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 "bus-common-errors.h"
29 #include "bus-error.h"
30 #include "bus-kernel.h"
32 #include "clean-ipc.h"
33 #include "clock-util.h"
34 #include "common-signal.h"
35 #include "confidential-virt.h"
36 #include "constants.h"
37 #include "core-varlink.h"
38 #include "creds-util.h"
40 #include "dbus-manager.h"
41 #include "dbus-unit.h"
43 #include "dirent-util.h"
46 #include "event-util.h"
47 #include "exec-util.h"
49 #include "exit-status.h"
52 #include "generator-setup.h"
54 #include "initrd-util.h"
55 #include "inotify-util.h"
58 #include "label-util.h"
59 #include "load-fragment.h"
60 #include "locale-setup.h"
64 #include "manager-dump.h"
65 #include "manager-serialize.h"
66 #include "memory-util.h"
67 #include "mkdir-label.h"
68 #include "mount-util.h"
70 #include "parse-util.h"
71 #include "path-lookup.h"
72 #include "path-util.h"
73 #include "plymouth-util.h"
74 #include "pretty-print.h"
75 #include "process-util.h"
77 #include "ratelimit.h"
78 #include "rlimit-util.h"
80 #include "selinux-util.h"
81 #include "signal-util.h"
82 #include "socket-util.h"
84 #include "stat-util.h"
85 #include "string-table.h"
86 #include "string-util.h"
89 #include "sysctl-util.h"
90 #include "syslog-util.h"
91 #include "terminal-util.h"
92 #include "time-util.h"
93 #include "transaction.h"
94 #include "uid-range.h"
95 #include "umask-util.h"
96 #include "unit-name.h"
97 #include "user-util.h"
101 #define NOTIFY_RCVBUF_SIZE (8*1024*1024)
102 #define CGROUPS_AGENT_RCVBUF_SIZE (8*1024*1024)
104 /* Initial delay and the interval for printing status messages about running jobs */
105 #define JOBS_IN_PROGRESS_WAIT_USEC (2*USEC_PER_SEC)
106 #define JOBS_IN_PROGRESS_QUIET_WAIT_USEC (25*USEC_PER_SEC)
107 #define JOBS_IN_PROGRESS_PERIOD_USEC (USEC_PER_SEC / 3)
108 #define JOBS_IN_PROGRESS_PERIOD_DIVISOR 3
110 /* If there are more than 1K bus messages queue across our API and direct buses, then let's not add more on top until
111 * the queue gets more empty. */
112 #define MANAGER_BUS_BUSY_THRESHOLD 1024LU
114 /* How many units and jobs to process of the bus queue before returning to the event loop. */
115 #define MANAGER_BUS_MESSAGE_BUDGET 100U
117 #define DEFAULT_TASKS_MAX ((CGroupTasksMax) { 15U, 100U }) /* 15% */
119 static int manager_dispatch_notify_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
120 static int manager_dispatch_cgroups_agent_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
121 static int manager_dispatch_signal_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
122 static int manager_dispatch_time_change_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
123 static int manager_dispatch_idle_pipe_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
124 static int manager_dispatch_user_lookup_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
125 static int manager_dispatch_jobs_in_progress(sd_event_source
*source
, usec_t usec
, void *userdata
);
126 static int manager_dispatch_run_queue(sd_event_source
*source
, void *userdata
);
127 static int manager_dispatch_sigchld(sd_event_source
*source
, void *userdata
);
128 static int manager_dispatch_timezone_change(sd_event_source
*source
, const struct inotify_event
*event
, void *userdata
);
129 static int manager_run_environment_generators(Manager
*m
);
130 static int manager_run_generators(Manager
*m
);
131 static void manager_vacuum(Manager
*m
);
133 static usec_t
manager_watch_jobs_next_time(Manager
*m
) {
136 if (MANAGER_IS_USER(m
))
137 /* Let the user manager without a timeout show status quickly, so the system manager can make
138 * use of it, if it wants to. */
139 timeout
= JOBS_IN_PROGRESS_WAIT_USEC
* 2 / 3;
140 else if (show_status_on(m
->show_status
))
141 /* When status is on, just use the usual timeout. */
142 timeout
= JOBS_IN_PROGRESS_WAIT_USEC
;
144 timeout
= JOBS_IN_PROGRESS_QUIET_WAIT_USEC
;
146 return usec_add(now(CLOCK_MONOTONIC
), timeout
);
149 static bool manager_is_confirm_spawn_disabled(Manager
*m
) {
152 if (!m
->confirm_spawn
)
155 return access("/run/systemd/confirm_spawn_disabled", F_OK
) >= 0;
158 static void manager_watch_jobs_in_progress(Manager
*m
) {
164 /* We do not want to show the cylon animation if the user
165 * needs to confirm service executions otherwise confirmation
166 * messages will be screwed by the cylon animation. */
167 if (!manager_is_confirm_spawn_disabled(m
))
170 if (m
->jobs_in_progress_event_source
)
173 next
= manager_watch_jobs_next_time(m
);
174 r
= sd_event_add_time(
176 &m
->jobs_in_progress_event_source
,
179 manager_dispatch_jobs_in_progress
, m
);
183 (void) sd_event_source_set_description(m
->jobs_in_progress_event_source
, "manager-jobs-in-progress");
186 static void manager_flip_auto_status(Manager
*m
, bool enable
, const char *reason
) {
190 if (m
->show_status
== SHOW_STATUS_AUTO
)
191 manager_set_show_status(m
, SHOW_STATUS_TEMPORARY
, reason
);
193 if (m
->show_status
== SHOW_STATUS_TEMPORARY
)
194 manager_set_show_status(m
, SHOW_STATUS_AUTO
, reason
);
198 static void manager_print_jobs_in_progress(Manager
*m
) {
200 unsigned counter
= 0, print_nr
;
201 char cylon
[6 + CYLON_BUFFER_EXTRA
+ 1];
203 uint64_t timeout
= 0;
206 assert(m
->n_running_jobs
> 0);
208 manager_flip_auto_status(m
, true, "delay");
210 print_nr
= (m
->jobs_in_progress_iteration
/ JOBS_IN_PROGRESS_PERIOD_DIVISOR
) % m
->n_running_jobs
;
212 HASHMAP_FOREACH(j
, m
->jobs
)
213 if (j
->state
== JOB_RUNNING
&& counter
++ == print_nr
)
216 /* m->n_running_jobs must be consistent with the contents of m->jobs,
217 * so the above loop must have succeeded in finding j. */
218 assert(counter
== print_nr
+ 1);
221 cylon_pos
= m
->jobs_in_progress_iteration
% 14;
223 cylon_pos
= 14 - cylon_pos
;
224 draw_cylon(cylon
, sizeof(cylon
), 6, cylon_pos
);
226 m
->jobs_in_progress_iteration
++;
228 char job_of_n
[STRLEN("( of ) ") + DECIMAL_STR_MAX(unsigned)*2] = "";
229 if (m
->n_running_jobs
> 1)
230 xsprintf(job_of_n
, "(%u of %u) ", counter
, m
->n_running_jobs
);
232 (void) job_get_timeout(j
, &timeout
);
234 /* We want to use enough information for the user to identify previous lines talking about the same
235 * unit, but keep the message as short as possible. So if 'Starting foo.service' or 'Starting
236 * foo.service - Description' were used, 'foo.service' is enough here. On the other hand, if we used
237 * 'Starting Description' before, then we shall also use 'Description' here. So we pass NULL as the
238 * second argument to unit_status_string(). */
239 const char *ident
= unit_status_string(j
->unit
, NULL
);
241 const char *time
= FORMAT_TIMESPAN(now(CLOCK_MONOTONIC
) - j
->begin_usec
, 1*USEC_PER_SEC
);
242 const char *limit
= timeout
> 0 ? FORMAT_TIMESPAN(timeout
- j
->begin_usec
, 1*USEC_PER_SEC
) : "no limit";
244 if (m
->status_unit_format
== STATUS_UNIT_FORMAT_DESCRIPTION
)
245 /* When using 'Description', we effectively don't have enough space to show the nested status
246 * without ellipsization, so let's not even try. */
247 manager_status_printf(m
, STATUS_TYPE_EPHEMERAL
, cylon
,
248 "%sA %s job is running for %s (%s / %s)",
250 job_type_to_string(j
->type
),
254 const char *status_text
= unit_status_text(j
->unit
);
256 manager_status_printf(m
, STATUS_TYPE_EPHEMERAL
, cylon
,
257 "%sJob %s/%s running (%s / %s)%s%s",
260 job_type_to_string(j
->type
),
262 status_text
? ": " : "",
263 strempty(status_text
));
267 "STATUS=%sUser job %s/%s running (%s / %s)...",
270 job_type_to_string(j
->type
),
272 m
->status_ready
= false;
275 static int have_ask_password(void) {
276 _cleanup_closedir_
DIR *dir
= NULL
;
278 dir
= opendir("/run/systemd/ask-password");
286 FOREACH_DIRENT_ALL(de
, dir
, return -errno
)
287 if (startswith(de
->d_name
, "ask."))
292 static int manager_dispatch_ask_password_fd(sd_event_source
*source
,
293 int fd
, uint32_t revents
, void *userdata
) {
294 Manager
*m
= ASSERT_PTR(userdata
);
298 m
->have_ask_password
= have_ask_password();
299 if (m
->have_ask_password
< 0)
300 /* Log error but continue. Negative have_ask_password
301 * is treated as unknown status. */
302 log_error_errno(m
->have_ask_password
, "Failed to list /run/systemd/ask-password: %m");
307 static void manager_close_ask_password(Manager
*m
) {
310 m
->ask_password_event_source
= sd_event_source_disable_unref(m
->ask_password_event_source
);
311 m
->ask_password_inotify_fd
= safe_close(m
->ask_password_inotify_fd
);
312 m
->have_ask_password
= -EINVAL
;
315 static int manager_check_ask_password(Manager
*m
) {
320 if (!m
->ask_password_event_source
) {
321 assert(m
->ask_password_inotify_fd
< 0);
323 (void) mkdir_p_label("/run/systemd/ask-password", 0755);
325 m
->ask_password_inotify_fd
= inotify_init1(IN_NONBLOCK
|IN_CLOEXEC
);
326 if (m
->ask_password_inotify_fd
< 0)
327 return log_error_errno(errno
, "Failed to create inotify object: %m");
329 r
= inotify_add_watch_and_warn(m
->ask_password_inotify_fd
,
330 "/run/systemd/ask-password",
331 IN_CREATE
|IN_DELETE
|IN_MOVE
);
333 manager_close_ask_password(m
);
337 r
= sd_event_add_io(m
->event
, &m
->ask_password_event_source
,
338 m
->ask_password_inotify_fd
, EPOLLIN
,
339 manager_dispatch_ask_password_fd
, m
);
341 log_error_errno(r
, "Failed to add event source for /run/systemd/ask-password: %m");
342 manager_close_ask_password(m
);
346 (void) sd_event_source_set_description(m
->ask_password_event_source
, "manager-ask-password");
348 /* Queries might have been added meanwhile... */
349 manager_dispatch_ask_password_fd(m
->ask_password_event_source
,
350 m
->ask_password_inotify_fd
, EPOLLIN
, m
);
353 return m
->have_ask_password
;
356 static int manager_watch_idle_pipe(Manager
*m
) {
361 if (m
->idle_pipe_event_source
)
364 if (m
->idle_pipe
[2] < 0)
367 r
= sd_event_add_io(m
->event
, &m
->idle_pipe_event_source
, m
->idle_pipe
[2], EPOLLIN
, manager_dispatch_idle_pipe_fd
, m
);
369 return log_error_errno(r
, "Failed to watch idle pipe: %m");
371 (void) sd_event_source_set_description(m
->idle_pipe_event_source
, "manager-idle-pipe");
376 static void manager_close_idle_pipe(Manager
*m
) {
379 m
->idle_pipe_event_source
= sd_event_source_disable_unref(m
->idle_pipe_event_source
);
381 safe_close_pair(m
->idle_pipe
);
382 safe_close_pair(m
->idle_pipe
+ 2);
385 static int manager_setup_time_change(Manager
*m
) {
390 if (MANAGER_IS_TEST_RUN(m
))
393 m
->time_change_event_source
= sd_event_source_disable_unref(m
->time_change_event_source
);
395 r
= event_add_time_change(m
->event
, &m
->time_change_event_source
, manager_dispatch_time_change_fd
, m
);
397 return log_error_errno(r
, "Failed to create time change event source: %m");
399 /* Schedule this slightly earlier than the .timer event sources */
400 r
= sd_event_source_set_priority(m
->time_change_event_source
, SD_EVENT_PRIORITY_NORMAL
-1);
402 return log_error_errno(r
, "Failed to set priority of time change event sources: %m");
404 log_debug("Set up TFD_TIMER_CANCEL_ON_SET timerfd.");
409 static int manager_read_timezone_stat(Manager
*m
) {
415 /* Read the current stat() data of /etc/localtime so that we detect changes */
416 if (lstat("/etc/localtime", &st
) < 0) {
417 log_debug_errno(errno
, "Failed to stat /etc/localtime, ignoring: %m");
418 changed
= m
->etc_localtime_accessible
;
419 m
->etc_localtime_accessible
= false;
423 k
= timespec_load(&st
.st_mtim
);
424 changed
= !m
->etc_localtime_accessible
|| k
!= m
->etc_localtime_mtime
;
426 m
->etc_localtime_mtime
= k
;
427 m
->etc_localtime_accessible
= true;
433 static int manager_setup_timezone_change(Manager
*m
) {
434 _cleanup_(sd_event_source_unrefp
) sd_event_source
*new_event
= NULL
;
439 if (MANAGER_IS_TEST_RUN(m
))
442 /* We watch /etc/localtime for three events: change of the link count (which might mean removal from /etc even
443 * though another link might be kept), renames, and file close operations after writing. Note we don't bother
444 * with IN_DELETE_SELF, as that would just report when the inode is removed entirely, i.e. after the link count
445 * went to zero and all fds to it are closed.
447 * Note that we never follow symlinks here. This is a simplification, but should cover almost all cases
450 * Note that we create the new event source first here, before releasing the old one. This should optimize
451 * behaviour as this way sd-event can reuse the old watch in case the inode didn't change. */
453 r
= sd_event_add_inotify(m
->event
, &new_event
, "/etc/localtime",
454 IN_ATTRIB
|IN_MOVE_SELF
|IN_CLOSE_WRITE
|IN_DONT_FOLLOW
, manager_dispatch_timezone_change
, m
);
456 /* If the file doesn't exist yet, subscribe to /etc instead, and wait until it is created either by
457 * O_CREATE or by rename() */
459 log_debug_errno(r
, "/etc/localtime doesn't exist yet, watching /etc instead.");
460 r
= sd_event_add_inotify(m
->event
, &new_event
, "/etc",
461 IN_CREATE
|IN_MOVED_TO
|IN_ONLYDIR
, manager_dispatch_timezone_change
, m
);
464 return log_error_errno(r
, "Failed to create timezone change event source: %m");
466 /* Schedule this slightly earlier than the .timer event sources */
467 r
= sd_event_source_set_priority(new_event
, SD_EVENT_PRIORITY_NORMAL
-1);
469 return log_error_errno(r
, "Failed to set priority of timezone change event sources: %m");
471 sd_event_source_unref(m
->timezone_change_event_source
);
472 m
->timezone_change_event_source
= TAKE_PTR(new_event
);
477 static int enable_special_signals(Manager
*m
) {
478 _cleanup_close_
int fd
= -EBADF
;
482 if (MANAGER_IS_TEST_RUN(m
))
485 /* Enable that we get SIGINT on control-alt-del. In containers
486 * this will fail with EPERM (older) or EINVAL (newer), so
488 if (reboot(RB_DISABLE_CAD
) < 0 && !IN_SET(errno
, EPERM
, EINVAL
))
489 log_warning_errno(errno
, "Failed to enable ctrl-alt-del handling: %m");
491 fd
= open_terminal("/dev/tty0", O_RDWR
|O_NOCTTY
|O_CLOEXEC
);
493 /* Support systems without virtual console */
495 log_warning_errno(errno
, "Failed to open /dev/tty0: %m");
497 /* Enable that we get SIGWINCH on kbrequest */
498 if (ioctl(fd
, KDSIGACCEPT
, SIGWINCH
) < 0)
499 log_warning_errno(errno
, "Failed to enable kbrequest handling: %m");
505 #define RTSIG_IF_AVAILABLE(signum) (signum <= SIGRTMAX ? signum : -1)
507 static int manager_setup_signals(Manager
*m
) {
508 struct sigaction sa
= {
509 .sa_handler
= SIG_DFL
,
510 .sa_flags
= SA_NOCLDSTOP
|SA_RESTART
,
517 assert_se(sigaction(SIGCHLD
, &sa
, NULL
) == 0);
519 /* We make liberal use of realtime signals here. On
520 * Linux/glibc we have 30 of them (with the exception of Linux
521 * on hppa, see below), between SIGRTMIN+0 ... SIGRTMIN+30
524 assert_se(sigemptyset(&mask
) == 0);
525 sigset_add_many(&mask
,
526 SIGCHLD
, /* Child died */
527 SIGTERM
, /* Reexecute daemon */
528 SIGHUP
, /* Reload configuration */
529 SIGUSR1
, /* systemd: reconnect to D-Bus */
530 SIGUSR2
, /* systemd: dump status */
531 SIGINT
, /* Kernel sends us this on control-alt-del */
532 SIGWINCH
, /* Kernel sends us this on kbrequest (alt-arrowup) */
533 SIGPWR
, /* Some kernel drivers and upsd send us this on power failure */
535 SIGRTMIN
+0, /* systemd: start default.target */
536 SIGRTMIN
+1, /* systemd: isolate rescue.target */
537 SIGRTMIN
+2, /* systemd: isolate emergency.target */
538 SIGRTMIN
+3, /* systemd: start halt.target */
539 SIGRTMIN
+4, /* systemd: start poweroff.target */
540 SIGRTMIN
+5, /* systemd: start reboot.target */
541 SIGRTMIN
+6, /* systemd: start kexec.target */
542 SIGRTMIN
+7, /* systemd: start soft-reboot.target */
544 /* ... space for more special targets ... */
546 SIGRTMIN
+13, /* systemd: Immediate halt */
547 SIGRTMIN
+14, /* systemd: Immediate poweroff */
548 SIGRTMIN
+15, /* systemd: Immediate reboot */
549 SIGRTMIN
+16, /* systemd: Immediate kexec */
550 SIGRTMIN
+17, /* systemd: Immediate soft-reboot */
551 SIGRTMIN
+18, /* systemd: control command */
555 SIGRTMIN
+20, /* systemd: enable status messages */
556 SIGRTMIN
+21, /* systemd: disable status messages */
557 SIGRTMIN
+22, /* systemd: set log level to LOG_DEBUG */
558 SIGRTMIN
+23, /* systemd: set log level to LOG_INFO */
559 SIGRTMIN
+24, /* systemd: Immediate exit (--user only) */
560 SIGRTMIN
+25, /* systemd: reexecute manager */
562 /* Apparently Linux on hppa had fewer RT signals until v3.18,
563 * SIGRTMAX was SIGRTMIN+25, and then SIGRTMIN was lowered,
564 * see commit v3.17-7614-g1f25df2eff.
566 * We cannot unconditionally make use of those signals here,
567 * so let's use a runtime check. Since these commands are
568 * accessible by different means and only really a safety
569 * net, the missing functionality on hppa shouldn't matter.
572 RTSIG_IF_AVAILABLE(SIGRTMIN
+26), /* systemd: set log target to journal-or-kmsg */
573 RTSIG_IF_AVAILABLE(SIGRTMIN
+27), /* systemd: set log target to console */
574 RTSIG_IF_AVAILABLE(SIGRTMIN
+28), /* systemd: set log target to kmsg */
575 RTSIG_IF_AVAILABLE(SIGRTMIN
+29), /* systemd: set log target to syslog-or-kmsg (obsolete) */
577 /* ... one free signal here SIGRTMIN+30 ... */
579 assert_se(sigprocmask(SIG_SETMASK
, &mask
, NULL
) == 0);
581 m
->signal_fd
= signalfd(-1, &mask
, SFD_NONBLOCK
|SFD_CLOEXEC
);
582 if (m
->signal_fd
< 0)
585 r
= sd_event_add_io(m
->event
, &m
->signal_event_source
, m
->signal_fd
, EPOLLIN
, manager_dispatch_signal_fd
, m
);
589 (void) sd_event_source_set_description(m
->signal_event_source
, "manager-signal");
591 /* Process signals a bit earlier than the rest of things, but later than notify_fd processing, so that the
592 * notify processing can still figure out to which process/service a message belongs, before we reap the
593 * process. Also, process this before handling cgroup notifications, so that we always collect child exit
594 * status information before detecting that there's no process in a cgroup. */
595 r
= sd_event_source_set_priority(m
->signal_event_source
, SD_EVENT_PRIORITY_NORMAL
-6);
599 if (MANAGER_IS_SYSTEM(m
))
600 return enable_special_signals(m
);
605 static char** sanitize_environment(char **l
) {
607 /* Let's remove some environment variables that we need ourselves to communicate with our clients */
611 "CONFIGURATION_DIRECTORY",
612 "CREDENTIALS_DIRECTORY",
624 "MEMORY_PRESSURE_WATCH",
625 "MEMORY_PRESSURE_WRITE",
627 "MONITOR_EXIT_STATUS",
628 "MONITOR_INVOCATION_ID",
629 "MONITOR_SERVICE_RESULT",
640 "TRIGGER_TIMER_MONOTONIC_USEC",
641 "TRIGGER_TIMER_REALTIME_USEC",
647 /* Let's order the environment alphabetically, just to make it pretty */
651 int manager_default_environment(Manager
*m
) {
656 m
->transient_environment
= strv_free(m
->transient_environment
);
658 if (MANAGER_IS_SYSTEM(m
)) {
659 /* The system manager always starts with a clean environment for its children. It does not
660 * import the kernel's or the parents' exported variables.
662 * The initial passed environment is untouched to keep /proc/self/environ valid; it is used
663 * for tagging the init process inside containers. */
664 m
->transient_environment
= strv_new("PATH=" DEFAULT_PATH
);
665 if (!m
->transient_environment
)
668 /* Import locale variables LC_*= from configuration */
669 (void) locale_setup(&m
->transient_environment
);
671 /* The user manager passes its own environment along to its children, except for $PATH. */
672 m
->transient_environment
= strv_copy(environ
);
673 if (!m
->transient_environment
)
676 r
= strv_env_replace_strdup(&m
->transient_environment
, "PATH=" DEFAULT_USER_PATH
);
681 sanitize_environment(m
->transient_environment
);
685 static int manager_setup_prefix(Manager
*m
) {
691 static const struct table_entry paths_system
[_EXEC_DIRECTORY_TYPE_MAX
] = {
692 [EXEC_DIRECTORY_RUNTIME
] = { SD_PATH_SYSTEM_RUNTIME
, NULL
},
693 [EXEC_DIRECTORY_STATE
] = { SD_PATH_SYSTEM_STATE_PRIVATE
, NULL
},
694 [EXEC_DIRECTORY_CACHE
] = { SD_PATH_SYSTEM_STATE_CACHE
, NULL
},
695 [EXEC_DIRECTORY_LOGS
] = { SD_PATH_SYSTEM_STATE_LOGS
, NULL
},
696 [EXEC_DIRECTORY_CONFIGURATION
] = { SD_PATH_SYSTEM_CONFIGURATION
, NULL
},
699 static const struct table_entry paths_user
[_EXEC_DIRECTORY_TYPE_MAX
] = {
700 [EXEC_DIRECTORY_RUNTIME
] = { SD_PATH_USER_RUNTIME
, NULL
},
701 [EXEC_DIRECTORY_STATE
] = { SD_PATH_USER_STATE_PRIVATE
, NULL
},
702 [EXEC_DIRECTORY_CACHE
] = { SD_PATH_USER_STATE_CACHE
, NULL
},
703 [EXEC_DIRECTORY_LOGS
] = { SD_PATH_USER_STATE_PRIVATE
, "log" },
704 [EXEC_DIRECTORY_CONFIGURATION
] = { SD_PATH_USER_CONFIGURATION
, NULL
},
709 const struct table_entry
*p
= MANAGER_IS_SYSTEM(m
) ? paths_system
: paths_user
;
712 for (ExecDirectoryType i
= 0; i
< _EXEC_DIRECTORY_TYPE_MAX
; i
++) {
713 r
= sd_path_lookup(p
[i
].type
, p
[i
].suffix
, &m
->prefix
[i
]);
715 return log_warning_errno(r
, "Failed to lookup %s path: %m",
716 exec_directory_type_to_string(i
));
722 static void manager_free_unit_name_maps(Manager
*m
) {
723 m
->unit_id_map
= hashmap_free(m
->unit_id_map
);
724 m
->unit_name_map
= hashmap_free(m
->unit_name_map
);
725 m
->unit_path_cache
= set_free(m
->unit_path_cache
);
726 m
->unit_cache_timestamp_hash
= 0;
729 static int manager_setup_run_queue(Manager
*m
) {
733 assert(!m
->run_queue_event_source
);
735 r
= sd_event_add_defer(m
->event
, &m
->run_queue_event_source
, manager_dispatch_run_queue
, m
);
739 r
= sd_event_source_set_priority(m
->run_queue_event_source
, SD_EVENT_PRIORITY_IDLE
);
743 r
= sd_event_source_set_enabled(m
->run_queue_event_source
, SD_EVENT_OFF
);
747 (void) sd_event_source_set_description(m
->run_queue_event_source
, "manager-run-queue");
752 static int manager_setup_sigchld_event_source(Manager
*m
) {
756 assert(!m
->sigchld_event_source
);
758 r
= sd_event_add_defer(m
->event
, &m
->sigchld_event_source
, manager_dispatch_sigchld
, m
);
762 r
= sd_event_source_set_priority(m
->sigchld_event_source
, SD_EVENT_PRIORITY_NORMAL
-7);
766 r
= sd_event_source_set_enabled(m
->sigchld_event_source
, SD_EVENT_OFF
);
770 (void) sd_event_source_set_description(m
->sigchld_event_source
, "manager-sigchld");
775 int manager_setup_memory_pressure_event_source(Manager
*m
) {
780 m
->memory_pressure_event_source
= sd_event_source_disable_unref(m
->memory_pressure_event_source
);
782 r
= sd_event_add_memory_pressure(m
->event
, &m
->memory_pressure_event_source
, NULL
, NULL
);
784 log_full_errno(ERRNO_IS_NOT_SUPPORTED(r
) || ERRNO_IS_PRIVILEGE(r
) || (r
== -EHOSTDOWN
) ? LOG_DEBUG
: LOG_NOTICE
, r
,
785 "Failed to establish memory pressure event source, ignoring: %m");
786 else if (m
->defaults
.memory_pressure_threshold_usec
!= USEC_INFINITY
) {
788 /* If there's a default memory pressure threshold set, also apply it to the service manager itself */
789 r
= sd_event_source_set_memory_pressure_period(
790 m
->memory_pressure_event_source
,
791 m
->defaults
.memory_pressure_threshold_usec
,
792 MEMORY_PRESSURE_DEFAULT_WINDOW_USEC
);
794 log_warning_errno(r
, "Failed to adjust memory pressure threshold, ignoring: %m");
800 static int manager_find_credentials_dirs(Manager
*m
) {
806 r
= get_credentials_dir(&e
);
809 log_debug_errno(r
, "Failed to determine credentials directory, ignoring: %m");
811 m
->received_credentials_directory
= strdup(e
);
812 if (!m
->received_credentials_directory
)
816 r
= get_encrypted_credentials_dir(&e
);
819 log_debug_errno(r
, "Failed to determine encrypted credentials directory, ignoring: %m");
821 m
->received_encrypted_credentials_directory
= strdup(e
);
822 if (!m
->received_encrypted_credentials_directory
)
829 void manager_set_switching_root(Manager
*m
, bool switching_root
) {
832 m
->switching_root
= MANAGER_IS_SYSTEM(m
) && switching_root
;
835 double manager_get_progress(Manager
*m
) {
838 if (MANAGER_IS_FINISHED(m
) || m
->n_installed_jobs
== 0)
841 return 1.0 - ((double) hashmap_size(m
->jobs
) / (double) m
->n_installed_jobs
);
844 static int compare_job_priority(const void *a
, const void *b
) {
845 const Job
*x
= a
, *y
= b
;
847 return unit_compare_priority(x
->unit
, y
->unit
);
850 int manager_new(RuntimeScope runtime_scope
, ManagerTestRunFlags test_run_flags
, Manager
**ret
) {
851 _cleanup_(manager_freep
) Manager
*m
= NULL
;
854 assert(IN_SET(runtime_scope
, RUNTIME_SCOPE_SYSTEM
, RUNTIME_SCOPE_USER
));
862 .runtime_scope
= runtime_scope
,
863 .objective
= _MANAGER_OBJECTIVE_INVALID
,
865 .status_unit_format
= STATUS_UNIT_FORMAT_DEFAULT
,
867 .original_log_level
= -1,
868 .original_log_target
= _LOG_TARGET_INVALID
,
870 .watchdog_overridden
[WATCHDOG_RUNTIME
] = USEC_INFINITY
,
871 .watchdog_overridden
[WATCHDOG_REBOOT
] = USEC_INFINITY
,
872 .watchdog_overridden
[WATCHDOG_KEXEC
] = USEC_INFINITY
,
873 .watchdog_overridden
[WATCHDOG_PRETIMEOUT
] = USEC_INFINITY
,
875 .show_status_overridden
= _SHOW_STATUS_INVALID
,
878 .cgroups_agent_fd
= -EBADF
,
880 .user_lookup_fds
= EBADF_PAIR
,
881 .private_listen_fd
= -EBADF
,
882 .dev_autofs_fd
= -EBADF
,
883 .cgroup_inotify_fd
= -EBADF
,
884 .pin_cgroupfs_fd
= -EBADF
,
885 .ask_password_inotify_fd
= -EBADF
,
886 .idle_pipe
= { -EBADF
, -EBADF
, -EBADF
, -EBADF
},
888 /* start as id #1, so that we can leave #0 around as "null-like" value */
891 .have_ask_password
= -EINVAL
, /* we don't know */
893 .test_run_flags
= test_run_flags
,
895 .dump_ratelimit
= (const RateLimit
) { .interval
= 10 * USEC_PER_MINUTE
, .burst
= 10 },
897 .executor_fd
= -EBADF
,
900 unit_defaults_init(&m
->defaults
, runtime_scope
);
903 if (MANAGER_IS_SYSTEM(m
) && detect_container() <= 0)
904 boot_timestamps(m
->timestamps
+ MANAGER_TIMESTAMP_USERSPACE
,
905 m
->timestamps
+ MANAGER_TIMESTAMP_FIRMWARE
,
906 m
->timestamps
+ MANAGER_TIMESTAMP_LOADER
);
909 /* Prepare log fields we can use for structured logging */
910 if (MANAGER_IS_SYSTEM(m
)) {
911 m
->unit_log_field
= "UNIT=";
912 m
->unit_log_format_string
= "UNIT=%s";
914 m
->invocation_log_field
= "INVOCATION_ID=";
915 m
->invocation_log_format_string
= "INVOCATION_ID=%s";
917 m
->unit_log_field
= "USER_UNIT=";
918 m
->unit_log_format_string
= "USER_UNIT=%s";
920 m
->invocation_log_field
= "USER_INVOCATION_ID=";
921 m
->invocation_log_format_string
= "USER_INVOCATION_ID=%s";
924 /* Reboot immediately if the user hits C-A-D more often than 7x per 2s */
925 m
->ctrl_alt_del_ratelimit
= (const RateLimit
) { .interval
= 2 * USEC_PER_SEC
, .burst
= 7 };
927 r
= manager_default_environment(m
);
931 r
= hashmap_ensure_allocated(&m
->units
, &string_hash_ops
);
935 r
= hashmap_ensure_allocated(&m
->cgroup_unit
, &path_hash_ops
);
939 r
= hashmap_ensure_allocated(&m
->watch_bus
, &string_hash_ops
);
943 r
= prioq_ensure_allocated(&m
->run_queue
, compare_job_priority
);
947 r
= manager_setup_prefix(m
);
951 r
= manager_find_credentials_dirs(m
);
955 r
= sd_event_default(&m
->event
);
959 r
= manager_setup_run_queue(m
);
963 if (FLAGS_SET(test_run_flags
, MANAGER_TEST_RUN_MINIMAL
)) {
964 m
->cgroup_root
= strdup("");
968 r
= manager_setup_signals(m
);
972 r
= manager_setup_cgroup(m
);
976 r
= manager_setup_time_change(m
);
980 r
= manager_read_timezone_stat(m
);
984 (void) manager_setup_timezone_change(m
);
986 r
= manager_setup_sigchld_event_source(m
);
990 r
= manager_setup_memory_pressure_event_source(m
);
995 if (MANAGER_IS_SYSTEM(m
) && lsm_bpf_supported(/* initialize = */ true)) {
996 r
= lsm_bpf_setup(m
);
998 log_warning_errno(r
, "Failed to setup LSM BPF, ignoring: %m");
1003 if (test_run_flags
== 0) {
1004 if (MANAGER_IS_SYSTEM(m
))
1005 r
= mkdir_label("/run/systemd/units", 0755);
1007 _cleanup_free_
char *units_path
= NULL
;
1008 r
= xdg_user_runtime_dir(&units_path
, "/systemd/units");
1011 r
= mkdir_p_label(units_path
, 0755);
1014 if (r
< 0 && r
!= -EEXIST
)
1017 m
->executor_fd
= open(SYSTEMD_EXECUTOR_BINARY_PATH
, O_CLOEXEC
|O_PATH
);
1018 if (m
->executor_fd
< 0)
1019 return log_warning_errno(errno
,
1020 "Failed to open executor binary '%s': %m",
1021 SYSTEMD_EXECUTOR_BINARY_PATH
);
1022 } else if (!FLAGS_SET(test_run_flags
, MANAGER_TEST_DONT_OPEN_EXECUTOR
)) {
1023 _cleanup_free_
char *self_exe
= NULL
, *executor_path
= NULL
;
1024 _cleanup_close_
int self_dir_fd
= -EBADF
;
1025 int level
= LOG_DEBUG
;
1027 /* Prefer sd-executor from the same directory as the test, e.g.: when running unit tests from the
1028 * build directory. Fallback to working directory and then the installation path. */
1029 r
= readlink_and_make_absolute("/proc/self/exe", &self_exe
);
1033 self_dir_fd
= open_parent(self_exe
, O_CLOEXEC
|O_PATH
|O_DIRECTORY
, 0);
1034 if (self_dir_fd
< 0)
1037 m
->executor_fd
= RET_NERRNO(openat(self_dir_fd
, "systemd-executor", O_CLOEXEC
|O_PATH
));
1038 if (m
->executor_fd
== -ENOENT
)
1039 m
->executor_fd
= RET_NERRNO(openat(AT_FDCWD
, "systemd-executor", O_CLOEXEC
|O_PATH
));
1040 if (m
->executor_fd
== -ENOENT
) {
1041 m
->executor_fd
= RET_NERRNO(open(SYSTEMD_EXECUTOR_BINARY_PATH
, O_CLOEXEC
|O_PATH
));
1042 level
= LOG_WARNING
; /* Tests should normally use local builds */
1044 if (m
->executor_fd
< 0)
1045 return m
->executor_fd
;
1047 r
= fd_get_path(m
->executor_fd
, &executor_path
);
1051 log_full(level
, "Using systemd-executor binary from '%s'.", executor_path
);
1054 /* Note that we do not set up the notify fd here. We do that after deserialization,
1055 * since they might have gotten serialized across the reexec. */
1062 static int manager_setup_notify(Manager
*m
) {
1065 if (MANAGER_IS_TEST_RUN(m
))
1068 if (m
->notify_fd
< 0) {
1069 _cleanup_close_
int fd
= -EBADF
;
1070 union sockaddr_union sa
;
1073 /* First free all secondary fields */
1074 m
->notify_socket
= mfree(m
->notify_socket
);
1075 m
->notify_event_source
= sd_event_source_disable_unref(m
->notify_event_source
);
1077 fd
= socket(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
|SOCK_NONBLOCK
, 0);
1079 return log_error_errno(errno
, "Failed to allocate notification socket: %m");
1081 fd_increase_rxbuf(fd
, NOTIFY_RCVBUF_SIZE
);
1083 m
->notify_socket
= path_join(m
->prefix
[EXEC_DIRECTORY_RUNTIME
], "systemd/notify");
1084 if (!m
->notify_socket
)
1087 r
= sockaddr_un_set_path(&sa
.un
, m
->notify_socket
);
1089 return log_error_errno(r
, "Notify socket '%s' not valid for AF_UNIX socket address, refusing.",
1093 (void) mkdir_parents_label(m
->notify_socket
, 0755);
1094 (void) sockaddr_un_unlink(&sa
.un
);
1096 r
= mac_selinux_bind(fd
, &sa
.sa
, sa_len
);
1098 return log_error_errno(r
, "bind(%s) failed: %m", m
->notify_socket
);
1100 r
= setsockopt_int(fd
, SOL_SOCKET
, SO_PASSCRED
, true);
1102 return log_error_errno(r
, "SO_PASSCRED failed: %m");
1104 m
->notify_fd
= TAKE_FD(fd
);
1106 log_debug("Using notification socket %s", m
->notify_socket
);
1109 if (!m
->notify_event_source
) {
1110 r
= sd_event_add_io(m
->event
, &m
->notify_event_source
, m
->notify_fd
, EPOLLIN
, manager_dispatch_notify_fd
, m
);
1112 return log_error_errno(r
, "Failed to allocate notify event source: %m");
1114 /* Process notification messages a bit earlier than SIGCHLD, so that we can still identify to which
1115 * service an exit message belongs. */
1116 r
= sd_event_source_set_priority(m
->notify_event_source
, SD_EVENT_PRIORITY_NORMAL
-8);
1118 return log_error_errno(r
, "Failed to set priority of notify event source: %m");
1120 (void) sd_event_source_set_description(m
->notify_event_source
, "manager-notify");
1126 static int manager_setup_cgroups_agent(Manager
*m
) {
1128 static const union sockaddr_union sa
= {
1129 .un
.sun_family
= AF_UNIX
,
1130 .un
.sun_path
= "/run/systemd/cgroups-agent",
1134 /* This creates a listening socket we receive cgroups agent messages on. We do not use D-Bus for delivering
1135 * these messages from the cgroups agent binary to PID 1, as the cgroups agent binary is very short-living, and
1136 * each instance of it needs a new D-Bus connection. Since D-Bus connections are SOCK_STREAM/AF_UNIX, on
1137 * overloaded systems the backlog of the D-Bus socket becomes relevant, as not more than the configured number
1138 * of D-Bus connections may be queued until the kernel will start dropping further incoming connections,
1139 * possibly resulting in lost cgroups agent messages. To avoid this, we'll use a private SOCK_DGRAM/AF_UNIX
1140 * socket, where no backlog is relevant as communication may take place without an actual connect() cycle, and
1141 * we thus won't lose messages.
1143 * Note that PID 1 will forward the agent message to system bus, so that the user systemd instance may listen
1144 * to it. The system instance hence listens on this special socket, but the user instances listen on the system
1145 * bus for these messages. */
1147 if (MANAGER_IS_TEST_RUN(m
))
1150 if (!MANAGER_IS_SYSTEM(m
))
1153 r
= cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER
);
1155 return log_error_errno(r
, "Failed to determine whether unified cgroups hierarchy is used: %m");
1156 if (r
> 0) /* We don't need this anymore on the unified hierarchy */
1159 if (m
->cgroups_agent_fd
< 0) {
1160 _cleanup_close_
int fd
= -EBADF
;
1162 /* First free all secondary fields */
1163 m
->cgroups_agent_event_source
= sd_event_source_disable_unref(m
->cgroups_agent_event_source
);
1165 fd
= socket(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
|SOCK_NONBLOCK
, 0);
1167 return log_error_errno(errno
, "Failed to allocate cgroups agent socket: %m");
1169 fd_increase_rxbuf(fd
, CGROUPS_AGENT_RCVBUF_SIZE
);
1171 (void) sockaddr_un_unlink(&sa
.un
);
1173 /* Only allow root to connect to this socket */
1175 r
= bind(fd
, &sa
.sa
, SOCKADDR_UN_LEN(sa
.un
));
1177 return log_error_errno(errno
, "bind(%s) failed: %m", sa
.un
.sun_path
);
1179 m
->cgroups_agent_fd
= TAKE_FD(fd
);
1182 if (!m
->cgroups_agent_event_source
) {
1183 r
= sd_event_add_io(m
->event
, &m
->cgroups_agent_event_source
, m
->cgroups_agent_fd
, EPOLLIN
, manager_dispatch_cgroups_agent_fd
, m
);
1185 return log_error_errno(r
, "Failed to allocate cgroups agent event source: %m");
1187 /* Process cgroups notifications early. Note that when the agent notification is received
1188 * we'll just enqueue the unit in the cgroup empty queue, hence pick a high priority than
1189 * that. Also see handling of cgroup inotify for the unified cgroup stuff. */
1190 r
= sd_event_source_set_priority(m
->cgroups_agent_event_source
, SD_EVENT_PRIORITY_NORMAL
-9);
1192 return log_error_errno(r
, "Failed to set priority of cgroups agent event source: %m");
1194 (void) sd_event_source_set_description(m
->cgroups_agent_event_source
, "manager-cgroups-agent");
1200 static int manager_setup_user_lookup_fd(Manager
*m
) {
1205 /* Set up the socket pair used for passing UID/GID resolution results from forked off processes to PID
1206 * 1. Background: we can't do name lookups (NSS) from PID 1, since it might involve IPC and thus activation,
1207 * and we might hence deadlock on ourselves. Hence we do all user/group lookups asynchronously from the forked
1208 * off processes right before executing the binaries to start. In order to be able to clean up any IPC objects
1209 * created by a unit (see RemoveIPC=) we need to know in PID 1 the used UID/GID of the executed processes,
1210 * hence we establish this communication channel so that forked off processes can pass their UID/GID
1211 * information back to PID 1. The forked off processes send their resolved UID/GID to PID 1 in a simple
1212 * datagram, along with their unit name, so that we can share one communication socket pair among all units for
1215 * You might wonder why we need a communication channel for this that is independent of the usual notification
1216 * socket scheme (i.e. $NOTIFY_SOCKET). The primary difference is about trust: data sent via the $NOTIFY_SOCKET
1217 * channel is only accepted if it originates from the right unit and if reception was enabled for it. The user
1218 * lookup socket OTOH is only accessible by PID 1 and its children until they exec(), and always available.
1220 * Note that this function is called under two circumstances: when we first initialize (in which case we
1221 * allocate both the socket pair and the event source to listen on it), and when we deserialize after a reload
1222 * (in which case the socket pair already exists but we still need to allocate the event source for it). */
1224 if (m
->user_lookup_fds
[0] < 0) {
1226 /* Free all secondary fields */
1227 safe_close_pair(m
->user_lookup_fds
);
1228 m
->user_lookup_event_source
= sd_event_source_disable_unref(m
->user_lookup_event_source
);
1230 if (socketpair(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
, 0, m
->user_lookup_fds
) < 0)
1231 return log_error_errno(errno
, "Failed to allocate user lookup socket: %m");
1233 (void) fd_increase_rxbuf(m
->user_lookup_fds
[0], NOTIFY_RCVBUF_SIZE
);
1236 if (!m
->user_lookup_event_source
) {
1237 r
= sd_event_add_io(m
->event
, &m
->user_lookup_event_source
, m
->user_lookup_fds
[0], EPOLLIN
, manager_dispatch_user_lookup_fd
, m
);
1239 return log_error_errno(errno
, "Failed to allocate user lookup event source: %m");
1241 /* Process even earlier than the notify event source, so that we always know first about valid UID/GID
1243 r
= sd_event_source_set_priority(m
->user_lookup_event_source
, SD_EVENT_PRIORITY_NORMAL
-11);
1245 return log_error_errno(errno
, "Failed to set priority of user lookup event source: %m");
1247 (void) sd_event_source_set_description(m
->user_lookup_event_source
, "user-lookup");
1253 static unsigned manager_dispatch_cleanup_queue(Manager
*m
) {
1259 while ((u
= m
->cleanup_queue
)) {
1260 assert(u
->in_cleanup_queue
);
1269 static unsigned manager_dispatch_release_resources_queue(Manager
*m
) {
1275 while ((u
= LIST_POP(release_resources_queue
, m
->release_resources_queue
))) {
1276 assert(u
->in_release_resources_queue
);
1277 u
->in_release_resources_queue
= false;
1281 unit_release_resources(u
);
1288 GC_OFFSET_IN_PATH
, /* This one is on the path we were traveling */
1289 GC_OFFSET_UNSURE
, /* No clue */
1290 GC_OFFSET_GOOD
, /* We still need this unit */
1291 GC_OFFSET_BAD
, /* We don't need this unit anymore */
1295 static void unit_gc_mark_good(Unit
*u
, unsigned gc_marker
) {
1298 u
->gc_marker
= gc_marker
+ GC_OFFSET_GOOD
;
1300 /* Recursively mark referenced units as GOOD as well */
1301 UNIT_FOREACH_DEPENDENCY(other
, u
, UNIT_ATOM_REFERENCES
)
1302 if (other
->gc_marker
== gc_marker
+ GC_OFFSET_UNSURE
)
1303 unit_gc_mark_good(other
, gc_marker
);
1306 static void unit_gc_sweep(Unit
*u
, unsigned gc_marker
) {
1312 if (IN_SET(u
->gc_marker
- gc_marker
,
1313 GC_OFFSET_GOOD
, GC_OFFSET_BAD
, GC_OFFSET_UNSURE
, GC_OFFSET_IN_PATH
))
1316 if (u
->in_cleanup_queue
)
1319 if (!unit_may_gc(u
))
1322 u
->gc_marker
= gc_marker
+ GC_OFFSET_IN_PATH
;
1326 UNIT_FOREACH_DEPENDENCY(other
, u
, UNIT_ATOM_REFERENCED_BY
) {
1327 unit_gc_sweep(other
, gc_marker
);
1329 if (other
->gc_marker
== gc_marker
+ GC_OFFSET_GOOD
)
1332 if (other
->gc_marker
!= gc_marker
+ GC_OFFSET_BAD
)
1336 LIST_FOREACH(refs_by_target
, ref
, u
->refs_by_target
) {
1337 unit_gc_sweep(ref
->source
, gc_marker
);
1339 if (ref
->source
->gc_marker
== gc_marker
+ GC_OFFSET_GOOD
)
1342 if (ref
->source
->gc_marker
!= gc_marker
+ GC_OFFSET_BAD
)
1349 /* We were unable to find anything out about this entry, so
1350 * let's investigate it later */
1351 u
->gc_marker
= gc_marker
+ GC_OFFSET_UNSURE
;
1352 unit_add_to_gc_queue(u
);
1356 /* We definitely know that this one is not useful anymore, so
1357 * let's mark it for deletion */
1358 u
->gc_marker
= gc_marker
+ GC_OFFSET_BAD
;
1359 unit_add_to_cleanup_queue(u
);
1363 unit_gc_mark_good(u
, gc_marker
);
1366 static unsigned manager_dispatch_gc_unit_queue(Manager
*m
) {
1367 unsigned n
= 0, gc_marker
;
1372 /* log_debug("Running GC..."); */
1374 m
->gc_marker
+= _GC_OFFSET_MAX
;
1375 if (m
->gc_marker
+ _GC_OFFSET_MAX
<= _GC_OFFSET_MAX
)
1378 gc_marker
= m
->gc_marker
;
1380 while ((u
= LIST_POP(gc_queue
, m
->gc_unit_queue
))) {
1381 assert(u
->in_gc_queue
);
1383 unit_gc_sweep(u
, gc_marker
);
1385 u
->in_gc_queue
= false;
1389 if (IN_SET(u
->gc_marker
- gc_marker
,
1390 GC_OFFSET_BAD
, GC_OFFSET_UNSURE
)) {
1392 log_unit_debug(u
, "Collecting.");
1393 u
->gc_marker
= gc_marker
+ GC_OFFSET_BAD
;
1394 unit_add_to_cleanup_queue(u
);
1401 static unsigned manager_dispatch_gc_job_queue(Manager
*m
) {
1407 while ((j
= LIST_POP(gc_queue
, m
->gc_job_queue
))) {
1408 assert(j
->in_gc_queue
);
1409 j
->in_gc_queue
= false;
1416 log_unit_debug(j
->unit
, "Collecting job.");
1417 (void) job_finish_and_invalidate(j
, JOB_COLLECTED
, false, false);
1423 static int manager_ratelimit_requeue(sd_event_source
*s
, uint64_t usec
, void *userdata
) {
1427 assert(s
== u
->auto_start_stop_event_source
);
1429 u
->auto_start_stop_event_source
= sd_event_source_unref(u
->auto_start_stop_event_source
);
1431 /* Re-queue to all queues, if the rate limit hit we might have been throttled on any of them. */
1432 unit_submit_to_stop_when_unneeded_queue(u
);
1433 unit_submit_to_start_when_upheld_queue(u
);
1434 unit_submit_to_stop_when_bound_queue(u
);
1439 static int manager_ratelimit_check_and_queue(Unit
*u
) {
1444 if (ratelimit_below(&u
->auto_start_stop_ratelimit
))
1447 /* Already queued, no need to requeue */
1448 if (u
->auto_start_stop_event_source
)
1451 r
= sd_event_add_time(
1453 &u
->auto_start_stop_event_source
,
1455 ratelimit_end(&u
->auto_start_stop_ratelimit
),
1457 manager_ratelimit_requeue
,
1460 return log_unit_error_errno(u
, r
, "Failed to queue timer on event loop: %m");
1465 static unsigned manager_dispatch_stop_when_unneeded_queue(Manager
*m
) {
1472 while ((u
= LIST_POP(stop_when_unneeded_queue
, m
->stop_when_unneeded_queue
))) {
1473 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
1475 assert(u
->in_stop_when_unneeded_queue
);
1476 u
->in_stop_when_unneeded_queue
= false;
1480 if (!unit_is_unneeded(u
))
1483 log_unit_debug(u
, "Unit is not needed anymore.");
1485 /* If stopping a unit fails continuously we might enter a stop loop here, hence stop acting on the
1486 * service being unnecessary after a while. */
1488 r
= manager_ratelimit_check_and_queue(u
);
1491 "Unit not needed anymore, but not stopping since we tried this too often recently.%s",
1492 r
== 0 ? " Will retry later." : "");
1496 /* Ok, nobody needs us anymore. Sniff. Then let's commit suicide */
1497 r
= manager_add_job(u
->manager
, JOB_STOP
, u
, JOB_FAIL
, NULL
, &error
, NULL
);
1499 log_unit_warning_errno(u
, r
, "Failed to enqueue stop job, ignoring: %s", bus_error_message(&error
, r
));
1505 static unsigned manager_dispatch_start_when_upheld_queue(Manager
*m
) {
1512 while ((u
= LIST_POP(start_when_upheld_queue
, m
->start_when_upheld_queue
))) {
1513 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
1514 Unit
*culprit
= NULL
;
1516 assert(u
->in_start_when_upheld_queue
);
1517 u
->in_start_when_upheld_queue
= false;
1521 if (!unit_is_upheld_by_active(u
, &culprit
))
1524 log_unit_debug(u
, "Unit is started because upheld by active unit %s.", culprit
->id
);
1526 /* If stopping a unit fails continuously we might enter a stop loop here, hence stop acting on the
1527 * service being unnecessary after a while. */
1529 r
= manager_ratelimit_check_and_queue(u
);
1532 "Unit needs to be started because active unit %s upholds it, but not starting since we tried this too often recently.%s",
1534 r
== 0 ? " Will retry later." : "");
1538 r
= manager_add_job(u
->manager
, JOB_START
, u
, JOB_FAIL
, NULL
, &error
, NULL
);
1540 log_unit_warning_errno(u
, r
, "Failed to enqueue start job, ignoring: %s", bus_error_message(&error
, r
));
1546 static unsigned manager_dispatch_stop_when_bound_queue(Manager
*m
) {
1553 while ((u
= LIST_POP(stop_when_bound_queue
, m
->stop_when_bound_queue
))) {
1554 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
1555 Unit
*culprit
= NULL
;
1557 assert(u
->in_stop_when_bound_queue
);
1558 u
->in_stop_when_bound_queue
= false;
1562 if (!unit_is_bound_by_inactive(u
, &culprit
))
1565 log_unit_debug(u
, "Unit is stopped because bound to inactive unit %s.", culprit
->id
);
1567 /* If stopping a unit fails continuously we might enter a stop loop here, hence stop acting on the
1568 * service being unnecessary after a while. */
1570 r
= manager_ratelimit_check_and_queue(u
);
1573 "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",
1575 r
== 0 ? " Will retry later." : "");
1579 r
= manager_add_job(u
->manager
, JOB_STOP
, u
, JOB_REPLACE
, NULL
, &error
, NULL
);
1581 log_unit_warning_errno(u
, r
, "Failed to enqueue stop job, ignoring: %s", bus_error_message(&error
, r
));
1587 static void manager_clear_jobs_and_units(Manager
*m
) {
1592 while ((u
= hashmap_first(m
->units
)))
1595 manager_dispatch_cleanup_queue(m
);
1597 assert(!m
->load_queue
);
1598 assert(prioq_isempty(m
->run_queue
));
1599 assert(!m
->dbus_unit_queue
);
1600 assert(!m
->dbus_job_queue
);
1601 assert(!m
->cleanup_queue
);
1602 assert(!m
->gc_unit_queue
);
1603 assert(!m
->gc_job_queue
);
1604 assert(!m
->cgroup_realize_queue
);
1605 assert(!m
->cgroup_empty_queue
);
1606 assert(!m
->cgroup_oom_queue
);
1607 assert(!m
->target_deps_queue
);
1608 assert(!m
->stop_when_unneeded_queue
);
1609 assert(!m
->start_when_upheld_queue
);
1610 assert(!m
->stop_when_bound_queue
);
1611 assert(!m
->release_resources_queue
);
1613 assert(hashmap_isempty(m
->jobs
));
1614 assert(hashmap_isempty(m
->units
));
1616 m
->n_on_console
= 0;
1617 m
->n_running_jobs
= 0;
1618 m
->n_installed_jobs
= 0;
1619 m
->n_failed_jobs
= 0;
1622 Manager
* manager_free(Manager
*m
) {
1626 manager_clear_jobs_and_units(m
);
1628 for (UnitType c
= 0; c
< _UNIT_TYPE_MAX
; c
++)
1629 if (unit_vtable
[c
]->shutdown
)
1630 unit_vtable
[c
]->shutdown(m
);
1632 /* Keep the cgroup hierarchy in place except when we know we are going down for good */
1633 manager_shutdown_cgroup(m
, /* delete= */ IN_SET(m
->objective
, MANAGER_EXIT
, MANAGER_REBOOT
, MANAGER_POWEROFF
, MANAGER_HALT
, MANAGER_KEXEC
));
1635 lookup_paths_flush_generator(&m
->lookup_paths
);
1638 manager_varlink_done(m
);
1640 exec_shared_runtime_vacuum(m
);
1641 hashmap_free(m
->exec_shared_runtime_by_id
);
1643 dynamic_user_vacuum(m
, false);
1644 hashmap_free(m
->dynamic_users
);
1646 hashmap_free(m
->units
);
1647 hashmap_free(m
->units_by_invocation_id
);
1648 hashmap_free(m
->jobs
);
1649 hashmap_free(m
->watch_pids
);
1650 hashmap_free(m
->watch_pids_more
);
1651 hashmap_free(m
->watch_bus
);
1653 prioq_free(m
->run_queue
);
1655 set_free(m
->startup_units
);
1656 set_free(m
->failed_units
);
1658 sd_event_source_unref(m
->signal_event_source
);
1659 sd_event_source_unref(m
->sigchld_event_source
);
1660 sd_event_source_unref(m
->notify_event_source
);
1661 sd_event_source_unref(m
->cgroups_agent_event_source
);
1662 sd_event_source_unref(m
->time_change_event_source
);
1663 sd_event_source_unref(m
->timezone_change_event_source
);
1664 sd_event_source_unref(m
->jobs_in_progress_event_source
);
1665 sd_event_source_unref(m
->run_queue_event_source
);
1666 sd_event_source_unref(m
->user_lookup_event_source
);
1667 sd_event_source_unref(m
->memory_pressure_event_source
);
1669 safe_close(m
->signal_fd
);
1670 safe_close(m
->notify_fd
);
1671 safe_close(m
->cgroups_agent_fd
);
1672 safe_close_pair(m
->user_lookup_fds
);
1674 manager_close_ask_password(m
);
1676 manager_close_idle_pipe(m
);
1678 sd_event_unref(m
->event
);
1680 free(m
->notify_socket
);
1682 lookup_paths_free(&m
->lookup_paths
);
1683 strv_free(m
->transient_environment
);
1684 strv_free(m
->client_environment
);
1686 hashmap_free(m
->cgroup_unit
);
1687 manager_free_unit_name_maps(m
);
1689 free(m
->switch_root
);
1690 free(m
->switch_root_init
);
1692 unit_defaults_done(&m
->defaults
);
1694 FOREACH_ARRAY(map
, m
->units_needing_mounts_for
, _UNIT_MOUNT_DEPENDENCY_TYPE_MAX
) {
1695 assert(hashmap_isempty(*map
));
1699 hashmap_free(m
->uid_refs
);
1700 hashmap_free(m
->gid_refs
);
1702 for (ExecDirectoryType dt
= 0; dt
< _EXEC_DIRECTORY_TYPE_MAX
; dt
++)
1703 m
->prefix
[dt
] = mfree(m
->prefix
[dt
]);
1704 free(m
->received_credentials_directory
);
1705 free(m
->received_encrypted_credentials_directory
);
1707 free(m
->watchdog_pretimeout_governor
);
1708 free(m
->watchdog_pretimeout_governor_overridden
);
1710 m
->fw_ctx
= fw_ctx_free(m
->fw_ctx
);
1713 lsm_bpf_destroy(m
->restrict_fs
);
1716 safe_close(m
->executor_fd
);
1721 static void manager_enumerate_perpetual(Manager
*m
) {
1724 if (FLAGS_SET(m
->test_run_flags
, MANAGER_TEST_RUN_MINIMAL
))
1727 /* Let's ask every type to load all units from disk/kernel that it might know */
1728 for (UnitType c
= 0; c
< _UNIT_TYPE_MAX
; c
++) {
1729 if (!unit_type_supported(c
)) {
1730 log_debug("Unit type .%s is not supported on this system.", unit_type_to_string(c
));
1734 if (unit_vtable
[c
]->enumerate_perpetual
)
1735 unit_vtable
[c
]->enumerate_perpetual(m
);
1739 static void manager_enumerate(Manager
*m
) {
1742 if (FLAGS_SET(m
->test_run_flags
, MANAGER_TEST_RUN_MINIMAL
))
1745 /* Let's ask every type to load all units from disk/kernel that it might know */
1746 for (UnitType c
= 0; c
< _UNIT_TYPE_MAX
; c
++) {
1747 if (!unit_type_supported(c
)) {
1748 log_debug("Unit type .%s is not supported on this system.", unit_type_to_string(c
));
1752 if (unit_vtable
[c
]->enumerate
)
1753 unit_vtable
[c
]->enumerate(m
);
1756 manager_dispatch_load_queue(m
);
1759 static void manager_coldplug(Manager
*m
) {
1766 log_debug("Invoking unit coldplug() handlers%s", special_glyph(SPECIAL_GLYPH_ELLIPSIS
));
1768 /* Let's place the units back into their deserialized state */
1769 HASHMAP_FOREACH_KEY(u
, k
, m
->units
) {
1771 /* ignore aliases */
1775 r
= unit_coldplug(u
);
1777 log_warning_errno(r
, "We couldn't coldplug %s, proceeding anyway: %m", u
->id
);
1781 static void manager_catchup(Manager
*m
) {
1787 log_debug("Invoking unit catchup() handlers%s", special_glyph(SPECIAL_GLYPH_ELLIPSIS
));
1789 /* Let's catch up on any state changes that happened while we were reloading/reexecing */
1790 HASHMAP_FOREACH_KEY(u
, k
, m
->units
) {
1792 /* ignore aliases */
1800 static void manager_distribute_fds(Manager
*m
, FDSet
*fds
) {
1805 HASHMAP_FOREACH(u
, m
->units
) {
1807 if (fdset_size(fds
) <= 0)
1810 if (!UNIT_VTABLE(u
)->distribute_fds
)
1813 UNIT_VTABLE(u
)->distribute_fds(u
, fds
);
1817 static bool manager_dbus_is_running(Manager
*m
, bool deserialized
) {
1822 /* This checks whether the dbus instance we are supposed to expose our APIs on is up. We check both the socket
1823 * and the service unit. If the 'deserialized' parameter is true we'll check the deserialized state of the unit
1824 * rather than the current one. */
1826 if (MANAGER_IS_TEST_RUN(m
))
1829 u
= manager_get_unit(m
, SPECIAL_DBUS_SOCKET
);
1832 if ((deserialized
? SOCKET(u
)->deserialized_state
: SOCKET(u
)->state
) != SOCKET_RUNNING
)
1835 u
= manager_get_unit(m
, SPECIAL_DBUS_SERVICE
);
1838 if (!IN_SET((deserialized
? SERVICE(u
)->deserialized_state
: SERVICE(u
)->state
),
1841 SERVICE_RELOAD_NOTIFY
,
1842 SERVICE_RELOAD_SIGNAL
))
1848 static void manager_setup_bus(Manager
*m
) {
1851 /* Let's set up our private bus connection now, unconditionally */
1852 (void) bus_init_private(m
);
1854 /* If we are in --user mode also connect to the system bus now */
1855 if (MANAGER_IS_USER(m
))
1856 (void) bus_init_system(m
);
1858 /* Let's connect to the bus now, but only if the unit is supposed to be up */
1859 if (manager_dbus_is_running(m
, MANAGER_IS_RELOADING(m
))) {
1860 (void) bus_init_api(m
);
1862 if (MANAGER_IS_SYSTEM(m
))
1863 (void) bus_init_system(m
);
1867 static void manager_preset_all(Manager
*m
) {
1872 if (m
->first_boot
<= 0)
1875 if (!MANAGER_IS_SYSTEM(m
))
1878 if (MANAGER_IS_TEST_RUN(m
))
1881 /* If this is the first boot, and we are in the host system, then preset everything */
1882 UnitFilePresetMode mode
=
1883 ENABLE_FIRST_BOOT_FULL_PRESET
? UNIT_FILE_PRESET_FULL
: UNIT_FILE_PRESET_ENABLE_ONLY
;
1885 r
= unit_file_preset_all(RUNTIME_SCOPE_SYSTEM
, 0, NULL
, mode
, NULL
, 0);
1887 log_full_errno(r
== -EEXIST
? LOG_NOTICE
: LOG_WARNING
, r
,
1888 "Failed to populate /etc with preset unit settings, ignoring: %m");
1890 log_info("Populated /etc with preset unit settings.");
1893 static void manager_ready(Manager
*m
) {
1896 /* After having loaded everything, do the final round of catching up with what might have changed */
1898 m
->objective
= MANAGER_OK
; /* Tell everyone we are up now */
1900 /* It might be safe to log to the journal now and connect to dbus */
1901 manager_recheck_journal(m
);
1902 manager_recheck_dbus(m
);
1904 /* Let's finally catch up with any changes that took place while we were reloading/reexecing */
1907 /* Create a file which will indicate when the manager started loading units the last time. */
1908 if (MANAGER_IS_SYSTEM(m
))
1909 (void) touch_file("/run/systemd/systemd-units-load", false,
1910 m
->timestamps
[MANAGER_TIMESTAMP_UNITS_LOAD
].realtime
?: now(CLOCK_REALTIME
),
1911 UID_INVALID
, GID_INVALID
, 0444);
1914 Manager
* manager_reloading_start(Manager
*m
) {
1916 dual_timestamp_now(m
->timestamps
+ MANAGER_TIMESTAMP_UNITS_LOAD
);
1920 void manager_reloading_stopp(Manager
**m
) {
1922 assert((*m
)->n_reloading
> 0);
1923 (*m
)->n_reloading
--;
1927 int manager_startup(Manager
*m
, FILE *serialization
, FDSet
*fds
, const char *root
) {
1932 /* If we are running in test mode, we still want to run the generators,
1933 * but we should not touch the real generator directories. */
1934 r
= lookup_paths_init_or_warn(&m
->lookup_paths
, m
->runtime_scope
,
1935 MANAGER_IS_TEST_RUN(m
) ? LOOKUP_PATHS_TEMPORARY_GENERATED
: 0,
1940 dual_timestamp_now(m
->timestamps
+ manager_timestamp_initrd_mangle(MANAGER_TIMESTAMP_GENERATORS_START
));
1941 r
= manager_run_environment_generators(m
);
1943 r
= manager_run_generators(m
);
1944 dual_timestamp_now(m
->timestamps
+ manager_timestamp_initrd_mangle(MANAGER_TIMESTAMP_GENERATORS_FINISH
));
1948 manager_preset_all(m
);
1950 lookup_paths_log(&m
->lookup_paths
);
1953 /* This block is (optionally) done with the reloading counter bumped */
1954 _unused_
_cleanup_(manager_reloading_stopp
) Manager
*reloading
= NULL
;
1956 /* Make sure we don't have a left-over from a previous run */
1958 (void) rm_rf(m
->lookup_paths
.transient
, 0);
1960 /* If we will deserialize make sure that during enumeration this is already known, so we increase the
1961 * counter here already */
1963 reloading
= manager_reloading_start(m
);
1965 /* First, enumerate what we can from all config files */
1966 dual_timestamp_now(m
->timestamps
+ manager_timestamp_initrd_mangle(MANAGER_TIMESTAMP_UNITS_LOAD_START
));
1967 manager_enumerate_perpetual(m
);
1968 manager_enumerate(m
);
1969 dual_timestamp_now(m
->timestamps
+ manager_timestamp_initrd_mangle(MANAGER_TIMESTAMP_UNITS_LOAD_FINISH
));
1971 /* Second, deserialize if there is something to deserialize */
1972 if (serialization
) {
1973 r
= manager_deserialize(m
, serialization
, fds
);
1975 return log_error_errno(r
, "Deserialization failed: %m");
1978 /* Any fds left? Find some unit which wants them. This is useful to allow container managers to pass
1979 * some file descriptors to us pre-initialized. This enables socket-based activation of entire
1981 manager_distribute_fds(m
, fds
);
1983 /* We might have deserialized the notify fd, but if we didn't then let's create the bus now */
1984 r
= manager_setup_notify(m
);
1986 /* No sense to continue without notifications, our children would fail anyway. */
1989 r
= manager_setup_cgroups_agent(m
);
1991 /* Likewise, no sense to continue without empty cgroup notifications. */
1994 r
= manager_setup_user_lookup_fd(m
);
1996 /* This shouldn't fail, except if things are really broken. */
1999 /* Connect to the bus if we are good for it */
2000 manager_setup_bus(m
);
2002 /* Now that we are connected to all possible buses, let's deserialize who is tracking us. */
2003 r
= bus_track_coldplug(m
, &m
->subscribed
, false, m
->deserialized_subscribed
);
2005 log_warning_errno(r
, "Failed to deserialized tracked clients, ignoring: %m");
2006 m
->deserialized_subscribed
= strv_free(m
->deserialized_subscribed
);
2008 r
= manager_varlink_init(m
);
2010 log_warning_errno(r
, "Failed to set up Varlink, ignoring: %m");
2012 /* Third, fire things up! */
2013 manager_coldplug(m
);
2015 /* Clean up runtime objects */
2019 /* Let's wait for the UnitNew/JobNew messages being sent, before we notify that the
2020 * reload is finished */
2021 m
->send_reloading_done
= true;
2026 manager_set_switching_root(m
, false);
2031 int manager_add_job(
2037 sd_bus_error
*error
,
2040 _cleanup_(transaction_abort_and_freep
) Transaction
*tr
= NULL
;
2044 assert(type
< _JOB_TYPE_MAX
);
2046 assert(mode
< _JOB_MODE_MAX
);
2048 if (mode
== JOB_ISOLATE
&& type
!= JOB_START
)
2049 return sd_bus_error_set(error
, SD_BUS_ERROR_INVALID_ARGS
, "Isolate is only valid for start.");
2051 if (mode
== JOB_ISOLATE
&& !unit
->allow_isolate
)
2052 return sd_bus_error_set(error
, BUS_ERROR_NO_ISOLATION
, "Operation refused, unit may not be isolated.");
2054 if (mode
== JOB_TRIGGERING
&& type
!= JOB_STOP
)
2055 return sd_bus_error_set(error
, SD_BUS_ERROR_INVALID_ARGS
, "--job-mode=triggering is only valid for stop.");
2057 if (mode
== JOB_RESTART_DEPENDENCIES
&& type
!= JOB_START
)
2058 return sd_bus_error_set(error
, SD_BUS_ERROR_INVALID_ARGS
, "--job-mode=restart-dependencies is only valid for start.");
2060 log_unit_debug(unit
, "Trying to enqueue job %s/%s/%s", unit
->id
, job_type_to_string(type
), job_mode_to_string(mode
));
2062 type
= job_type_collapse(type
, unit
);
2064 tr
= transaction_new(mode
== JOB_REPLACE_IRREVERSIBLY
);
2068 r
= transaction_add_job_and_dependencies(
2073 TRANSACTION_MATTERS
|
2074 (IN_SET(mode
, JOB_IGNORE_DEPENDENCIES
, JOB_IGNORE_REQUIREMENTS
) ? TRANSACTION_IGNORE_REQUIREMENTS
: 0) |
2075 (mode
== JOB_IGNORE_DEPENDENCIES
? TRANSACTION_IGNORE_ORDER
: 0) |
2076 (mode
== JOB_RESTART_DEPENDENCIES
? TRANSACTION_PROPAGATE_START_AS_RESTART
: 0),
2081 if (mode
== JOB_ISOLATE
) {
2082 r
= transaction_add_isolate_jobs(tr
, m
);
2087 if (mode
== JOB_TRIGGERING
) {
2088 r
= transaction_add_triggering_jobs(tr
, unit
);
2093 r
= transaction_activate(tr
, m
, mode
, affected_jobs
, error
);
2097 log_unit_debug(unit
,
2098 "Enqueued job %s/%s as %u", unit
->id
,
2099 job_type_to_string(type
), (unsigned) tr
->anchor_job
->id
);
2102 *ret
= tr
->anchor_job
;
2104 tr
= transaction_free(tr
);
2108 int manager_add_job_by_name(Manager
*m
, JobType type
, const char *name
, JobMode mode
, Set
*affected_jobs
, sd_bus_error
*e
, Job
**ret
) {
2109 Unit
*unit
= NULL
; /* just to appease gcc, initialization is not really necessary */
2113 assert(type
< _JOB_TYPE_MAX
);
2115 assert(mode
< _JOB_MODE_MAX
);
2117 r
= manager_load_unit(m
, name
, NULL
, NULL
, &unit
);
2122 return manager_add_job(m
, type
, unit
, mode
, affected_jobs
, e
, ret
);
2125 int manager_add_job_by_name_and_warn(Manager
*m
, JobType type
, const char *name
, JobMode mode
, Set
*affected_jobs
, Job
**ret
) {
2126 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
2130 assert(type
< _JOB_TYPE_MAX
);
2132 assert(mode
< _JOB_MODE_MAX
);
2134 r
= manager_add_job_by_name(m
, type
, name
, mode
, affected_jobs
, &error
, ret
);
2136 return log_warning_errno(r
, "Failed to enqueue %s job for %s: %s", job_mode_to_string(mode
), name
, bus_error_message(&error
, r
));
2141 int manager_propagate_reload(Manager
*m
, Unit
*unit
, JobMode mode
, sd_bus_error
*e
) {
2143 _cleanup_(transaction_abort_and_freep
) Transaction
*tr
= NULL
;
2147 assert(mode
< _JOB_MODE_MAX
);
2148 assert(mode
!= JOB_ISOLATE
); /* Isolate is only valid for start */
2150 tr
= transaction_new(mode
== JOB_REPLACE_IRREVERSIBLY
);
2154 /* We need an anchor job */
2155 r
= transaction_add_job_and_dependencies(tr
, JOB_NOP
, unit
, NULL
, TRANSACTION_IGNORE_REQUIREMENTS
|TRANSACTION_IGNORE_ORDER
, e
);
2159 /* Failure in adding individual dependencies is ignored, so this always succeeds. */
2160 transaction_add_propagate_reload_jobs(
2164 mode
== JOB_IGNORE_DEPENDENCIES
? TRANSACTION_IGNORE_ORDER
: 0);
2166 r
= transaction_activate(tr
, m
, mode
, NULL
, e
);
2170 tr
= transaction_free(tr
);
2174 Job
*manager_get_job(Manager
*m
, uint32_t id
) {
2177 return hashmap_get(m
->jobs
, UINT32_TO_PTR(id
));
2180 Unit
*manager_get_unit(Manager
*m
, const char *name
) {
2184 return hashmap_get(m
->units
, name
);
2187 static int manager_dispatch_target_deps_queue(Manager
*m
) {
2193 while ((u
= LIST_POP(target_deps_queue
, m
->target_deps_queue
))) {
2194 _cleanup_free_ Unit
**targets
= NULL
;
2197 assert(u
->in_target_deps_queue
);
2199 u
->in_target_deps_queue
= false;
2201 /* Take an "atomic" snapshot of dependencies here, as the call below will likely modify the
2202 * dependencies, and we can't have it that hash tables we iterate through are modified while
2203 * we are iterating through them. */
2204 n_targets
= unit_get_dependency_array(u
, UNIT_ATOM_DEFAULT_TARGET_DEPENDENCIES
, &targets
);
2208 for (int i
= 0; i
< n_targets
; i
++) {
2209 r
= unit_add_default_target_dependency(u
, targets
[i
]);
2218 unsigned manager_dispatch_load_queue(Manager
*m
) {
2224 /* Make sure we are not run recursively */
2225 if (m
->dispatching_load_queue
)
2228 m
->dispatching_load_queue
= true;
2230 /* Dispatches the load queue. Takes a unit from the queue and
2231 * tries to load its data until the queue is empty */
2233 while ((u
= m
->load_queue
)) {
2234 assert(u
->in_load_queue
);
2240 m
->dispatching_load_queue
= false;
2242 /* Dispatch the units waiting for their target dependencies to be added now, as all targets that we know about
2243 * should be loaded and have aliases resolved */
2244 (void) manager_dispatch_target_deps_queue(m
);
2249 bool manager_unit_cache_should_retry_load(Unit
*u
) {
2252 /* Automatic reloading from disk only applies to units which were not found sometime in the past, and
2253 * the not-found stub is kept pinned in the unit graph by dependencies. For units that were
2254 * previously loaded, we don't do automatic reloading, and daemon-reload is necessary to update. */
2255 if (u
->load_state
!= UNIT_NOT_FOUND
)
2258 /* The cache has been updated since the last time we tried to load the unit. There might be new
2259 * fragment paths to read. */
2260 if (u
->manager
->unit_cache_timestamp_hash
!= u
->fragment_not_found_timestamp_hash
)
2263 /* The cache needs to be updated because there are modifications on disk. */
2264 return !lookup_paths_timestamp_hash_same(&u
->manager
->lookup_paths
, u
->manager
->unit_cache_timestamp_hash
, NULL
);
2267 int manager_load_unit_prepare(
2274 _cleanup_(unit_freep
) Unit
*cleanup_unit
= NULL
;
2275 _cleanup_free_
char *nbuf
= NULL
;
2280 assert(name
|| path
);
2282 /* This will prepare the unit for loading, but not actually load anything from disk. */
2284 if (path
&& !path_is_absolute(path
))
2285 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
, "Path %s is not absolute.", path
);
2288 r
= path_extract_filename(path
, &nbuf
);
2291 if (r
== O_DIRECTORY
)
2292 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
, "Path '%s' refers to directory, refusing.", path
);
2297 UnitType t
= unit_name_to_type(name
);
2299 if (t
== _UNIT_TYPE_INVALID
|| !unit_name_is_valid(name
, UNIT_NAME_PLAIN
|UNIT_NAME_INSTANCE
)) {
2300 if (unit_name_is_valid(name
, UNIT_NAME_TEMPLATE
))
2301 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
, "Unit name %s is missing the instance name.", name
);
2303 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
, "Unit name %s is not valid.", name
);
2306 Unit
*unit
= manager_get_unit(m
, name
);
2308 /* The time-based cache allows to start new units without daemon-reload,
2309 * but if they are already referenced (because of dependencies or ordering)
2310 * then we have to force a load of the fragment. As an optimization, check
2311 * first if anything in the usual paths was modified since the last time
2312 * the cache was loaded. Also check if the last time an attempt to load the
2313 * unit was made was before the most recent cache refresh, so that we know
2314 * we need to try again — even if the cache is current, it might have been
2315 * updated in a different context before we had a chance to retry loading
2316 * this particular unit. */
2317 if (manager_unit_cache_should_retry_load(unit
))
2318 unit
->load_state
= UNIT_STUB
;
2321 return 0; /* The unit was already loaded */
2324 unit
= cleanup_unit
= unit_new(m
, unit_vtable
[t
]->object_size
);
2330 r
= free_and_strdup(&unit
->fragment_path
, path
);
2335 r
= unit_add_name(unit
, name
);
2339 unit_add_to_load_queue(unit
);
2340 unit_add_to_dbus_queue(unit
);
2341 unit_add_to_gc_queue(unit
);
2344 TAKE_PTR(cleanup_unit
);
2346 return 1; /* The unit was added the load queue */
2349 int manager_load_unit(
2360 /* This will load the unit config, but not actually start any services or anything. */
2362 r
= manager_load_unit_prepare(m
, name
, path
, e
, ret
);
2366 /* Unit was newly loaded */
2367 manager_dispatch_load_queue(m
);
2368 *ret
= unit_follow_merge(*ret
);
2372 int manager_load_startable_unit_or_warn(
2378 /* Load a unit, make sure it loaded fully and is not masked. */
2380 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
2384 r
= manager_load_unit(m
, name
, path
, &error
, &unit
);
2386 return log_error_errno(r
, "Failed to load %s %s: %s",
2387 name
? "unit" : "unit file", name
?: path
,
2388 bus_error_message(&error
, r
));
2390 r
= bus_unit_validate_load_state(unit
, &error
);
2392 return log_error_errno(r
, "%s", bus_error_message(&error
, r
));
2398 void manager_clear_jobs(Manager
*m
) {
2403 while ((j
= hashmap_first(m
->jobs
)))
2404 /* No need to recurse. We're cancelling all jobs. */
2405 job_finish_and_invalidate(j
, JOB_CANCELED
, false, false);
2408 void manager_unwatch_pidref(Manager
*m
, PidRef
*pid
) {
2414 u
= manager_get_unit_by_pidref_watching(m
, pid
);
2418 unit_unwatch_pidref(u
, pid
);
2422 static int manager_dispatch_run_queue(sd_event_source
*source
, void *userdata
) {
2423 Manager
*m
= ASSERT_PTR(userdata
);
2428 while ((j
= prioq_peek(m
->run_queue
))) {
2429 assert(j
->installed
);
2430 assert(j
->in_run_queue
);
2432 (void) job_run_and_invalidate(j
);
2435 if (m
->n_running_jobs
> 0)
2436 manager_watch_jobs_in_progress(m
);
2438 if (m
->n_on_console
> 0)
2439 manager_watch_idle_pipe(m
);
2444 void manager_trigger_run_queue(Manager
*m
) {
2449 r
= sd_event_source_set_enabled(
2450 m
->run_queue_event_source
,
2451 prioq_isempty(m
->run_queue
) ? SD_EVENT_OFF
: SD_EVENT_ONESHOT
);
2453 log_warning_errno(r
, "Failed to enable job run queue event source, ignoring: %m");
2456 static unsigned manager_dispatch_dbus_queue(Manager
*m
) {
2457 unsigned n
= 0, budget
;
2463 /* When we are reloading, let's not wait with generating signals, since we need to exit the manager as quickly
2464 * as we can. There's no point in throttling generation of signals in that case. */
2465 if (MANAGER_IS_RELOADING(m
) || m
->send_reloading_done
|| m
->pending_reload_message
)
2466 budget
= UINT_MAX
; /* infinite budget in this case */
2468 /* Anything to do at all? */
2469 if (!m
->dbus_unit_queue
&& !m
->dbus_job_queue
)
2472 /* Do we have overly many messages queued at the moment? If so, let's not enqueue more on top, let's
2473 * sit this cycle out, and process things in a later cycle when the queues got a bit emptier. */
2474 if (manager_bus_n_queued_write(m
) > MANAGER_BUS_BUSY_THRESHOLD
)
2477 /* Only process a certain number of units/jobs per event loop iteration. Even if the bus queue wasn't
2478 * overly full before this call we shouldn't increase it in size too wildly in one step, and we
2479 * shouldn't monopolize CPU time with generating these messages. Note the difference in counting of
2480 * this "budget" and the "threshold" above: the "budget" is decreased only once per generated message,
2481 * regardless how many buses/direct connections it is enqueued on, while the "threshold" is applied to
2482 * each queued instance of bus message, i.e. if the same message is enqueued to five buses/direct
2483 * connections it will be counted five times. This difference in counting ("references"
2484 * vs. "instances") is primarily a result of the fact that it's easier to implement it this way,
2485 * however it also reflects the thinking that the "threshold" should put a limit on used queue memory,
2486 * i.e. space, while the "budget" should put a limit on time. Also note that the "threshold" is
2487 * currently chosen much higher than the "budget". */
2488 budget
= MANAGER_BUS_MESSAGE_BUDGET
;
2491 while (budget
!= 0 && (u
= m
->dbus_unit_queue
)) {
2493 assert(u
->in_dbus_queue
);
2495 bus_unit_send_change_signal(u
);
2498 if (budget
!= UINT_MAX
)
2502 while (budget
!= 0 && (j
= m
->dbus_job_queue
)) {
2503 assert(j
->in_dbus_queue
);
2505 bus_job_send_change_signal(j
);
2508 if (budget
!= UINT_MAX
)
2512 if (m
->send_reloading_done
) {
2513 m
->send_reloading_done
= false;
2514 bus_manager_send_reloading(m
, false);
2518 if (m
->pending_reload_message
) {
2519 bus_send_pending_reload_message(m
);
2526 static int manager_dispatch_cgroups_agent_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
2527 Manager
*m
= userdata
;
2531 n
= recv(fd
, buf
, sizeof(buf
), 0);
2533 return log_error_errno(errno
, "Failed to read cgroups agent message: %m");
2535 log_error("Got zero-length cgroups agent message, ignoring.");
2538 if ((size_t) n
>= sizeof(buf
)) {
2539 log_error("Got overly long cgroups agent message, ignoring.");
2543 if (memchr(buf
, 0, n
)) {
2544 log_error("Got cgroups agent message with embedded NUL byte, ignoring.");
2549 manager_notify_cgroup_empty(m
, buf
);
2550 (void) bus_forward_agent_released(m
, buf
);
2555 static bool manager_process_barrier_fd(char * const *tags
, FDSet
*fds
) {
2557 /* nothing else must be sent when using BARRIER=1 */
2558 if (strv_contains(tags
, "BARRIER=1")) {
2559 if (strv_length(tags
) != 1)
2560 log_warning("Extra notification messages sent with BARRIER=1, ignoring everything.");
2561 else if (fdset_size(fds
) != 1)
2562 log_warning("Got incorrect number of fds with BARRIER=1, closing them.");
2564 /* Drop the message if BARRIER=1 was found */
2571 static void manager_invoke_notify_message(
2574 const struct ucred
*ucred
,
2583 if (u
->notifygen
== m
->notifygen
) /* Already invoked on this same unit in this same iteration? */
2585 u
->notifygen
= m
->notifygen
;
2587 if (UNIT_VTABLE(u
)->notify_message
)
2588 UNIT_VTABLE(u
)->notify_message(u
, ucred
, tags
, fds
);
2590 else if (DEBUG_LOGGING
) {
2591 _cleanup_free_
char *buf
= NULL
, *x
= NULL
, *y
= NULL
;
2593 buf
= strv_join(tags
, ", ");
2595 x
= ellipsize(buf
, 20, 90);
2599 log_unit_debug(u
, "Got notification message \"%s\", ignoring.", strnull(y
));
2603 static int manager_dispatch_notify_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
2605 _cleanup_fdset_free_ FDSet
*fds
= NULL
;
2606 Manager
*m
= ASSERT_PTR(userdata
);
2607 char buf
[NOTIFY_BUFFER_MAX
+1];
2608 struct iovec iovec
= {
2610 .iov_len
= sizeof(buf
)-1,
2612 CMSG_BUFFER_TYPE(CMSG_SPACE(sizeof(struct ucred
)) +
2613 CMSG_SPACE(sizeof(int) * NOTIFY_FD_MAX
)) control
;
2614 struct msghdr msghdr
= {
2617 .msg_control
= &control
,
2618 .msg_controllen
= sizeof(control
),
2621 struct cmsghdr
*cmsg
;
2622 struct ucred
*ucred
= NULL
;
2623 _cleanup_free_ Unit
**array_copy
= NULL
;
2624 _cleanup_strv_free_
char **tags
= NULL
;
2625 Unit
*u1
, *u2
, **array
;
2626 int r
, *fd_array
= NULL
;
2631 assert(m
->notify_fd
== fd
);
2633 if (revents
!= EPOLLIN
) {
2634 log_warning("Got unexpected poll event for notify fd.");
2638 n
= recvmsg_safe(m
->notify_fd
, &msghdr
, MSG_DONTWAIT
|MSG_CMSG_CLOEXEC
|MSG_TRUNC
);
2639 if (ERRNO_IS_NEG_TRANSIENT(n
))
2640 return 0; /* Spurious wakeup, try again */
2642 log_warning("Got message with truncated control data (too many fds sent?), ignoring.");
2646 /* If this is any other, real error, then stop processing this socket. This of course means
2647 * we won't take notification messages anymore, but that's still better than busy looping:
2648 * being woken up over and over again, but being unable to actually read the message from the
2650 return log_error_errno(n
, "Failed to receive notification message: %m");
2652 CMSG_FOREACH(cmsg
, &msghdr
)
2653 if (cmsg
->cmsg_level
== SOL_SOCKET
&& cmsg
->cmsg_type
== SCM_RIGHTS
) {
2656 fd_array
= CMSG_TYPED_DATA(cmsg
, int);
2657 n_fds
= (cmsg
->cmsg_len
- CMSG_LEN(0)) / sizeof(int);
2659 } else if (cmsg
->cmsg_level
== SOL_SOCKET
&&
2660 cmsg
->cmsg_type
== SCM_CREDENTIALS
&&
2661 cmsg
->cmsg_len
== CMSG_LEN(sizeof(struct ucred
))) {
2664 ucred
= CMSG_TYPED_DATA(cmsg
, struct ucred
);
2670 r
= fdset_new_array(&fds
, fd_array
, n_fds
);
2672 close_many(fd_array
, n_fds
);
2678 if (!ucred
|| !pid_is_valid(ucred
->pid
)) {
2679 log_warning("Received notify message without valid credentials. Ignoring.");
2683 if ((size_t) n
>= sizeof(buf
) || (msghdr
.msg_flags
& MSG_TRUNC
)) {
2684 log_warning("Received notify message exceeded maximum size. Ignoring.");
2688 /* As extra safety check, let's make sure the string we get doesn't contain embedded NUL bytes.
2689 * We permit one trailing NUL byte in the message, but don't expect it. */
2690 if (n
> 1 && memchr(buf
, 0, n
-1)) {
2691 log_warning("Received notify message with embedded NUL bytes. Ignoring.");
2695 /* Make sure it's NUL-terminated, then parse it to obtain the tags list. */
2697 tags
= strv_split_newlines(buf
);
2703 /* Possibly a barrier fd, let's see. */
2704 if (manager_process_barrier_fd(tags
, fds
)) {
2705 log_debug("Received barrier notification message from PID " PID_FMT
".", ucred
->pid
);
2709 /* Increase the generation counter used for filtering out duplicate unit invocations. */
2712 /* Generate lookup key from the PID (we have no pidfd here, after all) */
2713 PidRef pidref
= PIDREF_MAKE_FROM_PID(ucred
->pid
);
2715 /* Notify every unit that might be interested, which might be multiple. */
2716 u1
= manager_get_unit_by_pidref_cgroup(m
, &pidref
);
2717 u2
= hashmap_get(m
->watch_pids
, &pidref
);
2718 array
= hashmap_get(m
->watch_pids_more
, &pidref
);
2725 array_copy
= newdup(Unit
*, array
, k
+1);
2729 /* And now invoke the per-unit callbacks. Note that manager_invoke_notify_message() will handle
2730 * duplicate units make sure we only invoke each unit's handler once. */
2732 manager_invoke_notify_message(m
, u1
, ucred
, tags
, fds
);
2736 manager_invoke_notify_message(m
, u2
, ucred
, tags
, fds
);
2740 for (size_t i
= 0; array_copy
[i
]; i
++) {
2741 manager_invoke_notify_message(m
, array_copy
[i
], ucred
, tags
, fds
);
2746 log_warning("Cannot find unit for notify message of PID "PID_FMT
", ignoring.", ucred
->pid
);
2748 if (fdset_size(fds
) > 0)
2749 log_warning("Got extra auxiliary fds with notification message, closing them.");
2754 static void manager_invoke_sigchld_event(
2757 const siginfo_t
*si
) {
2763 /* Already invoked the handler of this unit in this iteration? Then don't process this again */
2764 if (u
->sigchldgen
== m
->sigchldgen
)
2766 u
->sigchldgen
= m
->sigchldgen
;
2768 log_unit_debug(u
, "Child "PID_FMT
" belongs to %s.", si
->si_pid
, u
->id
);
2769 unit_unwatch_pid(u
, si
->si_pid
);
2771 if (UNIT_VTABLE(u
)->sigchld_event
)
2772 UNIT_VTABLE(u
)->sigchld_event(u
, si
->si_pid
, si
->si_code
, si
->si_status
);
2775 static int manager_dispatch_sigchld(sd_event_source
*source
, void *userdata
) {
2776 Manager
*m
= ASSERT_PTR(userdata
);
2782 /* First we call waitid() for a PID and do not reap the zombie. That way we can still access
2783 * /proc/$PID for it while it is a zombie. */
2785 if (waitid(P_ALL
, 0, &si
, WEXITED
|WNOHANG
|WNOWAIT
) < 0) {
2787 if (errno
!= ECHILD
)
2788 log_error_errno(errno
, "Failed to peek for child with waitid(), ignoring: %m");
2796 if (IN_SET(si
.si_code
, CLD_EXITED
, CLD_KILLED
, CLD_DUMPED
)) {
2797 _cleanup_free_ Unit
**array_copy
= NULL
;
2798 _cleanup_free_
char *name
= NULL
;
2799 Unit
*u1
, *u2
, **array
;
2801 (void) pid_get_comm(si
.si_pid
, &name
);
2803 log_debug("Child "PID_FMT
" (%s) died (code=%s, status=%i/%s)",
2804 si
.si_pid
, strna(name
),
2805 sigchld_code_to_string(si
.si_code
),
2807 strna(si
.si_code
== CLD_EXITED
2808 ? exit_status_to_string(si
.si_status
, EXIT_STATUS_FULL
)
2809 : signal_to_string(si
.si_status
)));
2811 /* Increase the generation counter used for filtering out duplicate unit invocations */
2814 /* We look this up by a PidRef that only consists of the PID. After all we couldn't create a
2815 * pidfd here any more even if we wanted (since the process just exited). */
2816 PidRef pidref
= PIDREF_MAKE_FROM_PID(si
.si_pid
);
2818 /* And now figure out the unit this belongs to, it might be multiple... */
2819 u1
= manager_get_unit_by_pidref_cgroup(m
, &pidref
);
2820 u2
= hashmap_get(m
->watch_pids
, &pidref
);
2821 array
= hashmap_get(m
->watch_pids_more
, &pidref
);
2825 /* Count how many entries the array has */
2829 /* Make a copy of the array so that we don't trip up on the array changing beneath us */
2830 array_copy
= newdup(Unit
*, array
, n
+1);
2835 /* Finally, execute them all. Note that u1, u2 and the array might contain duplicates, but
2836 * that's fine, manager_invoke_sigchld_event() will ensure we only invoke the handlers once for
2837 * each iteration. */
2839 /* We check for oom condition, in case we got SIGCHLD before the oom notification.
2840 * We only do this for the cgroup the PID belonged to. */
2841 (void) unit_check_oom(u1
);
2843 /* We check if systemd-oomd performed a kill so that we log and notify appropriately */
2844 (void) unit_check_oomd_kill(u1
);
2846 manager_invoke_sigchld_event(m
, u1
, &si
);
2849 manager_invoke_sigchld_event(m
, u2
, &si
);
2851 for (size_t i
= 0; array_copy
[i
]; i
++)
2852 manager_invoke_sigchld_event(m
, array_copy
[i
], &si
);
2855 /* And now, we actually reap the zombie. */
2856 if (waitid(P_PID
, si
.si_pid
, &si
, WEXITED
) < 0) {
2857 log_error_errno(errno
, "Failed to dequeue child, ignoring: %m");
2864 /* All children processed for now, turn off event source */
2866 r
= sd_event_source_set_enabled(m
->sigchld_event_source
, SD_EVENT_OFF
);
2868 return log_error_errno(r
, "Failed to disable SIGCHLD event source: %m");
2873 static void manager_start_special(Manager
*m
, const char *name
, JobMode mode
) {
2876 if (manager_add_job_by_name_and_warn(m
, JOB_START
, name
, mode
, NULL
, &job
) < 0)
2879 const char *s
= unit_status_string(job
->unit
, NULL
);
2881 log_info("Activating special unit %s...", s
);
2884 "STATUS=Activating special unit %s...", s
);
2885 m
->status_ready
= false;
2888 static void manager_handle_ctrl_alt_del(Manager
*m
) {
2889 /* If the user presses C-A-D more than
2890 * 7 times within 2s, we reboot/shutdown immediately,
2891 * unless it was disabled in system.conf */
2893 if (ratelimit_below(&m
->ctrl_alt_del_ratelimit
) || m
->cad_burst_action
== EMERGENCY_ACTION_NONE
)
2894 manager_start_special(m
, SPECIAL_CTRL_ALT_DEL_TARGET
, JOB_REPLACE_IRREVERSIBLY
);
2896 emergency_action(m
, m
->cad_burst_action
, EMERGENCY_ACTION_WARN
, NULL
, -1,
2897 "Ctrl-Alt-Del was pressed more than 7 times within 2s");
2900 static int manager_dispatch_signal_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
2901 Manager
*m
= ASSERT_PTR(userdata
);
2903 struct signalfd_siginfo sfsi
;
2906 assert(m
->signal_fd
== fd
);
2908 if (revents
!= EPOLLIN
) {
2909 log_warning("Got unexpected events from signal file descriptor.");
2913 n
= read(m
->signal_fd
, &sfsi
, sizeof(sfsi
));
2915 if (ERRNO_IS_TRANSIENT(errno
))
2918 /* We return an error here, which will kill this handler,
2919 * to avoid a busy loop on read error. */
2920 return log_error_errno(errno
, "Reading from signal fd failed: %m");
2922 if (n
!= sizeof(sfsi
)) {
2923 log_warning("Truncated read from signal fd (%zi bytes), ignoring!", n
);
2927 log_received_signal(sfsi
.ssi_signo
== SIGCHLD
||
2928 (sfsi
.ssi_signo
== SIGTERM
&& MANAGER_IS_USER(m
))
2929 ? LOG_DEBUG
: LOG_INFO
,
2932 switch (sfsi
.ssi_signo
) {
2935 r
= sd_event_source_set_enabled(m
->sigchld_event_source
, SD_EVENT_ON
);
2937 log_warning_errno(r
, "Failed to enable SIGCHLD event source, ignoring: %m");
2942 if (MANAGER_IS_SYSTEM(m
)) {
2943 /* This is for compatibility with the original sysvinit */
2944 if (verify_run_space_and_log("Refusing to reexecute") < 0)
2947 m
->objective
= MANAGER_REEXECUTE
;
2953 if (MANAGER_IS_SYSTEM(m
))
2954 manager_handle_ctrl_alt_del(m
);
2956 manager_start_special(m
, SPECIAL_EXIT_TARGET
, JOB_REPLACE_IRREVERSIBLY
);
2960 /* This is a nop on non-init */
2961 if (MANAGER_IS_SYSTEM(m
))
2962 manager_start_special(m
, SPECIAL_KBREQUEST_TARGET
, JOB_REPLACE
);
2967 /* This is a nop on non-init */
2968 if (MANAGER_IS_SYSTEM(m
))
2969 manager_start_special(m
, SPECIAL_SIGPWR_TARGET
, JOB_REPLACE
);
2974 if (manager_dbus_is_running(m
, false)) {
2975 log_info("Trying to reconnect to bus...");
2977 (void) bus_init_api(m
);
2979 if (MANAGER_IS_SYSTEM(m
))
2980 (void) bus_init_system(m
);
2982 manager_start_special(m
, SPECIAL_DBUS_SERVICE
, JOB_REPLACE
);
2987 _cleanup_free_
char *dump
= NULL
;
2989 r
= manager_get_dump_string(m
, /* patterns= */ NULL
, &dump
);
2991 log_warning_errno(errno
, "Failed to acquire manager dump: %m");
2995 log_dump(LOG_INFO
, dump
);
3000 if (verify_run_space_and_log("Refusing to reload") < 0)
3003 m
->objective
= MANAGER_RELOAD
;
3008 /* Starting SIGRTMIN+0 */
3009 static const struct {
3012 } target_table
[] = {
3013 [0] = { SPECIAL_DEFAULT_TARGET
, JOB_ISOLATE
},
3014 [1] = { SPECIAL_RESCUE_TARGET
, JOB_ISOLATE
},
3015 [2] = { SPECIAL_EMERGENCY_TARGET
, JOB_ISOLATE
},
3016 [3] = { SPECIAL_HALT_TARGET
, JOB_REPLACE_IRREVERSIBLY
},
3017 [4] = { SPECIAL_POWEROFF_TARGET
, JOB_REPLACE_IRREVERSIBLY
},
3018 [5] = { SPECIAL_REBOOT_TARGET
, JOB_REPLACE_IRREVERSIBLY
},
3019 [6] = { SPECIAL_KEXEC_TARGET
, JOB_REPLACE_IRREVERSIBLY
},
3020 [7] = { SPECIAL_SOFT_REBOOT_TARGET
, JOB_REPLACE_IRREVERSIBLY
},
3023 /* Starting SIGRTMIN+13, so that target halt and system halt are 10 apart */
3024 static const ManagerObjective objective_table
[] = {
3026 [1] = MANAGER_POWEROFF
,
3027 [2] = MANAGER_REBOOT
,
3028 [3] = MANAGER_KEXEC
,
3029 [4] = MANAGER_SOFT_REBOOT
,
3032 if ((int) sfsi
.ssi_signo
>= SIGRTMIN
+0 &&
3033 (int) sfsi
.ssi_signo
< SIGRTMIN
+(int) ELEMENTSOF(target_table
)) {
3034 int idx
= (int) sfsi
.ssi_signo
- SIGRTMIN
;
3035 manager_start_special(m
, target_table
[idx
].target
, target_table
[idx
].mode
);
3039 if ((int) sfsi
.ssi_signo
>= SIGRTMIN
+13 &&
3040 (int) sfsi
.ssi_signo
< SIGRTMIN
+13+(int) ELEMENTSOF(objective_table
)) {
3041 m
->objective
= objective_table
[sfsi
.ssi_signo
- SIGRTMIN
- 13];
3045 switch (sfsi
.ssi_signo
- SIGRTMIN
) {
3048 bool generic
= false;
3050 if (sfsi
.ssi_code
!= SI_QUEUE
)
3053 /* Override a few select commands by our own PID1-specific logic */
3055 switch (sfsi
.ssi_int
) {
3057 case _COMMON_SIGNAL_COMMAND_LOG_LEVEL_BASE
..._COMMON_SIGNAL_COMMAND_LOG_LEVEL_END
:
3058 manager_override_log_level(m
, sfsi
.ssi_int
- _COMMON_SIGNAL_COMMAND_LOG_LEVEL_BASE
);
3061 case COMMON_SIGNAL_COMMAND_CONSOLE
:
3062 manager_override_log_target(m
, LOG_TARGET_CONSOLE
);
3065 case COMMON_SIGNAL_COMMAND_JOURNAL
:
3066 manager_override_log_target(m
, LOG_TARGET_JOURNAL
);
3069 case COMMON_SIGNAL_COMMAND_KMSG
:
3070 manager_override_log_target(m
, LOG_TARGET_KMSG
);
3073 case COMMON_SIGNAL_COMMAND_NULL
:
3074 manager_override_log_target(m
, LOG_TARGET_NULL
);
3077 case MANAGER_SIGNAL_COMMAND_DUMP_JOBS
: {
3078 _cleanup_free_
char *dump_jobs
= NULL
;
3080 r
= manager_get_dump_jobs_string(m
, /* patterns= */ NULL
, " ", &dump_jobs
);
3082 log_warning_errno(errno
, "Failed to acquire manager jobs dump: %m");
3086 log_dump(LOG_INFO
, dump_jobs
);
3096 return sigrtmin18_handler(source
, &sfsi
, NULL
);
3102 manager_override_show_status(m
, SHOW_STATUS_YES
, "signal");
3106 manager_override_show_status(m
, SHOW_STATUS_NO
, "signal");
3110 manager_override_log_level(m
, LOG_DEBUG
);
3114 manager_restore_original_log_level(m
);
3118 if (MANAGER_IS_USER(m
)) {
3119 m
->objective
= MANAGER_EXIT
;
3123 /* This is a nop on init */
3127 m
->objective
= MANAGER_REEXECUTE
;
3131 case 29: /* compatibility: used to be mapped to LOG_TARGET_SYSLOG_OR_KMSG */
3132 manager_restore_original_log_target(m
);
3136 manager_override_log_target(m
, LOG_TARGET_CONSOLE
);
3140 manager_override_log_target(m
, LOG_TARGET_KMSG
);
3144 log_warning("Got unhandled signal <%s>.", signal_to_string(sfsi
.ssi_signo
));
3151 static int manager_dispatch_time_change_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
3152 Manager
*m
= ASSERT_PTR(userdata
);
3155 log_struct(LOG_DEBUG
,
3156 "MESSAGE_ID=" SD_MESSAGE_TIME_CHANGE_STR
,
3157 LOG_MESSAGE("Time has been changed"));
3159 /* Restart the watch */
3160 (void) manager_setup_time_change(m
);
3162 HASHMAP_FOREACH(u
, m
->units
)
3163 if (UNIT_VTABLE(u
)->time_change
)
3164 UNIT_VTABLE(u
)->time_change(u
);
3169 static int manager_dispatch_timezone_change(
3170 sd_event_source
*source
,
3171 const struct inotify_event
*e
,
3174 Manager
*m
= ASSERT_PTR(userdata
);
3178 log_debug("inotify event for /etc/localtime");
3180 changed
= manager_read_timezone_stat(m
);
3184 /* Something changed, restart the watch, to ensure we watch the new /etc/localtime if it changed */
3185 (void) manager_setup_timezone_change(m
);
3187 /* Read the new timezone */
3190 log_debug("Timezone has been changed (now: %s).", tzname
[daylight
]);
3192 HASHMAP_FOREACH(u
, m
->units
)
3193 if (UNIT_VTABLE(u
)->timezone_change
)
3194 UNIT_VTABLE(u
)->timezone_change(u
);
3199 static int manager_dispatch_idle_pipe_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
3200 Manager
*m
= ASSERT_PTR(userdata
);
3202 assert(m
->idle_pipe
[2] == fd
);
3204 /* There's at least one Type=idle child that just gave up on us waiting for the boot process to
3205 * complete. Let's now turn off any further console output if there's at least one service that needs
3206 * console access, so that from now on our own output should not spill into that service's output
3207 * anymore. After all, we support Type=idle only to beautify console output and it generally is set
3208 * on services that want to own the console exclusively without our interference. */
3209 m
->no_console_output
= m
->n_on_console
> 0;
3211 /* Acknowledge the child's request, and let all other children know too that they shouldn't wait
3212 * any longer by closing the pipes towards them, which is what they are waiting for. */
3213 manager_close_idle_pipe(m
);
3218 static int manager_dispatch_jobs_in_progress(sd_event_source
*source
, usec_t usec
, void *userdata
) {
3219 Manager
*m
= ASSERT_PTR(userdata
);
3224 manager_print_jobs_in_progress(m
);
3226 r
= sd_event_source_set_time_relative(source
, JOBS_IN_PROGRESS_PERIOD_USEC
);
3230 return sd_event_source_set_enabled(source
, SD_EVENT_ONESHOT
);
3233 int manager_loop(Manager
*m
) {
3234 RateLimit rl
= { .interval
= 1*USEC_PER_SEC
, .burst
= 50000 };
3238 assert(m
->objective
== MANAGER_OK
); /* Ensure manager_startup() has been called */
3240 manager_check_finished(m
);
3242 /* There might still be some zombies hanging around from before we were exec()'ed. Let's reap them. */
3243 r
= sd_event_source_set_enabled(m
->sigchld_event_source
, SD_EVENT_ON
);
3245 return log_error_errno(r
, "Failed to enable SIGCHLD event source: %m");
3247 while (m
->objective
== MANAGER_OK
) {
3249 (void) watchdog_ping();
3251 if (!ratelimit_below(&rl
)) {
3252 /* Yay, something is going seriously wrong, pause a little */
3253 log_warning("Looping too fast. Throttling execution a little.");
3257 if (manager_dispatch_load_queue(m
) > 0)
3260 if (manager_dispatch_gc_job_queue(m
) > 0)
3263 if (manager_dispatch_gc_unit_queue(m
) > 0)
3266 if (manager_dispatch_cleanup_queue(m
) > 0)
3269 if (manager_dispatch_cgroup_realize_queue(m
) > 0)
3272 if (manager_dispatch_start_when_upheld_queue(m
) > 0)
3275 if (manager_dispatch_stop_when_bound_queue(m
) > 0)
3278 if (manager_dispatch_stop_when_unneeded_queue(m
) > 0)
3281 if (manager_dispatch_release_resources_queue(m
) > 0)
3284 if (manager_dispatch_dbus_queue(m
) > 0)
3287 /* Sleep for watchdog runtime wait time */
3288 r
= sd_event_run(m
->event
, watchdog_runtime_wait());
3290 return log_error_errno(r
, "Failed to run event loop: %m");
3293 return m
->objective
;
3296 int manager_load_unit_from_dbus_path(Manager
*m
, const char *s
, sd_bus_error
*e
, Unit
**_u
) {
3297 _cleanup_free_
char *n
= NULL
;
3298 sd_id128_t invocation_id
;
3306 r
= unit_name_from_dbus_path(s
, &n
);
3310 /* Permit addressing units by invocation ID: if the passed bus path is suffixed by a 128-bit ID then
3311 * we use it as invocation ID. */
3312 r
= sd_id128_from_string(n
, &invocation_id
);
3314 u
= hashmap_get(m
->units_by_invocation_id
, &invocation_id
);
3320 return sd_bus_error_setf(e
, BUS_ERROR_NO_UNIT_FOR_INVOCATION_ID
,
3321 "No unit with the specified invocation ID " SD_ID128_FORMAT_STR
" known.",
3322 SD_ID128_FORMAT_VAL(invocation_id
));
3325 /* If this didn't work, we check if this is a unit name */
3326 if (!unit_name_is_valid(n
, UNIT_NAME_PLAIN
|UNIT_NAME_INSTANCE
)) {
3327 _cleanup_free_
char *nn
= NULL
;
3330 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
,
3331 "Unit name %s is neither a valid invocation ID nor unit name.", strnull(nn
));
3334 r
= manager_load_unit(m
, n
, NULL
, e
, &u
);
3342 int manager_get_job_from_dbus_path(Manager
*m
, const char *s
, Job
**_j
) {
3352 p
= startswith(s
, "/org/freedesktop/systemd1/job/");
3356 r
= safe_atou(p
, &id
);
3360 j
= manager_get_job(m
, id
);
3369 void manager_send_unit_audit(Manager
*m
, Unit
*u
, int type
, bool success
) {
3372 _cleanup_free_
char *p
= NULL
;
3376 if (!MANAGER_IS_SYSTEM(m
))
3379 audit_fd
= get_audit_fd();
3383 /* Don't generate audit events if the service was already
3384 * started and we're just deserializing */
3385 if (MANAGER_IS_RELOADING(m
))
3388 r
= unit_name_to_prefix_and_instance(u
->id
, &p
);
3390 log_warning_errno(r
, "Failed to extract prefix and instance of unit name, ignoring: %m");
3394 msg
= strjoina("unit=", p
);
3395 if (audit_log_user_comm_message(audit_fd
, type
, msg
, "systemd", NULL
, NULL
, NULL
, success
) < 0) {
3396 if (ERRNO_IS_PRIVILEGE(errno
)) {
3397 /* We aren't allowed to send audit messages? Then let's not retry again. */
3398 log_debug_errno(errno
, "Failed to send audit message, closing audit socket: %m");
3401 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
;
3411 /* Don't generate plymouth events if the service was already
3412 * started and we're just deserializing */
3413 if (MANAGER_IS_RELOADING(m
))
3416 if (!MANAGER_IS_SYSTEM(m
))
3419 if (detect_container() > 0)
3422 if (!UNIT_VTABLE(u
)->notify_plymouth
)
3425 c
= asprintf(&message
, "U\x02%c%s%c", (int) (strlen(u
->id
) + 1), u
->id
, '\x00');
3427 return (void) log_oom();
3429 /* We set SOCK_NONBLOCK here so that we rather drop the message then wait for plymouth */
3430 r
= plymouth_send_raw(message
, c
, SOCK_NONBLOCK
);
3432 log_full_errno(ERRNO_IS_NO_PLYMOUTH(r
) ? LOG_DEBUG
: LOG_WARNING
, r
,
3433 "Failed to communicate with plymouth: %m");
3436 usec_t
manager_get_watchdog(Manager
*m
, WatchdogType t
) {
3439 if (MANAGER_IS_USER(m
))
3440 return USEC_INFINITY
;
3442 if (m
->watchdog_overridden
[t
] != USEC_INFINITY
)
3443 return m
->watchdog_overridden
[t
];
3445 return m
->watchdog
[t
];
3448 void manager_set_watchdog(Manager
*m
, WatchdogType t
, usec_t timeout
) {
3452 if (MANAGER_IS_USER(m
))
3455 if (m
->watchdog
[t
] == timeout
)
3458 if (m
->watchdog_overridden
[t
] == USEC_INFINITY
) {
3459 if (t
== WATCHDOG_RUNTIME
)
3460 (void) watchdog_setup(timeout
);
3461 else if (t
== WATCHDOG_PRETIMEOUT
)
3462 (void) watchdog_setup_pretimeout(timeout
);
3465 m
->watchdog
[t
] = timeout
;
3468 void manager_override_watchdog(Manager
*m
, WatchdogType t
, usec_t timeout
) {
3473 if (MANAGER_IS_USER(m
))
3476 if (m
->watchdog_overridden
[t
] == timeout
)
3479 usec
= timeout
== USEC_INFINITY
? m
->watchdog
[t
] : timeout
;
3480 if (t
== WATCHDOG_RUNTIME
)
3481 (void) watchdog_setup(usec
);
3482 else if (t
== WATCHDOG_PRETIMEOUT
)
3483 (void) watchdog_setup_pretimeout(usec
);
3485 m
->watchdog_overridden
[t
] = timeout
;
3488 int manager_set_watchdog_pretimeout_governor(Manager
*m
, const char *governor
) {
3489 _cleanup_free_
char *p
= NULL
;
3494 if (MANAGER_IS_USER(m
))
3497 if (streq_ptr(m
->watchdog_pretimeout_governor
, governor
))
3500 p
= strdup(governor
);
3504 r
= watchdog_setup_pretimeout_governor(governor
);
3508 return free_and_replace(m
->watchdog_pretimeout_governor
, p
);
3511 int manager_override_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_overridden
, governor
))
3523 p
= strdup(governor
);
3527 r
= watchdog_setup_pretimeout_governor(governor
);
3531 return free_and_replace(m
->watchdog_pretimeout_governor_overridden
, p
);
3534 int manager_reload(Manager
*m
) {
3535 _unused_
_cleanup_(manager_reloading_stopp
) Manager
*reloading
= NULL
;
3536 _cleanup_fdset_free_ FDSet
*fds
= NULL
;
3537 _cleanup_fclose_
FILE *f
= NULL
;
3542 r
= manager_open_serialization(m
, &f
);
3544 return log_error_errno(r
, "Failed to create serialization file: %m");
3550 /* We are officially in reload mode from here on. */
3551 reloading
= manager_reloading_start(m
);
3553 r
= manager_serialize(m
, f
, fds
, false);
3557 if (fseeko(f
, 0, SEEK_SET
) < 0)
3558 return log_error_errno(errno
, "Failed to seek to beginning of serialization: %m");
3560 /* 💀 This is the point of no return, from here on there is no way back. 💀 */
3563 bus_manager_send_reloading(m
, true);
3565 /* Start by flushing out all jobs and units, all generated units, all runtime environments, all dynamic users
3566 * and everything else that is worth flushing out. We'll get it all back from the serialization — if we need
3569 manager_clear_jobs_and_units(m
);
3570 lookup_paths_flush_generator(&m
->lookup_paths
);
3571 lookup_paths_free(&m
->lookup_paths
);
3572 exec_shared_runtime_vacuum(m
);
3573 dynamic_user_vacuum(m
, false);
3574 m
->uid_refs
= hashmap_free(m
->uid_refs
);
3575 m
->gid_refs
= hashmap_free(m
->gid_refs
);
3577 r
= lookup_paths_init_or_warn(&m
->lookup_paths
, m
->runtime_scope
, 0, NULL
);
3581 (void) manager_run_environment_generators(m
);
3582 (void) manager_run_generators(m
);
3584 lookup_paths_log(&m
->lookup_paths
);
3586 /* We flushed out generated files, for which we don't watch mtime, so we should flush the old map. */
3587 manager_free_unit_name_maps(m
);
3588 m
->unit_file_state_outdated
= false;
3590 /* First, enumerate what we can from kernel and suchlike */
3591 manager_enumerate_perpetual(m
);
3592 manager_enumerate(m
);
3594 /* Second, deserialize our stored data */
3595 r
= manager_deserialize(m
, f
, fds
);
3597 log_warning_errno(r
, "Deserialization failed, proceeding anyway: %m");
3599 /* We don't need the serialization anymore */
3602 /* Re-register notify_fd as event source, and set up other sockets/communication channels we might need */
3603 (void) manager_setup_notify(m
);
3604 (void) manager_setup_cgroups_agent(m
);
3605 (void) manager_setup_user_lookup_fd(m
);
3607 /* Third, fire things up! */
3608 manager_coldplug(m
);
3610 /* Clean up runtime objects no longer referenced */
3613 /* Clean up deserialized tracked clients */
3614 m
->deserialized_subscribed
= strv_free(m
->deserialized_subscribed
);
3616 /* Consider the reload process complete now. */
3617 assert(m
->n_reloading
> 0);
3622 m
->send_reloading_done
= true;
3626 void manager_reset_failed(Manager
*m
) {
3631 HASHMAP_FOREACH(u
, m
->units
)
3632 unit_reset_failed(u
);
3635 bool manager_unit_inactive_or_pending(Manager
*m
, const char *name
) {
3641 /* Returns true if the unit is inactive or going down */
3642 u
= manager_get_unit(m
, name
);
3646 return unit_inactive_or_pending(u
);
3649 static void log_taint_string(Manager
*m
) {
3650 _cleanup_free_
char *taint
= NULL
;
3654 if (MANAGER_IS_USER(m
) || m
->taint_logged
)
3657 m
->taint_logged
= true; /* only check for taint once */
3659 taint
= manager_taint_string(m
);
3663 log_struct(LOG_NOTICE
,
3664 LOG_MESSAGE("System is tainted: %s", taint
),
3666 "MESSAGE_ID=" SD_MESSAGE_TAINTED_STR
);
3669 static void manager_notify_finished(Manager
*m
) {
3670 usec_t firmware_usec
, loader_usec
, kernel_usec
, initrd_usec
, userspace_usec
, total_usec
;
3672 if (MANAGER_IS_TEST_RUN(m
))
3675 if (MANAGER_IS_SYSTEM(m
) && detect_container() <= 0) {
3676 char buf
[FORMAT_TIMESPAN_MAX
+ STRLEN(" (firmware) + ") + FORMAT_TIMESPAN_MAX
+ STRLEN(" (loader) + ")]
3679 size_t size
= sizeof buf
;
3681 /* Note that MANAGER_TIMESTAMP_KERNEL's monotonic value is always at 0, and
3682 * MANAGER_TIMESTAMP_FIRMWARE's and MANAGER_TIMESTAMP_LOADER's monotonic value should be considered
3683 * negative values. */
3685 firmware_usec
= m
->timestamps
[MANAGER_TIMESTAMP_FIRMWARE
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_LOADER
].monotonic
;
3686 loader_usec
= m
->timestamps
[MANAGER_TIMESTAMP_LOADER
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_KERNEL
].monotonic
;
3687 userspace_usec
= m
->timestamps
[MANAGER_TIMESTAMP_FINISH
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_USERSPACE
].monotonic
;
3688 total_usec
= m
->timestamps
[MANAGER_TIMESTAMP_FIRMWARE
].monotonic
+ m
->timestamps
[MANAGER_TIMESTAMP_FINISH
].monotonic
;
3690 if (firmware_usec
> 0)
3691 size
= strpcpyf(&p
, size
, "%s (firmware) + ", FORMAT_TIMESPAN(firmware_usec
, USEC_PER_MSEC
));
3692 if (loader_usec
> 0)
3693 size
= strpcpyf(&p
, size
, "%s (loader) + ", FORMAT_TIMESPAN(loader_usec
, USEC_PER_MSEC
));
3695 if (dual_timestamp_is_set(&m
->timestamps
[MANAGER_TIMESTAMP_INITRD
])) {
3697 /* The initrd case on bare-metal */
3698 kernel_usec
= m
->timestamps
[MANAGER_TIMESTAMP_INITRD
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_KERNEL
].monotonic
;
3699 initrd_usec
= m
->timestamps
[MANAGER_TIMESTAMP_USERSPACE
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_INITRD
].monotonic
;
3701 log_struct(LOG_INFO
,
3702 "MESSAGE_ID=" SD_MESSAGE_STARTUP_FINISHED_STR
,
3703 "KERNEL_USEC="USEC_FMT
, kernel_usec
,
3704 "INITRD_USEC="USEC_FMT
, initrd_usec
,
3705 "USERSPACE_USEC="USEC_FMT
, userspace_usec
,
3706 LOG_MESSAGE("Startup finished in %s%s (kernel) + %s (initrd) + %s (userspace) = %s.",
3708 FORMAT_TIMESPAN(kernel_usec
, USEC_PER_MSEC
),
3709 FORMAT_TIMESPAN(initrd_usec
, USEC_PER_MSEC
),
3710 FORMAT_TIMESPAN(userspace_usec
, USEC_PER_MSEC
),
3711 FORMAT_TIMESPAN(total_usec
, USEC_PER_MSEC
)));
3713 /* The initrd-less case on bare-metal */
3715 kernel_usec
= m
->timestamps
[MANAGER_TIMESTAMP_USERSPACE
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_KERNEL
].monotonic
;
3718 log_struct(LOG_INFO
,
3719 "MESSAGE_ID=" SD_MESSAGE_STARTUP_FINISHED_STR
,
3720 "KERNEL_USEC="USEC_FMT
, kernel_usec
,
3721 "USERSPACE_USEC="USEC_FMT
, userspace_usec
,
3722 LOG_MESSAGE("Startup finished in %s%s (kernel) + %s (userspace) = %s.",
3724 FORMAT_TIMESPAN(kernel_usec
, USEC_PER_MSEC
),
3725 FORMAT_TIMESPAN(userspace_usec
, USEC_PER_MSEC
),
3726 FORMAT_TIMESPAN(total_usec
, USEC_PER_MSEC
)));
3729 /* The container and --user case */
3730 firmware_usec
= loader_usec
= initrd_usec
= kernel_usec
= 0;
3731 total_usec
= userspace_usec
= m
->timestamps
[MANAGER_TIMESTAMP_FINISH
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_USERSPACE
].monotonic
;
3733 log_struct(LOG_INFO
,
3734 "MESSAGE_ID=" SD_MESSAGE_USER_STARTUP_FINISHED_STR
,
3735 "USERSPACE_USEC="USEC_FMT
, userspace_usec
,
3736 LOG_MESSAGE("Startup finished in %s.",
3737 FORMAT_TIMESPAN(total_usec
, USEC_PER_MSEC
)));
3740 bus_manager_send_finished(m
, firmware_usec
, loader_usec
, kernel_usec
, initrd_usec
, userspace_usec
, total_usec
);
3742 log_taint_string(m
);
3745 static void user_manager_send_ready(Manager
*m
) {
3750 /* We send READY=1 on reaching basic.target only when running in --user mode. */
3751 if (!MANAGER_IS_USER(m
) || m
->ready_sent
)
3754 r
= sd_notify(false,
3756 "STATUS=Reached " SPECIAL_BASIC_TARGET
".");
3758 log_warning_errno(r
, "Failed to send readiness notification, ignoring: %m");
3760 m
->ready_sent
= true;
3761 m
->status_ready
= false;
3764 static void manager_send_ready(Manager
*m
) {
3767 if (m
->ready_sent
&& m
->status_ready
)
3768 /* Skip the notification if nothing changed. */
3771 r
= sd_notify(false,
3775 log_full_errno(m
->ready_sent
? LOG_DEBUG
: LOG_WARNING
, r
,
3776 "Failed to send readiness notification, ignoring: %m");
3778 m
->ready_sent
= m
->status_ready
= true;
3781 static void manager_check_basic_target(Manager
*m
) {
3786 /* Small shortcut */
3787 if (m
->ready_sent
&& m
->taint_logged
)
3790 u
= manager_get_unit(m
, SPECIAL_BASIC_TARGET
);
3791 if (!u
|| !UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u
)))
3794 /* For user managers, send out READY=1 as soon as we reach basic.target */
3795 user_manager_send_ready(m
);
3797 /* Log the taint string as soon as we reach basic.target */
3798 log_taint_string(m
);
3801 void manager_check_finished(Manager
*m
) {
3804 if (MANAGER_IS_RELOADING(m
))
3807 /* Verify that we have entered the event loop already, and not left it again. */
3808 if (!MANAGER_IS_RUNNING(m
))
3811 manager_check_basic_target(m
);
3813 if (hashmap_size(m
->jobs
) > 0) {
3814 if (m
->jobs_in_progress_event_source
)
3815 /* Ignore any failure, this is only for feedback */
3816 (void) sd_event_source_set_time(m
->jobs_in_progress_event_source
,
3817 manager_watch_jobs_next_time(m
));
3821 /* The jobs hashmap tends to grow a lot during boot, and then it's not reused until shutdown. Let's
3822 kill the hashmap if it is relatively large. */
3823 if (hashmap_buckets(m
->jobs
) > hashmap_size(m
->units
) / 10)
3824 m
->jobs
= hashmap_free(m
->jobs
);
3826 manager_send_ready(m
);
3828 /* Notify Type=idle units that we are done now */
3829 manager_close_idle_pipe(m
);
3831 if (MANAGER_IS_FINISHED(m
))
3834 manager_flip_auto_status(m
, false, "boot finished");
3836 /* Turn off confirm spawn now */
3837 m
->confirm_spawn
= NULL
;
3839 /* No need to update ask password status when we're going non-interactive */
3840 manager_close_ask_password(m
);
3842 /* This is no longer the first boot */
3843 manager_set_first_boot(m
, false);
3845 dual_timestamp_now(m
->timestamps
+ MANAGER_TIMESTAMP_FINISH
);
3847 manager_notify_finished(m
);
3849 manager_invalidate_startup_units(m
);
3852 void manager_send_reloading(Manager
*m
) {
3855 /* Let whoever invoked us know that we are now reloading */
3856 (void) sd_notifyf(/* unset= */ false,
3858 "MONOTONIC_USEC=" USEC_FMT
"\n", now(CLOCK_MONOTONIC
));
3860 /* And ensure that we'll send READY=1 again as soon as we are ready again */
3861 m
->ready_sent
= false;
3864 static bool generator_path_any(const char* const* paths
) {
3867 /* Optimize by skipping the whole process by not creating output directories
3868 * if no generators are found. */
3869 STRV_FOREACH(path
, paths
)
3870 if (access(*path
, F_OK
) == 0)
3872 else if (errno
!= ENOENT
)
3873 log_warning_errno(errno
, "Failed to open generator directory %s: %m", *path
);
3878 static int manager_run_environment_generators(Manager
*m
) {
3879 char **tmp
= NULL
; /* this is only used in the forked process, no cleanup here */
3880 _cleanup_strv_free_
char **paths
= NULL
;
3882 [STDOUT_GENERATE
] = &tmp
,
3883 [STDOUT_COLLECT
] = &tmp
,
3884 [STDOUT_CONSUME
] = &m
->transient_environment
,
3888 if (MANAGER_IS_TEST_RUN(m
) && !(m
->test_run_flags
& MANAGER_TEST_RUN_ENV_GENERATORS
))
3891 paths
= env_generator_binary_paths(m
->runtime_scope
);
3895 if (!generator_path_any((const char* const*) paths
))
3899 r
= execute_directories((const char* const*) paths
, DEFAULT_TIMEOUT_USEC
, gather_environment
,
3900 args
, NULL
, m
->transient_environment
,
3901 EXEC_DIR_PARALLEL
| EXEC_DIR_IGNORE_ERRORS
| EXEC_DIR_SET_SYSTEMD_EXEC_PID
);
3905 static int build_generator_environment(Manager
*m
, char ***ret
) {
3906 _cleanup_strv_free_
char **nl
= NULL
;
3908 ConfidentialVirtualization cv
;
3914 /* Generators oftentimes want to know some basic facts about the environment they run in, in order to
3915 * adjust generated units to that. Let's pass down some bits of information that are easy for us to
3916 * determine (but a bit harder for generator scripts to determine), as environment variables. */
3918 nl
= strv_copy(m
->transient_environment
);
3922 r
= strv_env_assign(&nl
, "SYSTEMD_SCOPE", runtime_scope_to_string(m
->runtime_scope
));
3926 if (MANAGER_IS_SYSTEM(m
)) {
3927 /* Note that $SYSTEMD_IN_INITRD may be used to override the initrd detection in much of our
3928 * codebase. This is hence more than purely informational. It will shortcut detection of the
3929 * initrd state if generators invoke our own tools. But that's OK, as it would come to the
3930 * same results (hopefully). */
3931 r
= strv_env_assign(&nl
, "SYSTEMD_IN_INITRD", one_zero(in_initrd()));
3935 if (m
->first_boot
>= 0) {
3936 r
= strv_env_assign(&nl
, "SYSTEMD_FIRST_BOOT", one_zero(m
->first_boot
));
3942 v
= detect_virtualization();
3944 log_debug_errno(v
, "Failed to detect virtualization, ignoring: %m");
3948 s
= strjoina(VIRTUALIZATION_IS_VM(v
) ? "vm:" :
3949 VIRTUALIZATION_IS_CONTAINER(v
) ? "container:" : ":",
3950 virtualization_to_string(v
));
3952 r
= strv_env_assign(&nl
, "SYSTEMD_VIRTUALIZATION", s
);
3957 cv
= detect_confidential_virtualization();
3959 log_debug_errno(cv
, "Failed to detect confidential virtualization, ignoring: %m");
3961 r
= strv_env_assign(&nl
, "SYSTEMD_CONFIDENTIAL_VIRTUALIZATION", confidential_virtualization_to_string(cv
));
3966 r
= strv_env_assign(&nl
, "SYSTEMD_ARCHITECTURE", architecture_to_string(uname_architecture()));
3970 *ret
= TAKE_PTR(nl
);
3974 static int manager_execute_generators(Manager
*m
, char **paths
, bool remount_ro
) {
3975 _cleanup_strv_free_
char **ge
= NULL
;
3976 const char *argv
[] = {
3977 NULL
, /* Leave this empty, execute_directory() will fill something in */
3978 m
->lookup_paths
.generator
,
3979 m
->lookup_paths
.generator_early
,
3980 m
->lookup_paths
.generator_late
,
3985 r
= build_generator_environment(m
, &ge
);
3987 return log_error_errno(r
, "Failed to build generator environment: %m");
3990 /* Remount most of the filesystem tree read-only. We leave /sys/ as-is, because our code
3991 * checks whether it is read-only to detect containerized execution environments. We leave
3992 * /run/ as-is too, because that's where our output goes. We also leave /proc/ and /dev/shm/
3993 * because they're API, and /tmp/ that safe_fork() mounted for us.
3995 r
= bind_remount_recursive("/", MS_RDONLY
, MS_RDONLY
,
3996 STRV_MAKE("/sys", "/run", "/proc", "/dev/shm", "/tmp"));
3998 log_warning_errno(r
, "Read-only bind remount failed, ignoring: %m");
4001 BLOCK_WITH_UMASK(0022);
4002 return execute_directories(
4003 (const char* const*) paths
,
4004 DEFAULT_TIMEOUT_USEC
,
4005 /* callbacks= */ NULL
, /* callback_args= */ NULL
,
4008 EXEC_DIR_PARALLEL
| EXEC_DIR_IGNORE_ERRORS
| EXEC_DIR_SET_SYSTEMD_EXEC_PID
);
4011 static int manager_run_generators(Manager
*m
) {
4012 ForkFlags flags
= FORK_RESET_SIGNALS
| FORK_WAIT
| FORK_NEW_MOUNTNS
| FORK_MOUNTNS_SLAVE
;
4013 _cleanup_strv_free_
char **paths
= NULL
;
4018 if (MANAGER_IS_TEST_RUN(m
) && !(m
->test_run_flags
& MANAGER_TEST_RUN_GENERATORS
))
4021 paths
= generator_binary_paths(m
->runtime_scope
);
4025 if (!generator_path_any((const char* const*) paths
))
4028 r
= lookup_paths_mkdir_generator(&m
->lookup_paths
);
4030 log_error_errno(r
, "Failed to create generator directories: %m");
4034 /* If we are the system manager, we fork and invoke the generators in a sanitized mount namespace. If
4035 * we are the user manager, let's just execute the generators directly. We might not have the
4036 * necessary privileges, and the system manager has already mounted /tmp/ and everything else for us.
4038 if (MANAGER_IS_USER(m
)) {
4039 r
= manager_execute_generators(m
, paths
, /* remount_ro= */ false);
4043 /* On some systems /tmp/ doesn't exist, and on some other systems we cannot create it at all. Avoid
4044 * trying to mount a private tmpfs on it as there's no one size fits all. */
4045 if (is_dir("/tmp", /* follow= */ false) > 0)
4046 flags
|= FORK_PRIVATE_TMP
;
4048 r
= safe_fork("(sd-gens)", flags
, NULL
);
4050 r
= manager_execute_generators(m
, paths
, /* remount_ro= */ true);
4051 _exit(r
>= 0 ? EXIT_SUCCESS
: EXIT_FAILURE
);
4054 if (!ERRNO_IS_PRIVILEGE(r
) && r
!= -EINVAL
) {
4055 log_error_errno(r
, "Failed to fork off sandboxing environment for executing generators: %m");
4059 /* Failed to fork with new mount namespace? Maybe, running in a container environment with
4060 * seccomp or without capability.
4062 * We also allow -EINVAL to allow running without CLONE_NEWNS.
4064 * Also, when running on non-native userland architecture via systemd-nspawn and
4065 * qemu-user-static QEMU-emulator, clone() with CLONE_NEWNS fails with EINVAL, see
4066 * https://github.com/systemd/systemd/issues/28901.
4069 "Failed to fork off sandboxing environment for executing generators. "
4070 "Falling back to execute generators without sandboxing: %m");
4071 r
= manager_execute_generators(m
, paths
, /* remount_ro= */ false);
4075 lookup_paths_trim_generator(&m
->lookup_paths
);
4079 int manager_transient_environment_add(Manager
*m
, char **plus
) {
4084 if (strv_isempty(plus
))
4087 a
= strv_env_merge(m
->transient_environment
, plus
);
4091 sanitize_environment(a
);
4093 return strv_free_and_replace(m
->transient_environment
, a
);
4096 int manager_client_environment_modify(
4101 char **a
= NULL
, **b
= NULL
, **l
;
4105 if (strv_isempty(minus
) && strv_isempty(plus
))
4108 l
= m
->client_environment
;
4110 if (!strv_isempty(minus
)) {
4111 a
= strv_env_delete(l
, 1, minus
);
4118 if (!strv_isempty(plus
)) {
4119 b
= strv_env_merge(l
, plus
);
4128 if (m
->client_environment
!= l
)
4129 strv_free(m
->client_environment
);
4136 m
->client_environment
= sanitize_environment(l
);
4140 int manager_get_effective_environment(Manager
*m
, char ***ret
) {
4146 l
= strv_env_merge(m
->transient_environment
, m
->client_environment
);
4154 int manager_set_unit_defaults(Manager
*m
, const UnitDefaults
*defaults
) {
4155 _cleanup_free_
char *label
= NULL
;
4156 struct rlimit
*rlimit
[_RLIMIT_MAX
];
4162 if (streq_ptr(defaults
->smack_process_label
, "/"))
4165 const char *l
= defaults
->smack_process_label
;
4166 #ifdef SMACK_DEFAULT_PROCESS_LABEL
4168 l
= SMACK_DEFAULT_PROCESS_LABEL
;
4178 r
= rlimit_copy_all(rlimit
, defaults
->rlimit
);
4182 m
->defaults
.std_output
= defaults
->std_output
;
4183 m
->defaults
.std_error
= defaults
->std_error
;
4185 m
->defaults
.restart_usec
= defaults
->restart_usec
;
4186 m
->defaults
.timeout_start_usec
= defaults
->timeout_start_usec
;
4187 m
->defaults
.timeout_stop_usec
= defaults
->timeout_stop_usec
;
4188 m
->defaults
.timeout_abort_usec
= defaults
->timeout_abort_usec
;
4189 m
->defaults
.timeout_abort_set
= defaults
->timeout_abort_set
;
4190 m
->defaults
.device_timeout_usec
= defaults
->device_timeout_usec
;
4192 m
->defaults
.start_limit_interval
= defaults
->start_limit_interval
;
4193 m
->defaults
.start_limit_burst
= defaults
->start_limit_burst
;
4195 m
->defaults
.cpu_accounting
= defaults
->cpu_accounting
;
4196 m
->defaults
.memory_accounting
= defaults
->memory_accounting
;
4197 m
->defaults
.io_accounting
= defaults
->io_accounting
;
4198 m
->defaults
.blockio_accounting
= defaults
->blockio_accounting
;
4199 m
->defaults
.tasks_accounting
= defaults
->tasks_accounting
;
4200 m
->defaults
.ip_accounting
= defaults
->ip_accounting
;
4202 m
->defaults
.tasks_max
= defaults
->tasks_max
;
4203 m
->defaults
.timer_accuracy_usec
= defaults
->timer_accuracy_usec
;
4205 m
->defaults
.oom_policy
= defaults
->oom_policy
;
4206 m
->defaults
.oom_score_adjust
= defaults
->oom_score_adjust
;
4207 m
->defaults
.oom_score_adjust_set
= defaults
->oom_score_adjust_set
;
4209 m
->defaults
.memory_pressure_watch
= defaults
->memory_pressure_watch
;
4210 m
->defaults
.memory_pressure_threshold_usec
= defaults
->memory_pressure_threshold_usec
;
4212 free_and_replace(m
->defaults
.smack_process_label
, label
);
4213 rlimit_free_all(m
->defaults
.rlimit
);
4214 memcpy(m
->defaults
.rlimit
, rlimit
, sizeof(struct rlimit
*) * _RLIMIT_MAX
);
4219 void manager_recheck_dbus(Manager
*m
) {
4222 /* Connects to the bus if the dbus service and socket are running. If we are running in user mode
4223 * this is all it does. In system mode we'll also connect to the system bus (which will most likely
4224 * just reuse the connection of the API bus). That's because the system bus after all runs as service
4225 * of the system instance, while in the user instance we can assume it's already there. */
4227 if (MANAGER_IS_RELOADING(m
))
4228 return; /* don't check while we are reloading… */
4230 if (manager_dbus_is_running(m
, false)) {
4231 (void) bus_init_api(m
);
4233 if (MANAGER_IS_SYSTEM(m
))
4234 (void) bus_init_system(m
);
4236 (void) bus_done_api(m
);
4238 if (MANAGER_IS_SYSTEM(m
))
4239 (void) bus_done_system(m
);
4243 static bool manager_journal_is_running(Manager
*m
) {
4248 if (MANAGER_IS_TEST_RUN(m
))
4251 /* If we are the user manager we can safely assume that the journal is up */
4252 if (!MANAGER_IS_SYSTEM(m
))
4255 /* Check that the socket is not only up, but in RUNNING state */
4256 u
= manager_get_unit(m
, SPECIAL_JOURNALD_SOCKET
);
4259 if (SOCKET(u
)->state
!= SOCKET_RUNNING
)
4262 /* Similar, check if the daemon itself is fully up, too */
4263 u
= manager_get_unit(m
, SPECIAL_JOURNALD_SERVICE
);
4266 if (!IN_SET(SERVICE(u
)->state
, SERVICE_RELOAD
, SERVICE_RUNNING
))
4272 void disable_printk_ratelimit(void) {
4273 /* Disable kernel's printk ratelimit.
4275 * Logging to /dev/kmsg is most useful during early boot and shutdown, where normal logging
4276 * mechanisms are not available. The semantics of this sysctl are such that any kernel command-line
4277 * setting takes precedence. */
4280 r
= sysctl_write("kernel/printk_devkmsg", "on");
4282 log_debug_errno(r
, "Failed to set sysctl kernel.printk_devkmsg=on: %m");
4285 void manager_recheck_journal(Manager
*m
) {
4289 /* Don't bother with this unless we are in the special situation of being PID 1 */
4290 if (getpid_cached() != 1)
4293 /* Don't check this while we are reloading, things might still change */
4294 if (MANAGER_IS_RELOADING(m
))
4297 /* The journal is fully and entirely up? If so, let's permit logging to it, if that's configured. If
4298 * the journal is down, don't ever log to it, otherwise we might end up deadlocking ourselves as we
4299 * might trigger an activation ourselves we can't fulfill. */
4300 log_set_prohibit_ipc(!manager_journal_is_running(m
));
4304 static ShowStatus
manager_get_show_status(Manager
*m
) {
4307 if (MANAGER_IS_USER(m
))
4308 return _SHOW_STATUS_INVALID
;
4310 if (m
->show_status_overridden
!= _SHOW_STATUS_INVALID
)
4311 return m
->show_status_overridden
;
4313 return m
->show_status
;
4316 bool manager_get_show_status_on(Manager
*m
) {
4319 return show_status_on(manager_get_show_status(m
));
4322 static void set_show_status_marker(bool b
) {
4324 (void) touch("/run/systemd/show-status");
4326 (void) unlink("/run/systemd/show-status");
4329 void manager_set_show_status(Manager
*m
, ShowStatus mode
, const char *reason
) {
4332 assert(mode
>= 0 && mode
< _SHOW_STATUS_MAX
);
4334 if (MANAGER_IS_USER(m
))
4337 if (mode
== m
->show_status
)
4340 if (m
->show_status_overridden
== _SHOW_STATUS_INVALID
) {
4343 enabled
= show_status_on(mode
);
4344 log_debug("%s (%s) showing of status (%s).",
4345 enabled
? "Enabling" : "Disabling",
4346 strna(show_status_to_string(mode
)),
4349 set_show_status_marker(enabled
);
4352 m
->show_status
= mode
;
4355 void manager_override_show_status(Manager
*m
, ShowStatus mode
, const char *reason
) {
4357 assert(mode
< _SHOW_STATUS_MAX
);
4359 if (MANAGER_IS_USER(m
))
4362 if (mode
== m
->show_status_overridden
)
4365 m
->show_status_overridden
= mode
;
4367 if (mode
== _SHOW_STATUS_INVALID
)
4368 mode
= m
->show_status
;
4370 log_debug("%s (%s) showing of status (%s).",
4371 m
->show_status_overridden
!= _SHOW_STATUS_INVALID
? "Overriding" : "Restoring",
4372 strna(show_status_to_string(mode
)),
4375 set_show_status_marker(show_status_on(mode
));
4378 const char *manager_get_confirm_spawn(Manager
*m
) {
4379 static int last_errno
= 0;
4385 /* Here's the deal: we want to test the validity of the console but don't want
4386 * PID1 to go through the whole console process which might block. But we also
4387 * want to warn the user only once if something is wrong with the console so we
4388 * cannot do the sanity checks after spawning our children. So here we simply do
4389 * really basic tests to hopefully trap common errors.
4391 * If the console suddenly disappear at the time our children will really it
4392 * then they will simply fail to acquire it and a positive answer will be
4393 * assumed. New children will fall back to /dev/console though.
4395 * Note: TTYs are devices that can come and go any time, and frequently aren't
4396 * available yet during early boot (consider a USB rs232 dongle...). If for any
4397 * reason the configured console is not ready, we fall back to the default
4400 if (!m
->confirm_spawn
|| path_equal(m
->confirm_spawn
, "/dev/console"))
4401 return m
->confirm_spawn
;
4403 if (stat(m
->confirm_spawn
, &st
) < 0) {
4408 if (!S_ISCHR(st
.st_mode
)) {
4414 return m
->confirm_spawn
;
4417 if (last_errno
!= r
)
4418 last_errno
= log_warning_errno(r
, "Failed to open %s, using default console: %m", m
->confirm_spawn
);
4420 return "/dev/console";
4423 void manager_set_first_boot(Manager
*m
, bool b
) {
4426 if (!MANAGER_IS_SYSTEM(m
))
4429 if (m
->first_boot
!= (int) b
) {
4431 (void) touch("/run/systemd/first-boot");
4433 (void) unlink("/run/systemd/first-boot");
4439 void manager_disable_confirm_spawn(void) {
4440 (void) touch("/run/systemd/confirm_spawn_disabled");
4443 static bool manager_should_show_status(Manager
*m
, StatusType type
) {
4446 if (!MANAGER_IS_SYSTEM(m
))
4449 if (m
->no_console_output
)
4452 if (!IN_SET(manager_state(m
), MANAGER_INITIALIZING
, MANAGER_STARTING
, MANAGER_STOPPING
))
4455 /* If we cannot find out the status properly, just proceed. */
4456 if (type
!= STATUS_TYPE_EMERGENCY
&& manager_check_ask_password(m
) > 0)
4459 if (type
== STATUS_TYPE_NOTICE
&& m
->show_status
!= SHOW_STATUS_NO
)
4462 return manager_get_show_status_on(m
);
4465 void manager_status_printf(Manager
*m
, StatusType type
, const char *status
, const char *format
, ...) {
4468 /* If m is NULL, assume we're after shutdown and let the messages through. */
4470 if (m
&& !manager_should_show_status(m
, type
))
4473 /* XXX We should totally drop the check for ephemeral here
4474 * and thus effectively make 'Type=idle' pointless. */
4475 if (type
== STATUS_TYPE_EPHEMERAL
&& m
&& m
->n_on_console
> 0)
4478 va_start(ap
, format
);
4479 status_vprintf(status
, SHOW_STATUS_ELLIPSIZE
|(type
== STATUS_TYPE_EPHEMERAL
? SHOW_STATUS_EPHEMERAL
: 0), format
, ap
);
4483 Set
* manager_get_units_needing_mounts_for(Manager
*m
, const char *path
, UnitMountDependencyType t
) {
4486 assert(t
>= 0 && t
< _UNIT_MOUNT_DEPENDENCY_TYPE_MAX
);
4488 if (path_equal(path
, "/"))
4491 return hashmap_get(m
->units_needing_mounts_for
[t
], path
);
4494 int manager_update_failed_units(Manager
*m
, Unit
*u
, bool failed
) {
4499 assert(u
->manager
== m
);
4501 size
= set_size(m
->failed_units
);
4504 r
= set_ensure_put(&m
->failed_units
, NULL
, u
);
4508 (void) set_remove(m
->failed_units
, u
);
4510 if (set_size(m
->failed_units
) != size
)
4511 bus_manager_send_change_signal(m
);
4516 ManagerState
manager_state(Manager
*m
) {
4521 /* Is the special shutdown target active or queued? If so, we are in shutdown state */
4522 u
= manager_get_unit(m
, SPECIAL_SHUTDOWN_TARGET
);
4523 if (u
&& unit_active_or_pending(u
))
4524 return MANAGER_STOPPING
;
4526 /* Did we ever finish booting? If not then we are still starting up */
4527 if (!MANAGER_IS_FINISHED(m
)) {
4529 u
= manager_get_unit(m
, SPECIAL_BASIC_TARGET
);
4530 if (!u
|| !UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u
)))
4531 return MANAGER_INITIALIZING
;
4533 return MANAGER_STARTING
;
4536 if (MANAGER_IS_SYSTEM(m
)) {
4537 /* Are the rescue or emergency targets active or queued? If so we are in maintenance state */
4538 u
= manager_get_unit(m
, SPECIAL_RESCUE_TARGET
);
4539 if (u
&& unit_active_or_pending(u
))
4540 return MANAGER_MAINTENANCE
;
4542 u
= manager_get_unit(m
, SPECIAL_EMERGENCY_TARGET
);
4543 if (u
&& unit_active_or_pending(u
))
4544 return MANAGER_MAINTENANCE
;
4547 /* Are there any failed units? If so, we are in degraded mode */
4548 if (set_size(m
->failed_units
) > 0)
4549 return MANAGER_DEGRADED
;
4551 return MANAGER_RUNNING
;
4554 static void manager_unref_uid_internal(
4558 int (*_clean_ipc
)(uid_t uid
)) {
4562 assert(uid_is_valid(uid
));
4565 /* A generic implementation, covering both manager_unref_uid() and manager_unref_gid(), under the
4566 * assumption that uid_t and gid_t are actually defined the same way, with the same validity rules.
4568 * We store a hashmap where the key is the UID/GID and the value is a 32-bit reference counter, whose
4569 * highest bit is used as flag for marking UIDs/GIDs whose IPC objects to remove when the last
4570 * reference to the UID/GID is dropped. The flag is set to on, once at least one reference from a
4571 * unit where RemoveIPC= is set is added on a UID/GID. It is reset when the UID's/GID's reference
4572 * counter drops to 0 again. */
4574 assert_cc(sizeof(uid_t
) == sizeof(gid_t
));
4575 assert_cc(UID_INVALID
== (uid_t
) GID_INVALID
);
4577 if (uid
== 0) /* We don't keep track of root, and will never destroy it */
4580 c
= PTR_TO_UINT32(hashmap_get(uid_refs
, UID_TO_PTR(uid
)));
4582 n
= c
& ~DESTROY_IPC_FLAG
;
4586 if (destroy_now
&& n
== 0) {
4587 hashmap_remove(uid_refs
, UID_TO_PTR(uid
));
4589 if (c
& DESTROY_IPC_FLAG
) {
4590 log_debug("%s " UID_FMT
" is no longer referenced, cleaning up its IPC.",
4591 _clean_ipc
== clean_ipc_by_uid
? "UID" : "GID",
4593 (void) _clean_ipc(uid
);
4596 c
= n
| (c
& DESTROY_IPC_FLAG
);
4597 assert_se(hashmap_update(uid_refs
, UID_TO_PTR(uid
), UINT32_TO_PTR(c
)) >= 0);
4601 void manager_unref_uid(Manager
*m
, uid_t uid
, bool destroy_now
) {
4602 manager_unref_uid_internal(m
->uid_refs
, uid
, destroy_now
, clean_ipc_by_uid
);
4605 void manager_unref_gid(Manager
*m
, gid_t gid
, bool destroy_now
) {
4606 manager_unref_uid_internal(m
->gid_refs
, (uid_t
) gid
, destroy_now
, clean_ipc_by_gid
);
4609 static int manager_ref_uid_internal(
4618 assert(uid_is_valid(uid
));
4620 /* A generic implementation, covering both manager_ref_uid() and manager_ref_gid(), under the
4621 * assumption that uid_t and gid_t are actually defined the same way, with the same validity
4624 assert_cc(sizeof(uid_t
) == sizeof(gid_t
));
4625 assert_cc(UID_INVALID
== (uid_t
) GID_INVALID
);
4627 if (uid
== 0) /* We don't keep track of root, and will never destroy it */
4630 r
= hashmap_ensure_allocated(uid_refs
, &trivial_hash_ops
);
4634 c
= PTR_TO_UINT32(hashmap_get(*uid_refs
, UID_TO_PTR(uid
)));
4636 n
= c
& ~DESTROY_IPC_FLAG
;
4639 if (n
& DESTROY_IPC_FLAG
) /* check for overflow */
4642 c
= n
| (c
& DESTROY_IPC_FLAG
) | (clean_ipc
? DESTROY_IPC_FLAG
: 0);
4644 return hashmap_replace(*uid_refs
, UID_TO_PTR(uid
), UINT32_TO_PTR(c
));
4647 int manager_ref_uid(Manager
*m
, uid_t uid
, bool clean_ipc
) {
4648 return manager_ref_uid_internal(&m
->uid_refs
, uid
, clean_ipc
);
4651 int manager_ref_gid(Manager
*m
, gid_t gid
, bool clean_ipc
) {
4652 return manager_ref_uid_internal(&m
->gid_refs
, (uid_t
) gid
, clean_ipc
);
4655 static void manager_vacuum_uid_refs_internal(
4657 int (*_clean_ipc
)(uid_t uid
)) {
4663 HASHMAP_FOREACH_KEY(p
, k
, uid_refs
) {
4667 uid
= PTR_TO_UID(k
);
4668 c
= PTR_TO_UINT32(p
);
4670 n
= c
& ~DESTROY_IPC_FLAG
;
4674 if (c
& DESTROY_IPC_FLAG
) {
4675 log_debug("Found unreferenced %s " UID_FMT
" after reload/reexec. Cleaning up.",
4676 _clean_ipc
== clean_ipc_by_uid
? "UID" : "GID",
4678 (void) _clean_ipc(uid
);
4681 assert_se(hashmap_remove(uid_refs
, k
) == p
);
4685 static void manager_vacuum_uid_refs(Manager
*m
) {
4686 manager_vacuum_uid_refs_internal(m
->uid_refs
, clean_ipc_by_uid
);
4689 static void manager_vacuum_gid_refs(Manager
*m
) {
4690 manager_vacuum_uid_refs_internal(m
->gid_refs
, clean_ipc_by_gid
);
4693 static void manager_vacuum(Manager
*m
) {
4696 /* Release any dynamic users no longer referenced */
4697 dynamic_user_vacuum(m
, true);
4699 /* Release any references to UIDs/GIDs no longer referenced, and destroy any IPC owned by them */
4700 manager_vacuum_uid_refs(m
);
4701 manager_vacuum_gid_refs(m
);
4703 /* Release any runtimes no longer referenced */
4704 exec_shared_runtime_vacuum(m
);
4707 int manager_dispatch_user_lookup_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
4711 char unit_name
[UNIT_NAME_MAX
+1];
4714 Manager
*m
= userdata
;
4722 /* Invoked whenever a child process succeeded resolving its user/group to use and sent us the
4723 * resulting UID/GID in a datagram. We parse the datagram here and pass it off to the unit, so that
4724 * it can add a reference to the UID/GID so that it can destroy the UID/GID's IPC objects when the
4725 * reference counter drops to 0. */
4727 l
= recv(fd
, &buffer
, sizeof(buffer
), MSG_DONTWAIT
);
4729 if (ERRNO_IS_TRANSIENT(errno
))
4732 return log_error_errno(errno
, "Failed to read from user lookup fd: %m");
4735 if ((size_t) l
<= offsetof(struct buffer
, unit_name
)) {
4736 log_warning("Received too short user lookup message, ignoring.");
4740 if ((size_t) l
> offsetof(struct buffer
, unit_name
) + UNIT_NAME_MAX
) {
4741 log_warning("Received too long user lookup message, ignoring.");
4745 if (!uid_is_valid(buffer
.uid
) && !gid_is_valid(buffer
.gid
)) {
4746 log_warning("Got user lookup message with invalid UID/GID pair, ignoring.");
4750 n
= (size_t) l
- offsetof(struct buffer
, unit_name
);
4751 if (memchr(buffer
.unit_name
, 0, n
)) {
4752 log_warning("Received lookup message with embedded NUL character, ignoring.");
4756 buffer
.unit_name
[n
] = 0;
4757 u
= manager_get_unit(m
, buffer
.unit_name
);
4759 log_debug("Got user lookup message but unit doesn't exist, ignoring.");
4763 log_unit_debug(u
, "User lookup succeeded: uid=" UID_FMT
" gid=" GID_FMT
, buffer
.uid
, buffer
.gid
);
4765 unit_notify_user_lookup(u
, buffer
.uid
, buffer
.gid
);
4769 static int short_uid_range(const char *path
) {
4770 _cleanup_(uid_range_freep
) UidRange
*p
= NULL
;
4775 /* Taint systemd if we the UID range assigned to this environment doesn't at least cover 0…65534,
4776 * i.e. from root to nobody. */
4778 r
= uid_range_load_userns(&p
, path
);
4779 if (ERRNO_IS_NEG_NOT_SUPPORTED(r
))
4782 return log_debug_errno(r
, "Failed to load %s: %m", path
);
4784 return !uid_range_covers(p
, 0, 65535);
4787 char* manager_taint_string(const Manager
*m
) {
4788 /* Returns a "taint string", e.g. "local-hwclock:var-run-bad". Only things that are detected at
4789 * runtime should be tagged here. For stuff that is known during compilation, emit a warning in the
4790 * configuration phase. */
4794 const char* stage
[12] = {};
4797 _cleanup_free_
char *usrbin
= NULL
;
4798 if (readlink_malloc("/bin", &usrbin
) < 0 || !PATH_IN_SET(usrbin
, "usr/bin", "/usr/bin"))
4799 stage
[n
++] = "unmerged-usr";
4801 if (access("/proc/cgroups", F_OK
) < 0)
4802 stage
[n
++] = "cgroups-missing";
4804 if (cg_all_unified() == 0)
4805 stage
[n
++] = "cgroupsv1";
4807 if (clock_is_localtime(NULL
) > 0)
4808 stage
[n
++] = "local-hwclock";
4810 if (os_release_support_ended(NULL
, /* quiet= */ true, NULL
) > 0)
4811 stage
[n
++] = "support-ended";
4813 _cleanup_free_
char *destination
= NULL
;
4814 if (readlink_malloc("/var/run", &destination
) < 0 ||
4815 !PATH_IN_SET(destination
, "../run", "/run"))
4816 stage
[n
++] = "var-run-bad";
4818 _cleanup_free_
char *overflowuid
= NULL
, *overflowgid
= NULL
;
4819 if (read_one_line_file("/proc/sys/kernel/overflowuid", &overflowuid
) >= 0 &&
4820 !streq(overflowuid
, "65534"))
4821 stage
[n
++] = "overflowuid-not-65534";
4822 if (read_one_line_file("/proc/sys/kernel/overflowgid", &overflowgid
) >= 0 &&
4823 !streq(overflowgid
, "65534"))
4824 stage
[n
++] = "overflowgid-not-65534";
4827 assert_se(uname(&uts
) >= 0);
4828 if (strverscmp_improved(uts
.release
, KERNEL_BASELINE_VERSION
) < 0)
4829 stage
[n
++] = "old-kernel";
4831 if (short_uid_range("/proc/self/uid_map") > 0)
4832 stage
[n
++] = "short-uid-range";
4833 if (short_uid_range("/proc/self/gid_map") > 0)
4834 stage
[n
++] = "short-gid-range";
4836 assert(n
< ELEMENTSOF(stage
) - 1); /* One extra for NULL terminator */
4838 return strv_join((char**) stage
, ":");
4841 void manager_ref_console(Manager
*m
) {
4847 void manager_unref_console(Manager
*m
) {
4849 assert(m
->n_on_console
> 0);
4852 if (m
->n_on_console
== 0)
4853 m
->no_console_output
= false; /* unset no_console_output flag, since the console is definitely free now */
4856 void manager_override_log_level(Manager
*m
, int level
) {
4857 _cleanup_free_
char *s
= NULL
;
4860 if (!m
->log_level_overridden
) {
4861 m
->original_log_level
= log_get_max_level();
4862 m
->log_level_overridden
= true;
4865 (void) log_level_to_string_alloc(level
, &s
);
4866 log_info("Setting log level to %s.", strna(s
));
4868 log_set_max_level(level
);
4871 void manager_restore_original_log_level(Manager
*m
) {
4872 _cleanup_free_
char *s
= NULL
;
4875 if (!m
->log_level_overridden
)
4878 (void) log_level_to_string_alloc(m
->original_log_level
, &s
);
4879 log_info("Restoring log level to original (%s).", strna(s
));
4881 log_set_max_level(m
->original_log_level
);
4882 m
->log_level_overridden
= false;
4885 void manager_override_log_target(Manager
*m
, LogTarget target
) {
4888 if (!m
->log_target_overridden
) {
4889 m
->original_log_target
= log_get_target();
4890 m
->log_target_overridden
= true;
4893 log_info("Setting log target to %s.", log_target_to_string(target
));
4894 log_set_target(target
);
4897 void manager_restore_original_log_target(Manager
*m
) {
4900 if (!m
->log_target_overridden
)
4903 log_info("Restoring log target to original %s.", log_target_to_string(m
->original_log_target
));
4905 log_set_target(m
->original_log_target
);
4906 m
->log_target_overridden
= false;
4909 ManagerTimestamp
manager_timestamp_initrd_mangle(ManagerTimestamp s
) {
4911 s
>= MANAGER_TIMESTAMP_SECURITY_START
&&
4912 s
<= MANAGER_TIMESTAMP_UNITS_LOAD_FINISH
)
4913 return s
- MANAGER_TIMESTAMP_SECURITY_START
+ MANAGER_TIMESTAMP_INITRD_SECURITY_START
;
4917 int manager_allocate_idle_pipe(Manager
*m
) {
4922 if (m
->idle_pipe
[0] >= 0) {
4923 assert(m
->idle_pipe
[1] >= 0);
4924 assert(m
->idle_pipe
[2] >= 0);
4925 assert(m
->idle_pipe
[3] >= 0);
4929 assert(m
->idle_pipe
[1] < 0);
4930 assert(m
->idle_pipe
[2] < 0);
4931 assert(m
->idle_pipe
[3] < 0);
4933 r
= RET_NERRNO(pipe2(m
->idle_pipe
+ 0, O_NONBLOCK
|O_CLOEXEC
));
4937 r
= RET_NERRNO(pipe2(m
->idle_pipe
+ 2, O_NONBLOCK
|O_CLOEXEC
));
4939 safe_close_pair(m
->idle_pipe
+ 0);
4946 void unit_defaults_init(UnitDefaults
*defaults
, RuntimeScope scope
) {
4949 assert(scope
< _RUNTIME_SCOPE_MAX
);
4951 *defaults
= (UnitDefaults
) {
4952 .std_output
= EXEC_OUTPUT_JOURNAL
,
4953 .std_error
= EXEC_OUTPUT_INHERIT
,
4954 .restart_usec
= DEFAULT_RESTART_USEC
,
4955 .timeout_start_usec
= manager_default_timeout(scope
),
4956 .timeout_stop_usec
= manager_default_timeout(scope
),
4957 .timeout_abort_usec
= manager_default_timeout(scope
),
4958 .timeout_abort_set
= false,
4959 .device_timeout_usec
= manager_default_timeout(scope
),
4960 .start_limit_interval
= DEFAULT_START_LIMIT_INTERVAL
,
4961 .start_limit_burst
= DEFAULT_START_LIMIT_BURST
,
4963 /* On 4.15+ with unified hierarchy, CPU accounting is essentially free as it doesn't require the CPU
4964 * controller to be enabled, so the default is to enable it unless we got told otherwise. */
4965 .cpu_accounting
= cpu_accounting_is_cheap(),
4966 .memory_accounting
= MEMORY_ACCOUNTING_DEFAULT
,
4967 .io_accounting
= false,
4968 .blockio_accounting
= false,
4969 .tasks_accounting
= true,
4970 .ip_accounting
= false,
4972 .tasks_max
= DEFAULT_TASKS_MAX
,
4973 .timer_accuracy_usec
= 1 * USEC_PER_MINUTE
,
4975 .memory_pressure_watch
= CGROUP_PRESSURE_WATCH_AUTO
,
4976 .memory_pressure_threshold_usec
= MEMORY_PRESSURE_DEFAULT_THRESHOLD_USEC
,
4978 .oom_policy
= OOM_STOP
,
4979 .oom_score_adjust_set
= false,
4983 void unit_defaults_done(UnitDefaults
*defaults
) {
4986 defaults
->smack_process_label
= mfree(defaults
->smack_process_label
);
4987 rlimit_free_all(defaults
->rlimit
);
4990 LogTarget
manager_get_executor_log_target(Manager
*m
) {
4993 /* If journald is not available tell sd-executor to go to kmsg, as it might be starting journald */
4995 if (manager_journal_is_running(m
))
4996 return log_get_target();
4998 return LOG_TARGET_KMSG
;
5001 static const char *const manager_state_table
[_MANAGER_STATE_MAX
] = {
5002 [MANAGER_INITIALIZING
] = "initializing",
5003 [MANAGER_STARTING
] = "starting",
5004 [MANAGER_RUNNING
] = "running",
5005 [MANAGER_DEGRADED
] = "degraded",
5006 [MANAGER_MAINTENANCE
] = "maintenance",
5007 [MANAGER_STOPPING
] = "stopping",
5010 DEFINE_STRING_TABLE_LOOKUP(manager_state
, ManagerState
);
5012 static const char *const manager_timestamp_table
[_MANAGER_TIMESTAMP_MAX
] = {
5013 [MANAGER_TIMESTAMP_FIRMWARE
] = "firmware",
5014 [MANAGER_TIMESTAMP_LOADER
] = "loader",
5015 [MANAGER_TIMESTAMP_KERNEL
] = "kernel",
5016 [MANAGER_TIMESTAMP_INITRD
] = "initrd",
5017 [MANAGER_TIMESTAMP_USERSPACE
] = "userspace",
5018 [MANAGER_TIMESTAMP_FINISH
] = "finish",
5019 [MANAGER_TIMESTAMP_SECURITY_START
] = "security-start",
5020 [MANAGER_TIMESTAMP_SECURITY_FINISH
] = "security-finish",
5021 [MANAGER_TIMESTAMP_GENERATORS_START
] = "generators-start",
5022 [MANAGER_TIMESTAMP_GENERATORS_FINISH
] = "generators-finish",
5023 [MANAGER_TIMESTAMP_UNITS_LOAD_START
] = "units-load-start",
5024 [MANAGER_TIMESTAMP_UNITS_LOAD_FINISH
] = "units-load-finish",
5025 [MANAGER_TIMESTAMP_UNITS_LOAD
] = "units-load",
5026 [MANAGER_TIMESTAMP_INITRD_SECURITY_START
] = "initrd-security-start",
5027 [MANAGER_TIMESTAMP_INITRD_SECURITY_FINISH
] = "initrd-security-finish",
5028 [MANAGER_TIMESTAMP_INITRD_GENERATORS_START
] = "initrd-generators-start",
5029 [MANAGER_TIMESTAMP_INITRD_GENERATORS_FINISH
] = "initrd-generators-finish",
5030 [MANAGER_TIMESTAMP_INITRD_UNITS_LOAD_START
] = "initrd-units-load-start",
5031 [MANAGER_TIMESTAMP_INITRD_UNITS_LOAD_FINISH
] = "initrd-units-load-finish",
5034 DEFINE_STRING_TABLE_LOOKUP(manager_timestamp
, ManagerTimestamp
);
5036 static const char* const oom_policy_table
[_OOM_POLICY_MAX
] = {
5037 [OOM_CONTINUE
] = "continue",
5038 [OOM_STOP
] = "stop",
5039 [OOM_KILL
] = "kill",
5042 DEFINE_STRING_TABLE_LOOKUP(oom_policy
, OOMPolicy
);