OXYGENHUB
◆ oh-comment-body-file oxygenhub / crates / hub / src / lib.rs
14.5 KiB · Rust · last changed Insights: OxygenFlow's standard charts, for the site, an owner and a repository 17 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 insights;
21
pub mod junit;
22
pub mod keyless;
23
pub mod keys;
24
pub mod levels;
25
pub mod live;
26
pub mod mail;
27
pub mod maintenance;
28
pub mod metrics;
29
pub mod mirror;
30
pub mod net;
31
pub mod notify;
32
pub mod oidc;
33
pub mod protocol;
34
pub mod provenance;
35
pub mod render;
36
pub mod reports;
37
pub mod repos;
38
pub mod secrets;
39
pub mod setup;
40
pub mod shutdown;
41
pub mod signin;
42
pub mod social;
43
pub mod storage;
44
pub mod swarm;
45
pub mod terraform;
46
pub mod throttle;
47
pub mod tlog;
48
pub mod variables;
49
pub mod web;
50
pub mod wiki;
51
52
use crate::config::Config;
53
use crate::repos::Repos;
54
use std::sync::Arc;
55
56
/// Everything a request handler can reach.
57
pub struct App {
58
    pub config: Config,
59
    pub db: db::Pool,
60
    pub repos: Arc<Repos>,
61
    /// Wikis, one Oxygen repository per repository (hub spec §6.11).
62
    pub wikis: Arc<Repos>,
63
    pub server: Arc<oxy_server::Server>,
64
    pub hub_key: keys::HubKey,
65
    pub cookie_key: auth::CookieKey,
66
    /// Seals AI assistants' tokens at rest (hub spec §13.4).
67
    pub assistant_key: assistants::SecretKey,
68
    /// Seals Flow secrets at rest (hub spec §13.2, #107).
69
    pub secret_key: assistants::SecretKey,
70
    /// The OIDC issuer for jobs (hub spec §13.5), made on first use.
71
    pub oidc: tokio::sync::OnceCell<oidc::Issuer>,
72
    /// Which assistants are answering which proposals, and recent calls
73
    /// per repository (hub spec §13.4).
74
    pub assistant_gate: std::sync::Mutex<assistants::Gate>,
75
    /// Failed sign-ins and the like, per account and address (oxygenhub#221).
76
    pub throttle: std::sync::Mutex<throttle::Throttle>,
77
    pub events: events::Events,
78
    /// Who is resolving which conflict together (hub spec §6.10).
79
    pub rooms: Arc<live::Rooms>,
80
    /// Outgoing mail: hub.toml's `[email]`, overridden by what the admin
81
    /// page saved (#101).
82
    pub mail: std::sync::RwLock<config::Email>,
83
    /// The image swarm's tracker (oxygenhub#144).
84
    pub swarm: swarm::Tracker,
85
    pub started: std::time::Instant,
86
    /// The transparency log (#238).
87
    pub tlog: Arc<tlog::Log>,
88
}
89
90
impl App {
91
    /// The rootfs images this hub keeps for runners (oxygenflow#39).
92
    pub fn images(&self) -> images::Store {
93
        images::Store::new(&self.config.data)
94
    }
95
96
    /// The OIDC issuer, its key loaded or made the first time it is asked for.
97
    pub async fn oidc(&self) -> anyhow::Result<&oidc::Issuer> {
98
        let keys = self.config.keys_dir();
99
        self.oidc.get_or_try_init(|| async move { tokio::task::spawn_blocking(move || oidc::Issuer::load_or_create(&keys)).await? }).await
100
    }
101
}
102
103
pub type Shared = Arc<App>;
104
105
/// Open the data directory and build the application state. `init` must
106
/// have run once; `serve` and the tests call this.
107
pub async fn build(config: Config) -> anyhow::Result<Shared> {
108
    std::fs::create_dir_all(config.repos_dir())?;
109
    let db = db::open(&config).await?;
110
    let repos = Arc::new(Repos::new(config.repos_dir()));
111
    let wikis = Arc::new(Repos::new(config.data.join("wikis")));
112
    let hub_key = keys::ensure_identity(&config.keys_dir(), &config.site.name, config.base_url())?;
113
    let cookie_key = auth::CookieKey::load_or_create(&config.keys_dir())?;
114
    let assistant_key = assistants::SecretKey::load_or_create(&config.keys_dir())?;
115
    let secret_key = assistants::SecretKey::load_or_create_named(&config.keys_dir(), "secrets.key")?;
116
    let backend = protocol::HubBackend::new(
117
        repos.clone(),
118
        wikis.clone(),
119
        db.clone(),
120
        config.auth.push_tokens.clone(),
121
        tokio::runtime::Handle::current(),
122
    )
123
    .with_acceptor(hub_key.identity.clone(), keys::load_secret(&config.keys_dir())?.1);
124
    let slot = backend.app_slot();
125
    let server = oxy_server::Server::new(backend);
126
    let mail = mail::load(&db, &config.email).await;
127
    let tlog = Arc::new(tlog::Log::open(&config.data.join("tlog"), &tlog::origin_of(config.base_url()))?);
128
    tlog.configure(&config.tlog)?;
129
    let app = Arc::new(App {
130
        mail: std::sync::RwLock::new(mail),
131
        config,
132
        db,
133
        repos,
134
        wikis,
135
        server,
136
        hub_key,
137
        cookie_key,
138
        assistant_key,
139
        secret_key,
140
        oidc: tokio::sync::OnceCell::new(),
141
        assistant_gate: std::sync::Mutex::new(assistants::Gate::default()),
142
        throttle: std::sync::Mutex::new(throttle::Throttle::default()),
143
        events: events::Events::default(),
144
        rooms: Arc::new(live::Rooms::default()),
145
        swarm: Default::default(),
146
        started: std::time::Instant::now(),
147
        tlog,
148
    });
149
    *slot.lock().unwrap() = Some(Arc::downgrade(&app));
150
    storage::load(&app).await?;
151
    coverage::load(&app).await?;
152
    junit::load(&app).await?;
153
    reports::load(&app).await?;
154
    tokio::spawn(hooks::worker(app.clone()));
155
    tokio::spawn(tlog::checkpointer(app.clone()));
156
    tokio::spawn(flow::reaper(app.clone()));
157
    tokio::spawn(flow::scheduler(app.clone()));
158
    tokio::spawn(flow::placement(app.clone()));
159
    tokio::spawn(maintenance::nightly(app.clone()));
160
    tokio::spawn(backups::scheduler(app.clone()));
161
    // The reverse pin index for repositories pushed before it existed (#148).
162
    tokio::spawn({
163
        let app = app.clone();
164
        async move { diffusion::backfill(&app).await }
165
    });
166
    // Mend the pin index and the bumps it drives (oxygenhub#210).
167
    tokio::spawn(diffusion::reconciler(app.clone()));
168
    // What each image is for, read or learned once (oxygenhub#173).
169
    tokio::spawn({
170
        let app = app.clone();
171
        async move {
172
            match images::learn_archs(&app).await {
173
                Ok(n) if n > 0 => tracing::info!(images = n, "read the architecture of kept images"),
174
                Ok(_) => {}
175
                Err(e) => tracing::warn!(error = %e, "learning image architectures"),
176
            }
177
        }
178
    });
179
    setup::ensure(&app).await?;
180
    Ok(app)
181
}
182
183
/// The full router: pages, API, protocol and static assets.
184
pub fn router(app: Shared) -> axum::Router {
185
    let https = app_https(&app);
186
    let hsts = headers::hsts(app.config.base_url(), app.config.server.hsts_max_age);
187
    web::router(app.clone())
188
        .layer(axum::middleware::from_fn_with_state(app, agent_requests))
189
        .layer(axum::middleware::from_fn(private_pages))
190
        .layer(axum::middleware::from_fn(slow_requests))
191
        .layer(axum::middleware::from_fn_with_state(https, https_only))
192
        .layer(axum::middleware::from_fn_with_state(hsts, headers::security_headers))
193
}
194
195
/// The site's https address, when it has one: what [`https_only`] redirects to.
196
fn app_https(app: &Shared) -> Option<String> {
197
    let base = app.config.base_url();
198
    base.starts_with("https://").then(|| base.trim_end_matches('/').to_string())
199
}
200
201
/// The scheme a proxy in front of the hub says the visitor used
202
/// (`X-Forwarded-Proto`, else Cloudflare's `CF-Visitor`); `None` for a
203
/// request made to the hub directly (runners and tools on its host).
204
fn forwarded_scheme(headers: &axum::http::HeaderMap) -> Option<String> {
205
    if let Some(p) = headers.get("x-forwarded-proto").and_then(|v| v.to_str().ok()) {
206
        return p.split(',').next().map(|s| s.trim().to_ascii_lowercase());
207
    }
208
    let visitor = headers.get("cf-visitor").and_then(|v| v.to_str().ok())?;
209
    serde_json::from_str::<serde_json::Value>(visitor).ok()?["scheme"].as_str().map(str::to_ascii_lowercase)
210
}
211
212
/// HTTPS only (oxygenhub#218): on a site whose address is https, a request
213
/// a proxy reports as plain http is sent to the https address (GET and
214
/// HEAD) or refused (anything else, so a form or a password is never taken
215
/// over http). Requests made to the hub directly carry no such report and
216
/// are untouched.
217
async fn https_only(
218
    axum::extract::State(https): axum::extract::State<Option<String>>,
219
    req: axum::extract::Request,
220
    next: axum::middleware::Next,
221
) -> axum::response::Response {
222
    use axum::response::IntoResponse;
223
    let Some(base) = https else { return next.run(req).await };
224
    if forwarded_scheme(req.headers()).as_deref() != Some("http") {
225
        return next.run(req).await;
226
    }
227
    let to = format!("{base}{}", req.uri().path_and_query().map(|p| p.as_str()).unwrap_or("/"));
228
    if matches!(*req.method(), axum::http::Method::GET | axum::http::Method::HEAD) {
229
        return axum::response::Redirect::permanent(&to).into_response();
230
    }
231
    (axum::http::StatusCode::BAD_REQUEST, format!("This site takes requests over HTTPS only: send it to {to}\n")).into_response()
232
}
233
234
/// A request made with a token its owner marked as an agent's runs with
235
/// that token's name in [`notify::AGENT`], so what the agent does is told
236
/// to its owner (oxygenhub#183).
237
async fn agent_requests(
238
    axum::extract::State(app): axum::extract::State<Shared>,
239
    req: axum::extract::Request,
240
    next: axum::middleware::Next,
241
) -> axum::response::Response {
242
    let token = req
243
        .headers()
244
        .get(axum::http::header::AUTHORIZATION)
245
        .and_then(|v| v.to_str().ok())
246
        .and_then(|v| v.strip_prefix("Bearer "))
247
        .map(|t| t.trim().to_string());
248
    let agent = match token {
249
        // A keyless agent's grant token (#253) is always an agent's.
250
        Some(t) if t.starts_with(keyless::GRANT_TOKEN_PREFIX) => db::accounts::agent_grant_by_token(&app.db, &auth::token_hash(&t))
251
            .await
252
            .ok()
253
            .flatten()
254
            .filter(|g| g.live(db::now_ms()))
255
            .map(|g| g.agent),
256
        Some(t) => db::accounts::token_by_hash(&app.db, &auth::token_hash(&t)).await.ok().flatten().filter(|t| t.agent).map(|t| t.name),
257
        None => None,
258
    };
259
    notify::AGENT.scope(agent, next.run(req)).await
260
}
261
262
/// Requests slower than this are logged, with their path (#145).
263
const SLOW: std::time::Duration = std::time::Duration::from_secs(2);
264
265
/// Log a slow request's method, path and time, so a slow page can be
266
/// found afterwards. Never the query, which may hold a token.
267
async fn slow_requests(req: axum::extract::Request, next: axum::middleware::Next) -> axum::response::Response {
268
    let (method, path) = (req.method().clone(), req.uri().path().to_string());
269
    let start = std::time::Instant::now();
270
    let res = next.run(req).await;
271
    let took = start.elapsed();
272
    metrics::observe(method.as_str(), &path, res.status().as_u16(), took.as_secs_f64());
273
    if took >= SLOW {
274
        tracing::warn!("slow request: {method} {path} took {:.1} s ({})", took.as_secs_f64(), res.status().as_u16());
275
    }
276
    res
277
}
278
279
/// A page for a signed-in viewer is theirs alone: it may list private
280
/// repositories, and the dashboard's sky carries their activity (#157). An
281
/// HTML response to a request with a session cookie, which has not said
282
/// how to cache itself, says `private, no-store`, so no shared cache (a
283
/// proxy or a CDN in front of the hub) ever serves it to anyone else.
284
async fn private_pages(req: axum::extract::Request, next: axum::middleware::Next) -> axum::response::Response {
285
    use axum::http::header;
286
    let signed_in = req
287
        .headers()
288
        .get_all(header::COOKIE)
289
        .iter()
290
        .filter_map(|v| v.to_str().ok())
291
        .any(|c| c.split(';').any(|kv| kv.trim_start().starts_with("oxyhub=")));
292
    let mut res = next.run(req).await;
293
    let html = res.headers().get(header::CONTENT_TYPE).and_then(|v| v.to_str().ok()).is_some_and(|t| t.starts_with("text/html"));
294
    if signed_in && html && !res.headers().contains_key(header::CACHE_CONTROL) {
295
        res.headers_mut().insert(header::CACHE_CONTROL, header::HeaderValue::from_static("private, no-store"));
296
    }
297
    res
298
}
299
300
/// Serve on the configured addresses until the process is told to stop.
301
pub async fn serve(app: Shared) -> anyhow::Result<()> {
302
    let router = router(app.clone());
303
    let http = app.config.bind_addr()?;
304
    let https = app.config.tls_bind_addr()?;
305
    if http.is_none() && https.is_none() {
306
        anyhow::bail!("nothing to bind: set server.bind or server.tls_bind");
307
    }
308
    // Installed first: from here a SIGTERM drains rather than kills.
309
    let signal = shutdown::on_signal()?;
310
    let mut tasks = tokio::task::JoinSet::new();
311
    if let Some(addr) = http {
312
        let listener = match shutdown::systemd_listener() {
313
            Some(l) => {
314
                tracing::info!("http on the socket systemd holds ({})", l.local_addr().map(|a| a.to_string()).unwrap_or_default());
315
                tokio::net::TcpListener::from_std(l)?
316
            }
317
            None => {
318
                tracing::info!("http on {addr}");
319
                tokio::net::TcpListener::bind(addr).await?
320
            }
321
        };
322
        let r = router.clone();
323
        tasks.spawn(async move {
324
            axum::serve(listener, r.into_make_service_with_connect_info::<std::net::SocketAddr>())
325
                .with_graceful_shutdown(shutdown::stopping())
326
                .await
327
                .map_err(anyhow::Error::from)
328
        });
329
    }
330
    if let Some(addr) = https {
331
        let _ = rustls::crypto::ring::default_provider().install_default();
332
        let tls = axum_server::tls_rustls::RustlsConfig::from_pem_file(app.config.tls_cert(), app.config.tls_key())
333
            .await
334
            .map_err(|e| anyhow::anyhow!("loading TLS files: {e}"))?;
335
        tracing::info!("https on {addr}");
336
        let r = router.clone();
337
        let handle = axum_server::Handle::new();
338
        let draining = handle.clone();
339
        tokio::spawn(async move {
340
            shutdown::stopping().await;
341
            draining.graceful_shutdown(Some(shutdown::DRAIN));
342
        });
343
        tasks.spawn(async move {
344
            axum_server::bind_rustls(addr, tls)
345
                .handle(handle)
346
                .serve(r.into_make_service_with_connect_info::<std::net::SocketAddr>())
347
                .await
348
                .map_err(anyhow::Error::from)
349
        });
350
    }
351
    tokio::select! {
352
        r = tasks.join_next() => { r.transpose()?.transpose()?; Ok(()) }
353
        _ = signal => {
354
            // The servers stop taking connections and finish what they
355
            // have; whatever is still running after DRAIN is cut off.
356
            let drained = tokio::time::timeout(shutdown::DRAIN, async { while tasks.join_next().await.is_some() {} }).await;
357
            if drained.is_err() {
358
                tracing::warn!("requests still running after {} s: stopping anyway", shutdown::DRAIN.as_secs());
359
            }
360
            tracing::info!("stopped");
361
            Ok(())
362
        }
363
    }
364
}