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/mc_environ.h"
9 #include "xbt/config.hpp"
10 #include "xbt/system_error.hpp"
12 #if SIMGRID_HAVE_STATEFUL_MC
13 #include "src/mc/explo/LivenessChecker.hpp"
14 #include "src/mc/sosp/RemoteProcessMemory.hpp"
18 #include <sys/prctl.h>
21 #include <boost/tokenizer.hpp>
24 #include <sys/ptrace.h>
28 #define WAITPID_CHECKED_FLAGS __WALL
30 #define WAITPID_CHECKED_FLAGS 0
33 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(mc_checkerside, mc, "MC communication with the application");
35 static simgrid::config::Flag<std::string> _sg_mc_setenv{
36 "model-check/setenv", "Extra environment variables to pass to the child process (ex: 'AZE=aze;QWE=qwe').", "",
37 [](std::string_view value) {
38 xbt_assert(value.empty() || value.find('=', 0) != std::string_view::npos,
39 "The 'model-check/setenv' parameter must be like 'AZE=aze', but it does not contain an equal sign.");
42 namespace simgrid::mc {
44 XBT_ATTRIB_NORETURN static void run_child_process(int socket, const std::vector<char*>& args, bool need_ptrace)
46 /* On startup, simix_global_init() calls simgrid::mc::Client::initialize(), which checks whether the MC_ENV_SOCKET_FD
47 * env variable is set. If so, MC mode is assumed, and the client is setup from its side
51 // Make sure we do not outlive our parent
54 xbt_assert(sigprocmask(SIG_SETMASK, &mask, nullptr) >= 0, "Could not unblock signals");
55 xbt_assert(prctl(PR_SET_PDEATHSIG, SIGHUP) == 0, "Could not PR_SET_PDEATHSIG");
58 setenv(MC_ENV_SOCKET_FD, std::to_string(socket).c_str(), 1);
60 setenv(MC_ENV_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 xbt_die("no ptrace equivalent coded for this platform, stateful model-checking is impossible.");
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.",
120 XBT_DEBUG("%d ptrace correctly setup.", getpid());
123 void CheckerSide::setup_events(bool socket_only)
125 auto* base = event_base_new();
128 socket_event_ = event_new(
129 base, get_channel().get_socket(), EV_READ | EV_PERSIST,
130 [](evutil_socket_t, short events, void* arg) {
131 auto checker = static_cast<simgrid::mc::CheckerSide*>(arg);
132 if (events == EV_READ) {
134 std::array<char, MC_MESSAGE_LENGTH> buffer;
135 ssize_t size = checker->get_channel().receive(buffer.data(), buffer.size(), MSG_DONTWAIT);
137 XBT_ERROR("Channel::receive failure: %s", strerror(errno));
139 throw simgrid::xbt::errno_error();
142 if (size == 0) // The app closed the socket. It must be dead by now.
143 checker->handle_waitpid();
144 else if (not checker->handle_message(buffer.data(), size)) {
145 checker->break_loop();
148 } while (checker->get_channel().has_pending_data());
150 xbt_die("Unexpected event");
154 event_add(socket_event_, nullptr);
157 signal_event_ = nullptr;
159 signal_event_ = event_new(
160 base, SIGCHLD, EV_SIGNAL | EV_PERSIST,
161 [](evutil_socket_t sig, short events, void* arg) {
162 auto checker = static_cast<simgrid::mc::CheckerSide*>(arg);
163 if (events == EV_SIGNAL) {
165 checker->handle_waitpid();
167 xbt_die("Unexpected signal: %d", sig);
169 xbt_die("Unexpected event");
173 event_add(signal_event_, nullptr);
177 /* When this constructor is called, no other checkerside exists */
178 CheckerSide::CheckerSide(const std::vector<char*>& args, bool need_memory_info) : running_(true)
180 XBT_DEBUG("Create a CheckerSide. Needs_meminfo: %s", need_memory_info ? "YES" : "no");
182 // Create an AF_UNIX socketpair used for exchanging messages between the model-checker process (ancestor)
183 // and the application process (child)
185 xbt_assert(socketpair(AF_UNIX,
187 SOCK_STREAM, /* Mac OSX does not have AF_UNIX + SOCK_SEQPACKET, even if that's faster */
192 "Could not create socketpair: %s", strerror(errno));
195 xbt_assert(pid_ >= 0, "Could not fork application process");
197 if (pid_ == 0) { // Child
199 run_child_process(sockets[0], args, need_memory_info); // We need ptrace if we need the mem info
203 // Parent (model-checker):
205 channel_.reset_socket(sockets[1]);
207 setup_events(false); /* we need a signal handler too */
208 if (need_memory_info) {
209 #if SIMGRID_HAVE_STATEFUL_MC
210 // setup ptrace and sync with the app
211 wait_application_process(pid_);
213 // Request the initial memory on need
214 channel_.send(MessageType::NEED_MEMINFO);
215 s_mc_message_need_meminfo_reply_t answer;
216 ssize_t answer_size = channel_.receive(answer);
217 xbt_assert(answer_size != -1, "Could not receive message");
218 xbt_assert(answer.type == MessageType::NEED_MEMINFO_REPLY,
219 "The received message is not the NEED_MEMINFO_REPLY I was expecting but of type %s",
220 to_c_str(answer.type));
221 xbt_assert(answer_size == sizeof answer, "Broken message (size=%zd; expected %zu)", answer_size, sizeof answer);
223 /* We now have enough info to create the memory address space */
224 remote_memory_ = std::make_unique<simgrid::mc::RemoteProcessMemory>(pid_, answer.mmalloc_default_mdp);
226 xbt_die("Cannot introspect memory without MC support");
233 CheckerSide::~CheckerSide()
235 event_del(socket_event_);
236 event_free(socket_event_);
237 if (signal_event_ != nullptr) {
238 event_del(signal_event_);
239 event_free(signal_event_);
243 /* This constructor is called when cloning a checkerside to get its application to fork away */
244 CheckerSide::CheckerSide(int socket, CheckerSide* child_checker)
245 : channel_(socket, child_checker->channel_), running_(true), child_checker_(child_checker)
247 setup_events(true); // We already have a signal handled in that case
249 s_mc_message_int_t answer;
250 ssize_t s = get_channel().receive(answer);
251 xbt_assert(s != -1, "Could not receive answer to FORK_REPLY");
252 xbt_assert(s == sizeof answer, "Broken message (size=%zd; expected %zu)", s, sizeof answer);
253 xbt_assert(answer.type == MessageType::FORK_REPLY,
254 "Received unexpected message %s (%i); expected MessageType::FORK_REPLY (%i)", to_c_str(answer.type),
255 (int)answer.type, (int)MessageType::FORK_REPLY);
261 std::unique_ptr<CheckerSide> CheckerSide::clone(int master_socket)
263 s_mc_message_int_t m = {};
264 m.type = MessageType::FORK;
266 xbt_assert(get_channel().send(m) == 0, "Could not ask the app to fork on need.");
268 int sock = accept(master_socket, nullptr /* I know who's connecting*/, nullptr);
269 xbt_assert(sock > 0, "Cannot accept the incomming connection of the forked app: %s.", strerror(errno));
271 return std::make_unique<CheckerSide>(sock, this);
274 void CheckerSide::finalize(bool terminate_asap)
276 s_mc_message_int_t m = {};
277 m.type = MessageType::FINALIZE;
278 m.value = terminate_asap;
279 xbt_assert(get_channel().send(m) == 0, "Could not ask the app to finalize on need");
281 s_mc_message_t answer;
282 ssize_t s = get_channel().receive(answer);
283 xbt_assert(s != -1, "Could not receive answer to FINALIZE");
284 xbt_assert(s == sizeof answer, "Broken message (size=%zd; expected %zu)", s, sizeof answer);
285 xbt_assert(answer.type == MessageType::FINALIZE_REPLY,
286 "Received unexpected message %s (%i); expected MessageType::FINALIZE_REPLY (%i)", to_c_str(answer.type),
287 (int)answer.type, (int)MessageType::FINALIZE_REPLY);
290 void CheckerSide::dispatch_events() const
292 event_base_dispatch(base_.get());
295 void CheckerSide::break_loop() const
297 event_base_loopbreak(base_.get());
300 bool CheckerSide::handle_message(const char* buffer, ssize_t size)
302 s_mc_message_t base_message;
304 xbt_assert(size >= (ssize_t)sizeof(base_message), "Broken message. Got only %ld bytes.", size);
305 memcpy(&base_message, buffer, sizeof(base_message));
307 switch (base_message.type) {
308 case MessageType::IGNORE_HEAP: {
309 consumed = sizeof(s_mc_message_ignore_heap_t);
310 #if SIMGRID_HAVE_STATEFUL_MC
311 if (remote_memory_ != nullptr) {
312 s_mc_message_ignore_heap_t message;
313 xbt_assert(size >= static_cast<ssize_t>(sizeof(message)), "Broken message");
314 memcpy(&message, buffer, sizeof(message));
316 IgnoredHeapRegion region;
317 region.block = message.block;
318 region.fragment = message.fragment;
319 region.address = message.address;
320 region.size = message.size;
321 get_remote_memory()->ignore_heap(region);
324 XBT_INFO("Ignoring a IGNORE_HEAP message because we don't need to introspect memory.");
328 case MessageType::UNIGNORE_HEAP: {
329 consumed = sizeof(s_mc_message_ignore_memory_t);
330 #if SIMGRID_HAVE_STATEFUL_MC
331 if (remote_memory_ != nullptr) {
332 s_mc_message_ignore_memory_t message;
333 xbt_assert(size == static_cast<ssize_t>(sizeof(message)), "Broken message");
334 memcpy(&message, buffer, sizeof(message));
335 get_remote_memory()->unignore_heap((void*)message.addr, message.size);
338 XBT_INFO("Ignoring an UNIGNORE_HEAP message because we don't need to introspect memory.");
342 case MessageType::IGNORE_MEMORY: {
343 consumed = sizeof(s_mc_message_ignore_memory_t);
344 #if SIMGRID_HAVE_STATEFUL_MC
345 if (remote_memory_ != nullptr) {
346 s_mc_message_ignore_memory_t message;
347 xbt_assert(size >= static_cast<ssize_t>(sizeof(message)), "Broken message");
348 memcpy(&message, buffer, sizeof(message));
349 get_remote_memory()->ignore_region(message.addr, message.size);
352 XBT_INFO("Ignoring an IGNORE_MEMORY message because we don't need to introspect memory.");
356 case MessageType::STACK_REGION: {
357 consumed = sizeof(s_mc_message_stack_region_t);
358 #if SIMGRID_HAVE_STATEFUL_MC
359 if (remote_memory_ != nullptr) {
360 s_mc_message_stack_region_t message;
361 xbt_assert(size >= static_cast<ssize_t>(sizeof(message)), "Broken message");
362 memcpy(&message, buffer, sizeof(message));
363 get_remote_memory()->stack_areas().push_back(message.stack_region);
366 XBT_INFO("Ignoring an STACK_REGION message because we don't need to introspect memory.");
370 case MessageType::REGISTER_SYMBOL: {
371 consumed = sizeof(s_mc_message_register_symbol_t);
372 #if SIMGRID_HAVE_STATEFUL_MC
373 s_mc_message_register_symbol_t message;
374 xbt_assert(size >= static_cast<ssize_t>(sizeof(message)), "Broken message");
375 memcpy(&message, buffer, sizeof(message));
376 xbt_assert(not message.callback, "Support for client-side function proposition is not implemented.");
377 XBT_DEBUG("Received symbol: %s", message.name.data());
379 LivenessChecker::automaton_register_symbol(*get_remote_memory(), message.name.data(), remote((int*)message.data));
381 xbt_die("Please don't use liveness properties when MC is compiled out.");
386 case MessageType::WAITING:
387 consumed = sizeof(s_mc_message_t);
388 if (size > consumed) {
389 XBT_DEBUG("%d reinject %d bytes after a %s message", getpid(), (int)(size - consumed),
390 to_c_str(base_message.type));
391 channel_.reinject(&buffer[consumed], size - consumed);
396 case MessageType::ASSERTION_FAILED:
397 // report_assertion_failure() is NORETURN, but it may change when we report more than one error per run,
398 // so please keep the consumed computation even if clang-static detects it as a dead affectation.
399 consumed = sizeof(s_mc_message_t);
400 Exploration::get_instance()->report_assertion_failure();
404 xbt_die("Unexpected message from the application");
406 if (size > consumed) {
407 XBT_DEBUG("%d reinject %d bytes after a %s message", getpid(), (int)(size - consumed), to_c_str(base_message.type));
408 channel_.reinject(&buffer[consumed], size - consumed);
413 void CheckerSide::wait_for_requests()
415 XBT_DEBUG("Resume the application");
416 if (get_channel().send(MessageType::CONTINUE) != 0)
417 throw xbt::errno_error();
418 clear_memory_cache();
424 void CheckerSide::clear_memory_cache()
426 #if SIMGRID_HAVE_STATEFUL_MC
428 remote_memory_->clear_cache();
432 void CheckerSide::handle_dead_child(int status)
434 // From PTRACE_O_TRACEEXIT:
436 if (status >> 8 == (SIGTRAP | (PTRACE_EVENT_EXIT << 8))) {
437 unsigned long eventmsg;
438 xbt_assert(ptrace(PTRACE_GETEVENTMSG, pid_, 0, &eventmsg) != -1, "Could not get exit status");
439 status = static_cast<int>(eventmsg);
440 if (WIFSIGNALED(status)) {
442 Exploration::get_instance()->report_crash(status);
447 // We don't care about non-lethal signals, just reinject them:
448 if (WIFSTOPPED(status)) {
449 XBT_DEBUG("Stopped with signal %i", (int)WSTOPSIG(status));
452 ptrace(PTRACE_CONT, pid_, 0, WSTOPSIG(status));
454 ptrace(PT_CONTINUE, pid_, (caddr_t)1, WSTOPSIG(status));
456 xbt_assert(errno == 0, "Could not PTRACE_CONT: %s", strerror(errno));
459 else if (WIFSIGNALED(status)) {
461 Exploration::get_instance()->report_crash(status);
462 } else if (WIFEXITED(status)) {
463 XBT_DEBUG("Child process is over");
468 void CheckerSide::handle_waitpid()
470 XBT_DEBUG("%d checks for wait event. %s", getpid(),
471 child_checker_ == nullptr ? "Wait directly." : "Ask our proxy to wait for its child.");
473 if (child_checker_ == nullptr) { // Wait directly
476 while ((pid = waitpid(-1, &status, WNOHANG)) != 0) {
478 if (errno == ECHILD) { // No more children:
479 xbt_assert(not this->running(), "Inconsistent state");
482 xbt_die("Could not wait for pid: %s", strerror(errno));
486 if (pid == get_pid())
487 handle_dead_child(status);
490 } else { // Ask our proxy to wait for us
491 s_mc_message_int_t request = {};
492 request.type = MessageType::WAIT_CHILD;
493 request.value = pid_;
494 xbt_assert(child_checker_->get_channel().send(request) == 0,
495 "Could not ask my child to waitpid its child for me: %s", strerror(errno));
497 s_mc_message_int_t answer;
498 ssize_t answer_size = child_checker_->get_channel().receive(answer);
499 xbt_assert(answer_size != -1, "Could not receive message");
500 xbt_assert(answer.type == MessageType::WAIT_CHILD_REPLY,
501 "The received message is not the WAIT_CHILD_REPLY I was expecting but of type %s",
502 to_c_str(answer.type));
503 xbt_assert(answer_size == sizeof answer, "Broken message (size=%zd; expected %zu)", answer_size, sizeof answer);
504 handle_dead_child(answer.value);
507 } // namespace simgrid::mc