1 /* SPDX-License-Identifier: LGPL-2.1-or-later */
7 #include <sys/inotify.h>
10 #include <sys/reboot.h>
11 #include <sys/timerfd.h>
12 #include <sys/utsname.h>
20 #include "sd-daemon.h"
21 #include "sd-messages.h"
24 #include "all-units.h"
25 #include "alloc-util.h"
27 #include "boot-timestamps.h"
28 #include "build-path.h"
29 #include "bus-common-errors.h"
30 #include "bus-error.h"
31 #include "bus-kernel.h"
33 #include "clean-ipc.h"
34 #include "clock-util.h"
35 #include "common-signal.h"
36 #include "confidential-virt.h"
37 #include "constants.h"
38 #include "core-varlink.h"
39 #include "creds-util.h"
40 #include "daemon-util.h"
42 #include "dbus-manager.h"
43 #include "dbus-unit.h"
45 #include "dirent-util.h"
48 #include "event-util.h"
49 #include "exec-util.h"
51 #include "exit-status.h"
54 #include "generator-setup.h"
56 #include "initrd-util.h"
57 #include "inotify-util.h"
60 #include "label-util.h"
61 #include "load-fragment.h"
62 #include "locale-setup.h"
66 #include "manager-dump.h"
67 #include "manager-serialize.h"
68 #include "memory-util.h"
69 #include "mkdir-label.h"
70 #include "mount-util.h"
72 #include "parse-util.h"
73 #include "path-lookup.h"
74 #include "path-util.h"
75 #include "plymouth-util.h"
76 #include "pretty-print.h"
77 #include "process-util.h"
79 #include "ratelimit.h"
80 #include "rlimit-util.h"
82 #include "selinux-util.h"
83 #include "signal-util.h"
84 #include "socket-util.h"
86 #include "stat-util.h"
87 #include "string-table.h"
88 #include "string-util.h"
91 #include "sysctl-util.h"
92 #include "syslog-util.h"
93 #include "terminal-util.h"
94 #include "time-util.h"
95 #include "transaction.h"
96 #include "uid-range.h"
97 #include "umask-util.h"
98 #include "unit-name.h"
99 #include "user-util.h"
101 #include "watchdog.h"
103 #define NOTIFY_RCVBUF_SIZE (8*1024*1024)
104 #define CGROUPS_AGENT_RCVBUF_SIZE (8*1024*1024)
106 /* Initial delay and the interval for printing status messages about running jobs */
107 #define JOBS_IN_PROGRESS_WAIT_USEC (2*USEC_PER_SEC)
108 #define JOBS_IN_PROGRESS_QUIET_WAIT_USEC (25*USEC_PER_SEC)
109 #define JOBS_IN_PROGRESS_PERIOD_USEC (USEC_PER_SEC / 3)
110 #define JOBS_IN_PROGRESS_PERIOD_DIVISOR 3
112 /* If there are more than 1K bus messages queue across our API and direct buses, then let's not add more on top until
113 * the queue gets more empty. */
114 #define MANAGER_BUS_BUSY_THRESHOLD 1024LU
116 /* How many units and jobs to process of the bus queue before returning to the event loop. */
117 #define MANAGER_BUS_MESSAGE_BUDGET 100U
119 #define DEFAULT_TASKS_MAX ((CGroupTasksMax) { 15U, 100U }) /* 15% */
121 static int manager_dispatch_notify_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
122 static int manager_dispatch_cgroups_agent_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
123 static int manager_dispatch_signal_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
124 static int manager_dispatch_time_change_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
125 static int manager_dispatch_idle_pipe_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
126 static int manager_dispatch_user_lookup_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
127 static int manager_dispatch_jobs_in_progress(sd_event_source
*source
, usec_t usec
, void *userdata
);
128 static int manager_dispatch_run_queue(sd_event_source
*source
, void *userdata
);
129 static int manager_dispatch_sigchld(sd_event_source
*source
, void *userdata
);
130 static int manager_dispatch_timezone_change(sd_event_source
*source
, const struct inotify_event
*event
, void *userdata
);
131 static int manager_run_environment_generators(Manager
*m
);
132 static int manager_run_generators(Manager
*m
);
133 static void manager_vacuum(Manager
*m
);
135 static usec_t
manager_watch_jobs_next_time(Manager
*m
) {
138 if (MANAGER_IS_USER(m
))
139 /* Let the user manager without a timeout show status quickly, so the system manager can make
140 * use of it, if it wants to. */
141 timeout
= JOBS_IN_PROGRESS_WAIT_USEC
* 2 / 3;
142 else if (show_status_on(m
->show_status
))
143 /* When status is on, just use the usual timeout. */
144 timeout
= JOBS_IN_PROGRESS_WAIT_USEC
;
146 timeout
= JOBS_IN_PROGRESS_QUIET_WAIT_USEC
;
148 return usec_add(now(CLOCK_MONOTONIC
), timeout
);
151 static bool manager_is_confirm_spawn_disabled(Manager
*m
) {
154 if (!m
->confirm_spawn
)
157 return access("/run/systemd/confirm_spawn_disabled", F_OK
) >= 0;
160 static void manager_watch_jobs_in_progress(Manager
*m
) {
166 /* We do not want to show the cylon animation if the user
167 * needs to confirm service executions otherwise confirmation
168 * messages will be screwed by the cylon animation. */
169 if (!manager_is_confirm_spawn_disabled(m
))
172 if (m
->jobs_in_progress_event_source
)
175 next
= manager_watch_jobs_next_time(m
);
176 r
= sd_event_add_time(
178 &m
->jobs_in_progress_event_source
,
181 manager_dispatch_jobs_in_progress
, m
);
185 (void) sd_event_source_set_description(m
->jobs_in_progress_event_source
, "manager-jobs-in-progress");
188 static void manager_flip_auto_status(Manager
*m
, bool enable
, const char *reason
) {
192 if (m
->show_status
== SHOW_STATUS_AUTO
)
193 manager_set_show_status(m
, SHOW_STATUS_TEMPORARY
, reason
);
195 if (m
->show_status
== SHOW_STATUS_TEMPORARY
)
196 manager_set_show_status(m
, SHOW_STATUS_AUTO
, reason
);
200 static void manager_print_jobs_in_progress(Manager
*m
) {
202 unsigned counter
= 0, print_nr
;
203 char cylon
[6 + CYLON_BUFFER_EXTRA
+ 1];
205 uint64_t timeout
= 0;
208 assert(m
->n_running_jobs
> 0);
210 manager_flip_auto_status(m
, true, "delay");
212 print_nr
= (m
->jobs_in_progress_iteration
/ JOBS_IN_PROGRESS_PERIOD_DIVISOR
) % m
->n_running_jobs
;
214 HASHMAP_FOREACH(j
, m
->jobs
)
215 if (j
->state
== JOB_RUNNING
&& counter
++ == print_nr
)
218 /* m->n_running_jobs must be consistent with the contents of m->jobs,
219 * so the above loop must have succeeded in finding j. */
220 assert(counter
== print_nr
+ 1);
223 cylon_pos
= m
->jobs_in_progress_iteration
% 14;
225 cylon_pos
= 14 - cylon_pos
;
226 draw_cylon(cylon
, sizeof(cylon
), 6, cylon_pos
);
228 m
->jobs_in_progress_iteration
++;
230 char job_of_n
[STRLEN("( of ) ") + DECIMAL_STR_MAX(unsigned)*2] = "";
231 if (m
->n_running_jobs
> 1)
232 xsprintf(job_of_n
, "(%u of %u) ", counter
, m
->n_running_jobs
);
234 (void) job_get_timeout(j
, &timeout
);
236 /* We want to use enough information for the user to identify previous lines talking about the same
237 * unit, but keep the message as short as possible. So if 'Starting foo.service' or 'Starting
238 * foo.service - Description' were used, 'foo.service' is enough here. On the other hand, if we used
239 * 'Starting Description' before, then we shall also use 'Description' here. So we pass NULL as the
240 * second argument to unit_status_string(). */
241 const char *ident
= unit_status_string(j
->unit
, NULL
);
243 const char *time
= FORMAT_TIMESPAN(now(CLOCK_MONOTONIC
) - j
->begin_usec
, 1*USEC_PER_SEC
);
244 const char *limit
= timeout
> 0 ? FORMAT_TIMESPAN(timeout
- j
->begin_usec
, 1*USEC_PER_SEC
) : "no limit";
246 if (m
->status_unit_format
== STATUS_UNIT_FORMAT_DESCRIPTION
)
247 /* When using 'Description', we effectively don't have enough space to show the nested status
248 * without ellipsization, so let's not even try. */
249 manager_status_printf(m
, STATUS_TYPE_EPHEMERAL
, cylon
,
250 "%sA %s job is running for %s (%s / %s)",
252 job_type_to_string(j
->type
),
256 const char *status_text
= unit_status_text(j
->unit
);
258 manager_status_printf(m
, STATUS_TYPE_EPHEMERAL
, cylon
,
259 "%sJob %s/%s running (%s / %s)%s%s",
262 job_type_to_string(j
->type
),
264 status_text
? ": " : "",
265 strempty(status_text
));
268 (void) sd_notifyf(/* unset_environment= */ false,
269 "STATUS=%sUser job %s/%s running (%s / %s)...",
271 ident
, job_type_to_string(j
->type
),
273 m
->status_ready
= false;
276 static int have_ask_password(void) {
277 _cleanup_closedir_
DIR *dir
= NULL
;
279 dir
= opendir("/run/systemd/ask-password");
287 FOREACH_DIRENT_ALL(de
, dir
, return -errno
)
288 if (startswith(de
->d_name
, "ask."))
293 static int manager_dispatch_ask_password_fd(sd_event_source
*source
,
294 int fd
, uint32_t revents
, void *userdata
) {
295 Manager
*m
= ASSERT_PTR(userdata
);
299 m
->have_ask_password
= have_ask_password();
300 if (m
->have_ask_password
< 0)
301 /* Log error but continue. Negative have_ask_password
302 * is treated as unknown status. */
303 log_error_errno(m
->have_ask_password
, "Failed to list /run/systemd/ask-password: %m");
308 static void manager_close_ask_password(Manager
*m
) {
311 m
->ask_password_event_source
= sd_event_source_disable_unref(m
->ask_password_event_source
);
312 m
->ask_password_inotify_fd
= safe_close(m
->ask_password_inotify_fd
);
313 m
->have_ask_password
= -EINVAL
;
316 static int manager_check_ask_password(Manager
*m
) {
321 if (!m
->ask_password_event_source
) {
322 assert(m
->ask_password_inotify_fd
< 0);
324 (void) mkdir_p_label("/run/systemd/ask-password", 0755);
326 m
->ask_password_inotify_fd
= inotify_init1(IN_NONBLOCK
|IN_CLOEXEC
);
327 if (m
->ask_password_inotify_fd
< 0)
328 return log_error_errno(errno
, "Failed to create inotify object: %m");
330 r
= inotify_add_watch_and_warn(m
->ask_password_inotify_fd
,
331 "/run/systemd/ask-password",
332 IN_CREATE
|IN_DELETE
|IN_MOVE
);
334 manager_close_ask_password(m
);
338 r
= sd_event_add_io(m
->event
, &m
->ask_password_event_source
,
339 m
->ask_password_inotify_fd
, EPOLLIN
,
340 manager_dispatch_ask_password_fd
, m
);
342 log_error_errno(r
, "Failed to add event source for /run/systemd/ask-password: %m");
343 manager_close_ask_password(m
);
347 (void) sd_event_source_set_description(m
->ask_password_event_source
, "manager-ask-password");
349 /* Queries might have been added meanwhile... */
350 manager_dispatch_ask_password_fd(m
->ask_password_event_source
,
351 m
->ask_password_inotify_fd
, EPOLLIN
, m
);
354 return m
->have_ask_password
;
357 static int manager_watch_idle_pipe(Manager
*m
) {
362 if (m
->idle_pipe_event_source
)
365 if (m
->idle_pipe
[2] < 0)
368 r
= sd_event_add_io(m
->event
, &m
->idle_pipe_event_source
, m
->idle_pipe
[2], EPOLLIN
, manager_dispatch_idle_pipe_fd
, m
);
370 return log_error_errno(r
, "Failed to watch idle pipe: %m");
372 (void) sd_event_source_set_description(m
->idle_pipe_event_source
, "manager-idle-pipe");
377 static void manager_close_idle_pipe(Manager
*m
) {
380 m
->idle_pipe_event_source
= sd_event_source_disable_unref(m
->idle_pipe_event_source
);
382 safe_close_pair(m
->idle_pipe
);
383 safe_close_pair(m
->idle_pipe
+ 2);
386 static int manager_setup_time_change(Manager
*m
) {
391 if (MANAGER_IS_TEST_RUN(m
))
394 m
->time_change_event_source
= sd_event_source_disable_unref(m
->time_change_event_source
);
396 r
= event_add_time_change(m
->event
, &m
->time_change_event_source
, manager_dispatch_time_change_fd
, m
);
398 return log_error_errno(r
, "Failed to create time change event source: %m");
400 /* Schedule this slightly earlier than the .timer event sources */
401 r
= sd_event_source_set_priority(m
->time_change_event_source
, EVENT_PRIORITY_TIME_CHANGE
);
403 return log_error_errno(r
, "Failed to set priority of time change event sources: %m");
405 log_debug("Set up TFD_TIMER_CANCEL_ON_SET timerfd.");
410 static int manager_read_timezone_stat(Manager
*m
) {
416 /* Read the current stat() data of /etc/localtime so that we detect changes */
417 if (lstat("/etc/localtime", &st
) < 0) {
418 log_debug_errno(errno
, "Failed to stat /etc/localtime, ignoring: %m");
419 changed
= m
->etc_localtime_accessible
;
420 m
->etc_localtime_accessible
= false;
424 k
= timespec_load(&st
.st_mtim
);
425 changed
= !m
->etc_localtime_accessible
|| k
!= m
->etc_localtime_mtime
;
427 m
->etc_localtime_mtime
= k
;
428 m
->etc_localtime_accessible
= true;
434 static int manager_setup_timezone_change(Manager
*m
) {
435 _cleanup_(sd_event_source_unrefp
) sd_event_source
*new_event
= NULL
;
440 if (MANAGER_IS_TEST_RUN(m
))
443 /* We watch /etc/localtime for three events: change of the link count (which might mean removal from /etc even
444 * though another link might be kept), renames, and file close operations after writing. Note we don't bother
445 * with IN_DELETE_SELF, as that would just report when the inode is removed entirely, i.e. after the link count
446 * went to zero and all fds to it are closed.
448 * Note that we never follow symlinks here. This is a simplification, but should cover almost all cases
451 * Note that we create the new event source first here, before releasing the old one. This should optimize
452 * behaviour as this way sd-event can reuse the old watch in case the inode didn't change. */
454 r
= sd_event_add_inotify(m
->event
, &new_event
, "/etc/localtime",
455 IN_ATTRIB
|IN_MOVE_SELF
|IN_CLOSE_WRITE
|IN_DONT_FOLLOW
, manager_dispatch_timezone_change
, m
);
457 /* If the file doesn't exist yet, subscribe to /etc instead, and wait until it is created either by
458 * O_CREATE or by rename() */
460 log_debug_errno(r
, "/etc/localtime doesn't exist yet, watching /etc instead.");
461 r
= sd_event_add_inotify(m
->event
, &new_event
, "/etc",
462 IN_CREATE
|IN_MOVED_TO
|IN_ONLYDIR
, manager_dispatch_timezone_change
, m
);
465 return log_error_errno(r
, "Failed to create timezone change event source: %m");
467 /* Schedule this slightly earlier than the .timer event sources */
468 r
= sd_event_source_set_priority(new_event
, EVENT_PRIORITY_TIME_ZONE
);
470 return log_error_errno(r
, "Failed to set priority of timezone change event sources: %m");
472 sd_event_source_unref(m
->timezone_change_event_source
);
473 m
->timezone_change_event_source
= TAKE_PTR(new_event
);
478 static int enable_special_signals(Manager
*m
) {
479 _cleanup_close_
int fd
= -EBADF
;
483 if (MANAGER_IS_TEST_RUN(m
))
486 /* Enable that we get SIGINT on control-alt-del. In containers this will fail with EPERM (older) or
487 * EINVAL (newer), so ignore that. */
488 if (reboot(RB_DISABLE_CAD
) < 0 && !IN_SET(errno
, EPERM
, EINVAL
))
489 log_warning_errno(errno
, "Failed to enable ctrl-alt-del handling, ignoring: %m");
491 fd
= open_terminal("/dev/tty0", O_RDWR
|O_NOCTTY
|O_CLOEXEC
);
493 /* Support systems without virtual console (ENOENT) gracefully */
494 log_full_errno(fd
== -ENOENT
? LOG_DEBUG
: LOG_WARNING
, fd
, "Failed to open /dev/tty0, ignoring: %m");
496 /* Enable that we get SIGWINCH on kbrequest */
497 if (ioctl(fd
, KDSIGACCEPT
, SIGWINCH
) < 0)
498 log_warning_errno(errno
, "Failed to enable kbrequest handling, ignoring: %m");
504 #define RTSIG_IF_AVAILABLE(signum) (signum <= SIGRTMAX ? signum : -1)
506 static int manager_setup_signals(Manager
*m
) {
507 struct sigaction sa
= {
508 .sa_handler
= SIG_DFL
,
509 .sa_flags
= SA_NOCLDSTOP
|SA_RESTART
,
516 assert_se(sigaction(SIGCHLD
, &sa
, NULL
) == 0);
518 /* We make liberal use of realtime signals here. On
519 * Linux/glibc we have 30 of them (with the exception of Linux
520 * on hppa, see below), between SIGRTMIN+0 ... SIGRTMIN+30
523 assert_se(sigemptyset(&mask
) == 0);
524 sigset_add_many(&mask
,
525 SIGCHLD
, /* Child died */
526 SIGTERM
, /* Reexecute daemon */
527 SIGHUP
, /* Reload configuration */
528 SIGUSR1
, /* systemd: reconnect to D-Bus */
529 SIGUSR2
, /* systemd: dump status */
530 SIGINT
, /* Kernel sends us this on control-alt-del */
531 SIGWINCH
, /* Kernel sends us this on kbrequest (alt-arrowup) */
532 SIGPWR
, /* Some kernel drivers and upsd send us this on power failure */
534 SIGRTMIN
+0, /* systemd: start default.target */
535 SIGRTMIN
+1, /* systemd: isolate rescue.target */
536 SIGRTMIN
+2, /* systemd: isolate emergency.target */
537 SIGRTMIN
+3, /* systemd: start halt.target */
538 SIGRTMIN
+4, /* systemd: start poweroff.target */
539 SIGRTMIN
+5, /* systemd: start reboot.target */
540 SIGRTMIN
+6, /* systemd: start kexec.target */
541 SIGRTMIN
+7, /* systemd: start soft-reboot.target */
543 /* ... space for more special targets ... */
545 SIGRTMIN
+13, /* systemd: Immediate halt */
546 SIGRTMIN
+14, /* systemd: Immediate poweroff */
547 SIGRTMIN
+15, /* systemd: Immediate reboot */
548 SIGRTMIN
+16, /* systemd: Immediate kexec */
549 SIGRTMIN
+17, /* systemd: Immediate soft-reboot */
550 SIGRTMIN
+18, /* systemd: control command */
554 SIGRTMIN
+20, /* systemd: enable status messages */
555 SIGRTMIN
+21, /* systemd: disable status messages */
556 SIGRTMIN
+22, /* systemd: set log level to LOG_DEBUG */
557 SIGRTMIN
+23, /* systemd: set log level to LOG_INFO */
558 SIGRTMIN
+24, /* systemd: Immediate exit (--user only) */
559 SIGRTMIN
+25, /* systemd: reexecute manager */
561 /* Apparently Linux on hppa had fewer RT signals until v3.18,
562 * SIGRTMAX was SIGRTMIN+25, and then SIGRTMIN was lowered,
563 * see commit v3.17-7614-g1f25df2eff.
565 * We cannot unconditionally make use of those signals here,
566 * so let's use a runtime check. Since these commands are
567 * accessible by different means and only really a safety
568 * net, the missing functionality on hppa shouldn't matter.
571 RTSIG_IF_AVAILABLE(SIGRTMIN
+26), /* systemd: set log target to journal-or-kmsg */
572 RTSIG_IF_AVAILABLE(SIGRTMIN
+27), /* systemd: set log target to console */
573 RTSIG_IF_AVAILABLE(SIGRTMIN
+28), /* systemd: set log target to kmsg */
574 RTSIG_IF_AVAILABLE(SIGRTMIN
+29), /* systemd: set log target to syslog-or-kmsg (obsolete) */
576 /* ... one free signal here SIGRTMIN+30 ... */
578 assert_se(sigprocmask(SIG_SETMASK
, &mask
, NULL
) == 0);
580 m
->signal_fd
= signalfd(-1, &mask
, SFD_NONBLOCK
|SFD_CLOEXEC
);
581 if (m
->signal_fd
< 0)
584 r
= sd_event_add_io(m
->event
, &m
->signal_event_source
, m
->signal_fd
, EPOLLIN
, manager_dispatch_signal_fd
, m
);
588 (void) sd_event_source_set_description(m
->signal_event_source
, "manager-signal");
590 /* Process signals a bit earlier than the rest of things, but later than notify_fd processing, so that the
591 * notify processing can still figure out to which process/service a message belongs, before we reap the
592 * process. Also, process this before handling cgroup notifications, so that we always collect child exit
593 * status information before detecting that there's no process in a cgroup. */
594 r
= sd_event_source_set_priority(m
->signal_event_source
, EVENT_PRIORITY_SIGNALS
);
598 /* Report to supervisor that we now process the above signals. We report this as level "2", to
599 * indicate that we support more than sysvinit's signals (of course, sysvinit never sent this
600 * message, but conceptually it makes sense to consider level "1" to be equivalent to sysvinit's
601 * signal handling). Also, by setting this to "2" people looking for this hopefully won't
602 * misunderstand this as a boolean concept. Signal level 2 shall refer to the signals PID 1
603 * understands at the time of release of systemd v256, i.e. including basic SIGRTMIN+18 handling for
604 * memory pressure and stuff. When more signals are hooked up (or more SIGRTMIN+18 multiplex
605 * operations added, this level should be increased). */
606 (void) sd_notify(/* unset_environment= */ false,
607 "X_SYSTEMD_SIGNALS_LEVEL=2");
609 if (MANAGER_IS_SYSTEM(m
))
610 return enable_special_signals(m
);
615 static char** sanitize_environment(char **l
) {
617 /* Let's remove some environment variables that we need ourselves to communicate with our clients */
621 "CONFIGURATION_DIRECTORY",
622 "CREDENTIALS_DIRECTORY",
634 "MEMORY_PRESSURE_WATCH",
635 "MEMORY_PRESSURE_WRITE",
637 "MONITOR_EXIT_STATUS",
638 "MONITOR_INVOCATION_ID",
639 "MONITOR_SERVICE_RESULT",
650 "TRIGGER_TIMER_MONOTONIC_USEC",
651 "TRIGGER_TIMER_REALTIME_USEC",
656 /* Let's order the environment alphabetically, just to make it pretty */
660 int manager_default_environment(Manager
*m
) {
665 m
->transient_environment
= strv_free(m
->transient_environment
);
667 if (MANAGER_IS_SYSTEM(m
)) {
668 /* The system manager always starts with a clean environment for its children. It does not
669 * import the kernel's or the parents' exported variables.
671 * The initial passed environment is untouched to keep /proc/self/environ valid; it is used
672 * for tagging the init process inside containers. */
673 m
->transient_environment
= strv_new("PATH=" DEFAULT_PATH
);
674 if (!m
->transient_environment
)
677 /* Import locale variables LC_*= from configuration */
678 (void) locale_setup(&m
->transient_environment
);
680 /* The user manager passes its own environment along to its children, except for $PATH and
683 m
->transient_environment
= strv_copy(environ
);
684 if (!m
->transient_environment
)
687 r
= strv_env_replace_strdup(&m
->transient_environment
, "PATH=" DEFAULT_USER_PATH
);
691 /* Envvars set for our 'manager' class session are private and should not be propagated
692 * to children. Also it's likely that the graphical session will set these on their own. */
693 strv_env_unset_many(m
->transient_environment
,
697 "XDG_SESSION_DESKTOP",
702 sanitize_environment(m
->transient_environment
);
706 static int manager_setup_prefix(Manager
*m
) {
712 static const struct table_entry paths_system
[_EXEC_DIRECTORY_TYPE_MAX
] = {
713 [EXEC_DIRECTORY_RUNTIME
] = { SD_PATH_SYSTEM_RUNTIME
, NULL
},
714 [EXEC_DIRECTORY_STATE
] = { SD_PATH_SYSTEM_STATE_PRIVATE
, NULL
},
715 [EXEC_DIRECTORY_CACHE
] = { SD_PATH_SYSTEM_STATE_CACHE
, NULL
},
716 [EXEC_DIRECTORY_LOGS
] = { SD_PATH_SYSTEM_STATE_LOGS
, NULL
},
717 [EXEC_DIRECTORY_CONFIGURATION
] = { SD_PATH_SYSTEM_CONFIGURATION
, NULL
},
720 static const struct table_entry paths_user
[_EXEC_DIRECTORY_TYPE_MAX
] = {
721 [EXEC_DIRECTORY_RUNTIME
] = { SD_PATH_USER_RUNTIME
, NULL
},
722 [EXEC_DIRECTORY_STATE
] = { SD_PATH_USER_STATE_PRIVATE
, NULL
},
723 [EXEC_DIRECTORY_CACHE
] = { SD_PATH_USER_STATE_CACHE
, NULL
},
724 [EXEC_DIRECTORY_LOGS
] = { SD_PATH_USER_STATE_PRIVATE
, "log" },
725 [EXEC_DIRECTORY_CONFIGURATION
] = { SD_PATH_USER_CONFIGURATION
, NULL
},
730 const struct table_entry
*p
= MANAGER_IS_SYSTEM(m
) ? paths_system
: paths_user
;
733 for (ExecDirectoryType i
= 0; i
< _EXEC_DIRECTORY_TYPE_MAX
; i
++) {
734 r
= sd_path_lookup(p
[i
].type
, p
[i
].suffix
, &m
->prefix
[i
]);
736 return log_warning_errno(r
, "Failed to lookup %s path: %m",
737 exec_directory_type_to_string(i
));
743 static void manager_free_unit_name_maps(Manager
*m
) {
744 m
->unit_id_map
= hashmap_free(m
->unit_id_map
);
745 m
->unit_name_map
= hashmap_free(m
->unit_name_map
);
746 m
->unit_path_cache
= set_free(m
->unit_path_cache
);
747 m
->unit_cache_timestamp_hash
= 0;
750 static int manager_setup_run_queue(Manager
*m
) {
754 assert(!m
->run_queue_event_source
);
756 r
= sd_event_add_defer(m
->event
, &m
->run_queue_event_source
, manager_dispatch_run_queue
, m
);
760 r
= sd_event_source_set_priority(m
->run_queue_event_source
, EVENT_PRIORITY_RUN_QUEUE
);
764 r
= sd_event_source_set_enabled(m
->run_queue_event_source
, SD_EVENT_OFF
);
768 (void) sd_event_source_set_description(m
->run_queue_event_source
, "manager-run-queue");
773 static int manager_setup_sigchld_event_source(Manager
*m
) {
777 assert(!m
->sigchld_event_source
);
779 r
= sd_event_add_defer(m
->event
, &m
->sigchld_event_source
, manager_dispatch_sigchld
, m
);
783 r
= sd_event_source_set_priority(m
->sigchld_event_source
, EVENT_PRIORITY_SIGCHLD
);
787 r
= sd_event_source_set_enabled(m
->sigchld_event_source
, SD_EVENT_OFF
);
791 (void) sd_event_source_set_description(m
->sigchld_event_source
, "manager-sigchld");
796 int manager_setup_memory_pressure_event_source(Manager
*m
) {
801 m
->memory_pressure_event_source
= sd_event_source_disable_unref(m
->memory_pressure_event_source
);
803 r
= sd_event_add_memory_pressure(m
->event
, &m
->memory_pressure_event_source
, NULL
, NULL
);
805 log_full_errno(ERRNO_IS_NOT_SUPPORTED(r
) || ERRNO_IS_PRIVILEGE(r
) || (r
== -EHOSTDOWN
) ? LOG_DEBUG
: LOG_NOTICE
, r
,
806 "Failed to establish memory pressure event source, ignoring: %m");
807 else if (m
->defaults
.memory_pressure_threshold_usec
!= USEC_INFINITY
) {
809 /* If there's a default memory pressure threshold set, also apply it to the service manager itself */
810 r
= sd_event_source_set_memory_pressure_period(
811 m
->memory_pressure_event_source
,
812 m
->defaults
.memory_pressure_threshold_usec
,
813 MEMORY_PRESSURE_DEFAULT_WINDOW_USEC
);
815 log_warning_errno(r
, "Failed to adjust memory pressure threshold, ignoring: %m");
821 static int manager_find_credentials_dirs(Manager
*m
) {
827 r
= get_credentials_dir(&e
);
830 log_debug_errno(r
, "Failed to determine credentials directory, ignoring: %m");
832 m
->received_credentials_directory
= strdup(e
);
833 if (!m
->received_credentials_directory
)
837 r
= get_encrypted_credentials_dir(&e
);
840 log_debug_errno(r
, "Failed to determine encrypted credentials directory, ignoring: %m");
842 m
->received_encrypted_credentials_directory
= strdup(e
);
843 if (!m
->received_encrypted_credentials_directory
)
850 void manager_set_switching_root(Manager
*m
, bool switching_root
) {
853 m
->switching_root
= MANAGER_IS_SYSTEM(m
) && switching_root
;
856 double manager_get_progress(Manager
*m
) {
859 if (MANAGER_IS_FINISHED(m
) || m
->n_installed_jobs
== 0)
862 return 1.0 - ((double) hashmap_size(m
->jobs
) / (double) m
->n_installed_jobs
);
865 static int compare_job_priority(const void *a
, const void *b
) {
866 const Job
*x
= a
, *y
= b
;
868 return unit_compare_priority(x
->unit
, y
->unit
);
871 int manager_new(RuntimeScope runtime_scope
, ManagerTestRunFlags test_run_flags
, Manager
**ret
) {
872 _cleanup_(manager_freep
) Manager
*m
= NULL
;
875 assert(IN_SET(runtime_scope
, RUNTIME_SCOPE_SYSTEM
, RUNTIME_SCOPE_USER
));
883 .runtime_scope
= runtime_scope
,
884 .objective
= _MANAGER_OBJECTIVE_INVALID
,
885 .previous_objective
= _MANAGER_OBJECTIVE_INVALID
,
887 .status_unit_format
= STATUS_UNIT_FORMAT_DEFAULT
,
889 .original_log_level
= -1,
890 .original_log_target
= _LOG_TARGET_INVALID
,
892 .watchdog_overridden
[WATCHDOG_RUNTIME
] = USEC_INFINITY
,
893 .watchdog_overridden
[WATCHDOG_REBOOT
] = USEC_INFINITY
,
894 .watchdog_overridden
[WATCHDOG_KEXEC
] = USEC_INFINITY
,
895 .watchdog_overridden
[WATCHDOG_PRETIMEOUT
] = USEC_INFINITY
,
897 .show_status_overridden
= _SHOW_STATUS_INVALID
,
900 .cgroups_agent_fd
= -EBADF
,
902 .user_lookup_fds
= EBADF_PAIR
,
903 .private_listen_fd
= -EBADF
,
904 .dev_autofs_fd
= -EBADF
,
905 .cgroup_inotify_fd
= -EBADF
,
906 .pin_cgroupfs_fd
= -EBADF
,
907 .ask_password_inotify_fd
= -EBADF
,
908 .idle_pipe
= { -EBADF
, -EBADF
, -EBADF
, -EBADF
},
910 /* start as id #1, so that we can leave #0 around as "null-like" value */
913 .have_ask_password
= -EINVAL
, /* we don't know */
915 .test_run_flags
= test_run_flags
,
917 .dump_ratelimit
= (const RateLimit
) { .interval
= 10 * USEC_PER_MINUTE
, .burst
= 10 },
919 .executor_fd
= -EBADF
,
922 unit_defaults_init(&m
->defaults
, runtime_scope
);
925 if (MANAGER_IS_SYSTEM(m
) && detect_container() <= 0)
926 boot_timestamps(m
->timestamps
+ MANAGER_TIMESTAMP_USERSPACE
,
927 m
->timestamps
+ MANAGER_TIMESTAMP_FIRMWARE
,
928 m
->timestamps
+ MANAGER_TIMESTAMP_LOADER
);
931 /* Prepare log fields we can use for structured logging */
932 if (MANAGER_IS_SYSTEM(m
)) {
933 m
->unit_log_field
= "UNIT=";
934 m
->unit_log_format_string
= "UNIT=%s";
936 m
->invocation_log_field
= "INVOCATION_ID=";
937 m
->invocation_log_format_string
= "INVOCATION_ID=%s";
939 m
->unit_log_field
= "USER_UNIT=";
940 m
->unit_log_format_string
= "USER_UNIT=%s";
942 m
->invocation_log_field
= "USER_INVOCATION_ID=";
943 m
->invocation_log_format_string
= "USER_INVOCATION_ID=%s";
946 /* Reboot immediately if the user hits C-A-D more often than 7x per 2s */
947 m
->ctrl_alt_del_ratelimit
= (const RateLimit
) { .interval
= 2 * USEC_PER_SEC
, .burst
= 7 };
949 r
= manager_default_environment(m
);
953 r
= hashmap_ensure_allocated(&m
->units
, &string_hash_ops
);
957 r
= hashmap_ensure_allocated(&m
->cgroup_unit
, &path_hash_ops
);
961 r
= hashmap_ensure_allocated(&m
->watch_bus
, &string_hash_ops
);
965 r
= prioq_ensure_allocated(&m
->run_queue
, compare_job_priority
);
969 r
= manager_setup_prefix(m
);
973 r
= manager_find_credentials_dirs(m
);
977 r
= sd_event_default(&m
->event
);
981 r
= manager_setup_run_queue(m
);
985 if (FLAGS_SET(test_run_flags
, MANAGER_TEST_RUN_MINIMAL
)) {
986 m
->cgroup_root
= strdup("");
990 r
= manager_setup_signals(m
);
994 r
= manager_setup_cgroup(m
);
998 r
= manager_setup_time_change(m
);
1002 r
= manager_read_timezone_stat(m
);
1006 (void) manager_setup_timezone_change(m
);
1008 r
= manager_setup_sigchld_event_source(m
);
1012 r
= manager_setup_memory_pressure_event_source(m
);
1017 if (MANAGER_IS_SYSTEM(m
) && bpf_restrict_fs_supported(/* initialize = */ true)) {
1018 r
= bpf_restrict_fs_setup(m
);
1020 log_warning_errno(r
, "Failed to setup LSM BPF, ignoring: %m");
1025 if (test_run_flags
== 0) {
1026 if (MANAGER_IS_SYSTEM(m
))
1027 r
= mkdir_label("/run/systemd/units", 0755);
1029 _cleanup_free_
char *units_path
= NULL
;
1030 r
= xdg_user_runtime_dir(&units_path
, "/systemd/units");
1033 r
= mkdir_p_label(units_path
, 0755);
1036 if (r
< 0 && r
!= -EEXIST
)
1040 if (!FLAGS_SET(test_run_flags
, MANAGER_TEST_DONT_OPEN_EXECUTOR
)) {
1041 m
->executor_fd
= pin_callout_binary(SYSTEMD_EXECUTOR_BINARY_PATH
);
1042 if (m
->executor_fd
< 0)
1043 return log_debug_errno(m
->executor_fd
, "Failed to pin executor binary: %m");
1045 _cleanup_free_
char *executor_path
= NULL
;
1046 r
= fd_get_path(m
->executor_fd
, &executor_path
);
1050 log_debug("Using systemd-executor binary from '%s'.", executor_path
);
1053 /* Note that we do not set up the notify fd here. We do that after deserialization,
1054 * since they might have gotten serialized across the reexec. */
1061 static int manager_setup_notify(Manager
*m
) {
1064 if (MANAGER_IS_TEST_RUN(m
))
1067 if (m
->notify_fd
< 0) {
1068 _cleanup_close_
int fd
= -EBADF
;
1069 union sockaddr_union sa
;
1072 /* First free all secondary fields */
1073 m
->notify_socket
= mfree(m
->notify_socket
);
1074 m
->notify_event_source
= sd_event_source_disable_unref(m
->notify_event_source
);
1076 fd
= socket(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
|SOCK_NONBLOCK
, 0);
1078 return log_error_errno(errno
, "Failed to allocate notification socket: %m");
1080 fd_increase_rxbuf(fd
, NOTIFY_RCVBUF_SIZE
);
1082 m
->notify_socket
= path_join(m
->prefix
[EXEC_DIRECTORY_RUNTIME
], "systemd/notify");
1083 if (!m
->notify_socket
)
1086 r
= sockaddr_un_set_path(&sa
.un
, m
->notify_socket
);
1088 return log_error_errno(r
, "Notify socket '%s' not valid for AF_UNIX socket address, refusing.",
1092 (void) mkdir_parents_label(m
->notify_socket
, 0755);
1093 (void) sockaddr_un_unlink(&sa
.un
);
1095 r
= mac_selinux_bind(fd
, &sa
.sa
, sa_len
);
1097 return log_error_errno(r
, "bind(%s) failed: %m", m
->notify_socket
);
1099 r
= setsockopt_int(fd
, SOL_SOCKET
, SO_PASSCRED
, true);
1101 return log_error_errno(r
, "SO_PASSCRED failed: %m");
1103 m
->notify_fd
= TAKE_FD(fd
);
1105 log_debug("Using notification socket %s", m
->notify_socket
);
1108 if (!m
->notify_event_source
) {
1109 r
= sd_event_add_io(m
->event
, &m
->notify_event_source
, m
->notify_fd
, EPOLLIN
, manager_dispatch_notify_fd
, m
);
1111 return log_error_errno(r
, "Failed to allocate notify event source: %m");
1113 /* Process notification messages a bit earlier than SIGCHLD, so that we can still identify to which
1114 * service an exit message belongs. */
1115 r
= sd_event_source_set_priority(m
->notify_event_source
, EVENT_PRIORITY_NOTIFY
);
1117 return log_error_errno(r
, "Failed to set priority of notify event source: %m");
1119 (void) sd_event_source_set_description(m
->notify_event_source
, "manager-notify");
1125 static int manager_setup_cgroups_agent(Manager
*m
) {
1127 static const union sockaddr_union sa
= {
1128 .un
.sun_family
= AF_UNIX
,
1129 .un
.sun_path
= "/run/systemd/cgroups-agent",
1133 /* This creates a listening socket we receive cgroups agent messages on. We do not use D-Bus for delivering
1134 * these messages from the cgroups agent binary to PID 1, as the cgroups agent binary is very short-living, and
1135 * each instance of it needs a new D-Bus connection. Since D-Bus connections are SOCK_STREAM/AF_UNIX, on
1136 * overloaded systems the backlog of the D-Bus socket becomes relevant, as not more than the configured number
1137 * of D-Bus connections may be queued until the kernel will start dropping further incoming connections,
1138 * possibly resulting in lost cgroups agent messages. To avoid this, we'll use a private SOCK_DGRAM/AF_UNIX
1139 * socket, where no backlog is relevant as communication may take place without an actual connect() cycle, and
1140 * we thus won't lose messages.
1142 * Note that PID 1 will forward the agent message to system bus, so that the user systemd instance may listen
1143 * to it. The system instance hence listens on this special socket, but the user instances listen on the system
1144 * bus for these messages. */
1146 if (MANAGER_IS_TEST_RUN(m
))
1149 if (!MANAGER_IS_SYSTEM(m
))
1152 r
= cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER
);
1154 return log_error_errno(r
, "Failed to determine whether unified cgroups hierarchy is used: %m");
1155 if (r
> 0) /* We don't need this anymore on the unified hierarchy */
1158 if (m
->cgroups_agent_fd
< 0) {
1159 _cleanup_close_
int fd
= -EBADF
;
1161 /* First free all secondary fields */
1162 m
->cgroups_agent_event_source
= sd_event_source_disable_unref(m
->cgroups_agent_event_source
);
1164 fd
= socket(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
|SOCK_NONBLOCK
, 0);
1166 return log_error_errno(errno
, "Failed to allocate cgroups agent socket: %m");
1168 fd_increase_rxbuf(fd
, CGROUPS_AGENT_RCVBUF_SIZE
);
1170 (void) sockaddr_un_unlink(&sa
.un
);
1172 /* Only allow root to connect to this socket */
1174 r
= bind(fd
, &sa
.sa
, SOCKADDR_UN_LEN(sa
.un
));
1176 return log_error_errno(errno
, "bind(%s) failed: %m", sa
.un
.sun_path
);
1178 m
->cgroups_agent_fd
= TAKE_FD(fd
);
1181 if (!m
->cgroups_agent_event_source
) {
1182 r
= sd_event_add_io(m
->event
, &m
->cgroups_agent_event_source
, m
->cgroups_agent_fd
, EPOLLIN
, manager_dispatch_cgroups_agent_fd
, m
);
1184 return log_error_errno(r
, "Failed to allocate cgroups agent event source: %m");
1186 /* Process cgroups notifications early. Note that when the agent notification is received
1187 * we'll just enqueue the unit in the cgroup empty queue, hence pick a high priority than
1188 * that. Also see handling of cgroup inotify for the unified cgroup stuff. */
1189 r
= sd_event_source_set_priority(m
->cgroups_agent_event_source
, EVENT_PRIORITY_CGROUP_AGENT
);
1191 return log_error_errno(r
, "Failed to set priority of cgroups agent event source: %m");
1193 (void) sd_event_source_set_description(m
->cgroups_agent_event_source
, "manager-cgroups-agent");
1199 static int manager_setup_user_lookup_fd(Manager
*m
) {
1204 /* Set up the socket pair used for passing UID/GID resolution results from forked off processes to PID
1205 * 1. Background: we can't do name lookups (NSS) from PID 1, since it might involve IPC and thus activation,
1206 * and we might hence deadlock on ourselves. Hence we do all user/group lookups asynchronously from the forked
1207 * off processes right before executing the binaries to start. In order to be able to clean up any IPC objects
1208 * created by a unit (see RemoveIPC=) we need to know in PID 1 the used UID/GID of the executed processes,
1209 * hence we establish this communication channel so that forked off processes can pass their UID/GID
1210 * information back to PID 1. The forked off processes send their resolved UID/GID to PID 1 in a simple
1211 * datagram, along with their unit name, so that we can share one communication socket pair among all units for
1214 * You might wonder why we need a communication channel for this that is independent of the usual notification
1215 * socket scheme (i.e. $NOTIFY_SOCKET). The primary difference is about trust: data sent via the $NOTIFY_SOCKET
1216 * channel is only accepted if it originates from the right unit and if reception was enabled for it. The user
1217 * lookup socket OTOH is only accessible by PID 1 and its children until they exec(), and always available.
1219 * Note that this function is called under two circumstances: when we first initialize (in which case we
1220 * allocate both the socket pair and the event source to listen on it), and when we deserialize after a reload
1221 * (in which case the socket pair already exists but we still need to allocate the event source for it). */
1223 if (m
->user_lookup_fds
[0] < 0) {
1225 /* Free all secondary fields */
1226 safe_close_pair(m
->user_lookup_fds
);
1227 m
->user_lookup_event_source
= sd_event_source_disable_unref(m
->user_lookup_event_source
);
1229 if (socketpair(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
, 0, m
->user_lookup_fds
) < 0)
1230 return log_error_errno(errno
, "Failed to allocate user lookup socket: %m");
1232 (void) fd_increase_rxbuf(m
->user_lookup_fds
[0], NOTIFY_RCVBUF_SIZE
);
1235 if (!m
->user_lookup_event_source
) {
1236 r
= sd_event_add_io(m
->event
, &m
->user_lookup_event_source
, m
->user_lookup_fds
[0], EPOLLIN
, manager_dispatch_user_lookup_fd
, m
);
1238 return log_error_errno(errno
, "Failed to allocate user lookup event source: %m");
1240 /* Process even earlier than the notify event source, so that we always know first about valid UID/GID
1242 r
= sd_event_source_set_priority(m
->user_lookup_event_source
, EVENT_PRIORITY_USER_LOOKUP
);
1244 return log_error_errno(errno
, "Failed to set priority of user lookup event source: %m");
1246 (void) sd_event_source_set_description(m
->user_lookup_event_source
, "user-lookup");
1252 static unsigned manager_dispatch_cleanup_queue(Manager
*m
) {
1258 while ((u
= m
->cleanup_queue
)) {
1259 assert(u
->in_cleanup_queue
);
1268 static unsigned manager_dispatch_release_resources_queue(Manager
*m
) {
1274 while ((u
= LIST_POP(release_resources_queue
, m
->release_resources_queue
))) {
1275 assert(u
->in_release_resources_queue
);
1276 u
->in_release_resources_queue
= false;
1280 unit_release_resources(u
);
1287 GC_OFFSET_IN_PATH
, /* This one is on the path we were traveling */
1288 GC_OFFSET_UNSURE
, /* No clue */
1289 GC_OFFSET_GOOD
, /* We still need this unit */
1290 GC_OFFSET_BAD
, /* We don't need this unit anymore */
1294 static void unit_gc_mark_good(Unit
*u
, unsigned gc_marker
) {
1297 u
->gc_marker
= gc_marker
+ GC_OFFSET_GOOD
;
1299 /* Recursively mark referenced units as GOOD as well */
1300 UNIT_FOREACH_DEPENDENCY(other
, u
, UNIT_ATOM_REFERENCES
)
1301 if (other
->gc_marker
== gc_marker
+ GC_OFFSET_UNSURE
)
1302 unit_gc_mark_good(other
, gc_marker
);
1305 static void unit_gc_sweep(Unit
*u
, unsigned gc_marker
) {
1311 if (IN_SET(u
->gc_marker
- gc_marker
,
1312 GC_OFFSET_GOOD
, GC_OFFSET_BAD
, GC_OFFSET_UNSURE
, GC_OFFSET_IN_PATH
))
1315 if (u
->in_cleanup_queue
)
1318 if (!unit_may_gc(u
))
1321 u
->gc_marker
= gc_marker
+ GC_OFFSET_IN_PATH
;
1325 UNIT_FOREACH_DEPENDENCY(other
, u
, UNIT_ATOM_REFERENCED_BY
) {
1326 unit_gc_sweep(other
, gc_marker
);
1328 if (other
->gc_marker
== gc_marker
+ GC_OFFSET_GOOD
)
1331 if (other
->gc_marker
!= gc_marker
+ GC_OFFSET_BAD
)
1335 LIST_FOREACH(refs_by_target
, ref
, u
->refs_by_target
) {
1336 unit_gc_sweep(ref
->source
, gc_marker
);
1338 if (ref
->source
->gc_marker
== gc_marker
+ GC_OFFSET_GOOD
)
1341 if (ref
->source
->gc_marker
!= gc_marker
+ GC_OFFSET_BAD
)
1348 /* We were unable to find anything out about this entry, so
1349 * let's investigate it later */
1350 u
->gc_marker
= gc_marker
+ GC_OFFSET_UNSURE
;
1351 unit_add_to_gc_queue(u
);
1355 /* We definitely know that this one is not useful anymore, so
1356 * let's mark it for deletion */
1357 u
->gc_marker
= gc_marker
+ GC_OFFSET_BAD
;
1358 unit_add_to_cleanup_queue(u
);
1362 unit_gc_mark_good(u
, gc_marker
);
1365 static unsigned manager_dispatch_gc_unit_queue(Manager
*m
) {
1366 unsigned n
= 0, gc_marker
;
1371 /* log_debug("Running GC..."); */
1373 m
->gc_marker
+= _GC_OFFSET_MAX
;
1374 if (m
->gc_marker
+ _GC_OFFSET_MAX
<= _GC_OFFSET_MAX
)
1377 gc_marker
= m
->gc_marker
;
1379 while ((u
= LIST_POP(gc_queue
, m
->gc_unit_queue
))) {
1380 assert(u
->in_gc_queue
);
1382 unit_gc_sweep(u
, gc_marker
);
1384 u
->in_gc_queue
= false;
1388 if (IN_SET(u
->gc_marker
- gc_marker
,
1389 GC_OFFSET_BAD
, GC_OFFSET_UNSURE
)) {
1391 log_unit_debug(u
, "Collecting.");
1392 u
->gc_marker
= gc_marker
+ GC_OFFSET_BAD
;
1393 unit_add_to_cleanup_queue(u
);
1400 static unsigned manager_dispatch_gc_job_queue(Manager
*m
) {
1406 while ((j
= LIST_POP(gc_queue
, m
->gc_job_queue
))) {
1407 assert(j
->in_gc_queue
);
1408 j
->in_gc_queue
= false;
1415 log_unit_debug(j
->unit
, "Collecting job.");
1416 (void) job_finish_and_invalidate(j
, JOB_COLLECTED
, false, false);
1422 static int manager_ratelimit_requeue(sd_event_source
*s
, uint64_t usec
, void *userdata
) {
1426 assert(s
== u
->auto_start_stop_event_source
);
1428 u
->auto_start_stop_event_source
= sd_event_source_unref(u
->auto_start_stop_event_source
);
1430 /* Re-queue to all queues, if the rate limit hit we might have been throttled on any of them. */
1431 unit_submit_to_stop_when_unneeded_queue(u
);
1432 unit_submit_to_start_when_upheld_queue(u
);
1433 unit_submit_to_stop_when_bound_queue(u
);
1438 static int manager_ratelimit_check_and_queue(Unit
*u
) {
1443 if (ratelimit_below(&u
->auto_start_stop_ratelimit
))
1446 /* Already queued, no need to requeue */
1447 if (u
->auto_start_stop_event_source
)
1450 r
= sd_event_add_time(
1452 &u
->auto_start_stop_event_source
,
1454 ratelimit_end(&u
->auto_start_stop_ratelimit
),
1456 manager_ratelimit_requeue
,
1459 return log_unit_error_errno(u
, r
, "Failed to queue timer on event loop: %m");
1464 static unsigned manager_dispatch_stop_when_unneeded_queue(Manager
*m
) {
1471 while ((u
= LIST_POP(stop_when_unneeded_queue
, m
->stop_when_unneeded_queue
))) {
1472 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
1474 assert(u
->in_stop_when_unneeded_queue
);
1475 u
->in_stop_when_unneeded_queue
= false;
1479 if (!unit_is_unneeded(u
))
1482 log_unit_debug(u
, "Unit is not needed anymore.");
1484 /* If stopping a unit fails continuously we might enter a stop loop here, hence stop acting on the
1485 * service being unnecessary after a while. */
1487 r
= manager_ratelimit_check_and_queue(u
);
1490 "Unit not needed anymore, but not stopping since we tried this too often recently.%s",
1491 r
== 0 ? " Will retry later." : "");
1495 /* Ok, nobody needs us anymore. Sniff. Then let's commit suicide */
1496 r
= manager_add_job(u
->manager
, JOB_STOP
, u
, JOB_FAIL
, NULL
, &error
, NULL
);
1498 log_unit_warning_errno(u
, r
, "Failed to enqueue stop job, ignoring: %s", bus_error_message(&error
, r
));
1504 static unsigned manager_dispatch_start_when_upheld_queue(Manager
*m
) {
1511 while ((u
= LIST_POP(start_when_upheld_queue
, m
->start_when_upheld_queue
))) {
1512 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
1513 Unit
*culprit
= NULL
;
1515 assert(u
->in_start_when_upheld_queue
);
1516 u
->in_start_when_upheld_queue
= false;
1520 if (!unit_is_upheld_by_active(u
, &culprit
))
1523 log_unit_debug(u
, "Unit is started because upheld by active unit %s.", culprit
->id
);
1525 /* If stopping a unit fails continuously we might enter a stop loop here, hence stop acting on the
1526 * service being unnecessary after a while. */
1528 r
= manager_ratelimit_check_and_queue(u
);
1531 "Unit needs to be started because active unit %s upholds it, but not starting since we tried this too often recently.%s",
1533 r
== 0 ? " Will retry later." : "");
1537 r
= manager_add_job(u
->manager
, JOB_START
, u
, JOB_FAIL
, NULL
, &error
, NULL
);
1539 log_unit_warning_errno(u
, r
, "Failed to enqueue start job, ignoring: %s", bus_error_message(&error
, r
));
1545 static unsigned manager_dispatch_stop_when_bound_queue(Manager
*m
) {
1552 while ((u
= LIST_POP(stop_when_bound_queue
, m
->stop_when_bound_queue
))) {
1553 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
1554 Unit
*culprit
= NULL
;
1556 assert(u
->in_stop_when_bound_queue
);
1557 u
->in_stop_when_bound_queue
= false;
1561 if (!unit_is_bound_by_inactive(u
, &culprit
))
1564 log_unit_debug(u
, "Unit is stopped because bound to inactive unit %s.", culprit
->id
);
1566 /* If stopping a unit fails continuously we might enter a stop loop here, hence stop acting on the
1567 * service being unnecessary after a while. */
1569 r
= manager_ratelimit_check_and_queue(u
);
1572 "Unit needs to be stopped because it is bound to inactive unit %s it, but not stopping since we tried this too often recently.%s",
1574 r
== 0 ? " Will retry later." : "");
1578 r
= manager_add_job(u
->manager
, JOB_STOP
, u
, JOB_REPLACE
, NULL
, &error
, NULL
);
1580 log_unit_warning_errno(u
, r
, "Failed to enqueue stop job, ignoring: %s", bus_error_message(&error
, r
));
1586 static void manager_clear_jobs_and_units(Manager
*m
) {
1591 while ((u
= hashmap_first(m
->units
)))
1594 manager_dispatch_cleanup_queue(m
);
1596 assert(!m
->load_queue
);
1597 assert(prioq_isempty(m
->run_queue
));
1598 assert(!m
->dbus_unit_queue
);
1599 assert(!m
->dbus_job_queue
);
1600 assert(!m
->cleanup_queue
);
1601 assert(!m
->gc_unit_queue
);
1602 assert(!m
->gc_job_queue
);
1603 assert(!m
->cgroup_realize_queue
);
1604 assert(!m
->cgroup_empty_queue
);
1605 assert(!m
->cgroup_oom_queue
);
1606 assert(!m
->target_deps_queue
);
1607 assert(!m
->stop_when_unneeded_queue
);
1608 assert(!m
->start_when_upheld_queue
);
1609 assert(!m
->stop_when_bound_queue
);
1610 assert(!m
->release_resources_queue
);
1612 assert(hashmap_isempty(m
->jobs
));
1613 assert(hashmap_isempty(m
->units
));
1615 m
->n_on_console
= 0;
1616 m
->n_running_jobs
= 0;
1617 m
->n_installed_jobs
= 0;
1618 m
->n_failed_jobs
= 0;
1621 Manager
* manager_free(Manager
*m
) {
1625 manager_clear_jobs_and_units(m
);
1627 for (UnitType c
= 0; c
< _UNIT_TYPE_MAX
; c
++)
1628 if (unit_vtable
[c
]->shutdown
)
1629 unit_vtable
[c
]->shutdown(m
);
1631 /* Keep the cgroup hierarchy in place except when we know we are going down for good */
1632 manager_shutdown_cgroup(m
, /* delete= */ IN_SET(m
->objective
, MANAGER_EXIT
, MANAGER_REBOOT
, MANAGER_POWEROFF
, MANAGER_HALT
, MANAGER_KEXEC
));
1634 lookup_paths_flush_generator(&m
->lookup_paths
);
1637 manager_varlink_done(m
);
1639 exec_shared_runtime_vacuum(m
);
1640 hashmap_free(m
->exec_shared_runtime_by_id
);
1642 dynamic_user_vacuum(m
, false);
1643 hashmap_free(m
->dynamic_users
);
1645 hashmap_free(m
->units
);
1646 hashmap_free(m
->units_by_invocation_id
);
1647 hashmap_free(m
->jobs
);
1648 hashmap_free(m
->watch_pids
);
1649 hashmap_free(m
->watch_pids_more
);
1650 hashmap_free(m
->watch_bus
);
1652 prioq_free(m
->run_queue
);
1654 set_free(m
->startup_units
);
1655 set_free(m
->failed_units
);
1657 sd_event_source_unref(m
->signal_event_source
);
1658 sd_event_source_unref(m
->sigchld_event_source
);
1659 sd_event_source_unref(m
->notify_event_source
);
1660 sd_event_source_unref(m
->cgroups_agent_event_source
);
1661 sd_event_source_unref(m
->time_change_event_source
);
1662 sd_event_source_unref(m
->timezone_change_event_source
);
1663 sd_event_source_unref(m
->jobs_in_progress_event_source
);
1664 sd_event_source_unref(m
->run_queue_event_source
);
1665 sd_event_source_unref(m
->user_lookup_event_source
);
1666 sd_event_source_unref(m
->memory_pressure_event_source
);
1668 safe_close(m
->signal_fd
);
1669 safe_close(m
->notify_fd
);
1670 safe_close(m
->cgroups_agent_fd
);
1671 safe_close_pair(m
->user_lookup_fds
);
1673 manager_close_ask_password(m
);
1675 manager_close_idle_pipe(m
);
1677 sd_event_unref(m
->event
);
1679 free(m
->notify_socket
);
1681 lookup_paths_done(&m
->lookup_paths
);
1682 strv_free(m
->transient_environment
);
1683 strv_free(m
->client_environment
);
1685 hashmap_free(m
->cgroup_unit
);
1686 manager_free_unit_name_maps(m
);
1688 free(m
->switch_root
);
1689 free(m
->switch_root_init
);
1691 unit_defaults_done(&m
->defaults
);
1693 FOREACH_ARRAY(map
, m
->units_needing_mounts_for
, _UNIT_MOUNT_DEPENDENCY_TYPE_MAX
) {
1694 assert(hashmap_isempty(*map
));
1698 hashmap_free(m
->uid_refs
);
1699 hashmap_free(m
->gid_refs
);
1701 for (ExecDirectoryType dt
= 0; dt
< _EXEC_DIRECTORY_TYPE_MAX
; dt
++)
1702 m
->prefix
[dt
] = mfree(m
->prefix
[dt
]);
1703 free(m
->received_credentials_directory
);
1704 free(m
->received_encrypted_credentials_directory
);
1706 free(m
->watchdog_pretimeout_governor
);
1707 free(m
->watchdog_pretimeout_governor_overridden
);
1709 m
->fw_ctx
= fw_ctx_free(m
->fw_ctx
);
1712 bpf_restrict_fs_destroy(m
->restrict_fs
);
1715 safe_close(m
->executor_fd
);
1720 static void manager_enumerate_perpetual(Manager
*m
) {
1723 if (FLAGS_SET(m
->test_run_flags
, MANAGER_TEST_RUN_MINIMAL
))
1726 /* Let's ask every type to load all units from disk/kernel that it might know */
1727 for (UnitType c
= 0; c
< _UNIT_TYPE_MAX
; c
++) {
1728 if (!unit_type_supported(c
)) {
1729 log_debug("Unit type .%s is not supported on this system.", unit_type_to_string(c
));
1733 if (unit_vtable
[c
]->enumerate_perpetual
)
1734 unit_vtable
[c
]->enumerate_perpetual(m
);
1738 static void manager_enumerate(Manager
*m
) {
1741 if (FLAGS_SET(m
->test_run_flags
, MANAGER_TEST_RUN_MINIMAL
))
1744 /* Let's ask every type to load all units from disk/kernel that it might know */
1745 for (UnitType c
= 0; c
< _UNIT_TYPE_MAX
; c
++) {
1746 if (!unit_type_supported(c
)) {
1747 log_debug("Unit type .%s is not supported on this system.", unit_type_to_string(c
));
1751 if (unit_vtable
[c
]->enumerate
)
1752 unit_vtable
[c
]->enumerate(m
);
1755 manager_dispatch_load_queue(m
);
1758 static void manager_coldplug(Manager
*m
) {
1765 log_debug("Invoking unit coldplug() handlers%s", special_glyph(SPECIAL_GLYPH_ELLIPSIS
));
1767 /* Let's place the units back into their deserialized state */
1768 HASHMAP_FOREACH_KEY(u
, k
, m
->units
) {
1770 /* ignore aliases */
1774 r
= unit_coldplug(u
);
1776 log_warning_errno(r
, "We couldn't coldplug %s, proceeding anyway: %m", u
->id
);
1780 static void manager_catchup(Manager
*m
) {
1786 log_debug("Invoking unit catchup() handlers%s", special_glyph(SPECIAL_GLYPH_ELLIPSIS
));
1788 /* Let's catch up on any state changes that happened while we were reloading/reexecing */
1789 HASHMAP_FOREACH_KEY(u
, k
, m
->units
) {
1791 /* ignore aliases */
1799 static void manager_distribute_fds(Manager
*m
, FDSet
*fds
) {
1804 HASHMAP_FOREACH(u
, m
->units
) {
1806 if (fdset_isempty(fds
))
1809 if (!UNIT_VTABLE(u
)->distribute_fds
)
1812 UNIT_VTABLE(u
)->distribute_fds(u
, fds
);
1816 static bool manager_dbus_is_running(Manager
*m
, bool deserialized
) {
1821 /* This checks whether the dbus instance we are supposed to expose our APIs on is up. We check both the socket
1822 * and the service unit. If the 'deserialized' parameter is true we'll check the deserialized state of the unit
1823 * rather than the current one. */
1825 if (MANAGER_IS_TEST_RUN(m
))
1828 u
= manager_get_unit(m
, SPECIAL_DBUS_SOCKET
);
1831 if ((deserialized
? SOCKET(u
)->deserialized_state
: SOCKET(u
)->state
) != SOCKET_RUNNING
)
1834 u
= manager_get_unit(m
, SPECIAL_DBUS_SERVICE
);
1837 if (!IN_SET((deserialized
? SERVICE(u
)->deserialized_state
: SERVICE(u
)->state
),
1840 SERVICE_RELOAD_NOTIFY
,
1841 SERVICE_RELOAD_SIGNAL
))
1847 static void manager_setup_bus(Manager
*m
) {
1850 /* Let's set up our private bus connection now, unconditionally */
1851 (void) bus_init_private(m
);
1853 /* If we are in --user mode also connect to the system bus now */
1854 if (MANAGER_IS_USER(m
))
1855 (void) bus_init_system(m
);
1857 /* Let's connect to the bus now, but only if the unit is supposed to be up */
1858 if (manager_dbus_is_running(m
, MANAGER_IS_RELOADING(m
))) {
1859 (void) bus_init_api(m
);
1861 if (MANAGER_IS_SYSTEM(m
))
1862 (void) bus_init_system(m
);
1866 static void manager_preset_all(Manager
*m
) {
1871 if (m
->first_boot
<= 0)
1874 if (!MANAGER_IS_SYSTEM(m
))
1877 if (MANAGER_IS_TEST_RUN(m
))
1880 /* If this is the first boot, and we are in the host system, then preset everything */
1881 UnitFilePresetMode mode
=
1882 ENABLE_FIRST_BOOT_FULL_PRESET
? UNIT_FILE_PRESET_FULL
: UNIT_FILE_PRESET_ENABLE_ONLY
;
1884 r
= unit_file_preset_all(RUNTIME_SCOPE_SYSTEM
, 0, NULL
, mode
, NULL
, 0);
1886 log_full_errno(r
== -EEXIST
? LOG_NOTICE
: LOG_WARNING
, r
,
1887 "Failed to populate /etc with preset unit settings, ignoring: %m");
1889 log_info("Populated /etc with preset unit settings.");
1892 static void manager_ready(Manager
*m
) {
1895 /* After having loaded everything, do the final round of catching up with what might have changed */
1897 m
->objective
= MANAGER_OK
; /* Tell everyone we are up now */
1899 /* It might be safe to log to the journal now and connect to dbus */
1900 manager_recheck_journal(m
);
1901 manager_recheck_dbus(m
);
1903 /* Let's finally catch up with any changes that took place while we were reloading/reexecing */
1906 /* Create a file which will indicate when the manager started loading units the last time. */
1907 if (MANAGER_IS_SYSTEM(m
))
1908 (void) touch_file("/run/systemd/systemd-units-load", false,
1909 m
->timestamps
[MANAGER_TIMESTAMP_UNITS_LOAD
].realtime
?: now(CLOCK_REALTIME
),
1910 UID_INVALID
, GID_INVALID
, 0444);
1913 Manager
* manager_reloading_start(Manager
*m
) {
1915 dual_timestamp_now(m
->timestamps
+ MANAGER_TIMESTAMP_UNITS_LOAD
);
1919 void manager_reloading_stopp(Manager
**m
) {
1921 assert((*m
)->n_reloading
> 0);
1922 (*m
)->n_reloading
--;
1926 int manager_startup(Manager
*m
, FILE *serialization
, FDSet
*fds
, const char *root
) {
1931 /* If we are running in test mode, we still want to run the generators,
1932 * but we should not touch the real generator directories. */
1933 r
= lookup_paths_init_or_warn(&m
->lookup_paths
, m
->runtime_scope
,
1934 MANAGER_IS_TEST_RUN(m
) ? LOOKUP_PATHS_TEMPORARY_GENERATED
: 0,
1939 dual_timestamp_now(m
->timestamps
+ manager_timestamp_initrd_mangle(MANAGER_TIMESTAMP_GENERATORS_START
));
1940 r
= manager_run_environment_generators(m
);
1942 r
= manager_run_generators(m
);
1943 dual_timestamp_now(m
->timestamps
+ manager_timestamp_initrd_mangle(MANAGER_TIMESTAMP_GENERATORS_FINISH
));
1947 manager_preset_all(m
);
1949 lookup_paths_log(&m
->lookup_paths
);
1952 /* This block is (optionally) done with the reloading counter bumped */
1953 _unused_
_cleanup_(manager_reloading_stopp
) Manager
*reloading
= NULL
;
1955 /* Make sure we don't have a left-over from a previous run */
1957 (void) rm_rf(m
->lookup_paths
.transient
, 0);
1959 /* If we will deserialize make sure that during enumeration this is already known, so we increase the
1960 * counter here already */
1962 reloading
= manager_reloading_start(m
);
1964 /* First, enumerate what we can from all config files */
1965 dual_timestamp_now(m
->timestamps
+ manager_timestamp_initrd_mangle(MANAGER_TIMESTAMP_UNITS_LOAD_START
));
1966 manager_enumerate_perpetual(m
);
1967 manager_enumerate(m
);
1968 dual_timestamp_now(m
->timestamps
+ manager_timestamp_initrd_mangle(MANAGER_TIMESTAMP_UNITS_LOAD_FINISH
));
1970 /* Second, deserialize if there is something to deserialize */
1971 if (serialization
) {
1972 r
= manager_deserialize(m
, serialization
, fds
);
1974 return log_error_errno(r
, "Deserialization failed: %m");
1977 /* If we are in a new soft-reboot iteration bump the counter now before starting units, so
1978 * that they can reliably read it. We get the previous objective from serialized state. */
1979 if (m
->previous_objective
== MANAGER_SOFT_REBOOT
)
1980 m
->soft_reboots_count
++;
1982 /* Any fds left? Find some unit which wants them. This is useful to allow container managers to pass
1983 * some file descriptors to us pre-initialized. This enables socket-based activation of entire
1985 manager_distribute_fds(m
, fds
);
1987 /* We might have deserialized the notify fd, but if we didn't then let's create the bus now */
1988 r
= manager_setup_notify(m
);
1990 /* No sense to continue without notifications, our children would fail anyway. */
1993 r
= manager_setup_cgroups_agent(m
);
1995 /* Likewise, no sense to continue without empty cgroup notifications. */
1998 r
= manager_setup_user_lookup_fd(m
);
2000 /* This shouldn't fail, except if things are really broken. */
2003 /* Connect to the bus if we are good for it */
2004 manager_setup_bus(m
);
2006 /* Now that we are connected to all possible buses, let's deserialize who is tracking us. */
2007 r
= bus_track_coldplug(m
, &m
->subscribed
, false, m
->deserialized_subscribed
);
2009 log_warning_errno(r
, "Failed to deserialized tracked clients, ignoring: %m");
2010 m
->deserialized_subscribed
= strv_free(m
->deserialized_subscribed
);
2012 r
= manager_varlink_init(m
);
2014 log_warning_errno(r
, "Failed to set up Varlink, ignoring: %m");
2016 /* Third, fire things up! */
2017 manager_coldplug(m
);
2019 /* Clean up runtime objects */
2023 /* Let's wait for the UnitNew/JobNew messages being sent, before we notify that the
2024 * reload is finished */
2025 m
->send_reloading_done
= true;
2030 manager_set_switching_root(m
, false);
2035 int manager_add_job(
2041 sd_bus_error
*error
,
2044 _cleanup_(transaction_abort_and_freep
) Transaction
*tr
= NULL
;
2048 assert(type
< _JOB_TYPE_MAX
);
2050 assert(mode
< _JOB_MODE_MAX
);
2052 if (mode
== JOB_ISOLATE
&& type
!= JOB_START
)
2053 return sd_bus_error_set(error
, SD_BUS_ERROR_INVALID_ARGS
, "Isolate is only valid for start.");
2055 if (mode
== JOB_ISOLATE
&& !unit
->allow_isolate
)
2056 return sd_bus_error_set(error
, BUS_ERROR_NO_ISOLATION
, "Operation refused, unit may not be isolated.");
2058 if (mode
== JOB_TRIGGERING
&& type
!= JOB_STOP
)
2059 return sd_bus_error_set(error
, SD_BUS_ERROR_INVALID_ARGS
, "--job-mode=triggering is only valid for stop.");
2061 if (mode
== JOB_RESTART_DEPENDENCIES
&& type
!= JOB_START
)
2062 return sd_bus_error_set(error
, SD_BUS_ERROR_INVALID_ARGS
, "--job-mode=restart-dependencies is only valid for start.");
2064 log_unit_debug(unit
, "Trying to enqueue job %s/%s/%s", unit
->id
, job_type_to_string(type
), job_mode_to_string(mode
));
2066 type
= job_type_collapse(type
, unit
);
2068 tr
= transaction_new(mode
== JOB_REPLACE_IRREVERSIBLY
);
2072 r
= transaction_add_job_and_dependencies(
2077 TRANSACTION_MATTERS
|
2078 (IN_SET(mode
, JOB_IGNORE_DEPENDENCIES
, JOB_IGNORE_REQUIREMENTS
) ? TRANSACTION_IGNORE_REQUIREMENTS
: 0) |
2079 (mode
== JOB_IGNORE_DEPENDENCIES
? TRANSACTION_IGNORE_ORDER
: 0) |
2080 (mode
== JOB_RESTART_DEPENDENCIES
? TRANSACTION_PROPAGATE_START_AS_RESTART
: 0),
2085 if (mode
== JOB_ISOLATE
) {
2086 r
= transaction_add_isolate_jobs(tr
, m
);
2091 if (mode
== JOB_TRIGGERING
) {
2092 r
= transaction_add_triggering_jobs(tr
, unit
);
2097 r
= transaction_activate(tr
, m
, mode
, affected_jobs
, error
);
2101 log_unit_debug(unit
,
2102 "Enqueued job %s/%s as %u", unit
->id
,
2103 job_type_to_string(type
), (unsigned) tr
->anchor_job
->id
);
2106 *ret
= tr
->anchor_job
;
2108 tr
= transaction_free(tr
);
2112 int manager_add_job_by_name(Manager
*m
, JobType type
, const char *name
, JobMode mode
, Set
*affected_jobs
, sd_bus_error
*e
, Job
**ret
) {
2113 Unit
*unit
= NULL
; /* just to appease gcc, initialization is not really necessary */
2117 assert(type
< _JOB_TYPE_MAX
);
2119 assert(mode
< _JOB_MODE_MAX
);
2121 r
= manager_load_unit(m
, name
, NULL
, NULL
, &unit
);
2126 return manager_add_job(m
, type
, unit
, mode
, affected_jobs
, e
, ret
);
2129 int manager_add_job_by_name_and_warn(Manager
*m
, JobType type
, const char *name
, JobMode mode
, Set
*affected_jobs
, Job
**ret
) {
2130 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
2134 assert(type
< _JOB_TYPE_MAX
);
2136 assert(mode
< _JOB_MODE_MAX
);
2138 r
= manager_add_job_by_name(m
, type
, name
, mode
, affected_jobs
, &error
, ret
);
2140 return log_warning_errno(r
, "Failed to enqueue %s job for %s: %s", job_mode_to_string(mode
), name
, bus_error_message(&error
, r
));
2145 int manager_propagate_reload(Manager
*m
, Unit
*unit
, JobMode mode
, sd_bus_error
*e
) {
2147 _cleanup_(transaction_abort_and_freep
) Transaction
*tr
= NULL
;
2151 assert(mode
< _JOB_MODE_MAX
);
2152 assert(mode
!= JOB_ISOLATE
); /* Isolate is only valid for start */
2154 tr
= transaction_new(mode
== JOB_REPLACE_IRREVERSIBLY
);
2158 /* We need an anchor job */
2159 r
= transaction_add_job_and_dependencies(tr
, JOB_NOP
, unit
, NULL
, TRANSACTION_IGNORE_REQUIREMENTS
|TRANSACTION_IGNORE_ORDER
, e
);
2163 /* Failure in adding individual dependencies is ignored, so this always succeeds. */
2164 transaction_add_propagate_reload_jobs(
2168 mode
== JOB_IGNORE_DEPENDENCIES
? TRANSACTION_IGNORE_ORDER
: 0);
2170 r
= transaction_activate(tr
, m
, mode
, NULL
, e
);
2174 tr
= transaction_free(tr
);
2178 Job
*manager_get_job(Manager
*m
, uint32_t id
) {
2181 return hashmap_get(m
->jobs
, UINT32_TO_PTR(id
));
2184 Unit
*manager_get_unit(Manager
*m
, const char *name
) {
2188 return hashmap_get(m
->units
, name
);
2191 static int manager_dispatch_target_deps_queue(Manager
*m
) {
2197 while ((u
= LIST_POP(target_deps_queue
, m
->target_deps_queue
))) {
2198 _cleanup_free_ Unit
**targets
= NULL
;
2201 assert(u
->in_target_deps_queue
);
2203 u
->in_target_deps_queue
= false;
2205 /* Take an "atomic" snapshot of dependencies here, as the call below will likely modify the
2206 * dependencies, and we can't have it that hash tables we iterate through are modified while
2207 * we are iterating through them. */
2208 n_targets
= unit_get_dependency_array(u
, UNIT_ATOM_DEFAULT_TARGET_DEPENDENCIES
, &targets
);
2212 FOREACH_ARRAY(i
, targets
, n_targets
) {
2213 r
= unit_add_default_target_dependency(u
, *i
);
2222 unsigned manager_dispatch_load_queue(Manager
*m
) {
2228 /* Make sure we are not run recursively */
2229 if (m
->dispatching_load_queue
)
2232 m
->dispatching_load_queue
= true;
2234 /* Dispatches the load queue. Takes a unit from the queue and
2235 * tries to load its data until the queue is empty */
2237 while ((u
= m
->load_queue
)) {
2238 assert(u
->in_load_queue
);
2244 m
->dispatching_load_queue
= false;
2246 /* Dispatch the units waiting for their target dependencies to be added now, as all targets that we know about
2247 * should be loaded and have aliases resolved */
2248 (void) manager_dispatch_target_deps_queue(m
);
2253 bool manager_unit_cache_should_retry_load(Unit
*u
) {
2256 /* Automatic reloading from disk only applies to units which were not found sometime in the past, and
2257 * the not-found stub is kept pinned in the unit graph by dependencies. For units that were
2258 * previously loaded, we don't do automatic reloading, and daemon-reload is necessary to update. */
2259 if (u
->load_state
!= UNIT_NOT_FOUND
)
2262 /* The cache has been updated since the last time we tried to load the unit. There might be new
2263 * fragment paths to read. */
2264 if (u
->manager
->unit_cache_timestamp_hash
!= u
->fragment_not_found_timestamp_hash
)
2267 /* The cache needs to be updated because there are modifications on disk. */
2268 return !lookup_paths_timestamp_hash_same(&u
->manager
->lookup_paths
, u
->manager
->unit_cache_timestamp_hash
, NULL
);
2271 int manager_load_unit_prepare(
2278 _cleanup_(unit_freep
) Unit
*cleanup_unit
= NULL
;
2279 _cleanup_free_
char *nbuf
= NULL
;
2284 assert(name
|| path
);
2286 /* This will prepare the unit for loading, but not actually load anything from disk. */
2288 if (path
&& !path_is_absolute(path
))
2289 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
, "Path %s is not absolute.", path
);
2292 r
= path_extract_filename(path
, &nbuf
);
2295 if (r
== O_DIRECTORY
)
2296 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
, "Path '%s' refers to directory, refusing.", path
);
2301 UnitType t
= unit_name_to_type(name
);
2303 if (t
== _UNIT_TYPE_INVALID
|| !unit_name_is_valid(name
, UNIT_NAME_PLAIN
|UNIT_NAME_INSTANCE
)) {
2304 if (unit_name_is_valid(name
, UNIT_NAME_TEMPLATE
))
2305 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
, "Unit name %s is missing the instance name.", name
);
2307 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
, "Unit name %s is not valid.", name
);
2310 Unit
*unit
= manager_get_unit(m
, name
);
2312 /* The time-based cache allows to start new units without daemon-reload,
2313 * but if they are already referenced (because of dependencies or ordering)
2314 * then we have to force a load of the fragment. As an optimization, check
2315 * first if anything in the usual paths was modified since the last time
2316 * the cache was loaded. Also check if the last time an attempt to load the
2317 * unit was made was before the most recent cache refresh, so that we know
2318 * we need to try again — even if the cache is current, it might have been
2319 * updated in a different context before we had a chance to retry loading
2320 * this particular unit. */
2321 if (manager_unit_cache_should_retry_load(unit
))
2322 unit
->load_state
= UNIT_STUB
;
2325 return 0; /* The unit was already loaded */
2328 unit
= cleanup_unit
= unit_new(m
, unit_vtable
[t
]->object_size
);
2334 r
= free_and_strdup(&unit
->fragment_path
, path
);
2339 r
= unit_add_name(unit
, name
);
2343 unit_add_to_load_queue(unit
);
2344 unit_add_to_dbus_queue(unit
);
2345 unit_add_to_gc_queue(unit
);
2348 TAKE_PTR(cleanup_unit
);
2350 return 1; /* The unit was added the load queue */
2353 int manager_load_unit(
2364 /* This will load the unit config, but not actually start any services or anything. */
2366 r
= manager_load_unit_prepare(m
, name
, path
, e
, ret
);
2370 /* Unit was newly loaded */
2371 manager_dispatch_load_queue(m
);
2372 *ret
= unit_follow_merge(*ret
);
2376 int manager_load_startable_unit_or_warn(
2382 /* Load a unit, make sure it loaded fully and is not masked. */
2384 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
2388 r
= manager_load_unit(m
, name
, path
, &error
, &unit
);
2390 return log_error_errno(r
, "Failed to load %s %s: %s",
2391 name
? "unit" : "unit file", name
?: path
,
2392 bus_error_message(&error
, r
));
2394 r
= bus_unit_validate_load_state(unit
, &error
);
2396 return log_error_errno(r
, "%s", bus_error_message(&error
, r
));
2402 void manager_clear_jobs(Manager
*m
) {
2407 while ((j
= hashmap_first(m
->jobs
)))
2408 /* No need to recurse. We're cancelling all jobs. */
2409 job_finish_and_invalidate(j
, JOB_CANCELED
, false, false);
2412 void manager_unwatch_pidref(Manager
*m
, const PidRef
*pid
) {
2418 u
= manager_get_unit_by_pidref_watching(m
, pid
);
2422 unit_unwatch_pidref(u
, pid
);
2426 static int manager_dispatch_run_queue(sd_event_source
*source
, void *userdata
) {
2427 Manager
*m
= ASSERT_PTR(userdata
);
2432 while ((j
= prioq_peek(m
->run_queue
))) {
2433 assert(j
->installed
);
2434 assert(j
->in_run_queue
);
2436 (void) job_run_and_invalidate(j
);
2439 if (m
->n_running_jobs
> 0)
2440 manager_watch_jobs_in_progress(m
);
2442 if (m
->n_on_console
> 0)
2443 manager_watch_idle_pipe(m
);
2448 void manager_trigger_run_queue(Manager
*m
) {
2453 r
= sd_event_source_set_enabled(
2454 m
->run_queue_event_source
,
2455 prioq_isempty(m
->run_queue
) ? SD_EVENT_OFF
: SD_EVENT_ONESHOT
);
2457 log_warning_errno(r
, "Failed to enable job run queue event source, ignoring: %m");
2460 static unsigned manager_dispatch_dbus_queue(Manager
*m
) {
2461 unsigned n
= 0, budget
;
2467 /* When we are reloading, let's not wait with generating signals, since we need to exit the manager as quickly
2468 * as we can. There's no point in throttling generation of signals in that case. */
2469 if (MANAGER_IS_RELOADING(m
) || m
->send_reloading_done
|| m
->pending_reload_message
)
2470 budget
= UINT_MAX
; /* infinite budget in this case */
2472 /* Anything to do at all? */
2473 if (!m
->dbus_unit_queue
&& !m
->dbus_job_queue
)
2476 /* Do we have overly many messages queued at the moment? If so, let's not enqueue more on top, let's
2477 * sit this cycle out, and process things in a later cycle when the queues got a bit emptier. */
2478 if (manager_bus_n_queued_write(m
) > MANAGER_BUS_BUSY_THRESHOLD
)
2481 /* Only process a certain number of units/jobs per event loop iteration. Even if the bus queue wasn't
2482 * overly full before this call we shouldn't increase it in size too wildly in one step, and we
2483 * shouldn't monopolize CPU time with generating these messages. Note the difference in counting of
2484 * this "budget" and the "threshold" above: the "budget" is decreased only once per generated message,
2485 * regardless how many buses/direct connections it is enqueued on, while the "threshold" is applied to
2486 * each queued instance of bus message, i.e. if the same message is enqueued to five buses/direct
2487 * connections it will be counted five times. This difference in counting ("references"
2488 * vs. "instances") is primarily a result of the fact that it's easier to implement it this way,
2489 * however it also reflects the thinking that the "threshold" should put a limit on used queue memory,
2490 * i.e. space, while the "budget" should put a limit on time. Also note that the "threshold" is
2491 * currently chosen much higher than the "budget". */
2492 budget
= MANAGER_BUS_MESSAGE_BUDGET
;
2495 while (budget
!= 0 && (u
= m
->dbus_unit_queue
)) {
2497 assert(u
->in_dbus_queue
);
2499 bus_unit_send_change_signal(u
);
2502 if (budget
!= UINT_MAX
)
2506 while (budget
!= 0 && (j
= m
->dbus_job_queue
)) {
2507 assert(j
->in_dbus_queue
);
2509 bus_job_send_change_signal(j
);
2512 if (budget
!= UINT_MAX
)
2516 if (m
->send_reloading_done
) {
2517 m
->send_reloading_done
= false;
2518 bus_manager_send_reloading(m
, false);
2522 if (m
->pending_reload_message
) {
2523 bus_send_pending_reload_message(m
);
2530 static int manager_dispatch_cgroups_agent_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
2531 Manager
*m
= userdata
;
2535 n
= recv(fd
, buf
, sizeof(buf
), 0);
2537 return log_error_errno(errno
, "Failed to read cgroups agent message: %m");
2539 log_error("Got zero-length cgroups agent message, ignoring.");
2542 if ((size_t) n
>= sizeof(buf
)) {
2543 log_error("Got overly long cgroups agent message, ignoring.");
2547 if (memchr(buf
, 0, n
)) {
2548 log_error("Got cgroups agent message with embedded NUL byte, ignoring.");
2553 manager_notify_cgroup_empty(m
, buf
);
2554 (void) bus_forward_agent_released(m
, buf
);
2559 static bool manager_process_barrier_fd(char * const *tags
, FDSet
*fds
) {
2561 /* nothing else must be sent when using BARRIER=1 */
2562 if (strv_contains(tags
, "BARRIER=1")) {
2563 if (strv_length(tags
) != 1)
2564 log_warning("Extra notification messages sent with BARRIER=1, ignoring everything.");
2565 else if (fdset_size(fds
) != 1)
2566 log_warning("Got incorrect number of fds with BARRIER=1, closing them.");
2568 /* Drop the message if BARRIER=1 was found */
2575 static void manager_invoke_notify_message(
2578 const struct ucred
*ucred
,
2587 if (u
->notifygen
== m
->notifygen
) /* Already invoked on this same unit in this same iteration? */
2589 u
->notifygen
= m
->notifygen
;
2591 if (UNIT_VTABLE(u
)->notify_message
)
2592 UNIT_VTABLE(u
)->notify_message(u
, ucred
, tags
, fds
);
2594 else if (DEBUG_LOGGING
) {
2595 _cleanup_free_
char *joined
= strv_join(tags
, ", ");
2596 char buf
[CELLESCAPE_DEFAULT_LENGTH
];
2598 log_unit_debug(u
, "Got notification message from unexpected unit type, ignoring: %s",
2599 joined
? cellescape(buf
, sizeof(buf
), joined
) : "(null)");
2603 static int manager_dispatch_notify_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
2604 Manager
*m
= ASSERT_PTR(userdata
);
2605 _cleanup_fdset_free_ FDSet
*fds
= NULL
;
2606 char buf
[NOTIFY_BUFFER_MAX
+1];
2607 struct iovec iovec
= {
2609 .iov_len
= sizeof(buf
)-1,
2611 CMSG_BUFFER_TYPE(CMSG_SPACE(sizeof(struct ucred
)) +
2612 CMSG_SPACE(sizeof(int) * NOTIFY_FD_MAX
)) control
;
2613 struct msghdr msghdr
= {
2616 .msg_control
= &control
,
2617 .msg_controllen
= sizeof(control
),
2620 struct cmsghdr
*cmsg
;
2621 struct ucred
*ucred
= NULL
;
2622 _cleanup_free_ Unit
**array_copy
= NULL
;
2623 _cleanup_strv_free_
char **tags
= NULL
;
2624 Unit
*u1
, *u2
, **array
;
2625 int r
, *fd_array
= NULL
;
2630 assert(m
->notify_fd
== fd
);
2632 if (revents
!= EPOLLIN
) {
2633 log_warning("Got unexpected poll event for notify fd.");
2637 n
= recvmsg_safe(m
->notify_fd
, &msghdr
, MSG_DONTWAIT
|MSG_CMSG_CLOEXEC
|MSG_TRUNC
);
2638 if (ERRNO_IS_NEG_TRANSIENT(n
))
2639 return 0; /* Spurious wakeup, try again */
2641 log_warning("Got message with truncated control data (too many fds sent?), ignoring.");
2645 /* If this is any other, real error, then stop processing this socket. This of course means
2646 * we won't take notification messages anymore, but that's still better than busy looping:
2647 * being woken up over and over again, but being unable to actually read the message from the
2649 return log_error_errno(n
, "Failed to receive notification message: %m");
2651 CMSG_FOREACH(cmsg
, &msghdr
)
2652 if (cmsg
->cmsg_level
== SOL_SOCKET
&& cmsg
->cmsg_type
== SCM_RIGHTS
) {
2655 fd_array
= CMSG_TYPED_DATA(cmsg
, int);
2656 n_fds
= (cmsg
->cmsg_len
- CMSG_LEN(0)) / sizeof(int);
2658 } else if (cmsg
->cmsg_level
== SOL_SOCKET
&&
2659 cmsg
->cmsg_type
== SCM_CREDENTIALS
&&
2660 cmsg
->cmsg_len
== CMSG_LEN(sizeof(struct ucred
))) {
2663 ucred
= CMSG_TYPED_DATA(cmsg
, struct ucred
);
2669 r
= fdset_new_array(&fds
, fd_array
, n_fds
);
2671 close_many(fd_array
, n_fds
);
2677 if (!ucred
|| !pid_is_valid(ucred
->pid
)) {
2678 log_warning("Received notify message without valid credentials. Ignoring.");
2682 if ((size_t) n
>= sizeof(buf
) || (msghdr
.msg_flags
& MSG_TRUNC
)) {
2683 log_warning("Received notify message exceeded maximum size. Ignoring.");
2687 /* As extra safety check, let's make sure the string we get doesn't contain embedded NUL bytes.
2688 * We permit one trailing NUL byte in the message, but don't expect it. */
2689 if (n
> 1 && memchr(buf
, 0, n
-1)) {
2690 log_warning("Received notify message with embedded NUL bytes. Ignoring.");
2694 /* Make sure it's NUL-terminated, then parse it to obtain the tags list. */
2696 tags
= strv_split_newlines(buf
);
2702 /* Possibly a barrier fd, let's see. */
2703 if (manager_process_barrier_fd(tags
, fds
)) {
2704 log_debug("Received barrier notification message from PID " PID_FMT
".", ucred
->pid
);
2708 /* Increase the generation counter used for filtering out duplicate unit invocations. */
2711 /* Generate lookup key from the PID (we have no pidfd here, after all) */
2712 PidRef pidref
= PIDREF_MAKE_FROM_PID(ucred
->pid
);
2714 /* Notify every unit that might be interested, which might be multiple. */
2715 u1
= manager_get_unit_by_pidref_cgroup(m
, &pidref
);
2716 u2
= hashmap_get(m
->watch_pids
, &pidref
);
2717 array
= hashmap_get(m
->watch_pids_more
, &pidref
);
2724 array_copy
= newdup(Unit
*, array
, k
+1);
2728 /* And now invoke the per-unit callbacks. Note that manager_invoke_notify_message() will handle
2729 * duplicate units make sure we only invoke each unit's handler once. */
2731 manager_invoke_notify_message(m
, u1
, ucred
, tags
, fds
);
2735 manager_invoke_notify_message(m
, u2
, ucred
, tags
, fds
);
2739 for (size_t i
= 0; array_copy
[i
]; i
++) {
2740 manager_invoke_notify_message(m
, array_copy
[i
], ucred
, tags
, fds
);
2745 log_warning("Cannot find unit for notify message of PID "PID_FMT
", ignoring.", ucred
->pid
);
2747 if (!fdset_isempty(fds
))
2748 log_warning("Got extra auxiliary fds with notification message, closing them.");
2753 static void manager_invoke_sigchld_event(
2756 const siginfo_t
*si
) {
2762 /* Already invoked the handler of this unit in this iteration? Then don't process this again */
2763 if (u
->sigchldgen
== m
->sigchldgen
)
2765 u
->sigchldgen
= m
->sigchldgen
;
2767 log_unit_debug(u
, "Child "PID_FMT
" belongs to %s.", si
->si_pid
, u
->id
);
2768 unit_unwatch_pid(u
, si
->si_pid
);
2770 if (UNIT_VTABLE(u
)->sigchld_event
)
2771 UNIT_VTABLE(u
)->sigchld_event(u
, si
->si_pid
, si
->si_code
, si
->si_status
);
2774 static int manager_dispatch_sigchld(sd_event_source
*source
, void *userdata
) {
2775 Manager
*m
= ASSERT_PTR(userdata
);
2781 /* First we call waitid() for a PID and do not reap the zombie. That way we can still access
2782 * /proc/$PID for it while it is a zombie. */
2784 if (waitid(P_ALL
, 0, &si
, WEXITED
|WNOHANG
|WNOWAIT
) < 0) {
2786 if (errno
!= ECHILD
)
2787 log_error_errno(errno
, "Failed to peek for child with waitid(), ignoring: %m");
2795 if (IN_SET(si
.si_code
, CLD_EXITED
, CLD_KILLED
, CLD_DUMPED
)) {
2796 _cleanup_free_ Unit
**array_copy
= NULL
;
2797 _cleanup_free_
char *name
= NULL
;
2798 Unit
*u1
, *u2
, **array
;
2800 (void) pid_get_comm(si
.si_pid
, &name
);
2802 log_debug("Child "PID_FMT
" (%s) died (code=%s, status=%i/%s)",
2803 si
.si_pid
, strna(name
),
2804 sigchld_code_to_string(si
.si_code
),
2806 strna(si
.si_code
== CLD_EXITED
2807 ? exit_status_to_string(si
.si_status
, EXIT_STATUS_FULL
)
2808 : signal_to_string(si
.si_status
)));
2810 /* Increase the generation counter used for filtering out duplicate unit invocations */
2813 /* We look this up by a PidRef that only consists of the PID. After all we couldn't create a
2814 * pidfd here any more even if we wanted (since the process just exited). */
2815 PidRef pidref
= PIDREF_MAKE_FROM_PID(si
.si_pid
);
2817 /* And now figure out the unit this belongs to, it might be multiple... */
2818 u1
= manager_get_unit_by_pidref_cgroup(m
, &pidref
);
2819 u2
= hashmap_get(m
->watch_pids
, &pidref
);
2820 array
= hashmap_get(m
->watch_pids_more
, &pidref
);
2824 /* Count how many entries the array has */
2828 /* Make a copy of the array so that we don't trip up on the array changing beneath us */
2829 array_copy
= newdup(Unit
*, array
, n
+1);
2834 /* Finally, execute them all. Note that u1, u2 and the array might contain duplicates, but
2835 * that's fine, manager_invoke_sigchld_event() will ensure we only invoke the handlers once for
2836 * each iteration. */
2838 /* We check for oom condition, in case we got SIGCHLD before the oom notification.
2839 * We only do this for the cgroup the PID belonged to. */
2840 (void) unit_check_oom(u1
);
2842 /* We check if systemd-oomd performed a kill so that we log and notify appropriately */
2843 (void) unit_check_oomd_kill(u1
);
2845 manager_invoke_sigchld_event(m
, u1
, &si
);
2848 manager_invoke_sigchld_event(m
, u2
, &si
);
2850 for (size_t i
= 0; array_copy
[i
]; i
++)
2851 manager_invoke_sigchld_event(m
, array_copy
[i
], &si
);
2854 /* And now, we actually reap the zombie. */
2855 if (waitid(P_PID
, si
.si_pid
, &si
, WEXITED
) < 0) {
2856 log_error_errno(errno
, "Failed to dequeue child, ignoring: %m");
2863 /* All children processed for now, turn off event source */
2865 r
= sd_event_source_set_enabled(m
->sigchld_event_source
, SD_EVENT_OFF
);
2867 return log_error_errno(r
, "Failed to disable SIGCHLD event source: %m");
2872 static void manager_start_special(Manager
*m
, const char *name
, JobMode mode
) {
2875 if (manager_add_job_by_name_and_warn(m
, JOB_START
, name
, mode
, NULL
, &job
) < 0)
2878 const char *s
= unit_status_string(job
->unit
, NULL
);
2880 log_info("Activating special unit %s...", s
);
2882 (void) sd_notifyf(/* unset_environment= */ false,
2883 "STATUS=Activating special unit %s...", s
);
2884 m
->status_ready
= false;
2887 static void manager_handle_ctrl_alt_del(Manager
*m
) {
2888 /* If the user presses C-A-D more than
2889 * 7 times within 2s, we reboot/shutdown immediately,
2890 * unless it was disabled in system.conf */
2892 if (ratelimit_below(&m
->ctrl_alt_del_ratelimit
) || m
->cad_burst_action
== EMERGENCY_ACTION_NONE
)
2893 manager_start_special(m
, SPECIAL_CTRL_ALT_DEL_TARGET
, JOB_REPLACE_IRREVERSIBLY
);
2895 emergency_action(m
, m
->cad_burst_action
, EMERGENCY_ACTION_WARN
, NULL
, -1,
2896 "Ctrl-Alt-Del was pressed more than 7 times within 2s");
2899 static int manager_dispatch_signal_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
2900 Manager
*m
= ASSERT_PTR(userdata
);
2902 struct signalfd_siginfo sfsi
;
2905 assert(m
->signal_fd
== fd
);
2907 if (revents
!= EPOLLIN
) {
2908 log_warning("Got unexpected events from signal file descriptor.");
2912 n
= read(m
->signal_fd
, &sfsi
, sizeof(sfsi
));
2914 if (ERRNO_IS_TRANSIENT(errno
))
2917 /* We return an error here, which will kill this handler,
2918 * to avoid a busy loop on read error. */
2919 return log_error_errno(errno
, "Reading from signal fd failed: %m");
2921 if (n
!= sizeof(sfsi
)) {
2922 log_warning("Truncated read from signal fd (%zi bytes), ignoring!", n
);
2926 log_received_signal(sfsi
.ssi_signo
== SIGCHLD
||
2927 (sfsi
.ssi_signo
== SIGTERM
&& MANAGER_IS_USER(m
))
2928 ? LOG_DEBUG
: LOG_INFO
,
2931 switch (sfsi
.ssi_signo
) {
2934 r
= sd_event_source_set_enabled(m
->sigchld_event_source
, SD_EVENT_ON
);
2936 log_warning_errno(r
, "Failed to enable SIGCHLD event source, ignoring: %m");
2941 if (MANAGER_IS_SYSTEM(m
)) {
2942 /* This is for compatibility with the original sysvinit */
2943 if (verify_run_space_and_log("Refusing to reexecute") < 0)
2946 m
->objective
= MANAGER_REEXECUTE
;
2952 if (MANAGER_IS_SYSTEM(m
))
2953 manager_handle_ctrl_alt_del(m
);
2955 manager_start_special(m
, SPECIAL_EXIT_TARGET
, JOB_REPLACE_IRREVERSIBLY
);
2959 /* This is a nop on non-init */
2960 if (MANAGER_IS_SYSTEM(m
))
2961 manager_start_special(m
, SPECIAL_KBREQUEST_TARGET
, JOB_REPLACE
);
2966 /* This is a nop on non-init */
2967 if (MANAGER_IS_SYSTEM(m
))
2968 manager_start_special(m
, SPECIAL_SIGPWR_TARGET
, JOB_REPLACE
);
2973 if (manager_dbus_is_running(m
, false)) {
2974 log_info("Trying to reconnect to bus...");
2976 (void) bus_init_api(m
);
2978 if (MANAGER_IS_SYSTEM(m
))
2979 (void) bus_init_system(m
);
2981 manager_start_special(m
, SPECIAL_DBUS_SERVICE
, JOB_REPLACE
);
2986 _cleanup_free_
char *dump
= NULL
;
2988 r
= manager_get_dump_string(m
, /* patterns= */ NULL
, &dump
);
2990 log_warning_errno(errno
, "Failed to acquire manager dump: %m");
2994 log_dump(LOG_INFO
, dump
);
2999 if (verify_run_space_and_log("Refusing to reload") < 0)
3002 m
->objective
= MANAGER_RELOAD
;
3007 /* Starting SIGRTMIN+0 */
3008 static const struct {
3011 } target_table
[] = {
3012 [0] = { SPECIAL_DEFAULT_TARGET
, JOB_ISOLATE
},
3013 [1] = { SPECIAL_RESCUE_TARGET
, JOB_ISOLATE
},
3014 [2] = { SPECIAL_EMERGENCY_TARGET
, JOB_ISOLATE
},
3015 [3] = { SPECIAL_HALT_TARGET
, JOB_REPLACE_IRREVERSIBLY
},
3016 [4] = { SPECIAL_POWEROFF_TARGET
, JOB_REPLACE_IRREVERSIBLY
},
3017 [5] = { SPECIAL_REBOOT_TARGET
, JOB_REPLACE_IRREVERSIBLY
},
3018 [6] = { SPECIAL_KEXEC_TARGET
, JOB_REPLACE_IRREVERSIBLY
},
3019 [7] = { SPECIAL_SOFT_REBOOT_TARGET
, JOB_REPLACE_IRREVERSIBLY
},
3022 /* Starting SIGRTMIN+13, so that target halt and system halt are 10 apart */
3023 static const ManagerObjective objective_table
[] = {
3025 [1] = MANAGER_POWEROFF
,
3026 [2] = MANAGER_REBOOT
,
3027 [3] = MANAGER_KEXEC
,
3028 [4] = MANAGER_SOFT_REBOOT
,
3031 if ((int) sfsi
.ssi_signo
>= SIGRTMIN
+0 &&
3032 (int) sfsi
.ssi_signo
< SIGRTMIN
+(int) ELEMENTSOF(target_table
)) {
3033 int idx
= (int) sfsi
.ssi_signo
- SIGRTMIN
;
3034 manager_start_special(m
, target_table
[idx
].target
, target_table
[idx
].mode
);
3038 if ((int) sfsi
.ssi_signo
>= SIGRTMIN
+13 &&
3039 (int) sfsi
.ssi_signo
< SIGRTMIN
+13+(int) ELEMENTSOF(objective_table
)) {
3040 m
->objective
= objective_table
[sfsi
.ssi_signo
- SIGRTMIN
- 13];
3044 switch (sfsi
.ssi_signo
- SIGRTMIN
) {
3047 bool generic
= false;
3049 if (sfsi
.ssi_code
!= SI_QUEUE
)
3052 /* Override a few select commands by our own PID1-specific logic */
3054 switch (sfsi
.ssi_int
) {
3056 case _COMMON_SIGNAL_COMMAND_LOG_LEVEL_BASE
..._COMMON_SIGNAL_COMMAND_LOG_LEVEL_END
:
3057 manager_override_log_level(m
, sfsi
.ssi_int
- _COMMON_SIGNAL_COMMAND_LOG_LEVEL_BASE
);
3060 case COMMON_SIGNAL_COMMAND_CONSOLE
:
3061 manager_override_log_target(m
, LOG_TARGET_CONSOLE
);
3064 case COMMON_SIGNAL_COMMAND_JOURNAL
:
3065 manager_override_log_target(m
, LOG_TARGET_JOURNAL
);
3068 case COMMON_SIGNAL_COMMAND_KMSG
:
3069 manager_override_log_target(m
, LOG_TARGET_KMSG
);
3072 case COMMON_SIGNAL_COMMAND_NULL
:
3073 manager_override_log_target(m
, LOG_TARGET_NULL
);
3076 case MANAGER_SIGNAL_COMMAND_DUMP_JOBS
: {
3077 _cleanup_free_
char *dump_jobs
= NULL
;
3079 r
= manager_get_dump_jobs_string(m
, /* patterns= */ NULL
, " ", &dump_jobs
);
3081 log_warning_errno(errno
, "Failed to acquire manager jobs dump: %m");
3085 log_dump(LOG_INFO
, dump_jobs
);
3095 return sigrtmin18_handler(source
, &sfsi
, NULL
);
3101 manager_override_show_status(m
, SHOW_STATUS_YES
, "signal");
3105 manager_override_show_status(m
, SHOW_STATUS_NO
, "signal");
3109 manager_override_log_level(m
, LOG_DEBUG
);
3113 manager_restore_original_log_level(m
);
3117 if (MANAGER_IS_USER(m
)) {
3118 m
->objective
= MANAGER_EXIT
;
3122 /* This is a nop on init */
3126 m
->objective
= MANAGER_REEXECUTE
;
3130 case 29: /* compatibility: used to be mapped to LOG_TARGET_SYSLOG_OR_KMSG */
3131 manager_restore_original_log_target(m
);
3135 manager_override_log_target(m
, LOG_TARGET_CONSOLE
);
3139 manager_override_log_target(m
, LOG_TARGET_KMSG
);
3143 log_warning("Got unhandled signal <%s>.", signal_to_string(sfsi
.ssi_signo
));
3150 static int manager_dispatch_time_change_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
3151 Manager
*m
= ASSERT_PTR(userdata
);
3154 log_struct(LOG_DEBUG
,
3155 "MESSAGE_ID=" SD_MESSAGE_TIME_CHANGE_STR
,
3156 LOG_MESSAGE("Time has been changed"));
3158 /* Restart the watch */
3159 (void) manager_setup_time_change(m
);
3161 HASHMAP_FOREACH(u
, m
->units
)
3162 if (UNIT_VTABLE(u
)->time_change
)
3163 UNIT_VTABLE(u
)->time_change(u
);
3168 static int manager_dispatch_timezone_change(
3169 sd_event_source
*source
,
3170 const struct inotify_event
*e
,
3173 Manager
*m
= ASSERT_PTR(userdata
);
3177 log_debug("inotify event for /etc/localtime");
3179 changed
= manager_read_timezone_stat(m
);
3183 /* Something changed, restart the watch, to ensure we watch the new /etc/localtime if it changed */
3184 (void) manager_setup_timezone_change(m
);
3186 /* Read the new timezone */
3189 log_debug("Timezone has been changed (now: %s).", tzname
[daylight
]);
3191 HASHMAP_FOREACH(u
, m
->units
)
3192 if (UNIT_VTABLE(u
)->timezone_change
)
3193 UNIT_VTABLE(u
)->timezone_change(u
);
3198 static int manager_dispatch_idle_pipe_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
3199 Manager
*m
= ASSERT_PTR(userdata
);
3201 assert(m
->idle_pipe
[2] == fd
);
3203 /* There's at least one Type=idle child that just gave up on us waiting for the boot process to
3204 * complete. Let's now turn off any further console output if there's at least one service that needs
3205 * console access, so that from now on our own output should not spill into that service's output
3206 * anymore. After all, we support Type=idle only to beautify console output and it generally is set
3207 * on services that want to own the console exclusively without our interference. */
3208 m
->no_console_output
= m
->n_on_console
> 0;
3210 /* Acknowledge the child's request, and let all other children know too that they shouldn't wait
3211 * any longer by closing the pipes towards them, which is what they are waiting for. */
3212 manager_close_idle_pipe(m
);
3217 static int manager_dispatch_jobs_in_progress(sd_event_source
*source
, usec_t usec
, void *userdata
) {
3218 Manager
*m
= ASSERT_PTR(userdata
);
3223 manager_print_jobs_in_progress(m
);
3225 r
= sd_event_source_set_time_relative(source
, JOBS_IN_PROGRESS_PERIOD_USEC
);
3229 return sd_event_source_set_enabled(source
, SD_EVENT_ONESHOT
);
3232 int manager_loop(Manager
*m
) {
3233 RateLimit rl
= { .interval
= 1*USEC_PER_SEC
, .burst
= 50000 };
3237 assert(m
->objective
== MANAGER_OK
); /* Ensure manager_startup() has been called */
3239 manager_check_finished(m
);
3241 /* There might still be some zombies hanging around from before we were exec()'ed. Let's reap them. */
3242 r
= sd_event_source_set_enabled(m
->sigchld_event_source
, SD_EVENT_ON
);
3244 return log_error_errno(r
, "Failed to enable SIGCHLD event source: %m");
3246 while (m
->objective
== MANAGER_OK
) {
3248 (void) watchdog_ping();
3250 if (!ratelimit_below(&rl
)) {
3251 /* Yay, something is going seriously wrong, pause a little */
3252 log_warning("Looping too fast. Throttling execution a little.");
3256 if (manager_dispatch_load_queue(m
) > 0)
3259 if (manager_dispatch_gc_job_queue(m
) > 0)
3262 if (manager_dispatch_gc_unit_queue(m
) > 0)
3265 if (manager_dispatch_cleanup_queue(m
) > 0)
3268 if (manager_dispatch_cgroup_realize_queue(m
) > 0)
3271 if (manager_dispatch_start_when_upheld_queue(m
) > 0)
3274 if (manager_dispatch_stop_when_bound_queue(m
) > 0)
3277 if (manager_dispatch_stop_when_unneeded_queue(m
) > 0)
3280 if (manager_dispatch_release_resources_queue(m
) > 0)
3283 if (manager_dispatch_dbus_queue(m
) > 0)
3286 /* Sleep for watchdog runtime wait time */
3287 r
= sd_event_run(m
->event
, watchdog_runtime_wait());
3289 return log_error_errno(r
, "Failed to run event loop: %m");
3292 return m
->objective
;
3295 int manager_load_unit_from_dbus_path(Manager
*m
, const char *s
, sd_bus_error
*e
, Unit
**_u
) {
3296 _cleanup_free_
char *n
= NULL
;
3297 sd_id128_t invocation_id
;
3305 r
= unit_name_from_dbus_path(s
, &n
);
3309 /* Permit addressing units by invocation ID: if the passed bus path is suffixed by a 128-bit ID then
3310 * we use it as invocation ID. */
3311 r
= sd_id128_from_string(n
, &invocation_id
);
3313 u
= hashmap_get(m
->units_by_invocation_id
, &invocation_id
);
3319 return sd_bus_error_setf(e
, BUS_ERROR_NO_UNIT_FOR_INVOCATION_ID
,
3320 "No unit with the specified invocation ID " SD_ID128_FORMAT_STR
" known.",
3321 SD_ID128_FORMAT_VAL(invocation_id
));
3324 /* If this didn't work, we check if this is a unit name */
3325 if (!unit_name_is_valid(n
, UNIT_NAME_PLAIN
|UNIT_NAME_INSTANCE
)) {
3326 _cleanup_free_
char *nn
= NULL
;
3329 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
,
3330 "Unit name %s is neither a valid invocation ID nor unit name.", strnull(nn
));
3333 r
= manager_load_unit(m
, n
, NULL
, e
, &u
);
3341 int manager_get_job_from_dbus_path(Manager
*m
, const char *s
, Job
**_j
) {
3351 p
= startswith(s
, "/org/freedesktop/systemd1/job/");
3355 r
= safe_atou(p
, &id
);
3359 j
= manager_get_job(m
, id
);
3368 void manager_send_unit_audit(Manager
*m
, Unit
*u
, int type
, bool success
) {
3371 _cleanup_free_
char *p
= NULL
;
3378 if (!MANAGER_IS_SYSTEM(m
))
3381 /* Don't generate audit events if the service was already started and we're just deserializing */
3382 if (MANAGER_IS_RELOADING(m
))
3385 audit_fd
= get_audit_fd();
3389 r
= unit_name_to_prefix_and_instance(u
->id
, &p
);
3391 log_warning_errno(r
, "Failed to extract prefix and instance of unit name, ignoring: %m");
3395 msg
= strjoina("unit=", p
);
3396 if (audit_log_user_comm_message(audit_fd
, type
, msg
, "systemd", NULL
, NULL
, NULL
, success
) < 0) {
3397 if (ERRNO_IS_PRIVILEGE(errno
)) {
3398 /* We aren't allowed to send audit messages? Then let's not retry again. */
3399 log_debug_errno(errno
, "Failed to send audit message, closing audit socket: %m");
3402 log_warning_errno(errno
, "Failed to send audit message, ignoring: %m");
3407 void manager_send_unit_plymouth(Manager
*m
, Unit
*u
) {
3408 _cleanup_free_
char *message
= NULL
;
3414 if (!MANAGER_IS_SYSTEM(m
))
3417 /* Don't generate plymouth events if the service was already started and we're just deserializing */
3418 if (MANAGER_IS_RELOADING(m
))
3421 if (detect_container() > 0)
3424 if (!UNIT_VTABLE(u
)->notify_plymouth
)
3427 c
= asprintf(&message
, "U\x02%c%s%c", (int) (strlen(u
->id
) + 1), u
->id
, '\x00');
3429 return (void) log_oom();
3431 /* We set SOCK_NONBLOCK here so that we rather drop the message then wait for plymouth */
3432 r
= plymouth_send_raw(message
, c
, SOCK_NONBLOCK
);
3434 log_full_errno(ERRNO_IS_NO_PLYMOUTH(r
) ? LOG_DEBUG
: LOG_WARNING
, r
,
3435 "Failed to communicate with plymouth: %m");
3438 void manager_send_unit_supervisor(Manager
*m
, Unit
*u
, bool active
) {
3442 /* Notify a "supervisor" process about our progress, i.e. a container manager, hypervisor, or
3443 * surrounding service manager. */
3445 if (MANAGER_IS_RELOADING(m
))
3448 if (!UNIT_VTABLE(u
)->notify_supervisor
)
3451 if (in_initrd()) /* Only send these once we left the initrd */
3454 (void) sd_notifyf(/* unset_environment= */ false,
3455 active
? "X_SYSTEMD_UNIT_ACTIVE=%s" : "X_SYSTEMD_UNIT_INACTIVE=%s",
3459 usec_t
manager_get_watchdog(Manager
*m
, WatchdogType t
) {
3462 if (MANAGER_IS_USER(m
))
3463 return USEC_INFINITY
;
3465 if (m
->watchdog_overridden
[t
] != USEC_INFINITY
)
3466 return m
->watchdog_overridden
[t
];
3468 return m
->watchdog
[t
];
3471 void manager_set_watchdog(Manager
*m
, WatchdogType t
, usec_t timeout
) {
3475 if (MANAGER_IS_USER(m
))
3478 if (m
->watchdog
[t
] == timeout
)
3481 if (m
->watchdog_overridden
[t
] == USEC_INFINITY
) {
3482 if (t
== WATCHDOG_RUNTIME
)
3483 (void) watchdog_setup(timeout
);
3484 else if (t
== WATCHDOG_PRETIMEOUT
)
3485 (void) watchdog_setup_pretimeout(timeout
);
3488 m
->watchdog
[t
] = timeout
;
3491 void manager_override_watchdog(Manager
*m
, WatchdogType t
, usec_t timeout
) {
3496 if (MANAGER_IS_USER(m
))
3499 if (m
->watchdog_overridden
[t
] == timeout
)
3502 usec
= timeout
== USEC_INFINITY
? m
->watchdog
[t
] : timeout
;
3503 if (t
== WATCHDOG_RUNTIME
)
3504 (void) watchdog_setup(usec
);
3505 else if (t
== WATCHDOG_PRETIMEOUT
)
3506 (void) watchdog_setup_pretimeout(usec
);
3508 m
->watchdog_overridden
[t
] = timeout
;
3511 int manager_set_watchdog_pretimeout_governor(Manager
*m
, const char *governor
) {
3512 _cleanup_free_
char *p
= NULL
;
3517 if (MANAGER_IS_USER(m
))
3520 if (streq_ptr(m
->watchdog_pretimeout_governor
, governor
))
3523 p
= strdup(governor
);
3527 r
= watchdog_setup_pretimeout_governor(governor
);
3531 return free_and_replace(m
->watchdog_pretimeout_governor
, p
);
3534 int manager_override_watchdog_pretimeout_governor(Manager
*m
, const char *governor
) {
3535 _cleanup_free_
char *p
= NULL
;
3540 if (MANAGER_IS_USER(m
))
3543 if (streq_ptr(m
->watchdog_pretimeout_governor_overridden
, governor
))
3546 p
= strdup(governor
);
3550 r
= watchdog_setup_pretimeout_governor(governor
);
3554 return free_and_replace(m
->watchdog_pretimeout_governor_overridden
, p
);
3557 int manager_reload(Manager
*m
) {
3558 _unused_
_cleanup_(manager_reloading_stopp
) Manager
*reloading
= NULL
;
3559 _cleanup_fdset_free_ FDSet
*fds
= NULL
;
3560 _cleanup_fclose_
FILE *f
= NULL
;
3565 r
= manager_open_serialization(m
, &f
);
3567 return log_error_errno(r
, "Failed to create serialization file: %m");
3573 /* We are officially in reload mode from here on. */
3574 reloading
= manager_reloading_start(m
);
3576 r
= manager_serialize(m
, f
, fds
, false);
3580 if (fseeko(f
, 0, SEEK_SET
) < 0)
3581 return log_error_errno(errno
, "Failed to seek to beginning of serialization: %m");
3583 /* 💀 This is the point of no return, from here on there is no way back. 💀 */
3586 bus_manager_send_reloading(m
, true);
3588 /* Start by flushing out all jobs and units, all generated units, all runtime environments, all dynamic users
3589 * and everything else that is worth flushing out. We'll get it all back from the serialization — if we need
3592 manager_clear_jobs_and_units(m
);
3593 lookup_paths_flush_generator(&m
->lookup_paths
);
3594 lookup_paths_done(&m
->lookup_paths
);
3595 exec_shared_runtime_vacuum(m
);
3596 dynamic_user_vacuum(m
, false);
3597 m
->uid_refs
= hashmap_free(m
->uid_refs
);
3598 m
->gid_refs
= hashmap_free(m
->gid_refs
);
3600 r
= lookup_paths_init_or_warn(&m
->lookup_paths
, m
->runtime_scope
, 0, NULL
);
3604 (void) manager_run_environment_generators(m
);
3605 (void) manager_run_generators(m
);
3607 lookup_paths_log(&m
->lookup_paths
);
3609 /* We flushed out generated files, for which we don't watch mtime, so we should flush the old map. */
3610 manager_free_unit_name_maps(m
);
3611 m
->unit_file_state_outdated
= false;
3613 /* First, enumerate what we can from kernel and suchlike */
3614 manager_enumerate_perpetual(m
);
3615 manager_enumerate(m
);
3617 /* Second, deserialize our stored data */
3618 r
= manager_deserialize(m
, f
, fds
);
3620 log_warning_errno(r
, "Deserialization failed, proceeding anyway: %m");
3622 /* We don't need the serialization anymore */
3625 /* Re-register notify_fd as event source, and set up other sockets/communication channels we might need */
3626 (void) manager_setup_notify(m
);
3627 (void) manager_setup_cgroups_agent(m
);
3628 (void) manager_setup_user_lookup_fd(m
);
3630 /* Third, fire things up! */
3631 manager_coldplug(m
);
3633 /* Clean up runtime objects no longer referenced */
3636 /* Clean up deserialized tracked clients */
3637 m
->deserialized_subscribed
= strv_free(m
->deserialized_subscribed
);
3639 /* Consider the reload process complete now. */
3640 assert(m
->n_reloading
> 0);
3645 m
->send_reloading_done
= true;
3649 void manager_reset_failed(Manager
*m
) {
3654 HASHMAP_FOREACH(u
, m
->units
)
3655 unit_reset_failed(u
);
3658 bool manager_unit_inactive_or_pending(Manager
*m
, const char *name
) {
3664 /* Returns true if the unit is inactive or going down */
3665 u
= manager_get_unit(m
, name
);
3669 return unit_inactive_or_pending(u
);
3672 static void log_taint_string(Manager
*m
) {
3673 _cleanup_free_
char *taint
= NULL
;
3677 if (MANAGER_IS_USER(m
) || m
->taint_logged
)
3680 m
->taint_logged
= true; /* only check for taint once */
3682 taint
= manager_taint_string(m
);
3686 log_struct(LOG_NOTICE
,
3687 LOG_MESSAGE("System is tainted: %s", taint
),
3689 "MESSAGE_ID=" SD_MESSAGE_TAINTED_STR
);
3692 static void manager_notify_finished(Manager
*m
) {
3693 usec_t firmware_usec
, loader_usec
, kernel_usec
, initrd_usec
, userspace_usec
, total_usec
;
3695 if (MANAGER_IS_TEST_RUN(m
))
3698 if (MANAGER_IS_SYSTEM(m
) && m
->soft_reboots_count
> 0) {
3699 /* The soft-reboot case, where we only report data for the last reboot */
3700 firmware_usec
= loader_usec
= initrd_usec
= kernel_usec
= 0;
3701 total_usec
= userspace_usec
= usec_sub_unsigned(m
->timestamps
[MANAGER_TIMESTAMP_FINISH
].monotonic
, m
->timestamps
[MANAGER_TIMESTAMP_SHUTDOWN_START
].monotonic
);
3703 log_struct(LOG_INFO
,
3704 "MESSAGE_ID=" SD_MESSAGE_STARTUP_FINISHED_STR
,
3705 "USERSPACE_USEC="USEC_FMT
, userspace_usec
,
3706 LOG_MESSAGE("Soft-reboot finished in %s.",
3707 FORMAT_TIMESPAN(total_usec
, USEC_PER_MSEC
)));
3708 } else if (MANAGER_IS_SYSTEM(m
) && detect_container() <= 0) {
3709 char buf
[FORMAT_TIMESPAN_MAX
+ STRLEN(" (firmware) + ") + FORMAT_TIMESPAN_MAX
+ STRLEN(" (loader) + ")]
3712 size_t size
= sizeof buf
;
3714 /* Note that MANAGER_TIMESTAMP_KERNEL's monotonic value is always at 0, and
3715 * MANAGER_TIMESTAMP_FIRMWARE's and MANAGER_TIMESTAMP_LOADER's monotonic value should be considered
3716 * negative values. */
3718 firmware_usec
= m
->timestamps
[MANAGER_TIMESTAMP_FIRMWARE
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_LOADER
].monotonic
;
3719 loader_usec
= m
->timestamps
[MANAGER_TIMESTAMP_LOADER
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_KERNEL
].monotonic
;
3720 userspace_usec
= m
->timestamps
[MANAGER_TIMESTAMP_FINISH
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_USERSPACE
].monotonic
;
3721 total_usec
= m
->timestamps
[MANAGER_TIMESTAMP_FIRMWARE
].monotonic
+ m
->timestamps
[MANAGER_TIMESTAMP_FINISH
].monotonic
;
3723 if (firmware_usec
> 0)
3724 size
= strpcpyf(&p
, size
, "%s (firmware) + ", FORMAT_TIMESPAN(firmware_usec
, USEC_PER_MSEC
));
3725 if (loader_usec
> 0)
3726 size
= strpcpyf(&p
, size
, "%s (loader) + ", FORMAT_TIMESPAN(loader_usec
, USEC_PER_MSEC
));
3728 if (dual_timestamp_is_set(&m
->timestamps
[MANAGER_TIMESTAMP_INITRD
])) {
3730 /* The initrd case on bare-metal */
3731 kernel_usec
= m
->timestamps
[MANAGER_TIMESTAMP_INITRD
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_KERNEL
].monotonic
;
3732 initrd_usec
= m
->timestamps
[MANAGER_TIMESTAMP_USERSPACE
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_INITRD
].monotonic
;
3734 log_struct(LOG_INFO
,
3735 "MESSAGE_ID=" SD_MESSAGE_STARTUP_FINISHED_STR
,
3736 "KERNEL_USEC="USEC_FMT
, kernel_usec
,
3737 "INITRD_USEC="USEC_FMT
, initrd_usec
,
3738 "USERSPACE_USEC="USEC_FMT
, userspace_usec
,
3739 LOG_MESSAGE("Startup finished in %s%s (kernel) + %s (initrd) + %s (userspace) = %s.",
3741 FORMAT_TIMESPAN(kernel_usec
, USEC_PER_MSEC
),
3742 FORMAT_TIMESPAN(initrd_usec
, USEC_PER_MSEC
),
3743 FORMAT_TIMESPAN(userspace_usec
, USEC_PER_MSEC
),
3744 FORMAT_TIMESPAN(total_usec
, USEC_PER_MSEC
)));
3746 /* The initrd-less case on bare-metal */
3748 kernel_usec
= m
->timestamps
[MANAGER_TIMESTAMP_USERSPACE
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_KERNEL
].monotonic
;
3751 log_struct(LOG_INFO
,
3752 "MESSAGE_ID=" SD_MESSAGE_STARTUP_FINISHED_STR
,
3753 "KERNEL_USEC="USEC_FMT
, kernel_usec
,
3754 "USERSPACE_USEC="USEC_FMT
, userspace_usec
,
3755 LOG_MESSAGE("Startup finished in %s%s (kernel) + %s (userspace) = %s.",
3757 FORMAT_TIMESPAN(kernel_usec
, USEC_PER_MSEC
),
3758 FORMAT_TIMESPAN(userspace_usec
, USEC_PER_MSEC
),
3759 FORMAT_TIMESPAN(total_usec
, USEC_PER_MSEC
)));
3762 /* The container and --user case */
3763 firmware_usec
= loader_usec
= initrd_usec
= kernel_usec
= 0;
3764 total_usec
= userspace_usec
= m
->timestamps
[MANAGER_TIMESTAMP_FINISH
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_USERSPACE
].monotonic
;
3766 log_struct(LOG_INFO
,
3767 "MESSAGE_ID=" SD_MESSAGE_USER_STARTUP_FINISHED_STR
,
3768 "USERSPACE_USEC="USEC_FMT
, userspace_usec
,
3769 LOG_MESSAGE("Startup finished in %s.",
3770 FORMAT_TIMESPAN(total_usec
, USEC_PER_MSEC
)));
3773 bus_manager_send_finished(m
, firmware_usec
, loader_usec
, kernel_usec
, initrd_usec
, userspace_usec
, total_usec
);
3775 log_taint_string(m
);
3778 static void manager_send_ready_user_scope(Manager
*m
) {
3783 /* We send READY=1 on reaching basic.target only when running in --user mode. */
3784 if (!MANAGER_IS_USER(m
) || m
->ready_sent
)
3787 r
= sd_notify(/* unset_environment= */ false,
3789 "STATUS=Reached " SPECIAL_BASIC_TARGET
".");
3791 log_warning_errno(r
, "Failed to send readiness notification, ignoring: %m");
3793 m
->ready_sent
= true;
3794 m
->status_ready
= false;
3797 static void manager_send_ready_system_scope(Manager
*m
) {
3802 if (!MANAGER_IS_SYSTEM(m
))
3805 /* Skip the notification if nothing changed. */
3806 if (m
->ready_sent
&& m
->status_ready
)
3809 r
= sd_notify(/* unset_environment= */ false,
3813 log_full_errno(m
->ready_sent
? LOG_DEBUG
: LOG_WARNING
, r
,
3814 "Failed to send readiness notification, ignoring: %m");
3816 m
->ready_sent
= m
->status_ready
= true;
3819 static void manager_check_basic_target(Manager
*m
) {
3824 /* Small shortcut */
3825 if (m
->ready_sent
&& m
->taint_logged
)
3828 u
= manager_get_unit(m
, SPECIAL_BASIC_TARGET
);
3829 if (!u
|| !UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u
)))
3832 /* For user managers, send out READY=1 as soon as we reach basic.target */
3833 manager_send_ready_user_scope(m
);
3835 /* Log the taint string as soon as we reach basic.target */
3836 log_taint_string(m
);
3839 void manager_check_finished(Manager
*m
) {
3842 if (MANAGER_IS_RELOADING(m
))
3845 /* Verify that we have entered the event loop already, and not left it again. */
3846 if (!MANAGER_IS_RUNNING(m
))
3849 manager_check_basic_target(m
);
3851 if (!hashmap_isempty(m
->jobs
)) {
3852 if (m
->jobs_in_progress_event_source
)
3853 /* Ignore any failure, this is only for feedback */
3854 (void) sd_event_source_set_time(m
->jobs_in_progress_event_source
,
3855 manager_watch_jobs_next_time(m
));
3859 /* The jobs hashmap tends to grow a lot during boot, and then it's not reused until shutdown. Let's
3860 kill the hashmap if it is relatively large. */
3861 if (hashmap_buckets(m
->jobs
) > hashmap_size(m
->units
) / 10)
3862 m
->jobs
= hashmap_free(m
->jobs
);
3864 manager_send_ready_system_scope(m
);
3866 /* Notify Type=idle units that we are done now */
3867 manager_close_idle_pipe(m
);
3869 if (MANAGER_IS_FINISHED(m
))
3872 manager_flip_auto_status(m
, false, "boot finished");
3874 /* Turn off confirm spawn now */
3875 m
->confirm_spawn
= NULL
;
3877 /* No need to update ask password status when we're going non-interactive */
3878 manager_close_ask_password(m
);
3880 /* This is no longer the first boot */
3881 manager_set_first_boot(m
, false);
3883 dual_timestamp_now(m
->timestamps
+ MANAGER_TIMESTAMP_FINISH
);
3885 manager_notify_finished(m
);
3887 manager_invalidate_startup_units(m
);
3890 void manager_send_reloading(Manager
*m
) {
3893 /* Let whoever invoked us know that we are now reloading */
3894 (void) notify_reloading_full(/* status = */ NULL
);
3896 /* And ensure that we'll send READY=1 again as soon as we are ready again */
3897 m
->ready_sent
= false;
3900 static bool generator_path_any(const char* const* paths
) {
3903 /* Optimize by skipping the whole process by not creating output directories
3904 * if no generators are found. */
3905 STRV_FOREACH(path
, paths
)
3906 if (access(*path
, F_OK
) == 0)
3908 else if (errno
!= ENOENT
)
3909 log_warning_errno(errno
, "Failed to open generator directory %s: %m", *path
);
3914 static int manager_run_environment_generators(Manager
*m
) {
3915 char **tmp
= NULL
; /* this is only used in the forked process, no cleanup here */
3916 _cleanup_strv_free_
char **paths
= NULL
;
3918 [STDOUT_GENERATE
] = &tmp
,
3919 [STDOUT_COLLECT
] = &tmp
,
3920 [STDOUT_CONSUME
] = &m
->transient_environment
,
3924 if (MANAGER_IS_TEST_RUN(m
) && !(m
->test_run_flags
& MANAGER_TEST_RUN_ENV_GENERATORS
))
3927 paths
= env_generator_binary_paths(m
->runtime_scope
);
3931 if (!generator_path_any((const char* const*) paths
))
3935 r
= execute_directories((const char* const*) paths
, DEFAULT_TIMEOUT_USEC
, gather_environment
,
3936 args
, NULL
, m
->transient_environment
,
3937 EXEC_DIR_PARALLEL
| EXEC_DIR_IGNORE_ERRORS
| EXEC_DIR_SET_SYSTEMD_EXEC_PID
);
3941 static int build_generator_environment(Manager
*m
, char ***ret
) {
3942 _cleanup_strv_free_
char **nl
= NULL
;
3944 ConfidentialVirtualization cv
;
3950 /* Generators oftentimes want to know some basic facts about the environment they run in, in order to
3951 * adjust generated units to that. Let's pass down some bits of information that are easy for us to
3952 * determine (but a bit harder for generator scripts to determine), as environment variables. */
3954 nl
= strv_copy(m
->transient_environment
);
3958 r
= strv_env_assign(&nl
, "SYSTEMD_SCOPE", runtime_scope_to_string(m
->runtime_scope
));
3962 if (MANAGER_IS_SYSTEM(m
)) {
3963 /* Note that $SYSTEMD_IN_INITRD may be used to override the initrd detection in much of our
3964 * codebase. This is hence more than purely informational. It will shortcut detection of the
3965 * initrd state if generators invoke our own tools. But that's OK, as it would come to the
3966 * same results (hopefully). */
3967 r
= strv_env_assign(&nl
, "SYSTEMD_IN_INITRD", one_zero(in_initrd()));
3971 if (m
->first_boot
>= 0) {
3972 r
= strv_env_assign(&nl
, "SYSTEMD_FIRST_BOOT", one_zero(m
->first_boot
));
3978 v
= detect_virtualization();
3980 log_debug_errno(v
, "Failed to detect virtualization, ignoring: %m");
3984 s
= strjoina(VIRTUALIZATION_IS_VM(v
) ? "vm:" :
3985 VIRTUALIZATION_IS_CONTAINER(v
) ? "container:" : ":",
3986 virtualization_to_string(v
));
3988 r
= strv_env_assign(&nl
, "SYSTEMD_VIRTUALIZATION", s
);
3993 cv
= detect_confidential_virtualization();
3995 log_debug_errno(cv
, "Failed to detect confidential virtualization, ignoring: %m");
3997 r
= strv_env_assign(&nl
, "SYSTEMD_CONFIDENTIAL_VIRTUALIZATION", confidential_virtualization_to_string(cv
));
4002 r
= strv_env_assign(&nl
, "SYSTEMD_ARCHITECTURE", architecture_to_string(uname_architecture()));
4006 *ret
= TAKE_PTR(nl
);
4010 static int manager_execute_generators(Manager
*m
, char **paths
, bool remount_ro
) {
4011 _cleanup_strv_free_
char **ge
= NULL
;
4012 const char *argv
[] = {
4013 NULL
, /* Leave this empty, execute_directory() will fill something in */
4014 m
->lookup_paths
.generator
,
4015 m
->lookup_paths
.generator_early
,
4016 m
->lookup_paths
.generator_late
,
4021 r
= build_generator_environment(m
, &ge
);
4023 return log_error_errno(r
, "Failed to build generator environment: %m");
4026 /* Remount most of the filesystem tree read-only. We leave /sys/ as-is, because our code
4027 * checks whether it is read-only to detect containerized execution environments. We leave
4028 * /run/ as-is too, because that's where our output goes. We also leave /proc/ and /dev/shm/
4029 * because they're API, and /tmp/ that safe_fork() mounted for us.
4031 r
= bind_remount_recursive("/", MS_RDONLY
, MS_RDONLY
,
4032 STRV_MAKE("/sys", "/run", "/proc", "/dev/shm", "/tmp"));
4034 log_warning_errno(r
, "Read-only bind remount failed, ignoring: %m");
4037 BLOCK_WITH_UMASK(0022);
4038 return execute_directories(
4039 (const char* const*) paths
,
4040 DEFAULT_TIMEOUT_USEC
,
4041 /* callbacks= */ NULL
, /* callback_args= */ NULL
,
4044 EXEC_DIR_PARALLEL
| EXEC_DIR_IGNORE_ERRORS
| EXEC_DIR_SET_SYSTEMD_EXEC_PID
);
4047 static int manager_run_generators(Manager
*m
) {
4048 ForkFlags flags
= FORK_RESET_SIGNALS
| FORK_WAIT
| FORK_NEW_MOUNTNS
| FORK_MOUNTNS_SLAVE
;
4049 _cleanup_strv_free_
char **paths
= NULL
;
4054 if (MANAGER_IS_TEST_RUN(m
) && !(m
->test_run_flags
& MANAGER_TEST_RUN_GENERATORS
))
4057 paths
= generator_binary_paths(m
->runtime_scope
);
4061 if (!generator_path_any((const char* const*) paths
))
4064 r
= lookup_paths_mkdir_generator(&m
->lookup_paths
);
4066 log_error_errno(r
, "Failed to create generator directories: %m");
4070 /* If we are the system manager, we fork and invoke the generators in a sanitized mount namespace. If
4071 * we are the user manager, let's just execute the generators directly. We might not have the
4072 * necessary privileges, and the system manager has already mounted /tmp/ and everything else for us.
4074 if (MANAGER_IS_USER(m
)) {
4075 r
= manager_execute_generators(m
, paths
, /* remount_ro= */ false);
4079 /* On some systems /tmp/ doesn't exist, and on some other systems we cannot create it at all. Avoid
4080 * trying to mount a private tmpfs on it as there's no one size fits all. */
4081 if (is_dir("/tmp", /* follow= */ false) > 0)
4082 flags
|= FORK_PRIVATE_TMP
;
4084 r
= safe_fork("(sd-gens)", flags
, NULL
);
4086 r
= manager_execute_generators(m
, paths
, /* remount_ro= */ true);
4087 _exit(r
>= 0 ? EXIT_SUCCESS
: EXIT_FAILURE
);
4090 if (!ERRNO_IS_PRIVILEGE(r
) && r
!= -EINVAL
) {
4091 log_error_errno(r
, "Failed to fork off sandboxing environment for executing generators: %m");
4095 /* Failed to fork with new mount namespace? Maybe, running in a container environment with
4096 * seccomp or without capability.
4098 * We also allow -EINVAL to allow running without CLONE_NEWNS.
4100 * Also, when running on non-native userland architecture via systemd-nspawn and
4101 * qemu-user-static QEMU-emulator, clone() with CLONE_NEWNS fails with EINVAL, see
4102 * https://github.com/systemd/systemd/issues/28901.
4105 "Failed to fork off sandboxing environment for executing generators. "
4106 "Falling back to execute generators without sandboxing: %m");
4107 r
= manager_execute_generators(m
, paths
, /* remount_ro= */ false);
4111 lookup_paths_trim_generator(&m
->lookup_paths
);
4115 int manager_transient_environment_add(Manager
*m
, char **plus
) {
4120 if (strv_isempty(plus
))
4123 a
= strv_env_merge(m
->transient_environment
, plus
);
4127 sanitize_environment(a
);
4129 return strv_free_and_replace(m
->transient_environment
, a
);
4132 int manager_client_environment_modify(
4137 char **a
= NULL
, **b
= NULL
, **l
;
4141 if (strv_isempty(minus
) && strv_isempty(plus
))
4144 l
= m
->client_environment
;
4146 if (!strv_isempty(minus
)) {
4147 a
= strv_env_delete(l
, 1, minus
);
4154 if (!strv_isempty(plus
)) {
4155 b
= strv_env_merge(l
, plus
);
4164 if (m
->client_environment
!= l
)
4165 strv_free(m
->client_environment
);
4172 m
->client_environment
= sanitize_environment(l
);
4176 int manager_get_effective_environment(Manager
*m
, char ***ret
) {
4182 l
= strv_env_merge(m
->transient_environment
, m
->client_environment
);
4190 int manager_set_unit_defaults(Manager
*m
, const UnitDefaults
*defaults
) {
4191 _cleanup_free_
char *label
= NULL
;
4192 struct rlimit
*rlimit
[_RLIMIT_MAX
];
4198 if (streq_ptr(defaults
->smack_process_label
, "/"))
4201 const char *l
= defaults
->smack_process_label
;
4202 #ifdef SMACK_DEFAULT_PROCESS_LABEL
4204 l
= SMACK_DEFAULT_PROCESS_LABEL
;
4214 r
= rlimit_copy_all(rlimit
, defaults
->rlimit
);
4218 m
->defaults
.std_output
= defaults
->std_output
;
4219 m
->defaults
.std_error
= defaults
->std_error
;
4221 m
->defaults
.restart_usec
= defaults
->restart_usec
;
4222 m
->defaults
.timeout_start_usec
= defaults
->timeout_start_usec
;
4223 m
->defaults
.timeout_stop_usec
= defaults
->timeout_stop_usec
;
4224 m
->defaults
.timeout_abort_usec
= defaults
->timeout_abort_usec
;
4225 m
->defaults
.timeout_abort_set
= defaults
->timeout_abort_set
;
4226 m
->defaults
.device_timeout_usec
= defaults
->device_timeout_usec
;
4228 m
->defaults
.start_limit_interval
= defaults
->start_limit_interval
;
4229 m
->defaults
.start_limit_burst
= defaults
->start_limit_burst
;
4231 m
->defaults
.cpu_accounting
= defaults
->cpu_accounting
;
4232 m
->defaults
.memory_accounting
= defaults
->memory_accounting
;
4233 m
->defaults
.io_accounting
= defaults
->io_accounting
;
4234 m
->defaults
.blockio_accounting
= defaults
->blockio_accounting
;
4235 m
->defaults
.tasks_accounting
= defaults
->tasks_accounting
;
4236 m
->defaults
.ip_accounting
= defaults
->ip_accounting
;
4238 m
->defaults
.tasks_max
= defaults
->tasks_max
;
4239 m
->defaults
.timer_accuracy_usec
= defaults
->timer_accuracy_usec
;
4241 m
->defaults
.oom_policy
= defaults
->oom_policy
;
4242 m
->defaults
.oom_score_adjust
= defaults
->oom_score_adjust
;
4243 m
->defaults
.oom_score_adjust_set
= defaults
->oom_score_adjust_set
;
4245 m
->defaults
.memory_pressure_watch
= defaults
->memory_pressure_watch
;
4246 m
->defaults
.memory_pressure_threshold_usec
= defaults
->memory_pressure_threshold_usec
;
4248 free_and_replace(m
->defaults
.smack_process_label
, label
);
4249 rlimit_free_all(m
->defaults
.rlimit
);
4250 memcpy(m
->defaults
.rlimit
, rlimit
, sizeof(struct rlimit
*) * _RLIMIT_MAX
);
4255 void manager_recheck_dbus(Manager
*m
) {
4258 /* Connects to the bus if the dbus service and socket are running. If we are running in user mode
4259 * this is all it does. In system mode we'll also connect to the system bus (which will most likely
4260 * just reuse the connection of the API bus). That's because the system bus after all runs as service
4261 * of the system instance, while in the user instance we can assume it's already there. */
4263 if (MANAGER_IS_RELOADING(m
))
4264 return; /* don't check while we are reloading… */
4266 if (manager_dbus_is_running(m
, false)) {
4267 (void) bus_init_api(m
);
4269 if (MANAGER_IS_SYSTEM(m
))
4270 (void) bus_init_system(m
);
4272 (void) bus_done_api(m
);
4274 if (MANAGER_IS_SYSTEM(m
))
4275 (void) bus_done_system(m
);
4279 static bool manager_journal_is_running(Manager
*m
) {
4284 if (MANAGER_IS_TEST_RUN(m
))
4287 /* If we are the user manager we can safely assume that the journal is up */
4288 if (!MANAGER_IS_SYSTEM(m
))
4291 /* Check that the socket is not only up, but in RUNNING state */
4292 u
= manager_get_unit(m
, SPECIAL_JOURNALD_SOCKET
);
4295 if (SOCKET(u
)->state
!= SOCKET_RUNNING
)
4298 /* Similar, check if the daemon itself is fully up, too */
4299 u
= manager_get_unit(m
, SPECIAL_JOURNALD_SERVICE
);
4302 if (!IN_SET(SERVICE(u
)->state
, SERVICE_RELOAD
, SERVICE_RUNNING
))
4308 void disable_printk_ratelimit(void) {
4309 /* Disable kernel's printk ratelimit.
4311 * Logging to /dev/kmsg is most useful during early boot and shutdown, where normal logging
4312 * mechanisms are not available. The semantics of this sysctl are such that any kernel command-line
4313 * setting takes precedence. */
4316 r
= sysctl_write("kernel/printk_devkmsg", "on");
4318 log_debug_errno(r
, "Failed to set sysctl kernel.printk_devkmsg=on: %m");
4321 void manager_recheck_journal(Manager
*m
) {
4325 /* Don't bother with this unless we are in the special situation of being PID 1 */
4326 if (getpid_cached() != 1)
4329 /* Don't check this while we are reloading, things might still change */
4330 if (MANAGER_IS_RELOADING(m
))
4333 /* The journal is fully and entirely up? If so, let's permit logging to it, if that's configured. If
4334 * the journal is down, don't ever log to it, otherwise we might end up deadlocking ourselves as we
4335 * might trigger an activation ourselves we can't fulfill. */
4336 log_set_prohibit_ipc(!manager_journal_is_running(m
));
4340 static ShowStatus
manager_get_show_status(Manager
*m
) {
4343 if (MANAGER_IS_USER(m
))
4344 return _SHOW_STATUS_INVALID
;
4346 if (m
->show_status_overridden
!= _SHOW_STATUS_INVALID
)
4347 return m
->show_status_overridden
;
4349 return m
->show_status
;
4352 bool manager_get_show_status_on(Manager
*m
) {
4355 return show_status_on(manager_get_show_status(m
));
4358 static void set_show_status_marker(bool b
) {
4360 (void) touch("/run/systemd/show-status");
4362 (void) unlink("/run/systemd/show-status");
4365 void manager_set_show_status(Manager
*m
, ShowStatus mode
, const char *reason
) {
4368 assert(mode
>= 0 && mode
< _SHOW_STATUS_MAX
);
4370 if (MANAGER_IS_USER(m
))
4373 if (mode
== m
->show_status
)
4376 if (m
->show_status_overridden
== _SHOW_STATUS_INVALID
) {
4379 enabled
= show_status_on(mode
);
4380 log_debug("%s (%s) showing of status (%s).",
4381 enabled
? "Enabling" : "Disabling",
4382 strna(show_status_to_string(mode
)),
4385 set_show_status_marker(enabled
);
4388 m
->show_status
= mode
;
4391 void manager_override_show_status(Manager
*m
, ShowStatus mode
, const char *reason
) {
4393 assert(mode
< _SHOW_STATUS_MAX
);
4395 if (MANAGER_IS_USER(m
))
4398 if (mode
== m
->show_status_overridden
)
4401 m
->show_status_overridden
= mode
;
4403 if (mode
== _SHOW_STATUS_INVALID
)
4404 mode
= m
->show_status
;
4406 log_debug("%s (%s) showing of status (%s).",
4407 m
->show_status_overridden
!= _SHOW_STATUS_INVALID
? "Overriding" : "Restoring",
4408 strna(show_status_to_string(mode
)),
4411 set_show_status_marker(show_status_on(mode
));
4414 const char *manager_get_confirm_spawn(Manager
*m
) {
4415 static int last_errno
= 0;
4421 /* Here's the deal: we want to test the validity of the console but don't want
4422 * PID1 to go through the whole console process which might block. But we also
4423 * want to warn the user only once if something is wrong with the console so we
4424 * cannot do the sanity checks after spawning our children. So here we simply do
4425 * really basic tests to hopefully trap common errors.
4427 * If the console suddenly disappear at the time our children will really it
4428 * then they will simply fail to acquire it and a positive answer will be
4429 * assumed. New children will fall back to /dev/console though.
4431 * Note: TTYs are devices that can come and go any time, and frequently aren't
4432 * available yet during early boot (consider a USB rs232 dongle...). If for any
4433 * reason the configured console is not ready, we fall back to the default
4436 if (!m
->confirm_spawn
|| path_equal(m
->confirm_spawn
, "/dev/console"))
4437 return m
->confirm_spawn
;
4439 if (stat(m
->confirm_spawn
, &st
) < 0) {
4444 if (!S_ISCHR(st
.st_mode
)) {
4450 return m
->confirm_spawn
;
4453 if (last_errno
!= r
)
4454 last_errno
= log_warning_errno(r
, "Failed to open %s, using default console: %m", m
->confirm_spawn
);
4456 return "/dev/console";
4459 void manager_set_first_boot(Manager
*m
, bool b
) {
4462 if (!MANAGER_IS_SYSTEM(m
))
4465 if (m
->first_boot
!= (int) b
) {
4467 (void) touch("/run/systemd/first-boot");
4469 (void) unlink("/run/systemd/first-boot");
4475 void manager_disable_confirm_spawn(void) {
4476 (void) touch("/run/systemd/confirm_spawn_disabled");
4479 static bool manager_should_show_status(Manager
*m
, StatusType type
) {
4482 if (!MANAGER_IS_SYSTEM(m
))
4485 if (m
->no_console_output
)
4488 if (!IN_SET(manager_state(m
), MANAGER_INITIALIZING
, MANAGER_STARTING
, MANAGER_STOPPING
))
4491 /* If we cannot find out the status properly, just proceed. */
4492 if (type
!= STATUS_TYPE_EMERGENCY
&& manager_check_ask_password(m
) > 0)
4495 if (type
== STATUS_TYPE_NOTICE
&& m
->show_status
!= SHOW_STATUS_NO
)
4498 return manager_get_show_status_on(m
);
4501 void manager_status_printf(Manager
*m
, StatusType type
, const char *status
, const char *format
, ...) {
4504 /* If m is NULL, assume we're after shutdown and let the messages through. */
4506 if (m
&& !manager_should_show_status(m
, type
))
4509 /* XXX We should totally drop the check for ephemeral here
4510 * and thus effectively make 'Type=idle' pointless. */
4511 if (type
== STATUS_TYPE_EPHEMERAL
&& m
&& m
->n_on_console
> 0)
4514 va_start(ap
, format
);
4515 status_vprintf(status
, SHOW_STATUS_ELLIPSIZE
|(type
== STATUS_TYPE_EPHEMERAL
? SHOW_STATUS_EPHEMERAL
: 0), format
, ap
);
4519 Set
* manager_get_units_needing_mounts_for(Manager
*m
, const char *path
, UnitMountDependencyType t
) {
4522 assert(t
>= 0 && t
< _UNIT_MOUNT_DEPENDENCY_TYPE_MAX
);
4524 if (path_equal(path
, "/"))
4527 return hashmap_get(m
->units_needing_mounts_for
[t
], path
);
4530 int manager_update_failed_units(Manager
*m
, Unit
*u
, bool failed
) {
4535 assert(u
->manager
== m
);
4537 size
= set_size(m
->failed_units
);
4540 r
= set_ensure_put(&m
->failed_units
, NULL
, u
);
4544 (void) set_remove(m
->failed_units
, u
);
4546 if (set_size(m
->failed_units
) != size
)
4547 bus_manager_send_change_signal(m
);
4552 ManagerState
manager_state(Manager
*m
) {
4557 /* Is the special shutdown target active or queued? If so, we are in shutdown state */
4558 u
= manager_get_unit(m
, SPECIAL_SHUTDOWN_TARGET
);
4559 if (u
&& unit_active_or_pending(u
))
4560 return MANAGER_STOPPING
;
4562 /* Did we ever finish booting? If not then we are still starting up */
4563 if (!MANAGER_IS_FINISHED(m
)) {
4565 u
= manager_get_unit(m
, SPECIAL_BASIC_TARGET
);
4566 if (!u
|| !UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u
)))
4567 return MANAGER_INITIALIZING
;
4569 return MANAGER_STARTING
;
4572 if (MANAGER_IS_SYSTEM(m
)) {
4573 /* Are the rescue or emergency targets active or queued? If so we are in maintenance state */
4574 u
= manager_get_unit(m
, SPECIAL_RESCUE_TARGET
);
4575 if (u
&& unit_active_or_pending(u
))
4576 return MANAGER_MAINTENANCE
;
4578 u
= manager_get_unit(m
, SPECIAL_EMERGENCY_TARGET
);
4579 if (u
&& unit_active_or_pending(u
))
4580 return MANAGER_MAINTENANCE
;
4583 /* Are there any failed units? If so, we are in degraded mode */
4584 if (!set_isempty(m
->failed_units
))
4585 return MANAGER_DEGRADED
;
4587 return MANAGER_RUNNING
;
4590 static void manager_unref_uid_internal(
4594 int (*_clean_ipc
)(uid_t uid
)) {
4598 assert(uid_is_valid(uid
));
4601 /* A generic implementation, covering both manager_unref_uid() and manager_unref_gid(), under the
4602 * assumption that uid_t and gid_t are actually defined the same way, with the same validity rules.
4604 * We store a hashmap where the key is the UID/GID and the value is a 32-bit reference counter, whose
4605 * highest bit is used as flag for marking UIDs/GIDs whose IPC objects to remove when the last
4606 * reference to the UID/GID is dropped. The flag is set to on, once at least one reference from a
4607 * unit where RemoveIPC= is set is added on a UID/GID. It is reset when the UID's/GID's reference
4608 * counter drops to 0 again. */
4610 assert_cc(sizeof(uid_t
) == sizeof(gid_t
));
4611 assert_cc(UID_INVALID
== (uid_t
) GID_INVALID
);
4613 if (uid
== 0) /* We don't keep track of root, and will never destroy it */
4616 c
= PTR_TO_UINT32(hashmap_get(uid_refs
, UID_TO_PTR(uid
)));
4618 n
= c
& ~DESTROY_IPC_FLAG
;
4622 if (destroy_now
&& n
== 0) {
4623 hashmap_remove(uid_refs
, UID_TO_PTR(uid
));
4625 if (c
& DESTROY_IPC_FLAG
) {
4626 log_debug("%s " UID_FMT
" is no longer referenced, cleaning up its IPC.",
4627 _clean_ipc
== clean_ipc_by_uid
? "UID" : "GID",
4629 (void) _clean_ipc(uid
);
4632 c
= n
| (c
& DESTROY_IPC_FLAG
);
4633 assert_se(hashmap_update(uid_refs
, UID_TO_PTR(uid
), UINT32_TO_PTR(c
)) >= 0);
4637 void manager_unref_uid(Manager
*m
, uid_t uid
, bool destroy_now
) {
4638 manager_unref_uid_internal(m
->uid_refs
, uid
, destroy_now
, clean_ipc_by_uid
);
4641 void manager_unref_gid(Manager
*m
, gid_t gid
, bool destroy_now
) {
4642 manager_unref_uid_internal(m
->gid_refs
, (uid_t
) gid
, destroy_now
, clean_ipc_by_gid
);
4645 static int manager_ref_uid_internal(
4654 assert(uid_is_valid(uid
));
4656 /* A generic implementation, covering both manager_ref_uid() and manager_ref_gid(), under the
4657 * assumption that uid_t and gid_t are actually defined the same way, with the same validity
4660 assert_cc(sizeof(uid_t
) == sizeof(gid_t
));
4661 assert_cc(UID_INVALID
== (uid_t
) GID_INVALID
);
4663 if (uid
== 0) /* We don't keep track of root, and will never destroy it */
4666 r
= hashmap_ensure_allocated(uid_refs
, &trivial_hash_ops
);
4670 c
= PTR_TO_UINT32(hashmap_get(*uid_refs
, UID_TO_PTR(uid
)));
4672 n
= c
& ~DESTROY_IPC_FLAG
;
4675 if (n
& DESTROY_IPC_FLAG
) /* check for overflow */
4678 c
= n
| (c
& DESTROY_IPC_FLAG
) | (clean_ipc
? DESTROY_IPC_FLAG
: 0);
4680 return hashmap_replace(*uid_refs
, UID_TO_PTR(uid
), UINT32_TO_PTR(c
));
4683 int manager_ref_uid(Manager
*m
, uid_t uid
, bool clean_ipc
) {
4684 return manager_ref_uid_internal(&m
->uid_refs
, uid
, clean_ipc
);
4687 int manager_ref_gid(Manager
*m
, gid_t gid
, bool clean_ipc
) {
4688 return manager_ref_uid_internal(&m
->gid_refs
, (uid_t
) gid
, clean_ipc
);
4691 static void manager_vacuum_uid_refs_internal(
4693 int (*_clean_ipc
)(uid_t uid
)) {
4699 HASHMAP_FOREACH_KEY(p
, k
, uid_refs
) {
4703 uid
= PTR_TO_UID(k
);
4704 c
= PTR_TO_UINT32(p
);
4706 n
= c
& ~DESTROY_IPC_FLAG
;
4710 if (c
& DESTROY_IPC_FLAG
) {
4711 log_debug("Found unreferenced %s " UID_FMT
" after reload/reexec. Cleaning up.",
4712 _clean_ipc
== clean_ipc_by_uid
? "UID" : "GID",
4714 (void) _clean_ipc(uid
);
4717 assert_se(hashmap_remove(uid_refs
, k
) == p
);
4721 static void manager_vacuum_uid_refs(Manager
*m
) {
4722 manager_vacuum_uid_refs_internal(m
->uid_refs
, clean_ipc_by_uid
);
4725 static void manager_vacuum_gid_refs(Manager
*m
) {
4726 manager_vacuum_uid_refs_internal(m
->gid_refs
, clean_ipc_by_gid
);
4729 static void manager_vacuum(Manager
*m
) {
4732 /* Release any dynamic users no longer referenced */
4733 dynamic_user_vacuum(m
, true);
4735 /* Release any references to UIDs/GIDs no longer referenced, and destroy any IPC owned by them */
4736 manager_vacuum_uid_refs(m
);
4737 manager_vacuum_gid_refs(m
);
4739 /* Release any runtimes no longer referenced */
4740 exec_shared_runtime_vacuum(m
);
4743 int manager_dispatch_user_lookup_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
4747 char unit_name
[UNIT_NAME_MAX
+1];
4750 Manager
*m
= userdata
;
4758 /* Invoked whenever a child process succeeded resolving its user/group to use and sent us the
4759 * resulting UID/GID in a datagram. We parse the datagram here and pass it off to the unit, so that
4760 * it can add a reference to the UID/GID so that it can destroy the UID/GID's IPC objects when the
4761 * reference counter drops to 0. */
4763 l
= recv(fd
, &buffer
, sizeof(buffer
), MSG_DONTWAIT
);
4765 if (ERRNO_IS_TRANSIENT(errno
))
4768 return log_error_errno(errno
, "Failed to read from user lookup fd: %m");
4771 if ((size_t) l
<= offsetof(struct buffer
, unit_name
)) {
4772 log_warning("Received too short user lookup message, ignoring.");
4776 if ((size_t) l
> offsetof(struct buffer
, unit_name
) + UNIT_NAME_MAX
) {
4777 log_warning("Received too long user lookup message, ignoring.");
4781 if (!uid_is_valid(buffer
.uid
) && !gid_is_valid(buffer
.gid
)) {
4782 log_warning("Got user lookup message with invalid UID/GID pair, ignoring.");
4786 n
= (size_t) l
- offsetof(struct buffer
, unit_name
);
4787 if (memchr(buffer
.unit_name
, 0, n
)) {
4788 log_warning("Received lookup message with embedded NUL character, ignoring.");
4792 buffer
.unit_name
[n
] = 0;
4793 u
= manager_get_unit(m
, buffer
.unit_name
);
4795 log_debug("Got user lookup message but unit doesn't exist, ignoring.");
4799 log_unit_debug(u
, "User lookup succeeded: uid=" UID_FMT
" gid=" GID_FMT
, buffer
.uid
, buffer
.gid
);
4801 unit_notify_user_lookup(u
, buffer
.uid
, buffer
.gid
);
4805 static int short_uid_range(const char *path
) {
4806 _cleanup_(uid_range_freep
) UIDRange
*p
= NULL
;
4811 /* Taint systemd if we the UID range assigned to this environment doesn't at least cover 0…65534,
4812 * i.e. from root to nobody. */
4814 r
= uid_range_load_userns(path
, UID_RANGE_USERNS_INSIDE
, &p
);
4815 if (ERRNO_IS_NEG_NOT_SUPPORTED(r
))
4818 return log_debug_errno(r
, "Failed to load %s: %m", path
);
4820 return !uid_range_covers(p
, 0, 65535);
4823 char* manager_taint_string(const Manager
*m
) {
4824 /* Returns a "taint string", e.g. "local-hwclock:var-run-bad". Only things that are detected at
4825 * runtime should be tagged here. For stuff that is known during compilation, emit a warning in the
4826 * configuration phase. */
4830 const char* stage
[12] = {};
4833 _cleanup_free_
char *usrbin
= NULL
;
4834 if (readlink_malloc("/bin", &usrbin
) < 0 || !PATH_IN_SET(usrbin
, "usr/bin", "/usr/bin"))
4835 stage
[n
++] = "unmerged-usr";
4837 if (access("/proc/cgroups", F_OK
) < 0)
4838 stage
[n
++] = "cgroups-missing";
4840 if (cg_all_unified() == 0)
4841 stage
[n
++] = "cgroupsv1";
4843 if (clock_is_localtime(NULL
) > 0)
4844 stage
[n
++] = "local-hwclock";
4846 if (os_release_support_ended(NULL
, /* quiet= */ true, NULL
) > 0)
4847 stage
[n
++] = "support-ended";
4849 _cleanup_free_
char *destination
= NULL
;
4850 if (readlink_malloc("/var/run", &destination
) < 0 ||
4851 !PATH_IN_SET(destination
, "../run", "/run"))
4852 stage
[n
++] = "var-run-bad";
4854 _cleanup_free_
char *overflowuid
= NULL
, *overflowgid
= NULL
;
4855 if (read_one_line_file("/proc/sys/kernel/overflowuid", &overflowuid
) >= 0 &&
4856 !streq(overflowuid
, "65534"))
4857 stage
[n
++] = "overflowuid-not-65534";
4858 if (read_one_line_file("/proc/sys/kernel/overflowgid", &overflowgid
) >= 0 &&
4859 !streq(overflowgid
, "65534"))
4860 stage
[n
++] = "overflowgid-not-65534";
4863 assert_se(uname(&uts
) >= 0);
4864 if (strverscmp_improved(uts
.release
, KERNEL_BASELINE_VERSION
) < 0)
4865 stage
[n
++] = "old-kernel";
4867 if (short_uid_range("/proc/self/uid_map") > 0)
4868 stage
[n
++] = "short-uid-range";
4869 if (short_uid_range("/proc/self/gid_map") > 0)
4870 stage
[n
++] = "short-gid-range";
4872 assert(n
< ELEMENTSOF(stage
) - 1); /* One extra for NULL terminator */
4874 return strv_join((char**) stage
, ":");
4877 void manager_ref_console(Manager
*m
) {
4883 void manager_unref_console(Manager
*m
) {
4885 assert(m
->n_on_console
> 0);
4888 if (m
->n_on_console
== 0)
4889 m
->no_console_output
= false; /* unset no_console_output flag, since the console is definitely free now */
4892 void manager_override_log_level(Manager
*m
, int level
) {
4893 _cleanup_free_
char *s
= NULL
;
4896 if (!m
->log_level_overridden
) {
4897 m
->original_log_level
= log_get_max_level();
4898 m
->log_level_overridden
= true;
4901 (void) log_level_to_string_alloc(level
, &s
);
4902 log_info("Setting log level to %s.", strna(s
));
4904 log_set_max_level(level
);
4907 void manager_restore_original_log_level(Manager
*m
) {
4908 _cleanup_free_
char *s
= NULL
;
4911 if (!m
->log_level_overridden
)
4914 (void) log_level_to_string_alloc(m
->original_log_level
, &s
);
4915 log_info("Restoring log level to original (%s).", strna(s
));
4917 log_set_max_level(m
->original_log_level
);
4918 m
->log_level_overridden
= false;
4921 void manager_override_log_target(Manager
*m
, LogTarget target
) {
4924 if (!m
->log_target_overridden
) {
4925 m
->original_log_target
= log_get_target();
4926 m
->log_target_overridden
= true;
4929 log_info("Setting log target to %s.", log_target_to_string(target
));
4930 log_set_target(target
);
4933 void manager_restore_original_log_target(Manager
*m
) {
4936 if (!m
->log_target_overridden
)
4939 log_info("Restoring log target to original %s.", log_target_to_string(m
->original_log_target
));
4941 log_set_target(m
->original_log_target
);
4942 m
->log_target_overridden
= false;
4945 ManagerTimestamp
manager_timestamp_initrd_mangle(ManagerTimestamp s
) {
4947 s
>= MANAGER_TIMESTAMP_SECURITY_START
&&
4948 s
<= MANAGER_TIMESTAMP_UNITS_LOAD_FINISH
)
4949 return s
- MANAGER_TIMESTAMP_SECURITY_START
+ MANAGER_TIMESTAMP_INITRD_SECURITY_START
;
4953 int manager_allocate_idle_pipe(Manager
*m
) {
4958 if (m
->idle_pipe
[0] >= 0) {
4959 assert(m
->idle_pipe
[1] >= 0);
4960 assert(m
->idle_pipe
[2] >= 0);
4961 assert(m
->idle_pipe
[3] >= 0);
4965 assert(m
->idle_pipe
[1] < 0);
4966 assert(m
->idle_pipe
[2] < 0);
4967 assert(m
->idle_pipe
[3] < 0);
4969 r
= RET_NERRNO(pipe2(m
->idle_pipe
+ 0, O_NONBLOCK
|O_CLOEXEC
));
4973 r
= RET_NERRNO(pipe2(m
->idle_pipe
+ 2, O_NONBLOCK
|O_CLOEXEC
));
4975 safe_close_pair(m
->idle_pipe
+ 0);
4982 void unit_defaults_init(UnitDefaults
*defaults
, RuntimeScope scope
) {
4985 assert(scope
< _RUNTIME_SCOPE_MAX
);
4987 *defaults
= (UnitDefaults
) {
4988 .std_output
= EXEC_OUTPUT_JOURNAL
,
4989 .std_error
= EXEC_OUTPUT_INHERIT
,
4990 .restart_usec
= DEFAULT_RESTART_USEC
,
4991 .timeout_start_usec
= manager_default_timeout(scope
),
4992 .timeout_stop_usec
= manager_default_timeout(scope
),
4993 .timeout_abort_usec
= manager_default_timeout(scope
),
4994 .timeout_abort_set
= false,
4995 .device_timeout_usec
= manager_default_timeout(scope
),
4996 .start_limit_interval
= DEFAULT_START_LIMIT_INTERVAL
,
4997 .start_limit_burst
= DEFAULT_START_LIMIT_BURST
,
4999 /* On 4.15+ with unified hierarchy, CPU accounting is essentially free as it doesn't require the CPU
5000 * controller to be enabled, so the default is to enable it unless we got told otherwise. */
5001 .cpu_accounting
= cpu_accounting_is_cheap(),
5002 .memory_accounting
= MEMORY_ACCOUNTING_DEFAULT
,
5003 .io_accounting
= false,
5004 .blockio_accounting
= false,
5005 .tasks_accounting
= true,
5006 .ip_accounting
= false,
5008 .tasks_max
= DEFAULT_TASKS_MAX
,
5009 .timer_accuracy_usec
= 1 * USEC_PER_MINUTE
,
5011 .memory_pressure_watch
= CGROUP_PRESSURE_WATCH_AUTO
,
5012 .memory_pressure_threshold_usec
= MEMORY_PRESSURE_DEFAULT_THRESHOLD_USEC
,
5014 .oom_policy
= OOM_STOP
,
5015 .oom_score_adjust_set
= false,
5019 void unit_defaults_done(UnitDefaults
*defaults
) {
5022 defaults
->smack_process_label
= mfree(defaults
->smack_process_label
);
5023 rlimit_free_all(defaults
->rlimit
);
5026 LogTarget
manager_get_executor_log_target(Manager
*m
) {
5029 /* If journald is not available tell sd-executor to go to kmsg, as it might be starting journald */
5031 if (manager_journal_is_running(m
))
5032 return log_get_target();
5034 return LOG_TARGET_KMSG
;
5037 static const char *const manager_state_table
[_MANAGER_STATE_MAX
] = {
5038 [MANAGER_INITIALIZING
] = "initializing",
5039 [MANAGER_STARTING
] = "starting",
5040 [MANAGER_RUNNING
] = "running",
5041 [MANAGER_DEGRADED
] = "degraded",
5042 [MANAGER_MAINTENANCE
] = "maintenance",
5043 [MANAGER_STOPPING
] = "stopping",
5046 DEFINE_STRING_TABLE_LOOKUP(manager_state
, ManagerState
);
5048 static const char *const manager_timestamp_table
[_MANAGER_TIMESTAMP_MAX
] = {
5049 [MANAGER_TIMESTAMP_FIRMWARE
] = "firmware",
5050 [MANAGER_TIMESTAMP_LOADER
] = "loader",
5051 [MANAGER_TIMESTAMP_KERNEL
] = "kernel",
5052 [MANAGER_TIMESTAMP_INITRD
] = "initrd",
5053 [MANAGER_TIMESTAMP_USERSPACE
] = "userspace",
5054 [MANAGER_TIMESTAMP_FINISH
] = "finish",
5055 [MANAGER_TIMESTAMP_SECURITY_START
] = "security-start",
5056 [MANAGER_TIMESTAMP_SECURITY_FINISH
] = "security-finish",
5057 [MANAGER_TIMESTAMP_GENERATORS_START
] = "generators-start",
5058 [MANAGER_TIMESTAMP_GENERATORS_FINISH
] = "generators-finish",
5059 [MANAGER_TIMESTAMP_UNITS_LOAD_START
] = "units-load-start",
5060 [MANAGER_TIMESTAMP_UNITS_LOAD_FINISH
] = "units-load-finish",
5061 [MANAGER_TIMESTAMP_UNITS_LOAD
] = "units-load",
5062 [MANAGER_TIMESTAMP_INITRD_SECURITY_START
] = "initrd-security-start",
5063 [MANAGER_TIMESTAMP_INITRD_SECURITY_FINISH
] = "initrd-security-finish",
5064 [MANAGER_TIMESTAMP_INITRD_GENERATORS_START
] = "initrd-generators-start",
5065 [MANAGER_TIMESTAMP_INITRD_GENERATORS_FINISH
] = "initrd-generators-finish",
5066 [MANAGER_TIMESTAMP_INITRD_UNITS_LOAD_START
] = "initrd-units-load-start",
5067 [MANAGER_TIMESTAMP_INITRD_UNITS_LOAD_FINISH
] = "initrd-units-load-finish",
5068 [MANAGER_TIMESTAMP_SHUTDOWN_START
] = "shutdown-start",
5071 DEFINE_STRING_TABLE_LOOKUP(manager_timestamp
, ManagerTimestamp
);
5073 static const char* const oom_policy_table
[_OOM_POLICY_MAX
] = {
5074 [OOM_CONTINUE
] = "continue",
5075 [OOM_STOP
] = "stop",
5076 [OOM_KILL
] = "kill",
5079 DEFINE_STRING_TABLE_LOOKUP(oom_policy
, OOMPolicy
);