From 0804cb765dc5736e4eb5213d3f51546a76b0b48e Mon Sep 17 00:00:00 2001 From: William Grzybowski Date: Mon, 16 Nov 2026 21:53:49 -0300 Subject: [PATCH] truenas-appliance-runner: network: bridge, the appliance on a bridge with the runner Under qemu's user-mode network the appliance is 10.0.2.15 behind qemu's NAT, and only its ports are forwarded, to 8443, 8080 and 2222 on the runner's loopback. A TrueNAS Connect stack on the runner cannot work with that: registration publishes the NAS's own IPs, so its truenas.direct name resolves to 10.0.2.15, and a client then connects to it on 443 (truenas-connect/cui#121). network: bridge (APPLIANCE_NETWORK) puts the appliance on a bridge, tnapp-br, with the runner at 10.232.0.1. qemu's NIC is a tap on it with a fixed MAC, and dnsmasq gives that MAC 10.232.0.10 with a lease that never ends, so a revert keeps the address. dnsmasq also answers the appliance's DNS, forwarding to the runner's resolvers, which may be a loopback stub the appliance cannot reach. iptables accepts the bridge's traffic through Docker's FORWARD policy and NATs the appliance out. TN_HOST is then 10.232.0.10:443. Both modes now print TN_NETWORK and TN_RUNNER_IP: 10.0.2.2 under qemu's NAT, which is the runner's loopback, and 10.232.0.1 on the bridge. Bridge also prints TN_APPLIANCE_IP. Nothing is forwarded to 127.0.0.1 with bridge, so consumers that hard-code it keep user, the default, which is unchanged. stop now also takes the bridge network down, each part only if it is there, from what start noted before it changed anything, and start refuses while tnapp-br exists. The NIC stays a -nic in both modes, so it keeps its PCI slot and its name in the guest. --- .../truenas-appliance-runner/action.yml | 36 ++++ .../truenas-appliance-runner/appliance.sh | 190 ++++++++++++++++-- README.md | 45 ++++- 3 files changed, 252 insertions(+), 19 deletions(-) diff --git a/.github/actions/truenas-appliance-runner/action.yml b/.github/actions/truenas-appliance-runner/action.yml index e871e46..6221b6b 100644 --- a/.github/actions/truenas-appliance-runner/action.yml +++ b/.github/actions/truenas-appliance-runner/action.yml @@ -30,6 +30,13 @@ description: >- # action sets them; truenas-appliance (appliance.sh) is on PATH for # snapshot, revert, probe, call and console. # +# `network: bridge` puts the appliance on a bridge with the runner instead +# of behind qemu's NAT: TN_HOST is its own address, 10.232.0.10, on port +# 443, which the runner, its containers and a browser reach directly, and +# the appliance reaches the runner at TN_RUNNER_IP, 10.232.0.1. For what +# needs the appliance's real address and ports, such as a TrueNAS Connect +# stack on the runner. The default, `user`, is unchanged. +# # `ha: true` boots an HA pair instead (ha.sh): two appliances, from a pool # that attaches the appliance disk twice, # @@ -76,6 +83,15 @@ inputs: apply interface changes on without cutting itself off. Their MACs, 52:54:00:e2:e0:01 and up, are in TN_SPARE_NICS. Not for an HA pair. default: '0' + network: + description: >- + `user`: behind qemu's NAT, its ports forwarded to the runner's + loopback (TN_HOST 127.0.0.1:8443), and reaching the runner's loopback + at 10.0.2.2. `bridge`: on a bridge with the runner, at 10.232.0.10 on + its real ports (TN_HOST 10.232.0.10:443), reached directly from the + runner, its containers and a browser, and reaching what the runner + serves on 10.232.0.1. Needs passwordless sudo. Not for an HA pair. + default: user ssh: description: '`true` enables ssh for the admin, on TN_SSH.' default: 'false' @@ -102,6 +118,18 @@ outputs: ssh: description: host:port of its ssh, when asked for. value: ${{ steps.start.outputs.ssh }} + network: + description: '`user` or `bridge`, as asked; empty for an HA pair.' + value: ${{ steps.start.outputs.network }} + runner-ip: + description: >- + Where the appliance reaches the runner: 10.0.2.2 with `user`, which + qemu maps to the runner's loopback, or 10.232.0.1 with `bridge`. + Empty for an HA pair. + value: ${{ steps.start.outputs.runner-ip }} + appliance-ip: + description: The appliance's own address with `bridge`, 10.232.0.10; empty otherwise. + value: ${{ steps.start.outputs.appliance-ip }} boot-seconds: description: From qemu starting to the API taking a login. value: ${{ steps.start.outputs.boot-seconds }} @@ -122,6 +150,7 @@ runs: APPLIANCE_DATA_DISK_GB: ${{ inputs.data-disk-gb }} APPLIANCE_EXTRA_DISK_GB: ${{ inputs.extra-disk-gb }} APPLIANCE_SPARE_NICS: ${{ inputs.spare-nics }} + APPLIANCE_NETWORK: ${{ inputs.network }} APPLIANCE_SSH: ${{ inputs.ssh }} APPLIANCE_DISK_INPUT: ${{ inputs.disk }} EXPORT_ENV: ${{ inputs.export-env }} @@ -140,6 +169,10 @@ runs: echo "::error::spare-nics is for one appliance; ha.sh wires an HA pair's NICs itself" exit 1 fi + if [ "$HA" = true ] && [ "${APPLIANCE_NETWORK:-user}" != user ]; then + echo "::error::network is for one appliance; ha.sh puts an HA pair on bridges of its own" + exit 1 + fi export APPLIANCE_DIR="$RUNNER_TEMP/truenas-appliance" # The password is made here, so it is masked before anything can # print it. @@ -169,6 +202,9 @@ runs: echo "username=$(get TN_USERNAME)" echo "password=$(get TN_PASSWORD)" echo "ssh=$(get TN_SSH)" + echo "network=$(get TN_NETWORK)" + echo "runner-ip=$(get TN_RUNNER_IP)" + echo "appliance-ip=$(get TN_APPLIANCE_IP)" echo "boot-seconds=$(get APPLIANCE_BOOT_SECONDS)" echo "snapshot-seconds=$(get APPLIANCE_SNAPSHOT_SECONDS)" } >> "$GITHUB_OUTPUT" diff --git a/.github/actions/truenas-appliance-runner/appliance.sh b/.github/actions/truenas-appliance-runner/appliance.sh index d00f244..4cc7d74 100755 --- a/.github/actions/truenas-appliance-runner/appliance.sh +++ b/.github/actions/truenas-appliance-runner/appliance.sh @@ -25,7 +25,7 @@ # result as JSON on stdout # appliance.sh snapshots the names there are # appliance.sh console [lines] the end of its serial console -# appliance.sh stop +# appliance.sh stop quit it, and take its bridge network down # # Settings, environment variables, all optional: # @@ -42,8 +42,16 @@ # alone on a qemu hub of its own, so no DHCP, no # route and no peer: for a suite to reconfigure # without cutting itself off +# APPLIANCE_NETWORK user: qemu's user-mode network, the guest behind +# its NAT and its ports forwarded to this machine's +# loopback, on the ports below. bridge: the guest +# on a bridge with this machine (bridge_up), on its +# own address and its real ports, reachable from +# here, from containers and from a browser; needs +# passwordless sudo # APPLIANCE_HTTP_PORT 8080 APPLIANCE_HTTPS_PORT 8443 # APPLIANCE_SSH_PORT 2222 APPLIANCE_SSH false +# (the ports forwarded with user; bridge has none) # APPLIANCE_BASELINE fresh-install: the snapshot start takes; empty # takes none # APPLIANCE_PASSWORD the claim's admin password (default: generated) @@ -55,8 +63,10 @@ # TN_USERNAME, TN_PASSWORD, TN_DOMAIN, TN_BASELINE (the suite's contract, # as the lab's claim action gives it, less its TN_GUEST_ISO), TN_SSH when # APPLIANCE_SSH is true, TN_SPARE_NICS (the spare NICs' MAC addresses, -# space-separated) when there are any, and what it measured: -# APPLIANCE_BOOT_SECONDS, APPLIANCE_SNAPSHOT_SECONDS. +# space-separated) when there are any, TN_NETWORK (user or bridge), +# TN_RUNNER_IP (where the guest reaches this machine), TN_APPLIANCE_IP +# with bridge, and what it measured: APPLIANCE_BOOT_SECONDS, +# APPLIANCE_SNAPSHOT_SECONDS. set -euo pipefail # shellcheck source=lib.sh @@ -69,11 +79,47 @@ QMP="$DIR/qmp.sock" QGA="$DIR/qga.sock" STATE="$DIR/state.env" +# The bridge network (APPLIANCE_NETWORK=bridge): this machine at .1 and the +# appliance at .10, by its MAC. Not ha.sh's 10.231.0.0/24, Docker's +# 172.17.0.0/16 or compose's networks. The MAC is fixed so that the address +# is the same after a revert; the spare NICs' are e0:01 and up. +BR=tnapp-br +TAP=tnapp-tap +BR_NET=10.232.0 +BR_RUNNER_IP=$BR_NET.1 +BR_APPLIANCE_IP=$BR_NET.10 +BR_MAC=52:54:00:e2:e0:00 +# What bridge_up changed, written before it changes anything: bridge_down +# reads it, even after a start that failed halfway. +NETSTATE="$DIR/network.env" +# The rules bridge_up adds, as "table chain rule...". Docker loads +# br_netfilter and drops FORWARD by default, which would drop the +# appliance's traffic to containers and beyond this machine, and theirs to +# it; the NAT takes it beyond this machine, as qemu's user-mode NAT does. +BR_RULES=( + "filter FORWARD -i $BR -j ACCEPT" + "filter FORWARD -o $BR -j ACCEPT" + "nat POSTROUTING -s $BR_NET.0/24 ! -o $BR -j MASQUERADE" +) + hmp() { "${TNCTL[@]}" hmp "$QMP" "$*"; } # resume: a failed loadvm leaves the VM paused, and later loadvms leave it # so, however they go; cont is a no-op on one that runs. resume() { "${TNCTL[@]}" qmp "$QMP" cont >/dev/null; } -url() { echo "ws://127.0.0.1:${APPLIANCE_HTTP_PORT:-8080}"; } +# addr : host:port where this machine reaches that port of the +# appliance: its own address on the bridge, or qemu's forward of it. +addr() { + if [ "${APPLIANCE_NETWORK:-user}" = bridge ]; then + echo "$BR_APPLIANCE_IP:$1" + return + fi + case "$1" in + 80) echo "127.0.0.1:${APPLIANCE_HTTP_PORT:-8080}" ;; + 443) echo "127.0.0.1:${APPLIANCE_HTTPS_PORT:-8443}" ;; + 22) echo "127.0.0.1:${APPLIANCE_SSH_PORT:-2222}" ;; + esac +} +url() { echo "ws://$(addr 80)"; } running() { [ -f "$DIR/qemu.pid" ] && kill -0 "$(cat "$DIR/qemu.pid")" 2>/dev/null; } # shellcheck source=/dev/null state() { [ -r "$STATE" ] || die "no appliance started here ($STATE)"; . "$STATE"; } @@ -104,6 +150,70 @@ print(found[0]) ' } +# bridge_up: the bridge network. A bridge with this machine on it, a tap +# for qemu (this user's, so qemu runs without sudo), BR_RULES, and dnsmasq +# giving the appliance's MAC its address and answering its DNS. +bridge_up() { + say "network: bridge $BR, this machine at $BR_RUNNER_IP, the appliance at $BR_APPLIANCE_IP" + local forward rule r + forward="$(cat /proc/sys/net/ipv4/ip_forward)" + echo "BR_FORWARD_WAS=$forward" > "$NETSTATE" + sudo -n ip link add "$BR" type bridge + sudo -n ip addr add "$BR_RUNNER_IP/24" dev "$BR" + sudo -n ip link set "$BR" up + sudo -n ip tuntap add dev "$TAP" mode tap user "$(id -u)" + sudo -n ip link set "$TAP" master "$BR" up + for rule in "${BR_RULES[@]}"; do + read -r -a r <<< "$rule" + sudo -n iptables -t "${r[0]}" -I "${r[1]}" "${r[@]:2}" + done + # Docker turns it on; a runner without Docker may not have. + if [ "$forward" != 1 ]; then + sudo -n sysctl -q -w net.ipv4.ip_forward=1 + fi + # DNS as well as DHCP, unlike ha.sh's: the appliance needs a resolver + # for NTP and for what middleware looks up as it starts, and this + # machine's may be a stub on its loopback (systemd-resolved's + # 127.0.0.53) that the appliance cannot reach. dnsmasq forwards to + # whatever /etc/resolv.conf names, and its offer names it as the router + # and the DNS server; --no-hosts keeps this machine's /etc/hosts out of + # its answers. Static: it answers this MAC alone. The lease never ends, + # so a revert, which restores the guest's lease with its memory and + # moves its clock on, never finds it expired. + sudo -n dnsmasq --interface="$BR" --bind-interfaces --except-interface=lo --no-hosts \ + --dhcp-range="$BR_APPLIANCE_IP,static" --dhcp-host="$BR_MAC,$BR_APPLIANCE_IP,infinite" \ + --dhcp-leasefile="$DIR/dnsmasq.leases" --pid-file="$DIR/dnsmasq.pid" \ + --log-dhcp --log-facility="$DIR/dnsmasq.log" +} + +# bridge_down: undo bridge_up, each part only if it is there, so that a +# self-hosted runner's next start finds nothing of this one. Nothing to do +# for a user-mode appliance. +bridge_down() { + if [ ! -f "$NETSTATE" ] && ! ip link show "$BR" >/dev/null 2>&1; then + return 0 + fi + say "network: taking $BR down" + local BR_FORWARD_WAS=1 rule r + # shellcheck source=/dev/null + if [ -r "$NETSTATE" ]; then . "$NETSTATE"; fi + if [ -s "$DIR/dnsmasq.pid" ]; then + sudo -n kill "$(cat "$DIR/dnsmasq.pid")" 2>/dev/null || true + fi + for rule in "${BR_RULES[@]}"; do + read -r -a r <<< "$rule" + while sudo -n iptables -t "${r[0]}" -C "${r[1]}" "${r[@]:2}" 2>/dev/null; do + sudo -n iptables -t "${r[0]}" -D "${r[1]}" "${r[@]:2}" + done + done + if ip link show "$TAP" >/dev/null 2>&1; then sudo -n ip link del "$TAP"; fi + if ip link show "$BR" >/dev/null 2>&1; then sudo -n ip link del "$BR"; fi + if [ "$BR_FORWARD_WAS" != 1 ]; then + sudo -n sysctl -q -w net.ipv4.ip_forward="$BR_FORWARD_WAS" + fi + rm -f "$NETSTATE" +} + start() { local disk="${APPLIANCE_DISK:-}" local memory="${APPLIANCE_MEMORY_MB:-6144}" vcpus="${APPLIANCE_VCPUS:-4}" @@ -117,6 +227,18 @@ start() { local name="appliance-${GITHUB_RUN_ID:-local}-${GITHUB_RUN_ATTEMPT:-1}-${GITHUB_JOB:-job}" running && die "an appliance is running already ($DIR)" + APPLIANCE_NETWORK="${APPLIANCE_NETWORK:-user}" + case "$APPLIANCE_NETWORK" in + user) ;; + bridge) + if ip link show "$BR" >/dev/null 2>&1; then + die "$BR exists already: an earlier appliance's network is still up, and truenas-appliance stop takes it down" + fi + sudo -n true 2>/dev/null || die "APPLIANCE_NETWORK=bridge needs passwordless sudo, for the bridge, the tap, iptables and dnsmasq" + needs dnsmasq dnsmasq-base iptables iptables + ;; + *) die "APPLIANCE_NETWORK: \"$APPLIANCE_NETWORK\" is not user or bridge" ;; + esac local gb for gb in "${extragb[@]}"; do [[ "$gb" =~ ^[1-9][0-9]*$ ]] || die "APPLIANCE_EXTRA_DISK_GB: \"$gb\" is not a whole number of GB" @@ -185,6 +307,15 @@ start() { elif [ "$nspare" -gt 1 ]; then nicssaid=", $nspare spare NICs" fi + # -nic either way, so the NIC is made the same way, keeps its PCI slot, + # and with it its name in the guest. + local nic + if [ "$APPLIANCE_NETWORK" = bridge ]; then + bridge_up + nic="tap,ifname=$TAP,script=no,downscript=no,model=virtio-net-pci,mac=$BR_MAC" + else + nic="user,model=virtio-net-pci$(hostfwd 80 "$http")$(hostfwd 443 "$https")$(hostfwd 22 "$sshp")" + fi local booted t0 say "booting $disk as $name: ${memory} MB, $vcpus vCPUs, $ndata data disks of ${datagb}G$extrasaid$nicssaid" @@ -196,7 +327,7 @@ start() { -drive "file=$DIR/os.qcow2,if=none,id=os,format=qcow2,discard=unmap" \ -device "virtio-blk-pci,drive=os,serial=os,bootindex=0" \ "${drives[@]}" \ - -nic "user,model=virtio-net-pci$(hostfwd 80 "$http")$(hostfwd 443 "$https")$(hostfwd 22 "$sshp")" \ + -nic "$nic" \ "${nics[@]}" \ -device virtio-rng-pci \ -device virtio-serial-pci,id=vser0 \ @@ -215,16 +346,21 @@ start() { --since "$t0" --timeout "${APPLIANCE_BOOT_TIMEOUT:-900}"); then say "--- its console, last 60 lines" console 60 >&2 || true + if [ "$APPLIANCE_NETWORK" = bridge ]; then + say "--- dnsmasq" + cat "$DIR/dnsmasq.log" >&2 2>/dev/null || true + fi die "the appliance's API never came up" fi say "API up ${booted}s after qemu started" "${TNCTL[@]}" rotate "$(url)" "$APPLIANCE_USER" "$APPLIANCE_TEMPLATE_PASSWORD" "$password" if [ "${APPLIANCE_SSH:-false}" = true ]; then "${TNCTL[@]}" enable-ssh "$(url)" "$APPLIANCE_USER" "$password" - say "ssh on 127.0.0.1:$sshp" + say "ssh on $(addr 22)" fi { echo "APPLIANCE_PASSWORD_NOW=$(printf %q "$password")" + echo "APPLIANCE_NETWORK=$APPLIANCE_NETWORK" echo "APPLIANCE_HTTP_PORT=$http" echo "APPLIANCE_SSH_PORT=$sshp" } > "$STATE" @@ -238,21 +374,31 @@ start() { say "snapshot $baseline taken in ${snapped}s" fi + # TN_RUNNER_IP: where the guest reaches this machine. Under qemu's NAT + # that is 10.0.2.2, which qemu maps to this machine's loopback; on the + # bridge, what listens on the bridge's address or on every address. + local runner_ip=10.0.2.2 + [ "$APPLIANCE_NETWORK" = bridge ] && runner_ip=$BR_RUNNER_IP cat <&2 - local port="${APPLIANCE_SSH_PORT:-2222}" + local at + at="$(addr 22)" local -a ssh_ - ssh_=(ssh -i "$key" -p "$port" -o BatchMode=yes -o ConnectTimeout=5 + ssh_=(ssh -i "$key" -p "${at##*:}" -o BatchMode=yes -o ConnectTimeout=5 -o StrictHostKeyChecking=no -o UserKnownHostsFile=/dev/null -o LogLevel=ERROR - "$APPLIANCE_USER@127.0.0.1") + "$APPLIANCE_USER@${at%:*}") local i for i in $(seq 1 30); do "${ssh_[@]}" true 2>/dev/null && break - [ "$i" = 30 ] && die "no ssh into the appliance on 127.0.0.1:$port" + [ "$i" = 30 ] && die "no ssh into the appliance on $at" sleep 2 done "${ssh_[@]}" 'bash -s' <<'REPORT' @@ -394,6 +541,21 @@ print(f"{ms / 1000 - time.time():.1f}")')" echo "APPLIANCE_REVERT_SECONDS=$took" } +# stop: quit qemu, then take down the bridge network, if there is one: the +# tap is qemu's until it exits. +stop() { + if running; then + local pid i + pid="$(cat "$DIR/qemu.pid")" + "${TNCTL[@]}" qmp "$QMP" quit >/dev/null || true + for i in $(seq 1 50); do + kill -0 "$pid" 2>/dev/null || break + sleep 0.2 + done + fi + bridge_down +} + case "${1:-}" in start) start ;; call) shift; [ $# -gt 0 ] || die "call [params...]"; state @@ -404,6 +566,6 @@ revert) shift; revert "$@" ;; snapshots) running || die "no appliance running"; hmp info snapshots ;; console) console "${2:-100}" ;; boot-report) boot_report ;; -stop) if running; then "${TNCTL[@]}" qmp "$QMP" quit >/dev/null || true; fi ;; +stop) stop ;; *) die "usage: appliance.sh start | snapshot | revert | probe [baseline] | call [params...] | boot-report | snapshots | console [lines] | stop" ;; esac diff --git a/README.md b/README.md index d225467..ad03081 100644 --- a/README.md +++ b/README.md @@ -564,16 +564,51 @@ jobs: | `data-disk-count`, `data-disk-gb` | `10`, `10` | Sparse qcow2 files | | `extra-disk-gb` | none | More sparse disks after the data disks, one per size in GB: `'32'`, `'12 12 15'`. Not for an HA pair | | `spare-nics` | `'0'` | NICs after the one the API is reached on, each alone on a qemu hub: no DHCP, no route. For applying interface changes without losing the appliance. MACs `52:54:00:e2:e0:01` and up, in `TN_SPARE_NICS`. Not for an HA pair | +| `network` | `user` | `user`: behind qemu's NAT, its ports forwarded to the runner's loopback. `bridge`: on a bridge with the runner, on its own address and real ports; see below. Not for an HA pair | | `ssh` | `'false'` | `true` enables ssh for the admin, on `TN_SSH` | | `disk` | the attached one | Found by its serial, `virtio-truenas_appliance`, where the box keeps it; TrueNAS replaces it, so otherwise the one disk besides the root disk that holds a ZFS `boot-pool` | | `export-env` | `'true'` | Write `TN_*` to the job environment | Outputs: `host`, `host-http`, `username`, `password` (masked), `ssh`, -`boot-seconds`, `snapshot-seconds`. `truenas-appliance` is put on `PATH`: -`snapshot `, `revert `, `call [params...]`, -`snapshots`, `console [lines]`, `probe`, and `boot-report`: where the boot's -time went, from systemd and middleware's own log, over ssh (it leaves ssh on -with a key and sudo for the admin, so revert afterwards if that matters). +`network`, `runner-ip`, `appliance-ip`, `boot-seconds`, `snapshot-seconds`. +`truenas-appliance` is put on `PATH`: `snapshot `, `revert `, +`call [params...]`, `snapshots`, `console [lines]`, `probe`, +`boot-report`: where the boot's time went, from systemd and middleware's own +log, over ssh (it leaves ssh on with a key and sudo for the admin, so revert +afterwards if that matters), and `stop`. + +| Variable | `network: user` | `network: bridge` | +|---|---|---| +| `TN_HOST` | `127.0.0.1:8443` | `10.232.0.10:443` | +| `TN_HOST_HTTP` | `127.0.0.1:8080` | `10.232.0.10:80` | +| `TN_SSH` (with `ssh: true`) | `127.0.0.1:2222` | `10.232.0.10:22` | +| `TN_NETWORK` | `user` | `bridge` | +| `TN_RUNNER_IP`: where the appliance reaches the runner | `10.0.2.2` | `10.232.0.1` | +| `TN_APPLIANCE_IP` | not set | `10.232.0.10` | + +**`network: bridge`** is for what needs the appliance's own address and real +ports, such as a TrueNAS Connect stack on the runner: registration publishes +the NAS's own IPs, and a client then connects to them on 443. The runner is +`10.232.0.1` on a bridge, `tnapp-br`; qemu's NIC is a tap on it with the +fixed MAC `52:54:00:e2:e0:00`, and dnsmasq gives that MAC `10.232.0.10` for +good and forwards the appliance's DNS to the runner's resolvers. iptables +accepts the bridge's traffic through Docker's `FORWARD` policy and NATs the +appliance out, as qemu's NAT does under `user`. The appliance is reached on +`10.232.0.10` from the runner, from containers on Docker's networks and from +a browser on the runner, and keeps its address across `revert`. + +- **Nothing is forwarded to `127.0.0.1`** with `bridge`. A consumer that + hard-codes it, or reaches the appliance as `host.docker.internal` + (webui's nginx container), keeps `user`. +- **What the appliance reaches on the runner differs.** Under `user`, + `10.0.2.2` is the runner's loopback, so a port published on `127.0.0.1` + is reachable. On the bridge, the appliance reaches what listens on + `10.232.0.1` or on every address; publish a container's port there. +- **It needs passwordless sudo**, for the bridge, the tap, iptables and + dnsmasq (GARM's runners have it); qemu itself still runs as the job's user. +- **`truenas-appliance stop`** quits qemu and removes the bridge, the tap, + dnsmasq and the iptables rules. `start` refuses while `tnapp-br` exists. + An ephemeral runner needs neither; a self-hosted one re-running jobs does. `revert` restores the appliance's memory, and with it the clock it had when the snapshot was taken. It then sets the clock to the runner's through the