]> git.ipfire.org Git - thirdparty/systemd.git/blob - src/core/main.c
projects / thirdparty / systemd.git / blob
? search:


5da47a0bf65f008816a4067dd71042f9545c7a01
[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 <stdio.h>
7 #include <sys/mount.h>
8 #include <sys/prctl.h>
9 #include <sys/reboot.h>
10 #include <unistd.h>
11 #if HAVE_SECCOMP
12 #include <seccomp.h>
13 #endif
14 #if HAVE_VALGRIND_VALGRIND_H
15 #include <valgrind/valgrind.h>
16 #endif
17
18 #include "sd-bus.h"
19 #include "sd-daemon.h"
20 #include "sd-messages.h"
21
22 #include "alloc-util.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 "efi-random.h"
36 #include "emergency-action.h"
37 #include "env-util.h"
38 #include "exit-status.h"
39 #include "fd-util.h"
40 #include "fdset.h"
41 #include "fileio.h"
42 #include "format-util.h"
43 #include "fs-util.h"
44 #include "hostname-setup.h"
45 #include "ima-setup.h"
46 #include "killall.h"
47 #include "kmod-setup.h"
48 #include "limits-util.h"
49 #include "load-fragment.h"
50 #include "log.h"
51 #include "loopback-setup.h"
52 #include "machine-id-setup.h"
53 #include "manager.h"
54 #include "mount-setup.h"
55 #include "os-util.h"
56 #include "pager.h"
57 #include "parse-util.h"
58 #include "path-util.h"
59 #include "pretty-print.h"
60 #include "proc-cmdline.h"
61 #include "process-util.h"
62 #include "raw-clone.h"
63 #include "rlimit-util.h"
64 #if HAVE_SECCOMP
65 #include "seccomp-util.h"
66 #endif
67 #include "selinux-setup.h"
68 #include "selinux-util.h"
69 #include "signal-util.h"
70 #include "smack-setup.h"
71 #include "special.h"
72 #include "stat-util.h"
73 #include "stdio-util.h"
74 #include "strv.h"
75 #include "switch-root.h"
76 #include "sysctl-util.h"
77 #include "terminal-util.h"
78 #include "umask-util.h"
79 #include "user-util.h"
80 #include "util.h"
81 #include "virt.h"
82 #include "watchdog.h"
83
84 #if HAS_FEATURE_ADDRESS_SANITIZER
85 #include <sanitizer/lsan_interface.h>
86 #endif
87
88 static enum {
89 ACTION_RUN,
90 ACTION_HELP,
91 ACTION_VERSION,
92 ACTION_TEST,
93 ACTION_DUMP_CONFIGURATION_ITEMS,
94 ACTION_DUMP_BUS_PROPERTIES,
95 } arg_action = ACTION_RUN;
96
97 /* Those variables are initalized to 0 automatically, so we avoid uninitialized memory access.
98 * Real defaults are assigned in reset_arguments() below. */
99 static char *arg_default_unit;
100 static bool arg_system;
101 static bool arg_dump_core;
102 static int arg_crash_chvt;
103 static bool arg_crash_shell;
104 static bool arg_crash_reboot;
105 static char *arg_confirm_spawn;
106 static ShowStatus arg_show_status;
107 static StatusUnitFormat arg_status_unit_format;
108 static bool arg_switched_root;
109 static PagerFlags arg_pager_flags;
110 static bool arg_service_watchdogs;
111 static ExecOutput arg_default_std_output;
112 static ExecOutput arg_default_std_error;
113 static usec_t arg_default_restart_usec;
114 static usec_t arg_default_timeout_start_usec;
115 static usec_t arg_default_timeout_stop_usec;
116 static usec_t arg_default_timeout_abort_usec;
117 static bool arg_default_timeout_abort_set;
118 static usec_t arg_default_start_limit_interval;
119 static unsigned arg_default_start_limit_burst;
120 static usec_t arg_runtime_watchdog;
121 static usec_t arg_reboot_watchdog;
122 static usec_t arg_kexec_watchdog;
123 static char *arg_early_core_pattern;
124 static char *arg_watchdog_device;
125 static char **arg_default_environment;
126 static struct rlimit *arg_default_rlimit[_RLIMIT_MAX];
127 static uint64_t arg_capability_bounding_set;
128 static bool arg_no_new_privs;
129 static nsec_t arg_timer_slack_nsec;
130 static usec_t arg_default_timer_accuracy_usec;
131 static Set* arg_syscall_archs;
132 static FILE* arg_serialization;
133 static int arg_default_cpu_accounting;
134 static bool arg_default_io_accounting;
135 static bool arg_default_ip_accounting;
136 static bool arg_default_blockio_accounting;
137 static bool arg_default_memory_accounting;
138 static bool arg_default_tasks_accounting;
139 static uint64_t arg_default_tasks_max;
140 static sd_id128_t arg_machine_id;
141 static EmergencyAction arg_cad_burst_action;
142 static OOMPolicy arg_default_oom_policy;
143 static CPUSet arg_cpu_affinity;
144 static NUMAPolicy arg_numa_policy;
145
146 /* A copy of the original environment block */
147 static char **saved_env = NULL;
148
149 static int parse_configuration(const struct rlimit *saved_rlimit_nofile,
150 const struct rlimit *saved_rlimit_memlock);
151
152 _noreturn_ static void freeze_or_exit_or_reboot(void) {
153
154 /* If we are running in a container, let's prefer exiting, after all we can propagate an exit code to
155 * the container manager, and thus inform it that something went wrong. */
156 if (detect_container() > 0) {
157 log_emergency("Exiting PID 1...");
158 _exit(EXIT_EXCEPTION);
159 }
160
161 if (arg_crash_reboot) {
162 log_notice("Rebooting in 10s...");
163 (void) sleep(10);
164
165 log_notice("Rebooting now...");
166 (void) reboot(RB_AUTOBOOT);
167 log_emergency_errno(errno, "Failed to reboot: %m");
168 }
169
170 log_emergency("Freezing execution.");
171 freeze();
172 }
173
174 _noreturn_ static void crash(int sig) {
175 struct sigaction sa;
176 pid_t pid;
177
178 if (getpid_cached() != 1)
179 /* Pass this on immediately, if this is not PID 1 */
180 (void) raise(sig);
181 else if (!arg_dump_core)
182 log_emergency("Caught <%s>, not dumping core.", signal_to_string(sig));
183 else {
184 sa = (struct sigaction) {
185 .sa_handler = nop_signal_handler,
186 .sa_flags = SA_NOCLDSTOP|SA_RESTART,
187 };
188
189 /* We want to wait for the core process, hence let's enable SIGCHLD */
190 (void) sigaction(SIGCHLD, &sa, NULL);
191
192 pid = raw_clone(SIGCHLD);
193 if (pid < 0)
194 log_emergency_errno(errno, "Caught <%s>, cannot fork for core dump: %m", signal_to_string(sig));
195 else if (pid == 0) {
196 /* Enable default signal handler for core dump */
197
198 sa = (struct sigaction) {
199 .sa_handler = SIG_DFL,
200 };
201 (void) sigaction(sig, &sa, NULL);
202
203 /* Don't limit the coredump size */
204 (void) setrlimit(RLIMIT_CORE, &RLIMIT_MAKE_CONST(RLIM_INFINITY));
205
206 /* Just to be sure... */
207 (void) chdir("/");
208
209 /* Raise the signal again */
210 pid = raw_getpid();
211 (void) kill(pid, sig); /* raise() would kill the parent */
212
213 assert_not_reached("We shouldn't be here...");
214 _exit(EXIT_EXCEPTION);
215 } else {
216 siginfo_t status;
217 int r;
218
219 /* Order things nicely. */
220 r = wait_for_terminate(pid, &status);
221 if (r < 0)
222 log_emergency_errno(r, "Caught <%s>, waitpid() failed: %m", signal_to_string(sig));
223 else if (status.si_code != CLD_DUMPED) {
224 const char *s = status.si_code == CLD_EXITED
225 ? exit_status_to_string(status.si_status, EXIT_STATUS_LIBC)
226 : signal_to_string(status.si_status);
227
228 log_emergency("Caught <%s>, core dump failed (child "PID_FMT", code=%s, status=%i/%s).",
229 signal_to_string(sig),
230 pid,
231 sigchld_code_to_string(status.si_code),
232 status.si_status, strna(s));
233 } else
234 log_emergency("Caught <%s>, dumped core as pid "PID_FMT".",
235 signal_to_string(sig), pid);
236 }
237 }
238
239 if (arg_crash_chvt >= 0)
240 (void) chvt(arg_crash_chvt);
241
242 sa = (struct sigaction) {
243 .sa_handler = SIG_IGN,
244 .sa_flags = SA_NOCLDSTOP|SA_NOCLDWAIT|SA_RESTART,
245 };
246
247 /* Let the kernel reap children for us */
248 (void) sigaction(SIGCHLD, &sa, NULL);
249
250 if (arg_crash_shell) {
251 log_notice("Executing crash shell in 10s...");
252 (void) sleep(10);
253
254 pid = raw_clone(SIGCHLD);
255 if (pid < 0)
256 log_emergency_errno(errno, "Failed to fork off crash shell: %m");
257 else if (pid == 0) {
258 (void) setsid();
259 (void) make_console_stdio();
260 (void) rlimit_nofile_safe();
261 (void) execle("/bin/sh", "/bin/sh", NULL, environ);
262
263 log_emergency_errno(errno, "execle() failed: %m");
264 _exit(EXIT_EXCEPTION);
265 } else {
266 log_info("Spawned crash shell as PID "PID_FMT".", pid);
267 (void) wait_for_terminate(pid, NULL);
268 }
269 }
270
271 freeze_or_exit_or_reboot();
272 }
273
274 static void install_crash_handler(void) {
275 static const struct sigaction sa = {
276 .sa_handler = crash,
277 .sa_flags = SA_NODEFER, /* So that we can raise the signal again from the signal handler */
278 };
279 int r;
280
281 /* We ignore the return value here, since, we don't mind if we
282 * cannot set up a crash handler */
283 r = sigaction_many(&sa, SIGNALS_CRASH_HANDLER, -1);
284 if (r < 0)
285 log_debug_errno(r, "I had trouble setting up the crash handler, ignoring: %m");
286 }
287
288 static int console_setup(void) {
289 _cleanup_close_ int tty_fd = -1;
290 int r;
291
292 tty_fd = open_terminal("/dev/console", O_WRONLY|O_NOCTTY|O_CLOEXEC);
293 if (tty_fd < 0)
294 return log_error_errno(tty_fd, "Failed to open /dev/console: %m");
295
296 /* We don't want to force text mode. plymouth may be showing
297 * pictures already from initrd. */
298 r = reset_terminal_fd(tty_fd, false);
299 if (r < 0)
300 return log_error_errno(r, "Failed to reset /dev/console: %m");
301
302 return 0;
303 }
304
305 static int set_machine_id(const char *m) {
306 sd_id128_t t;
307 assert(m);
308
309 if (sd_id128_from_string(m, &t) < 0)
310 return -EINVAL;
311
312 if (sd_id128_is_null(t))
313 return -EINVAL;
314
315 arg_machine_id = t;
316 return 0;
317 }
318
319 static int parse_proc_cmdline_item(const char *key, const char *value, void *data) {
320
321 int r;
322
323 assert(key);
324
325 if (STR_IN_SET(key, "systemd.unit", "rd.systemd.unit")) {
326
327 if (proc_cmdline_value_missing(key, value))
328 return 0;
329
330 if (!unit_name_is_valid(value, UNIT_NAME_PLAIN|UNIT_NAME_INSTANCE))
331 log_warning("Unit name specified on %s= is not valid, ignoring: %s", key, value);
332 else if (in_initrd() == !!startswith(key, "rd.")) {
333 if (free_and_strdup(&arg_default_unit, value) < 0)
334 return log_oom();
335 }
336
337 } else if (proc_cmdline_key_streq(key, "systemd.dump_core")) {
338
339 r = value ? parse_boolean(value) : true;
340 if (r < 0)
341 log_warning_errno(r, "Failed to parse dump core switch %s, ignoring: %m", value);
342 else
343 arg_dump_core = r;
344
345 } else if (proc_cmdline_key_streq(key, "systemd.early_core_pattern")) {
346
347 if (proc_cmdline_value_missing(key, value))
348 return 0;
349
350 if (path_is_absolute(value))
351 (void) parse_path_argument_and_warn(value, false, &arg_early_core_pattern);
352 else
353 log_warning("Specified core pattern '%s' is not an absolute path, ignoring.", value);
354
355 } else if (proc_cmdline_key_streq(key, "systemd.crash_chvt")) {
356
357 if (!value)
358 arg_crash_chvt = 0; /* turn on */
359 else {
360 r = parse_crash_chvt(value, &arg_crash_chvt);
361 if (r < 0)
362 log_warning_errno(r, "Failed to parse crash chvt switch %s, ignoring: %m", value);
363 }
364
365 } else if (proc_cmdline_key_streq(key, "systemd.crash_shell")) {
366
367 r = value ? parse_boolean(value) : true;
368 if (r < 0)
369 log_warning_errno(r, "Failed to parse crash shell switch %s, ignoring: %m", value);
370 else
371 arg_crash_shell = r;
372
373 } else if (proc_cmdline_key_streq(key, "systemd.crash_reboot")) {
374
375 r = value ? parse_boolean(value) : true;
376 if (r < 0)
377 log_warning_errno(r, "Failed to parse crash reboot switch %s, ignoring: %m", value);
378 else
379 arg_crash_reboot = r;
380
381 } else if (proc_cmdline_key_streq(key, "systemd.confirm_spawn")) {
382 char *s;
383
384 r = parse_confirm_spawn(value, &s);
385 if (r < 0)
386 log_warning_errno(r, "Failed to parse confirm_spawn switch %s, ignoring: %m", value);
387 else
388 free_and_replace(arg_confirm_spawn, s);
389
390 } else if (proc_cmdline_key_streq(key, "systemd.service_watchdogs")) {
391
392 r = value ? parse_boolean(value) : true;
393 if (r < 0)
394 log_warning_errno(r, "Failed to parse service watchdog switch %s, ignoring: %m", value);
395 else
396 arg_service_watchdogs = r;
397
398 } else if (proc_cmdline_key_streq(key, "systemd.show_status")) {
399
400 if (value) {
401 r = parse_show_status(value, &arg_show_status);
402 if (r < 0)
403 log_warning_errno(r, "Failed to parse show status switch %s, ignoring: %m", value);
404 } else
405 arg_show_status = SHOW_STATUS_YES;
406
407 } else if (proc_cmdline_key_streq(key, "systemd.status_unit_format")) {
408
409 if (proc_cmdline_value_missing(key, value))
410 return 0;
411
412 r = status_unit_format_from_string(value);
413 if (r < 0)
414 log_warning_errno(r, "Failed to parse %s=%s, ignoring: %m", key, value);
415 else
416 arg_status_unit_format = r;
417
418 } else if (proc_cmdline_key_streq(key, "systemd.default_standard_output")) {
419
420 if (proc_cmdline_value_missing(key, value))
421 return 0;
422
423 r = exec_output_from_string(value);
424 if (r < 0)
425 log_warning_errno(r, "Failed to parse default standard output switch %s, ignoring: %m", value);
426 else
427 arg_default_std_output = r;
428
429 } else if (proc_cmdline_key_streq(key, "systemd.default_standard_error")) {
430
431 if (proc_cmdline_value_missing(key, value))
432 return 0;
433
434 r = exec_output_from_string(value);
435 if (r < 0)
436 log_warning_errno(r, "Failed to parse default standard error switch %s, ignoring: %m", value);
437 else
438 arg_default_std_error = r;
439
440 } else if (streq(key, "systemd.setenv")) {
441
442 if (proc_cmdline_value_missing(key, value))
443 return 0;
444
445 if (env_assignment_is_valid(value)) {
446 char **env;
447
448 env = strv_env_set(arg_default_environment, value);
449 if (!env)
450 return log_oom();
451
452 arg_default_environment = env;
453 } else
454 log_warning("Environment variable name '%s' is not valid. Ignoring.", value);
455
456 } else if (proc_cmdline_key_streq(key, "systemd.machine_id")) {
457
458 if (proc_cmdline_value_missing(key, value))
459 return 0;
460
461 r = set_machine_id(value);
462 if (r < 0)
463 log_warning_errno(r, "MachineID '%s' is not valid, ignoring: %m", value);
464
465 } else if (proc_cmdline_key_streq(key, "systemd.default_timeout_start_sec")) {
466
467 if (proc_cmdline_value_missing(key, value))
468 return 0;
469
470 r = parse_sec(value, &arg_default_timeout_start_usec);
471 if (r < 0)
472 log_warning_errno(r, "Failed to parse default start timeout '%s', ignoring: %m", value);
473
474 if (arg_default_timeout_start_usec <= 0)
475 arg_default_timeout_start_usec = USEC_INFINITY;
476
477 } else if (proc_cmdline_key_streq(key, "systemd.watchdog_device")) {
478
479 if (proc_cmdline_value_missing(key, value))
480 return 0;
481
482 (void) parse_path_argument_and_warn(value, false, &arg_watchdog_device);
483
484 } else if (streq(key, "quiet") && !value) {
485
486 if (arg_show_status == _SHOW_STATUS_INVALID)
487 arg_show_status = SHOW_STATUS_AUTO;
488
489 } else if (streq(key, "debug") && !value) {
490
491 /* Note that log_parse_environment() handles 'debug'
492 * too, and sets the log level to LOG_DEBUG. */
493
494 if (detect_container() > 0)
495 log_set_target(LOG_TARGET_CONSOLE);
496
497 } else if (!value) {
498 const char *target;
499
500 /* SysV compatibility */
501 target = runlevel_to_target(key);
502 if (target)
503 return free_and_strdup(&arg_default_unit, target);
504 }
505
506 return 0;
507 }
508
509 #define DEFINE_SETTER(name, func, descr) \
510 static int name(const char *unit, \
511 const char *filename, \
512 unsigned line, \
513 const char *section, \
514 unsigned section_line, \
515 const char *lvalue, \
516 int ltype, \
517 const char *rvalue, \
518 void *data, \
519 void *userdata) { \
520 \
521 int r; \
522 \
523 assert(filename); \
524 assert(lvalue); \
525 assert(rvalue); \
526 \
527 r = func(rvalue); \
528 if (r < 0) \
529 log_syntax(unit, LOG_ERR, filename, line, r, \
530 "Invalid " descr "'%s': %m", \
531 rvalue); \
532 \
533 return 0; \
534 }
535
536 DEFINE_SETTER(config_parse_level2, log_set_max_level_from_string, "log level");
537 DEFINE_SETTER(config_parse_target, log_set_target_from_string, "target");
538 DEFINE_SETTER(config_parse_color, log_show_color_from_string, "color" );
539 DEFINE_SETTER(config_parse_location, log_show_location_from_string, "location");
540
541 static int parse_config_file(void) {
542
543 const ConfigTableItem items[] = {
544 { "Manager", "LogLevel", config_parse_level2, 0, NULL },
545 { "Manager", "LogTarget", config_parse_target, 0, NULL },
546 { "Manager", "LogColor", config_parse_color, 0, NULL },
547 { "Manager", "LogLocation", config_parse_location, 0, NULL },
548 { "Manager", "DumpCore", config_parse_bool, 0, &arg_dump_core },
549 { "Manager", "CrashChVT", /* legacy */ config_parse_crash_chvt, 0, &arg_crash_chvt },
550 { "Manager", "CrashChangeVT", config_parse_crash_chvt, 0, &arg_crash_chvt },
551 { "Manager", "CrashShell", config_parse_bool, 0, &arg_crash_shell },
552 { "Manager", "CrashReboot", config_parse_bool, 0, &arg_crash_reboot },
553 { "Manager", "ShowStatus", config_parse_show_status, 0, &arg_show_status },
554 { "Manager", "StatusUnitFormat", config_parse_status_unit_format, 0, &arg_status_unit_format },
555 { "Manager", "CPUAffinity", config_parse_cpu_affinity2, 0, &arg_cpu_affinity },
556 { "Manager", "NUMAPolicy", config_parse_numa_policy, 0, &arg_numa_policy.type },
557 { "Manager", "NUMAMask", config_parse_numa_mask, 0, &arg_numa_policy },
558 { "Manager", "JoinControllers", config_parse_warn_compat, DISABLED_CONFIGURATION, NULL },
559 { "Manager", "RuntimeWatchdogSec", config_parse_sec, 0, &arg_runtime_watchdog },
560 { "Manager", "RebootWatchdogSec", config_parse_sec, 0, &arg_reboot_watchdog },
561 { "Manager", "ShutdownWatchdogSec", config_parse_sec, 0, &arg_reboot_watchdog }, /* obsolete alias */
562 { "Manager", "KExecWatchdogSec", config_parse_sec, 0, &arg_kexec_watchdog },
563 { "Manager", "WatchdogDevice", config_parse_path, 0, &arg_watchdog_device },
564 { "Manager", "CapabilityBoundingSet", config_parse_capability_set, 0, &arg_capability_bounding_set },
565 { "Manager", "NoNewPrivileges", config_parse_bool, 0, &arg_no_new_privs },
566 #if HAVE_SECCOMP
567 { "Manager", "SystemCallArchitectures", config_parse_syscall_archs, 0, &arg_syscall_archs },
568 #endif
569 { "Manager", "TimerSlackNSec", config_parse_nsec, 0, &arg_timer_slack_nsec },
570 { "Manager", "DefaultTimerAccuracySec", config_parse_sec, 0, &arg_default_timer_accuracy_usec },
571 { "Manager", "DefaultStandardOutput", config_parse_output_restricted, 0, &arg_default_std_output },
572 { "Manager", "DefaultStandardError", config_parse_output_restricted, 0, &arg_default_std_error },
573 { "Manager", "DefaultTimeoutStartSec", config_parse_sec, 0, &arg_default_timeout_start_usec },
574 { "Manager", "DefaultTimeoutStopSec", config_parse_sec, 0, &arg_default_timeout_stop_usec },
575 { "Manager", "DefaultTimeoutAbortSec", config_parse_timeout_abort, 0, &arg_default_timeout_abort_set },
576 { "Manager", "DefaultRestartSec", config_parse_sec, 0, &arg_default_restart_usec },
577 { "Manager", "DefaultStartLimitInterval", config_parse_sec, 0, &arg_default_start_limit_interval }, /* obsolete alias */
578 { "Manager", "DefaultStartLimitIntervalSec", config_parse_sec, 0, &arg_default_start_limit_interval },
579 { "Manager", "DefaultStartLimitBurst", config_parse_unsigned, 0, &arg_default_start_limit_burst },
580 { "Manager", "DefaultEnvironment", config_parse_environ, 0, &arg_default_environment },
581 { "Manager", "DefaultLimitCPU", config_parse_rlimit, RLIMIT_CPU, arg_default_rlimit },
582 { "Manager", "DefaultLimitFSIZE", config_parse_rlimit, RLIMIT_FSIZE, arg_default_rlimit },
583 { "Manager", "DefaultLimitDATA", config_parse_rlimit, RLIMIT_DATA, arg_default_rlimit },
584 { "Manager", "DefaultLimitSTACK", config_parse_rlimit, RLIMIT_STACK, arg_default_rlimit },
585 { "Manager", "DefaultLimitCORE", config_parse_rlimit, RLIMIT_CORE, arg_default_rlimit },
586 { "Manager", "DefaultLimitRSS", config_parse_rlimit, RLIMIT_RSS, arg_default_rlimit },
587 { "Manager", "DefaultLimitNOFILE", config_parse_rlimit, RLIMIT_NOFILE, arg_default_rlimit },
588 { "Manager", "DefaultLimitAS", config_parse_rlimit, RLIMIT_AS, arg_default_rlimit },
589 { "Manager", "DefaultLimitNPROC", config_parse_rlimit, RLIMIT_NPROC, arg_default_rlimit },
590 { "Manager", "DefaultLimitMEMLOCK", config_parse_rlimit, RLIMIT_MEMLOCK, arg_default_rlimit },
591 { "Manager", "DefaultLimitLOCKS", config_parse_rlimit, RLIMIT_LOCKS, arg_default_rlimit },
592 { "Manager", "DefaultLimitSIGPENDING", config_parse_rlimit, RLIMIT_SIGPENDING, arg_default_rlimit },
593 { "Manager", "DefaultLimitMSGQUEUE", config_parse_rlimit, RLIMIT_MSGQUEUE, arg_default_rlimit },
594 { "Manager", "DefaultLimitNICE", config_parse_rlimit, RLIMIT_NICE, arg_default_rlimit },
595 { "Manager", "DefaultLimitRTPRIO", config_parse_rlimit, RLIMIT_RTPRIO, arg_default_rlimit },
596 { "Manager", "DefaultLimitRTTIME", config_parse_rlimit, RLIMIT_RTTIME, arg_default_rlimit },
597 { "Manager", "DefaultCPUAccounting", config_parse_tristate, 0, &arg_default_cpu_accounting },
598 { "Manager", "DefaultIOAccounting", config_parse_bool, 0, &arg_default_io_accounting },
599 { "Manager", "DefaultIPAccounting", config_parse_bool, 0, &arg_default_ip_accounting },
600 { "Manager", "DefaultBlockIOAccounting", config_parse_bool, 0, &arg_default_blockio_accounting },
601 { "Manager", "DefaultMemoryAccounting", config_parse_bool, 0, &arg_default_memory_accounting },
602 { "Manager", "DefaultTasksAccounting", config_parse_bool, 0, &arg_default_tasks_accounting },
603 { "Manager", "DefaultTasksMax", config_parse_tasks_max, 0, &arg_default_tasks_max },
604 { "Manager", "CtrlAltDelBurstAction", config_parse_emergency_action, 0, &arg_cad_burst_action },
605 { "Manager", "DefaultOOMPolicy", config_parse_oom_policy, 0, &arg_default_oom_policy },
606 {}
607 };
608
609 const char *fn, *conf_dirs_nulstr;
610
611 fn = arg_system ?
612 PKGSYSCONFDIR "/system.conf" :
613 PKGSYSCONFDIR "/user.conf";
614
615 conf_dirs_nulstr = arg_system ?
616 CONF_PATHS_NULSTR("systemd/system.conf.d") :
617 CONF_PATHS_NULSTR("systemd/user.conf.d");
618
619 (void) config_parse_many_nulstr(fn, conf_dirs_nulstr, "Manager\0", config_item_table_lookup, items, CONFIG_PARSE_WARN, NULL);
620
621 /* Traditionally "0" was used to turn off the default unit timeouts. Fix this up so that we used USEC_INFINITY
622 * like everywhere else. */
623 if (arg_default_timeout_start_usec <= 0)
624 arg_default_timeout_start_usec = USEC_INFINITY;
625 if (arg_default_timeout_stop_usec <= 0)
626 arg_default_timeout_stop_usec = USEC_INFINITY;
627
628 return 0;
629 }
630
631 static void set_manager_defaults(Manager *m) {
632
633 assert(m);
634
635 /* Propagates the various default unit property settings into the manager object, i.e. properties that do not
636 * affect the manager itself, but are just what newly allocated units will have set if they haven't set
637 * anything else. (Also see set_manager_settings() for the settings that affect the manager's own behaviour) */
638
639 m->default_timer_accuracy_usec = arg_default_timer_accuracy_usec;
640 m->default_std_output = arg_default_std_output;
641 m->default_std_error = arg_default_std_error;
642 m->default_timeout_start_usec = arg_default_timeout_start_usec;
643 m->default_timeout_stop_usec = arg_default_timeout_stop_usec;
644 m->default_timeout_abort_usec = arg_default_timeout_abort_usec;
645 m->default_timeout_abort_set = arg_default_timeout_abort_set;
646 m->default_restart_usec = arg_default_restart_usec;
647 m->default_start_limit_interval = arg_default_start_limit_interval;
648 m->default_start_limit_burst = arg_default_start_limit_burst;
649
650 /* On 4.15+ with unified hierarchy, CPU accounting is essentially free as it doesn't require the CPU
651 * controller to be enabled, so the default is to enable it unless we got told otherwise. */
652 if (arg_default_cpu_accounting >= 0)
653 m->default_cpu_accounting = arg_default_cpu_accounting;
654 else
655 m->default_cpu_accounting = cpu_accounting_is_cheap();
656
657 m->default_io_accounting = arg_default_io_accounting;
658 m->default_ip_accounting = arg_default_ip_accounting;
659 m->default_blockio_accounting = arg_default_blockio_accounting;
660 m->default_memory_accounting = arg_default_memory_accounting;
661 m->default_tasks_accounting = arg_default_tasks_accounting;
662 m->default_tasks_max = arg_default_tasks_max;
663 m->default_oom_policy = arg_default_oom_policy;
664
665 (void) manager_set_default_rlimits(m, arg_default_rlimit);
666
667 (void) manager_default_environment(m);
668 (void) manager_transient_environment_add(m, arg_default_environment);
669 }
670
671 static void set_manager_settings(Manager *m) {
672
673 assert(m);
674
675 /* Propagates the various manager settings into the manager object, i.e. properties that effect the manager
676 * itself (as opposed to just being inherited into newly allocated units, see set_manager_defaults() above). */
677
678 m->confirm_spawn = arg_confirm_spawn;
679 m->service_watchdogs = arg_service_watchdogs;
680 m->runtime_watchdog = arg_runtime_watchdog;
681 m->reboot_watchdog = arg_reboot_watchdog;
682 m->kexec_watchdog = arg_kexec_watchdog;
683 m->cad_burst_action = arg_cad_burst_action;
684
685 manager_set_show_status(m, arg_show_status);
686 m->status_unit_format = arg_status_unit_format;
687 }
688
689 static int parse_argv(int argc, char *argv[]) {
690 enum {
691 ARG_LOG_LEVEL = 0x100,
692 ARG_LOG_TARGET,
693 ARG_LOG_COLOR,
694 ARG_LOG_LOCATION,
695 ARG_UNIT,
696 ARG_SYSTEM,
697 ARG_USER,
698 ARG_TEST,
699 ARG_NO_PAGER,
700 ARG_VERSION,
701 ARG_DUMP_CONFIGURATION_ITEMS,
702 ARG_DUMP_BUS_PROPERTIES,
703 ARG_DUMP_CORE,
704 ARG_CRASH_CHVT,
705 ARG_CRASH_SHELL,
706 ARG_CRASH_REBOOT,
707 ARG_CONFIRM_SPAWN,
708 ARG_SHOW_STATUS,
709 ARG_DESERIALIZE,
710 ARG_SWITCHED_ROOT,
711 ARG_DEFAULT_STD_OUTPUT,
712 ARG_DEFAULT_STD_ERROR,
713 ARG_MACHINE_ID,
714 ARG_SERVICE_WATCHDOGS,
715 };
716
717 static const struct option options[] = {
718 { "log-level", required_argument, NULL, ARG_LOG_LEVEL },
719 { "log-target", required_argument, NULL, ARG_LOG_TARGET },
720 { "log-color", optional_argument, NULL, ARG_LOG_COLOR },
721 { "log-location", optional_argument, NULL, ARG_LOG_LOCATION },
722 { "unit", required_argument, NULL, ARG_UNIT },
723 { "system", no_argument, NULL, ARG_SYSTEM },
724 { "user", no_argument, NULL, ARG_USER },
725 { "test", no_argument, NULL, ARG_TEST },
726 { "no-pager", no_argument, NULL, ARG_NO_PAGER },
727 { "help", no_argument, NULL, 'h' },
728 { "version", no_argument, NULL, ARG_VERSION },
729 { "dump-configuration-items", no_argument, NULL, ARG_DUMP_CONFIGURATION_ITEMS },
730 { "dump-bus-properties", no_argument, NULL, ARG_DUMP_BUS_PROPERTIES },
731 { "dump-core", optional_argument, NULL, ARG_DUMP_CORE },
732 { "crash-chvt", required_argument, NULL, ARG_CRASH_CHVT },
733 { "crash-shell", optional_argument, NULL, ARG_CRASH_SHELL },
734 { "crash-reboot", optional_argument, NULL, ARG_CRASH_REBOOT },
735 { "confirm-spawn", optional_argument, NULL, ARG_CONFIRM_SPAWN },
736 { "show-status", optional_argument, NULL, ARG_SHOW_STATUS },
737 { "deserialize", required_argument, NULL, ARG_DESERIALIZE },
738 { "switched-root", no_argument, NULL, ARG_SWITCHED_ROOT },
739 { "default-standard-output", required_argument, NULL, ARG_DEFAULT_STD_OUTPUT, },
740 { "default-standard-error", required_argument, NULL, ARG_DEFAULT_STD_ERROR, },
741 { "machine-id", required_argument, NULL, ARG_MACHINE_ID },
742 { "service-watchdogs", required_argument, NULL, ARG_SERVICE_WATCHDOGS },
743 {}
744 };
745
746 int c, r;
747
748 assert(argc >= 1);
749 assert(argv);
750
751 if (getpid_cached() == 1)
752 opterr = 0;
753
754 while ((c = getopt_long(argc, argv, "hDbsz:", options, NULL)) >= 0)
755
756 switch (c) {
757
758 case ARG_LOG_LEVEL:
759 r = log_set_max_level_from_string(optarg);
760 if (r < 0)
761 return log_error_errno(r, "Failed to parse log level \"%s\": %m", optarg);
762
763 break;
764
765 case ARG_LOG_TARGET:
766 r = log_set_target_from_string(optarg);
767 if (r < 0)
768 return log_error_errno(r, "Failed to parse log target \"%s\": %m", optarg);
769
770 break;
771
772 case ARG_LOG_COLOR:
773
774 if (optarg) {
775 r = log_show_color_from_string(optarg);
776 if (r < 0)
777 return log_error_errno(r, "Failed to parse log color setting \"%s\": %m",
778 optarg);
779 } else
780 log_show_color(true);
781
782 break;
783
784 case ARG_LOG_LOCATION:
785 if (optarg) {
786 r = log_show_location_from_string(optarg);
787 if (r < 0)
788 return log_error_errno(r, "Failed to parse log location setting \"%s\": %m",
789 optarg);
790 } else
791 log_show_location(true);
792
793 break;
794
795 case ARG_DEFAULT_STD_OUTPUT:
796 r = exec_output_from_string(optarg);
797 if (r < 0)
798 return log_error_errno(r, "Failed to parse default standard output setting \"%s\": %m",
799 optarg);
800 arg_default_std_output = r;
801 break;
802
803 case ARG_DEFAULT_STD_ERROR:
804 r = exec_output_from_string(optarg);
805 if (r < 0)
806 return log_error_errno(r, "Failed to parse default standard error output setting \"%s\": %m",
807 optarg);
808 arg_default_std_error = r;
809 break;
810
811 case ARG_UNIT:
812 r = free_and_strdup(&arg_default_unit, optarg);
813 if (r < 0)
814 return log_error_errno(r, "Failed to set default unit \"%s\": %m", optarg);
815
816 break;
817
818 case ARG_SYSTEM:
819 arg_system = true;
820 break;
821
822 case ARG_USER:
823 arg_system = false;
824 break;
825
826 case ARG_TEST:
827 arg_action = ACTION_TEST;
828 break;
829
830 case ARG_NO_PAGER:
831 arg_pager_flags |= PAGER_DISABLE;
832 break;
833
834 case ARG_VERSION:
835 arg_action = ACTION_VERSION;
836 break;
837
838 case ARG_DUMP_CONFIGURATION_ITEMS:
839 arg_action = ACTION_DUMP_CONFIGURATION_ITEMS;
840 break;
841
842 case ARG_DUMP_BUS_PROPERTIES:
843 arg_action = ACTION_DUMP_BUS_PROPERTIES;
844 break;
845
846 case ARG_DUMP_CORE:
847 if (!optarg)
848 arg_dump_core = true;
849 else {
850 r = parse_boolean(optarg);
851 if (r < 0)
852 return log_error_errno(r, "Failed to parse dump core boolean: \"%s\": %m",
853 optarg);
854 arg_dump_core = r;
855 }
856 break;
857
858 case ARG_CRASH_CHVT:
859 r = parse_crash_chvt(optarg, &arg_crash_chvt);
860 if (r < 0)
861 return log_error_errno(r, "Failed to parse crash virtual terminal index: \"%s\": %m",
862 optarg);
863 break;
864
865 case ARG_CRASH_SHELL:
866 if (!optarg)
867 arg_crash_shell = true;
868 else {
869 r = parse_boolean(optarg);
870 if (r < 0)
871 return log_error_errno(r, "Failed to parse crash shell boolean: \"%s\": %m",
872 optarg);
873 arg_crash_shell = r;
874 }
875 break;
876
877 case ARG_CRASH_REBOOT:
878 if (!optarg)
879 arg_crash_reboot = true;
880 else {
881 r = parse_boolean(optarg);
882 if (r < 0)
883 return log_error_errno(r, "Failed to parse crash shell boolean: \"%s\": %m",
884 optarg);
885 arg_crash_reboot = r;
886 }
887 break;
888
889 case ARG_CONFIRM_SPAWN:
890 arg_confirm_spawn = mfree(arg_confirm_spawn);
891
892 r = parse_confirm_spawn(optarg, &arg_confirm_spawn);
893 if (r < 0)
894 return log_error_errno(r, "Failed to parse confirm spawn option: \"%s\": %m",
895 optarg);
896 break;
897
898 case ARG_SERVICE_WATCHDOGS:
899 r = parse_boolean(optarg);
900 if (r < 0)
901 return log_error_errno(r, "Failed to parse service watchdogs boolean: \"%s\": %m",
902 optarg);
903 arg_service_watchdogs = r;
904 break;
905
906 case ARG_SHOW_STATUS:
907 if (optarg) {
908 r = parse_show_status(optarg, &arg_show_status);
909 if (r < 0)
910 return log_error_errno(r, "Failed to parse show status boolean: \"%s\": %m",
911 optarg);
912 } else
913 arg_show_status = SHOW_STATUS_YES;
914 break;
915
916 case ARG_DESERIALIZE: {
917 int fd;
918 FILE *f;
919
920 r = safe_atoi(optarg, &fd);
921 if (r < 0)
922 log_error_errno(r, "Failed to parse deserialize option \"%s\": %m", optarg);
923 if (fd < 0)
924 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
925 "Invalid deserialize fd: %d",
926 fd);
927
928 (void) fd_cloexec(fd, true);
929
930 f = fdopen(fd, "r");
931 if (!f)
932 return log_error_errno(errno, "Failed to open serialization fd %d: %m", fd);
933
934 safe_fclose(arg_serialization);
935 arg_serialization = f;
936
937 break;
938 }
939
940 case ARG_SWITCHED_ROOT:
941 arg_switched_root = true;
942 break;
943
944 case ARG_MACHINE_ID:
945 r = set_machine_id(optarg);
946 if (r < 0)
947 return log_error_errno(r, "MachineID '%s' is not valid: %m", optarg);
948 break;
949
950 case 'h':
951 arg_action = ACTION_HELP;
952 break;
953
954 case 'D':
955 log_set_max_level(LOG_DEBUG);
956 break;
957
958 case 'b':
959 case 's':
960 case 'z':
961 /* Just to eat away the sysvinit kernel
962 * cmdline args without getopt() error
963 * messages that we'll parse in
964 * parse_proc_cmdline_word() or ignore. */
965
966 case '?':
967 if (getpid_cached() != 1)
968 return -EINVAL;
969 else
970 return 0;
971
972 default:
973 assert_not_reached("Unhandled option code.");
974 }
975
976 if (optind < argc && getpid_cached() != 1) {
977 /* Hmm, when we aren't run as init system
978 * let's complain about excess arguments */
979
980 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
981 "Excess arguments.");
982 }
983
984 return 0;
985 }
986
987 static int help(void) {
988 _cleanup_free_ char *link = NULL;
989 int r;
990
991 r = terminal_urlify_man("systemd", "1", &link);
992 if (r < 0)
993 return log_oom();
994
995 printf("%s [OPTIONS...]\n\n"
996 "Starts up and maintains the system or user services.\n\n"
997 " -h --help Show this help\n"
998 " --version Show version\n"
999 " --test Determine initial transaction, dump it and exit\n"
1000 " --system In combination with --test: operate as system service manager\n"
1001 " --user In combination with --test: operate as per-user service manager\n"
1002 " --no-pager Do not pipe output into a pager\n"
1003 " --dump-configuration-items Dump understood unit configuration items\n"
1004 " --dump-bus-properties Dump exposed bus properties\n"
1005 " --unit=UNIT Set default unit\n"
1006 " --dump-core[=BOOL] Dump core on crash\n"
1007 " --crash-vt=NR Change to specified VT on crash\n"
1008 " --crash-reboot[=BOOL] Reboot on crash\n"
1009 " --crash-shell[=BOOL] Run shell on crash\n"
1010 " --confirm-spawn[=BOOL] Ask for confirmation when spawning processes\n"
1011 " --show-status[=BOOL] Show status updates on the console during bootup\n"
1012 " --log-target=TARGET Set log target (console, journal, kmsg, journal-or-kmsg, null)\n"
1013 " --log-level=LEVEL Set log level (debug, info, notice, warning, err, crit, alert, emerg)\n"
1014 " --log-color[=BOOL] Highlight important log messages\n"
1015 " --log-location[=BOOL] Include code location in log messages\n"
1016 " --default-standard-output= Set default standard output for services\n"
1017 " --default-standard-error= Set default standard error output for services\n"
1018 "\nSee the %s for details.\n"
1019 , program_invocation_short_name
1020 , link
1021 );
1022
1023 return 0;
1024 }
1025
1026 static int prepare_reexecute(
1027 Manager *m,
1028 FILE **ret_f,
1029 FDSet **ret_fds,
1030 bool switching_root) {
1031
1032 _cleanup_fdset_free_ FDSet *fds = NULL;
1033 _cleanup_fclose_ FILE *f = NULL;
1034 int r;
1035
1036 assert(m);
1037 assert(ret_f);
1038 assert(ret_fds);
1039
1040 r = manager_open_serialization(m, &f);
1041 if (r < 0)
1042 return log_error_errno(r, "Failed to create serialization file: %m");
1043
1044 /* Make sure nothing is really destructed when we shut down */
1045 m->n_reloading++;
1046 bus_manager_send_reloading(m, true);
1047
1048 fds = fdset_new();
1049 if (!fds)
1050 return log_oom();
1051
1052 r = manager_serialize(m, f, fds, switching_root);
1053 if (r < 0)
1054 return r;
1055
1056 if (fseeko(f, 0, SEEK_SET) == (off_t) -1)
1057 return log_error_errno(errno, "Failed to rewind serialization fd: %m");
1058
1059 r = fd_cloexec(fileno(f), false);
1060 if (r < 0)
1061 return log_error_errno(r, "Failed to disable O_CLOEXEC for serialization: %m");
1062
1063 r = fdset_cloexec(fds, false);
1064 if (r < 0)
1065 return log_error_errno(r, "Failed to disable O_CLOEXEC for serialization fds: %m");
1066
1067 *ret_f = TAKE_PTR(f);
1068 *ret_fds = TAKE_PTR(fds);
1069
1070 return 0;
1071 }
1072
1073 static void bump_file_max_and_nr_open(void) {
1074
1075 /* Let's bump fs.file-max and fs.nr_open to their respective maximums. On current kernels large numbers of file
1076 * descriptors are no longer a performance problem and their memory is properly tracked by memcg, thus counting
1077 * them and limiting them in another two layers of limits is unnecessary and just complicates things. This
1078 * function hence turns off 2 of the 4 levels of limits on file descriptors, and makes RLIMIT_NOLIMIT (soft +
1079 * hard) the only ones that really matter. */
1080
1081 #if BUMP_PROC_SYS_FS_FILE_MAX || BUMP_PROC_SYS_FS_NR_OPEN
1082 int r;
1083 #endif
1084
1085 #if BUMP_PROC_SYS_FS_FILE_MAX
1086 /* The maximum the kernel allows for this since 5.2 is LONG_MAX, use that. (Previously thing where
1087 * different but the operation would fail silently.) */
1088 r = sysctl_writef("fs/file-max", "%li\n", LONG_MAX);
1089 if (r < 0)
1090 log_full_errno(IN_SET(r, -EROFS, -EPERM, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, "Failed to bump fs.file-max, ignoring: %m");
1091 #endif
1092
1093 #if BUMP_PROC_SYS_FS_NR_OPEN
1094 int v = INT_MAX;
1095
1096 /* Arg! The kernel enforces maximum and minimum values on the fs.nr_open, but we don't really know what they
1097 * are. The expression by which the maximum is determined is dependent on the architecture, and is something we
1098 * don't really want to copy to userspace, as it is dependent on implementation details of the kernel. Since
1099 * the kernel doesn't expose the maximum value to us, we can only try and hope. Hence, let's start with
1100 * INT_MAX, and then keep halving the value until we find one that works. Ugly? Yes, absolutely, but kernel
1101 * APIs are kernel APIs, so what do can we do... 🤯 */
1102
1103 for (;;) {
1104 int k;
1105
1106 v &= ~(__SIZEOF_POINTER__ - 1); /* Round down to next multiple of the pointer size */
1107 if (v < 1024) {
1108 log_warning("Can't bump fs.nr_open, value too small.");
1109 break;
1110 }
1111
1112 k = read_nr_open();
1113 if (k < 0) {
1114 log_error_errno(k, "Failed to read fs.nr_open: %m");
1115 break;
1116 }
1117 if (k >= v) { /* Already larger */
1118 log_debug("Skipping bump, value is already larger.");
1119 break;
1120 }
1121
1122 r = sysctl_writef("fs/nr_open", "%i\n", v);
1123 if (r == -EINVAL) {
1124 log_debug("Couldn't write fs.nr_open as %i, halving it.", v);
1125 v /= 2;
1126 continue;
1127 }
1128 if (r < 0) {
1129 log_full_errno(IN_SET(r, -EROFS, -EPERM, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, "Failed to bump fs.nr_open, ignoring: %m");
1130 break;
1131 }
1132
1133 log_debug("Successfully bumped fs.nr_open to %i", v);
1134 break;
1135 }
1136 #endif
1137 }
1138
1139 static int bump_rlimit_nofile(struct rlimit *saved_rlimit) {
1140 struct rlimit new_rlimit;
1141 int r, nr;
1142
1143 /* Get the underlying absolute limit the kernel enforces */
1144 nr = read_nr_open();
1145
1146 /* Calculate the new limits to use for us. Never lower from what we inherited. */
1147 new_rlimit = (struct rlimit) {
1148 .rlim_cur = MAX((rlim_t) nr, saved_rlimit->rlim_cur),
1149 .rlim_max = MAX((rlim_t) nr, saved_rlimit->rlim_max),
1150 };
1151
1152 /* Shortcut if nothing changes. */
1153 if (saved_rlimit->rlim_max >= new_rlimit.rlim_max &&
1154 saved_rlimit->rlim_cur >= new_rlimit.rlim_cur) {
1155 log_debug("RLIMIT_NOFILE is already as high or higher than we need it, not bumping.");
1156 return 0;
1157 }
1158
1159 /* Bump up the resource limit for ourselves substantially, all the way to the maximum the kernel allows, for
1160 * both hard and soft. */
1161 r = setrlimit_closest(RLIMIT_NOFILE, &new_rlimit);
1162 if (r < 0)
1163 return log_warning_errno(r, "Setting RLIMIT_NOFILE failed, ignoring: %m");
1164
1165 return 0;
1166 }
1167
1168 static int bump_rlimit_memlock(struct rlimit *saved_rlimit) {
1169 struct rlimit new_rlimit;
1170 int r;
1171
1172 /* BPF_MAP_TYPE_LPM_TRIE bpf maps are charged against RLIMIT_MEMLOCK, even if we have CAP_IPC_LOCK which should
1173 * normally disable such checks. We need them to implement IPAccessAllow= and IPAccessDeny=, hence let's bump
1174 * the value high enough for our user. */
1175
1176 /* Using MAX() on resource limits only is safe if RLIM_INFINITY is > 0. POSIX declares that rlim_t
1177 * must be unsigned, hence this is a given, but let's make this clear here. */
1178 assert_cc(RLIM_INFINITY > 0);
1179
1180 new_rlimit = (struct rlimit) {
1181 .rlim_cur = MAX(HIGH_RLIMIT_MEMLOCK, saved_rlimit->rlim_cur),
1182 .rlim_max = MAX(HIGH_RLIMIT_MEMLOCK, saved_rlimit->rlim_max),
1183 };
1184
1185 if (saved_rlimit->rlim_max >= new_rlimit.rlim_cur &&
1186 saved_rlimit->rlim_cur >= new_rlimit.rlim_max) {
1187 log_debug("RLIMIT_MEMLOCK is already as high or higher than we need it, not bumping.");
1188 return 0;
1189 }
1190
1191 r = setrlimit_closest(RLIMIT_MEMLOCK, &new_rlimit);
1192 if (r < 0)
1193 return log_warning_errno(r, "Setting RLIMIT_MEMLOCK failed, ignoring: %m");
1194
1195 return 0;
1196 }
1197
1198 static void test_usr(void) {
1199
1200 /* Check that /usr is either on the same file system as / or mounted already. */
1201
1202 if (dir_is_empty("/usr") <= 0)
1203 return;
1204
1205 log_warning("/usr appears to be on its own filesystem and is not already mounted. This is not a supported setup. "
1206 "Some things will probably break (sometimes even silently) in mysterious ways. "
1207 "Consult http://freedesktop.org/wiki/Software/systemd/separate-usr-is-broken for more information.");
1208 }
1209
1210 static int enforce_syscall_archs(Set *archs) {
1211 #if HAVE_SECCOMP
1212 int r;
1213
1214 if (!is_seccomp_available())
1215 return 0;
1216
1217 r = seccomp_restrict_archs(arg_syscall_archs);
1218 if (r < 0)
1219 return log_error_errno(r, "Failed to enforce system call architecture restrication: %m");
1220 #endif
1221 return 0;
1222 }
1223
1224 static int status_welcome(void) {
1225 _cleanup_free_ char *pretty_name = NULL, *ansi_color = NULL;
1226 int r;
1227
1228 if (IN_SET(arg_show_status, SHOW_STATUS_NO, SHOW_STATUS_AUTO))
1229 return 0;
1230
1231 r = parse_os_release(NULL,
1232 "PRETTY_NAME", &pretty_name,
1233 "ANSI_COLOR", &ansi_color,
1234 NULL);
1235 if (r < 0)
1236 log_full_errno(r == -ENOENT ? LOG_DEBUG : LOG_WARNING, r,
1237 "Failed to read os-release file, ignoring: %m");
1238
1239 if (log_get_show_color())
1240 return status_printf(NULL, 0,
1241 "\nWelcome to \x1B[%sm%s\x1B[0m!\n",
1242 isempty(ansi_color) ? "1" : ansi_color,
1243 isempty(pretty_name) ? "Linux" : pretty_name);
1244 else
1245 return status_printf(NULL, 0,
1246 "\nWelcome to %s!\n",
1247 isempty(pretty_name) ? "Linux" : pretty_name);
1248 }
1249
1250 static int write_container_id(void) {
1251 const char *c;
1252 int r;
1253
1254 c = getenv("container");
1255 if (isempty(c))
1256 return 0;
1257
1258 RUN_WITH_UMASK(0022)
1259 r = write_string_file("/run/systemd/container", c, WRITE_STRING_FILE_CREATE);
1260 if (r < 0)
1261 return log_warning_errno(r, "Failed to write /run/systemd/container, ignoring: %m");
1262
1263 return 1;
1264 }
1265
1266 static int bump_unix_max_dgram_qlen(void) {
1267 _cleanup_free_ char *qlen = NULL;
1268 unsigned long v;
1269 int r;
1270
1271 /* Let's bump the net.unix.max_dgram_qlen sysctl. The kernel default of 16 is simply too low. We set the value
1272 * really really early during boot, so that it is actually applied to all our sockets, including the
1273 * $NOTIFY_SOCKET one. */
1274
1275 r = read_one_line_file("/proc/sys/net/unix/max_dgram_qlen", &qlen);
1276 if (r < 0)
1277 return log_full_errno(r == -ENOENT ? LOG_DEBUG : LOG_WARNING, r, "Failed to read AF_UNIX datagram queue length, ignoring: %m");
1278
1279 r = safe_atolu(qlen, &v);
1280 if (r < 0)
1281 return log_warning_errno(r, "Failed to parse AF_UNIX datagram queue length '%s', ignoring: %m", qlen);
1282
1283 if (v >= DEFAULT_UNIX_MAX_DGRAM_QLEN)
1284 return 0;
1285
1286 r = write_string_filef("/proc/sys/net/unix/max_dgram_qlen", WRITE_STRING_FILE_DISABLE_BUFFER, "%lu", DEFAULT_UNIX_MAX_DGRAM_QLEN);
1287 if (r < 0)
1288 return log_full_errno(IN_SET(r, -EROFS, -EPERM, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
1289 "Failed to bump AF_UNIX datagram queue length, ignoring: %m");
1290
1291 return 1;
1292 }
1293
1294 static int fixup_environment(void) {
1295 _cleanup_free_ char *term = NULL;
1296 const char *t;
1297 int r;
1298
1299 /* Only fix up the environment when we are started as PID 1 */
1300 if (getpid_cached() != 1)
1301 return 0;
1302
1303 /* We expect the environment to be set correctly if run inside a container. */
1304 if (detect_container() > 0)
1305 return 0;
1306
1307 /* When started as PID1, the kernel uses /dev/console for our stdios and uses TERM=linux whatever the backend
1308 * device used by the console. We try to make a better guess here since some consoles might not have support
1309 * for color mode for example.
1310 *
1311 * However if TERM was configured through the kernel command line then leave it alone. */
1312 r = proc_cmdline_get_key("TERM", 0, &term);
1313 if (r < 0)
1314 return r;
1315
1316 t = term ?: default_term_for_tty("/dev/console");
1317
1318 if (setenv("TERM", t, 1) < 0)
1319 return -errno;
1320
1321 /* The kernels sets HOME=/ for init. Let's undo this. */
1322 if (path_equal_ptr(getenv("HOME"), "/") &&
1323 unsetenv("HOME") < 0)
1324 log_warning_errno(errno, "Failed to unset $HOME: %m");
1325
1326 return 0;
1327 }
1328
1329 static void redirect_telinit(int argc, char *argv[]) {
1330
1331 /* This is compatibility support for SysV, where calling init as a user is identical to telinit. */
1332
1333 #if HAVE_SYSV_COMPAT
1334 if (getpid_cached() == 1)
1335 return;
1336
1337 if (!strstr(program_invocation_short_name, "init"))
1338 return;
1339
1340 execv(SYSTEMCTL_BINARY_PATH, argv);
1341 log_error_errno(errno, "Failed to exec " SYSTEMCTL_BINARY_PATH ": %m");
1342 exit(EXIT_FAILURE);
1343 #endif
1344 }
1345
1346 static int become_shutdown(
1347 const char *shutdown_verb,
1348 int retval) {
1349
1350 char log_level[DECIMAL_STR_MAX(int) + 1],
1351 exit_code[DECIMAL_STR_MAX(uint8_t) + 1],
1352 timeout[DECIMAL_STR_MAX(usec_t) + 1];
1353
1354 const char* command_line[13] = {
1355 SYSTEMD_SHUTDOWN_BINARY_PATH,
1356 shutdown_verb,
1357 "--timeout", timeout,
1358 "--log-level", log_level,
1359 "--log-target",
1360 };
1361
1362 _cleanup_strv_free_ char **env_block = NULL;
1363 size_t pos = 7;
1364 int r;
1365 usec_t watchdog_timer = 0;
1366
1367 assert(shutdown_verb);
1368 assert(!command_line[pos]);
1369 env_block = strv_copy(environ);
1370
1371 xsprintf(log_level, "%d", log_get_max_level());
1372 xsprintf(timeout, "%" PRI_USEC "us", arg_default_timeout_stop_usec);
1373
1374 switch (log_get_target()) {
1375
1376 case LOG_TARGET_KMSG:
1377 case LOG_TARGET_JOURNAL_OR_KMSG:
1378 case LOG_TARGET_SYSLOG_OR_KMSG:
1379 command_line[pos++] = "kmsg";
1380 break;
1381
1382 case LOG_TARGET_NULL:
1383 command_line[pos++] = "null";
1384 break;
1385
1386 case LOG_TARGET_CONSOLE:
1387 default:
1388 command_line[pos++] = "console";
1389 break;
1390 };
1391
1392 if (log_get_show_color())
1393 command_line[pos++] = "--log-color";
1394
1395 if (log_get_show_location())
1396 command_line[pos++] = "--log-location";
1397
1398 if (streq(shutdown_verb, "exit")) {
1399 command_line[pos++] = "--exit-code";
1400 command_line[pos++] = exit_code;
1401 xsprintf(exit_code, "%d", retval);
1402 }
1403
1404 assert(pos < ELEMENTSOF(command_line));
1405
1406 if (streq(shutdown_verb, "reboot"))
1407 watchdog_timer = arg_reboot_watchdog;
1408 else if (streq(shutdown_verb, "kexec"))
1409 watchdog_timer = arg_kexec_watchdog;
1410
1411 if (watchdog_timer > 0 && watchdog_timer != USEC_INFINITY) {
1412
1413 char *e;
1414
1415 /* If we reboot or kexec let's set the shutdown
1416 * watchdog and tell the shutdown binary to
1417 * repeatedly ping it */
1418 r = watchdog_set_timeout(&watchdog_timer);
1419 watchdog_close(r < 0);
1420
1421 /* Tell the binary how often to ping, ignore failure */
1422 if (asprintf(&e, "WATCHDOG_USEC="USEC_FMT, watchdog_timer) > 0)
1423 (void) strv_consume(&env_block, e);
1424
1425 if (arg_watchdog_device &&
1426 asprintf(&e, "WATCHDOG_DEVICE=%s", arg_watchdog_device) > 0)
1427 (void) strv_consume(&env_block, e);
1428 } else
1429 watchdog_close(true);
1430
1431 /* Avoid the creation of new processes forked by the
1432 * kernel; at this point, we will not listen to the
1433 * signals anyway */
1434 if (detect_container() <= 0)
1435 (void) cg_uninstall_release_agent(SYSTEMD_CGROUP_CONTROLLER);
1436
1437 execve(SYSTEMD_SHUTDOWN_BINARY_PATH, (char **) command_line, env_block);
1438 return -errno;
1439 }
1440
1441 static void initialize_clock(void) {
1442 int r;
1443
1444 if (clock_is_localtime(NULL) > 0) {
1445 int min;
1446
1447 /*
1448 * The very first call of settimeofday() also does a time warp in the kernel.
1449 *
1450 * In the rtc-in-local time mode, we set the kernel's timezone, and rely on external tools to take care
1451 * of maintaining the RTC and do all adjustments. This matches the behavior of Windows, which leaves
1452 * the RTC alone if the registry tells that the RTC runs in UTC.
1453 */
1454 r = clock_set_timezone(&min);
1455 if (r < 0)
1456 log_error_errno(r, "Failed to apply local time delta, ignoring: %m");
1457 else
1458 log_info("RTC configured in localtime, applying delta of %i minutes to system time.", min);
1459
1460 } else if (!in_initrd()) {
1461 /*
1462 * Do a dummy very first call to seal the kernel's time warp magic.
1463 *
1464 * Do not call this from inside the initrd. The initrd might not carry /etc/adjtime with LOCAL, but the
1465 * real system could be set up that way. In such case, we need to delay the time-warp or the sealing
1466 * until we reach the real system.
1467 *
1468 * Do no set the kernel's timezone. The concept of local time cannot be supported reliably, the time
1469 * will jump or be incorrect at every daylight saving time change. All kernel local time concepts will
1470 * be treated as UTC that way.
1471 */
1472 (void) clock_reset_timewarp();
1473 }
1474
1475 r = clock_apply_epoch();
1476 if (r < 0)
1477 log_error_errno(r, "Current system time is before build time, but cannot correct: %m");
1478 else if (r > 0)
1479 log_info("System time before build time, advancing clock.");
1480 }
1481
1482 static void initialize_coredump(bool skip_setup) {
1483 #if ENABLE_COREDUMP
1484 if (getpid_cached() != 1)
1485 return;
1486
1487 /* Don't limit the core dump size, so that coredump handlers such as systemd-coredump (which honour the limit)
1488 * will process core dumps for system services by default. */
1489 if (setrlimit(RLIMIT_CORE, &RLIMIT_MAKE_CONST(RLIM_INFINITY)) < 0)
1490 log_warning_errno(errno, "Failed to set RLIMIT_CORE: %m");
1491
1492 /* But at the same time, turn off the core_pattern logic by default, so that no
1493 * coredumps are stored until the systemd-coredump tool is enabled via
1494 * sysctl. However it can be changed via the kernel command line later so core
1495 * dumps can still be generated during early startup and in initramfs. */
1496 if (!skip_setup)
1497 disable_coredumps();
1498 #endif
1499 }
1500
1501 static void initialize_core_pattern(bool skip_setup) {
1502 int r;
1503
1504 if (skip_setup || !arg_early_core_pattern)
1505 return;
1506
1507 if (getpid_cached() != 1)
1508 return;
1509
1510 r = write_string_file("/proc/sys/kernel/core_pattern", arg_early_core_pattern, WRITE_STRING_FILE_DISABLE_BUFFER);
1511 if (r < 0)
1512 log_warning_errno(r, "Failed to write '%s' to /proc/sys/kernel/core_pattern, ignoring: %m", arg_early_core_pattern);
1513 }
1514
1515 static void update_cpu_affinity(bool skip_setup) {
1516 _cleanup_free_ char *mask = NULL;
1517
1518 if (skip_setup || !arg_cpu_affinity.set)
1519 return;
1520
1521 assert(arg_cpu_affinity.allocated > 0);
1522
1523 mask = cpu_set_to_string(&arg_cpu_affinity);
1524 log_debug("Setting CPU affinity to %s.", strnull(mask));
1525
1526 if (sched_setaffinity(0, arg_cpu_affinity.allocated, arg_cpu_affinity.set) < 0)
1527 log_warning_errno(errno, "Failed to set CPU affinity: %m");
1528 }
1529
1530 static void update_numa_policy(bool skip_setup) {
1531 int r;
1532 _cleanup_free_ char *nodes = NULL;
1533 const char * policy = NULL;
1534
1535 if (skip_setup || !mpol_is_valid(numa_policy_get_type(&arg_numa_policy)))
1536 return;
1537
1538 if (DEBUG_LOGGING) {
1539 policy = mpol_to_string(numa_policy_get_type(&arg_numa_policy));
1540 nodes = cpu_set_to_range_string(&arg_numa_policy.nodes);
1541 log_debug("Setting NUMA policy to %s, with nodes %s.", strnull(policy), strnull(nodes));
1542 }
1543
1544 r = apply_numa_policy(&arg_numa_policy);
1545 if (r == -EOPNOTSUPP)
1546 log_debug_errno(r, "NUMA support not available, ignoring.");
1547 else if (r < 0)
1548 log_warning_errno(r, "Failed to set NUMA memory policy: %m");
1549 }
1550
1551 static void do_reexecute(
1552 int argc,
1553 char *argv[],
1554 const struct rlimit *saved_rlimit_nofile,
1555 const struct rlimit *saved_rlimit_memlock,
1556 FDSet *fds,
1557 const char *switch_root_dir,
1558 const char *switch_root_init,
1559 const char **ret_error_message) {
1560
1561 unsigned i, j, args_size;
1562 const char **args;
1563 int r;
1564
1565 assert(saved_rlimit_nofile);
1566 assert(saved_rlimit_memlock);
1567 assert(ret_error_message);
1568
1569 /* Close and disarm the watchdog, so that the new instance can reinitialize it, but doesn't get rebooted while
1570 * we do that */
1571 watchdog_close(true);
1572
1573 /* Reset RLIMIT_NOFILE + RLIMIT_MEMLOCK back to the kernel defaults, so that the new systemd can pass
1574 * the kernel default to its child processes */
1575 if (saved_rlimit_nofile->rlim_cur != 0)
1576 (void) setrlimit(RLIMIT_NOFILE, saved_rlimit_nofile);
1577 if (saved_rlimit_memlock->rlim_cur != RLIM_INFINITY)
1578 (void) setrlimit(RLIMIT_MEMLOCK, saved_rlimit_memlock);
1579
1580 if (switch_root_dir) {
1581 /* Kill all remaining processes from the initrd, but don't wait for them, so that we can handle the
1582 * SIGCHLD for them after deserializing. */
1583 broadcast_signal(SIGTERM, false, true, arg_default_timeout_stop_usec);
1584
1585 /* And switch root with MS_MOVE, because we remove the old directory afterwards and detach it. */
1586 r = switch_root(switch_root_dir, "/mnt", true, MS_MOVE);
1587 if (r < 0)
1588 log_error_errno(r, "Failed to switch root, trying to continue: %m");
1589 }
1590
1591 args_size = MAX(6, argc+1);
1592 args = newa(const char*, args_size);
1593
1594 if (!switch_root_init) {
1595 char sfd[DECIMAL_STR_MAX(int) + 1];
1596
1597 /* First try to spawn ourselves with the right path, and with full serialization. We do this only if
1598 * the user didn't specify an explicit init to spawn. */
1599
1600 assert(arg_serialization);
1601 assert(fds);
1602
1603 xsprintf(sfd, "%i", fileno(arg_serialization));
1604
1605 i = 0;
1606 args[i++] = SYSTEMD_BINARY_PATH;
1607 if (switch_root_dir)
1608 args[i++] = "--switched-root";
1609 args[i++] = arg_system ? "--system" : "--user";
1610 args[i++] = "--deserialize";
1611 args[i++] = sfd;
1612 args[i++] = NULL;
1613
1614 assert(i <= args_size);
1615
1616 /*
1617 * We want valgrind to print its memory usage summary before reexecution. Valgrind won't do this is on
1618 * its own on exec(), but it will do it on exit(). Hence, to ensure we get a summary here, fork() off
1619 * a child, let it exit() cleanly, so that it prints the summary, and wait() for it in the parent,
1620 * before proceeding into the exec().
1621 */
1622 valgrind_summary_hack();
1623
1624 (void) execv(args[0], (char* const*) args);
1625 log_debug_errno(errno, "Failed to execute our own binary, trying fallback: %m");
1626 }
1627
1628 /* Try the fallback, if there is any, without any serialization. We pass the original argv[] and envp[]. (Well,
1629 * modulo the ordering changes due to getopt() in argv[], and some cleanups in envp[], but let's hope that
1630 * doesn't matter.) */
1631
1632 arg_serialization = safe_fclose(arg_serialization);
1633 fds = fdset_free(fds);
1634
1635 /* Reopen the console */
1636 (void) make_console_stdio();
1637
1638 for (j = 1, i = 1; j < (unsigned) argc; j++)
1639 args[i++] = argv[j];
1640 args[i++] = NULL;
1641 assert(i <= args_size);
1642
1643 /* Re-enable any blocked signals, especially important if we switch from initial ramdisk to init=... */
1644 (void) reset_all_signal_handlers();
1645 (void) reset_signal_mask();
1646 (void) rlimit_nofile_safe();
1647
1648 if (switch_root_init) {
1649 args[0] = switch_root_init;
1650 (void) execve(args[0], (char* const*) args, saved_env);
1651 log_warning_errno(errno, "Failed to execute configured init, trying fallback: %m");
1652 }
1653
1654 args[0] = "/sbin/init";
1655 (void) execv(args[0], (char* const*) args);
1656 r = -errno;
1657
1658 manager_status_printf(NULL, STATUS_TYPE_EMERGENCY,
1659 ANSI_HIGHLIGHT_RED " !! " ANSI_NORMAL,
1660 "Failed to execute /sbin/init");
1661
1662 if (r == -ENOENT) {
1663 log_warning("No /sbin/init, trying fallback");
1664
1665 args[0] = "/bin/sh";
1666 args[1] = NULL;
1667 (void) execve(args[0], (char* const*) args, saved_env);
1668 log_error_errno(errno, "Failed to execute /bin/sh, giving up: %m");
1669 } else
1670 log_warning_errno(r, "Failed to execute /sbin/init, giving up: %m");
1671
1672 *ret_error_message = "Failed to execute fallback shell";
1673 }
1674
1675 static int invoke_main_loop(
1676 Manager *m,
1677 const struct rlimit *saved_rlimit_nofile,
1678 const struct rlimit *saved_rlimit_memlock,
1679 bool *ret_reexecute,
1680 int *ret_retval, /* Return parameters relevant for shutting down */
1681 const char **ret_shutdown_verb, /* … */
1682 FDSet **ret_fds, /* Return parameters for reexecuting */
1683 char **ret_switch_root_dir, /* … */
1684 char **ret_switch_root_init, /* … */
1685 const char **ret_error_message) {
1686
1687 int r;
1688
1689 assert(m);
1690 assert(saved_rlimit_nofile);
1691 assert(saved_rlimit_memlock);
1692 assert(ret_reexecute);
1693 assert(ret_retval);
1694 assert(ret_shutdown_verb);
1695 assert(ret_fds);
1696 assert(ret_switch_root_dir);
1697 assert(ret_switch_root_init);
1698 assert(ret_error_message);
1699
1700 for (;;) {
1701 r = manager_loop(m);
1702 if (r < 0) {
1703 *ret_error_message = "Failed to run main loop";
1704 return log_emergency_errno(r, "Failed to run main loop: %m");
1705 }
1706
1707 switch ((ManagerObjective) r) {
1708
1709 case MANAGER_RELOAD: {
1710 LogTarget saved_log_target;
1711 int saved_log_level;
1712
1713 log_info("Reloading.");
1714
1715 /* First, save any overridden log level/target, then parse the configuration file, which might
1716 * change the log level to new settings. */
1717
1718 saved_log_level = m->log_level_overridden ? log_get_max_level() : -1;
1719 saved_log_target = m->log_target_overridden ? log_get_target() : _LOG_TARGET_INVALID;
1720
1721 (void) parse_configuration(saved_rlimit_nofile, saved_rlimit_memlock);
1722
1723 set_manager_defaults(m);
1724
1725 update_cpu_affinity(false);
1726 update_numa_policy(false);
1727
1728 if (saved_log_level >= 0)
1729 manager_override_log_level(m, saved_log_level);
1730 if (saved_log_target >= 0)
1731 manager_override_log_target(m, saved_log_target);
1732
1733 r = manager_reload(m);
1734 if (r < 0)
1735 /* Reloading failed before the point of no return. Let's continue running as if nothing happened. */
1736 m->objective = MANAGER_OK;
1737
1738 break;
1739 }
1740
1741 case MANAGER_REEXECUTE:
1742
1743 r = prepare_reexecute(m, &arg_serialization, ret_fds, false);
1744 if (r < 0) {
1745 *ret_error_message = "Failed to prepare for reexecution";
1746 return r;
1747 }
1748
1749 log_notice("Reexecuting.");
1750
1751 *ret_reexecute = true;
1752 *ret_retval = EXIT_SUCCESS;
1753 *ret_shutdown_verb = NULL;
1754 *ret_switch_root_dir = *ret_switch_root_init = NULL;
1755
1756 return 0;
1757
1758 case MANAGER_SWITCH_ROOT:
1759 if (!m->switch_root_init) {
1760 r = prepare_reexecute(m, &arg_serialization, ret_fds, true);
1761 if (r < 0) {
1762 *ret_error_message = "Failed to prepare for reexecution";
1763 return r;
1764 }
1765 } else
1766 *ret_fds = NULL;
1767
1768 log_notice("Switching root.");
1769
1770 *ret_reexecute = true;
1771 *ret_retval = EXIT_SUCCESS;
1772 *ret_shutdown_verb = NULL;
1773
1774 /* Steal the switch root parameters */
1775 *ret_switch_root_dir = TAKE_PTR(m->switch_root);
1776 *ret_switch_root_init = TAKE_PTR(m->switch_root_init);
1777
1778 return 0;
1779
1780 case MANAGER_EXIT:
1781
1782 if (MANAGER_IS_USER(m)) {
1783 log_debug("Exit.");
1784
1785 *ret_reexecute = false;
1786 *ret_retval = m->return_value;
1787 *ret_shutdown_verb = NULL;
1788 *ret_fds = NULL;
1789 *ret_switch_root_dir = *ret_switch_root_init = NULL;
1790
1791 return 0;
1792 }
1793
1794 _fallthrough_;
1795 case MANAGER_REBOOT:
1796 case MANAGER_POWEROFF:
1797 case MANAGER_HALT:
1798 case MANAGER_KEXEC: {
1799 static const char * const table[_MANAGER_OBJECTIVE_MAX] = {
1800 [MANAGER_EXIT] = "exit",
1801 [MANAGER_REBOOT] = "reboot",
1802 [MANAGER_POWEROFF] = "poweroff",
1803 [MANAGER_HALT] = "halt",
1804 [MANAGER_KEXEC] = "kexec",
1805 };
1806
1807 log_notice("Shutting down.");
1808
1809 *ret_reexecute = false;
1810 *ret_retval = m->return_value;
1811 assert_se(*ret_shutdown_verb = table[m->objective]);
1812 *ret_fds = NULL;
1813 *ret_switch_root_dir = *ret_switch_root_init = NULL;
1814
1815 return 0;
1816 }
1817
1818 default:
1819 assert_not_reached("Unknown or unexpected manager objective.");
1820 }
1821 }
1822 }
1823
1824 static void log_execution_mode(bool *ret_first_boot) {
1825 assert(ret_first_boot);
1826
1827 if (arg_system) {
1828 int v;
1829
1830 log_info("systemd " GIT_VERSION " running in %ssystem mode. (" SYSTEMD_FEATURES ")",
1831 arg_action == ACTION_TEST ? "test " : "" );
1832
1833 v = detect_virtualization();
1834 if (v > 0)
1835 log_info("Detected virtualization %s.", virtualization_to_string(v));
1836
1837 log_info("Detected architecture %s.", architecture_to_string(uname_architecture()));
1838
1839 if (in_initrd()) {
1840 *ret_first_boot = false;
1841 log_info("Running in initial RAM disk.");
1842 } else {
1843 /* Let's check whether we are in first boot, i.e. whether /etc is still unpopulated. We use
1844 * /etc/machine-id as flag file, for this: if it exists we assume /etc is populated, if it
1845 * doesn't it's unpopulated. This allows container managers and installers to provision a
1846 * couple of files already. If the container manager wants to provision the machine ID itself
1847 * it should pass $container_uuid to PID 1. */
1848
1849 *ret_first_boot = access("/etc/machine-id", F_OK) < 0;
1850 if (*ret_first_boot)
1851 log_info("Running with unpopulated /etc.");
1852 }
1853 } else {
1854 if (DEBUG_LOGGING) {
1855 _cleanup_free_ char *t;
1856
1857 t = uid_to_name(getuid());
1858 log_debug("systemd " GIT_VERSION " running in %suser mode for user " UID_FMT "/%s. (" SYSTEMD_FEATURES ")",
1859 arg_action == ACTION_TEST ? " test" : "", getuid(), strna(t));
1860 }
1861
1862 *ret_first_boot = false;
1863 }
1864 }
1865
1866 static int initialize_runtime(
1867 bool skip_setup,
1868 struct rlimit *saved_rlimit_nofile,
1869 struct rlimit *saved_rlimit_memlock,
1870 const char **ret_error_message) {
1871 int r;
1872
1873 assert(ret_error_message);
1874
1875 /* Sets up various runtime parameters. Many of these initializations are conditionalized:
1876 *
1877 * - Some only apply to --system instances
1878 * - Some only apply to --user instances
1879 * - Some only apply when we first start up, but not when we reexecute
1880 */
1881
1882 if (arg_action != ACTION_RUN)
1883 return 0;
1884
1885 update_cpu_affinity(skip_setup);
1886 update_numa_policy(skip_setup);
1887
1888 if (arg_system) {
1889 /* Make sure we leave a core dump without panicking the kernel. */
1890 install_crash_handler();
1891
1892 if (!skip_setup) {
1893 r = mount_cgroup_controllers();
1894 if (r < 0) {
1895 *ret_error_message = "Failed to mount cgroup hierarchies";
1896 return r;
1897 }
1898
1899 status_welcome();
1900 hostname_setup();
1901 machine_id_setup(NULL, arg_machine_id, NULL);
1902 loopback_setup();
1903 bump_unix_max_dgram_qlen();
1904 bump_file_max_and_nr_open();
1905 test_usr();
1906 write_container_id();
1907 }
1908
1909 if (arg_watchdog_device) {
1910 r = watchdog_set_device(arg_watchdog_device);
1911 if (r < 0)
1912 log_warning_errno(r, "Failed to set watchdog device to %s, ignoring: %m", arg_watchdog_device);
1913 }
1914
1915 if (timestamp_is_set(arg_runtime_watchdog))
1916 watchdog_set_timeout(&arg_runtime_watchdog);
1917 }
1918
1919 if (arg_timer_slack_nsec != NSEC_INFINITY)
1920 if (prctl(PR_SET_TIMERSLACK, arg_timer_slack_nsec) < 0)
1921 log_warning_errno(errno, "Failed to adjust timer slack, ignoring: %m");
1922
1923 if (arg_system && !cap_test_all(arg_capability_bounding_set)) {
1924 r = capability_bounding_set_drop_usermode(arg_capability_bounding_set);
1925 if (r < 0) {
1926 *ret_error_message = "Failed to drop capability bounding set of usermode helpers";
1927 return log_emergency_errno(r, "Failed to drop capability bounding set of usermode helpers: %m");
1928 }
1929
1930 r = capability_bounding_set_drop(arg_capability_bounding_set, true);
1931 if (r < 0) {
1932 *ret_error_message = "Failed to drop capability bounding set";
1933 return log_emergency_errno(r, "Failed to drop capability bounding set: %m");
1934 }
1935 }
1936
1937 if (arg_system && arg_no_new_privs) {
1938 if (prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0) < 0) {
1939 *ret_error_message = "Failed to disable new privileges";
1940 return log_emergency_errno(errno, "Failed to disable new privileges: %m");
1941 }
1942 }
1943
1944 if (arg_syscall_archs) {
1945 r = enforce_syscall_archs(arg_syscall_archs);
1946 if (r < 0) {
1947 *ret_error_message = "Failed to set syscall architectures";
1948 return r;
1949 }
1950 }
1951
1952 if (!arg_system)
1953 /* Become reaper of our children */
1954 if (prctl(PR_SET_CHILD_SUBREAPER, 1) < 0)
1955 log_warning_errno(errno, "Failed to make us a subreaper: %m");
1956
1957 /* Bump up RLIMIT_NOFILE for systemd itself */
1958 (void) bump_rlimit_nofile(saved_rlimit_nofile);
1959 (void) bump_rlimit_memlock(saved_rlimit_memlock);
1960
1961 return 0;
1962 }
1963
1964 static int do_queue_default_job(
1965 Manager *m,
1966 const char **ret_error_message) {
1967
1968 _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL;
1969 Job *default_unit_job;
1970 Unit *target = NULL;
1971 int r;
1972
1973 log_debug("Activating default unit: %s", arg_default_unit);
1974
1975 r = manager_load_startable_unit_or_warn(m, arg_default_unit, NULL, &target);
1976 if (r < 0) {
1977 log_info("Falling back to rescue target: " SPECIAL_RESCUE_TARGET);
1978
1979 r = manager_load_startable_unit_or_warn(m, SPECIAL_RESCUE_TARGET, NULL, &target);
1980 if (r < 0) {
1981 *ret_error_message = r == -ERFKILL ? "Rescue target masked"
1982 : "Failed to load rescue target";
1983 return r;
1984 }
1985 }
1986
1987 assert(target->load_state == UNIT_LOADED);
1988
1989 r = manager_add_job(m, JOB_START, target, JOB_ISOLATE, NULL, &error, &default_unit_job);
1990 if (r == -EPERM) {
1991 log_debug_errno(r, "Default target could not be isolated, starting instead: %s", bus_error_message(&error, r));
1992
1993 sd_bus_error_free(&error);
1994
1995 r = manager_add_job(m, JOB_START, target, JOB_REPLACE, NULL, &error, &default_unit_job);
1996 if (r < 0) {
1997 *ret_error_message = "Failed to start default target";
1998 return log_emergency_errno(r, "Failed to start default target: %s", bus_error_message(&error, r));
1999 }
2000
2001 } else if (r < 0) {
2002 *ret_error_message = "Failed to isolate default target";
2003 return log_emergency_errno(r, "Failed to isolate default target: %s", bus_error_message(&error, r));
2004 }
2005
2006 m->default_unit_job_id = default_unit_job->id;
2007
2008 return 0;
2009 }
2010
2011 static void save_rlimits(struct rlimit *saved_rlimit_nofile,
2012 struct rlimit *saved_rlimit_memlock) {
2013
2014 assert(saved_rlimit_nofile);
2015 assert(saved_rlimit_memlock);
2016
2017 if (getrlimit(RLIMIT_NOFILE, saved_rlimit_nofile) < 0)
2018 log_warning_errno(errno, "Reading RLIMIT_NOFILE failed, ignoring: %m");
2019
2020 if (getrlimit(RLIMIT_MEMLOCK, saved_rlimit_memlock) < 0)
2021 log_warning_errno(errno, "Reading RLIMIT_MEMLOCK failed, ignoring: %m");
2022 }
2023
2024 static void fallback_rlimit_nofile(const struct rlimit *saved_rlimit_nofile) {
2025 struct rlimit *rl;
2026
2027 if (arg_default_rlimit[RLIMIT_NOFILE])
2028 return;
2029
2030 /* Make sure forked processes get limits based on the original kernel setting */
2031
2032 rl = newdup(struct rlimit, saved_rlimit_nofile, 1);
2033 if (!rl) {
2034 log_oom();
2035 return;
2036 }
2037
2038 /* Bump the hard limit for system services to a substantially higher value. The default
2039 * hard limit current kernels set is pretty low (4K), mostly for historical
2040 * reasons. According to kernel developers, the fd handling in recent kernels has been
2041 * optimized substantially enough, so that we can bump the limit now, without paying too
2042 * high a price in memory or performance. Note however that we only bump the hard limit,
2043 * not the soft limit. That's because select() works the way it works, and chokes on fds
2044 * >= 1024. If we'd bump the soft limit globally, it might accidentally happen to
2045 * unexpecting programs that they get fds higher than what they can process using
2046 * select(). By only bumping the hard limit but leaving the low limit as it is we avoid
2047 * this pitfall: programs that are written by folks aware of the select() problem in mind
2048 * (and thus use poll()/epoll instead of select(), the way everybody should) can
2049 * explicitly opt into high fds by bumping their soft limit beyond 1024, to the hard limit
2050 * we pass. */
2051 if (arg_system) {
2052 int nr;
2053
2054 /* Get the underlying absolute limit the kernel enforces */
2055 nr = read_nr_open();
2056
2057 rl->rlim_max = MIN((rlim_t) nr, MAX(rl->rlim_max, (rlim_t) HIGH_RLIMIT_NOFILE));
2058 }
2059
2060 /* If for some reason we were invoked with a soft limit above 1024 (which should never
2061 * happen!, but who knows what we get passed in from pam_limit when invoked as --user
2062 * instance), then lower what we pass on to not confuse our children */
2063 rl->rlim_cur = MIN(rl->rlim_cur, (rlim_t) FD_SETSIZE);
2064
2065 arg_default_rlimit[RLIMIT_NOFILE] = rl;
2066 }
2067
2068 static void fallback_rlimit_memlock(const struct rlimit *saved_rlimit_memlock) {
2069 struct rlimit *rl;
2070
2071 /* Pass the original value down to invoked processes */
2072
2073 if (arg_default_rlimit[RLIMIT_MEMLOCK])
2074 return;
2075
2076 rl = newdup(struct rlimit, saved_rlimit_memlock, 1);
2077 if (!rl) {
2078 log_oom();
2079 return;
2080 }
2081
2082 arg_default_rlimit[RLIMIT_MEMLOCK] = rl;
2083 }
2084
2085 static void reset_arguments(void) {
2086 /* Frees/resets arg_* variables, with a few exceptions commented below. */
2087
2088 arg_default_unit = mfree(arg_default_unit);
2089
2090 /* arg_system — ignore */
2091
2092 arg_dump_core = true;
2093 arg_crash_chvt = -1;
2094 arg_crash_shell = false;
2095 arg_crash_reboot = false;
2096 arg_confirm_spawn = mfree(arg_confirm_spawn);
2097 arg_show_status = _SHOW_STATUS_INVALID;
2098 arg_status_unit_format = STATUS_UNIT_FORMAT_DEFAULT;
2099 arg_switched_root = false;
2100 arg_pager_flags = 0;
2101 arg_service_watchdogs = true;
2102 arg_default_std_output = EXEC_OUTPUT_JOURNAL;
2103 arg_default_std_error = EXEC_OUTPUT_INHERIT;
2104 arg_default_restart_usec = DEFAULT_RESTART_USEC;
2105 arg_default_timeout_start_usec = DEFAULT_TIMEOUT_USEC;
2106 arg_default_timeout_stop_usec = DEFAULT_TIMEOUT_USEC;
2107 arg_default_timeout_abort_usec = DEFAULT_TIMEOUT_USEC;
2108 arg_default_timeout_abort_set = false;
2109 arg_default_start_limit_interval = DEFAULT_START_LIMIT_INTERVAL;
2110 arg_default_start_limit_burst = DEFAULT_START_LIMIT_BURST;
2111 arg_runtime_watchdog = 0;
2112 arg_reboot_watchdog = 10 * USEC_PER_MINUTE;
2113 arg_kexec_watchdog = 0;
2114 arg_early_core_pattern = NULL;
2115 arg_watchdog_device = NULL;
2116
2117 arg_default_environment = strv_free(arg_default_environment);
2118 rlimit_free_all(arg_default_rlimit);
2119
2120 arg_capability_bounding_set = CAP_ALL;
2121 arg_no_new_privs = false;
2122 arg_timer_slack_nsec = NSEC_INFINITY;
2123 arg_default_timer_accuracy_usec = 1 * USEC_PER_MINUTE;
2124
2125 arg_syscall_archs = set_free(arg_syscall_archs);
2126
2127 /* arg_serialization — ignore */
2128
2129 arg_default_cpu_accounting = -1;
2130 arg_default_io_accounting = false;
2131 arg_default_ip_accounting = false;
2132 arg_default_blockio_accounting = false;
2133 arg_default_memory_accounting = MEMORY_ACCOUNTING_DEFAULT;
2134 arg_default_tasks_accounting = true;
2135 arg_default_tasks_max = system_tasks_max_scale(DEFAULT_TASKS_MAX_PERCENTAGE, 100U);
2136 arg_machine_id = (sd_id128_t) {};
2137 arg_cad_burst_action = EMERGENCY_ACTION_REBOOT_FORCE;
2138 arg_default_oom_policy = OOM_STOP;
2139
2140 cpu_set_reset(&arg_cpu_affinity);
2141 numa_policy_reset(&arg_numa_policy);
2142 }
2143
2144 static int parse_configuration(const struct rlimit *saved_rlimit_nofile,
2145 const struct rlimit *saved_rlimit_memlock) {
2146 int r;
2147
2148 assert(saved_rlimit_nofile);
2149 assert(saved_rlimit_memlock);
2150
2151 /* Assign configuration defaults */
2152 reset_arguments();
2153
2154 r = parse_config_file();
2155 if (r < 0)
2156 log_warning_errno(r, "Failed to parse config file, ignoring: %m");
2157
2158 if (arg_system) {
2159 r = proc_cmdline_parse(parse_proc_cmdline_item, NULL, 0);
2160 if (r < 0)
2161 log_warning_errno(r, "Failed to parse kernel command line, ignoring: %m");
2162 }
2163
2164 /* Initialize some default rlimits for services if they haven't been configured */
2165 fallback_rlimit_nofile(saved_rlimit_nofile);
2166 fallback_rlimit_memlock(saved_rlimit_memlock);
2167
2168 /* Note that this also parses bits from the kernel command line, including "debug". */
2169 log_parse_environment();
2170
2171 return 0;
2172 }
2173
2174 static int load_configuration(
2175 int argc,
2176 char **argv,
2177 const struct rlimit *saved_rlimit_nofile,
2178 const struct rlimit *saved_rlimit_memlock,
2179 const char **ret_error_message) {
2180 int r;
2181
2182 assert(saved_rlimit_nofile);
2183 assert(saved_rlimit_memlock);
2184 assert(ret_error_message);
2185
2186 (void) parse_configuration(saved_rlimit_nofile, saved_rlimit_memlock);
2187
2188 r = parse_argv(argc, argv);
2189 if (r < 0) {
2190 *ret_error_message = "Failed to parse commandline arguments";
2191 return r;
2192 }
2193
2194 /* Initialize default unit */
2195 if (!arg_default_unit) {
2196 arg_default_unit = strdup(SPECIAL_DEFAULT_TARGET);
2197 if (!arg_default_unit) {
2198 *ret_error_message = "Failed to set default unit";
2199 return log_oom();
2200 }
2201 }
2202
2203 /* Initialize the show status setting if it hasn't been set explicitly yet */
2204 if (arg_show_status == _SHOW_STATUS_INVALID)
2205 arg_show_status = SHOW_STATUS_YES;
2206
2207 return 0;
2208 }
2209
2210 static int safety_checks(void) {
2211
2212 if (getpid_cached() == 1 &&
2213 arg_action != ACTION_RUN)
2214 return log_error_errno(SYNTHETIC_ERRNO(EPERM),
2215 "Unsupported execution mode while PID 1.");
2216
2217 if (getpid_cached() == 1 &&
2218 !arg_system)
2219 return log_error_errno(SYNTHETIC_ERRNO(EPERM),
2220 "Can't run --user mode as PID 1.");
2221
2222 if (arg_action == ACTION_RUN &&
2223 arg_system &&
2224 getpid_cached() != 1)
2225 return log_error_errno(SYNTHETIC_ERRNO(EPERM),
2226 "Can't run system mode unless PID 1.");
2227
2228 if (arg_action == ACTION_TEST &&
2229 geteuid() == 0)
2230 return log_error_errno(SYNTHETIC_ERRNO(EPERM),
2231 "Don't run test mode as root.");
2232
2233 if (!arg_system &&
2234 arg_action == ACTION_RUN &&
2235 sd_booted() <= 0)
2236 return log_error_errno(SYNTHETIC_ERRNO(EOPNOTSUPP),
2237 "Trying to run as user instance, but the system has not been booted with systemd.");
2238
2239 if (!arg_system &&
2240 arg_action == ACTION_RUN &&
2241 !getenv("XDG_RUNTIME_DIR"))
2242 return log_error_errno(SYNTHETIC_ERRNO(EUNATCH),
2243 "Trying to run as user instance, but $XDG_RUNTIME_DIR is not set.");
2244
2245 if (arg_system &&
2246 arg_action == ACTION_RUN &&
2247 running_in_chroot() > 0)
2248 return log_error_errno(SYNTHETIC_ERRNO(EOPNOTSUPP),
2249 "Cannot be run in a chroot() environment.");
2250
2251 return 0;
2252 }
2253
2254 static int initialize_security(
2255 bool *loaded_policy,
2256 dual_timestamp *security_start_timestamp,
2257 dual_timestamp *security_finish_timestamp,
2258 const char **ret_error_message) {
2259
2260 int r;
2261
2262 assert(loaded_policy);
2263 assert(security_start_timestamp);
2264 assert(security_finish_timestamp);
2265 assert(ret_error_message);
2266
2267 dual_timestamp_get(security_start_timestamp);
2268
2269 r = mac_selinux_setup(loaded_policy);
2270 if (r < 0) {
2271 *ret_error_message = "Failed to load SELinux policy";
2272 return r;
2273 }
2274
2275 r = mac_smack_setup(loaded_policy);
2276 if (r < 0) {
2277 *ret_error_message = "Failed to load SMACK policy";
2278 return r;
2279 }
2280
2281 r = ima_setup();
2282 if (r < 0) {
2283 *ret_error_message = "Failed to load IMA policy";
2284 return r;
2285 }
2286
2287 dual_timestamp_get(security_finish_timestamp);
2288 return 0;
2289 }
2290
2291 static void test_summary(Manager *m) {
2292 assert(m);
2293
2294 printf("-> By units:\n");
2295 manager_dump_units(m, stdout, "\t");
2296
2297 printf("-> By jobs:\n");
2298 manager_dump_jobs(m, stdout, "\t");
2299 }
2300
2301 static int collect_fds(FDSet **ret_fds, const char **ret_error_message) {
2302 int r;
2303
2304 assert(ret_fds);
2305 assert(ret_error_message);
2306
2307 r = fdset_new_fill(ret_fds);
2308 if (r < 0) {
2309 *ret_error_message = "Failed to allocate fd set";
2310 return log_emergency_errno(r, "Failed to allocate fd set: %m");
2311 }
2312
2313 fdset_cloexec(*ret_fds, true);
2314
2315 if (arg_serialization)
2316 assert_se(fdset_remove(*ret_fds, fileno(arg_serialization)) >= 0);
2317
2318 return 0;
2319 }
2320
2321 static void setup_console_terminal(bool skip_setup) {
2322
2323 if (!arg_system)
2324 return;
2325
2326 /* Become a session leader if we aren't one yet. */
2327 (void) setsid();
2328
2329 /* If we are init, we connect stdin/stdout/stderr to /dev/null and make sure we don't have a controlling
2330 * tty. */
2331 (void) release_terminal();
2332
2333 /* Reset the console, but only if this is really init and we are freshly booted */
2334 if (getpid_cached() == 1 && !skip_setup)
2335 (void) console_setup();
2336 }
2337
2338 static bool early_skip_setup_check(int argc, char *argv[]) {
2339 bool found_deserialize = false;
2340 int i;
2341
2342 /* Determine if this is a reexecution or normal bootup. We do the full command line parsing much later, so
2343 * let's just have a quick peek here. Note that if we have switched root, do all the special setup things
2344 * anyway, even if in that case we also do deserialization. */
2345
2346 for (i = 1; i < argc; i++) {
2347 if (streq(argv[i], "--switched-root"))
2348 return false; /* If we switched root, don't skip the setup. */
2349 else if (streq(argv[i], "--deserialize"))
2350 found_deserialize = true;
2351 }
2352
2353 return found_deserialize; /* When we are deserializing, then we are reexecuting, hence avoid the extensive setup */
2354 }
2355
2356 static int save_env(void) {
2357 char **l;
2358
2359 l = strv_copy(environ);
2360 if (!l)
2361 return -ENOMEM;
2362
2363 strv_free_and_replace(saved_env, l);
2364 return 0;
2365 }
2366
2367 int main(int argc, char *argv[]) {
2368
2369 dual_timestamp initrd_timestamp = DUAL_TIMESTAMP_NULL, userspace_timestamp = DUAL_TIMESTAMP_NULL, kernel_timestamp = DUAL_TIMESTAMP_NULL,
2370 security_start_timestamp = DUAL_TIMESTAMP_NULL, security_finish_timestamp = DUAL_TIMESTAMP_NULL;
2371 struct rlimit saved_rlimit_nofile = RLIMIT_MAKE_CONST(0),
2372 saved_rlimit_memlock = RLIMIT_MAKE_CONST(RLIM_INFINITY); /* The original rlimits we passed
2373 * in. Note we use different values
2374 * for the two that indicate whether
2375 * these fields are initialized! */
2376 bool skip_setup, loaded_policy = false, queue_default_job = false, first_boot = false, reexecute = false;
2377 char *switch_root_dir = NULL, *switch_root_init = NULL;
2378 usec_t before_startup, after_startup;
2379 static char systemd[] = "systemd";
2380 char timespan[FORMAT_TIMESPAN_MAX];
2381 const char *shutdown_verb = NULL, *error_message = NULL;
2382 int r, retval = EXIT_FAILURE;
2383 Manager *m = NULL;
2384 FDSet *fds = NULL;
2385
2386 /* SysV compatibility: redirect init → telinit */
2387 redirect_telinit(argc, argv);
2388
2389 /* Take timestamps early on */
2390 dual_timestamp_from_monotonic(&kernel_timestamp, 0);
2391 dual_timestamp_get(&userspace_timestamp);
2392
2393 /* Figure out whether we need to do initialize the system, or if we already did that because we are
2394 * reexecuting */
2395 skip_setup = early_skip_setup_check(argc, argv);
2396
2397 /* If we get started via the /sbin/init symlink then we are called 'init'. After a subsequent reexecution we
2398 * are then called 'systemd'. That is confusing, hence let's call us systemd right-away. */
2399 program_invocation_short_name = systemd;
2400 (void) prctl(PR_SET_NAME, systemd);
2401
2402 /* Save the original command line */
2403 save_argc_argv(argc, argv);
2404
2405 /* Save the original environment as we might need to restore it if we're requested to execute another
2406 * system manager later. */
2407 r = save_env();
2408 if (r < 0) {
2409 error_message = "Failed to copy environment block";
2410 goto finish;
2411 }
2412
2413 /* Make sure that if the user says "syslog" we actually log to the journal. */
2414 log_set_upgrade_syslog_to_journal(true);
2415
2416 if (getpid_cached() == 1) {
2417 /* When we run as PID 1 force system mode */
2418 arg_system = true;
2419
2420 /* Disable the umask logic */
2421 umask(0);
2422
2423 /* Make sure that at least initially we do not ever log to journald/syslogd, because it might not be
2424 * activated yet (even though the log socket for it exists). */
2425 log_set_prohibit_ipc(true);
2426
2427 /* Always reopen /dev/console when running as PID 1 or one of its pre-execve() children. This is
2428 * important so that we never end up logging to any foreign stderr, for example if we have to log in a
2429 * child process right before execve()'ing the actual binary, at a point in time where socket
2430 * activation stderr/stdout area already set up. */
2431 log_set_always_reopen_console(true);
2432
2433 if (detect_container() <= 0) {
2434
2435 /* Running outside of a container as PID 1 */
2436 log_set_target(LOG_TARGET_KMSG);
2437 log_open();
2438
2439 if (in_initrd())
2440 initrd_timestamp = userspace_timestamp;
2441
2442 if (!skip_setup) {
2443 r = mount_setup_early();
2444 if (r < 0) {
2445 error_message = "Failed to mount early API filesystems";
2446 goto finish;
2447 }
2448
2449 /* Let's open the log backend a second time, in case the first time didn't
2450 * work. Quite possibly we have mounted /dev just now, so /dev/kmsg became
2451 * available, and it previously wasn't. */
2452 log_open();
2453
2454 disable_printk_ratelimit();
2455
2456 r = initialize_security(
2457 &loaded_policy,
2458 &security_start_timestamp,
2459 &security_finish_timestamp,
2460 &error_message);
2461 if (r < 0)
2462 goto finish;
2463 }
2464
2465 if (mac_selinux_init() < 0) {
2466 error_message = "Failed to initialize SELinux policy";
2467 goto finish;
2468 }
2469
2470 if (!skip_setup)
2471 initialize_clock();
2472
2473 /* Set the default for later on, but don't actually open the logs like this for now. Note that
2474 * if we are transitioning from the initrd there might still be journal fd open, and we
2475 * shouldn't attempt opening that before we parsed /proc/cmdline which might redirect output
2476 * elsewhere. */
2477 log_set_target(LOG_TARGET_JOURNAL_OR_KMSG);
2478
2479 } else {
2480 /* Running inside a container, as PID 1 */
2481 log_set_target(LOG_TARGET_CONSOLE);
2482 log_open();
2483
2484 /* For later on, see above... */
2485 log_set_target(LOG_TARGET_JOURNAL);
2486
2487 /* clear the kernel timestamp,
2488 * because we are in a container */
2489 kernel_timestamp = DUAL_TIMESTAMP_NULL;
2490 }
2491
2492 initialize_coredump(skip_setup);
2493
2494 r = fixup_environment();
2495 if (r < 0) {
2496 log_emergency_errno(r, "Failed to fix up PID 1 environment: %m");
2497 error_message = "Failed to fix up PID1 environment";
2498 goto finish;
2499 }
2500
2501 } else {
2502 /* Running as user instance */
2503 arg_system = false;
2504 log_set_target(LOG_TARGET_AUTO);
2505 log_open();
2506
2507 /* clear the kernel timestamp,
2508 * because we are not PID 1 */
2509 kernel_timestamp = DUAL_TIMESTAMP_NULL;
2510 }
2511
2512 if (arg_system) {
2513 /* Try to figure out if we can use colors with the console. No need to do that for user instances since
2514 * they never log into the console. */
2515 log_show_color(colors_enabled());
2516
2517 r = make_null_stdio();
2518 if (r < 0)
2519 log_warning_errno(r, "Failed to redirect standard streams to /dev/null, ignoring: %m");
2520 }
2521
2522 /* Mount /proc, /sys and friends, so that /proc/cmdline and
2523 * /proc/$PID/fd is available. */
2524 if (getpid_cached() == 1) {
2525
2526 /* Load the kernel modules early. */
2527 if (!skip_setup)
2528 kmod_setup();
2529
2530 r = mount_setup(loaded_policy);
2531 if (r < 0) {
2532 error_message = "Failed to mount API filesystems";
2533 goto finish;
2534 }
2535
2536 /* The efivarfs is now mounted, let's read the random seed off it */
2537 (void) efi_take_random_seed();
2538 }
2539
2540 /* Save the original RLIMIT_NOFILE/RLIMIT_MEMLOCK so that we can reset it later when
2541 * transitioning from the initrd to the main systemd or suchlike. */
2542 save_rlimits(&saved_rlimit_nofile, &saved_rlimit_memlock);
2543
2544 /* Reset all signal handlers. */
2545 (void) reset_all_signal_handlers();
2546 (void) ignore_signals(SIGNALS_IGNORE, -1);
2547
2548 r = load_configuration(argc, argv, &saved_rlimit_nofile, &saved_rlimit_memlock, &error_message);
2549 if (r < 0)
2550 goto finish;
2551
2552 r = safety_checks();
2553 if (r < 0)
2554 goto finish;
2555
2556 if (IN_SET(arg_action, ACTION_TEST, ACTION_HELP, ACTION_DUMP_CONFIGURATION_ITEMS, ACTION_DUMP_BUS_PROPERTIES))
2557 (void) pager_open(arg_pager_flags);
2558
2559 if (arg_action != ACTION_RUN)
2560 skip_setup = true;
2561
2562 if (arg_action == ACTION_HELP) {
2563 retval = help() < 0 ? EXIT_FAILURE : EXIT_SUCCESS;
2564 goto finish;
2565 } else if (arg_action == ACTION_VERSION) {
2566 retval = version();
2567 goto finish;
2568 } else if (arg_action == ACTION_DUMP_CONFIGURATION_ITEMS) {
2569 unit_dump_config_items(stdout);
2570 retval = EXIT_SUCCESS;
2571 goto finish;
2572 } else if (arg_action == ACTION_DUMP_BUS_PROPERTIES) {
2573 dump_bus_properties(stdout);
2574 retval = EXIT_SUCCESS;
2575 goto finish;
2576 }
2577
2578 assert_se(IN_SET(arg_action, ACTION_RUN, ACTION_TEST));
2579
2580 /* Move out of the way, so that we won't block unmounts */
2581 assert_se(chdir("/") == 0);
2582
2583 if (arg_action == ACTION_RUN) {
2584
2585 /* A core pattern might have been specified via the cmdline. */
2586 initialize_core_pattern(skip_setup);
2587
2588 /* Close logging fds, in order not to confuse collecting passed fds and terminal logic below */
2589 log_close();
2590
2591 /* Remember open file descriptors for later deserialization */
2592 r = collect_fds(&fds, &error_message);
2593 if (r < 0)
2594 goto finish;
2595
2596 /* Give up any control of the console, but make sure its initialized. */
2597 setup_console_terminal(skip_setup);
2598
2599 /* Open the logging devices, if possible and necessary */
2600 log_open();
2601 }
2602
2603 log_execution_mode(&first_boot);
2604
2605 r = initialize_runtime(skip_setup,
2606 &saved_rlimit_nofile,
2607 &saved_rlimit_memlock,
2608 &error_message);
2609 if (r < 0)
2610 goto finish;
2611
2612 r = manager_new(arg_system ? UNIT_FILE_SYSTEM : UNIT_FILE_USER,
2613 arg_action == ACTION_TEST ? MANAGER_TEST_FULL : 0,
2614 &m);
2615 if (r < 0) {
2616 log_emergency_errno(r, "Failed to allocate manager object: %m");
2617 error_message = "Failed to allocate manager object";
2618 goto finish;
2619 }
2620
2621 m->timestamps[MANAGER_TIMESTAMP_KERNEL] = kernel_timestamp;
2622 m->timestamps[MANAGER_TIMESTAMP_INITRD] = initrd_timestamp;
2623 m->timestamps[MANAGER_TIMESTAMP_USERSPACE] = userspace_timestamp;
2624 m->timestamps[manager_timestamp_initrd_mangle(MANAGER_TIMESTAMP_SECURITY_START)] = security_start_timestamp;
2625 m->timestamps[manager_timestamp_initrd_mangle(MANAGER_TIMESTAMP_SECURITY_FINISH)] = security_finish_timestamp;
2626
2627 set_manager_defaults(m);
2628 set_manager_settings(m);
2629 manager_set_first_boot(m, first_boot);
2630
2631 /* Remember whether we should queue the default job */
2632 queue_default_job = !arg_serialization || arg_switched_root;
2633
2634 before_startup = now(CLOCK_MONOTONIC);
2635
2636 r = manager_startup(m, arg_serialization, fds);
2637 if (r < 0) {
2638 error_message = "Failed to start up manager";
2639 goto finish;
2640 }
2641
2642 /* This will close all file descriptors that were opened, but not claimed by any unit. */
2643 fds = fdset_free(fds);
2644 arg_serialization = safe_fclose(arg_serialization);
2645
2646 if (queue_default_job) {
2647 r = do_queue_default_job(m, &error_message);
2648 if (r < 0)
2649 goto finish;
2650 }
2651
2652 after_startup = now(CLOCK_MONOTONIC);
2653
2654 log_full(arg_action == ACTION_TEST ? LOG_INFO : LOG_DEBUG,
2655 "Loaded units and determined initial transaction in %s.",
2656 format_timespan(timespan, sizeof(timespan), after_startup - before_startup, 100 * USEC_PER_MSEC));
2657
2658 if (arg_action == ACTION_TEST) {
2659 test_summary(m);
2660 retval = EXIT_SUCCESS;
2661 goto finish;
2662 }
2663
2664 (void) invoke_main_loop(m,
2665 &saved_rlimit_nofile,
2666 &saved_rlimit_memlock,
2667 &reexecute,
2668 &retval,
2669 &shutdown_verb,
2670 &fds,
2671 &switch_root_dir,
2672 &switch_root_init,
2673 &error_message);
2674
2675 finish:
2676 pager_close();
2677
2678 if (m) {
2679 arg_reboot_watchdog = m->reboot_watchdog;
2680 arg_kexec_watchdog = m->kexec_watchdog;
2681 m = manager_free(m);
2682 }
2683
2684 reset_arguments();
2685 mac_selinux_finish();
2686
2687 if (reexecute)
2688 do_reexecute(argc, argv,
2689 &saved_rlimit_nofile,
2690 &saved_rlimit_memlock,
2691 fds,
2692 switch_root_dir,
2693 switch_root_init,
2694 &error_message); /* This only returns if reexecution failed */
2695
2696 arg_serialization = safe_fclose(arg_serialization);
2697 fds = fdset_free(fds);
2698
2699 saved_env = strv_free(saved_env);
2700
2701 #if HAVE_VALGRIND_VALGRIND_H
2702 /* If we are PID 1 and running under valgrind, then let's exit
2703 * here explicitly. valgrind will only generate nice output on
2704 * exit(), not on exec(), hence let's do the former not the
2705 * latter here. */
2706 if (getpid_cached() == 1 && RUNNING_ON_VALGRIND) {
2707 /* Cleanup watchdog_device strings for valgrind. We need them
2708 * in become_shutdown() so normally we cannot free them yet. */
2709 watchdog_free_device();
2710 arg_watchdog_device = mfree(arg_watchdog_device);
2711 return retval;
2712 }
2713 #endif
2714
2715 #if HAS_FEATURE_ADDRESS_SANITIZER
2716 __lsan_do_leak_check();
2717 #endif
2718
2719 if (shutdown_verb) {
2720 r = become_shutdown(shutdown_verb, retval);
2721 log_error_errno(r, "Failed to execute shutdown binary, %s: %m", getpid_cached() == 1 ? "freezing" : "quitting");
2722 error_message = "Failed to execute shutdown binary";
2723 }
2724
2725 watchdog_free_device();
2726 arg_watchdog_device = mfree(arg_watchdog_device);
2727
2728 if (getpid_cached() == 1) {
2729 if (error_message)
2730 manager_status_printf(NULL, STATUS_TYPE_EMERGENCY,
2731 ANSI_HIGHLIGHT_RED "!!!!!!" ANSI_NORMAL,
2732 "%s.", error_message);
2733 freeze_or_exit_or_reboot();
2734 }
2735
2736 return retval;
2737 }
Unnamed repository; edit this file 'description' to name the repository.