]> git.ipfire.org Git - thirdparty/systemd.git/blob - src/core/main.c
projects / thirdparty / systemd.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
systemd-analyze: add --root option for 'verify' verb and allow path parsing
[thirdparty/systemd.git] / src / core / main.c
1 /* SPDX-License-Identifier: LGPL-2.1-or-later */
2
3 #include <errno.h>
4 #include <fcntl.h>
5 #include <getopt.h>
6 #include <sys/mount.h>
7 #include <sys/prctl.h>
8 #include <sys/reboot.h>
9 #include <unistd.h>
10 #if HAVE_SECCOMP
11 #include <seccomp.h>
12 #endif
13 #if HAVE_VALGRIND_VALGRIND_H
14 #include <valgrind/valgrind.h>
15 #endif
16
17 #include "sd-bus.h"
18 #include "sd-daemon.h"
19 #include "sd-messages.h"
20
21 #include "alloc-util.h"
22 #include "apparmor-setup.h"
23 #include "architecture.h"
24 #include "build.h"
25 #include "bus-error.h"
26 #include "bus-util.h"
27 #include "capability-util.h"
28 #include "cgroup-util.h"
29 #include "clock-util.h"
30 #include "conf-parser.h"
31 #include "cpu-set-util.h"
32 #include "dbus-manager.h"
33 #include "dbus.h"
34 #include "def.h"
35 #include "dev-setup.h"
36 #include "efi-random.h"
37 #include "efivars.h"
38 #include "emergency-action.h"
39 #include "env-util.h"
40 #include "exit-status.h"
41 #include "fd-util.h"
42 #include "fdset.h"
43 #include "fileio.h"
44 #include "format-util.h"
45 #include "fs-util.h"
46 #include "hexdecoct.h"
47 #include "hostname-setup.h"
48 #include "ima-setup.h"
49 #include "killall.h"
50 #include "kmod-setup.h"
51 #include "limits-util.h"
52 #include "load-fragment.h"
53 #include "log.h"
54 #include "loopback-setup.h"
55 #include "machine-id-setup.h"
56 #include "manager.h"
57 #include "manager-dump.h"
58 #include "manager-serialize.h"
59 #include "mkdir.h"
60 #include "mount-setup.h"
61 #include "os-util.h"
62 #include "pager.h"
63 #include "parse-argument.h"
64 #include "parse-util.h"
65 #include "path-util.h"
66 #include "pretty-print.h"
67 #include "proc-cmdline.h"
68 #include "process-util.h"
69 #include "random-util.h"
70 #include "raw-clone.h"
71 #include "rlimit-util.h"
72 #if HAVE_SECCOMP
73 #include "seccomp-util.h"
74 #endif
75 #include "selinux-setup.h"
76 #include "selinux-util.h"
77 #include "signal-util.h"
78 #include "smack-setup.h"
79 #include "special.h"
80 #include "stat-util.h"
81 #include "stdio-util.h"
82 #include "strv.h"
83 #include "switch-root.h"
84 #include "sysctl-util.h"
85 #include "terminal-util.h"
86 #include "time-util.h"
87 #include "umask-util.h"
88 #include "user-util.h"
89 #include "util.h"
90 #include "virt.h"
91 #include "watchdog.h"
92
93 #if HAS_FEATURE_ADDRESS_SANITIZER
94 #include <sanitizer/lsan_interface.h>
95 #endif
96
97 #define DEFAULT_TASKS_MAX ((TasksMax) { 15U, 100U }) /* 15% */
98
99 static enum {
100 ACTION_RUN,
101 ACTION_HELP,
102 ACTION_VERSION,
103 ACTION_TEST,
104 ACTION_DUMP_CONFIGURATION_ITEMS,
105 ACTION_DUMP_BUS_PROPERTIES,
106 ACTION_BUS_INTROSPECT,
107 } arg_action = ACTION_RUN;
108
109 static const char *arg_bus_introspect = NULL;
110
111 /* Those variables are initialized to 0 automatically, so we avoid uninitialized memory access. Real
112 * defaults are assigned in reset_arguments() below. */
113 static char *arg_default_unit;
114 static bool arg_system;
115 static bool arg_dump_core;
116 static int arg_crash_chvt;
117 static bool arg_crash_shell;
118 static bool arg_crash_reboot;
119 static char *arg_confirm_spawn;
120 static ShowStatus arg_show_status;
121 static StatusUnitFormat arg_status_unit_format;
122 static bool arg_switched_root;
123 static PagerFlags arg_pager_flags;
124 static bool arg_service_watchdogs;
125 static ExecOutput arg_default_std_output;
126 static ExecOutput arg_default_std_error;
127 static usec_t arg_default_restart_usec;
128 static usec_t arg_default_timeout_start_usec;
129 static usec_t arg_default_timeout_stop_usec;
130 static usec_t arg_default_timeout_abort_usec;
131 static bool arg_default_timeout_abort_set;
132 static usec_t arg_default_start_limit_interval;
133 static unsigned arg_default_start_limit_burst;
134 static usec_t arg_runtime_watchdog;
135 static usec_t arg_reboot_watchdog;
136 static usec_t arg_kexec_watchdog;
137 static char *arg_early_core_pattern;
138 static char *arg_watchdog_device;
139 static char **arg_default_environment;
140 static char **arg_manager_environment;
141 static struct rlimit *arg_default_rlimit[_RLIMIT_MAX];
142 static uint64_t arg_capability_bounding_set;
143 static bool arg_no_new_privs;
144 static nsec_t arg_timer_slack_nsec;
145 static usec_t arg_default_timer_accuracy_usec;
146 static Set* arg_syscall_archs;
147 static FILE* arg_serialization;
148 static int arg_default_cpu_accounting;
149 static bool arg_default_io_accounting;
150 static bool arg_default_ip_accounting;
151 static bool arg_default_blockio_accounting;
152 static bool arg_default_memory_accounting;
153 static bool arg_default_tasks_accounting;
154 static TasksMax arg_default_tasks_max;
155 static sd_id128_t arg_machine_id;
156 static EmergencyAction arg_cad_burst_action;
157 static OOMPolicy arg_default_oom_policy;
158 static CPUSet arg_cpu_affinity;
159 static NUMAPolicy arg_numa_policy;
160 static usec_t arg_clock_usec;
161 static void *arg_random_seed;
162 static size_t arg_random_seed_size;
163
164 /* A copy of the original environment block */
165 static char **saved_env = NULL;
166
167 static int parse_configuration(const struct rlimit *saved_rlimit_nofile,
168 const struct rlimit *saved_rlimit_memlock);
169
170 static int manager_find_user_config_paths(char ***ret_files, char ***ret_dirs) {
171 _cleanup_free_ char *base = NULL;
172 _cleanup_strv_free_ char **files = NULL, **dirs = NULL;
173 int r;
174
175 r = xdg_user_config_dir(&base, "/systemd");
176 if (r < 0)
177 return r;
178
179 r = strv_extendf(&files, "%s/user.conf", base);
180 if (r < 0)
181 return r;
182
183 r = strv_extend(&files, PKGSYSCONFDIR "/user.conf");
184 if (r < 0)
185 return r;
186
187 r = strv_consume(&dirs, TAKE_PTR(base));
188 if (r < 0)
189 return r;
190
191 r = strv_extend_strv(&dirs, CONF_PATHS_STRV("systemd"), false);
192 if (r < 0)
193 return r;
194
195 *ret_files = TAKE_PTR(files);
196 *ret_dirs = TAKE_PTR(dirs);
197 return 0;
198 }
199
200 _noreturn_ static void freeze_or_exit_or_reboot(void) {
201
202 /* If we are running in a container, let's prefer exiting, after all we can propagate an exit code to
203 * the container manager, and thus inform it that something went wrong. */
204 if (detect_container() > 0) {
205 log_emergency("Exiting PID 1...");
206 _exit(EXIT_EXCEPTION);
207 }
208
209 if (arg_crash_reboot) {
210 log_notice("Rebooting in 10s...");
211 (void) sleep(10);
212
213 log_notice("Rebooting now...");
214 (void) reboot(RB_AUTOBOOT);
215 log_emergency_errno(errno, "Failed to reboot: %m");
216 }
217
218 log_emergency("Freezing execution.");
219 freeze();
220 }
221
222 _noreturn_ static void crash(int sig) {
223 struct sigaction sa;
224 pid_t pid;
225
226 if (getpid_cached() != 1)
227 /* Pass this on immediately, if this is not PID 1 */
228 (void) raise(sig);
229 else if (!arg_dump_core)
230 log_emergency("Caught <%s>, not dumping core.", signal_to_string(sig));
231 else {
232 sa = (struct sigaction) {
233 .sa_handler = nop_signal_handler,
234 .sa_flags = SA_NOCLDSTOP|SA_RESTART,
235 };
236
237 /* We want to wait for the core process, hence let's enable SIGCHLD */
238 (void) sigaction(SIGCHLD, &sa, NULL);
239
240 pid = raw_clone(SIGCHLD);
241 if (pid < 0)
242 log_emergency_errno(errno, "Caught <%s>, cannot fork for core dump: %m", signal_to_string(sig));
243 else if (pid == 0) {
244 /* Enable default signal handler for core dump */
245
246 sa = (struct sigaction) {
247 .sa_handler = SIG_DFL,
248 };
249 (void) sigaction(sig, &sa, NULL);
250
251 /* Don't limit the coredump size */
252 (void) setrlimit(RLIMIT_CORE, &RLIMIT_MAKE_CONST(RLIM_INFINITY));
253
254 /* Just to be sure... */
255 (void) chdir("/");
256
257 /* Raise the signal again */
258 pid = raw_getpid();
259 (void) kill(pid, sig); /* raise() would kill the parent */
260
261 assert_not_reached();
262 _exit(EXIT_EXCEPTION);
263 } else {
264 siginfo_t status;
265 int r;
266
267 /* Order things nicely. */
268 r = wait_for_terminate(pid, &status);
269 if (r < 0)
270 log_emergency_errno(r, "Caught <%s>, waitpid() failed: %m", signal_to_string(sig));
271 else if (status.si_code != CLD_DUMPED) {
272 const char *s = status.si_code == CLD_EXITED
273 ? exit_status_to_string(status.si_status, EXIT_STATUS_LIBC)
274 : signal_to_string(status.si_status);
275
276 log_emergency("Caught <%s>, core dump failed (child "PID_FMT", code=%s, status=%i/%s).",
277 signal_to_string(sig),
278 pid,
279 sigchld_code_to_string(status.si_code),
280 status.si_status, strna(s));
281 } else
282 log_emergency("Caught <%s>, dumped core as pid "PID_FMT".",
283 signal_to_string(sig), pid);
284 }
285 }
286
287 if (arg_crash_chvt >= 0)
288 (void) chvt(arg_crash_chvt);
289
290 sa = (struct sigaction) {
291 .sa_handler = SIG_IGN,
292 .sa_flags = SA_NOCLDSTOP|SA_NOCLDWAIT|SA_RESTART,
293 };
294
295 /* Let the kernel reap children for us */
296 (void) sigaction(SIGCHLD, &sa, NULL);
297
298 if (arg_crash_shell) {
299 log_notice("Executing crash shell in 10s...");
300 (void) sleep(10);
301
302 pid = raw_clone(SIGCHLD);
303 if (pid < 0)
304 log_emergency_errno(errno, "Failed to fork off crash shell: %m");
305 else if (pid == 0) {
306 (void) setsid();
307 (void) make_console_stdio();
308 (void) rlimit_nofile_safe();
309 (void) execle("/bin/sh", "/bin/sh", NULL, environ);
310
311 log_emergency_errno(errno, "execle() failed: %m");
312 _exit(EXIT_EXCEPTION);
313 } else {
314 log_info("Spawned crash shell as PID "PID_FMT".", pid);
315 (void) wait_for_terminate(pid, NULL);
316 }
317 }
318
319 freeze_or_exit_or_reboot();
320 }
321
322 static void install_crash_handler(void) {
323 static const struct sigaction sa = {
324 .sa_handler = crash,
325 .sa_flags = SA_NODEFER, /* So that we can raise the signal again from the signal handler */
326 };
327 int r;
328
329 /* We ignore the return value here, since, we don't mind if we cannot set up a crash handler */
330 r = sigaction_many(&sa, SIGNALS_CRASH_HANDLER);
331 if (r < 0)
332 log_debug_errno(r, "I had trouble setting up the crash handler, ignoring: %m");
333 }
334
335 static int console_setup(void) {
336 _cleanup_close_ int tty_fd = -1;
337 int r;
338
339 tty_fd = open_terminal("/dev/console", O_WRONLY|O_NOCTTY|O_CLOEXEC);
340 if (tty_fd < 0)
341 return log_error_errno(tty_fd, "Failed to open /dev/console: %m");
342
343 /* We don't want to force text mode. plymouth may be showing
344 * pictures already from initrd. */
345 r = reset_terminal_fd(tty_fd, false);
346 if (r < 0)
347 return log_error_errno(r, "Failed to reset /dev/console: %m");
348
349 return 0;
350 }
351
352 static int set_machine_id(const char *m) {
353 sd_id128_t t;
354 assert(m);
355
356 if (sd_id128_from_string(m, &t) < 0)
357 return -EINVAL;
358
359 if (sd_id128_is_null(t))
360 return -EINVAL;
361
362 arg_machine_id = t;
363 return 0;
364 }
365
366 static int parse_proc_cmdline_item(const char *key, const char *value, void *data) {
367 int r;
368
369 assert(key);
370
371 if (STR_IN_SET(key, "systemd.unit", "rd.systemd.unit")) {
372
373 if (proc_cmdline_value_missing(key, value))
374 return 0;
375
376 if (!unit_name_is_valid(value, UNIT_NAME_PLAIN|UNIT_NAME_INSTANCE))
377 log_warning("Unit name specified on %s= is not valid, ignoring: %s", key, value);
378 else if (in_initrd() == !!startswith(key, "rd."))
379 return free_and_strdup_warn(&arg_default_unit, value);
380
381 } else if (proc_cmdline_key_streq(key, "systemd.dump_core")) {
382
383 r = value ? parse_boolean(value) : true;
384 if (r < 0)
385 log_warning_errno(r, "Failed to parse dump core switch %s, ignoring: %m", value);
386 else
387 arg_dump_core = r;
388
389 } else if (proc_cmdline_key_streq(key, "systemd.early_core_pattern")) {
390
391 if (proc_cmdline_value_missing(key, value))
392 return 0;
393
394 if (path_is_absolute(value))
395 (void) parse_path_argument(value, false, &arg_early_core_pattern);
396 else
397 log_warning("Specified core pattern '%s' is not an absolute path, ignoring.", value);
398
399 } else if (proc_cmdline_key_streq(key, "systemd.crash_chvt")) {
400
401 if (!value)
402 arg_crash_chvt = 0; /* turn on */
403 else {
404 r = parse_crash_chvt(value, &arg_crash_chvt);
405 if (r < 0)
406 log_warning_errno(r, "Failed to parse crash chvt switch %s, ignoring: %m", value);
407 }
408
409 } else if (proc_cmdline_key_streq(key, "systemd.crash_shell")) {
410
411 r = value ? parse_boolean(value) : true;
412 if (r < 0)
413 log_warning_errno(r, "Failed to parse crash shell switch %s, ignoring: %m", value);
414 else
415 arg_crash_shell = r;
416
417 } else if (proc_cmdline_key_streq(key, "systemd.crash_reboot")) {
418
419 r = value ? parse_boolean(value) : true;
420 if (r < 0)
421 log_warning_errno(r, "Failed to parse crash reboot switch %s, ignoring: %m", value);
422 else
423 arg_crash_reboot = r;
424
425 } else if (proc_cmdline_key_streq(key, "systemd.confirm_spawn")) {
426 char *s;
427
428 r = parse_confirm_spawn(value, &s);
429 if (r < 0)
430 log_warning_errno(r, "Failed to parse confirm_spawn switch %s, ignoring: %m", value);
431 else
432 free_and_replace(arg_confirm_spawn, s);
433
434 } else if (proc_cmdline_key_streq(key, "systemd.service_watchdogs")) {
435
436 r = value ? parse_boolean(value) : true;
437 if (r < 0)
438 log_warning_errno(r, "Failed to parse service watchdog switch %s, ignoring: %m", value);
439 else
440 arg_service_watchdogs = r;
441
442 } else if (proc_cmdline_key_streq(key, "systemd.show_status")) {
443
444 if (value) {
445 r = parse_show_status(value, &arg_show_status);
446 if (r < 0)
447 log_warning_errno(r, "Failed to parse show status switch %s, ignoring: %m", value);
448 } else
449 arg_show_status = SHOW_STATUS_YES;
450
451 } else if (proc_cmdline_key_streq(key, "systemd.status_unit_format")) {
452
453 if (proc_cmdline_value_missing(key, value))
454 return 0;
455
456 r = status_unit_format_from_string(value);
457 if (r < 0)
458 log_warning_errno(r, "Failed to parse %s=%s, ignoring: %m", key, value);
459 else
460 arg_status_unit_format = r;
461
462 } else if (proc_cmdline_key_streq(key, "systemd.default_standard_output")) {
463
464 if (proc_cmdline_value_missing(key, value))
465 return 0;
466
467 r = exec_output_from_string(value);
468 if (r < 0)
469 log_warning_errno(r, "Failed to parse default standard output switch %s, ignoring: %m", value);
470 else
471 arg_default_std_output = r;
472
473 } else if (proc_cmdline_key_streq(key, "systemd.default_standard_error")) {
474
475 if (proc_cmdline_value_missing(key, value))
476 return 0;
477
478 r = exec_output_from_string(value);
479 if (r < 0)
480 log_warning_errno(r, "Failed to parse default standard error switch %s, ignoring: %m", value);
481 else
482 arg_default_std_error = r;
483
484 } else if (streq(key, "systemd.setenv")) {
485
486 if (proc_cmdline_value_missing(key, value))
487 return 0;
488
489 if (!env_assignment_is_valid(value))
490 log_warning("Environment variable assignment '%s' is not valid. Ignoring.", value);
491 else {
492 r = strv_env_replace_strdup(&arg_default_environment, value);
493 if (r < 0)
494 return log_oom();
495 }
496
497 } else if (proc_cmdline_key_streq(key, "systemd.machine_id")) {
498
499 if (proc_cmdline_value_missing(key, value))
500 return 0;
501
502 r = set_machine_id(value);
503 if (r < 0)
504 log_warning_errno(r, "MachineID '%s' is not valid, ignoring: %m", value);
505
506 } else if (proc_cmdline_key_streq(key, "systemd.default_timeout_start_sec")) {
507
508 if (proc_cmdline_value_missing(key, value))
509 return 0;
510
511 r = parse_sec(value, &arg_default_timeout_start_usec);
512 if (r < 0)
513 log_warning_errno(r, "Failed to parse default start timeout '%s', ignoring: %m", value);
514
515 if (arg_default_timeout_start_usec <= 0)
516 arg_default_timeout_start_usec = USEC_INFINITY;
517
518 } else if (proc_cmdline_key_streq(key, "systemd.cpu_affinity")) {
519
520 if (proc_cmdline_value_missing(key, value))
521 return 0;
522
523 r = parse_cpu_set(value, &arg_cpu_affinity);
524 if (r < 0)
525 log_warning_errno(r, "Failed to parse CPU affinity mask '%s', ignoring: %m", value);
526
527 } else if (proc_cmdline_key_streq(key, "systemd.watchdog_device")) {
528
529 if (proc_cmdline_value_missing(key, value))
530 return 0;
531
532 (void) parse_path_argument(value, false, &arg_watchdog_device);
533
534 } else if (proc_cmdline_key_streq(key, "systemd.clock_usec")) {
535
536 if (proc_cmdline_value_missing(key, value))
537 return 0;
538
539 r = safe_atou64(value, &arg_clock_usec);
540 if (r < 0)
541 log_warning_errno(r, "Failed to parse systemd.clock_usec= argument, ignoring: %s", value);
542
543 } else if (proc_cmdline_key_streq(key, "systemd.random_seed")) {
544 void *p;
545 size_t sz;
546
547 if (proc_cmdline_value_missing(key, value))
548 return 0;
549
550 r = unbase64mem(value, SIZE_MAX, &p, &sz);
551 if (r < 0)
552 log_warning_errno(r, "Failed to parse systemd.random_seed= argument, ignoring: %s", value);
553
554 free(arg_random_seed);
555 arg_random_seed = sz > 0 ? p : mfree(p);
556 arg_random_seed_size = sz;
557
558 } else if (streq(key, "quiet") && !value) {
559
560 if (arg_show_status == _SHOW_STATUS_INVALID)
561 arg_show_status = SHOW_STATUS_ERROR;
562
563 } else if (streq(key, "debug") && !value) {
564
565 /* Note that log_parse_environment() handles 'debug'
566 * too, and sets the log level to LOG_DEBUG. */
567
568 if (detect_container() > 0)
569 log_set_target(LOG_TARGET_CONSOLE);
570
571 } else if (!value) {
572 const char *target;
573
574 /* Compatible with SysV, but supported independently even if SysV compatibility is disabled. */
575 target = runlevel_to_target(key);
576 if (target)
577 return free_and_strdup_warn(&arg_default_unit, target);
578 }
579
580 return 0;
581 }
582
583 #define DEFINE_SETTER(name, func, descr) \
584 static int name(const char *unit, \
585 const char *filename, \
586 unsigned line, \
587 const char *section, \
588 unsigned section_line, \
589 const char *lvalue, \
590 int ltype, \
591 const char *rvalue, \
592 void *data, \
593 void *userdata) { \
594 \
595 int r; \
596 \
597 assert(filename); \
598 assert(lvalue); \
599 assert(rvalue); \
600 \
601 r = func(rvalue); \
602 if (r < 0) \
603 log_syntax(unit, LOG_ERR, filename, line, r, \
604 "Invalid " descr "'%s': %m", \
605 rvalue); \
606 \
607 return 0; \
608 }
609
610 DEFINE_SETTER(config_parse_level2, log_set_max_level_from_string, "log level");
611 DEFINE_SETTER(config_parse_target, log_set_target_from_string, "target");
612 DEFINE_SETTER(config_parse_color, log_show_color_from_string, "color");
613 DEFINE_SETTER(config_parse_location, log_show_location_from_string, "location");
614 DEFINE_SETTER(config_parse_time, log_show_time_from_string, "time");
615
616 static int config_parse_default_timeout_abort(
617 const char *unit,
618 const char *filename,
619 unsigned line,
620 const char *section,
621 unsigned section_line,
622 const char *lvalue,
623 int ltype,
624 const char *rvalue,
625 void *data,
626 void *userdata) {
627 int r;
628
629 r = config_parse_timeout_abort(unit, filename, line, section, section_line, lvalue, ltype, rvalue,
630 &arg_default_timeout_abort_usec, userdata);
631 if (r >= 0)
632 arg_default_timeout_abort_set = r;
633 return 0;
634 }
635
636 static int parse_config_file(void) {
637 const ConfigTableItem items[] = {
638 { "Manager", "LogLevel", config_parse_level2, 0, NULL },
639 { "Manager", "LogTarget", config_parse_target, 0, NULL },
640 { "Manager", "LogColor", config_parse_color, 0, NULL },
641 { "Manager", "LogLocation", config_parse_location, 0, NULL },
642 { "Manager", "LogTime", config_parse_time, 0, NULL },
643 { "Manager", "DumpCore", config_parse_bool, 0, &arg_dump_core },
644 { "Manager", "CrashChVT", /* legacy */ config_parse_crash_chvt, 0, &arg_crash_chvt },
645 { "Manager", "CrashChangeVT", config_parse_crash_chvt, 0, &arg_crash_chvt },
646 { "Manager", "CrashShell", config_parse_bool, 0, &arg_crash_shell },
647 { "Manager", "CrashReboot", config_parse_bool, 0, &arg_crash_reboot },
648 { "Manager", "ShowStatus", config_parse_show_status, 0, &arg_show_status },
649 { "Manager", "StatusUnitFormat", config_parse_status_unit_format, 0, &arg_status_unit_format },
650 { "Manager", "CPUAffinity", config_parse_cpu_affinity2, 0, &arg_cpu_affinity },
651 { "Manager", "NUMAPolicy", config_parse_numa_policy, 0, &arg_numa_policy.type },
652 { "Manager", "NUMAMask", config_parse_numa_mask, 0, &arg_numa_policy },
653 { "Manager", "JoinControllers", config_parse_warn_compat, DISABLED_CONFIGURATION, NULL },
654 { "Manager", "RuntimeWatchdogSec", config_parse_sec, 0, &arg_runtime_watchdog },
655 { "Manager", "RebootWatchdogSec", config_parse_sec, 0, &arg_reboot_watchdog },
656 { "Manager", "ShutdownWatchdogSec", config_parse_sec, 0, &arg_reboot_watchdog }, /* obsolete alias */
657 { "Manager", "KExecWatchdogSec", config_parse_sec, 0, &arg_kexec_watchdog },
658 { "Manager", "WatchdogDevice", config_parse_path, 0, &arg_watchdog_device },
659 { "Manager", "CapabilityBoundingSet", config_parse_capability_set, 0, &arg_capability_bounding_set },
660 { "Manager", "NoNewPrivileges", config_parse_bool, 0, &arg_no_new_privs },
661 #if HAVE_SECCOMP
662 { "Manager", "SystemCallArchitectures", config_parse_syscall_archs, 0, &arg_syscall_archs },
663 #endif
664 { "Manager", "TimerSlackNSec", config_parse_nsec, 0, &arg_timer_slack_nsec },
665 { "Manager", "DefaultTimerAccuracySec", config_parse_sec, 0, &arg_default_timer_accuracy_usec },
666 { "Manager", "DefaultStandardOutput", config_parse_output_restricted, 0, &arg_default_std_output },
667 { "Manager", "DefaultStandardError", config_parse_output_restricted, 0, &arg_default_std_error },
668 { "Manager", "DefaultTimeoutStartSec", config_parse_sec, 0, &arg_default_timeout_start_usec },
669 { "Manager", "DefaultTimeoutStopSec", config_parse_sec, 0, &arg_default_timeout_stop_usec },
670 { "Manager", "DefaultTimeoutAbortSec", config_parse_default_timeout_abort, 0, NULL },
671 { "Manager", "DefaultRestartSec", config_parse_sec, 0, &arg_default_restart_usec },
672 { "Manager", "DefaultStartLimitInterval", config_parse_sec, 0, &arg_default_start_limit_interval }, /* obsolete alias */
673 { "Manager", "DefaultStartLimitIntervalSec", config_parse_sec, 0, &arg_default_start_limit_interval },
674 { "Manager", "DefaultStartLimitBurst", config_parse_unsigned, 0, &arg_default_start_limit_burst },
675 { "Manager", "DefaultEnvironment", config_parse_environ, 0, &arg_default_environment },
676 { "Manager", "ManagerEnvironment", config_parse_environ, 0, &arg_manager_environment },
677 { "Manager", "DefaultLimitCPU", config_parse_rlimit, RLIMIT_CPU, arg_default_rlimit },
678 { "Manager", "DefaultLimitFSIZE", config_parse_rlimit, RLIMIT_FSIZE, arg_default_rlimit },
679 { "Manager", "DefaultLimitDATA", config_parse_rlimit, RLIMIT_DATA, arg_default_rlimit },
680 { "Manager", "DefaultLimitSTACK", config_parse_rlimit, RLIMIT_STACK, arg_default_rlimit },
681 { "Manager", "DefaultLimitCORE", config_parse_rlimit, RLIMIT_CORE, arg_default_rlimit },
682 { "Manager", "DefaultLimitRSS", config_parse_rlimit, RLIMIT_RSS, arg_default_rlimit },
683 { "Manager", "DefaultLimitNOFILE", config_parse_rlimit, RLIMIT_NOFILE, arg_default_rlimit },
684 { "Manager", "DefaultLimitAS", config_parse_rlimit, RLIMIT_AS, arg_default_rlimit },
685 { "Manager", "DefaultLimitNPROC", config_parse_rlimit, RLIMIT_NPROC, arg_default_rlimit },
686 { "Manager", "DefaultLimitMEMLOCK", config_parse_rlimit, RLIMIT_MEMLOCK, arg_default_rlimit },
687 { "Manager", "DefaultLimitLOCKS", config_parse_rlimit, RLIMIT_LOCKS, arg_default_rlimit },
688 { "Manager", "DefaultLimitSIGPENDING", config_parse_rlimit, RLIMIT_SIGPENDING, arg_default_rlimit },
689 { "Manager", "DefaultLimitMSGQUEUE", config_parse_rlimit, RLIMIT_MSGQUEUE, arg_default_rlimit },
690 { "Manager", "DefaultLimitNICE", config_parse_rlimit, RLIMIT_NICE, arg_default_rlimit },
691 { "Manager", "DefaultLimitRTPRIO", config_parse_rlimit, RLIMIT_RTPRIO, arg_default_rlimit },
692 { "Manager", "DefaultLimitRTTIME", config_parse_rlimit, RLIMIT_RTTIME, arg_default_rlimit },
693 { "Manager", "DefaultCPUAccounting", config_parse_tristate, 0, &arg_default_cpu_accounting },
694 { "Manager", "DefaultIOAccounting", config_parse_bool, 0, &arg_default_io_accounting },
695 { "Manager", "DefaultIPAccounting", config_parse_bool, 0, &arg_default_ip_accounting },
696 { "Manager", "DefaultBlockIOAccounting", config_parse_bool, 0, &arg_default_blockio_accounting },
697 { "Manager", "DefaultMemoryAccounting", config_parse_bool, 0, &arg_default_memory_accounting },
698 { "Manager", "DefaultTasksAccounting", config_parse_bool, 0, &arg_default_tasks_accounting },
699 { "Manager", "DefaultTasksMax", config_parse_tasks_max, 0, &arg_default_tasks_max },
700 { "Manager", "CtrlAltDelBurstAction", config_parse_emergency_action, 0, &arg_cad_burst_action },
701 { "Manager", "DefaultOOMPolicy", config_parse_oom_policy, 0, &arg_default_oom_policy },
702 {}
703 };
704
705 _cleanup_strv_free_ char **files = NULL, **dirs = NULL;
706 const char *suffix;
707 int r;
708
709 if (arg_system)
710 suffix = "system.conf.d";
711 else {
712 r = manager_find_user_config_paths(&files, &dirs);
713 if (r < 0)
714 return log_error_errno(r, "Failed to determine config file paths: %m");
715
716 suffix = "user.conf.d";
717 }
718
719 (void) config_parse_many(
720 (const char* const*) (files ?: STRV_MAKE(PKGSYSCONFDIR "/system.conf")),
721 (const char* const*) (dirs ?: CONF_PATHS_STRV("systemd")),
722 suffix,
723 "Manager\0",
724 config_item_table_lookup, items,
725 CONFIG_PARSE_WARN,
726 NULL,
727 NULL);
728
729 /* Traditionally "0" was used to turn off the default unit timeouts. Fix this up so that we use
730 * USEC_INFINITY like everywhere else. */
731 if (arg_default_timeout_start_usec <= 0)
732 arg_default_timeout_start_usec = USEC_INFINITY;
733 if (arg_default_timeout_stop_usec <= 0)
734 arg_default_timeout_stop_usec = USEC_INFINITY;
735
736 return 0;
737 }
738
739 static void set_manager_defaults(Manager *m) {
740
741 assert(m);
742
743 /* Propagates the various default unit property settings into the manager object, i.e. properties that do not
744 * affect the manager itself, but are just what newly allocated units will have set if they haven't set
745 * anything else. (Also see set_manager_settings() for the settings that affect the manager's own behaviour) */
746
747 m->default_timer_accuracy_usec = arg_default_timer_accuracy_usec;
748 m->default_std_output = arg_default_std_output;
749 m->default_std_error = arg_default_std_error;
750 m->default_timeout_start_usec = arg_default_timeout_start_usec;
751 m->default_timeout_stop_usec = arg_default_timeout_stop_usec;
752 m->default_timeout_abort_usec = arg_default_timeout_abort_usec;
753 m->default_timeout_abort_set = arg_default_timeout_abort_set;
754 m->default_restart_usec = arg_default_restart_usec;
755 m->default_start_limit_interval = arg_default_start_limit_interval;
756 m->default_start_limit_burst = arg_default_start_limit_burst;
757
758 /* On 4.15+ with unified hierarchy, CPU accounting is essentially free as it doesn't require the CPU
759 * controller to be enabled, so the default is to enable it unless we got told otherwise. */
760 if (arg_default_cpu_accounting >= 0)
761 m->default_cpu_accounting = arg_default_cpu_accounting;
762 else
763 m->default_cpu_accounting = cpu_accounting_is_cheap();
764
765 m->default_io_accounting = arg_default_io_accounting;
766 m->default_ip_accounting = arg_default_ip_accounting;
767 m->default_blockio_accounting = arg_default_blockio_accounting;
768 m->default_memory_accounting = arg_default_memory_accounting;
769 m->default_tasks_accounting = arg_default_tasks_accounting;
770 m->default_tasks_max = arg_default_tasks_max;
771 m->default_oom_policy = arg_default_oom_policy;
772
773 (void) manager_set_default_rlimits(m, arg_default_rlimit);
774
775 (void) manager_default_environment(m);
776 (void) manager_transient_environment_add(m, arg_default_environment);
777 }
778
779 static void set_manager_settings(Manager *m) {
780
781 assert(m);
782
783 /* Propagates the various manager settings into the manager object, i.e. properties that
784 * effect the manager itself (as opposed to just being inherited into newly allocated
785 * units, see set_manager_defaults() above). */
786
787 m->confirm_spawn = arg_confirm_spawn;
788 m->service_watchdogs = arg_service_watchdogs;
789 m->cad_burst_action = arg_cad_burst_action;
790
791 manager_set_watchdog(m, WATCHDOG_RUNTIME, arg_runtime_watchdog);
792 manager_set_watchdog(m, WATCHDOG_REBOOT, arg_reboot_watchdog);
793 manager_set_watchdog(m, WATCHDOG_KEXEC, arg_kexec_watchdog);
794
795 manager_set_show_status(m, arg_show_status, "commandline");
796 m->status_unit_format = arg_status_unit_format;
797 }
798
799 static int parse_argv(int argc, char *argv[]) {
800 enum {
801 ARG_LOG_LEVEL = 0x100,
802 ARG_LOG_TARGET,
803 ARG_LOG_COLOR,
804 ARG_LOG_LOCATION,
805 ARG_LOG_TIME,
806 ARG_UNIT,
807 ARG_SYSTEM,
808 ARG_USER,
809 ARG_TEST,
810 ARG_NO_PAGER,
811 ARG_VERSION,
812 ARG_DUMP_CONFIGURATION_ITEMS,
813 ARG_DUMP_BUS_PROPERTIES,
814 ARG_BUS_INTROSPECT,
815 ARG_DUMP_CORE,
816 ARG_CRASH_CHVT,
817 ARG_CRASH_SHELL,
818 ARG_CRASH_REBOOT,
819 ARG_CONFIRM_SPAWN,
820 ARG_SHOW_STATUS,
821 ARG_DESERIALIZE,
822 ARG_SWITCHED_ROOT,
823 ARG_DEFAULT_STD_OUTPUT,
824 ARG_DEFAULT_STD_ERROR,
825 ARG_MACHINE_ID,
826 ARG_SERVICE_WATCHDOGS,
827 };
828
829 static const struct option options[] = {
830 { "log-level", required_argument, NULL, ARG_LOG_LEVEL },
831 { "log-target", required_argument, NULL, ARG_LOG_TARGET },
832 { "log-color", optional_argument, NULL, ARG_LOG_COLOR },
833 { "log-location", optional_argument, NULL, ARG_LOG_LOCATION },
834 { "log-time", optional_argument, NULL, ARG_LOG_TIME },
835 { "unit", required_argument, NULL, ARG_UNIT },
836 { "system", no_argument, NULL, ARG_SYSTEM },
837 { "user", no_argument, NULL, ARG_USER },
838 { "test", no_argument, NULL, ARG_TEST },
839 { "no-pager", no_argument, NULL, ARG_NO_PAGER },
840 { "help", no_argument, NULL, 'h' },
841 { "version", no_argument, NULL, ARG_VERSION },
842 { "dump-configuration-items", no_argument, NULL, ARG_DUMP_CONFIGURATION_ITEMS },
843 { "dump-bus-properties", no_argument, NULL, ARG_DUMP_BUS_PROPERTIES },
844 { "bus-introspect", required_argument, NULL, ARG_BUS_INTROSPECT },
845 { "dump-core", optional_argument, NULL, ARG_DUMP_CORE },
846 { "crash-chvt", required_argument, NULL, ARG_CRASH_CHVT },
847 { "crash-shell", optional_argument, NULL, ARG_CRASH_SHELL },
848 { "crash-reboot", optional_argument, NULL, ARG_CRASH_REBOOT },
849 { "confirm-spawn", optional_argument, NULL, ARG_CONFIRM_SPAWN },
850 { "show-status", optional_argument, NULL, ARG_SHOW_STATUS },
851 { "deserialize", required_argument, NULL, ARG_DESERIALIZE },
852 { "switched-root", no_argument, NULL, ARG_SWITCHED_ROOT },
853 { "default-standard-output", required_argument, NULL, ARG_DEFAULT_STD_OUTPUT, },
854 { "default-standard-error", required_argument, NULL, ARG_DEFAULT_STD_ERROR, },
855 { "machine-id", required_argument, NULL, ARG_MACHINE_ID },
856 { "service-watchdogs", required_argument, NULL, ARG_SERVICE_WATCHDOGS },
857 {}
858 };
859
860 int c, r;
861 bool user_arg_seen = false;
862
863 assert(argc >= 1);
864 assert(argv);
865
866 if (getpid_cached() == 1)
867 opterr = 0;
868
869 while ((c = getopt_long(argc, argv, "hDbsz:", options, NULL)) >= 0)
870
871 switch (c) {
872
873 case ARG_LOG_LEVEL:
874 r = log_set_max_level_from_string(optarg);
875 if (r < 0)
876 return log_error_errno(r, "Failed to parse log level \"%s\": %m", optarg);
877
878 break;
879
880 case ARG_LOG_TARGET:
881 r = log_set_target_from_string(optarg);
882 if (r < 0)
883 return log_error_errno(r, "Failed to parse log target \"%s\": %m", optarg);
884
885 break;
886
887 case ARG_LOG_COLOR:
888
889 if (optarg) {
890 r = log_show_color_from_string(optarg);
891 if (r < 0)
892 return log_error_errno(r, "Failed to parse log color setting \"%s\": %m",
893 optarg);
894 } else
895 log_show_color(true);
896
897 break;
898
899 case ARG_LOG_LOCATION:
900 if (optarg) {
901 r = log_show_location_from_string(optarg);
902 if (r < 0)
903 return log_error_errno(r, "Failed to parse log location setting \"%s\": %m",
904 optarg);
905 } else
906 log_show_location(true);
907
908 break;
909
910 case ARG_LOG_TIME:
911
912 if (optarg) {
913 r = log_show_time_from_string(optarg);
914 if (r < 0)
915 return log_error_errno(r, "Failed to parse log time setting \"%s\": %m",
916 optarg);
917 } else
918 log_show_time(true);
919
920 break;
921
922 case ARG_DEFAULT_STD_OUTPUT:
923 r = exec_output_from_string(optarg);
924 if (r < 0)
925 return log_error_errno(r, "Failed to parse default standard output setting \"%s\": %m",
926 optarg);
927 arg_default_std_output = r;
928 break;
929
930 case ARG_DEFAULT_STD_ERROR:
931 r = exec_output_from_string(optarg);
932 if (r < 0)
933 return log_error_errno(r, "Failed to parse default standard error output setting \"%s\": %m",
934 optarg);
935 arg_default_std_error = r;
936 break;
937
938 case ARG_UNIT:
939 r = free_and_strdup(&arg_default_unit, optarg);
940 if (r < 0)
941 return log_error_errno(r, "Failed to set default unit \"%s\": %m", optarg);
942
943 break;
944
945 case ARG_SYSTEM:
946 arg_system = true;
947 break;
948
949 case ARG_USER:
950 arg_system = false;
951 user_arg_seen = true;
952 break;
953
954 case ARG_TEST:
955 arg_action = ACTION_TEST;
956 break;
957
958 case ARG_NO_PAGER:
959 arg_pager_flags |= PAGER_DISABLE;
960 break;
961
962 case ARG_VERSION:
963 arg_action = ACTION_VERSION;
964 break;
965
966 case ARG_DUMP_CONFIGURATION_ITEMS:
967 arg_action = ACTION_DUMP_CONFIGURATION_ITEMS;
968 break;
969
970 case ARG_DUMP_BUS_PROPERTIES:
971 arg_action = ACTION_DUMP_BUS_PROPERTIES;
972 break;
973
974 case ARG_BUS_INTROSPECT:
975 arg_bus_introspect = optarg;
976 arg_action = ACTION_BUS_INTROSPECT;
977 break;
978
979 case ARG_DUMP_CORE:
980 r = parse_boolean_argument("--dump-core", optarg, &arg_dump_core);
981 if (r < 0)
982 return r;
983 break;
984
985 case ARG_CRASH_CHVT:
986 r = parse_crash_chvt(optarg, &arg_crash_chvt);
987 if (r < 0)
988 return log_error_errno(r, "Failed to parse crash virtual terminal index: \"%s\": %m",
989 optarg);
990 break;
991
992 case ARG_CRASH_SHELL:
993 r = parse_boolean_argument("--crash-shell", optarg, &arg_crash_shell);
994 if (r < 0)
995 return r;
996 break;
997
998 case ARG_CRASH_REBOOT:
999 r = parse_boolean_argument("--crash-reboot", optarg, &arg_crash_reboot);
1000 if (r < 0)
1001 return r;
1002 break;
1003
1004 case ARG_CONFIRM_SPAWN:
1005 arg_confirm_spawn = mfree(arg_confirm_spawn);
1006
1007 r = parse_confirm_spawn(optarg, &arg_confirm_spawn);
1008 if (r < 0)
1009 return log_error_errno(r, "Failed to parse confirm spawn option: \"%s\": %m",
1010 optarg);
1011 break;
1012
1013 case ARG_SERVICE_WATCHDOGS:
1014 r = parse_boolean_argument("--service-watchdogs=", optarg, &arg_service_watchdogs);
1015 if (r < 0)
1016 return r;
1017 break;
1018
1019 case ARG_SHOW_STATUS:
1020 if (optarg) {
1021 r = parse_show_status(optarg, &arg_show_status);
1022 if (r < 0)
1023 return log_error_errno(r, "Failed to parse show status boolean: \"%s\": %m",
1024 optarg);
1025 } else
1026 arg_show_status = SHOW_STATUS_YES;
1027 break;
1028
1029 case ARG_DESERIALIZE: {
1030 int fd;
1031 FILE *f;
1032
1033 r = safe_atoi(optarg, &fd);
1034 if (r < 0)
1035 log_error_errno(r, "Failed to parse deserialize option \"%s\": %m", optarg);
1036 if (fd < 0)
1037 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
1038 "Invalid deserialize fd: %d",
1039 fd);
1040
1041 (void) fd_cloexec(fd, true);
1042
1043 f = fdopen(fd, "r");
1044 if (!f)
1045 return log_error_errno(errno, "Failed to open serialization fd %d: %m", fd);
1046
1047 safe_fclose(arg_serialization);
1048 arg_serialization = f;
1049
1050 break;
1051 }
1052
1053 case ARG_SWITCHED_ROOT:
1054 arg_switched_root = true;
1055 break;
1056
1057 case ARG_MACHINE_ID:
1058 r = set_machine_id(optarg);
1059 if (r < 0)
1060 return log_error_errno(r, "MachineID '%s' is not valid: %m", optarg);
1061 break;
1062
1063 case 'h':
1064 arg_action = ACTION_HELP;
1065 break;
1066
1067 case 'D':
1068 log_set_max_level(LOG_DEBUG);
1069 break;
1070
1071 case 'b':
1072 case 's':
1073 case 'z':
1074 /* Just to eat away the sysvinit kernel cmdline args that we'll parse in
1075 * parse_proc_cmdline_item() or ignore, without any getopt() error messages.
1076 */
1077 case '?':
1078 if (getpid_cached() != 1)
1079 return -EINVAL;
1080 else
1081 return 0;
1082
1083 default:
1084 assert_not_reached();
1085 }
1086
1087 if (optind < argc && getpid_cached() != 1)
1088 /* Hmm, when we aren't run as init system let's complain about excess arguments */
1089 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Excess arguments.");
1090
1091 if (arg_action == ACTION_RUN && !arg_system && !user_arg_seen)
1092 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
1093 "Explicit --user argument required to run as user manager.");
1094
1095 return 0;
1096 }
1097
1098 static int help(void) {
1099 _cleanup_free_ char *link = NULL;
1100 int r;
1101
1102 r = terminal_urlify_man("systemd", "1", &link);
1103 if (r < 0)
1104 return log_oom();
1105
1106 printf("%s [OPTIONS...]\n\n"
1107 "%sStarts and monitors system and user services.%s\n\n"
1108 "This program takes no positional arguments.\n\n"
1109 "%sOptions%s:\n"
1110 " -h --help Show this help\n"
1111 " --version Show version\n"
1112 " --test Determine initial transaction, dump it and exit\n"
1113 " --system In combination with --test: operate as system service manager\n"
1114 " --user In combination with --test: operate as per-user service manager\n"
1115 " --no-pager Do not pipe output into a pager\n"
1116 " --dump-configuration-items Dump understood unit configuration items\n"
1117 " --dump-bus-properties Dump exposed bus properties\n"
1118 " --bus-introspect=PATH Write XML introspection data\n"
1119 " --unit=UNIT Set default unit\n"
1120 " --dump-core[=BOOL] Dump core on crash\n"
1121 " --crash-vt=NR Change to specified VT on crash\n"
1122 " --crash-reboot[=BOOL] Reboot on crash\n"
1123 " --crash-shell[=BOOL] Run shell on crash\n"
1124 " --confirm-spawn[=BOOL] Ask for confirmation when spawning processes\n"
1125 " --show-status[=BOOL] Show status updates on the console during bootup\n"
1126 " --log-target=TARGET Set log target (console, journal, kmsg, journal-or-kmsg, null)\n"
1127 " --log-level=LEVEL Set log level (debug, info, notice, warning, err, crit, alert, emerg)\n"
1128 " --log-color[=BOOL] Highlight important log messages\n"
1129 " --log-location[=BOOL] Include code location in log messages\n"
1130 " --log-time[=BOOL] Prefix log messages with current time\n"
1131 " --default-standard-output= Set default standard output for services\n"
1132 " --default-standard-error= Set default standard error output for services\n"
1133 "\nSee the %s for details.\n",
1134 program_invocation_short_name,
1135 ansi_highlight(),
1136 ansi_normal(),
1137 ansi_underline(),
1138 ansi_normal(),
1139 link);
1140
1141 return 0;
1142 }
1143
1144 static int prepare_reexecute(
1145 Manager *m,
1146 FILE **ret_f,
1147 FDSet **ret_fds,
1148 bool switching_root) {
1149
1150 _cleanup_fdset_free_ FDSet *fds = NULL;
1151 _cleanup_fclose_ FILE *f = NULL;
1152 int r;
1153
1154 assert(m);
1155 assert(ret_f);
1156 assert(ret_fds);
1157
1158 r = manager_open_serialization(m, &f);
1159 if (r < 0)
1160 return log_error_errno(r, "Failed to create serialization file: %m");
1161
1162 /* Make sure nothing is really destructed when we shut down */
1163 m->n_reloading++;
1164 bus_manager_send_reloading(m, true);
1165
1166 fds = fdset_new();
1167 if (!fds)
1168 return log_oom();
1169
1170 r = manager_serialize(m, f, fds, switching_root);
1171 if (r < 0)
1172 return r;
1173
1174 if (fseeko(f, 0, SEEK_SET) == (off_t) -1)
1175 return log_error_errno(errno, "Failed to rewind serialization fd: %m");
1176
1177 r = fd_cloexec(fileno(f), false);
1178 if (r < 0)
1179 return log_error_errno(r, "Failed to disable O_CLOEXEC for serialization: %m");
1180
1181 r = fdset_cloexec(fds, false);
1182 if (r < 0)
1183 return log_error_errno(r, "Failed to disable O_CLOEXEC for serialization fds: %m");
1184
1185 *ret_f = TAKE_PTR(f);
1186 *ret_fds = TAKE_PTR(fds);
1187
1188 return 0;
1189 }
1190
1191 static void bump_file_max_and_nr_open(void) {
1192
1193 /* Let's bump fs.file-max and fs.nr_open to their respective maximums. On current kernels large
1194 * numbers of file descriptors are no longer a performance problem and their memory is properly
1195 * tracked by memcg, thus counting them and limiting them in another two layers of limits is
1196 * unnecessary and just complicates things. This function hence turns off 2 of the 4 levels of limits
1197 * on file descriptors, and makes RLIMIT_NOLIMIT (soft + hard) the only ones that really matter. */
1198
1199 #if BUMP_PROC_SYS_FS_FILE_MAX || BUMP_PROC_SYS_FS_NR_OPEN
1200 int r;
1201 #endif
1202
1203 #if BUMP_PROC_SYS_FS_FILE_MAX
1204 /* The maximum the kernel allows for this since 5.2 is LONG_MAX, use that. (Previously things were
1205 * different, but the operation would fail silently.) */
1206 r = sysctl_writef("fs/file-max", "%li\n", LONG_MAX);
1207 if (r < 0)
1208 log_full_errno(IN_SET(r, -EROFS, -EPERM, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, "Failed to bump fs.file-max, ignoring: %m");
1209 #endif
1210
1211 #if BUMP_PROC_SYS_FS_NR_OPEN
1212 int v = INT_MAX;
1213
1214 /* Argh! The kernel enforces maximum and minimum values on the fs.nr_open, but we don't really know
1215 * what they are. The expression by which the maximum is determined is dependent on the architecture,
1216 * and is something we don't really want to copy to userspace, as it is dependent on implementation
1217 * details of the kernel. Since the kernel doesn't expose the maximum value to us, we can only try
1218 * and hope. Hence, let's start with INT_MAX, and then keep halving the value until we find one that
1219 * works. Ugly? Yes, absolutely, but kernel APIs are kernel APIs, so what do can we do... 🤯 */
1220
1221 for (;;) {
1222 int k;
1223
1224 v &= ~(__SIZEOF_POINTER__ - 1); /* Round down to next multiple of the pointer size */
1225 if (v < 1024) {
1226 log_warning("Can't bump fs.nr_open, value too small.");
1227 break;
1228 }
1229
1230 k = read_nr_open();
1231 if (k < 0) {
1232 log_error_errno(k, "Failed to read fs.nr_open: %m");
1233 break;
1234 }
1235 if (k >= v) { /* Already larger */
1236 log_debug("Skipping bump, value is already larger.");
1237 break;
1238 }
1239
1240 r = sysctl_writef("fs/nr_open", "%i\n", v);
1241 if (r == -EINVAL) {
1242 log_debug("Couldn't write fs.nr_open as %i, halving it.", v);
1243 v /= 2;
1244 continue;
1245 }
1246 if (r < 0) {
1247 log_full_errno(IN_SET(r, -EROFS, -EPERM, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, "Failed to bump fs.nr_open, ignoring: %m");
1248 break;
1249 }
1250
1251 log_debug("Successfully bumped fs.nr_open to %i", v);
1252 break;
1253 }
1254 #endif
1255 }
1256
1257 static int bump_rlimit_nofile(struct rlimit *saved_rlimit) {
1258 struct rlimit new_rlimit;
1259 int r, nr;
1260
1261 /* Get the underlying absolute limit the kernel enforces */
1262 nr = read_nr_open();
1263
1264 /* Calculate the new limits to use for us. Never lower from what we inherited. */
1265 new_rlimit = (struct rlimit) {
1266 .rlim_cur = MAX((rlim_t) nr, saved_rlimit->rlim_cur),
1267 .rlim_max = MAX((rlim_t) nr, saved_rlimit->rlim_max),
1268 };
1269
1270 /* Shortcut if nothing changes. */
1271 if (saved_rlimit->rlim_max >= new_rlimit.rlim_max &&
1272 saved_rlimit->rlim_cur >= new_rlimit.rlim_cur) {
1273 log_debug("RLIMIT_NOFILE is already as high or higher than we need it, not bumping.");
1274 return 0;
1275 }
1276
1277 /* Bump up the resource limit for ourselves substantially, all the way to the maximum the kernel allows, for
1278 * both hard and soft. */
1279 r = setrlimit_closest(RLIMIT_NOFILE, &new_rlimit);
1280 if (r < 0)
1281 return log_warning_errno(r, "Setting RLIMIT_NOFILE failed, ignoring: %m");
1282
1283 return 0;
1284 }
1285
1286 static int bump_rlimit_memlock(struct rlimit *saved_rlimit) {
1287 struct rlimit new_rlimit;
1288 uint64_t mm;
1289 int r;
1290
1291 /* BPF_MAP_TYPE_LPM_TRIE bpf maps are charged against RLIMIT_MEMLOCK, even if we have CAP_IPC_LOCK
1292 * which should normally disable such checks. We need them to implement IPAddressAllow= and
1293 * IPAddressDeny=, hence let's bump the value high enough for our user. */
1294
1295 /* Using MAX() on resource limits only is safe if RLIM_INFINITY is > 0. POSIX declares that rlim_t
1296 * must be unsigned, hence this is a given, but let's make this clear here. */
1297 assert_cc(RLIM_INFINITY > 0);
1298
1299 mm = physical_memory_scale(1, 8); /* Let's scale how much we allow to be locked by the amount of
1300 * physical RAM. We allow an eighth to be locked by us, just to
1301 * pick a value. */
1302
1303 new_rlimit = (struct rlimit) {
1304 .rlim_cur = MAX3(HIGH_RLIMIT_MEMLOCK, saved_rlimit->rlim_cur, mm),
1305 .rlim_max = MAX3(HIGH_RLIMIT_MEMLOCK, saved_rlimit->rlim_max, mm),
1306 };
1307
1308 if (saved_rlimit->rlim_max >= new_rlimit.rlim_cur &&
1309 saved_rlimit->rlim_cur >= new_rlimit.rlim_max) {
1310 log_debug("RLIMIT_MEMLOCK is already as high or higher than we need it, not bumping.");
1311 return 0;
1312 }
1313
1314 r = setrlimit_closest(RLIMIT_MEMLOCK, &new_rlimit);
1315 if (r < 0)
1316 return log_warning_errno(r, "Setting RLIMIT_MEMLOCK failed, ignoring: %m");
1317
1318 return 0;
1319 }
1320
1321 static void test_usr(void) {
1322
1323 /* Check that /usr is either on the same file system as / or mounted already. */
1324
1325 if (dir_is_empty("/usr") <= 0)
1326 return;
1327
1328 log_warning("/usr appears to be on its own filesystem and is not already mounted. This is not a supported setup. "
1329 "Some things will probably break (sometimes even silently) in mysterious ways. "
1330 "Consult http://freedesktop.org/wiki/Software/systemd/separate-usr-is-broken for more information.");
1331 }
1332
1333 static int enforce_syscall_archs(Set *archs) {
1334 #if HAVE_SECCOMP
1335 int r;
1336
1337 if (!is_seccomp_available())
1338 return 0;
1339
1340 r = seccomp_restrict_archs(arg_syscall_archs);
1341 if (r < 0)
1342 return log_error_errno(r, "Failed to enforce system call architecture restrication: %m");
1343 #endif
1344 return 0;
1345 }
1346
1347 static int status_welcome(void) {
1348 _cleanup_free_ char *pretty_name = NULL, *ansi_color = NULL;
1349 int r;
1350
1351 if (!show_status_on(arg_show_status))
1352 return 0;
1353
1354 r = parse_os_release(NULL,
1355 "PRETTY_NAME", &pretty_name,
1356 "ANSI_COLOR", &ansi_color);
1357 if (r < 0)
1358 log_full_errno(r == -ENOENT ? LOG_DEBUG : LOG_WARNING, r,
1359 "Failed to read os-release file, ignoring: %m");
1360
1361 if (log_get_show_color())
1362 return status_printf(NULL, 0,
1363 "\nWelcome to \x1B[%sm%s\x1B[0m!\n",
1364 isempty(ansi_color) ? "1" : ansi_color,
1365 isempty(pretty_name) ? "Linux" : pretty_name);
1366 else
1367 return status_printf(NULL, 0,
1368 "\nWelcome to %s!\n",
1369 isempty(pretty_name) ? "Linux" : pretty_name);
1370 }
1371
1372 static int write_container_id(void) {
1373 const char *c;
1374 int r = 0; /* avoid false maybe-uninitialized warning */
1375
1376 c = getenv("container");
1377 if (isempty(c))
1378 return 0;
1379
1380 RUN_WITH_UMASK(0022)
1381 r = write_string_file("/run/systemd/container", c, WRITE_STRING_FILE_CREATE);
1382 if (r < 0)
1383 return log_warning_errno(r, "Failed to write /run/systemd/container, ignoring: %m");
1384
1385 return 1;
1386 }
1387
1388 static int bump_unix_max_dgram_qlen(void) {
1389 _cleanup_free_ char *qlen = NULL;
1390 unsigned long v;
1391 int r;
1392
1393 /* Let's bump the net.unix.max_dgram_qlen sysctl. The kernel default of 16 is simply too low. We set
1394 * the value really really early during boot, so that it is actually applied to all our sockets,
1395 * including the $NOTIFY_SOCKET one. */
1396
1397 r = read_one_line_file("/proc/sys/net/unix/max_dgram_qlen", &qlen);
1398 if (r < 0)
1399 return log_full_errno(r == -ENOENT ? LOG_DEBUG : LOG_WARNING, r,
1400 "Failed to read AF_UNIX datagram queue length, ignoring: %m");
1401
1402 r = safe_atolu(qlen, &v);
1403 if (r < 0)
1404 return log_warning_errno(r, "Failed to parse AF_UNIX datagram queue length '%s', ignoring: %m", qlen);
1405
1406 if (v >= DEFAULT_UNIX_MAX_DGRAM_QLEN)
1407 return 0;
1408
1409 r = write_string_filef("/proc/sys/net/unix/max_dgram_qlen", WRITE_STRING_FILE_DISABLE_BUFFER,
1410 "%lu", DEFAULT_UNIX_MAX_DGRAM_QLEN);
1411 if (r < 0)
1412 return log_full_errno(IN_SET(r, -EROFS, -EPERM, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
1413 "Failed to bump AF_UNIX datagram queue length, ignoring: %m");
1414
1415 return 1;
1416 }
1417
1418 static int fixup_environment(void) {
1419 _cleanup_free_ char *term = NULL;
1420 const char *t;
1421 int r;
1422
1423 /* Only fix up the environment when we are started as PID 1 */
1424 if (getpid_cached() != 1)
1425 return 0;
1426
1427 /* We expect the environment to be set correctly if run inside a container. */
1428 if (detect_container() > 0)
1429 return 0;
1430
1431 /* When started as PID1, the kernel uses /dev/console for our stdios and uses TERM=linux whatever the
1432 * backend device used by the console. We try to make a better guess here since some consoles might
1433 * not have support for color mode for example.
1434 *
1435 * However if TERM was configured through the kernel command line then leave it alone. */
1436 r = proc_cmdline_get_key("TERM", 0, &term);
1437 if (r < 0)
1438 return r;
1439
1440 t = term ?: default_term_for_tty("/dev/console");
1441
1442 if (setenv("TERM", t, 1) < 0)
1443 return -errno;
1444
1445 /* The kernels sets HOME=/ for init. Let's undo this. */
1446 if (path_equal_ptr(getenv("HOME"), "/"))
1447 assert_se(unsetenv("HOME") == 0);
1448
1449 return 0;
1450 }
1451
1452 static void redirect_telinit(int argc, char *argv[]) {
1453
1454 /* This is compatibility support for SysV, where calling init as a user is identical to telinit. */
1455
1456 #if HAVE_SYSV_COMPAT
1457 if (getpid_cached() == 1)
1458 return;
1459
1460 if (!invoked_as(argv, "init"))
1461 return;
1462
1463 execv(SYSTEMCTL_BINARY_PATH, argv);
1464 log_error_errno(errno, "Failed to exec " SYSTEMCTL_BINARY_PATH ": %m");
1465 exit(EXIT_FAILURE);
1466 #endif
1467 }
1468
1469 static int become_shutdown(
1470 const char *shutdown_verb,
1471 int retval) {
1472
1473 char log_level[DECIMAL_STR_MAX(int) + 1],
1474 exit_code[DECIMAL_STR_MAX(uint8_t) + 1],
1475 timeout[DECIMAL_STR_MAX(usec_t) + 1];
1476
1477 const char* command_line[13] = {
1478 SYSTEMD_SHUTDOWN_BINARY_PATH,
1479 shutdown_verb,
1480 "--timeout", timeout,
1481 "--log-level", log_level,
1482 "--log-target",
1483 };
1484
1485 _cleanup_strv_free_ char **env_block = NULL;
1486 size_t pos = 7;
1487 int r;
1488 usec_t watchdog_timer = 0;
1489
1490 assert(shutdown_verb);
1491 assert(!command_line[pos]);
1492 env_block = strv_copy(environ);
1493
1494 xsprintf(log_level, "%d", log_get_max_level());
1495 xsprintf(timeout, "%" PRI_USEC "us", arg_default_timeout_stop_usec);
1496
1497 switch (log_get_target()) {
1498
1499 case LOG_TARGET_KMSG:
1500 case LOG_TARGET_JOURNAL_OR_KMSG:
1501 case LOG_TARGET_SYSLOG_OR_KMSG:
1502 command_line[pos++] = "kmsg";
1503 break;
1504
1505 case LOG_TARGET_NULL:
1506 command_line[pos++] = "null";
1507 break;
1508
1509 case LOG_TARGET_CONSOLE:
1510 default:
1511 command_line[pos++] = "console";
1512 break;
1513 };
1514
1515 if (log_get_show_color())
1516 command_line[pos++] = "--log-color";
1517
1518 if (log_get_show_location())
1519 command_line[pos++] = "--log-location";
1520
1521 if (log_get_show_time())
1522 command_line[pos++] = "--log-time";
1523
1524 if (streq(shutdown_verb, "exit")) {
1525 command_line[pos++] = "--exit-code";
1526 command_line[pos++] = exit_code;
1527 xsprintf(exit_code, "%d", retval);
1528 }
1529
1530 assert(pos < ELEMENTSOF(command_line));
1531
1532 if (streq(shutdown_verb, "reboot"))
1533 watchdog_timer = arg_reboot_watchdog;
1534 else if (streq(shutdown_verb, "kexec"))
1535 watchdog_timer = arg_kexec_watchdog;
1536
1537 if (watchdog_timer > 0 && watchdog_timer != USEC_INFINITY) {
1538
1539 char *e;
1540
1541 /* If we reboot or kexec let's set the shutdown
1542 * watchdog and tell the shutdown binary to
1543 * repeatedly ping it */
1544 r = watchdog_set_timeout(&watchdog_timer);
1545 watchdog_close(r < 0);
1546
1547 /* Tell the binary how often to ping, ignore failure */
1548 if (asprintf(&e, "WATCHDOG_USEC="USEC_FMT, watchdog_timer) > 0)
1549 (void) strv_consume(&env_block, e);
1550
1551 if (arg_watchdog_device &&
1552 asprintf(&e, "WATCHDOG_DEVICE=%s", arg_watchdog_device) > 0)
1553 (void) strv_consume(&env_block, e);
1554 } else
1555 watchdog_close(true);
1556
1557 /* Avoid the creation of new processes forked by the
1558 * kernel; at this point, we will not listen to the
1559 * signals anyway */
1560 if (detect_container() <= 0)
1561 (void) cg_uninstall_release_agent(SYSTEMD_CGROUP_CONTROLLER);
1562
1563 execve(SYSTEMD_SHUTDOWN_BINARY_PATH, (char **) command_line, env_block);
1564 return -errno;
1565 }
1566
1567 static void initialize_clock(void) {
1568 int r;
1569
1570 /* This is called very early on, before we parse the kernel command line or otherwise figure out why
1571 * we are running, but only once. */
1572
1573 if (clock_is_localtime(NULL) > 0) {
1574 int min;
1575
1576 /* The very first call of settimeofday() also does a time warp in the kernel.
1577 *
1578 * In the rtc-in-local time mode, we set the kernel's timezone, and rely on external tools to
1579 * take care of maintaining the RTC and do all adjustments. This matches the behavior of
1580 * Windows, which leaves the RTC alone if the registry tells that the RTC runs in UTC.
1581 */
1582 r = clock_set_timezone(&min);
1583 if (r < 0)
1584 log_error_errno(r, "Failed to apply local time delta, ignoring: %m");
1585 else
1586 log_info("RTC configured in localtime, applying delta of %i minutes to system time.", min);
1587
1588 } else if (!in_initrd())
1589 /*
1590 * Do a dummy very first call to seal the kernel's time warp magic.
1591 *
1592 * Do not call this from inside the initrd. The initrd might not carry /etc/adjtime with
1593 * LOCAL, but the real system could be set up that way. In such case, we need to delay the
1594 * time-warp or the sealing until we reach the real system.
1595 *
1596 * Do no set the kernel's timezone. The concept of local time cannot be supported reliably,
1597 * the time will jump or be incorrect at every daylight saving time change. All kernel local
1598 * time concepts will be treated as UTC that way.
1599 */
1600 (void) clock_reset_timewarp();
1601
1602 ClockChangeDirection change_dir;
1603 r = clock_apply_epoch(&change_dir);
1604 if (r > 0 && change_dir == CLOCK_CHANGE_FORWARD)
1605 log_info("System time before build time, advancing clock.");
1606 else if (r > 0 && change_dir == CLOCK_CHANGE_BACKWARD)
1607 log_info("System time is further ahead than %s after build time, resetting clock to build time.",
1608 FORMAT_TIMESPAN(CLOCK_VALID_RANGE_USEC_MAX, USEC_PER_DAY));
1609 else if (r < 0 && change_dir == CLOCK_CHANGE_FORWARD)
1610 log_error_errno(r, "Current system time is before build time, but cannot correct: %m");
1611 else if (r < 0 && change_dir == CLOCK_CHANGE_BACKWARD)
1612 log_error_errno(r, "Current system time is further ahead %s after build time, but cannot correct: %m",
1613 FORMAT_TIMESPAN(CLOCK_VALID_RANGE_USEC_MAX, USEC_PER_DAY));
1614 }
1615
1616 static void apply_clock_update(void) {
1617 struct timespec ts;
1618
1619 /* This is called later than initialize_clock(), i.e. after we parsed configuration files/kernel
1620 * command line and such. */
1621
1622 if (arg_clock_usec == 0)
1623 return;
1624
1625 if (getpid_cached() != 1)
1626 return;
1627
1628 if (clock_settime(CLOCK_REALTIME, timespec_store(&ts, arg_clock_usec)) < 0)
1629 log_error_errno(errno, "Failed to set system clock to time specified on kernel command line: %m");
1630 else
1631 log_info("Set system clock to %s, as specified on the kernel command line.",
1632 FORMAT_TIMESTAMP(arg_clock_usec));
1633 }
1634
1635 static void cmdline_take_random_seed(void) {
1636 size_t suggested;
1637 int r;
1638
1639 if (arg_random_seed_size == 0)
1640 return;
1641
1642 if (getpid_cached() != 1)
1643 return;
1644
1645 assert(arg_random_seed);
1646 suggested = random_pool_size();
1647
1648 if (arg_random_seed_size < suggested)
1649 log_warning("Random seed specified on kernel command line has size %zu, but %zu bytes required to fill entropy pool.",
1650 arg_random_seed_size, suggested);
1651
1652 r = random_write_entropy(-1, arg_random_seed, arg_random_seed_size, true);
1653 if (r < 0) {
1654 log_warning_errno(r, "Failed to credit entropy specified on kernel command line, ignoring: %m");
1655 return;
1656 }
1657
1658 log_notice("Successfully credited entropy passed on kernel command line.\n"
1659 "Note that the seed provided this way is accessible to unprivileged programs. "
1660 "This functionality should not be used outside of testing environments.");
1661 }
1662
1663 static void initialize_coredump(bool skip_setup) {
1664 #if ENABLE_COREDUMP
1665 if (getpid_cached() != 1)
1666 return;
1667
1668 /* Don't limit the core dump size, so that coredump handlers such as systemd-coredump (which honour
1669 * the limit) will process core dumps for system services by default. */
1670 if (setrlimit(RLIMIT_CORE, &RLIMIT_MAKE_CONST(RLIM_INFINITY)) < 0)
1671 log_warning_errno(errno, "Failed to set RLIMIT_CORE: %m");
1672
1673 /* But at the same time, turn off the core_pattern logic by default, so that no coredumps are stored
1674 * until the systemd-coredump tool is enabled via sysctl. However it can be changed via the kernel
1675 * command line later so core dumps can still be generated during early startup and in initramfs. */
1676 if (!skip_setup)
1677 disable_coredumps();
1678 #endif
1679 }
1680
1681 static void initialize_core_pattern(bool skip_setup) {
1682 int r;
1683
1684 if (skip_setup || !arg_early_core_pattern)
1685 return;
1686
1687 if (getpid_cached() != 1)
1688 return;
1689
1690 r = write_string_file("/proc/sys/kernel/core_pattern", arg_early_core_pattern, WRITE_STRING_FILE_DISABLE_BUFFER);
1691 if (r < 0)
1692 log_warning_errno(r, "Failed to write '%s' to /proc/sys/kernel/core_pattern, ignoring: %m",
1693 arg_early_core_pattern);
1694 }
1695
1696 static void update_cpu_affinity(bool skip_setup) {
1697 _cleanup_free_ char *mask = NULL;
1698
1699 if (skip_setup || !arg_cpu_affinity.set)
1700 return;
1701
1702 assert(arg_cpu_affinity.allocated > 0);
1703
1704 mask = cpu_set_to_string(&arg_cpu_affinity);
1705 log_debug("Setting CPU affinity to %s.", strnull(mask));
1706
1707 if (sched_setaffinity(0, arg_cpu_affinity.allocated, arg_cpu_affinity.set) < 0)
1708 log_warning_errno(errno, "Failed to set CPU affinity: %m");
1709 }
1710
1711 static void update_numa_policy(bool skip_setup) {
1712 int r;
1713 _cleanup_free_ char *nodes = NULL;
1714 const char * policy = NULL;
1715
1716 if (skip_setup || !mpol_is_valid(numa_policy_get_type(&arg_numa_policy)))
1717 return;
1718
1719 if (DEBUG_LOGGING) {
1720 policy = mpol_to_string(numa_policy_get_type(&arg_numa_policy));
1721 nodes = cpu_set_to_range_string(&arg_numa_policy.nodes);
1722 log_debug("Setting NUMA policy to %s, with nodes %s.", strnull(policy), strnull(nodes));
1723 }
1724
1725 r = apply_numa_policy(&arg_numa_policy);
1726 if (r == -EOPNOTSUPP)
1727 log_debug_errno(r, "NUMA support not available, ignoring.");
1728 else if (r < 0)
1729 log_warning_errno(r, "Failed to set NUMA memory policy: %m");
1730 }
1731
1732 static void filter_args(
1733 const char* dst[],
1734 size_t *dst_index,
1735 char **src,
1736 int argc) {
1737
1738 assert(dst);
1739 assert(dst_index);
1740
1741 /* Copy some filtered arguments into the dst array from src. */
1742 for (int i = 1; i < argc; i++) {
1743 if (STR_IN_SET(src[i],
1744 "--switched-root",
1745 "--system",
1746 "--user"))
1747 continue;
1748
1749 if (startswith(src[i], "--deserialize="))
1750 continue;
1751 if (streq(src[i], "--deserialize")) {
1752 i++; /* Skip the argument too */
1753 continue;
1754 }
1755
1756 /* Skip target unit designators. We already acted upon this information and have queued
1757 * appropriate jobs. We don't want to redo all this after reexecution. */
1758 if (startswith(src[i], "--unit="))
1759 continue;
1760 if (streq(src[i], "--unit")) {
1761 i++; /* Skip the argument too */
1762 continue;
1763 }
1764
1765 if (startswith(src[i],
1766 in_initrd() ? "rd.systemd.unit=" : "systemd.unit="))
1767 continue;
1768
1769 if (runlevel_to_target(src[i]))
1770 continue;
1771
1772 /* Seems we have a good old option. Let's pass it over to the new instance. */
1773 dst[(*dst_index)++] = src[i];
1774 }
1775 }
1776
1777 static void do_reexecute(
1778 int argc,
1779 char* argv[],
1780 const struct rlimit *saved_rlimit_nofile,
1781 const struct rlimit *saved_rlimit_memlock,
1782 FDSet *fds,
1783 const char *switch_root_dir,
1784 const char *switch_root_init,
1785 const char **ret_error_message) {
1786
1787 size_t i, args_size;
1788 const char **args;
1789 int r;
1790
1791 assert(argc >= 0);
1792 assert(saved_rlimit_nofile);
1793 assert(saved_rlimit_memlock);
1794 assert(ret_error_message);
1795
1796 /* Close and disarm the watchdog, so that the new instance can reinitialize it, but doesn't get
1797 * rebooted while we do that */
1798 watchdog_close(true);
1799
1800 /* Reset RLIMIT_NOFILE + RLIMIT_MEMLOCK back to the kernel defaults, so that the new systemd can pass
1801 * the kernel default to its child processes */
1802 if (saved_rlimit_nofile->rlim_cur != 0)
1803 (void) setrlimit(RLIMIT_NOFILE, saved_rlimit_nofile);
1804 if (saved_rlimit_memlock->rlim_cur != RLIM_INFINITY)
1805 (void) setrlimit(RLIMIT_MEMLOCK, saved_rlimit_memlock);
1806
1807 if (switch_root_dir) {
1808 /* Kill all remaining processes from the initrd, but don't wait for them, so that we can
1809 * handle the SIGCHLD for them after deserializing. */
1810 broadcast_signal(SIGTERM, false, true, arg_default_timeout_stop_usec);
1811
1812 /* And switch root with MS_MOVE, because we remove the old directory afterwards and detach it. */
1813 r = switch_root(switch_root_dir, "/mnt", true, MS_MOVE);
1814 if (r < 0)
1815 log_error_errno(r, "Failed to switch root, trying to continue: %m");
1816 }
1817
1818 args_size = argc + 6;
1819 args = newa(const char*, args_size);
1820
1821 if (!switch_root_init) {
1822 char sfd[DECIMAL_STR_MAX(int)];
1823
1824 /* First try to spawn ourselves with the right path, and with full serialization. We do this
1825 * only if the user didn't specify an explicit init to spawn. */
1826
1827 assert(arg_serialization);
1828 assert(fds);
1829
1830 xsprintf(sfd, "%i", fileno(arg_serialization));
1831
1832 i = 1; /* Leave args[0] empty for now. */
1833 filter_args(args, &i, argv, argc);
1834
1835 if (switch_root_dir)
1836 args[i++] = "--switched-root";
1837 args[i++] = arg_system ? "--system" : "--user";
1838 args[i++] = "--deserialize";
1839 args[i++] = sfd;
1840 args[i++] = NULL;
1841
1842 assert(i <= args_size);
1843
1844 /*
1845 * We want valgrind to print its memory usage summary before reexecution. Valgrind won't do
1846 * this is on its own on exec(), but it will do it on exit(). Hence, to ensure we get a
1847 * summary here, fork() off a child, let it exit() cleanly, so that it prints the summary,
1848 * and wait() for it in the parent, before proceeding into the exec().
1849 */
1850 valgrind_summary_hack();
1851
1852 args[0] = SYSTEMD_BINARY_PATH;
1853 (void) execv(args[0], (char* const*) args);
1854 log_debug_errno(errno, "Failed to execute our own binary %s, trying fallback: %m", args[0]);
1855 }
1856
1857 /* Try the fallback, if there is any, without any serialization. We pass the original argv[] and
1858 * envp[]. (Well, modulo the ordering changes due to getopt() in argv[], and some cleanups in envp[],
1859 * but let's hope that doesn't matter.) */
1860
1861 arg_serialization = safe_fclose(arg_serialization);
1862 fds = fdset_free(fds);
1863
1864 /* Reopen the console */
1865 (void) make_console_stdio();
1866
1867 i = 1; /* Leave args[0] empty for now. */
1868 for (int j = 1; j <= argc; j++)
1869 args[i++] = argv[j];
1870 assert(i <= args_size);
1871
1872 /* Re-enable any blocked signals, especially important if we switch from initial ramdisk to init=... */
1873 (void) reset_all_signal_handlers();
1874 (void) reset_signal_mask();
1875 (void) rlimit_nofile_safe();
1876
1877 if (switch_root_init) {
1878 args[0] = switch_root_init;
1879 (void) execve(args[0], (char* const*) args, saved_env);
1880 log_warning_errno(errno, "Failed to execute configured init %s, trying fallback: %m", args[0]);
1881 }
1882
1883 args[0] = "/sbin/init";
1884 (void) execv(args[0], (char* const*) args);
1885 r = -errno;
1886
1887 manager_status_printf(NULL, STATUS_TYPE_EMERGENCY,
1888 ANSI_HIGHLIGHT_RED " !! " ANSI_NORMAL,
1889 "Failed to execute /sbin/init");
1890
1891 if (r == -ENOENT) {
1892 log_warning("No /sbin/init, trying fallback");
1893
1894 args[0] = "/bin/sh";
1895 args[1] = NULL;
1896 (void) execve(args[0], (char* const*) args, saved_env);
1897 log_error_errno(errno, "Failed to execute /bin/sh, giving up: %m");
1898 } else
1899 log_warning_errno(r, "Failed to execute /sbin/init, giving up: %m");
1900
1901 *ret_error_message = "Failed to execute fallback shell";
1902 }
1903
1904 static int invoke_main_loop(
1905 Manager *m,
1906 const struct rlimit *saved_rlimit_nofile,
1907 const struct rlimit *saved_rlimit_memlock,
1908 bool *ret_reexecute,
1909 int *ret_retval, /* Return parameters relevant for shutting down */
1910 const char **ret_shutdown_verb, /* … */
1911 FDSet **ret_fds, /* Return parameters for reexecuting */
1912 char **ret_switch_root_dir, /* … */
1913 char **ret_switch_root_init, /* … */
1914 const char **ret_error_message) {
1915
1916 int r;
1917
1918 assert(m);
1919 assert(saved_rlimit_nofile);
1920 assert(saved_rlimit_memlock);
1921 assert(ret_reexecute);
1922 assert(ret_retval);
1923 assert(ret_shutdown_verb);
1924 assert(ret_fds);
1925 assert(ret_switch_root_dir);
1926 assert(ret_switch_root_init);
1927 assert(ret_error_message);
1928
1929 for (;;) {
1930 r = manager_loop(m);
1931 if (r < 0) {
1932 *ret_error_message = "Failed to run main loop";
1933 return log_emergency_errno(r, "Failed to run main loop: %m");
1934 }
1935
1936 switch ((ManagerObjective) r) {
1937
1938 case MANAGER_RELOAD: {
1939 LogTarget saved_log_target;
1940 int saved_log_level;
1941
1942 log_info("Reloading.");
1943
1944 /* First, save any overridden log level/target, then parse the configuration file,
1945 * which might change the log level to new settings. */
1946
1947 saved_log_level = m->log_level_overridden ? log_get_max_level() : -1;
1948 saved_log_target = m->log_target_overridden ? log_get_target() : _LOG_TARGET_INVALID;
1949
1950 (void) parse_configuration(saved_rlimit_nofile, saved_rlimit_memlock);
1951
1952 set_manager_defaults(m);
1953 set_manager_settings(m);
1954
1955 update_cpu_affinity(false);
1956 update_numa_policy(false);
1957
1958 if (saved_log_level >= 0)
1959 manager_override_log_level(m, saved_log_level);
1960 if (saved_log_target >= 0)
1961 manager_override_log_target(m, saved_log_target);
1962
1963 r = manager_reload(m);
1964 if (r < 0)
1965 /* Reloading failed before the point of no return.
1966 * Let's continue running as if nothing happened. */
1967 m->objective = MANAGER_OK;
1968
1969 break;
1970 }
1971
1972 case MANAGER_REEXECUTE:
1973
1974 r = prepare_reexecute(m, &arg_serialization, ret_fds, false);
1975 if (r < 0) {
1976 *ret_error_message = "Failed to prepare for reexecution";
1977 return r;
1978 }
1979
1980 log_notice("Reexecuting.");
1981
1982 *ret_reexecute = true;
1983 *ret_retval = EXIT_SUCCESS;
1984 *ret_shutdown_verb = NULL;
1985 *ret_switch_root_dir = *ret_switch_root_init = NULL;
1986
1987 return 0;
1988
1989 case MANAGER_SWITCH_ROOT:
1990 if (!m->switch_root_init) {
1991 r = prepare_reexecute(m, &arg_serialization, ret_fds, true);
1992 if (r < 0) {
1993 *ret_error_message = "Failed to prepare for reexecution";
1994 return r;
1995 }
1996 } else
1997 *ret_fds = NULL;
1998
1999 log_notice("Switching root.");
2000
2001 *ret_reexecute = true;
2002 *ret_retval = EXIT_SUCCESS;
2003 *ret_shutdown_verb = NULL;
2004
2005 /* Steal the switch root parameters */
2006 *ret_switch_root_dir = TAKE_PTR(m->switch_root);
2007 *ret_switch_root_init = TAKE_PTR(m->switch_root_init);
2008
2009 return 0;
2010
2011 case MANAGER_EXIT:
2012
2013 if (MANAGER_IS_USER(m)) {
2014 log_debug("Exit.");
2015
2016 *ret_reexecute = false;
2017 *ret_retval = m->return_value;
2018 *ret_shutdown_verb = NULL;
2019 *ret_fds = NULL;
2020 *ret_switch_root_dir = *ret_switch_root_init = NULL;
2021
2022 return 0;
2023 }
2024
2025 _fallthrough_;
2026 case MANAGER_REBOOT:
2027 case MANAGER_POWEROFF:
2028 case MANAGER_HALT:
2029 case MANAGER_KEXEC: {
2030 static const char * const table[_MANAGER_OBJECTIVE_MAX] = {
2031 [MANAGER_EXIT] = "exit",
2032 [MANAGER_REBOOT] = "reboot",
2033 [MANAGER_POWEROFF] = "poweroff",
2034 [MANAGER_HALT] = "halt",
2035 [MANAGER_KEXEC] = "kexec",
2036 };
2037
2038 log_notice("Shutting down.");
2039
2040 *ret_reexecute = false;
2041 *ret_retval = m->return_value;
2042 assert_se(*ret_shutdown_verb = table[m->objective]);
2043 *ret_fds = NULL;
2044 *ret_switch_root_dir = *ret_switch_root_init = NULL;
2045
2046 return 0;
2047 }
2048
2049 default:
2050 assert_not_reached();
2051 }
2052 }
2053 }
2054
2055 static void log_execution_mode(bool *ret_first_boot) {
2056 assert(ret_first_boot);
2057
2058 if (arg_system) {
2059 int v;
2060
2061 log_info("systemd " GIT_VERSION " running in %ssystem mode (%s)",
2062 arg_action == ACTION_TEST ? "test " : "",
2063 systemd_features);
2064
2065 v = detect_virtualization();
2066 if (v > 0)
2067 log_info("Detected virtualization %s.", virtualization_to_string(v));
2068
2069 log_info("Detected architecture %s.", architecture_to_string(uname_architecture()));
2070
2071 if (in_initrd()) {
2072 *ret_first_boot = false;
2073 log_info("Running in initial RAM disk.");
2074 } else {
2075 int r;
2076 _cleanup_free_ char *id_text = NULL;
2077
2078 /* Let's check whether we are in first boot. We use /etc/machine-id as flag file
2079 * for this: If it is missing or contains the value "uninitialized", this is the
2080 * first boot. In any other case, it is not. This allows container managers and
2081 * installers to provision a couple of files already. If the container manager
2082 * wants to provision the machine ID itself it should pass $container_uuid to PID 1. */
2083
2084 r = read_one_line_file("/etc/machine-id", &id_text);
2085 if (r < 0 || streq(id_text, "uninitialized")) {
2086 if (r < 0 && r != -ENOENT)
2087 log_warning_errno(r, "Unexpected error while reading /etc/machine-id, ignoring: %m");
2088
2089 *ret_first_boot = true;
2090 log_info("Detected first boot.");
2091 } else {
2092 *ret_first_boot = false;
2093 log_debug("Detected initialized system, this is not the first boot.");
2094 }
2095 }
2096 } else {
2097 if (DEBUG_LOGGING) {
2098 _cleanup_free_ char *t = NULL;
2099
2100 t = uid_to_name(getuid());
2101 log_debug("systemd " GIT_VERSION " running in %suser mode for user " UID_FMT "/%s. (%s)",
2102 arg_action == ACTION_TEST ? " test" : "",
2103 getuid(), strna(t), systemd_features);
2104 }
2105
2106 *ret_first_boot = false;
2107 }
2108 }
2109
2110 static int initialize_runtime(
2111 bool skip_setup,
2112 bool first_boot,
2113 struct rlimit *saved_rlimit_nofile,
2114 struct rlimit *saved_rlimit_memlock,
2115 const char **ret_error_message) {
2116 int r;
2117
2118 assert(ret_error_message);
2119
2120 /* Sets up various runtime parameters. Many of these initializations are conditionalized:
2121 *
2122 * - Some only apply to --system instances
2123 * - Some only apply to --user instances
2124 * - Some only apply when we first start up, but not when we reexecute
2125 */
2126
2127 if (arg_action != ACTION_RUN)
2128 return 0;
2129
2130 update_cpu_affinity(skip_setup);
2131 update_numa_policy(skip_setup);
2132
2133 if (arg_system) {
2134 /* Make sure we leave a core dump without panicking the kernel. */
2135 install_crash_handler();
2136
2137 if (!skip_setup) {
2138 r = mount_cgroup_controllers();
2139 if (r < 0) {
2140 *ret_error_message = "Failed to mount cgroup hierarchies";
2141 return r;
2142 }
2143
2144 status_welcome();
2145 (void) hostname_setup(true);
2146 /* Force transient machine-id on first boot. */
2147 machine_id_setup(NULL, first_boot, arg_machine_id, NULL);
2148 (void) loopback_setup();
2149 bump_unix_max_dgram_qlen();
2150 bump_file_max_and_nr_open();
2151 test_usr();
2152 write_container_id();
2153 }
2154
2155 if (arg_watchdog_device) {
2156 r = watchdog_set_device(arg_watchdog_device);
2157 if (r < 0)
2158 log_warning_errno(r, "Failed to set watchdog device to %s, ignoring: %m", arg_watchdog_device);
2159 }
2160 } else {
2161 _cleanup_free_ char *p = NULL;
2162
2163 /* Create the runtime directory and place the inaccessible device nodes there, if we run in
2164 * user mode. In system mode mount_setup() already did that. */
2165
2166 r = xdg_user_runtime_dir(&p, "/systemd");
2167 if (r < 0) {
2168 *ret_error_message = "$XDG_RUNTIME_DIR is not set";
2169 return log_emergency_errno(r, "Failed to determine $XDG_RUNTIME_DIR path: %m");
2170 }
2171
2172 (void) mkdir_p_label(p, 0755);
2173 (void) make_inaccessible_nodes(p, UID_INVALID, GID_INVALID);
2174 }
2175
2176 if (arg_timer_slack_nsec != NSEC_INFINITY)
2177 if (prctl(PR_SET_TIMERSLACK, arg_timer_slack_nsec) < 0)
2178 log_warning_errno(errno, "Failed to adjust timer slack, ignoring: %m");
2179
2180 if (arg_system && !cap_test_all(arg_capability_bounding_set)) {
2181 r = capability_bounding_set_drop_usermode(arg_capability_bounding_set);
2182 if (r < 0) {
2183 *ret_error_message = "Failed to drop capability bounding set of usermode helpers";
2184 return log_emergency_errno(r, "Failed to drop capability bounding set of usermode helpers: %m");
2185 }
2186
2187 r = capability_bounding_set_drop(arg_capability_bounding_set, true);
2188 if (r < 0) {
2189 *ret_error_message = "Failed to drop capability bounding set";
2190 return log_emergency_errno(r, "Failed to drop capability bounding set: %m");
2191 }
2192 }
2193
2194 if (arg_system && arg_no_new_privs) {
2195 if (prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0) < 0) {
2196 *ret_error_message = "Failed to disable new privileges";
2197 return log_emergency_errno(errno, "Failed to disable new privileges: %m");
2198 }
2199 }
2200
2201 if (arg_syscall_archs) {
2202 r = enforce_syscall_archs(arg_syscall_archs);
2203 if (r < 0) {
2204 *ret_error_message = "Failed to set syscall architectures";
2205 return r;
2206 }
2207 }
2208
2209 if (!arg_system)
2210 /* Become reaper of our children */
2211 if (prctl(PR_SET_CHILD_SUBREAPER, 1) < 0)
2212 log_warning_errno(errno, "Failed to make us a subreaper: %m");
2213
2214 /* Bump up RLIMIT_NOFILE for systemd itself */
2215 (void) bump_rlimit_nofile(saved_rlimit_nofile);
2216 (void) bump_rlimit_memlock(saved_rlimit_memlock);
2217
2218 return 0;
2219 }
2220
2221 static int do_queue_default_job(
2222 Manager *m,
2223 const char **ret_error_message) {
2224
2225 _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL;
2226 const char *unit;
2227 Job *job;
2228 Unit *target;
2229 int r;
2230
2231 if (arg_default_unit)
2232 unit = arg_default_unit;
2233 else if (in_initrd())
2234 unit = SPECIAL_INITRD_TARGET;
2235 else
2236 unit = SPECIAL_DEFAULT_TARGET;
2237
2238 log_debug("Activating default unit: %s", unit);
2239
2240 r = manager_load_startable_unit_or_warn(m, unit, NULL, &target);
2241 if (r < 0 && in_initrd() && !arg_default_unit) {
2242 /* Fall back to default.target, which we used to always use by default. Only do this if no
2243 * explicit configuration was given. */
2244
2245 log_info("Falling back to " SPECIAL_DEFAULT_TARGET ".");
2246
2247 r = manager_load_startable_unit_or_warn(m, SPECIAL_DEFAULT_TARGET, NULL, &target);
2248 }
2249 if (r < 0) {
2250 log_info("Falling back to " SPECIAL_RESCUE_TARGET ".");
2251
2252 r = manager_load_startable_unit_or_warn(m, SPECIAL_RESCUE_TARGET, NULL, &target);
2253 if (r < 0) {
2254 *ret_error_message = r == -ERFKILL ? SPECIAL_RESCUE_TARGET " masked"
2255 : "Failed to load " SPECIAL_RESCUE_TARGET;
2256 return r;
2257 }
2258 }
2259
2260 assert(target->load_state == UNIT_LOADED);
2261
2262 r = manager_add_job(m, JOB_START, target, JOB_ISOLATE, NULL, &error, &job);
2263 if (r == -EPERM) {
2264 log_debug_errno(r, "Default target could not be isolated, starting instead: %s", bus_error_message(&error, r));
2265
2266 sd_bus_error_free(&error);
2267
2268 r = manager_add_job(m, JOB_START, target, JOB_REPLACE, NULL, &error, &job);
2269 if (r < 0) {
2270 *ret_error_message = "Failed to start default target";
2271 return log_emergency_errno(r, "Failed to start default target: %s", bus_error_message(&error, r));
2272 }
2273
2274 } else if (r < 0) {
2275 *ret_error_message = "Failed to isolate default target";
2276 return log_emergency_errno(r, "Failed to isolate default target: %s", bus_error_message(&error, r));
2277 } else
2278 log_info("Queued %s job for default target %s.",
2279 job_type_to_string(job->type),
2280 unit_status_string(job->unit, NULL));
2281
2282 m->default_unit_job_id = job->id;
2283
2284 return 0;
2285 }
2286
2287 static void save_rlimits(struct rlimit *saved_rlimit_nofile,
2288 struct rlimit *saved_rlimit_memlock) {
2289
2290 assert(saved_rlimit_nofile);
2291 assert(saved_rlimit_memlock);
2292
2293 if (getrlimit(RLIMIT_NOFILE, saved_rlimit_nofile) < 0)
2294 log_warning_errno(errno, "Reading RLIMIT_NOFILE failed, ignoring: %m");
2295
2296 if (getrlimit(RLIMIT_MEMLOCK, saved_rlimit_memlock) < 0)
2297 log_warning_errno(errno, "Reading RLIMIT_MEMLOCK failed, ignoring: %m");
2298 }
2299
2300 static void fallback_rlimit_nofile(const struct rlimit *saved_rlimit_nofile) {
2301 struct rlimit *rl;
2302
2303 if (arg_default_rlimit[RLIMIT_NOFILE])
2304 return;
2305
2306 /* Make sure forked processes get limits based on the original kernel setting */
2307
2308 rl = newdup(struct rlimit, saved_rlimit_nofile, 1);
2309 if (!rl) {
2310 log_oom();
2311 return;
2312 }
2313
2314 /* Bump the hard limit for system services to a substantially higher value. The default
2315 * hard limit current kernels set is pretty low (4K), mostly for historical
2316 * reasons. According to kernel developers, the fd handling in recent kernels has been
2317 * optimized substantially enough, so that we can bump the limit now, without paying too
2318 * high a price in memory or performance. Note however that we only bump the hard limit,
2319 * not the soft limit. That's because select() works the way it works, and chokes on fds
2320 * >= 1024. If we'd bump the soft limit globally, it might accidentally happen to
2321 * unexpecting programs that they get fds higher than what they can process using
2322 * select(). By only bumping the hard limit but leaving the low limit as it is we avoid
2323 * this pitfall: programs that are written by folks aware of the select() problem in mind
2324 * (and thus use poll()/epoll instead of select(), the way everybody should) can
2325 * explicitly opt into high fds by bumping their soft limit beyond 1024, to the hard limit
2326 * we pass. */
2327 if (arg_system) {
2328 int nr;
2329
2330 /* Get the underlying absolute limit the kernel enforces */
2331 nr = read_nr_open();
2332
2333 rl->rlim_max = MIN((rlim_t) nr, MAX(rl->rlim_max, (rlim_t) HIGH_RLIMIT_NOFILE));
2334 }
2335
2336 /* If for some reason we were invoked with a soft limit above 1024 (which should never
2337 * happen!, but who knows what we get passed in from pam_limit when invoked as --user
2338 * instance), then lower what we pass on to not confuse our children */
2339 rl->rlim_cur = MIN(rl->rlim_cur, (rlim_t) FD_SETSIZE);
2340
2341 arg_default_rlimit[RLIMIT_NOFILE] = rl;
2342 }
2343
2344 static void fallback_rlimit_memlock(const struct rlimit *saved_rlimit_memlock) {
2345 struct rlimit *rl;
2346
2347 /* Pass the original value down to invoked processes */
2348
2349 if (arg_default_rlimit[RLIMIT_MEMLOCK])
2350 return;
2351
2352 rl = newdup(struct rlimit, saved_rlimit_memlock, 1);
2353 if (!rl) {
2354 log_oom();
2355 return;
2356 }
2357
2358 arg_default_rlimit[RLIMIT_MEMLOCK] = rl;
2359 }
2360
2361 static void setenv_manager_environment(void) {
2362 char **p;
2363 int r;
2364
2365 STRV_FOREACH(p, arg_manager_environment) {
2366 log_debug("Setting '%s' in our own environment.", *p);
2367
2368 r = putenv_dup(*p, true);
2369 if (r < 0)
2370 log_warning_errno(errno, "Failed to setenv \"%s\", ignoring: %m", *p);
2371 }
2372 }
2373
2374 static void reset_arguments(void) {
2375 /* Frees/resets arg_* variables, with a few exceptions commented below. */
2376
2377 arg_default_unit = mfree(arg_default_unit);
2378
2379 /* arg_system — ignore */
2380
2381 arg_dump_core = true;
2382 arg_crash_chvt = -1;
2383 arg_crash_shell = false;
2384 arg_crash_reboot = false;
2385 arg_confirm_spawn = mfree(arg_confirm_spawn);
2386 arg_show_status = _SHOW_STATUS_INVALID;
2387 arg_status_unit_format = STATUS_UNIT_FORMAT_DEFAULT;
2388 arg_switched_root = false;
2389 arg_pager_flags = 0;
2390 arg_service_watchdogs = true;
2391 arg_default_std_output = EXEC_OUTPUT_JOURNAL;
2392 arg_default_std_error = EXEC_OUTPUT_INHERIT;
2393 arg_default_restart_usec = DEFAULT_RESTART_USEC;
2394 arg_default_timeout_start_usec = DEFAULT_TIMEOUT_USEC;
2395 arg_default_timeout_stop_usec = DEFAULT_TIMEOUT_USEC;
2396 arg_default_timeout_abort_usec = DEFAULT_TIMEOUT_USEC;
2397 arg_default_timeout_abort_set = false;
2398 arg_default_start_limit_interval = DEFAULT_START_LIMIT_INTERVAL;
2399 arg_default_start_limit_burst = DEFAULT_START_LIMIT_BURST;
2400 arg_runtime_watchdog = 0;
2401 arg_reboot_watchdog = 10 * USEC_PER_MINUTE;
2402 arg_kexec_watchdog = 0;
2403 arg_early_core_pattern = NULL;
2404 arg_watchdog_device = NULL;
2405
2406 arg_default_environment = strv_free(arg_default_environment);
2407 arg_manager_environment = strv_free(arg_manager_environment);
2408 rlimit_free_all(arg_default_rlimit);
2409
2410 arg_capability_bounding_set = CAP_ALL;
2411 arg_no_new_privs = false;
2412 arg_timer_slack_nsec = NSEC_INFINITY;
2413 arg_default_timer_accuracy_usec = 1 * USEC_PER_MINUTE;
2414
2415 arg_syscall_archs = set_free(arg_syscall_archs);
2416
2417 /* arg_serialization — ignore */
2418
2419 arg_default_cpu_accounting = -1;
2420 arg_default_io_accounting = false;
2421 arg_default_ip_accounting = false;
2422 arg_default_blockio_accounting = false;
2423 arg_default_memory_accounting = MEMORY_ACCOUNTING_DEFAULT;
2424 arg_default_tasks_accounting = true;
2425 arg_default_tasks_max = DEFAULT_TASKS_MAX;
2426 arg_machine_id = (sd_id128_t) {};
2427 arg_cad_burst_action = EMERGENCY_ACTION_REBOOT_FORCE;
2428 arg_default_oom_policy = OOM_STOP;
2429
2430 cpu_set_reset(&arg_cpu_affinity);
2431 numa_policy_reset(&arg_numa_policy);
2432
2433 arg_random_seed = mfree(arg_random_seed);
2434 arg_random_seed_size = 0;
2435 arg_clock_usec = 0;
2436 }
2437
2438 static int parse_configuration(const struct rlimit *saved_rlimit_nofile,
2439 const struct rlimit *saved_rlimit_memlock) {
2440 int r;
2441
2442 assert(saved_rlimit_nofile);
2443 assert(saved_rlimit_memlock);
2444
2445 /* Assign configuration defaults */
2446 reset_arguments();
2447
2448 r = parse_config_file();
2449 if (r < 0)
2450 log_warning_errno(r, "Failed to parse config file, ignoring: %m");
2451
2452 if (arg_system) {
2453 r = proc_cmdline_parse(parse_proc_cmdline_item, NULL, 0);
2454 if (r < 0)
2455 log_warning_errno(r, "Failed to parse kernel command line, ignoring: %m");
2456 }
2457
2458 /* Initialize some default rlimits for services if they haven't been configured */
2459 fallback_rlimit_nofile(saved_rlimit_nofile);
2460 fallback_rlimit_memlock(saved_rlimit_memlock);
2461
2462 /* Note that this also parses bits from the kernel command line, including "debug". */
2463 log_parse_environment();
2464
2465 /* Initialize the show status setting if it hasn't been set explicitly yet */
2466 if (arg_show_status == _SHOW_STATUS_INVALID)
2467 arg_show_status = SHOW_STATUS_YES;
2468
2469 /* Push variables into the manager environment block */
2470 setenv_manager_environment();
2471
2472 return 0;
2473 }
2474
2475 static int safety_checks(void) {
2476
2477 if (getpid_cached() == 1 &&
2478 arg_action != ACTION_RUN)
2479 return log_error_errno(SYNTHETIC_ERRNO(EPERM),
2480 "Unsupported execution mode while PID 1.");
2481
2482 if (getpid_cached() == 1 &&
2483 !arg_system)
2484 return log_error_errno(SYNTHETIC_ERRNO(EPERM),
2485 "Can't run --user mode as PID 1.");
2486
2487 if (arg_action == ACTION_RUN &&
2488 arg_system &&
2489 getpid_cached() != 1)
2490 return log_error_errno(SYNTHETIC_ERRNO(EPERM),
2491 "Can't run system mode unless PID 1.");
2492
2493 if (arg_action == ACTION_TEST &&
2494 geteuid() == 0)
2495 return log_error_errno(SYNTHETIC_ERRNO(EPERM),
2496 "Don't run test mode as root.");
2497
2498 if (!arg_system &&
2499 arg_action == ACTION_RUN &&
2500 sd_booted() <= 0)
2501 return log_error_errno(SYNTHETIC_ERRNO(EOPNOTSUPP),
2502 "Trying to run as user instance, but the system has not been booted with systemd.");
2503
2504 if (!arg_system &&
2505 arg_action == ACTION_RUN &&
2506 !getenv("XDG_RUNTIME_DIR"))
2507 return log_error_errno(SYNTHETIC_ERRNO(EUNATCH),
2508 "Trying to run as user instance, but $XDG_RUNTIME_DIR is not set.");
2509
2510 if (arg_system &&
2511 arg_action == ACTION_RUN &&
2512 running_in_chroot() > 0)
2513 return log_error_errno(SYNTHETIC_ERRNO(EOPNOTSUPP),
2514 "Cannot be run in a chroot() environment.");
2515
2516 return 0;
2517 }
2518
2519 static int initialize_security(
2520 bool *loaded_policy,
2521 dual_timestamp *security_start_timestamp,
2522 dual_timestamp *security_finish_timestamp,
2523 const char **ret_error_message) {
2524
2525 int r;
2526
2527 assert(loaded_policy);
2528 assert(security_start_timestamp);
2529 assert(security_finish_timestamp);
2530 assert(ret_error_message);
2531
2532 dual_timestamp_get(security_start_timestamp);
2533
2534 r = mac_selinux_setup(loaded_policy);
2535 if (r < 0) {
2536 *ret_error_message = "Failed to load SELinux policy";
2537 return r;
2538 }
2539
2540 r = mac_smack_setup(loaded_policy);
2541 if (r < 0) {
2542 *ret_error_message = "Failed to load SMACK policy";
2543 return r;
2544 }
2545
2546 r = mac_apparmor_setup();
2547 if (r < 0) {
2548 *ret_error_message = "Failed to load AppArmor policy";
2549 return r;
2550 }
2551
2552 r = ima_setup();
2553 if (r < 0) {
2554 *ret_error_message = "Failed to load IMA policy";
2555 return r;
2556 }
2557
2558 dual_timestamp_get(security_finish_timestamp);
2559 return 0;
2560 }
2561
2562 static int collect_fds(FDSet **ret_fds, const char **ret_error_message) {
2563 int r;
2564
2565 assert(ret_fds);
2566 assert(ret_error_message);
2567
2568 r = fdset_new_fill(ret_fds);
2569 if (r < 0) {
2570 *ret_error_message = "Failed to allocate fd set";
2571 return log_emergency_errno(r, "Failed to allocate fd set: %m");
2572 }
2573
2574 fdset_cloexec(*ret_fds, true);
2575
2576 if (arg_serialization)
2577 assert_se(fdset_remove(*ret_fds, fileno(arg_serialization)) >= 0);
2578
2579 return 0;
2580 }
2581
2582 static void setup_console_terminal(bool skip_setup) {
2583
2584 if (!arg_system)
2585 return;
2586
2587 /* Become a session leader if we aren't one yet. */
2588 (void) setsid();
2589
2590 /* If we are init, we connect stdin/stdout/stderr to /dev/null and make sure we don't have a
2591 * controlling tty. */
2592 (void) release_terminal();
2593
2594 /* Reset the console, but only if this is really init and we are freshly booted */
2595 if (getpid_cached() == 1 && !skip_setup)
2596 (void) console_setup();
2597 }
2598
2599 static bool early_skip_setup_check(int argc, char *argv[]) {
2600 bool found_deserialize = false;
2601
2602 /* Determine if this is a reexecution or normal bootup. We do the full command line parsing much
2603 * later, so let's just have a quick peek here. Note that if we have switched root, do all the
2604 * special setup things anyway, even if in that case we also do deserialization. */
2605
2606 for (int i = 1; i < argc; i++)
2607 if (streq(argv[i], "--switched-root"))
2608 return false; /* If we switched root, don't skip the setup. */
2609 else if (streq(argv[i], "--deserialize"))
2610 found_deserialize = true;
2611
2612 return found_deserialize; /* When we are deserializing, then we are reexecuting, hence avoid the extensive setup */
2613 }
2614
2615 static int save_env(void) {
2616 char **l;
2617
2618 l = strv_copy(environ);
2619 if (!l)
2620 return -ENOMEM;
2621
2622 strv_free_and_replace(saved_env, l);
2623 return 0;
2624 }
2625
2626 int main(int argc, char *argv[]) {
2627
2628 dual_timestamp initrd_timestamp = DUAL_TIMESTAMP_NULL, userspace_timestamp = DUAL_TIMESTAMP_NULL, kernel_timestamp = DUAL_TIMESTAMP_NULL,
2629 security_start_timestamp = DUAL_TIMESTAMP_NULL, security_finish_timestamp = DUAL_TIMESTAMP_NULL;
2630 struct rlimit saved_rlimit_nofile = RLIMIT_MAKE_CONST(0),
2631 saved_rlimit_memlock = RLIMIT_MAKE_CONST(RLIM_INFINITY); /* The original rlimits we passed
2632 * in. Note we use different values
2633 * for the two that indicate whether
2634 * these fields are initialized! */
2635 bool skip_setup, loaded_policy = false, queue_default_job = false, first_boot = false, reexecute = false;
2636 char *switch_root_dir = NULL, *switch_root_init = NULL;
2637 usec_t before_startup, after_startup;
2638 static char systemd[] = "systemd";
2639 const char *shutdown_verb = NULL, *error_message = NULL;
2640 int r, retval = EXIT_FAILURE;
2641 Manager *m = NULL;
2642 FDSet *fds = NULL;
2643
2644 /* SysV compatibility: redirect init → telinit */
2645 redirect_telinit(argc, argv);
2646
2647 /* Take timestamps early on */
2648 dual_timestamp_from_monotonic(&kernel_timestamp, 0);
2649 dual_timestamp_get(&userspace_timestamp);
2650
2651 /* Figure out whether we need to do initialize the system, or if we already did that because we are
2652 * reexecuting. */
2653 skip_setup = early_skip_setup_check(argc, argv);
2654
2655 /* If we get started via the /sbin/init symlink then we are called 'init'. After a subsequent
2656 * reexecution we are then called 'systemd'. That is confusing, hence let's call us systemd
2657 * right-away. */
2658 program_invocation_short_name = systemd;
2659 (void) prctl(PR_SET_NAME, systemd);
2660
2661 /* Save the original command line */
2662 save_argc_argv(argc, argv);
2663
2664 /* Save the original environment as we might need to restore it if we're requested to execute another
2665 * system manager later. */
2666 r = save_env();
2667 if (r < 0) {
2668 error_message = "Failed to copy environment block";
2669 goto finish;
2670 }
2671
2672 /* Make sure that if the user says "syslog" we actually log to the journal. */
2673 log_set_upgrade_syslog_to_journal(true);
2674
2675 if (getpid_cached() == 1) {
2676 /* When we run as PID 1 force system mode */
2677 arg_system = true;
2678
2679 /* Disable the umask logic */
2680 umask(0);
2681
2682 /* Make sure that at least initially we do not ever log to journald/syslogd, because it might
2683 * not be activated yet (even though the log socket for it exists). */
2684 log_set_prohibit_ipc(true);
2685
2686 /* Always reopen /dev/console when running as PID 1 or one of its pre-execve() children. This
2687 * is important so that we never end up logging to any foreign stderr, for example if we have
2688 * to log in a child process right before execve()'ing the actual binary, at a point in time
2689 * where socket activation stderr/stdout area already set up. */
2690 log_set_always_reopen_console(true);
2691
2692 if (detect_container() <= 0) {
2693
2694 /* Running outside of a container as PID 1 */
2695 log_set_target(LOG_TARGET_KMSG);
2696 log_open();
2697
2698 if (in_initrd())
2699 initrd_timestamp = userspace_timestamp;
2700
2701 if (!skip_setup) {
2702 r = mount_setup_early();
2703 if (r < 0) {
2704 error_message = "Failed to mount early API filesystems";
2705 goto finish;
2706 }
2707
2708 /* Let's open the log backend a second time, in case the first time didn't
2709 * work. Quite possibly we have mounted /dev just now, so /dev/kmsg became
2710 * available, and it previously wasn't. */
2711 log_open();
2712
2713 disable_printk_ratelimit();
2714
2715 r = initialize_security(
2716 &loaded_policy,
2717 &security_start_timestamp,
2718 &security_finish_timestamp,
2719 &error_message);
2720 if (r < 0)
2721 goto finish;
2722 }
2723
2724 if (mac_selinux_init() < 0) {
2725 error_message = "Failed to initialize SELinux support";
2726 goto finish;
2727 }
2728
2729 if (!skip_setup)
2730 initialize_clock();
2731
2732 /* Set the default for later on, but don't actually open the logs like this for
2733 * now. Note that if we are transitioning from the initrd there might still be
2734 * journal fd open, and we shouldn't attempt opening that before we parsed
2735 * /proc/cmdline which might redirect output elsewhere. */
2736 log_set_target(LOG_TARGET_JOURNAL_OR_KMSG);
2737
2738 } else {
2739 /* Running inside a container, as PID 1 */
2740 log_set_target(LOG_TARGET_CONSOLE);
2741 log_open();
2742
2743 /* For later on, see above... */
2744 log_set_target(LOG_TARGET_JOURNAL);
2745
2746 /* clear the kernel timestamp, because we are in a container */
2747 kernel_timestamp = DUAL_TIMESTAMP_NULL;
2748 }
2749
2750 initialize_coredump(skip_setup);
2751
2752 r = fixup_environment();
2753 if (r < 0) {
2754 log_emergency_errno(r, "Failed to fix up PID 1 environment: %m");
2755 error_message = "Failed to fix up PID1 environment";
2756 goto finish;
2757 }
2758
2759 /* Try to figure out if we can use colors with the console. No need to do that for user
2760 * instances since they never log into the console. */
2761 log_show_color(colors_enabled());
2762
2763 r = make_null_stdio();
2764 if (r < 0)
2765 log_warning_errno(r, "Failed to redirect standard streams to /dev/null, ignoring: %m");
2766
2767 /* Load the kernel modules early. */
2768 if (!skip_setup)
2769 kmod_setup();
2770
2771 /* Mount /proc, /sys and friends, so that /proc/cmdline and /proc/$PID/fd is available. */
2772 r = mount_setup(loaded_policy, skip_setup);
2773 if (r < 0) {
2774 error_message = "Failed to mount API filesystems";
2775 goto finish;
2776 }
2777
2778 /* The efivarfs is now mounted, let's read the random seed off it */
2779 (void) efi_take_random_seed();
2780
2781 /* Cache command-line options passed from EFI variables */
2782 if (!skip_setup)
2783 (void) cache_efi_options_variable();
2784 } else {
2785 /* Running as user instance */
2786 arg_system = false;
2787 log_set_target(LOG_TARGET_AUTO);
2788 log_open();
2789
2790 /* clear the kernel timestamp, because we are not PID 1 */
2791 kernel_timestamp = DUAL_TIMESTAMP_NULL;
2792
2793 if (mac_selinux_init() < 0) {
2794 error_message = "Failed to initialize SELinux support";
2795 goto finish;
2796 }
2797 }
2798
2799 /* Save the original RLIMIT_NOFILE/RLIMIT_MEMLOCK so that we can reset it later when
2800 * transitioning from the initrd to the main systemd or suchlike. */
2801 save_rlimits(&saved_rlimit_nofile, &saved_rlimit_memlock);
2802
2803 /* Reset all signal handlers. */
2804 (void) reset_all_signal_handlers();
2805 (void) ignore_signals(SIGNALS_IGNORE);
2806
2807 (void) parse_configuration(&saved_rlimit_nofile, &saved_rlimit_memlock);
2808
2809 r = parse_argv(argc, argv);
2810 if (r < 0) {
2811 error_message = "Failed to parse commandline arguments";
2812 goto finish;
2813 }
2814
2815 r = safety_checks();
2816 if (r < 0)
2817 goto finish;
2818
2819 if (IN_SET(arg_action, ACTION_TEST, ACTION_HELP, ACTION_DUMP_CONFIGURATION_ITEMS, ACTION_DUMP_BUS_PROPERTIES, ACTION_BUS_INTROSPECT))
2820 (void) pager_open(arg_pager_flags);
2821
2822 if (arg_action != ACTION_RUN)
2823 skip_setup = true;
2824
2825 if (arg_action == ACTION_HELP) {
2826 retval = help() < 0 ? EXIT_FAILURE : EXIT_SUCCESS;
2827 goto finish;
2828 } else if (arg_action == ACTION_VERSION) {
2829 retval = version();
2830 goto finish;
2831 } else if (arg_action == ACTION_DUMP_CONFIGURATION_ITEMS) {
2832 unit_dump_config_items(stdout);
2833 retval = EXIT_SUCCESS;
2834 goto finish;
2835 } else if (arg_action == ACTION_DUMP_BUS_PROPERTIES) {
2836 dump_bus_properties(stdout);
2837 retval = EXIT_SUCCESS;
2838 goto finish;
2839 } else if (arg_action == ACTION_BUS_INTROSPECT) {
2840 r = bus_manager_introspect_implementations(stdout, arg_bus_introspect);
2841 retval = r >= 0 ? EXIT_SUCCESS : EXIT_FAILURE;
2842 goto finish;
2843 }
2844
2845 assert_se(IN_SET(arg_action, ACTION_RUN, ACTION_TEST));
2846
2847 /* Move out of the way, so that we won't block unmounts */
2848 assert_se(chdir("/") == 0);
2849
2850 if (arg_action == ACTION_RUN) {
2851 if (!skip_setup) {
2852 /* Apply the systemd.clock_usec= kernel command line switch */
2853 apply_clock_update();
2854
2855 /* Apply random seed from kernel command line */
2856 cmdline_take_random_seed();
2857 }
2858
2859 /* A core pattern might have been specified via the cmdline. */
2860 initialize_core_pattern(skip_setup);
2861
2862 /* Close logging fds, in order not to confuse collecting passed fds and terminal logic below */
2863 log_close();
2864
2865 /* Remember open file descriptors for later deserialization */
2866 r = collect_fds(&fds, &error_message);
2867 if (r < 0)
2868 goto finish;
2869
2870 /* Give up any control of the console, but make sure its initialized. */
2871 setup_console_terminal(skip_setup);
2872
2873 /* Open the logging devices, if possible and necessary */
2874 log_open();
2875 }
2876
2877 log_execution_mode(&first_boot);
2878
2879 r = initialize_runtime(skip_setup,
2880 first_boot,
2881 &saved_rlimit_nofile,
2882 &saved_rlimit_memlock,
2883 &error_message);
2884 if (r < 0)
2885 goto finish;
2886
2887 r = manager_new(arg_system ? UNIT_FILE_SYSTEM : UNIT_FILE_USER,
2888 arg_action == ACTION_TEST ? MANAGER_TEST_FULL : 0,
2889 &m);
2890 if (r < 0) {
2891 log_emergency_errno(r, "Failed to allocate manager object: %m");
2892 error_message = "Failed to allocate manager object";
2893 goto finish;
2894 }
2895
2896 m->timestamps[MANAGER_TIMESTAMP_KERNEL] = kernel_timestamp;
2897 m->timestamps[MANAGER_TIMESTAMP_INITRD] = initrd_timestamp;
2898 m->timestamps[MANAGER_TIMESTAMP_USERSPACE] = userspace_timestamp;
2899 m->timestamps[manager_timestamp_initrd_mangle(MANAGER_TIMESTAMP_SECURITY_START)] = security_start_timestamp;
2900 m->timestamps[manager_timestamp_initrd_mangle(MANAGER_TIMESTAMP_SECURITY_FINISH)] = security_finish_timestamp;
2901
2902 set_manager_defaults(m);
2903 set_manager_settings(m);
2904 manager_set_first_boot(m, first_boot);
2905
2906 /* Remember whether we should queue the default job */
2907 queue_default_job = !arg_serialization || arg_switched_root;
2908
2909 before_startup = now(CLOCK_MONOTONIC);
2910
2911 r = manager_startup(m, arg_serialization, fds, /* root= */ NULL);
2912 if (r < 0) {
2913 error_message = "Failed to start up manager";
2914 goto finish;
2915 }
2916
2917 /* This will close all file descriptors that were opened, but not claimed by any unit. */
2918 fds = fdset_free(fds);
2919 arg_serialization = safe_fclose(arg_serialization);
2920
2921 if (queue_default_job) {
2922 r = do_queue_default_job(m, &error_message);
2923 if (r < 0)
2924 goto finish;
2925 }
2926
2927 after_startup = now(CLOCK_MONOTONIC);
2928
2929 log_full(arg_action == ACTION_TEST ? LOG_INFO : LOG_DEBUG,
2930 "Loaded units and determined initial transaction in %s.",
2931 FORMAT_TIMESPAN(after_startup - before_startup, 100 * USEC_PER_MSEC));
2932
2933 if (arg_action == ACTION_TEST) {
2934 manager_test_summary(m);
2935 retval = EXIT_SUCCESS;
2936 goto finish;
2937 }
2938
2939 (void) invoke_main_loop(m,
2940 &saved_rlimit_nofile,
2941 &saved_rlimit_memlock,
2942 &reexecute,
2943 &retval,
2944 &shutdown_verb,
2945 &fds,
2946 &switch_root_dir,
2947 &switch_root_init,
2948 &error_message);
2949
2950 finish:
2951 pager_close();
2952
2953 if (m) {
2954 arg_reboot_watchdog = manager_get_watchdog(m, WATCHDOG_REBOOT);
2955 arg_kexec_watchdog = manager_get_watchdog(m, WATCHDOG_KEXEC);
2956 m = manager_free(m);
2957 }
2958
2959 mac_selinux_finish();
2960
2961 if (reexecute)
2962 do_reexecute(argc, argv,
2963 &saved_rlimit_nofile,
2964 &saved_rlimit_memlock,
2965 fds,
2966 switch_root_dir,
2967 switch_root_init,
2968 &error_message); /* This only returns if reexecution failed */
2969
2970 arg_serialization = safe_fclose(arg_serialization);
2971 fds = fdset_free(fds);
2972
2973 saved_env = strv_free(saved_env);
2974
2975 #if HAVE_VALGRIND_VALGRIND_H
2976 /* If we are PID 1 and running under valgrind, then let's exit
2977 * here explicitly. valgrind will only generate nice output on
2978 * exit(), not on exec(), hence let's do the former not the
2979 * latter here. */
2980 if (getpid_cached() == 1 && RUNNING_ON_VALGRIND) {
2981 /* Cleanup watchdog_device strings for valgrind. We need them
2982 * in become_shutdown() so normally we cannot free them yet. */
2983 watchdog_free_device();
2984 arg_watchdog_device = mfree(arg_watchdog_device);
2985 reset_arguments();
2986 return retval;
2987 }
2988 #endif
2989
2990 #if HAS_FEATURE_ADDRESS_SANITIZER
2991 __lsan_do_leak_check();
2992 #endif
2993
2994 if (shutdown_verb) {
2995 r = become_shutdown(shutdown_verb, retval);
2996 log_error_errno(r, "Failed to execute shutdown binary, %s: %m", getpid_cached() == 1 ? "freezing" : "quitting");
2997 error_message = "Failed to execute shutdown binary";
2998 }
2999
3000 watchdog_free_device();
3001 arg_watchdog_device = mfree(arg_watchdog_device);
3002
3003 if (getpid_cached() == 1) {
3004 if (error_message)
3005 manager_status_printf(NULL, STATUS_TYPE_EMERGENCY,
3006 ANSI_HIGHLIGHT_RED "!!!!!!" ANSI_NORMAL,
3007 "%s.", error_message);
3008 freeze_or_exit_or_reboot();
3009 }
3010
3011 reset_arguments();
3012 return retval;
3013 }
Unnamed repository; edit this file 'description' to name the repository.