]> git.ipfire.org Git - thirdparty/systemd.git/blob - src/core/unit.c
a85987c74bfc915b96d3cba40c642ad655213d83
[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 "load-dropin.h"
39 #include "load-fragment.h"
40 #include "log.h"
41 #include "logarithm.h"
42 #include "mkdir-label.h"
43 #include "manager.h"
44 #include "mount-util.h"
45 #include "mountpoint-util.h"
46 #include "netlink-internal.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(
1829 Unit *u,
1830 StatusType status_type,
1831 const char *status,
1832 const char *format,
1833 const char *ident) {
1834
1835 assert(u);
1836
1837 if (log_get_show_color()) {
1838 if (u->manager->status_unit_format == STATUS_UNIT_FORMAT_COMBINED && strchr(ident, ' '))
1839 ident = strjoina(ANSI_HIGHLIGHT, u->id, ANSI_NORMAL, " - ", u->description);
1840 else
1841 ident = strjoina(ANSI_HIGHLIGHT, ident, ANSI_NORMAL);
1842 }
1843
1844 DISABLE_WARNING_FORMAT_NONLITERAL;
1845 manager_status_printf(u->manager, status_type, status, format, ident);
1846 REENABLE_WARNING;
1847 }
1848
1849 int unit_test_start_limit(Unit *u) {
1850 const char *reason;
1851
1852 assert(u);
1853
1854 if (ratelimit_below(&u->start_ratelimit)) {
1855 u->start_limit_hit = false;
1856 return 0;
1857 }
1858
1859 log_unit_warning(u, "Start request repeated too quickly.");
1860 u->start_limit_hit = true;
1861
1862 reason = strjoina("unit ", u->id, " failed");
1863
1864 emergency_action(
1865 u->manager,
1866 u->start_limit_action,
1867 EMERGENCY_ACTION_IS_WATCHDOG|EMERGENCY_ACTION_WARN|EMERGENCY_ACTION_SLEEP_5S,
1868 u->reboot_arg,
1869 /* exit_status= */ -1,
1870 reason);
1871
1872 return -ECANCELED;
1873 }
1874
1875 static bool unit_verify_deps(Unit *u) {
1876 Unit *other;
1877
1878 assert(u);
1879
1880 /* Checks whether all BindsTo= dependencies of this unit are fulfilled — if they are also combined
1881 * with After=. We do not check Requires= or Requisite= here as they only should have an effect on
1882 * the job processing, but do not have any effect afterwards. We don't check BindsTo= dependencies
1883 * that are not used in conjunction with After= as for them any such check would make things entirely
1884 * racy. */
1885
1886 UNIT_FOREACH_DEPENDENCY(other, u, UNIT_ATOM_CANNOT_BE_ACTIVE_WITHOUT) {
1887
1888 if (!unit_has_dependency(u, UNIT_ATOM_AFTER, other))
1889 continue;
1890
1891 if (!UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(other))) {
1892 log_unit_notice(u, "Bound to unit %s, but unit isn't active.", other->id);
1893 return false;
1894 }
1895 }
1896
1897 return true;
1898 }
1899
1900 /* Errors that aren't really errors:
1901 * -EALREADY: Unit is already started.
1902 * -ECOMM: Condition failed
1903 * -EAGAIN: An operation is already in progress. Retry later.
1904 *
1905 * Errors that are real errors:
1906 * -EBADR: This unit type does not support starting.
1907 * -ECANCELED: Start limit hit, too many requests for now
1908 * -EPROTO: Assert failed
1909 * -EINVAL: Unit not loaded
1910 * -EOPNOTSUPP: Unit type not supported
1911 * -ENOLINK: The necessary dependencies are not fulfilled.
1912 * -ESTALE: This unit has been started before and can't be started a second time
1913 * -EDEADLK: This unit is frozen
1914 * -ENOENT: This is a triggering unit and unit to trigger is not loaded
1915 * -ETOOMANYREFS: The hard concurrency limit of at least one of the slices the unit is contained in has been reached
1916 */
1917 int unit_start(Unit *u, ActivationDetails *details) {
1918 UnitActiveState state;
1919 Unit *following;
1920 int r;
1921
1922 assert(u);
1923
1924 /* Let's hold off running start jobs for mount units when /proc/self/mountinfo monitor is ratelimited. */
1925 if (UNIT_VTABLE(u)->subsystem_ratelimited) {
1926 r = UNIT_VTABLE(u)->subsystem_ratelimited(u->manager);
1927 if (r < 0)
1928 return r;
1929 if (r > 0)
1930 return -EAGAIN;
1931 }
1932
1933 /* If this is already started, then this will succeed. Note that this will even succeed if this unit
1934 * is not startable by the user. This is relied on to detect when we need to wait for units and when
1935 * waiting is finished. */
1936 state = unit_active_state(u);
1937 if (UNIT_IS_ACTIVE_OR_RELOADING(state))
1938 return -EALREADY;
1939 if (IN_SET(state, UNIT_DEACTIVATING, UNIT_MAINTENANCE))
1940 return -EAGAIN;
1941
1942 /* Units that aren't loaded cannot be started */
1943 if (u->load_state != UNIT_LOADED)
1944 return -EINVAL;
1945
1946 /* Refuse starting scope units more than once */
1947 if (UNIT_VTABLE(u)->once_only && dual_timestamp_is_set(&u->inactive_enter_timestamp))
1948 return -ESTALE;
1949
1950 /* If the conditions were unmet, don't do anything at all. If we already are activating this call might
1951 * still be useful to speed up activation in case there is some hold-off time, but we don't want to
1952 * recheck the condition in that case. */
1953 if (state != UNIT_ACTIVATING &&
1954 !unit_test_condition(u))
1955 return log_unit_debug_errno(u, SYNTHETIC_ERRNO(ECOMM), "Starting requested but condition not met. Not starting unit.");
1956
1957 /* If the asserts failed, fail the entire job */
1958 if (state != UNIT_ACTIVATING &&
1959 !unit_test_assert(u))
1960 return log_unit_notice_errno(u, SYNTHETIC_ERRNO(EPROTO), "Starting requested but asserts failed.");
1961
1962 /* Units of types that aren't supported cannot be started. Note that we do this test only after the
1963 * condition checks, so that we rather return condition check errors (which are usually not
1964 * considered a true failure) than "not supported" errors (which are considered a failure).
1965 */
1966 if (!unit_type_supported(u->type))
1967 return -EOPNOTSUPP;
1968
1969 /* Let's make sure that the deps really are in order before we start this. Normally the job engine
1970 * should have taken care of this already, but let's check this here again. After all, our
1971 * dependencies might not be in effect anymore, due to a reload or due to an unmet condition. */
1972 if (!unit_verify_deps(u))
1973 return -ENOLINK;
1974
1975 /* Forward to the main object, if we aren't it. */
1976 following = unit_following(u);
1977 if (following) {
1978 log_unit_debug(u, "Redirecting start request from %s to %s.", u->id, following->id);
1979 return unit_start(following, details);
1980 }
1981
1982 /* Check to make sure the unit isn't frozen */
1983 if (u->freezer_state != FREEZER_RUNNING)
1984 return -EDEADLK;
1985
1986 /* Check our ability to start early so that ratelimited or already starting/started units don't
1987 * cause us to enter a busy loop. */
1988 if (UNIT_VTABLE(u)->test_startable) {
1989 r = UNIT_VTABLE(u)->test_startable(u);
1990 if (r <= 0)
1991 return r;
1992 }
1993
1994 /* If it is stopped, but we cannot start it, then fail */
1995 if (!UNIT_VTABLE(u)->start)
1996 return -EBADR;
1997
1998 if (UNIT_IS_INACTIVE_OR_FAILED(state)) {
1999 Slice *slice = SLICE(UNIT_GET_SLICE(u));
2000
2001 if (slice) {
2002 /* Check hard concurrency limit. Note this is partially redundant, we already checked
2003 * this when enqueuing jobs. However, between the time when we enqueued this and the
2004 * time we are dispatching the queue the configuration might have changed, hence
2005 * check here again */
2006 if (slice_concurrency_hard_max_reached(slice, u))
2007 return -ETOOMANYREFS;
2008
2009 /* Also check soft concurrenty limit, and return EAGAIN so that the job is kept in
2010 * the queue */
2011 if (slice_concurrency_soft_max_reached(slice, u))
2012 return -EAGAIN; /* Try again, keep in queue */
2013 }
2014 }
2015
2016 /* We don't suppress calls to ->start() here when we are already starting, to allow this request to
2017 * be used as a "hurry up" call, for example when the unit is in some "auto restart" state where it
2018 * waits for a holdoff timer to elapse before it will start again. */
2019
2020 unit_add_to_dbus_queue(u);
2021
2022 if (!u->activation_details) /* Older details object wins */
2023 u->activation_details = activation_details_ref(details);
2024
2025 return UNIT_VTABLE(u)->start(u);
2026 }
2027
2028 bool unit_can_start(Unit *u) {
2029 assert(u);
2030
2031 if (u->load_state != UNIT_LOADED)
2032 return false;
2033
2034 if (!unit_type_supported(u->type))
2035 return false;
2036
2037 /* Scope units may be started only once */
2038 if (UNIT_VTABLE(u)->once_only && dual_timestamp_is_set(&u->inactive_exit_timestamp))
2039 return false;
2040
2041 return !!UNIT_VTABLE(u)->start;
2042 }
2043
2044 bool unit_can_isolate(Unit *u) {
2045 assert(u);
2046
2047 return unit_can_start(u) &&
2048 u->allow_isolate;
2049 }
2050
2051 /* Errors:
2052 * -EBADR: This unit type does not support stopping.
2053 * -EALREADY: Unit is already stopped.
2054 * -EAGAIN: An operation is already in progress. Retry later.
2055 * -EDEADLK: Unit is frozen
2056 */
2057 int unit_stop(Unit *u) {
2058 UnitActiveState state;
2059 Unit *following;
2060
2061 assert(u);
2062
2063 state = unit_active_state(u);
2064 if (UNIT_IS_INACTIVE_OR_FAILED(state))
2065 return -EALREADY;
2066
2067 following = unit_following(u);
2068 if (following) {
2069 log_unit_debug(u, "Redirecting stop request from %s to %s.", u->id, following->id);
2070 return unit_stop(following);
2071 }
2072
2073 /* Check to make sure the unit isn't frozen */
2074 if (u->freezer_state != FREEZER_RUNNING)
2075 return -EDEADLK;
2076
2077 if (!UNIT_VTABLE(u)->stop)
2078 return -EBADR;
2079
2080 unit_add_to_dbus_queue(u);
2081
2082 return UNIT_VTABLE(u)->stop(u);
2083 }
2084
2085 bool unit_can_stop(Unit *u) {
2086 assert(u);
2087
2088 /* Note: if we return true here, it does not mean that the unit may be successfully stopped.
2089 * Extrinsic units follow external state and they may stop following external state changes
2090 * (hence we return true here), but an attempt to do this through the manager will fail. */
2091
2092 if (!unit_type_supported(u->type))
2093 return false;
2094
2095 if (u->perpetual)
2096 return false;
2097
2098 return !!UNIT_VTABLE(u)->stop;
2099 }
2100
2101 /* Errors:
2102 * -EBADR: This unit type does not support reloading.
2103 * -ENOEXEC: Unit is not started.
2104 * -EAGAIN: An operation is already in progress. Retry later.
2105 * -EDEADLK: Unit is frozen.
2106 */
2107 int unit_reload(Unit *u) {
2108 UnitActiveState state;
2109 Unit *following;
2110
2111 assert(u);
2112
2113 if (u->load_state != UNIT_LOADED)
2114 return -EINVAL;
2115
2116 if (!unit_can_reload(u))
2117 return -EBADR;
2118
2119 state = unit_active_state(u);
2120 if (IN_SET(state, UNIT_RELOADING, UNIT_REFRESHING))
2121 /* "refreshing" means some resources in the unit namespace is being updated. Unlike reload,
2122 * the unit processes aren't made aware of refresh. Let's put the job back to queue
2123 * in both cases, as refresh typically takes place before reload and it's better to wait
2124 * for it rather than failing. */
2125 return -EAGAIN;
2126
2127 if (state != UNIT_ACTIVE)
2128 return log_unit_warning_errno(u, SYNTHETIC_ERRNO(ENOEXEC), "Unit cannot be reloaded because it is inactive.");
2129
2130 following = unit_following(u);
2131 if (following) {
2132 log_unit_debug(u, "Redirecting reload request from %s to %s.", u->id, following->id);
2133 return unit_reload(following);
2134 }
2135
2136 /* Check to make sure the unit isn't frozen */
2137 if (u->freezer_state != FREEZER_RUNNING)
2138 return -EDEADLK;
2139
2140 unit_add_to_dbus_queue(u);
2141
2142 if (!UNIT_VTABLE(u)->reload) {
2143 /* Unit doesn't have a reload function, but we need to propagate the reload anyway */
2144 unit_notify(u, unit_active_state(u), unit_active_state(u), /* reload_success = */ true);
2145 return 0;
2146 }
2147
2148 return UNIT_VTABLE(u)->reload(u);
2149 }
2150
2151 bool unit_can_reload(Unit *u) {
2152 assert(u);
2153
2154 if (UNIT_VTABLE(u)->can_reload)
2155 return UNIT_VTABLE(u)->can_reload(u);
2156
2157 if (unit_has_dependency(u, UNIT_ATOM_PROPAGATES_RELOAD_TO, NULL))
2158 return true;
2159
2160 return UNIT_VTABLE(u)->reload;
2161 }
2162
2163 bool unit_is_unneeded(Unit *u) {
2164 Unit *other;
2165 assert(u);
2166
2167 if (!u->stop_when_unneeded)
2168 return false;
2169
2170 /* Don't clean up while the unit is transitioning or is even inactive. */
2171 if (unit_active_state(u) != UNIT_ACTIVE)
2172 return false;
2173 if (u->job)
2174 return false;
2175
2176 UNIT_FOREACH_DEPENDENCY(other, u, UNIT_ATOM_PINS_STOP_WHEN_UNNEEDED) {
2177 /* If a dependent unit has a job queued, is active or transitioning, or is marked for
2178 * restart, then don't clean this one up. */
2179
2180 if (other->job)
2181 return false;
2182
2183 if (!UNIT_IS_INACTIVE_OR_FAILED(unit_active_state(other)))
2184 return false;
2185
2186 if (unit_will_restart(other))
2187 return false;
2188 }
2189
2190 return true;
2191 }
2192
2193 bool unit_is_upheld_by_active(Unit *u, Unit **ret_culprit) {
2194 Unit *other;
2195
2196 assert(u);
2197
2198 /* Checks if the unit needs to be started because it currently is not running, but some other unit
2199 * that is active declared an Uphold= dependencies on it */
2200
2201 if (!UNIT_IS_INACTIVE_OR_FAILED(unit_active_state(u)) || u->job) {
2202 if (ret_culprit)
2203 *ret_culprit = NULL;
2204 return false;
2205 }
2206
2207 UNIT_FOREACH_DEPENDENCY(other, u, UNIT_ATOM_START_STEADILY) {
2208 if (other->job)
2209 continue;
2210
2211 if (UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(other))) {
2212 if (ret_culprit)
2213 *ret_culprit = other;
2214 return true;
2215 }
2216 }
2217
2218 if (ret_culprit)
2219 *ret_culprit = NULL;
2220 return false;
2221 }
2222
2223 bool unit_is_bound_by_inactive(Unit *u, Unit **ret_culprit) {
2224 Unit *other;
2225
2226 assert(u);
2227
2228 /* Checks whether this unit is bound to another unit that is inactive, i.e. whether we should stop
2229 * because the other unit is down. */
2230
2231 if (unit_active_state(u) != UNIT_ACTIVE || u->job) {
2232 /* Don't clean up while the unit is transitioning or is even inactive. */
2233 if (ret_culprit)
2234 *ret_culprit = NULL;
2235 return false;
2236 }
2237
2238 UNIT_FOREACH_DEPENDENCY(other, u, UNIT_ATOM_CANNOT_BE_ACTIVE_WITHOUT) {
2239 if (other->job)
2240 continue;
2241
2242 if (UNIT_IS_INACTIVE_OR_FAILED(unit_active_state(other))) {
2243 if (ret_culprit)
2244 *ret_culprit = other;
2245
2246 return true;
2247 }
2248 }
2249
2250 if (ret_culprit)
2251 *ret_culprit = NULL;
2252 return false;
2253 }
2254
2255 static void check_unneeded_dependencies(Unit *u) {
2256 Unit *other;
2257 assert(u);
2258
2259 /* Add all units this unit depends on to the queue that processes StopWhenUnneeded= behaviour. */
2260
2261 UNIT_FOREACH_DEPENDENCY(other, u, UNIT_ATOM_ADD_STOP_WHEN_UNNEEDED_QUEUE)
2262 unit_submit_to_stop_when_unneeded_queue(other);
2263 }
2264
2265 static void check_uphold_dependencies(Unit *u) {
2266 Unit *other;
2267 assert(u);
2268
2269 /* Add all units this unit depends on to the queue that processes Uphold= behaviour. */
2270
2271 UNIT_FOREACH_DEPENDENCY(other, u, UNIT_ATOM_ADD_START_WHEN_UPHELD_QUEUE)
2272 unit_submit_to_start_when_upheld_queue(other);
2273 }
2274
2275 static void check_bound_by_dependencies(Unit *u) {
2276 Unit *other;
2277 assert(u);
2278
2279 /* Add all units this unit depends on to the queue that processes BindsTo= stop behaviour. */
2280
2281 UNIT_FOREACH_DEPENDENCY(other, u, UNIT_ATOM_ADD_CANNOT_BE_ACTIVE_WITHOUT_QUEUE)
2282 unit_submit_to_stop_when_bound_queue(other);
2283 }
2284
2285 static void retroactively_start_dependencies(Unit *u) {
2286 Unit *other;
2287
2288 assert(u);
2289 assert(UNIT_IS_ACTIVE_OR_ACTIVATING(unit_active_state(u)));
2290
2291 UNIT_FOREACH_DEPENDENCY_SAFE(other, u, UNIT_ATOM_RETROACTIVE_START_REPLACE) /* Requires= + BindsTo= */
2292 if (!unit_has_dependency(u, UNIT_ATOM_AFTER, other) &&
2293 !UNIT_IS_ACTIVE_OR_ACTIVATING(unit_active_state(other)))
2294 (void) manager_add_job(u->manager, JOB_START, other, JOB_REPLACE, /* error = */ NULL, /* ret = */ NULL);
2295
2296 UNIT_FOREACH_DEPENDENCY_SAFE(other, u, UNIT_ATOM_RETROACTIVE_START_FAIL) /* Wants= */
2297 if (!unit_has_dependency(u, UNIT_ATOM_AFTER, other) &&
2298 !UNIT_IS_ACTIVE_OR_ACTIVATING(unit_active_state(other)))
2299 (void) manager_add_job(u->manager, JOB_START, other, JOB_FAIL, /* error = */ NULL, /* ret = */ NULL);
2300
2301 UNIT_FOREACH_DEPENDENCY_SAFE(other, u, UNIT_ATOM_RETROACTIVE_STOP_ON_START) /* Conflicts= (and inverse) */
2302 if (!UNIT_IS_INACTIVE_OR_DEACTIVATING(unit_active_state(other)))
2303 (void) manager_add_job(u->manager, JOB_STOP, other, JOB_REPLACE, /* error = */ NULL, /* ret = */ NULL);
2304 }
2305
2306 static void retroactively_stop_dependencies(Unit *u) {
2307 Unit *other;
2308
2309 assert(u);
2310 assert(UNIT_IS_INACTIVE_OR_DEACTIVATING(unit_active_state(u)));
2311
2312 /* Pull down units which are bound to us recursively if enabled */
2313 UNIT_FOREACH_DEPENDENCY_SAFE(other, u, UNIT_ATOM_RETROACTIVE_STOP_ON_STOP) /* BoundBy= */
2314 if (!UNIT_IS_INACTIVE_OR_DEACTIVATING(unit_active_state(other)))
2315 (void) manager_add_job(u->manager, JOB_STOP, other, JOB_REPLACE, /* error = */ NULL, /* ret = */ NULL);
2316 }
2317
2318 void unit_start_on_termination_deps(Unit *u, UnitDependencyAtom atom) {
2319 const char *dependency_name = NULL;
2320 JobMode job_mode;
2321 unsigned n_jobs = 0;
2322 int r;
2323
2324 /* Act on OnFailure= and OnSuccess= dependencies */
2325
2326 assert(u);
2327 assert(u->manager);
2328 assert(IN_SET(atom, UNIT_ATOM_ON_SUCCESS, UNIT_ATOM_ON_FAILURE));
2329
2330 FOREACH_ELEMENT(setting, on_termination_settings)
2331 if (atom == setting->atom) {
2332 job_mode = *(JobMode*) ((uint8_t*) u + setting->job_mode_offset);
2333 dependency_name = setting->dependency_name;
2334 break;
2335 }
2336
2337 assert(dependency_name);
2338
2339 Unit *other;
2340 UNIT_FOREACH_DEPENDENCY_SAFE(other, u, atom) {
2341 _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL;
2342
2343 if (n_jobs == 0)
2344 log_unit_info(u, "Triggering %s dependencies.", dependency_name);
2345
2346 r = manager_add_job(u->manager, JOB_START, other, job_mode, &error, /* ret = */ NULL);
2347 if (r < 0)
2348 log_unit_warning_errno(u, r, "Failed to enqueue %s%s job, ignoring: %s",
2349 dependency_name, other->id, bus_error_message(&error, r));
2350 n_jobs++;
2351 }
2352
2353 if (n_jobs > 0)
2354 log_unit_debug(u, "Triggering %s dependencies done (%u %s).",
2355 dependency_name, n_jobs, n_jobs == 1 ? "job" : "jobs");
2356 }
2357
2358 void unit_trigger_notify(Unit *u) {
2359 Unit *other;
2360
2361 assert(u);
2362
2363 UNIT_FOREACH_DEPENDENCY_SAFE(other, u, UNIT_ATOM_TRIGGERED_BY)
2364 if (UNIT_VTABLE(other)->trigger_notify)
2365 UNIT_VTABLE(other)->trigger_notify(other, u);
2366 }
2367
2368 static int raise_level(int log_level, bool condition_info, bool condition_notice) {
2369 if (condition_notice && log_level > LOG_NOTICE)
2370 return LOG_NOTICE;
2371 if (condition_info && log_level > LOG_INFO)
2372 return LOG_INFO;
2373 return log_level;
2374 }
2375
2376 static int unit_log_resources(Unit *u) {
2377
2378 static const struct {
2379 const char *journal_field;
2380 const char *message_suffix;
2381 } memory_fields[_CGROUP_MEMORY_ACCOUNTING_METRIC_CACHED_LAST + 1] = {
2382 [CGROUP_MEMORY_PEAK] = { "MEMORY_PEAK", "memory peak" },
2383 [CGROUP_MEMORY_SWAP_PEAK] = { "MEMORY_SWAP_PEAK", "memory swap peak" },
2384 }, ip_fields[_CGROUP_IP_ACCOUNTING_METRIC_MAX] = {
2385 [CGROUP_IP_INGRESS_BYTES] = { "IP_METRIC_INGRESS_BYTES", "incoming IP traffic" },
2386 [CGROUP_IP_EGRESS_BYTES] = { "IP_METRIC_EGRESS_BYTES", "outgoing IP traffic" },
2387 [CGROUP_IP_INGRESS_PACKETS] = { "IP_METRIC_INGRESS_PACKETS", NULL },
2388 [CGROUP_IP_EGRESS_PACKETS] = { "IP_METRIC_EGRESS_PACKETS", NULL },
2389 }, io_fields[_CGROUP_IO_ACCOUNTING_METRIC_MAX] = {
2390 [CGROUP_IO_READ_BYTES] = { "IO_METRIC_READ_BYTES", "read from disk" },
2391 [CGROUP_IO_WRITE_BYTES] = { "IO_METRIC_WRITE_BYTES", "written to disk" },
2392 [CGROUP_IO_READ_OPERATIONS] = { "IO_METRIC_READ_OPERATIONS", NULL },
2393 [CGROUP_IO_WRITE_OPERATIONS] = { "IO_METRIC_WRITE_OPERATIONS", NULL },
2394 };
2395
2396 struct iovec *iovec = NULL;
2397 size_t n_iovec = 0;
2398 _cleanup_free_ char *message = NULL, *t = NULL;
2399 nsec_t cpu_nsec = NSEC_INFINITY;
2400 int log_level = LOG_DEBUG; /* May be raised if resources consumed over a threshold */
2401
2402 assert(u);
2403
2404 CLEANUP_ARRAY(iovec, n_iovec, iovec_array_free);
2405
2406 iovec = new(struct iovec, 1 + (_CGROUP_MEMORY_ACCOUNTING_METRIC_CACHED_LAST + 1) +
2407 _CGROUP_IP_ACCOUNTING_METRIC_MAX + _CGROUP_IO_ACCOUNTING_METRIC_MAX + 4);
2408 if (!iovec)
2409 return log_oom();
2410
2411 /* Invoked whenever a unit enters failed or dead state. Logs information about consumed resources if resource
2412 * accounting was enabled for a unit. It does this in two ways: a friendly human-readable string with reduced
2413 * information and the complete data in structured fields. */
2414
2415 (void) unit_get_cpu_usage(u, &cpu_nsec);
2416 if (cpu_nsec != NSEC_INFINITY) {
2417 /* Format the CPU time for inclusion in the structured log message */
2418 if (asprintf(&t, "CPU_USAGE_NSEC=%" PRIu64, cpu_nsec) < 0)
2419 return log_oom();
2420 iovec[n_iovec++] = IOVEC_MAKE_STRING(TAKE_PTR(t));
2421
2422 /* Format the CPU time for inclusion in the human language message string */
2423 if (dual_timestamp_is_set(&u->inactive_exit_timestamp) &&
2424 dual_timestamp_is_set(&u->inactive_enter_timestamp)) {
2425 usec_t wall_clock_usec = usec_sub_unsigned(u->inactive_enter_timestamp.monotonic, u->inactive_exit_timestamp.monotonic);
2426 if (strextendf_with_separator(&message, ", ",
2427 "Consumed %s CPU time over %s wall clock time",
2428 FORMAT_TIMESPAN(cpu_nsec / NSEC_PER_USEC, USEC_PER_MSEC),
2429 FORMAT_TIMESPAN(wall_clock_usec, USEC_PER_MSEC)) < 0)
2430 return log_oom();
2431 } else {
2432 if (strextendf_with_separator(&message, ", ",
2433 "Consumed %s CPU time",
2434 FORMAT_TIMESPAN(cpu_nsec / NSEC_PER_USEC, USEC_PER_MSEC)) < 0)
2435 return log_oom();
2436 }
2437
2438 log_level = raise_level(log_level,
2439 cpu_nsec > MENTIONWORTHY_CPU_NSEC,
2440 cpu_nsec > NOTICEWORTHY_CPU_NSEC);
2441 }
2442
2443 for (CGroupMemoryAccountingMetric metric = 0; metric <= _CGROUP_MEMORY_ACCOUNTING_METRIC_CACHED_LAST; metric++) {
2444 uint64_t value = UINT64_MAX;
2445
2446 assert(memory_fields[metric].journal_field);
2447 assert(memory_fields[metric].message_suffix);
2448
2449 (void) unit_get_memory_accounting(u, metric, &value);
2450 if (value == UINT64_MAX)
2451 continue;
2452
2453 if (asprintf(&t, "%s=%" PRIu64, memory_fields[metric].journal_field, value) < 0)
2454 return log_oom();
2455 iovec[n_iovec++] = IOVEC_MAKE_STRING(TAKE_PTR(t));
2456
2457 /* If value is 0, we don't log it in the MESSAGE= field. */
2458 if (value == 0)
2459 continue;
2460
2461 if (strextendf_with_separator(&message, ", ", "%s %s",
2462 FORMAT_BYTES(value), memory_fields[metric].message_suffix) < 0)
2463 return log_oom();
2464
2465 log_level = raise_level(log_level,
2466 value > MENTIONWORTHY_MEMORY_BYTES,
2467 value > NOTICEWORTHY_MEMORY_BYTES);
2468 }
2469
2470 for (CGroupIOAccountingMetric k = 0; k < _CGROUP_IO_ACCOUNTING_METRIC_MAX; k++) {
2471 uint64_t value = UINT64_MAX;
2472
2473 assert(io_fields[k].journal_field);
2474
2475 (void) unit_get_io_accounting(u, k, &value);
2476 if (value == UINT64_MAX)
2477 continue;
2478
2479 /* Format IO accounting data for inclusion in the structured log message */
2480 if (asprintf(&t, "%s=%" PRIu64, io_fields[k].journal_field, value) < 0)
2481 return log_oom();
2482 iovec[n_iovec++] = IOVEC_MAKE_STRING(TAKE_PTR(t));
2483
2484 /* If value is 0, we don't log it in the MESSAGE= field. */
2485 if (value == 0)
2486 continue;
2487
2488 /* Format the IO accounting data for inclusion in the human language message string, but only
2489 * for the bytes counters (and not for the operations counters) */
2490 if (io_fields[k].message_suffix) {
2491 if (strextendf_with_separator(&message, ", ", "%s %s",
2492 FORMAT_BYTES(value), io_fields[k].message_suffix) < 0)
2493 return log_oom();
2494
2495 log_level = raise_level(log_level,
2496 value > MENTIONWORTHY_IO_BYTES,
2497 value > NOTICEWORTHY_IO_BYTES);
2498 }
2499 }
2500
2501 for (CGroupIPAccountingMetric m = 0; m < _CGROUP_IP_ACCOUNTING_METRIC_MAX; m++) {
2502 uint64_t value = UINT64_MAX;
2503
2504 assert(ip_fields[m].journal_field);
2505
2506 (void) unit_get_ip_accounting(u, m, &value);
2507 if (value == UINT64_MAX)
2508 continue;
2509
2510 /* Format IP accounting data for inclusion in the structured log message */
2511 if (asprintf(&t, "%s=%" PRIu64, ip_fields[m].journal_field, value) < 0)
2512 return log_oom();
2513 iovec[n_iovec++] = IOVEC_MAKE_STRING(TAKE_PTR(t));
2514
2515 /* If value is 0, we don't log it in the MESSAGE= field. */
2516 if (value == 0)
2517 continue;
2518
2519 /* Format the IP accounting data for inclusion in the human language message string, but only
2520 * for the bytes counters (and not for the packets counters) */
2521 if (ip_fields[m].message_suffix) {
2522 if (strextendf_with_separator(&message, ", ", "%s %s",
2523 FORMAT_BYTES(value), ip_fields[m].message_suffix) < 0)
2524 return log_oom();
2525
2526 log_level = raise_level(log_level,
2527 value > MENTIONWORTHY_IP_BYTES,
2528 value > NOTICEWORTHY_IP_BYTES);
2529 }
2530 }
2531
2532 /* This check is here because it is the earliest point following all possible log_level assignments.
2533 * (If log_level is assigned anywhere after this point, move this check.) */
2534 if (!unit_log_level_test(u, log_level))
2535 return 0;
2536
2537 /* Is there any accounting data available at all? */
2538 if (n_iovec == 0) {
2539 assert(!message);
2540 return 0;
2541 }
2542
2543 t = strjoin("MESSAGE=", u->id, ": ", message ?: "Completed", ".");
2544 if (!t)
2545 return log_oom();
2546 iovec[n_iovec++] = IOVEC_MAKE_STRING(TAKE_PTR(t));
2547
2548 if (!set_iovec_string_field(iovec, &n_iovec, "MESSAGE_ID=", SD_MESSAGE_UNIT_RESOURCES_STR))
2549 return log_oom();
2550
2551 if (!set_iovec_string_field(iovec, &n_iovec, unit_log_field(u), u->id))
2552 return log_oom();
2553
2554 if (!set_iovec_string_field(iovec, &n_iovec, unit_invocation_log_field(u), u->invocation_id_string))
2555 return log_oom();
2556
2557 log_unit_struct_iovec(u, log_level, iovec, n_iovec);
2558
2559 return 0;
2560 }
2561
2562 static void unit_update_on_console(Unit *u) {
2563 bool b;
2564
2565 assert(u);
2566
2567 b = unit_needs_console(u);
2568 if (u->on_console == b)
2569 return;
2570
2571 u->on_console = b;
2572 if (b)
2573 manager_ref_console(u->manager);
2574 else
2575 manager_unref_console(u->manager);
2576 }
2577
2578 static void unit_emit_audit_start(Unit *u) {
2579 assert(u);
2580
2581 if (UNIT_VTABLE(u)->audit_start_message_type <= 0)
2582 return;
2583
2584 /* Write audit record if we have just finished starting up */
2585 manager_send_unit_audit(u->manager, u, UNIT_VTABLE(u)->audit_start_message_type, /* success= */ true);
2586 u->in_audit = true;
2587 }
2588
2589 static void unit_emit_audit_stop(Unit *u, UnitActiveState state) {
2590 assert(u);
2591
2592 if (UNIT_VTABLE(u)->audit_start_message_type <= 0)
2593 return;
2594
2595 if (u->in_audit) {
2596 /* Write audit record if we have just finished shutting down */
2597 manager_send_unit_audit(u->manager, u, UNIT_VTABLE(u)->audit_stop_message_type, /* success= */ state == UNIT_INACTIVE);
2598 u->in_audit = false;
2599 } else {
2600 /* Hmm, if there was no start record written write it now, so that we always have a nice pair */
2601 manager_send_unit_audit(u->manager, u, UNIT_VTABLE(u)->audit_start_message_type, /* success= */ state == UNIT_INACTIVE);
2602
2603 if (state == UNIT_INACTIVE)
2604 manager_send_unit_audit(u->manager, u, UNIT_VTABLE(u)->audit_stop_message_type, /* success= */ true);
2605 }
2606 }
2607
2608 static bool unit_process_job(Job *j, UnitActiveState ns, bool reload_success) {
2609 bool unexpected = false;
2610 JobResult result;
2611
2612 assert(j);
2613
2614 if (j->state == JOB_WAITING)
2615 /* So we reached a different state for this job. Let's see if we can run it now if it failed previously
2616 * due to EAGAIN. */
2617 job_add_to_run_queue(j);
2618
2619 /* Let's check whether the unit's new state constitutes a finished job, or maybe contradicts a running job and
2620 * hence needs to invalidate jobs. */
2621
2622 switch (j->type) {
2623
2624 case JOB_START:
2625 case JOB_VERIFY_ACTIVE:
2626
2627 if (UNIT_IS_ACTIVE_OR_RELOADING(ns))
2628 job_finish_and_invalidate(j, JOB_DONE, true, false);
2629 else if (j->state == JOB_RUNNING && ns != UNIT_ACTIVATING) {
2630 unexpected = true;
2631
2632 if (UNIT_IS_INACTIVE_OR_FAILED(ns)) {
2633 if (ns == UNIT_FAILED)
2634 result = JOB_FAILED;
2635 else
2636 result = JOB_DONE;
2637
2638 job_finish_and_invalidate(j, result, true, false);
2639 }
2640 }
2641
2642 break;
2643
2644 case JOB_RELOAD:
2645 case JOB_RELOAD_OR_START:
2646 case JOB_TRY_RELOAD:
2647
2648 if (j->state == JOB_RUNNING) {
2649 if (ns == UNIT_ACTIVE)
2650 job_finish_and_invalidate(j, reload_success ? JOB_DONE : JOB_FAILED, true, false);
2651 else if (!IN_SET(ns, UNIT_ACTIVATING, UNIT_RELOADING, UNIT_REFRESHING)) {
2652 unexpected = true;
2653
2654 if (UNIT_IS_INACTIVE_OR_FAILED(ns))
2655 job_finish_and_invalidate(j, ns == UNIT_FAILED ? JOB_FAILED : JOB_DONE, true, false);
2656 }
2657 }
2658
2659 break;
2660
2661 case JOB_STOP:
2662 case JOB_RESTART:
2663 case JOB_TRY_RESTART:
2664
2665 if (UNIT_IS_INACTIVE_OR_FAILED(ns))
2666 job_finish_and_invalidate(j, JOB_DONE, true, false);
2667 else if (j->state == JOB_RUNNING && ns != UNIT_DEACTIVATING) {
2668 unexpected = true;
2669 job_finish_and_invalidate(j, JOB_FAILED, true, false);
2670 }
2671
2672 break;
2673
2674 default:
2675 assert_not_reached();
2676 }
2677
2678 return unexpected;
2679 }
2680
2681 static void unit_recursive_add_to_run_queue(Unit *u) {
2682 assert(u);
2683
2684 if (u->job)
2685 job_add_to_run_queue(u->job);
2686
2687 Unit *child;
2688 UNIT_FOREACH_DEPENDENCY(child, u, UNIT_ATOM_SLICE_OF) {
2689
2690 if (!child->job)
2691 continue;
2692
2693 unit_recursive_add_to_run_queue(child);
2694 }
2695 }
2696
2697 static void unit_check_concurrency_limit(Unit *u) {
2698 assert(u);
2699
2700 Unit *slice = UNIT_GET_SLICE(u);
2701 if (!slice)
2702 return;
2703
2704 /* If a unit was stopped, maybe it has pending siblings (or children thereof) that can be started now */
2705
2706 if (SLICE(slice)->concurrency_soft_max != UINT_MAX) {
2707 Unit *sibling;
2708 UNIT_FOREACH_DEPENDENCY(sibling, slice, UNIT_ATOM_SLICE_OF) {
2709 if (sibling == u)
2710 continue;
2711
2712 unit_recursive_add_to_run_queue(sibling);
2713 }
2714 }
2715
2716 /* Also go up the tree. */
2717 unit_check_concurrency_limit(slice);
2718 }
2719
2720 void unit_notify(Unit *u, UnitActiveState os, UnitActiveState ns, bool reload_success) {
2721 assert(u);
2722 assert(os < _UNIT_ACTIVE_STATE_MAX);
2723 assert(ns < _UNIT_ACTIVE_STATE_MAX);
2724
2725 /* Note that this is called for all low-level state changes, even if they might map to the same high-level
2726 * UnitActiveState! That means that ns == os is an expected behavior here. For example: if a mount point is
2727 * remounted this function will be called too! */
2728
2729 Manager *m = ASSERT_PTR(u->manager);
2730
2731 /* Let's enqueue the change signal early. In case this unit has a job associated we want that this unit is in
2732 * the bus queue, so that any job change signal queued will force out the unit change signal first. */
2733 unit_add_to_dbus_queue(u);
2734
2735 /* Update systemd-oomd on the property/state change.
2736 *
2737 * Always send an update if the unit is going into an inactive state so systemd-oomd knows to
2738 * stop monitoring.
2739 * Also send an update whenever the unit goes active; this is to handle a case where an override file
2740 * sets one of the ManagedOOM*= properties to "kill", then later removes it. systemd-oomd needs to
2741 * know to stop monitoring when the unit changes from "kill" -> "auto" on daemon-reload, but we don't
2742 * have the information on the property. Thus, indiscriminately send an update. */
2743 if (os != ns && (UNIT_IS_INACTIVE_OR_FAILED(ns) || UNIT_IS_ACTIVE_OR_RELOADING(ns)))
2744 (void) manager_varlink_send_managed_oom_update(u);
2745
2746 /* Update timestamps for state changes */
2747 if (!MANAGER_IS_RELOADING(m)) {
2748 dual_timestamp_now(&u->state_change_timestamp);
2749
2750 if (UNIT_IS_INACTIVE_OR_FAILED(os) && !UNIT_IS_INACTIVE_OR_FAILED(ns))
2751 u->inactive_exit_timestamp = u->state_change_timestamp;
2752 else if (!UNIT_IS_INACTIVE_OR_FAILED(os) && UNIT_IS_INACTIVE_OR_FAILED(ns))
2753 u->inactive_enter_timestamp = u->state_change_timestamp;
2754
2755 if (!UNIT_IS_ACTIVE_OR_RELOADING(os) && UNIT_IS_ACTIVE_OR_RELOADING(ns))
2756 u->active_enter_timestamp = u->state_change_timestamp;
2757 else if (UNIT_IS_ACTIVE_OR_RELOADING(os) && !UNIT_IS_ACTIVE_OR_RELOADING(ns))
2758 u->active_exit_timestamp = u->state_change_timestamp;
2759 }
2760
2761 /* Keep track of failed units */
2762 (void) manager_update_failed_units(m, u, ns == UNIT_FAILED);
2763
2764 /* Make sure the cgroup and state files are always removed when we become inactive */
2765 if (UNIT_IS_INACTIVE_OR_FAILED(ns)) {
2766 SET_FLAG(u->markers,
2767 (1u << UNIT_MARKER_NEEDS_RELOAD)|(1u << UNIT_MARKER_NEEDS_RESTART),
2768 false);
2769 unit_prune_cgroup(u);
2770 unit_unlink_state_files(u);
2771 } else if (ns != os && ns == UNIT_RELOADING)
2772 SET_FLAG(u->markers, 1u << UNIT_MARKER_NEEDS_RELOAD, false);
2773
2774 unit_update_on_console(u);
2775
2776 if (!MANAGER_IS_RELOADING(m)) {
2777 bool unexpected;
2778
2779 /* Let's propagate state changes to the job */
2780 if (u->job)
2781 unexpected = unit_process_job(u->job, ns, reload_success);
2782 else
2783 unexpected = true;
2784
2785 /* If this state change happened without being requested by a job, then let's retroactively start or
2786 * stop dependencies. We skip that step when deserializing, since we don't want to create any
2787 * additional jobs just because something is already activated. */
2788
2789 if (unexpected) {
2790 if (UNIT_IS_INACTIVE_OR_FAILED(os) && UNIT_IS_ACTIVE_OR_ACTIVATING(ns))
2791 retroactively_start_dependencies(u);
2792 else if (UNIT_IS_ACTIVE_OR_ACTIVATING(os) && UNIT_IS_INACTIVE_OR_DEACTIVATING(ns))
2793 retroactively_stop_dependencies(u);
2794 }
2795
2796 if (UNIT_IS_ACTIVE_OR_RELOADING(ns) && !UNIT_IS_ACTIVE_OR_RELOADING(os)) {
2797 /* This unit just finished starting up */
2798
2799 unit_emit_audit_start(u);
2800 manager_send_unit_plymouth(m, u);
2801 manager_send_unit_supervisor(m, u, /* active= */ true);
2802
2803 } else if (UNIT_IS_INACTIVE_OR_FAILED(ns) && !UNIT_IS_INACTIVE_OR_FAILED(os)) {
2804 /* This unit just stopped/failed. */
2805
2806 unit_emit_audit_stop(u, ns);
2807 manager_send_unit_supervisor(m, u, /* active= */ false);
2808 unit_log_resources(u);
2809 }
2810
2811 if (ns == UNIT_INACTIVE && !IN_SET(os, UNIT_FAILED, UNIT_INACTIVE, UNIT_MAINTENANCE))
2812 unit_start_on_termination_deps(u, UNIT_ATOM_ON_SUCCESS);
2813 else if (ns != os && ns == UNIT_FAILED)
2814 unit_start_on_termination_deps(u, UNIT_ATOM_ON_FAILURE);
2815 }
2816
2817 manager_recheck_journal(m);
2818 manager_recheck_dbus(m);
2819
2820 unit_trigger_notify(u);
2821
2822 if (!MANAGER_IS_RELOADING(m)) {
2823 const char *reason;
2824
2825 if (os != UNIT_FAILED && ns == UNIT_FAILED) {
2826 reason = strjoina("unit ", u->id, " failed");
2827 emergency_action(m, u->failure_action, EMERGENCY_ACTION_WARN|EMERGENCY_ACTION_SLEEP_5S, u->reboot_arg, unit_failure_action_exit_status(u), reason);
2828 } else if (!UNIT_IS_INACTIVE_OR_FAILED(os) && ns == UNIT_INACTIVE) {
2829 reason = strjoina("unit ", u->id, " succeeded");
2830 emergency_action(m, u->success_action, /* flags= */ 0, u->reboot_arg, unit_success_action_exit_status(u), reason);
2831 }
2832 }
2833
2834 /* And now, add the unit or depending units to various queues that will act on the new situation if
2835 * needed. These queues generally check for continuous state changes rather than events (like most of
2836 * the state propagation above), and do work deferred instead of instantly, since they typically
2837 * don't want to run during reloading, and usually involve checking combined state of multiple units
2838 * at once. */
2839
2840 if (UNIT_IS_INACTIVE_OR_FAILED(ns)) {
2841 /* Stop unneeded units and bound-by units regardless if going down was expected or not */
2842 check_unneeded_dependencies(u);
2843 check_bound_by_dependencies(u);
2844
2845 /* Maybe someone wants us to remain up? */
2846 unit_submit_to_start_when_upheld_queue(u);
2847
2848 /* Maybe the unit should be GC'ed now? */
2849 unit_add_to_gc_queue(u);
2850
2851 /* Maybe we can release some resources now? */
2852 unit_submit_to_release_resources_queue(u);
2853
2854 /* Maybe the concurrency limits now allow dispatching of another start job in this slice? */
2855 unit_check_concurrency_limit(u);
2856
2857 /* Maybe someone else has been waiting for us to stop? */
2858 m->may_dispatch_stop_notify_queue = true;
2859
2860 } else if (UNIT_IS_ACTIVE_OR_RELOADING(ns)) {
2861 /* Start uphold units regardless if going up was expected or not */
2862 check_uphold_dependencies(u);
2863
2864 /* Maybe we finished startup and are now ready for being stopped because unneeded? */
2865 unit_submit_to_stop_when_unneeded_queue(u);
2866
2867 /* Maybe we finished startup, but something we needed has vanished? Let's die then. (This happens
2868 * when something BindsTo= to a Type=oneshot unit, as these units go directly from starting to
2869 * inactive, without ever entering started.) */
2870 unit_submit_to_stop_when_bound_queue(u);
2871 }
2872 }
2873
2874 int unit_watch_pidref(Unit *u, const PidRef *pid, bool exclusive) {
2875 _cleanup_(pidref_freep) PidRef *pid_dup = NULL;
2876 int r;
2877
2878 /* Adds a specific PID to the set of PIDs this unit watches. */
2879
2880 assert(u);
2881 assert(pidref_is_set(pid));
2882
2883 /* Caller might be sure that this PID belongs to this unit only. Let's take this
2884 * opportunity to remove any stalled references to this PID as they can be created
2885 * easily (when watching a process which is not our direct child). */
2886 if (exclusive)
2887 manager_unwatch_pidref(u->manager, pid);
2888
2889 if (set_contains(u->pids, pid)) { /* early exit if already being watched */
2890 assert(!exclusive);
2891 return 0;
2892 }
2893
2894 r = pidref_dup(pid, &pid_dup);
2895 if (r < 0)
2896 return r;
2897
2898 /* First, insert into the set of PIDs maintained by the unit */
2899 r = set_ensure_put(&u->pids, &pidref_hash_ops_free, pid_dup);
2900 if (r < 0)
2901 return r;
2902
2903 pid = TAKE_PTR(pid_dup); /* continue with our copy now that we have installed it properly in our set */
2904
2905 /* Second, insert it into the simple global table, see if that works */
2906 r = hashmap_ensure_put(&u->manager->watch_pids, &pidref_hash_ops, pid, u);
2907 if (r != -EEXIST)
2908 return r;
2909
2910 /* OK, the key is already assigned to a different unit. That's fine, then add us via the second
2911 * hashmap that points to an array. */
2912
2913 PidRef *old_pid = NULL;
2914 Unit **array = hashmap_get2(u->manager->watch_pids_more, pid, (void**) &old_pid);
2915
2916 /* Count entries in array */
2917 size_t n = 0;
2918 for (; array && array[n]; n++)
2919 ;
2920
2921 /* Allocate a new array */
2922 _cleanup_free_ Unit **new_array = new(Unit*, n + 2);
2923 if (!new_array)
2924 return -ENOMEM;
2925
2926 /* Append us to the end */
2927 memcpy_safe(new_array, array, sizeof(Unit*) * n);
2928 new_array[n] = u;
2929 new_array[n+1] = NULL;
2930
2931 /* Add or replace the old array */
2932 r = hashmap_ensure_replace(&u->manager->watch_pids_more, &pidref_hash_ops, old_pid ?: pid, new_array);
2933 if (r < 0)
2934 return r;
2935
2936 TAKE_PTR(new_array); /* Now part of the hash table */
2937 free(array); /* Which means we can now delete the old version */
2938 return 0;
2939 }
2940
2941 void unit_unwatch_pidref(Unit *u, const PidRef *pid) {
2942 assert(u);
2943 assert(pidref_is_set(pid));
2944
2945 /* Remove from the set we maintain for this unit. (And destroy the returned pid eventually) */
2946 _cleanup_(pidref_freep) PidRef *pid1 = set_remove(u->pids, pid);
2947 if (!pid1)
2948 return; /* Early exit if this PID was never watched by us */
2949
2950 /* First let's drop the unit from the simple hash table, if it is included there */
2951 PidRef *pid2 = NULL;
2952 Unit *uu = hashmap_get2(u->manager->watch_pids, pid, (void**) &pid2);
2953
2954 /* Quick validation: iff we are in the watch_pids table then the PidRef object must be the same as in our local pids set */
2955 assert((uu == u) == (pid1 == pid2));
2956
2957 if (uu == u)
2958 /* OK, we are in the first table. Let's remove it there then, and we are done already. */
2959 assert_se(hashmap_remove_value(u->manager->watch_pids, pid2, uu));
2960 else {
2961 /* We weren't in the first table, then let's consult the 2nd table that points to an array */
2962 PidRef *pid3 = NULL;
2963 Unit **array = hashmap_get2(u->manager->watch_pids_more, pid, (void**) &pid3);
2964
2965 /* Let's iterate through the array, dropping our own entry */
2966 size_t m = 0, n = 0;
2967 for (; array && array[n]; n++)
2968 if (array[n] != u)
2969 array[m++] = array[n];
2970 if (n == m)
2971 return; /* Not there */
2972
2973 array[m] = NULL; /* set trailing NULL marker on the new end */
2974
2975 if (m == 0) {
2976 /* The array is now empty, remove the entire entry */
2977 assert_se(hashmap_remove_value(u->manager->watch_pids_more, pid3, array));
2978 free(array);
2979 } else {
2980 /* The array is not empty, but let's make sure the entry is not keyed by the PidRef
2981 * we will delete, but by the PidRef object of the Unit that is now first in the
2982 * array. */
2983
2984 PidRef *new_pid3 = ASSERT_PTR(set_get(array[0]->pids, pid));
2985 assert_se(hashmap_replace(u->manager->watch_pids_more, new_pid3, array) >= 0);
2986 }
2987 }
2988 }
2989
2990 void unit_unwatch_all_pids(Unit *u) {
2991 assert(u);
2992
2993 while (!set_isempty(u->pids))
2994 unit_unwatch_pidref(u, set_first(u->pids));
2995
2996 u->pids = set_free(u->pids);
2997 }
2998
2999 void unit_unwatch_pidref_done(Unit *u, PidRef *pidref) {
3000 assert(u);
3001
3002 if (!pidref_is_set(pidref))
3003 return;
3004
3005 unit_unwatch_pidref(u, pidref);
3006 pidref_done(pidref);
3007 }
3008
3009 bool unit_job_is_applicable(Unit *u, JobType j) {
3010 assert(u);
3011 assert(j >= 0 && j < _JOB_TYPE_MAX);
3012
3013 switch (j) {
3014
3015 case JOB_VERIFY_ACTIVE:
3016 case JOB_START:
3017 case JOB_NOP:
3018 /* Note that we don't check unit_can_start() here. That's because .device units and suchlike are not
3019 * startable by us but may appear due to external events, and it thus makes sense to permit enqueuing
3020 * jobs for it. */
3021 return true;
3022
3023 case JOB_STOP:
3024 /* Similar as above. However, perpetual units can never be stopped (neither explicitly nor due to
3025 * external events), hence it makes no sense to permit enqueuing such a request either. */
3026 return !u->perpetual;
3027
3028 case JOB_RESTART:
3029 case JOB_TRY_RESTART:
3030 return unit_can_stop(u) && unit_can_start(u);
3031
3032 case JOB_RELOAD:
3033 case JOB_TRY_RELOAD:
3034 return unit_can_reload(u);
3035
3036 case JOB_RELOAD_OR_START:
3037 return unit_can_reload(u) && unit_can_start(u);
3038
3039 default:
3040 assert_not_reached();
3041 }
3042 }
3043
3044 static Hashmap *unit_get_dependency_hashmap_per_type(Unit *u, UnitDependency d) {
3045 Hashmap *deps;
3046
3047 assert(u);
3048 assert(d >= 0 && d < _UNIT_DEPENDENCY_MAX);
3049
3050 deps = hashmap_get(u->dependencies, UNIT_DEPENDENCY_TO_PTR(d));
3051 if (!deps) {
3052 _cleanup_hashmap_free_ Hashmap *h = NULL;
3053
3054 h = hashmap_new(NULL);
3055 if (!h)
3056 return NULL;
3057
3058 if (hashmap_ensure_put(&u->dependencies, NULL, UNIT_DEPENDENCY_TO_PTR(d), h) < 0)
3059 return NULL;
3060
3061 deps = TAKE_PTR(h);
3062 }
3063
3064 return deps;
3065 }
3066
3067 typedef enum NotifyDependencyFlags {
3068 NOTIFY_DEPENDENCY_UPDATE_FROM = 1 << 0,
3069 NOTIFY_DEPENDENCY_UPDATE_TO = 1 << 1,
3070 } NotifyDependencyFlags;
3071
3072 static int unit_add_dependency_impl(
3073 Unit *u,
3074 UnitDependency d,
3075 Unit *other,
3076 UnitDependencyMask mask) {
3077
3078 static const UnitDependency inverse_table[_UNIT_DEPENDENCY_MAX] = {
3079 [UNIT_REQUIRES] = UNIT_REQUIRED_BY,
3080 [UNIT_REQUISITE] = UNIT_REQUISITE_OF,
3081 [UNIT_WANTS] = UNIT_WANTED_BY,
3082 [UNIT_BINDS_TO] = UNIT_BOUND_BY,
3083 [UNIT_PART_OF] = UNIT_CONSISTS_OF,
3084 [UNIT_UPHOLDS] = UNIT_UPHELD_BY,
3085 [UNIT_REQUIRED_BY] = UNIT_REQUIRES,
3086 [UNIT_REQUISITE_OF] = UNIT_REQUISITE,
3087 [UNIT_WANTED_BY] = UNIT_WANTS,
3088 [UNIT_BOUND_BY] = UNIT_BINDS_TO,
3089 [UNIT_CONSISTS_OF] = UNIT_PART_OF,
3090 [UNIT_UPHELD_BY] = UNIT_UPHOLDS,
3091 [UNIT_CONFLICTS] = UNIT_CONFLICTED_BY,
3092 [UNIT_CONFLICTED_BY] = UNIT_CONFLICTS,
3093 [UNIT_BEFORE] = UNIT_AFTER,
3094 [UNIT_AFTER] = UNIT_BEFORE,
3095 [UNIT_ON_SUCCESS] = UNIT_ON_SUCCESS_OF,
3096 [UNIT_ON_SUCCESS_OF] = UNIT_ON_SUCCESS,
3097 [UNIT_ON_FAILURE] = UNIT_ON_FAILURE_OF,
3098 [UNIT_ON_FAILURE_OF] = UNIT_ON_FAILURE,
3099 [UNIT_TRIGGERS] = UNIT_TRIGGERED_BY,
3100 [UNIT_TRIGGERED_BY] = UNIT_TRIGGERS,
3101 [UNIT_PROPAGATES_RELOAD_TO] = UNIT_RELOAD_PROPAGATED_FROM,
3102 [UNIT_RELOAD_PROPAGATED_FROM] = UNIT_PROPAGATES_RELOAD_TO,
3103 [UNIT_PROPAGATES_STOP_TO] = UNIT_STOP_PROPAGATED_FROM,
3104 [UNIT_STOP_PROPAGATED_FROM] = UNIT_PROPAGATES_STOP_TO,
3105 [UNIT_JOINS_NAMESPACE_OF] = UNIT_JOINS_NAMESPACE_OF, /* symmetric! 👓 */
3106 [UNIT_REFERENCES] = UNIT_REFERENCED_BY,
3107 [UNIT_REFERENCED_BY] = UNIT_REFERENCES,
3108 [UNIT_IN_SLICE] = UNIT_SLICE_OF,
3109 [UNIT_SLICE_OF] = UNIT_IN_SLICE,
3110 };
3111
3112 Hashmap *u_deps, *other_deps;
3113 UnitDependencyInfo u_info, u_info_old, other_info, other_info_old;
3114 NotifyDependencyFlags flags = 0;
3115 int r;
3116
3117 assert(u);
3118 assert(other);
3119 assert(d >= 0 && d < _UNIT_DEPENDENCY_MAX);
3120 assert(inverse_table[d] >= 0 && inverse_table[d] < _UNIT_DEPENDENCY_MAX);
3121 assert(mask > 0 && mask < _UNIT_DEPENDENCY_MASK_FULL);
3122
3123 /* Ensure the following two hashmaps for each unit exist:
3124 * - the top-level dependency hashmap that maps UnitDependency → Hashmap(Unit* → UnitDependencyInfo),
3125 * - the inner hashmap, that maps Unit* → UnitDependencyInfo, for the specified dependency type. */
3126 u_deps = unit_get_dependency_hashmap_per_type(u, d);
3127 if (!u_deps)
3128 return -ENOMEM;
3129
3130 other_deps = unit_get_dependency_hashmap_per_type(other, inverse_table[d]);
3131 if (!other_deps)
3132 return -ENOMEM;
3133
3134 /* Save the original dependency info. */
3135 u_info.data = u_info_old.data = hashmap_get(u_deps, other);
3136 other_info.data = other_info_old.data = hashmap_get(other_deps, u);
3137
3138 /* Update dependency info. */
3139 u_info.origin_mask |= mask;
3140 other_info.destination_mask |= mask;
3141
3142 /* Save updated dependency info. */
3143 if (u_info.data != u_info_old.data) {
3144 r = hashmap_replace(u_deps, other, u_info.data);
3145 if (r < 0)
3146 return r;
3147
3148 flags = NOTIFY_DEPENDENCY_UPDATE_FROM;
3149 u->dependency_generation++;
3150 }
3151
3152 if (other_info.data != other_info_old.data) {
3153 r = hashmap_replace(other_deps, u, other_info.data);
3154 if (r < 0) {
3155 if (u_info.data != u_info_old.data) {
3156 /* Restore the old dependency. */
3157 if (u_info_old.data)
3158 (void) hashmap_update(u_deps, other, u_info_old.data);
3159 else
3160 hashmap_remove(u_deps, other);
3161 }
3162 return r;
3163 }
3164
3165 flags |= NOTIFY_DEPENDENCY_UPDATE_TO;
3166 other->dependency_generation++;
3167 }
3168
3169 return flags;
3170 }
3171
3172 int unit_add_dependency(
3173 Unit *u,
3174 UnitDependency d,
3175 Unit *other,
3176 bool add_reference,
3177 UnitDependencyMask mask) {
3178
3179 UnitDependencyAtom a;
3180 int r;
3181
3182 /* Helper to know whether sending a notification is necessary or not: if the dependency is already
3183 * there, no need to notify! */
3184 NotifyDependencyFlags notify_flags;
3185
3186 assert(u);
3187 assert(d >= 0 && d < _UNIT_DEPENDENCY_MAX);
3188 assert(other);
3189
3190 u = unit_follow_merge(u);
3191 other = unit_follow_merge(other);
3192 a = unit_dependency_to_atom(d);
3193 assert(a >= 0);
3194
3195 /* We won't allow dependencies on ourselves. We will not consider them an error however. */
3196 if (u == other) {
3197 if (unit_should_warn_about_dependency(d))
3198 log_unit_warning(u, "Dependency %s=%s is dropped.",
3199 unit_dependency_to_string(d), u->id);
3200 return 0;
3201 }
3202
3203 if (u->manager && FLAGS_SET(u->manager->test_run_flags, MANAGER_TEST_RUN_IGNORE_DEPENDENCIES))
3204 return 0;
3205
3206 /* Note that ordering a device unit after a unit is permitted since it allows its job running
3207 * timeout to be started at a specific time. */
3208 if (FLAGS_SET(a, UNIT_ATOM_BEFORE) && other->type == UNIT_DEVICE) {
3209 log_unit_warning(u, "Dependency Before=%s ignored (.device units cannot be delayed)", other->id);
3210 return 0;
3211 }
3212
3213 if (FLAGS_SET(a, UNIT_ATOM_ON_FAILURE) && !UNIT_VTABLE(u)->can_fail) {
3214 log_unit_warning(u, "Requested dependency OnFailure=%s ignored (%s units cannot fail).", other->id, unit_type_to_string(u->type));
3215 return 0;
3216 }
3217
3218 if (FLAGS_SET(a, UNIT_ATOM_TRIGGERS) && !UNIT_VTABLE(u)->can_trigger)
3219 return log_unit_error_errno(u, SYNTHETIC_ERRNO(EINVAL),
3220 "Requested dependency Triggers=%s refused (%s units cannot trigger other units).", other->id, unit_type_to_string(u->type));
3221 if (FLAGS_SET(a, UNIT_ATOM_TRIGGERED_BY) && !UNIT_VTABLE(other)->can_trigger)
3222 return log_unit_error_errno(u, SYNTHETIC_ERRNO(EINVAL),
3223 "Requested dependency TriggeredBy=%s refused (%s units cannot trigger other units).", other->id, unit_type_to_string(other->type));
3224
3225 if (FLAGS_SET(a, UNIT_ATOM_IN_SLICE) && other->type != UNIT_SLICE)
3226 return log_unit_error_errno(u, SYNTHETIC_ERRNO(EINVAL),
3227 "Requested dependency Slice=%s refused (%s is not a slice unit).", other->id, other->id);
3228 if (FLAGS_SET(a, UNIT_ATOM_SLICE_OF) && u->type != UNIT_SLICE)
3229 return log_unit_error_errno(u, SYNTHETIC_ERRNO(EINVAL),
3230 "Requested dependency SliceOf=%s refused (%s is not a slice unit).", other->id, u->id);
3231
3232 if (FLAGS_SET(a, UNIT_ATOM_IN_SLICE) && !UNIT_HAS_CGROUP_CONTEXT(u))
3233 return log_unit_error_errno(u, SYNTHETIC_ERRNO(EINVAL),
3234 "Requested dependency Slice=%s refused (%s is not a cgroup unit).", other->id, u->id);
3235
3236 if (FLAGS_SET(a, UNIT_ATOM_SLICE_OF) && !UNIT_HAS_CGROUP_CONTEXT(other))
3237 return log_unit_error_errno(u, SYNTHETIC_ERRNO(EINVAL),
3238 "Requested dependency SliceOf=%s refused (%s is not a cgroup unit).", other->id, other->id);
3239
3240 r = unit_add_dependency_impl(u, d, other, mask);
3241 if (r < 0)
3242 return r;
3243 notify_flags = r;
3244
3245 if (add_reference) {
3246 r = unit_add_dependency_impl(u, UNIT_REFERENCES, other, mask);
3247 if (r < 0)
3248 return r;
3249 notify_flags |= r;
3250 }
3251
3252 if (FLAGS_SET(notify_flags, NOTIFY_DEPENDENCY_UPDATE_FROM))
3253 unit_add_to_dbus_queue(u);
3254 if (FLAGS_SET(notify_flags, NOTIFY_DEPENDENCY_UPDATE_TO))
3255 unit_add_to_dbus_queue(other);
3256
3257 return notify_flags != 0;
3258 }
3259
3260 int unit_add_two_dependencies(Unit *u, UnitDependency d, UnitDependency e, Unit *other, bool add_reference, UnitDependencyMask mask) {
3261 int r = 0, s = 0;
3262
3263 assert(u);
3264 assert(d >= 0 || e >= 0);
3265
3266 if (d >= 0) {
3267 r = unit_add_dependency(u, d, other, add_reference, mask);
3268 if (r < 0)
3269 return r;
3270 }
3271
3272 if (e >= 0) {
3273 s = unit_add_dependency(u, e, other, add_reference, mask);
3274 if (s < 0)
3275 return s;
3276 }
3277
3278 return r > 0 || s > 0;
3279 }
3280
3281 static int resolve_template(Unit *u, const char *name, char **buf, const char **ret) {
3282 int r;
3283
3284 assert(u);
3285 assert(name);
3286 assert(buf);
3287 assert(ret);
3288
3289 if (!unit_name_is_valid(name, UNIT_NAME_TEMPLATE)) {
3290 *buf = NULL;
3291 *ret = name;
3292 return 0;
3293 }
3294
3295 if (u->instance)
3296 r = unit_name_replace_instance(name, u->instance, buf);
3297 else {
3298 _cleanup_free_ char *i = NULL;
3299
3300 r = unit_name_to_prefix(u->id, &i);
3301 if (r < 0)
3302 return r;
3303
3304 r = unit_name_replace_instance(name, i, buf);
3305 }
3306 if (r < 0)
3307 return r;
3308
3309 *ret = *buf;
3310 return 0;
3311 }
3312
3313 int unit_add_dependency_by_name(Unit *u, UnitDependency d, const char *name, bool add_reference, UnitDependencyMask mask) {
3314 _cleanup_free_ char *buf = NULL;
3315 Unit *other;
3316 int r;
3317
3318 assert(u);
3319 assert(name);
3320
3321 r = resolve_template(u, name, &buf, &name);
3322 if (r < 0)
3323 return r;
3324
3325 if (u->manager && FLAGS_SET(u->manager->test_run_flags, MANAGER_TEST_RUN_IGNORE_DEPENDENCIES))
3326 return 0;
3327
3328 r = manager_load_unit(u->manager, name, NULL, NULL, &other);
3329 if (r < 0)
3330 return r;
3331
3332 return unit_add_dependency(u, d, other, add_reference, mask);
3333 }
3334
3335 int unit_add_two_dependencies_by_name(Unit *u, UnitDependency d, UnitDependency e, const char *name, bool add_reference, UnitDependencyMask mask) {
3336 _cleanup_free_ char *buf = NULL;
3337 Unit *other;
3338 int r;
3339
3340 assert(u);
3341 assert(name);
3342
3343 r = resolve_template(u, name, &buf, &name);
3344 if (r < 0)
3345 return r;
3346
3347 if (u->manager && FLAGS_SET(u->manager->test_run_flags, MANAGER_TEST_RUN_IGNORE_DEPENDENCIES))
3348 return 0;
3349
3350 r = manager_load_unit(u->manager, name, NULL, NULL, &other);
3351 if (r < 0)
3352 return r;
3353
3354 return unit_add_two_dependencies(u, d, e, other, add_reference, mask);
3355 }
3356
3357 int setenv_unit_path(const char *p) {
3358 assert(p);
3359
3360 /* This is mostly for debug purposes */
3361 return RET_NERRNO(setenv("SYSTEMD_UNIT_PATH", p, /* overwrite = */ true));
3362 }
3363
3364 char* unit_dbus_path(Unit *u) {
3365 assert(u);
3366
3367 if (!u->id)
3368 return NULL;
3369
3370 return unit_dbus_path_from_name(u->id);
3371 }
3372
3373 char* unit_dbus_path_invocation_id(Unit *u) {
3374 assert(u);
3375
3376 if (sd_id128_is_null(u->invocation_id))
3377 return NULL;
3378
3379 return unit_dbus_path_from_name(u->invocation_id_string);
3380 }
3381
3382 int unit_set_invocation_id(Unit *u, sd_id128_t id) {
3383 int r;
3384
3385 assert(u);
3386
3387 /* Set the invocation ID for this unit. If we cannot, this will not roll back, but reset the whole thing. */
3388
3389 if (sd_id128_equal(u->invocation_id, id))
3390 return 0;
3391
3392 if (!sd_id128_is_null(u->invocation_id))
3393 (void) hashmap_remove_value(u->manager->units_by_invocation_id, &u->invocation_id, u);
3394
3395 if (sd_id128_is_null(id)) {
3396 r = 0;
3397 goto reset;
3398 }
3399
3400 r = hashmap_ensure_allocated(&u->manager->units_by_invocation_id, &id128_hash_ops);
3401 if (r < 0)
3402 goto reset;
3403
3404 u->invocation_id = id;
3405 sd_id128_to_string(id, u->invocation_id_string);
3406
3407 r = hashmap_put(u->manager->units_by_invocation_id, &u->invocation_id, u);
3408 if (r < 0)
3409 goto reset;
3410
3411 return 0;
3412
3413 reset:
3414 u->invocation_id = SD_ID128_NULL;
3415 u->invocation_id_string[0] = 0;
3416 return r;
3417 }
3418
3419 int unit_set_slice(Unit *u, Unit *slice) {
3420 int r;
3421
3422 assert(u);
3423 assert(slice);
3424
3425 /* Sets the unit slice if it has not been set before. Is extra careful, to only allow this for units
3426 * that actually have a cgroup context. Also, we don't allow to set this for slices (since the parent
3427 * slice is derived from the name). Make sure the unit we set is actually a slice. */
3428
3429 if (!UNIT_HAS_CGROUP_CONTEXT(u))
3430 return -EOPNOTSUPP;
3431
3432 if (u->type == UNIT_SLICE)
3433 return -EINVAL;
3434
3435 if (unit_active_state(u) != UNIT_INACTIVE)
3436 return -EBUSY;
3437
3438 if (slice->type != UNIT_SLICE)
3439 return -EINVAL;
3440
3441 if (unit_has_name(u, SPECIAL_INIT_SCOPE) &&
3442 !unit_has_name(slice, SPECIAL_ROOT_SLICE))
3443 return -EPERM;
3444
3445 if (UNIT_GET_SLICE(u) == slice)
3446 return 0;
3447
3448 /* Disallow slice changes if @u is already bound to cgroups */
3449 if (UNIT_GET_SLICE(u)) {
3450 CGroupRuntime *crt = unit_get_cgroup_runtime(u);
3451 if (crt && crt->cgroup_path)
3452 return -EBUSY;
3453 }
3454
3455 /* Remove any slices assigned prior; we should only have one UNIT_IN_SLICE dependency */
3456 if (UNIT_GET_SLICE(u))
3457 unit_remove_dependencies(u, UNIT_DEPENDENCY_SLICE_PROPERTY);
3458
3459 r = unit_add_dependency(u, UNIT_IN_SLICE, slice, true, UNIT_DEPENDENCY_SLICE_PROPERTY);
3460 if (r < 0)
3461 return r;
3462
3463 return 1;
3464 }
3465
3466 int unit_set_default_slice(Unit *u) {
3467 const char *slice_name;
3468 Unit *slice;
3469 int r;
3470
3471 assert(u);
3472
3473 if (u->manager && FLAGS_SET(u->manager->test_run_flags, MANAGER_TEST_RUN_IGNORE_DEPENDENCIES))
3474 return 0;
3475
3476 if (UNIT_GET_SLICE(u))
3477 return 0;
3478
3479 if (u->instance) {
3480 _cleanup_free_ char *prefix = NULL, *escaped = NULL;
3481
3482 /* Implicitly place all instantiated units in their
3483 * own per-template slice */
3484
3485 r = unit_name_to_prefix(u->id, &prefix);
3486 if (r < 0)
3487 return r;
3488
3489 /* The prefix is already escaped, but it might include
3490 * "-" which has a special meaning for slice units,
3491 * hence escape it here extra. */
3492 escaped = unit_name_escape(prefix);
3493 if (!escaped)
3494 return -ENOMEM;
3495
3496 if (MANAGER_IS_SYSTEM(u->manager))
3497 slice_name = strjoina("system-", escaped, ".slice");
3498 else
3499 slice_name = strjoina("app-", escaped, ".slice");
3500
3501 } else if (unit_is_extrinsic(u))
3502 /* Keep all extrinsic units (e.g. perpetual units and swap and mount units in user mode) in
3503 * the root slice. They don't really belong in one of the subslices. */
3504 slice_name = SPECIAL_ROOT_SLICE;
3505
3506 else if (MANAGER_IS_SYSTEM(u->manager))
3507 slice_name = SPECIAL_SYSTEM_SLICE;
3508 else
3509 slice_name = SPECIAL_APP_SLICE;
3510
3511 r = manager_load_unit(u->manager, slice_name, NULL, NULL, &slice);
3512 if (r < 0)
3513 return r;
3514
3515 return unit_set_slice(u, slice);
3516 }
3517
3518 const char* unit_slice_name(Unit *u) {
3519 Unit *slice;
3520 assert(u);
3521
3522 slice = UNIT_GET_SLICE(u);
3523 if (!slice)
3524 return NULL;
3525
3526 return slice->id;
3527 }
3528
3529 int unit_load_related_unit(Unit *u, const char *type, Unit **_found) {
3530 _cleanup_free_ char *t = NULL;
3531 int r;
3532
3533 assert(u);
3534 assert(type);
3535 assert(_found);
3536
3537 r = unit_name_change_suffix(u->id, type, &t);
3538 if (r < 0)
3539 return r;
3540 if (unit_has_name(u, t))
3541 return -EINVAL;
3542
3543 r = manager_load_unit(u->manager, t, NULL, NULL, _found);
3544 assert(r < 0 || *_found != u);
3545 return r;
3546 }
3547
3548 static int signal_name_owner_changed_install_handler(sd_bus_message *message, void *userdata, sd_bus_error *error) {
3549 Unit *u = ASSERT_PTR(userdata);
3550 const sd_bus_error *e;
3551 int r;
3552
3553 e = sd_bus_message_get_error(message);
3554 if (!e) {
3555 log_unit_trace(u, "Successfully installed NameOwnerChanged signal match.");
3556 return 0;
3557 }
3558
3559 r = sd_bus_error_get_errno(e);
3560 log_unit_error_errno(u, r,
3561 "Unexpected error response on installing NameOwnerChanged signal match: %s",
3562 bus_error_message(e, r));
3563
3564 /* If we failed to install NameOwnerChanged signal, also unref the bus slot of GetNameOwner(). */
3565 u->match_bus_slot = sd_bus_slot_unref(u->match_bus_slot);
3566 u->get_name_owner_slot = sd_bus_slot_unref(u->get_name_owner_slot);
3567
3568 if (UNIT_VTABLE(u)->bus_name_owner_change)
3569 UNIT_VTABLE(u)->bus_name_owner_change(u, NULL);
3570
3571 return 0;
3572 }
3573
3574 static int signal_name_owner_changed(sd_bus_message *message, void *userdata, sd_bus_error *error) {
3575 const char *new_owner;
3576 Unit *u = ASSERT_PTR(userdata);
3577 int r;
3578
3579 assert(message);
3580
3581 r = sd_bus_message_read(message, "sss", NULL, NULL, &new_owner);
3582 if (r < 0) {
3583 bus_log_parse_error(r);
3584 return 0;
3585 }
3586
3587 if (UNIT_VTABLE(u)->bus_name_owner_change)
3588 UNIT_VTABLE(u)->bus_name_owner_change(u, empty_to_null(new_owner));
3589
3590 return 0;
3591 }
3592
3593 static int get_name_owner_handler(sd_bus_message *message, void *userdata, sd_bus_error *error) {
3594 const sd_bus_error *e;
3595 const char *new_owner;
3596 Unit *u = ASSERT_PTR(userdata);
3597 int r;
3598
3599 assert(message);
3600
3601 u->get_name_owner_slot = sd_bus_slot_unref(u->get_name_owner_slot);
3602
3603 e = sd_bus_message_get_error(message);
3604 if (e) {
3605 if (!sd_bus_error_has_name(e, SD_BUS_ERROR_NAME_HAS_NO_OWNER)) {
3606 r = sd_bus_error_get_errno(e);
3607 log_unit_error_errno(u, r,
3608 "Unexpected error response from GetNameOwner(): %s",
3609 bus_error_message(e, r));
3610 }
3611
3612 new_owner = NULL;
3613 } else {
3614 r = sd_bus_message_read(message, "s", &new_owner);
3615 if (r < 0)
3616 return bus_log_parse_error(r);
3617
3618 assert(!isempty(new_owner));
3619 }
3620
3621 if (UNIT_VTABLE(u)->bus_name_owner_change)
3622 UNIT_VTABLE(u)->bus_name_owner_change(u, new_owner);
3623
3624 return 0;
3625 }
3626
3627 int unit_install_bus_match(Unit *u, sd_bus *bus, const char *name) {
3628 _cleanup_(sd_bus_message_unrefp) sd_bus_message *m = NULL;
3629 const char *match;
3630 usec_t timeout_usec = 0;
3631 int r;
3632
3633 assert(u);
3634 assert(bus);
3635 assert(name);
3636
3637 if (u->match_bus_slot || u->get_name_owner_slot)
3638 return -EBUSY;
3639
3640 /* NameOwnerChanged and GetNameOwner is used to detect when a service finished starting up. The dbus
3641 * call timeout shouldn't be earlier than that. If we couldn't get the start timeout, use the default
3642 * value defined above. */
3643 if (UNIT_VTABLE(u)->get_timeout_start_usec)
3644 timeout_usec = UNIT_VTABLE(u)->get_timeout_start_usec(u);
3645
3646 match = strjoina("type='signal',"
3647 "sender='org.freedesktop.DBus',"
3648 "path='/org/freedesktop/DBus',"
3649 "interface='org.freedesktop.DBus',"
3650 "member='NameOwnerChanged',"
3651 "arg0='", name, "'");
3652
3653 r = bus_add_match_full(
3654 bus,
3655 &u->match_bus_slot,
3656 /* asynchronous = */ true,
3657 match,
3658 signal_name_owner_changed,
3659 signal_name_owner_changed_install_handler,
3660 u,
3661 timeout_usec);
3662 if (r < 0)
3663 return r;
3664
3665 r = sd_bus_message_new_method_call(
3666 bus,
3667 &m,
3668 "org.freedesktop.DBus",
3669 "/org/freedesktop/DBus",
3670 "org.freedesktop.DBus",
3671 "GetNameOwner");
3672 if (r < 0)
3673 return r;
3674
3675 r = sd_bus_message_append(m, "s", name);
3676 if (r < 0)
3677 return r;
3678
3679 r = sd_bus_call_async(
3680 bus,
3681 &u->get_name_owner_slot,
3682 m,
3683 get_name_owner_handler,
3684 u,
3685 timeout_usec);
3686 if (r < 0) {
3687 u->match_bus_slot = sd_bus_slot_unref(u->match_bus_slot);
3688 return r;
3689 }
3690
3691 log_unit_debug(u, "Watching D-Bus name '%s'.", name);
3692 return 0;
3693 }
3694
3695 int unit_watch_bus_name(Unit *u, const char *name) {
3696 int r;
3697
3698 assert(u);
3699 assert(name);
3700
3701 /* Watch a specific name on the bus. We only support one unit
3702 * watching each name for now. */
3703
3704 if (u->manager->api_bus) {
3705 /* If the bus is already available, install the match directly.
3706 * Otherwise, just put the name in the list. bus_setup_api() will take care later. */
3707 r = unit_install_bus_match(u, u->manager->api_bus, name);
3708 if (r < 0)
3709 return log_warning_errno(r, "Failed to subscribe to NameOwnerChanged signal for '%s': %m", name);
3710 }
3711
3712 r = hashmap_put(u->manager->watch_bus, name, u);
3713 if (r < 0) {
3714 u->match_bus_slot = sd_bus_slot_unref(u->match_bus_slot);
3715 u->get_name_owner_slot = sd_bus_slot_unref(u->get_name_owner_slot);
3716 return log_warning_errno(r, "Failed to put bus name to hashmap: %m");
3717 }
3718
3719 return 0;
3720 }
3721
3722 void unit_unwatch_bus_name(Unit *u, const char *name) {
3723 assert(u);
3724 assert(name);
3725
3726 (void) hashmap_remove_value(u->manager->watch_bus, name, u);
3727 u->match_bus_slot = sd_bus_slot_unref(u->match_bus_slot);
3728 u->get_name_owner_slot = sd_bus_slot_unref(u->get_name_owner_slot);
3729 }
3730
3731 int unit_add_node_dependency(Unit *u, const char *what, UnitDependency dep, UnitDependencyMask mask) {
3732 _cleanup_free_ char *e = NULL;
3733 Unit *device;
3734 int r;
3735
3736 assert(u);
3737
3738 /* Adds in links to the device node that this unit is based on */
3739 if (isempty(what))
3740 return 0;
3741
3742 if (!is_device_path(what))
3743 return 0;
3744
3745 /* When device units aren't supported (such as in a container), don't create dependencies on them. */
3746 if (!unit_type_supported(UNIT_DEVICE))
3747 return 0;
3748
3749 r = unit_name_from_path(what, ".device", &e);
3750 if (r < 0)
3751 return r;
3752
3753 r = manager_load_unit(u->manager, e, NULL, NULL, &device);
3754 if (r < 0)
3755 return r;
3756
3757 if (dep == UNIT_REQUIRES && device_shall_be_bound_by(device, u))
3758 dep = UNIT_BINDS_TO;
3759
3760 return unit_add_two_dependencies(u, UNIT_AFTER,
3761 MANAGER_IS_SYSTEM(u->manager) ? dep : UNIT_WANTS,
3762 device, true, mask);
3763 }
3764
3765 int unit_add_blockdev_dependency(Unit *u, const char *what, UnitDependencyMask mask) {
3766 _cleanup_free_ char *escaped = NULL, *target = NULL;
3767 int r;
3768
3769 assert(u);
3770
3771 if (isempty(what))
3772 return 0;
3773
3774 if (!path_startswith(what, "/dev/"))
3775 return 0;
3776
3777 /* If we don't support devices, then also don't bother with blockdev@.target */
3778 if (!unit_type_supported(UNIT_DEVICE))
3779 return 0;
3780
3781 r = unit_name_path_escape(what, &escaped);
3782 if (r < 0)
3783 return r;
3784
3785 r = unit_name_build("blockdev", escaped, ".target", &target);
3786 if (r < 0)
3787 return r;
3788
3789 return unit_add_dependency_by_name(u, UNIT_AFTER, target, true, mask);
3790 }
3791
3792 int unit_coldplug(Unit *u) {
3793 int r = 0;
3794
3795 assert(u);
3796
3797 /* Make sure we don't enter a loop, when coldplugging recursively. */
3798 if (u->coldplugged)
3799 return 0;
3800
3801 u->coldplugged = true;
3802
3803 STRV_FOREACH(i, u->deserialized_refs)
3804 RET_GATHER(r, bus_unit_track_add_name(u, *i));
3805
3806 u->deserialized_refs = strv_free(u->deserialized_refs);
3807
3808 if (UNIT_VTABLE(u)->coldplug)
3809 RET_GATHER(r, UNIT_VTABLE(u)->coldplug(u));
3810
3811 if (u->job)
3812 RET_GATHER(r, job_coldplug(u->job));
3813 if (u->nop_job)
3814 RET_GATHER(r, job_coldplug(u->nop_job));
3815
3816 unit_modify_nft_set(u, /* add = */ true);
3817 return r;
3818 }
3819
3820 void unit_catchup(Unit *u) {
3821 assert(u);
3822
3823 if (UNIT_VTABLE(u)->catchup)
3824 UNIT_VTABLE(u)->catchup(u);
3825
3826 unit_cgroup_catchup(u);
3827 }
3828
3829 static bool fragment_mtime_newer(const char *path, usec_t mtime, bool path_masked) {
3830 struct stat st;
3831
3832 if (!path)
3833 return false;
3834
3835 /* If the source is some virtual kernel file system, then we assume we watch it anyway, and hence pretend we
3836 * are never out-of-date. */
3837 if (PATH_STARTSWITH_SET(path, "/proc", "/sys"))
3838 return false;
3839
3840 if (stat(path, &st) < 0)
3841 /* What, cannot access this anymore? */
3842 return true;
3843
3844 if (path_masked)
3845 /* For masked files check if they are still so */
3846 return !null_or_empty(&st);
3847 else
3848 /* For non-empty files check the mtime */
3849 return timespec_load(&st.st_mtim) > mtime;
3850
3851 return false;
3852 }
3853
3854 bool unit_need_daemon_reload(Unit *u) {
3855 assert(u);
3856 assert(u->manager);
3857
3858 if (u->manager->unit_file_state_outdated)
3859 return true;
3860
3861 /* For unit files, we allow masking… */
3862 if (fragment_mtime_newer(u->fragment_path, u->fragment_mtime,
3863 u->load_state == UNIT_MASKED))
3864 return true;
3865
3866 /* Source paths should not be masked… */
3867 if (fragment_mtime_newer(u->source_path, u->source_mtime, false))
3868 return true;
3869
3870 if (u->load_state == UNIT_LOADED) {
3871 _cleanup_strv_free_ char **dropins = NULL;
3872
3873 (void) unit_find_dropin_paths(u, /* use_unit_path_cache = */ false, &dropins);
3874
3875 if (!strv_equal(u->dropin_paths, dropins))
3876 return true;
3877
3878 /* … any drop-ins that are masked are simply omitted from the list. */
3879 STRV_FOREACH(path, u->dropin_paths)
3880 if (fragment_mtime_newer(*path, u->dropin_mtime, false))
3881 return true;
3882 }
3883
3884 return false;
3885 }
3886
3887 void unit_reset_failed(Unit *u) {
3888 assert(u);
3889
3890 if (UNIT_VTABLE(u)->reset_failed)
3891 UNIT_VTABLE(u)->reset_failed(u);
3892
3893 ratelimit_reset(&u->start_ratelimit);
3894 u->start_limit_hit = false;
3895
3896 (void) unit_set_debug_invocation(u, /* enable= */ false);
3897 }
3898
3899 Unit *unit_following(Unit *u) {
3900 assert(u);
3901
3902 if (UNIT_VTABLE(u)->following)
3903 return UNIT_VTABLE(u)->following(u);
3904
3905 return NULL;
3906 }
3907
3908 bool unit_stop_pending(Unit *u) {
3909 assert(u);
3910
3911 /* This call does check the current state of the unit. It's
3912 * hence useful to be called from state change calls of the
3913 * unit itself, where the state isn't updated yet. This is
3914 * different from unit_inactive_or_pending() which checks both
3915 * the current state and for a queued job. */
3916
3917 return unit_has_job_type(u, JOB_STOP);
3918 }
3919
3920 bool unit_inactive_or_pending(Unit *u) {
3921 assert(u);
3922
3923 /* Returns true if the unit is inactive or going down */
3924
3925 if (UNIT_IS_INACTIVE_OR_DEACTIVATING(unit_active_state(u)))
3926 return true;
3927
3928 if (unit_stop_pending(u))
3929 return true;
3930
3931 return false;
3932 }
3933
3934 bool unit_active_or_pending(Unit *u) {
3935 assert(u);
3936
3937 /* Returns true if the unit is active or going up */
3938
3939 if (UNIT_IS_ACTIVE_OR_ACTIVATING(unit_active_state(u)))
3940 return true;
3941
3942 if (u->job &&
3943 IN_SET(u->job->type, JOB_START, JOB_RELOAD_OR_START, JOB_RESTART))
3944 return true;
3945
3946 return false;
3947 }
3948
3949 bool unit_will_restart_default(Unit *u) {
3950 assert(u);
3951
3952 return unit_has_job_type(u, JOB_START);
3953 }
3954
3955 bool unit_will_restart(Unit *u) {
3956 assert(u);
3957
3958 if (!UNIT_VTABLE(u)->will_restart)
3959 return false;
3960
3961 return UNIT_VTABLE(u)->will_restart(u);
3962 }
3963
3964 void unit_notify_cgroup_oom(Unit *u, bool managed_oom) {
3965 assert(u);
3966
3967 if (UNIT_VTABLE(u)->notify_cgroup_oom)
3968 UNIT_VTABLE(u)->notify_cgroup_oom(u, managed_oom);
3969 }
3970
3971 static int unit_pid_set(Unit *u, Set **pid_set) {
3972 int r;
3973
3974 assert(u);
3975 assert(pid_set);
3976
3977 set_clear(*pid_set); /* This updates input. */
3978
3979 /* Exclude the main/control pids from being killed via the cgroup */
3980
3981 PidRef *pid;
3982 FOREACH_ARGUMENT(pid, unit_main_pid(u), unit_control_pid(u))
3983 if (pidref_is_set(pid)) {
3984 r = set_ensure_put(pid_set, NULL, PID_TO_PTR(pid->pid));
3985 if (r < 0)
3986 return r;
3987 }
3988
3989 return 0;
3990 }
3991
3992 static int kill_common_log(const PidRef *pid, int signo, void *userdata) {
3993 _cleanup_free_ char *comm = NULL;
3994 Unit *u = ASSERT_PTR(userdata);
3995
3996 (void) pidref_get_comm(pid, &comm);
3997
3998 log_unit_info(u, "Sending signal SIG%s to process " PID_FMT " (%s) on client request.",
3999 signal_to_string(signo), pid->pid, strna(comm));
4000
4001 return 1;
4002 }
4003
4004 static int kill_or_sigqueue(PidRef *pidref, int signo, int code, int value) {
4005 assert(pidref_is_set(pidref));
4006 assert(SIGNAL_VALID(signo));
4007
4008 switch (code) {
4009
4010 case SI_USER:
4011 log_debug("Killing " PID_FMT " with signal SIG%s.", pidref->pid, signal_to_string(signo));
4012 return pidref_kill(pidref, signo);
4013
4014 case SI_QUEUE:
4015 log_debug("Enqueuing value %i to " PID_FMT " on signal SIG%s.", value, pidref->pid, signal_to_string(signo));
4016 return pidref_sigqueue(pidref, signo, value);
4017
4018 default:
4019 assert_not_reached();
4020 }
4021 }
4022
4023 static int unit_kill_one(
4024 Unit *u,
4025 PidRef *pidref,
4026 const char *type,
4027 int signo,
4028 int code,
4029 int value,
4030 sd_bus_error *ret_error) {
4031
4032 int r;
4033
4034 assert(u);
4035 assert(type);
4036
4037 if (!pidref_is_set(pidref))
4038 return 0;
4039
4040 _cleanup_free_ char *comm = NULL;
4041 (void) pidref_get_comm(pidref, &comm);
4042
4043 r = kill_or_sigqueue(pidref, signo, code, value);
4044 if (r == -ESRCH)
4045 return 0;
4046 if (r < 0) {
4047 /* Report this failure both to the logs and to the client */
4048 if (ret_error)
4049 sd_bus_error_set_errnof(
4050 ret_error, r,
4051 "Failed to send signal SIG%s to %s process " PID_FMT " (%s): %m",
4052 signal_to_string(signo), type, pidref->pid, strna(comm));
4053
4054 return log_unit_warning_errno(
4055 u, r,
4056 "Failed to send signal SIG%s to %s process " PID_FMT " (%s) on client request: %m",
4057 signal_to_string(signo), type, pidref->pid, strna(comm));
4058 }
4059
4060 log_unit_info(u, "Sent signal SIG%s to %s process " PID_FMT " (%s) on client request.",
4061 signal_to_string(signo), type, pidref->pid, strna(comm));
4062 return 1; /* killed */
4063 }
4064
4065 int unit_kill(
4066 Unit *u,
4067 KillWhom whom,
4068 const char *subgroup,
4069 int signo,
4070 int code,
4071 int value,
4072 sd_bus_error *ret_error) {
4073
4074 PidRef *main_pid, *control_pid;
4075 bool killed = false;
4076 int ret = 0, r;
4077
4078 /* This is the common implementation for explicit user-requested killing of unit processes, shared by
4079 * various unit types. Do not confuse with unit_kill_context(), which is what we use when we want to
4080 * stop a service ourselves. */
4081
4082 assert(u);
4083 assert(whom >= 0);
4084 assert(whom < _KILL_WHOM_MAX);
4085 assert(SIGNAL_VALID(signo));
4086 assert(IN_SET(code, SI_USER, SI_QUEUE));
4087
4088 if (subgroup) {
4089 if (!IN_SET(whom, KILL_CGROUP, KILL_CGROUP_FAIL))
4090 return sd_bus_error_set(ret_error, SD_BUS_ERROR_NOT_SUPPORTED,
4091 "Killing by subgroup is only supported for 'cgroup' or 'cgroup-kill' modes.");
4092
4093 if (!unit_cgroup_delegate(u))
4094 return sd_bus_error_set(ret_error, SD_BUS_ERROR_NOT_SUPPORTED,
4095 "Killing by subgroup is only available for units with control group delegation enabled.");
4096 }
4097
4098 main_pid = unit_main_pid(u);
4099 control_pid = unit_control_pid(u);
4100
4101 if (!UNIT_HAS_CGROUP_CONTEXT(u) && !main_pid && !control_pid)
4102 return sd_bus_error_set(ret_error, SD_BUS_ERROR_NOT_SUPPORTED, "Unit type does not support process killing.");
4103
4104 if (IN_SET(whom, KILL_MAIN, KILL_MAIN_FAIL)) {
4105 if (!main_pid)
4106 return sd_bus_error_setf(ret_error, BUS_ERROR_NO_SUCH_PROCESS, "%s units have no main processes", unit_type_to_string(u->type));
4107 if (!pidref_is_set(main_pid))
4108 return sd_bus_error_set_const(ret_error, BUS_ERROR_NO_SUCH_PROCESS, "No main process to kill");
4109 }
4110
4111 if (IN_SET(whom, KILL_CONTROL, KILL_CONTROL_FAIL)) {
4112 if (!control_pid)
4113 return sd_bus_error_setf(ret_error, BUS_ERROR_NO_SUCH_PROCESS, "%s units have no control processes", unit_type_to_string(u->type));
4114 if (!pidref_is_set(control_pid))
4115 return sd_bus_error_set_const(ret_error, BUS_ERROR_NO_SUCH_PROCESS, "No control process to kill");
4116 }
4117
4118 if (IN_SET(whom, KILL_CONTROL, KILL_CONTROL_FAIL, KILL_ALL, KILL_ALL_FAIL)) {
4119 r = unit_kill_one(u, control_pid, "control", signo, code, value, ret_error);
4120 RET_GATHER(ret, r);
4121 killed = killed || r > 0;
4122 }
4123
4124 if (IN_SET(whom, KILL_MAIN, KILL_MAIN_FAIL, KILL_ALL, KILL_ALL_FAIL)) {
4125 r = unit_kill_one(u, main_pid, "main", signo, code, value, ret >= 0 ? ret_error : NULL);
4126 RET_GATHER(ret, r);
4127 killed = killed || r > 0;
4128 }
4129
4130 /* Note: if we shall enqueue rather than kill we won't do this via the cgroup mechanism, since it
4131 * doesn't really make much sense (and given that enqueued values are a relatively expensive
4132 * resource, and we shouldn't allow us to be subjects for such allocation sprees) */
4133 if (IN_SET(whom, KILL_ALL, KILL_ALL_FAIL, KILL_CGROUP, KILL_CGROUP_FAIL) && code == SI_USER) {
4134 CGroupRuntime *crt = unit_get_cgroup_runtime(u);
4135 if (crt && crt->cgroup_path) {
4136 _cleanup_set_free_ Set *pid_set = NULL;
4137 _cleanup_free_ char *joined = NULL;
4138 const char *p;
4139
4140 if (empty_or_root(subgroup))
4141 p = crt->cgroup_path;
4142 else {
4143 joined = path_join(crt->cgroup_path, subgroup);
4144 if (!joined)
4145 return -ENOMEM;
4146
4147 p = joined;
4148 }
4149
4150 if (signo == SIGKILL) {
4151 r = cg_kill_kernel_sigkill(p);
4152 if (r >= 0) {
4153 killed = true;
4154 log_unit_info(u, "Killed unit cgroup '%s' with SIGKILL on client request.", p);
4155 goto finish;
4156 }
4157 if (r != -EOPNOTSUPP) {
4158 if (ret >= 0)
4159 sd_bus_error_set_errnof(ret_error, r,
4160 "Failed to kill unit cgroup: %m");
4161 RET_GATHER(ret, log_unit_warning_errno(u, r, "Failed to kill unit cgroup '%s': %m", p));
4162 goto finish;
4163 }
4164 /* Fall back to manual enumeration */
4165 } else if (IN_SET(whom, KILL_ALL, KILL_ALL_FAIL)) {
4166 /* Exclude the main/control pids from being killed via the cgroup if not
4167 * SIGKILL */
4168 r = unit_pid_set(u, &pid_set);
4169 if (r < 0)
4170 return log_oom();
4171 }
4172
4173 r = cg_kill_recursive(p, signo, /* flags= */ 0, pid_set, kill_common_log, u);
4174 if (r < 0 && !IN_SET(r, -ESRCH, -ENOENT)) {
4175 if (ret >= 0)
4176 sd_bus_error_set_errnof(
4177 ret_error, r,
4178 "Failed to send signal SIG%s to processes in unit cgroup '%s': %m",
4179 signal_to_string(signo), p);
4180
4181 RET_GATHER(ret, log_unit_warning_errno(
4182 u, r,
4183 "Failed to send signal SIG%s to processes in unit cgroup '%s' on client request: %m",
4184 signal_to_string(signo), p));
4185 }
4186 killed = killed || r > 0;
4187 }
4188 }
4189
4190 finish:
4191 /* If the "fail" versions of the operation are requested, then complain if the set of processes we killed is empty */
4192 if (ret >= 0 && !killed && IN_SET(whom, KILL_ALL_FAIL, KILL_CONTROL_FAIL, KILL_MAIN_FAIL, KILL_CGROUP_FAIL))
4193 return sd_bus_error_set_const(ret_error, BUS_ERROR_NO_SUCH_PROCESS, "No matching processes to kill");
4194
4195 return ret;
4196 }
4197
4198 int unit_following_set(Unit *u, Set **s) {
4199 assert(u);
4200 assert(s);
4201
4202 if (UNIT_VTABLE(u)->following_set)
4203 return UNIT_VTABLE(u)->following_set(u, s);
4204
4205 *s = NULL;
4206 return 0;
4207 }
4208
4209 UnitFileState unit_get_unit_file_state(Unit *u) {
4210 int r;
4211
4212 assert(u);
4213
4214 if (u->unit_file_state >= 0 || !u->fragment_path)
4215 return u->unit_file_state;
4216
4217 /* If we know this is a transient unit no need to ask the unit file state for details. Let's bypass
4218 * the more expensive on-disk check. */
4219 if (u->transient)
4220 return (u->unit_file_state = UNIT_FILE_TRANSIENT);
4221
4222 r = unit_file_get_state(
4223 u->manager->runtime_scope,
4224 /* root_dir= */ NULL,
4225 u->id,
4226 &u->unit_file_state);
4227 if (r < 0)
4228 u->unit_file_state = UNIT_FILE_BAD;
4229
4230 return u->unit_file_state;
4231 }
4232
4233 PresetAction unit_get_unit_file_preset(Unit *u) {
4234 int r;
4235
4236 assert(u);
4237
4238 if (u->unit_file_preset >= 0)
4239 return u->unit_file_preset;
4240
4241 /* If this is a transient or perpetual unit file it doesn't make much sense to ask the preset
4242 * database about this, because enabling/disabling makes no sense for either. Hence don't. */
4243 if (!u->fragment_path || u->transient || u->perpetual)
4244 return (u->unit_file_preset = -ENOEXEC);
4245
4246 _cleanup_free_ char *bn = NULL;
4247 r = path_extract_filename(u->fragment_path, &bn);
4248 if (r < 0)
4249 return (u->unit_file_preset = r);
4250 if (r == O_DIRECTORY)
4251 return (u->unit_file_preset = -EISDIR);
4252
4253 return (u->unit_file_preset = unit_file_query_preset(
4254 u->manager->runtime_scope,
4255 /* root_dir= */ NULL,
4256 bn,
4257 /* cached= */ NULL));
4258 }
4259
4260 Unit* unit_ref_set(UnitRef *ref, Unit *source, Unit *target) {
4261 assert(ref);
4262 assert(source);
4263 assert(target);
4264
4265 if (ref->target)
4266 unit_ref_unset(ref);
4267
4268 ref->source = source;
4269 ref->target = target;
4270 LIST_PREPEND(refs_by_target, target->refs_by_target, ref);
4271 return target;
4272 }
4273
4274 void unit_ref_unset(UnitRef *ref) {
4275 assert(ref);
4276
4277 if (!ref->target)
4278 return;
4279
4280 /* We are about to drop a reference to the unit, make sure the garbage collection has a look at it as it might
4281 * be unreferenced now. */
4282 unit_add_to_gc_queue(ref->target);
4283
4284 LIST_REMOVE(refs_by_target, ref->target->refs_by_target, ref);
4285 ref->source = ref->target = NULL;
4286 }
4287
4288 static int user_from_unit_name(Unit *u, char **ret) {
4289
4290 static const uint8_t hash_key[] = {
4291 0x58, 0x1a, 0xaf, 0xe6, 0x28, 0x58, 0x4e, 0x96,
4292 0xb4, 0x4e, 0xf5, 0x3b, 0x8c, 0x92, 0x07, 0xec
4293 };
4294
4295 _cleanup_free_ char *n = NULL;
4296 int r;
4297
4298 r = unit_name_to_prefix(u->id, &n);
4299 if (r < 0)
4300 return r;
4301
4302 if (valid_user_group_name(n, 0)) {
4303 *ret = TAKE_PTR(n);
4304 return 0;
4305 }
4306
4307 /* If we can't use the unit name as a user name, then let's hash it and use that */
4308 if (asprintf(ret, "_du%016" PRIx64, siphash24(n, strlen(n), hash_key)) < 0)
4309 return -ENOMEM;
4310
4311 return 0;
4312 }
4313
4314 static int unit_verify_contexts(const Unit *u) {
4315 assert(u);
4316
4317 const ExecContext *ec = unit_get_exec_context(u);
4318 if (!ec)
4319 return 0;
4320
4321 if (MANAGER_IS_USER(u->manager) && ec->dynamic_user)
4322 return log_unit_error_errno(u, SYNTHETIC_ERRNO(ENOEXEC), "DynamicUser= enabled for user unit, which is not supported. Refusing.");
4323
4324 if (ec->dynamic_user && ec->working_directory_home)
4325 return log_unit_error_errno(u, SYNTHETIC_ERRNO(ENOEXEC), "WorkingDirectory=~ is not allowed under DynamicUser=yes. Refusing.");
4326
4327 if (ec->working_directory && path_below_api_vfs(ec->working_directory) &&
4328 exec_needs_mount_namespace(ec, /* params = */ NULL, /* runtime = */ NULL))
4329 return log_unit_error_errno(u, SYNTHETIC_ERRNO(ENOEXEC), "WorkingDirectory= may not be below /proc/, /sys/ or /dev/ when using mount namespacing. Refusing.");
4330
4331 if (exec_needs_pid_namespace(ec, /* params= */ NULL) && !UNIT_VTABLE(u)->notify_pidref)
4332 return log_unit_error_errno(u, SYNTHETIC_ERRNO(ENOEXEC), "PrivatePIDs= setting is only supported for service units. Refusing.");
4333
4334 const KillContext *kc = unit_get_kill_context(u);
4335
4336 if (ec->pam_name && kc && !IN_SET(kc->kill_mode, KILL_CONTROL_GROUP, KILL_MIXED))
4337 return log_unit_error_errno(u, SYNTHETIC_ERRNO(ENOEXEC), "Unit has PAM enabled. Kill mode must be set to 'control-group' or 'mixed'. Refusing.");
4338
4339 return 0;
4340 }
4341
4342 static PrivateTmp unit_get_private_var_tmp(const Unit *u, const ExecContext *c) {
4343 assert(u);
4344 assert(c);
4345 assert(c->private_tmp >= 0 && c->private_tmp < _PRIVATE_TMP_MAX);
4346
4347 /* Disable disconnected private tmpfs on /var/tmp/ when DefaultDependencies=no and
4348 * RootImage=/RootDirectory= are not set, as /var/ may be a separated partition.
4349 * See issue #37258. */
4350
4351 /* PrivateTmp=yes/no also enables/disables private tmpfs on /var/tmp/. */
4352 if (c->private_tmp != PRIVATE_TMP_DISCONNECTED)
4353 return c->private_tmp;
4354
4355 /* When DefaultDependencies=yes, disconnected tmpfs is also enabled on /var/tmp/, and an explicit
4356 * dependency to the mount on /var/ will be added in unit_add_exec_dependencies(). */
4357 if (u->default_dependencies)
4358 return PRIVATE_TMP_DISCONNECTED;
4359
4360 /* When RootImage=/RootDirectory= is enabled, /var/ should be prepared by the image or directory,
4361 * hence we can mount a disconnected tmpfs on /var/tmp/. */
4362 if (exec_context_with_rootfs(c))
4363 return PRIVATE_TMP_DISCONNECTED;
4364
4365 /* Even if DefaultDependencies=no, enable disconnected tmpfs when
4366 * RequiresMountsFor=/WantsMountsFor=/var/ is explicitly set. */
4367 for (UnitMountDependencyType t = 0; t < _UNIT_MOUNT_DEPENDENCY_TYPE_MAX; t++)
4368 if (hashmap_contains(u->mounts_for[t], "/var/"))
4369 return PRIVATE_TMP_DISCONNECTED;
4370
4371 /* Check the same but for After= with Requires=/Requisite=/Wants= or friends. */
4372 Unit *m = manager_get_unit(u->manager, "var.mount");
4373 if (!m)
4374 return PRIVATE_TMP_NO;
4375
4376 if (!unit_has_dependency(u, UNIT_ATOM_AFTER, m))
4377 return PRIVATE_TMP_NO;
4378
4379 if (unit_has_dependency(u, UNIT_ATOM_PULL_IN_START, m) ||
4380 unit_has_dependency(u, UNIT_ATOM_PULL_IN_VERIFY, m) ||
4381 unit_has_dependency(u, UNIT_ATOM_PULL_IN_START_IGNORED, m))
4382 return PRIVATE_TMP_DISCONNECTED;
4383
4384 return PRIVATE_TMP_NO;
4385 }
4386
4387 int unit_patch_contexts(Unit *u) {
4388 CGroupContext *cc;
4389 ExecContext *ec;
4390 int r;
4391
4392 assert(u);
4393
4394 /* Patch in the manager defaults into the exec and cgroup
4395 * contexts, _after_ the rest of the settings have been
4396 * initialized */
4397
4398 ec = unit_get_exec_context(u);
4399 if (ec) {
4400 /* This only copies in the ones that need memory */
4401 for (unsigned i = 0; i < _RLIMIT_MAX; i++)
4402 if (u->manager->defaults.rlimit[i] && !ec->rlimit[i]) {
4403 ec->rlimit[i] = newdup(struct rlimit, u->manager->defaults.rlimit[i], 1);
4404 if (!ec->rlimit[i])
4405 return -ENOMEM;
4406 }
4407
4408 if (MANAGER_IS_USER(u->manager) && !ec->working_directory) {
4409 r = get_home_dir(&ec->working_directory);
4410 if (r < 0)
4411 return r;
4412
4413 if (!ec->working_directory_home)
4414 /* If home directory is implied by us, allow it to be missing. */
4415 ec->working_directory_missing_ok = true;
4416 }
4417
4418 if (ec->private_devices)
4419 ec->capability_bounding_set &= ~((UINT64_C(1) << CAP_MKNOD) | (UINT64_C(1) << CAP_SYS_RAWIO));
4420
4421 if (ec->protect_kernel_modules)
4422 ec->capability_bounding_set &= ~(UINT64_C(1) << CAP_SYS_MODULE);
4423
4424 if (ec->protect_kernel_logs)
4425 ec->capability_bounding_set &= ~(UINT64_C(1) << CAP_SYSLOG);
4426
4427 if (ec->protect_clock)
4428 ec->capability_bounding_set &= ~((UINT64_C(1) << CAP_SYS_TIME) | (UINT64_C(1) << CAP_WAKE_ALARM));
4429
4430 if (ec->dynamic_user) {
4431 if (!ec->user) {
4432 r = user_from_unit_name(u, &ec->user);
4433 if (r < 0)
4434 return r;
4435 }
4436
4437 if (!ec->group) {
4438 ec->group = strdup(ec->user);
4439 if (!ec->group)
4440 return -ENOMEM;
4441 }
4442
4443 /* If the dynamic user option is on, let's make sure that the unit can't leave its
4444 * UID/GID around in the file system or on IPC objects. Hence enforce a strict
4445 * sandbox. */
4446
4447 /* With DynamicUser= we want private directories, so if the user hasn't manually
4448 * selected PrivateTmp=, enable it, but to a fully private (disconnected) tmpfs
4449 * instance. */
4450 if (ec->private_tmp == PRIVATE_TMP_NO)
4451 ec->private_tmp = PRIVATE_TMP_DISCONNECTED;
4452 ec->remove_ipc = true;
4453 ec->protect_system = PROTECT_SYSTEM_STRICT;
4454 if (ec->protect_home == PROTECT_HOME_NO)
4455 ec->protect_home = PROTECT_HOME_READ_ONLY;
4456
4457 /* Make sure this service can neither benefit from SUID/SGID binaries nor create
4458 * them. */
4459 ec->no_new_privileges = true;
4460 ec->restrict_suid_sgid = true;
4461 }
4462
4463 ec->private_var_tmp = unit_get_private_var_tmp(u, ec);
4464
4465 FOREACH_ARRAY(d, ec->directories, _EXEC_DIRECTORY_TYPE_MAX)
4466 exec_directory_sort(d);
4467 }
4468
4469 cc = unit_get_cgroup_context(u);
4470 if (cc && ec) {
4471
4472 if (ec->private_devices &&
4473 cc->device_policy == CGROUP_DEVICE_POLICY_AUTO)
4474 cc->device_policy = CGROUP_DEVICE_POLICY_CLOSED;
4475
4476 /* Only add these if needed, as they imply that everything else is blocked. */
4477 if (cgroup_context_has_device_policy(cc)) {
4478 if (ec->root_image || ec->mount_images) {
4479
4480 /* When RootImage= or MountImages= is specified, the following devices are touched. */
4481 FOREACH_STRING(p, "/dev/loop-control", "/dev/mapper/control") {
4482 r = cgroup_context_add_device_allow(cc, p, CGROUP_DEVICE_READ|CGROUP_DEVICE_WRITE);
4483 if (r < 0)
4484 return r;
4485 }
4486 FOREACH_STRING(p, "block-loop", "block-blkext", "block-device-mapper") {
4487 r = cgroup_context_add_device_allow(cc, p, CGROUP_DEVICE_READ|CGROUP_DEVICE_WRITE|CGROUP_DEVICE_MKNOD);
4488 if (r < 0)
4489 return r;
4490 }
4491
4492 /* Make sure "block-loop" can be resolved, i.e. make sure "loop" shows up in /proc/devices.
4493 * Same for mapper and verity. */
4494 FOREACH_STRING(p, "modprobe@loop.service", "modprobe@dm_mod.service", "modprobe@dm_verity.service") {
4495 r = unit_add_two_dependencies_by_name(u, UNIT_AFTER, UNIT_WANTS, p, true, UNIT_DEPENDENCY_FILE);
4496 if (r < 0)
4497 return r;
4498 }
4499 }
4500
4501 if (ec->protect_clock) {
4502 r = cgroup_context_add_device_allow(cc, "char-rtc", CGROUP_DEVICE_READ);
4503 if (r < 0)
4504 return r;
4505 }
4506 }
4507 }
4508
4509 return unit_verify_contexts(u);
4510 }
4511
4512 ExecContext *unit_get_exec_context(const Unit *u) {
4513 size_t offset;
4514 assert(u);
4515
4516 if (u->type < 0)
4517 return NULL;
4518
4519 offset = UNIT_VTABLE(u)->exec_context_offset;
4520 if (offset <= 0)
4521 return NULL;
4522
4523 return (ExecContext*) ((uint8_t*) u + offset);
4524 }
4525
4526 KillContext *unit_get_kill_context(const Unit *u) {
4527 size_t offset;
4528 assert(u);
4529
4530 if (u->type < 0)
4531 return NULL;
4532
4533 offset = UNIT_VTABLE(u)->kill_context_offset;
4534 if (offset <= 0)
4535 return NULL;
4536
4537 return (KillContext*) ((uint8_t*) u + offset);
4538 }
4539
4540 CGroupContext *unit_get_cgroup_context(const Unit *u) {
4541 size_t offset;
4542
4543 if (u->type < 0)
4544 return NULL;
4545
4546 offset = UNIT_VTABLE(u)->cgroup_context_offset;
4547 if (offset <= 0)
4548 return NULL;
4549
4550 return (CGroupContext*) ((uint8_t*) u + offset);
4551 }
4552
4553 ExecRuntime *unit_get_exec_runtime(const Unit *u) {
4554 size_t offset;
4555
4556 if (u->type < 0)
4557 return NULL;
4558
4559 offset = UNIT_VTABLE(u)->exec_runtime_offset;
4560 if (offset <= 0)
4561 return NULL;
4562
4563 return *(ExecRuntime**) ((uint8_t*) u + offset);
4564 }
4565
4566 CGroupRuntime *unit_get_cgroup_runtime(const Unit *u) {
4567 size_t offset;
4568
4569 if (u->type < 0)
4570 return NULL;
4571
4572 offset = UNIT_VTABLE(u)->cgroup_runtime_offset;
4573 if (offset <= 0)
4574 return NULL;
4575
4576 return *(CGroupRuntime**) ((uint8_t*) u + offset);
4577 }
4578
4579 static const char* unit_drop_in_dir(Unit *u, UnitWriteFlags flags) {
4580 assert(u);
4581
4582 if (UNIT_WRITE_FLAGS_NOOP(flags))
4583 return NULL;
4584
4585 if (u->transient) /* Redirect drop-ins for transient units always into the transient directory. */
4586 return u->manager->lookup_paths.transient;
4587
4588 if (flags & UNIT_PERSISTENT)
4589 return u->manager->lookup_paths.persistent_control;
4590
4591 if (flags & UNIT_RUNTIME)
4592 return u->manager->lookup_paths.runtime_control;
4593
4594 return NULL;
4595 }
4596
4597 const char* unit_escape_setting(const char *s, UnitWriteFlags flags, char **buf) {
4598 assert(s);
4599 assert(popcount(flags & (UNIT_ESCAPE_EXEC_SYNTAX_ENV | UNIT_ESCAPE_EXEC_SYNTAX | UNIT_ESCAPE_C)) <= 1);
4600 assert(buf);
4601
4602 _cleanup_free_ char *t = NULL;
4603
4604 /* Returns a string with any escaping done. If no escaping was necessary, *buf is set to NULL, and
4605 * the input pointer is returned as-is. If an allocation was needed, the return buffer pointer is
4606 * written to *buf. This means the return value always contains a properly escaped version, but *buf
4607 * only contains a pointer if an allocation was made. Callers can use this to optimize memory
4608 * allocations. */
4609
4610 if (flags & UNIT_ESCAPE_SPECIFIERS) {
4611 t = specifier_escape(s);
4612 if (!t)
4613 return NULL;
4614
4615 s = t;
4616 }
4617
4618 /* We either do C-escaping or shell-escaping, to additionally escape characters that we parse for
4619 * ExecStart= and friends, i.e. '$' and quotes. */
4620
4621 if (flags & (UNIT_ESCAPE_EXEC_SYNTAX_ENV | UNIT_ESCAPE_EXEC_SYNTAX)) {
4622 char *t2;
4623
4624 if (flags & UNIT_ESCAPE_EXEC_SYNTAX_ENV) {
4625 t2 = strreplace(s, "$", "$$");
4626 if (!t2)
4627 return NULL;
4628 free_and_replace(t, t2);
4629 }
4630
4631 t2 = shell_escape(t ?: s, "\"");
4632 if (!t2)
4633 return NULL;
4634 free_and_replace(t, t2);
4635
4636 s = t;
4637
4638 } else if (flags & UNIT_ESCAPE_C) {
4639 char *t2;
4640
4641 t2 = cescape(s);
4642 if (!t2)
4643 return NULL;
4644 free_and_replace(t, t2);
4645
4646 s = t;
4647 }
4648
4649 *buf = TAKE_PTR(t);
4650 return s;
4651 }
4652
4653 char* unit_concat_strv(char **l, UnitWriteFlags flags) {
4654 _cleanup_free_ char *result = NULL;
4655 size_t n = 0;
4656
4657 /* Takes a list of strings, escapes them, and concatenates them. This may be used to format command
4658 * lines in a way suitable for ExecStart= stanzas. */
4659
4660 STRV_FOREACH(i, l) {
4661 _cleanup_free_ char *buf = NULL;
4662 const char *p;
4663 size_t a;
4664 char *q;
4665
4666 p = unit_escape_setting(*i, flags, &buf);
4667 if (!p)
4668 return NULL;
4669
4670 a = (n > 0) + 1 + strlen(p) + 1; /* separating space + " + entry + " */
4671 if (!GREEDY_REALLOC(result, n + a + 1))
4672 return NULL;
4673
4674 q = result + n;
4675 if (n > 0)
4676 *(q++) = ' ';
4677
4678 *(q++) = '"';
4679 q = stpcpy(q, p);
4680 *(q++) = '"';
4681
4682 n += a;
4683 }
4684
4685 if (!GREEDY_REALLOC(result, n + 1))
4686 return NULL;
4687
4688 result[n] = 0;
4689
4690 return TAKE_PTR(result);
4691 }
4692
4693 int unit_write_setting(Unit *u, UnitWriteFlags flags, const char *name, const char *data) {
4694 _cleanup_free_ char *p = NULL, *q = NULL, *escaped = NULL;
4695 const char *dir, *wrapped;
4696 int r;
4697
4698 assert(u);
4699 assert(name);
4700 assert(data);
4701
4702 if (UNIT_WRITE_FLAGS_NOOP(flags))
4703 return 0;
4704
4705 data = unit_escape_setting(data, flags, &escaped);
4706 if (!data)
4707 return -ENOMEM;
4708
4709 /* Prefix the section header. If we are writing this out as transient file, then let's suppress this if the
4710 * previous section header is the same */
4711
4712 if (flags & UNIT_PRIVATE) {
4713 if (!UNIT_VTABLE(u)->private_section)
4714 return -EINVAL;
4715
4716 if (!u->transient_file || u->last_section_private < 0)
4717 data = strjoina("[", UNIT_VTABLE(u)->private_section, "]\n", data);
4718 else if (u->last_section_private == 0)
4719 data = strjoina("\n[", UNIT_VTABLE(u)->private_section, "]\n", data);
4720 } else {
4721 if (!u->transient_file || u->last_section_private < 0)
4722 data = strjoina("[Unit]\n", data);
4723 else if (u->last_section_private > 0)
4724 data = strjoina("\n[Unit]\n", data);
4725 }
4726
4727 if (u->transient_file) {
4728 /* When this is a transient unit file in creation, then let's not create a new drop-in,
4729 * but instead write to the transient unit file. */
4730 fputs_with_newline(u->transient_file, data);
4731
4732 /* Remember which section we wrote this entry to */
4733 u->last_section_private = !!(flags & UNIT_PRIVATE);
4734 return 0;
4735 }
4736
4737 dir = unit_drop_in_dir(u, flags);
4738 if (!dir)
4739 return -EINVAL;
4740
4741 wrapped = strjoina("# This is a drop-in unit file extension, created via \"systemctl set-property\"\n"
4742 "# or an equivalent operation. Do not edit.\n",
4743 data,
4744 "\n");
4745
4746 r = drop_in_file(dir, u->id, 50, name, &p, &q);
4747 if (r < 0)
4748 return r;
4749
4750 (void) mkdir_p_label(p, 0755);
4751
4752 /* Make sure the drop-in dir is registered in our path cache. This way we don't need to stupidly
4753 * recreate the cache after every drop-in we write. */
4754 if (u->manager->unit_path_cache) {
4755 r = set_put_strdup_full(&u->manager->unit_path_cache, &path_hash_ops_free, p);
4756 if (r < 0)
4757 return r;
4758 }
4759
4760 r = write_string_file(q, wrapped, WRITE_STRING_FILE_CREATE|WRITE_STRING_FILE_ATOMIC|WRITE_STRING_FILE_LABEL);
4761 if (r < 0)
4762 return r;
4763
4764 r = strv_push(&u->dropin_paths, q);
4765 if (r < 0)
4766 return r;
4767 q = NULL;
4768
4769 strv_uniq(u->dropin_paths);
4770
4771 u->dropin_mtime = now(CLOCK_REALTIME);
4772
4773 return 0;
4774 }
4775
4776 int unit_write_settingf(Unit *u, UnitWriteFlags flags, const char *name, const char *format, ...) {
4777 _cleanup_free_ char *p = NULL;
4778 va_list ap;
4779 int r;
4780
4781 assert(u);
4782 assert(name);
4783 assert(format);
4784
4785 if (UNIT_WRITE_FLAGS_NOOP(flags))
4786 return 0;
4787
4788 va_start(ap, format);
4789 r = vasprintf(&p, format, ap);
4790 va_end(ap);
4791
4792 if (r < 0)
4793 return -ENOMEM;
4794
4795 return unit_write_setting(u, flags, name, p);
4796 }
4797
4798 int unit_make_transient(Unit *u) {
4799 _cleanup_free_ char *path = NULL;
4800 FILE *f;
4801
4802 assert(u);
4803
4804 if (!UNIT_VTABLE(u)->can_transient)
4805 return -EOPNOTSUPP;
4806
4807 (void) mkdir_p_label(u->manager->lookup_paths.transient, 0755);
4808
4809 path = path_join(u->manager->lookup_paths.transient, u->id);
4810 if (!path)
4811 return -ENOMEM;
4812
4813 /* Let's open the file we'll write the transient settings into. This file is kept open as long as we are
4814 * creating the transient, and is closed in unit_load(), as soon as we start loading the file. */
4815
4816 WITH_UMASK(0022) {
4817 f = fopen(path, "we");
4818 if (!f)
4819 return -errno;
4820 }
4821
4822 safe_fclose(u->transient_file);
4823 u->transient_file = f;
4824
4825 free_and_replace(u->fragment_path, path);
4826
4827 u->source_path = mfree(u->source_path);
4828 u->dropin_paths = strv_free(u->dropin_paths);
4829 u->fragment_mtime = u->source_mtime = u->dropin_mtime = 0;
4830
4831 u->load_state = UNIT_STUB;
4832 u->load_error = 0;
4833 u->transient = true;
4834
4835 unit_add_to_dbus_queue(u);
4836 unit_add_to_gc_queue(u);
4837
4838 fputs("# This is a transient unit file, created programmatically via the systemd API. Do not edit.\n",
4839 u->transient_file);
4840
4841 return 0;
4842 }
4843
4844 static bool ignore_leftover_process(const char *comm) {
4845 return comm && comm[0] == '('; /* Most likely our own helper process (PAM?), ignore */
4846 }
4847
4848 static int log_kill(const PidRef *pid, int sig, void *userdata) {
4849 const Unit *u = ASSERT_PTR(userdata);
4850 _cleanup_free_ char *comm = NULL;
4851
4852 assert(pidref_is_set(pid));
4853
4854 (void) pidref_get_comm(pid, &comm);
4855
4856 if (ignore_leftover_process(comm))
4857 /* Although we didn't log anything, as this callback is used in unit_kill_context we must return 1
4858 * here to let the manager know that a process was killed. */
4859 return 1;
4860
4861 log_unit_notice(u,
4862 "Killing process " PID_FMT " (%s) with signal SIG%s.",
4863 pid->pid,
4864 strna(comm),
4865 signal_to_string(sig));
4866
4867 return 1;
4868 }
4869
4870 static int operation_to_signal(
4871 const KillContext *c,
4872 KillOperation k,
4873 bool *ret_noteworthy) {
4874
4875 assert(c);
4876 assert(ret_noteworthy);
4877
4878 switch (k) {
4879
4880 case KILL_TERMINATE:
4881 case KILL_TERMINATE_AND_LOG:
4882 *ret_noteworthy = k == KILL_TERMINATE_AND_LOG;
4883 return c->kill_signal;
4884
4885 case KILL_RESTART:
4886 *ret_noteworthy = false;
4887 return restart_kill_signal(c);
4888
4889 case KILL_KILL:
4890 *ret_noteworthy = true;
4891 return c->final_kill_signal;
4892
4893 case KILL_WATCHDOG:
4894 *ret_noteworthy = true;
4895 return c->watchdog_signal;
4896
4897 default:
4898 assert_not_reached();
4899 }
4900 }
4901
4902 static int unit_kill_context_one(
4903 Unit *u,
4904 const PidRef *pidref,
4905 const char *type,
4906 bool is_alien,
4907 int sig,
4908 bool send_sighup,
4909 cg_kill_log_func_t log_func) {
4910
4911 int r;
4912
4913 assert(u);
4914 assert(type);
4915
4916 /* This returns > 0 if it makes sense to wait for SIGCHLD for the process, == 0 if not. */
4917
4918 if (!pidref_is_set(pidref))
4919 return 0;
4920
4921 if (log_func)
4922 log_func(pidref, sig, u);
4923
4924 r = pidref_kill_and_sigcont(pidref, sig);
4925 if (r == -ESRCH)
4926 return !is_alien;
4927 if (r < 0) {
4928 _cleanup_free_ char *comm = NULL;
4929
4930 (void) pidref_get_comm(pidref, &comm);
4931 return log_unit_warning_errno(u, r, "Failed to kill %s process " PID_FMT " (%s), ignoring: %m", type, pidref->pid, strna(comm));
4932 }
4933
4934 if (send_sighup)
4935 (void) pidref_kill(pidref, SIGHUP);
4936
4937 return !is_alien;
4938 }
4939
4940 int unit_kill_context(Unit *u, KillOperation k) {
4941 bool wait_for_exit = false, send_sighup;
4942 cg_kill_log_func_t log_func = NULL;
4943 int sig, r;
4944
4945 assert(u);
4946
4947 /* Kill the processes belonging to this unit, in preparation for shutting the unit down. Returns > 0
4948 * if we killed something worth waiting for, 0 otherwise. Do not confuse with unit_kill_common()
4949 * which is used for user-requested killing of unit processes. */
4950
4951 KillContext *c = unit_get_kill_context(u);
4952 if (!c || c->kill_mode == KILL_NONE)
4953 return 0;
4954
4955 bool noteworthy;
4956 sig = operation_to_signal(c, k, &noteworthy);
4957 if (noteworthy)
4958 log_func = log_kill;
4959
4960 send_sighup =
4961 c->send_sighup &&
4962 IN_SET(k, KILL_TERMINATE, KILL_TERMINATE_AND_LOG) &&
4963 sig != SIGHUP;
4964
4965 bool is_alien;
4966 PidRef *main_pid = unit_main_pid_full(u, &is_alien);
4967 r = unit_kill_context_one(u, main_pid, "main", is_alien, sig, send_sighup, log_func);
4968 wait_for_exit = wait_for_exit || r > 0;
4969
4970 r = unit_kill_context_one(u, unit_control_pid(u), "control", /* is_alien = */ false, sig, send_sighup, log_func);
4971 wait_for_exit = wait_for_exit || r > 0;
4972
4973 CGroupRuntime *crt = unit_get_cgroup_runtime(u);
4974 if (crt && crt->cgroup_path &&
4975 (c->kill_mode == KILL_CONTROL_GROUP || (c->kill_mode == KILL_MIXED && k == KILL_KILL))) {
4976 _cleanup_set_free_ Set *pid_set = NULL;
4977
4978 /* Exclude the main/control pids from being killed via the cgroup */
4979 r = unit_pid_set(u, &pid_set);
4980 if (r < 0)
4981 return r;
4982
4983 r = cg_kill_recursive(
4984 crt->cgroup_path,
4985 sig,
4986 CGROUP_SIGCONT|CGROUP_IGNORE_SELF,
4987 pid_set,
4988 log_func, u);
4989 if (r < 0) {
4990 if (!IN_SET(r, -EAGAIN, -ESRCH, -ENOENT))
4991 log_unit_warning_errno(u, r, "Failed to kill control group %s, ignoring: %m", empty_to_root(crt->cgroup_path));
4992
4993 } else if (r > 0) {
4994
4995 wait_for_exit = true;
4996
4997 if (send_sighup) {
4998 r = unit_pid_set(u, &pid_set);
4999 if (r < 0)
5000 return r;
5001
5002 (void) cg_kill_recursive(
5003 crt->cgroup_path,
5004 SIGHUP,
5005 CGROUP_IGNORE_SELF,
5006 pid_set,
5007 /* log_kill= */ NULL,
5008 /* userdata= */ NULL);
5009 }
5010 }
5011 }
5012
5013 return wait_for_exit;
5014 }
5015
5016 int unit_add_mounts_for(Unit *u, const char *path, UnitDependencyMask mask, UnitMountDependencyType type) {
5017 Hashmap **unit_map, **manager_map;
5018 int r;
5019
5020 assert(u);
5021 assert(path);
5022 assert(type >= 0 && type < _UNIT_MOUNT_DEPENDENCY_TYPE_MAX);
5023
5024 unit_map = &u->mounts_for[type];
5025 manager_map = &u->manager->units_needing_mounts_for[type];
5026
5027 /* Registers a unit for requiring a certain path and all its prefixes. We keep a hashtable of these
5028 * paths in the unit (from the path to the UnitDependencyInfo structure indicating how to the
5029 * dependency came to be). However, we build a prefix table for all possible prefixes so that new
5030 * appearing mount units can easily determine which units to make themselves a dependency of. */
5031
5032 if (!path_is_absolute(path))
5033 return -EINVAL;
5034
5035 if (hashmap_contains(*unit_map, path)) /* Exit quickly if the path is already covered. */
5036 return 0;
5037
5038 /* Use the canonical form of the path as the stored key. We call path_is_normalized()
5039 * only after simplification, since path_is_normalized() rejects paths with '.'.
5040 * path_is_normalized() also verifies that the path fits in PATH_MAX. */
5041 _cleanup_free_ char *p = NULL;
5042 r = path_simplify_alloc(path, &p);
5043 if (r < 0)
5044 return r;
5045 path = p;
5046
5047 if (!path_is_normalized(path))
5048 return -EPERM;
5049
5050 UnitDependencyInfo di = {
5051 .origin_mask = mask
5052 };
5053
5054 r = hashmap_ensure_put(unit_map, &path_hash_ops, p, di.data);
5055 if (r < 0)
5056 return r;
5057 assert(r > 0);
5058 TAKE_PTR(p); /* path remains a valid pointer to the string stored in the hashmap */
5059
5060 char prefix[strlen(path) + 1];
5061 PATH_FOREACH_PREFIX_MORE(prefix, path) {
5062 Set *x;
5063
5064 x = hashmap_get(*manager_map, prefix);
5065 if (!x) {
5066 _cleanup_free_ char *q = NULL;
5067
5068 r = hashmap_ensure_allocated(manager_map, &path_hash_ops);
5069 if (r < 0)
5070 return r;
5071
5072 q = strdup(prefix);
5073 if (!q)
5074 return -ENOMEM;
5075
5076 x = set_new(NULL);
5077 if (!x)
5078 return -ENOMEM;
5079
5080 r = hashmap_put(*manager_map, q, x);
5081 if (r < 0) {
5082 set_free(x);
5083 return r;
5084 }
5085 q = NULL;
5086 }
5087
5088 r = set_put(x, u);
5089 if (r < 0)
5090 return r;
5091 }
5092
5093 return 0;
5094 }
5095
5096 int unit_setup_exec_runtime(Unit *u) {
5097 _cleanup_(exec_shared_runtime_unrefp) ExecSharedRuntime *esr = NULL;
5098 _cleanup_(dynamic_creds_unrefp) DynamicCreds *dcreds = NULL;
5099 _cleanup_set_free_ Set *units = NULL;
5100 ExecRuntime **rt;
5101 ExecContext *ec;
5102 size_t offset;
5103 Unit *other;
5104 int r;
5105
5106 offset = UNIT_VTABLE(u)->exec_runtime_offset;
5107 assert(offset > 0);
5108
5109 /* Check if there already is an ExecRuntime for this unit? */
5110 rt = (ExecRuntime**) ((uint8_t*) u + offset);
5111 if (*rt)
5112 return 0;
5113
5114 ec = ASSERT_PTR(unit_get_exec_context(u));
5115
5116 r = unit_get_transitive_dependency_set(u, UNIT_ATOM_JOINS_NAMESPACE_OF, &units);
5117 if (r < 0)
5118 return r;
5119
5120 /* Try to get it from somebody else */
5121 SET_FOREACH(other, units) {
5122 r = exec_shared_runtime_acquire(u->manager, NULL, other->id, false, &esr);
5123 if (r < 0)
5124 return r;
5125 if (r > 0)
5126 break;
5127 }
5128
5129 if (!esr) {
5130 r = exec_shared_runtime_acquire(u->manager, ec, u->id, true, &esr);
5131 if (r < 0)
5132 return r;
5133 }
5134
5135 if (ec->dynamic_user) {
5136 r = dynamic_creds_make(u->manager, ec->user, ec->group, &dcreds);
5137 if (r < 0)
5138 return r;
5139 }
5140
5141 r = exec_runtime_make(u, ec, esr, dcreds, rt);
5142 if (r < 0)
5143 return r;
5144
5145 TAKE_PTR(esr);
5146 TAKE_PTR(dcreds);
5147
5148 return r;
5149 }
5150
5151 CGroupRuntime *unit_setup_cgroup_runtime(Unit *u) {
5152 size_t offset;
5153
5154 assert(u);
5155
5156 offset = UNIT_VTABLE(u)->cgroup_runtime_offset;
5157 assert(offset > 0);
5158
5159 CGroupRuntime **rt = (CGroupRuntime**) ((uint8_t*) u + offset);
5160 if (*rt)
5161 return *rt;
5162
5163 return (*rt = cgroup_runtime_new());
5164 }
5165
5166 bool unit_type_supported(UnitType t) {
5167 static int8_t cache[_UNIT_TYPE_MAX] = {}; /* -1: disabled, 1: enabled: 0: don't know */
5168 int r;
5169
5170 assert(t >= 0 && t < _UNIT_TYPE_MAX);
5171
5172 if (cache[t] == 0) {
5173 char *e;
5174
5175 e = strjoina("SYSTEMD_SUPPORT_", unit_type_to_string(t));
5176
5177 r = getenv_bool(ascii_strupper(e));
5178 if (r < 0 && r != -ENXIO)
5179 log_debug_errno(r, "Failed to parse $%s, ignoring: %m", e);
5180
5181 cache[t] = r == 0 ? -1 : 1;
5182 }
5183 if (cache[t] < 0)
5184 return false;
5185
5186 if (!unit_vtable[t]->supported)
5187 return true;
5188
5189 return unit_vtable[t]->supported();
5190 }
5191
5192 void unit_warn_if_dir_nonempty(Unit *u, const char* where) {
5193 int r;
5194
5195 assert(u);
5196 assert(where);
5197
5198 if (!unit_log_level_test(u, LOG_NOTICE))
5199 return;
5200
5201 r = dir_is_empty(where, /* ignore_hidden_or_backup= */ false);
5202 if (r > 0 || r == -ENOTDIR)
5203 return;
5204 if (r < 0) {
5205 log_unit_warning_errno(u, r, "Failed to check directory %s: %m", where);
5206 return;
5207 }
5208
5209 log_unit_struct(u, LOG_NOTICE,
5210 LOG_MESSAGE_ID(SD_MESSAGE_OVERMOUNTING_STR),
5211 LOG_UNIT_INVOCATION_ID(u),
5212 LOG_UNIT_MESSAGE(u, "Directory %s to mount over is not empty, mounting anyway.", where),
5213 LOG_ITEM("WHERE=%s", where));
5214 }
5215
5216 int unit_log_noncanonical_mount_path(Unit *u, const char *where) {
5217 assert(u);
5218 assert(where);
5219
5220 /* No need to mention "." or "..", they would already have been rejected by unit_name_from_path() */
5221 log_unit_struct(u, LOG_ERR,
5222 LOG_MESSAGE_ID(SD_MESSAGE_NON_CANONICAL_MOUNT_STR),
5223 LOG_UNIT_INVOCATION_ID(u),
5224 LOG_UNIT_MESSAGE(u, "Mount path %s is not canonical (contains a symlink).", where),
5225 LOG_ITEM("WHERE=%s", where));
5226
5227 return -ELOOP;
5228 }
5229
5230 int unit_fail_if_noncanonical_mount_path(Unit *u, const char* where) {
5231 int r;
5232
5233 assert(u);
5234 assert(where);
5235
5236 _cleanup_free_ char *canonical_where = NULL;
5237 r = chase(where, /* root= */ NULL, CHASE_NONEXISTENT, &canonical_where, /* ret_fd= */ NULL);
5238 if (r < 0) {
5239 log_unit_debug_errno(u, r, "Failed to check %s for symlinks, ignoring: %m", where);
5240 return 0;
5241 }
5242
5243 /* We will happily ignore a trailing slash (or any redundant slashes) */
5244 if (path_equal(where, canonical_where))
5245 return 0;
5246
5247 return unit_log_noncanonical_mount_path(u, where);
5248 }
5249
5250 bool unit_is_pristine(Unit *u) {
5251 assert(u);
5252
5253 /* Check if the unit already exists or is already around, in a number of different ways. Note that to
5254 * cater for unit types such as slice, we are generally fine with units that are marked UNIT_LOADED
5255 * even though nothing was actually loaded, as those unit types don't require a file on disk.
5256 *
5257 * Note that we don't check for drop-ins here, because we allow drop-ins for transient units
5258 * identically to non-transient units, both unit-specific and hierarchical. E.g. for a-b-c.service:
5259 * service.d/….conf, a-.service.d/….conf, a-b-.service.d/….conf, a-b-c.service.d/….conf.
5260 */
5261
5262 return IN_SET(u->load_state, UNIT_NOT_FOUND, UNIT_LOADED) &&
5263 !u->fragment_path &&
5264 !u->source_path &&
5265 !u->job &&
5266 !u->merged_into;
5267 }
5268
5269 PidRef* unit_control_pid(Unit *u) {
5270 assert(u);
5271
5272 if (UNIT_VTABLE(u)->control_pid)
5273 return UNIT_VTABLE(u)->control_pid(u);
5274
5275 return NULL;
5276 }
5277
5278 PidRef* unit_main_pid_full(Unit *u, bool *ret_is_alien) {
5279 assert(u);
5280
5281 if (UNIT_VTABLE(u)->main_pid)
5282 return UNIT_VTABLE(u)->main_pid(u, ret_is_alien);
5283
5284 if (ret_is_alien)
5285 *ret_is_alien = false;
5286 return NULL;
5287 }
5288
5289 static void unit_modify_user_nft_set(Unit *u, bool add, NFTSetSource source, uint32_t element) {
5290 int r;
5291
5292 assert(u);
5293
5294 if (!MANAGER_IS_SYSTEM(u->manager))
5295 return;
5296
5297 CGroupContext *c;
5298 c = unit_get_cgroup_context(u);
5299 if (!c)
5300 return;
5301
5302 if (!u->manager->nfnl) {
5303 r = sd_nfnl_socket_open(&u->manager->nfnl);
5304 if (r < 0)
5305 return;
5306 }
5307
5308 FOREACH_ARRAY(nft_set, c->nft_set_context.sets, c->nft_set_context.n_sets) {
5309 if (nft_set->source != source)
5310 continue;
5311
5312 r = nft_set_element_modify_any(u->manager->nfnl, add, nft_set->nfproto, nft_set->table, nft_set->set, &element, sizeof(element));
5313 if (r < 0)
5314 log_warning_errno(r, "Failed to %s NFT set entry: family %s, table %s, set %s, ID %u, ignoring: %m",
5315 add? "add" : "delete", nfproto_to_string(nft_set->nfproto), nft_set->table, nft_set->set, element);
5316 else
5317 log_debug("%s NFT set entry: family %s, table %s, set %s, ID %u",
5318 add? "Added" : "Deleted", nfproto_to_string(nft_set->nfproto), nft_set->table, nft_set->set, element);
5319 }
5320 }
5321
5322 static void unit_unref_uid_internal(
5323 Unit *u,
5324 uid_t *ref_uid,
5325 bool destroy_now,
5326 void (*_manager_unref_uid)(Manager *m, uid_t uid, bool destroy_now)) {
5327
5328 assert(u);
5329 assert(ref_uid);
5330 assert(_manager_unref_uid);
5331
5332 /* Generic implementation of both unit_unref_uid() and unit_unref_gid(), under the assumption that uid_t and
5333 * gid_t are actually the same time, with the same validity rules.
5334 *
5335 * Drops a reference to UID/GID from a unit. */
5336
5337 assert_cc(sizeof(uid_t) == sizeof(gid_t));
5338 assert_cc(UID_INVALID == (uid_t) GID_INVALID);
5339
5340 if (!uid_is_valid(*ref_uid))
5341 return;
5342
5343 _manager_unref_uid(u->manager, *ref_uid, destroy_now);
5344 *ref_uid = UID_INVALID;
5345 }
5346
5347 static void unit_unref_uid(Unit *u, bool destroy_now) {
5348 assert(u);
5349
5350 unit_modify_user_nft_set(u, /* add = */ false, NFT_SET_SOURCE_USER, u->ref_uid);
5351
5352 unit_unref_uid_internal(u, &u->ref_uid, destroy_now, manager_unref_uid);
5353 }
5354
5355 static void unit_unref_gid(Unit *u, bool destroy_now) {
5356 assert(u);
5357
5358 unit_modify_user_nft_set(u, /* add = */ false, NFT_SET_SOURCE_GROUP, u->ref_gid);
5359
5360 unit_unref_uid_internal(u, (uid_t*) &u->ref_gid, destroy_now, manager_unref_gid);
5361 }
5362
5363 void unit_unref_uid_gid(Unit *u, bool destroy_now) {
5364 assert(u);
5365
5366 unit_unref_uid(u, destroy_now);
5367 unit_unref_gid(u, destroy_now);
5368 }
5369
5370 static int unit_ref_uid_internal(
5371 Unit *u,
5372 uid_t *ref_uid,
5373 uid_t uid,
5374 bool clean_ipc,
5375 int (*_manager_ref_uid)(Manager *m, uid_t uid, bool clean_ipc)) {
5376
5377 int r;
5378
5379 assert(u);
5380 assert(ref_uid);
5381 assert(uid_is_valid(uid));
5382 assert(_manager_ref_uid);
5383
5384 /* Generic implementation of both unit_ref_uid() and unit_ref_guid(), under the assumption that uid_t and gid_t
5385 * are actually the same type, and have the same validity rules.
5386 *
5387 * Adds a reference on a specific UID/GID to this unit. Each unit referencing the same UID/GID maintains a
5388 * reference so that we can destroy the UID/GID's IPC resources as soon as this is requested and the counter
5389 * drops to zero. */
5390
5391 assert_cc(sizeof(uid_t) == sizeof(gid_t));
5392 assert_cc(UID_INVALID == (uid_t) GID_INVALID);
5393
5394 if (*ref_uid == uid)
5395 return 0;
5396
5397 if (uid_is_valid(*ref_uid)) /* Already set? */
5398 return -EBUSY;
5399
5400 r = _manager_ref_uid(u->manager, uid, clean_ipc);
5401 if (r < 0)
5402 return r;
5403
5404 *ref_uid = uid;
5405 return 1;
5406 }
5407
5408 static int unit_ref_uid(Unit *u, uid_t uid, bool clean_ipc) {
5409 return unit_ref_uid_internal(u, &u->ref_uid, uid, clean_ipc, manager_ref_uid);
5410 }
5411
5412 static int unit_ref_gid(Unit *u, gid_t gid, bool clean_ipc) {
5413 return unit_ref_uid_internal(u, (uid_t*) &u->ref_gid, (uid_t) gid, clean_ipc, manager_ref_gid);
5414 }
5415
5416 static int unit_ref_uid_gid_internal(Unit *u, uid_t uid, gid_t gid, bool clean_ipc) {
5417 int r = 0, q = 0;
5418
5419 assert(u);
5420
5421 /* Reference both a UID and a GID in one go. Either references both, or neither. */
5422
5423 if (uid_is_valid(uid)) {
5424 r = unit_ref_uid(u, uid, clean_ipc);
5425 if (r < 0)
5426 return r;
5427 }
5428
5429 if (gid_is_valid(gid)) {
5430 q = unit_ref_gid(u, gid, clean_ipc);
5431 if (q < 0) {
5432 if (r > 0)
5433 unit_unref_uid(u, false);
5434
5435 return q;
5436 }
5437 }
5438
5439 return r > 0 || q > 0;
5440 }
5441
5442 int unit_ref_uid_gid(Unit *u, uid_t uid, gid_t gid) {
5443 ExecContext *c;
5444 int r;
5445
5446 assert(u);
5447
5448 c = unit_get_exec_context(u);
5449
5450 r = unit_ref_uid_gid_internal(u, uid, gid, c ? c->remove_ipc : false);
5451 if (r < 0)
5452 return log_unit_warning_errno(u, r, "Couldn't add UID/GID reference to unit, proceeding without: %m");
5453
5454 unit_modify_user_nft_set(u, /* add = */ true, NFT_SET_SOURCE_USER, uid);
5455 unit_modify_user_nft_set(u, /* add = */ true, NFT_SET_SOURCE_GROUP, gid);
5456
5457 return r;
5458 }
5459
5460 void unit_notify_user_lookup(Unit *u, uid_t uid, gid_t gid) {
5461 int r;
5462
5463 assert(u);
5464
5465 /* This is invoked whenever one of the forked off processes let's us know the UID/GID its user name/group names
5466 * resolved to. We keep track of which UID/GID is currently assigned in order to be able to destroy its IPC
5467 * objects when no service references the UID/GID anymore. */
5468
5469 r = unit_ref_uid_gid(u, uid, gid);
5470 if (r > 0)
5471 unit_add_to_dbus_queue(u);
5472 }
5473
5474 int unit_acquire_invocation_id(Unit *u) {
5475 sd_id128_t id;
5476 int r;
5477
5478 assert(u);
5479
5480 r = sd_id128_randomize(&id);
5481 if (r < 0)
5482 return log_unit_error_errno(u, r, "Failed to generate invocation ID for unit: %m");
5483
5484 r = unit_set_invocation_id(u, id);
5485 if (r < 0)
5486 return log_unit_error_errno(u, r, "Failed to set invocation ID for unit: %m");
5487
5488 unit_add_to_dbus_queue(u);
5489 return 0;
5490 }
5491
5492 int unit_set_exec_params(Unit *u, ExecParameters *p) {
5493 int r;
5494
5495 assert(u);
5496 assert(p);
5497
5498 /* Copy parameters from manager */
5499 r = manager_get_effective_environment(u->manager, &p->environment);
5500 if (r < 0)
5501 return r;
5502
5503 p->runtime_scope = u->manager->runtime_scope;
5504
5505 r = strdup_to(&p->confirm_spawn, manager_get_confirm_spawn(u->manager));
5506 if (r < 0)
5507 return r;
5508
5509 p->prefix = u->manager->prefix;
5510 SET_FLAG(p->flags, EXEC_PASS_LOG_UNIT|EXEC_CHOWN_DIRECTORIES, MANAGER_IS_SYSTEM(u->manager));
5511
5512 /* Copy parameters from unit */
5513 CGroupRuntime *crt = unit_get_cgroup_runtime(u);
5514 p->cgroup_path = crt ? crt->cgroup_path : NULL;
5515 SET_FLAG(p->flags, EXEC_CGROUP_DELEGATE, unit_cgroup_delegate(u));
5516
5517 p->received_credentials_directory = u->manager->received_credentials_directory;
5518 p->received_encrypted_credentials_directory = u->manager->received_encrypted_credentials_directory;
5519
5520 p->shall_confirm_spawn = u->manager->confirm_spawn;
5521
5522 p->fallback_smack_process_label = u->manager->defaults.smack_process_label;
5523
5524 if (u->manager->restrict_fs && p->bpf_restrict_fs_map_fd < 0) {
5525 int fd = bpf_restrict_fs_map_fd(u);
5526 if (fd < 0)
5527 return fd;
5528
5529 p->bpf_restrict_fs_map_fd = fd;
5530 }
5531
5532 p->user_lookup_fd = u->manager->user_lookup_fds[1];
5533 p->handoff_timestamp_fd = u->manager->handoff_timestamp_fds[1];
5534 if (UNIT_VTABLE(u)->notify_pidref)
5535 p->pidref_transport_fd = u->manager->pidref_transport_fds[1];
5536
5537 p->cgroup_id = crt ? crt->cgroup_id : 0;
5538 p->invocation_id = u->invocation_id;
5539 sd_id128_to_string(p->invocation_id, p->invocation_id_string);
5540 p->unit_id = strdup(u->id);
5541 if (!p->unit_id)
5542 return -ENOMEM;
5543
5544 p->debug_invocation = u->debug_invocation;
5545
5546 return 0;
5547 }
5548
5549 int unit_fork_helper_process(Unit *u, const char *name, bool into_cgroup, PidRef *ret) {
5550 CGroupRuntime *crt = NULL;
5551 pid_t pid;
5552 int r;
5553
5554 assert(u);
5555 assert(ret);
5556
5557 /* Forks off a helper process and makes sure it is a member of the unit's cgroup, if configured to
5558 * do so. Returns == 0 in the child, and > 0 in the parent. The pid parameter is always filled in
5559 * with the child's PID. */
5560
5561 if (into_cgroup) {
5562 r = unit_realize_cgroup(u);
5563 if (r < 0)
5564 return r;
5565
5566 crt = unit_get_cgroup_runtime(u);
5567 }
5568
5569 r = safe_fork(name, FORK_REOPEN_LOG|FORK_DEATHSIG_SIGTERM, &pid);
5570 if (r < 0)
5571 return r;
5572 if (r > 0) {
5573 _cleanup_(pidref_done) PidRef pidref = PIDREF_NULL;
5574 int q;
5575
5576 /* Parent */
5577
5578 q = pidref_set_pid(&pidref, pid);
5579 if (q < 0)
5580 return q;
5581
5582 *ret = TAKE_PIDREF(pidref);
5583 return r;
5584 }
5585
5586 /* Child */
5587
5588 (void) default_signals(SIGNALS_CRASH_HANDLER, SIGNALS_IGNORE);
5589 (void) ignore_signals(SIGPIPE);
5590
5591 if (crt && crt->cgroup_path) {
5592 r = cg_attach(crt->cgroup_path, 0);
5593 if (r < 0) {
5594 log_unit_error_errno(u, r, "Failed to join unit cgroup %s: %m", empty_to_root(crt->cgroup_path));
5595 _exit(EXIT_CGROUP);
5596 }
5597 }
5598
5599 return 0;
5600 }
5601
5602 int unit_fork_and_watch_rm_rf(Unit *u, char **paths, PidRef *ret_pid) {
5603 _cleanup_(pidref_done) PidRef pid = PIDREF_NULL;
5604 int r;
5605
5606 assert(u);
5607 assert(ret_pid);
5608
5609 r = unit_fork_helper_process(u, "(sd-rmrf)", /* into_cgroup= */ true, &pid);
5610 if (r < 0)
5611 return r;
5612 if (r == 0) {
5613 int ret = EXIT_SUCCESS;
5614
5615 STRV_FOREACH(i, paths) {
5616 r = rm_rf(*i, REMOVE_ROOT|REMOVE_PHYSICAL|REMOVE_MISSING_OK);
5617 if (r < 0) {
5618 log_error_errno(r, "Failed to remove '%s': %m", *i);
5619 ret = EXIT_FAILURE;
5620 }
5621 }
5622
5623 _exit(ret);
5624 }
5625
5626 r = unit_watch_pidref(u, &pid, /* exclusive= */ true);
5627 if (r < 0)
5628 return r;
5629
5630 *ret_pid = TAKE_PIDREF(pid);
5631 return 0;
5632 }
5633
5634 static void unit_update_dependency_mask(Hashmap *deps, Unit *other, UnitDependencyInfo di) {
5635 assert(deps);
5636 assert(other);
5637
5638 if (di.origin_mask == 0 && di.destination_mask == 0)
5639 /* No bit set anymore, let's drop the whole entry */
5640 assert_se(hashmap_remove(deps, other));
5641 else
5642 /* Mask was reduced, let's update the entry */
5643 assert_se(hashmap_update(deps, other, di.data) == 0);
5644 }
5645
5646 void unit_remove_dependencies(Unit *u, UnitDependencyMask mask) {
5647 Hashmap *deps;
5648 assert(u);
5649
5650 /* Removes all dependencies u has on other units marked for ownership by 'mask'. */
5651
5652 if (mask == 0)
5653 return;
5654
5655 HASHMAP_FOREACH(deps, u->dependencies) {
5656 bool done;
5657
5658 do {
5659 UnitDependencyInfo di;
5660 Unit *other;
5661
5662 done = true;
5663
5664 HASHMAP_FOREACH_KEY(di.data, other, deps) {
5665 Hashmap *other_deps;
5666
5667 if (FLAGS_SET(~mask, di.origin_mask))
5668 continue;
5669
5670 di.origin_mask &= ~mask;
5671 unit_update_dependency_mask(deps, other, di);
5672
5673 /* We updated the dependency from our unit to the other unit now. But most
5674 * dependencies imply a reverse dependency. Hence, let's delete that one
5675 * too. For that we go through all dependency types on the other unit and
5676 * delete all those which point to us and have the right mask set. */
5677
5678 HASHMAP_FOREACH(other_deps, other->dependencies) {
5679 UnitDependencyInfo dj;
5680
5681 dj.data = hashmap_get(other_deps, u);
5682 if (FLAGS_SET(~mask, dj.destination_mask))
5683 continue;
5684
5685 dj.destination_mask &= ~mask;
5686 unit_update_dependency_mask(other_deps, u, dj);
5687 }
5688
5689 unit_add_to_gc_queue(other);
5690
5691 /* The unit 'other' may not be wanted by the unit 'u'. */
5692 unit_submit_to_stop_when_unneeded_queue(other);
5693
5694 u->dependency_generation++;
5695 other->dependency_generation++;
5696
5697 done = false;
5698 break;
5699 }
5700
5701 } while (!done);
5702 }
5703 }
5704
5705 static int unit_get_invocation_path(Unit *u, char **ret) {
5706 char *p;
5707 int r;
5708
5709 assert(u);
5710 assert(ret);
5711
5712 if (MANAGER_IS_SYSTEM(u->manager))
5713 p = strjoin("/run/systemd/units/invocation:", u->id);
5714 else {
5715 _cleanup_free_ char *user_path = NULL;
5716
5717 r = xdg_user_runtime_dir("/systemd/units/invocation:", &user_path);
5718 if (r < 0)
5719 return r;
5720
5721 p = strjoin(user_path, u->id);
5722 }
5723 if (!p)
5724 return -ENOMEM;
5725
5726 *ret = p;
5727 return 0;
5728 }
5729
5730 static int unit_export_invocation_id(Unit *u) {
5731 _cleanup_free_ char *p = NULL;
5732 int r;
5733
5734 assert(u);
5735
5736 if (u->exported_invocation_id)
5737 return 0;
5738
5739 if (sd_id128_is_null(u->invocation_id))
5740 return 0;
5741
5742 r = unit_get_invocation_path(u, &p);
5743 if (r < 0)
5744 return log_unit_debug_errno(u, r, "Failed to get invocation path: %m");
5745
5746 r = symlinkat_atomic_full(u->invocation_id_string, AT_FDCWD, p, SYMLINK_LABEL);
5747 if (r < 0)
5748 return log_unit_debug_errno(u, r, "Failed to create invocation ID symlink %s: %m", p);
5749
5750 u->exported_invocation_id = true;
5751 return 0;
5752 }
5753
5754 static int unit_export_log_level_max(Unit *u, int log_level_max, bool overwrite) {
5755 const char *p;
5756 char buf[2];
5757 int r;
5758
5759 assert(u);
5760
5761 /* When the debug_invocation logic runs, overwrite will be true as we always want to switch the max
5762 * log level that the journal applies, and we want to always restore the previous level once done */
5763
5764 if (!overwrite && u->exported_log_level_max)
5765 return 0;
5766
5767 if (log_level_max < 0)
5768 return 0;
5769
5770 assert(log_level_max <= 7);
5771
5772 buf[0] = '0' + log_level_max;
5773 buf[1] = 0;
5774
5775 p = strjoina("/run/systemd/units/log-level-max:", u->id);
5776 r = symlink_atomic(buf, p);
5777 if (r < 0)
5778 return log_unit_debug_errno(u, r, "Failed to create maximum log level symlink %s: %m", p);
5779
5780 u->exported_log_level_max = true;
5781 return 0;
5782 }
5783
5784 static int unit_export_log_extra_fields(Unit *u, const ExecContext *c) {
5785 _cleanup_close_ int fd = -EBADF;
5786 struct iovec *iovec;
5787 const char *p;
5788 char *pattern;
5789 le64_t *sizes;
5790 ssize_t n;
5791 int r;
5792
5793 if (u->exported_log_extra_fields)
5794 return 0;
5795
5796 if (c->n_log_extra_fields <= 0)
5797 return 0;
5798
5799 sizes = newa(le64_t, c->n_log_extra_fields);
5800 iovec = newa(struct iovec, c->n_log_extra_fields * 2);
5801
5802 for (size_t i = 0; i < c->n_log_extra_fields; i++) {
5803 sizes[i] = htole64(c->log_extra_fields[i].iov_len);
5804
5805 iovec[i*2] = IOVEC_MAKE(sizes + i, sizeof(le64_t));
5806 iovec[i*2+1] = c->log_extra_fields[i];
5807 }
5808
5809 p = strjoina("/run/systemd/units/log-extra-fields:", u->id);
5810 pattern = strjoina(p, ".XXXXXX");
5811
5812 fd = mkostemp_safe(pattern);
5813 if (fd < 0)
5814 return log_unit_debug_errno(u, fd, "Failed to create extra fields file %s: %m", p);
5815
5816 n = writev(fd, iovec, c->n_log_extra_fields*2);
5817 if (n < 0) {
5818 r = log_unit_debug_errno(u, errno, "Failed to write extra fields: %m");
5819 goto fail;
5820 }
5821
5822 (void) fchmod(fd, 0644);
5823
5824 if (rename(pattern, p) < 0) {
5825 r = log_unit_debug_errno(u, errno, "Failed to rename extra fields file: %m");
5826 goto fail;
5827 }
5828
5829 u->exported_log_extra_fields = true;
5830 return 0;
5831
5832 fail:
5833 (void) unlink(pattern);
5834 return r;
5835 }
5836
5837 static int unit_export_log_ratelimit_interval(Unit *u, const ExecContext *c) {
5838 _cleanup_free_ char *buf = NULL;
5839 const char *p;
5840 int r;
5841
5842 assert(u);
5843 assert(c);
5844
5845 if (u->exported_log_ratelimit_interval)
5846 return 0;
5847
5848 if (c->log_ratelimit.interval == 0)
5849 return 0;
5850
5851 p = strjoina("/run/systemd/units/log-rate-limit-interval:", u->id);
5852
5853 if (asprintf(&buf, "%" PRIu64, c->log_ratelimit.interval) < 0)
5854 return log_oom();
5855
5856 r = symlink_atomic(buf, p);
5857 if (r < 0)
5858 return log_unit_debug_errno(u, r, "Failed to create log rate limit interval symlink %s: %m", p);
5859
5860 u->exported_log_ratelimit_interval = true;
5861 return 0;
5862 }
5863
5864 static int unit_export_log_ratelimit_burst(Unit *u, const ExecContext *c) {
5865 _cleanup_free_ char *buf = NULL;
5866 const char *p;
5867 int r;
5868
5869 assert(u);
5870 assert(c);
5871
5872 if (u->exported_log_ratelimit_burst)
5873 return 0;
5874
5875 if (c->log_ratelimit.burst == 0)
5876 return 0;
5877
5878 p = strjoina("/run/systemd/units/log-rate-limit-burst:", u->id);
5879
5880 if (asprintf(&buf, "%u", c->log_ratelimit.burst) < 0)
5881 return log_oom();
5882
5883 r = symlink_atomic(buf, p);
5884 if (r < 0)
5885 return log_unit_debug_errno(u, r, "Failed to create log rate limit burst symlink %s: %m", p);
5886
5887 u->exported_log_ratelimit_burst = true;
5888 return 0;
5889 }
5890
5891 void unit_export_state_files(Unit *u) {
5892 const ExecContext *c;
5893
5894 assert(u);
5895
5896 if (!u->id)
5897 return;
5898
5899 if (MANAGER_IS_TEST_RUN(u->manager))
5900 return;
5901
5902 /* Exports a couple of unit properties to /run/systemd/units/, so that journald can quickly query this data
5903 * from there. Ideally, journald would use IPC to query this, like everybody else, but that's hard, as long as
5904 * the IPC system itself and PID 1 also log to the journal.
5905 *
5906 * Note that these files really shouldn't be considered API for anyone else, as use a runtime file system as
5907 * IPC replacement is not compatible with today's world of file system namespaces. However, this doesn't really
5908 * apply to communication between the journal and systemd, as we assume that these two daemons live in the same
5909 * namespace at least.
5910 *
5911 * Note that some of the "files" exported here are actually symlinks and not regular files. Symlinks work
5912 * better for storing small bits of data, in particular as we can write them with two system calls, and read
5913 * them with one. */
5914
5915 (void) unit_export_invocation_id(u);
5916
5917 if (!MANAGER_IS_SYSTEM(u->manager))
5918 return;
5919
5920 c = unit_get_exec_context(u);
5921 if (c) {
5922 (void) unit_export_log_level_max(u, c->log_level_max, /* overwrite= */ false);
5923 (void) unit_export_log_extra_fields(u, c);
5924 (void) unit_export_log_ratelimit_interval(u, c);
5925 (void) unit_export_log_ratelimit_burst(u, c);
5926 }
5927 }
5928
5929 void unit_unlink_state_files(Unit *u) {
5930 const char *p;
5931
5932 assert(u);
5933
5934 if (!u->id)
5935 return;
5936
5937 /* Undoes the effect of unit_export_state() */
5938
5939 if (u->exported_invocation_id) {
5940 _cleanup_free_ char *invocation_path = NULL;
5941 int r = unit_get_invocation_path(u, &invocation_path);
5942 if (r >= 0) {
5943 (void) unlink(invocation_path);
5944 u->exported_invocation_id = false;
5945 }
5946 }
5947
5948 if (!MANAGER_IS_SYSTEM(u->manager))
5949 return;
5950
5951 if (u->exported_log_level_max) {
5952 p = strjoina("/run/systemd/units/log-level-max:", u->id);
5953 (void) unlink(p);
5954
5955 u->exported_log_level_max = false;
5956 }
5957
5958 if (u->exported_log_extra_fields) {
5959 p = strjoina("/run/systemd/units/extra-fields:", u->id);
5960 (void) unlink(p);
5961
5962 u->exported_log_extra_fields = false;
5963 }
5964
5965 if (u->exported_log_ratelimit_interval) {
5966 p = strjoina("/run/systemd/units/log-rate-limit-interval:", u->id);
5967 (void) unlink(p);
5968
5969 u->exported_log_ratelimit_interval = false;
5970 }
5971
5972 if (u->exported_log_ratelimit_burst) {
5973 p = strjoina("/run/systemd/units/log-rate-limit-burst:", u->id);
5974 (void) unlink(p);
5975
5976 u->exported_log_ratelimit_burst = false;
5977 }
5978 }
5979
5980 int unit_set_debug_invocation(Unit *u, bool enable) {
5981 int r;
5982
5983 assert(u);
5984
5985 if (u->debug_invocation == enable)
5986 return 0; /* Nothing to do */
5987
5988 u->debug_invocation = enable;
5989
5990 /* Ensure that the new log level is exported for the journal, in place of the previous one */
5991 if (u->exported_log_level_max) {
5992 const ExecContext *ec = unit_get_exec_context(u);
5993 if (ec) {
5994 r = unit_export_log_level_max(u, enable ? LOG_PRI(LOG_DEBUG) : ec->log_level_max, /* overwrite= */ true);
5995 if (r < 0)
5996 return r;
5997 }
5998 }
5999
6000 return 1;
6001 }
6002
6003 int unit_prepare_exec(Unit *u) {
6004 int r;
6005
6006 assert(u);
6007
6008 /* Prepares everything so that we can fork of a process for this unit */
6009
6010 r = unit_realize_cgroup(u);
6011 if (r < 0)
6012 return r;
6013
6014 CGroupRuntime *crt = unit_get_cgroup_runtime(u);
6015 if (crt && crt->reset_accounting) {
6016 (void) unit_reset_accounting(u);
6017 crt->reset_accounting = false;
6018 }
6019
6020 unit_export_state_files(u);
6021
6022 r = unit_setup_exec_runtime(u);
6023 if (r < 0)
6024 return r;
6025
6026 return 0;
6027 }
6028
6029 static int unit_log_leftover_process_start(const PidRef *pid, int sig, void *userdata) {
6030 const Unit *u = ASSERT_PTR(userdata);
6031 _cleanup_free_ char *comm = NULL;
6032
6033 assert(pidref_is_set(pid));
6034
6035 (void) pidref_get_comm(pid, &comm);
6036
6037 if (ignore_leftover_process(comm))
6038 return 0;
6039
6040 /* During start we print a warning */
6041
6042 log_unit_warning(u,
6043 "Found left-over process " PID_FMT " (%s) in control group while starting unit. Ignoring.\n"
6044 "This usually indicates unclean termination of a previous run, or service implementation deficiencies.",
6045 pid->pid, strna(comm));
6046
6047 return 1;
6048 }
6049
6050 static int unit_log_leftover_process_stop(const PidRef *pid, int sig, void *userdata) {
6051 const Unit *u = ASSERT_PTR(userdata);
6052 _cleanup_free_ char *comm = NULL;
6053
6054 assert(pidref_is_set(pid));
6055
6056 (void) pidref_get_comm(pid, &comm);
6057
6058 if (ignore_leftover_process(comm))
6059 return 0;
6060
6061 /* During stop we only print an informational message */
6062
6063 log_unit_info(u,
6064 "Unit process " PID_FMT " (%s) remains running after unit stopped.",
6065 pid->pid, strna(comm));
6066
6067 return 1;
6068 }
6069
6070 int unit_warn_leftover_processes(Unit *u, bool start) {
6071 _cleanup_free_ char *cgroup = NULL;
6072 int r;
6073
6074 assert(u);
6075
6076 r = unit_get_cgroup_path_with_fallback(u, &cgroup);
6077 if (r < 0)
6078 return r;
6079
6080 return cg_kill_recursive(
6081 cgroup,
6082 /* sig= */ 0,
6083 /* flags= */ 0,
6084 /* killed_pids= */ NULL,
6085 start ? unit_log_leftover_process_start : unit_log_leftover_process_stop,
6086 u);
6087 }
6088
6089 bool unit_needs_console(Unit *u) {
6090 ExecContext *ec;
6091 UnitActiveState state;
6092
6093 assert(u);
6094
6095 state = unit_active_state(u);
6096
6097 if (UNIT_IS_INACTIVE_OR_FAILED(state))
6098 return false;
6099
6100 if (UNIT_VTABLE(u)->needs_console)
6101 return UNIT_VTABLE(u)->needs_console(u);
6102
6103 /* If this unit type doesn't implement this call, let's use a generic fallback implementation: */
6104 ec = unit_get_exec_context(u);
6105 if (!ec)
6106 return false;
6107
6108 return exec_context_may_touch_console(ec);
6109 }
6110
6111 int unit_pid_attachable(Unit *u, PidRef *pid, sd_bus_error *error) {
6112 int r;
6113
6114 assert(u);
6115
6116 /* Checks whether the specified PID is generally good for attaching, i.e. a valid PID, not our manager itself,
6117 * and not a kernel thread either */
6118
6119 /* First, a simple range check */
6120 if (!pidref_is_set(pid))
6121 return sd_bus_error_set(error, SD_BUS_ERROR_INVALID_ARGS, "Process identifier is not valid.");
6122
6123 /* Some extra safety check */
6124 if (pid->pid == 1 || pidref_is_self(pid))
6125 return sd_bus_error_setf(error, SD_BUS_ERROR_INVALID_ARGS, "Process " PID_FMT " is a manager process, refusing.", pid->pid);
6126
6127 /* Don't even begin to bother with kernel threads */
6128 r = pidref_is_kernel_thread(pid);
6129 if (r == -ESRCH)
6130 return sd_bus_error_setf(error, SD_BUS_ERROR_UNIX_PROCESS_ID_UNKNOWN, "Process with ID " PID_FMT " does not exist.", pid->pid);
6131 if (r < 0)
6132 return sd_bus_error_set_errnof(error, r, "Failed to determine whether process " PID_FMT " is a kernel thread: %m", pid->pid);
6133 if (r > 0)
6134 return sd_bus_error_setf(error, SD_BUS_ERROR_INVALID_ARGS, "Process " PID_FMT " is a kernel thread, refusing.", pid->pid);
6135
6136 return 0;
6137 }
6138
6139 int unit_get_log_level_max(const Unit *u) {
6140 if (u) {
6141 if (u->debug_invocation)
6142 return LOG_DEBUG;
6143
6144 ExecContext *ec = unit_get_exec_context(u);
6145 if (ec && ec->log_level_max >= 0)
6146 return ec->log_level_max;
6147 }
6148
6149 return log_get_max_level();
6150 }
6151
6152 bool unit_log_level_test(const Unit *u, int level) {
6153 assert(u);
6154 return LOG_PRI(level) <= unit_get_log_level_max(u);
6155 }
6156
6157 void unit_log_success(Unit *u) {
6158 assert(u);
6159
6160 /* Let's show message "Deactivated successfully" in debug mode (when manager is user) rather than in info mode.
6161 * This message has low information value for regular users and it might be a bit overwhelming on a system with
6162 * a lot of devices. */
6163 log_unit_struct(u,
6164 MANAGER_IS_USER(u->manager) ? LOG_DEBUG : LOG_INFO,
6165 LOG_MESSAGE_ID(SD_MESSAGE_UNIT_SUCCESS_STR),
6166 LOG_UNIT_INVOCATION_ID(u),
6167 LOG_UNIT_MESSAGE(u, "Deactivated successfully."));
6168 }
6169
6170 void unit_log_failure(Unit *u, const char *result) {
6171 assert(u);
6172 assert(result);
6173
6174 log_unit_struct(u, LOG_WARNING,
6175 LOG_MESSAGE_ID(SD_MESSAGE_UNIT_FAILURE_RESULT_STR),
6176 LOG_UNIT_INVOCATION_ID(u),
6177 LOG_UNIT_MESSAGE(u, "Failed with result '%s'.", result),
6178 LOG_ITEM("UNIT_RESULT=%s", result));
6179 }
6180
6181 void unit_log_skip(Unit *u, const char *result) {
6182 assert(u);
6183 assert(result);
6184
6185 log_unit_struct(u, LOG_INFO,
6186 LOG_MESSAGE_ID(SD_MESSAGE_UNIT_SKIPPED_STR),
6187 LOG_UNIT_INVOCATION_ID(u),
6188 LOG_UNIT_MESSAGE(u, "Skipped due to '%s'.", result),
6189 LOG_ITEM("UNIT_RESULT=%s", result));
6190 }
6191
6192 void unit_log_process_exit(
6193 Unit *u,
6194 const char *kind,
6195 const char *command,
6196 bool success,
6197 int code,
6198 int status) {
6199
6200 int level;
6201
6202 assert(u);
6203 assert(kind);
6204
6205 /* If this is a successful exit, let's log about the exit code on DEBUG level. If this is a failure
6206 * and the process exited on its own via exit(), then let's make this a NOTICE, under the assumption
6207 * that the service already logged the reason at a higher log level on its own. Otherwise, make it a
6208 * WARNING. */
6209 if (success)
6210 level = LOG_DEBUG;
6211 else if (code == CLD_EXITED)
6212 level = LOG_NOTICE;
6213 else
6214 level = LOG_WARNING;
6215
6216 log_unit_struct(u, level,
6217 LOG_MESSAGE_ID(SD_MESSAGE_UNIT_PROCESS_EXIT_STR),
6218 LOG_UNIT_MESSAGE(u, "%s exited, code=%s, status=%i/%s%s",
6219 kind,
6220 sigchld_code_to_string(code), status,
6221 strna(code == CLD_EXITED
6222 ? exit_status_to_string(status, EXIT_STATUS_FULL)
6223 : signal_to_string(status)),
6224 success ? " (success)" : ""),
6225 LOG_ITEM("EXIT_CODE=%s", sigchld_code_to_string(code)),
6226 LOG_ITEM("EXIT_STATUS=%i", status),
6227 LOG_ITEM("COMMAND=%s", strna(command)),
6228 LOG_UNIT_INVOCATION_ID(u));
6229 }
6230
6231 int unit_exit_status(Unit *u) {
6232 assert(u);
6233
6234 /* Returns the exit status to propagate for the most recent cycle of this unit. Returns a value in the range
6235 * 0…255 if there's something to propagate. EOPNOTSUPP if the concept does not apply to this unit type, ENODATA
6236 * if no data is currently known (for example because the unit hasn't deactivated yet) and EBADE if the main
6237 * service process has exited abnormally (signal/coredump). */
6238
6239 if (!UNIT_VTABLE(u)->exit_status)
6240 return -EOPNOTSUPP;
6241
6242 return UNIT_VTABLE(u)->exit_status(u);
6243 }
6244
6245 int unit_failure_action_exit_status(Unit *u) {
6246 int r;
6247
6248 assert(u);
6249
6250 /* Returns the exit status to propagate on failure, or an error if there's nothing to propagate */
6251
6252 if (u->failure_action_exit_status >= 0)
6253 return u->failure_action_exit_status;
6254
6255 r = unit_exit_status(u);
6256 if (r == -EBADE) /* Exited, but not cleanly (i.e. by signal or such) */
6257 return 255;
6258
6259 return r;
6260 }
6261
6262 int unit_success_action_exit_status(Unit *u) {
6263 int r;
6264
6265 assert(u);
6266
6267 /* Returns the exit status to propagate on success, or an error if there's nothing to propagate */
6268
6269 if (u->success_action_exit_status >= 0)
6270 return u->success_action_exit_status;
6271
6272 r = unit_exit_status(u);
6273 if (r == -EBADE) /* Exited, but not cleanly (i.e. by signal or such) */
6274 return 255;
6275
6276 return r;
6277 }
6278
6279 int unit_test_trigger_loaded(Unit *u) {
6280 Unit *trigger;
6281
6282 /* Tests whether the unit to trigger is loaded */
6283
6284 trigger = UNIT_TRIGGER(u);
6285 if (!trigger)
6286 return log_unit_error_errno(u, SYNTHETIC_ERRNO(ENOENT),
6287 "Refusing to start, no unit to trigger.");
6288 if (trigger->load_state != UNIT_LOADED)
6289 return log_unit_error_errno(u, SYNTHETIC_ERRNO(ENOENT),
6290 "Refusing to start, unit %s to trigger not loaded.", trigger->id);
6291
6292 return 0;
6293 }
6294
6295 void unit_destroy_runtime_data(Unit *u, const ExecContext *context, bool destroy_runtime_dir) {
6296 assert(u);
6297 assert(u->manager);
6298 assert(context);
6299
6300 /* EXEC_PRESERVE_RESTART is handled via unit_release_resources()! */
6301 if (destroy_runtime_dir && context->runtime_directory_preserve_mode == EXEC_PRESERVE_NO)
6302 exec_context_destroy_runtime_directory(context, u->manager->prefix[EXEC_DIRECTORY_RUNTIME]);
6303
6304 exec_context_destroy_credentials(context, u->manager->prefix[EXEC_DIRECTORY_RUNTIME], u->id);
6305 exec_context_destroy_mount_ns_dir(u);
6306 }
6307
6308 int unit_clean(Unit *u, ExecCleanMask mask) {
6309 UnitActiveState state;
6310
6311 assert(u);
6312
6313 /* Special return values:
6314 *
6315 * -EOPNOTSUPP → cleaning not supported for this unit type
6316 * -EUNATCH → cleaning not defined for this resource type
6317 * -EBUSY → unit currently can't be cleaned since it's running or not properly loaded, or has
6318 * a job queued or similar
6319 */
6320
6321 if (!UNIT_VTABLE(u)->clean)
6322 return -EOPNOTSUPP;
6323
6324 if (mask == 0)
6325 return -EUNATCH;
6326
6327 if (u->load_state != UNIT_LOADED)
6328 return -EBUSY;
6329
6330 if (u->job)
6331 return -EBUSY;
6332
6333 state = unit_active_state(u);
6334 if (state != UNIT_INACTIVE)
6335 return -EBUSY;
6336
6337 return UNIT_VTABLE(u)->clean(u, mask);
6338 }
6339
6340 int unit_can_clean(Unit *u, ExecCleanMask *ret) {
6341 assert(u);
6342
6343 if (!UNIT_VTABLE(u)->clean ||
6344 u->load_state != UNIT_LOADED) {
6345 *ret = 0;
6346 return 0;
6347 }
6348
6349 /* When the clean() method is set, can_clean() really should be set too */
6350 assert(UNIT_VTABLE(u)->can_clean);
6351
6352 return UNIT_VTABLE(u)->can_clean(u, ret);
6353 }
6354
6355 bool unit_can_start_refuse_manual(Unit *u) {
6356 return unit_can_start(u) && !u->refuse_manual_start;
6357 }
6358
6359 bool unit_can_stop_refuse_manual(Unit *u) {
6360 return unit_can_stop(u) && !u->refuse_manual_stop;
6361 }
6362
6363 bool unit_can_isolate_refuse_manual(Unit *u) {
6364 return unit_can_isolate(u) && !u->refuse_manual_start;
6365 }
6366
6367 void unit_next_freezer_state(Unit *u, FreezerAction action, FreezerState *ret_next, FreezerState *ret_objective) {
6368 FreezerState current, parent, next, objective;
6369
6370 assert(u);
6371 assert(action >= 0);
6372 assert(action < _FREEZER_ACTION_MAX);
6373 assert(ret_next);
6374 assert(ret_objective);
6375
6376 /* This function determines the correct freezer state transitions for a unit
6377 * given the action being requested. It returns the next state, and also the "objective",
6378 * which is either FREEZER_FROZEN or FREEZER_RUNNING, depending on what actual state we
6379 * ultimately want to achieve. */
6380
6381 current = u->freezer_state;
6382
6383 Unit *slice = UNIT_GET_SLICE(u);
6384 if (slice)
6385 parent = slice->freezer_state;
6386 else
6387 parent = FREEZER_RUNNING;
6388
6389 switch (action) {
6390
6391 case FREEZER_FREEZE:
6392 /* We always "promote" a freeze initiated by parent into a normal freeze */
6393 if (IN_SET(current, FREEZER_FROZEN, FREEZER_FROZEN_BY_PARENT))
6394 next = FREEZER_FROZEN;
6395 else
6396 next = FREEZER_FREEZING;
6397 break;
6398
6399 case FREEZER_THAW:
6400 /* Thawing is the most complicated operation here, because we can't thaw a unit
6401 * if its parent is frozen. So we instead "demote" a normal freeze into a freeze
6402 * initiated by parent if the parent is frozen */
6403 if (IN_SET(current, FREEZER_RUNNING, FREEZER_THAWING,
6404 FREEZER_FREEZING_BY_PARENT, FREEZER_FROZEN_BY_PARENT)) /* Should usually be refused by unit_freezer_action */
6405 next = current;
6406 else if (current == FREEZER_FREEZING) {
6407 if (IN_SET(parent, FREEZER_RUNNING, FREEZER_THAWING))
6408 next = FREEZER_THAWING;
6409 else
6410 next = FREEZER_FREEZING_BY_PARENT;
6411 } else if (current == FREEZER_FROZEN) {
6412 if (IN_SET(parent, FREEZER_RUNNING, FREEZER_THAWING))
6413 next = FREEZER_THAWING;
6414 else
6415 next = FREEZER_FROZEN_BY_PARENT;
6416 } else
6417 assert_not_reached();
6418 break;
6419
6420 case FREEZER_PARENT_FREEZE:
6421 /* We need to avoid accidentally demoting units frozen manually */
6422 if (IN_SET(current, FREEZER_FREEZING, FREEZER_FROZEN, FREEZER_FROZEN_BY_PARENT))
6423 next = current;
6424 else
6425 next = FREEZER_FREEZING_BY_PARENT;
6426 break;
6427
6428 case FREEZER_PARENT_THAW:
6429 /* We don't want to thaw units from a parent if they were frozen
6430 * manually, so for such units this action is a no-op */
6431 if (IN_SET(current, FREEZER_RUNNING, FREEZER_FREEZING, FREEZER_FROZEN))
6432 next = current;
6433 else
6434 next = FREEZER_THAWING;
6435 break;
6436
6437 default:
6438 assert_not_reached();
6439 }
6440
6441 objective = freezer_state_objective(next);
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 }