]> git.ipfire.org Git - thirdparty/systemd.git/blob - src/core/unit.c
Merge pull request #20199 from ddstreet/unit_cgroup_catchup
[thirdparty/systemd.git] / src / core / unit.c
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-util.h"
18 #include "cgroup-setup.h"
19 #include "cgroup-util.h"
20 #include "core-varlink.h"
21 #include "dbus-unit.h"
22 #include "dbus.h"
23 #include "dropin.h"
24 #include "escape.h"
25 #include "execute.h"
26 #include "fd-util.h"
27 #include "fileio-label.h"
28 #include "fileio.h"
29 #include "format-util.h"
30 #include "id128-util.h"
31 #include "install.h"
32 #include "io-util.h"
33 #include "label.h"
34 #include "load-dropin.h"
35 #include "load-fragment.h"
36 #include "log.h"
37 #include "macro.h"
38 #include "missing_audit.h"
39 #include "mkdir.h"
40 #include "path-util.h"
41 #include "process-util.h"
42 #include "rm-rf.h"
43 #include "set.h"
44 #include "signal-util.h"
45 #include "sparse-endian.h"
46 #include "special.h"
47 #include "specifier.h"
48 #include "stat-util.h"
49 #include "stdio-util.h"
50 #include "string-table.h"
51 #include "string-util.h"
52 #include "strv.h"
53 #include "terminal-util.h"
54 #include "tmpfile-util.h"
55 #include "umask-util.h"
56 #include "unit-name.h"
57 #include "unit.h"
58 #include "user-util.h"
59 #include "virt.h"
60 #if BPF_FRAMEWORK
61 #include "bpf-link.h"
62 #endif
63
64 /* Thresholds for logging at INFO level about resource consumption */
65 #define MENTIONWORTHY_CPU_NSEC (1 * NSEC_PER_SEC)
66 #define MENTIONWORTHY_IO_BYTES (1024 * 1024ULL)
67 #define MENTIONWORTHY_IP_BYTES (0ULL)
68
69 /* Thresholds for logging at INFO level about resource consumption */
70 #define NOTICEWORTHY_CPU_NSEC (10*60 * NSEC_PER_SEC) /* 10 minutes */
71 #define NOTICEWORTHY_IO_BYTES (10 * 1024 * 1024ULL) /* 10 MB */
72 #define NOTICEWORTHY_IP_BYTES (128 * 1024 * 1024ULL) /* 128 MB */
73
74 const UnitVTable * const unit_vtable[_UNIT_TYPE_MAX] = {
75 [UNIT_SERVICE] = &service_vtable,
76 [UNIT_SOCKET] = &socket_vtable,
77 [UNIT_TARGET] = &target_vtable,
78 [UNIT_DEVICE] = &device_vtable,
79 [UNIT_MOUNT] = &mount_vtable,
80 [UNIT_AUTOMOUNT] = &automount_vtable,
81 [UNIT_SWAP] = &swap_vtable,
82 [UNIT_TIMER] = &timer_vtable,
83 [UNIT_PATH] = &path_vtable,
84 [UNIT_SLICE] = &slice_vtable,
85 [UNIT_SCOPE] = &scope_vtable,
86 };
87
88 Unit* unit_new(Manager *m, size_t size) {
89 Unit *u;
90
91 assert(m);
92 assert(size >= sizeof(Unit));
93
94 u = malloc0(size);
95 if (!u)
96 return NULL;
97
98 u->manager = m;
99 u->type = _UNIT_TYPE_INVALID;
100 u->default_dependencies = true;
101 u->unit_file_state = _UNIT_FILE_STATE_INVALID;
102 u->unit_file_preset = -1;
103 u->on_failure_job_mode = JOB_REPLACE;
104 u->on_success_job_mode = JOB_FAIL;
105 u->cgroup_control_inotify_wd = -1;
106 u->cgroup_memory_inotify_wd = -1;
107 u->job_timeout = USEC_INFINITY;
108 u->job_running_timeout = USEC_INFINITY;
109 u->ref_uid = UID_INVALID;
110 u->ref_gid = GID_INVALID;
111 u->cpu_usage_last = NSEC_INFINITY;
112 u->cgroup_invalidated_mask |= CGROUP_MASK_BPF_FIREWALL;
113 u->failure_action_exit_status = u->success_action_exit_status = -1;
114
115 u->ip_accounting_ingress_map_fd = -1;
116 u->ip_accounting_egress_map_fd = -1;
117 for (CGroupIOAccountingMetric i = 0; i < _CGROUP_IO_ACCOUNTING_METRIC_MAX; i++)
118 u->io_accounting_last[i] = UINT64_MAX;
119
120 u->ipv4_allow_map_fd = -1;
121 u->ipv6_allow_map_fd = -1;
122 u->ipv4_deny_map_fd = -1;
123 u->ipv6_deny_map_fd = -1;
124
125 u->last_section_private = -1;
126
127 u->start_ratelimit = (RateLimit) { m->default_start_limit_interval, m->default_start_limit_burst };
128 u->auto_start_stop_ratelimit = (RateLimit) { 10 * USEC_PER_SEC, 16 };
129
130 return u;
131 }
132
133 int unit_new_for_name(Manager *m, size_t size, const char *name, Unit **ret) {
134 _cleanup_(unit_freep) Unit *u = NULL;
135 int r;
136
137 u = unit_new(m, size);
138 if (!u)
139 return -ENOMEM;
140
141 r = unit_add_name(u, name);
142 if (r < 0)
143 return r;
144
145 *ret = TAKE_PTR(u);
146
147 return r;
148 }
149
150 bool unit_has_name(const Unit *u, const char *name) {
151 assert(u);
152 assert(name);
153
154 return streq_ptr(name, u->id) ||
155 set_contains(u->aliases, name);
156 }
157
158 static void unit_init(Unit *u) {
159 CGroupContext *cc;
160 ExecContext *ec;
161 KillContext *kc;
162
163 assert(u);
164 assert(u->manager);
165 assert(u->type >= 0);
166
167 cc = unit_get_cgroup_context(u);
168 if (cc) {
169 cgroup_context_init(cc);
170
171 /* Copy in the manager defaults into the cgroup
172 * context, _before_ the rest of the settings have
173 * been initialized */
174
175 cc->cpu_accounting = u->manager->default_cpu_accounting;
176 cc->io_accounting = u->manager->default_io_accounting;
177 cc->blockio_accounting = u->manager->default_blockio_accounting;
178 cc->memory_accounting = u->manager->default_memory_accounting;
179 cc->tasks_accounting = u->manager->default_tasks_accounting;
180 cc->ip_accounting = u->manager->default_ip_accounting;
181
182 if (u->type != UNIT_SLICE)
183 cc->tasks_max = u->manager->default_tasks_max;
184 }
185
186 ec = unit_get_exec_context(u);
187 if (ec) {
188 exec_context_init(ec);
189
190 if (MANAGER_IS_SYSTEM(u->manager))
191 ec->keyring_mode = EXEC_KEYRING_SHARED;
192 else {
193 ec->keyring_mode = EXEC_KEYRING_INHERIT;
194
195 /* User manager might have its umask redefined by PAM or UMask=. In this
196 * case let the units it manages inherit this value by default. They can
197 * still tune this value through their own unit file */
198 (void) get_process_umask(getpid_cached(), &ec->umask);
199 }
200 }
201
202 kc = unit_get_kill_context(u);
203 if (kc)
204 kill_context_init(kc);
205
206 if (UNIT_VTABLE(u)->init)
207 UNIT_VTABLE(u)->init(u);
208 }
209
210 static int unit_add_alias(Unit *u, char *donated_name) {
211 int r;
212
213 /* Make sure that u->names is allocated. We may leave u->names
214 * empty if we fail later, but this is not a problem. */
215 r = set_ensure_put(&u->aliases, &string_hash_ops, donated_name);
216 if (r < 0)
217 return r;
218 assert(r > 0);
219
220 return 0;
221 }
222
223 int unit_add_name(Unit *u, const char *text) {
224 _cleanup_free_ char *name = NULL, *instance = NULL;
225 UnitType t;
226 int r;
227
228 assert(u);
229 assert(text);
230
231 if (unit_name_is_valid(text, UNIT_NAME_TEMPLATE)) {
232 if (!u->instance)
233 return log_unit_debug_errno(u, SYNTHETIC_ERRNO(EINVAL),
234 "instance is not set when adding name '%s': %m", text);
235
236 r = unit_name_replace_instance(text, u->instance, &name);
237 if (r < 0)
238 return log_unit_debug_errno(u, r,
239 "failed to build instance name from '%s': %m", text);
240 } else {
241 name = strdup(text);
242 if (!name)
243 return -ENOMEM;
244 }
245
246 if (unit_has_name(u, name))
247 return 0;
248
249 if (hashmap_contains(u->manager->units, name))
250 return log_unit_debug_errno(u, SYNTHETIC_ERRNO(EEXIST),
251 "unit already exist when adding name '%s': %m", name);
252
253 if (!unit_name_is_valid(name, UNIT_NAME_PLAIN|UNIT_NAME_INSTANCE))
254 return log_unit_debug_errno(u, SYNTHETIC_ERRNO(EINVAL),
255 "name '%s' is invalid: %m", name);
256
257 t = unit_name_to_type(name);
258 if (t < 0)
259 return log_unit_debug_errno(u, SYNTHETIC_ERRNO(EINVAL),
260 "failed to derive unit type from name '%s': %m", name);
261
262 if (u->type != _UNIT_TYPE_INVALID && t != u->type)
263 return log_unit_debug_errno(u, SYNTHETIC_ERRNO(EINVAL),
264 "unit type is illegal: u->type(%d) and t(%d) for name '%s': %m",
265 u->type, t, name);
266
267 r = unit_name_to_instance(name, &instance);
268 if (r < 0)
269 return log_unit_debug_errno(u, r, "failed to extract instance from name '%s': %m", name);
270
271 if (instance && !unit_type_may_template(t))
272 return log_unit_debug_errno(u, SYNTHETIC_ERRNO(EINVAL), "templates are not allowed for name '%s': %m", name);
273
274 /* Ensure that this unit either has no instance, or that the instance matches. */
275 if (u->type != _UNIT_TYPE_INVALID && !streq_ptr(u->instance, instance))
276 return log_unit_debug_errno(u, SYNTHETIC_ERRNO(EINVAL),
277 "cannot add name %s, the instances don't match (\"%s\" != \"%s\").",
278 name, instance, u->instance);
279
280 if (u->id && !unit_type_may_alias(t))
281 return log_unit_debug_errno(u, SYNTHETIC_ERRNO(EEXIST),
282 "cannot add name %s, aliases are not allowed for %s units.",
283 name, unit_type_to_string(t));
284
285 if (hashmap_size(u->manager->units) >= MANAGER_MAX_NAMES)
286 return log_unit_warning_errno(u, SYNTHETIC_ERRNO(E2BIG), "cannot add name, manager has too many units: %m");
287
288 /* Add name to the global hashmap first, because that's easier to undo */
289 r = hashmap_put(u->manager->units, name, u);
290 if (r < 0)
291 return log_unit_debug_errno(u, r, "add unit to hashmap failed for name '%s': %m", text);
292
293 if (u->id) {
294 r = unit_add_alias(u, name); /* unit_add_alias() takes ownership of the name on success */
295 if (r < 0) {
296 hashmap_remove(u->manager->units, name);
297 return r;
298 }
299 TAKE_PTR(name);
300
301 } else {
302 /* A new name, we don't need the set yet. */
303 assert(u->type == _UNIT_TYPE_INVALID);
304 assert(!u->instance);
305
306 u->type = t;
307 u->id = TAKE_PTR(name);
308 u->instance = TAKE_PTR(instance);
309
310 LIST_PREPEND(units_by_type, u->manager->units_by_type[t], u);
311 unit_init(u);
312 }
313
314 unit_add_to_dbus_queue(u);
315 return 0;
316 }
317
318 int unit_choose_id(Unit *u, const char *name) {
319 _cleanup_free_ char *t = NULL;
320 char *s;
321 int r;
322
323 assert(u);
324 assert(name);
325
326 if (unit_name_is_valid(name, UNIT_NAME_TEMPLATE)) {
327 if (!u->instance)
328 return -EINVAL;
329
330 r = unit_name_replace_instance(name, u->instance, &t);
331 if (r < 0)
332 return r;
333
334 name = t;
335 }
336
337 if (streq_ptr(u->id, name))
338 return 0; /* Nothing to do. */
339
340 /* Selects one of the aliases of this unit as the id */
341 s = set_get(u->aliases, (char*) name);
342 if (!s)
343 return -ENOENT;
344
345 if (u->id) {
346 r = set_remove_and_put(u->aliases, name, u->id);
347 if (r < 0)
348 return r;
349 } else
350 assert_se(set_remove(u->aliases, name)); /* see set_get() above… */
351
352 u->id = s; /* Old u->id is now stored in the set, and s is not stored anywhere */
353 unit_add_to_dbus_queue(u);
354
355 return 0;
356 }
357
358 int unit_set_description(Unit *u, const char *description) {
359 int r;
360
361 assert(u);
362
363 r = free_and_strdup(&u->description, empty_to_null(description));
364 if (r < 0)
365 return r;
366 if (r > 0)
367 unit_add_to_dbus_queue(u);
368
369 return 0;
370 }
371
372 bool unit_may_gc(Unit *u) {
373 UnitActiveState state;
374 int r;
375
376 assert(u);
377
378 /* Checks whether the unit is ready to be unloaded for garbage collection.
379 * Returns true when the unit may be collected, and false if there's some
380 * reason to keep it loaded.
381 *
382 * References from other units are *not* checked here. Instead, this is done
383 * in unit_gc_sweep(), but using markers to properly collect dependency loops.
384 */
385
386 if (u->job)
387 return false;
388
389 if (u->nop_job)
390 return false;
391
392 state = unit_active_state(u);
393
394 /* If the unit is inactive and failed and no job is queued for it, then release its runtime resources */
395 if (UNIT_IS_INACTIVE_OR_FAILED(state) &&
396 UNIT_VTABLE(u)->release_resources)
397 UNIT_VTABLE(u)->release_resources(u);
398
399 if (u->perpetual)
400 return false;
401
402 if (sd_bus_track_count(u->bus_track) > 0)
403 return false;
404
405 /* But we keep the unit object around for longer when it is referenced or configured to not be gc'ed */
406 switch (u->collect_mode) {
407
408 case COLLECT_INACTIVE:
409 if (state != UNIT_INACTIVE)
410 return false;
411
412 break;
413
414 case COLLECT_INACTIVE_OR_FAILED:
415 if (!IN_SET(state, UNIT_INACTIVE, UNIT_FAILED))
416 return false;
417
418 break;
419
420 default:
421 assert_not_reached();
422 }
423
424 if (u->cgroup_path) {
425 /* If the unit has a cgroup, then check whether there's anything in it. If so, we should stay
426 * around. Units with active processes should never be collected. */
427
428 r = cg_is_empty_recursive(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path);
429 if (r < 0)
430 log_unit_debug_errno(u, r, "Failed to determine whether cgroup %s is empty: %m", empty_to_root(u->cgroup_path));
431 if (r <= 0)
432 return false;
433 }
434
435 if (UNIT_VTABLE(u)->may_gc && !UNIT_VTABLE(u)->may_gc(u))
436 return false;
437
438 return true;
439 }
440
441 void unit_add_to_load_queue(Unit *u) {
442 assert(u);
443 assert(u->type != _UNIT_TYPE_INVALID);
444
445 if (u->load_state != UNIT_STUB || u->in_load_queue)
446 return;
447
448 LIST_PREPEND(load_queue, u->manager->load_queue, u);
449 u->in_load_queue = true;
450 }
451
452 void unit_add_to_cleanup_queue(Unit *u) {
453 assert(u);
454
455 if (u->in_cleanup_queue)
456 return;
457
458 LIST_PREPEND(cleanup_queue, u->manager->cleanup_queue, u);
459 u->in_cleanup_queue = true;
460 }
461
462 void unit_add_to_gc_queue(Unit *u) {
463 assert(u);
464
465 if (u->in_gc_queue || u->in_cleanup_queue)
466 return;
467
468 if (!unit_may_gc(u))
469 return;
470
471 LIST_PREPEND(gc_queue, u->manager->gc_unit_queue, u);
472 u->in_gc_queue = true;
473 }
474
475 void unit_add_to_dbus_queue(Unit *u) {
476 assert(u);
477 assert(u->type != _UNIT_TYPE_INVALID);
478
479 if (u->load_state == UNIT_STUB || u->in_dbus_queue)
480 return;
481
482 /* Shortcut things if nobody cares */
483 if (sd_bus_track_count(u->manager->subscribed) <= 0 &&
484 sd_bus_track_count(u->bus_track) <= 0 &&
485 set_isempty(u->manager->private_buses)) {
486 u->sent_dbus_new_signal = true;
487 return;
488 }
489
490 LIST_PREPEND(dbus_queue, u->manager->dbus_unit_queue, u);
491 u->in_dbus_queue = true;
492 }
493
494 void unit_submit_to_stop_when_unneeded_queue(Unit *u) {
495 assert(u);
496
497 if (u->in_stop_when_unneeded_queue)
498 return;
499
500 if (!u->stop_when_unneeded)
501 return;
502
503 if (!UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u)))
504 return;
505
506 LIST_PREPEND(stop_when_unneeded_queue, u->manager->stop_when_unneeded_queue, u);
507 u->in_stop_when_unneeded_queue = true;
508 }
509
510 void unit_submit_to_start_when_upheld_queue(Unit *u) {
511 assert(u);
512
513 if (u->in_start_when_upheld_queue)
514 return;
515
516 if (!UNIT_IS_INACTIVE_OR_FAILED(unit_active_state(u)))
517 return;
518
519 if (!unit_has_dependency(u, UNIT_ATOM_START_STEADILY, NULL))
520 return;
521
522 LIST_PREPEND(start_when_upheld_queue, u->manager->start_when_upheld_queue, u);
523 u->in_start_when_upheld_queue = true;
524 }
525
526 void unit_submit_to_stop_when_bound_queue(Unit *u) {
527 assert(u);
528
529 if (u->in_stop_when_bound_queue)
530 return;
531
532 if (!UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u)))
533 return;
534
535 if (!unit_has_dependency(u, UNIT_ATOM_CANNOT_BE_ACTIVE_WITHOUT, NULL))
536 return;
537
538 LIST_PREPEND(stop_when_bound_queue, u->manager->stop_when_bound_queue, u);
539 u->in_stop_when_bound_queue = true;
540 }
541
542 static void unit_clear_dependencies(Unit *u) {
543 assert(u);
544
545 /* Removes all dependencies configured on u and their reverse dependencies. */
546
547 for (Hashmap *deps; (deps = hashmap_steal_first(u->dependencies));) {
548
549 for (Unit *other; (other = hashmap_steal_first_key(deps));) {
550 Hashmap *other_deps;
551
552 HASHMAP_FOREACH(other_deps, other->dependencies)
553 hashmap_remove(other_deps, u);
554
555 unit_add_to_gc_queue(other);
556 }
557
558 hashmap_free(deps);
559 }
560
561 u->dependencies = hashmap_free(u->dependencies);
562 }
563
564 static void unit_remove_transient(Unit *u) {
565 char **i;
566
567 assert(u);
568
569 if (!u->transient)
570 return;
571
572 if (u->fragment_path)
573 (void) unlink(u->fragment_path);
574
575 STRV_FOREACH(i, u->dropin_paths) {
576 _cleanup_free_ char *p = NULL, *pp = NULL;
577
578 p = dirname_malloc(*i); /* Get the drop-in directory from the drop-in file */
579 if (!p)
580 continue;
581
582 pp = dirname_malloc(p); /* Get the config directory from the drop-in directory */
583 if (!pp)
584 continue;
585
586 /* Only drop transient drop-ins */
587 if (!path_equal(u->manager->lookup_paths.transient, pp))
588 continue;
589
590 (void) unlink(*i);
591 (void) rmdir(p);
592 }
593 }
594
595 static void unit_free_requires_mounts_for(Unit *u) {
596 assert(u);
597
598 for (;;) {
599 _cleanup_free_ char *path = NULL;
600
601 path = hashmap_steal_first_key(u->requires_mounts_for);
602 if (!path)
603 break;
604 else {
605 char s[strlen(path) + 1];
606
607 PATH_FOREACH_PREFIX_MORE(s, path) {
608 char *y;
609 Set *x;
610
611 x = hashmap_get2(u->manager->units_requiring_mounts_for, s, (void**) &y);
612 if (!x)
613 continue;
614
615 (void) set_remove(x, u);
616
617 if (set_isempty(x)) {
618 (void) hashmap_remove(u->manager->units_requiring_mounts_for, y);
619 free(y);
620 set_free(x);
621 }
622 }
623 }
624 }
625
626 u->requires_mounts_for = hashmap_free(u->requires_mounts_for);
627 }
628
629 static void unit_done(Unit *u) {
630 ExecContext *ec;
631 CGroupContext *cc;
632
633 assert(u);
634
635 if (u->type < 0)
636 return;
637
638 if (UNIT_VTABLE(u)->done)
639 UNIT_VTABLE(u)->done(u);
640
641 ec = unit_get_exec_context(u);
642 if (ec)
643 exec_context_done(ec);
644
645 cc = unit_get_cgroup_context(u);
646 if (cc)
647 cgroup_context_done(cc);
648 }
649
650 Unit* unit_free(Unit *u) {
651 Unit *slice;
652 char *t;
653
654 if (!u)
655 return NULL;
656
657 u->transient_file = safe_fclose(u->transient_file);
658
659 if (!MANAGER_IS_RELOADING(u->manager))
660 unit_remove_transient(u);
661
662 bus_unit_send_removed_signal(u);
663
664 unit_done(u);
665
666 unit_dequeue_rewatch_pids(u);
667
668 sd_bus_slot_unref(u->match_bus_slot);
669 sd_bus_track_unref(u->bus_track);
670 u->deserialized_refs = strv_free(u->deserialized_refs);
671 u->pending_freezer_message = sd_bus_message_unref(u->pending_freezer_message);
672
673 unit_free_requires_mounts_for(u);
674
675 SET_FOREACH(t, u->aliases)
676 hashmap_remove_value(u->manager->units, t, u);
677 if (u->id)
678 hashmap_remove_value(u->manager->units, u->id, u);
679
680 if (!sd_id128_is_null(u->invocation_id))
681 hashmap_remove_value(u->manager->units_by_invocation_id, &u->invocation_id, u);
682
683 if (u->job) {
684 Job *j = u->job;
685 job_uninstall(j);
686 job_free(j);
687 }
688
689 if (u->nop_job) {
690 Job *j = u->nop_job;
691 job_uninstall(j);
692 job_free(j);
693 }
694
695 /* A unit is being dropped from the tree, make sure our family is realized properly. Do this after we
696 * detach the unit from slice tree in order to eliminate its effect on controller masks. */
697 slice = UNIT_GET_SLICE(u);
698 unit_clear_dependencies(u);
699 if (slice)
700 unit_add_family_to_cgroup_realize_queue(slice);
701
702 if (u->on_console)
703 manager_unref_console(u->manager);
704
705
706 fdset_free(u->initial_socket_bind_link_fds);
707 #if BPF_FRAMEWORK
708 bpf_link_free(u->ipv4_socket_bind_link);
709 bpf_link_free(u->ipv6_socket_bind_link);
710 #endif
711
712 unit_release_cgroup(u);
713
714 if (!MANAGER_IS_RELOADING(u->manager))
715 unit_unlink_state_files(u);
716
717 unit_unref_uid_gid(u, false);
718
719 (void) manager_update_failed_units(u->manager, u, false);
720 set_remove(u->manager->startup_units, u);
721
722 unit_unwatch_all_pids(u);
723
724 while (u->refs_by_target)
725 unit_ref_unset(u->refs_by_target);
726
727 if (u->type != _UNIT_TYPE_INVALID)
728 LIST_REMOVE(units_by_type, u->manager->units_by_type[u->type], u);
729
730 if (u->in_load_queue)
731 LIST_REMOVE(load_queue, u->manager->load_queue, u);
732
733 if (u->in_dbus_queue)
734 LIST_REMOVE(dbus_queue, u->manager->dbus_unit_queue, u);
735
736 if (u->in_cleanup_queue)
737 LIST_REMOVE(cleanup_queue, u->manager->cleanup_queue, u);
738
739 if (u->in_gc_queue)
740 LIST_REMOVE(gc_queue, u->manager->gc_unit_queue, u);
741
742 if (u->in_cgroup_realize_queue)
743 LIST_REMOVE(cgroup_realize_queue, u->manager->cgroup_realize_queue, u);
744
745 if (u->in_cgroup_empty_queue)
746 LIST_REMOVE(cgroup_empty_queue, u->manager->cgroup_empty_queue, u);
747
748 if (u->in_cgroup_oom_queue)
749 LIST_REMOVE(cgroup_oom_queue, u->manager->cgroup_oom_queue, u);
750
751 if (u->in_target_deps_queue)
752 LIST_REMOVE(target_deps_queue, u->manager->target_deps_queue, u);
753
754 if (u->in_stop_when_unneeded_queue)
755 LIST_REMOVE(stop_when_unneeded_queue, u->manager->stop_when_unneeded_queue, u);
756
757 if (u->in_start_when_upheld_queue)
758 LIST_REMOVE(start_when_upheld_queue, u->manager->start_when_upheld_queue, u);
759
760 if (u->in_stop_when_bound_queue)
761 LIST_REMOVE(stop_when_bound_queue, u->manager->stop_when_bound_queue, u);
762
763 bpf_firewall_close(u);
764
765 hashmap_free(u->bpf_foreign_by_key);
766
767 bpf_program_unref(u->bpf_device_control_installed);
768
769 condition_free_list(u->conditions);
770 condition_free_list(u->asserts);
771
772 free(u->description);
773 strv_free(u->documentation);
774 free(u->fragment_path);
775 free(u->source_path);
776 strv_free(u->dropin_paths);
777 free(u->instance);
778
779 free(u->job_timeout_reboot_arg);
780 free(u->reboot_arg);
781
782 set_free_free(u->aliases);
783 free(u->id);
784
785 return mfree(u);
786 }
787
788 FreezerState unit_freezer_state(Unit *u) {
789 assert(u);
790
791 return u->freezer_state;
792 }
793
794 int unit_freezer_state_kernel(Unit *u, FreezerState *ret) {
795 char *values[1] = {};
796 int r;
797
798 assert(u);
799
800 r = cg_get_keyed_attribute(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, "cgroup.events",
801 STRV_MAKE("frozen"), values);
802 if (r < 0)
803 return r;
804
805 r = _FREEZER_STATE_INVALID;
806
807 if (values[0]) {
808 if (streq(values[0], "0"))
809 r = FREEZER_RUNNING;
810 else if (streq(values[0], "1"))
811 r = FREEZER_FROZEN;
812 }
813
814 free(values[0]);
815 *ret = r;
816
817 return 0;
818 }
819
820 UnitActiveState unit_active_state(Unit *u) {
821 assert(u);
822
823 if (u->load_state == UNIT_MERGED)
824 return unit_active_state(unit_follow_merge(u));
825
826 /* After a reload it might happen that a unit is not correctly
827 * loaded but still has a process around. That's why we won't
828 * shortcut failed loading to UNIT_INACTIVE_FAILED. */
829
830 return UNIT_VTABLE(u)->active_state(u);
831 }
832
833 const char* unit_sub_state_to_string(Unit *u) {
834 assert(u);
835
836 return UNIT_VTABLE(u)->sub_state_to_string(u);
837 }
838
839 static int unit_merge_names(Unit *u, Unit *other) {
840 char *name;
841 int r;
842
843 assert(u);
844 assert(other);
845
846 r = unit_add_alias(u, other->id);
847 if (r < 0)
848 return r;
849
850 r = set_move(u->aliases, other->aliases);
851 if (r < 0) {
852 set_remove(u->aliases, other->id);
853 return r;
854 }
855
856 TAKE_PTR(other->id);
857 other->aliases = set_free_free(other->aliases);
858
859 SET_FOREACH(name, u->aliases)
860 assert_se(hashmap_replace(u->manager->units, name, u) == 0);
861
862 return 0;
863 }
864
865 static int unit_reserve_dependencies(Unit *u, Unit *other) {
866 size_t n_reserve;
867 Hashmap* deps;
868 void *d;
869 int r;
870
871 assert(u);
872 assert(other);
873
874 /* Let's reserve some space in the dependency hashmaps so that later on merging the units cannot
875 * fail.
876 *
877 * First make some room in the per dependency type hashmaps. Using the summed size of both unit's
878 * hashmaps is an estimate that is likely too high since they probably use some of the same
879 * types. But it's never too low, and that's all we need. */
880
881 n_reserve = MIN(hashmap_size(other->dependencies), LESS_BY((size_t) _UNIT_DEPENDENCY_MAX, hashmap_size(u->dependencies)));
882 if (n_reserve > 0) {
883 r = hashmap_ensure_allocated(&u->dependencies, NULL);
884 if (r < 0)
885 return r;
886
887 r = hashmap_reserve(u->dependencies, n_reserve);
888 if (r < 0)
889 return r;
890 }
891
892 /* Now, enlarge our per dependency type hashmaps by the number of entries in the same hashmap of the
893 * other unit's dependencies.
894 *
895 * NB: If u does not have a dependency set allocated for some dependency type, there is no need to
896 * reserve anything for. In that case other's set will be transferred as a whole to u by
897 * complete_move(). */
898
899 HASHMAP_FOREACH_KEY(deps, d, u->dependencies) {
900 Hashmap *other_deps;
901
902 other_deps = hashmap_get(other->dependencies, d);
903
904 r = hashmap_reserve(deps, hashmap_size(other_deps));
905 if (r < 0)
906 return r;
907 }
908
909 return 0;
910 }
911
912 static void unit_maybe_warn_about_dependency(
913 Unit *u,
914 const char *other_id,
915 UnitDependency dependency) {
916
917 assert(u);
918
919 /* Only warn about some unit types */
920 if (!IN_SET(dependency,
921 UNIT_CONFLICTS,
922 UNIT_CONFLICTED_BY,
923 UNIT_BEFORE,
924 UNIT_AFTER,
925 UNIT_ON_SUCCESS,
926 UNIT_ON_FAILURE,
927 UNIT_TRIGGERS,
928 UNIT_TRIGGERED_BY))
929 return;
930
931 if (streq_ptr(u->id, other_id))
932 log_unit_warning(u, "Dependency %s=%s dropped", unit_dependency_to_string(dependency), u->id);
933 else
934 log_unit_warning(u, "Dependency %s=%s dropped, merged into %s", unit_dependency_to_string(dependency), strna(other_id), u->id);
935 }
936
937 static int unit_per_dependency_type_hashmap_update(
938 Hashmap *per_type,
939 Unit *other,
940 UnitDependencyMask origin_mask,
941 UnitDependencyMask destination_mask) {
942
943 UnitDependencyInfo info;
944 int r;
945
946 assert(other);
947 assert_cc(sizeof(void*) == sizeof(info));
948
949 /* Acquire the UnitDependencyInfo entry for the Unit* we are interested in, and update it if it
950 * exists, or insert it anew if not. */
951
952 info.data = hashmap_get(per_type, other);
953 if (info.data) {
954 /* Entry already exists. Add in our mask. */
955
956 if (FLAGS_SET(origin_mask, info.origin_mask) &&
957 FLAGS_SET(destination_mask, info.destination_mask))
958 return 0; /* NOP */
959
960 info.origin_mask |= origin_mask;
961 info.destination_mask |= destination_mask;
962
963 r = hashmap_update(per_type, other, info.data);
964 } else {
965 info = (UnitDependencyInfo) {
966 .origin_mask = origin_mask,
967 .destination_mask = destination_mask,
968 };
969
970 r = hashmap_put(per_type, other, info.data);
971 }
972 if (r < 0)
973 return r;
974
975
976 return 1;
977 }
978
979 static int unit_add_dependency_hashmap(
980 Hashmap **dependencies,
981 UnitDependency d,
982 Unit *other,
983 UnitDependencyMask origin_mask,
984 UnitDependencyMask destination_mask) {
985
986 Hashmap *per_type;
987 int r;
988
989 assert(dependencies);
990 assert(other);
991 assert(origin_mask < _UNIT_DEPENDENCY_MASK_FULL);
992 assert(destination_mask < _UNIT_DEPENDENCY_MASK_FULL);
993 assert(origin_mask > 0 || destination_mask > 0);
994
995 /* Ensure the top-level dependency hashmap exists that maps UnitDependency → Hashmap(Unit* →
996 * UnitDependencyInfo) */
997 r = hashmap_ensure_allocated(dependencies, NULL);
998 if (r < 0)
999 return r;
1000
1001 /* Acquire the inner hashmap, that maps Unit* → UnitDependencyInfo, for the specified dependency
1002 * type, and if it's missing allocate it and insert it. */
1003 per_type = hashmap_get(*dependencies, UNIT_DEPENDENCY_TO_PTR(d));
1004 if (!per_type) {
1005 per_type = hashmap_new(NULL);
1006 if (!per_type)
1007 return -ENOMEM;
1008
1009 r = hashmap_put(*dependencies, UNIT_DEPENDENCY_TO_PTR(d), per_type);
1010 if (r < 0) {
1011 hashmap_free(per_type);
1012 return r;
1013 }
1014 }
1015
1016 return unit_per_dependency_type_hashmap_update(per_type, other, origin_mask, destination_mask);
1017 }
1018
1019 static void unit_merge_dependencies(
1020 Unit *u,
1021 Unit *other) {
1022
1023 int r;
1024
1025 assert(u);
1026 assert(other);
1027
1028 if (u == other)
1029 return;
1030
1031 for (;;) {
1032 _cleanup_(hashmap_freep) Hashmap *other_deps = NULL;
1033 UnitDependencyInfo di_back;
1034 Unit *back;
1035 void *dt; /* Actually of type UnitDependency, except that we don't bother casting it here,
1036 * since the hashmaps all want it as void pointer. */
1037
1038 /* Let's focus on one dependency type at a time, that 'other' has defined. */
1039 other_deps = hashmap_steal_first_key_and_value(other->dependencies, &dt);
1040 if (!other_deps)
1041 break; /* done! */
1042
1043 /* Now iterate through all dependencies of this dependency type, of 'other'. We refer to the
1044 * referenced units as 'back'. */
1045 HASHMAP_FOREACH_KEY(di_back.data, back, other_deps) {
1046 Hashmap *back_deps;
1047 void *back_dt;
1048
1049 if (back == u) {
1050 /* This is a dependency pointing back to the unit we want to merge with?
1051 * Suppress it (but warn) */
1052 unit_maybe_warn_about_dependency(u, other->id, UNIT_DEPENDENCY_FROM_PTR(dt));
1053 continue;
1054 }
1055
1056 /* Now iterate through all deps of 'back', and fix the ones pointing to 'other' to
1057 * point to 'u' instead. */
1058 HASHMAP_FOREACH_KEY(back_deps, back_dt, back->dependencies) {
1059 UnitDependencyInfo di_move;
1060
1061 di_move.data = hashmap_remove(back_deps, other);
1062 if (!di_move.data)
1063 continue;
1064
1065 assert_se(unit_per_dependency_type_hashmap_update(
1066 back_deps,
1067 u,
1068 di_move.origin_mask,
1069 di_move.destination_mask) >= 0);
1070 }
1071 }
1072
1073 /* Now all references towards 'other' of the current type 'dt' are corrected to point to
1074 * 'u'. Lets's now move the deps of type 'dt' from 'other' to 'u'. First, let's try to move
1075 * them per type wholesale. */
1076 r = hashmap_put(u->dependencies, dt, other_deps);
1077 if (r == -EEXIST) {
1078 Hashmap *deps;
1079
1080 /* The target unit already has dependencies of this type, let's then merge this individually. */
1081
1082 assert_se(deps = hashmap_get(u->dependencies, dt));
1083
1084 for (;;) {
1085 UnitDependencyInfo di_move;
1086
1087 /* Get first dep */
1088 di_move.data = hashmap_steal_first_key_and_value(other_deps, (void**) &back);
1089 if (!di_move.data)
1090 break; /* done */
1091 if (back == u) {
1092 /* Would point back to us, ignore */
1093 unit_maybe_warn_about_dependency(u, other->id, UNIT_DEPENDENCY_FROM_PTR(dt));
1094 continue;
1095 }
1096
1097 assert_se(unit_per_dependency_type_hashmap_update(deps, back, di_move.origin_mask, di_move.destination_mask) >= 0);
1098 }
1099 } else {
1100 assert_se(r >= 0);
1101 TAKE_PTR(other_deps);
1102
1103 if (hashmap_remove(other_deps, u))
1104 unit_maybe_warn_about_dependency(u, other->id, UNIT_DEPENDENCY_FROM_PTR(dt));
1105 }
1106 }
1107
1108 other->dependencies = hashmap_free(other->dependencies);
1109 }
1110
1111 int unit_merge(Unit *u, Unit *other) {
1112 int r;
1113
1114 assert(u);
1115 assert(other);
1116 assert(u->manager == other->manager);
1117 assert(u->type != _UNIT_TYPE_INVALID);
1118
1119 other = unit_follow_merge(other);
1120
1121 if (other == u)
1122 return 0;
1123
1124 if (u->type != other->type)
1125 return -EINVAL;
1126
1127 if (!unit_type_may_alias(u->type)) /* Merging only applies to unit names that support aliases */
1128 return -EEXIST;
1129
1130 if (!IN_SET(other->load_state, UNIT_STUB, UNIT_NOT_FOUND))
1131 return -EEXIST;
1132
1133 if (!streq_ptr(u->instance, other->instance))
1134 return -EINVAL;
1135
1136 if (other->job)
1137 return -EEXIST;
1138
1139 if (other->nop_job)
1140 return -EEXIST;
1141
1142 if (!UNIT_IS_INACTIVE_OR_FAILED(unit_active_state(other)))
1143 return -EEXIST;
1144
1145 /* Make reservations to ensure merge_dependencies() won't fail. We don't rollback reservations if we
1146 * fail. We don't have a way to undo reservations. A reservation is not a leak. */
1147 r = unit_reserve_dependencies(u, other);
1148 if (r < 0)
1149 return r;
1150
1151 /* Merge names */
1152 r = unit_merge_names(u, other);
1153 if (r < 0)
1154 return r;
1155
1156 /* Redirect all references */
1157 while (other->refs_by_target)
1158 unit_ref_set(other->refs_by_target, other->refs_by_target->source, u);
1159
1160 /* Merge dependencies */
1161 unit_merge_dependencies(u, other);
1162
1163 other->load_state = UNIT_MERGED;
1164 other->merged_into = u;
1165
1166 /* If there is still some data attached to the other node, we
1167 * don't need it anymore, and can free it. */
1168 if (other->load_state != UNIT_STUB)
1169 if (UNIT_VTABLE(other)->done)
1170 UNIT_VTABLE(other)->done(other);
1171
1172 unit_add_to_dbus_queue(u);
1173 unit_add_to_cleanup_queue(other);
1174
1175 return 0;
1176 }
1177
1178 int unit_merge_by_name(Unit *u, const char *name) {
1179 _cleanup_free_ char *s = NULL;
1180 Unit *other;
1181 int r;
1182
1183 /* Either add name to u, or if a unit with name already exists, merge it with u.
1184 * If name is a template, do the same for name@instance, where instance is u's instance. */
1185
1186 assert(u);
1187 assert(name);
1188
1189 if (unit_name_is_valid(name, UNIT_NAME_TEMPLATE)) {
1190 if (!u->instance)
1191 return -EINVAL;
1192
1193 r = unit_name_replace_instance(name, u->instance, &s);
1194 if (r < 0)
1195 return r;
1196
1197 name = s;
1198 }
1199
1200 other = manager_get_unit(u->manager, name);
1201 if (other)
1202 return unit_merge(u, other);
1203
1204 return unit_add_name(u, name);
1205 }
1206
1207 Unit* unit_follow_merge(Unit *u) {
1208 assert(u);
1209
1210 while (u->load_state == UNIT_MERGED)
1211 assert_se(u = u->merged_into);
1212
1213 return u;
1214 }
1215
1216 int unit_add_exec_dependencies(Unit *u, ExecContext *c) {
1217 int r;
1218
1219 assert(u);
1220 assert(c);
1221
1222 if (c->working_directory && !c->working_directory_missing_ok) {
1223 r = unit_require_mounts_for(u, c->working_directory, UNIT_DEPENDENCY_FILE);
1224 if (r < 0)
1225 return r;
1226 }
1227
1228 if (c->root_directory) {
1229 r = unit_require_mounts_for(u, c->root_directory, UNIT_DEPENDENCY_FILE);
1230 if (r < 0)
1231 return r;
1232 }
1233
1234 if (c->root_image) {
1235 r = unit_require_mounts_for(u, c->root_image, UNIT_DEPENDENCY_FILE);
1236 if (r < 0)
1237 return r;
1238 }
1239
1240 for (ExecDirectoryType dt = 0; dt < _EXEC_DIRECTORY_TYPE_MAX; dt++) {
1241 if (!u->manager->prefix[dt])
1242 continue;
1243
1244 char **dp;
1245 STRV_FOREACH(dp, c->directories[dt].paths) {
1246 _cleanup_free_ char *p = NULL;
1247
1248 p = path_join(u->manager->prefix[dt], *dp);
1249 if (!p)
1250 return -ENOMEM;
1251
1252 r = unit_require_mounts_for(u, p, UNIT_DEPENDENCY_FILE);
1253 if (r < 0)
1254 return r;
1255 }
1256 }
1257
1258 if (!MANAGER_IS_SYSTEM(u->manager))
1259 return 0;
1260
1261 /* For the following three directory types we need write access, and /var/ is possibly on the root
1262 * fs. Hence order after systemd-remount-fs.service, to ensure things are writable. */
1263 if (!strv_isempty(c->directories[EXEC_DIRECTORY_STATE].paths) ||
1264 !strv_isempty(c->directories[EXEC_DIRECTORY_CACHE].paths) ||
1265 !strv_isempty(c->directories[EXEC_DIRECTORY_LOGS].paths)) {
1266 r = unit_add_dependency_by_name(u, UNIT_AFTER, SPECIAL_REMOUNT_FS_SERVICE, true, UNIT_DEPENDENCY_FILE);
1267 if (r < 0)
1268 return r;
1269 }
1270
1271 if (c->private_tmp) {
1272
1273 /* FIXME: for now we make a special case for /tmp and add a weak dependency on
1274 * tmp.mount so /tmp being masked is supported. However there's no reason to treat
1275 * /tmp specifically and masking other mount units should be handled more
1276 * gracefully too, see PR#16894. */
1277 r = unit_add_two_dependencies_by_name(u, UNIT_AFTER, UNIT_WANTS, "tmp.mount", true, UNIT_DEPENDENCY_FILE);
1278 if (r < 0)
1279 return r;
1280
1281 r = unit_require_mounts_for(u, "/var/tmp", UNIT_DEPENDENCY_FILE);
1282 if (r < 0)
1283 return r;
1284
1285 r = unit_add_dependency_by_name(u, UNIT_AFTER, SPECIAL_TMPFILES_SETUP_SERVICE, true, UNIT_DEPENDENCY_FILE);
1286 if (r < 0)
1287 return r;
1288 }
1289
1290 if (c->root_image) {
1291 /* We need to wait for /dev/loopX to appear when doing RootImage=, hence let's add an
1292 * implicit dependency on udev */
1293
1294 r = unit_add_dependency_by_name(u, UNIT_AFTER, SPECIAL_UDEVD_SERVICE, true, UNIT_DEPENDENCY_FILE);
1295 if (r < 0)
1296 return r;
1297 }
1298
1299 if (!IN_SET(c->std_output,
1300 EXEC_OUTPUT_JOURNAL, EXEC_OUTPUT_JOURNAL_AND_CONSOLE,
1301 EXEC_OUTPUT_KMSG, EXEC_OUTPUT_KMSG_AND_CONSOLE) &&
1302 !IN_SET(c->std_error,
1303 EXEC_OUTPUT_JOURNAL, EXEC_OUTPUT_JOURNAL_AND_CONSOLE,
1304 EXEC_OUTPUT_KMSG, EXEC_OUTPUT_KMSG_AND_CONSOLE) &&
1305 !c->log_namespace)
1306 return 0;
1307
1308 /* If syslog or kernel logging is requested (or log namespacing is), make sure our own logging daemon
1309 * is run first. */
1310
1311 if (c->log_namespace) {
1312 _cleanup_free_ char *socket_unit = NULL, *varlink_socket_unit = NULL;
1313
1314 r = unit_name_build_from_type("systemd-journald", c->log_namespace, UNIT_SOCKET, &socket_unit);
1315 if (r < 0)
1316 return r;
1317
1318 r = unit_add_two_dependencies_by_name(u, UNIT_AFTER, UNIT_REQUIRES, socket_unit, true, UNIT_DEPENDENCY_FILE);
1319 if (r < 0)
1320 return r;
1321
1322 r = unit_name_build_from_type("systemd-journald-varlink", c->log_namespace, UNIT_SOCKET, &varlink_socket_unit);
1323 if (r < 0)
1324 return r;
1325
1326 r = unit_add_two_dependencies_by_name(u, UNIT_AFTER, UNIT_REQUIRES, varlink_socket_unit, true, UNIT_DEPENDENCY_FILE);
1327 if (r < 0)
1328 return r;
1329 } else
1330 r = unit_add_dependency_by_name(u, UNIT_AFTER, SPECIAL_JOURNALD_SOCKET, true, UNIT_DEPENDENCY_FILE);
1331 if (r < 0)
1332 return r;
1333
1334 return 0;
1335 }
1336
1337 const char* unit_description(Unit *u) {
1338 assert(u);
1339
1340 if (u->description)
1341 return u->description;
1342
1343 return strna(u->id);
1344 }
1345
1346 const char* unit_status_string(Unit *u, char **ret_combined_buffer) {
1347 assert(u);
1348 assert(u->id);
1349
1350 /* Return u->id, u->description, or "{u->id} - {u->description}".
1351 * Versions with u->description are only used if it is set.
1352 * The last option is used if configured and the caller provided the 'ret_combined_buffer'
1353 * pointer.
1354 *
1355 * Note that *ret_combined_buffer may be set to NULL. */
1356
1357 if (!u->description ||
1358 u->manager->status_unit_format == STATUS_UNIT_FORMAT_NAME ||
1359 (u->manager->status_unit_format == STATUS_UNIT_FORMAT_COMBINED && !ret_combined_buffer) ||
1360 streq(u->description, u->id)) {
1361
1362 if (ret_combined_buffer)
1363 *ret_combined_buffer = NULL;
1364 return u->id;
1365 }
1366
1367 if (ret_combined_buffer) {
1368 if (u->manager->status_unit_format == STATUS_UNIT_FORMAT_COMBINED) {
1369 *ret_combined_buffer = strjoin(u->id, " - ", u->description);
1370 if (*ret_combined_buffer)
1371 return *ret_combined_buffer;
1372 log_oom(); /* Fall back to ->description */
1373 } else
1374 *ret_combined_buffer = NULL;
1375 }
1376
1377 return u->description;
1378 }
1379
1380 /* Common implementation for multiple backends */
1381 int unit_load_fragment_and_dropin(Unit *u, bool fragment_required) {
1382 int r;
1383
1384 assert(u);
1385
1386 /* Load a .{service,socket,...} file */
1387 r = unit_load_fragment(u);
1388 if (r < 0)
1389 return r;
1390
1391 if (u->load_state == UNIT_STUB) {
1392 if (fragment_required)
1393 return -ENOENT;
1394
1395 u->load_state = UNIT_LOADED;
1396 }
1397
1398 /* Load drop-in directory data. If u is an alias, we might be reloading the
1399 * target unit needlessly. But we cannot be sure which drops-ins have already
1400 * been loaded and which not, at least without doing complicated book-keeping,
1401 * so let's always reread all drop-ins. */
1402 r = unit_load_dropin(unit_follow_merge(u));
1403 if (r < 0)
1404 return r;
1405
1406 if (u->source_path) {
1407 struct stat st;
1408
1409 if (stat(u->source_path, &st) >= 0)
1410 u->source_mtime = timespec_load(&st.st_mtim);
1411 else
1412 u->source_mtime = 0;
1413 }
1414
1415 return 0;
1416 }
1417
1418 void unit_add_to_target_deps_queue(Unit *u) {
1419 Manager *m = u->manager;
1420
1421 assert(u);
1422
1423 if (u->in_target_deps_queue)
1424 return;
1425
1426 LIST_PREPEND(target_deps_queue, m->target_deps_queue, u);
1427 u->in_target_deps_queue = true;
1428 }
1429
1430 int unit_add_default_target_dependency(Unit *u, Unit *target) {
1431 assert(u);
1432 assert(target);
1433
1434 if (target->type != UNIT_TARGET)
1435 return 0;
1436
1437 /* Only add the dependency if both units are loaded, so that
1438 * that loop check below is reliable */
1439 if (u->load_state != UNIT_LOADED ||
1440 target->load_state != UNIT_LOADED)
1441 return 0;
1442
1443 /* If either side wants no automatic dependencies, then let's
1444 * skip this */
1445 if (!u->default_dependencies ||
1446 !target->default_dependencies)
1447 return 0;
1448
1449 /* Don't create loops */
1450 if (unit_has_dependency(target, UNIT_ATOM_BEFORE, u))
1451 return 0;
1452
1453 return unit_add_dependency(target, UNIT_AFTER, u, true, UNIT_DEPENDENCY_DEFAULT);
1454 }
1455
1456 static int unit_add_slice_dependencies(Unit *u) {
1457 Unit *slice;
1458 assert(u);
1459
1460 if (!UNIT_HAS_CGROUP_CONTEXT(u))
1461 return 0;
1462
1463 /* Slice units are implicitly ordered against their parent slices (as this relationship is encoded in the
1464 name), while all other units are ordered based on configuration (as in their case Slice= configures the
1465 relationship). */
1466 UnitDependencyMask mask = u->type == UNIT_SLICE ? UNIT_DEPENDENCY_IMPLICIT : UNIT_DEPENDENCY_FILE;
1467
1468 slice = UNIT_GET_SLICE(u);
1469 if (slice)
1470 return unit_add_two_dependencies(u, UNIT_AFTER, UNIT_REQUIRES, slice, true, mask);
1471
1472 if (unit_has_name(u, SPECIAL_ROOT_SLICE))
1473 return 0;
1474
1475 return unit_add_two_dependencies_by_name(u, UNIT_AFTER, UNIT_REQUIRES, SPECIAL_ROOT_SLICE, true, mask);
1476 }
1477
1478 static int unit_add_mount_dependencies(Unit *u) {
1479 UnitDependencyInfo di;
1480 const char *path;
1481 int r;
1482
1483 assert(u);
1484
1485 HASHMAP_FOREACH_KEY(di.data, path, u->requires_mounts_for) {
1486 char prefix[strlen(path) + 1];
1487
1488 PATH_FOREACH_PREFIX_MORE(prefix, path) {
1489 _cleanup_free_ char *p = NULL;
1490 Unit *m;
1491
1492 r = unit_name_from_path(prefix, ".mount", &p);
1493 if (IN_SET(r, -EINVAL, -ENAMETOOLONG))
1494 continue; /* If the path cannot be converted to a mount unit name, then it's
1495 * not manageable as a unit by systemd, and hence we don't need a
1496 * dependency on it. Let's thus silently ignore the issue. */
1497 if (r < 0)
1498 return r;
1499
1500 m = manager_get_unit(u->manager, p);
1501 if (!m) {
1502 /* Make sure to load the mount unit if it exists. If so the dependencies on
1503 * this unit will be added later during the loading of the mount unit. */
1504 (void) manager_load_unit_prepare(u->manager, p, NULL, NULL, &m);
1505 continue;
1506 }
1507 if (m == u)
1508 continue;
1509
1510 if (m->load_state != UNIT_LOADED)
1511 continue;
1512
1513 r = unit_add_dependency(u, UNIT_AFTER, m, true, di.origin_mask);
1514 if (r < 0)
1515 return r;
1516
1517 if (m->fragment_path) {
1518 r = unit_add_dependency(u, UNIT_REQUIRES, m, true, di.origin_mask);
1519 if (r < 0)
1520 return r;
1521 }
1522 }
1523 }
1524
1525 return 0;
1526 }
1527
1528 static int unit_add_oomd_dependencies(Unit *u) {
1529 CGroupContext *c;
1530 bool wants_oomd;
1531 int r;
1532
1533 assert(u);
1534
1535 if (!u->default_dependencies)
1536 return 0;
1537
1538 c = unit_get_cgroup_context(u);
1539 if (!c)
1540 return 0;
1541
1542 wants_oomd = (c->moom_swap == MANAGED_OOM_KILL || c->moom_mem_pressure == MANAGED_OOM_KILL);
1543 if (!wants_oomd)
1544 return 0;
1545
1546 r = unit_add_two_dependencies_by_name(u, UNIT_AFTER, UNIT_WANTS, "systemd-oomd.service", true, UNIT_DEPENDENCY_FILE);
1547 if (r < 0)
1548 return r;
1549
1550 return 0;
1551 }
1552
1553 static int unit_add_startup_units(Unit *u) {
1554 CGroupContext *c;
1555
1556 c = unit_get_cgroup_context(u);
1557 if (!c)
1558 return 0;
1559
1560 if (c->startup_cpu_shares == CGROUP_CPU_SHARES_INVALID &&
1561 c->startup_io_weight == CGROUP_WEIGHT_INVALID &&
1562 c->startup_blockio_weight == CGROUP_BLKIO_WEIGHT_INVALID)
1563 return 0;
1564
1565 return set_ensure_put(&u->manager->startup_units, NULL, u);
1566 }
1567
1568 static int unit_validate_on_failure_job_mode(
1569 Unit *u,
1570 const char *job_mode_setting,
1571 JobMode job_mode,
1572 const char *dependency_name,
1573 UnitDependencyAtom atom) {
1574
1575 Unit *other, *found = NULL;
1576
1577 if (job_mode != JOB_ISOLATE)
1578 return 0;
1579
1580 UNIT_FOREACH_DEPENDENCY(other, u, atom) {
1581 if (!found)
1582 found = other;
1583 else if (found != other)
1584 return log_unit_error_errno(
1585 u, SYNTHETIC_ERRNO(ENOEXEC),
1586 "More than one %s dependencies specified but %sisolate set. Refusing.",
1587 dependency_name, job_mode_setting);
1588 }
1589
1590 return 0;
1591 }
1592
1593 int unit_load(Unit *u) {
1594 int r;
1595
1596 assert(u);
1597
1598 if (u->in_load_queue) {
1599 LIST_REMOVE(load_queue, u->manager->load_queue, u);
1600 u->in_load_queue = false;
1601 }
1602
1603 if (u->type == _UNIT_TYPE_INVALID)
1604 return -EINVAL;
1605
1606 if (u->load_state != UNIT_STUB)
1607 return 0;
1608
1609 if (u->transient_file) {
1610 /* Finalize transient file: if this is a transient unit file, as soon as we reach unit_load() the setup
1611 * is complete, hence let's synchronize the unit file we just wrote to disk. */
1612
1613 r = fflush_and_check(u->transient_file);
1614 if (r < 0)
1615 goto fail;
1616
1617 u->transient_file = safe_fclose(u->transient_file);
1618 u->fragment_mtime = now(CLOCK_REALTIME);
1619 }
1620
1621 r = UNIT_VTABLE(u)->load(u);
1622 if (r < 0)
1623 goto fail;
1624
1625 assert(u->load_state != UNIT_STUB);
1626
1627 if (u->load_state == UNIT_LOADED) {
1628 unit_add_to_target_deps_queue(u);
1629
1630 r = unit_add_slice_dependencies(u);
1631 if (r < 0)
1632 goto fail;
1633
1634 r = unit_add_mount_dependencies(u);
1635 if (r < 0)
1636 goto fail;
1637
1638 r = unit_add_oomd_dependencies(u);
1639 if (r < 0)
1640 goto fail;
1641
1642 r = unit_add_startup_units(u);
1643 if (r < 0)
1644 goto fail;
1645
1646 r = unit_validate_on_failure_job_mode(u, "OnSuccessJobMode=", u->on_success_job_mode, "OnSuccess=", UNIT_ATOM_ON_SUCCESS);
1647 if (r < 0)
1648 goto fail;
1649
1650 r = unit_validate_on_failure_job_mode(u, "OnFailureJobMode=", u->on_failure_job_mode, "OnFailure=", UNIT_ATOM_ON_FAILURE);
1651 if (r < 0)
1652 goto fail;
1653
1654 if (u->job_running_timeout != USEC_INFINITY && u->job_running_timeout > u->job_timeout)
1655 log_unit_warning(u, "JobRunningTimeoutSec= is greater than JobTimeoutSec=, it has no effect.");
1656
1657 /* We finished loading, let's ensure our parents recalculate the members mask */
1658 unit_invalidate_cgroup_members_masks(u);
1659 }
1660
1661 assert((u->load_state != UNIT_MERGED) == !u->merged_into);
1662
1663 unit_add_to_dbus_queue(unit_follow_merge(u));
1664 unit_add_to_gc_queue(u);
1665 (void) manager_varlink_send_managed_oom_update(u);
1666
1667 return 0;
1668
1669 fail:
1670 /* We convert ENOEXEC errors to the UNIT_BAD_SETTING load state here. Configuration parsing code
1671 * should hence return ENOEXEC to ensure units are placed in this state after loading. */
1672
1673 u->load_state = u->load_state == UNIT_STUB ? UNIT_NOT_FOUND :
1674 r == -ENOEXEC ? UNIT_BAD_SETTING :
1675 UNIT_ERROR;
1676 u->load_error = r;
1677
1678 /* Record the timestamp on the cache, so that if the cache gets updated between now and the next time
1679 * an attempt is made to load this unit, we know we need to check again. */
1680 if (u->load_state == UNIT_NOT_FOUND)
1681 u->fragment_not_found_timestamp_hash = u->manager->unit_cache_timestamp_hash;
1682
1683 unit_add_to_dbus_queue(u);
1684 unit_add_to_gc_queue(u);
1685
1686 return log_unit_debug_errno(u, r, "Failed to load configuration: %m");
1687 }
1688
1689 _printf_(7, 8)
1690 static int log_unit_internal(void *userdata, int level, int error, const char *file, int line, const char *func, const char *format, ...) {
1691 Unit *u = userdata;
1692 va_list ap;
1693 int r;
1694
1695 if (u && !unit_log_level_test(u, level))
1696 return -ERRNO_VALUE(error);
1697
1698 va_start(ap, format);
1699 if (u)
1700 r = log_object_internalv(level, error, file, line, func,
1701 u->manager->unit_log_field,
1702 u->id,
1703 u->manager->invocation_log_field,
1704 u->invocation_id_string,
1705 format, ap);
1706 else
1707 r = log_internalv(level, error, file, line, func, format, ap);
1708 va_end(ap);
1709
1710 return r;
1711 }
1712
1713 static bool unit_test_condition(Unit *u) {
1714 _cleanup_strv_free_ char **env = NULL;
1715 int r;
1716
1717 assert(u);
1718
1719 dual_timestamp_get(&u->condition_timestamp);
1720
1721 r = manager_get_effective_environment(u->manager, &env);
1722 if (r < 0) {
1723 log_unit_error_errno(u, r, "Failed to determine effective environment: %m");
1724 u->condition_result = CONDITION_ERROR;
1725 } else
1726 u->condition_result = condition_test_list(
1727 u->conditions,
1728 env,
1729 condition_type_to_string,
1730 log_unit_internal,
1731 u);
1732
1733 unit_add_to_dbus_queue(u);
1734 return u->condition_result;
1735 }
1736
1737 static bool unit_test_assert(Unit *u) {
1738 _cleanup_strv_free_ char **env = NULL;
1739 int r;
1740
1741 assert(u);
1742
1743 dual_timestamp_get(&u->assert_timestamp);
1744
1745 r = manager_get_effective_environment(u->manager, &env);
1746 if (r < 0) {
1747 log_unit_error_errno(u, r, "Failed to determine effective environment: %m");
1748 u->assert_result = CONDITION_ERROR;
1749 } else
1750 u->assert_result = condition_test_list(
1751 u->asserts,
1752 env,
1753 assert_type_to_string,
1754 log_unit_internal,
1755 u);
1756
1757 unit_add_to_dbus_queue(u);
1758 return u->assert_result;
1759 }
1760
1761 void unit_status_printf(Unit *u, StatusType status_type, const char *status, const char *format, const char *ident) {
1762 if (log_get_show_color()) {
1763 if (u->manager->status_unit_format == STATUS_UNIT_FORMAT_COMBINED && strchr(ident, ' '))
1764 ident = strjoina(ANSI_HIGHLIGHT, u->id, ANSI_NORMAL, " - ", u->description);
1765 else
1766 ident = strjoina(ANSI_HIGHLIGHT, ident, ANSI_NORMAL);
1767 }
1768
1769 DISABLE_WARNING_FORMAT_NONLITERAL;
1770 manager_status_printf(u->manager, status_type, status, format, ident);
1771 REENABLE_WARNING;
1772 }
1773
1774 int unit_test_start_limit(Unit *u) {
1775 const char *reason;
1776
1777 assert(u);
1778
1779 if (ratelimit_below(&u->start_ratelimit)) {
1780 u->start_limit_hit = false;
1781 return 0;
1782 }
1783
1784 log_unit_warning(u, "Start request repeated too quickly.");
1785 u->start_limit_hit = true;
1786
1787 reason = strjoina("unit ", u->id, " failed");
1788
1789 emergency_action(u->manager, u->start_limit_action,
1790 EMERGENCY_ACTION_IS_WATCHDOG|EMERGENCY_ACTION_WARN,
1791 u->reboot_arg, -1, reason);
1792
1793 return -ECANCELED;
1794 }
1795
1796 bool unit_shall_confirm_spawn(Unit *u) {
1797 assert(u);
1798
1799 if (manager_is_confirm_spawn_disabled(u->manager))
1800 return false;
1801
1802 /* For some reasons units remaining in the same process group
1803 * as PID 1 fail to acquire the console even if it's not used
1804 * by any process. So skip the confirmation question for them. */
1805 return !unit_get_exec_context(u)->same_pgrp;
1806 }
1807
1808 static bool unit_verify_deps(Unit *u) {
1809 Unit *other;
1810
1811 assert(u);
1812
1813 /* Checks whether all BindsTo= dependencies of this unit are fulfilled — if they are also combined
1814 * with After=. We do not check Requires= or Requisite= here as they only should have an effect on
1815 * the job processing, but do not have any effect afterwards. We don't check BindsTo= dependencies
1816 * that are not used in conjunction with After= as for them any such check would make things entirely
1817 * racy. */
1818
1819 UNIT_FOREACH_DEPENDENCY(other, u, UNIT_ATOM_CANNOT_BE_ACTIVE_WITHOUT) {
1820
1821 if (!unit_has_dependency(u, UNIT_ATOM_AFTER, other))
1822 continue;
1823
1824 if (!UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(other))) {
1825 log_unit_notice(u, "Bound to unit %s, but unit isn't active.", other->id);
1826 return false;
1827 }
1828 }
1829
1830 return true;
1831 }
1832
1833 /* Errors that aren't really errors:
1834 * -EALREADY: Unit is already started.
1835 * -ECOMM: Condition failed
1836 * -EAGAIN: An operation is already in progress. Retry later.
1837 *
1838 * Errors that are real errors:
1839 * -EBADR: This unit type does not support starting.
1840 * -ECANCELED: Start limit hit, too many requests for now
1841 * -EPROTO: Assert failed
1842 * -EINVAL: Unit not loaded
1843 * -EOPNOTSUPP: Unit type not supported
1844 * -ENOLINK: The necessary dependencies are not fulfilled.
1845 * -ESTALE: This unit has been started before and can't be started a second time
1846 * -ENOENT: This is a triggering unit and unit to trigger is not loaded
1847 */
1848 int unit_start(Unit *u) {
1849 UnitActiveState state;
1850 Unit *following;
1851
1852 assert(u);
1853
1854 /* If this is already started, then this will succeed. Note that this will even succeed if this unit
1855 * is not startable by the user. This is relied on to detect when we need to wait for units and when
1856 * waiting is finished. */
1857 state = unit_active_state(u);
1858 if (UNIT_IS_ACTIVE_OR_RELOADING(state))
1859 return -EALREADY;
1860 if (state == UNIT_MAINTENANCE)
1861 return -EAGAIN;
1862
1863 /* Units that aren't loaded cannot be started */
1864 if (u->load_state != UNIT_LOADED)
1865 return -EINVAL;
1866
1867 /* Refuse starting scope units more than once */
1868 if (UNIT_VTABLE(u)->once_only && dual_timestamp_is_set(&u->inactive_enter_timestamp))
1869 return -ESTALE;
1870
1871 /* If the conditions failed, don't do anything at all. If we already are activating this call might
1872 * still be useful to speed up activation in case there is some hold-off time, but we don't want to
1873 * recheck the condition in that case. */
1874 if (state != UNIT_ACTIVATING &&
1875 !unit_test_condition(u))
1876 return log_unit_debug_errno(u, SYNTHETIC_ERRNO(ECOMM), "Starting requested but condition failed. Not starting unit.");
1877
1878 /* If the asserts failed, fail the entire job */
1879 if (state != UNIT_ACTIVATING &&
1880 !unit_test_assert(u))
1881 return log_unit_notice_errno(u, SYNTHETIC_ERRNO(EPROTO), "Starting requested but asserts failed.");
1882
1883 /* Units of types that aren't supported cannot be started. Note that we do this test only after the
1884 * condition checks, so that we rather return condition check errors (which are usually not
1885 * considered a true failure) than "not supported" errors (which are considered a failure).
1886 */
1887 if (!unit_type_supported(u->type))
1888 return -EOPNOTSUPP;
1889
1890 /* Let's make sure that the deps really are in order before we start this. Normally the job engine
1891 * should have taken care of this already, but let's check this here again. After all, our
1892 * dependencies might not be in effect anymore, due to a reload or due to a failed condition. */
1893 if (!unit_verify_deps(u))
1894 return -ENOLINK;
1895
1896 /* Forward to the main object, if we aren't it. */
1897 following = unit_following(u);
1898 if (following) {
1899 log_unit_debug(u, "Redirecting start request from %s to %s.", u->id, following->id);
1900 return unit_start(following);
1901 }
1902
1903 /* If it is stopped, but we cannot start it, then fail */
1904 if (!UNIT_VTABLE(u)->start)
1905 return -EBADR;
1906
1907 /* We don't suppress calls to ->start() here when we are already starting, to allow this request to
1908 * be used as a "hurry up" call, for example when the unit is in some "auto restart" state where it
1909 * waits for a holdoff timer to elapse before it will start again. */
1910
1911 unit_add_to_dbus_queue(u);
1912 unit_cgroup_freezer_action(u, FREEZER_THAW);
1913
1914 return UNIT_VTABLE(u)->start(u);
1915 }
1916
1917 bool unit_can_start(Unit *u) {
1918 assert(u);
1919
1920 if (u->load_state != UNIT_LOADED)
1921 return false;
1922
1923 if (!unit_type_supported(u->type))
1924 return false;
1925
1926 /* Scope units may be started only once */
1927 if (UNIT_VTABLE(u)->once_only && dual_timestamp_is_set(&u->inactive_exit_timestamp))
1928 return false;
1929
1930 return !!UNIT_VTABLE(u)->start;
1931 }
1932
1933 bool unit_can_isolate(Unit *u) {
1934 assert(u);
1935
1936 return unit_can_start(u) &&
1937 u->allow_isolate;
1938 }
1939
1940 /* Errors:
1941 * -EBADR: This unit type does not support stopping.
1942 * -EALREADY: Unit is already stopped.
1943 * -EAGAIN: An operation is already in progress. Retry later.
1944 */
1945 int unit_stop(Unit *u) {
1946 UnitActiveState state;
1947 Unit *following;
1948
1949 assert(u);
1950
1951 state = unit_active_state(u);
1952 if (UNIT_IS_INACTIVE_OR_FAILED(state))
1953 return -EALREADY;
1954
1955 following = unit_following(u);
1956 if (following) {
1957 log_unit_debug(u, "Redirecting stop request from %s to %s.", u->id, following->id);
1958 return unit_stop(following);
1959 }
1960
1961 if (!UNIT_VTABLE(u)->stop)
1962 return -EBADR;
1963
1964 unit_add_to_dbus_queue(u);
1965 unit_cgroup_freezer_action(u, FREEZER_THAW);
1966
1967 return UNIT_VTABLE(u)->stop(u);
1968 }
1969
1970 bool unit_can_stop(Unit *u) {
1971 assert(u);
1972
1973 /* Note: if we return true here, it does not mean that the unit may be successfully stopped.
1974 * Extrinsic units follow external state and they may stop following external state changes
1975 * (hence we return true here), but an attempt to do this through the manager will fail. */
1976
1977 if (!unit_type_supported(u->type))
1978 return false;
1979
1980 if (u->perpetual)
1981 return false;
1982
1983 return !!UNIT_VTABLE(u)->stop;
1984 }
1985
1986 /* Errors:
1987 * -EBADR: This unit type does not support reloading.
1988 * -ENOEXEC: Unit is not started.
1989 * -EAGAIN: An operation is already in progress. Retry later.
1990 */
1991 int unit_reload(Unit *u) {
1992 UnitActiveState state;
1993 Unit *following;
1994
1995 assert(u);
1996
1997 if (u->load_state != UNIT_LOADED)
1998 return -EINVAL;
1999
2000 if (!unit_can_reload(u))
2001 return -EBADR;
2002
2003 state = unit_active_state(u);
2004 if (state == UNIT_RELOADING)
2005 return -EAGAIN;
2006
2007 if (state != UNIT_ACTIVE)
2008 return log_unit_warning_errno(u, SYNTHETIC_ERRNO(ENOEXEC), "Unit cannot be reloaded because it is inactive.");
2009
2010 following = unit_following(u);
2011 if (following) {
2012 log_unit_debug(u, "Redirecting reload request from %s to %s.", u->id, following->id);
2013 return unit_reload(following);
2014 }
2015
2016 unit_add_to_dbus_queue(u);
2017
2018 if (!UNIT_VTABLE(u)->reload) {
2019 /* Unit doesn't have a reload function, but we need to propagate the reload anyway */
2020 unit_notify(u, unit_active_state(u), unit_active_state(u), 0);
2021 return 0;
2022 }
2023
2024 unit_cgroup_freezer_action(u, FREEZER_THAW);
2025
2026 return UNIT_VTABLE(u)->reload(u);
2027 }
2028
2029 bool unit_can_reload(Unit *u) {
2030 assert(u);
2031
2032 if (UNIT_VTABLE(u)->can_reload)
2033 return UNIT_VTABLE(u)->can_reload(u);
2034
2035 if (unit_has_dependency(u, UNIT_ATOM_PROPAGATES_RELOAD_TO, NULL))
2036 return true;
2037
2038 return UNIT_VTABLE(u)->reload;
2039 }
2040
2041 bool unit_is_unneeded(Unit *u) {
2042 Unit *other;
2043 assert(u);
2044
2045 if (!u->stop_when_unneeded)
2046 return false;
2047
2048 /* Don't clean up while the unit is transitioning or is even inactive. */
2049 if (unit_active_state(u) != UNIT_ACTIVE)
2050 return false;
2051 if (u->job)
2052 return false;
2053
2054 UNIT_FOREACH_DEPENDENCY(other, u, UNIT_ATOM_PINS_STOP_WHEN_UNNEEDED) {
2055 /* If a dependent unit has a job queued, is active or transitioning, or is marked for
2056 * restart, then don't clean this one up. */
2057
2058 if (other->job)
2059 return false;
2060
2061 if (!UNIT_IS_INACTIVE_OR_FAILED(unit_active_state(other)))
2062 return false;
2063
2064 if (unit_will_restart(other))
2065 return false;
2066 }
2067
2068 return true;
2069 }
2070
2071 bool unit_is_upheld_by_active(Unit *u, Unit **ret_culprit) {
2072 Unit *other;
2073
2074 assert(u);
2075
2076 /* Checks if the unit needs to be started because it currently is not running, but some other unit
2077 * that is active declared an Uphold= dependencies on it */
2078
2079 if (!UNIT_IS_INACTIVE_OR_FAILED(unit_active_state(u)) || u->job) {
2080 if (ret_culprit)
2081 *ret_culprit = NULL;
2082 return false;
2083 }
2084
2085 UNIT_FOREACH_DEPENDENCY(other, u, UNIT_ATOM_START_STEADILY) {
2086 if (other->job)
2087 continue;
2088
2089 if (UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(other))) {
2090 if (ret_culprit)
2091 *ret_culprit = other;
2092 return true;
2093 }
2094 }
2095
2096 if (ret_culprit)
2097 *ret_culprit = NULL;
2098 return false;
2099 }
2100
2101 bool unit_is_bound_by_inactive(Unit *u, Unit **ret_culprit) {
2102 Unit *other;
2103
2104 assert(u);
2105
2106 /* Checks whether this unit is bound to another unit that is inactive, i.e. whether we should stop
2107 * because the other unit is down. */
2108
2109 if (unit_active_state(u) != UNIT_ACTIVE || u->job) {
2110 /* Don't clean up while the unit is transitioning or is even inactive. */
2111 if (ret_culprit)
2112 *ret_culprit = NULL;
2113 return false;
2114 }
2115
2116 UNIT_FOREACH_DEPENDENCY(other, u, UNIT_ATOM_CANNOT_BE_ACTIVE_WITHOUT) {
2117 if (other->job)
2118 continue;
2119
2120 if (UNIT_IS_INACTIVE_OR_FAILED(unit_active_state(other))) {
2121 if (*ret_culprit)
2122 *ret_culprit = other;
2123
2124 return true;
2125 }
2126 }
2127
2128 if (ret_culprit)
2129 *ret_culprit = NULL;
2130 return false;
2131 }
2132
2133 static void check_unneeded_dependencies(Unit *u) {
2134 Unit *other;
2135 assert(u);
2136
2137 /* Add all units this unit depends on to the queue that processes StopWhenUnneeded= behaviour. */
2138
2139 UNIT_FOREACH_DEPENDENCY(other, u, UNIT_ATOM_ADD_STOP_WHEN_UNNEEDED_QUEUE)
2140 unit_submit_to_stop_when_unneeded_queue(other);
2141 }
2142
2143 static void check_uphold_dependencies(Unit *u) {
2144 Unit *other;
2145 assert(u);
2146
2147 /* Add all units this unit depends on to the queue that processes Uphold= behaviour. */
2148
2149 UNIT_FOREACH_DEPENDENCY(other, u, UNIT_ATOM_ADD_START_WHEN_UPHELD_QUEUE)
2150 unit_submit_to_start_when_upheld_queue(other);
2151 }
2152
2153 static void check_bound_by_dependencies(Unit *u) {
2154 Unit *other;
2155 assert(u);
2156
2157 /* Add all units this unit depends on to the queue that processes BindsTo= stop behaviour. */
2158
2159 UNIT_FOREACH_DEPENDENCY(other, u, UNIT_ATOM_ADD_CANNOT_BE_ACTIVE_WITHOUT_QUEUE)
2160 unit_submit_to_stop_when_bound_queue(other);
2161 }
2162
2163 static void retroactively_start_dependencies(Unit *u) {
2164 Unit *other;
2165
2166 assert(u);
2167 assert(UNIT_IS_ACTIVE_OR_ACTIVATING(unit_active_state(u)));
2168
2169 UNIT_FOREACH_DEPENDENCY(other, u, UNIT_ATOM_RETROACTIVE_START_REPLACE) /* Requires= + BindsTo= */
2170 if (!unit_has_dependency(u, UNIT_ATOM_AFTER, other) &&
2171 !UNIT_IS_ACTIVE_OR_ACTIVATING(unit_active_state(other)))
2172 manager_add_job(u->manager, JOB_START, other, JOB_REPLACE, NULL, NULL, NULL);
2173
2174 UNIT_FOREACH_DEPENDENCY(other, u, UNIT_ATOM_RETROACTIVE_START_FAIL) /* Wants= */
2175 if (!unit_has_dependency(u, UNIT_ATOM_AFTER, other) &&
2176 !UNIT_IS_ACTIVE_OR_ACTIVATING(unit_active_state(other)))
2177 manager_add_job(u->manager, JOB_START, other, JOB_FAIL, NULL, NULL, NULL);
2178
2179 UNIT_FOREACH_DEPENDENCY(other, u, UNIT_ATOM_RETROACTIVE_STOP_ON_START) /* Conflicts= (and inverse) */
2180 if (!UNIT_IS_INACTIVE_OR_DEACTIVATING(unit_active_state(other)))
2181 manager_add_job(u->manager, JOB_STOP, other, JOB_REPLACE, NULL, NULL, NULL);
2182 }
2183
2184 static void retroactively_stop_dependencies(Unit *u) {
2185 Unit *other;
2186
2187 assert(u);
2188 assert(UNIT_IS_INACTIVE_OR_DEACTIVATING(unit_active_state(u)));
2189
2190 /* Pull down units which are bound to us recursively if enabled */
2191 UNIT_FOREACH_DEPENDENCY(other, u, UNIT_ATOM_RETROACTIVE_STOP_ON_STOP) /* BoundBy= */
2192 if (!UNIT_IS_INACTIVE_OR_DEACTIVATING(unit_active_state(other)))
2193 manager_add_job(u->manager, JOB_STOP, other, JOB_REPLACE, NULL, NULL, NULL);
2194 }
2195
2196 void unit_start_on_failure(
2197 Unit *u,
2198 const char *dependency_name,
2199 UnitDependencyAtom atom,
2200 JobMode job_mode) {
2201
2202 bool logged = false;
2203 Unit *other;
2204 int r;
2205
2206 assert(u);
2207 assert(dependency_name);
2208 assert(IN_SET(atom, UNIT_ATOM_ON_SUCCESS, UNIT_ATOM_ON_FAILURE));
2209
2210 /* Act on OnFailure= and OnSuccess= dependencies */
2211
2212 UNIT_FOREACH_DEPENDENCY(other, u, atom) {
2213 _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL;
2214
2215 if (!logged) {
2216 log_unit_info(u, "Triggering %s dependencies.", dependency_name);
2217 logged = true;
2218 }
2219
2220 r = manager_add_job(u->manager, JOB_START, other, job_mode, NULL, &error, NULL);
2221 if (r < 0)
2222 log_unit_warning_errno(
2223 u, r, "Failed to enqueue %s job, ignoring: %s",
2224 dependency_name, bus_error_message(&error, r));
2225 }
2226
2227 if (logged)
2228 log_unit_debug(u, "Triggering %s dependencies done.", dependency_name);
2229 }
2230
2231 void unit_trigger_notify(Unit *u) {
2232 Unit *other;
2233
2234 assert(u);
2235
2236 UNIT_FOREACH_DEPENDENCY(other, u, UNIT_ATOM_TRIGGERED_BY)
2237 if (UNIT_VTABLE(other)->trigger_notify)
2238 UNIT_VTABLE(other)->trigger_notify(other, u);
2239 }
2240
2241 static int raise_level(int log_level, bool condition_info, bool condition_notice) {
2242 if (condition_notice && log_level > LOG_NOTICE)
2243 return LOG_NOTICE;
2244 if (condition_info && log_level > LOG_INFO)
2245 return LOG_INFO;
2246 return log_level;
2247 }
2248
2249 static int unit_log_resources(Unit *u) {
2250 struct iovec iovec[1 + _CGROUP_IP_ACCOUNTING_METRIC_MAX + _CGROUP_IO_ACCOUNTING_METRIC_MAX + 4];
2251 bool any_traffic = false, have_ip_accounting = false, any_io = false, have_io_accounting = false;
2252 _cleanup_free_ char *igress = NULL, *egress = NULL, *rr = NULL, *wr = NULL;
2253 int log_level = LOG_DEBUG; /* May be raised if resources consumed over a threshold */
2254 size_t n_message_parts = 0, n_iovec = 0;
2255 char* message_parts[1 + 2 + 2 + 1], *t;
2256 nsec_t nsec = NSEC_INFINITY;
2257 int r;
2258 const char* const ip_fields[_CGROUP_IP_ACCOUNTING_METRIC_MAX] = {
2259 [CGROUP_IP_INGRESS_BYTES] = "IP_METRIC_INGRESS_BYTES",
2260 [CGROUP_IP_INGRESS_PACKETS] = "IP_METRIC_INGRESS_PACKETS",
2261 [CGROUP_IP_EGRESS_BYTES] = "IP_METRIC_EGRESS_BYTES",
2262 [CGROUP_IP_EGRESS_PACKETS] = "IP_METRIC_EGRESS_PACKETS",
2263 };
2264 const char* const io_fields[_CGROUP_IO_ACCOUNTING_METRIC_MAX] = {
2265 [CGROUP_IO_READ_BYTES] = "IO_METRIC_READ_BYTES",
2266 [CGROUP_IO_WRITE_BYTES] = "IO_METRIC_WRITE_BYTES",
2267 [CGROUP_IO_READ_OPERATIONS] = "IO_METRIC_READ_OPERATIONS",
2268 [CGROUP_IO_WRITE_OPERATIONS] = "IO_METRIC_WRITE_OPERATIONS",
2269 };
2270
2271 assert(u);
2272
2273 /* Invoked whenever a unit enters failed or dead state. Logs information about consumed resources if resource
2274 * accounting was enabled for a unit. It does this in two ways: a friendly human readable string with reduced
2275 * information and the complete data in structured fields. */
2276
2277 (void) unit_get_cpu_usage(u, &nsec);
2278 if (nsec != NSEC_INFINITY) {
2279 /* Format the CPU time for inclusion in the structured log message */
2280 if (asprintf(&t, "CPU_USAGE_NSEC=%" PRIu64, nsec) < 0) {
2281 r = log_oom();
2282 goto finish;
2283 }
2284 iovec[n_iovec++] = IOVEC_MAKE_STRING(t);
2285
2286 /* Format the CPU time for inclusion in the human language message string */
2287 t = strjoin("consumed ", FORMAT_TIMESPAN(nsec / NSEC_PER_USEC, USEC_PER_MSEC), " CPU time");
2288 if (!t) {
2289 r = log_oom();
2290 goto finish;
2291 }
2292
2293 message_parts[n_message_parts++] = t;
2294
2295 log_level = raise_level(log_level,
2296 nsec > NOTICEWORTHY_CPU_NSEC,
2297 nsec > MENTIONWORTHY_CPU_NSEC);
2298 }
2299
2300 for (CGroupIOAccountingMetric k = 0; k < _CGROUP_IO_ACCOUNTING_METRIC_MAX; k++) {
2301 uint64_t value = UINT64_MAX;
2302
2303 assert(io_fields[k]);
2304
2305 (void) unit_get_io_accounting(u, k, k > 0, &value);
2306 if (value == UINT64_MAX)
2307 continue;
2308
2309 have_io_accounting = true;
2310 if (value > 0)
2311 any_io = true;
2312
2313 /* Format IO accounting data for inclusion in the structured log message */
2314 if (asprintf(&t, "%s=%" PRIu64, io_fields[k], value) < 0) {
2315 r = log_oom();
2316 goto finish;
2317 }
2318 iovec[n_iovec++] = IOVEC_MAKE_STRING(t);
2319
2320 /* Format the IO accounting data for inclusion in the human language message string, but only
2321 * for the bytes counters (and not for the operations counters) */
2322 if (k == CGROUP_IO_READ_BYTES) {
2323 assert(!rr);
2324 rr = strjoin("read ", strna(FORMAT_BYTES(value)), " from disk");
2325 if (!rr) {
2326 r = log_oom();
2327 goto finish;
2328 }
2329 } else if (k == CGROUP_IO_WRITE_BYTES) {
2330 assert(!wr);
2331 wr = strjoin("written ", strna(FORMAT_BYTES(value)), " to disk");
2332 if (!wr) {
2333 r = log_oom();
2334 goto finish;
2335 }
2336 }
2337
2338 if (IN_SET(k, CGROUP_IO_READ_BYTES, CGROUP_IO_WRITE_BYTES))
2339 log_level = raise_level(log_level,
2340 value > MENTIONWORTHY_IO_BYTES,
2341 value > NOTICEWORTHY_IO_BYTES);
2342 }
2343
2344 if (have_io_accounting) {
2345 if (any_io) {
2346 if (rr)
2347 message_parts[n_message_parts++] = TAKE_PTR(rr);
2348 if (wr)
2349 message_parts[n_message_parts++] = TAKE_PTR(wr);
2350
2351 } else {
2352 char *k;
2353
2354 k = strdup("no IO");
2355 if (!k) {
2356 r = log_oom();
2357 goto finish;
2358 }
2359
2360 message_parts[n_message_parts++] = k;
2361 }
2362 }
2363
2364 for (CGroupIPAccountingMetric m = 0; m < _CGROUP_IP_ACCOUNTING_METRIC_MAX; m++) {
2365 uint64_t value = UINT64_MAX;
2366
2367 assert(ip_fields[m]);
2368
2369 (void) unit_get_ip_accounting(u, m, &value);
2370 if (value == UINT64_MAX)
2371 continue;
2372
2373 have_ip_accounting = true;
2374 if (value > 0)
2375 any_traffic = true;
2376
2377 /* Format IP accounting data for inclusion in the structured log message */
2378 if (asprintf(&t, "%s=%" PRIu64, ip_fields[m], value) < 0) {
2379 r = log_oom();
2380 goto finish;
2381 }
2382 iovec[n_iovec++] = IOVEC_MAKE_STRING(t);
2383
2384 /* Format the IP accounting data for inclusion in the human language message string, but only for the
2385 * bytes counters (and not for the packets counters) */
2386 if (m == CGROUP_IP_INGRESS_BYTES) {
2387 assert(!igress);
2388 igress = strjoin("received ", strna(FORMAT_BYTES(value)), " IP traffic");
2389 if (!igress) {
2390 r = log_oom();
2391 goto finish;
2392 }
2393 } else if (m == CGROUP_IP_EGRESS_BYTES) {
2394 assert(!egress);
2395 egress = strjoin("sent ", strna(FORMAT_BYTES(value)), " IP traffic");
2396 if (!egress) {
2397 r = log_oom();
2398 goto finish;
2399 }
2400 }
2401
2402 if (IN_SET(m, CGROUP_IP_INGRESS_BYTES, CGROUP_IP_EGRESS_BYTES))
2403 log_level = raise_level(log_level,
2404 value > MENTIONWORTHY_IP_BYTES,
2405 value > NOTICEWORTHY_IP_BYTES);
2406 }
2407
2408 /* This check is here because it is the earliest point following all possible log_level assignments. If
2409 * log_level is assigned anywhere after this point, move this check. */
2410 if (!unit_log_level_test(u, log_level)) {
2411 r = 0;
2412 goto finish;
2413 }
2414
2415 if (have_ip_accounting) {
2416 if (any_traffic) {
2417 if (igress)
2418 message_parts[n_message_parts++] = TAKE_PTR(igress);
2419 if (egress)
2420 message_parts[n_message_parts++] = TAKE_PTR(egress);
2421
2422 } else {
2423 char *k;
2424
2425 k = strdup("no IP traffic");
2426 if (!k) {
2427 r = log_oom();
2428 goto finish;
2429 }
2430
2431 message_parts[n_message_parts++] = k;
2432 }
2433 }
2434
2435 /* Is there any accounting data available at all? */
2436 if (n_iovec == 0) {
2437 r = 0;
2438 goto finish;
2439 }
2440
2441 if (n_message_parts == 0)
2442 t = strjoina("MESSAGE=", u->id, ": Completed.");
2443 else {
2444 _cleanup_free_ char *joined = NULL;
2445
2446 message_parts[n_message_parts] = NULL;
2447
2448 joined = strv_join(message_parts, ", ");
2449 if (!joined) {
2450 r = log_oom();
2451 goto finish;
2452 }
2453
2454 joined[0] = ascii_toupper(joined[0]);
2455 t = strjoina("MESSAGE=", u->id, ": ", joined, ".");
2456 }
2457
2458 /* The following four fields we allocate on the stack or are static strings, we hence don't want to free them,
2459 * and hence don't increase n_iovec for them */
2460 iovec[n_iovec] = IOVEC_MAKE_STRING(t);
2461 iovec[n_iovec + 1] = IOVEC_MAKE_STRING("MESSAGE_ID=" SD_MESSAGE_UNIT_RESOURCES_STR);
2462
2463 t = strjoina(u->manager->unit_log_field, u->id);
2464 iovec[n_iovec + 2] = IOVEC_MAKE_STRING(t);
2465
2466 t = strjoina(u->manager->invocation_log_field, u->invocation_id_string);
2467 iovec[n_iovec + 3] = IOVEC_MAKE_STRING(t);
2468
2469 log_unit_struct_iovec(u, log_level, iovec, n_iovec + 4);
2470 r = 0;
2471
2472 finish:
2473 for (size_t i = 0; i < n_message_parts; i++)
2474 free(message_parts[i]);
2475
2476 for (size_t i = 0; i < n_iovec; i++)
2477 free(iovec[i].iov_base);
2478
2479 return r;
2480
2481 }
2482
2483 static void unit_update_on_console(Unit *u) {
2484 bool b;
2485
2486 assert(u);
2487
2488 b = unit_needs_console(u);
2489 if (u->on_console == b)
2490 return;
2491
2492 u->on_console = b;
2493 if (b)
2494 manager_ref_console(u->manager);
2495 else
2496 manager_unref_console(u->manager);
2497 }
2498
2499 static void unit_emit_audit_start(Unit *u) {
2500 assert(u);
2501
2502 if (u->type != UNIT_SERVICE)
2503 return;
2504
2505 /* Write audit record if we have just finished starting up */
2506 manager_send_unit_audit(u->manager, u, AUDIT_SERVICE_START, true);
2507 u->in_audit = true;
2508 }
2509
2510 static void unit_emit_audit_stop(Unit *u, UnitActiveState state) {
2511 assert(u);
2512
2513 if (u->type != UNIT_SERVICE)
2514 return;
2515
2516 if (u->in_audit) {
2517 /* Write audit record if we have just finished shutting down */
2518 manager_send_unit_audit(u->manager, u, AUDIT_SERVICE_STOP, state == UNIT_INACTIVE);
2519 u->in_audit = false;
2520 } else {
2521 /* Hmm, if there was no start record written write it now, so that we always have a nice pair */
2522 manager_send_unit_audit(u->manager, u, AUDIT_SERVICE_START, state == UNIT_INACTIVE);
2523
2524 if (state == UNIT_INACTIVE)
2525 manager_send_unit_audit(u->manager, u, AUDIT_SERVICE_STOP, true);
2526 }
2527 }
2528
2529 static bool unit_process_job(Job *j, UnitActiveState ns, UnitNotifyFlags flags) {
2530 bool unexpected = false;
2531 JobResult result;
2532
2533 assert(j);
2534
2535 if (j->state == JOB_WAITING)
2536
2537 /* So we reached a different state for this job. Let's see if we can run it now if it failed previously
2538 * due to EAGAIN. */
2539 job_add_to_run_queue(j);
2540
2541 /* Let's check whether the unit's new state constitutes a finished job, or maybe contradicts a running job and
2542 * hence needs to invalidate jobs. */
2543
2544 switch (j->type) {
2545
2546 case JOB_START:
2547 case JOB_VERIFY_ACTIVE:
2548
2549 if (UNIT_IS_ACTIVE_OR_RELOADING(ns))
2550 job_finish_and_invalidate(j, JOB_DONE, true, false);
2551 else if (j->state == JOB_RUNNING && ns != UNIT_ACTIVATING) {
2552 unexpected = true;
2553
2554 if (UNIT_IS_INACTIVE_OR_FAILED(ns)) {
2555 if (ns == UNIT_FAILED)
2556 result = JOB_FAILED;
2557 else
2558 result = JOB_DONE;
2559
2560 job_finish_and_invalidate(j, result, true, false);
2561 }
2562 }
2563
2564 break;
2565
2566 case JOB_RELOAD:
2567 case JOB_RELOAD_OR_START:
2568 case JOB_TRY_RELOAD:
2569
2570 if (j->state == JOB_RUNNING) {
2571 if (ns == UNIT_ACTIVE)
2572 job_finish_and_invalidate(j, (flags & UNIT_NOTIFY_RELOAD_FAILURE) ? JOB_FAILED : JOB_DONE, true, false);
2573 else if (!IN_SET(ns, UNIT_ACTIVATING, UNIT_RELOADING)) {
2574 unexpected = true;
2575
2576 if (UNIT_IS_INACTIVE_OR_FAILED(ns))
2577 job_finish_and_invalidate(j, ns == UNIT_FAILED ? JOB_FAILED : JOB_DONE, true, false);
2578 }
2579 }
2580
2581 break;
2582
2583 case JOB_STOP:
2584 case JOB_RESTART:
2585 case JOB_TRY_RESTART:
2586
2587 if (UNIT_IS_INACTIVE_OR_FAILED(ns))
2588 job_finish_and_invalidate(j, JOB_DONE, true, false);
2589 else if (j->state == JOB_RUNNING && ns != UNIT_DEACTIVATING) {
2590 unexpected = true;
2591 job_finish_and_invalidate(j, JOB_FAILED, true, false);
2592 }
2593
2594 break;
2595
2596 default:
2597 assert_not_reached();
2598 }
2599
2600 return unexpected;
2601 }
2602
2603 void unit_notify(Unit *u, UnitActiveState os, UnitActiveState ns, UnitNotifyFlags flags) {
2604 const char *reason;
2605 Manager *m;
2606
2607 assert(u);
2608 assert(os < _UNIT_ACTIVE_STATE_MAX);
2609 assert(ns < _UNIT_ACTIVE_STATE_MAX);
2610
2611 /* Note that this is called for all low-level state changes, even if they might map to the same high-level
2612 * UnitActiveState! That means that ns == os is an expected behavior here. For example: if a mount point is
2613 * remounted this function will be called too! */
2614
2615 m = u->manager;
2616
2617 /* Let's enqueue the change signal early. In case this unit has a job associated we want that this unit is in
2618 * the bus queue, so that any job change signal queued will force out the unit change signal first. */
2619 unit_add_to_dbus_queue(u);
2620
2621 /* Update systemd-oomd on the property/state change */
2622 if (os != ns) {
2623 /* Always send an update if the unit is going into an inactive state so systemd-oomd knows to stop
2624 * monitoring.
2625 * Also send an update whenever the unit goes active; this is to handle a case where an override file
2626 * sets one of the ManagedOOM*= properties to "kill", then later removes it. systemd-oomd needs to
2627 * know to stop monitoring when the unit changes from "kill" -> "auto" on daemon-reload, but we don't
2628 * have the information on the property. Thus, indiscriminately send an update. */
2629 if (UNIT_IS_INACTIVE_OR_FAILED(ns) || UNIT_IS_ACTIVE_OR_RELOADING(ns))
2630 (void) manager_varlink_send_managed_oom_update(u);
2631 }
2632
2633 /* Update timestamps for state changes */
2634 if (!MANAGER_IS_RELOADING(m)) {
2635 dual_timestamp_get(&u->state_change_timestamp);
2636
2637 if (UNIT_IS_INACTIVE_OR_FAILED(os) && !UNIT_IS_INACTIVE_OR_FAILED(ns))
2638 u->inactive_exit_timestamp = u->state_change_timestamp;
2639 else if (!UNIT_IS_INACTIVE_OR_FAILED(os) && UNIT_IS_INACTIVE_OR_FAILED(ns))
2640 u->inactive_enter_timestamp = u->state_change_timestamp;
2641
2642 if (!UNIT_IS_ACTIVE_OR_RELOADING(os) && UNIT_IS_ACTIVE_OR_RELOADING(ns))
2643 u->active_enter_timestamp = u->state_change_timestamp;
2644 else if (UNIT_IS_ACTIVE_OR_RELOADING(os) && !UNIT_IS_ACTIVE_OR_RELOADING(ns))
2645 u->active_exit_timestamp = u->state_change_timestamp;
2646 }
2647
2648 /* Keep track of failed units */
2649 (void) manager_update_failed_units(m, u, ns == UNIT_FAILED);
2650
2651 /* Make sure the cgroup and state files are always removed when we become inactive */
2652 if (UNIT_IS_INACTIVE_OR_FAILED(ns)) {
2653 SET_FLAG(u->markers,
2654 (1u << UNIT_MARKER_NEEDS_RELOAD)|(1u << UNIT_MARKER_NEEDS_RESTART),
2655 false);
2656 unit_prune_cgroup(u);
2657 unit_unlink_state_files(u);
2658 } else if (ns != os && ns == UNIT_RELOADING)
2659 SET_FLAG(u->markers, 1u << UNIT_MARKER_NEEDS_RELOAD, false);
2660
2661 unit_update_on_console(u);
2662
2663 if (!MANAGER_IS_RELOADING(m)) {
2664 bool unexpected;
2665
2666 /* Let's propagate state changes to the job */
2667 if (u->job)
2668 unexpected = unit_process_job(u->job, ns, flags);
2669 else
2670 unexpected = true;
2671
2672 /* If this state change happened without being requested by a job, then let's retroactively start or
2673 * stop dependencies. We skip that step when deserializing, since we don't want to create any
2674 * additional jobs just because something is already activated. */
2675
2676 if (unexpected) {
2677 if (UNIT_IS_INACTIVE_OR_FAILED(os) && UNIT_IS_ACTIVE_OR_ACTIVATING(ns))
2678 retroactively_start_dependencies(u);
2679 else if (UNIT_IS_ACTIVE_OR_ACTIVATING(os) && UNIT_IS_INACTIVE_OR_DEACTIVATING(ns))
2680 retroactively_stop_dependencies(u);
2681 }
2682
2683 if (ns != os && ns == UNIT_FAILED) {
2684 log_unit_debug(u, "Unit entered failed state.");
2685
2686 if (!(flags & UNIT_NOTIFY_WILL_AUTO_RESTART))
2687 unit_start_on_failure(u, "OnFailure=", UNIT_ATOM_ON_FAILURE, u->on_failure_job_mode);
2688 }
2689
2690 if (UNIT_IS_ACTIVE_OR_RELOADING(ns) && !UNIT_IS_ACTIVE_OR_RELOADING(os)) {
2691 /* This unit just finished starting up */
2692
2693 unit_emit_audit_start(u);
2694 manager_send_unit_plymouth(m, u);
2695 }
2696
2697 if (UNIT_IS_INACTIVE_OR_FAILED(ns) && !UNIT_IS_INACTIVE_OR_FAILED(os)) {
2698 /* This unit just stopped/failed. */
2699
2700 unit_emit_audit_stop(u, ns);
2701 unit_log_resources(u);
2702 }
2703
2704 if (ns == UNIT_INACTIVE && !IN_SET(os, UNIT_FAILED, UNIT_INACTIVE, UNIT_MAINTENANCE) &&
2705 !(flags & UNIT_NOTIFY_WILL_AUTO_RESTART))
2706 unit_start_on_failure(u, "OnSuccess=", UNIT_ATOM_ON_SUCCESS, u->on_success_job_mode);
2707 }
2708
2709 manager_recheck_journal(m);
2710 manager_recheck_dbus(m);
2711
2712 unit_trigger_notify(u);
2713
2714 if (!MANAGER_IS_RELOADING(m)) {
2715 if (os != UNIT_FAILED && ns == UNIT_FAILED) {
2716 reason = strjoina("unit ", u->id, " failed");
2717 emergency_action(m, u->failure_action, 0, u->reboot_arg, unit_failure_action_exit_status(u), reason);
2718 } else if (!UNIT_IS_INACTIVE_OR_FAILED(os) && ns == UNIT_INACTIVE) {
2719 reason = strjoina("unit ", u->id, " succeeded");
2720 emergency_action(m, u->success_action, 0, u->reboot_arg, unit_success_action_exit_status(u), reason);
2721 }
2722 }
2723
2724 /* And now, add the unit or depending units to various queues that will act on the new situation if
2725 * needed. These queues generally check for continuous state changes rather than events (like most of
2726 * the state propagation above), and do work deferred instead of instantly, since they typically
2727 * don't want to run during reloading, and usually involve checking combined state of multiple units
2728 * at once. */
2729
2730 if (UNIT_IS_INACTIVE_OR_FAILED(ns)) {
2731 /* Stop unneeded units and bound-by units regardless if going down was expected or not */
2732 check_unneeded_dependencies(u);
2733 check_bound_by_dependencies(u);
2734
2735 /* Maybe someone wants us to remain up? */
2736 unit_submit_to_start_when_upheld_queue(u);
2737
2738 /* Maybe the unit should be GC'ed now? */
2739 unit_add_to_gc_queue(u);
2740 }
2741
2742 if (UNIT_IS_ACTIVE_OR_RELOADING(ns)) {
2743 /* Start uphold units regardless if going up was expected or not */
2744 check_uphold_dependencies(u);
2745
2746 /* Maybe we finished startup and are now ready for being stopped because unneeded? */
2747 unit_submit_to_stop_when_unneeded_queue(u);
2748
2749 /* Maybe we finished startup, but something we needed has vanished? Let's die then. (This happens
2750 * when something BindsTo= to a Type=oneshot unit, as these units go directly from starting to
2751 * inactive, without ever entering started.) */
2752 unit_submit_to_stop_when_bound_queue(u);
2753 }
2754 }
2755
2756 int unit_watch_pid(Unit *u, pid_t pid, bool exclusive) {
2757 int r;
2758
2759 assert(u);
2760 assert(pid_is_valid(pid));
2761
2762 /* Watch a specific PID */
2763
2764 /* Caller might be sure that this PID belongs to this unit only. Let's take this
2765 * opportunity to remove any stalled references to this PID as they can be created
2766 * easily (when watching a process which is not our direct child). */
2767 if (exclusive)
2768 manager_unwatch_pid(u->manager, pid);
2769
2770 r = set_ensure_allocated(&u->pids, NULL);
2771 if (r < 0)
2772 return r;
2773
2774 r = hashmap_ensure_allocated(&u->manager->watch_pids, NULL);
2775 if (r < 0)
2776 return r;
2777
2778 /* First try, let's add the unit keyed by "pid". */
2779 r = hashmap_put(u->manager->watch_pids, PID_TO_PTR(pid), u);
2780 if (r == -EEXIST) {
2781 Unit **array;
2782 bool found = false;
2783 size_t n = 0;
2784
2785 /* OK, the "pid" key is already assigned to a different unit. Let's see if the "-pid" key (which points
2786 * to an array of Units rather than just a Unit), lists us already. */
2787
2788 array = hashmap_get(u->manager->watch_pids, PID_TO_PTR(-pid));
2789 if (array)
2790 for (; array[n]; n++)
2791 if (array[n] == u)
2792 found = true;
2793
2794 if (found) /* Found it already? if so, do nothing */
2795 r = 0;
2796 else {
2797 Unit **new_array;
2798
2799 /* Allocate a new array */
2800 new_array = new(Unit*, n + 2);
2801 if (!new_array)
2802 return -ENOMEM;
2803
2804 memcpy_safe(new_array, array, sizeof(Unit*) * n);
2805 new_array[n] = u;
2806 new_array[n+1] = NULL;
2807
2808 /* Add or replace the old array */
2809 r = hashmap_replace(u->manager->watch_pids, PID_TO_PTR(-pid), new_array);
2810 if (r < 0) {
2811 free(new_array);
2812 return r;
2813 }
2814
2815 free(array);
2816 }
2817 } else if (r < 0)
2818 return r;
2819
2820 r = set_put(u->pids, PID_TO_PTR(pid));
2821 if (r < 0)
2822 return r;
2823
2824 return 0;
2825 }
2826
2827 void unit_unwatch_pid(Unit *u, pid_t pid) {
2828 Unit **array;
2829
2830 assert(u);
2831 assert(pid_is_valid(pid));
2832
2833 /* First let's drop the unit in case it's keyed as "pid". */
2834 (void) hashmap_remove_value(u->manager->watch_pids, PID_TO_PTR(pid), u);
2835
2836 /* Then, let's also drop the unit, in case it's in the array keyed by -pid */
2837 array = hashmap_get(u->manager->watch_pids, PID_TO_PTR(-pid));
2838 if (array) {
2839 /* Let's iterate through the array, dropping our own entry */
2840
2841 size_t m = 0;
2842 for (size_t n = 0; array[n]; n++)
2843 if (array[n] != u)
2844 array[m++] = array[n];
2845 array[m] = NULL;
2846
2847 if (m == 0) {
2848 /* The array is now empty, remove the entire entry */
2849 assert_se(hashmap_remove(u->manager->watch_pids, PID_TO_PTR(-pid)) == array);
2850 free(array);
2851 }
2852 }
2853
2854 (void) set_remove(u->pids, PID_TO_PTR(pid));
2855 }
2856
2857 void unit_unwatch_all_pids(Unit *u) {
2858 assert(u);
2859
2860 while (!set_isempty(u->pids))
2861 unit_unwatch_pid(u, PTR_TO_PID(set_first(u->pids)));
2862
2863 u->pids = set_free(u->pids);
2864 }
2865
2866 static void unit_tidy_watch_pids(Unit *u) {
2867 pid_t except1, except2;
2868 void *e;
2869
2870 assert(u);
2871
2872 /* Cleans dead PIDs from our list */
2873
2874 except1 = unit_main_pid(u);
2875 except2 = unit_control_pid(u);
2876
2877 SET_FOREACH(e, u->pids) {
2878 pid_t pid = PTR_TO_PID(e);
2879
2880 if (pid == except1 || pid == except2)
2881 continue;
2882
2883 if (!pid_is_unwaited(pid))
2884 unit_unwatch_pid(u, pid);
2885 }
2886 }
2887
2888 static int on_rewatch_pids_event(sd_event_source *s, void *userdata) {
2889 Unit *u = userdata;
2890
2891 assert(s);
2892 assert(u);
2893
2894 unit_tidy_watch_pids(u);
2895 unit_watch_all_pids(u);
2896
2897 /* If the PID set is empty now, then let's finish this off. */
2898 unit_synthesize_cgroup_empty_event(u);
2899
2900 return 0;
2901 }
2902
2903 int unit_enqueue_rewatch_pids(Unit *u) {
2904 int r;
2905
2906 assert(u);
2907
2908 if (!u->cgroup_path)
2909 return -ENOENT;
2910
2911 r = cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER);
2912 if (r < 0)
2913 return r;
2914 if (r > 0) /* On unified we can use proper notifications */
2915 return 0;
2916
2917 /* Enqueues a low-priority job that will clean up dead PIDs from our list of PIDs to watch and subscribe to new
2918 * PIDs that might have appeared. We do this in a delayed job because the work might be quite slow, as it
2919 * involves issuing kill(pid, 0) on all processes we watch. */
2920
2921 if (!u->rewatch_pids_event_source) {
2922 _cleanup_(sd_event_source_unrefp) sd_event_source *s = NULL;
2923
2924 r = sd_event_add_defer(u->manager->event, &s, on_rewatch_pids_event, u);
2925 if (r < 0)
2926 return log_error_errno(r, "Failed to allocate event source for tidying watched PIDs: %m");
2927
2928 r = sd_event_source_set_priority(s, SD_EVENT_PRIORITY_IDLE);
2929 if (r < 0)
2930 return log_error_errno(r, "Failed to adjust priority of event source for tidying watched PIDs: %m");
2931
2932 (void) sd_event_source_set_description(s, "tidy-watch-pids");
2933
2934 u->rewatch_pids_event_source = TAKE_PTR(s);
2935 }
2936
2937 r = sd_event_source_set_enabled(u->rewatch_pids_event_source, SD_EVENT_ONESHOT);
2938 if (r < 0)
2939 return log_error_errno(r, "Failed to enable event source for tidying watched PIDs: %m");
2940
2941 return 0;
2942 }
2943
2944 void unit_dequeue_rewatch_pids(Unit *u) {
2945 int r;
2946 assert(u);
2947
2948 if (!u->rewatch_pids_event_source)
2949 return;
2950
2951 r = sd_event_source_set_enabled(u->rewatch_pids_event_source, SD_EVENT_OFF);
2952 if (r < 0)
2953 log_warning_errno(r, "Failed to disable event source for tidying watched PIDs, ignoring: %m");
2954
2955 u->rewatch_pids_event_source = sd_event_source_disable_unref(u->rewatch_pids_event_source);
2956 }
2957
2958 bool unit_job_is_applicable(Unit *u, JobType j) {
2959 assert(u);
2960 assert(j >= 0 && j < _JOB_TYPE_MAX);
2961
2962 switch (j) {
2963
2964 case JOB_VERIFY_ACTIVE:
2965 case JOB_START:
2966 case JOB_NOP:
2967 /* Note that we don't check unit_can_start() here. That's because .device units and suchlike are not
2968 * startable by us but may appear due to external events, and it thus makes sense to permit enqueuing
2969 * jobs for it. */
2970 return true;
2971
2972 case JOB_STOP:
2973 /* Similar as above. However, perpetual units can never be stopped (neither explicitly nor due to
2974 * external events), hence it makes no sense to permit enqueuing such a request either. */
2975 return !u->perpetual;
2976
2977 case JOB_RESTART:
2978 case JOB_TRY_RESTART:
2979 return unit_can_stop(u) && unit_can_start(u);
2980
2981 case JOB_RELOAD:
2982 case JOB_TRY_RELOAD:
2983 return unit_can_reload(u);
2984
2985 case JOB_RELOAD_OR_START:
2986 return unit_can_reload(u) && unit_can_start(u);
2987
2988 default:
2989 assert_not_reached();
2990 }
2991 }
2992
2993 int unit_add_dependency(
2994 Unit *u,
2995 UnitDependency d,
2996 Unit *other,
2997 bool add_reference,
2998 UnitDependencyMask mask) {
2999
3000 static const UnitDependency inverse_table[_UNIT_DEPENDENCY_MAX] = {
3001 [UNIT_REQUIRES] = UNIT_REQUIRED_BY,
3002 [UNIT_REQUISITE] = UNIT_REQUISITE_OF,
3003 [UNIT_WANTS] = UNIT_WANTED_BY,
3004 [UNIT_BINDS_TO] = UNIT_BOUND_BY,
3005 [UNIT_PART_OF] = UNIT_CONSISTS_OF,
3006 [UNIT_UPHOLDS] = UNIT_UPHELD_BY,
3007 [UNIT_REQUIRED_BY] = UNIT_REQUIRES,
3008 [UNIT_REQUISITE_OF] = UNIT_REQUISITE,
3009 [UNIT_WANTED_BY] = UNIT_WANTS,
3010 [UNIT_BOUND_BY] = UNIT_BINDS_TO,
3011 [UNIT_CONSISTS_OF] = UNIT_PART_OF,
3012 [UNIT_UPHELD_BY] = UNIT_UPHOLDS,
3013 [UNIT_CONFLICTS] = UNIT_CONFLICTED_BY,
3014 [UNIT_CONFLICTED_BY] = UNIT_CONFLICTS,
3015 [UNIT_BEFORE] = UNIT_AFTER,
3016 [UNIT_AFTER] = UNIT_BEFORE,
3017 [UNIT_ON_SUCCESS] = UNIT_ON_SUCCESS_OF,
3018 [UNIT_ON_SUCCESS_OF] = UNIT_ON_SUCCESS,
3019 [UNIT_ON_FAILURE] = UNIT_ON_FAILURE_OF,
3020 [UNIT_ON_FAILURE_OF] = UNIT_ON_FAILURE,
3021 [UNIT_TRIGGERS] = UNIT_TRIGGERED_BY,
3022 [UNIT_TRIGGERED_BY] = UNIT_TRIGGERS,
3023 [UNIT_PROPAGATES_RELOAD_TO] = UNIT_RELOAD_PROPAGATED_FROM,
3024 [UNIT_RELOAD_PROPAGATED_FROM] = UNIT_PROPAGATES_RELOAD_TO,
3025 [UNIT_PROPAGATES_STOP_TO] = UNIT_STOP_PROPAGATED_FROM,
3026 [UNIT_STOP_PROPAGATED_FROM] = UNIT_PROPAGATES_STOP_TO,
3027 [UNIT_JOINS_NAMESPACE_OF] = UNIT_JOINS_NAMESPACE_OF, /* symmetric! 👓 */
3028 [UNIT_REFERENCES] = UNIT_REFERENCED_BY,
3029 [UNIT_REFERENCED_BY] = UNIT_REFERENCES,
3030 [UNIT_IN_SLICE] = UNIT_SLICE_OF,
3031 [UNIT_SLICE_OF] = UNIT_IN_SLICE,
3032 };
3033 Unit *original_u = u, *original_other = other;
3034 UnitDependencyAtom a;
3035 int r;
3036
3037 /* Helper to know whether sending a notification is necessary or not: if the dependency is already
3038 * there, no need to notify! */
3039 bool noop;
3040
3041 assert(u);
3042 assert(d >= 0 && d < _UNIT_DEPENDENCY_MAX);
3043 assert(other);
3044
3045 u = unit_follow_merge(u);
3046 other = unit_follow_merge(other);
3047 a = unit_dependency_to_atom(d);
3048 assert(a >= 0);
3049
3050 /* We won't allow dependencies on ourselves. We will not consider them an error however. */
3051 if (u == other) {
3052 unit_maybe_warn_about_dependency(original_u, original_other->id, d);
3053 return 0;
3054 }
3055
3056 /* Note that ordering a device unit after a unit is permitted since it allows to start its job
3057 * running timeout at a specific time. */
3058 if (FLAGS_SET(a, UNIT_ATOM_BEFORE) && other->type == UNIT_DEVICE) {
3059 log_unit_warning(u, "Dependency Before=%s ignored (.device units cannot be delayed)", other->id);
3060 return 0;
3061 }
3062
3063 if (FLAGS_SET(a, UNIT_ATOM_ON_FAILURE) && !UNIT_VTABLE(u)->can_fail) {
3064 log_unit_warning(u, "Requested dependency OnFailure=%s ignored (%s units cannot fail).", other->id, unit_type_to_string(u->type));
3065 return 0;
3066 }
3067
3068 if (FLAGS_SET(a, UNIT_ATOM_TRIGGERS) && !UNIT_VTABLE(u)->can_trigger)
3069 return log_unit_error_errno(u, SYNTHETIC_ERRNO(EINVAL),
3070 "Requested dependency Triggers=%s refused (%s units cannot trigger other units).", other->id, unit_type_to_string(u->type));
3071 if (FLAGS_SET(a, UNIT_ATOM_TRIGGERED_BY) && !UNIT_VTABLE(other)->can_trigger)
3072 return log_unit_error_errno(u, SYNTHETIC_ERRNO(EINVAL),
3073 "Requested dependency TriggeredBy=%s refused (%s units cannot trigger other units).", other->id, unit_type_to_string(other->type));
3074
3075 if (FLAGS_SET(a, UNIT_ATOM_IN_SLICE) && other->type != UNIT_SLICE)
3076 return log_unit_error_errno(u, SYNTHETIC_ERRNO(EINVAL),
3077 "Requested dependency Slice=%s refused (%s is not a slice unit).", other->id, other->id);
3078 if (FLAGS_SET(a, UNIT_ATOM_SLICE_OF) && u->type != UNIT_SLICE)
3079 return log_unit_error_errno(u, SYNTHETIC_ERRNO(EINVAL),
3080 "Requested dependency SliceOf=%s refused (%s is not a slice unit).", other->id, u->id);
3081
3082 if (FLAGS_SET(a, UNIT_ATOM_IN_SLICE) && !UNIT_HAS_CGROUP_CONTEXT(u))
3083 return log_unit_error_errno(u, SYNTHETIC_ERRNO(EINVAL),
3084 "Requested dependency Slice=%s refused (%s is not a cgroup unit).", other->id, u->id);
3085
3086 if (FLAGS_SET(a, UNIT_ATOM_SLICE_OF) && !UNIT_HAS_CGROUP_CONTEXT(other))
3087 return log_unit_error_errno(u, SYNTHETIC_ERRNO(EINVAL),
3088 "Requested dependency SliceOf=%s refused (%s is not a cgroup unit).", other->id, other->id);
3089
3090 r = unit_add_dependency_hashmap(&u->dependencies, d, other, mask, 0);
3091 if (r < 0)
3092 return r;
3093 noop = !r;
3094
3095 if (inverse_table[d] != _UNIT_DEPENDENCY_INVALID && inverse_table[d] != d) {
3096 r = unit_add_dependency_hashmap(&other->dependencies, inverse_table[d], u, 0, mask);
3097 if (r < 0)
3098 return r;
3099 if (r)
3100 noop = false;
3101 }
3102
3103 if (add_reference) {
3104 r = unit_add_dependency_hashmap(&u->dependencies, UNIT_REFERENCES, other, mask, 0);
3105 if (r < 0)
3106 return r;
3107 if (r)
3108 noop = false;
3109
3110 r = unit_add_dependency_hashmap(&other->dependencies, UNIT_REFERENCED_BY, u, 0, mask);
3111 if (r < 0)
3112 return r;
3113 if (r)
3114 noop = false;
3115 }
3116
3117 if (!noop)
3118 unit_add_to_dbus_queue(u);
3119
3120 return 0;
3121 }
3122
3123 int unit_add_two_dependencies(Unit *u, UnitDependency d, UnitDependency e, Unit *other, bool add_reference, UnitDependencyMask mask) {
3124 int r;
3125
3126 assert(u);
3127
3128 r = unit_add_dependency(u, d, other, add_reference, mask);
3129 if (r < 0)
3130 return r;
3131
3132 return unit_add_dependency(u, e, other, add_reference, mask);
3133 }
3134
3135 static int resolve_template(Unit *u, const char *name, char **buf, const char **ret) {
3136 int r;
3137
3138 assert(u);
3139 assert(name);
3140 assert(buf);
3141 assert(ret);
3142
3143 if (!unit_name_is_valid(name, UNIT_NAME_TEMPLATE)) {
3144 *buf = NULL;
3145 *ret = name;
3146 return 0;
3147 }
3148
3149 if (u->instance)
3150 r = unit_name_replace_instance(name, u->instance, buf);
3151 else {
3152 _cleanup_free_ char *i = NULL;
3153
3154 r = unit_name_to_prefix(u->id, &i);
3155 if (r < 0)
3156 return r;
3157
3158 r = unit_name_replace_instance(name, i, buf);
3159 }
3160 if (r < 0)
3161 return r;
3162
3163 *ret = *buf;
3164 return 0;
3165 }
3166
3167 int unit_add_dependency_by_name(Unit *u, UnitDependency d, const char *name, bool add_reference, UnitDependencyMask mask) {
3168 _cleanup_free_ char *buf = NULL;
3169 Unit *other;
3170 int r;
3171
3172 assert(u);
3173 assert(name);
3174
3175 r = resolve_template(u, name, &buf, &name);
3176 if (r < 0)
3177 return r;
3178
3179 r = manager_load_unit(u->manager, name, NULL, NULL, &other);
3180 if (r < 0)
3181 return r;
3182
3183 return unit_add_dependency(u, d, other, add_reference, mask);
3184 }
3185
3186 int unit_add_two_dependencies_by_name(Unit *u, UnitDependency d, UnitDependency e, const char *name, bool add_reference, UnitDependencyMask mask) {
3187 _cleanup_free_ char *buf = NULL;
3188 Unit *other;
3189 int r;
3190
3191 assert(u);
3192 assert(name);
3193
3194 r = resolve_template(u, name, &buf, &name);
3195 if (r < 0)
3196 return r;
3197
3198 r = manager_load_unit(u->manager, name, NULL, NULL, &other);
3199 if (r < 0)
3200 return r;
3201
3202 return unit_add_two_dependencies(u, d, e, other, add_reference, mask);
3203 }
3204
3205 int set_unit_path(const char *p) {
3206 /* This is mostly for debug purposes */
3207 if (setenv("SYSTEMD_UNIT_PATH", p, 1) < 0)
3208 return -errno;
3209
3210 return 0;
3211 }
3212
3213 char *unit_dbus_path(Unit *u) {
3214 assert(u);
3215
3216 if (!u->id)
3217 return NULL;
3218
3219 return unit_dbus_path_from_name(u->id);
3220 }
3221
3222 char *unit_dbus_path_invocation_id(Unit *u) {
3223 assert(u);
3224
3225 if (sd_id128_is_null(u->invocation_id))
3226 return NULL;
3227
3228 return unit_dbus_path_from_name(u->invocation_id_string);
3229 }
3230
3231 int unit_set_invocation_id(Unit *u, sd_id128_t id) {
3232 int r;
3233
3234 assert(u);
3235
3236 /* Set the invocation ID for this unit. If we cannot, this will not roll back, but reset the whole thing. */
3237
3238 if (sd_id128_equal(u->invocation_id, id))
3239 return 0;
3240
3241 if (!sd_id128_is_null(u->invocation_id))
3242 (void) hashmap_remove_value(u->manager->units_by_invocation_id, &u->invocation_id, u);
3243
3244 if (sd_id128_is_null(id)) {
3245 r = 0;
3246 goto reset;
3247 }
3248
3249 r = hashmap_ensure_allocated(&u->manager->units_by_invocation_id, &id128_hash_ops);
3250 if (r < 0)
3251 goto reset;
3252
3253 u->invocation_id = id;
3254 sd_id128_to_string(id, u->invocation_id_string);
3255
3256 r = hashmap_put(u->manager->units_by_invocation_id, &u->invocation_id, u);
3257 if (r < 0)
3258 goto reset;
3259
3260 return 0;
3261
3262 reset:
3263 u->invocation_id = SD_ID128_NULL;
3264 u->invocation_id_string[0] = 0;
3265 return r;
3266 }
3267
3268 int unit_set_slice(Unit *u, Unit *slice, UnitDependencyMask mask) {
3269 int r;
3270
3271 assert(u);
3272 assert(slice);
3273
3274 /* Sets the unit slice if it has not been set before. Is extra careful, to only allow this for units
3275 * that actually have a cgroup context. Also, we don't allow to set this for slices (since the parent
3276 * slice is derived from the name). Make sure the unit we set is actually a slice. */
3277
3278 if (!UNIT_HAS_CGROUP_CONTEXT(u))
3279 return -EOPNOTSUPP;
3280
3281 if (u->type == UNIT_SLICE)
3282 return -EINVAL;
3283
3284 if (unit_active_state(u) != UNIT_INACTIVE)
3285 return -EBUSY;
3286
3287 if (slice->type != UNIT_SLICE)
3288 return -EINVAL;
3289
3290 if (unit_has_name(u, SPECIAL_INIT_SCOPE) &&
3291 !unit_has_name(slice, SPECIAL_ROOT_SLICE))
3292 return -EPERM;
3293
3294 if (UNIT_GET_SLICE(u) == slice)
3295 return 0;
3296
3297 /* Disallow slice changes if @u is already bound to cgroups */
3298 if (UNIT_GET_SLICE(u) && u->cgroup_realized)
3299 return -EBUSY;
3300
3301 r = unit_add_dependency(u, UNIT_IN_SLICE, slice, true, mask);
3302 if (r < 0)
3303 return r;
3304
3305 return 1;
3306 }
3307
3308 int unit_set_default_slice(Unit *u) {
3309 const char *slice_name;
3310 Unit *slice;
3311 int r;
3312
3313 assert(u);
3314
3315 if (UNIT_GET_SLICE(u))
3316 return 0;
3317
3318 if (u->instance) {
3319 _cleanup_free_ char *prefix = NULL, *escaped = NULL;
3320
3321 /* Implicitly place all instantiated units in their
3322 * own per-template slice */
3323
3324 r = unit_name_to_prefix(u->id, &prefix);
3325 if (r < 0)
3326 return r;
3327
3328 /* The prefix is already escaped, but it might include
3329 * "-" which has a special meaning for slice units,
3330 * hence escape it here extra. */
3331 escaped = unit_name_escape(prefix);
3332 if (!escaped)
3333 return -ENOMEM;
3334
3335 if (MANAGER_IS_SYSTEM(u->manager))
3336 slice_name = strjoina("system-", escaped, ".slice");
3337 else
3338 slice_name = strjoina("app-", escaped, ".slice");
3339
3340 } else if (unit_is_extrinsic(u))
3341 /* Keep all extrinsic units (e.g. perpetual units and swap and mount units in user mode) in
3342 * the root slice. They don't really belong in one of the subslices. */
3343 slice_name = SPECIAL_ROOT_SLICE;
3344
3345 else if (MANAGER_IS_SYSTEM(u->manager))
3346 slice_name = SPECIAL_SYSTEM_SLICE;
3347 else
3348 slice_name = SPECIAL_APP_SLICE;
3349
3350 r = manager_load_unit(u->manager, slice_name, NULL, NULL, &slice);
3351 if (r < 0)
3352 return r;
3353
3354 return unit_set_slice(u, slice, UNIT_DEPENDENCY_FILE);
3355 }
3356
3357 const char *unit_slice_name(Unit *u) {
3358 Unit *slice;
3359 assert(u);
3360
3361 slice = UNIT_GET_SLICE(u);
3362 if (!slice)
3363 return NULL;
3364
3365 return slice->id;
3366 }
3367
3368 int unit_load_related_unit(Unit *u, const char *type, Unit **_found) {
3369 _cleanup_free_ char *t = NULL;
3370 int r;
3371
3372 assert(u);
3373 assert(type);
3374 assert(_found);
3375
3376 r = unit_name_change_suffix(u->id, type, &t);
3377 if (r < 0)
3378 return r;
3379 if (unit_has_name(u, t))
3380 return -EINVAL;
3381
3382 r = manager_load_unit(u->manager, t, NULL, NULL, _found);
3383 assert(r < 0 || *_found != u);
3384 return r;
3385 }
3386
3387 static int signal_name_owner_changed(sd_bus_message *message, void *userdata, sd_bus_error *error) {
3388 const char *new_owner;
3389 Unit *u = userdata;
3390 int r;
3391
3392 assert(message);
3393 assert(u);
3394
3395 r = sd_bus_message_read(message, "sss", NULL, NULL, &new_owner);
3396 if (r < 0) {
3397 bus_log_parse_error(r);
3398 return 0;
3399 }
3400
3401 if (UNIT_VTABLE(u)->bus_name_owner_change)
3402 UNIT_VTABLE(u)->bus_name_owner_change(u, empty_to_null(new_owner));
3403
3404 return 0;
3405 }
3406
3407 static int get_name_owner_handler(sd_bus_message *message, void *userdata, sd_bus_error *error) {
3408 const sd_bus_error *e;
3409 const char *new_owner;
3410 Unit *u = userdata;
3411 int r;
3412
3413 assert(message);
3414 assert(u);
3415
3416 u->get_name_owner_slot = sd_bus_slot_unref(u->get_name_owner_slot);
3417
3418 e = sd_bus_message_get_error(message);
3419 if (e) {
3420 if (!sd_bus_error_has_name(e, "org.freedesktop.DBus.Error.NameHasNoOwner"))
3421 log_unit_error(u, "Unexpected error response from GetNameOwner(): %s", e->message);
3422
3423 new_owner = NULL;
3424 } else {
3425 r = sd_bus_message_read(message, "s", &new_owner);
3426 if (r < 0)
3427 return bus_log_parse_error(r);
3428
3429 assert(!isempty(new_owner));
3430 }
3431
3432 if (UNIT_VTABLE(u)->bus_name_owner_change)
3433 UNIT_VTABLE(u)->bus_name_owner_change(u, new_owner);
3434
3435 return 0;
3436 }
3437
3438 int unit_install_bus_match(Unit *u, sd_bus *bus, const char *name) {
3439 const char *match;
3440 int r;
3441
3442 assert(u);
3443 assert(bus);
3444 assert(name);
3445
3446 if (u->match_bus_slot || u->get_name_owner_slot)
3447 return -EBUSY;
3448
3449 match = strjoina("type='signal',"
3450 "sender='org.freedesktop.DBus',"
3451 "path='/org/freedesktop/DBus',"
3452 "interface='org.freedesktop.DBus',"
3453 "member='NameOwnerChanged',"
3454 "arg0='", name, "'");
3455
3456 r = sd_bus_add_match_async(bus, &u->match_bus_slot, match, signal_name_owner_changed, NULL, u);
3457 if (r < 0)
3458 return r;
3459
3460 r = sd_bus_call_method_async(
3461 bus,
3462 &u->get_name_owner_slot,
3463 "org.freedesktop.DBus",
3464 "/org/freedesktop/DBus",
3465 "org.freedesktop.DBus",
3466 "GetNameOwner",
3467 get_name_owner_handler,
3468 u,
3469 "s", name);
3470 if (r < 0) {
3471 u->match_bus_slot = sd_bus_slot_unref(u->match_bus_slot);
3472 return r;
3473 }
3474
3475 log_unit_debug(u, "Watching D-Bus name '%s'.", name);
3476 return 0;
3477 }
3478
3479 int unit_watch_bus_name(Unit *u, const char *name) {
3480 int r;
3481
3482 assert(u);
3483 assert(name);
3484
3485 /* Watch a specific name on the bus. We only support one unit
3486 * watching each name for now. */
3487
3488 if (u->manager->api_bus) {
3489 /* If the bus is already available, install the match directly.
3490 * Otherwise, just put the name in the list. bus_setup_api() will take care later. */
3491 r = unit_install_bus_match(u, u->manager->api_bus, name);
3492 if (r < 0)
3493 return log_warning_errno(r, "Failed to subscribe to NameOwnerChanged signal for '%s': %m", name);
3494 }
3495
3496 r = hashmap_put(u->manager->watch_bus, name, u);
3497 if (r < 0) {
3498 u->match_bus_slot = sd_bus_slot_unref(u->match_bus_slot);
3499 u->get_name_owner_slot = sd_bus_slot_unref(u->get_name_owner_slot);
3500 return log_warning_errno(r, "Failed to put bus name to hashmap: %m");
3501 }
3502
3503 return 0;
3504 }
3505
3506 void unit_unwatch_bus_name(Unit *u, const char *name) {
3507 assert(u);
3508 assert(name);
3509
3510 (void) hashmap_remove_value(u->manager->watch_bus, name, u);
3511 u->match_bus_slot = sd_bus_slot_unref(u->match_bus_slot);
3512 u->get_name_owner_slot = sd_bus_slot_unref(u->get_name_owner_slot);
3513 }
3514
3515 int unit_add_node_dependency(Unit *u, const char *what, UnitDependency dep, UnitDependencyMask mask) {
3516 _cleanup_free_ char *e = NULL;
3517 Unit *device;
3518 int r;
3519
3520 assert(u);
3521
3522 /* Adds in links to the device node that this unit is based on */
3523 if (isempty(what))
3524 return 0;
3525
3526 if (!is_device_path(what))
3527 return 0;
3528
3529 /* When device units aren't supported (such as in a container), don't create dependencies on them. */
3530 if (!unit_type_supported(UNIT_DEVICE))
3531 return 0;
3532
3533 r = unit_name_from_path(what, ".device", &e);
3534 if (r < 0)
3535 return r;
3536
3537 r = manager_load_unit(u->manager, e, NULL, NULL, &device);
3538 if (r < 0)
3539 return r;
3540
3541 if (dep == UNIT_REQUIRES && device_shall_be_bound_by(device, u))
3542 dep = UNIT_BINDS_TO;
3543
3544 return unit_add_two_dependencies(u, UNIT_AFTER,
3545 MANAGER_IS_SYSTEM(u->manager) ? dep : UNIT_WANTS,
3546 device, true, mask);
3547 }
3548
3549 int unit_add_blockdev_dependency(Unit *u, const char *what, UnitDependencyMask mask) {
3550 _cleanup_free_ char *escaped = NULL, *target = NULL;
3551 int r;
3552
3553 assert(u);
3554
3555 if (isempty(what))
3556 return 0;
3557
3558 if (!path_startswith(what, "/dev/"))
3559 return 0;
3560
3561 /* If we don't support devices, then also don't bother with blockdev@.target */
3562 if (!unit_type_supported(UNIT_DEVICE))
3563 return 0;
3564
3565 r = unit_name_path_escape(what, &escaped);
3566 if (r < 0)
3567 return r;
3568
3569 r = unit_name_build("blockdev", escaped, ".target", &target);
3570 if (r < 0)
3571 return r;
3572
3573 return unit_add_dependency_by_name(u, UNIT_AFTER, target, true, mask);
3574 }
3575
3576 int unit_coldplug(Unit *u) {
3577 int r = 0, q;
3578 char **i;
3579 Job *uj;
3580
3581 assert(u);
3582
3583 /* Make sure we don't enter a loop, when coldplugging recursively. */
3584 if (u->coldplugged)
3585 return 0;
3586
3587 u->coldplugged = true;
3588
3589 STRV_FOREACH(i, u->deserialized_refs) {
3590 q = bus_unit_track_add_name(u, *i);
3591 if (q < 0 && r >= 0)
3592 r = q;
3593 }
3594 u->deserialized_refs = strv_free(u->deserialized_refs);
3595
3596 if (UNIT_VTABLE(u)->coldplug) {
3597 q = UNIT_VTABLE(u)->coldplug(u);
3598 if (q < 0 && r >= 0)
3599 r = q;
3600 }
3601
3602 uj = u->job ?: u->nop_job;
3603 if (uj) {
3604 q = job_coldplug(uj);
3605 if (q < 0 && r >= 0)
3606 r = q;
3607 }
3608
3609 return r;
3610 }
3611
3612 void unit_catchup(Unit *u) {
3613 assert(u);
3614
3615 if (UNIT_VTABLE(u)->catchup)
3616 UNIT_VTABLE(u)->catchup(u);
3617
3618 unit_cgroup_catchup(u);
3619 }
3620
3621 static bool fragment_mtime_newer(const char *path, usec_t mtime, bool path_masked) {
3622 struct stat st;
3623
3624 if (!path)
3625 return false;
3626
3627 /* If the source is some virtual kernel file system, then we assume we watch it anyway, and hence pretend we
3628 * are never out-of-date. */
3629 if (PATH_STARTSWITH_SET(path, "/proc", "/sys"))
3630 return false;
3631
3632 if (stat(path, &st) < 0)
3633 /* What, cannot access this anymore? */
3634 return true;
3635
3636 if (path_masked)
3637 /* For masked files check if they are still so */
3638 return !null_or_empty(&st);
3639 else
3640 /* For non-empty files check the mtime */
3641 return timespec_load(&st.st_mtim) > mtime;
3642
3643 return false;
3644 }
3645
3646 bool unit_need_daemon_reload(Unit *u) {
3647 _cleanup_strv_free_ char **t = NULL;
3648 char **path;
3649
3650 assert(u);
3651
3652 /* For unit files, we allow masking… */
3653 if (fragment_mtime_newer(u->fragment_path, u->fragment_mtime,
3654 u->load_state == UNIT_MASKED))
3655 return true;
3656
3657 /* Source paths should not be masked… */
3658 if (fragment_mtime_newer(u->source_path, u->source_mtime, false))
3659 return true;
3660
3661 if (u->load_state == UNIT_LOADED)
3662 (void) unit_find_dropin_paths(u, &t);
3663 if (!strv_equal(u->dropin_paths, t))
3664 return true;
3665
3666 /* … any drop-ins that are masked are simply omitted from the list. */
3667 STRV_FOREACH(path, u->dropin_paths)
3668 if (fragment_mtime_newer(*path, u->dropin_mtime, false))
3669 return true;
3670
3671 return false;
3672 }
3673
3674 void unit_reset_failed(Unit *u) {
3675 assert(u);
3676
3677 if (UNIT_VTABLE(u)->reset_failed)
3678 UNIT_VTABLE(u)->reset_failed(u);
3679
3680 ratelimit_reset(&u->start_ratelimit);
3681 u->start_limit_hit = false;
3682 }
3683
3684 Unit *unit_following(Unit *u) {
3685 assert(u);
3686
3687 if (UNIT_VTABLE(u)->following)
3688 return UNIT_VTABLE(u)->following(u);
3689
3690 return NULL;
3691 }
3692
3693 bool unit_stop_pending(Unit *u) {
3694 assert(u);
3695
3696 /* This call does check the current state of the unit. It's
3697 * hence useful to be called from state change calls of the
3698 * unit itself, where the state isn't updated yet. This is
3699 * different from unit_inactive_or_pending() which checks both
3700 * the current state and for a queued job. */
3701
3702 return unit_has_job_type(u, JOB_STOP);
3703 }
3704
3705 bool unit_inactive_or_pending(Unit *u) {
3706 assert(u);
3707
3708 /* Returns true if the unit is inactive or going down */
3709
3710 if (UNIT_IS_INACTIVE_OR_DEACTIVATING(unit_active_state(u)))
3711 return true;
3712
3713 if (unit_stop_pending(u))
3714 return true;
3715
3716 return false;
3717 }
3718
3719 bool unit_active_or_pending(Unit *u) {
3720 assert(u);
3721
3722 /* Returns true if the unit is active or going up */
3723
3724 if (UNIT_IS_ACTIVE_OR_ACTIVATING(unit_active_state(u)))
3725 return true;
3726
3727 if (u->job &&
3728 IN_SET(u->job->type, JOB_START, JOB_RELOAD_OR_START, JOB_RESTART))
3729 return true;
3730
3731 return false;
3732 }
3733
3734 bool unit_will_restart_default(Unit *u) {
3735 assert(u);
3736
3737 return unit_has_job_type(u, JOB_START);
3738 }
3739
3740 bool unit_will_restart(Unit *u) {
3741 assert(u);
3742
3743 if (!UNIT_VTABLE(u)->will_restart)
3744 return false;
3745
3746 return UNIT_VTABLE(u)->will_restart(u);
3747 }
3748
3749 int unit_kill(Unit *u, KillWho w, int signo, sd_bus_error *error) {
3750 assert(u);
3751 assert(w >= 0 && w < _KILL_WHO_MAX);
3752 assert(SIGNAL_VALID(signo));
3753
3754 if (!UNIT_VTABLE(u)->kill)
3755 return -EOPNOTSUPP;
3756
3757 return UNIT_VTABLE(u)->kill(u, w, signo, error);
3758 }
3759
3760 static Set *unit_pid_set(pid_t main_pid, pid_t control_pid) {
3761 _cleanup_set_free_ Set *pid_set = NULL;
3762 int r;
3763
3764 pid_set = set_new(NULL);
3765 if (!pid_set)
3766 return NULL;
3767
3768 /* Exclude the main/control pids from being killed via the cgroup */
3769 if (main_pid > 0) {
3770 r = set_put(pid_set, PID_TO_PTR(main_pid));
3771 if (r < 0)
3772 return NULL;
3773 }
3774
3775 if (control_pid > 0) {
3776 r = set_put(pid_set, PID_TO_PTR(control_pid));
3777 if (r < 0)
3778 return NULL;
3779 }
3780
3781 return TAKE_PTR(pid_set);
3782 }
3783
3784 static int kill_common_log(pid_t pid, int signo, void *userdata) {
3785 _cleanup_free_ char *comm = NULL;
3786 Unit *u = userdata;
3787
3788 assert(u);
3789
3790 (void) get_process_comm(pid, &comm);
3791 log_unit_info(u, "Sending signal SIG%s to process " PID_FMT " (%s) on client request.",
3792 signal_to_string(signo), pid, strna(comm));
3793
3794 return 1;
3795 }
3796
3797 int unit_kill_common(
3798 Unit *u,
3799 KillWho who,
3800 int signo,
3801 pid_t main_pid,
3802 pid_t control_pid,
3803 sd_bus_error *error) {
3804
3805 int r = 0;
3806 bool killed = false;
3807
3808 /* This is the common implementation for explicit user-requested killing of unit processes, shared by
3809 * various unit types. Do not confuse with unit_kill_context(), which is what we use when we want to
3810 * stop a service ourselves. */
3811
3812 if (IN_SET(who, KILL_MAIN, KILL_MAIN_FAIL)) {
3813 if (main_pid < 0)
3814 return sd_bus_error_setf(error, BUS_ERROR_NO_SUCH_PROCESS, "%s units have no main processes", unit_type_to_string(u->type));
3815 if (main_pid == 0)
3816 return sd_bus_error_set_const(error, BUS_ERROR_NO_SUCH_PROCESS, "No main process to kill");
3817 }
3818
3819 if (IN_SET(who, KILL_CONTROL, KILL_CONTROL_FAIL)) {
3820 if (control_pid < 0)
3821 return sd_bus_error_setf(error, BUS_ERROR_NO_SUCH_PROCESS, "%s units have no control processes", unit_type_to_string(u->type));
3822 if (control_pid == 0)
3823 return sd_bus_error_set_const(error, BUS_ERROR_NO_SUCH_PROCESS, "No control process to kill");
3824 }
3825
3826 if (IN_SET(who, KILL_CONTROL, KILL_CONTROL_FAIL, KILL_ALL, KILL_ALL_FAIL))
3827 if (control_pid > 0) {
3828 _cleanup_free_ char *comm = NULL;
3829 (void) get_process_comm(control_pid, &comm);
3830
3831 if (kill(control_pid, signo) < 0) {
3832 /* Report this failure both to the logs and to the client */
3833 sd_bus_error_set_errnof(
3834 error, errno,
3835 "Failed to send signal SIG%s to control process " PID_FMT " (%s): %m",
3836 signal_to_string(signo), control_pid, strna(comm));
3837 r = log_unit_warning_errno(
3838 u, errno,
3839 "Failed to send signal SIG%s to control process " PID_FMT " (%s) on client request: %m",
3840 signal_to_string(signo), control_pid, strna(comm));
3841 } else {
3842 log_unit_info(u, "Sent signal SIG%s to control process " PID_FMT " (%s) on client request.",
3843 signal_to_string(signo), control_pid, strna(comm));
3844 killed = true;
3845 }
3846 }
3847
3848 if (IN_SET(who, KILL_MAIN, KILL_MAIN_FAIL, KILL_ALL, KILL_ALL_FAIL))
3849 if (main_pid > 0) {
3850 _cleanup_free_ char *comm = NULL;
3851 (void) get_process_comm(main_pid, &comm);
3852
3853 if (kill(main_pid, signo) < 0) {
3854 if (r == 0)
3855 sd_bus_error_set_errnof(
3856 error, errno,
3857 "Failed to send signal SIG%s to main process " PID_FMT " (%s): %m",
3858 signal_to_string(signo), main_pid, strna(comm));
3859
3860 r = log_unit_warning_errno(
3861 u, errno,
3862 "Failed to send signal SIG%s to main process " PID_FMT " (%s) on client request: %m",
3863 signal_to_string(signo), main_pid, strna(comm));
3864 } else {
3865 log_unit_info(u, "Sent signal SIG%s to main process " PID_FMT " (%s) on client request.",
3866 signal_to_string(signo), main_pid, strna(comm));
3867 killed = true;
3868 }
3869 }
3870
3871 if (IN_SET(who, KILL_ALL, KILL_ALL_FAIL) && u->cgroup_path) {
3872 _cleanup_set_free_ Set *pid_set = NULL;
3873 int q;
3874
3875 /* Exclude the main/control pids from being killed via the cgroup */
3876 pid_set = unit_pid_set(main_pid, control_pid);
3877 if (!pid_set)
3878 return log_oom();
3879
3880 q = cg_kill_recursive(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, signo, 0, pid_set, kill_common_log, u);
3881 if (q < 0) {
3882 if (!IN_SET(q, -ESRCH, -ENOENT)) {
3883 if (r == 0)
3884 sd_bus_error_set_errnof(
3885 error, q,
3886 "Failed to send signal SIG%s to auxiliary processes: %m",
3887 signal_to_string(signo));
3888
3889 r = log_unit_warning_errno(
3890 u, q,
3891 "Failed to send signal SIG%s to auxiliary processes on client request: %m",
3892 signal_to_string(signo));
3893 }
3894 } else
3895 killed = true;
3896 }
3897
3898 /* If the "fail" versions of the operation are requested, then complain if the set of processes we killed is empty */
3899 if (r == 0 && !killed && IN_SET(who, KILL_ALL_FAIL, KILL_CONTROL_FAIL, KILL_MAIN_FAIL))
3900 return sd_bus_error_set_const(error, BUS_ERROR_NO_SUCH_PROCESS, "No matching processes to kill");
3901
3902 return r;
3903 }
3904
3905 int unit_following_set(Unit *u, Set **s) {
3906 assert(u);
3907 assert(s);
3908
3909 if (UNIT_VTABLE(u)->following_set)
3910 return UNIT_VTABLE(u)->following_set(u, s);
3911
3912 *s = NULL;
3913 return 0;
3914 }
3915
3916 UnitFileState unit_get_unit_file_state(Unit *u) {
3917 int r;
3918
3919 assert(u);
3920
3921 if (u->unit_file_state < 0 && u->fragment_path) {
3922 r = unit_file_get_state(
3923 u->manager->unit_file_scope,
3924 NULL,
3925 u->id,
3926 &u->unit_file_state);
3927 if (r < 0)
3928 u->unit_file_state = UNIT_FILE_BAD;
3929 }
3930
3931 return u->unit_file_state;
3932 }
3933
3934 int unit_get_unit_file_preset(Unit *u) {
3935 assert(u);
3936
3937 if (u->unit_file_preset < 0 && u->fragment_path)
3938 u->unit_file_preset = unit_file_query_preset(
3939 u->manager->unit_file_scope,
3940 NULL,
3941 basename(u->fragment_path),
3942 NULL);
3943
3944 return u->unit_file_preset;
3945 }
3946
3947 Unit* unit_ref_set(UnitRef *ref, Unit *source, Unit *target) {
3948 assert(ref);
3949 assert(source);
3950 assert(target);
3951
3952 if (ref->target)
3953 unit_ref_unset(ref);
3954
3955 ref->source = source;
3956 ref->target = target;
3957 LIST_PREPEND(refs_by_target, target->refs_by_target, ref);
3958 return target;
3959 }
3960
3961 void unit_ref_unset(UnitRef *ref) {
3962 assert(ref);
3963
3964 if (!ref->target)
3965 return;
3966
3967 /* We are about to drop a reference to the unit, make sure the garbage collection has a look at it as it might
3968 * be unreferenced now. */
3969 unit_add_to_gc_queue(ref->target);
3970
3971 LIST_REMOVE(refs_by_target, ref->target->refs_by_target, ref);
3972 ref->source = ref->target = NULL;
3973 }
3974
3975 static int user_from_unit_name(Unit *u, char **ret) {
3976
3977 static const uint8_t hash_key[] = {
3978 0x58, 0x1a, 0xaf, 0xe6, 0x28, 0x58, 0x4e, 0x96,
3979 0xb4, 0x4e, 0xf5, 0x3b, 0x8c, 0x92, 0x07, 0xec
3980 };
3981
3982 _cleanup_free_ char *n = NULL;
3983 int r;
3984
3985 r = unit_name_to_prefix(u->id, &n);
3986 if (r < 0)
3987 return r;
3988
3989 if (valid_user_group_name(n, 0)) {
3990 *ret = TAKE_PTR(n);
3991 return 0;
3992 }
3993
3994 /* If we can't use the unit name as a user name, then let's hash it and use that */
3995 if (asprintf(ret, "_du%016" PRIx64, siphash24(n, strlen(n), hash_key)) < 0)
3996 return -ENOMEM;
3997
3998 return 0;
3999 }
4000
4001 int unit_patch_contexts(Unit *u) {
4002 CGroupContext *cc;
4003 ExecContext *ec;
4004 int r;
4005
4006 assert(u);
4007
4008 /* Patch in the manager defaults into the exec and cgroup
4009 * contexts, _after_ the rest of the settings have been
4010 * initialized */
4011
4012 ec = unit_get_exec_context(u);
4013 if (ec) {
4014 /* This only copies in the ones that need memory */
4015 for (unsigned i = 0; i < _RLIMIT_MAX; i++)
4016 if (u->manager->rlimit[i] && !ec->rlimit[i]) {
4017 ec->rlimit[i] = newdup(struct rlimit, u->manager->rlimit[i], 1);
4018 if (!ec->rlimit[i])
4019 return -ENOMEM;
4020 }
4021
4022 if (MANAGER_IS_USER(u->manager) &&
4023 !ec->working_directory) {
4024
4025 r = get_home_dir(&ec->working_directory);
4026 if (r < 0)
4027 return r;
4028
4029 /* Allow user services to run, even if the
4030 * home directory is missing */
4031 ec->working_directory_missing_ok = true;
4032 }
4033
4034 if (ec->private_devices)
4035 ec->capability_bounding_set &= ~((UINT64_C(1) << CAP_MKNOD) | (UINT64_C(1) << CAP_SYS_RAWIO));
4036
4037 if (ec->protect_kernel_modules)
4038 ec->capability_bounding_set &= ~(UINT64_C(1) << CAP_SYS_MODULE);
4039
4040 if (ec->protect_kernel_logs)
4041 ec->capability_bounding_set &= ~(UINT64_C(1) << CAP_SYSLOG);
4042
4043 if (ec->protect_clock)
4044 ec->capability_bounding_set &= ~((UINT64_C(1) << CAP_SYS_TIME) | (UINT64_C(1) << CAP_WAKE_ALARM));
4045
4046 if (ec->dynamic_user) {
4047 if (!ec->user) {
4048 r = user_from_unit_name(u, &ec->user);
4049 if (r < 0)
4050 return r;
4051 }
4052
4053 if (!ec->group) {
4054 ec->group = strdup(ec->user);
4055 if (!ec->group)
4056 return -ENOMEM;
4057 }
4058
4059 /* If the dynamic user option is on, let's make sure that the unit can't leave its
4060 * UID/GID around in the file system or on IPC objects. Hence enforce a strict
4061 * sandbox. */
4062
4063 ec->private_tmp = true;
4064 ec->remove_ipc = true;
4065 ec->protect_system = PROTECT_SYSTEM_STRICT;
4066 if (ec->protect_home == PROTECT_HOME_NO)
4067 ec->protect_home = PROTECT_HOME_READ_ONLY;
4068
4069 /* Make sure this service can neither benefit from SUID/SGID binaries nor create
4070 * them. */
4071 ec->no_new_privileges = true;
4072 ec->restrict_suid_sgid = true;
4073 }
4074 }
4075
4076 cc = unit_get_cgroup_context(u);
4077 if (cc && ec) {
4078
4079 if (ec->private_devices &&
4080 cc->device_policy == CGROUP_DEVICE_POLICY_AUTO)
4081 cc->device_policy = CGROUP_DEVICE_POLICY_CLOSED;
4082
4083 if ((ec->root_image || !LIST_IS_EMPTY(ec->mount_images)) &&
4084 (cc->device_policy != CGROUP_DEVICE_POLICY_AUTO || cc->device_allow)) {
4085 const char *p;
4086
4087 /* When RootImage= or MountImages= is specified, the following devices are touched. */
4088 FOREACH_STRING(p, "/dev/loop-control", "/dev/mapper/control") {
4089 r = cgroup_add_device_allow(cc, p, "rw");
4090 if (r < 0)
4091 return r;
4092 }
4093 FOREACH_STRING(p, "block-loop", "block-blkext", "block-device-mapper") {
4094 r = cgroup_add_device_allow(cc, p, "rwm");
4095 if (r < 0)
4096 return r;
4097 }
4098
4099 /* Make sure "block-loop" can be resolved, i.e. make sure "loop" shows up in /proc/devices.
4100 * Same for mapper and verity. */
4101 FOREACH_STRING(p, "modprobe@loop.service", "modprobe@dm_mod.service", "modprobe@dm_verity.service") {
4102 r = unit_add_two_dependencies_by_name(u, UNIT_AFTER, UNIT_WANTS, p, true, UNIT_DEPENDENCY_FILE);
4103 if (r < 0)
4104 return r;
4105 }
4106 }
4107
4108 if (ec->protect_clock) {
4109 r = cgroup_add_device_allow(cc, "char-rtc", "r");
4110 if (r < 0)
4111 return r;
4112 }
4113 }
4114
4115 return 0;
4116 }
4117
4118 ExecContext *unit_get_exec_context(const Unit *u) {
4119 size_t offset;
4120 assert(u);
4121
4122 if (u->type < 0)
4123 return NULL;
4124
4125 offset = UNIT_VTABLE(u)->exec_context_offset;
4126 if (offset <= 0)
4127 return NULL;
4128
4129 return (ExecContext*) ((uint8_t*) u + offset);
4130 }
4131
4132 KillContext *unit_get_kill_context(Unit *u) {
4133 size_t offset;
4134 assert(u);
4135
4136 if (u->type < 0)
4137 return NULL;
4138
4139 offset = UNIT_VTABLE(u)->kill_context_offset;
4140 if (offset <= 0)
4141 return NULL;
4142
4143 return (KillContext*) ((uint8_t*) u + offset);
4144 }
4145
4146 CGroupContext *unit_get_cgroup_context(Unit *u) {
4147 size_t offset;
4148
4149 if (u->type < 0)
4150 return NULL;
4151
4152 offset = UNIT_VTABLE(u)->cgroup_context_offset;
4153 if (offset <= 0)
4154 return NULL;
4155
4156 return (CGroupContext*) ((uint8_t*) u + offset);
4157 }
4158
4159 ExecRuntime *unit_get_exec_runtime(Unit *u) {
4160 size_t offset;
4161
4162 if (u->type < 0)
4163 return NULL;
4164
4165 offset = UNIT_VTABLE(u)->exec_runtime_offset;
4166 if (offset <= 0)
4167 return NULL;
4168
4169 return *(ExecRuntime**) ((uint8_t*) u + offset);
4170 }
4171
4172 static const char* unit_drop_in_dir(Unit *u, UnitWriteFlags flags) {
4173 assert(u);
4174
4175 if (UNIT_WRITE_FLAGS_NOOP(flags))
4176 return NULL;
4177
4178 if (u->transient) /* Redirect drop-ins for transient units always into the transient directory. */
4179 return u->manager->lookup_paths.transient;
4180
4181 if (flags & UNIT_PERSISTENT)
4182 return u->manager->lookup_paths.persistent_control;
4183
4184 if (flags & UNIT_RUNTIME)
4185 return u->manager->lookup_paths.runtime_control;
4186
4187 return NULL;
4188 }
4189
4190 char* unit_escape_setting(const char *s, UnitWriteFlags flags, char **buf) {
4191 char *ret = NULL;
4192
4193 if (!s)
4194 return NULL;
4195
4196 /* Escapes the input string as requested. Returns the escaped string. If 'buf' is specified then the allocated
4197 * return buffer pointer is also written to *buf, except if no escaping was necessary, in which case *buf is
4198 * set to NULL, and the input pointer is returned as-is. This means the return value always contains a properly
4199 * escaped version, but *buf when passed only contains a pointer if an allocation was necessary. If *buf is
4200 * not specified, then the return value always needs to be freed. Callers can use this to optimize memory
4201 * allocations. */
4202
4203 if (flags & UNIT_ESCAPE_SPECIFIERS) {
4204 ret = specifier_escape(s);
4205 if (!ret)
4206 return NULL;
4207
4208 s = ret;
4209 }
4210
4211 if (flags & UNIT_ESCAPE_C) {
4212 char *a;
4213
4214 a = cescape(s);
4215 free(ret);
4216 if (!a)
4217 return NULL;
4218
4219 ret = a;
4220 }
4221
4222 if (buf) {
4223 *buf = ret;
4224 return ret ?: (char*) s;
4225 }
4226
4227 return ret ?: strdup(s);
4228 }
4229
4230 char* unit_concat_strv(char **l, UnitWriteFlags flags) {
4231 _cleanup_free_ char *result = NULL;
4232 size_t n = 0;
4233 char **i;
4234
4235 /* Takes a list of strings, escapes them, and concatenates them. This may be used to format command lines in a
4236 * way suitable for ExecStart= stanzas */
4237
4238 STRV_FOREACH(i, l) {
4239 _cleanup_free_ char *buf = NULL;
4240 const char *p;
4241 size_t a;
4242 char *q;
4243
4244 p = unit_escape_setting(*i, flags, &buf);
4245 if (!p)
4246 return NULL;
4247
4248 a = (n > 0) + 1 + strlen(p) + 1; /* separating space + " + entry + " */
4249 if (!GREEDY_REALLOC(result, n + a + 1))
4250 return NULL;
4251
4252 q = result + n;
4253 if (n > 0)
4254 *(q++) = ' ';
4255
4256 *(q++) = '"';
4257 q = stpcpy(q, p);
4258 *(q++) = '"';
4259
4260 n += a;
4261 }
4262
4263 if (!GREEDY_REALLOC(result, n + 1))
4264 return NULL;
4265
4266 result[n] = 0;
4267
4268 return TAKE_PTR(result);
4269 }
4270
4271 int unit_write_setting(Unit *u, UnitWriteFlags flags, const char *name, const char *data) {
4272 _cleanup_free_ char *p = NULL, *q = NULL, *escaped = NULL;
4273 const char *dir, *wrapped;
4274 int r;
4275
4276 assert(u);
4277 assert(name);
4278 assert(data);
4279
4280 if (UNIT_WRITE_FLAGS_NOOP(flags))
4281 return 0;
4282
4283 data = unit_escape_setting(data, flags, &escaped);
4284 if (!data)
4285 return -ENOMEM;
4286
4287 /* Prefix the section header. If we are writing this out as transient file, then let's suppress this if the
4288 * previous section header is the same */
4289
4290 if (flags & UNIT_PRIVATE) {
4291 if (!UNIT_VTABLE(u)->private_section)
4292 return -EINVAL;
4293
4294 if (!u->transient_file || u->last_section_private < 0)
4295 data = strjoina("[", UNIT_VTABLE(u)->private_section, "]\n", data);
4296 else if (u->last_section_private == 0)
4297 data = strjoina("\n[", UNIT_VTABLE(u)->private_section, "]\n", data);
4298 } else {
4299 if (!u->transient_file || u->last_section_private < 0)
4300 data = strjoina("[Unit]\n", data);
4301 else if (u->last_section_private > 0)
4302 data = strjoina("\n[Unit]\n", data);
4303 }
4304
4305 if (u->transient_file) {
4306 /* When this is a transient unit file in creation, then let's not create a new drop-in but instead
4307 * write to the transient unit file. */
4308 fputs(data, u->transient_file);
4309
4310 if (!endswith(data, "\n"))
4311 fputc('\n', u->transient_file);
4312
4313 /* Remember which section we wrote this entry to */
4314 u->last_section_private = !!(flags & UNIT_PRIVATE);
4315 return 0;
4316 }
4317
4318 dir = unit_drop_in_dir(u, flags);
4319 if (!dir)
4320 return -EINVAL;
4321
4322 wrapped = strjoina("# This is a drop-in unit file extension, created via \"systemctl set-property\"\n"
4323 "# or an equivalent operation. Do not edit.\n",
4324 data,
4325 "\n");
4326
4327 r = drop_in_file(dir, u->id, 50, name, &p, &q);
4328 if (r < 0)
4329 return r;
4330
4331 (void) mkdir_p_label(p, 0755);
4332
4333 /* Make sure the drop-in dir is registered in our path cache. This way we don't need to stupidly
4334 * recreate the cache after every drop-in we write. */
4335 if (u->manager->unit_path_cache) {
4336 r = set_put_strdup(&u->manager->unit_path_cache, p);
4337 if (r < 0)
4338 return r;
4339 }
4340
4341 r = write_string_file_atomic_label(q, wrapped);
4342 if (r < 0)
4343 return r;
4344
4345 r = strv_push(&u->dropin_paths, q);
4346 if (r < 0)
4347 return r;
4348 q = NULL;
4349
4350 strv_uniq(u->dropin_paths);
4351
4352 u->dropin_mtime = now(CLOCK_REALTIME);
4353
4354 return 0;
4355 }
4356
4357 int unit_write_settingf(Unit *u, UnitWriteFlags flags, const char *name, const char *format, ...) {
4358 _cleanup_free_ char *p = NULL;
4359 va_list ap;
4360 int r;
4361
4362 assert(u);
4363 assert(name);
4364 assert(format);
4365
4366 if (UNIT_WRITE_FLAGS_NOOP(flags))
4367 return 0;
4368
4369 va_start(ap, format);
4370 r = vasprintf(&p, format, ap);
4371 va_end(ap);
4372
4373 if (r < 0)
4374 return -ENOMEM;
4375
4376 return unit_write_setting(u, flags, name, p);
4377 }
4378
4379 int unit_make_transient(Unit *u) {
4380 _cleanup_free_ char *path = NULL;
4381 FILE *f;
4382
4383 assert(u);
4384
4385 if (!UNIT_VTABLE(u)->can_transient)
4386 return -EOPNOTSUPP;
4387
4388 (void) mkdir_p_label(u->manager->lookup_paths.transient, 0755);
4389
4390 path = path_join(u->manager->lookup_paths.transient, u->id);
4391 if (!path)
4392 return -ENOMEM;
4393
4394 /* Let's open the file we'll write the transient settings into. This file is kept open as long as we are
4395 * creating the transient, and is closed in unit_load(), as soon as we start loading the file. */
4396
4397 RUN_WITH_UMASK(0022) {
4398 f = fopen(path, "we");
4399 if (!f)
4400 return -errno;
4401 }
4402
4403 safe_fclose(u->transient_file);
4404 u->transient_file = f;
4405
4406 free_and_replace(u->fragment_path, path);
4407
4408 u->source_path = mfree(u->source_path);
4409 u->dropin_paths = strv_free(u->dropin_paths);
4410 u->fragment_mtime = u->source_mtime = u->dropin_mtime = 0;
4411
4412 u->load_state = UNIT_STUB;
4413 u->load_error = 0;
4414 u->transient = true;
4415
4416 unit_add_to_dbus_queue(u);
4417 unit_add_to_gc_queue(u);
4418
4419 fputs("# This is a transient unit file, created programmatically via the systemd API. Do not edit.\n",
4420 u->transient_file);
4421
4422 return 0;
4423 }
4424
4425 static int log_kill(pid_t pid, int sig, void *userdata) {
4426 _cleanup_free_ char *comm = NULL;
4427
4428 (void) get_process_comm(pid, &comm);
4429
4430 /* Don't log about processes marked with brackets, under the assumption that these are temporary processes
4431 only, like for example systemd's own PAM stub process. */
4432 if (comm && comm[0] == '(')
4433 return 0;
4434
4435 log_unit_notice(userdata,
4436 "Killing process " PID_FMT " (%s) with signal SIG%s.",
4437 pid,
4438 strna(comm),
4439 signal_to_string(sig));
4440
4441 return 1;
4442 }
4443
4444 static int operation_to_signal(const KillContext *c, KillOperation k, bool *noteworthy) {
4445 assert(c);
4446
4447 switch (k) {
4448
4449 case KILL_TERMINATE:
4450 case KILL_TERMINATE_AND_LOG:
4451 *noteworthy = false;
4452 return c->kill_signal;
4453
4454 case KILL_RESTART:
4455 *noteworthy = false;
4456 return restart_kill_signal(c);
4457
4458 case KILL_KILL:
4459 *noteworthy = true;
4460 return c->final_kill_signal;
4461
4462 case KILL_WATCHDOG:
4463 *noteworthy = true;
4464 return c->watchdog_signal;
4465
4466 default:
4467 assert_not_reached();
4468 }
4469 }
4470
4471 int unit_kill_context(
4472 Unit *u,
4473 KillContext *c,
4474 KillOperation k,
4475 pid_t main_pid,
4476 pid_t control_pid,
4477 bool main_pid_alien) {
4478
4479 bool wait_for_exit = false, send_sighup;
4480 cg_kill_log_func_t log_func = NULL;
4481 int sig, r;
4482
4483 assert(u);
4484 assert(c);
4485
4486 /* Kill the processes belonging to this unit, in preparation for shutting the unit down. Returns > 0
4487 * if we killed something worth waiting for, 0 otherwise. Do not confuse with unit_kill_common()
4488 * which is used for user-requested killing of unit processes. */
4489
4490 if (c->kill_mode == KILL_NONE)
4491 return 0;
4492
4493 bool noteworthy;
4494 sig = operation_to_signal(c, k, &noteworthy);
4495 if (noteworthy)
4496 log_func = log_kill;
4497
4498 send_sighup =
4499 c->send_sighup &&
4500 IN_SET(k, KILL_TERMINATE, KILL_TERMINATE_AND_LOG) &&
4501 sig != SIGHUP;
4502
4503 if (main_pid > 0) {
4504 if (log_func)
4505 log_func(main_pid, sig, u);
4506
4507 r = kill_and_sigcont(main_pid, sig);
4508 if (r < 0 && r != -ESRCH) {
4509 _cleanup_free_ char *comm = NULL;
4510 (void) get_process_comm(main_pid, &comm);
4511
4512 log_unit_warning_errno(u, r, "Failed to kill main process " PID_FMT " (%s), ignoring: %m", main_pid, strna(comm));
4513 } else {
4514 if (!main_pid_alien)
4515 wait_for_exit = true;
4516
4517 if (r != -ESRCH && send_sighup)
4518 (void) kill(main_pid, SIGHUP);
4519 }
4520 }
4521
4522 if (control_pid > 0) {
4523 if (log_func)
4524 log_func(control_pid, sig, u);
4525
4526 r = kill_and_sigcont(control_pid, sig);
4527 if (r < 0 && r != -ESRCH) {
4528 _cleanup_free_ char *comm = NULL;
4529 (void) get_process_comm(control_pid, &comm);
4530
4531 log_unit_warning_errno(u, r, "Failed to kill control process " PID_FMT " (%s), ignoring: %m", control_pid, strna(comm));
4532 } else {
4533 wait_for_exit = true;
4534
4535 if (r != -ESRCH && send_sighup)
4536 (void) kill(control_pid, SIGHUP);
4537 }
4538 }
4539
4540 if (u->cgroup_path &&
4541 (c->kill_mode == KILL_CONTROL_GROUP || (c->kill_mode == KILL_MIXED && k == KILL_KILL))) {
4542 _cleanup_set_free_ Set *pid_set = NULL;
4543
4544 /* Exclude the main/control pids from being killed via the cgroup */
4545 pid_set = unit_pid_set(main_pid, control_pid);
4546 if (!pid_set)
4547 return -ENOMEM;
4548
4549 r = cg_kill_recursive(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path,
4550 sig,
4551 CGROUP_SIGCONT|CGROUP_IGNORE_SELF,
4552 pid_set,
4553 log_func, u);
4554 if (r < 0) {
4555 if (!IN_SET(r, -EAGAIN, -ESRCH, -ENOENT))
4556 log_unit_warning_errno(u, r, "Failed to kill control group %s, ignoring: %m", empty_to_root(u->cgroup_path));
4557
4558 } else if (r > 0) {
4559
4560 /* FIXME: For now, on the legacy hierarchy, we will not wait for the cgroup members to die if
4561 * we are running in a container or if this is a delegation unit, simply because cgroup
4562 * notification is unreliable in these cases. It doesn't work at all in containers, and outside
4563 * of containers it can be confused easily by left-over directories in the cgroup — which
4564 * however should not exist in non-delegated units. On the unified hierarchy that's different,
4565 * there we get proper events. Hence rely on them. */
4566
4567 if (cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER) > 0 ||
4568 (detect_container() == 0 && !unit_cgroup_delegate(u)))
4569 wait_for_exit = true;
4570
4571 if (send_sighup) {
4572 set_free(pid_set);
4573
4574 pid_set = unit_pid_set(main_pid, control_pid);
4575 if (!pid_set)
4576 return -ENOMEM;
4577
4578 (void) cg_kill_recursive(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path,
4579 SIGHUP,
4580 CGROUP_IGNORE_SELF,
4581 pid_set,
4582 NULL, NULL);
4583 }
4584 }
4585 }
4586
4587 return wait_for_exit;
4588 }
4589
4590 int unit_require_mounts_for(Unit *u, const char *path, UnitDependencyMask mask) {
4591 int r;
4592
4593 assert(u);
4594 assert(path);
4595
4596 /* Registers a unit for requiring a certain path and all its prefixes. We keep a hashtable of these
4597 * paths in the unit (from the path to the UnitDependencyInfo structure indicating how to the
4598 * dependency came to be). However, we build a prefix table for all possible prefixes so that new
4599 * appearing mount units can easily determine which units to make themselves a dependency of. */
4600
4601 if (!path_is_absolute(path))
4602 return -EINVAL;
4603
4604 if (hashmap_contains(u->requires_mounts_for, path)) /* Exit quickly if the path is already covered. */
4605 return 0;
4606
4607 _cleanup_free_ char *p = strdup(path);
4608 if (!p)
4609 return -ENOMEM;
4610
4611 /* Use the canonical form of the path as the stored key. We call path_is_normalized()
4612 * only after simplification, since path_is_normalized() rejects paths with '.'.
4613 * path_is_normalized() also verifies that the path fits in PATH_MAX. */
4614 path = path_simplify(p);
4615
4616 if (!path_is_normalized(path))
4617 return -EPERM;
4618
4619 UnitDependencyInfo di = {
4620 .origin_mask = mask
4621 };
4622
4623 r = hashmap_ensure_put(&u->requires_mounts_for, &path_hash_ops, p, di.data);
4624 if (r < 0)
4625 return r;
4626 assert(r > 0);
4627 TAKE_PTR(p); /* path remains a valid pointer to the string stored in the hashmap */
4628
4629 char prefix[strlen(path) + 1];
4630 PATH_FOREACH_PREFIX_MORE(prefix, path) {
4631 Set *x;
4632
4633 x = hashmap_get(u->manager->units_requiring_mounts_for, prefix);
4634 if (!x) {
4635 _cleanup_free_ char *q = NULL;
4636
4637 r = hashmap_ensure_allocated(&u->manager->units_requiring_mounts_for, &path_hash_ops);
4638 if (r < 0)
4639 return r;
4640
4641 q = strdup(prefix);
4642 if (!q)
4643 return -ENOMEM;
4644
4645 x = set_new(NULL);
4646 if (!x)
4647 return -ENOMEM;
4648
4649 r = hashmap_put(u->manager->units_requiring_mounts_for, q, x);
4650 if (r < 0) {
4651 set_free(x);
4652 return r;
4653 }
4654 q = NULL;
4655 }
4656
4657 r = set_put(x, u);
4658 if (r < 0)
4659 return r;
4660 }
4661
4662 return 0;
4663 }
4664
4665 int unit_setup_exec_runtime(Unit *u) {
4666 ExecRuntime **rt;
4667 size_t offset;
4668 Unit *other;
4669 int r;
4670
4671 offset = UNIT_VTABLE(u)->exec_runtime_offset;
4672 assert(offset > 0);
4673
4674 /* Check if there already is an ExecRuntime for this unit? */
4675 rt = (ExecRuntime**) ((uint8_t*) u + offset);
4676 if (*rt)
4677 return 0;
4678
4679 /* Try to get it from somebody else */
4680 UNIT_FOREACH_DEPENDENCY(other, u, UNIT_ATOM_JOINS_NAMESPACE_OF) {
4681 r = exec_runtime_acquire(u->manager, NULL, other->id, false, rt);
4682 if (r == 1)
4683 return 1;
4684 }
4685
4686 return exec_runtime_acquire(u->manager, unit_get_exec_context(u), u->id, true, rt);
4687 }
4688
4689 int unit_setup_dynamic_creds(Unit *u) {
4690 ExecContext *ec;
4691 DynamicCreds *dcreds;
4692 size_t offset;
4693
4694 assert(u);
4695
4696 offset = UNIT_VTABLE(u)->dynamic_creds_offset;
4697 assert(offset > 0);
4698 dcreds = (DynamicCreds*) ((uint8_t*) u + offset);
4699
4700 ec = unit_get_exec_context(u);
4701 assert(ec);
4702
4703 if (!ec->dynamic_user)
4704 return 0;
4705
4706 return dynamic_creds_acquire(dcreds, u->manager, ec->user, ec->group);
4707 }
4708
4709 bool unit_type_supported(UnitType t) {
4710 if (_unlikely_(t < 0))
4711 return false;
4712 if (_unlikely_(t >= _UNIT_TYPE_MAX))
4713 return false;
4714
4715 if (!unit_vtable[t]->supported)
4716 return true;
4717
4718 return unit_vtable[t]->supported();
4719 }
4720
4721 void unit_warn_if_dir_nonempty(Unit *u, const char* where) {
4722 int r;
4723
4724 assert(u);
4725 assert(where);
4726
4727 if (!unit_log_level_test(u, LOG_NOTICE))
4728 return;
4729
4730 r = dir_is_empty(where);
4731 if (r > 0 || r == -ENOTDIR)
4732 return;
4733 if (r < 0) {
4734 log_unit_warning_errno(u, r, "Failed to check directory %s: %m", where);
4735 return;
4736 }
4737
4738 log_unit_struct(u, LOG_NOTICE,
4739 "MESSAGE_ID=" SD_MESSAGE_OVERMOUNTING_STR,
4740 LOG_UNIT_INVOCATION_ID(u),
4741 LOG_UNIT_MESSAGE(u, "Directory %s to mount over is not empty, mounting anyway.", where),
4742 "WHERE=%s", where);
4743 }
4744
4745 int unit_fail_if_noncanonical(Unit *u, const char* where) {
4746 _cleanup_free_ char *canonical_where = NULL;
4747 int r;
4748
4749 assert(u);
4750 assert(where);
4751
4752 r = chase_symlinks(where, NULL, CHASE_NONEXISTENT, &canonical_where, NULL);
4753 if (r < 0) {
4754 log_unit_debug_errno(u, r, "Failed to check %s for symlinks, ignoring: %m", where);
4755 return 0;
4756 }
4757
4758 /* We will happily ignore a trailing slash (or any redundant slashes) */
4759 if (path_equal(where, canonical_where))
4760 return 0;
4761
4762 /* No need to mention "." or "..", they would already have been rejected by unit_name_from_path() */
4763 log_unit_struct(u, LOG_ERR,
4764 "MESSAGE_ID=" SD_MESSAGE_OVERMOUNTING_STR,
4765 LOG_UNIT_INVOCATION_ID(u),
4766 LOG_UNIT_MESSAGE(u, "Mount path %s is not canonical (contains a symlink).", where),
4767 "WHERE=%s", where);
4768
4769 return -ELOOP;
4770 }
4771
4772 bool unit_is_pristine(Unit *u) {
4773 assert(u);
4774
4775 /* Check if the unit already exists or is already around,
4776 * in a number of different ways. Note that to cater for unit
4777 * types such as slice, we are generally fine with units that
4778 * are marked UNIT_LOADED even though nothing was actually
4779 * loaded, as those unit types don't require a file on disk. */
4780
4781 return !(!IN_SET(u->load_state, UNIT_NOT_FOUND, UNIT_LOADED) ||
4782 u->fragment_path ||
4783 u->source_path ||
4784 !strv_isempty(u->dropin_paths) ||
4785 u->job ||
4786 u->merged_into);
4787 }
4788
4789 pid_t unit_control_pid(Unit *u) {
4790 assert(u);
4791
4792 if (UNIT_VTABLE(u)->control_pid)
4793 return UNIT_VTABLE(u)->control_pid(u);
4794
4795 return 0;
4796 }
4797
4798 pid_t unit_main_pid(Unit *u) {
4799 assert(u);
4800
4801 if (UNIT_VTABLE(u)->main_pid)
4802 return UNIT_VTABLE(u)->main_pid(u);
4803
4804 return 0;
4805 }
4806
4807 static void unit_unref_uid_internal(
4808 Unit *u,
4809 uid_t *ref_uid,
4810 bool destroy_now,
4811 void (*_manager_unref_uid)(Manager *m, uid_t uid, bool destroy_now)) {
4812
4813 assert(u);
4814 assert(ref_uid);
4815 assert(_manager_unref_uid);
4816
4817 /* Generic implementation of both unit_unref_uid() and unit_unref_gid(), under the assumption that uid_t and
4818 * gid_t are actually the same time, with the same validity rules.
4819 *
4820 * Drops a reference to UID/GID from a unit. */
4821
4822 assert_cc(sizeof(uid_t) == sizeof(gid_t));
4823 assert_cc(UID_INVALID == (uid_t) GID_INVALID);
4824
4825 if (!uid_is_valid(*ref_uid))
4826 return;
4827
4828 _manager_unref_uid(u->manager, *ref_uid, destroy_now);
4829 *ref_uid = UID_INVALID;
4830 }
4831
4832 static void unit_unref_uid(Unit *u, bool destroy_now) {
4833 unit_unref_uid_internal(u, &u->ref_uid, destroy_now, manager_unref_uid);
4834 }
4835
4836 static void unit_unref_gid(Unit *u, bool destroy_now) {
4837 unit_unref_uid_internal(u, (uid_t*) &u->ref_gid, destroy_now, manager_unref_gid);
4838 }
4839
4840 void unit_unref_uid_gid(Unit *u, bool destroy_now) {
4841 assert(u);
4842
4843 unit_unref_uid(u, destroy_now);
4844 unit_unref_gid(u, destroy_now);
4845 }
4846
4847 static int unit_ref_uid_internal(
4848 Unit *u,
4849 uid_t *ref_uid,
4850 uid_t uid,
4851 bool clean_ipc,
4852 int (*_manager_ref_uid)(Manager *m, uid_t uid, bool clean_ipc)) {
4853
4854 int r;
4855
4856 assert(u);
4857 assert(ref_uid);
4858 assert(uid_is_valid(uid));
4859 assert(_manager_ref_uid);
4860
4861 /* Generic implementation of both unit_ref_uid() and unit_ref_guid(), under the assumption that uid_t and gid_t
4862 * are actually the same type, and have the same validity rules.
4863 *
4864 * Adds a reference on a specific UID/GID to this unit. Each unit referencing the same UID/GID maintains a
4865 * reference so that we can destroy the UID/GID's IPC resources as soon as this is requested and the counter
4866 * drops to zero. */
4867
4868 assert_cc(sizeof(uid_t) == sizeof(gid_t));
4869 assert_cc(UID_INVALID == (uid_t) GID_INVALID);
4870
4871 if (*ref_uid == uid)
4872 return 0;
4873
4874 if (uid_is_valid(*ref_uid)) /* Already set? */
4875 return -EBUSY;
4876
4877 r = _manager_ref_uid(u->manager, uid, clean_ipc);
4878 if (r < 0)
4879 return r;
4880
4881 *ref_uid = uid;
4882 return 1;
4883 }
4884
4885 static int unit_ref_uid(Unit *u, uid_t uid, bool clean_ipc) {
4886 return unit_ref_uid_internal(u, &u->ref_uid, uid, clean_ipc, manager_ref_uid);
4887 }
4888
4889 static int unit_ref_gid(Unit *u, gid_t gid, bool clean_ipc) {
4890 return unit_ref_uid_internal(u, (uid_t*) &u->ref_gid, (uid_t) gid, clean_ipc, manager_ref_gid);
4891 }
4892
4893 static int unit_ref_uid_gid_internal(Unit *u, uid_t uid, gid_t gid, bool clean_ipc) {
4894 int r = 0, q = 0;
4895
4896 assert(u);
4897
4898 /* Reference both a UID and a GID in one go. Either references both, or neither. */
4899
4900 if (uid_is_valid(uid)) {
4901 r = unit_ref_uid(u, uid, clean_ipc);
4902 if (r < 0)
4903 return r;
4904 }
4905
4906 if (gid_is_valid(gid)) {
4907 q = unit_ref_gid(u, gid, clean_ipc);
4908 if (q < 0) {
4909 if (r > 0)
4910 unit_unref_uid(u, false);
4911
4912 return q;
4913 }
4914 }
4915
4916 return r > 0 || q > 0;
4917 }
4918
4919 int unit_ref_uid_gid(Unit *u, uid_t uid, gid_t gid) {
4920 ExecContext *c;
4921 int r;
4922
4923 assert(u);
4924
4925 c = unit_get_exec_context(u);
4926
4927 r = unit_ref_uid_gid_internal(u, uid, gid, c ? c->remove_ipc : false);
4928 if (r < 0)
4929 return log_unit_warning_errno(u, r, "Couldn't add UID/GID reference to unit, proceeding without: %m");
4930
4931 return r;
4932 }
4933
4934 void unit_notify_user_lookup(Unit *u, uid_t uid, gid_t gid) {
4935 int r;
4936
4937 assert(u);
4938
4939 /* This is invoked whenever one of the forked off processes let's us know the UID/GID its user name/group names
4940 * resolved to. We keep track of which UID/GID is currently assigned in order to be able to destroy its IPC
4941 * objects when no service references the UID/GID anymore. */
4942
4943 r = unit_ref_uid_gid(u, uid, gid);
4944 if (r > 0)
4945 unit_add_to_dbus_queue(u);
4946 }
4947
4948 int unit_acquire_invocation_id(Unit *u) {
4949 sd_id128_t id;
4950 int r;
4951
4952 assert(u);
4953
4954 r = sd_id128_randomize(&id);
4955 if (r < 0)
4956 return log_unit_error_errno(u, r, "Failed to generate invocation ID for unit: %m");
4957
4958 r = unit_set_invocation_id(u, id);
4959 if (r < 0)
4960 return log_unit_error_errno(u, r, "Failed to set invocation ID for unit: %m");
4961
4962 unit_add_to_dbus_queue(u);
4963 return 0;
4964 }
4965
4966 int unit_set_exec_params(Unit *u, ExecParameters *p) {
4967 int r;
4968
4969 assert(u);
4970 assert(p);
4971
4972 /* Copy parameters from manager */
4973 r = manager_get_effective_environment(u->manager, &p->environment);
4974 if (r < 0)
4975 return r;
4976
4977 p->confirm_spawn = manager_get_confirm_spawn(u->manager);
4978 p->cgroup_supported = u->manager->cgroup_supported;
4979 p->prefix = u->manager->prefix;
4980 SET_FLAG(p->flags, EXEC_PASS_LOG_UNIT|EXEC_CHOWN_DIRECTORIES, MANAGER_IS_SYSTEM(u->manager));
4981
4982 /* Copy parameters from unit */
4983 p->cgroup_path = u->cgroup_path;
4984 SET_FLAG(p->flags, EXEC_CGROUP_DELEGATE, unit_cgroup_delegate(u));
4985
4986 p->received_credentials = u->manager->received_credentials;
4987
4988 return 0;
4989 }
4990
4991 int unit_fork_helper_process(Unit *u, const char *name, pid_t *ret) {
4992 int r;
4993
4994 assert(u);
4995 assert(ret);
4996
4997 /* Forks off a helper process and makes sure it is a member of the unit's cgroup. Returns == 0 in the child,
4998 * and > 0 in the parent. The pid parameter is always filled in with the child's PID. */
4999
5000 (void) unit_realize_cgroup(u);
5001
5002 r = safe_fork(name, FORK_REOPEN_LOG, ret);
5003 if (r != 0)
5004 return r;
5005
5006 (void) default_signals(SIGNALS_CRASH_HANDLER, SIGNALS_IGNORE);
5007 (void) ignore_signals(SIGPIPE);
5008
5009 (void) prctl(PR_SET_PDEATHSIG, SIGTERM);
5010
5011 if (u->cgroup_path) {
5012 r = cg_attach_everywhere(u->manager->cgroup_supported, u->cgroup_path, 0, NULL, NULL);
5013 if (r < 0) {
5014 log_unit_error_errno(u, r, "Failed to join unit cgroup %s: %m", empty_to_root(u->cgroup_path));
5015 _exit(EXIT_CGROUP);
5016 }
5017 }
5018
5019 return 0;
5020 }
5021
5022 int unit_fork_and_watch_rm_rf(Unit *u, char **paths, pid_t *ret_pid) {
5023 pid_t pid;
5024 int r;
5025
5026 assert(u);
5027 assert(ret_pid);
5028
5029 r = unit_fork_helper_process(u, "(sd-rmrf)", &pid);
5030 if (r < 0)
5031 return r;
5032 if (r == 0) {
5033 int ret = EXIT_SUCCESS;
5034 char **i;
5035
5036 STRV_FOREACH(i, paths) {
5037 r = rm_rf(*i, REMOVE_ROOT|REMOVE_PHYSICAL|REMOVE_MISSING_OK);
5038 if (r < 0) {
5039 log_error_errno(r, "Failed to remove '%s': %m", *i);
5040 ret = EXIT_FAILURE;
5041 }
5042 }
5043
5044 _exit(ret);
5045 }
5046
5047 r = unit_watch_pid(u, pid, true);
5048 if (r < 0)
5049 return r;
5050
5051 *ret_pid = pid;
5052 return 0;
5053 }
5054
5055 static void unit_update_dependency_mask(Hashmap *deps, Unit *other, UnitDependencyInfo di) {
5056 assert(deps);
5057 assert(other);
5058
5059 if (di.origin_mask == 0 && di.destination_mask == 0)
5060 /* No bit set anymore, let's drop the whole entry */
5061 assert_se(hashmap_remove(deps, other));
5062 else
5063 /* Mask was reduced, let's update the entry */
5064 assert_se(hashmap_update(deps, other, di.data) == 0);
5065 }
5066
5067 void unit_remove_dependencies(Unit *u, UnitDependencyMask mask) {
5068 Hashmap *deps;
5069 assert(u);
5070
5071 /* Removes all dependencies u has on other units marked for ownership by 'mask'. */
5072
5073 if (mask == 0)
5074 return;
5075
5076 HASHMAP_FOREACH(deps, u->dependencies) {
5077 bool done;
5078
5079 do {
5080 UnitDependencyInfo di;
5081 Unit *other;
5082
5083 done = true;
5084
5085 HASHMAP_FOREACH_KEY(di.data, other, deps) {
5086 Hashmap *other_deps;
5087
5088 if (FLAGS_SET(~mask, di.origin_mask))
5089 continue;
5090
5091 di.origin_mask &= ~mask;
5092 unit_update_dependency_mask(deps, other, di);
5093
5094 /* We updated the dependency from our unit to the other unit now. But most
5095 * dependencies imply a reverse dependency. Hence, let's delete that one
5096 * too. For that we go through all dependency types on the other unit and
5097 * delete all those which point to us and have the right mask set. */
5098
5099 HASHMAP_FOREACH(other_deps, other->dependencies) {
5100 UnitDependencyInfo dj;
5101
5102 dj.data = hashmap_get(other_deps, u);
5103 if (FLAGS_SET(~mask, dj.destination_mask))
5104 continue;
5105
5106 dj.destination_mask &= ~mask;
5107 unit_update_dependency_mask(other_deps, u, dj);
5108 }
5109
5110 unit_add_to_gc_queue(other);
5111
5112 done = false;
5113 break;
5114 }
5115
5116 } while (!done);
5117 }
5118 }
5119
5120 static int unit_get_invocation_path(Unit *u, char **ret) {
5121 char *p;
5122 int r;
5123
5124 assert(u);
5125 assert(ret);
5126
5127 if (MANAGER_IS_SYSTEM(u->manager))
5128 p = strjoin("/run/systemd/units/invocation:", u->id);
5129 else {
5130 _cleanup_free_ char *user_path = NULL;
5131 r = xdg_user_runtime_dir(&user_path, "/systemd/units/invocation:");
5132 if (r < 0)
5133 return r;
5134 p = strjoin(user_path, u->id);
5135 }
5136
5137 if (!p)
5138 return -ENOMEM;
5139
5140 *ret = p;
5141 return 0;
5142 }
5143
5144 static int unit_export_invocation_id(Unit *u) {
5145 _cleanup_free_ char *p = NULL;
5146 int r;
5147
5148 assert(u);
5149
5150 if (u->exported_invocation_id)
5151 return 0;
5152
5153 if (sd_id128_is_null(u->invocation_id))
5154 return 0;
5155
5156 r = unit_get_invocation_path(u, &p);
5157 if (r < 0)
5158 return log_unit_debug_errno(u, r, "Failed to get invocation path: %m");
5159
5160 r = symlink_atomic_label(u->invocation_id_string, p);
5161 if (r < 0)
5162 return log_unit_debug_errno(u, r, "Failed to create invocation ID symlink %s: %m", p);
5163
5164 u->exported_invocation_id = true;
5165 return 0;
5166 }
5167
5168 static int unit_export_log_level_max(Unit *u, const ExecContext *c) {
5169 const char *p;
5170 char buf[2];
5171 int r;
5172
5173 assert(u);
5174 assert(c);
5175
5176 if (u->exported_log_level_max)
5177 return 0;
5178
5179 if (c->log_level_max < 0)
5180 return 0;
5181
5182 assert(c->log_level_max <= 7);
5183
5184 buf[0] = '0' + c->log_level_max;
5185 buf[1] = 0;
5186
5187 p = strjoina("/run/systemd/units/log-level-max:", u->id);
5188 r = symlink_atomic(buf, p);
5189 if (r < 0)
5190 return log_unit_debug_errno(u, r, "Failed to create maximum log level symlink %s: %m", p);
5191
5192 u->exported_log_level_max = true;
5193 return 0;
5194 }
5195
5196 static int unit_export_log_extra_fields(Unit *u, const ExecContext *c) {
5197 _cleanup_close_ int fd = -1;
5198 struct iovec *iovec;
5199 const char *p;
5200 char *pattern;
5201 le64_t *sizes;
5202 ssize_t n;
5203 int r;
5204
5205 if (u->exported_log_extra_fields)
5206 return 0;
5207
5208 if (c->n_log_extra_fields <= 0)
5209 return 0;
5210
5211 sizes = newa(le64_t, c->n_log_extra_fields);
5212 iovec = newa(struct iovec, c->n_log_extra_fields * 2);
5213
5214 for (size_t i = 0; i < c->n_log_extra_fields; i++) {
5215 sizes[i] = htole64(c->log_extra_fields[i].iov_len);
5216
5217 iovec[i*2] = IOVEC_MAKE(sizes + i, sizeof(le64_t));
5218 iovec[i*2+1] = c->log_extra_fields[i];
5219 }
5220
5221 p = strjoina("/run/systemd/units/log-extra-fields:", u->id);
5222 pattern = strjoina(p, ".XXXXXX");
5223
5224 fd = mkostemp_safe(pattern);
5225 if (fd < 0)
5226 return log_unit_debug_errno(u, fd, "Failed to create extra fields file %s: %m", p);
5227
5228 n = writev(fd, iovec, c->n_log_extra_fields*2);
5229 if (n < 0) {
5230 r = log_unit_debug_errno(u, errno, "Failed to write extra fields: %m");
5231 goto fail;
5232 }
5233
5234 (void) fchmod(fd, 0644);
5235
5236 if (rename(pattern, p) < 0) {
5237 r = log_unit_debug_errno(u, errno, "Failed to rename extra fields file: %m");
5238 goto fail;
5239 }
5240
5241 u->exported_log_extra_fields = true;
5242 return 0;
5243
5244 fail:
5245 (void) unlink(pattern);
5246 return r;
5247 }
5248
5249 static int unit_export_log_ratelimit_interval(Unit *u, const ExecContext *c) {
5250 _cleanup_free_ char *buf = NULL;
5251 const char *p;
5252 int r;
5253
5254 assert(u);
5255 assert(c);
5256
5257 if (u->exported_log_ratelimit_interval)
5258 return 0;
5259
5260 if (c->log_ratelimit_interval_usec == 0)
5261 return 0;
5262
5263 p = strjoina("/run/systemd/units/log-rate-limit-interval:", u->id);
5264
5265 if (asprintf(&buf, "%" PRIu64, c->log_ratelimit_interval_usec) < 0)
5266 return log_oom();
5267
5268 r = symlink_atomic(buf, p);
5269 if (r < 0)
5270 return log_unit_debug_errno(u, r, "Failed to create log rate limit interval symlink %s: %m", p);
5271
5272 u->exported_log_ratelimit_interval = true;
5273 return 0;
5274 }
5275
5276 static int unit_export_log_ratelimit_burst(Unit *u, const ExecContext *c) {
5277 _cleanup_free_ char *buf = NULL;
5278 const char *p;
5279 int r;
5280
5281 assert(u);
5282 assert(c);
5283
5284 if (u->exported_log_ratelimit_burst)
5285 return 0;
5286
5287 if (c->log_ratelimit_burst == 0)
5288 return 0;
5289
5290 p = strjoina("/run/systemd/units/log-rate-limit-burst:", u->id);
5291
5292 if (asprintf(&buf, "%u", c->log_ratelimit_burst) < 0)
5293 return log_oom();
5294
5295 r = symlink_atomic(buf, p);
5296 if (r < 0)
5297 return log_unit_debug_errno(u, r, "Failed to create log rate limit burst symlink %s: %m", p);
5298
5299 u->exported_log_ratelimit_burst = true;
5300 return 0;
5301 }
5302
5303 void unit_export_state_files(Unit *u) {
5304 const ExecContext *c;
5305
5306 assert(u);
5307
5308 if (!u->id)
5309 return;
5310
5311 if (MANAGER_IS_TEST_RUN(u->manager))
5312 return;
5313
5314 /* Exports a couple of unit properties to /run/systemd/units/, so that journald can quickly query this data
5315 * from there. Ideally, journald would use IPC to query this, like everybody else, but that's hard, as long as
5316 * the IPC system itself and PID 1 also log to the journal.
5317 *
5318 * Note that these files really shouldn't be considered API for anyone else, as use a runtime file system as
5319 * IPC replacement is not compatible with today's world of file system namespaces. However, this doesn't really
5320 * apply to communication between the journal and systemd, as we assume that these two daemons live in the same
5321 * namespace at least.
5322 *
5323 * Note that some of the "files" exported here are actually symlinks and not regular files. Symlinks work
5324 * better for storing small bits of data, in particular as we can write them with two system calls, and read
5325 * them with one. */
5326
5327 (void) unit_export_invocation_id(u);
5328
5329 if (!MANAGER_IS_SYSTEM(u->manager))
5330 return;
5331
5332 c = unit_get_exec_context(u);
5333 if (c) {
5334 (void) unit_export_log_level_max(u, c);
5335 (void) unit_export_log_extra_fields(u, c);
5336 (void) unit_export_log_ratelimit_interval(u, c);
5337 (void) unit_export_log_ratelimit_burst(u, c);
5338 }
5339 }
5340
5341 void unit_unlink_state_files(Unit *u) {
5342 const char *p;
5343
5344 assert(u);
5345
5346 if (!u->id)
5347 return;
5348
5349 /* Undoes the effect of unit_export_state() */
5350
5351 if (u->exported_invocation_id) {
5352 _cleanup_free_ char *invocation_path = NULL;
5353 int r = unit_get_invocation_path(u, &invocation_path);
5354 if (r >= 0) {
5355 (void) unlink(invocation_path);
5356 u->exported_invocation_id = false;
5357 }
5358 }
5359
5360 if (!MANAGER_IS_SYSTEM(u->manager))
5361 return;
5362
5363 if (u->exported_log_level_max) {
5364 p = strjoina("/run/systemd/units/log-level-max:", u->id);
5365 (void) unlink(p);
5366
5367 u->exported_log_level_max = false;
5368 }
5369
5370 if (u->exported_log_extra_fields) {
5371 p = strjoina("/run/systemd/units/extra-fields:", u->id);
5372 (void) unlink(p);
5373
5374 u->exported_log_extra_fields = false;
5375 }
5376
5377 if (u->exported_log_ratelimit_interval) {
5378 p = strjoina("/run/systemd/units/log-rate-limit-interval:", u->id);
5379 (void) unlink(p);
5380
5381 u->exported_log_ratelimit_interval = false;
5382 }
5383
5384 if (u->exported_log_ratelimit_burst) {
5385 p = strjoina("/run/systemd/units/log-rate-limit-burst:", u->id);
5386 (void) unlink(p);
5387
5388 u->exported_log_ratelimit_burst = false;
5389 }
5390 }
5391
5392 int unit_prepare_exec(Unit *u) {
5393 int r;
5394
5395 assert(u);
5396
5397 /* Load any custom firewall BPF programs here once to test if they are existing and actually loadable.
5398 * Fail here early since later errors in the call chain unit_realize_cgroup to cgroup_context_apply are ignored. */
5399 r = bpf_firewall_load_custom(u);
5400 if (r < 0)
5401 return r;
5402
5403 /* Prepares everything so that we can fork of a process for this unit */
5404
5405 (void) unit_realize_cgroup(u);
5406
5407 if (u->reset_accounting) {
5408 (void) unit_reset_accounting(u);
5409 u->reset_accounting = false;
5410 }
5411
5412 unit_export_state_files(u);
5413
5414 r = unit_setup_exec_runtime(u);
5415 if (r < 0)
5416 return r;
5417
5418 r = unit_setup_dynamic_creds(u);
5419 if (r < 0)
5420 return r;
5421
5422 return 0;
5423 }
5424
5425 static bool ignore_leftover_process(const char *comm) {
5426 return comm && comm[0] == '('; /* Most likely our own helper process (PAM?), ignore */
5427 }
5428
5429 int unit_log_leftover_process_start(pid_t pid, int sig, void *userdata) {
5430 _cleanup_free_ char *comm = NULL;
5431
5432 (void) get_process_comm(pid, &comm);
5433
5434 if (ignore_leftover_process(comm))
5435 return 0;
5436
5437 /* During start we print a warning */
5438
5439 log_unit_warning(userdata,
5440 "Found left-over process " PID_FMT " (%s) in control group while starting unit. Ignoring.\n"
5441 "This usually indicates unclean termination of a previous run, or service implementation deficiencies.",
5442 pid, strna(comm));
5443
5444 return 1;
5445 }
5446
5447 int unit_log_leftover_process_stop(pid_t pid, int sig, void *userdata) {
5448 _cleanup_free_ char *comm = NULL;
5449
5450 (void) get_process_comm(pid, &comm);
5451
5452 if (ignore_leftover_process(comm))
5453 return 0;
5454
5455 /* During stop we only print an informational message */
5456
5457 log_unit_info(userdata,
5458 "Unit process " PID_FMT " (%s) remains running after unit stopped.",
5459 pid, strna(comm));
5460
5461 return 1;
5462 }
5463
5464 int unit_warn_leftover_processes(Unit *u, cg_kill_log_func_t log_func) {
5465 assert(u);
5466
5467 (void) unit_pick_cgroup_path(u);
5468
5469 if (!u->cgroup_path)
5470 return 0;
5471
5472 return cg_kill_recursive(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, 0, 0, NULL, log_func, u);
5473 }
5474
5475 bool unit_needs_console(Unit *u) {
5476 ExecContext *ec;
5477 UnitActiveState state;
5478
5479 assert(u);
5480
5481 state = unit_active_state(u);
5482
5483 if (UNIT_IS_INACTIVE_OR_FAILED(state))
5484 return false;
5485
5486 if (UNIT_VTABLE(u)->needs_console)
5487 return UNIT_VTABLE(u)->needs_console(u);
5488
5489 /* If this unit type doesn't implement this call, let's use a generic fallback implementation: */
5490 ec = unit_get_exec_context(u);
5491 if (!ec)
5492 return false;
5493
5494 return exec_context_may_touch_console(ec);
5495 }
5496
5497 const char *unit_label_path(const Unit *u) {
5498 const char *p;
5499
5500 assert(u);
5501
5502 /* Returns the file system path to use for MAC access decisions, i.e. the file to read the SELinux label off
5503 * when validating access checks. */
5504
5505 if (IN_SET(u->load_state, UNIT_MASKED, UNIT_NOT_FOUND, UNIT_MERGED))
5506 return NULL; /* Shortcut things if we know there is no real, relevant unit file around */
5507
5508 p = u->source_path ?: u->fragment_path;
5509 if (!p)
5510 return NULL;
5511
5512 if (IN_SET(u->load_state, UNIT_LOADED, UNIT_BAD_SETTING, UNIT_ERROR))
5513 return p; /* Shortcut things, if we successfully loaded at least some stuff from the unit file */
5514
5515 /* Not loaded yet, we need to go to disk */
5516 assert(u->load_state == UNIT_STUB);
5517
5518 /* If a unit is masked, then don't read the SELinux label of /dev/null, as that really makes no sense */
5519 if (null_or_empty_path(p) > 0)
5520 return NULL;
5521
5522 return p;
5523 }
5524
5525 int unit_pid_attachable(Unit *u, pid_t pid, sd_bus_error *error) {
5526 int r;
5527
5528 assert(u);
5529
5530 /* Checks whether the specified PID is generally good for attaching, i.e. a valid PID, not our manager itself,
5531 * and not a kernel thread either */
5532
5533 /* First, a simple range check */
5534 if (!pid_is_valid(pid))
5535 return sd_bus_error_setf(error, SD_BUS_ERROR_INVALID_ARGS, "Process identifier " PID_FMT " is not valid.", pid);
5536
5537 /* Some extra safety check */
5538 if (pid == 1 || pid == getpid_cached())
5539 return sd_bus_error_setf(error, SD_BUS_ERROR_INVALID_ARGS, "Process " PID_FMT " is a manager process, refusing.", pid);
5540
5541 /* Don't even begin to bother with kernel threads */
5542 r = is_kernel_thread(pid);
5543 if (r == -ESRCH)
5544 return sd_bus_error_setf(error, SD_BUS_ERROR_UNIX_PROCESS_ID_UNKNOWN, "Process with ID " PID_FMT " does not exist.", pid);
5545 if (r < 0)
5546 return sd_bus_error_set_errnof(error, r, "Failed to determine whether process " PID_FMT " is a kernel thread: %m", pid);
5547 if (r > 0)
5548 return sd_bus_error_setf(error, SD_BUS_ERROR_INVALID_ARGS, "Process " PID_FMT " is a kernel thread, refusing.", pid);
5549
5550 return 0;
5551 }
5552
5553 void unit_log_success(Unit *u) {
5554 assert(u);
5555
5556 /* Let's show message "Deactivated successfully" in debug mode (when manager is user) rather than in info mode.
5557 * This message has low information value for regular users and it might be a bit overwhelming on a system with
5558 * a lot of devices. */
5559 log_unit_struct(u,
5560 MANAGER_IS_USER(u->manager) ? LOG_DEBUG : LOG_INFO,
5561 "MESSAGE_ID=" SD_MESSAGE_UNIT_SUCCESS_STR,
5562 LOG_UNIT_INVOCATION_ID(u),
5563 LOG_UNIT_MESSAGE(u, "Deactivated successfully."));
5564 }
5565
5566 void unit_log_failure(Unit *u, const char *result) {
5567 assert(u);
5568 assert(result);
5569
5570 log_unit_struct(u, LOG_WARNING,
5571 "MESSAGE_ID=" SD_MESSAGE_UNIT_FAILURE_RESULT_STR,
5572 LOG_UNIT_INVOCATION_ID(u),
5573 LOG_UNIT_MESSAGE(u, "Failed with result '%s'.", result),
5574 "UNIT_RESULT=%s", result);
5575 }
5576
5577 void unit_log_skip(Unit *u, const char *result) {
5578 assert(u);
5579 assert(result);
5580
5581 log_unit_struct(u, LOG_INFO,
5582 "MESSAGE_ID=" SD_MESSAGE_UNIT_SKIPPED_STR,
5583 LOG_UNIT_INVOCATION_ID(u),
5584 LOG_UNIT_MESSAGE(u, "Skipped due to '%s'.", result),
5585 "UNIT_RESULT=%s", result);
5586 }
5587
5588 void unit_log_process_exit(
5589 Unit *u,
5590 const char *kind,
5591 const char *command,
5592 bool success,
5593 int code,
5594 int status) {
5595
5596 int level;
5597
5598 assert(u);
5599 assert(kind);
5600
5601 /* If this is a successful exit, let's log about the exit code on DEBUG level. If this is a failure
5602 * and the process exited on its own via exit(), then let's make this a NOTICE, under the assumption
5603 * that the service already logged the reason at a higher log level on its own. Otherwise, make it a
5604 * WARNING. */
5605 if (success)
5606 level = LOG_DEBUG;
5607 else if (code == CLD_EXITED)
5608 level = LOG_NOTICE;
5609 else
5610 level = LOG_WARNING;
5611
5612 log_unit_struct(u, level,
5613 "MESSAGE_ID=" SD_MESSAGE_UNIT_PROCESS_EXIT_STR,
5614 LOG_UNIT_MESSAGE(u, "%s exited, code=%s, status=%i/%s%s",
5615 kind,
5616 sigchld_code_to_string(code), status,
5617 strna(code == CLD_EXITED
5618 ? exit_status_to_string(status, EXIT_STATUS_FULL)
5619 : signal_to_string(status)),
5620 success ? " (success)" : ""),
5621 "EXIT_CODE=%s", sigchld_code_to_string(code),
5622 "EXIT_STATUS=%i", status,
5623 "COMMAND=%s", strna(command),
5624 LOG_UNIT_INVOCATION_ID(u));
5625 }
5626
5627 int unit_exit_status(Unit *u) {
5628 assert(u);
5629
5630 /* Returns the exit status to propagate for the most recent cycle of this unit. Returns a value in the range
5631 * 0…255 if there's something to propagate. EOPNOTSUPP if the concept does not apply to this unit type, ENODATA
5632 * if no data is currently known (for example because the unit hasn't deactivated yet) and EBADE if the main
5633 * service process has exited abnormally (signal/coredump). */
5634
5635 if (!UNIT_VTABLE(u)->exit_status)
5636 return -EOPNOTSUPP;
5637
5638 return UNIT_VTABLE(u)->exit_status(u);
5639 }
5640
5641 int unit_failure_action_exit_status(Unit *u) {
5642 int r;
5643
5644 assert(u);
5645
5646 /* Returns the exit status to propagate on failure, or an error if there's nothing to propagate */
5647
5648 if (u->failure_action_exit_status >= 0)
5649 return u->failure_action_exit_status;
5650
5651 r = unit_exit_status(u);
5652 if (r == -EBADE) /* Exited, but not cleanly (i.e. by signal or such) */
5653 return 255;
5654
5655 return r;
5656 }
5657
5658 int unit_success_action_exit_status(Unit *u) {
5659 int r;
5660
5661 assert(u);
5662
5663 /* Returns the exit status to propagate on success, or an error if there's nothing to propagate */
5664
5665 if (u->success_action_exit_status >= 0)
5666 return u->success_action_exit_status;
5667
5668 r = unit_exit_status(u);
5669 if (r == -EBADE) /* Exited, but not cleanly (i.e. by signal or such) */
5670 return 255;
5671
5672 return r;
5673 }
5674
5675 int unit_test_trigger_loaded(Unit *u) {
5676 Unit *trigger;
5677
5678 /* Tests whether the unit to trigger is loaded */
5679
5680 trigger = UNIT_TRIGGER(u);
5681 if (!trigger)
5682 return log_unit_error_errno(u, SYNTHETIC_ERRNO(ENOENT),
5683 "Refusing to start, no unit to trigger.");
5684 if (trigger->load_state != UNIT_LOADED)
5685 return log_unit_error_errno(u, SYNTHETIC_ERRNO(ENOENT),
5686 "Refusing to start, unit %s to trigger not loaded.", trigger->id);
5687
5688 return 0;
5689 }
5690
5691 void unit_destroy_runtime_data(Unit *u, const ExecContext *context) {
5692 assert(u);
5693 assert(context);
5694
5695 if (context->runtime_directory_preserve_mode == EXEC_PRESERVE_NO ||
5696 (context->runtime_directory_preserve_mode == EXEC_PRESERVE_RESTART && !unit_will_restart(u)))
5697 exec_context_destroy_runtime_directory(context, u->manager->prefix[EXEC_DIRECTORY_RUNTIME]);
5698
5699 exec_context_destroy_credentials(context, u->manager->prefix[EXEC_DIRECTORY_RUNTIME], u->id);
5700 }
5701
5702 int unit_clean(Unit *u, ExecCleanMask mask) {
5703 UnitActiveState state;
5704
5705 assert(u);
5706
5707 /* Special return values:
5708 *
5709 * -EOPNOTSUPP → cleaning not supported for this unit type
5710 * -EUNATCH → cleaning not defined for this resource type
5711 * -EBUSY → unit currently can't be cleaned since it's running or not properly loaded, or has
5712 * a job queued or similar
5713 */
5714
5715 if (!UNIT_VTABLE(u)->clean)
5716 return -EOPNOTSUPP;
5717
5718 if (mask == 0)
5719 return -EUNATCH;
5720
5721 if (u->load_state != UNIT_LOADED)
5722 return -EBUSY;
5723
5724 if (u->job)
5725 return -EBUSY;
5726
5727 state = unit_active_state(u);
5728 if (!IN_SET(state, UNIT_INACTIVE))
5729 return -EBUSY;
5730
5731 return UNIT_VTABLE(u)->clean(u, mask);
5732 }
5733
5734 int unit_can_clean(Unit *u, ExecCleanMask *ret) {
5735 assert(u);
5736
5737 if (!UNIT_VTABLE(u)->clean ||
5738 u->load_state != UNIT_LOADED) {
5739 *ret = 0;
5740 return 0;
5741 }
5742
5743 /* When the clean() method is set, can_clean() really should be set too */
5744 assert(UNIT_VTABLE(u)->can_clean);
5745
5746 return UNIT_VTABLE(u)->can_clean(u, ret);
5747 }
5748
5749 bool unit_can_freeze(Unit *u) {
5750 assert(u);
5751
5752 if (UNIT_VTABLE(u)->can_freeze)
5753 return UNIT_VTABLE(u)->can_freeze(u);
5754
5755 return UNIT_VTABLE(u)->freeze;
5756 }
5757
5758 void unit_frozen(Unit *u) {
5759 assert(u);
5760
5761 u->freezer_state = FREEZER_FROZEN;
5762
5763 bus_unit_send_pending_freezer_message(u);
5764 }
5765
5766 void unit_thawed(Unit *u) {
5767 assert(u);
5768
5769 u->freezer_state = FREEZER_RUNNING;
5770
5771 bus_unit_send_pending_freezer_message(u);
5772 }
5773
5774 static int unit_freezer_action(Unit *u, FreezerAction action) {
5775 UnitActiveState s;
5776 int (*method)(Unit*);
5777 int r;
5778
5779 assert(u);
5780 assert(IN_SET(action, FREEZER_FREEZE, FREEZER_THAW));
5781
5782 method = action == FREEZER_FREEZE ? UNIT_VTABLE(u)->freeze : UNIT_VTABLE(u)->thaw;
5783 if (!method || !cg_freezer_supported())
5784 return -EOPNOTSUPP;
5785
5786 if (u->job)
5787 return -EBUSY;
5788
5789 if (u->load_state != UNIT_LOADED)
5790 return -EHOSTDOWN;
5791
5792 s = unit_active_state(u);
5793 if (s != UNIT_ACTIVE)
5794 return -EHOSTDOWN;
5795
5796 if (IN_SET(u->freezer_state, FREEZER_FREEZING, FREEZER_THAWING))
5797 return -EALREADY;
5798
5799 r = method(u);
5800 if (r <= 0)
5801 return r;
5802
5803 return 1;
5804 }
5805
5806 int unit_freeze(Unit *u) {
5807 return unit_freezer_action(u, FREEZER_FREEZE);
5808 }
5809
5810 int unit_thaw(Unit *u) {
5811 return unit_freezer_action(u, FREEZER_THAW);
5812 }
5813
5814 /* Wrappers around low-level cgroup freezer operations common for service and scope units */
5815 int unit_freeze_vtable_common(Unit *u) {
5816 return unit_cgroup_freezer_action(u, FREEZER_FREEZE);
5817 }
5818
5819 int unit_thaw_vtable_common(Unit *u) {
5820 return unit_cgroup_freezer_action(u, FREEZER_THAW);
5821 }
5822
5823 static const char* const collect_mode_table[_COLLECT_MODE_MAX] = {
5824 [COLLECT_INACTIVE] = "inactive",
5825 [COLLECT_INACTIVE_OR_FAILED] = "inactive-or-failed",
5826 };
5827
5828 DEFINE_STRING_TABLE_LOOKUP(collect_mode, CollectMode);
5829
5830 Unit* unit_has_dependency(const Unit *u, UnitDependencyAtom atom, Unit *other) {
5831 Unit *i;
5832
5833 assert(u);
5834
5835 /* Checks if the unit has a dependency on 'other' with the specified dependency atom. If 'other' is
5836 * NULL checks if the unit has *any* dependency of that atom. Returns 'other' if found (or if 'other'
5837 * is NULL the first entry found), or NULL if not found. */
5838
5839 UNIT_FOREACH_DEPENDENCY(i, u, atom)
5840 if (!other || other == i)
5841 return i;
5842
5843 return NULL;
5844 }
5845
5846 int unit_get_dependency_array(const Unit *u, UnitDependencyAtom atom, Unit ***ret_array) {
5847 _cleanup_free_ Unit **array = NULL;
5848 size_t n = 0;
5849 Unit *other;
5850
5851 assert(u);
5852 assert(ret_array);
5853
5854 /* Gets a list of units matching a specific atom as array. This is useful when iterating through
5855 * dependencies while modifying them: the array is an "atomic snapshot" of sorts, that can be read
5856 * while the dependency table is continuously updated. */
5857
5858 UNIT_FOREACH_DEPENDENCY(other, u, atom) {
5859 if (!GREEDY_REALLOC(array, n + 1))
5860 return -ENOMEM;
5861
5862 array[n++] = other;
5863 }
5864
5865 *ret_array = TAKE_PTR(array);
5866
5867 assert(n <= INT_MAX);
5868 return (int) n;
5869 }