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 auto memory = std::make_unique<simgrid::mc::RemoteProcessMemory>(pid);
127 checker_side_ = std::make_unique<simgrid::mc::CheckerSide>(sockets[1], std::move(memory));
131 /* Take the initial snapshot */
133 initial_snapshot_ = std::make_shared<simgrid::mc::Snapshot>(0, page_store_, checker_side_->get_remote_memory());
136 RemoteApp::~RemoteApp()
138 initial_snapshot_ = nullptr;
141 void RemoteApp::start()
143 XBT_DEBUG("Waiting for the model-checked process");
146 // The model-checked process SIGSTOP itself to signal it's ready:
147 const pid_t pid = get_remote_process_memory().pid();
149 xbt_assert(waitpid(pid, &status, WAITPID_CHECKED_FLAGS) == pid && WIFSTOPPED(status) && WSTOPSIG(status) == SIGSTOP,
150 "Could not wait model-checked process");
154 ptrace(PTRACE_SETOPTIONS, pid, nullptr, PTRACE_O_TRACEEXIT);
155 ptrace(PTRACE_CONT, pid, 0, 0);
157 ptrace(PT_CONTINUE, pid, (caddr_t)1, 0);
159 #error "no ptrace equivalent coded for this platform"
161 xbt_assert(errno == 0,
162 "Ptrace does not seem to be usable in your setup (errno: %d). "
163 "If you run from within a docker, adding `--cap-add SYS_PTRACE` to the docker line may help. "
164 "If it does not help, please report this bug.",
167 void RemoteApp::restore_initial_state() const
169 this->initial_snapshot_->restore(checker_side_->get_remote_memory());
172 unsigned long RemoteApp::get_maxpid() const
174 // note: we could maybe cache it and count the actor creation on checker side too.
175 // But counting correctly accross state checkpoint/restore would be annoying.
177 checker_side_->get_channel().send(MessageType::ACTORS_MAXPID);
178 s_mc_message_int_t answer;
179 ssize_t answer_size = checker_side_->get_channel().receive(answer);
180 xbt_assert(answer_size != -1, "Could not receive message");
181 xbt_assert(answer_size == sizeof answer, "Broken message (size=%zd; expected %zu)", answer_size, sizeof answer);
182 xbt_assert(answer.type == MessageType::ACTORS_MAXPID_REPLY,
183 "Received unexpected message %s (%i); expected MessageType::ACTORS_MAXPID_REPLY (%i)",
184 to_c_str(answer.type), (int)answer.type, (int)MessageType::ACTORS_MAXPID_REPLY);
189 void RemoteApp::get_actors_status(std::map<aid_t, ActorState>& whereto) const
191 // The messaging happens as follows:
193 // CheckerSide AppSide
194 // send ACTORS_STATUS ---->
195 // <----- send ACTORS_STATUS_REPLY
196 // <----- send `N` `s_mc_message_actors_status_one_t` structs
197 // <----- send `M` `s_mc_message_simcall_probe_one_t` structs
198 checker_side_->get_channel().send(MessageType::ACTORS_STATUS);
200 s_mc_message_actors_status_answer_t answer;
201 ssize_t answer_size = checker_side_->get_channel().receive(answer);
202 xbt_assert(answer_size != -1, "Could not receive message");
203 xbt_assert(answer_size == sizeof answer, "Broken message (size=%zd; expected %zu)", answer_size, sizeof answer);
204 xbt_assert(answer.type == MessageType::ACTORS_STATUS_REPLY,
205 "Received unexpected message %s (%i); expected MessageType::ACTORS_STATUS_REPLY (%i)",
206 to_c_str(answer.type), (int)answer.type, (int)MessageType::ACTORS_STATUS_REPLY);
208 // Message sanity checks
209 xbt_assert(answer.count >= 0, "Received an ACTOR_STATUS_REPLY message with an actor count of '%d' < 0", answer.count);
210 xbt_assert(answer.transition_count >= 0, "Received an ACTOR_STATUS_REPLY message with transition_count '%d' < 0",
211 answer.transition_count);
212 xbt_assert(answer.transition_count == 0 || answer.count >= 0,
213 "Received an ACTOR_STATUS_REPLY message with no actor data "
214 "but with transition data nonetheless");
216 std::vector<s_mc_message_actors_status_one_t> status(answer.count);
217 if (answer.count > 0) {
218 size_t size = status.size() * sizeof(s_mc_message_actors_status_one_t);
219 ssize_t received = checker_side_->get_channel().receive(status.data(), size);
220 xbt_assert(static_cast<size_t>(received) == size);
223 // Ensures that each actor sends precisely `answer.transition_count` transitions. While technically
224 // this doesn't catch the edge case where actor A sends 3 instead of 2 and actor B sends 2 instead
225 // of 3 transitions, that is ignored here since that invariant needs to be enforced on the AppSide
226 const auto expected_transitions = std::accumulate(
227 status.begin(), status.end(), 0, [](int total, const auto& actor) { return total + actor.n_transitions; });
228 xbt_assert(expected_transitions == answer.transition_count,
229 "Expected to receive %d transition(s) but was only notified of %d by the app side", expected_transitions,
230 answer.transition_count);
232 std::vector<s_mc_message_simcall_probe_one_t> probes(answer.transition_count);
233 if (answer.transition_count > 0) {
234 for (auto& probe : probes) {
235 ssize_t received = checker_side_->get_channel().receive(probe);
236 xbt_assert(received >= 0, "Could not receive response to ACTORS_PROBE message (%s)", strerror(errno));
237 xbt_assert(static_cast<size_t>(received) == sizeof probe,
238 "Could not receive response to ACTORS_PROBE message (%zd bytes received != %zu bytes expected",
239 received, sizeof probe);
244 std::move_iterator probes_iter(probes.begin());
246 for (const auto& actor : status) {
247 xbt_assert(actor.n_transitions == 0 || actor.n_transitions == actor.max_considered,
248 "If any transitions are serialized for an actor, it must match the "
249 "total number of transitions that can be considered for the actor "
250 "(currently %d), but only %d transition(s) was/were said to be encoded",
251 actor.max_considered, actor.n_transitions);
253 std::vector<std::shared_ptr<Transition>> actor_transitions;
254 for (int times_considered = 0; times_considered < actor.n_transitions; times_considered++, probes_iter++) {
255 std::stringstream stream((*probes_iter).buffer.data());
256 actor_transitions.emplace_back(deserialize_transition(actor.aid, times_considered, stream));
259 XBT_DEBUG("Received %zu transitions for actor %ld", actor_transitions.size(), actor.aid);
260 whereto.try_emplace(actor.aid, actor.aid, actor.enabled, actor.max_considered, std::move(actor_transitions));
264 void RemoteApp::check_deadlock() const
266 xbt_assert(checker_side_->get_channel().send(MessageType::DEADLOCK_CHECK) == 0, "Could not check deadlock state");
267 s_mc_message_int_t message;
268 ssize_t received = checker_side_->get_channel().receive(message);
269 xbt_assert(received != -1, "Could not receive message");
270 xbt_assert(received == sizeof message, "Broken message (size=%zd; expected %zu)", received, sizeof message);
271 xbt_assert(message.type == MessageType::DEADLOCK_CHECK_REPLY,
272 "Received unexpected message %s (%i); expected MessageType::DEADLOCK_CHECK_REPLY (%i)",
273 to_c_str(message.type), (int)message.type, (int)MessageType::DEADLOCK_CHECK_REPLY);
275 if (message.value != 0) {
276 auto* explo = Exploration::get_instance();
277 XBT_CINFO(mc_global, "Counter-example execution trace:");
278 for (auto const& frame : explo->get_textual_trace())
279 XBT_CINFO(mc_global, " %s", frame.c_str());
280 XBT_INFO("You can debug the problem (and see the whole details) by rerunning out of simgrid-mc with "
281 "--cfg=model-check/replay:'%s'",
282 explo->get_record_trace().to_string().c_str());
284 throw DeadlockError();
288 void RemoteApp::wait_for_requests()
290 /* Resume the application */
291 if (checker_side_->get_channel().send(MessageType::CONTINUE) != 0)
292 throw xbt::errno_error();
293 get_remote_process_memory().clear_cache();
295 if (this->get_remote_process_memory().running())
296 checker_side_->dispatch_events();
299 void RemoteApp::shutdown()
301 XBT_DEBUG("Shutting down model-checker");
303 RemoteProcessMemory& process = get_remote_process_memory();
304 if (process.running()) {
305 XBT_DEBUG("Killing process");
307 kill(process.pid(), SIGKILL);
312 Transition* RemoteApp::handle_simcall(aid_t aid, int times_considered, bool new_transition)
314 s_mc_message_simcall_execute_t m = {};
315 m.type = MessageType::SIMCALL_EXECUTE;
317 m.times_considered_ = times_considered;
318 checker_side_->get_channel().send(m);
320 get_remote_process_memory().clear_cache();
321 if (this->get_remote_process_memory().running())
322 checker_side_->dispatch_events(); // The app may send messages while processing the transition
324 s_mc_message_simcall_execute_answer_t answer;
325 ssize_t s = checker_side_->get_channel().receive(answer);
326 xbt_assert(s != -1, "Could not receive message");
327 xbt_assert(s == sizeof answer, "Broken message (size=%zd; expected %zu)", s, sizeof answer);
328 xbt_assert(answer.type == MessageType::SIMCALL_EXECUTE_ANSWER,
329 "Received unexpected message %s (%i); expected MessageType::SIMCALL_EXECUTE_ANSWER (%i)",
330 to_c_str(answer.type), (int)answer.type, (int)MessageType::SIMCALL_EXECUTE_ANSWER);
332 if (new_transition) {
333 std::stringstream stream(answer.buffer.data());
334 return deserialize_transition(aid, times_considered, stream);
339 void RemoteApp::finalize_app(bool terminate_asap)
341 s_mc_message_int_t m = {};
342 m.type = MessageType::FINALIZE;
343 m.value = terminate_asap;
344 xbt_assert(checker_side_->get_channel().send(m) == 0, "Could not ask the app to finalize on need");
346 s_mc_message_t answer;
347 ssize_t s = checker_side_->get_channel().receive(answer);
348 xbt_assert(s != -1, "Could not receive answer to FINALIZE");
349 xbt_assert(s == sizeof answer, "Broken message (size=%zd; expected %zu)", s, sizeof answer);
350 xbt_assert(answer.type == MessageType::FINALIZE_REPLY,
351 "Received unexpected message %s (%i); expected MessageType::FINALIZE_REPLY (%i)", to_c_str(answer.type),
352 (int)answer.type, (int)MessageType::FINALIZE_REPLY);
355 } // namespace simgrid::mc