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;  exit 1; }
  49
  50 # latest $MAIL_HOST
  51 if [[ -e /b/bash_unpublished/source-state ]]; then
  52   source /b/bash_unpublished/source-state
  53 fi
  54
  55 # note q is owned by root:1000
  56
  57 mountpoints=()
  58
  59 rsync_mountpoint=/q
  60
  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 pull_reexec=false
  68
  69 default_args_file=/etc/btrbk-run.conf
  70 if [[ -s $default_args_file ]]; then
  71   set -- $(< $default_args_file) "$@"
  72   # i havent used this feature yet, so warn about it
  73   echo "$0: warning: default btrbk-run options set in $default_args_file (sleeping 5 seconds):"
  74   cat $default_args_file
  75   sleep 5
  76 fi
  77
  78 cron=false
  79 orig_args=("$@")
  80 temp=$(getopt -l cron,pull-reexec,help cl:m:npqs:t:vh "$@") || usage 1
  81 eval set -- "$temp"
  82 while true; do
  83   case $1 in
  84     --cron)
  85       cron=true
  86       pre=
  87       shift
  88       ;;
  89     # only creates the config file, does not run btrbk
  90     -c) conf_only=true; shift ;;
  91     # bytes per second, suffix k m g
  92     -l) rate_limit=$2; shift 2 ;;
  93     # Comma separated mountpoints to backup. This has defaults set below.
  94     -m) IFS=, mountpoints=($2); unset IFS; shift 2 ;;
  95     -n) dry_run=true; dry_run_arg=-n; shift ;;
  96     -p) progress_arg="--progress"; shift ;;
  97     --pull-reexec) pull_reexec=true; shift ;;
  98     -q) verbose=false; verbose_arg=; progress_arg=; shift ;;
  99     # source host to receive a backup from
 100     -s) source=$2; shift 2 ;;
 101     # target hosts to send to. empty is valid for just doing local
 102     # snapshot. we have default hosts we will populate.
 103     -t) IFS=, targets=($2); unset IFS; shift 2 ;;
 104     -v) verbose=true; verbose_arg=-v; shift ;;
 105     -h|--help) usage ;;
 106     --) shift; break ;;
 107     *) die "Internal error!" ;;
 108   esac
 109 done
 110
 111 # only tested commands are resume and archive
 112 cmd_arg=${1:-run}
 113
 114 case $cmd_arg in
 115   run|resume|archive) : ;;
 116   *) die "untested command arg" ;;
 117 esac
 118
 119 if (( $# > 1 )); then
 120   die: "only 1 nonoption arg is supported"
 121 fi
 122
 123 if [[ -v targets && $source ]]; then
 124   die "error: -t and -s are mutually exclusive"
 125 fi
 126
 127 if $verbose; then
 128   printf "options: conf_only=%s\ndry_run=%s\nrate_limit=%s\nverbose=%s\ncmd_arg=%s" "$conf_only" "$dry_run" "$rate_limit" "$verbose" "$cmd_arg"
 129 fi
 130 ### end options parsing
 131
 132 # TODO: i wonder if there should be an option to send to the default
 133 # targets, plus any given on the command line.
 134
 135 # set default targets
 136 if [[ ! -v targets && ! $source ]]; then
 137   if [[ $HOSTNAME != "$MAIL_HOST" ]] && $cron ; then
 138     echo "MAIL_HOST=$MAIL_HOST, nothing to do"
 139     exit 0
 140   fi
 141   case $HOSTNAME in
 142     kw|x2)
 143       if ping -q -c1 -w1 iank.vpn.office.fsf.org &>/dev/null; then
 144         home=iank.vpn.office.fsf.org
 145       else
 146         home=b8.nz
 147       fi
 148       ;;&
 149     kw)
 150       targets=($home x2)
 151       ;;
 152     x2)
 153       targets=($home kw)
 154       ;;
 155     x3)
 156       targets=($home)
 157       ;;
 158     tp)
 159       # kd disabled temporarily while its hot and i plan to work on it.
 160       #targets=(frodo kd)
 161       targets=(frodo x3.b8.nz)
 162       # might not be connected to the vpn
 163       if timeout -s 9 10 ssh kw :; then
 164         targets+=(kw)
 165       fi
 166       ;;
 167     kd)
 168       targets=(frodo tp)
 169       # might not be connected to the vpn
 170       if timeout -s 9 10 ssh kw :; then
 171         targets+=(kw)
 172       fi
 173       ;;
 174     *)
 175       die "error: no default targets for this host, use -t"
 176       ;;
 177   esac
 178 fi
 179
 180 if [[ -v targets ]]; then
 181   echo "targets: ${targets[*]}"
 182 fi
 183
 184 if [[ $source ]]; then
 185   echo "source: $source"
 186 fi
 187
 188
 189 if [[ $mountpoints ]]; then
 190   for mp in ${mountpoints[@]}; do # default mountpoints to sync
 191     if [[ -e /nocow/btrfs-stale/$mp ]]; then
 192       die "error: $mp is stale, mount-latest-subvol first"
 193     fi
 194   done
 195 else
 196   # set default mountpoints
 197   case $HOSTNAME in
 198     # no remote backups atm. note, if we do enable this, configuration below will need some changes.
 199     #  frodo)
 200     #    prospective_mps=(/i)
 201     #    ;;
 202     *)
 203       prospective_mps=(/a /q)
 204       if [[ $source ]]; then
 205         source_state="$(ssh $source cat /a/bin/bash_unpublished/source-state)"
 206         eval "$source_state"
 207         source_host="$(ssh $source cat /etc/hostname)"
 208         if [[ $source_host == "$MAIL_HOST" ]]; then
 209           prospective_mps+=(/o)
 210         fi
 211       else
 212         if [[ $HOSTNAME == "$MAIL_HOST" ]]; then
 213           prospective_mps+=(/o)
 214         fi
 215       fi
 216       ;;
 217   esac
 218   for mp in ${prospective_mps[@]}; do # default mountpoints to sync
 219     if [[ -e /nocow/btrfs-stale/$mp ]]; then
 220       e "warning: $mp stale, not adding to default mountpoints"
 221       continue
 222     fi
 223     if awk '{print $2}' /etc/fstab | grep -xF $mp &>/dev/null; then
 224       mountpoints+=($mp)
 225     fi
 226   done
 227 fi
 228
 229 echo "mountpoints: ${mountpoints[*]}"
 230
 231 ##### end command line parsing ########
 232
 233 if [[ $source ]]; then
 234   if [[ $(ssh $source systemctl is-active btrbk.service) != inactive ]]; then
 235     echo "$0: error: cron btrbk is running on source. exiting out of caution"
 236     exit 1
 237   fi
 238 fi
 239
 240 # pull_reexec stops us from getting into an infinite loop if there is some
 241 # kind of weird problem
 242 pulla=false
 243 for m in "${mountpoints[@]}"; do
 244   if [[ $m == /a ]]; then
 245     pulla=true
 246     break
 247   fi
 248 done
 249 if ! $pull_reexec && [[ $source ]] && $pulla ; then
 250   tmpf=$(mktemp)
 251   scp $source:/a/bin/distro-setup/btrbk-run $tmpf
 252   if ! diff -q $tmpf $BASH_SOURCE; then
 253     e "found newer version on host $source. reexecing"
 254     install -T $tmpf /usr/local/bin/btrbk-run
 255     m /usr/local/bin/btrbk-run --pull-reexec "${orig_args[@]}"
 256     exit
 257   fi
 258 fi
 259
 260
 261
 262
 263
 264 if ! which btrbk &>/dev/null; then
 265   die "error: no btrbk binary found"
 266 fi
 267 # if our mountpoints are from stale snapshots,
 268 # it doesn't make sense to do a backup.
 269 check-subvol-stale ${mountpoints[@]} || die "found stale mountpoints in ${mountpoints[*]}"
 270
 271 # for an initial run, btrbk requires the dir to exist.
 272 mkdir -p /mnt/root/btrbk
 273 local_zone=$(date +%z)
 274
 275 if [[ $source ]]; then
 276   if ! zone=$(ssh root@$source date +%z); then
 277     die failed to ssh to root@$source
 278   fi
 279   if [[ $zone != $local_zone ]]; then
 280     die "error: dont confuse yourself with multiple time zones. $h has different timezone than localhost"
 281   fi
 282
 283 else
 284
 285   sshable=()
 286   sshfail=()
 287   min_idle_ms=$((1000 * 60 * 15))
 288   for h in ${targets[@]}; do
 289     if zone=$(ssh root@$h "mkdir -p /mnt/root/btrbk && date +%z"); then
 290       # This is a separate ssh because xprintidle can fail and thats ok.
 291       if $cron && idle_ms=$(ssh $h DISPLAY=:0 xprintidle); then
 292         if (( idle_ms < min_idle_ms )); then
 293
 294           # Ignore this host. i sometimes use a non-main machine for
 295           # testing or web browsing, knowing that everything will be wiped
 296           # by the next backup, but I dont want it to happen as Im using
 297           # it from cronjob.
 298           e "warning: $h: active X session in the last 15 minutes, skipping for now"
 299           continue
 300         fi
 301       fi
 302       sshable+=($h)
 303       if [[ $zone != $local_zone ]]; then
 304         die "error: dont confuse yourself with multiple time zones. $h has different timezone than localhost"
 305       fi
 306     else
 307       sshfail+=($h)
 308     fi
 309   done
 310   if [[ ! ${sshable[@]} ]] || { ! $cron && [[ ${sshfail[@]} ]]; }; then
 311     die "failed to ssh to hosts: ${sshfail[*]}"
 312   else
 313     if [[ ${sshfail[@]} ]]; then
 314       ret=1
 315       e "error: failed to ssh to ${sshfail[*]} but continuing with other hosts"
 316     fi
 317     targets=(${sshable[@]})
 318   fi
 319 fi
 320
 321
 322 cat >/etc/btrbk.conf <<EOF
 323 ssh_identity /root/.ssh/home
 324 # Just a guess that local7 is a good facility to pick.
 325 # It's a bit odd that the transaction log has to be logged to
 326 # a file or syslog, while other output is sent to std out.
 327 # The man does not mention a way for them to be together, but
 328 # I dunno if setting a log level like warn might also output
 329 # transaction info.
 330 transaction_syslog local7
 331
 332 # note, i had this because man said 20% speedup, but ran into
 333 # this issue, https://github.com/digint/btrbk/issues/275
 334 #stream_buffer 512m
 335
 336 # so we only run one at a time
 337 lockfile                   /var/lock/btrbk.lock
 338
 339 # default format of short does not accomidate hourly preservation setting
 340 timestamp_format long-iso
 341
 342 # only make a snapshot if things have changed
 343 snapshot_create onchange
 344 # I could make this different from target_preserve,
 345 # if one disk had less space.
 346 # for now, keeping them equal.
 347 snapshot_preserve 36h 14d 8w 24m
 348 snapshot_preserve_min 4h
 349 snapshot_dir btrbk
 350
 351 # so, total backups = ~89
 352 target_preserve 36h 14d 8w 24m
 353 target_preserve_min 4h
 354
 355 # if something fails and it's not obvious, try doing
 356 # btrbk -l debug -v dryrun
 357
 358 rate_limit $rate_limit
 359 EOF
 360
 361
 362
 363
 364
 365 vol=/mnt/root
 366 for m in ${mountpoints[@]}; do
 367   sub=${m##*/}
 368   if [[ $source ]]; then
 369     cat >>/etc/btrbk.conf <<EOF
 370 volume ssh://$source$vol
 371 subvolume $sub
 372 target send-receive $vol/btrbk
 373 EOF
 374   else
 375     cat >>/etc/btrbk.conf <<EOF
 376 volume $vol
 377 subvolume $sub
 378 EOF
 379     for tg in ${targets[@]}; do
 380       cat >>/etc/btrbk.conf <<EOF
 381 target send-receive ssh://$tg$vol/btrbk
 382 EOF
 383     done
 384   fi
 385 done
 386
 387
 388 # todo: umount first to ensure we don't have any errors
 389 # todo: do some kill fuser stuff to make umount more reliable
 390
 391
 392 if $conf_only; then
 393   exit
 394 fi
 395
 396
 397
 398 if $dry_run; then
 399   m btrbk -v -n $cmd_arg
 400   exit 0
 401 elif [[ $cmd_arg == archive ]]; then
 402   if [[ $source ]]; then
 403     m btrbk $verbose_arg $progress_arg $cmd_arg ssh://$source$vol $vol
 404   else
 405     for tg in ${targets[@]}; do
 406       m btrbk $verbose_arg $progress_arg $cmd_arg $vol ssh://$tg$vol
 407     done
 408   fi
 409   exit 0
 410 fi
 411 # -q and just using the syslog option seemed nice,
 412 # but it doesn't show when a send has a parent and when it doesn't.
 413 m btrbk $verbose_arg $progress_arg $cmd_arg
 414
 415 # if we have it, sync to systems which don't
 416 if mountpoint $rsync_mountpoint >/dev/null; then
 417   for tg in ${targets[@]}; do
 418     case $tg in
 419       li|lk)
 420         for x in /p/c/machine_specific/*.hosts; do
 421           if grep -qxF $tg $x; then
 422             dir=${x%.hosts}
 423             rsync-dirs $tg $dir
 424           fi
 425         done
 426         ;;
 427     esac
 428   done
 429 fi
 430
 431 if [[ $source ]]; then
 432   m mount-latest-subvol $verbose_arg
 433 else
 434   m /a/exe/mount-latest-remote ${targets[@]}
 435 fi
 436
 437 exit $ret
 438
 439 # todo: move variable data we don't care about backing up
 440 # to /nocow and symlink it.
 441
 442
 443 # background on btrbk timezones. with short/long, timestamps use local time.
 444 # for long, if your local time moves backwards, by moving timezones or
 445 # for an hour when daylight savings changes it, you will temporarily get
 446 # a more aggressive retention policy for the overlapping period, and
 447 # vice versa for the opposite timezone move. The alternative is using
 448 # long-iso, which puts timezone info into the timestamp, which means
 449 # that instead of shifting time, you shift the start of day/week/month
 450 # which is used for retention to your new local time, which means for
 451 # example, if you moved forward by 8 hours, the daily/weekly/monthly
 452 # retention will be 8 hours more aggressive since midnight is at a new
 453 # time, unless you fake the timzeone using the TZ env variable.
 454 # However, in the short term, there will be no inconsistencies.
 455 # I don't see any problem with shifting when the day starts for
 456 # retention, so I'm using long-iso.
 457
 458 # note to create a long-iso timestamp: date +%Y%m%dT%H%M%S%z