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 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=($2); unset IFS; shift 2 ;;
 104     -v) verbose=true; verbose_arg=-v; shift ;;
 105     -h|--help) usage ;;
 106     --) shift; break ;;
 107     *) die "Internal error!" ;;
 108   esac
 109 done
 110
 111 # usefull commands are resume and archive
 112 cmd_arg=${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 # TODO: i wonder if there should be an option to send to the default
 124 # targets, plus any given on the command line.
 125
 126 # set default targets
 127 if [[ ! -v targets && ! $source ]]; then
 128   if [[ $HOSTNAME != "$MAIL_HOST" ]] && $cron ; then
 129     echo "We are not MAIL_HOST, nothing to do"
 130     exit 0
 131   fi
 132   case $HOSTNAME in
 133     kw|x2)
 134       if ping -q -c1 -w1 iank.vpn.office.fsf.org &>/dev/null; then
 135         home=iank.vpn.office.fsf.org
 136       else
 137         home=$HOME_DOMAIN
 138       fi
 139       ;;&
 140     kw)
 141       targets=($home x2)
 142       ;;
 143     x2)
 144       targets=($home kw)
 145       ;;
 146     tp)
 147       targets=(frodo kd)
 148       # might not be connected to the vpn
 149       if timeout -s 9 10 ssh kw :; then
 150         targets+=(kw)
 151       fi
 152       ;;
 153     kd)
 154       targets=(frodo tp)
 155       # might not be connected to the vpn
 156       if timeout -s 9 10 ssh kw :; then
 157         targets+=(kw)
 158       fi
 159       ;;
 160     *)
 161       die "error: no default targets for this host, use -t"
 162       ;;
 163   esac
 164 fi
 165
 166 if [[ -v targets ]]; then
 167   echo "targets: ${targets[*]}"
 168 fi
 169
 170 if [[ $source ]]; then
 171   echo "source: $source"
 172 fi
 173
 174
 175 if [[ $mountpoints ]]; then
 176   for mp in ${mountpoints[@]}; do # default mountpoints to sync
 177     if [[ -e /nocow/btrfs-stale/$mp ]]; then
 178       die "error: $mp is stale, mount-latest-subvol first"
 179     fi
 180   done
 181 else
 182   # set default mountpoints
 183   case $HOSTNAME in
 184     # no remote backups atm. note, if we do enable this, configuration below will need some changes.
 185     #  frodo)
 186     #    prospective_mps=(/i)
 187     #    ;;
 188     *)
 189       prospective_mps=(/a /q)
 190       if [[ $HOSTNAME == "$MAIL_HOST" ]]; then
 191         prospective_mps+=(/o)
 192       fi
 193       ;;
 194   esac
 195   for mp in ${prospective_mps[@]}; do # default mountpoints to sync
 196     if [[ -e /nocow/btrfs-stale/$mp ]]; then
 197       echo "$pre warning: $mp stale, not adding to default mountpoints"
 198       continue
 199     fi
 200     if awk '{print $2}' /etc/fstab | grep -xF $mp &>/dev/null; then
 201       mountpoints+=($mp)
 202     fi
 203   done
 204 fi
 205
 206 echo "mountpoints: ${mountpoints[*]}"
 207
 208
 209
 210 # pull_reexec stops us from getting into an infinite loop if there is some
 211 # kind of weird problem
 212 pulla=false
 213 for m in "${mountpoints[@]}"; do
 214   if [[ $m == /a ]]; then
 215     pulla=true
 216     break
 217   fi
 218 done
 219 if ! $pull_reexec && [[ $source ]] && $pulla ; then
 220   tmpf=$(mktemp)
 221   scp $source:/a/bin/distro-setup/btrbk-run $tmpf
 222   if ! diff -q $tmpf $BASH_SOURCE; then
 223     echo "$pre found newer version on host $source. reexecing"
 224     install -T $tmpf /usr/local/bin/btrbk-run
 225     m /usr/local/bin/btrbk-run --pull-reexec "${orig_args[@]}"
 226     exit
 227   fi
 228 fi
 229
 230
 231 ##### end command line parsing ########
 232
 233
 234
 235 if ! which btrbk &>/dev/null; then
 236   die "error: no btrbk binary found"
 237 fi
 238 # if our mountpoints are from stale snapshots,
 239 # it doesn't make sense to do a backup.
 240 check-subvol-stale ${mountpoints[@]} || die "found stale mountpoints in ${mountpoints[*]}"
 241
 242 # for an initial run, btrbk requires the dir to exist.
 243 mkdir -p /mnt/root/btrbk
 244 local_zone=$(date +%z)
 245
 246 if [[ $source ]]; then
 247   if ! zone=$(ssh root@$source date +%z); then
 248     die failed to ssh to root@$source
 249   fi
 250   if [[ $zone != $local_zone ]]; then
 251     die "error: dont confuse yourself with multiple time zones. $h has different timezone than localhost"
 252   fi
 253
 254 else
 255
 256   sshable=()
 257   sshfail=()
 258   for h in ${targets[@]}; do
 259     if zone=$(ssh root@$h "mkdir -p /mnt/root/btrbk && date +%z"); then
 260       sshable+=($h)
 261       if [[ $zone != $local_zone ]]; then
 262         die "error: dont confuse yourself with multiple time zones. $h has different timezone than localhost"
 263       fi
 264     else
 265       sshfail+=($h)
 266     fi
 267   done
 268   if [[ ! $sshable ]] || { ! $cron && [[ $sshfail ]]; }; then
 269     die "failed to ssh to hosts: ${sshfail[*]}"
 270   else
 271     if [[ $sshfail ]]; then
 272       ret=1
 273       echo "$pre error: failed to ssh to ${sshfail[*]} but continuing with other hosts"
 274     fi
 275     targets=(${sshable[@]})
 276   fi
 277 fi
 278
 279
 280 cat >/etc/btrbk.conf <<EOF
 281 ssh_identity /root/.ssh/home
 282 # Just a guess that local7 is a good facility to pick.
 283 # It's a bit odd that the transaction log has to be logged to
 284 # a file or syslog, while other output is sent to std out.
 285 # The man does not mention a way for them to be together, but
 286 # I dunno if setting a log level like warn might also output
 287 # transaction info.
 288 transaction_syslog local7
 289
 290 # note, i had this because man said 20% speedup, but ran into
 291 # this issue, https://github.com/digint/btrbk/issues/275
 292 #stream_buffer 512m
 293
 294 # so we only run one at a time
 295 lockfile                   /var/lock/btrbk.lock
 296
 297 # default format of short does not accomidate hourly preservation setting
 298 timestamp_format long-iso
 299
 300 # only make a snapshot if things have changed
 301 snapshot_create onchange
 302 # I could make this different from target_preserve,
 303 # if one disk had less space.
 304 # for now, keeping them equal.
 305 snapshot_preserve 36h 14d 8w 24m
 306 snapshot_preserve_min 4h
 307 snapshot_dir btrbk
 308
 309 # so, total backups = ~89
 310 target_preserve 36h 14d 8w 24m
 311 target_preserve_min 4h
 312
 313 # if something fails and it's not obvious, try doing
 314 # btrbk -l debug -v dryrun
 315
 316 rate_limit $rate_limit
 317 EOF
 318
 319
 320
 321
 322
 323 vol=/mnt/root
 324 for m in ${mountpoints[@]}; do
 325   sub=${m##*/}
 326   if [[ $source ]]; then
 327     cat >>/etc/btrbk.conf <<EOF
 328 volume ssh://$source$vol
 329 subvolume $sub
 330 target send-receive $vol/btrbk
 331 EOF
 332   else
 333     cat >>/etc/btrbk.conf <<EOF
 334 volume $vol
 335 subvolume $sub
 336 EOF
 337     for tg in ${targets[@]}; do
 338       cat >>/etc/btrbk.conf <<EOF
 339 target send-receive ssh://$tg$vol/btrbk
 340 EOF
 341     done
 342   fi
 343 done
 344
 345
 346 # todo: umount first to ensure we don't have any errors
 347 # todo: do some kill fuser stuff to make umount more reliable
 348
 349
 350 if $conf_only; then
 351   exit
 352 fi
 353
 354
 355
 356 if $dry_run; then
 357   m btrbk -v -n $cmd_arg
 358   exit 0
 359 elif [[ $cmd_arg == archive ]]; then
 360   if [[ $source ]]; then
 361     m btrbk $verbose_arg $progress_arg $cmd_arg ssh://$source$vol $vol
 362   else
 363     for tg in ${targets[@]}; do
 364       m btrbk $verbose_arg $progress_arg $cmd_arg $vol ssh://$tg$vol
 365     done
 366   fi
 367   exit 0
 368 fi
 369 # -q and just using the syslog option seemed nice,
 370 # but it doesn't show when a send has a parent and when it doesn't.
 371 m btrbk $verbose_arg $progress_arg $cmd_arg
 372
 373 # if we have it, sync to systems which don't
 374 if mountpoint $rsync_mountpoint >/dev/null; then
 375   for tg in ${targets[@]}; do
 376     case $tg in
 377       li|lk)
 378         for x in /p/c/machine_specific/*.hosts; do
 379           if grep -qxF $tg $x; then
 380             dir=${x%.hosts}
 381             rsync-dirs $tg $dir
 382           fi
 383         done
 384         ;;
 385     esac
 386   done
 387 fi
 388
 389 if [[ $source ]]; then
 390   m mount-latest-subvol $verbose_arg
 391 else
 392   m /a/exe/mount-latest-remote ${targets[@]}
 393 fi
 394
 395 exit $ret
 396
 397 # todo: move variable data we don't care about backing up
 398 # to /nocow and symlink it.
 399
 400
 401 # background on btrbk timezones. with short/long, timestamps use local time.
 402 # for long, if your local time moves backwards, by moving timezones or
 403 # for an hour when daylight savings changes it, you will temporarily get
 404 # a more aggressive retention policy for the overlapping period, and
 405 # vice versa for the opposite timezone move. The alternative is using
 406 # long-iso, which puts timezone info into the timestamp, which means
 407 # that instead of shifting time, you shift the start of day/week/month
 408 # which is used for retention to your new local time, which means for
 409 # example, if you moved forward by 8 hours, the daily/weekly/monthly
 410 # retention will be 8 hours more aggressive since midnight is at a new
 411 # time, unless you fake the timzeone using the TZ env variable.
 412 # However, in the short term, there will be no inconsistencies.
 413 # I don't see any problem with shifting when the day starts for
 414 # retention, so I'm using long-iso.
 415
 416 # note to create a long-iso timestamp: date +%Y%m%dT%H%M%S%z