OXYGENHUB

Coverage / crates/hub/src/lib.rs 87.2%

246 of 282 lines run, 42 of 96 functions · run #510 on main · run missed

1
//! OxygenHub: a forge for Oxygen repositories (`docs/spec.md`).
2
//!
3
//! One process, or several sharing a database: an axum router serving
4
//! pages, the JSON API and the wire protocol (the web role); the background
5
//! tasks (the worker role, singletons under a leader lock: docs/roles.md);
6
//! SQLite or PostgreSQL for hub state; repositories as bare Oxygen stores.
7
8
pub mod api;
9
pub mod apt;
10
pub mod assistants;
11
pub mod auth;
12
pub mod autoland;
13
pub mod backups;
14
pub mod bus;
15
pub mod config;
16
pub mod coverage;
17
pub mod db;
18
pub mod diffusion;
19
pub mod drops;
20
pub mod events;
21
pub mod flow;
22
pub mod headers;
23
pub mod history_scan;
24
pub mod hooks;
25
pub mod images;
26
pub mod insights;
27
pub mod junit;
28
pub mod keyless;
29
pub mod keys;
30
pub mod leader;
31
pub mod levels;
32
pub mod live;
33
pub mod mail;
34
pub mod maintenance;
35
pub mod metrics;
36
pub mod mirror;
37
pub mod net;
38
pub mod notify;
39
pub mod oidc;
40
pub mod protocol;
41
pub mod provenance;
42
pub mod push;
43
pub mod push_protection;
44
pub mod queue;
45
pub mod render;
46
pub mod reports;
47
pub mod repos;
48
pub mod runner_updates;
49
pub mod secrets;
50
pub mod setup;
51
pub mod shutdown;
52
pub mod signin;
53
pub mod social;
54
pub mod storage;
55
pub mod swarm;
56
pub mod tags;
57
pub mod terraform;
58
pub mod throttle;
59
pub mod tlog;
60
pub mod variables;
61
pub mod web;
62
pub mod wiki;
63
64
use crate::config::Config;
65
use crate::repos::Repos;
66
use std::sync::Arc;
67
68
/// Everything a request handler can reach.
69
pub struct App {
70
    pub config: Config,
71
    pub db: db::Pool,
72
    pub repos: Arc<Repos>,
73
    /// Wikis, one Oxygen repository per repository (hub spec §6.11).
74
    pub wikis: Arc<Repos>,
75
    pub server: Arc<oxy_server::Server>,
76
    pub hub_key: keys::HubKey,
77
    pub cookie_key: auth::CookieKey,
78
    /// Seals AI assistants' tokens at rest (hub spec §13.4).
79
    pub assistant_key: assistants::SecretKey,
80
    /// Seals Flow secrets at rest (hub spec §13.2, #107).
81
    pub secret_key: assistants::SecretKey,
82
    /// The OIDC issuer for jobs (hub spec §13.5), made on first use.
83
    pub oidc: tokio::sync::OnceCell<oidc::Issuer>,
84
    /// Which assistants are answering which proposals, and recent calls
85
    /// per repository (hub spec §13.4).
86
    pub assistant_gate: std::sync::Mutex<assistants::Gate>,
87
    /// Failed sign-ins and the like, per account and address (oxygenhub#221).
88
    pub throttle: std::sync::Mutex<throttle::Throttle>,
89
    /// Events for live pages and workers, here and in the other processes
90
    /// sharing the database (docs/roles.md, #273).
91
    pub bus: bus::Bus,
92
    /// Who is resolving which conflict together (hub spec §6.10).
93
    pub rooms: Arc<live::Rooms>,
94
    /// Outgoing mail: hub.toml's `[email]`, overridden by what the admin
95
    /// page saved (#101).
96
    pub mail: std::sync::RwLock<config::Email>,
97
    /// The image swarm's tracker (oxygenhub#144).
98
    pub swarm: swarm::Tracker,
99
    pub started: std::time::Instant,
100
    /// The transparency log (#238).
101
    pub tlog: Arc<tlog::Log>,
102
    /// What this process does: web, worker or both (docs/roles.md, #272).
103
    pub roles: config::Roles,
104
    /// The background tasks started here and the leader locks held (#272).
105
    pub leader: leader::Leader,
106
}
107
108
impl App {
109
    /// The rootfs images this hub keeps for runners (oxygenflow#39).
11031×
    pub fn images(&self) -> images::Store {
11131×
        images::Store::new(&self.config.data)
11231×
    }
113
114
    /// The OIDC issuer, its key loaded or made the first time it is asked for.
11526×
    pub async fn oidc(&self) -> anyhow::Result<&oidc::Issuer> {
11626×
        let keys = self.config.keys_dir();
11726×
        self.oidc.get_or_try_init(|| async move { tokio::task::spawn_blocking(move || oidc::Issuer::load_or_create(&keys)).await? }).await
11826×
    }
119
}
120
121
pub type Shared = Arc<App>;
122
123
/// Open the data directory and build the application state. `init` must
124
/// have run once; `serve` and the tests call this.
125200×
pub async fn build(config: Config) -> anyhow::Result<Shared> {
126200×
    let roles = config.roles()?;
127199×
    std::fs::create_dir_all(config.repos_dir())?;
128199×
    let db = db::open(&config).await?;
129199×
    let repos = Arc::new(Repos::new(config.repos_dir()));
130199×
    let wikis = Arc::new(Repos::new(config.data.join("wikis")));
131199×
    let hub_key = keys::ensure_identity(&config.keys_dir(), &config.site.name, config.base_url())?;
132199×
    let cookie_key = auth::CookieKey::load_or_create(&config.keys_dir())?;
133199×
    let assistant_key = assistants::SecretKey::load_or_create(&config.keys_dir())?;
134199×
    let secret_key = assistants::SecretKey::load_or_create_named(&config.keys_dir(), "secrets.key")?;
135199×
    let backend = protocol::HubBackend::new(
136199×
        repos.clone(),
137199×
        wikis.clone(),
138199×
        db.clone(),
139199×
        config.auth.push_tokens.clone(),
140199×
        tokio::runtime::Handle::current(),
141
    )
142199×
    .with_acceptor(hub_key.identity.clone(), keys::load_secret(&config.keys_dir())?.1)
143199×
    .with_scan_budget(push_protection::Budget {
144199×
        bytes: config.limits.scan_limit.0,
145199×
        file: config.limits.scan_file_limit.0,
146199×
        time: std::time::Duration::from_millis(config.limits.scan_time_ms),
147199×
    });
148199×
    let slot = backend.app_slot();
149199×
    let server = oxy_server::Server::new(backend);
150199×
    let mail = mail::load(&db, &config.email).await;
151
    // The web role appends to the log; a worker alone only reads it (#272).
152199×
    let (tlog_dir, origin) = (config.data.join("tlog"), tlog::origin_of(config.base_url()));
153199×
    let tlog = Arc::new(if roles.web { tlog::Log::open(&tlog_dir, &origin)? } else { tlog::Log::open_reader(&tlog_dir, &origin)? });
154199×
    tlog.configure(&config.tlog)?;
155199×
    let app = Arc::new(App {
156199×
        mail: std::sync::RwLock::new(mail),
157199×
        config,
158199×
        db,
159199×
        repos,
160199×
        wikis,
161199×
        server,
162199×
        hub_key,
163199×
        cookie_key,
164199×
        assistant_key,
165199×
        secret_key,
166199×
        oidc: tokio::sync::OnceCell::new(),
167199×
        assistant_gate: std::sync::Mutex::new(assistants::Gate::default()),
168199×
        throttle: std::sync::Mutex::new(throttle::Throttle::default()),
169199×
        bus: bus::Bus::default(),
170199×
        rooms: Arc::new(live::Rooms::default()),
171199×
        swarm: Default::default(),
172199×
        started: std::time::Instant::now(),
173199×
        tlog,
174199×
        roles,
175199×
        leader: leader::Leader::default(),
176199×
    });
177199×
    *slot.lock().unwrap() = Some(Arc::downgrade(&app));
178199×
    bus::start(&app).await?;
179199×
    storage::load(&app).await?;
180199×
    coverage::load(&app).await?;
181199×
    junit::load(&app).await?;
182199×
    reports::load(&app).await?;
183199×
    start_background(&app);
184199×
    setup::ensure(&app).await?;
185199×
    Ok(app)
186200×
}
187
188
/// Start this process's background tasks (docs/roles.md, #272). The
189
/// worker role's are singletons across every process sharing the database,
190
/// each run under its leader lock; the transparency log's checkpointer goes
191
/// with the web role, which appends to the log, and is a singleton too.
192199×
fn start_background(app: &Shared) {
193199×
    if app.roles.web {
194198×
        // The checkpointer signs the log this process appends to (#238).
195198×
        leader::run(app, "tlog.checkpointer", tlog::checkpointer);
196198×
    }
197199×
    if !app.roles.worker {
1982×
        return;
199197×
    }
200197×
    leader::run(app, "hooks.worker", hooks::worker);
201197×
    leader::run(app, "flow.reaper", flow::reaper);
202197×
    leader::run(app, "flow.scheduler", flow::scheduler);
203197×
    leader::run(app, "flow.placement", flow::placement);
204197×
    leader::run(app, "maintenance.nightly", maintenance::nightly);
205197×
    leader::run(app, "projects.archiver", maintenance::archiver);
206197×
    leader::run(app, "backups.scheduler", backups::scheduler);
207
    // The reverse pin index for repositories pushed before it existed (#148):
208
    // once, by whichever worker leads it first (idempotent if run again).
209393×
    leader::run(app, "diffusion.backfill", |app| async move { diffusion::backfill(&app).await });
210
    // Mend the pin index and the bumps it drives (oxygenhub#210).
211197×
    leader::run(app, "diffusion.reconciler", diffusion::reconciler);
212
    // Each repository's history scanned for secrets (#277), a budget at a
213
    // time, at low priority.
214197×
    leader::run(app, "secrets.history_scan", history_scan::worker);
215
    // The durable work queue (#276): every worker takes its due items.
216197×
    app.leader.note("queue.worker");
217197×
    tokio::spawn(queue::worker(app.clone()));
218
    // What each image is for, read or learned once (oxygenhub#173): safe
219
    // in every worker, so it takes no lock.
220197×
    app.leader.note("images.learn_archs");
221197×
    tokio::spawn({
222197×
        let app = app.clone();
223197×
        async move {
224197×
            match images::learn_archs(&app).await {
225196×
                Ok(n) if n > 0 => tracing::info!(images = n, "read the architecture of kept images"),
226196×
                Ok(_) => {}
2270×
                Err(e) => tracing::warn!(error = %e, "learning image architectures"),
228
            }
229196×
        }
230
    });
231199×
}
232
233
/// The router this process serves: everything for the web role; for a
234
/// worker alone, `/health` and `/metrics`, so it can be watched.
2352727×
pub fn router(app: Shared) -> axum::Router {
2362727×
    if !app.roles.web {
2378×
        return worker_router(app);
2382719×
    }
2392719×
    full_router(app)
2402727×
}
241
242
/// What a worker-only process answers: its health and its metrics.
2438×
fn worker_router(app: Shared) -> axum::Router {
244
    use axum::routing::get;
2458×
    let hsts = headers::hsts(app.config.base_url(), app.config.server.hsts_max_age);
2468×
    axum::Router::new()
2478×
        .route("/health", get(web::home::health))
2488×
        .route("/metrics", get(metrics::metrics))
2498×
        .with_state(app)
2508×
        .layer(axum::middleware::from_fn_with_state(hsts, headers::security_headers))
2518×
}
252
253
/// The full router: pages, API, protocol and static assets.
2542719×
fn full_router(app: Shared) -> axum::Router {
2552719×
    let https = app_https(&app);
2562719×
    let hsts = headers::hsts(app.config.base_url(), app.config.server.hsts_max_age);
2572719×
    web::router(app.clone())
2582719×
        .layer(axum::middleware::from_fn_with_state(app, agent_requests))
2592719×
        .layer(axum::middleware::from_fn(private_pages))
2602719×
        .layer(axum::middleware::from_fn(slow_requests))
2612719×
        .layer(axum::middleware::from_fn_with_state(https, https_only))
2622719×
        .layer(axum::middleware::from_fn_with_state(hsts, headers::security_headers))
2632719×
}
264
265
/// The site's https address, when it has one: what [`https_only`] redirects to.
2662719×
fn app_https(app: &Shared) -> Option<String> {
2672719×
    let base = app.config.base_url();
2682719×
    base.starts_with("https://").then(|| base.trim_end_matches('/').to_string())
2692719×
}
270
271
/// The scheme a proxy in front of the hub says the visitor used
272
/// (`X-Forwarded-Proto`, else Cloudflare's `CF-Visitor`); `None` for a
273
/// request made to the hub directly (runners and tools on its host).
2746×
fn forwarded_scheme(headers: &axum::http::HeaderMap) -> Option<String> {
2756×
    if let Some(p) = headers.get("x-forwarded-proto").and_then(|v| v.to_str().ok()) {
2764×
        return p.split(',').next().map(|s| s.trim().to_ascii_lowercase());
2772×
    }
2782×
    let visitor = headers.get("cf-visitor").and_then(|v| v.to_str().ok())?;
2791×
    serde_json::from_str::<serde_json::Value>(visitor).ok()?["scheme"].as_str().map(str::to_ascii_lowercase)
2806×
}
281
282
/// HTTPS only (oxygenhub#218): on a site whose address is https, a request
283
/// a proxy reports as plain http is sent to the https address (GET and
284
/// HEAD) or refused (anything else, so a form or a password is never taken
285
/// over http). Requests made to the hub directly carry no such report and
286
/// are untouched.
2873723×
async fn https_only(
2883723×
    axum::extract::State(https): axum::extract::State<Option<String>>,
2893723×
    req: axum::extract::Request,
2903723×
    next: axum::middleware::Next,
2913723×
) -> axum::response::Response {
292
    use axum::response::IntoResponse;
2933723×
    let Some(base) = https else { return next.run(req).await };
2946×
    if forwarded_scheme(req.headers()).as_deref() != Some("http") {
2954×
        return next.run(req).await;
2962×
    }
2972×
    let to = format!("{base}{}", req.uri().path_and_query().map(|p| p.as_str()).unwrap_or("/"));
2982×
    if matches!(*req.method(), axum::http::Method::GET | axum::http::Method::HEAD) {
2991×
        return axum::response::Redirect::permanent(&to).into_response();
3001×
    }
3011×
    (axum::http::StatusCode::BAD_REQUEST, format!("This site takes requests over HTTPS only: send it to {to}\n")).into_response()
3023723×
}
303
304
/// A request made with a token its owner marked as an agent's runs with
305
/// that token's name in [`notify::AGENT`], so what the agent does is told
306
/// to its owner (oxygenhub#183).
3073721×
async fn agent_requests(
3083721×
    axum::extract::State(app): axum::extract::State<Shared>,
3093721×
    req: axum::extract::Request,
3103721×
    next: axum::middleware::Next,
3113721×
) -> axum::response::Response {
3123721×
    let token = req
3133721×
        .headers()
3143721×
        .get(axum::http::header::AUTHORIZATION)
3153721×
        .and_then(|v| v.to_str().ok())
3163721×
        .and_then(|v| v.strip_prefix("Bearer "))
3173721×
        .map(|t| t.trim().to_string());
3183721×
    let agent = match token {
319
        // A keyless agent's grant token (#253) is always an agent's.
3201762×
        Some(t) if t.starts_with(keyless::GRANT_TOKEN_PREFIX) => db::accounts::agent_grant_by_token(&app.db, &auth::token_hash(&t))
32120×
            .await
32220×
            .ok()
32320×
            .flatten()
32420×
            .filter(|g| g.live(db::now_ms()))
32520×
            .map(|g| g.agent),
3261742×
        Some(t) => db::accounts::token_by_hash(&app.db, &auth::token_hash(&t)).await.ok().flatten().filter(|t| t.agent).map(|t| t.name),
3271959×
        None => None,
328
    };
3293721×
    notify::AGENT.scope(agent, next.run(req)).await
3303721×
}
331
332
/// Requests slower than this are logged, with their path (#145).
333
const SLOW: std::time::Duration = std::time::Duration::from_secs(2);
334
335
/// Log a slow request's method, path and time, so a slow page can be
336
/// found afterwards. Never the query, which may hold a token.
3373721×
async fn slow_requests(req: axum::extract::Request, next: axum::middleware::Next) -> axum::response::Response {
3383721×
    let (method, path) = (req.method().clone(), req.uri().path().to_string());
3393721×
    let start = std::time::Instant::now();
3403721×
    let res = next.run(req).await;
3413721×
    let took = start.elapsed();
3423721×
    metrics::observe(method.as_str(), &path, res.status().as_u16(), took.as_secs_f64());
3433721×
    if took >= SLOW {
3443×
        tracing::warn!("slow request: {method} {path} took {:.1} s ({})", took.as_secs_f64(), res.status().as_u16());
3453718×
    }
3463721×
    res
3473721×
}
348
349
/// A page for a signed-in viewer is theirs alone: it may list private
350
/// repositories, and the dashboard's sky carries their activity (#157). An
351
/// HTML response to a request with a session cookie, which has not said
352
/// how to cache itself, says `private, no-store`, so no shared cache (a
353
/// proxy or a CDN in front of the hub) ever serves it to anyone else.
3543721×
async fn private_pages(req: axum::extract::Request, next: axum::middleware::Next) -> axum::response::Response {
355
    use axum::http::header;
3563721×
    let signed_in = req
3573721×
        .headers()
3583721×
        .get_all(header::COOKIE)
3593721×
        .iter()
3603721×
        .filter_map(|v| v.to_str().ok())
3613721×
        .any(|c| c.split(';').any(|kv| kv.trim_start().starts_with("oxyhub=")));
3623721×
    let mut res = next.run(req).await;
3633721×
    let html = res.headers().get(header::CONTENT_TYPE).and_then(|v| v.to_str().ok()).is_some_and(|t| t.starts_with("text/html"));
3643721×
    if signed_in && html && !res.headers().contains_key(header::CACHE_CONTROL) {
365843×
        res.headers_mut().insert(header::CACHE_CONTROL, header::HeaderValue::from_static("private, no-store"));
3662878×
    }
3673721×
    res
3683721×
}
369
370
/// Serve on the configured addresses until the process is told to stop.
3711×
pub async fn serve(app: Shared) -> anyhow::Result<()> {
3721×
    let router = router(app.clone());
3731×
    let http = app.config.bind_addr()?;
3741×
    let https = app.config.tls_bind_addr()?;
3751×
    if http.is_none() && https.is_none() && app.roles.web {
3760×
        anyhow::bail!("nothing to bind: set server.bind or server.tls_bind");
3771×
    }
3781×
    tracing::info!("roles: {}", app.roles.names().join(", "));
379
    // Installed first: from here a SIGTERM drains rather than kills.
3801×
    let signal = shutdown::on_signal()?;
3811×
    let mut tasks = tokio::task::JoinSet::new();
3821×
    if let Some(addr) = http {
3831×
        let listener = match shutdown::systemd_listener() {
3840×
            Some(l) => {
3850×
                tracing::info!("http on the socket systemd holds ({})", l.local_addr().map(|a| a.to_string()).unwrap_or_default());
3860×
                tokio::net::TcpListener::from_std(l)?
387
            }
388
            None => {
3891×
                tracing::info!("http on {addr}");
3901×
                tokio::net::TcpListener::bind(addr).await?
391
            }
392
        };
3931×
        let r = router.clone();
3941×
        tasks.spawn(async move {
3951×
            axum::serve(listener, r.into_make_service_with_connect_info::<std::net::SocketAddr>())
3961×
                .with_graceful_shutdown(shutdown::stopping())
3971×
                .await
3981×
                .map_err(anyhow::Error::from)
3991×
        });
4000×
    }
4011×
    if let Some(addr) = https {
4020×
        let _ = rustls::crypto::ring::default_provider().install_default();
4030×
        let tls = axum_server::tls_rustls::RustlsConfig::from_pem_file(app.config.tls_cert(), app.config.tls_key())
4040×
            .await
4050×
            .map_err(|e| anyhow::anyhow!("loading TLS files: {e}"))?;
4060×
        tracing::info!("https on {addr}");
4070×
        let r = router.clone();
4080×
        let handle = axum_server::Handle::new();
4090×
        let draining = handle.clone();
4100×
        tokio::spawn(async move {
4110×
            shutdown::stopping().await;
4120×
            draining.graceful_shutdown(Some(shutdown::DRAIN));
4130×
        });
4140×
        tasks.spawn(async move {
4150×
            axum_server::bind_rustls(addr, tls)
4160×
                .handle(handle)
4170×
                .serve(r.into_make_service_with_connect_info::<std::net::SocketAddr>())
4180×
                .await
4190×
                .map_err(anyhow::Error::from)
4200×
        });
4211×
    }
4221×
    tokio::select! {
423
        // A worker with no address to serve on runs until it is stopped.
4241×
        r = tasks.join_next(), if !tasks.is_empty() => { r.transpose()?.transpose()?; Ok(()) }
4251×
        _ = signal => {
426
            // The servers stop taking connections and finish what they
427
            // have; whatever is still running after DRAIN is cut off.
4280×
            let drained = tokio::time::timeout(shutdown::DRAIN, async { while tasks.join_next().await.is_some() {} }).await;
4290×
            if drained.is_err() {
4300×
                tracing::warn!("requests still running after {} s: stopping anyway", shutdown::DRAIN.as_secs());
4310×
            }
4320×
            tracing::info!("stopped");
4330×
            Ok(())
434
        }
435
    }
4361×
}