src/mc/api/RemoteApp.cpp

   1 /* Copyright (c) 2015-2023. The SimGrid Team. All rights reserved.          */
   2
   3 /* This program is free software; you can redistribute it and/or modify it
   4  * under the terms of the license (GNU LGPL) which comes with this package. */
   5
   6 #include "src/mc/api/RemoteApp.hpp"
   7 #include "src/mc/explo/Exploration.hpp"
   8 #include "src/mc/mc_config.hpp"
   9 #include "xbt/asserts.h"
  10 #include "src/mc/api/State.hpp"
  11 #include "src/mc/mc_config.hpp"
  12 #include "src/mc/mc_exit.hpp"
  13 #include "src/mc/mc_private.hpp"
  14 #include "xbt/log.h"
  15 #include "xbt/system_error.hpp"
  16 #include <signal.h>
  17
  18 #include <algorithm>
  19 #include <array>
  20 #include <memory>
  21 #include <numeric>
  22 #include <string>
  23
  24 #include <sys/ptrace.h>
  25 #include <sys/wait.h>
  26
  27 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(mc_Session, mc, "Model-checker session");
  28 XBT_LOG_EXTERNAL_CATEGORY(mc_global);
  29
  30 namespace simgrid::mc {
  31
  32 RemoteApp::RemoteApp(const std::vector<char*>& args, bool need_memory_introspection) : app_args_(args)
  33 {
  34   checker_side_ = std::make_unique<simgrid::mc::CheckerSide>(app_args_, need_memory_introspection);
  35
  36   if (need_memory_introspection)
  37     initial_snapshot_ = std::make_shared<simgrid::mc::Snapshot>(0, page_store_, *checker_side_->get_remote_memory());
  38 }
  39
  40 RemoteApp::~RemoteApp()
  41 {
  42   initial_snapshot_ = nullptr;
  43   checker_side_     = nullptr;
  44 }
  45
  46 void RemoteApp::restore_initial_state()
  47 {
  48   if (initial_snapshot_ == nullptr) { // No memory introspection
  49     // We need to destroy the existing CheckerSide before creating the new one, or libevent gets crazy
  50     checker_side_.reset(nullptr);
  51     checker_side_.reset(new simgrid::mc::CheckerSide(app_args_, false));
  52   } else
  53     initial_snapshot_->restore(*checker_side_->get_remote_memory());
  54 }
  55
  56 unsigned long RemoteApp::get_maxpid() const
  57 {
  58   // note: we could maybe cache it and count the actor creation on checker side too.
  59   // But counting correctly accross state checkpoint/restore would be annoying.
  60
  61   checker_side_->get_channel().send(MessageType::ACTORS_MAXPID);
  62   s_mc_message_int_t answer;
  63   ssize_t answer_size = checker_side_->get_channel().receive(answer);
  64   xbt_assert(answer_size != -1, "Could not receive message");
  65   xbt_assert(answer_size == sizeof answer, "Broken message (size=%zd; expected %zu)", answer_size, sizeof answer);
  66   xbt_assert(answer.type == MessageType::ACTORS_MAXPID_REPLY,
  67              "Received unexpected message %s (%i); expected MessageType::ACTORS_MAXPID_REPLY (%i)",
  68              to_c_str(answer.type), (int)answer.type, (int)MessageType::ACTORS_MAXPID_REPLY);
  69
  70   return answer.value;
  71 }
  72
  73 void RemoteApp::get_actors_status(std::map<aid_t, ActorState>& whereto) const
  74 {
  75   // The messaging happens as follows:
  76   //
  77   // CheckerSide                  AppSide
  78   // send ACTORS_STATUS ---->
  79   //                    <----- send ACTORS_STATUS_REPLY
  80   //                    <----- send `N` `s_mc_message_actors_status_one_t` structs
  81   //                    <----- send `M` `s_mc_message_simcall_probe_one_t` structs
  82   checker_side_->get_channel().send(MessageType::ACTORS_STATUS);
  83
  84   s_mc_message_actors_status_answer_t answer;
  85   ssize_t answer_size = checker_side_->get_channel().receive(answer);
  86   xbt_assert(answer_size != -1, "Could not receive message");
  87   xbt_assert(answer_size == sizeof answer, "Broken message (size=%zd; expected %zu)", answer_size, sizeof answer);
  88   xbt_assert(answer.type == MessageType::ACTORS_STATUS_REPLY,
  89              "Received unexpected message %s (%i); expected MessageType::ACTORS_STATUS_REPLY (%i)",
  90              to_c_str(answer.type), (int)answer.type, (int)MessageType::ACTORS_STATUS_REPLY);
  91
  92   // Message sanity checks
  93   xbt_assert(answer.count >= 0, "Received an ACTOR_STATUS_REPLY message with an actor count of '%d' < 0", answer.count);
  94
  95   std::vector<s_mc_message_actors_status_one_t> status(answer.count);
  96   if (answer.count > 0) {
  97     size_t size      = status.size() * sizeof(s_mc_message_actors_status_one_t);
  98     ssize_t received = checker_side_->get_channel().receive(status.data(), size);
  99     xbt_assert(static_cast<size_t>(received) == size);
 100   }
 101
 102   whereto.clear();
 103
 104   for (const auto& actor : status) {
 105     std::vector<std::shared_ptr<Transition>> actor_transitions;
 106     int n_transitions = actor.enabled ? actor.max_considered : 0;
 107     for (int times_considered = 0; times_considered < n_transitions; times_considered++) {
 108       s_mc_message_simcall_probe_one_t probe;
 109       ssize_t received = checker_side_->get_channel().receive(probe);
 110       xbt_assert(received >= 0, "Could not receive response to ACTORS_PROBE message (%s)", strerror(errno));
 111       xbt_assert(static_cast<size_t>(received) == sizeof probe,
 112                  "Could not receive response to ACTORS_PROBE message (%zd bytes received != %zu bytes expected",
 113                  received, sizeof probe);
 114
 115       std::stringstream stream(probe.buffer.data());
 116       actor_transitions.emplace_back(deserialize_transition(actor.aid, times_considered, stream));
 117     }
 118
 119     XBT_DEBUG("Received %zu transitions for actor %ld", actor_transitions.size(), actor.aid);
 120     whereto.try_emplace(actor.aid, actor.aid, actor.enabled, actor.max_considered, std::move(actor_transitions));
 121   }
 122 }
 123
 124 void RemoteApp::check_deadlock() const
 125 {
 126   xbt_assert(checker_side_->get_channel().send(MessageType::DEADLOCK_CHECK) == 0, "Could not check deadlock state");
 127   s_mc_message_int_t message;
 128   ssize_t received = checker_side_->get_channel().receive(message);
 129   xbt_assert(received != -1, "Could not receive message");
 130   xbt_assert(received == sizeof message, "Broken message (size=%zd; expected %zu)", received, sizeof message);
 131   xbt_assert(message.type == MessageType::DEADLOCK_CHECK_REPLY,
 132              "Received unexpected message %s (%i); expected MessageType::DEADLOCK_CHECK_REPLY (%i)",
 133              to_c_str(message.type), (int)message.type, (int)MessageType::DEADLOCK_CHECK_REPLY);
 134
 135   if (message.value != 0) {
 136     auto* explo = Exploration::get_instance();
 137     XBT_CINFO(mc_global, "Counter-example execution trace:");
 138     for (auto const& frame : explo->get_textual_trace())
 139       XBT_CINFO(mc_global, "  %s", frame.c_str());
 140     XBT_INFO("You can debug the problem (and see the whole details) by rerunning out of simgrid-mc with "
 141              "--cfg=model-check/replay:'%s'",
 142              explo->get_record_trace().to_string().c_str());
 143     explo->log_state();
 144     throw DeadlockError();
 145   }
 146 }
 147
 148 void RemoteApp::wait_for_requests()
 149 {
 150   checker_side_->wait_for_requests();
 151 }
 152
 153 Transition* RemoteApp::handle_simcall(aid_t aid, int times_considered, bool new_transition)
 154 {
 155   s_mc_message_simcall_execute_t m = {};
 156   m.type                           = MessageType::SIMCALL_EXECUTE;
 157   m.aid_                           = aid;
 158   m.times_considered_              = times_considered;
 159   checker_side_->get_channel().send(m);
 160
 161   if (auto* memory = get_remote_process_memory(); memory != nullptr)
 162     memory->clear_cache();
 163   if (checker_side_->running())
 164     checker_side_->dispatch_events(); // The app may send messages while processing the transition
 165
 166   s_mc_message_simcall_execute_answer_t answer;
 167   ssize_t s = checker_side_->get_channel().receive(answer);
 168   xbt_assert(s != -1, "Could not receive message");
 169   xbt_assert(s == sizeof answer, "Broken message (size=%zd; expected %zu)", s, sizeof answer);
 170   xbt_assert(answer.type == MessageType::SIMCALL_EXECUTE_ANSWER,
 171              "Received unexpected message %s (%i); expected MessageType::SIMCALL_EXECUTE_ANSWER (%i)",
 172              to_c_str(answer.type), (int)answer.type, (int)MessageType::SIMCALL_EXECUTE_ANSWER);
 173
 174   if (new_transition) {
 175     std::stringstream stream(answer.buffer.data());
 176     return deserialize_transition(aid, times_considered, stream);
 177   } else
 178     return nullptr;
 179 }
 180
 181 void RemoteApp::finalize_app(bool terminate_asap)
 182 {
 183   checker_side_->finalize(terminate_asap);
 184 }
 185
 186 } // namespace simgrid::mc