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 set -eE -o pipefail
  20 trap 'echo "$0:$LINENO:error: \"$BASH_COMMAND\" returned $?" >&2' ERR
  21
  22 [[ $EUID == 0 ]] || exec sudo -E "$BASH_SOURCE" "$@"
  23
  24
  25 usage() {
  26   cat <<'EOF'
  27 btrbk-run [OPTIONS]
  28 usually -t TARGET_HOST or -s SOURCE_HOST
  29
  30 Note, at source location, intentionally not executable, run and read
  31 install-my-scripts.
  32
  33 EOF
  34   echo "top of script file:"
  35   sed -n '1,/^[# ]*end command line/{p;b};q' "$0"
  36   exit $1
  37 }
  38
  39 rsync-dirs() {
  40   local host=$1
  41   local path=$2
  42   m rsync $dry_run_arg -ahi --relative --delete "$path" "root@$host:/"
  43 }
  44
  45
  46 m() { if $verbose; then printf "$pre%s\n"  "$*"; fi;  "$@"; }
  47 die() { printf "$pre%s\n" "$*" >&2;  exit 1; }
  48
  49 # latest $MAIL_HOST
  50 if [[ -e /b/bash_unpublished/source-semi-priv ]]; then
  51   source /b/bash_unpublished/source-semi-priv
  52 fi
  53
  54 # note q is owned by root:1000
  55
  56 mountpoints=()
  57
  58 rsync_mountpoint=/q
  59
  60 # default options
  61 conf_only=false
  62 dry_run=false # mostly for testing
  63 rate_limit=no
  64 verbose=true; verbose_arg=-v
  65 progress_arg="--progress"
  66 pull_reexec=false
  67
  68 default_args_file=/etc/btrbk-run.conf
  69 if [[ -s $default_args_file ]]; then
  70   set -- $(< $default_args_file) "$@"
  71   echo "$0: warning: default btrbk-run options set in $default_args_file (sleeping 5 seconds):"
  72   cat $default_args_file
  73   sleep 5
  74 fi
  75
  76 cron=false
  77 orig_args=("$@")
  78 temp=$(getopt -l cpull-reexec,help cl:m:npqs:t:vh "$@") || usage 1
  79 eval set -- "$temp"
  80 while true; do
  81   case $1 in
  82     --cron)
  83       cron=true
  84       pre="${0##*/}: "
  85       shift
  86       ;;
  87     # only creates the config file, does not run btrbk
  88     -c) conf_only=true; shift ;;
  89     # bytes per second, suffix k m g
  90     -l) rate_limit=$2; shift 2 ;;
  91     # Comma separated mountpoints to backup. This has defaults set below.
  92     -m) IFS=, mountpoints=($2); unset IFS; shift 2 ;;
  93     -n) dry_run=true; dry_run_arg=-n; shift ;;
  94     -p) progress_arg="--progress"; shift ;;
  95     --pull-reexec) pull_reexec=true; shift ;;
  96     -q) verbose=false; verbose_arg=; progress_arg=; shift ;;
  97     # source host to receive a backup from
  98     -s) source=$2; shift 2 ;;
  99     # target hosts to send to. empty is valid for just doing local
 100     # snapshot. we have default hosts we will populate.
 101     -t) IFS=, targets=($2); unset IFS; shift 2 ;;
 102     -v) verbose=true; verbose_arg=-v; shift ;;
 103     -h|--help) usage ;;
 104     --) shift; break ;;
 105     *) die "Internal error!" ;;
 106   esac
 107 done
 108
 109 # usefull commands are resume and archive
 110 cmd_arg=${1:-run}
 111
 112 if [[ -v targets && $source ]]; then
 113   die "error: -t and -s are mutually exclusive"
 114 fi
 115
 116 if $verbose; then
 117   printf "$0: options: conf_only=%s\ndry_run=%s\nrate_limit=%s\nverbose=%s\ncmd_arg=%s" "$conf_only" "$dry_run" "$rate_limit" "$verbose" "$cmd_arg"
 118 fi
 119 ### end options parsing
 120
 121
 122 # set default targets
 123 if [[ ! -v targets && ! $source && $HOSTNAME == $MAIL_HOST ]]; then
 124   case $HOSTNAME in
 125     kw|x2)
 126       if ping -q -c1 -w1 iank.vpn.office.fsf.org &>/dev/null; then
 127         home=iank.vpn.office.fsf.org
 128       else
 129         home=$HOME_DOMAIN
 130       fi
 131       ;;&
 132     kw)
 133       targets=($home x2)
 134       ;;
 135     x2)
 136       targets=($home kw)
 137       ;;
 138     tp)
 139       targets=(frodo kd)
 140       # might not be connected to the vpn
 141       if timeout -s 9 10 ssh kw :; then
 142         targets+=(kw)
 143       fi
 144       ;;
 145     kd)
 146       targets=(frodo tp)
 147       # might not be connected to the vpn
 148       if timeout -s 9 10 ssh kw :; then
 149         targets+=(kw)
 150       fi
 151       ;;
 152     *)
 153       die "error: no default targets for this host, use -t"
 154       ;;
 155   esac
 156 fi
 157
 158 if [[ -v targets ]]; then
 159   echo "targets: ${targets[*]}"
 160 fi
 161
 162 if [[ $source ]]; then
 163   echo "source: $source"
 164 fi
 165
 166
 167 if [[ $mountpoints ]]; then
 168   for mp in ${mountpoints[@]}; do # default mountpoints to sync
 169     if [[ -e /nocow/btrfs-stale/$mp ]]; then
 170       die "error: $mp is stale, mount-latest-subvol first"
 171     fi
 172   done
 173 else
 174   # set default mountpoints
 175   case $HOSTNAME in
 176     # no remote backups atm. note, if we do enable this, configuration below will need some changes.
 177     #  frodo)
 178     #    prospective_mps=(/i)
 179     #    ;;
 180     *)
 181       prospective_mps=(/a /q)
 182       if [[ $HOSTNAME == "$MAIL_HOST" ]]; then
 183         prospective_mps+=(/o)
 184       fi
 185       ;;
 186   esac
 187   for mp in ${prospective_mps[@]}; do # default mountpoints to sync
 188     if [[ -e /nocow/btrfs-stale/$mp ]]; then
 189       echo "$0: warning: $mp stale, not adding to default mountpoints"
 190       continue
 191     fi
 192     if awk '{print $2}' /etc/fstab | grep -xF $mp &>/dev/null; then
 193       mountpoints+=($mp)
 194     fi
 195   done
 196 fi
 197
 198 echo "mountpoints: ${mountpoints[*]}"
 199
 200
 201
 202 # pull_reexec stops us from getting into an infinite loop if there is some
 203 # kind of weird problem
 204 pulla=false
 205 for m in "${mountpoints[@]}"; do
 206   if [[ $m == /a ]]; then
 207     pulla=true
 208     break
 209   fi
 210 done
 211 if ! $pull_reexec && [[ $source ]] && ! $pulla ; then
 212   tmpf=$(mktemp)
 213   scp $source:/a/bin/distro-setup/btrbk-run $tmpf
 214   if ! diff -q $tmpf $BASH_SOURCE; then
 215     echo "$0: found newer version on host $source. reexecing"
 216     install -T $tmpf /usr/local/bin/btrbk-run
 217     m /usr/local/bin/btrbk-run --pull-reexec "${orig_args[@]}"
 218     exit
 219   fi
 220 fi
 221
 222
 223 ##### end command line parsing ########
 224
 225
 226
 227 if ! which btrbk &>/dev/null; then
 228   die "error: no btrbk binary found"
 229 fi
 230 # if our mountpoints are from stale snapshots,
 231 # it doesn't make sense to do a backup.
 232 check-subvol-stale ${mountpoints[@]} || die "found stale mountpoints in ${mountpoints[*]}"
 233
 234
 235 cat >/etc/btrbk.conf <<EOF
 236 ssh_identity /root/.ssh/home
 237 # Just a guess that local7 is a good facility to pick.
 238 # It's a bit odd that the transaction log has to be logged to
 239 # a file or syslog, while other output is sent to std out.
 240 # The man does not mention a way for them to be together, but
 241 # I dunno if setting a log level like warn might also output
 242 # transaction info.
 243 transaction_syslog local7
 244
 245 # note, i had this because man said 20% speedup, but ran into
 246 # this issue, https://github.com/digint/btrbk/issues/275
 247 #stream_buffer 512m
 248
 249 # so we only run one at a time
 250 lockfile                   /var/lock/btrbk.lock
 251
 252 # default format of short does not accomidate hourly preservation setting
 253 timestamp_format long-iso
 254
 255 # only make a snapshot if things have changed
 256 snapshot_create onchange
 257 # I could make this different from target_preserve,
 258 # if one disk had less space.
 259 # for now, keeping them equal.
 260 snapshot_preserve 36h 14d 8w 24m
 261 snapshot_preserve_min 4h
 262 snapshot_dir btrbk
 263
 264 # so, total backups = ~89
 265 target_preserve 36h 14d 8w 24m
 266 target_preserve_min 4h
 267
 268 # if something fails and it's not obvious, try doing
 269 # btrbk -l debug -v dryrun
 270
 271 rate_limit $rate_limit
 272 EOF
 273
 274
 275
 276 # for an initial run, btrbk requires the dir to exist.
 277 mkdir -p /mnt/root/btrbk
 278 local_zone=$(date +%z)
 279 for h in ${targets[@]} $source; do
 280   cmd="ssh root@$h mkdir -p /mnt/root/btrbk && date +%z"
 281   if ! zone=$($cmd); then
 282     die "error running: $cmd"
 283   fi
 284   if [[ $zone != $local_zone ]]; then
 285     die "error: dont confuse yourself with multiple time zones. $h has different timezone than localhost"
 286   fi
 287 done
 288
 289
 290
 291 vol=/mnt/root
 292 for m in ${mountpoints[@]}; do
 293   sub=${m##*/}
 294   if [[ $source ]]; then
 295     cat >>/etc/btrbk.conf <<EOF
 296 volume ssh://$source$vol
 297 subvolume $sub
 298 target send-receive $vol/btrbk
 299 EOF
 300   else
 301     cat >>/etc/btrbk.conf <<EOF
 302 volume $vol
 303 subvolume $sub
 304 EOF
 305     for tg in ${targets[@]}; do
 306       cat >>/etc/btrbk.conf <<EOF
 307 target send-receive ssh://$tg$vol/btrbk
 308 EOF
 309     done
 310   fi
 311 done
 312
 313
 314 # todo: umount first to ensure we don't have any errors
 315 # todo: do some kill fuser stuff to make umount more reliable
 316
 317
 318 if $conf_only; then
 319   exit
 320 fi
 321
 322
 323
 324 if $dry_run; then
 325   m btrbk -v -n $cmd_arg
 326   exit 0
 327 elif [[ $cmd_arg == archive ]]; then
 328   if [[ $source ]]; then
 329     m btrbk $verbose_arg $progress_arg $cmd_arg ssh://$source$vol $vol
 330   else
 331     for tg in ${targets[@]}; do
 332       m btrbk $verbose_arg $progress_arg $cmd_arg $vol ssh://$tg$vol
 333     done
 334   fi
 335   exit 0
 336 fi
 337 # -q and just using the syslog option seemed nice,
 338 # but it doesn't show when a send has a parent and when it doesn't.
 339 m btrbk $verbose_arg $progress_arg $cmd_arg
 340
 341 # if we have it, sync to systems which don't
 342 if mountpoint $rsync_mountpoint >/dev/null; then
 343   for tg in ${targets[@]}; do
 344     case $tg in
 345       li|lk)
 346         for x in /p/c/machine_specific/*.hosts; do
 347           if grep -qxF $tg $x; then
 348             dir=${x%.hosts}
 349             rsync-dirs $tg $dir
 350           fi
 351         done
 352         ;;
 353     esac
 354   done
 355 fi
 356
 357 if [[ $source ]]; then
 358   m mount-latest-subvol $verbose_arg
 359 else
 360   m /a/exe/mount-latest-remote ${targets[@]}
 361 fi
 362
 363 # todo: move variable data we don't care about backing up
 364 # to /nocow and symlink it.
 365
 366
 367 # background on btrbk timezones. with short/long, timestamps use local time.
 368 # for long, if your local time moves backwards, by moving timezones or
 369 # for an hour when daylight savings changes it, you will temporarily get
 370 # a more aggressive retention policy for the overlapping period, and
 371 # vice versa for the opposite timezone move. The alternative is using
 372 # long-iso, which puts timezone info into the timestamp, which means
 373 # that instead of shifting time, you shift the start of day/week/month
 374 # which is used for retention to your new local time, which means for
 375 # example, if you moved forward by 8 hours, the daily/weekly/monthly
 376 # retention will be 8 hours more aggressive since midnight is at a new
 377 # time, unless you fake the timzeone using the TZ env variable.
 378 # However, in the short term, there will be no inconsistencies.
 379 # I don't see any problem with shifting when the day starts for
 380 # retention, so I'm using long-iso.
 381
 382 # note to create a long-iso timestamp: date +%Y%m%dT%H%M%S%z