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_);
92 xbt_assert(actor->simcall_.observer_ == nullptr || actor->simcall_.observer_->is_enabled(),
93 "Please, model-checker, don't execute disabled transitions.");
95 // The client may send some messages to the server while processing the transition
96 actor->simcall_handle(message->times_considered_);
97 // Say the server that the transition is over and that it should proceed
98 xbt_assert(channel_.send(MessageType::WAITING) == 0, "Could not send MESSAGE_WAITING to model-checker: %s",
101 // Finish the RPC from the server: return a serialized observer, to build a Transition on Checker side
102 s_mc_message_simcall_execute_answer_t answer = {};
103 answer.type = MessageType::SIMCALL_EXECUTE_REPLY;
104 std::stringstream stream;
105 if (actor->simcall_.observer_ != nullptr) {
106 actor->simcall_.observer_->serialize(stream);
108 stream << (short)mc::Transition::Type::UNKNOWN;
110 std::string str = stream.str();
111 xbt_assert(str.size() + 1 <= answer.buffer.size(),
112 "The serialized simcall is too large for the buffer. Please fix the code.");
113 strncpy(answer.buffer.data(), str.c_str(), answer.buffer.size() - 1);
114 answer.buffer.back() = '\0';
116 XBT_DEBUG("send SIMCALL_EXECUTE_ANSWER(%s) ~> '%s'", actor->get_cname(), str.c_str());
117 xbt_assert(channel_.send(answer) == 0, "Could not send response: %s", strerror(errno));
120 void AppSide::handle_finalize(const s_mc_message_int_t* msg) const
122 bool terminate_asap = msg->value;
123 XBT_DEBUG("Finalize (terminate = %d)", (int)terminate_asap);
124 if (not terminate_asap) {
125 if (XBT_LOG_ISENABLED(mc_client, xbt_log_priority_debug))
126 kernel::EngineImpl::get_instance()->display_all_actor_status();
128 XBT_DEBUG("Smpi_enabled: %d", SMPI_is_inited());
129 if (SMPI_is_inited())
133 coverage_checkpoint();
134 xbt_assert(channel_.send(MessageType::FINALIZE_REPLY) == 0, "Could not answer to FINALIZE: %s", strerror(errno));
139 void AppSide::handle_fork(const s_mc_message_fork_t* msg)
143 /* Reap any zombie child, saving its status for later use in AppSide::handle_wait_child() */
144 while ((pid = waitpid(-1, &status, WNOHANG)) > 0)
145 child_statuses_[pid] = status;
148 xbt_assert(pid >= 0, "Could not fork application sub-process: %s.", strerror(errno));
150 if (pid == 0) { // Child
151 int sock = socket(AF_UNIX,
153 SOCK_STREAM, /* Mac OSX does not have AF_UNIX + SOCK_SEQPACKET, even if that's faster*/
159 struct sockaddr_un addr = {};
160 addr.sun_family = AF_UNIX;
161 std::copy_n(begin(msg->socket_name), MC_SOCKET_NAME_LEN, addr.sun_path);
163 xbt_assert(connect(sock, (struct sockaddr*)&addr, sizeof addr) >= 0, "Cannot connect to Checker on %c%s: %s.",
164 (addr.sun_path[0] ? addr.sun_path[0] : '@'), addr.sun_path + 1, strerror(errno));
166 channel_.reset_socket(sock);
168 s_mc_message_int_t answer = {};
169 answer.type = MessageType::FORK_REPLY;
170 answer.value = getpid();
171 xbt_assert(channel_.send(answer) == 0, "Could not send response to WAIT_CHILD_REPLY: %s", strerror(errno));
173 XBT_VERB("App %d forks subprocess %d.", getpid(), pid);
176 void AppSide::handle_wait_child(const s_mc_message_int_t* msg)
180 if (auto search = child_statuses_.find(msg->value); search != child_statuses_.end()) {
181 status = search->second;
182 child_statuses_.erase(search); // We only need this info once
184 waitpid(msg->value, &status, 0);
186 xbt_assert(errno == 0, "Cannot wait on behalf of the checker: %s.", strerror(errno));
188 s_mc_message_int_t answer = {};
189 answer.type = MessageType::WAIT_CHILD_REPLY;
190 answer.value = status;
191 xbt_assert(channel_.send(answer) == 0, "Could not send response to WAIT_CHILD: %s", strerror(errno));
193 void AppSide::handle_actors_status() const
195 auto const& actor_list = kernel::EngineImpl::get_instance()->get_actor_list();
196 XBT_DEBUG("Serialize the actors to answer ACTORS_STATUS from the checker. %zu actors to go.", actor_list.size());
198 std::vector<s_mc_message_actors_status_one_t> status;
199 for (auto const& [aid, actor] : actor_list) {
201 xbt_assert(actor->simcall_.observer_, "simcall %s in actor %s has no observer.", actor->simcall_.get_cname(),
203 s_mc_message_actors_status_one_t one = {};
204 one.type = MessageType::ACTORS_STATUS_REPLY_TRANSITION;
206 one.enabled = mc::actor_is_enabled(actor);
207 one.max_considered = actor->simcall_.observer_->get_max_consider();
208 status.push_back(one);
211 struct s_mc_message_actors_status_answer_t answer = {};
212 answer.type = MessageType::ACTORS_STATUS_REPLY_COUNT;
213 answer.count = static_cast<int>(status.size());
215 xbt_assert(channel_.send(answer) == 0, "Could not send ACTORS_STATUS_REPLY msg: %s", strerror(errno));
216 if (answer.count > 0) {
217 size_t size = status.size() * sizeof(s_mc_message_actors_status_one_t);
218 xbt_assert(channel_.send(status.data(), size) == 0, "Could not send ACTORS_STATUS_REPLY data: %s", strerror(errno));
221 // Serialize each transition to describe what each actor is doing
222 XBT_DEBUG("Deliver ACTOR_TRANSITION_PROBE payload");
223 for (const auto& actor_status : status) {
224 const auto& actor = actor_list.at(actor_status.aid);
225 const int max_considered = actor_status.max_considered;
227 for (int times_considered = 0; times_considered < max_considered; times_considered++) {
228 std::stringstream stream;
229 s_mc_message_simcall_probe_one_t probe;
230 probe.type = MessageType::ACTORS_STATUS_REPLY_SIMCALL;
232 if (actor->simcall_.observer_ != nullptr) {
233 actor->simcall_.observer_->prepare(times_considered);
234 actor->simcall_.observer_->serialize(stream);
236 stream << (short)mc::Transition::Type::UNKNOWN;
239 std::string str = stream.str();
240 xbt_assert(str.size() + 1 <= probe.buffer.size(),
241 "The serialized transition is too large for the buffer. Please fix the code.");
242 strncpy(probe.buffer.data(), str.c_str(), probe.buffer.size() - 1);
243 probe.buffer.back() = '\0';
245 XBT_DEBUG("send ACTOR_TRANSITION_PROBE(%s) ~> '%s'", actor->get_cname(), str.c_str());
246 xbt_assert(channel_.send(probe) == 0, "Could not send ACTOR_TRANSITION_PROBE payload: %s", strerror(errno));
248 // NOTE: We do NOT need to reset `times_considered` for each actor's
249 // simcall observer here to the "original" value (i.e. the value BEFORE
250 // multiple prepare() calls were made for serialization purposes) since
251 // each SIMCALL_EXECUTE provides a `times_considered` to be used to prepare
252 // the transition before execution.
255 void AppSide::handle_actors_maxpid() const
257 s_mc_message_int_t answer = {};
258 answer.type = MessageType::ACTORS_MAXPID_REPLY;
259 answer.value = kernel::actor::ActorImpl::get_maxpid();
260 xbt_assert(channel_.send(answer) == 0, "Could not send response: %s", strerror(errno));
263 #define assert_msg_size(_name_, _type_) \
264 xbt_assert(received_size == sizeof(_type_), "Unexpected size for " _name_ " (%zd != %zu)", received_size, \
267 void AppSide::handle_messages()
269 while (true) { // Until we get a CONTINUE message
270 XBT_DEBUG("Waiting messages from the model-checker");
272 std::array<char, MC_MESSAGE_LENGTH> message_buffer;
273 ssize_t received_size = channel_.receive(message_buffer.data(), message_buffer.size());
275 if (received_size == 0) {
276 XBT_DEBUG("Socket closed on the Checker side, bailing out.");
277 ::_Exit(0); // Nobody's listening to that process anymore => exit as quickly as possible.
279 xbt_assert(received_size >= 0, "Could not receive commands from the model-checker: %s", strerror(errno));
280 xbt_assert(static_cast<size_t>(received_size) >= sizeof(s_mc_message_t), "Cannot handle short message (size=%zd)",
283 const s_mc_message_t* message = (s_mc_message_t*)message_buffer.data();
284 switch (message->type) {
285 case MessageType::CONTINUE:
286 assert_msg_size("MESSAGE_CONTINUE", s_mc_message_t);
289 case MessageType::DEADLOCK_CHECK:
290 assert_msg_size("DEADLOCK_CHECK", s_mc_message_t);
291 handle_deadlock_check(message);
294 case MessageType::SIMCALL_EXECUTE:
295 assert_msg_size("SIMCALL_EXECUTE", s_mc_message_simcall_execute_t);
296 handle_simcall_execute((s_mc_message_simcall_execute_t*)message_buffer.data());
299 case MessageType::FINALIZE:
300 assert_msg_size("FINALIZE", s_mc_message_int_t);
301 handle_finalize((s_mc_message_int_t*)message_buffer.data());
304 case MessageType::FORK:
305 assert_msg_size("FORK", s_mc_message_fork_t);
306 handle_fork((s_mc_message_fork_t*)message_buffer.data());
309 case MessageType::WAIT_CHILD:
310 assert_msg_size("WAIT_CHILD", s_mc_message_int_t);
311 handle_wait_child((s_mc_message_int_t*)message_buffer.data());
314 case MessageType::ACTORS_STATUS:
315 assert_msg_size("ACTORS_STATUS", s_mc_message_t);
316 handle_actors_status();
319 case MessageType::ACTORS_MAXPID:
320 assert_msg_size("ACTORS_MAXPID", s_mc_message_t);
321 handle_actors_maxpid();
325 xbt_die("Received unexpected message %s (%i)", to_c_str(message->type), static_cast<int>(message->type));
331 void AppSide::main_loop()
333 simgrid::mc::processes_time.resize(simgrid::kernel::actor::ActorImpl::get_maxpid());
336 coverage_checkpoint();
339 simgrid::mc::execute_actors();
340 xbt_assert(channel_.send(MessageType::WAITING) == 0, "Could not send WAITING message to model-checker: %s",
342 this->handle_messages();
346 void AppSide::report_assertion_failure()
348 xbt_assert(channel_.send(MessageType::ASSERTION_FAILED) == 0, "Could not send assertion to model-checker: %s",
350 this->handle_messages();
353 } // namespace simgrid::mc