Logo AND Algorithmique Numérique Distribuée

Public GIT Repository
Add new entry in Release_Notes.
[simgrid.git] / src / mc / api / RemoteApp.cpp
index fc496ad..dc6c7d0 100644 (file)
@@ -30,52 +30,50 @@ XBT_LOG_EXTERNAL_CATEGORY(mc_global);
 
 namespace simgrid::mc {
 
-static char master_socket_name[65] = {};
-static void cleanup_master_socket()
-{
-  if (master_socket_name[0] != '\0')
-    unlink(master_socket_name);
-  master_socket_name[0] = '\0';
-}
+static std::string master_socket_name;
 
-RemoteApp::RemoteApp(const std::vector<char*>& args, bool need_memory_introspection) : app_args_(args)
+RemoteApp::RemoteApp(const std::vector<char*>& args) : app_args_(args)
 {
-  if (need_memory_introspection) {
-    checker_side_     = std::make_unique<simgrid::mc::CheckerSide>(app_args_, need_memory_introspection);
-    initial_snapshot_ = std::make_shared<simgrid::mc::Snapshot>(0, page_store_, *checker_side_->get_remote_memory());
-  } else {
-    master_socket_ = socket(AF_LOCAL, SOCK_SEQPACKET | SOCK_CLOEXEC, 0);
+  master_socket_ = socket(AF_UNIX,
+#ifdef __APPLE__
+                          SOCK_STREAM, /* Mac OSX does not have AF_UNIX + SOCK_SEQPACKET, even if that's faster */
+#else
+                          SOCK_SEQPACKET,
+#endif
+                          0);
     xbt_assert(master_socket_ != -1, "Cannot create the master socket: %s", strerror(errno));
 
+    master_socket_name = "/tmp/simgrid-mc-" + std::to_string(getpid());
+    master_socket_name.resize(MC_SOCKET_NAME_LEN); // truncate socket name if it's too long
+    master_socket_name.back() = '\0';              // ensure the data are null-terminated
+#ifdef __linux__
+    master_socket_name[0] = '\0'; // abstract socket, automatically removed after close
+#else
+    unlink(master_socket_name.c_str()); // remove possible stale socket before bind
+    atexit([]() {
+      if (not master_socket_name.empty())
+        unlink(master_socket_name.c_str());
+      master_socket_name.clear();
+    });
+#endif
+
     struct sockaddr_un serv_addr = {};
-    serv_addr.sun_family         = AF_LOCAL;
-    snprintf(serv_addr.sun_path, 64, "/tmp/simgrid-mc-%d", getpid());
-    strcpy(master_socket_name, serv_addr.sun_path);
-    auto addr_size = offsetof(struct sockaddr_un, sun_path) + strlen(serv_addr.sun_path);
+    serv_addr.sun_family         = AF_UNIX;
+    master_socket_name.copy(serv_addr.sun_path, MC_SOCKET_NAME_LEN);
 
-    xbt_assert(bind(master_socket_, (struct sockaddr*)&serv_addr, addr_size) >= 0,
-               "Cannot bind the master socket to %s: %s.", serv_addr.sun_path, strerror(errno));
-    atexit(cleanup_master_socket);
+    xbt_assert(bind(master_socket_, (struct sockaddr*)&serv_addr, sizeof serv_addr) >= 0,
+               "Cannot bind the master socket to %c%s: %s.", (serv_addr.sun_path[0] ? serv_addr.sun_path[0] : '@'),
+               serv_addr.sun_path + 1, strerror(errno));
 
     xbt_assert(listen(master_socket_, SOMAXCONN) >= 0, "Cannot listen to the master socket: %s.", strerror(errno));
 
-    application_factory_ = std::make_unique<simgrid::mc::CheckerSide>(app_args_, need_memory_introspection);
-    checker_side_        = application_factory_->clone(master_socket_);
-  }
-}
-
-RemoteApp::~RemoteApp()
-{
-  initial_snapshot_ = nullptr;
-  checker_side_     = nullptr;
+    application_factory_ = std::make_unique<simgrid::mc::CheckerSide>(app_args_);
+    checker_side_        = application_factory_->clone(master_socket_, master_socket_name);
 }
 
 void RemoteApp::restore_initial_state()
 {
-  if (initial_snapshot_ == nullptr) { // No memory introspection
-    checker_side_ = application_factory_->clone(master_socket_);
-  } else
-    initial_snapshot_->restore(*checker_side_->get_remote_memory());
+    checker_side_ = application_factory_->clone(master_socket_, master_socket_name);
 }
 
 unsigned long RemoteApp::get_maxpid() const
@@ -104,6 +102,10 @@ void RemoteApp::get_actors_status(std::map<aid_t, ActorState>& whereto) const
   //                    <----- send ACTORS_STATUS_REPLY_COUNT
   //                    <----- send `N` ACTORS_STATUS_REPLY_TRANSITION (s_mc_message_actors_status_one_t)
   //                    <----- send `M` ACTORS_STATUS_REPLY_SIMCALL (s_mc_message_simcall_probe_one_t)
+  //
+  // Note that we also receive disabled transitions, because the guiding strategies need them to decide what could
+  // unlock actors.
+
   checker_side_->get_channel().send(MessageType::ACTORS_STATUS);
 
   s_mc_message_actors_status_answer_t answer;
@@ -111,7 +113,7 @@ void RemoteApp::get_actors_status(std::map<aid_t, ActorState>& whereto) const
   xbt_assert(answer_size != -1, "Could not receive message");
   xbt_assert(answer_size == sizeof answer, "Broken message (size=%zd; expected %zu)", answer_size, sizeof answer);
   xbt_assert(answer.type == MessageType::ACTORS_STATUS_REPLY_COUNT,
-             "Received unexpected message %s (%i); expected MessageType::ACTORS_STATUS_REPLY_COUNT (%i)",
+             "%d Received unexpected message %s (%i); expected MessageType::ACTORS_STATUS_REPLY_COUNT (%i)", getpid(),
              to_c_str(answer.type), (int)answer.type, (int)MessageType::ACTORS_STATUS_REPLY_COUNT);
 
   // Message sanity checks
@@ -129,7 +131,7 @@ void RemoteApp::get_actors_status(std::map<aid_t, ActorState>& whereto) const
 
   for (const auto& actor : status) {
     std::vector<std::shared_ptr<Transition>> actor_transitions;
-    int n_transitions = actor.enabled ? actor.max_considered : 0;
+    int n_transitions = actor.max_considered;
     for (int times_considered = 0; times_considered < n_transitions; times_considered++) {
       s_mc_message_simcall_probe_one_t probe;
       ssize_t received = checker_side_->get_channel().receive(probe);
@@ -142,7 +144,8 @@ void RemoteApp::get_actors_status(std::map<aid_t, ActorState>& whereto) const
       actor_transitions.emplace_back(deserialize_transition(actor.aid, times_considered, stream));
     }
 
-    XBT_DEBUG("Received %zu transitions for actor %ld", actor_transitions.size(), actor.aid);
+    XBT_DEBUG("Received %zu transitions for actor %ld. The first one is %s", actor_transitions.size(), actor.aid,
+              (actor_transitions.size() > 0 ? actor_transitions[0]->to_string().c_str() : "null"));
     whereto.try_emplace(actor.aid, actor.aid, actor.enabled, actor.max_considered, std::move(actor_transitions));
   }
 }
@@ -167,7 +170,7 @@ void RemoteApp::check_deadlock() const
              "--cfg=model-check/replay:'%s'",
              explo->get_record_trace().to_string().c_str());
     explo->log_state();
-    throw DeadlockError();
+    throw McError(ExitStatus::DEADLOCK);
   }
 }
 
@@ -184,8 +187,6 @@ Transition* RemoteApp::handle_simcall(aid_t aid, int times_considered, bool new_
   m.times_considered_              = times_considered;
   checker_side_->get_channel().send(m);
 
-  if (auto* memory = get_remote_process_memory(); memory != nullptr)
-    memory->clear_cache();
   if (checker_side_->running())
     checker_side_->dispatch_events(); // The app may send messages while processing the transition