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   # i havent used this feature yet, so warn about it
  72   echo "$0: warning: default btrbk-run options set in $default_args_file (sleeping 5 seconds):"
  73   cat $default_args_file
  74   sleep 5
  75 fi
  76
  77 pre="${0##*/}:"
  78 cron=false
  79 orig_args=("$@")
  80 temp=$(getopt -l cpull-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=($2); unset IFS; shift 2 ;;
 104     -v) verbose=true; verbose_arg=-v; shift ;;
 105     -h|--help) usage ;;
 106     --) shift; break ;;
 107     *) die "Internal error!" ;;
 108   esac
 109 done
 110
 111 # usefull commands are resume and archive
 112 cmd_arg=${1:-run}
 113
 114 if [[ -v targets && $source ]]; then
 115   die "error: -t and -s are mutually exclusive"
 116 fi
 117
 118 if $verbose; then
 119   printf "$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"
 120 fi
 121 ### end options parsing
 122
 123
 124 # set default targets
 125 if [[ ! -v targets && ! $source && $HOSTNAME == "$MAIL_HOST" ]]; then
 126   case $HOSTNAME in
 127     kw|x2)
 128       if ping -q -c1 -w1 iank.vpn.office.fsf.org &>/dev/null; then
 129         home=iank.vpn.office.fsf.org
 130       else
 131         home=$HOME_DOMAIN
 132       fi
 133       ;;&
 134     kw)
 135       targets=($home x2)
 136       ;;
 137     x2)
 138       targets=($home kw)
 139       ;;
 140     tp)
 141       targets=(frodo kd)
 142       # might not be connected to the vpn
 143       if timeout -s 9 10 ssh kw :; then
 144         targets+=(kw)
 145       fi
 146       ;;
 147     kd)
 148       targets=(frodo tp)
 149       # might not be connected to the vpn
 150       if timeout -s 9 10 ssh kw :; then
 151         targets+=(kw)
 152       fi
 153       ;;
 154     *)
 155       die "error: no default targets for this host, use -t"
 156       ;;
 157   esac
 158 fi
 159
 160 if [[ -v targets ]]; then
 161   echo "targets: ${targets[*]}"
 162 fi
 163
 164 if [[ $source ]]; then
 165   echo "source: $source"
 166 fi
 167
 168
 169 if [[ $mountpoints ]]; then
 170   for mp in ${mountpoints[@]}; do # default mountpoints to sync
 171     if [[ -e /nocow/btrfs-stale/$mp ]]; then
 172       die "error: $mp is stale, mount-latest-subvol first"
 173     fi
 174   done
 175 else
 176   # set default mountpoints
 177   case $HOSTNAME in
 178     # no remote backups atm. note, if we do enable this, configuration below will need some changes.
 179     #  frodo)
 180     #    prospective_mps=(/i)
 181     #    ;;
 182     *)
 183       prospective_mps=(/a /q)
 184       if [[ $HOSTNAME == "$MAIL_HOST" ]]; then
 185         prospective_mps+=(/o)
 186       fi
 187       ;;
 188   esac
 189   for mp in ${prospective_mps[@]}; do # default mountpoints to sync
 190     if [[ -e /nocow/btrfs-stale/$mp ]]; then
 191       echo "$pre warning: $mp stale, not adding to default mountpoints"
 192       continue
 193     fi
 194     if awk '{print $2}' /etc/fstab | grep -xF $mp &>/dev/null; then
 195       mountpoints+=($mp)
 196     fi
 197   done
 198 fi
 199
 200 echo "mountpoints: ${mountpoints[*]}"
 201
 202
 203
 204 # pull_reexec stops us from getting into an infinite loop if there is some
 205 # kind of weird problem
 206 pulla=false
 207 for m in "${mountpoints[@]}"; do
 208   if [[ $m == /a ]]; then
 209     pulla=true
 210     break
 211   fi
 212 done
 213 if ! $pull_reexec && [[ $source ]] && ! $pulla ; then
 214   tmpf=$(mktemp)
 215   scp $source:/a/bin/distro-setup/btrbk-run $tmpf
 216   if ! diff -q $tmpf $BASH_SOURCE; then
 217     echo "$pre found newer version on host $source. reexecing"
 218     install -T $tmpf /usr/local/bin/btrbk-run
 219     m /usr/local/bin/btrbk-run --pull-reexec "${orig_args[@]}"
 220     exit
 221   fi
 222 fi
 223
 224
 225 ##### end command line parsing ########
 226
 227
 228
 229 if ! which btrbk &>/dev/null; then
 230   die "error: no btrbk binary found"
 231 fi
 232 # if our mountpoints are from stale snapshots,
 233 # it doesn't make sense to do a backup.
 234 check-subvol-stale ${mountpoints[@]} || die "found stale mountpoints in ${mountpoints[*]}"
 235
 236 # for an initial run, btrbk requires the dir to exist.
 237 mkdir -p /mnt/root/btrbk
 238 local_zone=$(date +%z)
 239
 240 if [[ $source ]]; then
 241   if ! zone=$(ssh root$source date +%z); then
 242     die failed to ssh to root@$source
 243   fi
 244   if [[ $zone != $local_zone ]]; then
 245     die "error: dont confuse yourself with multiple time zones. $h has different timezone than localhost"
 246   fi
 247
 248 else
 249
 250   sshable=()
 251   sshfail=()
 252   for h in ${targets[@]}; do
 253     if zone=$(ssh root@$h "mkdir -p /mnt/root/btrbk && date +%z"); then
 254       sshable+=($h)
 255     else
 256       sshfail+=($h)
 257     fi
 258     if [[ $zone != $local_zone ]]; then
 259       die "error: dont confuse yourself with multiple time zones. $h has different timezone than localhost"
 260     fi
 261   done
 262   if [[ ! $sshable ]] || { ! $cron && [[ $sshfail ]]; }; then
 263     die "failed to ssh to hosts: ${sshfail[*]}"
 264   else
 265     if [[ $sshfail ]]; then
 266       ret=1
 267       echo "$pre error: failed to ssh to ${sshfail[*]} but continuing with other hosts"
 268     fi
 269     targets=(${sshable[@]})
 270   fi
 271 fi
 272
 273
 274 cat >/etc/btrbk.conf <<EOF
 275 ssh_identity /root/.ssh/home
 276 # Just a guess that local7 is a good facility to pick.
 277 # It's a bit odd that the transaction log has to be logged to
 278 # a file or syslog, while other output is sent to std out.
 279 # The man does not mention a way for them to be together, but
 280 # I dunno if setting a log level like warn might also output
 281 # transaction info.
 282 transaction_syslog local7
 283
 284 # note, i had this because man said 20% speedup, but ran into
 285 # this issue, https://github.com/digint/btrbk/issues/275
 286 #stream_buffer 512m
 287
 288 # so we only run one at a time
 289 lockfile                   /var/lock/btrbk.lock
 290
 291 # default format of short does not accomidate hourly preservation setting
 292 timestamp_format long-iso
 293
 294 # only make a snapshot if things have changed
 295 snapshot_create onchange
 296 # I could make this different from target_preserve,
 297 # if one disk had less space.
 298 # for now, keeping them equal.
 299 snapshot_preserve 36h 14d 8w 24m
 300 snapshot_preserve_min 4h
 301 snapshot_dir btrbk
 302
 303 # so, total backups = ~89
 304 target_preserve 36h 14d 8w 24m
 305 target_preserve_min 4h
 306
 307 # if something fails and it's not obvious, try doing
 308 # btrbk -l debug -v dryrun
 309
 310 rate_limit $rate_limit
 311 EOF
 312
 313
 314
 315
 316
 317 vol=/mnt/root
 318 for m in ${mountpoints[@]}; do
 319   sub=${m##*/}
 320   if [[ $source ]]; then
 321     cat >>/etc/btrbk.conf <<EOF
 322 volume ssh://$source$vol
 323 subvolume $sub
 324 target send-receive $vol/btrbk
 325 EOF
 326   else
 327     cat >>/etc/btrbk.conf <<EOF
 328 volume $vol
 329 subvolume $sub
 330 EOF
 331     for tg in ${targets[@]}; do
 332       cat >>/etc/btrbk.conf <<EOF
 333 target send-receive ssh://$tg$vol/btrbk
 334 EOF
 335     done
 336   fi
 337 done
 338
 339
 340 # todo: umount first to ensure we don't have any errors
 341 # todo: do some kill fuser stuff to make umount more reliable
 342
 343
 344 if $conf_only; then
 345   exit
 346 fi
 347
 348
 349
 350 if $dry_run; then
 351   m btrbk -v -n $cmd_arg
 352   exit 0
 353 elif [[ $cmd_arg == archive ]]; then
 354   if [[ $source ]]; then
 355     m btrbk $verbose_arg $progress_arg $cmd_arg ssh://$source$vol $vol
 356   else
 357     for tg in ${targets[@]}; do
 358       m btrbk $verbose_arg $progress_arg $cmd_arg $vol ssh://$tg$vol
 359     done
 360   fi
 361   exit 0
 362 fi
 363 # -q and just using the syslog option seemed nice,
 364 # but it doesn't show when a send has a parent and when it doesn't.
 365 m btrbk $verbose_arg $progress_arg $cmd_arg
 366
 367 # if we have it, sync to systems which don't
 368 if mountpoint $rsync_mountpoint >/dev/null; then
 369   for tg in ${targets[@]}; do
 370     case $tg in
 371       li|lk)
 372         for x in /p/c/machine_specific/*.hosts; do
 373           if grep -qxF $tg $x; then
 374             dir=${x%.hosts}
 375             rsync-dirs $tg $dir
 376           fi
 377         done
 378         ;;
 379     esac
 380   done
 381 fi
 382
 383 if [[ $source ]]; then
 384   m mount-latest-subvol $verbose_arg
 385 else
 386   m /a/exe/mount-latest-remote ${targets[@]}
 387 fi
 388
 389 exit $ret
 390
 391 # todo: move variable data we don't care about backing up
 392 # to /nocow and symlink it.
 393
 394
 395 # background on btrbk timezones. with short/long, timestamps use local time.
 396 # for long, if your local time moves backwards, by moving timezones or
 397 # for an hour when daylight savings changes it, you will temporarily get
 398 # a more aggressive retention policy for the overlapping period, and
 399 # vice versa for the opposite timezone move. The alternative is using
 400 # long-iso, which puts timezone info into the timestamp, which means
 401 # that instead of shifting time, you shift the start of day/week/month
 402 # which is used for retention to your new local time, which means for
 403 # example, if you moved forward by 8 hours, the daily/weekly/monthly
 404 # retention will be 8 hours more aggressive since midnight is at a new
 405 # time, unless you fake the timzeone using the TZ env variable.
 406 # However, in the short term, there will be no inconsistencies.
 407 # I don't see any problem with shifting when the day starts for
 408 # retention, so I'm using long-iso.
 409
 410 # note to create a long-iso timestamp: date +%Y%m%dT%H%M%S%z