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 std::string master_socket_name;
  34
  35 RemoteApp::RemoteApp(const std::vector<char*>& args, bool need_memory_introspection) : app_args_(args)
  36 {
  37   if (need_memory_introspection) {
  38 #if SIMGRID_HAVE_STATEFUL_MC
  39     checker_side_     = std::make_unique<simgrid::mc::CheckerSide>(app_args_, need_memory_introspection);
  40     initial_snapshot_ = std::make_shared<simgrid::mc::Snapshot>(0, page_store_, *checker_side_->get_remote_memory());
  41 #else
  42     xbt_die("SimGrid MC was compiled without memory introspection support.");
  43 #endif
  44   } else {
  45     master_socket_ = socket(AF_UNIX,
  46 #ifdef __APPLE__
  47                             SOCK_STREAM, /* Mac OSX does not have AF_UNIX + SOCK_SEQPACKET, even if that's faster */
  48 #else
  49                             SOCK_SEQPACKET,
  50 #endif
  51                             0);
  52     xbt_assert(master_socket_ != -1, "Cannot create the master socket: %s", strerror(errno));
  53
  54     master_socket_name = "/tmp/simgrid-mc-" + std::to_string(getpid());
  55     master_socket_name.resize(MC_SOCKET_NAME_LEN); // truncate socket name if it's too long
  56     master_socket_name.back() = '\0';              // ensure the data are null-terminated
  57 #ifdef __linux__
  58     master_socket_name[0] = '\0'; // abstract socket, automatically removed after close
  59 #else
  60     unlink(master_socket_name.c_str()); // remove possible stale socket before bind
  61     atexit([]() {
  62       if (not master_socket_name.empty())
  63         unlink(master_socket_name.c_str());
  64       master_socket_name.clear();
  65     });
  66 #endif
  67
  68     struct sockaddr_un serv_addr = {};
  69     serv_addr.sun_family         = AF_UNIX;
  70     master_socket_name.copy(serv_addr.sun_path, MC_SOCKET_NAME_LEN);
  71
  72     xbt_assert(bind(master_socket_, (struct sockaddr*)&serv_addr, sizeof serv_addr) >= 0,
  73                "Cannot bind the master socket to %c%s: %s.", (serv_addr.sun_path[0] ? serv_addr.sun_path[0] : '@'),
  74                serv_addr.sun_path + 1, strerror(errno));
  75
  76     xbt_assert(listen(master_socket_, SOMAXCONN) >= 0, "Cannot listen to the master socket: %s.", strerror(errno));
  77
  78     application_factory_ = std::make_unique<simgrid::mc::CheckerSide>(app_args_, need_memory_introspection);
  79     checker_side_        = application_factory_->clone(master_socket_, master_socket_name);
  80   }
  81 }
  82
  83 void RemoteApp::restore_initial_state()
  84 {
  85   if (initial_snapshot_ == nullptr) // No memory introspection
  86     checker_side_ = application_factory_->clone(master_socket_, master_socket_name);
  87 #if SIMGRID_HAVE_STATEFUL_MC
  88   else
  89     initial_snapshot_->restore(*checker_side_->get_remote_memory());
  90 #endif
  91 }
  92
  93 unsigned long RemoteApp::get_maxpid() const
  94 {
  95   // note: we could maybe cache it and count the actor creation on checker side too.
  96   // But counting correctly accross state checkpoint/restore would be annoying.
  97
  98   checker_side_->get_channel().send(MessageType::ACTORS_MAXPID);
  99   s_mc_message_int_t answer;
 100   ssize_t answer_size = checker_side_->get_channel().receive(answer);
 101   xbt_assert(answer_size != -1, "Could not receive message");
 102   xbt_assert(answer_size == sizeof answer, "Broken message (size=%zd; expected %zu)", answer_size, sizeof answer);
 103   xbt_assert(answer.type == MessageType::ACTORS_MAXPID_REPLY,
 104              "Received unexpected message %s (%i); expected MessageType::ACTORS_MAXPID_REPLY (%i)",
 105              to_c_str(answer.type), (int)answer.type, (int)MessageType::ACTORS_MAXPID_REPLY);
 106
 107   return answer.value;
 108 }
 109
 110 void RemoteApp::get_actors_status(std::map<aid_t, ActorState>& whereto) const
 111 {
 112   // The messaging happens as follows:
 113   //
 114   // CheckerSide                  AppSide
 115   // send ACTORS_STATUS ---->
 116   //                    <----- send ACTORS_STATUS_REPLY_COUNT
 117   //                    <----- send `N` ACTORS_STATUS_REPLY_TRANSITION (s_mc_message_actors_status_one_t)
 118   //                    <----- send `M` ACTORS_STATUS_REPLY_SIMCALL (s_mc_message_simcall_probe_one_t)
 119   //
 120   // Note that we also receive disabled transitions, because the guiding strategies need them to decide what could
 121   // unlock actors.
 122
 123   checker_side_->get_channel().send(MessageType::ACTORS_STATUS);
 124
 125   s_mc_message_actors_status_answer_t answer;
 126   ssize_t answer_size = checker_side_->get_channel().receive(answer);
 127   xbt_assert(answer_size != -1, "Could not receive message");
 128   xbt_assert(answer_size == sizeof answer, "Broken message (size=%zd; expected %zu)", answer_size, sizeof answer);
 129   xbt_assert(answer.type == MessageType::ACTORS_STATUS_REPLY_COUNT,
 130              "%d Received unexpected message %s (%i); expected MessageType::ACTORS_STATUS_REPLY_COUNT (%i)", getpid(),
 131              to_c_str(answer.type), (int)answer.type, (int)MessageType::ACTORS_STATUS_REPLY_COUNT);
 132
 133   // Message sanity checks
 134   xbt_assert(answer.count >= 0, "Received an ACTORS_STATUS_REPLY_COUNT message with an actor count of '%d' < 0",
 135              answer.count);
 136
 137   std::vector<s_mc_message_actors_status_one_t> status(answer.count);
 138   if (answer.count > 0) {
 139     size_t size      = status.size() * sizeof(s_mc_message_actors_status_one_t);
 140     ssize_t received = checker_side_->get_channel().receive(status.data(), size);
 141     xbt_assert(static_cast<size_t>(received) == size);
 142   }
 143
 144   whereto.clear();
 145
 146   for (const auto& actor : status) {
 147     std::vector<std::shared_ptr<Transition>> actor_transitions;
 148     int n_transitions = actor.max_considered;
 149     for (int times_considered = 0; times_considered < n_transitions; times_considered++) {
 150       s_mc_message_simcall_probe_one_t probe;
 151       ssize_t received = checker_side_->get_channel().receive(probe);
 152       xbt_assert(received >= 0, "Could not receive response to ACTORS_PROBE message (%s)", strerror(errno));
 153       xbt_assert(static_cast<size_t>(received) == sizeof probe,
 154                  "Could not receive response to ACTORS_PROBE message (%zd bytes received != %zu bytes expected",
 155                  received, sizeof probe);
 156
 157       std::stringstream stream(probe.buffer.data());
 158       actor_transitions.emplace_back(deserialize_transition(actor.aid, times_considered, stream));
 159     }
 160
 161     XBT_DEBUG("Received %zu transitions for actor %ld. The first one is %s", actor_transitions.size(), actor.aid,
 162               (actor_transitions.size() > 0 ? actor_transitions[0]->to_string().c_str() : "null"));
 163     whereto.try_emplace(actor.aid, actor.aid, actor.enabled, actor.max_considered, std::move(actor_transitions));
 164   }
 165 }
 166
 167 void RemoteApp::check_deadlock() const
 168 {
 169   xbt_assert(checker_side_->get_channel().send(MessageType::DEADLOCK_CHECK) == 0, "Could not check deadlock state");
 170   s_mc_message_int_t message;
 171   ssize_t received = checker_side_->get_channel().receive(message);
 172   xbt_assert(received != -1, "Could not receive message");
 173   xbt_assert(received == sizeof message, "Broken message (size=%zd; expected %zu)", received, sizeof message);
 174   xbt_assert(message.type == MessageType::DEADLOCK_CHECK_REPLY,
 175              "Received unexpected message %s (%i); expected MessageType::DEADLOCK_CHECK_REPLY (%i)",
 176              to_c_str(message.type), (int)message.type, (int)MessageType::DEADLOCK_CHECK_REPLY);
 177
 178   if (message.value != 0) {
 179     auto* explo = Exploration::get_instance();
 180     XBT_CINFO(mc_global, "Counter-example execution trace:");
 181     for (auto const& frame : explo->get_textual_trace())
 182       XBT_CINFO(mc_global, "  %s", frame.c_str());
 183     XBT_INFO("You can debug the problem (and see the whole details) by rerunning out of simgrid-mc with "
 184              "--cfg=model-check/replay:'%s'",
 185              explo->get_record_trace().to_string().c_str());
 186     explo->log_state();
 187     throw McError(ExitStatus::DEADLOCK);
 188   }
 189 }
 190
 191 void RemoteApp::wait_for_requests()
 192 {
 193   checker_side_->wait_for_requests();
 194 }
 195
 196 Transition* RemoteApp::handle_simcall(aid_t aid, int times_considered, bool new_transition)
 197 {
 198   s_mc_message_simcall_execute_t m = {};
 199   m.type                           = MessageType::SIMCALL_EXECUTE;
 200   m.aid_                           = aid;
 201   m.times_considered_              = times_considered;
 202   checker_side_->get_channel().send(m);
 203
 204 #if SIMGRID_HAVE_STATEFUL_MC
 205   if (auto* memory = get_remote_process_memory(); memory != nullptr)
 206     memory->clear_cache();
 207 #endif
 208   if (checker_side_->running())
 209     checker_side_->dispatch_events(); // The app may send messages while processing the transition
 210
 211   s_mc_message_simcall_execute_answer_t answer;
 212   ssize_t s = checker_side_->get_channel().receive(answer);
 213   xbt_assert(s != -1, "Could not receive message");
 214   xbt_assert(s > 0 && answer.type == MessageType::SIMCALL_EXECUTE_REPLY,
 215              "%d Received unexpected message %s (%i); expected MessageType::SIMCALL_EXECUTE_REPLY (%i)", getpid(),
 216              to_c_str(answer.type), (int)answer.type, (int)MessageType::SIMCALL_EXECUTE_REPLY);
 217   xbt_assert(s == sizeof answer, "Broken message (size=%zd; expected %zu)", s, sizeof answer);
 218
 219   if (new_transition) {
 220     std::stringstream stream(answer.buffer.data());
 221     return deserialize_transition(aid, times_considered, stream);
 222   } else
 223     return nullptr;
 224 }
 225
 226 void RemoteApp::finalize_app(bool terminate_asap)
 227 {
 228   checker_side_->finalize(terminate_asap);
 229 }
 230
 231 } // namespace simgrid::mc