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/remote/AppSide.hpp"
7 #include "simgrid/s4u/Host.hpp"
8 #include "src/internal_config.h"
9 #include "src/kernel/EngineImpl.hpp"
10 #include "src/kernel/actor/ActorImpl.hpp"
11 #include "src/kernel/actor/SimcallObserver.hpp"
12 #include "src/mc/mc_base.hpp"
13 #include "src/mc/mc_config.hpp"
14 #include "src/mc/mc_environ.h"
16 #include "src/smpi/include/private.hpp"
18 #include "src/sthread/sthread.h"
19 #include "src/xbt/coverage.h"
21 #include <simgrid/modelchecker.h>
23 #include <boost/lexical_cast.hpp>
26 #include <cstdio> // setvbuf
30 #include <sys/ptrace.h>
31 #include <sys/socket.h>
32 #include <sys/types.h>
36 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(mc_client, mc, "MC client logic");
37 XBT_LOG_EXTERNAL_CATEGORY(mc_global);
39 namespace simgrid::mc {
41 std::unique_ptr<AppSide> AppSide::instance_;
43 AppSide* AppSide::get()
45 // Only initialize the MC world once
46 if (instance_ != nullptr)
47 return instance_.get();
49 if (std::getenv(MC_ENV_SOCKET_FD) == nullptr) // We are not in MC mode: don't initialize the MC world
52 XBT_DEBUG("Initialize the MC world.");
54 simgrid::mc::set_model_checking_mode(ModelCheckingMode::APP_SIDE);
56 setvbuf(stdout, nullptr, _IOLBF, 0);
58 // Fetch socket from MC_ENV_SOCKET_FD:
59 const char* fd_env = std::getenv(MC_ENV_SOCKET_FD);
62 fd = boost::lexical_cast<int>(fd_env);
63 } catch (boost::bad_lexical_cast const&) {
64 xbt_die("Not a number in variable '%s': %s", MC_ENV_SOCKET_FD, fd_env);
66 XBT_DEBUG("Model-checked application found socket FD %i", fd);
68 instance_ = std::make_unique<simgrid::mc::AppSide>(fd);
70 instance_->handle_messages();
71 return instance_.get();
74 void AppSide::handle_deadlock_check(const s_mc_message_t*) const
76 const auto* engine = kernel::EngineImpl::get_instance();
77 const auto& actor_list = engine->get_actor_list();
78 bool deadlock = not actor_list.empty() && std::none_of(begin(actor_list), end(actor_list), [](const auto& kv) {
79 return mc::actor_is_enabled(kv.second);
83 XBT_CINFO(mc_global, "**************************");
84 XBT_CINFO(mc_global, "*** DEADLOCK DETECTED ***");
85 XBT_CINFO(mc_global, "**************************");
86 engine->display_all_actor_status();
89 s_mc_message_int_t answer = {};
90 answer.type = MessageType::DEADLOCK_CHECK_REPLY;
91 answer.value = deadlock;
92 xbt_assert(channel_.send(answer) == 0, "Could not send response: %s", strerror(errno));
94 void AppSide::handle_simcall_execute(const s_mc_message_simcall_execute_t* message) const
96 kernel::actor::ActorImpl* actor = kernel::EngineImpl::get_instance()->get_actor_by_pid(message->aid_);
97 xbt_assert(actor != nullptr, "Invalid pid %ld", message->aid_);
98 xbt_assert(actor->simcall_.observer_ == nullptr || actor->simcall_.observer_->is_enabled(),
99 "Please, model-checker, don't execute disabled transitions.");
101 // The client may send some messages to the server while processing the transition
102 actor->simcall_handle(message->times_considered_);
103 // Say the server that the transition is over and that it should proceed
104 xbt_assert(channel_.send(MessageType::WAITING) == 0, "Could not send MESSAGE_WAITING to model-checker: %s",
107 // Finish the RPC from the server: return a serialized observer, to build a Transition on Checker side
108 s_mc_message_simcall_execute_answer_t answer = {};
109 answer.type = MessageType::SIMCALL_EXECUTE_REPLY;
110 std::stringstream stream;
111 if (actor->simcall_.observer_ != nullptr) {
112 actor->simcall_.observer_->serialize(stream);
114 stream << (short)mc::Transition::Type::UNKNOWN;
116 std::string str = stream.str();
117 xbt_assert(str.size() + 1 <= answer.buffer.size(),
118 "The serialized simcall is too large for the buffer. Please fix the code.");
119 strncpy(answer.buffer.data(), str.c_str(), answer.buffer.size() - 1);
120 answer.buffer.back() = '\0';
122 XBT_DEBUG("send SIMCALL_EXECUTE_ANSWER(%s) ~> '%s'", actor->get_cname(), str.c_str());
123 xbt_assert(channel_.send(answer) == 0, "Could not send response: %s", strerror(errno));
126 void AppSide::handle_finalize(const s_mc_message_int_t* msg) const
128 bool terminate_asap = msg->value;
129 XBT_DEBUG("Finalize (terminate = %d)", (int)terminate_asap);
130 if (not terminate_asap) {
131 if (XBT_LOG_ISENABLED(mc_client, xbt_log_priority_debug))
132 kernel::EngineImpl::get_instance()->display_all_actor_status();
134 XBT_DEBUG("Smpi_enabled: %d", SMPI_is_inited());
135 if (SMPI_is_inited())
139 coverage_checkpoint();
140 xbt_assert(channel_.send(MessageType::FINALIZE_REPLY) == 0, "Could not answer to FINALIZE: %s", strerror(errno));
145 void AppSide::handle_fork(const s_mc_message_fork_t* msg)
149 /* Reap any zombie child, saving its status for later use in AppSide::handle_wait_child() */
150 while ((pid = waitpid(-1, &status, WNOHANG)) > 0)
151 child_statuses_[pid] = status;
154 xbt_assert(pid >= 0, "Could not fork application sub-process: %s.", strerror(errno));
156 if (pid == 0) { // Child
157 int sock = socket(AF_UNIX,
159 SOCK_STREAM, /* Mac OSX does not have AF_UNIX + SOCK_SEQPACKET, even if that's faster*/
165 struct sockaddr_un addr = {};
166 addr.sun_family = AF_UNIX;
167 std::copy_n(begin(msg->socket_name), MC_SOCKET_NAME_LEN, addr.sun_path);
169 xbt_assert(connect(sock, (struct sockaddr*)&addr, sizeof addr) >= 0, "Cannot connect to Checker on %c%s: %s.",
170 (addr.sun_path[0] ? addr.sun_path[0] : '@'), addr.sun_path + 1, strerror(errno));
172 channel_.reset_socket(sock);
174 s_mc_message_int_t answer = {};
175 answer.type = MessageType::FORK_REPLY;
176 answer.value = getpid();
177 xbt_assert(channel_.send(answer) == 0, "Could not send response to WAIT_CHILD_REPLY: %s", strerror(errno));
179 XBT_VERB("App %d forks subprocess %d.", getpid(), pid);
182 void AppSide::handle_wait_child(const s_mc_message_int_t* msg)
186 if (auto search = child_statuses_.find(msg->value); search != child_statuses_.end()) {
187 status = search->second;
188 child_statuses_.erase(search); // We only need this info once
190 waitpid(msg->value, &status, 0);
192 xbt_assert(errno == 0, "Cannot wait on behalf of the checker: %s.", strerror(errno));
194 s_mc_message_int_t answer = {};
195 answer.type = MessageType::WAIT_CHILD_REPLY;
196 answer.value = status;
197 xbt_assert(channel_.send(answer) == 0, "Could not send response to WAIT_CHILD: %s", strerror(errno));
199 void AppSide::handle_actors_status() const
201 auto const& actor_list = kernel::EngineImpl::get_instance()->get_actor_list();
202 XBT_DEBUG("Serialize the actors to answer ACTORS_STATUS from the checker. %zu actors to go.", actor_list.size());
204 std::vector<s_mc_message_actors_status_one_t> status;
205 for (auto const& [aid, actor] : actor_list) {
207 xbt_assert(actor->simcall_.observer_, "simcall %s in actor %s has no observer.", actor->simcall_.get_cname(),
209 s_mc_message_actors_status_one_t one = {};
210 one.type = MessageType::ACTORS_STATUS_REPLY_TRANSITION;
212 one.enabled = mc::actor_is_enabled(actor);
213 one.max_considered = actor->simcall_.observer_->get_max_consider();
214 status.push_back(one);
217 struct s_mc_message_actors_status_answer_t answer = {};
218 answer.type = MessageType::ACTORS_STATUS_REPLY_COUNT;
219 answer.count = static_cast<int>(status.size());
221 xbt_assert(channel_.send(answer) == 0, "Could not send ACTORS_STATUS_REPLY msg: %s", strerror(errno));
222 if (answer.count > 0) {
223 size_t size = status.size() * sizeof(s_mc_message_actors_status_one_t);
224 xbt_assert(channel_.send(status.data(), size) == 0, "Could not send ACTORS_STATUS_REPLY data: %s", strerror(errno));
227 // Serialize each transition to describe what each actor is doing
228 XBT_DEBUG("Deliver ACTOR_TRANSITION_PROBE payload");
229 for (const auto& actor_status : status) {
230 const auto& actor = actor_list.at(actor_status.aid);
231 const int max_considered = actor_status.max_considered;
233 for (int times_considered = 0; times_considered < max_considered; times_considered++) {
234 std::stringstream stream;
235 s_mc_message_simcall_probe_one_t probe;
236 probe.type = MessageType::ACTORS_STATUS_REPLY_SIMCALL;
238 if (actor->simcall_.observer_ != nullptr) {
239 actor->simcall_.observer_->prepare(times_considered);
240 actor->simcall_.observer_->serialize(stream);
242 stream << (short)mc::Transition::Type::UNKNOWN;
245 std::string str = stream.str();
246 xbt_assert(str.size() + 1 <= probe.buffer.size(),
247 "The serialized transition is too large for the buffer. Please fix the code.");
248 strncpy(probe.buffer.data(), str.c_str(), probe.buffer.size() - 1);
249 probe.buffer.back() = '\0';
251 XBT_DEBUG("send ACTOR_TRANSITION_PROBE(%s) ~> '%s'", actor->get_cname(), str.c_str());
252 xbt_assert(channel_.send(probe) == 0, "Could not send ACTOR_TRANSITION_PROBE payload: %s", strerror(errno));
254 // NOTE: We do NOT need to reset `times_considered` for each actor's
255 // simcall observer here to the "original" value (i.e. the value BEFORE
256 // multiple prepare() calls were made for serialization purposes) since
257 // each SIMCALL_EXECUTE provides a `times_considered` to be used to prepare
258 // the transition before execution.
261 void AppSide::handle_actors_maxpid() const
263 s_mc_message_int_t answer = {};
264 answer.type = MessageType::ACTORS_MAXPID_REPLY;
265 answer.value = kernel::actor::ActorImpl::get_maxpid();
266 xbt_assert(channel_.send(answer) == 0, "Could not send response: %s", strerror(errno));
269 #define assert_msg_size(_name_, _type_) \
270 xbt_assert(received_size == sizeof(_type_), "Unexpected size for " _name_ " (%zd != %zu)", received_size, \
273 void AppSide::handle_messages()
275 while (true) { // Until we get a CONTINUE message
276 XBT_DEBUG("Waiting messages from the model-checker");
278 std::array<char, MC_MESSAGE_LENGTH> message_buffer;
279 ssize_t received_size = channel_.receive(message_buffer.data(), message_buffer.size());
281 if (received_size == 0) {
282 XBT_DEBUG("Socket closed on the Checker side, bailing out.");
283 ::_Exit(0); // Nobody's listening to that process anymore => exit as quickly as possible.
285 xbt_assert(received_size >= 0, "Could not receive commands from the model-checker: %s", strerror(errno));
286 xbt_assert(static_cast<size_t>(received_size) >= sizeof(s_mc_message_t), "Cannot handle short message (size=%zd)",
289 const s_mc_message_t* message = (s_mc_message_t*)message_buffer.data();
290 switch (message->type) {
291 case MessageType::CONTINUE:
292 assert_msg_size("MESSAGE_CONTINUE", s_mc_message_t);
295 case MessageType::DEADLOCK_CHECK:
296 assert_msg_size("DEADLOCK_CHECK", s_mc_message_t);
297 handle_deadlock_check(message);
300 case MessageType::SIMCALL_EXECUTE:
301 assert_msg_size("SIMCALL_EXECUTE", s_mc_message_simcall_execute_t);
302 handle_simcall_execute((s_mc_message_simcall_execute_t*)message_buffer.data());
305 case MessageType::FINALIZE:
306 assert_msg_size("FINALIZE", s_mc_message_int_t);
307 handle_finalize((s_mc_message_int_t*)message_buffer.data());
310 case MessageType::FORK:
311 assert_msg_size("FORK", s_mc_message_fork_t);
312 handle_fork((s_mc_message_fork_t*)message_buffer.data());
315 case MessageType::WAIT_CHILD:
316 assert_msg_size("WAIT_CHILD", s_mc_message_int_t);
317 handle_wait_child((s_mc_message_int_t*)message_buffer.data());
320 case MessageType::ACTORS_STATUS:
321 assert_msg_size("ACTORS_STATUS", s_mc_message_t);
322 handle_actors_status();
325 case MessageType::ACTORS_MAXPID:
326 assert_msg_size("ACTORS_MAXPID", s_mc_message_t);
327 handle_actors_maxpid();
331 xbt_die("Received unexpected message %s (%i)", to_c_str(message->type), static_cast<int>(message->type));
337 void AppSide::main_loop()
339 simgrid::mc::processes_time.resize(simgrid::kernel::actor::ActorImpl::get_maxpid());
342 coverage_checkpoint();
345 simgrid::mc::execute_actors();
346 xbt_assert(channel_.send(MessageType::WAITING) == 0, "Could not send WAITING message to model-checker: %s",
348 this->handle_messages();
352 void AppSide::report_assertion_failure()
354 xbt_assert(channel_.send(MessageType::ASSERTION_FAILED) == 0, "Could not send assertion to model-checker: %s",
356 this->handle_messages();
359 } // namespace simgrid::mc