1 /* SPDX-License-Identifier: LGPL-2.1+ */
10 #include <sys/inotify.h>
11 #include <sys/ioctl.h>
12 #include <sys/reboot.h>
13 #include <sys/timerfd.h>
21 #include "sd-daemon.h"
22 #include "sd-messages.h"
25 #include "alloc-util.h"
26 #include "all-units.h"
28 #include "boot-timestamps.h"
29 #include "bus-common-errors.h"
30 #include "bus-error.h"
31 #include "bus-kernel.h"
33 #include "clean-ipc.h"
34 #include "clock-util.h"
36 #include "dbus-manager.h"
37 #include "dbus-unit.h"
39 #include "dirent-util.h"
42 #include "exec-util.h"
44 #include "exit-status.h"
51 #include "locale-setup.h"
57 #include "parse-util.h"
58 #include "path-lookup.h"
59 #include "path-util.h"
60 #include "process-util.h"
61 #include "ratelimit.h"
62 #include "rlimit-util.h"
64 #include "signal-util.h"
65 #include "socket-util.h"
67 #include "stat-util.h"
68 #include "string-table.h"
69 #include "string-util.h"
72 #include "syslog-util.h"
73 #include "terminal-util.h"
74 #include "time-util.h"
75 #include "transaction.h"
76 #include "umask-util.h"
77 #include "unit-name.h"
78 #include "user-util.h"
83 #define NOTIFY_RCVBUF_SIZE (8*1024*1024)
84 #define CGROUPS_AGENT_RCVBUF_SIZE (8*1024*1024)
86 /* Initial delay and the interval for printing status messages about running jobs */
87 #define JOBS_IN_PROGRESS_WAIT_USEC (5*USEC_PER_SEC)
88 #define JOBS_IN_PROGRESS_PERIOD_USEC (USEC_PER_SEC / 3)
89 #define JOBS_IN_PROGRESS_PERIOD_DIVISOR 3
91 /* If there are more than 1K bus messages queue across our API and direct busses, then let's not add more on top until
92 * the queue gets more empty. */
93 #define MANAGER_BUS_BUSY_THRESHOLD 1024LU
95 /* How many units and jobs to process of the bus queue before returning to the event loop. */
96 #define MANAGER_BUS_MESSAGE_BUDGET 100U
98 static int manager_dispatch_notify_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
99 static int manager_dispatch_cgroups_agent_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
100 static int manager_dispatch_signal_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
101 static int manager_dispatch_time_change_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
102 static int manager_dispatch_idle_pipe_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
103 static int manager_dispatch_user_lookup_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
104 static int manager_dispatch_jobs_in_progress(sd_event_source
*source
, usec_t usec
, void *userdata
);
105 static int manager_dispatch_run_queue(sd_event_source
*source
, void *userdata
);
106 static int manager_dispatch_sigchld(sd_event_source
*source
, void *userdata
);
107 static int manager_dispatch_timezone_change(sd_event_source
*source
, const struct inotify_event
*event
, void *userdata
);
108 static int manager_run_environment_generators(Manager
*m
);
109 static int manager_run_generators(Manager
*m
);
111 static void manager_watch_jobs_in_progress(Manager
*m
) {
117 /* We do not want to show the cylon animation if the user
118 * needs to confirm service executions otherwise confirmation
119 * messages will be screwed by the cylon animation. */
120 if (!manager_is_confirm_spawn_disabled(m
))
123 if (m
->jobs_in_progress_event_source
)
126 next
= now(CLOCK_MONOTONIC
) + JOBS_IN_PROGRESS_WAIT_USEC
;
127 r
= sd_event_add_time(
129 &m
->jobs_in_progress_event_source
,
132 manager_dispatch_jobs_in_progress
, m
);
136 (void) sd_event_source_set_description(m
->jobs_in_progress_event_source
, "manager-jobs-in-progress");
139 #define CYLON_BUFFER_EXTRA (2*STRLEN(ANSI_RED) + STRLEN(ANSI_HIGHLIGHT_RED) + 2*STRLEN(ANSI_NORMAL))
141 static void draw_cylon(char buffer
[], size_t buflen
, unsigned width
, unsigned pos
) {
144 assert(buflen
>= CYLON_BUFFER_EXTRA
+ width
+ 1);
145 assert(pos
<= width
+1); /* 0 or width+1 mean that the center light is behind the corner */
149 p
= mempset(p
, ' ', pos
-2);
150 if (log_get_show_color())
151 p
= stpcpy(p
, ANSI_RED
);
155 if (pos
> 0 && pos
<= width
) {
156 if (log_get_show_color())
157 p
= stpcpy(p
, ANSI_HIGHLIGHT_RED
);
161 if (log_get_show_color())
162 p
= stpcpy(p
, ANSI_NORMAL
);
165 if (log_get_show_color())
166 p
= stpcpy(p
, ANSI_RED
);
169 p
= mempset(p
, ' ', width
-1-pos
);
170 if (log_get_show_color())
171 strcpy(p
, ANSI_NORMAL
);
175 void manager_flip_auto_status(Manager
*m
, bool enable
) {
179 if (m
->show_status
== SHOW_STATUS_AUTO
)
180 manager_set_show_status(m
, SHOW_STATUS_TEMPORARY
);
182 if (m
->show_status
== SHOW_STATUS_TEMPORARY
)
183 manager_set_show_status(m
, SHOW_STATUS_AUTO
);
187 static void manager_print_jobs_in_progress(Manager
*m
) {
188 _cleanup_free_
char *job_of_n
= NULL
;
191 unsigned counter
= 0, print_nr
;
192 char cylon
[6 + CYLON_BUFFER_EXTRA
+ 1];
194 char time
[FORMAT_TIMESPAN_MAX
], limit
[FORMAT_TIMESPAN_MAX
] = "no limit";
198 assert(m
->n_running_jobs
> 0);
200 manager_flip_auto_status(m
, true);
202 print_nr
= (m
->jobs_in_progress_iteration
/ JOBS_IN_PROGRESS_PERIOD_DIVISOR
) % m
->n_running_jobs
;
204 HASHMAP_FOREACH(j
, m
->jobs
, i
)
205 if (j
->state
== JOB_RUNNING
&& counter
++ == print_nr
)
208 /* m->n_running_jobs must be consistent with the contents of m->jobs,
209 * so the above loop must have succeeded in finding j. */
210 assert(counter
== print_nr
+ 1);
213 cylon_pos
= m
->jobs_in_progress_iteration
% 14;
215 cylon_pos
= 14 - cylon_pos
;
216 draw_cylon(cylon
, sizeof(cylon
), 6, cylon_pos
);
218 m
->jobs_in_progress_iteration
++;
220 if (m
->n_running_jobs
> 1) {
221 if (asprintf(&job_of_n
, "(%u of %u) ", counter
, m
->n_running_jobs
) < 0)
225 format_timespan(time
, sizeof(time
), now(CLOCK_MONOTONIC
) - j
->begin_usec
, 1*USEC_PER_SEC
);
226 if (job_get_timeout(j
, &x
) > 0)
227 format_timespan(limit
, sizeof(limit
), x
- j
->begin_usec
, 1*USEC_PER_SEC
);
229 manager_status_printf(m
, STATUS_TYPE_EPHEMERAL
, cylon
,
230 "%sA %s job is running for %s (%s / %s)",
232 job_type_to_string(j
->type
),
233 unit_description(j
->unit
),
237 static int have_ask_password(void) {
238 _cleanup_closedir_
DIR *dir
;
241 dir
= opendir("/run/systemd/ask-password");
249 FOREACH_DIRENT_ALL(de
, dir
, return -errno
) {
250 if (startswith(de
->d_name
, "ask."))
256 static int manager_dispatch_ask_password_fd(sd_event_source
*source
,
257 int fd
, uint32_t revents
, void *userdata
) {
258 Manager
*m
= userdata
;
264 m
->have_ask_password
= have_ask_password();
265 if (m
->have_ask_password
< 0)
266 /* Log error but continue. Negative have_ask_password
267 * is treated as unknown status. */
268 log_error_errno(m
->have_ask_password
, "Failed to list /run/systemd/ask-password: %m");
273 static void manager_close_ask_password(Manager
*m
) {
276 m
->ask_password_event_source
= sd_event_source_unref(m
->ask_password_event_source
);
277 m
->ask_password_inotify_fd
= safe_close(m
->ask_password_inotify_fd
);
278 m
->have_ask_password
= -EINVAL
;
281 static int manager_check_ask_password(Manager
*m
) {
286 if (!m
->ask_password_event_source
) {
287 assert(m
->ask_password_inotify_fd
< 0);
289 mkdir_p_label("/run/systemd/ask-password", 0755);
291 m
->ask_password_inotify_fd
= inotify_init1(IN_NONBLOCK
|IN_CLOEXEC
);
292 if (m
->ask_password_inotify_fd
< 0)
293 return log_error_errno(errno
, "inotify_init1() failed: %m");
295 if (inotify_add_watch(m
->ask_password_inotify_fd
, "/run/systemd/ask-password", IN_CREATE
|IN_DELETE
|IN_MOVE
) < 0) {
296 log_error_errno(errno
, "Failed to add watch on /run/systemd/ask-password: %m");
297 manager_close_ask_password(m
);
301 r
= sd_event_add_io(m
->event
, &m
->ask_password_event_source
,
302 m
->ask_password_inotify_fd
, EPOLLIN
,
303 manager_dispatch_ask_password_fd
, m
);
305 log_error_errno(errno
, "Failed to add event source for /run/systemd/ask-password: %m");
306 manager_close_ask_password(m
);
310 (void) sd_event_source_set_description(m
->ask_password_event_source
, "manager-ask-password");
312 /* Queries might have been added meanwhile... */
313 manager_dispatch_ask_password_fd(m
->ask_password_event_source
,
314 m
->ask_password_inotify_fd
, EPOLLIN
, m
);
317 return m
->have_ask_password
;
320 static int manager_watch_idle_pipe(Manager
*m
) {
325 if (m
->idle_pipe_event_source
)
328 if (m
->idle_pipe
[2] < 0)
331 r
= sd_event_add_io(m
->event
, &m
->idle_pipe_event_source
, m
->idle_pipe
[2], EPOLLIN
, manager_dispatch_idle_pipe_fd
, m
);
333 return log_error_errno(r
, "Failed to watch idle pipe: %m");
335 (void) sd_event_source_set_description(m
->idle_pipe_event_source
, "manager-idle-pipe");
340 static void manager_close_idle_pipe(Manager
*m
) {
343 m
->idle_pipe_event_source
= sd_event_source_unref(m
->idle_pipe_event_source
);
345 safe_close_pair(m
->idle_pipe
);
346 safe_close_pair(m
->idle_pipe
+ 2);
349 static int manager_setup_time_change(Manager
*m
) {
354 if (MANAGER_IS_TEST_RUN(m
))
357 m
->time_change_event_source
= sd_event_source_unref(m
->time_change_event_source
);
358 m
->time_change_fd
= safe_close(m
->time_change_fd
);
360 m
->time_change_fd
= time_change_fd();
361 if (m
->time_change_fd
< 0)
362 return log_error_errno(m
->time_change_fd
, "Failed to create timer change timer fd: %m");
364 r
= sd_event_add_io(m
->event
, &m
->time_change_event_source
, m
->time_change_fd
, EPOLLIN
, manager_dispatch_time_change_fd
, m
);
366 return log_error_errno(r
, "Failed to create time change event source: %m");
368 /* Schedule this slightly earlier than the .timer event sources */
369 r
= sd_event_source_set_priority(m
->time_change_event_source
, SD_EVENT_PRIORITY_NORMAL
-1);
371 return log_error_errno(r
, "Failed to set priority of time change event sources: %m");
373 (void) sd_event_source_set_description(m
->time_change_event_source
, "manager-time-change");
375 log_debug("Set up TFD_TIMER_CANCEL_ON_SET timerfd.");
380 static int manager_read_timezone_stat(Manager
*m
) {
386 /* Read the current stat() data of /etc/localtime so that we detect changes */
387 if (lstat("/etc/localtime", &st
) < 0) {
388 log_debug_errno(errno
, "Failed to stat /etc/localtime, ignoring: %m");
389 changed
= m
->etc_localtime_accessible
;
390 m
->etc_localtime_accessible
= false;
394 k
= timespec_load(&st
.st_mtim
);
395 changed
= !m
->etc_localtime_accessible
|| k
!= m
->etc_localtime_mtime
;
397 m
->etc_localtime_mtime
= k
;
398 m
->etc_localtime_accessible
= true;
404 static int manager_setup_timezone_change(Manager
*m
) {
405 _cleanup_(sd_event_source_unrefp
) sd_event_source
*new_event
= NULL
;
410 if (MANAGER_IS_TEST_RUN(m
))
413 /* We watch /etc/localtime for three events: change of the link count (which might mean removal from /etc even
414 * though another link might be kept), renames, and file close operations after writing. Note we don't bother
415 * with IN_DELETE_SELF, as that would just report when the inode is removed entirely, i.e. after the link count
416 * went to zero and all fds to it are closed.
418 * Note that we never follow symlinks here. This is a simplification, but should cover almost all cases
421 * Note that we create the new event source first here, before releasing the old one. This should optimize
422 * behaviour as this way sd-event can reuse the old watch in case the inode didn't change. */
424 r
= sd_event_add_inotify(m
->event
, &new_event
, "/etc/localtime",
425 IN_ATTRIB
|IN_MOVE_SELF
|IN_CLOSE_WRITE
|IN_DONT_FOLLOW
, manager_dispatch_timezone_change
, m
);
426 if (r
== -ENOENT
) /* If the file doesn't exist yet, subscribe to /etc instead, and wait until it is created
427 * either by O_CREATE or by rename() */
428 r
= sd_event_add_inotify(m
->event
, &new_event
, "/etc",
429 IN_CREATE
|IN_MOVED_TO
|IN_ONLYDIR
, manager_dispatch_timezone_change
, m
);
431 return log_error_errno(r
, "Failed to create timezone change event source: %m");
433 /* Schedule this slightly earlier than the .timer event sources */
434 r
= sd_event_source_set_priority(new_event
, SD_EVENT_PRIORITY_NORMAL
-1);
436 return log_error_errno(r
, "Failed to set priority of timezone change event sources: %m");
438 sd_event_source_unref(m
->timezone_change_event_source
);
439 m
->timezone_change_event_source
= TAKE_PTR(new_event
);
444 static int enable_special_signals(Manager
*m
) {
445 _cleanup_close_
int fd
= -1;
449 if (MANAGER_IS_TEST_RUN(m
))
452 /* Enable that we get SIGINT on control-alt-del. In containers
453 * this will fail with EPERM (older) or EINVAL (newer), so
455 if (reboot(RB_DISABLE_CAD
) < 0 && !IN_SET(errno
, EPERM
, EINVAL
))
456 log_warning_errno(errno
, "Failed to enable ctrl-alt-del handling: %m");
458 fd
= open_terminal("/dev/tty0", O_RDWR
|O_NOCTTY
|O_CLOEXEC
);
460 /* Support systems without virtual console */
462 log_warning_errno(errno
, "Failed to open /dev/tty0: %m");
464 /* Enable that we get SIGWINCH on kbrequest */
465 if (ioctl(fd
, KDSIGACCEPT
, SIGWINCH
) < 0)
466 log_warning_errno(errno
, "Failed to enable kbrequest handling: %m");
472 #define RTSIG_IF_AVAILABLE(signum) (signum <= SIGRTMAX ? signum : -1)
474 static int manager_setup_signals(Manager
*m
) {
475 struct sigaction sa
= {
476 .sa_handler
= SIG_DFL
,
477 .sa_flags
= SA_NOCLDSTOP
|SA_RESTART
,
484 assert_se(sigaction(SIGCHLD
, &sa
, NULL
) == 0);
486 /* We make liberal use of realtime signals here. On
487 * Linux/glibc we have 30 of them (with the exception of Linux
488 * on hppa, see below), between SIGRTMIN+0 ... SIGRTMIN+30
491 assert_se(sigemptyset(&mask
) == 0);
492 sigset_add_many(&mask
,
493 SIGCHLD
, /* Child died */
494 SIGTERM
, /* Reexecute daemon */
495 SIGHUP
, /* Reload configuration */
496 SIGUSR1
, /* systemd/upstart: reconnect to D-Bus */
497 SIGUSR2
, /* systemd: dump status */
498 SIGINT
, /* Kernel sends us this on control-alt-del */
499 SIGWINCH
, /* Kernel sends us this on kbrequest (alt-arrowup) */
500 SIGPWR
, /* Some kernel drivers and upsd send us this on power failure */
502 SIGRTMIN
+0, /* systemd: start default.target */
503 SIGRTMIN
+1, /* systemd: isolate rescue.target */
504 SIGRTMIN
+2, /* systemd: isolate emergency.target */
505 SIGRTMIN
+3, /* systemd: start halt.target */
506 SIGRTMIN
+4, /* systemd: start poweroff.target */
507 SIGRTMIN
+5, /* systemd: start reboot.target */
508 SIGRTMIN
+6, /* systemd: start kexec.target */
510 /* ... space for more special targets ... */
512 SIGRTMIN
+13, /* systemd: Immediate halt */
513 SIGRTMIN
+14, /* systemd: Immediate poweroff */
514 SIGRTMIN
+15, /* systemd: Immediate reboot */
515 SIGRTMIN
+16, /* systemd: Immediate kexec */
517 /* ... space for more immediate system state changes ... */
519 SIGRTMIN
+20, /* systemd: enable status messages */
520 SIGRTMIN
+21, /* systemd: disable status messages */
521 SIGRTMIN
+22, /* systemd: set log level to LOG_DEBUG */
522 SIGRTMIN
+23, /* systemd: set log level to LOG_INFO */
523 SIGRTMIN
+24, /* systemd: Immediate exit (--user only) */
525 /* .. one free signal here ... */
527 /* Apparently Linux on hppa had fewer RT signals until v3.18,
528 * SIGRTMAX was SIGRTMIN+25, and then SIGRTMIN was lowered,
529 * see commit v3.17-7614-g1f25df2eff.
531 * We cannot unconditionally make use of those signals here,
532 * so let's use a runtime check. Since these commands are
533 * accessible by different means and only really a safety
534 * net, the missing functionality on hppa shouldn't matter.
537 RTSIG_IF_AVAILABLE(SIGRTMIN
+26), /* systemd: set log target to journal-or-kmsg */
538 RTSIG_IF_AVAILABLE(SIGRTMIN
+27), /* systemd: set log target to console */
539 RTSIG_IF_AVAILABLE(SIGRTMIN
+28), /* systemd: set log target to kmsg */
540 RTSIG_IF_AVAILABLE(SIGRTMIN
+29), /* systemd: set log target to syslog-or-kmsg (obsolete) */
542 /* ... one free signal here SIGRTMIN+30 ... */
544 assert_se(sigprocmask(SIG_SETMASK
, &mask
, NULL
) == 0);
546 m
->signal_fd
= signalfd(-1, &mask
, SFD_NONBLOCK
|SFD_CLOEXEC
);
547 if (m
->signal_fd
< 0)
550 r
= sd_event_add_io(m
->event
, &m
->signal_event_source
, m
->signal_fd
, EPOLLIN
, manager_dispatch_signal_fd
, m
);
554 (void) sd_event_source_set_description(m
->signal_event_source
, "manager-signal");
556 /* Process signals a bit earlier than the rest of things, but later than notify_fd processing, so that the
557 * notify processing can still figure out to which process/service a message belongs, before we reap the
558 * process. Also, process this before handling cgroup notifications, so that we always collect child exit
559 * status information before detecting that there's no process in a cgroup. */
560 r
= sd_event_source_set_priority(m
->signal_event_source
, SD_EVENT_PRIORITY_NORMAL
-6);
564 if (MANAGER_IS_SYSTEM(m
))
565 return enable_special_signals(m
);
570 static void manager_sanitize_environment(Manager
*m
) {
573 /* Let's remove some environment variables that we need ourselves to communicate with our clients */
593 /* Let's order the environment alphabetically, just to make it pretty */
594 strv_sort(m
->environment
);
597 static int manager_default_environment(Manager
*m
) {
600 if (MANAGER_IS_SYSTEM(m
)) {
601 /* The system manager always starts with a clean
602 * environment for its children. It does not import
603 * the kernel's or the parents' exported variables.
605 * The initial passed environment is untouched to keep
606 * /proc/self/environ valid; it is used for tagging
607 * the init process inside containers. */
608 m
->environment
= strv_new("PATH=" DEFAULT_PATH
,
611 /* Import locale variables LC_*= from configuration */
612 locale_setup(&m
->environment
);
614 /* The user manager passes its own environment
615 * along to its children. */
616 m
->environment
= strv_copy(environ
);
621 manager_sanitize_environment(m
);
626 static int manager_setup_prefix(Manager
*m
) {
632 static const struct table_entry paths_system
[_EXEC_DIRECTORY_TYPE_MAX
] = {
633 [EXEC_DIRECTORY_RUNTIME
] = { SD_PATH_SYSTEM_RUNTIME
, NULL
},
634 [EXEC_DIRECTORY_STATE
] = { SD_PATH_SYSTEM_STATE_PRIVATE
, NULL
},
635 [EXEC_DIRECTORY_CACHE
] = { SD_PATH_SYSTEM_STATE_CACHE
, NULL
},
636 [EXEC_DIRECTORY_LOGS
] = { SD_PATH_SYSTEM_STATE_LOGS
, NULL
},
637 [EXEC_DIRECTORY_CONFIGURATION
] = { SD_PATH_SYSTEM_CONFIGURATION
, NULL
},
640 static const struct table_entry paths_user
[_EXEC_DIRECTORY_TYPE_MAX
] = {
641 [EXEC_DIRECTORY_RUNTIME
] = { SD_PATH_USER_RUNTIME
, NULL
},
642 [EXEC_DIRECTORY_STATE
] = { SD_PATH_USER_CONFIGURATION
, NULL
},
643 [EXEC_DIRECTORY_CACHE
] = { SD_PATH_USER_STATE_CACHE
, NULL
},
644 [EXEC_DIRECTORY_LOGS
] = { SD_PATH_USER_CONFIGURATION
, "log" },
645 [EXEC_DIRECTORY_CONFIGURATION
] = { SD_PATH_USER_CONFIGURATION
, NULL
},
648 const struct table_entry
*p
;
654 if (MANAGER_IS_SYSTEM(m
))
659 for (i
= 0; i
< _EXEC_DIRECTORY_TYPE_MAX
; i
++) {
660 r
= sd_path_home(p
[i
].type
, p
[i
].suffix
, &m
->prefix
[i
]);
668 static int manager_setup_run_queue(Manager
*m
) {
672 assert(!m
->run_queue_event_source
);
674 r
= sd_event_add_defer(m
->event
, &m
->run_queue_event_source
, manager_dispatch_run_queue
, m
);
678 r
= sd_event_source_set_priority(m
->run_queue_event_source
, SD_EVENT_PRIORITY_IDLE
);
682 r
= sd_event_source_set_enabled(m
->run_queue_event_source
, SD_EVENT_OFF
);
686 (void) sd_event_source_set_description(m
->run_queue_event_source
, "manager-run-queue");
691 static int manager_setup_sigchld_event_source(Manager
*m
) {
695 assert(!m
->sigchld_event_source
);
697 r
= sd_event_add_defer(m
->event
, &m
->sigchld_event_source
, manager_dispatch_sigchld
, m
);
701 r
= sd_event_source_set_priority(m
->sigchld_event_source
, SD_EVENT_PRIORITY_NORMAL
-7);
705 r
= sd_event_source_set_enabled(m
->sigchld_event_source
, SD_EVENT_OFF
);
709 (void) sd_event_source_set_description(m
->sigchld_event_source
, "manager-sigchld");
714 int manager_new(UnitFileScope scope
, ManagerTestRunFlags test_run_flags
, Manager
**_m
) {
715 _cleanup_(manager_freep
) Manager
*m
= NULL
;
719 assert(IN_SET(scope
, UNIT_FILE_SYSTEM
, UNIT_FILE_USER
));
726 .unit_file_scope
= scope
,
727 .objective
= _MANAGER_OBJECTIVE_INVALID
,
729 .default_timer_accuracy_usec
= USEC_PER_MINUTE
,
730 .default_memory_accounting
= MEMORY_ACCOUNTING_DEFAULT
,
731 .default_tasks_accounting
= true,
732 .default_tasks_max
= UINT64_MAX
,
733 .default_timeout_start_usec
= DEFAULT_TIMEOUT_USEC
,
734 .default_timeout_stop_usec
= DEFAULT_TIMEOUT_USEC
,
735 .default_restart_usec
= DEFAULT_RESTART_USEC
,
737 .original_log_level
= -1,
738 .original_log_target
= _LOG_TARGET_INVALID
,
741 .cgroups_agent_fd
= -1,
743 .time_change_fd
= -1,
744 .user_lookup_fds
= { -1, -1 },
745 .private_listen_fd
= -1,
747 .cgroup_inotify_fd
= -1,
748 .pin_cgroupfs_fd
= -1,
749 .ask_password_inotify_fd
= -1,
750 .idle_pipe
= { -1, -1, -1, -1},
752 /* start as id #1, so that we can leave #0 around as "null-like" value */
755 .have_ask_password
= -EINVAL
, /* we don't know */
757 .test_run_flags
= test_run_flags
,
761 if (MANAGER_IS_SYSTEM(m
) && detect_container() <= 0)
762 boot_timestamps(m
->timestamps
+ MANAGER_TIMESTAMP_USERSPACE
,
763 m
->timestamps
+ MANAGER_TIMESTAMP_FIRMWARE
,
764 m
->timestamps
+ MANAGER_TIMESTAMP_LOADER
);
767 /* Prepare log fields we can use for structured logging */
768 if (MANAGER_IS_SYSTEM(m
)) {
769 m
->unit_log_field
= "UNIT=";
770 m
->unit_log_format_string
= "UNIT=%s";
772 m
->invocation_log_field
= "INVOCATION_ID=";
773 m
->invocation_log_format_string
= "INVOCATION_ID=%s";
775 m
->unit_log_field
= "USER_UNIT=";
776 m
->unit_log_format_string
= "USER_UNIT=%s";
778 m
->invocation_log_field
= "USER_INVOCATION_ID=";
779 m
->invocation_log_format_string
= "USER_INVOCATION_ID=%s";
782 /* Reboot immediately if the user hits C-A-D more often than 7x per 2s */
783 RATELIMIT_INIT(m
->ctrl_alt_del_ratelimit
, 2 * USEC_PER_SEC
, 7);
785 r
= manager_default_environment(m
);
789 r
= hashmap_ensure_allocated(&m
->units
, &string_hash_ops
);
793 r
= hashmap_ensure_allocated(&m
->jobs
, NULL
);
797 r
= hashmap_ensure_allocated(&m
->cgroup_unit
, &path_hash_ops
);
801 r
= hashmap_ensure_allocated(&m
->watch_bus
, &string_hash_ops
);
805 r
= manager_setup_prefix(m
);
809 r
= sd_event_default(&m
->event
);
813 r
= manager_setup_run_queue(m
);
817 if (test_run_flags
== MANAGER_TEST_RUN_MINIMAL
) {
818 m
->cgroup_root
= strdup("");
822 r
= manager_setup_signals(m
);
826 r
= manager_setup_cgroup(m
);
830 r
= manager_setup_time_change(m
);
834 r
= manager_read_timezone_stat(m
);
838 r
= manager_setup_timezone_change(m
);
842 r
= manager_setup_sigchld_event_source(m
);
847 if (MANAGER_IS_SYSTEM(m
) && test_run_flags
== 0) {
848 r
= mkdir_label("/run/systemd/units", 0755);
849 if (r
< 0 && r
!= -EEXIST
)
855 dir_is_empty("/usr") > 0;
857 /* Note that we do not set up the notify fd here. We do that after deserialization,
858 * since they might have gotten serialized across the reexec. */
865 static int manager_setup_notify(Manager
*m
) {
868 if (MANAGER_IS_TEST_RUN(m
))
871 if (m
->notify_fd
< 0) {
872 _cleanup_close_
int fd
= -1;
873 union sockaddr_union sa
= {
874 .sa
.sa_family
= AF_UNIX
,
876 static const int one
= 1;
878 /* First free all secondary fields */
879 m
->notify_socket
= mfree(m
->notify_socket
);
880 m
->notify_event_source
= sd_event_source_unref(m
->notify_event_source
);
882 fd
= socket(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
|SOCK_NONBLOCK
, 0);
884 return log_error_errno(errno
, "Failed to allocate notification socket: %m");
886 fd_inc_rcvbuf(fd
, NOTIFY_RCVBUF_SIZE
);
888 m
->notify_socket
= strappend(m
->prefix
[EXEC_DIRECTORY_RUNTIME
], "/systemd/notify");
889 if (!m
->notify_socket
)
892 (void) mkdir_parents_label(m
->notify_socket
, 0755);
893 (void) unlink(m
->notify_socket
);
895 strncpy(sa
.un
.sun_path
, m
->notify_socket
, sizeof(sa
.un
.sun_path
)-1);
896 r
= bind(fd
, &sa
.sa
, SOCKADDR_UN_LEN(sa
.un
));
898 return log_error_errno(errno
, "bind(%s) failed: %m", sa
.un
.sun_path
);
900 r
= setsockopt(fd
, SOL_SOCKET
, SO_PASSCRED
, &one
, sizeof(one
));
902 return log_error_errno(errno
, "SO_PASSCRED failed: %m");
904 m
->notify_fd
= TAKE_FD(fd
);
906 log_debug("Using notification socket %s", m
->notify_socket
);
909 if (!m
->notify_event_source
) {
910 r
= sd_event_add_io(m
->event
, &m
->notify_event_source
, m
->notify_fd
, EPOLLIN
, manager_dispatch_notify_fd
, m
);
912 return log_error_errno(r
, "Failed to allocate notify event source: %m");
914 /* Process notification messages a bit earlier than SIGCHLD, so that we can still identify to which
915 * service an exit message belongs. */
916 r
= sd_event_source_set_priority(m
->notify_event_source
, SD_EVENT_PRIORITY_NORMAL
-8);
918 return log_error_errno(r
, "Failed to set priority of notify event source: %m");
920 (void) sd_event_source_set_description(m
->notify_event_source
, "manager-notify");
926 static int manager_setup_cgroups_agent(Manager
*m
) {
928 static const union sockaddr_union sa
= {
929 .un
.sun_family
= AF_UNIX
,
930 .un
.sun_path
= "/run/systemd/cgroups-agent",
934 /* This creates a listening socket we receive cgroups agent messages on. We do not use D-Bus for delivering
935 * these messages from the cgroups agent binary to PID 1, as the cgroups agent binary is very short-living, and
936 * each instance of it needs a new D-Bus connection. Since D-Bus connections are SOCK_STREAM/AF_UNIX, on
937 * overloaded systems the backlog of the D-Bus socket becomes relevant, as not more than the configured number
938 * of D-Bus connections may be queued until the kernel will start dropping further incoming connections,
939 * possibly resulting in lost cgroups agent messages. To avoid this, we'll use a private SOCK_DGRAM/AF_UNIX
940 * socket, where no backlog is relevant as communication may take place without an actual connect() cycle, and
941 * we thus won't lose messages.
943 * Note that PID 1 will forward the agent message to system bus, so that the user systemd instance may listen
944 * to it. The system instance hence listens on this special socket, but the user instances listen on the system
945 * bus for these messages. */
947 if (MANAGER_IS_TEST_RUN(m
))
950 if (!MANAGER_IS_SYSTEM(m
))
953 r
= cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER
);
955 return log_error_errno(r
, "Failed to determine whether unified cgroups hierarchy is used: %m");
956 if (r
> 0) /* We don't need this anymore on the unified hierarchy */
959 if (m
->cgroups_agent_fd
< 0) {
960 _cleanup_close_
int fd
= -1;
962 /* First free all secondary fields */
963 m
->cgroups_agent_event_source
= sd_event_source_unref(m
->cgroups_agent_event_source
);
965 fd
= socket(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
|SOCK_NONBLOCK
, 0);
967 return log_error_errno(errno
, "Failed to allocate cgroups agent socket: %m");
969 fd_inc_rcvbuf(fd
, CGROUPS_AGENT_RCVBUF_SIZE
);
971 (void) unlink(sa
.un
.sun_path
);
973 /* Only allow root to connect to this socket */
975 r
= bind(fd
, &sa
.sa
, SOCKADDR_UN_LEN(sa
.un
));
977 return log_error_errno(errno
, "bind(%s) failed: %m", sa
.un
.sun_path
);
979 m
->cgroups_agent_fd
= fd
;
983 if (!m
->cgroups_agent_event_source
) {
984 r
= sd_event_add_io(m
->event
, &m
->cgroups_agent_event_source
, m
->cgroups_agent_fd
, EPOLLIN
, manager_dispatch_cgroups_agent_fd
, m
);
986 return log_error_errno(r
, "Failed to allocate cgroups agent event source: %m");
988 /* Process cgroups notifications early, but after having processed service notification messages or
989 * SIGCHLD signals, so that a cgroup running empty is always just the last safety net of notification,
990 * and we collected the metadata the notification and SIGCHLD stuff offers first. Also see handling of
991 * cgroup inotify for the unified cgroup stuff. */
992 r
= sd_event_source_set_priority(m
->cgroups_agent_event_source
, SD_EVENT_PRIORITY_NORMAL
-4);
994 return log_error_errno(r
, "Failed to set priority of cgroups agent event source: %m");
996 (void) sd_event_source_set_description(m
->cgroups_agent_event_source
, "manager-cgroups-agent");
1002 static int manager_setup_user_lookup_fd(Manager
*m
) {
1007 /* Set up the socket pair used for passing UID/GID resolution results from forked off processes to PID
1008 * 1. Background: we can't do name lookups (NSS) from PID 1, since it might involve IPC and thus activation,
1009 * and we might hence deadlock on ourselves. Hence we do all user/group lookups asynchronously from the forked
1010 * off processes right before executing the binaries to start. In order to be able to clean up any IPC objects
1011 * created by a unit (see RemoveIPC=) we need to know in PID 1 the used UID/GID of the executed processes,
1012 * hence we establish this communication channel so that forked off processes can pass their UID/GID
1013 * information back to PID 1. The forked off processes send their resolved UID/GID to PID 1 in a simple
1014 * datagram, along with their unit name, so that we can share one communication socket pair among all units for
1017 * You might wonder why we need a communication channel for this that is independent of the usual notification
1018 * socket scheme (i.e. $NOTIFY_SOCKET). The primary difference is about trust: data sent via the $NOTIFY_SOCKET
1019 * channel is only accepted if it originates from the right unit and if reception was enabled for it. The user
1020 * lookup socket OTOH is only accessible by PID 1 and its children until they exec(), and always available.
1022 * Note that this function is called under two circumstances: when we first initialize (in which case we
1023 * allocate both the socket pair and the event source to listen on it), and when we deserialize after a reload
1024 * (in which case the socket pair already exists but we still need to allocate the event source for it). */
1026 if (m
->user_lookup_fds
[0] < 0) {
1028 /* Free all secondary fields */
1029 safe_close_pair(m
->user_lookup_fds
);
1030 m
->user_lookup_event_source
= sd_event_source_unref(m
->user_lookup_event_source
);
1032 if (socketpair(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
, 0, m
->user_lookup_fds
) < 0)
1033 return log_error_errno(errno
, "Failed to allocate user lookup socket: %m");
1035 (void) fd_inc_rcvbuf(m
->user_lookup_fds
[0], NOTIFY_RCVBUF_SIZE
);
1038 if (!m
->user_lookup_event_source
) {
1039 r
= sd_event_add_io(m
->event
, &m
->user_lookup_event_source
, m
->user_lookup_fds
[0], EPOLLIN
, manager_dispatch_user_lookup_fd
, m
);
1041 return log_error_errno(errno
, "Failed to allocate user lookup event source: %m");
1043 /* Process even earlier than the notify event source, so that we always know first about valid UID/GID
1045 r
= sd_event_source_set_priority(m
->user_lookup_event_source
, SD_EVENT_PRIORITY_NORMAL
-11);
1047 return log_error_errno(errno
, "Failed to set priority ot user lookup event source: %m");
1049 (void) sd_event_source_set_description(m
->user_lookup_event_source
, "user-lookup");
1055 static unsigned manager_dispatch_cleanup_queue(Manager
*m
) {
1061 while ((u
= m
->cleanup_queue
)) {
1062 assert(u
->in_cleanup_queue
);
1072 GC_OFFSET_IN_PATH
, /* This one is on the path we were traveling */
1073 GC_OFFSET_UNSURE
, /* No clue */
1074 GC_OFFSET_GOOD
, /* We still need this unit */
1075 GC_OFFSET_BAD
, /* We don't need this unit anymore */
1079 static void unit_gc_mark_good(Unit
*u
, unsigned gc_marker
) {
1084 u
->gc_marker
= gc_marker
+ GC_OFFSET_GOOD
;
1086 /* Recursively mark referenced units as GOOD as well */
1087 HASHMAP_FOREACH_KEY(v
, other
, u
->dependencies
[UNIT_REFERENCES
], i
)
1088 if (other
->gc_marker
== gc_marker
+ GC_OFFSET_UNSURE
)
1089 unit_gc_mark_good(other
, gc_marker
);
1092 static void unit_gc_sweep(Unit
*u
, unsigned gc_marker
) {
1100 if (IN_SET(u
->gc_marker
- gc_marker
,
1101 GC_OFFSET_GOOD
, GC_OFFSET_BAD
, GC_OFFSET_UNSURE
, GC_OFFSET_IN_PATH
))
1104 if (u
->in_cleanup_queue
)
1107 if (!unit_may_gc(u
))
1110 u
->gc_marker
= gc_marker
+ GC_OFFSET_IN_PATH
;
1114 HASHMAP_FOREACH_KEY(v
, other
, u
->dependencies
[UNIT_REFERENCED_BY
], i
) {
1115 unit_gc_sweep(other
, gc_marker
);
1117 if (other
->gc_marker
== gc_marker
+ GC_OFFSET_GOOD
)
1120 if (other
->gc_marker
!= gc_marker
+ GC_OFFSET_BAD
)
1124 if (u
->refs_by_target
) {
1127 LIST_FOREACH(refs_by_target
, ref
, u
->refs_by_target
) {
1128 unit_gc_sweep(ref
->source
, gc_marker
);
1130 if (ref
->source
->gc_marker
== gc_marker
+ GC_OFFSET_GOOD
)
1133 if (ref
->source
->gc_marker
!= gc_marker
+ GC_OFFSET_BAD
)
1141 /* We were unable to find anything out about this entry, so
1142 * let's investigate it later */
1143 u
->gc_marker
= gc_marker
+ GC_OFFSET_UNSURE
;
1144 unit_add_to_gc_queue(u
);
1148 /* We definitely know that this one is not useful anymore, so
1149 * let's mark it for deletion */
1150 u
->gc_marker
= gc_marker
+ GC_OFFSET_BAD
;
1151 unit_add_to_cleanup_queue(u
);
1155 unit_gc_mark_good(u
, gc_marker
);
1158 static unsigned manager_dispatch_gc_unit_queue(Manager
*m
) {
1159 unsigned n
= 0, gc_marker
;
1164 /* log_debug("Running GC..."); */
1166 m
->gc_marker
+= _GC_OFFSET_MAX
;
1167 if (m
->gc_marker
+ _GC_OFFSET_MAX
<= _GC_OFFSET_MAX
)
1170 gc_marker
= m
->gc_marker
;
1172 while ((u
= m
->gc_unit_queue
)) {
1173 assert(u
->in_gc_queue
);
1175 unit_gc_sweep(u
, gc_marker
);
1177 LIST_REMOVE(gc_queue
, m
->gc_unit_queue
, u
);
1178 u
->in_gc_queue
= false;
1182 if (IN_SET(u
->gc_marker
- gc_marker
,
1183 GC_OFFSET_BAD
, GC_OFFSET_UNSURE
)) {
1185 log_unit_debug(u
, "Collecting.");
1186 u
->gc_marker
= gc_marker
+ GC_OFFSET_BAD
;
1187 unit_add_to_cleanup_queue(u
);
1194 static unsigned manager_dispatch_gc_job_queue(Manager
*m
) {
1200 while ((j
= m
->gc_job_queue
)) {
1201 assert(j
->in_gc_queue
);
1203 LIST_REMOVE(gc_queue
, m
->gc_job_queue
, j
);
1204 j
->in_gc_queue
= false;
1211 log_unit_debug(j
->unit
, "Collecting job.");
1212 (void) job_finish_and_invalidate(j
, JOB_COLLECTED
, false, false);
1218 static unsigned manager_dispatch_stop_when_unneeded_queue(Manager
*m
) {
1225 while ((u
= m
->stop_when_unneeded_queue
)) {
1226 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
1227 assert(m
->stop_when_unneeded_queue
);
1229 assert(u
->in_stop_when_unneeded_queue
);
1230 LIST_REMOVE(stop_when_unneeded_queue
, m
->stop_when_unneeded_queue
, u
);
1231 u
->in_stop_when_unneeded_queue
= false;
1235 if (!unit_is_unneeded(u
))
1238 log_unit_debug(u
, "Unit is not needed anymore.");
1240 /* If stopping a unit fails continuously we might enter a stop loop here, hence stop acting on the
1241 * service being unnecessary after a while. */
1243 if (!ratelimit_below(&u
->auto_stop_ratelimit
)) {
1244 log_unit_warning(u
, "Unit not needed anymore, but not stopping since we tried this too often recently.");
1248 /* Ok, nobody needs us anymore. Sniff. Then let's commit suicide */
1249 r
= manager_add_job(u
->manager
, JOB_STOP
, u
, JOB_FAIL
, &error
, NULL
);
1251 log_unit_warning_errno(u
, r
, "Failed to enqueue stop job, ignoring: %s", bus_error_message(&error
, r
));
1257 static void manager_clear_jobs_and_units(Manager
*m
) {
1262 while ((u
= hashmap_first(m
->units
)))
1265 manager_dispatch_cleanup_queue(m
);
1267 assert(!m
->load_queue
);
1268 assert(!m
->run_queue
);
1269 assert(!m
->dbus_unit_queue
);
1270 assert(!m
->dbus_job_queue
);
1271 assert(!m
->cleanup_queue
);
1272 assert(!m
->gc_unit_queue
);
1273 assert(!m
->gc_job_queue
);
1274 assert(!m
->stop_when_unneeded_queue
);
1276 assert(hashmap_isempty(m
->jobs
));
1277 assert(hashmap_isempty(m
->units
));
1279 m
->n_on_console
= 0;
1280 m
->n_running_jobs
= 0;
1281 m
->n_installed_jobs
= 0;
1282 m
->n_failed_jobs
= 0;
1285 Manager
* manager_free(Manager
*m
) {
1286 ExecDirectoryType dt
;
1292 manager_clear_jobs_and_units(m
);
1294 for (c
= 0; c
< _UNIT_TYPE_MAX
; c
++)
1295 if (unit_vtable
[c
]->shutdown
)
1296 unit_vtable
[c
]->shutdown(m
);
1298 /* Keep the cgroup hierarchy in place except when we know we are going down for good */
1299 manager_shutdown_cgroup(m
, IN_SET(m
->objective
, MANAGER_EXIT
, MANAGER_REBOOT
, MANAGER_POWEROFF
, MANAGER_HALT
, MANAGER_KEXEC
));
1301 lookup_paths_flush_generator(&m
->lookup_paths
);
1305 exec_runtime_vacuum(m
);
1306 hashmap_free(m
->exec_runtime_by_id
);
1308 dynamic_user_vacuum(m
, false);
1309 hashmap_free(m
->dynamic_users
);
1311 hashmap_free(m
->units
);
1312 hashmap_free(m
->units_by_invocation_id
);
1313 hashmap_free(m
->jobs
);
1314 hashmap_free(m
->watch_pids
);
1315 hashmap_free(m
->watch_bus
);
1317 set_free(m
->startup_units
);
1318 set_free(m
->failed_units
);
1320 sd_event_source_unref(m
->signal_event_source
);
1321 sd_event_source_unref(m
->sigchld_event_source
);
1322 sd_event_source_unref(m
->notify_event_source
);
1323 sd_event_source_unref(m
->cgroups_agent_event_source
);
1324 sd_event_source_unref(m
->time_change_event_source
);
1325 sd_event_source_unref(m
->timezone_change_event_source
);
1326 sd_event_source_unref(m
->jobs_in_progress_event_source
);
1327 sd_event_source_unref(m
->run_queue_event_source
);
1328 sd_event_source_unref(m
->user_lookup_event_source
);
1329 sd_event_source_unref(m
->sync_bus_names_event_source
);
1331 safe_close(m
->signal_fd
);
1332 safe_close(m
->notify_fd
);
1333 safe_close(m
->cgroups_agent_fd
);
1334 safe_close(m
->time_change_fd
);
1335 safe_close_pair(m
->user_lookup_fds
);
1337 manager_close_ask_password(m
);
1339 manager_close_idle_pipe(m
);
1341 sd_event_unref(m
->event
);
1343 free(m
->notify_socket
);
1345 lookup_paths_free(&m
->lookup_paths
);
1346 strv_free(m
->environment
);
1348 hashmap_free(m
->cgroup_unit
);
1349 set_free_free(m
->unit_path_cache
);
1351 free(m
->switch_root
);
1352 free(m
->switch_root_init
);
1354 rlimit_free_all(m
->rlimit
);
1356 assert(hashmap_isempty(m
->units_requiring_mounts_for
));
1357 hashmap_free(m
->units_requiring_mounts_for
);
1359 hashmap_free(m
->uid_refs
);
1360 hashmap_free(m
->gid_refs
);
1362 for (dt
= 0; dt
< _EXEC_DIRECTORY_TYPE_MAX
; dt
++)
1363 m
->prefix
[dt
] = mfree(m
->prefix
[dt
]);
1368 static void manager_enumerate_perpetual(Manager
*m
) {
1373 /* Let's ask every type to load all units from disk/kernel that it might know */
1374 for (c
= 0; c
< _UNIT_TYPE_MAX
; c
++) {
1375 if (!unit_type_supported(c
)) {
1376 log_debug("Unit type .%s is not supported on this system.", unit_type_to_string(c
));
1380 if (unit_vtable
[c
]->enumerate_perpetual
)
1381 unit_vtable
[c
]->enumerate_perpetual(m
);
1385 static void manager_enumerate(Manager
*m
) {
1390 /* Let's ask every type to load all units from disk/kernel that it might know */
1391 for (c
= 0; c
< _UNIT_TYPE_MAX
; c
++) {
1392 if (!unit_type_supported(c
)) {
1393 log_debug("Unit type .%s is not supported on this system.", unit_type_to_string(c
));
1397 if (unit_vtable
[c
]->enumerate
)
1398 unit_vtable
[c
]->enumerate(m
);
1401 manager_dispatch_load_queue(m
);
1404 static void manager_coldplug(Manager
*m
) {
1412 log_debug("Invoking unit coldplug() handlers…");
1414 /* Let's place the units back into their deserialized state */
1415 HASHMAP_FOREACH_KEY(u
, k
, m
->units
, i
) {
1417 /* ignore aliases */
1421 r
= unit_coldplug(u
);
1423 log_warning_errno(r
, "We couldn't coldplug %s, proceeding anyway: %m", u
->id
);
1427 static void manager_catchup(Manager
*m
) {
1434 log_debug("Invoking unit catchup() handlers…");
1436 /* Let's catch up on any state changes that happened while we were reloading/reexecing */
1437 HASHMAP_FOREACH_KEY(u
, k
, m
->units
, i
) {
1439 /* ignore aliases */
1447 static void manager_build_unit_path_cache(Manager
*m
) {
1453 set_free_free(m
->unit_path_cache
);
1455 m
->unit_path_cache
= set_new(&path_hash_ops
);
1456 if (!m
->unit_path_cache
) {
1461 /* This simply builds a list of files we know exist, so that
1462 * we don't always have to go to disk */
1464 STRV_FOREACH(i
, m
->lookup_paths
.search_path
) {
1465 _cleanup_closedir_
DIR *d
= NULL
;
1470 if (errno
!= ENOENT
)
1471 log_warning_errno(errno
, "Failed to open directory %s, ignoring: %m", *i
);
1475 FOREACH_DIRENT(de
, d
, r
= -errno
; goto fail
) {
1478 p
= strjoin(streq(*i
, "/") ? "" : *i
, "/", de
->d_name
);
1484 r
= set_consume(m
->unit_path_cache
, p
);
1493 log_warning_errno(r
, "Failed to build unit path cache, proceeding without: %m");
1494 m
->unit_path_cache
= set_free_free(m
->unit_path_cache
);
1497 static void manager_distribute_fds(Manager
*m
, FDSet
*fds
) {
1503 HASHMAP_FOREACH(u
, m
->units
, i
) {
1505 if (fdset_size(fds
) <= 0)
1508 if (!UNIT_VTABLE(u
)->distribute_fds
)
1511 UNIT_VTABLE(u
)->distribute_fds(u
, fds
);
1515 static bool manager_dbus_is_running(Manager
*m
, bool deserialized
) {
1520 /* This checks whether the dbus instance we are supposed to expose our APIs on is up. We check both the socket
1521 * and the service unit. If the 'deserialized' parameter is true we'll check the deserialized state of the unit
1522 * rather than the current one. */
1524 if (MANAGER_IS_TEST_RUN(m
))
1527 u
= manager_get_unit(m
, SPECIAL_DBUS_SOCKET
);
1530 if ((deserialized
? SOCKET(u
)->deserialized_state
: SOCKET(u
)->state
) != SOCKET_RUNNING
)
1533 u
= manager_get_unit(m
, SPECIAL_DBUS_SERVICE
);
1536 if (!IN_SET((deserialized
? SERVICE(u
)->deserialized_state
: SERVICE(u
)->state
), SERVICE_RUNNING
, SERVICE_RELOAD
))
1542 static void manager_setup_bus(Manager
*m
) {
1545 /* Let's set up our private bus connection now, unconditionally */
1546 (void) bus_init_private(m
);
1548 /* If we are in --user mode also connect to the system bus now */
1549 if (MANAGER_IS_USER(m
))
1550 (void) bus_init_system(m
);
1552 /* Let's connect to the bus now, but only if the unit is supposed to be up */
1553 if (manager_dbus_is_running(m
, MANAGER_IS_RELOADING(m
))) {
1554 (void) bus_init_api(m
);
1556 if (MANAGER_IS_SYSTEM(m
))
1557 (void) bus_init_system(m
);
1561 static void manager_preset_all(Manager
*m
) {
1566 if (m
->first_boot
<= 0)
1569 if (!MANAGER_IS_SYSTEM(m
))
1572 if (MANAGER_IS_TEST_RUN(m
))
1575 /* If this is the first boot, and we are in the host system, then preset everything */
1576 r
= unit_file_preset_all(UNIT_FILE_SYSTEM
, 0, NULL
, UNIT_FILE_PRESET_ENABLE_ONLY
, NULL
, 0);
1578 log_full_errno(r
== -EEXIST
? LOG_NOTICE
: LOG_WARNING
, r
,
1579 "Failed to populate /etc with preset unit settings, ignoring: %m");
1581 log_info("Populated /etc with preset unit settings.");
1584 int manager_startup(Manager
*m
, FILE *serialization
, FDSet
*fds
) {
1589 /* If we are running in test mode, we still want to run the generators,
1590 * but we should not touch the real generator directories. */
1591 r
= lookup_paths_init(&m
->lookup_paths
, m
->unit_file_scope
,
1592 MANAGER_IS_TEST_RUN(m
) ? LOOKUP_PATHS_TEMPORARY_GENERATED
: 0,
1595 return log_error_errno(r
, "Failed to initialize path lookup table: %m");
1597 r
= manager_run_environment_generators(m
);
1601 dual_timestamp_get(m
->timestamps
+ manager_timestamp_initrd_mangle(MANAGER_TIMESTAMP_GENERATORS_START
));
1602 r
= manager_run_generators(m
);
1603 dual_timestamp_get(m
->timestamps
+ manager_timestamp_initrd_mangle(MANAGER_TIMESTAMP_GENERATORS_FINISH
));
1607 manager_preset_all(m
);
1609 r
= lookup_paths_reduce(&m
->lookup_paths
);
1611 log_warning_errno(r
, "Failed ot reduce unit file paths, ignoring: %m");
1613 manager_build_unit_path_cache(m
);
1615 /* If we will deserialize make sure that during enumeration
1616 * this is already known, so we increase the counter here
1621 /* First, enumerate what we can from all config files */
1622 dual_timestamp_get(m
->timestamps
+ manager_timestamp_initrd_mangle(MANAGER_TIMESTAMP_UNITS_LOAD_START
));
1623 manager_enumerate_perpetual(m
);
1624 manager_enumerate(m
);
1625 dual_timestamp_get(m
->timestamps
+ manager_timestamp_initrd_mangle(MANAGER_TIMESTAMP_UNITS_LOAD_FINISH
));
1627 /* Second, deserialize if there is something to deserialize */
1628 if (serialization
) {
1629 r
= manager_deserialize(m
, serialization
, fds
);
1631 return log_error_errno(r
, "Deserialization failed: %m");
1634 /* Any fds left? Find some unit which wants them. This is
1635 * useful to allow container managers to pass some file
1636 * descriptors to us pre-initialized. This enables
1637 * socket-based activation of entire containers. */
1638 manager_distribute_fds(m
, fds
);
1640 /* We might have deserialized the notify fd, but if we didn't
1641 * then let's create the bus now */
1642 r
= manager_setup_notify(m
);
1644 /* No sense to continue without notifications, our children would fail anyway. */
1647 r
= manager_setup_cgroups_agent(m
);
1649 /* Likewise, no sense to continue without empty cgroup notifications. */
1652 r
= manager_setup_user_lookup_fd(m
);
1654 /* This shouldn't fail, except if things are really broken. */
1657 /* Connect to the bus if we are good for it */
1658 manager_setup_bus(m
);
1660 /* Now that we are connected to all possible busses, let's deserialize who is tracking us. */
1661 r
= bus_track_coldplug(m
, &m
->subscribed
, false, m
->deserialized_subscribed
);
1663 log_warning_errno(r
, "Failed to deserialized tracked clients, ignoring: %m");
1664 m
->deserialized_subscribed
= strv_free(m
->deserialized_subscribed
);
1666 /* Third, fire things up! */
1667 manager_coldplug(m
);
1669 /* Release any dynamic users no longer referenced */
1670 dynamic_user_vacuum(m
, true);
1672 exec_runtime_vacuum(m
);
1674 /* Release any references to UIDs/GIDs no longer referenced, and destroy any IPC owned by them */
1675 manager_vacuum_uid_refs(m
);
1676 manager_vacuum_gid_refs(m
);
1678 if (serialization
) {
1679 assert(m
->n_reloading
> 0);
1682 /* Let's wait for the UnitNew/JobNew messages being
1683 * sent, before we notify that the reload is
1685 m
->send_reloading_done
= true;
1688 /* Let's finally catch up with any changes that took place while we were reloading/reexecing */
1694 int manager_add_job(Manager
*m
, JobType type
, Unit
*unit
, JobMode mode
, sd_bus_error
*e
, Job
**_ret
) {
1699 assert(type
< _JOB_TYPE_MAX
);
1701 assert(mode
< _JOB_MODE_MAX
);
1703 if (mode
== JOB_ISOLATE
&& type
!= JOB_START
)
1704 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
, "Isolate is only valid for start.");
1706 if (mode
== JOB_ISOLATE
&& !unit
->allow_isolate
)
1707 return sd_bus_error_setf(e
, BUS_ERROR_NO_ISOLATION
, "Operation refused, unit may not be isolated.");
1709 log_unit_debug(unit
, "Trying to enqueue job %s/%s/%s", unit
->id
, job_type_to_string(type
), job_mode_to_string(mode
));
1711 type
= job_type_collapse(type
, unit
);
1713 tr
= transaction_new(mode
== JOB_REPLACE_IRREVERSIBLY
);
1717 r
= transaction_add_job_and_dependencies(tr
, type
, unit
, NULL
, true, false,
1718 IN_SET(mode
, JOB_IGNORE_DEPENDENCIES
, JOB_IGNORE_REQUIREMENTS
),
1719 mode
== JOB_IGNORE_DEPENDENCIES
, e
);
1723 if (mode
== JOB_ISOLATE
) {
1724 r
= transaction_add_isolate_jobs(tr
, m
);
1729 r
= transaction_activate(tr
, m
, mode
, e
);
1733 log_unit_debug(unit
,
1734 "Enqueued job %s/%s as %u", unit
->id
,
1735 job_type_to_string(type
), (unsigned) tr
->anchor_job
->id
);
1738 *_ret
= tr
->anchor_job
;
1740 transaction_free(tr
);
1744 transaction_abort(tr
);
1745 transaction_free(tr
);
1749 int manager_add_job_by_name(Manager
*m
, JobType type
, const char *name
, JobMode mode
, sd_bus_error
*e
, Job
**ret
) {
1750 Unit
*unit
= NULL
; /* just to appease gcc, initialization is not really necessary */
1754 assert(type
< _JOB_TYPE_MAX
);
1756 assert(mode
< _JOB_MODE_MAX
);
1758 r
= manager_load_unit(m
, name
, NULL
, NULL
, &unit
);
1763 return manager_add_job(m
, type
, unit
, mode
, e
, ret
);
1766 int manager_add_job_by_name_and_warn(Manager
*m
, JobType type
, const char *name
, JobMode mode
, Job
**ret
) {
1767 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
1771 assert(type
< _JOB_TYPE_MAX
);
1773 assert(mode
< _JOB_MODE_MAX
);
1775 r
= manager_add_job_by_name(m
, type
, name
, mode
, &error
, ret
);
1777 return log_warning_errno(r
, "Failed to enqueue %s job for %s: %s", job_mode_to_string(mode
), name
, bus_error_message(&error
, r
));
1782 int manager_propagate_reload(Manager
*m
, Unit
*unit
, JobMode mode
, sd_bus_error
*e
) {
1788 assert(mode
< _JOB_MODE_MAX
);
1789 assert(mode
!= JOB_ISOLATE
); /* Isolate is only valid for start */
1791 tr
= transaction_new(mode
== JOB_REPLACE_IRREVERSIBLY
);
1795 /* We need an anchor job */
1796 r
= transaction_add_job_and_dependencies(tr
, JOB_NOP
, unit
, NULL
, false, false, true, true, e
);
1800 /* Failure in adding individual dependencies is ignored, so this always succeeds. */
1801 transaction_add_propagate_reload_jobs(tr
, unit
, tr
->anchor_job
, mode
== JOB_IGNORE_DEPENDENCIES
, e
);
1803 r
= transaction_activate(tr
, m
, mode
, e
);
1807 transaction_free(tr
);
1811 transaction_abort(tr
);
1812 transaction_free(tr
);
1816 Job
*manager_get_job(Manager
*m
, uint32_t id
) {
1819 return hashmap_get(m
->jobs
, UINT32_TO_PTR(id
));
1822 Unit
*manager_get_unit(Manager
*m
, const char *name
) {
1826 return hashmap_get(m
->units
, name
);
1829 static int manager_dispatch_target_deps_queue(Manager
*m
) {
1834 static const UnitDependency deps
[] = {
1843 while ((u
= m
->target_deps_queue
)) {
1844 assert(u
->in_target_deps_queue
);
1846 LIST_REMOVE(target_deps_queue
, u
->manager
->target_deps_queue
, u
);
1847 u
->in_target_deps_queue
= false;
1849 for (k
= 0; k
< ELEMENTSOF(deps
); k
++) {
1854 HASHMAP_FOREACH_KEY(v
, target
, u
->dependencies
[deps
[k
]], i
) {
1855 r
= unit_add_default_target_dependency(u
, target
);
1865 unsigned manager_dispatch_load_queue(Manager
*m
) {
1871 /* Make sure we are not run recursively */
1872 if (m
->dispatching_load_queue
)
1875 m
->dispatching_load_queue
= true;
1877 /* Dispatches the load queue. Takes a unit from the queue and
1878 * tries to load its data until the queue is empty */
1880 while ((u
= m
->load_queue
)) {
1881 assert(u
->in_load_queue
);
1887 m
->dispatching_load_queue
= false;
1889 /* Dispatch the units waiting for their target dependencies to be added now, as all targets that we know about
1890 * should be loaded and have aliases resolved */
1891 (void) manager_dispatch_target_deps_queue(m
);
1896 int manager_load_unit_prepare(
1903 _cleanup_(unit_freep
) Unit
*cleanup_ret
= NULL
;
1909 assert(name
|| path
);
1912 /* This will prepare the unit for loading, but not actually
1913 * load anything from disk. */
1915 if (path
&& !is_path(path
))
1916 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
, "Path %s is not absolute.", path
);
1919 name
= basename(path
);
1921 t
= unit_name_to_type(name
);
1923 if (t
== _UNIT_TYPE_INVALID
|| !unit_name_is_valid(name
, UNIT_NAME_PLAIN
|UNIT_NAME_INSTANCE
)) {
1924 if (unit_name_is_valid(name
, UNIT_NAME_TEMPLATE
))
1925 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
, "Unit name %s is missing the instance name.", name
);
1927 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
, "Unit name %s is not valid.", name
);
1930 ret
= manager_get_unit(m
, name
);
1936 ret
= cleanup_ret
= unit_new(m
, unit_vtable
[t
]->object_size
);
1941 ret
->fragment_path
= strdup(path
);
1942 if (!ret
->fragment_path
)
1946 r
= unit_add_name(ret
, name
);
1950 unit_add_to_load_queue(ret
);
1951 unit_add_to_dbus_queue(ret
);
1952 unit_add_to_gc_queue(ret
);
1960 int manager_load_unit(
1972 /* This will load the service information files, but not actually
1973 * start any services or anything. */
1975 r
= manager_load_unit_prepare(m
, name
, path
, e
, _ret
);
1979 manager_dispatch_load_queue(m
);
1981 *_ret
= unit_follow_merge(*_ret
);
1985 int manager_load_startable_unit_or_warn(
1991 /* Load a unit, make sure it loaded fully and is not masked. */
1993 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
1997 r
= manager_load_unit(m
, name
, path
, &error
, &unit
);
1999 return log_error_errno(r
, "Failed to load %s %s: %s",
2000 name
? "unit" : "unit file", name
?: path
,
2001 bus_error_message(&error
, r
));
2003 r
= bus_unit_validate_load_state(unit
, &error
);
2005 return log_error_errno(r
, "%s", bus_error_message(&error
, r
));
2011 void manager_dump_jobs(Manager
*s
, FILE *f
, const char *prefix
) {
2018 HASHMAP_FOREACH(j
, s
->jobs
, i
)
2019 job_dump(j
, f
, prefix
);
2022 void manager_dump_units(Manager
*s
, FILE *f
, const char *prefix
) {
2030 HASHMAP_FOREACH_KEY(u
, t
, s
->units
, i
)
2032 unit_dump(u
, f
, prefix
);
2035 void manager_dump(Manager
*m
, FILE *f
, const char *prefix
) {
2041 for (q
= 0; q
< _MANAGER_TIMESTAMP_MAX
; q
++) {
2042 char buf
[FORMAT_TIMESTAMP_MAX
];
2044 if (dual_timestamp_is_set(m
->timestamps
+ q
))
2045 fprintf(f
, "%sTimestamp %s: %s\n",
2047 manager_timestamp_to_string(q
),
2048 format_timestamp(buf
, sizeof(buf
), m
->timestamps
[q
].realtime
));
2051 manager_dump_units(m
, f
, prefix
);
2052 manager_dump_jobs(m
, f
, prefix
);
2055 int manager_get_dump_string(Manager
*m
, char **ret
) {
2056 _cleanup_free_
char *dump
= NULL
;
2057 _cleanup_fclose_
FILE *f
= NULL
;
2064 f
= open_memstream(&dump
, &size
);
2068 (void) __fsetlocking(f
, FSETLOCKING_BYCALLER
);
2070 manager_dump(m
, f
, NULL
);
2072 r
= fflush_and_check(f
);
2078 *ret
= TAKE_PTR(dump
);
2083 void manager_clear_jobs(Manager
*m
) {
2088 while ((j
= hashmap_first(m
->jobs
)))
2089 /* No need to recurse. We're cancelling all jobs. */
2090 job_finish_and_invalidate(j
, JOB_CANCELED
, false, false);
2093 static int manager_dispatch_run_queue(sd_event_source
*source
, void *userdata
) {
2094 Manager
*m
= userdata
;
2100 while ((j
= m
->run_queue
)) {
2101 assert(j
->installed
);
2102 assert(j
->in_run_queue
);
2104 job_run_and_invalidate(j
);
2107 if (m
->n_running_jobs
> 0)
2108 manager_watch_jobs_in_progress(m
);
2110 if (m
->n_on_console
> 0)
2111 manager_watch_idle_pipe(m
);
2116 static unsigned manager_dispatch_dbus_queue(Manager
*m
) {
2117 unsigned n
= 0, budget
;
2123 if (m
->dispatching_dbus_queue
)
2126 /* Anything to do at all? */
2127 if (!m
->dbus_unit_queue
&& !m
->dbus_job_queue
&& !m
->send_reloading_done
&& !m
->queued_message
)
2130 /* Do we have overly many messages queued at the moment? If so, let's not enqueue more on top, let's sit this
2131 * cycle out, and process things in a later cycle when the queues got a bit emptier. */
2132 if (manager_bus_n_queued_write(m
) > MANAGER_BUS_BUSY_THRESHOLD
)
2135 /* Only process a certain number of units/jobs per event loop iteration. Even if the bus queue wasn't overly
2136 * full before this call we shouldn't increase it in size too wildly in one step, and we shouldn't monopolize
2137 * CPU time with generating these messages. Note the difference in counting of this "budget" and the
2138 * "threshold" above: the "budget" is decreased only once per generated message, regardless how many
2139 * busses/direct connections it is enqueued on, while the "threshold" is applied to each queued instance of bus
2140 * message, i.e. if the same message is enqueued to five busses/direct connections it will be counted five
2141 * times. This difference in counting ("references" vs. "instances") is primarily a result of the fact that
2142 * it's easier to implement it this way, however it also reflects the thinking that the "threshold" should put
2143 * a limit on used queue memory, i.e. space, while the "budget" should put a limit on time. Also note that
2144 * the "threshold" is currently chosen much higher than the "budget". */
2145 budget
= MANAGER_BUS_MESSAGE_BUDGET
;
2147 m
->dispatching_dbus_queue
= true;
2149 while (budget
> 0 && (u
= m
->dbus_unit_queue
)) {
2151 assert(u
->in_dbus_queue
);
2153 bus_unit_send_change_signal(u
);
2157 while (budget
> 0 && (j
= m
->dbus_job_queue
)) {
2158 assert(j
->in_dbus_queue
);
2160 bus_job_send_change_signal(j
);
2164 m
->dispatching_dbus_queue
= false;
2166 if (budget
> 0 && m
->send_reloading_done
) {
2167 m
->send_reloading_done
= false;
2168 bus_manager_send_reloading(m
, false);
2172 if (budget
> 0 && m
->queued_message
) {
2173 bus_send_queued_message(m
);
2180 static int manager_dispatch_cgroups_agent_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
2181 Manager
*m
= userdata
;
2182 char buf
[PATH_MAX
+1];
2185 n
= recv(fd
, buf
, sizeof(buf
), 0);
2187 return log_error_errno(errno
, "Failed to read cgroups agent message: %m");
2189 log_error("Got zero-length cgroups agent message, ignoring.");
2192 if ((size_t) n
>= sizeof(buf
)) {
2193 log_error("Got overly long cgroups agent message, ignoring.");
2197 if (memchr(buf
, 0, n
)) {
2198 log_error("Got cgroups agent message with embedded NUL byte, ignoring.");
2203 manager_notify_cgroup_empty(m
, buf
);
2204 (void) bus_forward_agent_released(m
, buf
);
2209 static void manager_invoke_notify_message(
2212 const struct ucred
*ucred
,
2221 if (u
->notifygen
== m
->notifygen
) /* Already invoked on this same unit in this same iteration? */
2223 u
->notifygen
= m
->notifygen
;
2225 if (UNIT_VTABLE(u
)->notify_message
) {
2226 _cleanup_strv_free_
char **tags
= NULL
;
2228 tags
= strv_split(buf
, NEWLINE
);
2234 UNIT_VTABLE(u
)->notify_message(u
, ucred
, tags
, fds
);
2236 } else if (DEBUG_LOGGING
) {
2237 _cleanup_free_
char *x
= NULL
, *y
= NULL
;
2239 x
= ellipsize(buf
, 20, 90);
2243 log_unit_debug(u
, "Got notification message \"%s\", ignoring.", strnull(y
));
2247 static int manager_dispatch_notify_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
2249 _cleanup_fdset_free_ FDSet
*fds
= NULL
;
2250 Manager
*m
= userdata
;
2251 char buf
[NOTIFY_BUFFER_MAX
+1];
2252 struct iovec iovec
= {
2254 .iov_len
= sizeof(buf
)-1,
2257 struct cmsghdr cmsghdr
;
2258 uint8_t buf
[CMSG_SPACE(sizeof(struct ucred
)) +
2259 CMSG_SPACE(sizeof(int) * NOTIFY_FD_MAX
)];
2261 struct msghdr msghdr
= {
2264 .msg_control
= &control
,
2265 .msg_controllen
= sizeof(control
),
2268 struct cmsghdr
*cmsg
;
2269 struct ucred
*ucred
= NULL
;
2270 _cleanup_free_ Unit
**array_copy
= NULL
;
2271 Unit
*u1
, *u2
, **array
;
2272 int r
, *fd_array
= NULL
;
2278 assert(m
->notify_fd
== fd
);
2280 if (revents
!= EPOLLIN
) {
2281 log_warning("Got unexpected poll event for notify fd.");
2285 n
= recvmsg(m
->notify_fd
, &msghdr
, MSG_DONTWAIT
|MSG_CMSG_CLOEXEC
|MSG_TRUNC
);
2287 if (IN_SET(errno
, EAGAIN
, EINTR
))
2288 return 0; /* Spurious wakeup, try again */
2290 /* If this is any other, real error, then let's stop processing this socket. This of course means we
2291 * won't take notification messages anymore, but that's still better than busy looping around this:
2292 * being woken up over and over again but being unable to actually read the message off the socket. */
2293 return log_error_errno(errno
, "Failed to receive notification message: %m");
2296 CMSG_FOREACH(cmsg
, &msghdr
) {
2297 if (cmsg
->cmsg_level
== SOL_SOCKET
&& cmsg
->cmsg_type
== SCM_RIGHTS
) {
2299 fd_array
= (int*) CMSG_DATA(cmsg
);
2300 n_fds
= (cmsg
->cmsg_len
- CMSG_LEN(0)) / sizeof(int);
2302 } else if (cmsg
->cmsg_level
== SOL_SOCKET
&&
2303 cmsg
->cmsg_type
== SCM_CREDENTIALS
&&
2304 cmsg
->cmsg_len
== CMSG_LEN(sizeof(struct ucred
))) {
2306 ucred
= (struct ucred
*) CMSG_DATA(cmsg
);
2313 r
= fdset_new_array(&fds
, fd_array
, n_fds
);
2315 close_many(fd_array
, n_fds
);
2321 if (!ucred
|| !pid_is_valid(ucred
->pid
)) {
2322 log_warning("Received notify message without valid credentials. Ignoring.");
2326 if ((size_t) n
>= sizeof(buf
) || (msghdr
.msg_flags
& MSG_TRUNC
)) {
2327 log_warning("Received notify message exceeded maximum size. Ignoring.");
2331 /* As extra safety check, let's make sure the string we get doesn't contain embedded NUL bytes. We permit one
2332 * trailing NUL byte in the message, but don't expect it. */
2333 if (n
> 1 && memchr(buf
, 0, n
-1)) {
2334 log_warning("Received notify message with embedded NUL bytes. Ignoring.");
2338 /* Make sure it's NUL-terminated. */
2341 /* Increase the generation counter used for filtering out duplicate unit invocations. */
2344 /* Notify every unit that might be interested, which might be multiple. */
2345 u1
= manager_get_unit_by_pid_cgroup(m
, ucred
->pid
);
2346 u2
= hashmap_get(m
->watch_pids
, PID_TO_PTR(ucred
->pid
));
2347 array
= hashmap_get(m
->watch_pids
, PID_TO_PTR(-ucred
->pid
));
2354 array_copy
= newdup(Unit
*, array
, k
+1);
2358 /* And now invoke the per-unit callbacks. Note that manager_invoke_notify_message() will handle duplicate units
2359 * make sure we only invoke each unit's handler once. */
2361 manager_invoke_notify_message(m
, u1
, ucred
, buf
, fds
);
2365 manager_invoke_notify_message(m
, u2
, ucred
, buf
, fds
);
2369 for (size_t i
= 0; array_copy
[i
]; i
++) {
2370 manager_invoke_notify_message(m
, array_copy
[i
], ucred
, buf
, fds
);
2375 log_warning("Cannot find unit for notify message of PID "PID_FMT
", ignoring.", ucred
->pid
);
2377 if (fdset_size(fds
) > 0)
2378 log_warning("Got extra auxiliary fds with notification message, closing them.");
2383 static void manager_invoke_sigchld_event(
2386 const siginfo_t
*si
) {
2392 /* Already invoked the handler of this unit in this iteration? Then don't process this again */
2393 if (u
->sigchldgen
== m
->sigchldgen
)
2395 u
->sigchldgen
= m
->sigchldgen
;
2397 log_unit_debug(u
, "Child "PID_FMT
" belongs to %s.", si
->si_pid
, u
->id
);
2398 unit_unwatch_pid(u
, si
->si_pid
);
2400 if (UNIT_VTABLE(u
)->sigchld_event
)
2401 UNIT_VTABLE(u
)->sigchld_event(u
, si
->si_pid
, si
->si_code
, si
->si_status
);
2404 static int manager_dispatch_sigchld(sd_event_source
*source
, void *userdata
) {
2405 Manager
*m
= userdata
;
2412 /* First we call waitd() for a PID and do not reap the zombie. That way we can still access /proc/$PID for it
2413 * while it is a zombie. */
2415 if (waitid(P_ALL
, 0, &si
, WEXITED
|WNOHANG
|WNOWAIT
) < 0) {
2417 if (errno
!= ECHILD
)
2418 log_error_errno(errno
, "Failed to peek for child with waitid(), ignoring: %m");
2426 if (IN_SET(si
.si_code
, CLD_EXITED
, CLD_KILLED
, CLD_DUMPED
)) {
2427 _cleanup_free_ Unit
**array_copy
= NULL
;
2428 _cleanup_free_
char *name
= NULL
;
2429 Unit
*u1
, *u2
, **array
;
2431 (void) get_process_comm(si
.si_pid
, &name
);
2433 log_debug("Child "PID_FMT
" (%s) died (code=%s, status=%i/%s)",
2434 si
.si_pid
, strna(name
),
2435 sigchld_code_to_string(si
.si_code
),
2437 strna(si
.si_code
== CLD_EXITED
2438 ? exit_status_to_string(si
.si_status
, EXIT_STATUS_FULL
)
2439 : signal_to_string(si
.si_status
)));
2441 /* Increase the generation counter used for filtering out duplicate unit invocations */
2444 /* And now figure out the unit this belongs to, it might be multiple... */
2445 u1
= manager_get_unit_by_pid_cgroup(m
, si
.si_pid
);
2446 u2
= hashmap_get(m
->watch_pids
, PID_TO_PTR(si
.si_pid
));
2447 array
= hashmap_get(m
->watch_pids
, PID_TO_PTR(-si
.si_pid
));
2451 /* Cound how many entries the array has */
2455 /* Make a copy of the array so that we don't trip up on the array changing beneath us */
2456 array_copy
= newdup(Unit
*, array
, n
+1);
2461 /* Finally, execute them all. Note that u1, u2 and the array might contain duplicates, but
2462 * that's fine, manager_invoke_sigchld_event() will ensure we only invoke the handlers once for
2463 * each iteration. */
2465 manager_invoke_sigchld_event(m
, u1
, &si
);
2467 manager_invoke_sigchld_event(m
, u2
, &si
);
2469 for (size_t i
= 0; array_copy
[i
]; i
++)
2470 manager_invoke_sigchld_event(m
, array_copy
[i
], &si
);
2473 /* And now, we actually reap the zombie. */
2474 if (waitid(P_PID
, si
.si_pid
, &si
, WEXITED
) < 0) {
2475 log_error_errno(errno
, "Failed to dequeue child, ignoring: %m");
2482 /* All children processed for now, turn off event source */
2484 r
= sd_event_source_set_enabled(m
->sigchld_event_source
, SD_EVENT_OFF
);
2486 return log_error_errno(r
, "Failed to disable SIGCHLD event source: %m");
2491 static void manager_start_target(Manager
*m
, const char *name
, JobMode mode
) {
2492 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
2495 log_debug("Activating special unit %s", name
);
2497 r
= manager_add_job_by_name(m
, JOB_START
, name
, mode
, &error
, NULL
);
2499 log_error("Failed to enqueue %s job: %s", name
, bus_error_message(&error
, r
));
2502 static void manager_handle_ctrl_alt_del(Manager
*m
) {
2503 /* If the user presses C-A-D more than
2504 * 7 times within 2s, we reboot/shutdown immediately,
2505 * unless it was disabled in system.conf */
2507 if (ratelimit_below(&m
->ctrl_alt_del_ratelimit
) || m
->cad_burst_action
== EMERGENCY_ACTION_NONE
)
2508 manager_start_target(m
, SPECIAL_CTRL_ALT_DEL_TARGET
, JOB_REPLACE_IRREVERSIBLY
);
2510 emergency_action(m
, m
->cad_burst_action
, NULL
,
2511 "Ctrl-Alt-Del was pressed more than 7 times within 2s");
2514 static int manager_dispatch_signal_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
2515 Manager
*m
= userdata
;
2517 struct signalfd_siginfo sfsi
;
2521 assert(m
->signal_fd
== fd
);
2523 if (revents
!= EPOLLIN
) {
2524 log_warning("Got unexpected events from signal file descriptor.");
2528 n
= read(m
->signal_fd
, &sfsi
, sizeof(sfsi
));
2529 if (n
!= sizeof(sfsi
)) {
2531 log_warning("Truncated read from signal fd (%zu bytes), ignoring!", n
);
2535 if (IN_SET(errno
, EINTR
, EAGAIN
))
2538 /* We return an error here, which will kill this handler,
2539 * to avoid a busy loop on read error. */
2540 return log_error_errno(errno
, "Reading from signal fd failed: %m");
2543 log_received_signal(sfsi
.ssi_signo
== SIGCHLD
||
2544 (sfsi
.ssi_signo
== SIGTERM
&& MANAGER_IS_USER(m
))
2545 ? LOG_DEBUG
: LOG_INFO
,
2548 switch (sfsi
.ssi_signo
) {
2551 r
= sd_event_source_set_enabled(m
->sigchld_event_source
, SD_EVENT_ON
);
2553 log_warning_errno(r
, "Failed to enable SIGCHLD event source, ignoring: %m");
2558 if (MANAGER_IS_SYSTEM(m
)) {
2559 /* This is for compatibility with the original sysvinit */
2560 if (verify_run_space_and_log("Refusing to reexecute") < 0)
2563 m
->objective
= MANAGER_REEXECUTE
;
2569 if (MANAGER_IS_SYSTEM(m
))
2570 manager_handle_ctrl_alt_del(m
);
2572 manager_start_target(m
, SPECIAL_EXIT_TARGET
,
2573 JOB_REPLACE_IRREVERSIBLY
);
2577 /* This is a nop on non-init */
2578 if (MANAGER_IS_SYSTEM(m
))
2579 manager_start_target(m
, SPECIAL_KBREQUEST_TARGET
, JOB_REPLACE
);
2584 /* This is a nop on non-init */
2585 if (MANAGER_IS_SYSTEM(m
))
2586 manager_start_target(m
, SPECIAL_SIGPWR_TARGET
, JOB_REPLACE
);
2591 if (manager_dbus_is_running(m
, false)) {
2592 log_info("Trying to reconnect to bus...");
2594 (void) bus_init_api(m
);
2596 if (MANAGER_IS_SYSTEM(m
))
2597 (void) bus_init_system(m
);
2599 log_info("Starting D-Bus service...");
2600 manager_start_target(m
, SPECIAL_DBUS_SERVICE
, JOB_REPLACE
);
2606 _cleanup_free_
char *dump
= NULL
;
2608 r
= manager_get_dump_string(m
, &dump
);
2610 log_warning_errno(errno
, "Failed to acquire manager dump: %m");
2614 log_dump(LOG_INFO
, dump
);
2619 if (verify_run_space_and_log("Refusing to reload") < 0)
2622 m
->objective
= MANAGER_RELOAD
;
2627 /* Starting SIGRTMIN+0 */
2628 static const struct {
2631 } target_table
[] = {
2632 [0] = { SPECIAL_DEFAULT_TARGET
, JOB_ISOLATE
},
2633 [1] = { SPECIAL_RESCUE_TARGET
, JOB_ISOLATE
},
2634 [2] = { SPECIAL_EMERGENCY_TARGET
, JOB_ISOLATE
},
2635 [3] = { SPECIAL_HALT_TARGET
, JOB_REPLACE_IRREVERSIBLY
},
2636 [4] = { SPECIAL_POWEROFF_TARGET
, JOB_REPLACE_IRREVERSIBLY
},
2637 [5] = { SPECIAL_REBOOT_TARGET
, JOB_REPLACE_IRREVERSIBLY
},
2638 [6] = { SPECIAL_KEXEC_TARGET
, JOB_REPLACE_IRREVERSIBLY
},
2641 /* Starting SIGRTMIN+13, so that target halt and system halt are 10 apart */
2642 static const ManagerObjective objective_table
[] = {
2644 [1] = MANAGER_POWEROFF
,
2645 [2] = MANAGER_REBOOT
,
2646 [3] = MANAGER_KEXEC
,
2649 if ((int) sfsi
.ssi_signo
>= SIGRTMIN
+0 &&
2650 (int) sfsi
.ssi_signo
< SIGRTMIN
+(int) ELEMENTSOF(target_table
)) {
2651 int idx
= (int) sfsi
.ssi_signo
- SIGRTMIN
;
2652 manager_start_target(m
, target_table
[idx
].target
,
2653 target_table
[idx
].mode
);
2657 if ((int) sfsi
.ssi_signo
>= SIGRTMIN
+13 &&
2658 (int) sfsi
.ssi_signo
< SIGRTMIN
+13+(int) ELEMENTSOF(objective_table
)) {
2659 m
->objective
= objective_table
[sfsi
.ssi_signo
- SIGRTMIN
- 13];
2663 switch (sfsi
.ssi_signo
- SIGRTMIN
) {
2666 manager_set_show_status(m
, SHOW_STATUS_YES
);
2670 manager_set_show_status(m
, SHOW_STATUS_NO
);
2674 manager_override_log_level(m
, LOG_DEBUG
);
2678 manager_restore_original_log_level(m
);
2682 if (MANAGER_IS_USER(m
)) {
2683 m
->objective
= MANAGER_EXIT
;
2687 /* This is a nop on init */
2691 case 29: /* compatibility: used to be mapped to LOG_TARGET_SYSLOG_OR_KMSG */
2692 manager_restore_original_log_target(m
);
2696 manager_override_log_target(m
, LOG_TARGET_CONSOLE
);
2700 manager_override_log_target(m
, LOG_TARGET_KMSG
);
2704 log_warning("Got unhandled signal <%s>.", signal_to_string(sfsi
.ssi_signo
));
2711 static int manager_dispatch_time_change_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
2712 Manager
*m
= userdata
;
2717 assert(m
->time_change_fd
== fd
);
2719 log_struct(LOG_DEBUG
,
2720 "MESSAGE_ID=" SD_MESSAGE_TIME_CHANGE_STR
,
2721 LOG_MESSAGE("Time has been changed"));
2723 /* Restart the watch */
2724 (void) manager_setup_time_change(m
);
2726 HASHMAP_FOREACH(u
, m
->units
, i
)
2727 if (UNIT_VTABLE(u
)->time_change
)
2728 UNIT_VTABLE(u
)->time_change(u
);
2733 static int manager_dispatch_timezone_change(
2734 sd_event_source
*source
,
2735 const struct inotify_event
*e
,
2738 Manager
*m
= userdata
;
2745 log_debug("inotify event for /etc/localtime");
2747 changed
= manager_read_timezone_stat(m
);
2753 /* Something changed, restart the watch, to ensure we watch the new /etc/localtime if it changed */
2754 (void) manager_setup_timezone_change(m
);
2756 /* Read the new timezone */
2759 log_debug("Timezone has been changed (now: %s).", tzname
[daylight
]);
2761 HASHMAP_FOREACH(u
, m
->units
, i
)
2762 if (UNIT_VTABLE(u
)->timezone_change
)
2763 UNIT_VTABLE(u
)->timezone_change(u
);
2768 static int manager_dispatch_idle_pipe_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
2769 Manager
*m
= userdata
;
2772 assert(m
->idle_pipe
[2] == fd
);
2774 /* There's at least one Type=idle child that just gave up on us waiting for the boot process to complete. Let's
2775 * now turn off any further console output if there's at least one service that needs console access, so that
2776 * from now on our own output should not spill into that service's output anymore. After all, we support
2777 * Type=idle only to beautify console output and it generally is set on services that want to own the console
2778 * exclusively without our interference. */
2779 m
->no_console_output
= m
->n_on_console
> 0;
2781 /* Acknowledge the child's request, and let all all other children know too that they shouldn't wait any longer
2782 * by closing the pipes towards them, which is what they are waiting for. */
2783 manager_close_idle_pipe(m
);
2788 static int manager_dispatch_jobs_in_progress(sd_event_source
*source
, usec_t usec
, void *userdata
) {
2789 Manager
*m
= userdata
;
2796 manager_print_jobs_in_progress(m
);
2798 next
= now(CLOCK_MONOTONIC
) + JOBS_IN_PROGRESS_PERIOD_USEC
;
2799 r
= sd_event_source_set_time(source
, next
);
2803 return sd_event_source_set_enabled(source
, SD_EVENT_ONESHOT
);
2806 int manager_loop(Manager
*m
) {
2809 RATELIMIT_DEFINE(rl
, 1*USEC_PER_SEC
, 50000);
2812 m
->objective
= MANAGER_OK
;
2814 /* Release the path cache */
2815 m
->unit_path_cache
= set_free_free(m
->unit_path_cache
);
2817 manager_check_finished(m
);
2819 /* There might still be some zombies hanging around from before we were exec()'ed. Let's reap them. */
2820 r
= sd_event_source_set_enabled(m
->sigchld_event_source
, SD_EVENT_ON
);
2822 return log_error_errno(r
, "Failed to enable SIGCHLD event source: %m");
2824 while (m
->objective
== MANAGER_OK
) {
2827 if (m
->runtime_watchdog
> 0 && m
->runtime_watchdog
!= USEC_INFINITY
&& MANAGER_IS_SYSTEM(m
))
2830 if (!ratelimit_below(&rl
)) {
2831 /* Yay, something is going seriously wrong, pause a little */
2832 log_warning("Looping too fast. Throttling execution a little.");
2836 if (manager_dispatch_load_queue(m
) > 0)
2839 if (manager_dispatch_gc_job_queue(m
) > 0)
2842 if (manager_dispatch_gc_unit_queue(m
) > 0)
2845 if (manager_dispatch_cleanup_queue(m
) > 0)
2848 if (manager_dispatch_cgroup_realize_queue(m
) > 0)
2851 if (manager_dispatch_stop_when_unneeded_queue(m
) > 0)
2854 if (manager_dispatch_dbus_queue(m
) > 0)
2857 /* Sleep for half the watchdog time */
2858 if (m
->runtime_watchdog
> 0 && m
->runtime_watchdog
!= USEC_INFINITY
&& MANAGER_IS_SYSTEM(m
)) {
2859 wait_usec
= m
->runtime_watchdog
/ 2;
2863 wait_usec
= USEC_INFINITY
;
2865 r
= sd_event_run(m
->event
, wait_usec
);
2867 return log_error_errno(r
, "Failed to run event loop: %m");
2870 return m
->objective
;
2873 int manager_load_unit_from_dbus_path(Manager
*m
, const char *s
, sd_bus_error
*e
, Unit
**_u
) {
2874 _cleanup_free_
char *n
= NULL
;
2875 sd_id128_t invocation_id
;
2883 r
= unit_name_from_dbus_path(s
, &n
);
2887 /* Permit addressing units by invocation ID: if the passed bus path is suffixed by a 128bit ID then we use it
2888 * as invocation ID. */
2889 r
= sd_id128_from_string(n
, &invocation_id
);
2891 u
= hashmap_get(m
->units_by_invocation_id
, &invocation_id
);
2897 return sd_bus_error_setf(e
, BUS_ERROR_NO_UNIT_FOR_INVOCATION_ID
,
2898 "No unit with the specified invocation ID " SD_ID128_FORMAT_STR
" known.",
2899 SD_ID128_FORMAT_VAL(invocation_id
));
2902 /* If this didn't work, we check if this is a unit name */
2903 if (!unit_name_is_valid(n
, UNIT_NAME_PLAIN
|UNIT_NAME_INSTANCE
)) {
2904 _cleanup_free_
char *nn
= NULL
;
2907 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
,
2908 "Unit name %s is neither a valid invocation ID nor unit name.", strnull(nn
));
2911 r
= manager_load_unit(m
, n
, NULL
, e
, &u
);
2919 int manager_get_job_from_dbus_path(Manager
*m
, const char *s
, Job
**_j
) {
2929 p
= startswith(s
, "/org/freedesktop/systemd1/job/");
2933 r
= safe_atou(p
, &id
);
2937 j
= manager_get_job(m
, id
);
2946 void manager_send_unit_audit(Manager
*m
, Unit
*u
, int type
, bool success
) {
2949 _cleanup_free_
char *p
= NULL
;
2953 if (!MANAGER_IS_SYSTEM(m
))
2956 audit_fd
= get_audit_fd();
2960 /* Don't generate audit events if the service was already
2961 * started and we're just deserializing */
2962 if (MANAGER_IS_RELOADING(m
))
2965 if (u
->type
!= UNIT_SERVICE
)
2968 r
= unit_name_to_prefix_and_instance(u
->id
, &p
);
2970 log_error_errno(r
, "Failed to extract prefix and instance of unit name: %m");
2974 msg
= strjoina("unit=", p
);
2975 if (audit_log_user_comm_message(audit_fd
, type
, msg
, "systemd", NULL
, NULL
, NULL
, success
) < 0) {
2977 /* We aren't allowed to send audit messages?
2978 * Then let's not retry again. */
2981 log_warning_errno(errno
, "Failed to send audit message: %m");
2987 void manager_send_unit_plymouth(Manager
*m
, Unit
*u
) {
2988 static const union sockaddr_union sa
= PLYMOUTH_SOCKET
;
2989 _cleanup_free_
char *message
= NULL
;
2990 _cleanup_close_
int fd
= -1;
2993 /* Don't generate plymouth events if the service was already
2994 * started and we're just deserializing */
2995 if (MANAGER_IS_RELOADING(m
))
2998 if (!MANAGER_IS_SYSTEM(m
))
3001 if (detect_container() > 0)
3004 if (!IN_SET(u
->type
, UNIT_SERVICE
, UNIT_MOUNT
, UNIT_SWAP
))
3007 /* We set SOCK_NONBLOCK here so that we rather drop the
3008 * message then wait for plymouth */
3009 fd
= socket(AF_UNIX
, SOCK_STREAM
|SOCK_CLOEXEC
|SOCK_NONBLOCK
, 0);
3011 log_error_errno(errno
, "socket() failed: %m");
3015 if (connect(fd
, &sa
.sa
, SOCKADDR_UN_LEN(sa
.un
)) < 0) {
3016 if (!IN_SET(errno
, EPIPE
, EAGAIN
, ENOENT
, ECONNREFUSED
, ECONNRESET
, ECONNABORTED
))
3017 log_error_errno(errno
, "connect() failed: %m");
3021 if (asprintf(&message
, "U\002%c%s%n", (int) (strlen(u
->id
) + 1), u
->id
, &n
) < 0) {
3027 if (write(fd
, message
, n
+ 1) != n
+ 1)
3028 if (!IN_SET(errno
, EPIPE
, EAGAIN
, ENOENT
, ECONNREFUSED
, ECONNRESET
, ECONNABORTED
))
3029 log_error_errno(errno
, "Failed to write Plymouth message: %m");
3032 int manager_open_serialization(Manager
*m
, FILE **_f
) {
3038 fd
= open_serialization_fd("systemd-state");
3042 f
= fdopen(fd
, "w+");
3052 int manager_serialize(
3056 bool switching_root
) {
3070 fprintf(f
, "current-job-id=%"PRIu32
"\n", m
->current_job_id
);
3071 fprintf(f
, "n-installed-jobs=%u\n", m
->n_installed_jobs
);
3072 fprintf(f
, "n-failed-jobs=%u\n", m
->n_failed_jobs
);
3073 fprintf(f
, "taint-usr=%s\n", yes_no(m
->taint_usr
));
3074 fprintf(f
, "ready-sent=%s\n", yes_no(m
->ready_sent
));
3075 fprintf(f
, "taint-logged=%s\n", yes_no(m
->taint_logged
));
3076 fprintf(f
, "service-watchdogs=%s\n", yes_no(m
->service_watchdogs
));
3078 t
= show_status_to_string(m
->show_status
);
3080 fprintf(f
, "show-status=%s\n", t
);
3082 if (m
->log_level_overridden
)
3083 fprintf(f
, "log-level-override=%i\n", log_get_max_level());
3084 if (m
->log_target_overridden
)
3085 fprintf(f
, "log-target-override=%s\n", log_target_to_string(log_get_target()));
3087 for (q
= 0; q
< _MANAGER_TIMESTAMP_MAX
; q
++) {
3088 /* The following timestamps only apply to the host system, hence only serialize them there */
3090 IN_SET(q
, MANAGER_TIMESTAMP_USERSPACE
, MANAGER_TIMESTAMP_FINISH
,
3091 MANAGER_TIMESTAMP_SECURITY_START
, MANAGER_TIMESTAMP_SECURITY_FINISH
,
3092 MANAGER_TIMESTAMP_GENERATORS_START
, MANAGER_TIMESTAMP_GENERATORS_FINISH
,
3093 MANAGER_TIMESTAMP_UNITS_LOAD_START
, MANAGER_TIMESTAMP_UNITS_LOAD_FINISH
))
3096 t
= manager_timestamp_to_string(q
);
3098 char field
[strlen(t
) + STRLEN("-timestamp") + 1];
3099 strcpy(stpcpy(field
, t
), "-timestamp");
3100 dual_timestamp_serialize(f
, field
, m
->timestamps
+ q
);
3104 if (!switching_root
)
3105 (void) serialize_environment(f
, m
->environment
);
3107 if (m
->notify_fd
>= 0) {
3110 copy
= fdset_put_dup(fds
, m
->notify_fd
);
3116 fprintf(f
, "notify-fd=%i\n", copy
);
3117 fprintf(f
, "notify-socket=%s\n", m
->notify_socket
);
3120 if (m
->cgroups_agent_fd
>= 0) {
3123 copy
= fdset_put_dup(fds
, m
->cgroups_agent_fd
);
3129 fprintf(f
, "cgroups-agent-fd=%i\n", copy
);
3132 if (m
->user_lookup_fds
[0] >= 0) {
3135 copy0
= fdset_put_dup(fds
, m
->user_lookup_fds
[0]);
3141 copy1
= fdset_put_dup(fds
, m
->user_lookup_fds
[1]);
3147 fprintf(f
, "user-lookup=%i %i\n", copy0
, copy1
);
3150 bus_track_serialize(m
->subscribed
, f
, "subscribed");
3152 r
= dynamic_user_serialize(m
, f
, fds
);
3156 manager_serialize_uid_refs(m
, f
);
3157 manager_serialize_gid_refs(m
, f
);
3159 r
= exec_runtime_serialize(m
, f
, fds
);
3163 (void) fputc('\n', f
);
3165 HASHMAP_FOREACH_KEY(u
, t
, m
->units
, i
) {
3173 r
= unit_serialize(u
, f
, fds
, !switching_root
);
3178 r
= fflush_and_check(f
);
3182 r
= bus_fdset_add_all(m
, fds
);
3189 assert(m
->n_reloading
> 0);
3195 int manager_deserialize(Manager
*m
, FILE *f
, FDSet
*fds
) {
3201 log_debug("Deserializing state...");
3203 /* If we are not in reload mode yet, enter it now. Not that this is recursive, a caller might already have
3204 * increased it to non-zero, which is why we just increase it by one here and down again at the end of this
3209 char line
[LINE_MAX
];
3210 const char *val
, *l
;
3212 if (!fgets(line
, sizeof(line
), f
)) {
3227 if ((val
= startswith(l
, "current-job-id="))) {
3230 if (safe_atou32(val
, &id
) < 0)
3231 log_notice("Failed to parse current job id value '%s', ignoring.", val
);
3233 m
->current_job_id
= MAX(m
->current_job_id
, id
);
3235 } else if ((val
= startswith(l
, "n-installed-jobs="))) {
3238 if (safe_atou32(val
, &n
) < 0)
3239 log_notice("Failed to parse installed jobs counter '%s', ignoring.", val
);
3241 m
->n_installed_jobs
+= n
;
3243 } else if ((val
= startswith(l
, "n-failed-jobs="))) {
3246 if (safe_atou32(val
, &n
) < 0)
3247 log_notice("Failed to parse failed jobs counter '%s', ignoring.", val
);
3249 m
->n_failed_jobs
+= n
;
3251 } else if ((val
= startswith(l
, "taint-usr="))) {
3254 b
= parse_boolean(val
);
3256 log_notice("Failed to parse taint /usr flag '%s', ignoring.", val
);
3258 m
->taint_usr
= m
->taint_usr
|| b
;
3260 } else if ((val
= startswith(l
, "ready-sent="))) {
3263 b
= parse_boolean(val
);
3265 log_notice("Failed to parse ready-sent flag '%s', ignoring.", val
);
3267 m
->ready_sent
= m
->ready_sent
|| b
;
3269 } else if ((val
= startswith(l
, "taint-logged="))) {
3272 b
= parse_boolean(val
);
3274 log_notice("Failed to parse taint-logged flag '%s', ignoring.", val
);
3276 m
->taint_logged
= m
->taint_logged
|| b
;
3278 } else if ((val
= startswith(l
, "service-watchdogs="))) {
3281 b
= parse_boolean(val
);
3283 log_notice("Failed to parse service-watchdogs flag '%s', ignoring.", val
);
3285 m
->service_watchdogs
= b
;
3287 } else if ((val
= startswith(l
, "show-status="))) {
3290 s
= show_status_from_string(val
);
3292 log_notice("Failed to parse show-status flag '%s', ignoring.", val
);
3294 manager_set_show_status(m
, s
);
3296 } else if ((val
= startswith(l
, "log-level-override="))) {
3299 level
= log_level_from_string(val
);
3301 log_notice("Failed to parse log-level-override value '%s', ignoring.", val
);
3303 manager_override_log_level(m
, level
);
3305 } else if ((val
= startswith(l
, "log-target-override="))) {
3308 target
= log_target_from_string(val
);
3310 log_notice("Failed to parse log-target-override value '%s', ignoring.", val
);
3312 manager_override_log_target(m
, target
);
3314 } else if (startswith(l
, "env=")) {
3315 r
= deserialize_environment(&m
->environment
, l
);
3319 log_notice_errno(r
, "Failed to parse environment entry: \"%s\", ignoring: %m", l
);
3321 } else if ((val
= startswith(l
, "notify-fd="))) {
3324 if (safe_atoi(val
, &fd
) < 0 || fd
< 0 || !fdset_contains(fds
, fd
))
3325 log_notice("Failed to parse notify fd, ignoring: \"%s\"", val
);
3327 m
->notify_event_source
= sd_event_source_unref(m
->notify_event_source
);
3328 safe_close(m
->notify_fd
);
3329 m
->notify_fd
= fdset_remove(fds
, fd
);
3332 } else if ((val
= startswith(l
, "notify-socket="))) {
3341 free(m
->notify_socket
);
3342 m
->notify_socket
= n
;
3344 } else if ((val
= startswith(l
, "cgroups-agent-fd="))) {
3347 if (safe_atoi(val
, &fd
) < 0 || fd
< 0 || !fdset_contains(fds
, fd
))
3348 log_notice("Failed to parse cgroups agent fd, ignoring.: %s", val
);
3350 m
->cgroups_agent_event_source
= sd_event_source_unref(m
->cgroups_agent_event_source
);
3351 safe_close(m
->cgroups_agent_fd
);
3352 m
->cgroups_agent_fd
= fdset_remove(fds
, fd
);
3355 } else if ((val
= startswith(l
, "user-lookup="))) {
3358 if (sscanf(val
, "%i %i", &fd0
, &fd1
) != 2 || fd0
< 0 || fd1
< 0 || fd0
== fd1
|| !fdset_contains(fds
, fd0
) || !fdset_contains(fds
, fd1
))
3359 log_notice("Failed to parse user lookup fd, ignoring: %s", val
);
3361 m
->user_lookup_event_source
= sd_event_source_unref(m
->user_lookup_event_source
);
3362 safe_close_pair(m
->user_lookup_fds
);
3363 m
->user_lookup_fds
[0] = fdset_remove(fds
, fd0
);
3364 m
->user_lookup_fds
[1] = fdset_remove(fds
, fd1
);
3367 } else if ((val
= startswith(l
, "dynamic-user=")))
3368 dynamic_user_deserialize_one(m
, val
, fds
);
3369 else if ((val
= startswith(l
, "destroy-ipc-uid=")))
3370 manager_deserialize_uid_refs_one(m
, val
);
3371 else if ((val
= startswith(l
, "destroy-ipc-gid=")))
3372 manager_deserialize_gid_refs_one(m
, val
);
3373 else if ((val
= startswith(l
, "exec-runtime=")))
3374 exec_runtime_deserialize_one(m
, val
, fds
);
3375 else if ((val
= startswith(l
, "subscribed="))) {
3377 if (strv_extend(&m
->deserialized_subscribed
, val
) < 0) {
3385 for (q
= 0; q
< _MANAGER_TIMESTAMP_MAX
; q
++) {
3386 val
= startswith(l
, manager_timestamp_to_string(q
));
3390 val
= startswith(val
, "-timestamp=");
3395 if (q
< _MANAGER_TIMESTAMP_MAX
) /* found it */
3396 dual_timestamp_deserialize(val
, m
->timestamps
+ q
);
3397 else if (!startswith(l
, "kdbus-fd=")) /* ignore kdbus */
3398 log_notice("Unknown serialization item '%s', ignoring.", l
);
3404 char name
[UNIT_NAME_MAX
+2];
3405 const char* unit_name
;
3408 if (!fgets(name
, sizeof(name
), f
)) {
3418 unit_name
= strstrip(name
);
3420 r
= manager_load_unit(m
, unit_name
, NULL
, NULL
, &u
);
3425 log_notice_errno(r
, "Failed to load unit \"%s\", skipping deserialization: %m", unit_name
);
3426 unit_deserialize_skip(f
);
3430 r
= unit_deserialize(u
, f
, fds
);
3435 log_notice_errno(r
, "Failed to deserialize unit \"%s\": %m", unit_name
);
3443 /* We are done with reloading, decrease counter again */
3444 assert(m
->n_reloading
> 0);
3450 static void manager_flush_finished_jobs(Manager
*m
) {
3453 while ((j
= set_steal_first(m
->pending_finished_jobs
))) {
3454 bus_job_send_removed_signal(j
);
3458 m
->pending_finished_jobs
= set_free(m
->pending_finished_jobs
);
3461 int manager_reload(Manager
*m
) {
3462 _cleanup_fdset_free_ FDSet
*fds
= NULL
;
3463 _cleanup_fclose_
FILE *f
= NULL
;
3468 r
= manager_open_serialization(m
, &f
);
3470 return log_error_errno(r
, "Failed to create serialization file: %m");
3476 /* We are officially in reload mode from here on */
3479 r
= manager_serialize(m
, f
, fds
, false);
3481 log_error_errno(r
, "Failed to serialize manager: %m");
3485 if (fseeko(f
, 0, SEEK_SET
) < 0) {
3486 r
= log_error_errno(errno
, "Failed to seek to beginning of serialization: %m");
3490 /* 💀 This is the point of no return, from here on there is no way back. 💀 */
3492 bus_manager_send_reloading(m
, true);
3494 /* Start by flushing out all jobs and units, all generated units, all runtime environments, all dynamic users
3495 * and everything else that is worth flushing out. We'll get it all back from the serialization — if we need
3498 manager_clear_jobs_and_units(m
);
3499 lookup_paths_flush_generator(&m
->lookup_paths
);
3500 lookup_paths_free(&m
->lookup_paths
);
3501 exec_runtime_vacuum(m
);
3502 dynamic_user_vacuum(m
, false);
3503 m
->uid_refs
= hashmap_free(m
->uid_refs
);
3504 m
->gid_refs
= hashmap_free(m
->gid_refs
);
3506 r
= lookup_paths_init(&m
->lookup_paths
, m
->unit_file_scope
, 0, NULL
);
3508 log_warning_errno(r
, "Failed to initialize path lookup table, ignoring: %m");
3510 (void) manager_run_environment_generators(m
);
3511 (void) manager_run_generators(m
);
3513 r
= lookup_paths_reduce(&m
->lookup_paths
);
3515 log_warning_errno(r
, "Failed ot reduce unit file paths, ignoring: %m");
3517 manager_build_unit_path_cache(m
);
3519 /* First, enumerate what we can from kernel and suchlike */
3520 manager_enumerate(m
);
3522 /* Second, deserialize our stored data */
3523 r
= manager_deserialize(m
, f
, fds
);
3525 log_warning_errno(r
, "Deserialization failed, proceeding anyway: %m");
3527 /* We don't need the serialization anymore */
3530 /* Re-register notify_fd as event source, and set up other sockets/communication channels we might need */
3531 (void) manager_setup_notify(m
);
3532 (void) manager_setup_cgroups_agent(m
);
3533 (void) manager_setup_user_lookup_fd(m
);
3535 /* Third, fire things up! */
3536 manager_coldplug(m
);
3538 /* Release any dynamic users no longer referenced */
3539 dynamic_user_vacuum(m
, true);
3541 /* Release any references to UIDs/GIDs no longer referenced, and destroy any IPC owned by them */
3542 manager_vacuum_uid_refs(m
);
3543 manager_vacuum_gid_refs(m
);
3545 exec_runtime_vacuum(m
);
3547 /* Consider the reload process complete now. */
3548 assert(m
->n_reloading
> 0);
3551 /* It might be safe to log to the journal now and connect to dbus */
3552 manager_recheck_journal(m
);
3553 manager_recheck_dbus(m
);
3555 /* Let's finally catch up with any changes that took place while we were reloading/reexecing */
3558 /* Sync current state of bus names with our set of listening units */
3559 (void) manager_enqueue_sync_bus_names(m
);
3561 if (!MANAGER_IS_RELOADING(m
))
3562 manager_flush_finished_jobs(m
);
3564 m
->send_reloading_done
= true;
3568 /* Fail the call. Note that we hit this only if we fail before the point of no return, i.e. when the error is
3569 * still something that can be handled. */
3570 assert(m
->n_reloading
> 0);
3576 void manager_reset_failed(Manager
*m
) {
3582 HASHMAP_FOREACH(u
, m
->units
, i
)
3583 unit_reset_failed(u
);
3586 bool manager_unit_inactive_or_pending(Manager
*m
, const char *name
) {
3592 /* Returns true if the unit is inactive or going down */
3593 u
= manager_get_unit(m
, name
);
3597 return unit_inactive_or_pending(u
);
3600 static void log_taint_string(Manager
*m
) {
3601 _cleanup_free_
char *taint
= NULL
;
3605 if (MANAGER_IS_USER(m
) || m
->taint_logged
)
3608 m
->taint_logged
= true; /* only check for taint once */
3610 taint
= manager_taint_string(m
);
3614 log_struct(LOG_NOTICE
,
3615 LOG_MESSAGE("System is tainted: %s", taint
),
3617 "MESSAGE_ID=" SD_MESSAGE_TAINTED_STR
);
3620 static void manager_notify_finished(Manager
*m
) {
3621 char userspace
[FORMAT_TIMESPAN_MAX
], initrd
[FORMAT_TIMESPAN_MAX
], kernel
[FORMAT_TIMESPAN_MAX
], sum
[FORMAT_TIMESPAN_MAX
];
3622 usec_t firmware_usec
, loader_usec
, kernel_usec
, initrd_usec
, userspace_usec
, total_usec
;
3624 if (MANAGER_IS_TEST_RUN(m
))
3627 if (MANAGER_IS_SYSTEM(m
) && detect_container() <= 0) {
3628 char ts
[FORMAT_TIMESPAN_MAX
];
3629 char buf
[FORMAT_TIMESPAN_MAX
+ STRLEN(" (firmware) + ") + FORMAT_TIMESPAN_MAX
+ STRLEN(" (loader) + ")]
3632 size_t size
= sizeof buf
;
3634 /* Note that MANAGER_TIMESTAMP_KERNEL's monotonic value is always at 0, and
3635 * MANAGER_TIMESTAMP_FIRMWARE's and MANAGER_TIMESTAMP_LOADER's monotonic value should be considered
3636 * negative values. */
3638 firmware_usec
= m
->timestamps
[MANAGER_TIMESTAMP_FIRMWARE
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_LOADER
].monotonic
;
3639 loader_usec
= m
->timestamps
[MANAGER_TIMESTAMP_LOADER
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_KERNEL
].monotonic
;
3640 userspace_usec
= m
->timestamps
[MANAGER_TIMESTAMP_FINISH
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_USERSPACE
].monotonic
;
3641 total_usec
= m
->timestamps
[MANAGER_TIMESTAMP_FIRMWARE
].monotonic
+ m
->timestamps
[MANAGER_TIMESTAMP_FINISH
].monotonic
;
3643 if (firmware_usec
> 0)
3644 size
= strpcpyf(&p
, size
, "%s (firmware) + ", format_timespan(ts
, sizeof(ts
), firmware_usec
, USEC_PER_MSEC
));
3645 if (loader_usec
> 0)
3646 size
= strpcpyf(&p
, size
, "%s (loader) + ", format_timespan(ts
, sizeof(ts
), loader_usec
, USEC_PER_MSEC
));
3648 if (dual_timestamp_is_set(&m
->timestamps
[MANAGER_TIMESTAMP_INITRD
])) {
3650 /* The initrd case on bare-metal*/
3651 kernel_usec
= m
->timestamps
[MANAGER_TIMESTAMP_INITRD
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_KERNEL
].monotonic
;
3652 initrd_usec
= m
->timestamps
[MANAGER_TIMESTAMP_USERSPACE
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_INITRD
].monotonic
;
3654 log_struct(LOG_INFO
,
3655 "MESSAGE_ID=" SD_MESSAGE_STARTUP_FINISHED_STR
,
3656 "KERNEL_USEC="USEC_FMT
, kernel_usec
,
3657 "INITRD_USEC="USEC_FMT
, initrd_usec
,
3658 "USERSPACE_USEC="USEC_FMT
, userspace_usec
,
3659 LOG_MESSAGE("Startup finished in %s%s (kernel) + %s (initrd) + %s (userspace) = %s.",
3661 format_timespan(kernel
, sizeof(kernel
), kernel_usec
, USEC_PER_MSEC
),
3662 format_timespan(initrd
, sizeof(initrd
), initrd_usec
, USEC_PER_MSEC
),
3663 format_timespan(userspace
, sizeof(userspace
), userspace_usec
, USEC_PER_MSEC
),
3664 format_timespan(sum
, sizeof(sum
), total_usec
, USEC_PER_MSEC
)));
3666 /* The initrd-less case on bare-metal*/
3668 kernel_usec
= m
->timestamps
[MANAGER_TIMESTAMP_USERSPACE
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_KERNEL
].monotonic
;
3671 log_struct(LOG_INFO
,
3672 "MESSAGE_ID=" SD_MESSAGE_STARTUP_FINISHED_STR
,
3673 "KERNEL_USEC="USEC_FMT
, kernel_usec
,
3674 "USERSPACE_USEC="USEC_FMT
, userspace_usec
,
3675 LOG_MESSAGE("Startup finished in %s%s (kernel) + %s (userspace) = %s.",
3677 format_timespan(kernel
, sizeof(kernel
), kernel_usec
, USEC_PER_MSEC
),
3678 format_timespan(userspace
, sizeof(userspace
), userspace_usec
, USEC_PER_MSEC
),
3679 format_timespan(sum
, sizeof(sum
), total_usec
, USEC_PER_MSEC
)));
3682 /* The container and --user case */
3683 firmware_usec
= loader_usec
= initrd_usec
= kernel_usec
= 0;
3684 total_usec
= userspace_usec
= m
->timestamps
[MANAGER_TIMESTAMP_FINISH
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_USERSPACE
].monotonic
;
3686 log_struct(LOG_INFO
,
3687 "MESSAGE_ID=" SD_MESSAGE_USER_STARTUP_FINISHED_STR
,
3688 "USERSPACE_USEC="USEC_FMT
, userspace_usec
,
3689 LOG_MESSAGE("Startup finished in %s.",
3690 format_timespan(sum
, sizeof(sum
), total_usec
, USEC_PER_MSEC
)));
3693 bus_manager_send_finished(m
, firmware_usec
, loader_usec
, kernel_usec
, initrd_usec
, userspace_usec
, total_usec
);
3696 m
->ready_sent
? "STATUS=Startup finished in %s."
3698 "STATUS=Startup finished in %s.",
3699 format_timespan(sum
, sizeof(sum
), total_usec
, USEC_PER_MSEC
));
3700 m
->ready_sent
= true;
3702 log_taint_string(m
);
3705 static void manager_send_ready(Manager
*m
) {
3708 /* We send READY=1 on reaching basic.target only when running in --user mode. */
3709 if (!MANAGER_IS_USER(m
) || m
->ready_sent
)
3712 m
->ready_sent
= true;
3716 "STATUS=Reached " SPECIAL_BASIC_TARGET
".");
3719 static void manager_check_basic_target(Manager
*m
) {
3724 /* Small shortcut */
3725 if (m
->ready_sent
&& m
->taint_logged
)
3728 u
= manager_get_unit(m
, SPECIAL_BASIC_TARGET
);
3729 if (!u
|| !UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u
)))
3732 /* For user managers, send out READY=1 as soon as we reach basic.target */
3733 manager_send_ready(m
);
3735 /* Log the taint string as soon as we reach basic.target */
3736 log_taint_string(m
);
3739 void manager_check_finished(Manager
*m
) {
3742 if (MANAGER_IS_RELOADING(m
))
3745 /* Verify that we have entered the event loop already, and not left it again. */
3746 if (!MANAGER_IS_RUNNING(m
))
3749 manager_check_basic_target(m
);
3751 if (hashmap_size(m
->jobs
) > 0) {
3752 if (m
->jobs_in_progress_event_source
)
3753 /* Ignore any failure, this is only for feedback */
3754 (void) sd_event_source_set_time(m
->jobs_in_progress_event_source
, now(CLOCK_MONOTONIC
) + JOBS_IN_PROGRESS_WAIT_USEC
);
3759 manager_flip_auto_status(m
, false);
3761 /* Notify Type=idle units that we are done now */
3762 manager_close_idle_pipe(m
);
3764 /* Turn off confirm spawn now */
3765 m
->confirm_spawn
= NULL
;
3767 /* No need to update ask password status when we're going non-interactive */
3768 manager_close_ask_password(m
);
3770 /* This is no longer the first boot */
3771 manager_set_first_boot(m
, false);
3773 if (MANAGER_IS_FINISHED(m
))
3776 dual_timestamp_get(m
->timestamps
+ MANAGER_TIMESTAMP_FINISH
);
3778 manager_notify_finished(m
);
3780 manager_invalidate_startup_units(m
);
3783 static bool generator_path_any(const char* const* paths
) {
3787 /* Optimize by skipping the whole process by not creating output directories
3788 * if no generators are found. */
3789 STRV_FOREACH(path
, (char**) paths
)
3790 if (access(*path
, F_OK
) == 0)
3792 else if (errno
!= ENOENT
)
3793 log_warning_errno(errno
, "Failed to open generator directory %s: %m", *path
);
3798 static const char* system_env_generator_binary_paths
[] = {
3799 "/run/systemd/system-environment-generators",
3800 "/etc/systemd/system-environment-generators",
3801 "/usr/local/lib/systemd/system-environment-generators",
3802 SYSTEM_ENV_GENERATOR_PATH
,
3806 static const char* user_env_generator_binary_paths
[] = {
3807 "/run/systemd/user-environment-generators",
3808 "/etc/systemd/user-environment-generators",
3809 "/usr/local/lib/systemd/user-environment-generators",
3810 USER_ENV_GENERATOR_PATH
,
3814 static int manager_run_environment_generators(Manager
*m
) {
3815 char **tmp
= NULL
; /* this is only used in the forked process, no cleanup here */
3817 void* args
[] = {&tmp
, &tmp
, &m
->environment
};
3819 if (MANAGER_IS_TEST_RUN(m
) && !(m
->test_run_flags
& MANAGER_TEST_RUN_ENV_GENERATORS
))
3822 paths
= MANAGER_IS_SYSTEM(m
) ? system_env_generator_binary_paths
: user_env_generator_binary_paths
;
3824 if (!generator_path_any(paths
))
3827 return execute_directories(paths
, DEFAULT_TIMEOUT_USEC
, gather_environment
, args
, NULL
, m
->environment
);
3830 static int manager_run_generators(Manager
*m
) {
3831 _cleanup_strv_free_
char **paths
= NULL
;
3832 const char *argv
[5];
3837 if (MANAGER_IS_TEST_RUN(m
) && !(m
->test_run_flags
& MANAGER_TEST_RUN_GENERATORS
))
3840 paths
= generator_binary_paths(m
->unit_file_scope
);
3844 if (!generator_path_any((const char* const*) paths
))
3847 r
= lookup_paths_mkdir_generator(&m
->lookup_paths
);
3849 log_error_errno(r
, "Failed to create generator directories: %m");
3853 argv
[0] = NULL
; /* Leave this empty, execute_directory() will fill something in */
3854 argv
[1] = m
->lookup_paths
.generator
;
3855 argv
[2] = m
->lookup_paths
.generator_early
;
3856 argv
[3] = m
->lookup_paths
.generator_late
;
3859 RUN_WITH_UMASK(0022)
3860 (void) execute_directories((const char* const*) paths
, DEFAULT_TIMEOUT_USEC
,
3861 NULL
, NULL
, (char**) argv
, m
->environment
);
3866 lookup_paths_trim_generator(&m
->lookup_paths
);
3870 int manager_environment_add(Manager
*m
, char **minus
, char **plus
) {
3871 char **a
= NULL
, **b
= NULL
, **l
;
3876 if (!strv_isempty(minus
)) {
3877 a
= strv_env_delete(l
, 1, minus
);
3884 if (!strv_isempty(plus
)) {
3885 b
= strv_env_merge(2, l
, plus
);
3894 if (m
->environment
!= l
)
3895 strv_free(m
->environment
);
3902 manager_sanitize_environment(m
);
3907 int manager_set_default_rlimits(Manager
*m
, struct rlimit
**default_rlimit
) {
3912 for (i
= 0; i
< _RLIMIT_MAX
; i
++) {
3913 m
->rlimit
[i
] = mfree(m
->rlimit
[i
]);
3915 if (!default_rlimit
[i
])
3918 m
->rlimit
[i
] = newdup(struct rlimit
, default_rlimit
[i
], 1);
3926 void manager_recheck_dbus(Manager
*m
) {
3929 /* Connects to the bus if the dbus service and socket are running. If we are running in user mode this is all
3930 * it does. In system mode we'll also connect to the system bus (which will most likely just reuse the
3931 * connection of the API bus). That's because the system bus after all runs as service of the system instance,
3932 * while in the user instance we can assume it's already there. */
3934 if (MANAGER_IS_RELOADING(m
))
3935 return; /* don't check while we are reloading… */
3937 if (manager_dbus_is_running(m
, false)) {
3938 (void) bus_init_api(m
);
3940 if (MANAGER_IS_SYSTEM(m
))
3941 (void) bus_init_system(m
);
3943 (void) bus_done_api(m
);
3945 if (MANAGER_IS_SYSTEM(m
))
3946 (void) bus_done_system(m
);
3950 static bool manager_journal_is_running(Manager
*m
) {
3955 if (MANAGER_IS_TEST_RUN(m
))
3958 /* If we are the user manager we can safely assume that the journal is up */
3959 if (!MANAGER_IS_SYSTEM(m
))
3962 /* Check that the socket is not only up, but in RUNNING state */
3963 u
= manager_get_unit(m
, SPECIAL_JOURNALD_SOCKET
);
3966 if (SOCKET(u
)->state
!= SOCKET_RUNNING
)
3969 /* Similar, check if the daemon itself is fully up, too */
3970 u
= manager_get_unit(m
, SPECIAL_JOURNALD_SERVICE
);
3973 if (!IN_SET(SERVICE(u
)->state
, SERVICE_RELOAD
, SERVICE_RUNNING
))
3979 void manager_recheck_journal(Manager
*m
) {
3983 /* Don't bother with this unless we are in the special situation of being PID 1 */
3984 if (getpid_cached() != 1)
3987 /* Don't check this while we are reloading, things might still change */
3988 if (MANAGER_IS_RELOADING(m
))
3991 /* The journal is fully and entirely up? If so, let's permit logging to it, if that's configured. If the
3992 * journal is down, don't ever log to it, otherwise we might end up deadlocking ourselves as we might trigger
3993 * an activation ourselves we can't fulfill. */
3994 log_set_prohibit_ipc(!manager_journal_is_running(m
));
3998 void manager_set_show_status(Manager
*m
, ShowStatus mode
) {
4000 assert(IN_SET(mode
, SHOW_STATUS_AUTO
, SHOW_STATUS_NO
, SHOW_STATUS_YES
, SHOW_STATUS_TEMPORARY
));
4002 if (!MANAGER_IS_SYSTEM(m
))
4005 if (m
->show_status
!= mode
)
4006 log_debug("%s showing of status.",
4007 mode
== SHOW_STATUS_NO
? "Disabling" : "Enabling");
4008 m
->show_status
= mode
;
4010 if (IN_SET(mode
, SHOW_STATUS_TEMPORARY
, SHOW_STATUS_YES
))
4011 (void) touch("/run/systemd/show-status");
4013 (void) unlink("/run/systemd/show-status");
4016 static bool manager_get_show_status(Manager
*m
, StatusType type
) {
4019 if (!MANAGER_IS_SYSTEM(m
))
4022 if (m
->no_console_output
)
4025 if (!IN_SET(manager_state(m
), MANAGER_INITIALIZING
, MANAGER_STARTING
, MANAGER_STOPPING
))
4028 /* If we cannot find out the status properly, just proceed. */
4029 if (type
!= STATUS_TYPE_EMERGENCY
&& manager_check_ask_password(m
) > 0)
4032 return IN_SET(m
->show_status
, SHOW_STATUS_TEMPORARY
, SHOW_STATUS_YES
);
4035 const char *manager_get_confirm_spawn(Manager
*m
) {
4036 static int last_errno
= 0;
4037 const char *vc
= m
->confirm_spawn
;
4041 /* Here's the deal: we want to test the validity of the console but don't want
4042 * PID1 to go through the whole console process which might block. But we also
4043 * want to warn the user only once if something is wrong with the console so we
4044 * cannot do the sanity checks after spawning our children. So here we simply do
4045 * really basic tests to hopefully trap common errors.
4047 * If the console suddenly disappear at the time our children will really it
4048 * then they will simply fail to acquire it and a positive answer will be
4049 * assumed. New children will fallback to /dev/console though.
4051 * Note: TTYs are devices that can come and go any time, and frequently aren't
4052 * available yet during early boot (consider a USB rs232 dongle...). If for any
4053 * reason the configured console is not ready, we fallback to the default
4056 if (!vc
|| path_equal(vc
, "/dev/console"))
4063 if (!S_ISCHR(st
.st_mode
)) {
4071 if (last_errno
!= errno
) {
4073 log_warning_errno(errno
, "Failed to open %s: %m, using default console", vc
);
4075 return "/dev/console";
4078 void manager_set_first_boot(Manager
*m
, bool b
) {
4081 if (!MANAGER_IS_SYSTEM(m
))
4084 if (m
->first_boot
!= (int) b
) {
4086 (void) touch("/run/systemd/first-boot");
4088 (void) unlink("/run/systemd/first-boot");
4094 void manager_disable_confirm_spawn(void) {
4095 (void) touch("/run/systemd/confirm_spawn_disabled");
4098 bool manager_is_confirm_spawn_disabled(Manager
*m
) {
4099 if (!m
->confirm_spawn
)
4102 return access("/run/systemd/confirm_spawn_disabled", F_OK
) >= 0;
4105 void manager_status_printf(Manager
*m
, StatusType type
, const char *status
, const char *format
, ...) {
4108 /* If m is NULL, assume we're after shutdown and let the messages through. */
4110 if (m
&& !manager_get_show_status(m
, type
))
4113 /* XXX We should totally drop the check for ephemeral here
4114 * and thus effectively make 'Type=idle' pointless. */
4115 if (type
== STATUS_TYPE_EPHEMERAL
&& m
&& m
->n_on_console
> 0)
4118 va_start(ap
, format
);
4119 status_vprintf(status
, true, type
== STATUS_TYPE_EPHEMERAL
, format
, ap
);
4123 Set
*manager_get_units_requiring_mounts_for(Manager
*m
, const char *path
) {
4124 char p
[strlen(path
)+1];
4130 path_simplify(p
, false);
4132 return hashmap_get(m
->units_requiring_mounts_for
, streq(p
, "/") ? "" : p
);
4135 int manager_update_failed_units(Manager
*m
, Unit
*u
, bool failed
) {
4140 assert(u
->manager
== m
);
4142 size
= set_size(m
->failed_units
);
4145 r
= set_ensure_allocated(&m
->failed_units
, NULL
);
4149 if (set_put(m
->failed_units
, u
) < 0)
4152 (void) set_remove(m
->failed_units
, u
);
4154 if (set_size(m
->failed_units
) != size
)
4155 bus_manager_send_change_signal(m
);
4160 ManagerState
manager_state(Manager
*m
) {
4165 /* Did we ever finish booting? If not then we are still starting up */
4166 if (!MANAGER_IS_FINISHED(m
)) {
4168 u
= manager_get_unit(m
, SPECIAL_BASIC_TARGET
);
4169 if (!u
|| !UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u
)))
4170 return MANAGER_INITIALIZING
;
4172 return MANAGER_STARTING
;
4175 /* Is the special shutdown target active or queued? If so, we are in shutdown state */
4176 u
= manager_get_unit(m
, SPECIAL_SHUTDOWN_TARGET
);
4177 if (u
&& unit_active_or_pending(u
))
4178 return MANAGER_STOPPING
;
4180 if (MANAGER_IS_SYSTEM(m
)) {
4181 /* Are the rescue or emergency targets active or queued? If so we are in maintenance state */
4182 u
= manager_get_unit(m
, SPECIAL_RESCUE_TARGET
);
4183 if (u
&& unit_active_or_pending(u
))
4184 return MANAGER_MAINTENANCE
;
4186 u
= manager_get_unit(m
, SPECIAL_EMERGENCY_TARGET
);
4187 if (u
&& unit_active_or_pending(u
))
4188 return MANAGER_MAINTENANCE
;
4191 /* Are there any failed units? If so, we are in degraded mode */
4192 if (set_size(m
->failed_units
) > 0)
4193 return MANAGER_DEGRADED
;
4195 return MANAGER_RUNNING
;
4198 #define DESTROY_IPC_FLAG (UINT32_C(1) << 31)
4200 static void manager_unref_uid_internal(
4205 int (*_clean_ipc
)(uid_t uid
)) {
4211 assert(uid_is_valid(uid
));
4214 /* A generic implementation, covering both manager_unref_uid() and manager_unref_gid(), under the assumption
4215 * that uid_t and gid_t are actually defined the same way, with the same validity rules.
4217 * We store a hashmap where the UID/GID is they key and the value is a 32bit reference counter, whose highest
4218 * bit is used as flag for marking UIDs/GIDs whose IPC objects to remove when the last reference to the UID/GID
4219 * is dropped. The flag is set to on, once at least one reference from a unit where RemoveIPC= is set is added
4220 * on a UID/GID. It is reset when the UID's/GID's reference counter drops to 0 again. */
4222 assert_cc(sizeof(uid_t
) == sizeof(gid_t
));
4223 assert_cc(UID_INVALID
== (uid_t
) GID_INVALID
);
4225 if (uid
== 0) /* We don't keep track of root, and will never destroy it */
4228 c
= PTR_TO_UINT32(hashmap_get(*uid_refs
, UID_TO_PTR(uid
)));
4230 n
= c
& ~DESTROY_IPC_FLAG
;
4234 if (destroy_now
&& n
== 0) {
4235 hashmap_remove(*uid_refs
, UID_TO_PTR(uid
));
4237 if (c
& DESTROY_IPC_FLAG
) {
4238 log_debug("%s " UID_FMT
" is no longer referenced, cleaning up its IPC.",
4239 _clean_ipc
== clean_ipc_by_uid
? "UID" : "GID",
4241 (void) _clean_ipc(uid
);
4244 c
= n
| (c
& DESTROY_IPC_FLAG
);
4245 assert_se(hashmap_update(*uid_refs
, UID_TO_PTR(uid
), UINT32_TO_PTR(c
)) >= 0);
4249 void manager_unref_uid(Manager
*m
, uid_t uid
, bool destroy_now
) {
4250 manager_unref_uid_internal(m
, &m
->uid_refs
, uid
, destroy_now
, clean_ipc_by_uid
);
4253 void manager_unref_gid(Manager
*m
, gid_t gid
, bool destroy_now
) {
4254 manager_unref_uid_internal(m
, &m
->gid_refs
, (uid_t
) gid
, destroy_now
, clean_ipc_by_gid
);
4257 static int manager_ref_uid_internal(
4268 assert(uid_is_valid(uid
));
4270 /* A generic implementation, covering both manager_ref_uid() and manager_ref_gid(), under the assumption
4271 * that uid_t and gid_t are actually defined the same way, with the same validity rules. */
4273 assert_cc(sizeof(uid_t
) == sizeof(gid_t
));
4274 assert_cc(UID_INVALID
== (uid_t
) GID_INVALID
);
4276 if (uid
== 0) /* We don't keep track of root, and will never destroy it */
4279 r
= hashmap_ensure_allocated(uid_refs
, &trivial_hash_ops
);
4283 c
= PTR_TO_UINT32(hashmap_get(*uid_refs
, UID_TO_PTR(uid
)));
4285 n
= c
& ~DESTROY_IPC_FLAG
;
4288 if (n
& DESTROY_IPC_FLAG
) /* check for overflow */
4291 c
= n
| (c
& DESTROY_IPC_FLAG
) | (clean_ipc
? DESTROY_IPC_FLAG
: 0);
4293 return hashmap_replace(*uid_refs
, UID_TO_PTR(uid
), UINT32_TO_PTR(c
));
4296 int manager_ref_uid(Manager
*m
, uid_t uid
, bool clean_ipc
) {
4297 return manager_ref_uid_internal(m
, &m
->uid_refs
, uid
, clean_ipc
);
4300 int manager_ref_gid(Manager
*m
, gid_t gid
, bool clean_ipc
) {
4301 return manager_ref_uid_internal(m
, &m
->gid_refs
, (uid_t
) gid
, clean_ipc
);
4304 static void manager_vacuum_uid_refs_internal(
4307 int (*_clean_ipc
)(uid_t uid
)) {
4316 HASHMAP_FOREACH_KEY(p
, k
, *uid_refs
, i
) {
4320 uid
= PTR_TO_UID(k
);
4321 c
= PTR_TO_UINT32(p
);
4323 n
= c
& ~DESTROY_IPC_FLAG
;
4327 if (c
& DESTROY_IPC_FLAG
) {
4328 log_debug("Found unreferenced %s " UID_FMT
" after reload/reexec. Cleaning up.",
4329 _clean_ipc
== clean_ipc_by_uid
? "UID" : "GID",
4331 (void) _clean_ipc(uid
);
4334 assert_se(hashmap_remove(*uid_refs
, k
) == p
);
4338 void manager_vacuum_uid_refs(Manager
*m
) {
4339 manager_vacuum_uid_refs_internal(m
, &m
->uid_refs
, clean_ipc_by_uid
);
4342 void manager_vacuum_gid_refs(Manager
*m
) {
4343 manager_vacuum_uid_refs_internal(m
, &m
->gid_refs
, clean_ipc_by_gid
);
4346 static void manager_serialize_uid_refs_internal(
4350 const char *field_name
) {
4360 /* Serialize the UID reference table. Or actually, just the IPC destruction flag of it, as the actual counter
4361 * of it is better rebuild after a reload/reexec. */
4363 HASHMAP_FOREACH_KEY(p
, k
, *uid_refs
, i
) {
4367 uid
= PTR_TO_UID(k
);
4368 c
= PTR_TO_UINT32(p
);
4370 if (!(c
& DESTROY_IPC_FLAG
))
4373 fprintf(f
, "%s=" UID_FMT
"\n", field_name
, uid
);
4377 void manager_serialize_uid_refs(Manager
*m
, FILE *f
) {
4378 manager_serialize_uid_refs_internal(m
, f
, &m
->uid_refs
, "destroy-ipc-uid");
4381 void manager_serialize_gid_refs(Manager
*m
, FILE *f
) {
4382 manager_serialize_uid_refs_internal(m
, f
, &m
->gid_refs
, "destroy-ipc-gid");
4385 static void manager_deserialize_uid_refs_one_internal(
4388 const char *value
) {
4398 r
= parse_uid(value
, &uid
);
4399 if (r
< 0 || uid
== 0) {
4400 log_debug("Unable to parse UID reference serialization");
4404 r
= hashmap_ensure_allocated(uid_refs
, &trivial_hash_ops
);
4410 c
= PTR_TO_UINT32(hashmap_get(*uid_refs
, UID_TO_PTR(uid
)));
4411 if (c
& DESTROY_IPC_FLAG
)
4414 c
|= DESTROY_IPC_FLAG
;
4416 r
= hashmap_replace(*uid_refs
, UID_TO_PTR(uid
), UINT32_TO_PTR(c
));
4418 log_debug("Failed to add UID reference entry");
4423 void manager_deserialize_uid_refs_one(Manager
*m
, const char *value
) {
4424 manager_deserialize_uid_refs_one_internal(m
, &m
->uid_refs
, value
);
4427 void manager_deserialize_gid_refs_one(Manager
*m
, const char *value
) {
4428 manager_deserialize_uid_refs_one_internal(m
, &m
->gid_refs
, value
);
4431 int manager_dispatch_user_lookup_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
4435 char unit_name
[UNIT_NAME_MAX
+1];
4438 Manager
*m
= userdata
;
4446 /* Invoked whenever a child process succeeded resolving its user/group to use and sent us the resulting UID/GID
4447 * in a datagram. We parse the datagram here and pass it off to the unit, so that it can add a reference to the
4448 * UID/GID so that it can destroy the UID/GID's IPC objects when the reference counter drops to 0. */
4450 l
= recv(fd
, &buffer
, sizeof(buffer
), MSG_DONTWAIT
);
4452 if (IN_SET(errno
, EINTR
, EAGAIN
))
4455 return log_error_errno(errno
, "Failed to read from user lookup fd: %m");
4458 if ((size_t) l
<= offsetof(struct buffer
, unit_name
)) {
4459 log_warning("Received too short user lookup message, ignoring.");
4463 if ((size_t) l
> offsetof(struct buffer
, unit_name
) + UNIT_NAME_MAX
) {
4464 log_warning("Received too long user lookup message, ignoring.");
4468 if (!uid_is_valid(buffer
.uid
) && !gid_is_valid(buffer
.gid
)) {
4469 log_warning("Got user lookup message with invalid UID/GID pair, ignoring.");
4473 n
= (size_t) l
- offsetof(struct buffer
, unit_name
);
4474 if (memchr(buffer
.unit_name
, 0, n
)) {
4475 log_warning("Received lookup message with embedded NUL character, ignoring.");
4479 buffer
.unit_name
[n
] = 0;
4480 u
= manager_get_unit(m
, buffer
.unit_name
);
4482 log_debug("Got user lookup message but unit doesn't exist, ignoring.");
4486 log_unit_debug(u
, "User lookup succeeded: uid=" UID_FMT
" gid=" GID_FMT
, buffer
.uid
, buffer
.gid
);
4488 unit_notify_user_lookup(u
, buffer
.uid
, buffer
.gid
);
4492 char *manager_taint_string(Manager
*m
) {
4493 _cleanup_free_
char *destination
= NULL
, *overflowuid
= NULL
, *overflowgid
= NULL
;
4497 /* Returns a "taint string", e.g. "local-hwclock:var-run-bad".
4498 * Only things that are detected at runtime should be tagged
4499 * here. For stuff that is set during compilation, emit a warning
4500 * in the configuration phase. */
4504 buf
= new(char, sizeof("split-usr:"
4508 "overflowuid-not-65534:"
4509 "overflowgid-not-65534:"));
4517 e
= stpcpy(e
, "split-usr:");
4519 if (access("/proc/cgroups", F_OK
) < 0)
4520 e
= stpcpy(e
, "cgroups-missing:");
4522 if (clock_is_localtime(NULL
) > 0)
4523 e
= stpcpy(e
, "local-hwclock:");
4525 r
= readlink_malloc("/var/run", &destination
);
4526 if (r
< 0 || !PATH_IN_SET(destination
, "../run", "/run"))
4527 e
= stpcpy(e
, "var-run-bad:");
4529 r
= read_one_line_file("/proc/sys/kernel/overflowuid", &overflowuid
);
4530 if (r
>= 0 && !streq(overflowuid
, "65534"))
4531 e
= stpcpy(e
, "overflowuid-not-65534:");
4533 r
= read_one_line_file("/proc/sys/kernel/overflowgid", &overflowgid
);
4534 if (r
>= 0 && !streq(overflowgid
, "65534"))
4535 e
= stpcpy(e
, "overflowgid-not-65534:");
4537 /* remove the last ':' */
4544 void manager_ref_console(Manager
*m
) {
4550 void manager_unref_console(Manager
*m
) {
4552 assert(m
->n_on_console
> 0);
4555 if (m
->n_on_console
== 0)
4556 m
->no_console_output
= false; /* unset no_console_output flag, since the console is definitely free now */
4559 void manager_override_log_level(Manager
*m
, int level
) {
4560 _cleanup_free_
char *s
= NULL
;
4563 if (!m
->log_level_overridden
) {
4564 m
->original_log_level
= log_get_max_level();
4565 m
->log_level_overridden
= true;
4568 (void) log_level_to_string_alloc(level
, &s
);
4569 log_info("Setting log level to %s.", strna(s
));
4571 log_set_max_level(level
);
4574 void manager_restore_original_log_level(Manager
*m
) {
4575 _cleanup_free_
char *s
= NULL
;
4578 if (!m
->log_level_overridden
)
4581 (void) log_level_to_string_alloc(m
->original_log_level
, &s
);
4582 log_info("Restoring log level to original (%s).", strna(s
));
4584 log_set_max_level(m
->original_log_level
);
4585 m
->log_level_overridden
= false;
4588 void manager_override_log_target(Manager
*m
, LogTarget target
) {
4591 if (!m
->log_target_overridden
) {
4592 m
->original_log_target
= log_get_target();
4593 m
->log_target_overridden
= true;
4596 log_info("Setting log target to %s.", log_target_to_string(target
));
4597 log_set_target(target
);
4600 void manager_restore_original_log_target(Manager
*m
) {
4603 if (!m
->log_target_overridden
)
4606 log_info("Restoring log target to original %s.", log_target_to_string(m
->original_log_target
));
4608 log_set_target(m
->original_log_target
);
4609 m
->log_target_overridden
= false;
4612 ManagerTimestamp
manager_timestamp_initrd_mangle(ManagerTimestamp s
) {
4614 s
>= MANAGER_TIMESTAMP_SECURITY_START
&&
4615 s
<= MANAGER_TIMESTAMP_UNITS_LOAD_FINISH
)
4616 return s
- MANAGER_TIMESTAMP_SECURITY_START
+ MANAGER_TIMESTAMP_INITRD_SECURITY_START
;
4620 static const char *const manager_state_table
[_MANAGER_STATE_MAX
] = {
4621 [MANAGER_INITIALIZING
] = "initializing",
4622 [MANAGER_STARTING
] = "starting",
4623 [MANAGER_RUNNING
] = "running",
4624 [MANAGER_DEGRADED
] = "degraded",
4625 [MANAGER_MAINTENANCE
] = "maintenance",
4626 [MANAGER_STOPPING
] = "stopping",
4629 DEFINE_STRING_TABLE_LOOKUP(manager_state
, ManagerState
);
4631 static const char *const manager_timestamp_table
[_MANAGER_TIMESTAMP_MAX
] = {
4632 [MANAGER_TIMESTAMP_FIRMWARE
] = "firmware",
4633 [MANAGER_TIMESTAMP_LOADER
] = "loader",
4634 [MANAGER_TIMESTAMP_KERNEL
] = "kernel",
4635 [MANAGER_TIMESTAMP_INITRD
] = "initrd",
4636 [MANAGER_TIMESTAMP_USERSPACE
] = "userspace",
4637 [MANAGER_TIMESTAMP_FINISH
] = "finish",
4638 [MANAGER_TIMESTAMP_SECURITY_START
] = "security-start",
4639 [MANAGER_TIMESTAMP_SECURITY_FINISH
] = "security-finish",
4640 [MANAGER_TIMESTAMP_GENERATORS_START
] = "generators-start",
4641 [MANAGER_TIMESTAMP_GENERATORS_FINISH
] = "generators-finish",
4642 [MANAGER_TIMESTAMP_UNITS_LOAD_START
] = "units-load-start",
4643 [MANAGER_TIMESTAMP_UNITS_LOAD_FINISH
] = "units-load-finish",
4644 [MANAGER_TIMESTAMP_INITRD_SECURITY_START
] = "initrd-security-start",
4645 [MANAGER_TIMESTAMP_INITRD_SECURITY_FINISH
] = "initrd-security-finish",
4646 [MANAGER_TIMESTAMP_INITRD_GENERATORS_START
] = "initrd-generators-start",
4647 [MANAGER_TIMESTAMP_INITRD_GENERATORS_FINISH
] = "initrd-generators-finish",
4648 [MANAGER_TIMESTAMP_INITRD_UNITS_LOAD_START
] = "initrd-units-load-start",
4649 [MANAGER_TIMESTAMP_INITRD_UNITS_LOAD_FINISH
] = "initrd-units-load-finish",
4652 DEFINE_STRING_TABLE_LOOKUP(manager_timestamp
, ManagerTimestamp
);