[distro-setup] / btrbk-run

#!/bin/bash

# Configure & run btrbk & related work on Ian's computers.
# Copyright (C) 2024  Ian Kelling

# This program is free software: you can redistribute it and/or modify
# it under the terms of the GNU General Public License as published by
# the Free Software Foundation, either version 3 of the License, or
# (at your option) any later version.

# This program is distributed in the hope that it will be useful,
# but WITHOUT ANY WARRANTY; without even the implied warranty of
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
# GNU General Public License for more details.

# You should have received a copy of the GNU General Public License
# along with this program.  If not, see <http://www.gnu.org/licenses/>.

# SPDX-License-Identifier: GPL-3.0-or-later


# 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.


[[ $EUID == 0 ]] || exec sudo -E "${BASH_SOURCE[0]}" "$@"

set -e; . /usr/local/lib/bash-bear; set +e
shopt -s nullglob

usage() {
  cat <<'EOF'
btrbk-run [OPTIONS] [run|resume|archive]
usually -t TARGET_HOST or -s SOURCE_HOST

Note, at source location, intentionally not executable, run and read
install-my-scripts.

EOF
  echo "top of script file:"
  sed -n '1,/^[# ]*end command line/{p;b};q' "$0"
  exit $1
}



pre=btrbk-run



script_name="${BASH_SOURCE[0]}"
script_name="${script_name##*/}"


log-setup() {
  if [[ ! $log_path ]]; then
    mkdir -p /var/log/btrbk
    log_path=/var/log/btrbk/$(date +%F_%T%:::z).log
  fi
}
d() {
  if $dry_run || $conf_only; then
    printf "$pre dry-run: %s\n"  "$*"
  else
    log-setup
    printf "$pre running: %s\n"  "$*" |& pee cat 'ts "%F %T" >>'$log_path
    "$@" |& pee cat 'ts "%F %T" >>'$log_path
  fi
}
m() { if $verbose; then printf "$pre %s\n"  "$*"; fi;  "$@"; }
e() { printf "$pre %s\n"  "$*"; }

logq() {
  local exit_code
  exit_code=0
  log-setup
  printf "$pre running: %s\n"  "$*" | pee cat 'ts "%F %T" >>'$log_path
  e logging to $log_path
  "$@" |& ts "%F %T" >>$log_path || exit_code=$?
  printf "$pre exit code:%s of %s\n" "$exit_code"  "$*" | pee cat 'ts "%F %T" >>'$log_path
  if (( exit_code > 0 )); then
    e "error: command exit code: $exit_code. exiting after tail -n50 $log_path"
    tail -n50 $log_path
    exit $exit_code
  fi
}

die() { printf "$pre error: %s\n" "$*" >&2; echo "$pre exiting with status 1" >&2; exit 1; }
mexit() { echo "$pre exiting with status $1"; exit $1; }

uninstalled-file-die() {
  die "file $1 is not latest. run install-my-scripts or rerun with -f"
}

set-location() {
  case $HOSTNAME in
    kw)
      at_work=true
      ;;
    kd|frodo)
      at_home=true
      ;;
    x2|x3|sy|so)
      if [[ $(dig +short @10.2.0.1 -x 10.2.0.2 2>&1 ||:) == kd.b8.nz. ]] \
           && ip n show 10.2.0.1 | grep . &>/dev/null; then
        # note: logic duplicated in 11-iank
        at_home=true
      elif ping -q -c1 -w1 hal.office.fsf.org &>/dev/null \
          && ip n show 192.168.0.26 | grep . &>/dev/null; then
        at_work=true
      fi
      ;;
  esac
}

exit-if-no-default-targets() {
  if ! $force && [[ $HOSTNAME != "$MAIL_HOST" ]]; then
    echo "MAIL_HOST=$MAIL_HOST, nothing to do"
    mexit 0
  fi
  case $HOSTNAME in
    kw|kd|frodo|x2|x3|sy|so) : ;;
    *)
      die "error: no default targets for this host, use -t"
      ;;
  esac
}

add-x3-target() {
  # main work machine
  if ping -q -c1 -w1 x3.office.fsf.org &>/dev/null; then
    targets+=(x3.office.fsf.org)
  elif ping -q -c1 -w1 x3.b8.nz &>/dev/null; then
    # in case we took it home
    targets+=(x3.b8.nz)
  elif ping -q -c1 -w1 x3w.b8.nz &>/dev/null; then
    targets+=(x3w.b8.nz)
  else
    targets+=(x3wg.b8.nz)
  fi
}

add-wireless-target() {
  local host
  if [[ ! $1 ]]; then
    set -- $h
  fi
  for host; do
    # c = cabled, w = wireless
    if ping -q -c1 -w1 ${host}c.b8.nz &>/dev/null; then
      targets+=(${host}c.b8.nz)
    elif ping -q -c1 -w1 $host.b8.nz &>/dev/null; then
      targets+=($host.b8.nz)
    elif ping -q -c1 -w1 ${host}w.b8.nz &>/dev/null; then
      targets+=(${host}w.b8.nz)
    fi
  done
}

qconf() {
  case $sub in
    q)
      # q has sensitive data i dont want to backup for so long
      cat >>/etc/btrbk$conf_suf.conf <<EOF
snapshot_preserve $q_preserve
snapshot_preserve_min 2h
snapshot_dir btrbk
target_preserve $q_preserve
target_preserve_min 2h
EOF
      ;;
  esac

}


# latest $MAIL_HOST
if [[ -e /b/bash_unpublished/source-state ]]; then
  source /b/bash_unpublished/source-state
fi

# note q is owned by root:1000

mountpoints=()

rsync_mountpoint=/q

ret=0
# default options
conf_only=false
dry_run=false # mostly for testing
rate_limit=no
verbose=true; verbose_arg="-l trace"
force=false
if [[ $INVOCATION_ID ]]; then
  # INVOCATION_ID means running as a systemd service. we cant show progress in this case,
  # but if we pass the arg, it will insert mbuffer into the command.
  progress_arg=
else
  progress_arg="--progress"
fi
incremental_strict=false
pull_reexec=false

default_args_file=/etc/btrbk-run.conf
if [[ -s $default_args_file ]]; then
  # shellcheck disable=SC2046 # we want word splitting
  set -- $(< $default_args_file) "$@"
  # i havent used this feature yet, so warn about it
  echo "$0: warning: default btrbk-run options set in $default_args_file (sleeping 5 seconds):"
  cat $default_args_file
  sleep 5
fi

once_args_file=/etc/btrbk-run-once.conf
if mv -f $once_args_file $once_args_file-tmp 2> >(sed '/No such file or directory/d'); then
  # shellcheck disable=SC2046 # we want word splitting
  set -- $(< $once_args_file-tmp) "$@"
  # i havent used this feature yet, so warn about it
  echo "$0: btrbk-run options set in $once_args_file:"
  cat $once_args_file-tmp
  rm -f $once_args_file-tmp
fi


targets=()
early=false
fast=false
kd_spread=false
check_installed=false
orig_args=("$@")
temp=$(getopt -l check-installed,fast,pull-reexec,help 23acefikl:m:npqrs:t:vh "$@") || usage 1
eval set -- "$temp"
while true; do
  case $1 in
    # for the rare case we want to run multiple instances at the same time
    -2) conf_suf=2 ;;
    -3) conf_suf=3 ;;
    -a)
      # all moiuntpoints
      mountpoints=(/a /o /qr /qd /q)
      ;;
    # only creates the config file, does not run btrbk
    -c) conf_only=true ;;
    --check-installed)
      check_installed=true
      ;;
    # quit early, just btrbk, no extra remounting etc.
    -e) early=true ;;
    # avoids some default behaviors:
    # - no skipping hosts where xprintidle haven't been idle recently
    # - exit if we can't ssh to 1 or more hosts
    # - still set default hosts despite MAIL_HOST status
    -f) force=true ;;
    # skip various checks. when we run twice in a row for
    # switch mail-host, no need to repeat the same checks again.
    --fast) fast=true ;;
    -i) incremental_strict=true ;;
    # note this implies resume and -p because it is just meant to make
    # other hosts have the same snapshots, not do any expiry or new
    # backups.
    -k) kd_spread=true ;;
    # bytes per second, suffix k m g
    -l) rate_limit=$2; shift ;;
    # Comma separated mountpoints to backup. This has defaults set below.
    -m) IFS=, mountpoints=($2); unset IFS; shift ;;
    -n) dry_run=true ;;
    # preserve existing snapshots and backups
    -p) preserve_arg=-p ;;
    # internal option for rerunning under newer SOURCE_HOST version.
    --pull-reexec) pull_reexec=true;;
    # quiet
    -q) verbose=false; verbose_arg=; progress_arg= ;;
    # source host to receive a backup from
    -s)
      source=$2
      bbksource=$source
      if [[ $source == *:* ]]; then
        bbksource="[$source]"
      fi
      shift
      ;;
    # target hosts to send to. empty is valid for just doing local
    # snapshot. we have default hosts we will populate.
    -t) IFS=, targets=($2); unset IFS; shift ;;
    # verbose.
    -v)
      verbose=true; verbose_arg="-l trace"
      ;;
    -h|--help) usage ;;
    --) shift; break ;;
    *) die "Internal error!" ;;
  esac
  shift
done

cmd_arg="$1"



if ! $force && { $check_installed || [[ ! $source ]]; } ; then
  install_bin_files=(
    mount-latest-subvol
    check-subvol-stale
    btrbk-run
  )
  for f in ${install_bin_files[@]}; do
    if ! diff -q /a/bin/ds/$f /usr/local/bin/$f; then
      uninstalled-file-die $f
    fi
  done
  if ! diff -q /a/bin/bash-bear-trap/bash-bear /usr/local/lib/bash-bear; then
    uninstalled-file-die bash-bear
  fi
  if $check_installed; then
    exit 0
  fi
fi


if $kd_spread; then
  if [[ $cmd_arg && $cmd_arg != resume ]]; then
    die "dont pass -k without resume or empty run arg"
  fi
  if [[ $HOSTNAME == "$MAIL_HOST" ]]; then
    die "something went wrong, -k not meant to be run on MAIL_HOST"
  fi
  if [[ $HOSTNAME != kd ]]; then
    die "something went wrong, -k only meant to run on kd"
  fi
  cmd_arg=resume
  preserve_arg=-p
  add-wireless-target sy so
fi

if [[ ! $cmd_arg ]]; then
  cmd_arg=run
fi


std_preserve="36h 14d 8w 24m"
q_preserve="18h 14d 8w"

case $cmd_arg in
  run|resume) : ;;

  # This works better than the normal archive command. We have to
  # specify the mount points, but that is what we are used to doing and
  # we prefer it. Another difference is that archive works recursively
  # and we don't care about that. Sometimes we may still want to run
  # btrbk archive, but it doesn't even use the config file, so just
  # run it directly, eg:
  # time s btrbk -v archive /mnt/r7/amy/boot/btrbk ssh://bo/mnt/boot2/btrbk
  archive)
    cmd_arg=resume
    std_preserve="999h 999d 999w 999m"
    q_preserve="$std_preserve"
    preserve_arg=-p
    ;;
  *) die "untested command arg" ;;
esac

if (( $# > 1 )); then
  die: "only 1 nonoption arg is supported"
fi

if [[ -v targets && $source ]]; then
  # note, this doesnt need to be the case, but
  # we would need to think about it.
  die "error: -t and -s are mutually exclusive"
fi

### end options parsing

# remove path from earlier version of btrbk
rm -f /usr/sbin/btrbk
# note, this still works as intended if there is no /usr/bin/btrbk
if [[ /a/opt/btrbk/btrbk -nt /usr/bin/btrbk ]]; then
  if [[ -e /b/distro-functions/src/package-manager-abstractions ]]; then
    . /b/distro-functions/src/package-manager-abstractions
    pi asciidoctor
  fi
  cd /a/opt/btrbk
  m make install
  cd /
fi

# TODO: i wonder if there should be an option to send to the default
# targets, plus any given on the command line.


at_work=false
at_home=false


# set default targets
if [[ ! -v targets && ! $source ]]; then
  exit-if-no-default-targets
  set-location
  if $at_home; then
    if ! $kd_spread && [[ $HOSTNAME != x3 ]]; then
      add-x3-target
    fi
    if [[ $HOSTNAME != kd ]]; then
      targets+=(kd.b8.nz)
    fi
    wireless_home_hosts=(
      x2
      sy
      so
    )
    for h in ${wireless_home_hosts[@]}; do
      if [[ $HOSTNAME != "$h" ]]; then
        add-wireless-target
      fi
    done
  elif $at_work; then
    targets+=(b8.nz)
    for h in x2 x3 kw; do
      if [[ $HOSTNAME == "$h" ]]; then
        continue
      fi
      if ping -q -c1 -w1 $h.office.fsf.org &>/dev/null; then
        targets+=($h.office.fsf.org)
      fi
    done
  else
    targets+=(b8.nz)
  fi
fi

if [[ ${mountpoints[0]} ]]; then
  for mp in ${mountpoints[@]}; do
    if [[ -e /nocow/btrfs-stale/$mp ]]; then
      die "error: $mp is stale, mount-latest-subvol first"
    fi
  done
else
  # set default mountpoints
  if [[ ${targets[0]} == tp ]]; then
    prospective_mps=(/a)
  else
    case $HOSTNAME in
      *)
        prospective_mps=()
        if [[ $source ]]; then
          source_state="$(ssh $source 'cat /a/bin/bash_unpublished/source-state; echo source_host=$HOSTNAME')"
          eval "$source_state"
          # shellcheck disable=SC2154 # assigned in the above eval.
          if [[ $source_host == "$MAIL_HOST" ]]; then
            prospective_mps+=(/o)
          fi
          if [[ $source_host == "$HOST2" ]]; then
            prospective_mps+=(/a /qr /qd /q)
          fi
        else
          if [[ $HOSTNAME == "$MAIL_HOST" ]]; then
            prospective_mps+=(/o)
          fi
          if [[ $HOSTNAME == "$HOST2" ]]; then
            prospective_mps+=(/a /qr /qd /q)
          fi
          if $kd_spread; then
            prospective_mps=(/a /o /qr /qd /q)
          fi
        fi
        # note: put q last just in case its specific retention options were to
        # affect other config sections. I havent tested if that is the case.
        ;;
    esac
  fi
  for mp in ${prospective_mps[@]}; do # default mountpoints to sync
    if [[ -e /nocow/btrfs-stale/$mp ]]; then
      e "warning: $mp stale, not adding to default mountpoints"
      continue
    fi
    if awk '{print $2}' /etc/fstab | grep -xF $mp &>/dev/null; then
      mountpoints+=($mp)
    fi
  done
fi

tmp=$(( ${#mountpoints[@]} == 0 ))
if (( tmp )); then
  die didnt get mountpoint arg and had no defaults
fi

##### end command line parsing ########

#### begin pre-checks #####

# todo: this has a timing problem, since btrbk.timer could activate the service after this check.
if ! $fast && [[ $source ]]; then
  if [[ $(ssh $source ps --no-headers -o comm 1) == systemd ]]; then
    status=$(ssh $source systemctl is-active btrbk.service) || : # normally returns 3
    case $status in
      inactive|failed) : ;;
      *)
        echo "$0: error: btrbk is running on source. exiting out of caution"
        mexit 1
    esac
  fi
fi

if ! command -v btrbk &>/dev/null; then
  die "error: no btrbk binary found"
fi

# pull_reexec stops us from getting into an infinite loop if there is some
# kind of weird problem
pulla=false
for m in "${mountpoints[@]}"; do
  if [[ $m == /a ]]; then
    pulla=true
    break
  fi
done

if ! $pull_reexec && [[ $source ]] && $pulla && ! $force ; then
  ssh root@$source btrbk-run --check-installed
fi

#### end pre-checks #####




# print some non-default opts
if $verbose; then
  opts_show=()
  if ! $conf_only; then
    opts_show+=(conf_only=true)
  fi
  if ! $dry_run; then
    opts_show+=(dry_run=true)
  fi
  if [[ $rate_limit != no ]]; then
    opts_show+=("rate_limit=$rate_limit")
  fi
  if [[ $cmd_arg != run ]]; then
    opts_show+=(cmd_arg=$cmd_arg)
  fi
  if (( ${#opts_show[@]} >= 1 )); then
    first=true
    for opt in ${opts_show[@]}; do
      if $first; then
        printf "%s" "$opt"
        first=false
      else
        printf " %s" "$opt"
      fi
    done
    echo
  fi
fi


if ! $pull_reexec && [[ $source ]] && $pulla ; then
  tmpf=$(mktemp)
  m rsync -ra $source:/usr/local/bin/{mount-latest-subvol,check-subvol-stale} /usr/local/bin
  m rsync -ra $source:/usr/local/lib/bash-bear /usr/local/lib
  m scp $source:/a/bin/distro-setup/btrbk-run $tmpf
  if ! diff -q $tmpf ${BASH_SOURCE[0]}; then
    e "found different version on host $source. reexecing"
    install -T $tmpf /usr/local/bin/btrbk-run
    m /usr/local/bin/btrbk-run --pull-reexec "${orig_args[@]}"
    mexit 0
  fi
fi


if [[ -v targets ]]; then
  echo "targets: ${targets[*]}"
fi
if [[ $source ]]; then
  echo "source: $source"
fi
echo "mountpoints: ${mountpoints[*]}"


# todo: check if we have no snapshots yet, because I always want to run
# archive instead of run. Likely, I should give an error unless a cli
# override is passed. perhaps check-subvol-stale could give the error.
# see the error message "no snapshots found" in that file.
if ! $fast; then
  # if our mountpoints are from stale snapshots,
  # it doesn't make sense to do a backup.
  m check-subvol-stale ${mountpoints[@]} || die "found stale mountpoints in ${mountpoints[*]}"

  # for an initial run, btrbk requires the dir to exist.
  mkdir -p /mnt/{root,o}/btrbk
fi
local_zone=$(date +%z)

if [[ $source ]]; then
  if $fast; then
    zone=$local_zone
  else
    if ! zone=$(ssh root@$source date +%z); then
      if $conf_only; then
        echo "$0: warning: failed to ssh to root@$source"
      else
        die failed to ssh to root@$source
      fi
    fi
    if [[ $zone != "$local_zone" ]]; then
      die "error: dont confuse yourself with multiple time zones. $h has different timezone than localhost"
    fi
  fi
else

  sshable=()
  sshfail=()
  remote_str_cmd="mkdir -p /mnt/root/btrbk /mnt/o/btrbk && \
date +%z && \
df --output=size,pcent / | tail -n1"

  for h in ${targets[@]}; do
    if $fast || $conf_only; then
      # Use some typical values in this case
      root_size=$(( 1024 * 1024 * 2000 )) #2tb
      percent_used=10
      zone=$(date +%z)
    elif remote_str=$(timeout -s 9 6 ssh root@$h "$remote_str_cmd"); then
      mapfile -t tmp_array <<<"$remote_str"
      zone="${tmp_array[0]}"
      IFS=" " read -r root_size percent_used <<<"${tmp_array[1]}"
      percent_used=${percent_used%%%}

      tmp=$(( ${#tmp_array[@]} != 2 ))
      if (( tmp )); then
        die "error: didnt get 2 lines in test ssh to target $h. investigate"
      fi
      case $percent_used in
        [0-9]|[1-9][0-9]) : ;;
        *)
          die "error: didnt get percent disk use in test ssh to target $h. investigate"
          ;;
      esac
    else
      sshfail+=($h)
      continue
    fi

    # we may be booted into a bootstrap fs or something
    min_root_kb=$(( 1024 * 1024 * 200 )) # 200 gb
    tmp=$(( root_size < min_root_kb ))
    if (( tmp )); then
      e "warning: $h: root_size=$root_size < 200gb, perhaps it is booted to bootstrap vol. skipping for now"
      continue
    fi

    tmp=$(( percent_used >= 98 ))
    if (( tmp )); then
      die "error: filesystem on target $h is $percent_used % full"
    fi

    # on sy, xprintidle is resetting every 12 seconds even when not
    # idle, i dunno why, instead we are checking if the screen is locked,
    # which is good enough.
    #
    # This is a separate ssh because the command can fail and thatis ok.
    if ! $force; then
      locked=false
      if lock_info=$(timeout -s 9 6 ssh $h DISPLAY=:0 xscreensaver-command -time); then
        if [[ $lock_info != *non-blanked* ]]; then
          locked=true
        fi
      else
        locked=true
      fi
      if ! $locked; then
        # Ignore this host. i sometimes use a non-main machine for
        # testing or web browsing, knowing that everything will be wiped
        # by the next backup, but I dont want it to happen as Im using
        # it from cronjob.
        e "warning: $h: seems to be actively in use, skipping for now"
        continue
      fi
    fi
    sshable+=($h)
    if [[ $zone != "$local_zone" ]]; then
      die "error: dont confuse yourself with multiple time zones. $h has different timezone than localhost"
    fi
  done
  if [[ ! ${sshable[*]} ]] || { $force && [[ ${sshfail[*]} ]]; }; then
    die "see skipped host warning above or sshfail hosts: ${sshfail[*]}"
  else
    if [[ ${sshfail[*]} ]]; then
      ret=1
      e "error: failed to ssh to ${sshfail[*]} but continuing with other hosts"
    fi
    targets=(${sshable[@]})
  fi
fi


cat >/etc/btrbk$conf_suf.conf <<EOF
ssh_identity /q/root/h
#ssh_identity /root/.ssh/home

# trying this out
#stream_compress zstd

# so we only run one at a time
lockfile                   /var/lock/btrbk$conf_suf.lock

# default format of short does not accomidate hourly preservation setting
timestamp_format long-iso

# only make a snapshot if things have changed
snapshot_create onchange
# I could make this different from target_preserve,
# if one disk had less space.
# for now, keeping them equal.
snapshot_preserve $std_preserve
snapshot_preserve_min 6h
snapshot_dir btrbk
# so, total backups = ~58
target_preserve $std_preserve
target_preserve_min 6h

# it seems very likely that not doing this could result in clone source not found
# errors, for example when expiry happens differently on different hosts,
# also, as btrbk does by default, if a failed send happens, on the next run it
# will warn about a stray subvolume, but then create a backup of a newer subvol
# and use an older subvol as the parent.
incremental_prefs sao:1

# if something fails and it's not obvious, try doing
# btrbk -l trace -v dryrun

rate_limit $rate_limit
EOF

if $incremental_strict; then
  cat >>/etc/btrbk$conf_suf.conf <<EOF
incremental strict
EOF
fi


# make /q be last
mp_count=${#mountpoints[@]}
for (( i=0; i < mp_count - 1 ; i++ )); do
  if [[ ${mountpoints[i]} == /q ]]; then
    unset "mountpoints[i]"
    mountpoints+=(/q)
  fi
done



snap_list_cmds=()
tg_snaps=()
declare -A source_snaps

for m in ${mountpoints[@]}; do
  case $m in
    /o)
      vol=/mnt/o
      ;;
    *)
      vol=/mnt/root
      ;;
  esac

  sub=${m#/}
  snap_list_cmds+=("echo $vol/btrbk/$sub.*;")

  if [[ $source ]]; then
    tmp_a=($vol/btrbk/$sub.*)
    tg_snaps+=("${tmp_a[*]}")
    cat >>/etc/btrbk$conf_suf.conf <<EOF
volume ssh://$bbksource$vol
subvolume $sub
EOF
    qconf
    cat >>/etc/btrbk$conf_suf.conf <<EOF
target send-receive $vol/btrbk
EOF
  else # we have targets
    for snap in "$vol/btrbk/$sub."*; do
      source_snaps[$snap]=t
    done

    cat >>/etc/btrbk$conf_suf.conf <<EOF
volume $vol
subvolume $sub
EOF
    qconf
    for tg in ${targets[@]}; do
      # handle ipv6
      if [[ $tg == *:* ]]; then
        tg="[$tg]"
      fi
      cat >>/etc/btrbk$conf_suf.conf <<EOF
target send-receive ssh://$tg$vol/btrbk
EOF
    done
  fi
done

# Delete any subvols on the receiving host that don't exist on the
# sending host.  Otherwise, the receiving host could have snapshots that
# aren't on the sending side, and thus become odd leaf subvols, and then
# btrbk could try to use them when we sync back, creating a weird tree
# instead of linear parent/child relationship. Maybe this could lead to
# a missing source subvol error, so lets avoid it.

get-orphan-tg-snaps() {
  orphan_tg_snaps=()
  for (( i=0; i < ${#mountpoints[@]}; i++ )); do
    orphan_start_count=${#orphan_tg_snaps[@]}
    tg_snap_count=0
    for tg_snap in ${tg_snaps[$i]}; do
      tg_snap_count=$(( tg_snap_count + 1 ))
      if [[ ! ${source_snaps[$tg_snap]} ]]; then
        orphan_tg_snaps+=("$tg_snap")
      fi
    done
    orphan_mp_count=$(( ${#orphan_tg_snaps[@]} - orphan_start_count ))
    # sanity checking
    tmp=$(( tg_snap_count  > 1 && tg_snap_count == orphan_mp_count ))
    if (( tmp )) ; then
      die "something went wrong checking orphans on $tg: for mountpoint ${mountpoints[$i]}, $orphan_mp_count"
    fi
  done
}

if [[ $source ]]; then
  for snap in $(ssh root@$source "shopt -s nullglob; ${snap_list_cmds[*]}"); do
    source_snaps[$snap]=t
  done
  get-orphan-tg-snaps
  tmp=$(( ${#orphan_tg_snaps[*]} >= 1 ))
  if (( tmp )); then
    d btrfs sub del ${orphan_tg_snaps[*]}
  fi
else # we have targets
  for tg in ${targets[@]}; do
    tmp_str=$(ssh root@$tg "shopt -s nullglob; ${snap_list_cmds[*]}")
    mapfile -t tg_snaps <<<"$tmp_str"
    get-orphan-tg-snaps
    tmp=$(( ${#orphan_tg_snaps[*]} >= 1 ))
    if (( tmp )); then
      d ssh root@$tg "btrfs sub del ${orphan_tg_snaps[*]}"
    fi
  done
fi

# todo: umount first to ensure we don't have any errors
# todo: do some kill fuser stuff to make umount more reliable


if $conf_only; then
  mexit 0
fi



if $dry_run; then
  m btrbk -c /etc/btrbk$conf_suf.conf -v -n $cmd_arg
  mexit 0
fi
# -q and just using the syslog option seemed nice,
# but it doesn't show when a send has a parent and when it doesn't.
logq btrbk -c /etc/btrbk$conf_suf.conf $preserve_arg $verbose_arg $progress_arg $cmd_arg

if $early; then
  exit 0
fi

# todo: tp not valid anymore.
# if we have it, sync to systems which don't
if mountpoint $rsync_mountpoint >/dev/null; then
  for tg in ${targets[@]}; do
    case $tg in
      tp)
        dirs=(/p/c/machine_specific/tp)
        for x in /p/c/machine_specific/*.hosts; do
          if grep -qxF $tg $x; then
            dirs+=(${x%.hosts})
          fi
        done
        m rsync -aSAXPH --specials --devices --delete --relative ${dirs[@]} root@$tg:/
        ;;
    esac
  done
fi

subvols=()
for mp in "${mountpoints[@]}"; do
  subvols+=("${mp##*/}")
done
if [[ $source ]]; then
  d mount-latest-subvol "${subvols[@]}"
else
  for tg in ${targets[@]}; do
    d /a/exe/mount-latest-remote "$tg" "${subvols[@]}" || ret=$?
  done
fi

# todo, we get hostnames earlier, reuse that.
if [[ $ret == 0 ]]; then
  for tg in ${targets[@]}; do
    h=$(ssh $tg hostname)
    if [[ $h == kd && $HOSTNAME == x3 && $HOSTNAME == "$MAIL_HOST" ]]; then
      d ssh root@$tg 'btrbk-spread-wrap &>/dev/null </dev/null &'
    fi
    rsync --mkpath -a -f"- */" -f"+ *" /var/log/btrbk/ root@$tg:/var/log/btrbk/$tg
    cmd=/usr/local/bin/mail-backup-clean
    ssh root@$tg "if test -x $cmd; then $cmd; fi"
  done
  if [[ $source ]]; then
    rsync --mkpath -a -f"- */" -f"+ *" $source:/var/log/btrbk/ /var/log/btrbk/$source
  fi
fi

mexit $ret

# todo: move variable data we don't care about backing up
# to /nocow and symlink it.


# background on btrbk timezones. with short/long, timestamps use local time.
# for long, if your local time moves backwards, by moving timezones or
# for an hour when daylight savings changes it, you will temporarily get
# a more aggressive retention policy for the overlapping period, and
# vice versa for the opposite timezone move. The alternative is using
# long-iso, which puts timezone info into the timestamp, which means
# that instead of shifting time, you shift the start of day/week/month
# which is used for retention to your new local time, which means for
# example, if you moved forward by 8 hours, the daily/weekly/monthly
# retention will be 8 hours more aggressive since midnight is at a new
# time, unless you fake the timzeone using the TZ env variable.
# However, in the short term, there will be no inconsistencies.
# I don't see any problem with shifting when the day starts for
# retention, so I'm using long-iso.

# note to create a long-iso timestamp: date +%Y%m%dT%H%M%S%z