#include "service.h"
#include "dinit-log.h"
#include "dinit-socket.h"
+#include "dinit-util.h"
/*
* service.cc - Service management.
// from dinit.cc:
void open_control_socket(bool report_ro_failure = true) noexcept;
void setup_external_log() noexcept;
-extern EventLoop_t eventLoop;
+extern eventloop_t eventLoop;
// Find the requested service by name
-static ServiceRecord * find_service(const std::list<ServiceRecord *> & records,
+static service_record * find_service(const std::list<service_record *> & records,
const char *name) noexcept
{
using std::list;
- list<ServiceRecord *>::const_iterator i = records.begin();
+ list<service_record *>::const_iterator i = records.begin();
for ( ; i != records.end(); i++ ) {
- if (strcmp((*i)->getServiceName().c_str(), name) == 0) {
+ if (strcmp((*i)->get_service_name().c_str(), name) == 0) {
return *i;
}
}
- return (ServiceRecord *)0;
+ return nullptr;
}
-ServiceRecord * ServiceSet::find_service(const std::string &name) noexcept
+service_record * service_set::find_service(const std::string &name) noexcept
{
return ::find_service(records, name.c_str());
}
-void ServiceSet::startService(const char *name)
+void service_set::stop_service(const std::string & name) noexcept
{
- using namespace std;
- ServiceRecord *record = loadServiceRecord(name);
-
- record->start();
- processQueues(true);
-}
-
-void ServiceSet::stopService(const std::string & name) noexcept
-{
- ServiceRecord *record = find_service(name);
+ service_record *record = find_service(name);
if (record != nullptr) {
record->stop();
- processQueues(false);
+ process_queues();
}
}
-// Called when a service has actually stopped; dependents have stopped already.
-void ServiceRecord::stopped() noexcept
+// Called when a service has actually stopped; dependents have stopped already, unless this stop
+// is due to an unexpected process termination.
+void service_record::stopped() noexcept
{
- if (service_type != ServiceType::SCRIPTED && service_type != ServiceType::BGPROCESS && onstart_flags.runs_on_console) {
+ if (onstart_flags.runs_on_console) {
tcsetpgrp(0, getpgrp());
discard_console_log_buffer();
- releaseConsole();
+ release_console();
}
force_stop = false;
// If we are a soft dependency of another target, break the acquisition from that target now:
- bool will_restart = (desired_state == ServiceState::STARTED) && service_set->get_auto_restart();
- if (! will_restart) {
- for (auto dependency : soft_dpts) {
- if (dependency->holding_acq) {
- dependency->holding_acq = false;
+ for (auto & dependent : dependents) {
+ if (dependent->dep_type == dependency_type::SOFT) {
+ if (dependent->holding_acq) {
+ dependent->holding_acq = false;
release();
}
}
}
- will_restart &= (desired_state == ServiceState::STARTED);
+ bool will_restart = (desired_state == service_state_t::STARTED)
+ && services->get_auto_restart();
+
for (auto dependency : depends_on) {
- if (! will_restart || ! dependency->can_interrupt_stop()) {
- dependency->dependentStopped();
- }
+ // we signal dependencies in case they are waiting for us to stop:
+ dependency.get_to()->dependent_stopped();
}
+ service_state = service_state_t::STOPPED;
+
if (will_restart) {
// Desired state is "started".
- service_state = ServiceState::STOPPED;
- service_set->addToStartQueue(this);
+ restarting = true;
+ start(false);
}
else {
if (socket_fd != -1) {
start_explicit = false;
release();
}
-
- service_state = ServiceState::STOPPED;
- if (required_by == 0) {
- // Since state wasn't STOPPED until now, any release performed above won't have marked
- // the service inactive. We check for that now:
- service_set->service_inactive(this);
+ else if (required_by == 0) {
+ services->service_inactive(this);
}
}
- logServiceStopped(service_name);
- notifyListeners(ServiceEvent::STOPPED);
+ log_service_stopped(service_name);
+ notify_listeners(service_event::STOPPED);
}
-dasynq::rearm ServiceChildWatcher::child_status(EventLoop_t &loop, pid_t child, int status) noexcept
+dasynq::rearm service_child_watcher::status_change(eventloop_t &loop, pid_t child, int status) noexcept
{
- ServiceRecord *sr = service;
+ base_process_service *sr = service;
sr->pid = -1;
sr->exit_status = status;
if (sr->waiting_for_execstat) {
// We still don't have an exec() status from the forked child, wait for that
// before doing any further processing.
- return rearm::REMOVE;
+ return rearm::NOOP; // hold watch reservation
}
- // Must deregister now since handle_exit_status might result in re-launch:
- deregister(loop, child);
+ // Must stop watch now since handle_exit_status might result in re-launch:
+ // (stop_watch instead of deregister, so that we hold watch reservation).
+ stop_watch(loop);
+ if (sr->stop_timer_armed) {
+ sr->restart_timer.stop_timer(loop);
+ sr->stop_timer_armed = false;
+ }
+
sr->handle_exit_status(status);
- return rearm::REMOVED;
+ return rearm::NOOP;
}
-bool ServiceRecord::do_auto_restart() noexcept
+bool service_record::do_auto_restart() noexcept
{
if (auto_restart) {
- return service_set->get_auto_restart();
+ return services->get_auto_restart();
}
return false;
}
-void ServiceRecord::handle_exit_status(int exit_status) noexcept
+void service_record::emergency_stop() noexcept
{
- // TODO make abstract
+ if (! do_auto_restart() && start_explicit) {
+ start_explicit = false;
+ release();
+ }
+ forced_stop();
+ stop_dependents();
+ stopped();
}
void process_service::handle_exit_status(int exit_status) noexcept
{
bool did_exit = WIFEXITED(exit_status);
bool was_signalled = WIFSIGNALED(exit_status);
+ restarting = false;
- if (exit_status != 0 && service_state != ServiceState::STOPPING) {
+ if (exit_status != 0 && service_state != service_state_t::STOPPING) {
if (did_exit) {
- log(LogLevel::ERROR, "Service ", service_name, " process terminated with exit code ", WEXITSTATUS(exit_status));
+ log(loglevel_t::ERROR, "Service ", service_name, " process terminated with exit code ",
+ WEXITSTATUS(exit_status));
}
else if (was_signalled) {
- log(LogLevel::ERROR, "Service ", service_name, " terminated due to signal ", WTERMSIG(exit_status));
+ log(loglevel_t::ERROR, "Service ", service_name, " terminated due to signal ",
+ WTERMSIG(exit_status));
}
}
- if (service_state == ServiceState::STARTING) {
+ if (service_state == service_state_t::STARTING) {
if (did_exit && WEXITSTATUS(exit_status) == 0) {
started();
}
failed_to_start();
}
}
- else if (service_state == ServiceState::STOPPING) {
+ else if (service_state == service_state_t::STOPPING) {
// We won't log a non-zero exit status or termination due to signal here -
// we assume that the process died because we signalled it.
stopped();
}
- else if (smooth_recovery && service_state == ServiceState::STARTED && desired_state == ServiceState::STARTED) {
- // TODO ensure a minimum time between restarts
- // TODO if we are pinned-started then we should probably check
- // that dependencies have started before trying to re-start the
- // service process.
- start_ps_process();
+ else if (smooth_recovery && service_state == service_state_t::STARTED
+ && desired_state == service_state_t::STARTED) {
+ do_smooth_recovery();
return;
}
else {
- if (! do_auto_restart()) desired_state = ServiceState::STOPPED;
- forceStop();
+ emergency_stop();
+ }
+ services->process_queues();
+}
+
+void base_process_service::do_smooth_recovery() noexcept
+{
+ if (! restart_ps_process()) {
+ emergency_stop();
+ services->process_queues();
}
- service_set->processQueues(false);
}
void bgproc_service::handle_exit_status(int exit_status) noexcept
{
+ begin:
bool did_exit = WIFEXITED(exit_status);
bool was_signalled = WIFSIGNALED(exit_status);
- if (exit_status != 0 && service_state != ServiceState::STOPPING) {
+ if (exit_status != 0 && service_state != service_state_t::STOPPING) {
if (did_exit) {
- log(LogLevel::ERROR, "Service ", service_name, " process terminated with exit code ", WEXITSTATUS(exit_status));
+ log(loglevel_t::ERROR, "Service ", service_name, " process terminated with exit code ",
+ WEXITSTATUS(exit_status));
}
else if (was_signalled) {
- log(LogLevel::ERROR, "Service ", service_name, " terminated due to signal ", WTERMSIG(exit_status));
+ log(loglevel_t::ERROR, "Service ", service_name, " terminated due to signal ",
+ WTERMSIG(exit_status));
}
}
- if (doing_recovery) {
- // (BGPROCESS only)
- doing_recovery = false;
+ // This may be a "smooth recovery" where we are restarting the process while leaving the
+ // service in the STARTED state.
+ if (restarting && service_state == service_state_t::STARTED) {
+ restarting = false;
bool need_stop = false;
if ((did_exit && WEXITSTATUS(exit_status) != 0) || was_signalled) {
need_stop = true;
}
else {
// We need to re-read the PID, since it has now changed.
- if (service_type == ServiceType::BGPROCESS && pid_file.length() != 0) {
- if (! read_pid_file()) {
- need_stop = true;
+ if (pid_file.length() != 0) {
+ auto pid_result = read_pid_file(&exit_status);
+ switch (pid_result) {
+ case pid_result_t::FAILED:
+ // Failed startup: no auto-restart.
+ need_stop = true;
+ break;
+ case pid_result_t::TERMINATED:
+ goto begin;
+ case pid_result_t::OK:
+ break;
}
}
}
if (need_stop) {
// Failed startup: no auto-restart.
- desired_state = ServiceState::STOPPED;
- forceStop();
- service_set->processQueues(false);
+ emergency_stop();
+ services->process_queues();
}
return;
}
- if (service_state == ServiceState::STARTING) {
+ restarting = false;
+ if (service_state == service_state_t::STARTING) {
// POSIX requires that if the process exited clearly with a status code of 0,
// the exit status value will be 0:
if (exit_status == 0) {
- started();
+ auto pid_result = read_pid_file(&exit_status);
+ switch (pid_result) {
+ case pid_result_t::FAILED:
+ // Failed startup: no auto-restart.
+ failed_to_start();
+ break;
+ case pid_result_t::TERMINATED:
+ // started, but immediately terminated
+ started();
+ goto begin;
+ case pid_result_t::OK:
+ started();
+ break;
+ }
}
else {
failed_to_start();
}
}
- else if (service_state == ServiceState::STOPPING) {
+ else if (service_state == service_state_t::STOPPING) {
// We won't log a non-zero exit status or termination due to signal here -
// we assume that the process died because we signalled it.
stopped();
}
- else if (smooth_recovery && service_state == ServiceState::STARTED && desired_state == ServiceState::STARTED) {
- // TODO ensure a minimum time between restarts
- // TODO if we are pinned-started then we should probably check
- // that dependencies have started before trying to re-start the
- // service process.
- doing_recovery = true;
- start_ps_process();
+ else if (smooth_recovery && service_state == service_state_t::STARTED && desired_state == service_state_t::STARTED) {
+ do_smooth_recovery();
return;
}
else {
- if (! do_auto_restart()) desired_state = ServiceState::STOPPED;
- forceStop();
+ // we must be STARTED
+ if (! do_auto_restart() && start_explicit) {
+ start_explicit = false;
+ release();
+ }
+ forced_stop();
+ stop_dependents();
+ stopped();
}
- service_set->processQueues(false);
+ services->process_queues();
}
void scripted_service::handle_exit_status(int exit_status) noexcept
bool did_exit = WIFEXITED(exit_status);
bool was_signalled = WIFSIGNALED(exit_status);
- if (service_state == ServiceState::STOPPING) {
+ if (service_state == service_state_t::STOPPING) {
if (did_exit && WEXITSTATUS(exit_status) == 0) {
stopped();
}
else {
// ??? failed to stop! Let's log it as info:
if (did_exit) {
- log(LogLevel::INFO, "Service ", service_name, " stop command failed with exit code ", WEXITSTATUS(exit_status));
+ log(loglevel_t::INFO, "Service ", service_name, " stop command failed with exit code ",
+ WEXITSTATUS(exit_status));
}
else if (was_signalled) {
- log(LogLevel::INFO, "Serivice ", service_name, " stop command terminated due to signal ", WTERMSIG(exit_status));
+ log(loglevel_t::INFO, "Serivice ", service_name, " stop command terminated due to signal ",
+ WTERMSIG(exit_status));
}
// Just assume that we stopped, so that any dependencies
// can be stopped:
stopped();
}
- service_set->processQueues(false);
+ services->process_queues();
}
else { // STARTING
if (exit_status == 0) {
else {
// failed to start
if (did_exit) {
- log(LogLevel::ERROR, "Service ", service_name, " command failed with exit code ", WEXITSTATUS(exit_status));
+ log(loglevel_t::ERROR, "Service ", service_name, " command failed with exit code ",
+ WEXITSTATUS(exit_status));
}
else if (was_signalled) {
- log(LogLevel::ERROR, "Service ", service_name, " command terminated due to signal ", WTERMSIG(exit_status));
+ log(loglevel_t::ERROR, "Service ", service_name, " command terminated due to signal ",
+ WTERMSIG(exit_status));
}
failed_to_start();
}
- service_set->processQueues(true);
+ services->process_queues();
}
}
-rearm ServiceIoWatcher::fd_event(EventLoop_t &loop, int fd, int flags) noexcept
+rearm exec_status_pipe_watcher::fd_event(eventloop_t &loop, int fd, int flags) noexcept
{
- ServiceRecord *sr = service;
+ base_process_service *sr = service;
sr->waiting_for_execstat = false;
int exec_status;
if (r > 0) {
// We read an errno code; exec() failed, and the service startup failed.
+ if (sr->pid != -1) {
+ sr->child_listener.deregister(eventLoop, sr->pid);
+ sr->reserved_child_watch = false;
+ if (sr->stop_timer_armed) {
+ sr->restart_timer.stop_timer(loop);
+ sr->stop_timer_armed = false;
+ }
+ }
sr->pid = -1;
- log(LogLevel::ERROR, sr->service_name, ": execution failed: ", strerror(exec_status));
- if (sr->service_state == ServiceState::STARTING) {
+ log(loglevel_t::ERROR, sr->service_name, ": execution failed: ", strerror(exec_status));
+ if (sr->service_state == service_state_t::STARTING) {
sr->failed_to_start();
}
- else if (sr->service_state == ServiceState::STOPPING) {
+ else if (sr->service_state == service_state_t::STOPPING) {
// Must be a scripted service. We've logged the failure, but it's probably better
// not to leave the service in STARTED state:
sr->stopped();
}
else {
// exec() succeeded.
- if (sr->service_type == ServiceType::PROCESS) {
+ if (sr->record_type == service_type::PROCESS) {
// This could be a smooth recovery (state already STARTED). Even more, the process
// might be stopped (and killed via a signal) during smooth recovery. We don't to
// process startup again in either case, so we check for state STARTING:
- if (sr->service_state == ServiceState::STARTING) {
+ if (sr->service_state == service_state_t::STARTING) {
sr->started();
}
}
}
}
- sr->service_set->processQueues(true);
+ sr->services->process_queues();
return rearm::REMOVED;
}
-void ServiceRecord::require() noexcept
+void service_record::require() noexcept
{
if (required_by++ == 0) {
-
- if (! prop_require) {
- prop_require = true;
- prop_release = false;
- service_set->addToPropQueue(this);
- }
-
- if (service_state == ServiceState::STOPPED) {
- // (In any other state, the service is already considered active.)
- service_set->service_active(this);
- }
+ prop_require = !prop_release;
+ prop_release = false;
+ services->add_prop_queue(this);
}
}
-void ServiceRecord::release() noexcept
+void service_record::release() noexcept
{
if (--required_by == 0) {
- desired_state = ServiceState::STOPPED;
- // Can stop, and can release dependencies now:
- prop_release = true;
+ desired_state = service_state_t::STOPPED;
+
+ // Can stop, and can release dependencies now. We don't need to issue a release if
+ // the require was pending though:
+ prop_release = !prop_require;
prop_require = false;
- service_set->addToPropQueue(this);
- if (service_state != ServiceState::STOPPED) {
- service_set->addToStopQueue(this);
+ services->add_prop_queue(this);
+
+ if (service_state == service_state_t::STOPPED) {
+ services->service_inactive(this);
}
else {
- service_set->service_inactive(this);
+ do_stop();
}
}
}
-void ServiceRecord::release_dependencies() noexcept
+void service_record::release_dependencies() noexcept
{
- for (sr_iter i = depends_on.begin(); i != depends_on.end(); ++i) {
- (*i)->release();
- }
-
- for (auto i = soft_deps.begin(); i != soft_deps.end(); ++i) {
- ServiceRecord * to = i->getTo();
- if (i->holding_acq) {
- to->release();
- i->holding_acq = false;
+ for (auto & dependency : depends_on) {
+ service_record * dep_to = dependency.get_to();
+ if (dependency.holding_acq) {
+ dep_to->release();
+ dependency.holding_acq = false;
}
}
}
-void ServiceRecord::start(bool activate) noexcept
+void service_record::start(bool activate) noexcept
{
if (activate && ! start_explicit) {
require();
start_explicit = true;
}
- if (desired_state == ServiceState::STARTED && service_state != ServiceState::STOPPED) return;
+ if (desired_state == service_state_t::STARTED && service_state != service_state_t::STOPPED) return;
+
+ bool was_active = service_state != service_state_t::STOPPED || desired_state != service_state_t::STOPPED;
+ desired_state = service_state_t::STARTED;
- desired_state = ServiceState::STARTED;
- service_set->addToStartQueue(this);
+ if (service_state != service_state_t::STOPPED) {
+ // We're already starting/started, or we are stopping and need to wait for
+ // that the complete.
+ if (service_state != service_state_t::STOPPING || ! can_interrupt_stop()) {
+ return;
+ }
+ // We're STOPPING, and that can be interrupted. Our dependencies might be STOPPING,
+ // but if so they are waiting (for us), so they too can be instantly returned to
+ // STARTING state.
+ notify_listeners(service_event::STOPCANCELLED);
+ }
+ else if (! was_active) {
+ services->service_active(this);
+ }
+
+ service_state = service_state_t::STARTING;
+ waiting_for_deps = true;
+
+ if (start_check_dependencies()) {
+ services->add_transition_queue(this);
+ }
}
-void ServiceRecord::do_propagation() noexcept
+void service_record::do_propagation() noexcept
{
if (prop_require) {
// Need to require all our dependencies
- for (sr_iter i = depends_on.begin(); i != depends_on.end(); ++i) {
- (*i)->require();
- }
-
- for (auto i = soft_deps.begin(); i != soft_deps.end(); ++i) {
- ServiceRecord * to = i->getTo();
- to->require();
- i->holding_acq = true;
+ for (auto & dep : depends_on) {
+ dep.get_to()->require();
+ dep.holding_acq = true;
}
-
prop_require = false;
}
failed_to_start(true);
}
- if (waiting_for_deps) {
- if (service_state == ServiceState::STARTING) {
- if (startCheckDependencies(false)) {
- allDepsStarted();
- }
- }
- else if (service_state == ServiceState::STOPPING) {
- if (stopCheckDependents()) {
- allDepsStopped();
- }
- }
+ if (prop_start) {
+ prop_start = false;
+ start(false);
+ }
+
+ if (prop_stop) {
+ prop_stop = false;
+ do_stop();
}
}
-void ServiceRecord::execute_transition() noexcept
+void service_record::execute_transition() noexcept
{
- bool is_started = (service_state == ServiceState::STARTED)
- || (service_state == ServiceState::STARTING && can_interrupt_start());
- bool is_stopped = (service_state == ServiceState::STOPPED)
- || (service_state == ServiceState::STOPPING && can_interrupt_stop());
-
- if (is_started && (desired_state == ServiceState::STOPPED || force_stop)) {
- if (! pinned_started) {
- do_stop();
+ // state is STARTED with restarting set true if we are running a smooth recovery.
+ if (service_state == service_state_t::STARTING || (service_state == service_state_t::STARTED
+ && restarting)) {
+ if (check_deps_started()) {
+ bool have_console = service_state == service_state_t::STARTED && onstart_flags.runs_on_console;
+ all_deps_started(have_console);
}
}
- else if (is_stopped && desired_state == ServiceState::STARTED) {
- if (! pinned_stopped) {
- do_start();
+ else if (service_state == service_state_t::STOPPING) {
+ if (stop_check_dependents()) {
+ all_deps_stopped();
}
}
}
-void ServiceRecord::do_start() noexcept
+void service_record::do_start() noexcept
{
if (pinned_stopped) return;
- if (service_state != ServiceState::STOPPED) {
- // We're already starting/started, or we are stopping and need to wait for
- // that the complete.
- if (service_state != ServiceState::STOPPING || ! can_interrupt_stop()) {
- return;
- }
- // We're STOPPING, and that can be interrupted. Our dependencies might be STOPPING,
- // but if so they are waiting (for us), so they too can be instantly returned to
- // STARTING state.
- notifyListeners(ServiceEvent::STOPCANCELLED);
+ if (service_state != service_state_t::STARTING) {
+ return;
}
- service_state = ServiceState::STARTING;
+ service_state = service_state_t::STARTING;
waiting_for_deps = true;
// Ask dependencies to start, mark them as being waited on.
- if (! startCheckDependencies(true)) {
- return;
+ if (check_deps_started()) {
+ // Once all dependencies are started, we start properly:
+ all_deps_started();
}
-
- // Actually start this service.
- allDepsStarted();
}
-void ServiceRecord::dependencyStarted() noexcept
+void service_record::dependencyStarted() noexcept
{
- if (service_state == ServiceState::STARTING && waiting_for_deps) {
- service_set->addToPropQueue(this);
+ if ((service_state == service_state_t::STARTING || service_state == service_state_t::STARTED)
+ && waiting_for_deps) {
+ services->add_transition_queue(this);
}
}
-bool ServiceRecord::startCheckDependencies(bool start_deps) noexcept
+bool service_record::start_check_dependencies() noexcept
{
bool all_deps_started = true;
- for (sr_iter i = depends_on.begin(); i != depends_on.end(); ++i) {
- if ((*i)->service_state != ServiceState::STARTED) {
- if (start_deps) {
- all_deps_started = false;
- (*i)->start(false);
- }
- else {
- return false;
+ for (auto & dep : depends_on) {
+ service_record * to = dep.get_to();
+ if (to->service_state != service_state_t::STARTED) {
+ if (to->service_state != service_state_t::STARTING) {
+ to->prop_start = true;
+ services->add_prop_queue(to);
}
+ dep.waiting_on = true;
+ all_deps_started = false;
}
}
+
+ return all_deps_started;
+}
- for (auto i = soft_deps.begin(); i != soft_deps.end(); ++i) {
- ServiceRecord * to = i->getTo();
- if (start_deps) {
- if (to->service_state != ServiceState::STARTED) {
- to->start(false);
- i->waiting_on = true;
- all_deps_started = false;
- }
- else {
- i->waiting_on = false;
- }
- }
- else if (i->waiting_on) {
- if (to->service_state != ServiceState::STARTING) {
- // Service has either started or is no longer starting
- i->waiting_on = false;
- }
- else {
- // We are still waiting on this service
- return false;
- }
+bool service_record::check_deps_started() noexcept
+{
+ for (auto & dep : depends_on) {
+ if (dep.waiting_on) {
+ return false;
}
}
-
- return all_deps_started;
+
+ return true;
}
-bool ServiceRecord::open_socket() noexcept
+bool service_record::open_socket() noexcept
{
if (socket_path.empty() || socket_fd != -1) {
// No socket, or already open
}
const char * saddrname = socket_path.c_str();
- uint sockaddr_size = offsetof(struct sockaddr_un, sun_path) + socket_path.length() + 1;
+
+ // Check the specified socket path
+ struct stat stat_buf;
+ if (stat(saddrname, &stat_buf) == 0) {
+ if ((stat_buf.st_mode & S_IFSOCK) == 0) {
+ // Not a socket
+ log(loglevel_t::ERROR, service_name, ": Activation socket file exists (and is not a socket)");
+ return false;
+ }
+ }
+ else if (errno != ENOENT) {
+ // Other error
+ log(loglevel_t::ERROR, service_name, ": Error checking activation socket: ", strerror(errno));
+ return false;
+ }
+
+ // Remove stale socket file (if it exists).
+ // We won't test the return from unlink - if it fails other than due to ENOENT, we should get an
+ // error when we try to create the socket anyway.
+ unlink(saddrname);
+ uint sockaddr_size = offsetof(struct sockaddr_un, sun_path) + socket_path.length() + 1;
struct sockaddr_un * name = static_cast<sockaddr_un *>(malloc(sockaddr_size));
if (name == nullptr) {
- log(LogLevel::ERROR, service_name, ": Opening activation socket: out of memory");
+ log(loglevel_t::ERROR, service_name, ": Opening activation socket: out of memory");
return false;
}
-
- // Un-link any stale socket. TODO: safety check? should at least confirm the path is a socket.
- unlink(saddrname);
name->sun_family = AF_UNIX;
strcpy(name->sun_path, saddrname);
int sockfd = dinit_socket(AF_UNIX, SOCK_STREAM, 0, SOCK_NONBLOCK | SOCK_CLOEXEC);
if (sockfd == -1) {
- log(LogLevel::ERROR, service_name, ": Error creating activation socket: ", strerror(errno));
+ log(loglevel_t::ERROR, service_name, ": Error creating activation socket: ", strerror(errno));
free(name);
return false;
}
if (bind(sockfd, (struct sockaddr *) name, sockaddr_size) == -1) {
- log(LogLevel::ERROR, service_name, ": Error binding activation socket: ", strerror(errno));
+ log(loglevel_t::ERROR, service_name, ": Error binding activation socket: ", strerror(errno));
close(sockfd);
free(name);
return false;
free(name);
- // POSIX (1003.1, 2013) says that fchown and fchmod don't necesarily work on sockets. We have to
+ // POSIX (1003.1, 2013) says that fchown and fchmod don't necessarily work on sockets. We have to
// use chown and chmod instead.
if (chown(saddrname, socket_uid, socket_gid)) {
- log(LogLevel::ERROR, service_name, ": Error setting activation socket owner/group: ", strerror(errno));
+ log(loglevel_t::ERROR, service_name, ": Error setting activation socket owner/group: ", strerror(errno));
close(sockfd);
return false;
}
if (chmod(saddrname, socket_perms) == -1) {
- log(LogLevel::ERROR, service_name, ": Error setting activation socket permissions: ", strerror(errno));
+ log(loglevel_t::ERROR, service_name, ": Error setting activation socket permissions: ", strerror(errno));
close(sockfd);
return false;
}
if (listen(sockfd, 128) == -1) { // 128 "seems reasonable".
- log(LogLevel::ERROR, ": Error listening on activation socket: ", strerror(errno));
+ log(loglevel_t::ERROR, ": Error listening on activation socket: ", strerror(errno));
close(sockfd);
return false;
}
return true;
}
-void ServiceRecord::allDepsStarted(bool has_console) noexcept
+void service_record::all_deps_started(bool has_console) noexcept
{
- if (onstart_flags.runs_on_console && ! has_console) {
+ if (onstart_flags.starts_on_console && ! has_console) {
waiting_for_deps = true;
- queueForConsole();
+ queue_for_console();
return;
}
waiting_for_deps = false;
+ // We overload can_interrupt_start to check whether there is any other
+ // process (eg restart timer) that needs to finish before starting.
+ if (can_interrupt_start()) {
+ waiting_for_deps = true;
+ return;
+ }
+
if (! open_socket()) {
failed_to_start();
}
- if (service_type == ServiceType::PROCESS || service_type == ServiceType::BGPROCESS
- || service_type == ServiceType::SCRIPTED) {
- bool start_success = start_ps_process();
- if (! start_success) {
- failed_to_start();
- }
- }
- else {
- // "internal" service
- started();
+ bool start_success = start_ps_process();
+ if (! start_success) {
+ failed_to_start();
}
}
-void ServiceRecord::acquiredConsole() noexcept
+void service_record::acquired_console() noexcept
{
- if (service_state != ServiceState::STARTING) {
+ if (service_state != service_state_t::STARTING) {
// We got the console but no longer want it.
- releaseConsole();
+ release_console();
}
- else if (startCheckDependencies(false)) {
- allDepsStarted(true);
+ else if (check_deps_started()) {
+ all_deps_started(true);
}
else {
// We got the console but can't use it yet.
- releaseConsole();
+ release_console();
}
}
-bool ServiceRecord::read_pid_file() noexcept
+bgproc_service::pid_result_t
+bgproc_service::read_pid_file(int *exit_status) noexcept
{
const char *pid_file_c = pid_file.c_str();
int fd = open(pid_file_c, O_CLOEXEC);
- if (fd != -1) {
- char pidbuf[21]; // just enought to hold any 64-bit integer
- int r = read(fd, pidbuf, 20);
- if (r > 0) {
- pidbuf[r] = 0; // store nul terminator
- pid = std::atoi(pidbuf);
- if (kill(pid, 0) == 0) {
- child_listener.add_watch(eventLoop, pid);
+ if (fd == -1) {
+ log(loglevel_t::ERROR, service_name, ": read pid file: ", strerror(errno));
+ return pid_result_t::FAILED;
+ }
+
+ char pidbuf[21]; // just enough to hold any 64-bit integer
+ int r = ss_read(fd, pidbuf, 20);
+ if (r < 0) {
+ // Could not read from PID file
+ log(loglevel_t::ERROR, service_name, ": could not read from pidfile; ", strerror(errno));
+ close(fd);
+ return pid_result_t::FAILED;
+ }
+
+ close(fd);
+ pidbuf[r] = 0; // store nul terminator
+
+ bool valid_pid = false;
+ try {
+ unsigned long long v = std::stoull(pidbuf, nullptr, 0);
+ if (v <= std::numeric_limits<pid_t>::max()) {
+ pid = (pid_t) v;
+ valid_pid = true;
+ }
+ }
+ catch (std::out_of_range &exc) {
+ // Too large?
+ }
+ catch (std::invalid_argument &exc) {
+ // Ok, so it doesn't look like a number: proceed...
+ }
+
+ if (valid_pid) {
+ pid_t wait_r = waitpid(pid, exit_status, WNOHANG);
+ if (wait_r == -1 && errno == ECHILD) {
+ // We can't track this child - check process exists:
+ if (kill(pid, 0) == 0 || errno != ESRCH) {
+ tracking_child = false;
+ return pid_result_t::OK;
}
else {
- log(LogLevel::ERROR, service_name, ": pid read from pidfile (", pid, ") is not valid");
+ log(loglevel_t::ERROR, service_name, ": pid read from pidfile (", pid, ") is not valid");
pid = -1;
- close(fd);
- return false;
+ return pid_result_t::FAILED;
}
}
- close(fd);
- return true;
- }
- else {
- log(LogLevel::ERROR, service_name, ": read pid file: ", strerror(errno));
- return false;
+ else if (wait_r == pid) {
+ pid = -1;
+ return pid_result_t::TERMINATED;
+ }
+ else if (wait_r == 0) {
+ // We can track the child
+ child_listener.add_reserved(eventLoop, pid, DEFAULT_PRIORITY - 10);
+ tracking_child = true;
+ reserved_child_watch = true;
+ return pid_result_t::OK;
+ }
}
+
+ log(loglevel_t::ERROR, service_name, ": pid read from pidfile (", pid, ") is not valid");
+ pid = -1;
+ return pid_result_t::FAILED;
}
-void ServiceRecord::started() noexcept
+void service_record::started() noexcept
{
- if (onstart_flags.runs_on_console && (service_type == ServiceType::SCRIPTED || service_type == ServiceType::BGPROCESS)) {
+ if (onstart_flags.starts_on_console && ! onstart_flags.runs_on_console) {
tcsetpgrp(0, getpgrp());
- releaseConsole();
- }
-
- if (service_type == ServiceType::BGPROCESS && pid_file.length() != 0) {
- if (! read_pid_file()) {
- failed_to_start();
- return;
- }
+ release_console();
}
-
- logServiceStarted(service_name);
- service_state = ServiceState::STARTED;
- notifyListeners(ServiceEvent::STARTED);
+
+ log_service_started(service_name);
+ service_state = service_state_t::STARTED;
+ notify_listeners(service_event::STARTED);
if (onstart_flags.rw_ready) {
open_control_socket();
setup_external_log();
}
- if (force_stop || desired_state == ServiceState::STOPPED) {
+ if (force_stop || desired_state == service_state_t::STOPPED) {
// We must now stop.
- service_set->addToStopQueue(this);
+ do_stop();
return;
}
// Notify any dependents whose desired state is STARTED:
- for (auto i = dependents.begin(); i != dependents.end(); i++) {
- (*i)->dependencyStarted();
- }
- for (auto i = soft_dpts.begin(); i != soft_dpts.end(); i++) {
- (*i)->getFrom()->dependencyStarted();
+ for (auto dept : dependents) {
+ dept->get_from()->dependencyStarted();
+ dept->waiting_on = false;
}
}
-void ServiceRecord::failed_to_start(bool depfailed) noexcept
+void service_record::failed_to_start(bool depfailed) noexcept
{
- if (!depfailed && onstart_flags.runs_on_console) {
+ if (!depfailed && onstart_flags.starts_on_console) {
tcsetpgrp(0, getpgrp());
- releaseConsole();
+ release_console();
}
- logServiceFailed(service_name);
- service_state = ServiceState::STOPPED;
+ log_service_failed(service_name);
+ service_state = service_state_t::STOPPED;
if (start_explicit) {
start_explicit = false;
release();
}
- notifyListeners(ServiceEvent::FAILEDSTART);
+ notify_listeners(service_event::FAILEDSTART);
// Cancel start of dependents:
- for (sr_iter i = dependents.begin(); i != dependents.end(); i++) {
- if ((*i)->service_state == ServiceState::STARTING) {
- (*i)->prop_failure = true;
- service_set->addToPropQueue(*i);
- }
- }
- for (auto i = soft_dpts.begin(); i != soft_dpts.end(); i++) {
- // We can send 'start', because this is only a soft dependency.
- // Our startup failure means that they don't have to wait for us.
- if ((*i)->waiting_on) {
- (*i)->holding_acq = false;
- (*i)->waiting_on = false;
- (*i)->getFrom()->dependencyStarted();
- release();
+ for (auto & dept : dependents) {
+ if (dept->dep_type == dependency_type::REGULAR) {
+ if (dept->get_from()->service_state == service_state_t::STARTING) {
+ dept->get_from()->prop_failure = true;
+ services->add_prop_queue(dept->get_from());
+ }
+ }
+ else if (dept->dep_type == dependency_type::SOFT) {
+ if (dept->waiting_on) {
+ dept->waiting_on = false;
+ dept->get_from()->dependencyStarted();
+ }
+ if (dept->holding_acq) {
+ dept->holding_acq = false;
+ release();
+ }
}
}
}
-bool ServiceRecord::start_ps_process() noexcept
+bool service_record::start_ps_process() noexcept
{
- return start_ps_process(exec_arg_parts, onstart_flags.runs_on_console);
+ // default implementation: there is no process, so we are started.
+ started();
+ return true;
}
-bool ServiceRecord::start_ps_process(const std::vector<const char *> &cmd, bool on_console) noexcept
+bool base_process_service::start_ps_process() noexcept
+{
+ if (restarting) {
+ if (pid == -1) {
+ return restart_ps_process();
+ }
+ return true;
+ }
+ else {
+ eventLoop.get_time(restart_interval_time, clock_type::MONOTONIC);
+ restart_interval_count = 0;
+ return start_ps_process(exec_arg_parts, onstart_flags.starts_on_console);
+ }
+}
+
+bool base_process_service::start_ps_process(const std::vector<const char *> &cmd, bool on_console) noexcept
{
// In general, you can't tell whether fork/exec is successful. We use a pipe to communicate
// success/failure from the child to the parent. The pipe is set CLOEXEC so a successful
// exec closes the pipe, and the parent sees EOF. If the exec is unsuccessful, the errno
// is written to the pipe, and the parent can read it.
+ eventLoop.get_time(last_start_time, clock_type::MONOTONIC);
+
int pipefd[2];
if (pipe2(pipefd, O_CLOEXEC)) {
- log(LogLevel::ERROR, service_name, ": can't create status check pipe: ", strerror(errno));
+ log(loglevel_t::ERROR, service_name, ": can't create status check pipe: ", strerror(errno));
return false;
}
}
bool child_status_registered = false;
- ControlConn *control_conn = nullptr;
+ control_conn_t *control_conn = nullptr;
int control_socket[2] = {-1, -1};
if (onstart_flags.pass_cs_fd) {
if (dinit_socketpair(AF_UNIX, SOCK_STREAM, /* protocol */ 0, control_socket, SOCK_NONBLOCK)) {
- log(LogLevel::ERROR, service_name, ": can't create control socket: ", strerror(errno));
+ log(loglevel_t::ERROR, service_name, ": can't create control socket: ", strerror(errno));
goto out_p;
}
fcntl(control_socket[0], F_SETFD, fdflags | FD_CLOEXEC);
try {
- control_conn = new ControlConn(&eventLoop, service_set, control_socket[0]);
+ control_conn = new control_conn_t(&eventLoop, services, control_socket[0]);
}
catch (std::exception &exc) {
- log(LogLevel::ERROR, service_name, ": can't launch process; out of memory");
+ log(loglevel_t::ERROR, service_name, ": can't launch process; out of memory");
goto out_cs;
}
}
child_status_listener.add_watch(eventLoop, pipefd[0], IN_EVENTS);
child_status_registered = true;
- forkpid = child_listener.fork(eventLoop);
+ // We specify a high priority (i.e. low priority value) so that process termination is
+ // handled early. This means we have always recorded that the process is terminated by the
+ // time that we handle events that might otherwise cause us to signal the process, so we
+ // avoid sending a signal to an invalid (and possibly recycled) process ID.
+ forkpid = child_listener.fork(eventLoop, reserved_child_watch, DEFAULT_PRIORITY - 10);
+ reserved_child_watch = true;
}
catch (std::exception &e) {
- log(LogLevel::ERROR, service_name, ": Could not fork: ", e.what());
+ log(loglevel_t::ERROR, service_name, ": Could not fork: ", e.what());
goto out_cs_h;
}
return false;
}
-void ServiceRecord::run_child_proc(const char * const *args, const char *logfile, bool on_console,
+void service_record::run_child_proc(const char * const *args, const char *logfile, bool on_console,
int wpipefd, int csfd) noexcept
{
// Child process. Must not allocate memory (or otherwise risk throwing any exception)
sigdelset(&sigwait_set, SIGCHLD);
sigdelset(&sigwait_set, SIGINT);
sigdelset(&sigwait_set, SIGTERM);
+ sigdelset(&sigwait_set, SIGQUIT);
constexpr int bufsz = ((CHAR_BIT * sizeof(pid_t)) / 3 + 2) + 11;
// "LISTEN_PID=" - 11 characters; the expression above gives a conservative estimate
- // on the maxiumum number of bytes required for LISTEN=xxx, including nul terminator,
- // where xxx is a pid_t in decimal (i.e. one decimal digit is worth just over 3 bits).
+ // on the maxiumum number of bytes required for LISTEN=nnn, including nul terminator,
+ // where nnn is a pid_t in decimal (i.e. one decimal digit is worth just over 3 bits).
char nbuf[bufsz];
// "DINIT_CS_FD=" - 12 bytes. (we -1 from sizeof(int) in account of sign bit).
}
else goto failure_out;
- // We have the option of creating a new process group and/or session. If
- // we just create a new process group, the child process cannot make itself
- // a session leader if it wants to do that (eg getty/login will generally
- // want this). If we do neither, and we are running with a controlling
- // terminal, a ^C or similar will also affect the child process.
+ // We have the option of creating a session and process group, or just a new process
+ // group. If we just create a new process group, the child process cannot make itself
+ // a session leader if it wants to do that (eg getty/login will generally want this).
+ // If we do neither, and we are running with a controlling terminal, a ^C or similar
+ // will also affect the child process (which probably isn't so bad, though since we
+ // will handle the shutdown ourselves it's not necessary). Creating a new session
+ // (and a new process group as part of that) seems like a safe bet, and has the
+ // advantage of letting us signal the process as part of a process group.
setsid();
}
else {
}
// Mark this and all dependent services as force-stopped.
-void ServiceRecord::forceStop() noexcept
+void service_record::forced_stop() noexcept
{
- if (service_state != ServiceState::STOPPED) {
+ if (service_state != service_state_t::STOPPED) {
force_stop = true;
- service_set->addToStopQueue(this);
+ services->add_transition_queue(this);
}
}
-void ServiceRecord::dependentStopped() noexcept
+void service_record::dependent_stopped() noexcept
{
- if (service_state == ServiceState::STOPPING && waiting_for_deps) {
- service_set->addToPropQueue(this);
+ if (service_state == service_state_t::STOPPING && waiting_for_deps) {
+ services->add_transition_queue(this);
}
}
-void ServiceRecord::stop(bool bring_down) noexcept
+void service_record::stop(bool bring_down) noexcept
{
if (start_explicit) {
start_explicit = false;
release();
}
-
- if (bring_down && desired_state != ServiceState::STOPPED) {
- desired_state = ServiceState::STOPPED;
- service_set->addToStopQueue(this);
+
+ if (bring_down) {
+ do_stop();
}
}
-void ServiceRecord::do_stop() noexcept
+void service_record::do_stop() noexcept
{
if (pinned_started) return;
- if (service_state != ServiceState::STARTED) {
- if (service_state == ServiceState::STARTING) {
+ if (start_explicit && ! do_auto_restart()) {
+ start_explicit = false;
+ release();
+ if (required_by == 0) return; // release will re-call us anyway
+ }
+
+ if (service_state != service_state_t::STARTED) {
+ if (service_state == service_state_t::STARTING) {
if (! can_interrupt_start()) {
// Well this is awkward: we're going to have to continue
// starting, but we don't want any dependents to think that
// they are still waiting to start.
// Make sure they remain stopped:
- stopDependents();
+ stop_dependents();
return;
}
// We must have had desired_state == STARTED.
- notifyListeners(ServiceEvent::STARTCANCELLED);
+ notify_listeners(service_event::STARTCANCELLED);
- // Reaching this point, we have can_interrupt_start() == true. So,
- // we can stop. Dependents might be starting, but they must be
- // waiting on us, so they should also be immediately stoppable.
- // Fall through to below,.
+ interrupt_start();
+
+ // Reaching this point, we are starting interruptibly - so we
+ // stop now (by falling through to below).
}
else {
// If we're starting we need to wait for that to complete.
}
}
- service_state = ServiceState::STOPPING;
+ service_state = service_state_t::STOPPING;
waiting_for_deps = true;
-
- // If we get here, we are in STARTED state; stop all dependents.
- if (stopDependents()) {
- allDepsStopped();
+ if (stop_dependents()) {
+ services->add_transition_queue(this);
}
}
-bool ServiceRecord::stopCheckDependents() noexcept
+bool service_record::stop_check_dependents() noexcept
{
bool all_deps_stopped = true;
- for (sr_iter i = dependents.begin(); i != dependents.end(); ++i) {
- if (! (*i)->is_stopped()) {
+ for (auto dept : dependents) {
+ if (dept->dep_type == dependency_type::REGULAR && ! dept->get_from()->is_stopped()) {
all_deps_stopped = false;
break;
}
return all_deps_stopped;
}
-bool ServiceRecord::stopDependents() noexcept
+bool service_record::stop_dependents() noexcept
{
bool all_deps_stopped = true;
- for (sr_iter i = dependents.begin(); i != dependents.end(); ++i) {
- if (! (*i)->is_stopped()) {
- // Note we check *first* since if the dependent service is not stopped,
- // 1. We will issue a stop to it shortly and
- // 2. It will notify us when stopped, at which point the stopCheckDependents()
- // check is run anyway.
- all_deps_stopped = false;
- }
+ for (auto dept : dependents) {
+ if (dept->dep_type == dependency_type::REGULAR) {
+ if (! dept->get_from()->is_stopped()) {
+ // Note we check *first* since if the dependent service is not stopped,
+ // 1. We will issue a stop to it shortly and
+ // 2. It will notify us when stopped, at which point the stop_check_dependents()
+ // check is run anyway.
+ all_deps_stopped = false;
+ }
+
+ if (force_stop) {
+ // If this service is to be forcefully stopped, dependents must also be.
+ dept->get_from()->forced_stop();
+ }
- (*i)->forceStop();
+ dept->get_from()->prop_stop = true;
+ services->add_prop_queue(dept->get_from());
+ }
}
return all_deps_stopped;
}
// All dependents have stopped; we can stop now, too. Only called when STOPPING.
-void ServiceRecord::allDepsStopped()
+void service_record::all_deps_stopped() noexcept
{
waiting_for_deps = false;
- if (service_type == ServiceType::PROCESS || service_type == ServiceType::BGPROCESS) {
- if (pid != -1) {
- // The process is still kicking on - must actually kill it.
- if (! onstart_flags.no_sigterm) {
- kill(pid, SIGTERM);
- }
- if (term_signal != -1) {
- kill(pid, term_signal);
- }
-
- // In most cases, the rest is done in process_child_callback.
- // If we are a BGPROCESS and the process is not our immediate child, however, that
- // won't work - check for this now:
- if (service_type == ServiceType::BGPROCESS) {
- int status;
- pid_t r = waitpid(pid, &status, WNOHANG);
- if (r == -1 && errno == ECHILD) {
- // We can't track this child (or it's terminated already)
- stopped();
- }
- else if (r == pid) {
- // TODO, examine status and log anything unusual.
- stopped();
- }
- }
+ stopped();
+}
+
+void base_process_service::kill_pg(int signo) noexcept
+{
+ pid_t pgid = getpgid(pid);
+ if (pgid == -1) {
+ // only should happen if pid is invalid, which should never happen...
+ log(loglevel_t::ERROR, service_name, ": can't signal process: ", strerror(errno));
+ return;
+ }
+ kill(-pgid, signo);
+}
+
+void base_process_service::all_deps_stopped() noexcept
+{
+ waiting_for_deps = false;
+ if (pid != -1) {
+ // The process is still kicking on - must actually kill it. We signal the process
+ // group (-pid) rather than just the process as there's less risk then of creating
+ // an orphaned process group:
+ if (! onstart_flags.no_sigterm) {
+ kill_pg(SIGTERM);
}
- else {
- // The process is already dead.
- stopped();
+ if (term_signal != -1) {
+ kill_pg(term_signal);
}
- }
- else if (service_type == ServiceType::SCRIPTED) {
- // Scripted service.
- if (stop_command.length() == 0) {
+
+ // In most cases, the rest is done in handle_exit_status.
+ // If we are a BGPROCESS and the process is not our immediate child, however, that
+ // won't work - check for this now:
+ if (record_type == service_type::BGPROCESS && ! tracking_child) {
stopped();
}
- else if (! start_ps_process(stop_arg_parts, false)) {
- // Couldn't execute stop script, but there's not much we can do:
- stopped();
+ else if (stop_timeout != time_val(0,0)) {
+ restart_timer.arm_timer_rel(eventLoop, stop_timeout);
+ stop_timer_armed = true;
}
}
else {
+ // The process is already dead.
+ stopped();
+ }
+}
+
+void scripted_service::all_deps_stopped() noexcept
+{
+ waiting_for_deps = false;
+ if (stop_command.length() == 0) {
+ stopped();
+ }
+ else if (! start_ps_process(stop_arg_parts, false)) {
+ // Couldn't execute stop script, but there's not much we can do:
stopped();
}
+ else {
+ // successfully started stop script: start kill timer:
+ if (stop_timeout != time_val(0,0)) {
+ restart_timer.arm_timer_rel(eventLoop, stop_timeout);
+ stop_timer_armed = true;
+ }
+ }
}
-void ServiceRecord::unpin() noexcept
+void service_record::unpin() noexcept
{
if (pinned_started) {
pinned_started = false;
- if (desired_state == ServiceState::STOPPED) {
+ if (desired_state == service_state_t::STOPPED || force_stop) {
do_stop();
- service_set->processQueues(false);
+ services->process_queues();
}
}
if (pinned_stopped) {
pinned_stopped = false;
- if (desired_state == ServiceState::STARTED) {
+ if (desired_state == service_state_t::STARTED) {
do_start();
- service_set->processQueues(true);
+ services->process_queues();
}
}
}
-void ServiceRecord::queueForConsole() noexcept
+void service_record::queue_for_console() noexcept
{
- service_set->append_console_queue(this);
+ services->append_console_queue(this);
}
-void ServiceRecord::releaseConsole() noexcept
+void service_record::release_console() noexcept
{
- service_set->pullConsoleQueue();
+ services->pull_console_queue();
}
-void ServiceSet::service_active(ServiceRecord *sr) noexcept
+void service_record::interrupt_start() noexcept
+{
+ services->unqueue_console(this);
+}
+
+void service_set::service_active(service_record *sr) noexcept
{
active_services++;
}
-void ServiceSet::service_inactive(ServiceRecord *sr) noexcept
+void service_set::service_inactive(service_record *sr) noexcept
{
active_services--;
}
+
+base_process_service::base_process_service(service_set *sset, string name,
+ service_type service_type_p, string &&command,
+ std::list<std::pair<unsigned,unsigned>> &command_offsets,
+ const std::list<prelim_dep> &deplist_p)
+ : service_record(sset, name, service_type_p, std::move(command), command_offsets,
+ deplist_p), child_listener(this), child_status_listener(this)
+{
+ restart_interval_count = 0;
+ restart_interval_time = {0, 0};
+ restart_timer.service = this;
+ restart_timer.add_timer(eventLoop);
+
+ // By default, allow a maximum of 3 restarts within 10.0 seconds:
+ restart_interval.seconds() = 10;
+ restart_interval.nseconds() = 0;
+ max_restart_interval_count = 3;
+
+ waiting_restart_timer = false;
+ reserved_child_watch = false;
+ tracking_child = false;
+ stop_timer_armed = false;
+}
+
+void base_process_service::do_restart() noexcept
+{
+ waiting_restart_timer = false;
+ restart_interval_count++;
+
+ // We may be STARTING (regular restart) or STARTED ("smooth recovery"). This affects whether
+ // the process should be granted access to the console:
+ bool on_console = service_state == service_state_t::STARTING
+ ? onstart_flags.starts_on_console : onstart_flags.runs_on_console;
+
+ if (service_state == service_state_t::STARTING) {
+ // for a smooth recovery, we want to check dependencies are available before actually
+ // starting:
+ if (! check_deps_started()) {
+ waiting_for_deps = true;
+ return;
+ }
+ }
+
+ if (! start_ps_process(exec_arg_parts, on_console)) {
+ restarting = false;
+ if (service_state == service_state_t::STARTING) {
+ failed_to_start();
+ }
+ else {
+ desired_state = service_state_t::STOPPED;
+ forced_stop();
+ }
+ services->process_queues();
+ }
+}
+
+bool base_process_service::restart_ps_process() noexcept
+{
+ using time_val = eventloop_t::time_val;
+
+ time_val current_time;
+ eventLoop.get_time(current_time, clock_type::MONOTONIC);
+
+ if (max_restart_interval_count != 0) {
+ // Check whether we're still in the most recent restart check interval:
+ time_val int_diff = current_time - restart_interval_time;
+ if (int_diff < restart_interval) {
+ if (restart_interval_count >= max_restart_interval_count) {
+ log(loglevel_t::ERROR, "Service ", service_name, " restarting too quickly; stopping.");
+ return false;
+ }
+ }
+ else {
+ restart_interval_time = current_time;
+ restart_interval_count = 0;
+ }
+ }
+
+ // Check if enough time has lapsed since the prevous restart. If not, start a timer:
+ time_val tdiff = current_time - last_start_time;
+ if (restart_delay <= tdiff) {
+ // > restart delay (normally 200ms)
+ do_restart();
+ }
+ else {
+ time_val timeout = restart_delay - tdiff;
+ restart_timer.arm_timer_rel(eventLoop, timeout);
+ waiting_restart_timer = true;
+ }
+ return true;
+}
+
+void base_process_service::interrupt_start() noexcept
+{
+ // overridden in subclasses
+ if (waiting_restart_timer) {
+ restart_timer.stop_timer(eventLoop);
+ waiting_restart_timer = false;
+ }
+ service_record::interrupt_start();
+}
+
+void base_process_service::kill_with_fire() noexcept
+{
+ if (pid != -1) {
+ log(loglevel_t::WARN, "Service ", service_name, "with pid ", pid, " exceeded allowed stop time; killing.");
+ kill_pg(SIGKILL);
+ }
+}
+
+dasynq::rearm process_restart_timer::timer_expiry(eventloop_t &, int expiry_count)
+{
+ if (service->service_state == service_state_t::STOPPING) {
+ service->kill_with_fire();
+ service->stop_timer_armed = false;
+ }
+ else {
+ // STARTING / STARTED:
+ service->do_restart();
+ }
+ return dasynq::rearm::DISARM;
+}