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; echo "exiting with status 1" >&2; exit 1; }
  49 mexit() { echo "$pre: exiting with status $1"; exit $1; }
  50
  51 # latest $MAIL_HOST
  52 if [[ -e /b/bash_unpublished/source-state ]]; then
  53   source /b/bash_unpublished/source-state
  54 fi
  55
  56 # note q is owned by root:1000
  57
  58 mountpoints=()
  59
  60 rsync_mountpoint=/q
  61
  62 ret=0
  63 # default options
  64 conf_only=false
  65 dry_run=false # mostly for testing
  66 rate_limit=no
  67 verbose=true; verbose_arg=-v
  68 progress_arg="--progress"
  69 pull_reexec=false
  70
  71 default_args_file=/etc/btrbk-run.conf
  72 if [[ -s $default_args_file ]]; then
  73   set -- $(< $default_args_file) "$@"
  74   # i havent used this feature yet, so warn about it
  75   echo "$0: warning: default btrbk-run options set in $default_args_file (sleeping 5 seconds):"
  76   cat $default_args_file
  77   sleep 5
  78 fi
  79
  80 cron=false
  81 orig_args=("$@")
  82 temp=$(getopt -l cron,pull-reexec,help cl:m:npqs:t:vh "$@") || usage 1
  83 eval set -- "$temp"
  84 while true; do
  85   case $1 in
  86     --cron)
  87       cron=true
  88       pre=
  89       shift
  90       ;;
  91     # only creates the config file, does not run btrbk
  92     -c) conf_only=true; shift ;;
  93     # bytes per second, suffix k m g
  94     -l) rate_limit=$2; shift 2 ;;
  95     # Comma separated mountpoints to backup. This has defaults set below.
  96     -m) IFS=, mountpoints=($2); unset IFS; shift 2 ;;
  97     -n) dry_run=true; dry_run_arg=-n; shift ;;
  98     -p) progress_arg="--progress"; shift ;;
  99     --pull-reexec) pull_reexec=true; shift ;;
 100     -q) verbose=false; verbose_arg=; progress_arg=; shift ;;
 101     # source host to receive a backup from
 102     -s) source=$2; shift 2 ;;
 103     # target hosts to send to. empty is valid for just doing local
 104     # snapshot. we have default hosts we will populate.
 105     -t) IFS=, targets=($2); unset IFS; shift 2 ;;
 106     -v) verbose=true; verbose_arg=-v; shift ;;
 107     -h|--help) usage ;;
 108     --) shift; break ;;
 109     *) die "Internal error!" ;;
 110   esac
 111 done
 112
 113 # only tested commands are resume and archive
 114 cmd_arg=${1:-run}
 115
 116 case $cmd_arg in
 117   run|resume|archive) : ;;
 118   *) die "untested command arg" ;;
 119 esac
 120
 121 if (( $# > 1 )); then
 122   die: "only 1 nonoption arg is supported"
 123 fi
 124
 125 if [[ -v targets && $source ]]; then
 126   die "error: -t and -s are mutually exclusive"
 127 fi
 128
 129 if $verbose; then
 130   printf "options: conf_only=%s\ndry_run=%s\nrate_limit=%s\nverbose=%s\ncmd_arg=%s" "$conf_only" "$dry_run" "$rate_limit" "$verbose" "$cmd_arg"
 131 fi
 132 ### end options parsing
 133
 134
 135 # TODO: i wonder if there should be an option to send to the default
 136 # targets, plus any given on the command line.
 137
 138 # set default targets
 139 if [[ ! -v targets && ! $source ]]; then
 140   if [[ $HOSTNAME != "$MAIL_HOST" ]] && $cron ; then
 141     echo "MAIL_HOST=$MAIL_HOST, nothing to do"
 142     mexit 0
 143   fi
 144
 145   at_work=false
 146
 147   case $HOSTNAME in
 148     kw)
 149       at_work=true
 150       ;;&
 151     x2|x3)
 152       if ping -q -c1 -w1 hal.office.fsf.org \
 153               && ip n show 192.168.0.26 | grep . &>/dev/null; then
 154       at_work=true
 155       fi
 156       ;;&
 157     kw|x2|x3)
 158       if $at_work && ping -q -c1 -w1 iank.vpn.office.fsf.org &>/dev/null; then
 159         home=iank.vpn.office.fsf.org
 160       else
 161         home=b8.nz
 162       fi
 163       ;;&
 164     kw)
 165       targets=($home x2)
 166       ;;
 167     x2|x3)
 168       targets=($home kw)
 169       ;;
 170     tp)
 171       # kd disabled temporarily while its hot and i plan to work on it.
 172       #targets=(frodo kd)
 173       targets=(frodo x3.b8.nz)
 174       # might not be connected to the vpn
 175       if timeout -s 9 10 ssh kw :; then
 176         targets+=(kw)
 177       fi
 178       ;;
 179     kd)
 180       targets=(frodo tp)
 181       # might not be connected to the vpn
 182       if timeout -s 9 10 ssh kw :; then
 183         targets+=(kw)
 184       fi
 185       ;;
 186     *)
 187       die "error: no default targets for this host, use -t"
 188       ;;
 189   esac
 190 fi
 191
 192 if [[ -v targets ]]; then
 193   echo "targets: ${targets[*]}"
 194 fi
 195
 196 if [[ $source ]]; then
 197   echo "source: $source"
 198 fi
 199
 200
 201 if [[ $mountpoints ]]; then
 202   for mp in ${mountpoints[@]}; do # default mountpoints to sync
 203     if [[ -e /nocow/btrfs-stale/$mp ]]; then
 204       die "error: $mp is stale, mount-latest-subvol first"
 205     fi
 206   done
 207 else
 208   # set default mountpoints
 209   case $HOSTNAME in
 210     # no remote backups atm. note, if we do enable this, configuration below will need some changes.
 211     #  frodo)
 212     #    prospective_mps=(/i)
 213     #    ;;
 214     *)
 215       prospective_mps=(/a /q)
 216       if [[ $source ]]; then
 217         source_state="$(ssh $source cat /a/bin/bash_unpublished/source-state)"
 218         eval "$source_state"
 219         source_host="$(ssh $source cat /etc/hostname)"
 220         if [[ $source_host == "$MAIL_HOST" ]]; then
 221           prospective_mps+=(/o)
 222         fi
 223       else
 224         if [[ $HOSTNAME == "$MAIL_HOST" ]]; then
 225           prospective_mps+=(/o)
 226         fi
 227       fi
 228       ;;
 229   esac
 230   for mp in ${prospective_mps[@]}; do # default mountpoints to sync
 231     if [[ -e /nocow/btrfs-stale/$mp ]]; then
 232       e "warning: $mp stale, not adding to default mountpoints"
 233       continue
 234     fi
 235     if awk '{print $2}' /etc/fstab | grep -xF $mp &>/dev/null; then
 236       mountpoints+=($mp)
 237     fi
 238   done
 239 fi
 240
 241 echo "mountpoints: ${mountpoints[*]}"
 242
 243 ##### end command line parsing ########
 244
 245 if [[ $source ]]; then
 246   if [[ $(ssh $source systemctl is-active btrbk.service) != inactive ]]; then
 247     echo "$0: error: cron btrbk is running on source. exiting out of caution"
 248     mexit 1
 249   fi
 250 fi
 251
 252 # pull_reexec stops us from getting into an infinite loop if there is some
 253 # kind of weird problem
 254 pulla=false
 255 for m in "${mountpoints[@]}"; do
 256   if [[ $m == /a ]]; then
 257     pulla=true
 258     break
 259   fi
 260 done
 261 if ! $pull_reexec && [[ $source ]] && $pulla ; then
 262   tmpf=$(mktemp)
 263   scp $source:/a/bin/distro-setup/btrbk-run $tmpf
 264   if ! diff -q $tmpf $BASH_SOURCE; then
 265     e "found newer version on host $source. reexecing"
 266     install -T $tmpf /usr/local/bin/btrbk-run
 267     m /usr/local/bin/btrbk-run --pull-reexec "${orig_args[@]}"
 268     mexit 0
 269   fi
 270 fi
 271
 272
 273
 274
 275
 276 if ! which btrbk &>/dev/null; then
 277   die "error: no btrbk binary found"
 278 fi
 279 # if our mountpoints are from stale snapshots,
 280 # it doesn't make sense to do a backup.
 281 check-subvol-stale ${mountpoints[@]} || die "found stale mountpoints in ${mountpoints[*]}"
 282
 283 # for an initial run, btrbk requires the dir to exist.
 284 mkdir -p /mnt/root/btrbk
 285 local_zone=$(date +%z)
 286
 287 if [[ $source ]]; then
 288   if ! zone=$(ssh root@$source date +%z); then
 289     die failed to ssh to root@$source
 290   fi
 291   if [[ $zone != $local_zone ]]; then
 292     die "error: dont confuse yourself with multiple time zones. $h has different timezone than localhost"
 293   fi
 294
 295 else
 296
 297   sshable=()
 298   sshfail=()
 299   min_idle_ms=$((1000 * 60 * 15))
 300   for h in ${targets[@]}; do
 301     if zone=$(ssh root@$h "mkdir -p /mnt/root/btrbk && date +%z"); then
 302       # This is a separate ssh because xprintidle can fail and thats ok.
 303       if $cron && idle_ms=$(ssh $h DISPLAY=:0 xprintidle); then
 304         if (( idle_ms < min_idle_ms )); then
 305
 306           # Ignore this host. i sometimes use a non-main machine for
 307           # testing or web browsing, knowing that everything will be wiped
 308           # by the next backup, but I dont want it to happen as Im using
 309           # it from cronjob.
 310           e "warning: $h: active X session in the last 15 minutes, skipping for now"
 311           continue
 312         fi
 313       fi
 314       sshable+=($h)
 315       if [[ $zone != $local_zone ]]; then
 316         die "error: dont confuse yourself with multiple time zones. $h has different timezone than localhost"
 317       fi
 318     else
 319       sshfail+=($h)
 320     fi
 321   done
 322   if [[ ! ${sshable[@]} ]] || { ! $cron && [[ ${sshfail[@]} ]]; }; then
 323     die "failed to ssh to hosts: ${sshfail[*]}"
 324   else
 325     if [[ ${sshfail[@]} ]]; then
 326       ret=1
 327       e "error: failed to ssh to ${sshfail[*]} but continuing with other hosts"
 328     fi
 329     targets=(${sshable[@]})
 330   fi
 331 fi
 332
 333
 334 cat >/etc/btrbk.conf <<EOF
 335 ssh_identity /root/.ssh/home
 336 # Just a guess that local7 is a good facility to pick.
 337 # It's a bit odd that the transaction log has to be logged to
 338 # a file or syslog, while other output is sent to std out.
 339 # The man does not mention a way for them to be together, but
 340 # I dunno if setting a log level like warn might also output
 341 # transaction info.
 342 transaction_syslog local7
 343
 344 # note, i had this because man said 20% speedup, but ran into
 345 # this issue, https://github.com/digint/btrbk/issues/275
 346 #stream_buffer 512m
 347
 348 # so we only run one at a time
 349 lockfile                   /var/lock/btrbk.lock
 350
 351 # default format of short does not accomidate hourly preservation setting
 352 timestamp_format long-iso
 353
 354 # only make a snapshot if things have changed
 355 snapshot_create onchange
 356 # I could make this different from target_preserve,
 357 # if one disk had less space.
 358 # for now, keeping them equal.
 359 snapshot_preserve 36h 14d 8w 24m
 360 snapshot_preserve_min 4h
 361 snapshot_dir btrbk
 362
 363 # so, total backups = ~89
 364 target_preserve 36h 14d 8w 24m
 365 target_preserve_min 4h
 366
 367 # if something fails and it's not obvious, try doing
 368 # btrbk -l debug -v dryrun
 369
 370 rate_limit $rate_limit
 371 EOF
 372
 373
 374
 375
 376
 377 vol=/mnt/root
 378 for m in ${mountpoints[@]}; do
 379   sub=${m##*/}
 380   if [[ $source ]]; then
 381     cat >>/etc/btrbk.conf <<EOF
 382 volume ssh://$source$vol
 383 subvolume $sub
 384 target send-receive $vol/btrbk
 385 EOF
 386   else
 387     cat >>/etc/btrbk.conf <<EOF
 388 volume $vol
 389 subvolume $sub
 390 EOF
 391     for tg in ${targets[@]}; do
 392       cat >>/etc/btrbk.conf <<EOF
 393 target send-receive ssh://$tg$vol/btrbk
 394 EOF
 395     done
 396   fi
 397 done
 398
 399
 400 # todo: umount first to ensure we don't have any errors
 401 # todo: do some kill fuser stuff to make umount more reliable
 402
 403
 404 if $conf_only; then
 405   mexit 0
 406 fi
 407
 408
 409
 410 if $dry_run; then
 411   m btrbk -v -n $cmd_arg
 412   mexit 0
 413 elif [[ $cmd_arg == archive ]]; then
 414   if [[ $source ]]; then
 415     m btrbk $verbose_arg $progress_arg $cmd_arg ssh://$source$vol $vol
 416   else
 417     for tg in ${targets[@]}; do
 418       m btrbk $verbose_arg $progress_arg $cmd_arg $vol ssh://$tg$vol
 419     done
 420   fi
 421   mexit 0
 422 fi
 423 # -q and just using the syslog option seemed nice,
 424 # but it doesn't show when a send has a parent and when it doesn't.
 425 m btrbk $verbose_arg $progress_arg $cmd_arg
 426
 427 # if we have it, sync to systems which don't
 428 if mountpoint $rsync_mountpoint >/dev/null; then
 429   for tg in ${targets[@]}; do
 430     case $tg in
 431       li|lk)
 432         for x in /p/c/machine_specific/*.hosts; do
 433           if grep -qxF $tg $x; then
 434             dir=${x%.hosts}
 435             rsync-dirs $tg $dir
 436           fi
 437         done
 438         ;;
 439     esac
 440   done
 441 fi
 442
 443 if [[ $source ]]; then
 444   m mount-latest-subvol $verbose_arg
 445 else
 446   m /a/exe/mount-latest-remote ${targets[@]}
 447 fi
 448
 449 mexit $ret
 450
 451 # todo: move variable data we don't care about backing up
 452 # to /nocow and symlink it.
 453
 454
 455 # background on btrbk timezones. with short/long, timestamps use local time.
 456 # for long, if your local time moves backwards, by moving timezones or
 457 # for an hour when daylight savings changes it, you will temporarily get
 458 # a more aggressive retention policy for the overlapping period, and
 459 # vice versa for the opposite timezone move. The alternative is using
 460 # long-iso, which puts timezone info into the timestamp, which means
 461 # that instead of shifting time, you shift the start of day/week/month
 462 # which is used for retention to your new local time, which means for
 463 # example, if you moved forward by 8 hours, the daily/weekly/monthly
 464 # retention will be 8 hours more aggressive since midnight is at a new
 465 # time, unless you fake the timzeone using the TZ env variable.
 466 # However, in the short term, there will be no inconsistencies.
 467 # I don't see any problem with shifting when the day starts for
 468 # retention, so I'm using long-iso.
 469
 470 # note to create a long-iso timestamp: date +%Y%m%dT%H%M%S%z