boxlite-cli 0.10.8

//! End-to-end SSH control over a real `boxlite serve` and its guest.
//!
//! These tests need a VM: they run the real server as a child process and
//! verify what only the guest can prove — observation never boots a box,
//! reconfiguration advances the generation, an invalid configuration leaves
//! the listener running, disable is idempotent, and the startup restrictions
//! (explicit main command, AutoResume) survive the round trip.
//!
//! The no-VM boundaries — the public REST transport against a recording
//! stub, and the serve handlers behind real sockets — live in
//! `src/cli/src/commands/serve/tests/ssh.rs`, which the CI coverage pass
//! runs; this file it does not.

mod common;

use std::path::Path;
use std::process::{Command, Stdio};

use boxlite::{
    BoxOptions, BoxStatus, BoxliteError, BoxliteRestOptions, BoxliteRuntime, RootfsSpec,
    SshAccount, SshConfig, SshStatus,
};
use common::serve::ServeChild;

fn disabled() -> SshStatus {
    SshStatus {
        enabled: false,
        generation: 0,
        listen_address: String::new(),
        host_public_key: String::new(),
        host_key_fingerprint: String::new(),
    }
}

/// Remove the box even when an assertion panics. Declared after `ServeChild`,
/// so it drops first and the server is still available to service the removal.
struct BoxCleanup {
    url: String,
    api_key: Option<&'static str>,
    box_ids: Vec<String>,
}

impl Drop for BoxCleanup {
    fn drop(&mut self) {
        for box_id in self.box_ids.drain(..) {
            let _ = Command::new(env!("CARGO_BIN_EXE_boxlite"))
                .args(["--url", &self.url, "rm", "--force", &box_id])
                .envs(self.api_key.map(|k| ("BOXLITE_API_KEY", k)))
                .stdout(Stdio::null())
                .stderr(Stdio::null())
                .status();
        }
    }
}

fn generate_key(path: &Path) {
    let status = Command::new("ssh-keygen")
        .args(["-q", "-t", "ed25519", "-N", "", "-C", "ssh-rest-test", "-f"])
        .arg(path)
        .status()
        .expect("run ssh-keygen");
    assert!(status.success(), "ssh-keygen failed for {}", path.display());
}

fn local_key_config(keys: &Path) -> SshConfig {
    let host_key = keys.join("host");
    let user_key = keys.join("user");
    generate_key(&host_key);
    generate_key(&user_key);
    SshConfig {
        listen_address: "0.0.0.0:2222".into(),
        host_private_key: std::fs::read_to_string(&host_key).unwrap(),
        accounts: vec![SshAccount {
            login: "alice".into(),
            authorized_keys: vec![std::fs::read_to_string(user_key.with_extension("pub")).unwrap()],
            ca: None,
        }],
    }
}

/// The lifecycle the guide promises: observing never boots the box, every
/// configure restarts the listener (generation advances), an invalid
/// configuration leaves it running, disable is idempotent, and a stopped box
/// stays stopped through status and disable.
#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
async fn ssh_rest_journey() {
    let serve = ServeChild::start();
    let mut cleanup = BoxCleanup {
        url: serve.url(),
        api_key: None,
        box_ids: Vec::new(),
    };
    let keys = tempfile::tempdir().expect("key directory");
    let config = local_key_config(keys.path());

    let runtime = BoxliteRuntime::rest(BoxliteRestOptions::new(serve.url())).expect("rest runtime");
    let sandbox = runtime
        .create(
            BoxOptions {
                rootfs: RootfsSpec::Image("alpine:latest".into()),
                auto_delete: Some(0),
                ..Default::default()
            },
            None,
        )
        .await
        .expect("create box over REST");
    let box_id = sandbox.id().to_string();
    cleanup.box_ids.push(box_id.clone());

    // Creating the handle and observing SSH must not boot the box.
    assert_eq!(sandbox.ssh().status().await.unwrap(), disabled());
    let initial = sandbox.info().await.unwrap();
    assert_eq!(initial.status, BoxStatus::Configured);
    assert!(initial.pid.is_none());

    let first = sandbox
        .ssh()
        .configure(config.clone())
        .await
        .expect("configure");
    assert!(first.enabled);
    assert!(!first.listen_address.is_empty());
    assert!(!first.host_public_key.is_empty());
    assert!(!first.host_key_fingerprint.is_empty());

    // Every accepted configure restarts SSH; the host key is stable.
    let second = sandbox
        .ssh()
        .configure(config.clone())
        .await
        .expect("reconfigure");
    assert!(
        second.generation > first.generation,
        "repeated configure must advance the generation: {first:?} → {second:?}"
    );
    assert_eq!(second.host_key_fingerprint, first.host_key_fingerprint);

    // A semantically invalid configuration is rejected by the guest and
    // leaves the running listener untouched.
    let mut invalid = config;
    invalid.host_private_key = "not a private key".into();
    let error = sandbox
        .ssh()
        .configure(invalid)
        .await
        .expect_err("invalid configuration must be rejected");
    assert!(matches!(error, BoxliteError::InvalidArgument(_)), "{error}");
    let intact = sandbox
        .ssh()
        .status()
        .await
        .expect("status after rejection");
    assert!(intact.enabled);
    assert_eq!(intact.generation, second.generation);
    assert_eq!(intact.host_key_fingerprint, second.host_key_fingerprint);

    // Disable is idempotent.
    assert!(!sandbox.ssh().disable().await.unwrap().enabled);
    assert!(!sandbox.ssh().disable().await.unwrap().enabled);
    assert!(!sandbox.ssh().status().await.unwrap().enabled);

    // A stopped box is observable without being resumed.
    sandbox.stop().await.expect("stop box");
    let recovered = runtime
        .get(&box_id)
        .await
        .expect("get stopped box")
        .expect("stopped box still exists");
    assert_eq!(recovered.ssh().status().await.unwrap(), disabled());
    assert!(!recovered.ssh().disable().await.unwrap().enabled);
    let stopped = recovered.info().await.unwrap();
    assert_eq!(stopped.status, BoxStatus::Stopped);
    assert!(stopped.pid.is_none(), "observation must not restart the VM");

    let _ = runtime.remove(&box_id, true).await;
    serve.wait_until_no_boxes();
}

/// Authentication, missing boxes, and the sanitized body rejection, plus both
/// startup gates: an explicit main command and AutoResume disabled.
#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
async fn ssh_rest_auth_errors_and_startup_gates() {
    let serve = ServeChild::start_with(&["--api-key", "expected-key"], Some("expected-key"));
    let mut cleanup = BoxCleanup {
        url: serve.url(),
        api_key: Some("expected-key"),
        box_ids: Vec::new(),
    };
    let keys = tempfile::tempdir().expect("key directory");
    let config = local_key_config(keys.path());

    let options = |cmd: Option<Vec<String>>, auto_resume: bool| BoxOptions {
        rootfs: RootfsSpec::Image("alpine:latest".into()),
        cmd,
        auto_resume: Some(auto_resume),
        auto_delete: Some(0),
        ..Default::default()
    };
    let runtime =
        BoxliteRuntime::rest(BoxliteRestOptions::new(serve.url()).with_api_key("expected-key"))
            .expect("rest runtime");

    // The route is behind the key like every other box route.
    let http = reqwest::Client::new();
    let url = format!("{}/v1/boxes/x/ssh", serve.url());
    assert_eq!(http.get(&url).send().await.unwrap().status().as_u16(), 401);
    assert_eq!(
        http.get(&url)
            .bearer_auth("wrong-key")
            .send()
            .await
            .unwrap()
            .status()
            .as_u16(),
        401
    );

    // A malformed body is rejected with the fixed sanitized message before it
    // can reach anything that would echo it back.
    let box_for_body = runtime
        .create(options(None, true), None)
        .await
        .expect("create box");
    cleanup.box_ids.push(box_for_body.id().to_string());
    let body_url = format!("{}/v1/boxes/{}/ssh", serve.url(), box_for_body.id());
    let malformed = http
        .post(&body_url)
        .bearer_auth("expected-key")
        .header("content-type", "application/json")
        .body(r#"{"listen_address":"leak-me","host_private_key":"secret""#)
        .send()
        .await
        .unwrap();
    assert_eq!(malformed.status().as_u16(), 400);
    let body = malformed.text().await.unwrap();
    assert!(body.contains("invalid SSH configuration"), "{body}");
    assert!(
        !body.contains("leak-me") && !body.contains("secret"),
        "the rejection must not echo the body: {body}"
    );
    let wrong_type = http
        .post(&body_url)
        .bearer_auth("expected-key")
        .header("content-type", "text/plain")
        .body(r#"{"listen_address":"0.0.0.0:2222"}"#)
        .send()
        .await
        .unwrap();
    assert_eq!(wrong_type.status().as_u16(), 400);

    // A removed box reports not found through every SSH method. The handle is
    // taken before the removal, which is exactly how a stale client sees this.
    let stale_ssh = box_for_body.ssh();
    runtime
        .remove(box_for_body.id().as_ref(), true)
        .await
        .expect("remove box");
    let error = stale_ssh.status().await.unwrap_err();
    assert!(matches!(error, BoxliteError::NotFound(_)), "{error}");
    let error = stale_ssh.disable().await.unwrap_err();
    assert!(matches!(error, BoxliteError::NotFound(_)), "{error}");
    let error = stale_ssh.configure(config.clone()).await.unwrap_err();
    assert!(matches!(error, BoxliteError::NotFound(_)), "{error}");

    // An explicit main command keeps the startup gate: configure refuses
    // instead of booting the box.
    let explicit = runtime
        .create(
            options(Some(vec!["sleep".into(), "300".into()]), true),
            None,
        )
        .await
        .expect("create explicit-cmd box");
    cleanup.box_ids.push(explicit.id().to_string());
    let error = explicit
        .ssh()
        .configure(config.clone())
        .await
        .expect_err("explicit main command must refuse implicit start");
    assert!(matches!(error, BoxliteError::InvalidState(_)), "{error}");
    let info = explicit.info().await.unwrap();
    assert_eq!(info.status, BoxStatus::Configured);
    assert!(
        info.pid.is_none(),
        "the refused configure must not boot the VM"
    );

    // A stopped box with AutoResume disabled is refused for configure, but
    // status and disable still observe it without resuming.
    let frozen = runtime
        .create(options(None, false), None)
        .await
        .expect("create auto_resume=false box");
    cleanup.box_ids.push(frozen.id().to_string());
    frozen.start().await.expect("start box");
    frozen.stop().await.expect("stop box");
    let error = frozen
        .ssh()
        .configure(config.clone())
        .await
        .expect_err("stopped box with AutoResume disabled must be refused");
    assert!(error.to_string().contains("AutoResume disabled"), "{error}");
    assert_eq!(frozen.ssh().status().await.unwrap(), disabled());
    assert!(!frozen.ssh().disable().await.unwrap().enabled);
    let info = frozen.info().await.unwrap();
    assert_eq!(info.status, BoxStatus::Stopped);
    assert!(info.pid.is_none(), "observation must not resume the box");

    for box_id in &cleanup.box_ids {
        let _ = runtime.remove(box_id, true).await;
    }
    serve.wait_until_no_boxes();
}