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 void RemoteApp::restore_initial_state()
  78 {
  79   if (initial_snapshot_ == nullptr) // No memory introspection
  80     checker_side_ = application_factory_->clone(master_socket_);
  81 #if SIMGRID_HAVE_STATEFUL_MC
  82   else
  83     initial_snapshot_->restore(*checker_side_->get_remote_memory());
  84 #endif
  85 }
  86
  87 unsigned long RemoteApp::get_maxpid() const
  88 {
  89   // note: we could maybe cache it and count the actor creation on checker side too.
  90   // But counting correctly accross state checkpoint/restore would be annoying.
  91
  92   checker_side_->get_channel().send(MessageType::ACTORS_MAXPID);
  93   s_mc_message_int_t answer;
  94   ssize_t answer_size = checker_side_->get_channel().receive(answer);
  95   xbt_assert(answer_size != -1, "Could not receive message");
  96   xbt_assert(answer_size == sizeof answer, "Broken message (size=%zd; expected %zu)", answer_size, sizeof answer);
  97   xbt_assert(answer.type == MessageType::ACTORS_MAXPID_REPLY,
  98              "Received unexpected message %s (%i); expected MessageType::ACTORS_MAXPID_REPLY (%i)",
  99              to_c_str(answer.type), (int)answer.type, (int)MessageType::ACTORS_MAXPID_REPLY);
 100
 101   return answer.value;
 102 }
 103
 104 void RemoteApp::get_actors_status(std::map<aid_t, ActorState>& whereto) const
 105 {
 106   // The messaging happens as follows:
 107   //
 108   // CheckerSide                  AppSide
 109   // send ACTORS_STATUS ---->
 110   //                    <----- send ACTORS_STATUS_REPLY_COUNT
 111   //                    <----- send `N` ACTORS_STATUS_REPLY_TRANSITION (s_mc_message_actors_status_one_t)
 112   //                    <----- send `M` ACTORS_STATUS_REPLY_SIMCALL (s_mc_message_simcall_probe_one_t)
 113   //
 114   // Note that we also receive disabled transitions, because the guiding strategies need them to decide what could
 115   // unlock actors.
 116
 117   checker_side_->get_channel().send(MessageType::ACTORS_STATUS);
 118
 119   s_mc_message_actors_status_answer_t answer;
 120   ssize_t answer_size = checker_side_->get_channel().receive(answer);
 121   xbt_assert(answer_size != -1, "Could not receive message");
 122   xbt_assert(answer_size == sizeof answer, "Broken message (size=%zd; expected %zu)", answer_size, sizeof answer);
 123   xbt_assert(answer.type == MessageType::ACTORS_STATUS_REPLY_COUNT,
 124              "%d Received unexpected message %s (%i); expected MessageType::ACTORS_STATUS_REPLY_COUNT (%i)", getpid(),
 125              to_c_str(answer.type), (int)answer.type, (int)MessageType::ACTORS_STATUS_REPLY_COUNT);
 126
 127   // Message sanity checks
 128   xbt_assert(answer.count >= 0, "Received an ACTORS_STATUS_REPLY_COUNT message with an actor count of '%d' < 0",
 129              answer.count);
 130
 131   std::vector<s_mc_message_actors_status_one_t> status(answer.count);
 132   if (answer.count > 0) {
 133     size_t size      = status.size() * sizeof(s_mc_message_actors_status_one_t);
 134     ssize_t received = checker_side_->get_channel().receive(status.data(), size);
 135     xbt_assert(static_cast<size_t>(received) == size);
 136   }
 137
 138   whereto.clear();
 139
 140   for (const auto& actor : status) {
 141     std::vector<std::shared_ptr<Transition>> actor_transitions;
 142     int n_transitions = actor.max_considered;
 143     for (int times_considered = 0; times_considered < n_transitions; times_considered++) {
 144       s_mc_message_simcall_probe_one_t probe;
 145       ssize_t received = checker_side_->get_channel().receive(probe);
 146       xbt_assert(received >= 0, "Could not receive response to ACTORS_PROBE message (%s)", strerror(errno));
 147       xbt_assert(static_cast<size_t>(received) == sizeof probe,
 148                  "Could not receive response to ACTORS_PROBE message (%zd bytes received != %zu bytes expected",
 149                  received, sizeof probe);
 150
 151       std::stringstream stream(probe.buffer.data());
 152       actor_transitions.emplace_back(deserialize_transition(actor.aid, times_considered, stream));
 153     }
 154
 155     XBT_DEBUG("Received %zu transitions for actor %ld", actor_transitions.size(), actor.aid);
 156     whereto.try_emplace(actor.aid, actor.aid, actor.enabled, actor.max_considered, std::move(actor_transitions));
 157   }
 158 }
 159
 160 void RemoteApp::check_deadlock() const
 161 {
 162   xbt_assert(checker_side_->get_channel().send(MessageType::DEADLOCK_CHECK) == 0, "Could not check deadlock state");
 163   s_mc_message_int_t message;
 164   ssize_t received = checker_side_->get_channel().receive(message);
 165   xbt_assert(received != -1, "Could not receive message");
 166   xbt_assert(received == sizeof message, "Broken message (size=%zd; expected %zu)", received, sizeof message);
 167   xbt_assert(message.type == MessageType::DEADLOCK_CHECK_REPLY,
 168              "Received unexpected message %s (%i); expected MessageType::DEADLOCK_CHECK_REPLY (%i)",
 169              to_c_str(message.type), (int)message.type, (int)MessageType::DEADLOCK_CHECK_REPLY);
 170
 171   if (message.value != 0) {
 172     auto* explo = Exploration::get_instance();
 173     XBT_CINFO(mc_global, "Counter-example execution trace:");
 174     for (auto const& frame : explo->get_textual_trace())
 175       XBT_CINFO(mc_global, "  %s", frame.c_str());
 176     XBT_INFO("You can debug the problem (and see the whole details) by rerunning out of simgrid-mc with "
 177              "--cfg=model-check/replay:'%s'",
 178              explo->get_record_trace().to_string().c_str());
 179     explo->log_state();
 180     throw McError(ExitStatus::DEADLOCK);
 181   }
 182 }
 183
 184 void RemoteApp::wait_for_requests()
 185 {
 186   checker_side_->wait_for_requests();
 187 }
 188
 189 Transition* RemoteApp::handle_simcall(aid_t aid, int times_considered, bool new_transition)
 190 {
 191   s_mc_message_simcall_execute_t m = {};
 192   m.type                           = MessageType::SIMCALL_EXECUTE;
 193   m.aid_                           = aid;
 194   m.times_considered_              = times_considered;
 195   checker_side_->get_channel().send(m);
 196
 197 #if SIMGRID_HAVE_STATEFUL_MC
 198   if (auto* memory = get_remote_process_memory(); memory != nullptr)
 199     memory->clear_cache();
 200 #endif
 201   if (checker_side_->running())
 202     checker_side_->dispatch_events(); // The app may send messages while processing the transition
 203
 204   s_mc_message_simcall_execute_answer_t answer;
 205   ssize_t s = checker_side_->get_channel().receive(answer);
 206   xbt_assert(s != -1, "Could not receive message");
 207   xbt_assert(s > 0 && answer.type == MessageType::SIMCALL_EXECUTE_REPLY,
 208              "%d Received unexpected message %s (%i); expected MessageType::SIMCALL_EXECUTE_REPLY (%i)", getpid(),
 209              to_c_str(answer.type), (int)answer.type, (int)MessageType::SIMCALL_EXECUTE_REPLY);
 210   xbt_assert(s == sizeof answer, "Broken message (size=%zd; expected %zu)", s, sizeof answer);
 211
 212   if (new_transition) {
 213     std::stringstream stream(answer.buffer.data());
 214     return deserialize_transition(aid, times_considered, stream);
 215   } else
 216     return nullptr;
 217 }
 218
 219 void RemoteApp::finalize_app(bool terminate_asap)
 220 {
 221   checker_side_->finalize(terminate_asap);
 222 }
 223
 224 } // namespace simgrid::mc