]> git.ipfire.org Git - thirdparty/systemd.git/blob - src/core/job.c
core: add function to tell when job will time out
[thirdparty/systemd.git] / src / core / job.c
1 /*-*- Mode: C; c-basic-offset: 8; indent-tabs-mode: nil -*-*/
2
3 /***
4 This file is part of systemd.
5
6 Copyright 2010 Lennart Poettering
7
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.
12
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.
17
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/>.
20 ***/
21
22 #include <assert.h>
23 #include <errno.h>
24 #include <sys/timerfd.h>
25 #include <sys/epoll.h>
26
27 #include "sd-id128.h"
28 #include "sd-messages.h"
29 #include "set.h"
30 #include "unit.h"
31 #include "macro.h"
32 #include "strv.h"
33 #include "load-fragment.h"
34 #include "load-dropin.h"
35 #include "log.h"
36 #include "dbus-job.h"
37 #include "special.h"
38 #include "async.h"
39 #include "virt.h"
40 #include "dbus-client-track.h"
41
42 Job* job_new_raw(Unit *unit) {
43 Job *j;
44
45 /* used for deserialization */
46
47 assert(unit);
48
49 j = new0(Job, 1);
50 if (!j)
51 return NULL;
52
53 j->manager = unit->manager;
54 j->unit = unit;
55 j->type = _JOB_TYPE_INVALID;
56
57 return j;
58 }
59
60 Job* job_new(Unit *unit, JobType type) {
61 Job *j;
62
63 assert(type < _JOB_TYPE_MAX);
64
65 j = job_new_raw(unit);
66 if (!j)
67 return NULL;
68
69 j->id = j->manager->current_job_id++;
70 j->type = type;
71
72 /* We don't link it here, that's what job_dependency() is for */
73
74 return j;
75 }
76
77 void job_free(Job *j) {
78 assert(j);
79 assert(!j->installed);
80 assert(!j->transaction_prev);
81 assert(!j->transaction_next);
82 assert(!j->subject_list);
83 assert(!j->object_list);
84
85 if (j->in_run_queue)
86 LIST_REMOVE(run_queue, j->manager->run_queue, j);
87
88 if (j->in_dbus_queue)
89 LIST_REMOVE(dbus_queue, j->manager->dbus_job_queue, j);
90
91 sd_event_source_unref(j->timer_event_source);
92
93 bus_client_track_free(j->subscribed);
94
95 free(j);
96 }
97
98 void job_uninstall(Job *j) {
99 Job **pj;
100
101 assert(j->installed);
102
103 pj = (j->type == JOB_NOP) ? &j->unit->nop_job : &j->unit->job;
104 assert(*pj == j);
105
106 /* Detach from next 'bigger' objects */
107
108 /* daemon-reload should be transparent to job observers */
109 if (j->manager->n_reloading <= 0)
110 bus_job_send_removed_signal(j);
111
112 *pj = NULL;
113
114 unit_add_to_gc_queue(j->unit);
115
116 hashmap_remove(j->manager->jobs, UINT32_TO_PTR(j->id));
117 j->installed = false;
118 }
119
120 static bool job_type_allows_late_merge(JobType t) {
121 /* Tells whether it is OK to merge a job of type 't' with an already
122 * running job.
123 * Reloads cannot be merged this way. Think of the sequence:
124 * 1. Reload of a daemon is in progress; the daemon has already loaded
125 * its config file, but hasn't completed the reload operation yet.
126 * 2. Edit foo's config file.
127 * 3. Trigger another reload to have the daemon use the new config.
128 * Should the second reload job be merged into the first one, the daemon
129 * would not know about the new config.
130 * JOB_RESTART jobs on the other hand can be merged, because they get
131 * patched into JOB_START after stopping the unit. So if we see a
132 * JOB_RESTART running, it means the unit hasn't stopped yet and at
133 * this time the merge is still allowed. */
134 return t != JOB_RELOAD;
135 }
136
137 static void job_merge_into_installed(Job *j, Job *other) {
138 assert(j->installed);
139 assert(j->unit == other->unit);
140
141 if (j->type != JOB_NOP)
142 job_type_merge_and_collapse(&j->type, other->type, j->unit);
143 else
144 assert(other->type == JOB_NOP);
145
146 j->override = j->override || other->override;
147 j->irreversible = j->irreversible || other->irreversible;
148 j->ignore_order = j->ignore_order || other->ignore_order;
149 }
150
151 Job* job_install(Job *j) {
152 Job **pj;
153 Job *uj;
154
155 assert(!j->installed);
156 assert(j->type < _JOB_TYPE_MAX_IN_TRANSACTION);
157
158 pj = (j->type == JOB_NOP) ? &j->unit->nop_job : &j->unit->job;
159 uj = *pj;
160
161 if (uj) {
162 if (j->type != JOB_NOP && job_type_is_conflicting(uj->type, j->type))
163 job_finish_and_invalidate(uj, JOB_CANCELED, false);
164 else {
165 /* not conflicting, i.e. mergeable */
166
167 if (j->type == JOB_NOP || uj->state == JOB_WAITING ||
168 (job_type_allows_late_merge(j->type) && job_type_is_superset(uj->type, j->type))) {
169 job_merge_into_installed(uj, j);
170 log_debug_unit(uj->unit->id,
171 "Merged into installed job %s/%s as %u",
172 uj->unit->id, job_type_to_string(uj->type), (unsigned) uj->id);
173 return uj;
174 } else {
175 /* already running and not safe to merge into */
176 /* Patch uj to become a merged job and re-run it. */
177 /* XXX It should be safer to queue j to run after uj finishes, but it is
178 * not currently possible to have more than one installed job per unit. */
179 job_merge_into_installed(uj, j);
180 log_debug_unit(uj->unit->id,
181 "Merged into running job, re-running: %s/%s as %u",
182 uj->unit->id, job_type_to_string(uj->type), (unsigned) uj->id);
183 uj->state = JOB_WAITING;
184 uj->manager->n_running_jobs--;
185 return uj;
186 }
187 }
188 }
189
190 /* Install the job */
191 *pj = j;
192 j->installed = true;
193 j->manager->n_installed_jobs ++;
194 log_debug_unit(j->unit->id,
195 "Installed new job %s/%s as %u",
196 j->unit->id, job_type_to_string(j->type), (unsigned) j->id);
197 return j;
198 }
199
200 int job_install_deserialized(Job *j) {
201 Job **pj;
202
203 assert(!j->installed);
204
205 if (j->type < 0 || j->type >= _JOB_TYPE_MAX_IN_TRANSACTION) {
206 log_debug("Invalid job type %s in deserialization.", strna(job_type_to_string(j->type)));
207 return -EINVAL;
208 }
209
210 pj = (j->type == JOB_NOP) ? &j->unit->nop_job : &j->unit->job;
211
212 if (*pj) {
213 log_debug_unit(j->unit->id,
214 "Unit %s already has a job installed. Not installing deserialized job.",
215 j->unit->id);
216 return -EEXIST;
217 }
218 *pj = j;
219 j->installed = true;
220 log_debug_unit(j->unit->id,
221 "Reinstalled deserialized job %s/%s as %u",
222 j->unit->id, job_type_to_string(j->type), (unsigned) j->id);
223 return 0;
224 }
225
226 JobDependency* job_dependency_new(Job *subject, Job *object, bool matters, bool conflicts) {
227 JobDependency *l;
228
229 assert(object);
230
231 /* Adds a new job link, which encodes that the 'subject' job
232 * needs the 'object' job in some way. If 'subject' is NULL
233 * this means the 'anchor' job (i.e. the one the user
234 * explicitly asked for) is the requester. */
235
236 if (!(l = new0(JobDependency, 1)))
237 return NULL;
238
239 l->subject = subject;
240 l->object = object;
241 l->matters = matters;
242 l->conflicts = conflicts;
243
244 if (subject)
245 LIST_PREPEND(subject, subject->subject_list, l);
246
247 LIST_PREPEND(object, object->object_list, l);
248
249 return l;
250 }
251
252 void job_dependency_free(JobDependency *l) {
253 assert(l);
254
255 if (l->subject)
256 LIST_REMOVE(subject, l->subject->subject_list, l);
257
258 LIST_REMOVE(object, l->object->object_list, l);
259
260 free(l);
261 }
262
263 void job_dump(Job *j, FILE*f, const char *prefix) {
264 assert(j);
265 assert(f);
266
267 if (!prefix)
268 prefix = "";
269
270 fprintf(f,
271 "%s-> Job %u:\n"
272 "%s\tAction: %s -> %s\n"
273 "%s\tState: %s\n"
274 "%s\tForced: %s\n"
275 "%s\tIrreversible: %s\n",
276 prefix, j->id,
277 prefix, j->unit->id, job_type_to_string(j->type),
278 prefix, job_state_to_string(j->state),
279 prefix, yes_no(j->override),
280 prefix, yes_no(j->irreversible));
281 }
282
283 /*
284 * Merging is commutative, so imagine the matrix as symmetric. We store only
285 * its lower triangle to avoid duplication. We don't store the main diagonal,
286 * because A merged with A is simply A.
287 *
288 * If the resulting type is collapsed immediately afterwards (to get rid of
289 * the JOB_RELOAD_OR_START, which lies outside the lookup function's domain),
290 * the following properties hold:
291 *
292 * Merging is associative! A merged with B merged with C is the same as
293 * A merged with C merged with B.
294 *
295 * Mergeability is transitive! If A can be merged with B and B with C then
296 * A also with C.
297 *
298 * Also, if A merged with B cannot be merged with C, then either A or B cannot
299 * be merged with C either.
300 */
301 static const JobType job_merging_table[] = {
302 /* What \ With * JOB_START JOB_VERIFY_ACTIVE JOB_STOP JOB_RELOAD */
303 /*********************************************************************************/
304 /*JOB_START */
305 /*JOB_VERIFY_ACTIVE */ JOB_START,
306 /*JOB_STOP */ -1, -1,
307 /*JOB_RELOAD */ JOB_RELOAD_OR_START, JOB_RELOAD, -1,
308 /*JOB_RESTART */ JOB_RESTART, JOB_RESTART, -1, JOB_RESTART,
309 };
310
311 JobType job_type_lookup_merge(JobType a, JobType b) {
312 assert_cc(ELEMENTSOF(job_merging_table) == _JOB_TYPE_MAX_MERGING * (_JOB_TYPE_MAX_MERGING - 1) / 2);
313 assert(a >= 0 && a < _JOB_TYPE_MAX_MERGING);
314 assert(b >= 0 && b < _JOB_TYPE_MAX_MERGING);
315
316 if (a == b)
317 return a;
318
319 if (a < b) {
320 JobType tmp = a;
321 a = b;
322 b = tmp;
323 }
324
325 return job_merging_table[(a - 1) * a / 2 + b];
326 }
327
328 bool job_type_is_redundant(JobType a, UnitActiveState b) {
329 switch (a) {
330
331 case JOB_START:
332 return
333 b == UNIT_ACTIVE ||
334 b == UNIT_RELOADING;
335
336 case JOB_STOP:
337 return
338 b == UNIT_INACTIVE ||
339 b == UNIT_FAILED;
340
341 case JOB_VERIFY_ACTIVE:
342 return
343 b == UNIT_ACTIVE ||
344 b == UNIT_RELOADING;
345
346 case JOB_RELOAD:
347 return
348 b == UNIT_RELOADING;
349
350 case JOB_RESTART:
351 return
352 b == UNIT_ACTIVATING;
353
354 default:
355 assert_not_reached("Invalid job type");
356 }
357 }
358
359 void job_type_collapse(JobType *t, Unit *u) {
360 UnitActiveState s;
361
362 switch (*t) {
363
364 case JOB_TRY_RESTART:
365 s = unit_active_state(u);
366 if (UNIT_IS_INACTIVE_OR_DEACTIVATING(s))
367 *t = JOB_NOP;
368 else
369 *t = JOB_RESTART;
370 break;
371
372 case JOB_RELOAD_OR_START:
373 s = unit_active_state(u);
374 if (UNIT_IS_INACTIVE_OR_DEACTIVATING(s))
375 *t = JOB_START;
376 else
377 *t = JOB_RELOAD;
378 break;
379
380 default:
381 ;
382 }
383 }
384
385 int job_type_merge_and_collapse(JobType *a, JobType b, Unit *u) {
386 JobType t = job_type_lookup_merge(*a, b);
387 if (t < 0)
388 return -EEXIST;
389 *a = t;
390 job_type_collapse(a, u);
391 return 0;
392 }
393
394 static bool job_is_runnable(Job *j) {
395 Iterator i;
396 Unit *other;
397
398 assert(j);
399 assert(j->installed);
400
401 /* Checks whether there is any job running for the units this
402 * job needs to be running after (in the case of a 'positive'
403 * job type) or before (in the case of a 'negative' job
404 * type. */
405
406 /* Note that unit types have a say in what is runnable,
407 * too. For example, if they return -EAGAIN from
408 * unit_start() they can indicate they are not
409 * runnable yet. */
410
411 /* First check if there is an override */
412 if (j->ignore_order)
413 return true;
414
415 if (j->type == JOB_NOP)
416 return true;
417
418 if (j->type == JOB_START ||
419 j->type == JOB_VERIFY_ACTIVE ||
420 j->type == JOB_RELOAD) {
421
422 /* Immediate result is that the job is or might be
423 * started. In this case lets wait for the
424 * dependencies, regardless whether they are
425 * starting or stopping something. */
426
427 SET_FOREACH(other, j->unit->dependencies[UNIT_AFTER], i)
428 if (other->job)
429 return false;
430 }
431
432 /* Also, if something else is being stopped and we should
433 * change state after it, then lets wait. */
434
435 SET_FOREACH(other, j->unit->dependencies[UNIT_BEFORE], i)
436 if (other->job &&
437 (other->job->type == JOB_STOP ||
438 other->job->type == JOB_RESTART))
439 return false;
440
441 /* This means that for a service a and a service b where b
442 * shall be started after a:
443 *
444 * start a + start b → 1st step start a, 2nd step start b
445 * start a + stop b → 1st step stop b, 2nd step start a
446 * stop a + start b → 1st step stop a, 2nd step start b
447 * stop a + stop b → 1st step stop b, 2nd step stop a
448 *
449 * This has the side effect that restarts are properly
450 * synchronized too. */
451
452 return true;
453 }
454
455 static void job_change_type(Job *j, JobType newtype) {
456 log_debug_unit(j->unit->id,
457 "Converting job %s/%s -> %s/%s",
458 j->unit->id, job_type_to_string(j->type),
459 j->unit->id, job_type_to_string(newtype));
460
461 j->type = newtype;
462 }
463
464 int job_run_and_invalidate(Job *j) {
465 int r;
466 uint32_t id;
467 Manager *m = j->manager;
468
469 assert(j);
470 assert(j->installed);
471 assert(j->type < _JOB_TYPE_MAX_IN_TRANSACTION);
472 assert(j->in_run_queue);
473
474 LIST_REMOVE(run_queue, j->manager->run_queue, j);
475 j->in_run_queue = false;
476
477 if (j->state != JOB_WAITING)
478 return 0;
479
480 if (!job_is_runnable(j))
481 return -EAGAIN;
482
483 j->state = JOB_RUNNING;
484 m->n_running_jobs++;
485 job_add_to_dbus_queue(j);
486
487 /* While we execute this operation the job might go away (for
488 * example: because it is replaced by a new, conflicting
489 * job.) To make sure we don't access a freed job later on we
490 * store the id here, so that we can verify the job is still
491 * valid. */
492 id = j->id;
493
494 switch (j->type) {
495
496 case JOB_START:
497 r = unit_start(j->unit);
498
499 /* If this unit cannot be started, then simply wait */
500 if (r == -EBADR)
501 r = 0;
502 break;
503
504 case JOB_VERIFY_ACTIVE: {
505 UnitActiveState t = unit_active_state(j->unit);
506 if (UNIT_IS_ACTIVE_OR_RELOADING(t))
507 r = -EALREADY;
508 else if (t == UNIT_ACTIVATING)
509 r = -EAGAIN;
510 else
511 r = -EBADR;
512 break;
513 }
514
515 case JOB_STOP:
516 case JOB_RESTART:
517 r = unit_stop(j->unit);
518
519 /* If this unit cannot stopped, then simply wait. */
520 if (r == -EBADR)
521 r = 0;
522 break;
523
524 case JOB_RELOAD:
525 r = unit_reload(j->unit);
526 break;
527
528 case JOB_NOP:
529 r = -EALREADY;
530 break;
531
532 default:
533 assert_not_reached("Unknown job type");
534 }
535
536 j = manager_get_job(m, id);
537 if (j) {
538 if (r == -EALREADY)
539 r = job_finish_and_invalidate(j, JOB_DONE, true);
540 else if (r == -EBADR)
541 r = job_finish_and_invalidate(j, JOB_SKIPPED, true);
542 else if (r == -ENOEXEC)
543 r = job_finish_and_invalidate(j, JOB_INVALID, true);
544 else if (r == -EAGAIN) {
545 j->state = JOB_WAITING;
546 m->n_running_jobs--;
547 } else if (r < 0)
548 r = job_finish_and_invalidate(j, JOB_FAILED, true);
549 }
550
551 return r;
552 }
553
554 _pure_ static const char *job_get_status_message_format(Unit *u, JobType t, JobResult result) {
555 const UnitStatusMessageFormats *format_table;
556
557 assert(u);
558 assert(t >= 0);
559 assert(t < _JOB_TYPE_MAX);
560
561 format_table = &UNIT_VTABLE(u)->status_message_formats;
562 if (!format_table)
563 return NULL;
564
565 if (t == JOB_START)
566 return format_table->finished_start_job[result];
567 else if (t == JOB_STOP || t == JOB_RESTART)
568 return format_table->finished_stop_job[result];
569
570 return NULL;
571 }
572
573 _pure_ static const char *job_get_status_message_format_try_harder(Unit *u, JobType t, JobResult result) {
574 const char *format;
575
576 assert(u);
577 assert(t >= 0);
578 assert(t < _JOB_TYPE_MAX);
579
580 format = job_get_status_message_format(u, t, result);
581 if (format)
582 return format;
583
584 /* Return generic strings */
585 if (t == JOB_START) {
586 if (result == JOB_DONE)
587 return "Started %s.";
588 else if (result == JOB_FAILED)
589 return "Failed to start %s.";
590 else if (result == JOB_DEPENDENCY)
591 return "Dependency failed for %s.";
592 else if (result == JOB_TIMEOUT)
593 return "Timed out starting %s.";
594 } else if (t == JOB_STOP || t == JOB_RESTART) {
595 if (result == JOB_DONE)
596 return "Stopped %s.";
597 else if (result == JOB_FAILED)
598 return "Stopped (with error) %s.";
599 else if (result == JOB_TIMEOUT)
600 return "Timed out stoppping %s.";
601 } else if (t == JOB_RELOAD) {
602 if (result == JOB_DONE)
603 return "Reloaded %s.";
604 else if (result == JOB_FAILED)
605 return "Reload failed for %s.";
606 else if (result == JOB_TIMEOUT)
607 return "Timed out reloading %s.";
608 }
609
610 return NULL;
611 }
612
613 #pragma GCC diagnostic push
614 #pragma GCC diagnostic ignored "-Wformat-nonliteral"
615 static void job_print_status_message(Unit *u, JobType t, JobResult result) {
616 const char *format;
617
618 assert(u);
619 assert(t >= 0);
620 assert(t < _JOB_TYPE_MAX);
621
622 if (t == JOB_START) {
623 format = job_get_status_message_format(u, t, result);
624 if (!format)
625 return;
626
627 switch (result) {
628
629 case JOB_DONE:
630 if (u->condition_result)
631 unit_status_printf(u, ANSI_GREEN_ON " OK " ANSI_HIGHLIGHT_OFF, format);
632 break;
633
634 case JOB_FAILED:
635 unit_status_printf(u, ANSI_HIGHLIGHT_RED_ON "FAILED" ANSI_HIGHLIGHT_OFF, format);
636 manager_status_printf(u->manager, false, NULL, "See 'systemctl status %s' for details.", u->id);
637 break;
638
639 case JOB_DEPENDENCY:
640 unit_status_printf(u, ANSI_HIGHLIGHT_YELLOW_ON "DEPEND" ANSI_HIGHLIGHT_OFF, format);
641 break;
642
643 case JOB_TIMEOUT:
644 unit_status_printf(u, ANSI_HIGHLIGHT_RED_ON " TIME " ANSI_HIGHLIGHT_OFF, format);
645 break;
646
647 default:
648 ;
649 }
650
651 } else if (t == JOB_STOP || t == JOB_RESTART) {
652
653 format = job_get_status_message_format(u, t, result);
654 if (!format)
655 return;
656
657 switch (result) {
658
659 case JOB_TIMEOUT:
660 unit_status_printf(u, ANSI_HIGHLIGHT_RED_ON " TIME " ANSI_HIGHLIGHT_OFF, format);
661 break;
662
663 case JOB_DONE:
664 case JOB_FAILED:
665 unit_status_printf(u, ANSI_GREEN_ON " OK " ANSI_HIGHLIGHT_OFF, format);
666 break;
667
668 default:
669 ;
670 }
671
672 } else if (t == JOB_VERIFY_ACTIVE) {
673
674 /* When verify-active detects the unit is inactive, report it.
675 * Most likely a DEPEND warning from a requisiting unit will
676 * occur next and it's nice to see what was requisited. */
677 if (result == JOB_SKIPPED)
678 unit_status_printf(u, ANSI_HIGHLIGHT_ON " INFO " ANSI_HIGHLIGHT_OFF, "%s is not active.");
679 }
680 }
681 #pragma GCC diagnostic pop
682
683 #pragma GCC diagnostic push
684 #pragma GCC diagnostic ignored "-Wformat-nonliteral"
685 static void job_log_status_message(Unit *u, JobType t, JobResult result) {
686 const char *format;
687 char buf[LINE_MAX];
688
689 assert(u);
690 assert(t >= 0);
691 assert(t < _JOB_TYPE_MAX);
692
693 /* Skip this if it goes to the console. since we already print
694 * to the console anyway... */
695
696 if (log_on_console())
697 return;
698
699 format = job_get_status_message_format_try_harder(u, t, result);
700 if (!format)
701 return;
702
703 snprintf(buf, sizeof(buf), format, unit_description(u));
704 char_array_0(buf);
705
706 if (t == JOB_START) {
707 sd_id128_t mid;
708
709 mid = result == JOB_DONE ? SD_MESSAGE_UNIT_STARTED : SD_MESSAGE_UNIT_FAILED;
710 log_struct_unit(result == JOB_DONE ? LOG_INFO : LOG_ERR,
711 u->id,
712 MESSAGE_ID(mid),
713 "RESULT=%s", job_result_to_string(result),
714 "MESSAGE=%s", buf,
715 NULL);
716
717 } else if (t == JOB_STOP)
718 log_struct_unit(result == JOB_DONE ? LOG_INFO : LOG_ERR,
719 u->id,
720 MESSAGE_ID(SD_MESSAGE_UNIT_STOPPED),
721 "RESULT=%s", job_result_to_string(result),
722 "MESSAGE=%s", buf,
723 NULL);
724
725 else if (t == JOB_RELOAD)
726 log_struct_unit(result == JOB_DONE ? LOG_INFO : LOG_ERR,
727 u->id,
728 MESSAGE_ID(SD_MESSAGE_UNIT_RELOADED),
729 "RESULT=%s", job_result_to_string(result),
730 "MESSAGE=%s", buf,
731 NULL);
732 }
733 #pragma GCC diagnostic pop
734
735 int job_finish_and_invalidate(Job *j, JobResult result, bool recursive) {
736 Unit *u;
737 Unit *other;
738 JobType t;
739 Iterator i;
740
741 assert(j);
742 assert(j->installed);
743 assert(j->type < _JOB_TYPE_MAX_IN_TRANSACTION);
744
745 u = j->unit;
746 t = j->type;
747
748 j->result = result;
749
750 if (j->state == JOB_RUNNING)
751 j->manager->n_running_jobs--;
752
753 log_debug_unit(u->id, "Job %s/%s finished, result=%s",
754 u->id, job_type_to_string(t), job_result_to_string(result));
755
756 job_print_status_message(u, t, result);
757 job_log_status_message(u, t, result);
758
759 job_add_to_dbus_queue(j);
760
761 /* Patch restart jobs so that they become normal start jobs */
762 if (result == JOB_DONE && t == JOB_RESTART) {
763
764 job_change_type(j, JOB_START);
765 j->state = JOB_WAITING;
766
767 job_add_to_run_queue(j);
768
769 goto finish;
770 }
771
772 if (result == JOB_FAILED || result == JOB_INVALID)
773 j->manager->n_failed_jobs ++;
774
775 job_uninstall(j);
776 job_free(j);
777
778 /* Fail depending jobs on failure */
779 if (result != JOB_DONE && recursive) {
780
781 if (t == JOB_START ||
782 t == JOB_VERIFY_ACTIVE) {
783
784 SET_FOREACH(other, u->dependencies[UNIT_REQUIRED_BY], i)
785 if (other->job &&
786 (other->job->type == JOB_START ||
787 other->job->type == JOB_VERIFY_ACTIVE))
788 job_finish_and_invalidate(other->job, JOB_DEPENDENCY, true);
789
790 SET_FOREACH(other, u->dependencies[UNIT_BOUND_BY], i)
791 if (other->job &&
792 (other->job->type == JOB_START ||
793 other->job->type == JOB_VERIFY_ACTIVE))
794 job_finish_and_invalidate(other->job, JOB_DEPENDENCY, true);
795
796 SET_FOREACH(other, u->dependencies[UNIT_REQUIRED_BY_OVERRIDABLE], i)
797 if (other->job &&
798 !other->job->override &&
799 (other->job->type == JOB_START ||
800 other->job->type == JOB_VERIFY_ACTIVE))
801 job_finish_and_invalidate(other->job, JOB_DEPENDENCY, true);
802
803 } else if (t == JOB_STOP) {
804
805 SET_FOREACH(other, u->dependencies[UNIT_CONFLICTED_BY], i)
806 if (other->job &&
807 (other->job->type == JOB_START ||
808 other->job->type == JOB_VERIFY_ACTIVE))
809 job_finish_and_invalidate(other->job, JOB_DEPENDENCY, true);
810 }
811 }
812
813 /* Trigger OnFailure dependencies that are not generated by
814 * the unit itself. We don't treat JOB_CANCELED as failure in
815 * this context. And JOB_FAILURE is already handled by the
816 * unit itself. */
817 if (result == JOB_TIMEOUT || result == JOB_DEPENDENCY) {
818 log_struct_unit(LOG_NOTICE,
819 u->id,
820 "JOB_TYPE=%s", job_type_to_string(t),
821 "JOB_RESULT=%s", job_result_to_string(result),
822 "Job %s/%s failed with result '%s'.",
823 u->id,
824 job_type_to_string(t),
825 job_result_to_string(result),
826 NULL);
827
828 unit_start_on_failure(u);
829 }
830
831 unit_trigger_notify(u);
832
833 finish:
834 /* Try to start the next jobs that can be started */
835 SET_FOREACH(other, u->dependencies[UNIT_AFTER], i)
836 if (other->job)
837 job_add_to_run_queue(other->job);
838 SET_FOREACH(other, u->dependencies[UNIT_BEFORE], i)
839 if (other->job)
840 job_add_to_run_queue(other->job);
841
842 manager_check_finished(u->manager);
843
844 return 0;
845 }
846
847 static int job_dispatch_timer(sd_event_source *s, uint64_t monotonic, void *userdata) {
848 Job *j = userdata;
849
850 assert(j);
851 assert(s == j->timer_event_source);
852
853 log_warning_unit(j->unit->id, "Job %s/%s timed out.",
854 j->unit->id, job_type_to_string(j->type));
855
856 job_finish_and_invalidate(j, JOB_TIMEOUT, true);
857 return 0;
858 }
859
860 int job_start_timer(Job *j) {
861 int r;
862
863 if (j->unit->job_timeout <= 0 || j->timer_event_source)
864 return 0;
865
866 j->begin_usec = now(CLOCK_MONOTONIC);
867
868 r = sd_event_add_monotonic(j->manager->event, j->begin_usec + j->unit->job_timeout, 0, job_dispatch_timer, j, &j->timer_event_source);
869 if (r < 0)
870 return r;
871
872 return 0;
873 }
874
875 void job_add_to_run_queue(Job *j) {
876 assert(j);
877 assert(j->installed);
878
879 if (j->in_run_queue)
880 return;
881
882 if (!j->manager->run_queue)
883 sd_event_source_set_enabled(j->manager->run_queue_event_source, SD_EVENT_ONESHOT);
884
885 LIST_PREPEND(run_queue, j->manager->run_queue, j);
886 j->in_run_queue = true;
887 }
888
889 void job_add_to_dbus_queue(Job *j) {
890 assert(j);
891 assert(j->installed);
892
893 if (j->in_dbus_queue)
894 return;
895
896 /* We don't check if anybody is subscribed here, since this
897 * job might just have been created and not yet assigned to a
898 * connection/client. */
899
900 LIST_PREPEND(dbus_queue, j->manager->dbus_job_queue, j);
901 j->in_dbus_queue = true;
902 }
903
904 char *job_dbus_path(Job *j) {
905 char *p;
906
907 assert(j);
908
909 if (asprintf(&p, "/org/freedesktop/systemd1/job/%"PRIu32, j->id) < 0)
910 return NULL;
911
912 return p;
913 }
914
915 int job_serialize(Job *j, FILE *f, FDSet *fds) {
916 fprintf(f, "job-id=%u\n", j->id);
917 fprintf(f, "job-type=%s\n", job_type_to_string(j->type));
918 fprintf(f, "job-state=%s\n", job_state_to_string(j->state));
919 fprintf(f, "job-override=%s\n", yes_no(j->override));
920 fprintf(f, "job-irreversible=%s\n", yes_no(j->irreversible));
921 fprintf(f, "job-sent-dbus-new-signal=%s\n", yes_no(j->sent_dbus_new_signal));
922 fprintf(f, "job-ignore-order=%s\n", yes_no(j->ignore_order));
923
924 if (j->begin_usec > 0)
925 fprintf(f, "job-begin="USEC_FMT"\n", j->begin_usec);
926
927 bus_client_track_serialize(j->manager, f, j->subscribed);
928
929 /* End marker */
930 fputc('\n', f);
931 return 0;
932 }
933
934 int job_deserialize(Job *j, FILE *f, FDSet *fds) {
935 assert(j);
936
937 for (;;) {
938 char line[LINE_MAX], *l, *v;
939 size_t k;
940
941 if (!fgets(line, sizeof(line), f)) {
942 if (feof(f))
943 return 0;
944 return -errno;
945 }
946
947 char_array_0(line);
948 l = strstrip(line);
949
950 /* End marker */
951 if (l[0] == 0)
952 return 0;
953
954 k = strcspn(l, "=");
955
956 if (l[k] == '=') {
957 l[k] = 0;
958 v = l+k+1;
959 } else
960 v = l+k;
961
962 if (streq(l, "job-id")) {
963
964 if (safe_atou32(v, &j->id) < 0)
965 log_debug("Failed to parse job id value %s", v);
966
967 } else if (streq(l, "job-type")) {
968 JobType t;
969
970 t = job_type_from_string(v);
971 if (t < 0)
972 log_debug("Failed to parse job type %s", v);
973 else if (t >= _JOB_TYPE_MAX_IN_TRANSACTION)
974 log_debug("Cannot deserialize job of type %s", v);
975 else
976 j->type = t;
977
978 } else if (streq(l, "job-state")) {
979 JobState s;
980
981 s = job_state_from_string(v);
982 if (s < 0)
983 log_debug("Failed to parse job state %s", v);
984 else
985 j->state = s;
986
987 } else if (streq(l, "job-override")) {
988 int b;
989
990 b = parse_boolean(v);
991 if (b < 0)
992 log_debug("Failed to parse job override flag %s", v);
993 else
994 j->override = j->override || b;
995
996 } else if (streq(l, "job-irreversible")) {
997 int b;
998
999 b = parse_boolean(v);
1000 if (b < 0)
1001 log_debug("Failed to parse job irreversible flag %s", v);
1002 else
1003 j->irreversible = j->irreversible || b;
1004
1005 } else if (streq(l, "job-sent-dbus-new-signal")) {
1006 int b;
1007
1008 b = parse_boolean(v);
1009 if (b < 0)
1010 log_debug("Failed to parse job sent_dbus_new_signal flag %s", v);
1011 else
1012 j->sent_dbus_new_signal = j->sent_dbus_new_signal || b;
1013
1014 } else if (streq(l, "job-ignore-order")) {
1015 int b;
1016
1017 b = parse_boolean(v);
1018 if (b < 0)
1019 log_debug("Failed to parse job ignore_order flag %s", v);
1020 else
1021 j->ignore_order = j->ignore_order || b;
1022
1023 } else if (streq(l, "job-begin")) {
1024 unsigned long long ull;
1025
1026 if (sscanf(v, "%llu", &ull) != 1)
1027 log_debug("Failed to parse job-begin value %s", v);
1028 else
1029 j->begin_usec = ull;
1030
1031 } else {
1032 char t[strlen(l) + 1 + strlen(v) + 1];
1033
1034 strcpy(stpcpy(stpcpy(t, l), "="), v);
1035
1036 if (bus_client_track_deserialize_item(j->manager, &j->subscribed, t) == 0)
1037 log_debug("Unknown deserialization key '%s'", l);
1038 }
1039 }
1040 }
1041
1042 int job_coldplug(Job *j) {
1043 int r;
1044
1045 assert(j);
1046
1047 if (j->begin_usec <= 0)
1048 return 0;
1049
1050 if (j->timer_event_source)
1051 j->timer_event_source = sd_event_source_unref(j->timer_event_source);
1052
1053 r = sd_event_add_monotonic(j->manager->event, j->begin_usec + j->unit->job_timeout, 0, job_dispatch_timer, j, &j->timer_event_source);
1054 if (r < 0)
1055 log_debug("Failed to restart timeout for job: %s", strerror(-r));
1056
1057 return r;
1058 }
1059
1060 void job_shutdown_magic(Job *j) {
1061 assert(j);
1062
1063 /* The shutdown target gets some special treatment here: we
1064 * tell the kernel to begin with flushing its disk caches, to
1065 * optimize shutdown time a bit. Ideally we wouldn't hardcode
1066 * this magic into PID 1. However all other processes aren't
1067 * options either since they'd exit much sooner than PID 1 and
1068 * asynchronous sync() would cause their exit to be
1069 * delayed. */
1070
1071 if (j->type != JOB_START)
1072 return;
1073
1074 if (j->unit->manager->running_as != SYSTEMD_SYSTEM)
1075 return;
1076
1077 if (!unit_has_name(j->unit, SPECIAL_SHUTDOWN_TARGET))
1078 return;
1079
1080 /* In case messages on console has been disabled on boot */
1081 j->unit->manager->no_console_output = false;
1082
1083 if (detect_container(NULL) > 0)
1084 return;
1085
1086 asynchronous_sync();
1087 }
1088
1089 int job_get_timeout(Job *j, uint64_t *timeout) {
1090 Unit *u = j->unit;
1091 uint64_t x = -1, y = -1;
1092 int r = 0, q = 0;
1093
1094 assert(u);
1095
1096 if (j->timer_event_source) {
1097 r = sd_event_source_get_time(j->timer_event_source, &x);
1098 if (r < 0)
1099 return r;
1100 r = 1;
1101 }
1102
1103 if (UNIT_VTABLE(u)->get_timeout) {
1104 q = UNIT_VTABLE(u)->get_timeout(u, &y);
1105 if (q < 0)
1106 return q;
1107 }
1108
1109 if (r == 0 && q == 0)
1110 return 0;
1111
1112 *timeout = MIN(x, y);
1113
1114 log_info("job_get_timeout %s %d/%"PRIu64" %d/%"PRIu64" -> 1/%"PRIu64,
1115 j->unit->id, r, x, q, y, *timeout);
1116
1117 return 1;
1118 }
1119
1120 static const char* const job_state_table[_JOB_STATE_MAX] = {
1121 [JOB_WAITING] = "waiting",
1122 [JOB_RUNNING] = "running"
1123 };
1124
1125 DEFINE_STRING_TABLE_LOOKUP(job_state, JobState);
1126
1127 static const char* const job_type_table[_JOB_TYPE_MAX] = {
1128 [JOB_START] = "start",
1129 [JOB_VERIFY_ACTIVE] = "verify-active",
1130 [JOB_STOP] = "stop",
1131 [JOB_RELOAD] = "reload",
1132 [JOB_RELOAD_OR_START] = "reload-or-start",
1133 [JOB_RESTART] = "restart",
1134 [JOB_TRY_RESTART] = "try-restart",
1135 [JOB_NOP] = "nop",
1136 };
1137
1138 DEFINE_STRING_TABLE_LOOKUP(job_type, JobType);
1139
1140 static const char* const job_mode_table[_JOB_MODE_MAX] = {
1141 [JOB_FAIL] = "fail",
1142 [JOB_REPLACE] = "replace",
1143 [JOB_REPLACE_IRREVERSIBLY] = "replace-irreversibly",
1144 [JOB_ISOLATE] = "isolate",
1145 [JOB_IGNORE_DEPENDENCIES] = "ignore-dependencies",
1146 [JOB_IGNORE_REQUIREMENTS] = "ignore-requirements",
1147 [JOB_FLUSH] = "flush",
1148 };
1149
1150 DEFINE_STRING_TABLE_LOOKUP(job_mode, JobMode);
1151
1152 static const char* const job_result_table[_JOB_RESULT_MAX] = {
1153 [JOB_DONE] = "done",
1154 [JOB_CANCELED] = "canceled",
1155 [JOB_TIMEOUT] = "timeout",
1156 [JOB_FAILED] = "failed",
1157 [JOB_DEPENDENCY] = "dependency",
1158 [JOB_SKIPPED] = "skipped",
1159 [JOB_INVALID] = "invalid",
1160 };
1161
1162 DEFINE_STRING_TABLE_LOOKUP(job_result, JobResult);