1 /* SPDX-License-Identifier: LGPL-2.1-or-later */
7 #include "cgroup-setup.h"
8 #include "dbus-scope.h"
10 #include "exit-status.h"
13 #include "parse-util.h"
15 #include "random-util.h"
17 #include "serialize.h"
20 #include "string-table.h"
21 #include "string-util.h"
24 #include "user-util.h"
26 static const UnitActiveState state_translation_table
[_SCOPE_STATE_MAX
] = {
27 [SCOPE_DEAD
] = UNIT_INACTIVE
,
28 [SCOPE_START_CHOWN
] = UNIT_ACTIVATING
,
29 [SCOPE_RUNNING
] = UNIT_ACTIVE
,
30 [SCOPE_ABANDONED
] = UNIT_ACTIVE
,
31 [SCOPE_STOP_SIGTERM
] = UNIT_DEACTIVATING
,
32 [SCOPE_STOP_SIGKILL
] = UNIT_DEACTIVATING
,
33 [SCOPE_FAILED
] = UNIT_FAILED
,
36 static int scope_dispatch_timer(sd_event_source
*source
, usec_t usec
, void *userdata
);
38 static void scope_init(Unit
*u
) {
39 Scope
*s
= ASSERT_PTR(SCOPE(u
));
41 assert(u
->load_state
== UNIT_STUB
);
43 s
->runtime_max_usec
= USEC_INFINITY
;
44 s
->timeout_stop_usec
= u
->manager
->defaults
.timeout_stop_usec
;
45 u
->ignore_on_isolate
= true;
46 s
->user
= s
->group
= NULL
;
47 s
->oom_policy
= _OOM_POLICY_INVALID
;
50 static void scope_done(Unit
*u
) {
51 Scope
*s
= ASSERT_PTR(SCOPE(u
));
53 s
->controller
= mfree(s
->controller
);
54 s
->controller_track
= sd_bus_track_unref(s
->controller_track
);
56 s
->timer_event_source
= sd_event_source_disable_unref(s
->timer_event_source
);
58 s
->user
= mfree(s
->user
);
59 s
->group
= mfree(s
->group
);
62 static usec_t
scope_running_timeout(Scope
*s
) {
67 if (s
->runtime_rand_extra_usec
!= 0) {
68 delta
= random_u64_range(s
->runtime_rand_extra_usec
);
69 log_unit_debug(UNIT(s
), "Adding delta of %s sec to timeout", FORMAT_TIMESPAN(delta
, USEC_PER_SEC
));
72 return usec_add(usec_add(UNIT(s
)->active_enter_timestamp
.monotonic
,
77 static int scope_arm_timer(Scope
*s
, bool relative
, usec_t usec
) {
80 return unit_arm_timer(UNIT(s
), &s
->timer_event_source
, relative
, usec
, scope_dispatch_timer
);
83 static void scope_set_state(Scope
*s
, ScopeState state
) {
88 if (s
->state
!= state
)
89 bus_unit_send_pending_change_signal(UNIT(s
), false);
94 if (!IN_SET(state
, SCOPE_STOP_SIGTERM
, SCOPE_STOP_SIGKILL
, SCOPE_START_CHOWN
, SCOPE_RUNNING
))
95 s
->timer_event_source
= sd_event_source_disable_unref(s
->timer_event_source
);
97 if (!IN_SET(old_state
, SCOPE_DEAD
, SCOPE_FAILED
) && IN_SET(state
, SCOPE_DEAD
, SCOPE_FAILED
))
98 unit_unwatch_all_pids(UNIT(s
));
100 if (state
!= old_state
)
101 log_unit_debug(UNIT(s
), "Changed %s -> %s",
102 scope_state_to_string(old_state
), scope_state_to_string(state
));
104 unit_notify(UNIT(s
), state_translation_table
[old_state
], state_translation_table
[state
], /* reload_success = */ true);
107 static int scope_add_default_dependencies(Scope
*s
) {
112 if (!UNIT(s
)->default_dependencies
)
115 /* Make sure scopes are unloaded on shutdown */
116 r
= unit_add_two_dependencies_by_name(
118 UNIT_BEFORE
, UNIT_CONFLICTS
,
119 SPECIAL_SHUTDOWN_TARGET
, true,
120 UNIT_DEPENDENCY_DEFAULT
);
127 static int scope_verify(Scope
*s
) {
129 assert(UNIT(s
)->load_state
== UNIT_LOADED
);
131 if (set_isempty(UNIT(s
)->pids
) &&
132 !MANAGER_IS_RELOADING(UNIT(s
)->manager
) &&
133 !unit_has_name(UNIT(s
), SPECIAL_INIT_SCOPE
))
134 return log_unit_error_errno(UNIT(s
), SYNTHETIC_ERRNO(ENOENT
), "Scope has no PIDs. Refusing.");
139 static int scope_load_init_scope(Unit
*u
) {
142 if (!unit_has_name(u
, SPECIAL_INIT_SCOPE
))
148 /* init.scope is a bit special, as it has to stick around forever. Because of its special semantics we
149 * synthesize it here, instead of relying on the unit file on disk. */
151 u
->default_dependencies
= false;
153 /* Prettify things, if we can. */
155 u
->description
= strdup("System and Service Manager");
156 if (!u
->documentation
)
157 (void) strv_extend(&u
->documentation
, "man:systemd(1)");
162 static int scope_add_extras(Scope
*s
) {
165 r
= unit_patch_contexts(UNIT(s
));
169 r
= unit_set_default_slice(UNIT(s
));
173 if (s
->oom_policy
< 0)
174 s
->oom_policy
= s
->cgroup_context
.delegate
? OOM_CONTINUE
: UNIT(s
)->manager
->defaults
.oom_policy
;
176 s
->cgroup_context
.memory_oom_group
= s
->oom_policy
== OOM_KILL
;
178 return scope_add_default_dependencies(s
);
181 static int scope_load(Unit
*u
) {
182 Scope
*s
= ASSERT_PTR(SCOPE(u
));
185 assert(u
->load_state
== UNIT_STUB
);
187 if (!u
->transient
&& !MANAGER_IS_RELOADING(u
->manager
))
188 /* Refuse to load non-transient scope units, but allow them while reloading. */
191 r
= scope_load_init_scope(u
);
195 r
= unit_load_fragment_and_dropin(u
, false);
199 if (u
->load_state
!= UNIT_LOADED
)
202 r
= scope_add_extras(s
);
206 return scope_verify(s
);
209 static usec_t
scope_coldplug_timeout(Scope
*s
) {
212 switch (s
->deserialized_state
) {
215 return scope_running_timeout(s
);
217 case SCOPE_STOP_SIGKILL
:
218 case SCOPE_STOP_SIGTERM
:
219 return usec_add(UNIT(s
)->state_change_timestamp
.monotonic
, s
->timeout_stop_usec
);
222 return USEC_INFINITY
;
226 static int scope_coldplug(Unit
*u
) {
227 Scope
*s
= ASSERT_PTR(SCOPE(u
));
230 assert(s
->state
== SCOPE_DEAD
);
232 if (s
->deserialized_state
== s
->state
)
235 r
= scope_arm_timer(s
, /* relative= */ false, scope_coldplug_timeout(s
));
239 bus_scope_track_controller(s
);
241 scope_set_state(s
, s
->deserialized_state
);
245 static void scope_dump(Unit
*u
, FILE *f
, const char *prefix
) {
246 Scope
*s
= ASSERT_PTR(SCOPE(u
));
252 "%sScope State: %s\n"
254 "%sRuntimeMaxSec: %s\n"
255 "%sRuntimeRandomizedExtraSec: %s\n"
257 prefix
, scope_state_to_string(s
->state
),
258 prefix
, scope_result_to_string(s
->result
),
259 prefix
, FORMAT_TIMESPAN(s
->runtime_max_usec
, USEC_PER_SEC
),
260 prefix
, FORMAT_TIMESPAN(s
->runtime_rand_extra_usec
, USEC_PER_SEC
),
261 prefix
, oom_policy_to_string(s
->oom_policy
));
263 cgroup_context_dump(u
, f
, prefix
);
264 kill_context_dump(&s
->kill_context
, f
, prefix
);
267 static void scope_enter_dead(Scope
*s
, ScopeResult f
) {
270 if (s
->result
== SCOPE_SUCCESS
)
273 unit_log_result(UNIT(s
), s
->result
== SCOPE_SUCCESS
, scope_result_to_string(s
->result
));
274 scope_set_state(s
, s
->result
!= SCOPE_SUCCESS
? SCOPE_FAILED
: SCOPE_DEAD
);
277 static void scope_enter_signal(Scope
*s
, ScopeState state
, ScopeResult f
) {
278 bool skip_signal
= false;
283 if (s
->result
== SCOPE_SUCCESS
)
286 /* If we have a controller set let's ask the controller nicely to terminate the scope, instead of us going
287 * directly into SIGTERM berserk mode */
288 if (state
== SCOPE_STOP_SIGTERM
)
289 skip_signal
= bus_scope_send_request_stop(s
) > 0;
294 r
= unit_kill_context(
296 state
!= SCOPE_STOP_SIGTERM
? KILL_KILL
:
297 s
->was_abandoned
? KILL_TERMINATE_AND_LOG
:
300 log_unit_warning_errno(UNIT(s
), r
, "Failed to kill processes: %m");
306 r
= scope_arm_timer(s
, /* relative= */ true, s
->timeout_stop_usec
);
308 log_unit_warning_errno(UNIT(s
), r
, "Failed to install timer: %m");
312 scope_set_state(s
, state
);
313 } else if (state
== SCOPE_STOP_SIGTERM
)
314 scope_enter_signal(s
, SCOPE_STOP_SIGKILL
, SCOPE_SUCCESS
);
316 scope_enter_dead(s
, SCOPE_SUCCESS
);
321 scope_enter_dead(s
, SCOPE_FAILURE_RESOURCES
);
324 static int scope_enter_start_chown(Scope
*s
) {
325 Unit
*u
= UNIT(ASSERT_PTR(s
));
326 _cleanup_(pidref_done
) PidRef pidref
= PIDREF_NULL
;
331 if (!s
->cgroup_runtime
)
334 r
= scope_arm_timer(s
, /* relative= */ true, u
->manager
->defaults
.timeout_start_usec
);
338 r
= unit_fork_helper_process(u
, "(sd-chown-cgroup)", /* into_cgroup= */ true, &pidref
);
343 uid_t uid
= UID_INVALID
;
344 gid_t gid
= GID_INVALID
;
346 if (!isempty(s
->user
)) {
347 const char *user
= s
->user
;
349 r
= get_user_creds(&user
, &uid
, &gid
, NULL
, NULL
, 0);
351 log_unit_error_errno(UNIT(s
), r
, "Failed to resolve user \"%s\": %m", user
);
356 if (!isempty(s
->group
)) {
357 const char *group
= s
->group
;
359 r
= get_group_creds(&group
, &gid
, 0);
361 log_unit_error_errno(UNIT(s
), r
, "Failed to resolve group \"%s\": %m", group
);
366 r
= cg_set_access(s
->cgroup_runtime
->cgroup_path
, uid
, gid
);
368 log_unit_error_errno(UNIT(s
), r
, "Failed to adjust control group access: %m");
375 r
= unit_watch_pidref(UNIT(s
), &pidref
, /* exclusive= */ true);
379 scope_set_state(s
, SCOPE_START_CHOWN
);
383 s
->timer_event_source
= sd_event_source_disable_unref(s
->timer_event_source
);
387 static int scope_enter_running(Scope
*s
) {
388 Unit
*u
= UNIT(ASSERT_PTR(s
));
391 (void) bus_scope_track_controller(s
);
393 r
= unit_acquire_invocation_id(u
);
397 unit_export_state_files(u
);
399 r
= unit_attach_pids_to_cgroup(u
, u
->pids
, NULL
);
401 log_unit_warning_errno(u
, r
, "Failed to add PIDs to scope's control group: %m");
405 r
= log_unit_warning_errno(u
, SYNTHETIC_ERRNO(ECHILD
), "No PIDs left to attach to the scope's control group, refusing.");
408 log_unit_debug(u
, "%i %s added to scope's control group.", r
, r
== 1 ? "process" : "processes");
410 s
->result
= SCOPE_SUCCESS
;
412 scope_set_state(s
, SCOPE_RUNNING
);
414 /* Set the maximum runtime timeout. */
415 scope_arm_timer(s
, /* relative= */ false, scope_running_timeout(s
));
417 /* Unwatch all pids we've just added to cgroup. We rely on empty notifications there. */
418 unit_unwatch_all_pids(u
);
423 scope_enter_dead(s
, SCOPE_FAILURE_RESOURCES
);
427 static int scope_start(Unit
*u
) {
428 Scope
*s
= ASSERT_PTR(SCOPE(u
));
430 if (unit_has_name(u
, SPECIAL_INIT_SCOPE
))
433 if (s
->state
== SCOPE_FAILED
)
436 /* We can't fulfill this right now, please try again later */
437 if (IN_SET(s
->state
, SCOPE_STOP_SIGTERM
, SCOPE_STOP_SIGKILL
))
440 assert(s
->state
== SCOPE_DEAD
);
442 if (!u
->transient
&& !MANAGER_IS_RELOADING(u
->manager
))
445 (void) unit_realize_cgroup(u
);
446 (void) unit_reset_accounting(u
);
448 /* We check only for User= option to keep behavior consistent with logic for service units,
449 * i.e. having 'Delegate=true Group=foo' w/o specifying User= has no effect. */
450 if (s
->user
&& unit_cgroup_delegate(u
))
451 return scope_enter_start_chown(s
);
453 return scope_enter_running(s
);
456 static int scope_stop(Unit
*u
) {
457 Scope
*s
= ASSERT_PTR(SCOPE(u
));
459 if (IN_SET(s
->state
, SCOPE_STOP_SIGTERM
, SCOPE_STOP_SIGKILL
))
462 assert(IN_SET(s
->state
, SCOPE_RUNNING
, SCOPE_ABANDONED
));
464 scope_enter_signal(s
, SCOPE_STOP_SIGTERM
, SCOPE_SUCCESS
);
468 static void scope_reset_failed(Unit
*u
) {
469 Scope
*s
= ASSERT_PTR(SCOPE(u
));
471 if (s
->state
== SCOPE_FAILED
)
472 scope_set_state(s
, SCOPE_DEAD
);
474 s
->result
= SCOPE_SUCCESS
;
477 static int scope_get_timeout(Unit
*u
, usec_t
*timeout
) {
478 Scope
*s
= ASSERT_PTR(SCOPE(u
));
482 if (!s
->timer_event_source
)
485 r
= sd_event_source_get_time(s
->timer_event_source
, &t
);
488 if (t
== USEC_INFINITY
)
495 static int scope_serialize(Unit
*u
, FILE *f
, FDSet
*fds
) {
496 Scope
*s
= ASSERT_PTR(SCOPE(u
));
502 (void) serialize_item(f
, "state", scope_state_to_string(s
->state
));
503 (void) serialize_bool(f
, "was-abandoned", s
->was_abandoned
);
505 (void) serialize_item(f
, "controller", s
->controller
);
507 SET_FOREACH(pid
, u
->pids
)
508 serialize_pidref(f
, fds
, "pids", pid
);
513 static int scope_deserialize_item(Unit
*u
, const char *key
, const char *value
, FDSet
*fds
) {
514 Scope
*s
= ASSERT_PTR(SCOPE(u
));
521 if (streq(key
, "state")) {
524 state
= scope_state_from_string(value
);
526 log_unit_debug(u
, "Failed to parse state value: %s", value
);
528 s
->deserialized_state
= state
;
530 } else if (streq(key
, "was-abandoned")) {
533 k
= parse_boolean(value
);
535 log_unit_debug(u
, "Failed to parse boolean value: %s", value
);
537 s
->was_abandoned
= k
;
538 } else if (streq(key
, "controller")) {
540 r
= free_and_strdup(&s
->controller
, value
);
544 } else if (streq(key
, "pids")) {
545 _cleanup_(pidref_done
) PidRef pidref
= PIDREF_NULL
;
547 /* We don't check if we already received the pid before here because unit_watch_pidref()
548 * does this check internally and discards the new pidref if we already received it before. */
549 if (deserialize_pidref(fds
, value
, &pidref
) >= 0) {
550 r
= unit_watch_pidref(u
, &pidref
, /* exclusive= */ false);
552 log_unit_debug(u
, "Failed to watch PID, ignoring: %s", value
);
555 log_unit_debug(u
, "Unknown serialization key: %s", key
);
560 static void scope_notify_cgroup_empty_event(Unit
*u
) {
561 Scope
*s
= ASSERT_PTR(SCOPE(u
));
563 log_unit_debug(u
, "cgroup is empty");
565 if (IN_SET(s
->state
, SCOPE_RUNNING
, SCOPE_ABANDONED
, SCOPE_STOP_SIGTERM
, SCOPE_STOP_SIGKILL
))
566 scope_enter_dead(s
, SCOPE_SUCCESS
);
569 static void scope_notify_cgroup_oom_event(Unit
*u
, bool managed_oom
) {
570 Scope
*s
= ASSERT_PTR(SCOPE(u
));
573 log_unit_debug(u
, "Process(es) of control group were killed by systemd-oomd.");
575 log_unit_debug(u
, "Process of control group was killed by the OOM killer.");
577 if (s
->oom_policy
== OOM_CONTINUE
)
582 case SCOPE_START_CHOWN
:
584 scope_enter_signal(s
, SCOPE_STOP_SIGTERM
, SCOPE_FAILURE_OOM_KILL
);
587 case SCOPE_STOP_SIGTERM
:
588 scope_enter_signal(s
, SCOPE_STOP_SIGKILL
, SCOPE_FAILURE_OOM_KILL
);
591 case SCOPE_STOP_SIGKILL
:
592 if (s
->result
== SCOPE_SUCCESS
)
593 s
->result
= SCOPE_FAILURE_OOM_KILL
;
595 /* SCOPE_DEAD, SCOPE_ABANDONED, and SCOPE_FAILED end up in default */
601 static void scope_sigchld_event(Unit
*u
, pid_t pid
, int code
, int status
) {
602 Scope
*s
= ASSERT_PTR(SCOPE(u
));
604 if (s
->state
== SCOPE_START_CHOWN
) {
605 if (!is_clean_exit(code
, status
, EXIT_CLEAN_COMMAND
, NULL
))
606 scope_enter_dead(s
, SCOPE_FAILURE_RESOURCES
);
608 scope_enter_running(s
);
613 static int scope_dispatch_timer(sd_event_source
*source
, usec_t usec
, void *userdata
) {
614 Scope
*s
= ASSERT_PTR(SCOPE(userdata
));
616 assert(s
->timer_event_source
== source
);
621 log_unit_warning(UNIT(s
), "Scope reached runtime time limit. Stopping.");
622 scope_enter_signal(s
, SCOPE_STOP_SIGTERM
, SCOPE_FAILURE_TIMEOUT
);
625 case SCOPE_STOP_SIGTERM
:
626 if (s
->kill_context
.send_sigkill
) {
627 log_unit_warning(UNIT(s
), "Stopping timed out. Killing.");
628 scope_enter_signal(s
, SCOPE_STOP_SIGKILL
, SCOPE_FAILURE_TIMEOUT
);
630 log_unit_warning(UNIT(s
), "Stopping timed out. Skipping SIGKILL.");
631 scope_enter_dead(s
, SCOPE_FAILURE_TIMEOUT
);
636 case SCOPE_STOP_SIGKILL
:
637 log_unit_warning(UNIT(s
), "Still around after SIGKILL. Ignoring.");
638 scope_enter_dead(s
, SCOPE_FAILURE_TIMEOUT
);
641 case SCOPE_START_CHOWN
:
642 log_unit_warning(UNIT(s
), "User lookup timed out. Entering failed state.");
643 scope_enter_dead(s
, SCOPE_FAILURE_TIMEOUT
);
647 assert_not_reached();
653 int scope_abandon(Scope
*s
) {
656 if (unit_has_name(UNIT(s
), SPECIAL_INIT_SCOPE
))
659 if (!IN_SET(s
->state
, SCOPE_RUNNING
, SCOPE_ABANDONED
))
662 s
->was_abandoned
= true;
664 s
->controller
= mfree(s
->controller
);
665 s
->controller_track
= sd_bus_track_unref(s
->controller_track
);
667 scope_set_state(s
, SCOPE_ABANDONED
);
672 static UnitActiveState
scope_active_state(Unit
*u
) {
673 Scope
*s
= ASSERT_PTR(SCOPE(u
));
675 return state_translation_table
[s
->state
];
678 static const char *scope_sub_state_to_string(Unit
*u
) {
679 Scope
*s
= ASSERT_PTR(SCOPE(u
));
681 return scope_state_to_string(s
->state
);
684 static void scope_enumerate_perpetual(Manager
*m
) {
690 /* Let's unconditionally add the "init.scope" special unit
691 * that encapsulates PID 1. Note that PID 1 already is in the
692 * cgroup for this, we hence just need to allocate the object
693 * for it and that's it. */
695 u
= manager_get_unit(m
, SPECIAL_INIT_SCOPE
);
697 r
= unit_new_for_name(m
, sizeof(Scope
), SPECIAL_INIT_SCOPE
, &u
);
699 return (void) log_error_errno(r
, "Failed to allocate the special %s unit: %m",
705 SCOPE(u
)->deserialized_state
= SCOPE_RUNNING
;
707 unit_add_to_load_queue(u
);
708 unit_add_to_dbus_queue(u
);
709 /* Enqueue an explicit cgroup realization here. Unlike other cgroups this one already exists and is
710 * populated (by us, after all!) already, even when we are not in a reload cycle. Hence we cannot
711 * apply the settings at creation time anymore, but let's at least apply them asynchronously. */
712 unit_add_to_cgroup_realize_queue(u
);
715 static const char* const scope_result_table
[_SCOPE_RESULT_MAX
] = {
716 [SCOPE_SUCCESS
] = "success",
717 [SCOPE_FAILURE_RESOURCES
] = "resources",
718 [SCOPE_FAILURE_TIMEOUT
] = "timeout",
719 [SCOPE_FAILURE_OOM_KILL
] = "oom-kill",
722 DEFINE_STRING_TABLE_LOOKUP(scope_result
, ScopeResult
);
724 const UnitVTable scope_vtable
= {
725 .object_size
= sizeof(Scope
),
726 .cgroup_context_offset
= offsetof(Scope
, cgroup_context
),
727 .kill_context_offset
= offsetof(Scope
, kill_context
),
728 .cgroup_runtime_offset
= offsetof(Scope
, cgroup_runtime
),
734 .private_section
= "Scope",
736 .can_transient
= true,
737 .can_delegate
= true,
740 .can_set_managed_oom
= true,
746 .coldplug
= scope_coldplug
,
750 .start
= scope_start
,
753 .freezer_action
= unit_cgroup_freezer_action
,
755 .get_timeout
= scope_get_timeout
,
757 .serialize
= scope_serialize
,
758 .deserialize_item
= scope_deserialize_item
,
760 .active_state
= scope_active_state
,
761 .sub_state_to_string
= scope_sub_state_to_string
,
763 .sigchld_event
= scope_sigchld_event
,
765 .reset_failed
= scope_reset_failed
,
767 .notify_cgroup_empty
= scope_notify_cgroup_empty_event
,
768 .notify_cgroup_oom
= scope_notify_cgroup_oom_event
,
770 .bus_set_property
= bus_scope_set_property
,
771 .bus_commit_properties
= bus_scope_commit_properties
,
773 .enumerate_perpetual
= scope_enumerate_perpetual
,