X-Git-Url: http://info.iut-bm.univ-fcomte.fr/pub/gitweb/simgrid.git/blobdiff_plain/e9b99cc75875aaffe31d627aceb45a3583770d55..026045c5de7c9b574fc67c8b11a3cf64a35c3109:/src/mc/api/RemoteApp.cpp diff --git a/src/mc/api/RemoteApp.cpp b/src/mc/api/RemoteApp.cpp index a6fd5ab3c6..45e595eed6 100644 --- a/src/mc/api/RemoteApp.cpp +++ b/src/mc/api/RemoteApp.cpp @@ -4,159 +4,73 @@ * under the terms of the license (GNU LGPL) which comes with this package. */ #include "src/mc/api/RemoteApp.hpp" -#include "src/internal_config.h" // HAVE_SMPI #include "src/mc/explo/Exploration.hpp" #include "src/mc/mc_config.hpp" #include "xbt/asserts.h" -#if HAVE_SMPI -#include "smpi/smpi.h" -#include "src/smpi/include/private.hpp" -#endif -#include "signal.h" #include "src/mc/api/State.hpp" #include "src/mc/mc_config.hpp" #include "src/mc/mc_exit.hpp" #include "src/mc/mc_private.hpp" #include "xbt/log.h" #include "xbt/system_error.hpp" +#include #include #include -#include #include #include #include -#include -#ifdef __linux__ -#include -#endif +#include +#include XBT_LOG_NEW_DEFAULT_SUBCATEGORY(mc_Session, mc, "Model-checker session"); XBT_LOG_EXTERNAL_CATEGORY(mc_global); -static simgrid::config::Flag _sg_mc_setenv{ - "model-check/setenv", "Extra environment variables to pass to the child process (ex: 'AZE=aze;QWE=qwe').", "", - [](std::string_view value) { - xbt_assert(value.empty() || value.find('=', 0) != std::string_view::npos, - "The 'model-check/setenv' parameter must be like 'AZE=aze', but it does not contain an equal sign."); - }}; - namespace simgrid::mc { -XBT_ATTRIB_NORETURN static void run_child_process(int socket, const std::vector& args) -{ - /* On startup, simix_global_init() calls simgrid::mc::Client::initialize(), which checks whether the MC_ENV_SOCKET_FD - * env variable is set. If so, MC mode is assumed, and the client is setup from its side - */ - -#ifdef __linux__ - // Make sure we do not outlive our parent - sigset_t mask; - sigemptyset(&mask); - xbt_assert(sigprocmask(SIG_SETMASK, &mask, nullptr) >= 0, "Could not unblock signals"); - xbt_assert(prctl(PR_SET_PDEATHSIG, SIGHUP) == 0, "Could not PR_SET_PDEATHSIG"); -#endif - - // Remove CLOEXEC to pass the socket to the application - int fdflags = fcntl(socket, F_GETFD, 0); - xbt_assert(fdflags != -1 && fcntl(socket, F_SETFD, fdflags & ~FD_CLOEXEC) != -1, - "Could not remove CLOEXEC for socket"); - - setenv(MC_ENV_SOCKET_FD, std::to_string(socket).c_str(), 1); - - /* Setup the tokenizer that parses the cfg:model-check/setenv parameter */ - using Tokenizer = boost::tokenizer>; - boost::char_separator semicol_sep(";"); - boost::char_separator equal_sep("="); - Tokenizer token_vars(_sg_mc_setenv.get(), semicol_sep); /* Iterate over all FOO=foo parts */ - for (const auto& token : token_vars) { - std::vector kv; - Tokenizer token_kv(token, equal_sep); - for (const auto& t : token_kv) /* Iterate over 'FOO' and then 'foo' in that 'FOO=foo' */ - kv.push_back(t); - xbt_assert(kv.size() == 2, "Parse error on 'model-check/setenv' value %s. Does it contain an equal sign?", - token.c_str()); - XBT_INFO("setenv '%s'='%s'", kv[0].c_str(), kv[1].c_str()); - setenv(kv[0].c_str(), kv[1].c_str(), 1); - } - - /* And now, exec the child process */ - int i = 1; - while (args[i] != nullptr && args[i][0] == '-') - i++; - - xbt_assert(args[i] != nullptr, - "Unable to find a binary to exec on the command line. Did you only pass config flags?"); - - execvp(args[i], args.data() + i); - XBT_CRITICAL("The model-checked process failed to exec(%s): %s.\n" - " Make sure that your binary exists on disk and is executable.", - args[i], strerror(errno)); - if (strchr(args[i], '=') != nullptr) - XBT_CRITICAL("If you want to pass environment variables to the application, please use --cfg=model-check/setenv:%s", - args[i]); - - xbt_die("Aborting now."); -} - -RemoteApp::RemoteApp(const std::vector& args) +RemoteApp::RemoteApp(const std::vector& args, bool need_memory_introspection) { -#if HAVE_SMPI - smpi_init_options(); // only performed once - xbt_assert(smpi_cfg_privatization() != SmpiPrivStrategies::MMAP, - "Please use the dlopen privatization schema when model-checking SMPI code"); -#endif - - // Create an AF_LOCAL socketpair used for exchanging messages - // between the model-checker process (ourselves) and the model-checked - // process: - int sockets[2]; - xbt_assert(socketpair(AF_LOCAL, SOCK_SEQPACKET | SOCK_CLOEXEC, 0, sockets) != -1, "Could not create socketpair"); - - pid_t pid = fork(); - xbt_assert(pid >= 0, "Could not fork model-checked process"); - - if (pid == 0) { // Child - ::close(sockets[1]); - run_child_process(sockets[0], args); - DIE_IMPOSSIBLE; - } - - // Parent (model-checker): - ::close(sockets[0]); + for (auto* arg : args) + app_args_.push_back(arg); - xbt_assert(mc_model_checker == nullptr, "Did you manage to start the MC twice in this process?"); + checker_side_ = std::make_unique(app_args_, need_memory_introspection); - auto process = std::make_unique(pid); - model_checker_ = std::make_unique(std::move(process), sockets[1]); - - mc_model_checker = model_checker_.get(); - model_checker_->start(); - - /* Take the initial snapshot */ - model_checker_->wait_for_requests(); - initial_snapshot_ = std::make_shared(0, page_store_); + if (need_memory_introspection) + initial_snapshot_ = std::make_shared(0, page_store_, *checker_side_->get_remote_memory()); } RemoteApp::~RemoteApp() { initial_snapshot_ = nullptr; - if (model_checker_) { - model_checker_->shutdown(); - model_checker_ = nullptr; - mc_model_checker = nullptr; - } + checker_side_ = nullptr; } -void RemoteApp::restore_initial_state() const +void RemoteApp::restore_initial_state() { - this->initial_snapshot_->restore(&model_checker_->get_remote_process()); + if (initial_snapshot_ == nullptr) { // No memory introspection + // We need to destroy the existing CheckerSide before creating the new one, or libevent gets crazy + checker_side_.reset(nullptr); + checker_side_.reset(new simgrid::mc::CheckerSide(app_args_, true)); + } else + initial_snapshot_->restore(*checker_side_->get_remote_memory()); } unsigned long RemoteApp::get_maxpid() const { - return model_checker_->get_remote_process().get_maxpid(); + // note: we could maybe cache it and count the actor creation on checker side too. + // But counting correctly accross state checkpoint/restore would be annoying. + + checker_side_->get_channel().send(MessageType::ACTORS_MAXPID); + s_mc_message_int_t answer; + ssize_t answer_size = checker_side_->get_channel().receive(answer); + xbt_assert(answer_size != -1, "Could not receive message"); + xbt_assert(answer_size == sizeof answer, "Broken message (size=%zd; expected %zu)", answer_size, sizeof answer); + xbt_assert(answer.type == MessageType::ACTORS_MAXPID_REPLY, + "Received unexpected message %s (%i); expected MessageType::ACTORS_MAXPID_REPLY (%i)", + to_c_str(answer.type), (int)answer.type, (int)MessageType::ACTORS_MAXPID_REPLY); + + return answer.value; } void RemoteApp::get_actors_status(std::map& whereto) const @@ -168,16 +82,15 @@ void RemoteApp::get_actors_status(std::map& whereto) const // <----- send ACTORS_STATUS_REPLY // <----- send `N` `s_mc_message_actors_status_one_t` structs // <----- send `M` `s_mc_message_simcall_probe_one_t` structs - model_checker_->channel().send(MessageType::ACTORS_STATUS); + checker_side_->get_channel().send(MessageType::ACTORS_STATUS); s_mc_message_actors_status_answer_t answer; - ssize_t answer_size = model_checker_->channel().receive(answer); + ssize_t answer_size = checker_side_->get_channel().receive(answer); xbt_assert(answer_size != -1, "Could not receive message"); - xbt_assert(answer_size == sizeof answer && answer.type == MessageType::ACTORS_STATUS_REPLY, - "Received unexpected message %s (%i, size=%zd) " - "expected MessageType::ACTORS_STATUS_REPLY (%i, size=%zu)", - to_c_str(answer.type), (int)answer.type, answer_size, (int)MessageType::ACTORS_STATUS_REPLY, - sizeof answer); + xbt_assert(answer_size == sizeof answer, "Broken message (size=%zd; expected %zu)", answer_size, sizeof answer); + xbt_assert(answer.type == MessageType::ACTORS_STATUS_REPLY, + "Received unexpected message %s (%i); expected MessageType::ACTORS_STATUS_REPLY (%i)", + to_c_str(answer.type), (int)answer.type, (int)MessageType::ACTORS_STATUS_REPLY); // Message sanity checks xbt_assert(answer.count >= 0, "Received an ACTOR_STATUS_REPLY message with an actor count of '%d' < 0", answer.count); @@ -190,7 +103,7 @@ void RemoteApp::get_actors_status(std::map& whereto) const std::vector status(answer.count); if (answer.count > 0) { size_t size = status.size() * sizeof(s_mc_message_actors_status_one_t); - ssize_t received = model_checker_->channel().receive(status.data(), size); + ssize_t received = checker_side_->get_channel().receive(status.data(), size); xbt_assert(static_cast(received) == size); } @@ -206,7 +119,7 @@ void RemoteApp::get_actors_status(std::map& whereto) const std::vector probes(answer.transition_count); if (answer.transition_count > 0) { for (auto& probe : probes) { - ssize_t received = model_checker_->channel().receive(probe); + ssize_t received = checker_side_->get_channel().receive(probe); xbt_assert(received >= 0, "Could not receive response to ACTORS_PROBE message (%s)", strerror(errno)); xbt_assert(static_cast(received) == sizeof probe, "Could not receive response to ACTORS_PROBE message (%zd bytes received != %zu bytes expected", @@ -237,25 +150,64 @@ void RemoteApp::get_actors_status(std::map& whereto) const void RemoteApp::check_deadlock() const { - xbt_assert(model_checker_->channel().send(MessageType::DEADLOCK_CHECK) == 0, "Could not check deadlock state"); + xbt_assert(checker_side_->get_channel().send(MessageType::DEADLOCK_CHECK) == 0, "Could not check deadlock state"); s_mc_message_int_t message; - ssize_t received = model_checker_->channel().receive(message); + ssize_t received = checker_side_->get_channel().receive(message); xbt_assert(received != -1, "Could not receive message"); - xbt_assert(received == sizeof message && message.type == MessageType::DEADLOCK_CHECK_REPLY, - "Received unexpected message %s (%i, size=%zd) " - "expected MessageType::DEADLOCK_CHECK_REPLY (%i, size=%zu)", - to_c_str(message.type), (int)message.type, received, (int)MessageType::DEADLOCK_CHECK_REPLY, - sizeof message); + xbt_assert(received == sizeof message, "Broken message (size=%zd; expected %zu)", received, sizeof message); + xbt_assert(message.type == MessageType::DEADLOCK_CHECK_REPLY, + "Received unexpected message %s (%i); expected MessageType::DEADLOCK_CHECK_REPLY (%i)", + to_c_str(message.type), (int)message.type, (int)MessageType::DEADLOCK_CHECK_REPLY); if (message.value != 0) { + auto* explo = Exploration::get_instance(); XBT_CINFO(mc_global, "Counter-example execution trace:"); - for (auto const& frame : model_checker_->get_exploration()->get_textual_trace()) + for (auto const& frame : explo->get_textual_trace()) XBT_CINFO(mc_global, " %s", frame.c_str()); XBT_INFO("You can debug the problem (and see the whole details) by rerunning out of simgrid-mc with " "--cfg=model-check/replay:'%s'", - model_checker_->get_exploration()->get_record_trace().to_string().c_str()); - model_checker_->get_exploration()->log_state(); + explo->get_record_trace().to_string().c_str()); + explo->log_state(); throw DeadlockError(); } } + +void RemoteApp::wait_for_requests() +{ + checker_side_->wait_for_requests(); +} + +Transition* RemoteApp::handle_simcall(aid_t aid, int times_considered, bool new_transition) +{ + s_mc_message_simcall_execute_t m = {}; + m.type = MessageType::SIMCALL_EXECUTE; + m.aid_ = aid; + m.times_considered_ = times_considered; + checker_side_->get_channel().send(m); + + if (auto* memory = get_remote_process_memory(); memory != nullptr) + memory->clear_cache(); + if (checker_side_->running()) + checker_side_->dispatch_events(); // The app may send messages while processing the transition + + s_mc_message_simcall_execute_answer_t answer; + ssize_t s = checker_side_->get_channel().receive(answer); + xbt_assert(s != -1, "Could not receive message"); + xbt_assert(s == sizeof answer, "Broken message (size=%zd; expected %zu)", s, sizeof answer); + xbt_assert(answer.type == MessageType::SIMCALL_EXECUTE_ANSWER, + "Received unexpected message %s (%i); expected MessageType::SIMCALL_EXECUTE_ANSWER (%i)", + to_c_str(answer.type), (int)answer.type, (int)MessageType::SIMCALL_EXECUTE_ANSWER); + + if (new_transition) { + std::stringstream stream(answer.buffer.data()); + return deserialize_transition(aid, times_considered, stream); + } else + return nullptr; +} + +void RemoteApp::finalize_app(bool terminate_asap) +{ + checker_side_->finalize(terminate_asap); +} + } // namespace simgrid::mc