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)
  33 {
  34   for (auto* arg : args)
  35     app_args_.push_back(arg);
  36
  37   checker_side_ = std::make_unique<simgrid::mc::CheckerSide>(app_args_);
  38
  39   if (need_memory_introspection)
  40     initial_snapshot_ = std::make_shared<simgrid::mc::Snapshot>(0, page_store_, checker_side_->get_remote_memory());
  41 }
  42
  43 RemoteApp::~RemoteApp()
  44 {
  45   initial_snapshot_ = nullptr;
  46   shutdown();
  47 }
  48
  49 void RemoteApp::restore_initial_state()
  50 {
  51   if (initial_snapshot_ == nullptr) { // No memory introspection
  52     shutdown();
  53     checker_side_.reset(
  54         nullptr); // We need to destroy the existing CheckerSide before creating the new one, or libevent gets crazy
  55     checker_side_.reset(new simgrid::mc::CheckerSide(app_args_));
  56   } else
  57     initial_snapshot_->restore(checker_side_->get_remote_memory());
  58 }
  59
  60 unsigned long RemoteApp::get_maxpid() const
  61 {
  62   // note: we could maybe cache it and count the actor creation on checker side too.
  63   // But counting correctly accross state checkpoint/restore would be annoying.
  64
  65   checker_side_->get_channel().send(MessageType::ACTORS_MAXPID);
  66   s_mc_message_int_t answer;
  67   ssize_t answer_size = checker_side_->get_channel().receive(answer);
  68   xbt_assert(answer_size != -1, "Could not receive message");
  69   xbt_assert(answer_size == sizeof answer, "Broken message (size=%zd; expected %zu)", answer_size, sizeof answer);
  70   xbt_assert(answer.type == MessageType::ACTORS_MAXPID_REPLY,
  71              "Received unexpected message %s (%i); expected MessageType::ACTORS_MAXPID_REPLY (%i)",
  72              to_c_str(answer.type), (int)answer.type, (int)MessageType::ACTORS_MAXPID_REPLY);
  73
  74   return answer.value;
  75 }
  76
  77 void RemoteApp::get_actors_status(std::map<aid_t, ActorState>& whereto) const
  78 {
  79   // The messaging happens as follows:
  80   //
  81   // CheckerSide                  AppSide
  82   // send ACTORS_STATUS ---->
  83   //                    <----- send ACTORS_STATUS_REPLY
  84   //                    <----- send `N` `s_mc_message_actors_status_one_t` structs
  85   //                    <----- send `M` `s_mc_message_simcall_probe_one_t` structs
  86   checker_side_->get_channel().send(MessageType::ACTORS_STATUS);
  87
  88   s_mc_message_actors_status_answer_t answer;
  89   ssize_t answer_size = checker_side_->get_channel().receive(answer);
  90   xbt_assert(answer_size != -1, "Could not receive message");
  91   xbt_assert(answer_size == sizeof answer, "Broken message (size=%zd; expected %zu)", answer_size, sizeof answer);
  92   xbt_assert(answer.type == MessageType::ACTORS_STATUS_REPLY,
  93              "Received unexpected message %s (%i); expected MessageType::ACTORS_STATUS_REPLY (%i)",
  94              to_c_str(answer.type), (int)answer.type, (int)MessageType::ACTORS_STATUS_REPLY);
  95
  96   // Message sanity checks
  97   xbt_assert(answer.count >= 0, "Received an ACTOR_STATUS_REPLY message with an actor count of '%d' < 0", answer.count);
  98   xbt_assert(answer.transition_count >= 0, "Received an ACTOR_STATUS_REPLY message with transition_count '%d' < 0",
  99              answer.transition_count);
 100   xbt_assert(answer.transition_count == 0 || answer.count >= 0,
 101              "Received an ACTOR_STATUS_REPLY message with no actor data "
 102              "but with transition data nonetheless");
 103
 104   std::vector<s_mc_message_actors_status_one_t> status(answer.count);
 105   if (answer.count > 0) {
 106     size_t size      = status.size() * sizeof(s_mc_message_actors_status_one_t);
 107     ssize_t received = checker_side_->get_channel().receive(status.data(), size);
 108     xbt_assert(static_cast<size_t>(received) == size);
 109   }
 110
 111   // Ensures that each actor sends precisely `answer.transition_count` transitions. While technically
 112   // this doesn't catch the edge case where actor A sends 3 instead of 2 and actor B sends 2 instead
 113   // of 3 transitions, that is ignored here since that invariant needs to be enforced on the AppSide
 114   const auto expected_transitions = std::accumulate(
 115       status.begin(), status.end(), 0, [](int total, const auto& actor) { return total + actor.n_transitions; });
 116   xbt_assert(expected_transitions == answer.transition_count,
 117              "Expected to receive %d transition(s) but was only notified of %d by the app side", expected_transitions,
 118              answer.transition_count);
 119
 120   std::vector<s_mc_message_simcall_probe_one_t> probes(answer.transition_count);
 121   if (answer.transition_count > 0) {
 122     for (auto& probe : probes) {
 123       ssize_t received = checker_side_->get_channel().receive(probe);
 124       xbt_assert(received >= 0, "Could not receive response to ACTORS_PROBE message (%s)", strerror(errno));
 125       xbt_assert(static_cast<size_t>(received) == sizeof probe,
 126                  "Could not receive response to ACTORS_PROBE message (%zd bytes received != %zu bytes expected",
 127                  received, sizeof probe);
 128     }
 129   }
 130
 131   whereto.clear();
 132   std::move_iterator probes_iter(probes.begin());
 133
 134   for (const auto& actor : status) {
 135     xbt_assert(actor.n_transitions == 0 || actor.n_transitions == actor.max_considered,
 136                "If any transitions are serialized for an actor, it must match the "
 137                "total number of transitions that can be considered for the actor "
 138                "(currently %d), but only %d transition(s) was/were said to be encoded",
 139                actor.max_considered, actor.n_transitions);
 140
 141     std::vector<std::shared_ptr<Transition>> actor_transitions;
 142     for (int times_considered = 0; times_considered < actor.n_transitions; times_considered++, probes_iter++) {
 143       std::stringstream stream((*probes_iter).buffer.data());
 144       actor_transitions.emplace_back(deserialize_transition(actor.aid, times_considered, stream));
 145     }
 146
 147     XBT_DEBUG("Received %zu transitions for actor %ld", actor_transitions.size(), actor.aid);
 148     whereto.try_emplace(actor.aid, actor.aid, actor.enabled, actor.max_considered, std::move(actor_transitions));
 149   }
 150 }
 151
 152 void RemoteApp::check_deadlock() const
 153 {
 154   xbt_assert(checker_side_->get_channel().send(MessageType::DEADLOCK_CHECK) == 0, "Could not check deadlock state");
 155   s_mc_message_int_t message;
 156   ssize_t received = checker_side_->get_channel().receive(message);
 157   xbt_assert(received != -1, "Could not receive message");
 158   xbt_assert(received == sizeof message, "Broken message (size=%zd; expected %zu)", received, sizeof message);
 159   xbt_assert(message.type == MessageType::DEADLOCK_CHECK_REPLY,
 160              "Received unexpected message %s (%i); expected MessageType::DEADLOCK_CHECK_REPLY (%i)",
 161              to_c_str(message.type), (int)message.type, (int)MessageType::DEADLOCK_CHECK_REPLY);
 162
 163   if (message.value != 0) {
 164     auto* explo = Exploration::get_instance();
 165     XBT_CINFO(mc_global, "Counter-example execution trace:");
 166     for (auto const& frame : explo->get_textual_trace())
 167       XBT_CINFO(mc_global, "  %s", frame.c_str());
 168     XBT_INFO("You can debug the problem (and see the whole details) by rerunning out of simgrid-mc with "
 169              "--cfg=model-check/replay:'%s'",
 170              explo->get_record_trace().to_string().c_str());
 171     explo->log_state();
 172     throw DeadlockError();
 173   }
 174 }
 175
 176 void RemoteApp::wait_for_requests()
 177 {
 178   checker_side_->wait_for_requests();
 179 }
 180
 181 void RemoteApp::shutdown()
 182 {
 183   XBT_DEBUG("Shutting down model-checker");
 184
 185   if (checker_side_->running()) {
 186     XBT_DEBUG("Killing process");
 187     finalize_app(true);
 188     kill(checker_side_->get_pid(), SIGKILL);
 189     checker_side_->terminate();
 190   }
 191 }
 192
 193 Transition* RemoteApp::handle_simcall(aid_t aid, int times_considered, bool new_transition)
 194 {
 195   s_mc_message_simcall_execute_t m = {};
 196   m.type                           = MessageType::SIMCALL_EXECUTE;
 197   m.aid_                           = aid;
 198   m.times_considered_              = times_considered;
 199   checker_side_->get_channel().send(m);
 200
 201   get_remote_process_memory().clear_cache();
 202   if (checker_side_->running())
 203     checker_side_->dispatch_events(); // The app may send messages while processing the transition
 204
 205   s_mc_message_simcall_execute_answer_t answer;
 206   ssize_t s = checker_side_->get_channel().receive(answer);
 207   xbt_assert(s != -1, "Could not receive message");
 208   xbt_assert(s == sizeof answer, "Broken message (size=%zd; expected %zu)", s, sizeof answer);
 209   xbt_assert(answer.type == MessageType::SIMCALL_EXECUTE_ANSWER,
 210              "Received unexpected message %s (%i); expected MessageType::SIMCALL_EXECUTE_ANSWER (%i)",
 211              to_c_str(answer.type), (int)answer.type, (int)MessageType::SIMCALL_EXECUTE_ANSWER);
 212
 213   if (new_transition) {
 214     std::stringstream stream(answer.buffer.data());
 215     return deserialize_transition(aid, times_considered, stream);
 216   } else
 217     return nullptr;
 218 }
 219
 220 void RemoteApp::finalize_app(bool terminate_asap)
 221 {
 222   s_mc_message_int_t m = {};
 223   m.type               = MessageType::FINALIZE;
 224   m.value              = terminate_asap;
 225   xbt_assert(checker_side_->get_channel().send(m) == 0, "Could not ask the app to finalize on need");
 226
 227   s_mc_message_t answer;
 228   ssize_t s = checker_side_->get_channel().receive(answer);
 229   xbt_assert(s != -1, "Could not receive answer to FINALIZE");
 230   xbt_assert(s == sizeof answer, "Broken message (size=%zd; expected %zu)", s, sizeof answer);
 231   xbt_assert(answer.type == MessageType::FINALIZE_REPLY,
 232              "Received unexpected message %s (%i); expected MessageType::FINALIZE_REPLY (%i)", to_c_str(answer.type),
 233              (int)answer.type, (int)MessageType::FINALIZE_REPLY);
 234 }
 235
 236 } // namespace simgrid::mc