Logo AND Algorithmique Numérique Distribuée

Public GIT Repository
Merge branch 'master' of https://framagit.org/simgrid/simgrid
[simgrid.git] / src / mc / remote / AppSide.cpp
index 447f7db..a5d5ffa 100644 (file)
@@ -28,6 +28,8 @@
 #include <sys/ptrace.h>
 #include <sys/socket.h>
 #include <sys/types.h>
+#include <sys/un.h>
+#include <sys/wait.h>
 
 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(mc_client, mc, "MC client logic");
 XBT_LOG_EXTERNAL_CATEGORY(mc_global);
@@ -64,16 +66,19 @@ AppSide* AppSide::initialize()
   instance_ = std::make_unique<simgrid::mc::AppSide>(fd);
 
   // Wait for the model-checker:
-  errno = 0;
+  if (getenv("MC_NEED_PTRACE") != nullptr) {
+    errno = 0;
 #if defined __linux__
-  ptrace(PTRACE_TRACEME, 0, nullptr, nullptr);
+    ptrace(PTRACE_TRACEME, 0, nullptr, nullptr);
 #elif defined BSD
-  ptrace(PT_TRACE_ME, 0, nullptr, 0);
+    ptrace(PT_TRACE_ME, 0, nullptr, 0);
 #else
-#error "no ptrace equivalent coded for this platform"
+    xbt_die("no ptrace equivalent coded for this platform, please don't use the liveness checker here.");
 #endif
-  xbt_assert(errno == 0 && raise(SIGSTOP) == 0, "Could not wait for the model-checker (errno = %d: %s)", errno,
-             strerror(errno));
+
+    xbt_assert(errno == 0 && raise(SIGSTOP) == 0, "Could not wait for the model-checker (errno = %d: %s)", errno,
+               strerror(errno));
+  }
 
   instance_->handle_messages();
   return instance_.get();
@@ -111,7 +116,7 @@ void AppSide::handle_simcall_execute(const s_mc_message_simcall_execute_t* messa
 
   // Finish the RPC from the server: return a serialized observer, to build a Transition on Checker side
   s_mc_message_simcall_execute_answer_t answer = {};
-  answer.type = MessageType::SIMCALL_EXECUTE_ANSWER;
+  answer.type                                  = MessageType::SIMCALL_EXECUTE_REPLY;
   std::stringstream stream;
   if (actor->simcall_.observer_ != nullptr) {
     actor->simcall_.observer_->serialize(stream);
@@ -147,37 +152,68 @@ void AppSide::handle_finalize(const s_mc_message_int_t* msg) const
   if (terminate_asap)
     ::_Exit(0);
 }
-void AppSide::handle_initial_addresses()
+void AppSide::handle_fork(const s_mc_message_int_t* msg)
+{
+  int pid = fork();
+  xbt_assert(pid >= 0, "Could not fork application sub-process: %s.", strerror(errno));
+
+  if (pid == 0) { // Child
+    int sock = socket(AF_LOCAL, SOCK_SEQPACKET | SOCK_CLOEXEC, 0);
+
+    struct sockaddr_un addr = {};
+    addr.sun_family         = AF_LOCAL;
+    snprintf(addr.sun_path, 64, "/tmp/simgrid-mc-%lu", msg->value);
+    auto addr_size = offsetof(struct sockaddr_un, sun_path) + strlen(addr.sun_path);
+
+    xbt_assert(connect(sock, (struct sockaddr*)&addr, addr_size) >= 0,
+               "Cannot connect to Checker on /tmp/simgrid-mc-%lu: %s.", msg->value, strerror(errno));
+
+    channel_.reset_socket(sock);
+
+    s_mc_message_int_t answer = {};
+    answer.type               = MessageType::FORK_REPLY;
+    answer.value              = getpid();
+    xbt_assert(channel_.send(answer) == 0, "Could not send response to WAIT_CHILD_REPLY: %s", strerror(errno));
+  }
+}
+void AppSide::handle_wait_child(const s_mc_message_int_t* msg)
+{
+  int status;
+  errno = 0;
+  waitpid(msg->value, &status, 0);
+  xbt_assert(errno == 0, "Cannot wait on behalf of the checker: %s.", strerror(errno));
+
+  s_mc_message_int_t answer = {};
+  answer.type               = MessageType::WAIT_CHILD_REPLY;
+  answer.value              = status;
+  xbt_assert(channel_.send(answer) == 0, "Could not send response to WAIT_CHILD: %s", strerror(errno));
+}
+void AppSide::handle_need_meminfo()
 {
-  this->need_memory_info_                       = true;
-  s_mc_message_initial_addresses_reply_t answer = {};
-  answer.type                                   = MessageType::INITIAL_ADDRESSES_REPLY;
-  answer.mmalloc_default_mdp                    = mmalloc_get_current_heap();
-  xbt_assert(channel_.send(answer) == 0, "Could not send response with initial addresses.");
+  this->need_memory_info_                  = true;
+  s_mc_message_need_meminfo_reply_t answer = {};
+  answer.type                              = MessageType::NEED_MEMINFO_REPLY;
+  answer.mmalloc_default_mdp               = mmalloc_get_current_heap();
+  xbt_assert(channel_.send(answer) == 0, "Could not send response to the request for meminfo.");
 }
 void AppSide::handle_actors_status() const
 {
   auto const& actor_list = kernel::EngineImpl::get_instance()->get_actor_list();
-  const int num_actors   = actor_list.size();
-  XBT_DEBUG("Serialize the actors to answer ACTORS_STATUS from the checker. %d actors to go.", num_actors);
-
-  std::vector<s_mc_message_actors_status_one_t> status(num_actors);
-  int i                 = 0;
-  int total_transitions = 0;
+  XBT_DEBUG("Serialize the actors to answer ACTORS_STATUS from the checker. %zu actors to go.", actor_list.size());
 
+  std::vector<s_mc_message_actors_status_one_t> status;
   for (auto const& [aid, actor] : actor_list) {
-    status[i].aid            = aid;
-    status[i].enabled        = mc::actor_is_enabled(actor);
-    status[i].max_considered = actor->simcall_.observer_->get_max_consider();
-    status[i].n_transitions  = mc::actor_is_enabled(actor) ? status[i].max_considered : 0;
-    total_transitions += status[i].n_transitions;
-    i++;
+    s_mc_message_actors_status_one_t one = {};
+    one.type                             = MessageType::ACTORS_STATUS_REPLY_TRANSITION;
+    one.aid                              = aid;
+    one.enabled                          = mc::actor_is_enabled(actor);
+    one.max_considered                   = actor->simcall_.observer_->get_max_consider();
+    status.push_back(one);
   }
 
   struct s_mc_message_actors_status_answer_t answer = {};
-  answer.type             = MessageType::ACTORS_STATUS_REPLY;
-  answer.count            = num_actors;
-  answer.transition_count = total_transitions;
+  answer.type                                       = MessageType::ACTORS_STATUS_REPLY_COUNT;
+  answer.count                                      = static_cast<int>(status.size());
 
   xbt_assert(channel_.send(answer) == 0, "Could not send ACTORS_STATUS_REPLY msg");
   if (answer.count > 0) {
@@ -186,44 +222,39 @@ void AppSide::handle_actors_status() const
   }
 
   // Serialize each transition to describe what each actor is doing
-  if (total_transitions > 0) {
-    std::vector<s_mc_message_simcall_probe_one_t> probes(total_transitions);
-    auto probes_iter = probes.begin();
-
-    for (const auto& actor_status : status) {
-      if (not actor_status.enabled)
-        continue;
-
-      const auto& actor        = actor_list.at(actor_status.aid);
-      const int max_considered = actor_status.max_considered;
-
-      for (int times_considered = 0; times_considered < max_considered; times_considered++, probes_iter++) {
-        std::stringstream stream;
-        s_mc_message_simcall_probe_one_t& probe = *probes_iter;
-
-        if (actor->simcall_.observer_ != nullptr) {
-          actor->simcall_.observer_->prepare(times_considered);
-          actor->simcall_.observer_->serialize(stream);
-        } else {
-          stream << (short)mc::Transition::Type::UNKNOWN;
-        }
-
-        std::string str = stream.str();
-        xbt_assert(str.size() + 1 <= probe.buffer.size(),
-                   "The serialized transition is too large for the buffer. Please fix the code.");
-        strncpy(probe.buffer.data(), str.c_str(), probe.buffer.size() - 1);
-        probe.buffer.back() = '\0';
+  XBT_DEBUG("Deliver ACTOR_TRANSITION_PROBE payload");
+  for (const auto& actor_status : status) {
+    if (not actor_status.enabled)
+      continue;
+
+    const auto& actor        = actor_list.at(actor_status.aid);
+    const int max_considered = actor_status.max_considered;
+
+    for (int times_considered = 0; times_considered < max_considered; times_considered++) {
+      std::stringstream stream;
+      s_mc_message_simcall_probe_one_t probe;
+      probe.type = MessageType::ACTORS_STATUS_REPLY_SIMCALL;
+
+      if (actor->simcall_.observer_ != nullptr) {
+        actor->simcall_.observer_->prepare(times_considered);
+        actor->simcall_.observer_->serialize(stream);
+      } else {
+        stream << (short)mc::Transition::Type::UNKNOWN;
       }
-      // NOTE: We do NOT need to reset `times_considered` for each actor's
-      // simcall observer here to the "original" value (i.e. the value BEFORE
-      // multiple prepare() calls were made for serialization purposes) since
-      // each SIMCALL_EXECUTE provides a `times_considered` to be used to prepare
-      // the transition before execution.
-    }
-    XBT_DEBUG("Deliver ACTOR_TRANSITION_PROBE payload");
 
-    for (const auto& probe : probes)
+      std::string str = stream.str();
+      xbt_assert(str.size() + 1 <= probe.buffer.size(),
+                 "The serialized transition is too large for the buffer. Please fix the code.");
+      strncpy(probe.buffer.data(), str.c_str(), probe.buffer.size() - 1);
+      probe.buffer.back() = '\0';
+
       xbt_assert(channel_.send(probe) == 0, "Could not send ACTOR_TRANSITION_PROBE payload");
+    }
+    // NOTE: We do NOT need to reset `times_considered` for each actor's
+    // simcall observer here to the "original" value (i.e. the value BEFORE
+    // multiple prepare() calls were made for serialization purposes) since
+    // each SIMCALL_EXECUTE provides a `times_considered` to be used to prepare
+    // the transition before execution.
   }
 }
 void AppSide::handle_actors_maxpid() const
@@ -241,7 +272,7 @@ void AppSide::handle_actors_maxpid() const
 void AppSide::handle_messages()
 {
   while (true) { // Until we get a CONTINUE message
-    XBT_DEBUG("Waiting messages from model-checker");
+    XBT_DEBUG("Waiting messages from the model-checker");
 
     std::array<char, MC_MESSAGE_LENGTH> message_buffer;
     ssize_t received_size = channel_.receive(message_buffer.data(), message_buffer.size());
@@ -256,15 +287,15 @@ void AppSide::handle_messages()
 
     const s_mc_message_t* message = (s_mc_message_t*)message_buffer.data();
     switch (message->type) {
+      case MessageType::CONTINUE:
+        assert_msg_size("MESSAGE_CONTINUE", s_mc_message_t);
+        return;
+
       case MessageType::DEADLOCK_CHECK:
         assert_msg_size("DEADLOCK_CHECK", s_mc_message_t);
         handle_deadlock_check(message);
         break;
 
-      case MessageType::CONTINUE:
-        assert_msg_size("MESSAGE_CONTINUE", s_mc_message_t);
-        return;
-
       case MessageType::SIMCALL_EXECUTE:
         assert_msg_size("SIMCALL_EXECUTE", s_mc_message_simcall_execute_t);
         handle_simcall_execute((s_mc_message_simcall_execute_t*)message_buffer.data());
@@ -275,9 +306,19 @@ void AppSide::handle_messages()
         handle_finalize((s_mc_message_int_t*)message_buffer.data());
         break;
 
-      case MessageType::INITIAL_ADDRESSES:
-        assert_msg_size("INITIAL_ADDRESSES", s_mc_message_t);
-        handle_initial_addresses();
+      case MessageType::FORK:
+        assert_msg_size("FORK", s_mc_message_int_t);
+        handle_fork((s_mc_message_int_t*)message_buffer.data());
+        break;
+
+      case MessageType::WAIT_CHILD:
+        assert_msg_size("WAIT_CHILD", s_mc_message_int_t);
+        handle_wait_child((s_mc_message_int_t*)message_buffer.data());
+        break;
+
+      case MessageType::NEED_MEMINFO:
+        assert_msg_size("NEED_MEMINFO", s_mc_message_t);
+        handle_need_meminfo();
         break;
 
       case MessageType::ACTORS_STATUS: