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
  20 [[ $EUID == 0 ]] || exec sudo -E "${BASH_SOURCE[0]}" "$@"
  21
  22 source /usr/local/lib/err
  23
  24 usage() {
  25   cat <<'EOF'
  26 btrbk-run [OPTIONS]
  27 usually -t TARGET_HOST or -s SOURCE_HOST
  28
  29 Note, at source location, intentionally not executable, run and read
  30 install-my-scripts.
  31
  32 EOF
  33   echo "top of script file:"
  34   sed -n '1,/^[# ]*end command line/{p;b};q' "$0"
  35   exit $1
  36 }
  37
  38 rsync-dirs() {
  39   local host=$1
  40   local path=$2
  41   m rsync $dry_run_arg -ahi --relative --delete "$path" "root@$host:/"
  42 }
  43
  44
  45 pre="${0##*/}: "
  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 progress_arg="--progress"
  69 incremental_strict=true
  70 pull_reexec=false
  71
  72 default_args_file=/etc/btrbk-run.conf
  73 if [[ -s $default_args_file ]]; then
  74   set -- $(< $default_args_file) "$@"
  75   # i havent used this feature yet, so warn about it
  76   echo "$0: warning: default btrbk-run options set in $default_args_file (sleeping 5 seconds):"
  77   cat $default_args_file
  78   sleep 5
  79 fi
  80
  81 cron=false
  82 orig_args=("$@")
  83 temp=$(getopt -l cron,pull-reexec,help cil:m:npqs:t:vh "$@") || usage 1
  84 eval set -- "$temp"
  85 while true; do
  86   case $1 in
  87     --cron)
  88       cron=true
  89       pre=
  90       shift
  91       ;;
  92     # only creates the config file, does not run btrbk
  93     -c) conf_only=true; shift ;;
  94     -i) incremental_strict=false; 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) source=$2; shift 2 ;;
 105     # target hosts to send to. empty is valid for just doing local
 106     # snapshot. we have default hosts we will populate.
 107     -t) IFS=, targets=($2); unset IFS; shift 2 ;;
 108     -v) verbose=true; verbose_arg=-v; shift ;;
 109     -h|--help) usage ;;
 110     --) shift; break ;;
 111     *) die "Internal error!" ;;
 112   esac
 113 done
 114
 115 # only tested commands are resume and archive
 116 cmd_arg=${1:-run}
 117
 118 case $cmd_arg in
 119   run|resume|archive) : ;;
 120   *) die "untested command arg" ;;
 121 esac
 122
 123 if (( $# > 1 )); then
 124   die: "only 1 nonoption arg is supported"
 125 fi
 126
 127 if [[ -v targets && $source ]]; then
 128   die "error: -t and -s are mutually exclusive"
 129 fi
 130
 131 if $verbose; then
 132   printf "options: conf_only=%s\ndry_run=%s\nrate_limit=%s\nverbose=%s\ncmd_arg=%s" "$conf_only" "$dry_run" "$rate_limit" "$verbose" "$cmd_arg"
 133 fi
 134 ### end options parsing
 135
 136 if ! btrbk --version |& grep 0.29.0 >/dev/null; then
 137   cd /a/opt/btrbk
 138   m sudo make install
 139 fi
 140
 141 # TODO: i wonder if there should be an option to send to the default
 142 # targets, plus any given on the command line.
 143
 144 # set default targets
 145 if [[ ! -v targets && ! $source ]]; then
 146   if [[ $HOSTNAME != "$MAIL_HOST" ]] && $cron ; then
 147     echo "MAIL_HOST=$MAIL_HOST, nothing to do"
 148     mexit 0
 149   fi
 150
 151   at_work=false
 152
 153   case $HOSTNAME in
 154     kw)
 155       at_work=true
 156       ;;&
 157     x2|x3)
 158       if ping -q -c1 -w1 hal.office.fsf.org \
 159           && ip n show 192.168.0.26 | grep . &>/dev/null; then
 160         at_work=true
 161       fi
 162       ;;&
 163     kw|x2|x3)
 164       if $at_work && ping -q -c1 -w1 iank.vpn2.office.fsf.org &>/dev/null; then
 165         home=iank.vpn2.office.fsf.org
 166       else
 167         home=b8.nz
 168       fi
 169       ;;&
 170     kw)
 171       targets=($home x3)
 172       ;;
 173     x2|x3)
 174       targets=($home kw)
 175       ;;
 176     tp)
 177       # kd disabled temporarily while its hot and i plan to work on it.
 178       #targets=(frodo kd)
 179       #targets=(frodo x2.b8.nz)
 180       targets=(frodo)
 181       # might not be connected to the vpn
 182       if timeout -s 9 6 ssh kw :; then
 183         targets+=(kw)
 184       fi
 185       ;;
 186     kd)
 187       targets=(frodo tp)
 188       # might not be connected to the vpn
 189       if timeout -s 9 6 ssh kw :; then
 190         targets+=(kw)
 191       fi
 192       ;;
 193     *)
 194       die "error: no default targets for this host, use -t"
 195       ;;
 196   esac
 197 fi
 198
 199 if [[ -v targets ]]; then
 200   echo "targets: ${targets[*]}"
 201 fi
 202
 203 if [[ $source ]]; then
 204   echo "source: $source"
 205 fi
 206
 207
 208 if [[ $mountpoints ]]; then
 209   for mp in ${mountpoints[@]}; do # default mountpoints to sync
 210     if [[ -e /nocow/btrfs-stale/$mp ]]; then
 211       die "error: $mp is stale, mount-latest-subvol first"
 212     fi
 213   done
 214 else
 215   # set default mountpoints
 216   case $HOSTNAME in
 217     # no remote backups atm. note, if we do enable this, configuration below will need some changes.
 218     #  frodo)
 219     #    prospective_mps=(/i)
 220     #    ;;
 221     *)
 222       prospective_mps=(/a /q)
 223       if [[ $source ]]; then
 224         source_state="$(ssh $source cat /a/bin/bash_unpublished/source-state)"
 225         eval "$source_state"
 226         source_host="$(ssh $source cat /etc/hostname)"
 227         if [[ $source_host == "$MAIL_HOST" ]]; then
 228           prospective_mps+=(/o)
 229         fi
 230       else
 231         if [[ $HOSTNAME == "$MAIL_HOST" ]]; then
 232           prospective_mps+=(/o)
 233         fi
 234       fi
 235       ;;
 236   esac
 237   for mp in ${prospective_mps[@]}; do # default mountpoints to sync
 238     if [[ -e /nocow/btrfs-stale/$mp ]]; then
 239       e "warning: $mp stale, not adding to default mountpoints"
 240       continue
 241     fi
 242     if awk '{print $2}' /etc/fstab | grep -xF $mp &>/dev/null; then
 243       mountpoints+=($mp)
 244     fi
 245   done
 246 fi
 247
 248 echo "mountpoints: ${mountpoints[*]}"
 249
 250 ##### end command line parsing ########
 251
 252 if [[ $source ]]; then
 253   if [[ $(ssh $source systemctl is-active btrbk.service) != inactive ]]; then
 254     echo "$0: error: cron btrbk is running on source. exiting out of caution"
 255     mexit 1
 256   fi
 257 fi
 258
 259 # pull_reexec stops us from getting into an infinite loop if there is some
 260 # kind of weird problem
 261 pulla=false
 262 for m in "${mountpoints[@]}"; do
 263   if [[ $m == /a ]]; then
 264     pulla=true
 265     break
 266   fi
 267 done
 268 if ! $pull_reexec && [[ $source ]] && $pulla ; then
 269   tmpf=$(mktemp)
 270   scp $source:/a/bin/distro-setup/btrbk-run $tmpf
 271   if ! diff -q $tmpf $BASH_SOURCE; then
 272     e "found newer version on host $source. reexecing"
 273     install -T $tmpf /usr/local/bin/btrbk-run
 274     m /usr/local/bin/btrbk-run --pull-reexec "${orig_args[@]}"
 275     mexit 0
 276   fi
 277 fi
 278
 279
 280
 281
 282 if ! which btrbk &>/dev/null; then
 283   die "error: no btrbk binary found"
 284 fi
 285 # if our mountpoints are from stale snapshots,
 286 # it doesn't make sense to do a backup.
 287 check-subvol-stale ${mountpoints[@]} || die "found stale mountpoints in ${mountpoints[*]}"
 288
 289 # for an initial run, btrbk requires the dir to exist.
 290 mkdir -p /mnt/root/btrbk
 291 local_zone=$(date +%z)
 292
 293 if [[ $source ]]; then
 294   if ! zone=$(ssh root@$source date +%z); then
 295     die failed to ssh to root@$source
 296   fi
 297   if [[ $zone != $local_zone ]]; then
 298     die "error: dont confuse yourself with multiple time zones. $h has different timezone than localhost"
 299   fi
 300
 301 else
 302
 303   sshable=()
 304   sshfail=()
 305   min_idle_ms=$((1000 * 60 * 15))
 306   for h in ${targets[@]}; do
 307     if zone=$(timeout -s 9 6 ssh root@$h "mkdir -p /mnt/root/btrbk && date +%z"); then
 308       # This is a separate ssh because xprintidle can fail and thats ok.
 309       if $cron && idle_ms=$(timeout -s 9 6 ssh $h DISPLAY=:0 xprintidle); then
 310         if (( idle_ms < min_idle_ms )); then
 311
 312           # Ignore this host. i sometimes use a non-main machine for
 313           # testing or web browsing, knowing that everything will be wiped
 314           # by the next backup, but I dont want it to happen as Im using
 315           # it from cronjob.
 316           e "warning: $h: active X session in the last 15 minutes, skipping for now"
 317           continue
 318         fi
 319       fi
 320       sshable+=($h)
 321       if [[ $zone != $local_zone ]]; then
 322         die "error: dont confuse yourself with multiple time zones. $h has different timezone than localhost"
 323       fi
 324     else
 325       sshfail+=($h)
 326     fi
 327   done
 328   if [[ ! ${sshable[@]} ]] || { ! $cron && [[ ${sshfail[@]} ]]; }; then
 329     die "failed to ssh to hosts: ${sshfail[*]}"
 330   else
 331     if [[ ${sshfail[@]} ]]; then
 332       ret=1
 333       e "error: failed to ssh to ${sshfail[*]} but continuing with other hosts"
 334     fi
 335     targets=(${sshable[@]})
 336   fi
 337 fi
 338
 339
 340 cat >/etc/btrbk.conf <<EOF
 341 ssh_identity /root/.ssh/home
 342 # Just a guess that local7 is a good facility to pick.
 343 # It's a bit odd that the transaction log has to be logged to
 344 # a file or syslog, while other output is sent to std out.
 345 # The man does not mention a way for them to be together, but
 346 # I dunno if setting a log level like warn might also output
 347 # transaction info.
 348 transaction_syslog local7
 349
 350 # note, i had this because man said 20% speedup, but ran into
 351 # this issue, https://github.com/digint/btrbk/issues/275
 352 #stream_buffer 512m
 353
 354 # so we only run one at a time
 355 lockfile                   /var/lock/btrbk.lock
 356
 357 # default format of short does not accomidate hourly preservation setting
 358 timestamp_format long-iso
 359
 360 # only make a snapshot if things have changed
 361 snapshot_create onchange
 362 # I could make this different from target_preserve,
 363 # if one disk had less space.
 364 # for now, keeping them equal.
 365 snapshot_preserve 36h 14d 8w 24m
 366 snapshot_preserve_min 4h
 367 snapshot_dir btrbk
 368
 369 # so, total backups = ~89
 370 target_preserve 36h 14d 8w 24m
 371 target_preserve_min 4h
 372
 373 # if something fails and it's not obvious, try doing
 374 # btrbk -l debug -v dryrun
 375
 376 rate_limit $rate_limit
 377 EOF
 378
 379 if $incremental_strict; then
 380   cat >>/etc/btrbk.conf <<EOF
 381 incremental strict
 382 EOF
 383 fi
 384
 385 vol=/mnt/root
 386 for m in ${mountpoints[@]}; do
 387   sub=${m##*/}
 388   if [[ $source ]]; then
 389     cat >>/etc/btrbk.conf <<EOF
 390 volume ssh://$source$vol
 391 subvolume $sub
 392 target send-receive $vol/btrbk
 393 EOF
 394   else
 395     cat >>/etc/btrbk.conf <<EOF
 396 volume $vol
 397 subvolume $sub
 398 EOF
 399     for tg in ${targets[@]}; do
 400       cat >>/etc/btrbk.conf <<EOF
 401 target send-receive ssh://$tg$vol/btrbk
 402 EOF
 403     done
 404   fi
 405 done
 406
 407
 408 # todo: umount first to ensure we don't have any errors
 409 # todo: do some kill fuser stuff to make umount more reliable
 410
 411
 412 if $conf_only; then
 413   mexit 0
 414 fi
 415
 416
 417
 418 if $dry_run; then
 419   m btrbk -v -n $cmd_arg
 420   mexit 0
 421 elif [[ $cmd_arg == archive ]]; then
 422   if [[ $source ]]; then
 423     m btrbk $verbose_arg $progress_arg $cmd_arg ssh://$source$vol $vol
 424   else
 425     for tg in ${targets[@]}; do
 426       m btrbk $verbose_arg $progress_arg $cmd_arg $vol ssh://$tg$vol
 427     done
 428   fi
 429   mexit 0
 430 fi
 431 # -q and just using the syslog option seemed nice,
 432 # but it doesn't show when a send has a parent and when it doesn't.
 433 m btrbk $verbose_arg $progress_arg $cmd_arg
 434
 435 # if we have it, sync to systems which don't
 436 if mountpoint $rsync_mountpoint >/dev/null; then
 437   for tg in ${targets[@]}; do
 438     case $tg in
 439       li|lk)
 440         for x in /p/c/machine_specific/*.hosts; do
 441           if grep -qxF $tg $x; then
 442             dir=${x%.hosts}
 443             rsync-dirs $tg $dir
 444           fi
 445         done
 446         ;;
 447     esac
 448   done
 449 fi
 450
 451 if [[ $source ]]; then
 452   m mount-latest-subvol
 453 else
 454   m /a/exe/mount-latest-remote ${targets[@]}
 455 fi
 456
 457 mexit $ret
 458
 459 # todo: move variable data we don't care about backing up
 460 # to /nocow and symlink it.
 461
 462
 463 # background on btrbk timezones. with short/long, timestamps use local time.
 464 # for long, if your local time moves backwards, by moving timezones or
 465 # for an hour when daylight savings changes it, you will temporarily get
 466 # a more aggressive retention policy for the overlapping period, and
 467 # vice versa for the opposite timezone move. The alternative is using
 468 # long-iso, which puts timezone info into the timestamp, which means
 469 # that instead of shifting time, you shift the start of day/week/month
 470 # which is used for retention to your new local time, which means for
 471 # example, if you moved forward by 8 hours, the daily/weekly/monthly
 472 # retention will be 8 hours more aggressive since midnight is at a new
 473 # time, unless you fake the timzeone using the TZ env variable.
 474 # However, in the short term, there will be no inconsistencies.
 475 # I don't see any problem with shifting when the day starts for
 476 # retention, so I'm using long-iso.
 477
 478 # note to create a long-iso timestamp: date +%Y%m%dT%H%M%S%z