]> git.ipfire.org Git - thirdparty/systemd.git/blob - src/core/unit.c
Merge pull request #14194 from yuwata/network-multipath-routing-12541
[thirdparty/systemd.git] / src / core / unit.c
1 /* SPDX-License-Identifier: LGPL-2.1+ */
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 "bus-common-errors.h"
15 #include "bus-util.h"
16 #include "cgroup-setup.h"
17 #include "cgroup-util.h"
18 #include "dbus-unit.h"
19 #include "dbus.h"
20 #include "dropin.h"
21 #include "escape.h"
22 #include "execute.h"
23 #include "fd-util.h"
24 #include "fileio-label.h"
25 #include "fileio.h"
26 #include "format-util.h"
27 #include "fs-util.h"
28 #include "id128-util.h"
29 #include "io-util.h"
30 #include "install.h"
31 #include "load-dropin.h"
32 #include "load-fragment.h"
33 #include "log.h"
34 #include "macro.h"
35 #include "missing_audit.h"
36 #include "mkdir.h"
37 #include "parse-util.h"
38 #include "path-util.h"
39 #include "process-util.h"
40 #include "rm-rf.h"
41 #include "serialize.h"
42 #include "set.h"
43 #include "signal-util.h"
44 #include "sparse-endian.h"
45 #include "special.h"
46 #include "specifier.h"
47 #include "stat-util.h"
48 #include "stdio-util.h"
49 #include "string-table.h"
50 #include "string-util.h"
51 #include "strv.h"
52 #include "terminal-util.h"
53 #include "tmpfile-util.h"
54 #include "umask-util.h"
55 #include "unit-name.h"
56 #include "unit.h"
57 #include "user-util.h"
58 #include "virt.h"
59
60 /* Thresholds for logging at INFO level about resource consumption */
61 #define MENTIONWORTHY_CPU_NSEC (1 * NSEC_PER_SEC)
62 #define MENTIONWORTHY_IO_BYTES (1024 * 1024ULL)
63 #define MENTIONWORTHY_IP_BYTES (0ULL)
64
65 /* Thresholds for logging at INFO level about resource consumption */
66 #define NOTICEWORTHY_CPU_NSEC (10*60 * NSEC_PER_SEC) /* 10 minutes */
67 #define NOTICEWORTHY_IO_BYTES (10 * 1024 * 1024ULL) /* 10 MB */
68 #define NOTICEWORTHY_IP_BYTES (128 * 1024 * 1024ULL) /* 128 MB */
69
70 const UnitVTable * const unit_vtable[_UNIT_TYPE_MAX] = {
71 [UNIT_SERVICE] = &service_vtable,
72 [UNIT_SOCKET] = &socket_vtable,
73 [UNIT_TARGET] = &target_vtable,
74 [UNIT_DEVICE] = &device_vtable,
75 [UNIT_MOUNT] = &mount_vtable,
76 [UNIT_AUTOMOUNT] = &automount_vtable,
77 [UNIT_SWAP] = &swap_vtable,
78 [UNIT_TIMER] = &timer_vtable,
79 [UNIT_PATH] = &path_vtable,
80 [UNIT_SLICE] = &slice_vtable,
81 [UNIT_SCOPE] = &scope_vtable,
82 };
83
84 static void maybe_warn_about_dependency(Unit *u, const char *other, UnitDependency dependency);
85
86 Unit *unit_new(Manager *m, size_t size) {
87 Unit *u;
88
89 assert(m);
90 assert(size >= sizeof(Unit));
91
92 u = malloc0(size);
93 if (!u)
94 return NULL;
95
96 u->names = set_new(&string_hash_ops);
97 if (!u->names)
98 return mfree(u);
99
100 u->manager = m;
101 u->type = _UNIT_TYPE_INVALID;
102 u->default_dependencies = true;
103 u->unit_file_state = _UNIT_FILE_STATE_INVALID;
104 u->unit_file_preset = -1;
105 u->on_failure_job_mode = JOB_REPLACE;
106 u->cgroup_control_inotify_wd = -1;
107 u->cgroup_memory_inotify_wd = -1;
108 u->job_timeout = USEC_INFINITY;
109 u->job_running_timeout = USEC_INFINITY;
110 u->ref_uid = UID_INVALID;
111 u->ref_gid = GID_INVALID;
112 u->cpu_usage_last = NSEC_INFINITY;
113 u->cgroup_invalidated_mask |= CGROUP_MASK_BPF_FIREWALL;
114 u->failure_action_exit_status = u->success_action_exit_status = -1;
115
116 u->ip_accounting_ingress_map_fd = -1;
117 u->ip_accounting_egress_map_fd = -1;
118 u->ipv4_allow_map_fd = -1;
119 u->ipv6_allow_map_fd = -1;
120 u->ipv4_deny_map_fd = -1;
121 u->ipv6_deny_map_fd = -1;
122
123 u->last_section_private = -1;
124
125 u->start_ratelimit = (RateLimit) { m->default_start_limit_interval, m->default_start_limit_burst };
126 u->auto_stop_ratelimit = (RateLimit) { 10 * USEC_PER_SEC, 16 };
127
128 for (CGroupIOAccountingMetric i = 0; i < _CGROUP_IO_ACCOUNTING_METRIC_MAX; i++)
129 u->io_accounting_last[i] = UINT64_MAX;
130
131 return u;
132 }
133
134 int unit_new_for_name(Manager *m, size_t size, const char *name, Unit **ret) {
135 _cleanup_(unit_freep) Unit *u = NULL;
136 int r;
137
138 u = unit_new(m, size);
139 if (!u)
140 return -ENOMEM;
141
142 r = unit_add_name(u, name);
143 if (r < 0)
144 return r;
145
146 *ret = TAKE_PTR(u);
147
148 return r;
149 }
150
151 bool unit_has_name(const Unit *u, const char *name) {
152 assert(u);
153 assert(name);
154
155 return set_contains(u->names, (char*) 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 ec->keyring_mode = MANAGER_IS_SYSTEM(u->manager) ?
191 EXEC_KEYRING_SHARED : EXEC_KEYRING_INHERIT;
192 }
193
194 kc = unit_get_kill_context(u);
195 if (kc)
196 kill_context_init(kc);
197
198 if (UNIT_VTABLE(u)->init)
199 UNIT_VTABLE(u)->init(u);
200 }
201
202 int unit_add_name(Unit *u, const char *text) {
203 _cleanup_free_ char *s = NULL, *i = NULL;
204 UnitType t;
205 int r;
206
207 assert(u);
208 assert(text);
209
210 if (unit_name_is_valid(text, UNIT_NAME_TEMPLATE)) {
211
212 if (!u->instance)
213 return -EINVAL;
214
215 r = unit_name_replace_instance(text, u->instance, &s);
216 if (r < 0)
217 return r;
218 } else {
219 s = strdup(text);
220 if (!s)
221 return -ENOMEM;
222 }
223
224 if (set_contains(u->names, s))
225 return 0;
226 if (hashmap_contains(u->manager->units, s))
227 return -EEXIST;
228
229 if (!unit_name_is_valid(s, UNIT_NAME_PLAIN|UNIT_NAME_INSTANCE))
230 return -EINVAL;
231
232 t = unit_name_to_type(s);
233 if (t < 0)
234 return -EINVAL;
235
236 if (u->type != _UNIT_TYPE_INVALID && t != u->type)
237 return -EINVAL;
238
239 r = unit_name_to_instance(s, &i);
240 if (r < 0)
241 return r;
242
243 if (i && !unit_type_may_template(t))
244 return -EINVAL;
245
246 /* Ensure that this unit is either instanced or not instanced,
247 * but not both. Note that we do allow names with different
248 * instance names however! */
249 if (u->type != _UNIT_TYPE_INVALID && !u->instance != !i)
250 return -EINVAL;
251
252 if (!unit_type_may_alias(t) && !set_isempty(u->names))
253 return -EEXIST;
254
255 if (hashmap_size(u->manager->units) >= MANAGER_MAX_NAMES)
256 return -E2BIG;
257
258 r = set_put(u->names, s);
259 if (r < 0)
260 return r;
261 assert(r > 0);
262
263 r = hashmap_put(u->manager->units, s, u);
264 if (r < 0) {
265 (void) set_remove(u->names, s);
266 return r;
267 }
268
269 if (u->type == _UNIT_TYPE_INVALID) {
270 u->type = t;
271 u->id = s;
272 u->instance = TAKE_PTR(i);
273
274 LIST_PREPEND(units_by_type, u->manager->units_by_type[t], u);
275
276 unit_init(u);
277 }
278
279 s = NULL;
280
281 unit_add_to_dbus_queue(u);
282 return 0;
283 }
284
285 int unit_choose_id(Unit *u, const char *name) {
286 _cleanup_free_ char *t = NULL;
287 char *s, *i;
288 int r;
289
290 assert(u);
291 assert(name);
292
293 if (unit_name_is_valid(name, UNIT_NAME_TEMPLATE)) {
294
295 if (!u->instance)
296 return -EINVAL;
297
298 r = unit_name_replace_instance(name, u->instance, &t);
299 if (r < 0)
300 return r;
301
302 name = t;
303 }
304
305 /* Selects one of the names of this unit as the id */
306 s = set_get(u->names, (char*) name);
307 if (!s)
308 return -ENOENT;
309
310 /* Determine the new instance from the new id */
311 r = unit_name_to_instance(s, &i);
312 if (r < 0)
313 return r;
314
315 u->id = s;
316
317 free(u->instance);
318 u->instance = i;
319
320 unit_add_to_dbus_queue(u);
321
322 return 0;
323 }
324
325 int unit_set_description(Unit *u, const char *description) {
326 int r;
327
328 assert(u);
329
330 r = free_and_strdup(&u->description, empty_to_null(description));
331 if (r < 0)
332 return r;
333 if (r > 0)
334 unit_add_to_dbus_queue(u);
335
336 return 0;
337 }
338
339 bool unit_may_gc(Unit *u) {
340 UnitActiveState state;
341 int r;
342
343 assert(u);
344
345 /* Checks whether the unit is ready to be unloaded for garbage collection.
346 * Returns true when the unit may be collected, and false if there's some
347 * reason to keep it loaded.
348 *
349 * References from other units are *not* checked here. Instead, this is done
350 * in unit_gc_sweep(), but using markers to properly collect dependency loops.
351 */
352
353 if (u->job)
354 return false;
355
356 if (u->nop_job)
357 return false;
358
359 state = unit_active_state(u);
360
361 /* If the unit is inactive and failed and no job is queued for it, then release its runtime resources */
362 if (UNIT_IS_INACTIVE_OR_FAILED(state) &&
363 UNIT_VTABLE(u)->release_resources)
364 UNIT_VTABLE(u)->release_resources(u);
365
366 if (u->perpetual)
367 return false;
368
369 if (sd_bus_track_count(u->bus_track) > 0)
370 return false;
371
372 /* But we keep the unit object around for longer when it is referenced or configured to not be gc'ed */
373 switch (u->collect_mode) {
374
375 case COLLECT_INACTIVE:
376 if (state != UNIT_INACTIVE)
377 return false;
378
379 break;
380
381 case COLLECT_INACTIVE_OR_FAILED:
382 if (!IN_SET(state, UNIT_INACTIVE, UNIT_FAILED))
383 return false;
384
385 break;
386
387 default:
388 assert_not_reached("Unknown garbage collection mode");
389 }
390
391 if (u->cgroup_path) {
392 /* If the unit has a cgroup, then check whether there's anything in it. If so, we should stay
393 * around. Units with active processes should never be collected. */
394
395 r = cg_is_empty_recursive(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path);
396 if (r < 0)
397 log_unit_debug_errno(u, r, "Failed to determine whether cgroup %s is empty: %m", u->cgroup_path);
398 if (r <= 0)
399 return false;
400 }
401
402 if (UNIT_VTABLE(u)->may_gc && !UNIT_VTABLE(u)->may_gc(u))
403 return false;
404
405 return true;
406 }
407
408 void unit_add_to_load_queue(Unit *u) {
409 assert(u);
410 assert(u->type != _UNIT_TYPE_INVALID);
411
412 if (u->load_state != UNIT_STUB || u->in_load_queue)
413 return;
414
415 LIST_PREPEND(load_queue, u->manager->load_queue, u);
416 u->in_load_queue = true;
417 }
418
419 void unit_add_to_cleanup_queue(Unit *u) {
420 assert(u);
421
422 if (u->in_cleanup_queue)
423 return;
424
425 LIST_PREPEND(cleanup_queue, u->manager->cleanup_queue, u);
426 u->in_cleanup_queue = true;
427 }
428
429 void unit_add_to_gc_queue(Unit *u) {
430 assert(u);
431
432 if (u->in_gc_queue || u->in_cleanup_queue)
433 return;
434
435 if (!unit_may_gc(u))
436 return;
437
438 LIST_PREPEND(gc_queue, u->manager->gc_unit_queue, u);
439 u->in_gc_queue = true;
440 }
441
442 void unit_add_to_dbus_queue(Unit *u) {
443 assert(u);
444 assert(u->type != _UNIT_TYPE_INVALID);
445
446 if (u->load_state == UNIT_STUB || u->in_dbus_queue)
447 return;
448
449 /* Shortcut things if nobody cares */
450 if (sd_bus_track_count(u->manager->subscribed) <= 0 &&
451 sd_bus_track_count(u->bus_track) <= 0 &&
452 set_isempty(u->manager->private_buses)) {
453 u->sent_dbus_new_signal = true;
454 return;
455 }
456
457 LIST_PREPEND(dbus_queue, u->manager->dbus_unit_queue, u);
458 u->in_dbus_queue = true;
459 }
460
461 void unit_submit_to_stop_when_unneeded_queue(Unit *u) {
462 assert(u);
463
464 if (u->in_stop_when_unneeded_queue)
465 return;
466
467 if (!u->stop_when_unneeded)
468 return;
469
470 if (!UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u)))
471 return;
472
473 LIST_PREPEND(stop_when_unneeded_queue, u->manager->stop_when_unneeded_queue, u);
474 u->in_stop_when_unneeded_queue = true;
475 }
476
477 static void bidi_set_free(Unit *u, Hashmap *h) {
478 Unit *other;
479 Iterator i;
480 void *v;
481
482 assert(u);
483
484 /* Frees the hashmap and makes sure we are dropped from the inverse pointers */
485
486 HASHMAP_FOREACH_KEY(v, other, h, i) {
487 UnitDependency d;
488
489 for (d = 0; d < _UNIT_DEPENDENCY_MAX; d++)
490 hashmap_remove(other->dependencies[d], u);
491
492 unit_add_to_gc_queue(other);
493 }
494
495 hashmap_free(h);
496 }
497
498 static void unit_remove_transient(Unit *u) {
499 char **i;
500
501 assert(u);
502
503 if (!u->transient)
504 return;
505
506 if (u->fragment_path)
507 (void) unlink(u->fragment_path);
508
509 STRV_FOREACH(i, u->dropin_paths) {
510 _cleanup_free_ char *p = NULL, *pp = NULL;
511
512 p = dirname_malloc(*i); /* Get the drop-in directory from the drop-in file */
513 if (!p)
514 continue;
515
516 pp = dirname_malloc(p); /* Get the config directory from the drop-in directory */
517 if (!pp)
518 continue;
519
520 /* Only drop transient drop-ins */
521 if (!path_equal(u->manager->lookup_paths.transient, pp))
522 continue;
523
524 (void) unlink(*i);
525 (void) rmdir(p);
526 }
527 }
528
529 static void unit_free_requires_mounts_for(Unit *u) {
530 assert(u);
531
532 for (;;) {
533 _cleanup_free_ char *path;
534
535 path = hashmap_steal_first_key(u->requires_mounts_for);
536 if (!path)
537 break;
538 else {
539 char s[strlen(path) + 1];
540
541 PATH_FOREACH_PREFIX_MORE(s, path) {
542 char *y;
543 Set *x;
544
545 x = hashmap_get2(u->manager->units_requiring_mounts_for, s, (void**) &y);
546 if (!x)
547 continue;
548
549 (void) set_remove(x, u);
550
551 if (set_isempty(x)) {
552 (void) hashmap_remove(u->manager->units_requiring_mounts_for, y);
553 free(y);
554 set_free(x);
555 }
556 }
557 }
558 }
559
560 u->requires_mounts_for = hashmap_free(u->requires_mounts_for);
561 }
562
563 static void unit_done(Unit *u) {
564 ExecContext *ec;
565 CGroupContext *cc;
566
567 assert(u);
568
569 if (u->type < 0)
570 return;
571
572 if (UNIT_VTABLE(u)->done)
573 UNIT_VTABLE(u)->done(u);
574
575 ec = unit_get_exec_context(u);
576 if (ec)
577 exec_context_done(ec);
578
579 cc = unit_get_cgroup_context(u);
580 if (cc)
581 cgroup_context_done(cc);
582 }
583
584 void unit_free(Unit *u) {
585 UnitDependency d;
586 Iterator i;
587 char *t;
588
589 if (!u)
590 return;
591
592 if (UNIT_ISSET(u->slice)) {
593 /* A unit is being dropped from the tree, make sure our parent slice recalculates the member mask */
594 unit_invalidate_cgroup_members_masks(UNIT_DEREF(u->slice));
595
596 /* And make sure the parent is realized again, updating cgroup memberships */
597 unit_add_to_cgroup_realize_queue(UNIT_DEREF(u->slice));
598 }
599
600 u->transient_file = safe_fclose(u->transient_file);
601
602 if (!MANAGER_IS_RELOADING(u->manager))
603 unit_remove_transient(u);
604
605 bus_unit_send_removed_signal(u);
606
607 unit_done(u);
608
609 unit_dequeue_rewatch_pids(u);
610
611 sd_bus_slot_unref(u->match_bus_slot);
612 sd_bus_track_unref(u->bus_track);
613 u->deserialized_refs = strv_free(u->deserialized_refs);
614
615 unit_free_requires_mounts_for(u);
616
617 SET_FOREACH(t, u->names, i)
618 hashmap_remove_value(u->manager->units, t, u);
619
620 if (!sd_id128_is_null(u->invocation_id))
621 hashmap_remove_value(u->manager->units_by_invocation_id, &u->invocation_id, u);
622
623 if (u->job) {
624 Job *j = u->job;
625 job_uninstall(j);
626 job_free(j);
627 }
628
629 if (u->nop_job) {
630 Job *j = u->nop_job;
631 job_uninstall(j);
632 job_free(j);
633 }
634
635 for (d = 0; d < _UNIT_DEPENDENCY_MAX; d++)
636 bidi_set_free(u, u->dependencies[d]);
637
638 if (u->on_console)
639 manager_unref_console(u->manager);
640
641 unit_release_cgroup(u);
642
643 if (!MANAGER_IS_RELOADING(u->manager))
644 unit_unlink_state_files(u);
645
646 unit_unref_uid_gid(u, false);
647
648 (void) manager_update_failed_units(u->manager, u, false);
649 set_remove(u->manager->startup_units, u);
650
651 unit_unwatch_all_pids(u);
652
653 unit_ref_unset(&u->slice);
654 while (u->refs_by_target)
655 unit_ref_unset(u->refs_by_target);
656
657 if (u->type != _UNIT_TYPE_INVALID)
658 LIST_REMOVE(units_by_type, u->manager->units_by_type[u->type], u);
659
660 if (u->in_load_queue)
661 LIST_REMOVE(load_queue, u->manager->load_queue, u);
662
663 if (u->in_dbus_queue)
664 LIST_REMOVE(dbus_queue, u->manager->dbus_unit_queue, u);
665
666 if (u->in_gc_queue)
667 LIST_REMOVE(gc_queue, u->manager->gc_unit_queue, u);
668
669 if (u->in_cgroup_realize_queue)
670 LIST_REMOVE(cgroup_realize_queue, u->manager->cgroup_realize_queue, u);
671
672 if (u->in_cgroup_empty_queue)
673 LIST_REMOVE(cgroup_empty_queue, u->manager->cgroup_empty_queue, u);
674
675 if (u->in_cleanup_queue)
676 LIST_REMOVE(cleanup_queue, u->manager->cleanup_queue, u);
677
678 if (u->in_target_deps_queue)
679 LIST_REMOVE(target_deps_queue, u->manager->target_deps_queue, u);
680
681 if (u->in_stop_when_unneeded_queue)
682 LIST_REMOVE(stop_when_unneeded_queue, u->manager->stop_when_unneeded_queue, u);
683
684 safe_close(u->ip_accounting_ingress_map_fd);
685 safe_close(u->ip_accounting_egress_map_fd);
686
687 safe_close(u->ipv4_allow_map_fd);
688 safe_close(u->ipv6_allow_map_fd);
689 safe_close(u->ipv4_deny_map_fd);
690 safe_close(u->ipv6_deny_map_fd);
691
692 bpf_program_unref(u->ip_bpf_ingress);
693 bpf_program_unref(u->ip_bpf_ingress_installed);
694 bpf_program_unref(u->ip_bpf_egress);
695 bpf_program_unref(u->ip_bpf_egress_installed);
696
697 set_free(u->ip_bpf_custom_ingress);
698 set_free(u->ip_bpf_custom_egress);
699 set_free(u->ip_bpf_custom_ingress_installed);
700 set_free(u->ip_bpf_custom_egress_installed);
701
702 bpf_program_unref(u->bpf_device_control_installed);
703
704 condition_free_list(u->conditions);
705 condition_free_list(u->asserts);
706
707 free(u->description);
708 strv_free(u->documentation);
709 free(u->fragment_path);
710 free(u->source_path);
711 strv_free(u->dropin_paths);
712 free(u->instance);
713
714 free(u->job_timeout_reboot_arg);
715
716 set_free_free(u->names);
717
718 free(u->reboot_arg);
719
720 free(u);
721 }
722
723 UnitActiveState unit_active_state(Unit *u) {
724 assert(u);
725
726 if (u->load_state == UNIT_MERGED)
727 return unit_active_state(unit_follow_merge(u));
728
729 /* After a reload it might happen that a unit is not correctly
730 * loaded but still has a process around. That's why we won't
731 * shortcut failed loading to UNIT_INACTIVE_FAILED. */
732
733 return UNIT_VTABLE(u)->active_state(u);
734 }
735
736 const char* unit_sub_state_to_string(Unit *u) {
737 assert(u);
738
739 return UNIT_VTABLE(u)->sub_state_to_string(u);
740 }
741
742 static int set_complete_move(Set **s, Set **other) {
743 assert(s);
744 assert(other);
745
746 if (!other)
747 return 0;
748
749 if (*s)
750 return set_move(*s, *other);
751 else
752 *s = TAKE_PTR(*other);
753
754 return 0;
755 }
756
757 static int hashmap_complete_move(Hashmap **s, Hashmap **other) {
758 assert(s);
759 assert(other);
760
761 if (!*other)
762 return 0;
763
764 if (*s)
765 return hashmap_move(*s, *other);
766 else
767 *s = TAKE_PTR(*other);
768
769 return 0;
770 }
771
772 static int merge_names(Unit *u, Unit *other) {
773 char *t;
774 Iterator i;
775 int r;
776
777 assert(u);
778 assert(other);
779
780 r = set_complete_move(&u->names, &other->names);
781 if (r < 0)
782 return r;
783
784 set_free_free(other->names);
785 other->names = NULL;
786 other->id = NULL;
787
788 SET_FOREACH(t, u->names, i)
789 assert_se(hashmap_replace(u->manager->units, t, u) == 0);
790
791 return 0;
792 }
793
794 static int reserve_dependencies(Unit *u, Unit *other, UnitDependency d) {
795 unsigned n_reserve;
796
797 assert(u);
798 assert(other);
799 assert(d < _UNIT_DEPENDENCY_MAX);
800
801 /*
802 * If u does not have this dependency set allocated, there is no need
803 * to reserve anything. In that case other's set will be transferred
804 * as a whole to u by complete_move().
805 */
806 if (!u->dependencies[d])
807 return 0;
808
809 /* merge_dependencies() will skip a u-on-u dependency */
810 n_reserve = hashmap_size(other->dependencies[d]) - !!hashmap_get(other->dependencies[d], u);
811
812 return hashmap_reserve(u->dependencies[d], n_reserve);
813 }
814
815 static void merge_dependencies(Unit *u, Unit *other, const char *other_id, UnitDependency d) {
816 Iterator i;
817 Unit *back;
818 void *v;
819 int r;
820
821 /* Merges all dependencies of type 'd' of the unit 'other' into the deps of the unit 'u' */
822
823 assert(u);
824 assert(other);
825 assert(d < _UNIT_DEPENDENCY_MAX);
826
827 /* Fix backwards pointers. Let's iterate through all dependent units of the other unit. */
828 HASHMAP_FOREACH_KEY(v, back, other->dependencies[d], i) {
829 UnitDependency k;
830
831 /* Let's now iterate through the dependencies of that dependencies of the other units, looking for
832 * pointers back, and let's fix them up, to instead point to 'u'. */
833
834 for (k = 0; k < _UNIT_DEPENDENCY_MAX; k++) {
835 if (back == u) {
836 /* Do not add dependencies between u and itself. */
837 if (hashmap_remove(back->dependencies[k], other))
838 maybe_warn_about_dependency(u, other_id, k);
839 } else {
840 UnitDependencyInfo di_u, di_other, di_merged;
841
842 /* Let's drop this dependency between "back" and "other", and let's create it between
843 * "back" and "u" instead. Let's merge the bit masks of the dependency we are moving,
844 * and any such dependency which might already exist */
845
846 di_other.data = hashmap_get(back->dependencies[k], other);
847 if (!di_other.data)
848 continue; /* dependency isn't set, let's try the next one */
849
850 di_u.data = hashmap_get(back->dependencies[k], u);
851
852 di_merged = (UnitDependencyInfo) {
853 .origin_mask = di_u.origin_mask | di_other.origin_mask,
854 .destination_mask = di_u.destination_mask | di_other.destination_mask,
855 };
856
857 r = hashmap_remove_and_replace(back->dependencies[k], other, u, di_merged.data);
858 if (r < 0)
859 log_warning_errno(r, "Failed to remove/replace: back=%s other=%s u=%s: %m", back->id, other_id, u->id);
860 assert(r >= 0);
861
862 /* assert_se(hashmap_remove_and_replace(back->dependencies[k], other, u, di_merged.data) >= 0); */
863 }
864 }
865
866 }
867
868 /* Also do not move dependencies on u to itself */
869 back = hashmap_remove(other->dependencies[d], u);
870 if (back)
871 maybe_warn_about_dependency(u, other_id, d);
872
873 /* The move cannot fail. The caller must have performed a reservation. */
874 assert_se(hashmap_complete_move(&u->dependencies[d], &other->dependencies[d]) == 0);
875
876 other->dependencies[d] = hashmap_free(other->dependencies[d]);
877 }
878
879 int unit_merge(Unit *u, Unit *other) {
880 UnitDependency d;
881 const char *other_id = NULL;
882 int r;
883
884 assert(u);
885 assert(other);
886 assert(u->manager == other->manager);
887 assert(u->type != _UNIT_TYPE_INVALID);
888
889 other = unit_follow_merge(other);
890
891 if (other == u)
892 return 0;
893
894 if (u->type != other->type)
895 return -EINVAL;
896
897 if (!u->instance != !other->instance)
898 return -EINVAL;
899
900 if (!unit_type_may_alias(u->type)) /* Merging only applies to unit names that support aliases */
901 return -EEXIST;
902
903 if (!IN_SET(other->load_state, UNIT_STUB, UNIT_NOT_FOUND))
904 return -EEXIST;
905
906 if (other->job)
907 return -EEXIST;
908
909 if (other->nop_job)
910 return -EEXIST;
911
912 if (!UNIT_IS_INACTIVE_OR_FAILED(unit_active_state(other)))
913 return -EEXIST;
914
915 if (other->id)
916 other_id = strdupa(other->id);
917
918 /* Make reservations to ensure merge_dependencies() won't fail */
919 for (d = 0; d < _UNIT_DEPENDENCY_MAX; d++) {
920 r = reserve_dependencies(u, other, d);
921 /*
922 * We don't rollback reservations if we fail. We don't have
923 * a way to undo reservations. A reservation is not a leak.
924 */
925 if (r < 0)
926 return r;
927 }
928
929 /* Merge names */
930 r = merge_names(u, other);
931 if (r < 0)
932 return r;
933
934 /* Redirect all references */
935 while (other->refs_by_target)
936 unit_ref_set(other->refs_by_target, other->refs_by_target->source, u);
937
938 /* Merge dependencies */
939 for (d = 0; d < _UNIT_DEPENDENCY_MAX; d++)
940 merge_dependencies(u, other, other_id, d);
941
942 other->load_state = UNIT_MERGED;
943 other->merged_into = u;
944
945 /* If there is still some data attached to the other node, we
946 * don't need it anymore, and can free it. */
947 if (other->load_state != UNIT_STUB)
948 if (UNIT_VTABLE(other)->done)
949 UNIT_VTABLE(other)->done(other);
950
951 unit_add_to_dbus_queue(u);
952 unit_add_to_cleanup_queue(other);
953
954 return 0;
955 }
956
957 int unit_merge_by_name(Unit *u, const char *name) {
958 _cleanup_free_ char *s = NULL;
959 Unit *other;
960 int r;
961
962 /* Either add name to u, or if a unit with name already exists, merge it with u.
963 * If name is a template, do the same for name@instance, where instance is u's instance. */
964
965 assert(u);
966 assert(name);
967
968 if (unit_name_is_valid(name, UNIT_NAME_TEMPLATE)) {
969 if (!u->instance)
970 return -EINVAL;
971
972 r = unit_name_replace_instance(name, u->instance, &s);
973 if (r < 0)
974 return r;
975
976 name = s;
977 }
978
979 other = manager_get_unit(u->manager, name);
980 if (other)
981 return unit_merge(u, other);
982
983 return unit_add_name(u, name);
984 }
985
986 Unit* unit_follow_merge(Unit *u) {
987 assert(u);
988
989 while (u->load_state == UNIT_MERGED)
990 assert_se(u = u->merged_into);
991
992 return u;
993 }
994
995 int unit_add_exec_dependencies(Unit *u, ExecContext *c) {
996 ExecDirectoryType dt;
997 char **dp;
998 int r;
999
1000 assert(u);
1001 assert(c);
1002
1003 if (c->working_directory && !c->working_directory_missing_ok) {
1004 r = unit_require_mounts_for(u, c->working_directory, UNIT_DEPENDENCY_FILE);
1005 if (r < 0)
1006 return r;
1007 }
1008
1009 if (c->root_directory) {
1010 r = unit_require_mounts_for(u, c->root_directory, UNIT_DEPENDENCY_FILE);
1011 if (r < 0)
1012 return r;
1013 }
1014
1015 if (c->root_image) {
1016 r = unit_require_mounts_for(u, c->root_image, UNIT_DEPENDENCY_FILE);
1017 if (r < 0)
1018 return r;
1019 }
1020
1021 for (dt = 0; dt < _EXEC_DIRECTORY_TYPE_MAX; dt++) {
1022 if (!u->manager->prefix[dt])
1023 continue;
1024
1025 STRV_FOREACH(dp, c->directories[dt].paths) {
1026 _cleanup_free_ char *p;
1027
1028 p = path_join(u->manager->prefix[dt], *dp);
1029 if (!p)
1030 return -ENOMEM;
1031
1032 r = unit_require_mounts_for(u, p, UNIT_DEPENDENCY_FILE);
1033 if (r < 0)
1034 return r;
1035 }
1036 }
1037
1038 if (!MANAGER_IS_SYSTEM(u->manager))
1039 return 0;
1040
1041 if (c->private_tmp) {
1042 const char *p;
1043
1044 FOREACH_STRING(p, "/tmp", "/var/tmp") {
1045 r = unit_require_mounts_for(u, p, UNIT_DEPENDENCY_FILE);
1046 if (r < 0)
1047 return r;
1048 }
1049
1050 r = unit_add_dependency_by_name(u, UNIT_AFTER, SPECIAL_TMPFILES_SETUP_SERVICE, true, UNIT_DEPENDENCY_FILE);
1051 if (r < 0)
1052 return r;
1053 }
1054
1055 if (!IN_SET(c->std_output,
1056 EXEC_OUTPUT_JOURNAL, EXEC_OUTPUT_JOURNAL_AND_CONSOLE,
1057 EXEC_OUTPUT_KMSG, EXEC_OUTPUT_KMSG_AND_CONSOLE,
1058 EXEC_OUTPUT_SYSLOG, EXEC_OUTPUT_SYSLOG_AND_CONSOLE) &&
1059 !IN_SET(c->std_error,
1060 EXEC_OUTPUT_JOURNAL, EXEC_OUTPUT_JOURNAL_AND_CONSOLE,
1061 EXEC_OUTPUT_KMSG, EXEC_OUTPUT_KMSG_AND_CONSOLE,
1062 EXEC_OUTPUT_SYSLOG, EXEC_OUTPUT_SYSLOG_AND_CONSOLE))
1063 return 0;
1064
1065 /* If syslog or kernel logging is requested, make sure our own
1066 * logging daemon is run first. */
1067
1068 r = unit_add_dependency_by_name(u, UNIT_AFTER, SPECIAL_JOURNALD_SOCKET, true, UNIT_DEPENDENCY_FILE);
1069 if (r < 0)
1070 return r;
1071
1072 return 0;
1073 }
1074
1075 const char *unit_description(Unit *u) {
1076 assert(u);
1077
1078 if (u->description)
1079 return u->description;
1080
1081 return strna(u->id);
1082 }
1083
1084 const char *unit_status_string(Unit *u) {
1085 assert(u);
1086
1087 if (u->manager->status_unit_format == STATUS_UNIT_FORMAT_NAME && u->id)
1088 return u->id;
1089
1090 return unit_description(u);
1091 }
1092
1093 static void print_unit_dependency_mask(FILE *f, const char *kind, UnitDependencyMask mask, bool *space) {
1094 const struct {
1095 UnitDependencyMask mask;
1096 const char *name;
1097 } table[] = {
1098 { UNIT_DEPENDENCY_FILE, "file" },
1099 { UNIT_DEPENDENCY_IMPLICIT, "implicit" },
1100 { UNIT_DEPENDENCY_DEFAULT, "default" },
1101 { UNIT_DEPENDENCY_UDEV, "udev" },
1102 { UNIT_DEPENDENCY_PATH, "path" },
1103 { UNIT_DEPENDENCY_MOUNTINFO_IMPLICIT, "mountinfo-implicit" },
1104 { UNIT_DEPENDENCY_MOUNTINFO_DEFAULT, "mountinfo-default" },
1105 { UNIT_DEPENDENCY_PROC_SWAP, "proc-swap" },
1106 };
1107 size_t i;
1108
1109 assert(f);
1110 assert(kind);
1111 assert(space);
1112
1113 for (i = 0; i < ELEMENTSOF(table); i++) {
1114
1115 if (mask == 0)
1116 break;
1117
1118 if (FLAGS_SET(mask, table[i].mask)) {
1119 if (*space)
1120 fputc(' ', f);
1121 else
1122 *space = true;
1123
1124 fputs(kind, f);
1125 fputs("-", f);
1126 fputs(table[i].name, f);
1127
1128 mask &= ~table[i].mask;
1129 }
1130 }
1131
1132 assert(mask == 0);
1133 }
1134
1135 void unit_dump(Unit *u, FILE *f, const char *prefix) {
1136 char *t, **j;
1137 UnitDependency d;
1138 Iterator i;
1139 const char *prefix2;
1140 char timestamp[5][FORMAT_TIMESTAMP_MAX], timespan[FORMAT_TIMESPAN_MAX];
1141 Unit *following;
1142 _cleanup_set_free_ Set *following_set = NULL;
1143 const char *n;
1144 CGroupMask m;
1145 int r;
1146
1147 assert(u);
1148 assert(u->type >= 0);
1149
1150 prefix = strempty(prefix);
1151 prefix2 = strjoina(prefix, "\t");
1152
1153 fprintf(f,
1154 "%s-> Unit %s:\n",
1155 prefix, u->id);
1156
1157 SET_FOREACH(t, u->names, i)
1158 if (!streq(t, u->id))
1159 fprintf(f, "%s\tAlias: %s\n", prefix, t);
1160
1161 fprintf(f,
1162 "%s\tDescription: %s\n"
1163 "%s\tInstance: %s\n"
1164 "%s\tUnit Load State: %s\n"
1165 "%s\tUnit Active State: %s\n"
1166 "%s\tState Change Timestamp: %s\n"
1167 "%s\tInactive Exit Timestamp: %s\n"
1168 "%s\tActive Enter Timestamp: %s\n"
1169 "%s\tActive Exit Timestamp: %s\n"
1170 "%s\tInactive Enter Timestamp: %s\n"
1171 "%s\tMay GC: %s\n"
1172 "%s\tNeed Daemon Reload: %s\n"
1173 "%s\tTransient: %s\n"
1174 "%s\tPerpetual: %s\n"
1175 "%s\tGarbage Collection Mode: %s\n"
1176 "%s\tSlice: %s\n"
1177 "%s\tCGroup: %s\n"
1178 "%s\tCGroup realized: %s\n",
1179 prefix, unit_description(u),
1180 prefix, strna(u->instance),
1181 prefix, unit_load_state_to_string(u->load_state),
1182 prefix, unit_active_state_to_string(unit_active_state(u)),
1183 prefix, strna(format_timestamp(timestamp[0], sizeof(timestamp[0]), u->state_change_timestamp.realtime)),
1184 prefix, strna(format_timestamp(timestamp[1], sizeof(timestamp[1]), u->inactive_exit_timestamp.realtime)),
1185 prefix, strna(format_timestamp(timestamp[2], sizeof(timestamp[2]), u->active_enter_timestamp.realtime)),
1186 prefix, strna(format_timestamp(timestamp[3], sizeof(timestamp[3]), u->active_exit_timestamp.realtime)),
1187 prefix, strna(format_timestamp(timestamp[4], sizeof(timestamp[4]), u->inactive_enter_timestamp.realtime)),
1188 prefix, yes_no(unit_may_gc(u)),
1189 prefix, yes_no(unit_need_daemon_reload(u)),
1190 prefix, yes_no(u->transient),
1191 prefix, yes_no(u->perpetual),
1192 prefix, collect_mode_to_string(u->collect_mode),
1193 prefix, strna(unit_slice_name(u)),
1194 prefix, strna(u->cgroup_path),
1195 prefix, yes_no(u->cgroup_realized));
1196
1197 if (u->cgroup_realized_mask != 0) {
1198 _cleanup_free_ char *s = NULL;
1199 (void) cg_mask_to_string(u->cgroup_realized_mask, &s);
1200 fprintf(f, "%s\tCGroup realized mask: %s\n", prefix, strnull(s));
1201 }
1202
1203 if (u->cgroup_enabled_mask != 0) {
1204 _cleanup_free_ char *s = NULL;
1205 (void) cg_mask_to_string(u->cgroup_enabled_mask, &s);
1206 fprintf(f, "%s\tCGroup enabled mask: %s\n", prefix, strnull(s));
1207 }
1208
1209 m = unit_get_own_mask(u);
1210 if (m != 0) {
1211 _cleanup_free_ char *s = NULL;
1212 (void) cg_mask_to_string(m, &s);
1213 fprintf(f, "%s\tCGroup own mask: %s\n", prefix, strnull(s));
1214 }
1215
1216 m = unit_get_members_mask(u);
1217 if (m != 0) {
1218 _cleanup_free_ char *s = NULL;
1219 (void) cg_mask_to_string(m, &s);
1220 fprintf(f, "%s\tCGroup members mask: %s\n", prefix, strnull(s));
1221 }
1222
1223 m = unit_get_delegate_mask(u);
1224 if (m != 0) {
1225 _cleanup_free_ char *s = NULL;
1226 (void) cg_mask_to_string(m, &s);
1227 fprintf(f, "%s\tCGroup delegate mask: %s\n", prefix, strnull(s));
1228 }
1229
1230 if (!sd_id128_is_null(u->invocation_id))
1231 fprintf(f, "%s\tInvocation ID: " SD_ID128_FORMAT_STR "\n",
1232 prefix, SD_ID128_FORMAT_VAL(u->invocation_id));
1233
1234 STRV_FOREACH(j, u->documentation)
1235 fprintf(f, "%s\tDocumentation: %s\n", prefix, *j);
1236
1237 following = unit_following(u);
1238 if (following)
1239 fprintf(f, "%s\tFollowing: %s\n", prefix, following->id);
1240
1241 r = unit_following_set(u, &following_set);
1242 if (r >= 0) {
1243 Unit *other;
1244
1245 SET_FOREACH(other, following_set, i)
1246 fprintf(f, "%s\tFollowing Set Member: %s\n", prefix, other->id);
1247 }
1248
1249 if (u->fragment_path)
1250 fprintf(f, "%s\tFragment Path: %s\n", prefix, u->fragment_path);
1251
1252 if (u->source_path)
1253 fprintf(f, "%s\tSource Path: %s\n", prefix, u->source_path);
1254
1255 STRV_FOREACH(j, u->dropin_paths)
1256 fprintf(f, "%s\tDropIn Path: %s\n", prefix, *j);
1257
1258 if (u->failure_action != EMERGENCY_ACTION_NONE)
1259 fprintf(f, "%s\tFailure Action: %s\n", prefix, emergency_action_to_string(u->failure_action));
1260 if (u->failure_action_exit_status >= 0)
1261 fprintf(f, "%s\tFailure Action Exit Status: %i\n", prefix, u->failure_action_exit_status);
1262 if (u->success_action != EMERGENCY_ACTION_NONE)
1263 fprintf(f, "%s\tSuccess Action: %s\n", prefix, emergency_action_to_string(u->success_action));
1264 if (u->success_action_exit_status >= 0)
1265 fprintf(f, "%s\tSuccess Action Exit Status: %i\n", prefix, u->success_action_exit_status);
1266
1267 if (u->job_timeout != USEC_INFINITY)
1268 fprintf(f, "%s\tJob Timeout: %s\n", prefix, format_timespan(timespan, sizeof(timespan), u->job_timeout, 0));
1269
1270 if (u->job_timeout_action != EMERGENCY_ACTION_NONE)
1271 fprintf(f, "%s\tJob Timeout Action: %s\n", prefix, emergency_action_to_string(u->job_timeout_action));
1272
1273 if (u->job_timeout_reboot_arg)
1274 fprintf(f, "%s\tJob Timeout Reboot Argument: %s\n", prefix, u->job_timeout_reboot_arg);
1275
1276 condition_dump_list(u->conditions, f, prefix, condition_type_to_string);
1277 condition_dump_list(u->asserts, f, prefix, assert_type_to_string);
1278
1279 if (dual_timestamp_is_set(&u->condition_timestamp))
1280 fprintf(f,
1281 "%s\tCondition Timestamp: %s\n"
1282 "%s\tCondition Result: %s\n",
1283 prefix, strna(format_timestamp(timestamp[0], sizeof(timestamp[0]), u->condition_timestamp.realtime)),
1284 prefix, yes_no(u->condition_result));
1285
1286 if (dual_timestamp_is_set(&u->assert_timestamp))
1287 fprintf(f,
1288 "%s\tAssert Timestamp: %s\n"
1289 "%s\tAssert Result: %s\n",
1290 prefix, strna(format_timestamp(timestamp[0], sizeof(timestamp[0]), u->assert_timestamp.realtime)),
1291 prefix, yes_no(u->assert_result));
1292
1293 for (d = 0; d < _UNIT_DEPENDENCY_MAX; d++) {
1294 UnitDependencyInfo di;
1295 Unit *other;
1296
1297 HASHMAP_FOREACH_KEY(di.data, other, u->dependencies[d], i) {
1298 bool space = false;
1299
1300 fprintf(f, "%s\t%s: %s (", prefix, unit_dependency_to_string(d), other->id);
1301
1302 print_unit_dependency_mask(f, "origin", di.origin_mask, &space);
1303 print_unit_dependency_mask(f, "destination", di.destination_mask, &space);
1304
1305 fputs(")\n", f);
1306 }
1307 }
1308
1309 if (!hashmap_isempty(u->requires_mounts_for)) {
1310 UnitDependencyInfo di;
1311 const char *path;
1312
1313 HASHMAP_FOREACH_KEY(di.data, path, u->requires_mounts_for, i) {
1314 bool space = false;
1315
1316 fprintf(f, "%s\tRequiresMountsFor: %s (", prefix, path);
1317
1318 print_unit_dependency_mask(f, "origin", di.origin_mask, &space);
1319 print_unit_dependency_mask(f, "destination", di.destination_mask, &space);
1320
1321 fputs(")\n", f);
1322 }
1323 }
1324
1325 if (u->load_state == UNIT_LOADED) {
1326
1327 fprintf(f,
1328 "%s\tStopWhenUnneeded: %s\n"
1329 "%s\tRefuseManualStart: %s\n"
1330 "%s\tRefuseManualStop: %s\n"
1331 "%s\tDefaultDependencies: %s\n"
1332 "%s\tOnFailureJobMode: %s\n"
1333 "%s\tIgnoreOnIsolate: %s\n",
1334 prefix, yes_no(u->stop_when_unneeded),
1335 prefix, yes_no(u->refuse_manual_start),
1336 prefix, yes_no(u->refuse_manual_stop),
1337 prefix, yes_no(u->default_dependencies),
1338 prefix, job_mode_to_string(u->on_failure_job_mode),
1339 prefix, yes_no(u->ignore_on_isolate));
1340
1341 if (UNIT_VTABLE(u)->dump)
1342 UNIT_VTABLE(u)->dump(u, f, prefix2);
1343
1344 } else if (u->load_state == UNIT_MERGED)
1345 fprintf(f,
1346 "%s\tMerged into: %s\n",
1347 prefix, u->merged_into->id);
1348 else if (u->load_state == UNIT_ERROR)
1349 fprintf(f, "%s\tLoad Error Code: %s\n", prefix, strerror_safe(u->load_error));
1350
1351 for (n = sd_bus_track_first(u->bus_track); n; n = sd_bus_track_next(u->bus_track))
1352 fprintf(f, "%s\tBus Ref: %s\n", prefix, n);
1353
1354 if (u->job)
1355 job_dump(u->job, f, prefix2);
1356
1357 if (u->nop_job)
1358 job_dump(u->nop_job, f, prefix2);
1359 }
1360
1361 /* Common implementation for multiple backends */
1362 int unit_load_fragment_and_dropin(Unit *u, bool fragment_required) {
1363 int r;
1364
1365 assert(u);
1366
1367 /* Load a .{service,socket,...} file */
1368 r = unit_load_fragment(u);
1369 if (r < 0)
1370 return r;
1371
1372 if (u->load_state == UNIT_STUB) {
1373 if (fragment_required)
1374 return -ENOENT;
1375
1376 u->load_state = UNIT_LOADED;
1377 }
1378
1379 /* Load drop-in directory data. If u is an alias, we might be reloading the
1380 * target unit needlessly. But we cannot be sure which drops-ins have already
1381 * been loaded and which not, at least without doing complicated book-keeping,
1382 * so let's always reread all drop-ins. */
1383 return unit_load_dropin(unit_follow_merge(u));
1384 }
1385
1386 void unit_add_to_target_deps_queue(Unit *u) {
1387 Manager *m = u->manager;
1388
1389 assert(u);
1390
1391 if (u->in_target_deps_queue)
1392 return;
1393
1394 LIST_PREPEND(target_deps_queue, m->target_deps_queue, u);
1395 u->in_target_deps_queue = true;
1396 }
1397
1398 int unit_add_default_target_dependency(Unit *u, Unit *target) {
1399 assert(u);
1400 assert(target);
1401
1402 if (target->type != UNIT_TARGET)
1403 return 0;
1404
1405 /* Only add the dependency if both units are loaded, so that
1406 * that loop check below is reliable */
1407 if (u->load_state != UNIT_LOADED ||
1408 target->load_state != UNIT_LOADED)
1409 return 0;
1410
1411 /* If either side wants no automatic dependencies, then let's
1412 * skip this */
1413 if (!u->default_dependencies ||
1414 !target->default_dependencies)
1415 return 0;
1416
1417 /* Don't create loops */
1418 if (hashmap_get(target->dependencies[UNIT_BEFORE], u))
1419 return 0;
1420
1421 return unit_add_dependency(target, UNIT_AFTER, u, true, UNIT_DEPENDENCY_DEFAULT);
1422 }
1423
1424 static int unit_add_slice_dependencies(Unit *u) {
1425 UnitDependencyMask mask;
1426 assert(u);
1427
1428 if (!UNIT_HAS_CGROUP_CONTEXT(u))
1429 return 0;
1430
1431 /* Slice units are implicitly ordered against their parent slices (as this relationship is encoded in the
1432 name), while all other units are ordered based on configuration (as in their case Slice= configures the
1433 relationship). */
1434 mask = u->type == UNIT_SLICE ? UNIT_DEPENDENCY_IMPLICIT : UNIT_DEPENDENCY_FILE;
1435
1436 if (UNIT_ISSET(u->slice))
1437 return unit_add_two_dependencies(u, UNIT_AFTER, UNIT_REQUIRES, UNIT_DEREF(u->slice), true, mask);
1438
1439 if (unit_has_name(u, SPECIAL_ROOT_SLICE))
1440 return 0;
1441
1442 return unit_add_two_dependencies_by_name(u, UNIT_AFTER, UNIT_REQUIRES, SPECIAL_ROOT_SLICE, true, mask);
1443 }
1444
1445 static int unit_add_mount_dependencies(Unit *u) {
1446 UnitDependencyInfo di;
1447 const char *path;
1448 Iterator i;
1449 int r;
1450
1451 assert(u);
1452
1453 HASHMAP_FOREACH_KEY(di.data, path, u->requires_mounts_for, i) {
1454 char prefix[strlen(path) + 1];
1455
1456 PATH_FOREACH_PREFIX_MORE(prefix, path) {
1457 _cleanup_free_ char *p = NULL;
1458 Unit *m;
1459
1460 r = unit_name_from_path(prefix, ".mount", &p);
1461 if (r < 0)
1462 return r;
1463
1464 m = manager_get_unit(u->manager, p);
1465 if (!m) {
1466 /* Make sure to load the mount unit if
1467 * it exists. If so the dependencies
1468 * on this unit will be added later
1469 * during the loading of the mount
1470 * unit. */
1471 (void) manager_load_unit_prepare(u->manager, p, NULL, NULL, &m);
1472 continue;
1473 }
1474 if (m == u)
1475 continue;
1476
1477 if (m->load_state != UNIT_LOADED)
1478 continue;
1479
1480 r = unit_add_dependency(u, UNIT_AFTER, m, true, di.origin_mask);
1481 if (r < 0)
1482 return r;
1483
1484 if (m->fragment_path) {
1485 r = unit_add_dependency(u, UNIT_REQUIRES, m, true, di.origin_mask);
1486 if (r < 0)
1487 return r;
1488 }
1489 }
1490 }
1491
1492 return 0;
1493 }
1494
1495 static int unit_add_startup_units(Unit *u) {
1496 CGroupContext *c;
1497 int r;
1498
1499 c = unit_get_cgroup_context(u);
1500 if (!c)
1501 return 0;
1502
1503 if (c->startup_cpu_shares == CGROUP_CPU_SHARES_INVALID &&
1504 c->startup_io_weight == CGROUP_WEIGHT_INVALID &&
1505 c->startup_blockio_weight == CGROUP_BLKIO_WEIGHT_INVALID)
1506 return 0;
1507
1508 r = set_ensure_allocated(&u->manager->startup_units, NULL);
1509 if (r < 0)
1510 return r;
1511
1512 return set_put(u->manager->startup_units, u);
1513 }
1514
1515 int unit_load(Unit *u) {
1516 int r;
1517
1518 assert(u);
1519
1520 if (u->in_load_queue) {
1521 LIST_REMOVE(load_queue, u->manager->load_queue, u);
1522 u->in_load_queue = false;
1523 }
1524
1525 if (u->type == _UNIT_TYPE_INVALID)
1526 return -EINVAL;
1527
1528 if (u->load_state != UNIT_STUB)
1529 return 0;
1530
1531 if (u->transient_file) {
1532 /* Finalize transient file: if this is a transient unit file, as soon as we reach unit_load() the setup
1533 * is complete, hence let's synchronize the unit file we just wrote to disk. */
1534
1535 r = fflush_and_check(u->transient_file);
1536 if (r < 0)
1537 goto fail;
1538
1539 u->transient_file = safe_fclose(u->transient_file);
1540 u->fragment_mtime = now(CLOCK_REALTIME);
1541 }
1542
1543 r = UNIT_VTABLE(u)->load(u);
1544 if (r < 0)
1545 goto fail;
1546
1547 assert(u->load_state != UNIT_STUB);
1548
1549 if (u->load_state == UNIT_LOADED) {
1550 unit_add_to_target_deps_queue(u);
1551
1552 r = unit_add_slice_dependencies(u);
1553 if (r < 0)
1554 goto fail;
1555
1556 r = unit_add_mount_dependencies(u);
1557 if (r < 0)
1558 goto fail;
1559
1560 r = unit_add_startup_units(u);
1561 if (r < 0)
1562 goto fail;
1563
1564 if (u->on_failure_job_mode == JOB_ISOLATE && hashmap_size(u->dependencies[UNIT_ON_FAILURE]) > 1) {
1565 log_unit_error(u, "More than one OnFailure= dependencies specified but OnFailureJobMode=isolate set. Refusing.");
1566 r = -ENOEXEC;
1567 goto fail;
1568 }
1569
1570 if (u->job_running_timeout != USEC_INFINITY && u->job_running_timeout > u->job_timeout)
1571 log_unit_warning(u, "JobRunningTimeoutSec= is greater than JobTimeoutSec=, it has no effect.");
1572
1573 /* We finished loading, let's ensure our parents recalculate the members mask */
1574 unit_invalidate_cgroup_members_masks(u);
1575 }
1576
1577 assert((u->load_state != UNIT_MERGED) == !u->merged_into);
1578
1579 unit_add_to_dbus_queue(unit_follow_merge(u));
1580 unit_add_to_gc_queue(u);
1581
1582 return 0;
1583
1584 fail:
1585 /* We convert ENOEXEC errors to the UNIT_BAD_SETTING load state here. Configuration parsing code should hence
1586 * return ENOEXEC to ensure units are placed in this state after loading */
1587
1588 u->load_state = u->load_state == UNIT_STUB ? UNIT_NOT_FOUND :
1589 r == -ENOEXEC ? UNIT_BAD_SETTING :
1590 UNIT_ERROR;
1591 u->load_error = r;
1592
1593 unit_add_to_dbus_queue(u);
1594 unit_add_to_gc_queue(u);
1595
1596 return log_unit_debug_errno(u, r, "Failed to load configuration: %m");
1597 }
1598
1599 _printf_(7, 8)
1600 static int log_unit_internal(void *userdata, int level, int error, const char *file, int line, const char *func, const char *format, ...) {
1601 Unit *u = userdata;
1602 va_list ap;
1603 int r;
1604
1605 va_start(ap, format);
1606 if (u)
1607 r = log_object_internalv(level, error, file, line, func,
1608 u->manager->unit_log_field,
1609 u->id,
1610 u->manager->invocation_log_field,
1611 u->invocation_id_string,
1612 format, ap);
1613 else
1614 r = log_internalv(level, error, file, line, func, format, ap);
1615 va_end(ap);
1616
1617 return r;
1618 }
1619
1620 static bool unit_test_condition(Unit *u) {
1621 assert(u);
1622
1623 dual_timestamp_get(&u->condition_timestamp);
1624 u->condition_result = condition_test_list(u->conditions, condition_type_to_string, log_unit_internal, u);
1625
1626 unit_add_to_dbus_queue(u);
1627
1628 return u->condition_result;
1629 }
1630
1631 static bool unit_test_assert(Unit *u) {
1632 assert(u);
1633
1634 dual_timestamp_get(&u->assert_timestamp);
1635 u->assert_result = condition_test_list(u->asserts, assert_type_to_string, log_unit_internal, u);
1636
1637 unit_add_to_dbus_queue(u);
1638
1639 return u->assert_result;
1640 }
1641
1642 void unit_status_printf(Unit *u, const char *status, const char *unit_status_msg_format) {
1643 const char *d;
1644
1645 d = unit_status_string(u);
1646 if (log_get_show_color())
1647 d = strjoina(ANSI_HIGHLIGHT, d, ANSI_NORMAL);
1648
1649 DISABLE_WARNING_FORMAT_NONLITERAL;
1650 manager_status_printf(u->manager, STATUS_TYPE_NORMAL, status, unit_status_msg_format, d);
1651 REENABLE_WARNING;
1652 }
1653
1654 int unit_test_start_limit(Unit *u) {
1655 const char *reason;
1656
1657 assert(u);
1658
1659 if (ratelimit_below(&u->start_ratelimit)) {
1660 u->start_limit_hit = false;
1661 return 0;
1662 }
1663
1664 log_unit_warning(u, "Start request repeated too quickly.");
1665 u->start_limit_hit = true;
1666
1667 reason = strjoina("unit ", u->id, " failed");
1668
1669 emergency_action(u->manager, u->start_limit_action,
1670 EMERGENCY_ACTION_IS_WATCHDOG|EMERGENCY_ACTION_WARN,
1671 u->reboot_arg, -1, reason);
1672
1673 return -ECANCELED;
1674 }
1675
1676 bool unit_shall_confirm_spawn(Unit *u) {
1677 assert(u);
1678
1679 if (manager_is_confirm_spawn_disabled(u->manager))
1680 return false;
1681
1682 /* For some reasons units remaining in the same process group
1683 * as PID 1 fail to acquire the console even if it's not used
1684 * by any process. So skip the confirmation question for them. */
1685 return !unit_get_exec_context(u)->same_pgrp;
1686 }
1687
1688 static bool unit_verify_deps(Unit *u) {
1689 Unit *other;
1690 Iterator j;
1691 void *v;
1692
1693 assert(u);
1694
1695 /* Checks whether all BindsTo= dependencies of this unit are fulfilled — if they are also combined with
1696 * After=. We do not check Requires= or Requisite= here as they only should have an effect on the job
1697 * processing, but do not have any effect afterwards. We don't check BindsTo= dependencies that are not used in
1698 * conjunction with After= as for them any such check would make things entirely racy. */
1699
1700 HASHMAP_FOREACH_KEY(v, other, u->dependencies[UNIT_BINDS_TO], j) {
1701
1702 if (!hashmap_contains(u->dependencies[UNIT_AFTER], other))
1703 continue;
1704
1705 if (!UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(other))) {
1706 log_unit_notice(u, "Bound to unit %s, but unit isn't active.", other->id);
1707 return false;
1708 }
1709 }
1710
1711 return true;
1712 }
1713
1714 /* Errors that aren't really errors:
1715 * -EALREADY: Unit is already started.
1716 * -ECOMM: Condition failed
1717 * -EAGAIN: An operation is already in progress. Retry later.
1718 *
1719 * Errors that are real errors:
1720 * -EBADR: This unit type does not support starting.
1721 * -ECANCELED: Start limit hit, too many requests for now
1722 * -EPROTO: Assert failed
1723 * -EINVAL: Unit not loaded
1724 * -EOPNOTSUPP: Unit type not supported
1725 * -ENOLINK: The necessary dependencies are not fulfilled.
1726 * -ESTALE: This unit has been started before and can't be started a second time
1727 * -ENOENT: This is a triggering unit and unit to trigger is not loaded
1728 */
1729 int unit_start(Unit *u) {
1730 UnitActiveState state;
1731 Unit *following;
1732
1733 assert(u);
1734
1735 /* If this is already started, then this will succeed. Note that this will even succeed if this unit
1736 * is not startable by the user. This is relied on to detect when we need to wait for units and when
1737 * waiting is finished. */
1738 state = unit_active_state(u);
1739 if (UNIT_IS_ACTIVE_OR_RELOADING(state))
1740 return -EALREADY;
1741 if (state == UNIT_MAINTENANCE)
1742 return -EAGAIN;
1743
1744 /* Units that aren't loaded cannot be started */
1745 if (u->load_state != UNIT_LOADED)
1746 return -EINVAL;
1747
1748 /* Refuse starting scope units more than once */
1749 if (UNIT_VTABLE(u)->once_only && dual_timestamp_is_set(&u->inactive_enter_timestamp))
1750 return -ESTALE;
1751
1752 /* If the conditions failed, don't do anything at all. If we already are activating this call might
1753 * still be useful to speed up activation in case there is some hold-off time, but we don't want to
1754 * recheck the condition in that case. */
1755 if (state != UNIT_ACTIVATING &&
1756 !unit_test_condition(u))
1757 return log_unit_debug_errno(u, SYNTHETIC_ERRNO(ECOMM), "Starting requested but condition failed. Not starting unit.");
1758
1759 /* If the asserts failed, fail the entire job */
1760 if (state != UNIT_ACTIVATING &&
1761 !unit_test_assert(u))
1762 return log_unit_notice_errno(u, SYNTHETIC_ERRNO(EPROTO), "Starting requested but asserts failed.");
1763
1764 /* Units of types that aren't supported cannot be started. Note that we do this test only after the
1765 * condition checks, so that we rather return condition check errors (which are usually not
1766 * considered a true failure) than "not supported" errors (which are considered a failure).
1767 */
1768 if (!unit_type_supported(u->type))
1769 return -EOPNOTSUPP;
1770
1771 /* Let's make sure that the deps really are in order before we start this. Normally the job engine
1772 * should have taken care of this already, but let's check this here again. After all, our
1773 * dependencies might not be in effect anymore, due to a reload or due to a failed condition. */
1774 if (!unit_verify_deps(u))
1775 return -ENOLINK;
1776
1777 /* Forward to the main object, if we aren't it. */
1778 following = unit_following(u);
1779 if (following) {
1780 log_unit_debug(u, "Redirecting start request from %s to %s.", u->id, following->id);
1781 return unit_start(following);
1782 }
1783
1784 /* If it is stopped, but we cannot start it, then fail */
1785 if (!UNIT_VTABLE(u)->start)
1786 return -EBADR;
1787
1788 /* We don't suppress calls to ->start() here when we are already starting, to allow this request to
1789 * be used as a "hurry up" call, for example when the unit is in some "auto restart" state where it
1790 * waits for a holdoff timer to elapse before it will start again. */
1791
1792 unit_add_to_dbus_queue(u);
1793
1794 return UNIT_VTABLE(u)->start(u);
1795 }
1796
1797 bool unit_can_start(Unit *u) {
1798 assert(u);
1799
1800 if (u->load_state != UNIT_LOADED)
1801 return false;
1802
1803 if (!unit_type_supported(u->type))
1804 return false;
1805
1806 /* Scope units may be started only once */
1807 if (UNIT_VTABLE(u)->once_only && dual_timestamp_is_set(&u->inactive_exit_timestamp))
1808 return false;
1809
1810 return !!UNIT_VTABLE(u)->start;
1811 }
1812
1813 bool unit_can_isolate(Unit *u) {
1814 assert(u);
1815
1816 return unit_can_start(u) &&
1817 u->allow_isolate;
1818 }
1819
1820 /* Errors:
1821 * -EBADR: This unit type does not support stopping.
1822 * -EALREADY: Unit is already stopped.
1823 * -EAGAIN: An operation is already in progress. Retry later.
1824 */
1825 int unit_stop(Unit *u) {
1826 UnitActiveState state;
1827 Unit *following;
1828
1829 assert(u);
1830
1831 state = unit_active_state(u);
1832 if (UNIT_IS_INACTIVE_OR_FAILED(state))
1833 return -EALREADY;
1834
1835 following = unit_following(u);
1836 if (following) {
1837 log_unit_debug(u, "Redirecting stop request from %s to %s.", u->id, following->id);
1838 return unit_stop(following);
1839 }
1840
1841 if (!UNIT_VTABLE(u)->stop)
1842 return -EBADR;
1843
1844 unit_add_to_dbus_queue(u);
1845
1846 return UNIT_VTABLE(u)->stop(u);
1847 }
1848
1849 bool unit_can_stop(Unit *u) {
1850 assert(u);
1851
1852 if (!unit_type_supported(u->type))
1853 return false;
1854
1855 if (u->perpetual)
1856 return false;
1857
1858 return !!UNIT_VTABLE(u)->stop;
1859 }
1860
1861 /* Errors:
1862 * -EBADR: This unit type does not support reloading.
1863 * -ENOEXEC: Unit is not started.
1864 * -EAGAIN: An operation is already in progress. Retry later.
1865 */
1866 int unit_reload(Unit *u) {
1867 UnitActiveState state;
1868 Unit *following;
1869
1870 assert(u);
1871
1872 if (u->load_state != UNIT_LOADED)
1873 return -EINVAL;
1874
1875 if (!unit_can_reload(u))
1876 return -EBADR;
1877
1878 state = unit_active_state(u);
1879 if (state == UNIT_RELOADING)
1880 return -EAGAIN;
1881
1882 if (state != UNIT_ACTIVE) {
1883 log_unit_warning(u, "Unit cannot be reloaded because it is inactive.");
1884 return -ENOEXEC;
1885 }
1886
1887 following = unit_following(u);
1888 if (following) {
1889 log_unit_debug(u, "Redirecting reload request from %s to %s.", u->id, following->id);
1890 return unit_reload(following);
1891 }
1892
1893 unit_add_to_dbus_queue(u);
1894
1895 if (!UNIT_VTABLE(u)->reload) {
1896 /* Unit doesn't have a reload function, but we need to propagate the reload anyway */
1897 unit_notify(u, unit_active_state(u), unit_active_state(u), 0);
1898 return 0;
1899 }
1900
1901 return UNIT_VTABLE(u)->reload(u);
1902 }
1903
1904 bool unit_can_reload(Unit *u) {
1905 assert(u);
1906
1907 if (UNIT_VTABLE(u)->can_reload)
1908 return UNIT_VTABLE(u)->can_reload(u);
1909
1910 if (!hashmap_isempty(u->dependencies[UNIT_PROPAGATES_RELOAD_TO]))
1911 return true;
1912
1913 return UNIT_VTABLE(u)->reload;
1914 }
1915
1916 bool unit_is_unneeded(Unit *u) {
1917 static const UnitDependency deps[] = {
1918 UNIT_REQUIRED_BY,
1919 UNIT_REQUISITE_OF,
1920 UNIT_WANTED_BY,
1921 UNIT_BOUND_BY,
1922 };
1923 size_t j;
1924
1925 assert(u);
1926
1927 if (!u->stop_when_unneeded)
1928 return false;
1929
1930 /* Don't clean up while the unit is transitioning or is even inactive. */
1931 if (!UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u)))
1932 return false;
1933 if (u->job)
1934 return false;
1935
1936 for (j = 0; j < ELEMENTSOF(deps); j++) {
1937 Unit *other;
1938 Iterator i;
1939 void *v;
1940
1941 /* If a dependent unit has a job queued, is active or transitioning, or is marked for
1942 * restart, then don't clean this one up. */
1943
1944 HASHMAP_FOREACH_KEY(v, other, u->dependencies[deps[j]], i) {
1945 if (other->job)
1946 return false;
1947
1948 if (!UNIT_IS_INACTIVE_OR_FAILED(unit_active_state(other)))
1949 return false;
1950
1951 if (unit_will_restart(other))
1952 return false;
1953 }
1954 }
1955
1956 return true;
1957 }
1958
1959 static void check_unneeded_dependencies(Unit *u) {
1960
1961 static const UnitDependency deps[] = {
1962 UNIT_REQUIRES,
1963 UNIT_REQUISITE,
1964 UNIT_WANTS,
1965 UNIT_BINDS_TO,
1966 };
1967 size_t j;
1968
1969 assert(u);
1970
1971 /* Add all units this unit depends on to the queue that processes StopWhenUnneeded= behaviour. */
1972
1973 for (j = 0; j < ELEMENTSOF(deps); j++) {
1974 Unit *other;
1975 Iterator i;
1976 void *v;
1977
1978 HASHMAP_FOREACH_KEY(v, other, u->dependencies[deps[j]], i)
1979 unit_submit_to_stop_when_unneeded_queue(other);
1980 }
1981 }
1982
1983 static void unit_check_binds_to(Unit *u) {
1984 _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL;
1985 bool stop = false;
1986 Unit *other;
1987 Iterator i;
1988 void *v;
1989 int r;
1990
1991 assert(u);
1992
1993 if (u->job)
1994 return;
1995
1996 if (unit_active_state(u) != UNIT_ACTIVE)
1997 return;
1998
1999 HASHMAP_FOREACH_KEY(v, other, u->dependencies[UNIT_BINDS_TO], i) {
2000 if (other->job)
2001 continue;
2002
2003 if (!other->coldplugged)
2004 /* We might yet create a job for the other unit… */
2005 continue;
2006
2007 if (!UNIT_IS_INACTIVE_OR_FAILED(unit_active_state(other)))
2008 continue;
2009
2010 stop = true;
2011 break;
2012 }
2013
2014 if (!stop)
2015 return;
2016
2017 /* If stopping a unit fails continuously we might enter a stop
2018 * loop here, hence stop acting on the service being
2019 * unnecessary after a while. */
2020 if (!ratelimit_below(&u->auto_stop_ratelimit)) {
2021 log_unit_warning(u, "Unit is bound to inactive unit %s, but not stopping since we tried this too often recently.", other->id);
2022 return;
2023 }
2024
2025 assert(other);
2026 log_unit_info(u, "Unit is bound to inactive unit %s. Stopping, too.", other->id);
2027
2028 /* A unit we need to run is gone. Sniff. Let's stop this. */
2029 r = manager_add_job(u->manager, JOB_STOP, u, JOB_FAIL, NULL, &error, NULL);
2030 if (r < 0)
2031 log_unit_warning_errno(u, r, "Failed to enqueue stop job, ignoring: %s", bus_error_message(&error, r));
2032 }
2033
2034 static void retroactively_start_dependencies(Unit *u) {
2035 Iterator i;
2036 Unit *other;
2037 void *v;
2038
2039 assert(u);
2040 assert(UNIT_IS_ACTIVE_OR_ACTIVATING(unit_active_state(u)));
2041
2042 HASHMAP_FOREACH_KEY(v, other, u->dependencies[UNIT_REQUIRES], i)
2043 if (!hashmap_get(u->dependencies[UNIT_AFTER], other) &&
2044 !UNIT_IS_ACTIVE_OR_ACTIVATING(unit_active_state(other)))
2045 manager_add_job(u->manager, JOB_START, other, JOB_REPLACE, NULL, NULL, NULL);
2046
2047 HASHMAP_FOREACH_KEY(v, other, u->dependencies[UNIT_BINDS_TO], i)
2048 if (!hashmap_get(u->dependencies[UNIT_AFTER], other) &&
2049 !UNIT_IS_ACTIVE_OR_ACTIVATING(unit_active_state(other)))
2050 manager_add_job(u->manager, JOB_START, other, JOB_REPLACE, NULL, NULL, NULL);
2051
2052 HASHMAP_FOREACH_KEY(v, other, u->dependencies[UNIT_WANTS], i)
2053 if (!hashmap_get(u->dependencies[UNIT_AFTER], other) &&
2054 !UNIT_IS_ACTIVE_OR_ACTIVATING(unit_active_state(other)))
2055 manager_add_job(u->manager, JOB_START, other, JOB_FAIL, NULL, NULL, NULL);
2056
2057 HASHMAP_FOREACH_KEY(v, other, u->dependencies[UNIT_CONFLICTS], i)
2058 if (!UNIT_IS_INACTIVE_OR_DEACTIVATING(unit_active_state(other)))
2059 manager_add_job(u->manager, JOB_STOP, other, JOB_REPLACE, NULL, NULL, NULL);
2060
2061 HASHMAP_FOREACH_KEY(v, other, u->dependencies[UNIT_CONFLICTED_BY], i)
2062 if (!UNIT_IS_INACTIVE_OR_DEACTIVATING(unit_active_state(other)))
2063 manager_add_job(u->manager, JOB_STOP, other, JOB_REPLACE, NULL, NULL, NULL);
2064 }
2065
2066 static void retroactively_stop_dependencies(Unit *u) {
2067 Unit *other;
2068 Iterator i;
2069 void *v;
2070
2071 assert(u);
2072 assert(UNIT_IS_INACTIVE_OR_DEACTIVATING(unit_active_state(u)));
2073
2074 /* Pull down units which are bound to us recursively if enabled */
2075 HASHMAP_FOREACH_KEY(v, other, u->dependencies[UNIT_BOUND_BY], i)
2076 if (!UNIT_IS_INACTIVE_OR_DEACTIVATING(unit_active_state(other)))
2077 manager_add_job(u->manager, JOB_STOP, other, JOB_REPLACE, NULL, NULL, NULL);
2078 }
2079
2080 void unit_start_on_failure(Unit *u) {
2081 Unit *other;
2082 Iterator i;
2083 void *v;
2084 int r;
2085
2086 assert(u);
2087
2088 if (hashmap_size(u->dependencies[UNIT_ON_FAILURE]) <= 0)
2089 return;
2090
2091 log_unit_info(u, "Triggering OnFailure= dependencies.");
2092
2093 HASHMAP_FOREACH_KEY(v, other, u->dependencies[UNIT_ON_FAILURE], i) {
2094 _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL;
2095
2096 r = manager_add_job(u->manager, JOB_START, other, u->on_failure_job_mode, NULL, &error, NULL);
2097 if (r < 0)
2098 log_unit_warning_errno(u, r, "Failed to enqueue OnFailure= job, ignoring: %s", bus_error_message(&error, r));
2099 }
2100 }
2101
2102 void unit_trigger_notify(Unit *u) {
2103 Unit *other;
2104 Iterator i;
2105 void *v;
2106
2107 assert(u);
2108
2109 HASHMAP_FOREACH_KEY(v, other, u->dependencies[UNIT_TRIGGERED_BY], i)
2110 if (UNIT_VTABLE(other)->trigger_notify)
2111 UNIT_VTABLE(other)->trigger_notify(other, u);
2112 }
2113
2114 static int raise_level(int log_level, bool condition_info, bool condition_notice) {
2115 if (condition_notice && log_level > LOG_NOTICE)
2116 return LOG_NOTICE;
2117 if (condition_info && log_level > LOG_INFO)
2118 return LOG_INFO;
2119 return log_level;
2120 }
2121
2122 static int unit_log_resources(Unit *u) {
2123 struct iovec iovec[1 + _CGROUP_IP_ACCOUNTING_METRIC_MAX + _CGROUP_IO_ACCOUNTING_METRIC_MAX + 4];
2124 bool any_traffic = false, have_ip_accounting = false, any_io = false, have_io_accounting = false;
2125 _cleanup_free_ char *igress = NULL, *egress = NULL, *rr = NULL, *wr = NULL;
2126 int log_level = LOG_DEBUG; /* May be raised if resources consumed over a treshold */
2127 size_t n_message_parts = 0, n_iovec = 0;
2128 char* message_parts[1 + 2 + 2 + 1], *t;
2129 nsec_t nsec = NSEC_INFINITY;
2130 CGroupIPAccountingMetric m;
2131 size_t i;
2132 int r;
2133 const char* const ip_fields[_CGROUP_IP_ACCOUNTING_METRIC_MAX] = {
2134 [CGROUP_IP_INGRESS_BYTES] = "IP_METRIC_INGRESS_BYTES",
2135 [CGROUP_IP_INGRESS_PACKETS] = "IP_METRIC_INGRESS_PACKETS",
2136 [CGROUP_IP_EGRESS_BYTES] = "IP_METRIC_EGRESS_BYTES",
2137 [CGROUP_IP_EGRESS_PACKETS] = "IP_METRIC_EGRESS_PACKETS",
2138 };
2139 const char* const io_fields[_CGROUP_IO_ACCOUNTING_METRIC_MAX] = {
2140 [CGROUP_IO_READ_BYTES] = "IO_METRIC_READ_BYTES",
2141 [CGROUP_IO_WRITE_BYTES] = "IO_METRIC_WRITE_BYTES",
2142 [CGROUP_IO_READ_OPERATIONS] = "IO_METRIC_READ_OPERATIONS",
2143 [CGROUP_IO_WRITE_OPERATIONS] = "IO_METRIC_WRITE_OPERATIONS",
2144 };
2145
2146 assert(u);
2147
2148 /* Invoked whenever a unit enters failed or dead state. Logs information about consumed resources if resource
2149 * accounting was enabled for a unit. It does this in two ways: a friendly human readable string with reduced
2150 * information and the complete data in structured fields. */
2151
2152 (void) unit_get_cpu_usage(u, &nsec);
2153 if (nsec != NSEC_INFINITY) {
2154 char buf[FORMAT_TIMESPAN_MAX] = "";
2155
2156 /* Format the CPU time for inclusion in the structured log message */
2157 if (asprintf(&t, "CPU_USAGE_NSEC=%" PRIu64, nsec) < 0) {
2158 r = log_oom();
2159 goto finish;
2160 }
2161 iovec[n_iovec++] = IOVEC_MAKE_STRING(t);
2162
2163 /* Format the CPU time for inclusion in the human language message string */
2164 format_timespan(buf, sizeof(buf), nsec / NSEC_PER_USEC, USEC_PER_MSEC);
2165 t = strjoin("consumed ", buf, " CPU time");
2166 if (!t) {
2167 r = log_oom();
2168 goto finish;
2169 }
2170
2171 message_parts[n_message_parts++] = t;
2172
2173 log_level = raise_level(log_level,
2174 nsec > NOTICEWORTHY_CPU_NSEC,
2175 nsec > MENTIONWORTHY_CPU_NSEC);
2176 }
2177
2178 for (CGroupIOAccountingMetric k = 0; k < _CGROUP_IO_ACCOUNTING_METRIC_MAX; k++) {
2179 char buf[FORMAT_BYTES_MAX] = "";
2180 uint64_t value = UINT64_MAX;
2181
2182 assert(io_fields[k]);
2183
2184 (void) unit_get_io_accounting(u, k, k > 0, &value);
2185 if (value == UINT64_MAX)
2186 continue;
2187
2188 have_io_accounting = true;
2189 if (value > 0)
2190 any_io = true;
2191
2192 /* Format IO accounting data for inclusion in the structured log message */
2193 if (asprintf(&t, "%s=%" PRIu64, io_fields[k], value) < 0) {
2194 r = log_oom();
2195 goto finish;
2196 }
2197 iovec[n_iovec++] = IOVEC_MAKE_STRING(t);
2198
2199 /* Format the IO accounting data for inclusion in the human language message string, but only
2200 * for the bytes counters (and not for the operations counters) */
2201 if (k == CGROUP_IO_READ_BYTES) {
2202 assert(!rr);
2203 rr = strjoin("read ", format_bytes(buf, sizeof(buf), value), " from disk");
2204 if (!rr) {
2205 r = log_oom();
2206 goto finish;
2207 }
2208 } else if (k == CGROUP_IO_WRITE_BYTES) {
2209 assert(!wr);
2210 wr = strjoin("written ", format_bytes(buf, sizeof(buf), value), " to disk");
2211 if (!wr) {
2212 r = log_oom();
2213 goto finish;
2214 }
2215 }
2216
2217 if (IN_SET(k, CGROUP_IO_READ_BYTES, CGROUP_IO_WRITE_BYTES))
2218 log_level = raise_level(log_level,
2219 value > MENTIONWORTHY_IO_BYTES,
2220 value > NOTICEWORTHY_IO_BYTES);
2221 }
2222
2223 if (have_io_accounting) {
2224 if (any_io) {
2225 if (rr)
2226 message_parts[n_message_parts++] = TAKE_PTR(rr);
2227 if (wr)
2228 message_parts[n_message_parts++] = TAKE_PTR(wr);
2229
2230 } else {
2231 char *k;
2232
2233 k = strdup("no IO");
2234 if (!k) {
2235 r = log_oom();
2236 goto finish;
2237 }
2238
2239 message_parts[n_message_parts++] = k;
2240 }
2241 }
2242
2243 for (m = 0; m < _CGROUP_IP_ACCOUNTING_METRIC_MAX; m++) {
2244 char buf[FORMAT_BYTES_MAX] = "";
2245 uint64_t value = UINT64_MAX;
2246
2247 assert(ip_fields[m]);
2248
2249 (void) unit_get_ip_accounting(u, m, &value);
2250 if (value == UINT64_MAX)
2251 continue;
2252
2253 have_ip_accounting = true;
2254 if (value > 0)
2255 any_traffic = true;
2256
2257 /* Format IP accounting data for inclusion in the structured log message */
2258 if (asprintf(&t, "%s=%" PRIu64, ip_fields[m], value) < 0) {
2259 r = log_oom();
2260 goto finish;
2261 }
2262 iovec[n_iovec++] = IOVEC_MAKE_STRING(t);
2263
2264 /* Format the IP accounting data for inclusion in the human language message string, but only for the
2265 * bytes counters (and not for the packets counters) */
2266 if (m == CGROUP_IP_INGRESS_BYTES) {
2267 assert(!igress);
2268 igress = strjoin("received ", format_bytes(buf, sizeof(buf), value), " IP traffic");
2269 if (!igress) {
2270 r = log_oom();
2271 goto finish;
2272 }
2273 } else if (m == CGROUP_IP_EGRESS_BYTES) {
2274 assert(!egress);
2275 egress = strjoin("sent ", format_bytes(buf, sizeof(buf), value), " IP traffic");
2276 if (!egress) {
2277 r = log_oom();
2278 goto finish;
2279 }
2280 }
2281
2282 if (IN_SET(m, CGROUP_IP_INGRESS_BYTES, CGROUP_IP_EGRESS_BYTES))
2283 log_level = raise_level(log_level,
2284 value > MENTIONWORTHY_IP_BYTES,
2285 value > NOTICEWORTHY_IP_BYTES);
2286 }
2287
2288 if (have_ip_accounting) {
2289 if (any_traffic) {
2290 if (igress)
2291 message_parts[n_message_parts++] = TAKE_PTR(igress);
2292 if (egress)
2293 message_parts[n_message_parts++] = TAKE_PTR(egress);
2294
2295 } else {
2296 char *k;
2297
2298 k = strdup("no IP traffic");
2299 if (!k) {
2300 r = log_oom();
2301 goto finish;
2302 }
2303
2304 message_parts[n_message_parts++] = k;
2305 }
2306 }
2307
2308 /* Is there any accounting data available at all? */
2309 if (n_iovec == 0) {
2310 r = 0;
2311 goto finish;
2312 }
2313
2314 if (n_message_parts == 0)
2315 t = strjoina("MESSAGE=", u->id, ": Completed.");
2316 else {
2317 _cleanup_free_ char *joined;
2318
2319 message_parts[n_message_parts] = NULL;
2320
2321 joined = strv_join(message_parts, ", ");
2322 if (!joined) {
2323 r = log_oom();
2324 goto finish;
2325 }
2326
2327 joined[0] = ascii_toupper(joined[0]);
2328 t = strjoina("MESSAGE=", u->id, ": ", joined, ".");
2329 }
2330
2331 /* The following four fields we allocate on the stack or are static strings, we hence don't want to free them,
2332 * and hence don't increase n_iovec for them */
2333 iovec[n_iovec] = IOVEC_MAKE_STRING(t);
2334 iovec[n_iovec + 1] = IOVEC_MAKE_STRING("MESSAGE_ID=" SD_MESSAGE_UNIT_RESOURCES_STR);
2335
2336 t = strjoina(u->manager->unit_log_field, u->id);
2337 iovec[n_iovec + 2] = IOVEC_MAKE_STRING(t);
2338
2339 t = strjoina(u->manager->invocation_log_field, u->invocation_id_string);
2340 iovec[n_iovec + 3] = IOVEC_MAKE_STRING(t);
2341
2342 log_struct_iovec(log_level, iovec, n_iovec + 4);
2343 r = 0;
2344
2345 finish:
2346 for (i = 0; i < n_message_parts; i++)
2347 free(message_parts[i]);
2348
2349 for (i = 0; i < n_iovec; i++)
2350 free(iovec[i].iov_base);
2351
2352 return r;
2353
2354 }
2355
2356 static void unit_update_on_console(Unit *u) {
2357 bool b;
2358
2359 assert(u);
2360
2361 b = unit_needs_console(u);
2362 if (u->on_console == b)
2363 return;
2364
2365 u->on_console = b;
2366 if (b)
2367 manager_ref_console(u->manager);
2368 else
2369 manager_unref_console(u->manager);
2370 }
2371
2372 static void unit_emit_audit_start(Unit *u) {
2373 assert(u);
2374
2375 if (u->type != UNIT_SERVICE)
2376 return;
2377
2378 /* Write audit record if we have just finished starting up */
2379 manager_send_unit_audit(u->manager, u, AUDIT_SERVICE_START, true);
2380 u->in_audit = true;
2381 }
2382
2383 static void unit_emit_audit_stop(Unit *u, UnitActiveState state) {
2384 assert(u);
2385
2386 if (u->type != UNIT_SERVICE)
2387 return;
2388
2389 if (u->in_audit) {
2390 /* Write audit record if we have just finished shutting down */
2391 manager_send_unit_audit(u->manager, u, AUDIT_SERVICE_STOP, state == UNIT_INACTIVE);
2392 u->in_audit = false;
2393 } else {
2394 /* Hmm, if there was no start record written write it now, so that we always have a nice pair */
2395 manager_send_unit_audit(u->manager, u, AUDIT_SERVICE_START, state == UNIT_INACTIVE);
2396
2397 if (state == UNIT_INACTIVE)
2398 manager_send_unit_audit(u->manager, u, AUDIT_SERVICE_STOP, true);
2399 }
2400 }
2401
2402 static bool unit_process_job(Job *j, UnitActiveState ns, UnitNotifyFlags flags) {
2403 bool unexpected = false;
2404 JobResult result;
2405
2406 assert(j);
2407
2408 if (j->state == JOB_WAITING)
2409
2410 /* So we reached a different state for this job. Let's see if we can run it now if it failed previously
2411 * due to EAGAIN. */
2412 job_add_to_run_queue(j);
2413
2414 /* Let's check whether the unit's new state constitutes a finished job, or maybe contradicts a running job and
2415 * hence needs to invalidate jobs. */
2416
2417 switch (j->type) {
2418
2419 case JOB_START:
2420 case JOB_VERIFY_ACTIVE:
2421
2422 if (UNIT_IS_ACTIVE_OR_RELOADING(ns))
2423 job_finish_and_invalidate(j, JOB_DONE, true, false);
2424 else if (j->state == JOB_RUNNING && ns != UNIT_ACTIVATING) {
2425 unexpected = true;
2426
2427 if (UNIT_IS_INACTIVE_OR_FAILED(ns)) {
2428 if (ns == UNIT_FAILED)
2429 result = JOB_FAILED;
2430 else if (FLAGS_SET(flags, UNIT_NOTIFY_SKIP_CONDITION))
2431 result = JOB_SKIPPED;
2432 else
2433 result = JOB_DONE;
2434
2435 job_finish_and_invalidate(j, result, true, false);
2436 }
2437 }
2438
2439 break;
2440
2441 case JOB_RELOAD:
2442 case JOB_RELOAD_OR_START:
2443 case JOB_TRY_RELOAD:
2444
2445 if (j->state == JOB_RUNNING) {
2446 if (ns == UNIT_ACTIVE)
2447 job_finish_and_invalidate(j, (flags & UNIT_NOTIFY_RELOAD_FAILURE) ? JOB_FAILED : JOB_DONE, true, false);
2448 else if (!IN_SET(ns, UNIT_ACTIVATING, UNIT_RELOADING)) {
2449 unexpected = true;
2450
2451 if (UNIT_IS_INACTIVE_OR_FAILED(ns))
2452 job_finish_and_invalidate(j, ns == UNIT_FAILED ? JOB_FAILED : JOB_DONE, true, false);
2453 }
2454 }
2455
2456 break;
2457
2458 case JOB_STOP:
2459 case JOB_RESTART:
2460 case JOB_TRY_RESTART:
2461
2462 if (UNIT_IS_INACTIVE_OR_FAILED(ns))
2463 job_finish_and_invalidate(j, JOB_DONE, true, false);
2464 else if (j->state == JOB_RUNNING && ns != UNIT_DEACTIVATING) {
2465 unexpected = true;
2466 job_finish_and_invalidate(j, JOB_FAILED, true, false);
2467 }
2468
2469 break;
2470
2471 default:
2472 assert_not_reached("Job type unknown");
2473 }
2474
2475 return unexpected;
2476 }
2477
2478 void unit_notify(Unit *u, UnitActiveState os, UnitActiveState ns, UnitNotifyFlags flags) {
2479 const char *reason;
2480 Manager *m;
2481
2482 assert(u);
2483 assert(os < _UNIT_ACTIVE_STATE_MAX);
2484 assert(ns < _UNIT_ACTIVE_STATE_MAX);
2485
2486 /* Note that this is called for all low-level state changes, even if they might map to the same high-level
2487 * UnitActiveState! That means that ns == os is an expected behavior here. For example: if a mount point is
2488 * remounted this function will be called too! */
2489
2490 m = u->manager;
2491
2492 /* Let's enqueue the change signal early. In case this unit has a job associated we want that this unit is in
2493 * the bus queue, so that any job change signal queued will force out the unit change signal first. */
2494 unit_add_to_dbus_queue(u);
2495
2496 /* Update timestamps for state changes */
2497 if (!MANAGER_IS_RELOADING(m)) {
2498 dual_timestamp_get(&u->state_change_timestamp);
2499
2500 if (UNIT_IS_INACTIVE_OR_FAILED(os) && !UNIT_IS_INACTIVE_OR_FAILED(ns))
2501 u->inactive_exit_timestamp = u->state_change_timestamp;
2502 else if (!UNIT_IS_INACTIVE_OR_FAILED(os) && UNIT_IS_INACTIVE_OR_FAILED(ns))
2503 u->inactive_enter_timestamp = u->state_change_timestamp;
2504
2505 if (!UNIT_IS_ACTIVE_OR_RELOADING(os) && UNIT_IS_ACTIVE_OR_RELOADING(ns))
2506 u->active_enter_timestamp = u->state_change_timestamp;
2507 else if (UNIT_IS_ACTIVE_OR_RELOADING(os) && !UNIT_IS_ACTIVE_OR_RELOADING(ns))
2508 u->active_exit_timestamp = u->state_change_timestamp;
2509 }
2510
2511 /* Keep track of failed units */
2512 (void) manager_update_failed_units(m, u, ns == UNIT_FAILED);
2513
2514 /* Make sure the cgroup and state files are always removed when we become inactive */
2515 if (UNIT_IS_INACTIVE_OR_FAILED(ns)) {
2516 unit_prune_cgroup(u);
2517 unit_unlink_state_files(u);
2518 }
2519
2520 unit_update_on_console(u);
2521
2522 if (!MANAGER_IS_RELOADING(m)) {
2523 bool unexpected;
2524
2525 /* Let's propagate state changes to the job */
2526 if (u->job)
2527 unexpected = unit_process_job(u->job, ns, flags);
2528 else
2529 unexpected = true;
2530
2531 /* If this state change happened without being requested by a job, then let's retroactively start or
2532 * stop dependencies. We skip that step when deserializing, since we don't want to create any
2533 * additional jobs just because something is already activated. */
2534
2535 if (unexpected) {
2536 if (UNIT_IS_INACTIVE_OR_FAILED(os) && UNIT_IS_ACTIVE_OR_ACTIVATING(ns))
2537 retroactively_start_dependencies(u);
2538 else if (UNIT_IS_ACTIVE_OR_ACTIVATING(os) && UNIT_IS_INACTIVE_OR_DEACTIVATING(ns))
2539 retroactively_stop_dependencies(u);
2540 }
2541
2542 /* stop unneeded units regardless if going down was expected or not */
2543 if (UNIT_IS_INACTIVE_OR_FAILED(ns))
2544 check_unneeded_dependencies(u);
2545
2546 if (ns != os && ns == UNIT_FAILED) {
2547 log_unit_debug(u, "Unit entered failed state.");
2548
2549 if (!(flags & UNIT_NOTIFY_WILL_AUTO_RESTART))
2550 unit_start_on_failure(u);
2551 }
2552
2553 if (UNIT_IS_ACTIVE_OR_RELOADING(ns) && !UNIT_IS_ACTIVE_OR_RELOADING(os)) {
2554 /* This unit just finished starting up */
2555
2556 unit_emit_audit_start(u);
2557 manager_send_unit_plymouth(m, u);
2558 }
2559
2560 if (UNIT_IS_INACTIVE_OR_FAILED(ns) && !UNIT_IS_INACTIVE_OR_FAILED(os)) {
2561 /* This unit just stopped/failed. */
2562
2563 unit_emit_audit_stop(u, ns);
2564 unit_log_resources(u);
2565 }
2566 }
2567
2568 manager_recheck_journal(m);
2569 manager_recheck_dbus(m);
2570
2571 unit_trigger_notify(u);
2572
2573 if (!MANAGER_IS_RELOADING(m)) {
2574 /* Maybe we finished startup and are now ready for being stopped because unneeded? */
2575 unit_submit_to_stop_when_unneeded_queue(u);
2576
2577 /* Maybe we finished startup, but something we needed has vanished? Let's die then. (This happens when
2578 * something BindsTo= to a Type=oneshot unit, as these units go directly from starting to inactive,
2579 * without ever entering started.) */
2580 unit_check_binds_to(u);
2581
2582 if (os != UNIT_FAILED && ns == UNIT_FAILED) {
2583 reason = strjoina("unit ", u->id, " failed");
2584 emergency_action(m, u->failure_action, 0, u->reboot_arg, unit_failure_action_exit_status(u), reason);
2585 } else if (!UNIT_IS_INACTIVE_OR_FAILED(os) && ns == UNIT_INACTIVE) {
2586 reason = strjoina("unit ", u->id, " succeeded");
2587 emergency_action(m, u->success_action, 0, u->reboot_arg, unit_success_action_exit_status(u), reason);
2588 }
2589 }
2590
2591 unit_add_to_gc_queue(u);
2592 }
2593
2594 int unit_watch_pid(Unit *u, pid_t pid, bool exclusive) {
2595 int r;
2596
2597 assert(u);
2598 assert(pid_is_valid(pid));
2599
2600 /* Watch a specific PID */
2601
2602 /* Caller might be sure that this PID belongs to this unit only. Let's take this
2603 * opportunity to remove any stalled references to this PID as they can be created
2604 * easily (when watching a process which is not our direct child). */
2605 if (exclusive)
2606 manager_unwatch_pid(u->manager, pid);
2607
2608 r = set_ensure_allocated(&u->pids, NULL);
2609 if (r < 0)
2610 return r;
2611
2612 r = hashmap_ensure_allocated(&u->manager->watch_pids, NULL);
2613 if (r < 0)
2614 return r;
2615
2616 /* First try, let's add the unit keyed by "pid". */
2617 r = hashmap_put(u->manager->watch_pids, PID_TO_PTR(pid), u);
2618 if (r == -EEXIST) {
2619 Unit **array;
2620 bool found = false;
2621 size_t n = 0;
2622
2623 /* OK, the "pid" key is already assigned to a different unit. Let's see if the "-pid" key (which points
2624 * to an array of Units rather than just a Unit), lists us already. */
2625
2626 array = hashmap_get(u->manager->watch_pids, PID_TO_PTR(-pid));
2627 if (array)
2628 for (; array[n]; n++)
2629 if (array[n] == u)
2630 found = true;
2631
2632 if (found) /* Found it already? if so, do nothing */
2633 r = 0;
2634 else {
2635 Unit **new_array;
2636
2637 /* Allocate a new array */
2638 new_array = new(Unit*, n + 2);
2639 if (!new_array)
2640 return -ENOMEM;
2641
2642 memcpy_safe(new_array, array, sizeof(Unit*) * n);
2643 new_array[n] = u;
2644 new_array[n+1] = NULL;
2645
2646 /* Add or replace the old array */
2647 r = hashmap_replace(u->manager->watch_pids, PID_TO_PTR(-pid), new_array);
2648 if (r < 0) {
2649 free(new_array);
2650 return r;
2651 }
2652
2653 free(array);
2654 }
2655 } else if (r < 0)
2656 return r;
2657
2658 r = set_put(u->pids, PID_TO_PTR(pid));
2659 if (r < 0)
2660 return r;
2661
2662 return 0;
2663 }
2664
2665 void unit_unwatch_pid(Unit *u, pid_t pid) {
2666 Unit **array;
2667
2668 assert(u);
2669 assert(pid_is_valid(pid));
2670
2671 /* First let's drop the unit in case it's keyed as "pid". */
2672 (void) hashmap_remove_value(u->manager->watch_pids, PID_TO_PTR(pid), u);
2673
2674 /* Then, let's also drop the unit, in case it's in the array keyed by -pid */
2675 array = hashmap_get(u->manager->watch_pids, PID_TO_PTR(-pid));
2676 if (array) {
2677 size_t n, m = 0;
2678
2679 /* Let's iterate through the array, dropping our own entry */
2680 for (n = 0; array[n]; n++)
2681 if (array[n] != u)
2682 array[m++] = array[n];
2683 array[m] = NULL;
2684
2685 if (m == 0) {
2686 /* The array is now empty, remove the entire entry */
2687 assert(hashmap_remove(u->manager->watch_pids, PID_TO_PTR(-pid)) == array);
2688 free(array);
2689 }
2690 }
2691
2692 (void) set_remove(u->pids, PID_TO_PTR(pid));
2693 }
2694
2695 void unit_unwatch_all_pids(Unit *u) {
2696 assert(u);
2697
2698 while (!set_isempty(u->pids))
2699 unit_unwatch_pid(u, PTR_TO_PID(set_first(u->pids)));
2700
2701 u->pids = set_free(u->pids);
2702 }
2703
2704 static void unit_tidy_watch_pids(Unit *u) {
2705 pid_t except1, except2;
2706 Iterator i;
2707 void *e;
2708
2709 assert(u);
2710
2711 /* Cleans dead PIDs from our list */
2712
2713 except1 = unit_main_pid(u);
2714 except2 = unit_control_pid(u);
2715
2716 SET_FOREACH(e, u->pids, i) {
2717 pid_t pid = PTR_TO_PID(e);
2718
2719 if (pid == except1 || pid == except2)
2720 continue;
2721
2722 if (!pid_is_unwaited(pid))
2723 unit_unwatch_pid(u, pid);
2724 }
2725 }
2726
2727 static int on_rewatch_pids_event(sd_event_source *s, void *userdata) {
2728 Unit *u = userdata;
2729
2730 assert(s);
2731 assert(u);
2732
2733 unit_tidy_watch_pids(u);
2734 unit_watch_all_pids(u);
2735
2736 /* If the PID set is empty now, then let's finish this off. */
2737 unit_synthesize_cgroup_empty_event(u);
2738
2739 return 0;
2740 }
2741
2742 int unit_enqueue_rewatch_pids(Unit *u) {
2743 int r;
2744
2745 assert(u);
2746
2747 if (!u->cgroup_path)
2748 return -ENOENT;
2749
2750 r = cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER);
2751 if (r < 0)
2752 return r;
2753 if (r > 0) /* On unified we can use proper notifications */
2754 return 0;
2755
2756 /* Enqueues a low-priority job that will clean up dead PIDs from our list of PIDs to watch and subscribe to new
2757 * PIDs that might have appeared. We do this in a delayed job because the work might be quite slow, as it
2758 * involves issuing kill(pid, 0) on all processes we watch. */
2759
2760 if (!u->rewatch_pids_event_source) {
2761 _cleanup_(sd_event_source_unrefp) sd_event_source *s = NULL;
2762
2763 r = sd_event_add_defer(u->manager->event, &s, on_rewatch_pids_event, u);
2764 if (r < 0)
2765 return log_error_errno(r, "Failed to allocate event source for tidying watched PIDs: %m");
2766
2767 r = sd_event_source_set_priority(s, SD_EVENT_PRIORITY_IDLE);
2768 if (r < 0)
2769 return log_error_errno(r, "Failed to adjust priority of event source for tidying watched PIDs: %m");
2770
2771 (void) sd_event_source_set_description(s, "tidy-watch-pids");
2772
2773 u->rewatch_pids_event_source = TAKE_PTR(s);
2774 }
2775
2776 r = sd_event_source_set_enabled(u->rewatch_pids_event_source, SD_EVENT_ONESHOT);
2777 if (r < 0)
2778 return log_error_errno(r, "Failed to enable event source for tidying watched PIDs: %m");
2779
2780 return 0;
2781 }
2782
2783 void unit_dequeue_rewatch_pids(Unit *u) {
2784 int r;
2785 assert(u);
2786
2787 if (!u->rewatch_pids_event_source)
2788 return;
2789
2790 r = sd_event_source_set_enabled(u->rewatch_pids_event_source, SD_EVENT_OFF);
2791 if (r < 0)
2792 log_warning_errno(r, "Failed to disable event source for tidying watched PIDs, ignoring: %m");
2793
2794 u->rewatch_pids_event_source = sd_event_source_unref(u->rewatch_pids_event_source);
2795 }
2796
2797 bool unit_job_is_applicable(Unit *u, JobType j) {
2798 assert(u);
2799 assert(j >= 0 && j < _JOB_TYPE_MAX);
2800
2801 switch (j) {
2802
2803 case JOB_VERIFY_ACTIVE:
2804 case JOB_START:
2805 case JOB_NOP:
2806 /* Note that we don't check unit_can_start() here. That's because .device units and suchlike are not
2807 * startable by us but may appear due to external events, and it thus makes sense to permit enqueing
2808 * jobs for it. */
2809 return true;
2810
2811 case JOB_STOP:
2812 /* Similar as above. However, perpetual units can never be stopped (neither explicitly nor due to
2813 * external events), hence it makes no sense to permit enqueing such a request either. */
2814 return !u->perpetual;
2815
2816 case JOB_RESTART:
2817 case JOB_TRY_RESTART:
2818 return unit_can_stop(u) && unit_can_start(u);
2819
2820 case JOB_RELOAD:
2821 case JOB_TRY_RELOAD:
2822 return unit_can_reload(u);
2823
2824 case JOB_RELOAD_OR_START:
2825 return unit_can_reload(u) && unit_can_start(u);
2826
2827 default:
2828 assert_not_reached("Invalid job type");
2829 }
2830 }
2831
2832 static void maybe_warn_about_dependency(Unit *u, const char *other, UnitDependency dependency) {
2833 assert(u);
2834
2835 /* Only warn about some unit types */
2836 if (!IN_SET(dependency, UNIT_CONFLICTS, UNIT_CONFLICTED_BY, UNIT_BEFORE, UNIT_AFTER, UNIT_ON_FAILURE, UNIT_TRIGGERS, UNIT_TRIGGERED_BY))
2837 return;
2838
2839 if (streq_ptr(u->id, other))
2840 log_unit_warning(u, "Dependency %s=%s dropped", unit_dependency_to_string(dependency), u->id);
2841 else
2842 log_unit_warning(u, "Dependency %s=%s dropped, merged into %s", unit_dependency_to_string(dependency), strna(other), u->id);
2843 }
2844
2845 static int unit_add_dependency_hashmap(
2846 Hashmap **h,
2847 Unit *other,
2848 UnitDependencyMask origin_mask,
2849 UnitDependencyMask destination_mask) {
2850
2851 UnitDependencyInfo info;
2852 int r;
2853
2854 assert(h);
2855 assert(other);
2856 assert(origin_mask < _UNIT_DEPENDENCY_MASK_FULL);
2857 assert(destination_mask < _UNIT_DEPENDENCY_MASK_FULL);
2858 assert(origin_mask > 0 || destination_mask > 0);
2859
2860 r = hashmap_ensure_allocated(h, NULL);
2861 if (r < 0)
2862 return r;
2863
2864 assert_cc(sizeof(void*) == sizeof(info));
2865
2866 info.data = hashmap_get(*h, other);
2867 if (info.data) {
2868 /* Entry already exists. Add in our mask. */
2869
2870 if (FLAGS_SET(origin_mask, info.origin_mask) &&
2871 FLAGS_SET(destination_mask, info.destination_mask))
2872 return 0; /* NOP */
2873
2874 info.origin_mask |= origin_mask;
2875 info.destination_mask |= destination_mask;
2876
2877 r = hashmap_update(*h, other, info.data);
2878 } else {
2879 info = (UnitDependencyInfo) {
2880 .origin_mask = origin_mask,
2881 .destination_mask = destination_mask,
2882 };
2883
2884 r = hashmap_put(*h, other, info.data);
2885 }
2886 if (r < 0)
2887 return r;
2888
2889 return 1;
2890 }
2891
2892 int unit_add_dependency(
2893 Unit *u,
2894 UnitDependency d,
2895 Unit *other,
2896 bool add_reference,
2897 UnitDependencyMask mask) {
2898
2899 static const UnitDependency inverse_table[_UNIT_DEPENDENCY_MAX] = {
2900 [UNIT_REQUIRES] = UNIT_REQUIRED_BY,
2901 [UNIT_WANTS] = UNIT_WANTED_BY,
2902 [UNIT_REQUISITE] = UNIT_REQUISITE_OF,
2903 [UNIT_BINDS_TO] = UNIT_BOUND_BY,
2904 [UNIT_PART_OF] = UNIT_CONSISTS_OF,
2905 [UNIT_REQUIRED_BY] = UNIT_REQUIRES,
2906 [UNIT_REQUISITE_OF] = UNIT_REQUISITE,
2907 [UNIT_WANTED_BY] = UNIT_WANTS,
2908 [UNIT_BOUND_BY] = UNIT_BINDS_TO,
2909 [UNIT_CONSISTS_OF] = UNIT_PART_OF,
2910 [UNIT_CONFLICTS] = UNIT_CONFLICTED_BY,
2911 [UNIT_CONFLICTED_BY] = UNIT_CONFLICTS,
2912 [UNIT_BEFORE] = UNIT_AFTER,
2913 [UNIT_AFTER] = UNIT_BEFORE,
2914 [UNIT_ON_FAILURE] = _UNIT_DEPENDENCY_INVALID,
2915 [UNIT_REFERENCES] = UNIT_REFERENCED_BY,
2916 [UNIT_REFERENCED_BY] = UNIT_REFERENCES,
2917 [UNIT_TRIGGERS] = UNIT_TRIGGERED_BY,
2918 [UNIT_TRIGGERED_BY] = UNIT_TRIGGERS,
2919 [UNIT_PROPAGATES_RELOAD_TO] = UNIT_RELOAD_PROPAGATED_FROM,
2920 [UNIT_RELOAD_PROPAGATED_FROM] = UNIT_PROPAGATES_RELOAD_TO,
2921 [UNIT_JOINS_NAMESPACE_OF] = UNIT_JOINS_NAMESPACE_OF,
2922 };
2923 Unit *original_u = u, *original_other = other;
2924 int r;
2925
2926 assert(u);
2927 assert(d >= 0 && d < _UNIT_DEPENDENCY_MAX);
2928 assert(other);
2929
2930 u = unit_follow_merge(u);
2931 other = unit_follow_merge(other);
2932
2933 /* We won't allow dependencies on ourselves. We will not
2934 * consider them an error however. */
2935 if (u == other) {
2936 maybe_warn_about_dependency(original_u, original_other->id, d);
2937 return 0;
2938 }
2939
2940 if ((d == UNIT_BEFORE && other->type == UNIT_DEVICE) ||
2941 (d == UNIT_AFTER && u->type == UNIT_DEVICE)) {
2942 log_unit_warning(u, "Dependency Before=%s ignored (.device units cannot be delayed)", other->id);
2943 return 0;
2944 }
2945
2946 r = unit_add_dependency_hashmap(u->dependencies + d, other, mask, 0);
2947 if (r < 0)
2948 return r;
2949
2950 if (inverse_table[d] != _UNIT_DEPENDENCY_INVALID && inverse_table[d] != d) {
2951 r = unit_add_dependency_hashmap(other->dependencies + inverse_table[d], u, 0, mask);
2952 if (r < 0)
2953 return r;
2954 }
2955
2956 if (add_reference) {
2957 r = unit_add_dependency_hashmap(u->dependencies + UNIT_REFERENCES, other, mask, 0);
2958 if (r < 0)
2959 return r;
2960
2961 r = unit_add_dependency_hashmap(other->dependencies + UNIT_REFERENCED_BY, u, 0, mask);
2962 if (r < 0)
2963 return r;
2964 }
2965
2966 unit_add_to_dbus_queue(u);
2967 return 0;
2968 }
2969
2970 int unit_add_two_dependencies(Unit *u, UnitDependency d, UnitDependency e, Unit *other, bool add_reference, UnitDependencyMask mask) {
2971 int r;
2972
2973 assert(u);
2974
2975 r = unit_add_dependency(u, d, other, add_reference, mask);
2976 if (r < 0)
2977 return r;
2978
2979 return unit_add_dependency(u, e, other, add_reference, mask);
2980 }
2981
2982 static int resolve_template(Unit *u, const char *name, char **buf, const char **ret) {
2983 int r;
2984
2985 assert(u);
2986 assert(name);
2987 assert(buf);
2988 assert(ret);
2989
2990 if (!unit_name_is_valid(name, UNIT_NAME_TEMPLATE)) {
2991 *buf = NULL;
2992 *ret = name;
2993 return 0;
2994 }
2995
2996 if (u->instance)
2997 r = unit_name_replace_instance(name, u->instance, buf);
2998 else {
2999 _cleanup_free_ char *i = NULL;
3000
3001 r = unit_name_to_prefix(u->id, &i);
3002 if (r < 0)
3003 return r;
3004
3005 r = unit_name_replace_instance(name, i, buf);
3006 }
3007 if (r < 0)
3008 return r;
3009
3010 *ret = *buf;
3011 return 0;
3012 }
3013
3014 int unit_add_dependency_by_name(Unit *u, UnitDependency d, const char *name, bool add_reference, UnitDependencyMask mask) {
3015 _cleanup_free_ char *buf = NULL;
3016 Unit *other;
3017 int r;
3018
3019 assert(u);
3020 assert(name);
3021
3022 r = resolve_template(u, name, &buf, &name);
3023 if (r < 0)
3024 return r;
3025
3026 r = manager_load_unit(u->manager, name, NULL, NULL, &other);
3027 if (r < 0)
3028 return r;
3029
3030 return unit_add_dependency(u, d, other, add_reference, mask);
3031 }
3032
3033 int unit_add_two_dependencies_by_name(Unit *u, UnitDependency d, UnitDependency e, const char *name, bool add_reference, UnitDependencyMask mask) {
3034 _cleanup_free_ char *buf = NULL;
3035 Unit *other;
3036 int r;
3037
3038 assert(u);
3039 assert(name);
3040
3041 r = resolve_template(u, name, &buf, &name);
3042 if (r < 0)
3043 return r;
3044
3045 r = manager_load_unit(u->manager, name, NULL, NULL, &other);
3046 if (r < 0)
3047 return r;
3048
3049 return unit_add_two_dependencies(u, d, e, other, add_reference, mask);
3050 }
3051
3052 int set_unit_path(const char *p) {
3053 /* This is mostly for debug purposes */
3054 if (setenv("SYSTEMD_UNIT_PATH", p, 1) < 0)
3055 return -errno;
3056
3057 return 0;
3058 }
3059
3060 char *unit_dbus_path(Unit *u) {
3061 assert(u);
3062
3063 if (!u->id)
3064 return NULL;
3065
3066 return unit_dbus_path_from_name(u->id);
3067 }
3068
3069 char *unit_dbus_path_invocation_id(Unit *u) {
3070 assert(u);
3071
3072 if (sd_id128_is_null(u->invocation_id))
3073 return NULL;
3074
3075 return unit_dbus_path_from_name(u->invocation_id_string);
3076 }
3077
3078 int unit_set_slice(Unit *u, Unit *slice) {
3079 assert(u);
3080 assert(slice);
3081
3082 /* Sets the unit slice if it has not been set before. Is extra
3083 * careful, to only allow this for units that actually have a
3084 * cgroup context. Also, we don't allow to set this for slices
3085 * (since the parent slice is derived from the name). Make
3086 * sure the unit we set is actually a slice. */
3087
3088 if (!UNIT_HAS_CGROUP_CONTEXT(u))
3089 return -EOPNOTSUPP;
3090
3091 if (u->type == UNIT_SLICE)
3092 return -EINVAL;
3093
3094 if (unit_active_state(u) != UNIT_INACTIVE)
3095 return -EBUSY;
3096
3097 if (slice->type != UNIT_SLICE)
3098 return -EINVAL;
3099
3100 if (unit_has_name(u, SPECIAL_INIT_SCOPE) &&
3101 !unit_has_name(slice, SPECIAL_ROOT_SLICE))
3102 return -EPERM;
3103
3104 if (UNIT_DEREF(u->slice) == slice)
3105 return 0;
3106
3107 /* Disallow slice changes if @u is already bound to cgroups */
3108 if (UNIT_ISSET(u->slice) && u->cgroup_realized)
3109 return -EBUSY;
3110
3111 unit_ref_set(&u->slice, u, slice);
3112 return 1;
3113 }
3114
3115 int unit_set_default_slice(Unit *u) {
3116 const char *slice_name;
3117 Unit *slice;
3118 int r;
3119
3120 assert(u);
3121
3122 if (UNIT_ISSET(u->slice))
3123 return 0;
3124
3125 if (u->instance) {
3126 _cleanup_free_ char *prefix = NULL, *escaped = NULL;
3127
3128 /* Implicitly place all instantiated units in their
3129 * own per-template slice */
3130
3131 r = unit_name_to_prefix(u->id, &prefix);
3132 if (r < 0)
3133 return r;
3134
3135 /* The prefix is already escaped, but it might include
3136 * "-" which has a special meaning for slice units,
3137 * hence escape it here extra. */
3138 escaped = unit_name_escape(prefix);
3139 if (!escaped)
3140 return -ENOMEM;
3141
3142 if (MANAGER_IS_SYSTEM(u->manager))
3143 slice_name = strjoina("system-", escaped, ".slice");
3144 else
3145 slice_name = strjoina(escaped, ".slice");
3146 } else
3147 slice_name =
3148 MANAGER_IS_SYSTEM(u->manager) && !unit_has_name(u, SPECIAL_INIT_SCOPE)
3149 ? SPECIAL_SYSTEM_SLICE
3150 : SPECIAL_ROOT_SLICE;
3151
3152 r = manager_load_unit(u->manager, slice_name, NULL, NULL, &slice);
3153 if (r < 0)
3154 return r;
3155
3156 return unit_set_slice(u, slice);
3157 }
3158
3159 const char *unit_slice_name(Unit *u) {
3160 assert(u);
3161
3162 if (!UNIT_ISSET(u->slice))
3163 return NULL;
3164
3165 return UNIT_DEREF(u->slice)->id;
3166 }
3167
3168 int unit_load_related_unit(Unit *u, const char *type, Unit **_found) {
3169 _cleanup_free_ char *t = NULL;
3170 int r;
3171
3172 assert(u);
3173 assert(type);
3174 assert(_found);
3175
3176 r = unit_name_change_suffix(u->id, type, &t);
3177 if (r < 0)
3178 return r;
3179 if (unit_has_name(u, t))
3180 return -EINVAL;
3181
3182 r = manager_load_unit(u->manager, t, NULL, NULL, _found);
3183 assert(r < 0 || *_found != u);
3184 return r;
3185 }
3186
3187 static int signal_name_owner_changed(sd_bus_message *message, void *userdata, sd_bus_error *error) {
3188 const char *name, *old_owner, *new_owner;
3189 Unit *u = userdata;
3190 int r;
3191
3192 assert(message);
3193 assert(u);
3194
3195 r = sd_bus_message_read(message, "sss", &name, &old_owner, &new_owner);
3196 if (r < 0) {
3197 bus_log_parse_error(r);
3198 return 0;
3199 }
3200
3201 old_owner = empty_to_null(old_owner);
3202 new_owner = empty_to_null(new_owner);
3203
3204 if (UNIT_VTABLE(u)->bus_name_owner_change)
3205 UNIT_VTABLE(u)->bus_name_owner_change(u, old_owner, new_owner);
3206
3207 return 0;
3208 }
3209
3210 static int get_name_owner_handler(sd_bus_message *message, void *userdata, sd_bus_error *error) {
3211 const sd_bus_error *e;
3212 const char *new_owner;
3213 Unit *u = userdata;
3214 int r;
3215
3216 assert(message);
3217 assert(u);
3218
3219 u->get_name_owner_slot = sd_bus_slot_unref(u->get_name_owner_slot);
3220
3221 if (sd_bus_error_is_set(error)) {
3222 log_error("Failed to get name owner from bus: %s", error->message);
3223 return 0;
3224 }
3225
3226 e = sd_bus_message_get_error(message);
3227 if (sd_bus_error_has_name(e, "org.freedesktop.DBus.Error.NameHasNoOwner"))
3228 return 0;
3229
3230 if (e) {
3231 log_error("Unexpected error response from GetNameOwner: %s", e->message);
3232 return 0;
3233 }
3234
3235 r = sd_bus_message_read(message, "s", &new_owner);
3236 if (r < 0) {
3237 bus_log_parse_error(r);
3238 return 0;
3239 }
3240
3241 new_owner = empty_to_null(new_owner);
3242
3243 if (UNIT_VTABLE(u)->bus_name_owner_change)
3244 UNIT_VTABLE(u)->bus_name_owner_change(u, NULL, new_owner);
3245
3246 return 0;
3247 }
3248
3249 int unit_install_bus_match(Unit *u, sd_bus *bus, const char *name) {
3250 const char *match;
3251
3252 assert(u);
3253 assert(bus);
3254 assert(name);
3255
3256 if (u->match_bus_slot)
3257 return -EBUSY;
3258
3259 match = strjoina("type='signal',"
3260 "sender='org.freedesktop.DBus',"
3261 "path='/org/freedesktop/DBus',"
3262 "interface='org.freedesktop.DBus',"
3263 "member='NameOwnerChanged',"
3264 "arg0='", name, "'");
3265
3266 int r = sd_bus_add_match_async(bus, &u->match_bus_slot, match, signal_name_owner_changed, NULL, u);
3267 if (r < 0)
3268 return r;
3269
3270 return sd_bus_call_method_async(bus,
3271 &u->get_name_owner_slot,
3272 "org.freedesktop.DBus",
3273 "/org/freedesktop/DBus",
3274 "org.freedesktop.DBus",
3275 "GetNameOwner",
3276 get_name_owner_handler,
3277 u,
3278 "s", name);
3279 }
3280
3281 int unit_watch_bus_name(Unit *u, const char *name) {
3282 int r;
3283
3284 assert(u);
3285 assert(name);
3286
3287 /* Watch a specific name on the bus. We only support one unit
3288 * watching each name for now. */
3289
3290 if (u->manager->api_bus) {
3291 /* If the bus is already available, install the match directly.
3292 * Otherwise, just put the name in the list. bus_setup_api() will take care later. */
3293 r = unit_install_bus_match(u, u->manager->api_bus, name);
3294 if (r < 0)
3295 return log_warning_errno(r, "Failed to subscribe to NameOwnerChanged signal for '%s': %m", name);
3296 }
3297
3298 r = hashmap_put(u->manager->watch_bus, name, u);
3299 if (r < 0) {
3300 u->match_bus_slot = sd_bus_slot_unref(u->match_bus_slot);
3301 return log_warning_errno(r, "Failed to put bus name to hashmap: %m");
3302 }
3303
3304 return 0;
3305 }
3306
3307 void unit_unwatch_bus_name(Unit *u, const char *name) {
3308 assert(u);
3309 assert(name);
3310
3311 (void) hashmap_remove_value(u->manager->watch_bus, name, u);
3312 u->match_bus_slot = sd_bus_slot_unref(u->match_bus_slot);
3313 u->get_name_owner_slot = sd_bus_slot_unref(u->get_name_owner_slot);
3314 }
3315
3316 bool unit_can_serialize(Unit *u) {
3317 assert(u);
3318
3319 return UNIT_VTABLE(u)->serialize && UNIT_VTABLE(u)->deserialize_item;
3320 }
3321
3322 static int serialize_cgroup_mask(FILE *f, const char *key, CGroupMask mask) {
3323 _cleanup_free_ char *s = NULL;
3324 int r;
3325
3326 assert(f);
3327 assert(key);
3328
3329 if (mask == 0)
3330 return 0;
3331
3332 r = cg_mask_to_string(mask, &s);
3333 if (r < 0)
3334 return log_error_errno(r, "Failed to format cgroup mask: %m");
3335
3336 return serialize_item(f, key, s);
3337 }
3338
3339 static const char *const ip_accounting_metric_field[_CGROUP_IP_ACCOUNTING_METRIC_MAX] = {
3340 [CGROUP_IP_INGRESS_BYTES] = "ip-accounting-ingress-bytes",
3341 [CGROUP_IP_INGRESS_PACKETS] = "ip-accounting-ingress-packets",
3342 [CGROUP_IP_EGRESS_BYTES] = "ip-accounting-egress-bytes",
3343 [CGROUP_IP_EGRESS_PACKETS] = "ip-accounting-egress-packets",
3344 };
3345
3346 static const char *const io_accounting_metric_field_base[_CGROUP_IO_ACCOUNTING_METRIC_MAX] = {
3347 [CGROUP_IO_READ_BYTES] = "io-accounting-read-bytes-base",
3348 [CGROUP_IO_WRITE_BYTES] = "io-accounting-write-bytes-base",
3349 [CGROUP_IO_READ_OPERATIONS] = "io-accounting-read-operations-base",
3350 [CGROUP_IO_WRITE_OPERATIONS] = "io-accounting-write-operations-base",
3351 };
3352
3353 static const char *const io_accounting_metric_field_last[_CGROUP_IO_ACCOUNTING_METRIC_MAX] = {
3354 [CGROUP_IO_READ_BYTES] = "io-accounting-read-bytes-last",
3355 [CGROUP_IO_WRITE_BYTES] = "io-accounting-write-bytes-last",
3356 [CGROUP_IO_READ_OPERATIONS] = "io-accounting-read-operations-last",
3357 [CGROUP_IO_WRITE_OPERATIONS] = "io-accounting-write-operations-last",
3358 };
3359
3360 int unit_serialize(Unit *u, FILE *f, FDSet *fds, bool serialize_jobs) {
3361 CGroupIPAccountingMetric m;
3362 int r;
3363
3364 assert(u);
3365 assert(f);
3366 assert(fds);
3367
3368 if (unit_can_serialize(u)) {
3369 r = UNIT_VTABLE(u)->serialize(u, f, fds);
3370 if (r < 0)
3371 return r;
3372 }
3373
3374 (void) serialize_dual_timestamp(f, "state-change-timestamp", &u->state_change_timestamp);
3375
3376 (void) serialize_dual_timestamp(f, "inactive-exit-timestamp", &u->inactive_exit_timestamp);
3377 (void) serialize_dual_timestamp(f, "active-enter-timestamp", &u->active_enter_timestamp);
3378 (void) serialize_dual_timestamp(f, "active-exit-timestamp", &u->active_exit_timestamp);
3379 (void) serialize_dual_timestamp(f, "inactive-enter-timestamp", &u->inactive_enter_timestamp);
3380
3381 (void) serialize_dual_timestamp(f, "condition-timestamp", &u->condition_timestamp);
3382 (void) serialize_dual_timestamp(f, "assert-timestamp", &u->assert_timestamp);
3383
3384 if (dual_timestamp_is_set(&u->condition_timestamp))
3385 (void) serialize_bool(f, "condition-result", u->condition_result);
3386
3387 if (dual_timestamp_is_set(&u->assert_timestamp))
3388 (void) serialize_bool(f, "assert-result", u->assert_result);
3389
3390 (void) serialize_bool(f, "transient", u->transient);
3391 (void) serialize_bool(f, "in-audit", u->in_audit);
3392
3393 (void) serialize_bool(f, "exported-invocation-id", u->exported_invocation_id);
3394 (void) serialize_bool(f, "exported-log-level-max", u->exported_log_level_max);
3395 (void) serialize_bool(f, "exported-log-extra-fields", u->exported_log_extra_fields);
3396 (void) serialize_bool(f, "exported-log-rate-limit-interval", u->exported_log_ratelimit_interval);
3397 (void) serialize_bool(f, "exported-log-rate-limit-burst", u->exported_log_ratelimit_burst);
3398
3399 (void) serialize_item_format(f, "cpu-usage-base", "%" PRIu64, u->cpu_usage_base);
3400 if (u->cpu_usage_last != NSEC_INFINITY)
3401 (void) serialize_item_format(f, "cpu-usage-last", "%" PRIu64, u->cpu_usage_last);
3402
3403 if (u->oom_kill_last > 0)
3404 (void) serialize_item_format(f, "oom-kill-last", "%" PRIu64, u->oom_kill_last);
3405
3406 for (CGroupIOAccountingMetric im = 0; im < _CGROUP_IO_ACCOUNTING_METRIC_MAX; im++) {
3407 (void) serialize_item_format(f, io_accounting_metric_field_base[im], "%" PRIu64, u->io_accounting_base[im]);
3408
3409 if (u->io_accounting_last[im] != UINT64_MAX)
3410 (void) serialize_item_format(f, io_accounting_metric_field_last[im], "%" PRIu64, u->io_accounting_last[im]);
3411 }
3412
3413 if (u->cgroup_path)
3414 (void) serialize_item(f, "cgroup", u->cgroup_path);
3415
3416 (void) serialize_bool(f, "cgroup-realized", u->cgroup_realized);
3417 (void) serialize_cgroup_mask(f, "cgroup-realized-mask", u->cgroup_realized_mask);
3418 (void) serialize_cgroup_mask(f, "cgroup-enabled-mask", u->cgroup_enabled_mask);
3419 (void) serialize_cgroup_mask(f, "cgroup-invalidated-mask", u->cgroup_invalidated_mask);
3420
3421 if (uid_is_valid(u->ref_uid))
3422 (void) serialize_item_format(f, "ref-uid", UID_FMT, u->ref_uid);
3423 if (gid_is_valid(u->ref_gid))
3424 (void) serialize_item_format(f, "ref-gid", GID_FMT, u->ref_gid);
3425
3426 if (!sd_id128_is_null(u->invocation_id))
3427 (void) serialize_item_format(f, "invocation-id", SD_ID128_FORMAT_STR, SD_ID128_FORMAT_VAL(u->invocation_id));
3428
3429 bus_track_serialize(u->bus_track, f, "ref");
3430
3431 for (m = 0; m < _CGROUP_IP_ACCOUNTING_METRIC_MAX; m++) {
3432 uint64_t v;
3433
3434 r = unit_get_ip_accounting(u, m, &v);
3435 if (r >= 0)
3436 (void) serialize_item_format(f, ip_accounting_metric_field[m], "%" PRIu64, v);
3437 }
3438
3439 if (serialize_jobs) {
3440 if (u->job) {
3441 fputs("job\n", f);
3442 job_serialize(u->job, f);
3443 }
3444
3445 if (u->nop_job) {
3446 fputs("job\n", f);
3447 job_serialize(u->nop_job, f);
3448 }
3449 }
3450
3451 /* End marker */
3452 fputc('\n', f);
3453 return 0;
3454 }
3455
3456 static int unit_deserialize_job(Unit *u, FILE *f) {
3457 _cleanup_(job_freep) Job *j = NULL;
3458 int r;
3459
3460 assert(u);
3461 assert(f);
3462
3463 j = job_new_raw(u);
3464 if (!j)
3465 return log_oom();
3466
3467 r = job_deserialize(j, f);
3468 if (r < 0)
3469 return r;
3470
3471 r = job_install_deserialized(j);
3472 if (r < 0)
3473 return r;
3474
3475 TAKE_PTR(j);
3476 return 0;
3477 }
3478
3479 int unit_deserialize(Unit *u, FILE *f, FDSet *fds) {
3480 int r;
3481
3482 assert(u);
3483 assert(f);
3484 assert(fds);
3485
3486 for (;;) {
3487 _cleanup_free_ char *line = NULL;
3488 char *l, *v;
3489 ssize_t m;
3490 size_t k;
3491
3492 r = read_line(f, LONG_LINE_MAX, &line);
3493 if (r < 0)
3494 return log_error_errno(r, "Failed to read serialization line: %m");
3495 if (r == 0) /* eof */
3496 break;
3497
3498 l = strstrip(line);
3499 if (isempty(l)) /* End marker */
3500 break;
3501
3502 k = strcspn(l, "=");
3503
3504 if (l[k] == '=') {
3505 l[k] = 0;
3506 v = l+k+1;
3507 } else
3508 v = l+k;
3509
3510 if (streq(l, "job")) {
3511 if (v[0] == '\0') {
3512 /* New-style serialized job */
3513 r = unit_deserialize_job(u, f);
3514 if (r < 0)
3515 return r;
3516 } else /* Legacy for pre-44 */
3517 log_unit_warning(u, "Update from too old systemd versions are unsupported, cannot deserialize job: %s", v);
3518 continue;
3519 } else if (streq(l, "state-change-timestamp")) {
3520 (void) deserialize_dual_timestamp(v, &u->state_change_timestamp);
3521 continue;
3522 } else if (streq(l, "inactive-exit-timestamp")) {
3523 (void) deserialize_dual_timestamp(v, &u->inactive_exit_timestamp);
3524 continue;
3525 } else if (streq(l, "active-enter-timestamp")) {
3526 (void) deserialize_dual_timestamp(v, &u->active_enter_timestamp);
3527 continue;
3528 } else if (streq(l, "active-exit-timestamp")) {
3529 (void) deserialize_dual_timestamp(v, &u->active_exit_timestamp);
3530 continue;
3531 } else if (streq(l, "inactive-enter-timestamp")) {
3532 (void) deserialize_dual_timestamp(v, &u->inactive_enter_timestamp);
3533 continue;
3534 } else if (streq(l, "condition-timestamp")) {
3535 (void) deserialize_dual_timestamp(v, &u->condition_timestamp);
3536 continue;
3537 } else if (streq(l, "assert-timestamp")) {
3538 (void) deserialize_dual_timestamp(v, &u->assert_timestamp);
3539 continue;
3540 } else if (streq(l, "condition-result")) {
3541
3542 r = parse_boolean(v);
3543 if (r < 0)
3544 log_unit_debug(u, "Failed to parse condition result value %s, ignoring.", v);
3545 else
3546 u->condition_result = r;
3547
3548 continue;
3549
3550 } else if (streq(l, "assert-result")) {
3551
3552 r = parse_boolean(v);
3553 if (r < 0)
3554 log_unit_debug(u, "Failed to parse assert result value %s, ignoring.", v);
3555 else
3556 u->assert_result = r;
3557
3558 continue;
3559
3560 } else if (streq(l, "transient")) {
3561
3562 r = parse_boolean(v);
3563 if (r < 0)
3564 log_unit_debug(u, "Failed to parse transient bool %s, ignoring.", v);
3565 else
3566 u->transient = r;
3567
3568 continue;
3569
3570 } else if (streq(l, "in-audit")) {
3571
3572 r = parse_boolean(v);
3573 if (r < 0)
3574 log_unit_debug(u, "Failed to parse in-audit bool %s, ignoring.", v);
3575 else
3576 u->in_audit = r;
3577
3578 continue;
3579
3580 } else if (streq(l, "exported-invocation-id")) {
3581
3582 r = parse_boolean(v);
3583 if (r < 0)
3584 log_unit_debug(u, "Failed to parse exported invocation ID bool %s, ignoring.", v);
3585 else
3586 u->exported_invocation_id = r;
3587
3588 continue;
3589
3590 } else if (streq(l, "exported-log-level-max")) {
3591
3592 r = parse_boolean(v);
3593 if (r < 0)
3594 log_unit_debug(u, "Failed to parse exported log level max bool %s, ignoring.", v);
3595 else
3596 u->exported_log_level_max = r;
3597
3598 continue;
3599
3600 } else if (streq(l, "exported-log-extra-fields")) {
3601
3602 r = parse_boolean(v);
3603 if (r < 0)
3604 log_unit_debug(u, "Failed to parse exported log extra fields bool %s, ignoring.", v);
3605 else
3606 u->exported_log_extra_fields = r;
3607
3608 continue;
3609
3610 } else if (streq(l, "exported-log-rate-limit-interval")) {
3611
3612 r = parse_boolean(v);
3613 if (r < 0)
3614 log_unit_debug(u, "Failed to parse exported log rate limit interval %s, ignoring.", v);
3615 else
3616 u->exported_log_ratelimit_interval = r;
3617
3618 continue;
3619
3620 } else if (streq(l, "exported-log-rate-limit-burst")) {
3621
3622 r = parse_boolean(v);
3623 if (r < 0)
3624 log_unit_debug(u, "Failed to parse exported log rate limit burst %s, ignoring.", v);
3625 else
3626 u->exported_log_ratelimit_burst = r;
3627
3628 continue;
3629
3630 } else if (STR_IN_SET(l, "cpu-usage-base", "cpuacct-usage-base")) {
3631
3632 r = safe_atou64(v, &u->cpu_usage_base);
3633 if (r < 0)
3634 log_unit_debug(u, "Failed to parse CPU usage base %s, ignoring.", v);
3635
3636 continue;
3637
3638 } else if (streq(l, "cpu-usage-last")) {
3639
3640 r = safe_atou64(v, &u->cpu_usage_last);
3641 if (r < 0)
3642 log_unit_debug(u, "Failed to read CPU usage last %s, ignoring.", v);
3643
3644 continue;
3645
3646 } else if (streq(l, "oom-kill-last")) {
3647
3648 r = safe_atou64(v, &u->oom_kill_last);
3649 if (r < 0)
3650 log_unit_debug(u, "Failed to read OOM kill last %s, ignoring.", v);
3651
3652 continue;
3653
3654 } else if (streq(l, "cgroup")) {
3655
3656 r = unit_set_cgroup_path(u, v);
3657 if (r < 0)
3658 log_unit_debug_errno(u, r, "Failed to set cgroup path %s, ignoring: %m", v);
3659
3660 (void) unit_watch_cgroup(u);
3661 (void) unit_watch_cgroup_memory(u);
3662
3663 continue;
3664 } else if (streq(l, "cgroup-realized")) {
3665 int b;
3666
3667 b = parse_boolean(v);
3668 if (b < 0)
3669 log_unit_debug(u, "Failed to parse cgroup-realized bool %s, ignoring.", v);
3670 else
3671 u->cgroup_realized = b;
3672
3673 continue;
3674
3675 } else if (streq(l, "cgroup-realized-mask")) {
3676
3677 r = cg_mask_from_string(v, &u->cgroup_realized_mask);
3678 if (r < 0)
3679 log_unit_debug(u, "Failed to parse cgroup-realized-mask %s, ignoring.", v);
3680 continue;
3681
3682 } else if (streq(l, "cgroup-enabled-mask")) {
3683
3684 r = cg_mask_from_string(v, &u->cgroup_enabled_mask);
3685 if (r < 0)
3686 log_unit_debug(u, "Failed to parse cgroup-enabled-mask %s, ignoring.", v);
3687 continue;
3688
3689 } else if (streq(l, "cgroup-invalidated-mask")) {
3690
3691 r = cg_mask_from_string(v, &u->cgroup_invalidated_mask);
3692 if (r < 0)
3693 log_unit_debug(u, "Failed to parse cgroup-invalidated-mask %s, ignoring.", v);
3694 continue;
3695
3696 } else if (streq(l, "ref-uid")) {
3697 uid_t uid;
3698
3699 r = parse_uid(v, &uid);
3700 if (r < 0)
3701 log_unit_debug(u, "Failed to parse referenced UID %s, ignoring.", v);
3702 else
3703 unit_ref_uid_gid(u, uid, GID_INVALID);
3704
3705 continue;
3706
3707 } else if (streq(l, "ref-gid")) {
3708 gid_t gid;
3709
3710 r = parse_gid(v, &gid);
3711 if (r < 0)
3712 log_unit_debug(u, "Failed to parse referenced GID %s, ignoring.", v);
3713 else
3714 unit_ref_uid_gid(u, UID_INVALID, gid);
3715
3716 continue;
3717
3718 } else if (streq(l, "ref")) {
3719
3720 r = strv_extend(&u->deserialized_refs, v);
3721 if (r < 0)
3722 return log_oom();
3723
3724 continue;
3725 } else if (streq(l, "invocation-id")) {
3726 sd_id128_t id;
3727
3728 r = sd_id128_from_string(v, &id);
3729 if (r < 0)
3730 log_unit_debug(u, "Failed to parse invocation id %s, ignoring.", v);
3731 else {
3732 r = unit_set_invocation_id(u, id);
3733 if (r < 0)
3734 log_unit_warning_errno(u, r, "Failed to set invocation ID for unit: %m");
3735 }
3736
3737 continue;
3738 }
3739
3740 /* Check if this is an IP accounting metric serialization field */
3741 m = string_table_lookup(ip_accounting_metric_field, ELEMENTSOF(ip_accounting_metric_field), l);
3742 if (m >= 0) {
3743 uint64_t c;
3744
3745 r = safe_atou64(v, &c);
3746 if (r < 0)
3747 log_unit_debug(u, "Failed to parse IP accounting value %s, ignoring.", v);
3748 else
3749 u->ip_accounting_extra[m] = c;
3750 continue;
3751 }
3752
3753 m = string_table_lookup(io_accounting_metric_field_base, ELEMENTSOF(io_accounting_metric_field_base), l);
3754 if (m >= 0) {
3755 uint64_t c;
3756
3757 r = safe_atou64(v, &c);
3758 if (r < 0)
3759 log_unit_debug(u, "Failed to parse IO accounting base value %s, ignoring.", v);
3760 else
3761 u->io_accounting_base[m] = c;
3762 continue;
3763 }
3764
3765 m = string_table_lookup(io_accounting_metric_field_last, ELEMENTSOF(io_accounting_metric_field_last), l);
3766 if (m >= 0) {
3767 uint64_t c;
3768
3769 r = safe_atou64(v, &c);
3770 if (r < 0)
3771 log_unit_debug(u, "Failed to parse IO accounting last value %s, ignoring.", v);
3772 else
3773 u->io_accounting_last[m] = c;
3774 continue;
3775 }
3776
3777 if (unit_can_serialize(u)) {
3778 r = exec_runtime_deserialize_compat(u, l, v, fds);
3779 if (r < 0) {
3780 log_unit_warning(u, "Failed to deserialize runtime parameter '%s', ignoring.", l);
3781 continue;
3782 }
3783
3784 /* Returns positive if key was handled by the call */
3785 if (r > 0)
3786 continue;
3787
3788 r = UNIT_VTABLE(u)->deserialize_item(u, l, v, fds);
3789 if (r < 0)
3790 log_unit_warning(u, "Failed to deserialize unit parameter '%s', ignoring.", l);
3791 }
3792 }
3793
3794 /* Versions before 228 did not carry a state change timestamp. In this case, take the current time. This is
3795 * useful, so that timeouts based on this timestamp don't trigger too early, and is in-line with the logic from
3796 * before 228 where the base for timeouts was not persistent across reboots. */
3797
3798 if (!dual_timestamp_is_set(&u->state_change_timestamp))
3799 dual_timestamp_get(&u->state_change_timestamp);
3800
3801 /* Let's make sure that everything that is deserialized also gets any potential new cgroup settings applied
3802 * after we are done. For that we invalidate anything already realized, so that we can realize it again. */
3803 unit_invalidate_cgroup(u, _CGROUP_MASK_ALL);
3804 unit_invalidate_cgroup_bpf(u);
3805
3806 return 0;
3807 }
3808
3809 int unit_deserialize_skip(FILE *f) {
3810 int r;
3811 assert(f);
3812
3813 /* Skip serialized data for this unit. We don't know what it is. */
3814
3815 for (;;) {
3816 _cleanup_free_ char *line = NULL;
3817 char *l;
3818
3819 r = read_line(f, LONG_LINE_MAX, &line);
3820 if (r < 0)
3821 return log_error_errno(r, "Failed to read serialization line: %m");
3822 if (r == 0)
3823 return 0;
3824
3825 l = strstrip(line);
3826
3827 /* End marker */
3828 if (isempty(l))
3829 return 1;
3830 }
3831 }
3832
3833 int unit_add_node_dependency(Unit *u, const char *what, UnitDependency dep, UnitDependencyMask mask) {
3834 Unit *device;
3835 _cleanup_free_ char *e = NULL;
3836 int r;
3837
3838 assert(u);
3839
3840 /* Adds in links to the device node that this unit is based on */
3841 if (isempty(what))
3842 return 0;
3843
3844 if (!is_device_path(what))
3845 return 0;
3846
3847 /* When device units aren't supported (such as in a
3848 * container), don't create dependencies on them. */
3849 if (!unit_type_supported(UNIT_DEVICE))
3850 return 0;
3851
3852 r = unit_name_from_path(what, ".device", &e);
3853 if (r < 0)
3854 return r;
3855
3856 r = manager_load_unit(u->manager, e, NULL, NULL, &device);
3857 if (r < 0)
3858 return r;
3859
3860 if (dep == UNIT_REQUIRES && device_shall_be_bound_by(device, u))
3861 dep = UNIT_BINDS_TO;
3862
3863 return unit_add_two_dependencies(u, UNIT_AFTER,
3864 MANAGER_IS_SYSTEM(u->manager) ? dep : UNIT_WANTS,
3865 device, true, mask);
3866 }
3867
3868 int unit_coldplug(Unit *u) {
3869 int r = 0, q;
3870 char **i;
3871 Job *uj;
3872
3873 assert(u);
3874
3875 /* Make sure we don't enter a loop, when coldplugging recursively. */
3876 if (u->coldplugged)
3877 return 0;
3878
3879 u->coldplugged = true;
3880
3881 STRV_FOREACH(i, u->deserialized_refs) {
3882 q = bus_unit_track_add_name(u, *i);
3883 if (q < 0 && r >= 0)
3884 r = q;
3885 }
3886 u->deserialized_refs = strv_free(u->deserialized_refs);
3887
3888 if (UNIT_VTABLE(u)->coldplug) {
3889 q = UNIT_VTABLE(u)->coldplug(u);
3890 if (q < 0 && r >= 0)
3891 r = q;
3892 }
3893
3894 uj = u->job ?: u->nop_job;
3895 if (uj) {
3896 q = job_coldplug(uj);
3897 if (q < 0 && r >= 0)
3898 r = q;
3899 }
3900
3901 return r;
3902 }
3903
3904 void unit_catchup(Unit *u) {
3905 assert(u);
3906
3907 if (UNIT_VTABLE(u)->catchup)
3908 UNIT_VTABLE(u)->catchup(u);
3909 }
3910
3911 static bool fragment_mtime_newer(const char *path, usec_t mtime, bool path_masked) {
3912 struct stat st;
3913
3914 if (!path)
3915 return false;
3916
3917 /* If the source is some virtual kernel file system, then we assume we watch it anyway, and hence pretend we
3918 * are never out-of-date. */
3919 if (PATH_STARTSWITH_SET(path, "/proc", "/sys"))
3920 return false;
3921
3922 if (stat(path, &st) < 0)
3923 /* What, cannot access this anymore? */
3924 return true;
3925
3926 if (path_masked)
3927 /* For masked files check if they are still so */
3928 return !null_or_empty(&st);
3929 else
3930 /* For non-empty files check the mtime */
3931 return timespec_load(&st.st_mtim) > mtime;
3932
3933 return false;
3934 }
3935
3936 bool unit_need_daemon_reload(Unit *u) {
3937 _cleanup_strv_free_ char **t = NULL;
3938 char **path;
3939
3940 assert(u);
3941
3942 /* For unit files, we allow masking… */
3943 if (fragment_mtime_newer(u->fragment_path, u->fragment_mtime,
3944 u->load_state == UNIT_MASKED))
3945 return true;
3946
3947 /* Source paths should not be masked… */
3948 if (fragment_mtime_newer(u->source_path, u->source_mtime, false))
3949 return true;
3950
3951 if (u->load_state == UNIT_LOADED)
3952 (void) unit_find_dropin_paths(u, &t);
3953 if (!strv_equal(u->dropin_paths, t))
3954 return true;
3955
3956 /* … any drop-ins that are masked are simply omitted from the list. */
3957 STRV_FOREACH(path, u->dropin_paths)
3958 if (fragment_mtime_newer(*path, u->dropin_mtime, false))
3959 return true;
3960
3961 return false;
3962 }
3963
3964 void unit_reset_failed(Unit *u) {
3965 assert(u);
3966
3967 if (UNIT_VTABLE(u)->reset_failed)
3968 UNIT_VTABLE(u)->reset_failed(u);
3969
3970 ratelimit_reset(&u->start_ratelimit);
3971 u->start_limit_hit = false;
3972 }
3973
3974 Unit *unit_following(Unit *u) {
3975 assert(u);
3976
3977 if (UNIT_VTABLE(u)->following)
3978 return UNIT_VTABLE(u)->following(u);
3979
3980 return NULL;
3981 }
3982
3983 bool unit_stop_pending(Unit *u) {
3984 assert(u);
3985
3986 /* This call does check the current state of the unit. It's
3987 * hence useful to be called from state change calls of the
3988 * unit itself, where the state isn't updated yet. This is
3989 * different from unit_inactive_or_pending() which checks both
3990 * the current state and for a queued job. */
3991
3992 return unit_has_job_type(u, JOB_STOP);
3993 }
3994
3995 bool unit_inactive_or_pending(Unit *u) {
3996 assert(u);
3997
3998 /* Returns true if the unit is inactive or going down */
3999
4000 if (UNIT_IS_INACTIVE_OR_DEACTIVATING(unit_active_state(u)))
4001 return true;
4002
4003 if (unit_stop_pending(u))
4004 return true;
4005
4006 return false;
4007 }
4008
4009 bool unit_active_or_pending(Unit *u) {
4010 assert(u);
4011
4012 /* Returns true if the unit is active or going up */
4013
4014 if (UNIT_IS_ACTIVE_OR_ACTIVATING(unit_active_state(u)))
4015 return true;
4016
4017 if (u->job &&
4018 IN_SET(u->job->type, JOB_START, JOB_RELOAD_OR_START, JOB_RESTART))
4019 return true;
4020
4021 return false;
4022 }
4023
4024 bool unit_will_restart_default(Unit *u) {
4025 assert(u);
4026
4027 return unit_has_job_type(u, JOB_START);
4028 }
4029
4030 bool unit_will_restart(Unit *u) {
4031 assert(u);
4032
4033 if (!UNIT_VTABLE(u)->will_restart)
4034 return false;
4035
4036 return UNIT_VTABLE(u)->will_restart(u);
4037 }
4038
4039 int unit_kill(Unit *u, KillWho w, int signo, sd_bus_error *error) {
4040 assert(u);
4041 assert(w >= 0 && w < _KILL_WHO_MAX);
4042 assert(SIGNAL_VALID(signo));
4043
4044 if (!UNIT_VTABLE(u)->kill)
4045 return -EOPNOTSUPP;
4046
4047 return UNIT_VTABLE(u)->kill(u, w, signo, error);
4048 }
4049
4050 static Set *unit_pid_set(pid_t main_pid, pid_t control_pid) {
4051 _cleanup_set_free_ Set *pid_set = NULL;
4052 int r;
4053
4054 pid_set = set_new(NULL);
4055 if (!pid_set)
4056 return NULL;
4057
4058 /* Exclude the main/control pids from being killed via the cgroup */
4059 if (main_pid > 0) {
4060 r = set_put(pid_set, PID_TO_PTR(main_pid));
4061 if (r < 0)
4062 return NULL;
4063 }
4064
4065 if (control_pid > 0) {
4066 r = set_put(pid_set, PID_TO_PTR(control_pid));
4067 if (r < 0)
4068 return NULL;
4069 }
4070
4071 return TAKE_PTR(pid_set);
4072 }
4073
4074 int unit_kill_common(
4075 Unit *u,
4076 KillWho who,
4077 int signo,
4078 pid_t main_pid,
4079 pid_t control_pid,
4080 sd_bus_error *error) {
4081
4082 int r = 0;
4083 bool killed = false;
4084
4085 if (IN_SET(who, KILL_MAIN, KILL_MAIN_FAIL)) {
4086 if (main_pid < 0)
4087 return sd_bus_error_setf(error, BUS_ERROR_NO_SUCH_PROCESS, "%s units have no main processes", unit_type_to_string(u->type));
4088 else if (main_pid == 0)
4089 return sd_bus_error_set_const(error, BUS_ERROR_NO_SUCH_PROCESS, "No main process to kill");
4090 }
4091
4092 if (IN_SET(who, KILL_CONTROL, KILL_CONTROL_FAIL)) {
4093 if (control_pid < 0)
4094 return sd_bus_error_setf(error, BUS_ERROR_NO_SUCH_PROCESS, "%s units have no control processes", unit_type_to_string(u->type));
4095 else if (control_pid == 0)
4096 return sd_bus_error_set_const(error, BUS_ERROR_NO_SUCH_PROCESS, "No control process to kill");
4097 }
4098
4099 if (IN_SET(who, KILL_CONTROL, KILL_CONTROL_FAIL, KILL_ALL, KILL_ALL_FAIL))
4100 if (control_pid > 0) {
4101 if (kill(control_pid, signo) < 0)
4102 r = -errno;
4103 else
4104 killed = true;
4105 }
4106
4107 if (IN_SET(who, KILL_MAIN, KILL_MAIN_FAIL, KILL_ALL, KILL_ALL_FAIL))
4108 if (main_pid > 0) {
4109 if (kill(main_pid, signo) < 0)
4110 r = -errno;
4111 else
4112 killed = true;
4113 }
4114
4115 if (IN_SET(who, KILL_ALL, KILL_ALL_FAIL) && u->cgroup_path) {
4116 _cleanup_set_free_ Set *pid_set = NULL;
4117 int q;
4118
4119 /* Exclude the main/control pids from being killed via the cgroup */
4120 pid_set = unit_pid_set(main_pid, control_pid);
4121 if (!pid_set)
4122 return -ENOMEM;
4123
4124 q = cg_kill_recursive(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, signo, 0, pid_set, NULL, NULL);
4125 if (q < 0 && !IN_SET(q, -EAGAIN, -ESRCH, -ENOENT))
4126 r = q;
4127 else
4128 killed = true;
4129 }
4130
4131 if (r == 0 && !killed && IN_SET(who, KILL_ALL_FAIL, KILL_CONTROL_FAIL))
4132 return -ESRCH;
4133
4134 return r;
4135 }
4136
4137 int unit_following_set(Unit *u, Set **s) {
4138 assert(u);
4139 assert(s);
4140
4141 if (UNIT_VTABLE(u)->following_set)
4142 return UNIT_VTABLE(u)->following_set(u, s);
4143
4144 *s = NULL;
4145 return 0;
4146 }
4147
4148 UnitFileState unit_get_unit_file_state(Unit *u) {
4149 int r;
4150
4151 assert(u);
4152
4153 if (u->unit_file_state < 0 && u->fragment_path) {
4154 r = unit_file_get_state(
4155 u->manager->unit_file_scope,
4156 NULL,
4157 u->id,
4158 &u->unit_file_state);
4159 if (r < 0)
4160 u->unit_file_state = UNIT_FILE_BAD;
4161 }
4162
4163 return u->unit_file_state;
4164 }
4165
4166 int unit_get_unit_file_preset(Unit *u) {
4167 assert(u);
4168
4169 if (u->unit_file_preset < 0 && u->fragment_path)
4170 u->unit_file_preset = unit_file_query_preset(
4171 u->manager->unit_file_scope,
4172 NULL,
4173 basename(u->fragment_path));
4174
4175 return u->unit_file_preset;
4176 }
4177
4178 Unit* unit_ref_set(UnitRef *ref, Unit *source, Unit *target) {
4179 assert(ref);
4180 assert(source);
4181 assert(target);
4182
4183 if (ref->target)
4184 unit_ref_unset(ref);
4185
4186 ref->source = source;
4187 ref->target = target;
4188 LIST_PREPEND(refs_by_target, target->refs_by_target, ref);
4189 return target;
4190 }
4191
4192 void unit_ref_unset(UnitRef *ref) {
4193 assert(ref);
4194
4195 if (!ref->target)
4196 return;
4197
4198 /* We are about to drop a reference to the unit, make sure the garbage collection has a look at it as it might
4199 * be unreferenced now. */
4200 unit_add_to_gc_queue(ref->target);
4201
4202 LIST_REMOVE(refs_by_target, ref->target->refs_by_target, ref);
4203 ref->source = ref->target = NULL;
4204 }
4205
4206 static int user_from_unit_name(Unit *u, char **ret) {
4207
4208 static const uint8_t hash_key[] = {
4209 0x58, 0x1a, 0xaf, 0xe6, 0x28, 0x58, 0x4e, 0x96,
4210 0xb4, 0x4e, 0xf5, 0x3b, 0x8c, 0x92, 0x07, 0xec
4211 };
4212
4213 _cleanup_free_ char *n = NULL;
4214 int r;
4215
4216 r = unit_name_to_prefix(u->id, &n);
4217 if (r < 0)
4218 return r;
4219
4220 if (valid_user_group_name(n)) {
4221 *ret = TAKE_PTR(n);
4222 return 0;
4223 }
4224
4225 /* If we can't use the unit name as a user name, then let's hash it and use that */
4226 if (asprintf(ret, "_du%016" PRIx64, siphash24(n, strlen(n), hash_key)) < 0)
4227 return -ENOMEM;
4228
4229 return 0;
4230 }
4231
4232 int unit_patch_contexts(Unit *u) {
4233 CGroupContext *cc;
4234 ExecContext *ec;
4235 unsigned i;
4236 int r;
4237
4238 assert(u);
4239
4240 /* Patch in the manager defaults into the exec and cgroup
4241 * contexts, _after_ the rest of the settings have been
4242 * initialized */
4243
4244 ec = unit_get_exec_context(u);
4245 if (ec) {
4246 /* This only copies in the ones that need memory */
4247 for (i = 0; i < _RLIMIT_MAX; i++)
4248 if (u->manager->rlimit[i] && !ec->rlimit[i]) {
4249 ec->rlimit[i] = newdup(struct rlimit, u->manager->rlimit[i], 1);
4250 if (!ec->rlimit[i])
4251 return -ENOMEM;
4252 }
4253
4254 if (MANAGER_IS_USER(u->manager) &&
4255 !ec->working_directory) {
4256
4257 r = get_home_dir(&ec->working_directory);
4258 if (r < 0)
4259 return r;
4260
4261 /* Allow user services to run, even if the
4262 * home directory is missing */
4263 ec->working_directory_missing_ok = true;
4264 }
4265
4266 if (ec->private_devices)
4267 ec->capability_bounding_set &= ~((UINT64_C(1) << CAP_MKNOD) | (UINT64_C(1) << CAP_SYS_RAWIO));
4268
4269 if (ec->protect_kernel_modules)
4270 ec->capability_bounding_set &= ~(UINT64_C(1) << CAP_SYS_MODULE);
4271
4272 if (ec->protect_kernel_logs)
4273 ec->capability_bounding_set &= ~(UINT64_C(1) << CAP_SYSLOG);
4274
4275 if (ec->dynamic_user) {
4276 if (!ec->user) {
4277 r = user_from_unit_name(u, &ec->user);
4278 if (r < 0)
4279 return r;
4280 }
4281
4282 if (!ec->group) {
4283 ec->group = strdup(ec->user);
4284 if (!ec->group)
4285 return -ENOMEM;
4286 }
4287
4288 /* If the dynamic user option is on, let's make sure that the unit can't leave its
4289 * UID/GID around in the file system or on IPC objects. Hence enforce a strict
4290 * sandbox. */
4291
4292 ec->private_tmp = true;
4293 ec->remove_ipc = true;
4294 ec->protect_system = PROTECT_SYSTEM_STRICT;
4295 if (ec->protect_home == PROTECT_HOME_NO)
4296 ec->protect_home = PROTECT_HOME_READ_ONLY;
4297
4298 /* Make sure this service can neither benefit from SUID/SGID binaries nor create
4299 * them. */
4300 ec->no_new_privileges = true;
4301 ec->restrict_suid_sgid = true;
4302 }
4303 }
4304
4305 cc = unit_get_cgroup_context(u);
4306 if (cc && ec) {
4307
4308 if (ec->private_devices &&
4309 cc->device_policy == CGROUP_DEVICE_POLICY_AUTO)
4310 cc->device_policy = CGROUP_DEVICE_POLICY_CLOSED;
4311
4312 if (ec->root_image &&
4313 (cc->device_policy != CGROUP_DEVICE_POLICY_AUTO || cc->device_allow)) {
4314
4315 /* When RootImage= is specified, the following devices are touched. */
4316 r = cgroup_add_device_allow(cc, "/dev/loop-control", "rw");
4317 if (r < 0)
4318 return r;
4319
4320 r = cgroup_add_device_allow(cc, "block-loop", "rwm");
4321 if (r < 0)
4322 return r;
4323
4324 r = cgroup_add_device_allow(cc, "block-blkext", "rwm");
4325 if (r < 0)
4326 return r;
4327 }
4328 }
4329
4330 return 0;
4331 }
4332
4333 ExecContext *unit_get_exec_context(Unit *u) {
4334 size_t offset;
4335 assert(u);
4336
4337 if (u->type < 0)
4338 return NULL;
4339
4340 offset = UNIT_VTABLE(u)->exec_context_offset;
4341 if (offset <= 0)
4342 return NULL;
4343
4344 return (ExecContext*) ((uint8_t*) u + offset);
4345 }
4346
4347 KillContext *unit_get_kill_context(Unit *u) {
4348 size_t offset;
4349 assert(u);
4350
4351 if (u->type < 0)
4352 return NULL;
4353
4354 offset = UNIT_VTABLE(u)->kill_context_offset;
4355 if (offset <= 0)
4356 return NULL;
4357
4358 return (KillContext*) ((uint8_t*) u + offset);
4359 }
4360
4361 CGroupContext *unit_get_cgroup_context(Unit *u) {
4362 size_t offset;
4363
4364 if (u->type < 0)
4365 return NULL;
4366
4367 offset = UNIT_VTABLE(u)->cgroup_context_offset;
4368 if (offset <= 0)
4369 return NULL;
4370
4371 return (CGroupContext*) ((uint8_t*) u + offset);
4372 }
4373
4374 ExecRuntime *unit_get_exec_runtime(Unit *u) {
4375 size_t offset;
4376
4377 if (u->type < 0)
4378 return NULL;
4379
4380 offset = UNIT_VTABLE(u)->exec_runtime_offset;
4381 if (offset <= 0)
4382 return NULL;
4383
4384 return *(ExecRuntime**) ((uint8_t*) u + offset);
4385 }
4386
4387 static const char* unit_drop_in_dir(Unit *u, UnitWriteFlags flags) {
4388 assert(u);
4389
4390 if (UNIT_WRITE_FLAGS_NOOP(flags))
4391 return NULL;
4392
4393 if (u->transient) /* Redirect drop-ins for transient units always into the transient directory. */
4394 return u->manager->lookup_paths.transient;
4395
4396 if (flags & UNIT_PERSISTENT)
4397 return u->manager->lookup_paths.persistent_control;
4398
4399 if (flags & UNIT_RUNTIME)
4400 return u->manager->lookup_paths.runtime_control;
4401
4402 return NULL;
4403 }
4404
4405 char* unit_escape_setting(const char *s, UnitWriteFlags flags, char **buf) {
4406 char *ret = NULL;
4407
4408 if (!s)
4409 return NULL;
4410
4411 /* Escapes the input string as requested. Returns the escaped string. If 'buf' is specified then the allocated
4412 * return buffer pointer is also written to *buf, except if no escaping was necessary, in which case *buf is
4413 * set to NULL, and the input pointer is returned as-is. This means the return value always contains a properly
4414 * escaped version, but *buf when passed only contains a pointer if an allocation was necessary. If *buf is
4415 * not specified, then the return value always needs to be freed. Callers can use this to optimize memory
4416 * allocations. */
4417
4418 if (flags & UNIT_ESCAPE_SPECIFIERS) {
4419 ret = specifier_escape(s);
4420 if (!ret)
4421 return NULL;
4422
4423 s = ret;
4424 }
4425
4426 if (flags & UNIT_ESCAPE_C) {
4427 char *a;
4428
4429 a = cescape(s);
4430 free(ret);
4431 if (!a)
4432 return NULL;
4433
4434 ret = a;
4435 }
4436
4437 if (buf) {
4438 *buf = ret;
4439 return ret ?: (char*) s;
4440 }
4441
4442 return ret ?: strdup(s);
4443 }
4444
4445 char* unit_concat_strv(char **l, UnitWriteFlags flags) {
4446 _cleanup_free_ char *result = NULL;
4447 size_t n = 0, allocated = 0;
4448 char **i;
4449
4450 /* Takes a list of strings, escapes them, and concatenates them. This may be used to format command lines in a
4451 * way suitable for ExecStart= stanzas */
4452
4453 STRV_FOREACH(i, l) {
4454 _cleanup_free_ char *buf = NULL;
4455 const char *p;
4456 size_t a;
4457 char *q;
4458
4459 p = unit_escape_setting(*i, flags, &buf);
4460 if (!p)
4461 return NULL;
4462
4463 a = (n > 0) + 1 + strlen(p) + 1; /* separating space + " + entry + " */
4464 if (!GREEDY_REALLOC(result, allocated, n + a + 1))
4465 return NULL;
4466
4467 q = result + n;
4468 if (n > 0)
4469 *(q++) = ' ';
4470
4471 *(q++) = '"';
4472 q = stpcpy(q, p);
4473 *(q++) = '"';
4474
4475 n += a;
4476 }
4477
4478 if (!GREEDY_REALLOC(result, allocated, n + 1))
4479 return NULL;
4480
4481 result[n] = 0;
4482
4483 return TAKE_PTR(result);
4484 }
4485
4486 int unit_write_setting(Unit *u, UnitWriteFlags flags, const char *name, const char *data) {
4487 _cleanup_free_ char *p = NULL, *q = NULL, *escaped = NULL;
4488 const char *dir, *wrapped;
4489 int r;
4490
4491 assert(u);
4492 assert(name);
4493 assert(data);
4494
4495 if (UNIT_WRITE_FLAGS_NOOP(flags))
4496 return 0;
4497
4498 data = unit_escape_setting(data, flags, &escaped);
4499 if (!data)
4500 return -ENOMEM;
4501
4502 /* Prefix the section header. If we are writing this out as transient file, then let's suppress this if the
4503 * previous section header is the same */
4504
4505 if (flags & UNIT_PRIVATE) {
4506 if (!UNIT_VTABLE(u)->private_section)
4507 return -EINVAL;
4508
4509 if (!u->transient_file || u->last_section_private < 0)
4510 data = strjoina("[", UNIT_VTABLE(u)->private_section, "]\n", data);
4511 else if (u->last_section_private == 0)
4512 data = strjoina("\n[", UNIT_VTABLE(u)->private_section, "]\n", data);
4513 } else {
4514 if (!u->transient_file || u->last_section_private < 0)
4515 data = strjoina("[Unit]\n", data);
4516 else if (u->last_section_private > 0)
4517 data = strjoina("\n[Unit]\n", data);
4518 }
4519
4520 if (u->transient_file) {
4521 /* When this is a transient unit file in creation, then let's not create a new drop-in but instead
4522 * write to the transient unit file. */
4523 fputs(data, u->transient_file);
4524
4525 if (!endswith(data, "\n"))
4526 fputc('\n', u->transient_file);
4527
4528 /* Remember which section we wrote this entry to */
4529 u->last_section_private = !!(flags & UNIT_PRIVATE);
4530 return 0;
4531 }
4532
4533 dir = unit_drop_in_dir(u, flags);
4534 if (!dir)
4535 return -EINVAL;
4536
4537 wrapped = strjoina("# This is a drop-in unit file extension, created via \"systemctl set-property\"\n"
4538 "# or an equivalent operation. Do not edit.\n",
4539 data,
4540 "\n");
4541
4542 r = drop_in_file(dir, u->id, 50, name, &p, &q);
4543 if (r < 0)
4544 return r;
4545
4546 (void) mkdir_p_label(p, 0755);
4547
4548 /* Make sure the drop-in dir is registered in our path cache. This way we don't need to stupidly
4549 * recreate the cache after every drop-in we write. */
4550 if (u->manager->unit_path_cache) {
4551 r = set_put_strdup(u->manager->unit_path_cache, p);
4552 if (r < 0)
4553 return r;
4554 }
4555
4556 r = write_string_file_atomic_label(q, wrapped);
4557 if (r < 0)
4558 return r;
4559
4560 r = strv_push(&u->dropin_paths, q);
4561 if (r < 0)
4562 return r;
4563 q = NULL;
4564
4565 strv_uniq(u->dropin_paths);
4566
4567 u->dropin_mtime = now(CLOCK_REALTIME);
4568
4569 return 0;
4570 }
4571
4572 int unit_write_settingf(Unit *u, UnitWriteFlags flags, const char *name, const char *format, ...) {
4573 _cleanup_free_ char *p = NULL;
4574 va_list ap;
4575 int r;
4576
4577 assert(u);
4578 assert(name);
4579 assert(format);
4580
4581 if (UNIT_WRITE_FLAGS_NOOP(flags))
4582 return 0;
4583
4584 va_start(ap, format);
4585 r = vasprintf(&p, format, ap);
4586 va_end(ap);
4587
4588 if (r < 0)
4589 return -ENOMEM;
4590
4591 return unit_write_setting(u, flags, name, p);
4592 }
4593
4594 int unit_make_transient(Unit *u) {
4595 _cleanup_free_ char *path = NULL;
4596 FILE *f;
4597
4598 assert(u);
4599
4600 if (!UNIT_VTABLE(u)->can_transient)
4601 return -EOPNOTSUPP;
4602
4603 (void) mkdir_p_label(u->manager->lookup_paths.transient, 0755);
4604
4605 path = path_join(u->manager->lookup_paths.transient, u->id);
4606 if (!path)
4607 return -ENOMEM;
4608
4609 /* Let's open the file we'll write the transient settings into. This file is kept open as long as we are
4610 * creating the transient, and is closed in unit_load(), as soon as we start loading the file. */
4611
4612 RUN_WITH_UMASK(0022) {
4613 f = fopen(path, "we");
4614 if (!f)
4615 return -errno;
4616 }
4617
4618 safe_fclose(u->transient_file);
4619 u->transient_file = f;
4620
4621 free_and_replace(u->fragment_path, path);
4622
4623 u->source_path = mfree(u->source_path);
4624 u->dropin_paths = strv_free(u->dropin_paths);
4625 u->fragment_mtime = u->source_mtime = u->dropin_mtime = 0;
4626
4627 u->load_state = UNIT_STUB;
4628 u->load_error = 0;
4629 u->transient = true;
4630
4631 unit_add_to_dbus_queue(u);
4632 unit_add_to_gc_queue(u);
4633
4634 fputs("# This is a transient unit file, created programmatically via the systemd API. Do not edit.\n",
4635 u->transient_file);
4636
4637 return 0;
4638 }
4639
4640 static int log_kill(pid_t pid, int sig, void *userdata) {
4641 _cleanup_free_ char *comm = NULL;
4642
4643 (void) get_process_comm(pid, &comm);
4644
4645 /* Don't log about processes marked with brackets, under the assumption that these are temporary processes
4646 only, like for example systemd's own PAM stub process. */
4647 if (comm && comm[0] == '(')
4648 return 0;
4649
4650 log_unit_notice(userdata,
4651 "Killing process " PID_FMT " (%s) with signal SIG%s.",
4652 pid,
4653 strna(comm),
4654 signal_to_string(sig));
4655
4656 return 1;
4657 }
4658
4659 static int operation_to_signal(const KillContext *c, KillOperation k, bool *noteworthy) {
4660 assert(c);
4661
4662 switch (k) {
4663
4664 case KILL_TERMINATE:
4665 case KILL_TERMINATE_AND_LOG:
4666 *noteworthy = false;
4667 return c->kill_signal;
4668
4669 case KILL_RESTART:
4670 *noteworthy = false;
4671 return restart_kill_signal(c);
4672
4673 case KILL_KILL:
4674 *noteworthy = true;
4675 return c->final_kill_signal;
4676
4677 case KILL_WATCHDOG:
4678 *noteworthy = true;
4679 return c->watchdog_signal;
4680
4681 default:
4682 assert_not_reached("KillOperation unknown");
4683 }
4684 }
4685
4686 int unit_kill_context(
4687 Unit *u,
4688 KillContext *c,
4689 KillOperation k,
4690 pid_t main_pid,
4691 pid_t control_pid,
4692 bool main_pid_alien) {
4693
4694 bool wait_for_exit = false, send_sighup;
4695 cg_kill_log_func_t log_func = NULL;
4696 int sig, r;
4697
4698 assert(u);
4699 assert(c);
4700
4701 /* Kill the processes belonging to this unit, in preparation for shutting the unit down.
4702 * Returns > 0 if we killed something worth waiting for, 0 otherwise. */
4703
4704 if (c->kill_mode == KILL_NONE)
4705 return 0;
4706
4707 bool noteworthy;
4708 sig = operation_to_signal(c, k, &noteworthy);
4709 if (noteworthy)
4710 log_func = log_kill;
4711
4712 send_sighup =
4713 c->send_sighup &&
4714 IN_SET(k, KILL_TERMINATE, KILL_TERMINATE_AND_LOG) &&
4715 sig != SIGHUP;
4716
4717 if (main_pid > 0) {
4718 if (log_func)
4719 log_func(main_pid, sig, u);
4720
4721 r = kill_and_sigcont(main_pid, sig);
4722 if (r < 0 && r != -ESRCH) {
4723 _cleanup_free_ char *comm = NULL;
4724 (void) get_process_comm(main_pid, &comm);
4725
4726 log_unit_warning_errno(u, r, "Failed to kill main process " PID_FMT " (%s), ignoring: %m", main_pid, strna(comm));
4727 } else {
4728 if (!main_pid_alien)
4729 wait_for_exit = true;
4730
4731 if (r != -ESRCH && send_sighup)
4732 (void) kill(main_pid, SIGHUP);
4733 }
4734 }
4735
4736 if (control_pid > 0) {
4737 if (log_func)
4738 log_func(control_pid, sig, u);
4739
4740 r = kill_and_sigcont(control_pid, sig);
4741 if (r < 0 && r != -ESRCH) {
4742 _cleanup_free_ char *comm = NULL;
4743 (void) get_process_comm(control_pid, &comm);
4744
4745 log_unit_warning_errno(u, r, "Failed to kill control process " PID_FMT " (%s), ignoring: %m", control_pid, strna(comm));
4746 } else {
4747 wait_for_exit = true;
4748
4749 if (r != -ESRCH && send_sighup)
4750 (void) kill(control_pid, SIGHUP);
4751 }
4752 }
4753
4754 if (u->cgroup_path &&
4755 (c->kill_mode == KILL_CONTROL_GROUP || (c->kill_mode == KILL_MIXED && k == KILL_KILL))) {
4756 _cleanup_set_free_ Set *pid_set = NULL;
4757
4758 /* Exclude the main/control pids from being killed via the cgroup */
4759 pid_set = unit_pid_set(main_pid, control_pid);
4760 if (!pid_set)
4761 return -ENOMEM;
4762
4763 r = cg_kill_recursive(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path,
4764 sig,
4765 CGROUP_SIGCONT|CGROUP_IGNORE_SELF,
4766 pid_set,
4767 log_func, u);
4768 if (r < 0) {
4769 if (!IN_SET(r, -EAGAIN, -ESRCH, -ENOENT))
4770 log_unit_warning_errno(u, r, "Failed to kill control group %s, ignoring: %m", u->cgroup_path);
4771
4772 } else if (r > 0) {
4773
4774 /* FIXME: For now, on the legacy hierarchy, we will not wait for the cgroup members to die if
4775 * we are running in a container or if this is a delegation unit, simply because cgroup
4776 * notification is unreliable in these cases. It doesn't work at all in containers, and outside
4777 * of containers it can be confused easily by left-over directories in the cgroup — which
4778 * however should not exist in non-delegated units. On the unified hierarchy that's different,
4779 * there we get proper events. Hence rely on them. */
4780
4781 if (cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER) > 0 ||
4782 (detect_container() == 0 && !unit_cgroup_delegate(u)))
4783 wait_for_exit = true;
4784
4785 if (send_sighup) {
4786 set_free(pid_set);
4787
4788 pid_set = unit_pid_set(main_pid, control_pid);
4789 if (!pid_set)
4790 return -ENOMEM;
4791
4792 cg_kill_recursive(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path,
4793 SIGHUP,
4794 CGROUP_IGNORE_SELF,
4795 pid_set,
4796 NULL, NULL);
4797 }
4798 }
4799 }
4800
4801 return wait_for_exit;
4802 }
4803
4804 int unit_require_mounts_for(Unit *u, const char *path, UnitDependencyMask mask) {
4805 _cleanup_free_ char *p = NULL;
4806 UnitDependencyInfo di;
4807 int r;
4808
4809 assert(u);
4810 assert(path);
4811
4812 /* Registers a unit for requiring a certain path and all its prefixes. We keep a hashtable of these paths in
4813 * the unit (from the path to the UnitDependencyInfo structure indicating how to the dependency came to
4814 * be). However, we build a prefix table for all possible prefixes so that new appearing mount units can easily
4815 * determine which units to make themselves a dependency of. */
4816
4817 if (!path_is_absolute(path))
4818 return -EINVAL;
4819
4820 r = hashmap_ensure_allocated(&u->requires_mounts_for, &path_hash_ops);
4821 if (r < 0)
4822 return r;
4823
4824 p = strdup(path);
4825 if (!p)
4826 return -ENOMEM;
4827
4828 path = path_simplify(p, true);
4829
4830 if (!path_is_normalized(path))
4831 return -EPERM;
4832
4833 if (hashmap_contains(u->requires_mounts_for, path))
4834 return 0;
4835
4836 di = (UnitDependencyInfo) {
4837 .origin_mask = mask
4838 };
4839
4840 r = hashmap_put(u->requires_mounts_for, path, di.data);
4841 if (r < 0)
4842 return r;
4843 p = NULL;
4844
4845 char prefix[strlen(path) + 1];
4846 PATH_FOREACH_PREFIX_MORE(prefix, path) {
4847 Set *x;
4848
4849 x = hashmap_get(u->manager->units_requiring_mounts_for, prefix);
4850 if (!x) {
4851 _cleanup_free_ char *q = NULL;
4852
4853 r = hashmap_ensure_allocated(&u->manager->units_requiring_mounts_for, &path_hash_ops);
4854 if (r < 0)
4855 return r;
4856
4857 q = strdup(prefix);
4858 if (!q)
4859 return -ENOMEM;
4860
4861 x = set_new(NULL);
4862 if (!x)
4863 return -ENOMEM;
4864
4865 r = hashmap_put(u->manager->units_requiring_mounts_for, q, x);
4866 if (r < 0) {
4867 set_free(x);
4868 return r;
4869 }
4870 q = NULL;
4871 }
4872
4873 r = set_put(x, u);
4874 if (r < 0)
4875 return r;
4876 }
4877
4878 return 0;
4879 }
4880
4881 int unit_setup_exec_runtime(Unit *u) {
4882 ExecRuntime **rt;
4883 size_t offset;
4884 Unit *other;
4885 Iterator i;
4886 void *v;
4887 int r;
4888
4889 offset = UNIT_VTABLE(u)->exec_runtime_offset;
4890 assert(offset > 0);
4891
4892 /* Check if there already is an ExecRuntime for this unit? */
4893 rt = (ExecRuntime**) ((uint8_t*) u + offset);
4894 if (*rt)
4895 return 0;
4896
4897 /* Try to get it from somebody else */
4898 HASHMAP_FOREACH_KEY(v, other, u->dependencies[UNIT_JOINS_NAMESPACE_OF], i) {
4899 r = exec_runtime_acquire(u->manager, NULL, other->id, false, rt);
4900 if (r == 1)
4901 return 1;
4902 }
4903
4904 return exec_runtime_acquire(u->manager, unit_get_exec_context(u), u->id, true, rt);
4905 }
4906
4907 int unit_setup_dynamic_creds(Unit *u) {
4908 ExecContext *ec;
4909 DynamicCreds *dcreds;
4910 size_t offset;
4911
4912 assert(u);
4913
4914 offset = UNIT_VTABLE(u)->dynamic_creds_offset;
4915 assert(offset > 0);
4916 dcreds = (DynamicCreds*) ((uint8_t*) u + offset);
4917
4918 ec = unit_get_exec_context(u);
4919 assert(ec);
4920
4921 if (!ec->dynamic_user)
4922 return 0;
4923
4924 return dynamic_creds_acquire(dcreds, u->manager, ec->user, ec->group);
4925 }
4926
4927 bool unit_type_supported(UnitType t) {
4928 if (_unlikely_(t < 0))
4929 return false;
4930 if (_unlikely_(t >= _UNIT_TYPE_MAX))
4931 return false;
4932
4933 if (!unit_vtable[t]->supported)
4934 return true;
4935
4936 return unit_vtable[t]->supported();
4937 }
4938
4939 void unit_warn_if_dir_nonempty(Unit *u, const char* where) {
4940 int r;
4941
4942 assert(u);
4943 assert(where);
4944
4945 r = dir_is_empty(where);
4946 if (r > 0 || r == -ENOTDIR)
4947 return;
4948 if (r < 0) {
4949 log_unit_warning_errno(u, r, "Failed to check directory %s: %m", where);
4950 return;
4951 }
4952
4953 log_struct(LOG_NOTICE,
4954 "MESSAGE_ID=" SD_MESSAGE_OVERMOUNTING_STR,
4955 LOG_UNIT_ID(u),
4956 LOG_UNIT_INVOCATION_ID(u),
4957 LOG_UNIT_MESSAGE(u, "Directory %s to mount over is not empty, mounting anyway.", where),
4958 "WHERE=%s", where);
4959 }
4960
4961 int unit_fail_if_noncanonical(Unit *u, const char* where) {
4962 _cleanup_free_ char *canonical_where = NULL;
4963 int r;
4964
4965 assert(u);
4966 assert(where);
4967
4968 r = chase_symlinks(where, NULL, CHASE_NONEXISTENT, &canonical_where, NULL);
4969 if (r < 0) {
4970 log_unit_debug_errno(u, r, "Failed to check %s for symlinks, ignoring: %m", where);
4971 return 0;
4972 }
4973
4974 /* We will happily ignore a trailing slash (or any redundant slashes) */
4975 if (path_equal(where, canonical_where))
4976 return 0;
4977
4978 /* No need to mention "." or "..", they would already have been rejected by unit_name_from_path() */
4979 log_struct(LOG_ERR,
4980 "MESSAGE_ID=" SD_MESSAGE_OVERMOUNTING_STR,
4981 LOG_UNIT_ID(u),
4982 LOG_UNIT_INVOCATION_ID(u),
4983 LOG_UNIT_MESSAGE(u, "Mount path %s is not canonical (contains a symlink).", where),
4984 "WHERE=%s", where);
4985
4986 return -ELOOP;
4987 }
4988
4989 bool unit_is_pristine(Unit *u) {
4990 assert(u);
4991
4992 /* Check if the unit already exists or is already around,
4993 * in a number of different ways. Note that to cater for unit
4994 * types such as slice, we are generally fine with units that
4995 * are marked UNIT_LOADED even though nothing was actually
4996 * loaded, as those unit types don't require a file on disk. */
4997
4998 return !(!IN_SET(u->load_state, UNIT_NOT_FOUND, UNIT_LOADED) ||
4999 u->fragment_path ||
5000 u->source_path ||
5001 !strv_isempty(u->dropin_paths) ||
5002 u->job ||
5003 u->merged_into);
5004 }
5005
5006 pid_t unit_control_pid(Unit *u) {
5007 assert(u);
5008
5009 if (UNIT_VTABLE(u)->control_pid)
5010 return UNIT_VTABLE(u)->control_pid(u);
5011
5012 return 0;
5013 }
5014
5015 pid_t unit_main_pid(Unit *u) {
5016 assert(u);
5017
5018 if (UNIT_VTABLE(u)->main_pid)
5019 return UNIT_VTABLE(u)->main_pid(u);
5020
5021 return 0;
5022 }
5023
5024 static void unit_unref_uid_internal(
5025 Unit *u,
5026 uid_t *ref_uid,
5027 bool destroy_now,
5028 void (*_manager_unref_uid)(Manager *m, uid_t uid, bool destroy_now)) {
5029
5030 assert(u);
5031 assert(ref_uid);
5032 assert(_manager_unref_uid);
5033
5034 /* Generic implementation of both unit_unref_uid() and unit_unref_gid(), under the assumption that uid_t and
5035 * gid_t are actually the same time, with the same validity rules.
5036 *
5037 * Drops a reference to UID/GID from a unit. */
5038
5039 assert_cc(sizeof(uid_t) == sizeof(gid_t));
5040 assert_cc(UID_INVALID == (uid_t) GID_INVALID);
5041
5042 if (!uid_is_valid(*ref_uid))
5043 return;
5044
5045 _manager_unref_uid(u->manager, *ref_uid, destroy_now);
5046 *ref_uid = UID_INVALID;
5047 }
5048
5049 void unit_unref_uid(Unit *u, bool destroy_now) {
5050 unit_unref_uid_internal(u, &u->ref_uid, destroy_now, manager_unref_uid);
5051 }
5052
5053 void unit_unref_gid(Unit *u, bool destroy_now) {
5054 unit_unref_uid_internal(u, (uid_t*) &u->ref_gid, destroy_now, manager_unref_gid);
5055 }
5056
5057 static int unit_ref_uid_internal(
5058 Unit *u,
5059 uid_t *ref_uid,
5060 uid_t uid,
5061 bool clean_ipc,
5062 int (*_manager_ref_uid)(Manager *m, uid_t uid, bool clean_ipc)) {
5063
5064 int r;
5065
5066 assert(u);
5067 assert(ref_uid);
5068 assert(uid_is_valid(uid));
5069 assert(_manager_ref_uid);
5070
5071 /* Generic implementation of both unit_ref_uid() and unit_ref_guid(), under the assumption that uid_t and gid_t
5072 * are actually the same type, and have the same validity rules.
5073 *
5074 * Adds a reference on a specific UID/GID to this unit. Each unit referencing the same UID/GID maintains a
5075 * reference so that we can destroy the UID/GID's IPC resources as soon as this is requested and the counter
5076 * drops to zero. */
5077
5078 assert_cc(sizeof(uid_t) == sizeof(gid_t));
5079 assert_cc(UID_INVALID == (uid_t) GID_INVALID);
5080
5081 if (*ref_uid == uid)
5082 return 0;
5083
5084 if (uid_is_valid(*ref_uid)) /* Already set? */
5085 return -EBUSY;
5086
5087 r = _manager_ref_uid(u->manager, uid, clean_ipc);
5088 if (r < 0)
5089 return r;
5090
5091 *ref_uid = uid;
5092 return 1;
5093 }
5094
5095 int unit_ref_uid(Unit *u, uid_t uid, bool clean_ipc) {
5096 return unit_ref_uid_internal(u, &u->ref_uid, uid, clean_ipc, manager_ref_uid);
5097 }
5098
5099 int unit_ref_gid(Unit *u, gid_t gid, bool clean_ipc) {
5100 return unit_ref_uid_internal(u, (uid_t*) &u->ref_gid, (uid_t) gid, clean_ipc, manager_ref_gid);
5101 }
5102
5103 static int unit_ref_uid_gid_internal(Unit *u, uid_t uid, gid_t gid, bool clean_ipc) {
5104 int r = 0, q = 0;
5105
5106 assert(u);
5107
5108 /* Reference both a UID and a GID in one go. Either references both, or neither. */
5109
5110 if (uid_is_valid(uid)) {
5111 r = unit_ref_uid(u, uid, clean_ipc);
5112 if (r < 0)
5113 return r;
5114 }
5115
5116 if (gid_is_valid(gid)) {
5117 q = unit_ref_gid(u, gid, clean_ipc);
5118 if (q < 0) {
5119 if (r > 0)
5120 unit_unref_uid(u, false);
5121
5122 return q;
5123 }
5124 }
5125
5126 return r > 0 || q > 0;
5127 }
5128
5129 int unit_ref_uid_gid(Unit *u, uid_t uid, gid_t gid) {
5130 ExecContext *c;
5131 int r;
5132
5133 assert(u);
5134
5135 c = unit_get_exec_context(u);
5136
5137 r = unit_ref_uid_gid_internal(u, uid, gid, c ? c->remove_ipc : false);
5138 if (r < 0)
5139 return log_unit_warning_errno(u, r, "Couldn't add UID/GID reference to unit, proceeding without: %m");
5140
5141 return r;
5142 }
5143
5144 void unit_unref_uid_gid(Unit *u, bool destroy_now) {
5145 assert(u);
5146
5147 unit_unref_uid(u, destroy_now);
5148 unit_unref_gid(u, destroy_now);
5149 }
5150
5151 void unit_notify_user_lookup(Unit *u, uid_t uid, gid_t gid) {
5152 int r;
5153
5154 assert(u);
5155
5156 /* This is invoked whenever one of the forked off processes let's us know the UID/GID its user name/group names
5157 * resolved to. We keep track of which UID/GID is currently assigned in order to be able to destroy its IPC
5158 * objects when no service references the UID/GID anymore. */
5159
5160 r = unit_ref_uid_gid(u, uid, gid);
5161 if (r > 0)
5162 unit_add_to_dbus_queue(u);
5163 }
5164
5165 int unit_set_invocation_id(Unit *u, sd_id128_t id) {
5166 int r;
5167
5168 assert(u);
5169
5170 /* Set the invocation ID for this unit. If we cannot, this will not roll back, but reset the whole thing. */
5171
5172 if (sd_id128_equal(u->invocation_id, id))
5173 return 0;
5174
5175 if (!sd_id128_is_null(u->invocation_id))
5176 (void) hashmap_remove_value(u->manager->units_by_invocation_id, &u->invocation_id, u);
5177
5178 if (sd_id128_is_null(id)) {
5179 r = 0;
5180 goto reset;
5181 }
5182
5183 r = hashmap_ensure_allocated(&u->manager->units_by_invocation_id, &id128_hash_ops);
5184 if (r < 0)
5185 goto reset;
5186
5187 u->invocation_id = id;
5188 sd_id128_to_string(id, u->invocation_id_string);
5189
5190 r = hashmap_put(u->manager->units_by_invocation_id, &u->invocation_id, u);
5191 if (r < 0)
5192 goto reset;
5193
5194 return 0;
5195
5196 reset:
5197 u->invocation_id = SD_ID128_NULL;
5198 u->invocation_id_string[0] = 0;
5199 return r;
5200 }
5201
5202 int unit_acquire_invocation_id(Unit *u) {
5203 sd_id128_t id;
5204 int r;
5205
5206 assert(u);
5207
5208 r = sd_id128_randomize(&id);
5209 if (r < 0)
5210 return log_unit_error_errno(u, r, "Failed to generate invocation ID for unit: %m");
5211
5212 r = unit_set_invocation_id(u, id);
5213 if (r < 0)
5214 return log_unit_error_errno(u, r, "Failed to set invocation ID for unit: %m");
5215
5216 unit_add_to_dbus_queue(u);
5217 return 0;
5218 }
5219
5220 int unit_set_exec_params(Unit *u, ExecParameters *p) {
5221 int r;
5222
5223 assert(u);
5224 assert(p);
5225
5226 /* Copy parameters from manager */
5227 r = manager_get_effective_environment(u->manager, &p->environment);
5228 if (r < 0)
5229 return r;
5230
5231 p->confirm_spawn = manager_get_confirm_spawn(u->manager);
5232 p->cgroup_supported = u->manager->cgroup_supported;
5233 p->prefix = u->manager->prefix;
5234 SET_FLAG(p->flags, EXEC_PASS_LOG_UNIT|EXEC_CHOWN_DIRECTORIES, MANAGER_IS_SYSTEM(u->manager));
5235
5236 /* Copy parameters from unit */
5237 p->cgroup_path = u->cgroup_path;
5238 SET_FLAG(p->flags, EXEC_CGROUP_DELEGATE, unit_cgroup_delegate(u));
5239
5240 return 0;
5241 }
5242
5243 int unit_fork_helper_process(Unit *u, const char *name, pid_t *ret) {
5244 int r;
5245
5246 assert(u);
5247 assert(ret);
5248
5249 /* Forks off a helper process and makes sure it is a member of the unit's cgroup. Returns == 0 in the child,
5250 * and > 0 in the parent. The pid parameter is always filled in with the child's PID. */
5251
5252 (void) unit_realize_cgroup(u);
5253
5254 r = safe_fork(name, FORK_REOPEN_LOG, ret);
5255 if (r != 0)
5256 return r;
5257
5258 (void) default_signals(SIGNALS_CRASH_HANDLER, SIGNALS_IGNORE, -1);
5259 (void) ignore_signals(SIGPIPE, -1);
5260
5261 (void) prctl(PR_SET_PDEATHSIG, SIGTERM);
5262
5263 if (u->cgroup_path) {
5264 r = cg_attach_everywhere(u->manager->cgroup_supported, u->cgroup_path, 0, NULL, NULL);
5265 if (r < 0) {
5266 log_unit_error_errno(u, r, "Failed to join unit cgroup %s: %m", u->cgroup_path);
5267 _exit(EXIT_CGROUP);
5268 }
5269 }
5270
5271 return 0;
5272 }
5273
5274 int unit_fork_and_watch_rm_rf(Unit *u, char **paths, pid_t *ret_pid) {
5275 pid_t pid;
5276 int r;
5277
5278 assert(u);
5279 assert(ret_pid);
5280
5281 r = unit_fork_helper_process(u, "(sd-rmrf)", &pid);
5282 if (r < 0)
5283 return r;
5284 if (r == 0) {
5285 int ret = EXIT_SUCCESS;
5286 char **i;
5287
5288 STRV_FOREACH(i, paths) {
5289 r = rm_rf(*i, REMOVE_ROOT|REMOVE_PHYSICAL|REMOVE_MISSING_OK);
5290 if (r < 0) {
5291 log_error_errno(r, "Failed to remove '%s': %m", *i);
5292 ret = EXIT_FAILURE;
5293 }
5294 }
5295
5296 _exit(ret);
5297 }
5298
5299 r = unit_watch_pid(u, pid, true);
5300 if (r < 0)
5301 return r;
5302
5303 *ret_pid = pid;
5304 return 0;
5305 }
5306
5307 static void unit_update_dependency_mask(Unit *u, UnitDependency d, Unit *other, UnitDependencyInfo di) {
5308 assert(u);
5309 assert(d >= 0);
5310 assert(d < _UNIT_DEPENDENCY_MAX);
5311 assert(other);
5312
5313 if (di.origin_mask == 0 && di.destination_mask == 0) {
5314 /* No bit set anymore, let's drop the whole entry */
5315 assert_se(hashmap_remove(u->dependencies[d], other));
5316 log_unit_debug(u, "%s lost dependency %s=%s", u->id, unit_dependency_to_string(d), other->id);
5317 } else
5318 /* Mask was reduced, let's update the entry */
5319 assert_se(hashmap_update(u->dependencies[d], other, di.data) == 0);
5320 }
5321
5322 void unit_remove_dependencies(Unit *u, UnitDependencyMask mask) {
5323 UnitDependency d;
5324
5325 assert(u);
5326
5327 /* Removes all dependencies u has on other units marked for ownership by 'mask'. */
5328
5329 if (mask == 0)
5330 return;
5331
5332 for (d = 0; d < _UNIT_DEPENDENCY_MAX; d++) {
5333 bool done;
5334
5335 do {
5336 UnitDependencyInfo di;
5337 Unit *other;
5338 Iterator i;
5339
5340 done = true;
5341
5342 HASHMAP_FOREACH_KEY(di.data, other, u->dependencies[d], i) {
5343 UnitDependency q;
5344
5345 if ((di.origin_mask & ~mask) == di.origin_mask)
5346 continue;
5347 di.origin_mask &= ~mask;
5348 unit_update_dependency_mask(u, d, other, di);
5349
5350 /* We updated the dependency from our unit to the other unit now. But most dependencies
5351 * imply a reverse dependency. Hence, let's delete that one too. For that we go through
5352 * all dependency types on the other unit and delete all those which point to us and
5353 * have the right mask set. */
5354
5355 for (q = 0; q < _UNIT_DEPENDENCY_MAX; q++) {
5356 UnitDependencyInfo dj;
5357
5358 dj.data = hashmap_get(other->dependencies[q], u);
5359 if ((dj.destination_mask & ~mask) == dj.destination_mask)
5360 continue;
5361 dj.destination_mask &= ~mask;
5362
5363 unit_update_dependency_mask(other, q, u, dj);
5364 }
5365
5366 unit_add_to_gc_queue(other);
5367
5368 done = false;
5369 break;
5370 }
5371
5372 } while (!done);
5373 }
5374 }
5375
5376 static int unit_get_invocation_path(Unit *u, char **ret) {
5377 char *p;
5378 int r;
5379
5380 assert(u);
5381 assert(ret);
5382
5383 if (MANAGER_IS_SYSTEM(u->manager))
5384 p = strjoin("/run/systemd/units/invocation:", u->id);
5385 else {
5386 _cleanup_free_ char *user_path = NULL;
5387 r = xdg_user_runtime_dir(&user_path, "/systemd/units/invocation:");
5388 if (r < 0)
5389 return r;
5390 p = strjoin(user_path, u->id);
5391 }
5392
5393 if (!p)
5394 return -ENOMEM;
5395
5396 *ret = p;
5397 return 0;
5398 }
5399
5400 static int unit_export_invocation_id(Unit *u) {
5401 _cleanup_free_ char *p = NULL;
5402 int r;
5403
5404 assert(u);
5405
5406 if (u->exported_invocation_id)
5407 return 0;
5408
5409 if (sd_id128_is_null(u->invocation_id))
5410 return 0;
5411
5412 r = unit_get_invocation_path(u, &p);
5413 if (r < 0)
5414 return log_unit_debug_errno(u, r, "Failed to get invocation path: %m");
5415
5416 r = symlink_atomic(u->invocation_id_string, p);
5417 if (r < 0)
5418 return log_unit_debug_errno(u, r, "Failed to create invocation ID symlink %s: %m", p);
5419
5420 u->exported_invocation_id = true;
5421 return 0;
5422 }
5423
5424 static int unit_export_log_level_max(Unit *u, const ExecContext *c) {
5425 const char *p;
5426 char buf[2];
5427 int r;
5428
5429 assert(u);
5430 assert(c);
5431
5432 if (u->exported_log_level_max)
5433 return 0;
5434
5435 if (c->log_level_max < 0)
5436 return 0;
5437
5438 assert(c->log_level_max <= 7);
5439
5440 buf[0] = '0' + c->log_level_max;
5441 buf[1] = 0;
5442
5443 p = strjoina("/run/systemd/units/log-level-max:", u->id);
5444 r = symlink_atomic(buf, p);
5445 if (r < 0)
5446 return log_unit_debug_errno(u, r, "Failed to create maximum log level symlink %s: %m", p);
5447
5448 u->exported_log_level_max = true;
5449 return 0;
5450 }
5451
5452 static int unit_export_log_extra_fields(Unit *u, const ExecContext *c) {
5453 _cleanup_close_ int fd = -1;
5454 struct iovec *iovec;
5455 const char *p;
5456 char *pattern;
5457 le64_t *sizes;
5458 ssize_t n;
5459 size_t i;
5460 int r;
5461
5462 if (u->exported_log_extra_fields)
5463 return 0;
5464
5465 if (c->n_log_extra_fields <= 0)
5466 return 0;
5467
5468 sizes = newa(le64_t, c->n_log_extra_fields);
5469 iovec = newa(struct iovec, c->n_log_extra_fields * 2);
5470
5471 for (i = 0; i < c->n_log_extra_fields; i++) {
5472 sizes[i] = htole64(c->log_extra_fields[i].iov_len);
5473
5474 iovec[i*2] = IOVEC_MAKE(sizes + i, sizeof(le64_t));
5475 iovec[i*2+1] = c->log_extra_fields[i];
5476 }
5477
5478 p = strjoina("/run/systemd/units/log-extra-fields:", u->id);
5479 pattern = strjoina(p, ".XXXXXX");
5480
5481 fd = mkostemp_safe(pattern);
5482 if (fd < 0)
5483 return log_unit_debug_errno(u, fd, "Failed to create extra fields file %s: %m", p);
5484
5485 n = writev(fd, iovec, c->n_log_extra_fields*2);
5486 if (n < 0) {
5487 r = log_unit_debug_errno(u, errno, "Failed to write extra fields: %m");
5488 goto fail;
5489 }
5490
5491 (void) fchmod(fd, 0644);
5492
5493 if (rename(pattern, p) < 0) {
5494 r = log_unit_debug_errno(u, errno, "Failed to rename extra fields file: %m");
5495 goto fail;
5496 }
5497
5498 u->exported_log_extra_fields = true;
5499 return 0;
5500
5501 fail:
5502 (void) unlink(pattern);
5503 return r;
5504 }
5505
5506 static int unit_export_log_ratelimit_interval(Unit *u, const ExecContext *c) {
5507 _cleanup_free_ char *buf = NULL;
5508 const char *p;
5509 int r;
5510
5511 assert(u);
5512 assert(c);
5513
5514 if (u->exported_log_ratelimit_interval)
5515 return 0;
5516
5517 if (c->log_ratelimit_interval_usec == 0)
5518 return 0;
5519
5520 p = strjoina("/run/systemd/units/log-rate-limit-interval:", u->id);
5521
5522 if (asprintf(&buf, "%" PRIu64, c->log_ratelimit_interval_usec) < 0)
5523 return log_oom();
5524
5525 r = symlink_atomic(buf, p);
5526 if (r < 0)
5527 return log_unit_debug_errno(u, r, "Failed to create log rate limit interval symlink %s: %m", p);
5528
5529 u->exported_log_ratelimit_interval = true;
5530 return 0;
5531 }
5532
5533 static int unit_export_log_ratelimit_burst(Unit *u, const ExecContext *c) {
5534 _cleanup_free_ char *buf = NULL;
5535 const char *p;
5536 int r;
5537
5538 assert(u);
5539 assert(c);
5540
5541 if (u->exported_log_ratelimit_burst)
5542 return 0;
5543
5544 if (c->log_ratelimit_burst == 0)
5545 return 0;
5546
5547 p = strjoina("/run/systemd/units/log-rate-limit-burst:", u->id);
5548
5549 if (asprintf(&buf, "%u", c->log_ratelimit_burst) < 0)
5550 return log_oom();
5551
5552 r = symlink_atomic(buf, p);
5553 if (r < 0)
5554 return log_unit_debug_errno(u, r, "Failed to create log rate limit burst symlink %s: %m", p);
5555
5556 u->exported_log_ratelimit_burst = true;
5557 return 0;
5558 }
5559
5560 void unit_export_state_files(Unit *u) {
5561 const ExecContext *c;
5562
5563 assert(u);
5564
5565 if (!u->id)
5566 return;
5567
5568 if (MANAGER_IS_TEST_RUN(u->manager))
5569 return;
5570
5571 /* Exports a couple of unit properties to /run/systemd/units/, so that journald can quickly query this data
5572 * from there. Ideally, journald would use IPC to query this, like everybody else, but that's hard, as long as
5573 * the IPC system itself and PID 1 also log to the journal.
5574 *
5575 * Note that these files really shouldn't be considered API for anyone else, as use a runtime file system as
5576 * IPC replacement is not compatible with today's world of file system namespaces. However, this doesn't really
5577 * apply to communication between the journal and systemd, as we assume that these two daemons live in the same
5578 * namespace at least.
5579 *
5580 * Note that some of the "files" exported here are actually symlinks and not regular files. Symlinks work
5581 * better for storing small bits of data, in particular as we can write them with two system calls, and read
5582 * them with one. */
5583
5584 (void) unit_export_invocation_id(u);
5585
5586 if (!MANAGER_IS_SYSTEM(u->manager))
5587 return;
5588
5589 c = unit_get_exec_context(u);
5590 if (c) {
5591 (void) unit_export_log_level_max(u, c);
5592 (void) unit_export_log_extra_fields(u, c);
5593 (void) unit_export_log_ratelimit_interval(u, c);
5594 (void) unit_export_log_ratelimit_burst(u, c);
5595 }
5596 }
5597
5598 void unit_unlink_state_files(Unit *u) {
5599 const char *p;
5600
5601 assert(u);
5602
5603 if (!u->id)
5604 return;
5605
5606 /* Undoes the effect of unit_export_state() */
5607
5608 if (u->exported_invocation_id) {
5609 _cleanup_free_ char *invocation_path = NULL;
5610 int r = unit_get_invocation_path(u, &invocation_path);
5611 if (r >= 0) {
5612 (void) unlink(invocation_path);
5613 u->exported_invocation_id = false;
5614 }
5615 }
5616
5617 if (!MANAGER_IS_SYSTEM(u->manager))
5618 return;
5619
5620 if (u->exported_log_level_max) {
5621 p = strjoina("/run/systemd/units/log-level-max:", u->id);
5622 (void) unlink(p);
5623
5624 u->exported_log_level_max = false;
5625 }
5626
5627 if (u->exported_log_extra_fields) {
5628 p = strjoina("/run/systemd/units/extra-fields:", u->id);
5629 (void) unlink(p);
5630
5631 u->exported_log_extra_fields = false;
5632 }
5633
5634 if (u->exported_log_ratelimit_interval) {
5635 p = strjoina("/run/systemd/units/log-rate-limit-interval:", u->id);
5636 (void) unlink(p);
5637
5638 u->exported_log_ratelimit_interval = false;
5639 }
5640
5641 if (u->exported_log_ratelimit_burst) {
5642 p = strjoina("/run/systemd/units/log-rate-limit-burst:", u->id);
5643 (void) unlink(p);
5644
5645 u->exported_log_ratelimit_burst = false;
5646 }
5647 }
5648
5649 int unit_prepare_exec(Unit *u) {
5650 int r;
5651
5652 assert(u);
5653
5654 /* Load any custom firewall BPF programs here once to test if they are existing and actually loadable.
5655 * Fail here early since later errors in the call chain unit_realize_cgroup to cgroup_context_apply are ignored. */
5656 r = bpf_firewall_load_custom(u);
5657 if (r < 0)
5658 return r;
5659
5660 /* Prepares everything so that we can fork of a process for this unit */
5661
5662 (void) unit_realize_cgroup(u);
5663
5664 if (u->reset_accounting) {
5665 (void) unit_reset_accounting(u);
5666 u->reset_accounting = false;
5667 }
5668
5669 unit_export_state_files(u);
5670
5671 r = unit_setup_exec_runtime(u);
5672 if (r < 0)
5673 return r;
5674
5675 r = unit_setup_dynamic_creds(u);
5676 if (r < 0)
5677 return r;
5678
5679 return 0;
5680 }
5681
5682 static int log_leftover(pid_t pid, int sig, void *userdata) {
5683 _cleanup_free_ char *comm = NULL;
5684
5685 (void) get_process_comm(pid, &comm);
5686
5687 if (comm && comm[0] == '(') /* Most likely our own helper process (PAM?), ignore */
5688 return 0;
5689
5690 log_unit_warning(userdata,
5691 "Found left-over process " PID_FMT " (%s) in control group while starting unit. Ignoring.\n"
5692 "This usually indicates unclean termination of a previous run, or service implementation deficiencies.",
5693 pid, strna(comm));
5694
5695 return 1;
5696 }
5697
5698 int unit_warn_leftover_processes(Unit *u) {
5699 assert(u);
5700
5701 (void) unit_pick_cgroup_path(u);
5702
5703 if (!u->cgroup_path)
5704 return 0;
5705
5706 return cg_kill_recursive(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, 0, 0, NULL, log_leftover, u);
5707 }
5708
5709 bool unit_needs_console(Unit *u) {
5710 ExecContext *ec;
5711 UnitActiveState state;
5712
5713 assert(u);
5714
5715 state = unit_active_state(u);
5716
5717 if (UNIT_IS_INACTIVE_OR_FAILED(state))
5718 return false;
5719
5720 if (UNIT_VTABLE(u)->needs_console)
5721 return UNIT_VTABLE(u)->needs_console(u);
5722
5723 /* If this unit type doesn't implement this call, let's use a generic fallback implementation: */
5724 ec = unit_get_exec_context(u);
5725 if (!ec)
5726 return false;
5727
5728 return exec_context_may_touch_console(ec);
5729 }
5730
5731 const char *unit_label_path(Unit *u) {
5732 const char *p;
5733
5734 /* Returns the file system path to use for MAC access decisions, i.e. the file to read the SELinux label off
5735 * when validating access checks. */
5736
5737 p = u->source_path ?: u->fragment_path;
5738 if (!p)
5739 return NULL;
5740
5741 /* If a unit is masked, then don't read the SELinux label of /dev/null, as that really makes no sense */
5742 if (path_equal(p, "/dev/null"))
5743 return NULL;
5744
5745 return p;
5746 }
5747
5748 int unit_pid_attachable(Unit *u, pid_t pid, sd_bus_error *error) {
5749 int r;
5750
5751 assert(u);
5752
5753 /* Checks whether the specified PID is generally good for attaching, i.e. a valid PID, not our manager itself,
5754 * and not a kernel thread either */
5755
5756 /* First, a simple range check */
5757 if (!pid_is_valid(pid))
5758 return sd_bus_error_setf(error, SD_BUS_ERROR_INVALID_ARGS, "Process identifier " PID_FMT " is not valid.", pid);
5759
5760 /* Some extra safety check */
5761 if (pid == 1 || pid == getpid_cached())
5762 return sd_bus_error_setf(error, SD_BUS_ERROR_INVALID_ARGS, "Process " PID_FMT " is a manager process, refusing.", pid);
5763
5764 /* Don't even begin to bother with kernel threads */
5765 r = is_kernel_thread(pid);
5766 if (r == -ESRCH)
5767 return sd_bus_error_setf(error, SD_BUS_ERROR_UNIX_PROCESS_ID_UNKNOWN, "Process with ID " PID_FMT " does not exist.", pid);
5768 if (r < 0)
5769 return sd_bus_error_set_errnof(error, r, "Failed to determine whether process " PID_FMT " is a kernel thread: %m", pid);
5770 if (r > 0)
5771 return sd_bus_error_setf(error, SD_BUS_ERROR_INVALID_ARGS, "Process " PID_FMT " is a kernel thread, refusing.", pid);
5772
5773 return 0;
5774 }
5775
5776 void unit_log_success(Unit *u) {
5777 assert(u);
5778
5779 log_struct(LOG_INFO,
5780 "MESSAGE_ID=" SD_MESSAGE_UNIT_SUCCESS_STR,
5781 LOG_UNIT_ID(u),
5782 LOG_UNIT_INVOCATION_ID(u),
5783 LOG_UNIT_MESSAGE(u, "Succeeded."));
5784 }
5785
5786 void unit_log_failure(Unit *u, const char *result) {
5787 assert(u);
5788 assert(result);
5789
5790 log_struct(LOG_WARNING,
5791 "MESSAGE_ID=" SD_MESSAGE_UNIT_FAILURE_RESULT_STR,
5792 LOG_UNIT_ID(u),
5793 LOG_UNIT_INVOCATION_ID(u),
5794 LOG_UNIT_MESSAGE(u, "Failed with result '%s'.", result),
5795 "UNIT_RESULT=%s", result);
5796 }
5797
5798 void unit_log_skip(Unit *u, const char *result) {
5799 assert(u);
5800 assert(result);
5801
5802 log_struct(LOG_INFO,
5803 "MESSAGE_ID=" SD_MESSAGE_UNIT_SKIPPED_STR,
5804 LOG_UNIT_ID(u),
5805 LOG_UNIT_INVOCATION_ID(u),
5806 LOG_UNIT_MESSAGE(u, "Skipped due to '%s'.", result),
5807 "UNIT_RESULT=%s", result);
5808 }
5809
5810 void unit_log_process_exit(
5811 Unit *u,
5812 const char *kind,
5813 const char *command,
5814 bool success,
5815 int code,
5816 int status) {
5817
5818 int level;
5819
5820 assert(u);
5821 assert(kind);
5822
5823 /* If this is a successful exit, let's log about the exit code on DEBUG level. If this is a failure
5824 * and the process exited on its own via exit(), then let's make this a NOTICE, under the assumption
5825 * that the service already logged the reason at a higher log level on its own. Otherwise, make it a
5826 * WARNING. */
5827 if (success)
5828 level = LOG_DEBUG;
5829 else if (code == CLD_EXITED)
5830 level = LOG_NOTICE;
5831 else
5832 level = LOG_WARNING;
5833
5834 log_struct(level,
5835 "MESSAGE_ID=" SD_MESSAGE_UNIT_PROCESS_EXIT_STR,
5836 LOG_UNIT_MESSAGE(u, "%s exited, code=%s, status=%i/%s",
5837 kind,
5838 sigchld_code_to_string(code), status,
5839 strna(code == CLD_EXITED
5840 ? exit_status_to_string(status, EXIT_STATUS_FULL)
5841 : signal_to_string(status))),
5842 "EXIT_CODE=%s", sigchld_code_to_string(code),
5843 "EXIT_STATUS=%i", status,
5844 "COMMAND=%s", strna(command),
5845 LOG_UNIT_ID(u),
5846 LOG_UNIT_INVOCATION_ID(u));
5847 }
5848
5849 int unit_exit_status(Unit *u) {
5850 assert(u);
5851
5852 /* Returns the exit status to propagate for the most recent cycle of this unit. Returns a value in the range
5853 * 0…255 if there's something to propagate. EOPNOTSUPP if the concept does not apply to this unit type, ENODATA
5854 * if no data is currently known (for example because the unit hasn't deactivated yet) and EBADE if the main
5855 * service process has exited abnormally (signal/coredump). */
5856
5857 if (!UNIT_VTABLE(u)->exit_status)
5858 return -EOPNOTSUPP;
5859
5860 return UNIT_VTABLE(u)->exit_status(u);
5861 }
5862
5863 int unit_failure_action_exit_status(Unit *u) {
5864 int r;
5865
5866 assert(u);
5867
5868 /* Returns the exit status to propagate on failure, or an error if there's nothing to propagate */
5869
5870 if (u->failure_action_exit_status >= 0)
5871 return u->failure_action_exit_status;
5872
5873 r = unit_exit_status(u);
5874 if (r == -EBADE) /* Exited, but not cleanly (i.e. by signal or such) */
5875 return 255;
5876
5877 return r;
5878 }
5879
5880 int unit_success_action_exit_status(Unit *u) {
5881 int r;
5882
5883 assert(u);
5884
5885 /* Returns the exit status to propagate on success, or an error if there's nothing to propagate */
5886
5887 if (u->success_action_exit_status >= 0)
5888 return u->success_action_exit_status;
5889
5890 r = unit_exit_status(u);
5891 if (r == -EBADE) /* Exited, but not cleanly (i.e. by signal or such) */
5892 return 255;
5893
5894 return r;
5895 }
5896
5897 int unit_test_trigger_loaded(Unit *u) {
5898 Unit *trigger;
5899
5900 /* Tests whether the unit to trigger is loaded */
5901
5902 trigger = UNIT_TRIGGER(u);
5903 if (!trigger)
5904 return log_unit_error_errno(u, SYNTHETIC_ERRNO(ENOENT),
5905 "Refusing to start, no unit to trigger.");
5906 if (trigger->load_state != UNIT_LOADED)
5907 return log_unit_error_errno(u, SYNTHETIC_ERRNO(ENOENT),
5908 "Refusing to start, unit %s to trigger not loaded.", trigger->id);
5909
5910 return 0;
5911 }
5912
5913 void unit_destroy_runtime_directory(Unit *u, const ExecContext *context) {
5914 if (context->runtime_directory_preserve_mode == EXEC_PRESERVE_NO ||
5915 (context->runtime_directory_preserve_mode == EXEC_PRESERVE_RESTART && !unit_will_restart(u)))
5916 exec_context_destroy_runtime_directory(context, u->manager->prefix[EXEC_DIRECTORY_RUNTIME]);
5917 }
5918
5919 int unit_clean(Unit *u, ExecCleanMask mask) {
5920 UnitActiveState state;
5921
5922 assert(u);
5923
5924 /* Special return values:
5925 *
5926 * -EOPNOTSUPP → cleaning not supported for this unit type
5927 * -EUNATCH → cleaning not defined for this resource type
5928 * -EBUSY → unit currently can't be cleaned since it's running or not properly loaded, or has
5929 * a job queued or similar
5930 */
5931
5932 if (!UNIT_VTABLE(u)->clean)
5933 return -EOPNOTSUPP;
5934
5935 if (mask == 0)
5936 return -EUNATCH;
5937
5938 if (u->load_state != UNIT_LOADED)
5939 return -EBUSY;
5940
5941 if (u->job)
5942 return -EBUSY;
5943
5944 state = unit_active_state(u);
5945 if (!IN_SET(state, UNIT_INACTIVE))
5946 return -EBUSY;
5947
5948 return UNIT_VTABLE(u)->clean(u, mask);
5949 }
5950
5951 int unit_can_clean(Unit *u, ExecCleanMask *ret) {
5952 assert(u);
5953
5954 if (!UNIT_VTABLE(u)->clean ||
5955 u->load_state != UNIT_LOADED) {
5956 *ret = 0;
5957 return 0;
5958 }
5959
5960 /* When the clean() method is set, can_clean() really should be set too */
5961 assert(UNIT_VTABLE(u)->can_clean);
5962
5963 return UNIT_VTABLE(u)->can_clean(u, ret);
5964 }
5965
5966 static const char* const collect_mode_table[_COLLECT_MODE_MAX] = {
5967 [COLLECT_INACTIVE] = "inactive",
5968 [COLLECT_INACTIVE_OR_FAILED] = "inactive-or-failed",
5969 };
5970
5971 DEFINE_STRING_TABLE_LOOKUP(collect_mode, CollectMode);