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