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 "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 # default options
  63 conf_only=false
  64 dry_run=false # mostly for testing
  65 rate_limit=no
  66 verbose=true; verbose_arg=-v
  67 progress_arg="--progress"
  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 $default_args_file
  76   sleep 5
  77 fi
  78
  79 cron=false
  80 orig_args=("$@")
  81 temp=$(getopt -l cron,pull-reexec,help cl:m:npqs:t:vh "$@") || usage 1
  82 eval set -- "$temp"
  83 while true; do
  84   case $1 in
  85     --cron)
  86       cron=true
  87       pre=
  88       shift
  89       ;;
  90     # only creates the config file, does not run btrbk
  91     -c) conf_only=true; shift ;;
  92     # bytes per second, suffix k m g
  93     -l) rate_limit=$2; shift 2 ;;
  94     # Comma separated mountpoints to backup. This has defaults set below.
  95     -m) IFS=, mountpoints=($2); unset IFS; shift 2 ;;
  96     -n) dry_run=true; dry_run_arg=-n; shift ;;
  97     -p) progress_arg="--progress"; shift ;;
  98     --pull-reexec) pull_reexec=true; shift ;;
  99     -q) verbose=false; verbose_arg=; progress_arg=; shift ;;
 100     # source host to receive a backup from
 101     -s) source=$2; shift 2 ;;
 102     # target hosts to send to. empty is valid for just doing local
 103     # snapshot. we have default hosts we will populate.
 104     -t) IFS=, targets=($2); unset IFS; shift 2 ;;
 105     -v) verbose=true; verbose_arg=-v; shift ;;
 106     -h|--help) usage ;;
 107     --) shift; break ;;
 108     *) die "Internal error!" ;;
 109   esac
 110 done
 111
 112 # only tested commands are resume and archive
 113 cmd_arg=${1:-run}
 114
 115 case $cmd_arg in
 116   run|resume|archive) : ;;
 117   *) die "untested command arg" ;;
 118 esac
 119
 120 if (( $# > 1 )); then
 121   die: "only 1 nonoption arg is supported"
 122 fi
 123
 124 if [[ -v targets && $source ]]; then
 125   die "error: -t and -s are mutually exclusive"
 126 fi
 127
 128 if $verbose; then
 129   printf "options: conf_only=%s\ndry_run=%s\nrate_limit=%s\nverbose=%s\ncmd_arg=%s" "$conf_only" "$dry_run" "$rate_limit" "$verbose" "$cmd_arg"
 130 fi
 131 ### end options parsing
 132
 133
 134 # TODO: i wonder if there should be an option to send to the default
 135 # targets, plus any given on the command line.
 136
 137 # set default targets
 138 if [[ ! -v targets && ! $source ]]; then
 139   if [[ $HOSTNAME != "$MAIL_HOST" ]] && $cron ; then
 140     echo "MAIL_HOST=$MAIL_HOST, nothing to do"
 141     mexit 0
 142   fi
 143
 144   at_work=false
 145
 146   case $HOSTNAME in
 147     kw)
 148       at_work=true
 149       ;;&
 150     x2|x3)
 151       if ping -q -c1 -w1 hal.office.fsf.org \
 152               && ip n show 192.168.0.26 | grep . &>/dev/null; then
 153       at_work=true
 154       fi
 155       ;;&
 156     kw|x2|x3)
 157       if $at_work && ping -q -c1 -w1 iank.vpn.office.fsf.org &>/dev/null; then
 158         home=iank.vpn.office.fsf.org
 159       else
 160         home=b8.nz
 161       fi
 162       ;;&
 163     kw)
 164       targets=($home x2)
 165       ;;
 166     x2|x3)
 167       targets=($home kw)
 168       ;;
 169     tp)
 170       # kd disabled temporarily while its hot and i plan to work on it.
 171       #targets=(frodo kd)
 172       targets=(frodo x3.b8.nz)
 173       # might not be connected to the vpn
 174       if timeout -s 9 10 ssh kw :; then
 175         targets+=(kw)
 176       fi
 177       ;;
 178     kd)
 179       targets=(frodo tp)
 180       # might not be connected to the vpn
 181       if timeout -s 9 10 ssh kw :; then
 182         targets+=(kw)
 183       fi
 184       ;;
 185     *)
 186       die "error: no default targets for this host, use -t"
 187       ;;
 188   esac
 189 fi
 190
 191 if [[ -v targets ]]; then
 192   echo "targets: ${targets[*]}"
 193 fi
 194
 195 if [[ $source ]]; then
 196   echo "source: $source"
 197 fi
 198
 199
 200 if [[ $mountpoints ]]; then
 201   for mp in ${mountpoints[@]}; do # default mountpoints to sync
 202     if [[ -e /nocow/btrfs-stale/$mp ]]; then
 203       die "error: $mp is stale, mount-latest-subvol first"
 204     fi
 205   done
 206 else
 207   # set default mountpoints
 208   case $HOSTNAME in
 209     # no remote backups atm. note, if we do enable this, configuration below will need some changes.
 210     #  frodo)
 211     #    prospective_mps=(/i)
 212     #    ;;
 213     *)
 214       prospective_mps=(/a /q)
 215       if [[ $source ]]; then
 216         source_state="$(ssh $source cat /a/bin/bash_unpublished/source-state)"
 217         eval "$source_state"
 218         source_host="$(ssh $source cat /etc/hostname)"
 219         if [[ $source_host == "$MAIL_HOST" ]]; then
 220           prospective_mps+=(/o)
 221         fi
 222       else
 223         if [[ $HOSTNAME == "$MAIL_HOST" ]]; then
 224           prospective_mps+=(/o)
 225         fi
 226       fi
 227       ;;
 228   esac
 229   for mp in ${prospective_mps[@]}; do # default mountpoints to sync
 230     if [[ -e /nocow/btrfs-stale/$mp ]]; then
 231       e "warning: $mp stale, not adding to default mountpoints"
 232       continue
 233     fi
 234     if awk '{print $2}' /etc/fstab | grep -xF $mp &>/dev/null; then
 235       mountpoints+=($mp)
 236     fi
 237   done
 238 fi
 239
 240 echo "mountpoints: ${mountpoints[*]}"
 241
 242 ##### end command line parsing ########
 243
 244 if [[ $source ]]; then
 245   if [[ $(ssh $source systemctl is-active btrbk.service) != inactive ]]; then
 246     echo "$0: error: cron btrbk is running on source. exiting out of caution"
 247     mexit 1
 248   fi
 249 fi
 250
 251 # pull_reexec stops us from getting into an infinite loop if there is some
 252 # kind of weird problem
 253 pulla=false
 254 for m in "${mountpoints[@]}"; do
 255   if [[ $m == /a ]]; then
 256     pulla=true
 257     break
 258   fi
 259 done
 260 if ! $pull_reexec && [[ $source ]] && $pulla ; then
 261   tmpf=$(mktemp)
 262   scp $source:/a/bin/distro-setup/btrbk-run $tmpf
 263   if ! diff -q $tmpf $BASH_SOURCE; then
 264     e "found newer version on host $source. reexecing"
 265     install -T $tmpf /usr/local/bin/btrbk-run
 266     m /usr/local/bin/btrbk-run --pull-reexec "${orig_args[@]}"
 267     mexit 0
 268   fi
 269 fi
 270
 271
 272
 273
 274
 275 if ! which btrbk &>/dev/null; then
 276   die "error: no btrbk binary found"
 277 fi
 278 # if our mountpoints are from stale snapshots,
 279 # it doesn't make sense to do a backup.
 280 check-subvol-stale ${mountpoints[@]} || die "found stale mountpoints in ${mountpoints[*]}"
 281
 282 # for an initial run, btrbk requires the dir to exist.
 283 mkdir -p /mnt/root/btrbk
 284 local_zone=$(date +%z)
 285
 286 if [[ $source ]]; then
 287   if ! zone=$(ssh root@$source date +%z); then
 288     die failed to ssh to root@$source
 289   fi
 290   if [[ $zone != $local_zone ]]; then
 291     die "error: dont confuse yourself with multiple time zones. $h has different timezone than localhost"
 292   fi
 293
 294 else
 295
 296   sshable=()
 297   sshfail=()
 298   min_idle_ms=$((1000 * 60 * 15))
 299   for h in ${targets[@]}; do
 300     if zone=$(ssh root@$h "mkdir -p /mnt/root/btrbk && date +%z"); then
 301       # This is a separate ssh because xprintidle can fail and thats ok.
 302       if $cron && idle_ms=$(ssh $h DISPLAY=:0 xprintidle); then
 303         if (( idle_ms < min_idle_ms )); then
 304
 305           # Ignore this host. i sometimes use a non-main machine for
 306           # testing or web browsing, knowing that everything will be wiped
 307           # by the next backup, but I dont want it to happen as Im using
 308           # it from cronjob.
 309           e "warning: $h: active X session in the last 15 minutes, skipping for now"
 310           continue
 311         fi
 312       fi
 313       sshable+=($h)
 314       if [[ $zone != $local_zone ]]; then
 315         die "error: dont confuse yourself with multiple time zones. $h has different timezone than localhost"
 316       fi
 317     else
 318       sshfail+=($h)
 319     fi
 320   done
 321   if [[ ! ${sshable[@]} ]] || { ! $cron && [[ ${sshfail[@]} ]]; }; then
 322     die "failed to ssh to hosts: ${sshfail[*]}"
 323   else
 324     if [[ ${sshfail[@]} ]]; then
 325       ret=1
 326       e "error: failed to ssh to ${sshfail[*]} but continuing with other hosts"
 327     fi
 328     targets=(${sshable[@]})
 329   fi
 330 fi
 331
 332
 333 cat >/etc/btrbk.conf <<EOF
 334 ssh_identity /root/.ssh/home
 335 # Just a guess that local7 is a good facility to pick.
 336 # It's a bit odd that the transaction log has to be logged to
 337 # a file or syslog, while other output is sent to std out.
 338 # The man does not mention a way for them to be together, but
 339 # I dunno if setting a log level like warn might also output
 340 # transaction info.
 341 transaction_syslog local7
 342
 343 # note, i had this because man said 20% speedup, but ran into
 344 # this issue, https://github.com/digint/btrbk/issues/275
 345 #stream_buffer 512m
 346
 347 # so we only run one at a time
 348 lockfile                   /var/lock/btrbk.lock
 349
 350 # default format of short does not accomidate hourly preservation setting
 351 timestamp_format long-iso
 352
 353 # only make a snapshot if things have changed
 354 snapshot_create onchange
 355 # I could make this different from target_preserve,
 356 # if one disk had less space.
 357 # for now, keeping them equal.
 358 snapshot_preserve 36h 14d 8w 24m
 359 snapshot_preserve_min 4h
 360 snapshot_dir btrbk
 361
 362 # so, total backups = ~89
 363 target_preserve 36h 14d 8w 24m
 364 target_preserve_min 4h
 365
 366 # if something fails and it's not obvious, try doing
 367 # btrbk -l debug -v dryrun
 368
 369 rate_limit $rate_limit
 370 EOF
 371
 372
 373
 374
 375
 376 vol=/mnt/root
 377 for m in ${mountpoints[@]}; do
 378   sub=${m##*/}
 379   if [[ $source ]]; then
 380     cat >>/etc/btrbk.conf <<EOF
 381 volume ssh://$source$vol
 382 subvolume $sub
 383 target send-receive $vol/btrbk
 384 EOF
 385   else
 386     cat >>/etc/btrbk.conf <<EOF
 387 volume $vol
 388 subvolume $sub
 389 EOF
 390     for tg in ${targets[@]}; do
 391       cat >>/etc/btrbk.conf <<EOF
 392 target send-receive ssh://$tg$vol/btrbk
 393 EOF
 394     done
 395   fi
 396 done
 397
 398
 399 # todo: umount first to ensure we don't have any errors
 400 # todo: do some kill fuser stuff to make umount more reliable
 401
 402
 403 if $conf_only; then
 404   mexit 0
 405 fi
 406
 407
 408
 409 if $dry_run; then
 410   m btrbk -v -n $cmd_arg
 411   mexit 0
 412 elif [[ $cmd_arg == archive ]]; then
 413   if [[ $source ]]; then
 414     m btrbk $verbose_arg $progress_arg $cmd_arg ssh://$source$vol $vol
 415   else
 416     for tg in ${targets[@]}; do
 417       m btrbk $verbose_arg $progress_arg $cmd_arg $vol ssh://$tg$vol
 418     done
 419   fi
 420   mexit 0
 421 fi
 422 # -q and just using the syslog option seemed nice,
 423 # but it doesn't show when a send has a parent and when it doesn't.
 424 m btrbk $verbose_arg $progress_arg $cmd_arg
 425
 426 # if we have it, sync to systems which don't
 427 if mountpoint $rsync_mountpoint >/dev/null; then
 428   for tg in ${targets[@]}; do
 429     case $tg in
 430       li|lk)
 431         for x in /p/c/machine_specific/*.hosts; do
 432           if grep -qxF $tg $x; then
 433             dir=${x%.hosts}
 434             rsync-dirs $tg $dir
 435           fi
 436         done
 437         ;;
 438     esac
 439   done
 440 fi
 441
 442 if [[ $source ]]; then
 443   m mount-latest-subvol $verbose_arg
 444 else
 445   m /a/exe/mount-latest-remote ${targets[@]}
 446 fi
 447
 448 e btrbk-run COMPLETED with exit status $ret
 449 mexit $ret
 450
 451 # todo: move variable data we don't care about backing up
 452 # to /nocow and symlink it.
 453
 454
 455 # background on btrbk timezones. with short/long, timestamps use local time.
 456 # for long, if your local time moves backwards, by moving timezones or
 457 # for an hour when daylight savings changes it, you will temporarily get
 458 # a more aggressive retention policy for the overlapping period, and
 459 # vice versa for the opposite timezone move. The alternative is using
 460 # long-iso, which puts timezone info into the timestamp, which means
 461 # that instead of shifting time, you shift the start of day/week/month
 462 # which is used for retention to your new local time, which means for
 463 # example, if you moved forward by 8 hours, the daily/weekly/monthly
 464 # retention will be 8 hours more aggressive since midnight is at a new
 465 # time, unless you fake the timzeone using the TZ env variable.
 466 # However, in the short term, there will be no inconsistencies.
 467 # I don't see any problem with shifting when the day starts for
 468 # retention, so I'm using long-iso.
 469
 470 # note to create a long-iso timestamp: date +%Y%m%dT%H%M%S%z