1 /*-*- Mode: C; c-basic-offset: 8; indent-tabs-mode: nil -*-*/
4 This file is part of systemd.
6 Copyright 2010 Lennart Poettering
8 systemd is free software; you can redistribute it and/or modify it
9 under the terms of the GNU Lesser General Public License as published by
10 the Free Software Foundation; either version 2.1 of the License, or
11 (at your option) any later version.
13 systemd is distributed in the hope that it will be useful, but
14 WITHOUT ANY WARRANTY; without even the implied warranty of
15 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
16 Lesser General Public License for more details.
18 You should have received a copy of the GNU Lesser General Public License
19 along with systemd; If not, see <http://www.gnu.org/licenses/>.
24 #include <sys/timerfd.h>
25 #include <sys/epoll.h>
31 #include "load-fragment.h"
32 #include "load-dropin.h"
36 JobBusClient* job_bus_client_new(DBusConnection *connection, const char *name) {
40 name_len = strlen(name);
41 cl = malloc0(sizeof(JobBusClient) + name_len + 1);
46 memcpy(cl->name, name, name_len + 1);
50 Job* job_new_raw(Unit *unit) {
53 /* used for deserialization */
61 j->manager = unit->manager;
63 j->timer_watch.type = WATCH_INVALID;
68 Job* job_new(Unit *unit, JobType type) {
71 assert(type < _JOB_TYPE_MAX);
73 j = job_new_raw(unit);
77 j->id = j->manager->current_job_id++;
80 /* We don't link it here, that's what job_dependency() is for */
85 void job_free(Job *j) {
89 assert(!j->installed);
90 assert(!j->transaction_prev);
91 assert(!j->transaction_next);
92 assert(!j->subject_list);
93 assert(!j->object_list);
96 LIST_REMOVE(Job, run_queue, j->manager->run_queue, j);
99 LIST_REMOVE(Job, dbus_queue, j->manager->dbus_job_queue, j);
101 if (j->timer_watch.type != WATCH_INVALID) {
102 assert(j->timer_watch.type == WATCH_JOB_TIMER);
103 assert(j->timer_watch.data.job == j);
104 assert(j->timer_watch.fd >= 0);
106 assert_se(epoll_ctl(j->manager->epoll_fd, EPOLL_CTL_DEL, j->timer_watch.fd, NULL) >= 0);
107 close_nointr_nofail(j->timer_watch.fd);
110 while ((cl = j->bus_client_list)) {
111 LIST_REMOVE(JobBusClient, client, j->bus_client_list, cl);
117 void job_uninstall(Job *j) {
118 assert(j->installed);
119 assert(j->unit->job == j);
120 /* Detach from next 'bigger' objects */
122 /* daemon-reload should be transparent to job observers */
123 if (j->manager->n_reloading <= 0)
124 bus_job_send_removed_signal(j);
127 unit_add_to_gc_queue(j->unit);
129 hashmap_remove(j->manager->jobs, UINT32_TO_PTR(j->id));
130 j->installed = false;
133 static bool job_type_allows_late_merge(JobType t) {
134 /* Tells whether it is OK to merge a job of type 't' with an already
136 * Reloads cannot be merged this way. Think of the sequence:
137 * 1. Reload of a daemon is in progress; the daemon has already loaded
138 * its config file, but hasn't completed the reload operation yet.
139 * 2. Edit foo's config file.
140 * 3. Trigger another reload to have the daemon use the new config.
141 * Should the second reload job be merged into the first one, the daemon
142 * would not know about the new config.
143 * JOB_RESTART jobs on the other hand can be merged, because they get
144 * patched into JOB_START after stopping the unit. So if we see a
145 * JOB_RESTART running, it means the unit hasn't stopped yet and at
146 * this time the merge is still allowed. */
147 return !(t == JOB_RELOAD || t == JOB_RELOAD_OR_START);
150 static void job_merge_into_installed(Job *j, Job *other) {
151 assert(j->installed);
152 assert(j->unit == other->unit);
154 j->type = job_type_lookup_merge(j->type, other->type);
155 assert(j->type >= 0);
157 j->override = j->override || other->override;
160 Job* job_install(Job *j) {
161 Job *uj = j->unit->job;
163 assert(!j->installed);
166 if (job_type_is_conflicting(uj->type, j->type))
167 job_finish_and_invalidate(uj, JOB_CANCELED);
169 /* not conflicting, i.e. mergeable */
171 if (uj->state == JOB_WAITING ||
172 (job_type_allows_late_merge(j->type) && job_type_is_superset(uj->type, j->type))) {
173 job_merge_into_installed(uj, j);
174 log_debug("Merged into installed job %s/%s as %u",
175 uj->unit->id, job_type_to_string(uj->type), (unsigned) uj->id);
178 /* already running and not safe to merge into */
179 /* Patch uj to become a merged job and re-run it. */
180 /* XXX It should be safer to queue j to run after uj finishes, but it is
181 * not currently possible to have more than one installed job per unit. */
182 job_merge_into_installed(uj, j);
183 log_debug("Merged into running job, re-running: %s/%s as %u",
184 uj->unit->id, job_type_to_string(uj->type), (unsigned) uj->id);
185 uj->state = JOB_WAITING;
191 /* Install the job */
194 j->manager->n_installed_jobs ++;
195 log_debug("Installed new job %s/%s as %u", j->unit->id, job_type_to_string(j->type), (unsigned) j->id);
199 void job_install_deserialized(Job *j) {
200 assert(!j->installed);
203 log_debug("Unit %s already has a job installed. Not installing deserialized job.", j->unit->id);
208 log_debug("Reinstalled deserialized job %s/%s as %u", j->unit->id, job_type_to_string(j->type), (unsigned) j->id);
211 JobDependency* job_dependency_new(Job *subject, Job *object, bool matters, bool conflicts) {
216 /* Adds a new job link, which encodes that the 'subject' job
217 * needs the 'object' job in some way. If 'subject' is NULL
218 * this means the 'anchor' job (i.e. the one the user
219 * explicitly asked for) is the requester. */
221 if (!(l = new0(JobDependency, 1)))
224 l->subject = subject;
226 l->matters = matters;
227 l->conflicts = conflicts;
230 LIST_PREPEND(JobDependency, subject, subject->subject_list, l);
232 LIST_PREPEND(JobDependency, object, object->object_list, l);
237 void job_dependency_free(JobDependency *l) {
241 LIST_REMOVE(JobDependency, subject, l->subject->subject_list, l);
243 LIST_REMOVE(JobDependency, object, l->object->object_list, l);
248 void job_dump(Job *j, FILE*f, const char *prefix) {
257 "%s\tAction: %s -> %s\n"
261 prefix, j->unit->id, job_type_to_string(j->type),
262 prefix, job_state_to_string(j->state),
263 prefix, yes_no(j->override));
267 * Merging is commutative, so imagine the matrix as symmetric. We store only
268 * its lower triangle to avoid duplication. We don't store the main diagonal,
269 * because A merged with A is simply A.
271 * Merging is associative! A merged with B merged with C is the same as
272 * A merged with C merged with B.
274 * Mergeability is transitive! If A can be merged with B and B with C then
277 * Also, if A merged with B cannot be merged with C, then either A or B cannot
278 * be merged with C either.
280 static const JobType job_merging_table[] = {
281 /* What \ With * JOB_START JOB_VERIFY_ACTIVE JOB_STOP JOB_RELOAD JOB_RELOAD_OR_START JOB_RESTART JOB_TRY_RESTART */
282 /************************************************************************************************************************************/
284 /*JOB_VERIFY_ACTIVE */ JOB_START,
285 /*JOB_STOP */ -1, -1,
286 /*JOB_RELOAD */ JOB_RELOAD_OR_START, JOB_RELOAD, -1,
287 /*JOB_RELOAD_OR_START*/ JOB_RELOAD_OR_START, JOB_RELOAD_OR_START, -1, JOB_RELOAD_OR_START,
288 /*JOB_RESTART */ JOB_RESTART, JOB_RESTART, -1, JOB_RESTART, JOB_RESTART,
289 /*JOB_TRY_RESTART */ JOB_RESTART, JOB_TRY_RESTART, -1, JOB_TRY_RESTART, JOB_RESTART, JOB_RESTART,
292 JobType job_type_lookup_merge(JobType a, JobType b) {
293 assert_cc(ELEMENTSOF(job_merging_table) == _JOB_TYPE_MAX * (_JOB_TYPE_MAX - 1) / 2);
294 assert(a >= 0 && a < _JOB_TYPE_MAX);
295 assert(b >= 0 && b < _JOB_TYPE_MAX);
306 return job_merging_table[(a - 1) * a / 2 + b];
309 bool job_type_is_redundant(JobType a, UnitActiveState b) {
319 b == UNIT_INACTIVE ||
322 case JOB_VERIFY_ACTIVE:
331 case JOB_RELOAD_OR_START:
333 b == UNIT_ACTIVATING ||
338 b == UNIT_ACTIVATING;
340 case JOB_TRY_RESTART:
342 b == UNIT_ACTIVATING;
345 assert_not_reached("Invalid job type");
349 bool job_is_runnable(Job *j) {
354 assert(j->installed);
356 /* Checks whether there is any job running for the units this
357 * job needs to be running after (in the case of a 'positive'
358 * job type) or before (in the case of a 'negative' job
361 /* First check if there is an override */
365 if (j->type == JOB_START ||
366 j->type == JOB_VERIFY_ACTIVE ||
367 j->type == JOB_RELOAD ||
368 j->type == JOB_RELOAD_OR_START) {
370 /* Immediate result is that the job is or might be
371 * started. In this case lets wait for the
372 * dependencies, regardless whether they are
373 * starting or stopping something. */
375 SET_FOREACH(other, j->unit->dependencies[UNIT_AFTER], i)
380 /* Also, if something else is being stopped and we should
381 * change state after it, then lets wait. */
383 SET_FOREACH(other, j->unit->dependencies[UNIT_BEFORE], i)
385 (other->job->type == JOB_STOP ||
386 other->job->type == JOB_RESTART ||
387 other->job->type == JOB_TRY_RESTART))
390 /* This means that for a service a and a service b where b
391 * shall be started after a:
393 * start a + start b → 1st step start a, 2nd step start b
394 * start a + stop b → 1st step stop b, 2nd step start a
395 * stop a + start b → 1st step stop a, 2nd step start b
396 * stop a + stop b → 1st step stop b, 2nd step stop a
398 * This has the side effect that restarts are properly
399 * synchronized too. */
404 static void job_change_type(Job *j, JobType newtype) {
405 log_debug("Converting job %s/%s -> %s/%s",
406 j->unit->id, job_type_to_string(j->type),
407 j->unit->id, job_type_to_string(newtype));
412 int job_run_and_invalidate(Job *j) {
418 assert(j->installed);
420 if (j->in_run_queue) {
421 LIST_REMOVE(Job, run_queue, j->manager->run_queue, j);
422 j->in_run_queue = false;
425 if (j->state != JOB_WAITING)
428 if (!job_is_runnable(j))
431 j->state = JOB_RUNNING;
432 job_add_to_dbus_queue(j);
434 /* While we execute this operation the job might go away (for
435 * example: because it is replaced by a new, conflicting
436 * job.) To make sure we don't access a freed job later on we
437 * store the id here, so that we can verify the job is still
444 case JOB_RELOAD_OR_START:
445 if (unit_active_state(j->unit) == UNIT_ACTIVE) {
446 job_change_type(j, JOB_RELOAD);
447 r = unit_reload(j->unit);
450 job_change_type(j, JOB_START);
454 r = unit_start(j->unit);
456 /* If this unit cannot be started, then simply wait */
461 case JOB_VERIFY_ACTIVE: {
462 UnitActiveState t = unit_active_state(j->unit);
463 if (UNIT_IS_ACTIVE_OR_RELOADING(t))
465 else if (t == UNIT_ACTIVATING)
472 case JOB_TRY_RESTART:
473 if (UNIT_IS_INACTIVE_OR_DEACTIVATING(unit_active_state(j->unit))) {
477 job_change_type(j, JOB_RESTART);
482 r = unit_stop(j->unit);
484 /* If this unit cannot stopped, then simply wait. */
490 r = unit_reload(j->unit);
494 assert_not_reached("Unknown job type");
497 if ((j = manager_get_job(m, id))) {
499 r = job_finish_and_invalidate(j, JOB_DONE);
500 else if (r == -ENOEXEC)
501 r = job_finish_and_invalidate(j, JOB_SKIPPED);
502 else if (r == -EAGAIN)
503 j->state = JOB_WAITING;
505 r = job_finish_and_invalidate(j, JOB_FAILED);
511 static void job_print_status_message(Unit *u, JobType t, JobResult result) {
514 if (t == JOB_START) {
519 if (u->condition_result)
520 unit_status_printf(u, ANSI_HIGHLIGHT_GREEN_ON " OK " ANSI_HIGHLIGHT_OFF, "Started %s.", unit_description(u));
524 unit_status_printf(u, ANSI_HIGHLIGHT_RED_ON "FAILED" ANSI_HIGHLIGHT_OFF, "Failed to start %s.", unit_description(u));
525 unit_status_printf(u, NULL, "See 'systemctl status %s' for details.", u->id);
529 unit_status_printf(u, ANSI_HIGHLIGHT_RED_ON " ABORT" ANSI_HIGHLIGHT_OFF, "Dependency failed. Aborted start of %s.", unit_description(u));
533 unit_status_printf(u, ANSI_HIGHLIGHT_RED_ON " TIME " ANSI_HIGHLIGHT_OFF, "Timed out starting %s.", unit_description(u));
540 } else if (t == JOB_STOP) {
545 unit_status_printf(u, ANSI_HIGHLIGHT_RED_ON " TIME " ANSI_HIGHLIGHT_OFF, "Timed out stopping %s.", unit_description(u));
550 unit_status_printf(u, ANSI_HIGHLIGHT_GREEN_ON " OK " ANSI_HIGHLIGHT_OFF, "Stopped %s.", unit_description(u));
559 int job_finish_and_invalidate(Job *j, JobResult result) {
564 bool recursed = false;
567 assert(j->installed);
569 job_add_to_dbus_queue(j);
571 /* Patch restart jobs so that they become normal start jobs */
572 if (result == JOB_DONE && j->type == JOB_RESTART) {
574 job_change_type(j, JOB_START);
575 j->state = JOB_WAITING;
577 job_add_to_run_queue(j);
585 log_debug("Job %s/%s finished, result=%s", j->unit->id, job_type_to_string(j->type), job_result_to_string(result));
587 if (result == JOB_FAILED)
588 j->manager->n_failed_jobs ++;
595 job_print_status_message(u, t, result);
597 /* Fail depending jobs on failure */
598 if (result != JOB_DONE) {
600 if (t == JOB_START ||
601 t == JOB_VERIFY_ACTIVE ||
602 t == JOB_RELOAD_OR_START) {
604 SET_FOREACH(other, u->dependencies[UNIT_REQUIRED_BY], i)
606 (other->job->type == JOB_START ||
607 other->job->type == JOB_VERIFY_ACTIVE ||
608 other->job->type == JOB_RELOAD_OR_START)) {
609 job_finish_and_invalidate(other->job, JOB_DEPENDENCY);
613 SET_FOREACH(other, u->dependencies[UNIT_BOUND_BY], i)
615 (other->job->type == JOB_START ||
616 other->job->type == JOB_VERIFY_ACTIVE ||
617 other->job->type == JOB_RELOAD_OR_START)) {
618 job_finish_and_invalidate(other->job, JOB_DEPENDENCY);
622 SET_FOREACH(other, u->dependencies[UNIT_REQUIRED_BY_OVERRIDABLE], i)
624 !other->job->override &&
625 (other->job->type == JOB_START ||
626 other->job->type == JOB_VERIFY_ACTIVE ||
627 other->job->type == JOB_RELOAD_OR_START)) {
628 job_finish_and_invalidate(other->job, JOB_DEPENDENCY);
632 } else if (t == JOB_STOP) {
634 SET_FOREACH(other, u->dependencies[UNIT_CONFLICTED_BY], i)
636 (other->job->type == JOB_START ||
637 other->job->type == JOB_VERIFY_ACTIVE ||
638 other->job->type == JOB_RELOAD_OR_START)) {
639 job_finish_and_invalidate(other->job, JOB_DEPENDENCY);
645 /* Trigger OnFailure dependencies that are not generated by
646 * the unit itself. We don't tread JOB_CANCELED as failure in
647 * this context. And JOB_FAILURE is already handled by the
649 if (result == JOB_TIMEOUT || result == JOB_DEPENDENCY) {
650 log_notice("Job %s/%s failed with result '%s'.",
652 job_type_to_string(t),
653 job_result_to_string(result));
655 unit_trigger_on_failure(u);
659 /* Try to start the next jobs that can be started */
660 SET_FOREACH(other, u->dependencies[UNIT_AFTER], i)
662 job_add_to_run_queue(other->job);
663 SET_FOREACH(other, u->dependencies[UNIT_BEFORE], i)
665 job_add_to_run_queue(other->job);
667 manager_check_finished(u->manager);
672 int job_start_timer(Job *j) {
673 struct itimerspec its;
674 struct epoll_event ev;
678 if (j->unit->job_timeout <= 0 ||
679 j->timer_watch.type == WATCH_JOB_TIMER)
682 assert(j->timer_watch.type == WATCH_INVALID);
684 if ((fd = timerfd_create(CLOCK_MONOTONIC, TFD_NONBLOCK|TFD_CLOEXEC)) < 0) {
690 timespec_store(&its.it_value, j->unit->job_timeout);
692 if (timerfd_settime(fd, 0, &its, NULL) < 0) {
698 ev.data.ptr = &j->timer_watch;
701 if (epoll_ctl(j->manager->epoll_fd, EPOLL_CTL_ADD, fd, &ev) < 0) {
706 j->timer_watch.type = WATCH_JOB_TIMER;
707 j->timer_watch.fd = fd;
708 j->timer_watch.data.job = j;
714 close_nointr_nofail(fd);
719 void job_add_to_run_queue(Job *j) {
721 assert(j->installed);
726 LIST_PREPEND(Job, run_queue, j->manager->run_queue, j);
727 j->in_run_queue = true;
730 void job_add_to_dbus_queue(Job *j) {
732 assert(j->installed);
734 if (j->in_dbus_queue)
737 /* We don't check if anybody is subscribed here, since this
738 * job might just have been created and not yet assigned to a
739 * connection/client. */
741 LIST_PREPEND(Job, dbus_queue, j->manager->dbus_job_queue, j);
742 j->in_dbus_queue = true;
745 char *job_dbus_path(Job *j) {
750 if (asprintf(&p, "/org/freedesktop/systemd1/job/%lu", (unsigned long) j->id) < 0)
756 void job_timer_event(Job *j, uint64_t n_elapsed, Watch *w) {
758 assert(w == &j->timer_watch);
760 log_warning("Job %s/%s timed out.", j->unit->id, job_type_to_string(j->type));
761 job_finish_and_invalidate(j, JOB_TIMEOUT);
764 int job_serialize(Job *j, FILE *f, FDSet *fds) {
765 fprintf(f, "job-id=%u\n", j->id);
766 fprintf(f, "job-type=%s\n", job_type_to_string(j->type));
767 fprintf(f, "job-state=%s\n", job_state_to_string(j->state));
768 fprintf(f, "job-override=%s\n", yes_no(j->override));
769 fprintf(f, "job-sent-dbus-new-signal=%s\n", yes_no(j->sent_dbus_new_signal));
770 fprintf(f, "job-ignore-order=%s\n", yes_no(j->ignore_order));
771 /* Cannot save bus clients. Just note the fact that we're losing
772 * them. job_send_message() will fallback to broadcasting. */
773 fprintf(f, "job-forgot-bus-clients=%s\n",
774 yes_no(j->forgot_bus_clients || j->bus_client_list));
775 if (j->timer_watch.type == WATCH_JOB_TIMER) {
776 int copy = fdset_put_dup(fds, j->timer_watch.fd);
779 fprintf(f, "job-timer-watch-fd=%d\n", copy);
787 int job_deserialize(Job *j, FILE *f, FDSet *fds) {
789 char line[LINE_MAX], *l, *v;
792 if (!fgets(line, sizeof(line), f)) {
813 if (streq(l, "job-id")) {
814 if (safe_atou32(v, &j->id) < 0)
815 log_debug("Failed to parse job id value %s", v);
816 } else if (streq(l, "job-type")) {
817 JobType t = job_type_from_string(v);
819 log_debug("Failed to parse job type %s", v);
822 } else if (streq(l, "job-state")) {
823 JobState s = job_state_from_string(v);
825 log_debug("Failed to parse job state %s", v);
828 } else if (streq(l, "job-override")) {
829 int b = parse_boolean(v);
831 log_debug("Failed to parse job override flag %s", v);
833 j->override = j->override || b;
834 } else if (streq(l, "job-sent-dbus-new-signal")) {
835 int b = parse_boolean(v);
837 log_debug("Failed to parse job sent_dbus_new_signal flag %s", v);
839 j->sent_dbus_new_signal = j->sent_dbus_new_signal || b;
840 } else if (streq(l, "job-ignore-order")) {
841 int b = parse_boolean(v);
843 log_debug("Failed to parse job ignore_order flag %s", v);
845 j->ignore_order = j->ignore_order || b;
846 } else if (streq(l, "job-forgot-bus-clients")) {
847 int b = parse_boolean(v);
849 log_debug("Failed to parse job forgot_bus_clients flag %s", v);
851 j->forgot_bus_clients = j->forgot_bus_clients || b;
852 } else if (streq(l, "job-timer-watch-fd")) {
854 if (safe_atoi(v, &fd) < 0 || fd < 0 || !fdset_contains(fds, fd))
855 log_debug("Failed to parse job-timer-watch-fd value %s", v);
857 if (j->timer_watch.type == WATCH_JOB_TIMER)
858 close_nointr_nofail(j->timer_watch.fd);
860 j->timer_watch.type = WATCH_JOB_TIMER;
861 j->timer_watch.fd = fdset_remove(fds, fd);
862 j->timer_watch.data.job = j;
868 int job_coldplug(Job *j) {
869 struct epoll_event ev;
871 if (j->timer_watch.type != WATCH_JOB_TIMER)
875 ev.data.ptr = &j->timer_watch;
878 if (epoll_ctl(j->manager->epoll_fd, EPOLL_CTL_ADD, j->timer_watch.fd, &ev) < 0)
884 static const char* const job_state_table[_JOB_STATE_MAX] = {
885 [JOB_WAITING] = "waiting",
886 [JOB_RUNNING] = "running"
889 DEFINE_STRING_TABLE_LOOKUP(job_state, JobState);
891 static const char* const job_type_table[_JOB_TYPE_MAX] = {
892 [JOB_START] = "start",
893 [JOB_VERIFY_ACTIVE] = "verify-active",
895 [JOB_RELOAD] = "reload",
896 [JOB_RELOAD_OR_START] = "reload-or-start",
897 [JOB_RESTART] = "restart",
898 [JOB_TRY_RESTART] = "try-restart",
901 DEFINE_STRING_TABLE_LOOKUP(job_type, JobType);
903 static const char* const job_mode_table[_JOB_MODE_MAX] = {
905 [JOB_REPLACE] = "replace",
906 [JOB_ISOLATE] = "isolate",
907 [JOB_IGNORE_DEPENDENCIES] = "ignore-dependencies",
908 [JOB_IGNORE_REQUIREMENTS] = "ignore-requirements"
911 DEFINE_STRING_TABLE_LOOKUP(job_mode, JobMode);
913 static const char* const job_result_table[_JOB_RESULT_MAX] = {
915 [JOB_CANCELED] = "canceled",
916 [JOB_TIMEOUT] = "timeout",
917 [JOB_FAILED] = "failed",
918 [JOB_DEPENDENCY] = "dependency",
919 [JOB_SKIPPED] = "skipped"
922 DEFINE_STRING_TABLE_LOOKUP(job_result, JobResult);