Contractless/src/startup/windows_service.rs

#[cfg(windows)]
use crate::common::network_paths_and_settings::block_extension_and_paths;
#[cfg(windows)]
use crate::env;
#[cfg(windows)]
use crate::log::{error, info, warn};
#[cfg(windows)]
use crate::startup::initialize_startup::prepare_pre_wallet_startup;
#[cfg(windows)]
use crate::startup::node_runtime::{
    initialize_node_logging, install_panic_cleanup, run_unlocked_node,
};
#[cfg(windows)]
use crate::startup::unlock_pipe::{pipe_name, run_unlock_pipe_server};
#[cfg(windows)]
use crate::startup::unlock_structs::ServiceWaitState;
#[cfg(windows)]
use crate::Arc;
#[cfg(windows)]
use crate::Duration;
#[cfg(windows)]
use crate::Error;
#[cfg(windows)]
use crate::Runtime;
#[cfg(windows)]
use crate::RwLock;
#[cfg(windows)]
use crate::{AtomicBool, AtomicOrdering};
#[cfg(windows)]
use std::ffi::OsString;
#[cfg(windows)]
use std::path::{Path, PathBuf};
#[cfg(windows)]
use std::sync::Mutex as StdMutex;
#[cfg(windows)]
use std::thread::sleep as thread_sleep;
#[cfg(windows)]
use std::time::{Duration as StdDuration, Instant};
#[cfg(windows)]
use tokio::sync::mpsc;
#[cfg(windows)]
use tokio::task::JoinHandle;
#[cfg(windows)]
use windows_service::service::{
    ServiceAccess, ServiceErrorControl, ServiceInfo, ServiceStartType, ServiceType,
};
#[cfg(windows)]
use windows_service::service::{
    ServiceControl, ServiceControlAccept, ServiceExitCode, ServiceState, ServiceStatus,
};
#[cfg(windows)]
use windows_service::service_control_handler::{self, ServiceControlHandlerResult};
#[cfg(windows)]
use windows_service::service_dispatcher;
#[cfg(windows)]
use windows_service::service_manager::{ServiceManager, ServiceManagerAccess};
#[cfg(windows)]
use windows_sys::Win32::Foundation::{ERROR_SERVICE_ALREADY_RUNNING, ERROR_SERVICE_DOES_NOT_EXIST};
#[cfg(windows)]
use winreg::enums::HKEY_LOCAL_MACHINE;
#[cfg(windows)]
use winreg::RegKey;

#[cfg(windows)]
const WINDOWS_SERVICE_CONTROLLER_CONNECT_ERROR: i32 = 1063;

#[cfg(windows)]
windows_service::define_windows_service!(ffi_service_main, contractless_service_main);

#[cfg(windows)]
fn service_name() -> &'static str {
    #[cfg(feature = "mainnet")]
    {
        "ContractlessMainnet"
    }
    #[cfg(feature = "testnet")]
    {
        "ContractlessTestnet"
    }
}

#[cfg(windows)]
fn service_display_name() -> &'static str {
    #[cfg(feature = "mainnet")]
    {
        "Contractless Mainnet"
    }
    #[cfg(feature = "testnet")]
    {
        "Contractless Testnet"
    }
}

#[cfg(windows)]
fn uninstall_registry_key_name() -> String {
    // The uninstall entry shares the service identity so Windows can present the
    // running node as a normal removable application in Add/Remove Programs.
    format!(
        r"Software\Microsoft\Windows\CurrentVersion\Uninstall\{}",
        service_name()
    )
}

#[cfg(windows)]
fn submit_key_binary_path(service_binary_path: &Path) -> PathBuf {
    // The helper binary is expected next to the service executable.
    service_binary_path
        .parent()
        .unwrap_or_else(|| Path::new("."))
        .join("contractless-submit-key.exe")
}

#[cfg(windows)]
fn quoted_command(path: &Path, args: &[&str]) -> String {
    // Windows registry uninstall commands need quoted paths because Program Files contains spaces.
    let mut command = format!("\"{}\"", path.display());
    for arg in args {
        command.push(' ');
        command.push_str(arg);
    }
    command
}

#[cfg(windows)]
fn register_add_remove_entry(service_binary_path: &Path) -> Result<(), Box<dyn Error>> {
    // The uninstall registry entry points back to the same binary so the install,
    // service management, and uninstall logic all stay in one maintained path.
    let hklm = RegKey::predef(HKEY_LOCAL_MACHINE);
    let uninstall_path = uninstall_registry_key_name();
    let (key, _) = hklm.create_subkey(&uninstall_path)?;

    let install_location = service_binary_path
        .parent()
        .unwrap_or_else(|| Path::new("."))
        .to_string_lossy()
        .to_string();
    let uninstall_command = quoted_command(service_binary_path, &["--uninstall-service"]);
    let submit_key_path = submit_key_binary_path(service_binary_path);

    key.set_value(
        "DisplayName",
        &format!("{} Service", service_display_name()),
    )?;
    key.set_value("DisplayVersion", &env!("CARGO_PKG_VERSION"))?;
    key.set_value("Publisher", &"Contractless")?;
    key.set_value("InstallLocation", &install_location)?;
    key.set_value(
        "DisplayIcon",
        &service_binary_path.to_string_lossy().to_string(),
    )?;
    key.set_value("UninstallString", &uninstall_command)?;
    key.set_value("QuietUninstallString", &uninstall_command)?;
    key.set_value(
        "Comments",
        &format!(
            "{} background node service. Use contractless-submit-key.exe to unlock the wallet after the service starts.",
            service_display_name()
        ),
    )?;
    key.set_value("NoModify", &1u32)?;
    key.set_value("NoRepair", &1u32)?;

    if submit_key_path.exists() {
        key.set_value(
            "HelpLink",
            &format!(
                "Submit wallet key with {} after starting the service.",
                submit_key_path.display()
            ),
        )?;
    }

    Ok(())
}

#[cfg(windows)]
fn unregister_add_remove_entry() -> Result<(), Box<dyn Error>> {
    // Removing an already-missing uninstall entry should not make service uninstall fail.
    let hklm = RegKey::predef(HKEY_LOCAL_MACHINE);
    let uninstall_path = uninstall_registry_key_name();

    match hklm.delete_subkey_all(&uninstall_path) {
        Ok(()) => Ok(()),
        Err(err) if err.kind() == std::io::ErrorKind::NotFound => Ok(()),
        Err(err) => Err(err.into()),
    }
}

#[cfg(windows)]
pub fn try_run_as_windows_service() -> Result<bool, Box<dyn Error>> {
    // Console launches fall through to the normal startup path. Only SCM launches
    // should enter the dedicated Windows service runtime.
    match service_dispatcher::start(service_name(), ffi_service_main) {
        Ok(()) => Ok(true),
        Err(windows_service::Error::Winapi(err))
            if err.raw_os_error() == Some(WINDOWS_SERVICE_CONTROLLER_CONNECT_ERROR) =>
        {
            Ok(false)
        }
        Err(err) => Err(err.into()),
    }
}

#[cfg(windows)]
pub fn handle_windows_service_command() -> Result<bool, Box<dyn Error>> {
    let args: Vec<String> = env::args().skip(1).collect();

    // Service install and lifecycle commands are handled before any wallet prompt
    // or node startup so administrative setup never mixes with normal runtime flow.
    if args.iter().any(|arg| arg == "--install-service") {
        println!("{}", install_windows_service()?);
        return Ok(true);
    }

    if args.iter().any(|arg| arg == "--uninstall-service") {
        println!("{}", uninstall_windows_service()?);
        return Ok(true);
    }

    if args.iter().any(|arg| arg == "--start-service") {
        println!("{}", start_windows_service()?);
        return Ok(true);
    }

    if args.iter().any(|arg| arg == "--stop-service") {
        println!("{}", stop_windows_service()?);
        return Ok(true);
    }

    Ok(false)
}

#[cfg(windows)]
fn install_windows_service() -> Result<String, Box<dyn Error>> {
    // Installing the service also registers the uninstall entry so Windows has a
    // standard removal path in addition to the direct CLI commands.
    let manager_access = ServiceManagerAccess::CONNECT | ServiceManagerAccess::CREATE_SERVICE;
    let service_manager = ServiceManager::local_computer(None::<&str>, manager_access)?;

    let service_binary_path = env::current_exe()?;
    let service_info = ServiceInfo {
        name: OsString::from(service_name()),
        display_name: OsString::from(service_display_name()),
        service_type: ServiceType::OWN_PROCESS,
        start_type: ServiceStartType::OnDemand,
        error_control: ServiceErrorControl::Normal,
        executable_path: service_binary_path.clone(),
        launch_arguments: vec![],
        dependencies: vec![],
        account_name: None,
        account_password: None,
    };

    let service_access = ServiceAccess::QUERY_CONFIG
        | ServiceAccess::CHANGE_CONFIG
        | ServiceAccess::START
        | ServiceAccess::DELETE;

    let service = match service_manager.create_service(&service_info, service_access) {
        Ok(service) => {
            let _ = service.set_description("Contractless blockchain node service");
            register_add_remove_entry(&service_binary_path)?;
            return Ok(format!("Windows service {} installed.", service_name()));
        }
        Err(_) => service_manager.open_service(service_name(), service_access)?,
    };

    let _ = service.set_description("Contractless blockchain node service");
    register_add_remove_entry(&service_binary_path)?;
    Ok(format!(
        "Windows service {} is already installed.",
        service_name()
    ))
}

#[cfg(windows)]
fn uninstall_windows_service() -> Result<String, Box<dyn Error>> {
    // Uninstall first stops the service if needed, then removes both the SCM entry
    // and the Add/Remove Programs registration.
    let service_manager =
        ServiceManager::local_computer(None::<&str>, ServiceManagerAccess::CONNECT)?;

    let service_access = ServiceAccess::QUERY_STATUS | ServiceAccess::STOP | ServiceAccess::DELETE;
    let service = match service_manager.open_service(service_name(), service_access) {
        Ok(service) => service,
        Err(windows_service::Error::Winapi(err))
            if err.raw_os_error() == Some(ERROR_SERVICE_DOES_NOT_EXIST as i32) =>
        {
            unregister_add_remove_entry()?;
            return Ok(format!(
                "Windows service {} is not installed.",
                service_name()
            ));
        }
        Err(err) => return Err(err.into()),
    };

    stop_service_if_running(&service)?;
    service.delete()?;
    drop(service);

    let start = Instant::now();
    let timeout = StdDuration::from_secs(5);
    while start.elapsed() < timeout {
        match service_manager.open_service(service_name(), ServiceAccess::QUERY_STATUS) {
            Err(windows_service::Error::Winapi(err))
                if err.raw_os_error() == Some(ERROR_SERVICE_DOES_NOT_EXIST as i32) =>
            {
                unregister_add_remove_entry()?;
                return Ok(format!("Windows service {} uninstalled.", service_name()));
            }
            Ok(service) => {
                drop(service);
            }
            Err(err) => return Err(err.into()),
        }

        thread_sleep(StdDuration::from_millis(250));
    }

    unregister_add_remove_entry()?;
    Ok(format!(
        "Windows service {} is marked for deletion and will finish uninstalling once Windows releases the open handles.",
        service_name()
    ))
}

#[cfg(windows)]
fn start_windows_service() -> Result<String, Box<dyn Error>> {
    // Start is a management command; the service itself still waits for a wallet key over the pipe.
    let service_manager =
        ServiceManager::local_computer(None::<&str>, ServiceManagerAccess::CONNECT)?;
    let service = service_manager.open_service(
        service_name(),
        ServiceAccess::START | ServiceAccess::QUERY_STATUS,
    )?;

    match service.start(&[] as &[OsString]) {
        Ok(()) => Ok(format!("Windows service {} started.", service_name())),
        Err(windows_service::Error::Winapi(err))
            if err.raw_os_error() == Some(ERROR_SERVICE_ALREADY_RUNNING as i32) =>
        {
            Ok(format!(
                "Windows service {} is already running.",
                service_name()
            ))
        }
        Err(err) => Err(err.into()),
    }
}

#[cfg(windows)]
fn stop_windows_service() -> Result<String, Box<dyn Error>> {
    // Stop requests go through SCM so service shutdown follows the normal control handler path.
    let service_manager =
        ServiceManager::local_computer(None::<&str>, ServiceManagerAccess::CONNECT)?;
    let service = match service_manager.open_service(
        service_name(),
        ServiceAccess::QUERY_STATUS | ServiceAccess::STOP,
    ) {
        Ok(service) => service,
        Err(windows_service::Error::Winapi(err))
            if err.raw_os_error() == Some(ERROR_SERVICE_DOES_NOT_EXIST as i32) =>
        {
            return Ok(format!(
                "Windows service {} is not installed.",
                service_name()
            ));
        }
        Err(err) => return Err(err.into()),
    };

    if stop_service_if_running(&service)? {
        Ok(format!(
            "Windows service {} stop requested.",
            service_name()
        ))
    } else {
        Ok(format!(
            "Windows service {} is already stopped.",
            service_name()
        ))
    }
}

#[cfg(windows)]
fn stop_service_if_running(
    service: &windows_service::service::Service,
) -> Result<bool, Box<dyn Error>> {
    // Service deletion and rebuilds both depend on the process actually exiting, so
    // the stop path waits until Windows reports the service as fully stopped.
    let status = service.query_status()?;
    if status.current_state == ServiceState::Stopped {
        return Ok(false);
    }

    if status.current_state != ServiceState::StopPending {
        let _ = service.stop()?;
    }

    let start = Instant::now();
    let timeout = StdDuration::from_secs(15);
    while start.elapsed() < timeout {
        let status = service.query_status()?;
        if status.current_state == ServiceState::Stopped {
            return Ok(true);
        }

        thread_sleep(StdDuration::from_millis(250));
    }

    Err(format!(
        "Timed out waiting for Windows service {} to stop.",
        service_name()
    )
    .into())
}

#[cfg(windows)]
fn contractless_service_main(_arguments: Vec<OsString>) {
    // SCM enters here instead of the normal console main path.
    if let Err(err) = run_service() {
        eprintln!("Windows service error: {err}");
    }
}

#[cfg(windows)]
fn run_service() -> windows_service::Result<()> {
    // The Windows service starts in a locked state, completes only the pre-wallet
    // startup work, and then waits for the helper tool to submit the wallet key.
    let shutdown = Arc::new(AtomicBool::new(false));
    let shutdown_handler = shutdown.clone();

    let event_handler = move |control_event| -> ServiceControlHandlerResult {
        match control_event {
            ServiceControl::Stop => {
                shutdown_handler.store(true, AtomicOrdering::SeqCst);
                ServiceControlHandlerResult::NoError
            }
            ServiceControl::Interrogate => ServiceControlHandlerResult::NoError,
            _ => ServiceControlHandlerResult::NotImplemented,
        }
    };

    // Register the SCM control handler before reporting StartPending.
    let status_handle = service_control_handler::register(service_name(), event_handler)?;

    // Tell SCM that initialization has begun but the service is not ready yet.
    status_handle.set_service_status(ServiceStatus {
        service_type: ServiceType::OWN_PROCESS,
        current_state: ServiceState::StartPending,
        controls_accepted: ServiceControlAccept::empty(),
        exit_code: ServiceExitCode::Win32(0),
        checkpoint: 1,
        wait_hint: Duration::from_secs(10),
        process_id: None,
    })?;

    // The service owns its own Tokio runtime because SCM entry is synchronous.
    let runtime = Runtime::new().expect("Failed to create Windows service runtime");
    let _log_handle = runtime.block_on(async {
        match initialize_node_logging().await {
            Ok(handle) => Some(handle),
            Err(err) => {
                eprintln!("Failed to initialize logging: {err}");
                None
            }
        }
    });
    install_panic_cleanup();

    let service_state = Arc::new(RwLock::new(ServiceWaitState::WaitingForUnlock));
    let service_state_for_pipe = service_state.clone();
    let shutdown_for_pipe = shutdown.clone();
    let (unlock_tx, mut unlock_rx) = mpsc::unbounded_channel::<String>();
    let unlocked_node_task: Arc<StdMutex<Option<JoinHandle<()>>>> = Arc::new(StdMutex::new(None));
    let unlocked_node_task_for_loop = unlocked_node_task.clone();

    runtime.block_on(async {
        // Run only the wallet-independent startup work before accepting unlock requests.
        prepare_pre_wallet_startup().await;

        // The named pipe stays alive while the service waits for a valid wallet key.
        tokio::spawn(async move {
            run_unlock_pipe_server(service_state_for_pipe, shutdown_for_pipe, unlock_tx).await;
        });
    });

    info!(
        "Windows service {} for {} is running and waiting for wallet unlock on {}.",
        service_name(),
        block_extension_and_paths().0,
        pipe_name()
    );

    // At this point the service is healthy, but still locked until the helper submits a key.
    status_handle.set_service_status(ServiceStatus {
        service_type: ServiceType::OWN_PROCESS,
        current_state: ServiceState::Running,
        controls_accepted: ServiceControlAccept::STOP,
        exit_code: ServiceExitCode::Win32(0),
        checkpoint: 0,
        wait_hint: Duration::default(),
        process_id: None,
    })?;

    let shutdown_for_service = shutdown.clone();
    runtime.block_on(async move {
        loop {
            tokio::select! {
                maybe_wallet_key = unlock_rx.recv() => {
                    if let Some(wallet_key) = maybe_wallet_key {
                        // Once the wallet key is accepted, the shared unlocked-node
                        // runtime is launched inside the service process itself.
                        let handle = tokio::spawn(async move {
                            if let Err(err) = run_unlocked_node(wallet_key, false).await {
                                error!("Unlocked Windows service node failed during startup: {err}");
                            }
                        });

                        if let Ok(mut task_slot) = unlocked_node_task_for_loop.lock() {
                            // Keep the handle so service shutdown can abort the node task cleanly.
                            *task_slot = Some(handle);
                        }
                    }
                }
                _ = crate::sleep(Duration::from_millis(500)) => {
                    if shutdown_for_service.load(AtomicOrdering::SeqCst) {
                        break;
                    }
                }
            }
        }
    });

    runtime.block_on(async {
        // Publish the stopping state so pipe status checks see shutdown in progress.
        let mut state = service_state.write().await;
        *state = ServiceWaitState::Stopping;
    });

    warn!("Windows service received stop request.");

    status_handle.set_service_status(ServiceStatus {
        service_type: ServiceType::OWN_PROCESS,
        current_state: ServiceState::StopPending,
        controls_accepted: ServiceControlAccept::empty(),
        exit_code: ServiceExitCode::Win32(0),
        checkpoint: 1,
        wait_hint: Duration::from_secs(10),
        process_id: None,
    })?;

    // Stop the unlocked node task before the runtime is torn down.
    if let Ok(mut task_slot) = unlocked_node_task.lock() {
        if let Some(handle) = task_slot.take() {
            handle.abort();
        }
    }

    // Shutting down the runtime releases the remaining async work so the service
    // process can exit cleanly and the binary can be rebuilt or uninstalled.
    runtime.shutdown_timeout(StdDuration::from_secs(2));

    info!("Windows service {} stopped.", service_name());

    status_handle.set_service_status(ServiceStatus {
        service_type: ServiceType::OWN_PROCESS,
        current_state: ServiceState::Stopped,
        controls_accepted: ServiceControlAccept::empty(),
        exit_code: ServiceExitCode::Win32(0),
        checkpoint: 0,
        wait_hint: Duration::default(),
        process_id: None,
    })?;

    Ok(())
}

#[cfg(not(windows))]
pub fn try_run_as_windows_service() -> Result<bool, Box<dyn std::error::Error>> {
    Ok(false)
}

#[cfg(not(windows))]
pub fn handle_windows_service_command() -> Result<bool, Box<dyn std::error::Error>> {
    Ok(false)
}