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