1 /* Copyright (c) 2007-2023. 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/CheckerSide.hpp"
7 #include "src/mc/explo/Exploration.hpp"
8 #include "src/mc/explo/LivenessChecker.hpp"
9 #include "src/mc/sosp/RemoteProcessMemory.hpp"
10 #include "xbt/system_error.hpp"
13 #include <sys/prctl.h>
16 #include <boost/tokenizer.hpp>
19 #include <sys/ptrace.h>
23 #define WAITPID_CHECKED_FLAGS __WALL
25 #define WAITPID_CHECKED_FLAGS 0
28 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(mc_checkerside, mc, "MC communication with the application");
30 static simgrid::config::Flag<std::string> _sg_mc_setenv{
31 "model-check/setenv", "Extra environment variables to pass to the child process (ex: 'AZE=aze;QWE=qwe').", "",
32 [](std::string_view value) {
33 xbt_assert(value.empty() || value.find('=', 0) != std::string_view::npos,
34 "The 'model-check/setenv' parameter must be like 'AZE=aze', but it does not contain an equal sign.");
37 namespace simgrid::mc {
39 XBT_ATTRIB_NORETURN static void run_child_process(int socket, const std::vector<char*>& args, bool need_ptrace)
41 /* On startup, simix_global_init() calls simgrid::mc::Client::initialize(), which checks whether the MC_ENV_SOCKET_FD
42 * env variable is set. If so, MC mode is assumed, and the client is setup from its side
46 // Make sure we do not outlive our parent
49 xbt_assert(sigprocmask(SIG_SETMASK, &mask, nullptr) >= 0, "Could not unblock signals");
50 xbt_assert(prctl(PR_SET_PDEATHSIG, SIGHUP) == 0, "Could not PR_SET_PDEATHSIG");
53 // Remove CLOEXEC to pass the socket to the application
54 int fdflags = fcntl(socket, F_GETFD, 0);
55 xbt_assert(fdflags != -1 && fcntl(socket, F_SETFD, fdflags & ~FD_CLOEXEC) != -1,
56 "Could not remove CLOEXEC for socket");
58 setenv(MC_ENV_SOCKET_FD, std::to_string(socket).c_str(), 1);
60 setenv("MC_NEED_PTRACE", "1", 1);
62 /* Setup the tokenizer that parses the cfg:model-check/setenv parameter */
63 using Tokenizer = boost::tokenizer<boost::char_separator<char>>;
64 boost::char_separator<char> semicol_sep(";");
65 boost::char_separator<char> equal_sep("=");
66 Tokenizer token_vars(_sg_mc_setenv.get(), semicol_sep); /* Iterate over all FOO=foo parts */
67 for (const auto& token : token_vars) {
68 std::vector<std::string> kv;
69 Tokenizer token_kv(token, equal_sep);
70 for (const auto& t : token_kv) /* Iterate over 'FOO' and then 'foo' in that 'FOO=foo' */
72 xbt_assert(kv.size() == 2, "Parse error on 'model-check/setenv' value %s. Does it contain an equal sign?",
74 XBT_INFO("setenv '%s'='%s'", kv[0].c_str(), kv[1].c_str());
75 setenv(kv[0].c_str(), kv[1].c_str(), 1);
78 /* And now, exec the child process */
80 while (args[i] != nullptr && args[i][0] == '-')
83 xbt_assert(args[i] != nullptr,
84 "Unable to find a binary to exec on the command line. Did you only pass config flags?");
86 execvp(args[i], args.data() + i);
87 XBT_CRITICAL("The model-checked process failed to exec(%s): %s.\n"
88 " Make sure that your binary exists on disk and is executable.",
89 args[i], strerror(errno));
90 if (strchr(args[i], '=') != nullptr)
91 XBT_CRITICAL("If you want to pass environment variables to the application, please use --cfg=model-check/setenv:%s",
94 xbt_die("Aborting now.");
97 static void wait_application_process(pid_t pid)
99 XBT_DEBUG("Waiting for the model-checked process");
102 // The model-checked process SIGSTOP itself to signal it's ready:
103 xbt_assert(waitpid(pid, &status, WAITPID_CHECKED_FLAGS) == pid && WIFSTOPPED(status) && WSTOPSIG(status) == SIGSTOP,
104 "Could not wait model-checked process");
108 ptrace(PTRACE_SETOPTIONS, pid, nullptr, PTRACE_O_TRACEEXIT);
109 ptrace(PTRACE_CONT, pid, 0, 0);
111 ptrace(PT_CONTINUE, pid, (caddr_t)1, 0);
113 #error "no ptrace equivalent coded for this platform"
115 xbt_assert(errno == 0,
116 "Ptrace does not seem to be usable in your setup (errno: %d). "
117 "If you run from within a docker, adding `--cap-add SYS_PTRACE` to the docker line may help. "
118 "If it does not help, please report this bug.",
122 void CheckerSide::setup_events(bool socket_only)
124 auto* base = event_base_new();
127 socket_event_ = event_new(
128 base, get_channel().get_socket(), EV_READ | EV_PERSIST,
129 [](evutil_socket_t, short events, void* arg) {
130 auto checker = static_cast<simgrid::mc::CheckerSide*>(arg);
131 if (events == EV_READ) {
132 std::array<char, MC_MESSAGE_LENGTH> buffer;
133 ssize_t size = recv(checker->get_channel().get_socket(), buffer.data(), buffer.size(), MSG_DONTWAIT);
135 XBT_ERROR("Channel::receive failure: %s", strerror(errno));
137 throw simgrid::xbt::errno_error();
140 if (size == 0) // The app closed the socket. It must be dead by now.
141 checker->handle_waitpid();
142 else if (not checker->handle_message(buffer.data(), size))
143 checker->break_loop();
145 xbt_die("Unexpected event");
149 event_add(socket_event_, nullptr);
152 signal_event_ = nullptr;
154 signal_event_ = event_new(
155 base, SIGCHLD, EV_SIGNAL | EV_PERSIST,
156 [](evutil_socket_t sig, short events, void* arg) {
157 auto checker = static_cast<simgrid::mc::CheckerSide*>(arg);
158 if (events == EV_SIGNAL) {
160 checker->handle_waitpid();
162 xbt_die("Unexpected signal: %d", sig);
164 xbt_die("Unexpected event");
168 event_add(signal_event_, nullptr);
172 /* When this constructor is called, no other checkerside exists */
173 CheckerSide::CheckerSide(const std::vector<char*>& args, bool need_memory_info) : running_(true)
175 // Create an AF_LOCAL socketpair used for exchanging messages between the model-checker process (ancestor)
176 // and the application process (child)
178 xbt_assert(socketpair(AF_LOCAL, SOCK_SEQPACKET | SOCK_CLOEXEC, 0, sockets) != -1, "Could not create socketpair: %s",
182 xbt_assert(pid_ >= 0, "Could not fork application process");
184 if (pid_ == 0) { // Child
186 run_child_process(sockets[0], args, need_memory_info); // We need ptrace if we need the mem info
190 // Parent (model-checker):
192 channel_.reset_socket(sockets[1]);
194 setup_events(false); /* we need a signal handler too */
195 if (need_memory_info) {
196 // setup ptrace and sync with the app
197 wait_application_process(pid_);
199 // Request the initial memory on need
200 channel_.send(MessageType::NEED_MEMINFO);
201 s_mc_message_need_meminfo_reply_t answer;
202 ssize_t answer_size = channel_.receive(answer);
203 xbt_assert(answer_size != -1, "Could not receive message");
204 xbt_assert(answer.type == MessageType::NEED_MEMINFO_REPLY,
205 "The received message is not the NEED_MEMINFO_REPLY I was expecting but of type %s",
206 to_c_str(answer.type));
207 xbt_assert(answer_size == sizeof answer, "Broken message (size=%zd; expected %zu)", answer_size, sizeof answer);
209 /* We now have enough info to create the memory address space */
210 remote_memory_ = std::make_unique<simgrid::mc::RemoteProcessMemory>(pid_, answer.mmalloc_default_mdp);
216 CheckerSide::~CheckerSide()
218 event_del(socket_event_);
219 event_free(socket_event_);
220 if (signal_event_ != nullptr) {
221 event_del(signal_event_);
222 event_free(signal_event_);
226 /* This constructor is called when cloning a checkerside to get its application to fork away */
227 CheckerSide::CheckerSide(int socket, CheckerSide* child_checker)
228 : channel_(socket), running_(true), child_checker_(child_checker)
230 setup_events(true); // We already have a signal handled in that case
232 s_mc_message_int_t answer;
233 ssize_t s = get_channel().receive(answer);
234 xbt_assert(s != -1, "Could not receive answer to FORK_REPLY");
235 xbt_assert(s == sizeof answer, "Broken message (size=%zd; expected %zu)", s, sizeof answer);
236 xbt_assert(answer.type == MessageType::FORK_REPLY,
237 "Received unexpected message %s (%i); expected MessageType::FORK_REPLY (%i)", to_c_str(answer.type),
238 (int)answer.type, (int)MessageType::FORK_REPLY);
244 std::unique_ptr<CheckerSide> CheckerSide::clone(int master_socket)
246 s_mc_message_int_t m = {};
247 m.type = MessageType::FORK;
249 xbt_assert(get_channel().send(m) == 0, "Could not ask the app to fork on need.");
251 int sock = accept(master_socket, nullptr /* I know who's connecting*/, nullptr);
252 xbt_assert(sock > 0, "Cannot accept the incomming connection of the forked app: %s.", strerror(errno));
254 return std::make_unique<CheckerSide>(sock, this);
257 void CheckerSide::finalize(bool terminate_asap)
259 s_mc_message_int_t m = {};
260 m.type = MessageType::FINALIZE;
261 m.value = terminate_asap;
262 xbt_assert(get_channel().send(m) == 0, "Could not ask the app to finalize on need");
264 s_mc_message_t answer;
265 ssize_t s = get_channel().receive(answer);
266 xbt_assert(s != -1, "Could not receive answer to FINALIZE");
267 xbt_assert(s == sizeof answer, "Broken message (size=%zd; expected %zu)", s, sizeof answer);
268 xbt_assert(answer.type == MessageType::FINALIZE_REPLY,
269 "Received unexpected message %s (%i); expected MessageType::FINALIZE_REPLY (%i)", to_c_str(answer.type),
270 (int)answer.type, (int)MessageType::FINALIZE_REPLY);
273 void CheckerSide::dispatch_events() const
275 event_base_dispatch(base_.get());
278 void CheckerSide::break_loop() const
280 event_base_loopbreak(base_.get());
283 bool CheckerSide::handle_message(const char* buffer, ssize_t size)
285 s_mc_message_t base_message;
286 xbt_assert(size >= (ssize_t)sizeof(base_message), "Broken message. Got only %ld bytes.", size);
287 memcpy(&base_message, buffer, sizeof(base_message));
289 switch (base_message.type) {
290 case MessageType::IGNORE_HEAP: {
291 if (remote_memory_ != nullptr) {
292 s_mc_message_ignore_heap_t message;
293 xbt_assert(size == sizeof(message), "Broken message");
294 memcpy(&message, buffer, sizeof(message));
296 IgnoredHeapRegion region;
297 region.block = message.block;
298 region.fragment = message.fragment;
299 region.address = message.address;
300 region.size = message.size;
301 get_remote_memory()->ignore_heap(region);
303 XBT_INFO("Ignoring a IGNORE_HEAP message because we don't need to introspect memory.");
308 case MessageType::UNIGNORE_HEAP: {
309 if (remote_memory_ != nullptr) {
310 s_mc_message_ignore_memory_t message;
311 xbt_assert(size == sizeof(message), "Broken message");
312 memcpy(&message, buffer, sizeof(message));
313 get_remote_memory()->unignore_heap((void*)message.addr, message.size);
315 XBT_INFO("Ignoring an UNIGNORE_HEAP message because we don't need to introspect memory.");
320 case MessageType::IGNORE_MEMORY: {
321 if (remote_memory_ != nullptr) {
322 s_mc_message_ignore_memory_t message;
323 xbt_assert(size == sizeof(message), "Broken message");
324 memcpy(&message, buffer, sizeof(message));
325 get_remote_memory()->ignore_region(message.addr, message.size);
327 XBT_INFO("Ignoring an IGNORE_MEMORY message because we don't need to introspect memory.");
332 case MessageType::STACK_REGION: {
333 if (remote_memory_ != nullptr) {
334 s_mc_message_stack_region_t message;
335 xbt_assert(size == sizeof(message), "Broken message");
336 memcpy(&message, buffer, sizeof(message));
337 get_remote_memory()->stack_areas().push_back(message.stack_region);
339 XBT_INFO("Ignoring an STACK_REGION message because we don't need to introspect memory.");
344 case MessageType::REGISTER_SYMBOL: {
345 s_mc_message_register_symbol_t message;
346 xbt_assert(size == sizeof(message), "Broken message");
347 memcpy(&message, buffer, sizeof(message));
348 xbt_assert(not message.callback, "Support for client-side function proposition is not implemented.");
349 XBT_DEBUG("Received symbol: %s", message.name.data());
351 LivenessChecker::automaton_register_symbol(*get_remote_memory(), message.name.data(), remote((int*)message.data));
355 case MessageType::WAITING:
358 case MessageType::ASSERTION_FAILED:
359 Exploration::get_instance()->report_assertion_failure();
363 xbt_die("Unexpected message from the application");
368 void CheckerSide::wait_for_requests()
370 XBT_DEBUG("Resume the application");
371 if (get_channel().send(MessageType::CONTINUE) != 0)
372 throw xbt::errno_error();
373 clear_memory_cache();
379 void CheckerSide::clear_memory_cache()
382 remote_memory_->clear_cache();
385 void CheckerSide::handle_dead_child(int status)
387 // From PTRACE_O_TRACEEXIT:
389 if (status >> 8 == (SIGTRAP | (PTRACE_EVENT_EXIT << 8))) {
390 unsigned long eventmsg;
391 xbt_assert(ptrace(PTRACE_GETEVENTMSG, pid_, 0, &eventmsg) != -1, "Could not get exit status");
392 status = static_cast<int>(eventmsg);
393 if (WIFSIGNALED(status)) {
395 Exploration::get_instance()->report_crash(status);
400 // We don't care about non-lethal signals, just reinject them:
401 if (WIFSTOPPED(status)) {
402 XBT_DEBUG("Stopped with signal %i", (int)WSTOPSIG(status));
405 ptrace(PTRACE_CONT, pid_, 0, WSTOPSIG(status));
407 ptrace(PT_CONTINUE, pid_, (caddr_t)1, WSTOPSIG(status));
409 xbt_assert(errno == 0, "Could not PTRACE_CONT: %s", strerror(errno));
412 else if (WIFSIGNALED(status)) {
414 Exploration::get_instance()->report_crash(status);
415 } else if (WIFEXITED(status)) {
416 XBT_DEBUG("Child process is over");
421 void CheckerSide::handle_waitpid()
423 XBT_DEBUG("Check for wait event");
425 if (child_checker_ == nullptr) { // Wait directly
428 while ((pid = waitpid(-1, &status, WNOHANG)) != 0) {
430 if (errno == ECHILD) { // No more children:
431 xbt_assert(not this->running(), "Inconsistent state");
434 xbt_die("Could not wait for pid: %s", strerror(errno));
438 if (pid == get_pid())
439 handle_dead_child(status);
442 } else { // Ask our proxy to wait for us
444 s_mc_message_int_t request = {};
445 request.type = MessageType::WAIT_CHILD;
446 request.value = pid_;
447 xbt_assert(child_checker_->get_channel().send(request) == 0,
448 "Could not ask my child to waitpid its child for me: %s", strerror(errno));
450 s_mc_message_int_t answer;
451 ssize_t answer_size = child_checker_->get_channel().receive(answer);
452 xbt_assert(answer_size != -1, "Could not receive message");
453 xbt_assert(answer.type == MessageType::WAIT_CHILD_REPLY,
454 "The received message is not the WAIT_CHILD_REPLY I was expecting but of type %s",
455 to_c_str(answer.type));
456 xbt_assert(answer_size == sizeof answer, "Broken message (size=%zd; expected %zu)", answer_size, sizeof answer);
457 handle_dead_child(answer.value);
460 } // namespace simgrid::mc