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"
20 #include <boost/tokenizer.hpp>
27 #include <sys/prctl.h>
29 #include <sys/ptrace.h>
33 #define WAITPID_CHECKED_FLAGS __WALL
35 #define WAITPID_CHECKED_FLAGS 0
38 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(mc_Session, mc, "Model-checker session");
39 XBT_LOG_EXTERNAL_CATEGORY(mc_global);
41 static simgrid::config::Flag<std::string> _sg_mc_setenv{
42 "model-check/setenv", "Extra environment variables to pass to the child process (ex: 'AZE=aze;QWE=qwe').", "",
43 [](std::string_view value) {
44 xbt_assert(value.empty() || value.find('=', 0) != std::string_view::npos,
45 "The 'model-check/setenv' parameter must be like 'AZE=aze', but it does not contain an equal sign.");
48 namespace simgrid::mc {
50 XBT_ATTRIB_NORETURN static void run_child_process(int socket, const std::vector<char*>& args)
52 /* On startup, simix_global_init() calls simgrid::mc::Client::initialize(), which checks whether the MC_ENV_SOCKET_FD
53 * env variable is set. If so, MC mode is assumed, and the client is setup from its side
57 // Make sure we do not outlive our parent
60 xbt_assert(sigprocmask(SIG_SETMASK, &mask, nullptr) >= 0, "Could not unblock signals");
61 xbt_assert(prctl(PR_SET_PDEATHSIG, SIGHUP) == 0, "Could not PR_SET_PDEATHSIG");
64 // Remove CLOEXEC to pass the socket to the application
65 int fdflags = fcntl(socket, F_GETFD, 0);
66 xbt_assert(fdflags != -1 && fcntl(socket, F_SETFD, fdflags & ~FD_CLOEXEC) != -1,
67 "Could not remove CLOEXEC for socket");
69 setenv(MC_ENV_SOCKET_FD, std::to_string(socket).c_str(), 1);
71 /* Setup the tokenizer that parses the cfg:model-check/setenv parameter */
72 using Tokenizer = boost::tokenizer<boost::char_separator<char>>;
73 boost::char_separator<char> semicol_sep(";");
74 boost::char_separator<char> equal_sep("=");
75 Tokenizer token_vars(_sg_mc_setenv.get(), semicol_sep); /* Iterate over all FOO=foo parts */
76 for (const auto& token : token_vars) {
77 std::vector<std::string> kv;
78 Tokenizer token_kv(token, equal_sep);
79 for (const auto& t : token_kv) /* Iterate over 'FOO' and then 'foo' in that 'FOO=foo' */
81 xbt_assert(kv.size() == 2, "Parse error on 'model-check/setenv' value %s. Does it contain an equal sign?",
83 XBT_INFO("setenv '%s'='%s'", kv[0].c_str(), kv[1].c_str());
84 setenv(kv[0].c_str(), kv[1].c_str(), 1);
87 /* And now, exec the child process */
89 while (args[i] != nullptr && args[i][0] == '-')
92 xbt_assert(args[i] != nullptr,
93 "Unable to find a binary to exec on the command line. Did you only pass config flags?");
95 execvp(args[i], args.data() + i);
96 XBT_CRITICAL("The model-checked process failed to exec(%s): %s.\n"
97 " Make sure that your binary exists on disk and is executable.",
98 args[i], strerror(errno));
99 if (strchr(args[i], '=') != nullptr)
100 XBT_CRITICAL("If you want to pass environment variables to the application, please use --cfg=model-check/setenv:%s",
103 xbt_die("Aborting now.");
106 RemoteApp::RemoteApp(const std::vector<char*>& args)
108 // Create an AF_LOCAL socketpair used for exchanging messages
109 // between the model-checker process (ourselves) and the model-checked
112 xbt_assert(socketpair(AF_LOCAL, SOCK_SEQPACKET | SOCK_CLOEXEC, 0, sockets) != -1, "Could not create socketpair");
115 xbt_assert(pid >= 0, "Could not fork model-checked process");
117 if (pid == 0) { // Child
119 run_child_process(sockets[0], args);
123 // Parent (model-checker):
126 xbt_assert(mc_model_checker == nullptr, "Did you manage to start the MC twice in this process?");
128 checker_side_ = std::make_unique<simgrid::mc::CheckerSide>(sockets[1]);
129 auto process = std::make_unique<simgrid::mc::RemoteProcessMemory>(pid);
130 model_checker_ = std::make_unique<simgrid::mc::ModelChecker>(std::move(process));
132 mc_model_checker = model_checker_.get();
135 /* Take the initial snapshot */
138 std::make_shared<simgrid::mc::Snapshot>(0, page_store_, model_checker_->get_remote_process_memory());
141 RemoteApp::~RemoteApp()
143 initial_snapshot_ = nullptr;
144 if (model_checker_) {
146 model_checker_ = nullptr;
147 mc_model_checker = nullptr;
150 void RemoteApp::start()
152 checker_side_->start(
153 [](evutil_socket_t sig, short events, void* arg) {
154 auto checker = static_cast<simgrid::mc::CheckerSide*>(arg);
155 if (events == EV_READ) {
156 std::array<char, MC_MESSAGE_LENGTH> buffer;
157 ssize_t size = recv(checker->get_channel().get_socket(), buffer.data(), buffer.size(), MSG_DONTWAIT);
159 XBT_ERROR("Channel::receive failure: %s", strerror(errno));
161 throw simgrid::xbt::errno_error();
164 if (not mc_model_checker->handle_message(buffer.data(), size))
165 checker->break_loop();
167 xbt_die("Unexpected event");
170 [](evutil_socket_t sig, short events, void* arg) {
171 auto mc = static_cast<simgrid::mc::ModelChecker*>(arg);
172 if (events == EV_SIGNAL) {
174 mc->handle_waitpid();
176 xbt_die("Unexpected signal: %d", sig);
178 xbt_die("Unexpected event");
181 model_checker_.get());
183 XBT_DEBUG("Waiting for the model-checked process");
186 // The model-checked process SIGSTOP itself to signal it's ready:
187 const pid_t pid = get_remote_process_memory().pid();
189 xbt_assert(waitpid(pid, &status, WAITPID_CHECKED_FLAGS) == pid && WIFSTOPPED(status) && WSTOPSIG(status) == SIGSTOP,
190 "Could not wait model-checked process");
194 ptrace(PTRACE_SETOPTIONS, pid, nullptr, PTRACE_O_TRACEEXIT);
195 ptrace(PTRACE_CONT, pid, 0, 0);
197 ptrace(PT_CONTINUE, pid, (caddr_t)1, 0);
199 #error "no ptrace equivalent coded for this platform"
201 xbt_assert(errno == 0,
202 "Ptrace does not seem to be usable in your setup (errno: %d). "
203 "If you run from within a docker, adding `--cap-add SYS_PTRACE` to the docker line may help. "
204 "If it does not help, please report this bug.",
207 void RemoteApp::restore_initial_state() const
209 this->initial_snapshot_->restore(model_checker_->get_remote_process_memory());
212 unsigned long RemoteApp::get_maxpid() const
214 // note: we could maybe cache it and count the actor creation on checker side too.
215 // But counting correctly accross state checkpoint/restore would be annoying.
217 checker_side_->get_channel().send(MessageType::ACTORS_MAXPID);
218 s_mc_message_int_t answer;
219 ssize_t answer_size = checker_side_->get_channel().receive(answer);
220 xbt_assert(answer_size != -1, "Could not receive message");
221 xbt_assert(answer_size == sizeof answer, "Broken message (size=%zd; expected %zu)", answer_size, sizeof answer);
222 xbt_assert(answer.type == MessageType::ACTORS_MAXPID_REPLY,
223 "Received unexpected message %s (%i); expected MessageType::ACTORS_MAXPID_REPLY (%i)",
224 to_c_str(answer.type), (int)answer.type, (int)MessageType::ACTORS_MAXPID_REPLY);
229 void RemoteApp::get_actors_status(std::map<aid_t, ActorState>& whereto) const
231 // The messaging happens as follows:
233 // CheckerSide AppSide
234 // send ACTORS_STATUS ---->
235 // <----- send ACTORS_STATUS_REPLY
236 // <----- send `N` `s_mc_message_actors_status_one_t` structs
237 // <----- send `M` `s_mc_message_simcall_probe_one_t` structs
238 checker_side_->get_channel().send(MessageType::ACTORS_STATUS);
240 s_mc_message_actors_status_answer_t answer;
241 ssize_t answer_size = checker_side_->get_channel().receive(answer);
242 xbt_assert(answer_size != -1, "Could not receive message");
243 xbt_assert(answer_size == sizeof answer, "Broken message (size=%zd; expected %zu)", answer_size, sizeof answer);
244 xbt_assert(answer.type == MessageType::ACTORS_STATUS_REPLY,
245 "Received unexpected message %s (%i); expected MessageType::ACTORS_STATUS_REPLY (%i)",
246 to_c_str(answer.type), (int)answer.type, (int)MessageType::ACTORS_STATUS_REPLY);
248 // Message sanity checks
249 xbt_assert(answer.count >= 0, "Received an ACTOR_STATUS_REPLY message with an actor count of '%d' < 0", answer.count);
250 xbt_assert(answer.transition_count >= 0, "Received an ACTOR_STATUS_REPLY message with transition_count '%d' < 0",
251 answer.transition_count);
252 xbt_assert(answer.transition_count == 0 || answer.count >= 0,
253 "Received an ACTOR_STATUS_REPLY message with no actor data "
254 "but with transition data nonetheless");
256 std::vector<s_mc_message_actors_status_one_t> status(answer.count);
257 if (answer.count > 0) {
258 size_t size = status.size() * sizeof(s_mc_message_actors_status_one_t);
259 ssize_t received = checker_side_->get_channel().receive(status.data(), size);
260 xbt_assert(static_cast<size_t>(received) == size);
263 // Ensures that each actor sends precisely `answer.transition_count` transitions. While technically
264 // this doesn't catch the edge case where actor A sends 3 instead of 2 and actor B sends 2 instead
265 // of 3 transitions, that is ignored here since that invariant needs to be enforced on the AppSide
266 const auto expected_transitions = std::accumulate(
267 status.begin(), status.end(), 0, [](int total, const auto& actor) { return total + actor.n_transitions; });
268 xbt_assert(expected_transitions == answer.transition_count,
269 "Expected to receive %d transition(s) but was only notified of %d by the app side", expected_transitions,
270 answer.transition_count);
272 std::vector<s_mc_message_simcall_probe_one_t> probes(answer.transition_count);
273 if (answer.transition_count > 0) {
274 for (auto& probe : probes) {
275 ssize_t received = checker_side_->get_channel().receive(probe);
276 xbt_assert(received >= 0, "Could not receive response to ACTORS_PROBE message (%s)", strerror(errno));
277 xbt_assert(static_cast<size_t>(received) == sizeof probe,
278 "Could not receive response to ACTORS_PROBE message (%zd bytes received != %zu bytes expected",
279 received, sizeof probe);
284 std::move_iterator probes_iter(probes.begin());
286 for (const auto& actor : status) {
287 xbt_assert(actor.n_transitions == 0 || actor.n_transitions == actor.max_considered,
288 "If any transitions are serialized for an actor, it must match the "
289 "total number of transitions that can be considered for the actor "
290 "(currently %d), but only %d transition(s) was/were said to be encoded",
291 actor.max_considered, actor.n_transitions);
293 std::vector<std::shared_ptr<Transition>> actor_transitions;
294 for (int times_considered = 0; times_considered < actor.n_transitions; times_considered++, probes_iter++) {
295 std::stringstream stream((*probes_iter).buffer.data());
296 actor_transitions.emplace_back(deserialize_transition(actor.aid, times_considered, stream));
299 XBT_DEBUG("Received %zu transitions for actor %ld", actor_transitions.size(), actor.aid);
300 whereto.try_emplace(actor.aid, actor.aid, actor.enabled, actor.max_considered, std::move(actor_transitions));
304 void RemoteApp::check_deadlock() const
306 xbt_assert(checker_side_->get_channel().send(MessageType::DEADLOCK_CHECK) == 0, "Could not check deadlock state");
307 s_mc_message_int_t message;
308 ssize_t received = checker_side_->get_channel().receive(message);
309 xbt_assert(received != -1, "Could not receive message");
310 xbt_assert(received == sizeof message, "Broken message (size=%zd; expected %zu)", received, sizeof message);
311 xbt_assert(message.type == MessageType::DEADLOCK_CHECK_REPLY,
312 "Received unexpected message %s (%i); expected MessageType::DEADLOCK_CHECK_REPLY (%i)",
313 to_c_str(message.type), (int)message.type, (int)MessageType::DEADLOCK_CHECK_REPLY);
315 if (message.value != 0) {
316 XBT_CINFO(mc_global, "Counter-example execution trace:");
317 for (auto const& frame : model_checker_->get_exploration()->get_textual_trace())
318 XBT_CINFO(mc_global, " %s", frame.c_str());
319 XBT_INFO("You can debug the problem (and see the whole details) by rerunning out of simgrid-mc with "
320 "--cfg=model-check/replay:'%s'",
321 model_checker_->get_exploration()->get_record_trace().to_string().c_str());
322 model_checker_->get_exploration()->log_state();
323 throw DeadlockError();
327 void RemoteApp::wait_for_requests()
329 /* Resume the application */
330 if (checker_side_->get_channel().send(MessageType::CONTINUE) != 0)
331 throw xbt::errno_error();
332 get_remote_process_memory().clear_cache();
334 if (this->get_remote_process_memory().running())
335 checker_side_->dispatch();
338 void RemoteApp::shutdown()
340 XBT_DEBUG("Shutting down model-checker");
342 RemoteProcessMemory& process = get_remote_process_memory();
343 if (process.running()) {
344 XBT_DEBUG("Killing process");
346 kill(process.pid(), SIGKILL);
351 Transition* RemoteApp::handle_simcall(aid_t aid, int times_considered, bool new_transition)
353 s_mc_message_simcall_execute_t m = {};
354 m.type = MessageType::SIMCALL_EXECUTE;
356 m.times_considered_ = times_considered;
357 checker_side_->get_channel().send(m);
359 get_remote_process_memory().clear_cache();
360 if (this->get_remote_process_memory().running())
361 checker_side_->dispatch(); // The app may send messages while processing the transition
363 s_mc_message_simcall_execute_answer_t answer;
364 ssize_t s = checker_side_->get_channel().receive(answer);
365 xbt_assert(s != -1, "Could not receive message");
366 xbt_assert(s == sizeof answer, "Broken message (size=%zd; expected %zu)", s, sizeof answer);
367 xbt_assert(answer.type == MessageType::SIMCALL_EXECUTE_ANSWER,
368 "Received unexpected message %s (%i); expected MessageType::SIMCALL_EXECUTE_ANSWER (%i)",
369 to_c_str(answer.type), (int)answer.type, (int)MessageType::SIMCALL_EXECUTE_ANSWER);
371 if (new_transition) {
372 std::stringstream stream(answer.buffer.data());
373 return deserialize_transition(aid, times_considered, stream);
378 void RemoteApp::finalize_app(bool terminate_asap)
380 s_mc_message_int_t m = {};
381 m.type = MessageType::FINALIZE;
382 m.value = terminate_asap;
383 xbt_assert(checker_side_->get_channel().send(m) == 0, "Could not ask the app to finalize on need");
385 s_mc_message_t answer;
386 ssize_t s = checker_side_->get_channel().receive(answer);
387 xbt_assert(s != -1, "Could not receive answer to FINALIZE");
388 xbt_assert(s == sizeof answer, "Broken message (size=%zd; expected %zu)", s, sizeof answer);
389 xbt_assert(answer.type == MessageType::FINALIZE_REPLY,
390 "Received unexpected message %s (%i); expected MessageType::FINALIZE_REPLY (%i)", to_c_str(answer.type),
391 (int)answer.type, (int)MessageType::FINALIZE_REPLY);
394 } // namespace simgrid::mc