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->type = _JOB_TYPE_INVALID;
64 j->timer_watch.type = WATCH_INVALID;
69 Job* job_new(Unit *unit, JobType type) {
72 assert(type < _JOB_TYPE_MAX);
74 j = job_new_raw(unit);
78 j->id = j->manager->current_job_id++;
81 /* We don't link it here, that's what job_dependency() is for */
86 void job_free(Job *j) {
90 assert(!j->installed);
91 assert(!j->transaction_prev);
92 assert(!j->transaction_next);
93 assert(!j->subject_list);
94 assert(!j->object_list);
97 LIST_REMOVE(Job, run_queue, j->manager->run_queue, j);
100 LIST_REMOVE(Job, dbus_queue, j->manager->dbus_job_queue, j);
102 if (j->timer_watch.type != WATCH_INVALID) {
103 assert(j->timer_watch.type == WATCH_JOB_TIMER);
104 assert(j->timer_watch.data.job == j);
105 assert(j->timer_watch.fd >= 0);
107 assert_se(epoll_ctl(j->manager->epoll_fd, EPOLL_CTL_DEL, j->timer_watch.fd, NULL) >= 0);
108 close_nointr_nofail(j->timer_watch.fd);
111 while ((cl = j->bus_client_list)) {
112 LIST_REMOVE(JobBusClient, client, j->bus_client_list, cl);
118 void job_uninstall(Job *j) {
121 assert(j->installed);
123 pj = (j->type == JOB_NOP) ? &j->unit->nop_job : &j->unit->job;
126 /* Detach from next 'bigger' objects */
128 /* daemon-reload should be transparent to job observers */
129 if (j->manager->n_reloading <= 0)
130 bus_job_send_removed_signal(j);
134 unit_add_to_gc_queue(j->unit);
136 hashmap_remove(j->manager->jobs, UINT32_TO_PTR(j->id));
137 j->installed = false;
140 static bool job_type_allows_late_merge(JobType t) {
141 /* Tells whether it is OK to merge a job of type 't' with an already
143 * Reloads cannot be merged this way. Think of the sequence:
144 * 1. Reload of a daemon is in progress; the daemon has already loaded
145 * its config file, but hasn't completed the reload operation yet.
146 * 2. Edit foo's config file.
147 * 3. Trigger another reload to have the daemon use the new config.
148 * Should the second reload job be merged into the first one, the daemon
149 * would not know about the new config.
150 * JOB_RESTART jobs on the other hand can be merged, because they get
151 * patched into JOB_START after stopping the unit. So if we see a
152 * JOB_RESTART running, it means the unit hasn't stopped yet and at
153 * this time the merge is still allowed. */
154 return t != JOB_RELOAD;
157 static void job_merge_into_installed(Job *j, Job *other) {
158 assert(j->installed);
159 assert(j->unit == other->unit);
161 if (j->type != JOB_NOP)
162 job_type_merge_and_collapse(&j->type, other->type, j->unit);
164 assert(other->type == JOB_NOP);
166 j->override = j->override || other->override;
169 Job* job_install(Job *j) {
173 assert(!j->installed);
174 assert(j->type < _JOB_TYPE_MAX_IN_TRANSACTION);
176 pj = (j->type == JOB_NOP) ? &j->unit->nop_job : &j->unit->job;
180 if (j->type != JOB_NOP && job_type_is_conflicting(uj->type, j->type))
181 job_finish_and_invalidate(uj, JOB_CANCELED, true);
183 /* not conflicting, i.e. mergeable */
185 if (j->type == JOB_NOP || uj->state == JOB_WAITING ||
186 (job_type_allows_late_merge(j->type) && job_type_is_superset(uj->type, j->type))) {
187 job_merge_into_installed(uj, j);
188 log_debug("Merged into installed job %s/%s as %u",
189 uj->unit->id, job_type_to_string(uj->type), (unsigned) uj->id);
192 /* already running and not safe to merge into */
193 /* Patch uj to become a merged job and re-run it. */
194 /* XXX It should be safer to queue j to run after uj finishes, but it is
195 * not currently possible to have more than one installed job per unit. */
196 job_merge_into_installed(uj, j);
197 log_debug("Merged into running job, re-running: %s/%s as %u",
198 uj->unit->id, job_type_to_string(uj->type), (unsigned) uj->id);
199 uj->state = JOB_WAITING;
205 /* Install the job */
208 j->manager->n_installed_jobs ++;
209 log_debug("Installed new job %s/%s as %u", j->unit->id, job_type_to_string(j->type), (unsigned) j->id);
213 int job_install_deserialized(Job *j) {
216 assert(!j->installed);
218 if (j->type < 0 || j->type >= _JOB_TYPE_MAX_IN_TRANSACTION) {
219 log_debug("Invalid job type %s in deserialization.", strna(job_type_to_string(j->type)));
223 pj = (j->type == JOB_NOP) ? &j->unit->nop_job : &j->unit->job;
226 log_debug("Unit %s already has a job installed. Not installing deserialized job.", j->unit->id);
231 log_debug("Reinstalled deserialized job %s/%s as %u", j->unit->id, job_type_to_string(j->type), (unsigned) j->id);
235 JobDependency* job_dependency_new(Job *subject, Job *object, bool matters, bool conflicts) {
240 /* Adds a new job link, which encodes that the 'subject' job
241 * needs the 'object' job in some way. If 'subject' is NULL
242 * this means the 'anchor' job (i.e. the one the user
243 * explicitly asked for) is the requester. */
245 if (!(l = new0(JobDependency, 1)))
248 l->subject = subject;
250 l->matters = matters;
251 l->conflicts = conflicts;
254 LIST_PREPEND(JobDependency, subject, subject->subject_list, l);
256 LIST_PREPEND(JobDependency, object, object->object_list, l);
261 void job_dependency_free(JobDependency *l) {
265 LIST_REMOVE(JobDependency, subject, l->subject->subject_list, l);
267 LIST_REMOVE(JobDependency, object, l->object->object_list, l);
272 void job_dump(Job *j, FILE*f, const char *prefix) {
281 "%s\tAction: %s -> %s\n"
285 prefix, j->unit->id, job_type_to_string(j->type),
286 prefix, job_state_to_string(j->state),
287 prefix, yes_no(j->override));
291 * Merging is commutative, so imagine the matrix as symmetric. We store only
292 * its lower triangle to avoid duplication. We don't store the main diagonal,
293 * because A merged with A is simply A.
295 * If the resulting type is collapsed immediately afterwards (to get rid of
296 * the JOB_RELOAD_OR_START, which lies outside the lookup function's domain),
297 * the following properties hold:
299 * Merging is associative! A merged with B merged with C is the same as
300 * A merged with C merged with B.
302 * Mergeability is transitive! If A can be merged with B and B with C then
305 * Also, if A merged with B cannot be merged with C, then either A or B cannot
306 * be merged with C either.
308 static const JobType job_merging_table[] = {
309 /* What \ With * JOB_START JOB_VERIFY_ACTIVE JOB_STOP JOB_RELOAD */
310 /*********************************************************************************/
312 /*JOB_VERIFY_ACTIVE */ JOB_START,
313 /*JOB_STOP */ -1, -1,
314 /*JOB_RELOAD */ JOB_RELOAD_OR_START, JOB_RELOAD, -1,
315 /*JOB_RESTART */ JOB_RESTART, JOB_RESTART, -1, JOB_RESTART,
318 JobType job_type_lookup_merge(JobType a, JobType b) {
319 assert_cc(ELEMENTSOF(job_merging_table) == _JOB_TYPE_MAX_MERGING * (_JOB_TYPE_MAX_MERGING - 1) / 2);
320 assert(a >= 0 && a < _JOB_TYPE_MAX_MERGING);
321 assert(b >= 0 && b < _JOB_TYPE_MAX_MERGING);
332 return job_merging_table[(a - 1) * a / 2 + b];
335 bool job_type_is_redundant(JobType a, UnitActiveState b) {
345 b == UNIT_INACTIVE ||
348 case JOB_VERIFY_ACTIVE:
359 b == UNIT_ACTIVATING;
362 assert_not_reached("Invalid job type");
366 void job_type_collapse(JobType *t, Unit *u) {
371 case JOB_TRY_RESTART:
372 s = unit_active_state(u);
373 if (UNIT_IS_INACTIVE_OR_DEACTIVATING(s))
379 case JOB_RELOAD_OR_START:
380 s = unit_active_state(u);
381 if (UNIT_IS_INACTIVE_OR_DEACTIVATING(s))
392 int job_type_merge_and_collapse(JobType *a, JobType b, Unit *u) {
393 JobType t = job_type_lookup_merge(*a, b);
397 job_type_collapse(a, u);
401 bool job_is_runnable(Job *j) {
406 assert(j->installed);
408 /* Checks whether there is any job running for the units this
409 * job needs to be running after (in the case of a 'positive'
410 * job type) or before (in the case of a 'negative' job
413 /* First check if there is an override */
417 if (j->type == JOB_NOP)
420 if (j->type == JOB_START ||
421 j->type == JOB_VERIFY_ACTIVE ||
422 j->type == JOB_RELOAD) {
424 /* Immediate result is that the job is or might be
425 * started. In this case lets wait for the
426 * dependencies, regardless whether they are
427 * starting or stopping something. */
429 SET_FOREACH(other, j->unit->dependencies[UNIT_AFTER], i)
434 /* Also, if something else is being stopped and we should
435 * change state after it, then lets wait. */
437 SET_FOREACH(other, j->unit->dependencies[UNIT_BEFORE], i)
439 (other->job->type == JOB_STOP ||
440 other->job->type == JOB_RESTART))
443 /* This means that for a service a and a service b where b
444 * shall be started after a:
446 * start a + start b → 1st step start a, 2nd step start b
447 * start a + stop b → 1st step stop b, 2nd step start a
448 * stop a + start b → 1st step stop a, 2nd step start b
449 * stop a + stop b → 1st step stop b, 2nd step stop a
451 * This has the side effect that restarts are properly
452 * synchronized too. */
457 static void job_change_type(Job *j, JobType newtype) {
458 log_debug("Converting job %s/%s -> %s/%s",
459 j->unit->id, job_type_to_string(j->type),
460 j->unit->id, job_type_to_string(newtype));
465 int job_run_and_invalidate(Job *j) {
471 assert(j->installed);
472 assert(j->type < _JOB_TYPE_MAX_IN_TRANSACTION);
474 if (j->in_run_queue) {
475 LIST_REMOVE(Job, run_queue, j->manager->run_queue, j);
476 j->in_run_queue = false;
479 if (j->state != JOB_WAITING)
482 if (!job_is_runnable(j))
485 j->state = JOB_RUNNING;
486 job_add_to_dbus_queue(j);
488 /* While we execute this operation the job might go away (for
489 * example: because it is replaced by a new, conflicting
490 * job.) To make sure we don't access a freed job later on we
491 * store the id here, so that we can verify the job is still
499 r = unit_start(j->unit);
501 /* If this unit cannot be started, then simply wait */
506 case JOB_VERIFY_ACTIVE: {
507 UnitActiveState t = unit_active_state(j->unit);
508 if (UNIT_IS_ACTIVE_OR_RELOADING(t))
510 else if (t == UNIT_ACTIVATING)
519 r = unit_stop(j->unit);
521 /* If this unit cannot stopped, then simply wait. */
527 r = unit_reload(j->unit);
535 assert_not_reached("Unknown job type");
538 j = manager_get_job(m, id);
541 r = job_finish_and_invalidate(j, JOB_DONE, true);
542 else if (r == -ENOEXEC)
543 r = job_finish_and_invalidate(j, JOB_SKIPPED, true);
544 else if (r == -EAGAIN)
545 j->state = JOB_WAITING;
547 r = job_finish_and_invalidate(j, JOB_FAILED, true);
553 static void job_print_status_message(Unit *u, JobType t, JobResult result) {
556 if (t == JOB_START) {
561 if (u->condition_result)
562 unit_status_printf(u, ANSI_HIGHLIGHT_GREEN_ON " OK " ANSI_HIGHLIGHT_OFF, "Started %s.", unit_description(u));
566 unit_status_printf(u, ANSI_HIGHLIGHT_RED_ON "FAILED" ANSI_HIGHLIGHT_OFF, "Failed to start %s.", unit_description(u));
567 unit_status_printf(u, NULL, "See 'systemctl status %s' for details.", u->id);
571 unit_status_printf(u, ANSI_HIGHLIGHT_RED_ON " ABORT" ANSI_HIGHLIGHT_OFF, "Dependency failed. Aborted start of %s.", unit_description(u));
575 unit_status_printf(u, ANSI_HIGHLIGHT_RED_ON " TIME " ANSI_HIGHLIGHT_OFF, "Timed out starting %s.", unit_description(u));
582 } else if (t == JOB_STOP) {
587 unit_status_printf(u, ANSI_HIGHLIGHT_RED_ON " TIME " ANSI_HIGHLIGHT_OFF, "Timed out stopping %s.", unit_description(u));
592 unit_status_printf(u, ANSI_HIGHLIGHT_GREEN_ON " OK " ANSI_HIGHLIGHT_OFF, "Stopped %s.", unit_description(u));
601 int job_finish_and_invalidate(Job *j, JobResult result, bool recursive) {
608 assert(j->installed);
609 assert(j->type < _JOB_TYPE_MAX_IN_TRANSACTION);
611 job_add_to_dbus_queue(j);
613 /* Patch restart jobs so that they become normal start jobs */
614 if (result == JOB_DONE && j->type == JOB_RESTART) {
616 job_change_type(j, JOB_START);
617 j->state = JOB_WAITING;
619 job_add_to_run_queue(j);
627 log_debug("Job %s/%s finished, result=%s", j->unit->id, job_type_to_string(j->type), job_result_to_string(result));
629 if (result == JOB_FAILED)
630 j->manager->n_failed_jobs ++;
637 job_print_status_message(u, t, result);
639 /* Fail depending jobs on failure */
640 if (result != JOB_DONE && recursive) {
642 if (t == JOB_START ||
643 t == JOB_VERIFY_ACTIVE) {
645 SET_FOREACH(other, u->dependencies[UNIT_REQUIRED_BY], i)
647 (other->job->type == JOB_START ||
648 other->job->type == JOB_VERIFY_ACTIVE))
649 job_finish_and_invalidate(other->job, JOB_DEPENDENCY, true);
651 SET_FOREACH(other, u->dependencies[UNIT_BOUND_BY], i)
653 (other->job->type == JOB_START ||
654 other->job->type == JOB_VERIFY_ACTIVE))
655 job_finish_and_invalidate(other->job, JOB_DEPENDENCY, true);
657 SET_FOREACH(other, u->dependencies[UNIT_REQUIRED_BY_OVERRIDABLE], i)
659 !other->job->override &&
660 (other->job->type == JOB_START ||
661 other->job->type == JOB_VERIFY_ACTIVE))
662 job_finish_and_invalidate(other->job, JOB_DEPENDENCY, true);
664 } else if (t == JOB_STOP) {
666 SET_FOREACH(other, u->dependencies[UNIT_CONFLICTED_BY], i)
668 (other->job->type == JOB_START ||
669 other->job->type == JOB_VERIFY_ACTIVE))
670 job_finish_and_invalidate(other->job, JOB_DEPENDENCY, true);
674 /* Trigger OnFailure dependencies that are not generated by
675 * the unit itself. We don't tread JOB_CANCELED as failure in
676 * this context. And JOB_FAILURE is already handled by the
678 if (result == JOB_TIMEOUT || result == JOB_DEPENDENCY) {
679 log_notice("Job %s/%s failed with result '%s'.",
681 job_type_to_string(t),
682 job_result_to_string(result));
684 unit_trigger_on_failure(u);
688 /* Try to start the next jobs that can be started */
689 SET_FOREACH(other, u->dependencies[UNIT_AFTER], i)
691 job_add_to_run_queue(other->job);
692 SET_FOREACH(other, u->dependencies[UNIT_BEFORE], i)
694 job_add_to_run_queue(other->job);
696 manager_check_finished(u->manager);
701 int job_start_timer(Job *j) {
702 struct itimerspec its;
703 struct epoll_event ev;
707 if (j->unit->job_timeout <= 0 ||
708 j->timer_watch.type == WATCH_JOB_TIMER)
711 assert(j->timer_watch.type == WATCH_INVALID);
713 if ((fd = timerfd_create(CLOCK_MONOTONIC, TFD_NONBLOCK|TFD_CLOEXEC)) < 0) {
719 timespec_store(&its.it_value, j->unit->job_timeout);
721 if (timerfd_settime(fd, 0, &its, NULL) < 0) {
727 ev.data.ptr = &j->timer_watch;
730 if (epoll_ctl(j->manager->epoll_fd, EPOLL_CTL_ADD, fd, &ev) < 0) {
735 j->timer_watch.type = WATCH_JOB_TIMER;
736 j->timer_watch.fd = fd;
737 j->timer_watch.data.job = j;
743 close_nointr_nofail(fd);
748 void job_add_to_run_queue(Job *j) {
750 assert(j->installed);
755 LIST_PREPEND(Job, run_queue, j->manager->run_queue, j);
756 j->in_run_queue = true;
759 void job_add_to_dbus_queue(Job *j) {
761 assert(j->installed);
763 if (j->in_dbus_queue)
766 /* We don't check if anybody is subscribed here, since this
767 * job might just have been created and not yet assigned to a
768 * connection/client. */
770 LIST_PREPEND(Job, dbus_queue, j->manager->dbus_job_queue, j);
771 j->in_dbus_queue = true;
774 char *job_dbus_path(Job *j) {
779 if (asprintf(&p, "/org/freedesktop/systemd1/job/%lu", (unsigned long) j->id) < 0)
785 void job_timer_event(Job *j, uint64_t n_elapsed, Watch *w) {
787 assert(w == &j->timer_watch);
789 log_warning("Job %s/%s timed out.", j->unit->id, job_type_to_string(j->type));
790 job_finish_and_invalidate(j, JOB_TIMEOUT, true);
793 int job_serialize(Job *j, FILE *f, FDSet *fds) {
794 fprintf(f, "job-id=%u\n", j->id);
795 fprintf(f, "job-type=%s\n", job_type_to_string(j->type));
796 fprintf(f, "job-state=%s\n", job_state_to_string(j->state));
797 fprintf(f, "job-override=%s\n", yes_no(j->override));
798 fprintf(f, "job-sent-dbus-new-signal=%s\n", yes_no(j->sent_dbus_new_signal));
799 fprintf(f, "job-ignore-order=%s\n", yes_no(j->ignore_order));
800 /* Cannot save bus clients. Just note the fact that we're losing
801 * them. job_send_message() will fallback to broadcasting. */
802 fprintf(f, "job-forgot-bus-clients=%s\n",
803 yes_no(j->forgot_bus_clients || j->bus_client_list));
804 if (j->timer_watch.type == WATCH_JOB_TIMER) {
805 int copy = fdset_put_dup(fds, j->timer_watch.fd);
808 fprintf(f, "job-timer-watch-fd=%d\n", copy);
816 int job_deserialize(Job *j, FILE *f, FDSet *fds) {
818 char line[LINE_MAX], *l, *v;
821 if (!fgets(line, sizeof(line), f)) {
842 if (streq(l, "job-id")) {
843 if (safe_atou32(v, &j->id) < 0)
844 log_debug("Failed to parse job id value %s", v);
845 } else if (streq(l, "job-type")) {
846 JobType t = job_type_from_string(v);
848 log_debug("Failed to parse job type %s", v);
849 else if (t >= _JOB_TYPE_MAX_IN_TRANSACTION)
850 log_debug("Cannot deserialize job of type %s", v);
853 } else if (streq(l, "job-state")) {
854 JobState s = job_state_from_string(v);
856 log_debug("Failed to parse job state %s", v);
859 } else if (streq(l, "job-override")) {
860 int b = parse_boolean(v);
862 log_debug("Failed to parse job override flag %s", v);
864 j->override = j->override || b;
865 } else if (streq(l, "job-sent-dbus-new-signal")) {
866 int b = parse_boolean(v);
868 log_debug("Failed to parse job sent_dbus_new_signal flag %s", v);
870 j->sent_dbus_new_signal = j->sent_dbus_new_signal || b;
871 } else if (streq(l, "job-ignore-order")) {
872 int b = parse_boolean(v);
874 log_debug("Failed to parse job ignore_order flag %s", v);
876 j->ignore_order = j->ignore_order || b;
877 } else if (streq(l, "job-forgot-bus-clients")) {
878 int b = parse_boolean(v);
880 log_debug("Failed to parse job forgot_bus_clients flag %s", v);
882 j->forgot_bus_clients = j->forgot_bus_clients || b;
883 } else if (streq(l, "job-timer-watch-fd")) {
885 if (safe_atoi(v, &fd) < 0 || fd < 0 || !fdset_contains(fds, fd))
886 log_debug("Failed to parse job-timer-watch-fd value %s", v);
888 if (j->timer_watch.type == WATCH_JOB_TIMER)
889 close_nointr_nofail(j->timer_watch.fd);
891 j->timer_watch.type = WATCH_JOB_TIMER;
892 j->timer_watch.fd = fdset_remove(fds, fd);
893 j->timer_watch.data.job = j;
899 int job_coldplug(Job *j) {
900 struct epoll_event ev;
902 if (j->timer_watch.type != WATCH_JOB_TIMER)
906 ev.data.ptr = &j->timer_watch;
909 if (epoll_ctl(j->manager->epoll_fd, EPOLL_CTL_ADD, j->timer_watch.fd, &ev) < 0)
915 static const char* const job_state_table[_JOB_STATE_MAX] = {
916 [JOB_WAITING] = "waiting",
917 [JOB_RUNNING] = "running"
920 DEFINE_STRING_TABLE_LOOKUP(job_state, JobState);
922 static const char* const job_type_table[_JOB_TYPE_MAX] = {
923 [JOB_START] = "start",
924 [JOB_VERIFY_ACTIVE] = "verify-active",
926 [JOB_RELOAD] = "reload",
927 [JOB_RELOAD_OR_START] = "reload-or-start",
928 [JOB_RESTART] = "restart",
929 [JOB_TRY_RESTART] = "try-restart",
933 DEFINE_STRING_TABLE_LOOKUP(job_type, JobType);
935 static const char* const job_mode_table[_JOB_MODE_MAX] = {
937 [JOB_REPLACE] = "replace",
938 [JOB_ISOLATE] = "isolate",
939 [JOB_IGNORE_DEPENDENCIES] = "ignore-dependencies",
940 [JOB_IGNORE_REQUIREMENTS] = "ignore-requirements"
943 DEFINE_STRING_TABLE_LOOKUP(job_mode, JobMode);
945 static const char* const job_result_table[_JOB_RESULT_MAX] = {
947 [JOB_CANCELED] = "canceled",
948 [JOB_TIMEOUT] = "timeout",
949 [JOB_FAILED] = "failed",
950 [JOB_DEPENDENCY] = "dependency",
951 [JOB_SKIPPED] = "skipped"
954 DEFINE_STRING_TABLE_LOOKUP(job_result, JobResult);