]> git.ipfire.org Git - thirdparty/systemd.git/blob - src/core/unit.c
a6c3fd8bda588ae9fbddced353bfa7ca62055246
[thirdparty/systemd.git] / src / core / unit.c
1 /* SPDX-License-Identifier: LGPL-2.1-or-later */
2
3 #include <fnmatch.h>
4 #include <linux/capability.h>
5 #include <unistd.h>
6
7 #include "sd-bus.h"
8 #include "sd-id128.h"
9 #include "sd-messages.h"
10
11 #include "all-units.h"
12 #include "alloc-util.h"
13 #include "ansi-color.h"
14 #include "bpf-firewall.h"
15 #include "bpf-restrict-fs.h"
16 #include "bus-common-errors.h"
17 #include "bus-internal.h"
18 #include "bus-util.h"
19 #include "cgroup-setup.h"
20 #include "cgroup-util.h"
21 #include "chase.h"
22 #include "chattr-util.h"
23 #include "condition.h"
24 #include "dbus-unit.h"
25 #include "dropin.h"
26 #include "dynamic-user.h"
27 #include "env-util.h"
28 #include "escape.h"
29 #include "exec-credential.h"
30 #include "execute.h"
31 #include "fd-util.h"
32 #include "fileio.h"
33 #include "format-util.h"
34 #include "fs-util.h"
35 #include "id128-util.h"
36 #include "install.h"
37 #include "iovec-util.h"
38 #include "label-util.h"
39 #include "load-dropin.h"
40 #include "load-fragment.h"
41 #include "log.h"
42 #include "logarithm.h"
43 #include "mkdir-label.h"
44 #include "manager.h"
45 #include "mount-util.h"
46 #include "mountpoint-util.h"
47 #include "path-util.h"
48 #include "process-util.h"
49 #include "quota-util.h"
50 #include "rm-rf.h"
51 #include "serialize.h"
52 #include "set.h"
53 #include "signal-util.h"
54 #include "siphash24.h"
55 #include "sparse-endian.h"
56 #include "special.h"
57 #include "specifier.h"
58 #include "stat-util.h"
59 #include "string-table.h"
60 #include "string-util.h"
61 #include "strv.h"
62 #include "tmpfile-util.h"
63 #include "umask-util.h"
64 #include "unit.h"
65 #include "unit-name.h"
66 #include "user-util.h"
67 #include "varlink.h"
68
69 /* Thresholds for logging at INFO level about resource consumption */
70 #define MENTIONWORTHY_CPU_NSEC (1 * NSEC_PER_SEC)
71 #define MENTIONWORTHY_MEMORY_BYTES (64 * U64_MB)
72 #define MENTIONWORTHY_IO_BYTES (1 * U64_MB)
73 #define MENTIONWORTHY_IP_BYTES UINT64_C(0)
74
75 /* Thresholds for logging at NOTICE level about resource consumption */
76 #define NOTICEWORTHY_CPU_NSEC (10 * NSEC_PER_MINUTE)
77 #define NOTICEWORTHY_MEMORY_BYTES (512 * U64_MB)
78 #define NOTICEWORTHY_IO_BYTES (10 * U64_MB)
79 #define NOTICEWORTHY_IP_BYTES (128 * U64_MB)
80
81 const UnitVTable * const unit_vtable[_UNIT_TYPE_MAX] = {
82 [UNIT_SERVICE] = &service_vtable,
83 [UNIT_SOCKET] = &socket_vtable,
84 [UNIT_TARGET] = &target_vtable,
85 [UNIT_DEVICE] = &device_vtable,
86 [UNIT_MOUNT] = &mount_vtable,
87 [UNIT_AUTOMOUNT] = &automount_vtable,
88 [UNIT_SWAP] = &swap_vtable,
89 [UNIT_TIMER] = &timer_vtable,
90 [UNIT_PATH] = &path_vtable,
91 [UNIT_SLICE] = &slice_vtable,
92 [UNIT_SCOPE] = &scope_vtable,
93 };
94
95 Unit* unit_new(Manager *m, size_t size) {
96 Unit *u;
97
98 assert(m);
99 assert(size >= sizeof(Unit));
100
101 u = malloc0(size);
102 if (!u)
103 return NULL;
104
105 u->manager = m;
106 u->type = _UNIT_TYPE_INVALID;
107 u->default_dependencies = true;
108 u->unit_file_state = _UNIT_FILE_STATE_INVALID;
109 u->unit_file_preset = _PRESET_ACTION_INVALID;
110 u->on_failure_job_mode = JOB_REPLACE;
111 u->on_success_job_mode = JOB_FAIL;
112 u->job_timeout = USEC_INFINITY;
113 u->job_running_timeout = USEC_INFINITY;
114 u->ref_uid = UID_INVALID;
115 u->ref_gid = GID_INVALID;
116
117 u->failure_action_exit_status = u->success_action_exit_status = -1;
118
119 u->last_section_private = -1;
120
121 u->start_ratelimit = m->defaults.start_limit;
122
123 u->auto_start_stop_ratelimit = (const RateLimit) {
124 .interval = 10 * USEC_PER_SEC,
125 .burst = 16
126 };
127
128 return u;
129 }
130
131 int unit_new_for_name(Manager *m, size_t size, const char *name, Unit **ret) {
132 _cleanup_(unit_freep) Unit *u = NULL;
133 int r;
134
135 u = unit_new(m, size);
136 if (!u)
137 return -ENOMEM;
138
139 r = unit_add_name(u, name);
140 if (r < 0)
141 return r;
142
143 *ret = TAKE_PTR(u);
144
145 return r;
146 }
147
148 bool unit_has_name(const Unit *u, const char *name) {
149 assert(u);
150 assert(name);
151
152 return streq_ptr(name, u->id) ||
153 set_contains(u->aliases, name);
154 }
155
156 static void unit_init(Unit *u) {
157 CGroupContext *cc;
158 ExecContext *ec;
159 KillContext *kc;
160
161 assert(u);
162 assert(u->manager);
163 assert(u->type >= 0);
164
165 cc = unit_get_cgroup_context(u);
166 if (cc) {
167 cgroup_context_init(cc);
168
169 /* Copy in the manager defaults into the cgroup
170 * context, _before_ the rest of the settings have
171 * been initialized */
172
173 cc->io_accounting = u->manager->defaults.io_accounting;
174 cc->memory_accounting = u->manager->defaults.memory_accounting;
175 cc->tasks_accounting = u->manager->defaults.tasks_accounting;
176 cc->ip_accounting = u->manager->defaults.ip_accounting;
177
178 if (u->type != UNIT_SLICE)
179 cc->tasks_max = u->manager->defaults.tasks_max;
180
181 cc->memory_pressure_watch = u->manager->defaults.memory_pressure_watch;
182 cc->memory_pressure_threshold_usec = u->manager->defaults.memory_pressure_threshold_usec;
183 }
184
185 ec = unit_get_exec_context(u);
186 if (ec) {
187 exec_context_init(ec);
188
189 if (u->manager->defaults.oom_score_adjust_set) {
190 ec->oom_score_adjust = u->manager->defaults.oom_score_adjust;
191 ec->oom_score_adjust_set = true;
192 }
193
194 ec->restrict_suid_sgid = u->manager->defaults.restrict_suid_sgid;
195
196 if (MANAGER_IS_SYSTEM(u->manager))
197 ec->keyring_mode = EXEC_KEYRING_SHARED;
198 else {
199 ec->keyring_mode = EXEC_KEYRING_INHERIT;
200
201 /* User manager might have its umask redefined by PAM or UMask=. In this
202 * case let the units it manages inherit this value by default. They can
203 * still tune this value through their own unit file */
204 (void) get_process_umask(0, &ec->umask);
205 }
206 }
207
208 kc = unit_get_kill_context(u);
209 if (kc)
210 kill_context_init(kc);
211
212 if (UNIT_VTABLE(u)->init)
213 UNIT_VTABLE(u)->init(u);
214 }
215
216 static int unit_add_alias(Unit *u, char *donated_name) {
217 int r;
218
219 /* Make sure that u->names is allocated. We may leave u->names
220 * empty if we fail later, but this is not a problem. */
221 r = set_ensure_put(&u->aliases, &string_hash_ops_free, donated_name);
222 if (r < 0)
223 return r;
224 assert(r > 0);
225
226 return 0;
227 }
228
229 int unit_add_name(Unit *u, const char *text) {
230 _cleanup_free_ char *name = NULL, *instance = NULL;
231 UnitType t;
232 int r;
233
234 assert(u);
235 assert(text);
236
237 if (unit_name_is_valid(text, UNIT_NAME_TEMPLATE)) {
238 if (!u->instance)
239 return log_unit_debug_errno(u, SYNTHETIC_ERRNO(EINVAL),
240 "Instance is not set when adding name '%s'.", text);
241
242 r = unit_name_replace_instance(text, u->instance, &name);
243 if (r < 0)
244 return log_unit_debug_errno(u, r,
245 "Failed to build instance name from '%s': %m", text);
246 } else {
247 name = strdup(text);
248 if (!name)
249 return -ENOMEM;
250 }
251
252 if (unit_has_name(u, name))
253 return 0;
254
255 if (hashmap_contains(u->manager->units, name))
256 return log_unit_debug_errno(u, SYNTHETIC_ERRNO(EEXIST),
257 "Unit already exist when adding name '%s'.", name);
258
259 if (!unit_name_is_valid(name, UNIT_NAME_PLAIN|UNIT_NAME_INSTANCE))
260 return log_unit_debug_errno(u, SYNTHETIC_ERRNO(EINVAL),
261 "Name '%s' is invalid.", name);
262
263 t = unit_name_to_type(name);
264 if (t < 0)
265 return log_unit_debug_errno(u, SYNTHETIC_ERRNO(EINVAL),
266 "failed to derive unit type from name '%s'.", name);
267
268 if (u->type != _UNIT_TYPE_INVALID && t != u->type)
269 return log_unit_debug_errno(u, SYNTHETIC_ERRNO(EINVAL),
270 "Unit type is illegal: u->type(%d) and t(%d) for name '%s'.",
271 u->type, t, name);
272
273 r = unit_name_to_instance(name, &instance);
274 if (r < 0)
275 return log_unit_debug_errno(u, r, "Failed to extract instance from name '%s': %m", name);
276
277 if (instance && !unit_type_may_template(t))
278 return log_unit_debug_errno(u, SYNTHETIC_ERRNO(EINVAL), "Templates are not allowed for name '%s'.", name);
279
280 /* Ensure that this unit either has no instance, or that the instance matches. */
281 if (u->type != _UNIT_TYPE_INVALID && !streq_ptr(u->instance, instance))
282 return log_unit_debug_errno(u, SYNTHETIC_ERRNO(EINVAL),
283 "Cannot add name %s, the instances don't match (\"%s\" != \"%s\").",
284 name, instance, u->instance);
285
286 if (u->id && !unit_type_may_alias(t))
287 return log_unit_debug_errno(u, SYNTHETIC_ERRNO(EEXIST),
288 "Cannot add name %s, aliases are not allowed for %s units.",
289 name, unit_type_to_string(t));
290
291 if (hashmap_size(u->manager->units) >= MANAGER_MAX_NAMES)
292 return log_unit_warning_errno(u, SYNTHETIC_ERRNO(E2BIG), "Cannot add name, manager has too many units.");
293
294 /* Add name to the global hashmap first, because that's easier to undo */
295 r = hashmap_put(u->manager->units, name, u);
296 if (r < 0)
297 return log_unit_debug_errno(u, r, "Add unit to hashmap failed for name '%s': %m", text);
298
299 if (u->id) {
300 r = unit_add_alias(u, name); /* unit_add_alias() takes ownership of the name on success */
301 if (r < 0) {
302 hashmap_remove(u->manager->units, name);
303 return r;
304 }
305 TAKE_PTR(name);
306
307 } else {
308 /* A new name, we don't need the set yet. */
309 assert(u->type == _UNIT_TYPE_INVALID);
310 assert(!u->instance);
311
312 u->type = t;
313 u->id = TAKE_PTR(name);
314 u->instance = TAKE_PTR(instance);
315
316 LIST_PREPEND(units_by_type, u->manager->units_by_type[t], u);
317 unit_init(u);
318 }
319
320 unit_add_to_dbus_queue(u);
321 return 0;
322 }
323
324 int unit_choose_id(Unit *u, const char *name) {
325 _cleanup_free_ char *t = NULL;
326 char *s;
327 int r;
328
329 assert(u);
330 assert(name);
331
332 if (unit_name_is_valid(name, UNIT_NAME_TEMPLATE)) {
333 if (!u->instance)
334 return -EINVAL;
335
336 r = unit_name_replace_instance(name, u->instance, &t);
337 if (r < 0)
338 return r;
339
340 name = t;
341 }
342
343 if (streq_ptr(u->id, name))
344 return 0; /* Nothing to do. */
345
346 /* Selects one of the aliases of this unit as the id */
347 s = set_get(u->aliases, (char*) name);
348 if (!s)
349 return -ENOENT;
350
351 if (u->id) {
352 r = set_remove_and_put(u->aliases, name, u->id);
353 if (r < 0)
354 return r;
355 } else
356 assert_se(set_remove(u->aliases, name)); /* see set_get() above… */
357
358 u->id = s; /* Old u->id is now stored in the set, and s is not stored anywhere */
359 unit_add_to_dbus_queue(u);
360
361 return 0;
362 }
363
364 int unit_set_description(Unit *u, const char *description) {
365 int r;
366
367 assert(u);
368
369 r = free_and_strdup(&u->description, empty_to_null(description));
370 if (r < 0)
371 return r;
372 if (r > 0)
373 unit_add_to_dbus_queue(u);
374
375 return 0;
376 }
377
378 static bool unit_success_failure_handler_has_jobs(Unit *unit) {
379 Unit *other;
380
381 UNIT_FOREACH_DEPENDENCY(other, unit, UNIT_ATOM_ON_SUCCESS)
382 if (other->job || other->nop_job)
383 return true;
384
385 UNIT_FOREACH_DEPENDENCY(other, unit, UNIT_ATOM_ON_FAILURE)
386 if (other->job || other->nop_job)
387 return true;
388
389 return false;
390 }
391
392 void unit_release_resources(Unit *u) {
393 UnitActiveState state;
394 ExecContext *ec;
395
396 assert(u);
397
398 if (u->job || u->nop_job)
399 return;
400
401 if (u->perpetual)
402 return;
403
404 state = unit_active_state(u);
405 if (!UNIT_IS_INACTIVE_OR_FAILED(state))
406 return;
407
408 if (unit_will_restart(u))
409 return;
410
411 ec = unit_get_exec_context(u);
412 if (ec && ec->runtime_directory_preserve_mode == EXEC_PRESERVE_RESTART)
413 exec_context_destroy_runtime_directory(ec, u->manager->prefix[EXEC_DIRECTORY_RUNTIME]);
414
415 if (UNIT_VTABLE(u)->release_resources)
416 UNIT_VTABLE(u)->release_resources(u);
417 }
418
419 bool unit_may_gc(Unit *u) {
420 UnitActiveState state;
421 int r;
422
423 assert(u);
424
425 /* Checks whether the unit is ready to be unloaded for garbage collection. Returns true when the
426 * unit may be collected, and false if there's some reason to keep it loaded.
427 *
428 * References from other units are *not* checked here. Instead, this is done in unit_gc_sweep(), but
429 * using markers to properly collect dependency loops.
430 */
431
432 if (u->job || u->nop_job)
433 return false;
434
435 if (u->perpetual)
436 return false;
437
438 /* if we saw a cgroup empty event for this unit, stay around until we processed it so that we remove
439 * the empty cgroup if possible. Similar, process any pending OOM events if they are already queued
440 * before we release the unit. */
441 if (u->in_cgroup_empty_queue || u->in_cgroup_oom_queue)
442 return false;
443
444 /* Make sure to send out D-Bus events before we unload the unit */
445 if (u->in_dbus_queue)
446 return false;
447
448 if (sd_bus_track_count(u->bus_track) > 0)
449 return false;
450
451 state = unit_active_state(u);
452
453 /* But we keep the unit object around for longer when it is referenced or configured to not be
454 * gc'ed */
455 switch (u->collect_mode) {
456
457 case COLLECT_INACTIVE:
458 if (state != UNIT_INACTIVE)
459 return false;
460
461 break;
462
463 case COLLECT_INACTIVE_OR_FAILED:
464 if (!UNIT_IS_INACTIVE_OR_FAILED(state))
465 return false;
466
467 break;
468
469 default:
470 assert_not_reached();
471 }
472
473 /* Check if any OnFailure= or on Success= jobs may be pending */
474 if (unit_success_failure_handler_has_jobs(u))
475 return false;
476
477 /* If the unit has a cgroup, then check whether there's anything in it. If so, we should stay
478 * around. Units with active processes should never be collected. */
479 r = unit_cgroup_is_empty(u);
480 if (r <= 0 && !IN_SET(r, -ENXIO, -EOWNERDEAD))
481 return false; /* ENXIO/EOWNERDEAD means: currently not realized */
482
483 if (!UNIT_VTABLE(u)->may_gc)
484 return true;
485
486 return UNIT_VTABLE(u)->may_gc(u);
487 }
488
489 void unit_add_to_load_queue(Unit *u) {
490 assert(u);
491 assert(u->type != _UNIT_TYPE_INVALID);
492
493 if (u->load_state != UNIT_STUB || u->in_load_queue)
494 return;
495
496 LIST_PREPEND(load_queue, u->manager->load_queue, u);
497 u->in_load_queue = true;
498 }
499
500 void unit_add_to_cleanup_queue(Unit *u) {
501 assert(u);
502
503 if (u->in_cleanup_queue)
504 return;
505
506 LIST_PREPEND(cleanup_queue, u->manager->cleanup_queue, u);
507 u->in_cleanup_queue = true;
508 }
509
510 void unit_add_to_gc_queue(Unit *u) {
511 assert(u);
512
513 if (u->in_gc_queue || u->in_cleanup_queue)
514 return;
515
516 if (!unit_may_gc(u))
517 return;
518
519 LIST_PREPEND(gc_queue, u->manager->gc_unit_queue, u);
520 u->in_gc_queue = true;
521 }
522
523 void unit_add_to_dbus_queue(Unit *u) {
524 assert(u);
525 assert(u->type != _UNIT_TYPE_INVALID);
526
527 if (u->load_state == UNIT_STUB || u->in_dbus_queue)
528 return;
529
530 /* Shortcut things if nobody cares */
531 if (sd_bus_track_count(u->manager->subscribed) <= 0 &&
532 sd_bus_track_count(u->bus_track) <= 0 &&
533 set_isempty(u->manager->private_buses)) {
534 u->sent_dbus_new_signal = true;
535 return;
536 }
537
538 LIST_PREPEND(dbus_queue, u->manager->dbus_unit_queue, u);
539 u->in_dbus_queue = true;
540 }
541
542 void unit_submit_to_stop_when_unneeded_queue(Unit *u) {
543 assert(u);
544
545 if (u->in_stop_when_unneeded_queue)
546 return;
547
548 if (!u->stop_when_unneeded)
549 return;
550
551 if (!UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u)))
552 return;
553
554 LIST_PREPEND(stop_when_unneeded_queue, u->manager->stop_when_unneeded_queue, u);
555 u->in_stop_when_unneeded_queue = true;
556 }
557
558 void unit_submit_to_start_when_upheld_queue(Unit *u) {
559 assert(u);
560
561 if (u->in_start_when_upheld_queue)
562 return;
563
564 if (!UNIT_IS_INACTIVE_OR_FAILED(unit_active_state(u)))
565 return;
566
567 if (!unit_has_dependency(u, UNIT_ATOM_START_STEADILY, NULL))
568 return;
569
570 LIST_PREPEND(start_when_upheld_queue, u->manager->start_when_upheld_queue, u);
571 u->in_start_when_upheld_queue = true;
572 }
573
574 void unit_submit_to_stop_when_bound_queue(Unit *u) {
575 assert(u);
576
577 if (u->in_stop_when_bound_queue)
578 return;
579
580 if (!UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u)))
581 return;
582
583 if (!unit_has_dependency(u, UNIT_ATOM_CANNOT_BE_ACTIVE_WITHOUT, NULL))
584 return;
585
586 LIST_PREPEND(stop_when_bound_queue, u->manager->stop_when_bound_queue, u);
587 u->in_stop_when_bound_queue = true;
588 }
589
590 static bool unit_can_release_resources(Unit *u) {
591 ExecContext *ec;
592
593 assert(u);
594
595 if (UNIT_VTABLE(u)->release_resources)
596 return true;
597
598 ec = unit_get_exec_context(u);
599 if (ec && ec->runtime_directory_preserve_mode == EXEC_PRESERVE_RESTART)
600 return true;
601
602 return false;
603 }
604
605 void unit_submit_to_release_resources_queue(Unit *u) {
606 assert(u);
607
608 if (u->in_release_resources_queue)
609 return;
610
611 if (u->job || u->nop_job)
612 return;
613
614 if (u->perpetual)
615 return;
616
617 if (!unit_can_release_resources(u))
618 return;
619
620 LIST_PREPEND(release_resources_queue, u->manager->release_resources_queue, u);
621 u->in_release_resources_queue = true;
622 }
623
624 void unit_add_to_stop_notify_queue(Unit *u) {
625 assert(u);
626
627 if (u->in_stop_notify_queue)
628 return;
629
630 assert(UNIT_VTABLE(u)->stop_notify);
631
632 LIST_PREPEND(stop_notify_queue, u->manager->stop_notify_queue, u);
633 u->in_stop_notify_queue = true;
634 }
635
636 void unit_remove_from_stop_notify_queue(Unit *u) {
637 assert(u);
638
639 if (!u->in_stop_notify_queue)
640 return;
641
642 LIST_REMOVE(stop_notify_queue, u->manager->stop_notify_queue, u);
643 u->in_stop_notify_queue = false;
644 }
645
646 static void unit_clear_dependencies(Unit *u) {
647 assert(u);
648
649 /* Removes all dependencies configured on u and their reverse dependencies. */
650
651 for (Hashmap *deps; (deps = hashmap_steal_first(u->dependencies));) {
652
653 for (Unit *other; (other = hashmap_steal_first_key(deps));) {
654 Hashmap *other_deps;
655
656 HASHMAP_FOREACH(other_deps, other->dependencies)
657 hashmap_remove(other_deps, u);
658
659 unit_add_to_gc_queue(other);
660 other->dependency_generation++;
661 }
662
663 hashmap_free(deps);
664 }
665
666 u->dependencies = hashmap_free(u->dependencies);
667 u->dependency_generation++;
668 }
669
670 static void unit_remove_transient(Unit *u) {
671 assert(u);
672 assert(u->manager);
673
674 if (!u->transient)
675 return;
676
677 STRV_FOREACH(i, u->dropin_paths) {
678 _cleanup_free_ char *p = NULL, *pp = NULL;
679
680 if (path_extract_directory(*i, &p) < 0) /* Get the drop-in directory from the drop-in file */
681 continue;
682
683 if (path_extract_directory(p, &pp) < 0) /* Get the config directory from the drop-in directory */
684 continue;
685
686 /* Only drop transient drop-ins */
687 if (!path_equal(u->manager->lookup_paths.transient, pp))
688 continue;
689
690 (void) unlink(*i);
691 (void) rmdir(p);
692 }
693
694 if (u->fragment_path) {
695 (void) unlink(u->fragment_path);
696 (void) unit_file_remove_from_name_map(
697 &u->manager->lookup_paths,
698 &u->manager->unit_cache_timestamp_hash,
699 &u->manager->unit_id_map,
700 &u->manager->unit_name_map,
701 &u->manager->unit_path_cache,
702 u->fragment_path);
703 }
704 }
705
706 static void unit_free_mounts_for(Unit *u) {
707 assert(u);
708
709 for (UnitMountDependencyType t = 0; t < _UNIT_MOUNT_DEPENDENCY_TYPE_MAX; ++t) {
710 for (;;) {
711 _cleanup_free_ char *path = NULL;
712
713 path = hashmap_steal_first_key(u->mounts_for[t]);
714 if (!path)
715 break;
716
717 char s[strlen(path) + 1];
718
719 PATH_FOREACH_PREFIX_MORE(s, path) {
720 char *y;
721 Set *x;
722
723 x = hashmap_get2(u->manager->units_needing_mounts_for[t], s, (void**) &y);
724 if (!x)
725 continue;
726
727 (void) set_remove(x, u);
728
729 if (set_isempty(x)) {
730 assert_se(hashmap_remove(u->manager->units_needing_mounts_for[t], y));
731 free(y);
732 set_free(x);
733 }
734 }
735 }
736
737 u->mounts_for[t] = hashmap_free(u->mounts_for[t]);
738 }
739 }
740
741 static void unit_done(Unit *u) {
742 ExecContext *ec;
743 CGroupContext *cc;
744
745 assert(u);
746
747 if (u->type < 0)
748 return;
749
750 if (UNIT_VTABLE(u)->done)
751 UNIT_VTABLE(u)->done(u);
752
753 ec = unit_get_exec_context(u);
754 if (ec)
755 exec_context_done(ec);
756
757 cc = unit_get_cgroup_context(u);
758 if (cc)
759 cgroup_context_done(cc);
760 }
761
762 Unit* unit_free(Unit *u) {
763 Unit *slice;
764 char *t;
765
766 if (!u)
767 return NULL;
768
769 sd_event_source_disable_unref(u->auto_start_stop_event_source);
770
771 u->transient_file = safe_fclose(u->transient_file);
772
773 if (!MANAGER_IS_RELOADING(u->manager))
774 unit_remove_transient(u);
775
776 bus_unit_send_removed_signal(u);
777
778 unit_done(u);
779
780 u->match_bus_slot = sd_bus_slot_unref(u->match_bus_slot);
781 u->bus_track = sd_bus_track_unref(u->bus_track);
782 u->deserialized_refs = strv_free(u->deserialized_refs);
783 u->pending_freezer_invocation = sd_bus_message_unref(u->pending_freezer_invocation);
784
785 unit_free_mounts_for(u);
786
787 SET_FOREACH(t, u->aliases)
788 hashmap_remove_value(u->manager->units, t, u);
789 if (u->id)
790 hashmap_remove_value(u->manager->units, u->id, u);
791
792 if (!sd_id128_is_null(u->invocation_id))
793 hashmap_remove_value(u->manager->units_by_invocation_id, &u->invocation_id, u);
794
795 if (u->job) {
796 Job *j = u->job;
797 job_uninstall(j);
798 job_free(j);
799 }
800
801 if (u->nop_job) {
802 Job *j = u->nop_job;
803 job_uninstall(j);
804 job_free(j);
805 }
806
807 /* A unit is being dropped from the tree, make sure our family is realized properly. Do this after we
808 * detach the unit from slice tree in order to eliminate its effect on controller masks. */
809 slice = UNIT_GET_SLICE(u);
810 unit_clear_dependencies(u);
811 if (slice)
812 unit_add_family_to_cgroup_realize_queue(slice);
813
814 if (u->on_console)
815 manager_unref_console(u->manager);
816
817 unit_release_cgroup(u, /* drop_cgroup_runtime = */ true);
818
819 if (!MANAGER_IS_RELOADING(u->manager))
820 unit_unlink_state_files(u);
821
822 unit_unref_uid_gid(u, false);
823
824 (void) manager_update_failed_units(u->manager, u, false);
825 set_remove(u->manager->startup_units, u);
826
827 unit_unwatch_all_pids(u);
828
829 while (u->refs_by_target)
830 unit_ref_unset(u->refs_by_target);
831
832 if (u->type != _UNIT_TYPE_INVALID)
833 LIST_REMOVE(units_by_type, u->manager->units_by_type[u->type], u);
834
835 if (u->in_load_queue)
836 LIST_REMOVE(load_queue, u->manager->load_queue, u);
837
838 if (u->in_dbus_queue)
839 LIST_REMOVE(dbus_queue, u->manager->dbus_unit_queue, u);
840
841 if (u->in_cleanup_queue)
842 LIST_REMOVE(cleanup_queue, u->manager->cleanup_queue, u);
843
844 if (u->in_gc_queue)
845 LIST_REMOVE(gc_queue, u->manager->gc_unit_queue, u);
846
847 if (u->in_cgroup_realize_queue)
848 LIST_REMOVE(cgroup_realize_queue, u->manager->cgroup_realize_queue, u);
849
850 if (u->in_cgroup_empty_queue)
851 LIST_REMOVE(cgroup_empty_queue, u->manager->cgroup_empty_queue, u);
852
853 if (u->in_cgroup_oom_queue)
854 LIST_REMOVE(cgroup_oom_queue, u->manager->cgroup_oom_queue, u);
855
856 if (u->in_target_deps_queue)
857 LIST_REMOVE(target_deps_queue, u->manager->target_deps_queue, u);
858
859 if (u->in_stop_when_unneeded_queue)
860 LIST_REMOVE(stop_when_unneeded_queue, u->manager->stop_when_unneeded_queue, u);
861
862 if (u->in_start_when_upheld_queue)
863 LIST_REMOVE(start_when_upheld_queue, u->manager->start_when_upheld_queue, u);
864
865 if (u->in_stop_when_bound_queue)
866 LIST_REMOVE(stop_when_bound_queue, u->manager->stop_when_bound_queue, u);
867
868 if (u->in_release_resources_queue)
869 LIST_REMOVE(release_resources_queue, u->manager->release_resources_queue, u);
870
871 unit_remove_from_stop_notify_queue(u);
872
873 condition_free_list(u->conditions);
874 condition_free_list(u->asserts);
875
876 free(u->description);
877 strv_free(u->documentation);
878 free(u->fragment_path);
879 free(u->source_path);
880 strv_free(u->dropin_paths);
881 free(u->instance);
882
883 free(u->job_timeout_reboot_arg);
884 free(u->reboot_arg);
885
886 free(u->access_selinux_context);
887
888 set_free(u->aliases);
889 free(u->id);
890
891 activation_details_unref(u->activation_details);
892
893 return mfree(u);
894 }
895
896 UnitActiveState unit_active_state(Unit *u) {
897 assert(u);
898
899 if (u->load_state == UNIT_MERGED)
900 return unit_active_state(unit_follow_merge(u));
901
902 /* After a reload it might happen that a unit is not correctly
903 * loaded but still has a process around. That's why we won't
904 * shortcut failed loading to UNIT_INACTIVE_FAILED. */
905
906 return UNIT_VTABLE(u)->active_state(u);
907 }
908
909 const char* unit_sub_state_to_string(Unit *u) {
910 assert(u);
911
912 return UNIT_VTABLE(u)->sub_state_to_string(u);
913 }
914
915 static int unit_merge_names(Unit *u, Unit *other) {
916 char *name;
917 int r;
918
919 assert(u);
920 assert(other);
921
922 r = unit_add_alias(u, other->id);
923 if (r < 0)
924 return r;
925
926 r = set_move(u->aliases, other->aliases);
927 if (r < 0) {
928 set_remove(u->aliases, other->id);
929 return r;
930 }
931
932 TAKE_PTR(other->id);
933 other->aliases = set_free(other->aliases);
934
935 SET_FOREACH(name, u->aliases)
936 assert_se(hashmap_replace(u->manager->units, name, u) == 0);
937
938 return 0;
939 }
940
941 static int unit_reserve_dependencies(Unit *u, Unit *other) {
942 size_t n_reserve;
943 Hashmap* deps;
944 void *d;
945 int r;
946
947 assert(u);
948 assert(other);
949
950 /* Let's reserve some space in the dependency hashmaps so that later on merging the units cannot
951 * fail.
952 *
953 * First make some room in the per dependency type hashmaps. Using the summed size of both units'
954 * hashmaps is an estimate that is likely too high since they probably use some of the same
955 * types. But it's never too low, and that's all we need. */
956
957 n_reserve = MIN(hashmap_size(other->dependencies), LESS_BY((size_t) _UNIT_DEPENDENCY_MAX, hashmap_size(u->dependencies)));
958 if (n_reserve > 0) {
959 r = hashmap_ensure_allocated(&u->dependencies, NULL);
960 if (r < 0)
961 return r;
962
963 r = hashmap_reserve(u->dependencies, n_reserve);
964 if (r < 0)
965 return r;
966 }
967
968 /* Now, enlarge our per dependency type hashmaps by the number of entries in the same hashmap of the
969 * other unit's dependencies.
970 *
971 * NB: If u does not have a dependency set allocated for some dependency type, there is no need to
972 * reserve anything for. In that case other's set will be transferred as a whole to u by
973 * complete_move(). */
974
975 HASHMAP_FOREACH_KEY(deps, d, u->dependencies) {
976 Hashmap *other_deps;
977
978 other_deps = hashmap_get(other->dependencies, d);
979
980 r = hashmap_reserve(deps, hashmap_size(other_deps));
981 if (r < 0)
982 return r;
983 }
984
985 return 0;
986 }
987
988 static bool unit_should_warn_about_dependency(UnitDependency dependency) {
989 /* Only warn about some unit types */
990 return IN_SET(dependency,
991 UNIT_CONFLICTS,
992 UNIT_CONFLICTED_BY,
993 UNIT_BEFORE,
994 UNIT_AFTER,
995 UNIT_ON_SUCCESS,
996 UNIT_ON_FAILURE,
997 UNIT_TRIGGERS,
998 UNIT_TRIGGERED_BY);
999 }
1000
1001 static int unit_per_dependency_type_hashmap_update(
1002 Hashmap *per_type,
1003 Unit *other,
1004 UnitDependencyMask origin_mask,
1005 UnitDependencyMask destination_mask) {
1006
1007 UnitDependencyInfo info;
1008 int r;
1009
1010 assert(other);
1011 assert_cc(sizeof(void*) == sizeof(info));
1012
1013 /* Acquire the UnitDependencyInfo entry for the Unit* we are interested in, and update it if it
1014 * exists, or insert it anew if not. */
1015
1016 info.data = hashmap_get(per_type, other);
1017 if (info.data) {
1018 /* Entry already exists. Add in our mask. */
1019
1020 if (FLAGS_SET(origin_mask, info.origin_mask) &&
1021 FLAGS_SET(destination_mask, info.destination_mask))
1022 return 0; /* NOP */
1023
1024 info.origin_mask |= origin_mask;
1025 info.destination_mask |= destination_mask;
1026
1027 r = hashmap_update(per_type, other, info.data);
1028 } else {
1029 info = (UnitDependencyInfo) {
1030 .origin_mask = origin_mask,
1031 .destination_mask = destination_mask,
1032 };
1033
1034 r = hashmap_put(per_type, other, info.data);
1035 }
1036 if (r < 0)
1037 return r;
1038
1039 return 1;
1040 }
1041
1042 static void unit_merge_dependencies(Unit *u, Unit *other) {
1043 Hashmap *deps;
1044 void *dt; /* Actually of type UnitDependency, except that we don't bother casting it here,
1045 * since the hashmaps all want it as void pointer. */
1046
1047 assert(u);
1048 assert(other);
1049
1050 if (u == other)
1051 return;
1052
1053 /* First, remove dependency to other. */
1054 HASHMAP_FOREACH_KEY(deps, dt, u->dependencies) {
1055 if (hashmap_remove(deps, other) && unit_should_warn_about_dependency(UNIT_DEPENDENCY_FROM_PTR(dt)))
1056 log_unit_warning(u, "Dependency %s=%s is dropped, as %s is merged into %s.",
1057 unit_dependency_to_string(UNIT_DEPENDENCY_FROM_PTR(dt)),
1058 other->id, other->id, u->id);
1059
1060 if (hashmap_isempty(deps))
1061 hashmap_free(hashmap_remove(u->dependencies, dt));
1062 }
1063
1064 for (;;) {
1065 _cleanup_hashmap_free_ Hashmap *other_deps = NULL;
1066 UnitDependencyInfo di_back;
1067 Unit *back;
1068
1069 /* Let's focus on one dependency type at a time, that 'other' has defined. */
1070 other_deps = hashmap_steal_first_key_and_value(other->dependencies, &dt);
1071 if (!other_deps)
1072 break; /* done! */
1073
1074 deps = hashmap_get(u->dependencies, dt);
1075
1076 /* Now iterate through all dependencies of this dependency type, of 'other'. We refer to the
1077 * referenced units as 'back'. */
1078 HASHMAP_FOREACH_KEY(di_back.data, back, other_deps) {
1079 Hashmap *back_deps;
1080 void *back_dt;
1081
1082 if (back == u) {
1083 /* This is a dependency pointing back to the unit we want to merge with?
1084 * Suppress it (but warn) */
1085 if (unit_should_warn_about_dependency(UNIT_DEPENDENCY_FROM_PTR(dt)))
1086 log_unit_warning(u, "Dependency %s=%s in %s is dropped, as %s is merged into %s.",
1087 unit_dependency_to_string(UNIT_DEPENDENCY_FROM_PTR(dt)),
1088 u->id, other->id, other->id, u->id);
1089
1090 hashmap_remove(other_deps, back);
1091 continue;
1092 }
1093
1094 /* Now iterate through all deps of 'back', and fix the ones pointing to 'other' to
1095 * point to 'u' instead. */
1096 HASHMAP_FOREACH_KEY(back_deps, back_dt, back->dependencies) {
1097 UnitDependencyInfo di_move;
1098
1099 di_move.data = hashmap_remove(back_deps, other);
1100 if (!di_move.data)
1101 continue;
1102
1103 assert_se(unit_per_dependency_type_hashmap_update(
1104 back_deps,
1105 u,
1106 di_move.origin_mask,
1107 di_move.destination_mask) >= 0);
1108 }
1109
1110 /* The target unit already has dependencies of this type, let's then merge this individually. */
1111 if (deps)
1112 assert_se(unit_per_dependency_type_hashmap_update(
1113 deps,
1114 back,
1115 di_back.origin_mask,
1116 di_back.destination_mask) >= 0);
1117 }
1118
1119 /* Now all references towards 'other' of the current type 'dt' are corrected to point to 'u'.
1120 * Lets's now move the deps of type 'dt' from 'other' to 'u'. If the unit does not have
1121 * dependencies of this type, let's move them per type wholesale. */
1122 if (!deps)
1123 assert_se(hashmap_put(u->dependencies, dt, TAKE_PTR(other_deps)) >= 0);
1124 }
1125
1126 other->dependencies = hashmap_free(other->dependencies);
1127
1128 u->dependency_generation++;
1129 other->dependency_generation++;
1130 }
1131
1132 int unit_merge(Unit *u, Unit *other) {
1133 int r;
1134
1135 assert(u);
1136 assert(other);
1137 assert(u->manager == other->manager);
1138 assert(u->type != _UNIT_TYPE_INVALID);
1139
1140 other = unit_follow_merge(other);
1141
1142 if (other == u)
1143 return 0;
1144
1145 if (u->type != other->type)
1146 return -EINVAL;
1147
1148 if (!unit_type_may_alias(u->type)) /* Merging only applies to unit names that support aliases */
1149 return -EEXIST;
1150
1151 if (!IN_SET(other->load_state, UNIT_STUB, UNIT_NOT_FOUND))
1152 return -EEXIST;
1153
1154 if (!streq_ptr(u->instance, other->instance))
1155 return -EINVAL;
1156
1157 if (other->job)
1158 return -EEXIST;
1159
1160 if (other->nop_job)
1161 return -EEXIST;
1162
1163 if (!UNIT_IS_INACTIVE_OR_FAILED(unit_active_state(other)))
1164 return -EEXIST;
1165
1166 /* Make reservations to ensure merge_dependencies() won't fail. We don't rollback reservations if we
1167 * fail. We don't have a way to undo reservations. A reservation is not a leak. */
1168 r = unit_reserve_dependencies(u, other);
1169 if (r < 0)
1170 return r;
1171
1172 /* Redirect all references */
1173 while (other->refs_by_target)
1174 unit_ref_set(other->refs_by_target, other->refs_by_target->source, u);
1175
1176 /* Merge dependencies */
1177 unit_merge_dependencies(u, other);
1178
1179 /* Merge names. It is better to do that after merging deps, otherwise the log message contains n/a. */
1180 r = unit_merge_names(u, other);
1181 if (r < 0)
1182 return r;
1183
1184 other->load_state = UNIT_MERGED;
1185 other->merged_into = u;
1186
1187 if (!u->activation_details)
1188 u->activation_details = activation_details_ref(other->activation_details);
1189
1190 /* If there is still some data attached to the other node, we
1191 * don't need it anymore, and can free it. */
1192 if (other->load_state != UNIT_STUB)
1193 if (UNIT_VTABLE(other)->done)
1194 UNIT_VTABLE(other)->done(other);
1195
1196 unit_add_to_dbus_queue(u);
1197 unit_add_to_cleanup_queue(other);
1198
1199 return 0;
1200 }
1201
1202 int unit_merge_by_name(Unit *u, const char *name) {
1203 _cleanup_free_ char *s = NULL;
1204 Unit *other;
1205 int r;
1206
1207 /* Either add name to u, or if a unit with name already exists, merge it with u.
1208 * If name is a template, do the same for name@instance, where instance is u's instance. */
1209
1210 assert(u);
1211 assert(name);
1212
1213 if (unit_name_is_valid(name, UNIT_NAME_TEMPLATE)) {
1214 if (!u->instance)
1215 return -EINVAL;
1216
1217 r = unit_name_replace_instance(name, u->instance, &s);
1218 if (r < 0)
1219 return r;
1220
1221 name = s;
1222 }
1223
1224 other = manager_get_unit(u->manager, name);
1225 if (other)
1226 return unit_merge(u, other);
1227
1228 return unit_add_name(u, name);
1229 }
1230
1231 Unit* unit_follow_merge(Unit *u) {
1232 assert(u);
1233
1234 while (u->load_state == UNIT_MERGED)
1235 assert_se(u = u->merged_into);
1236
1237 return u;
1238 }
1239
1240 int unit_add_exec_dependencies(Unit *u, ExecContext *c) {
1241 int r;
1242
1243 assert(u);
1244 assert(c);
1245
1246 /* Unlike unit_add_dependency() or friends, this always returns 0 on success. */
1247
1248 if (c->working_directory) {
1249 r = unit_add_mounts_for(
1250 u,
1251 c->working_directory,
1252 UNIT_DEPENDENCY_FILE,
1253 c->working_directory_missing_ok ? UNIT_MOUNT_WANTS : UNIT_MOUNT_REQUIRES);
1254 if (r < 0)
1255 return r;
1256 }
1257
1258 if (c->root_directory) {
1259 r = unit_add_mounts_for(u, c->root_directory, UNIT_DEPENDENCY_FILE, UNIT_MOUNT_WANTS);
1260 if (r < 0)
1261 return r;
1262 }
1263
1264 if (c->root_image) {
1265 r = unit_add_mounts_for(u, c->root_image, UNIT_DEPENDENCY_FILE, UNIT_MOUNT_WANTS);
1266 if (r < 0)
1267 return r;
1268 }
1269
1270 for (ExecDirectoryType dt = 0; dt < _EXEC_DIRECTORY_TYPE_MAX; dt++) {
1271 if (!u->manager->prefix[dt])
1272 continue;
1273
1274 FOREACH_ARRAY(i, c->directories[dt].items, c->directories[dt].n_items) {
1275 _cleanup_free_ char *p = NULL;
1276
1277 p = path_join(u->manager->prefix[dt], i->path);
1278 if (!p)
1279 return -ENOMEM;
1280
1281 r = unit_add_mounts_for(u, p, UNIT_DEPENDENCY_FILE, UNIT_MOUNT_REQUIRES);
1282 if (r < 0)
1283 return r;
1284 }
1285 }
1286
1287 if (!MANAGER_IS_SYSTEM(u->manager))
1288 return 0;
1289
1290 /* For the following three directory types we need write access, and /var/ is possibly on the root
1291 * fs. Hence order after systemd-remount-fs.service, to ensure things are writable. */
1292 if (c->directories[EXEC_DIRECTORY_STATE].n_items > 0 ||
1293 c->directories[EXEC_DIRECTORY_CACHE].n_items > 0 ||
1294 c->directories[EXEC_DIRECTORY_LOGS].n_items > 0) {
1295 r = unit_add_dependency_by_name(u, UNIT_AFTER, SPECIAL_REMOUNT_FS_SERVICE, true, UNIT_DEPENDENCY_FILE);
1296 if (r < 0)
1297 return r;
1298 }
1299
1300 /* This must be already set in unit_patch_contexts(). */
1301 assert(c->private_var_tmp >= 0 && c->private_var_tmp < _PRIVATE_TMP_MAX);
1302
1303 if (c->private_tmp == PRIVATE_TMP_CONNECTED) {
1304 assert(c->private_var_tmp == PRIVATE_TMP_CONNECTED);
1305
1306 r = unit_add_mounts_for(u, "/tmp/", UNIT_DEPENDENCY_FILE, UNIT_MOUNT_WANTS);
1307 if (r < 0)
1308 return r;
1309
1310 r = unit_add_mounts_for(u, "/var/tmp/", UNIT_DEPENDENCY_FILE, UNIT_MOUNT_WANTS);
1311 if (r < 0)
1312 return r;
1313
1314 r = unit_add_dependency_by_name(u, UNIT_AFTER, SPECIAL_TMPFILES_SETUP_SERVICE, true, UNIT_DEPENDENCY_FILE);
1315 if (r < 0)
1316 return r;
1317
1318 } else if (c->private_var_tmp == PRIVATE_TMP_DISCONNECTED && !exec_context_with_rootfs(c)) {
1319 /* Even if PrivateTmp=disconnected, we still require /var/tmp/ mountpoint to be present,
1320 * i.e. /var/ needs to be mounted. See comments in unit_patch_contexts(). */
1321 r = unit_add_mounts_for(u, "/var/", UNIT_DEPENDENCY_FILE, UNIT_MOUNT_WANTS);
1322 if (r < 0)
1323 return r;
1324 }
1325
1326 if (c->root_image) {
1327 /* We need to wait for /dev/loopX to appear when doing RootImage=, hence let's add an
1328 * implicit dependency on udev */
1329
1330 r = unit_add_dependency_by_name(u, UNIT_AFTER, SPECIAL_UDEVD_SERVICE, true, UNIT_DEPENDENCY_FILE);
1331 if (r < 0)
1332 return r;
1333 }
1334
1335 /* If syslog or kernel logging is requested (or log namespacing is), make sure our own logging daemon
1336 * is run first. */
1337 if (c->log_namespace) {
1338 static const struct {
1339 const char *template;
1340 UnitType type;
1341 } deps[] = {
1342 { "systemd-journald", UNIT_SOCKET, },
1343 { "systemd-journald-varlink", UNIT_SOCKET, },
1344 { "systemd-journald-sync", UNIT_SERVICE, },
1345 };
1346
1347 FOREACH_ELEMENT(i, deps) {
1348 _cleanup_free_ char *unit = NULL;
1349
1350 r = unit_name_build_from_type(i->template, c->log_namespace, i->type, &unit);
1351 if (r < 0)
1352 return r;
1353
1354 r = unit_add_two_dependencies_by_name(u, UNIT_AFTER, UNIT_REQUIRES, unit, true, UNIT_DEPENDENCY_FILE);
1355 if (r < 0)
1356 return r;
1357 }
1358 } else if (IN_SET(c->std_output, EXEC_OUTPUT_JOURNAL, EXEC_OUTPUT_JOURNAL_AND_CONSOLE,
1359 EXEC_OUTPUT_KMSG, EXEC_OUTPUT_KMSG_AND_CONSOLE) ||
1360 IN_SET(c->std_error, EXEC_OUTPUT_JOURNAL, EXEC_OUTPUT_JOURNAL_AND_CONSOLE,
1361 EXEC_OUTPUT_KMSG, EXEC_OUTPUT_KMSG_AND_CONSOLE)) {
1362
1363 r = unit_add_dependency_by_name(u, UNIT_AFTER, SPECIAL_JOURNALD_SOCKET, true, UNIT_DEPENDENCY_FILE);
1364 if (r < 0)
1365 return r;
1366 }
1367
1368 return 0;
1369 }
1370
1371 const char* unit_description(Unit *u) {
1372 assert(u);
1373
1374 if (u->description)
1375 return u->description;
1376
1377 return strna(u->id);
1378 }
1379
1380 const char* unit_status_string(Unit *u, char **ret_combined_buffer) {
1381 assert(u);
1382 assert(u->id);
1383
1384 /* Return u->id, u->description, or "{u->id} - {u->description}".
1385 * Versions with u->description are only used if it is set.
1386 * The last option is used if configured and the caller provided the 'ret_combined_buffer'
1387 * pointer.
1388 *
1389 * Note that *ret_combined_buffer may be set to NULL. */
1390
1391 if (!u->description ||
1392 u->manager->status_unit_format == STATUS_UNIT_FORMAT_NAME ||
1393 (u->manager->status_unit_format == STATUS_UNIT_FORMAT_COMBINED && !ret_combined_buffer) ||
1394 streq(u->description, u->id)) {
1395
1396 if (ret_combined_buffer)
1397 *ret_combined_buffer = NULL;
1398 return u->id;
1399 }
1400
1401 if (ret_combined_buffer) {
1402 if (u->manager->status_unit_format == STATUS_UNIT_FORMAT_COMBINED) {
1403 *ret_combined_buffer = strjoin(u->id, " - ", u->description);
1404 if (*ret_combined_buffer)
1405 return *ret_combined_buffer;
1406 log_oom(); /* Fall back to ->description */
1407 } else
1408 *ret_combined_buffer = NULL;
1409 }
1410
1411 return u->description;
1412 }
1413
1414 /* Common implementation for multiple backends */
1415 int unit_load_fragment_and_dropin(Unit *u, bool fragment_required) {
1416 int r;
1417
1418 assert(u);
1419
1420 /* Load a .{service,socket,...} file */
1421 r = unit_load_fragment(u);
1422 if (r < 0)
1423 return r;
1424
1425 if (u->load_state == UNIT_MASKED)
1426 return 0;
1427
1428 if (u->load_state == UNIT_STUB) {
1429 if (fragment_required)
1430 return -ENOENT;
1431
1432 u->load_state = UNIT_LOADED;
1433 }
1434
1435 u = unit_follow_merge(u);
1436
1437 /* Load drop-in directory data. If u is an alias, we might be reloading the
1438 * target unit needlessly. But we cannot be sure which drops-ins have already
1439 * been loaded and which not, at least without doing complicated book-keeping,
1440 * so let's always reread all drop-ins. */
1441 r = unit_load_dropin(u);
1442 if (r < 0)
1443 return r;
1444
1445 if (u->source_path) {
1446 struct stat st;
1447
1448 if (stat(u->source_path, &st) >= 0)
1449 u->source_mtime = timespec_load(&st.st_mtim);
1450 else
1451 u->source_mtime = 0;
1452 }
1453
1454 return 0;
1455 }
1456
1457 void unit_add_to_target_deps_queue(Unit *u) {
1458 Manager *m = ASSERT_PTR(ASSERT_PTR(u)->manager);
1459
1460 if (u->in_target_deps_queue)
1461 return;
1462
1463 LIST_PREPEND(target_deps_queue, m->target_deps_queue, u);
1464 u->in_target_deps_queue = true;
1465 }
1466
1467 int unit_add_default_target_dependency(Unit *u, Unit *target) {
1468 assert(u);
1469 assert(target);
1470
1471 if (target->type != UNIT_TARGET)
1472 return 0;
1473
1474 /* Only add the dependency if both units are loaded, so that
1475 * that loop check below is reliable */
1476 if (u->load_state != UNIT_LOADED ||
1477 target->load_state != UNIT_LOADED)
1478 return 0;
1479
1480 /* If either side wants no automatic dependencies, then let's
1481 * skip this */
1482 if (!u->default_dependencies ||
1483 !target->default_dependencies)
1484 return 0;
1485
1486 /* Don't create loops */
1487 if (unit_has_dependency(target, UNIT_ATOM_BEFORE, u))
1488 return 0;
1489
1490 return unit_add_dependency(target, UNIT_AFTER, u, true, UNIT_DEPENDENCY_DEFAULT);
1491 }
1492
1493 static int unit_add_slice_dependencies(Unit *u) {
1494 Unit *slice;
1495
1496 assert(u);
1497
1498 if (!UNIT_HAS_CGROUP_CONTEXT(u))
1499 return 0;
1500
1501 /* Slice units are implicitly ordered against their parent slices (as this relationship is encoded in the
1502 name), while all other units are ordered based on configuration (as in their case Slice= configures the
1503 relationship). */
1504 UnitDependencyMask mask = u->type == UNIT_SLICE ? UNIT_DEPENDENCY_IMPLICIT : UNIT_DEPENDENCY_FILE;
1505
1506 slice = UNIT_GET_SLICE(u);
1507 if (slice) {
1508 if (!IN_SET(slice->freezer_state, FREEZER_RUNNING, FREEZER_THAWING))
1509 u->freezer_state = FREEZER_FROZEN_BY_PARENT;
1510
1511 return unit_add_two_dependencies(u, UNIT_AFTER, UNIT_REQUIRES, slice, true, mask);
1512 }
1513
1514 if (unit_has_name(u, SPECIAL_ROOT_SLICE))
1515 return 0;
1516
1517 return unit_add_two_dependencies_by_name(u, UNIT_AFTER, UNIT_REQUIRES, SPECIAL_ROOT_SLICE, true, mask);
1518 }
1519
1520 static int unit_add_mount_dependencies(Unit *u) {
1521 bool changed = false;
1522 int r;
1523
1524 assert(u);
1525
1526 for (UnitMountDependencyType t = 0; t < _UNIT_MOUNT_DEPENDENCY_TYPE_MAX; ++t) {
1527 UnitDependencyInfo di;
1528 const char *path;
1529
1530 HASHMAP_FOREACH_KEY(di.data, path, u->mounts_for[t]) {
1531
1532 char prefix[strlen(ASSERT_PTR(path)) + 1];
1533
1534 PATH_FOREACH_PREFIX_MORE(prefix, path) {
1535 _cleanup_free_ char *p = NULL;
1536 Unit *m;
1537
1538 r = unit_name_from_path(prefix, ".mount", &p);
1539 if (r == -EINVAL)
1540 continue; /* If the path cannot be converted to a mount unit name,
1541 * then it's not manageable as a unit by systemd, and
1542 * hence we don't need a dependency on it. Let's thus
1543 * silently ignore the issue. */
1544 if (r < 0)
1545 return r;
1546
1547 m = manager_get_unit(u->manager, p);
1548 if (!m) {
1549 /* Make sure to load the mount unit if it exists. If so the
1550 * dependencies on this unit will be added later during the loading
1551 * of the mount unit. */
1552 (void) manager_load_unit_prepare(
1553 u->manager,
1554 p,
1555 /* path= */NULL,
1556 /* e= */NULL,
1557 &m);
1558 continue;
1559 }
1560 if (m == u)
1561 continue;
1562
1563 if (m->load_state != UNIT_LOADED)
1564 continue;
1565
1566 r = unit_add_dependency(
1567 u,
1568 UNIT_AFTER,
1569 m,
1570 /* add_reference= */ true,
1571 di.origin_mask);
1572 if (r < 0)
1573 return r;
1574 changed = changed || r > 0;
1575
1576 if (m->fragment_path) {
1577 r = unit_add_dependency(
1578 u,
1579 unit_mount_dependency_type_to_dependency_type(t),
1580 m,
1581 /* add_reference= */ true,
1582 di.origin_mask);
1583 if (r < 0)
1584 return r;
1585 changed = changed || r > 0;
1586 }
1587 }
1588 }
1589 }
1590
1591 return changed;
1592 }
1593
1594 static int unit_add_oomd_dependencies(Unit *u) {
1595 CGroupContext *c;
1596 CGroupMask mask;
1597 int r;
1598
1599 assert(u);
1600
1601 if (!u->default_dependencies)
1602 return 0;
1603
1604 c = unit_get_cgroup_context(u);
1605 if (!c)
1606 return 0;
1607
1608 bool wants_oomd = c->moom_swap == MANAGED_OOM_KILL || c->moom_mem_pressure == MANAGED_OOM_KILL;
1609 if (!wants_oomd)
1610 return 0;
1611
1612 r = cg_mask_supported(&mask);
1613 if (r < 0)
1614 return log_debug_errno(r, "Failed to determine supported controllers: %m");
1615
1616 if (!FLAGS_SET(mask, CGROUP_MASK_MEMORY))
1617 return 0;
1618
1619 return unit_add_two_dependencies_by_name(u, UNIT_AFTER, UNIT_WANTS, "systemd-oomd.service", true, UNIT_DEPENDENCY_FILE);
1620 }
1621
1622 static int unit_add_startup_units(Unit *u) {
1623 if (!unit_has_startup_cgroup_constraints(u))
1624 return 0;
1625
1626 return set_ensure_put(&u->manager->startup_units, NULL, u);
1627 }
1628
1629 static const struct {
1630 UnitDependencyAtom atom;
1631 size_t job_mode_offset;
1632 const char *dependency_name;
1633 const char *job_mode_setting_name;
1634 } on_termination_settings[] = {
1635 { UNIT_ATOM_ON_SUCCESS, offsetof(Unit, on_success_job_mode), "OnSuccess=", "OnSuccessJobMode=" },
1636 { UNIT_ATOM_ON_FAILURE, offsetof(Unit, on_failure_job_mode), "OnFailure=", "OnFailureJobMode=" },
1637 };
1638
1639 static int unit_validate_on_termination_job_modes(Unit *u) {
1640 assert(u);
1641
1642 /* Verify that if On{Success,Failure}JobMode=isolate, only one unit gets specified. */
1643
1644 FOREACH_ELEMENT(setting, on_termination_settings) {
1645 JobMode job_mode = *(JobMode*) ((uint8_t*) u + setting->job_mode_offset);
1646
1647 if (job_mode != JOB_ISOLATE)
1648 continue;
1649
1650 Unit *other, *found = NULL;
1651 UNIT_FOREACH_DEPENDENCY(other, u, setting->atom) {
1652 if (!found)
1653 found = other;
1654 else if (found != other)
1655 return log_unit_error_errno(u, SYNTHETIC_ERRNO(ENOEXEC),
1656 "More than one %s dependencies specified but %sisolate set. Refusing.",
1657 setting->dependency_name, setting->job_mode_setting_name);
1658 }
1659 }
1660
1661 return 0;
1662 }
1663
1664 int unit_load(Unit *u) {
1665 int r;
1666
1667 assert(u);
1668
1669 if (u->in_load_queue) {
1670 LIST_REMOVE(load_queue, u->manager->load_queue, u);
1671 u->in_load_queue = false;
1672 }
1673
1674 if (u->type == _UNIT_TYPE_INVALID)
1675 return -EINVAL;
1676
1677 if (u->load_state != UNIT_STUB)
1678 return 0;
1679
1680 if (u->transient_file) {
1681 /* Finalize transient file: if this is a transient unit file, as soon as we reach unit_load() the setup
1682 * is complete, hence let's synchronize the unit file we just wrote to disk. */
1683
1684 r = fflush_and_check(u->transient_file);
1685 if (r < 0)
1686 goto fail;
1687
1688 u->transient_file = safe_fclose(u->transient_file);
1689 u->fragment_mtime = now(CLOCK_REALTIME);
1690 }
1691
1692 r = UNIT_VTABLE(u)->load(u);
1693 if (r < 0)
1694 goto fail;
1695
1696 assert(u->load_state != UNIT_STUB);
1697
1698 if (u->load_state == UNIT_LOADED) {
1699 unit_add_to_target_deps_queue(u);
1700
1701 r = unit_add_slice_dependencies(u);
1702 if (r < 0)
1703 goto fail;
1704
1705 r = unit_add_mount_dependencies(u);
1706 if (r < 0)
1707 goto fail;
1708
1709 r = unit_add_oomd_dependencies(u);
1710 if (r < 0)
1711 goto fail;
1712
1713 r = unit_add_startup_units(u);
1714 if (r < 0)
1715 goto fail;
1716
1717 r = unit_validate_on_termination_job_modes(u);
1718 if (r < 0)
1719 goto fail;
1720
1721 if (u->job_running_timeout != USEC_INFINITY && u->job_running_timeout > u->job_timeout)
1722 log_unit_warning(u, "JobRunningTimeoutSec= is greater than JobTimeoutSec=, it has no effect.");
1723
1724 /* We finished loading, let's ensure our parents recalculate the members mask */
1725 unit_invalidate_cgroup_members_masks(u);
1726 }
1727
1728 assert((u->load_state != UNIT_MERGED) == !u->merged_into);
1729
1730 unit_add_to_dbus_queue(unit_follow_merge(u));
1731 unit_add_to_gc_queue(u);
1732 (void) manager_varlink_send_managed_oom_update(u);
1733
1734 return 0;
1735
1736 fail:
1737 /* We convert ENOEXEC errors to the UNIT_BAD_SETTING load state here. Configuration parsing code
1738 * should hence return ENOEXEC to ensure units are placed in this state after loading. */
1739
1740 u->load_state = u->load_state == UNIT_STUB ? UNIT_NOT_FOUND :
1741 r == -ENOEXEC ? UNIT_BAD_SETTING :
1742 UNIT_ERROR;
1743 u->load_error = r;
1744
1745 /* Record the timestamp on the cache, so that if the cache gets updated between now and the next time
1746 * an attempt is made to load this unit, we know we need to check again. */
1747 if (u->load_state == UNIT_NOT_FOUND)
1748 u->fragment_not_found_timestamp_hash = u->manager->unit_cache_timestamp_hash;
1749
1750 unit_add_to_dbus_queue(u);
1751 unit_add_to_gc_queue(u);
1752
1753 return log_unit_debug_errno(u, r, "Failed to load configuration: %m");
1754 }
1755
1756 _printf_(7, 8)
1757 static int log_unit_internal(void *userdata, int level, int error, const char *file, int line, const char *func, const char *format, ...) {
1758 Unit *u = userdata;
1759 va_list ap;
1760 int r;
1761
1762 if (u && !unit_log_level_test(u, level))
1763 return -ERRNO_VALUE(error);
1764
1765 va_start(ap, format);
1766 if (u)
1767 r = log_object_internalv(level, error, file, line, func,
1768 unit_log_field(u),
1769 u->id,
1770 unit_invocation_log_field(u),
1771 u->invocation_id_string,
1772 format, ap);
1773 else
1774 r = log_internalv(level, error, file, line, func, format, ap);
1775 va_end(ap);
1776
1777 return r;
1778 }
1779
1780 static bool unit_test_condition(Unit *u) {
1781 _cleanup_strv_free_ char **env = NULL;
1782 int r;
1783
1784 assert(u);
1785
1786 dual_timestamp_now(&u->condition_timestamp);
1787
1788 r = manager_get_effective_environment(u->manager, &env);
1789 if (r < 0) {
1790 log_unit_error_errno(u, r, "Failed to determine effective environment: %m");
1791 u->condition_result = true;
1792 } else
1793 u->condition_result = condition_test_list(
1794 u->conditions,
1795 env,
1796 condition_type_to_string,
1797 log_unit_internal,
1798 u);
1799
1800 unit_add_to_dbus_queue(u);
1801 return u->condition_result;
1802 }
1803
1804 static bool unit_test_assert(Unit *u) {
1805 _cleanup_strv_free_ char **env = NULL;
1806 int r;
1807
1808 assert(u);
1809
1810 dual_timestamp_now(&u->assert_timestamp);
1811
1812 r = manager_get_effective_environment(u->manager, &env);
1813 if (r < 0) {
1814 log_unit_error_errno(u, r, "Failed to determine effective environment: %m");
1815 u->assert_result = CONDITION_ERROR;
1816 } else
1817 u->assert_result = condition_test_list(
1818 u->asserts,
1819 env,
1820 assert_type_to_string,
1821 log_unit_internal,
1822 u);
1823
1824 unit_add_to_dbus_queue(u);
1825 return u->assert_result;
1826 }
1827
1828 void unit_status_printf(Unit *u, StatusType status_type, const char *status, const char *format, const char *ident) {
1829 if (log_get_show_color()) {
1830 if (u->manager->status_unit_format == STATUS_UNIT_FORMAT_COMBINED && strchr(ident, ' '))
1831 ident = strjoina(ANSI_HIGHLIGHT, u->id, ANSI_NORMAL, " - ", u->description);
1832 else
1833 ident = strjoina(ANSI_HIGHLIGHT, ident, ANSI_NORMAL);
1834 }
1835
1836 DISABLE_WARNING_FORMAT_NONLITERAL;
1837 manager_status_printf(u->manager, status_type, status, format, ident);
1838 REENABLE_WARNING;
1839 }
1840
1841 int unit_test_start_limit(Unit *u) {
1842 const char *reason;
1843
1844 assert(u);
1845
1846 if (ratelimit_below(&u->start_ratelimit)) {
1847 u->start_limit_hit = false;
1848 return 0;
1849 }
1850
1851 log_unit_warning(u, "Start request repeated too quickly.");
1852 u->start_limit_hit = true;
1853
1854 reason = strjoina("unit ", u->id, " failed");
1855
1856 emergency_action(
1857 u->manager,
1858 u->start_limit_action,
1859 EMERGENCY_ACTION_IS_WATCHDOG|EMERGENCY_ACTION_WARN|EMERGENCY_ACTION_SLEEP_5S,
1860 u->reboot_arg,
1861 /* exit_status= */ -1,
1862 reason);
1863
1864 return -ECANCELED;
1865 }
1866
1867 static bool unit_verify_deps(Unit *u) {
1868 Unit *other;
1869
1870 assert(u);
1871
1872 /* Checks whether all BindsTo= dependencies of this unit are fulfilled — if they are also combined
1873 * with After=. We do not check Requires= or Requisite= here as they only should have an effect on
1874 * the job processing, but do not have any effect afterwards. We don't check BindsTo= dependencies
1875 * that are not used in conjunction with After= as for them any such check would make things entirely
1876 * racy. */
1877
1878 UNIT_FOREACH_DEPENDENCY(other, u, UNIT_ATOM_CANNOT_BE_ACTIVE_WITHOUT) {
1879
1880 if (!unit_has_dependency(u, UNIT_ATOM_AFTER, other))
1881 continue;
1882
1883 if (!UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(other))) {
1884 log_unit_notice(u, "Bound to unit %s, but unit isn't active.", other->id);
1885 return false;
1886 }
1887 }
1888
1889 return true;
1890 }
1891
1892 /* Errors that aren't really errors:
1893 * -EALREADY: Unit is already started.
1894 * -ECOMM: Condition failed
1895 * -EAGAIN: An operation is already in progress. Retry later.
1896 *
1897 * Errors that are real errors:
1898 * -EBADR: This unit type does not support starting.
1899 * -ECANCELED: Start limit hit, too many requests for now
1900 * -EPROTO: Assert failed
1901 * -EINVAL: Unit not loaded
1902 * -EOPNOTSUPP: Unit type not supported
1903 * -ENOLINK: The necessary dependencies are not fulfilled.
1904 * -ESTALE: This unit has been started before and can't be started a second time
1905 * -EDEADLK: This unit is frozen
1906 * -ENOENT: This is a triggering unit and unit to trigger is not loaded
1907 * -ETOOMANYREFS: The hard concurrency limit of at least one of the slices the unit is contained in has been reached
1908 */
1909 int unit_start(Unit *u, ActivationDetails *details) {
1910 UnitActiveState state;
1911 Unit *following;
1912 int r;
1913
1914 assert(u);
1915
1916 /* Let's hold off running start jobs for mount units when /proc/self/mountinfo monitor is ratelimited. */
1917 if (UNIT_VTABLE(u)->subsystem_ratelimited) {
1918 r = UNIT_VTABLE(u)->subsystem_ratelimited(u->manager);
1919 if (r < 0)
1920 return r;
1921 if (r > 0)
1922 return -EAGAIN;
1923 }
1924
1925 /* If this is already started, then this will succeed. Note that this will even succeed if this unit
1926 * is not startable by the user. This is relied on to detect when we need to wait for units and when
1927 * waiting is finished. */
1928 state = unit_active_state(u);
1929 if (UNIT_IS_ACTIVE_OR_RELOADING(state))
1930 return -EALREADY;
1931 if (state == UNIT_MAINTENANCE)
1932 return -EAGAIN;
1933
1934 /* Units that aren't loaded cannot be started */
1935 if (u->load_state != UNIT_LOADED)
1936 return -EINVAL;
1937
1938 /* Refuse starting scope units more than once */
1939 if (UNIT_VTABLE(u)->once_only && dual_timestamp_is_set(&u->inactive_enter_timestamp))
1940 return -ESTALE;
1941
1942 /* If the conditions were unmet, don't do anything at all. If we already are activating this call might
1943 * still be useful to speed up activation in case there is some hold-off time, but we don't want to
1944 * recheck the condition in that case. */
1945 if (state != UNIT_ACTIVATING &&
1946 !unit_test_condition(u))
1947 return log_unit_debug_errno(u, SYNTHETIC_ERRNO(ECOMM), "Starting requested but condition not met. Not starting unit.");
1948
1949 /* If the asserts failed, fail the entire job */
1950 if (state != UNIT_ACTIVATING &&
1951 !unit_test_assert(u))
1952 return log_unit_notice_errno(u, SYNTHETIC_ERRNO(EPROTO), "Starting requested but asserts failed.");
1953
1954 /* Units of types that aren't supported cannot be started. Note that we do this test only after the
1955 * condition checks, so that we rather return condition check errors (which are usually not
1956 * considered a true failure) than "not supported" errors (which are considered a failure).
1957 */
1958 if (!unit_type_supported(u->type))
1959 return -EOPNOTSUPP;
1960
1961 /* Let's make sure that the deps really are in order before we start this. Normally the job engine
1962 * should have taken care of this already, but let's check this here again. After all, our
1963 * dependencies might not be in effect anymore, due to a reload or due to an unmet condition. */
1964 if (!unit_verify_deps(u))
1965 return -ENOLINK;
1966
1967 /* Forward to the main object, if we aren't it. */
1968 following = unit_following(u);
1969 if (following) {
1970 log_unit_debug(u, "Redirecting start request from %s to %s.", u->id, following->id);
1971 return unit_start(following, details);
1972 }
1973
1974 /* Check to make sure the unit isn't frozen */
1975 if (u->freezer_state != FREEZER_RUNNING)
1976 return -EDEADLK;
1977
1978 /* Check our ability to start early so that failure conditions don't cause us to enter a busy loop. */
1979 if (UNIT_VTABLE(u)->can_start) {
1980 r = UNIT_VTABLE(u)->can_start(u);
1981 if (r < 0)
1982 return r;
1983 }
1984
1985 /* If it is stopped, but we cannot start it, then fail */
1986 if (!UNIT_VTABLE(u)->start)
1987 return -EBADR;
1988
1989 if (UNIT_IS_INACTIVE_OR_FAILED(state)) {
1990 Slice *slice = SLICE(UNIT_GET_SLICE(u));
1991
1992 if (slice) {
1993 /* Check hard concurrency limit. Note this is partially redundant, we already checked
1994 * this when enqueuing jobs. However, between the time when we enqueued this and the
1995 * time we are dispatching the queue the configuration might have changed, hence
1996 * check here again */
1997 if (slice_concurrency_hard_max_reached(slice, u))
1998 return -ETOOMANYREFS;
1999
2000 /* Also check soft concurrenty limit, and return EAGAIN so that the job is kept in
2001 * the queue */
2002 if (slice_concurrency_soft_max_reached(slice, u))
2003 return -EAGAIN; /* Try again, keep in queue */
2004 }
2005 }
2006
2007 /* We don't suppress calls to ->start() here when we are already starting, to allow this request to
2008 * be used as a "hurry up" call, for example when the unit is in some "auto restart" state where it
2009 * waits for a holdoff timer to elapse before it will start again. */
2010
2011 unit_add_to_dbus_queue(u);
2012
2013 if (!u->activation_details) /* Older details object wins */
2014 u->activation_details = activation_details_ref(details);
2015
2016 return UNIT_VTABLE(u)->start(u);
2017 }
2018
2019 bool unit_can_start(Unit *u) {
2020 assert(u);
2021
2022 if (u->load_state != UNIT_LOADED)
2023 return false;
2024
2025 if (!unit_type_supported(u->type))
2026 return false;
2027
2028 /* Scope units may be started only once */
2029 if (UNIT_VTABLE(u)->once_only && dual_timestamp_is_set(&u->inactive_exit_timestamp))
2030 return false;
2031
2032 return !!UNIT_VTABLE(u)->start;
2033 }
2034
2035 bool unit_can_isolate(Unit *u) {
2036 assert(u);
2037
2038 return unit_can_start(u) &&
2039 u->allow_isolate;
2040 }
2041
2042 /* Errors:
2043 * -EBADR: This unit type does not support stopping.
2044 * -EALREADY: Unit is already stopped.
2045 * -EAGAIN: An operation is already in progress. Retry later.
2046 * -EDEADLK: Unit is frozen
2047 */
2048 int unit_stop(Unit *u) {
2049 UnitActiveState state;
2050 Unit *following;
2051
2052 assert(u);
2053
2054 state = unit_active_state(u);
2055 if (UNIT_IS_INACTIVE_OR_FAILED(state))
2056 return -EALREADY;
2057
2058 following = unit_following(u);
2059 if (following) {
2060 log_unit_debug(u, "Redirecting stop request from %s to %s.", u->id, following->id);
2061 return unit_stop(following);
2062 }
2063
2064 /* Check to make sure the unit isn't frozen */
2065 if (u->freezer_state != FREEZER_RUNNING)
2066 return -EDEADLK;
2067
2068 if (!UNIT_VTABLE(u)->stop)
2069 return -EBADR;
2070
2071 unit_add_to_dbus_queue(u);
2072
2073 return UNIT_VTABLE(u)->stop(u);
2074 }
2075
2076 bool unit_can_stop(Unit *u) {
2077 assert(u);
2078
2079 /* Note: if we return true here, it does not mean that the unit may be successfully stopped.
2080 * Extrinsic units follow external state and they may stop following external state changes
2081 * (hence we return true here), but an attempt to do this through the manager will fail. */
2082
2083 if (!unit_type_supported(u->type))
2084 return false;
2085
2086 if (u->perpetual)
2087 return false;
2088
2089 return !!UNIT_VTABLE(u)->stop;
2090 }
2091
2092 /* Errors:
2093 * -EBADR: This unit type does not support reloading.
2094 * -ENOEXEC: Unit is not started.
2095 * -EAGAIN: An operation is already in progress. Retry later.
2096 * -EDEADLK: Unit is frozen.
2097 */
2098 int unit_reload(Unit *u) {
2099 UnitActiveState state;
2100 Unit *following;
2101
2102 assert(u);
2103
2104 if (u->load_state != UNIT_LOADED)
2105 return -EINVAL;
2106
2107 if (!unit_can_reload(u))
2108 return -EBADR;
2109
2110 state = unit_active_state(u);
2111 if (IN_SET(state, UNIT_RELOADING, UNIT_REFRESHING))
2112 /* "refreshing" means some resources in the unit namespace is being updated. Unlike reload,
2113 * the unit processes aren't made aware of refresh. Let's put the job back to queue
2114 * in both cases, as refresh typically takes place before reload and it's better to wait
2115 * for it rather than failing. */
2116 return -EAGAIN;
2117
2118 if (state != UNIT_ACTIVE)
2119 return log_unit_warning_errno(u, SYNTHETIC_ERRNO(ENOEXEC), "Unit cannot be reloaded because it is inactive.");
2120
2121 following = unit_following(u);
2122 if (following) {
2123 log_unit_debug(u, "Redirecting reload request from %s to %s.", u->id, following->id);
2124 return unit_reload(following);
2125 }
2126
2127 /* Check to make sure the unit isn't frozen */
2128 if (u->freezer_state != FREEZER_RUNNING)
2129 return -EDEADLK;
2130
2131 unit_add_to_dbus_queue(u);
2132
2133 if (!UNIT_VTABLE(u)->reload) {
2134 /* Unit doesn't have a reload function, but we need to propagate the reload anyway */
2135 unit_notify(u, unit_active_state(u), unit_active_state(u), /* reload_success = */ true);
2136 return 0;
2137 }
2138
2139 return UNIT_VTABLE(u)->reload(u);
2140 }
2141
2142 bool unit_can_reload(Unit *u) {
2143 assert(u);
2144
2145 if (UNIT_VTABLE(u)->can_reload)
2146 return UNIT_VTABLE(u)->can_reload(u);
2147
2148 if (unit_has_dependency(u, UNIT_ATOM_PROPAGATES_RELOAD_TO, NULL))
2149 return true;
2150
2151 return UNIT_VTABLE(u)->reload;
2152 }
2153
2154 bool unit_is_unneeded(Unit *u) {
2155 Unit *other;
2156 assert(u);
2157
2158 if (!u->stop_when_unneeded)
2159 return false;
2160
2161 /* Don't clean up while the unit is transitioning or is even inactive. */
2162 if (unit_active_state(u) != UNIT_ACTIVE)
2163 return false;
2164 if (u->job)
2165 return false;
2166
2167 UNIT_FOREACH_DEPENDENCY(other, u, UNIT_ATOM_PINS_STOP_WHEN_UNNEEDED) {
2168 /* If a dependent unit has a job queued, is active or transitioning, or is marked for
2169 * restart, then don't clean this one up. */
2170
2171 if (other->job)
2172 return false;
2173
2174 if (!UNIT_IS_INACTIVE_OR_FAILED(unit_active_state(other)))
2175 return false;
2176
2177 if (unit_will_restart(other))
2178 return false;
2179 }
2180
2181 return true;
2182 }
2183
2184 bool unit_is_upheld_by_active(Unit *u, Unit **ret_culprit) {
2185 Unit *other;
2186
2187 assert(u);
2188
2189 /* Checks if the unit needs to be started because it currently is not running, but some other unit
2190 * that is active declared an Uphold= dependencies on it */
2191
2192 if (!UNIT_IS_INACTIVE_OR_FAILED(unit_active_state(u)) || u->job) {
2193 if (ret_culprit)
2194 *ret_culprit = NULL;
2195 return false;
2196 }
2197
2198 UNIT_FOREACH_DEPENDENCY(other, u, UNIT_ATOM_START_STEADILY) {
2199 if (other->job)
2200 continue;
2201
2202 if (UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(other))) {
2203 if (ret_culprit)
2204 *ret_culprit = other;
2205 return true;
2206 }
2207 }
2208
2209 if (ret_culprit)
2210 *ret_culprit = NULL;
2211 return false;
2212 }
2213
2214 bool unit_is_bound_by_inactive(Unit *u, Unit **ret_culprit) {
2215 Unit *other;
2216
2217 assert(u);
2218
2219 /* Checks whether this unit is bound to another unit that is inactive, i.e. whether we should stop
2220 * because the other unit is down. */
2221
2222 if (unit_active_state(u) != UNIT_ACTIVE || u->job) {
2223 /* Don't clean up while the unit is transitioning or is even inactive. */
2224 if (ret_culprit)
2225 *ret_culprit = NULL;
2226 return false;
2227 }
2228
2229 UNIT_FOREACH_DEPENDENCY(other, u, UNIT_ATOM_CANNOT_BE_ACTIVE_WITHOUT) {
2230 if (other->job)
2231 continue;
2232
2233 if (UNIT_IS_INACTIVE_OR_FAILED(unit_active_state(other))) {
2234 if (ret_culprit)
2235 *ret_culprit = other;
2236
2237 return true;
2238 }
2239 }
2240
2241 if (ret_culprit)
2242 *ret_culprit = NULL;
2243 return false;
2244 }
2245
2246 static void check_unneeded_dependencies(Unit *u) {
2247 Unit *other;
2248 assert(u);
2249
2250 /* Add all units this unit depends on to the queue that processes StopWhenUnneeded= behaviour. */
2251
2252 UNIT_FOREACH_DEPENDENCY(other, u, UNIT_ATOM_ADD_STOP_WHEN_UNNEEDED_QUEUE)
2253 unit_submit_to_stop_when_unneeded_queue(other);
2254 }
2255
2256 static void check_uphold_dependencies(Unit *u) {
2257 Unit *other;
2258 assert(u);
2259
2260 /* Add all units this unit depends on to the queue that processes Uphold= behaviour. */
2261
2262 UNIT_FOREACH_DEPENDENCY(other, u, UNIT_ATOM_ADD_START_WHEN_UPHELD_QUEUE)
2263 unit_submit_to_start_when_upheld_queue(other);
2264 }
2265
2266 static void check_bound_by_dependencies(Unit *u) {
2267 Unit *other;
2268 assert(u);
2269
2270 /* Add all units this unit depends on to the queue that processes BindsTo= stop behaviour. */
2271
2272 UNIT_FOREACH_DEPENDENCY(other, u, UNIT_ATOM_ADD_CANNOT_BE_ACTIVE_WITHOUT_QUEUE)
2273 unit_submit_to_stop_when_bound_queue(other);
2274 }
2275
2276 static void retroactively_start_dependencies(Unit *u) {
2277 Unit *other;
2278
2279 assert(u);
2280 assert(UNIT_IS_ACTIVE_OR_ACTIVATING(unit_active_state(u)));
2281
2282 UNIT_FOREACH_DEPENDENCY_SAFE(other, u, UNIT_ATOM_RETROACTIVE_START_REPLACE) /* Requires= + BindsTo= */
2283 if (!unit_has_dependency(u, UNIT_ATOM_AFTER, other) &&
2284 !UNIT_IS_ACTIVE_OR_ACTIVATING(unit_active_state(other)))
2285 (void) manager_add_job(u->manager, JOB_START, other, JOB_REPLACE, /* error = */ NULL, /* ret = */ NULL);
2286
2287 UNIT_FOREACH_DEPENDENCY_SAFE(other, u, UNIT_ATOM_RETROACTIVE_START_FAIL) /* Wants= */
2288 if (!unit_has_dependency(u, UNIT_ATOM_AFTER, other) &&
2289 !UNIT_IS_ACTIVE_OR_ACTIVATING(unit_active_state(other)))
2290 (void) manager_add_job(u->manager, JOB_START, other, JOB_FAIL, /* error = */ NULL, /* ret = */ NULL);
2291
2292 UNIT_FOREACH_DEPENDENCY_SAFE(other, u, UNIT_ATOM_RETROACTIVE_STOP_ON_START) /* Conflicts= (and inverse) */
2293 if (!UNIT_IS_INACTIVE_OR_DEACTIVATING(unit_active_state(other)))
2294 (void) manager_add_job(u->manager, JOB_STOP, other, JOB_REPLACE, /* error = */ NULL, /* ret = */ NULL);
2295 }
2296
2297 static void retroactively_stop_dependencies(Unit *u) {
2298 Unit *other;
2299
2300 assert(u);
2301 assert(UNIT_IS_INACTIVE_OR_DEACTIVATING(unit_active_state(u)));
2302
2303 /* Pull down units which are bound to us recursively if enabled */
2304 UNIT_FOREACH_DEPENDENCY_SAFE(other, u, UNIT_ATOM_RETROACTIVE_STOP_ON_STOP) /* BoundBy= */
2305 if (!UNIT_IS_INACTIVE_OR_DEACTIVATING(unit_active_state(other)))
2306 (void) manager_add_job(u->manager, JOB_STOP, other, JOB_REPLACE, /* error = */ NULL, /* ret = */ NULL);
2307 }
2308
2309 void unit_start_on_termination_deps(Unit *u, UnitDependencyAtom atom) {
2310 const char *dependency_name = NULL;
2311 JobMode job_mode;
2312 unsigned n_jobs = 0;
2313 int r;
2314
2315 /* Act on OnFailure= and OnSuccess= dependencies */
2316
2317 assert(u);
2318 assert(u->manager);
2319 assert(IN_SET(atom, UNIT_ATOM_ON_SUCCESS, UNIT_ATOM_ON_FAILURE));
2320
2321 FOREACH_ELEMENT(setting, on_termination_settings)
2322 if (atom == setting->atom) {
2323 job_mode = *(JobMode*) ((uint8_t*) u + setting->job_mode_offset);
2324 dependency_name = setting->dependency_name;
2325 break;
2326 }
2327
2328 assert(dependency_name);
2329
2330 Unit *other;
2331 UNIT_FOREACH_DEPENDENCY_SAFE(other, u, atom) {
2332 _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL;
2333
2334 if (n_jobs == 0)
2335 log_unit_info(u, "Triggering %s dependencies.", dependency_name);
2336
2337 r = manager_add_job(u->manager, JOB_START, other, job_mode, &error, /* ret = */ NULL);
2338 if (r < 0)
2339 log_unit_warning_errno(u, r, "Failed to enqueue %s%s job, ignoring: %s",
2340 dependency_name, other->id, bus_error_message(&error, r));
2341 n_jobs++;
2342 }
2343
2344 if (n_jobs > 0)
2345 log_unit_debug(u, "Triggering %s dependencies done (%u %s).",
2346 dependency_name, n_jobs, n_jobs == 1 ? "job" : "jobs");
2347 }
2348
2349 void unit_trigger_notify(Unit *u) {
2350 Unit *other;
2351
2352 assert(u);
2353
2354 UNIT_FOREACH_DEPENDENCY_SAFE(other, u, UNIT_ATOM_TRIGGERED_BY)
2355 if (UNIT_VTABLE(other)->trigger_notify)
2356 UNIT_VTABLE(other)->trigger_notify(other, u);
2357 }
2358
2359 static int raise_level(int log_level, bool condition_info, bool condition_notice) {
2360 if (condition_notice && log_level > LOG_NOTICE)
2361 return LOG_NOTICE;
2362 if (condition_info && log_level > LOG_INFO)
2363 return LOG_INFO;
2364 return log_level;
2365 }
2366
2367 static int unit_log_resources(Unit *u) {
2368
2369 static const struct {
2370 const char *journal_field;
2371 const char *message_suffix;
2372 } memory_fields[_CGROUP_MEMORY_ACCOUNTING_METRIC_CACHED_LAST + 1] = {
2373 [CGROUP_MEMORY_PEAK] = { "MEMORY_PEAK", "memory peak" },
2374 [CGROUP_MEMORY_SWAP_PEAK] = { "MEMORY_SWAP_PEAK", "memory swap peak" },
2375 }, ip_fields[_CGROUP_IP_ACCOUNTING_METRIC_MAX] = {
2376 [CGROUP_IP_INGRESS_BYTES] = { "IP_METRIC_INGRESS_BYTES", "incoming IP traffic" },
2377 [CGROUP_IP_EGRESS_BYTES] = { "IP_METRIC_EGRESS_BYTES", "outgoing IP traffic" },
2378 [CGROUP_IP_INGRESS_PACKETS] = { "IP_METRIC_INGRESS_PACKETS", NULL },
2379 [CGROUP_IP_EGRESS_PACKETS] = { "IP_METRIC_EGRESS_PACKETS", NULL },
2380 }, io_fields[_CGROUP_IO_ACCOUNTING_METRIC_MAX] = {
2381 [CGROUP_IO_READ_BYTES] = { "IO_METRIC_READ_BYTES", "read from disk" },
2382 [CGROUP_IO_WRITE_BYTES] = { "IO_METRIC_WRITE_BYTES", "written to disk" },
2383 [CGROUP_IO_READ_OPERATIONS] = { "IO_METRIC_READ_OPERATIONS", NULL },
2384 [CGROUP_IO_WRITE_OPERATIONS] = { "IO_METRIC_WRITE_OPERATIONS", NULL },
2385 };
2386
2387 struct iovec *iovec = NULL;
2388 size_t n_iovec = 0;
2389 _cleanup_free_ char *message = NULL, *t = NULL;
2390 nsec_t cpu_nsec = NSEC_INFINITY;
2391 int log_level = LOG_DEBUG; /* May be raised if resources consumed over a threshold */
2392
2393 assert(u);
2394
2395 CLEANUP_ARRAY(iovec, n_iovec, iovec_array_free);
2396
2397 iovec = new(struct iovec, 1 + (_CGROUP_MEMORY_ACCOUNTING_METRIC_CACHED_LAST + 1) +
2398 _CGROUP_IP_ACCOUNTING_METRIC_MAX + _CGROUP_IO_ACCOUNTING_METRIC_MAX + 4);
2399 if (!iovec)
2400 return log_oom();
2401
2402 /* Invoked whenever a unit enters failed or dead state. Logs information about consumed resources if resource
2403 * accounting was enabled for a unit. It does this in two ways: a friendly human-readable string with reduced
2404 * information and the complete data in structured fields. */
2405
2406 (void) unit_get_cpu_usage(u, &cpu_nsec);
2407 if (cpu_nsec != NSEC_INFINITY) {
2408 /* Format the CPU time for inclusion in the structured log message */
2409 if (asprintf(&t, "CPU_USAGE_NSEC=%" PRIu64, cpu_nsec) < 0)
2410 return log_oom();
2411 iovec[n_iovec++] = IOVEC_MAKE_STRING(TAKE_PTR(t));
2412
2413 /* Format the CPU time for inclusion in the human language message string */
2414 if (dual_timestamp_is_set(&u->inactive_exit_timestamp) &&
2415 dual_timestamp_is_set(&u->inactive_enter_timestamp)) {
2416 usec_t wall_clock_usec = usec_sub_unsigned(u->inactive_enter_timestamp.monotonic, u->inactive_exit_timestamp.monotonic);
2417 if (strextendf_with_separator(&message, ", ",
2418 "Consumed %s CPU time over %s wall clock time",
2419 FORMAT_TIMESPAN(cpu_nsec / NSEC_PER_USEC, USEC_PER_MSEC),
2420 FORMAT_TIMESPAN(wall_clock_usec, USEC_PER_MSEC)) < 0)
2421 return log_oom();
2422 } else {
2423 if (strextendf_with_separator(&message, ", ",
2424 "Consumed %s CPU time",
2425 FORMAT_TIMESPAN(cpu_nsec / NSEC_PER_USEC, USEC_PER_MSEC)) < 0)
2426 return log_oom();
2427 }
2428
2429 log_level = raise_level(log_level,
2430 cpu_nsec > MENTIONWORTHY_CPU_NSEC,
2431 cpu_nsec > NOTICEWORTHY_CPU_NSEC);
2432 }
2433
2434 for (CGroupMemoryAccountingMetric metric = 0; metric <= _CGROUP_MEMORY_ACCOUNTING_METRIC_CACHED_LAST; metric++) {
2435 uint64_t value = UINT64_MAX;
2436
2437 assert(memory_fields[metric].journal_field);
2438 assert(memory_fields[metric].message_suffix);
2439
2440 (void) unit_get_memory_accounting(u, metric, &value);
2441 if (value == UINT64_MAX)
2442 continue;
2443
2444 if (asprintf(&t, "%s=%" PRIu64, memory_fields[metric].journal_field, value) < 0)
2445 return log_oom();
2446 iovec[n_iovec++] = IOVEC_MAKE_STRING(TAKE_PTR(t));
2447
2448 /* If value is 0, we don't log it in the MESSAGE= field. */
2449 if (value == 0)
2450 continue;
2451
2452 if (strextendf_with_separator(&message, ", ", "%s %s",
2453 FORMAT_BYTES(value), memory_fields[metric].message_suffix) < 0)
2454 return log_oom();
2455
2456 log_level = raise_level(log_level,
2457 value > MENTIONWORTHY_MEMORY_BYTES,
2458 value > NOTICEWORTHY_MEMORY_BYTES);
2459 }
2460
2461 for (CGroupIOAccountingMetric k = 0; k < _CGROUP_IO_ACCOUNTING_METRIC_MAX; k++) {
2462 uint64_t value = UINT64_MAX;
2463
2464 assert(io_fields[k].journal_field);
2465
2466 (void) unit_get_io_accounting(u, k, &value);
2467 if (value == UINT64_MAX)
2468 continue;
2469
2470 /* Format IO accounting data for inclusion in the structured log message */
2471 if (asprintf(&t, "%s=%" PRIu64, io_fields[k].journal_field, value) < 0)
2472 return log_oom();
2473 iovec[n_iovec++] = IOVEC_MAKE_STRING(TAKE_PTR(t));
2474
2475 /* If value is 0, we don't log it in the MESSAGE= field. */
2476 if (value == 0)
2477 continue;
2478
2479 /* Format the IO accounting data for inclusion in the human language message string, but only
2480 * for the bytes counters (and not for the operations counters) */
2481 if (io_fields[k].message_suffix) {
2482 if (strextendf_with_separator(&message, ", ", "%s %s",
2483 FORMAT_BYTES(value), io_fields[k].message_suffix) < 0)
2484 return log_oom();
2485
2486 log_level = raise_level(log_level,
2487 value > MENTIONWORTHY_IO_BYTES,
2488 value > NOTICEWORTHY_IO_BYTES);
2489 }
2490 }
2491
2492 for (CGroupIPAccountingMetric m = 0; m < _CGROUP_IP_ACCOUNTING_METRIC_MAX; m++) {
2493 uint64_t value = UINT64_MAX;
2494
2495 assert(ip_fields[m].journal_field);
2496
2497 (void) unit_get_ip_accounting(u, m, &value);
2498 if (value == UINT64_MAX)
2499 continue;
2500
2501 /* Format IP accounting data for inclusion in the structured log message */
2502 if (asprintf(&t, "%s=%" PRIu64, ip_fields[m].journal_field, value) < 0)
2503 return log_oom();
2504 iovec[n_iovec++] = IOVEC_MAKE_STRING(TAKE_PTR(t));
2505
2506 /* If value is 0, we don't log it in the MESSAGE= field. */
2507 if (value == 0)
2508 continue;
2509
2510 /* Format the IP accounting data for inclusion in the human language message string, but only
2511 * for the bytes counters (and not for the packets counters) */
2512 if (ip_fields[m].message_suffix) {
2513 if (strextendf_with_separator(&message, ", ", "%s %s",
2514 FORMAT_BYTES(value), ip_fields[m].message_suffix) < 0)
2515 return log_oom();
2516
2517 log_level = raise_level(log_level,
2518 value > MENTIONWORTHY_IP_BYTES,
2519 value > NOTICEWORTHY_IP_BYTES);
2520 }
2521 }
2522
2523 /* This check is here because it is the earliest point following all possible log_level assignments.
2524 * (If log_level is assigned anywhere after this point, move this check.) */
2525 if (!unit_log_level_test(u, log_level))
2526 return 0;
2527
2528 /* Is there any accounting data available at all? */
2529 if (n_iovec == 0) {
2530 assert(!message);
2531 return 0;
2532 }
2533
2534 t = strjoin("MESSAGE=", u->id, ": ", message ?: "Completed", ".");
2535 if (!t)
2536 return log_oom();
2537 iovec[n_iovec++] = IOVEC_MAKE_STRING(TAKE_PTR(t));
2538
2539 if (!set_iovec_string_field(iovec, &n_iovec, "MESSAGE_ID=", SD_MESSAGE_UNIT_RESOURCES_STR))
2540 return log_oom();
2541
2542 if (!set_iovec_string_field(iovec, &n_iovec, unit_log_field(u), u->id))
2543 return log_oom();
2544
2545 if (!set_iovec_string_field(iovec, &n_iovec, unit_invocation_log_field(u), u->invocation_id_string))
2546 return log_oom();
2547
2548 log_unit_struct_iovec(u, log_level, iovec, n_iovec);
2549
2550 return 0;
2551 }
2552
2553 static void unit_update_on_console(Unit *u) {
2554 bool b;
2555
2556 assert(u);
2557
2558 b = unit_needs_console(u);
2559 if (u->on_console == b)
2560 return;
2561
2562 u->on_console = b;
2563 if (b)
2564 manager_ref_console(u->manager);
2565 else
2566 manager_unref_console(u->manager);
2567 }
2568
2569 static void unit_emit_audit_start(Unit *u) {
2570 assert(u);
2571
2572 if (UNIT_VTABLE(u)->audit_start_message_type <= 0)
2573 return;
2574
2575 /* Write audit record if we have just finished starting up */
2576 manager_send_unit_audit(u->manager, u, UNIT_VTABLE(u)->audit_start_message_type, /* success= */ true);
2577 u->in_audit = true;
2578 }
2579
2580 static void unit_emit_audit_stop(Unit *u, UnitActiveState state) {
2581 assert(u);
2582
2583 if (UNIT_VTABLE(u)->audit_start_message_type <= 0)
2584 return;
2585
2586 if (u->in_audit) {
2587 /* Write audit record if we have just finished shutting down */
2588 manager_send_unit_audit(u->manager, u, UNIT_VTABLE(u)->audit_stop_message_type, /* success= */ state == UNIT_INACTIVE);
2589 u->in_audit = false;
2590 } else {
2591 /* Hmm, if there was no start record written write it now, so that we always have a nice pair */
2592 manager_send_unit_audit(u->manager, u, UNIT_VTABLE(u)->audit_start_message_type, /* success= */ state == UNIT_INACTIVE);
2593
2594 if (state == UNIT_INACTIVE)
2595 manager_send_unit_audit(u->manager, u, UNIT_VTABLE(u)->audit_stop_message_type, /* success= */ true);
2596 }
2597 }
2598
2599 static bool unit_process_job(Job *j, UnitActiveState ns, bool reload_success) {
2600 bool unexpected = false;
2601 JobResult result;
2602
2603 assert(j);
2604
2605 if (j->state == JOB_WAITING)
2606 /* So we reached a different state for this job. Let's see if we can run it now if it failed previously
2607 * due to EAGAIN. */
2608 job_add_to_run_queue(j);
2609
2610 /* Let's check whether the unit's new state constitutes a finished job, or maybe contradicts a running job and
2611 * hence needs to invalidate jobs. */
2612
2613 switch (j->type) {
2614
2615 case JOB_START:
2616 case JOB_VERIFY_ACTIVE:
2617
2618 if (UNIT_IS_ACTIVE_OR_RELOADING(ns))
2619 job_finish_and_invalidate(j, JOB_DONE, true, false);
2620 else if (j->state == JOB_RUNNING && ns != UNIT_ACTIVATING) {
2621 unexpected = true;
2622
2623 if (UNIT_IS_INACTIVE_OR_FAILED(ns)) {
2624 if (ns == UNIT_FAILED)
2625 result = JOB_FAILED;
2626 else
2627 result = JOB_DONE;
2628
2629 job_finish_and_invalidate(j, result, true, false);
2630 }
2631 }
2632
2633 break;
2634
2635 case JOB_RELOAD:
2636 case JOB_RELOAD_OR_START:
2637 case JOB_TRY_RELOAD:
2638
2639 if (j->state == JOB_RUNNING) {
2640 if (ns == UNIT_ACTIVE)
2641 job_finish_and_invalidate(j, reload_success ? JOB_DONE : JOB_FAILED, true, false);
2642 else if (!IN_SET(ns, UNIT_ACTIVATING, UNIT_RELOADING, UNIT_REFRESHING)) {
2643 unexpected = true;
2644
2645 if (UNIT_IS_INACTIVE_OR_FAILED(ns))
2646 job_finish_and_invalidate(j, ns == UNIT_FAILED ? JOB_FAILED : JOB_DONE, true, false);
2647 }
2648 }
2649
2650 break;
2651
2652 case JOB_STOP:
2653 case JOB_RESTART:
2654 case JOB_TRY_RESTART:
2655
2656 if (UNIT_IS_INACTIVE_OR_FAILED(ns))
2657 job_finish_and_invalidate(j, JOB_DONE, true, false);
2658 else if (j->state == JOB_RUNNING && ns != UNIT_DEACTIVATING) {
2659 unexpected = true;
2660 job_finish_and_invalidate(j, JOB_FAILED, true, false);
2661 }
2662
2663 break;
2664
2665 default:
2666 assert_not_reached();
2667 }
2668
2669 return unexpected;
2670 }
2671
2672 static void unit_recursive_add_to_run_queue(Unit *u) {
2673 assert(u);
2674
2675 if (u->job)
2676 job_add_to_run_queue(u->job);
2677
2678 Unit *child;
2679 UNIT_FOREACH_DEPENDENCY(child, u, UNIT_ATOM_SLICE_OF) {
2680
2681 if (!child->job)
2682 continue;
2683
2684 unit_recursive_add_to_run_queue(child);
2685 }
2686 }
2687
2688 static void unit_check_concurrency_limit(Unit *u) {
2689 assert(u);
2690
2691 Unit *slice = UNIT_GET_SLICE(u);
2692 if (!slice)
2693 return;
2694
2695 /* If a unit was stopped, maybe it has pending siblings (or children thereof) that can be started now */
2696
2697 if (SLICE(slice)->concurrency_soft_max != UINT_MAX) {
2698 Unit *sibling;
2699 UNIT_FOREACH_DEPENDENCY(sibling, slice, UNIT_ATOM_SLICE_OF) {
2700 if (sibling == u)
2701 continue;
2702
2703 unit_recursive_add_to_run_queue(sibling);
2704 }
2705 }
2706
2707 /* Also go up the tree. */
2708 unit_check_concurrency_limit(slice);
2709 }
2710
2711 void unit_notify(Unit *u, UnitActiveState os, UnitActiveState ns, bool reload_success) {
2712 assert(u);
2713 assert(os < _UNIT_ACTIVE_STATE_MAX);
2714 assert(ns < _UNIT_ACTIVE_STATE_MAX);
2715
2716 /* Note that this is called for all low-level state changes, even if they might map to the same high-level
2717 * UnitActiveState! That means that ns == os is an expected behavior here. For example: if a mount point is
2718 * remounted this function will be called too! */
2719
2720 Manager *m = ASSERT_PTR(u->manager);
2721
2722 /* Let's enqueue the change signal early. In case this unit has a job associated we want that this unit is in
2723 * the bus queue, so that any job change signal queued will force out the unit change signal first. */
2724 unit_add_to_dbus_queue(u);
2725
2726 /* Update systemd-oomd on the property/state change.
2727 *
2728 * Always send an update if the unit is going into an inactive state so systemd-oomd knows to
2729 * stop monitoring.
2730 * Also send an update whenever the unit goes active; this is to handle a case where an override file
2731 * sets one of the ManagedOOM*= properties to "kill", then later removes it. systemd-oomd needs to
2732 * know to stop monitoring when the unit changes from "kill" -> "auto" on daemon-reload, but we don't
2733 * have the information on the property. Thus, indiscriminately send an update. */
2734 if (os != ns && (UNIT_IS_INACTIVE_OR_FAILED(ns) || UNIT_IS_ACTIVE_OR_RELOADING(ns)))
2735 (void) manager_varlink_send_managed_oom_update(u);
2736
2737 /* Update timestamps for state changes */
2738 if (!MANAGER_IS_RELOADING(m)) {
2739 dual_timestamp_now(&u->state_change_timestamp);
2740
2741 if (UNIT_IS_INACTIVE_OR_FAILED(os) && !UNIT_IS_INACTIVE_OR_FAILED(ns))
2742 u->inactive_exit_timestamp = u->state_change_timestamp;
2743 else if (!UNIT_IS_INACTIVE_OR_FAILED(os) && UNIT_IS_INACTIVE_OR_FAILED(ns))
2744 u->inactive_enter_timestamp = u->state_change_timestamp;
2745
2746 if (!UNIT_IS_ACTIVE_OR_RELOADING(os) && UNIT_IS_ACTIVE_OR_RELOADING(ns))
2747 u->active_enter_timestamp = u->state_change_timestamp;
2748 else if (UNIT_IS_ACTIVE_OR_RELOADING(os) && !UNIT_IS_ACTIVE_OR_RELOADING(ns))
2749 u->active_exit_timestamp = u->state_change_timestamp;
2750 }
2751
2752 /* Keep track of failed units */
2753 (void) manager_update_failed_units(m, u, ns == UNIT_FAILED);
2754
2755 /* Make sure the cgroup and state files are always removed when we become inactive */
2756 if (UNIT_IS_INACTIVE_OR_FAILED(ns)) {
2757 SET_FLAG(u->markers,
2758 (1u << UNIT_MARKER_NEEDS_RELOAD)|(1u << UNIT_MARKER_NEEDS_RESTART),
2759 false);
2760 unit_prune_cgroup(u);
2761 unit_unlink_state_files(u);
2762 } else if (ns != os && ns == UNIT_RELOADING)
2763 SET_FLAG(u->markers, 1u << UNIT_MARKER_NEEDS_RELOAD, false);
2764
2765 unit_update_on_console(u);
2766
2767 if (!MANAGER_IS_RELOADING(m)) {
2768 bool unexpected;
2769
2770 /* Let's propagate state changes to the job */
2771 if (u->job)
2772 unexpected = unit_process_job(u->job, ns, reload_success);
2773 else
2774 unexpected = true;
2775
2776 /* If this state change happened without being requested by a job, then let's retroactively start or
2777 * stop dependencies. We skip that step when deserializing, since we don't want to create any
2778 * additional jobs just because something is already activated. */
2779
2780 if (unexpected) {
2781 if (UNIT_IS_INACTIVE_OR_FAILED(os) && UNIT_IS_ACTIVE_OR_ACTIVATING(ns))
2782 retroactively_start_dependencies(u);
2783 else if (UNIT_IS_ACTIVE_OR_ACTIVATING(os) && UNIT_IS_INACTIVE_OR_DEACTIVATING(ns))
2784 retroactively_stop_dependencies(u);
2785 }
2786
2787 if (UNIT_IS_ACTIVE_OR_RELOADING(ns) && !UNIT_IS_ACTIVE_OR_RELOADING(os)) {
2788 /* This unit just finished starting up */
2789
2790 unit_emit_audit_start(u);
2791 manager_send_unit_plymouth(m, u);
2792 manager_send_unit_supervisor(m, u, /* active= */ true);
2793
2794 } else if (UNIT_IS_INACTIVE_OR_FAILED(ns) && !UNIT_IS_INACTIVE_OR_FAILED(os)) {
2795 /* This unit just stopped/failed. */
2796
2797 unit_emit_audit_stop(u, ns);
2798 manager_send_unit_supervisor(m, u, /* active= */ false);
2799 unit_log_resources(u);
2800 }
2801
2802 if (ns == UNIT_INACTIVE && !IN_SET(os, UNIT_FAILED, UNIT_INACTIVE, UNIT_MAINTENANCE))
2803 unit_start_on_termination_deps(u, UNIT_ATOM_ON_SUCCESS);
2804 else if (ns != os && ns == UNIT_FAILED)
2805 unit_start_on_termination_deps(u, UNIT_ATOM_ON_FAILURE);
2806 }
2807
2808 manager_recheck_journal(m);
2809 manager_recheck_dbus(m);
2810
2811 unit_trigger_notify(u);
2812
2813 if (!MANAGER_IS_RELOADING(m)) {
2814 const char *reason;
2815
2816 if (os != UNIT_FAILED && ns == UNIT_FAILED) {
2817 reason = strjoina("unit ", u->id, " failed");
2818 emergency_action(m, u->failure_action, EMERGENCY_ACTION_WARN|EMERGENCY_ACTION_SLEEP_5S, u->reboot_arg, unit_failure_action_exit_status(u), reason);
2819 } else if (!UNIT_IS_INACTIVE_OR_FAILED(os) && ns == UNIT_INACTIVE) {
2820 reason = strjoina("unit ", u->id, " succeeded");
2821 emergency_action(m, u->success_action, /* flags= */ 0, u->reboot_arg, unit_success_action_exit_status(u), reason);
2822 }
2823 }
2824
2825 /* And now, add the unit or depending units to various queues that will act on the new situation if
2826 * needed. These queues generally check for continuous state changes rather than events (like most of
2827 * the state propagation above), and do work deferred instead of instantly, since they typically
2828 * don't want to run during reloading, and usually involve checking combined state of multiple units
2829 * at once. */
2830
2831 if (UNIT_IS_INACTIVE_OR_FAILED(ns)) {
2832 /* Stop unneeded units and bound-by units regardless if going down was expected or not */
2833 check_unneeded_dependencies(u);
2834 check_bound_by_dependencies(u);
2835
2836 /* Maybe someone wants us to remain up? */
2837 unit_submit_to_start_when_upheld_queue(u);
2838
2839 /* Maybe the unit should be GC'ed now? */
2840 unit_add_to_gc_queue(u);
2841
2842 /* Maybe we can release some resources now? */
2843 unit_submit_to_release_resources_queue(u);
2844
2845 /* Maybe the concurrency limits now allow dispatching of another start job in this slice? */
2846 unit_check_concurrency_limit(u);
2847
2848 /* Maybe someone else has been waiting for us to stop? */
2849 m->may_dispatch_stop_notify_queue = true;
2850
2851 } else if (UNIT_IS_ACTIVE_OR_RELOADING(ns)) {
2852 /* Start uphold units regardless if going up was expected or not */
2853 check_uphold_dependencies(u);
2854
2855 /* Maybe we finished startup and are now ready for being stopped because unneeded? */
2856 unit_submit_to_stop_when_unneeded_queue(u);
2857
2858 /* Maybe we finished startup, but something we needed has vanished? Let's die then. (This happens
2859 * when something BindsTo= to a Type=oneshot unit, as these units go directly from starting to
2860 * inactive, without ever entering started.) */
2861 unit_submit_to_stop_when_bound_queue(u);
2862 }
2863 }
2864
2865 int unit_watch_pidref(Unit *u, const PidRef *pid, bool exclusive) {
2866 _cleanup_(pidref_freep) PidRef *pid_dup = NULL;
2867 int r;
2868
2869 /* Adds a specific PID to the set of PIDs this unit watches. */
2870
2871 assert(u);
2872 assert(pidref_is_set(pid));
2873
2874 /* Caller might be sure that this PID belongs to this unit only. Let's take this
2875 * opportunity to remove any stalled references to this PID as they can be created
2876 * easily (when watching a process which is not our direct child). */
2877 if (exclusive)
2878 manager_unwatch_pidref(u->manager, pid);
2879
2880 if (set_contains(u->pids, pid)) { /* early exit if already being watched */
2881 assert(!exclusive);
2882 return 0;
2883 }
2884
2885 r = pidref_dup(pid, &pid_dup);
2886 if (r < 0)
2887 return r;
2888
2889 /* First, insert into the set of PIDs maintained by the unit */
2890 r = set_ensure_put(&u->pids, &pidref_hash_ops_free, pid_dup);
2891 if (r < 0)
2892 return r;
2893
2894 pid = TAKE_PTR(pid_dup); /* continue with our copy now that we have installed it properly in our set */
2895
2896 /* Second, insert it into the simple global table, see if that works */
2897 r = hashmap_ensure_put(&u->manager->watch_pids, &pidref_hash_ops, pid, u);
2898 if (r != -EEXIST)
2899 return r;
2900
2901 /* OK, the key is already assigned to a different unit. That's fine, then add us via the second
2902 * hashmap that points to an array. */
2903
2904 PidRef *old_pid = NULL;
2905 Unit **array = hashmap_get2(u->manager->watch_pids_more, pid, (void**) &old_pid);
2906
2907 /* Count entries in array */
2908 size_t n = 0;
2909 for (; array && array[n]; n++)
2910 ;
2911
2912 /* Allocate a new array */
2913 _cleanup_free_ Unit **new_array = new(Unit*, n + 2);
2914 if (!new_array)
2915 return -ENOMEM;
2916
2917 /* Append us to the end */
2918 memcpy_safe(new_array, array, sizeof(Unit*) * n);
2919 new_array[n] = u;
2920 new_array[n+1] = NULL;
2921
2922 /* Add or replace the old array */
2923 r = hashmap_ensure_replace(&u->manager->watch_pids_more, &pidref_hash_ops, old_pid ?: pid, new_array);
2924 if (r < 0)
2925 return r;
2926
2927 TAKE_PTR(new_array); /* Now part of the hash table */
2928 free(array); /* Which means we can now delete the old version */
2929 return 0;
2930 }
2931
2932 void unit_unwatch_pidref(Unit *u, const PidRef *pid) {
2933 assert(u);
2934 assert(pidref_is_set(pid));
2935
2936 /* Remove from the set we maintain for this unit. (And destroy the returned pid eventually) */
2937 _cleanup_(pidref_freep) PidRef *pid1 = set_remove(u->pids, pid);
2938 if (!pid1)
2939 return; /* Early exit if this PID was never watched by us */
2940
2941 /* First let's drop the unit from the simple hash table, if it is included there */
2942 PidRef *pid2 = NULL;
2943 Unit *uu = hashmap_get2(u->manager->watch_pids, pid, (void**) &pid2);
2944
2945 /* Quick validation: iff we are in the watch_pids table then the PidRef object must be the same as in our local pids set */
2946 assert((uu == u) == (pid1 == pid2));
2947
2948 if (uu == u)
2949 /* OK, we are in the first table. Let's remove it there then, and we are done already. */
2950 assert_se(hashmap_remove_value(u->manager->watch_pids, pid2, uu));
2951 else {
2952 /* We weren't in the first table, then let's consult the 2nd table that points to an array */
2953 PidRef *pid3 = NULL;
2954 Unit **array = hashmap_get2(u->manager->watch_pids_more, pid, (void**) &pid3);
2955
2956 /* Let's iterate through the array, dropping our own entry */
2957 size_t m = 0, n = 0;
2958 for (; array && array[n]; n++)
2959 if (array[n] != u)
2960 array[m++] = array[n];
2961 if (n == m)
2962 return; /* Not there */
2963
2964 array[m] = NULL; /* set trailing NULL marker on the new end */
2965
2966 if (m == 0) {
2967 /* The array is now empty, remove the entire entry */
2968 assert_se(hashmap_remove_value(u->manager->watch_pids_more, pid3, array));
2969 free(array);
2970 } else {
2971 /* The array is not empty, but let's make sure the entry is not keyed by the PidRef
2972 * we will delete, but by the PidRef object of the Unit that is now first in the
2973 * array. */
2974
2975 PidRef *new_pid3 = ASSERT_PTR(set_get(array[0]->pids, pid));
2976 assert_se(hashmap_replace(u->manager->watch_pids_more, new_pid3, array) >= 0);
2977 }
2978 }
2979 }
2980
2981 void unit_unwatch_all_pids(Unit *u) {
2982 assert(u);
2983
2984 while (!set_isempty(u->pids))
2985 unit_unwatch_pidref(u, set_first(u->pids));
2986
2987 u->pids = set_free(u->pids);
2988 }
2989
2990 void unit_unwatch_pidref_done(Unit *u, PidRef *pidref) {
2991 assert(u);
2992
2993 if (!pidref_is_set(pidref))
2994 return;
2995
2996 unit_unwatch_pidref(u, pidref);
2997 pidref_done(pidref);
2998 }
2999
3000 bool unit_job_is_applicable(Unit *u, JobType j) {
3001 assert(u);
3002 assert(j >= 0 && j < _JOB_TYPE_MAX);
3003
3004 switch (j) {
3005
3006 case JOB_VERIFY_ACTIVE:
3007 case JOB_START:
3008 case JOB_NOP:
3009 /* Note that we don't check unit_can_start() here. That's because .device units and suchlike are not
3010 * startable by us but may appear due to external events, and it thus makes sense to permit enqueuing
3011 * jobs for it. */
3012 return true;
3013
3014 case JOB_STOP:
3015 /* Similar as above. However, perpetual units can never be stopped (neither explicitly nor due to
3016 * external events), hence it makes no sense to permit enqueuing such a request either. */
3017 return !u->perpetual;
3018
3019 case JOB_RESTART:
3020 case JOB_TRY_RESTART:
3021 return unit_can_stop(u) && unit_can_start(u);
3022
3023 case JOB_RELOAD:
3024 case JOB_TRY_RELOAD:
3025 return unit_can_reload(u);
3026
3027 case JOB_RELOAD_OR_START:
3028 return unit_can_reload(u) && unit_can_start(u);
3029
3030 default:
3031 assert_not_reached();
3032 }
3033 }
3034
3035 static Hashmap *unit_get_dependency_hashmap_per_type(Unit *u, UnitDependency d) {
3036 Hashmap *deps;
3037
3038 assert(u);
3039 assert(d >= 0 && d < _UNIT_DEPENDENCY_MAX);
3040
3041 deps = hashmap_get(u->dependencies, UNIT_DEPENDENCY_TO_PTR(d));
3042 if (!deps) {
3043 _cleanup_hashmap_free_ Hashmap *h = NULL;
3044
3045 h = hashmap_new(NULL);
3046 if (!h)
3047 return NULL;
3048
3049 if (hashmap_ensure_put(&u->dependencies, NULL, UNIT_DEPENDENCY_TO_PTR(d), h) < 0)
3050 return NULL;
3051
3052 deps = TAKE_PTR(h);
3053 }
3054
3055 return deps;
3056 }
3057
3058 typedef enum NotifyDependencyFlags {
3059 NOTIFY_DEPENDENCY_UPDATE_FROM = 1 << 0,
3060 NOTIFY_DEPENDENCY_UPDATE_TO = 1 << 1,
3061 } NotifyDependencyFlags;
3062
3063 static int unit_add_dependency_impl(
3064 Unit *u,
3065 UnitDependency d,
3066 Unit *other,
3067 UnitDependencyMask mask) {
3068
3069 static const UnitDependency inverse_table[_UNIT_DEPENDENCY_MAX] = {
3070 [UNIT_REQUIRES] = UNIT_REQUIRED_BY,
3071 [UNIT_REQUISITE] = UNIT_REQUISITE_OF,
3072 [UNIT_WANTS] = UNIT_WANTED_BY,
3073 [UNIT_BINDS_TO] = UNIT_BOUND_BY,
3074 [UNIT_PART_OF] = UNIT_CONSISTS_OF,
3075 [UNIT_UPHOLDS] = UNIT_UPHELD_BY,
3076 [UNIT_REQUIRED_BY] = UNIT_REQUIRES,
3077 [UNIT_REQUISITE_OF] = UNIT_REQUISITE,
3078 [UNIT_WANTED_BY] = UNIT_WANTS,
3079 [UNIT_BOUND_BY] = UNIT_BINDS_TO,
3080 [UNIT_CONSISTS_OF] = UNIT_PART_OF,
3081 [UNIT_UPHELD_BY] = UNIT_UPHOLDS,
3082 [UNIT_CONFLICTS] = UNIT_CONFLICTED_BY,
3083 [UNIT_CONFLICTED_BY] = UNIT_CONFLICTS,
3084 [UNIT_BEFORE] = UNIT_AFTER,
3085 [UNIT_AFTER] = UNIT_BEFORE,
3086 [UNIT_ON_SUCCESS] = UNIT_ON_SUCCESS_OF,
3087 [UNIT_ON_SUCCESS_OF] = UNIT_ON_SUCCESS,
3088 [UNIT_ON_FAILURE] = UNIT_ON_FAILURE_OF,
3089 [UNIT_ON_FAILURE_OF] = UNIT_ON_FAILURE,
3090 [UNIT_TRIGGERS] = UNIT_TRIGGERED_BY,
3091 [UNIT_TRIGGERED_BY] = UNIT_TRIGGERS,
3092 [UNIT_PROPAGATES_RELOAD_TO] = UNIT_RELOAD_PROPAGATED_FROM,
3093 [UNIT_RELOAD_PROPAGATED_FROM] = UNIT_PROPAGATES_RELOAD_TO,
3094 [UNIT_PROPAGATES_STOP_TO] = UNIT_STOP_PROPAGATED_FROM,
3095 [UNIT_STOP_PROPAGATED_FROM] = UNIT_PROPAGATES_STOP_TO,
3096 [UNIT_JOINS_NAMESPACE_OF] = UNIT_JOINS_NAMESPACE_OF, /* symmetric! 👓 */
3097 [UNIT_REFERENCES] = UNIT_REFERENCED_BY,
3098 [UNIT_REFERENCED_BY] = UNIT_REFERENCES,
3099 [UNIT_IN_SLICE] = UNIT_SLICE_OF,
3100 [UNIT_SLICE_OF] = UNIT_IN_SLICE,
3101 };
3102
3103 Hashmap *u_deps, *other_deps;
3104 UnitDependencyInfo u_info, u_info_old, other_info, other_info_old;
3105 NotifyDependencyFlags flags = 0;
3106 int r;
3107
3108 assert(u);
3109 assert(other);
3110 assert(d >= 0 && d < _UNIT_DEPENDENCY_MAX);
3111 assert(inverse_table[d] >= 0 && inverse_table[d] < _UNIT_DEPENDENCY_MAX);
3112 assert(mask > 0 && mask < _UNIT_DEPENDENCY_MASK_FULL);
3113
3114 /* Ensure the following two hashmaps for each unit exist:
3115 * - the top-level dependency hashmap that maps UnitDependency → Hashmap(Unit* → UnitDependencyInfo),
3116 * - the inner hashmap, that maps Unit* → UnitDependencyInfo, for the specified dependency type. */
3117 u_deps = unit_get_dependency_hashmap_per_type(u, d);
3118 if (!u_deps)
3119 return -ENOMEM;
3120
3121 other_deps = unit_get_dependency_hashmap_per_type(other, inverse_table[d]);
3122 if (!other_deps)
3123 return -ENOMEM;
3124
3125 /* Save the original dependency info. */
3126 u_info.data = u_info_old.data = hashmap_get(u_deps, other);
3127 other_info.data = other_info_old.data = hashmap_get(other_deps, u);
3128
3129 /* Update dependency info. */
3130 u_info.origin_mask |= mask;
3131 other_info.destination_mask |= mask;
3132
3133 /* Save updated dependency info. */
3134 if (u_info.data != u_info_old.data) {
3135 r = hashmap_replace(u_deps, other, u_info.data);
3136 if (r < 0)
3137 return r;
3138
3139 flags = NOTIFY_DEPENDENCY_UPDATE_FROM;
3140 u->dependency_generation++;
3141 }
3142
3143 if (other_info.data != other_info_old.data) {
3144 r = hashmap_replace(other_deps, u, other_info.data);
3145 if (r < 0) {
3146 if (u_info.data != u_info_old.data) {
3147 /* Restore the old dependency. */
3148 if (u_info_old.data)
3149 (void) hashmap_update(u_deps, other, u_info_old.data);
3150 else
3151 hashmap_remove(u_deps, other);
3152 }
3153 return r;
3154 }
3155
3156 flags |= NOTIFY_DEPENDENCY_UPDATE_TO;
3157 other->dependency_generation++;
3158 }
3159
3160 return flags;
3161 }
3162
3163 int unit_add_dependency(
3164 Unit *u,
3165 UnitDependency d,
3166 Unit *other,
3167 bool add_reference,
3168 UnitDependencyMask mask) {
3169
3170 UnitDependencyAtom a;
3171 int r;
3172
3173 /* Helper to know whether sending a notification is necessary or not: if the dependency is already
3174 * there, no need to notify! */
3175 NotifyDependencyFlags notify_flags;
3176
3177 assert(u);
3178 assert(d >= 0 && d < _UNIT_DEPENDENCY_MAX);
3179 assert(other);
3180
3181 u = unit_follow_merge(u);
3182 other = unit_follow_merge(other);
3183 a = unit_dependency_to_atom(d);
3184 assert(a >= 0);
3185
3186 /* We won't allow dependencies on ourselves. We will not consider them an error however. */
3187 if (u == other) {
3188 if (unit_should_warn_about_dependency(d))
3189 log_unit_warning(u, "Dependency %s=%s is dropped.",
3190 unit_dependency_to_string(d), u->id);
3191 return 0;
3192 }
3193
3194 if (u->manager && FLAGS_SET(u->manager->test_run_flags, MANAGER_TEST_RUN_IGNORE_DEPENDENCIES))
3195 return 0;
3196
3197 /* Note that ordering a device unit after a unit is permitted since it allows its job running
3198 * timeout to be started at a specific time. */
3199 if (FLAGS_SET(a, UNIT_ATOM_BEFORE) && other->type == UNIT_DEVICE) {
3200 log_unit_warning(u, "Dependency Before=%s ignored (.device units cannot be delayed)", other->id);
3201 return 0;
3202 }
3203
3204 if (FLAGS_SET(a, UNIT_ATOM_ON_FAILURE) && !UNIT_VTABLE(u)->can_fail) {
3205 log_unit_warning(u, "Requested dependency OnFailure=%s ignored (%s units cannot fail).", other->id, unit_type_to_string(u->type));
3206 return 0;
3207 }
3208
3209 if (FLAGS_SET(a, UNIT_ATOM_TRIGGERS) && !UNIT_VTABLE(u)->can_trigger)
3210 return log_unit_error_errno(u, SYNTHETIC_ERRNO(EINVAL),
3211 "Requested dependency Triggers=%s refused (%s units cannot trigger other units).", other->id, unit_type_to_string(u->type));
3212 if (FLAGS_SET(a, UNIT_ATOM_TRIGGERED_BY) && !UNIT_VTABLE(other)->can_trigger)
3213 return log_unit_error_errno(u, SYNTHETIC_ERRNO(EINVAL),
3214 "Requested dependency TriggeredBy=%s refused (%s units cannot trigger other units).", other->id, unit_type_to_string(other->type));
3215
3216 if (FLAGS_SET(a, UNIT_ATOM_IN_SLICE) && other->type != UNIT_SLICE)
3217 return log_unit_error_errno(u, SYNTHETIC_ERRNO(EINVAL),
3218 "Requested dependency Slice=%s refused (%s is not a slice unit).", other->id, other->id);
3219 if (FLAGS_SET(a, UNIT_ATOM_SLICE_OF) && u->type != UNIT_SLICE)
3220 return log_unit_error_errno(u, SYNTHETIC_ERRNO(EINVAL),
3221 "Requested dependency SliceOf=%s refused (%s is not a slice unit).", other->id, u->id);
3222
3223 if (FLAGS_SET(a, UNIT_ATOM_IN_SLICE) && !UNIT_HAS_CGROUP_CONTEXT(u))
3224 return log_unit_error_errno(u, SYNTHETIC_ERRNO(EINVAL),
3225 "Requested dependency Slice=%s refused (%s is not a cgroup unit).", other->id, u->id);
3226
3227 if (FLAGS_SET(a, UNIT_ATOM_SLICE_OF) && !UNIT_HAS_CGROUP_CONTEXT(other))
3228 return log_unit_error_errno(u, SYNTHETIC_ERRNO(EINVAL),
3229 "Requested dependency SliceOf=%s refused (%s is not a cgroup unit).", other->id, other->id);
3230
3231 r = unit_add_dependency_impl(u, d, other, mask);
3232 if (r < 0)
3233 return r;
3234 notify_flags = r;
3235
3236 if (add_reference) {
3237 r = unit_add_dependency_impl(u, UNIT_REFERENCES, other, mask);
3238 if (r < 0)
3239 return r;
3240 notify_flags |= r;
3241 }
3242
3243 if (FLAGS_SET(notify_flags, NOTIFY_DEPENDENCY_UPDATE_FROM))
3244 unit_add_to_dbus_queue(u);
3245 if (FLAGS_SET(notify_flags, NOTIFY_DEPENDENCY_UPDATE_TO))
3246 unit_add_to_dbus_queue(other);
3247
3248 return notify_flags != 0;
3249 }
3250
3251 int unit_add_two_dependencies(Unit *u, UnitDependency d, UnitDependency e, Unit *other, bool add_reference, UnitDependencyMask mask) {
3252 int r = 0, s = 0;
3253
3254 assert(u);
3255 assert(d >= 0 || e >= 0);
3256
3257 if (d >= 0) {
3258 r = unit_add_dependency(u, d, other, add_reference, mask);
3259 if (r < 0)
3260 return r;
3261 }
3262
3263 if (e >= 0) {
3264 s = unit_add_dependency(u, e, other, add_reference, mask);
3265 if (s < 0)
3266 return s;
3267 }
3268
3269 return r > 0 || s > 0;
3270 }
3271
3272 static int resolve_template(Unit *u, const char *name, char **buf, const char **ret) {
3273 int r;
3274
3275 assert(u);
3276 assert(name);
3277 assert(buf);
3278 assert(ret);
3279
3280 if (!unit_name_is_valid(name, UNIT_NAME_TEMPLATE)) {
3281 *buf = NULL;
3282 *ret = name;
3283 return 0;
3284 }
3285
3286 if (u->instance)
3287 r = unit_name_replace_instance(name, u->instance, buf);
3288 else {
3289 _cleanup_free_ char *i = NULL;
3290
3291 r = unit_name_to_prefix(u->id, &i);
3292 if (r < 0)
3293 return r;
3294
3295 r = unit_name_replace_instance(name, i, buf);
3296 }
3297 if (r < 0)
3298 return r;
3299
3300 *ret = *buf;
3301 return 0;
3302 }
3303
3304 int unit_add_dependency_by_name(Unit *u, UnitDependency d, const char *name, bool add_reference, UnitDependencyMask mask) {
3305 _cleanup_free_ char *buf = NULL;
3306 Unit *other;
3307 int r;
3308
3309 assert(u);
3310 assert(name);
3311
3312 r = resolve_template(u, name, &buf, &name);
3313 if (r < 0)
3314 return r;
3315
3316 if (u->manager && FLAGS_SET(u->manager->test_run_flags, MANAGER_TEST_RUN_IGNORE_DEPENDENCIES))
3317 return 0;
3318
3319 r = manager_load_unit(u->manager, name, NULL, NULL, &other);
3320 if (r < 0)
3321 return r;
3322
3323 return unit_add_dependency(u, d, other, add_reference, mask);
3324 }
3325
3326 int unit_add_two_dependencies_by_name(Unit *u, UnitDependency d, UnitDependency e, const char *name, bool add_reference, UnitDependencyMask mask) {
3327 _cleanup_free_ char *buf = NULL;
3328 Unit *other;
3329 int r;
3330
3331 assert(u);
3332 assert(name);
3333
3334 r = resolve_template(u, name, &buf, &name);
3335 if (r < 0)
3336 return r;
3337
3338 if (u->manager && FLAGS_SET(u->manager->test_run_flags, MANAGER_TEST_RUN_IGNORE_DEPENDENCIES))
3339 return 0;
3340
3341 r = manager_load_unit(u->manager, name, NULL, NULL, &other);
3342 if (r < 0)
3343 return r;
3344
3345 return unit_add_two_dependencies(u, d, e, other, add_reference, mask);
3346 }
3347
3348 int setenv_unit_path(const char *p) {
3349 assert(p);
3350
3351 /* This is mostly for debug purposes */
3352 return RET_NERRNO(setenv("SYSTEMD_UNIT_PATH", p, /* overwrite = */ true));
3353 }
3354
3355 char* unit_dbus_path(Unit *u) {
3356 assert(u);
3357
3358 if (!u->id)
3359 return NULL;
3360
3361 return unit_dbus_path_from_name(u->id);
3362 }
3363
3364 char* unit_dbus_path_invocation_id(Unit *u) {
3365 assert(u);
3366
3367 if (sd_id128_is_null(u->invocation_id))
3368 return NULL;
3369
3370 return unit_dbus_path_from_name(u->invocation_id_string);
3371 }
3372
3373 int unit_set_invocation_id(Unit *u, sd_id128_t id) {
3374 int r;
3375
3376 assert(u);
3377
3378 /* Set the invocation ID for this unit. If we cannot, this will not roll back, but reset the whole thing. */
3379
3380 if (sd_id128_equal(u->invocation_id, id))
3381 return 0;
3382
3383 if (!sd_id128_is_null(u->invocation_id))
3384 (void) hashmap_remove_value(u->manager->units_by_invocation_id, &u->invocation_id, u);
3385
3386 if (sd_id128_is_null(id)) {
3387 r = 0;
3388 goto reset;
3389 }
3390
3391 r = hashmap_ensure_allocated(&u->manager->units_by_invocation_id, &id128_hash_ops);
3392 if (r < 0)
3393 goto reset;
3394
3395 u->invocation_id = id;
3396 sd_id128_to_string(id, u->invocation_id_string);
3397
3398 r = hashmap_put(u->manager->units_by_invocation_id, &u->invocation_id, u);
3399 if (r < 0)
3400 goto reset;
3401
3402 return 0;
3403
3404 reset:
3405 u->invocation_id = SD_ID128_NULL;
3406 u->invocation_id_string[0] = 0;
3407 return r;
3408 }
3409
3410 int unit_set_slice(Unit *u, Unit *slice) {
3411 int r;
3412
3413 assert(u);
3414 assert(slice);
3415
3416 /* Sets the unit slice if it has not been set before. Is extra careful, to only allow this for units
3417 * that actually have a cgroup context. Also, we don't allow to set this for slices (since the parent
3418 * slice is derived from the name). Make sure the unit we set is actually a slice. */
3419
3420 if (!UNIT_HAS_CGROUP_CONTEXT(u))
3421 return -EOPNOTSUPP;
3422
3423 if (u->type == UNIT_SLICE)
3424 return -EINVAL;
3425
3426 if (unit_active_state(u) != UNIT_INACTIVE)
3427 return -EBUSY;
3428
3429 if (slice->type != UNIT_SLICE)
3430 return -EINVAL;
3431
3432 if (unit_has_name(u, SPECIAL_INIT_SCOPE) &&
3433 !unit_has_name(slice, SPECIAL_ROOT_SLICE))
3434 return -EPERM;
3435
3436 if (UNIT_GET_SLICE(u) == slice)
3437 return 0;
3438
3439 /* Disallow slice changes if @u is already bound to cgroups */
3440 if (UNIT_GET_SLICE(u)) {
3441 CGroupRuntime *crt = unit_get_cgroup_runtime(u);
3442 if (crt && crt->cgroup_path)
3443 return -EBUSY;
3444 }
3445
3446 /* Remove any slices assigned prior; we should only have one UNIT_IN_SLICE dependency */
3447 if (UNIT_GET_SLICE(u))
3448 unit_remove_dependencies(u, UNIT_DEPENDENCY_SLICE_PROPERTY);
3449
3450 r = unit_add_dependency(u, UNIT_IN_SLICE, slice, true, UNIT_DEPENDENCY_SLICE_PROPERTY);
3451 if (r < 0)
3452 return r;
3453
3454 return 1;
3455 }
3456
3457 int unit_set_default_slice(Unit *u) {
3458 const char *slice_name;
3459 Unit *slice;
3460 int r;
3461
3462 assert(u);
3463
3464 if (u->manager && FLAGS_SET(u->manager->test_run_flags, MANAGER_TEST_RUN_IGNORE_DEPENDENCIES))
3465 return 0;
3466
3467 if (UNIT_GET_SLICE(u))
3468 return 0;
3469
3470 if (u->instance) {
3471 _cleanup_free_ char *prefix = NULL, *escaped = NULL;
3472
3473 /* Implicitly place all instantiated units in their
3474 * own per-template slice */
3475
3476 r = unit_name_to_prefix(u->id, &prefix);
3477 if (r < 0)
3478 return r;
3479
3480 /* The prefix is already escaped, but it might include
3481 * "-" which has a special meaning for slice units,
3482 * hence escape it here extra. */
3483 escaped = unit_name_escape(prefix);
3484 if (!escaped)
3485 return -ENOMEM;
3486
3487 if (MANAGER_IS_SYSTEM(u->manager))
3488 slice_name = strjoina("system-", escaped, ".slice");
3489 else
3490 slice_name = strjoina("app-", escaped, ".slice");
3491
3492 } else if (unit_is_extrinsic(u))
3493 /* Keep all extrinsic units (e.g. perpetual units and swap and mount units in user mode) in
3494 * the root slice. They don't really belong in one of the subslices. */
3495 slice_name = SPECIAL_ROOT_SLICE;
3496
3497 else if (MANAGER_IS_SYSTEM(u->manager))
3498 slice_name = SPECIAL_SYSTEM_SLICE;
3499 else
3500 slice_name = SPECIAL_APP_SLICE;
3501
3502 r = manager_load_unit(u->manager, slice_name, NULL, NULL, &slice);
3503 if (r < 0)
3504 return r;
3505
3506 return unit_set_slice(u, slice);
3507 }
3508
3509 const char* unit_slice_name(Unit *u) {
3510 Unit *slice;
3511 assert(u);
3512
3513 slice = UNIT_GET_SLICE(u);
3514 if (!slice)
3515 return NULL;
3516
3517 return slice->id;
3518 }
3519
3520 int unit_load_related_unit(Unit *u, const char *type, Unit **_found) {
3521 _cleanup_free_ char *t = NULL;
3522 int r;
3523
3524 assert(u);
3525 assert(type);
3526 assert(_found);
3527
3528 r = unit_name_change_suffix(u->id, type, &t);
3529 if (r < 0)
3530 return r;
3531 if (unit_has_name(u, t))
3532 return -EINVAL;
3533
3534 r = manager_load_unit(u->manager, t, NULL, NULL, _found);
3535 assert(r < 0 || *_found != u);
3536 return r;
3537 }
3538
3539 static int signal_name_owner_changed_install_handler(sd_bus_message *message, void *userdata, sd_bus_error *error) {
3540 Unit *u = ASSERT_PTR(userdata);
3541 const sd_bus_error *e;
3542 int r;
3543
3544 e = sd_bus_message_get_error(message);
3545 if (!e) {
3546 log_unit_trace(u, "Successfully installed NameOwnerChanged signal match.");
3547 return 0;
3548 }
3549
3550 r = sd_bus_error_get_errno(e);
3551 log_unit_error_errno(u, r,
3552 "Unexpected error response on installing NameOwnerChanged signal match: %s",
3553 bus_error_message(e, r));
3554
3555 /* If we failed to install NameOwnerChanged signal, also unref the bus slot of GetNameOwner(). */
3556 u->match_bus_slot = sd_bus_slot_unref(u->match_bus_slot);
3557 u->get_name_owner_slot = sd_bus_slot_unref(u->get_name_owner_slot);
3558
3559 if (UNIT_VTABLE(u)->bus_name_owner_change)
3560 UNIT_VTABLE(u)->bus_name_owner_change(u, NULL);
3561
3562 return 0;
3563 }
3564
3565 static int signal_name_owner_changed(sd_bus_message *message, void *userdata, sd_bus_error *error) {
3566 const char *new_owner;
3567 Unit *u = ASSERT_PTR(userdata);
3568 int r;
3569
3570 assert(message);
3571
3572 r = sd_bus_message_read(message, "sss", NULL, NULL, &new_owner);
3573 if (r < 0) {
3574 bus_log_parse_error(r);
3575 return 0;
3576 }
3577
3578 if (UNIT_VTABLE(u)->bus_name_owner_change)
3579 UNIT_VTABLE(u)->bus_name_owner_change(u, empty_to_null(new_owner));
3580
3581 return 0;
3582 }
3583
3584 static int get_name_owner_handler(sd_bus_message *message, void *userdata, sd_bus_error *error) {
3585 const sd_bus_error *e;
3586 const char *new_owner;
3587 Unit *u = ASSERT_PTR(userdata);
3588 int r;
3589
3590 assert(message);
3591
3592 u->get_name_owner_slot = sd_bus_slot_unref(u->get_name_owner_slot);
3593
3594 e = sd_bus_message_get_error(message);
3595 if (e) {
3596 if (!sd_bus_error_has_name(e, SD_BUS_ERROR_NAME_HAS_NO_OWNER)) {
3597 r = sd_bus_error_get_errno(e);
3598 log_unit_error_errno(u, r,
3599 "Unexpected error response from GetNameOwner(): %s",
3600 bus_error_message(e, r));
3601 }
3602
3603 new_owner = NULL;
3604 } else {
3605 r = sd_bus_message_read(message, "s", &new_owner);
3606 if (r < 0)
3607 return bus_log_parse_error(r);
3608
3609 assert(!isempty(new_owner));
3610 }
3611
3612 if (UNIT_VTABLE(u)->bus_name_owner_change)
3613 UNIT_VTABLE(u)->bus_name_owner_change(u, new_owner);
3614
3615 return 0;
3616 }
3617
3618 int unit_install_bus_match(Unit *u, sd_bus *bus, const char *name) {
3619 _cleanup_(sd_bus_message_unrefp) sd_bus_message *m = NULL;
3620 const char *match;
3621 usec_t timeout_usec = 0;
3622 int r;
3623
3624 assert(u);
3625 assert(bus);
3626 assert(name);
3627
3628 if (u->match_bus_slot || u->get_name_owner_slot)
3629 return -EBUSY;
3630
3631 /* NameOwnerChanged and GetNameOwner is used to detect when a service finished starting up. The dbus
3632 * call timeout shouldn't be earlier than that. If we couldn't get the start timeout, use the default
3633 * value defined above. */
3634 if (UNIT_VTABLE(u)->get_timeout_start_usec)
3635 timeout_usec = UNIT_VTABLE(u)->get_timeout_start_usec(u);
3636
3637 match = strjoina("type='signal',"
3638 "sender='org.freedesktop.DBus',"
3639 "path='/org/freedesktop/DBus',"
3640 "interface='org.freedesktop.DBus',"
3641 "member='NameOwnerChanged',"
3642 "arg0='", name, "'");
3643
3644 r = bus_add_match_full(
3645 bus,
3646 &u->match_bus_slot,
3647 /* asynchronous = */ true,
3648 match,
3649 signal_name_owner_changed,
3650 signal_name_owner_changed_install_handler,
3651 u,
3652 timeout_usec);
3653 if (r < 0)
3654 return r;
3655
3656 r = sd_bus_message_new_method_call(
3657 bus,
3658 &m,
3659 "org.freedesktop.DBus",
3660 "/org/freedesktop/DBus",
3661 "org.freedesktop.DBus",
3662 "GetNameOwner");
3663 if (r < 0)
3664 return r;
3665
3666 r = sd_bus_message_append(m, "s", name);
3667 if (r < 0)
3668 return r;
3669
3670 r = sd_bus_call_async(
3671 bus,
3672 &u->get_name_owner_slot,
3673 m,
3674 get_name_owner_handler,
3675 u,
3676 timeout_usec);
3677 if (r < 0) {
3678 u->match_bus_slot = sd_bus_slot_unref(u->match_bus_slot);
3679 return r;
3680 }
3681
3682 log_unit_debug(u, "Watching D-Bus name '%s'.", name);
3683 return 0;
3684 }
3685
3686 int unit_watch_bus_name(Unit *u, const char *name) {
3687 int r;
3688
3689 assert(u);
3690 assert(name);
3691
3692 /* Watch a specific name on the bus. We only support one unit
3693 * watching each name for now. */
3694
3695 if (u->manager->api_bus) {
3696 /* If the bus is already available, install the match directly.
3697 * Otherwise, just put the name in the list. bus_setup_api() will take care later. */
3698 r = unit_install_bus_match(u, u->manager->api_bus, name);
3699 if (r < 0)
3700 return log_warning_errno(r, "Failed to subscribe to NameOwnerChanged signal for '%s': %m", name);
3701 }
3702
3703 r = hashmap_put(u->manager->watch_bus, name, u);
3704 if (r < 0) {
3705 u->match_bus_slot = sd_bus_slot_unref(u->match_bus_slot);
3706 u->get_name_owner_slot = sd_bus_slot_unref(u->get_name_owner_slot);
3707 return log_warning_errno(r, "Failed to put bus name to hashmap: %m");
3708 }
3709
3710 return 0;
3711 }
3712
3713 void unit_unwatch_bus_name(Unit *u, const char *name) {
3714 assert(u);
3715 assert(name);
3716
3717 (void) hashmap_remove_value(u->manager->watch_bus, name, u);
3718 u->match_bus_slot = sd_bus_slot_unref(u->match_bus_slot);
3719 u->get_name_owner_slot = sd_bus_slot_unref(u->get_name_owner_slot);
3720 }
3721
3722 int unit_add_node_dependency(Unit *u, const char *what, UnitDependency dep, UnitDependencyMask mask) {
3723 _cleanup_free_ char *e = NULL;
3724 Unit *device;
3725 int r;
3726
3727 assert(u);
3728
3729 /* Adds in links to the device node that this unit is based on */
3730 if (isempty(what))
3731 return 0;
3732
3733 if (!is_device_path(what))
3734 return 0;
3735
3736 /* When device units aren't supported (such as in a container), don't create dependencies on them. */
3737 if (!unit_type_supported(UNIT_DEVICE))
3738 return 0;
3739
3740 r = unit_name_from_path(what, ".device", &e);
3741 if (r < 0)
3742 return r;
3743
3744 r = manager_load_unit(u->manager, e, NULL, NULL, &device);
3745 if (r < 0)
3746 return r;
3747
3748 if (dep == UNIT_REQUIRES && device_shall_be_bound_by(device, u))
3749 dep = UNIT_BINDS_TO;
3750
3751 return unit_add_two_dependencies(u, UNIT_AFTER,
3752 MANAGER_IS_SYSTEM(u->manager) ? dep : UNIT_WANTS,
3753 device, true, mask);
3754 }
3755
3756 int unit_add_blockdev_dependency(Unit *u, const char *what, UnitDependencyMask mask) {
3757 _cleanup_free_ char *escaped = NULL, *target = NULL;
3758 int r;
3759
3760 assert(u);
3761
3762 if (isempty(what))
3763 return 0;
3764
3765 if (!path_startswith(what, "/dev/"))
3766 return 0;
3767
3768 /* If we don't support devices, then also don't bother with blockdev@.target */
3769 if (!unit_type_supported(UNIT_DEVICE))
3770 return 0;
3771
3772 r = unit_name_path_escape(what, &escaped);
3773 if (r < 0)
3774 return r;
3775
3776 r = unit_name_build("blockdev", escaped, ".target", &target);
3777 if (r < 0)
3778 return r;
3779
3780 return unit_add_dependency_by_name(u, UNIT_AFTER, target, true, mask);
3781 }
3782
3783 int unit_coldplug(Unit *u) {
3784 int r = 0;
3785
3786 assert(u);
3787
3788 /* Make sure we don't enter a loop, when coldplugging recursively. */
3789 if (u->coldplugged)
3790 return 0;
3791
3792 u->coldplugged = true;
3793
3794 STRV_FOREACH(i, u->deserialized_refs)
3795 RET_GATHER(r, bus_unit_track_add_name(u, *i));
3796
3797 u->deserialized_refs = strv_free(u->deserialized_refs);
3798
3799 if (UNIT_VTABLE(u)->coldplug)
3800 RET_GATHER(r, UNIT_VTABLE(u)->coldplug(u));
3801
3802 if (u->job)
3803 RET_GATHER(r, job_coldplug(u->job));
3804 if (u->nop_job)
3805 RET_GATHER(r, job_coldplug(u->nop_job));
3806
3807 unit_modify_nft_set(u, /* add = */ true);
3808 return r;
3809 }
3810
3811 void unit_catchup(Unit *u) {
3812 assert(u);
3813
3814 if (UNIT_VTABLE(u)->catchup)
3815 UNIT_VTABLE(u)->catchup(u);
3816
3817 unit_cgroup_catchup(u);
3818 }
3819
3820 static bool fragment_mtime_newer(const char *path, usec_t mtime, bool path_masked) {
3821 struct stat st;
3822
3823 if (!path)
3824 return false;
3825
3826 /* If the source is some virtual kernel file system, then we assume we watch it anyway, and hence pretend we
3827 * are never out-of-date. */
3828 if (PATH_STARTSWITH_SET(path, "/proc", "/sys"))
3829 return false;
3830
3831 if (stat(path, &st) < 0)
3832 /* What, cannot access this anymore? */
3833 return true;
3834
3835 if (path_masked)
3836 /* For masked files check if they are still so */
3837 return !null_or_empty(&st);
3838 else
3839 /* For non-empty files check the mtime */
3840 return timespec_load(&st.st_mtim) > mtime;
3841
3842 return false;
3843 }
3844
3845 bool unit_need_daemon_reload(Unit *u) {
3846 assert(u);
3847 assert(u->manager);
3848
3849 if (u->manager->unit_file_state_outdated)
3850 return true;
3851
3852 /* For unit files, we allow masking… */
3853 if (fragment_mtime_newer(u->fragment_path, u->fragment_mtime,
3854 u->load_state == UNIT_MASKED))
3855 return true;
3856
3857 /* Source paths should not be masked… */
3858 if (fragment_mtime_newer(u->source_path, u->source_mtime, false))
3859 return true;
3860
3861 if (u->load_state == UNIT_LOADED) {
3862 _cleanup_strv_free_ char **dropins = NULL;
3863
3864 (void) unit_find_dropin_paths(u, /* use_unit_path_cache = */ false, &dropins);
3865
3866 if (!strv_equal(u->dropin_paths, dropins))
3867 return true;
3868
3869 /* … any drop-ins that are masked are simply omitted from the list. */
3870 STRV_FOREACH(path, u->dropin_paths)
3871 if (fragment_mtime_newer(*path, u->dropin_mtime, false))
3872 return true;
3873 }
3874
3875 return false;
3876 }
3877
3878 void unit_reset_failed(Unit *u) {
3879 assert(u);
3880
3881 if (UNIT_VTABLE(u)->reset_failed)
3882 UNIT_VTABLE(u)->reset_failed(u);
3883
3884 ratelimit_reset(&u->start_ratelimit);
3885 u->start_limit_hit = false;
3886
3887 (void) unit_set_debug_invocation(u, /* enable= */ false);
3888 }
3889
3890 Unit *unit_following(Unit *u) {
3891 assert(u);
3892
3893 if (UNIT_VTABLE(u)->following)
3894 return UNIT_VTABLE(u)->following(u);
3895
3896 return NULL;
3897 }
3898
3899 bool unit_stop_pending(Unit *u) {
3900 assert(u);
3901
3902 /* This call does check the current state of the unit. It's
3903 * hence useful to be called from state change calls of the
3904 * unit itself, where the state isn't updated yet. This is
3905 * different from unit_inactive_or_pending() which checks both
3906 * the current state and for a queued job. */
3907
3908 return unit_has_job_type(u, JOB_STOP);
3909 }
3910
3911 bool unit_inactive_or_pending(Unit *u) {
3912 assert(u);
3913
3914 /* Returns true if the unit is inactive or going down */
3915
3916 if (UNIT_IS_INACTIVE_OR_DEACTIVATING(unit_active_state(u)))
3917 return true;
3918
3919 if (unit_stop_pending(u))
3920 return true;
3921
3922 return false;
3923 }
3924
3925 bool unit_active_or_pending(Unit *u) {
3926 assert(u);
3927
3928 /* Returns true if the unit is active or going up */
3929
3930 if (UNIT_IS_ACTIVE_OR_ACTIVATING(unit_active_state(u)))
3931 return true;
3932
3933 if (u->job &&
3934 IN_SET(u->job->type, JOB_START, JOB_RELOAD_OR_START, JOB_RESTART))
3935 return true;
3936
3937 return false;
3938 }
3939
3940 bool unit_will_restart_default(Unit *u) {
3941 assert(u);
3942
3943 return unit_has_job_type(u, JOB_START);
3944 }
3945
3946 bool unit_will_restart(Unit *u) {
3947 assert(u);
3948
3949 if (!UNIT_VTABLE(u)->will_restart)
3950 return false;
3951
3952 return UNIT_VTABLE(u)->will_restart(u);
3953 }
3954
3955 void unit_notify_cgroup_oom(Unit *u, bool managed_oom) {
3956 assert(u);
3957
3958 if (UNIT_VTABLE(u)->notify_cgroup_oom)
3959 UNIT_VTABLE(u)->notify_cgroup_oom(u, managed_oom);
3960 }
3961
3962 static int unit_pid_set(Unit *u, Set **pid_set) {
3963 int r;
3964
3965 assert(u);
3966 assert(pid_set);
3967
3968 set_clear(*pid_set); /* This updates input. */
3969
3970 /* Exclude the main/control pids from being killed via the cgroup */
3971
3972 PidRef *pid;
3973 FOREACH_ARGUMENT(pid, unit_main_pid(u), unit_control_pid(u))
3974 if (pidref_is_set(pid)) {
3975 r = set_ensure_put(pid_set, NULL, PID_TO_PTR(pid->pid));
3976 if (r < 0)
3977 return r;
3978 }
3979
3980 return 0;
3981 }
3982
3983 static int kill_common_log(const PidRef *pid, int signo, void *userdata) {
3984 _cleanup_free_ char *comm = NULL;
3985 Unit *u = ASSERT_PTR(userdata);
3986
3987 (void) pidref_get_comm(pid, &comm);
3988
3989 log_unit_info(u, "Sending signal SIG%s to process " PID_FMT " (%s) on client request.",
3990 signal_to_string(signo), pid->pid, strna(comm));
3991
3992 return 1;
3993 }
3994
3995 static int kill_or_sigqueue(PidRef *pidref, int signo, int code, int value) {
3996 assert(pidref_is_set(pidref));
3997 assert(SIGNAL_VALID(signo));
3998
3999 switch (code) {
4000
4001 case SI_USER:
4002 log_debug("Killing " PID_FMT " with signal SIG%s.", pidref->pid, signal_to_string(signo));
4003 return pidref_kill(pidref, signo);
4004
4005 case SI_QUEUE:
4006 log_debug("Enqueuing value %i to " PID_FMT " on signal SIG%s.", value, pidref->pid, signal_to_string(signo));
4007 return pidref_sigqueue(pidref, signo, value);
4008
4009 default:
4010 assert_not_reached();
4011 }
4012 }
4013
4014 static int unit_kill_one(
4015 Unit *u,
4016 PidRef *pidref,
4017 const char *type,
4018 int signo,
4019 int code,
4020 int value,
4021 sd_bus_error *ret_error) {
4022
4023 int r;
4024
4025 assert(u);
4026 assert(type);
4027
4028 if (!pidref_is_set(pidref))
4029 return 0;
4030
4031 _cleanup_free_ char *comm = NULL;
4032 (void) pidref_get_comm(pidref, &comm);
4033
4034 r = kill_or_sigqueue(pidref, signo, code, value);
4035 if (r == -ESRCH)
4036 return 0;
4037 if (r < 0) {
4038 /* Report this failure both to the logs and to the client */
4039 if (ret_error)
4040 sd_bus_error_set_errnof(
4041 ret_error, r,
4042 "Failed to send signal SIG%s to %s process " PID_FMT " (%s): %m",
4043 signal_to_string(signo), type, pidref->pid, strna(comm));
4044
4045 return log_unit_warning_errno(
4046 u, r,
4047 "Failed to send signal SIG%s to %s process " PID_FMT " (%s) on client request: %m",
4048 signal_to_string(signo), type, pidref->pid, strna(comm));
4049 }
4050
4051 log_unit_info(u, "Sent signal SIG%s to %s process " PID_FMT " (%s) on client request.",
4052 signal_to_string(signo), type, pidref->pid, strna(comm));
4053 return 1; /* killed */
4054 }
4055
4056 int unit_kill(
4057 Unit *u,
4058 KillWhom whom,
4059 const char *subgroup,
4060 int signo,
4061 int code,
4062 int value,
4063 sd_bus_error *ret_error) {
4064
4065 PidRef *main_pid, *control_pid;
4066 bool killed = false;
4067 int ret = 0, r;
4068
4069 /* This is the common implementation for explicit user-requested killing of unit processes, shared by
4070 * various unit types. Do not confuse with unit_kill_context(), which is what we use when we want to
4071 * stop a service ourselves. */
4072
4073 assert(u);
4074 assert(whom >= 0);
4075 assert(whom < _KILL_WHOM_MAX);
4076 assert(SIGNAL_VALID(signo));
4077 assert(IN_SET(code, SI_USER, SI_QUEUE));
4078
4079 if (subgroup) {
4080 if (!IN_SET(whom, KILL_CGROUP, KILL_CGROUP_FAIL))
4081 return sd_bus_error_set(ret_error, SD_BUS_ERROR_NOT_SUPPORTED,
4082 "Killing by subgroup is only supported for 'cgroup' or 'cgroup-kill' modes.");
4083
4084 if (!unit_cgroup_delegate(u))
4085 return sd_bus_error_set(ret_error, SD_BUS_ERROR_NOT_SUPPORTED,
4086 "Killing by subgroup is only available for units with control group delegation enabled.");
4087 }
4088
4089 main_pid = unit_main_pid(u);
4090 control_pid = unit_control_pid(u);
4091
4092 if (!UNIT_HAS_CGROUP_CONTEXT(u) && !main_pid && !control_pid)
4093 return sd_bus_error_set(ret_error, SD_BUS_ERROR_NOT_SUPPORTED, "Unit type does not support process killing.");
4094
4095 if (IN_SET(whom, KILL_MAIN, KILL_MAIN_FAIL)) {
4096 if (!main_pid)
4097 return sd_bus_error_setf(ret_error, BUS_ERROR_NO_SUCH_PROCESS, "%s units have no main processes", unit_type_to_string(u->type));
4098 if (!pidref_is_set(main_pid))
4099 return sd_bus_error_set_const(ret_error, BUS_ERROR_NO_SUCH_PROCESS, "No main process to kill");
4100 }
4101
4102 if (IN_SET(whom, KILL_CONTROL, KILL_CONTROL_FAIL)) {
4103 if (!control_pid)
4104 return sd_bus_error_setf(ret_error, BUS_ERROR_NO_SUCH_PROCESS, "%s units have no control processes", unit_type_to_string(u->type));
4105 if (!pidref_is_set(control_pid))
4106 return sd_bus_error_set_const(ret_error, BUS_ERROR_NO_SUCH_PROCESS, "No control process to kill");
4107 }
4108
4109 if (IN_SET(whom, KILL_CONTROL, KILL_CONTROL_FAIL, KILL_ALL, KILL_ALL_FAIL)) {
4110 r = unit_kill_one(u, control_pid, "control", signo, code, value, ret_error);
4111 RET_GATHER(ret, r);
4112 killed = killed || r > 0;
4113 }
4114
4115 if (IN_SET(whom, KILL_MAIN, KILL_MAIN_FAIL, KILL_ALL, KILL_ALL_FAIL)) {
4116 r = unit_kill_one(u, main_pid, "main", signo, code, value, ret >= 0 ? ret_error : NULL);
4117 RET_GATHER(ret, r);
4118 killed = killed || r > 0;
4119 }
4120
4121 /* Note: if we shall enqueue rather than kill we won't do this via the cgroup mechanism, since it
4122 * doesn't really make much sense (and given that enqueued values are a relatively expensive
4123 * resource, and we shouldn't allow us to be subjects for such allocation sprees) */
4124 if (IN_SET(whom, KILL_ALL, KILL_ALL_FAIL, KILL_CGROUP, KILL_CGROUP_FAIL) && code == SI_USER) {
4125 CGroupRuntime *crt = unit_get_cgroup_runtime(u);
4126 if (crt && crt->cgroup_path) {
4127 _cleanup_set_free_ Set *pid_set = NULL;
4128 _cleanup_free_ char *joined = NULL;
4129 const char *p;
4130
4131 if (empty_or_root(subgroup))
4132 p = crt->cgroup_path;
4133 else {
4134 joined = path_join(crt->cgroup_path, subgroup);
4135 if (!joined)
4136 return -ENOMEM;
4137
4138 p = joined;
4139 }
4140
4141 if (signo == SIGKILL) {
4142 r = cg_kill_kernel_sigkill(p);
4143 if (r >= 0) {
4144 killed = true;
4145 log_unit_info(u, "Killed unit cgroup '%s' with SIGKILL on client request.", p);
4146 goto finish;
4147 }
4148 if (r != -EOPNOTSUPP) {
4149 if (ret >= 0)
4150 sd_bus_error_set_errnof(ret_error, r,
4151 "Failed to kill unit cgroup: %m");
4152 RET_GATHER(ret, log_unit_warning_errno(u, r, "Failed to kill unit cgroup '%s': %m", p));
4153 goto finish;
4154 }
4155 /* Fall back to manual enumeration */
4156 } else if (IN_SET(whom, KILL_ALL, KILL_ALL_FAIL)) {
4157 /* Exclude the main/control pids from being killed via the cgroup if not
4158 * SIGKILL */
4159 r = unit_pid_set(u, &pid_set);
4160 if (r < 0)
4161 return log_oom();
4162 }
4163
4164 r = cg_kill_recursive(p, signo, /* flags= */ 0, pid_set, kill_common_log, u);
4165 if (r < 0 && !IN_SET(r, -ESRCH, -ENOENT)) {
4166 if (ret >= 0)
4167 sd_bus_error_set_errnof(
4168 ret_error, r,
4169 "Failed to send signal SIG%s to processes in unit cgroup '%s': %m",
4170 signal_to_string(signo), p);
4171
4172 RET_GATHER(ret, log_unit_warning_errno(
4173 u, r,
4174 "Failed to send signal SIG%s to processes in unit cgroup '%s' on client request: %m",
4175 signal_to_string(signo), p));
4176 }
4177 killed = killed || r > 0;
4178 }
4179 }
4180
4181 finish:
4182 /* If the "fail" versions of the operation are requested, then complain if the set of processes we killed is empty */
4183 if (ret >= 0 && !killed && IN_SET(whom, KILL_ALL_FAIL, KILL_CONTROL_FAIL, KILL_MAIN_FAIL, KILL_CGROUP_FAIL))
4184 return sd_bus_error_set_const(ret_error, BUS_ERROR_NO_SUCH_PROCESS, "No matching processes to kill");
4185
4186 return ret;
4187 }
4188
4189 int unit_following_set(Unit *u, Set **s) {
4190 assert(u);
4191 assert(s);
4192
4193 if (UNIT_VTABLE(u)->following_set)
4194 return UNIT_VTABLE(u)->following_set(u, s);
4195
4196 *s = NULL;
4197 return 0;
4198 }
4199
4200 UnitFileState unit_get_unit_file_state(Unit *u) {
4201 int r;
4202
4203 assert(u);
4204
4205 if (u->unit_file_state >= 0 || !u->fragment_path)
4206 return u->unit_file_state;
4207
4208 /* If we know this is a transient unit no need to ask the unit file state for details. Let's bypass
4209 * the more expensive on-disk check. */
4210 if (u->transient)
4211 return (u->unit_file_state = UNIT_FILE_TRANSIENT);
4212
4213 r = unit_file_get_state(
4214 u->manager->runtime_scope,
4215 /* root_dir= */ NULL,
4216 u->id,
4217 &u->unit_file_state);
4218 if (r < 0)
4219 u->unit_file_state = UNIT_FILE_BAD;
4220
4221 return u->unit_file_state;
4222 }
4223
4224 PresetAction unit_get_unit_file_preset(Unit *u) {
4225 int r;
4226
4227 assert(u);
4228
4229 if (u->unit_file_preset >= 0)
4230 return u->unit_file_preset;
4231
4232 /* If this is a transient or perpetual unit file it doesn't make much sense to ask the preset
4233 * database about this, because enabling/disabling makes no sense for either. Hence don't. */
4234 if (!u->fragment_path || u->transient || u->perpetual)
4235 return (u->unit_file_preset = -ENOEXEC);
4236
4237 _cleanup_free_ char *bn = NULL;
4238 r = path_extract_filename(u->fragment_path, &bn);
4239 if (r < 0)
4240 return (u->unit_file_preset = r);
4241 if (r == O_DIRECTORY)
4242 return (u->unit_file_preset = -EISDIR);
4243
4244 return (u->unit_file_preset = unit_file_query_preset(
4245 u->manager->runtime_scope,
4246 /* root_dir= */ NULL,
4247 bn,
4248 /* cached= */ NULL));
4249 }
4250
4251 Unit* unit_ref_set(UnitRef *ref, Unit *source, Unit *target) {
4252 assert(ref);
4253 assert(source);
4254 assert(target);
4255
4256 if (ref->target)
4257 unit_ref_unset(ref);
4258
4259 ref->source = source;
4260 ref->target = target;
4261 LIST_PREPEND(refs_by_target, target->refs_by_target, ref);
4262 return target;
4263 }
4264
4265 void unit_ref_unset(UnitRef *ref) {
4266 assert(ref);
4267
4268 if (!ref->target)
4269 return;
4270
4271 /* We are about to drop a reference to the unit, make sure the garbage collection has a look at it as it might
4272 * be unreferenced now. */
4273 unit_add_to_gc_queue(ref->target);
4274
4275 LIST_REMOVE(refs_by_target, ref->target->refs_by_target, ref);
4276 ref->source = ref->target = NULL;
4277 }
4278
4279 static int user_from_unit_name(Unit *u, char **ret) {
4280
4281 static const uint8_t hash_key[] = {
4282 0x58, 0x1a, 0xaf, 0xe6, 0x28, 0x58, 0x4e, 0x96,
4283 0xb4, 0x4e, 0xf5, 0x3b, 0x8c, 0x92, 0x07, 0xec
4284 };
4285
4286 _cleanup_free_ char *n = NULL;
4287 int r;
4288
4289 r = unit_name_to_prefix(u->id, &n);
4290 if (r < 0)
4291 return r;
4292
4293 if (valid_user_group_name(n, 0)) {
4294 *ret = TAKE_PTR(n);
4295 return 0;
4296 }
4297
4298 /* If we can't use the unit name as a user name, then let's hash it and use that */
4299 if (asprintf(ret, "_du%016" PRIx64, siphash24(n, strlen(n), hash_key)) < 0)
4300 return -ENOMEM;
4301
4302 return 0;
4303 }
4304
4305 static int unit_verify_contexts(const Unit *u) {
4306 assert(u);
4307
4308 const ExecContext *ec = unit_get_exec_context(u);
4309 if (!ec)
4310 return 0;
4311
4312 if (MANAGER_IS_USER(u->manager) && ec->dynamic_user)
4313 return log_unit_error_errno(u, SYNTHETIC_ERRNO(ENOEXEC), "DynamicUser= enabled for user unit, which is not supported. Refusing.");
4314
4315 if (ec->dynamic_user && ec->working_directory_home)
4316 return log_unit_error_errno(u, SYNTHETIC_ERRNO(ENOEXEC), "WorkingDirectory=~ is not allowed under DynamicUser=yes. Refusing.");
4317
4318 if (ec->working_directory && path_below_api_vfs(ec->working_directory) &&
4319 exec_needs_mount_namespace(ec, /* params = */ NULL, /* runtime = */ NULL))
4320 return log_unit_error_errno(u, SYNTHETIC_ERRNO(ENOEXEC), "WorkingDirectory= may not be below /proc/, /sys/ or /dev/ when using mount namespacing. Refusing.");
4321
4322 if (exec_needs_pid_namespace(ec, /* params= */ NULL) && !UNIT_VTABLE(u)->notify_pidref)
4323 return log_unit_error_errno(u, SYNTHETIC_ERRNO(ENOEXEC), "PrivatePIDs= setting is only supported for service units. Refusing.");
4324
4325 const KillContext *kc = unit_get_kill_context(u);
4326
4327 if (ec->pam_name && kc && !IN_SET(kc->kill_mode, KILL_CONTROL_GROUP, KILL_MIXED))
4328 return log_unit_error_errno(u, SYNTHETIC_ERRNO(ENOEXEC), "Unit has PAM enabled. Kill mode must be set to 'control-group' or 'mixed'. Refusing.");
4329
4330 return 0;
4331 }
4332
4333 static PrivateTmp unit_get_private_var_tmp(const Unit *u, const ExecContext *c) {
4334 assert(u);
4335 assert(c);
4336 assert(c->private_tmp >= 0 && c->private_tmp < _PRIVATE_TMP_MAX);
4337
4338 /* Disable disconnected private tmpfs on /var/tmp/ when DefaultDependencies=no and
4339 * RootImage=/RootDirectory= are not set, as /var/ may be a separated partition.
4340 * See issue #37258. */
4341
4342 /* PrivateTmp=yes/no also enables/disables private tmpfs on /var/tmp/. */
4343 if (c->private_tmp != PRIVATE_TMP_DISCONNECTED)
4344 return c->private_tmp;
4345
4346 /* When DefaultDependencies=yes, disconnected tmpfs is also enabled on /var/tmp/, and an explicit
4347 * dependency to the mount on /var/ will be added in unit_add_exec_dependencies(). */
4348 if (u->default_dependencies)
4349 return PRIVATE_TMP_DISCONNECTED;
4350
4351 /* When RootImage=/RootDirectory= is enabled, /var/ should be prepared by the image or directory,
4352 * hence we can mount a disconnected tmpfs on /var/tmp/. */
4353 if (exec_context_with_rootfs(c))
4354 return PRIVATE_TMP_DISCONNECTED;
4355
4356 /* Even if DefaultDependencies=no, enable disconnected tmpfs when
4357 * RequiresMountsFor=/WantsMountsFor=/var/ is explicitly set. */
4358 for (UnitMountDependencyType t = 0; t < _UNIT_MOUNT_DEPENDENCY_TYPE_MAX; t++)
4359 if (hashmap_contains(u->mounts_for[t], "/var/"))
4360 return PRIVATE_TMP_DISCONNECTED;
4361
4362 /* Check the same but for After= with Requires=/Requisite=/Wants= or friends. */
4363 Unit *m = manager_get_unit(u->manager, "var.mount");
4364 if (!m)
4365 return PRIVATE_TMP_NO;
4366
4367 if (!unit_has_dependency(u, UNIT_ATOM_AFTER, m))
4368 return PRIVATE_TMP_NO;
4369
4370 if (unit_has_dependency(u, UNIT_ATOM_PULL_IN_START, m) ||
4371 unit_has_dependency(u, UNIT_ATOM_PULL_IN_VERIFY, m) ||
4372 unit_has_dependency(u, UNIT_ATOM_PULL_IN_START_IGNORED, m))
4373 return PRIVATE_TMP_DISCONNECTED;
4374
4375 return PRIVATE_TMP_NO;
4376 }
4377
4378 int unit_patch_contexts(Unit *u) {
4379 CGroupContext *cc;
4380 ExecContext *ec;
4381 int r;
4382
4383 assert(u);
4384
4385 /* Patch in the manager defaults into the exec and cgroup
4386 * contexts, _after_ the rest of the settings have been
4387 * initialized */
4388
4389 ec = unit_get_exec_context(u);
4390 if (ec) {
4391 /* This only copies in the ones that need memory */
4392 for (unsigned i = 0; i < _RLIMIT_MAX; i++)
4393 if (u->manager->defaults.rlimit[i] && !ec->rlimit[i]) {
4394 ec->rlimit[i] = newdup(struct rlimit, u->manager->defaults.rlimit[i], 1);
4395 if (!ec->rlimit[i])
4396 return -ENOMEM;
4397 }
4398
4399 if (MANAGER_IS_USER(u->manager) && !ec->working_directory) {
4400 r = get_home_dir(&ec->working_directory);
4401 if (r < 0)
4402 return r;
4403
4404 if (!ec->working_directory_home)
4405 /* If home directory is implied by us, allow it to be missing. */
4406 ec->working_directory_missing_ok = true;
4407 }
4408
4409 if (ec->private_devices)
4410 ec->capability_bounding_set &= ~((UINT64_C(1) << CAP_MKNOD) | (UINT64_C(1) << CAP_SYS_RAWIO));
4411
4412 if (ec->protect_kernel_modules)
4413 ec->capability_bounding_set &= ~(UINT64_C(1) << CAP_SYS_MODULE);
4414
4415 if (ec->protect_kernel_logs)
4416 ec->capability_bounding_set &= ~(UINT64_C(1) << CAP_SYSLOG);
4417
4418 if (ec->protect_clock)
4419 ec->capability_bounding_set &= ~((UINT64_C(1) << CAP_SYS_TIME) | (UINT64_C(1) << CAP_WAKE_ALARM));
4420
4421 if (ec->dynamic_user) {
4422 if (!ec->user) {
4423 r = user_from_unit_name(u, &ec->user);
4424 if (r < 0)
4425 return r;
4426 }
4427
4428 if (!ec->group) {
4429 ec->group = strdup(ec->user);
4430 if (!ec->group)
4431 return -ENOMEM;
4432 }
4433
4434 /* If the dynamic user option is on, let's make sure that the unit can't leave its
4435 * UID/GID around in the file system or on IPC objects. Hence enforce a strict
4436 * sandbox. */
4437
4438 /* With DynamicUser= we want private directories, so if the user hasn't manually
4439 * selected PrivateTmp=, enable it, but to a fully private (disconnected) tmpfs
4440 * instance. */
4441 if (ec->private_tmp == PRIVATE_TMP_NO)
4442 ec->private_tmp = PRIVATE_TMP_DISCONNECTED;
4443 ec->remove_ipc = true;
4444 ec->protect_system = PROTECT_SYSTEM_STRICT;
4445 if (ec->protect_home == PROTECT_HOME_NO)
4446 ec->protect_home = PROTECT_HOME_READ_ONLY;
4447
4448 /* Make sure this service can neither benefit from SUID/SGID binaries nor create
4449 * them. */
4450 ec->no_new_privileges = true;
4451 ec->restrict_suid_sgid = true;
4452 }
4453
4454 ec->private_var_tmp = unit_get_private_var_tmp(u, ec);
4455
4456 FOREACH_ARRAY(d, ec->directories, _EXEC_DIRECTORY_TYPE_MAX)
4457 exec_directory_sort(d);
4458 }
4459
4460 cc = unit_get_cgroup_context(u);
4461 if (cc && ec) {
4462
4463 if (ec->private_devices &&
4464 cc->device_policy == CGROUP_DEVICE_POLICY_AUTO)
4465 cc->device_policy = CGROUP_DEVICE_POLICY_CLOSED;
4466
4467 /* Only add these if needed, as they imply that everything else is blocked. */
4468 if (cgroup_context_has_device_policy(cc)) {
4469 if (ec->root_image || ec->mount_images) {
4470
4471 /* When RootImage= or MountImages= is specified, the following devices are touched. */
4472 FOREACH_STRING(p, "/dev/loop-control", "/dev/mapper/control") {
4473 r = cgroup_context_add_device_allow(cc, p, CGROUP_DEVICE_READ|CGROUP_DEVICE_WRITE);
4474 if (r < 0)
4475 return r;
4476 }
4477 FOREACH_STRING(p, "block-loop", "block-blkext", "block-device-mapper") {
4478 r = cgroup_context_add_device_allow(cc, p, CGROUP_DEVICE_READ|CGROUP_DEVICE_WRITE|CGROUP_DEVICE_MKNOD);
4479 if (r < 0)
4480 return r;
4481 }
4482
4483 /* Make sure "block-loop" can be resolved, i.e. make sure "loop" shows up in /proc/devices.
4484 * Same for mapper and verity. */
4485 FOREACH_STRING(p, "modprobe@loop.service", "modprobe@dm_mod.service", "modprobe@dm_verity.service") {
4486 r = unit_add_two_dependencies_by_name(u, UNIT_AFTER, UNIT_WANTS, p, true, UNIT_DEPENDENCY_FILE);
4487 if (r < 0)
4488 return r;
4489 }
4490 }
4491
4492 if (ec->protect_clock) {
4493 r = cgroup_context_add_device_allow(cc, "char-rtc", CGROUP_DEVICE_READ);
4494 if (r < 0)
4495 return r;
4496 }
4497 }
4498 }
4499
4500 return unit_verify_contexts(u);
4501 }
4502
4503 ExecContext *unit_get_exec_context(const Unit *u) {
4504 size_t offset;
4505 assert(u);
4506
4507 if (u->type < 0)
4508 return NULL;
4509
4510 offset = UNIT_VTABLE(u)->exec_context_offset;
4511 if (offset <= 0)
4512 return NULL;
4513
4514 return (ExecContext*) ((uint8_t*) u + offset);
4515 }
4516
4517 KillContext *unit_get_kill_context(const Unit *u) {
4518 size_t offset;
4519 assert(u);
4520
4521 if (u->type < 0)
4522 return NULL;
4523
4524 offset = UNIT_VTABLE(u)->kill_context_offset;
4525 if (offset <= 0)
4526 return NULL;
4527
4528 return (KillContext*) ((uint8_t*) u + offset);
4529 }
4530
4531 CGroupContext *unit_get_cgroup_context(const Unit *u) {
4532 size_t offset;
4533
4534 if (u->type < 0)
4535 return NULL;
4536
4537 offset = UNIT_VTABLE(u)->cgroup_context_offset;
4538 if (offset <= 0)
4539 return NULL;
4540
4541 return (CGroupContext*) ((uint8_t*) u + offset);
4542 }
4543
4544 ExecRuntime *unit_get_exec_runtime(const Unit *u) {
4545 size_t offset;
4546
4547 if (u->type < 0)
4548 return NULL;
4549
4550 offset = UNIT_VTABLE(u)->exec_runtime_offset;
4551 if (offset <= 0)
4552 return NULL;
4553
4554 return *(ExecRuntime**) ((uint8_t*) u + offset);
4555 }
4556
4557 CGroupRuntime *unit_get_cgroup_runtime(const Unit *u) {
4558 size_t offset;
4559
4560 if (u->type < 0)
4561 return NULL;
4562
4563 offset = UNIT_VTABLE(u)->cgroup_runtime_offset;
4564 if (offset <= 0)
4565 return NULL;
4566
4567 return *(CGroupRuntime**) ((uint8_t*) u + offset);
4568 }
4569
4570 static const char* unit_drop_in_dir(Unit *u, UnitWriteFlags flags) {
4571 assert(u);
4572
4573 if (UNIT_WRITE_FLAGS_NOOP(flags))
4574 return NULL;
4575
4576 if (u->transient) /* Redirect drop-ins for transient units always into the transient directory. */
4577 return u->manager->lookup_paths.transient;
4578
4579 if (flags & UNIT_PERSISTENT)
4580 return u->manager->lookup_paths.persistent_control;
4581
4582 if (flags & UNIT_RUNTIME)
4583 return u->manager->lookup_paths.runtime_control;
4584
4585 return NULL;
4586 }
4587
4588 const char* unit_escape_setting(const char *s, UnitWriteFlags flags, char **buf) {
4589 assert(s);
4590 assert(popcount(flags & (UNIT_ESCAPE_EXEC_SYNTAX_ENV | UNIT_ESCAPE_EXEC_SYNTAX | UNIT_ESCAPE_C)) <= 1);
4591 assert(buf);
4592
4593 _cleanup_free_ char *t = NULL;
4594
4595 /* Returns a string with any escaping done. If no escaping was necessary, *buf is set to NULL, and
4596 * the input pointer is returned as-is. If an allocation was needed, the return buffer pointer is
4597 * written to *buf. This means the return value always contains a properly escaped version, but *buf
4598 * only contains a pointer if an allocation was made. Callers can use this to optimize memory
4599 * allocations. */
4600
4601 if (flags & UNIT_ESCAPE_SPECIFIERS) {
4602 t = specifier_escape(s);
4603 if (!t)
4604 return NULL;
4605
4606 s = t;
4607 }
4608
4609 /* We either do C-escaping or shell-escaping, to additionally escape characters that we parse for
4610 * ExecStart= and friends, i.e. '$' and quotes. */
4611
4612 if (flags & (UNIT_ESCAPE_EXEC_SYNTAX_ENV | UNIT_ESCAPE_EXEC_SYNTAX)) {
4613 char *t2;
4614
4615 if (flags & UNIT_ESCAPE_EXEC_SYNTAX_ENV) {
4616 t2 = strreplace(s, "$", "$$");
4617 if (!t2)
4618 return NULL;
4619 free_and_replace(t, t2);
4620 }
4621
4622 t2 = shell_escape(t ?: s, "\"");
4623 if (!t2)
4624 return NULL;
4625 free_and_replace(t, t2);
4626
4627 s = t;
4628
4629 } else if (flags & UNIT_ESCAPE_C) {
4630 char *t2;
4631
4632 t2 = cescape(s);
4633 if (!t2)
4634 return NULL;
4635 free_and_replace(t, t2);
4636
4637 s = t;
4638 }
4639
4640 *buf = TAKE_PTR(t);
4641 return s;
4642 }
4643
4644 char* unit_concat_strv(char **l, UnitWriteFlags flags) {
4645 _cleanup_free_ char *result = NULL;
4646 size_t n = 0;
4647
4648 /* Takes a list of strings, escapes them, and concatenates them. This may be used to format command
4649 * lines in a way suitable for ExecStart= stanzas. */
4650
4651 STRV_FOREACH(i, l) {
4652 _cleanup_free_ char *buf = NULL;
4653 const char *p;
4654 size_t a;
4655 char *q;
4656
4657 p = unit_escape_setting(*i, flags, &buf);
4658 if (!p)
4659 return NULL;
4660
4661 a = (n > 0) + 1 + strlen(p) + 1; /* separating space + " + entry + " */
4662 if (!GREEDY_REALLOC(result, n + a + 1))
4663 return NULL;
4664
4665 q = result + n;
4666 if (n > 0)
4667 *(q++) = ' ';
4668
4669 *(q++) = '"';
4670 q = stpcpy(q, p);
4671 *(q++) = '"';
4672
4673 n += a;
4674 }
4675
4676 if (!GREEDY_REALLOC(result, n + 1))
4677 return NULL;
4678
4679 result[n] = 0;
4680
4681 return TAKE_PTR(result);
4682 }
4683
4684 int unit_write_setting(Unit *u, UnitWriteFlags flags, const char *name, const char *data) {
4685 _cleanup_free_ char *p = NULL, *q = NULL, *escaped = NULL;
4686 const char *dir, *wrapped;
4687 int r;
4688
4689 assert(u);
4690 assert(name);
4691 assert(data);
4692
4693 if (UNIT_WRITE_FLAGS_NOOP(flags))
4694 return 0;
4695
4696 data = unit_escape_setting(data, flags, &escaped);
4697 if (!data)
4698 return -ENOMEM;
4699
4700 /* Prefix the section header. If we are writing this out as transient file, then let's suppress this if the
4701 * previous section header is the same */
4702
4703 if (flags & UNIT_PRIVATE) {
4704 if (!UNIT_VTABLE(u)->private_section)
4705 return -EINVAL;
4706
4707 if (!u->transient_file || u->last_section_private < 0)
4708 data = strjoina("[", UNIT_VTABLE(u)->private_section, "]\n", data);
4709 else if (u->last_section_private == 0)
4710 data = strjoina("\n[", UNIT_VTABLE(u)->private_section, "]\n", data);
4711 } else {
4712 if (!u->transient_file || u->last_section_private < 0)
4713 data = strjoina("[Unit]\n", data);
4714 else if (u->last_section_private > 0)
4715 data = strjoina("\n[Unit]\n", data);
4716 }
4717
4718 if (u->transient_file) {
4719 /* When this is a transient unit file in creation, then let's not create a new drop-in,
4720 * but instead write to the transient unit file. */
4721 fputs_with_newline(u->transient_file, data);
4722
4723 /* Remember which section we wrote this entry to */
4724 u->last_section_private = !!(flags & UNIT_PRIVATE);
4725 return 0;
4726 }
4727
4728 dir = unit_drop_in_dir(u, flags);
4729 if (!dir)
4730 return -EINVAL;
4731
4732 wrapped = strjoina("# This is a drop-in unit file extension, created via \"systemctl set-property\"\n"
4733 "# or an equivalent operation. Do not edit.\n",
4734 data,
4735 "\n");
4736
4737 r = drop_in_file(dir, u->id, 50, name, &p, &q);
4738 if (r < 0)
4739 return r;
4740
4741 (void) mkdir_p_label(p, 0755);
4742
4743 /* Make sure the drop-in dir is registered in our path cache. This way we don't need to stupidly
4744 * recreate the cache after every drop-in we write. */
4745 if (u->manager->unit_path_cache) {
4746 r = set_put_strdup_full(&u->manager->unit_path_cache, &path_hash_ops_free, p);
4747 if (r < 0)
4748 return r;
4749 }
4750
4751 r = write_string_file(q, wrapped, WRITE_STRING_FILE_CREATE|WRITE_STRING_FILE_ATOMIC|WRITE_STRING_FILE_LABEL);
4752 if (r < 0)
4753 return r;
4754
4755 r = strv_push(&u->dropin_paths, q);
4756 if (r < 0)
4757 return r;
4758 q = NULL;
4759
4760 strv_uniq(u->dropin_paths);
4761
4762 u->dropin_mtime = now(CLOCK_REALTIME);
4763
4764 return 0;
4765 }
4766
4767 int unit_write_settingf(Unit *u, UnitWriteFlags flags, const char *name, const char *format, ...) {
4768 _cleanup_free_ char *p = NULL;
4769 va_list ap;
4770 int r;
4771
4772 assert(u);
4773 assert(name);
4774 assert(format);
4775
4776 if (UNIT_WRITE_FLAGS_NOOP(flags))
4777 return 0;
4778
4779 va_start(ap, format);
4780 r = vasprintf(&p, format, ap);
4781 va_end(ap);
4782
4783 if (r < 0)
4784 return -ENOMEM;
4785
4786 return unit_write_setting(u, flags, name, p);
4787 }
4788
4789 int unit_make_transient(Unit *u) {
4790 _cleanup_free_ char *path = NULL;
4791 FILE *f;
4792
4793 assert(u);
4794
4795 if (!UNIT_VTABLE(u)->can_transient)
4796 return -EOPNOTSUPP;
4797
4798 (void) mkdir_p_label(u->manager->lookup_paths.transient, 0755);
4799
4800 path = path_join(u->manager->lookup_paths.transient, u->id);
4801 if (!path)
4802 return -ENOMEM;
4803
4804 /* Let's open the file we'll write the transient settings into. This file is kept open as long as we are
4805 * creating the transient, and is closed in unit_load(), as soon as we start loading the file. */
4806
4807 WITH_UMASK(0022) {
4808 f = fopen(path, "we");
4809 if (!f)
4810 return -errno;
4811 }
4812
4813 safe_fclose(u->transient_file);
4814 u->transient_file = f;
4815
4816 free_and_replace(u->fragment_path, path);
4817
4818 u->source_path = mfree(u->source_path);
4819 u->dropin_paths = strv_free(u->dropin_paths);
4820 u->fragment_mtime = u->source_mtime = u->dropin_mtime = 0;
4821
4822 u->load_state = UNIT_STUB;
4823 u->load_error = 0;
4824 u->transient = true;
4825
4826 unit_add_to_dbus_queue(u);
4827 unit_add_to_gc_queue(u);
4828
4829 fputs("# This is a transient unit file, created programmatically via the systemd API. Do not edit.\n",
4830 u->transient_file);
4831
4832 return 0;
4833 }
4834
4835 static bool ignore_leftover_process(const char *comm) {
4836 return comm && comm[0] == '('; /* Most likely our own helper process (PAM?), ignore */
4837 }
4838
4839 static int log_kill(const PidRef *pid, int sig, void *userdata) {
4840 const Unit *u = ASSERT_PTR(userdata);
4841 _cleanup_free_ char *comm = NULL;
4842
4843 assert(pidref_is_set(pid));
4844
4845 (void) pidref_get_comm(pid, &comm);
4846
4847 if (ignore_leftover_process(comm))
4848 /* Although we didn't log anything, as this callback is used in unit_kill_context we must return 1
4849 * here to let the manager know that a process was killed. */
4850 return 1;
4851
4852 log_unit_notice(u,
4853 "Killing process " PID_FMT " (%s) with signal SIG%s.",
4854 pid->pid,
4855 strna(comm),
4856 signal_to_string(sig));
4857
4858 return 1;
4859 }
4860
4861 static int operation_to_signal(
4862 const KillContext *c,
4863 KillOperation k,
4864 bool *ret_noteworthy) {
4865
4866 assert(c);
4867 assert(ret_noteworthy);
4868
4869 switch (k) {
4870
4871 case KILL_TERMINATE:
4872 case KILL_TERMINATE_AND_LOG:
4873 *ret_noteworthy = k == KILL_TERMINATE_AND_LOG;
4874 return c->kill_signal;
4875
4876 case KILL_RESTART:
4877 *ret_noteworthy = false;
4878 return restart_kill_signal(c);
4879
4880 case KILL_KILL:
4881 *ret_noteworthy = true;
4882 return c->final_kill_signal;
4883
4884 case KILL_WATCHDOG:
4885 *ret_noteworthy = true;
4886 return c->watchdog_signal;
4887
4888 default:
4889 assert_not_reached();
4890 }
4891 }
4892
4893 static int unit_kill_context_one(
4894 Unit *u,
4895 const PidRef *pidref,
4896 const char *type,
4897 bool is_alien,
4898 int sig,
4899 bool send_sighup,
4900 cg_kill_log_func_t log_func) {
4901
4902 int r;
4903
4904 assert(u);
4905 assert(type);
4906
4907 /* This returns > 0 if it makes sense to wait for SIGCHLD for the process, == 0 if not. */
4908
4909 if (!pidref_is_set(pidref))
4910 return 0;
4911
4912 if (log_func)
4913 log_func(pidref, sig, u);
4914
4915 r = pidref_kill_and_sigcont(pidref, sig);
4916 if (r == -ESRCH)
4917 return !is_alien;
4918 if (r < 0) {
4919 _cleanup_free_ char *comm = NULL;
4920
4921 (void) pidref_get_comm(pidref, &comm);
4922 return log_unit_warning_errno(u, r, "Failed to kill %s process " PID_FMT " (%s), ignoring: %m", type, pidref->pid, strna(comm));
4923 }
4924
4925 if (send_sighup)
4926 (void) pidref_kill(pidref, SIGHUP);
4927
4928 return !is_alien;
4929 }
4930
4931 int unit_kill_context(Unit *u, KillOperation k) {
4932 bool wait_for_exit = false, send_sighup;
4933 cg_kill_log_func_t log_func = NULL;
4934 int sig, r;
4935
4936 assert(u);
4937
4938 /* Kill the processes belonging to this unit, in preparation for shutting the unit down. Returns > 0
4939 * if we killed something worth waiting for, 0 otherwise. Do not confuse with unit_kill_common()
4940 * which is used for user-requested killing of unit processes. */
4941
4942 KillContext *c = unit_get_kill_context(u);
4943 if (!c || c->kill_mode == KILL_NONE)
4944 return 0;
4945
4946 bool noteworthy;
4947 sig = operation_to_signal(c, k, &noteworthy);
4948 if (noteworthy)
4949 log_func = log_kill;
4950
4951 send_sighup =
4952 c->send_sighup &&
4953 IN_SET(k, KILL_TERMINATE, KILL_TERMINATE_AND_LOG) &&
4954 sig != SIGHUP;
4955
4956 bool is_alien;
4957 PidRef *main_pid = unit_main_pid_full(u, &is_alien);
4958 r = unit_kill_context_one(u, main_pid, "main", is_alien, sig, send_sighup, log_func);
4959 wait_for_exit = wait_for_exit || r > 0;
4960
4961 r = unit_kill_context_one(u, unit_control_pid(u), "control", /* is_alien = */ false, sig, send_sighup, log_func);
4962 wait_for_exit = wait_for_exit || r > 0;
4963
4964 CGroupRuntime *crt = unit_get_cgroup_runtime(u);
4965 if (crt && crt->cgroup_path &&
4966 (c->kill_mode == KILL_CONTROL_GROUP || (c->kill_mode == KILL_MIXED && k == KILL_KILL))) {
4967 _cleanup_set_free_ Set *pid_set = NULL;
4968
4969 /* Exclude the main/control pids from being killed via the cgroup */
4970 r = unit_pid_set(u, &pid_set);
4971 if (r < 0)
4972 return r;
4973
4974 r = cg_kill_recursive(
4975 crt->cgroup_path,
4976 sig,
4977 CGROUP_SIGCONT|CGROUP_IGNORE_SELF,
4978 pid_set,
4979 log_func, u);
4980 if (r < 0) {
4981 if (!IN_SET(r, -EAGAIN, -ESRCH, -ENOENT))
4982 log_unit_warning_errno(u, r, "Failed to kill control group %s, ignoring: %m", empty_to_root(crt->cgroup_path));
4983
4984 } else if (r > 0) {
4985
4986 wait_for_exit = true;
4987
4988 if (send_sighup) {
4989 r = unit_pid_set(u, &pid_set);
4990 if (r < 0)
4991 return r;
4992
4993 (void) cg_kill_recursive(
4994 crt->cgroup_path,
4995 SIGHUP,
4996 CGROUP_IGNORE_SELF,
4997 pid_set,
4998 /* log_kill= */ NULL,
4999 /* userdata= */ NULL);
5000 }
5001 }
5002 }
5003
5004 return wait_for_exit;
5005 }
5006
5007 int unit_add_mounts_for(Unit *u, const char *path, UnitDependencyMask mask, UnitMountDependencyType type) {
5008 Hashmap **unit_map, **manager_map;
5009 int r;
5010
5011 assert(u);
5012 assert(path);
5013 assert(type >= 0 && type < _UNIT_MOUNT_DEPENDENCY_TYPE_MAX);
5014
5015 unit_map = &u->mounts_for[type];
5016 manager_map = &u->manager->units_needing_mounts_for[type];
5017
5018 /* Registers a unit for requiring a certain path and all its prefixes. We keep a hashtable of these
5019 * paths in the unit (from the path to the UnitDependencyInfo structure indicating how to the
5020 * dependency came to be). However, we build a prefix table for all possible prefixes so that new
5021 * appearing mount units can easily determine which units to make themselves a dependency of. */
5022
5023 if (!path_is_absolute(path))
5024 return -EINVAL;
5025
5026 if (hashmap_contains(*unit_map, path)) /* Exit quickly if the path is already covered. */
5027 return 0;
5028
5029 /* Use the canonical form of the path as the stored key. We call path_is_normalized()
5030 * only after simplification, since path_is_normalized() rejects paths with '.'.
5031 * path_is_normalized() also verifies that the path fits in PATH_MAX. */
5032 _cleanup_free_ char *p = NULL;
5033 r = path_simplify_alloc(path, &p);
5034 if (r < 0)
5035 return r;
5036 path = p;
5037
5038 if (!path_is_normalized(path))
5039 return -EPERM;
5040
5041 UnitDependencyInfo di = {
5042 .origin_mask = mask
5043 };
5044
5045 r = hashmap_ensure_put(unit_map, &path_hash_ops, p, di.data);
5046 if (r < 0)
5047 return r;
5048 assert(r > 0);
5049 TAKE_PTR(p); /* path remains a valid pointer to the string stored in the hashmap */
5050
5051 char prefix[strlen(path) + 1];
5052 PATH_FOREACH_PREFIX_MORE(prefix, path) {
5053 Set *x;
5054
5055 x = hashmap_get(*manager_map, prefix);
5056 if (!x) {
5057 _cleanup_free_ char *q = NULL;
5058
5059 r = hashmap_ensure_allocated(manager_map, &path_hash_ops);
5060 if (r < 0)
5061 return r;
5062
5063 q = strdup(prefix);
5064 if (!q)
5065 return -ENOMEM;
5066
5067 x = set_new(NULL);
5068 if (!x)
5069 return -ENOMEM;
5070
5071 r = hashmap_put(*manager_map, q, x);
5072 if (r < 0) {
5073 set_free(x);
5074 return r;
5075 }
5076 q = NULL;
5077 }
5078
5079 r = set_put(x, u);
5080 if (r < 0)
5081 return r;
5082 }
5083
5084 return 0;
5085 }
5086
5087 int unit_setup_exec_runtime(Unit *u) {
5088 _cleanup_(exec_shared_runtime_unrefp) ExecSharedRuntime *esr = NULL;
5089 _cleanup_(dynamic_creds_unrefp) DynamicCreds *dcreds = NULL;
5090 _cleanup_set_free_ Set *units = NULL;
5091 ExecRuntime **rt;
5092 ExecContext *ec;
5093 size_t offset;
5094 Unit *other;
5095 int r;
5096
5097 offset = UNIT_VTABLE(u)->exec_runtime_offset;
5098 assert(offset > 0);
5099
5100 /* Check if there already is an ExecRuntime for this unit? */
5101 rt = (ExecRuntime**) ((uint8_t*) u + offset);
5102 if (*rt)
5103 return 0;
5104
5105 ec = ASSERT_PTR(unit_get_exec_context(u));
5106
5107 r = unit_get_transitive_dependency_set(u, UNIT_ATOM_JOINS_NAMESPACE_OF, &units);
5108 if (r < 0)
5109 return r;
5110
5111 /* Try to get it from somebody else */
5112 SET_FOREACH(other, units) {
5113 r = exec_shared_runtime_acquire(u->manager, NULL, other->id, false, &esr);
5114 if (r < 0)
5115 return r;
5116 if (r > 0)
5117 break;
5118 }
5119
5120 if (!esr) {
5121 r = exec_shared_runtime_acquire(u->manager, ec, u->id, true, &esr);
5122 if (r < 0)
5123 return r;
5124 }
5125
5126 if (ec->dynamic_user) {
5127 r = dynamic_creds_make(u->manager, ec->user, ec->group, &dcreds);
5128 if (r < 0)
5129 return r;
5130 }
5131
5132 r = exec_runtime_make(u, ec, esr, dcreds, rt);
5133 if (r < 0)
5134 return r;
5135
5136 TAKE_PTR(esr);
5137 TAKE_PTR(dcreds);
5138
5139 return r;
5140 }
5141
5142 CGroupRuntime *unit_setup_cgroup_runtime(Unit *u) {
5143 size_t offset;
5144
5145 assert(u);
5146
5147 offset = UNIT_VTABLE(u)->cgroup_runtime_offset;
5148 assert(offset > 0);
5149
5150 CGroupRuntime **rt = (CGroupRuntime**) ((uint8_t*) u + offset);
5151 if (*rt)
5152 return *rt;
5153
5154 return (*rt = cgroup_runtime_new());
5155 }
5156
5157 bool unit_type_supported(UnitType t) {
5158 static int8_t cache[_UNIT_TYPE_MAX] = {}; /* -1: disabled, 1: enabled: 0: don't know */
5159 int r;
5160
5161 assert(t >= 0 && t < _UNIT_TYPE_MAX);
5162
5163 if (cache[t] == 0) {
5164 char *e;
5165
5166 e = strjoina("SYSTEMD_SUPPORT_", unit_type_to_string(t));
5167
5168 r = getenv_bool(ascii_strupper(e));
5169 if (r < 0 && r != -ENXIO)
5170 log_debug_errno(r, "Failed to parse $%s, ignoring: %m", e);
5171
5172 cache[t] = r == 0 ? -1 : 1;
5173 }
5174 if (cache[t] < 0)
5175 return false;
5176
5177 if (!unit_vtable[t]->supported)
5178 return true;
5179
5180 return unit_vtable[t]->supported();
5181 }
5182
5183 void unit_warn_if_dir_nonempty(Unit *u, const char* where) {
5184 int r;
5185
5186 assert(u);
5187 assert(where);
5188
5189 if (!unit_log_level_test(u, LOG_NOTICE))
5190 return;
5191
5192 r = dir_is_empty(where, /* ignore_hidden_or_backup= */ false);
5193 if (r > 0 || r == -ENOTDIR)
5194 return;
5195 if (r < 0) {
5196 log_unit_warning_errno(u, r, "Failed to check directory %s: %m", where);
5197 return;
5198 }
5199
5200 log_unit_struct(u, LOG_NOTICE,
5201 LOG_MESSAGE_ID(SD_MESSAGE_OVERMOUNTING_STR),
5202 LOG_UNIT_INVOCATION_ID(u),
5203 LOG_UNIT_MESSAGE(u, "Directory %s to mount over is not empty, mounting anyway.", where),
5204 LOG_ITEM("WHERE=%s", where));
5205 }
5206
5207 int unit_log_noncanonical_mount_path(Unit *u, const char *where) {
5208 assert(u);
5209 assert(where);
5210
5211 /* No need to mention "." or "..", they would already have been rejected by unit_name_from_path() */
5212 log_unit_struct(u, LOG_ERR,
5213 LOG_MESSAGE_ID(SD_MESSAGE_NON_CANONICAL_MOUNT_STR),
5214 LOG_UNIT_INVOCATION_ID(u),
5215 LOG_UNIT_MESSAGE(u, "Mount path %s is not canonical (contains a symlink).", where),
5216 LOG_ITEM("WHERE=%s", where));
5217
5218 return -ELOOP;
5219 }
5220
5221 int unit_fail_if_noncanonical_mount_path(Unit *u, const char* where) {
5222 int r;
5223
5224 assert(u);
5225 assert(where);
5226
5227 _cleanup_free_ char *canonical_where = NULL;
5228 r = chase(where, /* root= */ NULL, CHASE_NONEXISTENT, &canonical_where, /* ret_fd= */ NULL);
5229 if (r < 0) {
5230 log_unit_debug_errno(u, r, "Failed to check %s for symlinks, ignoring: %m", where);
5231 return 0;
5232 }
5233
5234 /* We will happily ignore a trailing slash (or any redundant slashes) */
5235 if (path_equal(where, canonical_where))
5236 return 0;
5237
5238 return unit_log_noncanonical_mount_path(u, where);
5239 }
5240
5241 bool unit_is_pristine(Unit *u) {
5242 assert(u);
5243
5244 /* Check if the unit already exists or is already around, in a number of different ways. Note that to
5245 * cater for unit types such as slice, we are generally fine with units that are marked UNIT_LOADED
5246 * even though nothing was actually loaded, as those unit types don't require a file on disk.
5247 *
5248 * Note that we don't check for drop-ins here, because we allow drop-ins for transient units
5249 * identically to non-transient units, both unit-specific and hierarchical. E.g. for a-b-c.service:
5250 * service.d/….conf, a-.service.d/….conf, a-b-.service.d/….conf, a-b-c.service.d/….conf.
5251 */
5252
5253 return IN_SET(u->load_state, UNIT_NOT_FOUND, UNIT_LOADED) &&
5254 !u->fragment_path &&
5255 !u->source_path &&
5256 !u->job &&
5257 !u->merged_into;
5258 }
5259
5260 PidRef* unit_control_pid(Unit *u) {
5261 assert(u);
5262
5263 if (UNIT_VTABLE(u)->control_pid)
5264 return UNIT_VTABLE(u)->control_pid(u);
5265
5266 return NULL;
5267 }
5268
5269 PidRef* unit_main_pid_full(Unit *u, bool *ret_is_alien) {
5270 assert(u);
5271
5272 if (UNIT_VTABLE(u)->main_pid)
5273 return UNIT_VTABLE(u)->main_pid(u, ret_is_alien);
5274
5275 if (ret_is_alien)
5276 *ret_is_alien = false;
5277 return NULL;
5278 }
5279
5280 static void unit_modify_user_nft_set(Unit *u, bool add, NFTSetSource source, uint32_t element) {
5281 int r;
5282
5283 assert(u);
5284
5285 if (!MANAGER_IS_SYSTEM(u->manager))
5286 return;
5287
5288 CGroupContext *c;
5289 c = unit_get_cgroup_context(u);
5290 if (!c)
5291 return;
5292
5293 if (!u->manager->fw_ctx) {
5294 r = fw_ctx_new_full(&u->manager->fw_ctx, /* init_tables= */ false);
5295 if (r < 0)
5296 return;
5297
5298 assert(u->manager->fw_ctx);
5299 }
5300
5301 FOREACH_ARRAY(nft_set, c->nft_set_context.sets, c->nft_set_context.n_sets) {
5302 if (nft_set->source != source)
5303 continue;
5304
5305 r = nft_set_element_modify_any(u->manager->fw_ctx, add, nft_set->nfproto, nft_set->table, nft_set->set, &element, sizeof(element));
5306 if (r < 0)
5307 log_warning_errno(r, "Failed to %s NFT set entry: family %s, table %s, set %s, ID %u, ignoring: %m",
5308 add? "add" : "delete", nfproto_to_string(nft_set->nfproto), nft_set->table, nft_set->set, element);
5309 else
5310 log_debug("%s NFT set entry: family %s, table %s, set %s, ID %u",
5311 add? "Added" : "Deleted", nfproto_to_string(nft_set->nfproto), nft_set->table, nft_set->set, element);
5312 }
5313 }
5314
5315 static void unit_unref_uid_internal(
5316 Unit *u,
5317 uid_t *ref_uid,
5318 bool destroy_now,
5319 void (*_manager_unref_uid)(Manager *m, uid_t uid, bool destroy_now)) {
5320
5321 assert(u);
5322 assert(ref_uid);
5323 assert(_manager_unref_uid);
5324
5325 /* Generic implementation of both unit_unref_uid() and unit_unref_gid(), under the assumption that uid_t and
5326 * gid_t are actually the same time, with the same validity rules.
5327 *
5328 * Drops a reference to UID/GID from a unit. */
5329
5330 assert_cc(sizeof(uid_t) == sizeof(gid_t));
5331 assert_cc(UID_INVALID == (uid_t) GID_INVALID);
5332
5333 if (!uid_is_valid(*ref_uid))
5334 return;
5335
5336 _manager_unref_uid(u->manager, *ref_uid, destroy_now);
5337 *ref_uid = UID_INVALID;
5338 }
5339
5340 static void unit_unref_uid(Unit *u, bool destroy_now) {
5341 assert(u);
5342
5343 unit_modify_user_nft_set(u, /* add = */ false, NFT_SET_SOURCE_USER, u->ref_uid);
5344
5345 unit_unref_uid_internal(u, &u->ref_uid, destroy_now, manager_unref_uid);
5346 }
5347
5348 static void unit_unref_gid(Unit *u, bool destroy_now) {
5349 assert(u);
5350
5351 unit_modify_user_nft_set(u, /* add = */ false, NFT_SET_SOURCE_GROUP, u->ref_gid);
5352
5353 unit_unref_uid_internal(u, (uid_t*) &u->ref_gid, destroy_now, manager_unref_gid);
5354 }
5355
5356 void unit_unref_uid_gid(Unit *u, bool destroy_now) {
5357 assert(u);
5358
5359 unit_unref_uid(u, destroy_now);
5360 unit_unref_gid(u, destroy_now);
5361 }
5362
5363 static int unit_ref_uid_internal(
5364 Unit *u,
5365 uid_t *ref_uid,
5366 uid_t uid,
5367 bool clean_ipc,
5368 int (*_manager_ref_uid)(Manager *m, uid_t uid, bool clean_ipc)) {
5369
5370 int r;
5371
5372 assert(u);
5373 assert(ref_uid);
5374 assert(uid_is_valid(uid));
5375 assert(_manager_ref_uid);
5376
5377 /* Generic implementation of both unit_ref_uid() and unit_ref_guid(), under the assumption that uid_t and gid_t
5378 * are actually the same type, and have the same validity rules.
5379 *
5380 * Adds a reference on a specific UID/GID to this unit. Each unit referencing the same UID/GID maintains a
5381 * reference so that we can destroy the UID/GID's IPC resources as soon as this is requested and the counter
5382 * drops to zero. */
5383
5384 assert_cc(sizeof(uid_t) == sizeof(gid_t));
5385 assert_cc(UID_INVALID == (uid_t) GID_INVALID);
5386
5387 if (*ref_uid == uid)
5388 return 0;
5389
5390 if (uid_is_valid(*ref_uid)) /* Already set? */
5391 return -EBUSY;
5392
5393 r = _manager_ref_uid(u->manager, uid, clean_ipc);
5394 if (r < 0)
5395 return r;
5396
5397 *ref_uid = uid;
5398 return 1;
5399 }
5400
5401 static int unit_ref_uid(Unit *u, uid_t uid, bool clean_ipc) {
5402 return unit_ref_uid_internal(u, &u->ref_uid, uid, clean_ipc, manager_ref_uid);
5403 }
5404
5405 static int unit_ref_gid(Unit *u, gid_t gid, bool clean_ipc) {
5406 return unit_ref_uid_internal(u, (uid_t*) &u->ref_gid, (uid_t) gid, clean_ipc, manager_ref_gid);
5407 }
5408
5409 static int unit_ref_uid_gid_internal(Unit *u, uid_t uid, gid_t gid, bool clean_ipc) {
5410 int r = 0, q = 0;
5411
5412 assert(u);
5413
5414 /* Reference both a UID and a GID in one go. Either references both, or neither. */
5415
5416 if (uid_is_valid(uid)) {
5417 r = unit_ref_uid(u, uid, clean_ipc);
5418 if (r < 0)
5419 return r;
5420 }
5421
5422 if (gid_is_valid(gid)) {
5423 q = unit_ref_gid(u, gid, clean_ipc);
5424 if (q < 0) {
5425 if (r > 0)
5426 unit_unref_uid(u, false);
5427
5428 return q;
5429 }
5430 }
5431
5432 return r > 0 || q > 0;
5433 }
5434
5435 int unit_ref_uid_gid(Unit *u, uid_t uid, gid_t gid) {
5436 ExecContext *c;
5437 int r;
5438
5439 assert(u);
5440
5441 c = unit_get_exec_context(u);
5442
5443 r = unit_ref_uid_gid_internal(u, uid, gid, c ? c->remove_ipc : false);
5444 if (r < 0)
5445 return log_unit_warning_errno(u, r, "Couldn't add UID/GID reference to unit, proceeding without: %m");
5446
5447 unit_modify_user_nft_set(u, /* add = */ true, NFT_SET_SOURCE_USER, uid);
5448 unit_modify_user_nft_set(u, /* add = */ true, NFT_SET_SOURCE_GROUP, gid);
5449
5450 return r;
5451 }
5452
5453 void unit_notify_user_lookup(Unit *u, uid_t uid, gid_t gid) {
5454 int r;
5455
5456 assert(u);
5457
5458 /* This is invoked whenever one of the forked off processes let's us know the UID/GID its user name/group names
5459 * resolved to. We keep track of which UID/GID is currently assigned in order to be able to destroy its IPC
5460 * objects when no service references the UID/GID anymore. */
5461
5462 r = unit_ref_uid_gid(u, uid, gid);
5463 if (r > 0)
5464 unit_add_to_dbus_queue(u);
5465 }
5466
5467 int unit_acquire_invocation_id(Unit *u) {
5468 sd_id128_t id;
5469 int r;
5470
5471 assert(u);
5472
5473 r = sd_id128_randomize(&id);
5474 if (r < 0)
5475 return log_unit_error_errno(u, r, "Failed to generate invocation ID for unit: %m");
5476
5477 r = unit_set_invocation_id(u, id);
5478 if (r < 0)
5479 return log_unit_error_errno(u, r, "Failed to set invocation ID for unit: %m");
5480
5481 unit_add_to_dbus_queue(u);
5482 return 0;
5483 }
5484
5485 int unit_set_exec_params(Unit *u, ExecParameters *p) {
5486 int r;
5487
5488 assert(u);
5489 assert(p);
5490
5491 /* Copy parameters from manager */
5492 r = manager_get_effective_environment(u->manager, &p->environment);
5493 if (r < 0)
5494 return r;
5495
5496 p->runtime_scope = u->manager->runtime_scope;
5497
5498 r = strdup_to(&p->confirm_spawn, manager_get_confirm_spawn(u->manager));
5499 if (r < 0)
5500 return r;
5501
5502 p->cgroup_supported = u->manager->cgroup_supported;
5503 p->prefix = u->manager->prefix;
5504 SET_FLAG(p->flags, EXEC_PASS_LOG_UNIT|EXEC_CHOWN_DIRECTORIES, MANAGER_IS_SYSTEM(u->manager));
5505
5506 /* Copy parameters from unit */
5507 CGroupRuntime *crt = unit_get_cgroup_runtime(u);
5508 p->cgroup_path = crt ? crt->cgroup_path : NULL;
5509 SET_FLAG(p->flags, EXEC_CGROUP_DELEGATE, unit_cgroup_delegate(u));
5510
5511 p->received_credentials_directory = u->manager->received_credentials_directory;
5512 p->received_encrypted_credentials_directory = u->manager->received_encrypted_credentials_directory;
5513
5514 p->shall_confirm_spawn = u->manager->confirm_spawn;
5515
5516 p->fallback_smack_process_label = u->manager->defaults.smack_process_label;
5517
5518 if (u->manager->restrict_fs && p->bpf_restrict_fs_map_fd < 0) {
5519 int fd = bpf_restrict_fs_map_fd(u);
5520 if (fd < 0)
5521 return fd;
5522
5523 p->bpf_restrict_fs_map_fd = fd;
5524 }
5525
5526 p->user_lookup_fd = u->manager->user_lookup_fds[1];
5527 p->handoff_timestamp_fd = u->manager->handoff_timestamp_fds[1];
5528 if (UNIT_VTABLE(u)->notify_pidref)
5529 p->pidref_transport_fd = u->manager->pidref_transport_fds[1];
5530
5531 p->cgroup_id = crt ? crt->cgroup_id : 0;
5532 p->invocation_id = u->invocation_id;
5533 sd_id128_to_string(p->invocation_id, p->invocation_id_string);
5534 p->unit_id = strdup(u->id);
5535 if (!p->unit_id)
5536 return -ENOMEM;
5537
5538 p->debug_invocation = u->debug_invocation;
5539
5540 return 0;
5541 }
5542
5543 int unit_fork_helper_process(Unit *u, const char *name, bool into_cgroup, PidRef *ret) {
5544 CGroupRuntime *crt = NULL;
5545 pid_t pid;
5546 int r;
5547
5548 assert(u);
5549 assert(ret);
5550
5551 /* Forks off a helper process and makes sure it is a member of the unit's cgroup, if configured to
5552 * do so. Returns == 0 in the child, and > 0 in the parent. The pid parameter is always filled in
5553 * with the child's PID. */
5554
5555 if (into_cgroup) {
5556 (void) unit_realize_cgroup(u);
5557
5558 crt = unit_setup_cgroup_runtime(u);
5559 if (!crt)
5560 return -ENOMEM;
5561 }
5562
5563 r = safe_fork(name, FORK_REOPEN_LOG|FORK_DEATHSIG_SIGTERM, &pid);
5564 if (r < 0)
5565 return r;
5566 if (r > 0) {
5567 _cleanup_(pidref_done) PidRef pidref = PIDREF_NULL;
5568 int q;
5569
5570 /* Parent */
5571
5572 q = pidref_set_pid(&pidref, pid);
5573 if (q < 0)
5574 return q;
5575
5576 *ret = TAKE_PIDREF(pidref);
5577 return r;
5578 }
5579
5580 /* Child */
5581
5582 (void) default_signals(SIGNALS_CRASH_HANDLER, SIGNALS_IGNORE);
5583 (void) ignore_signals(SIGPIPE);
5584
5585 if (crt && crt->cgroup_path) {
5586 r = cg_attach(crt->cgroup_path, 0);
5587 if (r < 0) {
5588 log_unit_error_errno(u, r, "Failed to join unit cgroup %s: %m", empty_to_root(crt->cgroup_path));
5589 _exit(EXIT_CGROUP);
5590 }
5591 }
5592
5593 return 0;
5594 }
5595
5596 int unit_fork_and_watch_rm_rf(Unit *u, char **paths, PidRef *ret_pid) {
5597 _cleanup_(pidref_done) PidRef pid = PIDREF_NULL;
5598 int r;
5599
5600 assert(u);
5601 assert(ret_pid);
5602
5603 r = unit_fork_helper_process(u, "(sd-rmrf)", /* into_cgroup= */ true, &pid);
5604 if (r < 0)
5605 return r;
5606 if (r == 0) {
5607 int ret = EXIT_SUCCESS;
5608
5609 STRV_FOREACH(i, paths) {
5610 r = rm_rf(*i, REMOVE_ROOT|REMOVE_PHYSICAL|REMOVE_MISSING_OK);
5611 if (r < 0) {
5612 log_error_errno(r, "Failed to remove '%s': %m", *i);
5613 ret = EXIT_FAILURE;
5614 }
5615 }
5616
5617 _exit(ret);
5618 }
5619
5620 r = unit_watch_pidref(u, &pid, /* exclusive= */ true);
5621 if (r < 0)
5622 return r;
5623
5624 *ret_pid = TAKE_PIDREF(pid);
5625 return 0;
5626 }
5627
5628 static void unit_update_dependency_mask(Hashmap *deps, Unit *other, UnitDependencyInfo di) {
5629 assert(deps);
5630 assert(other);
5631
5632 if (di.origin_mask == 0 && di.destination_mask == 0)
5633 /* No bit set anymore, let's drop the whole entry */
5634 assert_se(hashmap_remove(deps, other));
5635 else
5636 /* Mask was reduced, let's update the entry */
5637 assert_se(hashmap_update(deps, other, di.data) == 0);
5638 }
5639
5640 void unit_remove_dependencies(Unit *u, UnitDependencyMask mask) {
5641 Hashmap *deps;
5642 assert(u);
5643
5644 /* Removes all dependencies u has on other units marked for ownership by 'mask'. */
5645
5646 if (mask == 0)
5647 return;
5648
5649 HASHMAP_FOREACH(deps, u->dependencies) {
5650 bool done;
5651
5652 do {
5653 UnitDependencyInfo di;
5654 Unit *other;
5655
5656 done = true;
5657
5658 HASHMAP_FOREACH_KEY(di.data, other, deps) {
5659 Hashmap *other_deps;
5660
5661 if (FLAGS_SET(~mask, di.origin_mask))
5662 continue;
5663
5664 di.origin_mask &= ~mask;
5665 unit_update_dependency_mask(deps, other, di);
5666
5667 /* We updated the dependency from our unit to the other unit now. But most
5668 * dependencies imply a reverse dependency. Hence, let's delete that one
5669 * too. For that we go through all dependency types on the other unit and
5670 * delete all those which point to us and have the right mask set. */
5671
5672 HASHMAP_FOREACH(other_deps, other->dependencies) {
5673 UnitDependencyInfo dj;
5674
5675 dj.data = hashmap_get(other_deps, u);
5676 if (FLAGS_SET(~mask, dj.destination_mask))
5677 continue;
5678
5679 dj.destination_mask &= ~mask;
5680 unit_update_dependency_mask(other_deps, u, dj);
5681 }
5682
5683 unit_add_to_gc_queue(other);
5684
5685 /* The unit 'other' may not be wanted by the unit 'u'. */
5686 unit_submit_to_stop_when_unneeded_queue(other);
5687
5688 u->dependency_generation++;
5689 other->dependency_generation++;
5690
5691 done = false;
5692 break;
5693 }
5694
5695 } while (!done);
5696 }
5697 }
5698
5699 static int unit_get_invocation_path(Unit *u, char **ret) {
5700 char *p;
5701 int r;
5702
5703 assert(u);
5704 assert(ret);
5705
5706 if (MANAGER_IS_SYSTEM(u->manager))
5707 p = strjoin("/run/systemd/units/invocation:", u->id);
5708 else {
5709 _cleanup_free_ char *user_path = NULL;
5710
5711 r = xdg_user_runtime_dir("/systemd/units/invocation:", &user_path);
5712 if (r < 0)
5713 return r;
5714
5715 p = strjoin(user_path, u->id);
5716 }
5717 if (!p)
5718 return -ENOMEM;
5719
5720 *ret = p;
5721 return 0;
5722 }
5723
5724 static int unit_export_invocation_id(Unit *u) {
5725 _cleanup_free_ char *p = NULL;
5726 int r;
5727
5728 assert(u);
5729
5730 if (u->exported_invocation_id)
5731 return 0;
5732
5733 if (sd_id128_is_null(u->invocation_id))
5734 return 0;
5735
5736 r = unit_get_invocation_path(u, &p);
5737 if (r < 0)
5738 return log_unit_debug_errno(u, r, "Failed to get invocation path: %m");
5739
5740 r = symlinkat_atomic_full(u->invocation_id_string, AT_FDCWD, p, SYMLINK_LABEL);
5741 if (r < 0)
5742 return log_unit_debug_errno(u, r, "Failed to create invocation ID symlink %s: %m", p);
5743
5744 u->exported_invocation_id = true;
5745 return 0;
5746 }
5747
5748 static int unit_export_log_level_max(Unit *u, int log_level_max, bool overwrite) {
5749 const char *p;
5750 char buf[2];
5751 int r;
5752
5753 assert(u);
5754
5755 /* When the debug_invocation logic runs, overwrite will be true as we always want to switch the max
5756 * log level that the journal applies, and we want to always restore the previous level once done */
5757
5758 if (!overwrite && u->exported_log_level_max)
5759 return 0;
5760
5761 if (log_level_max < 0)
5762 return 0;
5763
5764 assert(log_level_max <= 7);
5765
5766 buf[0] = '0' + log_level_max;
5767 buf[1] = 0;
5768
5769 p = strjoina("/run/systemd/units/log-level-max:", u->id);
5770 r = symlink_atomic(buf, p);
5771 if (r < 0)
5772 return log_unit_debug_errno(u, r, "Failed to create maximum log level symlink %s: %m", p);
5773
5774 u->exported_log_level_max = true;
5775 return 0;
5776 }
5777
5778 static int unit_export_log_extra_fields(Unit *u, const ExecContext *c) {
5779 _cleanup_close_ int fd = -EBADF;
5780 struct iovec *iovec;
5781 const char *p;
5782 char *pattern;
5783 le64_t *sizes;
5784 ssize_t n;
5785 int r;
5786
5787 if (u->exported_log_extra_fields)
5788 return 0;
5789
5790 if (c->n_log_extra_fields <= 0)
5791 return 0;
5792
5793 sizes = newa(le64_t, c->n_log_extra_fields);
5794 iovec = newa(struct iovec, c->n_log_extra_fields * 2);
5795
5796 for (size_t i = 0; i < c->n_log_extra_fields; i++) {
5797 sizes[i] = htole64(c->log_extra_fields[i].iov_len);
5798
5799 iovec[i*2] = IOVEC_MAKE(sizes + i, sizeof(le64_t));
5800 iovec[i*2+1] = c->log_extra_fields[i];
5801 }
5802
5803 p = strjoina("/run/systemd/units/log-extra-fields:", u->id);
5804 pattern = strjoina(p, ".XXXXXX");
5805
5806 fd = mkostemp_safe(pattern);
5807 if (fd < 0)
5808 return log_unit_debug_errno(u, fd, "Failed to create extra fields file %s: %m", p);
5809
5810 n = writev(fd, iovec, c->n_log_extra_fields*2);
5811 if (n < 0) {
5812 r = log_unit_debug_errno(u, errno, "Failed to write extra fields: %m");
5813 goto fail;
5814 }
5815
5816 (void) fchmod(fd, 0644);
5817
5818 if (rename(pattern, p) < 0) {
5819 r = log_unit_debug_errno(u, errno, "Failed to rename extra fields file: %m");
5820 goto fail;
5821 }
5822
5823 u->exported_log_extra_fields = true;
5824 return 0;
5825
5826 fail:
5827 (void) unlink(pattern);
5828 return r;
5829 }
5830
5831 static int unit_export_log_ratelimit_interval(Unit *u, const ExecContext *c) {
5832 _cleanup_free_ char *buf = NULL;
5833 const char *p;
5834 int r;
5835
5836 assert(u);
5837 assert(c);
5838
5839 if (u->exported_log_ratelimit_interval)
5840 return 0;
5841
5842 if (c->log_ratelimit.interval == 0)
5843 return 0;
5844
5845 p = strjoina("/run/systemd/units/log-rate-limit-interval:", u->id);
5846
5847 if (asprintf(&buf, "%" PRIu64, c->log_ratelimit.interval) < 0)
5848 return log_oom();
5849
5850 r = symlink_atomic(buf, p);
5851 if (r < 0)
5852 return log_unit_debug_errno(u, r, "Failed to create log rate limit interval symlink %s: %m", p);
5853
5854 u->exported_log_ratelimit_interval = true;
5855 return 0;
5856 }
5857
5858 static int unit_export_log_ratelimit_burst(Unit *u, const ExecContext *c) {
5859 _cleanup_free_ char *buf = NULL;
5860 const char *p;
5861 int r;
5862
5863 assert(u);
5864 assert(c);
5865
5866 if (u->exported_log_ratelimit_burst)
5867 return 0;
5868
5869 if (c->log_ratelimit.burst == 0)
5870 return 0;
5871
5872 p = strjoina("/run/systemd/units/log-rate-limit-burst:", u->id);
5873
5874 if (asprintf(&buf, "%u", c->log_ratelimit.burst) < 0)
5875 return log_oom();
5876
5877 r = symlink_atomic(buf, p);
5878 if (r < 0)
5879 return log_unit_debug_errno(u, r, "Failed to create log rate limit burst symlink %s: %m", p);
5880
5881 u->exported_log_ratelimit_burst = true;
5882 return 0;
5883 }
5884
5885 void unit_export_state_files(Unit *u) {
5886 const ExecContext *c;
5887
5888 assert(u);
5889
5890 if (!u->id)
5891 return;
5892
5893 if (MANAGER_IS_TEST_RUN(u->manager))
5894 return;
5895
5896 /* Exports a couple of unit properties to /run/systemd/units/, so that journald can quickly query this data
5897 * from there. Ideally, journald would use IPC to query this, like everybody else, but that's hard, as long as
5898 * the IPC system itself and PID 1 also log to the journal.
5899 *
5900 * Note that these files really shouldn't be considered API for anyone else, as use a runtime file system as
5901 * IPC replacement is not compatible with today's world of file system namespaces. However, this doesn't really
5902 * apply to communication between the journal and systemd, as we assume that these two daemons live in the same
5903 * namespace at least.
5904 *
5905 * Note that some of the "files" exported here are actually symlinks and not regular files. Symlinks work
5906 * better for storing small bits of data, in particular as we can write them with two system calls, and read
5907 * them with one. */
5908
5909 (void) unit_export_invocation_id(u);
5910
5911 if (!MANAGER_IS_SYSTEM(u->manager))
5912 return;
5913
5914 c = unit_get_exec_context(u);
5915 if (c) {
5916 (void) unit_export_log_level_max(u, c->log_level_max, /* overwrite= */ false);
5917 (void) unit_export_log_extra_fields(u, c);
5918 (void) unit_export_log_ratelimit_interval(u, c);
5919 (void) unit_export_log_ratelimit_burst(u, c);
5920 }
5921 }
5922
5923 void unit_unlink_state_files(Unit *u) {
5924 const char *p;
5925
5926 assert(u);
5927
5928 if (!u->id)
5929 return;
5930
5931 /* Undoes the effect of unit_export_state() */
5932
5933 if (u->exported_invocation_id) {
5934 _cleanup_free_ char *invocation_path = NULL;
5935 int r = unit_get_invocation_path(u, &invocation_path);
5936 if (r >= 0) {
5937 (void) unlink(invocation_path);
5938 u->exported_invocation_id = false;
5939 }
5940 }
5941
5942 if (!MANAGER_IS_SYSTEM(u->manager))
5943 return;
5944
5945 if (u->exported_log_level_max) {
5946 p = strjoina("/run/systemd/units/log-level-max:", u->id);
5947 (void) unlink(p);
5948
5949 u->exported_log_level_max = false;
5950 }
5951
5952 if (u->exported_log_extra_fields) {
5953 p = strjoina("/run/systemd/units/extra-fields:", u->id);
5954 (void) unlink(p);
5955
5956 u->exported_log_extra_fields = false;
5957 }
5958
5959 if (u->exported_log_ratelimit_interval) {
5960 p = strjoina("/run/systemd/units/log-rate-limit-interval:", u->id);
5961 (void) unlink(p);
5962
5963 u->exported_log_ratelimit_interval = false;
5964 }
5965
5966 if (u->exported_log_ratelimit_burst) {
5967 p = strjoina("/run/systemd/units/log-rate-limit-burst:", u->id);
5968 (void) unlink(p);
5969
5970 u->exported_log_ratelimit_burst = false;
5971 }
5972 }
5973
5974 int unit_set_debug_invocation(Unit *u, bool enable) {
5975 int r;
5976
5977 assert(u);
5978
5979 if (u->debug_invocation == enable)
5980 return 0; /* Nothing to do */
5981
5982 u->debug_invocation = enable;
5983
5984 /* Ensure that the new log level is exported for the journal, in place of the previous one */
5985 if (u->exported_log_level_max) {
5986 const ExecContext *ec = unit_get_exec_context(u);
5987 if (ec) {
5988 r = unit_export_log_level_max(u, enable ? LOG_PRI(LOG_DEBUG) : ec->log_level_max, /* overwrite= */ true);
5989 if (r < 0)
5990 return r;
5991 }
5992 }
5993
5994 return 1;
5995 }
5996
5997 int unit_prepare_exec(Unit *u) {
5998 int r;
5999
6000 assert(u);
6001
6002 /* Load any custom firewall BPF programs here once to test if they are existing and actually loadable.
6003 * Fail here early since later errors in the call chain unit_realize_cgroup to cgroup_context_apply are ignored. */
6004 r = bpf_firewall_load_custom(u);
6005 if (r < 0)
6006 return r;
6007
6008 /* Prepares everything so that we can fork of a process for this unit */
6009
6010 (void) unit_realize_cgroup(u);
6011
6012 CGroupRuntime *crt = unit_get_cgroup_runtime(u);
6013 if (crt && crt->reset_accounting) {
6014 (void) unit_reset_accounting(u);
6015 crt->reset_accounting = false;
6016 }
6017
6018 unit_export_state_files(u);
6019
6020 r = unit_setup_exec_runtime(u);
6021 if (r < 0)
6022 return r;
6023
6024 return 0;
6025 }
6026
6027 static int unit_log_leftover_process_start(const PidRef *pid, int sig, void *userdata) {
6028 const Unit *u = ASSERT_PTR(userdata);
6029 _cleanup_free_ char *comm = NULL;
6030
6031 assert(pidref_is_set(pid));
6032
6033 (void) pidref_get_comm(pid, &comm);
6034
6035 if (ignore_leftover_process(comm))
6036 return 0;
6037
6038 /* During start we print a warning */
6039
6040 log_unit_warning(u,
6041 "Found left-over process " PID_FMT " (%s) in control group while starting unit. Ignoring.\n"
6042 "This usually indicates unclean termination of a previous run, or service implementation deficiencies.",
6043 pid->pid, strna(comm));
6044
6045 return 1;
6046 }
6047
6048 static int unit_log_leftover_process_stop(const PidRef *pid, int sig, void *userdata) {
6049 const Unit *u = ASSERT_PTR(userdata);
6050 _cleanup_free_ char *comm = NULL;
6051
6052 assert(pidref_is_set(pid));
6053
6054 (void) pidref_get_comm(pid, &comm);
6055
6056 if (ignore_leftover_process(comm))
6057 return 0;
6058
6059 /* During stop we only print an informational message */
6060
6061 log_unit_info(u,
6062 "Unit process " PID_FMT " (%s) remains running after unit stopped.",
6063 pid->pid, strna(comm));
6064
6065 return 1;
6066 }
6067
6068 int unit_warn_leftover_processes(Unit *u, bool start) {
6069 _cleanup_free_ char *cgroup = NULL;
6070 int r;
6071
6072 assert(u);
6073
6074 r = unit_get_cgroup_path_with_fallback(u, &cgroup);
6075 if (r < 0)
6076 return r;
6077
6078 return cg_kill_recursive(
6079 cgroup,
6080 /* sig= */ 0,
6081 /* flags= */ 0,
6082 /* killed_pids= */ NULL,
6083 start ? unit_log_leftover_process_start : unit_log_leftover_process_stop,
6084 u);
6085 }
6086
6087 bool unit_needs_console(Unit *u) {
6088 ExecContext *ec;
6089 UnitActiveState state;
6090
6091 assert(u);
6092
6093 state = unit_active_state(u);
6094
6095 if (UNIT_IS_INACTIVE_OR_FAILED(state))
6096 return false;
6097
6098 if (UNIT_VTABLE(u)->needs_console)
6099 return UNIT_VTABLE(u)->needs_console(u);
6100
6101 /* If this unit type doesn't implement this call, let's use a generic fallback implementation: */
6102 ec = unit_get_exec_context(u);
6103 if (!ec)
6104 return false;
6105
6106 return exec_context_may_touch_console(ec);
6107 }
6108
6109 int unit_pid_attachable(Unit *u, PidRef *pid, sd_bus_error *error) {
6110 int r;
6111
6112 assert(u);
6113
6114 /* Checks whether the specified PID is generally good for attaching, i.e. a valid PID, not our manager itself,
6115 * and not a kernel thread either */
6116
6117 /* First, a simple range check */
6118 if (!pidref_is_set(pid))
6119 return sd_bus_error_set(error, SD_BUS_ERROR_INVALID_ARGS, "Process identifier is not valid.");
6120
6121 /* Some extra safety check */
6122 if (pid->pid == 1 || pidref_is_self(pid))
6123 return sd_bus_error_setf(error, SD_BUS_ERROR_INVALID_ARGS, "Process " PID_FMT " is a manager process, refusing.", pid->pid);
6124
6125 /* Don't even begin to bother with kernel threads */
6126 r = pidref_is_kernel_thread(pid);
6127 if (r == -ESRCH)
6128 return sd_bus_error_setf(error, SD_BUS_ERROR_UNIX_PROCESS_ID_UNKNOWN, "Process with ID " PID_FMT " does not exist.", pid->pid);
6129 if (r < 0)
6130 return sd_bus_error_set_errnof(error, r, "Failed to determine whether process " PID_FMT " is a kernel thread: %m", pid->pid);
6131 if (r > 0)
6132 return sd_bus_error_setf(error, SD_BUS_ERROR_INVALID_ARGS, "Process " PID_FMT " is a kernel thread, refusing.", pid->pid);
6133
6134 return 0;
6135 }
6136
6137 int unit_get_log_level_max(const Unit *u) {
6138 if (u) {
6139 if (u->debug_invocation)
6140 return LOG_DEBUG;
6141
6142 ExecContext *ec = unit_get_exec_context(u);
6143 if (ec && ec->log_level_max >= 0)
6144 return ec->log_level_max;
6145 }
6146
6147 return log_get_max_level();
6148 }
6149
6150 bool unit_log_level_test(const Unit *u, int level) {
6151 assert(u);
6152 return LOG_PRI(level) <= unit_get_log_level_max(u);
6153 }
6154
6155 void unit_log_success(Unit *u) {
6156 assert(u);
6157
6158 /* Let's show message "Deactivated successfully" in debug mode (when manager is user) rather than in info mode.
6159 * This message has low information value for regular users and it might be a bit overwhelming on a system with
6160 * a lot of devices. */
6161 log_unit_struct(u,
6162 MANAGER_IS_USER(u->manager) ? LOG_DEBUG : LOG_INFO,
6163 LOG_MESSAGE_ID(SD_MESSAGE_UNIT_SUCCESS_STR),
6164 LOG_UNIT_INVOCATION_ID(u),
6165 LOG_UNIT_MESSAGE(u, "Deactivated successfully."));
6166 }
6167
6168 void unit_log_failure(Unit *u, const char *result) {
6169 assert(u);
6170 assert(result);
6171
6172 log_unit_struct(u, LOG_WARNING,
6173 LOG_MESSAGE_ID(SD_MESSAGE_UNIT_FAILURE_RESULT_STR),
6174 LOG_UNIT_INVOCATION_ID(u),
6175 LOG_UNIT_MESSAGE(u, "Failed with result '%s'.", result),
6176 LOG_ITEM("UNIT_RESULT=%s", result));
6177 }
6178
6179 void unit_log_skip(Unit *u, const char *result) {
6180 assert(u);
6181 assert(result);
6182
6183 log_unit_struct(u, LOG_INFO,
6184 LOG_MESSAGE_ID(SD_MESSAGE_UNIT_SKIPPED_STR),
6185 LOG_UNIT_INVOCATION_ID(u),
6186 LOG_UNIT_MESSAGE(u, "Skipped due to '%s'.", result),
6187 LOG_ITEM("UNIT_RESULT=%s", result));
6188 }
6189
6190 void unit_log_process_exit(
6191 Unit *u,
6192 const char *kind,
6193 const char *command,
6194 bool success,
6195 int code,
6196 int status) {
6197
6198 int level;
6199
6200 assert(u);
6201 assert(kind);
6202
6203 /* If this is a successful exit, let's log about the exit code on DEBUG level. If this is a failure
6204 * and the process exited on its own via exit(), then let's make this a NOTICE, under the assumption
6205 * that the service already logged the reason at a higher log level on its own. Otherwise, make it a
6206 * WARNING. */
6207 if (success)
6208 level = LOG_DEBUG;
6209 else if (code == CLD_EXITED)
6210 level = LOG_NOTICE;
6211 else
6212 level = LOG_WARNING;
6213
6214 log_unit_struct(u, level,
6215 LOG_MESSAGE_ID(SD_MESSAGE_UNIT_PROCESS_EXIT_STR),
6216 LOG_UNIT_MESSAGE(u, "%s exited, code=%s, status=%i/%s%s",
6217 kind,
6218 sigchld_code_to_string(code), status,
6219 strna(code == CLD_EXITED
6220 ? exit_status_to_string(status, EXIT_STATUS_FULL)
6221 : signal_to_string(status)),
6222 success ? " (success)" : ""),
6223 LOG_ITEM("EXIT_CODE=%s", sigchld_code_to_string(code)),
6224 LOG_ITEM("EXIT_STATUS=%i", status),
6225 LOG_ITEM("COMMAND=%s", strna(command)),
6226 LOG_UNIT_INVOCATION_ID(u));
6227 }
6228
6229 int unit_exit_status(Unit *u) {
6230 assert(u);
6231
6232 /* Returns the exit status to propagate for the most recent cycle of this unit. Returns a value in the range
6233 * 0…255 if there's something to propagate. EOPNOTSUPP if the concept does not apply to this unit type, ENODATA
6234 * if no data is currently known (for example because the unit hasn't deactivated yet) and EBADE if the main
6235 * service process has exited abnormally (signal/coredump). */
6236
6237 if (!UNIT_VTABLE(u)->exit_status)
6238 return -EOPNOTSUPP;
6239
6240 return UNIT_VTABLE(u)->exit_status(u);
6241 }
6242
6243 int unit_failure_action_exit_status(Unit *u) {
6244 int r;
6245
6246 assert(u);
6247
6248 /* Returns the exit status to propagate on failure, or an error if there's nothing to propagate */
6249
6250 if (u->failure_action_exit_status >= 0)
6251 return u->failure_action_exit_status;
6252
6253 r = unit_exit_status(u);
6254 if (r == -EBADE) /* Exited, but not cleanly (i.e. by signal or such) */
6255 return 255;
6256
6257 return r;
6258 }
6259
6260 int unit_success_action_exit_status(Unit *u) {
6261 int r;
6262
6263 assert(u);
6264
6265 /* Returns the exit status to propagate on success, or an error if there's nothing to propagate */
6266
6267 if (u->success_action_exit_status >= 0)
6268 return u->success_action_exit_status;
6269
6270 r = unit_exit_status(u);
6271 if (r == -EBADE) /* Exited, but not cleanly (i.e. by signal or such) */
6272 return 255;
6273
6274 return r;
6275 }
6276
6277 int unit_test_trigger_loaded(Unit *u) {
6278 Unit *trigger;
6279
6280 /* Tests whether the unit to trigger is loaded */
6281
6282 trigger = UNIT_TRIGGER(u);
6283 if (!trigger)
6284 return log_unit_error_errno(u, SYNTHETIC_ERRNO(ENOENT),
6285 "Refusing to start, no unit to trigger.");
6286 if (trigger->load_state != UNIT_LOADED)
6287 return log_unit_error_errno(u, SYNTHETIC_ERRNO(ENOENT),
6288 "Refusing to start, unit %s to trigger not loaded.", trigger->id);
6289
6290 return 0;
6291 }
6292
6293 void unit_destroy_runtime_data(Unit *u, const ExecContext *context, bool destroy_runtime_dir) {
6294 assert(u);
6295 assert(u->manager);
6296 assert(context);
6297
6298 /* EXEC_PRESERVE_RESTART is handled via unit_release_resources()! */
6299 if (destroy_runtime_dir && context->runtime_directory_preserve_mode == EXEC_PRESERVE_NO)
6300 exec_context_destroy_runtime_directory(context, u->manager->prefix[EXEC_DIRECTORY_RUNTIME]);
6301
6302 exec_context_destroy_credentials(context, u->manager->prefix[EXEC_DIRECTORY_RUNTIME], u->id);
6303 exec_context_destroy_mount_ns_dir(u);
6304 }
6305
6306 int unit_clean(Unit *u, ExecCleanMask mask) {
6307 UnitActiveState state;
6308
6309 assert(u);
6310
6311 /* Special return values:
6312 *
6313 * -EOPNOTSUPP → cleaning not supported for this unit type
6314 * -EUNATCH → cleaning not defined for this resource type
6315 * -EBUSY → unit currently can't be cleaned since it's running or not properly loaded, or has
6316 * a job queued or similar
6317 */
6318
6319 if (!UNIT_VTABLE(u)->clean)
6320 return -EOPNOTSUPP;
6321
6322 if (mask == 0)
6323 return -EUNATCH;
6324
6325 if (u->load_state != UNIT_LOADED)
6326 return -EBUSY;
6327
6328 if (u->job)
6329 return -EBUSY;
6330
6331 state = unit_active_state(u);
6332 if (state != UNIT_INACTIVE)
6333 return -EBUSY;
6334
6335 return UNIT_VTABLE(u)->clean(u, mask);
6336 }
6337
6338 int unit_can_clean(Unit *u, ExecCleanMask *ret) {
6339 assert(u);
6340
6341 if (!UNIT_VTABLE(u)->clean ||
6342 u->load_state != UNIT_LOADED) {
6343 *ret = 0;
6344 return 0;
6345 }
6346
6347 /* When the clean() method is set, can_clean() really should be set too */
6348 assert(UNIT_VTABLE(u)->can_clean);
6349
6350 return UNIT_VTABLE(u)->can_clean(u, ret);
6351 }
6352
6353 bool unit_can_start_refuse_manual(Unit *u) {
6354 return unit_can_start(u) && !u->refuse_manual_start;
6355 }
6356
6357 bool unit_can_stop_refuse_manual(Unit *u) {
6358 return unit_can_stop(u) && !u->refuse_manual_stop;
6359 }
6360
6361 bool unit_can_isolate_refuse_manual(Unit *u) {
6362 return unit_can_isolate(u) && !u->refuse_manual_start;
6363 }
6364
6365 void unit_next_freezer_state(Unit *u, FreezerAction action, FreezerState *ret_next, FreezerState *ret_objective) {
6366 FreezerState current, parent, next, objective;
6367
6368 assert(u);
6369 assert(action >= 0);
6370 assert(action < _FREEZER_ACTION_MAX);
6371 assert(ret_next);
6372 assert(ret_objective);
6373
6374 /* This function determines the correct freezer state transitions for a unit
6375 * given the action being requested. It returns the next state, and also the "objective",
6376 * which is either FREEZER_FROZEN or FREEZER_RUNNING, depending on what actual state we
6377 * ultimately want to achieve. */
6378
6379 current = u->freezer_state;
6380
6381 Unit *slice = UNIT_GET_SLICE(u);
6382 if (slice)
6383 parent = slice->freezer_state;
6384 else
6385 parent = FREEZER_RUNNING;
6386
6387 switch (action) {
6388
6389 case FREEZER_FREEZE:
6390 /* We always "promote" a freeze initiated by parent into a normal freeze */
6391 if (IN_SET(current, FREEZER_FROZEN, FREEZER_FROZEN_BY_PARENT))
6392 next = FREEZER_FROZEN;
6393 else
6394 next = FREEZER_FREEZING;
6395 break;
6396
6397 case FREEZER_THAW:
6398 /* Thawing is the most complicated operation here, because we can't thaw a unit
6399 * if its parent is frozen. So we instead "demote" a normal freeze into a freeze
6400 * initiated by parent if the parent is frozen */
6401 if (IN_SET(current, FREEZER_RUNNING, FREEZER_THAWING,
6402 FREEZER_FREEZING_BY_PARENT, FREEZER_FROZEN_BY_PARENT)) /* Should usually be refused by unit_freezer_action */
6403 next = current;
6404 else if (current == FREEZER_FREEZING) {
6405 if (IN_SET(parent, FREEZER_RUNNING, FREEZER_THAWING))
6406 next = FREEZER_THAWING;
6407 else
6408 next = FREEZER_FREEZING_BY_PARENT;
6409 } else if (current == FREEZER_FROZEN) {
6410 if (IN_SET(parent, FREEZER_RUNNING, FREEZER_THAWING))
6411 next = FREEZER_THAWING;
6412 else
6413 next = FREEZER_FROZEN_BY_PARENT;
6414 } else
6415 assert_not_reached();
6416 break;
6417
6418 case FREEZER_PARENT_FREEZE:
6419 /* We need to avoid accidentally demoting units frozen manually */
6420 if (IN_SET(current, FREEZER_FREEZING, FREEZER_FROZEN, FREEZER_FROZEN_BY_PARENT))
6421 next = current;
6422 else
6423 next = FREEZER_FREEZING_BY_PARENT;
6424 break;
6425
6426 case FREEZER_PARENT_THAW:
6427 /* We don't want to thaw units from a parent if they were frozen
6428 * manually, so for such units this action is a no-op */
6429 if (IN_SET(current, FREEZER_RUNNING, FREEZER_FREEZING, FREEZER_FROZEN))
6430 next = current;
6431 else
6432 next = FREEZER_THAWING;
6433 break;
6434
6435 default:
6436 assert_not_reached();
6437 }
6438
6439 objective = freezer_state_finish(next);
6440 if (objective == FREEZER_FROZEN_BY_PARENT)
6441 objective = FREEZER_FROZEN;
6442 assert(IN_SET(objective, FREEZER_RUNNING, FREEZER_FROZEN));
6443
6444 *ret_next = next;
6445 *ret_objective = objective;
6446 }
6447
6448 bool unit_can_freeze(const Unit *u) {
6449 assert(u);
6450
6451 if (unit_has_name(u, SPECIAL_ROOT_SLICE) || unit_has_name(u, SPECIAL_INIT_SCOPE))
6452 return false;
6453
6454 if (UNIT_VTABLE(u)->can_freeze)
6455 return UNIT_VTABLE(u)->can_freeze(u);
6456
6457 return UNIT_VTABLE(u)->freezer_action;
6458 }
6459
6460 void unit_set_freezer_state(Unit *u, FreezerState state) {
6461 assert(u);
6462 assert(state >= 0);
6463 assert(state < _FREEZER_STATE_MAX);
6464
6465 if (u->freezer_state == state)
6466 return;
6467
6468 log_unit_debug(u, "Freezer state changed %s -> %s",
6469 freezer_state_to_string(u->freezer_state), freezer_state_to_string(state));
6470
6471 u->freezer_state = state;
6472
6473 unit_add_to_dbus_queue(u);
6474 }
6475
6476 void unit_freezer_complete(Unit *u, FreezerState kernel_state) {
6477 bool expected;
6478
6479 assert(u);
6480 assert(IN_SET(kernel_state, FREEZER_RUNNING, FREEZER_FROZEN));
6481
6482 expected = IN_SET(u->freezer_state, FREEZER_RUNNING, FREEZER_THAWING) == (kernel_state == FREEZER_RUNNING);
6483
6484 unit_set_freezer_state(u, expected ? freezer_state_finish(u->freezer_state) : kernel_state);
6485 log_unit_info(u, "Unit now %s.", u->freezer_state == FREEZER_RUNNING ? "thawed" :
6486 freezer_state_to_string(u->freezer_state));
6487
6488 /* If the cgroup's final state is against what's requested by us, report as canceled. */
6489 bus_unit_send_pending_freezer_message(u, /* canceled = */ !expected);
6490 }
6491
6492 int unit_freezer_action(Unit *u, FreezerAction action) {
6493 UnitActiveState s;
6494 int r;
6495
6496 assert(u);
6497 assert(IN_SET(action, FREEZER_FREEZE, FREEZER_THAW));
6498
6499 if (!unit_can_freeze(u))
6500 return -EOPNOTSUPP;
6501
6502 if (u->job)
6503 return -EBUSY;
6504
6505 if (u->load_state != UNIT_LOADED)
6506 return -EHOSTDOWN;
6507
6508 s = unit_active_state(u);
6509 if (s != UNIT_ACTIVE)
6510 return -EHOSTDOWN;
6511
6512 if (action == FREEZER_FREEZE && IN_SET(u->freezer_state, FREEZER_FREEZING, FREEZER_FREEZING_BY_PARENT))
6513 return -EALREADY;
6514 if (action == FREEZER_THAW && u->freezer_state == FREEZER_THAWING)
6515 return -EALREADY;
6516 if (action == FREEZER_THAW && IN_SET(u->freezer_state, FREEZER_FREEZING_BY_PARENT, FREEZER_FROZEN_BY_PARENT))
6517 return -EDEADLK;
6518
6519 r = UNIT_VTABLE(u)->freezer_action(u, action);
6520 if (r <= 0)
6521 return r;
6522
6523 assert(IN_SET(u->freezer_state, FREEZER_FREEZING, FREEZER_FREEZING_BY_PARENT, FREEZER_THAWING));
6524 return 1;
6525 }
6526
6527 Condition *unit_find_failed_condition(Unit *u) {
6528 Condition *failed_trigger = NULL;
6529 bool has_succeeded_trigger = false;
6530
6531 if (u->condition_result)
6532 return NULL;
6533
6534 LIST_FOREACH(conditions, c, u->conditions)
6535 if (c->trigger) {
6536 if (c->result == CONDITION_SUCCEEDED)
6537 has_succeeded_trigger = true;
6538 else if (!failed_trigger)
6539 failed_trigger = c;
6540 } else if (c->result != CONDITION_SUCCEEDED)
6541 return c;
6542
6543 return failed_trigger && !has_succeeded_trigger ? failed_trigger : NULL;
6544 }
6545
6546 int unit_can_live_mount(Unit *u, sd_bus_error *error) {
6547 assert(u);
6548
6549 if (!UNIT_VTABLE(u)->live_mount)
6550 return sd_bus_error_setf(
6551 error,
6552 SD_BUS_ERROR_NOT_SUPPORTED,
6553 "Live mounting not supported by unit type '%s'",
6554 unit_type_to_string(u->type));
6555
6556 if (u->load_state != UNIT_LOADED)
6557 return sd_bus_error_setf(
6558 error,
6559 BUS_ERROR_NO_SUCH_UNIT,
6560 "Unit '%s' not loaded, cannot live mount",
6561 u->id);
6562
6563 if (!UNIT_VTABLE(u)->can_live_mount)
6564 return 0;
6565
6566 return UNIT_VTABLE(u)->can_live_mount(u, error);
6567 }
6568
6569 int unit_live_mount(
6570 Unit *u,
6571 const char *src,
6572 const char *dst,
6573 sd_bus_message *message,
6574 MountInNamespaceFlags flags,
6575 const MountOptions *options,
6576 sd_bus_error *error) {
6577
6578 assert(u);
6579 assert(UNIT_VTABLE(u)->live_mount);
6580
6581 if (!UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u))) {
6582 log_unit_debug(u, "Unit not active, cannot perform live mount.");
6583 return sd_bus_error_setf(
6584 error,
6585 BUS_ERROR_UNIT_INACTIVE,
6586 "Live mounting '%s' on '%s' for unit '%s' cannot be scheduled: unit not active",
6587 src,
6588 dst,
6589 u->id);
6590 }
6591
6592 if (unit_active_state(u) == UNIT_REFRESHING) {
6593 log_unit_debug(u, "Unit already live mounting, refusing further requests.");
6594 return sd_bus_error_setf(
6595 error,
6596 BUS_ERROR_UNIT_BUSY,
6597 "Live mounting '%s' on '%s' for unit '%s' cannot be scheduled: another live mount in progress",
6598 src,
6599 dst,
6600 u->id);
6601 }
6602
6603 if (u->job) {
6604 log_unit_debug(u, "Unit already has a job in progress, cannot live mount");
6605 return sd_bus_error_setf(
6606 error,
6607 BUS_ERROR_UNIT_BUSY,
6608 "Live mounting '%s' on '%s' for unit '%s' cannot be scheduled: another operation in progress",
6609 src,
6610 dst,
6611 u->id);
6612 }
6613
6614 return UNIT_VTABLE(u)->live_mount(u, src, dst, message, flags, options, error);
6615 }
6616
6617 static const char* const collect_mode_table[_COLLECT_MODE_MAX] = {
6618 [COLLECT_INACTIVE] = "inactive",
6619 [COLLECT_INACTIVE_OR_FAILED] = "inactive-or-failed",
6620 };
6621
6622 DEFINE_STRING_TABLE_LOOKUP(collect_mode, CollectMode);
6623
6624 Unit* unit_has_dependency(const Unit *u, UnitDependencyAtom atom, Unit *other) {
6625 Unit *i;
6626
6627 assert(u);
6628
6629 /* Checks if the unit has a dependency on 'other' with the specified dependency atom. If 'other' is
6630 * NULL checks if the unit has *any* dependency of that atom. Returns 'other' if found (or if 'other'
6631 * is NULL the first entry found), or NULL if not found. */
6632
6633 UNIT_FOREACH_DEPENDENCY(i, u, atom)
6634 if (!other || other == i)
6635 return i;
6636
6637 return NULL;
6638 }
6639
6640 int unit_get_dependency_array(const Unit *u, UnitDependencyAtom atom, Unit ***ret_array) {
6641 _cleanup_free_ Unit **array = NULL;
6642 size_t n = 0;
6643 Unit *other;
6644
6645 assert(u);
6646 assert(ret_array);
6647
6648 /* Gets a list of units matching a specific atom as array. This is useful when iterating through
6649 * dependencies while modifying them: the array is an "atomic snapshot" of sorts, that can be read
6650 * while the dependency table is continuously updated. */
6651
6652 UNIT_FOREACH_DEPENDENCY(other, u, atom) {
6653 if (!GREEDY_REALLOC(array, n + 1))
6654 return -ENOMEM;
6655
6656 array[n++] = other;
6657 }
6658
6659 *ret_array = TAKE_PTR(array);
6660
6661 assert(n <= INT_MAX);
6662 return (int) n;
6663 }
6664
6665 int unit_get_transitive_dependency_set(Unit *u, UnitDependencyAtom atom, Set **ret) {
6666 _cleanup_set_free_ Set *units = NULL, *queue = NULL;
6667 Unit *other;
6668 int r;
6669
6670 assert(u);
6671 assert(ret);
6672
6673 /* Similar to unit_get_dependency_array(), but also search the same dependency in other units. */
6674
6675 do {
6676 UNIT_FOREACH_DEPENDENCY(other, u, atom) {
6677 r = set_ensure_put(&units, NULL, other);
6678 if (r < 0)
6679 return r;
6680 if (r == 0)
6681 continue;
6682 r = set_ensure_put(&queue, NULL, other);
6683 if (r < 0)
6684 return r;
6685 }
6686 } while ((u = set_steal_first(queue)));
6687
6688 *ret = TAKE_PTR(units);
6689 return 0;
6690 }
6691
6692 int unit_arm_timer(
6693 Unit *u,
6694 sd_event_source **source,
6695 bool relative,
6696 usec_t usec,
6697 sd_event_time_handler_t handler) {
6698
6699 int r;
6700
6701 assert(u);
6702 assert(source);
6703 assert(handler);
6704
6705 if (*source) {
6706 if (usec == USEC_INFINITY)
6707 return sd_event_source_set_enabled(*source, SD_EVENT_OFF);
6708
6709 r = (relative ? sd_event_source_set_time_relative : sd_event_source_set_time)(*source, usec);
6710 if (r < 0)
6711 return r;
6712
6713 return sd_event_source_set_enabled(*source, SD_EVENT_ONESHOT);
6714 }
6715
6716 if (usec == USEC_INFINITY)
6717 return 0;
6718
6719 r = (relative ? sd_event_add_time_relative : sd_event_add_time)(
6720 u->manager->event,
6721 source,
6722 CLOCK_MONOTONIC,
6723 usec, 0,
6724 handler,
6725 u);
6726 if (r < 0)
6727 return r;
6728
6729 const char *d = strjoina(unit_type_to_string(u->type), "-timer");
6730 (void) sd_event_source_set_description(*source, d);
6731
6732 return 0;
6733 }
6734
6735 bool unit_passes_filter(Unit *u, char * const *states, char * const *patterns) {
6736 assert(u);
6737
6738 if (!strv_isempty(states)) {
6739 char * const *unit_states = STRV_MAKE(
6740 unit_load_state_to_string(u->load_state),
6741 unit_active_state_to_string(unit_active_state(u)),
6742 unit_sub_state_to_string(u));
6743
6744 if (!strv_overlap(states, unit_states))
6745 return false;
6746 }
6747
6748 return strv_fnmatch_or_empty(patterns, u->id, FNM_NOESCAPE);
6749 }
6750
6751 static int unit_get_nice(Unit *u) {
6752 ExecContext *ec;
6753
6754 ec = unit_get_exec_context(u);
6755 return ec ? ec->nice : 0;
6756 }
6757
6758 static uint64_t unit_get_cpu_weight(Unit *u) {
6759 CGroupContext *cc;
6760
6761 cc = unit_get_cgroup_context(u);
6762 return cc ? cgroup_context_cpu_weight(cc, manager_state(u->manager)) : CGROUP_WEIGHT_DEFAULT;
6763 }
6764
6765 int unit_get_exec_quota_stats(Unit *u, ExecContext *c, ExecDirectoryType dt, uint64_t *ret_usage, uint64_t *ret_limit) {
6766 int r;
6767 _cleanup_close_ int fd = -EBADF;
6768 _cleanup_free_ char *p = NULL, *pp = NULL;
6769
6770 assert(u);
6771 assert(c);
6772
6773 if (c->directories[dt].n_items == 0) {
6774 *ret_usage = UINT64_MAX;
6775 *ret_limit = UINT64_MAX;
6776 return 0;
6777 }
6778
6779 ExecDirectoryItem *i = &c->directories[dt].items[0];
6780 p = path_join(u->manager->prefix[dt], i->path);
6781 if (!p)
6782 return log_oom_debug();
6783
6784 if (exec_directory_is_private(c, dt)) {
6785 pp = path_join(u->manager->prefix[dt], "private", i->path);
6786 if (!pp)
6787 return log_oom_debug();
6788 }
6789
6790 const char *target_dir = pp ?: p;
6791 fd = open(target_dir, O_PATH | O_CLOEXEC | O_DIRECTORY);
6792 if (fd < 0)
6793 return log_unit_debug_errno(u, errno, "Failed to get exec quota stats: %m");
6794
6795 uint32_t proj_id;
6796 r = read_fs_xattr_fd(fd, /* ret_xflags = */ NULL, &proj_id);
6797 if (r < 0)
6798 return log_unit_debug_errno(u, r, "Failed to get project ID for exec quota stats: %m");
6799
6800 struct dqblk req;
6801 r = quota_query_proj_id(fd, proj_id, &req);
6802 if (r <= 0)
6803 return log_unit_debug_errno(u, r, "Failed to query project ID for exec quota stats: %m");
6804
6805 *ret_usage = req.dqb_curspace;
6806 *ret_limit = req.dqb_bhardlimit * QIF_DQBLKSIZE;
6807
6808 return r;
6809 }
6810
6811 int unit_compare_priority(Unit *a, Unit *b) {
6812 int ret;
6813
6814 ret = CMP(a->type, b->type);
6815 if (ret != 0)
6816 return -ret;
6817
6818 ret = CMP(unit_get_cpu_weight(a), unit_get_cpu_weight(b));
6819 if (ret != 0)
6820 return -ret;
6821
6822 ret = CMP(unit_get_nice(a), unit_get_nice(b));
6823 if (ret != 0)
6824 return ret;
6825
6826 return strcmp(a->id, b->id);
6827 }
6828
6829 const char* unit_log_field(const Unit *u) {
6830 return MANAGER_IS_SYSTEM(ASSERT_PTR(u)->manager) ? "UNIT=" : "USER_UNIT=";
6831 }
6832
6833 const char* unit_invocation_log_field(const Unit *u) {
6834 return MANAGER_IS_SYSTEM(ASSERT_PTR(u)->manager) ? "INVOCATION_ID=" : "USER_INVOCATION_ID=";
6835 }
6836
6837 const ActivationDetailsVTable * const activation_details_vtable[_UNIT_TYPE_MAX] = {
6838 [UNIT_PATH] = &activation_details_path_vtable,
6839 [UNIT_TIMER] = &activation_details_timer_vtable,
6840 };
6841
6842 ActivationDetails *activation_details_new(Unit *trigger_unit) {
6843 _cleanup_free_ ActivationDetails *details = NULL;
6844
6845 assert(trigger_unit);
6846 assert(trigger_unit->type != _UNIT_TYPE_INVALID);
6847 assert(trigger_unit->id);
6848
6849 details = malloc0(activation_details_vtable[trigger_unit->type]->object_size);
6850 if (!details)
6851 return NULL;
6852
6853 *details = (ActivationDetails) {
6854 .n_ref = 1,
6855 .trigger_unit_type = trigger_unit->type,
6856 };
6857
6858 details->trigger_unit_name = strdup(trigger_unit->id);
6859 if (!details->trigger_unit_name)
6860 return NULL;
6861
6862 if (ACTIVATION_DETAILS_VTABLE(details)->init)
6863 ACTIVATION_DETAILS_VTABLE(details)->init(details, trigger_unit);
6864
6865 return TAKE_PTR(details);
6866 }
6867
6868 static ActivationDetails *activation_details_free(ActivationDetails *details) {
6869 if (!details)
6870 return NULL;
6871
6872 if (ACTIVATION_DETAILS_VTABLE(details)->done)
6873 ACTIVATION_DETAILS_VTABLE(details)->done(details);
6874
6875 free(details->trigger_unit_name);
6876
6877 return mfree(details);
6878 }
6879
6880 void activation_details_serialize(const ActivationDetails *details, FILE *f) {
6881 if (!details || details->trigger_unit_type == _UNIT_TYPE_INVALID)
6882 return;
6883
6884 (void) serialize_item(f, "activation-details-unit-type", unit_type_to_string(details->trigger_unit_type));
6885 if (details->trigger_unit_name)
6886 (void) serialize_item(f, "activation-details-unit-name", details->trigger_unit_name);
6887 if (ACTIVATION_DETAILS_VTABLE(details)->serialize)
6888 ACTIVATION_DETAILS_VTABLE(details)->serialize(details, f);
6889 }
6890
6891 int activation_details_deserialize(const char *key, const char *value, ActivationDetails **details) {
6892 int r;
6893
6894 assert(key);
6895 assert(value);
6896 assert(details);
6897
6898 if (!*details) {
6899 UnitType t;
6900
6901 if (!streq(key, "activation-details-unit-type"))
6902 return -EINVAL;
6903
6904 t = unit_type_from_string(value);
6905 if (t < 0)
6906 return t;
6907
6908 /* The activation details vtable has defined ops only for path and timer units */
6909 if (!activation_details_vtable[t])
6910 return -EINVAL;
6911
6912 *details = malloc0(activation_details_vtable[t]->object_size);
6913 if (!*details)
6914 return -ENOMEM;
6915
6916 **details = (ActivationDetails) {
6917 .n_ref = 1,
6918 .trigger_unit_type = t,
6919 };
6920
6921 return 0;
6922 }
6923
6924 if (streq(key, "activation-details-unit-name")) {
6925 r = free_and_strdup(&(*details)->trigger_unit_name, value);
6926 if (r < 0)
6927 return r;
6928
6929 return 0;
6930 }
6931
6932 if (ACTIVATION_DETAILS_VTABLE(*details)->deserialize)
6933 return ACTIVATION_DETAILS_VTABLE(*details)->deserialize(key, value, details);
6934
6935 return -EINVAL;
6936 }
6937
6938 int activation_details_append_env(const ActivationDetails *details, char ***strv) {
6939 int r = 0;
6940
6941 assert(strv);
6942
6943 if (!details)
6944 return 0;
6945
6946 if (!isempty(details->trigger_unit_name)) {
6947 char *s = strjoin("TRIGGER_UNIT=", details->trigger_unit_name);
6948 if (!s)
6949 return -ENOMEM;
6950
6951 r = strv_consume(strv, TAKE_PTR(s));
6952 if (r < 0)
6953 return r;
6954 }
6955
6956 if (ACTIVATION_DETAILS_VTABLE(details)->append_env) {
6957 r = ACTIVATION_DETAILS_VTABLE(details)->append_env(details, strv);
6958 if (r < 0)
6959 return r;
6960 }
6961
6962 return r + !isempty(details->trigger_unit_name); /* Return the number of variables added to the env block */
6963 }
6964
6965 int activation_details_append_pair(const ActivationDetails *details, char ***strv) {
6966 int r = 0;
6967
6968 assert(strv);
6969
6970 if (!details)
6971 return 0;
6972
6973 if (!isempty(details->trigger_unit_name)) {
6974 r = strv_extend_many(strv, "trigger_unit", details->trigger_unit_name);
6975 if (r < 0)
6976 return r;
6977 }
6978
6979 if (ACTIVATION_DETAILS_VTABLE(details)->append_pair) {
6980 r = ACTIVATION_DETAILS_VTABLE(details)->append_pair(details, strv);
6981 if (r < 0)
6982 return r;
6983 }
6984
6985 return r + !isempty(details->trigger_unit_name); /* Return the number of pairs added to the strv */
6986 }
6987
6988 DEFINE_TRIVIAL_REF_UNREF_FUNC(ActivationDetails, activation_details, activation_details_free);
6989
6990 static const char* const unit_mount_dependency_type_table[_UNIT_MOUNT_DEPENDENCY_TYPE_MAX] = {
6991 [UNIT_MOUNT_WANTS] = "WantsMountsFor",
6992 [UNIT_MOUNT_REQUIRES] = "RequiresMountsFor",
6993 };
6994
6995 DEFINE_STRING_TABLE_LOOKUP(unit_mount_dependency_type, UnitMountDependencyType);
6996
6997 static const char* const oom_policy_table[_OOM_POLICY_MAX] = {
6998 [OOM_CONTINUE] = "continue",
6999 [OOM_STOP] = "stop",
7000 [OOM_KILL] = "kill",
7001 };
7002
7003 DEFINE_STRING_TABLE_LOOKUP(oom_policy, OOMPolicy);
7004
7005 UnitDependency unit_mount_dependency_type_to_dependency_type(UnitMountDependencyType t) {
7006 switch (t) {
7007
7008 case UNIT_MOUNT_WANTS:
7009 return UNIT_WANTS;
7010
7011 case UNIT_MOUNT_REQUIRES:
7012 return UNIT_REQUIRES;
7013
7014 default:
7015 assert_not_reached();
7016 }
7017 }