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 (m
->test_run_flags
)
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 (m
->test_run_flags
!= 0)
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
);
426 if (r
== -ENOENT
) /* If the file doesn't exist yet, subscribe to /etc instead, and wait until it is created
427 * either by O_CREATE or by rename() */
428 r
= sd_event_add_inotify(m
->event
, &new_event
, "/etc",
429 IN_CREATE
|IN_MOVED_TO
|IN_ONLYDIR
, manager_dispatch_timezone_change
, m
);
431 return log_error_errno(r
, "Failed to create timezone change event source: %m");
433 /* Schedule this slightly earlier than the .timer event sources */
434 r
= sd_event_source_set_priority(new_event
, SD_EVENT_PRIORITY_NORMAL
-1);
436 return log_error_errno(r
, "Failed to set priority of timezone change event sources: %m");
438 sd_event_source_unref(m
->timezone_change_event_source
);
439 m
->timezone_change_event_source
= TAKE_PTR(new_event
);
444 static int enable_special_signals(Manager
*m
) {
445 _cleanup_close_
int fd
= -1;
449 if (m
->test_run_flags
)
452 /* Enable that we get SIGINT on control-alt-del. In containers
453 * this will fail with EPERM (older) or EINVAL (newer), so
455 if (reboot(RB_DISABLE_CAD
) < 0 && !IN_SET(errno
, EPERM
, EINVAL
))
456 log_warning_errno(errno
, "Failed to enable ctrl-alt-del handling: %m");
458 fd
= open_terminal("/dev/tty0", O_RDWR
|O_NOCTTY
|O_CLOEXEC
);
460 /* Support systems without virtual console */
462 log_warning_errno(errno
, "Failed to open /dev/tty0: %m");
464 /* Enable that we get SIGWINCH on kbrequest */
465 if (ioctl(fd
, KDSIGACCEPT
, SIGWINCH
) < 0)
466 log_warning_errno(errno
, "Failed to enable kbrequest handling: %m");
472 #define RTSIG_IF_AVAILABLE(signum) (signum <= SIGRTMAX ? signum : -1)
474 static int manager_setup_signals(Manager
*m
) {
475 struct sigaction sa
= {
476 .sa_handler
= SIG_DFL
,
477 .sa_flags
= SA_NOCLDSTOP
|SA_RESTART
,
484 assert_se(sigaction(SIGCHLD
, &sa
, NULL
) == 0);
486 /* We make liberal use of realtime signals here. On
487 * Linux/glibc we have 30 of them (with the exception of Linux
488 * on hppa, see below), between SIGRTMIN+0 ... SIGRTMIN+30
491 assert_se(sigemptyset(&mask
) == 0);
492 sigset_add_many(&mask
,
493 SIGCHLD
, /* Child died */
494 SIGTERM
, /* Reexecute daemon */
495 SIGHUP
, /* Reload configuration */
496 SIGUSR1
, /* systemd/upstart: reconnect to D-Bus */
497 SIGUSR2
, /* systemd: dump status */
498 SIGINT
, /* Kernel sends us this on control-alt-del */
499 SIGWINCH
, /* Kernel sends us this on kbrequest (alt-arrowup) */
500 SIGPWR
, /* Some kernel drivers and upsd send us this on power failure */
502 SIGRTMIN
+0, /* systemd: start default.target */
503 SIGRTMIN
+1, /* systemd: isolate rescue.target */
504 SIGRTMIN
+2, /* systemd: isolate emergency.target */
505 SIGRTMIN
+3, /* systemd: start halt.target */
506 SIGRTMIN
+4, /* systemd: start poweroff.target */
507 SIGRTMIN
+5, /* systemd: start reboot.target */
508 SIGRTMIN
+6, /* systemd: start kexec.target */
510 /* ... space for more special targets ... */
512 SIGRTMIN
+13, /* systemd: Immediate halt */
513 SIGRTMIN
+14, /* systemd: Immediate poweroff */
514 SIGRTMIN
+15, /* systemd: Immediate reboot */
515 SIGRTMIN
+16, /* systemd: Immediate kexec */
517 /* ... space for more immediate system state changes ... */
519 SIGRTMIN
+20, /* systemd: enable status messages */
520 SIGRTMIN
+21, /* systemd: disable status messages */
521 SIGRTMIN
+22, /* systemd: set log level to LOG_DEBUG */
522 SIGRTMIN
+23, /* systemd: set log level to LOG_INFO */
523 SIGRTMIN
+24, /* systemd: Immediate exit (--user only) */
525 /* .. one free signal here ... */
527 /* Apparently Linux on hppa had fewer RT signals until v3.18,
528 * SIGRTMAX was SIGRTMIN+25, and then SIGRTMIN was lowered,
529 * see commit v3.17-7614-g1f25df2eff.
531 * We cannot unconditionally make use of those signals here,
532 * so let's use a runtime check. Since these commands are
533 * accessible by different means and only really a safety
534 * net, the missing functionality on hppa shouldn't matter.
537 RTSIG_IF_AVAILABLE(SIGRTMIN
+26), /* systemd: set log target to journal-or-kmsg */
538 RTSIG_IF_AVAILABLE(SIGRTMIN
+27), /* systemd: set log target to console */
539 RTSIG_IF_AVAILABLE(SIGRTMIN
+28), /* systemd: set log target to kmsg */
540 RTSIG_IF_AVAILABLE(SIGRTMIN
+29), /* systemd: set log target to syslog-or-kmsg (obsolete) */
542 /* ... one free signal here SIGRTMIN+30 ... */
544 assert_se(sigprocmask(SIG_SETMASK
, &mask
, NULL
) == 0);
546 m
->signal_fd
= signalfd(-1, &mask
, SFD_NONBLOCK
|SFD_CLOEXEC
);
547 if (m
->signal_fd
< 0)
550 r
= sd_event_add_io(m
->event
, &m
->signal_event_source
, m
->signal_fd
, EPOLLIN
, manager_dispatch_signal_fd
, m
);
554 (void) sd_event_source_set_description(m
->signal_event_source
, "manager-signal");
556 /* Process signals a bit earlier than the rest of things, but later than notify_fd processing, so that the
557 * notify processing can still figure out to which process/service a message belongs, before we reap the
558 * process. Also, process this before handling cgroup notifications, so that we always collect child exit
559 * status information before detecting that there's no process in a cgroup. */
560 r
= sd_event_source_set_priority(m
->signal_event_source
, SD_EVENT_PRIORITY_NORMAL
-6);
564 if (MANAGER_IS_SYSTEM(m
))
565 return enable_special_signals(m
);
570 static void manager_sanitize_environment(Manager
*m
) {
573 /* Let's remove some environment variables that we need ourselves to communicate with our clients */
593 /* Let's order the environment alphabetically, just to make it pretty */
594 strv_sort(m
->environment
);
597 static int manager_default_environment(Manager
*m
) {
600 if (MANAGER_IS_SYSTEM(m
)) {
601 /* The system manager always starts with a clean
602 * environment for its children. It does not import
603 * the kernel's or the parents' exported variables.
605 * The initial passed environment is untouched to keep
606 * /proc/self/environ valid; it is used for tagging
607 * the init process inside containers. */
608 m
->environment
= strv_new("PATH=" DEFAULT_PATH
,
611 /* Import locale variables LC_*= from configuration */
612 locale_setup(&m
->environment
);
614 /* The user manager passes its own environment
615 * along to its children. */
616 m
->environment
= strv_copy(environ
);
621 manager_sanitize_environment(m
);
626 static int manager_setup_prefix(Manager
*m
) {
632 static const struct table_entry paths_system
[_EXEC_DIRECTORY_TYPE_MAX
] = {
633 [EXEC_DIRECTORY_RUNTIME
] = { SD_PATH_SYSTEM_RUNTIME
, NULL
},
634 [EXEC_DIRECTORY_STATE
] = { SD_PATH_SYSTEM_STATE_PRIVATE
, NULL
},
635 [EXEC_DIRECTORY_CACHE
] = { SD_PATH_SYSTEM_STATE_CACHE
, NULL
},
636 [EXEC_DIRECTORY_LOGS
] = { SD_PATH_SYSTEM_STATE_LOGS
, NULL
},
637 [EXEC_DIRECTORY_CONFIGURATION
] = { SD_PATH_SYSTEM_CONFIGURATION
, NULL
},
640 static const struct table_entry paths_user
[_EXEC_DIRECTORY_TYPE_MAX
] = {
641 [EXEC_DIRECTORY_RUNTIME
] = { SD_PATH_USER_RUNTIME
, NULL
},
642 [EXEC_DIRECTORY_STATE
] = { SD_PATH_USER_CONFIGURATION
, NULL
},
643 [EXEC_DIRECTORY_CACHE
] = { SD_PATH_USER_STATE_CACHE
, NULL
},
644 [EXEC_DIRECTORY_LOGS
] = { SD_PATH_USER_CONFIGURATION
, "log" },
645 [EXEC_DIRECTORY_CONFIGURATION
] = { SD_PATH_USER_CONFIGURATION
, NULL
},
648 const struct table_entry
*p
;
654 if (MANAGER_IS_SYSTEM(m
))
659 for (i
= 0; i
< _EXEC_DIRECTORY_TYPE_MAX
; i
++) {
660 r
= sd_path_home(p
[i
].type
, p
[i
].suffix
, &m
->prefix
[i
]);
668 static int manager_setup_run_queue(Manager
*m
) {
672 assert(!m
->run_queue_event_source
);
674 r
= sd_event_add_defer(m
->event
, &m
->run_queue_event_source
, manager_dispatch_run_queue
, m
);
678 r
= sd_event_source_set_priority(m
->run_queue_event_source
, SD_EVENT_PRIORITY_IDLE
);
682 r
= sd_event_source_set_enabled(m
->run_queue_event_source
, SD_EVENT_OFF
);
686 (void) sd_event_source_set_description(m
->run_queue_event_source
, "manager-run-queue");
691 static int manager_setup_sigchld_event_source(Manager
*m
) {
695 assert(!m
->sigchld_event_source
);
697 r
= sd_event_add_defer(m
->event
, &m
->sigchld_event_source
, manager_dispatch_sigchld
, m
);
701 r
= sd_event_source_set_priority(m
->sigchld_event_source
, SD_EVENT_PRIORITY_NORMAL
-7);
705 r
= sd_event_source_set_enabled(m
->sigchld_event_source
, SD_EVENT_OFF
);
709 (void) sd_event_source_set_description(m
->sigchld_event_source
, "manager-sigchld");
714 int manager_new(UnitFileScope scope
, unsigned test_run_flags
, Manager
**_m
) {
715 _cleanup_(manager_freep
) Manager
*m
= NULL
;
719 assert(IN_SET(scope
, UNIT_FILE_SYSTEM
, UNIT_FILE_USER
));
721 m
= new0(Manager
, 1);
725 m
->unit_file_scope
= scope
;
726 m
->exit_code
= _MANAGER_EXIT_CODE_INVALID
;
727 m
->default_timer_accuracy_usec
= USEC_PER_MINUTE
;
728 m
->default_memory_accounting
= MEMORY_ACCOUNTING_DEFAULT
;
729 m
->default_tasks_accounting
= true;
730 m
->default_tasks_max
= UINT64_MAX
;
731 m
->default_timeout_start_usec
= DEFAULT_TIMEOUT_USEC
;
732 m
->default_timeout_stop_usec
= DEFAULT_TIMEOUT_USEC
;
733 m
->default_restart_usec
= DEFAULT_RESTART_USEC
;
734 m
->original_log_level
= -1;
735 m
->original_log_target
= _LOG_TARGET_INVALID
;
738 if (MANAGER_IS_SYSTEM(m
) && detect_container() <= 0)
739 boot_timestamps(m
->timestamps
+ MANAGER_TIMESTAMP_USERSPACE
,
740 m
->timestamps
+ MANAGER_TIMESTAMP_FIRMWARE
,
741 m
->timestamps
+ MANAGER_TIMESTAMP_LOADER
);
744 /* Prepare log fields we can use for structured logging */
745 if (MANAGER_IS_SYSTEM(m
)) {
746 m
->unit_log_field
= "UNIT=";
747 m
->unit_log_format_string
= "UNIT=%s";
749 m
->invocation_log_field
= "INVOCATION_ID=";
750 m
->invocation_log_format_string
= "INVOCATION_ID=%s";
752 m
->unit_log_field
= "USER_UNIT=";
753 m
->unit_log_format_string
= "USER_UNIT=%s";
755 m
->invocation_log_field
= "USER_INVOCATION_ID=";
756 m
->invocation_log_format_string
= "USER_INVOCATION_ID=%s";
759 m
->idle_pipe
[0] = m
->idle_pipe
[1] = m
->idle_pipe
[2] = m
->idle_pipe
[3] = -1;
761 m
->pin_cgroupfs_fd
= m
->notify_fd
= m
->cgroups_agent_fd
= m
->signal_fd
= m
->time_change_fd
=
762 m
->dev_autofs_fd
= m
->private_listen_fd
= m
->cgroup_inotify_fd
=
763 m
->ask_password_inotify_fd
= -1;
765 m
->user_lookup_fds
[0] = m
->user_lookup_fds
[1] = -1;
767 m
->current_job_id
= 1; /* start as id #1, so that we can leave #0 around as "null-like" value */
769 m
->have_ask_password
= -EINVAL
; /* we don't know */
772 m
->test_run_flags
= test_run_flags
;
774 /* Reboot immediately if the user hits C-A-D more often than 7x per 2s */
775 RATELIMIT_INIT(m
->ctrl_alt_del_ratelimit
, 2 * USEC_PER_SEC
, 7);
777 r
= manager_default_environment(m
);
781 r
= hashmap_ensure_allocated(&m
->units
, &string_hash_ops
);
785 r
= hashmap_ensure_allocated(&m
->jobs
, NULL
);
789 r
= hashmap_ensure_allocated(&m
->cgroup_unit
, &path_hash_ops
);
793 r
= hashmap_ensure_allocated(&m
->watch_bus
, &string_hash_ops
);
797 r
= manager_setup_prefix(m
);
801 m
->udev
= udev_new();
805 r
= sd_event_default(&m
->event
);
809 r
= manager_setup_run_queue(m
);
813 if (test_run_flags
== MANAGER_TEST_RUN_MINIMAL
) {
814 m
->cgroup_root
= strdup("");
818 r
= manager_setup_signals(m
);
822 r
= manager_setup_cgroup(m
);
826 r
= manager_setup_time_change(m
);
830 r
= manager_read_timezone_stat(m
);
834 r
= manager_setup_timezone_change(m
);
838 r
= manager_setup_sigchld_event_source(m
);
843 if (MANAGER_IS_SYSTEM(m
) && test_run_flags
== 0) {
844 r
= mkdir_label("/run/systemd/units", 0755);
845 if (r
< 0 && r
!= -EEXIST
)
851 dir_is_empty("/usr") > 0;
853 /* Note that we do not set up the notify fd here. We do that after deserialization,
854 * since they might have gotten serialized across the reexec. */
861 static int manager_setup_notify(Manager
*m
) {
864 if (m
->test_run_flags
)
867 if (m
->notify_fd
< 0) {
868 _cleanup_close_
int fd
= -1;
869 union sockaddr_union sa
= {
870 .sa
.sa_family
= AF_UNIX
,
872 static const int one
= 1;
874 /* First free all secondary fields */
875 m
->notify_socket
= mfree(m
->notify_socket
);
876 m
->notify_event_source
= sd_event_source_unref(m
->notify_event_source
);
878 fd
= socket(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
|SOCK_NONBLOCK
, 0);
880 return log_error_errno(errno
, "Failed to allocate notification socket: %m");
882 fd_inc_rcvbuf(fd
, NOTIFY_RCVBUF_SIZE
);
884 m
->notify_socket
= strappend(m
->prefix
[EXEC_DIRECTORY_RUNTIME
], "/systemd/notify");
885 if (!m
->notify_socket
)
888 (void) mkdir_parents_label(m
->notify_socket
, 0755);
889 (void) unlink(m
->notify_socket
);
891 strncpy(sa
.un
.sun_path
, m
->notify_socket
, sizeof(sa
.un
.sun_path
)-1);
892 r
= bind(fd
, &sa
.sa
, SOCKADDR_UN_LEN(sa
.un
));
894 return log_error_errno(errno
, "bind(%s) failed: %m", sa
.un
.sun_path
);
896 r
= setsockopt(fd
, SOL_SOCKET
, SO_PASSCRED
, &one
, sizeof(one
));
898 return log_error_errno(errno
, "SO_PASSCRED failed: %m");
900 m
->notify_fd
= TAKE_FD(fd
);
902 log_debug("Using notification socket %s", m
->notify_socket
);
905 if (!m
->notify_event_source
) {
906 r
= sd_event_add_io(m
->event
, &m
->notify_event_source
, m
->notify_fd
, EPOLLIN
, manager_dispatch_notify_fd
, m
);
908 return log_error_errno(r
, "Failed to allocate notify event source: %m");
910 /* Process notification messages a bit earlier than SIGCHLD, so that we can still identify to which
911 * service an exit message belongs. */
912 r
= sd_event_source_set_priority(m
->notify_event_source
, SD_EVENT_PRIORITY_NORMAL
-8);
914 return log_error_errno(r
, "Failed to set priority of notify event source: %m");
916 (void) sd_event_source_set_description(m
->notify_event_source
, "manager-notify");
922 static int manager_setup_cgroups_agent(Manager
*m
) {
924 static const union sockaddr_union sa
= {
925 .un
.sun_family
= AF_UNIX
,
926 .un
.sun_path
= "/run/systemd/cgroups-agent",
930 /* This creates a listening socket we receive cgroups agent messages on. We do not use D-Bus for delivering
931 * these messages from the cgroups agent binary to PID 1, as the cgroups agent binary is very short-living, and
932 * each instance of it needs a new D-Bus connection. Since D-Bus connections are SOCK_STREAM/AF_UNIX, on
933 * overloaded systems the backlog of the D-Bus socket becomes relevant, as not more than the configured number
934 * of D-Bus connections may be queued until the kernel will start dropping further incoming connections,
935 * possibly resulting in lost cgroups agent messages. To avoid this, we'll use a private SOCK_DGRAM/AF_UNIX
936 * socket, where no backlog is relevant as communication may take place without an actual connect() cycle, and
937 * we thus won't lose messages.
939 * Note that PID 1 will forward the agent message to system bus, so that the user systemd instance may listen
940 * to it. The system instance hence listens on this special socket, but the user instances listen on the system
941 * bus for these messages. */
943 if (m
->test_run_flags
)
946 if (!MANAGER_IS_SYSTEM(m
))
949 r
= cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER
);
951 return log_error_errno(r
, "Failed to determine whether unified cgroups hierarchy is used: %m");
952 if (r
> 0) /* We don't need this anymore on the unified hierarchy */
955 if (m
->cgroups_agent_fd
< 0) {
956 _cleanup_close_
int fd
= -1;
958 /* First free all secondary fields */
959 m
->cgroups_agent_event_source
= sd_event_source_unref(m
->cgroups_agent_event_source
);
961 fd
= socket(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
|SOCK_NONBLOCK
, 0);
963 return log_error_errno(errno
, "Failed to allocate cgroups agent socket: %m");
965 fd_inc_rcvbuf(fd
, CGROUPS_AGENT_RCVBUF_SIZE
);
967 (void) unlink(sa
.un
.sun_path
);
969 /* Only allow root to connect to this socket */
971 r
= bind(fd
, &sa
.sa
, SOCKADDR_UN_LEN(sa
.un
));
973 return log_error_errno(errno
, "bind(%s) failed: %m", sa
.un
.sun_path
);
975 m
->cgroups_agent_fd
= fd
;
979 if (!m
->cgroups_agent_event_source
) {
980 r
= sd_event_add_io(m
->event
, &m
->cgroups_agent_event_source
, m
->cgroups_agent_fd
, EPOLLIN
, manager_dispatch_cgroups_agent_fd
, m
);
982 return log_error_errno(r
, "Failed to allocate cgroups agent event source: %m");
984 /* Process cgroups notifications early, but after having processed service notification messages or
985 * SIGCHLD signals, so that a cgroup running empty is always just the last safety net of notification,
986 * and we collected the metadata the notification and SIGCHLD stuff offers first. Also see handling of
987 * cgroup inotify for the unified cgroup stuff. */
988 r
= sd_event_source_set_priority(m
->cgroups_agent_event_source
, SD_EVENT_PRIORITY_NORMAL
-4);
990 return log_error_errno(r
, "Failed to set priority of cgroups agent event source: %m");
992 (void) sd_event_source_set_description(m
->cgroups_agent_event_source
, "manager-cgroups-agent");
998 static int manager_setup_user_lookup_fd(Manager
*m
) {
1003 /* Set up the socket pair used for passing UID/GID resolution results from forked off processes to PID
1004 * 1. Background: we can't do name lookups (NSS) from PID 1, since it might involve IPC and thus activation,
1005 * and we might hence deadlock on ourselves. Hence we do all user/group lookups asynchronously from the forked
1006 * off processes right before executing the binaries to start. In order to be able to clean up any IPC objects
1007 * created by a unit (see RemoveIPC=) we need to know in PID 1 the used UID/GID of the executed processes,
1008 * hence we establish this communication channel so that forked off processes can pass their UID/GID
1009 * information back to PID 1. The forked off processes send their resolved UID/GID to PID 1 in a simple
1010 * datagram, along with their unit name, so that we can share one communication socket pair among all units for
1013 * You might wonder why we need a communication channel for this that is independent of the usual notification
1014 * socket scheme (i.e. $NOTIFY_SOCKET). The primary difference is about trust: data sent via the $NOTIFY_SOCKET
1015 * channel is only accepted if it originates from the right unit and if reception was enabled for it. The user
1016 * lookup socket OTOH is only accessible by PID 1 and its children until they exec(), and always available.
1018 * Note that this function is called under two circumstances: when we first initialize (in which case we
1019 * allocate both the socket pair and the event source to listen on it), and when we deserialize after a reload
1020 * (in which case the socket pair already exists but we still need to allocate the event source for it). */
1022 if (m
->user_lookup_fds
[0] < 0) {
1024 /* Free all secondary fields */
1025 safe_close_pair(m
->user_lookup_fds
);
1026 m
->user_lookup_event_source
= sd_event_source_unref(m
->user_lookup_event_source
);
1028 if (socketpair(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
, 0, m
->user_lookup_fds
) < 0)
1029 return log_error_errno(errno
, "Failed to allocate user lookup socket: %m");
1031 (void) fd_inc_rcvbuf(m
->user_lookup_fds
[0], NOTIFY_RCVBUF_SIZE
);
1034 if (!m
->user_lookup_event_source
) {
1035 r
= sd_event_add_io(m
->event
, &m
->user_lookup_event_source
, m
->user_lookup_fds
[0], EPOLLIN
, manager_dispatch_user_lookup_fd
, m
);
1037 return log_error_errno(errno
, "Failed to allocate user lookup event source: %m");
1039 /* Process even earlier than the notify event source, so that we always know first about valid UID/GID
1041 r
= sd_event_source_set_priority(m
->user_lookup_event_source
, SD_EVENT_PRIORITY_NORMAL
-11);
1043 return log_error_errno(errno
, "Failed to set priority ot user lookup event source: %m");
1045 (void) sd_event_source_set_description(m
->user_lookup_event_source
, "user-lookup");
1051 static unsigned manager_dispatch_cleanup_queue(Manager
*m
) {
1057 while ((u
= m
->cleanup_queue
)) {
1058 assert(u
->in_cleanup_queue
);
1068 GC_OFFSET_IN_PATH
, /* This one is on the path we were traveling */
1069 GC_OFFSET_UNSURE
, /* No clue */
1070 GC_OFFSET_GOOD
, /* We still need this unit */
1071 GC_OFFSET_BAD
, /* We don't need this unit anymore */
1075 static void unit_gc_mark_good(Unit
*u
, unsigned gc_marker
) {
1080 u
->gc_marker
= gc_marker
+ GC_OFFSET_GOOD
;
1082 /* Recursively mark referenced units as GOOD as well */
1083 HASHMAP_FOREACH_KEY(v
, other
, u
->dependencies
[UNIT_REFERENCES
], i
)
1084 if (other
->gc_marker
== gc_marker
+ GC_OFFSET_UNSURE
)
1085 unit_gc_mark_good(other
, gc_marker
);
1088 static void unit_gc_sweep(Unit
*u
, unsigned gc_marker
) {
1096 if (IN_SET(u
->gc_marker
- gc_marker
,
1097 GC_OFFSET_GOOD
, GC_OFFSET_BAD
, GC_OFFSET_UNSURE
, GC_OFFSET_IN_PATH
))
1100 if (u
->in_cleanup_queue
)
1103 if (!unit_may_gc(u
))
1106 u
->gc_marker
= gc_marker
+ GC_OFFSET_IN_PATH
;
1110 HASHMAP_FOREACH_KEY(v
, other
, u
->dependencies
[UNIT_REFERENCED_BY
], i
) {
1111 unit_gc_sweep(other
, gc_marker
);
1113 if (other
->gc_marker
== gc_marker
+ GC_OFFSET_GOOD
)
1116 if (other
->gc_marker
!= gc_marker
+ GC_OFFSET_BAD
)
1120 if (u
->refs_by_target
) {
1123 LIST_FOREACH(refs_by_target
, ref
, u
->refs_by_target
) {
1124 unit_gc_sweep(ref
->source
, gc_marker
);
1126 if (ref
->source
->gc_marker
== gc_marker
+ GC_OFFSET_GOOD
)
1129 if (ref
->source
->gc_marker
!= gc_marker
+ GC_OFFSET_BAD
)
1137 /* We were unable to find anything out about this entry, so
1138 * let's investigate it later */
1139 u
->gc_marker
= gc_marker
+ GC_OFFSET_UNSURE
;
1140 unit_add_to_gc_queue(u
);
1144 /* We definitely know that this one is not useful anymore, so
1145 * let's mark it for deletion */
1146 u
->gc_marker
= gc_marker
+ GC_OFFSET_BAD
;
1147 unit_add_to_cleanup_queue(u
);
1151 unit_gc_mark_good(u
, gc_marker
);
1154 static unsigned manager_dispatch_gc_unit_queue(Manager
*m
) {
1155 unsigned n
= 0, gc_marker
;
1160 /* log_debug("Running GC..."); */
1162 m
->gc_marker
+= _GC_OFFSET_MAX
;
1163 if (m
->gc_marker
+ _GC_OFFSET_MAX
<= _GC_OFFSET_MAX
)
1166 gc_marker
= m
->gc_marker
;
1168 while ((u
= m
->gc_unit_queue
)) {
1169 assert(u
->in_gc_queue
);
1171 unit_gc_sweep(u
, gc_marker
);
1173 LIST_REMOVE(gc_queue
, m
->gc_unit_queue
, u
);
1174 u
->in_gc_queue
= false;
1178 if (IN_SET(u
->gc_marker
- gc_marker
,
1179 GC_OFFSET_BAD
, GC_OFFSET_UNSURE
)) {
1181 log_unit_debug(u
, "Collecting.");
1182 u
->gc_marker
= gc_marker
+ GC_OFFSET_BAD
;
1183 unit_add_to_cleanup_queue(u
);
1190 static unsigned manager_dispatch_gc_job_queue(Manager
*m
) {
1196 while ((j
= m
->gc_job_queue
)) {
1197 assert(j
->in_gc_queue
);
1199 LIST_REMOVE(gc_queue
, m
->gc_job_queue
, j
);
1200 j
->in_gc_queue
= false;
1207 log_unit_debug(j
->unit
, "Collecting job.");
1208 (void) job_finish_and_invalidate(j
, JOB_COLLECTED
, false, false);
1214 static void manager_clear_jobs_and_units(Manager
*m
) {
1219 while ((u
= hashmap_first(m
->units
)))
1222 manager_dispatch_cleanup_queue(m
);
1224 assert(!m
->load_queue
);
1225 assert(!m
->run_queue
);
1226 assert(!m
->dbus_unit_queue
);
1227 assert(!m
->dbus_job_queue
);
1228 assert(!m
->cleanup_queue
);
1229 assert(!m
->gc_unit_queue
);
1230 assert(!m
->gc_job_queue
);
1232 assert(hashmap_isempty(m
->jobs
));
1233 assert(hashmap_isempty(m
->units
));
1235 m
->n_on_console
= 0;
1236 m
->n_running_jobs
= 0;
1237 m
->n_installed_jobs
= 0;
1238 m
->n_failed_jobs
= 0;
1241 Manager
* manager_free(Manager
*m
) {
1243 ExecDirectoryType dt
;
1248 manager_clear_jobs_and_units(m
);
1250 for (c
= 0; c
< _UNIT_TYPE_MAX
; c
++)
1251 if (unit_vtable
[c
]->shutdown
)
1252 unit_vtable
[c
]->shutdown(m
);
1254 /* If we reexecute ourselves, we keep the root cgroup around */
1255 manager_shutdown_cgroup(m
, m
->exit_code
!= MANAGER_REEXECUTE
);
1257 lookup_paths_flush_generator(&m
->lookup_paths
);
1261 exec_runtime_vacuum(m
);
1262 hashmap_free(m
->exec_runtime_by_id
);
1264 dynamic_user_vacuum(m
, false);
1265 hashmap_free(m
->dynamic_users
);
1267 hashmap_free(m
->units
);
1268 hashmap_free(m
->units_by_invocation_id
);
1269 hashmap_free(m
->jobs
);
1270 hashmap_free(m
->watch_pids
);
1271 hashmap_free(m
->watch_bus
);
1273 set_free(m
->startup_units
);
1274 set_free(m
->failed_units
);
1276 sd_event_source_unref(m
->signal_event_source
);
1277 sd_event_source_unref(m
->sigchld_event_source
);
1278 sd_event_source_unref(m
->notify_event_source
);
1279 sd_event_source_unref(m
->cgroups_agent_event_source
);
1280 sd_event_source_unref(m
->time_change_event_source
);
1281 sd_event_source_unref(m
->timezone_change_event_source
);
1282 sd_event_source_unref(m
->jobs_in_progress_event_source
);
1283 sd_event_source_unref(m
->run_queue_event_source
);
1284 sd_event_source_unref(m
->user_lookup_event_source
);
1285 sd_event_source_unref(m
->sync_bus_names_event_source
);
1287 safe_close(m
->signal_fd
);
1288 safe_close(m
->notify_fd
);
1289 safe_close(m
->cgroups_agent_fd
);
1290 safe_close(m
->time_change_fd
);
1291 safe_close_pair(m
->user_lookup_fds
);
1293 manager_close_ask_password(m
);
1295 manager_close_idle_pipe(m
);
1297 udev_unref(m
->udev
);
1298 sd_event_unref(m
->event
);
1300 free(m
->notify_socket
);
1302 lookup_paths_free(&m
->lookup_paths
);
1303 strv_free(m
->environment
);
1305 hashmap_free(m
->cgroup_unit
);
1306 set_free_free(m
->unit_path_cache
);
1308 free(m
->switch_root
);
1309 free(m
->switch_root_init
);
1311 rlimit_free_all(m
->rlimit
);
1313 assert(hashmap_isempty(m
->units_requiring_mounts_for
));
1314 hashmap_free(m
->units_requiring_mounts_for
);
1316 hashmap_free(m
->uid_refs
);
1317 hashmap_free(m
->gid_refs
);
1319 for (dt
= 0; dt
< _EXEC_DIRECTORY_TYPE_MAX
; dt
++)
1320 m
->prefix
[dt
] = mfree(m
->prefix
[dt
]);
1325 static void manager_enumerate_perpetual(Manager
*m
) {
1330 /* Let's ask every type to load all units from disk/kernel that it might know */
1331 for (c
= 0; c
< _UNIT_TYPE_MAX
; c
++) {
1332 if (!unit_type_supported(c
)) {
1333 log_debug("Unit type .%s is not supported on this system.", unit_type_to_string(c
));
1337 if (unit_vtable
[c
]->enumerate_perpetual
)
1338 unit_vtable
[c
]->enumerate_perpetual(m
);
1342 static void manager_enumerate(Manager
*m
) {
1347 /* Let's ask every type to load all units from disk/kernel that it might know */
1348 for (c
= 0; c
< _UNIT_TYPE_MAX
; c
++) {
1349 if (!unit_type_supported(c
)) {
1350 log_debug("Unit type .%s is not supported on this system.", unit_type_to_string(c
));
1354 if (unit_vtable
[c
]->enumerate
)
1355 unit_vtable
[c
]->enumerate(m
);
1358 manager_dispatch_load_queue(m
);
1361 static void manager_coldplug(Manager
*m
) {
1369 log_debug("Invoking unit coldplug() handlers…");
1371 /* Let's place the units back into their deserialized state */
1372 HASHMAP_FOREACH_KEY(u
, k
, m
->units
, i
) {
1374 /* ignore aliases */
1378 r
= unit_coldplug(u
);
1380 log_warning_errno(r
, "We couldn't coldplug %s, proceeding anyway: %m", u
->id
);
1384 static void manager_catchup(Manager
*m
) {
1391 log_debug("Invoking unit catchup() handlers…");
1393 /* Let's catch up on any state changes that happened while we were reloading/reexecing */
1394 HASHMAP_FOREACH_KEY(u
, k
, m
->units
, i
) {
1396 /* ignore aliases */
1404 static void manager_build_unit_path_cache(Manager
*m
) {
1410 set_free_free(m
->unit_path_cache
);
1412 m
->unit_path_cache
= set_new(&path_hash_ops
);
1413 if (!m
->unit_path_cache
) {
1418 /* This simply builds a list of files we know exist, so that
1419 * we don't always have to go to disk */
1421 STRV_FOREACH(i
, m
->lookup_paths
.search_path
) {
1422 _cleanup_closedir_
DIR *d
= NULL
;
1427 if (errno
!= ENOENT
)
1428 log_warning_errno(errno
, "Failed to open directory %s, ignoring: %m", *i
);
1432 FOREACH_DIRENT(de
, d
, r
= -errno
; goto fail
) {
1435 p
= strjoin(streq(*i
, "/") ? "" : *i
, "/", de
->d_name
);
1441 r
= set_consume(m
->unit_path_cache
, p
);
1450 log_warning_errno(r
, "Failed to build unit path cache, proceeding without: %m");
1451 m
->unit_path_cache
= set_free_free(m
->unit_path_cache
);
1454 static void manager_distribute_fds(Manager
*m
, FDSet
*fds
) {
1460 HASHMAP_FOREACH(u
, m
->units
, i
) {
1462 if (fdset_size(fds
) <= 0)
1465 if (!UNIT_VTABLE(u
)->distribute_fds
)
1468 UNIT_VTABLE(u
)->distribute_fds(u
, fds
);
1472 static bool manager_dbus_is_running(Manager
*m
, bool deserialized
) {
1477 /* This checks whether the dbus instance we are supposed to expose our APIs on is up. We check both the socket
1478 * and the service unit. If the 'deserialized' parameter is true we'll check the deserialized state of the unit
1479 * rather than the current one. */
1481 if (m
->test_run_flags
!= 0)
1484 /* If we are in the user instance, and the env var is already set for us, then this means D-Bus is ran
1485 * somewhere outside of our own logic. Let's use it */
1486 if (MANAGER_IS_USER(m
) && getenv("DBUS_SESSION_BUS_ADDRESS"))
1489 u
= manager_get_unit(m
, SPECIAL_DBUS_SOCKET
);
1492 if ((deserialized
? SOCKET(u
)->deserialized_state
: SOCKET(u
)->state
) != SOCKET_RUNNING
)
1495 u
= manager_get_unit(m
, SPECIAL_DBUS_SERVICE
);
1498 if (!IN_SET((deserialized
? SERVICE(u
)->deserialized_state
: SERVICE(u
)->state
), SERVICE_RUNNING
, SERVICE_RELOAD
))
1504 static void manager_setup_bus(Manager
*m
) {
1507 /* Let's set up our private bus connection now, unconditionally */
1508 (void) bus_init_private(m
);
1510 /* If we are in --user mode also connect to the system bus now */
1511 if (MANAGER_IS_USER(m
))
1512 (void) bus_init_system(m
);
1514 /* Let's connect to the bus now, but only if the unit is supposed to be up */
1515 if (manager_dbus_is_running(m
, MANAGER_IS_RELOADING(m
))) {
1516 (void) bus_init_api(m
);
1518 if (MANAGER_IS_SYSTEM(m
))
1519 (void) bus_init_system(m
);
1523 static void manager_preset_all(Manager
*m
) {
1528 if (m
->first_boot
<= 0)
1531 if (!MANAGER_IS_SYSTEM(m
))
1534 if (m
->test_run_flags
!= 0)
1537 /* If this is the first boot, and we are in the host system, then preset everything */
1538 r
= unit_file_preset_all(UNIT_FILE_SYSTEM
, 0, NULL
, UNIT_FILE_PRESET_ENABLE_ONLY
, NULL
, 0);
1540 log_full_errno(r
== -EEXIST
? LOG_NOTICE
: LOG_WARNING
, r
,
1541 "Failed to populate /etc with preset unit settings, ignoring: %m");
1543 log_info("Populated /etc with preset unit settings.");
1546 int manager_startup(Manager
*m
, FILE *serialization
, FDSet
*fds
) {
1551 /* If we are running in test mode, we still want to run the generators,
1552 * but we should not touch the real generator directories. */
1553 r
= lookup_paths_init(&m
->lookup_paths
, m
->unit_file_scope
,
1554 m
->test_run_flags
? LOOKUP_PATHS_TEMPORARY_GENERATED
: 0,
1559 r
= manager_run_environment_generators(m
);
1563 dual_timestamp_get(m
->timestamps
+ MANAGER_TIMESTAMP_GENERATORS_START
);
1564 r
= manager_run_generators(m
);
1565 dual_timestamp_get(m
->timestamps
+ MANAGER_TIMESTAMP_GENERATORS_FINISH
);
1569 manager_preset_all(m
);
1570 lookup_paths_reduce(&m
->lookup_paths
);
1571 manager_build_unit_path_cache(m
);
1573 /* If we will deserialize make sure that during enumeration
1574 * this is already known, so we increase the counter here
1579 /* First, enumerate what we can from all config files */
1580 dual_timestamp_get(m
->timestamps
+ MANAGER_TIMESTAMP_UNITS_LOAD_START
);
1581 manager_enumerate_perpetual(m
);
1582 manager_enumerate(m
);
1583 dual_timestamp_get(m
->timestamps
+ MANAGER_TIMESTAMP_UNITS_LOAD_FINISH
);
1585 /* Second, deserialize if there is something to deserialize */
1586 if (serialization
) {
1587 r
= manager_deserialize(m
, serialization
, fds
);
1589 return log_error_errno(r
, "Deserialization failed: %m");
1592 /* Any fds left? Find some unit which wants them. This is
1593 * useful to allow container managers to pass some file
1594 * descriptors to us pre-initialized. This enables
1595 * socket-based activation of entire containers. */
1596 manager_distribute_fds(m
, fds
);
1598 /* We might have deserialized the notify fd, but if we didn't
1599 * then let's create the bus now */
1600 r
= manager_setup_notify(m
);
1602 /* No sense to continue without notifications, our children would fail anyway. */
1605 r
= manager_setup_cgroups_agent(m
);
1607 /* Likewise, no sense to continue without empty cgroup notifications. */
1610 r
= manager_setup_user_lookup_fd(m
);
1612 /* This shouldn't fail, except if things are really broken. */
1615 /* Connect to the bus if we are good for it */
1616 manager_setup_bus(m
);
1618 /* Now that we are connected to all possible busses, let's deserialize who is tracking us. */
1619 (void) bus_track_coldplug(m
, &m
->subscribed
, false, m
->deserialized_subscribed
);
1620 m
->deserialized_subscribed
= strv_free(m
->deserialized_subscribed
);
1622 /* Third, fire things up! */
1623 manager_coldplug(m
);
1625 /* Release any dynamic users no longer referenced */
1626 dynamic_user_vacuum(m
, true);
1628 exec_runtime_vacuum(m
);
1630 /* Release any references to UIDs/GIDs no longer referenced, and destroy any IPC owned by them */
1631 manager_vacuum_uid_refs(m
);
1632 manager_vacuum_gid_refs(m
);
1634 if (serialization
) {
1635 assert(m
->n_reloading
> 0);
1638 /* Let's wait for the UnitNew/JobNew messages being
1639 * sent, before we notify that the reload is
1641 m
->send_reloading_done
= true;
1644 /* Let's finally catch up with any changes that took place while we were reloading/reexecing */
1650 int manager_add_job(Manager
*m
, JobType type
, Unit
*unit
, JobMode mode
, sd_bus_error
*e
, Job
**_ret
) {
1655 assert(type
< _JOB_TYPE_MAX
);
1657 assert(mode
< _JOB_MODE_MAX
);
1659 if (mode
== JOB_ISOLATE
&& type
!= JOB_START
)
1660 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
, "Isolate is only valid for start.");
1662 if (mode
== JOB_ISOLATE
&& !unit
->allow_isolate
)
1663 return sd_bus_error_setf(e
, BUS_ERROR_NO_ISOLATION
, "Operation refused, unit may not be isolated.");
1665 log_unit_debug(unit
, "Trying to enqueue job %s/%s/%s", unit
->id
, job_type_to_string(type
), job_mode_to_string(mode
));
1667 type
= job_type_collapse(type
, unit
);
1669 tr
= transaction_new(mode
== JOB_REPLACE_IRREVERSIBLY
);
1673 r
= transaction_add_job_and_dependencies(tr
, type
, unit
, NULL
, true, false,
1674 IN_SET(mode
, JOB_IGNORE_DEPENDENCIES
, JOB_IGNORE_REQUIREMENTS
),
1675 mode
== JOB_IGNORE_DEPENDENCIES
, e
);
1679 if (mode
== JOB_ISOLATE
) {
1680 r
= transaction_add_isolate_jobs(tr
, m
);
1685 r
= transaction_activate(tr
, m
, mode
, e
);
1689 log_unit_debug(unit
,
1690 "Enqueued job %s/%s as %u", unit
->id
,
1691 job_type_to_string(type
), (unsigned) tr
->anchor_job
->id
);
1694 *_ret
= tr
->anchor_job
;
1696 transaction_free(tr
);
1700 transaction_abort(tr
);
1701 transaction_free(tr
);
1705 int manager_add_job_by_name(Manager
*m
, JobType type
, const char *name
, JobMode mode
, sd_bus_error
*e
, Job
**ret
) {
1706 Unit
*unit
= NULL
; /* just to appease gcc, initialization is not really necessary */
1710 assert(type
< _JOB_TYPE_MAX
);
1712 assert(mode
< _JOB_MODE_MAX
);
1714 r
= manager_load_unit(m
, name
, NULL
, NULL
, &unit
);
1719 return manager_add_job(m
, type
, unit
, mode
, e
, ret
);
1722 int manager_add_job_by_name_and_warn(Manager
*m
, JobType type
, const char *name
, JobMode mode
, Job
**ret
) {
1723 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
1727 assert(type
< _JOB_TYPE_MAX
);
1729 assert(mode
< _JOB_MODE_MAX
);
1731 r
= manager_add_job_by_name(m
, type
, name
, mode
, &error
, ret
);
1733 return log_warning_errno(r
, "Failed to enqueue %s job for %s: %s", job_mode_to_string(mode
), name
, bus_error_message(&error
, r
));
1738 int manager_propagate_reload(Manager
*m
, Unit
*unit
, JobMode mode
, sd_bus_error
*e
) {
1744 assert(mode
< _JOB_MODE_MAX
);
1745 assert(mode
!= JOB_ISOLATE
); /* Isolate is only valid for start */
1747 tr
= transaction_new(mode
== JOB_REPLACE_IRREVERSIBLY
);
1751 /* We need an anchor job */
1752 r
= transaction_add_job_and_dependencies(tr
, JOB_NOP
, unit
, NULL
, false, false, true, true, e
);
1756 /* Failure in adding individual dependencies is ignored, so this always succeeds. */
1757 transaction_add_propagate_reload_jobs(tr
, unit
, tr
->anchor_job
, mode
== JOB_IGNORE_DEPENDENCIES
, e
);
1759 r
= transaction_activate(tr
, m
, mode
, e
);
1763 transaction_free(tr
);
1767 transaction_abort(tr
);
1768 transaction_free(tr
);
1772 Job
*manager_get_job(Manager
*m
, uint32_t id
) {
1775 return hashmap_get(m
->jobs
, UINT32_TO_PTR(id
));
1778 Unit
*manager_get_unit(Manager
*m
, const char *name
) {
1782 return hashmap_get(m
->units
, name
);
1785 static int manager_dispatch_target_deps_queue(Manager
*m
) {
1790 static const UnitDependency deps
[] = {
1799 while ((u
= m
->target_deps_queue
)) {
1800 assert(u
->in_target_deps_queue
);
1802 LIST_REMOVE(target_deps_queue
, u
->manager
->target_deps_queue
, u
);
1803 u
->in_target_deps_queue
= false;
1805 for (k
= 0; k
< ELEMENTSOF(deps
); k
++) {
1810 HASHMAP_FOREACH_KEY(v
, target
, u
->dependencies
[deps
[k
]], i
) {
1811 r
= unit_add_default_target_dependency(u
, target
);
1821 unsigned manager_dispatch_load_queue(Manager
*m
) {
1827 /* Make sure we are not run recursively */
1828 if (m
->dispatching_load_queue
)
1831 m
->dispatching_load_queue
= true;
1833 /* Dispatches the load queue. Takes a unit from the queue and
1834 * tries to load its data until the queue is empty */
1836 while ((u
= m
->load_queue
)) {
1837 assert(u
->in_load_queue
);
1843 m
->dispatching_load_queue
= false;
1845 /* Dispatch the units waiting for their target dependencies to be added now, as all targets that we know about
1846 * should be loaded and have aliases resolved */
1847 (void) manager_dispatch_target_deps_queue(m
);
1852 int manager_load_unit_prepare(
1859 _cleanup_(unit_freep
) Unit
*cleanup_ret
= NULL
;
1865 assert(name
|| path
);
1868 /* This will prepare the unit for loading, but not actually
1869 * load anything from disk. */
1871 if (path
&& !is_path(path
))
1872 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
, "Path %s is not absolute.", path
);
1875 name
= basename(path
);
1877 t
= unit_name_to_type(name
);
1879 if (t
== _UNIT_TYPE_INVALID
|| !unit_name_is_valid(name
, UNIT_NAME_PLAIN
|UNIT_NAME_INSTANCE
)) {
1880 if (unit_name_is_valid(name
, UNIT_NAME_TEMPLATE
))
1881 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
, "Unit name %s is missing the instance name.", name
);
1883 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
, "Unit name %s is not valid.", name
);
1886 ret
= manager_get_unit(m
, name
);
1892 ret
= cleanup_ret
= unit_new(m
, unit_vtable
[t
]->object_size
);
1897 ret
->fragment_path
= strdup(path
);
1898 if (!ret
->fragment_path
)
1902 r
= unit_add_name(ret
, name
);
1906 unit_add_to_load_queue(ret
);
1907 unit_add_to_dbus_queue(ret
);
1908 unit_add_to_gc_queue(ret
);
1916 int manager_load_unit(
1928 /* This will load the service information files, but not actually
1929 * start any services or anything. */
1931 r
= manager_load_unit_prepare(m
, name
, path
, e
, _ret
);
1935 manager_dispatch_load_queue(m
);
1937 *_ret
= unit_follow_merge(*_ret
);
1941 int manager_load_startable_unit_or_warn(
1947 /* Load a unit, make sure it loaded fully and is not masked. */
1949 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
1953 r
= manager_load_unit(m
, name
, path
, &error
, &unit
);
1955 return log_error_errno(r
, "Failed to load %s %s: %s",
1956 name
? "unit" : "unit file", name
?: path
,
1957 bus_error_message(&error
, r
));
1959 r
= bus_unit_validate_load_state(unit
, &error
);
1961 return log_error_errno(r
, "%s", bus_error_message(&error
, r
));
1967 void manager_dump_jobs(Manager
*s
, FILE *f
, const char *prefix
) {
1974 HASHMAP_FOREACH(j
, s
->jobs
, i
)
1975 job_dump(j
, f
, prefix
);
1978 void manager_dump_units(Manager
*s
, FILE *f
, const char *prefix
) {
1986 HASHMAP_FOREACH_KEY(u
, t
, s
->units
, i
)
1988 unit_dump(u
, f
, prefix
);
1991 void manager_dump(Manager
*m
, FILE *f
, const char *prefix
) {
1997 for (q
= 0; q
< _MANAGER_TIMESTAMP_MAX
; q
++) {
1998 char buf
[FORMAT_TIMESTAMP_MAX
];
2000 if (dual_timestamp_is_set(m
->timestamps
+ q
))
2001 fprintf(f
, "%sTimestamp %s: %s\n",
2003 manager_timestamp_to_string(q
),
2004 format_timestamp(buf
, sizeof(buf
), m
->timestamps
[q
].realtime
));
2007 manager_dump_units(m
, f
, prefix
);
2008 manager_dump_jobs(m
, f
, prefix
);
2011 int manager_get_dump_string(Manager
*m
, char **ret
) {
2012 _cleanup_free_
char *dump
= NULL
;
2013 _cleanup_fclose_
FILE *f
= NULL
;
2020 f
= open_memstream(&dump
, &size
);
2024 (void) __fsetlocking(f
, FSETLOCKING_BYCALLER
);
2026 manager_dump(m
, f
, NULL
);
2028 r
= fflush_and_check(f
);
2034 *ret
= TAKE_PTR(dump
);
2039 void manager_clear_jobs(Manager
*m
) {
2044 while ((j
= hashmap_first(m
->jobs
)))
2045 /* No need to recurse. We're cancelling all jobs. */
2046 job_finish_and_invalidate(j
, JOB_CANCELED
, false, false);
2049 static int manager_dispatch_run_queue(sd_event_source
*source
, void *userdata
) {
2050 Manager
*m
= userdata
;
2056 while ((j
= m
->run_queue
)) {
2057 assert(j
->installed
);
2058 assert(j
->in_run_queue
);
2060 job_run_and_invalidate(j
);
2063 if (m
->n_running_jobs
> 0)
2064 manager_watch_jobs_in_progress(m
);
2066 if (m
->n_on_console
> 0)
2067 manager_watch_idle_pipe(m
);
2072 static unsigned manager_dispatch_dbus_queue(Manager
*m
) {
2073 unsigned n
= 0, budget
;
2079 if (m
->dispatching_dbus_queue
)
2082 /* Anything to do at all? */
2083 if (!m
->dbus_unit_queue
&& !m
->dbus_job_queue
&& !m
->send_reloading_done
&& !m
->queued_message
)
2086 /* Do we have overly many messages queued at the moment? If so, let's not enqueue more on top, let's sit this
2087 * cycle out, and process things in a later cycle when the queues got a bit emptier. */
2088 if (manager_bus_n_queued_write(m
) > MANAGER_BUS_BUSY_THRESHOLD
)
2091 /* Only process a certain number of units/jobs per event loop iteration. Even if the bus queue wasn't overly
2092 * full before this call we shouldn't increase it in size too wildly in one step, and we shouldn't monopolize
2093 * CPU time with generating these messages. Note the difference in counting of this "budget" and the
2094 * "threshold" above: the "budget" is decreased only once per generated message, regardless how many
2095 * busses/direct connections it is enqueued on, while the "threshold" is applied to each queued instance of bus
2096 * message, i.e. if the same message is enqueued to five busses/direct connections it will be counted five
2097 * times. This difference in counting ("references" vs. "instances") is primarily a result of the fact that
2098 * it's easier to implement it this way, however it also reflects the thinking that the "threshold" should put
2099 * a limit on used queue memory, i.e. space, while the "budget" should put a limit on time. Also note that
2100 * the "threshold" is currently chosen much higher than the "budget". */
2101 budget
= MANAGER_BUS_MESSAGE_BUDGET
;
2103 m
->dispatching_dbus_queue
= true;
2105 while (budget
> 0 && (u
= m
->dbus_unit_queue
)) {
2107 assert(u
->in_dbus_queue
);
2109 bus_unit_send_change_signal(u
);
2113 while (budget
> 0 && (j
= m
->dbus_job_queue
)) {
2114 assert(j
->in_dbus_queue
);
2116 bus_job_send_change_signal(j
);
2120 m
->dispatching_dbus_queue
= false;
2122 if (budget
> 0 && m
->send_reloading_done
) {
2123 m
->send_reloading_done
= false;
2124 bus_manager_send_reloading(m
, false);
2128 if (budget
> 0 && m
->queued_message
) {
2129 bus_send_queued_message(m
);
2136 static int manager_dispatch_cgroups_agent_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
2137 Manager
*m
= userdata
;
2138 char buf
[PATH_MAX
+1];
2141 n
= recv(fd
, buf
, sizeof(buf
), 0);
2143 return log_error_errno(errno
, "Failed to read cgroups agent message: %m");
2145 log_error("Got zero-length cgroups agent message, ignoring.");
2148 if ((size_t) n
>= sizeof(buf
)) {
2149 log_error("Got overly long cgroups agent message, ignoring.");
2153 if (memchr(buf
, 0, n
)) {
2154 log_error("Got cgroups agent message with embedded NUL byte, ignoring.");
2159 manager_notify_cgroup_empty(m
, buf
);
2160 (void) bus_forward_agent_released(m
, buf
);
2165 static void manager_invoke_notify_message(
2168 const struct ucred
*ucred
,
2177 if (u
->notifygen
== m
->notifygen
) /* Already invoked on this same unit in this same iteration? */
2179 u
->notifygen
= m
->notifygen
;
2181 if (UNIT_VTABLE(u
)->notify_message
) {
2182 _cleanup_strv_free_
char **tags
= NULL
;
2184 tags
= strv_split(buf
, NEWLINE
);
2190 UNIT_VTABLE(u
)->notify_message(u
, ucred
, tags
, fds
);
2192 } else if (DEBUG_LOGGING
) {
2193 _cleanup_free_
char *x
= NULL
, *y
= NULL
;
2195 x
= ellipsize(buf
, 20, 90);
2199 log_unit_debug(u
, "Got notification message \"%s\", ignoring.", strnull(y
));
2203 static int manager_dispatch_notify_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
2205 _cleanup_fdset_free_ FDSet
*fds
= NULL
;
2206 Manager
*m
= userdata
;
2207 char buf
[NOTIFY_BUFFER_MAX
+1];
2208 struct iovec iovec
= {
2210 .iov_len
= sizeof(buf
)-1,
2213 struct cmsghdr cmsghdr
;
2214 uint8_t buf
[CMSG_SPACE(sizeof(struct ucred
)) +
2215 CMSG_SPACE(sizeof(int) * NOTIFY_FD_MAX
)];
2217 struct msghdr msghdr
= {
2220 .msg_control
= &control
,
2221 .msg_controllen
= sizeof(control
),
2224 struct cmsghdr
*cmsg
;
2225 struct ucred
*ucred
= NULL
;
2226 _cleanup_free_ Unit
**array_copy
= NULL
;
2227 Unit
*u1
, *u2
, **array
;
2228 int r
, *fd_array
= NULL
;
2234 assert(m
->notify_fd
== fd
);
2236 if (revents
!= EPOLLIN
) {
2237 log_warning("Got unexpected poll event for notify fd.");
2241 n
= recvmsg(m
->notify_fd
, &msghdr
, MSG_DONTWAIT
|MSG_CMSG_CLOEXEC
|MSG_TRUNC
);
2243 if (IN_SET(errno
, EAGAIN
, EINTR
))
2244 return 0; /* Spurious wakeup, try again */
2246 /* If this is any other, real error, then let's stop processing this socket. This of course means we
2247 * won't take notification messages anymore, but that's still better than busy looping around this:
2248 * being woken up over and over again but being unable to actually read the message off the socket. */
2249 return log_error_errno(errno
, "Failed to receive notification message: %m");
2252 CMSG_FOREACH(cmsg
, &msghdr
) {
2253 if (cmsg
->cmsg_level
== SOL_SOCKET
&& cmsg
->cmsg_type
== SCM_RIGHTS
) {
2255 fd_array
= (int*) CMSG_DATA(cmsg
);
2256 n_fds
= (cmsg
->cmsg_len
- CMSG_LEN(0)) / sizeof(int);
2258 } else if (cmsg
->cmsg_level
== SOL_SOCKET
&&
2259 cmsg
->cmsg_type
== SCM_CREDENTIALS
&&
2260 cmsg
->cmsg_len
== CMSG_LEN(sizeof(struct ucred
))) {
2262 ucred
= (struct ucred
*) CMSG_DATA(cmsg
);
2269 r
= fdset_new_array(&fds
, fd_array
, n_fds
);
2271 close_many(fd_array
, n_fds
);
2277 if (!ucred
|| !pid_is_valid(ucred
->pid
)) {
2278 log_warning("Received notify message without valid credentials. Ignoring.");
2282 if ((size_t) n
>= sizeof(buf
) || (msghdr
.msg_flags
& MSG_TRUNC
)) {
2283 log_warning("Received notify message exceeded maximum size. Ignoring.");
2287 /* As extra safety check, let's make sure the string we get doesn't contain embedded NUL bytes. We permit one
2288 * trailing NUL byte in the message, but don't expect it. */
2289 if (n
> 1 && memchr(buf
, 0, n
-1)) {
2290 log_warning("Received notify message with embedded NUL bytes. Ignoring.");
2294 /* Make sure it's NUL-terminated. */
2297 /* Increase the generation counter used for filtering out duplicate unit invocations. */
2300 /* Notify every unit that might be interested, which might be multiple. */
2301 u1
= manager_get_unit_by_pid_cgroup(m
, ucred
->pid
);
2302 u2
= hashmap_get(m
->watch_pids
, PID_TO_PTR(ucred
->pid
));
2303 array
= hashmap_get(m
->watch_pids
, PID_TO_PTR(-ucred
->pid
));
2310 array_copy
= newdup(Unit
*, array
, k
+1);
2314 /* And now invoke the per-unit callbacks. Note that manager_invoke_notify_message() will handle duplicate units
2315 * make sure we only invoke each unit's handler once. */
2317 manager_invoke_notify_message(m
, u1
, ucred
, buf
, fds
);
2321 manager_invoke_notify_message(m
, u2
, ucred
, buf
, fds
);
2325 for (size_t i
= 0; array_copy
[i
]; i
++) {
2326 manager_invoke_notify_message(m
, array_copy
[i
], ucred
, buf
, fds
);
2331 log_warning("Cannot find unit for notify message of PID "PID_FMT
", ignoring.", ucred
->pid
);
2333 if (fdset_size(fds
) > 0)
2334 log_warning("Got extra auxiliary fds with notification message, closing them.");
2339 static void manager_invoke_sigchld_event(
2342 const siginfo_t
*si
) {
2348 /* Already invoked the handler of this unit in this iteration? Then don't process this again */
2349 if (u
->sigchldgen
== m
->sigchldgen
)
2351 u
->sigchldgen
= m
->sigchldgen
;
2353 log_unit_debug(u
, "Child "PID_FMT
" belongs to %s.", si
->si_pid
, u
->id
);
2354 unit_unwatch_pid(u
, si
->si_pid
);
2356 if (UNIT_VTABLE(u
)->sigchld_event
)
2357 UNIT_VTABLE(u
)->sigchld_event(u
, si
->si_pid
, si
->si_code
, si
->si_status
);
2360 static int manager_dispatch_sigchld(sd_event_source
*source
, void *userdata
) {
2361 Manager
*m
= userdata
;
2368 /* First we call waitd() for a PID and do not reap the zombie. That way we can still access /proc/$PID for it
2369 * while it is a zombie. */
2371 if (waitid(P_ALL
, 0, &si
, WEXITED
|WNOHANG
|WNOWAIT
) < 0) {
2373 if (errno
!= ECHILD
)
2374 log_error_errno(errno
, "Failed to peek for child with waitid(), ignoring: %m");
2382 if (IN_SET(si
.si_code
, CLD_EXITED
, CLD_KILLED
, CLD_DUMPED
)) {
2383 _cleanup_free_ Unit
**array_copy
= NULL
;
2384 _cleanup_free_
char *name
= NULL
;
2385 Unit
*u1
, *u2
, **array
;
2387 (void) get_process_comm(si
.si_pid
, &name
);
2389 log_debug("Child "PID_FMT
" (%s) died (code=%s, status=%i/%s)",
2390 si
.si_pid
, strna(name
),
2391 sigchld_code_to_string(si
.si_code
),
2393 strna(si
.si_code
== CLD_EXITED
2394 ? exit_status_to_string(si
.si_status
, EXIT_STATUS_FULL
)
2395 : signal_to_string(si
.si_status
)));
2397 /* Increase the generation counter used for filtering out duplicate unit invocations */
2400 /* And now figure out the unit this belongs to, it might be multiple... */
2401 u1
= manager_get_unit_by_pid_cgroup(m
, si
.si_pid
);
2402 u2
= hashmap_get(m
->watch_pids
, PID_TO_PTR(si
.si_pid
));
2403 array
= hashmap_get(m
->watch_pids
, PID_TO_PTR(-si
.si_pid
));
2407 /* Cound how many entries the array has */
2411 /* Make a copy of the array so that we don't trip up on the array changing beneath us */
2412 array_copy
= newdup(Unit
*, array
, n
+1);
2417 /* Finally, execute them all. Note that u1, u2 and the array might contain duplicates, but
2418 * that's fine, manager_invoke_sigchld_event() will ensure we only invoke the handlers once for
2419 * each iteration. */
2421 manager_invoke_sigchld_event(m
, u1
, &si
);
2423 manager_invoke_sigchld_event(m
, u2
, &si
);
2425 for (size_t i
= 0; array_copy
[i
]; i
++)
2426 manager_invoke_sigchld_event(m
, array_copy
[i
], &si
);
2429 /* And now, we actually reap the zombie. */
2430 if (waitid(P_PID
, si
.si_pid
, &si
, WEXITED
) < 0) {
2431 log_error_errno(errno
, "Failed to dequeue child, ignoring: %m");
2438 /* All children processed for now, turn off event source */
2440 r
= sd_event_source_set_enabled(m
->sigchld_event_source
, SD_EVENT_OFF
);
2442 return log_error_errno(r
, "Failed to disable SIGCHLD event source: %m");
2447 static void manager_start_target(Manager
*m
, const char *name
, JobMode mode
) {
2448 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
2451 log_debug("Activating special unit %s", name
);
2453 r
= manager_add_job_by_name(m
, JOB_START
, name
, mode
, &error
, NULL
);
2455 log_error("Failed to enqueue %s job: %s", name
, bus_error_message(&error
, r
));
2458 static void manager_handle_ctrl_alt_del(Manager
*m
) {
2459 /* If the user presses C-A-D more than
2460 * 7 times within 2s, we reboot/shutdown immediately,
2461 * unless it was disabled in system.conf */
2463 if (ratelimit_below(&m
->ctrl_alt_del_ratelimit
) || m
->cad_burst_action
== EMERGENCY_ACTION_NONE
)
2464 manager_start_target(m
, SPECIAL_CTRL_ALT_DEL_TARGET
, JOB_REPLACE_IRREVERSIBLY
);
2466 emergency_action(m
, m
->cad_burst_action
, NULL
,
2467 "Ctrl-Alt-Del was pressed more than 7 times within 2s");
2470 static int manager_dispatch_signal_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
2471 Manager
*m
= userdata
;
2473 struct signalfd_siginfo sfsi
;
2477 assert(m
->signal_fd
== fd
);
2479 if (revents
!= EPOLLIN
) {
2480 log_warning("Got unexpected events from signal file descriptor.");
2484 n
= read(m
->signal_fd
, &sfsi
, sizeof(sfsi
));
2485 if (n
!= sizeof(sfsi
)) {
2487 log_warning("Truncated read from signal fd (%zu bytes), ignoring!", n
);
2491 if (IN_SET(errno
, EINTR
, EAGAIN
))
2494 /* We return an error here, which will kill this handler,
2495 * to avoid a busy loop on read error. */
2496 return log_error_errno(errno
, "Reading from signal fd failed: %m");
2499 log_received_signal(sfsi
.ssi_signo
== SIGCHLD
||
2500 (sfsi
.ssi_signo
== SIGTERM
&& MANAGER_IS_USER(m
))
2501 ? LOG_DEBUG
: LOG_INFO
,
2504 switch (sfsi
.ssi_signo
) {
2507 r
= sd_event_source_set_enabled(m
->sigchld_event_source
, SD_EVENT_ON
);
2509 log_warning_errno(r
, "Failed to enable SIGCHLD event source, ignoring: %m");
2514 if (MANAGER_IS_SYSTEM(m
)) {
2515 /* This is for compatibility with the original sysvinit */
2516 r
= verify_run_space_and_log("Refusing to reexecute");
2518 m
->exit_code
= MANAGER_REEXECUTE
;
2524 if (MANAGER_IS_SYSTEM(m
))
2525 manager_handle_ctrl_alt_del(m
);
2527 manager_start_target(m
, SPECIAL_EXIT_TARGET
,
2528 JOB_REPLACE_IRREVERSIBLY
);
2532 /* This is a nop on non-init */
2533 if (MANAGER_IS_SYSTEM(m
))
2534 manager_start_target(m
, SPECIAL_KBREQUEST_TARGET
, JOB_REPLACE
);
2539 /* This is a nop on non-init */
2540 if (MANAGER_IS_SYSTEM(m
))
2541 manager_start_target(m
, SPECIAL_SIGPWR_TARGET
, JOB_REPLACE
);
2546 if (manager_dbus_is_running(m
, false)) {
2547 log_info("Trying to reconnect to bus...");
2549 (void) bus_init_api(m
);
2551 if (MANAGER_IS_SYSTEM(m
))
2552 (void) bus_init_system(m
);
2554 log_info("Starting D-Bus service...");
2555 manager_start_target(m
, SPECIAL_DBUS_SERVICE
, JOB_REPLACE
);
2561 _cleanup_free_
char *dump
= NULL
;
2563 r
= manager_get_dump_string(m
, &dump
);
2565 log_warning_errno(errno
, "Failed to acquire manager dump: %m");
2569 log_dump(LOG_INFO
, dump
);
2574 r
= verify_run_space_and_log("Refusing to reload");
2576 m
->exit_code
= MANAGER_RELOAD
;
2581 /* Starting SIGRTMIN+0 */
2582 static const struct {
2585 } target_table
[] = {
2586 [0] = { SPECIAL_DEFAULT_TARGET
, JOB_ISOLATE
},
2587 [1] = { SPECIAL_RESCUE_TARGET
, JOB_ISOLATE
},
2588 [2] = { SPECIAL_EMERGENCY_TARGET
, JOB_ISOLATE
},
2589 [3] = { SPECIAL_HALT_TARGET
, JOB_REPLACE_IRREVERSIBLY
},
2590 [4] = { SPECIAL_POWEROFF_TARGET
, JOB_REPLACE_IRREVERSIBLY
},
2591 [5] = { SPECIAL_REBOOT_TARGET
, JOB_REPLACE_IRREVERSIBLY
},
2592 [6] = { SPECIAL_KEXEC_TARGET
, JOB_REPLACE_IRREVERSIBLY
},
2595 /* Starting SIGRTMIN+13, so that target halt and system halt are 10 apart */
2596 static const ManagerExitCode code_table
[] = {
2598 [1] = MANAGER_POWEROFF
,
2599 [2] = MANAGER_REBOOT
,
2600 [3] = MANAGER_KEXEC
,
2603 if ((int) sfsi
.ssi_signo
>= SIGRTMIN
+0 &&
2604 (int) sfsi
.ssi_signo
< SIGRTMIN
+(int) ELEMENTSOF(target_table
)) {
2605 int idx
= (int) sfsi
.ssi_signo
- SIGRTMIN
;
2606 manager_start_target(m
, target_table
[idx
].target
,
2607 target_table
[idx
].mode
);
2611 if ((int) sfsi
.ssi_signo
>= SIGRTMIN
+13 &&
2612 (int) sfsi
.ssi_signo
< SIGRTMIN
+13+(int) ELEMENTSOF(code_table
)) {
2613 m
->exit_code
= code_table
[sfsi
.ssi_signo
- SIGRTMIN
- 13];
2617 switch (sfsi
.ssi_signo
- SIGRTMIN
) {
2620 manager_set_show_status(m
, SHOW_STATUS_YES
);
2624 manager_set_show_status(m
, SHOW_STATUS_NO
);
2628 manager_override_log_level(m
, LOG_DEBUG
);
2632 manager_restore_original_log_level(m
);
2636 if (MANAGER_IS_USER(m
)) {
2637 m
->exit_code
= MANAGER_EXIT
;
2641 /* This is a nop on init */
2645 case 29: /* compatibility: used to be mapped to LOG_TARGET_SYSLOG_OR_KMSG */
2646 manager_restore_original_log_target(m
);
2650 manager_override_log_target(m
, LOG_TARGET_CONSOLE
);
2654 manager_override_log_target(m
, LOG_TARGET_KMSG
);
2658 log_warning("Got unhandled signal <%s>.", signal_to_string(sfsi
.ssi_signo
));
2665 static int manager_dispatch_time_change_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
2666 Manager
*m
= userdata
;
2671 assert(m
->time_change_fd
== fd
);
2673 log_struct(LOG_DEBUG
,
2674 "MESSAGE_ID=" SD_MESSAGE_TIME_CHANGE_STR
,
2675 LOG_MESSAGE("Time has been changed"));
2677 /* Restart the watch */
2678 (void) manager_setup_time_change(m
);
2680 HASHMAP_FOREACH(u
, m
->units
, i
)
2681 if (UNIT_VTABLE(u
)->time_change
)
2682 UNIT_VTABLE(u
)->time_change(u
);
2687 static int manager_dispatch_timezone_change(
2688 sd_event_source
*source
,
2689 const struct inotify_event
*e
,
2692 Manager
*m
= userdata
;
2699 log_debug("inotify event for /etc/localtime");
2701 changed
= manager_read_timezone_stat(m
);
2707 /* Something changed, restart the watch, to ensure we watch the new /etc/localtime if it changed */
2708 (void) manager_setup_timezone_change(m
);
2710 /* Read the new timezone */
2713 log_debug("Timezone has been changed (now: %s).", tzname
[daylight
]);
2715 HASHMAP_FOREACH(u
, m
->units
, i
)
2716 if (UNIT_VTABLE(u
)->timezone_change
)
2717 UNIT_VTABLE(u
)->timezone_change(u
);
2722 static int manager_dispatch_idle_pipe_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
2723 Manager
*m
= userdata
;
2726 assert(m
->idle_pipe
[2] == fd
);
2728 /* There's at least one Type=idle child that just gave up on us waiting for the boot process to complete. Let's
2729 * now turn off any further console output if there's at least one service that needs console access, so that
2730 * from now on our own output should not spill into that service's output anymore. After all, we support
2731 * Type=idle only to beautify console output and it generally is set on services that want to own the console
2732 * exclusively without our interference. */
2733 m
->no_console_output
= m
->n_on_console
> 0;
2735 /* Acknowledge the child's request, and let all all other children know too that they shouldn't wait any longer
2736 * by closing the pipes towards them, which is what they are waiting for. */
2737 manager_close_idle_pipe(m
);
2742 static int manager_dispatch_jobs_in_progress(sd_event_source
*source
, usec_t usec
, void *userdata
) {
2743 Manager
*m
= userdata
;
2750 manager_print_jobs_in_progress(m
);
2752 next
= now(CLOCK_MONOTONIC
) + JOBS_IN_PROGRESS_PERIOD_USEC
;
2753 r
= sd_event_source_set_time(source
, next
);
2757 return sd_event_source_set_enabled(source
, SD_EVENT_ONESHOT
);
2760 int manager_loop(Manager
*m
) {
2763 RATELIMIT_DEFINE(rl
, 1*USEC_PER_SEC
, 50000);
2766 m
->exit_code
= MANAGER_OK
;
2768 /* Release the path cache */
2769 m
->unit_path_cache
= set_free_free(m
->unit_path_cache
);
2771 manager_check_finished(m
);
2773 /* There might still be some zombies hanging around from before we were exec()'ed. Let's reap them. */
2774 r
= sd_event_source_set_enabled(m
->sigchld_event_source
, SD_EVENT_ON
);
2776 return log_error_errno(r
, "Failed to enable SIGCHLD event source: %m");
2778 while (m
->exit_code
== MANAGER_OK
) {
2781 if (m
->runtime_watchdog
> 0 && m
->runtime_watchdog
!= USEC_INFINITY
&& MANAGER_IS_SYSTEM(m
))
2784 if (!ratelimit_below(&rl
)) {
2785 /* Yay, something is going seriously wrong, pause a little */
2786 log_warning("Looping too fast. Throttling execution a little.");
2790 if (manager_dispatch_load_queue(m
) > 0)
2793 if (manager_dispatch_gc_job_queue(m
) > 0)
2796 if (manager_dispatch_gc_unit_queue(m
) > 0)
2799 if (manager_dispatch_cleanup_queue(m
) > 0)
2802 if (manager_dispatch_cgroup_realize_queue(m
) > 0)
2805 if (manager_dispatch_dbus_queue(m
) > 0)
2808 /* Sleep for half the watchdog time */
2809 if (m
->runtime_watchdog
> 0 && m
->runtime_watchdog
!= USEC_INFINITY
&& MANAGER_IS_SYSTEM(m
)) {
2810 wait_usec
= m
->runtime_watchdog
/ 2;
2814 wait_usec
= USEC_INFINITY
;
2816 r
= sd_event_run(m
->event
, wait_usec
);
2818 return log_error_errno(r
, "Failed to run event loop: %m");
2821 return m
->exit_code
;
2824 int manager_load_unit_from_dbus_path(Manager
*m
, const char *s
, sd_bus_error
*e
, Unit
**_u
) {
2825 _cleanup_free_
char *n
= NULL
;
2826 sd_id128_t invocation_id
;
2834 r
= unit_name_from_dbus_path(s
, &n
);
2838 /* Permit addressing units by invocation ID: if the passed bus path is suffixed by a 128bit ID then we use it
2839 * as invocation ID. */
2840 r
= sd_id128_from_string(n
, &invocation_id
);
2842 u
= hashmap_get(m
->units_by_invocation_id
, &invocation_id
);
2848 return sd_bus_error_setf(e
, BUS_ERROR_NO_UNIT_FOR_INVOCATION_ID
,
2849 "No unit with the specified invocation ID " SD_ID128_FORMAT_STR
" known.",
2850 SD_ID128_FORMAT_VAL(invocation_id
));
2853 /* If this didn't work, we check if this is a unit name */
2854 if (!unit_name_is_valid(n
, UNIT_NAME_PLAIN
|UNIT_NAME_INSTANCE
)) {
2855 _cleanup_free_
char *nn
= NULL
;
2858 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
,
2859 "Unit name %s is neither a valid invocation ID nor unit name.", strnull(nn
));
2862 r
= manager_load_unit(m
, n
, NULL
, e
, &u
);
2870 int manager_get_job_from_dbus_path(Manager
*m
, const char *s
, Job
**_j
) {
2880 p
= startswith(s
, "/org/freedesktop/systemd1/job/");
2884 r
= safe_atou(p
, &id
);
2888 j
= manager_get_job(m
, id
);
2897 void manager_send_unit_audit(Manager
*m
, Unit
*u
, int type
, bool success
) {
2900 _cleanup_free_
char *p
= NULL
;
2904 if (!MANAGER_IS_SYSTEM(m
))
2907 audit_fd
= get_audit_fd();
2911 /* Don't generate audit events if the service was already
2912 * started and we're just deserializing */
2913 if (MANAGER_IS_RELOADING(m
))
2916 if (u
->type
!= UNIT_SERVICE
)
2919 r
= unit_name_to_prefix_and_instance(u
->id
, &p
);
2921 log_error_errno(r
, "Failed to extract prefix and instance of unit name: %m");
2925 msg
= strjoina("unit=", p
);
2926 if (audit_log_user_comm_message(audit_fd
, type
, msg
, "systemd", NULL
, NULL
, NULL
, success
) < 0) {
2928 /* We aren't allowed to send audit messages?
2929 * Then let's not retry again. */
2932 log_warning_errno(errno
, "Failed to send audit message: %m");
2938 void manager_send_unit_plymouth(Manager
*m
, Unit
*u
) {
2939 static const union sockaddr_union sa
= PLYMOUTH_SOCKET
;
2940 _cleanup_free_
char *message
= NULL
;
2941 _cleanup_close_
int fd
= -1;
2944 /* Don't generate plymouth events if the service was already
2945 * started and we're just deserializing */
2946 if (MANAGER_IS_RELOADING(m
))
2949 if (!MANAGER_IS_SYSTEM(m
))
2952 if (detect_container() > 0)
2955 if (!IN_SET(u
->type
, UNIT_SERVICE
, UNIT_MOUNT
, UNIT_SWAP
))
2958 /* We set SOCK_NONBLOCK here so that we rather drop the
2959 * message then wait for plymouth */
2960 fd
= socket(AF_UNIX
, SOCK_STREAM
|SOCK_CLOEXEC
|SOCK_NONBLOCK
, 0);
2962 log_error_errno(errno
, "socket() failed: %m");
2966 if (connect(fd
, &sa
.sa
, SOCKADDR_UN_LEN(sa
.un
)) < 0) {
2967 if (!IN_SET(errno
, EPIPE
, EAGAIN
, ENOENT
, ECONNREFUSED
, ECONNRESET
, ECONNABORTED
))
2968 log_error_errno(errno
, "connect() failed: %m");
2972 if (asprintf(&message
, "U\002%c%s%n", (int) (strlen(u
->id
) + 1), u
->id
, &n
) < 0) {
2978 if (write(fd
, message
, n
+ 1) != n
+ 1)
2979 if (!IN_SET(errno
, EPIPE
, EAGAIN
, ENOENT
, ECONNREFUSED
, ECONNRESET
, ECONNABORTED
))
2980 log_error_errno(errno
, "Failed to write Plymouth message: %m");
2983 int manager_open_serialization(Manager
*m
, FILE **_f
) {
2989 fd
= open_serialization_fd("systemd-state");
2993 f
= fdopen(fd
, "w+");
3003 int manager_serialize(Manager
*m
, FILE *f
, FDSet
*fds
, bool switching_root
) {
3016 fprintf(f
, "current-job-id=%"PRIu32
"\n", m
->current_job_id
);
3017 fprintf(f
, "n-installed-jobs=%u\n", m
->n_installed_jobs
);
3018 fprintf(f
, "n-failed-jobs=%u\n", m
->n_failed_jobs
);
3019 fprintf(f
, "taint-usr=%s\n", yes_no(m
->taint_usr
));
3020 fprintf(f
, "ready-sent=%s\n", yes_no(m
->ready_sent
));
3021 fprintf(f
, "taint-logged=%s\n", yes_no(m
->taint_logged
));
3022 fprintf(f
, "service-watchdogs=%s\n", yes_no(m
->service_watchdogs
));
3024 if (m
->log_level_overridden
)
3025 fprintf(f
, "log-level-override=%i\n", log_get_max_level());
3026 if (m
->log_target_overridden
)
3027 fprintf(f
, "log-target-override=%s\n", log_target_to_string(log_get_target()));
3029 for (q
= 0; q
< _MANAGER_TIMESTAMP_MAX
; q
++) {
3030 /* The userspace and finish timestamps only apply to the host system, hence only serialize them there */
3031 if (in_initrd() && IN_SET(q
, MANAGER_TIMESTAMP_USERSPACE
, MANAGER_TIMESTAMP_FINISH
))
3034 t
= manager_timestamp_to_string(q
);
3036 char field
[strlen(t
) + STRLEN("-timestamp") + 1];
3037 strcpy(stpcpy(field
, t
), "-timestamp");
3038 dual_timestamp_serialize(f
, field
, m
->timestamps
+ q
);
3042 if (!switching_root
)
3043 (void) serialize_environment(f
, m
->environment
);
3045 if (m
->notify_fd
>= 0) {
3048 copy
= fdset_put_dup(fds
, m
->notify_fd
);
3052 fprintf(f
, "notify-fd=%i\n", copy
);
3053 fprintf(f
, "notify-socket=%s\n", m
->notify_socket
);
3056 if (m
->cgroups_agent_fd
>= 0) {
3059 copy
= fdset_put_dup(fds
, m
->cgroups_agent_fd
);
3063 fprintf(f
, "cgroups-agent-fd=%i\n", copy
);
3066 if (m
->user_lookup_fds
[0] >= 0) {
3069 copy0
= fdset_put_dup(fds
, m
->user_lookup_fds
[0]);
3073 copy1
= fdset_put_dup(fds
, m
->user_lookup_fds
[1]);
3077 fprintf(f
, "user-lookup=%i %i\n", copy0
, copy1
);
3080 bus_track_serialize(m
->subscribed
, f
, "subscribed");
3082 r
= dynamic_user_serialize(m
, f
, fds
);
3086 manager_serialize_uid_refs(m
, f
);
3087 manager_serialize_gid_refs(m
, f
);
3089 r
= exec_runtime_serialize(m
, f
, fds
);
3093 (void) fputc('\n', f
);
3095 HASHMAP_FOREACH_KEY(u
, t
, m
->units
, i
) {
3103 r
= unit_serialize(u
, f
, fds
, !switching_root
);
3110 assert(m
->n_reloading
> 0);
3113 r
= fflush_and_check(f
);
3117 r
= bus_fdset_add_all(m
, fds
);
3124 int manager_deserialize(Manager
*m
, FILE *f
, FDSet
*fds
) {
3130 log_debug("Deserializing state...");
3135 char line
[LINE_MAX
];
3136 const char *val
, *l
;
3138 if (!fgets(line
, sizeof(line
), f
)) {
3153 if ((val
= startswith(l
, "current-job-id="))) {
3156 if (safe_atou32(val
, &id
) < 0)
3157 log_notice("Failed to parse current job id value %s", val
);
3159 m
->current_job_id
= MAX(m
->current_job_id
, id
);
3161 } else if ((val
= startswith(l
, "n-installed-jobs="))) {
3164 if (safe_atou32(val
, &n
) < 0)
3165 log_notice("Failed to parse installed jobs counter %s", val
);
3167 m
->n_installed_jobs
+= n
;
3169 } else if ((val
= startswith(l
, "n-failed-jobs="))) {
3172 if (safe_atou32(val
, &n
) < 0)
3173 log_notice("Failed to parse failed jobs counter %s", val
);
3175 m
->n_failed_jobs
+= n
;
3177 } else if ((val
= startswith(l
, "taint-usr="))) {
3180 b
= parse_boolean(val
);
3182 log_notice("Failed to parse taint /usr flag %s", val
);
3184 m
->taint_usr
= m
->taint_usr
|| b
;
3186 } else if ((val
= startswith(l
, "ready-sent="))) {
3189 b
= parse_boolean(val
);
3191 log_notice("Failed to parse ready-sent flag %s", val
);
3193 m
->ready_sent
= m
->ready_sent
|| b
;
3195 } else if ((val
= startswith(l
, "taint-logged="))) {
3198 b
= parse_boolean(val
);
3200 log_notice("Failed to parse taint-logged flag %s", val
);
3202 m
->taint_logged
= m
->taint_logged
|| b
;
3204 } else if ((val
= startswith(l
, "service-watchdogs="))) {
3207 b
= parse_boolean(val
);
3209 log_notice("Failed to parse service-watchdogs flag %s", val
);
3211 m
->service_watchdogs
= b
;
3213 } else if ((val
= startswith(l
, "log-level-override="))) {
3216 level
= log_level_from_string(val
);
3218 log_notice("Failed to parse log-level-override value '%s', ignoring.", val
);
3220 manager_override_log_level(m
, level
);
3222 } else if ((val
= startswith(l
, "log-target-override="))) {
3225 target
= log_target_from_string(val
);
3227 log_notice("Failed to parse log-target-override value '%s', ignoring.", val
);
3229 manager_override_log_target(m
, target
);
3231 } else if (startswith(l
, "env=")) {
3232 r
= deserialize_environment(&m
->environment
, l
);
3236 log_notice_errno(r
, "Failed to parse environment entry: \"%s\": %m", l
);
3238 } else if ((val
= startswith(l
, "notify-fd="))) {
3241 if (safe_atoi(val
, &fd
) < 0 || fd
< 0 || !fdset_contains(fds
, fd
))
3242 log_notice("Failed to parse notify fd: \"%s\"", val
);
3244 m
->notify_event_source
= sd_event_source_unref(m
->notify_event_source
);
3245 safe_close(m
->notify_fd
);
3246 m
->notify_fd
= fdset_remove(fds
, fd
);
3249 } else if ((val
= startswith(l
, "notify-socket="))) {
3258 free(m
->notify_socket
);
3259 m
->notify_socket
= n
;
3261 } else if ((val
= startswith(l
, "cgroups-agent-fd="))) {
3264 if (safe_atoi(val
, &fd
) < 0 || fd
< 0 || !fdset_contains(fds
, fd
))
3265 log_notice("Failed to parse cgroups agent fd: %s", val
);
3267 m
->cgroups_agent_event_source
= sd_event_source_unref(m
->cgroups_agent_event_source
);
3268 safe_close(m
->cgroups_agent_fd
);
3269 m
->cgroups_agent_fd
= fdset_remove(fds
, fd
);
3272 } else if ((val
= startswith(l
, "user-lookup="))) {
3275 if (sscanf(val
, "%i %i", &fd0
, &fd1
) != 2 || fd0
< 0 || fd1
< 0 || fd0
== fd1
|| !fdset_contains(fds
, fd0
) || !fdset_contains(fds
, fd1
))
3276 log_notice("Failed to parse user lookup fd: %s", val
);
3278 m
->user_lookup_event_source
= sd_event_source_unref(m
->user_lookup_event_source
);
3279 safe_close_pair(m
->user_lookup_fds
);
3280 m
->user_lookup_fds
[0] = fdset_remove(fds
, fd0
);
3281 m
->user_lookup_fds
[1] = fdset_remove(fds
, fd1
);
3284 } else if ((val
= startswith(l
, "dynamic-user=")))
3285 dynamic_user_deserialize_one(m
, val
, fds
);
3286 else if ((val
= startswith(l
, "destroy-ipc-uid=")))
3287 manager_deserialize_uid_refs_one(m
, val
);
3288 else if ((val
= startswith(l
, "destroy-ipc-gid=")))
3289 manager_deserialize_gid_refs_one(m
, val
);
3290 else if ((val
= startswith(l
, "exec-runtime=")))
3291 exec_runtime_deserialize_one(m
, val
, fds
);
3292 else if ((val
= startswith(l
, "subscribed="))) {
3294 if (strv_extend(&m
->deserialized_subscribed
, val
) < 0)
3299 for (q
= 0; q
< _MANAGER_TIMESTAMP_MAX
; q
++) {
3300 val
= startswith(l
, manager_timestamp_to_string(q
));
3304 val
= startswith(val
, "-timestamp=");
3309 if (q
< _MANAGER_TIMESTAMP_MAX
) /* found it */
3310 dual_timestamp_deserialize(val
, m
->timestamps
+ q
);
3311 else if (!startswith(l
, "kdbus-fd=")) /* ignore kdbus */
3312 log_notice("Unknown serialization item '%s'", l
);
3318 char name
[UNIT_NAME_MAX
+2];
3319 const char* unit_name
;
3322 if (!fgets(name
, sizeof(name
), f
)) {
3332 unit_name
= strstrip(name
);
3334 r
= manager_load_unit(m
, unit_name
, NULL
, NULL
, &u
);
3336 log_notice_errno(r
, "Failed to load unit \"%s\", skipping deserialization: %m", unit_name
);
3339 unit_deserialize_skip(f
);
3343 r
= unit_deserialize(u
, f
, fds
);
3345 log_notice_errno(r
, "Failed to deserialize unit \"%s\": %m", unit_name
);
3355 assert(m
->n_reloading
> 0);
3361 static void manager_flush_finished_jobs(Manager
*m
) {
3364 while ((j
= set_steal_first(m
->pending_finished_jobs
))) {
3365 bus_job_send_removed_signal(j
);
3369 m
->pending_finished_jobs
= set_free(m
->pending_finished_jobs
);
3372 int manager_reload(Manager
*m
) {
3374 _cleanup_fclose_
FILE *f
= NULL
;
3375 _cleanup_fdset_free_ FDSet
*fds
= NULL
;
3379 r
= manager_open_serialization(m
, &f
);
3384 bus_manager_send_reloading(m
, true);
3392 r
= manager_serialize(m
, f
, fds
, false);
3398 if (fseeko(f
, 0, SEEK_SET
) < 0) {
3403 /* From here on there is no way back. */
3404 manager_clear_jobs_and_units(m
);
3405 lookup_paths_flush_generator(&m
->lookup_paths
);
3406 lookup_paths_free(&m
->lookup_paths
);
3407 exec_runtime_vacuum(m
);
3408 dynamic_user_vacuum(m
, false);
3409 m
->uid_refs
= hashmap_free(m
->uid_refs
);
3410 m
->gid_refs
= hashmap_free(m
->gid_refs
);
3412 q
= lookup_paths_init(&m
->lookup_paths
, m
->unit_file_scope
, 0, NULL
);
3413 if (q
< 0 && r
>= 0)
3416 q
= manager_run_environment_generators(m
);
3417 if (q
< 0 && r
>= 0)
3420 /* Find new unit paths */
3421 q
= manager_run_generators(m
);
3422 if (q
< 0 && r
>= 0)
3425 lookup_paths_reduce(&m
->lookup_paths
);
3426 manager_build_unit_path_cache(m
);
3428 /* First, enumerate what we can from all config files */
3429 manager_enumerate(m
);
3431 /* Second, deserialize our stored data */
3432 q
= manager_deserialize(m
, f
, fds
);
3434 log_error_errno(q
, "Deserialization failed: %m");
3442 /* Re-register notify_fd as event source */
3443 q
= manager_setup_notify(m
);
3444 if (q
< 0 && r
>= 0)
3447 q
= manager_setup_cgroups_agent(m
);
3448 if (q
< 0 && r
>= 0)
3451 q
= manager_setup_user_lookup_fd(m
);
3452 if (q
< 0 && r
>= 0)
3455 /* Third, fire things up! */
3456 manager_coldplug(m
);
3458 /* Release any dynamic users no longer referenced */
3459 dynamic_user_vacuum(m
, true);
3461 /* Release any references to UIDs/GIDs no longer referenced, and destroy any IPC owned by them */
3462 manager_vacuum_uid_refs(m
);
3463 manager_vacuum_gid_refs(m
);
3465 exec_runtime_vacuum(m
);
3467 assert(m
->n_reloading
> 0);
3470 /* It might be safe to log to the journal now and connect to dbus */
3471 manager_recheck_journal(m
);
3472 manager_recheck_dbus(m
);
3474 /* Let's finally catch up with any changes that took place while we were reloading/reexecing */
3477 /* Sync current state of bus names with our set of listening units */
3478 q
= manager_enqueue_sync_bus_names(m
);
3479 if (q
< 0 && r
>= 0)
3482 if (!MANAGER_IS_RELOADING(m
))
3483 manager_flush_finished_jobs(m
);
3485 m
->send_reloading_done
= true;
3490 void manager_reset_failed(Manager
*m
) {
3496 HASHMAP_FOREACH(u
, m
->units
, i
)
3497 unit_reset_failed(u
);
3500 bool manager_unit_inactive_or_pending(Manager
*m
, const char *name
) {
3506 /* Returns true if the unit is inactive or going down */
3507 u
= manager_get_unit(m
, name
);
3511 return unit_inactive_or_pending(u
);
3514 static void log_taint_string(Manager
*m
) {
3515 _cleanup_free_
char *taint
= NULL
;
3519 if (MANAGER_IS_USER(m
) || m
->taint_logged
)
3522 m
->taint_logged
= true; /* only check for taint once */
3524 taint
= manager_taint_string(m
);
3528 log_struct(LOG_NOTICE
,
3529 LOG_MESSAGE("System is tainted: %s", taint
),
3531 "MESSAGE_ID=" SD_MESSAGE_TAINTED_STR
);
3534 static void manager_notify_finished(Manager
*m
) {
3535 char userspace
[FORMAT_TIMESPAN_MAX
], initrd
[FORMAT_TIMESPAN_MAX
], kernel
[FORMAT_TIMESPAN_MAX
], sum
[FORMAT_TIMESPAN_MAX
];
3536 usec_t firmware_usec
, loader_usec
, kernel_usec
, initrd_usec
, userspace_usec
, total_usec
;
3538 if (m
->test_run_flags
)
3541 if (MANAGER_IS_SYSTEM(m
) && detect_container() <= 0) {
3542 char ts
[FORMAT_TIMESPAN_MAX
];
3543 char buf
[FORMAT_TIMESPAN_MAX
+ STRLEN(" (firmware) + ") + FORMAT_TIMESPAN_MAX
+ STRLEN(" (loader) + ")]
3546 size_t size
= sizeof buf
;
3548 /* Note that MANAGER_TIMESTAMP_KERNEL's monotonic value is always at 0, and
3549 * MANAGER_TIMESTAMP_FIRMWARE's and MANAGER_TIMESTAMP_LOADER's monotonic value should be considered
3550 * negative values. */
3552 firmware_usec
= m
->timestamps
[MANAGER_TIMESTAMP_FIRMWARE
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_LOADER
].monotonic
;
3553 loader_usec
= m
->timestamps
[MANAGER_TIMESTAMP_LOADER
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_KERNEL
].monotonic
;
3554 userspace_usec
= m
->timestamps
[MANAGER_TIMESTAMP_FINISH
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_USERSPACE
].monotonic
;
3555 total_usec
= m
->timestamps
[MANAGER_TIMESTAMP_FIRMWARE
].monotonic
+ m
->timestamps
[MANAGER_TIMESTAMP_FINISH
].monotonic
;
3557 if (firmware_usec
> 0)
3558 size
= strpcpyf(&p
, size
, "%s (firmware) + ", format_timespan(ts
, sizeof(ts
), firmware_usec
, USEC_PER_MSEC
));
3559 if (loader_usec
> 0)
3560 size
= strpcpyf(&p
, size
, "%s (loader) + ", format_timespan(ts
, sizeof(ts
), loader_usec
, USEC_PER_MSEC
));
3562 if (dual_timestamp_is_set(&m
->timestamps
[MANAGER_TIMESTAMP_INITRD
])) {
3564 /* The initrd case on bare-metal*/
3565 kernel_usec
= m
->timestamps
[MANAGER_TIMESTAMP_INITRD
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_KERNEL
].monotonic
;
3566 initrd_usec
= m
->timestamps
[MANAGER_TIMESTAMP_USERSPACE
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_INITRD
].monotonic
;
3568 log_struct(LOG_INFO
,
3569 "MESSAGE_ID=" SD_MESSAGE_STARTUP_FINISHED_STR
,
3570 "KERNEL_USEC="USEC_FMT
, kernel_usec
,
3571 "INITRD_USEC="USEC_FMT
, initrd_usec
,
3572 "USERSPACE_USEC="USEC_FMT
, userspace_usec
,
3573 LOG_MESSAGE("Startup finished in %s%s (kernel) + %s (initrd) + %s (userspace) = %s.",
3575 format_timespan(kernel
, sizeof(kernel
), kernel_usec
, USEC_PER_MSEC
),
3576 format_timespan(initrd
, sizeof(initrd
), initrd_usec
, USEC_PER_MSEC
),
3577 format_timespan(userspace
, sizeof(userspace
), userspace_usec
, USEC_PER_MSEC
),
3578 format_timespan(sum
, sizeof(sum
), total_usec
, USEC_PER_MSEC
)));
3580 /* The initrd-less case on bare-metal*/
3582 kernel_usec
= m
->timestamps
[MANAGER_TIMESTAMP_USERSPACE
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_KERNEL
].monotonic
;
3585 log_struct(LOG_INFO
,
3586 "MESSAGE_ID=" SD_MESSAGE_STARTUP_FINISHED_STR
,
3587 "KERNEL_USEC="USEC_FMT
, kernel_usec
,
3588 "USERSPACE_USEC="USEC_FMT
, userspace_usec
,
3589 LOG_MESSAGE("Startup finished in %s%s (kernel) + %s (userspace) = %s.",
3591 format_timespan(kernel
, sizeof(kernel
), kernel_usec
, USEC_PER_MSEC
),
3592 format_timespan(userspace
, sizeof(userspace
), userspace_usec
, USEC_PER_MSEC
),
3593 format_timespan(sum
, sizeof(sum
), total_usec
, USEC_PER_MSEC
)));
3596 /* The container and --user case */
3597 firmware_usec
= loader_usec
= initrd_usec
= kernel_usec
= 0;
3598 total_usec
= userspace_usec
= m
->timestamps
[MANAGER_TIMESTAMP_FINISH
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_USERSPACE
].monotonic
;
3600 log_struct(LOG_INFO
,
3601 "MESSAGE_ID=" SD_MESSAGE_USER_STARTUP_FINISHED_STR
,
3602 "USERSPACE_USEC="USEC_FMT
, userspace_usec
,
3603 LOG_MESSAGE("Startup finished in %s.",
3604 format_timespan(sum
, sizeof(sum
), total_usec
, USEC_PER_MSEC
)));
3607 bus_manager_send_finished(m
, firmware_usec
, loader_usec
, kernel_usec
, initrd_usec
, userspace_usec
, total_usec
);
3610 m
->ready_sent
? "STATUS=Startup finished in %s."
3612 "STATUS=Startup finished in %s.",
3613 format_timespan(sum
, sizeof(sum
), total_usec
, USEC_PER_MSEC
));
3614 m
->ready_sent
= true;
3616 log_taint_string(m
);
3619 static void manager_send_ready(Manager
*m
) {
3622 /* We send READY=1 on reaching basic.target only when running in --user mode. */
3623 if (!MANAGER_IS_USER(m
) || m
->ready_sent
)
3626 m
->ready_sent
= true;
3630 "STATUS=Reached " SPECIAL_BASIC_TARGET
".");
3633 static void manager_check_basic_target(Manager
*m
) {
3638 /* Small shortcut */
3639 if (m
->ready_sent
&& m
->taint_logged
)
3642 u
= manager_get_unit(m
, SPECIAL_BASIC_TARGET
);
3643 if (!u
|| !UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u
)))
3646 /* For user managers, send out READY=1 as soon as we reach basic.target */
3647 manager_send_ready(m
);
3649 /* Log the taint string as soon as we reach basic.target */
3650 log_taint_string(m
);
3653 void manager_check_finished(Manager
*m
) {
3656 if (MANAGER_IS_RELOADING(m
))
3659 /* Verify that we have entered the event loop already, and not left it again. */
3660 if (!MANAGER_IS_RUNNING(m
))
3663 manager_check_basic_target(m
);
3665 if (hashmap_size(m
->jobs
) > 0) {
3666 if (m
->jobs_in_progress_event_source
)
3667 /* Ignore any failure, this is only for feedback */
3668 (void) sd_event_source_set_time(m
->jobs_in_progress_event_source
, now(CLOCK_MONOTONIC
) + JOBS_IN_PROGRESS_WAIT_USEC
);
3673 manager_flip_auto_status(m
, false);
3675 /* Notify Type=idle units that we are done now */
3676 manager_close_idle_pipe(m
);
3678 /* Turn off confirm spawn now */
3679 m
->confirm_spawn
= NULL
;
3681 /* No need to update ask password status when we're going non-interactive */
3682 manager_close_ask_password(m
);
3684 /* This is no longer the first boot */
3685 manager_set_first_boot(m
, false);
3687 if (MANAGER_IS_FINISHED(m
))
3690 dual_timestamp_get(m
->timestamps
+ MANAGER_TIMESTAMP_FINISH
);
3692 manager_notify_finished(m
);
3694 manager_invalidate_startup_units(m
);
3697 static bool generator_path_any(const char* const* paths
) {
3701 /* Optimize by skipping the whole process by not creating output directories
3702 * if no generators are found. */
3703 STRV_FOREACH(path
, (char**) paths
)
3704 if (access(*path
, F_OK
) == 0)
3706 else if (errno
!= ENOENT
)
3707 log_warning_errno(errno
, "Failed to open generator directory %s: %m", *path
);
3712 static const char* system_env_generator_binary_paths
[] = {
3713 "/run/systemd/system-environment-generators",
3714 "/etc/systemd/system-environment-generators",
3715 "/usr/local/lib/systemd/system-environment-generators",
3716 SYSTEM_ENV_GENERATOR_PATH
,
3720 static const char* user_env_generator_binary_paths
[] = {
3721 "/run/systemd/user-environment-generators",
3722 "/etc/systemd/user-environment-generators",
3723 "/usr/local/lib/systemd/user-environment-generators",
3724 USER_ENV_GENERATOR_PATH
,
3728 static int manager_run_environment_generators(Manager
*m
) {
3729 char **tmp
= NULL
; /* this is only used in the forked process, no cleanup here */
3731 void* args
[] = {&tmp
, &tmp
, &m
->environment
};
3733 if (m
->test_run_flags
&& !(m
->test_run_flags
& MANAGER_TEST_RUN_ENV_GENERATORS
))
3736 paths
= MANAGER_IS_SYSTEM(m
) ? system_env_generator_binary_paths
: user_env_generator_binary_paths
;
3738 if (!generator_path_any(paths
))
3741 return execute_directories(paths
, DEFAULT_TIMEOUT_USEC
, gather_environment
, args
, NULL
);
3744 static int manager_run_generators(Manager
*m
) {
3745 _cleanup_strv_free_
char **paths
= NULL
;
3746 const char *argv
[5];
3751 if (m
->test_run_flags
&& !(m
->test_run_flags
& MANAGER_TEST_RUN_GENERATORS
))
3754 paths
= generator_binary_paths(m
->unit_file_scope
);
3758 if (!generator_path_any((const char* const*) paths
))
3761 r
= lookup_paths_mkdir_generator(&m
->lookup_paths
);
3765 argv
[0] = NULL
; /* Leave this empty, execute_directory() will fill something in */
3766 argv
[1] = m
->lookup_paths
.generator
;
3767 argv
[2] = m
->lookup_paths
.generator_early
;
3768 argv
[3] = m
->lookup_paths
.generator_late
;
3771 RUN_WITH_UMASK(0022)
3772 execute_directories((const char* const*) paths
, DEFAULT_TIMEOUT_USEC
,
3773 NULL
, NULL
, (char**) argv
);
3776 lookup_paths_trim_generator(&m
->lookup_paths
);
3780 int manager_environment_add(Manager
*m
, char **minus
, char **plus
) {
3781 char **a
= NULL
, **b
= NULL
, **l
;
3786 if (!strv_isempty(minus
)) {
3787 a
= strv_env_delete(l
, 1, minus
);
3794 if (!strv_isempty(plus
)) {
3795 b
= strv_env_merge(2, l
, plus
);
3804 if (m
->environment
!= l
)
3805 strv_free(m
->environment
);
3812 manager_sanitize_environment(m
);
3817 int manager_set_default_rlimits(Manager
*m
, struct rlimit
**default_rlimit
) {
3822 for (i
= 0; i
< _RLIMIT_MAX
; i
++) {
3823 m
->rlimit
[i
] = mfree(m
->rlimit
[i
]);
3825 if (!default_rlimit
[i
])
3828 m
->rlimit
[i
] = newdup(struct rlimit
, default_rlimit
[i
], 1);
3836 void manager_recheck_dbus(Manager
*m
) {
3839 /* Connects to the bus if the dbus service and socket are running. If we are running in user mode this is all
3840 * it does. In system mode we'll also connect to the system bus (which will most likely just reuse the
3841 * connection of the API bus). That's because the system bus after all runs as service of the system instance,
3842 * while in the user instance we can assume it's already there. */
3844 if (MANAGER_IS_RELOADING(m
))
3845 return; /* don't check while we are reloading… */
3847 if (manager_dbus_is_running(m
, false)) {
3848 (void) bus_init_api(m
);
3850 if (MANAGER_IS_SYSTEM(m
))
3851 (void) bus_init_system(m
);
3853 (void) bus_done_api(m
);
3855 if (MANAGER_IS_SYSTEM(m
))
3856 (void) bus_done_system(m
);
3860 static bool manager_journal_is_running(Manager
*m
) {
3865 if (m
->test_run_flags
!= 0)
3868 /* If we are the user manager we can safely assume that the journal is up */
3869 if (!MANAGER_IS_SYSTEM(m
))
3872 /* Check that the socket is not only up, but in RUNNING state */
3873 u
= manager_get_unit(m
, SPECIAL_JOURNALD_SOCKET
);
3876 if (SOCKET(u
)->state
!= SOCKET_RUNNING
)
3879 /* Similar, check if the daemon itself is fully up, too */
3880 u
= manager_get_unit(m
, SPECIAL_JOURNALD_SERVICE
);
3883 if (!IN_SET(SERVICE(u
)->state
, SERVICE_RELOAD
, SERVICE_RUNNING
))
3889 void manager_recheck_journal(Manager
*m
) {
3893 /* Don't bother with this unless we are in the special situation of being PID 1 */
3894 if (getpid_cached() != 1)
3897 /* Don't check this while we are reloading, things might still change */
3898 if (MANAGER_IS_RELOADING(m
))
3901 /* The journal is fully and entirely up? If so, let's permit logging to it, if that's configured. If the
3902 * journal is down, don't ever log to it, otherwise we might end up deadlocking ourselves as we might trigger
3903 * an activation ourselves we can't fulfill. */
3904 log_set_prohibit_ipc(!manager_journal_is_running(m
));
3908 void manager_set_show_status(Manager
*m
, ShowStatus mode
) {
3910 assert(IN_SET(mode
, SHOW_STATUS_AUTO
, SHOW_STATUS_NO
, SHOW_STATUS_YES
, SHOW_STATUS_TEMPORARY
));
3912 if (!MANAGER_IS_SYSTEM(m
))
3915 if (m
->show_status
!= mode
)
3916 log_debug("%s showing of status.",
3917 mode
== SHOW_STATUS_NO
? "Disabling" : "Enabling");
3918 m
->show_status
= mode
;
3921 (void) touch("/run/systemd/show-status");
3923 (void) unlink("/run/systemd/show-status");
3926 static bool manager_get_show_status(Manager
*m
, StatusType type
) {
3929 if (!MANAGER_IS_SYSTEM(m
))
3932 if (m
->no_console_output
)
3935 if (!IN_SET(manager_state(m
), MANAGER_INITIALIZING
, MANAGER_STARTING
, MANAGER_STOPPING
))
3938 /* If we cannot find out the status properly, just proceed. */
3939 if (type
!= STATUS_TYPE_EMERGENCY
&& manager_check_ask_password(m
) > 0)
3942 return m
->show_status
> 0;
3945 const char *manager_get_confirm_spawn(Manager
*m
) {
3946 static int last_errno
= 0;
3947 const char *vc
= m
->confirm_spawn
;
3951 /* Here's the deal: we want to test the validity of the console but don't want
3952 * PID1 to go through the whole console process which might block. But we also
3953 * want to warn the user only once if something is wrong with the console so we
3954 * cannot do the sanity checks after spawning our children. So here we simply do
3955 * really basic tests to hopefully trap common errors.
3957 * If the console suddenly disappear at the time our children will really it
3958 * then they will simply fail to acquire it and a positive answer will be
3959 * assumed. New children will fallback to /dev/console though.
3961 * Note: TTYs are devices that can come and go any time, and frequently aren't
3962 * available yet during early boot (consider a USB rs232 dongle...). If for any
3963 * reason the configured console is not ready, we fallback to the default
3966 if (!vc
|| path_equal(vc
, "/dev/console"))
3973 if (!S_ISCHR(st
.st_mode
)) {
3981 if (last_errno
!= errno
) {
3983 log_warning_errno(errno
, "Failed to open %s: %m, using default console", vc
);
3985 return "/dev/console";
3988 void manager_set_first_boot(Manager
*m
, bool b
) {
3991 if (!MANAGER_IS_SYSTEM(m
))
3994 if (m
->first_boot
!= (int) b
) {
3996 (void) touch("/run/systemd/first-boot");
3998 (void) unlink("/run/systemd/first-boot");
4004 void manager_disable_confirm_spawn(void) {
4005 (void) touch("/run/systemd/confirm_spawn_disabled");
4008 bool manager_is_confirm_spawn_disabled(Manager
*m
) {
4009 if (!m
->confirm_spawn
)
4012 return access("/run/systemd/confirm_spawn_disabled", F_OK
) >= 0;
4015 void manager_status_printf(Manager
*m
, StatusType type
, const char *status
, const char *format
, ...) {
4018 /* If m is NULL, assume we're after shutdown and let the messages through. */
4020 if (m
&& !manager_get_show_status(m
, type
))
4023 /* XXX We should totally drop the check for ephemeral here
4024 * and thus effectively make 'Type=idle' pointless. */
4025 if (type
== STATUS_TYPE_EPHEMERAL
&& m
&& m
->n_on_console
> 0)
4028 va_start(ap
, format
);
4029 status_vprintf(status
, true, type
== STATUS_TYPE_EPHEMERAL
, format
, ap
);
4033 Set
*manager_get_units_requiring_mounts_for(Manager
*m
, const char *path
) {
4034 char p
[strlen(path
)+1];
4040 path_simplify(p
, false);
4042 return hashmap_get(m
->units_requiring_mounts_for
, streq(p
, "/") ? "" : p
);
4045 int manager_update_failed_units(Manager
*m
, Unit
*u
, bool failed
) {
4050 assert(u
->manager
== m
);
4052 size
= set_size(m
->failed_units
);
4055 r
= set_ensure_allocated(&m
->failed_units
, NULL
);
4059 if (set_put(m
->failed_units
, u
) < 0)
4062 (void) set_remove(m
->failed_units
, u
);
4064 if (set_size(m
->failed_units
) != size
)
4065 bus_manager_send_change_signal(m
);
4070 ManagerState
manager_state(Manager
*m
) {
4075 /* Did we ever finish booting? If not then we are still starting up */
4076 if (!MANAGER_IS_FINISHED(m
)) {
4078 u
= manager_get_unit(m
, SPECIAL_BASIC_TARGET
);
4079 if (!u
|| !UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u
)))
4080 return MANAGER_INITIALIZING
;
4082 return MANAGER_STARTING
;
4085 /* Is the special shutdown target active or queued? If so, we are in shutdown state */
4086 u
= manager_get_unit(m
, SPECIAL_SHUTDOWN_TARGET
);
4087 if (u
&& unit_active_or_pending(u
))
4088 return MANAGER_STOPPING
;
4090 if (MANAGER_IS_SYSTEM(m
)) {
4091 /* Are the rescue or emergency targets active or queued? If so we are in maintenance state */
4092 u
= manager_get_unit(m
, SPECIAL_RESCUE_TARGET
);
4093 if (u
&& unit_active_or_pending(u
))
4094 return MANAGER_MAINTENANCE
;
4096 u
= manager_get_unit(m
, SPECIAL_EMERGENCY_TARGET
);
4097 if (u
&& unit_active_or_pending(u
))
4098 return MANAGER_MAINTENANCE
;
4101 /* Are there any failed units? If so, we are in degraded mode */
4102 if (set_size(m
->failed_units
) > 0)
4103 return MANAGER_DEGRADED
;
4105 return MANAGER_RUNNING
;
4108 #define DESTROY_IPC_FLAG (UINT32_C(1) << 31)
4110 static void manager_unref_uid_internal(
4115 int (*_clean_ipc
)(uid_t uid
)) {
4121 assert(uid_is_valid(uid
));
4124 /* A generic implementation, covering both manager_unref_uid() and manager_unref_gid(), under the assumption
4125 * that uid_t and gid_t are actually defined the same way, with the same validity rules.
4127 * We store a hashmap where the UID/GID is they key and the value is a 32bit reference counter, whose highest
4128 * bit is used as flag for marking UIDs/GIDs whose IPC objects to remove when the last reference to the UID/GID
4129 * is dropped. The flag is set to on, once at least one reference from a unit where RemoveIPC= is set is added
4130 * on a UID/GID. It is reset when the UID's/GID's reference counter drops to 0 again. */
4132 assert_cc(sizeof(uid_t
) == sizeof(gid_t
));
4133 assert_cc(UID_INVALID
== (uid_t
) GID_INVALID
);
4135 if (uid
== 0) /* We don't keep track of root, and will never destroy it */
4138 c
= PTR_TO_UINT32(hashmap_get(*uid_refs
, UID_TO_PTR(uid
)));
4140 n
= c
& ~DESTROY_IPC_FLAG
;
4144 if (destroy_now
&& n
== 0) {
4145 hashmap_remove(*uid_refs
, UID_TO_PTR(uid
));
4147 if (c
& DESTROY_IPC_FLAG
) {
4148 log_debug("%s " UID_FMT
" is no longer referenced, cleaning up its IPC.",
4149 _clean_ipc
== clean_ipc_by_uid
? "UID" : "GID",
4151 (void) _clean_ipc(uid
);
4154 c
= n
| (c
& DESTROY_IPC_FLAG
);
4155 assert_se(hashmap_update(*uid_refs
, UID_TO_PTR(uid
), UINT32_TO_PTR(c
)) >= 0);
4159 void manager_unref_uid(Manager
*m
, uid_t uid
, bool destroy_now
) {
4160 manager_unref_uid_internal(m
, &m
->uid_refs
, uid
, destroy_now
, clean_ipc_by_uid
);
4163 void manager_unref_gid(Manager
*m
, gid_t gid
, bool destroy_now
) {
4164 manager_unref_uid_internal(m
, &m
->gid_refs
, (uid_t
) gid
, destroy_now
, clean_ipc_by_gid
);
4167 static int manager_ref_uid_internal(
4178 assert(uid_is_valid(uid
));
4180 /* A generic implementation, covering both manager_ref_uid() and manager_ref_gid(), under the assumption
4181 * that uid_t and gid_t are actually defined the same way, with the same validity rules. */
4183 assert_cc(sizeof(uid_t
) == sizeof(gid_t
));
4184 assert_cc(UID_INVALID
== (uid_t
) GID_INVALID
);
4186 if (uid
== 0) /* We don't keep track of root, and will never destroy it */
4189 r
= hashmap_ensure_allocated(uid_refs
, &trivial_hash_ops
);
4193 c
= PTR_TO_UINT32(hashmap_get(*uid_refs
, UID_TO_PTR(uid
)));
4195 n
= c
& ~DESTROY_IPC_FLAG
;
4198 if (n
& DESTROY_IPC_FLAG
) /* check for overflow */
4201 c
= n
| (c
& DESTROY_IPC_FLAG
) | (clean_ipc
? DESTROY_IPC_FLAG
: 0);
4203 return hashmap_replace(*uid_refs
, UID_TO_PTR(uid
), UINT32_TO_PTR(c
));
4206 int manager_ref_uid(Manager
*m
, uid_t uid
, bool clean_ipc
) {
4207 return manager_ref_uid_internal(m
, &m
->uid_refs
, uid
, clean_ipc
);
4210 int manager_ref_gid(Manager
*m
, gid_t gid
, bool clean_ipc
) {
4211 return manager_ref_uid_internal(m
, &m
->gid_refs
, (uid_t
) gid
, clean_ipc
);
4214 static void manager_vacuum_uid_refs_internal(
4217 int (*_clean_ipc
)(uid_t uid
)) {
4226 HASHMAP_FOREACH_KEY(p
, k
, *uid_refs
, i
) {
4230 uid
= PTR_TO_UID(k
);
4231 c
= PTR_TO_UINT32(p
);
4233 n
= c
& ~DESTROY_IPC_FLAG
;
4237 if (c
& DESTROY_IPC_FLAG
) {
4238 log_debug("Found unreferenced %s " UID_FMT
" after reload/reexec. Cleaning up.",
4239 _clean_ipc
== clean_ipc_by_uid
? "UID" : "GID",
4241 (void) _clean_ipc(uid
);
4244 assert_se(hashmap_remove(*uid_refs
, k
) == p
);
4248 void manager_vacuum_uid_refs(Manager
*m
) {
4249 manager_vacuum_uid_refs_internal(m
, &m
->uid_refs
, clean_ipc_by_uid
);
4252 void manager_vacuum_gid_refs(Manager
*m
) {
4253 manager_vacuum_uid_refs_internal(m
, &m
->gid_refs
, clean_ipc_by_gid
);
4256 static void manager_serialize_uid_refs_internal(
4260 const char *field_name
) {
4270 /* Serialize the UID reference table. Or actually, just the IPC destruction flag of it, as the actual counter
4271 * of it is better rebuild after a reload/reexec. */
4273 HASHMAP_FOREACH_KEY(p
, k
, *uid_refs
, i
) {
4277 uid
= PTR_TO_UID(k
);
4278 c
= PTR_TO_UINT32(p
);
4280 if (!(c
& DESTROY_IPC_FLAG
))
4283 fprintf(f
, "%s=" UID_FMT
"\n", field_name
, uid
);
4287 void manager_serialize_uid_refs(Manager
*m
, FILE *f
) {
4288 manager_serialize_uid_refs_internal(m
, f
, &m
->uid_refs
, "destroy-ipc-uid");
4291 void manager_serialize_gid_refs(Manager
*m
, FILE *f
) {
4292 manager_serialize_uid_refs_internal(m
, f
, &m
->gid_refs
, "destroy-ipc-gid");
4295 static void manager_deserialize_uid_refs_one_internal(
4298 const char *value
) {
4308 r
= parse_uid(value
, &uid
);
4309 if (r
< 0 || uid
== 0) {
4310 log_debug("Unable to parse UID reference serialization");
4314 r
= hashmap_ensure_allocated(uid_refs
, &trivial_hash_ops
);
4320 c
= PTR_TO_UINT32(hashmap_get(*uid_refs
, UID_TO_PTR(uid
)));
4321 if (c
& DESTROY_IPC_FLAG
)
4324 c
|= DESTROY_IPC_FLAG
;
4326 r
= hashmap_replace(*uid_refs
, UID_TO_PTR(uid
), UINT32_TO_PTR(c
));
4328 log_debug("Failed to add UID reference entry");
4333 void manager_deserialize_uid_refs_one(Manager
*m
, const char *value
) {
4334 manager_deserialize_uid_refs_one_internal(m
, &m
->uid_refs
, value
);
4337 void manager_deserialize_gid_refs_one(Manager
*m
, const char *value
) {
4338 manager_deserialize_uid_refs_one_internal(m
, &m
->gid_refs
, value
);
4341 int manager_dispatch_user_lookup_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
4345 char unit_name
[UNIT_NAME_MAX
+1];
4348 Manager
*m
= userdata
;
4356 /* Invoked whenever a child process succeeded resolving its user/group to use and sent us the resulting UID/GID
4357 * in a datagram. We parse the datagram here and pass it off to the unit, so that it can add a reference to the
4358 * UID/GID so that it can destroy the UID/GID's IPC objects when the reference counter drops to 0. */
4360 l
= recv(fd
, &buffer
, sizeof(buffer
), MSG_DONTWAIT
);
4362 if (IN_SET(errno
, EINTR
, EAGAIN
))
4365 return log_error_errno(errno
, "Failed to read from user lookup fd: %m");
4368 if ((size_t) l
<= offsetof(struct buffer
, unit_name
)) {
4369 log_warning("Received too short user lookup message, ignoring.");
4373 if ((size_t) l
> offsetof(struct buffer
, unit_name
) + UNIT_NAME_MAX
) {
4374 log_warning("Received too long user lookup message, ignoring.");
4378 if (!uid_is_valid(buffer
.uid
) && !gid_is_valid(buffer
.gid
)) {
4379 log_warning("Got user lookup message with invalid UID/GID pair, ignoring.");
4383 n
= (size_t) l
- offsetof(struct buffer
, unit_name
);
4384 if (memchr(buffer
.unit_name
, 0, n
)) {
4385 log_warning("Received lookup message with embedded NUL character, ignoring.");
4389 buffer
.unit_name
[n
] = 0;
4390 u
= manager_get_unit(m
, buffer
.unit_name
);
4392 log_debug("Got user lookup message but unit doesn't exist, ignoring.");
4396 log_unit_debug(u
, "User lookup succeeded: uid=" UID_FMT
" gid=" GID_FMT
, buffer
.uid
, buffer
.gid
);
4398 unit_notify_user_lookup(u
, buffer
.uid
, buffer
.gid
);
4402 char *manager_taint_string(Manager
*m
) {
4403 _cleanup_free_
char *destination
= NULL
, *overflowuid
= NULL
, *overflowgid
= NULL
;
4407 /* Returns a "taint string", e.g. "local-hwclock:var-run-bad".
4408 * Only things that are detected at runtime should be tagged
4409 * here. For stuff that is set during compilation, emit a warning
4410 * in the configuration phase. */
4414 buf
= new(char, sizeof("split-usr:"
4418 "overflowuid-not-65534:"
4419 "overflowgid-not-65534:"));
4427 e
= stpcpy(e
, "split-usr:");
4429 if (access("/proc/cgroups", F_OK
) < 0)
4430 e
= stpcpy(e
, "cgroups-missing:");
4432 if (clock_is_localtime(NULL
) > 0)
4433 e
= stpcpy(e
, "local-hwclock:");
4435 r
= readlink_malloc("/var/run", &destination
);
4436 if (r
< 0 || !PATH_IN_SET(destination
, "../run", "/run"))
4437 e
= stpcpy(e
, "var-run-bad:");
4439 r
= read_one_line_file("/proc/sys/kernel/overflowuid", &overflowuid
);
4440 if (r
>= 0 && !streq(overflowuid
, "65534"))
4441 e
= stpcpy(e
, "overflowuid-not-65534:");
4443 r
= read_one_line_file("/proc/sys/kernel/overflowgid", &overflowgid
);
4444 if (r
>= 0 && !streq(overflowgid
, "65534"))
4445 e
= stpcpy(e
, "overflowgid-not-65534:");
4447 /* remove the last ':' */
4454 void manager_ref_console(Manager
*m
) {
4460 void manager_unref_console(Manager
*m
) {
4462 assert(m
->n_on_console
> 0);
4465 if (m
->n_on_console
== 0)
4466 m
->no_console_output
= false; /* unset no_console_output flag, since the console is definitely free now */
4469 void manager_override_log_level(Manager
*m
, int level
) {
4470 _cleanup_free_
char *s
= NULL
;
4473 if (!m
->log_level_overridden
) {
4474 m
->original_log_level
= log_get_max_level();
4475 m
->log_level_overridden
= true;
4478 (void) log_level_to_string_alloc(level
, &s
);
4479 log_info("Setting log level to %s.", strna(s
));
4481 log_set_max_level(level
);
4484 void manager_restore_original_log_level(Manager
*m
) {
4485 _cleanup_free_
char *s
= NULL
;
4488 if (!m
->log_level_overridden
)
4491 (void) log_level_to_string_alloc(m
->original_log_level
, &s
);
4492 log_info("Restoring log level to original (%s).", strna(s
));
4494 log_set_max_level(m
->original_log_level
);
4495 m
->log_level_overridden
= false;
4498 void manager_override_log_target(Manager
*m
, LogTarget target
) {
4501 if (!m
->log_target_overridden
) {
4502 m
->original_log_target
= log_get_target();
4503 m
->log_target_overridden
= true;
4506 log_info("Setting log target to %s.", log_target_to_string(target
));
4507 log_set_target(target
);
4510 void manager_restore_original_log_target(Manager
*m
) {
4513 if (!m
->log_target_overridden
)
4516 log_info("Restoring log target to original %s.", log_target_to_string(m
->original_log_target
));
4518 log_set_target(m
->original_log_target
);
4519 m
->log_target_overridden
= false;
4522 static const char *const manager_state_table
[_MANAGER_STATE_MAX
] = {
4523 [MANAGER_INITIALIZING
] = "initializing",
4524 [MANAGER_STARTING
] = "starting",
4525 [MANAGER_RUNNING
] = "running",
4526 [MANAGER_DEGRADED
] = "degraded",
4527 [MANAGER_MAINTENANCE
] = "maintenance",
4528 [MANAGER_STOPPING
] = "stopping",
4531 DEFINE_STRING_TABLE_LOOKUP(manager_state
, ManagerState
);
4533 static const char *const manager_timestamp_table
[_MANAGER_TIMESTAMP_MAX
] = {
4534 [MANAGER_TIMESTAMP_FIRMWARE
] = "firmware",
4535 [MANAGER_TIMESTAMP_LOADER
] = "loader",
4536 [MANAGER_TIMESTAMP_KERNEL
] = "kernel",
4537 [MANAGER_TIMESTAMP_INITRD
] = "initrd",
4538 [MANAGER_TIMESTAMP_USERSPACE
] = "userspace",
4539 [MANAGER_TIMESTAMP_FINISH
] = "finish",
4540 [MANAGER_TIMESTAMP_SECURITY_START
] = "security-start",
4541 [MANAGER_TIMESTAMP_SECURITY_FINISH
] = "security-finish",
4542 [MANAGER_TIMESTAMP_GENERATORS_START
] = "generators-start",
4543 [MANAGER_TIMESTAMP_GENERATORS_FINISH
] = "generators-finish",
4544 [MANAGER_TIMESTAMP_UNITS_LOAD_START
] = "units-load-start",
4545 [MANAGER_TIMESTAMP_UNITS_LOAD_FINISH
] = "units-load-finish",
4548 DEFINE_STRING_TABLE_LOOKUP(manager_timestamp
, ManagerTimestamp
);