Skip to main content

trustee_api/
state.rs

1//! Shared server state: per-user session registry, broadcast channels, and auth state.
2//!
3//! ## Multi-User Architecture (TMU Phase 2)
4//!
5//! Each authenticated user gets their own [`UserSession`] containing:
6//! - An independent `Session` (workflow state, output, etc.)
7//! - A dedicated broadcast channel for WebSocket fan-out
8//! - A per-user token store for MCP credential isolation
9//!
10//! Sessions are keyed by user identity (`sub` claim from JWT, or `dev:email` for
11//! dev mode). Unauthenticated deployments use a single `"default"` key, preserving
12//! backward compatibility with single-user CLI operation.
13
14use std::sync::Arc;
15
16use dashmap::DashMap;
17use tokio::sync::{broadcast, mpsc, Mutex};
18use trustee_core::session::Session;
19use trustee_core::types::TuiMessage;
20
21use crate::auth::AuthState;
22
23/// Per-user session bundle.
24///
25/// Each user gets their own Session instance, broadcast channel, and
26/// token store. This struct is stored in the [`SessionRegistry`]
27/// and accessed via the user's identity key.
28pub struct UserSession {
29    /// The agent session, protected by a mutex.
30    pub session: Arc<Mutex<Session>>,
31    /// Broadcast sender for this user's WebSocket fan-out.
32    pub ws_tx: broadcast::Sender<String>,
33    /// Per-user in-memory token store for MCP credential isolation.
34    ///
35    /// Replaces the process-wide FileTokenStore that was vulnerable to
36    /// cross-user token leakage via __web_session.json. Each user's
37    /// MCP `web-session` tokens are stored here, isolated from other users.
38    pub token_store: Arc<pep::MemoryTokenStore>,
39}
40
41impl UserSession {
42    /// Create a new per-user session from an existing Session.
43    ///
44    /// Creates a fresh broadcast channel (256 capacity) for WebSocket fan-out
45    /// and a per-user MemoryTokenStore for MCP credential isolation.
46    pub fn new(session: Session) -> Self {
47        let (ws_tx, _ws_rx) = broadcast::channel::<String>(256);
48        let token_store = Arc::new(pep::MemoryTokenStore::new());
49        Self {
50            session: Arc::new(Mutex::new(session)),
51            ws_tx,
52            token_store,
53        }
54    }
55}
56
57/// Concurrent registry of per-user sessions.
58///
59/// Keyed by user identity string:
60/// - Authenticated: JWT `sub` claim (e.g., Kanidm UUID)
61/// - Dev mode: `dev:{email}`
62/// - No auth: `"default"`
63///
64/// Falls back to the `"default"` entry when no user key is provided,
65/// preserving backward compatibility.
66pub type SessionRegistry = Arc<DashMap<String, UserSession>>;
67
68/// Shared state accessible by all axum handlers.
69#[derive(Clone)]
70pub struct ServerState {
71    /// Per-user session registry (TMU Phase 2).
72    pub sessions: SessionRegistry,
73    /// Broadcast sender for backward compat — delegates to the default user's channel.
74    /// New code should use `user_ws_tx(user_key)` instead.
75    pub ws_tx: broadcast::Sender<String>,
76    /// Auth state (None = auth disabled, all endpoints open).
77    pub auth: Option<Arc<AuthState>>,
78    /// Shared config TOML (all users share the same agent config).
79    pub config_toml: Option<String>,
80    /// Shared secrets (injected into every per-user session).
81    pub secrets: Option<std::collections::HashMap<String, String>>,
82    /// Shared build info (injected into every per-user session).
83    pub build_info: Option<trustee_core::types::BuildInfo>,
84    /// Global concurrency limiter — limits the number of simultaneous workflows
85    /// across all users. Prevents resource exhaustion on shared infrastructure.
86    /// Default: 8 concurrent workflows.
87    pub workflow_semaphore: Arc<tokio::sync::Semaphore>,
88}
89
90impl ServerState {
91    /// Create new shared state from a default session, broadcast sender, and optional auth.
92    ///
93    /// The provided session becomes the `"default"` user's session. When auth is
94    /// enabled, authenticated users get their own sessions created on demand.
95    pub fn new(
96        session: Session,
97        ws_tx: broadcast::Sender<String>,
98        auth: Option<Arc<AuthState>>,
99    ) -> Self {
100        let sessions = Arc::new(DashMap::new());
101
102        // Store the default session under the "default" key
103        // Use the provided ws_tx as the default user's broadcast channel
104        let token_store = Arc::new(pep::MemoryTokenStore::new());
105        sessions.insert(
106            "default".to_string(),
107            UserSession {
108                session: Arc::new(Mutex::new(session)),
109                ws_tx: ws_tx.clone(),
110                token_store,
111            },
112        );
113
114        Self {
115            sessions,
116            ws_tx,
117            auth,
118            config_toml: None,
119            secrets: None,
120            build_info: None,
121            workflow_semaphore: Arc::new(tokio::sync::Semaphore::new(8)),
122        }
123    }
124
125    /// Set the shared config TOML.
126    pub fn with_config_toml(mut self, config_toml: String) -> Self {
127        self.config_toml = Some(config_toml);
128        self
129    }
130
131    /// Set the shared secrets.
132    pub fn with_secrets(mut self, secrets: std::collections::HashMap<String, String>) -> Self {
133        self.secrets = Some(secrets);
134        self
135    }
136
137    /// Set the shared build info.
138    pub fn with_build_info(mut self, build_info: trustee_core::types::BuildInfo) -> Self {
139        self.build_info = Some(build_info);
140        self
141    }
142
143    /// Set the max concurrent workflows.
144    pub fn with_max_concurrent_workflows(mut self, max: usize) -> Self {
145        self.workflow_semaphore = Arc::new(tokio::sync::Semaphore::new(max));
146        self
147    }
148
149    /// Get or create a session for the given user key, returning the session + ws_tx.
150    ///
151    /// This is the main entry point for route handlers. It ensures the user
152    /// has a session, spawns a drain task if newly created, and returns
153    /// references to the session mutex, broadcast sender, and token store.
154    pub async fn ensure_user_session(
155        &self,
156        user_key: &str,
157    ) -> (Arc<Mutex<Session>>, broadcast::Sender<String>, Arc<pep::MemoryTokenStore>) {
158        // Fast path: user already has a session
159        if let Some(entry) = self.sessions.get(user_key) {
160            return (
161                entry.session.clone(),
162                entry.ws_tx.clone(),
163                entry.token_store.clone(),
164            );
165        }
166
167        // Slow path: create new session for this user
168        let (mut session, workflow_rx) = Session::new();
169
170        // Copy shared config
171        if let Some(ref config_toml) = self.config_toml {
172            session.config_toml = Some(config_toml.clone());
173            session.parse_auto_handoff_config();
174
175            if let Ok(table) = config_toml.parse::<toml::Value>() {
176                if let Some(name) = table.get("agent").and_then(|a| a.get("name")).and_then(|n| n.as_str()) {
177                    session.agent_name = name.to_string();
178                }
179            }
180        }
181
182        // Copy shared secrets and build info
183        session.secrets = self.secrets.clone();
184        session.build_info = self.build_info.clone();
185
186        // Isolate checkpoint storage per user by setting a unique project_id.
187        // The project_id becomes the storage partition key in ABK's checkpoint
188        // system. By prefixing with the user_key, each user's checkpoints are
189        // stored in separate directories, preventing cross-user access.
190        // The "default" user (no auth) keeps the legacy behavior (no project_id).
191        if user_key != "default" {
192            session.project_id = Some(format!("user:{user_key}"));
193        }
194
195        let user_session = UserSession::new(session);
196        let session_arc = user_session.session.clone();
197        let ws_tx = user_session.ws_tx.clone();
198        let token_store = user_session.token_store.clone();
199
200        self.sessions.insert(user_key.to_string(), user_session);
201
202        // Spawn drain task for this user's workflow receiver
203        self.spawn_user_drain_task(
204            user_key.to_string(),
205            session_arc.clone(),
206            ws_tx.clone(),
207            workflow_rx,
208        );
209
210        (session_arc, ws_tx, token_store)
211    }
212
213    /// Spawn a background drain task for a specific user's workflow receiver.
214    ///
215    /// This replaces the old global drain task — each user gets their own.
216    fn spawn_user_drain_task(
217        &self,
218        user_key: String,
219        session: Arc<Mutex<Session>>,
220        ws_tx: broadcast::Sender<String>,
221        mut workflow_rx: mpsc::UnboundedReceiver<TuiMessage>,
222    ) {
223        tokio::spawn(async move {
224            while let Some(msg) = workflow_rx.recv().await {
225                // Process the message through Session's handler (updates state)
226                {
227                    let mut session = session.lock().await;
228                    session.handle_workflow_message(msg.clone());
229
230                    let state_str = match session.workflow_state {
231                        trustee_core::types::WorkflowState::Idle => "Idle",
232                        trustee_core::types::WorkflowState::Running => "Running",
233                        trustee_core::types::WorkflowState::Cancelling => "Cancelling",
234                    };
235                    let state_msg = serde_json::json!({
236                        "type": "StateChanged",
237                        "state": state_str
238                    });
239                    let _ = ws_tx.send(state_msg.to_string());
240                }
241
242                // Broadcast the raw message to WebSocket clients
243                let json = serde_json::to_string(&SerializableMessage(&msg)).unwrap_or_default();
244                let _ = ws_tx.send(json);
245            }
246            tracing::debug!("Drain task ended for user: {}", user_key);
247        });
248    }
249
250    /// Spawn the default user's drain task (backward compatibility).
251    ///
252    /// Called during server startup for the initial session.
253    pub fn spawn_drain_task(self, mut workflow_rx: mpsc::UnboundedReceiver<TuiMessage>) {
254        // Get the default session's arc
255        let default_entry = self.sessions.get("default").expect("default session must exist");
256        let session = default_entry.session.clone();
257        let ws_tx = default_entry.ws_tx.clone();
258        drop(default_entry);
259
260        tokio::spawn(async move {
261            while let Some(msg) = workflow_rx.recv().await {
262                {
263                    let mut session = session.lock().await;
264                    session.handle_workflow_message(msg.clone());
265
266                    let state_str = match session.workflow_state {
267                        trustee_core::types::WorkflowState::Idle => "Idle",
268                        trustee_core::types::WorkflowState::Running => "Running",
269                        trustee_core::types::WorkflowState::Cancelling => "Cancelling",
270                    };
271                    let state_msg = serde_json::json!({
272                        "type": "StateChanged",
273                        "state": state_str
274                    });
275                    let _ = ws_tx.send(state_msg.to_string());
276                }
277
278                let json = serde_json::to_string(&SerializableMessage(&msg)).unwrap_or_default();
279                let _ = ws_tx.send(json);
280            }
281        });
282    }
283
284    /// Resolve the user key from request headers.
285    ///
286    /// Returns `"default"` when auth is disabled.
287    /// Returns the JWT `sub` claim (or `dev:email` for dev mode) when auth is enabled.
288    pub async fn resolve_user_key(&self, headers: &axum::http::HeaderMap) -> String {
289        let Some(ref auth) = self.auth else {
290            return "default".to_string();
291        };
292
293        // Try Bearer header first
294        if let Some(token) = headers
295            .get(axum::http::header::AUTHORIZATION)
296            .and_then(|v| v.to_str().ok())
297            .and_then(|v| v.strip_prefix("Bearer "))
298            .map(|s| s.to_string())
299        {
300            // Dev token
301            if token.starts_with("dev:") {
302                let parts: Vec<&str> = token.splitn(4, ':').collect();
303                if parts.len() >= 4 {
304                    return format!("dev:{}", parts[1]);
305                }
306            }
307            // Real JWT — extract sub claim
308            if let Ok(claims) = auth.validate_token(&token).await {
309                return claims.sub;
310            }
311        }
312
313        // Try cookie
314        let cookie_session_id = headers
315            .get(axum::http::header::COOKIE)
316            .and_then(|v| v.to_str().ok())
317            .and_then(|cookies| {
318                cookies
319                    .split(';')
320                    .map(|c| c.trim())
321                    .find_map(|c| c.strip_prefix(&format!("{}=", auth.config.cookie_name)))
322                    .map(|s| s.to_string())
323            });
324
325        if let Some(session_id) = cookie_session_id {
326            // Dev token in cookie
327            if session_id.starts_with("dev:") {
328                let parts: Vec<&str> = session_id.splitn(4, ':').collect();
329                if parts.len() >= 4 {
330                    return format!("dev:{}", parts[1]);
331                }
332            }
333
334            // Resolve session_id → access_token → sub claim
335            if let Ok(access_token) = auth.session_manager.get_token(&session_id).await {
336                if let Ok(claims) = auth.validate_token(&access_token).await {
337                    return claims.sub;
338                }
339            }
340        }
341
342        "default".to_string()
343    }
344}
345
346/// Wrapper to serialize `TuiMessage` as JSON with a `type` discriminator.
347struct SerializableMessage<'a>(&'a TuiMessage);
348
349impl<'a> serde::Serialize for SerializableMessage<'a> {
350    fn serialize<S>(&self, serializer: S) -> Result<S::Ok, S::Error>
351    where
352        S: serde::Serializer,
353    {
354        use serde::ser::SerializeStruct;
355
356        match self.0 {
357            TuiMessage::OutputLine(line) => {
358                let mut s = serializer.serialize_struct("msg", 2)?;
359                s.serialize_field("type", "OutputLine")?;
360                s.serialize_field("line", line)?;
361                s.end()
362            }
363            TuiMessage::StreamDelta(delta) => {
364                let mut s = serializer.serialize_struct("msg", 2)?;
365                s.serialize_field("type", "StreamDelta")?;
366                s.serialize_field("delta", delta)?;
367                s.end()
368            }
369            TuiMessage::ReasoningDelta(delta) => {
370                let mut s = serializer.serialize_struct("msg", 2)?;
371                s.serialize_field("type", "ReasoningDelta")?;
372                s.serialize_field("delta", delta)?;
373                s.end()
374            }
375            TuiMessage::WorkflowCompleted => {
376                let mut s = serializer.serialize_struct("msg", 2)?;
377                s.serialize_field("type", "WorkflowCompleted")?;
378                s.serialize_field("state", "Idle")?;
379                s.end()
380            }
381            TuiMessage::WorkflowError(err) => {
382                let mut s = serializer.serialize_struct("msg", 2)?;
383                s.serialize_field("type", "WorkflowError")?;
384                s.serialize_field("error", err)?;
385                s.end()
386            }
387            TuiMessage::ResumeInfo(info) => {
388                match info {
389                    Some(ri) => {
390                        let mut s = serializer.serialize_struct("msg", 5)?;
391                        s.serialize_field("type", "ResumeInfo")?;
392                        s.serialize_field("state", "Idle")?;
393                        s.serialize_field("session_id", &ri.session_id)?;
394                        s.serialize_field("checkpoint_id", &ri.checkpoint_id)?;
395                        s.serialize_field("iteration", &ri.iteration)?;
396                        s.end()
397                    }
398                    None => {
399                        let mut s = serializer.serialize_struct("msg", 2)?;
400                        s.serialize_field("type", "ResumeInfo")?;
401                        s.serialize_field("state", "Idle")?;
402                        s.end()
403                    }
404                }
405            }
406            TuiMessage::TodoUpdate(content) => {
407                let mut s = serializer.serialize_struct("msg", 2)?;
408                s.serialize_field("type", "TodoUpdate")?;
409                s.serialize_field("content", content)?;
410                s.end()
411            }
412            TuiMessage::WorkflowCancelled => {
413                let mut s = serializer.serialize_struct("msg", 2)?;
414                s.serialize_field("type", "WorkflowCancelled")?;
415                s.serialize_field("state", "Idle")?;
416                s.end()
417            }
418            TuiMessage::HandoffReady(_) => {
419                let mut s = serializer.serialize_struct("msg", 2)?;
420                s.serialize_field("type", "HandoffReady")?;
421                s.serialize_field("state", "Idle")?;
422                s.end()
423            }
424            TuiMessage::ToolPending { tool_name, hint } => {
425                let mut s = serializer.serialize_struct("msg", 3)?;
426                s.serialize_field("type", "ToolPending")?;
427                s.serialize_field("tool_name", tool_name)?;
428                s.serialize_field("hint", hint)?;
429                s.end()
430            }
431            TuiMessage::ToolDone { tool_name, success, hint } => {
432                let mut s = serializer.serialize_struct("msg", 4)?;
433                s.serialize_field("type", "ToolDone")?;
434                s.serialize_field("tool_name", tool_name)?;
435                s.serialize_field("success", success)?;
436                s.serialize_field("hint", hint)?;
437                s.end()
438            }
439            TuiMessage::ContextTokensUpdated(count) => {
440                let mut s = serializer.serialize_struct("msg", 2)?;
441                s.serialize_field("type", "ContextTokensUpdated")?;
442                s.serialize_field("count", count)?;
443                s.end()
444            }
445            TuiMessage::McpServerStatus { name, connected, tool_count, error } => {
446                let mut s = serializer.serialize_struct("msg", 5)?;
447                s.serialize_field("type", "McpServerStatus")?;
448                s.serialize_field("name", name)?;
449                s.serialize_field("connected", connected)?;
450                s.serialize_field("tool_count", tool_count)?;
451                s.serialize_field("error", error)?;
452                s.end()
453            }
454        }
455    }
456}