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 "all-units.h"
26 #include "alloc-util.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 "serialize.h"
65 #include "signal-util.h"
66 #include "socket-util.h"
68 #include "stat-util.h"
69 #include "string-table.h"
70 #include "string-util.h"
73 #include "syslog-util.h"
74 #include "terminal-util.h"
75 #include "time-util.h"
76 #include "transaction.h"
77 #include "umask-util.h"
78 #include "unit-name.h"
79 #include "user-util.h"
84 #define NOTIFY_RCVBUF_SIZE (8*1024*1024)
85 #define CGROUPS_AGENT_RCVBUF_SIZE (8*1024*1024)
87 /* Initial delay and the interval for printing status messages about running jobs */
88 #define JOBS_IN_PROGRESS_WAIT_USEC (5*USEC_PER_SEC)
89 #define JOBS_IN_PROGRESS_PERIOD_USEC (USEC_PER_SEC / 3)
90 #define JOBS_IN_PROGRESS_PERIOD_DIVISOR 3
92 /* If there are more than 1K bus messages queue across our API and direct busses, then let's not add more on top until
93 * the queue gets more empty. */
94 #define MANAGER_BUS_BUSY_THRESHOLD 1024LU
96 /* How many units and jobs to process of the bus queue before returning to the event loop. */
97 #define MANAGER_BUS_MESSAGE_BUDGET 100U
99 static int manager_dispatch_notify_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
100 static int manager_dispatch_cgroups_agent_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
101 static int manager_dispatch_signal_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
102 static int manager_dispatch_time_change_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
103 static int manager_dispatch_idle_pipe_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
104 static int manager_dispatch_user_lookup_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
105 static int manager_dispatch_jobs_in_progress(sd_event_source
*source
, usec_t usec
, void *userdata
);
106 static int manager_dispatch_run_queue(sd_event_source
*source
, void *userdata
);
107 static int manager_dispatch_sigchld(sd_event_source
*source
, void *userdata
);
108 static int manager_dispatch_timezone_change(sd_event_source
*source
, const struct inotify_event
*event
, void *userdata
);
109 static int manager_run_environment_generators(Manager
*m
);
110 static int manager_run_generators(Manager
*m
);
112 static void manager_watch_jobs_in_progress(Manager
*m
) {
118 /* We do not want to show the cylon animation if the user
119 * needs to confirm service executions otherwise confirmation
120 * messages will be screwed by the cylon animation. */
121 if (!manager_is_confirm_spawn_disabled(m
))
124 if (m
->jobs_in_progress_event_source
)
127 next
= now(CLOCK_MONOTONIC
) + JOBS_IN_PROGRESS_WAIT_USEC
;
128 r
= sd_event_add_time(
130 &m
->jobs_in_progress_event_source
,
133 manager_dispatch_jobs_in_progress
, m
);
137 (void) sd_event_source_set_description(m
->jobs_in_progress_event_source
, "manager-jobs-in-progress");
140 #define CYLON_BUFFER_EXTRA (2*STRLEN(ANSI_RED) + STRLEN(ANSI_HIGHLIGHT_RED) + 2*STRLEN(ANSI_NORMAL))
142 static void draw_cylon(char buffer
[], size_t buflen
, unsigned width
, unsigned pos
) {
145 assert(buflen
>= CYLON_BUFFER_EXTRA
+ width
+ 1);
146 assert(pos
<= width
+1); /* 0 or width+1 mean that the center light is behind the corner */
150 p
= mempset(p
, ' ', pos
-2);
151 if (log_get_show_color())
152 p
= stpcpy(p
, ANSI_RED
);
156 if (pos
> 0 && pos
<= width
) {
157 if (log_get_show_color())
158 p
= stpcpy(p
, ANSI_HIGHLIGHT_RED
);
162 if (log_get_show_color())
163 p
= stpcpy(p
, ANSI_NORMAL
);
166 if (log_get_show_color())
167 p
= stpcpy(p
, ANSI_RED
);
170 p
= mempset(p
, ' ', width
-1-pos
);
171 if (log_get_show_color())
172 strcpy(p
, ANSI_NORMAL
);
176 void manager_flip_auto_status(Manager
*m
, bool enable
) {
180 if (m
->show_status
== SHOW_STATUS_AUTO
)
181 manager_set_show_status(m
, SHOW_STATUS_TEMPORARY
);
183 if (m
->show_status
== SHOW_STATUS_TEMPORARY
)
184 manager_set_show_status(m
, SHOW_STATUS_AUTO
);
188 static void manager_print_jobs_in_progress(Manager
*m
) {
189 _cleanup_free_
char *job_of_n
= NULL
;
192 unsigned counter
= 0, print_nr
;
193 char cylon
[6 + CYLON_BUFFER_EXTRA
+ 1];
195 char time
[FORMAT_TIMESPAN_MAX
], limit
[FORMAT_TIMESPAN_MAX
] = "no limit";
199 assert(m
->n_running_jobs
> 0);
201 manager_flip_auto_status(m
, true);
203 print_nr
= (m
->jobs_in_progress_iteration
/ JOBS_IN_PROGRESS_PERIOD_DIVISOR
) % m
->n_running_jobs
;
205 HASHMAP_FOREACH(j
, m
->jobs
, i
)
206 if (j
->state
== JOB_RUNNING
&& counter
++ == print_nr
)
209 /* m->n_running_jobs must be consistent with the contents of m->jobs,
210 * so the above loop must have succeeded in finding j. */
211 assert(counter
== print_nr
+ 1);
214 cylon_pos
= m
->jobs_in_progress_iteration
% 14;
216 cylon_pos
= 14 - cylon_pos
;
217 draw_cylon(cylon
, sizeof(cylon
), 6, cylon_pos
);
219 m
->jobs_in_progress_iteration
++;
221 if (m
->n_running_jobs
> 1) {
222 if (asprintf(&job_of_n
, "(%u of %u) ", counter
, m
->n_running_jobs
) < 0)
226 format_timespan(time
, sizeof(time
), now(CLOCK_MONOTONIC
) - j
->begin_usec
, 1*USEC_PER_SEC
);
227 if (job_get_timeout(j
, &x
) > 0)
228 format_timespan(limit
, sizeof(limit
), x
- j
->begin_usec
, 1*USEC_PER_SEC
);
230 manager_status_printf(m
, STATUS_TYPE_EPHEMERAL
, cylon
,
231 "%sA %s job is running for %s (%s / %s)",
233 job_type_to_string(j
->type
),
234 unit_description(j
->unit
),
238 static int have_ask_password(void) {
239 _cleanup_closedir_
DIR *dir
;
242 dir
= opendir("/run/systemd/ask-password");
250 FOREACH_DIRENT_ALL(de
, dir
, return -errno
) {
251 if (startswith(de
->d_name
, "ask."))
257 static int manager_dispatch_ask_password_fd(sd_event_source
*source
,
258 int fd
, uint32_t revents
, void *userdata
) {
259 Manager
*m
= userdata
;
265 m
->have_ask_password
= have_ask_password();
266 if (m
->have_ask_password
< 0)
267 /* Log error but continue. Negative have_ask_password
268 * is treated as unknown status. */
269 log_error_errno(m
->have_ask_password
, "Failed to list /run/systemd/ask-password: %m");
274 static void manager_close_ask_password(Manager
*m
) {
277 m
->ask_password_event_source
= sd_event_source_unref(m
->ask_password_event_source
);
278 m
->ask_password_inotify_fd
= safe_close(m
->ask_password_inotify_fd
);
279 m
->have_ask_password
= -EINVAL
;
282 static int manager_check_ask_password(Manager
*m
) {
287 if (!m
->ask_password_event_source
) {
288 assert(m
->ask_password_inotify_fd
< 0);
290 mkdir_p_label("/run/systemd/ask-password", 0755);
292 m
->ask_password_inotify_fd
= inotify_init1(IN_NONBLOCK
|IN_CLOEXEC
);
293 if (m
->ask_password_inotify_fd
< 0)
294 return log_error_errno(errno
, "inotify_init1() failed: %m");
296 if (inotify_add_watch(m
->ask_password_inotify_fd
, "/run/systemd/ask-password", IN_CREATE
|IN_DELETE
|IN_MOVE
) < 0) {
297 log_error_errno(errno
, "Failed to add watch on /run/systemd/ask-password: %m");
298 manager_close_ask_password(m
);
302 r
= sd_event_add_io(m
->event
, &m
->ask_password_event_source
,
303 m
->ask_password_inotify_fd
, EPOLLIN
,
304 manager_dispatch_ask_password_fd
, m
);
306 log_error_errno(errno
, "Failed to add event source for /run/systemd/ask-password: %m");
307 manager_close_ask_password(m
);
311 (void) sd_event_source_set_description(m
->ask_password_event_source
, "manager-ask-password");
313 /* Queries might have been added meanwhile... */
314 manager_dispatch_ask_password_fd(m
->ask_password_event_source
,
315 m
->ask_password_inotify_fd
, EPOLLIN
, m
);
318 return m
->have_ask_password
;
321 static int manager_watch_idle_pipe(Manager
*m
) {
326 if (m
->idle_pipe_event_source
)
329 if (m
->idle_pipe
[2] < 0)
332 r
= sd_event_add_io(m
->event
, &m
->idle_pipe_event_source
, m
->idle_pipe
[2], EPOLLIN
, manager_dispatch_idle_pipe_fd
, m
);
334 return log_error_errno(r
, "Failed to watch idle pipe: %m");
336 (void) sd_event_source_set_description(m
->idle_pipe_event_source
, "manager-idle-pipe");
341 static void manager_close_idle_pipe(Manager
*m
) {
344 m
->idle_pipe_event_source
= sd_event_source_unref(m
->idle_pipe_event_source
);
346 safe_close_pair(m
->idle_pipe
);
347 safe_close_pair(m
->idle_pipe
+ 2);
350 static int manager_setup_time_change(Manager
*m
) {
355 if (MANAGER_IS_TEST_RUN(m
))
358 m
->time_change_event_source
= sd_event_source_unref(m
->time_change_event_source
);
359 m
->time_change_fd
= safe_close(m
->time_change_fd
);
361 m
->time_change_fd
= time_change_fd();
362 if (m
->time_change_fd
< 0)
363 return log_error_errno(m
->time_change_fd
, "Failed to create timer change timer fd: %m");
365 r
= sd_event_add_io(m
->event
, &m
->time_change_event_source
, m
->time_change_fd
, EPOLLIN
, manager_dispatch_time_change_fd
, m
);
367 return log_error_errno(r
, "Failed to create time change event source: %m");
369 /* Schedule this slightly earlier than the .timer event sources */
370 r
= sd_event_source_set_priority(m
->time_change_event_source
, SD_EVENT_PRIORITY_NORMAL
-1);
372 return log_error_errno(r
, "Failed to set priority of time change event sources: %m");
374 (void) sd_event_source_set_description(m
->time_change_event_source
, "manager-time-change");
376 log_debug("Set up TFD_TIMER_CANCEL_ON_SET timerfd.");
381 static int manager_read_timezone_stat(Manager
*m
) {
387 /* Read the current stat() data of /etc/localtime so that we detect changes */
388 if (lstat("/etc/localtime", &st
) < 0) {
389 log_debug_errno(errno
, "Failed to stat /etc/localtime, ignoring: %m");
390 changed
= m
->etc_localtime_accessible
;
391 m
->etc_localtime_accessible
= false;
395 k
= timespec_load(&st
.st_mtim
);
396 changed
= !m
->etc_localtime_accessible
|| k
!= m
->etc_localtime_mtime
;
398 m
->etc_localtime_mtime
= k
;
399 m
->etc_localtime_accessible
= true;
405 static int manager_setup_timezone_change(Manager
*m
) {
406 _cleanup_(sd_event_source_unrefp
) sd_event_source
*new_event
= NULL
;
411 if (MANAGER_IS_TEST_RUN(m
))
414 /* We watch /etc/localtime for three events: change of the link count (which might mean removal from /etc even
415 * though another link might be kept), renames, and file close operations after writing. Note we don't bother
416 * with IN_DELETE_SELF, as that would just report when the inode is removed entirely, i.e. after the link count
417 * went to zero and all fds to it are closed.
419 * Note that we never follow symlinks here. This is a simplification, but should cover almost all cases
422 * Note that we create the new event source first here, before releasing the old one. This should optimize
423 * behaviour as this way sd-event can reuse the old watch in case the inode didn't change. */
425 r
= sd_event_add_inotify(m
->event
, &new_event
, "/etc/localtime",
426 IN_ATTRIB
|IN_MOVE_SELF
|IN_CLOSE_WRITE
|IN_DONT_FOLLOW
, manager_dispatch_timezone_change
, m
);
428 /* If the file doesn't exist yet, subscribe to /etc instead, and wait until it is created either by
429 * O_CREATE or by rename() */
431 log_debug_errno(r
, "/etc/localtime doesn't exist yet, watching /etc instead.");
432 r
= sd_event_add_inotify(m
->event
, &new_event
, "/etc",
433 IN_CREATE
|IN_MOVED_TO
|IN_ONLYDIR
, manager_dispatch_timezone_change
, m
);
436 return log_error_errno(r
, "Failed to create timezone change event source: %m");
438 /* Schedule this slightly earlier than the .timer event sources */
439 r
= sd_event_source_set_priority(new_event
, SD_EVENT_PRIORITY_NORMAL
-1);
441 return log_error_errno(r
, "Failed to set priority of timezone change event sources: %m");
443 sd_event_source_unref(m
->timezone_change_event_source
);
444 m
->timezone_change_event_source
= TAKE_PTR(new_event
);
449 static int enable_special_signals(Manager
*m
) {
450 _cleanup_close_
int fd
= -1;
454 if (MANAGER_IS_TEST_RUN(m
))
457 /* Enable that we get SIGINT on control-alt-del. In containers
458 * this will fail with EPERM (older) or EINVAL (newer), so
460 if (reboot(RB_DISABLE_CAD
) < 0 && !IN_SET(errno
, EPERM
, EINVAL
))
461 log_warning_errno(errno
, "Failed to enable ctrl-alt-del handling: %m");
463 fd
= open_terminal("/dev/tty0", O_RDWR
|O_NOCTTY
|O_CLOEXEC
);
465 /* Support systems without virtual console */
467 log_warning_errno(errno
, "Failed to open /dev/tty0: %m");
469 /* Enable that we get SIGWINCH on kbrequest */
470 if (ioctl(fd
, KDSIGACCEPT
, SIGWINCH
) < 0)
471 log_warning_errno(errno
, "Failed to enable kbrequest handling: %m");
477 #define RTSIG_IF_AVAILABLE(signum) (signum <= SIGRTMAX ? signum : -1)
479 static int manager_setup_signals(Manager
*m
) {
480 struct sigaction sa
= {
481 .sa_handler
= SIG_DFL
,
482 .sa_flags
= SA_NOCLDSTOP
|SA_RESTART
,
489 assert_se(sigaction(SIGCHLD
, &sa
, NULL
) == 0);
491 /* We make liberal use of realtime signals here. On
492 * Linux/glibc we have 30 of them (with the exception of Linux
493 * on hppa, see below), between SIGRTMIN+0 ... SIGRTMIN+30
496 assert_se(sigemptyset(&mask
) == 0);
497 sigset_add_many(&mask
,
498 SIGCHLD
, /* Child died */
499 SIGTERM
, /* Reexecute daemon */
500 SIGHUP
, /* Reload configuration */
501 SIGUSR1
, /* systemd/upstart: reconnect to D-Bus */
502 SIGUSR2
, /* systemd: dump status */
503 SIGINT
, /* Kernel sends us this on control-alt-del */
504 SIGWINCH
, /* Kernel sends us this on kbrequest (alt-arrowup) */
505 SIGPWR
, /* Some kernel drivers and upsd send us this on power failure */
507 SIGRTMIN
+0, /* systemd: start default.target */
508 SIGRTMIN
+1, /* systemd: isolate rescue.target */
509 SIGRTMIN
+2, /* systemd: isolate emergency.target */
510 SIGRTMIN
+3, /* systemd: start halt.target */
511 SIGRTMIN
+4, /* systemd: start poweroff.target */
512 SIGRTMIN
+5, /* systemd: start reboot.target */
513 SIGRTMIN
+6, /* systemd: start kexec.target */
515 /* ... space for more special targets ... */
517 SIGRTMIN
+13, /* systemd: Immediate halt */
518 SIGRTMIN
+14, /* systemd: Immediate poweroff */
519 SIGRTMIN
+15, /* systemd: Immediate reboot */
520 SIGRTMIN
+16, /* systemd: Immediate kexec */
522 /* ... space for more immediate system state changes ... */
524 SIGRTMIN
+20, /* systemd: enable status messages */
525 SIGRTMIN
+21, /* systemd: disable status messages */
526 SIGRTMIN
+22, /* systemd: set log level to LOG_DEBUG */
527 SIGRTMIN
+23, /* systemd: set log level to LOG_INFO */
528 SIGRTMIN
+24, /* systemd: Immediate exit (--user only) */
530 /* .. one free signal here ... */
532 /* Apparently Linux on hppa had fewer RT signals until v3.18,
533 * SIGRTMAX was SIGRTMIN+25, and then SIGRTMIN was lowered,
534 * see commit v3.17-7614-g1f25df2eff.
536 * We cannot unconditionally make use of those signals here,
537 * so let's use a runtime check. Since these commands are
538 * accessible by different means and only really a safety
539 * net, the missing functionality on hppa shouldn't matter.
542 RTSIG_IF_AVAILABLE(SIGRTMIN
+26), /* systemd: set log target to journal-or-kmsg */
543 RTSIG_IF_AVAILABLE(SIGRTMIN
+27), /* systemd: set log target to console */
544 RTSIG_IF_AVAILABLE(SIGRTMIN
+28), /* systemd: set log target to kmsg */
545 RTSIG_IF_AVAILABLE(SIGRTMIN
+29), /* systemd: set log target to syslog-or-kmsg (obsolete) */
547 /* ... one free signal here SIGRTMIN+30 ... */
549 assert_se(sigprocmask(SIG_SETMASK
, &mask
, NULL
) == 0);
551 m
->signal_fd
= signalfd(-1, &mask
, SFD_NONBLOCK
|SFD_CLOEXEC
);
552 if (m
->signal_fd
< 0)
555 r
= sd_event_add_io(m
->event
, &m
->signal_event_source
, m
->signal_fd
, EPOLLIN
, manager_dispatch_signal_fd
, m
);
559 (void) sd_event_source_set_description(m
->signal_event_source
, "manager-signal");
561 /* Process signals a bit earlier than the rest of things, but later than notify_fd processing, so that the
562 * notify processing can still figure out to which process/service a message belongs, before we reap the
563 * process. Also, process this before handling cgroup notifications, so that we always collect child exit
564 * status information before detecting that there's no process in a cgroup. */
565 r
= sd_event_source_set_priority(m
->signal_event_source
, SD_EVENT_PRIORITY_NORMAL
-6);
569 if (MANAGER_IS_SYSTEM(m
))
570 return enable_special_signals(m
);
575 static char** sanitize_environment(char **l
) {
577 /* Let's remove some environment variables that we need ourselves to communicate with our clients */
597 /* Let's order the environment alphabetically, just to make it pretty */
603 int manager_default_environment(Manager
*m
) {
606 m
->transient_environment
= strv_free(m
->transient_environment
);
608 if (MANAGER_IS_SYSTEM(m
)) {
609 /* The system manager always starts with a clean
610 * environment for its children. It does not import
611 * the kernel's or the parents' exported variables.
613 * The initial passed environment is untouched to keep
614 * /proc/self/environ valid; it is used for tagging
615 * the init process inside containers. */
616 m
->transient_environment
= strv_new("PATH=" DEFAULT_PATH
);
618 /* Import locale variables LC_*= from configuration */
619 (void) locale_setup(&m
->transient_environment
);
621 /* The user manager passes its own environment
622 * along to its children. */
623 m
->transient_environment
= strv_copy(environ
);
625 if (!m
->transient_environment
)
628 sanitize_environment(m
->transient_environment
);
633 static int manager_setup_prefix(Manager
*m
) {
639 static const struct table_entry paths_system
[_EXEC_DIRECTORY_TYPE_MAX
] = {
640 [EXEC_DIRECTORY_RUNTIME
] = { SD_PATH_SYSTEM_RUNTIME
, NULL
},
641 [EXEC_DIRECTORY_STATE
] = { SD_PATH_SYSTEM_STATE_PRIVATE
, NULL
},
642 [EXEC_DIRECTORY_CACHE
] = { SD_PATH_SYSTEM_STATE_CACHE
, NULL
},
643 [EXEC_DIRECTORY_LOGS
] = { SD_PATH_SYSTEM_STATE_LOGS
, NULL
},
644 [EXEC_DIRECTORY_CONFIGURATION
] = { SD_PATH_SYSTEM_CONFIGURATION
, NULL
},
647 static const struct table_entry paths_user
[_EXEC_DIRECTORY_TYPE_MAX
] = {
648 [EXEC_DIRECTORY_RUNTIME
] = { SD_PATH_USER_RUNTIME
, NULL
},
649 [EXEC_DIRECTORY_STATE
] = { SD_PATH_USER_CONFIGURATION
, NULL
},
650 [EXEC_DIRECTORY_CACHE
] = { SD_PATH_USER_STATE_CACHE
, NULL
},
651 [EXEC_DIRECTORY_LOGS
] = { SD_PATH_USER_CONFIGURATION
, "log" },
652 [EXEC_DIRECTORY_CONFIGURATION
] = { SD_PATH_USER_CONFIGURATION
, NULL
},
655 const struct table_entry
*p
;
661 if (MANAGER_IS_SYSTEM(m
))
666 for (i
= 0; i
< _EXEC_DIRECTORY_TYPE_MAX
; i
++) {
667 r
= sd_path_home(p
[i
].type
, p
[i
].suffix
, &m
->prefix
[i
]);
675 static int manager_setup_run_queue(Manager
*m
) {
679 assert(!m
->run_queue_event_source
);
681 r
= sd_event_add_defer(m
->event
, &m
->run_queue_event_source
, manager_dispatch_run_queue
, m
);
685 r
= sd_event_source_set_priority(m
->run_queue_event_source
, SD_EVENT_PRIORITY_IDLE
);
689 r
= sd_event_source_set_enabled(m
->run_queue_event_source
, SD_EVENT_OFF
);
693 (void) sd_event_source_set_description(m
->run_queue_event_source
, "manager-run-queue");
698 static int manager_setup_sigchld_event_source(Manager
*m
) {
702 assert(!m
->sigchld_event_source
);
704 r
= sd_event_add_defer(m
->event
, &m
->sigchld_event_source
, manager_dispatch_sigchld
, m
);
708 r
= sd_event_source_set_priority(m
->sigchld_event_source
, SD_EVENT_PRIORITY_NORMAL
-7);
712 r
= sd_event_source_set_enabled(m
->sigchld_event_source
, SD_EVENT_OFF
);
716 (void) sd_event_source_set_description(m
->sigchld_event_source
, "manager-sigchld");
721 int manager_new(UnitFileScope scope
, ManagerTestRunFlags test_run_flags
, Manager
**_m
) {
722 _cleanup_(manager_freep
) Manager
*m
= NULL
;
726 assert(IN_SET(scope
, UNIT_FILE_SYSTEM
, UNIT_FILE_USER
));
733 .unit_file_scope
= scope
,
734 .objective
= _MANAGER_OBJECTIVE_INVALID
,
736 .default_timer_accuracy_usec
= USEC_PER_MINUTE
,
737 .default_memory_accounting
= MEMORY_ACCOUNTING_DEFAULT
,
738 .default_tasks_accounting
= true,
739 .default_tasks_max
= UINT64_MAX
,
740 .default_timeout_start_usec
= DEFAULT_TIMEOUT_USEC
,
741 .default_timeout_stop_usec
= DEFAULT_TIMEOUT_USEC
,
742 .default_restart_usec
= DEFAULT_RESTART_USEC
,
744 .original_log_level
= -1,
745 .original_log_target
= _LOG_TARGET_INVALID
,
748 .cgroups_agent_fd
= -1,
750 .time_change_fd
= -1,
751 .user_lookup_fds
= { -1, -1 },
752 .private_listen_fd
= -1,
754 .cgroup_inotify_fd
= -1,
755 .pin_cgroupfs_fd
= -1,
756 .ask_password_inotify_fd
= -1,
757 .idle_pipe
= { -1, -1, -1, -1},
759 /* start as id #1, so that we can leave #0 around as "null-like" value */
762 .have_ask_password
= -EINVAL
, /* we don't know */
764 .test_run_flags
= test_run_flags
,
768 if (MANAGER_IS_SYSTEM(m
) && detect_container() <= 0)
769 boot_timestamps(m
->timestamps
+ MANAGER_TIMESTAMP_USERSPACE
,
770 m
->timestamps
+ MANAGER_TIMESTAMP_FIRMWARE
,
771 m
->timestamps
+ MANAGER_TIMESTAMP_LOADER
);
774 /* Prepare log fields we can use for structured logging */
775 if (MANAGER_IS_SYSTEM(m
)) {
776 m
->unit_log_field
= "UNIT=";
777 m
->unit_log_format_string
= "UNIT=%s";
779 m
->invocation_log_field
= "INVOCATION_ID=";
780 m
->invocation_log_format_string
= "INVOCATION_ID=%s";
782 m
->unit_log_field
= "USER_UNIT=";
783 m
->unit_log_format_string
= "USER_UNIT=%s";
785 m
->invocation_log_field
= "USER_INVOCATION_ID=";
786 m
->invocation_log_format_string
= "USER_INVOCATION_ID=%s";
789 /* Reboot immediately if the user hits C-A-D more often than 7x per 2s */
790 RATELIMIT_INIT(m
->ctrl_alt_del_ratelimit
, 2 * USEC_PER_SEC
, 7);
792 r
= manager_default_environment(m
);
796 r
= hashmap_ensure_allocated(&m
->units
, &string_hash_ops
);
800 r
= hashmap_ensure_allocated(&m
->jobs
, NULL
);
804 r
= hashmap_ensure_allocated(&m
->cgroup_unit
, &path_hash_ops
);
808 r
= hashmap_ensure_allocated(&m
->watch_bus
, &string_hash_ops
);
812 r
= manager_setup_prefix(m
);
816 r
= sd_event_default(&m
->event
);
820 r
= manager_setup_run_queue(m
);
824 if (test_run_flags
== MANAGER_TEST_RUN_MINIMAL
) {
825 m
->cgroup_root
= strdup("");
829 r
= manager_setup_signals(m
);
833 r
= manager_setup_cgroup(m
);
837 r
= manager_setup_time_change(m
);
841 r
= manager_read_timezone_stat(m
);
845 (void) manager_setup_timezone_change(m
);
847 r
= manager_setup_sigchld_event_source(m
);
852 if (MANAGER_IS_SYSTEM(m
) && test_run_flags
== 0) {
853 r
= mkdir_label("/run/systemd/units", 0755);
854 if (r
< 0 && r
!= -EEXIST
)
860 dir_is_empty("/usr") > 0;
862 /* Note that we do not set up the notify fd here. We do that after deserialization,
863 * since they might have gotten serialized across the reexec. */
870 static int manager_setup_notify(Manager
*m
) {
873 if (MANAGER_IS_TEST_RUN(m
))
876 if (m
->notify_fd
< 0) {
877 _cleanup_close_
int fd
= -1;
878 union sockaddr_union sa
= {};
881 /* First free all secondary fields */
882 m
->notify_socket
= mfree(m
->notify_socket
);
883 m
->notify_event_source
= sd_event_source_unref(m
->notify_event_source
);
885 fd
= socket(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
|SOCK_NONBLOCK
, 0);
887 return log_error_errno(errno
, "Failed to allocate notification socket: %m");
889 fd_inc_rcvbuf(fd
, NOTIFY_RCVBUF_SIZE
);
891 m
->notify_socket
= strappend(m
->prefix
[EXEC_DIRECTORY_RUNTIME
], "/systemd/notify");
892 if (!m
->notify_socket
)
895 salen
= sockaddr_un_set_path(&sa
.un
, m
->notify_socket
);
897 return log_error_errno(salen
, "Notify socket '%s' not valid for AF_UNIX socket address, refusing.", m
->notify_socket
);
899 (void) mkdir_parents_label(m
->notify_socket
, 0755);
900 (void) sockaddr_un_unlink(&sa
.un
);
902 r
= bind(fd
, &sa
.sa
, salen
);
904 return log_error_errno(errno
, "bind(%s) failed: %m", m
->notify_socket
);
906 r
= setsockopt_int(fd
, SOL_SOCKET
, SO_PASSCRED
, true);
908 return log_error_errno(r
, "SO_PASSCRED failed: %m");
910 m
->notify_fd
= TAKE_FD(fd
);
912 log_debug("Using notification socket %s", m
->notify_socket
);
915 if (!m
->notify_event_source
) {
916 r
= sd_event_add_io(m
->event
, &m
->notify_event_source
, m
->notify_fd
, EPOLLIN
, manager_dispatch_notify_fd
, m
);
918 return log_error_errno(r
, "Failed to allocate notify event source: %m");
920 /* Process notification messages a bit earlier than SIGCHLD, so that we can still identify to which
921 * service an exit message belongs. */
922 r
= sd_event_source_set_priority(m
->notify_event_source
, SD_EVENT_PRIORITY_NORMAL
-8);
924 return log_error_errno(r
, "Failed to set priority of notify event source: %m");
926 (void) sd_event_source_set_description(m
->notify_event_source
, "manager-notify");
932 static int manager_setup_cgroups_agent(Manager
*m
) {
934 static const union sockaddr_union sa
= {
935 .un
.sun_family
= AF_UNIX
,
936 .un
.sun_path
= "/run/systemd/cgroups-agent",
940 /* This creates a listening socket we receive cgroups agent messages on. We do not use D-Bus for delivering
941 * these messages from the cgroups agent binary to PID 1, as the cgroups agent binary is very short-living, and
942 * each instance of it needs a new D-Bus connection. Since D-Bus connections are SOCK_STREAM/AF_UNIX, on
943 * overloaded systems the backlog of the D-Bus socket becomes relevant, as not more than the configured number
944 * of D-Bus connections may be queued until the kernel will start dropping further incoming connections,
945 * possibly resulting in lost cgroups agent messages. To avoid this, we'll use a private SOCK_DGRAM/AF_UNIX
946 * socket, where no backlog is relevant as communication may take place without an actual connect() cycle, and
947 * we thus won't lose messages.
949 * Note that PID 1 will forward the agent message to system bus, so that the user systemd instance may listen
950 * to it. The system instance hence listens on this special socket, but the user instances listen on the system
951 * bus for these messages. */
953 if (MANAGER_IS_TEST_RUN(m
))
956 if (!MANAGER_IS_SYSTEM(m
))
959 r
= cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER
);
961 return log_error_errno(r
, "Failed to determine whether unified cgroups hierarchy is used: %m");
962 if (r
> 0) /* We don't need this anymore on the unified hierarchy */
965 if (m
->cgroups_agent_fd
< 0) {
966 _cleanup_close_
int fd
= -1;
968 /* First free all secondary fields */
969 m
->cgroups_agent_event_source
= sd_event_source_unref(m
->cgroups_agent_event_source
);
971 fd
= socket(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
|SOCK_NONBLOCK
, 0);
973 return log_error_errno(errno
, "Failed to allocate cgroups agent socket: %m");
975 fd_inc_rcvbuf(fd
, CGROUPS_AGENT_RCVBUF_SIZE
);
977 (void) sockaddr_un_unlink(&sa
.un
);
979 /* Only allow root to connect to this socket */
981 r
= bind(fd
, &sa
.sa
, SOCKADDR_UN_LEN(sa
.un
));
983 return log_error_errno(errno
, "bind(%s) failed: %m", sa
.un
.sun_path
);
985 m
->cgroups_agent_fd
= TAKE_FD(fd
);
988 if (!m
->cgroups_agent_event_source
) {
989 r
= sd_event_add_io(m
->event
, &m
->cgroups_agent_event_source
, m
->cgroups_agent_fd
, EPOLLIN
, manager_dispatch_cgroups_agent_fd
, m
);
991 return log_error_errno(r
, "Failed to allocate cgroups agent event source: %m");
993 /* Process cgroups notifications early, but after having processed service notification messages or
994 * SIGCHLD signals, so that a cgroup running empty is always just the last safety net of notification,
995 * and we collected the metadata the notification and SIGCHLD stuff offers first. Also see handling of
996 * cgroup inotify for the unified cgroup stuff. */
997 r
= sd_event_source_set_priority(m
->cgroups_agent_event_source
, SD_EVENT_PRIORITY_NORMAL
-4);
999 return log_error_errno(r
, "Failed to set priority of cgroups agent event source: %m");
1001 (void) sd_event_source_set_description(m
->cgroups_agent_event_source
, "manager-cgroups-agent");
1007 static int manager_setup_user_lookup_fd(Manager
*m
) {
1012 /* Set up the socket pair used for passing UID/GID resolution results from forked off processes to PID
1013 * 1. Background: we can't do name lookups (NSS) from PID 1, since it might involve IPC and thus activation,
1014 * and we might hence deadlock on ourselves. Hence we do all user/group lookups asynchronously from the forked
1015 * off processes right before executing the binaries to start. In order to be able to clean up any IPC objects
1016 * created by a unit (see RemoveIPC=) we need to know in PID 1 the used UID/GID of the executed processes,
1017 * hence we establish this communication channel so that forked off processes can pass their UID/GID
1018 * information back to PID 1. The forked off processes send their resolved UID/GID to PID 1 in a simple
1019 * datagram, along with their unit name, so that we can share one communication socket pair among all units for
1022 * You might wonder why we need a communication channel for this that is independent of the usual notification
1023 * socket scheme (i.e. $NOTIFY_SOCKET). The primary difference is about trust: data sent via the $NOTIFY_SOCKET
1024 * channel is only accepted if it originates from the right unit and if reception was enabled for it. The user
1025 * lookup socket OTOH is only accessible by PID 1 and its children until they exec(), and always available.
1027 * Note that this function is called under two circumstances: when we first initialize (in which case we
1028 * allocate both the socket pair and the event source to listen on it), and when we deserialize after a reload
1029 * (in which case the socket pair already exists but we still need to allocate the event source for it). */
1031 if (m
->user_lookup_fds
[0] < 0) {
1033 /* Free all secondary fields */
1034 safe_close_pair(m
->user_lookup_fds
);
1035 m
->user_lookup_event_source
= sd_event_source_unref(m
->user_lookup_event_source
);
1037 if (socketpair(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
, 0, m
->user_lookup_fds
) < 0)
1038 return log_error_errno(errno
, "Failed to allocate user lookup socket: %m");
1040 (void) fd_inc_rcvbuf(m
->user_lookup_fds
[0], NOTIFY_RCVBUF_SIZE
);
1043 if (!m
->user_lookup_event_source
) {
1044 r
= sd_event_add_io(m
->event
, &m
->user_lookup_event_source
, m
->user_lookup_fds
[0], EPOLLIN
, manager_dispatch_user_lookup_fd
, m
);
1046 return log_error_errno(errno
, "Failed to allocate user lookup event source: %m");
1048 /* Process even earlier than the notify event source, so that we always know first about valid UID/GID
1050 r
= sd_event_source_set_priority(m
->user_lookup_event_source
, SD_EVENT_PRIORITY_NORMAL
-11);
1052 return log_error_errno(errno
, "Failed to set priority ot user lookup event source: %m");
1054 (void) sd_event_source_set_description(m
->user_lookup_event_source
, "user-lookup");
1060 static unsigned manager_dispatch_cleanup_queue(Manager
*m
) {
1066 while ((u
= m
->cleanup_queue
)) {
1067 assert(u
->in_cleanup_queue
);
1077 GC_OFFSET_IN_PATH
, /* This one is on the path we were traveling */
1078 GC_OFFSET_UNSURE
, /* No clue */
1079 GC_OFFSET_GOOD
, /* We still need this unit */
1080 GC_OFFSET_BAD
, /* We don't need this unit anymore */
1084 static void unit_gc_mark_good(Unit
*u
, unsigned gc_marker
) {
1089 u
->gc_marker
= gc_marker
+ GC_OFFSET_GOOD
;
1091 /* Recursively mark referenced units as GOOD as well */
1092 HASHMAP_FOREACH_KEY(v
, other
, u
->dependencies
[UNIT_REFERENCES
], i
)
1093 if (other
->gc_marker
== gc_marker
+ GC_OFFSET_UNSURE
)
1094 unit_gc_mark_good(other
, gc_marker
);
1097 static void unit_gc_sweep(Unit
*u
, unsigned gc_marker
) {
1105 if (IN_SET(u
->gc_marker
- gc_marker
,
1106 GC_OFFSET_GOOD
, GC_OFFSET_BAD
, GC_OFFSET_UNSURE
, GC_OFFSET_IN_PATH
))
1109 if (u
->in_cleanup_queue
)
1112 if (!unit_may_gc(u
))
1115 u
->gc_marker
= gc_marker
+ GC_OFFSET_IN_PATH
;
1119 HASHMAP_FOREACH_KEY(v
, other
, u
->dependencies
[UNIT_REFERENCED_BY
], i
) {
1120 unit_gc_sweep(other
, gc_marker
);
1122 if (other
->gc_marker
== gc_marker
+ GC_OFFSET_GOOD
)
1125 if (other
->gc_marker
!= gc_marker
+ GC_OFFSET_BAD
)
1129 if (u
->refs_by_target
) {
1132 LIST_FOREACH(refs_by_target
, ref
, u
->refs_by_target
) {
1133 unit_gc_sweep(ref
->source
, gc_marker
);
1135 if (ref
->source
->gc_marker
== gc_marker
+ GC_OFFSET_GOOD
)
1138 if (ref
->source
->gc_marker
!= gc_marker
+ GC_OFFSET_BAD
)
1146 /* We were unable to find anything out about this entry, so
1147 * let's investigate it later */
1148 u
->gc_marker
= gc_marker
+ GC_OFFSET_UNSURE
;
1149 unit_add_to_gc_queue(u
);
1153 /* We definitely know that this one is not useful anymore, so
1154 * let's mark it for deletion */
1155 u
->gc_marker
= gc_marker
+ GC_OFFSET_BAD
;
1156 unit_add_to_cleanup_queue(u
);
1160 unit_gc_mark_good(u
, gc_marker
);
1163 static unsigned manager_dispatch_gc_unit_queue(Manager
*m
) {
1164 unsigned n
= 0, gc_marker
;
1169 /* log_debug("Running GC..."); */
1171 m
->gc_marker
+= _GC_OFFSET_MAX
;
1172 if (m
->gc_marker
+ _GC_OFFSET_MAX
<= _GC_OFFSET_MAX
)
1175 gc_marker
= m
->gc_marker
;
1177 while ((u
= m
->gc_unit_queue
)) {
1178 assert(u
->in_gc_queue
);
1180 unit_gc_sweep(u
, gc_marker
);
1182 LIST_REMOVE(gc_queue
, m
->gc_unit_queue
, u
);
1183 u
->in_gc_queue
= false;
1187 if (IN_SET(u
->gc_marker
- gc_marker
,
1188 GC_OFFSET_BAD
, GC_OFFSET_UNSURE
)) {
1190 log_unit_debug(u
, "Collecting.");
1191 u
->gc_marker
= gc_marker
+ GC_OFFSET_BAD
;
1192 unit_add_to_cleanup_queue(u
);
1199 static unsigned manager_dispatch_gc_job_queue(Manager
*m
) {
1205 while ((j
= m
->gc_job_queue
)) {
1206 assert(j
->in_gc_queue
);
1208 LIST_REMOVE(gc_queue
, m
->gc_job_queue
, j
);
1209 j
->in_gc_queue
= false;
1216 log_unit_debug(j
->unit
, "Collecting job.");
1217 (void) job_finish_and_invalidate(j
, JOB_COLLECTED
, false, false);
1223 static unsigned manager_dispatch_stop_when_unneeded_queue(Manager
*m
) {
1230 while ((u
= m
->stop_when_unneeded_queue
)) {
1231 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
1232 assert(m
->stop_when_unneeded_queue
);
1234 assert(u
->in_stop_when_unneeded_queue
);
1235 LIST_REMOVE(stop_when_unneeded_queue
, m
->stop_when_unneeded_queue
, u
);
1236 u
->in_stop_when_unneeded_queue
= false;
1240 if (!unit_is_unneeded(u
))
1243 log_unit_debug(u
, "Unit is not needed anymore.");
1245 /* If stopping a unit fails continuously we might enter a stop loop here, hence stop acting on the
1246 * service being unnecessary after a while. */
1248 if (!ratelimit_below(&u
->auto_stop_ratelimit
)) {
1249 log_unit_warning(u
, "Unit not needed anymore, but not stopping since we tried this too often recently.");
1253 /* Ok, nobody needs us anymore. Sniff. Then let's commit suicide */
1254 r
= manager_add_job(u
->manager
, JOB_STOP
, u
, JOB_FAIL
, &error
, NULL
);
1256 log_unit_warning_errno(u
, r
, "Failed to enqueue stop job, ignoring: %s", bus_error_message(&error
, r
));
1262 static void manager_clear_jobs_and_units(Manager
*m
) {
1267 while ((u
= hashmap_first(m
->units
)))
1270 manager_dispatch_cleanup_queue(m
);
1272 assert(!m
->load_queue
);
1273 assert(!m
->run_queue
);
1274 assert(!m
->dbus_unit_queue
);
1275 assert(!m
->dbus_job_queue
);
1276 assert(!m
->cleanup_queue
);
1277 assert(!m
->gc_unit_queue
);
1278 assert(!m
->gc_job_queue
);
1279 assert(!m
->stop_when_unneeded_queue
);
1281 assert(hashmap_isempty(m
->jobs
));
1282 assert(hashmap_isempty(m
->units
));
1284 m
->n_on_console
= 0;
1285 m
->n_running_jobs
= 0;
1286 m
->n_installed_jobs
= 0;
1287 m
->n_failed_jobs
= 0;
1290 Manager
* manager_free(Manager
*m
) {
1291 ExecDirectoryType dt
;
1297 manager_clear_jobs_and_units(m
);
1299 for (c
= 0; c
< _UNIT_TYPE_MAX
; c
++)
1300 if (unit_vtable
[c
]->shutdown
)
1301 unit_vtable
[c
]->shutdown(m
);
1303 /* Keep the cgroup hierarchy in place except when we know we are going down for good */
1304 manager_shutdown_cgroup(m
, IN_SET(m
->objective
, MANAGER_EXIT
, MANAGER_REBOOT
, MANAGER_POWEROFF
, MANAGER_HALT
, MANAGER_KEXEC
));
1306 lookup_paths_flush_generator(&m
->lookup_paths
);
1310 exec_runtime_vacuum(m
);
1311 hashmap_free(m
->exec_runtime_by_id
);
1313 dynamic_user_vacuum(m
, false);
1314 hashmap_free(m
->dynamic_users
);
1316 hashmap_free(m
->units
);
1317 hashmap_free(m
->units_by_invocation_id
);
1318 hashmap_free(m
->jobs
);
1319 hashmap_free(m
->watch_pids
);
1320 hashmap_free(m
->watch_bus
);
1322 set_free(m
->startup_units
);
1323 set_free(m
->failed_units
);
1325 sd_event_source_unref(m
->signal_event_source
);
1326 sd_event_source_unref(m
->sigchld_event_source
);
1327 sd_event_source_unref(m
->notify_event_source
);
1328 sd_event_source_unref(m
->cgroups_agent_event_source
);
1329 sd_event_source_unref(m
->time_change_event_source
);
1330 sd_event_source_unref(m
->timezone_change_event_source
);
1331 sd_event_source_unref(m
->jobs_in_progress_event_source
);
1332 sd_event_source_unref(m
->run_queue_event_source
);
1333 sd_event_source_unref(m
->user_lookup_event_source
);
1334 sd_event_source_unref(m
->sync_bus_names_event_source
);
1336 safe_close(m
->signal_fd
);
1337 safe_close(m
->notify_fd
);
1338 safe_close(m
->cgroups_agent_fd
);
1339 safe_close(m
->time_change_fd
);
1340 safe_close_pair(m
->user_lookup_fds
);
1342 manager_close_ask_password(m
);
1344 manager_close_idle_pipe(m
);
1346 sd_event_unref(m
->event
);
1348 free(m
->notify_socket
);
1350 lookup_paths_free(&m
->lookup_paths
);
1351 strv_free(m
->transient_environment
);
1352 strv_free(m
->client_environment
);
1354 hashmap_free(m
->cgroup_unit
);
1355 set_free_free(m
->unit_path_cache
);
1357 free(m
->switch_root
);
1358 free(m
->switch_root_init
);
1360 rlimit_free_all(m
->rlimit
);
1362 assert(hashmap_isempty(m
->units_requiring_mounts_for
));
1363 hashmap_free(m
->units_requiring_mounts_for
);
1365 hashmap_free(m
->uid_refs
);
1366 hashmap_free(m
->gid_refs
);
1368 for (dt
= 0; dt
< _EXEC_DIRECTORY_TYPE_MAX
; dt
++)
1369 m
->prefix
[dt
] = mfree(m
->prefix
[dt
]);
1374 static void manager_enumerate_perpetual(Manager
*m
) {
1379 /* Let's ask every type to load all units from disk/kernel that it might know */
1380 for (c
= 0; c
< _UNIT_TYPE_MAX
; c
++) {
1381 if (!unit_type_supported(c
)) {
1382 log_debug("Unit type .%s is not supported on this system.", unit_type_to_string(c
));
1386 if (unit_vtable
[c
]->enumerate_perpetual
)
1387 unit_vtable
[c
]->enumerate_perpetual(m
);
1391 static void manager_enumerate(Manager
*m
) {
1396 /* Let's ask every type to load all units from disk/kernel that it might know */
1397 for (c
= 0; c
< _UNIT_TYPE_MAX
; c
++) {
1398 if (!unit_type_supported(c
)) {
1399 log_debug("Unit type .%s is not supported on this system.", unit_type_to_string(c
));
1403 if (unit_vtable
[c
]->enumerate
)
1404 unit_vtable
[c
]->enumerate(m
);
1407 manager_dispatch_load_queue(m
);
1410 static void manager_coldplug(Manager
*m
) {
1418 log_debug("Invoking unit coldplug() handlers…");
1420 /* Let's place the units back into their deserialized state */
1421 HASHMAP_FOREACH_KEY(u
, k
, m
->units
, i
) {
1423 /* ignore aliases */
1427 r
= unit_coldplug(u
);
1429 log_warning_errno(r
, "We couldn't coldplug %s, proceeding anyway: %m", u
->id
);
1433 static void manager_catchup(Manager
*m
) {
1440 log_debug("Invoking unit catchup() handlers…");
1442 /* Let's catch up on any state changes that happened while we were reloading/reexecing */
1443 HASHMAP_FOREACH_KEY(u
, k
, m
->units
, i
) {
1445 /* ignore aliases */
1453 static void manager_build_unit_path_cache(Manager
*m
) {
1459 set_free_free(m
->unit_path_cache
);
1461 m
->unit_path_cache
= set_new(&path_hash_ops
);
1462 if (!m
->unit_path_cache
) {
1467 /* This simply builds a list of files we know exist, so that
1468 * we don't always have to go to disk */
1470 STRV_FOREACH(i
, m
->lookup_paths
.search_path
) {
1471 _cleanup_closedir_
DIR *d
= NULL
;
1476 if (errno
!= ENOENT
)
1477 log_warning_errno(errno
, "Failed to open directory %s, ignoring: %m", *i
);
1481 FOREACH_DIRENT(de
, d
, r
= -errno
; goto fail
) {
1484 p
= strjoin(streq(*i
, "/") ? "" : *i
, "/", de
->d_name
);
1490 r
= set_consume(m
->unit_path_cache
, p
);
1499 log_warning_errno(r
, "Failed to build unit path cache, proceeding without: %m");
1500 m
->unit_path_cache
= set_free_free(m
->unit_path_cache
);
1503 static void manager_distribute_fds(Manager
*m
, FDSet
*fds
) {
1509 HASHMAP_FOREACH(u
, m
->units
, i
) {
1511 if (fdset_size(fds
) <= 0)
1514 if (!UNIT_VTABLE(u
)->distribute_fds
)
1517 UNIT_VTABLE(u
)->distribute_fds(u
, fds
);
1521 static bool manager_dbus_is_running(Manager
*m
, bool deserialized
) {
1526 /* This checks whether the dbus instance we are supposed to expose our APIs on is up. We check both the socket
1527 * and the service unit. If the 'deserialized' parameter is true we'll check the deserialized state of the unit
1528 * rather than the current one. */
1530 if (MANAGER_IS_TEST_RUN(m
))
1533 u
= manager_get_unit(m
, SPECIAL_DBUS_SOCKET
);
1536 if ((deserialized
? SOCKET(u
)->deserialized_state
: SOCKET(u
)->state
) != SOCKET_RUNNING
)
1539 u
= manager_get_unit(m
, SPECIAL_DBUS_SERVICE
);
1542 if (!IN_SET((deserialized
? SERVICE(u
)->deserialized_state
: SERVICE(u
)->state
), SERVICE_RUNNING
, SERVICE_RELOAD
))
1548 static void manager_setup_bus(Manager
*m
) {
1551 /* Let's set up our private bus connection now, unconditionally */
1552 (void) bus_init_private(m
);
1554 /* If we are in --user mode also connect to the system bus now */
1555 if (MANAGER_IS_USER(m
))
1556 (void) bus_init_system(m
);
1558 /* Let's connect to the bus now, but only if the unit is supposed to be up */
1559 if (manager_dbus_is_running(m
, MANAGER_IS_RELOADING(m
))) {
1560 (void) bus_init_api(m
);
1562 if (MANAGER_IS_SYSTEM(m
))
1563 (void) bus_init_system(m
);
1567 static void manager_preset_all(Manager
*m
) {
1572 if (m
->first_boot
<= 0)
1575 if (!MANAGER_IS_SYSTEM(m
))
1578 if (MANAGER_IS_TEST_RUN(m
))
1581 /* If this is the first boot, and we are in the host system, then preset everything */
1582 r
= unit_file_preset_all(UNIT_FILE_SYSTEM
, 0, NULL
, UNIT_FILE_PRESET_ENABLE_ONLY
, NULL
, 0);
1584 log_full_errno(r
== -EEXIST
? LOG_NOTICE
: LOG_WARNING
, r
,
1585 "Failed to populate /etc with preset unit settings, ignoring: %m");
1587 log_info("Populated /etc with preset unit settings.");
1590 static void manager_vacuum(Manager
*m
) {
1593 /* Release any dynamic users no longer referenced */
1594 dynamic_user_vacuum(m
, true);
1596 /* Release any references to UIDs/GIDs no longer referenced, and destroy any IPC owned by them */
1597 manager_vacuum_uid_refs(m
);
1598 manager_vacuum_gid_refs(m
);
1600 /* Release any runtimes no longer referenced */
1601 exec_runtime_vacuum(m
);
1604 static void manager_ready(Manager
*m
) {
1607 /* After having loaded everything, do the final round of catching up with what might have changed */
1609 m
->objective
= MANAGER_OK
; /* Tell everyone we are up now */
1611 /* It might be safe to log to the journal now and connect to dbus */
1612 manager_recheck_journal(m
);
1613 manager_recheck_dbus(m
);
1615 /* Sync current state of bus names with our set of listening units */
1616 (void) manager_enqueue_sync_bus_names(m
);
1618 /* Let's finally catch up with any changes that took place while we were reloading/reexecing */
1622 static Manager
* manager_reloading_start(Manager
*m
) {
1626 static void manager_reloading_stopp(Manager
**m
) {
1628 assert((*m
)->n_reloading
> 0);
1629 (*m
)->n_reloading
--;
1633 int manager_startup(Manager
*m
, FILE *serialization
, FDSet
*fds
) {
1638 /* If we are running in test mode, we still want to run the generators,
1639 * but we should not touch the real generator directories. */
1640 r
= lookup_paths_init(&m
->lookup_paths
, m
->unit_file_scope
,
1641 MANAGER_IS_TEST_RUN(m
) ? LOOKUP_PATHS_TEMPORARY_GENERATED
: 0,
1644 return log_error_errno(r
, "Failed to initialize path lookup table: %m");
1646 dual_timestamp_get(m
->timestamps
+ manager_timestamp_initrd_mangle(MANAGER_TIMESTAMP_GENERATORS_START
));
1647 r
= manager_run_environment_generators(m
);
1649 r
= manager_run_generators(m
);
1650 dual_timestamp_get(m
->timestamps
+ manager_timestamp_initrd_mangle(MANAGER_TIMESTAMP_GENERATORS_FINISH
));
1654 manager_preset_all(m
);
1656 r
= lookup_paths_reduce(&m
->lookup_paths
);
1658 log_warning_errno(r
, "Failed ot reduce unit file paths, ignoring: %m");
1660 manager_build_unit_path_cache(m
);
1663 /* This block is (optionally) done with the reloading counter bumped */
1664 _cleanup_(manager_reloading_stopp
) Manager
*reloading
= NULL
;
1666 /* If we will deserialize make sure that during enumeration this is already known, so we increase the
1667 * counter here already */
1669 reloading
= manager_reloading_start(m
);
1671 /* First, enumerate what we can from all config files */
1672 dual_timestamp_get(m
->timestamps
+ manager_timestamp_initrd_mangle(MANAGER_TIMESTAMP_UNITS_LOAD_START
));
1673 manager_enumerate_perpetual(m
);
1674 manager_enumerate(m
);
1675 dual_timestamp_get(m
->timestamps
+ manager_timestamp_initrd_mangle(MANAGER_TIMESTAMP_UNITS_LOAD_FINISH
));
1677 /* Second, deserialize if there is something to deserialize */
1678 if (serialization
) {
1679 r
= manager_deserialize(m
, serialization
, fds
);
1681 return log_error_errno(r
, "Deserialization failed: %m");
1684 /* Any fds left? Find some unit which wants them. This is useful to allow container managers to pass
1685 * some file descriptors to us pre-initialized. This enables socket-based activation of entire
1687 manager_distribute_fds(m
, fds
);
1689 /* We might have deserialized the notify fd, but if we didn't then let's create the bus now */
1690 r
= manager_setup_notify(m
);
1692 /* No sense to continue without notifications, our children would fail anyway. */
1695 r
= manager_setup_cgroups_agent(m
);
1697 /* Likewise, no sense to continue without empty cgroup notifications. */
1700 r
= manager_setup_user_lookup_fd(m
);
1702 /* This shouldn't fail, except if things are really broken. */
1705 /* Connect to the bus if we are good for it */
1706 manager_setup_bus(m
);
1708 /* Now that we are connected to all possible busses, let's deserialize who is tracking us. */
1709 r
= bus_track_coldplug(m
, &m
->subscribed
, false, m
->deserialized_subscribed
);
1711 log_warning_errno(r
, "Failed to deserialized tracked clients, ignoring: %m");
1712 m
->deserialized_subscribed
= strv_free(m
->deserialized_subscribed
);
1714 /* Third, fire things up! */
1715 manager_coldplug(m
);
1717 /* Clean up runtime objects */
1721 /* Let's wait for the UnitNew/JobNew messages being sent, before we notify that the
1722 * reload is finished */
1723 m
->send_reloading_done
= true;
1731 int manager_add_job(Manager
*m
, JobType type
, Unit
*unit
, JobMode mode
, sd_bus_error
*e
, Job
**_ret
) {
1736 assert(type
< _JOB_TYPE_MAX
);
1738 assert(mode
< _JOB_MODE_MAX
);
1740 if (mode
== JOB_ISOLATE
&& type
!= JOB_START
)
1741 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
, "Isolate is only valid for start.");
1743 if (mode
== JOB_ISOLATE
&& !unit
->allow_isolate
)
1744 return sd_bus_error_setf(e
, BUS_ERROR_NO_ISOLATION
, "Operation refused, unit may not be isolated.");
1746 log_unit_debug(unit
, "Trying to enqueue job %s/%s/%s", unit
->id
, job_type_to_string(type
), job_mode_to_string(mode
));
1748 type
= job_type_collapse(type
, unit
);
1750 tr
= transaction_new(mode
== JOB_REPLACE_IRREVERSIBLY
);
1754 r
= transaction_add_job_and_dependencies(tr
, type
, unit
, NULL
, true, false,
1755 IN_SET(mode
, JOB_IGNORE_DEPENDENCIES
, JOB_IGNORE_REQUIREMENTS
),
1756 mode
== JOB_IGNORE_DEPENDENCIES
, e
);
1760 if (mode
== JOB_ISOLATE
) {
1761 r
= transaction_add_isolate_jobs(tr
, m
);
1766 r
= transaction_activate(tr
, m
, mode
, e
);
1770 log_unit_debug(unit
,
1771 "Enqueued job %s/%s as %u", unit
->id
,
1772 job_type_to_string(type
), (unsigned) tr
->anchor_job
->id
);
1775 *_ret
= tr
->anchor_job
;
1777 transaction_free(tr
);
1781 transaction_abort(tr
);
1782 transaction_free(tr
);
1786 int manager_add_job_by_name(Manager
*m
, JobType type
, const char *name
, JobMode mode
, sd_bus_error
*e
, Job
**ret
) {
1787 Unit
*unit
= NULL
; /* just to appease gcc, initialization is not really necessary */
1791 assert(type
< _JOB_TYPE_MAX
);
1793 assert(mode
< _JOB_MODE_MAX
);
1795 r
= manager_load_unit(m
, name
, NULL
, NULL
, &unit
);
1800 return manager_add_job(m
, type
, unit
, mode
, e
, ret
);
1803 int manager_add_job_by_name_and_warn(Manager
*m
, JobType type
, const char *name
, JobMode mode
, Job
**ret
) {
1804 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
1808 assert(type
< _JOB_TYPE_MAX
);
1810 assert(mode
< _JOB_MODE_MAX
);
1812 r
= manager_add_job_by_name(m
, type
, name
, mode
, &error
, ret
);
1814 return log_warning_errno(r
, "Failed to enqueue %s job for %s: %s", job_mode_to_string(mode
), name
, bus_error_message(&error
, r
));
1819 int manager_propagate_reload(Manager
*m
, Unit
*unit
, JobMode mode
, sd_bus_error
*e
) {
1825 assert(mode
< _JOB_MODE_MAX
);
1826 assert(mode
!= JOB_ISOLATE
); /* Isolate is only valid for start */
1828 tr
= transaction_new(mode
== JOB_REPLACE_IRREVERSIBLY
);
1832 /* We need an anchor job */
1833 r
= transaction_add_job_and_dependencies(tr
, JOB_NOP
, unit
, NULL
, false, false, true, true, e
);
1837 /* Failure in adding individual dependencies is ignored, so this always succeeds. */
1838 transaction_add_propagate_reload_jobs(tr
, unit
, tr
->anchor_job
, mode
== JOB_IGNORE_DEPENDENCIES
, e
);
1840 r
= transaction_activate(tr
, m
, mode
, e
);
1844 transaction_free(tr
);
1848 transaction_abort(tr
);
1849 transaction_free(tr
);
1853 Job
*manager_get_job(Manager
*m
, uint32_t id
) {
1856 return hashmap_get(m
->jobs
, UINT32_TO_PTR(id
));
1859 Unit
*manager_get_unit(Manager
*m
, const char *name
) {
1863 return hashmap_get(m
->units
, name
);
1866 static int manager_dispatch_target_deps_queue(Manager
*m
) {
1871 static const UnitDependency deps
[] = {
1880 while ((u
= m
->target_deps_queue
)) {
1881 assert(u
->in_target_deps_queue
);
1883 LIST_REMOVE(target_deps_queue
, u
->manager
->target_deps_queue
, u
);
1884 u
->in_target_deps_queue
= false;
1886 for (k
= 0; k
< ELEMENTSOF(deps
); k
++) {
1891 HASHMAP_FOREACH_KEY(v
, target
, u
->dependencies
[deps
[k
]], i
) {
1892 r
= unit_add_default_target_dependency(u
, target
);
1902 unsigned manager_dispatch_load_queue(Manager
*m
) {
1908 /* Make sure we are not run recursively */
1909 if (m
->dispatching_load_queue
)
1912 m
->dispatching_load_queue
= true;
1914 /* Dispatches the load queue. Takes a unit from the queue and
1915 * tries to load its data until the queue is empty */
1917 while ((u
= m
->load_queue
)) {
1918 assert(u
->in_load_queue
);
1924 m
->dispatching_load_queue
= false;
1926 /* Dispatch the units waiting for their target dependencies to be added now, as all targets that we know about
1927 * should be loaded and have aliases resolved */
1928 (void) manager_dispatch_target_deps_queue(m
);
1933 int manager_load_unit_prepare(
1940 _cleanup_(unit_freep
) Unit
*cleanup_ret
= NULL
;
1946 assert(name
|| path
);
1949 /* This will prepare the unit for loading, but not actually
1950 * load anything from disk. */
1952 if (path
&& !is_path(path
))
1953 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
, "Path %s is not absolute.", path
);
1956 name
= basename(path
);
1958 t
= unit_name_to_type(name
);
1960 if (t
== _UNIT_TYPE_INVALID
|| !unit_name_is_valid(name
, UNIT_NAME_PLAIN
|UNIT_NAME_INSTANCE
)) {
1961 if (unit_name_is_valid(name
, UNIT_NAME_TEMPLATE
))
1962 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
, "Unit name %s is missing the instance name.", name
);
1964 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
, "Unit name %s is not valid.", name
);
1967 ret
= manager_get_unit(m
, name
);
1973 ret
= cleanup_ret
= unit_new(m
, unit_vtable
[t
]->object_size
);
1978 ret
->fragment_path
= strdup(path
);
1979 if (!ret
->fragment_path
)
1983 r
= unit_add_name(ret
, name
);
1987 unit_add_to_load_queue(ret
);
1988 unit_add_to_dbus_queue(ret
);
1989 unit_add_to_gc_queue(ret
);
1997 int manager_load_unit(
2009 /* This will load the service information files, but not actually
2010 * start any services or anything. */
2012 r
= manager_load_unit_prepare(m
, name
, path
, e
, _ret
);
2016 manager_dispatch_load_queue(m
);
2018 *_ret
= unit_follow_merge(*_ret
);
2022 int manager_load_startable_unit_or_warn(
2028 /* Load a unit, make sure it loaded fully and is not masked. */
2030 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
2034 r
= manager_load_unit(m
, name
, path
, &error
, &unit
);
2036 return log_error_errno(r
, "Failed to load %s %s: %s",
2037 name
? "unit" : "unit file", name
?: path
,
2038 bus_error_message(&error
, r
));
2040 r
= bus_unit_validate_load_state(unit
, &error
);
2042 return log_error_errno(r
, "%s", bus_error_message(&error
, r
));
2048 void manager_dump_jobs(Manager
*s
, FILE *f
, const char *prefix
) {
2055 HASHMAP_FOREACH(j
, s
->jobs
, i
)
2056 job_dump(j
, f
, prefix
);
2059 void manager_dump_units(Manager
*s
, FILE *f
, const char *prefix
) {
2067 HASHMAP_FOREACH_KEY(u
, t
, s
->units
, i
)
2069 unit_dump(u
, f
, prefix
);
2072 void manager_dump(Manager
*m
, FILE *f
, const char *prefix
) {
2078 for (q
= 0; q
< _MANAGER_TIMESTAMP_MAX
; q
++) {
2079 char buf
[FORMAT_TIMESTAMP_MAX
];
2081 if (dual_timestamp_is_set(m
->timestamps
+ q
))
2082 fprintf(f
, "%sTimestamp %s: %s\n",
2084 manager_timestamp_to_string(q
),
2085 format_timestamp(buf
, sizeof(buf
), m
->timestamps
[q
].realtime
));
2088 manager_dump_units(m
, f
, prefix
);
2089 manager_dump_jobs(m
, f
, prefix
);
2092 int manager_get_dump_string(Manager
*m
, char **ret
) {
2093 _cleanup_free_
char *dump
= NULL
;
2094 _cleanup_fclose_
FILE *f
= NULL
;
2101 f
= open_memstream(&dump
, &size
);
2105 (void) __fsetlocking(f
, FSETLOCKING_BYCALLER
);
2107 manager_dump(m
, f
, NULL
);
2109 r
= fflush_and_check(f
);
2115 *ret
= TAKE_PTR(dump
);
2120 void manager_clear_jobs(Manager
*m
) {
2125 while ((j
= hashmap_first(m
->jobs
)))
2126 /* No need to recurse. We're cancelling all jobs. */
2127 job_finish_and_invalidate(j
, JOB_CANCELED
, false, false);
2130 static int manager_dispatch_run_queue(sd_event_source
*source
, void *userdata
) {
2131 Manager
*m
= userdata
;
2137 while ((j
= m
->run_queue
)) {
2138 assert(j
->installed
);
2139 assert(j
->in_run_queue
);
2141 (void) job_run_and_invalidate(j
);
2144 if (m
->n_running_jobs
> 0)
2145 manager_watch_jobs_in_progress(m
);
2147 if (m
->n_on_console
> 0)
2148 manager_watch_idle_pipe(m
);
2153 static unsigned manager_dispatch_dbus_queue(Manager
*m
) {
2154 unsigned n
= 0, budget
;
2160 /* When we are reloading, let's not wait with generating signals, since we need to exit the manager as quickly
2161 * as we can. There's no point in throttling generation of signals in that case. */
2162 if (MANAGER_IS_RELOADING(m
) || m
->send_reloading_done
|| m
->pending_reload_message
)
2163 budget
= (unsigned) -1; /* infinite budget in this case */
2165 /* Anything to do at all? */
2166 if (!m
->dbus_unit_queue
&& !m
->dbus_job_queue
)
2169 /* Do we have overly many messages queued at the moment? If so, let's not enqueue more on top, let's
2170 * sit this cycle out, and process things in a later cycle when the queues got a bit emptier. */
2171 if (manager_bus_n_queued_write(m
) > MANAGER_BUS_BUSY_THRESHOLD
)
2174 /* Only process a certain number of units/jobs per event loop iteration. Even if the bus queue wasn't
2175 * overly full before this call we shouldn't increase it in size too wildly in one step, and we
2176 * shouldn't monopolize CPU time with generating these messages. Note the difference in counting of
2177 * this "budget" and the "threshold" above: the "budget" is decreased only once per generated message,
2178 * regardless how many busses/direct connections it is enqueued on, while the "threshold" is applied to
2179 * each queued instance of bus message, i.e. if the same message is enqueued to five busses/direct
2180 * connections it will be counted five times. This difference in counting ("references"
2181 * vs. "instances") is primarily a result of the fact that it's easier to implement it this way,
2182 * however it also reflects the thinking that the "threshold" should put a limit on used queue memory,
2183 * i.e. space, while the "budget" should put a limit on time. Also note that the "threshold" is
2184 * currently chosen much higher than the "budget". */
2185 budget
= MANAGER_BUS_MESSAGE_BUDGET
;
2188 while (budget
!= 0 && (u
= m
->dbus_unit_queue
)) {
2190 assert(u
->in_dbus_queue
);
2192 bus_unit_send_change_signal(u
);
2195 if (budget
!= (unsigned) -1)
2199 while (budget
!= 0 && (j
= m
->dbus_job_queue
)) {
2200 assert(j
->in_dbus_queue
);
2202 bus_job_send_change_signal(j
);
2205 if (budget
!= (unsigned) -1)
2209 if (m
->send_reloading_done
) {
2210 m
->send_reloading_done
= false;
2211 bus_manager_send_reloading(m
, false);
2215 if (m
->pending_reload_message
) {
2216 bus_send_pending_reload_message(m
);
2223 static int manager_dispatch_cgroups_agent_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
2224 Manager
*m
= userdata
;
2228 n
= recv(fd
, buf
, sizeof(buf
), 0);
2230 return log_error_errno(errno
, "Failed to read cgroups agent message: %m");
2232 log_error("Got zero-length cgroups agent message, ignoring.");
2235 if ((size_t) n
>= sizeof(buf
)) {
2236 log_error("Got overly long cgroups agent message, ignoring.");
2240 if (memchr(buf
, 0, n
)) {
2241 log_error("Got cgroups agent message with embedded NUL byte, ignoring.");
2246 manager_notify_cgroup_empty(m
, buf
);
2247 (void) bus_forward_agent_released(m
, buf
);
2252 static void manager_invoke_notify_message(
2255 const struct ucred
*ucred
,
2264 if (u
->notifygen
== m
->notifygen
) /* Already invoked on this same unit in this same iteration? */
2266 u
->notifygen
= m
->notifygen
;
2268 if (UNIT_VTABLE(u
)->notify_message
) {
2269 _cleanup_strv_free_
char **tags
= NULL
;
2271 tags
= strv_split(buf
, NEWLINE
);
2277 UNIT_VTABLE(u
)->notify_message(u
, ucred
, tags
, fds
);
2279 } else if (DEBUG_LOGGING
) {
2280 _cleanup_free_
char *x
= NULL
, *y
= NULL
;
2282 x
= ellipsize(buf
, 20, 90);
2286 log_unit_debug(u
, "Got notification message \"%s\", ignoring.", strnull(y
));
2290 static int manager_dispatch_notify_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
2292 _cleanup_fdset_free_ FDSet
*fds
= NULL
;
2293 Manager
*m
= userdata
;
2294 char buf
[NOTIFY_BUFFER_MAX
+1];
2295 struct iovec iovec
= {
2297 .iov_len
= sizeof(buf
)-1,
2300 struct cmsghdr cmsghdr
;
2301 uint8_t buf
[CMSG_SPACE(sizeof(struct ucred
)) +
2302 CMSG_SPACE(sizeof(int) * NOTIFY_FD_MAX
)];
2304 struct msghdr msghdr
= {
2307 .msg_control
= &control
,
2308 .msg_controllen
= sizeof(control
),
2311 struct cmsghdr
*cmsg
;
2312 struct ucred
*ucred
= NULL
;
2313 _cleanup_free_ Unit
**array_copy
= NULL
;
2314 Unit
*u1
, *u2
, **array
;
2315 int r
, *fd_array
= NULL
;
2321 assert(m
->notify_fd
== fd
);
2323 if (revents
!= EPOLLIN
) {
2324 log_warning("Got unexpected poll event for notify fd.");
2328 n
= recvmsg(m
->notify_fd
, &msghdr
, MSG_DONTWAIT
|MSG_CMSG_CLOEXEC
|MSG_TRUNC
);
2330 if (IN_SET(errno
, EAGAIN
, EINTR
))
2331 return 0; /* Spurious wakeup, try again */
2333 /* If this is any other, real error, then let's stop processing this socket. This of course means we
2334 * won't take notification messages anymore, but that's still better than busy looping around this:
2335 * being woken up over and over again but being unable to actually read the message off the socket. */
2336 return log_error_errno(errno
, "Failed to receive notification message: %m");
2339 CMSG_FOREACH(cmsg
, &msghdr
) {
2340 if (cmsg
->cmsg_level
== SOL_SOCKET
&& cmsg
->cmsg_type
== SCM_RIGHTS
) {
2342 fd_array
= (int*) CMSG_DATA(cmsg
);
2343 n_fds
= (cmsg
->cmsg_len
- CMSG_LEN(0)) / sizeof(int);
2345 } else if (cmsg
->cmsg_level
== SOL_SOCKET
&&
2346 cmsg
->cmsg_type
== SCM_CREDENTIALS
&&
2347 cmsg
->cmsg_len
== CMSG_LEN(sizeof(struct ucred
))) {
2349 ucred
= (struct ucred
*) CMSG_DATA(cmsg
);
2356 r
= fdset_new_array(&fds
, fd_array
, n_fds
);
2358 close_many(fd_array
, n_fds
);
2364 if (!ucred
|| !pid_is_valid(ucred
->pid
)) {
2365 log_warning("Received notify message without valid credentials. Ignoring.");
2369 if ((size_t) n
>= sizeof(buf
) || (msghdr
.msg_flags
& MSG_TRUNC
)) {
2370 log_warning("Received notify message exceeded maximum size. Ignoring.");
2374 /* As extra safety check, let's make sure the string we get doesn't contain embedded NUL bytes. We permit one
2375 * trailing NUL byte in the message, but don't expect it. */
2376 if (n
> 1 && memchr(buf
, 0, n
-1)) {
2377 log_warning("Received notify message with embedded NUL bytes. Ignoring.");
2381 /* Make sure it's NUL-terminated. */
2384 /* Increase the generation counter used for filtering out duplicate unit invocations. */
2387 /* Notify every unit that might be interested, which might be multiple. */
2388 u1
= manager_get_unit_by_pid_cgroup(m
, ucred
->pid
);
2389 u2
= hashmap_get(m
->watch_pids
, PID_TO_PTR(ucred
->pid
));
2390 array
= hashmap_get(m
->watch_pids
, PID_TO_PTR(-ucred
->pid
));
2397 array_copy
= newdup(Unit
*, array
, k
+1);
2401 /* And now invoke the per-unit callbacks. Note that manager_invoke_notify_message() will handle duplicate units
2402 * make sure we only invoke each unit's handler once. */
2404 manager_invoke_notify_message(m
, u1
, ucred
, buf
, fds
);
2408 manager_invoke_notify_message(m
, u2
, ucred
, buf
, fds
);
2412 for (size_t i
= 0; array_copy
[i
]; i
++) {
2413 manager_invoke_notify_message(m
, array_copy
[i
], ucred
, buf
, fds
);
2418 log_warning("Cannot find unit for notify message of PID "PID_FMT
", ignoring.", ucred
->pid
);
2420 if (fdset_size(fds
) > 0)
2421 log_warning("Got extra auxiliary fds with notification message, closing them.");
2426 static void manager_invoke_sigchld_event(
2429 const siginfo_t
*si
) {
2435 /* Already invoked the handler of this unit in this iteration? Then don't process this again */
2436 if (u
->sigchldgen
== m
->sigchldgen
)
2438 u
->sigchldgen
= m
->sigchldgen
;
2440 log_unit_debug(u
, "Child "PID_FMT
" belongs to %s.", si
->si_pid
, u
->id
);
2441 unit_unwatch_pid(u
, si
->si_pid
);
2443 if (UNIT_VTABLE(u
)->sigchld_event
)
2444 UNIT_VTABLE(u
)->sigchld_event(u
, si
->si_pid
, si
->si_code
, si
->si_status
);
2447 static int manager_dispatch_sigchld(sd_event_source
*source
, void *userdata
) {
2448 Manager
*m
= userdata
;
2455 /* First we call waitd() for a PID and do not reap the zombie. That way we can still access /proc/$PID for it
2456 * while it is a zombie. */
2458 if (waitid(P_ALL
, 0, &si
, WEXITED
|WNOHANG
|WNOWAIT
) < 0) {
2460 if (errno
!= ECHILD
)
2461 log_error_errno(errno
, "Failed to peek for child with waitid(), ignoring: %m");
2469 if (IN_SET(si
.si_code
, CLD_EXITED
, CLD_KILLED
, CLD_DUMPED
)) {
2470 _cleanup_free_ Unit
**array_copy
= NULL
;
2471 _cleanup_free_
char *name
= NULL
;
2472 Unit
*u1
, *u2
, **array
;
2474 (void) get_process_comm(si
.si_pid
, &name
);
2476 log_debug("Child "PID_FMT
" (%s) died (code=%s, status=%i/%s)",
2477 si
.si_pid
, strna(name
),
2478 sigchld_code_to_string(si
.si_code
),
2480 strna(si
.si_code
== CLD_EXITED
2481 ? exit_status_to_string(si
.si_status
, EXIT_STATUS_FULL
)
2482 : signal_to_string(si
.si_status
)));
2484 /* Increase the generation counter used for filtering out duplicate unit invocations */
2487 /* And now figure out the unit this belongs to, it might be multiple... */
2488 u1
= manager_get_unit_by_pid_cgroup(m
, si
.si_pid
);
2489 u2
= hashmap_get(m
->watch_pids
, PID_TO_PTR(si
.si_pid
));
2490 array
= hashmap_get(m
->watch_pids
, PID_TO_PTR(-si
.si_pid
));
2494 /* Cound how many entries the array has */
2498 /* Make a copy of the array so that we don't trip up on the array changing beneath us */
2499 array_copy
= newdup(Unit
*, array
, n
+1);
2504 /* Finally, execute them all. Note that u1, u2 and the array might contain duplicates, but
2505 * that's fine, manager_invoke_sigchld_event() will ensure we only invoke the handlers once for
2506 * each iteration. */
2508 manager_invoke_sigchld_event(m
, u1
, &si
);
2510 manager_invoke_sigchld_event(m
, u2
, &si
);
2512 for (size_t i
= 0; array_copy
[i
]; i
++)
2513 manager_invoke_sigchld_event(m
, array_copy
[i
], &si
);
2516 /* And now, we actually reap the zombie. */
2517 if (waitid(P_PID
, si
.si_pid
, &si
, WEXITED
) < 0) {
2518 log_error_errno(errno
, "Failed to dequeue child, ignoring: %m");
2525 /* All children processed for now, turn off event source */
2527 r
= sd_event_source_set_enabled(m
->sigchld_event_source
, SD_EVENT_OFF
);
2529 return log_error_errno(r
, "Failed to disable SIGCHLD event source: %m");
2534 static void manager_start_target(Manager
*m
, const char *name
, JobMode mode
) {
2535 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
2538 log_debug("Activating special unit %s", name
);
2540 r
= manager_add_job_by_name(m
, JOB_START
, name
, mode
, &error
, NULL
);
2542 log_error("Failed to enqueue %s job: %s", name
, bus_error_message(&error
, r
));
2545 static void manager_handle_ctrl_alt_del(Manager
*m
) {
2546 /* If the user presses C-A-D more than
2547 * 7 times within 2s, we reboot/shutdown immediately,
2548 * unless it was disabled in system.conf */
2550 if (ratelimit_below(&m
->ctrl_alt_del_ratelimit
) || m
->cad_burst_action
== EMERGENCY_ACTION_NONE
)
2551 manager_start_target(m
, SPECIAL_CTRL_ALT_DEL_TARGET
, JOB_REPLACE_IRREVERSIBLY
);
2553 emergency_action(m
, m
->cad_burst_action
, EMERGENCY_ACTION_WARN
, NULL
, -1,
2554 "Ctrl-Alt-Del was pressed more than 7 times within 2s");
2557 static int manager_dispatch_signal_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
2558 Manager
*m
= userdata
;
2560 struct signalfd_siginfo sfsi
;
2564 assert(m
->signal_fd
== fd
);
2566 if (revents
!= EPOLLIN
) {
2567 log_warning("Got unexpected events from signal file descriptor.");
2571 n
= read(m
->signal_fd
, &sfsi
, sizeof(sfsi
));
2572 if (n
!= sizeof(sfsi
)) {
2574 log_warning("Truncated read from signal fd (%zu bytes), ignoring!", n
);
2578 if (IN_SET(errno
, EINTR
, EAGAIN
))
2581 /* We return an error here, which will kill this handler,
2582 * to avoid a busy loop on read error. */
2583 return log_error_errno(errno
, "Reading from signal fd failed: %m");
2586 log_received_signal(sfsi
.ssi_signo
== SIGCHLD
||
2587 (sfsi
.ssi_signo
== SIGTERM
&& MANAGER_IS_USER(m
))
2588 ? LOG_DEBUG
: LOG_INFO
,
2591 switch (sfsi
.ssi_signo
) {
2594 r
= sd_event_source_set_enabled(m
->sigchld_event_source
, SD_EVENT_ON
);
2596 log_warning_errno(r
, "Failed to enable SIGCHLD event source, ignoring: %m");
2601 if (MANAGER_IS_SYSTEM(m
)) {
2602 /* This is for compatibility with the original sysvinit */
2603 if (verify_run_space_and_log("Refusing to reexecute") < 0)
2606 m
->objective
= MANAGER_REEXECUTE
;
2612 if (MANAGER_IS_SYSTEM(m
))
2613 manager_handle_ctrl_alt_del(m
);
2615 manager_start_target(m
, SPECIAL_EXIT_TARGET
,
2616 JOB_REPLACE_IRREVERSIBLY
);
2620 /* This is a nop on non-init */
2621 if (MANAGER_IS_SYSTEM(m
))
2622 manager_start_target(m
, SPECIAL_KBREQUEST_TARGET
, JOB_REPLACE
);
2627 /* This is a nop on non-init */
2628 if (MANAGER_IS_SYSTEM(m
))
2629 manager_start_target(m
, SPECIAL_SIGPWR_TARGET
, JOB_REPLACE
);
2634 if (manager_dbus_is_running(m
, false)) {
2635 log_info("Trying to reconnect to bus...");
2637 (void) bus_init_api(m
);
2639 if (MANAGER_IS_SYSTEM(m
))
2640 (void) bus_init_system(m
);
2642 log_info("Starting D-Bus service...");
2643 manager_start_target(m
, SPECIAL_DBUS_SERVICE
, JOB_REPLACE
);
2649 _cleanup_free_
char *dump
= NULL
;
2651 r
= manager_get_dump_string(m
, &dump
);
2653 log_warning_errno(errno
, "Failed to acquire manager dump: %m");
2657 log_dump(LOG_INFO
, dump
);
2662 if (verify_run_space_and_log("Refusing to reload") < 0)
2665 m
->objective
= MANAGER_RELOAD
;
2670 /* Starting SIGRTMIN+0 */
2671 static const struct {
2674 } target_table
[] = {
2675 [0] = { SPECIAL_DEFAULT_TARGET
, JOB_ISOLATE
},
2676 [1] = { SPECIAL_RESCUE_TARGET
, JOB_ISOLATE
},
2677 [2] = { SPECIAL_EMERGENCY_TARGET
, JOB_ISOLATE
},
2678 [3] = { SPECIAL_HALT_TARGET
, JOB_REPLACE_IRREVERSIBLY
},
2679 [4] = { SPECIAL_POWEROFF_TARGET
, JOB_REPLACE_IRREVERSIBLY
},
2680 [5] = { SPECIAL_REBOOT_TARGET
, JOB_REPLACE_IRREVERSIBLY
},
2681 [6] = { SPECIAL_KEXEC_TARGET
, JOB_REPLACE_IRREVERSIBLY
},
2684 /* Starting SIGRTMIN+13, so that target halt and system halt are 10 apart */
2685 static const ManagerObjective objective_table
[] = {
2687 [1] = MANAGER_POWEROFF
,
2688 [2] = MANAGER_REBOOT
,
2689 [3] = MANAGER_KEXEC
,
2692 if ((int) sfsi
.ssi_signo
>= SIGRTMIN
+0 &&
2693 (int) sfsi
.ssi_signo
< SIGRTMIN
+(int) ELEMENTSOF(target_table
)) {
2694 int idx
= (int) sfsi
.ssi_signo
- SIGRTMIN
;
2695 manager_start_target(m
, target_table
[idx
].target
,
2696 target_table
[idx
].mode
);
2700 if ((int) sfsi
.ssi_signo
>= SIGRTMIN
+13 &&
2701 (int) sfsi
.ssi_signo
< SIGRTMIN
+13+(int) ELEMENTSOF(objective_table
)) {
2702 m
->objective
= objective_table
[sfsi
.ssi_signo
- SIGRTMIN
- 13];
2706 switch (sfsi
.ssi_signo
- SIGRTMIN
) {
2709 manager_set_show_status(m
, SHOW_STATUS_YES
);
2713 manager_set_show_status(m
, SHOW_STATUS_NO
);
2717 manager_override_log_level(m
, LOG_DEBUG
);
2721 manager_restore_original_log_level(m
);
2725 if (MANAGER_IS_USER(m
)) {
2726 m
->objective
= MANAGER_EXIT
;
2730 /* This is a nop on init */
2734 case 29: /* compatibility: used to be mapped to LOG_TARGET_SYSLOG_OR_KMSG */
2735 manager_restore_original_log_target(m
);
2739 manager_override_log_target(m
, LOG_TARGET_CONSOLE
);
2743 manager_override_log_target(m
, LOG_TARGET_KMSG
);
2747 log_warning("Got unhandled signal <%s>.", signal_to_string(sfsi
.ssi_signo
));
2754 static int manager_dispatch_time_change_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
2755 Manager
*m
= userdata
;
2760 assert(m
->time_change_fd
== fd
);
2762 log_struct(LOG_DEBUG
,
2763 "MESSAGE_ID=" SD_MESSAGE_TIME_CHANGE_STR
,
2764 LOG_MESSAGE("Time has been changed"));
2766 /* Restart the watch */
2767 (void) manager_setup_time_change(m
);
2769 HASHMAP_FOREACH(u
, m
->units
, i
)
2770 if (UNIT_VTABLE(u
)->time_change
)
2771 UNIT_VTABLE(u
)->time_change(u
);
2776 static int manager_dispatch_timezone_change(
2777 sd_event_source
*source
,
2778 const struct inotify_event
*e
,
2781 Manager
*m
= userdata
;
2788 log_debug("inotify event for /etc/localtime");
2790 changed
= manager_read_timezone_stat(m
);
2794 /* Something changed, restart the watch, to ensure we watch the new /etc/localtime if it changed */
2795 (void) manager_setup_timezone_change(m
);
2797 /* Read the new timezone */
2800 log_debug("Timezone has been changed (now: %s).", tzname
[daylight
]);
2802 HASHMAP_FOREACH(u
, m
->units
, i
)
2803 if (UNIT_VTABLE(u
)->timezone_change
)
2804 UNIT_VTABLE(u
)->timezone_change(u
);
2809 static int manager_dispatch_idle_pipe_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
2810 Manager
*m
= userdata
;
2813 assert(m
->idle_pipe
[2] == fd
);
2815 /* There's at least one Type=idle child that just gave up on us waiting for the boot process to complete. Let's
2816 * now turn off any further console output if there's at least one service that needs console access, so that
2817 * from now on our own output should not spill into that service's output anymore. After all, we support
2818 * Type=idle only to beautify console output and it generally is set on services that want to own the console
2819 * exclusively without our interference. */
2820 m
->no_console_output
= m
->n_on_console
> 0;
2822 /* Acknowledge the child's request, and let all all other children know too that they shouldn't wait any longer
2823 * by closing the pipes towards them, which is what they are waiting for. */
2824 manager_close_idle_pipe(m
);
2829 static int manager_dispatch_jobs_in_progress(sd_event_source
*source
, usec_t usec
, void *userdata
) {
2830 Manager
*m
= userdata
;
2837 manager_print_jobs_in_progress(m
);
2839 next
= now(CLOCK_MONOTONIC
) + JOBS_IN_PROGRESS_PERIOD_USEC
;
2840 r
= sd_event_source_set_time(source
, next
);
2844 return sd_event_source_set_enabled(source
, SD_EVENT_ONESHOT
);
2847 int manager_loop(Manager
*m
) {
2850 RATELIMIT_DEFINE(rl
, 1*USEC_PER_SEC
, 50000);
2853 assert(m
->objective
== MANAGER_OK
); /* Ensure manager_startup() has been called */
2855 /* Release the path cache */
2856 m
->unit_path_cache
= set_free_free(m
->unit_path_cache
);
2858 manager_check_finished(m
);
2860 /* There might still be some zombies hanging around from before we were exec()'ed. Let's reap them. */
2861 r
= sd_event_source_set_enabled(m
->sigchld_event_source
, SD_EVENT_ON
);
2863 return log_error_errno(r
, "Failed to enable SIGCHLD event source: %m");
2865 while (m
->objective
== MANAGER_OK
) {
2868 if (m
->runtime_watchdog
> 0 && m
->runtime_watchdog
!= USEC_INFINITY
&& MANAGER_IS_SYSTEM(m
))
2871 if (!ratelimit_below(&rl
)) {
2872 /* Yay, something is going seriously wrong, pause a little */
2873 log_warning("Looping too fast. Throttling execution a little.");
2877 if (manager_dispatch_load_queue(m
) > 0)
2880 if (manager_dispatch_gc_job_queue(m
) > 0)
2883 if (manager_dispatch_gc_unit_queue(m
) > 0)
2886 if (manager_dispatch_cleanup_queue(m
) > 0)
2889 if (manager_dispatch_cgroup_realize_queue(m
) > 0)
2892 if (manager_dispatch_stop_when_unneeded_queue(m
) > 0)
2895 if (manager_dispatch_dbus_queue(m
) > 0)
2898 /* Sleep for half the watchdog time */
2899 if (m
->runtime_watchdog
> 0 && m
->runtime_watchdog
!= USEC_INFINITY
&& MANAGER_IS_SYSTEM(m
)) {
2900 wait_usec
= m
->runtime_watchdog
/ 2;
2904 wait_usec
= USEC_INFINITY
;
2906 r
= sd_event_run(m
->event
, wait_usec
);
2908 return log_error_errno(r
, "Failed to run event loop: %m");
2911 return m
->objective
;
2914 int manager_load_unit_from_dbus_path(Manager
*m
, const char *s
, sd_bus_error
*e
, Unit
**_u
) {
2915 _cleanup_free_
char *n
= NULL
;
2916 sd_id128_t invocation_id
;
2924 r
= unit_name_from_dbus_path(s
, &n
);
2928 /* Permit addressing units by invocation ID: if the passed bus path is suffixed by a 128bit ID then we use it
2929 * as invocation ID. */
2930 r
= sd_id128_from_string(n
, &invocation_id
);
2932 u
= hashmap_get(m
->units_by_invocation_id
, &invocation_id
);
2938 return sd_bus_error_setf(e
, BUS_ERROR_NO_UNIT_FOR_INVOCATION_ID
,
2939 "No unit with the specified invocation ID " SD_ID128_FORMAT_STR
" known.",
2940 SD_ID128_FORMAT_VAL(invocation_id
));
2943 /* If this didn't work, we check if this is a unit name */
2944 if (!unit_name_is_valid(n
, UNIT_NAME_PLAIN
|UNIT_NAME_INSTANCE
)) {
2945 _cleanup_free_
char *nn
= NULL
;
2948 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
,
2949 "Unit name %s is neither a valid invocation ID nor unit name.", strnull(nn
));
2952 r
= manager_load_unit(m
, n
, NULL
, e
, &u
);
2960 int manager_get_job_from_dbus_path(Manager
*m
, const char *s
, Job
**_j
) {
2970 p
= startswith(s
, "/org/freedesktop/systemd1/job/");
2974 r
= safe_atou(p
, &id
);
2978 j
= manager_get_job(m
, id
);
2987 void manager_send_unit_audit(Manager
*m
, Unit
*u
, int type
, bool success
) {
2990 _cleanup_free_
char *p
= NULL
;
2994 if (!MANAGER_IS_SYSTEM(m
))
2997 audit_fd
= get_audit_fd();
3001 /* Don't generate audit events if the service was already
3002 * started and we're just deserializing */
3003 if (MANAGER_IS_RELOADING(m
))
3006 if (u
->type
!= UNIT_SERVICE
)
3009 r
= unit_name_to_prefix_and_instance(u
->id
, &p
);
3011 log_error_errno(r
, "Failed to extract prefix and instance of unit name: %m");
3015 msg
= strjoina("unit=", p
);
3016 if (audit_log_user_comm_message(audit_fd
, type
, msg
, "systemd", NULL
, NULL
, NULL
, success
) < 0) {
3018 /* We aren't allowed to send audit messages?
3019 * Then let's not retry again. */
3022 log_warning_errno(errno
, "Failed to send audit message: %m");
3028 void manager_send_unit_plymouth(Manager
*m
, Unit
*u
) {
3029 static const union sockaddr_union sa
= PLYMOUTH_SOCKET
;
3030 _cleanup_free_
char *message
= NULL
;
3031 _cleanup_close_
int fd
= -1;
3034 /* Don't generate plymouth events if the service was already
3035 * started and we're just deserializing */
3036 if (MANAGER_IS_RELOADING(m
))
3039 if (!MANAGER_IS_SYSTEM(m
))
3042 if (detect_container() > 0)
3045 if (!IN_SET(u
->type
, UNIT_SERVICE
, UNIT_MOUNT
, UNIT_SWAP
))
3048 /* We set SOCK_NONBLOCK here so that we rather drop the
3049 * message then wait for plymouth */
3050 fd
= socket(AF_UNIX
, SOCK_STREAM
|SOCK_CLOEXEC
|SOCK_NONBLOCK
, 0);
3052 log_error_errno(errno
, "socket() failed: %m");
3056 if (connect(fd
, &sa
.sa
, SOCKADDR_UN_LEN(sa
.un
)) < 0) {
3057 if (!IN_SET(errno
, EPIPE
, EAGAIN
, ENOENT
, ECONNREFUSED
, ECONNRESET
, ECONNABORTED
))
3058 log_error_errno(errno
, "connect() failed: %m");
3062 if (asprintf(&message
, "U\002%c%s%n", (int) (strlen(u
->id
) + 1), u
->id
, &n
) < 0) {
3068 if (write(fd
, message
, n
+ 1) != n
+ 1)
3069 if (!IN_SET(errno
, EPIPE
, EAGAIN
, ENOENT
, ECONNREFUSED
, ECONNRESET
, ECONNABORTED
))
3070 log_error_errno(errno
, "Failed to write Plymouth message: %m");
3073 int manager_open_serialization(Manager
*m
, FILE **_f
) {
3079 fd
= open_serialization_fd("systemd-state");
3083 f
= fdopen(fd
, "w+");
3093 static bool manager_timestamp_shall_serialize(ManagerTimestamp t
) {
3098 /* The following timestamps only apply to the host system, hence only serialize them there */
3100 MANAGER_TIMESTAMP_USERSPACE
, MANAGER_TIMESTAMP_FINISH
,
3101 MANAGER_TIMESTAMP_SECURITY_START
, MANAGER_TIMESTAMP_SECURITY_FINISH
,
3102 MANAGER_TIMESTAMP_GENERATORS_START
, MANAGER_TIMESTAMP_GENERATORS_FINISH
,
3103 MANAGER_TIMESTAMP_UNITS_LOAD_START
, MANAGER_TIMESTAMP_UNITS_LOAD_FINISH
);
3106 int manager_serialize(
3110 bool switching_root
) {
3122 _cleanup_(manager_reloading_stopp
) _unused_ Manager
*reloading
= manager_reloading_start(m
);
3124 (void) serialize_item_format(f
, "current-job-id", "%" PRIu32
, m
->current_job_id
);
3125 (void) serialize_item_format(f
, "n-installed-jobs", "%u", m
->n_installed_jobs
);
3126 (void) serialize_item_format(f
, "n-failed-jobs", "%u", m
->n_failed_jobs
);
3127 (void) serialize_bool(f
, "taint-usr", m
->taint_usr
);
3128 (void) serialize_bool(f
, "ready-sent", m
->ready_sent
);
3129 (void) serialize_bool(f
, "taint-logged", m
->taint_logged
);
3130 (void) serialize_bool(f
, "service-watchdogs", m
->service_watchdogs
);
3132 t
= show_status_to_string(m
->show_status
);
3134 (void) serialize_item(f
, "show-status", t
);
3136 if (m
->log_level_overridden
)
3137 (void) serialize_item_format(f
, "log-level-override", "%i", log_get_max_level());
3138 if (m
->log_target_overridden
)
3139 (void) serialize_item(f
, "log-target-override", log_target_to_string(log_get_target()));
3141 for (q
= 0; q
< _MANAGER_TIMESTAMP_MAX
; q
++) {
3142 _cleanup_free_
char *joined
= NULL
;
3144 if (!manager_timestamp_shall_serialize(q
))
3147 joined
= strjoin(manager_timestamp_to_string(q
), "-timestamp");
3151 (void) serialize_dual_timestamp(f
, joined
, m
->timestamps
+ q
);
3154 if (!switching_root
)
3155 (void) serialize_strv(f
, "env", m
->client_environment
);
3157 if (m
->notify_fd
>= 0) {
3158 r
= serialize_fd(f
, fds
, "notify-fd", m
->notify_fd
);
3162 (void) serialize_item(f
, "notify-socket", m
->notify_socket
);
3165 if (m
->cgroups_agent_fd
>= 0) {
3166 r
= serialize_fd(f
, fds
, "cgroups-agent-fd", m
->cgroups_agent_fd
);
3171 if (m
->user_lookup_fds
[0] >= 0) {
3174 copy0
= fdset_put_dup(fds
, m
->user_lookup_fds
[0]);
3176 return log_error_errno(copy0
, "Failed to add user lookup fd to serialization: %m");
3178 copy1
= fdset_put_dup(fds
, m
->user_lookup_fds
[1]);
3180 return log_error_errno(copy1
, "Failed to add user lookup fd to serialization: %m");
3182 (void) serialize_item_format(f
, "user-lookup", "%i %i", copy0
, copy1
);
3185 bus_track_serialize(m
->subscribed
, f
, "subscribed");
3187 r
= dynamic_user_serialize(m
, f
, fds
);
3191 manager_serialize_uid_refs(m
, f
);
3192 manager_serialize_gid_refs(m
, f
);
3194 r
= exec_runtime_serialize(m
, f
, fds
);
3198 (void) fputc('\n', f
);
3200 HASHMAP_FOREACH_KEY(u
, t
, m
->units
, i
) {
3208 r
= unit_serialize(u
, f
, fds
, !switching_root
);
3213 r
= fflush_and_check(f
);
3215 return log_error_errno(r
, "Failed to flush serialization: %m");
3217 r
= bus_fdset_add_all(m
, fds
);
3219 return log_error_errno(r
, "Failed to add bus sockets to serialization: %m");
3224 static int manager_deserialize_one_unit(Manager
*m
, const char *name
, FILE *f
, FDSet
*fds
) {
3228 r
= manager_load_unit(m
, name
, NULL
, NULL
, &u
);
3232 return log_notice_errno(r
, "Failed to load unit \"%s\", skipping deserialization: %m", name
);
3235 r
= unit_deserialize(u
, f
, fds
);
3239 return log_notice_errno(r
, "Failed to deserialize unit \"%s\", skipping: %m", name
);
3245 static int manager_deserialize_units(Manager
*m
, FILE *f
, FDSet
*fds
) {
3246 const char *unit_name
;
3250 _cleanup_free_
char *line
= NULL
;
3252 r
= read_line(f
, LONG_LINE_MAX
, &line
);
3254 return log_error_errno(r
, "Failed to read serialization line: %m");
3258 unit_name
= strstrip(line
);
3260 r
= manager_deserialize_one_unit(m
, unit_name
, f
, fds
);
3264 r
= unit_deserialize_skip(f
);
3273 int manager_deserialize(Manager
*m
, FILE *f
, FDSet
*fds
) {
3279 log_debug("Deserializing state...");
3281 /* If we are not in reload mode yet, enter it now. Not that this is recursive, a caller might already have
3282 * increased it to non-zero, which is why we just increase it by one here and down again at the end of this
3284 _cleanup_(manager_reloading_stopp
) _unused_ Manager
*reloading
= manager_reloading_start(m
);
3287 _cleanup_free_
char *line
= NULL
;
3288 const char *val
, *l
;
3290 r
= read_line(f
, LONG_LINE_MAX
, &line
);
3292 return log_error_errno(r
, "Failed to read serialization line: %m");
3297 if (isempty(l
)) /* end marker */
3300 if ((val
= startswith(l
, "current-job-id="))) {
3303 if (safe_atou32(val
, &id
) < 0)
3304 log_notice("Failed to parse current job id value '%s', ignoring.", val
);
3306 m
->current_job_id
= MAX(m
->current_job_id
, id
);
3308 } else if ((val
= startswith(l
, "n-installed-jobs="))) {
3311 if (safe_atou32(val
, &n
) < 0)
3312 log_notice("Failed to parse installed jobs counter '%s', ignoring.", val
);
3314 m
->n_installed_jobs
+= n
;
3316 } else if ((val
= startswith(l
, "n-failed-jobs="))) {
3319 if (safe_atou32(val
, &n
) < 0)
3320 log_notice("Failed to parse failed jobs counter '%s', ignoring.", val
);
3322 m
->n_failed_jobs
+= n
;
3324 } else if ((val
= startswith(l
, "taint-usr="))) {
3327 b
= parse_boolean(val
);
3329 log_notice("Failed to parse taint /usr flag '%s', ignoring.", val
);
3331 m
->taint_usr
= m
->taint_usr
|| b
;
3333 } else if ((val
= startswith(l
, "ready-sent="))) {
3336 b
= parse_boolean(val
);
3338 log_notice("Failed to parse ready-sent flag '%s', ignoring.", val
);
3340 m
->ready_sent
= m
->ready_sent
|| b
;
3342 } else if ((val
= startswith(l
, "taint-logged="))) {
3345 b
= parse_boolean(val
);
3347 log_notice("Failed to parse taint-logged flag '%s', ignoring.", val
);
3349 m
->taint_logged
= m
->taint_logged
|| b
;
3351 } else if ((val
= startswith(l
, "service-watchdogs="))) {
3354 b
= parse_boolean(val
);
3356 log_notice("Failed to parse service-watchdogs flag '%s', ignoring.", val
);
3358 m
->service_watchdogs
= b
;
3360 } else if ((val
= startswith(l
, "show-status="))) {
3363 s
= show_status_from_string(val
);
3365 log_notice("Failed to parse show-status flag '%s', ignoring.", val
);
3367 manager_set_show_status(m
, s
);
3369 } else if ((val
= startswith(l
, "log-level-override="))) {
3372 level
= log_level_from_string(val
);
3374 log_notice("Failed to parse log-level-override value '%s', ignoring.", val
);
3376 manager_override_log_level(m
, level
);
3378 } else if ((val
= startswith(l
, "log-target-override="))) {
3381 target
= log_target_from_string(val
);
3383 log_notice("Failed to parse log-target-override value '%s', ignoring.", val
);
3385 manager_override_log_target(m
, target
);
3387 } else if (startswith(l
, "env=")) {
3388 r
= deserialize_environment(l
+ 4, &m
->client_environment
);
3390 log_notice_errno(r
, "Failed to parse environment entry: \"%s\", ignoring: %m", l
);
3392 } else if ((val
= startswith(l
, "notify-fd="))) {
3395 if (safe_atoi(val
, &fd
) < 0 || fd
< 0 || !fdset_contains(fds
, fd
))
3396 log_notice("Failed to parse notify fd, ignoring: \"%s\"", val
);
3398 m
->notify_event_source
= sd_event_source_unref(m
->notify_event_source
);
3399 safe_close(m
->notify_fd
);
3400 m
->notify_fd
= fdset_remove(fds
, fd
);
3403 } else if ((val
= startswith(l
, "notify-socket="))) {
3404 r
= free_and_strdup(&m
->notify_socket
, val
);
3408 } else if ((val
= startswith(l
, "cgroups-agent-fd="))) {
3411 if (safe_atoi(val
, &fd
) < 0 || fd
< 0 || !fdset_contains(fds
, fd
))
3412 log_notice("Failed to parse cgroups agent fd, ignoring.: %s", val
);
3414 m
->cgroups_agent_event_source
= sd_event_source_unref(m
->cgroups_agent_event_source
);
3415 safe_close(m
->cgroups_agent_fd
);
3416 m
->cgroups_agent_fd
= fdset_remove(fds
, fd
);
3419 } else if ((val
= startswith(l
, "user-lookup="))) {
3422 if (sscanf(val
, "%i %i", &fd0
, &fd1
) != 2 || fd0
< 0 || fd1
< 0 || fd0
== fd1
|| !fdset_contains(fds
, fd0
) || !fdset_contains(fds
, fd1
))
3423 log_notice("Failed to parse user lookup fd, ignoring: %s", val
);
3425 m
->user_lookup_event_source
= sd_event_source_unref(m
->user_lookup_event_source
);
3426 safe_close_pair(m
->user_lookup_fds
);
3427 m
->user_lookup_fds
[0] = fdset_remove(fds
, fd0
);
3428 m
->user_lookup_fds
[1] = fdset_remove(fds
, fd1
);
3431 } else if ((val
= startswith(l
, "dynamic-user=")))
3432 dynamic_user_deserialize_one(m
, val
, fds
);
3433 else if ((val
= startswith(l
, "destroy-ipc-uid=")))
3434 manager_deserialize_uid_refs_one(m
, val
);
3435 else if ((val
= startswith(l
, "destroy-ipc-gid=")))
3436 manager_deserialize_gid_refs_one(m
, val
);
3437 else if ((val
= startswith(l
, "exec-runtime=")))
3438 exec_runtime_deserialize_one(m
, val
, fds
);
3439 else if ((val
= startswith(l
, "subscribed="))) {
3441 if (strv_extend(&m
->deserialized_subscribed
, val
) < 0)
3447 for (q
= 0; q
< _MANAGER_TIMESTAMP_MAX
; q
++) {
3448 val
= startswith(l
, manager_timestamp_to_string(q
));
3452 val
= startswith(val
, "-timestamp=");
3457 if (q
< _MANAGER_TIMESTAMP_MAX
) /* found it */
3458 (void) deserialize_dual_timestamp(val
, m
->timestamps
+ q
);
3459 else if (!startswith(l
, "kdbus-fd=")) /* ignore kdbus */
3460 log_notice("Unknown serialization item '%s', ignoring.", l
);
3464 return manager_deserialize_units(m
, f
, fds
);
3467 int manager_reload(Manager
*m
) {
3468 _cleanup_(manager_reloading_stopp
) Manager
*reloading
= NULL
;
3469 _cleanup_fdset_free_ FDSet
*fds
= NULL
;
3470 _cleanup_fclose_
FILE *f
= NULL
;
3475 r
= manager_open_serialization(m
, &f
);
3477 return log_error_errno(r
, "Failed to create serialization file: %m");
3483 /* We are officially in reload mode from here on. */
3484 reloading
= manager_reloading_start(m
);
3486 r
= manager_serialize(m
, f
, fds
, false);
3490 if (fseeko(f
, 0, SEEK_SET
) < 0)
3491 return log_error_errno(errno
, "Failed to seek to beginning of serialization: %m");
3493 /* 💀 This is the point of no return, from here on there is no way back. 💀 */
3496 bus_manager_send_reloading(m
, true);
3498 /* Start by flushing out all jobs and units, all generated units, all runtime environments, all dynamic users
3499 * and everything else that is worth flushing out. We'll get it all back from the serialization — if we need
3502 manager_clear_jobs_and_units(m
);
3503 lookup_paths_flush_generator(&m
->lookup_paths
);
3504 lookup_paths_free(&m
->lookup_paths
);
3505 exec_runtime_vacuum(m
);
3506 dynamic_user_vacuum(m
, false);
3507 m
->uid_refs
= hashmap_free(m
->uid_refs
);
3508 m
->gid_refs
= hashmap_free(m
->gid_refs
);
3510 r
= lookup_paths_init(&m
->lookup_paths
, m
->unit_file_scope
, 0, NULL
);
3512 log_warning_errno(r
, "Failed to initialize path lookup table, ignoring: %m");
3514 (void) manager_run_environment_generators(m
);
3515 (void) manager_run_generators(m
);
3517 r
= lookup_paths_reduce(&m
->lookup_paths
);
3519 log_warning_errno(r
, "Failed ot reduce unit file paths, ignoring: %m");
3521 manager_build_unit_path_cache(m
);
3523 /* First, enumerate what we can from kernel and suchlike */
3524 manager_enumerate_perpetual(m
);
3525 manager_enumerate(m
);
3527 /* Second, deserialize our stored data */
3528 r
= manager_deserialize(m
, f
, fds
);
3530 log_warning_errno(r
, "Deserialization failed, proceeding anyway: %m");
3532 /* We don't need the serialization anymore */
3535 /* Re-register notify_fd as event source, and set up other sockets/communication channels we might need */
3536 (void) manager_setup_notify(m
);
3537 (void) manager_setup_cgroups_agent(m
);
3538 (void) manager_setup_user_lookup_fd(m
);
3540 /* Third, fire things up! */
3541 manager_coldplug(m
);
3543 /* Clean up runtime objects no longer referenced */
3546 /* Consider the reload process complete now. */
3547 assert(m
->n_reloading
> 0);
3552 m
->send_reloading_done
= true;
3556 void manager_reset_failed(Manager
*m
) {
3562 HASHMAP_FOREACH(u
, m
->units
, i
)
3563 unit_reset_failed(u
);
3566 bool manager_unit_inactive_or_pending(Manager
*m
, const char *name
) {
3572 /* Returns true if the unit is inactive or going down */
3573 u
= manager_get_unit(m
, name
);
3577 return unit_inactive_or_pending(u
);
3580 static void log_taint_string(Manager
*m
) {
3581 _cleanup_free_
char *taint
= NULL
;
3585 if (MANAGER_IS_USER(m
) || m
->taint_logged
)
3588 m
->taint_logged
= true; /* only check for taint once */
3590 taint
= manager_taint_string(m
);
3594 log_struct(LOG_NOTICE
,
3595 LOG_MESSAGE("System is tainted: %s", taint
),
3597 "MESSAGE_ID=" SD_MESSAGE_TAINTED_STR
);
3600 static void manager_notify_finished(Manager
*m
) {
3601 char userspace
[FORMAT_TIMESPAN_MAX
], initrd
[FORMAT_TIMESPAN_MAX
], kernel
[FORMAT_TIMESPAN_MAX
], sum
[FORMAT_TIMESPAN_MAX
];
3602 usec_t firmware_usec
, loader_usec
, kernel_usec
, initrd_usec
, userspace_usec
, total_usec
;
3604 if (MANAGER_IS_TEST_RUN(m
))
3607 if (MANAGER_IS_SYSTEM(m
) && detect_container() <= 0) {
3608 char ts
[FORMAT_TIMESPAN_MAX
];
3609 char buf
[FORMAT_TIMESPAN_MAX
+ STRLEN(" (firmware) + ") + FORMAT_TIMESPAN_MAX
+ STRLEN(" (loader) + ")]
3612 size_t size
= sizeof buf
;
3614 /* Note that MANAGER_TIMESTAMP_KERNEL's monotonic value is always at 0, and
3615 * MANAGER_TIMESTAMP_FIRMWARE's and MANAGER_TIMESTAMP_LOADER's monotonic value should be considered
3616 * negative values. */
3618 firmware_usec
= m
->timestamps
[MANAGER_TIMESTAMP_FIRMWARE
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_LOADER
].monotonic
;
3619 loader_usec
= m
->timestamps
[MANAGER_TIMESTAMP_LOADER
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_KERNEL
].monotonic
;
3620 userspace_usec
= m
->timestamps
[MANAGER_TIMESTAMP_FINISH
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_USERSPACE
].monotonic
;
3621 total_usec
= m
->timestamps
[MANAGER_TIMESTAMP_FIRMWARE
].monotonic
+ m
->timestamps
[MANAGER_TIMESTAMP_FINISH
].monotonic
;
3623 if (firmware_usec
> 0)
3624 size
= strpcpyf(&p
, size
, "%s (firmware) + ", format_timespan(ts
, sizeof(ts
), firmware_usec
, USEC_PER_MSEC
));
3625 if (loader_usec
> 0)
3626 size
= strpcpyf(&p
, size
, "%s (loader) + ", format_timespan(ts
, sizeof(ts
), loader_usec
, USEC_PER_MSEC
));
3628 if (dual_timestamp_is_set(&m
->timestamps
[MANAGER_TIMESTAMP_INITRD
])) {
3630 /* The initrd case on bare-metal*/
3631 kernel_usec
= m
->timestamps
[MANAGER_TIMESTAMP_INITRD
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_KERNEL
].monotonic
;
3632 initrd_usec
= m
->timestamps
[MANAGER_TIMESTAMP_USERSPACE
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_INITRD
].monotonic
;
3634 log_struct(LOG_INFO
,
3635 "MESSAGE_ID=" SD_MESSAGE_STARTUP_FINISHED_STR
,
3636 "KERNEL_USEC="USEC_FMT
, kernel_usec
,
3637 "INITRD_USEC="USEC_FMT
, initrd_usec
,
3638 "USERSPACE_USEC="USEC_FMT
, userspace_usec
,
3639 LOG_MESSAGE("Startup finished in %s%s (kernel) + %s (initrd) + %s (userspace) = %s.",
3641 format_timespan(kernel
, sizeof(kernel
), kernel_usec
, USEC_PER_MSEC
),
3642 format_timespan(initrd
, sizeof(initrd
), initrd_usec
, USEC_PER_MSEC
),
3643 format_timespan(userspace
, sizeof(userspace
), userspace_usec
, USEC_PER_MSEC
),
3644 format_timespan(sum
, sizeof(sum
), total_usec
, USEC_PER_MSEC
)));
3646 /* The initrd-less case on bare-metal*/
3648 kernel_usec
= m
->timestamps
[MANAGER_TIMESTAMP_USERSPACE
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_KERNEL
].monotonic
;
3651 log_struct(LOG_INFO
,
3652 "MESSAGE_ID=" SD_MESSAGE_STARTUP_FINISHED_STR
,
3653 "KERNEL_USEC="USEC_FMT
, kernel_usec
,
3654 "USERSPACE_USEC="USEC_FMT
, userspace_usec
,
3655 LOG_MESSAGE("Startup finished in %s%s (kernel) + %s (userspace) = %s.",
3657 format_timespan(kernel
, sizeof(kernel
), kernel_usec
, USEC_PER_MSEC
),
3658 format_timespan(userspace
, sizeof(userspace
), userspace_usec
, USEC_PER_MSEC
),
3659 format_timespan(sum
, sizeof(sum
), total_usec
, USEC_PER_MSEC
)));
3662 /* The container and --user case */
3663 firmware_usec
= loader_usec
= initrd_usec
= kernel_usec
= 0;
3664 total_usec
= userspace_usec
= m
->timestamps
[MANAGER_TIMESTAMP_FINISH
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_USERSPACE
].monotonic
;
3666 log_struct(LOG_INFO
,
3667 "MESSAGE_ID=" SD_MESSAGE_USER_STARTUP_FINISHED_STR
,
3668 "USERSPACE_USEC="USEC_FMT
, userspace_usec
,
3669 LOG_MESSAGE("Startup finished in %s.",
3670 format_timespan(sum
, sizeof(sum
), total_usec
, USEC_PER_MSEC
)));
3673 bus_manager_send_finished(m
, firmware_usec
, loader_usec
, kernel_usec
, initrd_usec
, userspace_usec
, total_usec
);
3676 m
->ready_sent
? "STATUS=Startup finished in %s."
3678 "STATUS=Startup finished in %s.",
3679 format_timespan(sum
, sizeof(sum
), total_usec
, USEC_PER_MSEC
));
3680 m
->ready_sent
= true;
3682 log_taint_string(m
);
3685 static void manager_send_ready(Manager
*m
) {
3688 /* We send READY=1 on reaching basic.target only when running in --user mode. */
3689 if (!MANAGER_IS_USER(m
) || m
->ready_sent
)
3692 m
->ready_sent
= true;
3696 "STATUS=Reached " SPECIAL_BASIC_TARGET
".");
3699 static void manager_check_basic_target(Manager
*m
) {
3704 /* Small shortcut */
3705 if (m
->ready_sent
&& m
->taint_logged
)
3708 u
= manager_get_unit(m
, SPECIAL_BASIC_TARGET
);
3709 if (!u
|| !UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u
)))
3712 /* For user managers, send out READY=1 as soon as we reach basic.target */
3713 manager_send_ready(m
);
3715 /* Log the taint string as soon as we reach basic.target */
3716 log_taint_string(m
);
3719 void manager_check_finished(Manager
*m
) {
3722 if (MANAGER_IS_RELOADING(m
))
3725 /* Verify that we have entered the event loop already, and not left it again. */
3726 if (!MANAGER_IS_RUNNING(m
))
3729 manager_check_basic_target(m
);
3731 if (hashmap_size(m
->jobs
) > 0) {
3732 if (m
->jobs_in_progress_event_source
)
3733 /* Ignore any failure, this is only for feedback */
3734 (void) sd_event_source_set_time(m
->jobs_in_progress_event_source
, now(CLOCK_MONOTONIC
) + JOBS_IN_PROGRESS_WAIT_USEC
);
3739 manager_flip_auto_status(m
, false);
3741 /* Notify Type=idle units that we are done now */
3742 manager_close_idle_pipe(m
);
3744 /* Turn off confirm spawn now */
3745 m
->confirm_spawn
= NULL
;
3747 /* No need to update ask password status when we're going non-interactive */
3748 manager_close_ask_password(m
);
3750 /* This is no longer the first boot */
3751 manager_set_first_boot(m
, false);
3753 if (MANAGER_IS_FINISHED(m
))
3756 dual_timestamp_get(m
->timestamps
+ MANAGER_TIMESTAMP_FINISH
);
3758 manager_notify_finished(m
);
3760 manager_invalidate_startup_units(m
);
3763 static bool generator_path_any(const char* const* paths
) {
3767 /* Optimize by skipping the whole process by not creating output directories
3768 * if no generators are found. */
3769 STRV_FOREACH(path
, (char**) paths
)
3770 if (access(*path
, F_OK
) == 0)
3772 else if (errno
!= ENOENT
)
3773 log_warning_errno(errno
, "Failed to open generator directory %s: %m", *path
);
3778 static const char* system_env_generator_binary_paths
[] = {
3779 "/run/systemd/system-environment-generators",
3780 "/etc/systemd/system-environment-generators",
3781 "/usr/local/lib/systemd/system-environment-generators",
3782 SYSTEM_ENV_GENERATOR_PATH
,
3786 static const char* user_env_generator_binary_paths
[] = {
3787 "/run/systemd/user-environment-generators",
3788 "/etc/systemd/user-environment-generators",
3789 "/usr/local/lib/systemd/user-environment-generators",
3790 USER_ENV_GENERATOR_PATH
,
3794 static int manager_run_environment_generators(Manager
*m
) {
3795 char **tmp
= NULL
; /* this is only used in the forked process, no cleanup here */
3798 [STDOUT_GENERATE
] = &tmp
,
3799 [STDOUT_COLLECT
] = &tmp
,
3800 [STDOUT_CONSUME
] = &m
->transient_environment
,
3804 if (MANAGER_IS_TEST_RUN(m
) && !(m
->test_run_flags
& MANAGER_TEST_RUN_ENV_GENERATORS
))
3807 paths
= MANAGER_IS_SYSTEM(m
) ? system_env_generator_binary_paths
: user_env_generator_binary_paths
;
3809 if (!generator_path_any(paths
))
3812 RUN_WITH_UMASK(0022)
3813 r
= execute_directories(paths
, DEFAULT_TIMEOUT_USEC
, gather_environment
, args
, NULL
, m
->transient_environment
);
3818 static int manager_run_generators(Manager
*m
) {
3819 _cleanup_strv_free_
char **paths
= NULL
;
3820 const char *argv
[5];
3825 if (MANAGER_IS_TEST_RUN(m
) && !(m
->test_run_flags
& MANAGER_TEST_RUN_GENERATORS
))
3828 paths
= generator_binary_paths(m
->unit_file_scope
);
3832 if (!generator_path_any((const char* const*) paths
))
3835 r
= lookup_paths_mkdir_generator(&m
->lookup_paths
);
3837 log_error_errno(r
, "Failed to create generator directories: %m");
3841 argv
[0] = NULL
; /* Leave this empty, execute_directory() will fill something in */
3842 argv
[1] = m
->lookup_paths
.generator
;
3843 argv
[2] = m
->lookup_paths
.generator_early
;
3844 argv
[3] = m
->lookup_paths
.generator_late
;
3847 RUN_WITH_UMASK(0022)
3848 (void) execute_directories((const char* const*) paths
, DEFAULT_TIMEOUT_USEC
,
3849 NULL
, NULL
, (char**) argv
, m
->transient_environment
);
3854 lookup_paths_trim_generator(&m
->lookup_paths
);
3858 int manager_transient_environment_add(Manager
*m
, char **plus
) {
3863 if (strv_isempty(plus
))
3866 a
= strv_env_merge(2, m
->transient_environment
, plus
);
3870 sanitize_environment(a
);
3872 return strv_free_and_replace(m
->transient_environment
, a
);
3875 int manager_client_environment_modify(
3880 char **a
= NULL
, **b
= NULL
, **l
;
3884 if (strv_isempty(minus
) && strv_isempty(plus
))
3887 l
= m
->client_environment
;
3889 if (!strv_isempty(minus
)) {
3890 a
= strv_env_delete(l
, 1, minus
);
3897 if (!strv_isempty(plus
)) {
3898 b
= strv_env_merge(2, l
, plus
);
3907 if (m
->client_environment
!= l
)
3908 strv_free(m
->client_environment
);
3915 m
->client_environment
= sanitize_environment(l
);
3919 int manager_get_effective_environment(Manager
*m
, char ***ret
) {
3925 l
= strv_env_merge(2, m
->transient_environment
, m
->client_environment
);
3933 int manager_set_default_rlimits(Manager
*m
, struct rlimit
**default_rlimit
) {
3938 for (i
= 0; i
< _RLIMIT_MAX
; i
++) {
3939 m
->rlimit
[i
] = mfree(m
->rlimit
[i
]);
3941 if (!default_rlimit
[i
])
3944 m
->rlimit
[i
] = newdup(struct rlimit
, default_rlimit
[i
], 1);
3952 void manager_recheck_dbus(Manager
*m
) {
3955 /* Connects to the bus if the dbus service and socket are running. If we are running in user mode this is all
3956 * it does. In system mode we'll also connect to the system bus (which will most likely just reuse the
3957 * connection of the API bus). That's because the system bus after all runs as service of the system instance,
3958 * while in the user instance we can assume it's already there. */
3960 if (MANAGER_IS_RELOADING(m
))
3961 return; /* don't check while we are reloading… */
3963 if (manager_dbus_is_running(m
, false)) {
3964 (void) bus_init_api(m
);
3966 if (MANAGER_IS_SYSTEM(m
))
3967 (void) bus_init_system(m
);
3969 (void) bus_done_api(m
);
3971 if (MANAGER_IS_SYSTEM(m
))
3972 (void) bus_done_system(m
);
3976 static bool manager_journal_is_running(Manager
*m
) {
3981 if (MANAGER_IS_TEST_RUN(m
))
3984 /* If we are the user manager we can safely assume that the journal is up */
3985 if (!MANAGER_IS_SYSTEM(m
))
3988 /* Check that the socket is not only up, but in RUNNING state */
3989 u
= manager_get_unit(m
, SPECIAL_JOURNALD_SOCKET
);
3992 if (SOCKET(u
)->state
!= SOCKET_RUNNING
)
3995 /* Similar, check if the daemon itself is fully up, too */
3996 u
= manager_get_unit(m
, SPECIAL_JOURNALD_SERVICE
);
3999 if (!IN_SET(SERVICE(u
)->state
, SERVICE_RELOAD
, SERVICE_RUNNING
))
4005 void manager_recheck_journal(Manager
*m
) {
4009 /* Don't bother with this unless we are in the special situation of being PID 1 */
4010 if (getpid_cached() != 1)
4013 /* Don't check this while we are reloading, things might still change */
4014 if (MANAGER_IS_RELOADING(m
))
4017 /* The journal is fully and entirely up? If so, let's permit logging to it, if that's configured. If the
4018 * journal is down, don't ever log to it, otherwise we might end up deadlocking ourselves as we might trigger
4019 * an activation ourselves we can't fulfill. */
4020 log_set_prohibit_ipc(!manager_journal_is_running(m
));
4024 void manager_set_show_status(Manager
*m
, ShowStatus mode
) {
4026 assert(IN_SET(mode
, SHOW_STATUS_AUTO
, SHOW_STATUS_NO
, SHOW_STATUS_YES
, SHOW_STATUS_TEMPORARY
));
4028 if (!MANAGER_IS_SYSTEM(m
))
4031 if (m
->show_status
!= mode
)
4032 log_debug("%s showing of status.",
4033 mode
== SHOW_STATUS_NO
? "Disabling" : "Enabling");
4034 m
->show_status
= mode
;
4036 if (IN_SET(mode
, SHOW_STATUS_TEMPORARY
, SHOW_STATUS_YES
))
4037 (void) touch("/run/systemd/show-status");
4039 (void) unlink("/run/systemd/show-status");
4042 static bool manager_get_show_status(Manager
*m
, StatusType type
) {
4045 if (!MANAGER_IS_SYSTEM(m
))
4048 if (m
->no_console_output
)
4051 if (!IN_SET(manager_state(m
), MANAGER_INITIALIZING
, MANAGER_STARTING
, MANAGER_STOPPING
))
4054 /* If we cannot find out the status properly, just proceed. */
4055 if (type
!= STATUS_TYPE_EMERGENCY
&& manager_check_ask_password(m
) > 0)
4058 return IN_SET(m
->show_status
, SHOW_STATUS_TEMPORARY
, SHOW_STATUS_YES
);
4061 const char *manager_get_confirm_spawn(Manager
*m
) {
4062 static int last_errno
= 0;
4063 const char *vc
= m
->confirm_spawn
;
4067 /* Here's the deal: we want to test the validity of the console but don't want
4068 * PID1 to go through the whole console process which might block. But we also
4069 * want to warn the user only once if something is wrong with the console so we
4070 * cannot do the sanity checks after spawning our children. So here we simply do
4071 * really basic tests to hopefully trap common errors.
4073 * If the console suddenly disappear at the time our children will really it
4074 * then they will simply fail to acquire it and a positive answer will be
4075 * assumed. New children will fallback to /dev/console though.
4077 * Note: TTYs are devices that can come and go any time, and frequently aren't
4078 * available yet during early boot (consider a USB rs232 dongle...). If for any
4079 * reason the configured console is not ready, we fallback to the default
4082 if (!vc
|| path_equal(vc
, "/dev/console"))
4089 if (!S_ISCHR(st
.st_mode
)) {
4097 if (last_errno
!= errno
) {
4099 log_warning_errno(errno
, "Failed to open %s: %m, using default console", vc
);
4101 return "/dev/console";
4104 void manager_set_first_boot(Manager
*m
, bool b
) {
4107 if (!MANAGER_IS_SYSTEM(m
))
4110 if (m
->first_boot
!= (int) b
) {
4112 (void) touch("/run/systemd/first-boot");
4114 (void) unlink("/run/systemd/first-boot");
4120 void manager_disable_confirm_spawn(void) {
4121 (void) touch("/run/systemd/confirm_spawn_disabled");
4124 bool manager_is_confirm_spawn_disabled(Manager
*m
) {
4125 if (!m
->confirm_spawn
)
4128 return access("/run/systemd/confirm_spawn_disabled", F_OK
) >= 0;
4131 void manager_status_printf(Manager
*m
, StatusType type
, const char *status
, const char *format
, ...) {
4134 /* If m is NULL, assume we're after shutdown and let the messages through. */
4136 if (m
&& !manager_get_show_status(m
, type
))
4139 /* XXX We should totally drop the check for ephemeral here
4140 * and thus effectively make 'Type=idle' pointless. */
4141 if (type
== STATUS_TYPE_EPHEMERAL
&& m
&& m
->n_on_console
> 0)
4144 va_start(ap
, format
);
4145 status_vprintf(status
, SHOW_STATUS_ELLIPSIZE
|(type
== STATUS_TYPE_EPHEMERAL
? SHOW_STATUS_EPHEMERAL
: 0), format
, ap
);
4149 Set
*manager_get_units_requiring_mounts_for(Manager
*m
, const char *path
) {
4150 char p
[strlen(path
)+1];
4156 path_simplify(p
, false);
4158 return hashmap_get(m
->units_requiring_mounts_for
, streq(p
, "/") ? "" : p
);
4161 int manager_update_failed_units(Manager
*m
, Unit
*u
, bool failed
) {
4166 assert(u
->manager
== m
);
4168 size
= set_size(m
->failed_units
);
4171 r
= set_ensure_allocated(&m
->failed_units
, NULL
);
4175 if (set_put(m
->failed_units
, u
) < 0)
4178 (void) set_remove(m
->failed_units
, u
);
4180 if (set_size(m
->failed_units
) != size
)
4181 bus_manager_send_change_signal(m
);
4186 ManagerState
manager_state(Manager
*m
) {
4191 /* Did we ever finish booting? If not then we are still starting up */
4192 if (!MANAGER_IS_FINISHED(m
)) {
4194 u
= manager_get_unit(m
, SPECIAL_BASIC_TARGET
);
4195 if (!u
|| !UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u
)))
4196 return MANAGER_INITIALIZING
;
4198 return MANAGER_STARTING
;
4201 /* Is the special shutdown target active or queued? If so, we are in shutdown state */
4202 u
= manager_get_unit(m
, SPECIAL_SHUTDOWN_TARGET
);
4203 if (u
&& unit_active_or_pending(u
))
4204 return MANAGER_STOPPING
;
4206 if (MANAGER_IS_SYSTEM(m
)) {
4207 /* Are the rescue or emergency targets active or queued? If so we are in maintenance state */
4208 u
= manager_get_unit(m
, SPECIAL_RESCUE_TARGET
);
4209 if (u
&& unit_active_or_pending(u
))
4210 return MANAGER_MAINTENANCE
;
4212 u
= manager_get_unit(m
, SPECIAL_EMERGENCY_TARGET
);
4213 if (u
&& unit_active_or_pending(u
))
4214 return MANAGER_MAINTENANCE
;
4217 /* Are there any failed units? If so, we are in degraded mode */
4218 if (set_size(m
->failed_units
) > 0)
4219 return MANAGER_DEGRADED
;
4221 return MANAGER_RUNNING
;
4224 #define DESTROY_IPC_FLAG (UINT32_C(1) << 31)
4226 static void manager_unref_uid_internal(
4231 int (*_clean_ipc
)(uid_t uid
)) {
4237 assert(uid_is_valid(uid
));
4240 /* A generic implementation, covering both manager_unref_uid() and manager_unref_gid(), under the assumption
4241 * that uid_t and gid_t are actually defined the same way, with the same validity rules.
4243 * We store a hashmap where the UID/GID is they key and the value is a 32bit reference counter, whose highest
4244 * bit is used as flag for marking UIDs/GIDs whose IPC objects to remove when the last reference to the UID/GID
4245 * is dropped. The flag is set to on, once at least one reference from a unit where RemoveIPC= is set is added
4246 * on a UID/GID. It is reset when the UID's/GID's reference counter drops to 0 again. */
4248 assert_cc(sizeof(uid_t
) == sizeof(gid_t
));
4249 assert_cc(UID_INVALID
== (uid_t
) GID_INVALID
);
4251 if (uid
== 0) /* We don't keep track of root, and will never destroy it */
4254 c
= PTR_TO_UINT32(hashmap_get(*uid_refs
, UID_TO_PTR(uid
)));
4256 n
= c
& ~DESTROY_IPC_FLAG
;
4260 if (destroy_now
&& n
== 0) {
4261 hashmap_remove(*uid_refs
, UID_TO_PTR(uid
));
4263 if (c
& DESTROY_IPC_FLAG
) {
4264 log_debug("%s " UID_FMT
" is no longer referenced, cleaning up its IPC.",
4265 _clean_ipc
== clean_ipc_by_uid
? "UID" : "GID",
4267 (void) _clean_ipc(uid
);
4270 c
= n
| (c
& DESTROY_IPC_FLAG
);
4271 assert_se(hashmap_update(*uid_refs
, UID_TO_PTR(uid
), UINT32_TO_PTR(c
)) >= 0);
4275 void manager_unref_uid(Manager
*m
, uid_t uid
, bool destroy_now
) {
4276 manager_unref_uid_internal(m
, &m
->uid_refs
, uid
, destroy_now
, clean_ipc_by_uid
);
4279 void manager_unref_gid(Manager
*m
, gid_t gid
, bool destroy_now
) {
4280 manager_unref_uid_internal(m
, &m
->gid_refs
, (uid_t
) gid
, destroy_now
, clean_ipc_by_gid
);
4283 static int manager_ref_uid_internal(
4294 assert(uid_is_valid(uid
));
4296 /* A generic implementation, covering both manager_ref_uid() and manager_ref_gid(), under the assumption
4297 * that uid_t and gid_t are actually defined the same way, with the same validity rules. */
4299 assert_cc(sizeof(uid_t
) == sizeof(gid_t
));
4300 assert_cc(UID_INVALID
== (uid_t
) GID_INVALID
);
4302 if (uid
== 0) /* We don't keep track of root, and will never destroy it */
4305 r
= hashmap_ensure_allocated(uid_refs
, &trivial_hash_ops
);
4309 c
= PTR_TO_UINT32(hashmap_get(*uid_refs
, UID_TO_PTR(uid
)));
4311 n
= c
& ~DESTROY_IPC_FLAG
;
4314 if (n
& DESTROY_IPC_FLAG
) /* check for overflow */
4317 c
= n
| (c
& DESTROY_IPC_FLAG
) | (clean_ipc
? DESTROY_IPC_FLAG
: 0);
4319 return hashmap_replace(*uid_refs
, UID_TO_PTR(uid
), UINT32_TO_PTR(c
));
4322 int manager_ref_uid(Manager
*m
, uid_t uid
, bool clean_ipc
) {
4323 return manager_ref_uid_internal(m
, &m
->uid_refs
, uid
, clean_ipc
);
4326 int manager_ref_gid(Manager
*m
, gid_t gid
, bool clean_ipc
) {
4327 return manager_ref_uid_internal(m
, &m
->gid_refs
, (uid_t
) gid
, clean_ipc
);
4330 static void manager_vacuum_uid_refs_internal(
4333 int (*_clean_ipc
)(uid_t uid
)) {
4342 HASHMAP_FOREACH_KEY(p
, k
, *uid_refs
, i
) {
4346 uid
= PTR_TO_UID(k
);
4347 c
= PTR_TO_UINT32(p
);
4349 n
= c
& ~DESTROY_IPC_FLAG
;
4353 if (c
& DESTROY_IPC_FLAG
) {
4354 log_debug("Found unreferenced %s " UID_FMT
" after reload/reexec. Cleaning up.",
4355 _clean_ipc
== clean_ipc_by_uid
? "UID" : "GID",
4357 (void) _clean_ipc(uid
);
4360 assert_se(hashmap_remove(*uid_refs
, k
) == p
);
4364 void manager_vacuum_uid_refs(Manager
*m
) {
4365 manager_vacuum_uid_refs_internal(m
, &m
->uid_refs
, clean_ipc_by_uid
);
4368 void manager_vacuum_gid_refs(Manager
*m
) {
4369 manager_vacuum_uid_refs_internal(m
, &m
->gid_refs
, clean_ipc_by_gid
);
4372 static void manager_serialize_uid_refs_internal(
4376 const char *field_name
) {
4386 /* Serialize the UID reference table. Or actually, just the IPC destruction flag of it, as the actual counter
4387 * of it is better rebuild after a reload/reexec. */
4389 HASHMAP_FOREACH_KEY(p
, k
, *uid_refs
, i
) {
4393 uid
= PTR_TO_UID(k
);
4394 c
= PTR_TO_UINT32(p
);
4396 if (!(c
& DESTROY_IPC_FLAG
))
4399 (void) serialize_item_format(f
, field_name
, UID_FMT
, uid
);
4403 void manager_serialize_uid_refs(Manager
*m
, FILE *f
) {
4404 manager_serialize_uid_refs_internal(m
, f
, &m
->uid_refs
, "destroy-ipc-uid");
4407 void manager_serialize_gid_refs(Manager
*m
, FILE *f
) {
4408 manager_serialize_uid_refs_internal(m
, f
, &m
->gid_refs
, "destroy-ipc-gid");
4411 static void manager_deserialize_uid_refs_one_internal(
4414 const char *value
) {
4424 r
= parse_uid(value
, &uid
);
4425 if (r
< 0 || uid
== 0) {
4426 log_debug("Unable to parse UID reference serialization");
4430 r
= hashmap_ensure_allocated(uid_refs
, &trivial_hash_ops
);
4436 c
= PTR_TO_UINT32(hashmap_get(*uid_refs
, UID_TO_PTR(uid
)));
4437 if (c
& DESTROY_IPC_FLAG
)
4440 c
|= DESTROY_IPC_FLAG
;
4442 r
= hashmap_replace(*uid_refs
, UID_TO_PTR(uid
), UINT32_TO_PTR(c
));
4444 log_debug_errno(r
, "Failed to add UID reference entry: %m");
4449 void manager_deserialize_uid_refs_one(Manager
*m
, const char *value
) {
4450 manager_deserialize_uid_refs_one_internal(m
, &m
->uid_refs
, value
);
4453 void manager_deserialize_gid_refs_one(Manager
*m
, const char *value
) {
4454 manager_deserialize_uid_refs_one_internal(m
, &m
->gid_refs
, value
);
4457 int manager_dispatch_user_lookup_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
4461 char unit_name
[UNIT_NAME_MAX
+1];
4464 Manager
*m
= userdata
;
4472 /* Invoked whenever a child process succeeded resolving its user/group to use and sent us the resulting UID/GID
4473 * in a datagram. We parse the datagram here and pass it off to the unit, so that it can add a reference to the
4474 * UID/GID so that it can destroy the UID/GID's IPC objects when the reference counter drops to 0. */
4476 l
= recv(fd
, &buffer
, sizeof(buffer
), MSG_DONTWAIT
);
4478 if (IN_SET(errno
, EINTR
, EAGAIN
))
4481 return log_error_errno(errno
, "Failed to read from user lookup fd: %m");
4484 if ((size_t) l
<= offsetof(struct buffer
, unit_name
)) {
4485 log_warning("Received too short user lookup message, ignoring.");
4489 if ((size_t) l
> offsetof(struct buffer
, unit_name
) + UNIT_NAME_MAX
) {
4490 log_warning("Received too long user lookup message, ignoring.");
4494 if (!uid_is_valid(buffer
.uid
) && !gid_is_valid(buffer
.gid
)) {
4495 log_warning("Got user lookup message with invalid UID/GID pair, ignoring.");
4499 n
= (size_t) l
- offsetof(struct buffer
, unit_name
);
4500 if (memchr(buffer
.unit_name
, 0, n
)) {
4501 log_warning("Received lookup message with embedded NUL character, ignoring.");
4505 buffer
.unit_name
[n
] = 0;
4506 u
= manager_get_unit(m
, buffer
.unit_name
);
4508 log_debug("Got user lookup message but unit doesn't exist, ignoring.");
4512 log_unit_debug(u
, "User lookup succeeded: uid=" UID_FMT
" gid=" GID_FMT
, buffer
.uid
, buffer
.gid
);
4514 unit_notify_user_lookup(u
, buffer
.uid
, buffer
.gid
);
4518 char *manager_taint_string(Manager
*m
) {
4519 _cleanup_free_
char *destination
= NULL
, *overflowuid
= NULL
, *overflowgid
= NULL
;
4523 /* Returns a "taint string", e.g. "local-hwclock:var-run-bad".
4524 * Only things that are detected at runtime should be tagged
4525 * here. For stuff that is set during compilation, emit a warning
4526 * in the configuration phase. */
4530 buf
= new(char, sizeof("split-usr:"
4534 "overflowuid-not-65534:"
4535 "overflowgid-not-65534:"));
4543 e
= stpcpy(e
, "split-usr:");
4545 if (access("/proc/cgroups", F_OK
) < 0)
4546 e
= stpcpy(e
, "cgroups-missing:");
4548 if (clock_is_localtime(NULL
) > 0)
4549 e
= stpcpy(e
, "local-hwclock:");
4551 r
= readlink_malloc("/var/run", &destination
);
4552 if (r
< 0 || !PATH_IN_SET(destination
, "../run", "/run"))
4553 e
= stpcpy(e
, "var-run-bad:");
4555 r
= read_one_line_file("/proc/sys/kernel/overflowuid", &overflowuid
);
4556 if (r
>= 0 && !streq(overflowuid
, "65534"))
4557 e
= stpcpy(e
, "overflowuid-not-65534:");
4559 r
= read_one_line_file("/proc/sys/kernel/overflowgid", &overflowgid
);
4560 if (r
>= 0 && !streq(overflowgid
, "65534"))
4561 e
= stpcpy(e
, "overflowgid-not-65534:");
4563 /* remove the last ':' */
4570 void manager_ref_console(Manager
*m
) {
4576 void manager_unref_console(Manager
*m
) {
4578 assert(m
->n_on_console
> 0);
4581 if (m
->n_on_console
== 0)
4582 m
->no_console_output
= false; /* unset no_console_output flag, since the console is definitely free now */
4585 void manager_override_log_level(Manager
*m
, int level
) {
4586 _cleanup_free_
char *s
= NULL
;
4589 if (!m
->log_level_overridden
) {
4590 m
->original_log_level
= log_get_max_level();
4591 m
->log_level_overridden
= true;
4594 (void) log_level_to_string_alloc(level
, &s
);
4595 log_info("Setting log level to %s.", strna(s
));
4597 log_set_max_level(level
);
4600 void manager_restore_original_log_level(Manager
*m
) {
4601 _cleanup_free_
char *s
= NULL
;
4604 if (!m
->log_level_overridden
)
4607 (void) log_level_to_string_alloc(m
->original_log_level
, &s
);
4608 log_info("Restoring log level to original (%s).", strna(s
));
4610 log_set_max_level(m
->original_log_level
);
4611 m
->log_level_overridden
= false;
4614 void manager_override_log_target(Manager
*m
, LogTarget target
) {
4617 if (!m
->log_target_overridden
) {
4618 m
->original_log_target
= log_get_target();
4619 m
->log_target_overridden
= true;
4622 log_info("Setting log target to %s.", log_target_to_string(target
));
4623 log_set_target(target
);
4626 void manager_restore_original_log_target(Manager
*m
) {
4629 if (!m
->log_target_overridden
)
4632 log_info("Restoring log target to original %s.", log_target_to_string(m
->original_log_target
));
4634 log_set_target(m
->original_log_target
);
4635 m
->log_target_overridden
= false;
4638 ManagerTimestamp
manager_timestamp_initrd_mangle(ManagerTimestamp s
) {
4640 s
>= MANAGER_TIMESTAMP_SECURITY_START
&&
4641 s
<= MANAGER_TIMESTAMP_UNITS_LOAD_FINISH
)
4642 return s
- MANAGER_TIMESTAMP_SECURITY_START
+ MANAGER_TIMESTAMP_INITRD_SECURITY_START
;
4646 static const char *const manager_state_table
[_MANAGER_STATE_MAX
] = {
4647 [MANAGER_INITIALIZING
] = "initializing",
4648 [MANAGER_STARTING
] = "starting",
4649 [MANAGER_RUNNING
] = "running",
4650 [MANAGER_DEGRADED
] = "degraded",
4651 [MANAGER_MAINTENANCE
] = "maintenance",
4652 [MANAGER_STOPPING
] = "stopping",
4655 DEFINE_STRING_TABLE_LOOKUP(manager_state
, ManagerState
);
4657 static const char *const manager_timestamp_table
[_MANAGER_TIMESTAMP_MAX
] = {
4658 [MANAGER_TIMESTAMP_FIRMWARE
] = "firmware",
4659 [MANAGER_TIMESTAMP_LOADER
] = "loader",
4660 [MANAGER_TIMESTAMP_KERNEL
] = "kernel",
4661 [MANAGER_TIMESTAMP_INITRD
] = "initrd",
4662 [MANAGER_TIMESTAMP_USERSPACE
] = "userspace",
4663 [MANAGER_TIMESTAMP_FINISH
] = "finish",
4664 [MANAGER_TIMESTAMP_SECURITY_START
] = "security-start",
4665 [MANAGER_TIMESTAMP_SECURITY_FINISH
] = "security-finish",
4666 [MANAGER_TIMESTAMP_GENERATORS_START
] = "generators-start",
4667 [MANAGER_TIMESTAMP_GENERATORS_FINISH
] = "generators-finish",
4668 [MANAGER_TIMESTAMP_UNITS_LOAD_START
] = "units-load-start",
4669 [MANAGER_TIMESTAMP_UNITS_LOAD_FINISH
] = "units-load-finish",
4670 [MANAGER_TIMESTAMP_INITRD_SECURITY_START
] = "initrd-security-start",
4671 [MANAGER_TIMESTAMP_INITRD_SECURITY_FINISH
] = "initrd-security-finish",
4672 [MANAGER_TIMESTAMP_INITRD_GENERATORS_START
] = "initrd-generators-start",
4673 [MANAGER_TIMESTAMP_INITRD_GENERATORS_FINISH
] = "initrd-generators-finish",
4674 [MANAGER_TIMESTAMP_INITRD_UNITS_LOAD_START
] = "initrd-units-load-start",
4675 [MANAGER_TIMESTAMP_INITRD_UNITS_LOAD_FINISH
] = "initrd-units-load-finish",
4678 DEFINE_STRING_TABLE_LOOKUP(manager_timestamp
, ManagerTimestamp
);