]> git.ipfire.org Git - thirdparty/systemd.git/blame_incremental - src/core/unit.c
Merge pull request #33154 from yuwata/test-async
[thirdparty/systemd.git] / src / core / unit.c
... / ...
CommitLineData
1/* SPDX-License-Identifier: LGPL-2.1-or-later */
2
3#include <errno.h>
4#include <stdlib.h>
5#include <sys/prctl.h>
6#include <unistd.h>
7
8#include "sd-id128.h"
9#include "sd-messages.h"
10
11#include "all-units.h"
12#include "alloc-util.h"
13#include "bpf-firewall.h"
14#include "bpf-foreign.h"
15#include "bpf-socket-bind.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 "core-varlink.h"
23#include "dbus-unit.h"
24#include "dbus.h"
25#include "dropin.h"
26#include "env-util.h"
27#include "escape.h"
28#include "exec-credential.h"
29#include "execute.h"
30#include "fd-util.h"
31#include "fileio-label.h"
32#include "fileio.h"
33#include "format-util.h"
34#include "id128-util.h"
35#include "install.h"
36#include "iovec-util.h"
37#include "label-util.h"
38#include "load-dropin.h"
39#include "load-fragment.h"
40#include "log.h"
41#include "logarithm.h"
42#include "macro.h"
43#include "mkdir-label.h"
44#include "path-util.h"
45#include "process-util.h"
46#include "rm-rf.h"
47#include "serialize.h"
48#include "set.h"
49#include "signal-util.h"
50#include "sparse-endian.h"
51#include "special.h"
52#include "specifier.h"
53#include "stat-util.h"
54#include "stdio-util.h"
55#include "string-table.h"
56#include "string-util.h"
57#include "strv.h"
58#include "terminal-util.h"
59#include "tmpfile-util.h"
60#include "umask-util.h"
61#include "unit-name.h"
62#include "unit.h"
63#include "user-util.h"
64#include "virt.h"
65#if BPF_FRAMEWORK
66#include "bpf-link.h"
67#endif
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
81const 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
95Unit* 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 = -1;
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 = (const RateLimit) {
122 m->defaults.start_limit_interval,
123 m->defaults.start_limit_burst,
124 };
125
126 u->auto_start_stop_ratelimit = (const RateLimit) {
127 .interval = 10 * USEC_PER_SEC,
128 .burst = 16
129 };
130
131 unit_reset_memory_accounting_last(u);
132 unit_reset_io_accounting_last(u);
133
134 return u;
135}
136
137int unit_new_for_name(Manager *m, size_t size, const char *name, Unit **ret) {
138 _cleanup_(unit_freep) Unit *u = NULL;
139 int r;
140
141 u = unit_new(m, size);
142 if (!u)
143 return -ENOMEM;
144
145 r = unit_add_name(u, name);
146 if (r < 0)
147 return r;
148
149 *ret = TAKE_PTR(u);
150
151 return r;
152}
153
154bool unit_has_name(const Unit *u, const char *name) {
155 assert(u);
156 assert(name);
157
158 return streq_ptr(name, u->id) ||
159 set_contains(u->aliases, name);
160}
161
162static void unit_init(Unit *u) {
163 CGroupContext *cc;
164 ExecContext *ec;
165 KillContext *kc;
166
167 assert(u);
168 assert(u->manager);
169 assert(u->type >= 0);
170
171 cc = unit_get_cgroup_context(u);
172 if (cc) {
173 cgroup_context_init(cc);
174
175 /* Copy in the manager defaults into the cgroup
176 * context, _before_ the rest of the settings have
177 * been initialized */
178
179 cc->cpu_accounting = u->manager->defaults.cpu_accounting;
180 cc->io_accounting = u->manager->defaults.io_accounting;
181 cc->blockio_accounting = u->manager->defaults.blockio_accounting;
182 cc->memory_accounting = u->manager->defaults.memory_accounting;
183 cc->tasks_accounting = u->manager->defaults.tasks_accounting;
184 cc->ip_accounting = u->manager->defaults.ip_accounting;
185
186 if (u->type != UNIT_SLICE)
187 cc->tasks_max = u->manager->defaults.tasks_max;
188
189 cc->memory_pressure_watch = u->manager->defaults.memory_pressure_watch;
190 cc->memory_pressure_threshold_usec = u->manager->defaults.memory_pressure_threshold_usec;
191 }
192
193 ec = unit_get_exec_context(u);
194 if (ec) {
195 exec_context_init(ec);
196
197 if (u->manager->defaults.oom_score_adjust_set) {
198 ec->oom_score_adjust = u->manager->defaults.oom_score_adjust;
199 ec->oom_score_adjust_set = true;
200 }
201
202 if (MANAGER_IS_SYSTEM(u->manager))
203 ec->keyring_mode = EXEC_KEYRING_SHARED;
204 else {
205 ec->keyring_mode = EXEC_KEYRING_INHERIT;
206
207 /* User manager might have its umask redefined by PAM or UMask=. In this
208 * case let the units it manages inherit this value by default. They can
209 * still tune this value through their own unit file */
210 (void) get_process_umask(0, &ec->umask);
211 }
212 }
213
214 kc = unit_get_kill_context(u);
215 if (kc)
216 kill_context_init(kc);
217
218 if (UNIT_VTABLE(u)->init)
219 UNIT_VTABLE(u)->init(u);
220}
221
222static int unit_add_alias(Unit *u, char *donated_name) {
223 int r;
224
225 /* Make sure that u->names is allocated. We may leave u->names
226 * empty if we fail later, but this is not a problem. */
227 r = set_ensure_put(&u->aliases, &string_hash_ops, donated_name);
228 if (r < 0)
229 return r;
230 assert(r > 0);
231
232 return 0;
233}
234
235int unit_add_name(Unit *u, const char *text) {
236 _cleanup_free_ char *name = NULL, *instance = NULL;
237 UnitType t;
238 int r;
239
240 assert(u);
241 assert(text);
242
243 if (unit_name_is_valid(text, UNIT_NAME_TEMPLATE)) {
244 if (!u->instance)
245 return log_unit_debug_errno(u, SYNTHETIC_ERRNO(EINVAL),
246 "Instance is not set when adding name '%s'.", text);
247
248 r = unit_name_replace_instance(text, u->instance, &name);
249 if (r < 0)
250 return log_unit_debug_errno(u, r,
251 "Failed to build instance name from '%s': %m", text);
252 } else {
253 name = strdup(text);
254 if (!name)
255 return -ENOMEM;
256 }
257
258 if (unit_has_name(u, name))
259 return 0;
260
261 if (hashmap_contains(u->manager->units, name))
262 return log_unit_debug_errno(u, SYNTHETIC_ERRNO(EEXIST),
263 "Unit already exist when adding name '%s'.", name);
264
265 if (!unit_name_is_valid(name, UNIT_NAME_PLAIN|UNIT_NAME_INSTANCE))
266 return log_unit_debug_errno(u, SYNTHETIC_ERRNO(EINVAL),
267 "Name '%s' is invalid.", name);
268
269 t = unit_name_to_type(name);
270 if (t < 0)
271 return log_unit_debug_errno(u, SYNTHETIC_ERRNO(EINVAL),
272 "failed to derive unit type from name '%s'.", name);
273
274 if (u->type != _UNIT_TYPE_INVALID && t != u->type)
275 return log_unit_debug_errno(u, SYNTHETIC_ERRNO(EINVAL),
276 "Unit type is illegal: u->type(%d) and t(%d) for name '%s'.",
277 u->type, t, name);
278
279 r = unit_name_to_instance(name, &instance);
280 if (r < 0)
281 return log_unit_debug_errno(u, r, "Failed to extract instance from name '%s': %m", name);
282
283 if (instance && !unit_type_may_template(t))
284 return log_unit_debug_errno(u, SYNTHETIC_ERRNO(EINVAL), "Templates are not allowed for name '%s'.", name);
285
286 /* Ensure that this unit either has no instance, or that the instance matches. */
287 if (u->type != _UNIT_TYPE_INVALID && !streq_ptr(u->instance, instance))
288 return log_unit_debug_errno(u, SYNTHETIC_ERRNO(EINVAL),
289 "Cannot add name %s, the instances don't match (\"%s\" != \"%s\").",
290 name, instance, u->instance);
291
292 if (u->id && !unit_type_may_alias(t))
293 return log_unit_debug_errno(u, SYNTHETIC_ERRNO(EEXIST),
294 "Cannot add name %s, aliases are not allowed for %s units.",
295 name, unit_type_to_string(t));
296
297 if (hashmap_size(u->manager->units) >= MANAGER_MAX_NAMES)
298 return log_unit_warning_errno(u, SYNTHETIC_ERRNO(E2BIG), "Cannot add name, manager has too many units.");
299
300 /* Add name to the global hashmap first, because that's easier to undo */
301 r = hashmap_put(u->manager->units, name, u);
302 if (r < 0)
303 return log_unit_debug_errno(u, r, "Add unit to hashmap failed for name '%s': %m", text);
304
305 if (u->id) {
306 r = unit_add_alias(u, name); /* unit_add_alias() takes ownership of the name on success */
307 if (r < 0) {
308 hashmap_remove(u->manager->units, name);
309 return r;
310 }
311 TAKE_PTR(name);
312
313 } else {
314 /* A new name, we don't need the set yet. */
315 assert(u->type == _UNIT_TYPE_INVALID);
316 assert(!u->instance);
317
318 u->type = t;
319 u->id = TAKE_PTR(name);
320 u->instance = TAKE_PTR(instance);
321
322 LIST_PREPEND(units_by_type, u->manager->units_by_type[t], u);
323 unit_init(u);
324 }
325
326 unit_add_to_dbus_queue(u);
327 return 0;
328}
329
330int unit_choose_id(Unit *u, const char *name) {
331 _cleanup_free_ char *t = NULL;
332 char *s;
333 int r;
334
335 assert(u);
336 assert(name);
337
338 if (unit_name_is_valid(name, UNIT_NAME_TEMPLATE)) {
339 if (!u->instance)
340 return -EINVAL;
341
342 r = unit_name_replace_instance(name, u->instance, &t);
343 if (r < 0)
344 return r;
345
346 name = t;
347 }
348
349 if (streq_ptr(u->id, name))
350 return 0; /* Nothing to do. */
351
352 /* Selects one of the aliases of this unit as the id */
353 s = set_get(u->aliases, (char*) name);
354 if (!s)
355 return -ENOENT;
356
357 if (u->id) {
358 r = set_remove_and_put(u->aliases, name, u->id);
359 if (r < 0)
360 return r;
361 } else
362 assert_se(set_remove(u->aliases, name)); /* see set_get() above… */
363
364 u->id = s; /* Old u->id is now stored in the set, and s is not stored anywhere */
365 unit_add_to_dbus_queue(u);
366
367 return 0;
368}
369
370int unit_set_description(Unit *u, const char *description) {
371 int r;
372
373 assert(u);
374
375 r = free_and_strdup(&u->description, empty_to_null(description));
376 if (r < 0)
377 return r;
378 if (r > 0)
379 unit_add_to_dbus_queue(u);
380
381 return 0;
382}
383
384static bool unit_success_failure_handler_has_jobs(Unit *unit) {
385 Unit *other;
386
387 UNIT_FOREACH_DEPENDENCY(other, unit, UNIT_ATOM_ON_SUCCESS)
388 if (other->job || other->nop_job)
389 return true;
390
391 UNIT_FOREACH_DEPENDENCY(other, unit, UNIT_ATOM_ON_FAILURE)
392 if (other->job || other->nop_job)
393 return true;
394
395 return false;
396}
397
398void unit_release_resources(Unit *u) {
399 UnitActiveState state;
400 ExecContext *ec;
401
402 assert(u);
403
404 if (u->job || u->nop_job)
405 return;
406
407 if (u->perpetual)
408 return;
409
410 state = unit_active_state(u);
411 if (!IN_SET(state, UNIT_INACTIVE, UNIT_FAILED))
412 return;
413
414 if (unit_will_restart(u))
415 return;
416
417 ec = unit_get_exec_context(u);
418 if (ec && ec->runtime_directory_preserve_mode == EXEC_PRESERVE_RESTART)
419 exec_context_destroy_runtime_directory(ec, u->manager->prefix[EXEC_DIRECTORY_RUNTIME]);
420
421 if (UNIT_VTABLE(u)->release_resources)
422 UNIT_VTABLE(u)->release_resources(u);
423}
424
425bool unit_may_gc(Unit *u) {
426 UnitActiveState state;
427 int r;
428
429 assert(u);
430
431 /* Checks whether the unit is ready to be unloaded for garbage collection. Returns true when the
432 * unit may be collected, and false if there's some reason to keep it loaded.
433 *
434 * References from other units are *not* checked here. Instead, this is done in unit_gc_sweep(), but
435 * using markers to properly collect dependency loops.
436 */
437
438 if (u->job || u->nop_job)
439 return false;
440
441 if (u->perpetual)
442 return false;
443
444 /* if we saw a cgroup empty event for this unit, stay around until we processed it so that we remove
445 * the empty cgroup if possible. Similar, process any pending OOM events if they are already queued
446 * before we release the unit. */
447 if (u->in_cgroup_empty_queue || u->in_cgroup_oom_queue)
448 return false;
449
450 /* Make sure to send out D-Bus events before we unload the unit */
451 if (u->in_dbus_queue)
452 return false;
453
454 if (sd_bus_track_count(u->bus_track) > 0)
455 return false;
456
457 state = unit_active_state(u);
458
459 /* But we keep the unit object around for longer when it is referenced or configured to not be
460 * gc'ed */
461 switch (u->collect_mode) {
462
463 case COLLECT_INACTIVE:
464 if (state != UNIT_INACTIVE)
465 return false;
466
467 break;
468
469 case COLLECT_INACTIVE_OR_FAILED:
470 if (!UNIT_IS_INACTIVE_OR_FAILED(state))
471 return false;
472
473 break;
474
475 default:
476 assert_not_reached();
477 }
478
479 /* Check if any OnFailure= or on Success= jobs may be pending */
480 if (unit_success_failure_handler_has_jobs(u))
481 return false;
482
483 /* If the unit has a cgroup, then check whether there's anything in it. If so, we should stay
484 * around. Units with active processes should never be collected. */
485 r = unit_cgroup_is_empty(u);
486 if (r <= 0 && r != -ENXIO)
487 return false; /* ENXIO means: currently not realized */
488
489 if (!UNIT_VTABLE(u)->may_gc)
490 return true;
491
492 return UNIT_VTABLE(u)->may_gc(u);
493}
494
495void unit_add_to_load_queue(Unit *u) {
496 assert(u);
497 assert(u->type != _UNIT_TYPE_INVALID);
498
499 if (u->load_state != UNIT_STUB || u->in_load_queue)
500 return;
501
502 LIST_PREPEND(load_queue, u->manager->load_queue, u);
503 u->in_load_queue = true;
504}
505
506void unit_add_to_cleanup_queue(Unit *u) {
507 assert(u);
508
509 if (u->in_cleanup_queue)
510 return;
511
512 LIST_PREPEND(cleanup_queue, u->manager->cleanup_queue, u);
513 u->in_cleanup_queue = true;
514}
515
516void unit_add_to_gc_queue(Unit *u) {
517 assert(u);
518
519 if (u->in_gc_queue || u->in_cleanup_queue)
520 return;
521
522 if (!unit_may_gc(u))
523 return;
524
525 LIST_PREPEND(gc_queue, u->manager->gc_unit_queue, u);
526 u->in_gc_queue = true;
527}
528
529void unit_add_to_dbus_queue(Unit *u) {
530 assert(u);
531 assert(u->type != _UNIT_TYPE_INVALID);
532
533 if (u->load_state == UNIT_STUB || u->in_dbus_queue)
534 return;
535
536 /* Shortcut things if nobody cares */
537 if (sd_bus_track_count(u->manager->subscribed) <= 0 &&
538 sd_bus_track_count(u->bus_track) <= 0 &&
539 set_isempty(u->manager->private_buses)) {
540 u->sent_dbus_new_signal = true;
541 return;
542 }
543
544 LIST_PREPEND(dbus_queue, u->manager->dbus_unit_queue, u);
545 u->in_dbus_queue = true;
546}
547
548void unit_submit_to_stop_when_unneeded_queue(Unit *u) {
549 assert(u);
550
551 if (u->in_stop_when_unneeded_queue)
552 return;
553
554 if (!u->stop_when_unneeded)
555 return;
556
557 if (!UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u)))
558 return;
559
560 LIST_PREPEND(stop_when_unneeded_queue, u->manager->stop_when_unneeded_queue, u);
561 u->in_stop_when_unneeded_queue = true;
562}
563
564void unit_submit_to_start_when_upheld_queue(Unit *u) {
565 assert(u);
566
567 if (u->in_start_when_upheld_queue)
568 return;
569
570 if (!UNIT_IS_INACTIVE_OR_FAILED(unit_active_state(u)))
571 return;
572
573 if (!unit_has_dependency(u, UNIT_ATOM_START_STEADILY, NULL))
574 return;
575
576 LIST_PREPEND(start_when_upheld_queue, u->manager->start_when_upheld_queue, u);
577 u->in_start_when_upheld_queue = true;
578}
579
580void unit_submit_to_stop_when_bound_queue(Unit *u) {
581 assert(u);
582
583 if (u->in_stop_when_bound_queue)
584 return;
585
586 if (!UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u)))
587 return;
588
589 if (!unit_has_dependency(u, UNIT_ATOM_CANNOT_BE_ACTIVE_WITHOUT, NULL))
590 return;
591
592 LIST_PREPEND(stop_when_bound_queue, u->manager->stop_when_bound_queue, u);
593 u->in_stop_when_bound_queue = true;
594}
595
596static bool unit_can_release_resources(Unit *u) {
597 ExecContext *ec;
598
599 assert(u);
600
601 if (UNIT_VTABLE(u)->release_resources)
602 return true;
603
604 ec = unit_get_exec_context(u);
605 if (ec && ec->runtime_directory_preserve_mode == EXEC_PRESERVE_RESTART)
606 return true;
607
608 return false;
609}
610
611void unit_submit_to_release_resources_queue(Unit *u) {
612 assert(u);
613
614 if (u->in_release_resources_queue)
615 return;
616
617 if (u->job || u->nop_job)
618 return;
619
620 if (u->perpetual)
621 return;
622
623 if (!unit_can_release_resources(u))
624 return;
625
626 LIST_PREPEND(release_resources_queue, u->manager->release_resources_queue, u);
627 u->in_release_resources_queue = true;
628}
629
630static void unit_clear_dependencies(Unit *u) {
631 assert(u);
632
633 /* Removes all dependencies configured on u and their reverse dependencies. */
634
635 for (Hashmap *deps; (deps = hashmap_steal_first(u->dependencies));) {
636
637 for (Unit *other; (other = hashmap_steal_first_key(deps));) {
638 Hashmap *other_deps;
639
640 HASHMAP_FOREACH(other_deps, other->dependencies)
641 hashmap_remove(other_deps, u);
642
643 unit_add_to_gc_queue(other);
644 }
645
646 hashmap_free(deps);
647 }
648
649 u->dependencies = hashmap_free(u->dependencies);
650}
651
652static void unit_remove_transient(Unit *u) {
653 assert(u);
654
655 if (!u->transient)
656 return;
657
658 if (u->fragment_path)
659 (void) unlink(u->fragment_path);
660
661 STRV_FOREACH(i, u->dropin_paths) {
662 _cleanup_free_ char *p = NULL, *pp = NULL;
663
664 if (path_extract_directory(*i, &p) < 0) /* Get the drop-in directory from the drop-in file */
665 continue;
666
667 if (path_extract_directory(p, &pp) < 0) /* Get the config directory from the drop-in directory */
668 continue;
669
670 /* Only drop transient drop-ins */
671 if (!path_equal(u->manager->lookup_paths.transient, pp))
672 continue;
673
674 (void) unlink(*i);
675 (void) rmdir(p);
676 }
677}
678
679static void unit_free_mounts_for(Unit *u) {
680 assert(u);
681
682 for (UnitMountDependencyType t = 0; t < _UNIT_MOUNT_DEPENDENCY_TYPE_MAX; ++t) {
683 for (;;) {
684 _cleanup_free_ char *path = NULL;
685
686 path = hashmap_steal_first_key(u->mounts_for[t]);
687 if (!path)
688 break;
689
690 char s[strlen(path) + 1];
691
692 PATH_FOREACH_PREFIX_MORE(s, path) {
693 char *y;
694 Set *x;
695
696 x = hashmap_get2(u->manager->units_needing_mounts_for[t], s, (void**) &y);
697 if (!x)
698 continue;
699
700 (void) set_remove(x, u);
701
702 if (set_isempty(x)) {
703 assert_se(hashmap_remove(u->manager->units_needing_mounts_for[t], y));
704 free(y);
705 set_free(x);
706 }
707 }
708 }
709
710 u->mounts_for[t] = hashmap_free(u->mounts_for[t]);
711 }
712}
713
714static void unit_done(Unit *u) {
715 ExecContext *ec;
716 CGroupContext *cc;
717
718 assert(u);
719
720 if (u->type < 0)
721 return;
722
723 if (UNIT_VTABLE(u)->done)
724 UNIT_VTABLE(u)->done(u);
725
726 ec = unit_get_exec_context(u);
727 if (ec)
728 exec_context_done(ec);
729
730 cc = unit_get_cgroup_context(u);
731 if (cc)
732 cgroup_context_done(cc);
733}
734
735Unit* unit_free(Unit *u) {
736 Unit *slice;
737 char *t;
738
739 if (!u)
740 return NULL;
741
742 sd_event_source_disable_unref(u->auto_start_stop_event_source);
743
744 u->transient_file = safe_fclose(u->transient_file);
745
746 if (!MANAGER_IS_RELOADING(u->manager))
747 unit_remove_transient(u);
748
749 bus_unit_send_removed_signal(u);
750
751 unit_done(u);
752
753 unit_dequeue_rewatch_pids(u);
754
755 u->match_bus_slot = sd_bus_slot_unref(u->match_bus_slot);
756 u->bus_track = sd_bus_track_unref(u->bus_track);
757 u->deserialized_refs = strv_free(u->deserialized_refs);
758 u->pending_freezer_invocation = sd_bus_message_unref(u->pending_freezer_invocation);
759
760 unit_free_mounts_for(u);
761
762 SET_FOREACH(t, u->aliases)
763 hashmap_remove_value(u->manager->units, t, u);
764 if (u->id)
765 hashmap_remove_value(u->manager->units, u->id, u);
766
767 if (!sd_id128_is_null(u->invocation_id))
768 hashmap_remove_value(u->manager->units_by_invocation_id, &u->invocation_id, u);
769
770 if (u->job) {
771 Job *j = u->job;
772 job_uninstall(j);
773 job_free(j);
774 }
775
776 if (u->nop_job) {
777 Job *j = u->nop_job;
778 job_uninstall(j);
779 job_free(j);
780 }
781
782 /* A unit is being dropped from the tree, make sure our family is realized properly. Do this after we
783 * detach the unit from slice tree in order to eliminate its effect on controller masks. */
784 slice = UNIT_GET_SLICE(u);
785 unit_clear_dependencies(u);
786 if (slice)
787 unit_add_family_to_cgroup_realize_queue(slice);
788
789 if (u->on_console)
790 manager_unref_console(u->manager);
791
792 unit_release_cgroup(u);
793
794 if (!MANAGER_IS_RELOADING(u->manager))
795 unit_unlink_state_files(u);
796
797 unit_unref_uid_gid(u, false);
798
799 (void) manager_update_failed_units(u->manager, u, false);
800 set_remove(u->manager->startup_units, u);
801
802 unit_unwatch_all_pids(u);
803
804 while (u->refs_by_target)
805 unit_ref_unset(u->refs_by_target);
806
807 if (u->type != _UNIT_TYPE_INVALID)
808 LIST_REMOVE(units_by_type, u->manager->units_by_type[u->type], u);
809
810 if (u->in_load_queue)
811 LIST_REMOVE(load_queue, u->manager->load_queue, u);
812
813 if (u->in_dbus_queue)
814 LIST_REMOVE(dbus_queue, u->manager->dbus_unit_queue, u);
815
816 if (u->in_cleanup_queue)
817 LIST_REMOVE(cleanup_queue, u->manager->cleanup_queue, u);
818
819 if (u->in_gc_queue)
820 LIST_REMOVE(gc_queue, u->manager->gc_unit_queue, u);
821
822 if (u->in_cgroup_realize_queue)
823 LIST_REMOVE(cgroup_realize_queue, u->manager->cgroup_realize_queue, u);
824
825 if (u->in_cgroup_empty_queue)
826 LIST_REMOVE(cgroup_empty_queue, u->manager->cgroup_empty_queue, u);
827
828 if (u->in_cgroup_oom_queue)
829 LIST_REMOVE(cgroup_oom_queue, u->manager->cgroup_oom_queue, u);
830
831 if (u->in_target_deps_queue)
832 LIST_REMOVE(target_deps_queue, u->manager->target_deps_queue, u);
833
834 if (u->in_stop_when_unneeded_queue)
835 LIST_REMOVE(stop_when_unneeded_queue, u->manager->stop_when_unneeded_queue, u);
836
837 if (u->in_start_when_upheld_queue)
838 LIST_REMOVE(start_when_upheld_queue, u->manager->start_when_upheld_queue, u);
839
840 if (u->in_stop_when_bound_queue)
841 LIST_REMOVE(stop_when_bound_queue, u->manager->stop_when_bound_queue, u);
842
843 if (u->in_release_resources_queue)
844 LIST_REMOVE(release_resources_queue, u->manager->release_resources_queue, u);
845
846 bpf_firewall_close(u);
847
848 condition_free_list(u->conditions);
849 condition_free_list(u->asserts);
850
851 free(u->description);
852 strv_free(u->documentation);
853 free(u->fragment_path);
854 free(u->source_path);
855 strv_free(u->dropin_paths);
856 free(u->instance);
857
858 free(u->job_timeout_reboot_arg);
859 free(u->reboot_arg);
860
861 free(u->access_selinux_context);
862
863 set_free_free(u->aliases);
864 free(u->id);
865
866 activation_details_unref(u->activation_details);
867
868 return mfree(u);
869}
870
871FreezerState unit_freezer_state(Unit *u) {
872 assert(u);
873
874 return u->freezer_state;
875}
876
877UnitActiveState unit_active_state(Unit *u) {
878 assert(u);
879
880 if (u->load_state == UNIT_MERGED)
881 return unit_active_state(unit_follow_merge(u));
882
883 /* After a reload it might happen that a unit is not correctly
884 * loaded but still has a process around. That's why we won't
885 * shortcut failed loading to UNIT_INACTIVE_FAILED. */
886
887 return UNIT_VTABLE(u)->active_state(u);
888}
889
890const char* unit_sub_state_to_string(Unit *u) {
891 assert(u);
892
893 return UNIT_VTABLE(u)->sub_state_to_string(u);
894}
895
896static int unit_merge_names(Unit *u, Unit *other) {
897 char *name;
898 int r;
899
900 assert(u);
901 assert(other);
902
903 r = unit_add_alias(u, other->id);
904 if (r < 0)
905 return r;
906
907 r = set_move(u->aliases, other->aliases);
908 if (r < 0) {
909 set_remove(u->aliases, other->id);
910 return r;
911 }
912
913 TAKE_PTR(other->id);
914 other->aliases = set_free_free(other->aliases);
915
916 SET_FOREACH(name, u->aliases)
917 assert_se(hashmap_replace(u->manager->units, name, u) == 0);
918
919 return 0;
920}
921
922static int unit_reserve_dependencies(Unit *u, Unit *other) {
923 size_t n_reserve;
924 Hashmap* deps;
925 void *d;
926 int r;
927
928 assert(u);
929 assert(other);
930
931 /* Let's reserve some space in the dependency hashmaps so that later on merging the units cannot
932 * fail.
933 *
934 * First make some room in the per dependency type hashmaps. Using the summed size of both units'
935 * hashmaps is an estimate that is likely too high since they probably use some of the same
936 * types. But it's never too low, and that's all we need. */
937
938 n_reserve = MIN(hashmap_size(other->dependencies), LESS_BY((size_t) _UNIT_DEPENDENCY_MAX, hashmap_size(u->dependencies)));
939 if (n_reserve > 0) {
940 r = hashmap_ensure_allocated(&u->dependencies, NULL);
941 if (r < 0)
942 return r;
943
944 r = hashmap_reserve(u->dependencies, n_reserve);
945 if (r < 0)
946 return r;
947 }
948
949 /* Now, enlarge our per dependency type hashmaps by the number of entries in the same hashmap of the
950 * other unit's dependencies.
951 *
952 * NB: If u does not have a dependency set allocated for some dependency type, there is no need to
953 * reserve anything for. In that case other's set will be transferred as a whole to u by
954 * complete_move(). */
955
956 HASHMAP_FOREACH_KEY(deps, d, u->dependencies) {
957 Hashmap *other_deps;
958
959 other_deps = hashmap_get(other->dependencies, d);
960
961 r = hashmap_reserve(deps, hashmap_size(other_deps));
962 if (r < 0)
963 return r;
964 }
965
966 return 0;
967}
968
969static bool unit_should_warn_about_dependency(UnitDependency dependency) {
970 /* Only warn about some unit types */
971 return IN_SET(dependency,
972 UNIT_CONFLICTS,
973 UNIT_CONFLICTED_BY,
974 UNIT_BEFORE,
975 UNIT_AFTER,
976 UNIT_ON_SUCCESS,
977 UNIT_ON_FAILURE,
978 UNIT_TRIGGERS,
979 UNIT_TRIGGERED_BY);
980}
981
982static int unit_per_dependency_type_hashmap_update(
983 Hashmap *per_type,
984 Unit *other,
985 UnitDependencyMask origin_mask,
986 UnitDependencyMask destination_mask) {
987
988 UnitDependencyInfo info;
989 int r;
990
991 assert(other);
992 assert_cc(sizeof(void*) == sizeof(info));
993
994 /* Acquire the UnitDependencyInfo entry for the Unit* we are interested in, and update it if it
995 * exists, or insert it anew if not. */
996
997 info.data = hashmap_get(per_type, other);
998 if (info.data) {
999 /* Entry already exists. Add in our mask. */
1000
1001 if (FLAGS_SET(origin_mask, info.origin_mask) &&
1002 FLAGS_SET(destination_mask, info.destination_mask))
1003 return 0; /* NOP */
1004
1005 info.origin_mask |= origin_mask;
1006 info.destination_mask |= destination_mask;
1007
1008 r = hashmap_update(per_type, other, info.data);
1009 } else {
1010 info = (UnitDependencyInfo) {
1011 .origin_mask = origin_mask,
1012 .destination_mask = destination_mask,
1013 };
1014
1015 r = hashmap_put(per_type, other, info.data);
1016 }
1017 if (r < 0)
1018 return r;
1019
1020 return 1;
1021}
1022
1023static void unit_merge_dependencies(Unit *u, Unit *other) {
1024 Hashmap *deps;
1025 void *dt; /* Actually of type UnitDependency, except that we don't bother casting it here,
1026 * since the hashmaps all want it as void pointer. */
1027
1028 assert(u);
1029 assert(other);
1030
1031 if (u == other)
1032 return;
1033
1034 /* First, remove dependency to other. */
1035 HASHMAP_FOREACH_KEY(deps, dt, u->dependencies) {
1036 if (hashmap_remove(deps, other) && unit_should_warn_about_dependency(UNIT_DEPENDENCY_FROM_PTR(dt)))
1037 log_unit_warning(u, "Dependency %s=%s is dropped, as %s is merged into %s.",
1038 unit_dependency_to_string(UNIT_DEPENDENCY_FROM_PTR(dt)),
1039 other->id, other->id, u->id);
1040
1041 if (hashmap_isempty(deps))
1042 hashmap_free(hashmap_remove(u->dependencies, dt));
1043 }
1044
1045 for (;;) {
1046 _cleanup_hashmap_free_ Hashmap *other_deps = NULL;
1047 UnitDependencyInfo di_back;
1048 Unit *back;
1049
1050 /* Let's focus on one dependency type at a time, that 'other' has defined. */
1051 other_deps = hashmap_steal_first_key_and_value(other->dependencies, &dt);
1052 if (!other_deps)
1053 break; /* done! */
1054
1055 deps = hashmap_get(u->dependencies, dt);
1056
1057 /* Now iterate through all dependencies of this dependency type, of 'other'. We refer to the
1058 * referenced units as 'back'. */
1059 HASHMAP_FOREACH_KEY(di_back.data, back, other_deps) {
1060 Hashmap *back_deps;
1061 void *back_dt;
1062
1063 if (back == u) {
1064 /* This is a dependency pointing back to the unit we want to merge with?
1065 * Suppress it (but warn) */
1066 if (unit_should_warn_about_dependency(UNIT_DEPENDENCY_FROM_PTR(dt)))
1067 log_unit_warning(u, "Dependency %s=%s in %s is dropped, as %s is merged into %s.",
1068 unit_dependency_to_string(UNIT_DEPENDENCY_FROM_PTR(dt)),
1069 u->id, other->id, other->id, u->id);
1070
1071 hashmap_remove(other_deps, back);
1072 continue;
1073 }
1074
1075 /* Now iterate through all deps of 'back', and fix the ones pointing to 'other' to
1076 * point to 'u' instead. */
1077 HASHMAP_FOREACH_KEY(back_deps, back_dt, back->dependencies) {
1078 UnitDependencyInfo di_move;
1079
1080 di_move.data = hashmap_remove(back_deps, other);
1081 if (!di_move.data)
1082 continue;
1083
1084 assert_se(unit_per_dependency_type_hashmap_update(
1085 back_deps,
1086 u,
1087 di_move.origin_mask,
1088 di_move.destination_mask) >= 0);
1089 }
1090
1091 /* The target unit already has dependencies of this type, let's then merge this individually. */
1092 if (deps)
1093 assert_se(unit_per_dependency_type_hashmap_update(
1094 deps,
1095 back,
1096 di_back.origin_mask,
1097 di_back.destination_mask) >= 0);
1098 }
1099
1100 /* Now all references towards 'other' of the current type 'dt' are corrected to point to 'u'.
1101 * Lets's now move the deps of type 'dt' from 'other' to 'u'. If the unit does not have
1102 * dependencies of this type, let's move them per type wholesale. */
1103 if (!deps)
1104 assert_se(hashmap_put(u->dependencies, dt, TAKE_PTR(other_deps)) >= 0);
1105 }
1106
1107 other->dependencies = hashmap_free(other->dependencies);
1108}
1109
1110int unit_merge(Unit *u, Unit *other) {
1111 int r;
1112
1113 assert(u);
1114 assert(other);
1115 assert(u->manager == other->manager);
1116 assert(u->type != _UNIT_TYPE_INVALID);
1117
1118 other = unit_follow_merge(other);
1119
1120 if (other == u)
1121 return 0;
1122
1123 if (u->type != other->type)
1124 return -EINVAL;
1125
1126 if (!unit_type_may_alias(u->type)) /* Merging only applies to unit names that support aliases */
1127 return -EEXIST;
1128
1129 if (!IN_SET(other->load_state, UNIT_STUB, UNIT_NOT_FOUND))
1130 return -EEXIST;
1131
1132 if (!streq_ptr(u->instance, other->instance))
1133 return -EINVAL;
1134
1135 if (other->job)
1136 return -EEXIST;
1137
1138 if (other->nop_job)
1139 return -EEXIST;
1140
1141 if (!UNIT_IS_INACTIVE_OR_FAILED(unit_active_state(other)))
1142 return -EEXIST;
1143
1144 /* Make reservations to ensure merge_dependencies() won't fail. We don't rollback reservations if we
1145 * fail. We don't have a way to undo reservations. A reservation is not a leak. */
1146 r = unit_reserve_dependencies(u, other);
1147 if (r < 0)
1148 return r;
1149
1150 /* Redirect all references */
1151 while (other->refs_by_target)
1152 unit_ref_set(other->refs_by_target, other->refs_by_target->source, u);
1153
1154 /* Merge dependencies */
1155 unit_merge_dependencies(u, other);
1156
1157 /* Merge names. It is better to do that after merging deps, otherwise the log message contains n/a. */
1158 r = unit_merge_names(u, other);
1159 if (r < 0)
1160 return r;
1161
1162 other->load_state = UNIT_MERGED;
1163 other->merged_into = u;
1164
1165 if (!u->activation_details)
1166 u->activation_details = activation_details_ref(other->activation_details);
1167
1168 /* If there is still some data attached to the other node, we
1169 * don't need it anymore, and can free it. */
1170 if (other->load_state != UNIT_STUB)
1171 if (UNIT_VTABLE(other)->done)
1172 UNIT_VTABLE(other)->done(other);
1173
1174 unit_add_to_dbus_queue(u);
1175 unit_add_to_cleanup_queue(other);
1176
1177 return 0;
1178}
1179
1180int unit_merge_by_name(Unit *u, const char *name) {
1181 _cleanup_free_ char *s = NULL;
1182 Unit *other;
1183 int r;
1184
1185 /* Either add name to u, or if a unit with name already exists, merge it with u.
1186 * If name is a template, do the same for name@instance, where instance is u's instance. */
1187
1188 assert(u);
1189 assert(name);
1190
1191 if (unit_name_is_valid(name, UNIT_NAME_TEMPLATE)) {
1192 if (!u->instance)
1193 return -EINVAL;
1194
1195 r = unit_name_replace_instance(name, u->instance, &s);
1196 if (r < 0)
1197 return r;
1198
1199 name = s;
1200 }
1201
1202 other = manager_get_unit(u->manager, name);
1203 if (other)
1204 return unit_merge(u, other);
1205
1206 return unit_add_name(u, name);
1207}
1208
1209Unit* unit_follow_merge(Unit *u) {
1210 assert(u);
1211
1212 while (u->load_state == UNIT_MERGED)
1213 assert_se(u = u->merged_into);
1214
1215 return u;
1216}
1217
1218int unit_add_exec_dependencies(Unit *u, ExecContext *c) {
1219 int r;
1220
1221 assert(u);
1222 assert(c);
1223
1224 /* Unlike unit_add_dependency() or friends, this always returns 0 on success. */
1225
1226 if (c->working_directory) {
1227 r = unit_add_mounts_for(
1228 u,
1229 c->working_directory,
1230 UNIT_DEPENDENCY_FILE,
1231 c->working_directory_missing_ok ? UNIT_MOUNT_WANTS : UNIT_MOUNT_REQUIRES);
1232 if (r < 0)
1233 return r;
1234 }
1235
1236 if (c->root_directory) {
1237 r = unit_add_mounts_for(u, c->root_directory, UNIT_DEPENDENCY_FILE, UNIT_MOUNT_WANTS);
1238 if (r < 0)
1239 return r;
1240 }
1241
1242 if (c->root_image) {
1243 r = unit_add_mounts_for(u, c->root_image, UNIT_DEPENDENCY_FILE, UNIT_MOUNT_WANTS);
1244 if (r < 0)
1245 return r;
1246 }
1247
1248 for (ExecDirectoryType dt = 0; dt < _EXEC_DIRECTORY_TYPE_MAX; dt++) {
1249 if (!u->manager->prefix[dt])
1250 continue;
1251
1252 FOREACH_ARRAY(i, c->directories[dt].items, c->directories[dt].n_items) {
1253 _cleanup_free_ char *p = NULL;
1254
1255 p = path_join(u->manager->prefix[dt], i->path);
1256 if (!p)
1257 return -ENOMEM;
1258
1259 r = unit_add_mounts_for(u, p, UNIT_DEPENDENCY_FILE, UNIT_MOUNT_REQUIRES);
1260 if (r < 0)
1261 return r;
1262 }
1263 }
1264
1265 if (!MANAGER_IS_SYSTEM(u->manager))
1266 return 0;
1267
1268 /* For the following three directory types we need write access, and /var/ is possibly on the root
1269 * fs. Hence order after systemd-remount-fs.service, to ensure things are writable. */
1270 if (c->directories[EXEC_DIRECTORY_STATE].n_items > 0 ||
1271 c->directories[EXEC_DIRECTORY_CACHE].n_items > 0 ||
1272 c->directories[EXEC_DIRECTORY_LOGS].n_items > 0) {
1273 r = unit_add_dependency_by_name(u, UNIT_AFTER, SPECIAL_REMOUNT_FS_SERVICE, true, UNIT_DEPENDENCY_FILE);
1274 if (r < 0)
1275 return r;
1276 }
1277
1278 if (c->private_tmp) {
1279 r = unit_add_mounts_for(u, "/tmp", UNIT_DEPENDENCY_FILE, UNIT_MOUNT_WANTS);
1280 if (r < 0)
1281 return r;
1282
1283 r = unit_add_mounts_for(u, "/var/tmp", UNIT_DEPENDENCY_FILE, UNIT_MOUNT_WANTS);
1284 if (r < 0)
1285 return r;
1286
1287 r = unit_add_dependency_by_name(u, UNIT_AFTER, SPECIAL_TMPFILES_SETUP_SERVICE, true, UNIT_DEPENDENCY_FILE);
1288 if (r < 0)
1289 return r;
1290 }
1291
1292 if (c->root_image) {
1293 /* We need to wait for /dev/loopX to appear when doing RootImage=, hence let's add an
1294 * implicit dependency on udev */
1295
1296 r = unit_add_dependency_by_name(u, UNIT_AFTER, SPECIAL_UDEVD_SERVICE, true, UNIT_DEPENDENCY_FILE);
1297 if (r < 0)
1298 return r;
1299 }
1300
1301 if (!IN_SET(c->std_output,
1302 EXEC_OUTPUT_JOURNAL, EXEC_OUTPUT_JOURNAL_AND_CONSOLE,
1303 EXEC_OUTPUT_KMSG, EXEC_OUTPUT_KMSG_AND_CONSOLE) &&
1304 !IN_SET(c->std_error,
1305 EXEC_OUTPUT_JOURNAL, EXEC_OUTPUT_JOURNAL_AND_CONSOLE,
1306 EXEC_OUTPUT_KMSG, EXEC_OUTPUT_KMSG_AND_CONSOLE) &&
1307 !c->log_namespace)
1308 return 0;
1309
1310 /* If syslog or kernel logging is requested (or log namespacing is), make sure our own logging daemon
1311 * is run first. */
1312
1313 if (c->log_namespace) {
1314 static const struct {
1315 const char *template;
1316 UnitType type;
1317 } deps[] = {
1318 { "systemd-journald", UNIT_SOCKET, },
1319 { "systemd-journald-varlink", UNIT_SOCKET, },
1320 { "systemd-journald-sync", UNIT_SERVICE, },
1321 };
1322
1323 FOREACH_ELEMENT(i, deps) {
1324 _cleanup_free_ char *unit = NULL;
1325
1326 r = unit_name_build_from_type(i->template, c->log_namespace, i->type, &unit);
1327 if (r < 0)
1328 return r;
1329
1330 r = unit_add_two_dependencies_by_name(u, UNIT_AFTER, UNIT_REQUIRES, unit, true, UNIT_DEPENDENCY_FILE);
1331 if (r < 0)
1332 return r;
1333 }
1334 } else {
1335 r = unit_add_dependency_by_name(u, UNIT_AFTER, SPECIAL_JOURNALD_SOCKET, true, UNIT_DEPENDENCY_FILE);
1336 if (r < 0)
1337 return r;
1338 }
1339
1340 r = unit_add_default_credential_dependencies(u, c);
1341 if (r < 0)
1342 return r;
1343
1344 return 0;
1345}
1346
1347const char* unit_description(Unit *u) {
1348 assert(u);
1349
1350 if (u->description)
1351 return u->description;
1352
1353 return strna(u->id);
1354}
1355
1356const char* unit_status_string(Unit *u, char **ret_combined_buffer) {
1357 assert(u);
1358 assert(u->id);
1359
1360 /* Return u->id, u->description, or "{u->id} - {u->description}".
1361 * Versions with u->description are only used if it is set.
1362 * The last option is used if configured and the caller provided the 'ret_combined_buffer'
1363 * pointer.
1364 *
1365 * Note that *ret_combined_buffer may be set to NULL. */
1366
1367 if (!u->description ||
1368 u->manager->status_unit_format == STATUS_UNIT_FORMAT_NAME ||
1369 (u->manager->status_unit_format == STATUS_UNIT_FORMAT_COMBINED && !ret_combined_buffer) ||
1370 streq(u->description, u->id)) {
1371
1372 if (ret_combined_buffer)
1373 *ret_combined_buffer = NULL;
1374 return u->id;
1375 }
1376
1377 if (ret_combined_buffer) {
1378 if (u->manager->status_unit_format == STATUS_UNIT_FORMAT_COMBINED) {
1379 *ret_combined_buffer = strjoin(u->id, " - ", u->description);
1380 if (*ret_combined_buffer)
1381 return *ret_combined_buffer;
1382 log_oom(); /* Fall back to ->description */
1383 } else
1384 *ret_combined_buffer = NULL;
1385 }
1386
1387 return u->description;
1388}
1389
1390/* Common implementation for multiple backends */
1391int unit_load_fragment_and_dropin(Unit *u, bool fragment_required) {
1392 int r;
1393
1394 assert(u);
1395
1396 /* Load a .{service,socket,...} file */
1397 r = unit_load_fragment(u);
1398 if (r < 0)
1399 return r;
1400
1401 if (u->load_state == UNIT_STUB) {
1402 if (fragment_required)
1403 return -ENOENT;
1404
1405 u->load_state = UNIT_LOADED;
1406 }
1407
1408 /* Load drop-in directory data. If u is an alias, we might be reloading the
1409 * target unit needlessly. But we cannot be sure which drops-ins have already
1410 * been loaded and which not, at least without doing complicated book-keeping,
1411 * so let's always reread all drop-ins. */
1412 r = unit_load_dropin(unit_follow_merge(u));
1413 if (r < 0)
1414 return r;
1415
1416 if (u->source_path) {
1417 struct stat st;
1418
1419 if (stat(u->source_path, &st) >= 0)
1420 u->source_mtime = timespec_load(&st.st_mtim);
1421 else
1422 u->source_mtime = 0;
1423 }
1424
1425 return 0;
1426}
1427
1428void unit_add_to_target_deps_queue(Unit *u) {
1429 Manager *m = ASSERT_PTR(ASSERT_PTR(u)->manager);
1430
1431 if (u->in_target_deps_queue)
1432 return;
1433
1434 LIST_PREPEND(target_deps_queue, m->target_deps_queue, u);
1435 u->in_target_deps_queue = true;
1436}
1437
1438int unit_add_default_target_dependency(Unit *u, Unit *target) {
1439 assert(u);
1440 assert(target);
1441
1442 if (target->type != UNIT_TARGET)
1443 return 0;
1444
1445 /* Only add the dependency if both units are loaded, so that
1446 * that loop check below is reliable */
1447 if (u->load_state != UNIT_LOADED ||
1448 target->load_state != UNIT_LOADED)
1449 return 0;
1450
1451 /* If either side wants no automatic dependencies, then let's
1452 * skip this */
1453 if (!u->default_dependencies ||
1454 !target->default_dependencies)
1455 return 0;
1456
1457 /* Don't create loops */
1458 if (unit_has_dependency(target, UNIT_ATOM_BEFORE, u))
1459 return 0;
1460
1461 return unit_add_dependency(target, UNIT_AFTER, u, true, UNIT_DEPENDENCY_DEFAULT);
1462}
1463
1464static int unit_add_slice_dependencies(Unit *u) {
1465 Unit *slice;
1466
1467 assert(u);
1468
1469 if (!UNIT_HAS_CGROUP_CONTEXT(u))
1470 return 0;
1471
1472 /* Slice units are implicitly ordered against their parent slices (as this relationship is encoded in the
1473 name), while all other units are ordered based on configuration (as in their case Slice= configures the
1474 relationship). */
1475 UnitDependencyMask mask = u->type == UNIT_SLICE ? UNIT_DEPENDENCY_IMPLICIT : UNIT_DEPENDENCY_FILE;
1476
1477 slice = UNIT_GET_SLICE(u);
1478 if (slice) {
1479 if (!IN_SET(slice->freezer_state, FREEZER_RUNNING, FREEZER_THAWING))
1480 u->freezer_state = FREEZER_FROZEN_BY_PARENT;
1481
1482 return unit_add_two_dependencies(u, UNIT_AFTER, UNIT_REQUIRES, slice, true, mask);
1483 }
1484
1485 if (unit_has_name(u, SPECIAL_ROOT_SLICE))
1486 return 0;
1487
1488 return unit_add_two_dependencies_by_name(u, UNIT_AFTER, UNIT_REQUIRES, SPECIAL_ROOT_SLICE, true, mask);
1489}
1490
1491static int unit_add_mount_dependencies(Unit *u) {
1492 bool changed = false;
1493 int r;
1494
1495 assert(u);
1496
1497 for (UnitMountDependencyType t = 0; t < _UNIT_MOUNT_DEPENDENCY_TYPE_MAX; ++t) {
1498 UnitDependencyInfo di;
1499 const char *path;
1500
1501 HASHMAP_FOREACH_KEY(di.data, path, u->mounts_for[t]) {
1502
1503 char prefix[strlen(ASSERT_PTR(path)) + 1];
1504
1505 PATH_FOREACH_PREFIX_MORE(prefix, path) {
1506 _cleanup_free_ char *p = NULL;
1507 Unit *m;
1508
1509 r = unit_name_from_path(prefix, ".mount", &p);
1510 if (r == -EINVAL)
1511 continue; /* If the path cannot be converted to a mount unit name,
1512 * then it's not manageable as a unit by systemd, and
1513 * hence we don't need a dependency on it. Let's thus
1514 * silently ignore the issue. */
1515 if (r < 0)
1516 return r;
1517
1518 m = manager_get_unit(u->manager, p);
1519 if (!m) {
1520 /* Make sure to load the mount unit if it exists. If so the
1521 * dependencies on this unit will be added later during the loading
1522 * of the mount unit. */
1523 (void) manager_load_unit_prepare(
1524 u->manager,
1525 p,
1526 /* path= */NULL,
1527 /* e= */NULL,
1528 &m);
1529 continue;
1530 }
1531 if (m == u)
1532 continue;
1533
1534 if (m->load_state != UNIT_LOADED)
1535 continue;
1536
1537 r = unit_add_dependency(
1538 u,
1539 UNIT_AFTER,
1540 m,
1541 /* add_reference= */ true,
1542 di.origin_mask);
1543 if (r < 0)
1544 return r;
1545 changed = changed || r > 0;
1546
1547 if (m->fragment_path) {
1548 r = unit_add_dependency(
1549 u,
1550 unit_mount_dependency_type_to_dependency_type(t),
1551 m,
1552 /* add_reference= */ true,
1553 di.origin_mask);
1554 if (r < 0)
1555 return r;
1556 changed = changed || r > 0;
1557 }
1558 }
1559 }
1560 }
1561
1562 return changed;
1563}
1564
1565static int unit_add_oomd_dependencies(Unit *u) {
1566 CGroupContext *c;
1567 CGroupMask mask;
1568 int r;
1569
1570 assert(u);
1571
1572 if (!u->default_dependencies)
1573 return 0;
1574
1575 c = unit_get_cgroup_context(u);
1576 if (!c)
1577 return 0;
1578
1579 bool wants_oomd = c->moom_swap == MANAGED_OOM_KILL || c->moom_mem_pressure == MANAGED_OOM_KILL;
1580 if (!wants_oomd)
1581 return 0;
1582
1583 if (!cg_all_unified())
1584 return 0;
1585
1586 r = cg_mask_supported(&mask);
1587 if (r < 0)
1588 return log_debug_errno(r, "Failed to determine supported controllers: %m");
1589
1590 if (!FLAGS_SET(mask, CGROUP_MASK_MEMORY))
1591 return 0;
1592
1593 return unit_add_two_dependencies_by_name(u, UNIT_AFTER, UNIT_WANTS, "systemd-oomd.service", true, UNIT_DEPENDENCY_FILE);
1594}
1595
1596static int unit_add_startup_units(Unit *u) {
1597 if (!unit_has_startup_cgroup_constraints(u))
1598 return 0;
1599
1600 return set_ensure_put(&u->manager->startup_units, NULL, u);
1601}
1602
1603static int unit_validate_on_failure_job_mode(
1604 Unit *u,
1605 const char *job_mode_setting,
1606 JobMode job_mode,
1607 const char *dependency_name,
1608 UnitDependencyAtom atom) {
1609
1610 Unit *other, *found = NULL;
1611
1612 if (job_mode != JOB_ISOLATE)
1613 return 0;
1614
1615 UNIT_FOREACH_DEPENDENCY(other, u, atom) {
1616 if (!found)
1617 found = other;
1618 else if (found != other)
1619 return log_unit_error_errno(
1620 u, SYNTHETIC_ERRNO(ENOEXEC),
1621 "More than one %s dependencies specified but %sisolate set. Refusing.",
1622 dependency_name, job_mode_setting);
1623 }
1624
1625 return 0;
1626}
1627
1628int unit_load(Unit *u) {
1629 int r;
1630
1631 assert(u);
1632
1633 if (u->in_load_queue) {
1634 LIST_REMOVE(load_queue, u->manager->load_queue, u);
1635 u->in_load_queue = false;
1636 }
1637
1638 if (u->type == _UNIT_TYPE_INVALID)
1639 return -EINVAL;
1640
1641 if (u->load_state != UNIT_STUB)
1642 return 0;
1643
1644 if (u->transient_file) {
1645 /* Finalize transient file: if this is a transient unit file, as soon as we reach unit_load() the setup
1646 * is complete, hence let's synchronize the unit file we just wrote to disk. */
1647
1648 r = fflush_and_check(u->transient_file);
1649 if (r < 0)
1650 goto fail;
1651
1652 u->transient_file = safe_fclose(u->transient_file);
1653 u->fragment_mtime = now(CLOCK_REALTIME);
1654 }
1655
1656 r = UNIT_VTABLE(u)->load(u);
1657 if (r < 0)
1658 goto fail;
1659
1660 assert(u->load_state != UNIT_STUB);
1661
1662 if (u->load_state == UNIT_LOADED) {
1663 unit_add_to_target_deps_queue(u);
1664
1665 r = unit_add_slice_dependencies(u);
1666 if (r < 0)
1667 goto fail;
1668
1669 r = unit_add_mount_dependencies(u);
1670 if (r < 0)
1671 goto fail;
1672
1673 r = unit_add_oomd_dependencies(u);
1674 if (r < 0)
1675 goto fail;
1676
1677 r = unit_add_startup_units(u);
1678 if (r < 0)
1679 goto fail;
1680
1681 r = unit_validate_on_failure_job_mode(u, "OnSuccessJobMode=", u->on_success_job_mode, "OnSuccess=", UNIT_ATOM_ON_SUCCESS);
1682 if (r < 0)
1683 goto fail;
1684
1685 r = unit_validate_on_failure_job_mode(u, "OnFailureJobMode=", u->on_failure_job_mode, "OnFailure=", UNIT_ATOM_ON_FAILURE);
1686 if (r < 0)
1687 goto fail;
1688
1689 if (u->job_running_timeout != USEC_INFINITY && u->job_running_timeout > u->job_timeout)
1690 log_unit_warning(u, "JobRunningTimeoutSec= is greater than JobTimeoutSec=, it has no effect.");
1691
1692 /* We finished loading, let's ensure our parents recalculate the members mask */
1693 unit_invalidate_cgroup_members_masks(u);
1694 }
1695
1696 assert((u->load_state != UNIT_MERGED) == !u->merged_into);
1697
1698 unit_add_to_dbus_queue(unit_follow_merge(u));
1699 unit_add_to_gc_queue(u);
1700 (void) manager_varlink_send_managed_oom_update(u);
1701
1702 return 0;
1703
1704fail:
1705 /* We convert ENOEXEC errors to the UNIT_BAD_SETTING load state here. Configuration parsing code
1706 * should hence return ENOEXEC to ensure units are placed in this state after loading. */
1707
1708 u->load_state = u->load_state == UNIT_STUB ? UNIT_NOT_FOUND :
1709 r == -ENOEXEC ? UNIT_BAD_SETTING :
1710 UNIT_ERROR;
1711 u->load_error = r;
1712
1713 /* Record the timestamp on the cache, so that if the cache gets updated between now and the next time
1714 * an attempt is made to load this unit, we know we need to check again. */
1715 if (u->load_state == UNIT_NOT_FOUND)
1716 u->fragment_not_found_timestamp_hash = u->manager->unit_cache_timestamp_hash;
1717
1718 unit_add_to_dbus_queue(u);
1719 unit_add_to_gc_queue(u);
1720
1721 return log_unit_debug_errno(u, r, "Failed to load configuration: %m");
1722}
1723
1724_printf_(7, 8)
1725static int log_unit_internal(void *userdata, int level, int error, const char *file, int line, const char *func, const char *format, ...) {
1726 Unit *u = userdata;
1727 va_list ap;
1728 int r;
1729
1730 if (u && !unit_log_level_test(u, level))
1731 return -ERRNO_VALUE(error);
1732
1733 va_start(ap, format);
1734 if (u)
1735 r = log_object_internalv(level, error, file, line, func,
1736 u->manager->unit_log_field,
1737 u->id,
1738 u->manager->invocation_log_field,
1739 u->invocation_id_string,
1740 format, ap);
1741 else
1742 r = log_internalv(level, error, file, line, func, format, ap);
1743 va_end(ap);
1744
1745 return r;
1746}
1747
1748static bool unit_test_condition(Unit *u) {
1749 _cleanup_strv_free_ char **env = NULL;
1750 int r;
1751
1752 assert(u);
1753
1754 dual_timestamp_now(&u->condition_timestamp);
1755
1756 r = manager_get_effective_environment(u->manager, &env);
1757 if (r < 0) {
1758 log_unit_error_errno(u, r, "Failed to determine effective environment: %m");
1759 u->condition_result = true;
1760 } else
1761 u->condition_result = condition_test_list(
1762 u->conditions,
1763 env,
1764 condition_type_to_string,
1765 log_unit_internal,
1766 u);
1767
1768 unit_add_to_dbus_queue(u);
1769 return u->condition_result;
1770}
1771
1772static bool unit_test_assert(Unit *u) {
1773 _cleanup_strv_free_ char **env = NULL;
1774 int r;
1775
1776 assert(u);
1777
1778 dual_timestamp_now(&u->assert_timestamp);
1779
1780 r = manager_get_effective_environment(u->manager, &env);
1781 if (r < 0) {
1782 log_unit_error_errno(u, r, "Failed to determine effective environment: %m");
1783 u->assert_result = CONDITION_ERROR;
1784 } else
1785 u->assert_result = condition_test_list(
1786 u->asserts,
1787 env,
1788 assert_type_to_string,
1789 log_unit_internal,
1790 u);
1791
1792 unit_add_to_dbus_queue(u);
1793 return u->assert_result;
1794}
1795
1796void unit_status_printf(Unit *u, StatusType status_type, const char *status, const char *format, const char *ident) {
1797 if (log_get_show_color()) {
1798 if (u->manager->status_unit_format == STATUS_UNIT_FORMAT_COMBINED && strchr(ident, ' '))
1799 ident = strjoina(ANSI_HIGHLIGHT, u->id, ANSI_NORMAL, " - ", u->description);
1800 else
1801 ident = strjoina(ANSI_HIGHLIGHT, ident, ANSI_NORMAL);
1802 }
1803
1804 DISABLE_WARNING_FORMAT_NONLITERAL;
1805 manager_status_printf(u->manager, status_type, status, format, ident);
1806 REENABLE_WARNING;
1807}
1808
1809int unit_test_start_limit(Unit *u) {
1810 const char *reason;
1811
1812 assert(u);
1813
1814 if (ratelimit_below(&u->start_ratelimit)) {
1815 u->start_limit_hit = false;
1816 return 0;
1817 }
1818
1819 log_unit_warning(u, "Start request repeated too quickly.");
1820 u->start_limit_hit = true;
1821
1822 reason = strjoina("unit ", u->id, " failed");
1823
1824 emergency_action(u->manager, u->start_limit_action,
1825 EMERGENCY_ACTION_IS_WATCHDOG|EMERGENCY_ACTION_WARN,
1826 u->reboot_arg, -1, reason);
1827
1828 return -ECANCELED;
1829}
1830
1831static bool unit_verify_deps(Unit *u) {
1832 Unit *other;
1833
1834 assert(u);
1835
1836 /* Checks whether all BindsTo= dependencies of this unit are fulfilled — if they are also combined
1837 * with After=. We do not check Requires= or Requisite= here as they only should have an effect on
1838 * the job processing, but do not have any effect afterwards. We don't check BindsTo= dependencies
1839 * that are not used in conjunction with After= as for them any such check would make things entirely
1840 * racy. */
1841
1842 UNIT_FOREACH_DEPENDENCY(other, u, UNIT_ATOM_CANNOT_BE_ACTIVE_WITHOUT) {
1843
1844 if (!unit_has_dependency(u, UNIT_ATOM_AFTER, other))
1845 continue;
1846
1847 if (!UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(other))) {
1848 log_unit_notice(u, "Bound to unit %s, but unit isn't active.", other->id);
1849 return false;
1850 }
1851 }
1852
1853 return true;
1854}
1855
1856/* Errors that aren't really errors:
1857 * -EALREADY: Unit is already started.
1858 * -ECOMM: Condition failed
1859 * -EAGAIN: An operation is already in progress. Retry later.
1860 *
1861 * Errors that are real errors:
1862 * -EBADR: This unit type does not support starting.
1863 * -ECANCELED: Start limit hit, too many requests for now
1864 * -EPROTO: Assert failed
1865 * -EINVAL: Unit not loaded
1866 * -EOPNOTSUPP: Unit type not supported
1867 * -ENOLINK: The necessary dependencies are not fulfilled.
1868 * -ESTALE: This unit has been started before and can't be started a second time
1869 * -ENOENT: This is a triggering unit and unit to trigger is not loaded
1870 */
1871int unit_start(Unit *u, ActivationDetails *details) {
1872 UnitActiveState state;
1873 Unit *following;
1874 int r;
1875
1876 assert(u);
1877
1878 /* Let's hold off running start jobs for mount units when /proc/self/mountinfo monitor is ratelimited. */
1879 if (UNIT_VTABLE(u)->subsystem_ratelimited) {
1880 r = UNIT_VTABLE(u)->subsystem_ratelimited(u->manager);
1881 if (r < 0)
1882 return r;
1883 if (r > 0)
1884 return -EAGAIN;
1885 }
1886
1887 /* If this is already started, then this will succeed. Note that this will even succeed if this unit
1888 * is not startable by the user. This is relied on to detect when we need to wait for units and when
1889 * waiting is finished. */
1890 state = unit_active_state(u);
1891 if (UNIT_IS_ACTIVE_OR_RELOADING(state))
1892 return -EALREADY;
1893 if (state == UNIT_MAINTENANCE)
1894 return -EAGAIN;
1895
1896 /* Units that aren't loaded cannot be started */
1897 if (u->load_state != UNIT_LOADED)
1898 return -EINVAL;
1899
1900 /* Refuse starting scope units more than once */
1901 if (UNIT_VTABLE(u)->once_only && dual_timestamp_is_set(&u->inactive_enter_timestamp))
1902 return -ESTALE;
1903
1904 /* If the conditions were unmet, don't do anything at all. If we already are activating this call might
1905 * still be useful to speed up activation in case there is some hold-off time, but we don't want to
1906 * recheck the condition in that case. */
1907 if (state != UNIT_ACTIVATING &&
1908 !unit_test_condition(u))
1909 return log_unit_debug_errno(u, SYNTHETIC_ERRNO(ECOMM), "Starting requested but condition not met. Not starting unit.");
1910
1911 /* If the asserts failed, fail the entire job */
1912 if (state != UNIT_ACTIVATING &&
1913 !unit_test_assert(u))
1914 return log_unit_notice_errno(u, SYNTHETIC_ERRNO(EPROTO), "Starting requested but asserts failed.");
1915
1916 /* Units of types that aren't supported cannot be started. Note that we do this test only after the
1917 * condition checks, so that we rather return condition check errors (which are usually not
1918 * considered a true failure) than "not supported" errors (which are considered a failure).
1919 */
1920 if (!unit_type_supported(u->type))
1921 return -EOPNOTSUPP;
1922
1923 /* Let's make sure that the deps really are in order before we start this. Normally the job engine
1924 * should have taken care of this already, but let's check this here again. After all, our
1925 * dependencies might not be in effect anymore, due to a reload or due to an unmet condition. */
1926 if (!unit_verify_deps(u))
1927 return -ENOLINK;
1928
1929 /* Forward to the main object, if we aren't it. */
1930 following = unit_following(u);
1931 if (following) {
1932 log_unit_debug(u, "Redirecting start request from %s to %s.", u->id, following->id);
1933 return unit_start(following, details);
1934 }
1935
1936 /* Check to make sure the unit isn't frozen */
1937 if (u->freezer_state != FREEZER_RUNNING)
1938 return -EDEADLK;
1939
1940 /* Check our ability to start early so that failure conditions don't cause us to enter a busy loop. */
1941 if (UNIT_VTABLE(u)->can_start) {
1942 r = UNIT_VTABLE(u)->can_start(u);
1943 if (r < 0)
1944 return r;
1945 }
1946
1947 /* If it is stopped, but we cannot start it, then fail */
1948 if (!UNIT_VTABLE(u)->start)
1949 return -EBADR;
1950
1951 /* We don't suppress calls to ->start() here when we are already starting, to allow this request to
1952 * be used as a "hurry up" call, for example when the unit is in some "auto restart" state where it
1953 * waits for a holdoff timer to elapse before it will start again. */
1954
1955 unit_add_to_dbus_queue(u);
1956
1957 if (!u->activation_details) /* Older details object wins */
1958 u->activation_details = activation_details_ref(details);
1959
1960 return UNIT_VTABLE(u)->start(u);
1961}
1962
1963bool unit_can_start(Unit *u) {
1964 assert(u);
1965
1966 if (u->load_state != UNIT_LOADED)
1967 return false;
1968
1969 if (!unit_type_supported(u->type))
1970 return false;
1971
1972 /* Scope units may be started only once */
1973 if (UNIT_VTABLE(u)->once_only && dual_timestamp_is_set(&u->inactive_exit_timestamp))
1974 return false;
1975
1976 return !!UNIT_VTABLE(u)->start;
1977}
1978
1979bool unit_can_isolate(Unit *u) {
1980 assert(u);
1981
1982 return unit_can_start(u) &&
1983 u->allow_isolate;
1984}
1985
1986/* Errors:
1987 * -EBADR: This unit type does not support stopping.
1988 * -EALREADY: Unit is already stopped.
1989 * -EAGAIN: An operation is already in progress. Retry later.
1990 * -EDEADLK: Unit is frozen
1991 */
1992int unit_stop(Unit *u) {
1993 UnitActiveState state;
1994 Unit *following;
1995
1996 assert(u);
1997
1998 state = unit_active_state(u);
1999 if (UNIT_IS_INACTIVE_OR_FAILED(state))
2000 return -EALREADY;
2001
2002 following = unit_following(u);
2003 if (following) {
2004 log_unit_debug(u, "Redirecting stop request from %s to %s.", u->id, following->id);
2005 return unit_stop(following);
2006 }
2007
2008 /* Check to make sure the unit isn't frozen */
2009 if (u->freezer_state != FREEZER_RUNNING)
2010 return -EDEADLK;
2011
2012 if (!UNIT_VTABLE(u)->stop)
2013 return -EBADR;
2014
2015 unit_add_to_dbus_queue(u);
2016
2017 return UNIT_VTABLE(u)->stop(u);
2018}
2019
2020bool unit_can_stop(Unit *u) {
2021 assert(u);
2022
2023 /* Note: if we return true here, it does not mean that the unit may be successfully stopped.
2024 * Extrinsic units follow external state and they may stop following external state changes
2025 * (hence we return true here), but an attempt to do this through the manager will fail. */
2026
2027 if (!unit_type_supported(u->type))
2028 return false;
2029
2030 if (u->perpetual)
2031 return false;
2032
2033 return !!UNIT_VTABLE(u)->stop;
2034}
2035
2036/* Errors:
2037 * -EBADR: This unit type does not support reloading.
2038 * -ENOEXEC: Unit is not started.
2039 * -EAGAIN: An operation is already in progress. Retry later.
2040 * -EDEADLK: Unit is frozen.
2041 */
2042int unit_reload(Unit *u) {
2043 UnitActiveState state;
2044 Unit *following;
2045
2046 assert(u);
2047
2048 if (u->load_state != UNIT_LOADED)
2049 return -EINVAL;
2050
2051 if (!unit_can_reload(u))
2052 return -EBADR;
2053
2054 state = unit_active_state(u);
2055 if (state == UNIT_RELOADING)
2056 return -EAGAIN;
2057
2058 if (state != UNIT_ACTIVE)
2059 return log_unit_warning_errno(u, SYNTHETIC_ERRNO(ENOEXEC), "Unit cannot be reloaded because it is inactive.");
2060
2061 following = unit_following(u);
2062 if (following) {
2063 log_unit_debug(u, "Redirecting reload request from %s to %s.", u->id, following->id);
2064 return unit_reload(following);
2065 }
2066
2067 /* Check to make sure the unit isn't frozen */
2068 if (u->freezer_state != FREEZER_RUNNING)
2069 return -EDEADLK;
2070
2071 unit_add_to_dbus_queue(u);
2072
2073 if (!UNIT_VTABLE(u)->reload) {
2074 /* Unit doesn't have a reload function, but we need to propagate the reload anyway */
2075 unit_notify(u, unit_active_state(u), unit_active_state(u), /* reload_success = */ true);
2076 return 0;
2077 }
2078
2079 return UNIT_VTABLE(u)->reload(u);
2080}
2081
2082bool unit_can_reload(Unit *u) {
2083 assert(u);
2084
2085 if (UNIT_VTABLE(u)->can_reload)
2086 return UNIT_VTABLE(u)->can_reload(u);
2087
2088 if (unit_has_dependency(u, UNIT_ATOM_PROPAGATES_RELOAD_TO, NULL))
2089 return true;
2090
2091 return UNIT_VTABLE(u)->reload;
2092}
2093
2094bool unit_is_unneeded(Unit *u) {
2095 Unit *other;
2096 assert(u);
2097
2098 if (!u->stop_when_unneeded)
2099 return false;
2100
2101 /* Don't clean up while the unit is transitioning or is even inactive. */
2102 if (unit_active_state(u) != UNIT_ACTIVE)
2103 return false;
2104 if (u->job)
2105 return false;
2106
2107 UNIT_FOREACH_DEPENDENCY(other, u, UNIT_ATOM_PINS_STOP_WHEN_UNNEEDED) {
2108 /* If a dependent unit has a job queued, is active or transitioning, or is marked for
2109 * restart, then don't clean this one up. */
2110
2111 if (other->job)
2112 return false;
2113
2114 if (!UNIT_IS_INACTIVE_OR_FAILED(unit_active_state(other)))
2115 return false;
2116
2117 if (unit_will_restart(other))
2118 return false;
2119 }
2120
2121 return true;
2122}
2123
2124bool unit_is_upheld_by_active(Unit *u, Unit **ret_culprit) {
2125 Unit *other;
2126
2127 assert(u);
2128
2129 /* Checks if the unit needs to be started because it currently is not running, but some other unit
2130 * that is active declared an Uphold= dependencies on it */
2131
2132 if (!UNIT_IS_INACTIVE_OR_FAILED(unit_active_state(u)) || u->job) {
2133 if (ret_culprit)
2134 *ret_culprit = NULL;
2135 return false;
2136 }
2137
2138 UNIT_FOREACH_DEPENDENCY(other, u, UNIT_ATOM_START_STEADILY) {
2139 if (other->job)
2140 continue;
2141
2142 if (UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(other))) {
2143 if (ret_culprit)
2144 *ret_culprit = other;
2145 return true;
2146 }
2147 }
2148
2149 if (ret_culprit)
2150 *ret_culprit = NULL;
2151 return false;
2152}
2153
2154bool unit_is_bound_by_inactive(Unit *u, Unit **ret_culprit) {
2155 Unit *other;
2156
2157 assert(u);
2158
2159 /* Checks whether this unit is bound to another unit that is inactive, i.e. whether we should stop
2160 * because the other unit is down. */
2161
2162 if (unit_active_state(u) != UNIT_ACTIVE || u->job) {
2163 /* Don't clean up while the unit is transitioning or is even inactive. */
2164 if (ret_culprit)
2165 *ret_culprit = NULL;
2166 return false;
2167 }
2168
2169 UNIT_FOREACH_DEPENDENCY(other, u, UNIT_ATOM_CANNOT_BE_ACTIVE_WITHOUT) {
2170 if (other->job)
2171 continue;
2172
2173 if (UNIT_IS_INACTIVE_OR_FAILED(unit_active_state(other))) {
2174 if (ret_culprit)
2175 *ret_culprit = other;
2176
2177 return true;
2178 }
2179 }
2180
2181 if (ret_culprit)
2182 *ret_culprit = NULL;
2183 return false;
2184}
2185
2186static void check_unneeded_dependencies(Unit *u) {
2187 Unit *other;
2188 assert(u);
2189
2190 /* Add all units this unit depends on to the queue that processes StopWhenUnneeded= behaviour. */
2191
2192 UNIT_FOREACH_DEPENDENCY(other, u, UNIT_ATOM_ADD_STOP_WHEN_UNNEEDED_QUEUE)
2193 unit_submit_to_stop_when_unneeded_queue(other);
2194}
2195
2196static void check_uphold_dependencies(Unit *u) {
2197 Unit *other;
2198 assert(u);
2199
2200 /* Add all units this unit depends on to the queue that processes Uphold= behaviour. */
2201
2202 UNIT_FOREACH_DEPENDENCY(other, u, UNIT_ATOM_ADD_START_WHEN_UPHELD_QUEUE)
2203 unit_submit_to_start_when_upheld_queue(other);
2204}
2205
2206static void check_bound_by_dependencies(Unit *u) {
2207 Unit *other;
2208 assert(u);
2209
2210 /* Add all units this unit depends on to the queue that processes BindsTo= stop behaviour. */
2211
2212 UNIT_FOREACH_DEPENDENCY(other, u, UNIT_ATOM_ADD_CANNOT_BE_ACTIVE_WITHOUT_QUEUE)
2213 unit_submit_to_stop_when_bound_queue(other);
2214}
2215
2216static void retroactively_start_dependencies(Unit *u) {
2217 Unit *other;
2218
2219 assert(u);
2220 assert(UNIT_IS_ACTIVE_OR_ACTIVATING(unit_active_state(u)));
2221
2222 UNIT_FOREACH_DEPENDENCY(other, u, UNIT_ATOM_RETROACTIVE_START_REPLACE) /* Requires= + BindsTo= */
2223 if (!unit_has_dependency(u, UNIT_ATOM_AFTER, other) &&
2224 !UNIT_IS_ACTIVE_OR_ACTIVATING(unit_active_state(other)))
2225 (void) manager_add_job(u->manager, JOB_START, other, JOB_REPLACE, NULL, NULL, NULL);
2226
2227 UNIT_FOREACH_DEPENDENCY(other, u, UNIT_ATOM_RETROACTIVE_START_FAIL) /* Wants= */
2228 if (!unit_has_dependency(u, UNIT_ATOM_AFTER, other) &&
2229 !UNIT_IS_ACTIVE_OR_ACTIVATING(unit_active_state(other)))
2230 (void) manager_add_job(u->manager, JOB_START, other, JOB_FAIL, NULL, NULL, NULL);
2231
2232 UNIT_FOREACH_DEPENDENCY(other, u, UNIT_ATOM_RETROACTIVE_STOP_ON_START) /* Conflicts= (and inverse) */
2233 if (!UNIT_IS_INACTIVE_OR_DEACTIVATING(unit_active_state(other)))
2234 (void) manager_add_job(u->manager, JOB_STOP, other, JOB_REPLACE, NULL, NULL, NULL);
2235}
2236
2237static void retroactively_stop_dependencies(Unit *u) {
2238 Unit *other;
2239
2240 assert(u);
2241 assert(UNIT_IS_INACTIVE_OR_DEACTIVATING(unit_active_state(u)));
2242
2243 /* Pull down units which are bound to us recursively if enabled */
2244 UNIT_FOREACH_DEPENDENCY(other, u, UNIT_ATOM_RETROACTIVE_STOP_ON_STOP) /* BoundBy= */
2245 if (!UNIT_IS_INACTIVE_OR_DEACTIVATING(unit_active_state(other)))
2246 (void) manager_add_job(u->manager, JOB_STOP, other, JOB_REPLACE, NULL, NULL, NULL);
2247}
2248
2249void unit_start_on_failure(
2250 Unit *u,
2251 const char *dependency_name,
2252 UnitDependencyAtom atom,
2253 JobMode job_mode) {
2254
2255 int n_jobs = -1;
2256 Unit *other;
2257 int r;
2258
2259 assert(u);
2260 assert(dependency_name);
2261 assert(IN_SET(atom, UNIT_ATOM_ON_SUCCESS, UNIT_ATOM_ON_FAILURE));
2262
2263 /* Act on OnFailure= and OnSuccess= dependencies */
2264
2265 UNIT_FOREACH_DEPENDENCY(other, u, atom) {
2266 _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL;
2267
2268 if (n_jobs < 0) {
2269 log_unit_info(u, "Triggering %s dependencies.", dependency_name);
2270 n_jobs = 0;
2271 }
2272
2273 r = manager_add_job(u->manager, JOB_START, other, job_mode, NULL, &error, NULL);
2274 if (r < 0)
2275 log_unit_warning_errno(
2276 u, r, "Failed to enqueue %s job, ignoring: %s",
2277 dependency_name, bus_error_message(&error, r));
2278 n_jobs++;
2279 }
2280
2281 if (n_jobs >= 0)
2282 log_unit_debug(u, "Triggering %s dependencies done (%i %s).",
2283 dependency_name, n_jobs, n_jobs == 1 ? "job" : "jobs");
2284}
2285
2286void unit_trigger_notify(Unit *u) {
2287 Unit *other;
2288
2289 assert(u);
2290
2291 UNIT_FOREACH_DEPENDENCY(other, u, UNIT_ATOM_TRIGGERED_BY)
2292 if (UNIT_VTABLE(other)->trigger_notify)
2293 UNIT_VTABLE(other)->trigger_notify(other, u);
2294}
2295
2296static int raise_level(int log_level, bool condition_info, bool condition_notice) {
2297 if (condition_notice && log_level > LOG_NOTICE)
2298 return LOG_NOTICE;
2299 if (condition_info && log_level > LOG_INFO)
2300 return LOG_INFO;
2301 return log_level;
2302}
2303
2304static int unit_log_resources(Unit *u) {
2305
2306 static const struct {
2307 const char *journal_field;
2308 const char *message_suffix;
2309 } memory_fields[_CGROUP_MEMORY_ACCOUNTING_METRIC_CACHED_LAST + 1] = {
2310 [CGROUP_MEMORY_PEAK] = { "MEMORY_PEAK", "memory peak" },
2311 [CGROUP_MEMORY_SWAP_PEAK] = { "MEMORY_SWAP_PEAK", "memory swap peak" },
2312 }, ip_fields[_CGROUP_IP_ACCOUNTING_METRIC_MAX] = {
2313 [CGROUP_IP_INGRESS_BYTES] = { "IP_METRIC_INGRESS_BYTES", "incoming IP traffic" },
2314 [CGROUP_IP_EGRESS_BYTES] = { "IP_METRIC_EGRESS_BYTES", "outgoing IP traffic" },
2315 [CGROUP_IP_INGRESS_PACKETS] = { "IP_METRIC_INGRESS_PACKETS", NULL },
2316 [CGROUP_IP_EGRESS_PACKETS] = { "IP_METRIC_EGRESS_PACKETS", NULL },
2317 }, io_fields[_CGROUP_IO_ACCOUNTING_METRIC_MAX] = {
2318 [CGROUP_IO_READ_BYTES] = { "IO_METRIC_READ_BYTES", "read from disk" },
2319 [CGROUP_IO_WRITE_BYTES] = { "IO_METRIC_WRITE_BYTES", "written to disk" },
2320 [CGROUP_IO_READ_OPERATIONS] = { "IO_METRIC_READ_OPERATIONS", NULL },
2321 [CGROUP_IO_WRITE_OPERATIONS] = { "IO_METRIC_WRITE_OPERATIONS", NULL },
2322 };
2323
2324 struct iovec *iovec = NULL;
2325 size_t n_iovec = 0;
2326 _cleanup_free_ char *message = NULL, *t = NULL;
2327 nsec_t cpu_nsec = NSEC_INFINITY;
2328 int log_level = LOG_DEBUG; /* May be raised if resources consumed over a threshold */
2329
2330 assert(u);
2331
2332 CLEANUP_ARRAY(iovec, n_iovec, iovec_array_free);
2333
2334 iovec = new(struct iovec, 1 + (_CGROUP_MEMORY_ACCOUNTING_METRIC_CACHED_LAST + 1) +
2335 _CGROUP_IP_ACCOUNTING_METRIC_MAX + _CGROUP_IO_ACCOUNTING_METRIC_MAX + 4);
2336 if (!iovec)
2337 return log_oom();
2338
2339 /* Invoked whenever a unit enters failed or dead state. Logs information about consumed resources if resource
2340 * accounting was enabled for a unit. It does this in two ways: a friendly human readable string with reduced
2341 * information and the complete data in structured fields. */
2342
2343 (void) unit_get_cpu_usage(u, &cpu_nsec);
2344 if (cpu_nsec != NSEC_INFINITY) {
2345 /* Format the CPU time for inclusion in the structured log message */
2346 if (asprintf(&t, "CPU_USAGE_NSEC=%" PRIu64, cpu_nsec) < 0)
2347 return log_oom();
2348 iovec[n_iovec++] = IOVEC_MAKE_STRING(TAKE_PTR(t));
2349
2350 /* Format the CPU time for inclusion in the human language message string */
2351 if (strextendf_with_separator(&message, ", ",
2352 "Consumed %s CPU time",
2353 FORMAT_TIMESPAN(cpu_nsec / NSEC_PER_USEC, USEC_PER_MSEC)) < 0)
2354 return log_oom();
2355
2356 log_level = raise_level(log_level,
2357 cpu_nsec > MENTIONWORTHY_CPU_NSEC,
2358 cpu_nsec > NOTICEWORTHY_CPU_NSEC);
2359 }
2360
2361 for (CGroupMemoryAccountingMetric metric = 0; metric <= _CGROUP_MEMORY_ACCOUNTING_METRIC_CACHED_LAST; metric++) {
2362 uint64_t value = UINT64_MAX;
2363
2364 assert(memory_fields[metric].journal_field);
2365 assert(memory_fields[metric].message_suffix);
2366
2367 (void) unit_get_memory_accounting(u, metric, &value);
2368 if (value == UINT64_MAX)
2369 continue;
2370
2371 if (asprintf(&t, "%s=%" PRIu64, memory_fields[metric].journal_field, value) < 0)
2372 return log_oom();
2373 iovec[n_iovec++] = IOVEC_MAKE_STRING(TAKE_PTR(t));
2374
2375 /* If value is 0, we don't log it in the MESSAGE= field. */
2376 if (value == 0)
2377 continue;
2378
2379 if (strextendf_with_separator(&message, ", ", "%s %s",
2380 FORMAT_BYTES(value), memory_fields[metric].message_suffix) < 0)
2381 return log_oom();
2382
2383 log_level = raise_level(log_level,
2384 value > MENTIONWORTHY_MEMORY_BYTES,
2385 value > NOTICEWORTHY_MEMORY_BYTES);
2386 }
2387
2388 for (CGroupIOAccountingMetric k = 0; k < _CGROUP_IO_ACCOUNTING_METRIC_MAX; k++) {
2389 uint64_t value = UINT64_MAX;
2390
2391 assert(io_fields[k].journal_field);
2392
2393 (void) unit_get_io_accounting(u, k, k > 0, &value);
2394 if (value == UINT64_MAX)
2395 continue;
2396
2397 /* Format IO accounting data for inclusion in the structured log message */
2398 if (asprintf(&t, "%s=%" PRIu64, io_fields[k].journal_field, value) < 0)
2399 return log_oom();
2400 iovec[n_iovec++] = IOVEC_MAKE_STRING(TAKE_PTR(t));
2401
2402 /* If value is 0, we don't log it in the MESSAGE= field. */
2403 if (value == 0)
2404 continue;
2405
2406 /* Format the IO accounting data for inclusion in the human language message string, but only
2407 * for the bytes counters (and not for the operations counters) */
2408 if (io_fields[k].message_suffix) {
2409 if (strextendf_with_separator(&message, ", ", "%s %s",
2410 FORMAT_BYTES(value), io_fields[k].message_suffix) < 0)
2411 return log_oom();
2412
2413 log_level = raise_level(log_level,
2414 value > MENTIONWORTHY_IO_BYTES,
2415 value > NOTICEWORTHY_IO_BYTES);
2416 }
2417 }
2418
2419 for (CGroupIPAccountingMetric m = 0; m < _CGROUP_IP_ACCOUNTING_METRIC_MAX; m++) {
2420 uint64_t value = UINT64_MAX;
2421
2422 assert(ip_fields[m].journal_field);
2423
2424 (void) unit_get_ip_accounting(u, m, &value);
2425 if (value == UINT64_MAX)
2426 continue;
2427
2428 /* Format IP accounting data for inclusion in the structured log message */
2429 if (asprintf(&t, "%s=%" PRIu64, ip_fields[m].journal_field, value) < 0)
2430 return log_oom();
2431 iovec[n_iovec++] = IOVEC_MAKE_STRING(TAKE_PTR(t));
2432
2433 /* If value is 0, we don't log it in the MESSAGE= field. */
2434 if (value == 0)
2435 continue;
2436
2437 /* Format the IP accounting data for inclusion in the human language message string, but only
2438 * for the bytes counters (and not for the packets counters) */
2439 if (ip_fields[m].message_suffix) {
2440 if (strextendf_with_separator(&message, ", ", "%s %s",
2441 FORMAT_BYTES(value), ip_fields[m].message_suffix) < 0)
2442 return log_oom();
2443
2444 log_level = raise_level(log_level,
2445 value > MENTIONWORTHY_IP_BYTES,
2446 value > NOTICEWORTHY_IP_BYTES);
2447 }
2448 }
2449
2450 /* This check is here because it is the earliest point following all possible log_level assignments.
2451 * (If log_level is assigned anywhere after this point, move this check.) */
2452 if (!unit_log_level_test(u, log_level))
2453 return 0;
2454
2455 /* Is there any accounting data available at all? */
2456 if (n_iovec == 0) {
2457 assert(!message);
2458 return 0;
2459 }
2460
2461 t = strjoin("MESSAGE=", u->id, ": ", message ?: "Completed", ".");
2462 if (!t)
2463 return log_oom();
2464 iovec[n_iovec++] = IOVEC_MAKE_STRING(TAKE_PTR(t));
2465
2466 if (!set_iovec_string_field(iovec, &n_iovec, "MESSAGE_ID=", SD_MESSAGE_UNIT_RESOURCES_STR))
2467 return log_oom();
2468
2469 if (!set_iovec_string_field(iovec, &n_iovec, u->manager->unit_log_field, u->id))
2470 return log_oom();
2471
2472 if (!set_iovec_string_field(iovec, &n_iovec, u->manager->invocation_log_field, u->invocation_id_string))
2473 return log_oom();
2474
2475 log_unit_struct_iovec(u, log_level, iovec, n_iovec);
2476
2477 return 0;
2478}
2479
2480static void unit_update_on_console(Unit *u) {
2481 bool b;
2482
2483 assert(u);
2484
2485 b = unit_needs_console(u);
2486 if (u->on_console == b)
2487 return;
2488
2489 u->on_console = b;
2490 if (b)
2491 manager_ref_console(u->manager);
2492 else
2493 manager_unref_console(u->manager);
2494}
2495
2496static void unit_emit_audit_start(Unit *u) {
2497 assert(u);
2498
2499 if (UNIT_VTABLE(u)->audit_start_message_type <= 0)
2500 return;
2501
2502 /* Write audit record if we have just finished starting up */
2503 manager_send_unit_audit(u->manager, u, UNIT_VTABLE(u)->audit_start_message_type, /* success= */ true);
2504 u->in_audit = true;
2505}
2506
2507static void unit_emit_audit_stop(Unit *u, UnitActiveState state) {
2508 assert(u);
2509
2510 if (UNIT_VTABLE(u)->audit_start_message_type <= 0)
2511 return;
2512
2513 if (u->in_audit) {
2514 /* Write audit record if we have just finished shutting down */
2515 manager_send_unit_audit(u->manager, u, UNIT_VTABLE(u)->audit_stop_message_type, /* success= */ state == UNIT_INACTIVE);
2516 u->in_audit = false;
2517 } else {
2518 /* Hmm, if there was no start record written write it now, so that we always have a nice pair */
2519 manager_send_unit_audit(u->manager, u, UNIT_VTABLE(u)->audit_start_message_type, /* success= */ state == UNIT_INACTIVE);
2520
2521 if (state == UNIT_INACTIVE)
2522 manager_send_unit_audit(u->manager, u, UNIT_VTABLE(u)->audit_stop_message_type, /* success= */ true);
2523 }
2524}
2525
2526static bool unit_process_job(Job *j, UnitActiveState ns, bool reload_success) {
2527 bool unexpected = false;
2528 JobResult result;
2529
2530 assert(j);
2531
2532 if (j->state == JOB_WAITING)
2533 /* So we reached a different state for this job. Let's see if we can run it now if it failed previously
2534 * due to EAGAIN. */
2535 job_add_to_run_queue(j);
2536
2537 /* Let's check whether the unit's new state constitutes a finished job, or maybe contradicts a running job and
2538 * hence needs to invalidate jobs. */
2539
2540 switch (j->type) {
2541
2542 case JOB_START:
2543 case JOB_VERIFY_ACTIVE:
2544
2545 if (UNIT_IS_ACTIVE_OR_RELOADING(ns))
2546 job_finish_and_invalidate(j, JOB_DONE, true, false);
2547 else if (j->state == JOB_RUNNING && ns != UNIT_ACTIVATING) {
2548 unexpected = true;
2549
2550 if (UNIT_IS_INACTIVE_OR_FAILED(ns)) {
2551 if (ns == UNIT_FAILED)
2552 result = JOB_FAILED;
2553 else
2554 result = JOB_DONE;
2555
2556 job_finish_and_invalidate(j, result, true, false);
2557 }
2558 }
2559
2560 break;
2561
2562 case JOB_RELOAD:
2563 case JOB_RELOAD_OR_START:
2564 case JOB_TRY_RELOAD:
2565
2566 if (j->state == JOB_RUNNING) {
2567 if (ns == UNIT_ACTIVE)
2568 job_finish_and_invalidate(j, reload_success ? JOB_DONE : JOB_FAILED, true, false);
2569 else if (!IN_SET(ns, UNIT_ACTIVATING, UNIT_RELOADING)) {
2570 unexpected = true;
2571
2572 if (UNIT_IS_INACTIVE_OR_FAILED(ns))
2573 job_finish_and_invalidate(j, ns == UNIT_FAILED ? JOB_FAILED : JOB_DONE, true, false);
2574 }
2575 }
2576
2577 break;
2578
2579 case JOB_STOP:
2580 case JOB_RESTART:
2581 case JOB_TRY_RESTART:
2582
2583 if (UNIT_IS_INACTIVE_OR_FAILED(ns))
2584 job_finish_and_invalidate(j, JOB_DONE, true, false);
2585 else if (j->state == JOB_RUNNING && ns != UNIT_DEACTIVATING) {
2586 unexpected = true;
2587 job_finish_and_invalidate(j, JOB_FAILED, true, false);
2588 }
2589
2590 break;
2591
2592 default:
2593 assert_not_reached();
2594 }
2595
2596 return unexpected;
2597}
2598
2599void unit_notify(Unit *u, UnitActiveState os, UnitActiveState ns, bool reload_success) {
2600 const char *reason;
2601 Manager *m;
2602
2603 assert(u);
2604 assert(os < _UNIT_ACTIVE_STATE_MAX);
2605 assert(ns < _UNIT_ACTIVE_STATE_MAX);
2606
2607 /* Note that this is called for all low-level state changes, even if they might map to the same high-level
2608 * UnitActiveState! That means that ns == os is an expected behavior here. For example: if a mount point is
2609 * remounted this function will be called too! */
2610
2611 m = u->manager;
2612
2613 /* Let's enqueue the change signal early. In case this unit has a job associated we want that this unit is in
2614 * the bus queue, so that any job change signal queued will force out the unit change signal first. */
2615 unit_add_to_dbus_queue(u);
2616
2617 /* Update systemd-oomd on the property/state change */
2618 if (os != ns) {
2619 /* Always send an update if the unit is going into an inactive state so systemd-oomd knows to stop
2620 * monitoring.
2621 * Also send an update whenever the unit goes active; this is to handle a case where an override file
2622 * sets one of the ManagedOOM*= properties to "kill", then later removes it. systemd-oomd needs to
2623 * know to stop monitoring when the unit changes from "kill" -> "auto" on daemon-reload, but we don't
2624 * have the information on the property. Thus, indiscriminately send an update. */
2625 if (UNIT_IS_INACTIVE_OR_FAILED(ns) || UNIT_IS_ACTIVE_OR_RELOADING(ns))
2626 (void) manager_varlink_send_managed_oom_update(u);
2627 }
2628
2629 /* Update timestamps for state changes */
2630 if (!MANAGER_IS_RELOADING(m)) {
2631 dual_timestamp_now(&u->state_change_timestamp);
2632
2633 if (UNIT_IS_INACTIVE_OR_FAILED(os) && !UNIT_IS_INACTIVE_OR_FAILED(ns))
2634 u->inactive_exit_timestamp = u->state_change_timestamp;
2635 else if (!UNIT_IS_INACTIVE_OR_FAILED(os) && UNIT_IS_INACTIVE_OR_FAILED(ns))
2636 u->inactive_enter_timestamp = u->state_change_timestamp;
2637
2638 if (!UNIT_IS_ACTIVE_OR_RELOADING(os) && UNIT_IS_ACTIVE_OR_RELOADING(ns))
2639 u->active_enter_timestamp = u->state_change_timestamp;
2640 else if (UNIT_IS_ACTIVE_OR_RELOADING(os) && !UNIT_IS_ACTIVE_OR_RELOADING(ns))
2641 u->active_exit_timestamp = u->state_change_timestamp;
2642 }
2643
2644 /* Keep track of failed units */
2645 (void) manager_update_failed_units(m, u, ns == UNIT_FAILED);
2646
2647 /* Make sure the cgroup and state files are always removed when we become inactive */
2648 if (UNIT_IS_INACTIVE_OR_FAILED(ns)) {
2649 SET_FLAG(u->markers,
2650 (1u << UNIT_MARKER_NEEDS_RELOAD)|(1u << UNIT_MARKER_NEEDS_RESTART),
2651 false);
2652 unit_prune_cgroup(u);
2653 unit_unlink_state_files(u);
2654 } else if (ns != os && ns == UNIT_RELOADING)
2655 SET_FLAG(u->markers, 1u << UNIT_MARKER_NEEDS_RELOAD, false);
2656
2657 unit_update_on_console(u);
2658
2659 if (!MANAGER_IS_RELOADING(m)) {
2660 bool unexpected;
2661
2662 /* Let's propagate state changes to the job */
2663 if (u->job)
2664 unexpected = unit_process_job(u->job, ns, reload_success);
2665 else
2666 unexpected = true;
2667
2668 /* If this state change happened without being requested by a job, then let's retroactively start or
2669 * stop dependencies. We skip that step when deserializing, since we don't want to create any
2670 * additional jobs just because something is already activated. */
2671
2672 if (unexpected) {
2673 if (UNIT_IS_INACTIVE_OR_FAILED(os) && UNIT_IS_ACTIVE_OR_ACTIVATING(ns))
2674 retroactively_start_dependencies(u);
2675 else if (UNIT_IS_ACTIVE_OR_ACTIVATING(os) && UNIT_IS_INACTIVE_OR_DEACTIVATING(ns))
2676 retroactively_stop_dependencies(u);
2677 }
2678
2679 if (ns != os && ns == UNIT_FAILED) {
2680 log_unit_debug(u, "Unit entered failed state.");
2681 unit_start_on_failure(u, "OnFailure=", UNIT_ATOM_ON_FAILURE, u->on_failure_job_mode);
2682 }
2683
2684 if (UNIT_IS_ACTIVE_OR_RELOADING(ns) && !UNIT_IS_ACTIVE_OR_RELOADING(os)) {
2685 /* This unit just finished starting up */
2686
2687 unit_emit_audit_start(u);
2688 manager_send_unit_plymouth(m, u);
2689 manager_send_unit_supervisor(m, u, /* active= */ true);
2690 }
2691
2692 if (UNIT_IS_INACTIVE_OR_FAILED(ns) && !UNIT_IS_INACTIVE_OR_FAILED(os)) {
2693 /* This unit just stopped/failed. */
2694
2695 unit_emit_audit_stop(u, ns);
2696 manager_send_unit_supervisor(m, u, /* active= */ false);
2697 unit_log_resources(u);
2698 }
2699
2700 if (ns == UNIT_INACTIVE && !IN_SET(os, UNIT_FAILED, UNIT_INACTIVE, UNIT_MAINTENANCE))
2701 unit_start_on_failure(u, "OnSuccess=", UNIT_ATOM_ON_SUCCESS, u->on_success_job_mode);
2702 }
2703
2704 manager_recheck_journal(m);
2705 manager_recheck_dbus(m);
2706
2707 unit_trigger_notify(u);
2708
2709 if (!MANAGER_IS_RELOADING(m)) {
2710 if (os != UNIT_FAILED && ns == UNIT_FAILED) {
2711 reason = strjoina("unit ", u->id, " failed");
2712 emergency_action(m, u->failure_action, 0, u->reboot_arg, unit_failure_action_exit_status(u), reason);
2713 } else if (!UNIT_IS_INACTIVE_OR_FAILED(os) && ns == UNIT_INACTIVE) {
2714 reason = strjoina("unit ", u->id, " succeeded");
2715 emergency_action(m, u->success_action, 0, u->reboot_arg, unit_success_action_exit_status(u), reason);
2716 }
2717 }
2718
2719 /* And now, add the unit or depending units to various queues that will act on the new situation if
2720 * needed. These queues generally check for continuous state changes rather than events (like most of
2721 * the state propagation above), and do work deferred instead of instantly, since they typically
2722 * don't want to run during reloading, and usually involve checking combined state of multiple units
2723 * at once. */
2724
2725 if (UNIT_IS_INACTIVE_OR_FAILED(ns)) {
2726 /* Stop unneeded units and bound-by units regardless if going down was expected or not */
2727 check_unneeded_dependencies(u);
2728 check_bound_by_dependencies(u);
2729
2730 /* Maybe someone wants us to remain up? */
2731 unit_submit_to_start_when_upheld_queue(u);
2732
2733 /* Maybe the unit should be GC'ed now? */
2734 unit_add_to_gc_queue(u);
2735
2736 /* Maybe we can release some resources now? */
2737 unit_submit_to_release_resources_queue(u);
2738 }
2739
2740 if (UNIT_IS_ACTIVE_OR_RELOADING(ns)) {
2741 /* Start uphold units regardless if going up was expected or not */
2742 check_uphold_dependencies(u);
2743
2744 /* Maybe we finished startup and are now ready for being stopped because unneeded? */
2745 unit_submit_to_stop_when_unneeded_queue(u);
2746
2747 /* Maybe we finished startup, but something we needed has vanished? Let's die then. (This happens
2748 * when something BindsTo= to a Type=oneshot unit, as these units go directly from starting to
2749 * inactive, without ever entering started.) */
2750 unit_submit_to_stop_when_bound_queue(u);
2751 }
2752}
2753
2754int unit_watch_pidref(Unit *u, const PidRef *pid, bool exclusive) {
2755 _cleanup_(pidref_freep) PidRef *pid_dup = NULL;
2756 int r;
2757
2758 /* Adds a specific PID to the set of PIDs this unit watches. */
2759
2760 assert(u);
2761 assert(pidref_is_set(pid));
2762
2763 /* Caller might be sure that this PID belongs to this unit only. Let's take this
2764 * opportunity to remove any stalled references to this PID as they can be created
2765 * easily (when watching a process which is not our direct child). */
2766 if (exclusive)
2767 manager_unwatch_pidref(u->manager, pid);
2768
2769 if (set_contains(u->pids, pid)) /* early exit if already being watched */
2770 return 0;
2771
2772 r = pidref_dup(pid, &pid_dup);
2773 if (r < 0)
2774 return r;
2775
2776 /* First, insert into the set of PIDs maintained by the unit */
2777 r = set_ensure_put(&u->pids, &pidref_hash_ops_free, pid_dup);
2778 if (r < 0)
2779 return r;
2780
2781 pid = TAKE_PTR(pid_dup); /* continue with our copy now that we have installed it properly in our set */
2782
2783 /* Second, insert it into the simple global table, see if that works */
2784 r = hashmap_ensure_put(&u->manager->watch_pids, &pidref_hash_ops_free, pid, u);
2785 if (r != -EEXIST)
2786 return r;
2787
2788 /* OK, the key is already assigned to a different unit. That's fine, then add us via the second
2789 * hashmap that points to an array. */
2790
2791 PidRef *old_pid = NULL;
2792 Unit **array = hashmap_get2(u->manager->watch_pids_more, pid, (void**) &old_pid);
2793
2794 /* Count entries in array */
2795 size_t n = 0;
2796 for (; array && array[n]; n++)
2797 ;
2798
2799 /* Allocate a new array */
2800 _cleanup_free_ Unit **new_array = new(Unit*, n + 2);
2801 if (!new_array)
2802 return -ENOMEM;
2803
2804 /* Append us to the end */
2805 memcpy_safe(new_array, array, sizeof(Unit*) * n);
2806 new_array[n] = u;
2807 new_array[n+1] = NULL;
2808
2809 /* Make sure the hashmap is allocated */
2810 r = hashmap_ensure_allocated(&u->manager->watch_pids_more, &pidref_hash_ops_free);
2811 if (r < 0)
2812 return r;
2813
2814 /* Add or replace the old array */
2815 r = hashmap_replace(u->manager->watch_pids_more, old_pid ?: pid, new_array);
2816 if (r < 0)
2817 return r;
2818
2819 TAKE_PTR(new_array); /* Now part of the hash table */
2820 free(array); /* Which means we can now delete the old version */
2821 return 0;
2822}
2823
2824int unit_watch_pid(Unit *u, pid_t pid, bool exclusive) {
2825 _cleanup_(pidref_done) PidRef pidref = PIDREF_NULL;
2826 int r;
2827
2828 assert(u);
2829 assert(pid_is_valid(pid));
2830
2831 r = pidref_set_pid(&pidref, pid);
2832 if (r < 0)
2833 return r;
2834
2835 return unit_watch_pidref(u, &pidref, exclusive);
2836}
2837
2838void unit_unwatch_pidref(Unit *u, const PidRef *pid) {
2839 assert(u);
2840 assert(pidref_is_set(pid));
2841
2842 /* Remove from the set we maintain for this unit. (And destroy the returned pid eventually) */
2843 _cleanup_(pidref_freep) PidRef *pid1 = set_remove(u->pids, pid);
2844 if (!pid1)
2845 return; /* Early exit if this PID was never watched by us */
2846
2847 /* First let's drop the unit from the simple hash table, if it is included there */
2848 PidRef *pid2 = NULL;
2849 Unit *uu = hashmap_get2(u->manager->watch_pids, pid, (void**) &pid2);
2850
2851 /* Quick validation: iff we are in the watch_pids table then the PidRef object must be the same as in our local pids set */
2852 assert((uu == u) == (pid1 == pid2));
2853
2854 if (uu == u)
2855 /* OK, we are in the first table. Let's remove it there then, and we are done already. */
2856 assert_se(hashmap_remove_value(u->manager->watch_pids, pid2, uu));
2857 else {
2858 /* We weren't in the first table, then let's consult the 2nd table that points to an array */
2859 PidRef *pid3 = NULL;
2860 Unit **array = hashmap_get2(u->manager->watch_pids_more, pid, (void**) &pid3);
2861
2862 /* Let's iterate through the array, dropping our own entry */
2863 size_t m = 0, n = 0;
2864 for (; array && array[n]; n++)
2865 if (array[n] != u)
2866 array[m++] = array[n];
2867 if (n == m)
2868 return; /* Not there */
2869
2870 array[m] = NULL; /* set trailing NULL marker on the new end */
2871
2872 if (m == 0) {
2873 /* The array is now empty, remove the entire entry */
2874 assert_se(hashmap_remove_value(u->manager->watch_pids_more, pid3, array));
2875 free(array);
2876 } else {
2877 /* The array is not empty, but let's make sure the entry is not keyed by the PidRef
2878 * we will delete, but by the PidRef object of the Unit that is now first in the
2879 * array. */
2880
2881 PidRef *new_pid3 = ASSERT_PTR(set_get(array[0]->pids, pid));
2882 assert_se(hashmap_replace(u->manager->watch_pids_more, new_pid3, array) >= 0);
2883 }
2884 }
2885}
2886
2887void unit_unwatch_pid(Unit *u, pid_t pid) {
2888 return unit_unwatch_pidref(u, &PIDREF_MAKE_FROM_PID(pid));
2889}
2890
2891void unit_unwatch_all_pids(Unit *u) {
2892 assert(u);
2893
2894 while (!set_isempty(u->pids))
2895 unit_unwatch_pidref(u, set_first(u->pids));
2896
2897 u->pids = set_free(u->pids);
2898}
2899
2900void unit_unwatch_pidref_done(Unit *u, PidRef *pidref) {
2901 assert(u);
2902
2903 if (!pidref_is_set(pidref))
2904 return;
2905
2906 unit_unwatch_pidref(u, pidref);
2907 pidref_done(pidref);
2908}
2909
2910static void unit_tidy_watch_pids(Unit *u) {
2911 PidRef *except1, *except2, *e;
2912
2913 assert(u);
2914
2915 /* Cleans dead PIDs from our list */
2916
2917 except1 = unit_main_pid(u);
2918 except2 = unit_control_pid(u);
2919
2920 SET_FOREACH(e, u->pids) {
2921 if (pidref_equal(except1, e) || pidref_equal(except2, e))
2922 continue;
2923
2924 if (pidref_is_unwaited(e) <= 0)
2925 unit_unwatch_pidref(u, e);
2926 }
2927}
2928
2929static int on_rewatch_pids_event(sd_event_source *s, void *userdata) {
2930 Unit *u = ASSERT_PTR(userdata);
2931
2932 assert(s);
2933
2934 unit_tidy_watch_pids(u);
2935 (void) unit_watch_all_pids(u);
2936
2937 /* If the PID set is empty now, then let's finish this off. */
2938 unit_synthesize_cgroup_empty_event(u);
2939
2940 return 0;
2941}
2942
2943int unit_enqueue_rewatch_pids(Unit *u) {
2944 int r;
2945
2946 assert(u);
2947
2948 CGroupRuntime *crt = unit_get_cgroup_runtime(u);
2949 if (!crt || !crt->cgroup_path)
2950 return -ENOENT;
2951
2952 r = cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER);
2953 if (r < 0)
2954 return r;
2955 if (r > 0) /* On unified we can use proper notifications */
2956 return 0;
2957
2958 /* Enqueues a low-priority job that will clean up dead PIDs from our list of PIDs to watch and subscribe to new
2959 * PIDs that might have appeared. We do this in a delayed job because the work might be quite slow, as it
2960 * involves issuing kill(pid, 0) on all processes we watch. */
2961
2962 if (!u->rewatch_pids_event_source) {
2963 _cleanup_(sd_event_source_unrefp) sd_event_source *s = NULL;
2964
2965 r = sd_event_add_defer(u->manager->event, &s, on_rewatch_pids_event, u);
2966 if (r < 0)
2967 return log_error_errno(r, "Failed to allocate event source for tidying watched PIDs: %m");
2968
2969 r = sd_event_source_set_priority(s, EVENT_PRIORITY_REWATCH_PIDS);
2970 if (r < 0)
2971 return log_error_errno(r, "Failed to adjust priority of event source for tidying watched PIDs: %m");
2972
2973 (void) sd_event_source_set_description(s, "tidy-watch-pids");
2974
2975 u->rewatch_pids_event_source = TAKE_PTR(s);
2976 }
2977
2978 r = sd_event_source_set_enabled(u->rewatch_pids_event_source, SD_EVENT_ONESHOT);
2979 if (r < 0)
2980 return log_error_errno(r, "Failed to enable event source for tidying watched PIDs: %m");
2981
2982 return 0;
2983}
2984
2985void unit_dequeue_rewatch_pids(Unit *u) {
2986 int r;
2987 assert(u);
2988
2989 if (!u->rewatch_pids_event_source)
2990 return;
2991
2992 r = sd_event_source_set_enabled(u->rewatch_pids_event_source, SD_EVENT_OFF);
2993 if (r < 0)
2994 log_warning_errno(r, "Failed to disable event source for tidying watched PIDs, ignoring: %m");
2995
2996 u->rewatch_pids_event_source = sd_event_source_disable_unref(u->rewatch_pids_event_source);
2997}
2998
2999bool unit_job_is_applicable(Unit *u, JobType j) {
3000 assert(u);
3001 assert(j >= 0 && j < _JOB_TYPE_MAX);
3002
3003 switch (j) {
3004
3005 case JOB_VERIFY_ACTIVE:
3006 case JOB_START:
3007 case JOB_NOP:
3008 /* Note that we don't check unit_can_start() here. That's because .device units and suchlike are not
3009 * startable by us but may appear due to external events, and it thus makes sense to permit enqueuing
3010 * jobs for it. */
3011 return true;
3012
3013 case JOB_STOP:
3014 /* Similar as above. However, perpetual units can never be stopped (neither explicitly nor due to
3015 * external events), hence it makes no sense to permit enqueuing such a request either. */
3016 return !u->perpetual;
3017
3018 case JOB_RESTART:
3019 case JOB_TRY_RESTART:
3020 return unit_can_stop(u) && unit_can_start(u);
3021
3022 case JOB_RELOAD:
3023 case JOB_TRY_RELOAD:
3024 return unit_can_reload(u);
3025
3026 case JOB_RELOAD_OR_START:
3027 return unit_can_reload(u) && unit_can_start(u);
3028
3029 default:
3030 assert_not_reached();
3031 }
3032}
3033
3034static Hashmap *unit_get_dependency_hashmap_per_type(Unit *u, UnitDependency d) {
3035 Hashmap *deps;
3036
3037 assert(u);
3038 assert(d >= 0 && d < _UNIT_DEPENDENCY_MAX);
3039
3040 deps = hashmap_get(u->dependencies, UNIT_DEPENDENCY_TO_PTR(d));
3041 if (!deps) {
3042 _cleanup_hashmap_free_ Hashmap *h = NULL;
3043
3044 h = hashmap_new(NULL);
3045 if (!h)
3046 return NULL;
3047
3048 if (hashmap_ensure_put(&u->dependencies, NULL, UNIT_DEPENDENCY_TO_PTR(d), h) < 0)
3049 return NULL;
3050
3051 deps = TAKE_PTR(h);
3052 }
3053
3054 return deps;
3055}
3056
3057typedef enum NotifyDependencyFlags {
3058 NOTIFY_DEPENDENCY_UPDATE_FROM = 1 << 0,
3059 NOTIFY_DEPENDENCY_UPDATE_TO = 1 << 1,
3060} NotifyDependencyFlags;
3061
3062static int unit_add_dependency_impl(
3063 Unit *u,
3064 UnitDependency d,
3065 Unit *other,
3066 UnitDependencyMask mask) {
3067
3068 static const UnitDependency inverse_table[_UNIT_DEPENDENCY_MAX] = {
3069 [UNIT_REQUIRES] = UNIT_REQUIRED_BY,
3070 [UNIT_REQUISITE] = UNIT_REQUISITE_OF,
3071 [UNIT_WANTS] = UNIT_WANTED_BY,
3072 [UNIT_BINDS_TO] = UNIT_BOUND_BY,
3073 [UNIT_PART_OF] = UNIT_CONSISTS_OF,
3074 [UNIT_UPHOLDS] = UNIT_UPHELD_BY,
3075 [UNIT_REQUIRED_BY] = UNIT_REQUIRES,
3076 [UNIT_REQUISITE_OF] = UNIT_REQUISITE,
3077 [UNIT_WANTED_BY] = UNIT_WANTS,
3078 [UNIT_BOUND_BY] = UNIT_BINDS_TO,
3079 [UNIT_CONSISTS_OF] = UNIT_PART_OF,
3080 [UNIT_UPHELD_BY] = UNIT_UPHOLDS,
3081 [UNIT_CONFLICTS] = UNIT_CONFLICTED_BY,
3082 [UNIT_CONFLICTED_BY] = UNIT_CONFLICTS,
3083 [UNIT_BEFORE] = UNIT_AFTER,
3084 [UNIT_AFTER] = UNIT_BEFORE,
3085 [UNIT_ON_SUCCESS] = UNIT_ON_SUCCESS_OF,
3086 [UNIT_ON_SUCCESS_OF] = UNIT_ON_SUCCESS,
3087 [UNIT_ON_FAILURE] = UNIT_ON_FAILURE_OF,
3088 [UNIT_ON_FAILURE_OF] = UNIT_ON_FAILURE,
3089 [UNIT_TRIGGERS] = UNIT_TRIGGERED_BY,
3090 [UNIT_TRIGGERED_BY] = UNIT_TRIGGERS,
3091 [UNIT_PROPAGATES_RELOAD_TO] = UNIT_RELOAD_PROPAGATED_FROM,
3092 [UNIT_RELOAD_PROPAGATED_FROM] = UNIT_PROPAGATES_RELOAD_TO,
3093 [UNIT_PROPAGATES_STOP_TO] = UNIT_STOP_PROPAGATED_FROM,
3094 [UNIT_STOP_PROPAGATED_FROM] = UNIT_PROPAGATES_STOP_TO,
3095 [UNIT_JOINS_NAMESPACE_OF] = UNIT_JOINS_NAMESPACE_OF, /* symmetric! 👓 */
3096 [UNIT_REFERENCES] = UNIT_REFERENCED_BY,
3097 [UNIT_REFERENCED_BY] = UNIT_REFERENCES,
3098 [UNIT_IN_SLICE] = UNIT_SLICE_OF,
3099 [UNIT_SLICE_OF] = UNIT_IN_SLICE,
3100 };
3101
3102 Hashmap *u_deps, *other_deps;
3103 UnitDependencyInfo u_info, u_info_old, other_info, other_info_old;
3104 NotifyDependencyFlags flags = 0;
3105 int r;
3106
3107 assert(u);
3108 assert(other);
3109 assert(d >= 0 && d < _UNIT_DEPENDENCY_MAX);
3110 assert(inverse_table[d] >= 0 && inverse_table[d] < _UNIT_DEPENDENCY_MAX);
3111 assert(mask > 0 && mask < _UNIT_DEPENDENCY_MASK_FULL);
3112
3113 /* Ensure the following two hashmaps for each unit exist:
3114 * - the top-level dependency hashmap that maps UnitDependency → Hashmap(Unit* → UnitDependencyInfo),
3115 * - the inner hashmap, that maps Unit* → UnitDependencyInfo, for the specified dependency type. */
3116 u_deps = unit_get_dependency_hashmap_per_type(u, d);
3117 if (!u_deps)
3118 return -ENOMEM;
3119
3120 other_deps = unit_get_dependency_hashmap_per_type(other, inverse_table[d]);
3121 if (!other_deps)
3122 return -ENOMEM;
3123
3124 /* Save the original dependency info. */
3125 u_info.data = u_info_old.data = hashmap_get(u_deps, other);
3126 other_info.data = other_info_old.data = hashmap_get(other_deps, u);
3127
3128 /* Update dependency info. */
3129 u_info.origin_mask |= mask;
3130 other_info.destination_mask |= mask;
3131
3132 /* Save updated dependency info. */
3133 if (u_info.data != u_info_old.data) {
3134 r = hashmap_replace(u_deps, other, u_info.data);
3135 if (r < 0)
3136 return r;
3137
3138 flags = NOTIFY_DEPENDENCY_UPDATE_FROM;
3139 }
3140
3141 if (other_info.data != other_info_old.data) {
3142 r = hashmap_replace(other_deps, u, other_info.data);
3143 if (r < 0) {
3144 if (u_info.data != u_info_old.data) {
3145 /* Restore the old dependency. */
3146 if (u_info_old.data)
3147 (void) hashmap_update(u_deps, other, u_info_old.data);
3148 else
3149 hashmap_remove(u_deps, other);
3150 }
3151 return r;
3152 }
3153
3154 flags |= NOTIFY_DEPENDENCY_UPDATE_TO;
3155 }
3156
3157 return flags;
3158}
3159
3160int unit_add_dependency(
3161 Unit *u,
3162 UnitDependency d,
3163 Unit *other,
3164 bool add_reference,
3165 UnitDependencyMask mask) {
3166
3167 UnitDependencyAtom a;
3168 int r;
3169
3170 /* Helper to know whether sending a notification is necessary or not: if the dependency is already
3171 * there, no need to notify! */
3172 NotifyDependencyFlags notify_flags;
3173
3174 assert(u);
3175 assert(d >= 0 && d < _UNIT_DEPENDENCY_MAX);
3176 assert(other);
3177
3178 u = unit_follow_merge(u);
3179 other = unit_follow_merge(other);
3180 a = unit_dependency_to_atom(d);
3181 assert(a >= 0);
3182
3183 /* We won't allow dependencies on ourselves. We will not consider them an error however. */
3184 if (u == other) {
3185 if (unit_should_warn_about_dependency(d))
3186 log_unit_warning(u, "Dependency %s=%s is dropped.",
3187 unit_dependency_to_string(d), u->id);
3188 return 0;
3189 }
3190
3191 if (u->manager && FLAGS_SET(u->manager->test_run_flags, MANAGER_TEST_RUN_IGNORE_DEPENDENCIES))
3192 return 0;
3193
3194 /* Note that ordering a device unit after a unit is permitted since it allows its job running
3195 * timeout to be started at a specific time. */
3196 if (FLAGS_SET(a, UNIT_ATOM_BEFORE) && other->type == UNIT_DEVICE) {
3197 log_unit_warning(u, "Dependency Before=%s ignored (.device units cannot be delayed)", other->id);
3198 return 0;
3199 }
3200
3201 if (FLAGS_SET(a, UNIT_ATOM_ON_FAILURE) && !UNIT_VTABLE(u)->can_fail) {
3202 log_unit_warning(u, "Requested dependency OnFailure=%s ignored (%s units cannot fail).", other->id, unit_type_to_string(u->type));
3203 return 0;
3204 }
3205
3206 if (FLAGS_SET(a, UNIT_ATOM_TRIGGERS) && !UNIT_VTABLE(u)->can_trigger)
3207 return log_unit_error_errno(u, SYNTHETIC_ERRNO(EINVAL),
3208 "Requested dependency Triggers=%s refused (%s units cannot trigger other units).", other->id, unit_type_to_string(u->type));
3209 if (FLAGS_SET(a, UNIT_ATOM_TRIGGERED_BY) && !UNIT_VTABLE(other)->can_trigger)
3210 return log_unit_error_errno(u, SYNTHETIC_ERRNO(EINVAL),
3211 "Requested dependency TriggeredBy=%s refused (%s units cannot trigger other units).", other->id, unit_type_to_string(other->type));
3212
3213 if (FLAGS_SET(a, UNIT_ATOM_IN_SLICE) && other->type != UNIT_SLICE)
3214 return log_unit_error_errno(u, SYNTHETIC_ERRNO(EINVAL),
3215 "Requested dependency Slice=%s refused (%s is not a slice unit).", other->id, other->id);
3216 if (FLAGS_SET(a, UNIT_ATOM_SLICE_OF) && u->type != UNIT_SLICE)
3217 return log_unit_error_errno(u, SYNTHETIC_ERRNO(EINVAL),
3218 "Requested dependency SliceOf=%s refused (%s is not a slice unit).", other->id, u->id);
3219
3220 if (FLAGS_SET(a, UNIT_ATOM_IN_SLICE) && !UNIT_HAS_CGROUP_CONTEXT(u))
3221 return log_unit_error_errno(u, SYNTHETIC_ERRNO(EINVAL),
3222 "Requested dependency Slice=%s refused (%s is not a cgroup unit).", other->id, u->id);
3223
3224 if (FLAGS_SET(a, UNIT_ATOM_SLICE_OF) && !UNIT_HAS_CGROUP_CONTEXT(other))
3225 return log_unit_error_errno(u, SYNTHETIC_ERRNO(EINVAL),
3226 "Requested dependency SliceOf=%s refused (%s is not a cgroup unit).", other->id, other->id);
3227
3228 r = unit_add_dependency_impl(u, d, other, mask);
3229 if (r < 0)
3230 return r;
3231 notify_flags = r;
3232
3233 if (add_reference) {
3234 r = unit_add_dependency_impl(u, UNIT_REFERENCES, other, mask);
3235 if (r < 0)
3236 return r;
3237 notify_flags |= r;
3238 }
3239
3240 if (FLAGS_SET(notify_flags, NOTIFY_DEPENDENCY_UPDATE_FROM))
3241 unit_add_to_dbus_queue(u);
3242 if (FLAGS_SET(notify_flags, NOTIFY_DEPENDENCY_UPDATE_TO))
3243 unit_add_to_dbus_queue(other);
3244
3245 return notify_flags != 0;
3246}
3247
3248int unit_add_two_dependencies(Unit *u, UnitDependency d, UnitDependency e, Unit *other, bool add_reference, UnitDependencyMask mask) {
3249 int r = 0, s = 0;
3250
3251 assert(u);
3252 assert(d >= 0 || e >= 0);
3253
3254 if (d >= 0) {
3255 r = unit_add_dependency(u, d, other, add_reference, mask);
3256 if (r < 0)
3257 return r;
3258 }
3259
3260 if (e >= 0) {
3261 s = unit_add_dependency(u, e, other, add_reference, mask);
3262 if (s < 0)
3263 return s;
3264 }
3265
3266 return r > 0 || s > 0;
3267}
3268
3269static int resolve_template(Unit *u, const char *name, char **buf, const char **ret) {
3270 int r;
3271
3272 assert(u);
3273 assert(name);
3274 assert(buf);
3275 assert(ret);
3276
3277 if (!unit_name_is_valid(name, UNIT_NAME_TEMPLATE)) {
3278 *buf = NULL;
3279 *ret = name;
3280 return 0;
3281 }
3282
3283 if (u->instance)
3284 r = unit_name_replace_instance(name, u->instance, buf);
3285 else {
3286 _cleanup_free_ char *i = NULL;
3287
3288 r = unit_name_to_prefix(u->id, &i);
3289 if (r < 0)
3290 return r;
3291
3292 r = unit_name_replace_instance(name, i, buf);
3293 }
3294 if (r < 0)
3295 return r;
3296
3297 *ret = *buf;
3298 return 0;
3299}
3300
3301int unit_add_dependency_by_name(Unit *u, UnitDependency d, const char *name, bool add_reference, UnitDependencyMask mask) {
3302 _cleanup_free_ char *buf = NULL;
3303 Unit *other;
3304 int r;
3305
3306 assert(u);
3307 assert(name);
3308
3309 r = resolve_template(u, name, &buf, &name);
3310 if (r < 0)
3311 return r;
3312
3313 if (u->manager && FLAGS_SET(u->manager->test_run_flags, MANAGER_TEST_RUN_IGNORE_DEPENDENCIES))
3314 return 0;
3315
3316 r = manager_load_unit(u->manager, name, NULL, NULL, &other);
3317 if (r < 0)
3318 return r;
3319
3320 return unit_add_dependency(u, d, other, add_reference, mask);
3321}
3322
3323int unit_add_two_dependencies_by_name(Unit *u, UnitDependency d, UnitDependency e, const char *name, bool add_reference, UnitDependencyMask mask) {
3324 _cleanup_free_ char *buf = NULL;
3325 Unit *other;
3326 int r;
3327
3328 assert(u);
3329 assert(name);
3330
3331 r = resolve_template(u, name, &buf, &name);
3332 if (r < 0)
3333 return r;
3334
3335 if (u->manager && FLAGS_SET(u->manager->test_run_flags, MANAGER_TEST_RUN_IGNORE_DEPENDENCIES))
3336 return 0;
3337
3338 r = manager_load_unit(u->manager, name, NULL, NULL, &other);
3339 if (r < 0)
3340 return r;
3341
3342 return unit_add_two_dependencies(u, d, e, other, add_reference, mask);
3343}
3344
3345int set_unit_path(const char *p) {
3346 /* This is mostly for debug purposes */
3347 return RET_NERRNO(setenv("SYSTEMD_UNIT_PATH", p, 1));
3348}
3349
3350char *unit_dbus_path(Unit *u) {
3351 assert(u);
3352
3353 if (!u->id)
3354 return NULL;
3355
3356 return unit_dbus_path_from_name(u->id);
3357}
3358
3359char *unit_dbus_path_invocation_id(Unit *u) {
3360 assert(u);
3361
3362 if (sd_id128_is_null(u->invocation_id))
3363 return NULL;
3364
3365 return unit_dbus_path_from_name(u->invocation_id_string);
3366}
3367
3368int unit_set_invocation_id(Unit *u, sd_id128_t id) {
3369 int r;
3370
3371 assert(u);
3372
3373 /* Set the invocation ID for this unit. If we cannot, this will not roll back, but reset the whole thing. */
3374
3375 if (sd_id128_equal(u->invocation_id, id))
3376 return 0;
3377
3378 if (!sd_id128_is_null(u->invocation_id))
3379 (void) hashmap_remove_value(u->manager->units_by_invocation_id, &u->invocation_id, u);
3380
3381 if (sd_id128_is_null(id)) {
3382 r = 0;
3383 goto reset;
3384 }
3385
3386 r = hashmap_ensure_allocated(&u->manager->units_by_invocation_id, &id128_hash_ops);
3387 if (r < 0)
3388 goto reset;
3389
3390 u->invocation_id = id;
3391 sd_id128_to_string(id, u->invocation_id_string);
3392
3393 r = hashmap_put(u->manager->units_by_invocation_id, &u->invocation_id, u);
3394 if (r < 0)
3395 goto reset;
3396
3397 return 0;
3398
3399reset:
3400 u->invocation_id = SD_ID128_NULL;
3401 u->invocation_id_string[0] = 0;
3402 return r;
3403}
3404
3405int unit_set_slice(Unit *u, Unit *slice) {
3406 int r;
3407
3408 assert(u);
3409 assert(slice);
3410
3411 /* Sets the unit slice if it has not been set before. Is extra careful, to only allow this for units
3412 * that actually have a cgroup context. Also, we don't allow to set this for slices (since the parent
3413 * slice is derived from the name). Make sure the unit we set is actually a slice. */
3414
3415 if (!UNIT_HAS_CGROUP_CONTEXT(u))
3416 return -EOPNOTSUPP;
3417
3418 if (u->type == UNIT_SLICE)
3419 return -EINVAL;
3420
3421 if (unit_active_state(u) != UNIT_INACTIVE)
3422 return -EBUSY;
3423
3424 if (slice->type != UNIT_SLICE)
3425 return -EINVAL;
3426
3427 if (unit_has_name(u, SPECIAL_INIT_SCOPE) &&
3428 !unit_has_name(slice, SPECIAL_ROOT_SLICE))
3429 return -EPERM;
3430
3431 if (UNIT_GET_SLICE(u) == slice)
3432 return 0;
3433
3434 /* Disallow slice changes if @u is already bound to cgroups */
3435 if (UNIT_GET_SLICE(u)) {
3436 CGroupRuntime *crt = unit_get_cgroup_runtime(u);
3437 if (crt && crt->cgroup_realized)
3438 return -EBUSY;
3439 }
3440
3441 /* Remove any slices assigned prior; we should only have one UNIT_IN_SLICE dependency */
3442 if (UNIT_GET_SLICE(u))
3443 unit_remove_dependencies(u, UNIT_DEPENDENCY_SLICE_PROPERTY);
3444
3445 r = unit_add_dependency(u, UNIT_IN_SLICE, slice, true, UNIT_DEPENDENCY_SLICE_PROPERTY);
3446 if (r < 0)
3447 return r;
3448
3449 return 1;
3450}
3451
3452int unit_set_default_slice(Unit *u) {
3453 const char *slice_name;
3454 Unit *slice;
3455 int r;
3456
3457 assert(u);
3458
3459 if (u->manager && FLAGS_SET(u->manager->test_run_flags, MANAGER_TEST_RUN_IGNORE_DEPENDENCIES))
3460 return 0;
3461
3462 if (UNIT_GET_SLICE(u))
3463 return 0;
3464
3465 if (u->instance) {
3466 _cleanup_free_ char *prefix = NULL, *escaped = NULL;
3467
3468 /* Implicitly place all instantiated units in their
3469 * own per-template slice */
3470
3471 r = unit_name_to_prefix(u->id, &prefix);
3472 if (r < 0)
3473 return r;
3474
3475 /* The prefix is already escaped, but it might include
3476 * "-" which has a special meaning for slice units,
3477 * hence escape it here extra. */
3478 escaped = unit_name_escape(prefix);
3479 if (!escaped)
3480 return -ENOMEM;
3481
3482 if (MANAGER_IS_SYSTEM(u->manager))
3483 slice_name = strjoina("system-", escaped, ".slice");
3484 else
3485 slice_name = strjoina("app-", escaped, ".slice");
3486
3487 } else if (unit_is_extrinsic(u))
3488 /* Keep all extrinsic units (e.g. perpetual units and swap and mount units in user mode) in
3489 * the root slice. They don't really belong in one of the subslices. */
3490 slice_name = SPECIAL_ROOT_SLICE;
3491
3492 else if (MANAGER_IS_SYSTEM(u->manager))
3493 slice_name = SPECIAL_SYSTEM_SLICE;
3494 else
3495 slice_name = SPECIAL_APP_SLICE;
3496
3497 r = manager_load_unit(u->manager, slice_name, NULL, NULL, &slice);
3498 if (r < 0)
3499 return r;
3500
3501 return unit_set_slice(u, slice);
3502}
3503
3504const char *unit_slice_name(Unit *u) {
3505 Unit *slice;
3506 assert(u);
3507
3508 slice = UNIT_GET_SLICE(u);
3509 if (!slice)
3510 return NULL;
3511
3512 return slice->id;
3513}
3514
3515int unit_load_related_unit(Unit *u, const char *type, Unit **_found) {
3516 _cleanup_free_ char *t = NULL;
3517 int r;
3518
3519 assert(u);
3520 assert(type);
3521 assert(_found);
3522
3523 r = unit_name_change_suffix(u->id, type, &t);
3524 if (r < 0)
3525 return r;
3526 if (unit_has_name(u, t))
3527 return -EINVAL;
3528
3529 r = manager_load_unit(u->manager, t, NULL, NULL, _found);
3530 assert(r < 0 || *_found != u);
3531 return r;
3532}
3533
3534static int signal_name_owner_changed(sd_bus_message *message, void *userdata, sd_bus_error *error) {
3535 const char *new_owner;
3536 Unit *u = ASSERT_PTR(userdata);
3537 int r;
3538
3539 assert(message);
3540
3541 r = sd_bus_message_read(message, "sss", NULL, NULL, &new_owner);
3542 if (r < 0) {
3543 bus_log_parse_error(r);
3544 return 0;
3545 }
3546
3547 if (UNIT_VTABLE(u)->bus_name_owner_change)
3548 UNIT_VTABLE(u)->bus_name_owner_change(u, empty_to_null(new_owner));
3549
3550 return 0;
3551}
3552
3553static int get_name_owner_handler(sd_bus_message *message, void *userdata, sd_bus_error *error) {
3554 const sd_bus_error *e;
3555 const char *new_owner;
3556 Unit *u = ASSERT_PTR(userdata);
3557 int r;
3558
3559 assert(message);
3560
3561 u->get_name_owner_slot = sd_bus_slot_unref(u->get_name_owner_slot);
3562
3563 e = sd_bus_message_get_error(message);
3564 if (e) {
3565 if (!sd_bus_error_has_name(e, SD_BUS_ERROR_NAME_HAS_NO_OWNER)) {
3566 r = sd_bus_error_get_errno(e);
3567 log_unit_error_errno(u, r,
3568 "Unexpected error response from GetNameOwner(): %s",
3569 bus_error_message(e, r));
3570 }
3571
3572 new_owner = NULL;
3573 } else {
3574 r = sd_bus_message_read(message, "s", &new_owner);
3575 if (r < 0)
3576 return bus_log_parse_error(r);
3577
3578 assert(!isempty(new_owner));
3579 }
3580
3581 if (UNIT_VTABLE(u)->bus_name_owner_change)
3582 UNIT_VTABLE(u)->bus_name_owner_change(u, new_owner);
3583
3584 return 0;
3585}
3586
3587int unit_install_bus_match(Unit *u, sd_bus *bus, const char *name) {
3588 _cleanup_(sd_bus_message_unrefp) sd_bus_message *m = NULL;
3589 const char *match;
3590 usec_t timeout_usec = 0;
3591 int r;
3592
3593 assert(u);
3594 assert(bus);
3595 assert(name);
3596
3597 if (u->match_bus_slot || u->get_name_owner_slot)
3598 return -EBUSY;
3599
3600 /* NameOwnerChanged and GetNameOwner is used to detect when a service finished starting up. The dbus
3601 * call timeout shouldn't be earlier than that. If we couldn't get the start timeout, use the default
3602 * value defined above. */
3603 if (UNIT_VTABLE(u)->get_timeout_start_usec)
3604 timeout_usec = UNIT_VTABLE(u)->get_timeout_start_usec(u);
3605
3606 match = strjoina("type='signal',"
3607 "sender='org.freedesktop.DBus',"
3608 "path='/org/freedesktop/DBus',"
3609 "interface='org.freedesktop.DBus',"
3610 "member='NameOwnerChanged',"
3611 "arg0='", name, "'");
3612
3613 r = bus_add_match_full(
3614 bus,
3615 &u->match_bus_slot,
3616 true,
3617 match,
3618 signal_name_owner_changed,
3619 NULL,
3620 u,
3621 timeout_usec);
3622 if (r < 0)
3623 return r;
3624
3625 r = sd_bus_message_new_method_call(
3626 bus,
3627 &m,
3628 "org.freedesktop.DBus",
3629 "/org/freedesktop/DBus",
3630 "org.freedesktop.DBus",
3631 "GetNameOwner");
3632 if (r < 0)
3633 return r;
3634
3635 r = sd_bus_message_append(m, "s", name);
3636 if (r < 0)
3637 return r;
3638
3639 r = sd_bus_call_async(
3640 bus,
3641 &u->get_name_owner_slot,
3642 m,
3643 get_name_owner_handler,
3644 u,
3645 timeout_usec);
3646
3647 if (r < 0) {
3648 u->match_bus_slot = sd_bus_slot_unref(u->match_bus_slot);
3649 return r;
3650 }
3651
3652 log_unit_debug(u, "Watching D-Bus name '%s'.", name);
3653 return 0;
3654}
3655
3656int unit_watch_bus_name(Unit *u, const char *name) {
3657 int r;
3658
3659 assert(u);
3660 assert(name);
3661
3662 /* Watch a specific name on the bus. We only support one unit
3663 * watching each name for now. */
3664
3665 if (u->manager->api_bus) {
3666 /* If the bus is already available, install the match directly.
3667 * Otherwise, just put the name in the list. bus_setup_api() will take care later. */
3668 r = unit_install_bus_match(u, u->manager->api_bus, name);
3669 if (r < 0)
3670 return log_warning_errno(r, "Failed to subscribe to NameOwnerChanged signal for '%s': %m", name);
3671 }
3672
3673 r = hashmap_put(u->manager->watch_bus, name, u);
3674 if (r < 0) {
3675 u->match_bus_slot = sd_bus_slot_unref(u->match_bus_slot);
3676 u->get_name_owner_slot = sd_bus_slot_unref(u->get_name_owner_slot);
3677 return log_warning_errno(r, "Failed to put bus name to hashmap: %m");
3678 }
3679
3680 return 0;
3681}
3682
3683void unit_unwatch_bus_name(Unit *u, const char *name) {
3684 assert(u);
3685 assert(name);
3686
3687 (void) hashmap_remove_value(u->manager->watch_bus, name, u);
3688 u->match_bus_slot = sd_bus_slot_unref(u->match_bus_slot);
3689 u->get_name_owner_slot = sd_bus_slot_unref(u->get_name_owner_slot);
3690}
3691
3692int unit_add_node_dependency(Unit *u, const char *what, UnitDependency dep, UnitDependencyMask mask) {
3693 _cleanup_free_ char *e = NULL;
3694 Unit *device;
3695 int r;
3696
3697 assert(u);
3698
3699 /* Adds in links to the device node that this unit is based on */
3700 if (isempty(what))
3701 return 0;
3702
3703 if (!is_device_path(what))
3704 return 0;
3705
3706 /* When device units aren't supported (such as in a container), don't create dependencies on them. */
3707 if (!unit_type_supported(UNIT_DEVICE))
3708 return 0;
3709
3710 r = unit_name_from_path(what, ".device", &e);
3711 if (r < 0)
3712 return r;
3713
3714 r = manager_load_unit(u->manager, e, NULL, NULL, &device);
3715 if (r < 0)
3716 return r;
3717
3718 if (dep == UNIT_REQUIRES && device_shall_be_bound_by(device, u))
3719 dep = UNIT_BINDS_TO;
3720
3721 return unit_add_two_dependencies(u, UNIT_AFTER,
3722 MANAGER_IS_SYSTEM(u->manager) ? dep : UNIT_WANTS,
3723 device, true, mask);
3724}
3725
3726int unit_add_blockdev_dependency(Unit *u, const char *what, UnitDependencyMask mask) {
3727 _cleanup_free_ char *escaped = NULL, *target = NULL;
3728 int r;
3729
3730 assert(u);
3731
3732 if (isempty(what))
3733 return 0;
3734
3735 if (!path_startswith(what, "/dev/"))
3736 return 0;
3737
3738 /* If we don't support devices, then also don't bother with blockdev@.target */
3739 if (!unit_type_supported(UNIT_DEVICE))
3740 return 0;
3741
3742 r = unit_name_path_escape(what, &escaped);
3743 if (r < 0)
3744 return r;
3745
3746 r = unit_name_build("blockdev", escaped, ".target", &target);
3747 if (r < 0)
3748 return r;
3749
3750 return unit_add_dependency_by_name(u, UNIT_AFTER, target, true, mask);
3751}
3752
3753int unit_coldplug(Unit *u) {
3754 int r = 0;
3755
3756 assert(u);
3757
3758 /* Make sure we don't enter a loop, when coldplugging recursively. */
3759 if (u->coldplugged)
3760 return 0;
3761
3762 u->coldplugged = true;
3763
3764 STRV_FOREACH(i, u->deserialized_refs)
3765 RET_GATHER(r, bus_unit_track_add_name(u, *i));
3766
3767 u->deserialized_refs = strv_free(u->deserialized_refs);
3768
3769 if (UNIT_VTABLE(u)->coldplug)
3770 RET_GATHER(r, UNIT_VTABLE(u)->coldplug(u));
3771
3772 if (u->job)
3773 RET_GATHER(r, job_coldplug(u->job));
3774 if (u->nop_job)
3775 RET_GATHER(r, job_coldplug(u->nop_job));
3776
3777 unit_modify_nft_set(u, /* add = */ true);
3778 return r;
3779}
3780
3781void unit_catchup(Unit *u) {
3782 assert(u);
3783
3784 if (UNIT_VTABLE(u)->catchup)
3785 UNIT_VTABLE(u)->catchup(u);
3786
3787 unit_cgroup_catchup(u);
3788}
3789
3790static bool fragment_mtime_newer(const char *path, usec_t mtime, bool path_masked) {
3791 struct stat st;
3792
3793 if (!path)
3794 return false;
3795
3796 /* If the source is some virtual kernel file system, then we assume we watch it anyway, and hence pretend we
3797 * are never out-of-date. */
3798 if (PATH_STARTSWITH_SET(path, "/proc", "/sys"))
3799 return false;
3800
3801 if (stat(path, &st) < 0)
3802 /* What, cannot access this anymore? */
3803 return true;
3804
3805 if (path_masked)
3806 /* For masked files check if they are still so */
3807 return !null_or_empty(&st);
3808 else
3809 /* For non-empty files check the mtime */
3810 return timespec_load(&st.st_mtim) > mtime;
3811
3812 return false;
3813}
3814
3815bool unit_need_daemon_reload(Unit *u) {
3816 _cleanup_strv_free_ char **dropins = NULL;
3817
3818 assert(u);
3819 assert(u->manager);
3820
3821 if (u->manager->unit_file_state_outdated)
3822 return true;
3823
3824 /* For unit files, we allow masking… */
3825 if (fragment_mtime_newer(u->fragment_path, u->fragment_mtime,
3826 u->load_state == UNIT_MASKED))
3827 return true;
3828
3829 /* Source paths should not be masked… */
3830 if (fragment_mtime_newer(u->source_path, u->source_mtime, false))
3831 return true;
3832
3833 if (u->load_state == UNIT_LOADED)
3834 (void) unit_find_dropin_paths(u, &dropins);
3835 if (!strv_equal(u->dropin_paths, dropins))
3836 return true;
3837
3838 /* … any drop-ins that are masked are simply omitted from the list. */
3839 STRV_FOREACH(path, u->dropin_paths)
3840 if (fragment_mtime_newer(*path, u->dropin_mtime, false))
3841 return true;
3842
3843 return false;
3844}
3845
3846void unit_reset_failed(Unit *u) {
3847 assert(u);
3848
3849 if (UNIT_VTABLE(u)->reset_failed)
3850 UNIT_VTABLE(u)->reset_failed(u);
3851
3852 ratelimit_reset(&u->start_ratelimit);
3853 u->start_limit_hit = false;
3854}
3855
3856Unit *unit_following(Unit *u) {
3857 assert(u);
3858
3859 if (UNIT_VTABLE(u)->following)
3860 return UNIT_VTABLE(u)->following(u);
3861
3862 return NULL;
3863}
3864
3865bool unit_stop_pending(Unit *u) {
3866 assert(u);
3867
3868 /* This call does check the current state of the unit. It's
3869 * hence useful to be called from state change calls of the
3870 * unit itself, where the state isn't updated yet. This is
3871 * different from unit_inactive_or_pending() which checks both
3872 * the current state and for a queued job. */
3873
3874 return unit_has_job_type(u, JOB_STOP);
3875}
3876
3877bool unit_inactive_or_pending(Unit *u) {
3878 assert(u);
3879
3880 /* Returns true if the unit is inactive or going down */
3881
3882 if (UNIT_IS_INACTIVE_OR_DEACTIVATING(unit_active_state(u)))
3883 return true;
3884
3885 if (unit_stop_pending(u))
3886 return true;
3887
3888 return false;
3889}
3890
3891bool unit_active_or_pending(Unit *u) {
3892 assert(u);
3893
3894 /* Returns true if the unit is active or going up */
3895
3896 if (UNIT_IS_ACTIVE_OR_ACTIVATING(unit_active_state(u)))
3897 return true;
3898
3899 if (u->job &&
3900 IN_SET(u->job->type, JOB_START, JOB_RELOAD_OR_START, JOB_RESTART))
3901 return true;
3902
3903 return false;
3904}
3905
3906bool unit_will_restart_default(Unit *u) {
3907 assert(u);
3908
3909 return unit_has_job_type(u, JOB_START);
3910}
3911
3912bool unit_will_restart(Unit *u) {
3913 assert(u);
3914
3915 if (!UNIT_VTABLE(u)->will_restart)
3916 return false;
3917
3918 return UNIT_VTABLE(u)->will_restart(u);
3919}
3920
3921void unit_notify_cgroup_oom(Unit *u, bool managed_oom) {
3922 assert(u);
3923
3924 if (UNIT_VTABLE(u)->notify_cgroup_oom)
3925 UNIT_VTABLE(u)->notify_cgroup_oom(u, managed_oom);
3926}
3927
3928static int unit_pid_set(Unit *u, Set **pid_set) {
3929 int r;
3930
3931 assert(u);
3932 assert(pid_set);
3933
3934 set_clear(*pid_set); /* This updates input. */
3935
3936 /* Exclude the main/control pids from being killed via the cgroup */
3937
3938 PidRef *pid;
3939 FOREACH_ARGUMENT(pid, unit_main_pid(u), unit_control_pid(u))
3940 if (pidref_is_set(pid)) {
3941 r = set_ensure_put(pid_set, NULL, PID_TO_PTR(pid->pid));
3942 if (r < 0)
3943 return r;
3944 }
3945
3946 return 0;
3947}
3948
3949static int kill_common_log(const PidRef *pid, int signo, void *userdata) {
3950 _cleanup_free_ char *comm = NULL;
3951 Unit *u = ASSERT_PTR(userdata);
3952
3953 (void) pidref_get_comm(pid, &comm);
3954
3955 log_unit_info(u, "Sending signal SIG%s to process " PID_FMT " (%s) on client request.",
3956 signal_to_string(signo), pid->pid, strna(comm));
3957
3958 return 1;
3959}
3960
3961static int kill_or_sigqueue(PidRef* pidref, int signo, int code, int value) {
3962 assert(pidref_is_set(pidref));
3963 assert(SIGNAL_VALID(signo));
3964
3965 switch (code) {
3966
3967 case SI_USER:
3968 log_debug("Killing " PID_FMT " with signal SIG%s.", pidref->pid, signal_to_string(signo));
3969 return pidref_kill(pidref, signo);
3970
3971 case SI_QUEUE:
3972 log_debug("Enqueuing value %i to " PID_FMT " on signal SIG%s.", value, pidref->pid, signal_to_string(signo));
3973 return pidref_sigqueue(pidref, signo, value);
3974
3975 default:
3976 assert_not_reached();
3977 }
3978}
3979
3980static int unit_kill_one(
3981 Unit *u,
3982 PidRef *pidref,
3983 const char *type,
3984 int signo,
3985 int code,
3986 int value,
3987 sd_bus_error *ret_error) {
3988
3989 int r;
3990
3991 assert(u);
3992 assert(type);
3993
3994 if (!pidref_is_set(pidref))
3995 return 0;
3996
3997 _cleanup_free_ char *comm = NULL;
3998 (void) pidref_get_comm(pidref, &comm);
3999
4000 r = kill_or_sigqueue(pidref, signo, code, value);
4001 if (r == -ESRCH)
4002 return 0;
4003 if (r < 0) {
4004 /* Report this failure both to the logs and to the client */
4005 if (ret_error)
4006 sd_bus_error_set_errnof(
4007 ret_error, r,
4008 "Failed to send signal SIG%s to %s process " PID_FMT " (%s): %m",
4009 signal_to_string(signo), type, pidref->pid, strna(comm));
4010
4011 return log_unit_warning_errno(
4012 u, r,
4013 "Failed to send signal SIG%s to %s process " PID_FMT " (%s) on client request: %m",
4014 signal_to_string(signo), type, pidref->pid, strna(comm));
4015 }
4016
4017 log_unit_info(u, "Sent signal SIG%s to %s process " PID_FMT " (%s) on client request.",
4018 signal_to_string(signo), type, pidref->pid, strna(comm));
4019 return 1; /* killed */
4020}
4021
4022int unit_kill(
4023 Unit *u,
4024 KillWho who,
4025 int signo,
4026 int code,
4027 int value,
4028 sd_bus_error *ret_error) {
4029
4030 PidRef *main_pid, *control_pid;
4031 bool killed = false;
4032 int ret = 0, r;
4033
4034 /* This is the common implementation for explicit user-requested killing of unit processes, shared by
4035 * various unit types. Do not confuse with unit_kill_context(), which is what we use when we want to
4036 * stop a service ourselves. */
4037
4038 assert(u);
4039 assert(who >= 0);
4040 assert(who < _KILL_WHO_MAX);
4041 assert(SIGNAL_VALID(signo));
4042 assert(IN_SET(code, SI_USER, SI_QUEUE));
4043
4044 main_pid = unit_main_pid(u);
4045 control_pid = unit_control_pid(u);
4046
4047 if (!UNIT_HAS_CGROUP_CONTEXT(u) && !main_pid && !control_pid)
4048 return sd_bus_error_setf(ret_error, SD_BUS_ERROR_NOT_SUPPORTED, "Unit type does not support process killing.");
4049
4050 if (IN_SET(who, KILL_MAIN, KILL_MAIN_FAIL)) {
4051 if (!main_pid)
4052 return sd_bus_error_setf(ret_error, BUS_ERROR_NO_SUCH_PROCESS, "%s units have no main processes", unit_type_to_string(u->type));
4053 if (!pidref_is_set(main_pid))
4054 return sd_bus_error_set_const(ret_error, BUS_ERROR_NO_SUCH_PROCESS, "No main process to kill");
4055 }
4056
4057 if (IN_SET(who, KILL_CONTROL, KILL_CONTROL_FAIL)) {
4058 if (!control_pid)
4059 return sd_bus_error_setf(ret_error, BUS_ERROR_NO_SUCH_PROCESS, "%s units have no control processes", unit_type_to_string(u->type));
4060 if (!pidref_is_set(control_pid))
4061 return sd_bus_error_set_const(ret_error, BUS_ERROR_NO_SUCH_PROCESS, "No control process to kill");
4062 }
4063
4064 if (IN_SET(who, KILL_CONTROL, KILL_CONTROL_FAIL, KILL_ALL, KILL_ALL_FAIL)) {
4065 r = unit_kill_one(u, control_pid, "control", signo, code, value, ret_error);
4066 RET_GATHER(ret, r);
4067 killed = killed || r > 0;
4068 }
4069
4070 if (IN_SET(who, KILL_MAIN, KILL_MAIN_FAIL, KILL_ALL, KILL_ALL_FAIL)) {
4071 r = unit_kill_one(u, main_pid, "main", signo, code, value, ret >= 0 ? ret_error : NULL);
4072 RET_GATHER(ret, r);
4073 killed = killed || r > 0;
4074 }
4075
4076 /* Note: if we shall enqueue rather than kill we won't do this via the cgroup mechanism, since it
4077 * doesn't really make much sense (and given that enqueued values are a relatively expensive
4078 * resource, and we shouldn't allow us to be subjects for such allocation sprees) */
4079 if (IN_SET(who, KILL_ALL, KILL_ALL_FAIL) && code == SI_USER) {
4080 CGroupRuntime *crt = unit_get_cgroup_runtime(u);
4081
4082 if (crt && crt->cgroup_path) {
4083 _cleanup_set_free_ Set *pid_set = NULL;
4084
4085 /* Exclude the main/control pids from being killed via the cgroup */
4086 r = unit_pid_set(u, &pid_set);
4087 if (r < 0)
4088 return log_oom();
4089
4090 r = cg_kill_recursive(crt->cgroup_path, signo, 0, pid_set, kill_common_log, u);
4091 if (r < 0 && !IN_SET(r, -ESRCH, -ENOENT)) {
4092 if (ret >= 0)
4093 sd_bus_error_set_errnof(
4094 ret_error, r,
4095 "Failed to send signal SIG%s to auxiliary processes: %m",
4096 signal_to_string(signo));
4097
4098 log_unit_warning_errno(
4099 u, r,
4100 "Failed to send signal SIG%s to auxiliary processes on client request: %m",
4101 signal_to_string(signo));
4102
4103 RET_GATHER(ret, r);
4104 }
4105
4106 killed = killed || r >= 0;
4107 }
4108 }
4109
4110 /* If the "fail" versions of the operation are requested, then complain if the set of processes we killed is empty */
4111 if (ret >= 0 && !killed && IN_SET(who, KILL_ALL_FAIL, KILL_CONTROL_FAIL, KILL_MAIN_FAIL))
4112 return sd_bus_error_set_const(ret_error, BUS_ERROR_NO_SUCH_PROCESS, "No matching processes to kill");
4113
4114 return ret;
4115}
4116
4117int unit_following_set(Unit *u, Set **s) {
4118 assert(u);
4119 assert(s);
4120
4121 if (UNIT_VTABLE(u)->following_set)
4122 return UNIT_VTABLE(u)->following_set(u, s);
4123
4124 *s = NULL;
4125 return 0;
4126}
4127
4128UnitFileState unit_get_unit_file_state(Unit *u) {
4129 int r;
4130
4131 assert(u);
4132
4133 if (u->unit_file_state < 0 && u->fragment_path) {
4134 r = unit_file_get_state(
4135 u->manager->runtime_scope,
4136 NULL,
4137 u->id,
4138 &u->unit_file_state);
4139 if (r < 0)
4140 u->unit_file_state = UNIT_FILE_BAD;
4141 }
4142
4143 return u->unit_file_state;
4144}
4145
4146PresetAction unit_get_unit_file_preset(Unit *u) {
4147 int r;
4148
4149 assert(u);
4150
4151 if (u->unit_file_preset < 0 && u->fragment_path) {
4152 _cleanup_free_ char *bn = NULL;
4153
4154 r = path_extract_filename(u->fragment_path, &bn);
4155 if (r < 0)
4156 return (u->unit_file_preset = r);
4157
4158 if (r == O_DIRECTORY)
4159 return (u->unit_file_preset = -EISDIR);
4160
4161 u->unit_file_preset = unit_file_query_preset(
4162 u->manager->runtime_scope,
4163 NULL,
4164 bn,
4165 NULL);
4166 }
4167
4168 return u->unit_file_preset;
4169}
4170
4171Unit* unit_ref_set(UnitRef *ref, Unit *source, Unit *target) {
4172 assert(ref);
4173 assert(source);
4174 assert(target);
4175
4176 if (ref->target)
4177 unit_ref_unset(ref);
4178
4179 ref->source = source;
4180 ref->target = target;
4181 LIST_PREPEND(refs_by_target, target->refs_by_target, ref);
4182 return target;
4183}
4184
4185void unit_ref_unset(UnitRef *ref) {
4186 assert(ref);
4187
4188 if (!ref->target)
4189 return;
4190
4191 /* We are about to drop a reference to the unit, make sure the garbage collection has a look at it as it might
4192 * be unreferenced now. */
4193 unit_add_to_gc_queue(ref->target);
4194
4195 LIST_REMOVE(refs_by_target, ref->target->refs_by_target, ref);
4196 ref->source = ref->target = NULL;
4197}
4198
4199static int user_from_unit_name(Unit *u, char **ret) {
4200
4201 static const uint8_t hash_key[] = {
4202 0x58, 0x1a, 0xaf, 0xe6, 0x28, 0x58, 0x4e, 0x96,
4203 0xb4, 0x4e, 0xf5, 0x3b, 0x8c, 0x92, 0x07, 0xec
4204 };
4205
4206 _cleanup_free_ char *n = NULL;
4207 int r;
4208
4209 r = unit_name_to_prefix(u->id, &n);
4210 if (r < 0)
4211 return r;
4212
4213 if (valid_user_group_name(n, 0)) {
4214 *ret = TAKE_PTR(n);
4215 return 0;
4216 }
4217
4218 /* If we can't use the unit name as a user name, then let's hash it and use that */
4219 if (asprintf(ret, "_du%016" PRIx64, siphash24(n, strlen(n), hash_key)) < 0)
4220 return -ENOMEM;
4221
4222 return 0;
4223}
4224
4225static int unit_verify_contexts(const Unit *u, const ExecContext *ec) {
4226 assert(u);
4227
4228 if (!ec)
4229 return 0;
4230
4231 if (MANAGER_IS_USER(u->manager) && ec->dynamic_user)
4232 return log_unit_error_errno(u, SYNTHETIC_ERRNO(ENOEXEC), "DynamicUser= enabled for user unit, which is not supported. Refusing.");
4233
4234 if (ec->dynamic_user && ec->working_directory_home)
4235 return log_unit_error_errno(u, SYNTHETIC_ERRNO(ENOEXEC), "WorkingDirectory=~ is not allowed under DynamicUser=yes. Refusing.");
4236
4237 return 0;
4238}
4239
4240int unit_patch_contexts(Unit *u) {
4241 CGroupContext *cc;
4242 ExecContext *ec;
4243 int r;
4244
4245 assert(u);
4246
4247 /* Patch in the manager defaults into the exec and cgroup
4248 * contexts, _after_ the rest of the settings have been
4249 * initialized */
4250
4251 ec = unit_get_exec_context(u);
4252 if (ec) {
4253 /* This only copies in the ones that need memory */
4254 for (unsigned i = 0; i < _RLIMIT_MAX; i++)
4255 if (u->manager->defaults.rlimit[i] && !ec->rlimit[i]) {
4256 ec->rlimit[i] = newdup(struct rlimit, u->manager->defaults.rlimit[i], 1);
4257 if (!ec->rlimit[i])
4258 return -ENOMEM;
4259 }
4260
4261 if (MANAGER_IS_USER(u->manager) && !ec->working_directory) {
4262 r = get_home_dir(&ec->working_directory);
4263 if (r < 0)
4264 return r;
4265
4266 if (!ec->working_directory_home)
4267 /* If home directory is implied by us, allow it to be missing. */
4268 ec->working_directory_missing_ok = true;
4269 }
4270
4271 if (ec->private_devices)
4272 ec->capability_bounding_set &= ~((UINT64_C(1) << CAP_MKNOD) | (UINT64_C(1) << CAP_SYS_RAWIO));
4273
4274 if (ec->protect_kernel_modules)
4275 ec->capability_bounding_set &= ~(UINT64_C(1) << CAP_SYS_MODULE);
4276
4277 if (ec->protect_kernel_logs)
4278 ec->capability_bounding_set &= ~(UINT64_C(1) << CAP_SYSLOG);
4279
4280 if (ec->protect_clock)
4281 ec->capability_bounding_set &= ~((UINT64_C(1) << CAP_SYS_TIME) | (UINT64_C(1) << CAP_WAKE_ALARM));
4282
4283 if (ec->dynamic_user) {
4284 if (!ec->user) {
4285 r = user_from_unit_name(u, &ec->user);
4286 if (r < 0)
4287 return r;
4288 }
4289
4290 if (!ec->group) {
4291 ec->group = strdup(ec->user);
4292 if (!ec->group)
4293 return -ENOMEM;
4294 }
4295
4296 /* If the dynamic user option is on, let's make sure that the unit can't leave its
4297 * UID/GID around in the file system or on IPC objects. Hence enforce a strict
4298 * sandbox. */
4299
4300 ec->private_tmp = true;
4301 ec->remove_ipc = true;
4302 ec->protect_system = PROTECT_SYSTEM_STRICT;
4303 if (ec->protect_home == PROTECT_HOME_NO)
4304 ec->protect_home = PROTECT_HOME_READ_ONLY;
4305
4306 /* Make sure this service can neither benefit from SUID/SGID binaries nor create
4307 * them. */
4308 ec->no_new_privileges = true;
4309 ec->restrict_suid_sgid = true;
4310 }
4311
4312 FOREACH_ARRAY(d, ec->directories, _EXEC_DIRECTORY_TYPE_MAX)
4313 exec_directory_sort(d);
4314 }
4315
4316 cc = unit_get_cgroup_context(u);
4317 if (cc && ec) {
4318
4319 if (ec->private_devices &&
4320 cc->device_policy == CGROUP_DEVICE_POLICY_AUTO)
4321 cc->device_policy = CGROUP_DEVICE_POLICY_CLOSED;
4322
4323 /* Only add these if needed, as they imply that everything else is blocked. */
4324 if (cc->device_policy != CGROUP_DEVICE_POLICY_AUTO || cc->device_allow) {
4325 if (ec->root_image || ec->mount_images) {
4326
4327 /* When RootImage= or MountImages= is specified, the following devices are touched. */
4328 FOREACH_STRING(p, "/dev/loop-control", "/dev/mapper/control") {
4329 r = cgroup_context_add_device_allow(cc, p, CGROUP_DEVICE_READ|CGROUP_DEVICE_WRITE);
4330 if (r < 0)
4331 return r;
4332 }
4333 FOREACH_STRING(p, "block-loop", "block-blkext", "block-device-mapper") {
4334 r = cgroup_context_add_device_allow(cc, p, CGROUP_DEVICE_READ|CGROUP_DEVICE_WRITE|CGROUP_DEVICE_MKNOD);
4335 if (r < 0)
4336 return r;
4337 }
4338
4339 /* Make sure "block-loop" can be resolved, i.e. make sure "loop" shows up in /proc/devices.
4340 * Same for mapper and verity. */
4341 FOREACH_STRING(p, "modprobe@loop.service", "modprobe@dm_mod.service", "modprobe@dm_verity.service") {
4342 r = unit_add_two_dependencies_by_name(u, UNIT_AFTER, UNIT_WANTS, p, true, UNIT_DEPENDENCY_FILE);
4343 if (r < 0)
4344 return r;
4345 }
4346 }
4347
4348 if (ec->protect_clock) {
4349 r = cgroup_context_add_device_allow(cc, "char-rtc", CGROUP_DEVICE_READ);
4350 if (r < 0)
4351 return r;
4352 }
4353
4354 /* If there are encrypted credentials we might need to access the TPM. */
4355 if (exec_context_has_encrypted_credentials(ec)) {
4356 r = cgroup_context_add_device_allow(cc, "char-tpm", CGROUP_DEVICE_READ|CGROUP_DEVICE_WRITE);
4357 if (r < 0)
4358 return r;
4359 }
4360 }
4361 }
4362
4363 return unit_verify_contexts(u, ec);
4364}
4365
4366ExecContext *unit_get_exec_context(const Unit *u) {
4367 size_t offset;
4368 assert(u);
4369
4370 if (u->type < 0)
4371 return NULL;
4372
4373 offset = UNIT_VTABLE(u)->exec_context_offset;
4374 if (offset <= 0)
4375 return NULL;
4376
4377 return (ExecContext*) ((uint8_t*) u + offset);
4378}
4379
4380KillContext *unit_get_kill_context(const Unit *u) {
4381 size_t offset;
4382 assert(u);
4383
4384 if (u->type < 0)
4385 return NULL;
4386
4387 offset = UNIT_VTABLE(u)->kill_context_offset;
4388 if (offset <= 0)
4389 return NULL;
4390
4391 return (KillContext*) ((uint8_t*) u + offset);
4392}
4393
4394CGroupContext *unit_get_cgroup_context(const Unit *u) {
4395 size_t offset;
4396
4397 if (u->type < 0)
4398 return NULL;
4399
4400 offset = UNIT_VTABLE(u)->cgroup_context_offset;
4401 if (offset <= 0)
4402 return NULL;
4403
4404 return (CGroupContext*) ((uint8_t*) u + offset);
4405}
4406
4407ExecRuntime *unit_get_exec_runtime(const Unit *u) {
4408 size_t offset;
4409
4410 if (u->type < 0)
4411 return NULL;
4412
4413 offset = UNIT_VTABLE(u)->exec_runtime_offset;
4414 if (offset <= 0)
4415 return NULL;
4416
4417 return *(ExecRuntime**) ((uint8_t*) u + offset);
4418}
4419
4420CGroupRuntime *unit_get_cgroup_runtime(const Unit *u) {
4421 size_t offset;
4422
4423 if (u->type < 0)
4424 return NULL;
4425
4426 offset = UNIT_VTABLE(u)->cgroup_runtime_offset;
4427 if (offset <= 0)
4428 return NULL;
4429
4430 return *(CGroupRuntime**) ((uint8_t*) u + offset);
4431}
4432
4433static const char* unit_drop_in_dir(Unit *u, UnitWriteFlags flags) {
4434 assert(u);
4435
4436 if (UNIT_WRITE_FLAGS_NOOP(flags))
4437 return NULL;
4438
4439 if (u->transient) /* Redirect drop-ins for transient units always into the transient directory. */
4440 return u->manager->lookup_paths.transient;
4441
4442 if (flags & UNIT_PERSISTENT)
4443 return u->manager->lookup_paths.persistent_control;
4444
4445 if (flags & UNIT_RUNTIME)
4446 return u->manager->lookup_paths.runtime_control;
4447
4448 return NULL;
4449}
4450
4451const char* unit_escape_setting(const char *s, UnitWriteFlags flags, char **buf) {
4452 assert(s);
4453 assert(popcount(flags & (UNIT_ESCAPE_EXEC_SYNTAX_ENV | UNIT_ESCAPE_EXEC_SYNTAX | UNIT_ESCAPE_C)) <= 1);
4454 assert(buf);
4455
4456 _cleanup_free_ char *t = NULL;
4457
4458 /* Returns a string with any escaping done. If no escaping was necessary, *buf is set to NULL, and
4459 * the input pointer is returned as-is. If an allocation was needed, the return buffer pointer is
4460 * written to *buf. This means the return value always contains a properly escaped version, but *buf
4461 * only contains a pointer if an allocation was made. Callers can use this to optimize memory
4462 * allocations. */
4463
4464 if (flags & UNIT_ESCAPE_SPECIFIERS) {
4465 t = specifier_escape(s);
4466 if (!t)
4467 return NULL;
4468
4469 s = t;
4470 }
4471
4472 /* We either do C-escaping or shell-escaping, to additionally escape characters that we parse for
4473 * ExecStart= and friends, i.e. '$' and quotes. */
4474
4475 if (flags & (UNIT_ESCAPE_EXEC_SYNTAX_ENV | UNIT_ESCAPE_EXEC_SYNTAX)) {
4476 char *t2;
4477
4478 if (flags & UNIT_ESCAPE_EXEC_SYNTAX_ENV) {
4479 t2 = strreplace(s, "$", "$$");
4480 if (!t2)
4481 return NULL;
4482 free_and_replace(t, t2);
4483 }
4484
4485 t2 = shell_escape(t ?: s, "\"");
4486 if (!t2)
4487 return NULL;
4488 free_and_replace(t, t2);
4489
4490 s = t;
4491
4492 } else if (flags & UNIT_ESCAPE_C) {
4493 char *t2;
4494
4495 t2 = cescape(s);
4496 if (!t2)
4497 return NULL;
4498 free_and_replace(t, t2);
4499
4500 s = t;
4501 }
4502
4503 *buf = TAKE_PTR(t);
4504 return s;
4505}
4506
4507char* unit_concat_strv(char **l, UnitWriteFlags flags) {
4508 _cleanup_free_ char *result = NULL;
4509 size_t n = 0;
4510
4511 /* Takes a list of strings, escapes them, and concatenates them. This may be used to format command
4512 * lines in a way suitable for ExecStart= stanzas. */
4513
4514 STRV_FOREACH(i, l) {
4515 _cleanup_free_ char *buf = NULL;
4516 const char *p;
4517 size_t a;
4518 char *q;
4519
4520 p = unit_escape_setting(*i, flags, &buf);
4521 if (!p)
4522 return NULL;
4523
4524 a = (n > 0) + 1 + strlen(p) + 1; /* separating space + " + entry + " */
4525 if (!GREEDY_REALLOC(result, n + a + 1))
4526 return NULL;
4527
4528 q = result + n;
4529 if (n > 0)
4530 *(q++) = ' ';
4531
4532 *(q++) = '"';
4533 q = stpcpy(q, p);
4534 *(q++) = '"';
4535
4536 n += a;
4537 }
4538
4539 if (!GREEDY_REALLOC(result, n + 1))
4540 return NULL;
4541
4542 result[n] = 0;
4543
4544 return TAKE_PTR(result);
4545}
4546
4547int unit_write_setting(Unit *u, UnitWriteFlags flags, const char *name, const char *data) {
4548 _cleanup_free_ char *p = NULL, *q = NULL, *escaped = NULL;
4549 const char *dir, *wrapped;
4550 int r;
4551
4552 assert(u);
4553 assert(name);
4554 assert(data);
4555
4556 if (UNIT_WRITE_FLAGS_NOOP(flags))
4557 return 0;
4558
4559 data = unit_escape_setting(data, flags, &escaped);
4560 if (!data)
4561 return -ENOMEM;
4562
4563 /* Prefix the section header. If we are writing this out as transient file, then let's suppress this if the
4564 * previous section header is the same */
4565
4566 if (flags & UNIT_PRIVATE) {
4567 if (!UNIT_VTABLE(u)->private_section)
4568 return -EINVAL;
4569
4570 if (!u->transient_file || u->last_section_private < 0)
4571 data = strjoina("[", UNIT_VTABLE(u)->private_section, "]\n", data);
4572 else if (u->last_section_private == 0)
4573 data = strjoina("\n[", UNIT_VTABLE(u)->private_section, "]\n", data);
4574 } else {
4575 if (!u->transient_file || u->last_section_private < 0)
4576 data = strjoina("[Unit]\n", data);
4577 else if (u->last_section_private > 0)
4578 data = strjoina("\n[Unit]\n", data);
4579 }
4580
4581 if (u->transient_file) {
4582 /* When this is a transient unit file in creation, then let's not create a new drop-in but instead
4583 * write to the transient unit file. */
4584 fputs(data, u->transient_file);
4585
4586 if (!endswith(data, "\n"))
4587 fputc('\n', u->transient_file);
4588
4589 /* Remember which section we wrote this entry to */
4590 u->last_section_private = !!(flags & UNIT_PRIVATE);
4591 return 0;
4592 }
4593
4594 dir = unit_drop_in_dir(u, flags);
4595 if (!dir)
4596 return -EINVAL;
4597
4598 wrapped = strjoina("# This is a drop-in unit file extension, created via \"systemctl set-property\"\n"
4599 "# or an equivalent operation. Do not edit.\n",
4600 data,
4601 "\n");
4602
4603 r = drop_in_file(dir, u->id, 50, name, &p, &q);
4604 if (r < 0)
4605 return r;
4606
4607 (void) mkdir_p_label(p, 0755);
4608
4609 /* Make sure the drop-in dir is registered in our path cache. This way we don't need to stupidly
4610 * recreate the cache after every drop-in we write. */
4611 if (u->manager->unit_path_cache) {
4612 r = set_put_strdup(&u->manager->unit_path_cache, p);
4613 if (r < 0)
4614 return r;
4615 }
4616
4617 r = write_string_file_atomic_label(q, wrapped);
4618 if (r < 0)
4619 return r;
4620
4621 r = strv_push(&u->dropin_paths, q);
4622 if (r < 0)
4623 return r;
4624 q = NULL;
4625
4626 strv_uniq(u->dropin_paths);
4627
4628 u->dropin_mtime = now(CLOCK_REALTIME);
4629
4630 return 0;
4631}
4632
4633int unit_write_settingf(Unit *u, UnitWriteFlags flags, const char *name, const char *format, ...) {
4634 _cleanup_free_ char *p = NULL;
4635 va_list ap;
4636 int r;
4637
4638 assert(u);
4639 assert(name);
4640 assert(format);
4641
4642 if (UNIT_WRITE_FLAGS_NOOP(flags))
4643 return 0;
4644
4645 va_start(ap, format);
4646 r = vasprintf(&p, format, ap);
4647 va_end(ap);
4648
4649 if (r < 0)
4650 return -ENOMEM;
4651
4652 return unit_write_setting(u, flags, name, p);
4653}
4654
4655int unit_make_transient(Unit *u) {
4656 _cleanup_free_ char *path = NULL;
4657 FILE *f;
4658
4659 assert(u);
4660
4661 if (!UNIT_VTABLE(u)->can_transient)
4662 return -EOPNOTSUPP;
4663
4664 (void) mkdir_p_label(u->manager->lookup_paths.transient, 0755);
4665
4666 path = path_join(u->manager->lookup_paths.transient, u->id);
4667 if (!path)
4668 return -ENOMEM;
4669
4670 /* Let's open the file we'll write the transient settings into. This file is kept open as long as we are
4671 * creating the transient, and is closed in unit_load(), as soon as we start loading the file. */
4672
4673 WITH_UMASK(0022) {
4674 f = fopen(path, "we");
4675 if (!f)
4676 return -errno;
4677 }
4678
4679 safe_fclose(u->transient_file);
4680 u->transient_file = f;
4681
4682 free_and_replace(u->fragment_path, path);
4683
4684 u->source_path = mfree(u->source_path);
4685 u->dropin_paths = strv_free(u->dropin_paths);
4686 u->fragment_mtime = u->source_mtime = u->dropin_mtime = 0;
4687
4688 u->load_state = UNIT_STUB;
4689 u->load_error = 0;
4690 u->transient = true;
4691
4692 unit_add_to_dbus_queue(u);
4693 unit_add_to_gc_queue(u);
4694
4695 fputs("# This is a transient unit file, created programmatically via the systemd API. Do not edit.\n",
4696 u->transient_file);
4697
4698 return 0;
4699}
4700
4701static int log_kill(const PidRef *pid, int sig, void *userdata) {
4702 _cleanup_free_ char *comm = NULL;
4703
4704 assert(pidref_is_set(pid));
4705
4706 (void) pidref_get_comm(pid, &comm);
4707
4708 /* Don't log about processes marked with brackets, under the assumption that these are temporary processes
4709 only, like for example systemd's own PAM stub process. */
4710 if (comm && comm[0] == '(')
4711 /* Although we didn't log anything, as this callback is used in unit_kill_context we must return 1
4712 * here to let the manager know that a process was killed. */
4713 return 1;
4714
4715 log_unit_notice(userdata,
4716 "Killing process " PID_FMT " (%s) with signal SIG%s.",
4717 pid->pid,
4718 strna(comm),
4719 signal_to_string(sig));
4720
4721 return 1;
4722}
4723
4724static int operation_to_signal(
4725 const KillContext *c,
4726 KillOperation k,
4727 bool *ret_noteworthy) {
4728
4729 assert(c);
4730
4731 switch (k) {
4732
4733 case KILL_TERMINATE:
4734 case KILL_TERMINATE_AND_LOG:
4735 *ret_noteworthy = false;
4736 return c->kill_signal;
4737
4738 case KILL_RESTART:
4739 *ret_noteworthy = false;
4740 return restart_kill_signal(c);
4741
4742 case KILL_KILL:
4743 *ret_noteworthy = true;
4744 return c->final_kill_signal;
4745
4746 case KILL_WATCHDOG:
4747 *ret_noteworthy = true;
4748 return c->watchdog_signal;
4749
4750 default:
4751 assert_not_reached();
4752 }
4753}
4754
4755static int unit_kill_context_one(
4756 Unit *u,
4757 const PidRef *pidref,
4758 const char *type,
4759 bool is_alien,
4760 int sig,
4761 bool send_sighup,
4762 cg_kill_log_func_t log_func) {
4763
4764 int r;
4765
4766 assert(u);
4767 assert(type);
4768
4769 /* This returns > 0 if it makes sense to wait for SIGCHLD for the process, == 0 if not. */
4770
4771 if (!pidref_is_set(pidref))
4772 return 0;
4773
4774 if (log_func)
4775 log_func(pidref, sig, u);
4776
4777 r = pidref_kill_and_sigcont(pidref, sig);
4778 if (r == -ESRCH)
4779 return !is_alien;
4780 if (r < 0) {
4781 _cleanup_free_ char *comm = NULL;
4782
4783 (void) pidref_get_comm(pidref, &comm);
4784 return log_unit_warning_errno(u, r, "Failed to kill %s process " PID_FMT " (%s), ignoring: %m", type, pidref->pid, strna(comm));
4785 }
4786
4787 if (send_sighup)
4788 (void) pidref_kill(pidref, SIGHUP);
4789
4790 return !is_alien;
4791}
4792
4793int unit_kill_context(Unit *u, KillOperation k) {
4794 bool wait_for_exit = false, send_sighup;
4795 cg_kill_log_func_t log_func = NULL;
4796 int sig, r;
4797
4798 assert(u);
4799
4800 /* Kill the processes belonging to this unit, in preparation for shutting the unit down. Returns > 0
4801 * if we killed something worth waiting for, 0 otherwise. Do not confuse with unit_kill_common()
4802 * which is used for user-requested killing of unit processes. */
4803
4804 KillContext *c = unit_get_kill_context(u);
4805 if (!c || c->kill_mode == KILL_NONE)
4806 return 0;
4807
4808 bool noteworthy;
4809 sig = operation_to_signal(c, k, &noteworthy);
4810 if (noteworthy)
4811 log_func = log_kill;
4812
4813 send_sighup =
4814 c->send_sighup &&
4815 IN_SET(k, KILL_TERMINATE, KILL_TERMINATE_AND_LOG) &&
4816 sig != SIGHUP;
4817
4818 bool is_alien;
4819 PidRef *main_pid = unit_main_pid_full(u, &is_alien);
4820 r = unit_kill_context_one(u, main_pid, "main", is_alien, sig, send_sighup, log_func);
4821 wait_for_exit = wait_for_exit || r > 0;
4822
4823 r = unit_kill_context_one(u, unit_control_pid(u), "control", /* is_alien = */ false, sig, send_sighup, log_func);
4824 wait_for_exit = wait_for_exit || r > 0;
4825
4826 CGroupRuntime *crt = unit_get_cgroup_runtime(u);
4827 if (crt && crt->cgroup_path &&
4828 (c->kill_mode == KILL_CONTROL_GROUP || (c->kill_mode == KILL_MIXED && k == KILL_KILL))) {
4829 _cleanup_set_free_ Set *pid_set = NULL;
4830
4831 /* Exclude the main/control pids from being killed via the cgroup */
4832 r = unit_pid_set(u, &pid_set);
4833 if (r < 0)
4834 return r;
4835
4836 r = cg_kill_recursive(
4837 crt->cgroup_path,
4838 sig,
4839 CGROUP_SIGCONT|CGROUP_IGNORE_SELF,
4840 pid_set,
4841 log_func, u);
4842 if (r < 0) {
4843 if (!IN_SET(r, -EAGAIN, -ESRCH, -ENOENT))
4844 log_unit_warning_errno(u, r, "Failed to kill control group %s, ignoring: %m", empty_to_root(crt->cgroup_path));
4845
4846 } else if (r > 0) {
4847
4848 /* FIXME: For now, on the legacy hierarchy, we will not wait for the cgroup members to die if
4849 * we are running in a container or if this is a delegation unit, simply because cgroup
4850 * notification is unreliable in these cases. It doesn't work at all in containers, and outside
4851 * of containers it can be confused easily by left-over directories in the cgroup — which
4852 * however should not exist in non-delegated units. On the unified hierarchy that's different,
4853 * there we get proper events. Hence rely on them. */
4854
4855 if (cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER) > 0 ||
4856 (detect_container() == 0 && !unit_cgroup_delegate(u)))
4857 wait_for_exit = true;
4858
4859 if (send_sighup) {
4860 r = unit_pid_set(u, &pid_set);
4861 if (r < 0)
4862 return r;
4863
4864 (void) cg_kill_recursive(
4865 crt->cgroup_path,
4866 SIGHUP,
4867 CGROUP_IGNORE_SELF,
4868 pid_set,
4869 /* kill_log= */ NULL,
4870 /* userdata= */ NULL);
4871 }
4872 }
4873 }
4874
4875 return wait_for_exit;
4876}
4877
4878int unit_add_mounts_for(Unit *u, const char *path, UnitDependencyMask mask, UnitMountDependencyType type) {
4879 Hashmap **unit_map, **manager_map;
4880 int r;
4881
4882 assert(u);
4883 assert(path);
4884 assert(type >= 0 && type < _UNIT_MOUNT_DEPENDENCY_TYPE_MAX);
4885
4886 unit_map = &u->mounts_for[type];
4887 manager_map = &u->manager->units_needing_mounts_for[type];
4888
4889 /* Registers a unit for requiring a certain path and all its prefixes. We keep a hashtable of these
4890 * paths in the unit (from the path to the UnitDependencyInfo structure indicating how to the
4891 * dependency came to be). However, we build a prefix table for all possible prefixes so that new
4892 * appearing mount units can easily determine which units to make themselves a dependency of. */
4893
4894 if (!path_is_absolute(path))
4895 return -EINVAL;
4896
4897 if (hashmap_contains(*unit_map, path)) /* Exit quickly if the path is already covered. */
4898 return 0;
4899
4900 /* Use the canonical form of the path as the stored key. We call path_is_normalized()
4901 * only after simplification, since path_is_normalized() rejects paths with '.'.
4902 * path_is_normalized() also verifies that the path fits in PATH_MAX. */
4903 _cleanup_free_ char *p = NULL;
4904 r = path_simplify_alloc(path, &p);
4905 if (r < 0)
4906 return r;
4907 path = p;
4908
4909 if (!path_is_normalized(path))
4910 return -EPERM;
4911
4912 UnitDependencyInfo di = {
4913 .origin_mask = mask
4914 };
4915
4916 r = hashmap_ensure_put(unit_map, &path_hash_ops, p, di.data);
4917 if (r < 0)
4918 return r;
4919 assert(r > 0);
4920 TAKE_PTR(p); /* path remains a valid pointer to the string stored in the hashmap */
4921
4922 char prefix[strlen(path) + 1];
4923 PATH_FOREACH_PREFIX_MORE(prefix, path) {
4924 Set *x;
4925
4926 x = hashmap_get(*manager_map, prefix);
4927 if (!x) {
4928 _cleanup_free_ char *q = NULL;
4929
4930 r = hashmap_ensure_allocated(manager_map, &path_hash_ops);
4931 if (r < 0)
4932 return r;
4933
4934 q = strdup(prefix);
4935 if (!q)
4936 return -ENOMEM;
4937
4938 x = set_new(NULL);
4939 if (!x)
4940 return -ENOMEM;
4941
4942 r = hashmap_put(*manager_map, q, x);
4943 if (r < 0) {
4944 set_free(x);
4945 return r;
4946 }
4947 q = NULL;
4948 }
4949
4950 r = set_put(x, u);
4951 if (r < 0)
4952 return r;
4953 }
4954
4955 return 0;
4956}
4957
4958int unit_setup_exec_runtime(Unit *u) {
4959 _cleanup_(exec_shared_runtime_unrefp) ExecSharedRuntime *esr = NULL;
4960 _cleanup_(dynamic_creds_unrefp) DynamicCreds *dcreds = NULL;
4961 _cleanup_set_free_ Set *units = NULL;
4962 ExecRuntime **rt;
4963 ExecContext *ec;
4964 size_t offset;
4965 Unit *other;
4966 int r;
4967
4968 offset = UNIT_VTABLE(u)->exec_runtime_offset;
4969 assert(offset > 0);
4970
4971 /* Check if there already is an ExecRuntime for this unit? */
4972 rt = (ExecRuntime**) ((uint8_t*) u + offset);
4973 if (*rt)
4974 return 0;
4975
4976 ec = ASSERT_PTR(unit_get_exec_context(u));
4977
4978 r = unit_get_transitive_dependency_set(u, UNIT_ATOM_JOINS_NAMESPACE_OF, &units);
4979 if (r < 0)
4980 return r;
4981
4982 /* Try to get it from somebody else */
4983 SET_FOREACH(other, units) {
4984 r = exec_shared_runtime_acquire(u->manager, NULL, other->id, false, &esr);
4985 if (r < 0)
4986 return r;
4987 if (r > 0)
4988 break;
4989 }
4990
4991 if (!esr) {
4992 r = exec_shared_runtime_acquire(u->manager, ec, u->id, true, &esr);
4993 if (r < 0)
4994 return r;
4995 }
4996
4997 if (ec->dynamic_user) {
4998 r = dynamic_creds_make(u->manager, ec->user, ec->group, &dcreds);
4999 if (r < 0)
5000 return r;
5001 }
5002
5003 r = exec_runtime_make(u, ec, esr, dcreds, rt);
5004 if (r < 0)
5005 return r;
5006
5007 TAKE_PTR(esr);
5008 TAKE_PTR(dcreds);
5009
5010 return r;
5011}
5012
5013CGroupRuntime *unit_setup_cgroup_runtime(Unit *u) {
5014 size_t offset;
5015
5016 assert(u);
5017
5018 offset = UNIT_VTABLE(u)->cgroup_runtime_offset;
5019 assert(offset > 0);
5020
5021 CGroupRuntime **rt = (CGroupRuntime**) ((uint8_t*) u + offset);
5022 if (*rt)
5023 return *rt;
5024
5025 return (*rt = cgroup_runtime_new());
5026}
5027
5028bool unit_type_supported(UnitType t) {
5029 static int8_t cache[_UNIT_TYPE_MAX] = {}; /* -1: disabled, 1: enabled: 0: don't know */
5030 int r;
5031
5032 assert(t >= 0 && t < _UNIT_TYPE_MAX);
5033
5034 if (cache[t] == 0) {
5035 char *e;
5036
5037 e = strjoina("SYSTEMD_SUPPORT_", unit_type_to_string(t));
5038
5039 r = getenv_bool(ascii_strupper(e));
5040 if (r < 0 && r != -ENXIO)
5041 log_debug_errno(r, "Failed to parse $%s, ignoring: %m", e);
5042
5043 cache[t] = r == 0 ? -1 : 1;
5044 }
5045 if (cache[t] < 0)
5046 return false;
5047
5048 if (!unit_vtable[t]->supported)
5049 return true;
5050
5051 return unit_vtable[t]->supported();
5052}
5053
5054void unit_warn_if_dir_nonempty(Unit *u, const char* where) {
5055 int r;
5056
5057 assert(u);
5058 assert(where);
5059
5060 if (!unit_log_level_test(u, LOG_NOTICE))
5061 return;
5062
5063 r = dir_is_empty(where, /* ignore_hidden_or_backup= */ false);
5064 if (r > 0 || r == -ENOTDIR)
5065 return;
5066 if (r < 0) {
5067 log_unit_warning_errno(u, r, "Failed to check directory %s: %m", where);
5068 return;
5069 }
5070
5071 log_unit_struct(u, LOG_NOTICE,
5072 "MESSAGE_ID=" SD_MESSAGE_OVERMOUNTING_STR,
5073 LOG_UNIT_INVOCATION_ID(u),
5074 LOG_UNIT_MESSAGE(u, "Directory %s to mount over is not empty, mounting anyway.", where),
5075 "WHERE=%s", where);
5076}
5077
5078int unit_fail_if_noncanonical(Unit *u, const char* where) {
5079 _cleanup_free_ char *canonical_where = NULL;
5080 int r;
5081
5082 assert(u);
5083 assert(where);
5084
5085 r = chase(where, NULL, CHASE_NONEXISTENT, &canonical_where, NULL);
5086 if (r < 0) {
5087 log_unit_debug_errno(u, r, "Failed to check %s for symlinks, ignoring: %m", where);
5088 return 0;
5089 }
5090
5091 /* We will happily ignore a trailing slash (or any redundant slashes) */
5092 if (path_equal(where, canonical_where))
5093 return 0;
5094
5095 /* No need to mention "." or "..", they would already have been rejected by unit_name_from_path() */
5096 log_unit_struct(u, LOG_ERR,
5097 "MESSAGE_ID=" SD_MESSAGE_OVERMOUNTING_STR,
5098 LOG_UNIT_INVOCATION_ID(u),
5099 LOG_UNIT_MESSAGE(u, "Mount path %s is not canonical (contains a symlink).", where),
5100 "WHERE=%s", where);
5101
5102 return -ELOOP;
5103}
5104
5105bool unit_is_pristine(Unit *u) {
5106 assert(u);
5107
5108 /* Check if the unit already exists or is already around, in a number of different ways. Note that to
5109 * cater for unit types such as slice, we are generally fine with units that are marked UNIT_LOADED
5110 * even though nothing was actually loaded, as those unit types don't require a file on disk.
5111 *
5112 * Note that we don't check for drop-ins here, because we allow drop-ins for transient units
5113 * identically to non-transient units, both unit-specific and hierarchical. E.g. for a-b-c.service:
5114 * service.d/….conf, a-.service.d/….conf, a-b-.service.d/….conf, a-b-c.service.d/….conf.
5115 */
5116
5117 return IN_SET(u->load_state, UNIT_NOT_FOUND, UNIT_LOADED) &&
5118 !u->fragment_path &&
5119 !u->source_path &&
5120 !u->job &&
5121 !u->merged_into;
5122}
5123
5124PidRef* unit_control_pid(Unit *u) {
5125 assert(u);
5126
5127 if (UNIT_VTABLE(u)->control_pid)
5128 return UNIT_VTABLE(u)->control_pid(u);
5129
5130 return NULL;
5131}
5132
5133PidRef* unit_main_pid_full(Unit *u, bool *ret_is_alien) {
5134 assert(u);
5135
5136 if (UNIT_VTABLE(u)->main_pid)
5137 return UNIT_VTABLE(u)->main_pid(u, ret_is_alien);
5138
5139 if (ret_is_alien)
5140 *ret_is_alien = false;
5141 return NULL;
5142}
5143
5144static void unit_modify_user_nft_set(Unit *u, bool add, NFTSetSource source, uint32_t element) {
5145 int r;
5146
5147 assert(u);
5148
5149 if (!MANAGER_IS_SYSTEM(u->manager))
5150 return;
5151
5152 CGroupContext *c;
5153 c = unit_get_cgroup_context(u);
5154 if (!c)
5155 return;
5156
5157 if (!u->manager->fw_ctx) {
5158 r = fw_ctx_new_full(&u->manager->fw_ctx, /* init_tables= */ false);
5159 if (r < 0)
5160 return;
5161
5162 assert(u->manager->fw_ctx);
5163 }
5164
5165 FOREACH_ARRAY(nft_set, c->nft_set_context.sets, c->nft_set_context.n_sets) {
5166 if (nft_set->source != source)
5167 continue;
5168
5169 r = nft_set_element_modify_any(u->manager->fw_ctx, add, nft_set->nfproto, nft_set->table, nft_set->set, &element, sizeof(element));
5170 if (r < 0)
5171 log_warning_errno(r, "Failed to %s NFT set: family %s, table %s, set %s, ID %u, ignoring: %m",
5172 add? "add" : "delete", nfproto_to_string(nft_set->nfproto), nft_set->table, nft_set->set, element);
5173 else
5174 log_debug("%s NFT set: family %s, table %s, set %s, ID %u",
5175 add? "Added" : "Deleted", nfproto_to_string(nft_set->nfproto), nft_set->table, nft_set->set, element);
5176 }
5177}
5178
5179static void unit_unref_uid_internal(
5180 Unit *u,
5181 uid_t *ref_uid,
5182 bool destroy_now,
5183 void (*_manager_unref_uid)(Manager *m, uid_t uid, bool destroy_now)) {
5184
5185 assert(u);
5186 assert(ref_uid);
5187 assert(_manager_unref_uid);
5188
5189 /* Generic implementation of both unit_unref_uid() and unit_unref_gid(), under the assumption that uid_t and
5190 * gid_t are actually the same time, with the same validity rules.
5191 *
5192 * Drops a reference to UID/GID from a unit. */
5193
5194 assert_cc(sizeof(uid_t) == sizeof(gid_t));
5195 assert_cc(UID_INVALID == (uid_t) GID_INVALID);
5196
5197 if (!uid_is_valid(*ref_uid))
5198 return;
5199
5200 _manager_unref_uid(u->manager, *ref_uid, destroy_now);
5201 *ref_uid = UID_INVALID;
5202}
5203
5204static void unit_unref_uid(Unit *u, bool destroy_now) {
5205 assert(u);
5206
5207 unit_modify_user_nft_set(u, /* add = */ false, NFT_SET_SOURCE_USER, u->ref_uid);
5208
5209 unit_unref_uid_internal(u, &u->ref_uid, destroy_now, manager_unref_uid);
5210}
5211
5212static void unit_unref_gid(Unit *u, bool destroy_now) {
5213 assert(u);
5214
5215 unit_modify_user_nft_set(u, /* add = */ false, NFT_SET_SOURCE_GROUP, u->ref_gid);
5216
5217 unit_unref_uid_internal(u, (uid_t*) &u->ref_gid, destroy_now, manager_unref_gid);
5218}
5219
5220void unit_unref_uid_gid(Unit *u, bool destroy_now) {
5221 assert(u);
5222
5223 unit_unref_uid(u, destroy_now);
5224 unit_unref_gid(u, destroy_now);
5225}
5226
5227static int unit_ref_uid_internal(
5228 Unit *u,
5229 uid_t *ref_uid,
5230 uid_t uid,
5231 bool clean_ipc,
5232 int (*_manager_ref_uid)(Manager *m, uid_t uid, bool clean_ipc)) {
5233
5234 int r;
5235
5236 assert(u);
5237 assert(ref_uid);
5238 assert(uid_is_valid(uid));
5239 assert(_manager_ref_uid);
5240
5241 /* Generic implementation of both unit_ref_uid() and unit_ref_guid(), under the assumption that uid_t and gid_t
5242 * are actually the same type, and have the same validity rules.
5243 *
5244 * Adds a reference on a specific UID/GID to this unit. Each unit referencing the same UID/GID maintains a
5245 * reference so that we can destroy the UID/GID's IPC resources as soon as this is requested and the counter
5246 * drops to zero. */
5247
5248 assert_cc(sizeof(uid_t) == sizeof(gid_t));
5249 assert_cc(UID_INVALID == (uid_t) GID_INVALID);
5250
5251 if (*ref_uid == uid)
5252 return 0;
5253
5254 if (uid_is_valid(*ref_uid)) /* Already set? */
5255 return -EBUSY;
5256
5257 r = _manager_ref_uid(u->manager, uid, clean_ipc);
5258 if (r < 0)
5259 return r;
5260
5261 *ref_uid = uid;
5262 return 1;
5263}
5264
5265static int unit_ref_uid(Unit *u, uid_t uid, bool clean_ipc) {
5266 return unit_ref_uid_internal(u, &u->ref_uid, uid, clean_ipc, manager_ref_uid);
5267}
5268
5269static int unit_ref_gid(Unit *u, gid_t gid, bool clean_ipc) {
5270 return unit_ref_uid_internal(u, (uid_t*) &u->ref_gid, (uid_t) gid, clean_ipc, manager_ref_gid);
5271}
5272
5273static int unit_ref_uid_gid_internal(Unit *u, uid_t uid, gid_t gid, bool clean_ipc) {
5274 int r = 0, q = 0;
5275
5276 assert(u);
5277
5278 /* Reference both a UID and a GID in one go. Either references both, or neither. */
5279
5280 if (uid_is_valid(uid)) {
5281 r = unit_ref_uid(u, uid, clean_ipc);
5282 if (r < 0)
5283 return r;
5284 }
5285
5286 if (gid_is_valid(gid)) {
5287 q = unit_ref_gid(u, gid, clean_ipc);
5288 if (q < 0) {
5289 if (r > 0)
5290 unit_unref_uid(u, false);
5291
5292 return q;
5293 }
5294 }
5295
5296 return r > 0 || q > 0;
5297}
5298
5299int unit_ref_uid_gid(Unit *u, uid_t uid, gid_t gid) {
5300 ExecContext *c;
5301 int r;
5302
5303 assert(u);
5304
5305 c = unit_get_exec_context(u);
5306
5307 r = unit_ref_uid_gid_internal(u, uid, gid, c ? c->remove_ipc : false);
5308 if (r < 0)
5309 return log_unit_warning_errno(u, r, "Couldn't add UID/GID reference to unit, proceeding without: %m");
5310
5311 unit_modify_user_nft_set(u, /* add = */ true, NFT_SET_SOURCE_USER, uid);
5312 unit_modify_user_nft_set(u, /* add = */ true, NFT_SET_SOURCE_GROUP, gid);
5313
5314 return r;
5315}
5316
5317void unit_notify_user_lookup(Unit *u, uid_t uid, gid_t gid) {
5318 int r;
5319
5320 assert(u);
5321
5322 /* This is invoked whenever one of the forked off processes let's us know the UID/GID its user name/group names
5323 * resolved to. We keep track of which UID/GID is currently assigned in order to be able to destroy its IPC
5324 * objects when no service references the UID/GID anymore. */
5325
5326 r = unit_ref_uid_gid(u, uid, gid);
5327 if (r > 0)
5328 unit_add_to_dbus_queue(u);
5329}
5330
5331int unit_acquire_invocation_id(Unit *u) {
5332 sd_id128_t id;
5333 int r;
5334
5335 assert(u);
5336
5337 r = sd_id128_randomize(&id);
5338 if (r < 0)
5339 return log_unit_error_errno(u, r, "Failed to generate invocation ID for unit: %m");
5340
5341 r = unit_set_invocation_id(u, id);
5342 if (r < 0)
5343 return log_unit_error_errno(u, r, "Failed to set invocation ID for unit: %m");
5344
5345 unit_add_to_dbus_queue(u);
5346 return 0;
5347}
5348
5349int unit_set_exec_params(Unit *u, ExecParameters *p) {
5350 int r;
5351
5352 assert(u);
5353 assert(p);
5354
5355 /* Copy parameters from manager */
5356 r = manager_get_effective_environment(u->manager, &p->environment);
5357 if (r < 0)
5358 return r;
5359
5360 p->runtime_scope = u->manager->runtime_scope;
5361
5362 r = strdup_to(&p->confirm_spawn, manager_get_confirm_spawn(u->manager));
5363 if (r < 0)
5364 return r;
5365
5366 p->cgroup_supported = u->manager->cgroup_supported;
5367 p->prefix = u->manager->prefix;
5368 SET_FLAG(p->flags, EXEC_PASS_LOG_UNIT|EXEC_CHOWN_DIRECTORIES, MANAGER_IS_SYSTEM(u->manager));
5369
5370 /* Copy parameters from unit */
5371 CGroupRuntime *crt = unit_get_cgroup_runtime(u);
5372 p->cgroup_path = crt ? crt->cgroup_path : NULL;
5373 SET_FLAG(p->flags, EXEC_CGROUP_DELEGATE, unit_cgroup_delegate(u));
5374
5375 p->received_credentials_directory = u->manager->received_credentials_directory;
5376 p->received_encrypted_credentials_directory = u->manager->received_encrypted_credentials_directory;
5377
5378 p->shall_confirm_spawn = u->manager->confirm_spawn;
5379
5380 p->fallback_smack_process_label = u->manager->defaults.smack_process_label;
5381
5382 if (u->manager->restrict_fs && p->bpf_restrict_fs_map_fd < 0) {
5383 int fd = bpf_restrict_fs_map_fd(u);
5384 if (fd < 0)
5385 return fd;
5386
5387 p->bpf_restrict_fs_map_fd = fd;
5388 }
5389
5390 p->user_lookup_fd = u->manager->user_lookup_fds[1];
5391 p->handoff_timestamp_fd = u->manager->handoff_timestamp_fds[1];
5392
5393 p->cgroup_id = crt ? crt->cgroup_id : 0;
5394 p->invocation_id = u->invocation_id;
5395 sd_id128_to_string(p->invocation_id, p->invocation_id_string);
5396 p->unit_id = strdup(u->id);
5397 if (!p->unit_id)
5398 return -ENOMEM;
5399
5400 return 0;
5401}
5402
5403int unit_fork_helper_process(Unit *u, const char *name, PidRef *ret) {
5404 pid_t pid;
5405 int r;
5406
5407 assert(u);
5408 assert(ret);
5409
5410 /* Forks off a helper process and makes sure it is a member of the unit's cgroup. Returns == 0 in the child,
5411 * and > 0 in the parent. The pid parameter is always filled in with the child's PID. */
5412
5413 (void) unit_realize_cgroup(u);
5414
5415 CGroupRuntime *crt = unit_setup_cgroup_runtime(u);
5416 if (!crt)
5417 return -ENOMEM;
5418
5419 r = safe_fork(name, FORK_REOPEN_LOG|FORK_DEATHSIG_SIGTERM, &pid);
5420 if (r < 0)
5421 return r;
5422 if (r > 0) {
5423 _cleanup_(pidref_done) PidRef pidref = PIDREF_NULL;
5424 int q;
5425
5426 /* Parent */
5427
5428 q = pidref_set_pid(&pidref, pid);
5429 if (q < 0)
5430 return q;
5431
5432 *ret = TAKE_PIDREF(pidref);
5433 return r;
5434 }
5435
5436 /* Child */
5437
5438 (void) default_signals(SIGNALS_CRASH_HANDLER, SIGNALS_IGNORE);
5439 (void) ignore_signals(SIGPIPE);
5440
5441 if (crt->cgroup_path) {
5442 r = cg_attach_everywhere(u->manager->cgroup_supported, crt->cgroup_path, 0, NULL, NULL);
5443 if (r < 0) {
5444 log_unit_error_errno(u, r, "Failed to join unit cgroup %s: %m", empty_to_root(crt->cgroup_path));
5445 _exit(EXIT_CGROUP);
5446 }
5447 }
5448
5449 return 0;
5450}
5451
5452int unit_fork_and_watch_rm_rf(Unit *u, char **paths, PidRef *ret_pid) {
5453 _cleanup_(pidref_done) PidRef pid = PIDREF_NULL;
5454 int r;
5455
5456 assert(u);
5457 assert(ret_pid);
5458
5459 r = unit_fork_helper_process(u, "(sd-rmrf)", &pid);
5460 if (r < 0)
5461 return r;
5462 if (r == 0) {
5463 int ret = EXIT_SUCCESS;
5464
5465 STRV_FOREACH(i, paths) {
5466 r = rm_rf(*i, REMOVE_ROOT|REMOVE_PHYSICAL|REMOVE_MISSING_OK);
5467 if (r < 0) {
5468 log_error_errno(r, "Failed to remove '%s': %m", *i);
5469 ret = EXIT_FAILURE;
5470 }
5471 }
5472
5473 _exit(ret);
5474 }
5475
5476 r = unit_watch_pidref(u, &pid, /* exclusive= */ true);
5477 if (r < 0)
5478 return r;
5479
5480 *ret_pid = TAKE_PIDREF(pid);
5481 return 0;
5482}
5483
5484static void unit_update_dependency_mask(Hashmap *deps, Unit *other, UnitDependencyInfo di) {
5485 assert(deps);
5486 assert(other);
5487
5488 if (di.origin_mask == 0 && di.destination_mask == 0)
5489 /* No bit set anymore, let's drop the whole entry */
5490 assert_se(hashmap_remove(deps, other));
5491 else
5492 /* Mask was reduced, let's update the entry */
5493 assert_se(hashmap_update(deps, other, di.data) == 0);
5494}
5495
5496void unit_remove_dependencies(Unit *u, UnitDependencyMask mask) {
5497 Hashmap *deps;
5498 assert(u);
5499
5500 /* Removes all dependencies u has on other units marked for ownership by 'mask'. */
5501
5502 if (mask == 0)
5503 return;
5504
5505 HASHMAP_FOREACH(deps, u->dependencies) {
5506 bool done;
5507
5508 do {
5509 UnitDependencyInfo di;
5510 Unit *other;
5511
5512 done = true;
5513
5514 HASHMAP_FOREACH_KEY(di.data, other, deps) {
5515 Hashmap *other_deps;
5516
5517 if (FLAGS_SET(~mask, di.origin_mask))
5518 continue;
5519
5520 di.origin_mask &= ~mask;
5521 unit_update_dependency_mask(deps, other, di);
5522
5523 /* We updated the dependency from our unit to the other unit now. But most
5524 * dependencies imply a reverse dependency. Hence, let's delete that one
5525 * too. For that we go through all dependency types on the other unit and
5526 * delete all those which point to us and have the right mask set. */
5527
5528 HASHMAP_FOREACH(other_deps, other->dependencies) {
5529 UnitDependencyInfo dj;
5530
5531 dj.data = hashmap_get(other_deps, u);
5532 if (FLAGS_SET(~mask, dj.destination_mask))
5533 continue;
5534
5535 dj.destination_mask &= ~mask;
5536 unit_update_dependency_mask(other_deps, u, dj);
5537 }
5538
5539 unit_add_to_gc_queue(other);
5540
5541 /* The unit 'other' may not be wanted by the unit 'u'. */
5542 unit_submit_to_stop_when_unneeded_queue(other);
5543
5544 done = false;
5545 break;
5546 }
5547
5548 } while (!done);
5549 }
5550}
5551
5552static int unit_get_invocation_path(Unit *u, char **ret) {
5553 char *p;
5554 int r;
5555
5556 assert(u);
5557 assert(ret);
5558
5559 if (MANAGER_IS_SYSTEM(u->manager))
5560 p = strjoin("/run/systemd/units/invocation:", u->id);
5561 else {
5562 _cleanup_free_ char *user_path = NULL;
5563 r = xdg_user_runtime_dir(&user_path, "/systemd/units/invocation:");
5564 if (r < 0)
5565 return r;
5566 p = strjoin(user_path, u->id);
5567 }
5568
5569 if (!p)
5570 return -ENOMEM;
5571
5572 *ret = p;
5573 return 0;
5574}
5575
5576static int unit_export_invocation_id(Unit *u) {
5577 _cleanup_free_ char *p = NULL;
5578 int r;
5579
5580 assert(u);
5581
5582 if (u->exported_invocation_id)
5583 return 0;
5584
5585 if (sd_id128_is_null(u->invocation_id))
5586 return 0;
5587
5588 r = unit_get_invocation_path(u, &p);
5589 if (r < 0)
5590 return log_unit_debug_errno(u, r, "Failed to get invocation path: %m");
5591
5592 r = symlink_atomic_label(u->invocation_id_string, p);
5593 if (r < 0)
5594 return log_unit_debug_errno(u, r, "Failed to create invocation ID symlink %s: %m", p);
5595
5596 u->exported_invocation_id = true;
5597 return 0;
5598}
5599
5600static int unit_export_log_level_max(Unit *u, const ExecContext *c) {
5601 const char *p;
5602 char buf[2];
5603 int r;
5604
5605 assert(u);
5606 assert(c);
5607
5608 if (u->exported_log_level_max)
5609 return 0;
5610
5611 if (c->log_level_max < 0)
5612 return 0;
5613
5614 assert(c->log_level_max <= 7);
5615
5616 buf[0] = '0' + c->log_level_max;
5617 buf[1] = 0;
5618
5619 p = strjoina("/run/systemd/units/log-level-max:", u->id);
5620 r = symlink_atomic(buf, p);
5621 if (r < 0)
5622 return log_unit_debug_errno(u, r, "Failed to create maximum log level symlink %s: %m", p);
5623
5624 u->exported_log_level_max = true;
5625 return 0;
5626}
5627
5628static int unit_export_log_extra_fields(Unit *u, const ExecContext *c) {
5629 _cleanup_close_ int fd = -EBADF;
5630 struct iovec *iovec;
5631 const char *p;
5632 char *pattern;
5633 le64_t *sizes;
5634 ssize_t n;
5635 int r;
5636
5637 if (u->exported_log_extra_fields)
5638 return 0;
5639
5640 if (c->n_log_extra_fields <= 0)
5641 return 0;
5642
5643 sizes = newa(le64_t, c->n_log_extra_fields);
5644 iovec = newa(struct iovec, c->n_log_extra_fields * 2);
5645
5646 for (size_t i = 0; i < c->n_log_extra_fields; i++) {
5647 sizes[i] = htole64(c->log_extra_fields[i].iov_len);
5648
5649 iovec[i*2] = IOVEC_MAKE(sizes + i, sizeof(le64_t));
5650 iovec[i*2+1] = c->log_extra_fields[i];
5651 }
5652
5653 p = strjoina("/run/systemd/units/log-extra-fields:", u->id);
5654 pattern = strjoina(p, ".XXXXXX");
5655
5656 fd = mkostemp_safe(pattern);
5657 if (fd < 0)
5658 return log_unit_debug_errno(u, fd, "Failed to create extra fields file %s: %m", p);
5659
5660 n = writev(fd, iovec, c->n_log_extra_fields*2);
5661 if (n < 0) {
5662 r = log_unit_debug_errno(u, errno, "Failed to write extra fields: %m");
5663 goto fail;
5664 }
5665
5666 (void) fchmod(fd, 0644);
5667
5668 if (rename(pattern, p) < 0) {
5669 r = log_unit_debug_errno(u, errno, "Failed to rename extra fields file: %m");
5670 goto fail;
5671 }
5672
5673 u->exported_log_extra_fields = true;
5674 return 0;
5675
5676fail:
5677 (void) unlink(pattern);
5678 return r;
5679}
5680
5681static int unit_export_log_ratelimit_interval(Unit *u, const ExecContext *c) {
5682 _cleanup_free_ char *buf = NULL;
5683 const char *p;
5684 int r;
5685
5686 assert(u);
5687 assert(c);
5688
5689 if (u->exported_log_ratelimit_interval)
5690 return 0;
5691
5692 if (c->log_ratelimit_interval_usec == 0)
5693 return 0;
5694
5695 p = strjoina("/run/systemd/units/log-rate-limit-interval:", u->id);
5696
5697 if (asprintf(&buf, "%" PRIu64, c->log_ratelimit_interval_usec) < 0)
5698 return log_oom();
5699
5700 r = symlink_atomic(buf, p);
5701 if (r < 0)
5702 return log_unit_debug_errno(u, r, "Failed to create log rate limit interval symlink %s: %m", p);
5703
5704 u->exported_log_ratelimit_interval = true;
5705 return 0;
5706}
5707
5708static int unit_export_log_ratelimit_burst(Unit *u, const ExecContext *c) {
5709 _cleanup_free_ char *buf = NULL;
5710 const char *p;
5711 int r;
5712
5713 assert(u);
5714 assert(c);
5715
5716 if (u->exported_log_ratelimit_burst)
5717 return 0;
5718
5719 if (c->log_ratelimit_burst == 0)
5720 return 0;
5721
5722 p = strjoina("/run/systemd/units/log-rate-limit-burst:", u->id);
5723
5724 if (asprintf(&buf, "%u", c->log_ratelimit_burst) < 0)
5725 return log_oom();
5726
5727 r = symlink_atomic(buf, p);
5728 if (r < 0)
5729 return log_unit_debug_errno(u, r, "Failed to create log rate limit burst symlink %s: %m", p);
5730
5731 u->exported_log_ratelimit_burst = true;
5732 return 0;
5733}
5734
5735void unit_export_state_files(Unit *u) {
5736 const ExecContext *c;
5737
5738 assert(u);
5739
5740 if (!u->id)
5741 return;
5742
5743 if (MANAGER_IS_TEST_RUN(u->manager))
5744 return;
5745
5746 /* Exports a couple of unit properties to /run/systemd/units/, so that journald can quickly query this data
5747 * from there. Ideally, journald would use IPC to query this, like everybody else, but that's hard, as long as
5748 * the IPC system itself and PID 1 also log to the journal.
5749 *
5750 * Note that these files really shouldn't be considered API for anyone else, as use a runtime file system as
5751 * IPC replacement is not compatible with today's world of file system namespaces. However, this doesn't really
5752 * apply to communication between the journal and systemd, as we assume that these two daemons live in the same
5753 * namespace at least.
5754 *
5755 * Note that some of the "files" exported here are actually symlinks and not regular files. Symlinks work
5756 * better for storing small bits of data, in particular as we can write them with two system calls, and read
5757 * them with one. */
5758
5759 (void) unit_export_invocation_id(u);
5760
5761 if (!MANAGER_IS_SYSTEM(u->manager))
5762 return;
5763
5764 c = unit_get_exec_context(u);
5765 if (c) {
5766 (void) unit_export_log_level_max(u, c);
5767 (void) unit_export_log_extra_fields(u, c);
5768 (void) unit_export_log_ratelimit_interval(u, c);
5769 (void) unit_export_log_ratelimit_burst(u, c);
5770 }
5771}
5772
5773void unit_unlink_state_files(Unit *u) {
5774 const char *p;
5775
5776 assert(u);
5777
5778 if (!u->id)
5779 return;
5780
5781 /* Undoes the effect of unit_export_state() */
5782
5783 if (u->exported_invocation_id) {
5784 _cleanup_free_ char *invocation_path = NULL;
5785 int r = unit_get_invocation_path(u, &invocation_path);
5786 if (r >= 0) {
5787 (void) unlink(invocation_path);
5788 u->exported_invocation_id = false;
5789 }
5790 }
5791
5792 if (!MANAGER_IS_SYSTEM(u->manager))
5793 return;
5794
5795 if (u->exported_log_level_max) {
5796 p = strjoina("/run/systemd/units/log-level-max:", u->id);
5797 (void) unlink(p);
5798
5799 u->exported_log_level_max = false;
5800 }
5801
5802 if (u->exported_log_extra_fields) {
5803 p = strjoina("/run/systemd/units/extra-fields:", u->id);
5804 (void) unlink(p);
5805
5806 u->exported_log_extra_fields = false;
5807 }
5808
5809 if (u->exported_log_ratelimit_interval) {
5810 p = strjoina("/run/systemd/units/log-rate-limit-interval:", u->id);
5811 (void) unlink(p);
5812
5813 u->exported_log_ratelimit_interval = false;
5814 }
5815
5816 if (u->exported_log_ratelimit_burst) {
5817 p = strjoina("/run/systemd/units/log-rate-limit-burst:", u->id);
5818 (void) unlink(p);
5819
5820 u->exported_log_ratelimit_burst = false;
5821 }
5822}
5823
5824int unit_prepare_exec(Unit *u) {
5825 int r;
5826
5827 assert(u);
5828
5829 /* Load any custom firewall BPF programs here once to test if they are existing and actually loadable.
5830 * Fail here early since later errors in the call chain unit_realize_cgroup to cgroup_context_apply are ignored. */
5831 r = bpf_firewall_load_custom(u);
5832 if (r < 0)
5833 return r;
5834
5835 /* Prepares everything so that we can fork of a process for this unit */
5836
5837 (void) unit_realize_cgroup(u);
5838
5839 CGroupRuntime *crt = unit_get_cgroup_runtime(u);
5840 if (crt && crt->reset_accounting) {
5841 (void) unit_reset_accounting(u);
5842 crt->reset_accounting = false;
5843 }
5844
5845 unit_export_state_files(u);
5846
5847 r = unit_setup_exec_runtime(u);
5848 if (r < 0)
5849 return r;
5850
5851 return 0;
5852}
5853
5854static bool ignore_leftover_process(const char *comm) {
5855 return comm && comm[0] == '('; /* Most likely our own helper process (PAM?), ignore */
5856}
5857
5858int unit_log_leftover_process_start(const PidRef *pid, int sig, void *userdata) {
5859 _cleanup_free_ char *comm = NULL;
5860
5861 assert(pidref_is_set(pid));
5862
5863 (void) pidref_get_comm(pid, &comm);
5864
5865 if (ignore_leftover_process(comm))
5866 return 0;
5867
5868 /* During start we print a warning */
5869
5870 log_unit_warning(userdata,
5871 "Found left-over process " PID_FMT " (%s) in control group while starting unit. Ignoring.\n"
5872 "This usually indicates unclean termination of a previous run, or service implementation deficiencies.",
5873 pid->pid, strna(comm));
5874
5875 return 1;
5876}
5877
5878int unit_log_leftover_process_stop(const PidRef *pid, int sig, void *userdata) {
5879 _cleanup_free_ char *comm = NULL;
5880
5881 assert(pidref_is_set(pid));
5882
5883 (void) pidref_get_comm(pid, &comm);
5884
5885 if (ignore_leftover_process(comm))
5886 return 0;
5887
5888 /* During stop we only print an informational message */
5889
5890 log_unit_info(userdata,
5891 "Unit process " PID_FMT " (%s) remains running after unit stopped.",
5892 pid->pid, strna(comm));
5893
5894 return 1;
5895}
5896
5897int unit_warn_leftover_processes(Unit *u, cg_kill_log_func_t log_func) {
5898 assert(u);
5899
5900 (void) unit_pick_cgroup_path(u);
5901
5902 CGroupRuntime *crt = unit_get_cgroup_runtime(u);
5903
5904 if (!crt || !crt->cgroup_path)
5905 return 0;
5906
5907 return cg_kill_recursive(
5908 crt->cgroup_path,
5909 /* sig= */ 0,
5910 /* flags= */ 0,
5911 /* set= */ NULL,
5912 log_func,
5913 u);
5914}
5915
5916bool unit_needs_console(Unit *u) {
5917 ExecContext *ec;
5918 UnitActiveState state;
5919
5920 assert(u);
5921
5922 state = unit_active_state(u);
5923
5924 if (UNIT_IS_INACTIVE_OR_FAILED(state))
5925 return false;
5926
5927 if (UNIT_VTABLE(u)->needs_console)
5928 return UNIT_VTABLE(u)->needs_console(u);
5929
5930 /* If this unit type doesn't implement this call, let's use a generic fallback implementation: */
5931 ec = unit_get_exec_context(u);
5932 if (!ec)
5933 return false;
5934
5935 return exec_context_may_touch_console(ec);
5936}
5937
5938int unit_pid_attachable(Unit *u, const PidRef *pid, sd_bus_error *error) {
5939 int r;
5940
5941 assert(u);
5942
5943 /* Checks whether the specified PID is generally good for attaching, i.e. a valid PID, not our manager itself,
5944 * and not a kernel thread either */
5945
5946 /* First, a simple range check */
5947 if (!pidref_is_set(pid))
5948 return sd_bus_error_setf(error, SD_BUS_ERROR_INVALID_ARGS, "Process identifier is not valid.");
5949
5950 /* Some extra safety check */
5951 if (pid->pid == 1 || pidref_is_self(pid))
5952 return sd_bus_error_setf(error, SD_BUS_ERROR_INVALID_ARGS, "Process " PID_FMT " is a manager process, refusing.", pid->pid);
5953
5954 /* Don't even begin to bother with kernel threads */
5955 r = pidref_is_kernel_thread(pid);
5956 if (r == -ESRCH)
5957 return sd_bus_error_setf(error, SD_BUS_ERROR_UNIX_PROCESS_ID_UNKNOWN, "Process with ID " PID_FMT " does not exist.", pid->pid);
5958 if (r < 0)
5959 return sd_bus_error_set_errnof(error, r, "Failed to determine whether process " PID_FMT " is a kernel thread: %m", pid->pid);
5960 if (r > 0)
5961 return sd_bus_error_setf(error, SD_BUS_ERROR_INVALID_ARGS, "Process " PID_FMT " is a kernel thread, refusing.", pid->pid);
5962
5963 return 0;
5964}
5965
5966void unit_log_success(Unit *u) {
5967 assert(u);
5968
5969 /* Let's show message "Deactivated successfully" in debug mode (when manager is user) rather than in info mode.
5970 * This message has low information value for regular users and it might be a bit overwhelming on a system with
5971 * a lot of devices. */
5972 log_unit_struct(u,
5973 MANAGER_IS_USER(u->manager) ? LOG_DEBUG : LOG_INFO,
5974 "MESSAGE_ID=" SD_MESSAGE_UNIT_SUCCESS_STR,
5975 LOG_UNIT_INVOCATION_ID(u),
5976 LOG_UNIT_MESSAGE(u, "Deactivated successfully."));
5977}
5978
5979void unit_log_failure(Unit *u, const char *result) {
5980 assert(u);
5981 assert(result);
5982
5983 log_unit_struct(u, LOG_WARNING,
5984 "MESSAGE_ID=" SD_MESSAGE_UNIT_FAILURE_RESULT_STR,
5985 LOG_UNIT_INVOCATION_ID(u),
5986 LOG_UNIT_MESSAGE(u, "Failed with result '%s'.", result),
5987 "UNIT_RESULT=%s", result);
5988}
5989
5990void unit_log_skip(Unit *u, const char *result) {
5991 assert(u);
5992 assert(result);
5993
5994 log_unit_struct(u, LOG_INFO,
5995 "MESSAGE_ID=" SD_MESSAGE_UNIT_SKIPPED_STR,
5996 LOG_UNIT_INVOCATION_ID(u),
5997 LOG_UNIT_MESSAGE(u, "Skipped due to '%s'.", result),
5998 "UNIT_RESULT=%s", result);
5999}
6000
6001void unit_log_process_exit(
6002 Unit *u,
6003 const char *kind,
6004 const char *command,
6005 bool success,
6006 int code,
6007 int status) {
6008
6009 int level;
6010
6011 assert(u);
6012 assert(kind);
6013
6014 /* If this is a successful exit, let's log about the exit code on DEBUG level. If this is a failure
6015 * and the process exited on its own via exit(), then let's make this a NOTICE, under the assumption
6016 * that the service already logged the reason at a higher log level on its own. Otherwise, make it a
6017 * WARNING. */
6018 if (success)
6019 level = LOG_DEBUG;
6020 else if (code == CLD_EXITED)
6021 level = LOG_NOTICE;
6022 else
6023 level = LOG_WARNING;
6024
6025 log_unit_struct(u, level,
6026 "MESSAGE_ID=" SD_MESSAGE_UNIT_PROCESS_EXIT_STR,
6027 LOG_UNIT_MESSAGE(u, "%s exited, code=%s, status=%i/%s%s",
6028 kind,
6029 sigchld_code_to_string(code), status,
6030 strna(code == CLD_EXITED
6031 ? exit_status_to_string(status, EXIT_STATUS_FULL)
6032 : signal_to_string(status)),
6033 success ? " (success)" : ""),
6034 "EXIT_CODE=%s", sigchld_code_to_string(code),
6035 "EXIT_STATUS=%i", status,
6036 "COMMAND=%s", strna(command),
6037 LOG_UNIT_INVOCATION_ID(u));
6038}
6039
6040int unit_exit_status(Unit *u) {
6041 assert(u);
6042
6043 /* Returns the exit status to propagate for the most recent cycle of this unit. Returns a value in the range
6044 * 0…255 if there's something to propagate. EOPNOTSUPP if the concept does not apply to this unit type, ENODATA
6045 * if no data is currently known (for example because the unit hasn't deactivated yet) and EBADE if the main
6046 * service process has exited abnormally (signal/coredump). */
6047
6048 if (!UNIT_VTABLE(u)->exit_status)
6049 return -EOPNOTSUPP;
6050
6051 return UNIT_VTABLE(u)->exit_status(u);
6052}
6053
6054int unit_failure_action_exit_status(Unit *u) {
6055 int r;
6056
6057 assert(u);
6058
6059 /* Returns the exit status to propagate on failure, or an error if there's nothing to propagate */
6060
6061 if (u->failure_action_exit_status >= 0)
6062 return u->failure_action_exit_status;
6063
6064 r = unit_exit_status(u);
6065 if (r == -EBADE) /* Exited, but not cleanly (i.e. by signal or such) */
6066 return 255;
6067
6068 return r;
6069}
6070
6071int unit_success_action_exit_status(Unit *u) {
6072 int r;
6073
6074 assert(u);
6075
6076 /* Returns the exit status to propagate on success, or an error if there's nothing to propagate */
6077
6078 if (u->success_action_exit_status >= 0)
6079 return u->success_action_exit_status;
6080
6081 r = unit_exit_status(u);
6082 if (r == -EBADE) /* Exited, but not cleanly (i.e. by signal or such) */
6083 return 255;
6084
6085 return r;
6086}
6087
6088int unit_test_trigger_loaded(Unit *u) {
6089 Unit *trigger;
6090
6091 /* Tests whether the unit to trigger is loaded */
6092
6093 trigger = UNIT_TRIGGER(u);
6094 if (!trigger)
6095 return log_unit_error_errno(u, SYNTHETIC_ERRNO(ENOENT),
6096 "Refusing to start, no unit to trigger.");
6097 if (trigger->load_state != UNIT_LOADED)
6098 return log_unit_error_errno(u, SYNTHETIC_ERRNO(ENOENT),
6099 "Refusing to start, unit %s to trigger not loaded.", trigger->id);
6100
6101 return 0;
6102}
6103
6104void unit_destroy_runtime_data(Unit *u, const ExecContext *context) {
6105 assert(u);
6106 assert(context);
6107
6108 /* EXEC_PRESERVE_RESTART is handled via unit_release_resources()! */
6109 if (context->runtime_directory_preserve_mode == EXEC_PRESERVE_NO)
6110 exec_context_destroy_runtime_directory(context, u->manager->prefix[EXEC_DIRECTORY_RUNTIME]);
6111
6112 exec_context_destroy_credentials(u);
6113 exec_context_destroy_mount_ns_dir(u);
6114}
6115
6116int unit_clean(Unit *u, ExecCleanMask mask) {
6117 UnitActiveState state;
6118
6119 assert(u);
6120
6121 /* Special return values:
6122 *
6123 * -EOPNOTSUPP → cleaning not supported for this unit type
6124 * -EUNATCH → cleaning not defined for this resource type
6125 * -EBUSY → unit currently can't be cleaned since it's running or not properly loaded, or has
6126 * a job queued or similar
6127 */
6128
6129 if (!UNIT_VTABLE(u)->clean)
6130 return -EOPNOTSUPP;
6131
6132 if (mask == 0)
6133 return -EUNATCH;
6134
6135 if (u->load_state != UNIT_LOADED)
6136 return -EBUSY;
6137
6138 if (u->job)
6139 return -EBUSY;
6140
6141 state = unit_active_state(u);
6142 if (state != UNIT_INACTIVE)
6143 return -EBUSY;
6144
6145 return UNIT_VTABLE(u)->clean(u, mask);
6146}
6147
6148int unit_can_clean(Unit *u, ExecCleanMask *ret) {
6149 assert(u);
6150
6151 if (!UNIT_VTABLE(u)->clean ||
6152 u->load_state != UNIT_LOADED) {
6153 *ret = 0;
6154 return 0;
6155 }
6156
6157 /* When the clean() method is set, can_clean() really should be set too */
6158 assert(UNIT_VTABLE(u)->can_clean);
6159
6160 return UNIT_VTABLE(u)->can_clean(u, ret);
6161}
6162
6163bool unit_can_start_refuse_manual(Unit *u) {
6164 return unit_can_start(u) && !u->refuse_manual_start;
6165}
6166
6167bool unit_can_stop_refuse_manual(Unit *u) {
6168 return unit_can_stop(u) && !u->refuse_manual_stop;
6169}
6170
6171bool unit_can_isolate_refuse_manual(Unit *u) {
6172 return unit_can_isolate(u) && !u->refuse_manual_start;
6173}
6174
6175void unit_next_freezer_state(Unit *u, FreezerAction action, FreezerState *ret, FreezerState *ret_target) {
6176 Unit *slice;
6177 FreezerState curr, parent, next, tgt;
6178
6179 assert(u);
6180 assert(IN_SET(action, FREEZER_FREEZE, FREEZER_PARENT_FREEZE,
6181 FREEZER_THAW, FREEZER_PARENT_THAW));
6182 assert(ret);
6183 assert(ret_target);
6184
6185 /* This function determines the correct freezer state transitions for a unit
6186 * given the action being requested. It returns the next state, and also the "target",
6187 * which is either FREEZER_FROZEN or FREEZER_RUNNING, depending on what actual state we
6188 * ultimately want to achieve. */
6189
6190 curr = u->freezer_state;
6191 slice = UNIT_GET_SLICE(u);
6192 if (slice)
6193 parent = slice->freezer_state;
6194 else
6195 parent = FREEZER_RUNNING;
6196
6197 if (action == FREEZER_FREEZE) {
6198 /* We always "promote" a freeze initiated by parent into a normal freeze */
6199 if (IN_SET(curr, FREEZER_FROZEN, FREEZER_FROZEN_BY_PARENT))
6200 next = FREEZER_FROZEN;
6201 else
6202 next = FREEZER_FREEZING;
6203 } else if (action == FREEZER_THAW) {
6204 /* Thawing is the most complicated operation here, because we can't thaw a unit
6205 * if its parent is frozen. So we instead "demote" a normal freeze into a freeze
6206 * initiated by parent if the parent is frozen */
6207 if (IN_SET(curr, FREEZER_RUNNING, FREEZER_THAWING, FREEZER_FREEZING_BY_PARENT, FREEZER_FROZEN_BY_PARENT))
6208 next = curr;
6209 else if (curr == FREEZER_FREEZING) {
6210 if (IN_SET(parent, FREEZER_RUNNING, FREEZER_THAWING))
6211 next = FREEZER_THAWING;
6212 else
6213 next = FREEZER_FREEZING_BY_PARENT;
6214 } else {
6215 assert(curr == FREEZER_FROZEN);
6216 if (IN_SET(parent, FREEZER_RUNNING, FREEZER_THAWING))
6217 next = FREEZER_THAWING;
6218 else
6219 next = FREEZER_FROZEN_BY_PARENT;
6220 }
6221 } else if (action == FREEZER_PARENT_FREEZE) {
6222 /* We need to avoid accidentally demoting units frozen manually */
6223 if (IN_SET(curr, FREEZER_FREEZING, FREEZER_FROZEN, FREEZER_FROZEN_BY_PARENT))
6224 next = curr;
6225 else
6226 next = FREEZER_FREEZING_BY_PARENT;
6227 } else {
6228 assert(action == FREEZER_PARENT_THAW);
6229
6230 /* We don't want to thaw units from a parent if they were frozen
6231 * manually, so for such units this action is a no-op */
6232 if (IN_SET(curr, FREEZER_RUNNING, FREEZER_FREEZING, FREEZER_FROZEN))
6233 next = curr;
6234 else
6235 next = FREEZER_THAWING;
6236 }
6237
6238 tgt = freezer_state_finish(next);
6239 if (tgt == FREEZER_FROZEN_BY_PARENT)
6240 tgt = FREEZER_FROZEN;
6241 assert(IN_SET(tgt, FREEZER_RUNNING, FREEZER_FROZEN));
6242
6243 *ret = next;
6244 *ret_target = tgt;
6245}
6246
6247bool unit_can_freeze(Unit *u) {
6248 assert(u);
6249
6250 if (unit_has_name(u, SPECIAL_ROOT_SLICE) || unit_has_name(u, SPECIAL_INIT_SCOPE))
6251 return false;
6252
6253 if (UNIT_VTABLE(u)->can_freeze)
6254 return UNIT_VTABLE(u)->can_freeze(u);
6255
6256 return UNIT_VTABLE(u)->freezer_action;
6257}
6258
6259void unit_frozen(Unit *u) {
6260 assert(u);
6261
6262 u->freezer_state = u->freezer_state == FREEZER_FREEZING_BY_PARENT
6263 ? FREEZER_FROZEN_BY_PARENT
6264 : FREEZER_FROZEN;
6265
6266 log_unit_debug(u, "Unit now %s.", freezer_state_to_string(u->freezer_state));
6267
6268 bus_unit_send_pending_freezer_message(u, false);
6269}
6270
6271void unit_thawed(Unit *u) {
6272 assert(u);
6273
6274 u->freezer_state = FREEZER_RUNNING;
6275
6276 log_unit_debug(u, "Unit thawed.");
6277
6278 bus_unit_send_pending_freezer_message(u, false);
6279}
6280
6281int unit_freezer_action(Unit *u, FreezerAction action) {
6282 UnitActiveState s;
6283 int r;
6284
6285 assert(u);
6286 assert(IN_SET(action, FREEZER_FREEZE, FREEZER_THAW));
6287
6288 if (!cg_freezer_supported() || !unit_can_freeze(u))
6289 return -EOPNOTSUPP;
6290
6291 if (u->job)
6292 return -EBUSY;
6293
6294 if (u->load_state != UNIT_LOADED)
6295 return -EHOSTDOWN;
6296
6297 s = unit_active_state(u);
6298 if (s != UNIT_ACTIVE)
6299 return -EHOSTDOWN;
6300
6301 if (action == FREEZER_FREEZE && IN_SET(u->freezer_state, FREEZER_FREEZING, FREEZER_FREEZING_BY_PARENT))
6302 return -EALREADY;
6303 if (action == FREEZER_THAW && u->freezer_state == FREEZER_THAWING)
6304 return -EALREADY;
6305 if (action == FREEZER_THAW && IN_SET(u->freezer_state, FREEZER_FREEZING_BY_PARENT, FREEZER_FROZEN_BY_PARENT))
6306 return -ECHILD;
6307
6308 r = UNIT_VTABLE(u)->freezer_action(u, action);
6309 if (r <= 0)
6310 return r;
6311
6312 assert(IN_SET(u->freezer_state, FREEZER_FREEZING, FREEZER_FREEZING_BY_PARENT, FREEZER_THAWING));
6313 return 1;
6314}
6315
6316Condition *unit_find_failed_condition(Unit *u) {
6317 Condition *failed_trigger = NULL;
6318 bool has_succeeded_trigger = false;
6319
6320 if (u->condition_result)
6321 return NULL;
6322
6323 LIST_FOREACH(conditions, c, u->conditions)
6324 if (c->trigger) {
6325 if (c->result == CONDITION_SUCCEEDED)
6326 has_succeeded_trigger = true;
6327 else if (!failed_trigger)
6328 failed_trigger = c;
6329 } else if (c->result != CONDITION_SUCCEEDED)
6330 return c;
6331
6332 return failed_trigger && !has_succeeded_trigger ? failed_trigger : NULL;
6333}
6334
6335static const char* const collect_mode_table[_COLLECT_MODE_MAX] = {
6336 [COLLECT_INACTIVE] = "inactive",
6337 [COLLECT_INACTIVE_OR_FAILED] = "inactive-or-failed",
6338};
6339
6340DEFINE_STRING_TABLE_LOOKUP(collect_mode, CollectMode);
6341
6342Unit* unit_has_dependency(const Unit *u, UnitDependencyAtom atom, Unit *other) {
6343 Unit *i;
6344
6345 assert(u);
6346
6347 /* Checks if the unit has a dependency on 'other' with the specified dependency atom. If 'other' is
6348 * NULL checks if the unit has *any* dependency of that atom. Returns 'other' if found (or if 'other'
6349 * is NULL the first entry found), or NULL if not found. */
6350
6351 UNIT_FOREACH_DEPENDENCY(i, u, atom)
6352 if (!other || other == i)
6353 return i;
6354
6355 return NULL;
6356}
6357
6358int unit_get_dependency_array(const Unit *u, UnitDependencyAtom atom, Unit ***ret_array) {
6359 _cleanup_free_ Unit **array = NULL;
6360 size_t n = 0;
6361 Unit *other;
6362
6363 assert(u);
6364 assert(ret_array);
6365
6366 /* Gets a list of units matching a specific atom as array. This is useful when iterating through
6367 * dependencies while modifying them: the array is an "atomic snapshot" of sorts, that can be read
6368 * while the dependency table is continuously updated. */
6369
6370 UNIT_FOREACH_DEPENDENCY(other, u, atom) {
6371 if (!GREEDY_REALLOC(array, n + 1))
6372 return -ENOMEM;
6373
6374 array[n++] = other;
6375 }
6376
6377 *ret_array = TAKE_PTR(array);
6378
6379 assert(n <= INT_MAX);
6380 return (int) n;
6381}
6382
6383int unit_get_transitive_dependency_set(Unit *u, UnitDependencyAtom atom, Set **ret) {
6384 _cleanup_set_free_ Set *units = NULL, *queue = NULL;
6385 Unit *other;
6386 int r;
6387
6388 assert(u);
6389 assert(ret);
6390
6391 /* Similar to unit_get_dependency_array(), but also search the same dependency in other units. */
6392
6393 do {
6394 UNIT_FOREACH_DEPENDENCY(other, u, atom) {
6395 r = set_ensure_put(&units, NULL, other);
6396 if (r < 0)
6397 return r;
6398 if (r == 0)
6399 continue;
6400 r = set_ensure_put(&queue, NULL, other);
6401 if (r < 0)
6402 return r;
6403 }
6404 } while ((u = set_steal_first(queue)));
6405
6406 *ret = TAKE_PTR(units);
6407 return 0;
6408}
6409
6410int unit_arm_timer(
6411 Unit *u,
6412 sd_event_source **source,
6413 bool relative,
6414 usec_t usec,
6415 sd_event_time_handler_t handler) {
6416
6417 int r;
6418
6419 assert(u);
6420 assert(source);
6421 assert(handler);
6422
6423 if (*source) {
6424 if (usec == USEC_INFINITY)
6425 return sd_event_source_set_enabled(*source, SD_EVENT_OFF);
6426
6427 r = (relative ? sd_event_source_set_time_relative : sd_event_source_set_time)(*source, usec);
6428 if (r < 0)
6429 return r;
6430
6431 return sd_event_source_set_enabled(*source, SD_EVENT_ONESHOT);
6432 }
6433
6434 if (usec == USEC_INFINITY)
6435 return 0;
6436
6437 r = (relative ? sd_event_add_time_relative : sd_event_add_time)(
6438 u->manager->event,
6439 source,
6440 CLOCK_MONOTONIC,
6441 usec, 0,
6442 handler,
6443 u);
6444 if (r < 0)
6445 return r;
6446
6447 const char *d = strjoina(unit_type_to_string(u->type), "-timer");
6448 (void) sd_event_source_set_description(*source, d);
6449
6450 return 0;
6451}
6452
6453static int unit_get_nice(Unit *u) {
6454 ExecContext *ec;
6455
6456 ec = unit_get_exec_context(u);
6457 return ec ? ec->nice : 0;
6458}
6459
6460static uint64_t unit_get_cpu_weight(Unit *u) {
6461 CGroupContext *cc;
6462
6463 cc = unit_get_cgroup_context(u);
6464 return cc ? cgroup_context_cpu_weight(cc, manager_state(u->manager)) : CGROUP_WEIGHT_DEFAULT;
6465}
6466
6467int unit_compare_priority(Unit *a, Unit *b) {
6468 int ret;
6469
6470 ret = CMP(a->type, b->type);
6471 if (ret != 0)
6472 return -ret;
6473
6474 ret = CMP(unit_get_cpu_weight(a), unit_get_cpu_weight(b));
6475 if (ret != 0)
6476 return -ret;
6477
6478 ret = CMP(unit_get_nice(a), unit_get_nice(b));
6479 if (ret != 0)
6480 return ret;
6481
6482 return strcmp(a->id, b->id);
6483}
6484
6485const ActivationDetailsVTable * const activation_details_vtable[_UNIT_TYPE_MAX] = {
6486 [UNIT_PATH] = &activation_details_path_vtable,
6487 [UNIT_TIMER] = &activation_details_timer_vtable,
6488};
6489
6490ActivationDetails *activation_details_new(Unit *trigger_unit) {
6491 _cleanup_free_ ActivationDetails *details = NULL;
6492
6493 assert(trigger_unit);
6494 assert(trigger_unit->type != _UNIT_TYPE_INVALID);
6495 assert(trigger_unit->id);
6496
6497 details = malloc0(activation_details_vtable[trigger_unit->type]->object_size);
6498 if (!details)
6499 return NULL;
6500
6501 *details = (ActivationDetails) {
6502 .n_ref = 1,
6503 .trigger_unit_type = trigger_unit->type,
6504 };
6505
6506 details->trigger_unit_name = strdup(trigger_unit->id);
6507 if (!details->trigger_unit_name)
6508 return NULL;
6509
6510 if (ACTIVATION_DETAILS_VTABLE(details)->init)
6511 ACTIVATION_DETAILS_VTABLE(details)->init(details, trigger_unit);
6512
6513 return TAKE_PTR(details);
6514}
6515
6516static ActivationDetails *activation_details_free(ActivationDetails *details) {
6517 if (!details)
6518 return NULL;
6519
6520 if (ACTIVATION_DETAILS_VTABLE(details)->done)
6521 ACTIVATION_DETAILS_VTABLE(details)->done(details);
6522
6523 free(details->trigger_unit_name);
6524
6525 return mfree(details);
6526}
6527
6528void activation_details_serialize(ActivationDetails *details, FILE *f) {
6529 if (!details || details->trigger_unit_type == _UNIT_TYPE_INVALID)
6530 return;
6531
6532 (void) serialize_item(f, "activation-details-unit-type", unit_type_to_string(details->trigger_unit_type));
6533 if (details->trigger_unit_name)
6534 (void) serialize_item(f, "activation-details-unit-name", details->trigger_unit_name);
6535 if (ACTIVATION_DETAILS_VTABLE(details)->serialize)
6536 ACTIVATION_DETAILS_VTABLE(details)->serialize(details, f);
6537}
6538
6539int activation_details_deserialize(const char *key, const char *value, ActivationDetails **details) {
6540 int r;
6541
6542 assert(key);
6543 assert(value);
6544 assert(details);
6545
6546 if (!*details) {
6547 UnitType t;
6548
6549 if (!streq(key, "activation-details-unit-type"))
6550 return -EINVAL;
6551
6552 t = unit_type_from_string(value);
6553 if (t < 0)
6554 return t;
6555
6556 /* The activation details vtable has defined ops only for path and timer units */
6557 if (!activation_details_vtable[t])
6558 return -EINVAL;
6559
6560 *details = malloc0(activation_details_vtable[t]->object_size);
6561 if (!*details)
6562 return -ENOMEM;
6563
6564 **details = (ActivationDetails) {
6565 .n_ref = 1,
6566 .trigger_unit_type = t,
6567 };
6568
6569 return 0;
6570 }
6571
6572 if (streq(key, "activation-details-unit-name")) {
6573 r = free_and_strdup(&(*details)->trigger_unit_name, value);
6574 if (r < 0)
6575 return r;
6576
6577 return 0;
6578 }
6579
6580 if (ACTIVATION_DETAILS_VTABLE(*details)->deserialize)
6581 return ACTIVATION_DETAILS_VTABLE(*details)->deserialize(key, value, details);
6582
6583 return -EINVAL;
6584}
6585
6586int activation_details_append_env(ActivationDetails *details, char ***strv) {
6587 int r = 0;
6588
6589 assert(strv);
6590
6591 if (!details)
6592 return 0;
6593
6594 if (!isempty(details->trigger_unit_name)) {
6595 char *s = strjoin("TRIGGER_UNIT=", details->trigger_unit_name);
6596 if (!s)
6597 return -ENOMEM;
6598
6599 r = strv_consume(strv, TAKE_PTR(s));
6600 if (r < 0)
6601 return r;
6602 }
6603
6604 if (ACTIVATION_DETAILS_VTABLE(details)->append_env) {
6605 r = ACTIVATION_DETAILS_VTABLE(details)->append_env(details, strv);
6606 if (r < 0)
6607 return r;
6608 }
6609
6610 return r + !isempty(details->trigger_unit_name); /* Return the number of variables added to the env block */
6611}
6612
6613int activation_details_append_pair(ActivationDetails *details, char ***strv) {
6614 int r = 0;
6615
6616 assert(strv);
6617
6618 if (!details)
6619 return 0;
6620
6621 if (!isempty(details->trigger_unit_name)) {
6622 r = strv_extend_many(strv, "trigger_unit", details->trigger_unit_name);
6623 if (r < 0)
6624 return r;
6625 }
6626
6627 if (ACTIVATION_DETAILS_VTABLE(details)->append_pair) {
6628 r = ACTIVATION_DETAILS_VTABLE(details)->append_pair(details, strv);
6629 if (r < 0)
6630 return r;
6631 }
6632
6633 return r + !isempty(details->trigger_unit_name); /* Return the number of pairs added to the strv */
6634}
6635
6636DEFINE_TRIVIAL_REF_UNREF_FUNC(ActivationDetails, activation_details, activation_details_free);
6637
6638static const char* const unit_mount_dependency_type_table[_UNIT_MOUNT_DEPENDENCY_TYPE_MAX] = {
6639 [UNIT_MOUNT_WANTS] = "WantsMountsFor",
6640 [UNIT_MOUNT_REQUIRES] = "RequiresMountsFor",
6641};
6642
6643DEFINE_STRING_TABLE_LOOKUP(unit_mount_dependency_type, UnitMountDependencyType);
6644
6645UnitDependency unit_mount_dependency_type_to_dependency_type(UnitMountDependencyType t) {
6646 switch (t) {
6647
6648 case UNIT_MOUNT_WANTS:
6649 return UNIT_WANTS;
6650
6651 case UNIT_MOUNT_REQUIRES:
6652 return UNIT_REQUIRES;
6653
6654 default:
6655 assert_not_reached();
6656 }
6657}