harness: add /screen page and /screen/ws WebSocket VNC relay

Reads /etc/hyperhive/gui.json at startup to get the VNC port written
by the weston-vnc ExecStart script (issue #50). Adds:
- gui_vnc_port: Option<u16> on AppState
- gui_enabled: bool on StateSnapshot (for issue #52 screen link)
- GET /screen: serves a minimal RFB-over-WebSocket viewer (screen.html)
- GET /screen/ws: upgrades to WebSocket and byte-pumps to 127.0.0.1:<vnc_port>

The relay is a pure two-task byte pump (WS→TCP and TCP→WS), transparent
to any RFB variant including VeNCrypt. Returns 404 when gui is not
enabled.

screen.html is a self-contained RFB client: handshake, FramebufferUpdate
(Raw encoding), pointer and keyboard forwarding — enough to display the
desktop and interact with it. noVNC assets (issue #52) replace this.

Closes #51
This commit is contained in:
iris 2026-05-20 14:24:05 +02:00 committed by Mara
parent 29df223650
commit 2027e94432
5 changed files with 651 additions and 1 deletions

View file

@ -22,6 +22,7 @@ use axum::{
routing::{get, post},
};
use serde::{Deserialize, Serialize};
use tokio::io::{AsyncReadExt, AsyncWriteExt};
use tokio_stream::{Stream, StreamExt, wrappers::BroadcastStream};
use crate::client;
@ -56,6 +57,9 @@ struct AppState {
files: TurnFiles,
/// Prevents `/api/compact` from racing with an in-flight normal turn.
turn_lock: TurnLock,
/// VNC port read from `/etc/hyperhive/gui.json` at startup.
/// `None` when the file is absent (gui not enabled for this agent).
gui_vnc_port: Option<u16>,
}
impl AppState {
@ -80,6 +84,7 @@ pub async fn serve(
files: TurnFiles,
turn_lock: TurnLock,
) -> Result<()> {
let gui_vnc_port = read_gui_json();
let state = AppState {
label,
login,
@ -88,6 +93,7 @@ pub async fn serve(
socket,
files,
turn_lock,
gui_vnc_port,
};
let app = Router::new()
.route("/", get(serve_index))
@ -110,6 +116,8 @@ pub async fn serve(
.route("/stats", get(serve_stats))
.route("/static/stats.js", get(serve_stats_js))
.route("/api/stats", get(api_stats))
.route("/screen", get(serve_screen))
.route("/screen/ws", get(screen_ws))
.with_state(state);
let addr = SocketAddr::from(([0, 0, 0, 0], port));
let listener = bind_with_retry(addr, "web UI").await?;
@ -215,6 +223,91 @@ async fn serve_stats_js() -> impl IntoResponse {
)
}
async fn serve_screen() -> impl IntoResponse {
(
[("content-type", "text/html; charset=utf-8")],
include_str!("../assets/screen.html"),
)
}
/// Read `/etc/hyperhive/gui.json` and extract the `vnc_port` field.
/// Returns `None` if the file is absent or unparseable — GUI not enabled.
fn read_gui_json() -> Option<u16> {
let text = std::fs::read_to_string("/etc/hyperhive/gui.json").ok()?;
let val: serde_json::Value = serde_json::from_str(&text).ok()?;
val["vnc_port"].as_u64().and_then(|p| u16::try_from(p).ok())
}
/// WebSocket handler: upgrade then pump bytes between the WS client and
/// the VNC server on `127.0.0.1:<vnc_port>`. Returns 404 when gui is not
/// enabled for this agent.
async fn screen_ws(
ws: axum::extract::ws::WebSocketUpgrade,
State(state): State<AppState>,
) -> Response {
let Some(vnc_port) = state.gui_vnc_port else {
return (StatusCode::NOT_FOUND, "gui not enabled for this agent").into_response();
};
ws.on_upgrade(move |socket| relay_ws_vnc(socket, vnc_port))
}
/// Pure byte pump: forwards raw bytes between the WebSocket client and
/// the VNC TCP stream. Transparent to any RFB variant (plain, VeNCrypt).
async fn relay_ws_vnc(socket: axum::extract::ws::WebSocket, vnc_port: u16) {
// Import futures traits locally so they don't conflict with
// tokio_stream::StreamExt used at module scope.
use axum::extract::ws::Message;
use futures_util::{SinkExt, StreamExt as _};
let addr = format!("127.0.0.1:{vnc_port}");
let Ok(tcp) = tokio::net::TcpStream::connect(&addr).await else {
tracing::warn!(%addr, "screen/ws: could not connect to VNC server");
return;
};
let (mut tcp_rx, mut tcp_tx) = tcp.into_split();
let (mut ws_tx, mut ws_rx) = socket.split();
// WS → TCP
let ws_to_tcp = tokio::spawn(async move {
while let Some(Ok(msg)) = futures_util::StreamExt::next(&mut ws_rx).await {
match msg {
Message::Binary(data) => {
if tcp_tx.write_all(&data).await.is_err() {
break;
}
}
Message::Close(_) => break,
_ => {} // ping/pong/text: ignore
}
}
});
// TCP → WS
let tcp_to_ws = tokio::spawn(async move {
let mut buf = vec![0u8; 8192];
loop {
match tcp_rx.read(&mut buf).await {
Ok(0) | Err(_) => break,
Ok(n) => {
if ws_tx
.send(Message::Binary(buf[..n].to_vec().into()))
.await
.is_err()
{
break;
}
}
}
}
});
// Wait for either direction to close, then let both tasks drop.
tokio::select! {
_ = ws_to_tcp => {}
_ = tcp_to_ws => {}
}
}
#[derive(Deserialize)]
struct StatsQuery {
window: Option<String>,
@ -271,6 +364,10 @@ struct StateSnapshot {
/// Cumulative token usage across the most recent turn's inferences
/// (cost signal). `null` until the first turn finishes.
cost_usage: Option<crate::events::TokenUsage>,
/// Whether the weston VNC compositor is configured for this agent
/// (i.e. `/etc/hyperhive/gui.json` was present at harness startup).
/// When true, the UI may render a `🖥 screen` link to `/screen`.
gui_enabled: bool,
}
#[derive(Serialize)]
@ -367,6 +464,7 @@ async fn api_state(State(state): State<AppState>) -> axum::Json<StateSnapshot> {
model,
ctx_usage,
cost_usage,
gui_enabled: state.gui_vnc_port.is_some(),
})
}