X-Git-Url: http://bilbo.iut-bm.univ-fcomte.fr/pub/gitweb/simgrid.git/blobdiff_plain/ab209b5fb2592a3287497a9ea08773471005192f..a7cc502f224afb96f0203b94759cd0dabebf8de5:/src/mc/api/RemoteApp.cpp diff --git a/src/mc/api/RemoteApp.cpp b/src/mc/api/RemoteApp.cpp index b56d54d472..dc6c7d0d54 100644 --- a/src/mc/api/RemoteApp.cpp +++ b/src/mc/api/RemoteApp.cpp @@ -4,159 +4,93 @@ * 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 - -#include -#ifdef __linux__ -#include -#endif +#include +#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?"); +static std::string master_socket_name; - 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) : app_args_(args) { -#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"); + master_socket_ = socket(AF_UNIX, +#ifdef __APPLE__ + SOCK_STREAM, /* Mac OSX does not have AF_UNIX + SOCK_SEQPACKET, even if that's faster */ +#else + SOCK_SEQPACKET, #endif + 0); + xbt_assert(master_socket_ != -1, "Cannot create the master socket: %s", strerror(errno)); - // 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]); + master_socket_name = "/tmp/simgrid-mc-" + std::to_string(getpid()); + master_socket_name.resize(MC_SOCKET_NAME_LEN); // truncate socket name if it's too long + master_socket_name.back() = '\0'; // ensure the data are null-terminated +#ifdef __linux__ + master_socket_name[0] = '\0'; // abstract socket, automatically removed after close +#else + unlink(master_socket_name.c_str()); // remove possible stale socket before bind + atexit([]() { + if (not master_socket_name.empty()) + unlink(master_socket_name.c_str()); + master_socket_name.clear(); + }); +#endif - xbt_assert(mc_model_checker == nullptr, "Did you manage to start the MC twice in this process?"); + struct sockaddr_un serv_addr = {}; + serv_addr.sun_family = AF_UNIX; + master_socket_name.copy(serv_addr.sun_path, MC_SOCKET_NAME_LEN); - auto process = std::make_unique(pid); - model_checker_ = std::make_unique(std::move(process), sockets[1]); + xbt_assert(bind(master_socket_, (struct sockaddr*)&serv_addr, sizeof serv_addr) >= 0, + "Cannot bind the master socket to %c%s: %s.", (serv_addr.sun_path[0] ? serv_addr.sun_path[0] : '@'), + serv_addr.sun_path + 1, strerror(errno)); - mc_model_checker = model_checker_.get(); - model_checker_->start(); + xbt_assert(listen(master_socket_, SOMAXCONN) >= 0, "Cannot listen to the master socket: %s.", strerror(errno)); - /* Take the initial snapshot */ - model_checker_->wait_for_requests(); - initial_snapshot_ = std::make_shared(0); + application_factory_ = std::make_unique(app_args_); + checker_side_ = application_factory_->clone(master_socket_, master_socket_name); } -RemoteApp::~RemoteApp() +void RemoteApp::restore_initial_state() { - initial_snapshot_ = nullptr; - if (model_checker_) { - model_checker_->shutdown(); - model_checker_ = nullptr; - mc_model_checker = nullptr; - } -} - -void RemoteApp::restore_initial_state() const -{ - this->initial_snapshot_->restore(&model_checker_->get_remote_process()); + checker_side_ = application_factory_->clone(master_socket_, master_socket_name); } 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 @@ -165,89 +99,115 @@ void RemoteApp::get_actors_status(std::map& whereto) const // // CheckerSide AppSide // send ACTORS_STATUS ----> - // <----- send ACTORS_STATUS_REPLY - // <----- send `N` `s_mc_message_actors_status_one_t` structs - // <----- send `M` `s_mc_message_simcall_probe_one_t` structs - s_mc_message_t msg; - memset(&msg, 0, sizeof msg); - msg.type = simgrid::mc::MessageType::ACTORS_STATUS; - model_checker_->channel().send(msg); + // <----- send ACTORS_STATUS_REPLY_COUNT + // <----- send `N` ACTORS_STATUS_REPLY_TRANSITION (s_mc_message_actors_status_one_t) + // <----- send `M` ACTORS_STATUS_REPLY_SIMCALL (s_mc_message_simcall_probe_one_t) + // + // Note that we also receive disabled transitions, because the guiding strategies need them to decide what could + // unlock actors. + + checker_side_->get_channel().send(MessageType::ACTORS_STATUS); s_mc_message_actors_status_answer_t answer; - ssize_t received = model_checker_->channel().receive(answer); - xbt_assert(received != -1, "Could not receive message"); - xbt_assert(received == sizeof(answer) && answer.type == MessageType::ACTORS_STATUS_REPLY, - "Received unexpected message %s (%i, size=%i) " - "expected MessageType::ACTORS_STATUS_REPLY (%i, size=%i)", - to_c_str(answer.type), (int)answer.type, (int)received, (int)MessageType::ACTORS_STATUS_REPLY, - (int)sizeof(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_STATUS_REPLY_COUNT, + "%d Received unexpected message %s (%i); expected MessageType::ACTORS_STATUS_REPLY_COUNT (%i)", getpid(), + to_c_str(answer.type), (int)answer.type, (int)MessageType::ACTORS_STATUS_REPLY_COUNT); + + // Message sanity checks + xbt_assert(answer.count >= 0, "Received an ACTORS_STATUS_REPLY_COUNT message with an actor count of '%d' < 0", + answer.count); std::vector status(answer.count); if (answer.count > 0) { - size_t size = status.size() * sizeof(s_mc_message_actors_status_one_t); - received = model_checker_->channel().receive(status.data(), size); - xbt_assert(static_cast(received) == size); - } - - std::vector action_pool(answer.transition_count); - if (answer.transition_count > 0) { - size_t size = action_pool.size() * sizeof(s_mc_message_simcall_probe_one_t); - received = model_checker_->channel().receive(action_pool.data(), size); + size_t size = status.size() * sizeof(s_mc_message_actors_status_one_t); + ssize_t received = checker_side_->get_channel().receive(status.data(), size); xbt_assert(static_cast(received) == size); } - // Ensures that each actor sends precisely `actor.max_considered` transitions. While technically - // this doesn't catch the edge case where actor A sends 3 instead of 2 and actor B sends 2 instead - // of 3 transitions, that is ignored here since that invariant needs to be enforced on the AppSide - const auto expected_transitions = std::accumulate( - status.begin(), status.end(), 0, [](int total, const auto& actor) { return total + actor.n_transitions; }); - xbt_assert(expected_transitions == action_pool.size(), - "Expected to receive %d transition(s) but was only notified of %lu by the app side", expected_transitions, - action_pool.size()); - whereto.clear(); - auto action_pool_iter = std::move_iterator(action_pool.begin()); for (const auto& actor : status) { - xbt_assert(actor.n_transitions == 0 || actor.n_transitions == actor.max_considered, - "If any transitions are serialized for an actor, it must match the " - "total number of transitions that can be considered for the actor " - "(currently %d), but only %d transition(s) was/were said to be encoded", - actor.max_considered, actor.n_transitions); - - std::stringstream stream((*action_pool_iter).buffer.data()); - auto actor_transitions = std::vector>(actor.max_considered); - - for (int times_considered = 0; times_considered < actor.max_considered; times_considered++, action_pool_iter++) { - auto transition = std::unique_ptr(deserialize_transition(actor.aid, times_considered, stream)); - actor_transitions.push_back(std::move(transition)); + std::vector> actor_transitions; + int n_transitions = actor.max_considered; + for (int times_considered = 0; times_considered < n_transitions; times_considered++) { + s_mc_message_simcall_probe_one_t 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", + received, sizeof probe); + + std::stringstream stream(probe.buffer.data()); + actor_transitions.emplace_back(deserialize_transition(actor.aid, times_considered, stream)); } + XBT_DEBUG("Received %zu transitions for actor %ld. The first one is %s", actor_transitions.size(), actor.aid, + (actor_transitions.size() > 0 ? actor_transitions[0]->to_string().c_str() : "null")); whereto.try_emplace(actor.aid, actor.aid, actor.enabled, actor.max_considered, std::move(actor_transitions)); } } 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 s = model_checker_->channel().receive(message); - xbt_assert(s != -1, "Could not receive message"); - xbt_assert(s == sizeof(message) && message.type == MessageType::DEADLOCK_CHECK_REPLY, - "Received unexpected message %s (%i, size=%i) " - "expected MessageType::DEADLOCK_CHECK_REPLY (%i, size=%i)", - to_c_str(message.type), (int)message.type, (int)s, (int)MessageType::DEADLOCK_CHECK_REPLY, - (int)sizeof(message)); + ssize_t received = checker_side_->get_channel().receive(message); + xbt_assert(received != -1, "Could not receive 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(); - throw DeadlockError(); + explo->get_record_trace().to_string().c_str()); + explo->log_state(); + throw McError(ExitStatus::DEADLOCK); } } + +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 (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 > 0 && answer.type == MessageType::SIMCALL_EXECUTE_REPLY, + "%d Received unexpected message %s (%i); expected MessageType::SIMCALL_EXECUTE_REPLY (%i)", getpid(), + to_c_str(answer.type), (int)answer.type, (int)MessageType::SIMCALL_EXECUTE_REPLY); + xbt_assert(s == sizeof answer, "Broken message (size=%zd; expected %zu)", s, sizeof 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