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 err-bash-trace() {
  20   local -i argc_index=0 frame i start=${1:-0} max_indent=8 indent
  21   local source
  22   local extdebug=false
  23   if [[ $(shopt -p extdebug) == *-s* ]]; then
  24     extdebug=true
  25   fi
  26   for ((frame=0; frame < ${#FUNCNAME[@]}-1; frame++)); do
  27     argc=${BASH_ARGC[frame]}
  28     argc_index+=$argc
  29     ((frame < start)) && continue
  30     if (( ${#BASH_SOURCE[@]} > 1 )); then
  31       source="${BASH_SOURCE[frame+1]}:${BASH_LINENO[frame]}:"
  32     fi
  33     indent=$((frame-start + 1))
  34     indent=$((indent < max_indent ? indent : max_indent))
  35     printf "%${indent}s↳%sin \`%s" '' "$source" "${FUNCNAME[frame]}"
  36     if $extdebug; then
  37       for ((i=argc_index-1; i >= argc_index-argc; i--)); do
  38         printf " %s" "${BASH_ARGV[i]}"
  39       done
  40     fi
  41     echo \'
  42   done
  43   return 0
  44 }
  45 err-catch() {
  46   set -E; shopt -s extdebug
  47   _err-trap() {
  48     err=$?
  49     exec >&2
  50     set +x
  51     echo "${BASH_SOURCE[1]}:${BASH_LINENO[0]}: \`$BASH_COMMAND' returned $err"
  52     err-bash-trace 2
  53     set -e # err trap does not work within an error trap
  54     "${_errcatch_cleanup[@]:-:}" # note :-: is to be compatible with set -u
  55     echo "$0: exiting with code $err"
  56     exit $err
  57   }
  58   trap _err-trap ERR
  59   set -o pipefail
  60 }
  61 err-catch
  62
  63
  64 [[ $EUID == 0 ]] || exec sudo -E "$BASH_SOURCE" "$@"
  65
  66
  67 usage() {
  68   cat <<'EOF'
  69 btrbk-run [OPTIONS]
  70 usually -t TARGET_HOST or -s SOURCE_HOST
  71
  72 Note, at source location, intentionally not executable, run and read
  73 install-my-scripts.
  74
  75 EOF
  76   echo "top of script file:"
  77   sed -n '1,/^[# ]*end command line/{p;b};q' "$0"
  78   exit $1
  79 }
  80
  81 rsync-dirs() {
  82   local host=$1
  83   local path=$2
  84   m rsync $dry_run_arg -ahi --relative --delete "$path" "root@$host:/"
  85 }
  86
  87
  88 m() { if $verbose; then printf "$pre %s\n"  "$*"; fi;  "$@"; }
  89 die() { printf "$pre %s\n" "$*" >&2;  exit 1; }
  90
  91 # latest $MAIL_HOST
  92 if [[ -e /b/bash_unpublished/source-state ]]; then
  93   source /b/bash_unpublished/source-state
  94 fi
  95
  96 # note q is owned by root:1000
  97
  98 mountpoints=()
  99
 100 rsync_mountpoint=/q
 101
 102 # default options
 103 conf_only=false
 104 dry_run=false # mostly for testing
 105 rate_limit=no
 106 verbose=true; verbose_arg=-v
 107 progress_arg="--progress"
 108 pull_reexec=false
 109
 110 default_args_file=/etc/btrbk-run.conf
 111 if [[ -s $default_args_file ]]; then
 112   set -- $(< $default_args_file) "$@"
 113   # i havent used this feature yet, so warn about it
 114   echo "$0: warning: default btrbk-run options set in $default_args_file (sleeping 5 seconds):"
 115   cat $default_args_file
 116   sleep 5
 117 fi
 118
 119 pre="${0##*/}:"
 120 cron=false
 121 orig_args=("$@")
 122 temp=$(getopt -l cron,pull-reexec,help cl:m:npqs:t:vh "$@") || usage 1
 123 eval set -- "$temp"
 124 while true; do
 125   case $1 in
 126     --cron)
 127       cron=true
 128       pre=
 129       shift
 130       ;;
 131     # only creates the config file, does not run btrbk
 132     -c) conf_only=true; shift ;;
 133     # bytes per second, suffix k m g
 134     -l) rate_limit=$2; shift 2 ;;
 135     # Comma separated mountpoints to backup. This has defaults set below.
 136     -m) IFS=, mountpoints=($2); unset IFS; shift 2 ;;
 137     -n) dry_run=true; dry_run_arg=-n; shift ;;
 138     -p) progress_arg="--progress"; shift ;;
 139     --pull-reexec) pull_reexec=true; shift ;;
 140     -q) verbose=false; verbose_arg=; progress_arg=; shift ;;
 141     # source host to receive a backup from
 142     -s) source=$2; shift 2 ;;
 143     # target hosts to send to. empty is valid for just doing local
 144     # snapshot. we have default hosts we will populate.
 145     -t) IFS=, targets=($2); unset IFS; shift 2 ;;
 146     -v) verbose=true; verbose_arg=-v; shift ;;
 147     -h|--help) usage ;;
 148     --) shift; break ;;
 149     *) die "Internal error!" ;;
 150   esac
 151 done
 152
 153 # usefull commands are resume and archive
 154 cmd_arg=${1:-run}
 155
 156 if [[ -v targets && $source ]]; then
 157   die "error: -t and -s are mutually exclusive"
 158 fi
 159
 160 if $verbose; then
 161   printf "$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"
 162 fi
 163 ### end options parsing
 164
 165 # TODO: i wonder if there should be an option to send to the default
 166 # targets, plus any given on the command line.
 167
 168 # set default targets
 169 if [[ ! -v targets && ! $source ]]; then
 170   if [[ $HOSTNAME != "$MAIL_HOST" ]] && $cron ; then
 171     echo "MAIL_HOST=$MAIL_HOST, nothing to do"
 172     exit 0
 173   fi
 174   case $HOSTNAME in
 175     kw|x2)
 176       if ping -q -c1 -w1 iank.vpn.office.fsf.org &>/dev/null; then
 177         home=iank.vpn.office.fsf.org
 178       else
 179         home=b8.nz
 180       fi
 181       ;;&
 182     kw)
 183       targets=($home x2)
 184       ;;
 185     x2)
 186       targets=($home kw)
 187       ;;
 188     tp)
 189       targets=(frodo kd)
 190       # might not be connected to the vpn
 191       if timeout -s 9 10 ssh kw :; then
 192         targets+=(kw)
 193       fi
 194       ;;
 195     kd)
 196       targets=(frodo tp)
 197       # might not be connected to the vpn
 198       if timeout -s 9 10 ssh kw :; then
 199         targets+=(kw)
 200       fi
 201       ;;
 202     *)
 203       die "error: no default targets for this host, use -t"
 204       ;;
 205   esac
 206 fi
 207
 208 if [[ -v targets ]]; then
 209   echo "targets: ${targets[*]}"
 210 fi
 211
 212 if [[ $source ]]; then
 213   echo "source: $source"
 214 fi
 215
 216
 217 if [[ $mountpoints ]]; then
 218   for mp in ${mountpoints[@]}; do # default mountpoints to sync
 219     if [[ -e /nocow/btrfs-stale/$mp ]]; then
 220       die "error: $mp is stale, mount-latest-subvol first"
 221     fi
 222   done
 223 else
 224   # set default mountpoints
 225   case $HOSTNAME in
 226     # no remote backups atm. note, if we do enable this, configuration below will need some changes.
 227     #  frodo)
 228     #    prospective_mps=(/i)
 229     #    ;;
 230     *)
 231       prospective_mps=(/a /q)
 232       if [[ $HOSTNAME == "$MAIL_HOST" ]]; then
 233         prospective_mps+=(/o)
 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       echo "$pre 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 "$0: error: cron btrbk is running on source. exiting out of caution"
 255     exit 1
 256   fi
 257 fi
 258
 259 # pull_reexec stops us from getting into an infinite loop if there is some
 260 # kind of weird problem
 261 pulla=false
 262 for m in "${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 $BASH_SOURCE; then
 272     echo "$pre 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     exit
 276   fi
 277 fi
 278
 279
 280
 281
 282
 283 if ! which btrbk &>/dev/null; then
 284   die "error: no btrbk binary found"
 285 fi
 286 # if our mountpoints are from stale snapshots,
 287 # it doesn't make sense to do a backup.
 288 check-subvol-stale ${mountpoints[@]} || die "found stale mountpoints in ${mountpoints[*]}"
 289
 290 # for an initial run, btrbk requires the dir to exist.
 291 mkdir -p /mnt/root/btrbk
 292 local_zone=$(date +%z)
 293
 294 if [[ $source ]]; then
 295   if ! zone=$(ssh root@$source date +%z); then
 296     die failed to ssh to root@$source
 297   fi
 298   if [[ $zone != $local_zone ]]; then
 299     die "error: dont confuse yourself with multiple time zones. $h has different timezone than localhost"
 300   fi
 301
 302 else
 303
 304   sshable=()
 305   sshfail=()
 306   min_idle_ms=$((1000 * 60 * 15))
 307   for h in ${targets[@]}; do
 308     if zone=$(ssh root@$h "mkdir -p /mnt/root/btrbk && date +%z"); then
 309       if $cron && DISPLAY=:0 xprintidle; then
 310         # This is a separate ssh because xprintidle can fail and thats ok.
 311         # Ignore this host. i sometimes use a non-main machine for testing or web browsing, knowing that
 312         # everything will be wiped by the next backup, but I dont want it to happen as Im using
 313         # it from cronjob.
 314         continue
 315       fi
 316       sshable+=($h)
 317       if [[ $zone != $local_zone ]]; then
 318         die "error: dont confuse yourself with multiple time zones. $h has different timezone than localhost"
 319       fi
 320     else
 321       sshfail+=($h)
 322     fi
 323   done
 324   if [[ ! $sshable ]] || { ! $cron && [[ $sshfail ]]; }; then
 325     die "failed to ssh to hosts: ${sshfail[*]}"
 326   else
 327     if [[ $sshfail ]]; then
 328       ret=1
 329       echo "$pre error: failed to ssh to ${sshfail[*]} but continuing with other hosts"
 330     fi
 331     targets=(${sshable[@]})
 332   fi
 333 fi
 334
 335
 336 cat >/etc/btrbk.conf <<EOF
 337 ssh_identity /root/.ssh/home
 338 # Just a guess that local7 is a good facility to pick.
 339 # It's a bit odd that the transaction log has to be logged to
 340 # a file or syslog, while other output is sent to std out.
 341 # The man does not mention a way for them to be together, but
 342 # I dunno if setting a log level like warn might also output
 343 # transaction info.
 344 transaction_syslog local7
 345
 346 # note, i had this because man said 20% speedup, but ran into
 347 # this issue, https://github.com/digint/btrbk/issues/275
 348 #stream_buffer 512m
 349
 350 # so we only run one at a time
 351 lockfile                   /var/lock/btrbk.lock
 352
 353 # default format of short does not accomidate hourly preservation setting
 354 timestamp_format long-iso
 355
 356 # only make a snapshot if things have changed
 357 snapshot_create onchange
 358 # I could make this different from target_preserve,
 359 # if one disk had less space.
 360 # for now, keeping them equal.
 361 snapshot_preserve 36h 14d 8w 24m
 362 snapshot_preserve_min 4h
 363 snapshot_dir btrbk
 364
 365 # so, total backups = ~89
 366 target_preserve 36h 14d 8w 24m
 367 target_preserve_min 4h
 368
 369 # if something fails and it's not obvious, try doing
 370 # btrbk -l debug -v dryrun
 371
 372 rate_limit $rate_limit
 373 EOF
 374
 375
 376
 377
 378
 379 vol=/mnt/root
 380 for m in ${mountpoints[@]}; do
 381   sub=${m##*/}
 382   if [[ $source ]]; then
 383     cat >>/etc/btrbk.conf <<EOF
 384 volume ssh://$source$vol
 385 subvolume $sub
 386 target send-receive $vol/btrbk
 387 EOF
 388   else
 389     cat >>/etc/btrbk.conf <<EOF
 390 volume $vol
 391 subvolume $sub
 392 EOF
 393     for tg in ${targets[@]}; do
 394       cat >>/etc/btrbk.conf <<EOF
 395 target send-receive ssh://$tg$vol/btrbk
 396 EOF
 397     done
 398   fi
 399 done
 400
 401
 402 # todo: umount first to ensure we don't have any errors
 403 # todo: do some kill fuser stuff to make umount more reliable
 404
 405
 406 if $conf_only; then
 407   exit
 408 fi
 409
 410
 411
 412 if $dry_run; then
 413   m btrbk -v -n $cmd_arg
 414   exit 0
 415 elif [[ $cmd_arg == archive ]]; then
 416   if [[ $source ]]; then
 417     m btrbk $verbose_arg $progress_arg $cmd_arg ssh://$source$vol $vol
 418   else
 419     for tg in ${targets[@]}; do
 420       m btrbk $verbose_arg $progress_arg $cmd_arg $vol ssh://$tg$vol
 421     done
 422   fi
 423   exit 0
 424 fi
 425 # -q and just using the syslog option seemed nice,
 426 # but it doesn't show when a send has a parent and when it doesn't.
 427 m btrbk $verbose_arg $progress_arg $cmd_arg
 428
 429 # if we have it, sync to systems which don't
 430 if mountpoint $rsync_mountpoint >/dev/null; then
 431   for tg in ${targets[@]}; do
 432     case $tg in
 433       li|lk)
 434         for x in /p/c/machine_specific/*.hosts; do
 435           if grep -qxF $tg $x; then
 436             dir=${x%.hosts}
 437             rsync-dirs $tg $dir
 438           fi
 439         done
 440         ;;
 441     esac
 442   done
 443 fi
 444
 445 if [[ $source ]]; then
 446   m mount-latest-subvol $verbose_arg
 447 else
 448   m /a/exe/mount-latest-remote ${targets[@]}
 449 fi
 450
 451 exit $ret
 452
 453 # todo: move variable data we don't care about backing up
 454 # to /nocow and symlink it.
 455
 456
 457 # background on btrbk timezones. with short/long, timestamps use local time.
 458 # for long, if your local time moves backwards, by moving timezones or
 459 # for an hour when daylight savings changes it, you will temporarily get
 460 # a more aggressive retention policy for the overlapping period, and
 461 # vice versa for the opposite timezone move. The alternative is using
 462 # long-iso, which puts timezone info into the timestamp, which means
 463 # that instead of shifting time, you shift the start of day/week/month
 464 # which is used for retention to your new local time, which means for
 465 # example, if you moved forward by 8 hours, the daily/weekly/monthly
 466 # retention will be 8 hours more aggressive since midnight is at a new
 467 # time, unless you fake the timzeone using the TZ env variable.
 468 # However, in the short term, there will be no inconsistencies.
 469 # I don't see any problem with shifting when the day starts for
 470 # retention, so I'm using long-iso.
 471
 472 # note to create a long-iso timestamp: date +%Y%m%dT%H%M%S%z