[BUG] RPC readiness probe hardcodes localhost, so a non-loopback rpc-bind-address leaves container init incomplete and every stop is a SIGKILL

Start with svc-transmission/run and svc-transmission/finish, focusing on how each script reads rpc-port and targets the RPC service. Reproduce the non-loopback setup from the issue, then verify wildcard and specific bind addresses both complete init and that docker stop exits cleanly without SIGKILL or premature finish-script termination.

Is there an existing issue for this?

• I have searched the existing issues

I checked #80, #279, #261, #285, #306 and #324. #80 is the closest — it identified that transmission-remote --exit failed when USER/PASS were set, and was fixed by passing credentials — but it never identified the bind-address dependency described below. #324 is a different bug (the [::] default failing to bind on some Synology/QNAP kernels), though the workaround being given there — hand-editing rpc-bind-address — is exactly how a user can walk into this one.

Current Behavior

svc-transmission/run gates service readiness on a hardcoded loopback probe:

PORT=$(jq '.["rpc-port"]' /config/settings.json)

s6-notifyoncheck -d -n 300 -w 1000 -c "nc -z localhost ${PORT:-9091}" \
    s6-setuidgid abc /usr/bin/transmission-daemon \
    -g /config -f

It reads rpc-port from settings.json but ignores rpc-bind-address, always probing localhost.

With the default rpc-bind-address ([::], or 0.0.0.0) this is invisible, because localhost resolves to an address the wildcard socket accepts. But if rpc-bind-address is set to a specific non-loopback address, nc -z localhost can never succeed, and the container never finishes starting:

1. s6-notifyoncheck never signals readiness
2. s6-svlisten1 -U -- /run/s6-rc/servicedirs/svc-transmission blocks forever waiting for "up"
3. rc.init top / s6-rc -v1 -u -t 0 -- change top never complete — container init stays permanently in progress
4. because init never finished, s6's shutdown path never delivers SIGTERM to transmission-daemon
5. every docker stop therefore ends in SIGKILL

The damage is that the daemon is hard-killed on every single stop/restart: resume data is not flushed and no "stopped" announce is sent to trackers. Nothing in the logs indicates a problem — the daemon itself is completely healthy and serving RPC the whole time.

There is a second, independent loopback hardcode in the same service's finish script, so even once readiness is fixed the graceful exit still cannot land off-localhost:

/usr/bin/transmission-remote 127.0.0.1:${PORT:-9091} -n "$USER":"$PASS" --exit

Related: there is no timeout-finish file in the servicedir, so s6 SIGKILLs the finish script at its 5000 ms default. On a large library the daemon legitimately needs longer than that to exit (measured ~8–9 s for ~1300 torrents on my host), so tail --pid=${pid} is killed before the daemon is actually gone.

Expected Behavior

The readiness probe and the shutdown request should target the address Transmission is actually bound to, derived from rpc-bind-address. Wildcard binds (0.0.0.0, ::, [::]) should keep probing loopback as they do today.

Container init should complete, and docker stop should result in a graceful shutdown (exit 0) regardless of which address RPC is bound to.

Steps To Reproduce

1. Create a user-defined network so the container has a predictable non-loopback address:
   docker network create --subnet 172.30.0.0/16 tmnet
2. Seed /config/settings.json with "rpc-bind-address": "172.30.0.5"
3. Start the container on that network with --ip 172.30.0.5 (command below)
4. Observe the daemon is healthy and listening on 172.30.0.5:9091, but [ls.io-init] done. is never printed and the init processes are still alive:

$ docker top tm-base          # columns abridged for readability
1201  root  {rc.init} /bin/sh -e /run/s6/basedir/scripts/rc.init top
1228  root  s6-rc -v1 -u -t 0 -- change top
1344  root  s6-svlisten1 -U -- /run/s6-rc/servicedirs/svc-transmission ... s6-svc -u -- /run/s6-rc/servicedirs/svc-transmission
1361  root  s6-notifyoncheck -d -n 300 -w 1000 -c nc -z localhost 9091 s6-setuidgid abc /usr/bin/transmission-daemon -g /config -f

$ docker exec tm-base netstat -tln | grep 9091
tcp  0  0  172.30.0.5:9091  0.0.0.0:*  LISTEN

5. Stop it and observe the SIGKILL:

$ S=$(date +%s); docker stop -t 30 tm-base; E=$(date +%s); echo "elapsed: $((E-S))s"
elapsed: 31s
$ docker inspect -f '{{.State.ExitCode}}' tm-base
137

The full 30 s grace period elapses with no shutdown log lines at all — the daemon never received a signal.

For contrast, the identical container with the stock [::] default completes init in ~2 s and stops cleanly with exit 0 in ~4 s.

Environment

• OS: originally observed on Ubuntu 24.04 LTS; the isolated reproduction above was run on OrbStack (macOS) with the same image and is reproducible on both
• How docker service was installed: distro's package manager (Ubuntu) / OrbStack (macOS)

CPU architecture

x86-64

Docker creation

docker network create --subnet 172.30.0.0/16 tmnet

mkdir -p ./cfg && cat > ./cfg/settings.json <<'EOF'
{
    "rpc-bind-address": "172.30.0.5",
    "rpc-port": 9091,
    "rpc-url": "/transmission/"
}
EOF

docker run -d --name tm-base \
  --network tmnet --ip 172.30.0.5 \
  -e PUID=1000 -e PGID=1000 \
  -e USER=tu -e PASS=tp \
  -v "$PWD/cfg:/config" \
  ghcr.io/linuxserver/transmission:4.1.3@sha256:6b78970a0cb00709d3817fcb47b7cf74da56a43384bbd50d03955ca87e38c5f0

Image: 4.1.3-r0-ls356, built 2026-07-28.

Container logs

[migrations] started
[migrations] no migrations found
───────────────────────────────────────

      ██╗     ███████╗██╗ ██████╗
      ██║     ██╔════╝██║██╔═══██╗
      ██║     ███████╗██║██║   ██║
      ██║     ╚════██║██║██║   ██║
      ███████╗███████║██║╚██████╔╝
      ╚══════╝╚══════╝╚═╝ ╚═════╝

   Brought to you by linuxserver.io
───────────────────────────────────────

To support LSIO projects visit:
https://www.linuxserver.io/donate/

───────────────────────────────────────
GID/UID
───────────────────────────────────────

User UID:    1000
User GID:    1000
───────────────────────────────────────
Linuxserver.io version: 4.1.3-r0-ls356
Build-date: 2026-07-28T12:17:56+00:00
───────────────────────────────────────
    
[custom-init] No custom files found, skipping...
[2026-08-01T19:04:58.423+0000] inf ip-cache.cc:385 Cached IPv4 source address 172.30.0.5 (ip-cache.cc:385)
...
[2026-08-01T19:04:58.423+0000] inf session.cc:424 Listening to incoming peer connections on 0.0.0.0:51413 (session.cc:424)
[2026-08-01T19:04:58.423+0000] inf session.cc:424 Listening to incoming peer connections on [::]:51413 (session.cc:424)
[2026-08-01T19:04:58.423+0000] inf port-forwarding.cc:221 State changed from 'Not forwarded' to 'Starting' (port-forwarding.cc:221)
[2026-08-01T19:04:58.423+0000] inf tr-udp.cc:207 Bound UDP IPv4 address 0.0.0.0:51413 (tr-udp.cc:207)
[2026-08-01T19:04:58.423+0000] inf tr-udp.cc:254 Bound UDP IPv6 address [::]:51413 (tr-udp.cc:254)
[2026-08-01T19:04:58.423+0000] WRN tr-lpd.cc:288 Couldn't initialize IPv6 LPD: No such device (19) (tr-lpd.cc:288)
[2026-08-01T19:04:58.468+0000] inf rpc-server.cc:878 Added '127.0.0.1' to host whitelist (rpc-server.cc:878)
[2026-08-01T19:04:58.468+0000] inf rpc-server.cc:878 Added '::1' to host whitelist (rpc-server.cc:878)
[2026-08-01T19:04:58.468+0000] inf rpc-server.cc:1013 Serving RPC and Web requests on 172.30.0.5:9091/transmission/ (rpc-server.cc:1013)
[2026-08-01T19:04:58.468+0000] inf rpc-server.cc:825 Listening for RPC and Web requests on '172.30.0.5:9091' (rpc-server.cc:825)

[ls.io-init] done. never appears. On docker stop the log simply ends — there is no
"shutting down" and no "Closing transmission session" line, because the daemon is never
signalled.