1 /* SPDX-License-Identifier: LGPL-2.1-or-later */
9 #include <sys/utsname.h>
11 #if HAVE_VALGRIND_VALGRIND_H
12 # include <valgrind/valgrind.h>
16 #include "sd-daemon.h"
17 #include "sd-messages.h"
19 #include "alloc-util.h"
20 #include "apparmor-setup.h"
21 #include "architecture.h"
22 #include "argv-util.h"
27 #include "bus-error.h"
29 #include "capability-util.h"
30 #include "cgroup-util.h"
32 #include "clock-util.h"
33 #include "conf-parser.h"
34 #include "confidential-virt.h"
36 #include "cpu-set-util.h"
37 #include "crash-handler.h"
38 #include "dbus-manager.h"
40 #include "constants.h"
41 #include "dev-setup.h"
42 #include "efi-random.h"
44 #include "emergency-action.h"
46 #include "exit-status.h"
50 #include "format-util.h"
52 #include "getopt-defs.h"
53 #include "hexdecoct.h"
54 #include "hostname-setup.h"
55 #include "ima-setup.h"
56 #include "import-creds.h"
57 #include "initrd-util.h"
59 #include "kmod-setup.h"
60 #include "limits-util.h"
61 #include "load-fragment.h"
63 #include "loopback-setup.h"
64 #include "machine-id-setup.h"
67 #include "manager-dump.h"
68 #include "manager-serialize.h"
69 #include "mkdir-label.h"
70 #include "mount-setup.h"
73 #include "parse-argument.h"
74 #include "parse-util.h"
75 #include "path-util.h"
76 #include "pretty-print.h"
77 #include "proc-cmdline.h"
78 #include "process-util.h"
80 #include "random-util.h"
81 #include "rlimit-util.h"
82 #include "seccomp-util.h"
83 #include "selinux-setup.h"
84 #include "selinux-util.h"
85 #include "signal-util.h"
86 #include "smack-setup.h"
88 #include "stat-util.h"
89 #include "stdio-util.h"
91 #include "switch-root.h"
92 #include "sysctl-util.h"
93 #include "terminal-util.h"
94 #include "time-util.h"
95 #include "umask-util.h"
96 #include "user-util.h"
101 #if HAS_FEATURE_ADDRESS_SANITIZER
102 #include <sanitizer/lsan_interface.h>
110 ACTION_DUMP_CONFIGURATION_ITEMS
,
111 ACTION_DUMP_BUS_PROPERTIES
,
112 ACTION_BUS_INTROSPECT
,
113 } arg_action
= ACTION_RUN
;
115 static const char *arg_bus_introspect
= NULL
;
117 /* Those variables are initialized to 0 automatically, so we avoid uninitialized memory access. Real
118 * defaults are assigned in reset_arguments() below. */
119 static char *arg_default_unit
;
120 static RuntimeScope arg_runtime_scope
;
123 bool arg_crash_shell
;
124 bool arg_crash_reboot
;
125 static char *arg_confirm_spawn
;
126 static ShowStatus arg_show_status
;
127 static StatusUnitFormat arg_status_unit_format
;
128 static bool arg_switched_root
;
129 static PagerFlags arg_pager_flags
;
130 static bool arg_service_watchdogs
;
131 static UnitDefaults arg_defaults
;
132 static usec_t arg_runtime_watchdog
;
133 static usec_t arg_reboot_watchdog
;
134 static usec_t arg_kexec_watchdog
;
135 static usec_t arg_pretimeout_watchdog
;
136 static char *arg_early_core_pattern
;
137 static char *arg_watchdog_pretimeout_governor
;
138 static char *arg_watchdog_device
;
139 static char **arg_default_environment
;
140 static char **arg_manager_environment
;
141 static uint64_t arg_capability_bounding_set
;
142 static bool arg_no_new_privs
;
143 static nsec_t arg_timer_slack_nsec
;
144 static Set
* arg_syscall_archs
;
145 static FILE* arg_serialization
;
146 static sd_id128_t arg_machine_id
;
147 static EmergencyAction arg_cad_burst_action
;
148 static CPUSet arg_cpu_affinity
;
149 static NUMAPolicy arg_numa_policy
;
150 static usec_t arg_clock_usec
;
151 static void *arg_random_seed
;
152 static size_t arg_random_seed_size
;
153 static usec_t arg_reload_limit_interval_sec
;
154 static unsigned arg_reload_limit_burst
;
156 /* A copy of the original environment block */
157 static char **saved_env
= NULL
;
159 static int parse_configuration(const struct rlimit
*saved_rlimit_nofile
,
160 const struct rlimit
*saved_rlimit_memlock
);
162 static int manager_find_user_config_paths(char ***ret_files
, char ***ret_dirs
) {
163 _cleanup_free_
char *base
= NULL
;
164 _cleanup_strv_free_
char **files
= NULL
, **dirs
= NULL
;
167 r
= xdg_user_config_dir(&base
, "/systemd");
171 r
= strv_extendf(&files
, "%s/user.conf", base
);
175 r
= strv_extend(&files
, PKGSYSCONFDIR
"/user.conf");
179 r
= strv_consume(&dirs
, TAKE_PTR(base
));
183 r
= strv_extend_strv(&dirs
, CONF_PATHS_STRV("systemd"), false);
187 *ret_files
= TAKE_PTR(files
);
188 *ret_dirs
= TAKE_PTR(dirs
);
192 static int console_setup(void) {
193 _cleanup_close_
int tty_fd
= -EBADF
;
197 tty_fd
= open_terminal("/dev/console", O_WRONLY
|O_NOCTTY
|O_CLOEXEC
);
199 return log_error_errno(tty_fd
, "Failed to open /dev/console: %m");
201 /* We don't want to force text mode. plymouth may be showing
202 * pictures already from initrd. */
203 r
= reset_terminal_fd(tty_fd
, false);
205 return log_error_errno(r
, "Failed to reset /dev/console: %m");
207 r
= proc_cmdline_tty_size("/dev/console", &rows
, &cols
);
209 log_warning_errno(r
, "Failed to get terminal size, ignoring: %m");
211 r
= terminal_set_size_fd(tty_fd
, NULL
, rows
, cols
);
213 log_warning_errno(r
, "Failed to set terminal size, ignoring: %m");
219 static int parse_proc_cmdline_item(const char *key
, const char *value
, void *data
) {
224 if (STR_IN_SET(key
, "systemd.unit", "rd.systemd.unit")) {
226 if (proc_cmdline_value_missing(key
, value
))
229 if (!unit_name_is_valid(value
, UNIT_NAME_PLAIN
|UNIT_NAME_INSTANCE
))
230 log_warning("Unit name specified on %s= is not valid, ignoring: %s", key
, value
);
231 else if (in_initrd() == !!startswith(key
, "rd."))
232 return free_and_strdup_warn(&arg_default_unit
, value
);
234 } else if (proc_cmdline_key_streq(key
, "systemd.dump_core")) {
236 r
= value
? parse_boolean(value
) : true;
238 log_warning_errno(r
, "Failed to parse dump core switch %s, ignoring: %m", value
);
242 } else if (proc_cmdline_key_streq(key
, "systemd.early_core_pattern")) {
244 if (proc_cmdline_value_missing(key
, value
))
247 if (path_is_absolute(value
))
248 (void) parse_path_argument(value
, false, &arg_early_core_pattern
);
250 log_warning("Specified core pattern '%s' is not an absolute path, ignoring.", value
);
252 } else if (proc_cmdline_key_streq(key
, "systemd.crash_chvt")) {
255 arg_crash_chvt
= 0; /* turn on */
257 r
= parse_crash_chvt(value
, &arg_crash_chvt
);
259 log_warning_errno(r
, "Failed to parse crash chvt switch %s, ignoring: %m", value
);
262 } else if (proc_cmdline_key_streq(key
, "systemd.crash_shell")) {
264 r
= value
? parse_boolean(value
) : true;
266 log_warning_errno(r
, "Failed to parse crash shell switch %s, ignoring: %m", value
);
270 } else if (proc_cmdline_key_streq(key
, "systemd.crash_reboot")) {
272 r
= value
? parse_boolean(value
) : true;
274 log_warning_errno(r
, "Failed to parse crash reboot switch %s, ignoring: %m", value
);
276 arg_crash_reboot
= r
;
278 } else if (proc_cmdline_key_streq(key
, "systemd.confirm_spawn")) {
281 r
= parse_confirm_spawn(value
, &s
);
283 log_warning_errno(r
, "Failed to parse confirm_spawn switch %s, ignoring: %m", value
);
285 free_and_replace(arg_confirm_spawn
, s
);
287 } else if (proc_cmdline_key_streq(key
, "systemd.service_watchdogs")) {
289 r
= value
? parse_boolean(value
) : true;
291 log_warning_errno(r
, "Failed to parse service watchdog switch %s, ignoring: %m", value
);
293 arg_service_watchdogs
= r
;
295 } else if (proc_cmdline_key_streq(key
, "systemd.show_status")) {
298 r
= parse_show_status(value
, &arg_show_status
);
300 log_warning_errno(r
, "Failed to parse show status switch %s, ignoring: %m", value
);
302 arg_show_status
= SHOW_STATUS_YES
;
304 } else if (proc_cmdline_key_streq(key
, "systemd.status_unit_format")) {
306 if (proc_cmdline_value_missing(key
, value
))
309 r
= status_unit_format_from_string(value
);
311 log_warning_errno(r
, "Failed to parse %s=%s, ignoring: %m", key
, value
);
313 arg_status_unit_format
= r
;
315 } else if (proc_cmdline_key_streq(key
, "systemd.default_standard_output")) {
317 if (proc_cmdline_value_missing(key
, value
))
320 r
= exec_output_from_string(value
);
322 log_warning_errno(r
, "Failed to parse default standard output switch %s, ignoring: %m", value
);
324 arg_defaults
.std_output
= r
;
326 } else if (proc_cmdline_key_streq(key
, "systemd.default_standard_error")) {
328 if (proc_cmdline_value_missing(key
, value
))
331 r
= exec_output_from_string(value
);
333 log_warning_errno(r
, "Failed to parse default standard error switch %s, ignoring: %m", value
);
335 arg_defaults
.std_error
= r
;
337 } else if (streq(key
, "systemd.setenv")) {
339 if (proc_cmdline_value_missing(key
, value
))
342 if (!env_assignment_is_valid(value
))
343 log_warning("Environment variable assignment '%s' is not valid. Ignoring.", value
);
345 r
= strv_env_replace_strdup(&arg_default_environment
, value
);
350 } else if (proc_cmdline_key_streq(key
, "systemd.machine_id")) {
352 if (proc_cmdline_value_missing(key
, value
))
355 r
= id128_from_string_nonzero(value
, &arg_machine_id
);
357 log_warning_errno(r
, "MachineID '%s' is not valid, ignoring: %m", value
);
359 } else if (proc_cmdline_key_streq(key
, "systemd.default_timeout_start_sec")) {
361 if (proc_cmdline_value_missing(key
, value
))
364 r
= parse_sec(value
, &arg_defaults
.timeout_start_usec
);
366 log_warning_errno(r
, "Failed to parse default start timeout '%s', ignoring: %m", value
);
368 if (arg_defaults
.timeout_start_usec
<= 0)
369 arg_defaults
.timeout_start_usec
= USEC_INFINITY
;
371 } else if (proc_cmdline_key_streq(key
, "systemd.default_device_timeout_sec")) {
373 if (proc_cmdline_value_missing(key
, value
))
376 r
= parse_sec(value
, &arg_defaults
.device_timeout_usec
);
378 log_warning_errno(r
, "Failed to parse default device timeout '%s', ignoring: %m", value
);
380 if (arg_defaults
.device_timeout_usec
<= 0)
381 arg_defaults
.device_timeout_usec
= USEC_INFINITY
;
383 } else if (proc_cmdline_key_streq(key
, "systemd.cpu_affinity")) {
385 if (proc_cmdline_value_missing(key
, value
))
388 r
= parse_cpu_set(value
, &arg_cpu_affinity
);
390 log_warning_errno(r
, "Failed to parse CPU affinity mask '%s', ignoring: %m", value
);
392 } else if (proc_cmdline_key_streq(key
, "systemd.watchdog_device")) {
394 if (proc_cmdline_value_missing(key
, value
))
397 (void) parse_path_argument(value
, false, &arg_watchdog_device
);
399 } else if (proc_cmdline_key_streq(key
, "systemd.watchdog_sec")) {
401 if (proc_cmdline_value_missing(key
, value
))
404 if (streq(value
, "default"))
405 arg_runtime_watchdog
= USEC_INFINITY
;
406 else if (streq(value
, "off"))
407 arg_runtime_watchdog
= 0;
409 r
= parse_sec(value
, &arg_runtime_watchdog
);
411 log_warning_errno(r
, "Failed to parse systemd.watchdog_sec= argument '%s', ignoring: %m", value
);
416 arg_kexec_watchdog
= arg_reboot_watchdog
= arg_runtime_watchdog
;
418 } else if (proc_cmdline_key_streq(key
, "systemd.watchdog_pre_sec")) {
420 if (proc_cmdline_value_missing(key
, value
))
423 if (streq(value
, "default"))
424 arg_pretimeout_watchdog
= USEC_INFINITY
;
425 else if (streq(value
, "off"))
426 arg_pretimeout_watchdog
= 0;
428 r
= parse_sec(value
, &arg_pretimeout_watchdog
);
430 log_warning_errno(r
, "Failed to parse systemd.watchdog_pre_sec= argument '%s', ignoring: %m", value
);
435 } else if (proc_cmdline_key_streq(key
, "systemd.watchdog_pretimeout_governor")) {
437 if (proc_cmdline_value_missing(key
, value
) || isempty(value
)) {
438 arg_watchdog_pretimeout_governor
= mfree(arg_watchdog_pretimeout_governor
);
442 if (!string_is_safe(value
)) {
443 log_warning("Watchdog pretimeout governor '%s' is not valid, ignoring.", value
);
447 return free_and_strdup_warn(&arg_watchdog_pretimeout_governor
, value
);
449 } else if (proc_cmdline_key_streq(key
, "systemd.clock_usec")) {
451 if (proc_cmdline_value_missing(key
, value
))
454 r
= safe_atou64(value
, &arg_clock_usec
);
456 log_warning_errno(r
, "Failed to parse systemd.clock_usec= argument, ignoring: %s", value
);
458 } else if (proc_cmdline_key_streq(key
, "systemd.random_seed")) {
462 if (proc_cmdline_value_missing(key
, value
))
465 r
= unbase64mem(value
, SIZE_MAX
, &p
, &sz
);
467 log_warning_errno(r
, "Failed to parse systemd.random_seed= argument, ignoring: %s", value
);
469 free(arg_random_seed
);
470 arg_random_seed
= sz
> 0 ? p
: mfree(p
);
471 arg_random_seed_size
= sz
;
473 } else if (proc_cmdline_key_streq(key
, "systemd.reload_limit_interval_sec")) {
475 if (proc_cmdline_value_missing(key
, value
))
478 r
= parse_sec(value
, &arg_reload_limit_interval_sec
);
480 log_warning_errno(r
, "Failed to parse systemd.reload_limit_interval_sec= argument '%s', ignoring: %m", value
);
484 } else if (proc_cmdline_key_streq(key
, "systemd.reload_limit_burst")) {
486 if (proc_cmdline_value_missing(key
, value
))
489 r
= safe_atou(value
, &arg_reload_limit_burst
);
491 log_warning_errno(r
, "Failed to parse systemd.reload_limit_burst= argument '%s', ignoring: %m", value
);
495 } else if (streq(key
, "quiet") && !value
) {
497 if (arg_show_status
== _SHOW_STATUS_INVALID
)
498 arg_show_status
= SHOW_STATUS_ERROR
;
500 } else if (streq(key
, "debug") && !value
) {
502 /* Note that log_parse_environment() handles 'debug'
503 * too, and sets the log level to LOG_DEBUG. */
505 if (detect_container() > 0)
506 log_set_target(LOG_TARGET_CONSOLE
);
511 /* Compatible with SysV, but supported independently even if SysV compatibility is disabled. */
512 target
= runlevel_to_target(key
);
514 return free_and_strdup_warn(&arg_default_unit
, target
);
520 #define DEFINE_SETTER(name, func, descr) \
521 static int name(const char *unit, \
522 const char *filename, \
524 const char *section, \
525 unsigned section_line, \
526 const char *lvalue, \
528 const char *rvalue, \
540 log_syntax(unit, LOG_ERR, filename, line, r, \
541 "Invalid " descr "'%s': %m", \
547 DEFINE_SETTER(config_parse_level2
, log_set_max_level_from_string
, "log level");
548 DEFINE_SETTER(config_parse_target
, log_set_target_from_string
, "target");
549 DEFINE_SETTER(config_parse_color
, log_show_color_from_string
, "color");
550 DEFINE_SETTER(config_parse_location
, log_show_location_from_string
, "location");
551 DEFINE_SETTER(config_parse_time
, log_show_time_from_string
, "time");
553 static int config_parse_default_timeout_abort(
555 const char *filename
,
558 unsigned section_line
,
566 r
= config_parse_timeout_abort(
575 &arg_defaults
.timeout_abort_usec
,
578 arg_defaults
.timeout_abort_set
= r
;
582 static int config_parse_oom_score_adjust(
584 const char *filename
,
587 unsigned section_line
,
596 if (isempty(rvalue
)) {
597 arg_defaults
.oom_score_adjust_set
= false;
601 r
= parse_oom_score_adjust(rvalue
, &oa
);
603 log_syntax(unit
, LOG_WARNING
, filename
, line
, r
, "Failed to parse the OOM score adjust value '%s', ignoring: %m", rvalue
);
607 arg_defaults
.oom_score_adjust
= oa
;
608 arg_defaults
.oom_score_adjust_set
= true;
613 static int parse_config_file(void) {
614 const ConfigTableItem items
[] = {
615 { "Manager", "LogLevel", config_parse_level2
, 0, NULL
},
616 { "Manager", "LogTarget", config_parse_target
, 0, NULL
},
617 { "Manager", "LogColor", config_parse_color
, 0, NULL
},
618 { "Manager", "LogLocation", config_parse_location
, 0, NULL
},
619 { "Manager", "LogTime", config_parse_time
, 0, NULL
},
620 { "Manager", "DumpCore", config_parse_bool
, 0, &arg_dump_core
},
621 { "Manager", "CrashChVT", /* legacy */ config_parse_crash_chvt
, 0, &arg_crash_chvt
},
622 { "Manager", "CrashChangeVT", config_parse_crash_chvt
, 0, &arg_crash_chvt
},
623 { "Manager", "CrashShell", config_parse_bool
, 0, &arg_crash_shell
},
624 { "Manager", "CrashReboot", config_parse_bool
, 0, &arg_crash_reboot
},
625 { "Manager", "ShowStatus", config_parse_show_status
, 0, &arg_show_status
},
626 { "Manager", "StatusUnitFormat", config_parse_status_unit_format
, 0, &arg_status_unit_format
},
627 { "Manager", "CPUAffinity", config_parse_cpu_affinity2
, 0, &arg_cpu_affinity
},
628 { "Manager", "NUMAPolicy", config_parse_numa_policy
, 0, &arg_numa_policy
.type
},
629 { "Manager", "NUMAMask", config_parse_numa_mask
, 0, &arg_numa_policy
},
630 { "Manager", "JoinControllers", config_parse_warn_compat
, DISABLED_CONFIGURATION
, NULL
},
631 { "Manager", "RuntimeWatchdogSec", config_parse_watchdog_sec
, 0, &arg_runtime_watchdog
},
632 { "Manager", "RuntimeWatchdogPreSec", config_parse_watchdog_sec
, 0, &arg_pretimeout_watchdog
},
633 { "Manager", "RebootWatchdogSec", config_parse_watchdog_sec
, 0, &arg_reboot_watchdog
},
634 { "Manager", "ShutdownWatchdogSec", config_parse_watchdog_sec
, 0, &arg_reboot_watchdog
}, /* obsolete alias */
635 { "Manager", "KExecWatchdogSec", config_parse_watchdog_sec
, 0, &arg_kexec_watchdog
},
636 { "Manager", "WatchdogDevice", config_parse_path
, 0, &arg_watchdog_device
},
637 { "Manager", "RuntimeWatchdogPreGovernor", config_parse_string
, CONFIG_PARSE_STRING_SAFE
, &arg_watchdog_pretimeout_governor
},
638 { "Manager", "CapabilityBoundingSet", config_parse_capability_set
, 0, &arg_capability_bounding_set
},
639 { "Manager", "NoNewPrivileges", config_parse_bool
, 0, &arg_no_new_privs
},
641 { "Manager", "SystemCallArchitectures", config_parse_syscall_archs
, 0, &arg_syscall_archs
},
643 { "Manager", "SystemCallArchitectures", config_parse_warn_compat
, DISABLED_CONFIGURATION
, NULL
},
646 { "Manager", "TimerSlackNSec", config_parse_nsec
, 0, &arg_timer_slack_nsec
},
647 { "Manager", "DefaultTimerAccuracySec", config_parse_sec
, 0, &arg_defaults
.timer_accuracy_usec
},
648 { "Manager", "DefaultStandardOutput", config_parse_output_restricted
, 0, &arg_defaults
.std_output
},
649 { "Manager", "DefaultStandardError", config_parse_output_restricted
, 0, &arg_defaults
.std_error
},
650 { "Manager", "DefaultTimeoutStartSec", config_parse_sec
, 0, &arg_defaults
.timeout_start_usec
},
651 { "Manager", "DefaultTimeoutStopSec", config_parse_sec
, 0, &arg_defaults
.timeout_stop_usec
},
652 { "Manager", "DefaultTimeoutAbortSec", config_parse_default_timeout_abort
, 0, NULL
},
653 { "Manager", "DefaultDeviceTimeoutSec", config_parse_sec
, 0, &arg_defaults
.device_timeout_usec
},
654 { "Manager", "DefaultRestartSec", config_parse_sec
, 0, &arg_defaults
.restart_usec
},
655 { "Manager", "DefaultStartLimitInterval", config_parse_sec
, 0, &arg_defaults
.start_limit_interval
}, /* obsolete alias */
656 { "Manager", "DefaultStartLimitIntervalSec", config_parse_sec
, 0, &arg_defaults
.start_limit_interval
},
657 { "Manager", "DefaultStartLimitBurst", config_parse_unsigned
, 0, &arg_defaults
.start_limit_burst
},
658 { "Manager", "DefaultEnvironment", config_parse_environ
, arg_runtime_scope
, &arg_default_environment
},
659 { "Manager", "ManagerEnvironment", config_parse_environ
, arg_runtime_scope
, &arg_manager_environment
},
660 { "Manager", "DefaultLimitCPU", config_parse_rlimit
, RLIMIT_CPU
, arg_defaults
.rlimit
},
661 { "Manager", "DefaultLimitFSIZE", config_parse_rlimit
, RLIMIT_FSIZE
, arg_defaults
.rlimit
},
662 { "Manager", "DefaultLimitDATA", config_parse_rlimit
, RLIMIT_DATA
, arg_defaults
.rlimit
},
663 { "Manager", "DefaultLimitSTACK", config_parse_rlimit
, RLIMIT_STACK
, arg_defaults
.rlimit
},
664 { "Manager", "DefaultLimitCORE", config_parse_rlimit
, RLIMIT_CORE
, arg_defaults
.rlimit
},
665 { "Manager", "DefaultLimitRSS", config_parse_rlimit
, RLIMIT_RSS
, arg_defaults
.rlimit
},
666 { "Manager", "DefaultLimitNOFILE", config_parse_rlimit
, RLIMIT_NOFILE
, arg_defaults
.rlimit
},
667 { "Manager", "DefaultLimitAS", config_parse_rlimit
, RLIMIT_AS
, arg_defaults
.rlimit
},
668 { "Manager", "DefaultLimitNPROC", config_parse_rlimit
, RLIMIT_NPROC
, arg_defaults
.rlimit
},
669 { "Manager", "DefaultLimitMEMLOCK", config_parse_rlimit
, RLIMIT_MEMLOCK
, arg_defaults
.rlimit
},
670 { "Manager", "DefaultLimitLOCKS", config_parse_rlimit
, RLIMIT_LOCKS
, arg_defaults
.rlimit
},
671 { "Manager", "DefaultLimitSIGPENDING", config_parse_rlimit
, RLIMIT_SIGPENDING
, arg_defaults
.rlimit
},
672 { "Manager", "DefaultLimitMSGQUEUE", config_parse_rlimit
, RLIMIT_MSGQUEUE
, arg_defaults
.rlimit
},
673 { "Manager", "DefaultLimitNICE", config_parse_rlimit
, RLIMIT_NICE
, arg_defaults
.rlimit
},
674 { "Manager", "DefaultLimitRTPRIO", config_parse_rlimit
, RLIMIT_RTPRIO
, arg_defaults
.rlimit
},
675 { "Manager", "DefaultLimitRTTIME", config_parse_rlimit
, RLIMIT_RTTIME
, arg_defaults
.rlimit
},
676 { "Manager", "DefaultCPUAccounting", config_parse_bool
, 0, &arg_defaults
.cpu_accounting
},
677 { "Manager", "DefaultIOAccounting", config_parse_bool
, 0, &arg_defaults
.io_accounting
},
678 { "Manager", "DefaultIPAccounting", config_parse_bool
, 0, &arg_defaults
.ip_accounting
},
679 { "Manager", "DefaultBlockIOAccounting", config_parse_bool
, 0, &arg_defaults
.blockio_accounting
},
680 { "Manager", "DefaultMemoryAccounting", config_parse_bool
, 0, &arg_defaults
.memory_accounting
},
681 { "Manager", "DefaultTasksAccounting", config_parse_bool
, 0, &arg_defaults
.tasks_accounting
},
682 { "Manager", "DefaultTasksMax", config_parse_tasks_max
, 0, &arg_defaults
.tasks_max
},
683 { "Manager", "DefaultMemoryPressureThresholdSec", config_parse_sec
, 0, &arg_defaults
.memory_pressure_threshold_usec
},
684 { "Manager", "DefaultMemoryPressureWatch", config_parse_memory_pressure_watch
, 0, &arg_defaults
.memory_pressure_watch
},
685 { "Manager", "CtrlAltDelBurstAction", config_parse_emergency_action
, arg_runtime_scope
, &arg_cad_burst_action
},
686 { "Manager", "DefaultOOMPolicy", config_parse_oom_policy
, 0, &arg_defaults
.oom_policy
},
687 { "Manager", "DefaultOOMScoreAdjust", config_parse_oom_score_adjust
, 0, NULL
},
688 { "Manager", "ReloadLimitIntervalSec", config_parse_sec
, 0, &arg_reload_limit_interval_sec
},
689 { "Manager", "ReloadLimitBurst", config_parse_unsigned
, 0, &arg_reload_limit_burst
},
691 { "Manager", "DefaultSmackProcessLabel", config_parse_string
, 0, &arg_defaults
.smack_process_label
},
693 { "Manager", "DefaultSmackProcessLabel", config_parse_warn_compat
, DISABLED_CONFIGURATION
, NULL
},
698 if (arg_runtime_scope
== RUNTIME_SCOPE_SYSTEM
)
699 (void) config_parse_config_file("system.conf",
701 config_item_table_lookup
, items
,
705 _cleanup_strv_free_
char **files
= NULL
, **dirs
= NULL
;
708 assert(arg_runtime_scope
== RUNTIME_SCOPE_USER
);
710 r
= manager_find_user_config_paths(&files
, &dirs
);
712 return log_error_errno(r
, "Failed to determine config file paths: %m");
714 (void) config_parse_many(
715 (const char* const*) files
,
716 (const char* const*) dirs
,
720 config_item_table_lookup
, items
,
725 /* Traditionally "0" was used to turn off the default unit timeouts. Fix this up so that we use
726 * USEC_INFINITY like everywhere else. */
727 if (arg_defaults
.timeout_start_usec
<= 0)
728 arg_defaults
.timeout_start_usec
= USEC_INFINITY
;
729 if (arg_defaults
.timeout_stop_usec
<= 0)
730 arg_defaults
.timeout_stop_usec
= USEC_INFINITY
;
735 static void set_manager_defaults(Manager
*m
) {
740 /* Propagates the various default unit property settings into the manager object, i.e. properties
741 * that do not affect the manager itself, but are just what newly allocated units will have set if
742 * they haven't set anything else. (Also see set_manager_settings() for the settings that affect the
743 * manager's own behaviour) */
745 r
= manager_set_unit_defaults(m
, &arg_defaults
);
747 log_warning_errno(r
, "Failed to set manager defaults, ignoring: %m");
749 r
= manager_default_environment(m
);
751 log_warning_errno(r
, "Failed to set manager default environment, ignoring: %m");
753 r
= manager_transient_environment_add(m
, arg_default_environment
);
755 log_warning_errno(r
, "Failed to add to transient environment, ignoring: %m");
758 static void set_manager_settings(Manager
*m
) {
763 /* Propagates the various manager settings into the manager object, i.e. properties that
764 * effect the manager itself (as opposed to just being inherited into newly allocated
765 * units, see set_manager_defaults() above). */
767 m
->confirm_spawn
= arg_confirm_spawn
;
768 m
->service_watchdogs
= arg_service_watchdogs
;
769 m
->cad_burst_action
= arg_cad_burst_action
;
770 /* Note that we don't do structured initialization here, otherwise it will reset the rate limit
771 * counter on every daemon-reload. */
772 m
->reload_ratelimit
.interval
= arg_reload_limit_interval_sec
;
773 m
->reload_ratelimit
.burst
= arg_reload_limit_burst
;
775 manager_set_watchdog(m
, WATCHDOG_RUNTIME
, arg_runtime_watchdog
);
776 manager_set_watchdog(m
, WATCHDOG_REBOOT
, arg_reboot_watchdog
);
777 manager_set_watchdog(m
, WATCHDOG_KEXEC
, arg_kexec_watchdog
);
778 manager_set_watchdog(m
, WATCHDOG_PRETIMEOUT
, arg_pretimeout_watchdog
);
779 r
= manager_set_watchdog_pretimeout_governor(m
, arg_watchdog_pretimeout_governor
);
781 log_warning_errno(r
, "Failed to set watchdog pretimeout governor to '%s', ignoring: %m", arg_watchdog_pretimeout_governor
);
783 manager_set_show_status(m
, arg_show_status
, "command line");
784 m
->status_unit_format
= arg_status_unit_format
;
787 static int parse_argv(int argc
, char *argv
[]) {
793 static const struct option options
[] = {
794 COMMON_GETOPT_OPTIONS
,
795 SYSTEMD_GETOPT_OPTIONS
,
800 bool user_arg_seen
= false;
805 if (getpid_cached() == 1)
808 while ((c
= getopt_long(argc
, argv
, SYSTEMD_GETOPT_SHORT_OPTIONS
, options
, NULL
)) >= 0)
813 r
= log_set_max_level_from_string(optarg
);
815 return log_error_errno(r
, "Failed to parse log level \"%s\": %m", optarg
);
820 r
= log_set_target_from_string(optarg
);
822 return log_error_errno(r
, "Failed to parse log target \"%s\": %m", optarg
);
829 r
= log_show_color_from_string(optarg
);
831 return log_error_errno(r
, "Failed to parse log color setting \"%s\": %m",
834 log_show_color(true);
838 case ARG_LOG_LOCATION
:
840 r
= log_show_location_from_string(optarg
);
842 return log_error_errno(r
, "Failed to parse log location setting \"%s\": %m",
845 log_show_location(true);
852 r
= log_show_time_from_string(optarg
);
854 return log_error_errno(r
, "Failed to parse log time setting \"%s\": %m",
861 case ARG_DEFAULT_STD_OUTPUT
:
862 r
= exec_output_from_string(optarg
);
864 return log_error_errno(r
, "Failed to parse default standard output setting \"%s\": %m",
866 arg_defaults
.std_output
= r
;
869 case ARG_DEFAULT_STD_ERROR
:
870 r
= exec_output_from_string(optarg
);
872 return log_error_errno(r
, "Failed to parse default standard error output setting \"%s\": %m",
874 arg_defaults
.std_error
= r
;
878 r
= free_and_strdup(&arg_default_unit
, optarg
);
880 return log_error_errno(r
, "Failed to set default unit \"%s\": %m", optarg
);
885 arg_runtime_scope
= RUNTIME_SCOPE_SYSTEM
;
889 arg_runtime_scope
= RUNTIME_SCOPE_USER
;
890 user_arg_seen
= true;
894 arg_action
= ACTION_TEST
;
898 arg_pager_flags
|= PAGER_DISABLE
;
902 arg_action
= ACTION_VERSION
;
905 case ARG_DUMP_CONFIGURATION_ITEMS
:
906 arg_action
= ACTION_DUMP_CONFIGURATION_ITEMS
;
909 case ARG_DUMP_BUS_PROPERTIES
:
910 arg_action
= ACTION_DUMP_BUS_PROPERTIES
;
913 case ARG_BUS_INTROSPECT
:
914 arg_bus_introspect
= optarg
;
915 arg_action
= ACTION_BUS_INTROSPECT
;
919 r
= parse_boolean_argument("--dump-core", optarg
, &arg_dump_core
);
925 r
= parse_crash_chvt(optarg
, &arg_crash_chvt
);
927 return log_error_errno(r
, "Failed to parse crash virtual terminal index: \"%s\": %m",
931 case ARG_CRASH_SHELL
:
932 r
= parse_boolean_argument("--crash-shell", optarg
, &arg_crash_shell
);
937 case ARG_CRASH_REBOOT
:
938 r
= parse_boolean_argument("--crash-reboot", optarg
, &arg_crash_reboot
);
943 case ARG_CONFIRM_SPAWN
:
944 arg_confirm_spawn
= mfree(arg_confirm_spawn
);
946 r
= parse_confirm_spawn(optarg
, &arg_confirm_spawn
);
948 return log_error_errno(r
, "Failed to parse confirm spawn option: \"%s\": %m",
952 case ARG_SERVICE_WATCHDOGS
:
953 r
= parse_boolean_argument("--service-watchdogs=", optarg
, &arg_service_watchdogs
);
958 case ARG_SHOW_STATUS
:
960 r
= parse_show_status(optarg
, &arg_show_status
);
962 return log_error_errno(r
, "Failed to parse show status boolean: \"%s\": %m",
965 arg_show_status
= SHOW_STATUS_YES
;
968 case ARG_DESERIALIZE
: {
972 fd
= parse_fd(optarg
);
974 return log_error_errno(fd
, "Failed to parse serialization fd \"%s\": %m", optarg
);
976 (void) fd_cloexec(fd
, true);
980 return log_error_errno(errno
, "Failed to open serialization fd %d: %m", fd
);
982 safe_fclose(arg_serialization
);
983 arg_serialization
= f
;
988 case ARG_SWITCHED_ROOT
:
989 arg_switched_root
= true;
993 r
= id128_from_string_nonzero(optarg
, &arg_machine_id
);
995 return log_error_errno(r
, "MachineID '%s' is not valid: %m", optarg
);
999 arg_action
= ACTION_HELP
;
1003 log_set_max_level(LOG_DEBUG
);
1009 /* Just to eat away the sysvinit kernel cmdline args that we'll parse in
1010 * parse_proc_cmdline_item() or ignore, without any getopt() error messages.
1013 if (getpid_cached() != 1)
1019 assert_not_reached();
1022 if (optind
< argc
&& getpid_cached() != 1)
1023 /* Hmm, when we aren't run as init system let's complain about excess arguments */
1024 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Excess arguments.");
1026 if (arg_action
== ACTION_RUN
&& arg_runtime_scope
== RUNTIME_SCOPE_USER
&& !user_arg_seen
)
1027 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1028 "Explicit --user argument required to run as user manager.");
1033 static int help(void) {
1034 _cleanup_free_
char *link
= NULL
;
1037 r
= terminal_urlify_man("systemd", "1", &link
);
1041 printf("%s [OPTIONS...]\n\n"
1042 "%sStarts and monitors system and user services.%s\n\n"
1043 "This program takes no positional arguments.\n\n"
1045 " -h --help Show this help\n"
1046 " --version Show version\n"
1047 " --test Determine initial transaction, dump it and exit\n"
1048 " --system Combined with --test: operate in system mode\n"
1049 " --user Combined with --test: operate in user mode\n"
1050 " --dump-configuration-items Dump understood unit configuration items\n"
1051 " --dump-bus-properties Dump exposed bus properties\n"
1052 " --bus-introspect=PATH Write XML introspection data\n"
1053 " --unit=UNIT Set default unit\n"
1054 " --dump-core[=BOOL] Dump core on crash\n"
1055 " --crash-vt=NR Change to specified VT on crash\n"
1056 " --crash-reboot[=BOOL] Reboot on crash\n"
1057 " --crash-shell[=BOOL] Run shell on crash\n"
1058 " --confirm-spawn[=BOOL] Ask for confirmation when spawning processes\n"
1059 " --show-status[=BOOL] Show status updates on the console during boot\n"
1060 " --log-target=TARGET Set log target (console, journal, kmsg,\n"
1061 " journal-or-kmsg, null)\n"
1062 " --log-level=LEVEL Set log level (debug, info, notice, warning,\n"
1063 " err, crit, alert, emerg)\n"
1064 " --log-color[=BOOL] Highlight important log messages\n"
1065 " --log-location[=BOOL] Include code location in log messages\n"
1066 " --log-time[=BOOL] Prefix log messages with current time\n"
1067 " --default-standard-output= Set default standard output for services\n"
1068 " --default-standard-error= Set default standard error output for services\n"
1069 " --no-pager Do not pipe output into a pager\n"
1070 "\nSee the %s for details.\n",
1071 program_invocation_short_name
,
1081 static int prepare_reexecute(
1085 bool switching_root
) {
1087 _cleanup_fdset_free_ FDSet
*fds
= NULL
;
1088 _cleanup_fclose_
FILE *f
= NULL
;
1095 r
= manager_open_serialization(m
, &f
);
1097 return log_error_errno(r
, "Failed to create serialization file: %m");
1099 /* Make sure nothing is really destructed when we shut down */
1101 bus_manager_send_reloading(m
, true);
1107 r
= manager_serialize(m
, f
, fds
, switching_root
);
1111 if (fseeko(f
, 0, SEEK_SET
) < 0)
1112 return log_error_errno(errno
, "Failed to rewind serialization fd: %m");
1114 r
= fd_cloexec(fileno(f
), false);
1116 return log_error_errno(r
, "Failed to disable O_CLOEXEC for serialization: %m");
1118 r
= fdset_cloexec(fds
, false);
1120 return log_error_errno(r
, "Failed to disable O_CLOEXEC for serialization fds: %m");
1122 *ret_f
= TAKE_PTR(f
);
1123 *ret_fds
= TAKE_PTR(fds
);
1128 static void bump_file_max_and_nr_open(void) {
1130 /* Let's bump fs.file-max and fs.nr_open to their respective maximums. On current kernels large
1131 * numbers of file descriptors are no longer a performance problem and their memory is properly
1132 * tracked by memcg, thus counting them and limiting them in another two layers of limits is
1133 * unnecessary and just complicates things. This function hence turns off 2 of the 4 levels of limits
1134 * on file descriptors, and makes RLIMIT_NOLIMIT (soft + hard) the only ones that really matter. */
1136 #if BUMP_PROC_SYS_FS_FILE_MAX || BUMP_PROC_SYS_FS_NR_OPEN
1140 #if BUMP_PROC_SYS_FS_FILE_MAX
1141 /* The maximum the kernel allows for this since 5.2 is LONG_MAX, use that. (Previously things were
1142 * different, but the operation would fail silently.) */
1143 r
= sysctl_write("fs/file-max", LONG_MAX_STR
);
1145 log_full_errno(IN_SET(r
, -EROFS
, -EPERM
, -EACCES
) ? LOG_DEBUG
: LOG_WARNING
,
1146 r
, "Failed to bump fs.file-max, ignoring: %m");
1149 #if BUMP_PROC_SYS_FS_NR_OPEN
1152 /* Argh! The kernel enforces maximum and minimum values on the fs.nr_open, but we don't really know
1153 * what they are. The expression by which the maximum is determined is dependent on the architecture,
1154 * and is something we don't really want to copy to userspace, as it is dependent on implementation
1155 * details of the kernel. Since the kernel doesn't expose the maximum value to us, we can only try
1156 * and hope. Hence, let's start with INT_MAX, and then keep halving the value until we find one that
1157 * works. Ugly? Yes, absolutely, but kernel APIs are kernel APIs, so what do can we do... 🤯 */
1162 v
&= ~(__SIZEOF_POINTER__
- 1); /* Round down to next multiple of the pointer size */
1164 log_warning("Can't bump fs.nr_open, value too small.");
1170 log_error_errno(k
, "Failed to read fs.nr_open: %m");
1173 if (k
>= v
) { /* Already larger */
1174 log_debug("Skipping bump, value is already larger.");
1178 r
= sysctl_writef("fs/nr_open", "%i", v
);
1180 log_debug("Couldn't write fs.nr_open as %i, halving it.", v
);
1185 log_full_errno(IN_SET(r
, -EROFS
, -EPERM
, -EACCES
) ? LOG_DEBUG
: LOG_WARNING
, r
, "Failed to bump fs.nr_open, ignoring: %m");
1189 log_debug("Successfully bumped fs.nr_open to %i", v
);
1195 static int bump_rlimit_nofile(const struct rlimit
*saved_rlimit
) {
1196 struct rlimit new_rlimit
;
1199 /* Get the underlying absolute limit the kernel enforces */
1200 nr
= read_nr_open();
1202 /* Calculate the new limits to use for us. Never lower from what we inherited. */
1203 new_rlimit
= (struct rlimit
) {
1204 .rlim_cur
= MAX((rlim_t
) nr
, saved_rlimit
->rlim_cur
),
1205 .rlim_max
= MAX((rlim_t
) nr
, saved_rlimit
->rlim_max
),
1208 /* Shortcut if nothing changes. */
1209 if (saved_rlimit
->rlim_max
>= new_rlimit
.rlim_max
&&
1210 saved_rlimit
->rlim_cur
>= new_rlimit
.rlim_cur
) {
1211 log_debug("RLIMIT_NOFILE is already as high or higher than we need it, not bumping.");
1215 /* Bump up the resource limit for ourselves substantially, all the way to the maximum the kernel allows, for
1216 * both hard and soft. */
1217 r
= setrlimit_closest(RLIMIT_NOFILE
, &new_rlimit
);
1219 return log_warning_errno(r
, "Setting RLIMIT_NOFILE failed, ignoring: %m");
1224 static int bump_rlimit_memlock(const struct rlimit
*saved_rlimit
) {
1225 struct rlimit new_rlimit
;
1229 /* BPF_MAP_TYPE_LPM_TRIE bpf maps are charged against RLIMIT_MEMLOCK, even if we have CAP_IPC_LOCK
1230 * which should normally disable such checks. We need them to implement IPAddressAllow= and
1231 * IPAddressDeny=, hence let's bump the value high enough for our user. */
1233 /* Using MAX() on resource limits only is safe if RLIM_INFINITY is > 0. POSIX declares that rlim_t
1234 * must be unsigned, hence this is a given, but let's make this clear here. */
1235 assert_cc(RLIM_INFINITY
> 0);
1237 mm
= physical_memory_scale(1, 8); /* Let's scale how much we allow to be locked by the amount of
1238 * physical RAM. We allow an eighth to be locked by us, just to
1241 new_rlimit
= (struct rlimit
) {
1242 .rlim_cur
= MAX3(HIGH_RLIMIT_MEMLOCK
, saved_rlimit
->rlim_cur
, mm
),
1243 .rlim_max
= MAX3(HIGH_RLIMIT_MEMLOCK
, saved_rlimit
->rlim_max
, mm
),
1246 if (saved_rlimit
->rlim_max
>= new_rlimit
.rlim_cur
&&
1247 saved_rlimit
->rlim_cur
>= new_rlimit
.rlim_max
) {
1248 log_debug("RLIMIT_MEMLOCK is already as high or higher than we need it, not bumping.");
1252 r
= setrlimit_closest(RLIMIT_MEMLOCK
, &new_rlimit
);
1254 return log_warning_errno(r
, "Setting RLIMIT_MEMLOCK failed, ignoring: %m");
1259 static void test_usr(void) {
1261 /* Check that /usr is either on the same file system as / or mounted already. */
1263 if (dir_is_empty("/usr", /* ignore_hidden_or_backup= */ false) <= 0)
1266 log_warning("/usr appears to be on its own filesystem and is not already mounted. This is not a supported setup. "
1267 "Some things will probably break (sometimes even silently) in mysterious ways. "
1268 "Consult https://www.freedesktop.org/wiki/Software/systemd/separate-usr-is-broken for more information.");
1271 static int enforce_syscall_archs(Set
*archs
) {
1275 if (!is_seccomp_available())
1278 r
= seccomp_restrict_archs(arg_syscall_archs
);
1280 return log_error_errno(r
, "Failed to enforce system call architecture restrication: %m");
1285 static int os_release_status(void) {
1286 _cleanup_free_
char *pretty_name
= NULL
, *name
= NULL
, *version
= NULL
,
1287 *ansi_color
= NULL
, *support_end
= NULL
;
1290 r
= parse_os_release(NULL
,
1291 "PRETTY_NAME", &pretty_name
,
1293 "VERSION", &version
,
1294 "ANSI_COLOR", &ansi_color
,
1295 "SUPPORT_END", &support_end
);
1297 return log_full_errno(r
== -ENOENT
? LOG_DEBUG
: LOG_WARNING
, r
,
1298 "Failed to read os-release file, ignoring: %m");
1300 const char *label
= os_release_pretty_name(pretty_name
, name
);
1302 if (show_status_on(arg_show_status
)) {
1303 if (log_get_show_color())
1304 status_printf(NULL
, 0,
1305 "\nWelcome to \x1B[%sm%s\x1B[0m!\n",
1306 empty_to_null(ansi_color
) ?: "1",
1309 status_printf(NULL
, 0,
1310 "\nWelcome to %s!\n",
1314 if (support_end
&& os_release_support_ended(support_end
, /* quiet */ false, NULL
) > 0)
1315 /* pretty_name may include the version already, so we'll print the version only if we
1316 * have it and we're not using pretty_name. */
1317 status_printf(ANSI_HIGHLIGHT_RED
" !! " ANSI_NORMAL
, 0,
1318 "This OS version (%s%s%s) is past its end-of-support date (%s)",
1320 (pretty_name
|| !version
) ? "" : " version ",
1321 (pretty_name
|| !version
) ? "" : version
,
1327 static int setup_os_release(RuntimeScope scope
) {
1328 _cleanup_free_
char *os_release_dst
= NULL
;
1329 const char *os_release_src
= "/etc/os-release";
1332 if (access("/etc/os-release", F_OK
) < 0) {
1333 if (errno
!= ENOENT
)
1334 log_debug_errno(errno
, "Failed to check if /etc/os-release exists, ignoring: %m");
1336 os_release_src
= "/usr/lib/os-release";
1339 if (scope
== RUNTIME_SCOPE_SYSTEM
) {
1340 os_release_dst
= strdup("/run/systemd/propagate/.os-release-stage/os-release");
1341 if (!os_release_dst
)
1342 return log_oom_debug();
1344 if (asprintf(&os_release_dst
, "/run/user/" UID_FMT
"/systemd/propagate/.os-release-stage/os-release", geteuid()) < 0)
1345 return log_oom_debug();
1348 r
= mkdir_parents_label(os_release_dst
, 0755);
1350 return log_debug_errno(r
, "Failed to create parent directory of %s, ignoring: %m", os_release_dst
);
1352 r
= copy_file_atomic(os_release_src
, os_release_dst
, 0644, COPY_MAC_CREATE
|COPY_REPLACE
);
1354 return log_debug_errno(r
, "Failed to create %s, ignoring: %m", os_release_dst
);
1359 static int write_container_id(void) {
1361 int r
= 0; /* avoid false maybe-uninitialized warning */
1363 c
= getenv("container");
1368 r
= write_string_file("/run/systemd/container", c
, WRITE_STRING_FILE_CREATE
);
1370 return log_warning_errno(r
, "Failed to write /run/systemd/container, ignoring: %m");
1375 static int bump_unix_max_dgram_qlen(void) {
1376 _cleanup_free_
char *qlen
= NULL
;
1380 /* Let's bump the net.unix.max_dgram_qlen sysctl. The kernel default of 16 is simply too low. We set
1381 * the value really really early during boot, so that it is actually applied to all our sockets,
1382 * including the $NOTIFY_SOCKET one. */
1384 r
= read_one_line_file("/proc/sys/net/unix/max_dgram_qlen", &qlen
);
1386 return log_full_errno(r
== -ENOENT
? LOG_DEBUG
: LOG_WARNING
, r
,
1387 "Failed to read AF_UNIX datagram queue length, ignoring: %m");
1389 r
= safe_atolu(qlen
, &v
);
1391 return log_warning_errno(r
, "Failed to parse AF_UNIX datagram queue length '%s', ignoring: %m", qlen
);
1393 if (v
>= DEFAULT_UNIX_MAX_DGRAM_QLEN
)
1396 r
= sysctl_write("net/unix/max_dgram_qlen", STRINGIFY(DEFAULT_UNIX_MAX_DGRAM_QLEN
));
1398 return log_full_errno(IN_SET(r
, -EROFS
, -EPERM
, -EACCES
) ? LOG_DEBUG
: LOG_WARNING
, r
,
1399 "Failed to bump AF_UNIX datagram queue length, ignoring: %m");
1404 static int fixup_environment(void) {
1405 _cleanup_free_
char *term
= NULL
;
1409 /* Only fix up the environment when we are started as PID 1 */
1410 if (getpid_cached() != 1)
1413 /* We expect the environment to be set correctly if run inside a container. */
1414 if (detect_container() > 0)
1417 /* When started as PID1, the kernel uses /dev/console for our stdios and uses TERM=linux whatever the
1418 * backend device used by the console. We try to make a better guess here since some consoles might
1419 * not have support for color mode for example.
1421 * However if TERM was configured through the kernel command line then leave it alone. */
1422 r
= proc_cmdline_get_key("TERM", 0, &term
);
1427 r
= proc_cmdline_get_key("systemd.tty.term.console", 0, &term
);
1432 t
= term
?: default_term_for_tty("/dev/console");
1434 if (setenv("TERM", t
, 1) < 0)
1437 /* The kernels sets HOME=/ for init. Let's undo this. */
1438 if (path_equal_ptr(getenv("HOME"), "/"))
1439 assert_se(unsetenv("HOME") == 0);
1444 static void redirect_telinit(int argc
, char *argv
[]) {
1446 /* This is compatibility support for SysV, where calling init as a user is identical to telinit. */
1448 #if HAVE_SYSV_COMPAT
1449 if (getpid_cached() == 1)
1452 if (!invoked_as(argv
, "init"))
1455 execv(SYSTEMCTL_BINARY_PATH
, argv
);
1456 log_error_errno(errno
, "Failed to exec " SYSTEMCTL_BINARY_PATH
": %m");
1461 static int become_shutdown(int objective
, int retval
) {
1462 static const char* const table
[_MANAGER_OBJECTIVE_MAX
] = {
1463 [MANAGER_EXIT
] = "exit",
1464 [MANAGER_REBOOT
] = "reboot",
1465 [MANAGER_POWEROFF
] = "poweroff",
1466 [MANAGER_HALT
] = "halt",
1467 [MANAGER_KEXEC
] = "kexec",
1470 char log_level
[STRLEN("--log-level=") + DECIMAL_STR_MAX(int)],
1471 timeout
[STRLEN("--timeout=") + DECIMAL_STR_MAX(usec_t
) + STRLEN("us")],
1472 exit_code
[STRLEN("--exit-code=") + DECIMAL_STR_MAX(uint8_t)];
1474 _cleanup_strv_free_
char **env_block
= NULL
;
1475 usec_t watchdog_timer
= 0;
1478 assert(objective
>= 0 && objective
< _MANAGER_OBJECTIVE_MAX
);
1479 assert(table
[objective
]);
1481 xsprintf(log_level
, "--log-level=%d", log_get_max_level());
1482 xsprintf(timeout
, "--timeout=%" PRI_USEC
"us", arg_defaults
.timeout_stop_usec
);
1484 const char* command_line
[10] = {
1485 SYSTEMD_SHUTDOWN_BINARY_PATH
,
1489 /* Note that the last position is a terminator and must contain NULL. */
1493 assert(command_line
[pos
-1]);
1494 assert(!command_line
[pos
]);
1496 switch (log_get_target()) {
1498 case LOG_TARGET_KMSG
:
1499 case LOG_TARGET_JOURNAL_OR_KMSG
:
1500 case LOG_TARGET_SYSLOG_OR_KMSG
:
1501 command_line
[pos
++] = "--log-target=kmsg";
1504 case LOG_TARGET_NULL
:
1505 command_line
[pos
++] = "--log-target=null";
1508 case LOG_TARGET_CONSOLE
:
1510 command_line
[pos
++] = "--log-target=console";
1514 if (log_get_show_color())
1515 command_line
[pos
++] = "--log-color";
1517 if (log_get_show_location())
1518 command_line
[pos
++] = "--log-location";
1520 if (log_get_show_time())
1521 command_line
[pos
++] = "--log-time";
1523 xsprintf(exit_code
, "--exit-code=%d", retval
);
1524 command_line
[pos
++] = exit_code
;
1526 assert(pos
< ELEMENTSOF(command_line
));
1530 if (objective
== MANAGER_REBOOT
)
1531 watchdog_timer
= arg_reboot_watchdog
;
1532 else if (objective
== MANAGER_KEXEC
)
1533 watchdog_timer
= arg_kexec_watchdog
;
1535 /* If we reboot or kexec let's set the shutdown watchdog and tell the
1536 * shutdown binary to repeatedly ping it.
1537 * Disable the pretimeout watchdog, as we do not support it from the shutdown binary. */
1538 (void) watchdog_setup_pretimeout(0);
1539 (void) watchdog_setup_pretimeout_governor(NULL
);
1540 r
= watchdog_setup(watchdog_timer
);
1541 watchdog_close(r
< 0);
1543 /* The environment block: */
1545 env_block
= strv_copy(environ
);
1547 /* Tell the binary how often to ping, ignore failure */
1548 (void) strv_extendf(&env_block
, "WATCHDOG_USEC="USEC_FMT
, watchdog_timer
);
1550 if (arg_watchdog_device
)
1551 (void) strv_extendf(&env_block
, "WATCHDOG_DEVICE=%s", arg_watchdog_device
);
1553 /* Avoid the creation of new processes forked by the kernel; at this
1554 * point, we will not listen to the signals anyway */
1555 if (detect_container() <= 0)
1556 (void) cg_uninstall_release_agent(SYSTEMD_CGROUP_CONTROLLER
);
1558 execve(SYSTEMD_SHUTDOWN_BINARY_PATH
, (char **) command_line
, env_block
);
1562 static void initialize_clock(void) {
1565 /* This is called very early on, before we parse the kernel command line or otherwise figure out why
1566 * we are running, but only once. */
1568 if (clock_is_localtime(NULL
) > 0) {
1571 /* The very first call of settimeofday() also does a time warp in the kernel.
1573 * In the rtc-in-local time mode, we set the kernel's timezone, and rely on external tools to
1574 * take care of maintaining the RTC and do all adjustments. This matches the behavior of
1575 * Windows, which leaves the RTC alone if the registry tells that the RTC runs in UTC.
1577 r
= clock_set_timezone(&min
);
1579 log_error_errno(r
, "Failed to apply local time delta, ignoring: %m");
1581 log_info("RTC configured in localtime, applying delta of %i minutes to system time.", min
);
1583 } else if (!in_initrd())
1585 * Do a dummy very first call to seal the kernel's time warp magic.
1587 * Do not call this from inside the initrd. The initrd might not carry /etc/adjtime with
1588 * LOCAL, but the real system could be set up that way. In such case, we need to delay the
1589 * time-warp or the sealing until we reach the real system.
1591 * Do no set the kernel's timezone. The concept of local time cannot be supported reliably,
1592 * the time will jump or be incorrect at every daylight saving time change. All kernel local
1593 * time concepts will be treated as UTC that way.
1595 (void) clock_reset_timewarp();
1597 ClockChangeDirection change_dir
;
1598 r
= clock_apply_epoch(&change_dir
);
1599 if (r
> 0 && change_dir
== CLOCK_CHANGE_FORWARD
)
1600 log_info("System time before build time, advancing clock.");
1601 else if (r
> 0 && change_dir
== CLOCK_CHANGE_BACKWARD
)
1602 log_info("System time is further ahead than %s after build time, resetting clock to build time.",
1603 FORMAT_TIMESPAN(CLOCK_VALID_RANGE_USEC_MAX
, USEC_PER_DAY
));
1604 else if (r
< 0 && change_dir
== CLOCK_CHANGE_FORWARD
)
1605 log_error_errno(r
, "Current system time is before build time, but cannot correct: %m");
1606 else if (r
< 0 && change_dir
== CLOCK_CHANGE_BACKWARD
)
1607 log_error_errno(r
, "Current system time is further ahead %s after build time, but cannot correct: %m",
1608 FORMAT_TIMESPAN(CLOCK_VALID_RANGE_USEC_MAX
, USEC_PER_DAY
));
1611 static void apply_clock_update(void) {
1612 /* This is called later than initialize_clock(), i.e. after we parsed configuration files/kernel
1613 * command line and such. */
1615 if (arg_clock_usec
== 0)
1618 if (getpid_cached() != 1)
1621 if (clock_settime(CLOCK_REALTIME
, TIMESPEC_STORE(arg_clock_usec
)) < 0)
1622 log_error_errno(errno
, "Failed to set system clock to time specified on kernel command line: %m");
1624 log_info("Set system clock to %s, as specified on the kernel command line.",
1625 FORMAT_TIMESTAMP(arg_clock_usec
));
1628 static void cmdline_take_random_seed(void) {
1632 if (arg_random_seed_size
== 0)
1635 if (getpid_cached() != 1)
1638 assert(arg_random_seed
);
1639 suggested
= random_pool_size();
1641 if (arg_random_seed_size
< suggested
)
1642 log_warning("Random seed specified on kernel command line has size %zu, but %zu bytes required to fill entropy pool.",
1643 arg_random_seed_size
, suggested
);
1645 r
= random_write_entropy(-1, arg_random_seed
, arg_random_seed_size
, true);
1647 log_warning_errno(r
, "Failed to credit entropy specified on kernel command line, ignoring: %m");
1651 log_notice("Successfully credited entropy passed on kernel command line.\n"
1652 "Note that the seed provided this way is accessible to unprivileged programs. "
1653 "This functionality should not be used outside of testing environments.");
1656 static void initialize_coredump(bool skip_setup
) {
1657 if (getpid_cached() != 1)
1660 /* Don't limit the core dump size, so that coredump handlers such as systemd-coredump (which honour
1661 * the limit) will process core dumps for system services by default. */
1662 if (setrlimit(RLIMIT_CORE
, &RLIMIT_MAKE_CONST(RLIM_INFINITY
)) < 0)
1663 log_warning_errno(errno
, "Failed to set RLIMIT_CORE: %m");
1665 /* But at the same time, turn off the core_pattern logic by default, so that no coredumps are stored
1666 * until the systemd-coredump tool is enabled via sysctl. However it can be changed via the kernel
1667 * command line later so core dumps can still be generated during early startup and in initrd. */
1669 disable_coredumps();
1672 static void initialize_core_pattern(bool skip_setup
) {
1675 if (skip_setup
|| !arg_early_core_pattern
)
1678 if (getpid_cached() != 1)
1681 r
= write_string_file("/proc/sys/kernel/core_pattern", arg_early_core_pattern
, WRITE_STRING_FILE_DISABLE_BUFFER
);
1683 log_warning_errno(r
, "Failed to write '%s' to /proc/sys/kernel/core_pattern, ignoring: %m",
1684 arg_early_core_pattern
);
1687 static void update_cpu_affinity(bool skip_setup
) {
1688 _cleanup_free_
char *mask
= NULL
;
1690 if (skip_setup
|| !arg_cpu_affinity
.set
)
1693 assert(arg_cpu_affinity
.allocated
> 0);
1695 mask
= cpu_set_to_range_string(&arg_cpu_affinity
);
1696 log_debug("Setting CPU affinity to {%s}.", strnull(mask
));
1698 if (sched_setaffinity(0, arg_cpu_affinity
.allocated
, arg_cpu_affinity
.set
) < 0)
1699 log_warning_errno(errno
, "Failed to set CPU affinity, ignoring: %m");
1702 static void update_numa_policy(bool skip_setup
) {
1704 _cleanup_free_
char *nodes
= NULL
;
1705 const char * policy
= NULL
;
1707 if (skip_setup
|| !mpol_is_valid(numa_policy_get_type(&arg_numa_policy
)))
1710 if (DEBUG_LOGGING
) {
1711 policy
= mpol_to_string(numa_policy_get_type(&arg_numa_policy
));
1712 nodes
= cpu_set_to_range_string(&arg_numa_policy
.nodes
);
1713 log_debug("Setting NUMA policy to %s, with nodes {%s}.", strnull(policy
), strnull(nodes
));
1716 r
= apply_numa_policy(&arg_numa_policy
);
1717 if (r
== -EOPNOTSUPP
)
1718 log_debug_errno(r
, "NUMA support not available, ignoring.");
1720 log_warning_errno(r
, "Failed to set NUMA memory policy, ignoring: %m");
1723 static void filter_args(
1732 /* Copy some filtered arguments into the dst array from src. */
1733 for (int i
= 1; i
< argc
; i
++) {
1734 if (STR_IN_SET(src
[i
],
1740 if (startswith(src
[i
], "--deserialize="))
1742 if (streq(src
[i
], "--deserialize")) {
1743 i
++; /* Skip the argument too */
1747 /* Skip target unit designators. We already acted upon this information and have queued
1748 * appropriate jobs. We don't want to redo all this after reexecution. */
1749 if (startswith(src
[i
], "--unit="))
1751 if (streq(src
[i
], "--unit")) {
1752 i
++; /* Skip the argument too */
1756 /* Seems we have a good old option. Let's pass it over to the new instance. */
1757 dst
[(*dst_index
)++] = src
[i
];
1761 static void finish_remaining_processes(ManagerObjective objective
) {
1762 assert(objective
>= 0 && objective
< _MANAGER_OBJECTIVE_MAX
);
1764 /* Kill all remaining processes from the initrd, but don't wait for them, so that we can handle the
1765 * SIGCHLD for them after deserializing. */
1766 if (IN_SET(objective
, MANAGER_SWITCH_ROOT
, MANAGER_SOFT_REBOOT
))
1767 broadcast_signal(SIGTERM
, /* wait_for_exit= */ false, /* send_sighup= */ true, arg_defaults
.timeout_stop_usec
);
1769 /* On soft reboot really make sure nothing is left. Note that this will skip cgroups
1770 * of units that were configured with SurviveFinalKillSignal=yes. */
1771 if (objective
== MANAGER_SOFT_REBOOT
)
1772 broadcast_signal(SIGKILL
, /* wait_for_exit= */ false, /* send_sighup= */ false, arg_defaults
.timeout_stop_usec
);
1775 static int do_reexecute(
1776 ManagerObjective objective
,
1779 const struct rlimit
*saved_rlimit_nofile
,
1780 const struct rlimit
*saved_rlimit_memlock
,
1782 const char *switch_root_dir
,
1783 const char *switch_root_init
,
1784 const char **ret_error_message
) {
1786 size_t i
, args_size
;
1790 assert(IN_SET(objective
, MANAGER_REEXECUTE
, MANAGER_SWITCH_ROOT
, MANAGER_SOFT_REBOOT
));
1792 assert(saved_rlimit_nofile
);
1793 assert(saved_rlimit_memlock
);
1794 assert(ret_error_message
);
1796 if (switch_root_init
) {
1797 r
= chase(switch_root_init
, switch_root_dir
, CHASE_PREFIX_ROOT
, NULL
, NULL
);
1799 log_warning_errno(r
, "Failed to chase configured init %s/%s: %m",
1800 strempty(switch_root_dir
), switch_root_init
);
1802 r
= chase(SYSTEMD_BINARY_PATH
, switch_root_dir
, CHASE_PREFIX_ROOT
, NULL
, NULL
);
1804 log_debug_errno(r
, "Failed to chase our own binary %s/%s: %m",
1805 strempty(switch_root_dir
), SYSTEMD_BINARY_PATH
);
1809 r
= chase("/sbin/init", switch_root_dir
, CHASE_PREFIX_ROOT
, NULL
, NULL
);
1811 return log_error_errno(r
, "Failed to chase %s/sbin/init", strempty(switch_root_dir
));
1814 /* Close and disarm the watchdog, so that the new instance can reinitialize it, but doesn't get
1815 * rebooted while we do that */
1816 watchdog_close(true);
1818 /* Reset RLIMIT_NOFILE + RLIMIT_MEMLOCK back to the kernel defaults, so that the new systemd can pass
1819 * the kernel default to its child processes */
1820 if (saved_rlimit_nofile
->rlim_cur
!= 0)
1821 (void) setrlimit(RLIMIT_NOFILE
, saved_rlimit_nofile
);
1822 if (saved_rlimit_memlock
->rlim_cur
!= RLIM_INFINITY
)
1823 (void) setrlimit(RLIMIT_MEMLOCK
, saved_rlimit_memlock
);
1825 finish_remaining_processes(objective
);
1827 if (!switch_root_dir
&& objective
== MANAGER_SOFT_REBOOT
) {
1828 /* If no switch root dir is specified, then check if /run/nextroot/ qualifies and use that */
1829 r
= path_is_os_tree("/run/nextroot");
1830 if (r
< 0 && r
!= -ENOENT
)
1831 log_debug_errno(r
, "Failed to determine if /run/nextroot/ is a valid OS tree, ignoring: %m");
1833 switch_root_dir
= "/run/nextroot";
1836 if (switch_root_dir
) {
1837 r
= switch_root(/* new_root= */ switch_root_dir
,
1838 /* old_root_after= */ NULL
,
1839 /* flags= */ (objective
== MANAGER_SWITCH_ROOT
? SWITCH_ROOT_DESTROY_OLD_ROOT
: 0) |
1840 (objective
== MANAGER_SOFT_REBOOT
? 0 : SWITCH_ROOT_RECURSIVE_RUN
));
1842 log_error_errno(r
, "Failed to switch root, trying to continue: %m");
1845 args_size
= argc
+ 5;
1846 args
= newa(const char*, args_size
);
1848 if (!switch_root_init
) {
1849 char sfd
[STRLEN("--deserialize=") + DECIMAL_STR_MAX(int)];
1851 /* First try to spawn ourselves with the right path, and with full serialization. We do this
1852 * only if the user didn't specify an explicit init to spawn. */
1854 assert(arg_serialization
);
1857 xsprintf(sfd
, "--deserialize=%i", fileno(arg_serialization
));
1859 i
= 1; /* Leave args[0] empty for now. */
1861 /* Put our stuff first to make sure it always gets parsed in case
1862 * we get weird stuff from the kernel cmdline (like --) */
1863 if (IN_SET(objective
, MANAGER_SWITCH_ROOT
, MANAGER_SOFT_REBOOT
))
1864 args
[i
++] = "--switched-root";
1865 args
[i
++] = runtime_scope_cmdline_option_to_string(arg_runtime_scope
);
1868 filter_args(args
, &i
, argv
, argc
);
1872 assert(i
<= args_size
);
1875 * We want valgrind to print its memory usage summary before reexecution. Valgrind won't do
1876 * this is on its own on exec(), but it will do it on exit(). Hence, to ensure we get a
1877 * summary here, fork() off a child, let it exit() cleanly, so that it prints the summary,
1878 * and wait() for it in the parent, before proceeding into the exec().
1880 valgrind_summary_hack();
1882 args
[0] = SYSTEMD_BINARY_PATH
;
1883 (void) execv(args
[0], (char* const*) args
);
1885 if (objective
== MANAGER_REEXECUTE
) {
1886 *ret_error_message
= "Failed to execute our own binary";
1887 return log_error_errno(errno
, "Failed to execute our own binary %s: %m", args
[0]);
1890 log_debug_errno(errno
, "Failed to execute our own binary %s, trying fallback: %m", args
[0]);
1893 /* Try the fallback, if there is any, without any serialization. We pass the original argv[] and
1894 * envp[]. (Well, modulo the ordering changes due to getopt() in argv[], and some cleanups in envp[],
1895 * but let's hope that doesn't matter.) */
1897 arg_serialization
= safe_fclose(arg_serialization
);
1898 fds
= fdset_free(fds
);
1900 /* Reopen the console */
1901 (void) make_console_stdio();
1903 i
= 1; /* Leave args[0] empty for now. */
1904 for (int j
= 1; j
<= argc
; j
++)
1905 args
[i
++] = argv
[j
];
1906 assert(i
<= args_size
);
1908 /* Re-enable any blocked signals, especially important if we switch from initrd to init=... */
1909 (void) reset_all_signal_handlers();
1910 (void) reset_signal_mask();
1911 (void) rlimit_nofile_safe();
1913 if (switch_root_init
) {
1914 args
[0] = switch_root_init
;
1915 (void) execve(args
[0], (char* const*) args
, saved_env
);
1916 log_warning_errno(errno
, "Failed to execute configured init %s, trying fallback: %m", args
[0]);
1919 args
[0] = "/sbin/init";
1920 (void) execv(args
[0], (char* const*) args
);
1923 manager_status_printf(NULL
, STATUS_TYPE_EMERGENCY
,
1924 ANSI_HIGHLIGHT_RED
" !! " ANSI_NORMAL
,
1925 "Failed to execute /sbin/init");
1927 *ret_error_message
= "Failed to execute fallback shell";
1929 log_warning("No /sbin/init, trying fallback");
1931 args
[0] = "/bin/sh";
1933 (void) execve(args
[0], (char* const*) args
, saved_env
);
1934 return log_error_errno(errno
, "Failed to execute /bin/sh, giving up: %m");
1936 return log_error_errno(r
, "Failed to execute /sbin/init, giving up: %m");
1939 static int invoke_main_loop(
1941 const struct rlimit
*saved_rlimit_nofile
,
1942 const struct rlimit
*saved_rlimit_memlock
,
1943 int *ret_retval
, /* Return parameters relevant for shutting down */
1944 FDSet
**ret_fds
, /* Return parameters for reexecuting */
1945 char **ret_switch_root_dir
, /* … */
1946 char **ret_switch_root_init
, /* … */
1947 const char **ret_error_message
) {
1952 assert(saved_rlimit_nofile
);
1953 assert(saved_rlimit_memlock
);
1956 assert(ret_switch_root_dir
);
1957 assert(ret_switch_root_init
);
1958 assert(ret_error_message
);
1961 int objective
= manager_loop(m
);
1962 if (objective
< 0) {
1963 *ret_error_message
= "Failed to run main loop";
1964 return log_struct_errno(LOG_EMERG
, objective
,
1965 LOG_MESSAGE("Failed to run main loop: %m"),
1966 "MESSAGE_ID=" SD_MESSAGE_CORE_MAINLOOP_FAILED_STR
);
1969 switch (objective
) {
1971 case MANAGER_RELOAD
: {
1972 LogTarget saved_log_target
;
1973 int saved_log_level
;
1975 manager_send_reloading(m
);
1977 log_info("Reloading...");
1979 /* First, save any overridden log level/target, then parse the configuration file,
1980 * which might change the log level to new settings. */
1982 saved_log_level
= m
->log_level_overridden
? log_get_max_level() : -1;
1983 saved_log_target
= m
->log_target_overridden
? log_get_target() : _LOG_TARGET_INVALID
;
1985 (void) parse_configuration(saved_rlimit_nofile
, saved_rlimit_memlock
);
1987 set_manager_defaults(m
);
1988 set_manager_settings(m
);
1990 update_cpu_affinity(false);
1991 update_numa_policy(false);
1993 if (saved_log_level
>= 0)
1994 manager_override_log_level(m
, saved_log_level
);
1995 if (saved_log_target
>= 0)
1996 manager_override_log_target(m
, saved_log_target
);
1998 if (manager_reload(m
) < 0)
1999 /* Reloading failed before the point of no return.
2000 * Let's continue running as if nothing happened. */
2001 m
->objective
= MANAGER_OK
;
2003 log_info("Reloading finished in " USEC_FMT
" ms.",
2004 usec_sub_unsigned(now(CLOCK_MONOTONIC
), m
->timestamps
[MANAGER_TIMESTAMP_UNITS_LOAD
].monotonic
) / USEC_PER_MSEC
);
2009 case MANAGER_REEXECUTE
:
2011 manager_send_reloading(m
); /* From the perspective of the manager calling us this is
2012 * pretty much the same as a reload */
2014 r
= prepare_reexecute(m
, &arg_serialization
, ret_fds
, false);
2016 *ret_error_message
= "Failed to prepare for reexecution";
2020 log_notice("Reexecuting.");
2022 *ret_retval
= EXIT_SUCCESS
;
2023 *ret_switch_root_dir
= *ret_switch_root_init
= NULL
;
2027 case MANAGER_SWITCH_ROOT
:
2029 manager_send_reloading(m
); /* From the perspective of the manager calling us this is
2030 * pretty much the same as a reload */
2032 manager_set_switching_root(m
, true);
2034 if (!m
->switch_root_init
) {
2035 r
= prepare_reexecute(m
, &arg_serialization
, ret_fds
, true);
2037 *ret_error_message
= "Failed to prepare for reexecution";
2043 log_notice("Switching root.");
2045 *ret_retval
= EXIT_SUCCESS
;
2047 /* Steal the switch root parameters */
2048 *ret_switch_root_dir
= TAKE_PTR(m
->switch_root
);
2049 *ret_switch_root_init
= TAKE_PTR(m
->switch_root_init
);
2053 case MANAGER_SOFT_REBOOT
:
2054 manager_send_reloading(m
);
2055 manager_set_switching_root(m
, true);
2057 r
= prepare_reexecute(m
, &arg_serialization
, ret_fds
, /* switching_root= */ true);
2059 *ret_error_message
= "Failed to prepare for reexecution";
2063 log_notice("Soft-rebooting.");
2065 *ret_retval
= EXIT_SUCCESS
;
2066 *ret_switch_root_dir
= TAKE_PTR(m
->switch_root
);
2067 *ret_switch_root_init
= NULL
;
2072 if (MANAGER_IS_USER(m
)) {
2075 *ret_retval
= m
->return_value
;
2077 *ret_switch_root_dir
= *ret_switch_root_init
= NULL
;
2083 case MANAGER_REBOOT
:
2084 case MANAGER_POWEROFF
:
2086 case MANAGER_KEXEC
: {
2087 log_notice("Shutting down.");
2089 *ret_retval
= m
->return_value
;
2091 *ret_switch_root_dir
= *ret_switch_root_init
= NULL
;
2097 assert_not_reached();
2102 static void log_execution_mode(bool *ret_first_boot
) {
2103 bool first_boot
= false;
2106 assert(ret_first_boot
);
2108 switch (arg_runtime_scope
) {
2110 case RUNTIME_SCOPE_SYSTEM
: {
2114 log_info("systemd " GIT_VERSION
" running in %ssystem mode (%s)",
2115 arg_action
== ACTION_TEST
? "test " : "",
2118 v
= detect_virtualization();
2120 log_info("Detected virtualization %s.", virtualization_to_string(v
));
2122 v
= detect_confidential_virtualization();
2124 log_info("Detected confidential virtualization %s.", confidential_virtualization_to_string(v
));
2126 log_info("Detected architecture %s.", architecture_to_string(uname_architecture()));
2129 log_info("Running in initrd.");
2131 _cleanup_free_
char *id_text
= NULL
;
2133 /* Let's check whether we are in first boot. First, check if an override was
2134 * specified on the kernel command line. If yes, we honour that. */
2136 r
= proc_cmdline_get_bool("systemd.condition-first-boot", /* flags = */ 0, &first_boot
);
2138 log_debug_errno(r
, "Failed to parse systemd.condition-first-boot= kernel command line argument, ignoring: %m");
2141 log_full(first_boot
? LOG_INFO
: LOG_DEBUG
,
2142 "Kernel command line argument says we are %s first boot.",
2143 first_boot
? "in" : "not in");
2145 /* Second, perform autodetection. We use /etc/machine-id as flag file for
2146 * this: If it is missing or contains the value "uninitialized", this is the
2147 * first boot. In other cases, it is not. This allows container managers and
2148 * installers to provision a couple of files in /etc but still permit the
2149 * first-boot initialization to occur. If the container manager wants to
2150 * provision the machine ID it should pass $container_uuid to PID 1. */
2152 r
= read_one_line_file("/etc/machine-id", &id_text
);
2153 if (r
< 0 || streq(id_text
, "uninitialized")) {
2154 if (r
< 0 && r
!= -ENOENT
)
2155 log_warning_errno(r
, "Unexpected error while reading /etc/machine-id, assuming first boot: %m");
2158 log_info("Detected first boot.");
2160 log_debug("Detected initialized system, this is not the first boot.");
2164 assert_se(uname(&uts
) >= 0);
2166 if (strverscmp_improved(uts
.release
, KERNEL_BASELINE_VERSION
) < 0)
2167 log_warning("Warning! Reported kernel version %s is older than systemd's required baseline kernel version %s. "
2168 "Your mileage may vary.", uts
.release
, KERNEL_BASELINE_VERSION
);
2170 log_debug("Kernel version %s, our baseline is %s", uts
.release
, KERNEL_BASELINE_VERSION
);
2175 case RUNTIME_SCOPE_USER
:
2176 if (DEBUG_LOGGING
) {
2177 _cleanup_free_
char *t
= NULL
;
2179 t
= uid_to_name(getuid());
2180 log_debug("systemd " GIT_VERSION
" running in %suser mode for user " UID_FMT
"/%s. (%s)",
2181 arg_action
== ACTION_TEST
? " test" : "",
2182 getuid(), strna(t
), systemd_features
);
2188 assert_not_reached();
2191 *ret_first_boot
= first_boot
;
2194 static int initialize_runtime(
2197 struct rlimit
*saved_rlimit_nofile
,
2198 struct rlimit
*saved_rlimit_memlock
,
2199 const char **ret_error_message
) {
2202 assert(ret_error_message
);
2204 /* Sets up various runtime parameters. Many of these initializations are conditionalized:
2206 * - Some only apply to --system instances
2207 * - Some only apply to --user instances
2208 * - Some only apply when we first start up, but not when we reexecute
2211 if (arg_action
!= ACTION_RUN
)
2214 update_cpu_affinity(skip_setup
);
2215 update_numa_policy(skip_setup
);
2217 switch (arg_runtime_scope
) {
2219 case RUNTIME_SCOPE_SYSTEM
:
2220 /* Make sure we leave a core dump without panicking the kernel. */
2221 install_crash_handler();
2224 r
= mount_cgroup_controllers();
2226 *ret_error_message
= "Failed to mount cgroup hierarchies";
2230 /* Pull credentials from various sources into a common credential directory (we do
2231 * this here, before setting up the machine ID, so that we can use credential info
2232 * for setting up the machine ID) */
2233 (void) import_credentials();
2235 (void) os_release_status();
2236 (void) hostname_setup(true);
2237 /* Force transient machine-id on first boot. */
2238 machine_id_setup(/* root= */ NULL
, /* force_transient= */ first_boot
, arg_machine_id
, /* ret_machine_id */ NULL
);
2239 (void) loopback_setup();
2240 bump_unix_max_dgram_qlen();
2241 bump_file_max_and_nr_open();
2243 write_container_id();
2245 /* Copy os-release to the propagate directory, so that we update it for services running
2246 * under RootDirectory=/RootImage= when we do a soft reboot. */
2247 r
= setup_os_release(RUNTIME_SCOPE_SYSTEM
);
2249 log_warning_errno(r
, "Failed to copy os-release for propagation, ignoring: %m");
2252 r
= watchdog_set_device(arg_watchdog_device
);
2254 log_warning_errno(r
, "Failed to set watchdog device to %s, ignoring: %m", arg_watchdog_device
);
2258 case RUNTIME_SCOPE_USER
: {
2259 _cleanup_free_
char *p
= NULL
;
2261 /* Create the runtime directory and place the inaccessible device nodes there, if we run in
2262 * user mode. In system mode mount_setup() already did that. */
2264 r
= xdg_user_runtime_dir(&p
, "/systemd");
2266 *ret_error_message
= "$XDG_RUNTIME_DIR is not set";
2267 return log_struct_errno(LOG_EMERG
, r
,
2268 LOG_MESSAGE("Failed to determine $XDG_RUNTIME_DIR path: %m"),
2269 "MESSAGE_ID=" SD_MESSAGE_CORE_NO_XDGDIR_PATH_STR
);
2272 (void) mkdir_p_label(p
, 0755);
2273 (void) make_inaccessible_nodes(p
, UID_INVALID
, GID_INVALID
);
2274 r
= setup_os_release(RUNTIME_SCOPE_USER
);
2276 log_warning_errno(r
, "Failed to copy os-release for propagation, ignoring: %m");
2281 assert_not_reached();
2284 if (arg_timer_slack_nsec
!= NSEC_INFINITY
)
2285 if (prctl(PR_SET_TIMERSLACK
, arg_timer_slack_nsec
) < 0)
2286 log_warning_errno(errno
, "Failed to adjust timer slack, ignoring: %m");
2288 if (arg_runtime_scope
== RUNTIME_SCOPE_SYSTEM
) {
2290 if (!cap_test_all(arg_capability_bounding_set
)) {
2291 r
= capability_bounding_set_drop_usermode(arg_capability_bounding_set
);
2293 *ret_error_message
= "Failed to drop capability bounding set of usermode helpers";
2294 return log_struct_errno(LOG_EMERG
, r
,
2295 LOG_MESSAGE("Failed to drop capability bounding set of usermode helpers: %m"),
2296 "MESSAGE_ID=" SD_MESSAGE_CORE_CAPABILITY_BOUNDING_USER_STR
);
2299 r
= capability_bounding_set_drop(arg_capability_bounding_set
, true);
2301 *ret_error_message
= "Failed to drop capability bounding set";
2302 return log_struct_errno(LOG_EMERG
, r
,
2303 LOG_MESSAGE("Failed to drop capability bounding set: %m"),
2304 "MESSAGE_ID=" SD_MESSAGE_CORE_CAPABILITY_BOUNDING_STR
);
2308 if (arg_no_new_privs
) {
2309 if (prctl(PR_SET_NO_NEW_PRIVS
, 1, 0, 0, 0) < 0) {
2310 *ret_error_message
= "Failed to disable new privileges";
2311 return log_struct_errno(LOG_EMERG
, errno
,
2312 LOG_MESSAGE("Failed to disable new privileges: %m"),
2313 "MESSAGE_ID=" SD_MESSAGE_CORE_DISABLE_PRIVILEGES_STR
);
2318 if (arg_syscall_archs
) {
2319 r
= enforce_syscall_archs(arg_syscall_archs
);
2321 *ret_error_message
= "Failed to set syscall architectures";
2326 r
= make_reaper_process(true);
2328 log_warning_errno(r
, "Failed to make us a subreaper, ignoring: %m");
2330 /* Bump up RLIMIT_NOFILE for systemd itself */
2331 (void) bump_rlimit_nofile(saved_rlimit_nofile
);
2332 (void) bump_rlimit_memlock(saved_rlimit_memlock
);
2337 static int do_queue_default_job(
2339 const char **ret_error_message
) {
2341 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
2347 if (arg_default_unit
)
2348 unit
= arg_default_unit
;
2349 else if (in_initrd())
2350 unit
= SPECIAL_INITRD_TARGET
;
2352 unit
= SPECIAL_DEFAULT_TARGET
;
2354 log_debug("Activating default unit: %s", unit
);
2356 r
= manager_load_startable_unit_or_warn(m
, unit
, NULL
, &target
);
2357 if (r
< 0 && in_initrd() && !arg_default_unit
) {
2358 /* Fall back to default.target, which we used to always use by default. Only do this if no
2359 * explicit configuration was given. */
2361 log_info("Falling back to " SPECIAL_DEFAULT_TARGET
".");
2363 r
= manager_load_startable_unit_or_warn(m
, SPECIAL_DEFAULT_TARGET
, NULL
, &target
);
2366 log_info("Falling back to " SPECIAL_RESCUE_TARGET
".");
2368 r
= manager_load_startable_unit_or_warn(m
, SPECIAL_RESCUE_TARGET
, NULL
, &target
);
2370 *ret_error_message
= r
== -ERFKILL
? SPECIAL_RESCUE_TARGET
" masked"
2371 : "Failed to load " SPECIAL_RESCUE_TARGET
;
2376 assert(target
->load_state
== UNIT_LOADED
);
2378 r
= manager_add_job(m
, JOB_START
, target
, JOB_ISOLATE
, NULL
, &error
, &job
);
2380 log_debug_errno(r
, "Default target could not be isolated, starting instead: %s", bus_error_message(&error
, r
));
2382 sd_bus_error_free(&error
);
2384 r
= manager_add_job(m
, JOB_START
, target
, JOB_REPLACE
, NULL
, &error
, &job
);
2386 *ret_error_message
= "Failed to start default target";
2387 return log_struct_errno(LOG_EMERG
, r
,
2388 LOG_MESSAGE("Failed to start default target: %s", bus_error_message(&error
, r
)),
2389 "MESSAGE_ID=" SD_MESSAGE_CORE_START_TARGET_FAILED_STR
);
2393 *ret_error_message
= "Failed to isolate default target";
2394 return log_struct_errno(LOG_EMERG
, r
,
2395 LOG_MESSAGE("Failed to isolate default target: %s", bus_error_message(&error
, r
)),
2396 "MESSAGE_ID=" SD_MESSAGE_CORE_ISOLATE_TARGET_FAILED_STR
);
2398 log_info("Queued %s job for default target %s.",
2399 job_type_to_string(job
->type
),
2400 unit_status_string(job
->unit
, NULL
));
2402 m
->default_unit_job_id
= job
->id
;
2407 static void save_rlimits(struct rlimit
*saved_rlimit_nofile
,
2408 struct rlimit
*saved_rlimit_memlock
) {
2410 assert(saved_rlimit_nofile
);
2411 assert(saved_rlimit_memlock
);
2413 if (getrlimit(RLIMIT_NOFILE
, saved_rlimit_nofile
) < 0)
2414 log_warning_errno(errno
, "Reading RLIMIT_NOFILE failed, ignoring: %m");
2416 if (getrlimit(RLIMIT_MEMLOCK
, saved_rlimit_memlock
) < 0)
2417 log_warning_errno(errno
, "Reading RLIMIT_MEMLOCK failed, ignoring: %m");
2420 static void fallback_rlimit_nofile(const struct rlimit
*saved_rlimit_nofile
) {
2423 if (arg_defaults
.rlimit
[RLIMIT_NOFILE
])
2426 /* Make sure forked processes get limits based on the original kernel setting */
2428 rl
= newdup(struct rlimit
, saved_rlimit_nofile
, 1);
2434 /* Bump the hard limit for system services to a substantially higher value. The default
2435 * hard limit current kernels set is pretty low (4K), mostly for historical
2436 * reasons. According to kernel developers, the fd handling in recent kernels has been
2437 * optimized substantially enough, so that we can bump the limit now, without paying too
2438 * high a price in memory or performance. Note however that we only bump the hard limit,
2439 * not the soft limit. That's because select() works the way it works, and chokes on fds
2440 * >= 1024. If we'd bump the soft limit globally, it might accidentally happen to
2441 * unexpecting programs that they get fds higher than what they can process using
2442 * select(). By only bumping the hard limit but leaving the low limit as it is we avoid
2443 * this pitfall: programs that are written by folks aware of the select() problem in mind
2444 * (and thus use poll()/epoll instead of select(), the way everybody should) can
2445 * explicitly opt into high fds by bumping their soft limit beyond 1024, to the hard limit
2447 if (arg_runtime_scope
== RUNTIME_SCOPE_SYSTEM
) {
2450 /* Get the underlying absolute limit the kernel enforces */
2451 nr
= read_nr_open();
2453 rl
->rlim_max
= MIN((rlim_t
) nr
, MAX(rl
->rlim_max
, (rlim_t
) HIGH_RLIMIT_NOFILE
));
2456 /* If for some reason we were invoked with a soft limit above 1024 (which should never
2457 * happen!, but who knows what we get passed in from pam_limit when invoked as --user
2458 * instance), then lower what we pass on to not confuse our children */
2459 rl
->rlim_cur
= MIN(rl
->rlim_cur
, (rlim_t
) FD_SETSIZE
);
2461 arg_defaults
.rlimit
[RLIMIT_NOFILE
] = rl
;
2464 static void fallback_rlimit_memlock(const struct rlimit
*saved_rlimit_memlock
) {
2467 /* Pass the original value down to invoked processes */
2469 if (arg_defaults
.rlimit
[RLIMIT_MEMLOCK
])
2472 rl
= newdup(struct rlimit
, saved_rlimit_memlock
, 1);
2478 if (arg_runtime_scope
== RUNTIME_SCOPE_SYSTEM
) {
2479 /* Raise the default limit to 8M also on old kernels and in containers (8M is the kernel
2480 * default for this since kernel 5.16) */
2481 rl
->rlim_max
= MAX(rl
->rlim_max
, (rlim_t
) DEFAULT_RLIMIT_MEMLOCK
);
2482 rl
->rlim_cur
= MAX(rl
->rlim_cur
, (rlim_t
) DEFAULT_RLIMIT_MEMLOCK
);
2485 arg_defaults
.rlimit
[RLIMIT_MEMLOCK
] = rl
;
2488 static void setenv_manager_environment(void) {
2491 STRV_FOREACH(p
, arg_manager_environment
) {
2492 log_debug("Setting '%s' in our own environment.", *p
);
2494 r
= putenv_dup(*p
, true);
2496 log_warning_errno(errno
, "Failed to setenv \"%s\", ignoring: %m", *p
);
2500 static void reset_arguments(void) {
2501 /* Frees/resets arg_* variables, with a few exceptions commented below. */
2503 arg_default_unit
= mfree(arg_default_unit
);
2505 /* arg_runtime_scope — ignore */
2507 arg_dump_core
= true;
2508 arg_crash_chvt
= -1;
2509 arg_crash_shell
= false;
2510 arg_crash_reboot
= false;
2511 arg_confirm_spawn
= mfree(arg_confirm_spawn
);
2512 arg_show_status
= _SHOW_STATUS_INVALID
;
2513 arg_status_unit_format
= STATUS_UNIT_FORMAT_DEFAULT
;
2514 arg_switched_root
= false;
2515 arg_pager_flags
= 0;
2516 arg_service_watchdogs
= true;
2518 unit_defaults_done(&arg_defaults
);
2519 unit_defaults_init(&arg_defaults
, arg_runtime_scope
);
2521 arg_runtime_watchdog
= 0;
2522 arg_reboot_watchdog
= 10 * USEC_PER_MINUTE
;
2523 arg_kexec_watchdog
= 0;
2524 arg_pretimeout_watchdog
= 0;
2525 arg_early_core_pattern
= mfree(arg_early_core_pattern
);
2526 arg_watchdog_device
= mfree(arg_watchdog_device
);
2527 arg_watchdog_pretimeout_governor
= mfree(arg_watchdog_pretimeout_governor
);
2529 arg_default_environment
= strv_free(arg_default_environment
);
2530 arg_manager_environment
= strv_free(arg_manager_environment
);
2532 arg_capability_bounding_set
= CAP_MASK_UNSET
;
2533 arg_no_new_privs
= false;
2534 arg_timer_slack_nsec
= NSEC_INFINITY
;
2536 arg_syscall_archs
= set_free(arg_syscall_archs
);
2538 /* arg_serialization — ignore */
2540 arg_machine_id
= (sd_id128_t
) {};
2541 arg_cad_burst_action
= EMERGENCY_ACTION_REBOOT_FORCE
;
2543 cpu_set_reset(&arg_cpu_affinity
);
2544 numa_policy_reset(&arg_numa_policy
);
2546 arg_random_seed
= mfree(arg_random_seed
);
2547 arg_random_seed_size
= 0;
2550 arg_reload_limit_interval_sec
= 0;
2551 arg_reload_limit_burst
= 0;
2554 static void determine_default_oom_score_adjust(void) {
2557 /* Run our services at slightly higher OOM score than ourselves. But let's be conservative here, and
2558 * do this only if we don't run as root (i.e. only if we are run in user mode, for an unprivileged
2561 if (arg_defaults
.oom_score_adjust_set
)
2567 r
= get_oom_score_adjust(&a
);
2569 return (void) log_warning_errno(r
, "Failed to determine current OOM score adjustment value, ignoring: %m");
2571 assert_cc(100 <= OOM_SCORE_ADJ_MAX
);
2572 b
= a
>= OOM_SCORE_ADJ_MAX
- 100 ? OOM_SCORE_ADJ_MAX
: a
+ 100;
2577 arg_defaults
.oom_score_adjust
= b
;
2578 arg_defaults
.oom_score_adjust_set
= true;
2581 static int parse_configuration(const struct rlimit
*saved_rlimit_nofile
,
2582 const struct rlimit
*saved_rlimit_memlock
) {
2585 assert(saved_rlimit_nofile
);
2586 assert(saved_rlimit_memlock
);
2588 /* Assign configuration defaults */
2591 r
= parse_config_file();
2593 log_warning_errno(r
, "Failed to parse config file, ignoring: %m");
2595 if (arg_runtime_scope
== RUNTIME_SCOPE_SYSTEM
) {
2596 r
= proc_cmdline_parse(parse_proc_cmdline_item
, NULL
, 0);
2598 log_warning_errno(r
, "Failed to parse kernel command line, ignoring: %m");
2601 /* Initialize some default rlimits for services if they haven't been configured */
2602 fallback_rlimit_nofile(saved_rlimit_nofile
);
2603 fallback_rlimit_memlock(saved_rlimit_memlock
);
2605 /* Note that this also parses bits from the kernel command line, including "debug". */
2606 log_parse_environment();
2608 /* Initialize the show status setting if it hasn't been set explicitly yet */
2609 if (arg_show_status
== _SHOW_STATUS_INVALID
)
2610 arg_show_status
= SHOW_STATUS_YES
;
2612 /* Slightly raise the OOM score for our services if we are running for unprivileged users. */
2613 determine_default_oom_score_adjust();
2615 /* Push variables into the manager environment block */
2616 setenv_manager_environment();
2618 /* Parse log environment variables again to take into account any new environment variables. */
2619 log_parse_environment();
2624 static int safety_checks(void) {
2626 if (getpid_cached() == 1 &&
2627 arg_action
!= ACTION_RUN
)
2628 return log_error_errno(SYNTHETIC_ERRNO(EPERM
),
2629 "Unsupported execution mode while PID 1.");
2631 if (getpid_cached() == 1 &&
2632 arg_runtime_scope
== RUNTIME_SCOPE_USER
)
2633 return log_error_errno(SYNTHETIC_ERRNO(EPERM
),
2634 "Can't run --user mode as PID 1.");
2636 if (arg_action
== ACTION_RUN
&&
2637 arg_runtime_scope
== RUNTIME_SCOPE_SYSTEM
&&
2638 getpid_cached() != 1)
2639 return log_error_errno(SYNTHETIC_ERRNO(EPERM
),
2640 "Can't run system mode unless PID 1.");
2642 if (arg_action
== ACTION_TEST
&&
2644 return log_error_errno(SYNTHETIC_ERRNO(EPERM
),
2645 "Don't run test mode as root.");
2647 switch (arg_runtime_scope
) {
2649 case RUNTIME_SCOPE_USER
:
2651 if (arg_action
== ACTION_RUN
&&
2653 return log_error_errno(SYNTHETIC_ERRNO(EOPNOTSUPP
),
2654 "Trying to run as user instance, but the system has not been booted with systemd.");
2656 if (arg_action
== ACTION_RUN
&&
2657 !getenv("XDG_RUNTIME_DIR"))
2658 return log_error_errno(SYNTHETIC_ERRNO(EUNATCH
),
2659 "Trying to run as user instance, but $XDG_RUNTIME_DIR is not set.");
2663 case RUNTIME_SCOPE_SYSTEM
:
2664 if (arg_action
== ACTION_RUN
&&
2665 running_in_chroot() > 0)
2666 return log_error_errno(SYNTHETIC_ERRNO(EOPNOTSUPP
),
2667 "Cannot be run in a chroot() environment.");
2671 assert_not_reached();
2677 static int initialize_security(
2678 bool *loaded_policy
,
2679 dual_timestamp
*security_start_timestamp
,
2680 dual_timestamp
*security_finish_timestamp
,
2681 const char **ret_error_message
) {
2685 assert(loaded_policy
);
2686 assert(security_start_timestamp
);
2687 assert(security_finish_timestamp
);
2688 assert(ret_error_message
);
2690 dual_timestamp_now(security_start_timestamp
);
2692 r
= mac_selinux_setup(loaded_policy
);
2694 *ret_error_message
= "Failed to load SELinux policy";
2698 r
= mac_smack_setup(loaded_policy
);
2700 *ret_error_message
= "Failed to load SMACK policy";
2704 r
= mac_apparmor_setup();
2706 *ret_error_message
= "Failed to load AppArmor policy";
2712 *ret_error_message
= "Failed to load IMA policy";
2716 dual_timestamp_now(security_finish_timestamp
);
2720 static int collect_fds(FDSet
**ret_fds
, const char **ret_error_message
) {
2724 assert(ret_error_message
);
2726 /* Pick up all fds passed to us. We apply a filter here: we only take the fds that have O_CLOEXEC
2727 * off. All fds passed via execve() to us must have O_CLOEXEC off, and our own code and dependencies
2728 * should be clean enough to set O_CLOEXEC universally. Thus checking the bit should be a safe
2729 * mechanism to distinguish passed in fds from our own.
2731 * Why bother? Some subsystems we initialize early, specifically selinux might keep fds open in our
2732 * process behind our back. We should not take possession of that (and then accidentally close
2733 * it). SELinux thankfully sets O_CLOEXEC on its fds, so this test should work. */
2734 r
= fdset_new_fill(/* filter_cloexec= */ 0, ret_fds
);
2736 *ret_error_message
= "Failed to allocate fd set";
2737 return log_struct_errno(LOG_EMERG
, r
,
2738 LOG_MESSAGE("Failed to allocate fd set: %m"),
2739 "MESSAGE_ID=" SD_MESSAGE_CORE_FD_SET_FAILED_STR
);
2742 (void) fdset_cloexec(*ret_fds
, true);
2744 /* The serialization fd should have O_CLOEXEC turned on already, let's verify that we didn't pick it up here */
2745 assert_se(!arg_serialization
|| !fdset_contains(*ret_fds
, fileno(arg_serialization
)));
2750 static void setup_console_terminal(bool skip_setup
) {
2752 if (arg_runtime_scope
!= RUNTIME_SCOPE_SYSTEM
)
2755 /* Become a session leader if we aren't one yet. */
2758 /* If we are init, we connect stdin/stdout/stderr to /dev/null and make sure we don't have a
2759 * controlling tty. */
2760 (void) release_terminal();
2762 /* Reset the console, but only if this is really init and we are freshly booted */
2763 if (getpid_cached() == 1 && !skip_setup
)
2764 (void) console_setup();
2767 static bool early_skip_setup_check(int argc
, char *argv
[]) {
2768 bool found_deserialize
= false;
2770 /* Determine if this is a reexecution or normal bootup. We do the full command line parsing much
2771 * later, so let's just have a quick peek here. Note that if we have switched root, do all the
2772 * special setup things anyway, even if in that case we also do deserialization. */
2774 for (int i
= 1; i
< argc
; i
++)
2775 if (streq(argv
[i
], "--switched-root"))
2776 return false; /* If we switched root, don't skip the setup. */
2777 else if (startswith(argv
[i
], "--deserialize=") || streq(argv
[i
], "--deserialize"))
2778 found_deserialize
= true;
2780 return found_deserialize
; /* When we are deserializing, then we are reexecuting, hence avoid the extensive setup */
2783 static int save_env(void) {
2786 l
= strv_copy(environ
);
2790 strv_free_and_replace(saved_env
, l
);
2794 int main(int argc
, char *argv
[]) {
2796 initrd_timestamp
= DUAL_TIMESTAMP_NULL
,
2797 userspace_timestamp
= DUAL_TIMESTAMP_NULL
,
2798 kernel_timestamp
= DUAL_TIMESTAMP_NULL
,
2799 security_start_timestamp
= DUAL_TIMESTAMP_NULL
,
2800 security_finish_timestamp
= DUAL_TIMESTAMP_NULL
;
2801 struct rlimit saved_rlimit_nofile
= RLIMIT_MAKE_CONST(0),
2802 saved_rlimit_memlock
= RLIMIT_MAKE_CONST(RLIM_INFINITY
); /* The original rlimits we passed
2803 * in. Note we use different values
2804 * for the two that indicate whether
2805 * these fields are initialized! */
2806 bool skip_setup
, loaded_policy
= false, queue_default_job
= false, first_boot
= false;
2807 char *switch_root_dir
= NULL
, *switch_root_init
= NULL
;
2808 usec_t before_startup
, after_startup
;
2809 static char systemd
[] = "systemd";
2810 const char *error_message
= NULL
;
2811 int r
, retval
= EXIT_FAILURE
;
2815 assert_se(argc
> 0 && !isempty(argv
[0]));
2817 /* SysV compatibility: redirect init → telinit */
2818 redirect_telinit(argc
, argv
);
2820 /* Take timestamps early on */
2821 dual_timestamp_from_monotonic(&kernel_timestamp
, 0);
2822 dual_timestamp_now(&userspace_timestamp
);
2824 /* Figure out whether we need to do initialize the system, or if we already did that because we are
2826 skip_setup
= early_skip_setup_check(argc
, argv
);
2828 /* If we get started via the /sbin/init symlink then we are called 'init'. After a subsequent
2829 * reexecution we are then called 'systemd'. That is confusing, hence let's call us systemd
2831 program_invocation_short_name
= systemd
;
2832 (void) prctl(PR_SET_NAME
, systemd
);
2834 /* Save the original command line */
2835 save_argc_argv(argc
, argv
);
2837 /* Save the original environment as we might need to restore it if we're requested to execute another
2838 * system manager later. */
2841 error_message
= "Failed to copy environment block";
2845 /* Make sure that if the user says "syslog" we actually log to the journal. */
2846 log_set_upgrade_syslog_to_journal(true);
2848 if (getpid_cached() == 1) {
2849 /* When we run as PID 1 force system mode */
2850 arg_runtime_scope
= RUNTIME_SCOPE_SYSTEM
;
2852 /* Disable the umask logic */
2855 /* Make sure that at least initially we do not ever log to journald/syslogd, because it might
2856 * not be activated yet (even though the log socket for it exists). */
2857 log_set_prohibit_ipc(true);
2859 /* Always reopen /dev/console when running as PID 1 or one of its pre-execve() children. This
2860 * is important so that we never end up logging to any foreign stderr, for example if we have
2861 * to log in a child process right before execve()'ing the actual binary, at a point in time
2862 * where socket activation stderr/stdout area already set up. */
2863 log_set_always_reopen_console(true);
2865 if (detect_container() <= 0) {
2867 /* Running outside of a container as PID 1 */
2868 log_set_target_and_open(LOG_TARGET_KMSG
);
2871 initrd_timestamp
= userspace_timestamp
;
2874 r
= mount_setup_early();
2876 error_message
= "Failed to mount early API filesystems";
2881 /* We might have just mounted /proc, so let's try to parse the kernel
2882 * command line log arguments immediately. */
2883 log_parse_environment();
2885 /* Let's open the log backend a second time, in case the first time didn't
2886 * work. Quite possibly we have mounted /dev just now, so /dev/kmsg became
2887 * available, and it previously wasn't. */
2891 disable_printk_ratelimit();
2893 r
= initialize_security(
2895 &security_start_timestamp
,
2896 &security_finish_timestamp
,
2902 if (mac_init() < 0) {
2903 error_message
= "Failed to initialize MAC support";
2910 /* Set the default for later on, but don't actually open the logs like this for
2911 * now. Note that if we are transitioning from the initrd there might still be
2912 * journal fd open, and we shouldn't attempt opening that before we parsed
2913 * /proc/cmdline which might redirect output elsewhere. */
2914 log_set_target(LOG_TARGET_JOURNAL_OR_KMSG
);
2917 /* Running inside a container, as PID 1 */
2918 log_set_target_and_open(LOG_TARGET_CONSOLE
);
2920 /* For later on, see above... */
2921 log_set_target(LOG_TARGET_JOURNAL
);
2923 /* clear the kernel timestamp, because we are in a container */
2924 kernel_timestamp
= DUAL_TIMESTAMP_NULL
;
2927 initialize_coredump(skip_setup
);
2929 r
= fixup_environment();
2931 log_struct_errno(LOG_EMERG
, r
,
2932 LOG_MESSAGE("Failed to fix up PID 1 environment: %m"),
2933 "MESSAGE_ID=" SD_MESSAGE_CORE_PID1_ENVIRONMENT_STR
);
2934 error_message
= "Failed to fix up PID1 environment";
2938 /* Try to figure out if we can use colors with the console. No need to do that for user
2939 * instances since they never log into the console. */
2940 log_show_color(colors_enabled());
2942 r
= make_null_stdio();
2944 log_warning_errno(r
, "Failed to redirect standard streams to /dev/null, ignoring: %m");
2946 /* Load the kernel modules early. */
2948 (void) kmod_setup();
2950 /* Mount /proc, /sys and friends, so that /proc/cmdline and /proc/$PID/fd is available. */
2951 r
= mount_setup(loaded_policy
, skip_setup
);
2953 error_message
= "Failed to mount API filesystems";
2957 /* The efivarfs is now mounted, let's lock down the system token. */
2958 lock_down_efi_variables();
2960 /* Cache command-line options passed from EFI variables */
2962 (void) cache_efi_options_variable();
2964 /* Running as user instance */
2965 arg_runtime_scope
= RUNTIME_SCOPE_USER
;
2966 log_set_always_reopen_console(true);
2967 log_set_target_and_open(LOG_TARGET_AUTO
);
2969 /* clear the kernel timestamp, because we are not PID 1 */
2970 kernel_timestamp
= DUAL_TIMESTAMP_NULL
;
2972 /* Clear ambient capabilities, so services do not inherit them implicitly. Dropping them does
2973 * not affect the permitted and effective sets which are important for the manager itself to
2975 capability_ambient_set_apply(0, /* also_inherit= */ false);
2977 if (mac_init() < 0) {
2978 error_message
= "Failed to initialize MAC support";
2983 /* Save the original RLIMIT_NOFILE/RLIMIT_MEMLOCK so that we can reset it later when
2984 * transitioning from the initrd to the main systemd or suchlike. */
2985 save_rlimits(&saved_rlimit_nofile
, &saved_rlimit_memlock
);
2987 /* Reset all signal handlers. */
2988 (void) reset_all_signal_handlers();
2989 (void) ignore_signals(SIGNALS_IGNORE
);
2991 (void) parse_configuration(&saved_rlimit_nofile
, &saved_rlimit_memlock
);
2993 r
= parse_argv(argc
, argv
);
2995 error_message
= "Failed to parse command line arguments";
2999 r
= safety_checks();
3003 if (IN_SET(arg_action
, ACTION_TEST
, ACTION_HELP
, ACTION_DUMP_CONFIGURATION_ITEMS
, ACTION_DUMP_BUS_PROPERTIES
, ACTION_BUS_INTROSPECT
))
3004 pager_open(arg_pager_flags
);
3006 if (arg_action
!= ACTION_RUN
)
3009 if (arg_action
== ACTION_HELP
) {
3010 retval
= help() < 0 ? EXIT_FAILURE
: EXIT_SUCCESS
;
3012 } else if (arg_action
== ACTION_VERSION
) {
3015 } else if (arg_action
== ACTION_DUMP_CONFIGURATION_ITEMS
) {
3016 unit_dump_config_items(stdout
);
3017 retval
= EXIT_SUCCESS
;
3019 } else if (arg_action
== ACTION_DUMP_BUS_PROPERTIES
) {
3020 dump_bus_properties(stdout
);
3021 retval
= EXIT_SUCCESS
;
3023 } else if (arg_action
== ACTION_BUS_INTROSPECT
) {
3024 r
= bus_manager_introspect_implementations(stdout
, arg_bus_introspect
);
3025 retval
= r
>= 0 ? EXIT_SUCCESS
: EXIT_FAILURE
;
3029 assert_se(IN_SET(arg_action
, ACTION_RUN
, ACTION_TEST
));
3031 /* Move out of the way, so that we won't block unmounts */
3032 assert_se(chdir("/") == 0);
3034 if (arg_action
== ACTION_RUN
) {
3036 /* Apply the systemd.clock_usec= kernel command line switch */
3037 apply_clock_update();
3039 /* Apply random seed from kernel command line */
3040 cmdline_take_random_seed();
3043 /* A core pattern might have been specified via the cmdline. */
3044 initialize_core_pattern(skip_setup
);
3046 /* Close logging fds, in order not to confuse collecting passed fds and terminal logic below */
3049 /* Remember open file descriptors for later deserialization */
3050 r
= collect_fds(&fds
, &error_message
);
3054 /* Give up any control of the console, but make sure its initialized. */
3055 setup_console_terminal(skip_setup
);
3057 /* Open the logging devices, if possible and necessary */
3061 log_execution_mode(&first_boot
);
3063 r
= initialize_runtime(skip_setup
,
3065 &saved_rlimit_nofile
,
3066 &saved_rlimit_memlock
,
3071 r
= manager_new(arg_runtime_scope
,
3072 arg_action
== ACTION_TEST
? MANAGER_TEST_FULL
: 0,
3075 log_struct_errno(LOG_EMERG
, r
,
3076 LOG_MESSAGE("Failed to allocate manager object: %m"),
3077 "MESSAGE_ID=" SD_MESSAGE_CORE_MANAGER_ALLOCATE_STR
);
3078 error_message
= "Failed to allocate manager object";
3082 m
->timestamps
[MANAGER_TIMESTAMP_KERNEL
] = kernel_timestamp
;
3083 m
->timestamps
[MANAGER_TIMESTAMP_INITRD
] = initrd_timestamp
;
3084 m
->timestamps
[MANAGER_TIMESTAMP_USERSPACE
] = userspace_timestamp
;
3085 m
->timestamps
[manager_timestamp_initrd_mangle(MANAGER_TIMESTAMP_SECURITY_START
)] = security_start_timestamp
;
3086 m
->timestamps
[manager_timestamp_initrd_mangle(MANAGER_TIMESTAMP_SECURITY_FINISH
)] = security_finish_timestamp
;
3088 set_manager_defaults(m
);
3089 set_manager_settings(m
);
3090 manager_set_first_boot(m
, first_boot
);
3091 manager_set_switching_root(m
, arg_switched_root
);
3093 /* Remember whether we should queue the default job */
3094 queue_default_job
= !arg_serialization
|| arg_switched_root
;
3096 before_startup
= now(CLOCK_MONOTONIC
);
3098 r
= manager_startup(m
, arg_serialization
, fds
, /* root= */ NULL
);
3100 error_message
= "Failed to start up manager";
3104 /* This will close all file descriptors that were opened, but not claimed by any unit. */
3105 fds
= fdset_free(fds
);
3106 arg_serialization
= safe_fclose(arg_serialization
);
3108 if (queue_default_job
) {
3109 r
= do_queue_default_job(m
, &error_message
);
3114 after_startup
= now(CLOCK_MONOTONIC
);
3116 log_full(arg_action
== ACTION_TEST
? LOG_INFO
: LOG_DEBUG
,
3117 "Loaded units and determined initial transaction in %s.",
3118 FORMAT_TIMESPAN(after_startup
- before_startup
, 100 * USEC_PER_MSEC
));
3120 if (arg_action
== ACTION_TEST
) {
3121 manager_test_summary(m
);
3122 retval
= EXIT_SUCCESS
;
3126 r
= invoke_main_loop(m
,
3127 &saved_rlimit_nofile
,
3128 &saved_rlimit_memlock
,
3134 assert(r
< 0 || IN_SET(r
, MANAGER_EXIT
, /* MANAGER_OK is not expected here. */
3138 MANAGER_SOFT_REBOOT
,
3142 MANAGER_SWITCH_ROOT
));
3148 arg_reboot_watchdog
= manager_get_watchdog(m
, WATCHDOG_REBOOT
);
3149 arg_kexec_watchdog
= manager_get_watchdog(m
, WATCHDOG_KEXEC
);
3150 m
= manager_free(m
);
3153 mac_selinux_finish();
3155 if (IN_SET(r
, MANAGER_REEXECUTE
, MANAGER_SWITCH_ROOT
, MANAGER_SOFT_REBOOT
))
3158 &saved_rlimit_nofile
,
3159 &saved_rlimit_memlock
,
3163 &error_message
); /* This only returns if reexecution failed */
3165 arg_serialization
= safe_fclose(arg_serialization
);
3166 fds
= fdset_free(fds
);
3168 saved_env
= strv_free(saved_env
);
3170 #if HAVE_VALGRIND_VALGRIND_H
3171 /* If we are PID 1 and running under valgrind, then let's exit
3172 * here explicitly. valgrind will only generate nice output on
3173 * exit(), not on exec(), hence let's do the former not the
3175 if (getpid_cached() == 1 && RUNNING_ON_VALGRIND
) {
3176 /* Cleanup watchdog_device strings for valgrind. We need them
3177 * in become_shutdown() so normally we cannot free them yet. */
3178 watchdog_free_device();
3184 #if HAS_FEATURE_ADDRESS_SANITIZER
3185 /* At this stage we most likely don't have stdio/stderr open, so the following
3186 * LSan check would not print any actionable information and would just crash
3187 * PID 1. To make this a bit more helpful, let's try to open /dev/console,
3188 * and if we succeed redirect LSan's report there. */
3189 if (getpid_cached() == 1) {
3190 _cleanup_close_
int tty_fd
= -EBADF
;
3192 tty_fd
= open_terminal("/dev/console", O_WRONLY
|O_NOCTTY
|O_CLOEXEC
);
3194 __sanitizer_set_report_fd((void*) (intptr_t) tty_fd
);
3196 __lsan_do_leak_check();
3201 (void) sd_notifyf(0, "ERRNO=%i", -r
);
3203 /* Try to invoke the shutdown binary unless we already failed.
3204 * If we failed above, we want to freeze after finishing cleanup. */
3205 if (arg_runtime_scope
== RUNTIME_SCOPE_SYSTEM
&&
3206 IN_SET(r
, MANAGER_EXIT
, MANAGER_REBOOT
, MANAGER_POWEROFF
, MANAGER_HALT
, MANAGER_KEXEC
)) {
3207 r
= become_shutdown(r
, retval
);
3208 log_error_errno(r
, "Failed to execute shutdown binary, %s: %m", getpid_cached() == 1 ? "freezing" : "quitting");
3209 error_message
= "Failed to execute shutdown binary";
3212 /* This is primarily useful when running systemd in a VM, as it provides the user running the VM with
3213 * a mechanism to pick up systemd's exit status in the VM. */
3214 (void) sd_notifyf(0, "EXIT_STATUS=%i", retval
);
3216 watchdog_free_device();
3217 arg_watchdog_device
= mfree(arg_watchdog_device
);
3219 if (getpid_cached() == 1) {
3221 manager_status_printf(NULL
, STATUS_TYPE_EMERGENCY
,
3222 ANSI_HIGHLIGHT_RED
"!!!!!!" ANSI_NORMAL
,
3223 "%s.", error_message
);
3224 freeze_or_exit_or_reboot();