4 #include <sys/socket.h>
7 #include "dinit-socket.h"
8 #include "dinit-util.h"
10 #include "proc-service.h"
13 * Most of the implementation for process-based services (process, scripted, bgprocess) is here.
15 * See proc-service.h header for interface details.
18 // Given a string and a list of pairs of (start,end) indices for each argument in that string,
19 // store a null terminator for the argument. Return a `char *` vector containing the beginning
20 // of each argument and a trailing nullptr. (The returned array is invalidated if the string is
22 std::vector<const char *> separate_args(std::string &s,
23 const std::list<std::pair<unsigned,unsigned>> &arg_indices)
25 std::vector<const char *> r;
26 r.reserve(arg_indices.size() + 1);
28 // First store nul terminator for each part:
29 for (auto index_pair : arg_indices) {
30 if (index_pair.second < s.length()) {
31 s[index_pair.second] = 0;
35 // Now we can get the C string (c_str) and store offsets into it:
36 const char * cstr = s.c_str();
37 for (auto index_pair : arg_indices) {
38 r.push_back(cstr + index_pair.first);
44 void process_service::exec_succeeded() noexcept
46 // This could be a smooth recovery (state already STARTED). Even more, the process
47 // might be stopped (and killed via a signal) during smooth recovery. We don't to
48 // process startup again in either case, so we check for state STARTING:
49 if (get_state() == service_state_t::STARTING) {
52 else if (get_state() == service_state_t::STOPPING) {
53 // stopping, but smooth recovery was in process. That's now over so we can
54 // commence normal stop. Note that if pid == -1 the process already stopped(!),
55 // that's handled below.
56 if (pid != -1 && stop_check_dependents()) {
62 void scripted_service::exec_succeeded() noexcept
64 // For a scripted service, this means nothing other than that the start/stop
65 // script will now begin.
68 rearm exec_status_pipe_watcher::fd_event(eventloop_t &loop, int fd, int flags) noexcept
70 base_process_service *sr = service;
71 sr->waiting_for_execstat = false;
74 int r = read(get_watched_fd(), &exec_status, sizeof(int));
76 close(get_watched_fd());
79 // We read an errno code; exec() failed, and the service startup failed.
81 sr->child_listener.deregister(event_loop, sr->pid);
82 sr->reserved_child_watch = false;
83 if (sr->stop_timer_armed) {
84 sr->restart_timer.stop_timer(loop);
85 sr->stop_timer_armed = false;
89 sr->exec_failed(exec_status);
95 // Somehow the process managed to complete before we even saw the exec() status.
96 sr->handle_exit_status(sr->exit_status);
100 sr->services->process_queues();
102 return rearm::REMOVED;
105 dasynq::rearm service_child_watcher::status_change(eventloop_t &loop, pid_t child, int status) noexcept
107 base_process_service *sr = service;
110 sr->exit_status = bp_sys::exit_status(status);
112 // Ok, for a process service, any process death which we didn't rig
113 // ourselves is a bit... unexpected. Probably, the child died because
114 // we asked it to (sr->service_state == STOPPING). But even if
115 // we didn't, there's not much we can do.
117 if (sr->waiting_for_execstat) {
118 // We still don't have an exec() status from the forked child, wait for that
119 // before doing any further processing.
120 return dasynq::rearm::NOOP; // hold watch reservation
123 // Must stop watch now since handle_exit_status might result in re-launch:
124 // (stop_watch instead of deregister, so that we hold watch reservation).
127 if (sr->stop_timer_armed) {
128 sr->restart_timer.stop_timer(loop);
129 sr->stop_timer_armed = false;
132 sr->handle_exit_status(bp_sys::exit_status(status));
133 return dasynq::rearm::NOOP;
136 void process_service::handle_exit_status(bp_sys::exit_status exit_status) noexcept
138 bool did_exit = exit_status.did_exit();
139 bool was_signalled = exit_status.was_signalled();
141 auto service_state = get_state();
143 if (exit_status.did_exit_clean() && service_state != service_state_t::STOPPING) {
145 log(loglevel_t::ERROR, "Service ", get_name(), " process terminated with exit code ",
146 exit_status.get_exit_status());
148 else if (was_signalled) {
149 log(loglevel_t::ERROR, "Service ", get_name(), " terminated due to signal ",
150 exit_status.get_term_sig());
154 if (service_state == service_state_t::STARTING) {
155 if (exit_status.did_exit_clean()) {
159 stop_reason = stopped_reason_t::FAILED;
163 else if (service_state == service_state_t::STOPPING) {
164 // We won't log a non-zero exit status or termination due to signal here -
165 // we assume that the process died because we signalled it.
166 if (stop_timer_armed) {
167 restart_timer.stop_timer(event_loop);
171 else if (smooth_recovery && service_state == service_state_t::STARTED
172 && get_target_state() == service_state_t::STARTED) {
173 do_smooth_recovery();
177 stop_reason = stopped_reason_t::TERMINATED;
180 services->process_queues();
183 void process_service::exec_failed(int errcode) noexcept
185 log(loglevel_t::ERROR, get_name(), ": execution failed: ", strerror(errcode));
186 if (get_state() == service_state_t::STARTING) {
187 stop_reason = stopped_reason_t::EXECFAILED;
191 // Process service in smooth recovery:
192 stop_reason = stopped_reason_t::TERMINATED;
197 void bgproc_service::handle_exit_status(bp_sys::exit_status exit_status) noexcept
200 bool did_exit = exit_status.did_exit();
201 bool was_signalled = exit_status.was_signalled();
202 auto service_state = get_state();
204 if (!exit_status.did_exit_clean() && service_state != service_state_t::STOPPING) {
206 log(loglevel_t::ERROR, "Service ", get_name(), " process terminated with exit code ",
207 exit_status.get_exit_status());
209 else if (was_signalled) {
210 log(loglevel_t::ERROR, "Service ", get_name(), " terminated due to signal ",
211 exit_status.get_term_sig());
215 // This may be a "smooth recovery" where we are restarting the process while leaving the
216 // service in the STARTED state.
217 if (restarting && service_state == service_state_t::STARTED) {
219 bool need_stop = false;
220 if ((did_exit && exit_status.get_exit_status() != 0) || was_signalled) {
224 // We need to re-read the PID, since it has now changed.
225 if (pid_file.length() != 0) {
226 auto pid_result = read_pid_file(&exit_status);
227 switch (pid_result) {
228 case pid_result_t::FAILED:
229 // Failed startup: no auto-restart.
232 case pid_result_t::TERMINATED:
234 case pid_result_t::OK:
241 // Failed startup: no auto-restart.
242 stop_reason = stopped_reason_t::TERMINATED;
244 services->process_queues();
251 if (service_state == service_state_t::STARTING) {
252 // POSIX requires that if the process exited clearly with a status code of 0,
253 // the exit status value will be 0:
254 if (exit_status.did_exit_clean()) {
255 auto pid_result = read_pid_file(&exit_status);
256 switch (pid_result) {
257 case pid_result_t::FAILED:
258 // Failed startup: no auto-restart.
259 stop_reason = stopped_reason_t::FAILED;
262 case pid_result_t::TERMINATED:
263 // started, but immediately terminated
266 case pid_result_t::OK:
272 stop_reason = stopped_reason_t::FAILED;
276 else if (service_state == service_state_t::STOPPING) {
277 // We won't log a non-zero exit status or termination due to signal here -
278 // we assume that the process died because we signalled it.
282 // we must be STARTED
283 if (smooth_recovery && get_target_state() == service_state_t::STARTED) {
284 do_smooth_recovery();
287 if (! do_auto_restart() && start_explicit) {
288 start_explicit = false;
291 stop_reason = stopped_reason_t::TERMINATED;
296 services->process_queues();
299 void bgproc_service::exec_failed(int errcode) noexcept
301 log(loglevel_t::ERROR, get_name(), ": execution failed: ", strerror(errcode));
302 // Only time we execute is for startup:
303 stop_reason = stopped_reason_t::EXECFAILED;
307 void scripted_service::handle_exit_status(bp_sys::exit_status exit_status) noexcept
309 bool did_exit = exit_status.did_exit();
310 bool was_signalled = exit_status.was_signalled();
311 auto service_state = get_state();
313 // For a scripted service, a termination occurs in one of three main cases:
314 // - the start script completed (or failed), when service was STARTING
315 // - the start script was interrupted to cancel startup; state is STOPPING
316 // - the stop script complete (or failed), state is STOPPING
318 if (service_state == service_state_t::STOPPING) {
319 // We might be running the stop script, or we might be running the start script and have issued
320 // a cancel order via SIGINT:
321 if (interrupting_start) {
322 // We issued a start interrupt, so we expected this failure:
323 if (did_exit && exit_status.get_exit_status() != 0) {
324 log(loglevel_t::INFO, "Service ", get_name(), " start cancelled; exit code ",
325 exit_status.get_exit_status());
326 // Assume that a command terminating normally requires no cleanup:
331 log(loglevel_t::INFO, "Service ", get_name(), " start cancelled from signal ",
332 exit_status.get_term_sig());
334 // If the start script completed successfully, or was interrupted via our signal,
335 // we want to run the stop script to clean up:
338 interrupting_start = false;
340 else if (exit_status.did_exit_clean()) {
341 // We were running the stop script and finished successfully
345 // ??? failed to stop! Let's log it as warning:
347 log(loglevel_t::WARN, "Service ", get_name(), " stop command failed with exit code ",
348 exit_status.get_exit_status());
350 else if (was_signalled) {
351 log(loglevel_t::WARN, "Service ", get_name(), " stop command terminated due to signal ",
352 exit_status.get_term_sig());
354 // Even if the stop script failed, assume that service is now stopped, so that any dependencies
355 // can be stopped. There's not really any other useful course of action here.
358 services->process_queues();
361 if (exit_status.did_exit_clean()) {
364 else if (was_signalled && exit_status.get_term_sig() == SIGINT && onstart_flags.skippable) {
365 // A skippable service can be skipped by interrupting (eg by ^C if the service
366 // starts on the console).
367 start_skipped = true;
373 log(loglevel_t::ERROR, "Service ", get_name(), " command failed with exit code ",
374 exit_status.get_exit_status());
376 else if (was_signalled) {
377 log(loglevel_t::ERROR, "Service ", get_name(), " command terminated due to signal ",
378 exit_status.get_term_sig());
380 stop_reason = stopped_reason_t::FAILED;
383 services->process_queues();
387 void scripted_service::exec_failed(int errcode) noexcept
389 log(loglevel_t::ERROR, get_name(), ": execution failed: ", strerror(errcode));
390 auto service_state = get_state();
391 if (service_state == service_state_t::STARTING) {
392 stop_reason = stopped_reason_t::EXECFAILED;
395 else if (service_state == service_state_t::STOPPING) {
396 // We've logged the failure, but it's probably better not to leave the service in
402 bgproc_service::pid_result_t
403 bgproc_service::read_pid_file(bp_sys::exit_status *exit_status) noexcept
405 const char *pid_file_c = pid_file.c_str();
406 int fd = open(pid_file_c, O_CLOEXEC);
408 log(loglevel_t::ERROR, get_name(), ": read pid file: ", strerror(errno));
409 return pid_result_t::FAILED;
412 char pidbuf[21]; // just enough to hold any 64-bit integer
413 int r = complete_read(fd, pidbuf, 20);
415 // Could not read from PID file
416 log(loglevel_t::ERROR, get_name(), ": could not read from pidfile; ", strerror(errno));
418 return pid_result_t::FAILED;
422 pidbuf[r] = 0; // store nul terminator
424 bool valid_pid = false;
426 unsigned long long v = std::stoull(pidbuf, nullptr, 0);
427 if (v <= std::numeric_limits<pid_t>::max()) {
432 catch (std::out_of_range &exc) {
435 catch (std::invalid_argument &exc) {
436 // Ok, so it doesn't look like a number: proceed...
440 pid_t wait_r = waitpid(pid, exit_status, WNOHANG);
441 if (wait_r == -1 && errno == ECHILD) {
442 // We can't track this child - check process exists:
443 if (kill(pid, 0) == 0 || errno != ESRCH) {
444 tracking_child = false;
445 return pid_result_t::OK;
448 log(loglevel_t::ERROR, get_name(), ": pid read from pidfile (", pid, ") is not valid");
450 return pid_result_t::FAILED;
453 else if (wait_r == pid) {
455 return pid_result_t::TERMINATED;
457 else if (wait_r == 0) {
458 // We can track the child
459 child_listener.add_reserved(event_loop, pid, dasynq::DEFAULT_PRIORITY - 10);
460 tracking_child = true;
461 reserved_child_watch = true;
462 return pid_result_t::OK;
466 log(loglevel_t::ERROR, get_name(), ": pid read from pidfile (", pid, ") is not valid");
468 return pid_result_t::FAILED;
471 void process_service::bring_down() noexcept
473 if (waiting_for_execstat) {
474 // The process is still starting. This should be uncommon, but can occur during
475 // smooth recovery. We can't do much now; we have to wait until we get the
476 // status, and then act appropriately.
479 else if (pid != -1) {
480 // The process is still kicking on - must actually kill it. We signal the process
481 // group (-pid) rather than just the process as there's less risk then of creating
482 // an orphaned process group:
483 if (! onstart_flags.no_sigterm) {
486 if (term_signal != -1) {
487 kill_pg(term_signal);
490 // If there's a stop timeout, arm the timer now:
491 if (stop_timeout != time_val(0,0)) {
492 restart_timer.arm_timer_rel(event_loop, stop_timeout);
493 stop_timer_armed = true;
496 // The rest is done in handle_exit_status.
499 // The process is already dead.
504 void bgproc_service::bring_down() noexcept
507 // The process is still kicking on - must actually kill it. We signal the process
508 // group (-pid) rather than just the process as there's less risk then of creating
509 // an orphaned process group:
510 if (! onstart_flags.no_sigterm) {
513 if (term_signal != -1) {
514 kill_pg(term_signal);
517 // In most cases, the rest is done in handle_exit_status.
518 // If we are a BGPROCESS and the process is not our immediate child, however, that
519 // won't work - check for this now:
520 if (! tracking_child) {
523 else if (stop_timeout != time_val(0,0)) {
524 restart_timer.arm_timer_rel(event_loop, stop_timeout);
525 stop_timer_armed = true;
529 // The process is already dead.
534 void scripted_service::bring_down() noexcept
537 // We're already running the stop script; nothing to do.
541 if (stop_command.length() == 0) {
544 else if (! start_ps_process(stop_arg_parts, false)) {
545 // Couldn't execute stop script, but there's not much we can do:
549 // successfully started stop script: start kill timer:
550 if (stop_timeout != time_val(0,0)) {
551 restart_timer.arm_timer_rel(event_loop, stop_timeout);
552 stop_timer_armed = true;
557 dasynq::rearm process_restart_timer::timer_expiry(eventloop_t &, int expiry_count)
559 service->timer_expired();
561 // Leave the timer disabled, or, if it has been reset by any processing above, leave it armed:
562 return dasynq::rearm::NOOP;