]>
Commit | Line | Data |
---|---|---|
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 "systemd/sd-id128.h" | |
28 | #include "systemd/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 "sync.h" | |
39 | #include "virt.h" | |
40 | ||
41 | JobBusClient* job_bus_client_new(DBusConnection *connection, const char *name) { | |
42 | JobBusClient *cl; | |
43 | size_t name_len; | |
44 | ||
45 | name_len = strlen(name); | |
46 | cl = malloc0(sizeof(JobBusClient) + name_len + 1); | |
47 | if (!cl) | |
48 | return NULL; | |
49 | ||
50 | cl->bus = connection; | |
51 | memcpy(cl->name, name, name_len + 1); | |
52 | return cl; | |
53 | } | |
54 | ||
55 | Job* job_new_raw(Unit *unit) { | |
56 | Job *j; | |
57 | ||
58 | /* used for deserialization */ | |
59 | ||
60 | assert(unit); | |
61 | ||
62 | j = new0(Job, 1); | |
63 | if (!j) | |
64 | return NULL; | |
65 | ||
66 | j->manager = unit->manager; | |
67 | j->unit = unit; | |
68 | j->type = _JOB_TYPE_INVALID; | |
69 | j->timer_watch.type = WATCH_INVALID; | |
70 | ||
71 | return j; | |
72 | } | |
73 | ||
74 | Job* job_new(Unit *unit, JobType type) { | |
75 | Job *j; | |
76 | ||
77 | assert(type < _JOB_TYPE_MAX); | |
78 | ||
79 | j = job_new_raw(unit); | |
80 | if (!j) | |
81 | return NULL; | |
82 | ||
83 | j->id = j->manager->current_job_id++; | |
84 | j->type = type; | |
85 | ||
86 | /* We don't link it here, that's what job_dependency() is for */ | |
87 | ||
88 | return j; | |
89 | } | |
90 | ||
91 | void job_free(Job *j) { | |
92 | JobBusClient *cl; | |
93 | ||
94 | assert(j); | |
95 | assert(!j->installed); | |
96 | assert(!j->transaction_prev); | |
97 | assert(!j->transaction_next); | |
98 | assert(!j->subject_list); | |
99 | assert(!j->object_list); | |
100 | ||
101 | if (j->in_run_queue) | |
102 | LIST_REMOVE(Job, run_queue, j->manager->run_queue, j); | |
103 | ||
104 | if (j->in_dbus_queue) | |
105 | LIST_REMOVE(Job, dbus_queue, j->manager->dbus_job_queue, j); | |
106 | ||
107 | if (j->timer_watch.type != WATCH_INVALID) { | |
108 | assert(j->timer_watch.type == WATCH_JOB_TIMER); | |
109 | assert(j->timer_watch.data.job == j); | |
110 | assert(j->timer_watch.fd >= 0); | |
111 | ||
112 | assert_se(epoll_ctl(j->manager->epoll_fd, EPOLL_CTL_DEL, j->timer_watch.fd, NULL) >= 0); | |
113 | close_nointr_nofail(j->timer_watch.fd); | |
114 | } | |
115 | ||
116 | while ((cl = j->bus_client_list)) { | |
117 | LIST_REMOVE(JobBusClient, client, j->bus_client_list, cl); | |
118 | free(cl); | |
119 | } | |
120 | free(j); | |
121 | } | |
122 | ||
123 | void job_uninstall(Job *j) { | |
124 | Job **pj; | |
125 | ||
126 | assert(j->installed); | |
127 | ||
128 | pj = (j->type == JOB_NOP) ? &j->unit->nop_job : &j->unit->job; | |
129 | assert(*pj == j); | |
130 | ||
131 | /* Detach from next 'bigger' objects */ | |
132 | ||
133 | /* daemon-reload should be transparent to job observers */ | |
134 | if (j->manager->n_reloading <= 0) | |
135 | bus_job_send_removed_signal(j); | |
136 | ||
137 | *pj = NULL; | |
138 | ||
139 | unit_add_to_gc_queue(j->unit); | |
140 | ||
141 | hashmap_remove(j->manager->jobs, UINT32_TO_PTR(j->id)); | |
142 | j->installed = false; | |
143 | } | |
144 | ||
145 | static bool job_type_allows_late_merge(JobType t) { | |
146 | /* Tells whether it is OK to merge a job of type 't' with an already | |
147 | * running job. | |
148 | * Reloads cannot be merged this way. Think of the sequence: | |
149 | * 1. Reload of a daemon is in progress; the daemon has already loaded | |
150 | * its config file, but hasn't completed the reload operation yet. | |
151 | * 2. Edit foo's config file. | |
152 | * 3. Trigger another reload to have the daemon use the new config. | |
153 | * Should the second reload job be merged into the first one, the daemon | |
154 | * would not know about the new config. | |
155 | * JOB_RESTART jobs on the other hand can be merged, because they get | |
156 | * patched into JOB_START after stopping the unit. So if we see a | |
157 | * JOB_RESTART running, it means the unit hasn't stopped yet and at | |
158 | * this time the merge is still allowed. */ | |
159 | return t != JOB_RELOAD; | |
160 | } | |
161 | ||
162 | static void job_merge_into_installed(Job *j, Job *other) { | |
163 | assert(j->installed); | |
164 | assert(j->unit == other->unit); | |
165 | ||
166 | if (j->type != JOB_NOP) | |
167 | job_type_merge_and_collapse(&j->type, other->type, j->unit); | |
168 | else | |
169 | assert(other->type == JOB_NOP); | |
170 | ||
171 | j->override = j->override || other->override; | |
172 | j->irreversible = j->irreversible || other->irreversible; | |
173 | j->ignore_order = j->ignore_order || other->ignore_order; | |
174 | } | |
175 | ||
176 | Job* job_install(Job *j) { | |
177 | Job **pj; | |
178 | Job *uj; | |
179 | ||
180 | assert(!j->installed); | |
181 | assert(j->type < _JOB_TYPE_MAX_IN_TRANSACTION); | |
182 | ||
183 | pj = (j->type == JOB_NOP) ? &j->unit->nop_job : &j->unit->job; | |
184 | uj = *pj; | |
185 | ||
186 | if (uj) { | |
187 | if (j->type != JOB_NOP && job_type_is_conflicting(uj->type, j->type)) | |
188 | job_finish_and_invalidate(uj, JOB_CANCELED, false); | |
189 | else { | |
190 | /* not conflicting, i.e. mergeable */ | |
191 | ||
192 | if (j->type == JOB_NOP || uj->state == JOB_WAITING || | |
193 | (job_type_allows_late_merge(j->type) && job_type_is_superset(uj->type, j->type))) { | |
194 | job_merge_into_installed(uj, j); | |
195 | log_debug_unit(uj->unit->id, | |
196 | "Merged into installed job %s/%s as %u", | |
197 | uj->unit->id, job_type_to_string(uj->type), (unsigned) uj->id); | |
198 | return uj; | |
199 | } else { | |
200 | /* already running and not safe to merge into */ | |
201 | /* Patch uj to become a merged job and re-run it. */ | |
202 | /* XXX It should be safer to queue j to run after uj finishes, but it is | |
203 | * not currently possible to have more than one installed job per unit. */ | |
204 | job_merge_into_installed(uj, j); | |
205 | log_debug_unit(uj->unit->id, | |
206 | "Merged into running job, re-running: %s/%s as %u", | |
207 | uj->unit->id, job_type_to_string(uj->type), (unsigned) uj->id); | |
208 | uj->state = JOB_WAITING; | |
209 | uj->manager->n_running_jobs--; | |
210 | return uj; | |
211 | } | |
212 | } | |
213 | } | |
214 | ||
215 | /* Install the job */ | |
216 | *pj = j; | |
217 | j->installed = true; | |
218 | j->manager->n_installed_jobs ++; | |
219 | log_debug_unit(j->unit->id, | |
220 | "Installed new job %s/%s as %u", | |
221 | j->unit->id, job_type_to_string(j->type), (unsigned) j->id); | |
222 | return j; | |
223 | } | |
224 | ||
225 | int job_install_deserialized(Job *j) { | |
226 | Job **pj; | |
227 | ||
228 | assert(!j->installed); | |
229 | ||
230 | if (j->type < 0 || j->type >= _JOB_TYPE_MAX_IN_TRANSACTION) { | |
231 | log_debug("Invalid job type %s in deserialization.", strna(job_type_to_string(j->type))); | |
232 | return -EINVAL; | |
233 | } | |
234 | ||
235 | pj = (j->type == JOB_NOP) ? &j->unit->nop_job : &j->unit->job; | |
236 | ||
237 | if (*pj) { | |
238 | log_debug_unit(j->unit->id, | |
239 | "Unit %s already has a job installed. Not installing deserialized job.", | |
240 | j->unit->id); | |
241 | return -EEXIST; | |
242 | } | |
243 | *pj = j; | |
244 | j->installed = true; | |
245 | log_debug_unit(j->unit->id, | |
246 | "Reinstalled deserialized job %s/%s as %u", | |
247 | j->unit->id, job_type_to_string(j->type), (unsigned) j->id); | |
248 | return 0; | |
249 | } | |
250 | ||
251 | JobDependency* job_dependency_new(Job *subject, Job *object, bool matters, bool conflicts) { | |
252 | JobDependency *l; | |
253 | ||
254 | assert(object); | |
255 | ||
256 | /* Adds a new job link, which encodes that the 'subject' job | |
257 | * needs the 'object' job in some way. If 'subject' is NULL | |
258 | * this means the 'anchor' job (i.e. the one the user | |
259 | * explicitly asked for) is the requester. */ | |
260 | ||
261 | if (!(l = new0(JobDependency, 1))) | |
262 | return NULL; | |
263 | ||
264 | l->subject = subject; | |
265 | l->object = object; | |
266 | l->matters = matters; | |
267 | l->conflicts = conflicts; | |
268 | ||
269 | if (subject) | |
270 | LIST_PREPEND(JobDependency, subject, subject->subject_list, l); | |
271 | ||
272 | LIST_PREPEND(JobDependency, object, object->object_list, l); | |
273 | ||
274 | return l; | |
275 | } | |
276 | ||
277 | void job_dependency_free(JobDependency *l) { | |
278 | assert(l); | |
279 | ||
280 | if (l->subject) | |
281 | LIST_REMOVE(JobDependency, subject, l->subject->subject_list, l); | |
282 | ||
283 | LIST_REMOVE(JobDependency, object, l->object->object_list, l); | |
284 | ||
285 | free(l); | |
286 | } | |
287 | ||
288 | void job_dump(Job *j, FILE*f, const char *prefix) { | |
289 | assert(j); | |
290 | assert(f); | |
291 | ||
292 | if (!prefix) | |
293 | prefix = ""; | |
294 | ||
295 | fprintf(f, | |
296 | "%s-> Job %u:\n" | |
297 | "%s\tAction: %s -> %s\n" | |
298 | "%s\tState: %s\n" | |
299 | "%s\tForced: %s\n" | |
300 | "%s\tIrreversible: %s\n", | |
301 | prefix, j->id, | |
302 | prefix, j->unit->id, job_type_to_string(j->type), | |
303 | prefix, job_state_to_string(j->state), | |
304 | prefix, yes_no(j->override), | |
305 | prefix, yes_no(j->irreversible)); | |
306 | } | |
307 | ||
308 | /* | |
309 | * Merging is commutative, so imagine the matrix as symmetric. We store only | |
310 | * its lower triangle to avoid duplication. We don't store the main diagonal, | |
311 | * because A merged with A is simply A. | |
312 | * | |
313 | * If the resulting type is collapsed immediately afterwards (to get rid of | |
314 | * the JOB_RELOAD_OR_START, which lies outside the lookup function's domain), | |
315 | * the following properties hold: | |
316 | * | |
317 | * Merging is associative! A merged with B merged with C is the same as | |
318 | * A merged with C merged with B. | |
319 | * | |
320 | * Mergeability is transitive! If A can be merged with B and B with C then | |
321 | * A also with C. | |
322 | * | |
323 | * Also, if A merged with B cannot be merged with C, then either A or B cannot | |
324 | * be merged with C either. | |
325 | */ | |
326 | static const JobType job_merging_table[] = { | |
327 | /* What \ With * JOB_START JOB_VERIFY_ACTIVE JOB_STOP JOB_RELOAD */ | |
328 | /*********************************************************************************/ | |
329 | /*JOB_START */ | |
330 | /*JOB_VERIFY_ACTIVE */ JOB_START, | |
331 | /*JOB_STOP */ -1, -1, | |
332 | /*JOB_RELOAD */ JOB_RELOAD_OR_START, JOB_RELOAD, -1, | |
333 | /*JOB_RESTART */ JOB_RESTART, JOB_RESTART, -1, JOB_RESTART, | |
334 | }; | |
335 | ||
336 | JobType job_type_lookup_merge(JobType a, JobType b) { | |
337 | assert_cc(ELEMENTSOF(job_merging_table) == _JOB_TYPE_MAX_MERGING * (_JOB_TYPE_MAX_MERGING - 1) / 2); | |
338 | assert(a >= 0 && a < _JOB_TYPE_MAX_MERGING); | |
339 | assert(b >= 0 && b < _JOB_TYPE_MAX_MERGING); | |
340 | ||
341 | if (a == b) | |
342 | return a; | |
343 | ||
344 | if (a < b) { | |
345 | JobType tmp = a; | |
346 | a = b; | |
347 | b = tmp; | |
348 | } | |
349 | ||
350 | return job_merging_table[(a - 1) * a / 2 + b]; | |
351 | } | |
352 | ||
353 | bool job_type_is_redundant(JobType a, UnitActiveState b) { | |
354 | switch (a) { | |
355 | ||
356 | case JOB_START: | |
357 | return | |
358 | b == UNIT_ACTIVE || | |
359 | b == UNIT_RELOADING; | |
360 | ||
361 | case JOB_STOP: | |
362 | return | |
363 | b == UNIT_INACTIVE || | |
364 | b == UNIT_FAILED; | |
365 | ||
366 | case JOB_VERIFY_ACTIVE: | |
367 | return | |
368 | b == UNIT_ACTIVE || | |
369 | b == UNIT_RELOADING; | |
370 | ||
371 | case JOB_RELOAD: | |
372 | return | |
373 | b == UNIT_RELOADING; | |
374 | ||
375 | case JOB_RESTART: | |
376 | return | |
377 | b == UNIT_ACTIVATING; | |
378 | ||
379 | default: | |
380 | assert_not_reached("Invalid job type"); | |
381 | } | |
382 | } | |
383 | ||
384 | void job_type_collapse(JobType *t, Unit *u) { | |
385 | UnitActiveState s; | |
386 | ||
387 | switch (*t) { | |
388 | ||
389 | case JOB_TRY_RESTART: | |
390 | s = unit_active_state(u); | |
391 | if (UNIT_IS_INACTIVE_OR_DEACTIVATING(s)) | |
392 | *t = JOB_NOP; | |
393 | else | |
394 | *t = JOB_RESTART; | |
395 | break; | |
396 | ||
397 | case JOB_RELOAD_OR_START: | |
398 | s = unit_active_state(u); | |
399 | if (UNIT_IS_INACTIVE_OR_DEACTIVATING(s)) | |
400 | *t = JOB_START; | |
401 | else | |
402 | *t = JOB_RELOAD; | |
403 | break; | |
404 | ||
405 | default: | |
406 | ; | |
407 | } | |
408 | } | |
409 | ||
410 | int job_type_merge_and_collapse(JobType *a, JobType b, Unit *u) { | |
411 | JobType t = job_type_lookup_merge(*a, b); | |
412 | if (t < 0) | |
413 | return -EEXIST; | |
414 | *a = t; | |
415 | job_type_collapse(a, u); | |
416 | return 0; | |
417 | } | |
418 | ||
419 | bool job_is_runnable(Job *j) { | |
420 | Iterator i; | |
421 | Unit *other; | |
422 | ||
423 | assert(j); | |
424 | assert(j->installed); | |
425 | ||
426 | /* Checks whether there is any job running for the units this | |
427 | * job needs to be running after (in the case of a 'positive' | |
428 | * job type) or before (in the case of a 'negative' job | |
429 | * type. */ | |
430 | ||
431 | /* First check if there is an override */ | |
432 | if (j->ignore_order) | |
433 | return true; | |
434 | ||
435 | if (j->type == JOB_NOP) | |
436 | return true; | |
437 | ||
438 | if (j->type == JOB_START || | |
439 | j->type == JOB_VERIFY_ACTIVE || | |
440 | j->type == JOB_RELOAD) { | |
441 | ||
442 | /* Immediate result is that the job is or might be | |
443 | * started. In this case lets wait for the | |
444 | * dependencies, regardless whether they are | |
445 | * starting or stopping something. */ | |
446 | ||
447 | SET_FOREACH(other, j->unit->dependencies[UNIT_AFTER], i) | |
448 | if (other->job) | |
449 | return false; | |
450 | } | |
451 | ||
452 | /* Also, if something else is being stopped and we should | |
453 | * change state after it, then lets wait. */ | |
454 | ||
455 | SET_FOREACH(other, j->unit->dependencies[UNIT_BEFORE], i) | |
456 | if (other->job && | |
457 | (other->job->type == JOB_STOP || | |
458 | other->job->type == JOB_RESTART)) | |
459 | return false; | |
460 | ||
461 | /* This means that for a service a and a service b where b | |
462 | * shall be started after a: | |
463 | * | |
464 | * start a + start b → 1st step start a, 2nd step start b | |
465 | * start a + stop b → 1st step stop b, 2nd step start a | |
466 | * stop a + start b → 1st step stop a, 2nd step start b | |
467 | * stop a + stop b → 1st step stop b, 2nd step stop a | |
468 | * | |
469 | * This has the side effect that restarts are properly | |
470 | * synchronized too. */ | |
471 | ||
472 | return true; | |
473 | } | |
474 | ||
475 | static void job_change_type(Job *j, JobType newtype) { | |
476 | log_debug_unit(j->unit->id, | |
477 | "Converting job %s/%s -> %s/%s", | |
478 | j->unit->id, job_type_to_string(j->type), | |
479 | j->unit->id, job_type_to_string(newtype)); | |
480 | ||
481 | j->type = newtype; | |
482 | } | |
483 | ||
484 | int job_run_and_invalidate(Job *j) { | |
485 | int r; | |
486 | uint32_t id; | |
487 | Manager *m = j->manager; | |
488 | ||
489 | assert(j); | |
490 | assert(j->installed); | |
491 | assert(j->type < _JOB_TYPE_MAX_IN_TRANSACTION); | |
492 | assert(j->in_run_queue); | |
493 | ||
494 | LIST_REMOVE(Job, run_queue, j->manager->run_queue, j); | |
495 | j->in_run_queue = false; | |
496 | ||
497 | if (j->state != JOB_WAITING) | |
498 | return 0; | |
499 | ||
500 | if (!job_is_runnable(j)) | |
501 | return -EAGAIN; | |
502 | ||
503 | j->state = JOB_RUNNING; | |
504 | m->n_running_jobs++; | |
505 | job_add_to_dbus_queue(j); | |
506 | ||
507 | /* While we execute this operation the job might go away (for | |
508 | * example: because it is replaced by a new, conflicting | |
509 | * job.) To make sure we don't access a freed job later on we | |
510 | * store the id here, so that we can verify the job is still | |
511 | * valid. */ | |
512 | id = j->id; | |
513 | ||
514 | switch (j->type) { | |
515 | ||
516 | case JOB_START: | |
517 | r = unit_start(j->unit); | |
518 | ||
519 | /* If this unit cannot be started, then simply wait */ | |
520 | if (r == -EBADR) | |
521 | r = 0; | |
522 | break; | |
523 | ||
524 | case JOB_VERIFY_ACTIVE: { | |
525 | UnitActiveState t = unit_active_state(j->unit); | |
526 | if (UNIT_IS_ACTIVE_OR_RELOADING(t)) | |
527 | r = -EALREADY; | |
528 | else if (t == UNIT_ACTIVATING) | |
529 | r = -EAGAIN; | |
530 | else | |
531 | r = -ENOEXEC; | |
532 | break; | |
533 | } | |
534 | ||
535 | case JOB_STOP: | |
536 | case JOB_RESTART: | |
537 | r = unit_stop(j->unit); | |
538 | ||
539 | /* If this unit cannot stopped, then simply wait. */ | |
540 | if (r == -EBADR) | |
541 | r = 0; | |
542 | break; | |
543 | ||
544 | case JOB_RELOAD: | |
545 | r = unit_reload(j->unit); | |
546 | break; | |
547 | ||
548 | case JOB_NOP: | |
549 | r = -EALREADY; | |
550 | break; | |
551 | ||
552 | default: | |
553 | assert_not_reached("Unknown job type"); | |
554 | } | |
555 | ||
556 | j = manager_get_job(m, id); | |
557 | if (j) { | |
558 | if (r == -EALREADY) | |
559 | r = job_finish_and_invalidate(j, JOB_DONE, true); | |
560 | else if (r == -ENOEXEC) | |
561 | r = job_finish_and_invalidate(j, JOB_SKIPPED, true); | |
562 | else if (r == -EAGAIN) { | |
563 | j->state = JOB_WAITING; | |
564 | m->n_running_jobs--; | |
565 | } else if (r < 0) | |
566 | r = job_finish_and_invalidate(j, JOB_FAILED, true); | |
567 | } | |
568 | ||
569 | return r; | |
570 | } | |
571 | ||
572 | static const char *job_get_status_message_format(Unit *u, JobType t, JobResult result) { | |
573 | const UnitStatusMessageFormats *format_table; | |
574 | ||
575 | assert(u); | |
576 | assert(t >= 0); | |
577 | assert(t < _JOB_TYPE_MAX); | |
578 | ||
579 | format_table = &UNIT_VTABLE(u)->status_message_formats; | |
580 | if (!format_table) | |
581 | return NULL; | |
582 | ||
583 | if (t == JOB_START) | |
584 | return format_table->finished_start_job[result]; | |
585 | else if (t == JOB_STOP || t == JOB_RESTART) | |
586 | return format_table->finished_stop_job[result]; | |
587 | ||
588 | return NULL; | |
589 | } | |
590 | ||
591 | static const char *job_get_status_message_format_try_harder(Unit *u, JobType t, JobResult result) { | |
592 | const char *format; | |
593 | ||
594 | assert(u); | |
595 | assert(t >= 0); | |
596 | assert(t < _JOB_TYPE_MAX); | |
597 | ||
598 | format = job_get_status_message_format(u, t, result); | |
599 | if (format) | |
600 | return format; | |
601 | ||
602 | /* Return generic strings */ | |
603 | if (t == JOB_START) { | |
604 | if (result == JOB_DONE) | |
605 | return "Started %s."; | |
606 | else if (result == JOB_FAILED) | |
607 | return "Failed to start %s."; | |
608 | else if (result == JOB_DEPENDENCY) | |
609 | return "Dependency failed for %s."; | |
610 | else if (result == JOB_TIMEOUT) | |
611 | return "Timed out starting %s."; | |
612 | } else if (t == JOB_STOP || t == JOB_RESTART) { | |
613 | if (result == JOB_DONE) | |
614 | return "Stopped %s."; | |
615 | else if (result == JOB_FAILED) | |
616 | return "Stopped (with error) %s."; | |
617 | else if (result == JOB_TIMEOUT) | |
618 | return "Timed out stoppping %s."; | |
619 | } else if (t == JOB_RELOAD) { | |
620 | if (result == JOB_DONE) | |
621 | return "Reloaded %s."; | |
622 | else if (result == JOB_FAILED) | |
623 | return "Reload failed for %s."; | |
624 | else if (result == JOB_TIMEOUT) | |
625 | return "Timed out reloading %s."; | |
626 | } | |
627 | ||
628 | return NULL; | |
629 | } | |
630 | ||
631 | static void job_print_status_message(Unit *u, JobType t, JobResult result) { | |
632 | const char *format; | |
633 | ||
634 | assert(u); | |
635 | assert(t >= 0); | |
636 | assert(t < _JOB_TYPE_MAX); | |
637 | ||
638 | if (t == JOB_START) { | |
639 | format = job_get_status_message_format(u, t, result); | |
640 | if (!format) | |
641 | return; | |
642 | ||
643 | switch (result) { | |
644 | ||
645 | case JOB_DONE: | |
646 | if (u->condition_result) | |
647 | unit_status_printf(u, ANSI_HIGHLIGHT_GREEN_ON " OK " ANSI_HIGHLIGHT_OFF, format, unit_description(u)); | |
648 | break; | |
649 | ||
650 | case JOB_FAILED: | |
651 | unit_status_printf(u, ANSI_HIGHLIGHT_RED_ON "FAILED" ANSI_HIGHLIGHT_OFF, format, unit_description(u)); | |
652 | unit_status_printf(u, NULL, "See 'systemctl status %s' for details.", u->id); | |
653 | break; | |
654 | ||
655 | case JOB_DEPENDENCY: | |
656 | unit_status_printf(u, ANSI_HIGHLIGHT_YELLOW_ON "DEPEND" ANSI_HIGHLIGHT_OFF, format, unit_description(u)); | |
657 | break; | |
658 | ||
659 | case JOB_TIMEOUT: | |
660 | unit_status_printf(u, ANSI_HIGHLIGHT_RED_ON " TIME " ANSI_HIGHLIGHT_OFF, format, unit_description(u)); | |
661 | break; | |
662 | ||
663 | default: | |
664 | ; | |
665 | } | |
666 | ||
667 | } else if (t == JOB_STOP || t == JOB_RESTART) { | |
668 | ||
669 | format = job_get_status_message_format(u, t, result); | |
670 | if (!format) | |
671 | return; | |
672 | ||
673 | switch (result) { | |
674 | ||
675 | case JOB_TIMEOUT: | |
676 | unit_status_printf(u, ANSI_HIGHLIGHT_RED_ON " TIME " ANSI_HIGHLIGHT_OFF, format, unit_description(u)); | |
677 | break; | |
678 | ||
679 | case JOB_DONE: | |
680 | case JOB_FAILED: | |
681 | unit_status_printf(u, ANSI_HIGHLIGHT_GREEN_ON " OK " ANSI_HIGHLIGHT_OFF, format, unit_description(u)); | |
682 | break; | |
683 | ||
684 | default: | |
685 | ; | |
686 | } | |
687 | ||
688 | } else if (t == JOB_VERIFY_ACTIVE) { | |
689 | ||
690 | /* When verify-active detects the unit is inactive, report it. | |
691 | * Most likely a DEPEND warning from a requisiting unit will | |
692 | * occur next and it's nice to see what was requisited. */ | |
693 | if (result == JOB_SKIPPED) | |
694 | unit_status_printf(u, ANSI_HIGHLIGHT_ON " INFO " ANSI_HIGHLIGHT_OFF, "%s is not active.", unit_description(u)); | |
695 | } | |
696 | } | |
697 | ||
698 | #pragma GCC diagnostic push | |
699 | #pragma GCC diagnostic ignored "-Wformat-nonliteral" | |
700 | static void job_log_status_message(Unit *u, JobType t, JobResult result) { | |
701 | const char *format; | |
702 | char buf[LINE_MAX]; | |
703 | ||
704 | assert(u); | |
705 | assert(t >= 0); | |
706 | assert(t < _JOB_TYPE_MAX); | |
707 | ||
708 | /* Skip this if it goes to the console. since we already print | |
709 | * to the console anyway... */ | |
710 | ||
711 | if (log_on_console()) | |
712 | return; | |
713 | ||
714 | format = job_get_status_message_format_try_harder(u, t, result); | |
715 | if (!format) | |
716 | return; | |
717 | ||
718 | snprintf(buf, sizeof(buf), format, unit_description(u)); | |
719 | char_array_0(buf); | |
720 | ||
721 | if (t == JOB_START) { | |
722 | sd_id128_t mid; | |
723 | ||
724 | mid = result == JOB_DONE ? SD_MESSAGE_UNIT_STARTED : SD_MESSAGE_UNIT_FAILED; | |
725 | log_struct_unit(result == JOB_DONE ? LOG_INFO : LOG_ERR, | |
726 | u->id, | |
727 | MESSAGE_ID(mid), | |
728 | "RESULT=%s", job_result_to_string(result), | |
729 | "MESSAGE=%s", buf, | |
730 | NULL); | |
731 | ||
732 | } else if (t == JOB_STOP) | |
733 | log_struct_unit(result == JOB_DONE ? LOG_INFO : LOG_ERR, | |
734 | u->id, | |
735 | MESSAGE_ID(SD_MESSAGE_UNIT_STOPPED), | |
736 | "RESULT=%s", job_result_to_string(result), | |
737 | "MESSAGE=%s", buf, | |
738 | NULL); | |
739 | ||
740 | else if (t == JOB_RELOAD) | |
741 | log_struct_unit(result == JOB_DONE ? LOG_INFO : LOG_ERR, | |
742 | u->id, | |
743 | MESSAGE_ID(SD_MESSAGE_UNIT_RELOADED), | |
744 | "RESULT=%s", job_result_to_string(result), | |
745 | "MESSAGE=%s", buf, | |
746 | NULL); | |
747 | } | |
748 | #pragma GCC diagnostic pop | |
749 | ||
750 | int job_finish_and_invalidate(Job *j, JobResult result, bool recursive) { | |
751 | Unit *u; | |
752 | Unit *other; | |
753 | JobType t; | |
754 | Iterator i; | |
755 | ||
756 | assert(j); | |
757 | assert(j->installed); | |
758 | assert(j->type < _JOB_TYPE_MAX_IN_TRANSACTION); | |
759 | ||
760 | u = j->unit; | |
761 | t = j->type; | |
762 | ||
763 | j->result = result; | |
764 | ||
765 | if (j->state == JOB_RUNNING) | |
766 | j->manager->n_running_jobs--; | |
767 | ||
768 | log_debug_unit(u->id, "Job %s/%s finished, result=%s", | |
769 | u->id, job_type_to_string(t), job_result_to_string(result)); | |
770 | ||
771 | job_print_status_message(u, t, result); | |
772 | job_log_status_message(u, t, result); | |
773 | ||
774 | job_add_to_dbus_queue(j); | |
775 | ||
776 | /* Patch restart jobs so that they become normal start jobs */ | |
777 | if (result == JOB_DONE && t == JOB_RESTART) { | |
778 | ||
779 | job_change_type(j, JOB_START); | |
780 | j->state = JOB_WAITING; | |
781 | ||
782 | job_add_to_run_queue(j); | |
783 | ||
784 | goto finish; | |
785 | } | |
786 | ||
787 | if (result == JOB_FAILED) | |
788 | j->manager->n_failed_jobs ++; | |
789 | ||
790 | job_uninstall(j); | |
791 | job_free(j); | |
792 | ||
793 | /* Fail depending jobs on failure */ | |
794 | if (result != JOB_DONE && recursive) { | |
795 | ||
796 | if (t == JOB_START || | |
797 | t == JOB_VERIFY_ACTIVE) { | |
798 | ||
799 | SET_FOREACH(other, u->dependencies[UNIT_REQUIRED_BY], i) | |
800 | if (other->job && | |
801 | (other->job->type == JOB_START || | |
802 | other->job->type == JOB_VERIFY_ACTIVE)) | |
803 | job_finish_and_invalidate(other->job, JOB_DEPENDENCY, true); | |
804 | ||
805 | SET_FOREACH(other, u->dependencies[UNIT_BOUND_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 | SET_FOREACH(other, u->dependencies[UNIT_REQUIRED_BY_OVERRIDABLE], i) | |
812 | if (other->job && | |
813 | !other->job->override && | |
814 | (other->job->type == JOB_START || | |
815 | other->job->type == JOB_VERIFY_ACTIVE)) | |
816 | job_finish_and_invalidate(other->job, JOB_DEPENDENCY, true); | |
817 | ||
818 | } else if (t == JOB_STOP) { | |
819 | ||
820 | SET_FOREACH(other, u->dependencies[UNIT_CONFLICTED_BY], i) | |
821 | if (other->job && | |
822 | (other->job->type == JOB_START || | |
823 | other->job->type == JOB_VERIFY_ACTIVE)) | |
824 | job_finish_and_invalidate(other->job, JOB_DEPENDENCY, true); | |
825 | } | |
826 | } | |
827 | ||
828 | /* Trigger OnFailure dependencies that are not generated by | |
829 | * the unit itself. We don't treat JOB_CANCELED as failure in | |
830 | * this context. And JOB_FAILURE is already handled by the | |
831 | * unit itself. */ | |
832 | if (result == JOB_TIMEOUT || result == JOB_DEPENDENCY) { | |
833 | log_struct_unit(LOG_NOTICE, | |
834 | u->id, | |
835 | "JOB_TYPE=%s", job_type_to_string(t), | |
836 | "JOB_RESULT=%s", job_result_to_string(result), | |
837 | "Job %s/%s failed with result '%s'.", | |
838 | u->id, | |
839 | job_type_to_string(t), | |
840 | job_result_to_string(result), | |
841 | NULL); | |
842 | ||
843 | unit_trigger_on_failure(u); | |
844 | } | |
845 | ||
846 | finish: | |
847 | /* Try to start the next jobs that can be started */ | |
848 | SET_FOREACH(other, u->dependencies[UNIT_AFTER], i) | |
849 | if (other->job) | |
850 | job_add_to_run_queue(other->job); | |
851 | SET_FOREACH(other, u->dependencies[UNIT_BEFORE], i) | |
852 | if (other->job) | |
853 | job_add_to_run_queue(other->job); | |
854 | ||
855 | manager_check_finished(u->manager); | |
856 | ||
857 | return 0; | |
858 | } | |
859 | ||
860 | int job_start_timer(Job *j) { | |
861 | struct itimerspec its; | |
862 | struct epoll_event ev; | |
863 | int fd, r; | |
864 | assert(j); | |
865 | ||
866 | if (j->unit->job_timeout <= 0 || | |
867 | j->timer_watch.type == WATCH_JOB_TIMER) | |
868 | return 0; | |
869 | ||
870 | assert(j->timer_watch.type == WATCH_INVALID); | |
871 | ||
872 | if ((fd = timerfd_create(CLOCK_MONOTONIC, TFD_NONBLOCK|TFD_CLOEXEC)) < 0) { | |
873 | r = -errno; | |
874 | goto fail; | |
875 | } | |
876 | ||
877 | zero(its); | |
878 | timespec_store(&its.it_value, j->unit->job_timeout); | |
879 | ||
880 | if (timerfd_settime(fd, 0, &its, NULL) < 0) { | |
881 | r = -errno; | |
882 | goto fail; | |
883 | } | |
884 | ||
885 | zero(ev); | |
886 | ev.data.ptr = &j->timer_watch; | |
887 | ev.events = EPOLLIN; | |
888 | ||
889 | if (epoll_ctl(j->manager->epoll_fd, EPOLL_CTL_ADD, fd, &ev) < 0) { | |
890 | r = -errno; | |
891 | goto fail; | |
892 | } | |
893 | ||
894 | j->timer_watch.type = WATCH_JOB_TIMER; | |
895 | j->timer_watch.fd = fd; | |
896 | j->timer_watch.data.job = j; | |
897 | ||
898 | return 0; | |
899 | ||
900 | fail: | |
901 | if (fd >= 0) | |
902 | close_nointr_nofail(fd); | |
903 | ||
904 | return r; | |
905 | } | |
906 | ||
907 | void job_add_to_run_queue(Job *j) { | |
908 | assert(j); | |
909 | assert(j->installed); | |
910 | ||
911 | if (j->in_run_queue) | |
912 | return; | |
913 | ||
914 | LIST_PREPEND(Job, run_queue, j->manager->run_queue, j); | |
915 | j->in_run_queue = true; | |
916 | } | |
917 | ||
918 | void job_add_to_dbus_queue(Job *j) { | |
919 | assert(j); | |
920 | assert(j->installed); | |
921 | ||
922 | if (j->in_dbus_queue) | |
923 | return; | |
924 | ||
925 | /* We don't check if anybody is subscribed here, since this | |
926 | * job might just have been created and not yet assigned to a | |
927 | * connection/client. */ | |
928 | ||
929 | LIST_PREPEND(Job, dbus_queue, j->manager->dbus_job_queue, j); | |
930 | j->in_dbus_queue = true; | |
931 | } | |
932 | ||
933 | char *job_dbus_path(Job *j) { | |
934 | char *p; | |
935 | ||
936 | assert(j); | |
937 | ||
938 | if (asprintf(&p, "/org/freedesktop/systemd1/job/%lu", (unsigned long) j->id) < 0) | |
939 | return NULL; | |
940 | ||
941 | return p; | |
942 | } | |
943 | ||
944 | void job_timer_event(Job *j, uint64_t n_elapsed, Watch *w) { | |
945 | assert(j); | |
946 | assert(w == &j->timer_watch); | |
947 | ||
948 | log_warning_unit(j->unit->id, "Job %s/%s timed out.", | |
949 | j->unit->id, job_type_to_string(j->type)); | |
950 | job_finish_and_invalidate(j, JOB_TIMEOUT, true); | |
951 | } | |
952 | ||
953 | int job_serialize(Job *j, FILE *f, FDSet *fds) { | |
954 | fprintf(f, "job-id=%u\n", j->id); | |
955 | fprintf(f, "job-type=%s\n", job_type_to_string(j->type)); | |
956 | fprintf(f, "job-state=%s\n", job_state_to_string(j->state)); | |
957 | fprintf(f, "job-override=%s\n", yes_no(j->override)); | |
958 | fprintf(f, "job-irreversible=%s\n", yes_no(j->irreversible)); | |
959 | fprintf(f, "job-sent-dbus-new-signal=%s\n", yes_no(j->sent_dbus_new_signal)); | |
960 | fprintf(f, "job-ignore-order=%s\n", yes_no(j->ignore_order)); | |
961 | /* Cannot save bus clients. Just note the fact that we're losing | |
962 | * them. job_send_message() will fallback to broadcasting. */ | |
963 | fprintf(f, "job-forgot-bus-clients=%s\n", | |
964 | yes_no(j->forgot_bus_clients || j->bus_client_list)); | |
965 | if (j->timer_watch.type == WATCH_JOB_TIMER) { | |
966 | int copy = fdset_put_dup(fds, j->timer_watch.fd); | |
967 | if (copy < 0) | |
968 | return copy; | |
969 | fprintf(f, "job-timer-watch-fd=%d\n", copy); | |
970 | } | |
971 | ||
972 | /* End marker */ | |
973 | fputc('\n', f); | |
974 | return 0; | |
975 | } | |
976 | ||
977 | int job_deserialize(Job *j, FILE *f, FDSet *fds) { | |
978 | for (;;) { | |
979 | char line[LINE_MAX], *l, *v; | |
980 | size_t k; | |
981 | ||
982 | if (!fgets(line, sizeof(line), f)) { | |
983 | if (feof(f)) | |
984 | return 0; | |
985 | return -errno; | |
986 | } | |
987 | ||
988 | char_array_0(line); | |
989 | l = strstrip(line); | |
990 | ||
991 | /* End marker */ | |
992 | if (l[0] == 0) | |
993 | return 0; | |
994 | ||
995 | k = strcspn(l, "="); | |
996 | ||
997 | if (l[k] == '=') { | |
998 | l[k] = 0; | |
999 | v = l+k+1; | |
1000 | } else | |
1001 | v = l+k; | |
1002 | ||
1003 | if (streq(l, "job-id")) { | |
1004 | if (safe_atou32(v, &j->id) < 0) | |
1005 | log_debug("Failed to parse job id value %s", v); | |
1006 | } else if (streq(l, "job-type")) { | |
1007 | JobType t = job_type_from_string(v); | |
1008 | if (t < 0) | |
1009 | log_debug("Failed to parse job type %s", v); | |
1010 | else if (t >= _JOB_TYPE_MAX_IN_TRANSACTION) | |
1011 | log_debug("Cannot deserialize job of type %s", v); | |
1012 | else | |
1013 | j->type = t; | |
1014 | } else if (streq(l, "job-state")) { | |
1015 | JobState s = job_state_from_string(v); | |
1016 | if (s < 0) | |
1017 | log_debug("Failed to parse job state %s", v); | |
1018 | else | |
1019 | j->state = s; | |
1020 | } else if (streq(l, "job-override")) { | |
1021 | int b = parse_boolean(v); | |
1022 | if (b < 0) | |
1023 | log_debug("Failed to parse job override flag %s", v); | |
1024 | else | |
1025 | j->override = j->override || b; | |
1026 | } else if (streq(l, "job-irreversible")) { | |
1027 | int b = parse_boolean(v); | |
1028 | if (b < 0) | |
1029 | log_debug("Failed to parse job irreversible flag %s", v); | |
1030 | else | |
1031 | j->irreversible = j->irreversible || b; | |
1032 | } else if (streq(l, "job-sent-dbus-new-signal")) { | |
1033 | int b = parse_boolean(v); | |
1034 | if (b < 0) | |
1035 | log_debug("Failed to parse job sent_dbus_new_signal flag %s", v); | |
1036 | else | |
1037 | j->sent_dbus_new_signal = j->sent_dbus_new_signal || b; | |
1038 | } else if (streq(l, "job-ignore-order")) { | |
1039 | int b = parse_boolean(v); | |
1040 | if (b < 0) | |
1041 | log_debug("Failed to parse job ignore_order flag %s", v); | |
1042 | else | |
1043 | j->ignore_order = j->ignore_order || b; | |
1044 | } else if (streq(l, "job-forgot-bus-clients")) { | |
1045 | int b = parse_boolean(v); | |
1046 | if (b < 0) | |
1047 | log_debug("Failed to parse job forgot_bus_clients flag %s", v); | |
1048 | else | |
1049 | j->forgot_bus_clients = j->forgot_bus_clients || b; | |
1050 | } else if (streq(l, "job-timer-watch-fd")) { | |
1051 | int fd; | |
1052 | if (safe_atoi(v, &fd) < 0 || fd < 0 || !fdset_contains(fds, fd)) | |
1053 | log_debug("Failed to parse job-timer-watch-fd value %s", v); | |
1054 | else { | |
1055 | if (j->timer_watch.type == WATCH_JOB_TIMER) | |
1056 | close_nointr_nofail(j->timer_watch.fd); | |
1057 | ||
1058 | j->timer_watch.type = WATCH_JOB_TIMER; | |
1059 | j->timer_watch.fd = fdset_remove(fds, fd); | |
1060 | j->timer_watch.data.job = j; | |
1061 | } | |
1062 | } | |
1063 | } | |
1064 | } | |
1065 | ||
1066 | int job_coldplug(Job *j) { | |
1067 | struct epoll_event ev; | |
1068 | ||
1069 | if (j->timer_watch.type != WATCH_JOB_TIMER) | |
1070 | return 0; | |
1071 | ||
1072 | zero(ev); | |
1073 | ev.data.ptr = &j->timer_watch; | |
1074 | ev.events = EPOLLIN; | |
1075 | ||
1076 | if (epoll_ctl(j->manager->epoll_fd, EPOLL_CTL_ADD, j->timer_watch.fd, &ev) < 0) | |
1077 | return -errno; | |
1078 | ||
1079 | return 0; | |
1080 | } | |
1081 | ||
1082 | void job_shutdown_magic(Job *j) { | |
1083 | assert(j); | |
1084 | ||
1085 | /* The shutdown target gets some special treatment here: we | |
1086 | * tell the kernel to begin with flushing its disk caches, to | |
1087 | * optimize shutdown time a bit. Ideally we wouldn't hardcode | |
1088 | * this magic into PID 1. However all other processes aren't | |
1089 | * options either since they'd exit much sooner than PID 1 and | |
1090 | * asynchronous sync() would cause their exit to be | |
1091 | * delayed. */ | |
1092 | ||
1093 | if (!unit_has_name(j->unit, SPECIAL_SHUTDOWN_TARGET)) | |
1094 | return; | |
1095 | ||
1096 | if (j->type != JOB_START) | |
1097 | return; | |
1098 | ||
1099 | if (detect_container(NULL) > 0) | |
1100 | return; | |
1101 | ||
1102 | asynchronous_sync(); | |
1103 | } | |
1104 | ||
1105 | static const char* const job_state_table[_JOB_STATE_MAX] = { | |
1106 | [JOB_WAITING] = "waiting", | |
1107 | [JOB_RUNNING] = "running" | |
1108 | }; | |
1109 | ||
1110 | DEFINE_STRING_TABLE_LOOKUP(job_state, JobState); | |
1111 | ||
1112 | static const char* const job_type_table[_JOB_TYPE_MAX] = { | |
1113 | [JOB_START] = "start", | |
1114 | [JOB_VERIFY_ACTIVE] = "verify-active", | |
1115 | [JOB_STOP] = "stop", | |
1116 | [JOB_RELOAD] = "reload", | |
1117 | [JOB_RELOAD_OR_START] = "reload-or-start", | |
1118 | [JOB_RESTART] = "restart", | |
1119 | [JOB_TRY_RESTART] = "try-restart", | |
1120 | [JOB_NOP] = "nop", | |
1121 | }; | |
1122 | ||
1123 | DEFINE_STRING_TABLE_LOOKUP(job_type, JobType); | |
1124 | ||
1125 | static const char* const job_mode_table[_JOB_MODE_MAX] = { | |
1126 | [JOB_FAIL] = "fail", | |
1127 | [JOB_REPLACE] = "replace", | |
1128 | [JOB_REPLACE_IRREVERSIBLY] = "replace-irreversibly", | |
1129 | [JOB_ISOLATE] = "isolate", | |
1130 | [JOB_IGNORE_DEPENDENCIES] = "ignore-dependencies", | |
1131 | [JOB_IGNORE_REQUIREMENTS] = "ignore-requirements" | |
1132 | }; | |
1133 | ||
1134 | DEFINE_STRING_TABLE_LOOKUP(job_mode, JobMode); | |
1135 | ||
1136 | static const char* const job_result_table[_JOB_RESULT_MAX] = { | |
1137 | [JOB_DONE] = "done", | |
1138 | [JOB_CANCELED] = "canceled", | |
1139 | [JOB_TIMEOUT] = "timeout", | |
1140 | [JOB_FAILED] = "failed", | |
1141 | [JOB_DEPENDENCY] = "dependency", | |
1142 | [JOB_SKIPPED] = "skipped" | |
1143 | }; | |
1144 | ||
1145 | DEFINE_STRING_TABLE_LOOKUP(job_result, JobResult); |