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"
71 #include "mount-util.h"
74 #include "parse-argument.h"
75 #include "parse-util.h"
76 #include "path-util.h"
77 #include "pretty-print.h"
78 #include "proc-cmdline.h"
79 #include "process-util.h"
81 #include "random-util.h"
82 #include "rlimit-util.h"
83 #include "seccomp-util.h"
84 #include "selinux-setup.h"
85 #include "selinux-util.h"
86 #include "signal-util.h"
87 #include "smack-setup.h"
89 #include "stat-util.h"
90 #include "stdio-util.h"
92 #include "switch-root.h"
93 #include "sysctl-util.h"
94 #include "terminal-util.h"
95 #include "time-util.h"
96 #include "umask-util.h"
97 #include "user-util.h"
100 #include "watchdog.h"
102 #if HAS_FEATURE_ADDRESS_SANITIZER
103 #include <sanitizer/lsan_interface.h>
111 ACTION_DUMP_CONFIGURATION_ITEMS
,
112 ACTION_DUMP_BUS_PROPERTIES
,
113 ACTION_BUS_INTROSPECT
,
114 } arg_action
= ACTION_RUN
;
116 static const char *arg_bus_introspect
= NULL
;
118 /* Those variables are initialized to 0 automatically, so we avoid uninitialized memory access. Real
119 * defaults are assigned in reset_arguments() below. */
120 static char *arg_default_unit
;
121 static RuntimeScope arg_runtime_scope
;
124 bool arg_crash_shell
;
125 bool arg_crash_reboot
;
126 static char *arg_confirm_spawn
;
127 static ShowStatus arg_show_status
;
128 static StatusUnitFormat arg_status_unit_format
;
129 static bool arg_switched_root
;
130 static PagerFlags arg_pager_flags
;
131 static bool arg_service_watchdogs
;
132 static UnitDefaults arg_defaults
;
133 static usec_t arg_runtime_watchdog
;
134 static usec_t arg_reboot_watchdog
;
135 static usec_t arg_kexec_watchdog
;
136 static usec_t arg_pretimeout_watchdog
;
137 static char *arg_early_core_pattern
;
138 static char *arg_watchdog_pretimeout_governor
;
139 static char *arg_watchdog_device
;
140 static char **arg_default_environment
;
141 static char **arg_manager_environment
;
142 static uint64_t arg_capability_bounding_set
;
143 static bool arg_no_new_privs
;
144 static int arg_protect_system
;
145 static nsec_t arg_timer_slack_nsec
;
146 static Set
* arg_syscall_archs
;
147 static FILE* arg_serialization
;
148 static sd_id128_t arg_machine_id
;
149 static EmergencyAction arg_cad_burst_action
;
150 static CPUSet arg_cpu_affinity
;
151 static NUMAPolicy arg_numa_policy
;
152 static usec_t arg_clock_usec
;
153 static void *arg_random_seed
;
154 static size_t arg_random_seed_size
;
155 static usec_t arg_reload_limit_interval_sec
;
156 static unsigned arg_reload_limit_burst
;
158 /* A copy of the original environment block */
159 static char **saved_env
= NULL
;
161 static int parse_configuration(const struct rlimit
*saved_rlimit_nofile
,
162 const struct rlimit
*saved_rlimit_memlock
);
164 static int manager_find_user_config_paths(char ***ret_files
, char ***ret_dirs
) {
165 _cleanup_free_
char *base
= NULL
;
166 _cleanup_strv_free_
char **files
= NULL
, **dirs
= NULL
;
169 r
= xdg_user_config_dir(&base
, "/systemd");
173 r
= strv_extendf(&files
, "%s/user.conf", base
);
177 r
= strv_extend(&files
, PKGSYSCONFDIR
"/user.conf");
181 r
= strv_consume(&dirs
, TAKE_PTR(base
));
185 r
= strv_extend_strv(&dirs
, CONF_PATHS_STRV("systemd"), false);
189 *ret_files
= TAKE_PTR(files
);
190 *ret_dirs
= TAKE_PTR(dirs
);
194 static int console_setup(void) {
195 _cleanup_close_
int tty_fd
= -EBADF
;
199 tty_fd
= open_terminal("/dev/console", O_WRONLY
|O_NOCTTY
|O_CLOEXEC
);
201 return log_error_errno(tty_fd
, "Failed to open /dev/console: %m");
203 /* We don't want to force text mode. plymouth may be showing
204 * pictures already from initrd. */
205 r
= reset_terminal_fd(tty_fd
, false);
207 return log_error_errno(r
, "Failed to reset /dev/console: %m");
209 r
= proc_cmdline_tty_size("/dev/console", &rows
, &cols
);
211 log_warning_errno(r
, "Failed to get terminal size, ignoring: %m");
213 r
= terminal_set_size_fd(tty_fd
, NULL
, rows
, cols
);
215 log_warning_errno(r
, "Failed to set terminal size, ignoring: %m");
221 static int parse_proc_cmdline_item(const char *key
, const char *value
, void *data
) {
226 if (STR_IN_SET(key
, "systemd.unit", "rd.systemd.unit")) {
228 if (proc_cmdline_value_missing(key
, value
))
231 if (!unit_name_is_valid(value
, UNIT_NAME_PLAIN
|UNIT_NAME_INSTANCE
))
232 log_warning("Unit name specified on %s= is not valid, ignoring: %s", key
, value
);
233 else if (in_initrd() == !!startswith(key
, "rd."))
234 return free_and_strdup_warn(&arg_default_unit
, value
);
236 } else if (proc_cmdline_key_streq(key
, "systemd.dump_core")) {
238 r
= value
? parse_boolean(value
) : true;
240 log_warning_errno(r
, "Failed to parse dump core switch %s, ignoring: %m", value
);
244 } else if (proc_cmdline_key_streq(key
, "systemd.early_core_pattern")) {
246 if (proc_cmdline_value_missing(key
, value
))
249 if (path_is_absolute(value
))
250 (void) parse_path_argument(value
, false, &arg_early_core_pattern
);
252 log_warning("Specified core pattern '%s' is not an absolute path, ignoring.", value
);
254 } else if (proc_cmdline_key_streq(key
, "systemd.crash_chvt")) {
257 arg_crash_chvt
= 0; /* turn on */
259 r
= parse_crash_chvt(value
, &arg_crash_chvt
);
261 log_warning_errno(r
, "Failed to parse crash chvt switch %s, ignoring: %m", value
);
264 } else if (proc_cmdline_key_streq(key
, "systemd.crash_shell")) {
266 r
= value
? parse_boolean(value
) : true;
268 log_warning_errno(r
, "Failed to parse crash shell switch %s, ignoring: %m", value
);
272 } else if (proc_cmdline_key_streq(key
, "systemd.crash_reboot")) {
274 r
= value
? parse_boolean(value
) : true;
276 log_warning_errno(r
, "Failed to parse crash reboot switch %s, ignoring: %m", value
);
278 arg_crash_reboot
= r
;
280 } else if (proc_cmdline_key_streq(key
, "systemd.confirm_spawn")) {
283 r
= parse_confirm_spawn(value
, &s
);
285 log_warning_errno(r
, "Failed to parse confirm_spawn switch %s, ignoring: %m", value
);
287 free_and_replace(arg_confirm_spawn
, s
);
289 } else if (proc_cmdline_key_streq(key
, "systemd.service_watchdogs")) {
291 r
= value
? parse_boolean(value
) : true;
293 log_warning_errno(r
, "Failed to parse service watchdog switch %s, ignoring: %m", value
);
295 arg_service_watchdogs
= r
;
297 } else if (proc_cmdline_key_streq(key
, "systemd.show_status")) {
300 r
= parse_show_status(value
, &arg_show_status
);
302 log_warning_errno(r
, "Failed to parse show status switch %s, ignoring: %m", value
);
304 arg_show_status
= SHOW_STATUS_YES
;
306 } else if (proc_cmdline_key_streq(key
, "systemd.status_unit_format")) {
308 if (proc_cmdline_value_missing(key
, value
))
311 r
= status_unit_format_from_string(value
);
313 log_warning_errno(r
, "Failed to parse %s=%s, ignoring: %m", key
, value
);
315 arg_status_unit_format
= r
;
317 } else if (proc_cmdline_key_streq(key
, "systemd.default_standard_output")) {
319 if (proc_cmdline_value_missing(key
, value
))
322 r
= exec_output_from_string(value
);
324 log_warning_errno(r
, "Failed to parse default standard output switch %s, ignoring: %m", value
);
326 arg_defaults
.std_output
= r
;
328 } else if (proc_cmdline_key_streq(key
, "systemd.default_standard_error")) {
330 if (proc_cmdline_value_missing(key
, value
))
333 r
= exec_output_from_string(value
);
335 log_warning_errno(r
, "Failed to parse default standard error switch %s, ignoring: %m", value
);
337 arg_defaults
.std_error
= r
;
339 } else if (streq(key
, "systemd.setenv")) {
341 if (proc_cmdline_value_missing(key
, value
))
344 if (!env_assignment_is_valid(value
))
345 log_warning("Environment variable assignment '%s' is not valid. Ignoring.", value
);
347 r
= strv_env_replace_strdup(&arg_default_environment
, value
);
352 } else if (proc_cmdline_key_streq(key
, "systemd.machine_id")) {
354 if (proc_cmdline_value_missing(key
, value
))
357 r
= id128_from_string_nonzero(value
, &arg_machine_id
);
359 log_warning_errno(r
, "MachineID '%s' is not valid, ignoring: %m", value
);
361 } else if (proc_cmdline_key_streq(key
, "systemd.default_timeout_start_sec")) {
363 if (proc_cmdline_value_missing(key
, value
))
366 r
= parse_sec(value
, &arg_defaults
.timeout_start_usec
);
368 log_warning_errno(r
, "Failed to parse default start timeout '%s', ignoring: %m", value
);
370 if (arg_defaults
.timeout_start_usec
<= 0)
371 arg_defaults
.timeout_start_usec
= USEC_INFINITY
;
373 } else if (proc_cmdline_key_streq(key
, "systemd.default_device_timeout_sec")) {
375 if (proc_cmdline_value_missing(key
, value
))
378 r
= parse_sec(value
, &arg_defaults
.device_timeout_usec
);
380 log_warning_errno(r
, "Failed to parse default device timeout '%s', ignoring: %m", value
);
382 if (arg_defaults
.device_timeout_usec
<= 0)
383 arg_defaults
.device_timeout_usec
= USEC_INFINITY
;
385 } else if (proc_cmdline_key_streq(key
, "systemd.cpu_affinity")) {
387 if (proc_cmdline_value_missing(key
, value
))
390 r
= parse_cpu_set(value
, &arg_cpu_affinity
);
392 log_warning_errno(r
, "Failed to parse CPU affinity mask '%s', ignoring: %m", value
);
394 } else if (proc_cmdline_key_streq(key
, "systemd.watchdog_device")) {
396 if (proc_cmdline_value_missing(key
, value
))
399 (void) parse_path_argument(value
, false, &arg_watchdog_device
);
401 } else if (proc_cmdline_key_streq(key
, "systemd.watchdog_sec")) {
403 if (proc_cmdline_value_missing(key
, value
))
406 if (streq(value
, "default"))
407 arg_runtime_watchdog
= USEC_INFINITY
;
408 else if (streq(value
, "off"))
409 arg_runtime_watchdog
= 0;
411 r
= parse_sec(value
, &arg_runtime_watchdog
);
413 log_warning_errno(r
, "Failed to parse systemd.watchdog_sec= argument '%s', ignoring: %m", value
);
418 arg_kexec_watchdog
= arg_reboot_watchdog
= arg_runtime_watchdog
;
420 } else if (proc_cmdline_key_streq(key
, "systemd.watchdog_pre_sec")) {
422 if (proc_cmdline_value_missing(key
, value
))
425 if (streq(value
, "default"))
426 arg_pretimeout_watchdog
= USEC_INFINITY
;
427 else if (streq(value
, "off"))
428 arg_pretimeout_watchdog
= 0;
430 r
= parse_sec(value
, &arg_pretimeout_watchdog
);
432 log_warning_errno(r
, "Failed to parse systemd.watchdog_pre_sec= argument '%s', ignoring: %m", value
);
437 } else if (proc_cmdline_key_streq(key
, "systemd.watchdog_pretimeout_governor")) {
439 if (proc_cmdline_value_missing(key
, value
) || isempty(value
)) {
440 arg_watchdog_pretimeout_governor
= mfree(arg_watchdog_pretimeout_governor
);
444 if (!string_is_safe(value
)) {
445 log_warning("Watchdog pretimeout governor '%s' is not valid, ignoring.", value
);
449 return free_and_strdup_warn(&arg_watchdog_pretimeout_governor
, value
);
451 } else if (proc_cmdline_key_streq(key
, "systemd.clock_usec")) {
453 if (proc_cmdline_value_missing(key
, value
))
456 r
= safe_atou64(value
, &arg_clock_usec
);
458 log_warning_errno(r
, "Failed to parse systemd.clock_usec= argument, ignoring: %s", value
);
460 } else if (proc_cmdline_key_streq(key
, "systemd.random_seed")) {
464 if (proc_cmdline_value_missing(key
, value
))
467 r
= unbase64mem(value
, SIZE_MAX
, &p
, &sz
);
469 log_warning_errno(r
, "Failed to parse systemd.random_seed= argument, ignoring: %s", value
);
471 free(arg_random_seed
);
472 arg_random_seed
= sz
> 0 ? p
: mfree(p
);
473 arg_random_seed_size
= sz
;
475 } else if (proc_cmdline_key_streq(key
, "systemd.reload_limit_interval_sec")) {
477 if (proc_cmdline_value_missing(key
, value
))
480 r
= parse_sec(value
, &arg_reload_limit_interval_sec
);
482 log_warning_errno(r
, "Failed to parse systemd.reload_limit_interval_sec= argument '%s', ignoring: %m", value
);
486 } else if (proc_cmdline_key_streq(key
, "systemd.reload_limit_burst")) {
488 if (proc_cmdline_value_missing(key
, value
))
491 r
= safe_atou(value
, &arg_reload_limit_burst
);
493 log_warning_errno(r
, "Failed to parse systemd.reload_limit_burst= argument '%s', ignoring: %m", value
);
497 } else if (streq(key
, "quiet") && !value
) {
499 if (arg_show_status
== _SHOW_STATUS_INVALID
)
500 arg_show_status
= SHOW_STATUS_ERROR
;
502 } else if (streq(key
, "debug") && !value
) {
504 /* Note that log_parse_environment() handles 'debug'
505 * too, and sets the log level to LOG_DEBUG. */
507 if (detect_container() > 0)
508 log_set_target(LOG_TARGET_CONSOLE
);
513 /* Compatible with SysV, but supported independently even if SysV compatibility is disabled. */
514 target
= runlevel_to_target(key
);
516 return free_and_strdup_warn(&arg_default_unit
, target
);
522 #define DEFINE_SETTER(name, func, descr) \
523 static int name(const char *unit, \
524 const char *filename, \
526 const char *section, \
527 unsigned section_line, \
528 const char *lvalue, \
530 const char *rvalue, \
542 log_syntax(unit, LOG_ERR, filename, line, r, \
543 "Invalid " descr "'%s': %m", \
549 DEFINE_SETTER(config_parse_level2
, log_set_max_level_from_string
, "log level");
550 DEFINE_SETTER(config_parse_target
, log_set_target_from_string
, "target");
551 DEFINE_SETTER(config_parse_color
, log_show_color_from_string
, "color");
552 DEFINE_SETTER(config_parse_location
, log_show_location_from_string
, "location");
553 DEFINE_SETTER(config_parse_time
, log_show_time_from_string
, "time");
555 static int config_parse_default_timeout_abort(
557 const char *filename
,
560 unsigned section_line
,
568 r
= config_parse_timeout_abort(
577 &arg_defaults
.timeout_abort_usec
,
580 arg_defaults
.timeout_abort_set
= r
;
584 static int config_parse_oom_score_adjust(
586 const char *filename
,
589 unsigned section_line
,
598 if (isempty(rvalue
)) {
599 arg_defaults
.oom_score_adjust_set
= false;
603 r
= parse_oom_score_adjust(rvalue
, &oa
);
605 log_syntax(unit
, LOG_WARNING
, filename
, line
, r
, "Failed to parse the OOM score adjust value '%s', ignoring: %m", rvalue
);
609 arg_defaults
.oom_score_adjust
= oa
;
610 arg_defaults
.oom_score_adjust_set
= true;
615 static int config_parse_protect_system_pid1(
617 const char *filename
,
620 unsigned section_line
,
627 int *v
= ASSERT_PTR(data
), r
;
629 /* This is modelled after the per-service ProtectSystem= setting, but a bit more restricted on one
630 * hand, and more automatic in another. i.e. we currently only support yes/no (not "strict" or
631 * "full"). And we will enable this automatically for the initrd unless configured otherwise.
633 * We might extend this later to match more closely what the per-service ProtectSystem= can do, but
634 * this is not trivial, due to ordering constraints: besides /usr/ we don't really have much mounted
635 * at the moment we enable this logic. */
637 if (isempty(rvalue
) || streq(rvalue
, "auto")) {
642 r
= parse_boolean(rvalue
);
644 log_syntax(unit
, LOG_WARNING
, filename
, line
, r
, "Failed to parse ProtectSystem= argument '%s', ignoring: %m", rvalue
);
652 static int parse_config_file(void) {
653 const ConfigTableItem items
[] = {
654 { "Manager", "LogLevel", config_parse_level2
, 0, NULL
},
655 { "Manager", "LogTarget", config_parse_target
, 0, NULL
},
656 { "Manager", "LogColor", config_parse_color
, 0, NULL
},
657 { "Manager", "LogLocation", config_parse_location
, 0, NULL
},
658 { "Manager", "LogTime", config_parse_time
, 0, NULL
},
659 { "Manager", "DumpCore", config_parse_bool
, 0, &arg_dump_core
},
660 { "Manager", "CrashChVT", /* legacy */ config_parse_crash_chvt
, 0, &arg_crash_chvt
},
661 { "Manager", "CrashChangeVT", config_parse_crash_chvt
, 0, &arg_crash_chvt
},
662 { "Manager", "CrashShell", config_parse_bool
, 0, &arg_crash_shell
},
663 { "Manager", "CrashReboot", config_parse_bool
, 0, &arg_crash_reboot
},
664 { "Manager", "ShowStatus", config_parse_show_status
, 0, &arg_show_status
},
665 { "Manager", "StatusUnitFormat", config_parse_status_unit_format
, 0, &arg_status_unit_format
},
666 { "Manager", "CPUAffinity", config_parse_cpu_affinity2
, 0, &arg_cpu_affinity
},
667 { "Manager", "NUMAPolicy", config_parse_numa_policy
, 0, &arg_numa_policy
.type
},
668 { "Manager", "NUMAMask", config_parse_numa_mask
, 0, &arg_numa_policy
},
669 { "Manager", "JoinControllers", config_parse_warn_compat
, DISABLED_CONFIGURATION
, NULL
},
670 { "Manager", "RuntimeWatchdogSec", config_parse_watchdog_sec
, 0, &arg_runtime_watchdog
},
671 { "Manager", "RuntimeWatchdogPreSec", config_parse_watchdog_sec
, 0, &arg_pretimeout_watchdog
},
672 { "Manager", "RebootWatchdogSec", config_parse_watchdog_sec
, 0, &arg_reboot_watchdog
},
673 { "Manager", "ShutdownWatchdogSec", config_parse_watchdog_sec
, 0, &arg_reboot_watchdog
}, /* obsolete alias */
674 { "Manager", "KExecWatchdogSec", config_parse_watchdog_sec
, 0, &arg_kexec_watchdog
},
675 { "Manager", "WatchdogDevice", config_parse_path
, 0, &arg_watchdog_device
},
676 { "Manager", "RuntimeWatchdogPreGovernor", config_parse_string
, CONFIG_PARSE_STRING_SAFE
, &arg_watchdog_pretimeout_governor
},
677 { "Manager", "CapabilityBoundingSet", config_parse_capability_set
, 0, &arg_capability_bounding_set
},
678 { "Manager", "NoNewPrivileges", config_parse_bool
, 0, &arg_no_new_privs
},
679 { "Manager", "ProtectSystem", config_parse_protect_system_pid1
, 0, &arg_protect_system
},
681 { "Manager", "SystemCallArchitectures", config_parse_syscall_archs
, 0, &arg_syscall_archs
},
683 { "Manager", "SystemCallArchitectures", config_parse_warn_compat
, DISABLED_CONFIGURATION
, NULL
},
686 { "Manager", "TimerSlackNSec", config_parse_nsec
, 0, &arg_timer_slack_nsec
},
687 { "Manager", "DefaultTimerAccuracySec", config_parse_sec
, 0, &arg_defaults
.timer_accuracy_usec
},
688 { "Manager", "DefaultStandardOutput", config_parse_output_restricted
, 0, &arg_defaults
.std_output
},
689 { "Manager", "DefaultStandardError", config_parse_output_restricted
, 0, &arg_defaults
.std_error
},
690 { "Manager", "DefaultTimeoutStartSec", config_parse_sec
, 0, &arg_defaults
.timeout_start_usec
},
691 { "Manager", "DefaultTimeoutStopSec", config_parse_sec
, 0, &arg_defaults
.timeout_stop_usec
},
692 { "Manager", "DefaultTimeoutAbortSec", config_parse_default_timeout_abort
, 0, NULL
},
693 { "Manager", "DefaultDeviceTimeoutSec", config_parse_sec
, 0, &arg_defaults
.device_timeout_usec
},
694 { "Manager", "DefaultRestartSec", config_parse_sec
, 0, &arg_defaults
.restart_usec
},
695 { "Manager", "DefaultStartLimitInterval", config_parse_sec
, 0, &arg_defaults
.start_limit_interval
}, /* obsolete alias */
696 { "Manager", "DefaultStartLimitIntervalSec", config_parse_sec
, 0, &arg_defaults
.start_limit_interval
},
697 { "Manager", "DefaultStartLimitBurst", config_parse_unsigned
, 0, &arg_defaults
.start_limit_burst
},
698 { "Manager", "DefaultEnvironment", config_parse_environ
, arg_runtime_scope
, &arg_default_environment
},
699 { "Manager", "ManagerEnvironment", config_parse_environ
, arg_runtime_scope
, &arg_manager_environment
},
700 { "Manager", "DefaultLimitCPU", config_parse_rlimit
, RLIMIT_CPU
, arg_defaults
.rlimit
},
701 { "Manager", "DefaultLimitFSIZE", config_parse_rlimit
, RLIMIT_FSIZE
, arg_defaults
.rlimit
},
702 { "Manager", "DefaultLimitDATA", config_parse_rlimit
, RLIMIT_DATA
, arg_defaults
.rlimit
},
703 { "Manager", "DefaultLimitSTACK", config_parse_rlimit
, RLIMIT_STACK
, arg_defaults
.rlimit
},
704 { "Manager", "DefaultLimitCORE", config_parse_rlimit
, RLIMIT_CORE
, arg_defaults
.rlimit
},
705 { "Manager", "DefaultLimitRSS", config_parse_rlimit
, RLIMIT_RSS
, arg_defaults
.rlimit
},
706 { "Manager", "DefaultLimitNOFILE", config_parse_rlimit
, RLIMIT_NOFILE
, arg_defaults
.rlimit
},
707 { "Manager", "DefaultLimitAS", config_parse_rlimit
, RLIMIT_AS
, arg_defaults
.rlimit
},
708 { "Manager", "DefaultLimitNPROC", config_parse_rlimit
, RLIMIT_NPROC
, arg_defaults
.rlimit
},
709 { "Manager", "DefaultLimitMEMLOCK", config_parse_rlimit
, RLIMIT_MEMLOCK
, arg_defaults
.rlimit
},
710 { "Manager", "DefaultLimitLOCKS", config_parse_rlimit
, RLIMIT_LOCKS
, arg_defaults
.rlimit
},
711 { "Manager", "DefaultLimitSIGPENDING", config_parse_rlimit
, RLIMIT_SIGPENDING
, arg_defaults
.rlimit
},
712 { "Manager", "DefaultLimitMSGQUEUE", config_parse_rlimit
, RLIMIT_MSGQUEUE
, arg_defaults
.rlimit
},
713 { "Manager", "DefaultLimitNICE", config_parse_rlimit
, RLIMIT_NICE
, arg_defaults
.rlimit
},
714 { "Manager", "DefaultLimitRTPRIO", config_parse_rlimit
, RLIMIT_RTPRIO
, arg_defaults
.rlimit
},
715 { "Manager", "DefaultLimitRTTIME", config_parse_rlimit
, RLIMIT_RTTIME
, arg_defaults
.rlimit
},
716 { "Manager", "DefaultCPUAccounting", config_parse_bool
, 0, &arg_defaults
.cpu_accounting
},
717 { "Manager", "DefaultIOAccounting", config_parse_bool
, 0, &arg_defaults
.io_accounting
},
718 { "Manager", "DefaultIPAccounting", config_parse_bool
, 0, &arg_defaults
.ip_accounting
},
719 { "Manager", "DefaultBlockIOAccounting", config_parse_bool
, 0, &arg_defaults
.blockio_accounting
},
720 { "Manager", "DefaultMemoryAccounting", config_parse_bool
, 0, &arg_defaults
.memory_accounting
},
721 { "Manager", "DefaultTasksAccounting", config_parse_bool
, 0, &arg_defaults
.tasks_accounting
},
722 { "Manager", "DefaultTasksMax", config_parse_tasks_max
, 0, &arg_defaults
.tasks_max
},
723 { "Manager", "DefaultMemoryPressureThresholdSec", config_parse_sec
, 0, &arg_defaults
.memory_pressure_threshold_usec
},
724 { "Manager", "DefaultMemoryPressureWatch", config_parse_memory_pressure_watch
, 0, &arg_defaults
.memory_pressure_watch
},
725 { "Manager", "CtrlAltDelBurstAction", config_parse_emergency_action
, arg_runtime_scope
, &arg_cad_burst_action
},
726 { "Manager", "DefaultOOMPolicy", config_parse_oom_policy
, 0, &arg_defaults
.oom_policy
},
727 { "Manager", "DefaultOOMScoreAdjust", config_parse_oom_score_adjust
, 0, NULL
},
728 { "Manager", "ReloadLimitIntervalSec", config_parse_sec
, 0, &arg_reload_limit_interval_sec
},
729 { "Manager", "ReloadLimitBurst", config_parse_unsigned
, 0, &arg_reload_limit_burst
},
731 { "Manager", "DefaultSmackProcessLabel", config_parse_string
, 0, &arg_defaults
.smack_process_label
},
733 { "Manager", "DefaultSmackProcessLabel", config_parse_warn_compat
, DISABLED_CONFIGURATION
, NULL
},
738 if (arg_runtime_scope
== RUNTIME_SCOPE_SYSTEM
)
739 (void) config_parse_config_file("system.conf",
741 config_item_table_lookup
, items
,
745 _cleanup_strv_free_
char **files
= NULL
, **dirs
= NULL
;
748 assert(arg_runtime_scope
== RUNTIME_SCOPE_USER
);
750 r
= manager_find_user_config_paths(&files
, &dirs
);
752 return log_error_errno(r
, "Failed to determine config file paths: %m");
754 (void) config_parse_many(
755 (const char* const*) files
,
756 (const char* const*) dirs
,
760 config_item_table_lookup
, items
,
765 /* Traditionally "0" was used to turn off the default unit timeouts. Fix this up so that we use
766 * USEC_INFINITY like everywhere else. */
767 if (arg_defaults
.timeout_start_usec
<= 0)
768 arg_defaults
.timeout_start_usec
= USEC_INFINITY
;
769 if (arg_defaults
.timeout_stop_usec
<= 0)
770 arg_defaults
.timeout_stop_usec
= USEC_INFINITY
;
775 static void set_manager_defaults(Manager
*m
) {
780 /* Propagates the various default unit property settings into the manager object, i.e. properties
781 * that do not affect the manager itself, but are just what newly allocated units will have set if
782 * they haven't set anything else. (Also see set_manager_settings() for the settings that affect the
783 * manager's own behaviour) */
785 r
= manager_set_unit_defaults(m
, &arg_defaults
);
787 log_warning_errno(r
, "Failed to set manager defaults, ignoring: %m");
789 r
= manager_default_environment(m
);
791 log_warning_errno(r
, "Failed to set manager default environment, ignoring: %m");
793 r
= manager_transient_environment_add(m
, arg_default_environment
);
795 log_warning_errno(r
, "Failed to add to transient environment, ignoring: %m");
798 static void set_manager_settings(Manager
*m
) {
803 /* Propagates the various manager settings into the manager object, i.e. properties that
804 * effect the manager itself (as opposed to just being inherited into newly allocated
805 * units, see set_manager_defaults() above). */
807 m
->confirm_spawn
= arg_confirm_spawn
;
808 m
->service_watchdogs
= arg_service_watchdogs
;
809 m
->cad_burst_action
= arg_cad_burst_action
;
810 /* Note that we don't do structured initialization here, otherwise it will reset the rate limit
811 * counter on every daemon-reload. */
812 m
->reload_ratelimit
.interval
= arg_reload_limit_interval_sec
;
813 m
->reload_ratelimit
.burst
= arg_reload_limit_burst
;
815 manager_set_watchdog(m
, WATCHDOG_RUNTIME
, arg_runtime_watchdog
);
816 manager_set_watchdog(m
, WATCHDOG_REBOOT
, arg_reboot_watchdog
);
817 manager_set_watchdog(m
, WATCHDOG_KEXEC
, arg_kexec_watchdog
);
818 manager_set_watchdog(m
, WATCHDOG_PRETIMEOUT
, arg_pretimeout_watchdog
);
819 r
= manager_set_watchdog_pretimeout_governor(m
, arg_watchdog_pretimeout_governor
);
821 log_warning_errno(r
, "Failed to set watchdog pretimeout governor to '%s', ignoring: %m", arg_watchdog_pretimeout_governor
);
823 manager_set_show_status(m
, arg_show_status
, "command line");
824 m
->status_unit_format
= arg_status_unit_format
;
827 static int parse_argv(int argc
, char *argv
[]) {
833 static const struct option options
[] = {
834 COMMON_GETOPT_OPTIONS
,
835 SYSTEMD_GETOPT_OPTIONS
,
840 bool user_arg_seen
= false;
845 if (getpid_cached() == 1)
848 while ((c
= getopt_long(argc
, argv
, SYSTEMD_GETOPT_SHORT_OPTIONS
, options
, NULL
)) >= 0)
853 r
= log_set_max_level_from_string(optarg
);
855 return log_error_errno(r
, "Failed to parse log level \"%s\": %m", optarg
);
860 r
= log_set_target_from_string(optarg
);
862 return log_error_errno(r
, "Failed to parse log target \"%s\": %m", optarg
);
869 r
= log_show_color_from_string(optarg
);
871 return log_error_errno(r
, "Failed to parse log color setting \"%s\": %m",
874 log_show_color(true);
878 case ARG_LOG_LOCATION
:
880 r
= log_show_location_from_string(optarg
);
882 return log_error_errno(r
, "Failed to parse log location setting \"%s\": %m",
885 log_show_location(true);
892 r
= log_show_time_from_string(optarg
);
894 return log_error_errno(r
, "Failed to parse log time setting \"%s\": %m",
901 case ARG_DEFAULT_STD_OUTPUT
:
902 r
= exec_output_from_string(optarg
);
904 return log_error_errno(r
, "Failed to parse default standard output setting \"%s\": %m",
906 arg_defaults
.std_output
= r
;
909 case ARG_DEFAULT_STD_ERROR
:
910 r
= exec_output_from_string(optarg
);
912 return log_error_errno(r
, "Failed to parse default standard error output setting \"%s\": %m",
914 arg_defaults
.std_error
= r
;
918 r
= free_and_strdup(&arg_default_unit
, optarg
);
920 return log_error_errno(r
, "Failed to set default unit \"%s\": %m", optarg
);
925 arg_runtime_scope
= RUNTIME_SCOPE_SYSTEM
;
929 arg_runtime_scope
= RUNTIME_SCOPE_USER
;
930 user_arg_seen
= true;
934 arg_action
= ACTION_TEST
;
938 arg_pager_flags
|= PAGER_DISABLE
;
942 arg_action
= ACTION_VERSION
;
945 case ARG_DUMP_CONFIGURATION_ITEMS
:
946 arg_action
= ACTION_DUMP_CONFIGURATION_ITEMS
;
949 case ARG_DUMP_BUS_PROPERTIES
:
950 arg_action
= ACTION_DUMP_BUS_PROPERTIES
;
953 case ARG_BUS_INTROSPECT
:
954 arg_bus_introspect
= optarg
;
955 arg_action
= ACTION_BUS_INTROSPECT
;
959 r
= parse_boolean_argument("--dump-core", optarg
, &arg_dump_core
);
965 r
= parse_crash_chvt(optarg
, &arg_crash_chvt
);
967 return log_error_errno(r
, "Failed to parse crash virtual terminal index: \"%s\": %m",
971 case ARG_CRASH_SHELL
:
972 r
= parse_boolean_argument("--crash-shell", optarg
, &arg_crash_shell
);
977 case ARG_CRASH_REBOOT
:
978 r
= parse_boolean_argument("--crash-reboot", optarg
, &arg_crash_reboot
);
983 case ARG_CONFIRM_SPAWN
:
984 arg_confirm_spawn
= mfree(arg_confirm_spawn
);
986 r
= parse_confirm_spawn(optarg
, &arg_confirm_spawn
);
988 return log_error_errno(r
, "Failed to parse confirm spawn option: \"%s\": %m",
992 case ARG_SERVICE_WATCHDOGS
:
993 r
= parse_boolean_argument("--service-watchdogs=", optarg
, &arg_service_watchdogs
);
998 case ARG_SHOW_STATUS
:
1000 r
= parse_show_status(optarg
, &arg_show_status
);
1002 return log_error_errno(r
, "Failed to parse show status boolean: \"%s\": %m",
1005 arg_show_status
= SHOW_STATUS_YES
;
1008 case ARG_DESERIALIZE
: {
1012 fd
= parse_fd(optarg
);
1014 return log_error_errno(fd
, "Failed to parse serialization fd \"%s\": %m", optarg
);
1016 (void) fd_cloexec(fd
, true);
1018 f
= fdopen(fd
, "r");
1020 return log_error_errno(errno
, "Failed to open serialization fd %d: %m", fd
);
1022 safe_fclose(arg_serialization
);
1023 arg_serialization
= f
;
1028 case ARG_SWITCHED_ROOT
:
1029 arg_switched_root
= true;
1032 case ARG_MACHINE_ID
:
1033 r
= id128_from_string_nonzero(optarg
, &arg_machine_id
);
1035 return log_error_errno(r
, "MachineID '%s' is not valid: %m", optarg
);
1039 arg_action
= ACTION_HELP
;
1043 log_set_max_level(LOG_DEBUG
);
1049 /* Just to eat away the sysvinit kernel cmdline args that we'll parse in
1050 * parse_proc_cmdline_item() or ignore, without any getopt() error messages.
1053 if (getpid_cached() != 1)
1059 assert_not_reached();
1062 if (optind
< argc
&& getpid_cached() != 1)
1063 /* Hmm, when we aren't run as init system let's complain about excess arguments */
1064 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Excess arguments.");
1066 if (arg_action
== ACTION_RUN
&& arg_runtime_scope
== RUNTIME_SCOPE_USER
&& !user_arg_seen
)
1067 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1068 "Explicit --user argument required to run as user manager.");
1073 static int help(void) {
1074 _cleanup_free_
char *link
= NULL
;
1077 r
= terminal_urlify_man("systemd", "1", &link
);
1081 printf("%s [OPTIONS...]\n\n"
1082 "%sStarts and monitors system and user services.%s\n\n"
1083 "This program takes no positional arguments.\n\n"
1085 " -h --help Show this help\n"
1086 " --version Show version\n"
1087 " --test Determine initial transaction, dump it and exit\n"
1088 " --system Combined with --test: operate in system mode\n"
1089 " --user Combined with --test: operate in user mode\n"
1090 " --dump-configuration-items Dump understood unit configuration items\n"
1091 " --dump-bus-properties Dump exposed bus properties\n"
1092 " --bus-introspect=PATH Write XML introspection data\n"
1093 " --unit=UNIT Set default unit\n"
1094 " --dump-core[=BOOL] Dump core on crash\n"
1095 " --crash-vt=NR Change to specified VT on crash\n"
1096 " --crash-reboot[=BOOL] Reboot on crash\n"
1097 " --crash-shell[=BOOL] Run shell on crash\n"
1098 " --confirm-spawn[=BOOL] Ask for confirmation when spawning processes\n"
1099 " --show-status[=BOOL] Show status updates on the console during boot\n"
1100 " --log-target=TARGET Set log target (console, journal, kmsg,\n"
1101 " journal-or-kmsg, null)\n"
1102 " --log-level=LEVEL Set log level (debug, info, notice, warning,\n"
1103 " err, crit, alert, emerg)\n"
1104 " --log-color[=BOOL] Highlight important log messages\n"
1105 " --log-location[=BOOL] Include code location in log messages\n"
1106 " --log-time[=BOOL] Prefix log messages with current time\n"
1107 " --default-standard-output= Set default standard output for services\n"
1108 " --default-standard-error= Set default standard error output for services\n"
1109 " --no-pager Do not pipe output into a pager\n"
1110 "\nSee the %s for details.\n",
1111 program_invocation_short_name
,
1121 static int prepare_reexecute(
1125 bool switching_root
) {
1127 _cleanup_fdset_free_ FDSet
*fds
= NULL
;
1128 _cleanup_fclose_
FILE *f
= NULL
;
1135 r
= manager_open_serialization(m
, &f
);
1137 return log_error_errno(r
, "Failed to create serialization file: %m");
1139 /* Make sure nothing is really destructed when we shut down */
1141 bus_manager_send_reloading(m
, true);
1147 r
= manager_serialize(m
, f
, fds
, switching_root
);
1151 if (fseeko(f
, 0, SEEK_SET
) < 0)
1152 return log_error_errno(errno
, "Failed to rewind serialization fd: %m");
1154 r
= fd_cloexec(fileno(f
), false);
1156 return log_error_errno(r
, "Failed to disable O_CLOEXEC for serialization: %m");
1158 r
= fdset_cloexec(fds
, false);
1160 return log_error_errno(r
, "Failed to disable O_CLOEXEC for serialization fds: %m");
1162 *ret_f
= TAKE_PTR(f
);
1163 *ret_fds
= TAKE_PTR(fds
);
1168 static void bump_file_max_and_nr_open(void) {
1170 /* Let's bump fs.file-max and fs.nr_open to their respective maximums. On current kernels large
1171 * numbers of file descriptors are no longer a performance problem and their memory is properly
1172 * tracked by memcg, thus counting them and limiting them in another two layers of limits is
1173 * unnecessary and just complicates things. This function hence turns off 2 of the 4 levels of limits
1174 * on file descriptors, and makes RLIMIT_NOLIMIT (soft + hard) the only ones that really matter. */
1176 #if BUMP_PROC_SYS_FS_FILE_MAX || BUMP_PROC_SYS_FS_NR_OPEN
1180 #if BUMP_PROC_SYS_FS_FILE_MAX
1181 /* The maximum the kernel allows for this since 5.2 is LONG_MAX, use that. (Previously things were
1182 * different, but the operation would fail silently.) */
1183 r
= sysctl_write("fs/file-max", LONG_MAX_STR
);
1185 log_full_errno(IN_SET(r
, -EROFS
, -EPERM
, -EACCES
) ? LOG_DEBUG
: LOG_WARNING
,
1186 r
, "Failed to bump fs.file-max, ignoring: %m");
1189 #if BUMP_PROC_SYS_FS_NR_OPEN
1192 /* Argh! The kernel enforces maximum and minimum values on the fs.nr_open, but we don't really know
1193 * what they are. The expression by which the maximum is determined is dependent on the architecture,
1194 * and is something we don't really want to copy to userspace, as it is dependent on implementation
1195 * details of the kernel. Since the kernel doesn't expose the maximum value to us, we can only try
1196 * and hope. Hence, let's start with INT_MAX, and then keep halving the value until we find one that
1197 * works. Ugly? Yes, absolutely, but kernel APIs are kernel APIs, so what do can we do... 🤯 */
1202 v
&= ~(__SIZEOF_POINTER__
- 1); /* Round down to next multiple of the pointer size */
1204 log_warning("Can't bump fs.nr_open, value too small.");
1210 log_error_errno(k
, "Failed to read fs.nr_open: %m");
1213 if (k
>= v
) { /* Already larger */
1214 log_debug("Skipping bump, value is already larger.");
1218 r
= sysctl_writef("fs/nr_open", "%i", v
);
1220 log_debug("Couldn't write fs.nr_open as %i, halving it.", v
);
1225 log_full_errno(IN_SET(r
, -EROFS
, -EPERM
, -EACCES
) ? LOG_DEBUG
: LOG_WARNING
, r
, "Failed to bump fs.nr_open, ignoring: %m");
1229 log_debug("Successfully bumped fs.nr_open to %i", v
);
1235 static int bump_rlimit_nofile(const struct rlimit
*saved_rlimit
) {
1236 struct rlimit new_rlimit
;
1239 /* Get the underlying absolute limit the kernel enforces */
1240 nr
= read_nr_open();
1242 /* Calculate the new limits to use for us. Never lower from what we inherited. */
1243 new_rlimit
= (struct rlimit
) {
1244 .rlim_cur
= MAX((rlim_t
) nr
, saved_rlimit
->rlim_cur
),
1245 .rlim_max
= MAX((rlim_t
) nr
, saved_rlimit
->rlim_max
),
1248 /* Shortcut if nothing changes. */
1249 if (saved_rlimit
->rlim_max
>= new_rlimit
.rlim_max
&&
1250 saved_rlimit
->rlim_cur
>= new_rlimit
.rlim_cur
) {
1251 log_debug("RLIMIT_NOFILE is already as high or higher than we need it, not bumping.");
1255 /* Bump up the resource limit for ourselves substantially, all the way to the maximum the kernel allows, for
1256 * both hard and soft. */
1257 r
= setrlimit_closest(RLIMIT_NOFILE
, &new_rlimit
);
1259 return log_warning_errno(r
, "Setting RLIMIT_NOFILE failed, ignoring: %m");
1264 static int bump_rlimit_memlock(const struct rlimit
*saved_rlimit
) {
1265 struct rlimit new_rlimit
;
1269 /* BPF_MAP_TYPE_LPM_TRIE bpf maps are charged against RLIMIT_MEMLOCK, even if we have CAP_IPC_LOCK
1270 * which should normally disable such checks. We need them to implement IPAddressAllow= and
1271 * IPAddressDeny=, hence let's bump the value high enough for our user. */
1273 /* Using MAX() on resource limits only is safe if RLIM_INFINITY is > 0. POSIX declares that rlim_t
1274 * must be unsigned, hence this is a given, but let's make this clear here. */
1275 assert_cc(RLIM_INFINITY
> 0);
1277 mm
= physical_memory_scale(1, 8); /* Let's scale how much we allow to be locked by the amount of
1278 * physical RAM. We allow an eighth to be locked by us, just to
1281 new_rlimit
= (struct rlimit
) {
1282 .rlim_cur
= MAX3(HIGH_RLIMIT_MEMLOCK
, saved_rlimit
->rlim_cur
, mm
),
1283 .rlim_max
= MAX3(HIGH_RLIMIT_MEMLOCK
, saved_rlimit
->rlim_max
, mm
),
1286 if (saved_rlimit
->rlim_max
>= new_rlimit
.rlim_cur
&&
1287 saved_rlimit
->rlim_cur
>= new_rlimit
.rlim_max
) {
1288 log_debug("RLIMIT_MEMLOCK is already as high or higher than we need it, not bumping.");
1292 r
= setrlimit_closest(RLIMIT_MEMLOCK
, &new_rlimit
);
1294 return log_warning_errno(r
, "Setting RLIMIT_MEMLOCK failed, ignoring: %m");
1299 static void test_usr(void) {
1301 /* Check that /usr is either on the same file system as / or mounted already. */
1303 if (dir_is_empty("/usr", /* ignore_hidden_or_backup= */ false) <= 0)
1306 log_warning("/usr appears to be on its own filesystem and is not already mounted. This is not a supported setup. "
1307 "Some things will probably break (sometimes even silently) in mysterious ways. "
1308 "Consult https://www.freedesktop.org/wiki/Software/systemd/separate-usr-is-broken for more information.");
1311 static int enforce_syscall_archs(Set
*archs
) {
1315 if (!is_seccomp_available())
1318 r
= seccomp_restrict_archs(arg_syscall_archs
);
1320 return log_error_errno(r
, "Failed to enforce system call architecture restrication: %m");
1325 static int os_release_status(void) {
1326 _cleanup_free_
char *pretty_name
= NULL
, *name
= NULL
, *version
= NULL
,
1327 *ansi_color
= NULL
, *support_end
= NULL
;
1330 r
= parse_os_release(NULL
,
1331 "PRETTY_NAME", &pretty_name
,
1333 "VERSION", &version
,
1334 "ANSI_COLOR", &ansi_color
,
1335 "SUPPORT_END", &support_end
);
1337 return log_full_errno(r
== -ENOENT
? LOG_DEBUG
: LOG_WARNING
, r
,
1338 "Failed to read os-release file, ignoring: %m");
1340 const char *label
= os_release_pretty_name(pretty_name
, name
);
1342 if (show_status_on(arg_show_status
)) {
1343 if (log_get_show_color())
1344 status_printf(NULL
, 0,
1345 "\nWelcome to \x1B[%sm%s\x1B[0m!\n",
1346 empty_to_null(ansi_color
) ?: "1",
1349 status_printf(NULL
, 0,
1350 "\nWelcome to %s!\n",
1354 if (support_end
&& os_release_support_ended(support_end
, /* quiet */ false, NULL
) > 0)
1355 /* pretty_name may include the version already, so we'll print the version only if we
1356 * have it and we're not using pretty_name. */
1357 status_printf(ANSI_HIGHLIGHT_RED
" !! " ANSI_NORMAL
, 0,
1358 "This OS version (%s%s%s) is past its end-of-support date (%s)",
1360 (pretty_name
|| !version
) ? "" : " version ",
1361 (pretty_name
|| !version
) ? "" : version
,
1367 static int setup_os_release(RuntimeScope scope
) {
1368 _cleanup_free_
char *os_release_dst
= NULL
;
1369 const char *os_release_src
= "/etc/os-release";
1372 if (access("/etc/os-release", F_OK
) < 0) {
1373 if (errno
!= ENOENT
)
1374 log_debug_errno(errno
, "Failed to check if /etc/os-release exists, ignoring: %m");
1376 os_release_src
= "/usr/lib/os-release";
1379 if (scope
== RUNTIME_SCOPE_SYSTEM
) {
1380 os_release_dst
= strdup("/run/systemd/propagate/.os-release-stage/os-release");
1381 if (!os_release_dst
)
1382 return log_oom_debug();
1384 if (asprintf(&os_release_dst
, "/run/user/" UID_FMT
"/systemd/propagate/.os-release-stage/os-release", geteuid()) < 0)
1385 return log_oom_debug();
1388 r
= mkdir_parents_label(os_release_dst
, 0755);
1390 return log_debug_errno(r
, "Failed to create parent directory of %s, ignoring: %m", os_release_dst
);
1392 r
= copy_file_atomic(os_release_src
, os_release_dst
, 0644, COPY_MAC_CREATE
|COPY_REPLACE
);
1394 return log_debug_errno(r
, "Failed to create %s, ignoring: %m", os_release_dst
);
1399 static int write_container_id(void) {
1401 int r
= 0; /* avoid false maybe-uninitialized warning */
1403 c
= getenv("container");
1408 r
= write_string_file("/run/systemd/container", c
, WRITE_STRING_FILE_CREATE
);
1410 return log_warning_errno(r
, "Failed to write /run/systemd/container, ignoring: %m");
1415 static int bump_unix_max_dgram_qlen(void) {
1416 _cleanup_free_
char *qlen
= NULL
;
1420 /* Let's bump the net.unix.max_dgram_qlen sysctl. The kernel default of 16 is simply too low. We set
1421 * the value really really early during boot, so that it is actually applied to all our sockets,
1422 * including the $NOTIFY_SOCKET one. */
1424 r
= read_one_line_file("/proc/sys/net/unix/max_dgram_qlen", &qlen
);
1426 return log_full_errno(r
== -ENOENT
? LOG_DEBUG
: LOG_WARNING
, r
,
1427 "Failed to read AF_UNIX datagram queue length, ignoring: %m");
1429 r
= safe_atolu(qlen
, &v
);
1431 return log_warning_errno(r
, "Failed to parse AF_UNIX datagram queue length '%s', ignoring: %m", qlen
);
1433 if (v
>= DEFAULT_UNIX_MAX_DGRAM_QLEN
)
1436 r
= sysctl_write("net/unix/max_dgram_qlen", STRINGIFY(DEFAULT_UNIX_MAX_DGRAM_QLEN
));
1438 return log_full_errno(IN_SET(r
, -EROFS
, -EPERM
, -EACCES
) ? LOG_DEBUG
: LOG_WARNING
, r
,
1439 "Failed to bump AF_UNIX datagram queue length, ignoring: %m");
1444 static int fixup_environment(void) {
1445 _cleanup_free_
char *term
= NULL
;
1449 /* Only fix up the environment when we are started as PID 1 */
1450 if (getpid_cached() != 1)
1453 /* We expect the environment to be set correctly if run inside a container. */
1454 if (detect_container() > 0)
1457 /* When started as PID1, the kernel uses /dev/console for our stdios and uses TERM=linux whatever the
1458 * backend device used by the console. We try to make a better guess here since some consoles might
1459 * not have support for color mode for example.
1461 * However if TERM was configured through the kernel command line then leave it alone. */
1462 r
= proc_cmdline_get_key("TERM", 0, &term
);
1467 r
= proc_cmdline_get_key("systemd.tty.term.console", 0, &term
);
1472 t
= term
?: default_term_for_tty("/dev/console");
1474 if (setenv("TERM", t
, 1) < 0)
1477 /* The kernels sets HOME=/ for init. Let's undo this. */
1478 if (path_equal_ptr(getenv("HOME"), "/"))
1479 assert_se(unsetenv("HOME") == 0);
1484 static void redirect_telinit(int argc
, char *argv
[]) {
1486 /* This is compatibility support for SysV, where calling init as a user is identical to telinit. */
1488 #if HAVE_SYSV_COMPAT
1489 if (getpid_cached() == 1)
1492 if (!invoked_as(argv
, "init"))
1495 execv(SYSTEMCTL_BINARY_PATH
, argv
);
1496 log_error_errno(errno
, "Failed to exec " SYSTEMCTL_BINARY_PATH
": %m");
1501 static int become_shutdown(int objective
, int retval
) {
1502 static const char* const table
[_MANAGER_OBJECTIVE_MAX
] = {
1503 [MANAGER_EXIT
] = "exit",
1504 [MANAGER_REBOOT
] = "reboot",
1505 [MANAGER_POWEROFF
] = "poweroff",
1506 [MANAGER_HALT
] = "halt",
1507 [MANAGER_KEXEC
] = "kexec",
1510 char log_level
[STRLEN("--log-level=") + DECIMAL_STR_MAX(int)],
1511 timeout
[STRLEN("--timeout=") + DECIMAL_STR_MAX(usec_t
) + STRLEN("us")],
1512 exit_code
[STRLEN("--exit-code=") + DECIMAL_STR_MAX(uint8_t)];
1514 _cleanup_strv_free_
char **env_block
= NULL
;
1515 usec_t watchdog_timer
= 0;
1518 assert(objective
>= 0 && objective
< _MANAGER_OBJECTIVE_MAX
);
1519 assert(table
[objective
]);
1521 xsprintf(log_level
, "--log-level=%d", log_get_max_level());
1522 xsprintf(timeout
, "--timeout=%" PRI_USEC
"us", arg_defaults
.timeout_stop_usec
);
1524 const char* command_line
[10] = {
1525 SYSTEMD_SHUTDOWN_BINARY_PATH
,
1529 /* Note that the last position is a terminator and must contain NULL. */
1533 assert(command_line
[pos
-1]);
1534 assert(!command_line
[pos
]);
1536 switch (log_get_target()) {
1538 case LOG_TARGET_KMSG
:
1539 case LOG_TARGET_JOURNAL_OR_KMSG
:
1540 case LOG_TARGET_SYSLOG_OR_KMSG
:
1541 command_line
[pos
++] = "--log-target=kmsg";
1544 case LOG_TARGET_NULL
:
1545 command_line
[pos
++] = "--log-target=null";
1548 case LOG_TARGET_CONSOLE
:
1550 command_line
[pos
++] = "--log-target=console";
1554 if (log_get_show_color())
1555 command_line
[pos
++] = "--log-color";
1557 if (log_get_show_location())
1558 command_line
[pos
++] = "--log-location";
1560 if (log_get_show_time())
1561 command_line
[pos
++] = "--log-time";
1563 xsprintf(exit_code
, "--exit-code=%d", retval
);
1564 command_line
[pos
++] = exit_code
;
1566 assert(pos
< ELEMENTSOF(command_line
));
1570 if (objective
== MANAGER_REBOOT
)
1571 watchdog_timer
= arg_reboot_watchdog
;
1572 else if (objective
== MANAGER_KEXEC
)
1573 watchdog_timer
= arg_kexec_watchdog
;
1575 /* If we reboot or kexec let's set the shutdown watchdog and tell the
1576 * shutdown binary to repeatedly ping it.
1577 * Disable the pretimeout watchdog, as we do not support it from the shutdown binary. */
1578 (void) watchdog_setup_pretimeout(0);
1579 (void) watchdog_setup_pretimeout_governor(NULL
);
1580 r
= watchdog_setup(watchdog_timer
);
1581 watchdog_close(r
< 0);
1583 /* The environment block: */
1585 env_block
= strv_copy(environ
);
1587 /* Tell the binary how often to ping, ignore failure */
1588 (void) strv_extendf(&env_block
, "WATCHDOG_USEC="USEC_FMT
, watchdog_timer
);
1590 if (arg_watchdog_device
)
1591 (void) strv_extendf(&env_block
, "WATCHDOG_DEVICE=%s", arg_watchdog_device
);
1593 /* Avoid the creation of new processes forked by the kernel; at this
1594 * point, we will not listen to the signals anyway */
1595 if (detect_container() <= 0)
1596 (void) cg_uninstall_release_agent(SYSTEMD_CGROUP_CONTROLLER
);
1598 execve(SYSTEMD_SHUTDOWN_BINARY_PATH
, (char **) command_line
, env_block
);
1602 static void initialize_clock(void) {
1605 /* This is called very early on, before we parse the kernel command line or otherwise figure out why
1606 * we are running, but only once. */
1608 if (clock_is_localtime(NULL
) > 0) {
1611 /* The very first call of settimeofday() also does a time warp in the kernel.
1613 * In the rtc-in-local time mode, we set the kernel's timezone, and rely on external tools to
1614 * take care of maintaining the RTC and do all adjustments. This matches the behavior of
1615 * Windows, which leaves the RTC alone if the registry tells that the RTC runs in UTC.
1617 r
= clock_set_timezone(&min
);
1619 log_error_errno(r
, "Failed to apply local time delta, ignoring: %m");
1621 log_info("RTC configured in localtime, applying delta of %i minutes to system time.", min
);
1623 } else if (!in_initrd())
1625 * Do a dummy very first call to seal the kernel's time warp magic.
1627 * Do not call this from inside the initrd. The initrd might not carry /etc/adjtime with
1628 * LOCAL, but the real system could be set up that way. In such case, we need to delay the
1629 * time-warp or the sealing until we reach the real system.
1631 * Do no set the kernel's timezone. The concept of local time cannot be supported reliably,
1632 * the time will jump or be incorrect at every daylight saving time change. All kernel local
1633 * time concepts will be treated as UTC that way.
1635 (void) clock_reset_timewarp();
1637 ClockChangeDirection change_dir
;
1638 r
= clock_apply_epoch(&change_dir
);
1639 if (r
> 0 && change_dir
== CLOCK_CHANGE_FORWARD
)
1640 log_info("System time before build time, advancing clock.");
1641 else if (r
> 0 && change_dir
== CLOCK_CHANGE_BACKWARD
)
1642 log_info("System time is further ahead than %s after build time, resetting clock to build time.",
1643 FORMAT_TIMESPAN(CLOCK_VALID_RANGE_USEC_MAX
, USEC_PER_DAY
));
1644 else if (r
< 0 && change_dir
== CLOCK_CHANGE_FORWARD
)
1645 log_error_errno(r
, "Current system time is before build time, but cannot correct: %m");
1646 else if (r
< 0 && change_dir
== CLOCK_CHANGE_BACKWARD
)
1647 log_error_errno(r
, "Current system time is further ahead %s after build time, but cannot correct: %m",
1648 FORMAT_TIMESPAN(CLOCK_VALID_RANGE_USEC_MAX
, USEC_PER_DAY
));
1651 static void apply_clock_update(void) {
1652 /* This is called later than initialize_clock(), i.e. after we parsed configuration files/kernel
1653 * command line and such. */
1655 if (arg_clock_usec
== 0)
1658 if (getpid_cached() != 1)
1661 if (clock_settime(CLOCK_REALTIME
, TIMESPEC_STORE(arg_clock_usec
)) < 0)
1662 log_error_errno(errno
, "Failed to set system clock to time specified on kernel command line: %m");
1664 log_info("Set system clock to %s, as specified on the kernel command line.",
1665 FORMAT_TIMESTAMP(arg_clock_usec
));
1668 static void cmdline_take_random_seed(void) {
1672 if (arg_random_seed_size
== 0)
1675 if (getpid_cached() != 1)
1678 assert(arg_random_seed
);
1679 suggested
= random_pool_size();
1681 if (arg_random_seed_size
< suggested
)
1682 log_warning("Random seed specified on kernel command line has size %zu, but %zu bytes required to fill entropy pool.",
1683 arg_random_seed_size
, suggested
);
1685 r
= random_write_entropy(-1, arg_random_seed
, arg_random_seed_size
, true);
1687 log_warning_errno(r
, "Failed to credit entropy specified on kernel command line, ignoring: %m");
1691 log_notice("Successfully credited entropy passed on kernel command line.\n"
1692 "Note that the seed provided this way is accessible to unprivileged programs. "
1693 "This functionality should not be used outside of testing environments.");
1696 static void initialize_coredump(bool skip_setup
) {
1697 if (getpid_cached() != 1)
1700 /* Don't limit the core dump size, so that coredump handlers such as systemd-coredump (which honour
1701 * the limit) will process core dumps for system services by default. */
1702 if (setrlimit(RLIMIT_CORE
, &RLIMIT_MAKE_CONST(RLIM_INFINITY
)) < 0)
1703 log_warning_errno(errno
, "Failed to set RLIMIT_CORE: %m");
1705 /* But at the same time, turn off the core_pattern logic by default, so that no coredumps are stored
1706 * until the systemd-coredump tool is enabled via sysctl. However it can be changed via the kernel
1707 * command line later so core dumps can still be generated during early startup and in initrd. */
1709 disable_coredumps();
1712 static void initialize_core_pattern(bool skip_setup
) {
1715 if (skip_setup
|| !arg_early_core_pattern
)
1718 if (getpid_cached() != 1)
1721 r
= write_string_file("/proc/sys/kernel/core_pattern", arg_early_core_pattern
, WRITE_STRING_FILE_DISABLE_BUFFER
);
1723 log_warning_errno(r
, "Failed to write '%s' to /proc/sys/kernel/core_pattern, ignoring: %m",
1724 arg_early_core_pattern
);
1727 static void apply_protect_system(bool skip_setup
) {
1730 if (skip_setup
|| getpid_cached() != 1 || arg_protect_system
== 0)
1733 if (arg_protect_system
< 0 && !in_initrd()) {
1734 log_debug("ProtectSystem=auto selected, but not running in an initrd, skipping.");
1738 r
= make_mount_point("/usr");
1740 log_warning_errno(r
, "Failed to make /usr/ a mount point, ignoring: %m");
1744 if (mount_nofollow_verbose(
1749 MS_BIND
|MS_REMOUNT
|MS_RDONLY
,
1750 /* options= */ NULL
) < 0)
1753 log_info("Successfully made /usr/ read-only.");
1756 static void update_cpu_affinity(bool skip_setup
) {
1757 _cleanup_free_
char *mask
= NULL
;
1759 if (skip_setup
|| !arg_cpu_affinity
.set
)
1762 assert(arg_cpu_affinity
.allocated
> 0);
1764 mask
= cpu_set_to_range_string(&arg_cpu_affinity
);
1765 log_debug("Setting CPU affinity to {%s}.", strnull(mask
));
1767 if (sched_setaffinity(0, arg_cpu_affinity
.allocated
, arg_cpu_affinity
.set
) < 0)
1768 log_warning_errno(errno
, "Failed to set CPU affinity, ignoring: %m");
1771 static void update_numa_policy(bool skip_setup
) {
1773 _cleanup_free_
char *nodes
= NULL
;
1774 const char * policy
= NULL
;
1776 if (skip_setup
|| !mpol_is_valid(numa_policy_get_type(&arg_numa_policy
)))
1779 if (DEBUG_LOGGING
) {
1780 policy
= mpol_to_string(numa_policy_get_type(&arg_numa_policy
));
1781 nodes
= cpu_set_to_range_string(&arg_numa_policy
.nodes
);
1782 log_debug("Setting NUMA policy to %s, with nodes {%s}.", strnull(policy
), strnull(nodes
));
1785 r
= apply_numa_policy(&arg_numa_policy
);
1786 if (r
== -EOPNOTSUPP
)
1787 log_debug_errno(r
, "NUMA support not available, ignoring.");
1789 log_warning_errno(r
, "Failed to set NUMA memory policy, ignoring: %m");
1792 static void filter_args(
1801 /* Copy some filtered arguments into the dst array from src. */
1802 for (int i
= 1; i
< argc
; i
++) {
1803 if (STR_IN_SET(src
[i
],
1809 if (startswith(src
[i
], "--deserialize="))
1811 if (streq(src
[i
], "--deserialize")) {
1812 i
++; /* Skip the argument too */
1816 /* Skip target unit designators. We already acted upon this information and have queued
1817 * appropriate jobs. We don't want to redo all this after reexecution. */
1818 if (startswith(src
[i
], "--unit="))
1820 if (streq(src
[i
], "--unit")) {
1821 i
++; /* Skip the argument too */
1825 /* Seems we have a good old option. Let's pass it over to the new instance. */
1826 dst
[(*dst_index
)++] = src
[i
];
1830 static void finish_remaining_processes(ManagerObjective objective
) {
1831 assert(objective
>= 0 && objective
< _MANAGER_OBJECTIVE_MAX
);
1833 /* Kill all remaining processes from the initrd, but don't wait for them, so that we can handle the
1834 * SIGCHLD for them after deserializing. */
1835 if (IN_SET(objective
, MANAGER_SWITCH_ROOT
, MANAGER_SOFT_REBOOT
))
1836 broadcast_signal(SIGTERM
, /* wait_for_exit= */ false, /* send_sighup= */ true, arg_defaults
.timeout_stop_usec
);
1838 /* On soft reboot really make sure nothing is left. Note that this will skip cgroups
1839 * of units that were configured with SurviveFinalKillSignal=yes. */
1840 if (objective
== MANAGER_SOFT_REBOOT
)
1841 broadcast_signal(SIGKILL
, /* wait_for_exit= */ false, /* send_sighup= */ false, arg_defaults
.timeout_stop_usec
);
1844 static int do_reexecute(
1845 ManagerObjective objective
,
1848 const struct rlimit
*saved_rlimit_nofile
,
1849 const struct rlimit
*saved_rlimit_memlock
,
1851 const char *switch_root_dir
,
1852 const char *switch_root_init
,
1853 const char **ret_error_message
) {
1855 size_t i
, args_size
;
1859 assert(IN_SET(objective
, MANAGER_REEXECUTE
, MANAGER_SWITCH_ROOT
, MANAGER_SOFT_REBOOT
));
1861 assert(saved_rlimit_nofile
);
1862 assert(saved_rlimit_memlock
);
1863 assert(ret_error_message
);
1865 if (switch_root_init
) {
1866 r
= chase(switch_root_init
, switch_root_dir
, CHASE_PREFIX_ROOT
, NULL
, NULL
);
1868 log_warning_errno(r
, "Failed to chase configured init %s/%s: %m",
1869 strempty(switch_root_dir
), switch_root_init
);
1871 r
= chase(SYSTEMD_BINARY_PATH
, switch_root_dir
, CHASE_PREFIX_ROOT
, NULL
, NULL
);
1873 log_debug_errno(r
, "Failed to chase our own binary %s/%s: %m",
1874 strempty(switch_root_dir
), SYSTEMD_BINARY_PATH
);
1878 r
= chase("/sbin/init", switch_root_dir
, CHASE_PREFIX_ROOT
, NULL
, NULL
);
1880 return log_error_errno(r
, "Failed to chase %s/sbin/init", strempty(switch_root_dir
));
1883 /* Close and disarm the watchdog, so that the new instance can reinitialize it, but doesn't get
1884 * rebooted while we do that */
1885 watchdog_close(true);
1887 /* Reset RLIMIT_NOFILE + RLIMIT_MEMLOCK back to the kernel defaults, so that the new systemd can pass
1888 * the kernel default to its child processes */
1889 if (saved_rlimit_nofile
->rlim_cur
!= 0)
1890 (void) setrlimit(RLIMIT_NOFILE
, saved_rlimit_nofile
);
1891 if (saved_rlimit_memlock
->rlim_cur
!= RLIM_INFINITY
)
1892 (void) setrlimit(RLIMIT_MEMLOCK
, saved_rlimit_memlock
);
1894 finish_remaining_processes(objective
);
1896 if (!switch_root_dir
&& objective
== MANAGER_SOFT_REBOOT
) {
1897 /* If no switch root dir is specified, then check if /run/nextroot/ qualifies and use that */
1898 r
= path_is_os_tree("/run/nextroot");
1899 if (r
< 0 && r
!= -ENOENT
)
1900 log_debug_errno(r
, "Failed to determine if /run/nextroot/ is a valid OS tree, ignoring: %m");
1902 switch_root_dir
= "/run/nextroot";
1905 if (switch_root_dir
) {
1906 r
= switch_root(/* new_root= */ switch_root_dir
,
1907 /* old_root_after= */ NULL
,
1908 /* flags= */ (objective
== MANAGER_SWITCH_ROOT
? SWITCH_ROOT_DESTROY_OLD_ROOT
: 0) |
1909 (objective
== MANAGER_SOFT_REBOOT
? 0 : SWITCH_ROOT_RECURSIVE_RUN
));
1911 log_error_errno(r
, "Failed to switch root, trying to continue: %m");
1914 args_size
= argc
+ 5;
1915 args
= newa(const char*, args_size
);
1917 if (!switch_root_init
) {
1918 char sfd
[STRLEN("--deserialize=") + DECIMAL_STR_MAX(int)];
1920 /* First try to spawn ourselves with the right path, and with full serialization. We do this
1921 * only if the user didn't specify an explicit init to spawn. */
1923 assert(arg_serialization
);
1926 xsprintf(sfd
, "--deserialize=%i", fileno(arg_serialization
));
1928 i
= 1; /* Leave args[0] empty for now. */
1930 /* Put our stuff first to make sure it always gets parsed in case
1931 * we get weird stuff from the kernel cmdline (like --) */
1932 if (IN_SET(objective
, MANAGER_SWITCH_ROOT
, MANAGER_SOFT_REBOOT
))
1933 args
[i
++] = "--switched-root";
1934 args
[i
++] = runtime_scope_cmdline_option_to_string(arg_runtime_scope
);
1937 filter_args(args
, &i
, argv
, argc
);
1941 assert(i
<= args_size
);
1944 * We want valgrind to print its memory usage summary before reexecution. Valgrind won't do
1945 * this is on its own on exec(), but it will do it on exit(). Hence, to ensure we get a
1946 * summary here, fork() off a child, let it exit() cleanly, so that it prints the summary,
1947 * and wait() for it in the parent, before proceeding into the exec().
1949 valgrind_summary_hack();
1951 args
[0] = SYSTEMD_BINARY_PATH
;
1952 (void) execv(args
[0], (char* const*) args
);
1954 if (objective
== MANAGER_REEXECUTE
) {
1955 *ret_error_message
= "Failed to execute our own binary";
1956 return log_error_errno(errno
, "Failed to execute our own binary %s: %m", args
[0]);
1959 log_debug_errno(errno
, "Failed to execute our own binary %s, trying fallback: %m", args
[0]);
1962 /* Try the fallback, if there is any, without any serialization. We pass the original argv[] and
1963 * envp[]. (Well, modulo the ordering changes due to getopt() in argv[], and some cleanups in envp[],
1964 * but let's hope that doesn't matter.) */
1966 arg_serialization
= safe_fclose(arg_serialization
);
1967 fds
= fdset_free(fds
);
1969 /* Reopen the console */
1970 (void) make_console_stdio();
1972 i
= 1; /* Leave args[0] empty for now. */
1973 for (int j
= 1; j
<= argc
; j
++)
1974 args
[i
++] = argv
[j
];
1975 assert(i
<= args_size
);
1977 /* Re-enable any blocked signals, especially important if we switch from initrd to init=... */
1978 (void) reset_all_signal_handlers();
1979 (void) reset_signal_mask();
1980 (void) rlimit_nofile_safe();
1982 if (switch_root_init
) {
1983 args
[0] = switch_root_init
;
1984 (void) execve(args
[0], (char* const*) args
, saved_env
);
1985 log_warning_errno(errno
, "Failed to execute configured init %s, trying fallback: %m", args
[0]);
1988 args
[0] = "/sbin/init";
1989 (void) execv(args
[0], (char* const*) args
);
1992 manager_status_printf(NULL
, STATUS_TYPE_EMERGENCY
,
1993 ANSI_HIGHLIGHT_RED
" !! " ANSI_NORMAL
,
1994 "Failed to execute /sbin/init");
1996 *ret_error_message
= "Failed to execute fallback shell";
1998 log_warning("No /sbin/init, trying fallback");
2000 args
[0] = "/bin/sh";
2002 (void) execve(args
[0], (char* const*) args
, saved_env
);
2003 return log_error_errno(errno
, "Failed to execute /bin/sh, giving up: %m");
2005 return log_error_errno(r
, "Failed to execute /sbin/init, giving up: %m");
2008 static int invoke_main_loop(
2010 const struct rlimit
*saved_rlimit_nofile
,
2011 const struct rlimit
*saved_rlimit_memlock
,
2012 int *ret_retval
, /* Return parameters relevant for shutting down */
2013 FDSet
**ret_fds
, /* Return parameters for reexecuting */
2014 char **ret_switch_root_dir
, /* … */
2015 char **ret_switch_root_init
, /* … */
2016 const char **ret_error_message
) {
2021 assert(saved_rlimit_nofile
);
2022 assert(saved_rlimit_memlock
);
2025 assert(ret_switch_root_dir
);
2026 assert(ret_switch_root_init
);
2027 assert(ret_error_message
);
2030 int objective
= manager_loop(m
);
2031 if (objective
< 0) {
2032 *ret_error_message
= "Failed to run main loop";
2033 return log_struct_errno(LOG_EMERG
, objective
,
2034 LOG_MESSAGE("Failed to run main loop: %m"),
2035 "MESSAGE_ID=" SD_MESSAGE_CORE_MAINLOOP_FAILED_STR
);
2038 switch (objective
) {
2040 case MANAGER_RELOAD
: {
2041 LogTarget saved_log_target
;
2042 int saved_log_level
;
2044 manager_send_reloading(m
);
2046 log_info("Reloading...");
2048 /* First, save any overridden log level/target, then parse the configuration file,
2049 * which might change the log level to new settings. */
2051 saved_log_level
= m
->log_level_overridden
? log_get_max_level() : -1;
2052 saved_log_target
= m
->log_target_overridden
? log_get_target() : _LOG_TARGET_INVALID
;
2054 (void) parse_configuration(saved_rlimit_nofile
, saved_rlimit_memlock
);
2056 set_manager_defaults(m
);
2057 set_manager_settings(m
);
2059 update_cpu_affinity(false);
2060 update_numa_policy(false);
2062 if (saved_log_level
>= 0)
2063 manager_override_log_level(m
, saved_log_level
);
2064 if (saved_log_target
>= 0)
2065 manager_override_log_target(m
, saved_log_target
);
2067 if (manager_reload(m
) < 0)
2068 /* Reloading failed before the point of no return.
2069 * Let's continue running as if nothing happened. */
2070 m
->objective
= MANAGER_OK
;
2072 log_info("Reloading finished in " USEC_FMT
" ms.",
2073 usec_sub_unsigned(now(CLOCK_MONOTONIC
), m
->timestamps
[MANAGER_TIMESTAMP_UNITS_LOAD
].monotonic
) / USEC_PER_MSEC
);
2078 case MANAGER_REEXECUTE
:
2080 manager_send_reloading(m
); /* From the perspective of the manager calling us this is
2081 * pretty much the same as a reload */
2083 r
= prepare_reexecute(m
, &arg_serialization
, ret_fds
, false);
2085 *ret_error_message
= "Failed to prepare for reexecution";
2089 log_notice("Reexecuting.");
2091 *ret_retval
= EXIT_SUCCESS
;
2092 *ret_switch_root_dir
= *ret_switch_root_init
= NULL
;
2096 case MANAGER_SWITCH_ROOT
:
2098 manager_send_reloading(m
); /* From the perspective of the manager calling us this is
2099 * pretty much the same as a reload */
2101 manager_set_switching_root(m
, true);
2103 if (!m
->switch_root_init
) {
2104 r
= prepare_reexecute(m
, &arg_serialization
, ret_fds
, true);
2106 *ret_error_message
= "Failed to prepare for reexecution";
2112 log_notice("Switching root.");
2114 *ret_retval
= EXIT_SUCCESS
;
2116 /* Steal the switch root parameters */
2117 *ret_switch_root_dir
= TAKE_PTR(m
->switch_root
);
2118 *ret_switch_root_init
= TAKE_PTR(m
->switch_root_init
);
2122 case MANAGER_SOFT_REBOOT
:
2123 manager_send_reloading(m
);
2124 manager_set_switching_root(m
, true);
2126 r
= prepare_reexecute(m
, &arg_serialization
, ret_fds
, /* switching_root= */ true);
2128 *ret_error_message
= "Failed to prepare for reexecution";
2132 log_notice("Soft-rebooting.");
2134 *ret_retval
= EXIT_SUCCESS
;
2135 *ret_switch_root_dir
= TAKE_PTR(m
->switch_root
);
2136 *ret_switch_root_init
= NULL
;
2141 if (MANAGER_IS_USER(m
)) {
2144 *ret_retval
= m
->return_value
;
2146 *ret_switch_root_dir
= *ret_switch_root_init
= NULL
;
2152 case MANAGER_REBOOT
:
2153 case MANAGER_POWEROFF
:
2155 case MANAGER_KEXEC
: {
2156 log_notice("Shutting down.");
2158 *ret_retval
= m
->return_value
;
2160 *ret_switch_root_dir
= *ret_switch_root_init
= NULL
;
2166 assert_not_reached();
2171 static void log_execution_mode(bool *ret_first_boot
) {
2172 bool first_boot
= false;
2175 assert(ret_first_boot
);
2177 switch (arg_runtime_scope
) {
2179 case RUNTIME_SCOPE_SYSTEM
: {
2183 log_info("systemd " GIT_VERSION
" running in %ssystem mode (%s)",
2184 arg_action
== ACTION_TEST
? "test " : "",
2187 v
= detect_virtualization();
2189 log_info("Detected virtualization %s.", virtualization_to_string(v
));
2191 v
= detect_confidential_virtualization();
2193 log_info("Detected confidential virtualization %s.", confidential_virtualization_to_string(v
));
2195 log_info("Detected architecture %s.", architecture_to_string(uname_architecture()));
2198 log_info("Running in initrd.");
2200 _cleanup_free_
char *id_text
= NULL
;
2202 /* Let's check whether we are in first boot. First, check if an override was
2203 * specified on the kernel command line. If yes, we honour that. */
2205 r
= proc_cmdline_get_bool("systemd.condition-first-boot", /* flags = */ 0, &first_boot
);
2207 log_debug_errno(r
, "Failed to parse systemd.condition-first-boot= kernel command line argument, ignoring: %m");
2210 log_full(first_boot
? LOG_INFO
: LOG_DEBUG
,
2211 "Kernel command line argument says we are %s first boot.",
2212 first_boot
? "in" : "not in");
2214 /* Second, perform autodetection. We use /etc/machine-id as flag file for
2215 * this: If it is missing or contains the value "uninitialized", this is the
2216 * first boot. In other cases, it is not. This allows container managers and
2217 * installers to provision a couple of files in /etc but still permit the
2218 * first-boot initialization to occur. If the container manager wants to
2219 * provision the machine ID it should pass $container_uuid to PID 1. */
2221 r
= read_one_line_file("/etc/machine-id", &id_text
);
2222 if (r
< 0 || streq(id_text
, "uninitialized")) {
2223 if (r
< 0 && r
!= -ENOENT
)
2224 log_warning_errno(r
, "Unexpected error while reading /etc/machine-id, assuming first boot: %m");
2227 log_info("Detected first boot.");
2229 log_debug("Detected initialized system, this is not the first boot.");
2233 assert_se(uname(&uts
) >= 0);
2235 if (strverscmp_improved(uts
.release
, KERNEL_BASELINE_VERSION
) < 0)
2236 log_warning("Warning! Reported kernel version %s is older than systemd's required baseline kernel version %s. "
2237 "Your mileage may vary.", uts
.release
, KERNEL_BASELINE_VERSION
);
2239 log_debug("Kernel version %s, our baseline is %s", uts
.release
, KERNEL_BASELINE_VERSION
);
2244 case RUNTIME_SCOPE_USER
:
2245 if (DEBUG_LOGGING
) {
2246 _cleanup_free_
char *t
= NULL
;
2248 t
= uid_to_name(getuid());
2249 log_debug("systemd " GIT_VERSION
" running in %suser mode for user " UID_FMT
"/%s. (%s)",
2250 arg_action
== ACTION_TEST
? " test" : "",
2251 getuid(), strna(t
), systemd_features
);
2257 assert_not_reached();
2260 *ret_first_boot
= first_boot
;
2263 static int initialize_runtime(
2266 struct rlimit
*saved_rlimit_nofile
,
2267 struct rlimit
*saved_rlimit_memlock
,
2268 const char **ret_error_message
) {
2271 assert(ret_error_message
);
2273 /* Sets up various runtime parameters. Many of these initializations are conditionalized:
2275 * - Some only apply to --system instances
2276 * - Some only apply to --user instances
2277 * - Some only apply when we first start up, but not when we reexecute
2280 if (arg_action
!= ACTION_RUN
)
2283 update_cpu_affinity(skip_setup
);
2284 update_numa_policy(skip_setup
);
2286 switch (arg_runtime_scope
) {
2288 case RUNTIME_SCOPE_SYSTEM
:
2289 /* Make sure we leave a core dump without panicking the kernel. */
2290 install_crash_handler();
2293 r
= mount_cgroup_controllers();
2295 *ret_error_message
= "Failed to mount cgroup hierarchies";
2299 /* Pull credentials from various sources into a common credential directory (we do
2300 * this here, before setting up the machine ID, so that we can use credential info
2301 * for setting up the machine ID) */
2302 (void) import_credentials();
2304 (void) os_release_status();
2305 (void) hostname_setup(true);
2306 /* Force transient machine-id on first boot. */
2307 machine_id_setup(/* root= */ NULL
, /* force_transient= */ first_boot
, arg_machine_id
, /* ret_machine_id */ NULL
);
2308 (void) loopback_setup();
2309 bump_unix_max_dgram_qlen();
2310 bump_file_max_and_nr_open();
2312 write_container_id();
2314 /* Copy os-release to the propagate directory, so that we update it for services running
2315 * under RootDirectory=/RootImage= when we do a soft reboot. */
2316 r
= setup_os_release(RUNTIME_SCOPE_SYSTEM
);
2318 log_warning_errno(r
, "Failed to copy os-release for propagation, ignoring: %m");
2321 r
= watchdog_set_device(arg_watchdog_device
);
2323 log_warning_errno(r
, "Failed to set watchdog device to %s, ignoring: %m", arg_watchdog_device
);
2327 case RUNTIME_SCOPE_USER
: {
2328 _cleanup_free_
char *p
= NULL
;
2330 /* Create the runtime directory and place the inaccessible device nodes there, if we run in
2331 * user mode. In system mode mount_setup() already did that. */
2333 r
= xdg_user_runtime_dir(&p
, "/systemd");
2335 *ret_error_message
= "$XDG_RUNTIME_DIR is not set";
2336 return log_struct_errno(LOG_EMERG
, r
,
2337 LOG_MESSAGE("Failed to determine $XDG_RUNTIME_DIR path: %m"),
2338 "MESSAGE_ID=" SD_MESSAGE_CORE_NO_XDGDIR_PATH_STR
);
2341 (void) mkdir_p_label(p
, 0755);
2342 (void) make_inaccessible_nodes(p
, UID_INVALID
, GID_INVALID
);
2343 r
= setup_os_release(RUNTIME_SCOPE_USER
);
2345 log_warning_errno(r
, "Failed to copy os-release for propagation, ignoring: %m");
2350 assert_not_reached();
2353 if (arg_timer_slack_nsec
!= NSEC_INFINITY
)
2354 if (prctl(PR_SET_TIMERSLACK
, arg_timer_slack_nsec
) < 0)
2355 log_warning_errno(errno
, "Failed to adjust timer slack, ignoring: %m");
2357 if (arg_runtime_scope
== RUNTIME_SCOPE_SYSTEM
) {
2359 if (!cap_test_all(arg_capability_bounding_set
)) {
2360 r
= capability_bounding_set_drop_usermode(arg_capability_bounding_set
);
2362 *ret_error_message
= "Failed to drop capability bounding set of usermode helpers";
2363 return log_struct_errno(LOG_EMERG
, r
,
2364 LOG_MESSAGE("Failed to drop capability bounding set of usermode helpers: %m"),
2365 "MESSAGE_ID=" SD_MESSAGE_CORE_CAPABILITY_BOUNDING_USER_STR
);
2368 r
= capability_bounding_set_drop(arg_capability_bounding_set
, true);
2370 *ret_error_message
= "Failed to drop capability bounding set";
2371 return log_struct_errno(LOG_EMERG
, r
,
2372 LOG_MESSAGE("Failed to drop capability bounding set: %m"),
2373 "MESSAGE_ID=" SD_MESSAGE_CORE_CAPABILITY_BOUNDING_STR
);
2377 if (arg_no_new_privs
) {
2378 if (prctl(PR_SET_NO_NEW_PRIVS
, 1, 0, 0, 0) < 0) {
2379 *ret_error_message
= "Failed to disable new privileges";
2380 return log_struct_errno(LOG_EMERG
, errno
,
2381 LOG_MESSAGE("Failed to disable new privileges: %m"),
2382 "MESSAGE_ID=" SD_MESSAGE_CORE_DISABLE_PRIVILEGES_STR
);
2387 if (arg_syscall_archs
) {
2388 r
= enforce_syscall_archs(arg_syscall_archs
);
2390 *ret_error_message
= "Failed to set syscall architectures";
2395 r
= make_reaper_process(true);
2397 log_warning_errno(r
, "Failed to make us a subreaper, ignoring: %m");
2399 /* Bump up RLIMIT_NOFILE for systemd itself */
2400 (void) bump_rlimit_nofile(saved_rlimit_nofile
);
2401 (void) bump_rlimit_memlock(saved_rlimit_memlock
);
2406 static int do_queue_default_job(
2408 const char **ret_error_message
) {
2410 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
2416 if (arg_default_unit
)
2417 unit
= arg_default_unit
;
2418 else if (in_initrd())
2419 unit
= SPECIAL_INITRD_TARGET
;
2421 unit
= SPECIAL_DEFAULT_TARGET
;
2423 log_debug("Activating default unit: %s", unit
);
2425 r
= manager_load_startable_unit_or_warn(m
, unit
, NULL
, &target
);
2426 if (r
< 0 && in_initrd() && !arg_default_unit
) {
2427 /* Fall back to default.target, which we used to always use by default. Only do this if no
2428 * explicit configuration was given. */
2430 log_info("Falling back to " SPECIAL_DEFAULT_TARGET
".");
2432 r
= manager_load_startable_unit_or_warn(m
, SPECIAL_DEFAULT_TARGET
, NULL
, &target
);
2435 log_info("Falling back to " SPECIAL_RESCUE_TARGET
".");
2437 r
= manager_load_startable_unit_or_warn(m
, SPECIAL_RESCUE_TARGET
, NULL
, &target
);
2439 *ret_error_message
= r
== -ERFKILL
? SPECIAL_RESCUE_TARGET
" masked"
2440 : "Failed to load " SPECIAL_RESCUE_TARGET
;
2445 assert(target
->load_state
== UNIT_LOADED
);
2447 r
= manager_add_job(m
, JOB_START
, target
, JOB_ISOLATE
, NULL
, &error
, &job
);
2449 log_debug_errno(r
, "Default target could not be isolated, starting instead: %s", bus_error_message(&error
, r
));
2451 sd_bus_error_free(&error
);
2453 r
= manager_add_job(m
, JOB_START
, target
, JOB_REPLACE
, NULL
, &error
, &job
);
2455 *ret_error_message
= "Failed to start default target";
2456 return log_struct_errno(LOG_EMERG
, r
,
2457 LOG_MESSAGE("Failed to start default target: %s", bus_error_message(&error
, r
)),
2458 "MESSAGE_ID=" SD_MESSAGE_CORE_START_TARGET_FAILED_STR
);
2462 *ret_error_message
= "Failed to isolate default target";
2463 return log_struct_errno(LOG_EMERG
, r
,
2464 LOG_MESSAGE("Failed to isolate default target: %s", bus_error_message(&error
, r
)),
2465 "MESSAGE_ID=" SD_MESSAGE_CORE_ISOLATE_TARGET_FAILED_STR
);
2467 log_info("Queued %s job for default target %s.",
2468 job_type_to_string(job
->type
),
2469 unit_status_string(job
->unit
, NULL
));
2471 m
->default_unit_job_id
= job
->id
;
2476 static void save_rlimits(struct rlimit
*saved_rlimit_nofile
,
2477 struct rlimit
*saved_rlimit_memlock
) {
2479 assert(saved_rlimit_nofile
);
2480 assert(saved_rlimit_memlock
);
2482 if (getrlimit(RLIMIT_NOFILE
, saved_rlimit_nofile
) < 0)
2483 log_warning_errno(errno
, "Reading RLIMIT_NOFILE failed, ignoring: %m");
2485 if (getrlimit(RLIMIT_MEMLOCK
, saved_rlimit_memlock
) < 0)
2486 log_warning_errno(errno
, "Reading RLIMIT_MEMLOCK failed, ignoring: %m");
2489 static void fallback_rlimit_nofile(const struct rlimit
*saved_rlimit_nofile
) {
2492 if (arg_defaults
.rlimit
[RLIMIT_NOFILE
])
2495 /* Make sure forked processes get limits based on the original kernel setting */
2497 rl
= newdup(struct rlimit
, saved_rlimit_nofile
, 1);
2503 /* Bump the hard limit for system services to a substantially higher value. The default
2504 * hard limit current kernels set is pretty low (4K), mostly for historical
2505 * reasons. According to kernel developers, the fd handling in recent kernels has been
2506 * optimized substantially enough, so that we can bump the limit now, without paying too
2507 * high a price in memory or performance. Note however that we only bump the hard limit,
2508 * not the soft limit. That's because select() works the way it works, and chokes on fds
2509 * >= 1024. If we'd bump the soft limit globally, it might accidentally happen to
2510 * unexpecting programs that they get fds higher than what they can process using
2511 * select(). By only bumping the hard limit but leaving the low limit as it is we avoid
2512 * this pitfall: programs that are written by folks aware of the select() problem in mind
2513 * (and thus use poll()/epoll instead of select(), the way everybody should) can
2514 * explicitly opt into high fds by bumping their soft limit beyond 1024, to the hard limit
2516 if (arg_runtime_scope
== RUNTIME_SCOPE_SYSTEM
) {
2519 /* Get the underlying absolute limit the kernel enforces */
2520 nr
= read_nr_open();
2522 rl
->rlim_max
= MIN((rlim_t
) nr
, MAX(rl
->rlim_max
, (rlim_t
) HIGH_RLIMIT_NOFILE
));
2525 /* If for some reason we were invoked with a soft limit above 1024 (which should never
2526 * happen!, but who knows what we get passed in from pam_limit when invoked as --user
2527 * instance), then lower what we pass on to not confuse our children */
2528 rl
->rlim_cur
= MIN(rl
->rlim_cur
, (rlim_t
) FD_SETSIZE
);
2530 arg_defaults
.rlimit
[RLIMIT_NOFILE
] = rl
;
2533 static void fallback_rlimit_memlock(const struct rlimit
*saved_rlimit_memlock
) {
2536 /* Pass the original value down to invoked processes */
2538 if (arg_defaults
.rlimit
[RLIMIT_MEMLOCK
])
2541 rl
= newdup(struct rlimit
, saved_rlimit_memlock
, 1);
2547 if (arg_runtime_scope
== RUNTIME_SCOPE_SYSTEM
) {
2548 /* Raise the default limit to 8M also on old kernels and in containers (8M is the kernel
2549 * default for this since kernel 5.16) */
2550 rl
->rlim_max
= MAX(rl
->rlim_max
, (rlim_t
) DEFAULT_RLIMIT_MEMLOCK
);
2551 rl
->rlim_cur
= MAX(rl
->rlim_cur
, (rlim_t
) DEFAULT_RLIMIT_MEMLOCK
);
2554 arg_defaults
.rlimit
[RLIMIT_MEMLOCK
] = rl
;
2557 static void setenv_manager_environment(void) {
2560 STRV_FOREACH(p
, arg_manager_environment
) {
2561 log_debug("Setting '%s' in our own environment.", *p
);
2563 r
= putenv_dup(*p
, true);
2565 log_warning_errno(errno
, "Failed to setenv \"%s\", ignoring: %m", *p
);
2569 static void reset_arguments(void) {
2570 /* Frees/resets arg_* variables, with a few exceptions commented below. */
2572 arg_default_unit
= mfree(arg_default_unit
);
2574 /* arg_runtime_scope — ignore */
2576 arg_dump_core
= true;
2577 arg_crash_chvt
= -1;
2578 arg_crash_shell
= false;
2579 arg_crash_reboot
= false;
2580 arg_confirm_spawn
= mfree(arg_confirm_spawn
);
2581 arg_show_status
= _SHOW_STATUS_INVALID
;
2582 arg_status_unit_format
= STATUS_UNIT_FORMAT_DEFAULT
;
2583 arg_switched_root
= false;
2584 arg_pager_flags
= 0;
2585 arg_service_watchdogs
= true;
2587 unit_defaults_done(&arg_defaults
);
2588 unit_defaults_init(&arg_defaults
, arg_runtime_scope
);
2590 arg_runtime_watchdog
= 0;
2591 arg_reboot_watchdog
= 10 * USEC_PER_MINUTE
;
2592 arg_kexec_watchdog
= 0;
2593 arg_pretimeout_watchdog
= 0;
2594 arg_early_core_pattern
= mfree(arg_early_core_pattern
);
2595 arg_watchdog_device
= mfree(arg_watchdog_device
);
2596 arg_watchdog_pretimeout_governor
= mfree(arg_watchdog_pretimeout_governor
);
2598 arg_default_environment
= strv_free(arg_default_environment
);
2599 arg_manager_environment
= strv_free(arg_manager_environment
);
2601 arg_capability_bounding_set
= CAP_MASK_UNSET
;
2602 arg_no_new_privs
= false;
2603 arg_protect_system
= -1;
2604 arg_timer_slack_nsec
= NSEC_INFINITY
;
2606 arg_syscall_archs
= set_free(arg_syscall_archs
);
2608 /* arg_serialization — ignore */
2610 arg_machine_id
= (sd_id128_t
) {};
2611 arg_cad_burst_action
= EMERGENCY_ACTION_REBOOT_FORCE
;
2613 cpu_set_reset(&arg_cpu_affinity
);
2614 numa_policy_reset(&arg_numa_policy
);
2616 arg_random_seed
= mfree(arg_random_seed
);
2617 arg_random_seed_size
= 0;
2620 arg_reload_limit_interval_sec
= 0;
2621 arg_reload_limit_burst
= 0;
2624 static void determine_default_oom_score_adjust(void) {
2627 /* Run our services at slightly higher OOM score than ourselves. But let's be conservative here, and
2628 * do this only if we don't run as root (i.e. only if we are run in user mode, for an unprivileged
2631 if (arg_defaults
.oom_score_adjust_set
)
2637 r
= get_oom_score_adjust(&a
);
2639 return (void) log_warning_errno(r
, "Failed to determine current OOM score adjustment value, ignoring: %m");
2641 assert_cc(100 <= OOM_SCORE_ADJ_MAX
);
2642 b
= a
>= OOM_SCORE_ADJ_MAX
- 100 ? OOM_SCORE_ADJ_MAX
: a
+ 100;
2647 arg_defaults
.oom_score_adjust
= b
;
2648 arg_defaults
.oom_score_adjust_set
= true;
2651 static int parse_configuration(const struct rlimit
*saved_rlimit_nofile
,
2652 const struct rlimit
*saved_rlimit_memlock
) {
2655 assert(saved_rlimit_nofile
);
2656 assert(saved_rlimit_memlock
);
2658 /* Assign configuration defaults */
2661 r
= parse_config_file();
2663 log_warning_errno(r
, "Failed to parse config file, ignoring: %m");
2665 if (arg_runtime_scope
== RUNTIME_SCOPE_SYSTEM
) {
2666 r
= proc_cmdline_parse(parse_proc_cmdline_item
, NULL
, 0);
2668 log_warning_errno(r
, "Failed to parse kernel command line, ignoring: %m");
2671 /* Initialize some default rlimits for services if they haven't been configured */
2672 fallback_rlimit_nofile(saved_rlimit_nofile
);
2673 fallback_rlimit_memlock(saved_rlimit_memlock
);
2675 /* Note that this also parses bits from the kernel command line, including "debug". */
2676 log_parse_environment();
2678 /* Initialize the show status setting if it hasn't been set explicitly yet */
2679 if (arg_show_status
== _SHOW_STATUS_INVALID
)
2680 arg_show_status
= SHOW_STATUS_YES
;
2682 /* Slightly raise the OOM score for our services if we are running for unprivileged users. */
2683 determine_default_oom_score_adjust();
2685 /* Push variables into the manager environment block */
2686 setenv_manager_environment();
2688 /* Parse log environment variables again to take into account any new environment variables. */
2689 log_parse_environment();
2694 static int safety_checks(void) {
2696 if (getpid_cached() == 1 &&
2697 arg_action
!= ACTION_RUN
)
2698 return log_error_errno(SYNTHETIC_ERRNO(EPERM
),
2699 "Unsupported execution mode while PID 1.");
2701 if (getpid_cached() == 1 &&
2702 arg_runtime_scope
== RUNTIME_SCOPE_USER
)
2703 return log_error_errno(SYNTHETIC_ERRNO(EPERM
),
2704 "Can't run --user mode as PID 1.");
2706 if (arg_action
== ACTION_RUN
&&
2707 arg_runtime_scope
== RUNTIME_SCOPE_SYSTEM
&&
2708 getpid_cached() != 1)
2709 return log_error_errno(SYNTHETIC_ERRNO(EPERM
),
2710 "Can't run system mode unless PID 1.");
2712 if (arg_action
== ACTION_TEST
&&
2714 return log_error_errno(SYNTHETIC_ERRNO(EPERM
),
2715 "Don't run test mode as root.");
2717 switch (arg_runtime_scope
) {
2719 case RUNTIME_SCOPE_USER
:
2721 if (arg_action
== ACTION_RUN
&&
2723 return log_error_errno(SYNTHETIC_ERRNO(EOPNOTSUPP
),
2724 "Trying to run as user instance, but the system has not been booted with systemd.");
2726 if (arg_action
== ACTION_RUN
&&
2727 !getenv("XDG_RUNTIME_DIR"))
2728 return log_error_errno(SYNTHETIC_ERRNO(EUNATCH
),
2729 "Trying to run as user instance, but $XDG_RUNTIME_DIR is not set.");
2733 case RUNTIME_SCOPE_SYSTEM
:
2734 if (arg_action
== ACTION_RUN
&&
2735 running_in_chroot() > 0)
2736 return log_error_errno(SYNTHETIC_ERRNO(EOPNOTSUPP
),
2737 "Cannot be run in a chroot() environment.");
2741 assert_not_reached();
2747 static int initialize_security(
2748 bool *loaded_policy
,
2749 dual_timestamp
*security_start_timestamp
,
2750 dual_timestamp
*security_finish_timestamp
,
2751 const char **ret_error_message
) {
2755 assert(loaded_policy
);
2756 assert(security_start_timestamp
);
2757 assert(security_finish_timestamp
);
2758 assert(ret_error_message
);
2760 dual_timestamp_now(security_start_timestamp
);
2762 r
= mac_selinux_setup(loaded_policy
);
2764 *ret_error_message
= "Failed to load SELinux policy";
2768 r
= mac_smack_setup(loaded_policy
);
2770 *ret_error_message
= "Failed to load SMACK policy";
2774 r
= mac_apparmor_setup();
2776 *ret_error_message
= "Failed to load AppArmor policy";
2782 *ret_error_message
= "Failed to load IMA policy";
2786 dual_timestamp_now(security_finish_timestamp
);
2790 static int collect_fds(FDSet
**ret_fds
, const char **ret_error_message
) {
2794 assert(ret_error_message
);
2796 /* Pick up all fds passed to us. We apply a filter here: we only take the fds that have O_CLOEXEC
2797 * off. All fds passed via execve() to us must have O_CLOEXEC off, and our own code and dependencies
2798 * should be clean enough to set O_CLOEXEC universally. Thus checking the bit should be a safe
2799 * mechanism to distinguish passed in fds from our own.
2801 * Why bother? Some subsystems we initialize early, specifically selinux might keep fds open in our
2802 * process behind our back. We should not take possession of that (and then accidentally close
2803 * it). SELinux thankfully sets O_CLOEXEC on its fds, so this test should work. */
2804 r
= fdset_new_fill(/* filter_cloexec= */ 0, ret_fds
);
2806 *ret_error_message
= "Failed to allocate fd set";
2807 return log_struct_errno(LOG_EMERG
, r
,
2808 LOG_MESSAGE("Failed to allocate fd set: %m"),
2809 "MESSAGE_ID=" SD_MESSAGE_CORE_FD_SET_FAILED_STR
);
2812 (void) fdset_cloexec(*ret_fds
, true);
2814 /* The serialization fd should have O_CLOEXEC turned on already, let's verify that we didn't pick it up here */
2815 assert_se(!arg_serialization
|| !fdset_contains(*ret_fds
, fileno(arg_serialization
)));
2820 static void setup_console_terminal(bool skip_setup
) {
2822 if (arg_runtime_scope
!= RUNTIME_SCOPE_SYSTEM
)
2825 /* Become a session leader if we aren't one yet. */
2828 /* If we are init, we connect stdin/stdout/stderr to /dev/null and make sure we don't have a
2829 * controlling tty. */
2830 (void) release_terminal();
2832 /* Reset the console, but only if this is really init and we are freshly booted */
2833 if (getpid_cached() == 1 && !skip_setup
)
2834 (void) console_setup();
2837 static bool early_skip_setup_check(int argc
, char *argv
[]) {
2838 bool found_deserialize
= false;
2840 /* Determine if this is a reexecution or normal bootup. We do the full command line parsing much
2841 * later, so let's just have a quick peek here. Note that if we have switched root, do all the
2842 * special setup things anyway, even if in that case we also do deserialization. */
2844 for (int i
= 1; i
< argc
; i
++)
2845 if (streq(argv
[i
], "--switched-root"))
2846 return false; /* If we switched root, don't skip the setup. */
2847 else if (startswith(argv
[i
], "--deserialize=") || streq(argv
[i
], "--deserialize"))
2848 found_deserialize
= true;
2850 return found_deserialize
; /* When we are deserializing, then we are reexecuting, hence avoid the extensive setup */
2853 static int save_env(void) {
2856 l
= strv_copy(environ
);
2860 strv_free_and_replace(saved_env
, l
);
2864 int main(int argc
, char *argv
[]) {
2866 initrd_timestamp
= DUAL_TIMESTAMP_NULL
,
2867 userspace_timestamp
= DUAL_TIMESTAMP_NULL
,
2868 kernel_timestamp
= DUAL_TIMESTAMP_NULL
,
2869 security_start_timestamp
= DUAL_TIMESTAMP_NULL
,
2870 security_finish_timestamp
= DUAL_TIMESTAMP_NULL
;
2871 struct rlimit saved_rlimit_nofile
= RLIMIT_MAKE_CONST(0),
2872 saved_rlimit_memlock
= RLIMIT_MAKE_CONST(RLIM_INFINITY
); /* The original rlimits we passed
2873 * in. Note we use different values
2874 * for the two that indicate whether
2875 * these fields are initialized! */
2876 bool skip_setup
, loaded_policy
= false, queue_default_job
= false, first_boot
= false;
2877 char *switch_root_dir
= NULL
, *switch_root_init
= NULL
;
2878 usec_t before_startup
, after_startup
;
2879 static char systemd
[] = "systemd";
2880 const char *error_message
= NULL
;
2881 int r
, retval
= EXIT_FAILURE
;
2885 assert_se(argc
> 0 && !isempty(argv
[0]));
2887 /* SysV compatibility: redirect init → telinit */
2888 redirect_telinit(argc
, argv
);
2890 /* Take timestamps early on */
2891 dual_timestamp_from_monotonic(&kernel_timestamp
, 0);
2892 dual_timestamp_now(&userspace_timestamp
);
2894 /* Figure out whether we need to do initialize the system, or if we already did that because we are
2896 skip_setup
= early_skip_setup_check(argc
, argv
);
2898 /* If we get started via the /sbin/init symlink then we are called 'init'. After a subsequent
2899 * reexecution we are then called 'systemd'. That is confusing, hence let's call us systemd
2901 program_invocation_short_name
= systemd
;
2902 (void) prctl(PR_SET_NAME
, systemd
);
2904 /* Save the original command line */
2905 save_argc_argv(argc
, argv
);
2907 /* Save the original environment as we might need to restore it if we're requested to execute another
2908 * system manager later. */
2911 error_message
= "Failed to copy environment block";
2915 /* Make sure that if the user says "syslog" we actually log to the journal. */
2916 log_set_upgrade_syslog_to_journal(true);
2918 if (getpid_cached() == 1) {
2919 /* When we run as PID 1 force system mode */
2920 arg_runtime_scope
= RUNTIME_SCOPE_SYSTEM
;
2922 /* Disable the umask logic */
2925 /* Make sure that at least initially we do not ever log to journald/syslogd, because it might
2926 * not be activated yet (even though the log socket for it exists). */
2927 log_set_prohibit_ipc(true);
2929 /* Always reopen /dev/console when running as PID 1 or one of its pre-execve() children. This
2930 * is important so that we never end up logging to any foreign stderr, for example if we have
2931 * to log in a child process right before execve()'ing the actual binary, at a point in time
2932 * where socket activation stderr/stdout area already set up. */
2933 log_set_always_reopen_console(true);
2935 if (detect_container() <= 0) {
2937 /* Running outside of a container as PID 1 */
2938 log_set_target_and_open(LOG_TARGET_KMSG
);
2941 initrd_timestamp
= userspace_timestamp
;
2944 r
= mount_setup_early();
2946 error_message
= "Failed to mount early API filesystems";
2951 /* We might have just mounted /proc, so let's try to parse the kernel
2952 * command line log arguments immediately. */
2953 log_parse_environment();
2955 /* Let's open the log backend a second time, in case the first time didn't
2956 * work. Quite possibly we have mounted /dev just now, so /dev/kmsg became
2957 * available, and it previously wasn't. */
2961 disable_printk_ratelimit();
2963 r
= initialize_security(
2965 &security_start_timestamp
,
2966 &security_finish_timestamp
,
2972 if (mac_init() < 0) {
2973 error_message
= "Failed to initialize MAC support";
2980 /* Set the default for later on, but don't actually open the logs like this for
2981 * now. Note that if we are transitioning from the initrd there might still be
2982 * journal fd open, and we shouldn't attempt opening that before we parsed
2983 * /proc/cmdline which might redirect output elsewhere. */
2984 log_set_target(LOG_TARGET_JOURNAL_OR_KMSG
);
2987 /* Running inside a container, as PID 1 */
2988 log_set_target_and_open(LOG_TARGET_CONSOLE
);
2990 /* For later on, see above... */
2991 log_set_target(LOG_TARGET_JOURNAL
);
2993 /* clear the kernel timestamp, because we are in a container */
2994 kernel_timestamp
= DUAL_TIMESTAMP_NULL
;
2997 initialize_coredump(skip_setup
);
2999 r
= fixup_environment();
3001 log_struct_errno(LOG_EMERG
, r
,
3002 LOG_MESSAGE("Failed to fix up PID 1 environment: %m"),
3003 "MESSAGE_ID=" SD_MESSAGE_CORE_PID1_ENVIRONMENT_STR
);
3004 error_message
= "Failed to fix up PID1 environment";
3008 /* Try to figure out if we can use colors with the console. No need to do that for user
3009 * instances since they never log into the console. */
3010 log_show_color(colors_enabled());
3012 r
= make_null_stdio();
3014 log_warning_errno(r
, "Failed to redirect standard streams to /dev/null, ignoring: %m");
3016 /* Load the kernel modules early. */
3018 (void) kmod_setup();
3020 /* Mount /proc, /sys and friends, so that /proc/cmdline and /proc/$PID/fd is available. */
3021 r
= mount_setup(loaded_policy
, skip_setup
);
3023 error_message
= "Failed to mount API filesystems";
3027 /* The efivarfs is now mounted, let's lock down the system token. */
3028 lock_down_efi_variables();
3030 /* Cache command-line options passed from EFI variables */
3032 (void) cache_efi_options_variable();
3034 /* Running as user instance */
3035 arg_runtime_scope
= RUNTIME_SCOPE_USER
;
3036 log_set_always_reopen_console(true);
3037 log_set_target_and_open(LOG_TARGET_AUTO
);
3039 /* clear the kernel timestamp, because we are not PID 1 */
3040 kernel_timestamp
= DUAL_TIMESTAMP_NULL
;
3042 /* Clear ambient capabilities, so services do not inherit them implicitly. Dropping them does
3043 * not affect the permitted and effective sets which are important for the manager itself to
3045 capability_ambient_set_apply(0, /* also_inherit= */ false);
3047 if (mac_init() < 0) {
3048 error_message
= "Failed to initialize MAC support";
3053 /* Save the original RLIMIT_NOFILE/RLIMIT_MEMLOCK so that we can reset it later when
3054 * transitioning from the initrd to the main systemd or suchlike. */
3055 save_rlimits(&saved_rlimit_nofile
, &saved_rlimit_memlock
);
3057 /* Reset all signal handlers. */
3058 (void) reset_all_signal_handlers();
3059 (void) ignore_signals(SIGNALS_IGNORE
);
3061 (void) parse_configuration(&saved_rlimit_nofile
, &saved_rlimit_memlock
);
3063 r
= parse_argv(argc
, argv
);
3065 error_message
= "Failed to parse command line arguments";
3069 r
= safety_checks();
3073 if (IN_SET(arg_action
, ACTION_TEST
, ACTION_HELP
, ACTION_DUMP_CONFIGURATION_ITEMS
, ACTION_DUMP_BUS_PROPERTIES
, ACTION_BUS_INTROSPECT
))
3074 pager_open(arg_pager_flags
);
3076 if (arg_action
!= ACTION_RUN
)
3079 if (arg_action
== ACTION_HELP
) {
3080 retval
= help() < 0 ? EXIT_FAILURE
: EXIT_SUCCESS
;
3082 } else if (arg_action
== ACTION_VERSION
) {
3085 } else if (arg_action
== ACTION_DUMP_CONFIGURATION_ITEMS
) {
3086 unit_dump_config_items(stdout
);
3087 retval
= EXIT_SUCCESS
;
3089 } else if (arg_action
== ACTION_DUMP_BUS_PROPERTIES
) {
3090 dump_bus_properties(stdout
);
3091 retval
= EXIT_SUCCESS
;
3093 } else if (arg_action
== ACTION_BUS_INTROSPECT
) {
3094 r
= bus_manager_introspect_implementations(stdout
, arg_bus_introspect
);
3095 retval
= r
>= 0 ? EXIT_SUCCESS
: EXIT_FAILURE
;
3099 assert_se(IN_SET(arg_action
, ACTION_RUN
, ACTION_TEST
));
3101 /* Move out of the way, so that we won't block unmounts */
3102 assert_se(chdir("/") == 0);
3104 if (arg_action
== ACTION_RUN
) {
3106 /* Apply the systemd.clock_usec= kernel command line switch */
3107 apply_clock_update();
3109 /* Apply random seed from kernel command line */
3110 cmdline_take_random_seed();
3113 /* A core pattern might have been specified via the cmdline. */
3114 initialize_core_pattern(skip_setup
);
3116 /* Make /usr/ read-only */
3117 apply_protect_system(skip_setup
);
3119 /* Close logging fds, in order not to confuse collecting passed fds and terminal logic below */
3122 /* Remember open file descriptors for later deserialization */
3123 r
= collect_fds(&fds
, &error_message
);
3127 /* Give up any control of the console, but make sure its initialized. */
3128 setup_console_terminal(skip_setup
);
3130 /* Open the logging devices, if possible and necessary */
3134 log_execution_mode(&first_boot
);
3136 r
= initialize_runtime(skip_setup
,
3138 &saved_rlimit_nofile
,
3139 &saved_rlimit_memlock
,
3144 r
= manager_new(arg_runtime_scope
,
3145 arg_action
== ACTION_TEST
? MANAGER_TEST_FULL
: 0,
3148 log_struct_errno(LOG_EMERG
, r
,
3149 LOG_MESSAGE("Failed to allocate manager object: %m"),
3150 "MESSAGE_ID=" SD_MESSAGE_CORE_MANAGER_ALLOCATE_STR
);
3151 error_message
= "Failed to allocate manager object";
3155 m
->timestamps
[MANAGER_TIMESTAMP_KERNEL
] = kernel_timestamp
;
3156 m
->timestamps
[MANAGER_TIMESTAMP_INITRD
] = initrd_timestamp
;
3157 m
->timestamps
[MANAGER_TIMESTAMP_USERSPACE
] = userspace_timestamp
;
3158 m
->timestamps
[manager_timestamp_initrd_mangle(MANAGER_TIMESTAMP_SECURITY_START
)] = security_start_timestamp
;
3159 m
->timestamps
[manager_timestamp_initrd_mangle(MANAGER_TIMESTAMP_SECURITY_FINISH
)] = security_finish_timestamp
;
3161 set_manager_defaults(m
);
3162 set_manager_settings(m
);
3163 manager_set_first_boot(m
, first_boot
);
3164 manager_set_switching_root(m
, arg_switched_root
);
3166 /* Remember whether we should queue the default job */
3167 queue_default_job
= !arg_serialization
|| arg_switched_root
;
3169 before_startup
= now(CLOCK_MONOTONIC
);
3171 r
= manager_startup(m
, arg_serialization
, fds
, /* root= */ NULL
);
3173 error_message
= "Failed to start up manager";
3177 /* This will close all file descriptors that were opened, but not claimed by any unit. */
3178 fds
= fdset_free(fds
);
3179 arg_serialization
= safe_fclose(arg_serialization
);
3181 if (queue_default_job
) {
3182 r
= do_queue_default_job(m
, &error_message
);
3187 after_startup
= now(CLOCK_MONOTONIC
);
3189 log_full(arg_action
== ACTION_TEST
? LOG_INFO
: LOG_DEBUG
,
3190 "Loaded units and determined initial transaction in %s.",
3191 FORMAT_TIMESPAN(after_startup
- before_startup
, 100 * USEC_PER_MSEC
));
3193 if (arg_action
== ACTION_TEST
) {
3194 manager_test_summary(m
);
3195 retval
= EXIT_SUCCESS
;
3199 r
= invoke_main_loop(m
,
3200 &saved_rlimit_nofile
,
3201 &saved_rlimit_memlock
,
3207 assert(r
< 0 || IN_SET(r
, MANAGER_EXIT
, /* MANAGER_OK is not expected here. */
3211 MANAGER_SOFT_REBOOT
,
3215 MANAGER_SWITCH_ROOT
));
3221 arg_reboot_watchdog
= manager_get_watchdog(m
, WATCHDOG_REBOOT
);
3222 arg_kexec_watchdog
= manager_get_watchdog(m
, WATCHDOG_KEXEC
);
3223 m
= manager_free(m
);
3226 mac_selinux_finish();
3228 if (IN_SET(r
, MANAGER_REEXECUTE
, MANAGER_SWITCH_ROOT
, MANAGER_SOFT_REBOOT
))
3231 &saved_rlimit_nofile
,
3232 &saved_rlimit_memlock
,
3236 &error_message
); /* This only returns if reexecution failed */
3238 arg_serialization
= safe_fclose(arg_serialization
);
3239 fds
= fdset_free(fds
);
3241 saved_env
= strv_free(saved_env
);
3243 #if HAVE_VALGRIND_VALGRIND_H
3244 /* If we are PID 1 and running under valgrind, then let's exit
3245 * here explicitly. valgrind will only generate nice output on
3246 * exit(), not on exec(), hence let's do the former not the
3248 if (getpid_cached() == 1 && RUNNING_ON_VALGRIND
) {
3249 /* Cleanup watchdog_device strings for valgrind. We need them
3250 * in become_shutdown() so normally we cannot free them yet. */
3251 watchdog_free_device();
3257 #if HAS_FEATURE_ADDRESS_SANITIZER
3258 /* At this stage we most likely don't have stdio/stderr open, so the following
3259 * LSan check would not print any actionable information and would just crash
3260 * PID 1. To make this a bit more helpful, let's try to open /dev/console,
3261 * and if we succeed redirect LSan's report there. */
3262 if (getpid_cached() == 1) {
3263 _cleanup_close_
int tty_fd
= -EBADF
;
3265 tty_fd
= open_terminal("/dev/console", O_WRONLY
|O_NOCTTY
|O_CLOEXEC
);
3267 __sanitizer_set_report_fd((void*) (intptr_t) tty_fd
);
3269 __lsan_do_leak_check();
3274 (void) sd_notifyf(0, "ERRNO=%i", -r
);
3276 /* Try to invoke the shutdown binary unless we already failed.
3277 * If we failed above, we want to freeze after finishing cleanup. */
3278 if (arg_runtime_scope
== RUNTIME_SCOPE_SYSTEM
&&
3279 IN_SET(r
, MANAGER_EXIT
, MANAGER_REBOOT
, MANAGER_POWEROFF
, MANAGER_HALT
, MANAGER_KEXEC
)) {
3280 r
= become_shutdown(r
, retval
);
3281 log_error_errno(r
, "Failed to execute shutdown binary, %s: %m", getpid_cached() == 1 ? "freezing" : "quitting");
3282 error_message
= "Failed to execute shutdown binary";
3285 /* This is primarily useful when running systemd in a VM, as it provides the user running the VM with
3286 * a mechanism to pick up systemd's exit status in the VM. */
3287 (void) sd_notifyf(0, "EXIT_STATUS=%i", retval
);
3289 watchdog_free_device();
3290 arg_watchdog_device
= mfree(arg_watchdog_device
);
3292 if (getpid_cached() == 1) {
3294 manager_status_printf(NULL
, STATUS_TYPE_EMERGENCY
,
3295 ANSI_HIGHLIGHT_RED
"!!!!!!" ANSI_NORMAL
,
3296 "%s.", error_message
);
3297 freeze_or_exit_or_reboot();