]> git.ipfire.org Git - thirdparty/systemd.git/blob - src/core/manager.h
Merge pull request #8150 from poettering/memory-accounting-by-default
[thirdparty/systemd.git] / src / core / manager.h
1 /* SPDX-License-Identifier: LGPL-2.1+ */
2 #pragma once
3
4 /***
5 This file is part of systemd.
6
7 Copyright 2010 Lennart Poettering
8
9 systemd is free software; you can redistribute it and/or modify it
10 under the terms of the GNU Lesser General Public License as published by
11 the Free Software Foundation; either version 2.1 of the License, or
12 (at your option) any later version.
13
14 systemd is distributed in the hope that it will be useful, but
15 WITHOUT ANY WARRANTY; without even the implied warranty of
16 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
17 Lesser General Public License for more details.
18
19 You should have received a copy of the GNU Lesser General Public License
20 along with systemd; If not, see <http://www.gnu.org/licenses/>.
21 ***/
22
23 #include <libmount.h>
24 #include <stdbool.h>
25 #include <stdio.h>
26
27 #include "sd-bus.h"
28 #include "sd-event.h"
29
30 #include "cgroup-util.h"
31 #include "fdset.h"
32 #include "hashmap.h"
33 #include "ip-address-access.h"
34 #include "list.h"
35 #include "ratelimit.h"
36
37 /* Enforce upper limit how many names we allow */
38 #define MANAGER_MAX_NAMES 131072 /* 128K */
39
40 typedef struct Manager Manager;
41
42 typedef enum ManagerState {
43 MANAGER_INITIALIZING,
44 MANAGER_STARTING,
45 MANAGER_RUNNING,
46 MANAGER_DEGRADED,
47 MANAGER_MAINTENANCE,
48 MANAGER_STOPPING,
49 _MANAGER_STATE_MAX,
50 _MANAGER_STATE_INVALID = -1
51 } ManagerState;
52
53 typedef enum ManagerExitCode {
54 MANAGER_OK,
55 MANAGER_EXIT,
56 MANAGER_RELOAD,
57 MANAGER_REEXECUTE,
58 MANAGER_REBOOT,
59 MANAGER_POWEROFF,
60 MANAGER_HALT,
61 MANAGER_KEXEC,
62 MANAGER_SWITCH_ROOT,
63 _MANAGER_EXIT_CODE_MAX,
64 _MANAGER_EXIT_CODE_INVALID = -1
65 } ManagerExitCode;
66
67 typedef enum StatusType {
68 STATUS_TYPE_EPHEMERAL,
69 STATUS_TYPE_NORMAL,
70 STATUS_TYPE_EMERGENCY,
71 } StatusType;
72
73 typedef enum ManagerTimestamp {
74 MANAGER_TIMESTAMP_FIRMWARE,
75 MANAGER_TIMESTAMP_LOADER,
76 MANAGER_TIMESTAMP_KERNEL,
77 MANAGER_TIMESTAMP_INITRD,
78 MANAGER_TIMESTAMP_USERSPACE,
79 MANAGER_TIMESTAMP_FINISH,
80
81 MANAGER_TIMESTAMP_SECURITY_START,
82 MANAGER_TIMESTAMP_SECURITY_FINISH,
83 MANAGER_TIMESTAMP_GENERATORS_START,
84 MANAGER_TIMESTAMP_GENERATORS_FINISH,
85 MANAGER_TIMESTAMP_UNITS_LOAD_START,
86 MANAGER_TIMESTAMP_UNITS_LOAD_FINISH,
87 _MANAGER_TIMESTAMP_MAX,
88 _MANAGER_TIMESTAMP_INVALID = -1,
89 } ManagerTimestamp;
90
91 #include "execute.h"
92 #include "job.h"
93 #include "path-lookup.h"
94 #include "show-status.h"
95 #include "unit-name.h"
96
97 enum {
98 /* 0 = run normally */
99 MANAGER_TEST_RUN_MINIMAL = 1, /* run test w/o generators */
100 MANAGER_TEST_RUN_ENV_GENERATORS = 2, /* also run env generators */
101 MANAGER_TEST_RUN_GENERATORS = 4, /* also run unit generators */
102 MANAGER_TEST_FULL = MANAGER_TEST_RUN_ENV_GENERATORS | MANAGER_TEST_RUN_GENERATORS,
103 };
104 assert_cc((MANAGER_TEST_FULL & UINT8_MAX) == MANAGER_TEST_FULL);
105
106 struct Manager {
107 /* Note that the set of units we know of is allowed to be
108 * inconsistent. However the subset of it that is loaded may
109 * not, and the list of jobs may neither. */
110
111 /* Active jobs and units */
112 Hashmap *units; /* name string => Unit object n:1 */
113 Hashmap *units_by_invocation_id;
114 Hashmap *jobs; /* job id => Job object 1:1 */
115
116 /* To make it easy to iterate through the units of a specific
117 * type we maintain a per type linked list */
118 LIST_HEAD(Unit, units_by_type[_UNIT_TYPE_MAX]);
119
120 /* Units that need to be loaded */
121 LIST_HEAD(Unit, load_queue); /* this is actually more a stack than a queue, but uh. */
122
123 /* Jobs that need to be run */
124 LIST_HEAD(Job, run_queue); /* more a stack than a queue, too */
125
126 /* Units and jobs that have not yet been announced via
127 * D-Bus. When something about a job changes it is added here
128 * if it is not in there yet. This allows easy coalescing of
129 * D-Bus change signals. */
130 LIST_HEAD(Unit, dbus_unit_queue);
131 LIST_HEAD(Job, dbus_job_queue);
132
133 /* Units to remove */
134 LIST_HEAD(Unit, cleanup_queue);
135
136 /* Units and jobs to check when doing GC */
137 LIST_HEAD(Unit, gc_unit_queue);
138 LIST_HEAD(Job, gc_job_queue);
139
140 /* Units that should be realized */
141 LIST_HEAD(Unit, cgroup_realize_queue);
142
143 /* Units whose cgroup ran empty */
144 LIST_HEAD(Unit, cgroup_empty_queue);
145
146 sd_event *event;
147
148 /* This maps PIDs we care about to units that are interested in. We allow multiple units to he interested in
149 * the same PID and multiple PIDs to be relevant to the same unit. Since in most cases only a single unit will
150 * be interested in the same PID we use a somewhat special encoding here: the first unit interested in a PID is
151 * stored directly in the hashmap, keyed by the PID unmodified. If there are other units interested too they'll
152 * be stored in a NULL-terminated array, and keyed by the negative PID. This is safe as pid_t is signed and
153 * negative PIDs are not used for regular processes but process groups, which we don't care about in this
154 * context, but this allows us to use the negative range for our own purposes. */
155 Hashmap *watch_pids; /* pid => unit as well as -pid => array of units */
156
157 /* A set contains all units which cgroup should be refreshed after startup */
158 Set *startup_units;
159
160 /* A set which contains all currently failed units */
161 Set *failed_units;
162
163 sd_event_source *run_queue_event_source;
164
165 char *notify_socket;
166 int notify_fd;
167 sd_event_source *notify_event_source;
168
169 int cgroups_agent_fd;
170 sd_event_source *cgroups_agent_event_source;
171
172 int signal_fd;
173 sd_event_source *signal_event_source;
174
175 sd_event_source *sigchld_event_source;
176
177 int time_change_fd;
178 sd_event_source *time_change_event_source;
179
180 sd_event_source *jobs_in_progress_event_source;
181
182 int user_lookup_fds[2];
183 sd_event_source *user_lookup_event_source;
184
185 sd_event_source *sync_bus_names_event_source;
186
187 UnitFileScope unit_file_scope;
188 LookupPaths lookup_paths;
189 Set *unit_path_cache;
190
191 char **environment;
192
193 usec_t runtime_watchdog;
194 usec_t shutdown_watchdog;
195
196 dual_timestamp timestamps[_MANAGER_TIMESTAMP_MAX];
197
198 struct udev* udev;
199
200 /* Data specific to the device subsystem */
201 struct udev_monitor* udev_monitor;
202 sd_event_source *udev_event_source;
203 Hashmap *devices_by_sysfs;
204
205 /* Data specific to the mount subsystem */
206 struct libmnt_monitor *mount_monitor;
207 sd_event_source *mount_event_source;
208
209 /* Data specific to the swap filesystem */
210 FILE *proc_swaps;
211 sd_event_source *swap_event_source;
212 Hashmap *swaps_by_devnode;
213
214 /* Data specific to the D-Bus subsystem */
215 sd_bus *api_bus, *system_bus;
216 Set *private_buses;
217 int private_listen_fd;
218 sd_event_source *private_listen_event_source;
219
220 /* Contains all the clients that are subscribed to signals via
221 the API bus. Note that private bus connections are always
222 considered subscribes, since they last for very short only,
223 and it is much simpler that way. */
224 sd_bus_track *subscribed;
225 char **deserialized_subscribed;
226
227 /* This is used during reloading: before the reload we queue
228 * the reply message here, and afterwards we send it */
229 sd_bus_message *queued_message;
230
231 Hashmap *watch_bus; /* D-Bus names => Unit object n:1 */
232
233 bool send_reloading_done;
234
235 uint32_t current_job_id;
236 uint32_t default_unit_job_id;
237
238 /* Data specific to the Automount subsystem */
239 int dev_autofs_fd;
240
241 /* Data specific to the cgroup subsystem */
242 Hashmap *cgroup_unit;
243 CGroupMask cgroup_supported;
244 char *cgroup_root;
245
246 /* Notifications from cgroups, when the unified hierarchy is used is done via inotify. */
247 int cgroup_inotify_fd;
248 sd_event_source *cgroup_inotify_event_source;
249 Hashmap *cgroup_inotify_wd_unit;
250
251 /* A defer event for handling cgroup empty events and processing them after SIGCHLD in all cases. */
252 sd_event_source *cgroup_empty_event_source;
253
254 /* Make sure the user cannot accidentally unmount our cgroup
255 * file system */
256 int pin_cgroupfs_fd;
257
258 unsigned gc_marker;
259
260 /* Flags */
261 ManagerExitCode exit_code:5;
262
263 bool dispatching_load_queue:1;
264 bool dispatching_dbus_queue:1;
265
266 bool taint_usr:1;
267
268 /* Have we already sent out the READY=1 notification? */
269 bool ready_sent:1;
270
271 /* Have we already printed the taint line if necessary? */
272 bool taint_logged:1;
273
274 /* Have we ever changed the "kernel.pid_max" sysctl? */
275 bool sysctl_pid_max_changed:1;
276
277 unsigned test_run_flags:8;
278
279 /* If non-zero, exit with the following value when the systemd
280 * process terminate. Useful for containers: systemd-nspawn could get
281 * the return value. */
282 uint8_t return_value;
283
284 ShowStatus show_status;
285 char *confirm_spawn;
286 bool no_console_output;
287 bool service_watchdogs;
288
289 ExecOutput default_std_output, default_std_error;
290
291 usec_t default_restart_usec, default_timeout_start_usec, default_timeout_stop_usec;
292
293 usec_t default_start_limit_interval;
294 unsigned default_start_limit_burst;
295
296 bool default_cpu_accounting;
297 bool default_memory_accounting;
298 bool default_io_accounting;
299 bool default_blockio_accounting;
300 bool default_tasks_accounting;
301 bool default_ip_accounting;
302
303 uint64_t default_tasks_max;
304 usec_t default_timer_accuracy_usec;
305
306 struct rlimit *rlimit[_RLIMIT_MAX];
307
308 /* non-zero if we are reloading or reexecuting, */
309 int n_reloading;
310
311 unsigned n_installed_jobs;
312 unsigned n_failed_jobs;
313
314 /* Jobs in progress watching */
315 unsigned n_running_jobs;
316 unsigned n_on_console;
317 unsigned jobs_in_progress_iteration;
318
319 /* Do we have any outstanding password prompts? */
320 int have_ask_password;
321 int ask_password_inotify_fd;
322 sd_event_source *ask_password_event_source;
323
324 /* Type=idle pipes */
325 int idle_pipe[4];
326 sd_event_source *idle_pipe_event_source;
327
328 char *switch_root;
329 char *switch_root_init;
330
331 /* This maps all possible path prefixes to the units needing
332 * them. It's a hashmap with a path string as key and a Set as
333 * value where Unit objects are contained. */
334 Hashmap *units_requiring_mounts_for;
335
336 /* Used for processing polkit authorization responses */
337 Hashmap *polkit_registry;
338
339 /* Dynamic users/groups, indexed by their name */
340 Hashmap *dynamic_users;
341
342 /* Keep track of all UIDs and GIDs any of our services currently use. This is useful for the RemoveIPC= logic. */
343 Hashmap *uid_refs;
344 Hashmap *gid_refs;
345
346 /* ExecRuntime, indexed by their owner unit id */
347 Hashmap *exec_runtime_by_id;
348
349 /* When the user hits C-A-D more than 7 times per 2s, do something immediately... */
350 RateLimit ctrl_alt_del_ratelimit;
351 EmergencyAction cad_burst_action;
352
353 const char *unit_log_field;
354 const char *unit_log_format_string;
355
356 const char *invocation_log_field;
357 const char *invocation_log_format_string;
358
359 int first_boot; /* tri-state */
360
361 /* Prefixes of e.g. RuntimeDirectory= */
362 char *prefix[_EXEC_DIRECTORY_TYPE_MAX];
363
364 /* Used in the SIGCHLD and sd_notify() message invocation logic to avoid that we dispatch the same event
365 * multiple times on the same unit. */
366 unsigned sigchldgen;
367 unsigned notifygen;
368 };
369
370 #define MANAGER_IS_SYSTEM(m) ((m)->unit_file_scope == UNIT_FILE_SYSTEM)
371 #define MANAGER_IS_USER(m) ((m)->unit_file_scope != UNIT_FILE_SYSTEM)
372
373 #define MANAGER_IS_RELOADING(m) ((m)->n_reloading > 0)
374
375 #define MANAGER_IS_FINISHED(m) (dual_timestamp_is_set((m)->timestamps + MANAGER_TIMESTAMP_FINISH))
376
377 /* The exit code is set to OK as soon as we enter the main loop, and set otherwise as soon as we are done with it */
378 #define MANAGER_IS_RUNNING(m) ((m)->exit_code == MANAGER_OK)
379
380 int manager_new(UnitFileScope scope, unsigned test_run_flags, Manager **m);
381 Manager* manager_free(Manager *m);
382
383 void manager_enumerate(Manager *m);
384 int manager_startup(Manager *m, FILE *serialization, FDSet *fds);
385
386 Job *manager_get_job(Manager *m, uint32_t id);
387 Unit *manager_get_unit(Manager *m, const char *name);
388
389 int manager_get_job_from_dbus_path(Manager *m, const char *s, Job **_j);
390
391 int manager_load_unit_prepare(Manager *m, const char *name, const char *path, sd_bus_error *e, Unit **_ret);
392 int manager_load_unit(Manager *m, const char *name, const char *path, sd_bus_error *e, Unit **_ret);
393 int manager_load_unit_from_dbus_path(Manager *m, const char *s, sd_bus_error *e, Unit **_u);
394
395 int manager_add_job(Manager *m, JobType type, Unit *unit, JobMode mode, sd_bus_error *e, Job **_ret);
396 int manager_add_job_by_name(Manager *m, JobType type, const char *name, JobMode mode, sd_bus_error *e, Job **_ret);
397 int manager_add_job_by_name_and_warn(Manager *m, JobType type, const char *name, JobMode mode, Job **ret);
398 int manager_propagate_reload(Manager *m, Unit *unit, JobMode mode, sd_bus_error *e);
399
400 void manager_dump_units(Manager *s, FILE *f, const char *prefix);
401 void manager_dump_jobs(Manager *s, FILE *f, const char *prefix);
402 void manager_dump(Manager *s, FILE *f, const char *prefix);
403 int manager_get_dump_string(Manager *m, char **ret);
404
405 void manager_clear_jobs(Manager *m);
406
407 unsigned manager_dispatch_load_queue(Manager *m);
408
409 int manager_environment_add(Manager *m, char **minus, char **plus);
410 int manager_set_default_rlimits(Manager *m, struct rlimit **default_rlimit);
411
412 int manager_loop(Manager *m);
413
414 int manager_open_serialization(Manager *m, FILE **_f);
415
416 int manager_serialize(Manager *m, FILE *f, FDSet *fds, bool switching_root);
417 int manager_deserialize(Manager *m, FILE *f, FDSet *fds);
418
419 int manager_reload(Manager *m);
420
421 void manager_reset_failed(Manager *m);
422
423 void manager_send_unit_audit(Manager *m, Unit *u, int type, bool success);
424 void manager_send_unit_plymouth(Manager *m, Unit *u);
425
426 bool manager_unit_inactive_or_pending(Manager *m, const char *name);
427
428 void manager_check_finished(Manager *m);
429
430 void manager_recheck_dbus(Manager *m);
431 void manager_recheck_journal(Manager *m);
432
433 void manager_set_show_status(Manager *m, ShowStatus mode);
434 void manager_set_first_boot(Manager *m, bool b);
435
436 void manager_status_printf(Manager *m, StatusType type, const char *status, const char *format, ...) _printf_(4,5);
437 void manager_flip_auto_status(Manager *m, bool enable);
438
439 Set *manager_get_units_requiring_mounts_for(Manager *m, const char *path);
440
441 ManagerState manager_state(Manager *m);
442
443 int manager_update_failed_units(Manager *m, Unit *u, bool failed);
444
445 void manager_unref_uid(Manager *m, uid_t uid, bool destroy_now);
446 int manager_ref_uid(Manager *m, uid_t uid, bool clean_ipc);
447
448 void manager_unref_gid(Manager *m, gid_t gid, bool destroy_now);
449 int manager_ref_gid(Manager *m, gid_t gid, bool destroy_now);
450
451 void manager_vacuum_uid_refs(Manager *m);
452 void manager_vacuum_gid_refs(Manager *m);
453
454 void manager_serialize_uid_refs(Manager *m, FILE *f);
455 void manager_deserialize_uid_refs_one(Manager *m, const char *value);
456
457 void manager_serialize_gid_refs(Manager *m, FILE *f);
458 void manager_deserialize_gid_refs_one(Manager *m, const char *value);
459
460 char *manager_taint_string(Manager *m);
461
462 void manager_ref_console(Manager *m);
463 void manager_unref_console(Manager *m);
464
465 const char *manager_state_to_string(ManagerState m) _const_;
466 ManagerState manager_state_from_string(const char *s) _pure_;
467
468 const char *manager_get_confirm_spawn(Manager *m);
469 bool manager_is_confirm_spawn_disabled(Manager *m);
470 void manager_disable_confirm_spawn(void);
471
472 const char *manager_timestamp_to_string(ManagerTimestamp m) _const_;
473 ManagerTimestamp manager_timestamp_from_string(const char *s) _pure_;