1 /* Copyright (c) 2015-2023. The SimGrid Team. All rights reserved. */
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. */
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"
15 #include "xbt/system_error.hpp"
24 #include <sys/ptrace.h>
28 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(mc_Session, mc, "Model-checker session");
29 XBT_LOG_EXTERNAL_CATEGORY(mc_global);
31 namespace simgrid::mc {
33 static std::string master_socket_name;
35 RemoteApp::RemoteApp(const std::vector<char*>& args, bool need_memory_introspection) : app_args_(args)
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());
42 xbt_die("SimGrid MC was compiled without memory introspection support.");
45 master_socket_ = socket(AF_UNIX,
47 SOCK_STREAM, /* Mac OSX does not have AF_UNIX + SOCK_SEQPACKET, even if that's faster */
52 xbt_assert(master_socket_ != -1, "Cannot create the master socket: %s", strerror(errno));
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
58 master_socket_name[0] = '\0'; // abstract socket, automatically removed after close
60 unlink(master_socket_name.c_str()); // remove possible stale socket before bind
62 if (not master_socket_name.empty())
63 unlink(master_socket_name.c_str());
64 master_socket_name.clear();
69 struct sockaddr_un serv_addr = {};
70 serv_addr.sun_family = AF_UNIX;
71 master_socket_name.copy(serv_addr.sun_path, MC_SOCKET_NAME_LEN);
73 xbt_assert(bind(master_socket_, (struct sockaddr*)&serv_addr, sizeof serv_addr) >= 0,
74 "Cannot bind the master socket to %c%s: %s.", (serv_addr.sun_path[0] ? serv_addr.sun_path[0] : '@'),
75 serv_addr.sun_path + 1, strerror(errno));
77 xbt_assert(listen(master_socket_, SOMAXCONN) >= 0, "Cannot listen to the master socket: %s.", strerror(errno));
79 application_factory_ = std::make_unique<simgrid::mc::CheckerSide>(app_args_, need_memory_introspection);
80 checker_side_ = application_factory_->clone(master_socket_, master_socket_name);
84 void RemoteApp::restore_initial_state()
86 if (initial_snapshot_ == nullptr) // No memory introspection
87 checker_side_ = application_factory_->clone(master_socket_, master_socket_name);
88 #if SIMGRID_HAVE_STATEFUL_MC
90 initial_snapshot_->restore(*checker_side_->get_remote_memory());
94 unsigned long RemoteApp::get_maxpid() const
96 // note: we could maybe cache it and count the actor creation on checker side too.
97 // But counting correctly accross state checkpoint/restore would be annoying.
99 checker_side_->get_channel().send(MessageType::ACTORS_MAXPID);
100 s_mc_message_int_t answer;
101 ssize_t answer_size = checker_side_->get_channel().receive(answer);
102 xbt_assert(answer_size != -1, "Could not receive message");
103 xbt_assert(answer_size == sizeof answer, "Broken message (size=%zd; expected %zu)", answer_size, sizeof answer);
104 xbt_assert(answer.type == MessageType::ACTORS_MAXPID_REPLY,
105 "Received unexpected message %s (%i); expected MessageType::ACTORS_MAXPID_REPLY (%i)",
106 to_c_str(answer.type), (int)answer.type, (int)MessageType::ACTORS_MAXPID_REPLY);
111 void RemoteApp::get_actors_status(std::map<aid_t, ActorState>& whereto) const
113 // The messaging happens as follows:
115 // CheckerSide AppSide
116 // send ACTORS_STATUS ---->
117 // <----- send ACTORS_STATUS_REPLY_COUNT
118 // <----- send `N` ACTORS_STATUS_REPLY_TRANSITION (s_mc_message_actors_status_one_t)
119 // <----- send `M` ACTORS_STATUS_REPLY_SIMCALL (s_mc_message_simcall_probe_one_t)
121 // Note that we also receive disabled transitions, because the guiding strategies need them to decide what could
124 checker_side_->get_channel().send(MessageType::ACTORS_STATUS);
126 s_mc_message_actors_status_answer_t answer;
127 ssize_t answer_size = checker_side_->get_channel().receive(answer);
128 xbt_assert(answer_size != -1, "Could not receive message");
129 xbt_assert(answer_size == sizeof answer, "Broken message (size=%zd; expected %zu)", answer_size, sizeof answer);
130 xbt_assert(answer.type == MessageType::ACTORS_STATUS_REPLY_COUNT,
131 "%d Received unexpected message %s (%i); expected MessageType::ACTORS_STATUS_REPLY_COUNT (%i)", getpid(),
132 to_c_str(answer.type), (int)answer.type, (int)MessageType::ACTORS_STATUS_REPLY_COUNT);
134 // Message sanity checks
135 xbt_assert(answer.count >= 0, "Received an ACTORS_STATUS_REPLY_COUNT message with an actor count of '%d' < 0",
138 std::vector<s_mc_message_actors_status_one_t> status(answer.count);
139 if (answer.count > 0) {
140 size_t size = status.size() * sizeof(s_mc_message_actors_status_one_t);
141 ssize_t received = checker_side_->get_channel().receive(status.data(), size);
142 xbt_assert(static_cast<size_t>(received) == size);
147 for (const auto& actor : status) {
148 std::vector<std::shared_ptr<Transition>> actor_transitions;
149 int n_transitions = actor.max_considered;
150 for (int times_considered = 0; times_considered < n_transitions; times_considered++) {
151 s_mc_message_simcall_probe_one_t probe;
152 ssize_t received = checker_side_->get_channel().receive(probe);
153 xbt_assert(received >= 0, "Could not receive response to ACTORS_PROBE message (%s)", strerror(errno));
154 xbt_assert(static_cast<size_t>(received) == sizeof probe,
155 "Could not receive response to ACTORS_PROBE message (%zd bytes received != %zu bytes expected",
156 received, sizeof probe);
158 std::stringstream stream(probe.buffer.data());
159 actor_transitions.emplace_back(deserialize_transition(actor.aid, times_considered, stream));
162 XBT_DEBUG("Received %zu transitions for actor %ld", actor_transitions.size(), actor.aid);
163 whereto.try_emplace(actor.aid, actor.aid, actor.enabled, actor.max_considered, std::move(actor_transitions));
167 void RemoteApp::check_deadlock() const
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);
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());
187 throw McError(ExitStatus::DEADLOCK);
191 void RemoteApp::wait_for_requests()
193 checker_side_->wait_for_requests();
196 Transition* RemoteApp::handle_simcall(aid_t aid, int times_considered, bool new_transition)
198 s_mc_message_simcall_execute_t m = {};
199 m.type = MessageType::SIMCALL_EXECUTE;
201 m.times_considered_ = times_considered;
202 checker_side_->get_channel().send(m);
204 #if SIMGRID_HAVE_STATEFUL_MC
205 if (auto* memory = get_remote_process_memory(); memory != nullptr)
206 memory->clear_cache();
208 if (checker_side_->running())
209 checker_side_->dispatch_events(); // The app may send messages while processing the transition
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);
219 if (new_transition) {
220 std::stringstream stream(answer.buffer.data());
221 return deserialize_transition(aid, times_considered, stream);
226 void RemoteApp::finalize_app(bool terminate_asap)
228 checker_side_->finalize(terminate_asap);
231 } // namespace simgrid::mc