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()
124 if (base_ != nullptr)
125 event_base_free(base_.get());
126 auto* base = event_base_new();
129 socket_event_ = event_new(
130 base, get_channel().get_socket(), EV_READ | EV_PERSIST,
131 [](evutil_socket_t, short events, void* arg) {
132 auto checker = static_cast<simgrid::mc::CheckerSide*>(arg);
133 if (events == EV_READ) {
134 std::array<char, MC_MESSAGE_LENGTH> buffer;
135 ssize_t size = recv(checker->get_channel().get_socket(), 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();
147 xbt_die("Unexpected event");
151 event_add(socket_event_, nullptr);
153 signal_event_ = event_new(
154 base, SIGCHLD, EV_SIGNAL | EV_PERSIST,
155 [](evutil_socket_t sig, short events, void* arg) {
156 auto checker = static_cast<simgrid::mc::CheckerSide*>(arg);
157 if (events == EV_SIGNAL) {
159 checker->handle_waitpid();
161 xbt_die("Unexpected signal: %d", sig);
163 xbt_die("Unexpected event");
167 event_add(signal_event_, nullptr);
170 CheckerSide::CheckerSide(const std::vector<char*>& args, bool need_memory_info) : running_(true)
172 // Create an AF_LOCAL socketpair used for exchanging messages between the model-checker process (ancestor)
173 // and the application process (child)
175 xbt_assert(socketpair(AF_LOCAL, SOCK_SEQPACKET | SOCK_CLOEXEC, 0, sockets) != -1, "Could not create socketpair");
178 xbt_assert(pid_ >= 0, "Could not fork model-checked process");
180 if (pid_ == 0) { // Child
182 run_child_process(sockets[0], args, need_memory_info); // We need ptrace if we need the mem info
186 // Parent (model-checker):
188 channel_.reset_socket(sockets[1]);
191 if (need_memory_info)
192 wait_application_process(pid_);
194 // Request the initial memory on need
195 if (need_memory_info) {
196 channel_.send(MessageType::INITIAL_ADDRESSES);
197 s_mc_message_initial_addresses_reply_t answer;
198 ssize_t answer_size = channel_.receive(answer);
199 xbt_assert(answer_size != -1, "Could not receive message");
200 xbt_assert(answer.type == MessageType::INITIAL_ADDRESSES_REPLY,
201 "The received message is not the INITIAL_ADDRESS_REPLY I was expecting but of type %s",
202 to_c_str(answer.type));
203 xbt_assert(answer_size == sizeof answer, "Broken message (size=%zd; expected %zu)", answer_size, sizeof answer);
205 /* We now have enough info to create the memory address space */
206 remote_memory_ = std::make_unique<simgrid::mc::RemoteProcessMemory>(pid_, answer.mmalloc_default_mdp);
212 CheckerSide::~CheckerSide()
214 event_del(socket_event_);
215 event_free(socket_event_);
216 event_del(signal_event_);
217 event_free(signal_event_);
221 xbt_assert(kill(get_pid(), SIGKILL) == 0);
222 xbt_assert(errno == 0);
223 waitpid(get_pid(), nullptr, 0);
224 xbt_assert(errno == 0);
228 void CheckerSide::finalize(bool terminate_asap)
230 s_mc_message_int_t m = {};
231 m.type = MessageType::FINALIZE;
232 m.value = terminate_asap;
233 xbt_assert(get_channel().send(m) == 0, "Could not ask the app to finalize on need");
235 s_mc_message_t answer;
236 ssize_t s = get_channel().receive(answer);
237 xbt_assert(s != -1, "Could not receive answer to FINALIZE");
238 xbt_assert(s == sizeof answer, "Broken message (size=%zd; expected %zu)", s, sizeof answer);
239 xbt_assert(answer.type == MessageType::FINALIZE_REPLY,
240 "Received unexpected message %s (%i); expected MessageType::FINALIZE_REPLY (%i)", to_c_str(answer.type),
241 (int)answer.type, (int)MessageType::FINALIZE_REPLY);
244 void CheckerSide::dispatch_events() const
246 event_base_dispatch(base_.get());
249 void CheckerSide::break_loop() const
251 event_base_loopbreak(base_.get());
254 bool CheckerSide::handle_message(const char* buffer, ssize_t size)
256 s_mc_message_t base_message;
257 xbt_assert(size >= (ssize_t)sizeof(base_message), "Broken message. Got only %ld bytes.", size);
258 memcpy(&base_message, buffer, sizeof(base_message));
260 switch (base_message.type) {
261 case MessageType::IGNORE_HEAP: {
262 if (remote_memory_ != nullptr) {
263 s_mc_message_ignore_heap_t message;
264 xbt_assert(size == sizeof(message), "Broken message");
265 memcpy(&message, buffer, sizeof(message));
267 IgnoredHeapRegion region;
268 region.block = message.block;
269 region.fragment = message.fragment;
270 region.address = message.address;
271 region.size = message.size;
272 get_remote_memory()->ignore_heap(region);
274 XBT_INFO("Ignoring a IGNORE_HEAP message because we don't need to introspect memory.");
279 case MessageType::UNIGNORE_HEAP: {
280 if (remote_memory_ != nullptr) {
281 s_mc_message_ignore_memory_t message;
282 xbt_assert(size == sizeof(message), "Broken message");
283 memcpy(&message, buffer, sizeof(message));
284 get_remote_memory()->unignore_heap((void*)message.addr, message.size);
286 XBT_INFO("Ignoring an UNIGNORE_HEAP message because we don't need to introspect memory.");
291 case MessageType::IGNORE_MEMORY: {
292 if (remote_memory_ != nullptr) {
293 s_mc_message_ignore_memory_t message;
294 xbt_assert(size == sizeof(message), "Broken message");
295 memcpy(&message, buffer, sizeof(message));
296 get_remote_memory()->ignore_region(message.addr, message.size);
298 XBT_INFO("Ignoring an IGNORE_MEMORY message because we don't need to introspect memory.");
303 case MessageType::STACK_REGION: {
304 if (remote_memory_ != nullptr) {
305 s_mc_message_stack_region_t message;
306 xbt_assert(size == sizeof(message), "Broken message");
307 memcpy(&message, buffer, sizeof(message));
308 get_remote_memory()->stack_areas().push_back(message.stack_region);
310 XBT_INFO("Ignoring an STACK_REGION message because we don't need to introspect memory.");
315 case MessageType::REGISTER_SYMBOL: {
316 s_mc_message_register_symbol_t message;
317 xbt_assert(size == sizeof(message), "Broken message");
318 memcpy(&message, buffer, sizeof(message));
319 xbt_assert(not message.callback, "Support for client-side function proposition is not implemented.");
320 XBT_DEBUG("Received symbol: %s", message.name.data());
322 LivenessChecker::automaton_register_symbol(*get_remote_memory(), message.name.data(), remote((int*)message.data));
326 case MessageType::WAITING:
329 case MessageType::ASSERTION_FAILED:
330 Exploration::get_instance()->report_assertion_failure();
334 xbt_die("Unexpected message from model-checked application");
339 void CheckerSide::wait_for_requests()
341 /* Resume the application */
342 if (get_channel().send(MessageType::CONTINUE) != 0)
343 throw xbt::errno_error();
344 clear_memory_cache();
350 void CheckerSide::clear_memory_cache()
353 remote_memory_->clear_cache();
356 void CheckerSide::handle_waitpid()
358 XBT_DEBUG("Check for wait event");
361 while ((pid = waitpid(-1, &status, WNOHANG)) != 0) {
363 if (errno == ECHILD) { // No more children:
364 xbt_assert(not this->running(), "Inconsistent state");
367 xbt_die("Could not wait for pid: %s", strerror(errno));
371 if (pid == get_pid()) {
372 // From PTRACE_O_TRACEEXIT:
374 if (status >> 8 == (SIGTRAP | (PTRACE_EVENT_EXIT << 8))) {
375 unsigned long eventmsg;
376 xbt_assert(ptrace(PTRACE_GETEVENTMSG, pid, 0, &eventmsg) != -1, "Could not get exit status");
377 status = static_cast<int>(eventmsg);
378 if (WIFSIGNALED(status)) {
380 Exploration::get_instance()->report_crash(status);
385 // We don't care about non-lethal signals, just reinject them:
386 if (WIFSTOPPED(status)) {
387 XBT_DEBUG("Stopped with signal %i", (int)WSTOPSIG(status));
390 ptrace(PTRACE_CONT, pid, 0, WSTOPSIG(status));
392 ptrace(PT_CONTINUE, pid, (caddr_t)1, WSTOPSIG(status));
394 xbt_assert(errno == 0, "Could not PTRACE_CONT");
397 else if (WIFSIGNALED(status)) {
399 Exploration::get_instance()->report_crash(status);
400 } else if (WIFEXITED(status)) {
401 XBT_DEBUG("Child process is over");
408 } // namespace simgrid::mc