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 unsigned manager_dispatch_stop_when_unneeded_queue(Manager
*m
) {
1221 while ((u
= m
->stop_when_unneeded_queue
)) {
1222 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
1223 assert(m
->stop_when_unneeded_queue
);
1225 assert(u
->in_stop_when_unneeded_queue
);
1226 LIST_REMOVE(stop_when_unneeded_queue
, m
->stop_when_unneeded_queue
, u
);
1227 u
->in_stop_when_unneeded_queue
= false;
1231 if (!unit_is_unneeded(u
))
1234 log_unit_debug(u
, "Unit is not needed anymore.");
1236 /* If stopping a unit fails continuously we might enter a stop loop here, hence stop acting on the
1237 * service being unnecessary after a while. */
1239 if (!ratelimit_below(&u
->auto_stop_ratelimit
)) {
1240 log_unit_warning(u
, "Unit not needed anymore, but not stopping since we tried this too often recently.");
1244 /* Ok, nobody needs us anymore. Sniff. Then let's commit suicide */
1245 r
= manager_add_job(u
->manager
, JOB_STOP
, u
, JOB_FAIL
, &error
, NULL
);
1247 log_unit_warning_errno(u
, r
, "Failed to enqueue stop job, ignoring: %s", bus_error_message(&error
, r
));
1253 static void manager_clear_jobs_and_units(Manager
*m
) {
1258 while ((u
= hashmap_first(m
->units
)))
1261 manager_dispatch_cleanup_queue(m
);
1263 assert(!m
->load_queue
);
1264 assert(!m
->run_queue
);
1265 assert(!m
->dbus_unit_queue
);
1266 assert(!m
->dbus_job_queue
);
1267 assert(!m
->cleanup_queue
);
1268 assert(!m
->gc_unit_queue
);
1269 assert(!m
->gc_job_queue
);
1270 assert(!m
->stop_when_unneeded_queue
);
1272 assert(hashmap_isempty(m
->jobs
));
1273 assert(hashmap_isempty(m
->units
));
1275 m
->n_on_console
= 0;
1276 m
->n_running_jobs
= 0;
1277 m
->n_installed_jobs
= 0;
1278 m
->n_failed_jobs
= 0;
1281 Manager
* manager_free(Manager
*m
) {
1283 ExecDirectoryType dt
;
1288 manager_clear_jobs_and_units(m
);
1290 for (c
= 0; c
< _UNIT_TYPE_MAX
; c
++)
1291 if (unit_vtable
[c
]->shutdown
)
1292 unit_vtable
[c
]->shutdown(m
);
1294 /* If we reexecute ourselves, we keep the root cgroup around */
1295 manager_shutdown_cgroup(m
, m
->exit_code
!= MANAGER_REEXECUTE
);
1297 lookup_paths_flush_generator(&m
->lookup_paths
);
1301 exec_runtime_vacuum(m
);
1302 hashmap_free(m
->exec_runtime_by_id
);
1304 dynamic_user_vacuum(m
, false);
1305 hashmap_free(m
->dynamic_users
);
1307 hashmap_free(m
->units
);
1308 hashmap_free(m
->units_by_invocation_id
);
1309 hashmap_free(m
->jobs
);
1310 hashmap_free(m
->watch_pids
);
1311 hashmap_free(m
->watch_bus
);
1313 set_free(m
->startup_units
);
1314 set_free(m
->failed_units
);
1316 sd_event_source_unref(m
->signal_event_source
);
1317 sd_event_source_unref(m
->sigchld_event_source
);
1318 sd_event_source_unref(m
->notify_event_source
);
1319 sd_event_source_unref(m
->cgroups_agent_event_source
);
1320 sd_event_source_unref(m
->time_change_event_source
);
1321 sd_event_source_unref(m
->timezone_change_event_source
);
1322 sd_event_source_unref(m
->jobs_in_progress_event_source
);
1323 sd_event_source_unref(m
->run_queue_event_source
);
1324 sd_event_source_unref(m
->user_lookup_event_source
);
1325 sd_event_source_unref(m
->sync_bus_names_event_source
);
1327 safe_close(m
->signal_fd
);
1328 safe_close(m
->notify_fd
);
1329 safe_close(m
->cgroups_agent_fd
);
1330 safe_close(m
->time_change_fd
);
1331 safe_close_pair(m
->user_lookup_fds
);
1333 manager_close_ask_password(m
);
1335 manager_close_idle_pipe(m
);
1337 udev_unref(m
->udev
);
1338 sd_event_unref(m
->event
);
1340 free(m
->notify_socket
);
1342 lookup_paths_free(&m
->lookup_paths
);
1343 strv_free(m
->environment
);
1345 hashmap_free(m
->cgroup_unit
);
1346 set_free_free(m
->unit_path_cache
);
1348 free(m
->switch_root
);
1349 free(m
->switch_root_init
);
1351 rlimit_free_all(m
->rlimit
);
1353 assert(hashmap_isempty(m
->units_requiring_mounts_for
));
1354 hashmap_free(m
->units_requiring_mounts_for
);
1356 hashmap_free(m
->uid_refs
);
1357 hashmap_free(m
->gid_refs
);
1359 for (dt
= 0; dt
< _EXEC_DIRECTORY_TYPE_MAX
; dt
++)
1360 m
->prefix
[dt
] = mfree(m
->prefix
[dt
]);
1365 static void manager_enumerate_perpetual(Manager
*m
) {
1370 /* Let's ask every type to load all units from disk/kernel that it might know */
1371 for (c
= 0; c
< _UNIT_TYPE_MAX
; c
++) {
1372 if (!unit_type_supported(c
)) {
1373 log_debug("Unit type .%s is not supported on this system.", unit_type_to_string(c
));
1377 if (unit_vtable
[c
]->enumerate_perpetual
)
1378 unit_vtable
[c
]->enumerate_perpetual(m
);
1382 static void manager_enumerate(Manager
*m
) {
1387 /* Let's ask every type to load all units from disk/kernel that it might know */
1388 for (c
= 0; c
< _UNIT_TYPE_MAX
; c
++) {
1389 if (!unit_type_supported(c
)) {
1390 log_debug("Unit type .%s is not supported on this system.", unit_type_to_string(c
));
1394 if (unit_vtable
[c
]->enumerate
)
1395 unit_vtable
[c
]->enumerate(m
);
1398 manager_dispatch_load_queue(m
);
1401 static void manager_coldplug(Manager
*m
) {
1409 log_debug("Invoking unit coldplug() handlers…");
1411 /* Let's place the units back into their deserialized state */
1412 HASHMAP_FOREACH_KEY(u
, k
, m
->units
, i
) {
1414 /* ignore aliases */
1418 r
= unit_coldplug(u
);
1420 log_warning_errno(r
, "We couldn't coldplug %s, proceeding anyway: %m", u
->id
);
1424 static void manager_catchup(Manager
*m
) {
1431 log_debug("Invoking unit catchup() handlers…");
1433 /* Let's catch up on any state changes that happened while we were reloading/reexecing */
1434 HASHMAP_FOREACH_KEY(u
, k
, m
->units
, i
) {
1436 /* ignore aliases */
1444 static void manager_build_unit_path_cache(Manager
*m
) {
1450 set_free_free(m
->unit_path_cache
);
1452 m
->unit_path_cache
= set_new(&path_hash_ops
);
1453 if (!m
->unit_path_cache
) {
1458 /* This simply builds a list of files we know exist, so that
1459 * we don't always have to go to disk */
1461 STRV_FOREACH(i
, m
->lookup_paths
.search_path
) {
1462 _cleanup_closedir_
DIR *d
= NULL
;
1467 if (errno
!= ENOENT
)
1468 log_warning_errno(errno
, "Failed to open directory %s, ignoring: %m", *i
);
1472 FOREACH_DIRENT(de
, d
, r
= -errno
; goto fail
) {
1475 p
= strjoin(streq(*i
, "/") ? "" : *i
, "/", de
->d_name
);
1481 r
= set_consume(m
->unit_path_cache
, p
);
1490 log_warning_errno(r
, "Failed to build unit path cache, proceeding without: %m");
1491 m
->unit_path_cache
= set_free_free(m
->unit_path_cache
);
1494 static void manager_distribute_fds(Manager
*m
, FDSet
*fds
) {
1500 HASHMAP_FOREACH(u
, m
->units
, i
) {
1502 if (fdset_size(fds
) <= 0)
1505 if (!UNIT_VTABLE(u
)->distribute_fds
)
1508 UNIT_VTABLE(u
)->distribute_fds(u
, fds
);
1512 static bool manager_dbus_is_running(Manager
*m
, bool deserialized
) {
1517 /* This checks whether the dbus instance we are supposed to expose our APIs on is up. We check both the socket
1518 * and the service unit. If the 'deserialized' parameter is true we'll check the deserialized state of the unit
1519 * rather than the current one. */
1521 if (m
->test_run_flags
!= 0)
1524 u
= manager_get_unit(m
, SPECIAL_DBUS_SOCKET
);
1527 if ((deserialized
? SOCKET(u
)->deserialized_state
: SOCKET(u
)->state
) != SOCKET_RUNNING
)
1530 u
= manager_get_unit(m
, SPECIAL_DBUS_SERVICE
);
1533 if (!IN_SET((deserialized
? SERVICE(u
)->deserialized_state
: SERVICE(u
)->state
), SERVICE_RUNNING
, SERVICE_RELOAD
))
1539 static void manager_setup_bus(Manager
*m
) {
1542 /* Let's set up our private bus connection now, unconditionally */
1543 (void) bus_init_private(m
);
1545 /* If we are in --user mode also connect to the system bus now */
1546 if (MANAGER_IS_USER(m
))
1547 (void) bus_init_system(m
);
1549 /* Let's connect to the bus now, but only if the unit is supposed to be up */
1550 if (manager_dbus_is_running(m
, MANAGER_IS_RELOADING(m
))) {
1551 (void) bus_init_api(m
);
1553 if (MANAGER_IS_SYSTEM(m
))
1554 (void) bus_init_system(m
);
1558 static void manager_preset_all(Manager
*m
) {
1563 if (m
->first_boot
<= 0)
1566 if (!MANAGER_IS_SYSTEM(m
))
1569 if (m
->test_run_flags
!= 0)
1572 /* If this is the first boot, and we are in the host system, then preset everything */
1573 r
= unit_file_preset_all(UNIT_FILE_SYSTEM
, 0, NULL
, UNIT_FILE_PRESET_ENABLE_ONLY
, NULL
, 0);
1575 log_full_errno(r
== -EEXIST
? LOG_NOTICE
: LOG_WARNING
, r
,
1576 "Failed to populate /etc with preset unit settings, ignoring: %m");
1578 log_info("Populated /etc with preset unit settings.");
1581 int manager_startup(Manager
*m
, FILE *serialization
, FDSet
*fds
) {
1586 /* If we are running in test mode, we still want to run the generators,
1587 * but we should not touch the real generator directories. */
1588 r
= lookup_paths_init(&m
->lookup_paths
, m
->unit_file_scope
,
1589 m
->test_run_flags
? LOOKUP_PATHS_TEMPORARY_GENERATED
: 0,
1594 r
= manager_run_environment_generators(m
);
1598 dual_timestamp_get(m
->timestamps
+ manager_timestamp_initrd_mangle(MANAGER_TIMESTAMP_GENERATORS_START
));
1599 r
= manager_run_generators(m
);
1600 dual_timestamp_get(m
->timestamps
+ manager_timestamp_initrd_mangle(MANAGER_TIMESTAMP_GENERATORS_FINISH
));
1604 manager_preset_all(m
);
1605 lookup_paths_reduce(&m
->lookup_paths
);
1606 manager_build_unit_path_cache(m
);
1608 /* If we will deserialize make sure that during enumeration
1609 * this is already known, so we increase the counter here
1614 /* First, enumerate what we can from all config files */
1615 dual_timestamp_get(m
->timestamps
+ manager_timestamp_initrd_mangle(MANAGER_TIMESTAMP_UNITS_LOAD_START
));
1616 manager_enumerate_perpetual(m
);
1617 manager_enumerate(m
);
1618 dual_timestamp_get(m
->timestamps
+ manager_timestamp_initrd_mangle(MANAGER_TIMESTAMP_UNITS_LOAD_FINISH
));
1620 /* Second, deserialize if there is something to deserialize */
1621 if (serialization
) {
1622 r
= manager_deserialize(m
, serialization
, fds
);
1624 return log_error_errno(r
, "Deserialization failed: %m");
1627 /* Any fds left? Find some unit which wants them. This is
1628 * useful to allow container managers to pass some file
1629 * descriptors to us pre-initialized. This enables
1630 * socket-based activation of entire containers. */
1631 manager_distribute_fds(m
, fds
);
1633 /* We might have deserialized the notify fd, but if we didn't
1634 * then let's create the bus now */
1635 r
= manager_setup_notify(m
);
1637 /* No sense to continue without notifications, our children would fail anyway. */
1640 r
= manager_setup_cgroups_agent(m
);
1642 /* Likewise, no sense to continue without empty cgroup notifications. */
1645 r
= manager_setup_user_lookup_fd(m
);
1647 /* This shouldn't fail, except if things are really broken. */
1650 /* Connect to the bus if we are good for it */
1651 manager_setup_bus(m
);
1653 /* Now that we are connected to all possible busses, let's deserialize who is tracking us. */
1654 (void) bus_track_coldplug(m
, &m
->subscribed
, false, m
->deserialized_subscribed
);
1655 m
->deserialized_subscribed
= strv_free(m
->deserialized_subscribed
);
1657 /* Third, fire things up! */
1658 manager_coldplug(m
);
1660 /* Release any dynamic users no longer referenced */
1661 dynamic_user_vacuum(m
, true);
1663 exec_runtime_vacuum(m
);
1665 /* Release any references to UIDs/GIDs no longer referenced, and destroy any IPC owned by them */
1666 manager_vacuum_uid_refs(m
);
1667 manager_vacuum_gid_refs(m
);
1669 if (serialization
) {
1670 assert(m
->n_reloading
> 0);
1673 /* Let's wait for the UnitNew/JobNew messages being
1674 * sent, before we notify that the reload is
1676 m
->send_reloading_done
= true;
1679 /* Let's finally catch up with any changes that took place while we were reloading/reexecing */
1685 int manager_add_job(Manager
*m
, JobType type
, Unit
*unit
, JobMode mode
, sd_bus_error
*e
, Job
**_ret
) {
1690 assert(type
< _JOB_TYPE_MAX
);
1692 assert(mode
< _JOB_MODE_MAX
);
1694 if (mode
== JOB_ISOLATE
&& type
!= JOB_START
)
1695 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
, "Isolate is only valid for start.");
1697 if (mode
== JOB_ISOLATE
&& !unit
->allow_isolate
)
1698 return sd_bus_error_setf(e
, BUS_ERROR_NO_ISOLATION
, "Operation refused, unit may not be isolated.");
1700 log_unit_debug(unit
, "Trying to enqueue job %s/%s/%s", unit
->id
, job_type_to_string(type
), job_mode_to_string(mode
));
1702 type
= job_type_collapse(type
, unit
);
1704 tr
= transaction_new(mode
== JOB_REPLACE_IRREVERSIBLY
);
1708 r
= transaction_add_job_and_dependencies(tr
, type
, unit
, NULL
, true, false,
1709 IN_SET(mode
, JOB_IGNORE_DEPENDENCIES
, JOB_IGNORE_REQUIREMENTS
),
1710 mode
== JOB_IGNORE_DEPENDENCIES
, e
);
1714 if (mode
== JOB_ISOLATE
) {
1715 r
= transaction_add_isolate_jobs(tr
, m
);
1720 r
= transaction_activate(tr
, m
, mode
, e
);
1724 log_unit_debug(unit
,
1725 "Enqueued job %s/%s as %u", unit
->id
,
1726 job_type_to_string(type
), (unsigned) tr
->anchor_job
->id
);
1729 *_ret
= tr
->anchor_job
;
1731 transaction_free(tr
);
1735 transaction_abort(tr
);
1736 transaction_free(tr
);
1740 int manager_add_job_by_name(Manager
*m
, JobType type
, const char *name
, JobMode mode
, sd_bus_error
*e
, Job
**ret
) {
1741 Unit
*unit
= NULL
; /* just to appease gcc, initialization is not really necessary */
1745 assert(type
< _JOB_TYPE_MAX
);
1747 assert(mode
< _JOB_MODE_MAX
);
1749 r
= manager_load_unit(m
, name
, NULL
, NULL
, &unit
);
1754 return manager_add_job(m
, type
, unit
, mode
, e
, ret
);
1757 int manager_add_job_by_name_and_warn(Manager
*m
, JobType type
, const char *name
, JobMode mode
, Job
**ret
) {
1758 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
1762 assert(type
< _JOB_TYPE_MAX
);
1764 assert(mode
< _JOB_MODE_MAX
);
1766 r
= manager_add_job_by_name(m
, type
, name
, mode
, &error
, ret
);
1768 return log_warning_errno(r
, "Failed to enqueue %s job for %s: %s", job_mode_to_string(mode
), name
, bus_error_message(&error
, r
));
1773 int manager_propagate_reload(Manager
*m
, Unit
*unit
, JobMode mode
, sd_bus_error
*e
) {
1779 assert(mode
< _JOB_MODE_MAX
);
1780 assert(mode
!= JOB_ISOLATE
); /* Isolate is only valid for start */
1782 tr
= transaction_new(mode
== JOB_REPLACE_IRREVERSIBLY
);
1786 /* We need an anchor job */
1787 r
= transaction_add_job_and_dependencies(tr
, JOB_NOP
, unit
, NULL
, false, false, true, true, e
);
1791 /* Failure in adding individual dependencies is ignored, so this always succeeds. */
1792 transaction_add_propagate_reload_jobs(tr
, unit
, tr
->anchor_job
, mode
== JOB_IGNORE_DEPENDENCIES
, e
);
1794 r
= transaction_activate(tr
, m
, mode
, e
);
1798 transaction_free(tr
);
1802 transaction_abort(tr
);
1803 transaction_free(tr
);
1807 Job
*manager_get_job(Manager
*m
, uint32_t id
) {
1810 return hashmap_get(m
->jobs
, UINT32_TO_PTR(id
));
1813 Unit
*manager_get_unit(Manager
*m
, const char *name
) {
1817 return hashmap_get(m
->units
, name
);
1820 static int manager_dispatch_target_deps_queue(Manager
*m
) {
1825 static const UnitDependency deps
[] = {
1834 while ((u
= m
->target_deps_queue
)) {
1835 assert(u
->in_target_deps_queue
);
1837 LIST_REMOVE(target_deps_queue
, u
->manager
->target_deps_queue
, u
);
1838 u
->in_target_deps_queue
= false;
1840 for (k
= 0; k
< ELEMENTSOF(deps
); k
++) {
1845 HASHMAP_FOREACH_KEY(v
, target
, u
->dependencies
[deps
[k
]], i
) {
1846 r
= unit_add_default_target_dependency(u
, target
);
1856 unsigned manager_dispatch_load_queue(Manager
*m
) {
1862 /* Make sure we are not run recursively */
1863 if (m
->dispatching_load_queue
)
1866 m
->dispatching_load_queue
= true;
1868 /* Dispatches the load queue. Takes a unit from the queue and
1869 * tries to load its data until the queue is empty */
1871 while ((u
= m
->load_queue
)) {
1872 assert(u
->in_load_queue
);
1878 m
->dispatching_load_queue
= false;
1880 /* Dispatch the units waiting for their target dependencies to be added now, as all targets that we know about
1881 * should be loaded and have aliases resolved */
1882 (void) manager_dispatch_target_deps_queue(m
);
1887 int manager_load_unit_prepare(
1894 _cleanup_(unit_freep
) Unit
*cleanup_ret
= NULL
;
1900 assert(name
|| path
);
1903 /* This will prepare the unit for loading, but not actually
1904 * load anything from disk. */
1906 if (path
&& !is_path(path
))
1907 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
, "Path %s is not absolute.", path
);
1910 name
= basename(path
);
1912 t
= unit_name_to_type(name
);
1914 if (t
== _UNIT_TYPE_INVALID
|| !unit_name_is_valid(name
, UNIT_NAME_PLAIN
|UNIT_NAME_INSTANCE
)) {
1915 if (unit_name_is_valid(name
, UNIT_NAME_TEMPLATE
))
1916 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
, "Unit name %s is missing the instance name.", name
);
1918 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
, "Unit name %s is not valid.", name
);
1921 ret
= manager_get_unit(m
, name
);
1927 ret
= cleanup_ret
= unit_new(m
, unit_vtable
[t
]->object_size
);
1932 ret
->fragment_path
= strdup(path
);
1933 if (!ret
->fragment_path
)
1937 r
= unit_add_name(ret
, name
);
1941 unit_add_to_load_queue(ret
);
1942 unit_add_to_dbus_queue(ret
);
1943 unit_add_to_gc_queue(ret
);
1951 int manager_load_unit(
1963 /* This will load the service information files, but not actually
1964 * start any services or anything. */
1966 r
= manager_load_unit_prepare(m
, name
, path
, e
, _ret
);
1970 manager_dispatch_load_queue(m
);
1972 *_ret
= unit_follow_merge(*_ret
);
1976 int manager_load_startable_unit_or_warn(
1982 /* Load a unit, make sure it loaded fully and is not masked. */
1984 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
1988 r
= manager_load_unit(m
, name
, path
, &error
, &unit
);
1990 return log_error_errno(r
, "Failed to load %s %s: %s",
1991 name
? "unit" : "unit file", name
?: path
,
1992 bus_error_message(&error
, r
));
1994 r
= bus_unit_validate_load_state(unit
, &error
);
1996 return log_error_errno(r
, "%s", bus_error_message(&error
, r
));
2002 void manager_dump_jobs(Manager
*s
, FILE *f
, const char *prefix
) {
2009 HASHMAP_FOREACH(j
, s
->jobs
, i
)
2010 job_dump(j
, f
, prefix
);
2013 void manager_dump_units(Manager
*s
, FILE *f
, const char *prefix
) {
2021 HASHMAP_FOREACH_KEY(u
, t
, s
->units
, i
)
2023 unit_dump(u
, f
, prefix
);
2026 void manager_dump(Manager
*m
, FILE *f
, const char *prefix
) {
2032 for (q
= 0; q
< _MANAGER_TIMESTAMP_MAX
; q
++) {
2033 char buf
[FORMAT_TIMESTAMP_MAX
];
2035 if (dual_timestamp_is_set(m
->timestamps
+ q
))
2036 fprintf(f
, "%sTimestamp %s: %s\n",
2038 manager_timestamp_to_string(q
),
2039 format_timestamp(buf
, sizeof(buf
), m
->timestamps
[q
].realtime
));
2042 manager_dump_units(m
, f
, prefix
);
2043 manager_dump_jobs(m
, f
, prefix
);
2046 int manager_get_dump_string(Manager
*m
, char **ret
) {
2047 _cleanup_free_
char *dump
= NULL
;
2048 _cleanup_fclose_
FILE *f
= NULL
;
2055 f
= open_memstream(&dump
, &size
);
2059 (void) __fsetlocking(f
, FSETLOCKING_BYCALLER
);
2061 manager_dump(m
, f
, NULL
);
2063 r
= fflush_and_check(f
);
2069 *ret
= TAKE_PTR(dump
);
2074 void manager_clear_jobs(Manager
*m
) {
2079 while ((j
= hashmap_first(m
->jobs
)))
2080 /* No need to recurse. We're cancelling all jobs. */
2081 job_finish_and_invalidate(j
, JOB_CANCELED
, false, false);
2084 static int manager_dispatch_run_queue(sd_event_source
*source
, void *userdata
) {
2085 Manager
*m
= userdata
;
2091 while ((j
= m
->run_queue
)) {
2092 assert(j
->installed
);
2093 assert(j
->in_run_queue
);
2095 job_run_and_invalidate(j
);
2098 if (m
->n_running_jobs
> 0)
2099 manager_watch_jobs_in_progress(m
);
2101 if (m
->n_on_console
> 0)
2102 manager_watch_idle_pipe(m
);
2107 static unsigned manager_dispatch_dbus_queue(Manager
*m
) {
2108 unsigned n
= 0, budget
;
2114 if (m
->dispatching_dbus_queue
)
2117 /* Anything to do at all? */
2118 if (!m
->dbus_unit_queue
&& !m
->dbus_job_queue
&& !m
->send_reloading_done
&& !m
->queued_message
)
2121 /* Do we have overly many messages queued at the moment? If so, let's not enqueue more on top, let's sit this
2122 * cycle out, and process things in a later cycle when the queues got a bit emptier. */
2123 if (manager_bus_n_queued_write(m
) > MANAGER_BUS_BUSY_THRESHOLD
)
2126 /* Only process a certain number of units/jobs per event loop iteration. Even if the bus queue wasn't overly
2127 * full before this call we shouldn't increase it in size too wildly in one step, and we shouldn't monopolize
2128 * CPU time with generating these messages. Note the difference in counting of this "budget" and the
2129 * "threshold" above: the "budget" is decreased only once per generated message, regardless how many
2130 * busses/direct connections it is enqueued on, while the "threshold" is applied to each queued instance of bus
2131 * message, i.e. if the same message is enqueued to five busses/direct connections it will be counted five
2132 * times. This difference in counting ("references" vs. "instances") is primarily a result of the fact that
2133 * it's easier to implement it this way, however it also reflects the thinking that the "threshold" should put
2134 * a limit on used queue memory, i.e. space, while the "budget" should put a limit on time. Also note that
2135 * the "threshold" is currently chosen much higher than the "budget". */
2136 budget
= MANAGER_BUS_MESSAGE_BUDGET
;
2138 m
->dispatching_dbus_queue
= true;
2140 while (budget
> 0 && (u
= m
->dbus_unit_queue
)) {
2142 assert(u
->in_dbus_queue
);
2144 bus_unit_send_change_signal(u
);
2148 while (budget
> 0 && (j
= m
->dbus_job_queue
)) {
2149 assert(j
->in_dbus_queue
);
2151 bus_job_send_change_signal(j
);
2155 m
->dispatching_dbus_queue
= false;
2157 if (budget
> 0 && m
->send_reloading_done
) {
2158 m
->send_reloading_done
= false;
2159 bus_manager_send_reloading(m
, false);
2163 if (budget
> 0 && m
->queued_message
) {
2164 bus_send_queued_message(m
);
2171 static int manager_dispatch_cgroups_agent_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
2172 Manager
*m
= userdata
;
2173 char buf
[PATH_MAX
+1];
2176 n
= recv(fd
, buf
, sizeof(buf
), 0);
2178 return log_error_errno(errno
, "Failed to read cgroups agent message: %m");
2180 log_error("Got zero-length cgroups agent message, ignoring.");
2183 if ((size_t) n
>= sizeof(buf
)) {
2184 log_error("Got overly long cgroups agent message, ignoring.");
2188 if (memchr(buf
, 0, n
)) {
2189 log_error("Got cgroups agent message with embedded NUL byte, ignoring.");
2194 manager_notify_cgroup_empty(m
, buf
);
2195 (void) bus_forward_agent_released(m
, buf
);
2200 static void manager_invoke_notify_message(
2203 const struct ucred
*ucred
,
2212 if (u
->notifygen
== m
->notifygen
) /* Already invoked on this same unit in this same iteration? */
2214 u
->notifygen
= m
->notifygen
;
2216 if (UNIT_VTABLE(u
)->notify_message
) {
2217 _cleanup_strv_free_
char **tags
= NULL
;
2219 tags
= strv_split(buf
, NEWLINE
);
2225 UNIT_VTABLE(u
)->notify_message(u
, ucred
, tags
, fds
);
2227 } else if (DEBUG_LOGGING
) {
2228 _cleanup_free_
char *x
= NULL
, *y
= NULL
;
2230 x
= ellipsize(buf
, 20, 90);
2234 log_unit_debug(u
, "Got notification message \"%s\", ignoring.", strnull(y
));
2238 static int manager_dispatch_notify_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
2240 _cleanup_fdset_free_ FDSet
*fds
= NULL
;
2241 Manager
*m
= userdata
;
2242 char buf
[NOTIFY_BUFFER_MAX
+1];
2243 struct iovec iovec
= {
2245 .iov_len
= sizeof(buf
)-1,
2248 struct cmsghdr cmsghdr
;
2249 uint8_t buf
[CMSG_SPACE(sizeof(struct ucred
)) +
2250 CMSG_SPACE(sizeof(int) * NOTIFY_FD_MAX
)];
2252 struct msghdr msghdr
= {
2255 .msg_control
= &control
,
2256 .msg_controllen
= sizeof(control
),
2259 struct cmsghdr
*cmsg
;
2260 struct ucred
*ucred
= NULL
;
2261 _cleanup_free_ Unit
**array_copy
= NULL
;
2262 Unit
*u1
, *u2
, **array
;
2263 int r
, *fd_array
= NULL
;
2269 assert(m
->notify_fd
== fd
);
2271 if (revents
!= EPOLLIN
) {
2272 log_warning("Got unexpected poll event for notify fd.");
2276 n
= recvmsg(m
->notify_fd
, &msghdr
, MSG_DONTWAIT
|MSG_CMSG_CLOEXEC
|MSG_TRUNC
);
2278 if (IN_SET(errno
, EAGAIN
, EINTR
))
2279 return 0; /* Spurious wakeup, try again */
2281 /* If this is any other, real error, then let's stop processing this socket. This of course means we
2282 * won't take notification messages anymore, but that's still better than busy looping around this:
2283 * being woken up over and over again but being unable to actually read the message off the socket. */
2284 return log_error_errno(errno
, "Failed to receive notification message: %m");
2287 CMSG_FOREACH(cmsg
, &msghdr
) {
2288 if (cmsg
->cmsg_level
== SOL_SOCKET
&& cmsg
->cmsg_type
== SCM_RIGHTS
) {
2290 fd_array
= (int*) CMSG_DATA(cmsg
);
2291 n_fds
= (cmsg
->cmsg_len
- CMSG_LEN(0)) / sizeof(int);
2293 } else if (cmsg
->cmsg_level
== SOL_SOCKET
&&
2294 cmsg
->cmsg_type
== SCM_CREDENTIALS
&&
2295 cmsg
->cmsg_len
== CMSG_LEN(sizeof(struct ucred
))) {
2297 ucred
= (struct ucred
*) CMSG_DATA(cmsg
);
2304 r
= fdset_new_array(&fds
, fd_array
, n_fds
);
2306 close_many(fd_array
, n_fds
);
2312 if (!ucred
|| !pid_is_valid(ucred
->pid
)) {
2313 log_warning("Received notify message without valid credentials. Ignoring.");
2317 if ((size_t) n
>= sizeof(buf
) || (msghdr
.msg_flags
& MSG_TRUNC
)) {
2318 log_warning("Received notify message exceeded maximum size. Ignoring.");
2322 /* As extra safety check, let's make sure the string we get doesn't contain embedded NUL bytes. We permit one
2323 * trailing NUL byte in the message, but don't expect it. */
2324 if (n
> 1 && memchr(buf
, 0, n
-1)) {
2325 log_warning("Received notify message with embedded NUL bytes. Ignoring.");
2329 /* Make sure it's NUL-terminated. */
2332 /* Increase the generation counter used for filtering out duplicate unit invocations. */
2335 /* Notify every unit that might be interested, which might be multiple. */
2336 u1
= manager_get_unit_by_pid_cgroup(m
, ucred
->pid
);
2337 u2
= hashmap_get(m
->watch_pids
, PID_TO_PTR(ucred
->pid
));
2338 array
= hashmap_get(m
->watch_pids
, PID_TO_PTR(-ucred
->pid
));
2345 array_copy
= newdup(Unit
*, array
, k
+1);
2349 /* And now invoke the per-unit callbacks. Note that manager_invoke_notify_message() will handle duplicate units
2350 * make sure we only invoke each unit's handler once. */
2352 manager_invoke_notify_message(m
, u1
, ucred
, buf
, fds
);
2356 manager_invoke_notify_message(m
, u2
, ucred
, buf
, fds
);
2360 for (size_t i
= 0; array_copy
[i
]; i
++) {
2361 manager_invoke_notify_message(m
, array_copy
[i
], ucred
, buf
, fds
);
2366 log_warning("Cannot find unit for notify message of PID "PID_FMT
", ignoring.", ucred
->pid
);
2368 if (fdset_size(fds
) > 0)
2369 log_warning("Got extra auxiliary fds with notification message, closing them.");
2374 static void manager_invoke_sigchld_event(
2377 const siginfo_t
*si
) {
2383 /* Already invoked the handler of this unit in this iteration? Then don't process this again */
2384 if (u
->sigchldgen
== m
->sigchldgen
)
2386 u
->sigchldgen
= m
->sigchldgen
;
2388 log_unit_debug(u
, "Child "PID_FMT
" belongs to %s.", si
->si_pid
, u
->id
);
2389 unit_unwatch_pid(u
, si
->si_pid
);
2391 if (UNIT_VTABLE(u
)->sigchld_event
)
2392 UNIT_VTABLE(u
)->sigchld_event(u
, si
->si_pid
, si
->si_code
, si
->si_status
);
2395 static int manager_dispatch_sigchld(sd_event_source
*source
, void *userdata
) {
2396 Manager
*m
= userdata
;
2403 /* First we call waitd() for a PID and do not reap the zombie. That way we can still access /proc/$PID for it
2404 * while it is a zombie. */
2406 if (waitid(P_ALL
, 0, &si
, WEXITED
|WNOHANG
|WNOWAIT
) < 0) {
2408 if (errno
!= ECHILD
)
2409 log_error_errno(errno
, "Failed to peek for child with waitid(), ignoring: %m");
2417 if (IN_SET(si
.si_code
, CLD_EXITED
, CLD_KILLED
, CLD_DUMPED
)) {
2418 _cleanup_free_ Unit
**array_copy
= NULL
;
2419 _cleanup_free_
char *name
= NULL
;
2420 Unit
*u1
, *u2
, **array
;
2422 (void) get_process_comm(si
.si_pid
, &name
);
2424 log_debug("Child "PID_FMT
" (%s) died (code=%s, status=%i/%s)",
2425 si
.si_pid
, strna(name
),
2426 sigchld_code_to_string(si
.si_code
),
2428 strna(si
.si_code
== CLD_EXITED
2429 ? exit_status_to_string(si
.si_status
, EXIT_STATUS_FULL
)
2430 : signal_to_string(si
.si_status
)));
2432 /* Increase the generation counter used for filtering out duplicate unit invocations */
2435 /* And now figure out the unit this belongs to, it might be multiple... */
2436 u1
= manager_get_unit_by_pid_cgroup(m
, si
.si_pid
);
2437 u2
= hashmap_get(m
->watch_pids
, PID_TO_PTR(si
.si_pid
));
2438 array
= hashmap_get(m
->watch_pids
, PID_TO_PTR(-si
.si_pid
));
2442 /* Cound how many entries the array has */
2446 /* Make a copy of the array so that we don't trip up on the array changing beneath us */
2447 array_copy
= newdup(Unit
*, array
, n
+1);
2452 /* Finally, execute them all. Note that u1, u2 and the array might contain duplicates, but
2453 * that's fine, manager_invoke_sigchld_event() will ensure we only invoke the handlers once for
2454 * each iteration. */
2456 manager_invoke_sigchld_event(m
, u1
, &si
);
2458 manager_invoke_sigchld_event(m
, u2
, &si
);
2460 for (size_t i
= 0; array_copy
[i
]; i
++)
2461 manager_invoke_sigchld_event(m
, array_copy
[i
], &si
);
2464 /* And now, we actually reap the zombie. */
2465 if (waitid(P_PID
, si
.si_pid
, &si
, WEXITED
) < 0) {
2466 log_error_errno(errno
, "Failed to dequeue child, ignoring: %m");
2473 /* All children processed for now, turn off event source */
2475 r
= sd_event_source_set_enabled(m
->sigchld_event_source
, SD_EVENT_OFF
);
2477 return log_error_errno(r
, "Failed to disable SIGCHLD event source: %m");
2482 static void manager_start_target(Manager
*m
, const char *name
, JobMode mode
) {
2483 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
2486 log_debug("Activating special unit %s", name
);
2488 r
= manager_add_job_by_name(m
, JOB_START
, name
, mode
, &error
, NULL
);
2490 log_error("Failed to enqueue %s job: %s", name
, bus_error_message(&error
, r
));
2493 static void manager_handle_ctrl_alt_del(Manager
*m
) {
2494 /* If the user presses C-A-D more than
2495 * 7 times within 2s, we reboot/shutdown immediately,
2496 * unless it was disabled in system.conf */
2498 if (ratelimit_below(&m
->ctrl_alt_del_ratelimit
) || m
->cad_burst_action
== EMERGENCY_ACTION_NONE
)
2499 manager_start_target(m
, SPECIAL_CTRL_ALT_DEL_TARGET
, JOB_REPLACE_IRREVERSIBLY
);
2501 emergency_action(m
, m
->cad_burst_action
, NULL
,
2502 "Ctrl-Alt-Del was pressed more than 7 times within 2s");
2505 static int manager_dispatch_signal_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
2506 Manager
*m
= userdata
;
2508 struct signalfd_siginfo sfsi
;
2512 assert(m
->signal_fd
== fd
);
2514 if (revents
!= EPOLLIN
) {
2515 log_warning("Got unexpected events from signal file descriptor.");
2519 n
= read(m
->signal_fd
, &sfsi
, sizeof(sfsi
));
2520 if (n
!= sizeof(sfsi
)) {
2522 log_warning("Truncated read from signal fd (%zu bytes), ignoring!", n
);
2526 if (IN_SET(errno
, EINTR
, EAGAIN
))
2529 /* We return an error here, which will kill this handler,
2530 * to avoid a busy loop on read error. */
2531 return log_error_errno(errno
, "Reading from signal fd failed: %m");
2534 log_received_signal(sfsi
.ssi_signo
== SIGCHLD
||
2535 (sfsi
.ssi_signo
== SIGTERM
&& MANAGER_IS_USER(m
))
2536 ? LOG_DEBUG
: LOG_INFO
,
2539 switch (sfsi
.ssi_signo
) {
2542 r
= sd_event_source_set_enabled(m
->sigchld_event_source
, SD_EVENT_ON
);
2544 log_warning_errno(r
, "Failed to enable SIGCHLD event source, ignoring: %m");
2549 if (MANAGER_IS_SYSTEM(m
)) {
2550 /* This is for compatibility with the original sysvinit */
2551 r
= verify_run_space_and_log("Refusing to reexecute");
2553 m
->exit_code
= MANAGER_REEXECUTE
;
2559 if (MANAGER_IS_SYSTEM(m
))
2560 manager_handle_ctrl_alt_del(m
);
2562 manager_start_target(m
, SPECIAL_EXIT_TARGET
,
2563 JOB_REPLACE_IRREVERSIBLY
);
2567 /* This is a nop on non-init */
2568 if (MANAGER_IS_SYSTEM(m
))
2569 manager_start_target(m
, SPECIAL_KBREQUEST_TARGET
, JOB_REPLACE
);
2574 /* This is a nop on non-init */
2575 if (MANAGER_IS_SYSTEM(m
))
2576 manager_start_target(m
, SPECIAL_SIGPWR_TARGET
, JOB_REPLACE
);
2581 if (manager_dbus_is_running(m
, false)) {
2582 log_info("Trying to reconnect to bus...");
2584 (void) bus_init_api(m
);
2586 if (MANAGER_IS_SYSTEM(m
))
2587 (void) bus_init_system(m
);
2589 log_info("Starting D-Bus service...");
2590 manager_start_target(m
, SPECIAL_DBUS_SERVICE
, JOB_REPLACE
);
2596 _cleanup_free_
char *dump
= NULL
;
2598 r
= manager_get_dump_string(m
, &dump
);
2600 log_warning_errno(errno
, "Failed to acquire manager dump: %m");
2604 log_dump(LOG_INFO
, dump
);
2609 r
= verify_run_space_and_log("Refusing to reload");
2611 m
->exit_code
= MANAGER_RELOAD
;
2616 /* Starting SIGRTMIN+0 */
2617 static const struct {
2620 } target_table
[] = {
2621 [0] = { SPECIAL_DEFAULT_TARGET
, JOB_ISOLATE
},
2622 [1] = { SPECIAL_RESCUE_TARGET
, JOB_ISOLATE
},
2623 [2] = { SPECIAL_EMERGENCY_TARGET
, JOB_ISOLATE
},
2624 [3] = { SPECIAL_HALT_TARGET
, JOB_REPLACE_IRREVERSIBLY
},
2625 [4] = { SPECIAL_POWEROFF_TARGET
, JOB_REPLACE_IRREVERSIBLY
},
2626 [5] = { SPECIAL_REBOOT_TARGET
, JOB_REPLACE_IRREVERSIBLY
},
2627 [6] = { SPECIAL_KEXEC_TARGET
, JOB_REPLACE_IRREVERSIBLY
},
2630 /* Starting SIGRTMIN+13, so that target halt and system halt are 10 apart */
2631 static const ManagerExitCode code_table
[] = {
2633 [1] = MANAGER_POWEROFF
,
2634 [2] = MANAGER_REBOOT
,
2635 [3] = MANAGER_KEXEC
,
2638 if ((int) sfsi
.ssi_signo
>= SIGRTMIN
+0 &&
2639 (int) sfsi
.ssi_signo
< SIGRTMIN
+(int) ELEMENTSOF(target_table
)) {
2640 int idx
= (int) sfsi
.ssi_signo
- SIGRTMIN
;
2641 manager_start_target(m
, target_table
[idx
].target
,
2642 target_table
[idx
].mode
);
2646 if ((int) sfsi
.ssi_signo
>= SIGRTMIN
+13 &&
2647 (int) sfsi
.ssi_signo
< SIGRTMIN
+13+(int) ELEMENTSOF(code_table
)) {
2648 m
->exit_code
= code_table
[sfsi
.ssi_signo
- SIGRTMIN
- 13];
2652 switch (sfsi
.ssi_signo
- SIGRTMIN
) {
2655 manager_set_show_status(m
, SHOW_STATUS_YES
);
2659 manager_set_show_status(m
, SHOW_STATUS_NO
);
2663 manager_override_log_level(m
, LOG_DEBUG
);
2667 manager_restore_original_log_level(m
);
2671 if (MANAGER_IS_USER(m
)) {
2672 m
->exit_code
= MANAGER_EXIT
;
2676 /* This is a nop on init */
2680 case 29: /* compatibility: used to be mapped to LOG_TARGET_SYSLOG_OR_KMSG */
2681 manager_restore_original_log_target(m
);
2685 manager_override_log_target(m
, LOG_TARGET_CONSOLE
);
2689 manager_override_log_target(m
, LOG_TARGET_KMSG
);
2693 log_warning("Got unhandled signal <%s>.", signal_to_string(sfsi
.ssi_signo
));
2700 static int manager_dispatch_time_change_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
2701 Manager
*m
= userdata
;
2706 assert(m
->time_change_fd
== fd
);
2708 log_struct(LOG_DEBUG
,
2709 "MESSAGE_ID=" SD_MESSAGE_TIME_CHANGE_STR
,
2710 LOG_MESSAGE("Time has been changed"));
2712 /* Restart the watch */
2713 (void) manager_setup_time_change(m
);
2715 HASHMAP_FOREACH(u
, m
->units
, i
)
2716 if (UNIT_VTABLE(u
)->time_change
)
2717 UNIT_VTABLE(u
)->time_change(u
);
2722 static int manager_dispatch_timezone_change(
2723 sd_event_source
*source
,
2724 const struct inotify_event
*e
,
2727 Manager
*m
= userdata
;
2734 log_debug("inotify event for /etc/localtime");
2736 changed
= manager_read_timezone_stat(m
);
2742 /* Something changed, restart the watch, to ensure we watch the new /etc/localtime if it changed */
2743 (void) manager_setup_timezone_change(m
);
2745 /* Read the new timezone */
2748 log_debug("Timezone has been changed (now: %s).", tzname
[daylight
]);
2750 HASHMAP_FOREACH(u
, m
->units
, i
)
2751 if (UNIT_VTABLE(u
)->timezone_change
)
2752 UNIT_VTABLE(u
)->timezone_change(u
);
2757 static int manager_dispatch_idle_pipe_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
2758 Manager
*m
= userdata
;
2761 assert(m
->idle_pipe
[2] == fd
);
2763 /* There's at least one Type=idle child that just gave up on us waiting for the boot process to complete. Let's
2764 * now turn off any further console output if there's at least one service that needs console access, so that
2765 * from now on our own output should not spill into that service's output anymore. After all, we support
2766 * Type=idle only to beautify console output and it generally is set on services that want to own the console
2767 * exclusively without our interference. */
2768 m
->no_console_output
= m
->n_on_console
> 0;
2770 /* Acknowledge the child's request, and let all all other children know too that they shouldn't wait any longer
2771 * by closing the pipes towards them, which is what they are waiting for. */
2772 manager_close_idle_pipe(m
);
2777 static int manager_dispatch_jobs_in_progress(sd_event_source
*source
, usec_t usec
, void *userdata
) {
2778 Manager
*m
= userdata
;
2785 manager_print_jobs_in_progress(m
);
2787 next
= now(CLOCK_MONOTONIC
) + JOBS_IN_PROGRESS_PERIOD_USEC
;
2788 r
= sd_event_source_set_time(source
, next
);
2792 return sd_event_source_set_enabled(source
, SD_EVENT_ONESHOT
);
2795 int manager_loop(Manager
*m
) {
2798 RATELIMIT_DEFINE(rl
, 1*USEC_PER_SEC
, 50000);
2801 m
->exit_code
= MANAGER_OK
;
2803 /* Release the path cache */
2804 m
->unit_path_cache
= set_free_free(m
->unit_path_cache
);
2806 manager_check_finished(m
);
2808 /* There might still be some zombies hanging around from before we were exec()'ed. Let's reap them. */
2809 r
= sd_event_source_set_enabled(m
->sigchld_event_source
, SD_EVENT_ON
);
2811 return log_error_errno(r
, "Failed to enable SIGCHLD event source: %m");
2813 while (m
->exit_code
== MANAGER_OK
) {
2816 if (m
->runtime_watchdog
> 0 && m
->runtime_watchdog
!= USEC_INFINITY
&& MANAGER_IS_SYSTEM(m
))
2819 if (!ratelimit_below(&rl
)) {
2820 /* Yay, something is going seriously wrong, pause a little */
2821 log_warning("Looping too fast. Throttling execution a little.");
2825 if (manager_dispatch_load_queue(m
) > 0)
2828 if (manager_dispatch_gc_job_queue(m
) > 0)
2831 if (manager_dispatch_gc_unit_queue(m
) > 0)
2834 if (manager_dispatch_cleanup_queue(m
) > 0)
2837 if (manager_dispatch_cgroup_realize_queue(m
) > 0)
2840 if (manager_dispatch_stop_when_unneeded_queue(m
) > 0)
2843 if (manager_dispatch_dbus_queue(m
) > 0)
2846 /* Sleep for half the watchdog time */
2847 if (m
->runtime_watchdog
> 0 && m
->runtime_watchdog
!= USEC_INFINITY
&& MANAGER_IS_SYSTEM(m
)) {
2848 wait_usec
= m
->runtime_watchdog
/ 2;
2852 wait_usec
= USEC_INFINITY
;
2854 r
= sd_event_run(m
->event
, wait_usec
);
2856 return log_error_errno(r
, "Failed to run event loop: %m");
2859 return m
->exit_code
;
2862 int manager_load_unit_from_dbus_path(Manager
*m
, const char *s
, sd_bus_error
*e
, Unit
**_u
) {
2863 _cleanup_free_
char *n
= NULL
;
2864 sd_id128_t invocation_id
;
2872 r
= unit_name_from_dbus_path(s
, &n
);
2876 /* Permit addressing units by invocation ID: if the passed bus path is suffixed by a 128bit ID then we use it
2877 * as invocation ID. */
2878 r
= sd_id128_from_string(n
, &invocation_id
);
2880 u
= hashmap_get(m
->units_by_invocation_id
, &invocation_id
);
2886 return sd_bus_error_setf(e
, BUS_ERROR_NO_UNIT_FOR_INVOCATION_ID
,
2887 "No unit with the specified invocation ID " SD_ID128_FORMAT_STR
" known.",
2888 SD_ID128_FORMAT_VAL(invocation_id
));
2891 /* If this didn't work, we check if this is a unit name */
2892 if (!unit_name_is_valid(n
, UNIT_NAME_PLAIN
|UNIT_NAME_INSTANCE
)) {
2893 _cleanup_free_
char *nn
= NULL
;
2896 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
,
2897 "Unit name %s is neither a valid invocation ID nor unit name.", strnull(nn
));
2900 r
= manager_load_unit(m
, n
, NULL
, e
, &u
);
2908 int manager_get_job_from_dbus_path(Manager
*m
, const char *s
, Job
**_j
) {
2918 p
= startswith(s
, "/org/freedesktop/systemd1/job/");
2922 r
= safe_atou(p
, &id
);
2926 j
= manager_get_job(m
, id
);
2935 void manager_send_unit_audit(Manager
*m
, Unit
*u
, int type
, bool success
) {
2938 _cleanup_free_
char *p
= NULL
;
2942 if (!MANAGER_IS_SYSTEM(m
))
2945 audit_fd
= get_audit_fd();
2949 /* Don't generate audit events if the service was already
2950 * started and we're just deserializing */
2951 if (MANAGER_IS_RELOADING(m
))
2954 if (u
->type
!= UNIT_SERVICE
)
2957 r
= unit_name_to_prefix_and_instance(u
->id
, &p
);
2959 log_error_errno(r
, "Failed to extract prefix and instance of unit name: %m");
2963 msg
= strjoina("unit=", p
);
2964 if (audit_log_user_comm_message(audit_fd
, type
, msg
, "systemd", NULL
, NULL
, NULL
, success
) < 0) {
2966 /* We aren't allowed to send audit messages?
2967 * Then let's not retry again. */
2970 log_warning_errno(errno
, "Failed to send audit message: %m");
2976 void manager_send_unit_plymouth(Manager
*m
, Unit
*u
) {
2977 static const union sockaddr_union sa
= PLYMOUTH_SOCKET
;
2978 _cleanup_free_
char *message
= NULL
;
2979 _cleanup_close_
int fd
= -1;
2982 /* Don't generate plymouth events if the service was already
2983 * started and we're just deserializing */
2984 if (MANAGER_IS_RELOADING(m
))
2987 if (!MANAGER_IS_SYSTEM(m
))
2990 if (detect_container() > 0)
2993 if (!IN_SET(u
->type
, UNIT_SERVICE
, UNIT_MOUNT
, UNIT_SWAP
))
2996 /* We set SOCK_NONBLOCK here so that we rather drop the
2997 * message then wait for plymouth */
2998 fd
= socket(AF_UNIX
, SOCK_STREAM
|SOCK_CLOEXEC
|SOCK_NONBLOCK
, 0);
3000 log_error_errno(errno
, "socket() failed: %m");
3004 if (connect(fd
, &sa
.sa
, SOCKADDR_UN_LEN(sa
.un
)) < 0) {
3005 if (!IN_SET(errno
, EPIPE
, EAGAIN
, ENOENT
, ECONNREFUSED
, ECONNRESET
, ECONNABORTED
))
3006 log_error_errno(errno
, "connect() failed: %m");
3010 if (asprintf(&message
, "U\002%c%s%n", (int) (strlen(u
->id
) + 1), u
->id
, &n
) < 0) {
3016 if (write(fd
, message
, n
+ 1) != n
+ 1)
3017 if (!IN_SET(errno
, EPIPE
, EAGAIN
, ENOENT
, ECONNREFUSED
, ECONNRESET
, ECONNABORTED
))
3018 log_error_errno(errno
, "Failed to write Plymouth message: %m");
3021 int manager_open_serialization(Manager
*m
, FILE **_f
) {
3027 fd
= open_serialization_fd("systemd-state");
3031 f
= fdopen(fd
, "w+");
3041 int manager_serialize(Manager
*m
, FILE *f
, FDSet
*fds
, bool switching_root
) {
3054 fprintf(f
, "current-job-id=%"PRIu32
"\n", m
->current_job_id
);
3055 fprintf(f
, "n-installed-jobs=%u\n", m
->n_installed_jobs
);
3056 fprintf(f
, "n-failed-jobs=%u\n", m
->n_failed_jobs
);
3057 fprintf(f
, "taint-usr=%s\n", yes_no(m
->taint_usr
));
3058 fprintf(f
, "ready-sent=%s\n", yes_no(m
->ready_sent
));
3059 fprintf(f
, "taint-logged=%s\n", yes_no(m
->taint_logged
));
3060 fprintf(f
, "service-watchdogs=%s\n", yes_no(m
->service_watchdogs
));
3062 t
= show_status_to_string(m
->show_status
);
3064 fprintf(f
, "show-status=%s\n", t
);
3066 if (m
->log_level_overridden
)
3067 fprintf(f
, "log-level-override=%i\n", log_get_max_level());
3068 if (m
->log_target_overridden
)
3069 fprintf(f
, "log-target-override=%s\n", log_target_to_string(log_get_target()));
3071 for (q
= 0; q
< _MANAGER_TIMESTAMP_MAX
; q
++) {
3072 /* The following timestamps only apply to the host system, hence only serialize them there */
3074 IN_SET(q
, MANAGER_TIMESTAMP_USERSPACE
, MANAGER_TIMESTAMP_FINISH
,
3075 MANAGER_TIMESTAMP_SECURITY_START
, MANAGER_TIMESTAMP_SECURITY_FINISH
,
3076 MANAGER_TIMESTAMP_GENERATORS_START
, MANAGER_TIMESTAMP_GENERATORS_FINISH
,
3077 MANAGER_TIMESTAMP_UNITS_LOAD_START
, MANAGER_TIMESTAMP_UNITS_LOAD_FINISH
))
3080 t
= manager_timestamp_to_string(q
);
3082 char field
[strlen(t
) + STRLEN("-timestamp") + 1];
3083 strcpy(stpcpy(field
, t
), "-timestamp");
3084 dual_timestamp_serialize(f
, field
, m
->timestamps
+ q
);
3088 if (!switching_root
)
3089 (void) serialize_environment(f
, m
->environment
);
3091 if (m
->notify_fd
>= 0) {
3094 copy
= fdset_put_dup(fds
, m
->notify_fd
);
3098 fprintf(f
, "notify-fd=%i\n", copy
);
3099 fprintf(f
, "notify-socket=%s\n", m
->notify_socket
);
3102 if (m
->cgroups_agent_fd
>= 0) {
3105 copy
= fdset_put_dup(fds
, m
->cgroups_agent_fd
);
3109 fprintf(f
, "cgroups-agent-fd=%i\n", copy
);
3112 if (m
->user_lookup_fds
[0] >= 0) {
3115 copy0
= fdset_put_dup(fds
, m
->user_lookup_fds
[0]);
3119 copy1
= fdset_put_dup(fds
, m
->user_lookup_fds
[1]);
3123 fprintf(f
, "user-lookup=%i %i\n", copy0
, copy1
);
3126 bus_track_serialize(m
->subscribed
, f
, "subscribed");
3128 r
= dynamic_user_serialize(m
, f
, fds
);
3132 manager_serialize_uid_refs(m
, f
);
3133 manager_serialize_gid_refs(m
, f
);
3135 r
= exec_runtime_serialize(m
, f
, fds
);
3139 (void) fputc('\n', f
);
3141 HASHMAP_FOREACH_KEY(u
, t
, m
->units
, i
) {
3149 r
= unit_serialize(u
, f
, fds
, !switching_root
);
3156 assert(m
->n_reloading
> 0);
3159 r
= fflush_and_check(f
);
3163 r
= bus_fdset_add_all(m
, fds
);
3170 int manager_deserialize(Manager
*m
, FILE *f
, FDSet
*fds
) {
3176 log_debug("Deserializing state...");
3181 char line
[LINE_MAX
];
3182 const char *val
, *l
;
3184 if (!fgets(line
, sizeof(line
), f
)) {
3199 if ((val
= startswith(l
, "current-job-id="))) {
3202 if (safe_atou32(val
, &id
) < 0)
3203 log_notice("Failed to parse current job id value %s", val
);
3205 m
->current_job_id
= MAX(m
->current_job_id
, id
);
3207 } else if ((val
= startswith(l
, "n-installed-jobs="))) {
3210 if (safe_atou32(val
, &n
) < 0)
3211 log_notice("Failed to parse installed jobs counter %s", val
);
3213 m
->n_installed_jobs
+= n
;
3215 } else if ((val
= startswith(l
, "n-failed-jobs="))) {
3218 if (safe_atou32(val
, &n
) < 0)
3219 log_notice("Failed to parse failed jobs counter %s", val
);
3221 m
->n_failed_jobs
+= n
;
3223 } else if ((val
= startswith(l
, "taint-usr="))) {
3226 b
= parse_boolean(val
);
3228 log_notice("Failed to parse taint /usr flag %s", val
);
3230 m
->taint_usr
= m
->taint_usr
|| b
;
3232 } else if ((val
= startswith(l
, "ready-sent="))) {
3235 b
= parse_boolean(val
);
3237 log_notice("Failed to parse ready-sent flag %s", val
);
3239 m
->ready_sent
= m
->ready_sent
|| b
;
3241 } else if ((val
= startswith(l
, "taint-logged="))) {
3244 b
= parse_boolean(val
);
3246 log_notice("Failed to parse taint-logged flag %s", val
);
3248 m
->taint_logged
= m
->taint_logged
|| b
;
3250 } else if ((val
= startswith(l
, "service-watchdogs="))) {
3253 b
= parse_boolean(val
);
3255 log_notice("Failed to parse service-watchdogs flag %s", val
);
3257 m
->service_watchdogs
= b
;
3259 } else if ((val
= startswith(l
, "show-status="))) {
3262 s
= show_status_from_string(val
);
3264 log_notice("Failed to parse show-status flag %s", val
);
3266 manager_set_show_status(m
, s
);
3268 } else if ((val
= startswith(l
, "log-level-override="))) {
3271 level
= log_level_from_string(val
);
3273 log_notice("Failed to parse log-level-override value '%s', ignoring.", val
);
3275 manager_override_log_level(m
, level
);
3277 } else if ((val
= startswith(l
, "log-target-override="))) {
3280 target
= log_target_from_string(val
);
3282 log_notice("Failed to parse log-target-override value '%s', ignoring.", val
);
3284 manager_override_log_target(m
, target
);
3286 } else if (startswith(l
, "env=")) {
3287 r
= deserialize_environment(&m
->environment
, l
);
3291 log_notice_errno(r
, "Failed to parse environment entry: \"%s\": %m", l
);
3293 } else if ((val
= startswith(l
, "notify-fd="))) {
3296 if (safe_atoi(val
, &fd
) < 0 || fd
< 0 || !fdset_contains(fds
, fd
))
3297 log_notice("Failed to parse notify fd: \"%s\"", val
);
3299 m
->notify_event_source
= sd_event_source_unref(m
->notify_event_source
);
3300 safe_close(m
->notify_fd
);
3301 m
->notify_fd
= fdset_remove(fds
, fd
);
3304 } else if ((val
= startswith(l
, "notify-socket="))) {
3313 free(m
->notify_socket
);
3314 m
->notify_socket
= n
;
3316 } else if ((val
= startswith(l
, "cgroups-agent-fd="))) {
3319 if (safe_atoi(val
, &fd
) < 0 || fd
< 0 || !fdset_contains(fds
, fd
))
3320 log_notice("Failed to parse cgroups agent fd: %s", val
);
3322 m
->cgroups_agent_event_source
= sd_event_source_unref(m
->cgroups_agent_event_source
);
3323 safe_close(m
->cgroups_agent_fd
);
3324 m
->cgroups_agent_fd
= fdset_remove(fds
, fd
);
3327 } else if ((val
= startswith(l
, "user-lookup="))) {
3330 if (sscanf(val
, "%i %i", &fd0
, &fd1
) != 2 || fd0
< 0 || fd1
< 0 || fd0
== fd1
|| !fdset_contains(fds
, fd0
) || !fdset_contains(fds
, fd1
))
3331 log_notice("Failed to parse user lookup fd: %s", val
);
3333 m
->user_lookup_event_source
= sd_event_source_unref(m
->user_lookup_event_source
);
3334 safe_close_pair(m
->user_lookup_fds
);
3335 m
->user_lookup_fds
[0] = fdset_remove(fds
, fd0
);
3336 m
->user_lookup_fds
[1] = fdset_remove(fds
, fd1
);
3339 } else if ((val
= startswith(l
, "dynamic-user=")))
3340 dynamic_user_deserialize_one(m
, val
, fds
);
3341 else if ((val
= startswith(l
, "destroy-ipc-uid=")))
3342 manager_deserialize_uid_refs_one(m
, val
);
3343 else if ((val
= startswith(l
, "destroy-ipc-gid=")))
3344 manager_deserialize_gid_refs_one(m
, val
);
3345 else if ((val
= startswith(l
, "exec-runtime=")))
3346 exec_runtime_deserialize_one(m
, val
, fds
);
3347 else if ((val
= startswith(l
, "subscribed="))) {
3349 if (strv_extend(&m
->deserialized_subscribed
, val
) < 0)
3354 for (q
= 0; q
< _MANAGER_TIMESTAMP_MAX
; q
++) {
3355 val
= startswith(l
, manager_timestamp_to_string(q
));
3359 val
= startswith(val
, "-timestamp=");
3364 if (q
< _MANAGER_TIMESTAMP_MAX
) /* found it */
3365 dual_timestamp_deserialize(val
, m
->timestamps
+ q
);
3366 else if (!startswith(l
, "kdbus-fd=")) /* ignore kdbus */
3367 log_notice("Unknown serialization item '%s'", l
);
3373 char name
[UNIT_NAME_MAX
+2];
3374 const char* unit_name
;
3377 if (!fgets(name
, sizeof(name
), f
)) {
3387 unit_name
= strstrip(name
);
3389 r
= manager_load_unit(m
, unit_name
, NULL
, NULL
, &u
);
3391 log_notice_errno(r
, "Failed to load unit \"%s\", skipping deserialization: %m", unit_name
);
3394 unit_deserialize_skip(f
);
3398 r
= unit_deserialize(u
, f
, fds
);
3400 log_notice_errno(r
, "Failed to deserialize unit \"%s\": %m", unit_name
);
3410 assert(m
->n_reloading
> 0);
3416 static void manager_flush_finished_jobs(Manager
*m
) {
3419 while ((j
= set_steal_first(m
->pending_finished_jobs
))) {
3420 bus_job_send_removed_signal(j
);
3424 m
->pending_finished_jobs
= set_free(m
->pending_finished_jobs
);
3427 int manager_reload(Manager
*m
) {
3429 _cleanup_fclose_
FILE *f
= NULL
;
3430 _cleanup_fdset_free_ FDSet
*fds
= NULL
;
3434 r
= manager_open_serialization(m
, &f
);
3439 bus_manager_send_reloading(m
, true);
3447 r
= manager_serialize(m
, f
, fds
, false);
3453 if (fseeko(f
, 0, SEEK_SET
) < 0) {
3458 /* From here on there is no way back. */
3459 manager_clear_jobs_and_units(m
);
3460 lookup_paths_flush_generator(&m
->lookup_paths
);
3461 lookup_paths_free(&m
->lookup_paths
);
3462 exec_runtime_vacuum(m
);
3463 dynamic_user_vacuum(m
, false);
3464 m
->uid_refs
= hashmap_free(m
->uid_refs
);
3465 m
->gid_refs
= hashmap_free(m
->gid_refs
);
3467 q
= lookup_paths_init(&m
->lookup_paths
, m
->unit_file_scope
, 0, NULL
);
3468 if (q
< 0 && r
>= 0)
3471 q
= manager_run_environment_generators(m
);
3472 if (q
< 0 && r
>= 0)
3475 /* Find new unit paths */
3476 q
= manager_run_generators(m
);
3477 if (q
< 0 && r
>= 0)
3480 lookup_paths_reduce(&m
->lookup_paths
);
3481 manager_build_unit_path_cache(m
);
3483 /* First, enumerate what we can from all config files */
3484 manager_enumerate(m
);
3486 /* Second, deserialize our stored data */
3487 q
= manager_deserialize(m
, f
, fds
);
3489 log_error_errno(q
, "Deserialization failed: %m");
3497 /* Re-register notify_fd as event source */
3498 q
= manager_setup_notify(m
);
3499 if (q
< 0 && r
>= 0)
3502 q
= manager_setup_cgroups_agent(m
);
3503 if (q
< 0 && r
>= 0)
3506 q
= manager_setup_user_lookup_fd(m
);
3507 if (q
< 0 && r
>= 0)
3510 /* Third, fire things up! */
3511 manager_coldplug(m
);
3513 /* Release any dynamic users no longer referenced */
3514 dynamic_user_vacuum(m
, true);
3516 /* Release any references to UIDs/GIDs no longer referenced, and destroy any IPC owned by them */
3517 manager_vacuum_uid_refs(m
);
3518 manager_vacuum_gid_refs(m
);
3520 exec_runtime_vacuum(m
);
3522 assert(m
->n_reloading
> 0);
3525 /* It might be safe to log to the journal now and connect to dbus */
3526 manager_recheck_journal(m
);
3527 manager_recheck_dbus(m
);
3529 /* Let's finally catch up with any changes that took place while we were reloading/reexecing */
3532 /* Sync current state of bus names with our set of listening units */
3533 q
= manager_enqueue_sync_bus_names(m
);
3534 if (q
< 0 && r
>= 0)
3537 if (!MANAGER_IS_RELOADING(m
))
3538 manager_flush_finished_jobs(m
);
3540 m
->send_reloading_done
= true;
3545 void manager_reset_failed(Manager
*m
) {
3551 HASHMAP_FOREACH(u
, m
->units
, i
)
3552 unit_reset_failed(u
);
3555 bool manager_unit_inactive_or_pending(Manager
*m
, const char *name
) {
3561 /* Returns true if the unit is inactive or going down */
3562 u
= manager_get_unit(m
, name
);
3566 return unit_inactive_or_pending(u
);
3569 static void log_taint_string(Manager
*m
) {
3570 _cleanup_free_
char *taint
= NULL
;
3574 if (MANAGER_IS_USER(m
) || m
->taint_logged
)
3577 m
->taint_logged
= true; /* only check for taint once */
3579 taint
= manager_taint_string(m
);
3583 log_struct(LOG_NOTICE
,
3584 LOG_MESSAGE("System is tainted: %s", taint
),
3586 "MESSAGE_ID=" SD_MESSAGE_TAINTED_STR
);
3589 static void manager_notify_finished(Manager
*m
) {
3590 char userspace
[FORMAT_TIMESPAN_MAX
], initrd
[FORMAT_TIMESPAN_MAX
], kernel
[FORMAT_TIMESPAN_MAX
], sum
[FORMAT_TIMESPAN_MAX
];
3591 usec_t firmware_usec
, loader_usec
, kernel_usec
, initrd_usec
, userspace_usec
, total_usec
;
3593 if (m
->test_run_flags
)
3596 if (MANAGER_IS_SYSTEM(m
) && detect_container() <= 0) {
3597 char ts
[FORMAT_TIMESPAN_MAX
];
3598 char buf
[FORMAT_TIMESPAN_MAX
+ STRLEN(" (firmware) + ") + FORMAT_TIMESPAN_MAX
+ STRLEN(" (loader) + ")]
3601 size_t size
= sizeof buf
;
3603 /* Note that MANAGER_TIMESTAMP_KERNEL's monotonic value is always at 0, and
3604 * MANAGER_TIMESTAMP_FIRMWARE's and MANAGER_TIMESTAMP_LOADER's monotonic value should be considered
3605 * negative values. */
3607 firmware_usec
= m
->timestamps
[MANAGER_TIMESTAMP_FIRMWARE
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_LOADER
].monotonic
;
3608 loader_usec
= m
->timestamps
[MANAGER_TIMESTAMP_LOADER
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_KERNEL
].monotonic
;
3609 userspace_usec
= m
->timestamps
[MANAGER_TIMESTAMP_FINISH
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_USERSPACE
].monotonic
;
3610 total_usec
= m
->timestamps
[MANAGER_TIMESTAMP_FIRMWARE
].monotonic
+ m
->timestamps
[MANAGER_TIMESTAMP_FINISH
].monotonic
;
3612 if (firmware_usec
> 0)
3613 size
= strpcpyf(&p
, size
, "%s (firmware) + ", format_timespan(ts
, sizeof(ts
), firmware_usec
, USEC_PER_MSEC
));
3614 if (loader_usec
> 0)
3615 size
= strpcpyf(&p
, size
, "%s (loader) + ", format_timespan(ts
, sizeof(ts
), loader_usec
, USEC_PER_MSEC
));
3617 if (dual_timestamp_is_set(&m
->timestamps
[MANAGER_TIMESTAMP_INITRD
])) {
3619 /* The initrd case on bare-metal*/
3620 kernel_usec
= m
->timestamps
[MANAGER_TIMESTAMP_INITRD
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_KERNEL
].monotonic
;
3621 initrd_usec
= m
->timestamps
[MANAGER_TIMESTAMP_USERSPACE
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_INITRD
].monotonic
;
3623 log_struct(LOG_INFO
,
3624 "MESSAGE_ID=" SD_MESSAGE_STARTUP_FINISHED_STR
,
3625 "KERNEL_USEC="USEC_FMT
, kernel_usec
,
3626 "INITRD_USEC="USEC_FMT
, initrd_usec
,
3627 "USERSPACE_USEC="USEC_FMT
, userspace_usec
,
3628 LOG_MESSAGE("Startup finished in %s%s (kernel) + %s (initrd) + %s (userspace) = %s.",
3630 format_timespan(kernel
, sizeof(kernel
), kernel_usec
, USEC_PER_MSEC
),
3631 format_timespan(initrd
, sizeof(initrd
), initrd_usec
, USEC_PER_MSEC
),
3632 format_timespan(userspace
, sizeof(userspace
), userspace_usec
, USEC_PER_MSEC
),
3633 format_timespan(sum
, sizeof(sum
), total_usec
, USEC_PER_MSEC
)));
3635 /* The initrd-less case on bare-metal*/
3637 kernel_usec
= m
->timestamps
[MANAGER_TIMESTAMP_USERSPACE
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_KERNEL
].monotonic
;
3640 log_struct(LOG_INFO
,
3641 "MESSAGE_ID=" SD_MESSAGE_STARTUP_FINISHED_STR
,
3642 "KERNEL_USEC="USEC_FMT
, kernel_usec
,
3643 "USERSPACE_USEC="USEC_FMT
, userspace_usec
,
3644 LOG_MESSAGE("Startup finished in %s%s (kernel) + %s (userspace) = %s.",
3646 format_timespan(kernel
, sizeof(kernel
), kernel_usec
, USEC_PER_MSEC
),
3647 format_timespan(userspace
, sizeof(userspace
), userspace_usec
, USEC_PER_MSEC
),
3648 format_timespan(sum
, sizeof(sum
), total_usec
, USEC_PER_MSEC
)));
3651 /* The container and --user case */
3652 firmware_usec
= loader_usec
= initrd_usec
= kernel_usec
= 0;
3653 total_usec
= userspace_usec
= m
->timestamps
[MANAGER_TIMESTAMP_FINISH
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_USERSPACE
].monotonic
;
3655 log_struct(LOG_INFO
,
3656 "MESSAGE_ID=" SD_MESSAGE_USER_STARTUP_FINISHED_STR
,
3657 "USERSPACE_USEC="USEC_FMT
, userspace_usec
,
3658 LOG_MESSAGE("Startup finished in %s.",
3659 format_timespan(sum
, sizeof(sum
), total_usec
, USEC_PER_MSEC
)));
3662 bus_manager_send_finished(m
, firmware_usec
, loader_usec
, kernel_usec
, initrd_usec
, userspace_usec
, total_usec
);
3665 m
->ready_sent
? "STATUS=Startup finished in %s."
3667 "STATUS=Startup finished in %s.",
3668 format_timespan(sum
, sizeof(sum
), total_usec
, USEC_PER_MSEC
));
3669 m
->ready_sent
= true;
3671 log_taint_string(m
);
3674 static void manager_send_ready(Manager
*m
) {
3677 /* We send READY=1 on reaching basic.target only when running in --user mode. */
3678 if (!MANAGER_IS_USER(m
) || m
->ready_sent
)
3681 m
->ready_sent
= true;
3685 "STATUS=Reached " SPECIAL_BASIC_TARGET
".");
3688 static void manager_check_basic_target(Manager
*m
) {
3693 /* Small shortcut */
3694 if (m
->ready_sent
&& m
->taint_logged
)
3697 u
= manager_get_unit(m
, SPECIAL_BASIC_TARGET
);
3698 if (!u
|| !UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u
)))
3701 /* For user managers, send out READY=1 as soon as we reach basic.target */
3702 manager_send_ready(m
);
3704 /* Log the taint string as soon as we reach basic.target */
3705 log_taint_string(m
);
3708 void manager_check_finished(Manager
*m
) {
3711 if (MANAGER_IS_RELOADING(m
))
3714 /* Verify that we have entered the event loop already, and not left it again. */
3715 if (!MANAGER_IS_RUNNING(m
))
3718 manager_check_basic_target(m
);
3720 if (hashmap_size(m
->jobs
) > 0) {
3721 if (m
->jobs_in_progress_event_source
)
3722 /* Ignore any failure, this is only for feedback */
3723 (void) sd_event_source_set_time(m
->jobs_in_progress_event_source
, now(CLOCK_MONOTONIC
) + JOBS_IN_PROGRESS_WAIT_USEC
);
3728 manager_flip_auto_status(m
, false);
3730 /* Notify Type=idle units that we are done now */
3731 manager_close_idle_pipe(m
);
3733 /* Turn off confirm spawn now */
3734 m
->confirm_spawn
= NULL
;
3736 /* No need to update ask password status when we're going non-interactive */
3737 manager_close_ask_password(m
);
3739 /* This is no longer the first boot */
3740 manager_set_first_boot(m
, false);
3742 if (MANAGER_IS_FINISHED(m
))
3745 dual_timestamp_get(m
->timestamps
+ MANAGER_TIMESTAMP_FINISH
);
3747 manager_notify_finished(m
);
3749 manager_invalidate_startup_units(m
);
3752 static bool generator_path_any(const char* const* paths
) {
3756 /* Optimize by skipping the whole process by not creating output directories
3757 * if no generators are found. */
3758 STRV_FOREACH(path
, (char**) paths
)
3759 if (access(*path
, F_OK
) == 0)
3761 else if (errno
!= ENOENT
)
3762 log_warning_errno(errno
, "Failed to open generator directory %s: %m", *path
);
3767 static const char* system_env_generator_binary_paths
[] = {
3768 "/run/systemd/system-environment-generators",
3769 "/etc/systemd/system-environment-generators",
3770 "/usr/local/lib/systemd/system-environment-generators",
3771 SYSTEM_ENV_GENERATOR_PATH
,
3775 static const char* user_env_generator_binary_paths
[] = {
3776 "/run/systemd/user-environment-generators",
3777 "/etc/systemd/user-environment-generators",
3778 "/usr/local/lib/systemd/user-environment-generators",
3779 USER_ENV_GENERATOR_PATH
,
3783 static int manager_run_environment_generators(Manager
*m
) {
3784 char **tmp
= NULL
; /* this is only used in the forked process, no cleanup here */
3786 void* args
[] = {&tmp
, &tmp
, &m
->environment
};
3788 if (m
->test_run_flags
&& !(m
->test_run_flags
& MANAGER_TEST_RUN_ENV_GENERATORS
))
3791 paths
= MANAGER_IS_SYSTEM(m
) ? system_env_generator_binary_paths
: user_env_generator_binary_paths
;
3793 if (!generator_path_any(paths
))
3796 return execute_directories(paths
, DEFAULT_TIMEOUT_USEC
, gather_environment
, args
, NULL
);
3799 static int manager_run_generators(Manager
*m
) {
3800 _cleanup_strv_free_
char **paths
= NULL
;
3801 const char *argv
[5];
3806 if (m
->test_run_flags
&& !(m
->test_run_flags
& MANAGER_TEST_RUN_GENERATORS
))
3809 paths
= generator_binary_paths(m
->unit_file_scope
);
3813 if (!generator_path_any((const char* const*) paths
))
3816 r
= lookup_paths_mkdir_generator(&m
->lookup_paths
);
3820 argv
[0] = NULL
; /* Leave this empty, execute_directory() will fill something in */
3821 argv
[1] = m
->lookup_paths
.generator
;
3822 argv
[2] = m
->lookup_paths
.generator_early
;
3823 argv
[3] = m
->lookup_paths
.generator_late
;
3826 RUN_WITH_UMASK(0022)
3827 execute_directories((const char* const*) paths
, DEFAULT_TIMEOUT_USEC
,
3828 NULL
, NULL
, (char**) argv
);
3831 lookup_paths_trim_generator(&m
->lookup_paths
);
3835 int manager_environment_add(Manager
*m
, char **minus
, char **plus
) {
3836 char **a
= NULL
, **b
= NULL
, **l
;
3841 if (!strv_isempty(minus
)) {
3842 a
= strv_env_delete(l
, 1, minus
);
3849 if (!strv_isempty(plus
)) {
3850 b
= strv_env_merge(2, l
, plus
);
3859 if (m
->environment
!= l
)
3860 strv_free(m
->environment
);
3867 manager_sanitize_environment(m
);
3872 int manager_set_default_rlimits(Manager
*m
, struct rlimit
**default_rlimit
) {
3877 for (i
= 0; i
< _RLIMIT_MAX
; i
++) {
3878 m
->rlimit
[i
] = mfree(m
->rlimit
[i
]);
3880 if (!default_rlimit
[i
])
3883 m
->rlimit
[i
] = newdup(struct rlimit
, default_rlimit
[i
], 1);
3891 void manager_recheck_dbus(Manager
*m
) {
3894 /* Connects to the bus if the dbus service and socket are running. If we are running in user mode this is all
3895 * it does. In system mode we'll also connect to the system bus (which will most likely just reuse the
3896 * connection of the API bus). That's because the system bus after all runs as service of the system instance,
3897 * while in the user instance we can assume it's already there. */
3899 if (MANAGER_IS_RELOADING(m
))
3900 return; /* don't check while we are reloading… */
3902 if (manager_dbus_is_running(m
, false)) {
3903 (void) bus_init_api(m
);
3905 if (MANAGER_IS_SYSTEM(m
))
3906 (void) bus_init_system(m
);
3908 (void) bus_done_api(m
);
3910 if (MANAGER_IS_SYSTEM(m
))
3911 (void) bus_done_system(m
);
3915 static bool manager_journal_is_running(Manager
*m
) {
3920 if (m
->test_run_flags
!= 0)
3923 /* If we are the user manager we can safely assume that the journal is up */
3924 if (!MANAGER_IS_SYSTEM(m
))
3927 /* Check that the socket is not only up, but in RUNNING state */
3928 u
= manager_get_unit(m
, SPECIAL_JOURNALD_SOCKET
);
3931 if (SOCKET(u
)->state
!= SOCKET_RUNNING
)
3934 /* Similar, check if the daemon itself is fully up, too */
3935 u
= manager_get_unit(m
, SPECIAL_JOURNALD_SERVICE
);
3938 if (!IN_SET(SERVICE(u
)->state
, SERVICE_RELOAD
, SERVICE_RUNNING
))
3944 void manager_recheck_journal(Manager
*m
) {
3948 /* Don't bother with this unless we are in the special situation of being PID 1 */
3949 if (getpid_cached() != 1)
3952 /* Don't check this while we are reloading, things might still change */
3953 if (MANAGER_IS_RELOADING(m
))
3956 /* The journal is fully and entirely up? If so, let's permit logging to it, if that's configured. If the
3957 * journal is down, don't ever log to it, otherwise we might end up deadlocking ourselves as we might trigger
3958 * an activation ourselves we can't fulfill. */
3959 log_set_prohibit_ipc(!manager_journal_is_running(m
));
3963 void manager_set_show_status(Manager
*m
, ShowStatus mode
) {
3965 assert(IN_SET(mode
, SHOW_STATUS_AUTO
, SHOW_STATUS_NO
, SHOW_STATUS_YES
, SHOW_STATUS_TEMPORARY
));
3967 if (!MANAGER_IS_SYSTEM(m
))
3970 if (m
->show_status
!= mode
)
3971 log_debug("%s showing of status.",
3972 mode
== SHOW_STATUS_NO
? "Disabling" : "Enabling");
3973 m
->show_status
= mode
;
3975 if (IN_SET(mode
, SHOW_STATUS_TEMPORARY
, SHOW_STATUS_YES
))
3976 (void) touch("/run/systemd/show-status");
3978 (void) unlink("/run/systemd/show-status");
3981 static bool manager_get_show_status(Manager
*m
, StatusType type
) {
3984 if (!MANAGER_IS_SYSTEM(m
))
3987 if (m
->no_console_output
)
3990 if (!IN_SET(manager_state(m
), MANAGER_INITIALIZING
, MANAGER_STARTING
, MANAGER_STOPPING
))
3993 /* If we cannot find out the status properly, just proceed. */
3994 if (type
!= STATUS_TYPE_EMERGENCY
&& manager_check_ask_password(m
) > 0)
3997 return IN_SET(m
->show_status
, SHOW_STATUS_TEMPORARY
, SHOW_STATUS_YES
);
4000 const char *manager_get_confirm_spawn(Manager
*m
) {
4001 static int last_errno
= 0;
4002 const char *vc
= m
->confirm_spawn
;
4006 /* Here's the deal: we want to test the validity of the console but don't want
4007 * PID1 to go through the whole console process which might block. But we also
4008 * want to warn the user only once if something is wrong with the console so we
4009 * cannot do the sanity checks after spawning our children. So here we simply do
4010 * really basic tests to hopefully trap common errors.
4012 * If the console suddenly disappear at the time our children will really it
4013 * then they will simply fail to acquire it and a positive answer will be
4014 * assumed. New children will fallback to /dev/console though.
4016 * Note: TTYs are devices that can come and go any time, and frequently aren't
4017 * available yet during early boot (consider a USB rs232 dongle...). If for any
4018 * reason the configured console is not ready, we fallback to the default
4021 if (!vc
|| path_equal(vc
, "/dev/console"))
4028 if (!S_ISCHR(st
.st_mode
)) {
4036 if (last_errno
!= errno
) {
4038 log_warning_errno(errno
, "Failed to open %s: %m, using default console", vc
);
4040 return "/dev/console";
4043 void manager_set_first_boot(Manager
*m
, bool b
) {
4046 if (!MANAGER_IS_SYSTEM(m
))
4049 if (m
->first_boot
!= (int) b
) {
4051 (void) touch("/run/systemd/first-boot");
4053 (void) unlink("/run/systemd/first-boot");
4059 void manager_disable_confirm_spawn(void) {
4060 (void) touch("/run/systemd/confirm_spawn_disabled");
4063 bool manager_is_confirm_spawn_disabled(Manager
*m
) {
4064 if (!m
->confirm_spawn
)
4067 return access("/run/systemd/confirm_spawn_disabled", F_OK
) >= 0;
4070 void manager_status_printf(Manager
*m
, StatusType type
, const char *status
, const char *format
, ...) {
4073 /* If m is NULL, assume we're after shutdown and let the messages through. */
4075 if (m
&& !manager_get_show_status(m
, type
))
4078 /* XXX We should totally drop the check for ephemeral here
4079 * and thus effectively make 'Type=idle' pointless. */
4080 if (type
== STATUS_TYPE_EPHEMERAL
&& m
&& m
->n_on_console
> 0)
4083 va_start(ap
, format
);
4084 status_vprintf(status
, true, type
== STATUS_TYPE_EPHEMERAL
, format
, ap
);
4088 Set
*manager_get_units_requiring_mounts_for(Manager
*m
, const char *path
) {
4089 char p
[strlen(path
)+1];
4095 path_simplify(p
, false);
4097 return hashmap_get(m
->units_requiring_mounts_for
, streq(p
, "/") ? "" : p
);
4100 int manager_update_failed_units(Manager
*m
, Unit
*u
, bool failed
) {
4105 assert(u
->manager
== m
);
4107 size
= set_size(m
->failed_units
);
4110 r
= set_ensure_allocated(&m
->failed_units
, NULL
);
4114 if (set_put(m
->failed_units
, u
) < 0)
4117 (void) set_remove(m
->failed_units
, u
);
4119 if (set_size(m
->failed_units
) != size
)
4120 bus_manager_send_change_signal(m
);
4125 ManagerState
manager_state(Manager
*m
) {
4130 /* Did we ever finish booting? If not then we are still starting up */
4131 if (!MANAGER_IS_FINISHED(m
)) {
4133 u
= manager_get_unit(m
, SPECIAL_BASIC_TARGET
);
4134 if (!u
|| !UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u
)))
4135 return MANAGER_INITIALIZING
;
4137 return MANAGER_STARTING
;
4140 /* Is the special shutdown target active or queued? If so, we are in shutdown state */
4141 u
= manager_get_unit(m
, SPECIAL_SHUTDOWN_TARGET
);
4142 if (u
&& unit_active_or_pending(u
))
4143 return MANAGER_STOPPING
;
4145 if (MANAGER_IS_SYSTEM(m
)) {
4146 /* Are the rescue or emergency targets active or queued? If so we are in maintenance state */
4147 u
= manager_get_unit(m
, SPECIAL_RESCUE_TARGET
);
4148 if (u
&& unit_active_or_pending(u
))
4149 return MANAGER_MAINTENANCE
;
4151 u
= manager_get_unit(m
, SPECIAL_EMERGENCY_TARGET
);
4152 if (u
&& unit_active_or_pending(u
))
4153 return MANAGER_MAINTENANCE
;
4156 /* Are there any failed units? If so, we are in degraded mode */
4157 if (set_size(m
->failed_units
) > 0)
4158 return MANAGER_DEGRADED
;
4160 return MANAGER_RUNNING
;
4163 #define DESTROY_IPC_FLAG (UINT32_C(1) << 31)
4165 static void manager_unref_uid_internal(
4170 int (*_clean_ipc
)(uid_t uid
)) {
4176 assert(uid_is_valid(uid
));
4179 /* A generic implementation, covering both manager_unref_uid() and manager_unref_gid(), under the assumption
4180 * that uid_t and gid_t are actually defined the same way, with the same validity rules.
4182 * We store a hashmap where the UID/GID is they key and the value is a 32bit reference counter, whose highest
4183 * bit is used as flag for marking UIDs/GIDs whose IPC objects to remove when the last reference to the UID/GID
4184 * is dropped. The flag is set to on, once at least one reference from a unit where RemoveIPC= is set is added
4185 * on a UID/GID. It is reset when the UID's/GID's reference counter drops to 0 again. */
4187 assert_cc(sizeof(uid_t
) == sizeof(gid_t
));
4188 assert_cc(UID_INVALID
== (uid_t
) GID_INVALID
);
4190 if (uid
== 0) /* We don't keep track of root, and will never destroy it */
4193 c
= PTR_TO_UINT32(hashmap_get(*uid_refs
, UID_TO_PTR(uid
)));
4195 n
= c
& ~DESTROY_IPC_FLAG
;
4199 if (destroy_now
&& n
== 0) {
4200 hashmap_remove(*uid_refs
, UID_TO_PTR(uid
));
4202 if (c
& DESTROY_IPC_FLAG
) {
4203 log_debug("%s " UID_FMT
" is no longer referenced, cleaning up its IPC.",
4204 _clean_ipc
== clean_ipc_by_uid
? "UID" : "GID",
4206 (void) _clean_ipc(uid
);
4209 c
= n
| (c
& DESTROY_IPC_FLAG
);
4210 assert_se(hashmap_update(*uid_refs
, UID_TO_PTR(uid
), UINT32_TO_PTR(c
)) >= 0);
4214 void manager_unref_uid(Manager
*m
, uid_t uid
, bool destroy_now
) {
4215 manager_unref_uid_internal(m
, &m
->uid_refs
, uid
, destroy_now
, clean_ipc_by_uid
);
4218 void manager_unref_gid(Manager
*m
, gid_t gid
, bool destroy_now
) {
4219 manager_unref_uid_internal(m
, &m
->gid_refs
, (uid_t
) gid
, destroy_now
, clean_ipc_by_gid
);
4222 static int manager_ref_uid_internal(
4233 assert(uid_is_valid(uid
));
4235 /* A generic implementation, covering both manager_ref_uid() and manager_ref_gid(), under the assumption
4236 * that uid_t and gid_t are actually defined the same way, with the same validity rules. */
4238 assert_cc(sizeof(uid_t
) == sizeof(gid_t
));
4239 assert_cc(UID_INVALID
== (uid_t
) GID_INVALID
);
4241 if (uid
== 0) /* We don't keep track of root, and will never destroy it */
4244 r
= hashmap_ensure_allocated(uid_refs
, &trivial_hash_ops
);
4248 c
= PTR_TO_UINT32(hashmap_get(*uid_refs
, UID_TO_PTR(uid
)));
4250 n
= c
& ~DESTROY_IPC_FLAG
;
4253 if (n
& DESTROY_IPC_FLAG
) /* check for overflow */
4256 c
= n
| (c
& DESTROY_IPC_FLAG
) | (clean_ipc
? DESTROY_IPC_FLAG
: 0);
4258 return hashmap_replace(*uid_refs
, UID_TO_PTR(uid
), UINT32_TO_PTR(c
));
4261 int manager_ref_uid(Manager
*m
, uid_t uid
, bool clean_ipc
) {
4262 return manager_ref_uid_internal(m
, &m
->uid_refs
, uid
, clean_ipc
);
4265 int manager_ref_gid(Manager
*m
, gid_t gid
, bool clean_ipc
) {
4266 return manager_ref_uid_internal(m
, &m
->gid_refs
, (uid_t
) gid
, clean_ipc
);
4269 static void manager_vacuum_uid_refs_internal(
4272 int (*_clean_ipc
)(uid_t uid
)) {
4281 HASHMAP_FOREACH_KEY(p
, k
, *uid_refs
, i
) {
4285 uid
= PTR_TO_UID(k
);
4286 c
= PTR_TO_UINT32(p
);
4288 n
= c
& ~DESTROY_IPC_FLAG
;
4292 if (c
& DESTROY_IPC_FLAG
) {
4293 log_debug("Found unreferenced %s " UID_FMT
" after reload/reexec. Cleaning up.",
4294 _clean_ipc
== clean_ipc_by_uid
? "UID" : "GID",
4296 (void) _clean_ipc(uid
);
4299 assert_se(hashmap_remove(*uid_refs
, k
) == p
);
4303 void manager_vacuum_uid_refs(Manager
*m
) {
4304 manager_vacuum_uid_refs_internal(m
, &m
->uid_refs
, clean_ipc_by_uid
);
4307 void manager_vacuum_gid_refs(Manager
*m
) {
4308 manager_vacuum_uid_refs_internal(m
, &m
->gid_refs
, clean_ipc_by_gid
);
4311 static void manager_serialize_uid_refs_internal(
4315 const char *field_name
) {
4325 /* Serialize the UID reference table. Or actually, just the IPC destruction flag of it, as the actual counter
4326 * of it is better rebuild after a reload/reexec. */
4328 HASHMAP_FOREACH_KEY(p
, k
, *uid_refs
, i
) {
4332 uid
= PTR_TO_UID(k
);
4333 c
= PTR_TO_UINT32(p
);
4335 if (!(c
& DESTROY_IPC_FLAG
))
4338 fprintf(f
, "%s=" UID_FMT
"\n", field_name
, uid
);
4342 void manager_serialize_uid_refs(Manager
*m
, FILE *f
) {
4343 manager_serialize_uid_refs_internal(m
, f
, &m
->uid_refs
, "destroy-ipc-uid");
4346 void manager_serialize_gid_refs(Manager
*m
, FILE *f
) {
4347 manager_serialize_uid_refs_internal(m
, f
, &m
->gid_refs
, "destroy-ipc-gid");
4350 static void manager_deserialize_uid_refs_one_internal(
4353 const char *value
) {
4363 r
= parse_uid(value
, &uid
);
4364 if (r
< 0 || uid
== 0) {
4365 log_debug("Unable to parse UID reference serialization");
4369 r
= hashmap_ensure_allocated(uid_refs
, &trivial_hash_ops
);
4375 c
= PTR_TO_UINT32(hashmap_get(*uid_refs
, UID_TO_PTR(uid
)));
4376 if (c
& DESTROY_IPC_FLAG
)
4379 c
|= DESTROY_IPC_FLAG
;
4381 r
= hashmap_replace(*uid_refs
, UID_TO_PTR(uid
), UINT32_TO_PTR(c
));
4383 log_debug("Failed to add UID reference entry");
4388 void manager_deserialize_uid_refs_one(Manager
*m
, const char *value
) {
4389 manager_deserialize_uid_refs_one_internal(m
, &m
->uid_refs
, value
);
4392 void manager_deserialize_gid_refs_one(Manager
*m
, const char *value
) {
4393 manager_deserialize_uid_refs_one_internal(m
, &m
->gid_refs
, value
);
4396 int manager_dispatch_user_lookup_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
4400 char unit_name
[UNIT_NAME_MAX
+1];
4403 Manager
*m
= userdata
;
4411 /* Invoked whenever a child process succeeded resolving its user/group to use and sent us the resulting UID/GID
4412 * in a datagram. We parse the datagram here and pass it off to the unit, so that it can add a reference to the
4413 * UID/GID so that it can destroy the UID/GID's IPC objects when the reference counter drops to 0. */
4415 l
= recv(fd
, &buffer
, sizeof(buffer
), MSG_DONTWAIT
);
4417 if (IN_SET(errno
, EINTR
, EAGAIN
))
4420 return log_error_errno(errno
, "Failed to read from user lookup fd: %m");
4423 if ((size_t) l
<= offsetof(struct buffer
, unit_name
)) {
4424 log_warning("Received too short user lookup message, ignoring.");
4428 if ((size_t) l
> offsetof(struct buffer
, unit_name
) + UNIT_NAME_MAX
) {
4429 log_warning("Received too long user lookup message, ignoring.");
4433 if (!uid_is_valid(buffer
.uid
) && !gid_is_valid(buffer
.gid
)) {
4434 log_warning("Got user lookup message with invalid UID/GID pair, ignoring.");
4438 n
= (size_t) l
- offsetof(struct buffer
, unit_name
);
4439 if (memchr(buffer
.unit_name
, 0, n
)) {
4440 log_warning("Received lookup message with embedded NUL character, ignoring.");
4444 buffer
.unit_name
[n
] = 0;
4445 u
= manager_get_unit(m
, buffer
.unit_name
);
4447 log_debug("Got user lookup message but unit doesn't exist, ignoring.");
4451 log_unit_debug(u
, "User lookup succeeded: uid=" UID_FMT
" gid=" GID_FMT
, buffer
.uid
, buffer
.gid
);
4453 unit_notify_user_lookup(u
, buffer
.uid
, buffer
.gid
);
4457 char *manager_taint_string(Manager
*m
) {
4458 _cleanup_free_
char *destination
= NULL
, *overflowuid
= NULL
, *overflowgid
= NULL
;
4462 /* Returns a "taint string", e.g. "local-hwclock:var-run-bad".
4463 * Only things that are detected at runtime should be tagged
4464 * here. For stuff that is set during compilation, emit a warning
4465 * in the configuration phase. */
4469 buf
= new(char, sizeof("split-usr:"
4473 "overflowuid-not-65534:"
4474 "overflowgid-not-65534:"));
4482 e
= stpcpy(e
, "split-usr:");
4484 if (access("/proc/cgroups", F_OK
) < 0)
4485 e
= stpcpy(e
, "cgroups-missing:");
4487 if (clock_is_localtime(NULL
) > 0)
4488 e
= stpcpy(e
, "local-hwclock:");
4490 r
= readlink_malloc("/var/run", &destination
);
4491 if (r
< 0 || !PATH_IN_SET(destination
, "../run", "/run"))
4492 e
= stpcpy(e
, "var-run-bad:");
4494 r
= read_one_line_file("/proc/sys/kernel/overflowuid", &overflowuid
);
4495 if (r
>= 0 && !streq(overflowuid
, "65534"))
4496 e
= stpcpy(e
, "overflowuid-not-65534:");
4498 r
= read_one_line_file("/proc/sys/kernel/overflowgid", &overflowgid
);
4499 if (r
>= 0 && !streq(overflowgid
, "65534"))
4500 e
= stpcpy(e
, "overflowgid-not-65534:");
4502 /* remove the last ':' */
4509 void manager_ref_console(Manager
*m
) {
4515 void manager_unref_console(Manager
*m
) {
4517 assert(m
->n_on_console
> 0);
4520 if (m
->n_on_console
== 0)
4521 m
->no_console_output
= false; /* unset no_console_output flag, since the console is definitely free now */
4524 void manager_override_log_level(Manager
*m
, int level
) {
4525 _cleanup_free_
char *s
= NULL
;
4528 if (!m
->log_level_overridden
) {
4529 m
->original_log_level
= log_get_max_level();
4530 m
->log_level_overridden
= true;
4533 (void) log_level_to_string_alloc(level
, &s
);
4534 log_info("Setting log level to %s.", strna(s
));
4536 log_set_max_level(level
);
4539 void manager_restore_original_log_level(Manager
*m
) {
4540 _cleanup_free_
char *s
= NULL
;
4543 if (!m
->log_level_overridden
)
4546 (void) log_level_to_string_alloc(m
->original_log_level
, &s
);
4547 log_info("Restoring log level to original (%s).", strna(s
));
4549 log_set_max_level(m
->original_log_level
);
4550 m
->log_level_overridden
= false;
4553 void manager_override_log_target(Manager
*m
, LogTarget target
) {
4556 if (!m
->log_target_overridden
) {
4557 m
->original_log_target
= log_get_target();
4558 m
->log_target_overridden
= true;
4561 log_info("Setting log target to %s.", log_target_to_string(target
));
4562 log_set_target(target
);
4565 void manager_restore_original_log_target(Manager
*m
) {
4568 if (!m
->log_target_overridden
)
4571 log_info("Restoring log target to original %s.", log_target_to_string(m
->original_log_target
));
4573 log_set_target(m
->original_log_target
);
4574 m
->log_target_overridden
= false;
4577 ManagerTimestamp
manager_timestamp_initrd_mangle(ManagerTimestamp s
) {
4579 s
>= MANAGER_TIMESTAMP_SECURITY_START
&&
4580 s
<= MANAGER_TIMESTAMP_UNITS_LOAD_FINISH
)
4581 return s
- MANAGER_TIMESTAMP_SECURITY_START
+ MANAGER_TIMESTAMP_INITRD_SECURITY_START
;
4585 static const char *const manager_state_table
[_MANAGER_STATE_MAX
] = {
4586 [MANAGER_INITIALIZING
] = "initializing",
4587 [MANAGER_STARTING
] = "starting",
4588 [MANAGER_RUNNING
] = "running",
4589 [MANAGER_DEGRADED
] = "degraded",
4590 [MANAGER_MAINTENANCE
] = "maintenance",
4591 [MANAGER_STOPPING
] = "stopping",
4594 DEFINE_STRING_TABLE_LOOKUP(manager_state
, ManagerState
);
4596 static const char *const manager_timestamp_table
[_MANAGER_TIMESTAMP_MAX
] = {
4597 [MANAGER_TIMESTAMP_FIRMWARE
] = "firmware",
4598 [MANAGER_TIMESTAMP_LOADER
] = "loader",
4599 [MANAGER_TIMESTAMP_KERNEL
] = "kernel",
4600 [MANAGER_TIMESTAMP_INITRD
] = "initrd",
4601 [MANAGER_TIMESTAMP_USERSPACE
] = "userspace",
4602 [MANAGER_TIMESTAMP_FINISH
] = "finish",
4603 [MANAGER_TIMESTAMP_SECURITY_START
] = "security-start",
4604 [MANAGER_TIMESTAMP_SECURITY_FINISH
] = "security-finish",
4605 [MANAGER_TIMESTAMP_GENERATORS_START
] = "generators-start",
4606 [MANAGER_TIMESTAMP_GENERATORS_FINISH
] = "generators-finish",
4607 [MANAGER_TIMESTAMP_UNITS_LOAD_START
] = "units-load-start",
4608 [MANAGER_TIMESTAMP_UNITS_LOAD_FINISH
] = "units-load-finish",
4609 [MANAGER_TIMESTAMP_INITRD_SECURITY_START
] = "initrd-security-start",
4610 [MANAGER_TIMESTAMP_INITRD_SECURITY_FINISH
] = "initrd-security-finish",
4611 [MANAGER_TIMESTAMP_INITRD_GENERATORS_START
] = "initrd-generators-start",
4612 [MANAGER_TIMESTAMP_INITRD_GENERATORS_FINISH
] = "initrd-generators-finish",
4613 [MANAGER_TIMESTAMP_INITRD_UNITS_LOAD_START
] = "initrd-units-load-start",
4614 [MANAGER_TIMESTAMP_INITRD_UNITS_LOAD_FINISH
] = "initrd-units-load-finish",
4617 DEFINE_STRING_TABLE_LOOKUP(manager_timestamp
, ManagerTimestamp
);