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>
25 #include <cstdio> // setvbuf
29 #include <sys/ptrace.h>
30 #include <sys/socket.h>
31 #include <sys/types.h>
35 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(mc_client, mc, "MC client logic");
36 XBT_LOG_EXTERNAL_CATEGORY(mc_global);
38 namespace simgrid::mc {
40 std::unique_ptr<AppSide> AppSide::instance_;
42 AppSide* AppSide::get()
44 // Only initialize the MC world once
45 if (instance_ != nullptr)
46 return instance_.get();
48 if (std::getenv(MC_ENV_SOCKET_FD) == nullptr) // We are not in MC mode: don't initialize the MC world
51 XBT_DEBUG("Initialize the MC world.");
53 simgrid::mc::set_model_checking_mode(ModelCheckingMode::APP_SIDE);
55 setvbuf(stdout, nullptr, _IOLBF, 0);
57 // Fetch socket from MC_ENV_SOCKET_FD:
58 const char* fd_env = std::getenv(MC_ENV_SOCKET_FD);
59 int fd = xbt_str_parse_int(fd_env, "Not a number in variable '" MC_ENV_SOCKET_FD "'");
60 XBT_DEBUG("Model-checked application found socket FD %i", fd);
62 instance_ = std::make_unique<simgrid::mc::AppSide>(fd);
64 instance_->handle_messages();
65 return instance_.get();
68 void AppSide::handle_deadlock_check(const s_mc_message_t*) const
70 const auto* engine = kernel::EngineImpl::get_instance();
71 const auto& actor_list = engine->get_actor_list();
72 bool deadlock = not actor_list.empty() && std::none_of(begin(actor_list), end(actor_list), [](const auto& kv) {
73 return mc::actor_is_enabled(kv.second);
77 XBT_CINFO(mc_global, "**************************");
78 XBT_CINFO(mc_global, "*** DEADLOCK DETECTED ***");
79 XBT_CINFO(mc_global, "**************************");
80 engine->display_all_actor_status();
83 s_mc_message_int_t answer = {};
84 answer.type = MessageType::DEADLOCK_CHECK_REPLY;
85 answer.value = deadlock;
86 xbt_assert(channel_.send(answer) == 0, "Could not send response: %s", strerror(errno));
88 void AppSide::handle_simcall_execute(const s_mc_message_simcall_execute_t* message) const
90 kernel::actor::ActorImpl* actor = kernel::EngineImpl::get_instance()->get_actor_by_pid(message->aid_);
91 xbt_assert(actor != nullptr, "Invalid pid %ld", message->aid_);
93 // The client may send some messages to the server while processing the transition
94 actor->simcall_handle(message->times_considered_);
95 // Say the server that the transition is over and that it should proceed
96 xbt_assert(channel_.send(MessageType::WAITING) == 0, "Could not send MESSAGE_WAITING to model-checker: %s",
99 // Finish the RPC from the server: return a serialized observer, to build a Transition on Checker side
100 s_mc_message_simcall_execute_answer_t answer = {};
101 answer.type = MessageType::SIMCALL_EXECUTE_REPLY;
102 std::stringstream stream;
103 if (actor->simcall_.observer_ != nullptr) {
104 actor->simcall_.observer_->serialize(stream);
106 stream << (short)mc::Transition::Type::UNKNOWN;
108 std::string str = stream.str();
109 xbt_assert(str.size() + 1 <= answer.buffer.size(),
110 "The serialized simcall is too large for the buffer. Please fix the code.");
111 strncpy(answer.buffer.data(), str.c_str(), answer.buffer.size() - 1);
112 answer.buffer.back() = '\0';
114 XBT_DEBUG("send SIMCALL_EXECUTE_ANSWER(%s) ~> '%s'", actor->get_cname(), str.c_str());
115 xbt_assert(channel_.send(answer) == 0, "Could not send response: %s", strerror(errno));
118 void AppSide::handle_finalize(const s_mc_message_int_t* msg) const
120 bool terminate_asap = msg->value;
121 XBT_DEBUG("Finalize (terminate = %d)", (int)terminate_asap);
122 if (not terminate_asap) {
123 if (XBT_LOG_ISENABLED(mc_client, xbt_log_priority_debug))
124 kernel::EngineImpl::get_instance()->display_all_actor_status();
126 XBT_DEBUG("Smpi_enabled: %d", SMPI_is_inited());
127 if (SMPI_is_inited())
131 coverage_checkpoint();
132 xbt_assert(channel_.send(MessageType::FINALIZE_REPLY) == 0, "Could not answer to FINALIZE: %s", strerror(errno));
137 void AppSide::handle_fork(const s_mc_message_fork_t* msg)
141 /* Reap any zombie child, saving its status for later use in AppSide::handle_wait_child() */
142 while ((pid = waitpid(-1, &status, WNOHANG)) > 0)
143 child_statuses_[pid] = status;
146 xbt_assert(pid >= 0, "Could not fork application sub-process: %s.", strerror(errno));
148 if (pid == 0) { // Child
149 int sock = socket(AF_UNIX,
151 SOCK_STREAM, /* Mac OSX does not have AF_UNIX + SOCK_SEQPACKET, even if that's faster*/
157 struct sockaddr_un addr = {};
158 addr.sun_family = AF_UNIX;
159 std::copy_n(begin(msg->socket_name), MC_SOCKET_NAME_LEN, addr.sun_path);
161 xbt_assert(connect(sock, (struct sockaddr*)&addr, sizeof addr) >= 0, "Cannot connect to Checker on %c%s: %s.",
162 (addr.sun_path[0] ? addr.sun_path[0] : '@'), addr.sun_path + 1, strerror(errno));
164 channel_.reset_socket(sock);
166 s_mc_message_int_t answer = {};
167 answer.type = MessageType::FORK_REPLY;
168 answer.value = getpid();
169 xbt_assert(channel_.send(answer) == 0, "Could not send response to WAIT_CHILD_REPLY: %s", strerror(errno));
171 XBT_VERB("App %d forks subprocess %d.", getpid(), pid);
174 void AppSide::handle_wait_child(const s_mc_message_int_t* msg)
178 if (auto search = child_statuses_.find(msg->value); search != child_statuses_.end()) {
179 status = search->second;
180 child_statuses_.erase(search); // We only need this info once
182 waitpid(msg->value, &status, 0);
184 xbt_assert(errno == 0, "Cannot wait on behalf of the checker: %s.", strerror(errno));
186 s_mc_message_int_t answer = {};
187 answer.type = MessageType::WAIT_CHILD_REPLY;
188 answer.value = status;
189 xbt_assert(channel_.send(answer) == 0, "Could not send response to WAIT_CHILD: %s", strerror(errno));
191 void AppSide::handle_actors_status() const
193 auto const& actor_list = kernel::EngineImpl::get_instance()->get_actor_list();
194 XBT_DEBUG("Serialize the actors to answer ACTORS_STATUS from the checker. %zu actors to go.", actor_list.size());
196 std::vector<s_mc_message_actors_status_one_t> status;
197 for (auto const& [aid, actor] : actor_list) {
199 xbt_assert(actor->simcall_.observer_, "simcall %s in actor %s has no observer.", actor->simcall_.get_cname(),
201 s_mc_message_actors_status_one_t one = {};
202 one.type = MessageType::ACTORS_STATUS_REPLY_TRANSITION;
204 one.enabled = mc::actor_is_enabled(actor);
205 one.max_considered = actor->simcall_.observer_->get_max_consider();
206 status.push_back(one);
209 struct s_mc_message_actors_status_answer_t answer = {};
210 answer.type = MessageType::ACTORS_STATUS_REPLY_COUNT;
211 answer.count = static_cast<int>(status.size());
213 xbt_assert(channel_.send(answer) == 0, "Could not send ACTORS_STATUS_REPLY msg: %s", strerror(errno));
214 if (answer.count > 0) {
215 size_t size = status.size() * sizeof(s_mc_message_actors_status_one_t);
216 xbt_assert(channel_.send(status.data(), size) == 0, "Could not send ACTORS_STATUS_REPLY data: %s", strerror(errno));
219 // Serialize each transition to describe what each actor is doing
220 XBT_DEBUG("Deliver ACTOR_TRANSITION_PROBE payload");
221 for (const auto& actor_status : status) {
222 const auto& actor = actor_list.at(actor_status.aid);
223 const int max_considered = actor_status.max_considered;
225 for (int times_considered = 0; times_considered < max_considered; times_considered++) {
226 std::stringstream stream;
227 s_mc_message_simcall_probe_one_t probe;
228 probe.type = MessageType::ACTORS_STATUS_REPLY_SIMCALL;
230 if (actor->simcall_.observer_ != nullptr) {
231 actor->simcall_.observer_->prepare(times_considered);
232 actor->simcall_.observer_->serialize(stream);
234 stream << (short)mc::Transition::Type::UNKNOWN;
237 std::string str = stream.str();
238 xbt_assert(str.size() + 1 <= probe.buffer.size(),
239 "The serialized transition is too large for the buffer. Please fix the code.");
240 strncpy(probe.buffer.data(), str.c_str(), probe.buffer.size() - 1);
241 probe.buffer.back() = '\0';
243 XBT_DEBUG("send ACTOR_TRANSITION_PROBE(%s) ~> '%s'", actor->get_cname(), str.c_str());
244 xbt_assert(channel_.send(probe) == 0, "Could not send ACTOR_TRANSITION_PROBE payload: %s", strerror(errno));
246 // NOTE: We do NOT need to reset `times_considered` for each actor's
247 // simcall observer here to the "original" value (i.e. the value BEFORE
248 // multiple prepare() calls were made for serialization purposes) since
249 // each SIMCALL_EXECUTE provides a `times_considered` to be used to prepare
250 // the transition before execution.
253 void AppSide::handle_actors_maxpid() const
255 s_mc_message_int_t answer = {};
256 answer.type = MessageType::ACTORS_MAXPID_REPLY;
257 answer.value = kernel::actor::ActorImpl::get_maxpid();
258 xbt_assert(channel_.send(answer) == 0, "Could not send response: %s", strerror(errno));
261 #define assert_msg_size(_name_, _type_) \
262 xbt_assert(received_size == sizeof(_type_), "Unexpected size for " _name_ " (%zd != %zu)", received_size, \
265 void AppSide::handle_messages()
267 while (true) { // Until we get a CONTINUE message
268 XBT_DEBUG("Waiting messages from the model-checker");
270 std::array<char, MC_MESSAGE_LENGTH> message_buffer;
271 ssize_t received_size = channel_.receive(message_buffer.data(), message_buffer.size());
273 if (received_size == 0) {
274 XBT_DEBUG("Socket closed on the Checker side, bailing out.");
275 ::_Exit(0); // Nobody's listening to that process anymore => exit as quickly as possible.
277 xbt_assert(received_size >= 0, "Could not receive commands from the model-checker: %s", strerror(errno));
278 xbt_assert(static_cast<size_t>(received_size) >= sizeof(s_mc_message_t), "Cannot handle short message (size=%zd)",
281 const s_mc_message_t* message = (s_mc_message_t*)message_buffer.data();
282 switch (message->type) {
283 case MessageType::CONTINUE:
284 assert_msg_size("MESSAGE_CONTINUE", s_mc_message_t);
287 case MessageType::DEADLOCK_CHECK:
288 assert_msg_size("DEADLOCK_CHECK", s_mc_message_t);
289 handle_deadlock_check(message);
292 case MessageType::SIMCALL_EXECUTE:
293 assert_msg_size("SIMCALL_EXECUTE", s_mc_message_simcall_execute_t);
294 handle_simcall_execute((s_mc_message_simcall_execute_t*)message_buffer.data());
297 case MessageType::FINALIZE:
298 assert_msg_size("FINALIZE", s_mc_message_int_t);
299 handle_finalize((s_mc_message_int_t*)message_buffer.data());
302 case MessageType::FORK:
303 assert_msg_size("FORK", s_mc_message_fork_t);
304 handle_fork((s_mc_message_fork_t*)message_buffer.data());
307 case MessageType::WAIT_CHILD:
308 assert_msg_size("WAIT_CHILD", s_mc_message_int_t);
309 handle_wait_child((s_mc_message_int_t*)message_buffer.data());
312 case MessageType::ACTORS_STATUS:
313 assert_msg_size("ACTORS_STATUS", s_mc_message_t);
314 handle_actors_status();
317 case MessageType::ACTORS_MAXPID:
318 assert_msg_size("ACTORS_MAXPID", s_mc_message_t);
319 handle_actors_maxpid();
323 xbt_die("Received unexpected message %s (%i)", to_c_str(message->type), static_cast<int>(message->type));
329 void AppSide::main_loop()
331 simgrid::mc::processes_time.resize(simgrid::kernel::actor::ActorImpl::get_maxpid());
334 coverage_checkpoint();
337 simgrid::mc::execute_actors();
338 xbt_assert(channel_.send(MessageType::WAITING) == 0, "Could not send WAITING message to model-checker: %s",
340 this->handle_messages();
344 void AppSide::report_assertion_failure()
346 xbt_assert(channel_.send(MessageType::ASSERTION_FAILED) == 0, "Could not send assertion to model-checker: %s",
348 this->handle_messages();
351 } // namespace simgrid::mc