Files
soulseek-rs-modded/docs/daemon-protocol.md
T
Michel de Graaf d11326f149 fix(daemon): ship the OpenRPC document inside the crate
docs/openrpc.json lived at the repository root, outside the soulseek-rs
package, so the include_str! in proto compiled from a checkout but not
from the published tarball - cargo publish failed verification and the
v13.0.0 release stopped after publishing the lib. The document moves
into the crate's own docs/, where packaging can see it; the generator
test and every link follow.
2026-07-29 17:34:29 +02:00

7.6 KiB

The daemon control protocol

soulseek-rs daemon stays logged in and answers JSON-RPC over a socket, so several clients can share one Soulseek session. The machine-readable contract is openrpc.json, an OpenRPC document and the JSON-RPC counterpart to OpenAPI. Point a client generator at it, or at a running daemon via rpc.discover, and you have a client in your language.

This page covers what a schema cannot: framing, the handshake, and how pushed events behave.

Framing

Newline-delimited JSON-RPC 2.0. One JSON value per line, in both directions, UTF-8, no length prefix. A line that does not parse gets a -32700 reply; on an established connection it stays open, but the first line has to be a valid auth request, so a parse error there ends the connection.

Lines are capped at 1 MiB and a connection that has not authenticated within 30 seconds is closed. The daemon serves at most 64 control connections at once.

Transports and authentication

Transport Address Authentication
Unix socket <state-dir>/daemon.sock, mode 0600 The socket's permissions. token may be omitted.
TCP opt-in via daemon --bind ADDR The token in <state-dir>/daemon.token, mode 0600.

The local socket is the Docker and ssh-agent bargain: if the operating system let you open it, you are the user who started the daemon, and there is nothing further to prove. A local client, whether a script or an agent, works with no configuration at all.

TCP always requires the token, and a token that is offered is always checked, on either transport, so a wrong one never passes. Failed attempts sleep one second. The token is 256 bits from the OS, generated on first start and reused across restarts so existing clients keep working; soulseek-rs daemon token prints it.

There is no transport encryption. Binding non-loopback TCP means putting SSH or a TLS-terminating proxy in front of it.

<state-dir> is $SOULSEEK_STATE_DIR, else the platform data directory (~/.local/share/soulseek-rs/state on Linux and macOS).

The handshake

auth must be the first request on a connection. Anything else is answered -32600 and the connection is dropped. Until it succeeds the connection is told nothing and receives no events.

--> {"jsonrpc":"2.0","id":1,"method":"auth","params":{"protocol":1}}
<-- {"jsonrpc":"2.0","id":1,"result":{"protocol":1,"daemon_version":"13.0.0","username":"you"}}

Compare protocol with your own and fail loudly on a mismatch rather than guessing; it is bumped whenever a change would break a client written against the old shape.

Requests

Parameters are by name: a JSON object, never an array. Each method's fields are published as individual OpenRPC content descriptors, so a generated client sends exactly the object the daemon deserializes. rpc.discover returns that same document from the running daemon.

An event is the exception: its params object is the payload, published as a single payload descriptor because several payloads are tagged unions with no fixed field set.

Requests may be pipelined and are answered out of order, so match replies by id. A request without an id is a notification and is answered with silence, including when it fails.

Errors

{"jsonrpc":"2.0","id":4,"error":{"code":-32000,"message":"no results","data":{"exit":4}}}

An error always carries an id, explicitly null when the request was too malformed to have one. Do not read a missing id as "this is a notification" and drop it.

Standard codes (-32700 parse, -32600 invalid request, -32601 unknown method, -32602 bad parameters) mean what they always mean. Application failures use -32000 and carry data.exit: the CLI exit status the same operation would have produced locally, so a wrapper can branch on it exactly as it branches on the exit code.

exit Meaning
2 Usage or configuration
3 Could not reach the server, or the login was rejected
4 Succeeded but produced nothing
5 Timed out
6 A transfer started but did not complete
7 The server session ended mid-operation

Events

The daemon pushes notifications. They carry no id, so never answer them:

{"jsonrpc":"2.0","method":"event.room","params":{"event":"message","room":"lobby","username":"bob","message":"hi"}}

event.room, event.message, event.upload, event.download_status, event.browse, event.session_loss. Payload schemas are in openrpc.json, where they appear as entries marked x-notification. OpenRPC has no first-class notion of a server-pushed message, so that flag is how they are distinguished from methods you may call.

message.history and event.message describe the same conversation in different shapes, deliberately: a live event is the message as the server delivered it (id, timestamp, username, message), while history is a chat log and carries a direction (peer, outgoing, text, at) because it includes what this account sent.

Events are the only way to observe these things, and they are not replayable. The underlying library hands over its event buffer and empties it, so the daemon drains once and copies to every attached connection. A client that is not connected when something happens does not learn about it later, with one exception: private messages are also kept, and message.history returns the conversation including what arrived while nobody was attached.

Every connection gets its own queue, bounded at 1024 events. A client that stops reading its socket is disconnected once that queue fills. The daemon will not let one stalled consumer cost the others their events. Read continuously; do not treat the socket as request/response only.

A download.start you issued reports progress through event.download_status matched on username + filename.

Behaviour worth knowing

Files land on the daemon's filesystem. The directory is the daemon's configuration and a client cannot name one, which would be an arbitrary write on the daemon's host. daemon.status reports the directory in use. There is no method to fetch the bytes over the control socket.

Configuration is per machine, shares are per session. A client's config file is its own; changing it does not reach a daemon. The exception is shares.set, which the CLI's shares add/remove call so a folder is served without waiting for a restart.

Several clients, no locking. Last write wins. Two clients cancelling the same transfer is two hands on one keyboard, not an error.

The daemon owns durable state. Transfers, joined rooms, and the conversation survive a restart: unfinished transfers are re-queued and rooms rejoined on startup.

Searching is two steps. search.start puts the query on the wire and returns immediately; peers answer over their own connections for as long as you care to wait. Decide the window yourself, then read search.results. This is deliberate: a client's search window must not be time the daemon spends not answering everybody else.

A session by hand

$ nc -U ~/.local/share/soulseek-rs/state/daemon.sock
{"jsonrpc":"2.0","id":1,"method":"auth","params":{"protocol":1}}
{"jsonrpc":"2.0","id":1,"result":{"protocol":1,"daemon_version":"13.0.0","username":"you"}}
{"jsonrpc":"2.0","id":2,"method":"search.start","params":{"query":"aphex twin"}}
{"jsonrpc":"2.0","id":2,"result":{"ok":true}}
{"jsonrpc":"2.0","id":3,"method":"search.results","params":{"query":"aphex twin"}}
{"jsonrpc":"2.0","id":3,"result":{"results":[…]}}

That the whole protocol is reachable from nc and a text editor is the reason it is shaped this way.