]> git.ipfire.org Git - thirdparty/systemd.git/blob - src/core/main.c
projects / thirdparty / systemd.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
pid1: rework environment block copy logic
[thirdparty/systemd.git] / src / core / main.c
1 /* SPDX-License-Identifier: LGPL-2.1+ */
2
3 #include <errno.h>
4 #include <fcntl.h>
5 #include <getopt.h>
6 #include <signal.h>
7 #include <stdio.h>
8 #include <string.h>
9 #include <sys/mount.h>
10 #include <sys/prctl.h>
11 #include <sys/reboot.h>
12 #include <sys/stat.h>
13 #include <unistd.h>
14 #if HAVE_SECCOMP
15 #include <seccomp.h>
16 #endif
17 #if HAVE_VALGRIND_VALGRIND_H
18 #include <valgrind/valgrind.h>
19 #endif
20
21 #include "sd-bus.h"
22 #include "sd-daemon.h"
23 #include "sd-messages.h"
24
25 #include "alloc-util.h"
26 #include "architecture.h"
27 #include "build.h"
28 #include "bus-error.h"
29 #include "bus-util.h"
30 #include "capability-util.h"
31 #include "cgroup-util.h"
32 #include "clock-util.h"
33 #include "conf-parser.h"
34 #include "cpu-set-util.h"
35 #include "dbus-manager.h"
36 #include "dbus.h"
37 #include "def.h"
38 #include "efi-random.h"
39 #include "emergency-action.h"
40 #include "env-util.h"
41 #include "exit-status.h"
42 #include "fd-util.h"
43 #include "fdset.h"
44 #include "fileio.h"
45 #include "format-util.h"
46 #include "fs-util.h"
47 #include "hostname-setup.h"
48 #include "ima-setup.h"
49 #include "killall.h"
50 #include "kmod-setup.h"
51 #include "limits-util.h"
52 #include "load-fragment.h"
53 #include "log.h"
54 #include "loopback-setup.h"
55 #include "machine-id-setup.h"
56 #include "manager.h"
57 #include "mount-setup.h"
58 #include "os-util.h"
59 #include "pager.h"
60 #include "parse-util.h"
61 #include "path-util.h"
62 #include "pretty-print.h"
63 #include "proc-cmdline.h"
64 #include "process-util.h"
65 #include "raw-clone.h"
66 #include "rlimit-util.h"
67 #if HAVE_SECCOMP
68 #include "seccomp-util.h"
69 #endif
70 #include "selinux-setup.h"
71 #include "selinux-util.h"
72 #include "signal-util.h"
73 #include "smack-setup.h"
74 #include "special.h"
75 #include "stat-util.h"
76 #include "stdio-util.h"
77 #include "strv.h"
78 #include "switch-root.h"
79 #include "sysctl-util.h"
80 #include "terminal-util.h"
81 #include "umask-util.h"
82 #include "user-util.h"
83 #include "util.h"
84 #include "virt.h"
85 #include "watchdog.h"
86
87 #if HAS_FEATURE_ADDRESS_SANITIZER
88 #include <sanitizer/lsan_interface.h>
89 #endif
90
91 static enum {
92 ACTION_RUN,
93 ACTION_HELP,
94 ACTION_VERSION,
95 ACTION_TEST,
96 ACTION_DUMP_CONFIGURATION_ITEMS,
97 ACTION_DUMP_BUS_PROPERTIES,
98 } arg_action = ACTION_RUN;
99
100 /* Those variables are initalized to 0 automatically, so we avoid uninitialized memory access.
101 * Real defaults are assigned in reset_arguments() below. */
102 static char *arg_default_unit;
103 static bool arg_system;
104 static bool arg_dump_core;
105 static int arg_crash_chvt;
106 static bool arg_crash_shell;
107 static bool arg_crash_reboot;
108 static char *arg_confirm_spawn;
109 static ShowStatus arg_show_status;
110 static StatusUnitFormat arg_status_unit_format;
111 static bool arg_switched_root;
112 static PagerFlags arg_pager_flags;
113 static bool arg_service_watchdogs;
114 static ExecOutput arg_default_std_output;
115 static ExecOutput arg_default_std_error;
116 static usec_t arg_default_restart_usec;
117 static usec_t arg_default_timeout_start_usec;
118 static usec_t arg_default_timeout_stop_usec;
119 static usec_t arg_default_timeout_abort_usec;
120 static bool arg_default_timeout_abort_set;
121 static usec_t arg_default_start_limit_interval;
122 static unsigned arg_default_start_limit_burst;
123 static usec_t arg_runtime_watchdog;
124 static usec_t arg_reboot_watchdog;
125 static usec_t arg_kexec_watchdog;
126 static char *arg_early_core_pattern;
127 static char *arg_watchdog_device;
128 static char **arg_default_environment;
129 static struct rlimit *arg_default_rlimit[_RLIMIT_MAX];
130 static uint64_t arg_capability_bounding_set;
131 static bool arg_no_new_privs;
132 static nsec_t arg_timer_slack_nsec;
133 static usec_t arg_default_timer_accuracy_usec;
134 static Set* arg_syscall_archs;
135 static FILE* arg_serialization;
136 static int arg_default_cpu_accounting;
137 static bool arg_default_io_accounting;
138 static bool arg_default_ip_accounting;
139 static bool arg_default_blockio_accounting;
140 static bool arg_default_memory_accounting;
141 static bool arg_default_tasks_accounting;
142 static uint64_t arg_default_tasks_max;
143 static sd_id128_t arg_machine_id;
144 static EmergencyAction arg_cad_burst_action;
145 static OOMPolicy arg_default_oom_policy;
146 static CPUSet arg_cpu_affinity;
147 static NUMAPolicy arg_numa_policy;
148
149 /* A copy of the original environment block */
150 static char **saved_env = NULL;
151
152 static int parse_configuration(const struct rlimit *saved_rlimit_nofile,
153 const struct rlimit *saved_rlimit_memlock);
154
155 _noreturn_ static void freeze_or_exit_or_reboot(void) {
156
157 /* If we are running in a container, let's prefer exiting, after all we can propagate an exit code to
158 * the container manager, and thus inform it that something went wrong. */
159 if (detect_container() > 0) {
160 log_emergency("Exiting PID 1...");
161 _exit(EXIT_EXCEPTION);
162 }
163
164 if (arg_crash_reboot) {
165 log_notice("Rebooting in 10s...");
166 (void) sleep(10);
167
168 log_notice("Rebooting now...");
169 (void) reboot(RB_AUTOBOOT);
170 log_emergency_errno(errno, "Failed to reboot: %m");
171 }
172
173 log_emergency("Freezing execution.");
174 freeze();
175 }
176
177 _noreturn_ static void crash(int sig) {
178 struct sigaction sa;
179 pid_t pid;
180
181 if (getpid_cached() != 1)
182 /* Pass this on immediately, if this is not PID 1 */
183 (void) raise(sig);
184 else if (!arg_dump_core)
185 log_emergency("Caught <%s>, not dumping core.", signal_to_string(sig));
186 else {
187 sa = (struct sigaction) {
188 .sa_handler = nop_signal_handler,
189 .sa_flags = SA_NOCLDSTOP|SA_RESTART,
190 };
191
192 /* We want to wait for the core process, hence let's enable SIGCHLD */
193 (void) sigaction(SIGCHLD, &sa, NULL);
194
195 pid = raw_clone(SIGCHLD);
196 if (pid < 0)
197 log_emergency_errno(errno, "Caught <%s>, cannot fork for core dump: %m", signal_to_string(sig));
198 else if (pid == 0) {
199 /* Enable default signal handler for core dump */
200
201 sa = (struct sigaction) {
202 .sa_handler = SIG_DFL,
203 };
204 (void) sigaction(sig, &sa, NULL);
205
206 /* Don't limit the coredump size */
207 (void) setrlimit(RLIMIT_CORE, &RLIMIT_MAKE_CONST(RLIM_INFINITY));
208
209 /* Just to be sure... */
210 (void) chdir("/");
211
212 /* Raise the signal again */
213 pid = raw_getpid();
214 (void) kill(pid, sig); /* raise() would kill the parent */
215
216 assert_not_reached("We shouldn't be here...");
217 _exit(EXIT_EXCEPTION);
218 } else {
219 siginfo_t status;
220 int r;
221
222 /* Order things nicely. */
223 r = wait_for_terminate(pid, &status);
224 if (r < 0)
225 log_emergency_errno(r, "Caught <%s>, waitpid() failed: %m", signal_to_string(sig));
226 else if (status.si_code != CLD_DUMPED) {
227 const char *s = status.si_code == CLD_EXITED
228 ? exit_status_to_string(status.si_status, EXIT_STATUS_LIBC)
229 : signal_to_string(status.si_status);
230
231 log_emergency("Caught <%s>, core dump failed (child "PID_FMT", code=%s, status=%i/%s).",
232 signal_to_string(sig),
233 pid,
234 sigchld_code_to_string(status.si_code),
235 status.si_status, strna(s));
236 } else
237 log_emergency("Caught <%s>, dumped core as pid "PID_FMT".",
238 signal_to_string(sig), pid);
239 }
240 }
241
242 if (arg_crash_chvt >= 0)
243 (void) chvt(arg_crash_chvt);
244
245 sa = (struct sigaction) {
246 .sa_handler = SIG_IGN,
247 .sa_flags = SA_NOCLDSTOP|SA_NOCLDWAIT|SA_RESTART,
248 };
249
250 /* Let the kernel reap children for us */
251 (void) sigaction(SIGCHLD, &sa, NULL);
252
253 if (arg_crash_shell) {
254 log_notice("Executing crash shell in 10s...");
255 (void) sleep(10);
256
257 pid = raw_clone(SIGCHLD);
258 if (pid < 0)
259 log_emergency_errno(errno, "Failed to fork off crash shell: %m");
260 else if (pid == 0) {
261 (void) setsid();
262 (void) make_console_stdio();
263 (void) rlimit_nofile_safe();
264 (void) execle("/bin/sh", "/bin/sh", NULL, environ);
265
266 log_emergency_errno(errno, "execle() failed: %m");
267 _exit(EXIT_EXCEPTION);
268 } else {
269 log_info("Spawned crash shell as PID "PID_FMT".", pid);
270 (void) wait_for_terminate(pid, NULL);
271 }
272 }
273
274 freeze_or_exit_or_reboot();
275 }
276
277 static void install_crash_handler(void) {
278 static const struct sigaction sa = {
279 .sa_handler = crash,
280 .sa_flags = SA_NODEFER, /* So that we can raise the signal again from the signal handler */
281 };
282 int r;
283
284 /* We ignore the return value here, since, we don't mind if we
285 * cannot set up a crash handler */
286 r = sigaction_many(&sa, SIGNALS_CRASH_HANDLER, -1);
287 if (r < 0)
288 log_debug_errno(r, "I had trouble setting up the crash handler, ignoring: %m");
289 }
290
291 static int console_setup(void) {
292 _cleanup_close_ int tty_fd = -1;
293 int r;
294
295 tty_fd = open_terminal("/dev/console", O_WRONLY|O_NOCTTY|O_CLOEXEC);
296 if (tty_fd < 0)
297 return log_error_errno(tty_fd, "Failed to open /dev/console: %m");
298
299 /* We don't want to force text mode. plymouth may be showing
300 * pictures already from initrd. */
301 r = reset_terminal_fd(tty_fd, false);
302 if (r < 0)
303 return log_error_errno(r, "Failed to reset /dev/console: %m");
304
305 return 0;
306 }
307
308 static int set_machine_id(const char *m) {
309 sd_id128_t t;
310 assert(m);
311
312 if (sd_id128_from_string(m, &t) < 0)
313 return -EINVAL;
314
315 if (sd_id128_is_null(t))
316 return -EINVAL;
317
318 arg_machine_id = t;
319 return 0;
320 }
321
322 static int parse_proc_cmdline_item(const char *key, const char *value, void *data) {
323
324 int r;
325
326 assert(key);
327
328 if (STR_IN_SET(key, "systemd.unit", "rd.systemd.unit")) {
329
330 if (proc_cmdline_value_missing(key, value))
331 return 0;
332
333 if (!unit_name_is_valid(value, UNIT_NAME_PLAIN|UNIT_NAME_INSTANCE))
334 log_warning("Unit name specified on %s= is not valid, ignoring: %s", key, value);
335 else if (in_initrd() == !!startswith(key, "rd.")) {
336 if (free_and_strdup(&arg_default_unit, value) < 0)
337 return log_oom();
338 }
339
340 } else if (proc_cmdline_key_streq(key, "systemd.dump_core")) {
341
342 r = value ? parse_boolean(value) : true;
343 if (r < 0)
344 log_warning_errno(r, "Failed to parse dump core switch %s, ignoring: %m", value);
345 else
346 arg_dump_core = r;
347
348 } else if (proc_cmdline_key_streq(key, "systemd.early_core_pattern")) {
349
350 if (proc_cmdline_value_missing(key, value))
351 return 0;
352
353 if (path_is_absolute(value))
354 (void) parse_path_argument_and_warn(value, false, &arg_early_core_pattern);
355 else
356 log_warning("Specified core pattern '%s' is not an absolute path, ignoring.", value);
357
358 } else if (proc_cmdline_key_streq(key, "systemd.crash_chvt")) {
359
360 if (!value)
361 arg_crash_chvt = 0; /* turn on */
362 else {
363 r = parse_crash_chvt(value, &arg_crash_chvt);
364 if (r < 0)
365 log_warning_errno(r, "Failed to parse crash chvt switch %s, ignoring: %m", value);
366 }
367
368 } else if (proc_cmdline_key_streq(key, "systemd.crash_shell")) {
369
370 r = value ? parse_boolean(value) : true;
371 if (r < 0)
372 log_warning_errno(r, "Failed to parse crash shell switch %s, ignoring: %m", value);
373 else
374 arg_crash_shell = r;
375
376 } else if (proc_cmdline_key_streq(key, "systemd.crash_reboot")) {
377
378 r = value ? parse_boolean(value) : true;
379 if (r < 0)
380 log_warning_errno(r, "Failed to parse crash reboot switch %s, ignoring: %m", value);
381 else
382 arg_crash_reboot = r;
383
384 } else if (proc_cmdline_key_streq(key, "systemd.confirm_spawn")) {
385 char *s;
386
387 r = parse_confirm_spawn(value, &s);
388 if (r < 0)
389 log_warning_errno(r, "Failed to parse confirm_spawn switch %s, ignoring: %m", value);
390 else
391 free_and_replace(arg_confirm_spawn, s);
392
393 } else if (proc_cmdline_key_streq(key, "systemd.service_watchdogs")) {
394
395 r = value ? parse_boolean(value) : true;
396 if (r < 0)
397 log_warning_errno(r, "Failed to parse service watchdog switch %s, ignoring: %m", value);
398 else
399 arg_service_watchdogs = r;
400
401 } else if (proc_cmdline_key_streq(key, "systemd.show_status")) {
402
403 if (value) {
404 r = parse_show_status(value, &arg_show_status);
405 if (r < 0)
406 log_warning_errno(r, "Failed to parse show status switch %s, ignoring: %m", value);
407 } else
408 arg_show_status = SHOW_STATUS_YES;
409
410 } else if (proc_cmdline_key_streq(key, "systemd.status_unit_format")) {
411
412 if (proc_cmdline_value_missing(key, value))
413 return 0;
414
415 r = status_unit_format_from_string(value);
416 if (r < 0)
417 log_warning_errno(r, "Failed to parse %s=%s, ignoring: %m", key, value);
418 else
419 arg_status_unit_format = r;
420
421 } else if (proc_cmdline_key_streq(key, "systemd.default_standard_output")) {
422
423 if (proc_cmdline_value_missing(key, value))
424 return 0;
425
426 r = exec_output_from_string(value);
427 if (r < 0)
428 log_warning_errno(r, "Failed to parse default standard output switch %s, ignoring: %m", value);
429 else
430 arg_default_std_output = r;
431
432 } else if (proc_cmdline_key_streq(key, "systemd.default_standard_error")) {
433
434 if (proc_cmdline_value_missing(key, value))
435 return 0;
436
437 r = exec_output_from_string(value);
438 if (r < 0)
439 log_warning_errno(r, "Failed to parse default standard error switch %s, ignoring: %m", value);
440 else
441 arg_default_std_error = r;
442
443 } else if (streq(key, "systemd.setenv")) {
444
445 if (proc_cmdline_value_missing(key, value))
446 return 0;
447
448 if (env_assignment_is_valid(value)) {
449 char **env;
450
451 env = strv_env_set(arg_default_environment, value);
452 if (!env)
453 return log_oom();
454
455 arg_default_environment = env;
456 } else
457 log_warning("Environment variable name '%s' is not valid. Ignoring.", value);
458
459 } else if (proc_cmdline_key_streq(key, "systemd.machine_id")) {
460
461 if (proc_cmdline_value_missing(key, value))
462 return 0;
463
464 r = set_machine_id(value);
465 if (r < 0)
466 log_warning_errno(r, "MachineID '%s' is not valid, ignoring: %m", value);
467
468 } else if (proc_cmdline_key_streq(key, "systemd.default_timeout_start_sec")) {
469
470 if (proc_cmdline_value_missing(key, value))
471 return 0;
472
473 r = parse_sec(value, &arg_default_timeout_start_usec);
474 if (r < 0)
475 log_warning_errno(r, "Failed to parse default start timeout '%s', ignoring: %m", value);
476
477 if (arg_default_timeout_start_usec <= 0)
478 arg_default_timeout_start_usec = USEC_INFINITY;
479
480 } else if (proc_cmdline_key_streq(key, "systemd.watchdog_device")) {
481
482 if (proc_cmdline_value_missing(key, value))
483 return 0;
484
485 (void) parse_path_argument_and_warn(value, false, &arg_watchdog_device);
486
487 } else if (streq(key, "quiet") && !value) {
488
489 if (arg_show_status == _SHOW_STATUS_INVALID)
490 arg_show_status = SHOW_STATUS_AUTO;
491
492 } else if (streq(key, "debug") && !value) {
493
494 /* Note that log_parse_environment() handles 'debug'
495 * too, and sets the log level to LOG_DEBUG. */
496
497 if (detect_container() > 0)
498 log_set_target(LOG_TARGET_CONSOLE);
499
500 } else if (!value) {
501 const char *target;
502
503 /* SysV compatibility */
504 target = runlevel_to_target(key);
505 if (target)
506 return free_and_strdup(&arg_default_unit, target);
507 }
508
509 return 0;
510 }
511
512 #define DEFINE_SETTER(name, func, descr) \
513 static int name(const char *unit, \
514 const char *filename, \
515 unsigned line, \
516 const char *section, \
517 unsigned section_line, \
518 const char *lvalue, \
519 int ltype, \
520 const char *rvalue, \
521 void *data, \
522 void *userdata) { \
523 \
524 int r; \
525 \
526 assert(filename); \
527 assert(lvalue); \
528 assert(rvalue); \
529 \
530 r = func(rvalue); \
531 if (r < 0) \
532 log_syntax(unit, LOG_ERR, filename, line, r, \
533 "Invalid " descr "'%s': %m", \
534 rvalue); \
535 \
536 return 0; \
537 }
538
539 DEFINE_SETTER(config_parse_level2, log_set_max_level_from_string, "log level");
540 DEFINE_SETTER(config_parse_target, log_set_target_from_string, "target");
541 DEFINE_SETTER(config_parse_color, log_show_color_from_string, "color" );
542 DEFINE_SETTER(config_parse_location, log_show_location_from_string, "location");
543
544 static int parse_config_file(void) {
545
546 const ConfigTableItem items[] = {
547 { "Manager", "LogLevel", config_parse_level2, 0, NULL },
548 { "Manager", "LogTarget", config_parse_target, 0, NULL },
549 { "Manager", "LogColor", config_parse_color, 0, NULL },
550 { "Manager", "LogLocation", config_parse_location, 0, NULL },
551 { "Manager", "DumpCore", config_parse_bool, 0, &arg_dump_core },
552 { "Manager", "CrashChVT", /* legacy */ config_parse_crash_chvt, 0, &arg_crash_chvt },
553 { "Manager", "CrashChangeVT", config_parse_crash_chvt, 0, &arg_crash_chvt },
554 { "Manager", "CrashShell", config_parse_bool, 0, &arg_crash_shell },
555 { "Manager", "CrashReboot", config_parse_bool, 0, &arg_crash_reboot },
556 { "Manager", "ShowStatus", config_parse_show_status, 0, &arg_show_status },
557 { "Manager", "StatusUnitFormat", config_parse_status_unit_format, 0, &arg_status_unit_format },
558 { "Manager", "CPUAffinity", config_parse_cpu_affinity2, 0, &arg_cpu_affinity },
559 { "Manager", "NUMAPolicy", config_parse_numa_policy, 0, &arg_numa_policy.type },
560 { "Manager", "NUMAMask", config_parse_numa_mask, 0, &arg_numa_policy },
561 { "Manager", "JoinControllers", config_parse_warn_compat, DISABLED_CONFIGURATION, NULL },
562 { "Manager", "RuntimeWatchdogSec", config_parse_sec, 0, &arg_runtime_watchdog },
563 { "Manager", "RebootWatchdogSec", config_parse_sec, 0, &arg_reboot_watchdog },
564 { "Manager", "ShutdownWatchdogSec", config_parse_sec, 0, &arg_reboot_watchdog }, /* obsolete alias */
565 { "Manager", "KExecWatchdogSec", config_parse_sec, 0, &arg_kexec_watchdog },
566 { "Manager", "WatchdogDevice", config_parse_path, 0, &arg_watchdog_device },
567 { "Manager", "CapabilityBoundingSet", config_parse_capability_set, 0, &arg_capability_bounding_set },
568 { "Manager", "NoNewPrivileges", config_parse_bool, 0, &arg_no_new_privs },
569 #if HAVE_SECCOMP
570 { "Manager", "SystemCallArchitectures", config_parse_syscall_archs, 0, &arg_syscall_archs },
571 #endif
572 { "Manager", "TimerSlackNSec", config_parse_nsec, 0, &arg_timer_slack_nsec },
573 { "Manager", "DefaultTimerAccuracySec", config_parse_sec, 0, &arg_default_timer_accuracy_usec },
574 { "Manager", "DefaultStandardOutput", config_parse_output_restricted, 0, &arg_default_std_output },
575 { "Manager", "DefaultStandardError", config_parse_output_restricted, 0, &arg_default_std_error },
576 { "Manager", "DefaultTimeoutStartSec", config_parse_sec, 0, &arg_default_timeout_start_usec },
577 { "Manager", "DefaultTimeoutStopSec", config_parse_sec, 0, &arg_default_timeout_stop_usec },
578 { "Manager", "DefaultTimeoutAbortSec", config_parse_timeout_abort, 0, &arg_default_timeout_abort_set },
579 { "Manager", "DefaultRestartSec", config_parse_sec, 0, &arg_default_restart_usec },
580 { "Manager", "DefaultStartLimitInterval", config_parse_sec, 0, &arg_default_start_limit_interval }, /* obsolete alias */
581 { "Manager", "DefaultStartLimitIntervalSec", config_parse_sec, 0, &arg_default_start_limit_interval },
582 { "Manager", "DefaultStartLimitBurst", config_parse_unsigned, 0, &arg_default_start_limit_burst },
583 { "Manager", "DefaultEnvironment", config_parse_environ, 0, &arg_default_environment },
584 { "Manager", "DefaultLimitCPU", config_parse_rlimit, RLIMIT_CPU, arg_default_rlimit },
585 { "Manager", "DefaultLimitFSIZE", config_parse_rlimit, RLIMIT_FSIZE, arg_default_rlimit },
586 { "Manager", "DefaultLimitDATA", config_parse_rlimit, RLIMIT_DATA, arg_default_rlimit },
587 { "Manager", "DefaultLimitSTACK", config_parse_rlimit, RLIMIT_STACK, arg_default_rlimit },
588 { "Manager", "DefaultLimitCORE", config_parse_rlimit, RLIMIT_CORE, arg_default_rlimit },
589 { "Manager", "DefaultLimitRSS", config_parse_rlimit, RLIMIT_RSS, arg_default_rlimit },
590 { "Manager", "DefaultLimitNOFILE", config_parse_rlimit, RLIMIT_NOFILE, arg_default_rlimit },
591 { "Manager", "DefaultLimitAS", config_parse_rlimit, RLIMIT_AS, arg_default_rlimit },
592 { "Manager", "DefaultLimitNPROC", config_parse_rlimit, RLIMIT_NPROC, arg_default_rlimit },
593 { "Manager", "DefaultLimitMEMLOCK", config_parse_rlimit, RLIMIT_MEMLOCK, arg_default_rlimit },
594 { "Manager", "DefaultLimitLOCKS", config_parse_rlimit, RLIMIT_LOCKS, arg_default_rlimit },
595 { "Manager", "DefaultLimitSIGPENDING", config_parse_rlimit, RLIMIT_SIGPENDING, arg_default_rlimit },
596 { "Manager", "DefaultLimitMSGQUEUE", config_parse_rlimit, RLIMIT_MSGQUEUE, arg_default_rlimit },
597 { "Manager", "DefaultLimitNICE", config_parse_rlimit, RLIMIT_NICE, arg_default_rlimit },
598 { "Manager", "DefaultLimitRTPRIO", config_parse_rlimit, RLIMIT_RTPRIO, arg_default_rlimit },
599 { "Manager", "DefaultLimitRTTIME", config_parse_rlimit, RLIMIT_RTTIME, arg_default_rlimit },
600 { "Manager", "DefaultCPUAccounting", config_parse_tristate, 0, &arg_default_cpu_accounting },
601 { "Manager", "DefaultIOAccounting", config_parse_bool, 0, &arg_default_io_accounting },
602 { "Manager", "DefaultIPAccounting", config_parse_bool, 0, &arg_default_ip_accounting },
603 { "Manager", "DefaultBlockIOAccounting", config_parse_bool, 0, &arg_default_blockio_accounting },
604 { "Manager", "DefaultMemoryAccounting", config_parse_bool, 0, &arg_default_memory_accounting },
605 { "Manager", "DefaultTasksAccounting", config_parse_bool, 0, &arg_default_tasks_accounting },
606 { "Manager", "DefaultTasksMax", config_parse_tasks_max, 0, &arg_default_tasks_max },
607 { "Manager", "CtrlAltDelBurstAction", config_parse_emergency_action, 0, &arg_cad_burst_action },
608 { "Manager", "DefaultOOMPolicy", config_parse_oom_policy, 0, &arg_default_oom_policy },
609 {}
610 };
611
612 const char *fn, *conf_dirs_nulstr;
613
614 fn = arg_system ?
615 PKGSYSCONFDIR "/system.conf" :
616 PKGSYSCONFDIR "/user.conf";
617
618 conf_dirs_nulstr = arg_system ?
619 CONF_PATHS_NULSTR("systemd/system.conf.d") :
620 CONF_PATHS_NULSTR("systemd/user.conf.d");
621
622 (void) config_parse_many_nulstr(fn, conf_dirs_nulstr, "Manager\0", config_item_table_lookup, items, CONFIG_PARSE_WARN, NULL);
623
624 /* Traditionally "0" was used to turn off the default unit timeouts. Fix this up so that we used USEC_INFINITY
625 * like everywhere else. */
626 if (arg_default_timeout_start_usec <= 0)
627 arg_default_timeout_start_usec = USEC_INFINITY;
628 if (arg_default_timeout_stop_usec <= 0)
629 arg_default_timeout_stop_usec = USEC_INFINITY;
630
631 return 0;
632 }
633
634 static void set_manager_defaults(Manager *m) {
635
636 assert(m);
637
638 /* Propagates the various default unit property settings into the manager object, i.e. properties that do not
639 * affect the manager itself, but are just what newly allocated units will have set if they haven't set
640 * anything else. (Also see set_manager_settings() for the settings that affect the manager's own behaviour) */
641
642 m->default_timer_accuracy_usec = arg_default_timer_accuracy_usec;
643 m->default_std_output = arg_default_std_output;
644 m->default_std_error = arg_default_std_error;
645 m->default_timeout_start_usec = arg_default_timeout_start_usec;
646 m->default_timeout_stop_usec = arg_default_timeout_stop_usec;
647 m->default_timeout_abort_usec = arg_default_timeout_abort_usec;
648 m->default_timeout_abort_set = arg_default_timeout_abort_set;
649 m->default_restart_usec = arg_default_restart_usec;
650 m->default_start_limit_interval = arg_default_start_limit_interval;
651 m->default_start_limit_burst = arg_default_start_limit_burst;
652
653 /* On 4.15+ with unified hierarchy, CPU accounting is essentially free as it doesn't require the CPU
654 * controller to be enabled, so the default is to enable it unless we got told otherwise. */
655 if (arg_default_cpu_accounting >= 0)
656 m->default_cpu_accounting = arg_default_cpu_accounting;
657 else
658 m->default_cpu_accounting = cpu_accounting_is_cheap();
659
660 m->default_io_accounting = arg_default_io_accounting;
661 m->default_ip_accounting = arg_default_ip_accounting;
662 m->default_blockio_accounting = arg_default_blockio_accounting;
663 m->default_memory_accounting = arg_default_memory_accounting;
664 m->default_tasks_accounting = arg_default_tasks_accounting;
665 m->default_tasks_max = arg_default_tasks_max;
666 m->default_oom_policy = arg_default_oom_policy;
667
668 (void) manager_set_default_rlimits(m, arg_default_rlimit);
669
670 (void) manager_default_environment(m);
671 (void) manager_transient_environment_add(m, arg_default_environment);
672 }
673
674 static void set_manager_settings(Manager *m) {
675
676 assert(m);
677
678 /* Propagates the various manager settings into the manager object, i.e. properties that effect the manager
679 * itself (as opposed to just being inherited into newly allocated units, see set_manager_defaults() above). */
680
681 m->confirm_spawn = arg_confirm_spawn;
682 m->service_watchdogs = arg_service_watchdogs;
683 m->runtime_watchdog = arg_runtime_watchdog;
684 m->reboot_watchdog = arg_reboot_watchdog;
685 m->kexec_watchdog = arg_kexec_watchdog;
686 m->cad_burst_action = arg_cad_burst_action;
687
688 manager_set_show_status(m, arg_show_status);
689 m->status_unit_format = arg_status_unit_format;
690 }
691
692 static int parse_argv(int argc, char *argv[]) {
693 enum {
694 ARG_LOG_LEVEL = 0x100,
695 ARG_LOG_TARGET,
696 ARG_LOG_COLOR,
697 ARG_LOG_LOCATION,
698 ARG_UNIT,
699 ARG_SYSTEM,
700 ARG_USER,
701 ARG_TEST,
702 ARG_NO_PAGER,
703 ARG_VERSION,
704 ARG_DUMP_CONFIGURATION_ITEMS,
705 ARG_DUMP_BUS_PROPERTIES,
706 ARG_DUMP_CORE,
707 ARG_CRASH_CHVT,
708 ARG_CRASH_SHELL,
709 ARG_CRASH_REBOOT,
710 ARG_CONFIRM_SPAWN,
711 ARG_SHOW_STATUS,
712 ARG_DESERIALIZE,
713 ARG_SWITCHED_ROOT,
714 ARG_DEFAULT_STD_OUTPUT,
715 ARG_DEFAULT_STD_ERROR,
716 ARG_MACHINE_ID,
717 ARG_SERVICE_WATCHDOGS,
718 };
719
720 static const struct option options[] = {
721 { "log-level", required_argument, NULL, ARG_LOG_LEVEL },
722 { "log-target", required_argument, NULL, ARG_LOG_TARGET },
723 { "log-color", optional_argument, NULL, ARG_LOG_COLOR },
724 { "log-location", optional_argument, NULL, ARG_LOG_LOCATION },
725 { "unit", required_argument, NULL, ARG_UNIT },
726 { "system", no_argument, NULL, ARG_SYSTEM },
727 { "user", no_argument, NULL, ARG_USER },
728 { "test", no_argument, NULL, ARG_TEST },
729 { "no-pager", no_argument, NULL, ARG_NO_PAGER },
730 { "help", no_argument, NULL, 'h' },
731 { "version", no_argument, NULL, ARG_VERSION },
732 { "dump-configuration-items", no_argument, NULL, ARG_DUMP_CONFIGURATION_ITEMS },
733 { "dump-bus-properties", no_argument, NULL, ARG_DUMP_BUS_PROPERTIES },
734 { "dump-core", optional_argument, NULL, ARG_DUMP_CORE },
735 { "crash-chvt", required_argument, NULL, ARG_CRASH_CHVT },
736 { "crash-shell", optional_argument, NULL, ARG_CRASH_SHELL },
737 { "crash-reboot", optional_argument, NULL, ARG_CRASH_REBOOT },
738 { "confirm-spawn", optional_argument, NULL, ARG_CONFIRM_SPAWN },
739 { "show-status", optional_argument, NULL, ARG_SHOW_STATUS },
740 { "deserialize", required_argument, NULL, ARG_DESERIALIZE },
741 { "switched-root", no_argument, NULL, ARG_SWITCHED_ROOT },
742 { "default-standard-output", required_argument, NULL, ARG_DEFAULT_STD_OUTPUT, },
743 { "default-standard-error", required_argument, NULL, ARG_DEFAULT_STD_ERROR, },
744 { "machine-id", required_argument, NULL, ARG_MACHINE_ID },
745 { "service-watchdogs", required_argument, NULL, ARG_SERVICE_WATCHDOGS },
746 {}
747 };
748
749 int c, r;
750
751 assert(argc >= 1);
752 assert(argv);
753
754 if (getpid_cached() == 1)
755 opterr = 0;
756
757 while ((c = getopt_long(argc, argv, "hDbsz:", options, NULL)) >= 0)
758
759 switch (c) {
760
761 case ARG_LOG_LEVEL:
762 r = log_set_max_level_from_string(optarg);
763 if (r < 0)
764 return log_error_errno(r, "Failed to parse log level \"%s\": %m", optarg);
765
766 break;
767
768 case ARG_LOG_TARGET:
769 r = log_set_target_from_string(optarg);
770 if (r < 0)
771 return log_error_errno(r, "Failed to parse log target \"%s\": %m", optarg);
772
773 break;
774
775 case ARG_LOG_COLOR:
776
777 if (optarg) {
778 r = log_show_color_from_string(optarg);
779 if (r < 0)
780 return log_error_errno(r, "Failed to parse log color setting \"%s\": %m",
781 optarg);
782 } else
783 log_show_color(true);
784
785 break;
786
787 case ARG_LOG_LOCATION:
788 if (optarg) {
789 r = log_show_location_from_string(optarg);
790 if (r < 0)
791 return log_error_errno(r, "Failed to parse log location setting \"%s\": %m",
792 optarg);
793 } else
794 log_show_location(true);
795
796 break;
797
798 case ARG_DEFAULT_STD_OUTPUT:
799 r = exec_output_from_string(optarg);
800 if (r < 0)
801 return log_error_errno(r, "Failed to parse default standard output setting \"%s\": %m",
802 optarg);
803 arg_default_std_output = r;
804 break;
805
806 case ARG_DEFAULT_STD_ERROR:
807 r = exec_output_from_string(optarg);
808 if (r < 0)
809 return log_error_errno(r, "Failed to parse default standard error output setting \"%s\": %m",
810 optarg);
811 arg_default_std_error = r;
812 break;
813
814 case ARG_UNIT:
815 r = free_and_strdup(&arg_default_unit, optarg);
816 if (r < 0)
817 return log_error_errno(r, "Failed to set default unit \"%s\": %m", optarg);
818
819 break;
820
821 case ARG_SYSTEM:
822 arg_system = true;
823 break;
824
825 case ARG_USER:
826 arg_system = false;
827 break;
828
829 case ARG_TEST:
830 arg_action = ACTION_TEST;
831 break;
832
833 case ARG_NO_PAGER:
834 arg_pager_flags |= PAGER_DISABLE;
835 break;
836
837 case ARG_VERSION:
838 arg_action = ACTION_VERSION;
839 break;
840
841 case ARG_DUMP_CONFIGURATION_ITEMS:
842 arg_action = ACTION_DUMP_CONFIGURATION_ITEMS;
843 break;
844
845 case ARG_DUMP_BUS_PROPERTIES:
846 arg_action = ACTION_DUMP_BUS_PROPERTIES;
847 break;
848
849 case ARG_DUMP_CORE:
850 if (!optarg)
851 arg_dump_core = true;
852 else {
853 r = parse_boolean(optarg);
854 if (r < 0)
855 return log_error_errno(r, "Failed to parse dump core boolean: \"%s\": %m",
856 optarg);
857 arg_dump_core = r;
858 }
859 break;
860
861 case ARG_CRASH_CHVT:
862 r = parse_crash_chvt(optarg, &arg_crash_chvt);
863 if (r < 0)
864 return log_error_errno(r, "Failed to parse crash virtual terminal index: \"%s\": %m",
865 optarg);
866 break;
867
868 case ARG_CRASH_SHELL:
869 if (!optarg)
870 arg_crash_shell = true;
871 else {
872 r = parse_boolean(optarg);
873 if (r < 0)
874 return log_error_errno(r, "Failed to parse crash shell boolean: \"%s\": %m",
875 optarg);
876 arg_crash_shell = r;
877 }
878 break;
879
880 case ARG_CRASH_REBOOT:
881 if (!optarg)
882 arg_crash_reboot = true;
883 else {
884 r = parse_boolean(optarg);
885 if (r < 0)
886 return log_error_errno(r, "Failed to parse crash shell boolean: \"%s\": %m",
887 optarg);
888 arg_crash_reboot = r;
889 }
890 break;
891
892 case ARG_CONFIRM_SPAWN:
893 arg_confirm_spawn = mfree(arg_confirm_spawn);
894
895 r = parse_confirm_spawn(optarg, &arg_confirm_spawn);
896 if (r < 0)
897 return log_error_errno(r, "Failed to parse confirm spawn option: \"%s\": %m",
898 optarg);
899 break;
900
901 case ARG_SERVICE_WATCHDOGS:
902 r = parse_boolean(optarg);
903 if (r < 0)
904 return log_error_errno(r, "Failed to parse service watchdogs boolean: \"%s\": %m",
905 optarg);
906 arg_service_watchdogs = r;
907 break;
908
909 case ARG_SHOW_STATUS:
910 if (optarg) {
911 r = parse_show_status(optarg, &arg_show_status);
912 if (r < 0)
913 return log_error_errno(r, "Failed to parse show status boolean: \"%s\": %m",
914 optarg);
915 } else
916 arg_show_status = SHOW_STATUS_YES;
917 break;
918
919 case ARG_DESERIALIZE: {
920 int fd;
921 FILE *f;
922
923 r = safe_atoi(optarg, &fd);
924 if (r < 0)
925 log_error_errno(r, "Failed to parse deserialize option \"%s\": %m", optarg);
926 if (fd < 0)
927 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
928 "Invalid deserialize fd: %d",
929 fd);
930
931 (void) fd_cloexec(fd, true);
932
933 f = fdopen(fd, "r");
934 if (!f)
935 return log_error_errno(errno, "Failed to open serialization fd %d: %m", fd);
936
937 safe_fclose(arg_serialization);
938 arg_serialization = f;
939
940 break;
941 }
942
943 case ARG_SWITCHED_ROOT:
944 arg_switched_root = true;
945 break;
946
947 case ARG_MACHINE_ID:
948 r = set_machine_id(optarg);
949 if (r < 0)
950 return log_error_errno(r, "MachineID '%s' is not valid: %m", optarg);
951 break;
952
953 case 'h':
954 arg_action = ACTION_HELP;
955 break;
956
957 case 'D':
958 log_set_max_level(LOG_DEBUG);
959 break;
960
961 case 'b':
962 case 's':
963 case 'z':
964 /* Just to eat away the sysvinit kernel
965 * cmdline args without getopt() error
966 * messages that we'll parse in
967 * parse_proc_cmdline_word() or ignore. */
968
969 case '?':
970 if (getpid_cached() != 1)
971 return -EINVAL;
972 else
973 return 0;
974
975 default:
976 assert_not_reached("Unhandled option code.");
977 }
978
979 if (optind < argc && getpid_cached() != 1) {
980 /* Hmm, when we aren't run as init system
981 * let's complain about excess arguments */
982
983 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
984 "Excess arguments.");
985 }
986
987 return 0;
988 }
989
990 static int help(void) {
991 _cleanup_free_ char *link = NULL;
992 int r;
993
994 r = terminal_urlify_man("systemd", "1", &link);
995 if (r < 0)
996 return log_oom();
997
998 printf("%s [OPTIONS...]\n\n"
999 "Starts up and maintains the system or user services.\n\n"
1000 " -h --help Show this help\n"
1001 " --version Show version\n"
1002 " --test Determine initial transaction, dump it and exit\n"
1003 " --system In combination with --test: operate as system service manager\n"
1004 " --user In combination with --test: operate as per-user service manager\n"
1005 " --no-pager Do not pipe output into a pager\n"
1006 " --dump-configuration-items Dump understood unit configuration items\n"
1007 " --dump-bus-properties Dump exposed bus properties\n"
1008 " --unit=UNIT Set default unit\n"
1009 " --dump-core[=BOOL] Dump core on crash\n"
1010 " --crash-vt=NR Change to specified VT on crash\n"
1011 " --crash-reboot[=BOOL] Reboot on crash\n"
1012 " --crash-shell[=BOOL] Run shell on crash\n"
1013 " --confirm-spawn[=BOOL] Ask for confirmation when spawning processes\n"
1014 " --show-status[=BOOL] Show status updates on the console during bootup\n"
1015 " --log-target=TARGET Set log target (console, journal, kmsg, journal-or-kmsg, null)\n"
1016 " --log-level=LEVEL Set log level (debug, info, notice, warning, err, crit, alert, emerg)\n"
1017 " --log-color[=BOOL] Highlight important log messages\n"
1018 " --log-location[=BOOL] Include code location in log messages\n"
1019 " --default-standard-output= Set default standard output for services\n"
1020 " --default-standard-error= Set default standard error output for services\n"
1021 "\nSee the %s for details.\n"
1022 , program_invocation_short_name
1023 , link
1024 );
1025
1026 return 0;
1027 }
1028
1029 static int prepare_reexecute(
1030 Manager *m,
1031 FILE **ret_f,
1032 FDSet **ret_fds,
1033 bool switching_root) {
1034
1035 _cleanup_fdset_free_ FDSet *fds = NULL;
1036 _cleanup_fclose_ FILE *f = NULL;
1037 int r;
1038
1039 assert(m);
1040 assert(ret_f);
1041 assert(ret_fds);
1042
1043 r = manager_open_serialization(m, &f);
1044 if (r < 0)
1045 return log_error_errno(r, "Failed to create serialization file: %m");
1046
1047 /* Make sure nothing is really destructed when we shut down */
1048 m->n_reloading++;
1049 bus_manager_send_reloading(m, true);
1050
1051 fds = fdset_new();
1052 if (!fds)
1053 return log_oom();
1054
1055 r = manager_serialize(m, f, fds, switching_root);
1056 if (r < 0)
1057 return r;
1058
1059 if (fseeko(f, 0, SEEK_SET) == (off_t) -1)
1060 return log_error_errno(errno, "Failed to rewind serialization fd: %m");
1061
1062 r = fd_cloexec(fileno(f), false);
1063 if (r < 0)
1064 return log_error_errno(r, "Failed to disable O_CLOEXEC for serialization: %m");
1065
1066 r = fdset_cloexec(fds, false);
1067 if (r < 0)
1068 return log_error_errno(r, "Failed to disable O_CLOEXEC for serialization fds: %m");
1069
1070 *ret_f = TAKE_PTR(f);
1071 *ret_fds = TAKE_PTR(fds);
1072
1073 return 0;
1074 }
1075
1076 static void bump_file_max_and_nr_open(void) {
1077
1078 /* Let's bump fs.file-max and fs.nr_open to their respective maximums. On current kernels large numbers of file
1079 * descriptors are no longer a performance problem and their memory is properly tracked by memcg, thus counting
1080 * them and limiting them in another two layers of limits is unnecessary and just complicates things. This
1081 * function hence turns off 2 of the 4 levels of limits on file descriptors, and makes RLIMIT_NOLIMIT (soft +
1082 * hard) the only ones that really matter. */
1083
1084 #if BUMP_PROC_SYS_FS_FILE_MAX || BUMP_PROC_SYS_FS_NR_OPEN
1085 int r;
1086 #endif
1087
1088 #if BUMP_PROC_SYS_FS_FILE_MAX
1089 /* The maximum the kernel allows for this since 5.2 is LONG_MAX, use that. (Previously thing where
1090 * different but the operation would fail silently.) */
1091 r = sysctl_writef("fs/file-max", "%li\n", LONG_MAX);
1092 if (r < 0)
1093 log_full_errno(IN_SET(r, -EROFS, -EPERM, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, "Failed to bump fs.file-max, ignoring: %m");
1094 #endif
1095
1096 #if BUMP_PROC_SYS_FS_NR_OPEN
1097 int v = INT_MAX;
1098
1099 /* Arg! The kernel enforces maximum and minimum values on the fs.nr_open, but we don't really know what they
1100 * are. The expression by which the maximum is determined is dependent on the architecture, and is something we
1101 * don't really want to copy to userspace, as it is dependent on implementation details of the kernel. Since
1102 * the kernel doesn't expose the maximum value to us, we can only try and hope. Hence, let's start with
1103 * INT_MAX, and then keep halving the value until we find one that works. Ugly? Yes, absolutely, but kernel
1104 * APIs are kernel APIs, so what do can we do... 🤯 */
1105
1106 for (;;) {
1107 int k;
1108
1109 v &= ~(__SIZEOF_POINTER__ - 1); /* Round down to next multiple of the pointer size */
1110 if (v < 1024) {
1111 log_warning("Can't bump fs.nr_open, value too small.");
1112 break;
1113 }
1114
1115 k = read_nr_open();
1116 if (k < 0) {
1117 log_error_errno(k, "Failed to read fs.nr_open: %m");
1118 break;
1119 }
1120 if (k >= v) { /* Already larger */
1121 log_debug("Skipping bump, value is already larger.");
1122 break;
1123 }
1124
1125 r = sysctl_writef("fs/nr_open", "%i\n", v);
1126 if (r == -EINVAL) {
1127 log_debug("Couldn't write fs.nr_open as %i, halving it.", v);
1128 v /= 2;
1129 continue;
1130 }
1131 if (r < 0) {
1132 log_full_errno(IN_SET(r, -EROFS, -EPERM, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, "Failed to bump fs.nr_open, ignoring: %m");
1133 break;
1134 }
1135
1136 log_debug("Successfully bumped fs.nr_open to %i", v);
1137 break;
1138 }
1139 #endif
1140 }
1141
1142 static int bump_rlimit_nofile(struct rlimit *saved_rlimit) {
1143 struct rlimit new_rlimit;
1144 int r, nr;
1145
1146 /* Get the underlying absolute limit the kernel enforces */
1147 nr = read_nr_open();
1148
1149 /* Calculate the new limits to use for us. Never lower from what we inherited. */
1150 new_rlimit = (struct rlimit) {
1151 .rlim_cur = MAX((rlim_t) nr, saved_rlimit->rlim_cur),
1152 .rlim_max = MAX((rlim_t) nr, saved_rlimit->rlim_max),
1153 };
1154
1155 /* Shortcut if nothing changes. */
1156 if (saved_rlimit->rlim_max >= new_rlimit.rlim_max &&
1157 saved_rlimit->rlim_cur >= new_rlimit.rlim_cur) {
1158 log_debug("RLIMIT_NOFILE is already as high or higher than we need it, not bumping.");
1159 return 0;
1160 }
1161
1162 /* Bump up the resource limit for ourselves substantially, all the way to the maximum the kernel allows, for
1163 * both hard and soft. */
1164 r = setrlimit_closest(RLIMIT_NOFILE, &new_rlimit);
1165 if (r < 0)
1166 return log_warning_errno(r, "Setting RLIMIT_NOFILE failed, ignoring: %m");
1167
1168 return 0;
1169 }
1170
1171 static int bump_rlimit_memlock(struct rlimit *saved_rlimit) {
1172 struct rlimit new_rlimit;
1173 int r;
1174
1175 /* BPF_MAP_TYPE_LPM_TRIE bpf maps are charged against RLIMIT_MEMLOCK, even if we have CAP_IPC_LOCK which should
1176 * normally disable such checks. We need them to implement IPAccessAllow= and IPAccessDeny=, hence let's bump
1177 * the value high enough for our user. */
1178
1179 /* Using MAX() on resource limits only is safe if RLIM_INFINITY is > 0. POSIX declares that rlim_t
1180 * must be unsigned, hence this is a given, but let's make this clear here. */
1181 assert_cc(RLIM_INFINITY > 0);
1182
1183 new_rlimit = (struct rlimit) {
1184 .rlim_cur = MAX(HIGH_RLIMIT_MEMLOCK, saved_rlimit->rlim_cur),
1185 .rlim_max = MAX(HIGH_RLIMIT_MEMLOCK, saved_rlimit->rlim_max),
1186 };
1187
1188 if (saved_rlimit->rlim_max >= new_rlimit.rlim_cur &&
1189 saved_rlimit->rlim_cur >= new_rlimit.rlim_max) {
1190 log_debug("RLIMIT_MEMLOCK is already as high or higher than we need it, not bumping.");
1191 return 0;
1192 }
1193
1194 r = setrlimit_closest(RLIMIT_MEMLOCK, &new_rlimit);
1195 if (r < 0)
1196 return log_warning_errno(r, "Setting RLIMIT_MEMLOCK failed, ignoring: %m");
1197
1198 return 0;
1199 }
1200
1201 static void test_usr(void) {
1202
1203 /* Check that /usr is either on the same file system as / or mounted already. */
1204
1205 if (dir_is_empty("/usr") <= 0)
1206 return;
1207
1208 log_warning("/usr appears to be on its own filesystem and is not already mounted. This is not a supported setup. "
1209 "Some things will probably break (sometimes even silently) in mysterious ways. "
1210 "Consult http://freedesktop.org/wiki/Software/systemd/separate-usr-is-broken for more information.");
1211 }
1212
1213 static int enforce_syscall_archs(Set *archs) {
1214 #if HAVE_SECCOMP
1215 int r;
1216
1217 if (!is_seccomp_available())
1218 return 0;
1219
1220 r = seccomp_restrict_archs(arg_syscall_archs);
1221 if (r < 0)
1222 return log_error_errno(r, "Failed to enforce system call architecture restrication: %m");
1223 #endif
1224 return 0;
1225 }
1226
1227 static int status_welcome(void) {
1228 _cleanup_free_ char *pretty_name = NULL, *ansi_color = NULL;
1229 int r;
1230
1231 if (IN_SET(arg_show_status, SHOW_STATUS_NO, SHOW_STATUS_AUTO))
1232 return 0;
1233
1234 r = parse_os_release(NULL,
1235 "PRETTY_NAME", &pretty_name,
1236 "ANSI_COLOR", &ansi_color,
1237 NULL);
1238 if (r < 0)
1239 log_full_errno(r == -ENOENT ? LOG_DEBUG : LOG_WARNING, r,
1240 "Failed to read os-release file, ignoring: %m");
1241
1242 if (log_get_show_color())
1243 return status_printf(NULL, 0,
1244 "\nWelcome to \x1B[%sm%s\x1B[0m!\n",
1245 isempty(ansi_color) ? "1" : ansi_color,
1246 isempty(pretty_name) ? "Linux" : pretty_name);
1247 else
1248 return status_printf(NULL, 0,
1249 "\nWelcome to %s!\n",
1250 isempty(pretty_name) ? "Linux" : pretty_name);
1251 }
1252
1253 static int write_container_id(void) {
1254 const char *c;
1255 int r;
1256
1257 c = getenv("container");
1258 if (isempty(c))
1259 return 0;
1260
1261 RUN_WITH_UMASK(0022)
1262 r = write_string_file("/run/systemd/container", c, WRITE_STRING_FILE_CREATE);
1263 if (r < 0)
1264 return log_warning_errno(r, "Failed to write /run/systemd/container, ignoring: %m");
1265
1266 return 1;
1267 }
1268
1269 static int bump_unix_max_dgram_qlen(void) {
1270 _cleanup_free_ char *qlen = NULL;
1271 unsigned long v;
1272 int r;
1273
1274 /* Let's bump the net.unix.max_dgram_qlen sysctl. The kernel default of 16 is simply too low. We set the value
1275 * really really early during boot, so that it is actually applied to all our sockets, including the
1276 * $NOTIFY_SOCKET one. */
1277
1278 r = read_one_line_file("/proc/sys/net/unix/max_dgram_qlen", &qlen);
1279 if (r < 0)
1280 return log_full_errno(r == -ENOENT ? LOG_DEBUG : LOG_WARNING, r, "Failed to read AF_UNIX datagram queue length, ignoring: %m");
1281
1282 r = safe_atolu(qlen, &v);
1283 if (r < 0)
1284 return log_warning_errno(r, "Failed to parse AF_UNIX datagram queue length '%s', ignoring: %m", qlen);
1285
1286 if (v >= DEFAULT_UNIX_MAX_DGRAM_QLEN)
1287 return 0;
1288
1289 r = write_string_filef("/proc/sys/net/unix/max_dgram_qlen", WRITE_STRING_FILE_DISABLE_BUFFER, "%lu", DEFAULT_UNIX_MAX_DGRAM_QLEN);
1290 if (r < 0)
1291 return log_full_errno(IN_SET(r, -EROFS, -EPERM, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
1292 "Failed to bump AF_UNIX datagram queue length, ignoring: %m");
1293
1294 return 1;
1295 }
1296
1297 static int fixup_environment(void) {
1298 _cleanup_free_ char *term = NULL;
1299 const char *t;
1300 int r;
1301
1302 /* Only fix up the environment when we are started as PID 1 */
1303 if (getpid_cached() != 1)
1304 return 0;
1305
1306 /* We expect the environment to be set correctly if run inside a container. */
1307 if (detect_container() > 0)
1308 return 0;
1309
1310 /* When started as PID1, the kernel uses /dev/console for our stdios and uses TERM=linux whatever the backend
1311 * device used by the console. We try to make a better guess here since some consoles might not have support
1312 * for color mode for example.
1313 *
1314 * However if TERM was configured through the kernel command line then leave it alone. */
1315 r = proc_cmdline_get_key("TERM", 0, &term);
1316 if (r < 0)
1317 return r;
1318
1319 t = term ?: default_term_for_tty("/dev/console");
1320
1321 if (setenv("TERM", t, 1) < 0)
1322 return -errno;
1323
1324 /* The kernels sets HOME=/ for init. Let's undo this. */
1325 if (path_equal_ptr(getenv("HOME"), "/") &&
1326 unsetenv("HOME") < 0)
1327 log_warning_errno(errno, "Failed to unset $HOME: %m");
1328
1329 return 0;
1330 }
1331
1332 static void redirect_telinit(int argc, char *argv[]) {
1333
1334 /* This is compatibility support for SysV, where calling init as a user is identical to telinit. */
1335
1336 #if HAVE_SYSV_COMPAT
1337 if (getpid_cached() == 1)
1338 return;
1339
1340 if (!strstr(program_invocation_short_name, "init"))
1341 return;
1342
1343 execv(SYSTEMCTL_BINARY_PATH, argv);
1344 log_error_errno(errno, "Failed to exec " SYSTEMCTL_BINARY_PATH ": %m");
1345 exit(EXIT_FAILURE);
1346 #endif
1347 }
1348
1349 static int become_shutdown(
1350 const char *shutdown_verb,
1351 int retval) {
1352
1353 char log_level[DECIMAL_STR_MAX(int) + 1],
1354 exit_code[DECIMAL_STR_MAX(uint8_t) + 1],
1355 timeout[DECIMAL_STR_MAX(usec_t) + 1];
1356
1357 const char* command_line[13] = {
1358 SYSTEMD_SHUTDOWN_BINARY_PATH,
1359 shutdown_verb,
1360 "--timeout", timeout,
1361 "--log-level", log_level,
1362 "--log-target",
1363 };
1364
1365 _cleanup_strv_free_ char **env_block = NULL;
1366 size_t pos = 7;
1367 int r;
1368 usec_t watchdog_timer = 0;
1369
1370 assert(shutdown_verb);
1371 assert(!command_line[pos]);
1372 env_block = strv_copy(environ);
1373
1374 xsprintf(log_level, "%d", log_get_max_level());
1375 xsprintf(timeout, "%" PRI_USEC "us", arg_default_timeout_stop_usec);
1376
1377 switch (log_get_target()) {
1378
1379 case LOG_TARGET_KMSG:
1380 case LOG_TARGET_JOURNAL_OR_KMSG:
1381 case LOG_TARGET_SYSLOG_OR_KMSG:
1382 command_line[pos++] = "kmsg";
1383 break;
1384
1385 case LOG_TARGET_NULL:
1386 command_line[pos++] = "null";
1387 break;
1388
1389 case LOG_TARGET_CONSOLE:
1390 default:
1391 command_line[pos++] = "console";
1392 break;
1393 };
1394
1395 if (log_get_show_color())
1396 command_line[pos++] = "--log-color";
1397
1398 if (log_get_show_location())
1399 command_line[pos++] = "--log-location";
1400
1401 if (streq(shutdown_verb, "exit")) {
1402 command_line[pos++] = "--exit-code";
1403 command_line[pos++] = exit_code;
1404 xsprintf(exit_code, "%d", retval);
1405 }
1406
1407 assert(pos < ELEMENTSOF(command_line));
1408
1409 if (streq(shutdown_verb, "reboot"))
1410 watchdog_timer = arg_reboot_watchdog;
1411 else if (streq(shutdown_verb, "kexec"))
1412 watchdog_timer = arg_kexec_watchdog;
1413
1414 if (watchdog_timer > 0 && watchdog_timer != USEC_INFINITY) {
1415
1416 char *e;
1417
1418 /* If we reboot or kexec let's set the shutdown
1419 * watchdog and tell the shutdown binary to
1420 * repeatedly ping it */
1421 r = watchdog_set_timeout(&watchdog_timer);
1422 watchdog_close(r < 0);
1423
1424 /* Tell the binary how often to ping, ignore failure */
1425 if (asprintf(&e, "WATCHDOG_USEC="USEC_FMT, watchdog_timer) > 0)
1426 (void) strv_consume(&env_block, e);
1427
1428 if (arg_watchdog_device &&
1429 asprintf(&e, "WATCHDOG_DEVICE=%s", arg_watchdog_device) > 0)
1430 (void) strv_consume(&env_block, e);
1431 } else
1432 watchdog_close(true);
1433
1434 /* Avoid the creation of new processes forked by the
1435 * kernel; at this point, we will not listen to the
1436 * signals anyway */
1437 if (detect_container() <= 0)
1438 (void) cg_uninstall_release_agent(SYSTEMD_CGROUP_CONTROLLER);
1439
1440 execve(SYSTEMD_SHUTDOWN_BINARY_PATH, (char **) command_line, env_block);
1441 return -errno;
1442 }
1443
1444 static void initialize_clock(void) {
1445 int r;
1446
1447 if (clock_is_localtime(NULL) > 0) {
1448 int min;
1449
1450 /*
1451 * The very first call of settimeofday() also does a time warp in the kernel.
1452 *
1453 * In the rtc-in-local time mode, we set the kernel's timezone, and rely on external tools to take care
1454 * of maintaining the RTC and do all adjustments. This matches the behavior of Windows, which leaves
1455 * the RTC alone if the registry tells that the RTC runs in UTC.
1456 */
1457 r = clock_set_timezone(&min);
1458 if (r < 0)
1459 log_error_errno(r, "Failed to apply local time delta, ignoring: %m");
1460 else
1461 log_info("RTC configured in localtime, applying delta of %i minutes to system time.", min);
1462
1463 } else if (!in_initrd()) {
1464 /*
1465 * Do a dummy very first call to seal the kernel's time warp magic.
1466 *
1467 * Do not call this from inside the initrd. The initrd might not carry /etc/adjtime with LOCAL, but the
1468 * real system could be set up that way. In such case, we need to delay the time-warp or the sealing
1469 * until we reach the real system.
1470 *
1471 * Do no set the kernel's timezone. The concept of local time cannot be supported reliably, the time
1472 * will jump or be incorrect at every daylight saving time change. All kernel local time concepts will
1473 * be treated as UTC that way.
1474 */
1475 (void) clock_reset_timewarp();
1476 }
1477
1478 r = clock_apply_epoch();
1479 if (r < 0)
1480 log_error_errno(r, "Current system time is before build time, but cannot correct: %m");
1481 else if (r > 0)
1482 log_info("System time before build time, advancing clock.");
1483 }
1484
1485 static void initialize_coredump(bool skip_setup) {
1486 #if ENABLE_COREDUMP
1487 if (getpid_cached() != 1)
1488 return;
1489
1490 /* Don't limit the core dump size, so that coredump handlers such as systemd-coredump (which honour the limit)
1491 * will process core dumps for system services by default. */
1492 if (setrlimit(RLIMIT_CORE, &RLIMIT_MAKE_CONST(RLIM_INFINITY)) < 0)
1493 log_warning_errno(errno, "Failed to set RLIMIT_CORE: %m");
1494
1495 /* But at the same time, turn off the core_pattern logic by default, so that no
1496 * coredumps are stored until the systemd-coredump tool is enabled via
1497 * sysctl. However it can be changed via the kernel command line later so core
1498 * dumps can still be generated during early startup and in initramfs. */
1499 if (!skip_setup)
1500 disable_coredumps();
1501 #endif
1502 }
1503
1504 static void initialize_core_pattern(bool skip_setup) {
1505 int r;
1506
1507 if (skip_setup || !arg_early_core_pattern)
1508 return;
1509
1510 if (getpid_cached() != 1)
1511 return;
1512
1513 r = write_string_file("/proc/sys/kernel/core_pattern", arg_early_core_pattern, WRITE_STRING_FILE_DISABLE_BUFFER);
1514 if (r < 0)
1515 log_warning_errno(r, "Failed to write '%s' to /proc/sys/kernel/core_pattern, ignoring: %m", arg_early_core_pattern);
1516 }
1517
1518 static void update_cpu_affinity(bool skip_setup) {
1519 _cleanup_free_ char *mask = NULL;
1520
1521 if (skip_setup || !arg_cpu_affinity.set)
1522 return;
1523
1524 assert(arg_cpu_affinity.allocated > 0);
1525
1526 mask = cpu_set_to_string(&arg_cpu_affinity);
1527 log_debug("Setting CPU affinity to %s.", strnull(mask));
1528
1529 if (sched_setaffinity(0, arg_cpu_affinity.allocated, arg_cpu_affinity.set) < 0)
1530 log_warning_errno(errno, "Failed to set CPU affinity: %m");
1531 }
1532
1533 static void update_numa_policy(bool skip_setup) {
1534 int r;
1535 _cleanup_free_ char *nodes = NULL;
1536 const char * policy = NULL;
1537
1538 if (skip_setup || !mpol_is_valid(numa_policy_get_type(&arg_numa_policy)))
1539 return;
1540
1541 if (DEBUG_LOGGING) {
1542 policy = mpol_to_string(numa_policy_get_type(&arg_numa_policy));
1543 nodes = cpu_set_to_range_string(&arg_numa_policy.nodes);
1544 log_debug("Setting NUMA policy to %s, with nodes %s.", strnull(policy), strnull(nodes));
1545 }
1546
1547 r = apply_numa_policy(&arg_numa_policy);
1548 if (r == -EOPNOTSUPP)
1549 log_debug_errno(r, "NUMA support not available, ignoring.");
1550 else if (r < 0)
1551 log_warning_errno(r, "Failed to set NUMA memory policy: %m");
1552 }
1553
1554 static void do_reexecute(
1555 int argc,
1556 char *argv[],
1557 const struct rlimit *saved_rlimit_nofile,
1558 const struct rlimit *saved_rlimit_memlock,
1559 FDSet *fds,
1560 const char *switch_root_dir,
1561 const char *switch_root_init,
1562 const char **ret_error_message) {
1563
1564 unsigned i, j, args_size;
1565 const char **args;
1566 int r;
1567
1568 assert(saved_rlimit_nofile);
1569 assert(saved_rlimit_memlock);
1570 assert(ret_error_message);
1571
1572 /* Close and disarm the watchdog, so that the new instance can reinitialize it, but doesn't get rebooted while
1573 * we do that */
1574 watchdog_close(true);
1575
1576 /* Reset RLIMIT_NOFILE + RLIMIT_MEMLOCK back to the kernel defaults, so that the new systemd can pass
1577 * the kernel default to its child processes */
1578 if (saved_rlimit_nofile->rlim_cur != 0)
1579 (void) setrlimit(RLIMIT_NOFILE, saved_rlimit_nofile);
1580 if (saved_rlimit_memlock->rlim_cur != RLIM_INFINITY)
1581 (void) setrlimit(RLIMIT_MEMLOCK, saved_rlimit_memlock);
1582
1583 if (switch_root_dir) {
1584 /* Kill all remaining processes from the initrd, but don't wait for them, so that we can handle the
1585 * SIGCHLD for them after deserializing. */
1586 broadcast_signal(SIGTERM, false, true, arg_default_timeout_stop_usec);
1587
1588 /* And switch root with MS_MOVE, because we remove the old directory afterwards and detach it. */
1589 r = switch_root(switch_root_dir, "/mnt", true, MS_MOVE);
1590 if (r < 0)
1591 log_error_errno(r, "Failed to switch root, trying to continue: %m");
1592 }
1593
1594 args_size = MAX(6, argc+1);
1595 args = newa(const char*, args_size);
1596
1597 if (!switch_root_init) {
1598 char sfd[DECIMAL_STR_MAX(int) + 1];
1599
1600 /* First try to spawn ourselves with the right path, and with full serialization. We do this only if
1601 * the user didn't specify an explicit init to spawn. */
1602
1603 assert(arg_serialization);
1604 assert(fds);
1605
1606 xsprintf(sfd, "%i", fileno(arg_serialization));
1607
1608 i = 0;
1609 args[i++] = SYSTEMD_BINARY_PATH;
1610 if (switch_root_dir)
1611 args[i++] = "--switched-root";
1612 args[i++] = arg_system ? "--system" : "--user";
1613 args[i++] = "--deserialize";
1614 args[i++] = sfd;
1615 args[i++] = NULL;
1616
1617 assert(i <= args_size);
1618
1619 /*
1620 * We want valgrind to print its memory usage summary before reexecution. Valgrind won't do this is on
1621 * its own on exec(), but it will do it on exit(). Hence, to ensure we get a summary here, fork() off
1622 * a child, let it exit() cleanly, so that it prints the summary, and wait() for it in the parent,
1623 * before proceeding into the exec().
1624 */
1625 valgrind_summary_hack();
1626
1627 (void) execv(args[0], (char* const*) args);
1628 log_debug_errno(errno, "Failed to execute our own binary, trying fallback: %m");
1629 }
1630
1631 /* Try the fallback, if there is any, without any serialization. We pass the original argv[] and envp[]. (Well,
1632 * modulo the ordering changes due to getopt() in argv[], and some cleanups in envp[], but let's hope that
1633 * doesn't matter.) */
1634
1635 arg_serialization = safe_fclose(arg_serialization);
1636 fds = fdset_free(fds);
1637
1638 /* Reopen the console */
1639 (void) make_console_stdio();
1640
1641 for (j = 1, i = 1; j < (unsigned) argc; j++)
1642 args[i++] = argv[j];
1643 args[i++] = NULL;
1644 assert(i <= args_size);
1645
1646 /* Re-enable any blocked signals, especially important if we switch from initial ramdisk to init=... */
1647 (void) reset_all_signal_handlers();
1648 (void) reset_signal_mask();
1649 (void) rlimit_nofile_safe();
1650
1651 if (switch_root_init) {
1652 args[0] = switch_root_init;
1653 (void) execve(args[0], (char* const*) args, saved_env);
1654 log_warning_errno(errno, "Failed to execute configured init, trying fallback: %m");
1655 }
1656
1657 args[0] = "/sbin/init";
1658 (void) execv(args[0], (char* const*) args);
1659 r = -errno;
1660
1661 manager_status_printf(NULL, STATUS_TYPE_EMERGENCY,
1662 ANSI_HIGHLIGHT_RED " !! " ANSI_NORMAL,
1663 "Failed to execute /sbin/init");
1664
1665 if (r == -ENOENT) {
1666 log_warning("No /sbin/init, trying fallback");
1667
1668 args[0] = "/bin/sh";
1669 args[1] = NULL;
1670 (void) execve(args[0], (char* const*) args, saved_env);
1671 log_error_errno(errno, "Failed to execute /bin/sh, giving up: %m");
1672 } else
1673 log_warning_errno(r, "Failed to execute /sbin/init, giving up: %m");
1674
1675 *ret_error_message = "Failed to execute fallback shell";
1676 }
1677
1678 static int invoke_main_loop(
1679 Manager *m,
1680 const struct rlimit *saved_rlimit_nofile,
1681 const struct rlimit *saved_rlimit_memlock,
1682 bool *ret_reexecute,
1683 int *ret_retval, /* Return parameters relevant for shutting down */
1684 const char **ret_shutdown_verb, /* … */
1685 FDSet **ret_fds, /* Return parameters for reexecuting */
1686 char **ret_switch_root_dir, /* … */
1687 char **ret_switch_root_init, /* … */
1688 const char **ret_error_message) {
1689
1690 int r;
1691
1692 assert(m);
1693 assert(saved_rlimit_nofile);
1694 assert(saved_rlimit_memlock);
1695 assert(ret_reexecute);
1696 assert(ret_retval);
1697 assert(ret_shutdown_verb);
1698 assert(ret_fds);
1699 assert(ret_switch_root_dir);
1700 assert(ret_switch_root_init);
1701 assert(ret_error_message);
1702
1703 for (;;) {
1704 r = manager_loop(m);
1705 if (r < 0) {
1706 *ret_error_message = "Failed to run main loop";
1707 return log_emergency_errno(r, "Failed to run main loop: %m");
1708 }
1709
1710 switch ((ManagerObjective) r) {
1711
1712 case MANAGER_RELOAD: {
1713 LogTarget saved_log_target;
1714 int saved_log_level;
1715
1716 log_info("Reloading.");
1717
1718 /* First, save any overridden log level/target, then parse the configuration file, which might
1719 * change the log level to new settings. */
1720
1721 saved_log_level = m->log_level_overridden ? log_get_max_level() : -1;
1722 saved_log_target = m->log_target_overridden ? log_get_target() : _LOG_TARGET_INVALID;
1723
1724 (void) parse_configuration(saved_rlimit_nofile, saved_rlimit_memlock);
1725
1726 set_manager_defaults(m);
1727
1728 update_cpu_affinity(false);
1729 update_numa_policy(false);
1730
1731 if (saved_log_level >= 0)
1732 manager_override_log_level(m, saved_log_level);
1733 if (saved_log_target >= 0)
1734 manager_override_log_target(m, saved_log_target);
1735
1736 r = manager_reload(m);
1737 if (r < 0)
1738 /* Reloading failed before the point of no return. Let's continue running as if nothing happened. */
1739 m->objective = MANAGER_OK;
1740
1741 break;
1742 }
1743
1744 case MANAGER_REEXECUTE:
1745
1746 r = prepare_reexecute(m, &arg_serialization, ret_fds, false);
1747 if (r < 0) {
1748 *ret_error_message = "Failed to prepare for reexecution";
1749 return r;
1750 }
1751
1752 log_notice("Reexecuting.");
1753
1754 *ret_reexecute = true;
1755 *ret_retval = EXIT_SUCCESS;
1756 *ret_shutdown_verb = NULL;
1757 *ret_switch_root_dir = *ret_switch_root_init = NULL;
1758
1759 return 0;
1760
1761 case MANAGER_SWITCH_ROOT:
1762 if (!m->switch_root_init) {
1763 r = prepare_reexecute(m, &arg_serialization, ret_fds, true);
1764 if (r < 0) {
1765 *ret_error_message = "Failed to prepare for reexecution";
1766 return r;
1767 }
1768 } else
1769 *ret_fds = NULL;
1770
1771 log_notice("Switching root.");
1772
1773 *ret_reexecute = true;
1774 *ret_retval = EXIT_SUCCESS;
1775 *ret_shutdown_verb = NULL;
1776
1777 /* Steal the switch root parameters */
1778 *ret_switch_root_dir = TAKE_PTR(m->switch_root);
1779 *ret_switch_root_init = TAKE_PTR(m->switch_root_init);
1780
1781 return 0;
1782
1783 case MANAGER_EXIT:
1784
1785 if (MANAGER_IS_USER(m)) {
1786 log_debug("Exit.");
1787
1788 *ret_reexecute = false;
1789 *ret_retval = m->return_value;
1790 *ret_shutdown_verb = NULL;
1791 *ret_fds = NULL;
1792 *ret_switch_root_dir = *ret_switch_root_init = NULL;
1793
1794 return 0;
1795 }
1796
1797 _fallthrough_;
1798 case MANAGER_REBOOT:
1799 case MANAGER_POWEROFF:
1800 case MANAGER_HALT:
1801 case MANAGER_KEXEC: {
1802 static const char * const table[_MANAGER_OBJECTIVE_MAX] = {
1803 [MANAGER_EXIT] = "exit",
1804 [MANAGER_REBOOT] = "reboot",
1805 [MANAGER_POWEROFF] = "poweroff",
1806 [MANAGER_HALT] = "halt",
1807 [MANAGER_KEXEC] = "kexec",
1808 };
1809
1810 log_notice("Shutting down.");
1811
1812 *ret_reexecute = false;
1813 *ret_retval = m->return_value;
1814 assert_se(*ret_shutdown_verb = table[m->objective]);
1815 *ret_fds = NULL;
1816 *ret_switch_root_dir = *ret_switch_root_init = NULL;
1817
1818 return 0;
1819 }
1820
1821 default:
1822 assert_not_reached("Unknown or unexpected manager objective.");
1823 }
1824 }
1825 }
1826
1827 static void log_execution_mode(bool *ret_first_boot) {
1828 assert(ret_first_boot);
1829
1830 if (arg_system) {
1831 int v;
1832
1833 log_info("systemd " GIT_VERSION " running in %ssystem mode. (" SYSTEMD_FEATURES ")",
1834 arg_action == ACTION_TEST ? "test " : "" );
1835
1836 v = detect_virtualization();
1837 if (v > 0)
1838 log_info("Detected virtualization %s.", virtualization_to_string(v));
1839
1840 log_info("Detected architecture %s.", architecture_to_string(uname_architecture()));
1841
1842 if (in_initrd()) {
1843 *ret_first_boot = false;
1844 log_info("Running in initial RAM disk.");
1845 } else {
1846 /* Let's check whether we are in first boot, i.e. whether /etc is still unpopulated. We use
1847 * /etc/machine-id as flag file, for this: if it exists we assume /etc is populated, if it
1848 * doesn't it's unpopulated. This allows container managers and installers to provision a
1849 * couple of files already. If the container manager wants to provision the machine ID itself
1850 * it should pass $container_uuid to PID 1. */
1851
1852 *ret_first_boot = access("/etc/machine-id", F_OK) < 0;
1853 if (*ret_first_boot)
1854 log_info("Running with unpopulated /etc.");
1855 }
1856 } else {
1857 if (DEBUG_LOGGING) {
1858 _cleanup_free_ char *t;
1859
1860 t = uid_to_name(getuid());
1861 log_debug("systemd " GIT_VERSION " running in %suser mode for user " UID_FMT "/%s. (" SYSTEMD_FEATURES ")",
1862 arg_action == ACTION_TEST ? " test" : "", getuid(), strna(t));
1863 }
1864
1865 *ret_first_boot = false;
1866 }
1867 }
1868
1869 static int initialize_runtime(
1870 bool skip_setup,
1871 struct rlimit *saved_rlimit_nofile,
1872 struct rlimit *saved_rlimit_memlock,
1873 const char **ret_error_message) {
1874 int r;
1875
1876 assert(ret_error_message);
1877
1878 /* Sets up various runtime parameters. Many of these initializations are conditionalized:
1879 *
1880 * - Some only apply to --system instances
1881 * - Some only apply to --user instances
1882 * - Some only apply when we first start up, but not when we reexecute
1883 */
1884
1885 if (arg_action != ACTION_RUN)
1886 return 0;
1887
1888 update_cpu_affinity(skip_setup);
1889 update_numa_policy(skip_setup);
1890
1891 if (arg_system) {
1892 /* Make sure we leave a core dump without panicking the kernel. */
1893 install_crash_handler();
1894
1895 if (!skip_setup) {
1896 r = mount_cgroup_controllers();
1897 if (r < 0) {
1898 *ret_error_message = "Failed to mount cgroup hierarchies";
1899 return r;
1900 }
1901
1902 status_welcome();
1903 hostname_setup();
1904 machine_id_setup(NULL, arg_machine_id, NULL);
1905 loopback_setup();
1906 bump_unix_max_dgram_qlen();
1907 bump_file_max_and_nr_open();
1908 test_usr();
1909 write_container_id();
1910 }
1911
1912 if (arg_watchdog_device) {
1913 r = watchdog_set_device(arg_watchdog_device);
1914 if (r < 0)
1915 log_warning_errno(r, "Failed to set watchdog device to %s, ignoring: %m", arg_watchdog_device);
1916 }
1917
1918 if (timestamp_is_set(arg_runtime_watchdog))
1919 watchdog_set_timeout(&arg_runtime_watchdog);
1920 }
1921
1922 if (arg_timer_slack_nsec != NSEC_INFINITY)
1923 if (prctl(PR_SET_TIMERSLACK, arg_timer_slack_nsec) < 0)
1924 log_warning_errno(errno, "Failed to adjust timer slack, ignoring: %m");
1925
1926 if (arg_system && !cap_test_all(arg_capability_bounding_set)) {
1927 r = capability_bounding_set_drop_usermode(arg_capability_bounding_set);
1928 if (r < 0) {
1929 *ret_error_message = "Failed to drop capability bounding set of usermode helpers";
1930 return log_emergency_errno(r, "Failed to drop capability bounding set of usermode helpers: %m");
1931 }
1932
1933 r = capability_bounding_set_drop(arg_capability_bounding_set, true);
1934 if (r < 0) {
1935 *ret_error_message = "Failed to drop capability bounding set";
1936 return log_emergency_errno(r, "Failed to drop capability bounding set: %m");
1937 }
1938 }
1939
1940 if (arg_system && arg_no_new_privs) {
1941 if (prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0) < 0) {
1942 *ret_error_message = "Failed to disable new privileges";
1943 return log_emergency_errno(errno, "Failed to disable new privileges: %m");
1944 }
1945 }
1946
1947 if (arg_syscall_archs) {
1948 r = enforce_syscall_archs(arg_syscall_archs);
1949 if (r < 0) {
1950 *ret_error_message = "Failed to set syscall architectures";
1951 return r;
1952 }
1953 }
1954
1955 if (!arg_system)
1956 /* Become reaper of our children */
1957 if (prctl(PR_SET_CHILD_SUBREAPER, 1) < 0)
1958 log_warning_errno(errno, "Failed to make us a subreaper: %m");
1959
1960 /* Bump up RLIMIT_NOFILE for systemd itself */
1961 (void) bump_rlimit_nofile(saved_rlimit_nofile);
1962 (void) bump_rlimit_memlock(saved_rlimit_memlock);
1963
1964 return 0;
1965 }
1966
1967 static int do_queue_default_job(
1968 Manager *m,
1969 const char **ret_error_message) {
1970
1971 _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL;
1972 Job *default_unit_job;
1973 Unit *target = NULL;
1974 int r;
1975
1976 log_debug("Activating default unit: %s", arg_default_unit);
1977
1978 r = manager_load_startable_unit_or_warn(m, arg_default_unit, NULL, &target);
1979 if (r < 0) {
1980 log_info("Falling back to rescue target: " SPECIAL_RESCUE_TARGET);
1981
1982 r = manager_load_startable_unit_or_warn(m, SPECIAL_RESCUE_TARGET, NULL, &target);
1983 if (r < 0) {
1984 *ret_error_message = r == -ERFKILL ? "Rescue target masked"
1985 : "Failed to load rescue target";
1986 return r;
1987 }
1988 }
1989
1990 assert(target->load_state == UNIT_LOADED);
1991
1992 r = manager_add_job(m, JOB_START, target, JOB_ISOLATE, NULL, &error, &default_unit_job);
1993 if (r == -EPERM) {
1994 log_debug_errno(r, "Default target could not be isolated, starting instead: %s", bus_error_message(&error, r));
1995
1996 sd_bus_error_free(&error);
1997
1998 r = manager_add_job(m, JOB_START, target, JOB_REPLACE, NULL, &error, &default_unit_job);
1999 if (r < 0) {
2000 *ret_error_message = "Failed to start default target";
2001 return log_emergency_errno(r, "Failed to start default target: %s", bus_error_message(&error, r));
2002 }
2003
2004 } else if (r < 0) {
2005 *ret_error_message = "Failed to isolate default target";
2006 return log_emergency_errno(r, "Failed to isolate default target: %s", bus_error_message(&error, r));
2007 }
2008
2009 m->default_unit_job_id = default_unit_job->id;
2010
2011 return 0;
2012 }
2013
2014 static void save_rlimits(struct rlimit *saved_rlimit_nofile,
2015 struct rlimit *saved_rlimit_memlock) {
2016
2017 assert(saved_rlimit_nofile);
2018 assert(saved_rlimit_memlock);
2019
2020 if (getrlimit(RLIMIT_NOFILE, saved_rlimit_nofile) < 0)
2021 log_warning_errno(errno, "Reading RLIMIT_NOFILE failed, ignoring: %m");
2022
2023 if (getrlimit(RLIMIT_MEMLOCK, saved_rlimit_memlock) < 0)
2024 log_warning_errno(errno, "Reading RLIMIT_MEMLOCK failed, ignoring: %m");
2025 }
2026
2027 static void fallback_rlimit_nofile(const struct rlimit *saved_rlimit_nofile) {
2028 struct rlimit *rl;
2029
2030 if (arg_default_rlimit[RLIMIT_NOFILE])
2031 return;
2032
2033 /* Make sure forked processes get limits based on the original kernel setting */
2034
2035 rl = newdup(struct rlimit, saved_rlimit_nofile, 1);
2036 if (!rl) {
2037 log_oom();
2038 return;
2039 }
2040
2041 /* Bump the hard limit for system services to a substantially higher value. The default
2042 * hard limit current kernels set is pretty low (4K), mostly for historical
2043 * reasons. According to kernel developers, the fd handling in recent kernels has been
2044 * optimized substantially enough, so that we can bump the limit now, without paying too
2045 * high a price in memory or performance. Note however that we only bump the hard limit,
2046 * not the soft limit. That's because select() works the way it works, and chokes on fds
2047 * >= 1024. If we'd bump the soft limit globally, it might accidentally happen to
2048 * unexpecting programs that they get fds higher than what they can process using
2049 * select(). By only bumping the hard limit but leaving the low limit as it is we avoid
2050 * this pitfall: programs that are written by folks aware of the select() problem in mind
2051 * (and thus use poll()/epoll instead of select(), the way everybody should) can
2052 * explicitly opt into high fds by bumping their soft limit beyond 1024, to the hard limit
2053 * we pass. */
2054 if (arg_system) {
2055 int nr;
2056
2057 /* Get the underlying absolute limit the kernel enforces */
2058 nr = read_nr_open();
2059
2060 rl->rlim_max = MIN((rlim_t) nr, MAX(rl->rlim_max, (rlim_t) HIGH_RLIMIT_NOFILE));
2061 }
2062
2063 /* If for some reason we were invoked with a soft limit above 1024 (which should never
2064 * happen!, but who knows what we get passed in from pam_limit when invoked as --user
2065 * instance), then lower what we pass on to not confuse our children */
2066 rl->rlim_cur = MIN(rl->rlim_cur, (rlim_t) FD_SETSIZE);
2067
2068 arg_default_rlimit[RLIMIT_NOFILE] = rl;
2069 }
2070
2071 static void fallback_rlimit_memlock(const struct rlimit *saved_rlimit_memlock) {
2072 struct rlimit *rl;
2073
2074 /* Pass the original value down to invoked processes */
2075
2076 if (arg_default_rlimit[RLIMIT_MEMLOCK])
2077 return;
2078
2079 rl = newdup(struct rlimit, saved_rlimit_memlock, 1);
2080 if (!rl) {
2081 log_oom();
2082 return;
2083 }
2084
2085 arg_default_rlimit[RLIMIT_MEMLOCK] = rl;
2086 }
2087
2088 static void reset_arguments(void) {
2089 /* Frees/resets arg_* variables, with a few exceptions commented below. */
2090
2091 arg_default_unit = mfree(arg_default_unit);
2092
2093 /* arg_system — ignore */
2094
2095 arg_dump_core = true;
2096 arg_crash_chvt = -1;
2097 arg_crash_shell = false;
2098 arg_crash_reboot = false;
2099 arg_confirm_spawn = mfree(arg_confirm_spawn);
2100 arg_show_status = _SHOW_STATUS_INVALID;
2101 arg_status_unit_format = STATUS_UNIT_FORMAT_DEFAULT;
2102 arg_switched_root = false;
2103 arg_pager_flags = 0;
2104 arg_service_watchdogs = true;
2105 arg_default_std_output = EXEC_OUTPUT_JOURNAL;
2106 arg_default_std_error = EXEC_OUTPUT_INHERIT;
2107 arg_default_restart_usec = DEFAULT_RESTART_USEC;
2108 arg_default_timeout_start_usec = DEFAULT_TIMEOUT_USEC;
2109 arg_default_timeout_stop_usec = DEFAULT_TIMEOUT_USEC;
2110 arg_default_timeout_abort_usec = DEFAULT_TIMEOUT_USEC;
2111 arg_default_timeout_abort_set = false;
2112 arg_default_start_limit_interval = DEFAULT_START_LIMIT_INTERVAL;
2113 arg_default_start_limit_burst = DEFAULT_START_LIMIT_BURST;
2114 arg_runtime_watchdog = 0;
2115 arg_reboot_watchdog = 10 * USEC_PER_MINUTE;
2116 arg_kexec_watchdog = 0;
2117 arg_early_core_pattern = NULL;
2118 arg_watchdog_device = NULL;
2119
2120 arg_default_environment = strv_free(arg_default_environment);
2121 rlimit_free_all(arg_default_rlimit);
2122
2123 arg_capability_bounding_set = CAP_ALL;
2124 arg_no_new_privs = false;
2125 arg_timer_slack_nsec = NSEC_INFINITY;
2126 arg_default_timer_accuracy_usec = 1 * USEC_PER_MINUTE;
2127
2128 arg_syscall_archs = set_free(arg_syscall_archs);
2129
2130 /* arg_serialization — ignore */
2131
2132 arg_default_cpu_accounting = -1;
2133 arg_default_io_accounting = false;
2134 arg_default_ip_accounting = false;
2135 arg_default_blockio_accounting = false;
2136 arg_default_memory_accounting = MEMORY_ACCOUNTING_DEFAULT;
2137 arg_default_tasks_accounting = true;
2138 arg_default_tasks_max = system_tasks_max_scale(DEFAULT_TASKS_MAX_PERCENTAGE, 100U);
2139 arg_machine_id = (sd_id128_t) {};
2140 arg_cad_burst_action = EMERGENCY_ACTION_REBOOT_FORCE;
2141 arg_default_oom_policy = OOM_STOP;
2142
2143 cpu_set_reset(&arg_cpu_affinity);
2144 numa_policy_reset(&arg_numa_policy);
2145 }
2146
2147 static int parse_configuration(const struct rlimit *saved_rlimit_nofile,
2148 const struct rlimit *saved_rlimit_memlock) {
2149 int r;
2150
2151 assert(saved_rlimit_nofile);
2152 assert(saved_rlimit_memlock);
2153
2154 /* Assign configuration defaults */
2155 reset_arguments();
2156
2157 r = parse_config_file();
2158 if (r < 0)
2159 log_warning_errno(r, "Failed to parse config file, ignoring: %m");
2160
2161 if (arg_system) {
2162 r = proc_cmdline_parse(parse_proc_cmdline_item, NULL, 0);
2163 if (r < 0)
2164 log_warning_errno(r, "Failed to parse kernel command line, ignoring: %m");
2165 }
2166
2167 /* Initialize some default rlimits for services if they haven't been configured */
2168 fallback_rlimit_nofile(saved_rlimit_nofile);
2169 fallback_rlimit_memlock(saved_rlimit_memlock);
2170
2171 /* Note that this also parses bits from the kernel command line, including "debug". */
2172 log_parse_environment();
2173
2174 return 0;
2175 }
2176
2177 static int load_configuration(
2178 int argc,
2179 char **argv,
2180 const struct rlimit *saved_rlimit_nofile,
2181 const struct rlimit *saved_rlimit_memlock,
2182 const char **ret_error_message) {
2183 int r;
2184
2185 assert(saved_rlimit_nofile);
2186 assert(saved_rlimit_memlock);
2187 assert(ret_error_message);
2188
2189 (void) parse_configuration(saved_rlimit_nofile, saved_rlimit_memlock);
2190
2191 r = parse_argv(argc, argv);
2192 if (r < 0) {
2193 *ret_error_message = "Failed to parse commandline arguments";
2194 return r;
2195 }
2196
2197 /* Initialize default unit */
2198 if (!arg_default_unit) {
2199 arg_default_unit = strdup(SPECIAL_DEFAULT_TARGET);
2200 if (!arg_default_unit) {
2201 *ret_error_message = "Failed to set default unit";
2202 return log_oom();
2203 }
2204 }
2205
2206 /* Initialize the show status setting if it hasn't been set explicitly yet */
2207 if (arg_show_status == _SHOW_STATUS_INVALID)
2208 arg_show_status = SHOW_STATUS_YES;
2209
2210 return 0;
2211 }
2212
2213 static int safety_checks(void) {
2214
2215 if (getpid_cached() == 1 &&
2216 arg_action != ACTION_RUN)
2217 return log_error_errno(SYNTHETIC_ERRNO(EPERM),
2218 "Unsupported execution mode while PID 1.");
2219
2220 if (getpid_cached() == 1 &&
2221 !arg_system)
2222 return log_error_errno(SYNTHETIC_ERRNO(EPERM),
2223 "Can't run --user mode as PID 1.");
2224
2225 if (arg_action == ACTION_RUN &&
2226 arg_system &&
2227 getpid_cached() != 1)
2228 return log_error_errno(SYNTHETIC_ERRNO(EPERM),
2229 "Can't run system mode unless PID 1.");
2230
2231 if (arg_action == ACTION_TEST &&
2232 geteuid() == 0)
2233 return log_error_errno(SYNTHETIC_ERRNO(EPERM),
2234 "Don't run test mode as root.");
2235
2236 if (!arg_system &&
2237 arg_action == ACTION_RUN &&
2238 sd_booted() <= 0)
2239 return log_error_errno(SYNTHETIC_ERRNO(EOPNOTSUPP),
2240 "Trying to run as user instance, but the system has not been booted with systemd.");
2241
2242 if (!arg_system &&
2243 arg_action == ACTION_RUN &&
2244 !getenv("XDG_RUNTIME_DIR"))
2245 return log_error_errno(SYNTHETIC_ERRNO(EUNATCH),
2246 "Trying to run as user instance, but $XDG_RUNTIME_DIR is not set.");
2247
2248 if (arg_system &&
2249 arg_action == ACTION_RUN &&
2250 running_in_chroot() > 0)
2251 return log_error_errno(SYNTHETIC_ERRNO(EOPNOTSUPP),
2252 "Cannot be run in a chroot() environment.");
2253
2254 return 0;
2255 }
2256
2257 static int initialize_security(
2258 bool *loaded_policy,
2259 dual_timestamp *security_start_timestamp,
2260 dual_timestamp *security_finish_timestamp,
2261 const char **ret_error_message) {
2262
2263 int r;
2264
2265 assert(loaded_policy);
2266 assert(security_start_timestamp);
2267 assert(security_finish_timestamp);
2268 assert(ret_error_message);
2269
2270 dual_timestamp_get(security_start_timestamp);
2271
2272 r = mac_selinux_setup(loaded_policy);
2273 if (r < 0) {
2274 *ret_error_message = "Failed to load SELinux policy";
2275 return r;
2276 }
2277
2278 r = mac_smack_setup(loaded_policy);
2279 if (r < 0) {
2280 *ret_error_message = "Failed to load SMACK policy";
2281 return r;
2282 }
2283
2284 r = ima_setup();
2285 if (r < 0) {
2286 *ret_error_message = "Failed to load IMA policy";
2287 return r;
2288 }
2289
2290 dual_timestamp_get(security_finish_timestamp);
2291 return 0;
2292 }
2293
2294 static void test_summary(Manager *m) {
2295 assert(m);
2296
2297 printf("-> By units:\n");
2298 manager_dump_units(m, stdout, "\t");
2299
2300 printf("-> By jobs:\n");
2301 manager_dump_jobs(m, stdout, "\t");
2302 }
2303
2304 static int collect_fds(FDSet **ret_fds, const char **ret_error_message) {
2305 int r;
2306
2307 assert(ret_fds);
2308 assert(ret_error_message);
2309
2310 r = fdset_new_fill(ret_fds);
2311 if (r < 0) {
2312 *ret_error_message = "Failed to allocate fd set";
2313 return log_emergency_errno(r, "Failed to allocate fd set: %m");
2314 }
2315
2316 fdset_cloexec(*ret_fds, true);
2317
2318 if (arg_serialization)
2319 assert_se(fdset_remove(*ret_fds, fileno(arg_serialization)) >= 0);
2320
2321 return 0;
2322 }
2323
2324 static void setup_console_terminal(bool skip_setup) {
2325
2326 if (!arg_system)
2327 return;
2328
2329 /* Become a session leader if we aren't one yet. */
2330 (void) setsid();
2331
2332 /* If we are init, we connect stdin/stdout/stderr to /dev/null and make sure we don't have a controlling
2333 * tty. */
2334 (void) release_terminal();
2335
2336 /* Reset the console, but only if this is really init and we are freshly booted */
2337 if (getpid_cached() == 1 && !skip_setup)
2338 (void) console_setup();
2339 }
2340
2341 static bool early_skip_setup_check(int argc, char *argv[]) {
2342 bool found_deserialize = false;
2343 int i;
2344
2345 /* Determine if this is a reexecution or normal bootup. We do the full command line parsing much later, so
2346 * let's just have a quick peek here. Note that if we have switched root, do all the special setup things
2347 * anyway, even if in that case we also do deserialization. */
2348
2349 for (i = 1; i < argc; i++) {
2350 if (streq(argv[i], "--switched-root"))
2351 return false; /* If we switched root, don't skip the setup. */
2352 else if (streq(argv[i], "--deserialize"))
2353 found_deserialize = true;
2354 }
2355
2356 return found_deserialize; /* When we are deserializing, then we are reexecuting, hence avoid the extensive setup */
2357 }
2358
2359 static int save_env(void) {
2360 char **l;
2361
2362 l = strv_copy(environ);
2363 if (!l)
2364 return -ENOMEM;
2365
2366 strv_free_and_replace(saved_env, l);
2367 return 0;
2368 }
2369
2370 int main(int argc, char *argv[]) {
2371
2372 dual_timestamp initrd_timestamp = DUAL_TIMESTAMP_NULL, userspace_timestamp = DUAL_TIMESTAMP_NULL, kernel_timestamp = DUAL_TIMESTAMP_NULL,
2373 security_start_timestamp = DUAL_TIMESTAMP_NULL, security_finish_timestamp = DUAL_TIMESTAMP_NULL;
2374 struct rlimit saved_rlimit_nofile = RLIMIT_MAKE_CONST(0),
2375 saved_rlimit_memlock = RLIMIT_MAKE_CONST(RLIM_INFINITY); /* The original rlimits we passed
2376 * in. Note we use different values
2377 * for the two that indicate whether
2378 * these fields are initialized! */
2379 bool skip_setup, loaded_policy = false, queue_default_job = false, first_boot = false, reexecute = false;
2380 char *switch_root_dir = NULL, *switch_root_init = NULL;
2381 usec_t before_startup, after_startup;
2382 static char systemd[] = "systemd";
2383 char timespan[FORMAT_TIMESPAN_MAX];
2384 const char *shutdown_verb = NULL, *error_message = NULL;
2385 int r, retval = EXIT_FAILURE;
2386 Manager *m = NULL;
2387 FDSet *fds = NULL;
2388
2389 /* SysV compatibility: redirect init → telinit */
2390 redirect_telinit(argc, argv);
2391
2392 /* Take timestamps early on */
2393 dual_timestamp_from_monotonic(&kernel_timestamp, 0);
2394 dual_timestamp_get(&userspace_timestamp);
2395
2396 /* Figure out whether we need to do initialize the system, or if we already did that because we are
2397 * reexecuting */
2398 skip_setup = early_skip_setup_check(argc, argv);
2399
2400 /* If we get started via the /sbin/init symlink then we are called 'init'. After a subsequent reexecution we
2401 * are then called 'systemd'. That is confusing, hence let's call us systemd right-away. */
2402 program_invocation_short_name = systemd;
2403 (void) prctl(PR_SET_NAME, systemd);
2404
2405 /* Save the original command line */
2406 save_argc_argv(argc, argv);
2407
2408 /* Save the original environment as we might need to restore it if we're requested to execute another
2409 * system manager later. */
2410 r = save_env();
2411 if (r < 0) {
2412 error_message = "Failed to copy environment block";
2413 goto finish;
2414 }
2415
2416 /* Make sure that if the user says "syslog" we actually log to the journal. */
2417 log_set_upgrade_syslog_to_journal(true);
2418
2419 if (getpid_cached() == 1) {
2420 /* When we run as PID 1 force system mode */
2421 arg_system = true;
2422
2423 /* Disable the umask logic */
2424 umask(0);
2425
2426 /* Make sure that at least initially we do not ever log to journald/syslogd, because it might not be
2427 * activated yet (even though the log socket for it exists). */
2428 log_set_prohibit_ipc(true);
2429
2430 /* Always reopen /dev/console when running as PID 1 or one of its pre-execve() children. This is
2431 * important so that we never end up logging to any foreign stderr, for example if we have to log in a
2432 * child process right before execve()'ing the actual binary, at a point in time where socket
2433 * activation stderr/stdout area already set up. */
2434 log_set_always_reopen_console(true);
2435
2436 if (detect_container() <= 0) {
2437
2438 /* Running outside of a container as PID 1 */
2439 log_set_target(LOG_TARGET_KMSG);
2440 log_open();
2441
2442 if (in_initrd())
2443 initrd_timestamp = userspace_timestamp;
2444
2445 if (!skip_setup) {
2446 r = mount_setup_early();
2447 if (r < 0) {
2448 error_message = "Failed to mount early API filesystems";
2449 goto finish;
2450 }
2451
2452 /* Let's open the log backend a second time, in case the first time didn't
2453 * work. Quite possibly we have mounted /dev just now, so /dev/kmsg became
2454 * available, and it previously wasn't. */
2455 log_open();
2456
2457 disable_printk_ratelimit();
2458
2459 r = initialize_security(
2460 &loaded_policy,
2461 &security_start_timestamp,
2462 &security_finish_timestamp,
2463 &error_message);
2464 if (r < 0)
2465 goto finish;
2466 }
2467
2468 if (mac_selinux_init() < 0) {
2469 error_message = "Failed to initialize SELinux policy";
2470 goto finish;
2471 }
2472
2473 if (!skip_setup)
2474 initialize_clock();
2475
2476 /* Set the default for later on, but don't actually open the logs like this for now. Note that
2477 * if we are transitioning from the initrd there might still be journal fd open, and we
2478 * shouldn't attempt opening that before we parsed /proc/cmdline which might redirect output
2479 * elsewhere. */
2480 log_set_target(LOG_TARGET_JOURNAL_OR_KMSG);
2481
2482 } else {
2483 /* Running inside a container, as PID 1 */
2484 log_set_target(LOG_TARGET_CONSOLE);
2485 log_open();
2486
2487 /* For later on, see above... */
2488 log_set_target(LOG_TARGET_JOURNAL);
2489
2490 /* clear the kernel timestamp,
2491 * because we are in a container */
2492 kernel_timestamp = DUAL_TIMESTAMP_NULL;
2493 }
2494
2495 initialize_coredump(skip_setup);
2496
2497 r = fixup_environment();
2498 if (r < 0) {
2499 log_emergency_errno(r, "Failed to fix up PID 1 environment: %m");
2500 error_message = "Failed to fix up PID1 environment";
2501 goto finish;
2502 }
2503
2504 } else {
2505 /* Running as user instance */
2506 arg_system = false;
2507 log_set_target(LOG_TARGET_AUTO);
2508 log_open();
2509
2510 /* clear the kernel timestamp,
2511 * because we are not PID 1 */
2512 kernel_timestamp = DUAL_TIMESTAMP_NULL;
2513 }
2514
2515 if (arg_system) {
2516 /* Try to figure out if we can use colors with the console. No need to do that for user instances since
2517 * they never log into the console. */
2518 log_show_color(colors_enabled());
2519
2520 r = make_null_stdio();
2521 if (r < 0)
2522 log_warning_errno(r, "Failed to redirect standard streams to /dev/null, ignoring: %m");
2523 }
2524
2525 /* Mount /proc, /sys and friends, so that /proc/cmdline and
2526 * /proc/$PID/fd is available. */
2527 if (getpid_cached() == 1) {
2528
2529 /* Load the kernel modules early. */
2530 if (!skip_setup)
2531 kmod_setup();
2532
2533 r = mount_setup(loaded_policy);
2534 if (r < 0) {
2535 error_message = "Failed to mount API filesystems";
2536 goto finish;
2537 }
2538
2539 /* The efivarfs is now mounted, let's read the random seed off it */
2540 (void) efi_take_random_seed();
2541 }
2542
2543 /* Save the original RLIMIT_NOFILE/RLIMIT_MEMLOCK so that we can reset it later when
2544 * transitioning from the initrd to the main systemd or suchlike. */
2545 save_rlimits(&saved_rlimit_nofile, &saved_rlimit_memlock);
2546
2547 /* Reset all signal handlers. */
2548 (void) reset_all_signal_handlers();
2549 (void) ignore_signals(SIGNALS_IGNORE, -1);
2550
2551 r = load_configuration(argc, argv, &saved_rlimit_nofile, &saved_rlimit_memlock, &error_message);
2552 if (r < 0)
2553 goto finish;
2554
2555 r = safety_checks();
2556 if (r < 0)
2557 goto finish;
2558
2559 if (IN_SET(arg_action, ACTION_TEST, ACTION_HELP, ACTION_DUMP_CONFIGURATION_ITEMS, ACTION_DUMP_BUS_PROPERTIES))
2560 (void) pager_open(arg_pager_flags);
2561
2562 if (arg_action != ACTION_RUN)
2563 skip_setup = true;
2564
2565 if (arg_action == ACTION_HELP) {
2566 retval = help() < 0 ? EXIT_FAILURE : EXIT_SUCCESS;
2567 goto finish;
2568 } else if (arg_action == ACTION_VERSION) {
2569 retval = version();
2570 goto finish;
2571 } else if (arg_action == ACTION_DUMP_CONFIGURATION_ITEMS) {
2572 unit_dump_config_items(stdout);
2573 retval = EXIT_SUCCESS;
2574 goto finish;
2575 } else if (arg_action == ACTION_DUMP_BUS_PROPERTIES) {
2576 dump_bus_properties(stdout);
2577 retval = EXIT_SUCCESS;
2578 goto finish;
2579 }
2580
2581 assert_se(IN_SET(arg_action, ACTION_RUN, ACTION_TEST));
2582
2583 /* Move out of the way, so that we won't block unmounts */
2584 assert_se(chdir("/") == 0);
2585
2586 if (arg_action == ACTION_RUN) {
2587
2588 /* A core pattern might have been specified via the cmdline. */
2589 initialize_core_pattern(skip_setup);
2590
2591 /* Close logging fds, in order not to confuse collecting passed fds and terminal logic below */
2592 log_close();
2593
2594 /* Remember open file descriptors for later deserialization */
2595 r = collect_fds(&fds, &error_message);
2596 if (r < 0)
2597 goto finish;
2598
2599 /* Give up any control of the console, but make sure its initialized. */
2600 setup_console_terminal(skip_setup);
2601
2602 /* Open the logging devices, if possible and necessary */
2603 log_open();
2604 }
2605
2606 log_execution_mode(&first_boot);
2607
2608 r = initialize_runtime(skip_setup,
2609 &saved_rlimit_nofile,
2610 &saved_rlimit_memlock,
2611 &error_message);
2612 if (r < 0)
2613 goto finish;
2614
2615 r = manager_new(arg_system ? UNIT_FILE_SYSTEM : UNIT_FILE_USER,
2616 arg_action == ACTION_TEST ? MANAGER_TEST_FULL : 0,
2617 &m);
2618 if (r < 0) {
2619 log_emergency_errno(r, "Failed to allocate manager object: %m");
2620 error_message = "Failed to allocate manager object";
2621 goto finish;
2622 }
2623
2624 m->timestamps[MANAGER_TIMESTAMP_KERNEL] = kernel_timestamp;
2625 m->timestamps[MANAGER_TIMESTAMP_INITRD] = initrd_timestamp;
2626 m->timestamps[MANAGER_TIMESTAMP_USERSPACE] = userspace_timestamp;
2627 m->timestamps[manager_timestamp_initrd_mangle(MANAGER_TIMESTAMP_SECURITY_START)] = security_start_timestamp;
2628 m->timestamps[manager_timestamp_initrd_mangle(MANAGER_TIMESTAMP_SECURITY_FINISH)] = security_finish_timestamp;
2629
2630 set_manager_defaults(m);
2631 set_manager_settings(m);
2632 manager_set_first_boot(m, first_boot);
2633
2634 /* Remember whether we should queue the default job */
2635 queue_default_job = !arg_serialization || arg_switched_root;
2636
2637 before_startup = now(CLOCK_MONOTONIC);
2638
2639 r = manager_startup(m, arg_serialization, fds);
2640 if (r < 0) {
2641 error_message = "Failed to start up manager";
2642 goto finish;
2643 }
2644
2645 /* This will close all file descriptors that were opened, but not claimed by any unit. */
2646 fds = fdset_free(fds);
2647 arg_serialization = safe_fclose(arg_serialization);
2648
2649 if (queue_default_job) {
2650 r = do_queue_default_job(m, &error_message);
2651 if (r < 0)
2652 goto finish;
2653 }
2654
2655 after_startup = now(CLOCK_MONOTONIC);
2656
2657 log_full(arg_action == ACTION_TEST ? LOG_INFO : LOG_DEBUG,
2658 "Loaded units and determined initial transaction in %s.",
2659 format_timespan(timespan, sizeof(timespan), after_startup - before_startup, 100 * USEC_PER_MSEC));
2660
2661 if (arg_action == ACTION_TEST) {
2662 test_summary(m);
2663 retval = EXIT_SUCCESS;
2664 goto finish;
2665 }
2666
2667 (void) invoke_main_loop(m,
2668 &saved_rlimit_nofile,
2669 &saved_rlimit_memlock,
2670 &reexecute,
2671 &retval,
2672 &shutdown_verb,
2673 &fds,
2674 &switch_root_dir,
2675 &switch_root_init,
2676 &error_message);
2677
2678 finish:
2679 pager_close();
2680
2681 if (m) {
2682 arg_reboot_watchdog = m->reboot_watchdog;
2683 arg_kexec_watchdog = m->kexec_watchdog;
2684 m = manager_free(m);
2685 }
2686
2687 reset_arguments();
2688 mac_selinux_finish();
2689
2690 if (reexecute)
2691 do_reexecute(argc, argv,
2692 &saved_rlimit_nofile,
2693 &saved_rlimit_memlock,
2694 fds,
2695 switch_root_dir,
2696 switch_root_init,
2697 &error_message); /* This only returns if reexecution failed */
2698
2699 arg_serialization = safe_fclose(arg_serialization);
2700 fds = fdset_free(fds);
2701
2702 saved_env = strv_free(saved_env);
2703
2704 #if HAVE_VALGRIND_VALGRIND_H
2705 /* If we are PID 1 and running under valgrind, then let's exit
2706 * here explicitly. valgrind will only generate nice output on
2707 * exit(), not on exec(), hence let's do the former not the
2708 * latter here. */
2709 if (getpid_cached() == 1 && RUNNING_ON_VALGRIND) {
2710 /* Cleanup watchdog_device strings for valgrind. We need them
2711 * in become_shutdown() so normally we cannot free them yet. */
2712 watchdog_free_device();
2713 arg_watchdog_device = mfree(arg_watchdog_device);
2714 return retval;
2715 }
2716 #endif
2717
2718 #if HAS_FEATURE_ADDRESS_SANITIZER
2719 __lsan_do_leak_check();
2720 #endif
2721
2722 if (shutdown_verb) {
2723 r = become_shutdown(shutdown_verb, retval);
2724 log_error_errno(r, "Failed to execute shutdown binary, %s: %m", getpid_cached() == 1 ? "freezing" : "quitting");
2725 error_message = "Failed to execute shutdown binary";
2726 }
2727
2728 watchdog_free_device();
2729 arg_watchdog_device = mfree(arg_watchdog_device);
2730
2731 if (getpid_cached() == 1) {
2732 if (error_message)
2733 manager_status_printf(NULL, STATUS_TYPE_EMERGENCY,
2734 ANSI_HIGHLIGHT_RED "!!!!!!" ANSI_NORMAL,
2735 "%s.", error_message);
2736 freeze_or_exit_or_reboot();
2737 }
2738
2739 return retval;
2740 }
Unnamed repository; edit this file 'description' to name the repository.