hyperhive/swarm-controller/src/gitlab_account.rs

//! An agent's accounts on GitLab instances: an operator hands us a base URL and
//! a token, we put them in the swarm's secret store.
//!
//! The agent end is `nix/agent-modules/gitlab-accounts.nix`, which lists the
//! agent's accounts and reads each under the agent's own certificate, and writes
//! the `glab` configuration. No hive is in the path.

use axum::Json;
use axum::extract::State;
use axum::http::StatusCode;
use serde::{Deserialize, Serialize};
use swarm_secret_client::gitlab;
use utoipa::ToSchema;

use super::linked_accounts::link;
use crate::backend_egress::host_of;
use crate::routes::agents::swarm_hive;
use crate::routes::{AppState, error_problem};

/// The account to store for one agent's GitLab instance.
///
/// No `Debug` derive: this carries a token.
#[derive(Deserialize, ToSchema)]
pub struct PutGitlabAccountRequest {
    /// The instance's base URL, `http://` or `https://`, with a DNS name as its
    /// host. A trailing slash is dropped.
    #[schema(example = "https://gitlab.example.org")]
    url: String,
    /// The access token, scoped `api` and `write_repository`. Never logged, and
    /// never returned by this route.
    token: String,
}

/// `put_gitlab_account`'s success body.
#[derive(Debug, Serialize, ToSchema)]
pub struct PutGitlabAccountResponse {
    /// The base URL as stored.
    url: String,
}

/// Store an agent's GitLab account, unless one is stored under the label
/// already.
#[utoipa::path(
    put,
    path = "/api/hives/{hive}/agents/{agent}/gitlab-accounts/{label}",
    params(
        ("hive" = String, Path, description = "hive the agent runs on"),
        ("agent" = String, Path, description = "agent the account belongs to"),
        ("label" = String, Path, description = "the account's name"),
    ),
    request_body = PutGitlabAccountRequest,
    responses(
        (status = 200, description = "stored", body = PutGitlabAccountResponse),
        (status = 400, description = "the agent or label is not an identifier, the URL is not http(s) with a DNS name as its host, the token is empty, or the hive is not in this swarm (problem+json)", body = String),
        (status = 409, description = "an account is stored under the label already; nothing was written (problem+json)", body = String),
        (status = 500, description = "the store could not be read or written (problem+json)", body = String),
    ),
    tag = "agents"
)]
pub async fn put_gitlab_account(
    State(state): State<AppState>,
    axum::extract::Path((hive, agent, label)): axum::extract::Path<(String, String, String)>,
    Json(req): Json<PutGitlabAccountRequest>,
) -> Result<Json<PutGitlabAccountResponse>, problem_details::ProblemDetails> {
    let hive = swarm_hive(&state, &hive).map_err(|(s, d)| error_problem(s, &d))?;
    let agent = hive_types::Ident::parse(&agent)
        .map_err(|reason| error_problem(StatusCode::BAD_REQUEST, reason))?
        .into_string();
    let label = hive_types::Ident::parse(&label)
        .map_err(|reason| error_problem(StatusCode::BAD_REQUEST, reason))?
        .into_string();
    let secret_path = gitlab::account_path(&agent, &label)
        .map_err(|e| error_problem(StatusCode::BAD_REQUEST, &e.to_string()))?;
    let account = account(req).map_err(|e| error_problem(StatusCode::BAD_REQUEST, &e))?;

    let store = state.accounts.open().await.map_err(|e| {
        tracing::warn!(error = %e, "connecting to the swarm secret store failed");
        error_problem(StatusCode::INTERNAL_SERVER_ERROR, &e.to_string())
    })?;
    link(&store, &secret_path, &account).await.map_err(|e| {
        // The path names the agent and the label; the value is not in it.
        tracing::warn!(path = %secret_path, error = ?e, "linking the gitlab account failed");
        e.problem(&existing(&agent, &label))
    })?;
    crate::link_egress::added(
        &state,
        &agent,
        &crate::link_egress::Link::Gitlab {
            label: &label,
            url: &account.url,
        },
    )
    .await?;

    tracing::info!(%hive, %agent, %label, url = %account.url, "gitlab account stored");
    Ok(Json(PutGitlabAccountResponse { url: account.url }))
}

/// The account a refused link names.
fn existing(agent: &str, label: &str) -> String {
    format!("agent {agent} already has gitlab account {label:?}")
}

/// The request as it is stored, or why it cannot be.
///
/// The URL must match [`gitlab::URL_PATTERN`], the rule the agent end filters by,
/// so a stored account is never one the agent drops. The host must also be a
/// name the egress allow list takes: the agent reaches the instance through its
/// hive's proxy, so an address or `localhost` would be stored and never reachable.
fn account(req: PutGitlabAccountRequest) -> Result<gitlab::Account, String> {
    let url = req.url.trim().trim_end_matches('/');
    if !(url.starts_with("http://") || url.starts_with("https://")) {
        return Err("url must start with http:// or https://".to_owned());
    }
    let authority = url
        .split_once("://")
        .and_then(|(_, rest)| rest.split(['/', '?', '#']).next())
        .unwrap_or_default();
    if authority.contains('@') {
        return Err("url must not carry user information; the token is its own field".to_owned());
    }
    if !gitlab::is_instance_url(url) {
        return Err(format!(
            "url must match {}: http(s), a host name, an optional port, and an optional path of \
             letters, digits and . _ ~ / - (no query, fragment, '@' or spaces)",
            gitlab::URL_PATTERN
        ));
    }
    let host = host_of(url).ok_or("url has no host")?;
    swarm_wire::egress_allow::normalize_domain(host).map_err(|why| format!("url host: {why}"))?;
    if req.token.trim().is_empty() {
        return Err("token is required".to_owned());
    }
    Ok(gitlab::Account {
        value: req.token,
        url: url.to_owned(),
    })
}

#[cfg(test)]
mod tests {
    use std::sync::Arc;

    use super::super::linked_accounts::{Accounts, tests::FakeStore};
    use super::{PutGitlabAccountRequest, account};

    fn request(url: &str, token: &str) -> PutGitlabAccountRequest {
        PutGitlabAccountRequest {
            url: url.to_owned(),
            token: token.to_owned(),
        }
    }

    #[test]
    fn the_url_loses_its_trailing_slash_and_the_token_is_kept() {
        let a = account(request("https://gitlab.example.org/ ", "t0k3n")).expect("valid");
        assert_eq!(a.url, "https://gitlab.example.org");
        assert_eq!(a.value, "t0k3n");
        // A port and a subfolder are part of the instance's address.
        let a = account(request("https://gitlab.example.org:8443/gl/", "t")).expect("valid");
        assert_eq!(a.url, "https://gitlab.example.org:8443/gl");
    }

    #[test]
    fn a_url_that_is_not_http_is_refused() {
        assert!(account(request("gitlab.example.org", "t")).is_err());
        assert!(account(request("file:///etc/passwd", "t")).is_err());
        // The control: plain http with a name is accepted.
        assert!(account(request("http://gitlab.lan", "t")).is_ok());
    }

    #[test]
    fn a_host_the_egress_list_would_refuse_is_refused() {
        for url in [
            "http://10.0.0.5",
            "https://10.0.0.5:8443",
            "http://[::1]",
            "https://localhost",
            "https://gitlab",
            "https://*.example.org",
            "https://",
        ] {
            assert!(account(request(url, "t")).is_err(), "{url}");
        }
    }

    #[test]
    fn user_information_in_the_url_is_refused() {
        assert!(account(request("https://oauth2:t0k3n@gitlab.example.org", "t")).is_err());
        // A path character the agent end drops is refused here too.
        for url in [
            "https://gitlab.example.org/@x",
            "https://gitlab.example.org/?x=1",
            "https://gitlab.example.org/#f",
            "https://gitlab.example.org/a b",
        ] {
            assert!(account(request(url, "t")).is_err(), "{url}");
        }
    }

    #[test]
    fn an_empty_token_is_refused() {
        assert!(account(request("https://gitlab.example.org", " ")).is_err());
    }

    /// Bare-minimum `AppState`, as `forge_account`'s tests build it.
    fn state(store: &Arc<FakeStore>) -> crate::routes::AppState {
        crate::routes::AppState {
            hives: std::sync::Arc::new(vec![crate::routes::hives::HiveEntry {
                name: "pr1ma".to_owned(),
                domain: "pr1ma.example".to_owned(),
                dashboard: "pr1ma.hives.example".to_owned(),
            }]),
            links: std::sync::Arc::new(Vec::new()),
            status: None,
            wanted: None,
            agent_status: None,
            agent_icons: None,
            agents_md: None,
            reported_inference: None,
            agent_login: None,
            jobq: std::sync::Arc::new(std::sync::Mutex::new(hive_jobq::scheduler::Scheduler::new(
                hive_jobq::Graph::new(),
                hive_jobq::resources::ResourceTable::new(),
            ))),
            webhook_secret: std::sync::Arc::default(),
            config_prs: std::sync::Arc::default(),
            swarm_name: None,
            auth: None,
            forge: std::sync::Arc::default(),
            create_gate: std::sync::Arc::default(),
            accounts: Accounts::Fake(Arc::clone(store)),
            try_outs: super::super::review_classifier::TryOuts::default(),
            reviews: std::sync::Arc::default(),
            runs: crate::review_classifier::runs::Runs::Live(None),
            builds: crate::builder::runs::Builds::Live(None),
            alerts: None,
        }
    }

    async fn put(
        store: &Arc<FakeStore>,
        agent: &str,
        label: &str,
        url: &str,
    ) -> Result<axum::Json<super::PutGitlabAccountResponse>, problem_details::ProblemDetails> {
        super::put_gitlab_account(
            axum::extract::State(state(store)),
            axum::extract::Path(("pr1ma".to_owned(), agent.to_owned(), label.to_owned())),
            axum::Json(request(url, "t0k3n")),
        )
        .await
    }

    #[tokio::test]
    async fn a_plain_account_reaches_the_store_under_the_agents_gitlab_directory() {
        let store = Arc::default();
        let stored = put(&store, "atlas", "work", "https://gitlab.example.org")
            .await
            .expect("nothing is stored");
        assert_eq!(stored.url, "https://gitlab.example.org");
        assert_eq!(store.written(), ["swarm/agents/atlas/gitlab/work"]);
    }

    /// Refused before the store, whatever the part: an address, a label or an
    /// agent that is not an identifier.
    #[tokio::test]
    async fn a_request_the_swarm_would_refuse_is_a_400_and_writes_nothing() {
        let store = Arc::default();
        for (agent, label, url) in [
            ("atlas", "work", "http://10.0.0.5:8080"),
            ("atlas", "work", "https://localhost"),
            ("atlas", "work", "https://gitlab.example.org/@x"),
            ("atlas", "work", "https://gitlab.example.org/?x=1"),
            ("atlas", "My_Label", "https://gitlab.example.org"),
            ("atlas", "../x", "https://gitlab.example.org"),
            ("../x", "work", "https://gitlab.example.org"),
        ] {
            let problem = put(&store, agent, label, url).await.expect_err("refused");
            assert_eq!(
                problem.status,
                Some(axum::http::StatusCode::BAD_REQUEST),
                "{agent} {label} {url}: {problem:?}"
            );
        }
        assert!(store.written().is_empty());
    }

    #[tokio::test]
    async fn linking_a_label_twice_is_a_409_and_the_first_account_stays() {
        use super::super::linked_accounts::{AccountStore, link};
        use swarm_secret_client::gitlab;

        let store = FakeStore::default();
        let path = gitlab::account_path("atlas", "work").expect("a valid path");
        let first = account(request("https://gitlab.example.org", "t0k3n-first")).expect("valid");
        let second = account(request("https://gitlab.lan", "t0k3n-second")).expect("valid");

        link(&store, &path, &first)
            .await
            .expect("nothing is stored");
        let problem = link(&store, &path, &second)
            .await
            .expect_err("an account is stored")
            .problem(&super::existing("atlas", "work"));

        assert_eq!(problem.status, Some(axum::http::StatusCode::CONFLICT));
        let detail = problem.detail.expect("a detail");
        assert!(
            detail.contains(r#"agent atlas already has gitlab account "work""#),
            "{detail}"
        );
        assert_eq!(store.written(), std::slice::from_ref(&path));
        let kept: gitlab::Account = store
            .read_optional(&path)
            .await
            .expect("store answers")
            .expect("still stored");
        assert_eq!(kept.url, "https://gitlab.example.org");
        assert_eq!(kept.value, "t0k3n-first");
    }
}