system layer manager (llarp::sys::service_manager)

the win32 and sd_notify components provided a disjointed set of
similar high level functionality so we consolidate these duplicate
code paths into one that has the same lifecycle regardless of platform
to reduce complexity of this feature.

this new component is responsible for reporting state changes to the
system layer and optionally propagating state change to lokinet
requested by the system layer (used by windows service).

daemon/lokinet.cpp

@@ -7,7 +7,10 @@
 #include <llarp/util/str.hpp>
 
 #ifdef _WIN32
+#include <llarp/win32/service_manager.hpp>
 #include <dbghelp.h>
+#else
+#include <llarp/util/service_manager.hpp>
 #endif
 
 #include <csignal>
@@ -21,19 +24,18 @@ int
 lokinet_main(int, char**);
 
 #ifdef _WIN32
-#include <strsafe.h>
 extern "C" LONG FAR PASCAL
 win32_signal_handler(EXCEPTION_POINTERS*);
 extern "C" VOID FAR PASCAL
 win32_daemon_entry(DWORD, LPTSTR*);
-BOOL ReportSvcStatus(DWORD, DWORD, DWORD);
+
 VOID
 insert_description();
-SERVICE_STATUS SvcStatus;
-SERVICE_STATUS_HANDLE SvcStatusHandle;
-bool start_as_daemon = false;
+
 #endif
 
+bool run_as_daemon{false};
+
 static auto logcat = llarp::log::Cat("main");
 std::shared_ptr<llarp::Context> ctx;
 std::promise<int> exit_code;
@@ -84,9 +86,6 @@ install_win32_daemon()
     llarp::LogError("Cannot install service ", GetLastError());
     return;
   }
-  // just put the flag here. we eat it later on and specify the
-  // config path in the daemon entry point
-  StringCchCat(szPath.data(), 1024, " --win32-daemon");
 
   // Get a handle to the SCM database.
   schSCManager = OpenSCManager(
@@ -294,37 +293,6 @@ run_main_context(std::optional<fs::path> confFile, const llarp::RuntimeOptions o
 }
 
 #ifdef _WIN32
-void
-TellWindowsServiceStopped()
-{
-  ::WSACleanup();
-  if (not start_as_daemon)
-    return;
-
-  llarp::LogInfo("Telling Windows the service has stopped.");
-  if (not ReportSvcStatus(SERVICE_STOPPED, NO_ERROR, 0))
-  {
-    auto error_code = GetLastError();
-    if (error_code == ERROR_INVALID_DATA)
-      llarp::LogError(
-          "SetServiceStatus failed: \"The specified service status structure is invalid.\"");
-    else if (error_code == ERROR_INVALID_HANDLE)
-      llarp::LogError("SetServiceStatus failed: \"The specified handle is invalid.\"");
-    else
-      llarp::LogError("SetServiceStatus failed with an unknown error.");
-  }
-}
-
-class WindowsServiceStopped
-{
- public:
-  WindowsServiceStopped() = default;
-
-  ~WindowsServiceStopped()
-  {
-    TellWindowsServiceStopped();
-  }
-};
 
 /// minidump generation for windows jizz
 /// will make a coredump when there is an unhandled exception
@@ -370,9 +338,9 @@ main(int argc, char* argv[])
 #else
   SERVICE_TABLE_ENTRY DispatchTable[] = {
       {strdup("lokinet"), (LPSERVICE_MAIN_FUNCTION)win32_daemon_entry}, {NULL, NULL}};
-  if (lstrcmpi(argv[1], "--win32-daemon") == 0)
+  if (std::string{argv[1]} == "--win32-daemon")
   {
-    start_as_daemon = true;
+    run_as_daemon = true;
     StartServiceCtrlDispatcher(DispatchTable);
   }
   else
@@ -383,6 +351,10 @@ main(int argc, char* argv[])
 int
 lokinet_main(int argc, char** argv)
 {
+  // if we are not running as a service disable reporting
+  if (llarp::platform::is_windows and not run_as_daemon)
+    llarp::sys::service_manager->disable();
+
   if (auto result = Lokinet_INIT())
     return result;
 
@@ -398,7 +370,6 @@ lokinet_main(int argc, char** argv)
   opts.showBanner = false;
 
 #ifdef _WIN32
-  WindowsServiceStopped stopped_raii;
   if (startWinsock())
     return -1;
   SetConsoleCtrlHandler(handle_signal_win32, TRUE);
@@ -545,13 +516,9 @@ lokinet_main(int argc, char** argv)
   SetUnhandledExceptionFilter(&GenerateDump);
 #endif
 
-  std::thread main_thread{[&] { run_main_context(configFile, opts); }};
+  std::thread main_thread{[configFile, opts] { run_main_context(configFile, opts); }};
   auto ftr = exit_code.get_future();
 
-#ifdef _WIN32
-  ReportSvcStatus(SERVICE_RUNNING, NO_ERROR, 0);
-#endif
-
   do
   {
     // do periodic non lokinet related tasks here
@@ -582,9 +549,7 @@ lokinet_main(int argc, char** argv)
         llarp::log::critical(deadlock_cat, wtf);
         llarp::log::flush();
       }
-#ifdef _WIN32
-      TellWindowsServiceStopped();
-#endif
+      llarp::sys::service_manager->failed();
       std::abort();
     }
   } while (ftr.wait_for(std::chrono::seconds(1)) != std::future_status::ready);
@@ -609,6 +574,7 @@ lokinet_main(int argc, char** argv)
   }
 
   llarp::log::flush();
+  llarp::sys::service_manager->stopped();
   if (ctx)
   {
     ctx.reset();
@@ -617,29 +583,6 @@ lokinet_main(int argc, char** argv)
 }
 
 #ifdef _WIN32
-BOOL
-ReportSvcStatus(DWORD dwCurrentState, DWORD dwWin32ExitCode, DWORD dwWaitHint)
-{
-  static DWORD dwCheckPoint = 1;
-
-  // Fill in the SERVICE_STATUS structure.
-  SvcStatus.dwCurrentState = dwCurrentState;
-  SvcStatus.dwWin32ExitCode = dwWin32ExitCode;
-  SvcStatus.dwWaitHint = dwWaitHint;
-
-  if (dwCurrentState == SERVICE_START_PENDING)
-    SvcStatus.dwControlsAccepted = 0;
-  else
-    SvcStatus.dwControlsAccepted = SERVICE_ACCEPT_STOP;
-
-  if ((dwCurrentState == SERVICE_RUNNING) || (dwCurrentState == SERVICE_STOPPED))
-    SvcStatus.dwCheckPoint = 0;
-  else
-    SvcStatus.dwCheckPoint = dwCheckPoint++;
-
-  // Report the status of the service to the SCM.
-  return SetServiceStatus(SvcStatusHandle, &SvcStatus);
-}
 
 VOID FAR PASCAL
 SvcCtrlHandler(DWORD dwCtrl)
@@ -651,14 +594,13 @@ SvcCtrlHandler(DWORD dwCtrl)
     case SERVICE_CONTROL_STOP:
       // tell service we are stopping
       llarp::log::info(logcat, "Windows service controller gave SERVICE_CONTROL_STOP");
-      ReportSvcStatus(SERVICE_STOP_PENDING, NO_ERROR, 0);
-      // do the actual tear down
-      handle_signal(SIGINT);
+      llarp::sys::service_manager->system_changed_our_state(llarp::sys::ServiceState::Stopping);
       return;
 
     case SERVICE_CONTROL_INTERROGATE:
       // report status
-      SetServiceStatus(SvcStatusHandle, &SvcStatus);
+      llarp::log::debug(logcat, "Got win32 service interrogate signal");
+      llarp::sys::service_manager->report_changed_state();
       return;
 
     default:
@@ -673,21 +615,15 @@ VOID FAR PASCAL
 win32_daemon_entry(DWORD, LPTSTR* argv)
 {
   // Register the handler function for the service
-  SvcStatusHandle = RegisterServiceCtrlHandler("lokinet", SvcCtrlHandler);
+  auto* svc = dynamic_cast<llarp::sys::SVC_Manager*>(llarp::sys::service_manager);
+  svc->handle = RegisterServiceCtrlHandler("lokinet", SvcCtrlHandler);
 
-  if (!SvcStatusHandle)
+  if (svc->handle == nullptr)
   {
     llarp::LogError("failed to register daemon control handler");
     return;
   }
 
-  // These SERVICE_STATUS members remain as set here
-  SvcStatus.dwServiceType = SERVICE_WIN32_OWN_PROCESS;
-  SvcStatus.dwServiceSpecificExitCode = 0;
-
-  // Report initial status to the SCM
-  ReportSvcStatus(SERVICE_START_PENDING, NO_ERROR, 3000);
-
   // we hard code the args to lokinet_main.
   // we yoink argv[0] (lokinet.exe path) and pass in the new args.
   std::array args = {

include/llarp.hpp

@@ -45,6 +45,7 @@ namespace llarp
     std::shared_ptr<NodeDB> nodedb = nullptr;
     std::string nodedb_dir;
 
+    Context();
     virtual ~Context() = default;
 
     void

llarp/CMakeLists.txt

@@ -49,17 +49,23 @@ target_link_libraries(lokinet-platform PUBLIC lokinet-cryptography lokinet-util
 target_link_libraries(lokinet-platform PRIVATE oxenmq::oxenmq)
 
 if (ANDROID)
-  target_sources(lokinet-platform PRIVATE android/ifaddrs.c)
+  target_sources(lokinet-platform PRIVATE android/ifaddrs.c util/nop_service_manager.cpp)
 endif()
 
 if(CMAKE_SYSTEM_NAME MATCHES "Linux")
   target_sources(lokinet-platform PRIVATE linux/dbus.cpp)
+  if(WITH_SYSTEMD)
+    target_sources(lokinet-platform PRIVATE linux/sd_service_manager.cpp)
+  else()
+    target_sources(lokinet-platform PRIVATE util/nop_service_manager.cpp)
+  endif()
 endif()
 
 if (WIN32)
   target_sources(lokinet-platform PRIVATE
     net/win32.cpp
     vpn/win32.cpp
+    win32/service_manager.cpp
     win32/exec.cpp)
   add_library(lokinet-win32 STATIC
     win32/dll.cpp
@@ -312,6 +318,7 @@ endif()
 
 if(APPLE)
   add_subdirectory(apple)
+  target_sources(lokinet-platform PRIVATE util/nop_system_manager.cpp)
 endif()
 
 file(GLOB_RECURSE docs_SRC */*.hpp *.hpp)

llarp/context.cpp

@@ -12,6 +12,8 @@
 #include "service/context.hpp"
 #include "util/logging.hpp"
 
+#include <llarp/util/service_manager.hpp>
+
 #include <cxxopts.hpp>
 #include <csignal>
 #include <stdexcept>
@@ -213,4 +215,10 @@ namespace llarp
     loop.reset();
   }
 
+  Context::Context()
+  {
+    // service_manager is a global and context isnt
+    llarp::sys::service_manager->give_context(this);
+  }
+
 }  // namespace llarp

llarp/linux/sd_service_manager.cpp

@@ -0,0 +1,60 @@
+#include <llarp/util/service_manager.hpp>
+
+#include <systemd/sd-daemon.h>
+#include <cassert>
+#include <llarp.hpp>
+#include <llarp/router/router.hpp>
+#include <llarp/util/logging.hpp>
+
+namespace llarp::sys
+{
+  class SD_Manager : public I_SystemLayerManager
+  {
+    llarp::sys::ServiceState m_State{ServiceState::Initial};
+
+   public:
+    /// change our state and report it to the system layer
+    void
+    we_changed_our_state(ServiceState st) override
+    {
+      assert(m_State != st);
+      m_State = st;
+      report_changed_state();
+    }
+
+    void
+    report_changed_state() override
+    {
+      if (m_State == ServiceState::Running)
+      {
+        ::sd_notify(0, "READY=1");
+        return;
+      }
+      if (m_State == ServiceState::Stopping)
+      {
+        ::sd_notify(0, "STOPPING=1");
+        return;
+      }
+    }
+
+    void
+    report_periodic_stats() override
+    {
+      if (m_Context and m_Context->router and not m_disable)
+      {
+        auto status = fmt::format("WATCHDOG=1\nSTATUS={}", m_Context->router->status_line());
+        ::sd_notify(0, status.c_str());
+      }
+    }
+
+    void
+    system_changed_our_state(ServiceState) override
+    {
+      // not applicable on systemd
+    }
+  };
+
+  SD_Manager _manager{};
+  I_SystemLayerManager* const service_manager = &_manager;
+
+}  // namespace llarp::sys

llarp/router/abstractrouter.hpp

@@ -361,6 +361,9 @@ namespace llarp
     virtual void
     GossipRCIfNeeded(const RouterContact rc) = 0;
 
+    virtual std::string
+    status_line() = 0;
+
     /// Templated convenience function to generate a RouterHive event and
     /// delegate to non-templated (and overridable) function for handling.
     template <class EventType, class... Params>