X-Git-Url: http://bilbo.iut-bm.univ-fcomte.fr/pub/gitweb/simgrid.git/blobdiff_plain/16f7a7fb199ebea9d85106b74a5e7378038fc6df..d68e1c39ec0832cb2391aedd17a868c597dd399e:/src/mc/api/RemoteApp.cpp diff --git a/src/mc/api/RemoteApp.cpp b/src/mc/api/RemoteApp.cpp index c680ae4101..dc6c7d0d54 100644 --- a/src/mc/api/RemoteApp.cpp +++ b/src/mc/api/RemoteApp.cpp @@ -17,164 +17,63 @@ #include #include -#include +#include #include #include #include - -#include -#ifdef __linux__ -#include -#endif #include +#include #include -#ifdef __linux__ -#define WAITPID_CHECKED_FLAGS __WALL -#else -#define WAITPID_CHECKED_FLAGS 0 -#endif - 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) +static std::string master_socket_name; + +RemoteApp::RemoteApp(const std::vector& args) : app_args_(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 - */ + 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)); + 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__ - // 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"); + 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 - // 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++; + struct sockaddr_un serv_addr = {}; + serv_addr.sun_family = AF_UNIX; + master_socket_name.copy(serv_addr.sun_path, MC_SOCKET_NAME_LEN); - xbt_assert(args[i] != nullptr, - "Unable to find a binary to exec on the command line. Did you only pass config flags?"); + 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)); - 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_assert(listen(master_socket_, SOMAXCONN) >= 0, "Cannot listen to the master socket: %s.", strerror(errno)); - xbt_die("Aborting now."); + application_factory_ = std::make_unique(app_args_); + checker_side_ = application_factory_->clone(master_socket_, master_socket_name); } -RemoteApp::RemoteApp(const std::vector& args) +void RemoteApp::restore_initial_state() { - // 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]); - - xbt_assert(mc_model_checker == nullptr, "Did you manage to start the MC twice in this process?"); - - auto memory = std::make_unique(pid); - model_checker_ = std::make_unique(); - mc_model_checker = model_checker_.get(); - checker_side_ = std::make_unique(sockets[1], std::move(memory), model_checker_.get()); - - start(); - - /* Take the initial snapshot */ - wait_for_requests(); - initial_snapshot_ = std::make_shared(0, page_store_, checker_side_->get_remote_memory()); -} - -RemoteApp::~RemoteApp() -{ - initial_snapshot_ = nullptr; - if (model_checker_) { - shutdown(); - model_checker_ = nullptr; - mc_model_checker = nullptr; - } -} -void RemoteApp::start() -{ - XBT_DEBUG("Waiting for the model-checked process"); - int status; - - // The model-checked process SIGSTOP itself to signal it's ready: - const pid_t pid = get_remote_process_memory().pid(); - - xbt_assert(waitpid(pid, &status, WAITPID_CHECKED_FLAGS) == pid && WIFSTOPPED(status) && WSTOPSIG(status) == SIGSTOP, - "Could not wait model-checked process"); - - errno = 0; -#ifdef __linux__ - ptrace(PTRACE_SETOPTIONS, pid, nullptr, PTRACE_O_TRACEEXIT); - ptrace(PTRACE_CONT, pid, 0, 0); -#elif defined BSD - ptrace(PT_CONTINUE, pid, (caddr_t)1, 0); -#else -#error "no ptrace equivalent coded for this platform" -#endif - xbt_assert(errno == 0, - "Ptrace does not seem to be usable in your setup (errno: %d). " - "If you run from within a docker, adding `--cap-add SYS_PTRACE` to the docker line may help. " - "If it does not help, please report this bug.", - errno); -} -void RemoteApp::restore_initial_state() const -{ - this->initial_snapshot_->restore(checker_side_->get_remote_memory()); + checker_side_ = application_factory_->clone(master_socket_, master_socket_name); } unsigned long RemoteApp::get_maxpid() const @@ -200,26 +99,26 @@ 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 + // <----- 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 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, - "Received unexpected message %s (%i); expected MessageType::ACTORS_STATUS_REPLY (%i)", - to_c_str(answer.type), (int)answer.type, (int)MessageType::ACTORS_STATUS_REPLY); + 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 ACTOR_STATUS_REPLY message with an actor count of '%d' < 0", answer.count); - xbt_assert(answer.transition_count >= 0, "Received an ACTOR_STATUS_REPLY message with transition_count '%d' < 0", - answer.transition_count); - xbt_assert(answer.transition_count == 0 || answer.count >= 0, - "Received an ACTOR_STATUS_REPLY message with no actor data " - "but with transition data nonetheless"); + 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) { @@ -228,43 +127,25 @@ void RemoteApp::get_actors_status(std::map& whereto) const xbt_assert(static_cast(received) == size); } - // Ensures that each actor sends precisely `answer.transition_count` 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 == answer.transition_count, - "Expected to receive %d transition(s) but was only notified of %d by the app side", expected_transitions, - answer.transition_count); - - std::vector probes(answer.transition_count); - if (answer.transition_count > 0) { - for (auto& probe : probes) { + whereto.clear(); + + for (const auto& actor : status) { + 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); - } - } - whereto.clear(); - std::move_iterator probes_iter(probes.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::vector> actor_transitions; - for (int times_considered = 0; times_considered < actor.n_transitions; times_considered++, probes_iter++) { - std::stringstream stream((*probes_iter).buffer.data()); + 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", actor_transitions.size(), actor.aid); + 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)); } } @@ -289,32 +170,13 @@ void RemoteApp::check_deadlock() const "--cfg=model-check/replay:'%s'", explo->get_record_trace().to_string().c_str()); explo->log_state(); - throw DeadlockError(); + throw McError(ExitStatus::DEADLOCK); } } void RemoteApp::wait_for_requests() { - /* Resume the application */ - if (checker_side_->get_channel().send(MessageType::CONTINUE) != 0) - throw xbt::errno_error(); - get_remote_process_memory().clear_cache(); - - if (this->get_remote_process_memory().running()) - checker_side_->dispatch_events(); -} - -void RemoteApp::shutdown() -{ - XBT_DEBUG("Shutting down model-checker"); - - RemoteProcessMemory& process = get_remote_process_memory(); - if (process.running()) { - XBT_DEBUG("Killing process"); - finalize_app(true); - kill(process.pid(), SIGKILL); - process.terminate(); - } + checker_side_->wait_for_requests(); } Transition* RemoteApp::handle_simcall(aid_t aid, int times_considered, bool new_transition) @@ -325,17 +187,16 @@ Transition* RemoteApp::handle_simcall(aid_t aid, int times_considered, bool new_ m.times_considered_ = times_considered; checker_side_->get_channel().send(m); - get_remote_process_memory().clear_cache(); - if (this->get_remote_process_memory().running()) + 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); - 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()); @@ -346,18 +207,7 @@ Transition* RemoteApp::handle_simcall(aid_t aid, int times_considered, bool new_ void RemoteApp::finalize_app(bool terminate_asap) { - s_mc_message_int_t m = {}; - m.type = MessageType::FINALIZE; - m.value = terminate_asap; - xbt_assert(checker_side_->get_channel().send(m) == 0, "Could not ask the app to finalize on need"); - - s_mc_message_t answer; - ssize_t s = checker_side_->get_channel().receive(answer); - xbt_assert(s != -1, "Could not receive answer to FINALIZE"); - xbt_assert(s == sizeof answer, "Broken message (size=%zd; expected %zu)", s, sizeof answer); - xbt_assert(answer.type == MessageType::FINALIZE_REPLY, - "Received unexpected message %s (%i); expected MessageType::FINALIZE_REPLY (%i)", to_c_str(answer.type), - (int)answer.type, (int)MessageType::FINALIZE_REPLY); + checker_side_->finalize(terminate_asap); } } // namespace simgrid::mc