1 /* Copyright (c) 2015-2022. 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/internal_config.h" // HAVE_SMPI
8 #include "src/mc/explo/Exploration.hpp"
9 #include "src/mc/mc_config.hpp"
10 #include "xbt/asserts.h"
12 #include "smpi/smpi.h"
13 #include "src/smpi/include/private.hpp"
16 #include "src/mc/api/State.hpp"
17 #include "src/mc/mc_config.hpp"
18 #include "src/mc/mc_exit.hpp"
19 #include "src/mc/mc_private.hpp"
21 #include "xbt/system_error.hpp"
24 #include <boost/tokenizer.hpp>
30 #include <sys/prctl.h>
33 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(mc_Session, mc, "Model-checker session");
34 XBT_LOG_EXTERNAL_CATEGORY(mc_global);
36 static simgrid::config::Flag<std::string> _sg_mc_setenv{
37 "model-check/setenv", "Extra environment variables to pass to the child process (ex: 'AZE=aze;QWE=qwe').", "",
38 [](std::string_view value) {
39 xbt_assert(value.empty() || value.find('=', 0) != std::string_view::npos,
40 "The 'model-check/setenv' parameter must be like 'AZE=aze', but it does not contain an equal sign.");
43 namespace simgrid::mc {
45 static void run_child_process(int socket, const std::vector<char*>& args)
47 /* On startup, simix_global_init() calls simgrid::mc::Client::initialize(), which checks whether the MC_ENV_SOCKET_FD
48 * env variable is set. If so, MC mode is assumed, and the client is setup from its side
52 // Make sure we do not outlive our parent
55 xbt_assert(sigprocmask(SIG_SETMASK, &mask, nullptr) >= 0, "Could not unblock signals");
56 xbt_assert(prctl(PR_SET_PDEATHSIG, SIGHUP) == 0, "Could not PR_SET_PDEATHSIG");
59 // Remove CLOEXEC to pass the socket to the application
60 int fdflags = fcntl(socket, F_GETFD, 0);
61 xbt_assert(fdflags != -1 && fcntl(socket, F_SETFD, fdflags & ~FD_CLOEXEC) != -1,
62 "Could not remove CLOEXEC for socket");
64 setenv(MC_ENV_SOCKET_FD, std::to_string(socket).c_str(), 1);
66 /* Setup the tokenizer that parses the cfg:model-check/setenv parameter */
67 using Tokenizer = boost::tokenizer<boost::char_separator<char>>;
68 boost::char_separator<char> semicol_sep(";");
69 boost::char_separator<char> equal_sep("=");
70 Tokenizer token_vars(_sg_mc_setenv.get(), semicol_sep); /* Iterate over all FOO=foo parts */
71 for (const auto& token : token_vars) {
72 std::vector<std::string> kv;
73 Tokenizer token_kv(token, equal_sep);
74 for (const auto& t : token_kv) /* Iterate over 'FOO' and then 'foo' in that 'FOO=foo' */
76 xbt_assert(kv.size() == 2, "Parse error on 'model-check/setenv' value %s. Does it contain an equal sign?",
78 XBT_INFO("setenv '%s'='%s'", kv[0].c_str(), kv[1].c_str());
79 setenv(kv[0].c_str(), kv[1].c_str(), 1);
82 /* And now, exec the child process */
84 while (args[i] != nullptr && args[i][0] == '-')
87 xbt_assert(args[i] != nullptr,
88 "Unable to find a binary to exec on the command line. Did you only pass config flags?");
90 execvp(args[i], args.data() + i);
91 XBT_CRITICAL("The model-checked process failed to exec(%s): %s.\n"
92 " Make sure that your binary exists on disk and is executable.",
93 args[i], strerror(errno));
94 if (strchr(args[i], '=') != nullptr)
95 XBT_CRITICAL("If you want to pass environment variables to the application, please use --cfg=model-check/setenv:%s",
98 xbt_die("Aborting now.");
101 RemoteApp::RemoteApp(const std::vector<char*>& args)
104 smpi_init_options(); // only performed once
105 xbt_assert(smpi_cfg_privatization() != SmpiPrivStrategies::MMAP,
106 "Please use the dlopen privatization schema when model-checking SMPI code");
109 // Create an AF_LOCAL socketpair used for exchanging messages
110 // between the model-checker process (ourselves) and the model-checked
113 xbt_assert(socketpair(AF_LOCAL, SOCK_SEQPACKET | SOCK_CLOEXEC, 0, sockets) != -1, "Could not create socketpair");
116 xbt_assert(pid >= 0, "Could not fork model-checked process");
118 if (pid == 0) { // Child
120 run_child_process(sockets[0], args);
124 // Parent (model-checker):
127 xbt_assert(mc_model_checker == nullptr, "Did you manage to start the MC twice in this process?");
129 auto process = std::make_unique<simgrid::mc::RemoteProcess>(pid);
130 model_checker_ = std::make_unique<simgrid::mc::ModelChecker>(std::move(process), sockets[1]);
132 mc_model_checker = model_checker_.get();
133 model_checker_->start();
135 /* Take the initial snapshot */
136 model_checker_->wait_for_requests();
137 initial_snapshot_ = std::make_shared<simgrid::mc::Snapshot>(0);
140 RemoteApp::~RemoteApp()
142 initial_snapshot_ = nullptr;
143 if (model_checker_) {
144 model_checker_->shutdown();
145 model_checker_ = nullptr;
146 mc_model_checker = nullptr;
150 void RemoteApp::restore_initial_state() const
152 this->initial_snapshot_->restore(&model_checker_->get_remote_process());
155 unsigned long RemoteApp::get_maxpid() const
157 return model_checker_->get_remote_process().get_maxpid();
160 void RemoteApp::get_actors_status(std::map<aid_t, ActorState>& whereto) const
163 memset(&msg, 0, sizeof msg);
164 msg.type = simgrid::mc::MessageType::ACTORS_STATUS;
165 model_checker_->channel().send(msg);
167 s_mc_message_actors_status_answer_t answer;
168 ssize_t received = model_checker_->channel().receive(answer);
169 xbt_assert(received != -1, "Could not receive message");
170 xbt_assert(received == sizeof(answer) && answer.type == MessageType::ACTORS_STATUS_REPLY,
171 "Received unexpected message %s (%i, size=%i) "
172 "expected MessageType::ACTORS_STATUS_REPLY (%i, size=%i)",
173 to_c_str(answer.type), (int)answer.type, (int)received, (int)MessageType::ACTORS_STATUS_REPLY,
174 (int)sizeof(answer));
176 std::vector<s_mc_message_actors_status_one_t> status(answer.count);
177 if (answer.count > 0) {
178 size_t size = status.size() * sizeof(s_mc_message_actors_status_one_t);
179 received = model_checker_->channel().receive(status.data(), size);
180 xbt_assert(static_cast<size_t>(received) == size);
184 for (auto const& actor : status)
185 whereto.try_emplace(actor.aid, actor.aid, actor.enabled, actor.max_considered);
188 void RemoteApp::check_deadlock() const
190 xbt_assert(model_checker_->channel().send(MessageType::DEADLOCK_CHECK) == 0, "Could not check deadlock state");
191 s_mc_message_int_t message;
192 ssize_t s = model_checker_->channel().receive(message);
193 xbt_assert(s != -1, "Could not receive message");
194 xbt_assert(s == sizeof(message) && message.type == MessageType::DEADLOCK_CHECK_REPLY,
195 "Received unexpected message %s (%i, size=%i) "
196 "expected MessageType::DEADLOCK_CHECK_REPLY (%i, size=%i)",
197 to_c_str(message.type), (int)message.type, (int)s, (int)MessageType::DEADLOCK_CHECK_REPLY,
198 (int)sizeof(message));
200 if (message.value != 0) {
201 XBT_CINFO(mc_global, "**************************");
202 XBT_CINFO(mc_global, "*** DEADLOCK DETECTED ***");
203 XBT_CINFO(mc_global, "**************************");
204 XBT_CINFO(mc_global, "Counter-example execution trace:");
205 for (auto const& frame : model_checker_->get_exploration()->get_textual_trace())
206 XBT_CINFO(mc_global, " %s", frame.c_str());
207 XBT_INFO("You can debug the problem (and see the whole details) by rerunning out of simgrid-mc with "
208 "--cfg=model-check/replay:'%s'",
209 model_checker_->get_exploration()->get_record_trace().to_string().c_str());
210 model_checker_->get_exploration()->log_state();
211 throw DeadlockError();
214 } // namespace simgrid::mc