1 /* Copyright (c) 2015-2023. 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/AppSide.hpp"
7 #include "simgrid/s4u/Host.hpp"
8 #include "src/internal_config.h"
9 #include "src/kernel/EngineImpl.hpp"
10 #include "src/kernel/actor/ActorImpl.hpp"
11 #include "src/kernel/actor/SimcallObserver.hpp"
12 #include "src/mc/mc_base.hpp"
13 #include "src/mc/mc_config.hpp"
14 #if SIMGRID_HAVE_STATEFUL_MC
15 #include "src/mc/sosp/RemoteProcessMemory.hpp"
18 #include "src/smpi/include/private.hpp"
20 #include "src/sthread/sthread.h"
21 #include "src/xbt/coverage.h"
23 #include <simgrid/modelchecker.h>
26 #include <cstdio> // setvbuf
30 #include <sys/ptrace.h>
31 #include <sys/socket.h>
32 #include <sys/types.h>
36 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(mc_client, mc, "MC client logic");
37 XBT_LOG_EXTERNAL_CATEGORY(mc_global);
39 namespace simgrid::mc {
41 std::unique_ptr<AppSide> AppSide::instance_;
43 AppSide* AppSide::get()
45 // Only initialize the MC world once
46 if (instance_ != nullptr)
47 return instance_.get();
49 if (std::getenv(MC_ENV_SOCKET_FD) == nullptr) // We are not in MC mode: don't initialize the MC world
52 XBT_DEBUG("Initialize the MC world. MC_NEED_PTRACE=%s", std::getenv("MC_NEED_PTRACE"));
54 simgrid::mc::set_model_checking_mode(ModelCheckingMode::APP_SIDE);
56 setvbuf(stdout, nullptr, _IOLBF, 0);
58 // Fetch socket from MC_ENV_SOCKET_FD:
59 const char* fd_env = std::getenv(MC_ENV_SOCKET_FD);
60 int fd = xbt_str_parse_int(fd_env, "Not a number in variable '" MC_ENV_SOCKET_FD "'");
61 XBT_DEBUG("Model-checked application found socket FD %i", fd);
63 // Check the socket type/validity:
65 socklen_t socklen = sizeof(type);
66 xbt_assert(getsockopt(fd, SOL_SOCKET, SO_TYPE, &type, &socklen) == 0, "Could not check socket type");
67 xbt_assert(type == SOCK_SEQPACKET, "Unexpected socket type %i", type);
68 XBT_DEBUG("Model-checked application found expected socket type");
70 instance_ = std::make_unique<simgrid::mc::AppSide>(fd);
72 // Wait for the model-checker:
73 if (getenv("MC_NEED_PTRACE") != nullptr) {
76 ptrace(PTRACE_TRACEME, 0, nullptr, nullptr);
78 ptrace(PT_TRACE_ME, 0, nullptr, 0);
80 xbt_die("no ptrace equivalent coded for this platform, please don't use the liveness checker here.");
83 xbt_assert(errno == 0 && raise(SIGSTOP) == 0, "Could not wait for the model-checker (errno = %d: %s)", errno,
87 instance_->handle_messages();
88 return instance_.get();
91 void AppSide::handle_deadlock_check(const s_mc_message_t*) const
93 const auto* engine = kernel::EngineImpl::get_instance();
94 const auto& actor_list = engine->get_actor_list();
95 bool deadlock = not actor_list.empty() && std::none_of(begin(actor_list), end(actor_list), [](const auto& kv) {
96 return mc::actor_is_enabled(kv.second);
100 XBT_CINFO(mc_global, "**************************");
101 XBT_CINFO(mc_global, "*** DEADLOCK DETECTED ***");
102 XBT_CINFO(mc_global, "**************************");
103 engine->display_all_actor_status();
106 s_mc_message_int_t answer = {};
107 answer.type = MessageType::DEADLOCK_CHECK_REPLY;
108 answer.value = deadlock;
109 xbt_assert(channel_.send(answer) == 0, "Could not send response");
111 void AppSide::handle_simcall_execute(const s_mc_message_simcall_execute_t* message) const
113 kernel::actor::ActorImpl* actor = kernel::EngineImpl::get_instance()->get_actor_by_pid(message->aid_);
114 xbt_assert(actor != nullptr, "Invalid pid %ld", message->aid_);
116 // The client may send some messages to the server while processing the transition
117 actor->simcall_handle(message->times_considered_);
118 // Say the server that the transition is over and that it should proceed
119 xbt_assert(channel_.send(MessageType::WAITING) == 0, "Could not send MESSAGE_WAITING to model-checker");
121 // Finish the RPC from the server: return a serialized observer, to build a Transition on Checker side
122 s_mc_message_simcall_execute_answer_t answer = {};
123 answer.type = MessageType::SIMCALL_EXECUTE_REPLY;
124 std::stringstream stream;
125 if (actor->simcall_.observer_ != nullptr) {
126 actor->simcall_.observer_->serialize(stream);
128 stream << (short)mc::Transition::Type::UNKNOWN;
130 std::string str = stream.str();
131 xbt_assert(str.size() + 1 <= answer.buffer.size(),
132 "The serialized simcall is too large for the buffer. Please fix the code.");
133 strncpy(answer.buffer.data(), str.c_str(), answer.buffer.size() - 1);
134 answer.buffer.back() = '\0';
136 XBT_DEBUG("send SIMCALL_EXECUTE_ANSWER(%s) ~> '%s'", actor->get_cname(), str.c_str());
137 xbt_assert(channel_.send(answer) == 0, "Could not send response");
140 void AppSide::handle_finalize(const s_mc_message_int_t* msg) const
142 bool terminate_asap = msg->value;
143 XBT_DEBUG("Finalize (terminate = %d)", (int)terminate_asap);
144 if (not terminate_asap) {
145 if (XBT_LOG_ISENABLED(mc_client, xbt_log_priority_debug))
146 kernel::EngineImpl::get_instance()->display_all_actor_status();
148 XBT_DEBUG("Smpi_enabled: %d", SMPI_is_inited());
149 if (SMPI_is_inited())
153 coverage_checkpoint();
154 xbt_assert(channel_.send(MessageType::FINALIZE_REPLY) == 0, "Could not answer to FINALIZE");
159 void AppSide::handle_fork(const s_mc_message_int_t* msg)
162 xbt_assert(pid >= 0, "Could not fork application sub-process: %s.", strerror(errno));
164 if (pid == 0) { // Child
165 int sock = socket(AF_LOCAL, SOCK_SEQPACKET | SOCK_CLOEXEC, 0);
167 struct sockaddr_un addr = {};
168 addr.sun_family = AF_LOCAL;
169 snprintf(addr.sun_path, 64, "/tmp/simgrid-mc-%lu", msg->value);
170 auto addr_size = offsetof(struct sockaddr_un, sun_path) + strlen(addr.sun_path);
172 xbt_assert(connect(sock, (struct sockaddr*)&addr, addr_size) >= 0,
173 "Cannot connect to Checker on /tmp/simgrid-mc-%lu: %s.", msg->value, strerror(errno));
175 channel_.reset_socket(sock);
177 s_mc_message_int_t answer = {};
178 answer.type = MessageType::FORK_REPLY;
179 answer.value = getpid();
180 xbt_assert(channel_.send(answer) == 0, "Could not send response to WAIT_CHILD_REPLY: %s", strerror(errno));
182 XBT_DEBUG("App %d forks subprocess %d.", getpid(), pid);
185 void AppSide::handle_wait_child(const s_mc_message_int_t* msg)
189 waitpid(msg->value, &status, 0);
190 xbt_assert(errno == 0, "Cannot wait on behalf of the checker: %s.", strerror(errno));
192 s_mc_message_int_t answer = {};
193 answer.type = MessageType::WAIT_CHILD_REPLY;
194 answer.value = status;
195 xbt_assert(channel_.send(answer) == 0, "Could not send response to WAIT_CHILD: %s", strerror(errno));
197 void AppSide::handle_need_meminfo()
199 #if SIMGRID_HAVE_STATEFUL_MC
200 this->need_memory_info_ = true;
201 s_mc_message_need_meminfo_reply_t answer = {};
202 answer.type = MessageType::NEED_MEMINFO_REPLY;
203 answer.mmalloc_default_mdp = mmalloc_get_current_heap();
204 xbt_assert(channel_.send(answer) == 0, "Could not send response to the request for meminfo.");
206 xbt_die("SimGrid was compiled without MC suppport, so liveness and similar features are not available.");
209 void AppSide::handle_actors_status() const
211 auto const& actor_list = kernel::EngineImpl::get_instance()->get_actor_list();
212 XBT_DEBUG("Serialize the actors to answer ACTORS_STATUS from the checker. %zu actors to go.", actor_list.size());
214 std::vector<s_mc_message_actors_status_one_t> status;
215 for (auto const& [aid, actor] : actor_list) {
216 s_mc_message_actors_status_one_t one = {};
217 one.type = MessageType::ACTORS_STATUS_REPLY_TRANSITION;
219 one.enabled = mc::actor_is_enabled(actor);
220 one.max_considered = actor->simcall_.observer_->get_max_consider();
221 status.push_back(one);
224 struct s_mc_message_actors_status_answer_t answer = {};
225 answer.type = MessageType::ACTORS_STATUS_REPLY_COUNT;
226 answer.count = static_cast<int>(status.size());
228 xbt_assert(channel_.send(answer) == 0, "Could not send ACTORS_STATUS_REPLY msg");
229 if (answer.count > 0) {
230 size_t size = status.size() * sizeof(s_mc_message_actors_status_one_t);
231 xbt_assert(channel_.send(status.data(), size) == 0, "Could not send ACTORS_STATUS_REPLY data");
234 // Serialize each transition to describe what each actor is doing
235 XBT_DEBUG("Deliver ACTOR_TRANSITION_PROBE payload");
236 for (const auto& actor_status : status) {
237 if (not actor_status.enabled)
240 const auto& actor = actor_list.at(actor_status.aid);
241 const int max_considered = actor_status.max_considered;
243 for (int times_considered = 0; times_considered < max_considered; times_considered++) {
244 std::stringstream stream;
245 s_mc_message_simcall_probe_one_t probe;
246 probe.type = MessageType::ACTORS_STATUS_REPLY_SIMCALL;
248 if (actor->simcall_.observer_ != nullptr) {
249 actor->simcall_.observer_->prepare(times_considered);
250 actor->simcall_.observer_->serialize(stream);
252 stream << (short)mc::Transition::Type::UNKNOWN;
255 std::string str = stream.str();
256 xbt_assert(str.size() + 1 <= probe.buffer.size(),
257 "The serialized transition is too large for the buffer. Please fix the code.");
258 strncpy(probe.buffer.data(), str.c_str(), probe.buffer.size() - 1);
259 probe.buffer.back() = '\0';
261 xbt_assert(channel_.send(probe) == 0, "Could not send ACTOR_TRANSITION_PROBE payload");
263 // NOTE: We do NOT need to reset `times_considered` for each actor's
264 // simcall observer here to the "original" value (i.e. the value BEFORE
265 // multiple prepare() calls were made for serialization purposes) since
266 // each SIMCALL_EXECUTE provides a `times_considered` to be used to prepare
267 // the transition before execution.
270 void AppSide::handle_actors_maxpid() const
272 s_mc_message_int_t answer = {};
273 answer.type = MessageType::ACTORS_MAXPID_REPLY;
274 answer.value = kernel::actor::ActorImpl::get_maxpid();
275 xbt_assert(channel_.send(answer) == 0, "Could not send response");
278 #define assert_msg_size(_name_, _type_) \
279 xbt_assert(received_size == sizeof(_type_), "Unexpected size for " _name_ " (%zd != %zu)", received_size, \
282 void AppSide::handle_messages()
284 while (true) { // Until we get a CONTINUE message
285 XBT_DEBUG("Waiting messages from the model-checker");
287 std::array<char, MC_MESSAGE_LENGTH> message_buffer;
288 ssize_t received_size = channel_.receive(message_buffer.data(), message_buffer.size());
290 if (received_size == 0) {
291 XBT_DEBUG("Socket closed on the Checker side, bailing out.");
292 ::_Exit(0); // Nobody's listening to that process anymore => exit as quickly as possible.
294 xbt_assert(received_size >= 0, "Could not receive commands from the model-checker");
295 xbt_assert(static_cast<size_t>(received_size) >= sizeof(s_mc_message_t), "Cannot handle short message (size=%zd)",
298 const s_mc_message_t* message = (s_mc_message_t*)message_buffer.data();
299 switch (message->type) {
300 case MessageType::CONTINUE:
301 assert_msg_size("MESSAGE_CONTINUE", s_mc_message_t);
304 case MessageType::DEADLOCK_CHECK:
305 assert_msg_size("DEADLOCK_CHECK", s_mc_message_t);
306 handle_deadlock_check(message);
309 case MessageType::SIMCALL_EXECUTE:
310 assert_msg_size("SIMCALL_EXECUTE", s_mc_message_simcall_execute_t);
311 handle_simcall_execute((s_mc_message_simcall_execute_t*)message_buffer.data());
314 case MessageType::FINALIZE:
315 assert_msg_size("FINALIZE", s_mc_message_int_t);
316 handle_finalize((s_mc_message_int_t*)message_buffer.data());
319 case MessageType::FORK:
320 assert_msg_size("FORK", s_mc_message_int_t);
321 handle_fork((s_mc_message_int_t*)message_buffer.data());
324 case MessageType::WAIT_CHILD:
325 assert_msg_size("WAIT_CHILD", s_mc_message_int_t);
326 handle_wait_child((s_mc_message_int_t*)message_buffer.data());
329 case MessageType::NEED_MEMINFO:
330 assert_msg_size("NEED_MEMINFO", s_mc_message_t);
331 handle_need_meminfo();
334 case MessageType::ACTORS_STATUS:
335 assert_msg_size("ACTORS_STATUS", s_mc_message_t);
336 handle_actors_status();
339 case MessageType::ACTORS_MAXPID:
340 assert_msg_size("ACTORS_MAXPID", s_mc_message_t);
341 handle_actors_maxpid();
345 xbt_die("Received unexpected message %s (%i)", to_c_str(message->type), static_cast<int>(message->type));
351 void AppSide::main_loop()
353 simgrid::mc::processes_time.resize(simgrid::kernel::actor::ActorImpl::get_maxpid());
354 MC_ignore_heap(simgrid::mc::processes_time.data(),
355 simgrid::mc::processes_time.size() * sizeof(simgrid::mc::processes_time[0]));
358 coverage_checkpoint();
361 simgrid::mc::execute_actors();
362 xbt_assert(channel_.send(MessageType::WAITING) == 0, "Could not send WAITING message to model-checker");
363 this->handle_messages();
367 void AppSide::report_assertion_failure()
369 xbt_assert(channel_.send(MessageType::ASSERTION_FAILED) == 0, "Could not send assertion to model-checker");
370 this->handle_messages();
373 void AppSide::ignore_memory(void* addr, std::size_t size) const
375 if (not MC_is_active() || not need_memory_info_)
378 #if SIMGRID_HAVE_STATEFUL_MC
379 s_mc_message_ignore_memory_t message = {};
380 message.type = MessageType::IGNORE_MEMORY;
381 message.addr = (std::uintptr_t)addr;
383 xbt_assert(channel_.send(message) == 0, "Could not send IGNORE_MEMORY message to model-checker");
385 xbt_die("Cannot really call ignore_heap() in non-SIMGRID_MC mode.");
389 void AppSide::ignore_heap(void* address, std::size_t size) const
391 if (not MC_is_active() || not need_memory_info_)
394 #if SIMGRID_HAVE_STATEFUL_MC
395 const s_xbt_mheap_t* heap = mmalloc_get_current_heap();
397 s_mc_message_ignore_heap_t message = {};
398 message.type = MessageType::IGNORE_HEAP;
399 message.address = address;
401 message.block = ((char*)address - (char*)heap->heapbase) / BLOCKSIZE + 1;
402 if (heap->heapinfo[message.block].type == 0) {
403 message.fragment = -1;
404 heap->heapinfo[message.block].busy_block.ignore++;
406 message.fragment = (ADDR2UINT(address) % BLOCKSIZE) >> heap->heapinfo[message.block].type;
407 heap->heapinfo[message.block].busy_frag.ignore[message.fragment]++;
410 xbt_assert(channel_.send(message) == 0, "Could not send ignored region to MCer");
412 xbt_die("Cannot really call ignore_heap() in non-SIMGRID_MC mode.");
416 void AppSide::unignore_heap(void* address, std::size_t size) const
418 if (not MC_is_active() || not need_memory_info_)
421 #if SIMGRID_HAVE_STATEFUL_MC
422 s_mc_message_ignore_memory_t message = {};
423 message.type = MessageType::UNIGNORE_HEAP;
424 message.addr = (std::uintptr_t)address;
426 xbt_assert(channel_.send(message) == 0, "Could not send IGNORE_HEAP message to model-checker");
428 xbt_die("Cannot really call unignore_heap() in non-SIMGRID_MC mode.");
432 void AppSide::declare_symbol(const char* name, int* value) const
434 if (not MC_is_active() || not need_memory_info_) {
435 XBT_CRITICAL("Ignore AppSide::declare_symbol(%s)", name);
439 #if SIMGRID_HAVE_STATEFUL_MC
440 s_mc_message_register_symbol_t message = {};
441 message.type = MessageType::REGISTER_SYMBOL;
442 xbt_assert(strlen(name) + 1 <= message.name.size(), "Symbol is too long");
443 strncpy(message.name.data(), name, message.name.size() - 1);
444 message.callback = nullptr;
445 message.data = value;
446 xbt_assert(channel_.send(message) == 0, "Could send REGISTER_SYMBOL message to model-checker");
448 xbt_die("Cannot really call declare_symbol() in non-SIMGRID_MC mode.");
452 /** Register a stack in the model checker
454 * The stacks are allocated in the heap. The MC handle them specifically
455 * when we analyze/compare the content of the heap so it must be told where
456 * they are with this function.
458 void AppSide::declare_stack(void* stack, size_t size, ucontext_t* context) const
460 if (not MC_is_active() || not need_memory_info_)
463 #if SIMGRID_HAVE_STATEFUL_MC
464 const s_xbt_mheap_t* heap = mmalloc_get_current_heap();
466 s_stack_region_t region = {};
467 region.address = stack;
468 region.context = context;
470 region.block = ((char*)stack - (char*)heap->heapbase) / BLOCKSIZE + 1;
472 s_mc_message_stack_region_t message = {};
473 message.type = MessageType::STACK_REGION;
474 message.stack_region = region;
475 xbt_assert(channel_.send(message) == 0, "Could not send STACK_REGION to model-checker");
477 xbt_die("Cannot really call declare_stack() in non-SIMGRID_MC mode.");
480 } // namespace simgrid::mc