]> 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
core: implement OOMPolicy= and watch cgroups for OOM killings
[thirdparty/systemd.git] / src / core / main.c
1 /* SPDX-License-Identifier: LGPL-2.1+ */
2
3 #include <errno.h>
4 #include <fcntl.h>
5 #include <getopt.h>
6 #include <signal.h>
7 #include <stdio.h>
8 #include <string.h>
9 #include <sys/mount.h>
10 #include <sys/prctl.h>
11 #include <sys/reboot.h>
12 #include <sys/stat.h>
13 #include <unistd.h>
14 #if HAVE_SECCOMP
15 #include <seccomp.h>
16 #endif
17 #if HAVE_VALGRIND_VALGRIND_H
18 #include <valgrind/valgrind.h>
19 #endif
20
21 #include "sd-bus.h"
22 #include "sd-daemon.h"
23 #include "sd-messages.h"
24
25 #include "alloc-util.h"
26 #include "architecture.h"
27 #include "build.h"
28 #include "bus-error.h"
29 #include "bus-util.h"
30 #include "capability-util.h"
31 #include "cgroup-util.h"
32 #include "clock-util.h"
33 #include "conf-parser.h"
34 #include "cpu-set-util.h"
35 #include "dbus.h"
36 #include "dbus-manager.h"
37 #include "def.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 "hostname-setup.h"
47 #include "ima-setup.h"
48 #include "killall.h"
49 #include "kmod-setup.h"
50 #include "limits-util.h"
51 #include "load-fragment.h"
52 #include "log.h"
53 #include "loopback-setup.h"
54 #include "machine-id-setup.h"
55 #include "manager.h"
56 #include "missing.h"
57 #include "mount-setup.h"
58 #include "os-util.h"
59 #include "pager.h"
60 #include "parse-util.h"
61 #include "path-util.h"
62 #include "pretty-print.h"
63 #include "proc-cmdline.h"
64 #include "process-util.h"
65 #include "raw-clone.h"
66 #include "rlimit-util.h"
67 #if HAVE_SECCOMP
68 #include "seccomp-util.h"
69 #endif
70 #include "selinux-setup.h"
71 #include "selinux-util.h"
72 #include "signal-util.h"
73 #include "smack-setup.h"
74 #include "special.h"
75 #include "stat-util.h"
76 #include "stdio-util.h"
77 #include "strv.h"
78 #include "switch-root.h"
79 #include "sysctl-util.h"
80 #include "terminal-util.h"
81 #include "umask-util.h"
82 #include "user-util.h"
83 #include "util.h"
84 #include "virt.h"
85 #include "watchdog.h"
86
87 #if HAS_FEATURE_ADDRESS_SANITIZER
88 #include <sanitizer/lsan_interface.h>
89 #endif
90
91 static enum {
92 ACTION_RUN,
93 ACTION_HELP,
94 ACTION_VERSION,
95 ACTION_TEST,
96 ACTION_DUMP_CONFIGURATION_ITEMS,
97 ACTION_DUMP_BUS_PROPERTIES,
98 } arg_action = ACTION_RUN;
99 static char *arg_default_unit = NULL;
100 static bool arg_system = false;
101 static bool arg_dump_core = true;
102 static int arg_crash_chvt = -1;
103 static bool arg_crash_shell = false;
104 static bool arg_crash_reboot = false;
105 static char *arg_confirm_spawn = NULL;
106 static ShowStatus arg_show_status = _SHOW_STATUS_INVALID;
107 static bool arg_switched_root = false;
108 static PagerFlags arg_pager_flags = 0;
109 static bool arg_service_watchdogs = true;
110 static ExecOutput arg_default_std_output = EXEC_OUTPUT_JOURNAL;
111 static ExecOutput arg_default_std_error = EXEC_OUTPUT_INHERIT;
112 static usec_t arg_default_restart_usec = DEFAULT_RESTART_USEC;
113 static usec_t arg_default_timeout_start_usec = DEFAULT_TIMEOUT_USEC;
114 static usec_t arg_default_timeout_stop_usec = DEFAULT_TIMEOUT_USEC;
115 static usec_t arg_default_start_limit_interval = DEFAULT_START_LIMIT_INTERVAL;
116 static unsigned arg_default_start_limit_burst = DEFAULT_START_LIMIT_BURST;
117 static usec_t arg_runtime_watchdog = 0;
118 static usec_t arg_shutdown_watchdog = 10 * USEC_PER_MINUTE;
119 static char *arg_early_core_pattern = NULL;
120 static char *arg_watchdog_device = NULL;
121 static char **arg_default_environment = NULL;
122 static struct rlimit *arg_default_rlimit[_RLIMIT_MAX] = {};
123 static uint64_t arg_capability_bounding_set = CAP_ALL;
124 static bool arg_no_new_privs = false;
125 static nsec_t arg_timer_slack_nsec = NSEC_INFINITY;
126 static usec_t arg_default_timer_accuracy_usec = 1 * USEC_PER_MINUTE;
127 static Set* arg_syscall_archs = NULL;
128 static FILE* arg_serialization = NULL;
129 static int arg_default_cpu_accounting = -1;
130 static bool arg_default_io_accounting = false;
131 static bool arg_default_ip_accounting = false;
132 static bool arg_default_blockio_accounting = false;
133 static bool arg_default_memory_accounting = MEMORY_ACCOUNTING_DEFAULT;
134 static bool arg_default_tasks_accounting = true;
135 static uint64_t arg_default_tasks_max = UINT64_MAX;
136 static sd_id128_t arg_machine_id = {};
137 static EmergencyAction arg_cad_burst_action = EMERGENCY_ACTION_REBOOT_FORCE;
138 static OOMPolicy arg_default_oom_policy = OOM_STOP;
139
140 _noreturn_ static void freeze_or_exit_or_reboot(void) {
141
142 /* If we are running in a container, let's prefer exiting, after all we can propagate an exit code to
143 * the container manager, and thus inform it that something went wrong. */
144 if (detect_container() > 0) {
145 log_emergency("Exiting PID 1...");
146 _exit(EXIT_EXCEPTION);
147 }
148
149 if (arg_crash_reboot) {
150 log_notice("Rebooting in 10s...");
151 (void) sleep(10);
152
153 log_notice("Rebooting now...");
154 (void) reboot(RB_AUTOBOOT);
155 log_emergency_errno(errno, "Failed to reboot: %m");
156 }
157
158 log_emergency("Freezing execution.");
159 freeze();
160 }
161
162 _noreturn_ static void crash(int sig) {
163 struct sigaction sa;
164 pid_t pid;
165
166 if (getpid_cached() != 1)
167 /* Pass this on immediately, if this is not PID 1 */
168 (void) raise(sig);
169 else if (!arg_dump_core)
170 log_emergency("Caught <%s>, not dumping core.", signal_to_string(sig));
171 else {
172 sa = (struct sigaction) {
173 .sa_handler = nop_signal_handler,
174 .sa_flags = SA_NOCLDSTOP|SA_RESTART,
175 };
176
177 /* We want to wait for the core process, hence let's enable SIGCHLD */
178 (void) sigaction(SIGCHLD, &sa, NULL);
179
180 pid = raw_clone(SIGCHLD);
181 if (pid < 0)
182 log_emergency_errno(errno, "Caught <%s>, cannot fork for core dump: %m", signal_to_string(sig));
183 else if (pid == 0) {
184 /* Enable default signal handler for core dump */
185
186 sa = (struct sigaction) {
187 .sa_handler = SIG_DFL,
188 };
189 (void) sigaction(sig, &sa, NULL);
190
191 /* Don't limit the coredump size */
192 (void) setrlimit(RLIMIT_CORE, &RLIMIT_MAKE_CONST(RLIM_INFINITY));
193
194 /* Just to be sure... */
195 (void) chdir("/");
196
197 /* Raise the signal again */
198 pid = raw_getpid();
199 (void) kill(pid, sig); /* raise() would kill the parent */
200
201 assert_not_reached("We shouldn't be here...");
202 _exit(EXIT_EXCEPTION);
203 } else {
204 siginfo_t status;
205 int r;
206
207 /* Order things nicely. */
208 r = wait_for_terminate(pid, &status);
209 if (r < 0)
210 log_emergency_errno(r, "Caught <%s>, waitpid() failed: %m", signal_to_string(sig));
211 else if (status.si_code != CLD_DUMPED)
212 log_emergency("Caught <%s>, core dump failed (child "PID_FMT", code=%s, status=%i/%s).",
213 signal_to_string(sig),
214 pid, sigchld_code_to_string(status.si_code),
215 status.si_status,
216 strna(status.si_code == CLD_EXITED
217 ? exit_status_to_string(status.si_status, EXIT_STATUS_MINIMAL)
218 : signal_to_string(status.si_status)));
219 else
220 log_emergency("Caught <%s>, dumped core as pid "PID_FMT".", signal_to_string(sig), pid);
221 }
222 }
223
224 if (arg_crash_chvt >= 0)
225 (void) chvt(arg_crash_chvt);
226
227 sa = (struct sigaction) {
228 .sa_handler = SIG_IGN,
229 .sa_flags = SA_NOCLDSTOP|SA_NOCLDWAIT|SA_RESTART,
230 };
231
232 /* Let the kernel reap children for us */
233 (void) sigaction(SIGCHLD, &sa, NULL);
234
235 if (arg_crash_shell) {
236 log_notice("Executing crash shell in 10s...");
237 (void) sleep(10);
238
239 pid = raw_clone(SIGCHLD);
240 if (pid < 0)
241 log_emergency_errno(errno, "Failed to fork off crash shell: %m");
242 else if (pid == 0) {
243 (void) setsid();
244 (void) make_console_stdio();
245 (void) rlimit_nofile_safe();
246 (void) execle("/bin/sh", "/bin/sh", NULL, environ);
247
248 log_emergency_errno(errno, "execle() failed: %m");
249 _exit(EXIT_EXCEPTION);
250 } else {
251 log_info("Spawned crash shell as PID "PID_FMT".", pid);
252 (void) wait_for_terminate(pid, NULL);
253 }
254 }
255
256 freeze_or_exit_or_reboot();
257 }
258
259 static void install_crash_handler(void) {
260 static const struct sigaction sa = {
261 .sa_handler = crash,
262 .sa_flags = SA_NODEFER, /* So that we can raise the signal again from the signal handler */
263 };
264 int r;
265
266 /* We ignore the return value here, since, we don't mind if we
267 * cannot set up a crash handler */
268 r = sigaction_many(&sa, SIGNALS_CRASH_HANDLER, -1);
269 if (r < 0)
270 log_debug_errno(r, "I had trouble setting up the crash handler, ignoring: %m");
271 }
272
273 static int console_setup(void) {
274 _cleanup_close_ int tty_fd = -1;
275 int r;
276
277 tty_fd = open_terminal("/dev/console", O_WRONLY|O_NOCTTY|O_CLOEXEC);
278 if (tty_fd < 0)
279 return log_error_errno(tty_fd, "Failed to open /dev/console: %m");
280
281 /* We don't want to force text mode. plymouth may be showing
282 * pictures already from initrd. */
283 r = reset_terminal_fd(tty_fd, false);
284 if (r < 0)
285 return log_error_errno(r, "Failed to reset /dev/console: %m");
286
287 return 0;
288 }
289
290 static int parse_crash_chvt(const char *value) {
291 int b;
292
293 if (safe_atoi(value, &arg_crash_chvt) >= 0)
294 return 0;
295
296 b = parse_boolean(value);
297 if (b < 0)
298 return b;
299
300 if (b > 0)
301 arg_crash_chvt = 0; /* switch to where kmsg goes */
302 else
303 arg_crash_chvt = -1; /* turn off switching */
304
305 return 0;
306 }
307
308 static int parse_confirm_spawn(const char *value, char **console) {
309 char *s;
310 int r;
311
312 r = value ? parse_boolean(value) : 1;
313 if (r == 0) {
314 *console = NULL;
315 return 0;
316 }
317
318 if (r > 0) /* on with default tty */
319 s = strdup("/dev/console");
320 else if (is_path(value)) /* on with fully qualified path */
321 s = strdup(value);
322 else /* on with only a tty file name, not a fully qualified path */
323 s = strjoin("/dev/", value);
324 if (!s)
325 return -ENOMEM;
326
327 *console = s;
328 return 0;
329 }
330
331 static int set_machine_id(const char *m) {
332 sd_id128_t t;
333 assert(m);
334
335 if (sd_id128_from_string(m, &t) < 0)
336 return -EINVAL;
337
338 if (sd_id128_is_null(t))
339 return -EINVAL;
340
341 arg_machine_id = t;
342 return 0;
343 }
344
345 static int parse_proc_cmdline_item(const char *key, const char *value, void *data) {
346
347 int r;
348
349 assert(key);
350
351 if (STR_IN_SET(key, "systemd.unit", "rd.systemd.unit")) {
352
353 if (proc_cmdline_value_missing(key, value))
354 return 0;
355
356 if (!unit_name_is_valid(value, UNIT_NAME_PLAIN|UNIT_NAME_INSTANCE))
357 log_warning("Unit name specified on %s= is not valid, ignoring: %s", key, value);
358 else if (in_initrd() == !!startswith(key, "rd.")) {
359 if (free_and_strdup(&arg_default_unit, value) < 0)
360 return log_oom();
361 }
362
363 } else if (proc_cmdline_key_streq(key, "systemd.dump_core")) {
364
365 r = value ? parse_boolean(value) : true;
366 if (r < 0)
367 log_warning_errno(r, "Failed to parse dump core switch %s, ignoring: %m", value);
368 else
369 arg_dump_core = r;
370
371 } else if (proc_cmdline_key_streq(key, "systemd.early_core_pattern")) {
372
373 if (proc_cmdline_value_missing(key, value))
374 return 0;
375
376 if (path_is_absolute(value))
377 (void) parse_path_argument_and_warn(value, false, &arg_early_core_pattern);
378 else
379 log_warning("Specified core pattern '%s' is not an absolute path, ignoring.", value);
380
381 } else if (proc_cmdline_key_streq(key, "systemd.crash_chvt")) {
382
383 if (!value)
384 arg_crash_chvt = 0; /* turn on */
385 else {
386 r = parse_crash_chvt(value);
387 if (r < 0)
388 log_warning_errno(r, "Failed to parse crash chvt switch %s, ignoring: %m", value);
389 }
390
391 } else if (proc_cmdline_key_streq(key, "systemd.crash_shell")) {
392
393 r = value ? parse_boolean(value) : true;
394 if (r < 0)
395 log_warning_errno(r, "Failed to parse crash shell switch %s, ignoring: %m", value);
396 else
397 arg_crash_shell = r;
398
399 } else if (proc_cmdline_key_streq(key, "systemd.crash_reboot")) {
400
401 r = value ? parse_boolean(value) : true;
402 if (r < 0)
403 log_warning_errno(r, "Failed to parse crash reboot switch %s, ignoring: %m", value);
404 else
405 arg_crash_reboot = r;
406
407 } else if (proc_cmdline_key_streq(key, "systemd.confirm_spawn")) {
408 char *s;
409
410 r = parse_confirm_spawn(value, &s);
411 if (r < 0)
412 log_warning_errno(r, "Failed to parse confirm_spawn switch %s, ignoring: %m", value);
413 else
414 free_and_replace(arg_confirm_spawn, s);
415
416 } else if (proc_cmdline_key_streq(key, "systemd.service_watchdogs")) {
417
418 r = value ? parse_boolean(value) : true;
419 if (r < 0)
420 log_warning_errno(r, "Failed to parse service watchdog switch %s, ignoring: %m", value);
421 else
422 arg_service_watchdogs = r;
423
424 } else if (proc_cmdline_key_streq(key, "systemd.show_status")) {
425
426 if (value) {
427 r = parse_show_status(value, &arg_show_status);
428 if (r < 0)
429 log_warning_errno(r, "Failed to parse show status switch %s, ignoring: %m", value);
430 } else
431 arg_show_status = SHOW_STATUS_YES;
432
433 } else if (proc_cmdline_key_streq(key, "systemd.default_standard_output")) {
434
435 if (proc_cmdline_value_missing(key, value))
436 return 0;
437
438 r = exec_output_from_string(value);
439 if (r < 0)
440 log_warning_errno(r, "Failed to parse default standard output switch %s, ignoring: %m", value);
441 else
442 arg_default_std_output = r;
443
444 } else if (proc_cmdline_key_streq(key, "systemd.default_standard_error")) {
445
446 if (proc_cmdline_value_missing(key, value))
447 return 0;
448
449 r = exec_output_from_string(value);
450 if (r < 0)
451 log_warning_errno(r, "Failed to parse default standard error switch %s, ignoring: %m", value);
452 else
453 arg_default_std_error = r;
454
455 } else if (streq(key, "systemd.setenv")) {
456
457 if (proc_cmdline_value_missing(key, value))
458 return 0;
459
460 if (env_assignment_is_valid(value)) {
461 char **env;
462
463 env = strv_env_set(arg_default_environment, value);
464 if (!env)
465 return log_oom();
466
467 arg_default_environment = env;
468 } else
469 log_warning("Environment variable name '%s' is not valid. Ignoring.", value);
470
471 } else if (proc_cmdline_key_streq(key, "systemd.machine_id")) {
472
473 if (proc_cmdline_value_missing(key, value))
474 return 0;
475
476 r = set_machine_id(value);
477 if (r < 0)
478 log_warning_errno(r, "MachineID '%s' is not valid, ignoring: %m", value);
479
480 } else if (proc_cmdline_key_streq(key, "systemd.default_timeout_start_sec")) {
481
482 if (proc_cmdline_value_missing(key, value))
483 return 0;
484
485 r = parse_sec(value, &arg_default_timeout_start_usec);
486 if (r < 0)
487 log_warning_errno(r, "Failed to parse default start timeout '%s', ignoring: %m", value);
488
489 if (arg_default_timeout_start_usec <= 0)
490 arg_default_timeout_start_usec = USEC_INFINITY;
491
492 } else if (proc_cmdline_key_streq(key, "systemd.watchdog_device")) {
493
494 if (proc_cmdline_value_missing(key, value))
495 return 0;
496
497 (void) parse_path_argument_and_warn(value, false, &arg_watchdog_device);
498
499 } else if (streq(key, "quiet") && !value) {
500
501 if (arg_show_status == _SHOW_STATUS_INVALID)
502 arg_show_status = SHOW_STATUS_AUTO;
503
504 } else if (streq(key, "debug") && !value) {
505
506 /* Note that log_parse_environment() handles 'debug'
507 * too, and sets the log level to LOG_DEBUG. */
508
509 if (detect_container() > 0)
510 log_set_target(LOG_TARGET_CONSOLE);
511
512 } else if (!value) {
513 const char *target;
514
515 /* SysV compatibility */
516 target = runlevel_to_target(key);
517 if (target)
518 return free_and_strdup(&arg_default_unit, target);
519 }
520
521 return 0;
522 }
523
524 #define DEFINE_SETTER(name, func, descr) \
525 static int name(const char *unit, \
526 const char *filename, \
527 unsigned line, \
528 const char *section, \
529 unsigned section_line, \
530 const char *lvalue, \
531 int ltype, \
532 const char *rvalue, \
533 void *data, \
534 void *userdata) { \
535 \
536 int r; \
537 \
538 assert(filename); \
539 assert(lvalue); \
540 assert(rvalue); \
541 \
542 r = func(rvalue); \
543 if (r < 0) \
544 log_syntax(unit, LOG_ERR, filename, line, r, \
545 "Invalid " descr "'%s': %m", \
546 rvalue); \
547 \
548 return 0; \
549 }
550
551 DEFINE_SETTER(config_parse_level2, log_set_max_level_from_string, "log level");
552 DEFINE_SETTER(config_parse_target, log_set_target_from_string, "target");
553 DEFINE_SETTER(config_parse_color, log_show_color_from_string, "color" );
554 DEFINE_SETTER(config_parse_location, log_show_location_from_string, "location");
555
556 static int config_parse_cpu_affinity2(
557 const char *unit,
558 const char *filename,
559 unsigned line,
560 const char *section,
561 unsigned section_line,
562 const char *lvalue,
563 int ltype,
564 const char *rvalue,
565 void *data,
566 void *userdata) {
567
568 _cleanup_cpu_free_ cpu_set_t *c = NULL;
569 int ncpus;
570
571 ncpus = parse_cpu_set_and_warn(rvalue, &c, unit, filename, line, lvalue);
572 if (ncpus < 0)
573 return ncpus;
574
575 if (sched_setaffinity(0, CPU_ALLOC_SIZE(ncpus), c) < 0)
576 log_warning_errno(errno, "Failed to set CPU affinity: %m");
577
578 return 0;
579 }
580
581 static int config_parse_show_status(
582 const char* unit,
583 const char *filename,
584 unsigned line,
585 const char *section,
586 unsigned section_line,
587 const char *lvalue,
588 int ltype,
589 const char *rvalue,
590 void *data,
591 void *userdata) {
592
593 int k;
594 ShowStatus *b = data;
595
596 assert(filename);
597 assert(lvalue);
598 assert(rvalue);
599 assert(data);
600
601 k = parse_show_status(rvalue, b);
602 if (k < 0) {
603 log_syntax(unit, LOG_ERR, filename, line, k, "Failed to parse show status setting, ignoring: %s", rvalue);
604 return 0;
605 }
606
607 return 0;
608 }
609
610 static int config_parse_output_restricted(
611 const char* unit,
612 const char *filename,
613 unsigned line,
614 const char *section,
615 unsigned section_line,
616 const char *lvalue,
617 int ltype,
618 const char *rvalue,
619 void *data,
620 void *userdata) {
621
622 ExecOutput t, *eo = data;
623
624 assert(filename);
625 assert(lvalue);
626 assert(rvalue);
627 assert(data);
628
629 t = exec_output_from_string(rvalue);
630 if (t < 0) {
631 log_syntax(unit, LOG_ERR, filename, line, 0, "Failed to parse output type, ignoring: %s", rvalue);
632 return 0;
633 }
634
635 if (IN_SET(t, EXEC_OUTPUT_SOCKET, EXEC_OUTPUT_NAMED_FD, EXEC_OUTPUT_FILE, EXEC_OUTPUT_FILE_APPEND)) {
636 log_syntax(unit, LOG_ERR, filename, line, 0, "Standard output types socket, fd:, file:, append: are not supported as defaults, ignoring: %s", rvalue);
637 return 0;
638 }
639
640 *eo = t;
641 return 0;
642 }
643
644 static int config_parse_crash_chvt(
645 const char* unit,
646 const char *filename,
647 unsigned line,
648 const char *section,
649 unsigned section_line,
650 const char *lvalue,
651 int ltype,
652 const char *rvalue,
653 void *data,
654 void *userdata) {
655
656 int r;
657
658 assert(filename);
659 assert(lvalue);
660 assert(rvalue);
661
662 r = parse_crash_chvt(rvalue);
663 if (r < 0) {
664 log_syntax(unit, LOG_ERR, filename, line, r, "Failed to parse CrashChangeVT= setting, ignoring: %s", rvalue);
665 return 0;
666 }
667
668 return 0;
669 }
670
671 static int parse_config_file(void) {
672
673 const ConfigTableItem items[] = {
674 { "Manager", "LogLevel", config_parse_level2, 0, NULL },
675 { "Manager", "LogTarget", config_parse_target, 0, NULL },
676 { "Manager", "LogColor", config_parse_color, 0, NULL },
677 { "Manager", "LogLocation", config_parse_location, 0, NULL },
678 { "Manager", "DumpCore", config_parse_bool, 0, &arg_dump_core },
679 { "Manager", "CrashChVT", /* legacy */ config_parse_crash_chvt, 0, NULL },
680 { "Manager", "CrashChangeVT", config_parse_crash_chvt, 0, NULL },
681 { "Manager", "CrashShell", config_parse_bool, 0, &arg_crash_shell },
682 { "Manager", "CrashReboot", config_parse_bool, 0, &arg_crash_reboot },
683 { "Manager", "ShowStatus", config_parse_show_status, 0, &arg_show_status },
684 { "Manager", "CPUAffinity", config_parse_cpu_affinity2, 0, NULL },
685 { "Manager", "JoinControllers", config_parse_warn_compat, DISABLED_CONFIGURATION, NULL },
686 { "Manager", "RuntimeWatchdogSec", config_parse_sec, 0, &arg_runtime_watchdog },
687 { "Manager", "ShutdownWatchdogSec", config_parse_sec, 0, &arg_shutdown_watchdog },
688 { "Manager", "WatchdogDevice", config_parse_path, 0, &arg_watchdog_device },
689 { "Manager", "CapabilityBoundingSet", config_parse_capability_set, 0, &arg_capability_bounding_set },
690 { "Manager", "NoNewPrivileges", config_parse_bool, 0, &arg_no_new_privs },
691 #if HAVE_SECCOMP
692 { "Manager", "SystemCallArchitectures", config_parse_syscall_archs, 0, &arg_syscall_archs },
693 #endif
694 { "Manager", "TimerSlackNSec", config_parse_nsec, 0, &arg_timer_slack_nsec },
695 { "Manager", "DefaultTimerAccuracySec", config_parse_sec, 0, &arg_default_timer_accuracy_usec },
696 { "Manager", "DefaultStandardOutput", config_parse_output_restricted,0, &arg_default_std_output },
697 { "Manager", "DefaultStandardError", config_parse_output_restricted,0, &arg_default_std_error },
698 { "Manager", "DefaultTimeoutStartSec", config_parse_sec, 0, &arg_default_timeout_start_usec },
699 { "Manager", "DefaultTimeoutStopSec", config_parse_sec, 0, &arg_default_timeout_stop_usec },
700 { "Manager", "DefaultRestartSec", config_parse_sec, 0, &arg_default_restart_usec },
701 { "Manager", "DefaultStartLimitInterval", config_parse_sec, 0, &arg_default_start_limit_interval }, /* obsolete alias */
702 { "Manager", "DefaultStartLimitIntervalSec",config_parse_sec, 0, &arg_default_start_limit_interval },
703 { "Manager", "DefaultStartLimitBurst", config_parse_unsigned, 0, &arg_default_start_limit_burst },
704 { "Manager", "DefaultEnvironment", config_parse_environ, 0, &arg_default_environment },
705 { "Manager", "DefaultLimitCPU", config_parse_rlimit, RLIMIT_CPU, arg_default_rlimit },
706 { "Manager", "DefaultLimitFSIZE", config_parse_rlimit, RLIMIT_FSIZE, arg_default_rlimit },
707 { "Manager", "DefaultLimitDATA", config_parse_rlimit, RLIMIT_DATA, arg_default_rlimit },
708 { "Manager", "DefaultLimitSTACK", config_parse_rlimit, RLIMIT_STACK, arg_default_rlimit },
709 { "Manager", "DefaultLimitCORE", config_parse_rlimit, RLIMIT_CORE, arg_default_rlimit },
710 { "Manager", "DefaultLimitRSS", config_parse_rlimit, RLIMIT_RSS, arg_default_rlimit },
711 { "Manager", "DefaultLimitNOFILE", config_parse_rlimit, RLIMIT_NOFILE, arg_default_rlimit },
712 { "Manager", "DefaultLimitAS", config_parse_rlimit, RLIMIT_AS, arg_default_rlimit },
713 { "Manager", "DefaultLimitNPROC", config_parse_rlimit, RLIMIT_NPROC, arg_default_rlimit },
714 { "Manager", "DefaultLimitMEMLOCK", config_parse_rlimit, RLIMIT_MEMLOCK, arg_default_rlimit },
715 { "Manager", "DefaultLimitLOCKS", config_parse_rlimit, RLIMIT_LOCKS, arg_default_rlimit },
716 { "Manager", "DefaultLimitSIGPENDING", config_parse_rlimit, RLIMIT_SIGPENDING, arg_default_rlimit },
717 { "Manager", "DefaultLimitMSGQUEUE", config_parse_rlimit, RLIMIT_MSGQUEUE, arg_default_rlimit },
718 { "Manager", "DefaultLimitNICE", config_parse_rlimit, RLIMIT_NICE, arg_default_rlimit },
719 { "Manager", "DefaultLimitRTPRIO", config_parse_rlimit, RLIMIT_RTPRIO, arg_default_rlimit },
720 { "Manager", "DefaultLimitRTTIME", config_parse_rlimit, RLIMIT_RTTIME, arg_default_rlimit },
721 { "Manager", "DefaultCPUAccounting", config_parse_tristate, 0, &arg_default_cpu_accounting },
722 { "Manager", "DefaultIOAccounting", config_parse_bool, 0, &arg_default_io_accounting },
723 { "Manager", "DefaultIPAccounting", config_parse_bool, 0, &arg_default_ip_accounting },
724 { "Manager", "DefaultBlockIOAccounting", config_parse_bool, 0, &arg_default_blockio_accounting },
725 { "Manager", "DefaultMemoryAccounting", config_parse_bool, 0, &arg_default_memory_accounting },
726 { "Manager", "DefaultTasksAccounting", config_parse_bool, 0, &arg_default_tasks_accounting },
727 { "Manager", "DefaultTasksMax", config_parse_tasks_max, 0, &arg_default_tasks_max },
728 { "Manager", "CtrlAltDelBurstAction", config_parse_emergency_action, 0, &arg_cad_burst_action },
729 { "Manager", "DefaultOOMPolicy", config_parse_oom_policy, 0, &arg_default_oom_policy },
730 {}
731 };
732
733 const char *fn, *conf_dirs_nulstr;
734
735 fn = arg_system ?
736 PKGSYSCONFDIR "/system.conf" :
737 PKGSYSCONFDIR "/user.conf";
738
739 conf_dirs_nulstr = arg_system ?
740 CONF_PATHS_NULSTR("systemd/system.conf.d") :
741 CONF_PATHS_NULSTR("systemd/user.conf.d");
742
743 (void) config_parse_many_nulstr(fn, conf_dirs_nulstr, "Manager\0", config_item_table_lookup, items, CONFIG_PARSE_WARN, NULL);
744
745 /* Traditionally "0" was used to turn off the default unit timeouts. Fix this up so that we used USEC_INFINITY
746 * like everywhere else. */
747 if (arg_default_timeout_start_usec <= 0)
748 arg_default_timeout_start_usec = USEC_INFINITY;
749 if (arg_default_timeout_stop_usec <= 0)
750 arg_default_timeout_stop_usec = USEC_INFINITY;
751
752 return 0;
753 }
754
755 static void set_manager_defaults(Manager *m) {
756
757 assert(m);
758
759 /* Propagates the various default unit property settings into the manager object, i.e. properties that do not
760 * affect the manager itself, but are just what newly allocated units will have set if they haven't set
761 * anything else. (Also see set_manager_settings() for the settings that affect the manager's own behaviour) */
762
763 m->default_timer_accuracy_usec = arg_default_timer_accuracy_usec;
764 m->default_std_output = arg_default_std_output;
765 m->default_std_error = arg_default_std_error;
766 m->default_timeout_start_usec = arg_default_timeout_start_usec;
767 m->default_timeout_stop_usec = arg_default_timeout_stop_usec;
768 m->default_restart_usec = arg_default_restart_usec;
769 m->default_start_limit_interval = arg_default_start_limit_interval;
770 m->default_start_limit_burst = arg_default_start_limit_burst;
771
772 /* On 4.15+ with unified hierarchy, CPU accounting is essentially free as it doesn't require the CPU
773 * controller to be enabled, so the default is to enable it unless we got told otherwise. */
774 if (arg_default_cpu_accounting >= 0)
775 m->default_cpu_accounting = arg_default_cpu_accounting;
776 else
777 m->default_cpu_accounting = cpu_accounting_is_cheap();
778
779 m->default_io_accounting = arg_default_io_accounting;
780 m->default_ip_accounting = arg_default_ip_accounting;
781 m->default_blockio_accounting = arg_default_blockio_accounting;
782 m->default_memory_accounting = arg_default_memory_accounting;
783 m->default_tasks_accounting = arg_default_tasks_accounting;
784 m->default_tasks_max = arg_default_tasks_max;
785 m->default_oom_policy = arg_default_oom_policy;
786
787 (void) manager_set_default_rlimits(m, arg_default_rlimit);
788
789 (void) manager_default_environment(m);
790 (void) manager_transient_environment_add(m, arg_default_environment);
791 }
792
793 static void set_manager_settings(Manager *m) {
794
795 assert(m);
796
797 /* Propagates the various manager settings into the manager object, i.e. properties that effect the manager
798 * itself (as opposed to just being inherited into newly allocated units, see set_manager_defaults() above). */
799
800 m->confirm_spawn = arg_confirm_spawn;
801 m->service_watchdogs = arg_service_watchdogs;
802 m->runtime_watchdog = arg_runtime_watchdog;
803 m->shutdown_watchdog = arg_shutdown_watchdog;
804 m->cad_burst_action = arg_cad_burst_action;
805
806 manager_set_show_status(m, arg_show_status);
807 }
808
809 static int parse_argv(int argc, char *argv[]) {
810 enum {
811 ARG_LOG_LEVEL = 0x100,
812 ARG_LOG_TARGET,
813 ARG_LOG_COLOR,
814 ARG_LOG_LOCATION,
815 ARG_UNIT,
816 ARG_SYSTEM,
817 ARG_USER,
818 ARG_TEST,
819 ARG_NO_PAGER,
820 ARG_VERSION,
821 ARG_DUMP_CONFIGURATION_ITEMS,
822 ARG_DUMP_BUS_PROPERTIES,
823 ARG_DUMP_CORE,
824 ARG_CRASH_CHVT,
825 ARG_CRASH_SHELL,
826 ARG_CRASH_REBOOT,
827 ARG_CONFIRM_SPAWN,
828 ARG_SHOW_STATUS,
829 ARG_DESERIALIZE,
830 ARG_SWITCHED_ROOT,
831 ARG_DEFAULT_STD_OUTPUT,
832 ARG_DEFAULT_STD_ERROR,
833 ARG_MACHINE_ID,
834 ARG_SERVICE_WATCHDOGS,
835 };
836
837 static const struct option options[] = {
838 { "log-level", required_argument, NULL, ARG_LOG_LEVEL },
839 { "log-target", required_argument, NULL, ARG_LOG_TARGET },
840 { "log-color", optional_argument, NULL, ARG_LOG_COLOR },
841 { "log-location", optional_argument, NULL, ARG_LOG_LOCATION },
842 { "unit", required_argument, NULL, ARG_UNIT },
843 { "system", no_argument, NULL, ARG_SYSTEM },
844 { "user", no_argument, NULL, ARG_USER },
845 { "test", no_argument, NULL, ARG_TEST },
846 { "no-pager", no_argument, NULL, ARG_NO_PAGER },
847 { "help", no_argument, NULL, 'h' },
848 { "version", no_argument, NULL, ARG_VERSION },
849 { "dump-configuration-items", no_argument, NULL, ARG_DUMP_CONFIGURATION_ITEMS },
850 { "dump-bus-properties", no_argument, NULL, ARG_DUMP_BUS_PROPERTIES },
851 { "dump-core", optional_argument, NULL, ARG_DUMP_CORE },
852 { "crash-chvt", required_argument, NULL, ARG_CRASH_CHVT },
853 { "crash-shell", optional_argument, NULL, ARG_CRASH_SHELL },
854 { "crash-reboot", optional_argument, NULL, ARG_CRASH_REBOOT },
855 { "confirm-spawn", optional_argument, NULL, ARG_CONFIRM_SPAWN },
856 { "show-status", optional_argument, NULL, ARG_SHOW_STATUS },
857 { "deserialize", required_argument, NULL, ARG_DESERIALIZE },
858 { "switched-root", no_argument, NULL, ARG_SWITCHED_ROOT },
859 { "default-standard-output", required_argument, NULL, ARG_DEFAULT_STD_OUTPUT, },
860 { "default-standard-error", required_argument, NULL, ARG_DEFAULT_STD_ERROR, },
861 { "machine-id", required_argument, NULL, ARG_MACHINE_ID },
862 { "service-watchdogs", required_argument, NULL, ARG_SERVICE_WATCHDOGS },
863 {}
864 };
865
866 int c, r;
867
868 assert(argc >= 1);
869 assert(argv);
870
871 if (getpid_cached() == 1)
872 opterr = 0;
873
874 while ((c = getopt_long(argc, argv, "hDbsz:", options, NULL)) >= 0)
875
876 switch (c) {
877
878 case ARG_LOG_LEVEL:
879 r = log_set_max_level_from_string(optarg);
880 if (r < 0)
881 return log_error_errno(r, "Failed to parse log level \"%s\": %m", optarg);
882
883 break;
884
885 case ARG_LOG_TARGET:
886 r = log_set_target_from_string(optarg);
887 if (r < 0)
888 return log_error_errno(r, "Failed to parse log target \"%s\": %m", optarg);
889
890 break;
891
892 case ARG_LOG_COLOR:
893
894 if (optarg) {
895 r = log_show_color_from_string(optarg);
896 if (r < 0)
897 return log_error_errno(r, "Failed to parse log color setting \"%s\": %m",
898 optarg);
899 } else
900 log_show_color(true);
901
902 break;
903
904 case ARG_LOG_LOCATION:
905 if (optarg) {
906 r = log_show_location_from_string(optarg);
907 if (r < 0)
908 return log_error_errno(r, "Failed to parse log location setting \"%s\": %m",
909 optarg);
910 } else
911 log_show_location(true);
912
913 break;
914
915 case ARG_DEFAULT_STD_OUTPUT:
916 r = exec_output_from_string(optarg);
917 if (r < 0)
918 return log_error_errno(r, "Failed to parse default standard output setting \"%s\": %m",
919 optarg);
920 arg_default_std_output = r;
921 break;
922
923 case ARG_DEFAULT_STD_ERROR:
924 r = exec_output_from_string(optarg);
925 if (r < 0)
926 return log_error_errno(r, "Failed to parse default standard error output setting \"%s\": %m",
927 optarg);
928 arg_default_std_error = r;
929 break;
930
931 case ARG_UNIT:
932 r = free_and_strdup(&arg_default_unit, optarg);
933 if (r < 0)
934 return log_error_errno(r, "Failed to set default unit \"%s\": %m", optarg);
935
936 break;
937
938 case ARG_SYSTEM:
939 arg_system = true;
940 break;
941
942 case ARG_USER:
943 arg_system = false;
944 break;
945
946 case ARG_TEST:
947 arg_action = ACTION_TEST;
948 break;
949
950 case ARG_NO_PAGER:
951 arg_pager_flags |= PAGER_DISABLE;
952 break;
953
954 case ARG_VERSION:
955 arg_action = ACTION_VERSION;
956 break;
957
958 case ARG_DUMP_CONFIGURATION_ITEMS:
959 arg_action = ACTION_DUMP_CONFIGURATION_ITEMS;
960 break;
961
962 case ARG_DUMP_BUS_PROPERTIES:
963 arg_action = ACTION_DUMP_BUS_PROPERTIES;
964 break;
965
966 case ARG_DUMP_CORE:
967 if (!optarg)
968 arg_dump_core = true;
969 else {
970 r = parse_boolean(optarg);
971 if (r < 0)
972 return log_error_errno(r, "Failed to parse dump core boolean: \"%s\": %m",
973 optarg);
974 arg_dump_core = r;
975 }
976 break;
977
978 case ARG_CRASH_CHVT:
979 r = parse_crash_chvt(optarg);
980 if (r < 0)
981 return log_error_errno(r, "Failed to parse crash virtual terminal index: \"%s\": %m",
982 optarg);
983 break;
984
985 case ARG_CRASH_SHELL:
986 if (!optarg)
987 arg_crash_shell = true;
988 else {
989 r = parse_boolean(optarg);
990 if (r < 0)
991 return log_error_errno(r, "Failed to parse crash shell boolean: \"%s\": %m",
992 optarg);
993 arg_crash_shell = r;
994 }
995 break;
996
997 case ARG_CRASH_REBOOT:
998 if (!optarg)
999 arg_crash_reboot = true;
1000 else {
1001 r = parse_boolean(optarg);
1002 if (r < 0)
1003 return log_error_errno(r, "Failed to parse crash shell boolean: \"%s\": %m",
1004 optarg);
1005 arg_crash_reboot = r;
1006 }
1007 break;
1008
1009 case ARG_CONFIRM_SPAWN:
1010 arg_confirm_spawn = mfree(arg_confirm_spawn);
1011
1012 r = parse_confirm_spawn(optarg, &arg_confirm_spawn);
1013 if (r < 0)
1014 return log_error_errno(r, "Failed to parse confirm spawn option: \"%s\": %m",
1015 optarg);
1016 break;
1017
1018 case ARG_SERVICE_WATCHDOGS:
1019 r = parse_boolean(optarg);
1020 if (r < 0)
1021 return log_error_errno(r, "Failed to parse service watchdogs boolean: \"%s\": %m",
1022 optarg);
1023 arg_service_watchdogs = r;
1024 break;
1025
1026 case ARG_SHOW_STATUS:
1027 if (optarg) {
1028 r = parse_show_status(optarg, &arg_show_status);
1029 if (r < 0)
1030 return log_error_errno(r, "Failed to parse show status boolean: \"%s\": %m",
1031 optarg);
1032 } else
1033 arg_show_status = SHOW_STATUS_YES;
1034 break;
1035
1036 case ARG_DESERIALIZE: {
1037 int fd;
1038 FILE *f;
1039
1040 r = safe_atoi(optarg, &fd);
1041 if (r < 0)
1042 log_error_errno(r, "Failed to parse deserialize option \"%s\": %m", optarg);
1043 if (fd < 0)
1044 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
1045 "Invalid deserialize fd: %d",
1046 fd);
1047
1048 (void) fd_cloexec(fd, true);
1049
1050 f = fdopen(fd, "r");
1051 if (!f)
1052 return log_error_errno(errno, "Failed to open serialization fd %d: %m", fd);
1053
1054 safe_fclose(arg_serialization);
1055 arg_serialization = f;
1056
1057 break;
1058 }
1059
1060 case ARG_SWITCHED_ROOT:
1061 arg_switched_root = true;
1062 break;
1063
1064 case ARG_MACHINE_ID:
1065 r = set_machine_id(optarg);
1066 if (r < 0)
1067 return log_error_errno(r, "MachineID '%s' is not valid: %m", optarg);
1068 break;
1069
1070 case 'h':
1071 arg_action = ACTION_HELP;
1072 break;
1073
1074 case 'D':
1075 log_set_max_level(LOG_DEBUG);
1076 break;
1077
1078 case 'b':
1079 case 's':
1080 case 'z':
1081 /* Just to eat away the sysvinit kernel
1082 * cmdline args without getopt() error
1083 * messages that we'll parse in
1084 * parse_proc_cmdline_word() or ignore. */
1085
1086 case '?':
1087 if (getpid_cached() != 1)
1088 return -EINVAL;
1089 else
1090 return 0;
1091
1092 default:
1093 assert_not_reached("Unhandled option code.");
1094 }
1095
1096 if (optind < argc && getpid_cached() != 1) {
1097 /* Hmm, when we aren't run as init system
1098 * let's complain about excess arguments */
1099
1100 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
1101 "Excess arguments.");
1102 }
1103
1104 return 0;
1105 }
1106
1107 static int help(void) {
1108 _cleanup_free_ char *link = NULL;
1109 int r;
1110
1111 r = terminal_urlify_man("systemd", "1", &link);
1112 if (r < 0)
1113 return log_oom();
1114
1115 printf("%s [OPTIONS...]\n\n"
1116 "Starts up and maintains the system or user services.\n\n"
1117 " -h --help Show this help\n"
1118 " --version Show version\n"
1119 " --test Determine startup sequence, dump it and exit\n"
1120 " --no-pager Do not pipe output into a pager\n"
1121 " --dump-configuration-items Dump understood unit configuration items\n"
1122 " --dump-bus-properties Dump exposed bus properties\n"
1123 " --unit=UNIT Set default unit\n"
1124 " --system Run a system instance, even if PID != 1\n"
1125 " --user Run a user instance\n"
1126 " --dump-core[=BOOL] Dump core on crash\n"
1127 " --crash-vt=NR Change to specified VT on crash\n"
1128 " --crash-reboot[=BOOL] Reboot on crash\n"
1129 " --crash-shell[=BOOL] Run shell on crash\n"
1130 " --confirm-spawn[=BOOL] Ask for confirmation when spawning processes\n"
1131 " --show-status[=BOOL] Show status updates on the console during bootup\n"
1132 " --log-target=TARGET Set log target (console, journal, kmsg, journal-or-kmsg, null)\n"
1133 " --log-level=LEVEL Set log level (debug, info, notice, warning, err, crit, alert, emerg)\n"
1134 " --log-color[=BOOL] Highlight important log messages\n"
1135 " --log-location[=BOOL] Include code location in log messages\n"
1136 " --default-standard-output= Set default standard output for services\n"
1137 " --default-standard-error= Set default standard error output for services\n"
1138 "\nSee the %s for details.\n"
1139 , program_invocation_short_name
1140 , link
1141 );
1142
1143 return 0;
1144 }
1145
1146 static int prepare_reexecute(
1147 Manager *m,
1148 FILE **ret_f,
1149 FDSet **ret_fds,
1150 bool switching_root) {
1151
1152 _cleanup_fdset_free_ FDSet *fds = NULL;
1153 _cleanup_fclose_ FILE *f = NULL;
1154 int r;
1155
1156 assert(m);
1157 assert(ret_f);
1158 assert(ret_fds);
1159
1160 r = manager_open_serialization(m, &f);
1161 if (r < 0)
1162 return log_error_errno(r, "Failed to create serialization file: %m");
1163
1164 /* Make sure nothing is really destructed when we shut down */
1165 m->n_reloading++;
1166 bus_manager_send_reloading(m, true);
1167
1168 fds = fdset_new();
1169 if (!fds)
1170 return log_oom();
1171
1172 r = manager_serialize(m, f, fds, switching_root);
1173 if (r < 0)
1174 return r;
1175
1176 if (fseeko(f, 0, SEEK_SET) == (off_t) -1)
1177 return log_error_errno(errno, "Failed to rewind serialization fd: %m");
1178
1179 r = fd_cloexec(fileno(f), false);
1180 if (r < 0)
1181 return log_error_errno(r, "Failed to disable O_CLOEXEC for serialization: %m");
1182
1183 r = fdset_cloexec(fds, false);
1184 if (r < 0)
1185 return log_error_errno(r, "Failed to disable O_CLOEXEC for serialization fds: %m");
1186
1187 *ret_f = TAKE_PTR(f);
1188 *ret_fds = TAKE_PTR(fds);
1189
1190 return 0;
1191 }
1192
1193 static void bump_file_max_and_nr_open(void) {
1194
1195 /* Let's bump fs.file-max and fs.nr_open to their respective maximums. On current kernels large numbers of file
1196 * descriptors are no longer a performance problem and their memory is properly tracked by memcg, thus counting
1197 * them and limiting them in another two layers of limits is unnecessary and just complicates things. This
1198 * function hence turns off 2 of the 4 levels of limits on file descriptors, and makes RLIMIT_NOLIMIT (soft +
1199 * hard) the only ones that really matter. */
1200
1201 #if BUMP_PROC_SYS_FS_FILE_MAX || BUMP_PROC_SYS_FS_NR_OPEN
1202 _cleanup_free_ char *t = NULL;
1203 int r;
1204 #endif
1205
1206 #if BUMP_PROC_SYS_FS_FILE_MAX
1207 /* I so wanted to use STRINGIFY(ULONG_MAX) here, but alas we can't as glibc/gcc define that as
1208 * "(0x7fffffffffffffffL * 2UL + 1UL)". Seriously. 😢 */
1209 if (asprintf(&t, "%lu\n", ULONG_MAX) < 0) {
1210 log_oom();
1211 return;
1212 }
1213
1214 r = sysctl_write("fs/file-max", t);
1215 if (r < 0)
1216 log_full_errno(IN_SET(r, -EROFS, -EPERM, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, "Failed to bump fs.file-max, ignoring: %m");
1217 #endif
1218
1219 #if BUMP_PROC_SYS_FS_FILE_MAX && BUMP_PROC_SYS_FS_NR_OPEN
1220 t = mfree(t);
1221 #endif
1222
1223 #if BUMP_PROC_SYS_FS_NR_OPEN
1224 int v = INT_MAX;
1225
1226 /* Arg! The kernel enforces maximum and minimum values on the fs.nr_open, but we don't really know what they
1227 * are. The expression by which the maximum is determined is dependent on the architecture, and is something we
1228 * don't really want to copy to userspace, as it is dependent on implementation details of the kernel. Since
1229 * the kernel doesn't expose the maximum value to us, we can only try and hope. Hence, let's start with
1230 * INT_MAX, and then keep halving the value until we find one that works. Ugly? Yes, absolutely, but kernel
1231 * APIs are kernel APIs, so what do can we do... 🤯 */
1232
1233 for (;;) {
1234 int k;
1235
1236 v &= ~(__SIZEOF_POINTER__ - 1); /* Round down to next multiple of the pointer size */
1237 if (v < 1024) {
1238 log_warning("Can't bump fs.nr_open, value too small.");
1239 break;
1240 }
1241
1242 k = read_nr_open();
1243 if (k < 0) {
1244 log_error_errno(k, "Failed to read fs.nr_open: %m");
1245 break;
1246 }
1247 if (k >= v) { /* Already larger */
1248 log_debug("Skipping bump, value is already larger.");
1249 break;
1250 }
1251
1252 if (asprintf(&t, "%i\n", v) < 0) {
1253 log_oom();
1254 return;
1255 }
1256
1257 r = sysctl_write("fs/nr_open", t);
1258 t = mfree(t);
1259 if (r == -EINVAL) {
1260 log_debug("Couldn't write fs.nr_open as %i, halving it.", v);
1261 v /= 2;
1262 continue;
1263 }
1264 if (r < 0) {
1265 log_full_errno(IN_SET(r, -EROFS, -EPERM, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, "Failed to bump fs.nr_open, ignoring: %m");
1266 break;
1267 }
1268
1269 log_debug("Successfully bumped fs.nr_open to %i", v);
1270 break;
1271 }
1272 #endif
1273 }
1274
1275 static int bump_rlimit_nofile(struct rlimit *saved_rlimit) {
1276 struct rlimit new_rlimit;
1277 int r, nr;
1278
1279 assert(saved_rlimit);
1280
1281 /* Save the original RLIMIT_NOFILE so that we can reset it later when transitioning from the initrd to the main
1282 * systemd or suchlike. */
1283 if (getrlimit(RLIMIT_NOFILE, saved_rlimit) < 0)
1284 return log_warning_errno(errno, "Reading RLIMIT_NOFILE failed, ignoring: %m");
1285
1286 /* Get the underlying absolute limit the kernel enforces */
1287 nr = read_nr_open();
1288
1289 /* Make sure forked processes get limits based on the original kernel setting */
1290 if (!arg_default_rlimit[RLIMIT_NOFILE]) {
1291 struct rlimit *rl;
1292
1293 rl = newdup(struct rlimit, saved_rlimit, 1);
1294 if (!rl)
1295 return log_oom();
1296
1297 /* Bump the hard limit for system services to a substantially higher value. The default hard limit
1298 * current kernels set is pretty low (4K), mostly for historical reasons. According to kernel
1299 * developers, the fd handling in recent kernels has been optimized substantially enough, so that we
1300 * can bump the limit now, without paying too high a price in memory or performance. Note however that
1301 * we only bump the hard limit, not the soft limit. That's because select() works the way it works, and
1302 * chokes on fds >= 1024. If we'd bump the soft limit globally, it might accidentally happen to
1303 * unexpecting programs that they get fds higher than what they can process using select(). By only
1304 * bumping the hard limit but leaving the low limit as it is we avoid this pitfall: programs that are
1305 * written by folks aware of the select() problem in mind (and thus use poll()/epoll instead of
1306 * select(), the way everybody should) can explicitly opt into high fds by bumping their soft limit
1307 * beyond 1024, to the hard limit we pass. */
1308 if (arg_system)
1309 rl->rlim_max = MIN((rlim_t) nr, MAX(rl->rlim_max, (rlim_t) HIGH_RLIMIT_NOFILE));
1310
1311 /* If for some reason we were invoked with a soft limit above 1024 (which should never
1312 * happen!, but who knows what we get passed in from pam_limit when invoked as --user
1313 * instance), then lower what we pass on to not confuse our children */
1314 rl->rlim_cur = MIN(rl->rlim_cur, (rlim_t) FD_SETSIZE);
1315
1316 arg_default_rlimit[RLIMIT_NOFILE] = rl;
1317 }
1318
1319 /* Calculate the new limits to use for us. Never lower from what we inherited. */
1320 new_rlimit = (struct rlimit) {
1321 .rlim_cur = MAX((rlim_t) nr, saved_rlimit->rlim_cur),
1322 .rlim_max = MAX((rlim_t) nr, saved_rlimit->rlim_max),
1323 };
1324
1325 /* Shortcut if nothing changes. */
1326 if (saved_rlimit->rlim_max >= new_rlimit.rlim_max &&
1327 saved_rlimit->rlim_cur >= new_rlimit.rlim_cur) {
1328 log_debug("RLIMIT_NOFILE is already as high or higher than we need it, not bumping.");
1329 return 0;
1330 }
1331
1332 /* Bump up the resource limit for ourselves substantially, all the way to the maximum the kernel allows, for
1333 * both hard and soft. */
1334 r = setrlimit_closest(RLIMIT_NOFILE, &new_rlimit);
1335 if (r < 0)
1336 return log_warning_errno(r, "Setting RLIMIT_NOFILE failed, ignoring: %m");
1337
1338 return 0;
1339 }
1340
1341 static int bump_rlimit_memlock(struct rlimit *saved_rlimit) {
1342 struct rlimit new_rlimit;
1343 int r;
1344
1345 assert(saved_rlimit);
1346
1347 /* BPF_MAP_TYPE_LPM_TRIE bpf maps are charged against RLIMIT_MEMLOCK, even if we have CAP_IPC_LOCK which should
1348 * normally disable such checks. We need them to implement IPAccessAllow= and IPAccessDeny=, hence let's bump
1349 * the value high enough for our user. */
1350
1351 if (getrlimit(RLIMIT_MEMLOCK, saved_rlimit) < 0)
1352 return log_warning_errno(errno, "Reading RLIMIT_MEMLOCK failed, ignoring: %m");
1353
1354 /* Pass the original value down to invoked processes */
1355 if (!arg_default_rlimit[RLIMIT_MEMLOCK]) {
1356 struct rlimit *rl;
1357
1358 rl = newdup(struct rlimit, saved_rlimit, 1);
1359 if (!rl)
1360 return log_oom();
1361
1362 arg_default_rlimit[RLIMIT_MEMLOCK] = rl;
1363 }
1364
1365 /* Using MAX() on resource limits only is safe if RLIM_INFINITY is > 0. POSIX declares that rlim_t
1366 * must be unsigned, hence this is a given, but let's make this clear here. */
1367 assert_cc(RLIM_INFINITY > 0);
1368
1369 new_rlimit = (struct rlimit) {
1370 .rlim_cur = MAX(HIGH_RLIMIT_MEMLOCK, saved_rlimit->rlim_cur),
1371 .rlim_max = MAX(HIGH_RLIMIT_MEMLOCK, saved_rlimit->rlim_max),
1372 };
1373
1374 if (saved_rlimit->rlim_max >= new_rlimit.rlim_cur &&
1375 saved_rlimit->rlim_cur >= new_rlimit.rlim_max) {
1376 log_debug("RLIMIT_MEMLOCK is already as high or higher than we need it, not bumping.");
1377 return 0;
1378 }
1379
1380 r = setrlimit_closest(RLIMIT_MEMLOCK, &new_rlimit);
1381 if (r < 0)
1382 return log_warning_errno(r, "Setting RLIMIT_MEMLOCK failed, ignoring: %m");
1383
1384 return 0;
1385 }
1386
1387 static void test_usr(void) {
1388
1389 /* Check that /usr is either on the same file system as / or mounted already. */
1390
1391 if (dir_is_empty("/usr") <= 0)
1392 return;
1393
1394 log_warning("/usr appears to be on its own filesystem and is not already mounted. This is not a supported setup. "
1395 "Some things will probably break (sometimes even silently) in mysterious ways. "
1396 "Consult http://freedesktop.org/wiki/Software/systemd/separate-usr-is-broken for more information.");
1397 }
1398
1399 static int enforce_syscall_archs(Set *archs) {
1400 #if HAVE_SECCOMP
1401 int r;
1402
1403 if (!is_seccomp_available())
1404 return 0;
1405
1406 r = seccomp_restrict_archs(arg_syscall_archs);
1407 if (r < 0)
1408 return log_error_errno(r, "Failed to enforce system call architecture restrication: %m");
1409 #endif
1410 return 0;
1411 }
1412
1413 static int status_welcome(void) {
1414 _cleanup_free_ char *pretty_name = NULL, *ansi_color = NULL;
1415 int r;
1416
1417 if (IN_SET(arg_show_status, SHOW_STATUS_NO, SHOW_STATUS_AUTO))
1418 return 0;
1419
1420 r = parse_os_release(NULL,
1421 "PRETTY_NAME", &pretty_name,
1422 "ANSI_COLOR", &ansi_color,
1423 NULL);
1424 if (r < 0)
1425 log_full_errno(r == -ENOENT ? LOG_DEBUG : LOG_WARNING, r,
1426 "Failed to read os-release file, ignoring: %m");
1427
1428 if (log_get_show_color())
1429 return status_printf(NULL, 0,
1430 "\nWelcome to \x1B[%sm%s\x1B[0m!\n",
1431 isempty(ansi_color) ? "1" : ansi_color,
1432 isempty(pretty_name) ? "Linux" : pretty_name);
1433 else
1434 return status_printf(NULL, 0,
1435 "\nWelcome to %s!\n",
1436 isempty(pretty_name) ? "Linux" : pretty_name);
1437 }
1438
1439 static int write_container_id(void) {
1440 const char *c;
1441 int r;
1442
1443 c = getenv("container");
1444 if (isempty(c))
1445 return 0;
1446
1447 RUN_WITH_UMASK(0022)
1448 r = write_string_file("/run/systemd/container", c, WRITE_STRING_FILE_CREATE);
1449 if (r < 0)
1450 return log_warning_errno(r, "Failed to write /run/systemd/container, ignoring: %m");
1451
1452 return 1;
1453 }
1454
1455 static int bump_unix_max_dgram_qlen(void) {
1456 _cleanup_free_ char *qlen = NULL;
1457 unsigned long v;
1458 int r;
1459
1460 /* Let's bump the net.unix.max_dgram_qlen sysctl. The kernel default of 16 is simply too low. We set the value
1461 * really really early during boot, so that it is actually applied to all our sockets, including the
1462 * $NOTIFY_SOCKET one. */
1463
1464 r = read_one_line_file("/proc/sys/net/unix/max_dgram_qlen", &qlen);
1465 if (r < 0)
1466 return log_full_errno(r == -ENOENT ? LOG_DEBUG : LOG_WARNING, r, "Failed to read AF_UNIX datagram queue length, ignoring: %m");
1467
1468 r = safe_atolu(qlen, &v);
1469 if (r < 0)
1470 return log_warning_errno(r, "Failed to parse AF_UNIX datagram queue length '%s', ignoring: %m", qlen);
1471
1472 if (v >= DEFAULT_UNIX_MAX_DGRAM_QLEN)
1473 return 0;
1474
1475 r = write_string_filef("/proc/sys/net/unix/max_dgram_qlen", WRITE_STRING_FILE_DISABLE_BUFFER, "%lu", DEFAULT_UNIX_MAX_DGRAM_QLEN);
1476 if (r < 0)
1477 return log_full_errno(IN_SET(r, -EROFS, -EPERM, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
1478 "Failed to bump AF_UNIX datagram queue length, ignoring: %m");
1479
1480 return 1;
1481 }
1482
1483 static int fixup_environment(void) {
1484 _cleanup_free_ char *term = NULL;
1485 const char *t;
1486 int r;
1487
1488 /* Only fix up the environment when we are started as PID 1 */
1489 if (getpid_cached() != 1)
1490 return 0;
1491
1492 /* We expect the environment to be set correctly if run inside a container. */
1493 if (detect_container() > 0)
1494 return 0;
1495
1496 /* When started as PID1, the kernel uses /dev/console for our stdios and uses TERM=linux whatever the backend
1497 * device used by the console. We try to make a better guess here since some consoles might not have support
1498 * for color mode for example.
1499 *
1500 * However if TERM was configured through the kernel command line then leave it alone. */
1501 r = proc_cmdline_get_key("TERM", 0, &term);
1502 if (r < 0)
1503 return r;
1504
1505 t = term ?: default_term_for_tty("/dev/console");
1506
1507 if (setenv("TERM", t, 1) < 0)
1508 return -errno;
1509
1510 return 0;
1511 }
1512
1513 static void redirect_telinit(int argc, char *argv[]) {
1514
1515 /* This is compatibility support for SysV, where calling init as a user is identical to telinit. */
1516
1517 #if HAVE_SYSV_COMPAT
1518 if (getpid_cached() == 1)
1519 return;
1520
1521 if (!strstr(program_invocation_short_name, "init"))
1522 return;
1523
1524 execv(SYSTEMCTL_BINARY_PATH, argv);
1525 log_error_errno(errno, "Failed to exec " SYSTEMCTL_BINARY_PATH ": %m");
1526 exit(EXIT_FAILURE);
1527 #endif
1528 }
1529
1530 static int become_shutdown(
1531 const char *shutdown_verb,
1532 int retval) {
1533
1534 char log_level[DECIMAL_STR_MAX(int) + 1],
1535 exit_code[DECIMAL_STR_MAX(uint8_t) + 1],
1536 timeout[DECIMAL_STR_MAX(usec_t) + 1];
1537
1538 const char* command_line[13] = {
1539 SYSTEMD_SHUTDOWN_BINARY_PATH,
1540 shutdown_verb,
1541 "--timeout", timeout,
1542 "--log-level", log_level,
1543 "--log-target",
1544 };
1545
1546 _cleanup_strv_free_ char **env_block = NULL;
1547 size_t pos = 7;
1548 int r;
1549
1550 assert(shutdown_verb);
1551 assert(!command_line[pos]);
1552 env_block = strv_copy(environ);
1553
1554 xsprintf(log_level, "%d", log_get_max_level());
1555 xsprintf(timeout, "%" PRI_USEC "us", arg_default_timeout_stop_usec);
1556
1557 switch (log_get_target()) {
1558
1559 case LOG_TARGET_KMSG:
1560 case LOG_TARGET_JOURNAL_OR_KMSG:
1561 case LOG_TARGET_SYSLOG_OR_KMSG:
1562 command_line[pos++] = "kmsg";
1563 break;
1564
1565 case LOG_TARGET_NULL:
1566 command_line[pos++] = "null";
1567 break;
1568
1569 case LOG_TARGET_CONSOLE:
1570 default:
1571 command_line[pos++] = "console";
1572 break;
1573 };
1574
1575 if (log_get_show_color())
1576 command_line[pos++] = "--log-color";
1577
1578 if (log_get_show_location())
1579 command_line[pos++] = "--log-location";
1580
1581 if (streq(shutdown_verb, "exit")) {
1582 command_line[pos++] = "--exit-code";
1583 command_line[pos++] = exit_code;
1584 xsprintf(exit_code, "%d", retval);
1585 }
1586
1587 assert(pos < ELEMENTSOF(command_line));
1588
1589 if (streq(shutdown_verb, "reboot") &&
1590 arg_shutdown_watchdog > 0 &&
1591 arg_shutdown_watchdog != USEC_INFINITY) {
1592
1593 char *e;
1594
1595 /* If we reboot let's set the shutdown
1596 * watchdog and tell the shutdown binary to
1597 * repeatedly ping it */
1598 r = watchdog_set_timeout(&arg_shutdown_watchdog);
1599 watchdog_close(r < 0);
1600
1601 /* Tell the binary how often to ping, ignore failure */
1602 if (asprintf(&e, "WATCHDOG_USEC="USEC_FMT, arg_shutdown_watchdog) > 0)
1603 (void) strv_consume(&env_block, e);
1604
1605 if (arg_watchdog_device &&
1606 asprintf(&e, "WATCHDOG_DEVICE=%s", arg_watchdog_device) > 0)
1607 (void) strv_consume(&env_block, e);
1608 } else
1609 watchdog_close(true);
1610
1611 /* Avoid the creation of new processes forked by the
1612 * kernel; at this point, we will not listen to the
1613 * signals anyway */
1614 if (detect_container() <= 0)
1615 (void) cg_uninstall_release_agent(SYSTEMD_CGROUP_CONTROLLER);
1616
1617 execve(SYSTEMD_SHUTDOWN_BINARY_PATH, (char **) command_line, env_block);
1618 return -errno;
1619 }
1620
1621 static void initialize_clock(void) {
1622 int r;
1623
1624 if (clock_is_localtime(NULL) > 0) {
1625 int min;
1626
1627 /*
1628 * The very first call of settimeofday() also does a time warp in the kernel.
1629 *
1630 * In the rtc-in-local time mode, we set the kernel's timezone, and rely on external tools to take care
1631 * of maintaining the RTC and do all adjustments. This matches the behavior of Windows, which leaves
1632 * the RTC alone if the registry tells that the RTC runs in UTC.
1633 */
1634 r = clock_set_timezone(&min);
1635 if (r < 0)
1636 log_error_errno(r, "Failed to apply local time delta, ignoring: %m");
1637 else
1638 log_info("RTC configured in localtime, applying delta of %i minutes to system time.", min);
1639
1640 } else if (!in_initrd()) {
1641 /*
1642 * Do a dummy very first call to seal the kernel's time warp magic.
1643 *
1644 * Do not call this from inside the initrd. The initrd might not carry /etc/adjtime with LOCAL, but the
1645 * real system could be set up that way. In such case, we need to delay the time-warp or the sealing
1646 * until we reach the real system.
1647 *
1648 * Do no set the kernel's timezone. The concept of local time cannot be supported reliably, the time
1649 * will jump or be incorrect at every daylight saving time change. All kernel local time concepts will
1650 * be treated as UTC that way.
1651 */
1652 (void) clock_reset_timewarp();
1653 }
1654
1655 r = clock_apply_epoch();
1656 if (r < 0)
1657 log_error_errno(r, "Current system time is before build time, but cannot correct: %m");
1658 else if (r > 0)
1659 log_info("System time before build time, advancing clock.");
1660 }
1661
1662 static void initialize_coredump(bool skip_setup) {
1663 #if ENABLE_COREDUMP
1664 if (getpid_cached() != 1)
1665 return;
1666
1667 /* Don't limit the core dump size, so that coredump handlers such as systemd-coredump (which honour the limit)
1668 * will process core dumps for system services by default. */
1669 if (setrlimit(RLIMIT_CORE, &RLIMIT_MAKE_CONST(RLIM_INFINITY)) < 0)
1670 log_warning_errno(errno, "Failed to set RLIMIT_CORE: %m");
1671
1672 /* But at the same time, turn off the core_pattern logic by default, so that no
1673 * coredumps are stored until the systemd-coredump tool is enabled via
1674 * sysctl. However it can be changed via the kernel command line later so core
1675 * 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", arg_early_core_pattern);
1693 }
1694
1695 static void do_reexecute(
1696 int argc,
1697 char *argv[],
1698 const struct rlimit *saved_rlimit_nofile,
1699 const struct rlimit *saved_rlimit_memlock,
1700 FDSet *fds,
1701 const char *switch_root_dir,
1702 const char *switch_root_init,
1703 const char **ret_error_message) {
1704
1705 unsigned i, j, args_size;
1706 const char **args;
1707 int r;
1708
1709 assert(saved_rlimit_nofile);
1710 assert(saved_rlimit_memlock);
1711 assert(ret_error_message);
1712
1713 /* Close and disarm the watchdog, so that the new instance can reinitialize it, but doesn't get rebooted while
1714 * we do that */
1715 watchdog_close(true);
1716
1717 /* Reset RLIMIT_NOFILE + RLIMIT_MEMLOCK back to the kernel defaults, so that the new systemd can pass
1718 * the kernel default to its child processes */
1719 if (saved_rlimit_nofile->rlim_cur != 0)
1720 (void) setrlimit(RLIMIT_NOFILE, saved_rlimit_nofile);
1721 if (saved_rlimit_memlock->rlim_cur != RLIM_INFINITY)
1722 (void) setrlimit(RLIMIT_MEMLOCK, saved_rlimit_memlock);
1723
1724 if (switch_root_dir) {
1725 /* Kill all remaining processes from the initrd, but don't wait for them, so that we can handle the
1726 * SIGCHLD for them after deserializing. */
1727 broadcast_signal(SIGTERM, false, true, arg_default_timeout_stop_usec);
1728
1729 /* And switch root with MS_MOVE, because we remove the old directory afterwards and detach it. */
1730 r = switch_root(switch_root_dir, "/mnt", true, MS_MOVE);
1731 if (r < 0)
1732 log_error_errno(r, "Failed to switch root, trying to continue: %m");
1733 }
1734
1735 args_size = MAX(6, argc+1);
1736 args = newa(const char*, args_size);
1737
1738 if (!switch_root_init) {
1739 char sfd[DECIMAL_STR_MAX(int) + 1];
1740
1741 /* First try to spawn ourselves with the right path, and with full serialization. We do this only if
1742 * the user didn't specify an explicit init to spawn. */
1743
1744 assert(arg_serialization);
1745 assert(fds);
1746
1747 xsprintf(sfd, "%i", fileno(arg_serialization));
1748
1749 i = 0;
1750 args[i++] = SYSTEMD_BINARY_PATH;
1751 if (switch_root_dir)
1752 args[i++] = "--switched-root";
1753 args[i++] = arg_system ? "--system" : "--user";
1754 args[i++] = "--deserialize";
1755 args[i++] = sfd;
1756 args[i++] = NULL;
1757
1758 assert(i <= args_size);
1759
1760 /*
1761 * We want valgrind to print its memory usage summary before reexecution. Valgrind won't do this is on
1762 * its own on exec(), but it will do it on exit(). Hence, to ensure we get a summary here, fork() off
1763 * a child, let it exit() cleanly, so that it prints the summary, and wait() for it in the parent,
1764 * before proceeding into the exec().
1765 */
1766 valgrind_summary_hack();
1767
1768 (void) execv(args[0], (char* const*) args);
1769 log_debug_errno(errno, "Failed to execute our own binary, trying fallback: %m");
1770 }
1771
1772 /* Try the fallback, if there is any, without any serialization. We pass the original argv[] and envp[]. (Well,
1773 * modulo the ordering changes due to getopt() in argv[], and some cleanups in envp[], but let's hope that
1774 * doesn't matter.) */
1775
1776 arg_serialization = safe_fclose(arg_serialization);
1777 fds = fdset_free(fds);
1778
1779 /* Reopen the console */
1780 (void) make_console_stdio();
1781
1782 for (j = 1, i = 1; j < (unsigned) argc; j++)
1783 args[i++] = argv[j];
1784 args[i++] = NULL;
1785 assert(i <= args_size);
1786
1787 /* Reenable any blocked signals, especially important if we switch from initial ramdisk to init=... */
1788 (void) reset_all_signal_handlers();
1789 (void) reset_signal_mask();
1790 (void) rlimit_nofile_safe();
1791
1792 if (switch_root_init) {
1793 args[0] = switch_root_init;
1794 (void) execv(args[0], (char* const*) args);
1795 log_warning_errno(errno, "Failed to execute configured init, trying fallback: %m");
1796 }
1797
1798 args[0] = "/sbin/init";
1799 (void) execv(args[0], (char* const*) args);
1800 r = -errno;
1801
1802 manager_status_printf(NULL, STATUS_TYPE_EMERGENCY,
1803 ANSI_HIGHLIGHT_RED " !! " ANSI_NORMAL,
1804 "Failed to execute /sbin/init");
1805
1806 if (r == -ENOENT) {
1807 log_warning("No /sbin/init, trying fallback");
1808
1809 args[0] = "/bin/sh";
1810 args[1] = NULL;
1811 (void) execv(args[0], (char* const*) args);
1812 log_error_errno(errno, "Failed to execute /bin/sh, giving up: %m");
1813 } else
1814 log_warning_errno(r, "Failed to execute /sbin/init, giving up: %m");
1815
1816 *ret_error_message = "Failed to execute fallback shell";
1817 }
1818
1819 static int invoke_main_loop(
1820 Manager *m,
1821 bool *ret_reexecute,
1822 int *ret_retval, /* Return parameters relevant for shutting down */
1823 const char **ret_shutdown_verb, /* … */
1824 FDSet **ret_fds, /* Return parameters for reexecuting */
1825 char **ret_switch_root_dir, /* … */
1826 char **ret_switch_root_init, /* … */
1827 const char **ret_error_message) {
1828
1829 int r;
1830
1831 assert(m);
1832 assert(ret_reexecute);
1833 assert(ret_retval);
1834 assert(ret_shutdown_verb);
1835 assert(ret_fds);
1836 assert(ret_switch_root_dir);
1837 assert(ret_switch_root_init);
1838 assert(ret_error_message);
1839
1840 for (;;) {
1841 r = manager_loop(m);
1842 if (r < 0) {
1843 *ret_error_message = "Failed to run main loop";
1844 return log_emergency_errno(r, "Failed to run main loop: %m");
1845 }
1846
1847 switch ((ManagerObjective) r) {
1848
1849 case MANAGER_RELOAD: {
1850 LogTarget saved_log_target;
1851 int saved_log_level;
1852
1853 log_info("Reloading.");
1854
1855 /* First, save any overridden log level/target, then parse the configuration file, which might
1856 * change the log level to new settings. */
1857
1858 saved_log_level = m->log_level_overridden ? log_get_max_level() : -1;
1859 saved_log_target = m->log_target_overridden ? log_get_target() : _LOG_TARGET_INVALID;
1860
1861 r = parse_config_file();
1862 if (r < 0)
1863 log_warning_errno(r, "Failed to parse config file, ignoring: %m");
1864
1865 set_manager_defaults(m);
1866
1867 if (saved_log_level >= 0)
1868 manager_override_log_level(m, saved_log_level);
1869 if (saved_log_target >= 0)
1870 manager_override_log_target(m, saved_log_target);
1871
1872 r = manager_reload(m);
1873 if (r < 0)
1874 /* Reloading failed before the point of no return. Let's continue running as if nothing happened. */
1875 m->objective = MANAGER_OK;
1876
1877 break;
1878 }
1879
1880 case MANAGER_REEXECUTE:
1881
1882 r = prepare_reexecute(m, &arg_serialization, ret_fds, false);
1883 if (r < 0) {
1884 *ret_error_message = "Failed to prepare for reexecution";
1885 return r;
1886 }
1887
1888 log_notice("Reexecuting.");
1889
1890 *ret_reexecute = true;
1891 *ret_retval = EXIT_SUCCESS;
1892 *ret_shutdown_verb = NULL;
1893 *ret_switch_root_dir = *ret_switch_root_init = NULL;
1894
1895 return 0;
1896
1897 case MANAGER_SWITCH_ROOT:
1898 if (!m->switch_root_init) {
1899 r = prepare_reexecute(m, &arg_serialization, ret_fds, true);
1900 if (r < 0) {
1901 *ret_error_message = "Failed to prepare for reexecution";
1902 return r;
1903 }
1904 } else
1905 *ret_fds = NULL;
1906
1907 log_notice("Switching root.");
1908
1909 *ret_reexecute = true;
1910 *ret_retval = EXIT_SUCCESS;
1911 *ret_shutdown_verb = NULL;
1912
1913 /* Steal the switch root parameters */
1914 *ret_switch_root_dir = TAKE_PTR(m->switch_root);
1915 *ret_switch_root_init = TAKE_PTR(m->switch_root_init);
1916
1917 return 0;
1918
1919 case MANAGER_EXIT:
1920
1921 if (MANAGER_IS_USER(m)) {
1922 log_debug("Exit.");
1923
1924 *ret_reexecute = false;
1925 *ret_retval = m->return_value;
1926 *ret_shutdown_verb = NULL;
1927 *ret_fds = NULL;
1928 *ret_switch_root_dir = *ret_switch_root_init = NULL;
1929
1930 return 0;
1931 }
1932
1933 _fallthrough_;
1934 case MANAGER_REBOOT:
1935 case MANAGER_POWEROFF:
1936 case MANAGER_HALT:
1937 case MANAGER_KEXEC: {
1938 static const char * const table[_MANAGER_OBJECTIVE_MAX] = {
1939 [MANAGER_EXIT] = "exit",
1940 [MANAGER_REBOOT] = "reboot",
1941 [MANAGER_POWEROFF] = "poweroff",
1942 [MANAGER_HALT] = "halt",
1943 [MANAGER_KEXEC] = "kexec",
1944 };
1945
1946 log_notice("Shutting down.");
1947
1948 *ret_reexecute = false;
1949 *ret_retval = m->return_value;
1950 assert_se(*ret_shutdown_verb = table[m->objective]);
1951 *ret_fds = NULL;
1952 *ret_switch_root_dir = *ret_switch_root_init = NULL;
1953
1954 return 0;
1955 }
1956
1957 default:
1958 assert_not_reached("Unknown or unexpected manager objective.");
1959 }
1960 }
1961 }
1962
1963 static void log_execution_mode(bool *ret_first_boot) {
1964 assert(ret_first_boot);
1965
1966 if (arg_system) {
1967 int v;
1968
1969 log_info("systemd " GIT_VERSION " running in %ssystem mode. (" SYSTEMD_FEATURES ")",
1970 arg_action == ACTION_TEST ? "test " : "" );
1971
1972 v = detect_virtualization();
1973 if (v > 0)
1974 log_info("Detected virtualization %s.", virtualization_to_string(v));
1975
1976 log_info("Detected architecture %s.", architecture_to_string(uname_architecture()));
1977
1978 if (in_initrd()) {
1979 *ret_first_boot = false;
1980 log_info("Running in initial RAM disk.");
1981 } else {
1982 /* Let's check whether we are in first boot, i.e. whether /etc is still unpopulated. We use
1983 * /etc/machine-id as flag file, for this: if it exists we assume /etc is populated, if it
1984 * doesn't it's unpopulated. This allows container managers and installers to provision a
1985 * couple of files already. If the container manager wants to provision the machine ID itself
1986 * it should pass $container_uuid to PID 1. */
1987
1988 *ret_first_boot = access("/etc/machine-id", F_OK) < 0;
1989 if (*ret_first_boot)
1990 log_info("Running with unpopulated /etc.");
1991 }
1992 } else {
1993 if (DEBUG_LOGGING) {
1994 _cleanup_free_ char *t;
1995
1996 t = uid_to_name(getuid());
1997 log_debug("systemd " GIT_VERSION " running in %suser mode for user " UID_FMT "/%s. (" SYSTEMD_FEATURES ")",
1998 arg_action == ACTION_TEST ? " test" : "", getuid(), strna(t));
1999 }
2000
2001 *ret_first_boot = false;
2002 }
2003 }
2004
2005 static int initialize_runtime(
2006 bool skip_setup,
2007 struct rlimit *saved_rlimit_nofile,
2008 struct rlimit *saved_rlimit_memlock,
2009 const char **ret_error_message) {
2010
2011 int r;
2012
2013 assert(ret_error_message);
2014
2015 /* Sets up various runtime parameters. Many of these initializations are conditionalized:
2016 *
2017 * - Some only apply to --system instances
2018 * - Some only apply to --user instances
2019 * - Some only apply when we first start up, but not when we reexecute
2020 */
2021
2022 if (arg_action != ACTION_RUN)
2023 return 0;
2024
2025 if (arg_system) {
2026 /* Make sure we leave a core dump without panicing the kernel. */
2027 install_crash_handler();
2028
2029 if (!skip_setup) {
2030 r = mount_cgroup_controllers();
2031 if (r < 0) {
2032 *ret_error_message = "Failed to mount cgroup hierarchies";
2033 return r;
2034 }
2035
2036 status_welcome();
2037 hostname_setup();
2038 machine_id_setup(NULL, arg_machine_id, NULL);
2039 loopback_setup();
2040 bump_unix_max_dgram_qlen();
2041 bump_file_max_and_nr_open();
2042 test_usr();
2043 write_container_id();
2044 }
2045
2046 if (arg_watchdog_device) {
2047 r = watchdog_set_device(arg_watchdog_device);
2048 if (r < 0)
2049 log_warning_errno(r, "Failed to set watchdog device to %s, ignoring: %m", arg_watchdog_device);
2050 }
2051
2052 if (arg_runtime_watchdog > 0 && arg_runtime_watchdog != USEC_INFINITY)
2053 watchdog_set_timeout(&arg_runtime_watchdog);
2054 }
2055
2056 if (arg_timer_slack_nsec != NSEC_INFINITY)
2057 if (prctl(PR_SET_TIMERSLACK, arg_timer_slack_nsec) < 0)
2058 log_warning_errno(errno, "Failed to adjust timer slack, ignoring: %m");
2059
2060 if (arg_system && !cap_test_all(arg_capability_bounding_set)) {
2061 r = capability_bounding_set_drop_usermode(arg_capability_bounding_set);
2062 if (r < 0) {
2063 *ret_error_message = "Failed to drop capability bounding set of usermode helpers";
2064 return log_emergency_errno(r, "Failed to drop capability bounding set of usermode helpers: %m");
2065 }
2066
2067 r = capability_bounding_set_drop(arg_capability_bounding_set, true);
2068 if (r < 0) {
2069 *ret_error_message = "Failed to drop capability bounding set";
2070 return log_emergency_errno(r, "Failed to drop capability bounding set: %m");
2071 }
2072 }
2073
2074 if (arg_system && arg_no_new_privs) {
2075 if (prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0) < 0) {
2076 *ret_error_message = "Failed to disable new privileges";
2077 return log_emergency_errno(errno, "Failed to disable new privileges: %m");
2078 }
2079 }
2080
2081 if (arg_syscall_archs) {
2082 r = enforce_syscall_archs(arg_syscall_archs);
2083 if (r < 0) {
2084 *ret_error_message = "Failed to set syscall architectures";
2085 return r;
2086 }
2087 }
2088
2089 if (!arg_system)
2090 /* Become reaper of our children */
2091 if (prctl(PR_SET_CHILD_SUBREAPER, 1) < 0)
2092 log_warning_errno(errno, "Failed to make us a subreaper: %m");
2093
2094 /* Bump up RLIMIT_NOFILE for systemd itself */
2095 (void) bump_rlimit_nofile(saved_rlimit_nofile);
2096 (void) bump_rlimit_memlock(saved_rlimit_memlock);
2097
2098 return 0;
2099 }
2100
2101 static int do_queue_default_job(
2102 Manager *m,
2103 const char **ret_error_message) {
2104
2105 _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL;
2106 Job *default_unit_job;
2107 Unit *target = NULL;
2108 int r;
2109
2110 log_debug("Activating default unit: %s", arg_default_unit);
2111
2112 r = manager_load_startable_unit_or_warn(m, arg_default_unit, NULL, &target);
2113 if (r < 0) {
2114 log_info("Falling back to rescue target: " SPECIAL_RESCUE_TARGET);
2115
2116 r = manager_load_startable_unit_or_warn(m, SPECIAL_RESCUE_TARGET, NULL, &target);
2117 if (r < 0) {
2118 *ret_error_message = r == -ERFKILL ? "Rescue target masked"
2119 : "Failed to load rescue target";
2120 return r;
2121 }
2122 }
2123
2124 assert(target->load_state == UNIT_LOADED);
2125
2126 r = manager_add_job(m, JOB_START, target, JOB_ISOLATE, NULL, &error, &default_unit_job);
2127 if (r == -EPERM) {
2128 log_debug_errno(r, "Default target could not be isolated, starting instead: %s", bus_error_message(&error, r));
2129
2130 sd_bus_error_free(&error);
2131
2132 r = manager_add_job(m, JOB_START, target, JOB_REPLACE, NULL, &error, &default_unit_job);
2133 if (r < 0) {
2134 *ret_error_message = "Failed to start default target";
2135 return log_emergency_errno(r, "Failed to start default target: %s", bus_error_message(&error, r));
2136 }
2137
2138 } else if (r < 0) {
2139 *ret_error_message = "Failed to isolate default target";
2140 return log_emergency_errno(r, "Failed to isolate default target: %s", bus_error_message(&error, r));
2141 }
2142
2143 m->default_unit_job_id = default_unit_job->id;
2144
2145 return 0;
2146 }
2147
2148 static void free_arguments(void) {
2149
2150 /* Frees all arg_* variables, with the exception of arg_serialization */
2151 rlimit_free_all(arg_default_rlimit);
2152
2153 arg_default_unit = mfree(arg_default_unit);
2154 arg_confirm_spawn = mfree(arg_confirm_spawn);
2155 arg_default_environment = strv_free(arg_default_environment);
2156 arg_syscall_archs = set_free(arg_syscall_archs);
2157 }
2158
2159 static int load_configuration(int argc, char **argv, const char **ret_error_message) {
2160 int r;
2161
2162 assert(ret_error_message);
2163
2164 arg_default_tasks_max = system_tasks_max_scale(DEFAULT_TASKS_MAX_PERCENTAGE, 100U);
2165
2166 r = parse_config_file();
2167 if (r < 0) {
2168 *ret_error_message = "Failed to parse config file";
2169 return r;
2170 }
2171
2172 if (arg_system) {
2173 r = proc_cmdline_parse(parse_proc_cmdline_item, NULL, 0);
2174 if (r < 0)
2175 log_warning_errno(r, "Failed to parse kernel command line, ignoring: %m");
2176 }
2177
2178 /* Note that this also parses bits from the kernel command line, including "debug". */
2179 log_parse_environment();
2180
2181 r = parse_argv(argc, argv);
2182 if (r < 0) {
2183 *ret_error_message = "Failed to parse commandline arguments";
2184 return r;
2185 }
2186
2187 /* Initialize default unit */
2188 if (!arg_default_unit) {
2189 arg_default_unit = strdup(SPECIAL_DEFAULT_TARGET);
2190 if (!arg_default_unit) {
2191 *ret_error_message = "Failed to set default unit";
2192 return log_oom();
2193 }
2194 }
2195
2196 /* Initialize the show status setting if it hasn't been set explicitly yet */
2197 if (arg_show_status == _SHOW_STATUS_INVALID)
2198 arg_show_status = SHOW_STATUS_YES;
2199
2200 return 0;
2201 }
2202
2203 static int safety_checks(void) {
2204
2205 if (getpid_cached() == 1 &&
2206 arg_action != ACTION_RUN)
2207 return log_error_errno(SYNTHETIC_ERRNO(EPERM),
2208 "Unsupported execution mode while PID 1.");
2209
2210 if (getpid_cached() == 1 &&
2211 !arg_system)
2212 return log_error_errno(SYNTHETIC_ERRNO(EPERM),
2213 "Can't run --user mode as PID 1.");
2214
2215 if (arg_action == ACTION_RUN &&
2216 arg_system &&
2217 getpid_cached() != 1)
2218 return log_error_errno(SYNTHETIC_ERRNO(EPERM),
2219 "Can't run system mode unless PID 1.");
2220
2221 if (arg_action == ACTION_TEST &&
2222 geteuid() == 0)
2223 return log_error_errno(SYNTHETIC_ERRNO(EPERM),
2224 "Don't run test mode as root.");
2225
2226 if (!arg_system &&
2227 arg_action == ACTION_RUN &&
2228 sd_booted() <= 0)
2229 return log_error_errno(SYNTHETIC_ERRNO(EOPNOTSUPP),
2230 "Trying to run as user instance, but the system has not been booted with systemd.");
2231
2232 if (!arg_system &&
2233 arg_action == ACTION_RUN &&
2234 !getenv("XDG_RUNTIME_DIR"))
2235 return log_error_errno(SYNTHETIC_ERRNO(EUNATCH),
2236 "Trying to run as user instance, but $XDG_RUNTIME_DIR is not set.");
2237
2238 if (arg_system &&
2239 arg_action == ACTION_RUN &&
2240 running_in_chroot() > 0)
2241 return log_error_errno(SYNTHETIC_ERRNO(EOPNOTSUPP),
2242 "Cannot be run in a chroot() environment.");
2243
2244 return 0;
2245 }
2246
2247 static int initialize_security(
2248 bool *loaded_policy,
2249 dual_timestamp *security_start_timestamp,
2250 dual_timestamp *security_finish_timestamp,
2251 const char **ret_error_message) {
2252
2253 int r;
2254
2255 assert(loaded_policy);
2256 assert(security_start_timestamp);
2257 assert(security_finish_timestamp);
2258 assert(ret_error_message);
2259
2260 dual_timestamp_get(security_start_timestamp);
2261
2262 r = mac_selinux_setup(loaded_policy);
2263 if (r < 0) {
2264 *ret_error_message = "Failed to load SELinux policy";
2265 return r;
2266 }
2267
2268 r = mac_smack_setup(loaded_policy);
2269 if (r < 0) {
2270 *ret_error_message = "Failed to load SMACK policy";
2271 return r;
2272 }
2273
2274 r = ima_setup();
2275 if (r < 0) {
2276 *ret_error_message = "Failed to load IMA policy";
2277 return r;
2278 }
2279
2280 dual_timestamp_get(security_finish_timestamp);
2281 return 0;
2282 }
2283
2284 static void test_summary(Manager *m) {
2285 assert(m);
2286
2287 printf("-> By units:\n");
2288 manager_dump_units(m, stdout, "\t");
2289
2290 printf("-> By jobs:\n");
2291 manager_dump_jobs(m, stdout, "\t");
2292 }
2293
2294 static int collect_fds(FDSet **ret_fds, const char **ret_error_message) {
2295 int r;
2296
2297 assert(ret_fds);
2298 assert(ret_error_message);
2299
2300 r = fdset_new_fill(ret_fds);
2301 if (r < 0) {
2302 *ret_error_message = "Failed to allocate fd set";
2303 return log_emergency_errno(r, "Failed to allocate fd set: %m");
2304 }
2305
2306 fdset_cloexec(*ret_fds, true);
2307
2308 if (arg_serialization)
2309 assert_se(fdset_remove(*ret_fds, fileno(arg_serialization)) >= 0);
2310
2311 return 0;
2312 }
2313
2314 static void setup_console_terminal(bool skip_setup) {
2315
2316 if (!arg_system)
2317 return;
2318
2319 /* Become a session leader if we aren't one yet. */
2320 (void) setsid();
2321
2322 /* If we are init, we connect stdin/stdout/stderr to /dev/null and make sure we don't have a controlling
2323 * tty. */
2324 (void) release_terminal();
2325
2326 /* Reset the console, but only if this is really init and we are freshly booted */
2327 if (getpid_cached() == 1 && !skip_setup)
2328 (void) console_setup();
2329 }
2330
2331 static bool early_skip_setup_check(int argc, char *argv[]) {
2332 bool found_deserialize = false;
2333 int i;
2334
2335 /* Determine if this is a reexecution or normal bootup. We do the full command line parsing much later, so
2336 * let's just have a quick peek here. Note that if we have switched root, do all the special setup things
2337 * anyway, even if in that case we also do deserialization. */
2338
2339 for (i = 1; i < argc; i++) {
2340 if (streq(argv[i], "--switched-root"))
2341 return false; /* If we switched root, don't skip the setup. */
2342 else if (streq(argv[i], "--deserialize"))
2343 found_deserialize = true;
2344 }
2345
2346 return found_deserialize; /* When we are deserializing, then we are reexecuting, hence avoid the extensive setup */
2347 }
2348
2349 int main(int argc, char *argv[]) {
2350
2351 dual_timestamp initrd_timestamp = DUAL_TIMESTAMP_NULL, userspace_timestamp = DUAL_TIMESTAMP_NULL, kernel_timestamp = DUAL_TIMESTAMP_NULL,
2352 security_start_timestamp = DUAL_TIMESTAMP_NULL, security_finish_timestamp = DUAL_TIMESTAMP_NULL;
2353 struct rlimit saved_rlimit_nofile = RLIMIT_MAKE_CONST(0),
2354 saved_rlimit_memlock = RLIMIT_MAKE_CONST(RLIM_INFINITY); /* The original rlimits we passed
2355 * in. Note we use different values
2356 * for the two that indicate whether
2357 * these fields are initialized! */
2358 bool skip_setup, loaded_policy = false, queue_default_job = false, first_boot = false, reexecute = false;
2359 char *switch_root_dir = NULL, *switch_root_init = NULL;
2360 usec_t before_startup, after_startup;
2361 static char systemd[] = "systemd";
2362 char timespan[FORMAT_TIMESPAN_MAX];
2363 const char *shutdown_verb = NULL, *error_message = NULL;
2364 int r, retval = EXIT_FAILURE;
2365 Manager *m = NULL;
2366 FDSet *fds = NULL;
2367
2368 /* SysV compatibility: redirect init → telinit */
2369 redirect_telinit(argc, argv);
2370
2371 /* Take timestamps early on */
2372 dual_timestamp_from_monotonic(&kernel_timestamp, 0);
2373 dual_timestamp_get(&userspace_timestamp);
2374
2375 /* Figure out whether we need to do initialize the system, or if we already did that because we are
2376 * reexecuting */
2377 skip_setup = early_skip_setup_check(argc, argv);
2378
2379 /* If we get started via the /sbin/init symlink then we are called 'init'. After a subsequent reexecution we
2380 * are then called 'systemd'. That is confusing, hence let's call us systemd right-away. */
2381 program_invocation_short_name = systemd;
2382 (void) prctl(PR_SET_NAME, systemd);
2383
2384 /* Save the original command line */
2385 save_argc_argv(argc, argv);
2386
2387 /* Make sure that if the user says "syslog" we actually log to the journal. */
2388 log_set_upgrade_syslog_to_journal(true);
2389
2390 if (getpid_cached() == 1) {
2391 /* When we run as PID 1 force system mode */
2392 arg_system = true;
2393
2394 /* Disable the umask logic */
2395 umask(0);
2396
2397 /* Make sure that at least initially we do not ever log to journald/syslogd, because it might not be
2398 * activated yet (even though the log socket for it exists). */
2399 log_set_prohibit_ipc(true);
2400
2401 /* Always reopen /dev/console when running as PID 1 or one of its pre-execve() children. This is
2402 * important so that we never end up logging to any foreign stderr, for example if we have to log in a
2403 * child process right before execve()'ing the actual binary, at a point in time where socket
2404 * activation stderr/stdout area already set up. */
2405 log_set_always_reopen_console(true);
2406
2407 if (detect_container() <= 0) {
2408
2409 /* Running outside of a container as PID 1 */
2410 log_set_target(LOG_TARGET_KMSG);
2411 log_open();
2412
2413 if (in_initrd())
2414 initrd_timestamp = userspace_timestamp;
2415
2416 if (!skip_setup) {
2417 r = mount_setup_early();
2418 if (r < 0) {
2419 error_message = "Failed to mount early API filesystems";
2420 goto finish;
2421 }
2422
2423 r = initialize_security(
2424 &loaded_policy,
2425 &security_start_timestamp,
2426 &security_finish_timestamp,
2427 &error_message);
2428 if (r < 0)
2429 goto finish;
2430 }
2431
2432 if (mac_selinux_init() < 0) {
2433 error_message = "Failed to initialize SELinux policy";
2434 goto finish;
2435 }
2436
2437 if (!skip_setup)
2438 initialize_clock();
2439
2440 /* Set the default for later on, but don't actually open the logs like this for now. Note that
2441 * if we are transitioning from the initrd there might still be journal fd open, and we
2442 * shouldn't attempt opening that before we parsed /proc/cmdline which might redirect output
2443 * elsewhere. */
2444 log_set_target(LOG_TARGET_JOURNAL_OR_KMSG);
2445
2446 } else {
2447 /* Running inside a container, as PID 1 */
2448 log_set_target(LOG_TARGET_CONSOLE);
2449 log_open();
2450
2451 /* For later on, see above... */
2452 log_set_target(LOG_TARGET_JOURNAL);
2453
2454 /* clear the kernel timestamp,
2455 * because we are in a container */
2456 kernel_timestamp = DUAL_TIMESTAMP_NULL;
2457 }
2458
2459 initialize_coredump(skip_setup);
2460
2461 r = fixup_environment();
2462 if (r < 0) {
2463 log_emergency_errno(r, "Failed to fix up PID 1 environment: %m");
2464 error_message = "Failed to fix up PID1 environment";
2465 goto finish;
2466 }
2467
2468 } else {
2469 /* Running as user instance */
2470 arg_system = false;
2471 log_set_target(LOG_TARGET_AUTO);
2472 log_open();
2473
2474 /* clear the kernel timestamp,
2475 * because we are not PID 1 */
2476 kernel_timestamp = DUAL_TIMESTAMP_NULL;
2477 }
2478
2479 if (arg_system) {
2480 /* Try to figure out if we can use colors with the console. No need to do that for user instances since
2481 * they never log into the console. */
2482 log_show_color(colors_enabled());
2483
2484 r = make_null_stdio();
2485 if (r < 0)
2486 log_warning_errno(r, "Failed to redirect standard streams to /dev/null, ignoring: %m");
2487 }
2488
2489 /* Mount /proc, /sys and friends, so that /proc/cmdline and
2490 * /proc/$PID/fd is available. */
2491 if (getpid_cached() == 1) {
2492
2493 /* Load the kernel modules early. */
2494 if (!skip_setup)
2495 kmod_setup();
2496
2497 r = mount_setup(loaded_policy);
2498 if (r < 0) {
2499 error_message = "Failed to mount API filesystems";
2500 goto finish;
2501 }
2502 }
2503
2504 /* Reset all signal handlers. */
2505 (void) reset_all_signal_handlers();
2506 (void) ignore_signals(SIGNALS_IGNORE, -1);
2507
2508 r = load_configuration(argc, argv, &error_message);
2509 if (r < 0)
2510 goto finish;
2511
2512 r = safety_checks();
2513 if (r < 0)
2514 goto finish;
2515
2516 if (IN_SET(arg_action, ACTION_TEST, ACTION_HELP, ACTION_DUMP_CONFIGURATION_ITEMS, ACTION_DUMP_BUS_PROPERTIES))
2517 (void) pager_open(arg_pager_flags);
2518
2519 if (arg_action != ACTION_RUN)
2520 skip_setup = true;
2521
2522 if (arg_action == ACTION_HELP) {
2523 retval = help() < 0 ? EXIT_FAILURE : EXIT_SUCCESS;
2524 goto finish;
2525 } else if (arg_action == ACTION_VERSION) {
2526 retval = version();
2527 goto finish;
2528 } else if (arg_action == ACTION_DUMP_CONFIGURATION_ITEMS) {
2529 unit_dump_config_items(stdout);
2530 retval = EXIT_SUCCESS;
2531 goto finish;
2532 } else if (arg_action == ACTION_DUMP_BUS_PROPERTIES) {
2533 dump_bus_properties(stdout);
2534 retval = EXIT_SUCCESS;
2535 goto finish;
2536 }
2537
2538 assert_se(IN_SET(arg_action, ACTION_RUN, ACTION_TEST));
2539
2540 /* Move out of the way, so that we won't block unmounts */
2541 assert_se(chdir("/") == 0);
2542
2543 if (arg_action == ACTION_RUN) {
2544
2545 /* A core pattern might have been specified via the cmdline. */
2546 initialize_core_pattern(skip_setup);
2547
2548 /* Close logging fds, in order not to confuse collecting passed fds and terminal logic below */
2549 log_close();
2550
2551 /* Remember open file descriptors for later deserialization */
2552 r = collect_fds(&fds, &error_message);
2553 if (r < 0)
2554 goto finish;
2555
2556 /* Give up any control of the console, but make sure its initialized. */
2557 setup_console_terminal(skip_setup);
2558
2559 /* Open the logging devices, if possible and necessary */
2560 log_open();
2561 }
2562
2563 log_execution_mode(&first_boot);
2564
2565 r = initialize_runtime(skip_setup,
2566 &saved_rlimit_nofile,
2567 &saved_rlimit_memlock,
2568 &error_message);
2569 if (r < 0)
2570 goto finish;
2571
2572 r = manager_new(arg_system ? UNIT_FILE_SYSTEM : UNIT_FILE_USER,
2573 arg_action == ACTION_TEST ? MANAGER_TEST_FULL : 0,
2574 &m);
2575 if (r < 0) {
2576 log_emergency_errno(r, "Failed to allocate manager object: %m");
2577 error_message = "Failed to allocate manager object";
2578 goto finish;
2579 }
2580
2581 m->timestamps[MANAGER_TIMESTAMP_KERNEL] = kernel_timestamp;
2582 m->timestamps[MANAGER_TIMESTAMP_INITRD] = initrd_timestamp;
2583 m->timestamps[MANAGER_TIMESTAMP_USERSPACE] = userspace_timestamp;
2584 m->timestamps[manager_timestamp_initrd_mangle(MANAGER_TIMESTAMP_SECURITY_START)] = security_start_timestamp;
2585 m->timestamps[manager_timestamp_initrd_mangle(MANAGER_TIMESTAMP_SECURITY_FINISH)] = security_finish_timestamp;
2586
2587 set_manager_defaults(m);
2588 set_manager_settings(m);
2589 manager_set_first_boot(m, first_boot);
2590
2591 /* Remember whether we should queue the default job */
2592 queue_default_job = !arg_serialization || arg_switched_root;
2593
2594 before_startup = now(CLOCK_MONOTONIC);
2595
2596 r = manager_startup(m, arg_serialization, fds);
2597 if (r < 0) {
2598 error_message = "Failed to start up manager";
2599 goto finish;
2600 }
2601
2602 /* This will close all file descriptors that were opened, but not claimed by any unit. */
2603 fds = fdset_free(fds);
2604 arg_serialization = safe_fclose(arg_serialization);
2605
2606 if (queue_default_job) {
2607 r = do_queue_default_job(m, &error_message);
2608 if (r < 0)
2609 goto finish;
2610 }
2611
2612 after_startup = now(CLOCK_MONOTONIC);
2613
2614 log_full(arg_action == ACTION_TEST ? LOG_INFO : LOG_DEBUG,
2615 "Loaded units and determined initial transaction in %s.",
2616 format_timespan(timespan, sizeof(timespan), after_startup - before_startup, 100 * USEC_PER_MSEC));
2617
2618 if (arg_action == ACTION_TEST) {
2619 test_summary(m);
2620 retval = EXIT_SUCCESS;
2621 goto finish;
2622 }
2623
2624 (void) invoke_main_loop(m,
2625 &reexecute,
2626 &retval,
2627 &shutdown_verb,
2628 &fds,
2629 &switch_root_dir,
2630 &switch_root_init,
2631 &error_message);
2632
2633 finish:
2634 pager_close();
2635
2636 if (m) {
2637 arg_shutdown_watchdog = m->shutdown_watchdog;
2638 m = manager_free(m);
2639 }
2640
2641 free_arguments();
2642 mac_selinux_finish();
2643
2644 if (reexecute)
2645 do_reexecute(argc, argv,
2646 &saved_rlimit_nofile,
2647 &saved_rlimit_memlock,
2648 fds,
2649 switch_root_dir,
2650 switch_root_init,
2651 &error_message); /* This only returns if reexecution failed */
2652
2653 arg_serialization = safe_fclose(arg_serialization);
2654 fds = fdset_free(fds);
2655
2656 #if HAVE_VALGRIND_VALGRIND_H
2657 /* If we are PID 1 and running under valgrind, then let's exit
2658 * here explicitly. valgrind will only generate nice output on
2659 * exit(), not on exec(), hence let's do the former not the
2660 * latter here. */
2661 if (getpid_cached() == 1 && RUNNING_ON_VALGRIND) {
2662 /* Cleanup watchdog_device strings for valgrind. We need them
2663 * in become_shutdown() so normally we cannot free them yet. */
2664 watchdog_free_device();
2665 arg_watchdog_device = mfree(arg_watchdog_device);
2666 return retval;
2667 }
2668 #endif
2669
2670 #if HAS_FEATURE_ADDRESS_SANITIZER
2671 __lsan_do_leak_check();
2672 #endif
2673
2674 if (shutdown_verb) {
2675 r = become_shutdown(shutdown_verb, retval);
2676 log_error_errno(r, "Failed to execute shutdown binary, %s: %m", getpid_cached() == 1 ? "freezing" : "quitting");
2677 error_message = "Failed to execute shutdown binary";
2678 }
2679
2680 watchdog_free_device();
2681 arg_watchdog_device = mfree(arg_watchdog_device);
2682
2683 if (getpid_cached() == 1) {
2684 if (error_message)
2685 manager_status_printf(NULL, STATUS_TYPE_EMERGENCY,
2686 ANSI_HIGHLIGHT_RED "!!!!!!" ANSI_NORMAL,
2687 "%s.", error_message);
2688 freeze_or_exit_or_reboot();
2689 }
2690
2691 return retval;
2692 }
Unnamed repository; edit this file 'description' to name the repository.