]> git.ipfire.org Git - thirdparty/systemd.git/blob - src/core/manager.h
Merge pull request #7915 from poettering/pids-max-tweak
[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 UnitFileScope unit_file_scope;
186 LookupPaths lookup_paths;
187 Set *unit_path_cache;
188
189 char **environment;
190
191 usec_t runtime_watchdog;
192 usec_t shutdown_watchdog;
193
194 dual_timestamp timestamps[_MANAGER_TIMESTAMP_MAX];
195
196 struct udev* udev;
197
198 /* Data specific to the device subsystem */
199 struct udev_monitor* udev_monitor;
200 sd_event_source *udev_event_source;
201 Hashmap *devices_by_sysfs;
202
203 /* Data specific to the mount subsystem */
204 struct libmnt_monitor *mount_monitor;
205 sd_event_source *mount_event_source;
206
207 /* Data specific to the swap filesystem */
208 FILE *proc_swaps;
209 sd_event_source *swap_event_source;
210 Hashmap *swaps_by_devnode;
211
212 /* Data specific to the D-Bus subsystem */
213 sd_bus *api_bus, *system_bus;
214 Set *private_buses;
215 int private_listen_fd;
216 sd_event_source *private_listen_event_source;
217
218 /* Contains all the clients that are subscribed to signals via
219 the API bus. Note that private bus connections are always
220 considered subscribes, since they last for very short only,
221 and it is much simpler that way. */
222 sd_bus_track *subscribed;
223 char **deserialized_subscribed;
224
225 /* This is used during reloading: before the reload we queue
226 * the reply message here, and afterwards we send it */
227 sd_bus_message *queued_message;
228
229 Hashmap *watch_bus; /* D-Bus names => Unit object n:1 */
230
231 bool send_reloading_done;
232
233 uint32_t current_job_id;
234 uint32_t default_unit_job_id;
235
236 /* Data specific to the Automount subsystem */
237 int dev_autofs_fd;
238
239 /* Data specific to the cgroup subsystem */
240 Hashmap *cgroup_unit;
241 CGroupMask cgroup_supported;
242 char *cgroup_root;
243
244 /* Notifications from cgroups, when the unified hierarchy is used is done via inotify. */
245 int cgroup_inotify_fd;
246 sd_event_source *cgroup_inotify_event_source;
247 Hashmap *cgroup_inotify_wd_unit;
248
249 /* A defer event for handling cgroup empty events and processing them after SIGCHLD in all cases. */
250 sd_event_source *cgroup_empty_event_source;
251
252 /* Make sure the user cannot accidentally unmount our cgroup
253 * file system */
254 int pin_cgroupfs_fd;
255
256 unsigned gc_marker;
257
258 /* Flags */
259 ManagerExitCode exit_code:5;
260
261 bool dispatching_load_queue:1;
262 bool dispatching_dbus_queue:1;
263
264 bool taint_usr:1;
265
266 /* Have we already sent out the READY=1 notification? */
267 bool ready_sent:1;
268
269 /* Have we already printed the taint line if necessary? */
270 bool taint_logged:1;
271
272 /* Have we ever changed the "kernel.pid_max" sysctl? */
273 bool sysctl_pid_max_changed:1;
274
275 unsigned test_run_flags:8;
276
277 /* If non-zero, exit with the following value when the systemd
278 * process terminate. Useful for containers: systemd-nspawn could get
279 * the return value. */
280 uint8_t return_value;
281
282 ShowStatus show_status;
283 char *confirm_spawn;
284 bool no_console_output;
285 bool service_watchdogs;
286
287 ExecOutput default_std_output, default_std_error;
288
289 usec_t default_restart_usec, default_timeout_start_usec, default_timeout_stop_usec;
290
291 usec_t default_start_limit_interval;
292 unsigned default_start_limit_burst;
293
294 bool default_cpu_accounting;
295 bool default_memory_accounting;
296 bool default_io_accounting;
297 bool default_blockio_accounting;
298 bool default_tasks_accounting;
299 bool default_ip_accounting;
300
301 uint64_t default_tasks_max;
302 usec_t default_timer_accuracy_usec;
303
304 struct rlimit *rlimit[_RLIMIT_MAX];
305
306 /* non-zero if we are reloading or reexecuting, */
307 int n_reloading;
308
309 unsigned n_installed_jobs;
310 unsigned n_failed_jobs;
311
312 /* Jobs in progress watching */
313 unsigned n_running_jobs;
314 unsigned n_on_console;
315 unsigned jobs_in_progress_iteration;
316
317 /* Do we have any outstanding password prompts? */
318 int have_ask_password;
319 int ask_password_inotify_fd;
320 sd_event_source *ask_password_event_source;
321
322 /* Type=idle pipes */
323 int idle_pipe[4];
324 sd_event_source *idle_pipe_event_source;
325
326 char *switch_root;
327 char *switch_root_init;
328
329 /* This maps all possible path prefixes to the units needing
330 * them. It's a hashmap with a path string as key and a Set as
331 * value where Unit objects are contained. */
332 Hashmap *units_requiring_mounts_for;
333
334 /* Used for processing polkit authorization responses */
335 Hashmap *polkit_registry;
336
337 /* Dynamic users/groups, indexed by their name */
338 Hashmap *dynamic_users;
339
340 /* Keep track of all UIDs and GIDs any of our services currently use. This is useful for the RemoveIPC= logic. */
341 Hashmap *uid_refs;
342 Hashmap *gid_refs;
343
344 /* When the user hits C-A-D more than 7 times per 2s, do something immediately... */
345 RateLimit ctrl_alt_del_ratelimit;
346 EmergencyAction cad_burst_action;
347
348 const char *unit_log_field;
349 const char *unit_log_format_string;
350
351 const char *invocation_log_field;
352 const char *invocation_log_format_string;
353
354 int first_boot; /* tri-state */
355
356 /* Prefixes of e.g. RuntimeDirectory= */
357 char *prefix[_EXEC_DIRECTORY_TYPE_MAX];
358
359 /* Used in the SIGCHLD and sd_notify() message invocation logic to avoid that we dispatch the same event
360 * multiple times on the same unit. */
361 unsigned sigchldgen;
362 unsigned notifygen;
363 };
364
365 #define MANAGER_IS_SYSTEM(m) ((m)->unit_file_scope == UNIT_FILE_SYSTEM)
366 #define MANAGER_IS_USER(m) ((m)->unit_file_scope != UNIT_FILE_SYSTEM)
367
368 #define MANAGER_IS_RELOADING(m) ((m)->n_reloading > 0)
369
370 #define MANAGER_IS_FINISHED(m) (dual_timestamp_is_set((m)->timestamps + MANAGER_TIMESTAMP_FINISH))
371
372 /* 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 */
373 #define MANAGER_IS_RUNNING(m) ((m)->exit_code == MANAGER_OK)
374
375 int manager_new(UnitFileScope scope, unsigned test_run_flags, Manager **m);
376 Manager* manager_free(Manager *m);
377
378 void manager_enumerate(Manager *m);
379 int manager_startup(Manager *m, FILE *serialization, FDSet *fds);
380
381 Job *manager_get_job(Manager *m, uint32_t id);
382 Unit *manager_get_unit(Manager *m, const char *name);
383
384 int manager_get_job_from_dbus_path(Manager *m, const char *s, Job **_j);
385
386 int manager_load_unit_prepare(Manager *m, const char *name, const char *path, sd_bus_error *e, Unit **_ret);
387 int manager_load_unit(Manager *m, const char *name, const char *path, sd_bus_error *e, Unit **_ret);
388 int manager_load_unit_from_dbus_path(Manager *m, const char *s, sd_bus_error *e, Unit **_u);
389
390 int manager_add_job(Manager *m, JobType type, Unit *unit, JobMode mode, sd_bus_error *e, Job **_ret);
391 int manager_add_job_by_name(Manager *m, JobType type, const char *name, JobMode mode, sd_bus_error *e, Job **_ret);
392 int manager_add_job_by_name_and_warn(Manager *m, JobType type, const char *name, JobMode mode, Job **ret);
393 int manager_propagate_reload(Manager *m, Unit *unit, JobMode mode, sd_bus_error *e);
394
395 void manager_dump_units(Manager *s, FILE *f, const char *prefix);
396 void manager_dump_jobs(Manager *s, FILE *f, const char *prefix);
397 void manager_dump(Manager *s, FILE *f, const char *prefix);
398 int manager_get_dump_string(Manager *m, char **ret);
399
400 void manager_clear_jobs(Manager *m);
401
402 unsigned manager_dispatch_load_queue(Manager *m);
403
404 int manager_environment_add(Manager *m, char **minus, char **plus);
405 int manager_set_default_rlimits(Manager *m, struct rlimit **default_rlimit);
406
407 int manager_loop(Manager *m);
408
409 int manager_open_serialization(Manager *m, FILE **_f);
410
411 int manager_serialize(Manager *m, FILE *f, FDSet *fds, bool switching_root);
412 int manager_deserialize(Manager *m, FILE *f, FDSet *fds);
413
414 int manager_reload(Manager *m);
415
416 void manager_reset_failed(Manager *m);
417
418 void manager_send_unit_audit(Manager *m, Unit *u, int type, bool success);
419 void manager_send_unit_plymouth(Manager *m, Unit *u);
420
421 bool manager_unit_inactive_or_pending(Manager *m, const char *name);
422
423 void manager_check_finished(Manager *m);
424
425 void manager_recheck_journal(Manager *m);
426
427 void manager_set_show_status(Manager *m, ShowStatus mode);
428 void manager_set_first_boot(Manager *m, bool b);
429
430 void manager_status_printf(Manager *m, StatusType type, const char *status, const char *format, ...) _printf_(4,5);
431 void manager_flip_auto_status(Manager *m, bool enable);
432
433 Set *manager_get_units_requiring_mounts_for(Manager *m, const char *path);
434
435 void manager_set_exec_params(Manager *m, ExecParameters *p);
436
437 ManagerState manager_state(Manager *m);
438
439 int manager_update_failed_units(Manager *m, Unit *u, bool failed);
440
441 void manager_unref_uid(Manager *m, uid_t uid, bool destroy_now);
442 int manager_ref_uid(Manager *m, uid_t uid, bool clean_ipc);
443
444 void manager_unref_gid(Manager *m, gid_t gid, bool destroy_now);
445 int manager_ref_gid(Manager *m, gid_t gid, bool destroy_now);
446
447 void manager_vacuum_uid_refs(Manager *m);
448 void manager_vacuum_gid_refs(Manager *m);
449
450 void manager_serialize_uid_refs(Manager *m, FILE *f);
451 void manager_deserialize_uid_refs_one(Manager *m, const char *value);
452
453 void manager_serialize_gid_refs(Manager *m, FILE *f);
454 void manager_deserialize_gid_refs_one(Manager *m, const char *value);
455
456 char *manager_taint_string(Manager *m);
457
458 const char *manager_state_to_string(ManagerState m) _const_;
459 ManagerState manager_state_from_string(const char *s) _pure_;
460
461 const char *manager_get_confirm_spawn(Manager *m);
462 bool manager_is_confirm_spawn_disabled(Manager *m);
463 void manager_disable_confirm_spawn(void);
464
465 const char *manager_timestamp_to_string(ManagerTimestamp m) _const_;
466 ManagerTimestamp manager_timestamp_from_string(const char *s) _pure_;