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"
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::initialize()
45 if (not std::getenv(MC_ENV_SOCKET_FD)) // We are not in MC mode: don't initialize the MC world
48 // Do not break if we are called multiple times:
50 return instance_.get();
52 simgrid::mc::model_checking_mode = ModelCheckingMode::APP_SIDE;
54 setvbuf(stdout, nullptr, _IOLBF, 0);
56 // Fetch socket from MC_ENV_SOCKET_FD:
57 const char* fd_env = std::getenv(MC_ENV_SOCKET_FD);
58 int fd = xbt_str_parse_int(fd_env, "Not a number in variable '" MC_ENV_SOCKET_FD "'");
59 XBT_DEBUG("Model-checked application found socket FD %i", fd);
61 // Check the socket type/validity:
63 socklen_t socklen = sizeof(type);
64 xbt_assert(getsockopt(fd, SOL_SOCKET, SO_TYPE, &type, &socklen) == 0, "Could not check socket type");
65 xbt_assert(type == SOCK_SEQPACKET, "Unexpected socket type %i", type);
66 XBT_DEBUG("Model-checked application found expected socket type");
68 instance_ = std::make_unique<simgrid::mc::AppSide>(fd);
70 // Wait for the model-checker:
71 if (getenv("MC_NEED_PTRACE") != nullptr) {
74 ptrace(PTRACE_TRACEME, 0, nullptr, nullptr);
76 ptrace(PT_TRACE_ME, 0, nullptr, 0);
78 xbt_die("no ptrace equivalent coded for this platform, please don't use the liveness checker here.");
81 xbt_assert(errno == 0 && raise(SIGSTOP) == 0, "Could not wait for the model-checker (errno = %d: %s)", errno,
85 instance_->handle_messages();
86 return instance_.get();
89 void AppSide::handle_deadlock_check(const s_mc_message_t*) const
91 const auto* engine = kernel::EngineImpl::get_instance();
92 const auto& actor_list = engine->get_actor_list();
93 bool deadlock = not actor_list.empty() && std::none_of(begin(actor_list), end(actor_list), [](const auto& kv) {
94 return mc::actor_is_enabled(kv.second);
98 XBT_CINFO(mc_global, "**************************");
99 XBT_CINFO(mc_global, "*** DEADLOCK DETECTED ***");
100 XBT_CINFO(mc_global, "**************************");
101 engine->display_all_actor_status();
104 s_mc_message_int_t answer = {};
105 answer.type = MessageType::DEADLOCK_CHECK_REPLY;
106 answer.value = deadlock;
107 xbt_assert(channel_.send(answer) == 0, "Could not send response");
109 void AppSide::handle_simcall_execute(const s_mc_message_simcall_execute_t* message) const
111 kernel::actor::ActorImpl* actor = kernel::EngineImpl::get_instance()->get_actor_by_pid(message->aid_);
112 xbt_assert(actor != nullptr, "Invalid pid %ld", message->aid_);
114 // The client may send some messages to the server while processing the transition
115 actor->simcall_handle(message->times_considered_);
116 // Say the server that the transition is over and that it should proceed
117 xbt_assert(channel_.send(MessageType::WAITING) == 0, "Could not send MESSAGE_WAITING to model-checker");
119 // Finish the RPC from the server: return a serialized observer, to build a Transition on Checker side
120 s_mc_message_simcall_execute_answer_t answer = {};
121 answer.type = MessageType::SIMCALL_EXECUTE_REPLY;
122 std::stringstream stream;
123 if (actor->simcall_.observer_ != nullptr) {
124 actor->simcall_.observer_->serialize(stream);
126 stream << (short)mc::Transition::Type::UNKNOWN;
128 std::string str = stream.str();
129 xbt_assert(str.size() + 1 <= answer.buffer.size(),
130 "The serialized simcall is too large for the buffer. Please fix the code.");
131 strncpy(answer.buffer.data(), str.c_str(), answer.buffer.size() - 1);
132 answer.buffer.back() = '\0';
134 XBT_DEBUG("send SIMCALL_EXECUTE_ANSWER(%s) ~> '%s'", actor->get_cname(), str.c_str());
135 xbt_assert(channel_.send(answer) == 0, "Could not send response");
138 void AppSide::handle_finalize(const s_mc_message_int_t* msg) const
140 bool terminate_asap = msg->value;
141 XBT_DEBUG("Finalize (terminate = %d)", (int)terminate_asap);
142 if (not terminate_asap) {
143 if (XBT_LOG_ISENABLED(mc_client, xbt_log_priority_debug))
144 kernel::EngineImpl::get_instance()->display_all_actor_status();
146 XBT_DEBUG("Smpi_enabled: %d", SMPI_is_inited());
147 if (SMPI_is_inited())
151 coverage_checkpoint();
152 xbt_assert(channel_.send(MessageType::FINALIZE_REPLY) == 0, "Could not answer to FINALIZE");
157 void AppSide::handle_fork(const s_mc_message_int_t* msg)
160 xbt_assert(pid >= 0, "Could not fork application sub-process: %s.", strerror(errno));
162 if (pid == 0) { // Child
163 int sock = socket(AF_LOCAL, SOCK_SEQPACKET | SOCK_CLOEXEC, 0);
165 struct sockaddr_un addr = {};
166 addr.sun_family = AF_LOCAL;
167 snprintf(addr.sun_path, 64, "/tmp/simgrid-mc-%lu", msg->value);
168 auto addr_size = offsetof(struct sockaddr_un, sun_path) + strlen(addr.sun_path);
170 xbt_assert(connect(sock, (struct sockaddr*)&addr, addr_size) >= 0,
171 "Cannot connect to Checker on /tmp/simgrid-mc-%lu: %s.", msg->value, strerror(errno));
173 channel_.reset_socket(sock);
175 s_mc_message_int_t answer = {};
176 answer.type = MessageType::FORK_REPLY;
177 answer.value = getpid();
178 xbt_assert(channel_.send(answer) == 0, "Could not send response to WAIT_CHILD_REPLY: %s", strerror(errno));
181 void AppSide::handle_wait_child(const s_mc_message_int_t* msg)
185 waitpid(msg->value, &status, 0);
186 xbt_assert(errno == 0, "Cannot wait on behalf of the checker: %s.", strerror(errno));
188 s_mc_message_int_t answer = {};
189 answer.type = MessageType::WAIT_CHILD_REPLY;
190 answer.value = status;
191 xbt_assert(channel_.send(answer) == 0, "Could not send response to WAIT_CHILD: %s", strerror(errno));
193 void AppSide::handle_need_meminfo()
196 this->need_memory_info_ = true;
197 s_mc_message_need_meminfo_reply_t answer = {};
198 answer.type = MessageType::NEED_MEMINFO_REPLY;
199 answer.mmalloc_default_mdp = mmalloc_get_current_heap();
200 xbt_assert(channel_.send(answer) == 0, "Could not send response to the request for meminfo.");
202 xbt_die("SimGrid was compiled without MC suppport, so liveness and similar features are not available.");
205 void AppSide::handle_actors_status() const
207 auto const& actor_list = kernel::EngineImpl::get_instance()->get_actor_list();
208 XBT_DEBUG("Serialize the actors to answer ACTORS_STATUS from the checker. %zu actors to go.", actor_list.size());
210 std::vector<s_mc_message_actors_status_one_t> status;
211 for (auto const& [aid, actor] : actor_list) {
212 s_mc_message_actors_status_one_t one = {};
213 one.type = MessageType::ACTORS_STATUS_REPLY_TRANSITION;
215 one.enabled = mc::actor_is_enabled(actor);
216 one.max_considered = actor->simcall_.observer_->get_max_consider();
217 status.push_back(one);
220 struct s_mc_message_actors_status_answer_t answer = {};
221 answer.type = MessageType::ACTORS_STATUS_REPLY_COUNT;
222 answer.count = static_cast<int>(status.size());
224 xbt_assert(channel_.send(answer) == 0, "Could not send ACTORS_STATUS_REPLY msg");
225 if (answer.count > 0) {
226 size_t size = status.size() * sizeof(s_mc_message_actors_status_one_t);
227 xbt_assert(channel_.send(status.data(), size) == 0, "Could not send ACTORS_STATUS_REPLY data");
230 // Serialize each transition to describe what each actor is doing
231 XBT_DEBUG("Deliver ACTOR_TRANSITION_PROBE payload");
232 for (const auto& actor_status : status) {
233 if (not actor_status.enabled)
236 const auto& actor = actor_list.at(actor_status.aid);
237 const int max_considered = actor_status.max_considered;
239 for (int times_considered = 0; times_considered < max_considered; times_considered++) {
240 std::stringstream stream;
241 s_mc_message_simcall_probe_one_t probe;
242 probe.type = MessageType::ACTORS_STATUS_REPLY_SIMCALL;
244 if (actor->simcall_.observer_ != nullptr) {
245 actor->simcall_.observer_->prepare(times_considered);
246 actor->simcall_.observer_->serialize(stream);
248 stream << (short)mc::Transition::Type::UNKNOWN;
251 std::string str = stream.str();
252 xbt_assert(str.size() + 1 <= probe.buffer.size(),
253 "The serialized transition is too large for the buffer. Please fix the code.");
254 strncpy(probe.buffer.data(), str.c_str(), probe.buffer.size() - 1);
255 probe.buffer.back() = '\0';
257 xbt_assert(channel_.send(probe) == 0, "Could not send ACTOR_TRANSITION_PROBE payload");
259 // NOTE: We do NOT need to reset `times_considered` for each actor's
260 // simcall observer here to the "original" value (i.e. the value BEFORE
261 // multiple prepare() calls were made for serialization purposes) since
262 // each SIMCALL_EXECUTE provides a `times_considered` to be used to prepare
263 // the transition before execution.
266 void AppSide::handle_actors_maxpid() const
268 s_mc_message_int_t answer = {};
269 answer.type = MessageType::ACTORS_MAXPID_REPLY;
270 answer.value = kernel::actor::ActorImpl::get_maxpid();
271 xbt_assert(channel_.send(answer) == 0, "Could not send response");
274 #define assert_msg_size(_name_, _type_) \
275 xbt_assert(received_size == sizeof(_type_), "Unexpected size for " _name_ " (%zd != %zu)", received_size, \
278 void AppSide::handle_messages()
280 while (true) { // Until we get a CONTINUE message
281 XBT_DEBUG("Waiting messages from the model-checker");
283 std::array<char, MC_MESSAGE_LENGTH> message_buffer;
284 ssize_t received_size = channel_.receive(message_buffer.data(), message_buffer.size());
286 if (received_size == 0) {
287 XBT_DEBUG("Socket closed on the Checker side, bailing out.");
288 ::_Exit(0); // Nobody's listening to that process anymore => exit as quickly as possible.
290 xbt_assert(received_size >= 0, "Could not receive commands from the model-checker");
291 xbt_assert(static_cast<size_t>(received_size) >= sizeof(s_mc_message_t), "Cannot handle short message (size=%zd)",
294 const s_mc_message_t* message = (s_mc_message_t*)message_buffer.data();
295 switch (message->type) {
296 case MessageType::CONTINUE:
297 assert_msg_size("MESSAGE_CONTINUE", s_mc_message_t);
300 case MessageType::DEADLOCK_CHECK:
301 assert_msg_size("DEADLOCK_CHECK", s_mc_message_t);
302 handle_deadlock_check(message);
305 case MessageType::SIMCALL_EXECUTE:
306 assert_msg_size("SIMCALL_EXECUTE", s_mc_message_simcall_execute_t);
307 handle_simcall_execute((s_mc_message_simcall_execute_t*)message_buffer.data());
310 case MessageType::FINALIZE:
311 assert_msg_size("FINALIZE", s_mc_message_int_t);
312 handle_finalize((s_mc_message_int_t*)message_buffer.data());
315 case MessageType::FORK:
316 assert_msg_size("FORK", s_mc_message_int_t);
317 handle_fork((s_mc_message_int_t*)message_buffer.data());
320 case MessageType::WAIT_CHILD:
321 assert_msg_size("WAIT_CHILD", s_mc_message_int_t);
322 handle_wait_child((s_mc_message_int_t*)message_buffer.data());
325 case MessageType::NEED_MEMINFO:
326 assert_msg_size("NEED_MEMINFO", s_mc_message_t);
327 handle_need_meminfo();
330 case MessageType::ACTORS_STATUS:
331 assert_msg_size("ACTORS_STATUS", s_mc_message_t);
332 handle_actors_status();
335 case MessageType::ACTORS_MAXPID:
336 assert_msg_size("ACTORS_MAXPID", s_mc_message_t);
337 handle_actors_maxpid();
341 xbt_die("Received unexpected message %s (%i)", to_c_str(message->type), static_cast<int>(message->type));
347 void AppSide::main_loop()
349 simgrid::mc::processes_time.resize(simgrid::kernel::actor::ActorImpl::get_maxpid());
350 MC_ignore_heap(simgrid::mc::processes_time.data(),
351 simgrid::mc::processes_time.size() * sizeof(simgrid::mc::processes_time[0]));
354 coverage_checkpoint();
357 simgrid::mc::execute_actors();
358 xbt_assert(channel_.send(MessageType::WAITING) == 0, "Could not send WAITING message to model-checker");
359 this->handle_messages();
363 void AppSide::report_assertion_failure()
365 xbt_assert(channel_.send(MessageType::ASSERTION_FAILED) == 0, "Could not send assertion to model-checker");
366 this->handle_messages();
369 void AppSide::ignore_memory(void* addr, std::size_t size) const
371 if (not MC_is_active() || not need_memory_info_)
375 s_mc_message_ignore_memory_t message = {};
376 message.type = MessageType::IGNORE_MEMORY;
377 message.addr = (std::uintptr_t)addr;
379 xbt_assert(channel_.send(message) == 0, "Could not send IGNORE_MEMORY message to model-checker");
381 xbt_die("Cannot really call ignore_heap() in non-SIMGRID_MC mode.");
385 void AppSide::ignore_heap(void* address, std::size_t size) const
387 if (not MC_is_active() || not need_memory_info_)
391 const s_xbt_mheap_t* heap = mmalloc_get_current_heap();
393 s_mc_message_ignore_heap_t message = {};
394 message.type = MessageType::IGNORE_HEAP;
395 message.address = address;
397 message.block = ((char*)address - (char*)heap->heapbase) / BLOCKSIZE + 1;
398 if (heap->heapinfo[message.block].type == 0) {
399 message.fragment = -1;
400 heap->heapinfo[message.block].busy_block.ignore++;
402 message.fragment = (ADDR2UINT(address) % BLOCKSIZE) >> heap->heapinfo[message.block].type;
403 heap->heapinfo[message.block].busy_frag.ignore[message.fragment]++;
406 xbt_assert(channel_.send(message) == 0, "Could not send ignored region to MCer");
408 xbt_die("Cannot really call ignore_heap() in non-SIMGRID_MC mode.");
412 void AppSide::unignore_heap(void* address, std::size_t size) const
414 if (not MC_is_active() || not need_memory_info_)
418 s_mc_message_ignore_memory_t message = {};
419 message.type = MessageType::UNIGNORE_HEAP;
420 message.addr = (std::uintptr_t)address;
422 xbt_assert(channel_.send(message) == 0, "Could not send IGNORE_HEAP message to model-checker");
424 xbt_die("Cannot really call unignore_heap() in non-SIMGRID_MC mode.");
428 void AppSide::declare_symbol(const char* name, int* value) const
430 if (not MC_is_active() || not need_memory_info_) {
431 XBT_CRITICAL("Ignore AppSide::declare_symbol(%s)", name);
436 s_mc_message_register_symbol_t message = {};
437 message.type = MessageType::REGISTER_SYMBOL;
438 xbt_assert(strlen(name) + 1 <= message.name.size(), "Symbol is too long");
439 strncpy(message.name.data(), name, message.name.size() - 1);
440 message.callback = nullptr;
441 message.data = value;
442 xbt_assert(channel_.send(message) == 0, "Could send REGISTER_SYMBOL message to model-checker");
444 xbt_die("Cannot really call declare_symbol() in non-SIMGRID_MC mode.");
448 /** Register a stack in the model checker
450 * The stacks are allocated in the heap. The MC handle them specifically
451 * when we analyze/compare the content of the heap so it must be told where
452 * they are with this function.
454 void AppSide::declare_stack(void* stack, size_t size, ucontext_t* context) const
456 if (not MC_is_active() || not need_memory_info_)
460 const s_xbt_mheap_t* heap = mmalloc_get_current_heap();
462 s_stack_region_t region = {};
463 region.address = stack;
464 region.context = context;
466 region.block = ((char*)stack - (char*)heap->heapbase) / BLOCKSIZE + 1;
468 s_mc_message_stack_region_t message = {};
469 message.type = MessageType::STACK_REGION;
470 message.stack_region = region;
471 xbt_assert(channel_.send(message) == 0, "Could not send STACK_REGION to model-checker");
473 xbt_die("Cannot really call declare_stack() in non-SIMGRID_MC mode.");
476 } // namespace simgrid::mc