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