && $pulla ; then
 310   tmpf=$(mktemp)
 311   m rsync -ra $source:/usr/local/bin/{mount-latest-subvol,check-subvol-stale} /usr/local/bin
 312   m rsync -ra $source:/usr/local/lib/err /usr/local/lib
 313   m scp $source:/a/bin/distro-setup/btrbk-run $tmpf
 314   if ! diff -q $tmpf ${BASH_SOURCE[0]}; then
 315     e "found different version on host $source. reexecing"
 316     install -T $tmpf /usr/local/bin/btrbk-run
 317     m /usr/local/bin/btrbk-run --pull-reexec "${orig_args[@]}"
 318     mexit 0
 319   fi
 320 fi
 321
 322
 323 if ! command -v btrbk &>/dev/null; then
 324   die "error: no btrbk binary found"
 325 fi
 326 # if our mountpoints are from stale snapshots,
 327 # it doesn't make sense to do a backup.
 328 check-subvol-stale ${mountpoints[@]} || die "found stale mountpoints in ${mountpoints[*]}"
 329
 330 # for an initial run, btrbk requires the dir to exist.
 331 mkdir -p /mnt/{root,o}/btrbk
 332 local_zone=$(date +%z)
 333
 334 if [[ $source ]]; then
 335   if ! zone=$(ssh root@$source date +%z); then
 336     die failed to ssh to root@$source
 337   fi
 338   if [[ $zone != "$local_zone" ]]; then
 339     die "error: dont confuse yourself with multiple time zones. $h has different timezone than localhost"
 340   fi
 341
 342 else
 343
 344   sshable=()
 345   sshfail=()
 346   min_idle_ms=$((1000 * 60 * 15))
 347   for h in ${targets[@]}; do
 348     if remote_info=( $(timeout -s 9 6 ssh root@$h "mkdir -p /mnt/root/btrbk /mnt/o/btrbk && date +%z && df --output=size,pcent / | tail -n1") ); then
 349       zone=${remote_info[0]}
 350       root_size=${remote_info[1]}
 351       percent_used=${remote_info[2]%%%}
 352
 353       if (( ${#remote_info[@]} != 3 )); then
 354         die "error: didnt get 3 fields in test ssh to target $h. investigate"
 355       fi
 356
 357
 358       # we may be booted into a bootstrap fs or something
 359       min_root_kb=$(( 1024 * 1024 * 200 )) # 200 gb
 360       if (( root_size < min_root_kb )); then
 361         continue
 362       fi
 363
 364       if (( percent_used >= 98 )); then
 365         die "error: filesystem on target $h is $percent_used % full"
 366       fi
 367
 368       # This is a separate ssh because xprintidle can fail and thats ok.
 369       if $cron && idle_ms=$(timeout -s 9 6 ssh $h DISPLAY=:0 xprintidle); then
 370         if (( idle_ms < min_idle_ms )); then
 371
 372           # Ignore this host. i sometimes use a non-main machine for
 373           # testing or web browsing, knowing that everything will be wiped
 374           # by the next backup, but I dont want it to happen as Im using
 375           # it from cronjob.
 376           e "warning: $h: active X session in the last 15 minutes, skipping for now"
 377           continue
 378         fi
 379       fi
 380       sshable+=($h)
 381       if [[ $zone != "$local_zone" ]]; then
 382         die "error: dont confuse yourself with multiple time zones. $h has different timezone than localhost"
 383       fi
 384     else
 385       sshfail+=($h)
 386     fi
 387   done
 388   if [[ ! ${sshable[*]} ]] || { ! $cron && [[ ${sshfail[*]} ]]; }; then
 389     die "failed to ssh to hosts: ${sshfail[*]}"
 390   else
 391     if [[ ${sshfail[*]} ]]; then
 392       ret=1
 393       e "error: failed to ssh to ${sshfail[*]} but continuing with other hosts"
 394     fi
 395     targets=(${sshable[@]})
 396   fi
 397 fi
 398
 399
 400 cat >/etc/btrbk.conf <<EOF
 401 ssh_identity /q/root/h
 402 #ssh_identity /root/.ssh/home
 403
 404 # Just a guess that local7 is a good facility to pick.
 405 # It's a bit odd that the transaction log has to be logged to
 406 # a file or syslog, while other output is sent to std out.
 407 # The man does not mention a way for them to be together, but
 408 # I dunno if setting a log level like warn might also output
 409 # transaction info.
 410 transaction_syslog local7
 411
 412 # trying this out
 413 stream_compress zstd
 414
 415 # so we only run one at a time
 416 lockfile                   /var/lock/btrbk.lock
 417
 418 # default format of short does not accomidate hourly preservation setting
 419 timestamp_format long-iso
 420
 421 # only make a snapshot if things have changed
 422 snapshot_create onchange
 423 # I could make this different from target_preserve,
 424 # if one disk had less space.
 425 # for now, keeping them equal.
 426 snapshot_preserve 18h 14d 8w 24m
 427 snapshot_preserve_min 2h
 428 snapshot_dir btrbk
 429 # so, total backups = ~58
 430 target_preserve 18h 14d 8w 24m
 431 target_preserve_min 2h
 432
 433 # if something fails and it's not obvious, try doing
 434 # btrbk -l debug -v dryrun
 435
 436 rate_limit $rate_limit
 437 EOF
 438
 439 if $incremental_strict; then
 440   cat >>/etc/btrbk.conf <<EOF
 441 incremental strict
 442 EOF
 443 fi
 444
 445 for m in ${mountpoints[@]}; do
 446   case $m in
 447     /o)
 448       vol=/mnt/o
 449       ;;
 450     *)
 451       vol=/mnt/root
 452       ;;
 453   esac
 454
 455   sub=${m#/}
 456   if [[ $source ]]; then
 457     cat >>/etc/btrbk.conf <<EOF
 458 volume ssh://$bbksource$vol
 459 subvolume $sub
 460 target send-receive $vol/btrbk
 461 EOF
 462   fi
 463   if (( ${#targets[@]} )); then
 464     cat >>/etc/btrbk.conf <<EOF
 465 volume $vol
 466 subvolume $sub
 467 EOF
 468     case $sub in
 469       q)
 470         # q has sensitive data i dont want to backup for so long
 471         cat >>/etc/btrbk.conf <<EOF
 472 snapshot_preserve 18h 14d
 473 snapshot_preserve_min 2h
 474 snapshot_dir btrbk
 475 target_preserve 18h 14d
 476 target_preserve_min 2h
 477 EOF
 478         ;;
 479     esac
 480     for tg in ${targets[@]}; do
 481       # handle ipv6
 482       if [[ $tg == *:* ]]; then
 483         tg="[$tg]"
 484       fi
 485       cat >>/etc/btrbk.conf <<EOF
 486 target send-receive ssh://$tg$vol/btrbk
 487 EOF
 488     done
 489   fi
 490 done
 491
 492 # if $amy; then
 493 #   # to manually backup amy,
 494 #   # bbk -e -s amy -m root_ubuntubionic
 495 #   cat >>/etc/btrbk.conf <<'EOF'
 496 # volume ssh://amy/mnt/root
 497 # subvolume root_ubuntubionic
 498 # target send-receive /mnt/root/btrbk
 499 # EOF
 500 # fi
 501
 502 # todo: umount first to ensure we don't have any errors
 503 # todo: do some kill fuser stuff to make umount more reliable
 504
 505
 506 if $conf_only; then
 507   mexit 0
 508 fi
 509
 510
 511
 512 if $dry_run; then
 513   m btrbk -v -n $cmd_arg
 514   mexit 0
 515 elif [[ $cmd_arg == archive ]]; then
 516   if [[ $source ]]; then
 517     m btrbk $verbose_arg $progress_arg $cmd_arg ssh://$source$vol $vol
 518   else
 519     for tg in ${targets[@]}; do
 520       m btrbk $verbose_arg $progress_arg $cmd_arg $vol ssh://$tg$vol
 521     done
 522   fi
 523   mexit 0
 524 fi
 525 # -q and just using the syslog option seemed nice,
 526 # but it doesn't show when a send has a parent and when it doesn't.
 527 m btrbk $verbose_arg $progress_arg $cmd_arg
 528
 529 # todo: tp not valid anymore.
 530 # if we have it, sync to systems which don't
 531 if mountpoint $rsync_mountpoint >/dev/null; then
 532   for tg in ${targets[@]}; do
 533     case $tg in
 534       tp)
 535         dirs=(/p/c/machine_specific/tp)
 536         for x in /p/c/machine_specific/*.hosts; do
 537           if grep -qxF $tg $x; then
 538             dirs+=(${x%.hosts})
 539           fi
 540         done
 541         m rsync -aSAXPH --specials --devices --delete --relative ${dirs[@]} root@$tg:/
 542         ;;
 543     esac
 544   done
 545 fi
 546
 547 if [[ $source ]]; then
 548   m mount-latest-subvol
 549 else
 550   m /a/exe/mount-latest-remote ${targets[@]}
 551 fi
 552
 553 if [[ $ret == 0 ]]; then
 554   for tg in ${targets[@]}; do
 555     :
 556     #ssh root@$tg /a/exe/mail-backup-clean
 557   done
 558 fi
 559
 560 mexit $ret
 561
 562 # todo: move variable data we don't care about backing up
 563 # to /nocow and symlink it.
 564
 565
 566 # background on btrbk timezones. with short/long, timestamps use local time.
 567 # for long, if your local time moves backwards, by moving timezones or
 568 # for an hour when daylight savings changes it, you will temporarily get
 569 # a more aggressive retention policy for the overlapping period, and
 570 # vice versa for the opposite timezone move. The alternative is using
 571 # long-iso, which puts timezone info into the timestamp, which means
 572 # that instead of shifting time, you shift the start of day/week/month
 573 # which is used for retention to your new local time, which means for
 574 # example, if you moved forward by 8 hours, the daily/weekly/monthly
 575 # retention will be 8 hours more aggressive since midnight is at a new
 576 # time, unless you fake the timzeone using the TZ env variable.
 577 # However, in the short term, there will be no inconsistencies.
 578 # I don't see any problem with shifting when the day starts for
 579 # retention, so I'm using long-iso.
 580
 581 # note to create a long-iso timestamp: date +%Y%m%dT%H%M%S%z