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 <limits.h>
  21 #include <memory>
  22 #include <numeric>
  23 #include <string>
  24 #include <sys/ptrace.h>
  25 #include <sys/un.h>
  26 #include <sys/wait.h>
  27
  28 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(mc_Session, mc, "Model-checker session");
  29 XBT_LOG_EXTERNAL_CATEGORY(mc_global);
  30
  31 namespace simgrid::mc {
  32
  33 static char master_socket_name[65] = {};
  34 static void cleanup_master_socket()
  35 {
  36   if (master_socket_name[0] != '\0')
  37     unlink(master_socket_name);
  38   master_socket_name[0] = '\0';
  39 }
  40
  41 RemoteApp::RemoteApp(const std::vector<char*>& args, bool need_memory_introspection) : app_args_(args)
  42 {
  43   if (need_memory_introspection) {
  44     checker_side_     = std::make_unique<simgrid::mc::CheckerSide>(app_args_, need_memory_introspection);
  45     initial_snapshot_ = std::make_shared<simgrid::mc::Snapshot>(0, page_store_, *checker_side_->get_remote_memory());
  46   } else {
  47     master_socket_ = socket(AF_LOCAL, SOCK_SEQPACKET | SOCK_CLOEXEC, 0);
  48     xbt_assert(master_socket_ != -1, "Cannot create the master socket: %s", strerror(errno));
  49
  50     struct sockaddr_un serv_addr = {};
  51     serv_addr.sun_family         = AF_LOCAL;
  52     snprintf(serv_addr.sun_path, 64, "/tmp/simgrid-mc-%d", getpid());
  53     strcpy(master_socket_name, serv_addr.sun_path);
  54     auto addr_size = offsetof(struct sockaddr_un, sun_path) + strlen(serv_addr.sun_path);
  55
  56     xbt_assert(bind(master_socket_, (struct sockaddr*)&serv_addr, addr_size) >= 0,
  57                "Cannot bind the master socket to %s: %s.", serv_addr.sun_path, strerror(errno));
  58     atexit(cleanup_master_socket);
  59
  60     xbt_assert(listen(master_socket_, SOMAXCONN) >= 0, "Cannot listen to the master socket: %s.", strerror(errno));
  61
  62     application_factory_ = std::make_unique<simgrid::mc::CheckerSide>(app_args_, need_memory_introspection);
  63     checker_side_        = application_factory_->clone(master_socket_);
  64   }
  65 }
  66
  67 RemoteApp::~RemoteApp()
  68 {
  69   initial_snapshot_ = nullptr;
  70   checker_side_     = nullptr;
  71 }
  72
  73 void RemoteApp::restore_initial_state()
  74 {
  75   if (initial_snapshot_ == nullptr) { // No memory introspection
  76     checker_side_ = application_factory_->clone(master_socket_);
  77   } else
  78     initial_snapshot_->restore(*checker_side_->get_remote_memory());
  79 }
  80
  81 unsigned long RemoteApp::get_maxpid() const
  82 {
  83   // note: we could maybe cache it and count the actor creation on checker side too.
  84   // But counting correctly accross state checkpoint/restore would be annoying.
  85
  86   checker_side_->get_channel().send(MessageType::ACTORS_MAXPID);
  87   s_mc_message_int_t answer;
  88   ssize_t answer_size = checker_side_->get_channel().receive(answer);
  89   xbt_assert(answer_size != -1, "Could not receive message");
  90   xbt_assert(answer_size == sizeof answer, "Broken message (size=%zd; expected %zu)", answer_size, sizeof answer);
  91   xbt_assert(answer.type == MessageType::ACTORS_MAXPID_REPLY,
  92              "Received unexpected message %s (%i); expected MessageType::ACTORS_MAXPID_REPLY (%i)",
  93              to_c_str(answer.type), (int)answer.type, (int)MessageType::ACTORS_MAXPID_REPLY);
  94
  95   return answer.value;
  96 }
  97
  98 void RemoteApp::get_actors_status(std::map<aid_t, ActorState>& whereto) const
  99 {
 100   // The messaging happens as follows:
 101   //
 102   // CheckerSide                  AppSide
 103   // send ACTORS_STATUS ---->
 104   //                    <----- send ACTORS_STATUS_REPLY_COUNT
 105   //                    <----- send `N` ACTORS_STATUS_REPLY_TRANSITION (s_mc_message_actors_status_one_t)
 106   //                    <----- send `M` ACTORS_STATUS_REPLY_SIMCALL (s_mc_message_simcall_probe_one_t)
 107   checker_side_->get_channel().send(MessageType::ACTORS_STATUS);
 108
 109   s_mc_message_actors_status_answer_t answer;
 110   ssize_t answer_size = checker_side_->get_channel().receive(answer);
 111   xbt_assert(answer_size != -1, "Could not receive message");
 112   xbt_assert(answer_size == sizeof answer, "Broken message (size=%zd; expected %zu)", answer_size, sizeof answer);
 113   xbt_assert(answer.type == MessageType::ACTORS_STATUS_REPLY_COUNT,
 114              "Received unexpected message %s (%i); expected MessageType::ACTORS_STATUS_REPLY_COUNT (%i)",
 115              to_c_str(answer.type), (int)answer.type, (int)MessageType::ACTORS_STATUS_REPLY_COUNT);
 116
 117   // Message sanity checks
 118   xbt_assert(answer.count >= 0, "Received an ACTORS_STATUS_REPLY_COUNT message with an actor count of '%d' < 0",
 119              answer.count);
 120
 121   std::vector<s_mc_message_actors_status_one_t> status(answer.count);
 122   if (answer.count > 0) {
 123     size_t size      = status.size() * sizeof(s_mc_message_actors_status_one_t);
 124     ssize_t received = checker_side_->get_channel().receive(status.data(), size);
 125     xbt_assert(static_cast<size_t>(received) == size);
 126   }
 127
 128   whereto.clear();
 129
 130   for (const auto& actor : status) {
 131     std::vector<std::shared_ptr<Transition>> actor_transitions;
 132     int n_transitions = actor.enabled ? actor.max_considered : 0;
 133     for (int times_considered = 0; times_considered < n_transitions; times_considered++) {
 134       s_mc_message_simcall_probe_one_t probe;
 135       ssize_t received = checker_side_->get_channel().receive(probe);
 136       xbt_assert(received >= 0, "Could not receive response to ACTORS_PROBE message (%s)", strerror(errno));
 137       xbt_assert(static_cast<size_t>(received) == sizeof probe,
 138                  "Could not receive response to ACTORS_PROBE message (%zd bytes received != %zu bytes expected",
 139                  received, sizeof probe);
 140
 141       std::stringstream stream(probe.buffer.data());
 142       actor_transitions.emplace_back(deserialize_transition(actor.aid, times_considered, stream));
 143     }
 144
 145     XBT_DEBUG("Received %zu transitions for actor %ld", actor_transitions.size(), actor.aid);
 146     whereto.try_emplace(actor.aid, actor.aid, actor.enabled, actor.max_considered, std::move(actor_transitions));
 147   }
 148 }
 149
 150 void RemoteApp::check_deadlock() const
 151 {
 152   xbt_assert(checker_side_->get_channel().send(MessageType::DEADLOCK_CHECK) == 0, "Could not check deadlock state");
 153   s_mc_message_int_t message;
 154   ssize_t received = checker_side_->get_channel().receive(message);
 155   xbt_assert(received != -1, "Could not receive message");
 156   xbt_assert(received == sizeof message, "Broken message (size=%zd; expected %zu)", received, sizeof message);
 157   xbt_assert(message.type == MessageType::DEADLOCK_CHECK_REPLY,
 158              "Received unexpected message %s (%i); expected MessageType::DEADLOCK_CHECK_REPLY (%i)",
 159              to_c_str(message.type), (int)message.type, (int)MessageType::DEADLOCK_CHECK_REPLY);
 160
 161   if (message.value != 0) {
 162     auto* explo = Exploration::get_instance();
 163     XBT_CINFO(mc_global, "Counter-example execution trace:");
 164     for (auto const& frame : explo->get_textual_trace())
 165       XBT_CINFO(mc_global, "  %s", frame.c_str());
 166     XBT_INFO("You can debug the problem (and see the whole details) by rerunning out of simgrid-mc with "
 167              "--cfg=model-check/replay:'%s'",
 168              explo->get_record_trace().to_string().c_str());
 169     explo->log_state();
 170     throw DeadlockError();
 171   }
 172 }
 173
 174 void RemoteApp::wait_for_requests()
 175 {
 176   checker_side_->wait_for_requests();
 177 }
 178
 179 Transition* RemoteApp::handle_simcall(aid_t aid, int times_considered, bool new_transition)
 180 {
 181   s_mc_message_simcall_execute_t m = {};
 182   m.type                           = MessageType::SIMCALL_EXECUTE;
 183   m.aid_                           = aid;
 184   m.times_considered_              = times_considered;
 185   checker_side_->get_channel().send(m);
 186
 187   if (auto* memory = get_remote_process_memory(); memory != nullptr)
 188     memory->clear_cache();
 189   if (checker_side_->running())
 190     checker_side_->dispatch_events(); // The app may send messages while processing the transition
 191
 192   s_mc_message_simcall_execute_answer_t answer;
 193   ssize_t s = checker_side_->get_channel().receive(answer);
 194   xbt_assert(s != -1, "Could not receive message");
 195   xbt_assert(s > 0 && answer.type == MessageType::SIMCALL_EXECUTE_REPLY,
 196              "%d Received unexpected message %s (%i); expected MessageType::SIMCALL_EXECUTE_REPLY (%i)", getpid(),
 197              to_c_str(answer.type), (int)answer.type, (int)MessageType::SIMCALL_EXECUTE_REPLY);
 198   xbt_assert(s == sizeof answer, "Broken message (size=%zd; expected %zu)", s, sizeof answer);
 199
 200   if (new_transition) {
 201     std::stringstream stream(answer.buffer.data());
 202     return deserialize_transition(aid, times_considered, stream);
 203   } else
 204     return nullptr;
 205 }
 206
 207 void RemoteApp::finalize_app(bool terminate_asap)
 208 {
 209   checker_side_->finalize(terminate_asap);
 210 }
 211
 212 } // namespace simgrid::mc