btrbk: disable frodo temporarily, make more robust
[distro-setup] / 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
18 # immediately, the received subvolume doesn't get a Received UUID:
19 # field, and we won't mount it. Need to figure out a solution that will
20 # fix this.
21
22
23 [[ $EUID == 0 ]] || exec sudo -E "${BASH_SOURCE[0]}" "$@"
24
25 source /usr/local/lib/err
26
27 usage() {
28 cat <<'EOF'
29 btrbk-run [OPTIONS] [run|resume|archive]
30 usually -t TARGET_HOST or -s SOURCE_HOST
31
32 Note, at source location, intentionally not executable, run and read
33 install-my-scripts.
34
35 EOF
36 echo "top of script file:"
37 sed -n '1,/^[# ]*end command line/{p;b};q' "$0"
38 exit $1
39 }
40
41
42
43 script_name="${BASH_SOURCE[0]}"
44 script_name="${script_name##*/}"
45 pre="${SSH_CLIENT:+$HOSTNAME} $script_name:"
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 if [[ $INVOCATION_ID ]]; then
69 # INVOCATION_ID means running as a systemd service. we cant show progress in this case,
70 # but if we pass the arg, it will insert mbuffer into the command.
71 progress_arg=
72 else
73 progress_arg="--progress"
74 fi
75 incremental_strict=false
76 pull_reexec=false
77
78 default_args_file=/etc/btrbk-run.conf
79 if [[ -s $default_args_file ]]; then
80 # shellcheck disable=SC2046 # we want word splitting
81 set -- $(< $default_args_file) "$@"
82 # i havent used this feature yet, so warn about it
83 echo "$0: warning: default btrbk-run options set in $default_args_file (sleeping 5 seconds):"
84 cat $default_args_file
85 sleep 5
86 fi
87
88 targets=()
89 early=false
90 cron=false
91 fast=false
92 orig_args=("$@")
93 temp=$(getopt -l cron,fast,pull-reexec,help 23ceil:m:npqrs:t:vh "$@") || usage 1
94 eval set -- "$temp"
95 while true; do
96 case $1 in
97 # some behaviors specific to running under cron:
98 # - skip hosts where xprintidle haven't been idle recently
99 # - if we can't ssh to 1 or more hosts, still do the rest
100 # - if we aren't MAIL_HOST and no -m or -s, just exit
101 --cron)
102 cron=true
103 pre=
104 ;;
105 # for the rare case we want to run multiple instances at the same time
106 -2) conf_suf=2 ;;
107 -3) conf_suf=3 ;;
108 # only creates the config file, does not run btrbk
109 -c) conf_only=true ;;
110 # quit early, just btrbk, no extra remounting etc.
111 -e) early=true ;;
112 # skip various checks. when we run twice in a row for
113 # switch mail-host, no need to repeat the same checks again.
114 --fast) fast=true ;;
115 -i) incremental_strict=true ;;
116 # bytes per second, suffix k m g
117 -l) rate_limit=$2; shift ;;
118 # Comma separated mountpoints to backup. This has defaults set below.
119 -m) IFS=, mountpoints=($2); unset IFS; shift ;;
120 -n) dry_run=true ;;
121 # hide progress
122 -p) progress_arg= ;;
123 # internal option for rerunning under newer SOURCE_HOST version.
124 --pull-reexec) pull_reexec=true;;
125 # quiet
126 -q) verbose=false; verbose_arg=; progress_arg= ;;
127 # source host to receive a backup from
128 -s)
129 source=$2
130 bbksource=$source
131 if [[ $source == *:* ]]; then
132 bbksource="[$source]"
133 fi
134 shift
135 ;;
136 # target hosts to send to. empty is valid for just doing local
137 # snapshot. we have default hosts we will populate.
138 -t) IFS=, targets=($2); unset IFS; shift ;;
139 # verbose.
140 -v) verbose=true; verbose_arg=-v ;;
141 -h|--help) usage ;;
142 --) shift; break ;;
143 *) die "Internal error!" ;;
144 esac
145 shift
146 done
147
148 cmd_arg=${1:-run}
149
150
151 std_preserve="36h 14d 8w 24m"
152 q_preserve="18h 14d 8w"
153
154 case $cmd_arg in
155 run|resume) : ;;
156
157 # This works better than the normal archive command. We have to
158 # specify the mount points, but that is what we are used to doing and
159 # we prefer it. Another difference is that archive works recursively
160 # and we don't care about that. Sometimes we may still want to run
161 # btrbk archive, but it doesn't even use the config file, so just
162 # run it directly, eg:
163 # time s btrbk -v archive /mnt/r7/amy/boot/btrbk ssh://bo/mnt/boot2/btrbk
164 archive)
165 cmd_arg=resume
166 std_preserve="999h 999d 999w 999m"
167 q_preserve="$std_preserve"
168 preserve_arg=-p
169 ;;
170 *) die "untested command arg" ;;
171 esac
172
173 if (( $# > 1 )); then
174 die: "only 1 nonoption arg is supported"
175 fi
176
177 if [[ -v targets && $source ]]; then
178 # note, this doesnt need to be the case, but
179 # we would need to think about it.
180 die "error: -t and -s are mutually exclusive"
181 fi
182
183 if $verbose; then
184 printf "$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"
185 fi
186 ### end options parsing
187
188 # remove path from earlier version of btrbk
189 rm -f /usr/sbin/btrbk
190 # note, this still works as intended if there is no /usr/bin/btrbk
191 if [[ /a/opt/btrbk/btrbk -nt /usr/bin/btrbk ]]; then
192 if [[ -e /b/distro-functions/src/package-manager-abstractions ]]; then
193 . /b/distro-functions/src/package-manager-abstractions
194 pi asciidoctor
195 fi
196 cd /a/opt/btrbk
197 m make install
198 fi
199
200 # TODO: i wonder if there should be an option to send to the default
201 # targets, plus any given on the command line.
202
203
204
205 kd_spread=false
206 # set default targets
207 if [[ ! -v targets && ! $source ]]; then
208 if $cron; then
209 if [[ $HOSTNAME != "$MAIL_HOST" ]]; then
210 if [[ $HOSTNAME == kd && $MAIL_HOST == x3 ]]; then
211 if ping -q -c1 -w1 x3.office.fsf.org &>/dev/null; then
212 work_host=x3.office.fsf.org
213 elif ping -q -c1 -w1 x3wg.b8.nz &>/dev/null; then
214 work_host=x3wg.b8.nz
215 fi
216 if [[ $work_host ]]; then
217 source_state="$(ssh $work_host cat /a/bin/bash_unpublished/source-state)"
218 eval "$source_state"
219 if [[ $MAIL_HOST == x3 ]]; then
220 kd_spread=true
221 else
222 echo "MAIL_HOST=$MAIL_HOST, nothing to do"
223 mexit 0
224 fi
225 else
226 echo "MAIL_HOST=$MAIL_HOST, nothing to do"
227 mexit 0
228 fi
229 else
230 echo "MAIL_HOST=$MAIL_HOST, nothing to do"
231 mexit 0
232 fi
233 fi
234 fi
235
236 at_work=false
237 at_home=false
238
239 case $HOSTNAME in
240 kw|kd|frodo|x2|x3|sy) : ;;
241 *)
242 die "error: no default targets for this host, use -t"
243 ;;
244 esac
245
246 case $HOSTNAME in
247 kw)
248 at_work=true
249 ;;&
250 kd|frodo)
251 at_home=true
252 ;;&
253 x2|x3|sy)
254 if [[ $(dig +short @10.2.0.1 -x 10.2.0.2 2>&1 ||:) == kd.b8.nz. ]] \
255 && ip n show 10.2.0.1 | grep . &>/dev/null; then
256 at_home=true
257 elif ping -q -c1 -w1 hal.office.fsf.org &>/dev/null \
258 && ip n show 192.168.0.26 | grep . &>/dev/null; then
259 at_work=true
260 fi
261 ;;&
262 *)
263 if $at_home; then
264 if ! $kd_spread; then
265 # main work machine
266 if ping -q -c1 -w1 x3.office.fsf.org &>/dev/null; then
267 targets+=(x3.office.fsf.org)
268 else
269 targets+=(x3wg.b8.nz)
270 fi
271 fi
272 # temporarily disabled while doing recovery
273 # for h in frodo kd; do
274 for h in kd; do
275 if [[ $HOSTNAME == "$h" ]]; then
276 continue
277 fi
278 targets+=($h.b8.nz)
279 done
280 for h in x2 x3 sy; do
281 if [[ $HOSTNAME == "$h" ]]; then
282 continue
283 fi
284 if ping -q -c1 -w1 $h.b8.nz &>/dev/null; then
285 targets+=($h.b8.nz)
286 elif ping -q -c1 -w1 ${h}w.b8.nz &>/dev/null; then
287 targets+=(${h}w.b8.nz)
288 fi
289 done
290 elif $at_work; then
291 if ping -q -c1 -w1 iank.vpn.office.fsf.org &>/dev/null; then
292 targets+=(iank.vpn.office.fsf.org)
293 else
294 targets+=(i.b8.nz)
295 fi
296 for h in x2 x3 kw; do
297 if [[ $HOSTNAME == "$h" ]]; then
298 continue
299 fi
300 if ping -q -c1 -w1 $h.office.fsf.org &>/dev/null; then
301 targets+=($h.office.fsf.org)
302 fi
303 done
304 else
305 targets+=(i.b8.nz)
306 fi
307 ;;
308 esac
309 fi
310
311 if [[ -v targets ]]; then
312 echo "targets: ${targets[*]}"
313 fi
314
315 if [[ $source ]]; then
316 echo "source: $source"
317 fi
318
319 if [[ ${mountpoints[0]} ]]; then
320 for mp in ${mountpoints[@]}; do
321 if [[ -e /nocow/btrfs-stale/$mp ]]; then
322 die "error: $mp is stale, mount-latest-subvol first"
323 fi
324 done
325 else
326 # set default mountpoints
327 if [[ ${targets[0]} == tp ]]; then
328 prospective_mps=(/a)
329 else
330 case $HOSTNAME in
331 *)
332 prospective_mps=()
333 if [[ $source ]]; then
334 source_state="$(ssh $source cat /a/bin/bash_unpublished/source-state)"
335 eval "$source_state"
336 source_host="$(ssh $source cat /etc/hostname)"
337 if [[ $source_host == "$MAIL_HOST" ]]; then
338 prospective_mps+=(/o)
339 fi
340 if [[ $source_host == "$HOST2" ]]; then
341 prospective_mps+=(/a /ar /qr /q)
342 fi
343 else
344 if [[ $HOSTNAME == "$MAIL_HOST" ]]; then
345 prospective_mps+=(/o)
346 fi
347 if [[ $HOSTNAME == "$HOST2" ]]; then
348 prospective_mps+=(/a /ar /qr /q)
349 fi
350 fi
351 # note: put q last just in case its specific retention options were to
352 # affect other config sections. I havent tested if that is the case.
353 ;;
354 esac
355 fi
356 for mp in ${prospective_mps[@]}; do # default mountpoints to sync
357 if [[ -e /nocow/btrfs-stale/$mp ]]; then
358 e "warning: $mp stale, not adding to default mountpoints"
359 continue
360 fi
361 if awk '{print $2}' /etc/fstab | grep -xF $mp &>/dev/null; then
362 mountpoints+=($mp)
363 fi
364 done
365 fi
366
367 echo "mountpoints: ${mountpoints[*]}"
368
369 ##### end command line parsing ########
370
371 if ! $fast && [[ $source ]]; then
372 if [[ $(ssh $source ps --no-headers -o comm 1) == systemd ]]; then
373 status=$(ssh $source systemctl is-active btrbk.service) || : # normally returns 3
374 case $status in
375 inactive|failed) : ;;
376 *)
377 echo "$0: error: cron btrbk is running on source. exiting out of caution"
378 mexit 1
379 esac
380 fi
381 fi
382
383 # pull_reexec stops us from getting into an infinite loop if there is some
384 # kind of weird problem
385 pulla=false
386 for m in "${mountpoints[@]}"; do
387 if [[ $m == /a ]]; then
388 pulla=true
389 break
390 fi
391 done
392 if ! $pull_reexec && [[ $source ]] && $pulla ; then
393 tmpf=$(mktemp)
394 m rsync -ra $source:/usr/local/bin/{mount-latest-subvol,check-subvol-stale} /usr/local/bin
395 m rsync -ra $source:/usr/local/lib/err /usr/local/lib
396 m scp $source:/a/bin/distro-setup/btrbk-run $tmpf
397 if ! diff -q $tmpf ${BASH_SOURCE[0]}; then
398 e "found different version on host $source. reexecing"
399 install -T $tmpf /usr/local/bin/btrbk-run
400 m /usr/local/bin/btrbk-run --pull-reexec "${orig_args[@]}"
401 mexit 0
402 fi
403 fi
404
405
406 if ! command -v btrbk &>/dev/null; then
407 die "error: no btrbk binary found"
408 fi
409
410 if ! $fast; then
411 # if our mountpoints are from stale snapshots,
412 # it doesn't make sense to do a backup.
413 m check-subvol-stale ${mountpoints[@]} || die "found stale mountpoints in ${mountpoints[*]}"
414
415 # for an initial run, btrbk requires the dir to exist.
416 mkdir -p /mnt/{root,o}/btrbk
417 fi
418 local_zone=$(date +%z)
419
420 if [[ $source ]]; then
421 if $fast; then
422 zone=$local_zone
423 else
424 if ! zone=$(ssh root@$source date +%z); then
425 if $conf_only; then
426 echo "$0: warning: failed to ssh to root@$source"
427 else
428 die failed to ssh to root@$source
429 fi
430 fi
431 if [[ $zone != "$local_zone" ]]; then
432 die "error: dont confuse yourself with multiple time zones. $h has different timezone than localhost"
433 fi
434 fi
435 else
436
437 sshable=()
438 sshfail=()
439 min_idle_ms=$((1000 * 60 * 15))
440 for h in ${targets[@]}; do
441 if $fast || $conf_only; then
442 # Use some typical values in this case
443 root_size=$(( 1024 * 1024 * 2000 )) #2tb
444 percent_used=10
445 zone=$(date +%z)
446 elif remote_str=$(timeout -s 9 6 ssh root@$h "mkdir -p /mnt/root/btrbk /mnt/o/btrbk && date +%z && df --output=size,pcent / | tail -n1"); then
447 mapfile -t tmp_array <<<"$remote_str"
448 zone="${tmp_array[0]}"
449 IFS=" " read -r root_size percent_used <<<"${tmp_array[1]}"
450 percent_used=${percent_used%%%}
451
452 if (( ${#tmp_array[@]} != 2 )); then
453 die "error: didnt get 2 lines in test ssh to target $h. investigate"
454 fi
455 case $percent_used in
456 [0-9]|[1-9][0-9]) : ;;
457 *)
458 die "error: didnt get percent disk use in test ssh to target $h. investigate"
459 ;;
460 esac
461 else
462 sshfail+=($h)
463 continue
464 fi
465
466 # we may be booted into a bootstrap fs or something
467 min_root_kb=$(( 1024 * 1024 * 200 )) # 200 gb
468 if (( root_size < min_root_kb )); then
469 continue
470 fi
471
472 if (( percent_used >= 98 )); then
473 die "error: filesystem on target $h is $percent_used % full"
474 fi
475
476 # This is a separate ssh because xprintidle can fail and thats ok.
477 if $cron && idle_ms=$(timeout -s 9 6 ssh $h DISPLAY=:0 xprintidle); then
478 if (( idle_ms < min_idle_ms )); then
479
480 # Ignore this host. i sometimes use a non-main machine for
481 # testing or web browsing, knowing that everything will be wiped
482 # by the next backup, but I dont want it to happen as Im using
483 # it from cronjob.
484 e "warning: $h: active X session in the last 15 minutes, skipping for now"
485 continue
486 fi
487 fi
488 sshable+=($h)
489 if [[ $zone != "$local_zone" ]]; then
490 die "error: dont confuse yourself with multiple time zones. $h has different timezone than localhost"
491 fi
492 done
493 if [[ ! ${sshable[*]} ]] || { ! $cron && [[ ${sshfail[*]} ]]; }; then
494 die "failed to ssh to hosts: ${sshfail[*]}"
495 else
496 if [[ ${sshfail[*]} ]]; then
497 ret=1
498 e "error: failed to ssh to ${sshfail[*]} but continuing with other hosts"
499 fi
500 targets=(${sshable[@]})
501 fi
502 fi
503
504
505 cat >/etc/btrbk$conf_suf.conf <<EOF
506 ssh_identity /q/root/h
507 #ssh_identity /root/.ssh/home
508
509 # Just a guess that local7 is a good facility to pick.
510 # It's a bit odd that the transaction log has to be logged to
511 # a file or syslog, while other output is sent to std out.
512 # The man does not mention a way for them to be together, but
513 # I dunno if setting a log level like warn might also output
514 # transaction info.
515 transaction_syslog local7
516
517 # trying this out
518 #stream_compress zstd
519
520 # so we only run one at a time
521 lockfile /var/lock/btrbk$conf_suf.lock
522
523 # default format of short does not accomidate hourly preservation setting
524 timestamp_format long-iso
525
526 # only make a snapshot if things have changed
527 snapshot_create onchange
528 # I could make this different from target_preserve,
529 # if one disk had less space.
530 # for now, keeping them equal.
531 snapshot_preserve $std_preserve
532 snapshot_preserve_min 2h
533 snapshot_dir btrbk
534 # so, total backups = ~58
535 target_preserve $std_preserve
536 target_preserve_min 2h
537
538 # i tried this when investigating: clone no source subvolume found error
539 #incremental_prefs sro:1 srn:1 sao san:1 aro:1 arn:1
540
541 # if something fails and it's not obvious, try doing
542 # btrbk -l debug -v dryrun
543
544 rate_limit $rate_limit
545 EOF
546
547 if $incremental_strict; then
548 cat >>/etc/btrbk$conf_suf.conf <<EOF
549 incremental strict
550 EOF
551 fi
552
553 qconf() {
554 case $sub in
555 q)
556 # q has sensitive data i dont want to backup for so long
557 cat >>/etc/btrbk$conf_suf.conf <<EOF
558 snapshot_preserve $q_preserve
559 snapshot_preserve_min 2h
560 snapshot_dir btrbk
561 target_preserve $q_preserve
562 target_preserve_min 2h
563 EOF
564 ;;
565 esac
566
567 }
568
569 # make /q be last
570 mp_count=${#mountpoints[@]}
571 for (( i=0; i < mp_count - 1 ; i++ )); do
572 if [[ ${mountpoints[i]} == /q ]]; then
573 unset "mountpoints[i]"
574 mountpoints+=(/q)
575 fi
576 done
577
578 for m in ${mountpoints[@]}; do
579 case $m in
580 /o)
581 vol=/mnt/o
582 ;;
583 *)
584 vol=/mnt/root
585 ;;
586 esac
587
588 sub=${m#/}
589 if [[ $source ]]; then
590 cat >>/etc/btrbk$conf_suf.conf <<EOF
591 volume ssh://$bbksource$vol
592 subvolume $sub
593 EOF
594 qconf
595 cat >>/etc/btrbk$conf_suf.conf <<EOF
596 target send-receive $vol/btrbk
597 EOF
598 fi
599 if (( ${#targets[@]} )); then
600 cat >>/etc/btrbk$conf_suf.conf <<EOF
601 volume $vol
602 subvolume $sub
603 EOF
604 qconf
605 for tg in ${targets[@]}; do
606 # handle ipv6
607 if [[ $tg == *:* ]]; then
608 tg="[$tg]"
609 fi
610 cat >>/etc/btrbk$conf_suf.conf <<EOF
611 target send-receive ssh://$tg$vol/btrbk
612 EOF
613 done
614 fi
615 done
616
617 # todo: umount first to ensure we don't have any errors
618 # todo: do some kill fuser stuff to make umount more reliable
619
620
621 if $conf_only; then
622 mexit 0
623 fi
624
625
626
627 if $dry_run; then
628 m btrbk -c /etc/btrbk$conf_suf.conf -v -n $cmd_arg
629 mexit 0
630 fi
631 # -q and just using the syslog option seemed nice,
632 # but it doesn't show when a send has a parent and when it doesn't.
633 m btrbk -c /etc/btrbk$conf_suf.conf $preserve_arg $verbose_arg $progress_arg $cmd_arg
634
635 if $early; then
636 exit 0
637 fi
638
639 # todo: tp not valid anymore.
640 # if we have it, sync to systems which don't
641 if mountpoint $rsync_mountpoint >/dev/null; then
642 for tg in ${targets[@]}; do
643 case $tg in
644 tp)
645 dirs=(/p/c/machine_specific/tp)
646 for x in /p/c/machine_specific/*.hosts; do
647 if grep -qxF $tg $x; then
648 dirs+=(${x%.hosts})
649 fi
650 done
651 m rsync -aSAXPH --specials --devices --delete --relative ${dirs[@]} root@$tg:/
652 ;;
653 esac
654 done
655 fi
656
657 subvols=()
658 for mp in "${mountpoints[@]}"; do
659 subvols+=("${mp##*/}")
660 done
661 if [[ $source ]]; then
662 m mount-latest-subvol "${subvols[@]}"
663 else
664 for tg in ${targets[@]}; do
665 m /a/exe/mount-latest-remote "$tg" "${subvols[@]}" || ret=$?
666 done
667 fi
668
669 if [[ $ret == 0 ]]; then
670 for tg in ${targets[@]}; do
671 :
672 #ssh root@$tg /a/exe/mail-backup-clean
673 done
674 fi
675
676 mexit $ret
677
678 # todo: move variable data we don't care about backing up
679 # to /nocow and symlink it.
680
681
682 # background on btrbk timezones. with short/long, timestamps use local time.
683 # for long, if your local time moves backwards, by moving timezones or
684 # for an hour when daylight savings changes it, you will temporarily get
685 # a more aggressive retention policy for the overlapping period, and
686 # vice versa for the opposite timezone move. The alternative is using
687 # long-iso, which puts timezone info into the timestamp, which means
688 # that instead of shifting time, you shift the start of day/week/month
689 # which is used for retention to your new local time, which means for
690 # example, if you moved forward by 8 hours, the daily/weekly/monthly
691 # retention will be 8 hours more aggressive since midnight is at a new
692 # time, unless you fake the timzeone using the TZ env variable.
693 # However, in the short term, there will be no inconsistencies.
694 # I don't see any problem with shifting when the day starts for
695 # retention, so I'm using long-iso.
696
697 # note to create a long-iso timestamp: date +%Y%m%dT%H%M%S%z