OXYGENHUB
# 3eb491e9489b oxygenhub / crates / hub / src / lib.rs
13.6 KiB · Rust · last changed The hub countersigns updates by keys it certified (keyless step 4c) 18 days ago source blame history raw
1
//! OxygenHub: a forge for Oxygen repositories (`docs/spec.md`).
2
//!
3
//! One process: an axum router serving pages, the JSON API and the wire
4
//! protocol; SQLite for hub state; repositories as bare Oxygen stores.
5
6
pub mod api;
7
pub mod apt;
8
pub mod assistants;
9
pub mod auth;
10
pub mod backups;
11
pub mod config;
12
pub mod coverage;
13
pub mod db;
14
pub mod diffusion;
15
pub mod events;
16
pub mod flow;
17
pub mod headers;
18
pub mod hooks;
19
pub mod images;
20
pub mod junit;
21
pub mod keyless;
22
pub mod keys;
23
pub mod levels;
24
pub mod live;
25
pub mod mail;
26
pub mod maintenance;
27
pub mod mirror;
28
pub mod net;
29
pub mod notify;
30
pub mod oidc;
31
pub mod protocol;
32
pub mod render;
33
pub mod reports;
34
pub mod repos;
35
pub mod secrets;
36
pub mod setup;
37
pub mod shutdown;
38
pub mod signin;
39
pub mod social;
40
pub mod storage;
41
pub mod swarm;
42
pub mod throttle;
43
pub mod variables;
44
pub mod web;
45
pub mod wiki;
46
47
use crate::config::Config;
48
use crate::repos::Repos;
49
use std::sync::Arc;
50
51
/// Everything a request handler can reach.
52
pub struct App {
53
    pub config: Config,
54
    pub db: db::Pool,
55
    pub repos: Arc<Repos>,
56
    /// Wikis, one Oxygen repository per repository (hub spec §6.11).
57
    pub wikis: Arc<Repos>,
58
    pub server: Arc<oxy_server::Server>,
59
    pub hub_key: keys::HubKey,
60
    pub cookie_key: auth::CookieKey,
61
    /// Seals AI assistants' tokens at rest (hub spec §13.4).
62
    pub assistant_key: assistants::SecretKey,
63
    /// Seals Flow secrets at rest (hub spec §13.2, #107).
64
    pub secret_key: assistants::SecretKey,
65
    /// The OIDC issuer for jobs (hub spec §13.5), made on first use.
66
    pub oidc: tokio::sync::OnceCell<oidc::Issuer>,
67
    /// Which assistants are answering which proposals, and recent calls
68
    /// per repository (hub spec §13.4).
69
    pub assistant_gate: std::sync::Mutex<assistants::Gate>,
70
    /// Failed sign-ins and the like, per account and address (oxygenhub#221).
71
    pub throttle: std::sync::Mutex<throttle::Throttle>,
72
    pub events: events::Events,
73
    /// Who is resolving which conflict together (hub spec §6.10).
74
    pub rooms: Arc<live::Rooms>,
75
    /// Outgoing mail: hub.toml's `[email]`, overridden by what the admin
76
    /// page saved (#101).
77
    pub mail: std::sync::RwLock<config::Email>,
78
    /// The image swarm's tracker (oxygenhub#144).
79
    pub swarm: swarm::Tracker,
80
    pub started: std::time::Instant,
81
}
82
83
impl App {
84
    /// The rootfs images this hub keeps for runners (oxygenflow#39).
85
    pub fn images(&self) -> images::Store {
86
        images::Store::new(&self.config.data)
87
    }
88
89
    /// The OIDC issuer, its key loaded or made the first time it is asked for.
90
    pub async fn oidc(&self) -> anyhow::Result<&oidc::Issuer> {
91
        let keys = self.config.keys_dir();
92
        self.oidc.get_or_try_init(|| async move { tokio::task::spawn_blocking(move || oidc::Issuer::load_or_create(&keys)).await? }).await
93
    }
94
}
95
96
pub type Shared = Arc<App>;
97
98
/// Open the data directory and build the application state. `init` must
99
/// have run once; `serve` and the tests call this.
100
pub async fn build(config: Config) -> anyhow::Result<Shared> {
101
    std::fs::create_dir_all(config.repos_dir())?;
102
    let db = db::open(&config).await?;
103
    let repos = Arc::new(Repos::new(config.repos_dir()));
104
    let wikis = Arc::new(Repos::new(config.data.join("wikis")));
105
    let hub_key = keys::ensure_identity(&config.keys_dir(), &config.site.name, config.base_url())?;
106
    let cookie_key = auth::CookieKey::load_or_create(&config.keys_dir())?;
107
    let assistant_key = assistants::SecretKey::load_or_create(&config.keys_dir())?;
108
    let secret_key = assistants::SecretKey::load_or_create_named(&config.keys_dir(), "secrets.key")?;
109
    let backend = protocol::HubBackend::new(
110
        repos.clone(),
111
        wikis.clone(),
112
        db.clone(),
113
        config.auth.push_tokens.clone(),
114
        tokio::runtime::Handle::current(),
115
    )
116
    .with_acceptor(hub_key.identity.clone(), keys::load_secret(&config.keys_dir())?.1);
117
    let slot = backend.app_slot();
118
    let server = oxy_server::Server::new(backend);
119
    let mail = mail::load(&db, &config.email).await;
120
    let app = Arc::new(App {
121
        mail: std::sync::RwLock::new(mail),
122
        config,
123
        db,
124
        repos,
125
        wikis,
126
        server,
127
        hub_key,
128
        cookie_key,
129
        assistant_key,
130
        secret_key,
131
        oidc: tokio::sync::OnceCell::new(),
132
        assistant_gate: std::sync::Mutex::new(assistants::Gate::default()),
133
        throttle: std::sync::Mutex::new(throttle::Throttle::default()),
134
        events: events::Events::default(),
135
        rooms: Arc::new(live::Rooms::default()),
136
        swarm: Default::default(),
137
        started: std::time::Instant::now(),
138
    });
139
    *slot.lock().unwrap() = Some(Arc::downgrade(&app));
140
    storage::load(&app).await?;
141
    coverage::load(&app).await?;
142
    junit::load(&app).await?;
143
    reports::load(&app).await?;
144
    tokio::spawn(hooks::worker(app.clone()));
145
    tokio::spawn(flow::reaper(app.clone()));
146
    tokio::spawn(flow::scheduler(app.clone()));
147
    tokio::spawn(flow::placement(app.clone()));
148
    tokio::spawn(maintenance::nightly(app.clone()));
149
    tokio::spawn(backups::scheduler(app.clone()));
150
    // The reverse pin index for repositories pushed before it existed (#148).
151
    tokio::spawn({
152
        let app = app.clone();
153
        async move { diffusion::backfill(&app).await }
154
    });
155
    // Mend the pin index and the bumps it drives (oxygenhub#210).
156
    tokio::spawn(diffusion::reconciler(app.clone()));
157
    // What each image is for, read or learned once (oxygenhub#173).
158
    tokio::spawn({
159
        let app = app.clone();
160
        async move {
161
            match images::learn_archs(&app).await {
162
                Ok(n) if n > 0 => tracing::info!(images = n, "read the architecture of kept images"),
163
                Ok(_) => {}
164
                Err(e) => tracing::warn!(error = %e, "learning image architectures"),
165
            }
166
        }
167
    });
168
    setup::ensure(&app).await?;
169
    Ok(app)
170
}
171
172
/// The full router: pages, API, protocol and static assets.
173
pub fn router(app: Shared) -> axum::Router {
174
    let https = app_https(&app);
175
    web::router(app.clone())
176
        .layer(axum::middleware::from_fn_with_state(app, agent_requests))
177
        .layer(axum::middleware::from_fn(private_pages))
178
        .layer(axum::middleware::from_fn(slow_requests))
179
        .layer(axum::middleware::from_fn_with_state(https, https_only))
180
        .layer(axum::middleware::from_fn(headers::security_headers))
181
}
182
183
/// The site's https address, when it has one: what [`https_only`] redirects to.
184
fn app_https(app: &Shared) -> Option<String> {
185
    let base = app.config.base_url();
186
    base.starts_with("https://").then(|| base.trim_end_matches('/').to_string())
187
}
188
189
/// The scheme a proxy in front of the hub says the visitor used
190
/// (`X-Forwarded-Proto`, else Cloudflare's `CF-Visitor`); `None` for a
191
/// request made to the hub directly (runners and tools on its host).
192
fn forwarded_scheme(headers: &axum::http::HeaderMap) -> Option<String> {
193
    if let Some(p) = headers.get("x-forwarded-proto").and_then(|v| v.to_str().ok()) {
194
        return p.split(',').next().map(|s| s.trim().to_ascii_lowercase());
195
    }
196
    let visitor = headers.get("cf-visitor").and_then(|v| v.to_str().ok())?;
197
    serde_json::from_str::<serde_json::Value>(visitor).ok()?["scheme"].as_str().map(str::to_ascii_lowercase)
198
}
199
200
/// HTTPS only (oxygenhub#218): on a site whose address is https, a request
201
/// a proxy reports as plain http is sent to the https address (GET and
202
/// HEAD) or refused (anything else, so a form or a password is never taken
203
/// over http). Requests made to the hub directly carry no such report and
204
/// are untouched.
205
async fn https_only(
206
    axum::extract::State(https): axum::extract::State<Option<String>>,
207
    req: axum::extract::Request,
208
    next: axum::middleware::Next,
209
) -> axum::response::Response {
210
    use axum::response::IntoResponse;
211
    let Some(base) = https else { return next.run(req).await };
212
    if forwarded_scheme(req.headers()).as_deref() != Some("http") {
213
        return next.run(req).await;
214
    }
215
    let to = format!("{base}{}", req.uri().path_and_query().map(|p| p.as_str()).unwrap_or("/"));
216
    if matches!(*req.method(), axum::http::Method::GET | axum::http::Method::HEAD) {
217
        return axum::response::Redirect::permanent(&to).into_response();
218
    }
219
    (axum::http::StatusCode::BAD_REQUEST, format!("This site takes requests over HTTPS only: send it to {to}\n")).into_response()
220
}
221
222
/// A request made with a token its owner marked as an agent's runs with
223
/// that token's name in [`notify::AGENT`], so what the agent does is told
224
/// to its owner (oxygenhub#183).
225
async fn agent_requests(
226
    axum::extract::State(app): axum::extract::State<Shared>,
227
    req: axum::extract::Request,
228
    next: axum::middleware::Next,
229
) -> axum::response::Response {
230
    let token = req
231
        .headers()
232
        .get(axum::http::header::AUTHORIZATION)
233
        .and_then(|v| v.to_str().ok())
234
        .and_then(|v| v.strip_prefix("Bearer "))
235
        .map(|t| auth::token_hash(t.trim()));
236
    let agent = match token {
237
        Some(hash) => db::accounts::token_by_hash(&app.db, &hash).await.ok().flatten().filter(|t| t.agent).map(|t| t.name),
238
        None => None,
239
    };
240
    notify::AGENT.scope(agent, next.run(req)).await
241
}
242
243
/// Requests slower than this are logged, with their path (#145).
244
const SLOW: std::time::Duration = std::time::Duration::from_secs(2);
245
246
/// Log a slow request's method, path and time, so a slow page can be
247
/// found afterwards. Never the query, which may hold a token.
248
async fn slow_requests(req: axum::extract::Request, next: axum::middleware::Next) -> axum::response::Response {
249
    let (method, path) = (req.method().clone(), req.uri().path().to_string());
250
    let start = std::time::Instant::now();
251
    let res = next.run(req).await;
252
    let took = start.elapsed();
253
    if took >= SLOW {
254
        tracing::warn!("slow request: {method} {path} took {:.1} s ({})", took.as_secs_f64(), res.status().as_u16());
255
    }
256
    res
257
}
258
259
/// A page for a signed-in viewer is theirs alone: it may list private
260
/// repositories, and the dashboard's sky carries their activity (#157). An
261
/// HTML response to a request with a session cookie, which has not said
262
/// how to cache itself, says `private, no-store`, so no shared cache (a
263
/// proxy or a CDN in front of the hub) ever serves it to anyone else.
264
async fn private_pages(req: axum::extract::Request, next: axum::middleware::Next) -> axum::response::Response {
265
    use axum::http::header;
266
    let signed_in = req
267
        .headers()
268
        .get_all(header::COOKIE)
269
        .iter()
270
        .filter_map(|v| v.to_str().ok())
271
        .any(|c| c.split(';').any(|kv| kv.trim_start().starts_with("oxyhub=")));
272
    let mut res = next.run(req).await;
273
    let html = res.headers().get(header::CONTENT_TYPE).and_then(|v| v.to_str().ok()).is_some_and(|t| t.starts_with("text/html"));
274
    if signed_in && html && !res.headers().contains_key(header::CACHE_CONTROL) {
275
        res.headers_mut().insert(header::CACHE_CONTROL, header::HeaderValue::from_static("private, no-store"));
276
    }
277
    res
278
}
279
280
/// Serve on the configured addresses until the process is told to stop.
281
pub async fn serve(app: Shared) -> anyhow::Result<()> {
282
    let router = router(app.clone());
283
    let http = app.config.bind_addr()?;
284
    let https = app.config.tls_bind_addr()?;
285
    if http.is_none() && https.is_none() {
286
        anyhow::bail!("nothing to bind: set server.bind or server.tls_bind");
287
    }
288
    // Installed first: from here a SIGTERM drains rather than kills.
289
    let signal = shutdown::on_signal()?;
290
    let mut tasks = tokio::task::JoinSet::new();
291
    if let Some(addr) = http {
292
        let listener = match shutdown::systemd_listener() {
293
            Some(l) => {
294
                tracing::info!("http on the socket systemd holds ({})", l.local_addr().map(|a| a.to_string()).unwrap_or_default());
295
                tokio::net::TcpListener::from_std(l)?
296
            }
297
            None => {
298
                tracing::info!("http on {addr}");
299
                tokio::net::TcpListener::bind(addr).await?
300
            }
301
        };
302
        let r = router.clone();
303
        tasks.spawn(async move {
304
            axum::serve(listener, r.into_make_service_with_connect_info::<std::net::SocketAddr>())
305
                .with_graceful_shutdown(shutdown::stopping())
306
                .await
307
                .map_err(anyhow::Error::from)
308
        });
309
    }
310
    if let Some(addr) = https {
311
        let _ = rustls::crypto::ring::default_provider().install_default();
312
        let tls = axum_server::tls_rustls::RustlsConfig::from_pem_file(app.config.tls_cert(), app.config.tls_key())
313
            .await
314
            .map_err(|e| anyhow::anyhow!("loading TLS files: {e}"))?;
315
        tracing::info!("https on {addr}");
316
        let r = router.clone();
317
        let handle = axum_server::Handle::new();
318
        let draining = handle.clone();
319
        tokio::spawn(async move {
320
            shutdown::stopping().await;
321
            draining.graceful_shutdown(Some(shutdown::DRAIN));
322
        });
323
        tasks.spawn(async move {
324
            axum_server::bind_rustls(addr, tls)
325
                .handle(handle)
326
                .serve(r.into_make_service_with_connect_info::<std::net::SocketAddr>())
327
                .await
328
                .map_err(anyhow::Error::from)
329
        });
330
    }
331
    tokio::select! {
332
        r = tasks.join_next() => { r.transpose()?.transpose()?; Ok(()) }
333
        _ = signal => {
334
            // The servers stop taking connections and finish what they
335
            // have; whatever is still running after DRAIN is cut off.
336
            let drained = tokio::time::timeout(shutdown::DRAIN, async { while tasks.join_next().await.is_some() {} }).await;
337
            if drained.is_err() {
338
                tracing::warn!("requests still running after {} s: stopping anyway", shutdown::DRAIN.as_secs());
339
            }
340
            tracing::info!("stopped");
341
            Ok(())
342
        }
343
    }
344
}