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] [run|resume|archive]
  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 $1
  39 }
  40
  41
  42
  43 script_name="${BASH_SOURCE[0]}"
  44 script_name="${script_name##*/}"
  45 pre="${SSH_CLIENT:+$HOSTNAME} $script_name:"
  46 m() { if $verbose; then printf "$pre%s\n"  "$*"; fi;  "$@"; }
  47 e() { printf "$pre%s\n"  "$*"; }
  48 die() { printf "$pre%s\n" "$*" >&2; echo "exiting with status 1" >&2; exit 1; }
  49 mexit() { echo "$pre: exiting with status $1"; exit $1; }
  50
  51 # latest $MAIL_HOST
  52 if [[ -e /b/bash_unpublished/source-state ]]; then
  53   source /b/bash_unpublished/source-state
  54 fi
  55
  56 # note q is owned by root:1000
  57
  58 mountpoints=()
  59
  60 rsync_mountpoint=/q
  61
  62 ret=0
  63 # default options
  64 conf_only=false
  65 dry_run=false # mostly for testing
  66 rate_limit=no
  67 verbose=true; verbose_arg=-v
  68 if [[ $INVOCATION_ID ]]; then
  69   # INVOCATION_ID means running as a systemd service. we cant show progress in this case,
  70   # but if we pass the arg, it will insert mbuffer into the command.
  71   progress_arg=
  72 else
  73   progress_arg="--progress"
  74 fi
  75 incremental_strict=false
  76 pull_reexec=false
  77
  78 default_args_file=/etc/btrbk-run.conf
  79 if [[ -s $default_args_file ]]; then
  80   set -- $(< $default_args_file) "$@"
  81   # i havent used this feature yet, so warn about it
  82   echo "$0: warning: default btrbk-run options set in $default_args_file (sleeping 5 seconds):"
  83   cat $default_args_file
  84   sleep 5
  85 fi
  86
  87 targets=()
  88 early=false
  89 cron=false
  90 orig_args=("$@")
  91 temp=$(getopt -l cron,pull-reexec,help 23ceil:m:npqrs:t:vh "$@") || usage 1
  92 eval set -- "$temp"
  93 while true; do
  94   case $1 in
  95     # some behaviors specific to running under cron:
  96     # - skip hosts where xprintidle haven't been idle recently
  97     # - if we can't ssh to 1 or more hosts, still do the rest
  98     # - if we aren't MAIL_HOST and no -m or -s, just exit
  99     --cron)
 100       cron=true
 101       pre=
 102       ;;
 103     # for the rare case we want to run multiple instances at the same time
 104     -2) conf_suf=2 ;;
 105     -3) conf_suf=3 ;;
 106     # only creates the config file, does not run btrbk
 107     -c) conf_only=true ;;
 108     # quit early, just btrbk, no extra remounting etc.
 109     -e) early=true ;;
 110     -i) incremental_strict=true ;;
 111     # bytes per second, suffix k m g
 112     -l) rate_limit=$2; shift ;;
 113     # Comma separated mountpoints to backup. This has defaults set below.
 114     -m) IFS=, mountpoints=($2); unset IFS; shift ;;
 115     -n) dry_run=true ;;
 116     # hide progress
 117     -p) progress_arg= ;;
 118     # internal option for rerunning under newer SOURCE_HOST version.
 119     --pull-reexec) pull_reexec=true;;
 120     # quiet
 121     -q) verbose=false; verbose_arg=; progress_arg= ;;
 122     # source host to receive a backup from
 123     -s)
 124       source=$2
 125       bbksource=$source
 126       if [[ $source == *:* ]]; then
 127         bbksource="[$source]"
 128       fi
 129       shift
 130       ;;
 131     # target hosts to send to. empty is valid for just doing local
 132     # snapshot. we have default hosts we will populate.
 133     -t) IFS=, targets=($2); unset IFS; shift ;;
 134     # verbose.
 135     -v) verbose=true; verbose_arg=-v ;;
 136     -h|--help) usage ;;
 137     --) shift; break ;;
 138     *) die "Internal error!" ;;
 139   esac
 140   shift
 141 done
 142
 143 cmd_arg=${1:-run}
 144
 145
 146 std_preserve="36h 14d 8w 24m"
 147 q_preserve="18h 14d 8w"
 148
 149 case $cmd_arg in
 150   run|resume) : ;;
 151
 152   # This works better than the normal archive command. We have to
 153   # specify the mount points, but that is what we are used to doing and
 154   # we prefer it. Another difference is that archive works recursively
 155   # and we don't care about that. Sometimes we may still want to run
 156   # btrbk archive, but it doesn't even use the config file, so just
 157   # run it directly, eg:
 158   # time s btrbk -v archive /mnt/r7/amy/boot/btrbk ssh://bo/mnt/boot2/btrbk
 159   archive)
 160     cmd_arg=resume
 161     std_preserve="999h 999d 999w 999m"
 162     q_preserve="$std_preserve"
 163     preserve_arg=-p
 164     ;;
 165   *) die "untested command arg" ;;
 166 esac
 167
 168 if (( $# > 1 )); then
 169   die: "only 1 nonoption arg is supported"
 170 fi
 171
 172 if [[ -v targets && $source ]]; then
 173   # note, this doesnt need to be the case, but
 174   # we would need to think about it.
 175   die "error: -t and -s are mutually exclusive"
 176 fi
 177
 178 if $verbose; then
 179   printf "$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"
 180 fi
 181 ### end options parsing
 182
 183 # remove path from earlier version of btrbk
 184 rm -f /usr/sbin/btrbk
 185 # note, this still works as intended if there is no /usr/bin/btrbk
 186 if [[ /a/opt/btrbk/btrbk -nt /usr/bin/btrbk ]]; then
 187   if [[ -e /b/distro-functions/src/package-manager-abstractions ]]; then
 188     . /b/distro-functions/src/package-manager-abstractions
 189     pi asciidoctor
 190   fi
 191   cd /a/opt/btrbk
 192   m make install
 193 fi
 194
 195 # TODO: i wonder if there should be an option to send to the default
 196 # targets, plus any given on the command line.
 197
 198
 199
 200 kd_spread=false
 201 # set default targets
 202 if [[ ! -v targets && ! $source ]]; then
 203   if $cron; then
 204     if [[ $HOSTNAME != "$MAIL_HOST" ]]; then
 205       if [[ $HOSTNAME == kd && $MAIL_HOST = x2 ]]; then
 206         kd_spread=true
 207       else
 208         echo "MAIL_HOST=$MAIL_HOST, nothing to do"
 209         mexit 0
 210       fi
 211     fi
 212   fi
 213
 214   # x2 at home atm
 215   kd_spread=false
 216
 217   at_work=false
 218   at_home=false
 219
 220   case $HOSTNAME in
 221     kw|kd|frodo|x2|x3|sy) : ;;
 222     *)
 223       die "error: no default targets for this host, use -t"
 224       ;;
 225   esac
 226
 227   case $HOSTNAME in
 228     kw)
 229       at_work=true
 230       ;;&
 231     kd|frodo)
 232       at_home=true
 233       ;;&
 234     x2|x3|sy)
 235       if [[ $(dig +short @10.2.0.1 -x 10.2.0.2 2>&1 ||:) == kd.b8.nz. ]] \
 236            && ip n show 10.2.0.1 | grep . &>/dev/null; then
 237         at_home=true
 238       elif ping -q -c1 -w1 hal.office.fsf.org &>/dev/null \
 239           && ip n show 192.168.0.26 | grep . &>/dev/null; then
 240         at_work=true
 241       fi
 242       ;;&
 243     *)
 244       if $at_home; then
 245         # main work machine
 246         if ping -q -c1 -w1 x3.office.fsf.org &>/dev/null; then
 247           targets+=(x3.office.fsf.org)
 248         else
 249           targets+=(x3wg.b8.nz)
 250         fi
 251         for h in frodo kd; do
 252           if [[ $HOSTNAME == "$h" ]]; then
 253             continue
 254           fi
 255           targets+=($h.b8.nz)
 256         done
 257         for h in x2 x3 sy; do
 258           if [[ $HOSTNAME == "$h" ]]; then
 259             continue
 260           fi
 261           if ping -q -c1 -w1 $h.b8.nz &>/dev/null; then
 262             targets+=($h.b8.nz)
 263           elif ping -q -c1 -w1 ${h}w.b8.nz &>/dev/null; then
 264             targets+=(${h}w.b8.nz)
 265           fi
 266         done
 267       elif $at_work; then
 268         if ping -q -c1 -w1 iank.vpn.office.fsf.org &>/dev/null; then
 269           targets+=(iank.vpn.office.fsf.org)
 270         else
 271           targets+=(i.b8.nz)
 272         fi
 273         for h in x2 x3 kw; do
 274           if [[ $HOSTNAME == "$h" ]]; then
 275             continue
 276           fi
 277           if ping -q -c1 -w1 $h.office.fsf.org &>/dev/null; then
 278             targets+=($h.office.fsf.org)
 279           fi
 280         done
 281       else
 282         targets+=(i.b8.nz)
 283       fi
 284       ;;
 285   esac
 286 fi
 287
 288 if [[ -v targets ]]; then
 289   echo "targets: ${targets[*]}"
 290 fi
 291
 292 if [[ $source ]]; then
 293   echo "source: $source"
 294 fi
 295
 296 if [[ ${mountpoints[0]} ]]; then
 297   for mp in ${mountpoints[@]}; do
 298     if [[ -e /nocow/btrfs-stale/$mp ]]; then
 299       die "error: $mp is stale, mount-latest-subvol first"
 300     fi
 301   done
 302 else
 303   # set default mountpoints
 304   if [[ ${targets[0]} == tp ]]; then
 305     prospective_mps=(/a)
 306   else
 307     case $HOSTNAME in
 308       *)
 309         prospective_mps=()
 310         if [[ $source ]]; then
 311           source_state="$(ssh $source cat /a/bin/bash_unpublished/source-state)"
 312           eval "$source_state"
 313           source_host="$(ssh $source cat /etc/hostname)"
 314           if [[ $source_host == "$MAIL_HOST" ]]; then
 315             prospective_mps+=(/o)
 316           fi
 317           if [[ $source_host == "$HOST2" ]]; then
 318             prospective_mps+=(/a /ar /qr /q)
 319           fi
 320         else
 321           if [[ $HOSTNAME == "$MAIL_HOST" ]]; then
 322             prospective_mps+=(/o)
 323           fi
 324           if [[ $HOSTNAME == "$HOST2" ]]; then
 325             prospective_mps+=(/a /ar /qr /q)
 326           fi
 327         fi
 328         # note: put q last just in case its specific retention options were to
 329         # affect other config sections. I havent tested if that is the case.
 330         ;;
 331     esac
 332   fi
 333   for mp in ${prospective_mps[@]}; do # default mountpoints to sync
 334     if [[ -e /nocow/btrfs-stale/$mp ]]; then
 335       e "warning: $mp stale, not adding to default mountpoints"
 336       continue
 337     fi
 338     if awk '{print $2}' /etc/fstab | grep -xF $mp &>/dev/null; then
 339       mountpoints+=($mp)
 340     fi
 341   done
 342 fi
 343
 344 echo "mountpoints: ${mountpoints[*]}"
 345
 346 ##### end command line parsing ########
 347
 348 if [[ $source ]]; then
 349   if [[ $(ssh $source ps --no-headers -o comm 1) == systemd ]]; then
 350     status=$(ssh $source systemctl is-active btrbk.service) || : # normally returns 3
 351     case $status in
 352       inactive|failed) : ;;
 353       *)
 354         echo "$0: error: cron btrbk is running on source. exiting out of caution"
 355         mexit 1
 356     esac
 357   fi
 358 fi
 359
 360 # pull_reexec stops us from getting into an infinite loop if there is some
 361 # kind of weird problem
 362 pulla=false
 363 for m in "${mountpoints[@]}"; do
 364   if [[ $m == /a ]]; then
 365     pulla=true
 366     break
 367   fi
 368 done
 369 if ! $pull_reexec && [[ $source ]] && $pulla ; then
 370   tmpf=$(mktemp)
 371   m rsync -ra $source:/usr/local/bin/{mount-latest-subvol,check-subvol-stale} /usr/local/bin
 372   m rsync -ra $source:/usr/local/lib/err /usr/local/lib
 373   m scp $source:/a/bin/distro-setup/btrbk-run $tmpf
 374   if ! diff -q $tmpf ${BASH_SOURCE[0]}; then
 375     e "found different version on host $source. reexecing"
 376     install -T $tmpf /usr/local/bin/btrbk-run
 377     m /usr/local/bin/btrbk-run --pull-reexec "${orig_args[@]}"
 378     mexit 0
 379   fi
 380 fi
 381
 382
 383 if ! command -v btrbk &>/dev/null; then
 384   die "error: no btrbk binary found"
 385 fi
 386 # if our mountpoints are from stale snapshots,
 387 # it doesn't make sense to do a backup.
 388 m check-subvol-stale ${mountpoints[@]} || die "found stale mountpoints in ${mountpoints[*]}"
 389
 390 # for an initial run, btrbk requires the dir to exist.
 391 mkdir -p /mnt/{root,o}/btrbk
 392 local_zone=$(date +%z)
 393
 394 if [[ $source ]]; then
 395   if ! zone=$(ssh root@$source date +%z); then
 396     if $conf_only; then
 397       echo "$0: warning: failed to ssh to root@$source"
 398     else
 399       die failed to ssh to root@$source
 400     fi
 401   fi
 402   if [[ $zone != "$local_zone" ]]; then
 403     die "error: dont confuse yourself with multiple time zones. $h has different timezone than localhost"
 404   fi
 405
 406 else
 407
 408   sshable=()
 409   sshfail=()
 410   min_idle_ms=$((1000 * 60 * 15))
 411   for h in ${targets[@]}; do
 412     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
 413       zone=${remote_info[0]}
 414       root_size=${remote_info[1]}
 415       percent_used=${remote_info[2]%%%}
 416
 417       if (( ${#remote_info[@]} != 3 )); then
 418         die "error: didnt get 3 fields in test ssh to target $h. investigate"
 419       fi
 420     elif $conf_only; then
 421       # Use some typical values in this case
 422       root_size=$(( 1024 * 1024 * 2000 )) #2tb
 423       percent_used=10
 424       zone=$(date +%z)
 425     else
 426       sshfail+=($h)
 427       continue
 428     fi
 429
 430     # we may be booted into a bootstrap fs or something
 431     min_root_kb=$(( 1024 * 1024 * 200 )) # 200 gb
 432     if (( root_size < min_root_kb )); then
 433       continue
 434     fi
 435
 436     if (( percent_used >= 98 )); then
 437       die "error: filesystem on target $h is $percent_used % full"
 438     fi
 439
 440     # This is a separate ssh because xprintidle can fail and thats ok.
 441     if $cron && idle_ms=$(timeout -s 9 6 ssh $h DISPLAY=:0 xprintidle); then
 442       if (( idle_ms < min_idle_ms )); then
 443
 444         # Ignore this host. i sometimes use a non-main machine for
 445         # testing or web browsing, knowing that everything will be wiped
 446         # by the next backup, but I dont want it to happen as Im using
 447         # it from cronjob.
 448         e "warning: $h: active X session in the last 15 minutes, skipping for now"
 449         continue
 450       fi
 451     fi
 452     sshable+=($h)
 453     if [[ $zone != "$local_zone" ]]; then
 454       die "error: dont confuse yourself with multiple time zones. $h has different timezone than localhost"
 455     fi
 456   done
 457   if [[ ! ${sshable[*]} ]] || { ! $cron && [[ ${sshfail[*]} ]]; }; then
 458     die "failed to ssh to hosts: ${sshfail[*]}"
 459   else
 460     if [[ ${sshfail[*]} ]]; then
 461       ret=1
 462       e "error: failed to ssh to ${sshfail[*]} but continuing with other hosts"
 463     fi
 464     targets=(${sshable[@]})
 465   fi
 466 fi
 467
 468
 469 cat >/etc/btrbk$conf_suf.conf <<EOF
 470 ssh_identity /q/root/h
 471 #ssh_identity /root/.ssh/home
 472
 473 # Just a guess that local7 is a good facility to pick.
 474 # It's a bit odd that the transaction log has to be logged to
 475 # a file or syslog, while other output is sent to std out.
 476 # The man does not mention a way for them to be together, but
 477 # I dunno if setting a log level like warn might also output
 478 # transaction info.
 479 transaction_syslog local7
 480
 481 # trying this out
 482 #stream_compress zstd
 483
 484 # so we only run one at a time
 485 lockfile                   /var/lock/btrbk$conf_suf.lock
 486
 487 # default format of short does not accomidate hourly preservation setting
 488 timestamp_format long-iso
 489
 490 # only make a snapshot if things have changed
 491 snapshot_create onchange
 492 # I could make this different from target_preserve,
 493 # if one disk had less space.
 494 # for now, keeping them equal.
 495 snapshot_preserve $std_preserve
 496 snapshot_preserve_min 2h
 497 snapshot_dir btrbk
 498 # so, total backups = ~58
 499 target_preserve $std_preserve
 500 target_preserve_min 2h
 501
 502 # if something fails and it's not obvious, try doing
 503 # btrbk -l debug -v dryrun
 504
 505 rate_limit $rate_limit
 506 EOF
 507
 508 if $incremental_strict; then
 509   cat >>/etc/btrbk$conf_suf.conf <<EOF
 510 incremental strict
 511 EOF
 512 fi
 513
 514 qconf() {
 515   case $sub in
 516     q)
 517       # q has sensitive data i dont want to backup for so long
 518       cat >>/etc/btrbk$conf_suf.conf <<EOF
 519 snapshot_preserve $q_preserve
 520 snapshot_preserve_min 2h
 521 snapshot_dir btrbk
 522 target_preserve $q_preserve
 523 target_preserve_min 2h
 524 EOF
 525       ;;
 526   esac
 527
 528 }
 529
 530 # make /q be last
 531 mp_count=${#mountpoints[@]}
 532 for (( i=0; i < mp_count - 1 ; i++ )); do
 533   if [[ ${mountpoints[i]} == /q ]]; then
 534     unset mountpoints[i]
 535     mountpoints+=(/q)
 536   fi
 537 done
 538
 539 for m in ${mountpoints[@]}; do
 540   case $m in
 541     /o)
 542       vol=/mnt/o
 543       ;;
 544     *)
 545       vol=/mnt/root
 546       ;;
 547   esac
 548
 549   sub=${m#/}
 550   if [[ $source ]]; then
 551     cat >>/etc/btrbk$conf_suf.conf <<EOF
 552 volume ssh://$bbksource$vol
 553 subvolume $sub
 554 EOF
 555     qconf
 556     cat >>/etc/btrbk$conf_suf.conf <<EOF
 557 target send-receive $vol/btrbk
 558 EOF
 559   fi
 560   if (( ${#targets[@]} )); then
 561     cat >>/etc/btrbk$conf_suf.conf <<EOF
 562 volume $vol
 563 subvolume $sub
 564 EOF
 565     qconf
 566     for tg in ${targets[@]}; do
 567       # handle ipv6
 568       if [[ $tg == *:* ]]; then
 569         tg="[$tg]"
 570       fi
 571       cat >>/etc/btrbk$conf_suf.conf <<EOF
 572 target send-receive ssh://$tg$vol/btrbk
 573 EOF
 574     done
 575   fi
 576 done
 577
 578 # todo: umount first to ensure we don't have any errors
 579 # todo: do some kill fuser stuff to make umount more reliable
 580
 581
 582 if $conf_only; then
 583   mexit 0
 584 fi
 585
 586
 587
 588 if $dry_run; then
 589   m btrbk -c /etc/btrbk$conf_suf.conf -v -n $cmd_arg
 590   mexit 0
 591 fi
 592 # -q and just using the syslog option seemed nice,
 593 # but it doesn't show when a send has a parent and when it doesn't.
 594 m btrbk -c /etc/btrbk$conf_suf.conf $preserve_arg $verbose_arg $progress_arg $cmd_arg
 595
 596 if $early; then
 597   exit 0
 598 fi
 599
 600 # todo: tp not valid anymore.
 601 # if we have it, sync to systems which don't
 602 if mountpoint $rsync_mountpoint >/dev/null; then
 603   for tg in ${targets[@]}; do
 604     case $tg in
 605       tp)
 606         dirs=(/p/c/machine_specific/tp)
 607         for x in /p/c/machine_specific/*.hosts; do
 608           if grep -qxF $tg $x; then
 609             dirs+=(${x%.hosts})
 610           fi
 611         done
 612         m rsync -aSAXPH --specials --devices --delete --relative ${dirs[@]} root@$tg:/
 613         ;;
 614     esac
 615   done
 616 fi
 617
 618 if [[ $source ]]; then
 619   m mount-latest-subvol
 620 else
 621   m /a/exe/mount-latest-remote ${targets[@]}
 622 fi
 623
 624 if [[ $ret == 0 ]]; then
 625   for tg in ${targets[@]}; do
 626     :
 627     #ssh root@$tg /a/exe/mail-backup-clean
 628   done
 629 fi
 630
 631 mexit $ret
 632
 633 # todo: move variable data we don't care about backing up
 634 # to /nocow and symlink it.
 635
 636
 637 # background on btrbk timezones. with short/long, timestamps use local time.
 638 # for long, if your local time moves backwards, by moving timezones or
 639 # for an hour when daylight savings changes it, you will temporarily get
 640 # a more aggressive retention policy for the overlapping period, and
 641 # vice versa for the opposite timezone move. The alternative is using
 642 # long-iso, which puts timezone info into the timestamp, which means
 643 # that instead of shifting time, you shift the start of day/week/month
 644 # which is used for retention to your new local time, which means for
 645 # example, if you moved forward by 8 hours, the daily/weekly/monthly
 646 # retention will be 8 hours more aggressive since midnight is at a new
 647 # time, unless you fake the timzeone using the TZ env variable.
 648 # However, in the short term, there will be no inconsistencies.
 649 # I don't see any problem with shifting when the day starts for
 650 # retention, so I'm using long-iso.
 651
 652 # note to create a long-iso timestamp: date +%Y%m%dT%H%M%S%z