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 "xbt/config.hpp"
9 #include "xbt/system_error.hpp"
12 #include "src/mc/explo/LivenessChecker.hpp"
13 #include "src/mc/sosp/RemoteProcessMemory.hpp"
17 #include <sys/prctl.h>
20 #include <boost/tokenizer.hpp>
23 #include <sys/ptrace.h>
27 #define WAITPID_CHECKED_FLAGS __WALL
29 #define WAITPID_CHECKED_FLAGS 0
32 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(mc_checkerside, mc, "MC communication with the application");
34 static simgrid::config::Flag<std::string> _sg_mc_setenv{
35 "model-check/setenv", "Extra environment variables to pass to the child process (ex: 'AZE=aze;QWE=qwe').", "",
36 [](std::string_view value) {
37 xbt_assert(value.empty() || value.find('=', 0) != std::string_view::npos,
38 "The 'model-check/setenv' parameter must be like 'AZE=aze', but it does not contain an equal sign.");
41 namespace simgrid::mc {
43 XBT_ATTRIB_NORETURN static void run_child_process(int socket, const std::vector<char*>& args, bool need_ptrace)
45 /* On startup, simix_global_init() calls simgrid::mc::Client::initialize(), which checks whether the MC_ENV_SOCKET_FD
46 * env variable is set. If so, MC mode is assumed, and the client is setup from its side
50 // Make sure we do not outlive our parent
53 xbt_assert(sigprocmask(SIG_SETMASK, &mask, nullptr) >= 0, "Could not unblock signals");
54 xbt_assert(prctl(PR_SET_PDEATHSIG, SIGHUP) == 0, "Could not PR_SET_PDEATHSIG");
57 // Remove CLOEXEC to pass the socket to the application
58 int fdflags = fcntl(socket, F_GETFD, 0);
59 xbt_assert(fdflags != -1 && fcntl(socket, F_SETFD, fdflags & ~FD_CLOEXEC) != -1,
60 "Could not remove CLOEXEC for socket");
62 setenv(MC_ENV_SOCKET_FD, std::to_string(socket).c_str(), 1);
64 setenv("MC_NEED_PTRACE", "1", 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 static void wait_application_process(pid_t pid)
103 XBT_DEBUG("Waiting for the model-checked process");
106 // The model-checked process SIGSTOP itself to signal it's ready:
107 xbt_assert(waitpid(pid, &status, WAITPID_CHECKED_FLAGS) == pid && WIFSTOPPED(status) && WSTOPSIG(status) == SIGSTOP,
108 "Could not wait model-checked process");
112 ptrace(PTRACE_SETOPTIONS, pid, nullptr, PTRACE_O_TRACEEXIT);
113 ptrace(PTRACE_CONT, pid, 0, 0);
115 ptrace(PT_CONTINUE, pid, (caddr_t)1, 0);
117 #error "no ptrace equivalent coded for this platform"
119 xbt_assert(errno == 0,
120 "Ptrace does not seem to be usable in your setup (errno: %d). "
121 "If you run from within a docker, adding `--cap-add SYS_PTRACE` to the docker line may help. "
122 "If it does not help, please report this bug.",
126 void CheckerSide::setup_events(bool socket_only)
128 auto* base = event_base_new();
131 socket_event_ = event_new(
132 base, get_channel().get_socket(), EV_READ | EV_PERSIST,
133 [](evutil_socket_t, short events, void* arg) {
134 auto checker = static_cast<simgrid::mc::CheckerSide*>(arg);
135 if (events == EV_READ) {
136 std::array<char, MC_MESSAGE_LENGTH> buffer;
137 ssize_t size = recv(checker->get_channel().get_socket(), buffer.data(), buffer.size(), MSG_DONTWAIT);
139 XBT_ERROR("Channel::receive failure: %s", strerror(errno));
141 throw simgrid::xbt::errno_error();
144 if (size == 0) // The app closed the socket. It must be dead by now.
145 checker->handle_waitpid();
146 else if (not checker->handle_message(buffer.data(), size))
147 checker->break_loop();
149 xbt_die("Unexpected event");
153 event_add(socket_event_, nullptr);
156 signal_event_ = nullptr;
158 signal_event_ = event_new(
159 base, SIGCHLD, EV_SIGNAL | EV_PERSIST,
160 [](evutil_socket_t sig, short events, void* arg) {
161 auto checker = static_cast<simgrid::mc::CheckerSide*>(arg);
162 if (events == EV_SIGNAL) {
164 checker->handle_waitpid();
166 xbt_die("Unexpected signal: %d", sig);
168 xbt_die("Unexpected event");
172 event_add(signal_event_, nullptr);
176 /* When this constructor is called, no other checkerside exists */
177 CheckerSide::CheckerSide(const std::vector<char*>& args, bool need_memory_info) : running_(true)
179 // Create an AF_LOCAL socketpair used for exchanging messages between the model-checker process (ancestor)
180 // and the application process (child)
182 xbt_assert(socketpair(AF_LOCAL, SOCK_SEQPACKET | SOCK_CLOEXEC, 0, sockets) != -1, "Could not create socketpair: %s",
186 xbt_assert(pid_ >= 0, "Could not fork application process");
188 if (pid_ == 0) { // Child
190 run_child_process(sockets[0], args, need_memory_info); // We need ptrace if we need the mem info
194 // Parent (model-checker):
196 channel_.reset_socket(sockets[1]);
198 setup_events(false); /* we need a signal handler too */
199 if (need_memory_info) {
201 // setup ptrace and sync with the app
202 wait_application_process(pid_);
204 // Request the initial memory on need
205 channel_.send(MessageType::NEED_MEMINFO);
206 s_mc_message_need_meminfo_reply_t answer;
207 ssize_t answer_size = channel_.receive(answer);
208 xbt_assert(answer_size != -1, "Could not receive message");
209 xbt_assert(answer.type == MessageType::NEED_MEMINFO_REPLY,
210 "The received message is not the NEED_MEMINFO_REPLY I was expecting but of type %s",
211 to_c_str(answer.type));
212 xbt_assert(answer_size == sizeof answer, "Broken message (size=%zd; expected %zu)", answer_size, sizeof answer);
214 /* We now have enough info to create the memory address space */
215 remote_memory_ = std::make_unique<simgrid::mc::RemoteProcessMemory>(pid_, answer.mmalloc_default_mdp);
217 xbt_die("Cannot introspect memory without MC support");
224 CheckerSide::~CheckerSide()
226 event_del(socket_event_);
227 event_free(socket_event_);
228 if (signal_event_ != nullptr) {
229 event_del(signal_event_);
230 event_free(signal_event_);
234 /* This constructor is called when cloning a checkerside to get its application to fork away */
235 CheckerSide::CheckerSide(int socket, CheckerSide* child_checker)
236 : channel_(socket), running_(true), child_checker_(child_checker)
238 setup_events(true); // We already have a signal handled in that case
240 s_mc_message_int_t answer;
241 ssize_t s = get_channel().receive(answer);
242 xbt_assert(s != -1, "Could not receive answer to FORK_REPLY");
243 xbt_assert(s == sizeof answer, "Broken message (size=%zd; expected %zu)", s, sizeof answer);
244 xbt_assert(answer.type == MessageType::FORK_REPLY,
245 "Received unexpected message %s (%i); expected MessageType::FORK_REPLY (%i)", to_c_str(answer.type),
246 (int)answer.type, (int)MessageType::FORK_REPLY);
252 std::unique_ptr<CheckerSide> CheckerSide::clone(int master_socket)
254 s_mc_message_int_t m = {};
255 m.type = MessageType::FORK;
257 xbt_assert(get_channel().send(m) == 0, "Could not ask the app to fork on need.");
259 int sock = accept(master_socket, nullptr /* I know who's connecting*/, nullptr);
260 xbt_assert(sock > 0, "Cannot accept the incomming connection of the forked app: %s.", strerror(errno));
262 return std::make_unique<CheckerSide>(sock, this);
265 void CheckerSide::finalize(bool terminate_asap)
267 s_mc_message_int_t m = {};
268 m.type = MessageType::FINALIZE;
269 m.value = terminate_asap;
270 xbt_assert(get_channel().send(m) == 0, "Could not ask the app to finalize on need");
272 s_mc_message_t answer;
273 ssize_t s = get_channel().receive(answer);
274 xbt_assert(s != -1, "Could not receive answer to FINALIZE");
275 xbt_assert(s == sizeof answer, "Broken message (size=%zd; expected %zu)", s, sizeof answer);
276 xbt_assert(answer.type == MessageType::FINALIZE_REPLY,
277 "Received unexpected message %s (%i); expected MessageType::FINALIZE_REPLY (%i)", to_c_str(answer.type),
278 (int)answer.type, (int)MessageType::FINALIZE_REPLY);
281 void CheckerSide::dispatch_events() const
283 event_base_dispatch(base_.get());
286 void CheckerSide::break_loop() const
288 event_base_loopbreak(base_.get());
291 bool CheckerSide::handle_message(const char* buffer, ssize_t size)
293 s_mc_message_t base_message;
294 xbt_assert(size >= (ssize_t)sizeof(base_message), "Broken message. Got only %ld bytes.", size);
295 memcpy(&base_message, buffer, sizeof(base_message));
297 switch (base_message.type) {
298 case MessageType::IGNORE_HEAP: {
300 if (remote_memory_ != nullptr) {
301 s_mc_message_ignore_heap_t message;
302 xbt_assert(size == sizeof(message), "Broken message");
303 memcpy(&message, buffer, sizeof(message));
305 IgnoredHeapRegion region;
306 region.block = message.block;
307 region.fragment = message.fragment;
308 region.address = message.address;
309 region.size = message.size;
310 get_remote_memory()->ignore_heap(region);
313 XBT_INFO("Ignoring a IGNORE_HEAP message because we don't need to introspect memory.");
317 case MessageType::UNIGNORE_HEAP: {
319 if (remote_memory_ != nullptr) {
320 s_mc_message_ignore_memory_t message;
321 xbt_assert(size == sizeof(message), "Broken message");
322 memcpy(&message, buffer, sizeof(message));
323 get_remote_memory()->unignore_heap((void*)message.addr, message.size);
326 XBT_INFO("Ignoring an UNIGNORE_HEAP message because we don't need to introspect memory.");
330 case MessageType::IGNORE_MEMORY: {
332 if (remote_memory_ != nullptr) {
333 s_mc_message_ignore_memory_t message;
334 xbt_assert(size == sizeof(message), "Broken message");
335 memcpy(&message, buffer, sizeof(message));
336 get_remote_memory()->ignore_region(message.addr, message.size);
339 XBT_INFO("Ignoring an IGNORE_MEMORY message because we don't need to introspect memory.");
343 case MessageType::STACK_REGION: {
345 if (remote_memory_ != nullptr) {
346 s_mc_message_stack_region_t message;
347 xbt_assert(size == sizeof(message), "Broken message");
348 memcpy(&message, buffer, sizeof(message));
349 get_remote_memory()->stack_areas().push_back(message.stack_region);
352 XBT_INFO("Ignoring an STACK_REGION message because we don't need to introspect memory.");
356 case MessageType::REGISTER_SYMBOL: {
358 s_mc_message_register_symbol_t message;
359 xbt_assert(size == sizeof(message), "Broken message");
360 memcpy(&message, buffer, sizeof(message));
361 xbt_assert(not message.callback, "Support for client-side function proposition is not implemented.");
362 XBT_DEBUG("Received symbol: %s", message.name.data());
364 LivenessChecker::automaton_register_symbol(*get_remote_memory(), message.name.data(), remote((int*)message.data));
366 xbt_die("Please don't use liveness properties when MC is compiled out.");
371 case MessageType::WAITING:
374 case MessageType::ASSERTION_FAILED:
375 Exploration::get_instance()->report_assertion_failure();
379 xbt_die("Unexpected message from the application");
384 void CheckerSide::wait_for_requests()
386 XBT_DEBUG("Resume the application");
387 if (get_channel().send(MessageType::CONTINUE) != 0)
388 throw xbt::errno_error();
389 clear_memory_cache();
395 void CheckerSide::clear_memory_cache()
399 remote_memory_->clear_cache();
403 void CheckerSide::handle_dead_child(int status)
405 // From PTRACE_O_TRACEEXIT:
407 if (status >> 8 == (SIGTRAP | (PTRACE_EVENT_EXIT << 8))) {
408 unsigned long eventmsg;
409 xbt_assert(ptrace(PTRACE_GETEVENTMSG, pid_, 0, &eventmsg) != -1, "Could not get exit status");
410 status = static_cast<int>(eventmsg);
411 if (WIFSIGNALED(status)) {
413 Exploration::get_instance()->report_crash(status);
418 // We don't care about non-lethal signals, just reinject them:
419 if (WIFSTOPPED(status)) {
420 XBT_DEBUG("Stopped with signal %i", (int)WSTOPSIG(status));
423 ptrace(PTRACE_CONT, pid_, 0, WSTOPSIG(status));
425 ptrace(PT_CONTINUE, pid_, (caddr_t)1, WSTOPSIG(status));
427 xbt_assert(errno == 0, "Could not PTRACE_CONT: %s", strerror(errno));
430 else if (WIFSIGNALED(status)) {
432 Exploration::get_instance()->report_crash(status);
433 } else if (WIFEXITED(status)) {
434 XBT_DEBUG("Child process is over");
439 void CheckerSide::handle_waitpid()
441 XBT_DEBUG("Check for wait event");
443 if (child_checker_ == nullptr) { // Wait directly
446 while ((pid = waitpid(-1, &status, WNOHANG)) != 0) {
448 if (errno == ECHILD) { // No more children:
449 xbt_assert(not this->running(), "Inconsistent state");
452 xbt_die("Could not wait for pid: %s", strerror(errno));
456 if (pid == get_pid())
457 handle_dead_child(status);
460 } else { // Ask our proxy to wait for us
462 s_mc_message_int_t request = {};
463 request.type = MessageType::WAIT_CHILD;
464 request.value = pid_;
465 xbt_assert(child_checker_->get_channel().send(request) == 0,
466 "Could not ask my child to waitpid its child for me: %s", strerror(errno));
468 s_mc_message_int_t answer;
469 ssize_t answer_size = child_checker_->get_channel().receive(answer);
470 xbt_assert(answer_size != -1, "Could not receive message");
471 xbt_assert(answer.type == MessageType::WAIT_CHILD_REPLY,
472 "The received message is not the WAIT_CHILD_REPLY I was expecting but of type %s",
473 to_c_str(answer.type));
474 xbt_assert(answer_size == sizeof answer, "Broken message (size=%zd; expected %zu)", answer_size, sizeof answer);
475 handle_dead_child(answer.value);
478 } // namespace simgrid::mc