1 /* SPDX-License-Identifier: LGPL-2.1+ */
11 #include "cgroup-util.h"
14 #include "ip-address-access.h"
16 #include "ratelimit.h"
18 struct libmnt_monitor
;
19 typedef struct Unit Unit
;
21 /* Enforce upper limit how many names we allow */
22 #define MANAGER_MAX_NAMES 131072 /* 128K */
24 typedef struct Manager Manager
;
26 /* An externally visible state. We don't actually maintain this as state variable, but derive it from various fields
28 typedef enum ManagerState
{
36 _MANAGER_STATE_INVALID
= -1
39 typedef enum ManagerObjective
{
49 _MANAGER_OBJECTIVE_MAX
,
50 _MANAGER_OBJECTIVE_INVALID
= -1
53 typedef enum StatusType
{
54 STATUS_TYPE_EPHEMERAL
,
56 STATUS_TYPE_EMERGENCY
,
59 typedef enum OOMPolicy
{
60 OOM_CONTINUE
, /* The kernel kills the process it wants to kill, and that's it */
61 OOM_STOP
, /* The kernel kills the process it wants to kill, and we stop the unit */
62 OOM_KILL
, /* The kernel kills the process it wants to kill, and all others in the unit, and we stop the unit */
64 _OOM_POLICY_INVALID
= -1
68 * 1. TIMESTAMP_FIRMWARE, TIMESTAMP_LOADER, TIMESTAMP_KERNEL, TIMESTAMP_INITRD,
69 * TIMESTAMP_SECURITY_START, and TIMESTAMP_SECURITY_FINISH are set only when
70 * the manager is system and not running under container environment.
72 * 2. The monotonic timestamp of TIMESTAMP_KERNEL is always zero.
74 * 3. The realtime timestamp of TIMESTAMP_KERNEL will be unset if the system does not
77 * 4. TIMESTAMP_FIRMWARE and TIMESTAMP_LOADER will be unset if the system does not
78 * have RTC, or systemd is built without EFI support.
80 * 5. The monotonic timestamps of TIMESTAMP_FIRMWARE and TIMESTAMP_LOADER are stored as
81 * negative of the actual value.
83 * 6. TIMESTAMP_USERSPACE is the timestamp of when the manager was started.
85 * 7. TIMESTAMP_INITRD_* are set only when the system is booted with an initrd.
88 typedef enum ManagerTimestamp
{
89 MANAGER_TIMESTAMP_FIRMWARE
,
90 MANAGER_TIMESTAMP_LOADER
,
91 MANAGER_TIMESTAMP_KERNEL
,
92 MANAGER_TIMESTAMP_INITRD
,
93 MANAGER_TIMESTAMP_USERSPACE
,
94 MANAGER_TIMESTAMP_FINISH
,
96 MANAGER_TIMESTAMP_SECURITY_START
,
97 MANAGER_TIMESTAMP_SECURITY_FINISH
,
98 MANAGER_TIMESTAMP_GENERATORS_START
,
99 MANAGER_TIMESTAMP_GENERATORS_FINISH
,
100 MANAGER_TIMESTAMP_UNITS_LOAD_START
,
101 MANAGER_TIMESTAMP_UNITS_LOAD_FINISH
,
103 MANAGER_TIMESTAMP_INITRD_SECURITY_START
,
104 MANAGER_TIMESTAMP_INITRD_SECURITY_FINISH
,
105 MANAGER_TIMESTAMP_INITRD_GENERATORS_START
,
106 MANAGER_TIMESTAMP_INITRD_GENERATORS_FINISH
,
107 MANAGER_TIMESTAMP_INITRD_UNITS_LOAD_START
,
108 MANAGER_TIMESTAMP_INITRD_UNITS_LOAD_FINISH
,
109 _MANAGER_TIMESTAMP_MAX
,
110 _MANAGER_TIMESTAMP_INVALID
= -1,
115 #include "path-lookup.h"
116 #include "show-status.h"
117 #include "unit-name.h"
119 typedef enum ManagerTestRunFlags
{
120 MANAGER_TEST_NORMAL
= 0, /* run normally */
121 MANAGER_TEST_RUN_MINIMAL
= 1 << 0, /* create basic data structures */
122 MANAGER_TEST_RUN_BASIC
= 1 << 1, /* interact with the environment */
123 MANAGER_TEST_RUN_ENV_GENERATORS
= 1 << 2, /* also run env generators */
124 MANAGER_TEST_RUN_GENERATORS
= 1 << 3, /* also run unit generators */
125 MANAGER_TEST_FULL
= MANAGER_TEST_RUN_BASIC
| MANAGER_TEST_RUN_ENV_GENERATORS
| MANAGER_TEST_RUN_GENERATORS
,
126 } ManagerTestRunFlags
;
128 assert_cc((MANAGER_TEST_FULL
& UINT8_MAX
) == MANAGER_TEST_FULL
);
131 /* Note that the set of units we know of is allowed to be
132 * inconsistent. However the subset of it that is loaded may
133 * not, and the list of jobs may neither. */
135 /* Active jobs and units */
136 Hashmap
*units
; /* name string => Unit object n:1 */
137 Hashmap
*units_by_invocation_id
;
138 Hashmap
*jobs
; /* job id => Job object 1:1 */
140 /* To make it easy to iterate through the units of a specific
141 * type we maintain a per type linked list */
142 LIST_HEAD(Unit
, units_by_type
[_UNIT_TYPE_MAX
]);
144 /* Units that need to be loaded */
145 LIST_HEAD(Unit
, load_queue
); /* this is actually more a stack than a queue, but uh. */
147 /* Jobs that need to be run */
148 LIST_HEAD(Job
, run_queue
); /* more a stack than a queue, too */
150 /* Units and jobs that have not yet been announced via
151 * D-Bus. When something about a job changes it is added here
152 * if it is not in there yet. This allows easy coalescing of
153 * D-Bus change signals. */
154 LIST_HEAD(Unit
, dbus_unit_queue
);
155 LIST_HEAD(Job
, dbus_job_queue
);
157 /* Units to remove */
158 LIST_HEAD(Unit
, cleanup_queue
);
160 /* Units and jobs to check when doing GC */
161 LIST_HEAD(Unit
, gc_unit_queue
);
162 LIST_HEAD(Job
, gc_job_queue
);
164 /* Units that should be realized */
165 LIST_HEAD(Unit
, cgroup_realize_queue
);
167 /* Units whose cgroup ran empty */
168 LIST_HEAD(Unit
, cgroup_empty_queue
);
170 /* Units whose memory.event fired */
171 LIST_HEAD(Unit
, cgroup_oom_queue
);
173 /* Target units whose default target dependencies haven't been set yet */
174 LIST_HEAD(Unit
, target_deps_queue
);
176 /* Units that might be subject to StopWhenUnneeded= clean-up */
177 LIST_HEAD(Unit
, stop_when_unneeded_queue
);
181 /* This maps PIDs we care about to units that are interested in. We allow multiple units to he interested in
182 * the same PID and multiple PIDs to be relevant to the same unit. Since in most cases only a single unit will
183 * be interested in the same PID we use a somewhat special encoding here: the first unit interested in a PID is
184 * stored directly in the hashmap, keyed by the PID unmodified. If there are other units interested too they'll
185 * be stored in a NULL-terminated array, and keyed by the negative PID. This is safe as pid_t is signed and
186 * negative PIDs are not used for regular processes but process groups, which we don't care about in this
187 * context, but this allows us to use the negative range for our own purposes. */
188 Hashmap
*watch_pids
; /* pid => unit as well as -pid => array of units */
190 /* A set contains all units which cgroup should be refreshed after startup */
193 /* A set which contains all currently failed units */
196 sd_event_source
*run_queue_event_source
;
200 sd_event_source
*notify_event_source
;
202 int cgroups_agent_fd
;
203 sd_event_source
*cgroups_agent_event_source
;
206 sd_event_source
*signal_event_source
;
208 sd_event_source
*sigchld_event_source
;
211 sd_event_source
*time_change_event_source
;
213 sd_event_source
*timezone_change_event_source
;
215 sd_event_source
*jobs_in_progress_event_source
;
217 int user_lookup_fds
[2];
218 sd_event_source
*user_lookup_event_source
;
220 sd_event_source
*sync_bus_names_event_source
;
222 UnitFileScope unit_file_scope
;
223 LookupPaths lookup_paths
;
224 Set
*unit_path_cache
;
226 char **transient_environment
; /* The environment, as determined from config files, kernel cmdline and environment generators */
227 char **client_environment
; /* Environment variables created by clients through the bus API */
229 usec_t runtime_watchdog
;
230 usec_t shutdown_watchdog
;
232 dual_timestamp timestamps
[_MANAGER_TIMESTAMP_MAX
];
234 /* Data specific to the device subsystem */
235 sd_device_monitor
*device_monitor
;
236 Hashmap
*devices_by_sysfs
;
238 /* Data specific to the mount subsystem */
239 struct libmnt_monitor
*mount_monitor
;
240 sd_event_source
*mount_event_source
;
242 /* Data specific to the swap filesystem */
244 sd_event_source
*swap_event_source
;
245 Hashmap
*swaps_by_devnode
;
247 /* Data specific to the D-Bus subsystem */
248 sd_bus
*api_bus
, *system_bus
;
250 int private_listen_fd
;
251 sd_event_source
*private_listen_event_source
;
253 /* Contains all the clients that are subscribed to signals via
254 the API bus. Note that private bus connections are always
255 considered subscribes, since they last for very short only,
256 and it is much simpler that way. */
257 sd_bus_track
*subscribed
;
258 char **deserialized_subscribed
;
260 /* This is used during reloading: before the reload we queue
261 * the reply message here, and afterwards we send it */
262 sd_bus_message
*pending_reload_message
;
264 Hashmap
*watch_bus
; /* D-Bus names => Unit object n:1 */
266 bool send_reloading_done
;
268 uint32_t current_job_id
;
269 uint32_t default_unit_job_id
;
271 /* Data specific to the Automount subsystem */
274 /* Data specific to the cgroup subsystem */
275 Hashmap
*cgroup_unit
;
276 CGroupMask cgroup_supported
;
279 /* Notifications from cgroups, when the unified hierarchy is used is done via inotify. */
280 int cgroup_inotify_fd
;
281 sd_event_source
*cgroup_inotify_event_source
;
283 /* Maps for finding the unit for each inotify watch descriptor for the cgroup.events and
284 * memory.events cgroupv2 attributes. */
285 Hashmap
*cgroup_control_inotify_wd_unit
;
286 Hashmap
*cgroup_memory_inotify_wd_unit
;
288 /* A defer event for handling cgroup empty events and processing them after SIGCHLD in all cases. */
289 sd_event_source
*cgroup_empty_event_source
;
290 sd_event_source
*cgroup_oom_event_source
;
292 /* Make sure the user cannot accidentally unmount our cgroup
298 /* The stat() data the last time we saw /etc/localtime */
299 usec_t etc_localtime_mtime
;
300 bool etc_localtime_accessible
:1;
302 ManagerObjective objective
:5;
305 bool dispatching_load_queue
:1;
309 /* Have we already sent out the READY=1 notification? */
312 /* Have we already printed the taint line if necessary? */
315 /* Have we ever changed the "kernel.pid_max" sysctl? */
316 bool sysctl_pid_max_changed
:1;
318 ManagerTestRunFlags test_run_flags
:8;
320 /* If non-zero, exit with the following value when the systemd
321 * process terminate. Useful for containers: systemd-nspawn could get
322 * the return value. */
323 uint8_t return_value
;
325 ShowStatus show_status
;
326 StatusUnitFormat status_unit_format
;
328 bool no_console_output
;
329 bool service_watchdogs
;
331 ExecOutput default_std_output
, default_std_error
;
333 usec_t default_restart_usec
, default_timeout_start_usec
, default_timeout_stop_usec
;
334 usec_t default_timeout_abort_usec
;
335 bool default_timeout_abort_set
;
337 usec_t default_start_limit_interval
;
338 unsigned default_start_limit_burst
;
340 bool default_cpu_accounting
;
341 bool default_memory_accounting
;
342 bool default_io_accounting
;
343 bool default_blockio_accounting
;
344 bool default_tasks_accounting
;
345 bool default_ip_accounting
;
347 uint64_t default_tasks_max
;
348 usec_t default_timer_accuracy_usec
;
350 OOMPolicy default_oom_policy
;
352 int original_log_level
;
353 LogTarget original_log_target
;
354 bool log_level_overridden
:1;
355 bool log_target_overridden
:1;
357 struct rlimit
*rlimit
[_RLIMIT_MAX
];
359 /* non-zero if we are reloading or reexecuting, */
362 unsigned n_installed_jobs
;
363 unsigned n_failed_jobs
;
365 /* Jobs in progress watching */
366 unsigned n_running_jobs
;
367 unsigned n_on_console
;
368 unsigned jobs_in_progress_iteration
;
370 /* Do we have any outstanding password prompts? */
371 int have_ask_password
;
372 int ask_password_inotify_fd
;
373 sd_event_source
*ask_password_event_source
;
375 /* Type=idle pipes */
377 sd_event_source
*idle_pipe_event_source
;
380 char *switch_root_init
;
382 /* This maps all possible path prefixes to the units needing
383 * them. It's a hashmap with a path string as key and a Set as
384 * value where Unit objects are contained. */
385 Hashmap
*units_requiring_mounts_for
;
387 /* Used for processing polkit authorization responses */
388 Hashmap
*polkit_registry
;
390 /* Dynamic users/groups, indexed by their name */
391 Hashmap
*dynamic_users
;
393 /* Keep track of all UIDs and GIDs any of our services currently use. This is useful for the RemoveIPC= logic. */
397 /* ExecRuntime, indexed by their owner unit id */
398 Hashmap
*exec_runtime_by_id
;
400 /* When the user hits C-A-D more than 7 times per 2s, do something immediately... */
401 RateLimit ctrl_alt_del_ratelimit
;
402 EmergencyAction cad_burst_action
;
404 const char *unit_log_field
;
405 const char *unit_log_format_string
;
407 const char *invocation_log_field
;
408 const char *invocation_log_format_string
;
410 int first_boot
; /* tri-state */
412 /* Prefixes of e.g. RuntimeDirectory= */
413 char *prefix
[_EXEC_DIRECTORY_TYPE_MAX
];
415 /* Used in the SIGCHLD and sd_notify() message invocation logic to avoid that we dispatch the same event
416 * multiple times on the same unit. */
420 bool honor_device_enumeration
;
423 static inline usec_t
manager_default_timeout_abort_usec(Manager
*m
) {
425 return m
->default_timeout_abort_set
? m
->default_timeout_abort_usec
: m
->default_timeout_stop_usec
;
428 #define MANAGER_IS_SYSTEM(m) ((m)->unit_file_scope == UNIT_FILE_SYSTEM)
429 #define MANAGER_IS_USER(m) ((m)->unit_file_scope != UNIT_FILE_SYSTEM)
431 #define MANAGER_IS_RELOADING(m) ((m)->n_reloading > 0)
433 #define MANAGER_IS_FINISHED(m) (dual_timestamp_is_set((m)->timestamps + MANAGER_TIMESTAMP_FINISH))
435 /* The objective is set to OK as soon as we enter the main loop, and set otherwise as soon as we are done with it */
436 #define MANAGER_IS_RUNNING(m) ((m)->objective == MANAGER_OK)
438 #define MANAGER_IS_TEST_RUN(m) ((m)->test_run_flags != 0)
440 int manager_new(UnitFileScope scope
, ManagerTestRunFlags test_run_flags
, Manager
**m
);
441 Manager
* manager_free(Manager
*m
);
442 DEFINE_TRIVIAL_CLEANUP_FUNC(Manager
*, manager_free
);
444 int manager_startup(Manager
*m
, FILE *serialization
, FDSet
*fds
);
446 Job
*manager_get_job(Manager
*m
, uint32_t id
);
447 Unit
*manager_get_unit(Manager
*m
, const char *name
);
449 int manager_get_job_from_dbus_path(Manager
*m
, const char *s
, Job
**_j
);
451 int manager_load_unit_prepare(Manager
*m
, const char *name
, const char *path
, sd_bus_error
*e
, Unit
**_ret
);
452 int manager_load_unit(Manager
*m
, const char *name
, const char *path
, sd_bus_error
*e
, Unit
**_ret
);
453 int manager_load_startable_unit_or_warn(Manager
*m
, const char *name
, const char *path
, Unit
**ret
);
454 int manager_load_unit_from_dbus_path(Manager
*m
, const char *s
, sd_bus_error
*e
, Unit
**_u
);
456 int manager_add_job(Manager
*m
, JobType type
, Unit
*unit
, JobMode mode
, Set
*affected_jobs
, sd_bus_error
*e
, Job
**_ret
);
457 int manager_add_job_by_name(Manager
*m
, JobType type
, const char *name
, JobMode mode
, Set
*affected_jobs
, sd_bus_error
*e
, Job
**_ret
);
458 int manager_add_job_by_name_and_warn(Manager
*m
, JobType type
, const char *name
, JobMode mode
, Set
*affected_jobs
, Job
**ret
);
459 int manager_propagate_reload(Manager
*m
, Unit
*unit
, JobMode mode
, sd_bus_error
*e
);
461 void manager_dump_units(Manager
*s
, FILE *f
, const char *prefix
);
462 void manager_dump_jobs(Manager
*s
, FILE *f
, const char *prefix
);
463 void manager_dump(Manager
*s
, FILE *f
, const char *prefix
);
464 int manager_get_dump_string(Manager
*m
, char **ret
);
466 void manager_clear_jobs(Manager
*m
);
468 void manager_unwatch_pid(Manager
*m
, pid_t pid
);
470 unsigned manager_dispatch_load_queue(Manager
*m
);
472 int manager_default_environment(Manager
*m
);
473 int manager_transient_environment_add(Manager
*m
, char **plus
);
474 int manager_client_environment_modify(Manager
*m
, char **minus
, char **plus
);
475 int manager_get_effective_environment(Manager
*m
, char ***ret
);
477 int manager_set_default_rlimits(Manager
*m
, struct rlimit
**default_rlimit
);
479 int manager_loop(Manager
*m
);
481 int manager_open_serialization(Manager
*m
, FILE **_f
);
483 int manager_serialize(Manager
*m
, FILE *f
, FDSet
*fds
, bool switching_root
);
484 int manager_deserialize(Manager
*m
, FILE *f
, FDSet
*fds
);
486 int manager_reload(Manager
*m
);
488 void manager_reset_failed(Manager
*m
);
490 void manager_send_unit_audit(Manager
*m
, Unit
*u
, int type
, bool success
);
491 void manager_send_unit_plymouth(Manager
*m
, Unit
*u
);
493 bool manager_unit_inactive_or_pending(Manager
*m
, const char *name
);
495 void manager_check_finished(Manager
*m
);
497 void manager_recheck_dbus(Manager
*m
);
498 void manager_recheck_journal(Manager
*m
);
500 void manager_set_show_status(Manager
*m
, ShowStatus mode
);
501 void manager_set_first_boot(Manager
*m
, bool b
);
503 void manager_status_printf(Manager
*m
, StatusType type
, const char *status
, const char *format
, ...) _printf_(4,5);
504 void manager_flip_auto_status(Manager
*m
, bool enable
);
506 Set
*manager_get_units_requiring_mounts_for(Manager
*m
, const char *path
);
508 ManagerState
manager_state(Manager
*m
);
510 int manager_update_failed_units(Manager
*m
, Unit
*u
, bool failed
);
512 void manager_unref_uid(Manager
*m
, uid_t uid
, bool destroy_now
);
513 int manager_ref_uid(Manager
*m
, uid_t uid
, bool clean_ipc
);
515 void manager_unref_gid(Manager
*m
, gid_t gid
, bool destroy_now
);
516 int manager_ref_gid(Manager
*m
, gid_t gid
, bool destroy_now
);
518 void manager_vacuum_uid_refs(Manager
*m
);
519 void manager_vacuum_gid_refs(Manager
*m
);
521 void manager_serialize_uid_refs(Manager
*m
, FILE *f
);
522 void manager_deserialize_uid_refs_one(Manager
*m
, const char *value
);
524 void manager_serialize_gid_refs(Manager
*m
, FILE *f
);
525 void manager_deserialize_gid_refs_one(Manager
*m
, const char *value
);
527 char *manager_taint_string(Manager
*m
);
529 void manager_ref_console(Manager
*m
);
530 void manager_unref_console(Manager
*m
);
532 void manager_override_log_level(Manager
*m
, int level
);
533 void manager_restore_original_log_level(Manager
*m
);
535 void manager_override_log_target(Manager
*m
, LogTarget target
);
536 void manager_restore_original_log_target(Manager
*m
);
538 const char *manager_state_to_string(ManagerState m
) _const_
;
539 ManagerState
manager_state_from_string(const char *s
) _pure_
;
541 const char *manager_get_confirm_spawn(Manager
*m
);
542 bool manager_is_confirm_spawn_disabled(Manager
*m
);
543 void manager_disable_confirm_spawn(void);
545 const char *manager_timestamp_to_string(ManagerTimestamp m
) _const_
;
546 ManagerTimestamp
manager_timestamp_from_string(const char *s
) _pure_
;
547 ManagerTimestamp
manager_timestamp_initrd_mangle(ManagerTimestamp s
);
549 const char* oom_policy_to_string(OOMPolicy i
) _const_
;
550 OOMPolicy
oom_policy_from_string(const char *s
) _pure_
;