1 /* SPDX-License-Identifier: LGPL-2.1+ */
11 #include "cgroup-util.h"
15 #include "ip-address-access.h"
18 #include "ratelimit.h"
21 struct libmnt_monitor
;
22 typedef struct Unit Unit
;
24 /* Enforce upper limit how many names we allow */
25 #define MANAGER_MAX_NAMES 131072 /* 128K */
27 typedef struct Manager Manager
;
29 /* An externally visible state. We don't actually maintain this as state variable, but derive it from various fields
31 typedef enum ManagerState
{
39 _MANAGER_STATE_INVALID
= -1
42 typedef enum ManagerObjective
{
52 _MANAGER_OBJECTIVE_MAX
,
53 _MANAGER_OBJECTIVE_INVALID
= -1
56 typedef enum StatusType
{
57 STATUS_TYPE_EPHEMERAL
,
60 STATUS_TYPE_EMERGENCY
,
63 typedef enum OOMPolicy
{
64 OOM_CONTINUE
, /* The kernel kills the process it wants to kill, and that's it */
65 OOM_STOP
, /* The kernel kills the process it wants to kill, and we stop the unit */
66 OOM_KILL
, /* The kernel kills the process it wants to kill, and all others in the unit, and we stop the unit */
68 _OOM_POLICY_INVALID
= -1
72 * 1. TIMESTAMP_FIRMWARE, TIMESTAMP_LOADER, TIMESTAMP_KERNEL, TIMESTAMP_INITRD,
73 * TIMESTAMP_SECURITY_START, and TIMESTAMP_SECURITY_FINISH are set only when
74 * the manager is system and not running under container environment.
76 * 2. The monotonic timestamp of TIMESTAMP_KERNEL is always zero.
78 * 3. The realtime timestamp of TIMESTAMP_KERNEL will be unset if the system does not
81 * 4. TIMESTAMP_FIRMWARE and TIMESTAMP_LOADER will be unset if the system does not
82 * have RTC, or systemd is built without EFI support.
84 * 5. The monotonic timestamps of TIMESTAMP_FIRMWARE and TIMESTAMP_LOADER are stored as
85 * negative of the actual value.
87 * 6. TIMESTAMP_USERSPACE is the timestamp of when the manager was started.
89 * 7. TIMESTAMP_INITRD_* are set only when the system is booted with an initrd.
92 typedef enum ManagerTimestamp
{
93 MANAGER_TIMESTAMP_FIRMWARE
,
94 MANAGER_TIMESTAMP_LOADER
,
95 MANAGER_TIMESTAMP_KERNEL
,
96 MANAGER_TIMESTAMP_INITRD
,
97 MANAGER_TIMESTAMP_USERSPACE
,
98 MANAGER_TIMESTAMP_FINISH
,
100 MANAGER_TIMESTAMP_SECURITY_START
,
101 MANAGER_TIMESTAMP_SECURITY_FINISH
,
102 MANAGER_TIMESTAMP_GENERATORS_START
,
103 MANAGER_TIMESTAMP_GENERATORS_FINISH
,
104 MANAGER_TIMESTAMP_UNITS_LOAD_START
,
105 MANAGER_TIMESTAMP_UNITS_LOAD_FINISH
,
107 MANAGER_TIMESTAMP_INITRD_SECURITY_START
,
108 MANAGER_TIMESTAMP_INITRD_SECURITY_FINISH
,
109 MANAGER_TIMESTAMP_INITRD_GENERATORS_START
,
110 MANAGER_TIMESTAMP_INITRD_GENERATORS_FINISH
,
111 MANAGER_TIMESTAMP_INITRD_UNITS_LOAD_START
,
112 MANAGER_TIMESTAMP_INITRD_UNITS_LOAD_FINISH
,
113 _MANAGER_TIMESTAMP_MAX
,
114 _MANAGER_TIMESTAMP_INVALID
= -1,
117 typedef enum WatchdogType
{
126 #include "path-lookup.h"
127 #include "show-status.h"
128 #include "unit-name.h"
130 typedef enum ManagerTestRunFlags
{
131 MANAGER_TEST_NORMAL
= 0, /* run normally */
132 MANAGER_TEST_RUN_MINIMAL
= 1 << 0, /* create basic data structures */
133 MANAGER_TEST_RUN_BASIC
= 1 << 1, /* interact with the environment */
134 MANAGER_TEST_RUN_ENV_GENERATORS
= 1 << 2, /* also run env generators */
135 MANAGER_TEST_RUN_GENERATORS
= 1 << 3, /* also run unit generators */
136 MANAGER_TEST_FULL
= MANAGER_TEST_RUN_BASIC
| MANAGER_TEST_RUN_ENV_GENERATORS
| MANAGER_TEST_RUN_GENERATORS
,
137 } ManagerTestRunFlags
;
139 assert_cc((MANAGER_TEST_FULL
& UINT8_MAX
) == MANAGER_TEST_FULL
);
142 /* Note that the set of units we know of is allowed to be
143 * inconsistent. However the subset of it that is loaded may
144 * not, and the list of jobs may neither. */
146 /* Active jobs and units */
147 Hashmap
*units
; /* name string => Unit object n:1 */
148 Hashmap
*units_by_invocation_id
;
149 Hashmap
*jobs
; /* job id => Job object 1:1 */
151 /* To make it easy to iterate through the units of a specific
152 * type we maintain a per type linked list */
153 LIST_HEAD(Unit
, units_by_type
[_UNIT_TYPE_MAX
]);
155 /* Units that need to be loaded */
156 LIST_HEAD(Unit
, load_queue
); /* this is actually more a stack than a queue, but uh. */
158 /* Jobs that need to be run */
159 struct Prioq
*run_queue
;
161 /* Units and jobs that have not yet been announced via
162 * D-Bus. When something about a job changes it is added here
163 * if it is not in there yet. This allows easy coalescing of
164 * D-Bus change signals. */
165 LIST_HEAD(Unit
, dbus_unit_queue
);
166 LIST_HEAD(Job
, dbus_job_queue
);
168 /* Units to remove */
169 LIST_HEAD(Unit
, cleanup_queue
);
171 /* Units and jobs to check when doing GC */
172 LIST_HEAD(Unit
, gc_unit_queue
);
173 LIST_HEAD(Job
, gc_job_queue
);
175 /* Units that should be realized */
176 LIST_HEAD(Unit
, cgroup_realize_queue
);
178 /* Units whose cgroup ran empty */
179 LIST_HEAD(Unit
, cgroup_empty_queue
);
181 /* Units whose memory.event fired */
182 LIST_HEAD(Unit
, cgroup_oom_queue
);
184 /* Target units whose default target dependencies haven't been set yet */
185 LIST_HEAD(Unit
, target_deps_queue
);
187 /* Units that might be subject to StopWhenUnneeded= clean-up */
188 LIST_HEAD(Unit
, stop_when_unneeded_queue
);
192 /* This maps PIDs we care about to units that are interested in. We allow multiple units to he interested in
193 * the same PID and multiple PIDs to be relevant to the same unit. Since in most cases only a single unit will
194 * be interested in the same PID we use a somewhat special encoding here: the first unit interested in a PID is
195 * stored directly in the hashmap, keyed by the PID unmodified. If there are other units interested too they'll
196 * be stored in a NULL-terminated array, and keyed by the negative PID. This is safe as pid_t is signed and
197 * negative PIDs are not used for regular processes but process groups, which we don't care about in this
198 * context, but this allows us to use the negative range for our own purposes. */
199 Hashmap
*watch_pids
; /* pid => unit as well as -pid => array of units */
201 /* A set contains all units which cgroup should be refreshed after startup */
204 /* A set which contains all currently failed units */
207 sd_event_source
*run_queue_event_source
;
211 sd_event_source
*notify_event_source
;
213 int cgroups_agent_fd
;
214 sd_event_source
*cgroups_agent_event_source
;
217 sd_event_source
*signal_event_source
;
219 sd_event_source
*sigchld_event_source
;
222 sd_event_source
*time_change_event_source
;
224 sd_event_source
*timezone_change_event_source
;
226 sd_event_source
*jobs_in_progress_event_source
;
228 int user_lookup_fds
[2];
229 sd_event_source
*user_lookup_event_source
;
231 UnitFileScope unit_file_scope
;
232 LookupPaths lookup_paths
;
233 Hashmap
*unit_id_map
;
234 Hashmap
*unit_name_map
;
235 Set
*unit_path_cache
;
236 uint64_t unit_cache_timestamp_hash
;
238 char **transient_environment
; /* The environment, as determined from config files, kernel cmdline and environment generators */
239 char **client_environment
; /* Environment variables created by clients through the bus API */
241 usec_t watchdog
[_WATCHDOG_TYPE_MAX
];
242 usec_t watchdog_overridden
[_WATCHDOG_TYPE_MAX
];
244 dual_timestamp timestamps
[_MANAGER_TIMESTAMP_MAX
];
246 /* Data specific to the device subsystem */
247 sd_device_monitor
*device_monitor
;
248 Hashmap
*devices_by_sysfs
;
250 /* Data specific to the mount subsystem */
251 struct libmnt_monitor
*mount_monitor
;
252 sd_event_source
*mount_event_source
;
254 /* Data specific to the swap filesystem */
256 sd_event_source
*swap_event_source
;
257 Hashmap
*swaps_by_devnode
;
259 /* Data specific to the D-Bus subsystem */
260 sd_bus
*api_bus
, *system_bus
;
262 int private_listen_fd
;
263 sd_event_source
*private_listen_event_source
;
265 /* Contains all the clients that are subscribed to signals via
266 the API bus. Note that private bus connections are always
267 considered subscribes, since they last for very short only,
268 and it is much simpler that way. */
269 sd_bus_track
*subscribed
;
270 char **deserialized_subscribed
;
272 /* This is used during reloading: before the reload we queue
273 * the reply message here, and afterwards we send it */
274 sd_bus_message
*pending_reload_message
;
276 Hashmap
*watch_bus
; /* D-Bus names => Unit object n:1 */
278 bool send_reloading_done
;
280 uint32_t current_job_id
;
281 uint32_t default_unit_job_id
;
283 /* Data specific to the Automount subsystem */
286 /* Data specific to the cgroup subsystem */
287 Hashmap
*cgroup_unit
;
288 CGroupMask cgroup_supported
;
291 /* Notifications from cgroups, when the unified hierarchy is used is done via inotify. */
292 int cgroup_inotify_fd
;
293 sd_event_source
*cgroup_inotify_event_source
;
295 /* Maps for finding the unit for each inotify watch descriptor for the cgroup.events and
296 * memory.events cgroupv2 attributes. */
297 Hashmap
*cgroup_control_inotify_wd_unit
;
298 Hashmap
*cgroup_memory_inotify_wd_unit
;
300 /* A defer event for handling cgroup empty events and processing them after SIGCHLD in all cases. */
301 sd_event_source
*cgroup_empty_event_source
;
302 sd_event_source
*cgroup_oom_event_source
;
304 /* Make sure the user cannot accidentally unmount our cgroup
310 /* The stat() data the last time we saw /etc/localtime */
311 usec_t etc_localtime_mtime
;
312 bool etc_localtime_accessible
:1;
314 ManagerObjective objective
:5;
317 bool dispatching_load_queue
:1;
321 /* Have we already sent out the READY=1 notification? */
324 /* Have we already printed the taint line if necessary? */
327 /* Have we ever changed the "kernel.pid_max" sysctl? */
328 bool sysctl_pid_max_changed
:1;
330 ManagerTestRunFlags test_run_flags
:8;
332 /* If non-zero, exit with the following value when the systemd
333 * process terminate. Useful for containers: systemd-nspawn could get
334 * the return value. */
335 uint8_t return_value
;
337 ShowStatus show_status
;
338 ShowStatus show_status_overridden
;
339 StatusUnitFormat status_unit_format
;
341 bool no_console_output
;
342 bool service_watchdogs
;
344 ExecOutput default_std_output
, default_std_error
;
346 usec_t default_restart_usec
, default_timeout_start_usec
, default_timeout_stop_usec
;
347 usec_t default_timeout_abort_usec
;
348 bool default_timeout_abort_set
;
350 usec_t default_start_limit_interval
;
351 unsigned default_start_limit_burst
;
353 bool default_cpu_accounting
;
354 bool default_memory_accounting
;
355 bool default_io_accounting
;
356 bool default_blockio_accounting
;
357 bool default_tasks_accounting
;
358 bool default_ip_accounting
;
360 TasksMax default_tasks_max
;
361 usec_t default_timer_accuracy_usec
;
363 OOMPolicy default_oom_policy
;
365 int original_log_level
;
366 LogTarget original_log_target
;
367 bool log_level_overridden
:1;
368 bool log_target_overridden
:1;
370 struct rlimit
*rlimit
[_RLIMIT_MAX
];
372 /* non-zero if we are reloading or reexecuting, */
375 unsigned n_installed_jobs
;
376 unsigned n_failed_jobs
;
378 /* Jobs in progress watching */
379 unsigned n_running_jobs
;
380 unsigned n_on_console
;
381 unsigned jobs_in_progress_iteration
;
383 /* Do we have any outstanding password prompts? */
384 int have_ask_password
;
385 int ask_password_inotify_fd
;
386 sd_event_source
*ask_password_event_source
;
388 /* Type=idle pipes */
390 sd_event_source
*idle_pipe_event_source
;
393 char *switch_root_init
;
395 /* This maps all possible path prefixes to the units needing
396 * them. It's a hashmap with a path string as key and a Set as
397 * value where Unit objects are contained. */
398 Hashmap
*units_requiring_mounts_for
;
400 /* Used for processing polkit authorization responses */
401 Hashmap
*polkit_registry
;
403 /* Dynamic users/groups, indexed by their name */
404 Hashmap
*dynamic_users
;
406 /* Keep track of all UIDs and GIDs any of our services currently use. This is useful for the RemoveIPC= logic. */
410 /* ExecRuntime, indexed by their owner unit id */
411 Hashmap
*exec_runtime_by_id
;
413 /* When the user hits C-A-D more than 7 times per 2s, do something immediately... */
414 RateLimit ctrl_alt_del_ratelimit
;
415 EmergencyAction cad_burst_action
;
417 const char *unit_log_field
;
418 const char *unit_log_format_string
;
420 const char *invocation_log_field
;
421 const char *invocation_log_format_string
;
423 int first_boot
; /* tri-state */
425 /* Prefixes of e.g. RuntimeDirectory= */
426 char *prefix
[_EXEC_DIRECTORY_TYPE_MAX
];
427 char *received_credentials
;
429 /* Used in the SIGCHLD and sd_notify() message invocation logic to avoid that we dispatch the same event
430 * multiple times on the same unit. */
434 bool honor_device_enumeration
;
436 VarlinkServer
*varlink_server
;
439 static inline usec_t
manager_default_timeout_abort_usec(Manager
*m
) {
441 return m
->default_timeout_abort_set
? m
->default_timeout_abort_usec
: m
->default_timeout_stop_usec
;
444 #define MANAGER_IS_SYSTEM(m) ((m)->unit_file_scope == UNIT_FILE_SYSTEM)
445 #define MANAGER_IS_USER(m) ((m)->unit_file_scope != UNIT_FILE_SYSTEM)
447 #define MANAGER_IS_RELOADING(m) ((m)->n_reloading > 0)
449 #define MANAGER_IS_FINISHED(m) (dual_timestamp_is_set((m)->timestamps + MANAGER_TIMESTAMP_FINISH))
451 /* 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 */
452 #define MANAGER_IS_RUNNING(m) ((m)->objective == MANAGER_OK)
454 #define MANAGER_IS_TEST_RUN(m) ((m)->test_run_flags != 0)
456 int manager_new(UnitFileScope scope
, ManagerTestRunFlags test_run_flags
, Manager
**m
);
457 Manager
* manager_free(Manager
*m
);
458 DEFINE_TRIVIAL_CLEANUP_FUNC(Manager
*, manager_free
);
460 int manager_startup(Manager
*m
, FILE *serialization
, FDSet
*fds
);
462 Job
*manager_get_job(Manager
*m
, uint32_t id
);
463 Unit
*manager_get_unit(Manager
*m
, const char *name
);
465 int manager_get_job_from_dbus_path(Manager
*m
, const char *s
, Job
**_j
);
467 bool manager_unit_cache_should_retry_load(Unit
*u
);
468 int manager_load_unit_prepare(Manager
*m
, const char *name
, const char *path
, sd_bus_error
*e
, Unit
**_ret
);
469 int manager_load_unit(Manager
*m
, const char *name
, const char *path
, sd_bus_error
*e
, Unit
**_ret
);
470 int manager_load_startable_unit_or_warn(Manager
*m
, const char *name
, const char *path
, Unit
**ret
);
471 int manager_load_unit_from_dbus_path(Manager
*m
, const char *s
, sd_bus_error
*e
, Unit
**_u
);
473 int manager_add_job(Manager
*m
, JobType type
, Unit
*unit
, JobMode mode
, Set
*affected_jobs
, sd_bus_error
*e
, Job
**_ret
);
474 int manager_add_job_by_name(Manager
*m
, JobType type
, const char *name
, JobMode mode
, Set
*affected_jobs
, sd_bus_error
*e
, Job
**_ret
);
475 int manager_add_job_by_name_and_warn(Manager
*m
, JobType type
, const char *name
, JobMode mode
, Set
*affected_jobs
, Job
**ret
);
476 int manager_propagate_reload(Manager
*m
, Unit
*unit
, JobMode mode
, sd_bus_error
*e
);
478 void manager_dump_units(Manager
*s
, FILE *f
, const char *prefix
);
479 void manager_dump_jobs(Manager
*s
, FILE *f
, const char *prefix
);
480 void manager_dump(Manager
*s
, FILE *f
, const char *prefix
);
481 int manager_get_dump_string(Manager
*m
, char **ret
);
483 void manager_clear_jobs(Manager
*m
);
485 void manager_unwatch_pid(Manager
*m
, pid_t pid
);
487 unsigned manager_dispatch_load_queue(Manager
*m
);
489 int manager_default_environment(Manager
*m
);
490 int manager_transient_environment_add(Manager
*m
, char **plus
);
491 int manager_client_environment_modify(Manager
*m
, char **minus
, char **plus
);
492 int manager_get_effective_environment(Manager
*m
, char ***ret
);
494 int manager_set_default_rlimits(Manager
*m
, struct rlimit
**default_rlimit
);
496 int manager_loop(Manager
*m
);
498 int manager_open_serialization(Manager
*m
, FILE **_f
);
500 int manager_serialize(Manager
*m
, FILE *f
, FDSet
*fds
, bool switching_root
);
501 int manager_deserialize(Manager
*m
, FILE *f
, FDSet
*fds
);
503 int manager_reload(Manager
*m
);
505 void manager_reset_failed(Manager
*m
);
507 void manager_send_unit_audit(Manager
*m
, Unit
*u
, int type
, bool success
);
508 void manager_send_unit_plymouth(Manager
*m
, Unit
*u
);
510 bool manager_unit_inactive_or_pending(Manager
*m
, const char *name
);
512 void manager_check_finished(Manager
*m
);
514 void disable_printk_ratelimit(void);
515 void manager_recheck_dbus(Manager
*m
);
516 void manager_recheck_journal(Manager
*m
);
518 bool manager_get_show_status_on(Manager
*m
);
519 void manager_set_show_status(Manager
*m
, ShowStatus mode
, const char *reason
);
520 void manager_override_show_status(Manager
*m
, ShowStatus mode
, const char *reason
);
522 void manager_set_first_boot(Manager
*m
, bool b
);
524 void manager_status_printf(Manager
*m
, StatusType type
, const char *status
, const char *format
, ...) _printf_(4,5);
526 Set
*manager_get_units_requiring_mounts_for(Manager
*m
, const char *path
);
528 ManagerState
manager_state(Manager
*m
);
530 int manager_update_failed_units(Manager
*m
, Unit
*u
, bool failed
);
532 void manager_unref_uid(Manager
*m
, uid_t uid
, bool destroy_now
);
533 int manager_ref_uid(Manager
*m
, uid_t uid
, bool clean_ipc
);
535 void manager_unref_gid(Manager
*m
, gid_t gid
, bool destroy_now
);
536 int manager_ref_gid(Manager
*m
, gid_t gid
, bool destroy_now
);
538 char *manager_taint_string(Manager
*m
);
540 void manager_ref_console(Manager
*m
);
541 void manager_unref_console(Manager
*m
);
543 void manager_override_log_level(Manager
*m
, int level
);
544 void manager_restore_original_log_level(Manager
*m
);
546 void manager_override_log_target(Manager
*m
, LogTarget target
);
547 void manager_restore_original_log_target(Manager
*m
);
549 const char *manager_state_to_string(ManagerState m
) _const_
;
550 ManagerState
manager_state_from_string(const char *s
) _pure_
;
552 const char *manager_get_confirm_spawn(Manager
*m
);
553 bool manager_is_confirm_spawn_disabled(Manager
*m
);
554 void manager_disable_confirm_spawn(void);
556 const char *manager_timestamp_to_string(ManagerTimestamp m
) _const_
;
557 ManagerTimestamp
manager_timestamp_from_string(const char *s
) _pure_
;
558 ManagerTimestamp
manager_timestamp_initrd_mangle(ManagerTimestamp s
);
560 usec_t
manager_get_watchdog(Manager
*m
, WatchdogType t
);
561 void manager_set_watchdog(Manager
*m
, WatchdogType t
, usec_t timeout
);
562 int manager_override_watchdog(Manager
*m
, WatchdogType t
, usec_t timeout
);
564 const char* oom_policy_to_string(OOMPolicy i
) _const_
;
565 OOMPolicy
oom_policy_from_string(const char *s
) _pure_
;