OXYGENHUB

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

246 of 282 lines run, 42 of 96 functions · run #464 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_protection;
43
pub mod queue;
44
pub mod render;
45
pub mod reports;
46
pub mod repos;
47
pub mod secrets;
48
pub mod setup;
49
pub mod shutdown;
50
pub mod signin;
51
pub mod social;
52
pub mod storage;
53
pub mod swarm;
54
pub mod tags;
55
pub mod terraform;
56
pub mod throttle;
57
pub mod tlog;
58
pub mod variables;
59
pub mod web;
60
pub mod wiki;
61
62
use crate::config::Config;
63
use crate::repos::Repos;
64
use std::sync::Arc;
65
66
/// Everything a request handler can reach.
67
pub struct App {
68
    pub config: Config,
69
    pub db: db::Pool,
70
    pub repos: Arc<Repos>,
71
    /// Wikis, one Oxygen repository per repository (hub spec §6.11).
72
    pub wikis: Arc<Repos>,
73
    pub server: Arc<oxy_server::Server>,
74
    pub hub_key: keys::HubKey,
75
    pub cookie_key: auth::CookieKey,
76
    /// Seals AI assistants' tokens at rest (hub spec §13.4).
77
    pub assistant_key: assistants::SecretKey,
78
    /// Seals Flow secrets at rest (hub spec §13.2, #107).
79
    pub secret_key: assistants::SecretKey,
80
    /// The OIDC issuer for jobs (hub spec §13.5), made on first use.
81
    pub oidc: tokio::sync::OnceCell<oidc::Issuer>,
82
    /// Which assistants are answering which proposals, and recent calls
83
    /// per repository (hub spec §13.4).
84
    pub assistant_gate: std::sync::Mutex<assistants::Gate>,
85
    /// Failed sign-ins and the like, per account and address (oxygenhub#221).
86
    pub throttle: std::sync::Mutex<throttle::Throttle>,
87
    /// Events for live pages and workers, here and in the other processes
88
    /// sharing the database (docs/roles.md, #273).
89
    pub bus: bus::Bus,
90
    /// Who is resolving which conflict together (hub spec §6.10).
91
    pub rooms: Arc<live::Rooms>,
92
    /// Outgoing mail: hub.toml's `[email]`, overridden by what the admin
93
    /// page saved (#101).
94
    pub mail: std::sync::RwLock<config::Email>,
95
    /// The image swarm's tracker (oxygenhub#144).
96
    pub swarm: swarm::Tracker,
97
    pub started: std::time::Instant,
98
    /// The transparency log (#238).
99
    pub tlog: Arc<tlog::Log>,
100
    /// What this process does: web, worker or both (docs/roles.md, #272).
101
    pub roles: config::Roles,
102
    /// The background tasks started here and the leader locks held (#272).
103
    pub leader: leader::Leader,
104
}
105
106
impl App {
107
    /// The rootfs images this hub keeps for runners (oxygenflow#39).
10824×
    pub fn images(&self) -> images::Store {
10924×
        images::Store::new(&self.config.data)
11024×
    }
111
112
    /// The OIDC issuer, its key loaded or made the first time it is asked for.
11326×
    pub async fn oidc(&self) -> anyhow::Result<&oidc::Issuer> {
11426×
        let keys = self.config.keys_dir();
11526×
        self.oidc.get_or_try_init(|| async move { tokio::task::spawn_blocking(move || oidc::Issuer::load_or_create(&keys)).await? }).await
11626×
    }
117
}
118
119
pub type Shared = Arc<App>;
120
121
/// Open the data directory and build the application state. `init` must
122
/// have run once; `serve` and the tests call this.
123190×
pub async fn build(config: Config) -> anyhow::Result<Shared> {
124190×
    let roles = config.roles()?;
125189×
    std::fs::create_dir_all(config.repos_dir())?;
126189×
    let db = db::open(&config).await?;
127189×
    let repos = Arc::new(Repos::new(config.repos_dir()));
128189×
    let wikis = Arc::new(Repos::new(config.data.join("wikis")));
129189×
    let hub_key = keys::ensure_identity(&config.keys_dir(), &config.site.name, config.base_url())?;
130189×
    let cookie_key = auth::CookieKey::load_or_create(&config.keys_dir())?;
131189×
    let assistant_key = assistants::SecretKey::load_or_create(&config.keys_dir())?;
132189×
    let secret_key = assistants::SecretKey::load_or_create_named(&config.keys_dir(), "secrets.key")?;
133189×
    let backend = protocol::HubBackend::new(
134189×
        repos.clone(),
135189×
        wikis.clone(),
136189×
        db.clone(),
137189×
        config.auth.push_tokens.clone(),
138189×
        tokio::runtime::Handle::current(),
139
    )
140189×
    .with_acceptor(hub_key.identity.clone(), keys::load_secret(&config.keys_dir())?.1)
141189×
    .with_scan_budget(push_protection::Budget {
142189×
        bytes: config.limits.scan_limit.0,
143189×
        file: config.limits.scan_file_limit.0,
144189×
        time: std::time::Duration::from_millis(config.limits.scan_time_ms),
145189×
    });
146189×
    let slot = backend.app_slot();
147189×
    let server = oxy_server::Server::new(backend);
148189×
    let mail = mail::load(&db, &config.email).await;
149
    // The web role appends to the log; a worker alone only reads it (#272).
150189×
    let (tlog_dir, origin) = (config.data.join("tlog"), tlog::origin_of(config.base_url()));
151189×
    let tlog = Arc::new(if roles.web { tlog::Log::open(&tlog_dir, &origin)? } else { tlog::Log::open_reader(&tlog_dir, &origin)? });
152189×
    tlog.configure(&config.tlog)?;
153189×
    let app = Arc::new(App {
154189×
        mail: std::sync::RwLock::new(mail),
155189×
        config,
156189×
        db,
157189×
        repos,
158189×
        wikis,
159189×
        server,
160189×
        hub_key,
161189×
        cookie_key,
162189×
        assistant_key,
163189×
        secret_key,
164189×
        oidc: tokio::sync::OnceCell::new(),
165189×
        assistant_gate: std::sync::Mutex::new(assistants::Gate::default()),
166189×
        throttle: std::sync::Mutex::new(throttle::Throttle::default()),
167189×
        bus: bus::Bus::default(),
168189×
        rooms: Arc::new(live::Rooms::default()),
169189×
        swarm: Default::default(),
170189×
        started: std::time::Instant::now(),
171189×
        tlog,
172189×
        roles,
173189×
        leader: leader::Leader::default(),
174189×
    });
175189×
    *slot.lock().unwrap() = Some(Arc::downgrade(&app));
176189×
    bus::start(&app).await?;
177189×
    storage::load(&app).await?;
178189×
    coverage::load(&app).await?;
179189×
    junit::load(&app).await?;
180189×
    reports::load(&app).await?;
181189×
    start_background(&app);
182189×
    setup::ensure(&app).await?;
183189×
    Ok(app)
184190×
}
185
186
/// Start this process's background tasks (docs/roles.md, #272). The
187
/// worker role's are singletons across every process sharing the database,
188
/// each run under its leader lock; the transparency log's checkpointer goes
189
/// with the web role, which appends to the log, and is a singleton too.
190189×
fn start_background(app: &Shared) {
191189×
    if app.roles.web {
192188×
        // The checkpointer signs the log this process appends to (#238).
193188×
        leader::run(app, "tlog.checkpointer", tlog::checkpointer);
194188×
    }
195189×
    if !app.roles.worker {
1962×
        return;
197187×
    }
198187×
    leader::run(app, "hooks.worker", hooks::worker);
199187×
    leader::run(app, "flow.reaper", flow::reaper);
200187×
    leader::run(app, "flow.scheduler", flow::scheduler);
201187×
    leader::run(app, "flow.placement", flow::placement);
202187×
    leader::run(app, "maintenance.nightly", maintenance::nightly);
203187×
    leader::run(app, "projects.archiver", maintenance::archiver);
204187×
    leader::run(app, "backups.scheduler", backups::scheduler);
205
    // The reverse pin index for repositories pushed before it existed (#148):
206
    // once, by whichever worker leads it first (idempotent if run again).
207373×
    leader::run(app, "diffusion.backfill", |app| async move { diffusion::backfill(&app).await });
208
    // Mend the pin index and the bumps it drives (oxygenhub#210).
209187×
    leader::run(app, "diffusion.reconciler", diffusion::reconciler);
210
    // Each repository's history scanned for secrets (#277), a budget at a
211
    // time, at low priority.
212187×
    leader::run(app, "secrets.history_scan", history_scan::worker);
213
    // The durable work queue (#276): every worker takes its due items.
214187×
    app.leader.note("queue.worker");
215187×
    tokio::spawn(queue::worker(app.clone()));
216
    // What each image is for, read or learned once (oxygenhub#173): safe
217
    // in every worker, so it takes no lock.
218187×
    app.leader.note("images.learn_archs");
219187×
    tokio::spawn({
220187×
        let app = app.clone();
221187×
        async move {
222187×
            match images::learn_archs(&app).await {
223187×
                Ok(n) if n > 0 => tracing::info!(images = n, "read the architecture of kept images"),
224187×
                Ok(_) => {}
2250×
                Err(e) => tracing::warn!(error = %e, "learning image architectures"),
226
            }
227187×
        }
228
    });
229189×
}
230
231
/// The router this process serves: everything for the web role; for a
232
/// worker alone, `/health` and `/metrics`, so it can be watched.
2332581×
pub fn router(app: Shared) -> axum::Router {
2342581×
    if !app.roles.web {
2358×
        return worker_router(app);
2362573×
    }
2372573×
    full_router(app)
2382581×
}
239
240
/// What a worker-only process answers: its health and its metrics.
2418×
fn worker_router(app: Shared) -> axum::Router {
242
    use axum::routing::get;
2438×
    let hsts = headers::hsts(app.config.base_url(), app.config.server.hsts_max_age);
2448×
    axum::Router::new()
2458×
        .route("/health", get(web::home::health))
2468×
        .route("/metrics", get(metrics::metrics))
2478×
        .with_state(app)
2488×
        .layer(axum::middleware::from_fn_with_state(hsts, headers::security_headers))
2498×
}
250
251
/// The full router: pages, API, protocol and static assets.
2522573×
fn full_router(app: Shared) -> axum::Router {
2532573×
    let https = app_https(&app);
2542573×
    let hsts = headers::hsts(app.config.base_url(), app.config.server.hsts_max_age);
2552573×
    web::router(app.clone())
2562573×
        .layer(axum::middleware::from_fn_with_state(app, agent_requests))
2572573×
        .layer(axum::middleware::from_fn(private_pages))
2582573×
        .layer(axum::middleware::from_fn(slow_requests))
2592573×
        .layer(axum::middleware::from_fn_with_state(https, https_only))
2602573×
        .layer(axum::middleware::from_fn_with_state(hsts, headers::security_headers))
2612573×
}
262
263
/// The site's https address, when it has one: what [`https_only`] redirects to.
2642573×
fn app_https(app: &Shared) -> Option<String> {
2652573×
    let base = app.config.base_url();
2662573×
    base.starts_with("https://").then(|| base.trim_end_matches('/').to_string())
2672573×
}
268
269
/// The scheme a proxy in front of the hub says the visitor used
270
/// (`X-Forwarded-Proto`, else Cloudflare's `CF-Visitor`); `None` for a
271
/// request made to the hub directly (runners and tools on its host).
2726×
fn forwarded_scheme(headers: &axum::http::HeaderMap) -> Option<String> {
2736×
    if let Some(p) = headers.get("x-forwarded-proto").and_then(|v| v.to_str().ok()) {
2744×
        return p.split(',').next().map(|s| s.trim().to_ascii_lowercase());
2752×
    }
2762×
    let visitor = headers.get("cf-visitor").and_then(|v| v.to_str().ok())?;
2771×
    serde_json::from_str::<serde_json::Value>(visitor).ok()?["scheme"].as_str().map(str::to_ascii_lowercase)
2786×
}
279
280
/// HTTPS only (oxygenhub#218): on a site whose address is https, a request
281
/// a proxy reports as plain http is sent to the https address (GET and
282
/// HEAD) or refused (anything else, so a form or a password is never taken
283
/// over http). Requests made to the hub directly carry no such report and
284
/// are untouched.
2853535×
async fn https_only(
2863535×
    axum::extract::State(https): axum::extract::State<Option<String>>,
2873535×
    req: axum::extract::Request,
2883535×
    next: axum::middleware::Next,
2893535×
) -> axum::response::Response {
290
    use axum::response::IntoResponse;
2913535×
    let Some(base) = https else { return next.run(req).await };
2926×
    if forwarded_scheme(req.headers()).as_deref() != Some("http") {
2934×
        return next.run(req).await;
2942×
    }
2952×
    let to = format!("{base}{}", req.uri().path_and_query().map(|p| p.as_str()).unwrap_or("/"));
2962×
    if matches!(*req.method(), axum::http::Method::GET | axum::http::Method::HEAD) {
2971×
        return axum::response::Redirect::permanent(&to).into_response();
2981×
    }
2991×
    (axum::http::StatusCode::BAD_REQUEST, format!("This site takes requests over HTTPS only: send it to {to}\n")).into_response()
3003535×
}
301
302
/// A request made with a token its owner marked as an agent's runs with
303
/// that token's name in [`notify::AGENT`], so what the agent does is told
304
/// to its owner (oxygenhub#183).
3053533×
async fn agent_requests(
3063533×
    axum::extract::State(app): axum::extract::State<Shared>,
3073533×
    req: axum::extract::Request,
3083533×
    next: axum::middleware::Next,
3093533×
) -> axum::response::Response {
3103533×
    let token = req
3113533×
        .headers()
3123533×
        .get(axum::http::header::AUTHORIZATION)
3133533×
        .and_then(|v| v.to_str().ok())
3143533×
        .and_then(|v| v.strip_prefix("Bearer "))
3153533×
        .map(|t| t.trim().to_string());
3163533×
    let agent = match token {
317
        // A keyless agent's grant token (#253) is always an agent's.
3181690×
        Some(t) if t.starts_with(keyless::GRANT_TOKEN_PREFIX) => db::accounts::agent_grant_by_token(&app.db, &auth::token_hash(&t))
31912×
            .await
32012×
            .ok()
32112×
            .flatten()
32212×
            .filter(|g| g.live(db::now_ms()))
32312×
            .map(|g| g.agent),
3241678×
        Some(t) => db::accounts::token_by_hash(&app.db, &auth::token_hash(&t)).await.ok().flatten().filter(|t| t.agent).map(|t| t.name),
3251843×
        None => None,
326
    };
3273533×
    notify::AGENT.scope(agent, next.run(req)).await
3283533×
}
329
330
/// Requests slower than this are logged, with their path (#145).
331
const SLOW: std::time::Duration = std::time::Duration::from_secs(2);
332
333
/// Log a slow request's method, path and time, so a slow page can be
334
/// found afterwards. Never the query, which may hold a token.
3353533×
async fn slow_requests(req: axum::extract::Request, next: axum::middleware::Next) -> axum::response::Response {
3363533×
    let (method, path) = (req.method().clone(), req.uri().path().to_string());
3373533×
    let start = std::time::Instant::now();
3383533×
    let res = next.run(req).await;
3393533×
    let took = start.elapsed();
3403533×
    metrics::observe(method.as_str(), &path, res.status().as_u16(), took.as_secs_f64());
3413533×
    if took >= SLOW {
3424×
        tracing::warn!("slow request: {method} {path} took {:.1} s ({})", took.as_secs_f64(), res.status().as_u16());
3433529×
    }
3443533×
    res
3453533×
}
346
347
/// A page for a signed-in viewer is theirs alone: it may list private
348
/// repositories, and the dashboard's sky carries their activity (#157). An
349
/// HTML response to a request with a session cookie, which has not said
350
/// how to cache itself, says `private, no-store`, so no shared cache (a
351
/// proxy or a CDN in front of the hub) ever serves it to anyone else.
3523533×
async fn private_pages(req: axum::extract::Request, next: axum::middleware::Next) -> axum::response::Response {
353
    use axum::http::header;
3543533×
    let signed_in = req
3553533×
        .headers()
3563533×
        .get_all(header::COOKIE)
3573533×
        .iter()
3583533×
        .filter_map(|v| v.to_str().ok())
3593533×
        .any(|c| c.split(';').any(|kv| kv.trim_start().starts_with("oxyhub=")));
3603533×
    let mut res = next.run(req).await;
3613533×
    let html = res.headers().get(header::CONTENT_TYPE).and_then(|v| v.to_str().ok()).is_some_and(|t| t.starts_with("text/html"));
3623533×
    if signed_in && html && !res.headers().contains_key(header::CACHE_CONTROL) {
363777×
        res.headers_mut().insert(header::CACHE_CONTROL, header::HeaderValue::from_static("private, no-store"));
3642756×
    }
3653533×
    res
3663533×
}
367
368
/// Serve on the configured addresses until the process is told to stop.
3691×
pub async fn serve(app: Shared) -> anyhow::Result<()> {
3701×
    let router = router(app.clone());
3711×
    let http = app.config.bind_addr()?;
3721×
    let https = app.config.tls_bind_addr()?;
3731×
    if http.is_none() && https.is_none() && app.roles.web {
3740×
        anyhow::bail!("nothing to bind: set server.bind or server.tls_bind");
3751×
    }
3761×
    tracing::info!("roles: {}", app.roles.names().join(", "));
377
    // Installed first: from here a SIGTERM drains rather than kills.
3781×
    let signal = shutdown::on_signal()?;
3791×
    let mut tasks = tokio::task::JoinSet::new();
3801×
    if let Some(addr) = http {
3811×
        let listener = match shutdown::systemd_listener() {
3820×
            Some(l) => {
3830×
                tracing::info!("http on the socket systemd holds ({})", l.local_addr().map(|a| a.to_string()).unwrap_or_default());
3840×
                tokio::net::TcpListener::from_std(l)?
385
            }
386
            None => {
3871×
                tracing::info!("http on {addr}");
3881×
                tokio::net::TcpListener::bind(addr).await?
389
            }
390
        };
3911×
        let r = router.clone();
3921×
        tasks.spawn(async move {
3931×
            axum::serve(listener, r.into_make_service_with_connect_info::<std::net::SocketAddr>())
3941×
                .with_graceful_shutdown(shutdown::stopping())
3951×
                .await
3961×
                .map_err(anyhow::Error::from)
3971×
        });
3980×
    }
3991×
    if let Some(addr) = https {
4000×
        let _ = rustls::crypto::ring::default_provider().install_default();
4010×
        let tls = axum_server::tls_rustls::RustlsConfig::from_pem_file(app.config.tls_cert(), app.config.tls_key())
4020×
            .await
4030×
            .map_err(|e| anyhow::anyhow!("loading TLS files: {e}"))?;
4040×
        tracing::info!("https on {addr}");
4050×
        let r = router.clone();
4060×
        let handle = axum_server::Handle::new();
4070×
        let draining = handle.clone();
4080×
        tokio::spawn(async move {
4090×
            shutdown::stopping().await;
4100×
            draining.graceful_shutdown(Some(shutdown::DRAIN));
4110×
        });
4120×
        tasks.spawn(async move {
4130×
            axum_server::bind_rustls(addr, tls)
4140×
                .handle(handle)
4150×
                .serve(r.into_make_service_with_connect_info::<std::net::SocketAddr>())
4160×
                .await
4170×
                .map_err(anyhow::Error::from)
4180×
        });
4191×
    }
4201×
    tokio::select! {
421
        // A worker with no address to serve on runs until it is stopped.
4221×
        r = tasks.join_next(), if !tasks.is_empty() => { r.transpose()?.transpose()?; Ok(()) }
4231×
        _ = signal => {
424
            // The servers stop taking connections and finish what they
425
            // have; whatever is still running after DRAIN is cut off.
4260×
            let drained = tokio::time::timeout(shutdown::DRAIN, async { while tasks.join_next().await.is_some() {} }).await;
4270×
            if drained.is_err() {
4280×
                tracing::warn!("requests still running after {} s: stopping anyway", shutdown::DRAIN.as_secs());
4290×
            }
4300×
            tracing::info!("stopped");
4310×
            Ok(())
432
        }
433
    }
4341×
}