1 /* SPDX-License-Identifier: LGPL-2.1+ */
10 #include <sys/inotify.h>
11 #include <sys/ioctl.h>
12 #include <sys/reboot.h>
13 #include <sys/timerfd.h>
21 #include "sd-daemon.h"
22 #include "sd-messages.h"
25 #include "alloc-util.h"
26 #include "all-units.h"
28 #include "boot-timestamps.h"
29 #include "bus-common-errors.h"
30 #include "bus-error.h"
31 #include "bus-kernel.h"
33 #include "clean-ipc.h"
34 #include "clock-util.h"
36 #include "dbus-manager.h"
37 #include "dbus-unit.h"
39 #include "dirent-util.h"
42 #include "exec-util.h"
44 #include "exit-status.h"
51 #include "locale-setup.h"
57 #include "parse-util.h"
58 #include "path-lookup.h"
59 #include "path-util.h"
60 #include "process-util.h"
61 #include "ratelimit.h"
62 #include "rlimit-util.h"
64 #include "signal-util.h"
65 #include "socket-util.h"
67 #include "stat-util.h"
68 #include "string-table.h"
69 #include "string-util.h"
72 #include "syslog-util.h"
73 #include "terminal-util.h"
74 #include "time-util.h"
75 #include "transaction.h"
76 #include "umask-util.h"
77 #include "unit-name.h"
78 #include "user-util.h"
83 #define NOTIFY_RCVBUF_SIZE (8*1024*1024)
84 #define CGROUPS_AGENT_RCVBUF_SIZE (8*1024*1024)
86 /* Initial delay and the interval for printing status messages about running jobs */
87 #define JOBS_IN_PROGRESS_WAIT_USEC (5*USEC_PER_SEC)
88 #define JOBS_IN_PROGRESS_PERIOD_USEC (USEC_PER_SEC / 3)
89 #define JOBS_IN_PROGRESS_PERIOD_DIVISOR 3
91 /* If there are more than 1K bus messages queue across our API and direct busses, then let's not add more on top until
92 * the queue gets more empty. */
93 #define MANAGER_BUS_BUSY_THRESHOLD 1024LU
95 /* How many units and jobs to process of the bus queue before returning to the event loop. */
96 #define MANAGER_BUS_MESSAGE_BUDGET 100U
98 static int manager_dispatch_notify_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
99 static int manager_dispatch_cgroups_agent_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
100 static int manager_dispatch_signal_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
101 static int manager_dispatch_time_change_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
102 static int manager_dispatch_idle_pipe_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
103 static int manager_dispatch_user_lookup_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
104 static int manager_dispatch_jobs_in_progress(sd_event_source
*source
, usec_t usec
, void *userdata
);
105 static int manager_dispatch_run_queue(sd_event_source
*source
, void *userdata
);
106 static int manager_dispatch_sigchld(sd_event_source
*source
, void *userdata
);
107 static int manager_dispatch_timezone_change(sd_event_source
*source
, const struct inotify_event
*event
, void *userdata
);
108 static int manager_run_environment_generators(Manager
*m
);
109 static int manager_run_generators(Manager
*m
);
111 static void manager_watch_jobs_in_progress(Manager
*m
) {
117 /* We do not want to show the cylon animation if the user
118 * needs to confirm service executions otherwise confirmation
119 * messages will be screwed by the cylon animation. */
120 if (!manager_is_confirm_spawn_disabled(m
))
123 if (m
->jobs_in_progress_event_source
)
126 next
= now(CLOCK_MONOTONIC
) + JOBS_IN_PROGRESS_WAIT_USEC
;
127 r
= sd_event_add_time(
129 &m
->jobs_in_progress_event_source
,
132 manager_dispatch_jobs_in_progress
, m
);
136 (void) sd_event_source_set_description(m
->jobs_in_progress_event_source
, "manager-jobs-in-progress");
139 #define CYLON_BUFFER_EXTRA (2*STRLEN(ANSI_RED) + STRLEN(ANSI_HIGHLIGHT_RED) + 2*STRLEN(ANSI_NORMAL))
141 static void draw_cylon(char buffer
[], size_t buflen
, unsigned width
, unsigned pos
) {
144 assert(buflen
>= CYLON_BUFFER_EXTRA
+ width
+ 1);
145 assert(pos
<= width
+1); /* 0 or width+1 mean that the center light is behind the corner */
149 p
= mempset(p
, ' ', pos
-2);
150 if (log_get_show_color())
151 p
= stpcpy(p
, ANSI_RED
);
155 if (pos
> 0 && pos
<= width
) {
156 if (log_get_show_color())
157 p
= stpcpy(p
, ANSI_HIGHLIGHT_RED
);
161 if (log_get_show_color())
162 p
= stpcpy(p
, ANSI_NORMAL
);
165 if (log_get_show_color())
166 p
= stpcpy(p
, ANSI_RED
);
169 p
= mempset(p
, ' ', width
-1-pos
);
170 if (log_get_show_color())
171 strcpy(p
, ANSI_NORMAL
);
175 void manager_flip_auto_status(Manager
*m
, bool enable
) {
179 if (m
->show_status
== SHOW_STATUS_AUTO
)
180 manager_set_show_status(m
, SHOW_STATUS_TEMPORARY
);
182 if (m
->show_status
== SHOW_STATUS_TEMPORARY
)
183 manager_set_show_status(m
, SHOW_STATUS_AUTO
);
187 static void manager_print_jobs_in_progress(Manager
*m
) {
188 _cleanup_free_
char *job_of_n
= NULL
;
191 unsigned counter
= 0, print_nr
;
192 char cylon
[6 + CYLON_BUFFER_EXTRA
+ 1];
194 char time
[FORMAT_TIMESPAN_MAX
], limit
[FORMAT_TIMESPAN_MAX
] = "no limit";
198 assert(m
->n_running_jobs
> 0);
200 manager_flip_auto_status(m
, true);
202 print_nr
= (m
->jobs_in_progress_iteration
/ JOBS_IN_PROGRESS_PERIOD_DIVISOR
) % m
->n_running_jobs
;
204 HASHMAP_FOREACH(j
, m
->jobs
, i
)
205 if (j
->state
== JOB_RUNNING
&& counter
++ == print_nr
)
208 /* m->n_running_jobs must be consistent with the contents of m->jobs,
209 * so the above loop must have succeeded in finding j. */
210 assert(counter
== print_nr
+ 1);
213 cylon_pos
= m
->jobs_in_progress_iteration
% 14;
215 cylon_pos
= 14 - cylon_pos
;
216 draw_cylon(cylon
, sizeof(cylon
), 6, cylon_pos
);
218 m
->jobs_in_progress_iteration
++;
220 if (m
->n_running_jobs
> 1) {
221 if (asprintf(&job_of_n
, "(%u of %u) ", counter
, m
->n_running_jobs
) < 0)
225 format_timespan(time
, sizeof(time
), now(CLOCK_MONOTONIC
) - j
->begin_usec
, 1*USEC_PER_SEC
);
226 if (job_get_timeout(j
, &x
) > 0)
227 format_timespan(limit
, sizeof(limit
), x
- j
->begin_usec
, 1*USEC_PER_SEC
);
229 manager_status_printf(m
, STATUS_TYPE_EPHEMERAL
, cylon
,
230 "%sA %s job is running for %s (%s / %s)",
232 job_type_to_string(j
->type
),
233 unit_description(j
->unit
),
237 static int have_ask_password(void) {
238 _cleanup_closedir_
DIR *dir
;
241 dir
= opendir("/run/systemd/ask-password");
249 FOREACH_DIRENT_ALL(de
, dir
, return -errno
) {
250 if (startswith(de
->d_name
, "ask."))
256 static int manager_dispatch_ask_password_fd(sd_event_source
*source
,
257 int fd
, uint32_t revents
, void *userdata
) {
258 Manager
*m
= userdata
;
264 m
->have_ask_password
= have_ask_password();
265 if (m
->have_ask_password
< 0)
266 /* Log error but continue. Negative have_ask_password
267 * is treated as unknown status. */
268 log_error_errno(m
->have_ask_password
, "Failed to list /run/systemd/ask-password: %m");
273 static void manager_close_ask_password(Manager
*m
) {
276 m
->ask_password_event_source
= sd_event_source_unref(m
->ask_password_event_source
);
277 m
->ask_password_inotify_fd
= safe_close(m
->ask_password_inotify_fd
);
278 m
->have_ask_password
= -EINVAL
;
281 static int manager_check_ask_password(Manager
*m
) {
286 if (!m
->ask_password_event_source
) {
287 assert(m
->ask_password_inotify_fd
< 0);
289 mkdir_p_label("/run/systemd/ask-password", 0755);
291 m
->ask_password_inotify_fd
= inotify_init1(IN_NONBLOCK
|IN_CLOEXEC
);
292 if (m
->ask_password_inotify_fd
< 0)
293 return log_error_errno(errno
, "inotify_init1() failed: %m");
295 if (inotify_add_watch(m
->ask_password_inotify_fd
, "/run/systemd/ask-password", IN_CREATE
|IN_DELETE
|IN_MOVE
) < 0) {
296 log_error_errno(errno
, "Failed to add watch on /run/systemd/ask-password: %m");
297 manager_close_ask_password(m
);
301 r
= sd_event_add_io(m
->event
, &m
->ask_password_event_source
,
302 m
->ask_password_inotify_fd
, EPOLLIN
,
303 manager_dispatch_ask_password_fd
, m
);
305 log_error_errno(errno
, "Failed to add event source for /run/systemd/ask-password: %m");
306 manager_close_ask_password(m
);
310 (void) sd_event_source_set_description(m
->ask_password_event_source
, "manager-ask-password");
312 /* Queries might have been added meanwhile... */
313 manager_dispatch_ask_password_fd(m
->ask_password_event_source
,
314 m
->ask_password_inotify_fd
, EPOLLIN
, m
);
317 return m
->have_ask_password
;
320 static int manager_watch_idle_pipe(Manager
*m
) {
325 if (m
->idle_pipe_event_source
)
328 if (m
->idle_pipe
[2] < 0)
331 r
= sd_event_add_io(m
->event
, &m
->idle_pipe_event_source
, m
->idle_pipe
[2], EPOLLIN
, manager_dispatch_idle_pipe_fd
, m
);
333 return log_error_errno(r
, "Failed to watch idle pipe: %m");
335 (void) sd_event_source_set_description(m
->idle_pipe_event_source
, "manager-idle-pipe");
340 static void manager_close_idle_pipe(Manager
*m
) {
343 m
->idle_pipe_event_source
= sd_event_source_unref(m
->idle_pipe_event_source
);
345 safe_close_pair(m
->idle_pipe
);
346 safe_close_pair(m
->idle_pipe
+ 2);
349 static int manager_setup_time_change(Manager
*m
) {
354 if (MANAGER_IS_TEST_RUN(m
))
357 m
->time_change_event_source
= sd_event_source_unref(m
->time_change_event_source
);
358 m
->time_change_fd
= safe_close(m
->time_change_fd
);
360 m
->time_change_fd
= time_change_fd();
361 if (m
->time_change_fd
< 0)
362 return log_error_errno(m
->time_change_fd
, "Failed to create timer change timer fd: %m");
364 r
= sd_event_add_io(m
->event
, &m
->time_change_event_source
, m
->time_change_fd
, EPOLLIN
, manager_dispatch_time_change_fd
, m
);
366 return log_error_errno(r
, "Failed to create time change event source: %m");
368 /* Schedule this slightly earlier than the .timer event sources */
369 r
= sd_event_source_set_priority(m
->time_change_event_source
, SD_EVENT_PRIORITY_NORMAL
-1);
371 return log_error_errno(r
, "Failed to set priority of time change event sources: %m");
373 (void) sd_event_source_set_description(m
->time_change_event_source
, "manager-time-change");
375 log_debug("Set up TFD_TIMER_CANCEL_ON_SET timerfd.");
380 static int manager_read_timezone_stat(Manager
*m
) {
386 /* Read the current stat() data of /etc/localtime so that we detect changes */
387 if (lstat("/etc/localtime", &st
) < 0) {
388 log_debug_errno(errno
, "Failed to stat /etc/localtime, ignoring: %m");
389 changed
= m
->etc_localtime_accessible
;
390 m
->etc_localtime_accessible
= false;
394 k
= timespec_load(&st
.st_mtim
);
395 changed
= !m
->etc_localtime_accessible
|| k
!= m
->etc_localtime_mtime
;
397 m
->etc_localtime_mtime
= k
;
398 m
->etc_localtime_accessible
= true;
404 static int manager_setup_timezone_change(Manager
*m
) {
405 _cleanup_(sd_event_source_unrefp
) sd_event_source
*new_event
= NULL
;
410 if (MANAGER_IS_TEST_RUN(m
))
413 /* We watch /etc/localtime for three events: change of the link count (which might mean removal from /etc even
414 * though another link might be kept), renames, and file close operations after writing. Note we don't bother
415 * with IN_DELETE_SELF, as that would just report when the inode is removed entirely, i.e. after the link count
416 * went to zero and all fds to it are closed.
418 * Note that we never follow symlinks here. This is a simplification, but should cover almost all cases
421 * Note that we create the new event source first here, before releasing the old one. This should optimize
422 * behaviour as this way sd-event can reuse the old watch in case the inode didn't change. */
424 r
= sd_event_add_inotify(m
->event
, &new_event
, "/etc/localtime",
425 IN_ATTRIB
|IN_MOVE_SELF
|IN_CLOSE_WRITE
|IN_DONT_FOLLOW
, manager_dispatch_timezone_change
, m
);
427 /* If the file doesn't exist yet, subscribe to /etc instead, and wait until it is created either by
428 * O_CREATE or by rename() */
430 log_debug_errno(r
, "/etc/localtime doesn't exist yet, watching /etc instead.");
431 r
= sd_event_add_inotify(m
->event
, &new_event
, "/etc",
432 IN_CREATE
|IN_MOVED_TO
|IN_ONLYDIR
, manager_dispatch_timezone_change
, m
);
435 return log_error_errno(r
, "Failed to create timezone change event source: %m");
437 /* Schedule this slightly earlier than the .timer event sources */
438 r
= sd_event_source_set_priority(new_event
, SD_EVENT_PRIORITY_NORMAL
-1);
440 return log_error_errno(r
, "Failed to set priority of timezone change event sources: %m");
442 sd_event_source_unref(m
->timezone_change_event_source
);
443 m
->timezone_change_event_source
= TAKE_PTR(new_event
);
448 static int enable_special_signals(Manager
*m
) {
449 _cleanup_close_
int fd
= -1;
453 if (MANAGER_IS_TEST_RUN(m
))
456 /* Enable that we get SIGINT on control-alt-del. In containers
457 * this will fail with EPERM (older) or EINVAL (newer), so
459 if (reboot(RB_DISABLE_CAD
) < 0 && !IN_SET(errno
, EPERM
, EINVAL
))
460 log_warning_errno(errno
, "Failed to enable ctrl-alt-del handling: %m");
462 fd
= open_terminal("/dev/tty0", O_RDWR
|O_NOCTTY
|O_CLOEXEC
);
464 /* Support systems without virtual console */
466 log_warning_errno(errno
, "Failed to open /dev/tty0: %m");
468 /* Enable that we get SIGWINCH on kbrequest */
469 if (ioctl(fd
, KDSIGACCEPT
, SIGWINCH
) < 0)
470 log_warning_errno(errno
, "Failed to enable kbrequest handling: %m");
476 #define RTSIG_IF_AVAILABLE(signum) (signum <= SIGRTMAX ? signum : -1)
478 static int manager_setup_signals(Manager
*m
) {
479 struct sigaction sa
= {
480 .sa_handler
= SIG_DFL
,
481 .sa_flags
= SA_NOCLDSTOP
|SA_RESTART
,
488 assert_se(sigaction(SIGCHLD
, &sa
, NULL
) == 0);
490 /* We make liberal use of realtime signals here. On
491 * Linux/glibc we have 30 of them (with the exception of Linux
492 * on hppa, see below), between SIGRTMIN+0 ... SIGRTMIN+30
495 assert_se(sigemptyset(&mask
) == 0);
496 sigset_add_many(&mask
,
497 SIGCHLD
, /* Child died */
498 SIGTERM
, /* Reexecute daemon */
499 SIGHUP
, /* Reload configuration */
500 SIGUSR1
, /* systemd/upstart: reconnect to D-Bus */
501 SIGUSR2
, /* systemd: dump status */
502 SIGINT
, /* Kernel sends us this on control-alt-del */
503 SIGWINCH
, /* Kernel sends us this on kbrequest (alt-arrowup) */
504 SIGPWR
, /* Some kernel drivers and upsd send us this on power failure */
506 SIGRTMIN
+0, /* systemd: start default.target */
507 SIGRTMIN
+1, /* systemd: isolate rescue.target */
508 SIGRTMIN
+2, /* systemd: isolate emergency.target */
509 SIGRTMIN
+3, /* systemd: start halt.target */
510 SIGRTMIN
+4, /* systemd: start poweroff.target */
511 SIGRTMIN
+5, /* systemd: start reboot.target */
512 SIGRTMIN
+6, /* systemd: start kexec.target */
514 /* ... space for more special targets ... */
516 SIGRTMIN
+13, /* systemd: Immediate halt */
517 SIGRTMIN
+14, /* systemd: Immediate poweroff */
518 SIGRTMIN
+15, /* systemd: Immediate reboot */
519 SIGRTMIN
+16, /* systemd: Immediate kexec */
521 /* ... space for more immediate system state changes ... */
523 SIGRTMIN
+20, /* systemd: enable status messages */
524 SIGRTMIN
+21, /* systemd: disable status messages */
525 SIGRTMIN
+22, /* systemd: set log level to LOG_DEBUG */
526 SIGRTMIN
+23, /* systemd: set log level to LOG_INFO */
527 SIGRTMIN
+24, /* systemd: Immediate exit (--user only) */
529 /* .. one free signal here ... */
531 /* Apparently Linux on hppa had fewer RT signals until v3.18,
532 * SIGRTMAX was SIGRTMIN+25, and then SIGRTMIN was lowered,
533 * see commit v3.17-7614-g1f25df2eff.
535 * We cannot unconditionally make use of those signals here,
536 * so let's use a runtime check. Since these commands are
537 * accessible by different means and only really a safety
538 * net, the missing functionality on hppa shouldn't matter.
541 RTSIG_IF_AVAILABLE(SIGRTMIN
+26), /* systemd: set log target to journal-or-kmsg */
542 RTSIG_IF_AVAILABLE(SIGRTMIN
+27), /* systemd: set log target to console */
543 RTSIG_IF_AVAILABLE(SIGRTMIN
+28), /* systemd: set log target to kmsg */
544 RTSIG_IF_AVAILABLE(SIGRTMIN
+29), /* systemd: set log target to syslog-or-kmsg (obsolete) */
546 /* ... one free signal here SIGRTMIN+30 ... */
548 assert_se(sigprocmask(SIG_SETMASK
, &mask
, NULL
) == 0);
550 m
->signal_fd
= signalfd(-1, &mask
, SFD_NONBLOCK
|SFD_CLOEXEC
);
551 if (m
->signal_fd
< 0)
554 r
= sd_event_add_io(m
->event
, &m
->signal_event_source
, m
->signal_fd
, EPOLLIN
, manager_dispatch_signal_fd
, m
);
558 (void) sd_event_source_set_description(m
->signal_event_source
, "manager-signal");
560 /* Process signals a bit earlier than the rest of things, but later than notify_fd processing, so that the
561 * notify processing can still figure out to which process/service a message belongs, before we reap the
562 * process. Also, process this before handling cgroup notifications, so that we always collect child exit
563 * status information before detecting that there's no process in a cgroup. */
564 r
= sd_event_source_set_priority(m
->signal_event_source
, SD_EVENT_PRIORITY_NORMAL
-6);
568 if (MANAGER_IS_SYSTEM(m
))
569 return enable_special_signals(m
);
574 static void manager_sanitize_environment(Manager
*m
) {
577 /* Let's remove some environment variables that we need ourselves to communicate with our clients */
597 /* Let's order the environment alphabetically, just to make it pretty */
598 strv_sort(m
->environment
);
601 static int manager_default_environment(Manager
*m
) {
604 if (MANAGER_IS_SYSTEM(m
)) {
605 /* The system manager always starts with a clean
606 * environment for its children. It does not import
607 * the kernel's or the parents' exported variables.
609 * The initial passed environment is untouched to keep
610 * /proc/self/environ valid; it is used for tagging
611 * the init process inside containers. */
612 m
->environment
= strv_new("PATH=" DEFAULT_PATH
,
615 /* Import locale variables LC_*= from configuration */
616 locale_setup(&m
->environment
);
618 /* The user manager passes its own environment
619 * along to its children. */
620 m
->environment
= strv_copy(environ
);
625 manager_sanitize_environment(m
);
630 static int manager_setup_prefix(Manager
*m
) {
636 static const struct table_entry paths_system
[_EXEC_DIRECTORY_TYPE_MAX
] = {
637 [EXEC_DIRECTORY_RUNTIME
] = { SD_PATH_SYSTEM_RUNTIME
, NULL
},
638 [EXEC_DIRECTORY_STATE
] = { SD_PATH_SYSTEM_STATE_PRIVATE
, NULL
},
639 [EXEC_DIRECTORY_CACHE
] = { SD_PATH_SYSTEM_STATE_CACHE
, NULL
},
640 [EXEC_DIRECTORY_LOGS
] = { SD_PATH_SYSTEM_STATE_LOGS
, NULL
},
641 [EXEC_DIRECTORY_CONFIGURATION
] = { SD_PATH_SYSTEM_CONFIGURATION
, NULL
},
644 static const struct table_entry paths_user
[_EXEC_DIRECTORY_TYPE_MAX
] = {
645 [EXEC_DIRECTORY_RUNTIME
] = { SD_PATH_USER_RUNTIME
, NULL
},
646 [EXEC_DIRECTORY_STATE
] = { SD_PATH_USER_CONFIGURATION
, NULL
},
647 [EXEC_DIRECTORY_CACHE
] = { SD_PATH_USER_STATE_CACHE
, NULL
},
648 [EXEC_DIRECTORY_LOGS
] = { SD_PATH_USER_CONFIGURATION
, "log" },
649 [EXEC_DIRECTORY_CONFIGURATION
] = { SD_PATH_USER_CONFIGURATION
, NULL
},
652 const struct table_entry
*p
;
658 if (MANAGER_IS_SYSTEM(m
))
663 for (i
= 0; i
< _EXEC_DIRECTORY_TYPE_MAX
; i
++) {
664 r
= sd_path_home(p
[i
].type
, p
[i
].suffix
, &m
->prefix
[i
]);
672 static int manager_setup_run_queue(Manager
*m
) {
676 assert(!m
->run_queue_event_source
);
678 r
= sd_event_add_defer(m
->event
, &m
->run_queue_event_source
, manager_dispatch_run_queue
, m
);
682 r
= sd_event_source_set_priority(m
->run_queue_event_source
, SD_EVENT_PRIORITY_IDLE
);
686 r
= sd_event_source_set_enabled(m
->run_queue_event_source
, SD_EVENT_OFF
);
690 (void) sd_event_source_set_description(m
->run_queue_event_source
, "manager-run-queue");
695 static int manager_setup_sigchld_event_source(Manager
*m
) {
699 assert(!m
->sigchld_event_source
);
701 r
= sd_event_add_defer(m
->event
, &m
->sigchld_event_source
, manager_dispatch_sigchld
, m
);
705 r
= sd_event_source_set_priority(m
->sigchld_event_source
, SD_EVENT_PRIORITY_NORMAL
-7);
709 r
= sd_event_source_set_enabled(m
->sigchld_event_source
, SD_EVENT_OFF
);
713 (void) sd_event_source_set_description(m
->sigchld_event_source
, "manager-sigchld");
718 int manager_new(UnitFileScope scope
, ManagerTestRunFlags test_run_flags
, Manager
**_m
) {
719 _cleanup_(manager_freep
) Manager
*m
= NULL
;
723 assert(IN_SET(scope
, UNIT_FILE_SYSTEM
, UNIT_FILE_USER
));
730 .unit_file_scope
= scope
,
731 .objective
= _MANAGER_OBJECTIVE_INVALID
,
733 .default_timer_accuracy_usec
= USEC_PER_MINUTE
,
734 .default_memory_accounting
= MEMORY_ACCOUNTING_DEFAULT
,
735 .default_tasks_accounting
= true,
736 .default_tasks_max
= UINT64_MAX
,
737 .default_timeout_start_usec
= DEFAULT_TIMEOUT_USEC
,
738 .default_timeout_stop_usec
= DEFAULT_TIMEOUT_USEC
,
739 .default_restart_usec
= DEFAULT_RESTART_USEC
,
741 .original_log_level
= -1,
742 .original_log_target
= _LOG_TARGET_INVALID
,
745 .cgroups_agent_fd
= -1,
747 .time_change_fd
= -1,
748 .user_lookup_fds
= { -1, -1 },
749 .private_listen_fd
= -1,
751 .cgroup_inotify_fd
= -1,
752 .pin_cgroupfs_fd
= -1,
753 .ask_password_inotify_fd
= -1,
754 .idle_pipe
= { -1, -1, -1, -1},
756 /* start as id #1, so that we can leave #0 around as "null-like" value */
759 .have_ask_password
= -EINVAL
, /* we don't know */
761 .test_run_flags
= test_run_flags
,
765 if (MANAGER_IS_SYSTEM(m
) && detect_container() <= 0)
766 boot_timestamps(m
->timestamps
+ MANAGER_TIMESTAMP_USERSPACE
,
767 m
->timestamps
+ MANAGER_TIMESTAMP_FIRMWARE
,
768 m
->timestamps
+ MANAGER_TIMESTAMP_LOADER
);
771 /* Prepare log fields we can use for structured logging */
772 if (MANAGER_IS_SYSTEM(m
)) {
773 m
->unit_log_field
= "UNIT=";
774 m
->unit_log_format_string
= "UNIT=%s";
776 m
->invocation_log_field
= "INVOCATION_ID=";
777 m
->invocation_log_format_string
= "INVOCATION_ID=%s";
779 m
->unit_log_field
= "USER_UNIT=";
780 m
->unit_log_format_string
= "USER_UNIT=%s";
782 m
->invocation_log_field
= "USER_INVOCATION_ID=";
783 m
->invocation_log_format_string
= "USER_INVOCATION_ID=%s";
786 /* Reboot immediately if the user hits C-A-D more often than 7x per 2s */
787 RATELIMIT_INIT(m
->ctrl_alt_del_ratelimit
, 2 * USEC_PER_SEC
, 7);
789 r
= manager_default_environment(m
);
793 r
= hashmap_ensure_allocated(&m
->units
, &string_hash_ops
);
797 r
= hashmap_ensure_allocated(&m
->jobs
, NULL
);
801 r
= hashmap_ensure_allocated(&m
->cgroup_unit
, &path_hash_ops
);
805 r
= hashmap_ensure_allocated(&m
->watch_bus
, &string_hash_ops
);
809 r
= manager_setup_prefix(m
);
813 r
= sd_event_default(&m
->event
);
817 r
= manager_setup_run_queue(m
);
821 if (test_run_flags
== MANAGER_TEST_RUN_MINIMAL
) {
822 m
->cgroup_root
= strdup("");
826 r
= manager_setup_signals(m
);
830 r
= manager_setup_cgroup(m
);
834 r
= manager_setup_time_change(m
);
838 r
= manager_read_timezone_stat(m
);
842 (void) manager_setup_timezone_change(m
);
844 r
= manager_setup_sigchld_event_source(m
);
849 if (MANAGER_IS_SYSTEM(m
) && test_run_flags
== 0) {
850 r
= mkdir_label("/run/systemd/units", 0755);
851 if (r
< 0 && r
!= -EEXIST
)
857 dir_is_empty("/usr") > 0;
859 /* Note that we do not set up the notify fd here. We do that after deserialization,
860 * since they might have gotten serialized across the reexec. */
867 static int manager_setup_notify(Manager
*m
) {
870 if (MANAGER_IS_TEST_RUN(m
))
873 if (m
->notify_fd
< 0) {
874 _cleanup_close_
int fd
= -1;
875 union sockaddr_union sa
= {};
876 static const int one
= 1;
879 /* First free all secondary fields */
880 m
->notify_socket
= mfree(m
->notify_socket
);
881 m
->notify_event_source
= sd_event_source_unref(m
->notify_event_source
);
883 fd
= socket(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
|SOCK_NONBLOCK
, 0);
885 return log_error_errno(errno
, "Failed to allocate notification socket: %m");
887 fd_inc_rcvbuf(fd
, NOTIFY_RCVBUF_SIZE
);
889 m
->notify_socket
= strappend(m
->prefix
[EXEC_DIRECTORY_RUNTIME
], "/systemd/notify");
890 if (!m
->notify_socket
)
893 salen
= sockaddr_un_set_path(&sa
.un
, m
->notify_socket
);
895 return log_error_errno(salen
, "Notify socket '%s' not valid for AF_UNIX socket address, refusing.", m
->notify_socket
);
897 (void) mkdir_parents_label(m
->notify_socket
, 0755);
898 (void) unlink(m
->notify_socket
);
900 r
= bind(fd
, &sa
.sa
, salen
);
902 return log_error_errno(errno
, "bind(%s) failed: %m", m
->notify_socket
);
904 r
= setsockopt(fd
, SOL_SOCKET
, SO_PASSCRED
, &one
, sizeof(one
));
906 return log_error_errno(errno
, "SO_PASSCRED failed: %m");
908 m
->notify_fd
= TAKE_FD(fd
);
910 log_debug("Using notification socket %s", m
->notify_socket
);
913 if (!m
->notify_event_source
) {
914 r
= sd_event_add_io(m
->event
, &m
->notify_event_source
, m
->notify_fd
, EPOLLIN
, manager_dispatch_notify_fd
, m
);
916 return log_error_errno(r
, "Failed to allocate notify event source: %m");
918 /* Process notification messages a bit earlier than SIGCHLD, so that we can still identify to which
919 * service an exit message belongs. */
920 r
= sd_event_source_set_priority(m
->notify_event_source
, SD_EVENT_PRIORITY_NORMAL
-8);
922 return log_error_errno(r
, "Failed to set priority of notify event source: %m");
924 (void) sd_event_source_set_description(m
->notify_event_source
, "manager-notify");
930 static int manager_setup_cgroups_agent(Manager
*m
) {
932 static const union sockaddr_union sa
= {
933 .un
.sun_family
= AF_UNIX
,
934 .un
.sun_path
= "/run/systemd/cgroups-agent",
938 /* This creates a listening socket we receive cgroups agent messages on. We do not use D-Bus for delivering
939 * these messages from the cgroups agent binary to PID 1, as the cgroups agent binary is very short-living, and
940 * each instance of it needs a new D-Bus connection. Since D-Bus connections are SOCK_STREAM/AF_UNIX, on
941 * overloaded systems the backlog of the D-Bus socket becomes relevant, as not more than the configured number
942 * of D-Bus connections may be queued until the kernel will start dropping further incoming connections,
943 * possibly resulting in lost cgroups agent messages. To avoid this, we'll use a private SOCK_DGRAM/AF_UNIX
944 * socket, where no backlog is relevant as communication may take place without an actual connect() cycle, and
945 * we thus won't lose messages.
947 * Note that PID 1 will forward the agent message to system bus, so that the user systemd instance may listen
948 * to it. The system instance hence listens on this special socket, but the user instances listen on the system
949 * bus for these messages. */
951 if (MANAGER_IS_TEST_RUN(m
))
954 if (!MANAGER_IS_SYSTEM(m
))
957 r
= cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER
);
959 return log_error_errno(r
, "Failed to determine whether unified cgroups hierarchy is used: %m");
960 if (r
> 0) /* We don't need this anymore on the unified hierarchy */
963 if (m
->cgroups_agent_fd
< 0) {
964 _cleanup_close_
int fd
= -1;
966 /* First free all secondary fields */
967 m
->cgroups_agent_event_source
= sd_event_source_unref(m
->cgroups_agent_event_source
);
969 fd
= socket(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
|SOCK_NONBLOCK
, 0);
971 return log_error_errno(errno
, "Failed to allocate cgroups agent socket: %m");
973 fd_inc_rcvbuf(fd
, CGROUPS_AGENT_RCVBUF_SIZE
);
975 (void) sockaddr_un_unlink(&sa
.un
);
977 /* Only allow root to connect to this socket */
979 r
= bind(fd
, &sa
.sa
, SOCKADDR_UN_LEN(sa
.un
));
981 return log_error_errno(errno
, "bind(%s) failed: %m", sa
.un
.sun_path
);
983 m
->cgroups_agent_fd
= TAKE_FD(fd
);
986 if (!m
->cgroups_agent_event_source
) {
987 r
= sd_event_add_io(m
->event
, &m
->cgroups_agent_event_source
, m
->cgroups_agent_fd
, EPOLLIN
, manager_dispatch_cgroups_agent_fd
, m
);
989 return log_error_errno(r
, "Failed to allocate cgroups agent event source: %m");
991 /* Process cgroups notifications early, but after having processed service notification messages or
992 * SIGCHLD signals, so that a cgroup running empty is always just the last safety net of notification,
993 * and we collected the metadata the notification and SIGCHLD stuff offers first. Also see handling of
994 * cgroup inotify for the unified cgroup stuff. */
995 r
= sd_event_source_set_priority(m
->cgroups_agent_event_source
, SD_EVENT_PRIORITY_NORMAL
-4);
997 return log_error_errno(r
, "Failed to set priority of cgroups agent event source: %m");
999 (void) sd_event_source_set_description(m
->cgroups_agent_event_source
, "manager-cgroups-agent");
1005 static int manager_setup_user_lookup_fd(Manager
*m
) {
1010 /* Set up the socket pair used for passing UID/GID resolution results from forked off processes to PID
1011 * 1. Background: we can't do name lookups (NSS) from PID 1, since it might involve IPC and thus activation,
1012 * and we might hence deadlock on ourselves. Hence we do all user/group lookups asynchronously from the forked
1013 * off processes right before executing the binaries to start. In order to be able to clean up any IPC objects
1014 * created by a unit (see RemoveIPC=) we need to know in PID 1 the used UID/GID of the executed processes,
1015 * hence we establish this communication channel so that forked off processes can pass their UID/GID
1016 * information back to PID 1. The forked off processes send their resolved UID/GID to PID 1 in a simple
1017 * datagram, along with their unit name, so that we can share one communication socket pair among all units for
1020 * You might wonder why we need a communication channel for this that is independent of the usual notification
1021 * socket scheme (i.e. $NOTIFY_SOCKET). The primary difference is about trust: data sent via the $NOTIFY_SOCKET
1022 * channel is only accepted if it originates from the right unit and if reception was enabled for it. The user
1023 * lookup socket OTOH is only accessible by PID 1 and its children until they exec(), and always available.
1025 * Note that this function is called under two circumstances: when we first initialize (in which case we
1026 * allocate both the socket pair and the event source to listen on it), and when we deserialize after a reload
1027 * (in which case the socket pair already exists but we still need to allocate the event source for it). */
1029 if (m
->user_lookup_fds
[0] < 0) {
1031 /* Free all secondary fields */
1032 safe_close_pair(m
->user_lookup_fds
);
1033 m
->user_lookup_event_source
= sd_event_source_unref(m
->user_lookup_event_source
);
1035 if (socketpair(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
, 0, m
->user_lookup_fds
) < 0)
1036 return log_error_errno(errno
, "Failed to allocate user lookup socket: %m");
1038 (void) fd_inc_rcvbuf(m
->user_lookup_fds
[0], NOTIFY_RCVBUF_SIZE
);
1041 if (!m
->user_lookup_event_source
) {
1042 r
= sd_event_add_io(m
->event
, &m
->user_lookup_event_source
, m
->user_lookup_fds
[0], EPOLLIN
, manager_dispatch_user_lookup_fd
, m
);
1044 return log_error_errno(errno
, "Failed to allocate user lookup event source: %m");
1046 /* Process even earlier than the notify event source, so that we always know first about valid UID/GID
1048 r
= sd_event_source_set_priority(m
->user_lookup_event_source
, SD_EVENT_PRIORITY_NORMAL
-11);
1050 return log_error_errno(errno
, "Failed to set priority ot user lookup event source: %m");
1052 (void) sd_event_source_set_description(m
->user_lookup_event_source
, "user-lookup");
1058 static unsigned manager_dispatch_cleanup_queue(Manager
*m
) {
1064 while ((u
= m
->cleanup_queue
)) {
1065 assert(u
->in_cleanup_queue
);
1075 GC_OFFSET_IN_PATH
, /* This one is on the path we were traveling */
1076 GC_OFFSET_UNSURE
, /* No clue */
1077 GC_OFFSET_GOOD
, /* We still need this unit */
1078 GC_OFFSET_BAD
, /* We don't need this unit anymore */
1082 static void unit_gc_mark_good(Unit
*u
, unsigned gc_marker
) {
1087 u
->gc_marker
= gc_marker
+ GC_OFFSET_GOOD
;
1089 /* Recursively mark referenced units as GOOD as well */
1090 HASHMAP_FOREACH_KEY(v
, other
, u
->dependencies
[UNIT_REFERENCES
], i
)
1091 if (other
->gc_marker
== gc_marker
+ GC_OFFSET_UNSURE
)
1092 unit_gc_mark_good(other
, gc_marker
);
1095 static void unit_gc_sweep(Unit
*u
, unsigned gc_marker
) {
1103 if (IN_SET(u
->gc_marker
- gc_marker
,
1104 GC_OFFSET_GOOD
, GC_OFFSET_BAD
, GC_OFFSET_UNSURE
, GC_OFFSET_IN_PATH
))
1107 if (u
->in_cleanup_queue
)
1110 if (!unit_may_gc(u
))
1113 u
->gc_marker
= gc_marker
+ GC_OFFSET_IN_PATH
;
1117 HASHMAP_FOREACH_KEY(v
, other
, u
->dependencies
[UNIT_REFERENCED_BY
], i
) {
1118 unit_gc_sweep(other
, gc_marker
);
1120 if (other
->gc_marker
== gc_marker
+ GC_OFFSET_GOOD
)
1123 if (other
->gc_marker
!= gc_marker
+ GC_OFFSET_BAD
)
1127 if (u
->refs_by_target
) {
1130 LIST_FOREACH(refs_by_target
, ref
, u
->refs_by_target
) {
1131 unit_gc_sweep(ref
->source
, gc_marker
);
1133 if (ref
->source
->gc_marker
== gc_marker
+ GC_OFFSET_GOOD
)
1136 if (ref
->source
->gc_marker
!= gc_marker
+ GC_OFFSET_BAD
)
1144 /* We were unable to find anything out about this entry, so
1145 * let's investigate it later */
1146 u
->gc_marker
= gc_marker
+ GC_OFFSET_UNSURE
;
1147 unit_add_to_gc_queue(u
);
1151 /* We definitely know that this one is not useful anymore, so
1152 * let's mark it for deletion */
1153 u
->gc_marker
= gc_marker
+ GC_OFFSET_BAD
;
1154 unit_add_to_cleanup_queue(u
);
1158 unit_gc_mark_good(u
, gc_marker
);
1161 static unsigned manager_dispatch_gc_unit_queue(Manager
*m
) {
1162 unsigned n
= 0, gc_marker
;
1167 /* log_debug("Running GC..."); */
1169 m
->gc_marker
+= _GC_OFFSET_MAX
;
1170 if (m
->gc_marker
+ _GC_OFFSET_MAX
<= _GC_OFFSET_MAX
)
1173 gc_marker
= m
->gc_marker
;
1175 while ((u
= m
->gc_unit_queue
)) {
1176 assert(u
->in_gc_queue
);
1178 unit_gc_sweep(u
, gc_marker
);
1180 LIST_REMOVE(gc_queue
, m
->gc_unit_queue
, u
);
1181 u
->in_gc_queue
= false;
1185 if (IN_SET(u
->gc_marker
- gc_marker
,
1186 GC_OFFSET_BAD
, GC_OFFSET_UNSURE
)) {
1188 log_unit_debug(u
, "Collecting.");
1189 u
->gc_marker
= gc_marker
+ GC_OFFSET_BAD
;
1190 unit_add_to_cleanup_queue(u
);
1197 static unsigned manager_dispatch_gc_job_queue(Manager
*m
) {
1203 while ((j
= m
->gc_job_queue
)) {
1204 assert(j
->in_gc_queue
);
1206 LIST_REMOVE(gc_queue
, m
->gc_job_queue
, j
);
1207 j
->in_gc_queue
= false;
1214 log_unit_debug(j
->unit
, "Collecting job.");
1215 (void) job_finish_and_invalidate(j
, JOB_COLLECTED
, false, false);
1221 static unsigned manager_dispatch_stop_when_unneeded_queue(Manager
*m
) {
1228 while ((u
= m
->stop_when_unneeded_queue
)) {
1229 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
1230 assert(m
->stop_when_unneeded_queue
);
1232 assert(u
->in_stop_when_unneeded_queue
);
1233 LIST_REMOVE(stop_when_unneeded_queue
, m
->stop_when_unneeded_queue
, u
);
1234 u
->in_stop_when_unneeded_queue
= false;
1238 if (!unit_is_unneeded(u
))
1241 log_unit_debug(u
, "Unit is not needed anymore.");
1243 /* If stopping a unit fails continuously we might enter a stop loop here, hence stop acting on the
1244 * service being unnecessary after a while. */
1246 if (!ratelimit_below(&u
->auto_stop_ratelimit
)) {
1247 log_unit_warning(u
, "Unit not needed anymore, but not stopping since we tried this too often recently.");
1251 /* Ok, nobody needs us anymore. Sniff. Then let's commit suicide */
1252 r
= manager_add_job(u
->manager
, JOB_STOP
, u
, JOB_FAIL
, &error
, NULL
);
1254 log_unit_warning_errno(u
, r
, "Failed to enqueue stop job, ignoring: %s", bus_error_message(&error
, r
));
1260 static void manager_clear_jobs_and_units(Manager
*m
) {
1265 while ((u
= hashmap_first(m
->units
)))
1268 manager_dispatch_cleanup_queue(m
);
1270 assert(!m
->load_queue
);
1271 assert(!m
->run_queue
);
1272 assert(!m
->dbus_unit_queue
);
1273 assert(!m
->dbus_job_queue
);
1274 assert(!m
->cleanup_queue
);
1275 assert(!m
->gc_unit_queue
);
1276 assert(!m
->gc_job_queue
);
1277 assert(!m
->stop_when_unneeded_queue
);
1279 assert(hashmap_isempty(m
->jobs
));
1280 assert(hashmap_isempty(m
->units
));
1282 m
->n_on_console
= 0;
1283 m
->n_running_jobs
= 0;
1284 m
->n_installed_jobs
= 0;
1285 m
->n_failed_jobs
= 0;
1288 Manager
* manager_free(Manager
*m
) {
1289 ExecDirectoryType dt
;
1295 manager_clear_jobs_and_units(m
);
1297 for (c
= 0; c
< _UNIT_TYPE_MAX
; c
++)
1298 if (unit_vtable
[c
]->shutdown
)
1299 unit_vtable
[c
]->shutdown(m
);
1301 /* Keep the cgroup hierarchy in place except when we know we are going down for good */
1302 manager_shutdown_cgroup(m
, IN_SET(m
->objective
, MANAGER_EXIT
, MANAGER_REBOOT
, MANAGER_POWEROFF
, MANAGER_HALT
, MANAGER_KEXEC
));
1304 lookup_paths_flush_generator(&m
->lookup_paths
);
1308 exec_runtime_vacuum(m
);
1309 hashmap_free(m
->exec_runtime_by_id
);
1311 dynamic_user_vacuum(m
, false);
1312 hashmap_free(m
->dynamic_users
);
1314 hashmap_free(m
->units
);
1315 hashmap_free(m
->units_by_invocation_id
);
1316 hashmap_free(m
->jobs
);
1317 hashmap_free(m
->watch_pids
);
1318 hashmap_free(m
->watch_bus
);
1320 set_free(m
->startup_units
);
1321 set_free(m
->failed_units
);
1323 sd_event_source_unref(m
->signal_event_source
);
1324 sd_event_source_unref(m
->sigchld_event_source
);
1325 sd_event_source_unref(m
->notify_event_source
);
1326 sd_event_source_unref(m
->cgroups_agent_event_source
);
1327 sd_event_source_unref(m
->time_change_event_source
);
1328 sd_event_source_unref(m
->timezone_change_event_source
);
1329 sd_event_source_unref(m
->jobs_in_progress_event_source
);
1330 sd_event_source_unref(m
->run_queue_event_source
);
1331 sd_event_source_unref(m
->user_lookup_event_source
);
1332 sd_event_source_unref(m
->sync_bus_names_event_source
);
1334 safe_close(m
->signal_fd
);
1335 safe_close(m
->notify_fd
);
1336 safe_close(m
->cgroups_agent_fd
);
1337 safe_close(m
->time_change_fd
);
1338 safe_close_pair(m
->user_lookup_fds
);
1340 manager_close_ask_password(m
);
1342 manager_close_idle_pipe(m
);
1344 sd_event_unref(m
->event
);
1346 free(m
->notify_socket
);
1348 lookup_paths_free(&m
->lookup_paths
);
1349 strv_free(m
->environment
);
1351 hashmap_free(m
->cgroup_unit
);
1352 set_free_free(m
->unit_path_cache
);
1354 free(m
->switch_root
);
1355 free(m
->switch_root_init
);
1357 rlimit_free_all(m
->rlimit
);
1359 assert(hashmap_isempty(m
->units_requiring_mounts_for
));
1360 hashmap_free(m
->units_requiring_mounts_for
);
1362 hashmap_free(m
->uid_refs
);
1363 hashmap_free(m
->gid_refs
);
1365 for (dt
= 0; dt
< _EXEC_DIRECTORY_TYPE_MAX
; dt
++)
1366 m
->prefix
[dt
] = mfree(m
->prefix
[dt
]);
1371 static void manager_enumerate_perpetual(Manager
*m
) {
1376 /* Let's ask every type to load all units from disk/kernel that it might know */
1377 for (c
= 0; c
< _UNIT_TYPE_MAX
; c
++) {
1378 if (!unit_type_supported(c
)) {
1379 log_debug("Unit type .%s is not supported on this system.", unit_type_to_string(c
));
1383 if (unit_vtable
[c
]->enumerate_perpetual
)
1384 unit_vtable
[c
]->enumerate_perpetual(m
);
1388 static void manager_enumerate(Manager
*m
) {
1393 /* Let's ask every type to load all units from disk/kernel that it might know */
1394 for (c
= 0; c
< _UNIT_TYPE_MAX
; c
++) {
1395 if (!unit_type_supported(c
)) {
1396 log_debug("Unit type .%s is not supported on this system.", unit_type_to_string(c
));
1400 if (unit_vtable
[c
]->enumerate
)
1401 unit_vtable
[c
]->enumerate(m
);
1404 manager_dispatch_load_queue(m
);
1407 static void manager_coldplug(Manager
*m
) {
1415 log_debug("Invoking unit coldplug() handlers…");
1417 /* Let's place the units back into their deserialized state */
1418 HASHMAP_FOREACH_KEY(u
, k
, m
->units
, i
) {
1420 /* ignore aliases */
1424 r
= unit_coldplug(u
);
1426 log_warning_errno(r
, "We couldn't coldplug %s, proceeding anyway: %m", u
->id
);
1430 static void manager_catchup(Manager
*m
) {
1437 log_debug("Invoking unit catchup() handlers…");
1439 /* Let's catch up on any state changes that happened while we were reloading/reexecing */
1440 HASHMAP_FOREACH_KEY(u
, k
, m
->units
, i
) {
1442 /* ignore aliases */
1450 static void manager_build_unit_path_cache(Manager
*m
) {
1456 set_free_free(m
->unit_path_cache
);
1458 m
->unit_path_cache
= set_new(&path_hash_ops
);
1459 if (!m
->unit_path_cache
) {
1464 /* This simply builds a list of files we know exist, so that
1465 * we don't always have to go to disk */
1467 STRV_FOREACH(i
, m
->lookup_paths
.search_path
) {
1468 _cleanup_closedir_
DIR *d
= NULL
;
1473 if (errno
!= ENOENT
)
1474 log_warning_errno(errno
, "Failed to open directory %s, ignoring: %m", *i
);
1478 FOREACH_DIRENT(de
, d
, r
= -errno
; goto fail
) {
1481 p
= strjoin(streq(*i
, "/") ? "" : *i
, "/", de
->d_name
);
1487 r
= set_consume(m
->unit_path_cache
, p
);
1496 log_warning_errno(r
, "Failed to build unit path cache, proceeding without: %m");
1497 m
->unit_path_cache
= set_free_free(m
->unit_path_cache
);
1500 static void manager_distribute_fds(Manager
*m
, FDSet
*fds
) {
1506 HASHMAP_FOREACH(u
, m
->units
, i
) {
1508 if (fdset_size(fds
) <= 0)
1511 if (!UNIT_VTABLE(u
)->distribute_fds
)
1514 UNIT_VTABLE(u
)->distribute_fds(u
, fds
);
1518 static bool manager_dbus_is_running(Manager
*m
, bool deserialized
) {
1523 /* This checks whether the dbus instance we are supposed to expose our APIs on is up. We check both the socket
1524 * and the service unit. If the 'deserialized' parameter is true we'll check the deserialized state of the unit
1525 * rather than the current one. */
1527 if (MANAGER_IS_TEST_RUN(m
))
1530 u
= manager_get_unit(m
, SPECIAL_DBUS_SOCKET
);
1533 if ((deserialized
? SOCKET(u
)->deserialized_state
: SOCKET(u
)->state
) != SOCKET_RUNNING
)
1536 u
= manager_get_unit(m
, SPECIAL_DBUS_SERVICE
);
1539 if (!IN_SET((deserialized
? SERVICE(u
)->deserialized_state
: SERVICE(u
)->state
), SERVICE_RUNNING
, SERVICE_RELOAD
))
1545 static void manager_setup_bus(Manager
*m
) {
1548 /* Let's set up our private bus connection now, unconditionally */
1549 (void) bus_init_private(m
);
1551 /* If we are in --user mode also connect to the system bus now */
1552 if (MANAGER_IS_USER(m
))
1553 (void) bus_init_system(m
);
1555 /* Let's connect to the bus now, but only if the unit is supposed to be up */
1556 if (manager_dbus_is_running(m
, MANAGER_IS_RELOADING(m
))) {
1557 (void) bus_init_api(m
);
1559 if (MANAGER_IS_SYSTEM(m
))
1560 (void) bus_init_system(m
);
1564 static void manager_preset_all(Manager
*m
) {
1569 if (m
->first_boot
<= 0)
1572 if (!MANAGER_IS_SYSTEM(m
))
1575 if (MANAGER_IS_TEST_RUN(m
))
1578 /* If this is the first boot, and we are in the host system, then preset everything */
1579 r
= unit_file_preset_all(UNIT_FILE_SYSTEM
, 0, NULL
, UNIT_FILE_PRESET_ENABLE_ONLY
, NULL
, 0);
1581 log_full_errno(r
== -EEXIST
? LOG_NOTICE
: LOG_WARNING
, r
,
1582 "Failed to populate /etc with preset unit settings, ignoring: %m");
1584 log_info("Populated /etc with preset unit settings.");
1587 static void manager_vacuum(Manager
*m
) {
1590 /* Release any dynamic users no longer referenced */
1591 dynamic_user_vacuum(m
, true);
1593 /* Release any references to UIDs/GIDs no longer referenced, and destroy any IPC owned by them */
1594 manager_vacuum_uid_refs(m
);
1595 manager_vacuum_gid_refs(m
);
1597 /* Release any runtimes no longer referenced */
1598 exec_runtime_vacuum(m
);
1601 static void manager_ready(Manager
*m
) {
1604 /* After having loaded everything, do the final round of catching up with what might have changed */
1606 m
->objective
= MANAGER_OK
; /* Tell everyone we are up now */
1608 /* It might be safe to log to the journal now and connect to dbus */
1609 manager_recheck_journal(m
);
1610 manager_recheck_dbus(m
);
1612 /* Sync current state of bus names with our set of listening units */
1613 (void) manager_enqueue_sync_bus_names(m
);
1615 /* Let's finally catch up with any changes that took place while we were reloading/reexecing */
1619 static Manager
* manager_reloading_start(Manager
*m
) {
1623 static void manager_reloading_stopp(Manager
**m
) {
1625 assert((*m
)->n_reloading
> 0);
1626 (*m
)->n_reloading
--;
1630 int manager_startup(Manager
*m
, FILE *serialization
, FDSet
*fds
) {
1635 /* If we are running in test mode, we still want to run the generators,
1636 * but we should not touch the real generator directories. */
1637 r
= lookup_paths_init(&m
->lookup_paths
, m
->unit_file_scope
,
1638 MANAGER_IS_TEST_RUN(m
) ? LOOKUP_PATHS_TEMPORARY_GENERATED
: 0,
1641 return log_error_errno(r
, "Failed to initialize path lookup table: %m");
1643 dual_timestamp_get(m
->timestamps
+ manager_timestamp_initrd_mangle(MANAGER_TIMESTAMP_GENERATORS_START
));
1644 r
= manager_run_environment_generators(m
);
1646 r
= manager_run_generators(m
);
1647 dual_timestamp_get(m
->timestamps
+ manager_timestamp_initrd_mangle(MANAGER_TIMESTAMP_GENERATORS_FINISH
));
1651 manager_preset_all(m
);
1653 r
= lookup_paths_reduce(&m
->lookup_paths
);
1655 log_warning_errno(r
, "Failed ot reduce unit file paths, ignoring: %m");
1657 manager_build_unit_path_cache(m
);
1660 /* This block is (optionally) done with the reloading counter bumped */
1661 _cleanup_(manager_reloading_stopp
) Manager
*reloading
= NULL
;
1663 /* If we will deserialize make sure that during enumeration this is already known, so we increase the
1664 * counter here already */
1666 reloading
= manager_reloading_start(m
);
1668 /* First, enumerate what we can from all config files */
1669 dual_timestamp_get(m
->timestamps
+ manager_timestamp_initrd_mangle(MANAGER_TIMESTAMP_UNITS_LOAD_START
));
1670 manager_enumerate_perpetual(m
);
1671 manager_enumerate(m
);
1672 dual_timestamp_get(m
->timestamps
+ manager_timestamp_initrd_mangle(MANAGER_TIMESTAMP_UNITS_LOAD_FINISH
));
1674 /* Second, deserialize if there is something to deserialize */
1675 if (serialization
) {
1676 r
= manager_deserialize(m
, serialization
, fds
);
1678 return log_error_errno(r
, "Deserialization failed: %m");
1681 /* Any fds left? Find some unit which wants them. This is useful to allow container managers to pass
1682 * some file descriptors to us pre-initialized. This enables socket-based activation of entire
1684 manager_distribute_fds(m
, fds
);
1686 /* We might have deserialized the notify fd, but if we didn't then let's create the bus now */
1687 r
= manager_setup_notify(m
);
1689 /* No sense to continue without notifications, our children would fail anyway. */
1692 r
= manager_setup_cgroups_agent(m
);
1694 /* Likewise, no sense to continue without empty cgroup notifications. */
1697 r
= manager_setup_user_lookup_fd(m
);
1699 /* This shouldn't fail, except if things are really broken. */
1702 /* Connect to the bus if we are good for it */
1703 manager_setup_bus(m
);
1705 /* Now that we are connected to all possible busses, let's deserialize who is tracking us. */
1706 r
= bus_track_coldplug(m
, &m
->subscribed
, false, m
->deserialized_subscribed
);
1708 log_warning_errno(r
, "Failed to deserialized tracked clients, ignoring: %m");
1709 m
->deserialized_subscribed
= strv_free(m
->deserialized_subscribed
);
1711 /* Third, fire things up! */
1712 manager_coldplug(m
);
1714 /* Clean up runtime objects */
1718 /* Let's wait for the UnitNew/JobNew messages being sent, before we notify that the
1719 * reload is finished */
1720 m
->send_reloading_done
= true;
1728 int manager_add_job(Manager
*m
, JobType type
, Unit
*unit
, JobMode mode
, sd_bus_error
*e
, Job
**_ret
) {
1733 assert(type
< _JOB_TYPE_MAX
);
1735 assert(mode
< _JOB_MODE_MAX
);
1737 if (mode
== JOB_ISOLATE
&& type
!= JOB_START
)
1738 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
, "Isolate is only valid for start.");
1740 if (mode
== JOB_ISOLATE
&& !unit
->allow_isolate
)
1741 return sd_bus_error_setf(e
, BUS_ERROR_NO_ISOLATION
, "Operation refused, unit may not be isolated.");
1743 log_unit_debug(unit
, "Trying to enqueue job %s/%s/%s", unit
->id
, job_type_to_string(type
), job_mode_to_string(mode
));
1745 type
= job_type_collapse(type
, unit
);
1747 tr
= transaction_new(mode
== JOB_REPLACE_IRREVERSIBLY
);
1751 r
= transaction_add_job_and_dependencies(tr
, type
, unit
, NULL
, true, false,
1752 IN_SET(mode
, JOB_IGNORE_DEPENDENCIES
, JOB_IGNORE_REQUIREMENTS
),
1753 mode
== JOB_IGNORE_DEPENDENCIES
, e
);
1757 if (mode
== JOB_ISOLATE
) {
1758 r
= transaction_add_isolate_jobs(tr
, m
);
1763 r
= transaction_activate(tr
, m
, mode
, e
);
1767 log_unit_debug(unit
,
1768 "Enqueued job %s/%s as %u", unit
->id
,
1769 job_type_to_string(type
), (unsigned) tr
->anchor_job
->id
);
1772 *_ret
= tr
->anchor_job
;
1774 transaction_free(tr
);
1778 transaction_abort(tr
);
1779 transaction_free(tr
);
1783 int manager_add_job_by_name(Manager
*m
, JobType type
, const char *name
, JobMode mode
, sd_bus_error
*e
, Job
**ret
) {
1784 Unit
*unit
= NULL
; /* just to appease gcc, initialization is not really necessary */
1788 assert(type
< _JOB_TYPE_MAX
);
1790 assert(mode
< _JOB_MODE_MAX
);
1792 r
= manager_load_unit(m
, name
, NULL
, NULL
, &unit
);
1797 return manager_add_job(m
, type
, unit
, mode
, e
, ret
);
1800 int manager_add_job_by_name_and_warn(Manager
*m
, JobType type
, const char *name
, JobMode mode
, Job
**ret
) {
1801 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
1805 assert(type
< _JOB_TYPE_MAX
);
1807 assert(mode
< _JOB_MODE_MAX
);
1809 r
= manager_add_job_by_name(m
, type
, name
, mode
, &error
, ret
);
1811 return log_warning_errno(r
, "Failed to enqueue %s job for %s: %s", job_mode_to_string(mode
), name
, bus_error_message(&error
, r
));
1816 int manager_propagate_reload(Manager
*m
, Unit
*unit
, JobMode mode
, sd_bus_error
*e
) {
1822 assert(mode
< _JOB_MODE_MAX
);
1823 assert(mode
!= JOB_ISOLATE
); /* Isolate is only valid for start */
1825 tr
= transaction_new(mode
== JOB_REPLACE_IRREVERSIBLY
);
1829 /* We need an anchor job */
1830 r
= transaction_add_job_and_dependencies(tr
, JOB_NOP
, unit
, NULL
, false, false, true, true, e
);
1834 /* Failure in adding individual dependencies is ignored, so this always succeeds. */
1835 transaction_add_propagate_reload_jobs(tr
, unit
, tr
->anchor_job
, mode
== JOB_IGNORE_DEPENDENCIES
, e
);
1837 r
= transaction_activate(tr
, m
, mode
, e
);
1841 transaction_free(tr
);
1845 transaction_abort(tr
);
1846 transaction_free(tr
);
1850 Job
*manager_get_job(Manager
*m
, uint32_t id
) {
1853 return hashmap_get(m
->jobs
, UINT32_TO_PTR(id
));
1856 Unit
*manager_get_unit(Manager
*m
, const char *name
) {
1860 return hashmap_get(m
->units
, name
);
1863 static int manager_dispatch_target_deps_queue(Manager
*m
) {
1868 static const UnitDependency deps
[] = {
1877 while ((u
= m
->target_deps_queue
)) {
1878 assert(u
->in_target_deps_queue
);
1880 LIST_REMOVE(target_deps_queue
, u
->manager
->target_deps_queue
, u
);
1881 u
->in_target_deps_queue
= false;
1883 for (k
= 0; k
< ELEMENTSOF(deps
); k
++) {
1888 HASHMAP_FOREACH_KEY(v
, target
, u
->dependencies
[deps
[k
]], i
) {
1889 r
= unit_add_default_target_dependency(u
, target
);
1899 unsigned manager_dispatch_load_queue(Manager
*m
) {
1905 /* Make sure we are not run recursively */
1906 if (m
->dispatching_load_queue
)
1909 m
->dispatching_load_queue
= true;
1911 /* Dispatches the load queue. Takes a unit from the queue and
1912 * tries to load its data until the queue is empty */
1914 while ((u
= m
->load_queue
)) {
1915 assert(u
->in_load_queue
);
1921 m
->dispatching_load_queue
= false;
1923 /* Dispatch the units waiting for their target dependencies to be added now, as all targets that we know about
1924 * should be loaded and have aliases resolved */
1925 (void) manager_dispatch_target_deps_queue(m
);
1930 int manager_load_unit_prepare(
1937 _cleanup_(unit_freep
) Unit
*cleanup_ret
= NULL
;
1943 assert(name
|| path
);
1946 /* This will prepare the unit for loading, but not actually
1947 * load anything from disk. */
1949 if (path
&& !is_path(path
))
1950 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
, "Path %s is not absolute.", path
);
1953 name
= basename(path
);
1955 t
= unit_name_to_type(name
);
1957 if (t
== _UNIT_TYPE_INVALID
|| !unit_name_is_valid(name
, UNIT_NAME_PLAIN
|UNIT_NAME_INSTANCE
)) {
1958 if (unit_name_is_valid(name
, UNIT_NAME_TEMPLATE
))
1959 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
, "Unit name %s is missing the instance name.", name
);
1961 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
, "Unit name %s is not valid.", name
);
1964 ret
= manager_get_unit(m
, name
);
1970 ret
= cleanup_ret
= unit_new(m
, unit_vtable
[t
]->object_size
);
1975 ret
->fragment_path
= strdup(path
);
1976 if (!ret
->fragment_path
)
1980 r
= unit_add_name(ret
, name
);
1984 unit_add_to_load_queue(ret
);
1985 unit_add_to_dbus_queue(ret
);
1986 unit_add_to_gc_queue(ret
);
1994 int manager_load_unit(
2006 /* This will load the service information files, but not actually
2007 * start any services or anything. */
2009 r
= manager_load_unit_prepare(m
, name
, path
, e
, _ret
);
2013 manager_dispatch_load_queue(m
);
2015 *_ret
= unit_follow_merge(*_ret
);
2019 int manager_load_startable_unit_or_warn(
2025 /* Load a unit, make sure it loaded fully and is not masked. */
2027 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
2031 r
= manager_load_unit(m
, name
, path
, &error
, &unit
);
2033 return log_error_errno(r
, "Failed to load %s %s: %s",
2034 name
? "unit" : "unit file", name
?: path
,
2035 bus_error_message(&error
, r
));
2037 r
= bus_unit_validate_load_state(unit
, &error
);
2039 return log_error_errno(r
, "%s", bus_error_message(&error
, r
));
2045 void manager_dump_jobs(Manager
*s
, FILE *f
, const char *prefix
) {
2052 HASHMAP_FOREACH(j
, s
->jobs
, i
)
2053 job_dump(j
, f
, prefix
);
2056 void manager_dump_units(Manager
*s
, FILE *f
, const char *prefix
) {
2064 HASHMAP_FOREACH_KEY(u
, t
, s
->units
, i
)
2066 unit_dump(u
, f
, prefix
);
2069 void manager_dump(Manager
*m
, FILE *f
, const char *prefix
) {
2075 for (q
= 0; q
< _MANAGER_TIMESTAMP_MAX
; q
++) {
2076 char buf
[FORMAT_TIMESTAMP_MAX
];
2078 if (dual_timestamp_is_set(m
->timestamps
+ q
))
2079 fprintf(f
, "%sTimestamp %s: %s\n",
2081 manager_timestamp_to_string(q
),
2082 format_timestamp(buf
, sizeof(buf
), m
->timestamps
[q
].realtime
));
2085 manager_dump_units(m
, f
, prefix
);
2086 manager_dump_jobs(m
, f
, prefix
);
2089 int manager_get_dump_string(Manager
*m
, char **ret
) {
2090 _cleanup_free_
char *dump
= NULL
;
2091 _cleanup_fclose_
FILE *f
= NULL
;
2098 f
= open_memstream(&dump
, &size
);
2102 (void) __fsetlocking(f
, FSETLOCKING_BYCALLER
);
2104 manager_dump(m
, f
, NULL
);
2106 r
= fflush_and_check(f
);
2112 *ret
= TAKE_PTR(dump
);
2117 void manager_clear_jobs(Manager
*m
) {
2122 while ((j
= hashmap_first(m
->jobs
)))
2123 /* No need to recurse. We're cancelling all jobs. */
2124 job_finish_and_invalidate(j
, JOB_CANCELED
, false, false);
2127 static int manager_dispatch_run_queue(sd_event_source
*source
, void *userdata
) {
2128 Manager
*m
= userdata
;
2134 while ((j
= m
->run_queue
)) {
2135 assert(j
->installed
);
2136 assert(j
->in_run_queue
);
2138 job_run_and_invalidate(j
);
2141 if (m
->n_running_jobs
> 0)
2142 manager_watch_jobs_in_progress(m
);
2144 if (m
->n_on_console
> 0)
2145 manager_watch_idle_pipe(m
);
2150 static unsigned manager_dispatch_dbus_queue(Manager
*m
) {
2151 unsigned n
= 0, budget
;
2157 if (m
->dispatching_dbus_queue
)
2160 /* Anything to do at all? */
2161 if (!m
->dbus_unit_queue
&& !m
->dbus_job_queue
&& !m
->send_reloading_done
&& !m
->queued_message
)
2164 /* Do we have overly many messages queued at the moment? If so, let's not enqueue more on top, let's sit this
2165 * cycle out, and process things in a later cycle when the queues got a bit emptier. */
2166 if (manager_bus_n_queued_write(m
) > MANAGER_BUS_BUSY_THRESHOLD
)
2169 /* Only process a certain number of units/jobs per event loop iteration. Even if the bus queue wasn't overly
2170 * full before this call we shouldn't increase it in size too wildly in one step, and we shouldn't monopolize
2171 * CPU time with generating these messages. Note the difference in counting of this "budget" and the
2172 * "threshold" above: the "budget" is decreased only once per generated message, regardless how many
2173 * busses/direct connections it is enqueued on, while the "threshold" is applied to each queued instance of bus
2174 * message, i.e. if the same message is enqueued to five busses/direct connections it will be counted five
2175 * times. This difference in counting ("references" vs. "instances") is primarily a result of the fact that
2176 * it's easier to implement it this way, however it also reflects the thinking that the "threshold" should put
2177 * a limit on used queue memory, i.e. space, while the "budget" should put a limit on time. Also note that
2178 * the "threshold" is currently chosen much higher than the "budget". */
2179 budget
= MANAGER_BUS_MESSAGE_BUDGET
;
2181 m
->dispatching_dbus_queue
= true;
2183 while (budget
> 0 && (u
= m
->dbus_unit_queue
)) {
2185 assert(u
->in_dbus_queue
);
2187 bus_unit_send_change_signal(u
);
2191 while (budget
> 0 && (j
= m
->dbus_job_queue
)) {
2192 assert(j
->in_dbus_queue
);
2194 bus_job_send_change_signal(j
);
2198 m
->dispatching_dbus_queue
= false;
2200 if (budget
> 0 && m
->send_reloading_done
) {
2201 m
->send_reloading_done
= false;
2202 bus_manager_send_reloading(m
, false);
2206 if (budget
> 0 && m
->queued_message
) {
2207 bus_send_queued_message(m
);
2214 static int manager_dispatch_cgroups_agent_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
2215 Manager
*m
= userdata
;
2216 char buf
[PATH_MAX
+1];
2219 n
= recv(fd
, buf
, sizeof(buf
), 0);
2221 return log_error_errno(errno
, "Failed to read cgroups agent message: %m");
2223 log_error("Got zero-length cgroups agent message, ignoring.");
2226 if ((size_t) n
>= sizeof(buf
)) {
2227 log_error("Got overly long cgroups agent message, ignoring.");
2231 if (memchr(buf
, 0, n
)) {
2232 log_error("Got cgroups agent message with embedded NUL byte, ignoring.");
2237 manager_notify_cgroup_empty(m
, buf
);
2238 (void) bus_forward_agent_released(m
, buf
);
2243 static void manager_invoke_notify_message(
2246 const struct ucred
*ucred
,
2255 if (u
->notifygen
== m
->notifygen
) /* Already invoked on this same unit in this same iteration? */
2257 u
->notifygen
= m
->notifygen
;
2259 if (UNIT_VTABLE(u
)->notify_message
) {
2260 _cleanup_strv_free_
char **tags
= NULL
;
2262 tags
= strv_split(buf
, NEWLINE
);
2268 UNIT_VTABLE(u
)->notify_message(u
, ucred
, tags
, fds
);
2270 } else if (DEBUG_LOGGING
) {
2271 _cleanup_free_
char *x
= NULL
, *y
= NULL
;
2273 x
= ellipsize(buf
, 20, 90);
2277 log_unit_debug(u
, "Got notification message \"%s\", ignoring.", strnull(y
));
2281 static int manager_dispatch_notify_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
2283 _cleanup_fdset_free_ FDSet
*fds
= NULL
;
2284 Manager
*m
= userdata
;
2285 char buf
[NOTIFY_BUFFER_MAX
+1];
2286 struct iovec iovec
= {
2288 .iov_len
= sizeof(buf
)-1,
2291 struct cmsghdr cmsghdr
;
2292 uint8_t buf
[CMSG_SPACE(sizeof(struct ucred
)) +
2293 CMSG_SPACE(sizeof(int) * NOTIFY_FD_MAX
)];
2295 struct msghdr msghdr
= {
2298 .msg_control
= &control
,
2299 .msg_controllen
= sizeof(control
),
2302 struct cmsghdr
*cmsg
;
2303 struct ucred
*ucred
= NULL
;
2304 _cleanup_free_ Unit
**array_copy
= NULL
;
2305 Unit
*u1
, *u2
, **array
;
2306 int r
, *fd_array
= NULL
;
2312 assert(m
->notify_fd
== fd
);
2314 if (revents
!= EPOLLIN
) {
2315 log_warning("Got unexpected poll event for notify fd.");
2319 n
= recvmsg(m
->notify_fd
, &msghdr
, MSG_DONTWAIT
|MSG_CMSG_CLOEXEC
|MSG_TRUNC
);
2321 if (IN_SET(errno
, EAGAIN
, EINTR
))
2322 return 0; /* Spurious wakeup, try again */
2324 /* If this is any other, real error, then let's stop processing this socket. This of course means we
2325 * won't take notification messages anymore, but that's still better than busy looping around this:
2326 * being woken up over and over again but being unable to actually read the message off the socket. */
2327 return log_error_errno(errno
, "Failed to receive notification message: %m");
2330 CMSG_FOREACH(cmsg
, &msghdr
) {
2331 if (cmsg
->cmsg_level
== SOL_SOCKET
&& cmsg
->cmsg_type
== SCM_RIGHTS
) {
2333 fd_array
= (int*) CMSG_DATA(cmsg
);
2334 n_fds
= (cmsg
->cmsg_len
- CMSG_LEN(0)) / sizeof(int);
2336 } else if (cmsg
->cmsg_level
== SOL_SOCKET
&&
2337 cmsg
->cmsg_type
== SCM_CREDENTIALS
&&
2338 cmsg
->cmsg_len
== CMSG_LEN(sizeof(struct ucred
))) {
2340 ucred
= (struct ucred
*) CMSG_DATA(cmsg
);
2347 r
= fdset_new_array(&fds
, fd_array
, n_fds
);
2349 close_many(fd_array
, n_fds
);
2355 if (!ucred
|| !pid_is_valid(ucred
->pid
)) {
2356 log_warning("Received notify message without valid credentials. Ignoring.");
2360 if ((size_t) n
>= sizeof(buf
) || (msghdr
.msg_flags
& MSG_TRUNC
)) {
2361 log_warning("Received notify message exceeded maximum size. Ignoring.");
2365 /* As extra safety check, let's make sure the string we get doesn't contain embedded NUL bytes. We permit one
2366 * trailing NUL byte in the message, but don't expect it. */
2367 if (n
> 1 && memchr(buf
, 0, n
-1)) {
2368 log_warning("Received notify message with embedded NUL bytes. Ignoring.");
2372 /* Make sure it's NUL-terminated. */
2375 /* Increase the generation counter used for filtering out duplicate unit invocations. */
2378 /* Notify every unit that might be interested, which might be multiple. */
2379 u1
= manager_get_unit_by_pid_cgroup(m
, ucred
->pid
);
2380 u2
= hashmap_get(m
->watch_pids
, PID_TO_PTR(ucred
->pid
));
2381 array
= hashmap_get(m
->watch_pids
, PID_TO_PTR(-ucred
->pid
));
2388 array_copy
= newdup(Unit
*, array
, k
+1);
2392 /* And now invoke the per-unit callbacks. Note that manager_invoke_notify_message() will handle duplicate units
2393 * make sure we only invoke each unit's handler once. */
2395 manager_invoke_notify_message(m
, u1
, ucred
, buf
, fds
);
2399 manager_invoke_notify_message(m
, u2
, ucred
, buf
, fds
);
2403 for (size_t i
= 0; array_copy
[i
]; i
++) {
2404 manager_invoke_notify_message(m
, array_copy
[i
], ucred
, buf
, fds
);
2409 log_warning("Cannot find unit for notify message of PID "PID_FMT
", ignoring.", ucred
->pid
);
2411 if (fdset_size(fds
) > 0)
2412 log_warning("Got extra auxiliary fds with notification message, closing them.");
2417 static void manager_invoke_sigchld_event(
2420 const siginfo_t
*si
) {
2426 /* Already invoked the handler of this unit in this iteration? Then don't process this again */
2427 if (u
->sigchldgen
== m
->sigchldgen
)
2429 u
->sigchldgen
= m
->sigchldgen
;
2431 log_unit_debug(u
, "Child "PID_FMT
" belongs to %s.", si
->si_pid
, u
->id
);
2432 unit_unwatch_pid(u
, si
->si_pid
);
2434 if (UNIT_VTABLE(u
)->sigchld_event
)
2435 UNIT_VTABLE(u
)->sigchld_event(u
, si
->si_pid
, si
->si_code
, si
->si_status
);
2438 static int manager_dispatch_sigchld(sd_event_source
*source
, void *userdata
) {
2439 Manager
*m
= userdata
;
2446 /* First we call waitd() for a PID and do not reap the zombie. That way we can still access /proc/$PID for it
2447 * while it is a zombie. */
2449 if (waitid(P_ALL
, 0, &si
, WEXITED
|WNOHANG
|WNOWAIT
) < 0) {
2451 if (errno
!= ECHILD
)
2452 log_error_errno(errno
, "Failed to peek for child with waitid(), ignoring: %m");
2460 if (IN_SET(si
.si_code
, CLD_EXITED
, CLD_KILLED
, CLD_DUMPED
)) {
2461 _cleanup_free_ Unit
**array_copy
= NULL
;
2462 _cleanup_free_
char *name
= NULL
;
2463 Unit
*u1
, *u2
, **array
;
2465 (void) get_process_comm(si
.si_pid
, &name
);
2467 log_debug("Child "PID_FMT
" (%s) died (code=%s, status=%i/%s)",
2468 si
.si_pid
, strna(name
),
2469 sigchld_code_to_string(si
.si_code
),
2471 strna(si
.si_code
== CLD_EXITED
2472 ? exit_status_to_string(si
.si_status
, EXIT_STATUS_FULL
)
2473 : signal_to_string(si
.si_status
)));
2475 /* Increase the generation counter used for filtering out duplicate unit invocations */
2478 /* And now figure out the unit this belongs to, it might be multiple... */
2479 u1
= manager_get_unit_by_pid_cgroup(m
, si
.si_pid
);
2480 u2
= hashmap_get(m
->watch_pids
, PID_TO_PTR(si
.si_pid
));
2481 array
= hashmap_get(m
->watch_pids
, PID_TO_PTR(-si
.si_pid
));
2485 /* Cound how many entries the array has */
2489 /* Make a copy of the array so that we don't trip up on the array changing beneath us */
2490 array_copy
= newdup(Unit
*, array
, n
+1);
2495 /* Finally, execute them all. Note that u1, u2 and the array might contain duplicates, but
2496 * that's fine, manager_invoke_sigchld_event() will ensure we only invoke the handlers once for
2497 * each iteration. */
2499 manager_invoke_sigchld_event(m
, u1
, &si
);
2501 manager_invoke_sigchld_event(m
, u2
, &si
);
2503 for (size_t i
= 0; array_copy
[i
]; i
++)
2504 manager_invoke_sigchld_event(m
, array_copy
[i
], &si
);
2507 /* And now, we actually reap the zombie. */
2508 if (waitid(P_PID
, si
.si_pid
, &si
, WEXITED
) < 0) {
2509 log_error_errno(errno
, "Failed to dequeue child, ignoring: %m");
2516 /* All children processed for now, turn off event source */
2518 r
= sd_event_source_set_enabled(m
->sigchld_event_source
, SD_EVENT_OFF
);
2520 return log_error_errno(r
, "Failed to disable SIGCHLD event source: %m");
2525 static void manager_start_target(Manager
*m
, const char *name
, JobMode mode
) {
2526 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
2529 log_debug("Activating special unit %s", name
);
2531 r
= manager_add_job_by_name(m
, JOB_START
, name
, mode
, &error
, NULL
);
2533 log_error("Failed to enqueue %s job: %s", name
, bus_error_message(&error
, r
));
2536 static void manager_handle_ctrl_alt_del(Manager
*m
) {
2537 /* If the user presses C-A-D more than
2538 * 7 times within 2s, we reboot/shutdown immediately,
2539 * unless it was disabled in system.conf */
2541 if (ratelimit_below(&m
->ctrl_alt_del_ratelimit
) || m
->cad_burst_action
== EMERGENCY_ACTION_NONE
)
2542 manager_start_target(m
, SPECIAL_CTRL_ALT_DEL_TARGET
, JOB_REPLACE_IRREVERSIBLY
);
2544 emergency_action(m
, m
->cad_burst_action
, NULL
,
2545 "Ctrl-Alt-Del was pressed more than 7 times within 2s");
2548 static int manager_dispatch_signal_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
2549 Manager
*m
= userdata
;
2551 struct signalfd_siginfo sfsi
;
2555 assert(m
->signal_fd
== fd
);
2557 if (revents
!= EPOLLIN
) {
2558 log_warning("Got unexpected events from signal file descriptor.");
2562 n
= read(m
->signal_fd
, &sfsi
, sizeof(sfsi
));
2563 if (n
!= sizeof(sfsi
)) {
2565 log_warning("Truncated read from signal fd (%zu bytes), ignoring!", n
);
2569 if (IN_SET(errno
, EINTR
, EAGAIN
))
2572 /* We return an error here, which will kill this handler,
2573 * to avoid a busy loop on read error. */
2574 return log_error_errno(errno
, "Reading from signal fd failed: %m");
2577 log_received_signal(sfsi
.ssi_signo
== SIGCHLD
||
2578 (sfsi
.ssi_signo
== SIGTERM
&& MANAGER_IS_USER(m
))
2579 ? LOG_DEBUG
: LOG_INFO
,
2582 switch (sfsi
.ssi_signo
) {
2585 r
= sd_event_source_set_enabled(m
->sigchld_event_source
, SD_EVENT_ON
);
2587 log_warning_errno(r
, "Failed to enable SIGCHLD event source, ignoring: %m");
2592 if (MANAGER_IS_SYSTEM(m
)) {
2593 /* This is for compatibility with the original sysvinit */
2594 if (verify_run_space_and_log("Refusing to reexecute") < 0)
2597 m
->objective
= MANAGER_REEXECUTE
;
2603 if (MANAGER_IS_SYSTEM(m
))
2604 manager_handle_ctrl_alt_del(m
);
2606 manager_start_target(m
, SPECIAL_EXIT_TARGET
,
2607 JOB_REPLACE_IRREVERSIBLY
);
2611 /* This is a nop on non-init */
2612 if (MANAGER_IS_SYSTEM(m
))
2613 manager_start_target(m
, SPECIAL_KBREQUEST_TARGET
, JOB_REPLACE
);
2618 /* This is a nop on non-init */
2619 if (MANAGER_IS_SYSTEM(m
))
2620 manager_start_target(m
, SPECIAL_SIGPWR_TARGET
, JOB_REPLACE
);
2625 if (manager_dbus_is_running(m
, false)) {
2626 log_info("Trying to reconnect to bus...");
2628 (void) bus_init_api(m
);
2630 if (MANAGER_IS_SYSTEM(m
))
2631 (void) bus_init_system(m
);
2633 log_info("Starting D-Bus service...");
2634 manager_start_target(m
, SPECIAL_DBUS_SERVICE
, JOB_REPLACE
);
2640 _cleanup_free_
char *dump
= NULL
;
2642 r
= manager_get_dump_string(m
, &dump
);
2644 log_warning_errno(errno
, "Failed to acquire manager dump: %m");
2648 log_dump(LOG_INFO
, dump
);
2653 if (verify_run_space_and_log("Refusing to reload") < 0)
2656 m
->objective
= MANAGER_RELOAD
;
2661 /* Starting SIGRTMIN+0 */
2662 static const struct {
2665 } target_table
[] = {
2666 [0] = { SPECIAL_DEFAULT_TARGET
, JOB_ISOLATE
},
2667 [1] = { SPECIAL_RESCUE_TARGET
, JOB_ISOLATE
},
2668 [2] = { SPECIAL_EMERGENCY_TARGET
, JOB_ISOLATE
},
2669 [3] = { SPECIAL_HALT_TARGET
, JOB_REPLACE_IRREVERSIBLY
},
2670 [4] = { SPECIAL_POWEROFF_TARGET
, JOB_REPLACE_IRREVERSIBLY
},
2671 [5] = { SPECIAL_REBOOT_TARGET
, JOB_REPLACE_IRREVERSIBLY
},
2672 [6] = { SPECIAL_KEXEC_TARGET
, JOB_REPLACE_IRREVERSIBLY
},
2675 /* Starting SIGRTMIN+13, so that target halt and system halt are 10 apart */
2676 static const ManagerObjective objective_table
[] = {
2678 [1] = MANAGER_POWEROFF
,
2679 [2] = MANAGER_REBOOT
,
2680 [3] = MANAGER_KEXEC
,
2683 if ((int) sfsi
.ssi_signo
>= SIGRTMIN
+0 &&
2684 (int) sfsi
.ssi_signo
< SIGRTMIN
+(int) ELEMENTSOF(target_table
)) {
2685 int idx
= (int) sfsi
.ssi_signo
- SIGRTMIN
;
2686 manager_start_target(m
, target_table
[idx
].target
,
2687 target_table
[idx
].mode
);
2691 if ((int) sfsi
.ssi_signo
>= SIGRTMIN
+13 &&
2692 (int) sfsi
.ssi_signo
< SIGRTMIN
+13+(int) ELEMENTSOF(objective_table
)) {
2693 m
->objective
= objective_table
[sfsi
.ssi_signo
- SIGRTMIN
- 13];
2697 switch (sfsi
.ssi_signo
- SIGRTMIN
) {
2700 manager_set_show_status(m
, SHOW_STATUS_YES
);
2704 manager_set_show_status(m
, SHOW_STATUS_NO
);
2708 manager_override_log_level(m
, LOG_DEBUG
);
2712 manager_restore_original_log_level(m
);
2716 if (MANAGER_IS_USER(m
)) {
2717 m
->objective
= MANAGER_EXIT
;
2721 /* This is a nop on init */
2725 case 29: /* compatibility: used to be mapped to LOG_TARGET_SYSLOG_OR_KMSG */
2726 manager_restore_original_log_target(m
);
2730 manager_override_log_target(m
, LOG_TARGET_CONSOLE
);
2734 manager_override_log_target(m
, LOG_TARGET_KMSG
);
2738 log_warning("Got unhandled signal <%s>.", signal_to_string(sfsi
.ssi_signo
));
2745 static int manager_dispatch_time_change_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
2746 Manager
*m
= userdata
;
2751 assert(m
->time_change_fd
== fd
);
2753 log_struct(LOG_DEBUG
,
2754 "MESSAGE_ID=" SD_MESSAGE_TIME_CHANGE_STR
,
2755 LOG_MESSAGE("Time has been changed"));
2757 /* Restart the watch */
2758 (void) manager_setup_time_change(m
);
2760 HASHMAP_FOREACH(u
, m
->units
, i
)
2761 if (UNIT_VTABLE(u
)->time_change
)
2762 UNIT_VTABLE(u
)->time_change(u
);
2767 static int manager_dispatch_timezone_change(
2768 sd_event_source
*source
,
2769 const struct inotify_event
*e
,
2772 Manager
*m
= userdata
;
2779 log_debug("inotify event for /etc/localtime");
2781 changed
= manager_read_timezone_stat(m
);
2787 /* Something changed, restart the watch, to ensure we watch the new /etc/localtime if it changed */
2788 (void) manager_setup_timezone_change(m
);
2790 /* Read the new timezone */
2793 log_debug("Timezone has been changed (now: %s).", tzname
[daylight
]);
2795 HASHMAP_FOREACH(u
, m
->units
, i
)
2796 if (UNIT_VTABLE(u
)->timezone_change
)
2797 UNIT_VTABLE(u
)->timezone_change(u
);
2802 static int manager_dispatch_idle_pipe_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
2803 Manager
*m
= userdata
;
2806 assert(m
->idle_pipe
[2] == fd
);
2808 /* There's at least one Type=idle child that just gave up on us waiting for the boot process to complete. Let's
2809 * now turn off any further console output if there's at least one service that needs console access, so that
2810 * from now on our own output should not spill into that service's output anymore. After all, we support
2811 * Type=idle only to beautify console output and it generally is set on services that want to own the console
2812 * exclusively without our interference. */
2813 m
->no_console_output
= m
->n_on_console
> 0;
2815 /* Acknowledge the child's request, and let all all other children know too that they shouldn't wait any longer
2816 * by closing the pipes towards them, which is what they are waiting for. */
2817 manager_close_idle_pipe(m
);
2822 static int manager_dispatch_jobs_in_progress(sd_event_source
*source
, usec_t usec
, void *userdata
) {
2823 Manager
*m
= userdata
;
2830 manager_print_jobs_in_progress(m
);
2832 next
= now(CLOCK_MONOTONIC
) + JOBS_IN_PROGRESS_PERIOD_USEC
;
2833 r
= sd_event_source_set_time(source
, next
);
2837 return sd_event_source_set_enabled(source
, SD_EVENT_ONESHOT
);
2840 int manager_loop(Manager
*m
) {
2843 RATELIMIT_DEFINE(rl
, 1*USEC_PER_SEC
, 50000);
2846 assert(m
->objective
== MANAGER_OK
); /* Ensure manager_startup() has been called */
2848 /* Release the path cache */
2849 m
->unit_path_cache
= set_free_free(m
->unit_path_cache
);
2851 manager_check_finished(m
);
2853 /* There might still be some zombies hanging around from before we were exec()'ed. Let's reap them. */
2854 r
= sd_event_source_set_enabled(m
->sigchld_event_source
, SD_EVENT_ON
);
2856 return log_error_errno(r
, "Failed to enable SIGCHLD event source: %m");
2858 while (m
->objective
== MANAGER_OK
) {
2861 if (m
->runtime_watchdog
> 0 && m
->runtime_watchdog
!= USEC_INFINITY
&& MANAGER_IS_SYSTEM(m
))
2864 if (!ratelimit_below(&rl
)) {
2865 /* Yay, something is going seriously wrong, pause a little */
2866 log_warning("Looping too fast. Throttling execution a little.");
2870 if (manager_dispatch_load_queue(m
) > 0)
2873 if (manager_dispatch_gc_job_queue(m
) > 0)
2876 if (manager_dispatch_gc_unit_queue(m
) > 0)
2879 if (manager_dispatch_cleanup_queue(m
) > 0)
2882 if (manager_dispatch_cgroup_realize_queue(m
) > 0)
2885 if (manager_dispatch_stop_when_unneeded_queue(m
) > 0)
2888 if (manager_dispatch_dbus_queue(m
) > 0)
2891 /* Sleep for half the watchdog time */
2892 if (m
->runtime_watchdog
> 0 && m
->runtime_watchdog
!= USEC_INFINITY
&& MANAGER_IS_SYSTEM(m
)) {
2893 wait_usec
= m
->runtime_watchdog
/ 2;
2897 wait_usec
= USEC_INFINITY
;
2899 r
= sd_event_run(m
->event
, wait_usec
);
2901 return log_error_errno(r
, "Failed to run event loop: %m");
2904 return m
->objective
;
2907 int manager_load_unit_from_dbus_path(Manager
*m
, const char *s
, sd_bus_error
*e
, Unit
**_u
) {
2908 _cleanup_free_
char *n
= NULL
;
2909 sd_id128_t invocation_id
;
2917 r
= unit_name_from_dbus_path(s
, &n
);
2921 /* Permit addressing units by invocation ID: if the passed bus path is suffixed by a 128bit ID then we use it
2922 * as invocation ID. */
2923 r
= sd_id128_from_string(n
, &invocation_id
);
2925 u
= hashmap_get(m
->units_by_invocation_id
, &invocation_id
);
2931 return sd_bus_error_setf(e
, BUS_ERROR_NO_UNIT_FOR_INVOCATION_ID
,
2932 "No unit with the specified invocation ID " SD_ID128_FORMAT_STR
" known.",
2933 SD_ID128_FORMAT_VAL(invocation_id
));
2936 /* If this didn't work, we check if this is a unit name */
2937 if (!unit_name_is_valid(n
, UNIT_NAME_PLAIN
|UNIT_NAME_INSTANCE
)) {
2938 _cleanup_free_
char *nn
= NULL
;
2941 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
,
2942 "Unit name %s is neither a valid invocation ID nor unit name.", strnull(nn
));
2945 r
= manager_load_unit(m
, n
, NULL
, e
, &u
);
2953 int manager_get_job_from_dbus_path(Manager
*m
, const char *s
, Job
**_j
) {
2963 p
= startswith(s
, "/org/freedesktop/systemd1/job/");
2967 r
= safe_atou(p
, &id
);
2971 j
= manager_get_job(m
, id
);
2980 void manager_send_unit_audit(Manager
*m
, Unit
*u
, int type
, bool success
) {
2983 _cleanup_free_
char *p
= NULL
;
2987 if (!MANAGER_IS_SYSTEM(m
))
2990 audit_fd
= get_audit_fd();
2994 /* Don't generate audit events if the service was already
2995 * started and we're just deserializing */
2996 if (MANAGER_IS_RELOADING(m
))
2999 if (u
->type
!= UNIT_SERVICE
)
3002 r
= unit_name_to_prefix_and_instance(u
->id
, &p
);
3004 log_error_errno(r
, "Failed to extract prefix and instance of unit name: %m");
3008 msg
= strjoina("unit=", p
);
3009 if (audit_log_user_comm_message(audit_fd
, type
, msg
, "systemd", NULL
, NULL
, NULL
, success
) < 0) {
3011 /* We aren't allowed to send audit messages?
3012 * Then let's not retry again. */
3015 log_warning_errno(errno
, "Failed to send audit message: %m");
3021 void manager_send_unit_plymouth(Manager
*m
, Unit
*u
) {
3022 static const union sockaddr_union sa
= PLYMOUTH_SOCKET
;
3023 _cleanup_free_
char *message
= NULL
;
3024 _cleanup_close_
int fd
= -1;
3027 /* Don't generate plymouth events if the service was already
3028 * started and we're just deserializing */
3029 if (MANAGER_IS_RELOADING(m
))
3032 if (!MANAGER_IS_SYSTEM(m
))
3035 if (detect_container() > 0)
3038 if (!IN_SET(u
->type
, UNIT_SERVICE
, UNIT_MOUNT
, UNIT_SWAP
))
3041 /* We set SOCK_NONBLOCK here so that we rather drop the
3042 * message then wait for plymouth */
3043 fd
= socket(AF_UNIX
, SOCK_STREAM
|SOCK_CLOEXEC
|SOCK_NONBLOCK
, 0);
3045 log_error_errno(errno
, "socket() failed: %m");
3049 if (connect(fd
, &sa
.sa
, SOCKADDR_UN_LEN(sa
.un
)) < 0) {
3050 if (!IN_SET(errno
, EPIPE
, EAGAIN
, ENOENT
, ECONNREFUSED
, ECONNRESET
, ECONNABORTED
))
3051 log_error_errno(errno
, "connect() failed: %m");
3055 if (asprintf(&message
, "U\002%c%s%n", (int) (strlen(u
->id
) + 1), u
->id
, &n
) < 0) {
3061 if (write(fd
, message
, n
+ 1) != n
+ 1)
3062 if (!IN_SET(errno
, EPIPE
, EAGAIN
, ENOENT
, ECONNREFUSED
, ECONNRESET
, ECONNABORTED
))
3063 log_error_errno(errno
, "Failed to write Plymouth message: %m");
3066 int manager_open_serialization(Manager
*m
, FILE **_f
) {
3072 fd
= open_serialization_fd("systemd-state");
3076 f
= fdopen(fd
, "w+");
3086 int manager_serialize(
3090 bool switching_root
) {
3102 _cleanup_(manager_reloading_stopp
) _unused_ Manager
*reloading
= manager_reloading_start(m
);
3104 fprintf(f
, "current-job-id=%"PRIu32
"\n", m
->current_job_id
);
3105 fprintf(f
, "n-installed-jobs=%u\n", m
->n_installed_jobs
);
3106 fprintf(f
, "n-failed-jobs=%u\n", m
->n_failed_jobs
);
3107 fprintf(f
, "taint-usr=%s\n", yes_no(m
->taint_usr
));
3108 fprintf(f
, "ready-sent=%s\n", yes_no(m
->ready_sent
));
3109 fprintf(f
, "taint-logged=%s\n", yes_no(m
->taint_logged
));
3110 fprintf(f
, "service-watchdogs=%s\n", yes_no(m
->service_watchdogs
));
3112 t
= show_status_to_string(m
->show_status
);
3114 fprintf(f
, "show-status=%s\n", t
);
3116 if (m
->log_level_overridden
)
3117 fprintf(f
, "log-level-override=%i\n", log_get_max_level());
3118 if (m
->log_target_overridden
)
3119 fprintf(f
, "log-target-override=%s\n", log_target_to_string(log_get_target()));
3121 for (q
= 0; q
< _MANAGER_TIMESTAMP_MAX
; q
++) {
3122 /* The following timestamps only apply to the host system, hence only serialize them there */
3124 IN_SET(q
, MANAGER_TIMESTAMP_USERSPACE
, MANAGER_TIMESTAMP_FINISH
,
3125 MANAGER_TIMESTAMP_SECURITY_START
, MANAGER_TIMESTAMP_SECURITY_FINISH
,
3126 MANAGER_TIMESTAMP_GENERATORS_START
, MANAGER_TIMESTAMP_GENERATORS_FINISH
,
3127 MANAGER_TIMESTAMP_UNITS_LOAD_START
, MANAGER_TIMESTAMP_UNITS_LOAD_FINISH
))
3130 t
= manager_timestamp_to_string(q
);
3131 const char *field
= strjoina(t
, "-timestamp");
3132 dual_timestamp_serialize(f
, field
, m
->timestamps
+ q
);
3135 if (!switching_root
)
3136 (void) serialize_environment(f
, m
->environment
);
3138 if (m
->notify_fd
>= 0) {
3141 copy
= fdset_put_dup(fds
, m
->notify_fd
);
3145 fprintf(f
, "notify-fd=%i\n", copy
);
3146 fprintf(f
, "notify-socket=%s\n", m
->notify_socket
);
3149 if (m
->cgroups_agent_fd
>= 0) {
3152 copy
= fdset_put_dup(fds
, m
->cgroups_agent_fd
);
3156 fprintf(f
, "cgroups-agent-fd=%i\n", copy
);
3159 if (m
->user_lookup_fds
[0] >= 0) {
3162 copy0
= fdset_put_dup(fds
, m
->user_lookup_fds
[0]);
3166 copy1
= fdset_put_dup(fds
, m
->user_lookup_fds
[1]);
3170 fprintf(f
, "user-lookup=%i %i\n", copy0
, copy1
);
3173 bus_track_serialize(m
->subscribed
, f
, "subscribed");
3175 r
= dynamic_user_serialize(m
, f
, fds
);
3179 manager_serialize_uid_refs(m
, f
);
3180 manager_serialize_gid_refs(m
, f
);
3182 r
= exec_runtime_serialize(m
, f
, fds
);
3186 (void) fputc('\n', f
);
3188 HASHMAP_FOREACH_KEY(u
, t
, m
->units
, i
) {
3196 r
= unit_serialize(u
, f
, fds
, !switching_root
);
3201 r
= fflush_and_check(f
);
3205 r
= bus_fdset_add_all(m
, fds
);
3212 int manager_deserialize(Manager
*m
, FILE *f
, FDSet
*fds
) {
3218 log_debug("Deserializing state...");
3220 /* If we are not in reload mode yet, enter it now. Not that this is recursive, a caller might already have
3221 * increased it to non-zero, which is why we just increase it by one here and down again at the end of this
3223 _cleanup_(manager_reloading_stopp
) _unused_ Manager
*reloading
= manager_reloading_start(m
);
3226 char line
[LINE_MAX
];
3227 const char *val
, *l
;
3230 if (!fgets(line
, sizeof(line
), f
)) {
3232 return -errno
?: -EIO
;
3242 if ((val
= startswith(l
, "current-job-id="))) {
3245 if (safe_atou32(val
, &id
) < 0)
3246 log_notice("Failed to parse current job id value '%s', ignoring.", val
);
3248 m
->current_job_id
= MAX(m
->current_job_id
, id
);
3250 } else if ((val
= startswith(l
, "n-installed-jobs="))) {
3253 if (safe_atou32(val
, &n
) < 0)
3254 log_notice("Failed to parse installed jobs counter '%s', ignoring.", val
);
3256 m
->n_installed_jobs
+= n
;
3258 } else if ((val
= startswith(l
, "n-failed-jobs="))) {
3261 if (safe_atou32(val
, &n
) < 0)
3262 log_notice("Failed to parse failed jobs counter '%s', ignoring.", val
);
3264 m
->n_failed_jobs
+= n
;
3266 } else if ((val
= startswith(l
, "taint-usr="))) {
3269 b
= parse_boolean(val
);
3271 log_notice("Failed to parse taint /usr flag '%s', ignoring.", val
);
3273 m
->taint_usr
= m
->taint_usr
|| b
;
3275 } else if ((val
= startswith(l
, "ready-sent="))) {
3278 b
= parse_boolean(val
);
3280 log_notice("Failed to parse ready-sent flag '%s', ignoring.", val
);
3282 m
->ready_sent
= m
->ready_sent
|| b
;
3284 } else if ((val
= startswith(l
, "taint-logged="))) {
3287 b
= parse_boolean(val
);
3289 log_notice("Failed to parse taint-logged flag '%s', ignoring.", val
);
3291 m
->taint_logged
= m
->taint_logged
|| b
;
3293 } else if ((val
= startswith(l
, "service-watchdogs="))) {
3296 b
= parse_boolean(val
);
3298 log_notice("Failed to parse service-watchdogs flag '%s', ignoring.", val
);
3300 m
->service_watchdogs
= b
;
3302 } else if ((val
= startswith(l
, "show-status="))) {
3305 s
= show_status_from_string(val
);
3307 log_notice("Failed to parse show-status flag '%s', ignoring.", val
);
3309 manager_set_show_status(m
, s
);
3311 } else if ((val
= startswith(l
, "log-level-override="))) {
3314 level
= log_level_from_string(val
);
3316 log_notice("Failed to parse log-level-override value '%s', ignoring.", val
);
3318 manager_override_log_level(m
, level
);
3320 } else if ((val
= startswith(l
, "log-target-override="))) {
3323 target
= log_target_from_string(val
);
3325 log_notice("Failed to parse log-target-override value '%s', ignoring.", val
);
3327 manager_override_log_target(m
, target
);
3329 } else if (startswith(l
, "env=")) {
3330 r
= deserialize_environment(&m
->environment
, l
);
3334 log_notice_errno(r
, "Failed to parse environment entry: \"%s\", ignoring: %m", l
);
3336 } else if ((val
= startswith(l
, "notify-fd="))) {
3339 if (safe_atoi(val
, &fd
) < 0 || fd
< 0 || !fdset_contains(fds
, fd
))
3340 log_notice("Failed to parse notify fd, ignoring: \"%s\"", val
);
3342 m
->notify_event_source
= sd_event_source_unref(m
->notify_event_source
);
3343 safe_close(m
->notify_fd
);
3344 m
->notify_fd
= fdset_remove(fds
, fd
);
3347 } else if ((val
= startswith(l
, "notify-socket="))) {
3348 r
= free_and_strdup(&m
->notify_socket
, val
);
3352 } else if ((val
= startswith(l
, "cgroups-agent-fd="))) {
3355 if (safe_atoi(val
, &fd
) < 0 || fd
< 0 || !fdset_contains(fds
, fd
))
3356 log_notice("Failed to parse cgroups agent fd, ignoring.: %s", val
);
3358 m
->cgroups_agent_event_source
= sd_event_source_unref(m
->cgroups_agent_event_source
);
3359 safe_close(m
->cgroups_agent_fd
);
3360 m
->cgroups_agent_fd
= fdset_remove(fds
, fd
);
3363 } else if ((val
= startswith(l
, "user-lookup="))) {
3366 if (sscanf(val
, "%i %i", &fd0
, &fd1
) != 2 || fd0
< 0 || fd1
< 0 || fd0
== fd1
|| !fdset_contains(fds
, fd0
) || !fdset_contains(fds
, fd1
))
3367 log_notice("Failed to parse user lookup fd, ignoring: %s", val
);
3369 m
->user_lookup_event_source
= sd_event_source_unref(m
->user_lookup_event_source
);
3370 safe_close_pair(m
->user_lookup_fds
);
3371 m
->user_lookup_fds
[0] = fdset_remove(fds
, fd0
);
3372 m
->user_lookup_fds
[1] = fdset_remove(fds
, fd1
);
3375 } else if ((val
= startswith(l
, "dynamic-user=")))
3376 dynamic_user_deserialize_one(m
, val
, fds
);
3377 else if ((val
= startswith(l
, "destroy-ipc-uid=")))
3378 manager_deserialize_uid_refs_one(m
, val
);
3379 else if ((val
= startswith(l
, "destroy-ipc-gid=")))
3380 manager_deserialize_gid_refs_one(m
, val
);
3381 else if ((val
= startswith(l
, "exec-runtime=")))
3382 exec_runtime_deserialize_one(m
, val
, fds
);
3383 else if ((val
= startswith(l
, "subscribed="))) {
3385 if (strv_extend(&m
->deserialized_subscribed
, val
) < 0)
3391 for (q
= 0; q
< _MANAGER_TIMESTAMP_MAX
; q
++) {
3392 val
= startswith(l
, manager_timestamp_to_string(q
));
3396 val
= startswith(val
, "-timestamp=");
3401 if (q
< _MANAGER_TIMESTAMP_MAX
) /* found it */
3402 dual_timestamp_deserialize(val
, m
->timestamps
+ q
);
3403 else if (!startswith(l
, "kdbus-fd=")) /* ignore kdbus */
3404 log_notice("Unknown serialization item '%s', ignoring.", l
);
3410 char name
[UNIT_NAME_MAX
+2];
3411 const char* unit_name
;
3415 if (!fgets(name
, sizeof(name
), f
)) {
3417 return -errno
?: -EIO
;
3422 unit_name
= strstrip(name
);
3424 r
= manager_load_unit(m
, unit_name
, NULL
, NULL
, &u
);
3429 log_notice_errno(r
, "Failed to load unit \"%s\", skipping deserialization: %m", unit_name
);
3430 unit_deserialize_skip(f
);
3434 r
= unit_deserialize(u
, f
, fds
);
3439 log_notice_errno(r
, "Failed to deserialize unit \"%s\": %m", unit_name
);
3446 static void manager_flush_finished_jobs(Manager
*m
) {
3449 while ((j
= set_steal_first(m
->pending_finished_jobs
))) {
3450 bus_job_send_removed_signal(j
);
3454 m
->pending_finished_jobs
= set_free(m
->pending_finished_jobs
);
3457 int manager_reload(Manager
*m
) {
3458 _cleanup_(manager_reloading_stopp
) Manager
*reloading
= NULL
;
3459 _cleanup_fdset_free_ FDSet
*fds
= NULL
;
3460 _cleanup_fclose_
FILE *f
= NULL
;
3465 r
= manager_open_serialization(m
, &f
);
3467 return log_error_errno(r
, "Failed to create serialization file: %m");
3473 /* We are officially in reload mode from here on. */
3474 reloading
= manager_reloading_start(m
);
3476 r
= manager_serialize(m
, f
, fds
, false);
3478 return log_error_errno(r
, "Failed to serialize manager: %m");
3480 if (fseeko(f
, 0, SEEK_SET
) < 0)
3481 return log_error_errno(errno
, "Failed to seek to beginning of serialization: %m");
3483 /* 💀 This is the point of no return, from here on there is no way back. 💀 */
3486 bus_manager_send_reloading(m
, true);
3488 /* Start by flushing out all jobs and units, all generated units, all runtime environments, all dynamic users
3489 * and everything else that is worth flushing out. We'll get it all back from the serialization — if we need
3492 manager_clear_jobs_and_units(m
);
3493 lookup_paths_flush_generator(&m
->lookup_paths
);
3494 lookup_paths_free(&m
->lookup_paths
);
3495 exec_runtime_vacuum(m
);
3496 dynamic_user_vacuum(m
, false);
3497 m
->uid_refs
= hashmap_free(m
->uid_refs
);
3498 m
->gid_refs
= hashmap_free(m
->gid_refs
);
3500 r
= lookup_paths_init(&m
->lookup_paths
, m
->unit_file_scope
, 0, NULL
);
3502 log_warning_errno(r
, "Failed to initialize path lookup table, ignoring: %m");
3504 (void) manager_run_environment_generators(m
);
3505 (void) manager_run_generators(m
);
3507 r
= lookup_paths_reduce(&m
->lookup_paths
);
3509 log_warning_errno(r
, "Failed ot reduce unit file paths, ignoring: %m");
3511 manager_build_unit_path_cache(m
);
3513 /* First, enumerate what we can from kernel and suchlike */
3514 manager_enumerate_perpetual(m
);
3515 manager_enumerate(m
);
3517 /* Second, deserialize our stored data */
3518 r
= manager_deserialize(m
, f
, fds
);
3520 log_warning_errno(r
, "Deserialization failed, proceeding anyway: %m");
3522 /* We don't need the serialization anymore */
3525 /* Re-register notify_fd as event source, and set up other sockets/communication channels we might need */
3526 (void) manager_setup_notify(m
);
3527 (void) manager_setup_cgroups_agent(m
);
3528 (void) manager_setup_user_lookup_fd(m
);
3530 /* Third, fire things up! */
3531 manager_coldplug(m
);
3533 /* Clean up runtime objects no longer referenced */
3536 /* Consider the reload process complete now. */
3537 assert(m
->n_reloading
> 0);
3542 if (!MANAGER_IS_RELOADING(m
))
3543 manager_flush_finished_jobs(m
);
3545 m
->send_reloading_done
= true;
3549 void manager_reset_failed(Manager
*m
) {
3555 HASHMAP_FOREACH(u
, m
->units
, i
)
3556 unit_reset_failed(u
);
3559 bool manager_unit_inactive_or_pending(Manager
*m
, const char *name
) {
3565 /* Returns true if the unit is inactive or going down */
3566 u
= manager_get_unit(m
, name
);
3570 return unit_inactive_or_pending(u
);
3573 static void log_taint_string(Manager
*m
) {
3574 _cleanup_free_
char *taint
= NULL
;
3578 if (MANAGER_IS_USER(m
) || m
->taint_logged
)
3581 m
->taint_logged
= true; /* only check for taint once */
3583 taint
= manager_taint_string(m
);
3587 log_struct(LOG_NOTICE
,
3588 LOG_MESSAGE("System is tainted: %s", taint
),
3590 "MESSAGE_ID=" SD_MESSAGE_TAINTED_STR
);
3593 static void manager_notify_finished(Manager
*m
) {
3594 char userspace
[FORMAT_TIMESPAN_MAX
], initrd
[FORMAT_TIMESPAN_MAX
], kernel
[FORMAT_TIMESPAN_MAX
], sum
[FORMAT_TIMESPAN_MAX
];
3595 usec_t firmware_usec
, loader_usec
, kernel_usec
, initrd_usec
, userspace_usec
, total_usec
;
3597 if (MANAGER_IS_TEST_RUN(m
))
3600 if (MANAGER_IS_SYSTEM(m
) && detect_container() <= 0) {
3601 char ts
[FORMAT_TIMESPAN_MAX
];
3602 char buf
[FORMAT_TIMESPAN_MAX
+ STRLEN(" (firmware) + ") + FORMAT_TIMESPAN_MAX
+ STRLEN(" (loader) + ")]
3605 size_t size
= sizeof buf
;
3607 /* Note that MANAGER_TIMESTAMP_KERNEL's monotonic value is always at 0, and
3608 * MANAGER_TIMESTAMP_FIRMWARE's and MANAGER_TIMESTAMP_LOADER's monotonic value should be considered
3609 * negative values. */
3611 firmware_usec
= m
->timestamps
[MANAGER_TIMESTAMP_FIRMWARE
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_LOADER
].monotonic
;
3612 loader_usec
= m
->timestamps
[MANAGER_TIMESTAMP_LOADER
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_KERNEL
].monotonic
;
3613 userspace_usec
= m
->timestamps
[MANAGER_TIMESTAMP_FINISH
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_USERSPACE
].monotonic
;
3614 total_usec
= m
->timestamps
[MANAGER_TIMESTAMP_FIRMWARE
].monotonic
+ m
->timestamps
[MANAGER_TIMESTAMP_FINISH
].monotonic
;
3616 if (firmware_usec
> 0)
3617 size
= strpcpyf(&p
, size
, "%s (firmware) + ", format_timespan(ts
, sizeof(ts
), firmware_usec
, USEC_PER_MSEC
));
3618 if (loader_usec
> 0)
3619 size
= strpcpyf(&p
, size
, "%s (loader) + ", format_timespan(ts
, sizeof(ts
), loader_usec
, USEC_PER_MSEC
));
3621 if (dual_timestamp_is_set(&m
->timestamps
[MANAGER_TIMESTAMP_INITRD
])) {
3623 /* The initrd case on bare-metal*/
3624 kernel_usec
= m
->timestamps
[MANAGER_TIMESTAMP_INITRD
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_KERNEL
].monotonic
;
3625 initrd_usec
= m
->timestamps
[MANAGER_TIMESTAMP_USERSPACE
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_INITRD
].monotonic
;
3627 log_struct(LOG_INFO
,
3628 "MESSAGE_ID=" SD_MESSAGE_STARTUP_FINISHED_STR
,
3629 "KERNEL_USEC="USEC_FMT
, kernel_usec
,
3630 "INITRD_USEC="USEC_FMT
, initrd_usec
,
3631 "USERSPACE_USEC="USEC_FMT
, userspace_usec
,
3632 LOG_MESSAGE("Startup finished in %s%s (kernel) + %s (initrd) + %s (userspace) = %s.",
3634 format_timespan(kernel
, sizeof(kernel
), kernel_usec
, USEC_PER_MSEC
),
3635 format_timespan(initrd
, sizeof(initrd
), initrd_usec
, USEC_PER_MSEC
),
3636 format_timespan(userspace
, sizeof(userspace
), userspace_usec
, USEC_PER_MSEC
),
3637 format_timespan(sum
, sizeof(sum
), total_usec
, USEC_PER_MSEC
)));
3639 /* The initrd-less case on bare-metal*/
3641 kernel_usec
= m
->timestamps
[MANAGER_TIMESTAMP_USERSPACE
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_KERNEL
].monotonic
;
3644 log_struct(LOG_INFO
,
3645 "MESSAGE_ID=" SD_MESSAGE_STARTUP_FINISHED_STR
,
3646 "KERNEL_USEC="USEC_FMT
, kernel_usec
,
3647 "USERSPACE_USEC="USEC_FMT
, userspace_usec
,
3648 LOG_MESSAGE("Startup finished in %s%s (kernel) + %s (userspace) = %s.",
3650 format_timespan(kernel
, sizeof(kernel
), kernel_usec
, USEC_PER_MSEC
),
3651 format_timespan(userspace
, sizeof(userspace
), userspace_usec
, USEC_PER_MSEC
),
3652 format_timespan(sum
, sizeof(sum
), total_usec
, USEC_PER_MSEC
)));
3655 /* The container and --user case */
3656 firmware_usec
= loader_usec
= initrd_usec
= kernel_usec
= 0;
3657 total_usec
= userspace_usec
= m
->timestamps
[MANAGER_TIMESTAMP_FINISH
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_USERSPACE
].monotonic
;
3659 log_struct(LOG_INFO
,
3660 "MESSAGE_ID=" SD_MESSAGE_USER_STARTUP_FINISHED_STR
,
3661 "USERSPACE_USEC="USEC_FMT
, userspace_usec
,
3662 LOG_MESSAGE("Startup finished in %s.",
3663 format_timespan(sum
, sizeof(sum
), total_usec
, USEC_PER_MSEC
)));
3666 bus_manager_send_finished(m
, firmware_usec
, loader_usec
, kernel_usec
, initrd_usec
, userspace_usec
, total_usec
);
3669 m
->ready_sent
? "STATUS=Startup finished in %s."
3671 "STATUS=Startup finished in %s.",
3672 format_timespan(sum
, sizeof(sum
), total_usec
, USEC_PER_MSEC
));
3673 m
->ready_sent
= true;
3675 log_taint_string(m
);
3678 static void manager_send_ready(Manager
*m
) {
3681 /* We send READY=1 on reaching basic.target only when running in --user mode. */
3682 if (!MANAGER_IS_USER(m
) || m
->ready_sent
)
3685 m
->ready_sent
= true;
3689 "STATUS=Reached " SPECIAL_BASIC_TARGET
".");
3692 static void manager_check_basic_target(Manager
*m
) {
3697 /* Small shortcut */
3698 if (m
->ready_sent
&& m
->taint_logged
)
3701 u
= manager_get_unit(m
, SPECIAL_BASIC_TARGET
);
3702 if (!u
|| !UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u
)))
3705 /* For user managers, send out READY=1 as soon as we reach basic.target */
3706 manager_send_ready(m
);
3708 /* Log the taint string as soon as we reach basic.target */
3709 log_taint_string(m
);
3712 void manager_check_finished(Manager
*m
) {
3715 if (MANAGER_IS_RELOADING(m
))
3718 /* Verify that we have entered the event loop already, and not left it again. */
3719 if (!MANAGER_IS_RUNNING(m
))
3722 manager_check_basic_target(m
);
3724 if (hashmap_size(m
->jobs
) > 0) {
3725 if (m
->jobs_in_progress_event_source
)
3726 /* Ignore any failure, this is only for feedback */
3727 (void) sd_event_source_set_time(m
->jobs_in_progress_event_source
, now(CLOCK_MONOTONIC
) + JOBS_IN_PROGRESS_WAIT_USEC
);
3732 manager_flip_auto_status(m
, false);
3734 /* Notify Type=idle units that we are done now */
3735 manager_close_idle_pipe(m
);
3737 /* Turn off confirm spawn now */
3738 m
->confirm_spawn
= NULL
;
3740 /* No need to update ask password status when we're going non-interactive */
3741 manager_close_ask_password(m
);
3743 /* This is no longer the first boot */
3744 manager_set_first_boot(m
, false);
3746 if (MANAGER_IS_FINISHED(m
))
3749 dual_timestamp_get(m
->timestamps
+ MANAGER_TIMESTAMP_FINISH
);
3751 manager_notify_finished(m
);
3753 manager_invalidate_startup_units(m
);
3756 static bool generator_path_any(const char* const* paths
) {
3760 /* Optimize by skipping the whole process by not creating output directories
3761 * if no generators are found. */
3762 STRV_FOREACH(path
, (char**) paths
)
3763 if (access(*path
, F_OK
) == 0)
3765 else if (errno
!= ENOENT
)
3766 log_warning_errno(errno
, "Failed to open generator directory %s: %m", *path
);
3771 static const char* system_env_generator_binary_paths
[] = {
3772 "/run/systemd/system-environment-generators",
3773 "/etc/systemd/system-environment-generators",
3774 "/usr/local/lib/systemd/system-environment-generators",
3775 SYSTEM_ENV_GENERATOR_PATH
,
3779 static const char* user_env_generator_binary_paths
[] = {
3780 "/run/systemd/user-environment-generators",
3781 "/etc/systemd/user-environment-generators",
3782 "/usr/local/lib/systemd/user-environment-generators",
3783 USER_ENV_GENERATOR_PATH
,
3787 static int manager_run_environment_generators(Manager
*m
) {
3788 char **tmp
= NULL
; /* this is only used in the forked process, no cleanup here */
3790 void* args
[] = {&tmp
, &tmp
, &m
->environment
};
3792 if (MANAGER_IS_TEST_RUN(m
) && !(m
->test_run_flags
& MANAGER_TEST_RUN_ENV_GENERATORS
))
3795 paths
= MANAGER_IS_SYSTEM(m
) ? system_env_generator_binary_paths
: user_env_generator_binary_paths
;
3797 if (!generator_path_any(paths
))
3800 return execute_directories(paths
, DEFAULT_TIMEOUT_USEC
, gather_environment
, args
, NULL
, m
->environment
);
3803 static int manager_run_generators(Manager
*m
) {
3804 _cleanup_strv_free_
char **paths
= NULL
;
3805 const char *argv
[5];
3810 if (MANAGER_IS_TEST_RUN(m
) && !(m
->test_run_flags
& MANAGER_TEST_RUN_GENERATORS
))
3813 paths
= generator_binary_paths(m
->unit_file_scope
);
3817 if (!generator_path_any((const char* const*) paths
))
3820 r
= lookup_paths_mkdir_generator(&m
->lookup_paths
);
3822 log_error_errno(r
, "Failed to create generator directories: %m");
3826 argv
[0] = NULL
; /* Leave this empty, execute_directory() will fill something in */
3827 argv
[1] = m
->lookup_paths
.generator
;
3828 argv
[2] = m
->lookup_paths
.generator_early
;
3829 argv
[3] = m
->lookup_paths
.generator_late
;
3832 RUN_WITH_UMASK(0022)
3833 (void) execute_directories((const char* const*) paths
, DEFAULT_TIMEOUT_USEC
,
3834 NULL
, NULL
, (char**) argv
, m
->environment
);
3839 lookup_paths_trim_generator(&m
->lookup_paths
);
3843 int manager_environment_add(Manager
*m
, char **minus
, char **plus
) {
3844 char **a
= NULL
, **b
= NULL
, **l
;
3849 if (!strv_isempty(minus
)) {
3850 a
= strv_env_delete(l
, 1, minus
);
3857 if (!strv_isempty(plus
)) {
3858 b
= strv_env_merge(2, l
, plus
);
3867 if (m
->environment
!= l
)
3868 strv_free(m
->environment
);
3875 manager_sanitize_environment(m
);
3880 int manager_set_default_rlimits(Manager
*m
, struct rlimit
**default_rlimit
) {
3885 for (i
= 0; i
< _RLIMIT_MAX
; i
++) {
3886 m
->rlimit
[i
] = mfree(m
->rlimit
[i
]);
3888 if (!default_rlimit
[i
])
3891 m
->rlimit
[i
] = newdup(struct rlimit
, default_rlimit
[i
], 1);
3899 void manager_recheck_dbus(Manager
*m
) {
3902 /* Connects to the bus if the dbus service and socket are running. If we are running in user mode this is all
3903 * it does. In system mode we'll also connect to the system bus (which will most likely just reuse the
3904 * connection of the API bus). That's because the system bus after all runs as service of the system instance,
3905 * while in the user instance we can assume it's already there. */
3907 if (MANAGER_IS_RELOADING(m
))
3908 return; /* don't check while we are reloading… */
3910 if (manager_dbus_is_running(m
, false)) {
3911 (void) bus_init_api(m
);
3913 if (MANAGER_IS_SYSTEM(m
))
3914 (void) bus_init_system(m
);
3916 (void) bus_done_api(m
);
3918 if (MANAGER_IS_SYSTEM(m
))
3919 (void) bus_done_system(m
);
3923 static bool manager_journal_is_running(Manager
*m
) {
3928 if (MANAGER_IS_TEST_RUN(m
))
3931 /* If we are the user manager we can safely assume that the journal is up */
3932 if (!MANAGER_IS_SYSTEM(m
))
3935 /* Check that the socket is not only up, but in RUNNING state */
3936 u
= manager_get_unit(m
, SPECIAL_JOURNALD_SOCKET
);
3939 if (SOCKET(u
)->state
!= SOCKET_RUNNING
)
3942 /* Similar, check if the daemon itself is fully up, too */
3943 u
= manager_get_unit(m
, SPECIAL_JOURNALD_SERVICE
);
3946 if (!IN_SET(SERVICE(u
)->state
, SERVICE_RELOAD
, SERVICE_RUNNING
))
3952 void manager_recheck_journal(Manager
*m
) {
3956 /* Don't bother with this unless we are in the special situation of being PID 1 */
3957 if (getpid_cached() != 1)
3960 /* Don't check this while we are reloading, things might still change */
3961 if (MANAGER_IS_RELOADING(m
))
3964 /* The journal is fully and entirely up? If so, let's permit logging to it, if that's configured. If the
3965 * journal is down, don't ever log to it, otherwise we might end up deadlocking ourselves as we might trigger
3966 * an activation ourselves we can't fulfill. */
3967 log_set_prohibit_ipc(!manager_journal_is_running(m
));
3971 void manager_set_show_status(Manager
*m
, ShowStatus mode
) {
3973 assert(IN_SET(mode
, SHOW_STATUS_AUTO
, SHOW_STATUS_NO
, SHOW_STATUS_YES
, SHOW_STATUS_TEMPORARY
));
3975 if (!MANAGER_IS_SYSTEM(m
))
3978 if (m
->show_status
!= mode
)
3979 log_debug("%s showing of status.",
3980 mode
== SHOW_STATUS_NO
? "Disabling" : "Enabling");
3981 m
->show_status
= mode
;
3983 if (IN_SET(mode
, SHOW_STATUS_TEMPORARY
, SHOW_STATUS_YES
))
3984 (void) touch("/run/systemd/show-status");
3986 (void) unlink("/run/systemd/show-status");
3989 static bool manager_get_show_status(Manager
*m
, StatusType type
) {
3992 if (!MANAGER_IS_SYSTEM(m
))
3995 if (m
->no_console_output
)
3998 if (!IN_SET(manager_state(m
), MANAGER_INITIALIZING
, MANAGER_STARTING
, MANAGER_STOPPING
))
4001 /* If we cannot find out the status properly, just proceed. */
4002 if (type
!= STATUS_TYPE_EMERGENCY
&& manager_check_ask_password(m
) > 0)
4005 return IN_SET(m
->show_status
, SHOW_STATUS_TEMPORARY
, SHOW_STATUS_YES
);
4008 const char *manager_get_confirm_spawn(Manager
*m
) {
4009 static int last_errno
= 0;
4010 const char *vc
= m
->confirm_spawn
;
4014 /* Here's the deal: we want to test the validity of the console but don't want
4015 * PID1 to go through the whole console process which might block. But we also
4016 * want to warn the user only once if something is wrong with the console so we
4017 * cannot do the sanity checks after spawning our children. So here we simply do
4018 * really basic tests to hopefully trap common errors.
4020 * If the console suddenly disappear at the time our children will really it
4021 * then they will simply fail to acquire it and a positive answer will be
4022 * assumed. New children will fallback to /dev/console though.
4024 * Note: TTYs are devices that can come and go any time, and frequently aren't
4025 * available yet during early boot (consider a USB rs232 dongle...). If for any
4026 * reason the configured console is not ready, we fallback to the default
4029 if (!vc
|| path_equal(vc
, "/dev/console"))
4036 if (!S_ISCHR(st
.st_mode
)) {
4044 if (last_errno
!= errno
) {
4046 log_warning_errno(errno
, "Failed to open %s: %m, using default console", vc
);
4048 return "/dev/console";
4051 void manager_set_first_boot(Manager
*m
, bool b
) {
4054 if (!MANAGER_IS_SYSTEM(m
))
4057 if (m
->first_boot
!= (int) b
) {
4059 (void) touch("/run/systemd/first-boot");
4061 (void) unlink("/run/systemd/first-boot");
4067 void manager_disable_confirm_spawn(void) {
4068 (void) touch("/run/systemd/confirm_spawn_disabled");
4071 bool manager_is_confirm_spawn_disabled(Manager
*m
) {
4072 if (!m
->confirm_spawn
)
4075 return access("/run/systemd/confirm_spawn_disabled", F_OK
) >= 0;
4078 void manager_status_printf(Manager
*m
, StatusType type
, const char *status
, const char *format
, ...) {
4081 /* If m is NULL, assume we're after shutdown and let the messages through. */
4083 if (m
&& !manager_get_show_status(m
, type
))
4086 /* XXX We should totally drop the check for ephemeral here
4087 * and thus effectively make 'Type=idle' pointless. */
4088 if (type
== STATUS_TYPE_EPHEMERAL
&& m
&& m
->n_on_console
> 0)
4091 va_start(ap
, format
);
4092 status_vprintf(status
, true, type
== STATUS_TYPE_EPHEMERAL
, format
, ap
);
4096 Set
*manager_get_units_requiring_mounts_for(Manager
*m
, const char *path
) {
4097 char p
[strlen(path
)+1];
4103 path_simplify(p
, false);
4105 return hashmap_get(m
->units_requiring_mounts_for
, streq(p
, "/") ? "" : p
);
4108 int manager_update_failed_units(Manager
*m
, Unit
*u
, bool failed
) {
4113 assert(u
->manager
== m
);
4115 size
= set_size(m
->failed_units
);
4118 r
= set_ensure_allocated(&m
->failed_units
, NULL
);
4122 if (set_put(m
->failed_units
, u
) < 0)
4125 (void) set_remove(m
->failed_units
, u
);
4127 if (set_size(m
->failed_units
) != size
)
4128 bus_manager_send_change_signal(m
);
4133 ManagerState
manager_state(Manager
*m
) {
4138 /* Did we ever finish booting? If not then we are still starting up */
4139 if (!MANAGER_IS_FINISHED(m
)) {
4141 u
= manager_get_unit(m
, SPECIAL_BASIC_TARGET
);
4142 if (!u
|| !UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u
)))
4143 return MANAGER_INITIALIZING
;
4145 return MANAGER_STARTING
;
4148 /* Is the special shutdown target active or queued? If so, we are in shutdown state */
4149 u
= manager_get_unit(m
, SPECIAL_SHUTDOWN_TARGET
);
4150 if (u
&& unit_active_or_pending(u
))
4151 return MANAGER_STOPPING
;
4153 if (MANAGER_IS_SYSTEM(m
)) {
4154 /* Are the rescue or emergency targets active or queued? If so we are in maintenance state */
4155 u
= manager_get_unit(m
, SPECIAL_RESCUE_TARGET
);
4156 if (u
&& unit_active_or_pending(u
))
4157 return MANAGER_MAINTENANCE
;
4159 u
= manager_get_unit(m
, SPECIAL_EMERGENCY_TARGET
);
4160 if (u
&& unit_active_or_pending(u
))
4161 return MANAGER_MAINTENANCE
;
4164 /* Are there any failed units? If so, we are in degraded mode */
4165 if (set_size(m
->failed_units
) > 0)
4166 return MANAGER_DEGRADED
;
4168 return MANAGER_RUNNING
;
4171 #define DESTROY_IPC_FLAG (UINT32_C(1) << 31)
4173 static void manager_unref_uid_internal(
4178 int (*_clean_ipc
)(uid_t uid
)) {
4184 assert(uid_is_valid(uid
));
4187 /* A generic implementation, covering both manager_unref_uid() and manager_unref_gid(), under the assumption
4188 * that uid_t and gid_t are actually defined the same way, with the same validity rules.
4190 * We store a hashmap where the UID/GID is they key and the value is a 32bit reference counter, whose highest
4191 * bit is used as flag for marking UIDs/GIDs whose IPC objects to remove when the last reference to the UID/GID
4192 * is dropped. The flag is set to on, once at least one reference from a unit where RemoveIPC= is set is added
4193 * on a UID/GID. It is reset when the UID's/GID's reference counter drops to 0 again. */
4195 assert_cc(sizeof(uid_t
) == sizeof(gid_t
));
4196 assert_cc(UID_INVALID
== (uid_t
) GID_INVALID
);
4198 if (uid
== 0) /* We don't keep track of root, and will never destroy it */
4201 c
= PTR_TO_UINT32(hashmap_get(*uid_refs
, UID_TO_PTR(uid
)));
4203 n
= c
& ~DESTROY_IPC_FLAG
;
4207 if (destroy_now
&& n
== 0) {
4208 hashmap_remove(*uid_refs
, UID_TO_PTR(uid
));
4210 if (c
& DESTROY_IPC_FLAG
) {
4211 log_debug("%s " UID_FMT
" is no longer referenced, cleaning up its IPC.",
4212 _clean_ipc
== clean_ipc_by_uid
? "UID" : "GID",
4214 (void) _clean_ipc(uid
);
4217 c
= n
| (c
& DESTROY_IPC_FLAG
);
4218 assert_se(hashmap_update(*uid_refs
, UID_TO_PTR(uid
), UINT32_TO_PTR(c
)) >= 0);
4222 void manager_unref_uid(Manager
*m
, uid_t uid
, bool destroy_now
) {
4223 manager_unref_uid_internal(m
, &m
->uid_refs
, uid
, destroy_now
, clean_ipc_by_uid
);
4226 void manager_unref_gid(Manager
*m
, gid_t gid
, bool destroy_now
) {
4227 manager_unref_uid_internal(m
, &m
->gid_refs
, (uid_t
) gid
, destroy_now
, clean_ipc_by_gid
);
4230 static int manager_ref_uid_internal(
4241 assert(uid_is_valid(uid
));
4243 /* A generic implementation, covering both manager_ref_uid() and manager_ref_gid(), under the assumption
4244 * that uid_t and gid_t are actually defined the same way, with the same validity rules. */
4246 assert_cc(sizeof(uid_t
) == sizeof(gid_t
));
4247 assert_cc(UID_INVALID
== (uid_t
) GID_INVALID
);
4249 if (uid
== 0) /* We don't keep track of root, and will never destroy it */
4252 r
= hashmap_ensure_allocated(uid_refs
, &trivial_hash_ops
);
4256 c
= PTR_TO_UINT32(hashmap_get(*uid_refs
, UID_TO_PTR(uid
)));
4258 n
= c
& ~DESTROY_IPC_FLAG
;
4261 if (n
& DESTROY_IPC_FLAG
) /* check for overflow */
4264 c
= n
| (c
& DESTROY_IPC_FLAG
) | (clean_ipc
? DESTROY_IPC_FLAG
: 0);
4266 return hashmap_replace(*uid_refs
, UID_TO_PTR(uid
), UINT32_TO_PTR(c
));
4269 int manager_ref_uid(Manager
*m
, uid_t uid
, bool clean_ipc
) {
4270 return manager_ref_uid_internal(m
, &m
->uid_refs
, uid
, clean_ipc
);
4273 int manager_ref_gid(Manager
*m
, gid_t gid
, bool clean_ipc
) {
4274 return manager_ref_uid_internal(m
, &m
->gid_refs
, (uid_t
) gid
, clean_ipc
);
4277 static void manager_vacuum_uid_refs_internal(
4280 int (*_clean_ipc
)(uid_t uid
)) {
4289 HASHMAP_FOREACH_KEY(p
, k
, *uid_refs
, i
) {
4293 uid
= PTR_TO_UID(k
);
4294 c
= PTR_TO_UINT32(p
);
4296 n
= c
& ~DESTROY_IPC_FLAG
;
4300 if (c
& DESTROY_IPC_FLAG
) {
4301 log_debug("Found unreferenced %s " UID_FMT
" after reload/reexec. Cleaning up.",
4302 _clean_ipc
== clean_ipc_by_uid
? "UID" : "GID",
4304 (void) _clean_ipc(uid
);
4307 assert_se(hashmap_remove(*uid_refs
, k
) == p
);
4311 void manager_vacuum_uid_refs(Manager
*m
) {
4312 manager_vacuum_uid_refs_internal(m
, &m
->uid_refs
, clean_ipc_by_uid
);
4315 void manager_vacuum_gid_refs(Manager
*m
) {
4316 manager_vacuum_uid_refs_internal(m
, &m
->gid_refs
, clean_ipc_by_gid
);
4319 static void manager_serialize_uid_refs_internal(
4323 const char *field_name
) {
4333 /* Serialize the UID reference table. Or actually, just the IPC destruction flag of it, as the actual counter
4334 * of it is better rebuild after a reload/reexec. */
4336 HASHMAP_FOREACH_KEY(p
, k
, *uid_refs
, i
) {
4340 uid
= PTR_TO_UID(k
);
4341 c
= PTR_TO_UINT32(p
);
4343 if (!(c
& DESTROY_IPC_FLAG
))
4346 fprintf(f
, "%s=" UID_FMT
"\n", field_name
, uid
);
4350 void manager_serialize_uid_refs(Manager
*m
, FILE *f
) {
4351 manager_serialize_uid_refs_internal(m
, f
, &m
->uid_refs
, "destroy-ipc-uid");
4354 void manager_serialize_gid_refs(Manager
*m
, FILE *f
) {
4355 manager_serialize_uid_refs_internal(m
, f
, &m
->gid_refs
, "destroy-ipc-gid");
4358 static void manager_deserialize_uid_refs_one_internal(
4361 const char *value
) {
4371 r
= parse_uid(value
, &uid
);
4372 if (r
< 0 || uid
== 0) {
4373 log_debug("Unable to parse UID reference serialization");
4377 r
= hashmap_ensure_allocated(uid_refs
, &trivial_hash_ops
);
4383 c
= PTR_TO_UINT32(hashmap_get(*uid_refs
, UID_TO_PTR(uid
)));
4384 if (c
& DESTROY_IPC_FLAG
)
4387 c
|= DESTROY_IPC_FLAG
;
4389 r
= hashmap_replace(*uid_refs
, UID_TO_PTR(uid
), UINT32_TO_PTR(c
));
4391 log_debug("Failed to add UID reference entry");
4396 void manager_deserialize_uid_refs_one(Manager
*m
, const char *value
) {
4397 manager_deserialize_uid_refs_one_internal(m
, &m
->uid_refs
, value
);
4400 void manager_deserialize_gid_refs_one(Manager
*m
, const char *value
) {
4401 manager_deserialize_uid_refs_one_internal(m
, &m
->gid_refs
, value
);
4404 int manager_dispatch_user_lookup_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
4408 char unit_name
[UNIT_NAME_MAX
+1];
4411 Manager
*m
= userdata
;
4419 /* Invoked whenever a child process succeeded resolving its user/group to use and sent us the resulting UID/GID
4420 * in a datagram. We parse the datagram here and pass it off to the unit, so that it can add a reference to the
4421 * UID/GID so that it can destroy the UID/GID's IPC objects when the reference counter drops to 0. */
4423 l
= recv(fd
, &buffer
, sizeof(buffer
), MSG_DONTWAIT
);
4425 if (IN_SET(errno
, EINTR
, EAGAIN
))
4428 return log_error_errno(errno
, "Failed to read from user lookup fd: %m");
4431 if ((size_t) l
<= offsetof(struct buffer
, unit_name
)) {
4432 log_warning("Received too short user lookup message, ignoring.");
4436 if ((size_t) l
> offsetof(struct buffer
, unit_name
) + UNIT_NAME_MAX
) {
4437 log_warning("Received too long user lookup message, ignoring.");
4441 if (!uid_is_valid(buffer
.uid
) && !gid_is_valid(buffer
.gid
)) {
4442 log_warning("Got user lookup message with invalid UID/GID pair, ignoring.");
4446 n
= (size_t) l
- offsetof(struct buffer
, unit_name
);
4447 if (memchr(buffer
.unit_name
, 0, n
)) {
4448 log_warning("Received lookup message with embedded NUL character, ignoring.");
4452 buffer
.unit_name
[n
] = 0;
4453 u
= manager_get_unit(m
, buffer
.unit_name
);
4455 log_debug("Got user lookup message but unit doesn't exist, ignoring.");
4459 log_unit_debug(u
, "User lookup succeeded: uid=" UID_FMT
" gid=" GID_FMT
, buffer
.uid
, buffer
.gid
);
4461 unit_notify_user_lookup(u
, buffer
.uid
, buffer
.gid
);
4465 char *manager_taint_string(Manager
*m
) {
4466 _cleanup_free_
char *destination
= NULL
, *overflowuid
= NULL
, *overflowgid
= NULL
;
4470 /* Returns a "taint string", e.g. "local-hwclock:var-run-bad".
4471 * Only things that are detected at runtime should be tagged
4472 * here. For stuff that is set during compilation, emit a warning
4473 * in the configuration phase. */
4477 buf
= new(char, sizeof("split-usr:"
4481 "overflowuid-not-65534:"
4482 "overflowgid-not-65534:"));
4490 e
= stpcpy(e
, "split-usr:");
4492 if (access("/proc/cgroups", F_OK
) < 0)
4493 e
= stpcpy(e
, "cgroups-missing:");
4495 if (clock_is_localtime(NULL
) > 0)
4496 e
= stpcpy(e
, "local-hwclock:");
4498 r
= readlink_malloc("/var/run", &destination
);
4499 if (r
< 0 || !PATH_IN_SET(destination
, "../run", "/run"))
4500 e
= stpcpy(e
, "var-run-bad:");
4502 r
= read_one_line_file("/proc/sys/kernel/overflowuid", &overflowuid
);
4503 if (r
>= 0 && !streq(overflowuid
, "65534"))
4504 e
= stpcpy(e
, "overflowuid-not-65534:");
4506 r
= read_one_line_file("/proc/sys/kernel/overflowgid", &overflowgid
);
4507 if (r
>= 0 && !streq(overflowgid
, "65534"))
4508 e
= stpcpy(e
, "overflowgid-not-65534:");
4510 /* remove the last ':' */
4517 void manager_ref_console(Manager
*m
) {
4523 void manager_unref_console(Manager
*m
) {
4525 assert(m
->n_on_console
> 0);
4528 if (m
->n_on_console
== 0)
4529 m
->no_console_output
= false; /* unset no_console_output flag, since the console is definitely free now */
4532 void manager_override_log_level(Manager
*m
, int level
) {
4533 _cleanup_free_
char *s
= NULL
;
4536 if (!m
->log_level_overridden
) {
4537 m
->original_log_level
= log_get_max_level();
4538 m
->log_level_overridden
= true;
4541 (void) log_level_to_string_alloc(level
, &s
);
4542 log_info("Setting log level to %s.", strna(s
));
4544 log_set_max_level(level
);
4547 void manager_restore_original_log_level(Manager
*m
) {
4548 _cleanup_free_
char *s
= NULL
;
4551 if (!m
->log_level_overridden
)
4554 (void) log_level_to_string_alloc(m
->original_log_level
, &s
);
4555 log_info("Restoring log level to original (%s).", strna(s
));
4557 log_set_max_level(m
->original_log_level
);
4558 m
->log_level_overridden
= false;
4561 void manager_override_log_target(Manager
*m
, LogTarget target
) {
4564 if (!m
->log_target_overridden
) {
4565 m
->original_log_target
= log_get_target();
4566 m
->log_target_overridden
= true;
4569 log_info("Setting log target to %s.", log_target_to_string(target
));
4570 log_set_target(target
);
4573 void manager_restore_original_log_target(Manager
*m
) {
4576 if (!m
->log_target_overridden
)
4579 log_info("Restoring log target to original %s.", log_target_to_string(m
->original_log_target
));
4581 log_set_target(m
->original_log_target
);
4582 m
->log_target_overridden
= false;
4585 ManagerTimestamp
manager_timestamp_initrd_mangle(ManagerTimestamp s
) {
4587 s
>= MANAGER_TIMESTAMP_SECURITY_START
&&
4588 s
<= MANAGER_TIMESTAMP_UNITS_LOAD_FINISH
)
4589 return s
- MANAGER_TIMESTAMP_SECURITY_START
+ MANAGER_TIMESTAMP_INITRD_SECURITY_START
;
4593 static const char *const manager_state_table
[_MANAGER_STATE_MAX
] = {
4594 [MANAGER_INITIALIZING
] = "initializing",
4595 [MANAGER_STARTING
] = "starting",
4596 [MANAGER_RUNNING
] = "running",
4597 [MANAGER_DEGRADED
] = "degraded",
4598 [MANAGER_MAINTENANCE
] = "maintenance",
4599 [MANAGER_STOPPING
] = "stopping",
4602 DEFINE_STRING_TABLE_LOOKUP(manager_state
, ManagerState
);
4604 static const char *const manager_timestamp_table
[_MANAGER_TIMESTAMP_MAX
] = {
4605 [MANAGER_TIMESTAMP_FIRMWARE
] = "firmware",
4606 [MANAGER_TIMESTAMP_LOADER
] = "loader",
4607 [MANAGER_TIMESTAMP_KERNEL
] = "kernel",
4608 [MANAGER_TIMESTAMP_INITRD
] = "initrd",
4609 [MANAGER_TIMESTAMP_USERSPACE
] = "userspace",
4610 [MANAGER_TIMESTAMP_FINISH
] = "finish",
4611 [MANAGER_TIMESTAMP_SECURITY_START
] = "security-start",
4612 [MANAGER_TIMESTAMP_SECURITY_FINISH
] = "security-finish",
4613 [MANAGER_TIMESTAMP_GENERATORS_START
] = "generators-start",
4614 [MANAGER_TIMESTAMP_GENERATORS_FINISH
] = "generators-finish",
4615 [MANAGER_TIMESTAMP_UNITS_LOAD_START
] = "units-load-start",
4616 [MANAGER_TIMESTAMP_UNITS_LOAD_FINISH
] = "units-load-finish",
4617 [MANAGER_TIMESTAMP_INITRD_SECURITY_START
] = "initrd-security-start",
4618 [MANAGER_TIMESTAMP_INITRD_SECURITY_FINISH
] = "initrd-security-finish",
4619 [MANAGER_TIMESTAMP_INITRD_GENERATORS_START
] = "initrd-generators-start",
4620 [MANAGER_TIMESTAMP_INITRD_GENERATORS_FINISH
] = "initrd-generators-finish",
4621 [MANAGER_TIMESTAMP_INITRD_UNITS_LOAD_START
] = "initrd-units-load-start",
4622 [MANAGER_TIMESTAMP_INITRD_UNITS_LOAD_FINISH
] = "initrd-units-load-finish",
4625 DEFINE_STRING_TABLE_LOOKUP(manager_timestamp
, ManagerTimestamp
);