2 This file is part of systemd.
4 Copyright 2010 Lennart Poettering
6 systemd is free software; you can redistribute it and/or modify it
7 under the terms of the GNU Lesser General Public License as published by
8 the Free Software Foundation; either version 2.1 of the License, or
9 (at your option) any later version.
11 systemd is distributed in the hope that it will be useful, but
12 WITHOUT ANY WARRANTY; without even the implied warranty of
13 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 Lesser General Public License for more details.
16 You should have received a copy of the GNU Lesser General Public License
17 along with systemd; If not, see <http://www.gnu.org/licenses/>.
25 #include <sys/epoll.h>
26 #include <sys/inotify.h>
27 #include <sys/ioctl.h>
28 #include <sys/reboot.h>
29 #include <sys/timerfd.h>
37 #include "sd-daemon.h"
38 #include "sd-messages.h"
41 #include "alloc-util.h"
43 #include "boot-timestamps.h"
44 #include "bus-common-errors.h"
45 #include "bus-error.h"
46 #include "bus-kernel.h"
48 #include "clean-ipc.h"
50 #include "dbus-manager.h"
51 #include "dbus-unit.h"
53 #include "dirent-util.h"
56 #include "exec-util.h"
58 #include "exit-status.h"
65 #include "locale-setup.h"
71 #include "parse-util.h"
72 #include "path-lookup.h"
73 #include "path-util.h"
74 #include "process-util.h"
75 #include "ratelimit.h"
77 #include "signal-util.h"
79 #include "stat-util.h"
80 #include "string-table.h"
81 #include "string-util.h"
83 #include "terminal-util.h"
84 #include "time-util.h"
85 #include "transaction.h"
86 #include "umask-util.h"
87 #include "unit-name.h"
88 #include "user-util.h"
93 #define NOTIFY_RCVBUF_SIZE (8*1024*1024)
94 #define CGROUPS_AGENT_RCVBUF_SIZE (8*1024*1024)
96 /* Initial delay and the interval for printing status messages about running jobs */
97 #define JOBS_IN_PROGRESS_WAIT_USEC (5*USEC_PER_SEC)
98 #define JOBS_IN_PROGRESS_PERIOD_USEC (USEC_PER_SEC / 3)
99 #define JOBS_IN_PROGRESS_PERIOD_DIVISOR 3
101 static int manager_dispatch_notify_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
102 static int manager_dispatch_cgroups_agent_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
103 static int manager_dispatch_signal_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
104 static int manager_dispatch_time_change_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
105 static int manager_dispatch_idle_pipe_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
106 static int manager_dispatch_user_lookup_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
107 static int manager_dispatch_jobs_in_progress(sd_event_source
*source
, usec_t usec
, void *userdata
);
108 static int manager_dispatch_run_queue(sd_event_source
*source
, void *userdata
);
109 static int manager_run_environment_generators(Manager
*m
);
110 static int manager_run_generators(Manager
*m
);
112 static void manager_watch_jobs_in_progress(Manager
*m
) {
118 /* We do not want to show the cylon animation if the user
119 * needs to confirm service executions otherwise confirmation
120 * messages will be screwed by the cylon animation. */
121 if (!manager_is_confirm_spawn_disabled(m
))
124 if (m
->jobs_in_progress_event_source
)
127 next
= now(CLOCK_MONOTONIC
) + JOBS_IN_PROGRESS_WAIT_USEC
;
128 r
= sd_event_add_time(
130 &m
->jobs_in_progress_event_source
,
133 manager_dispatch_jobs_in_progress
, m
);
137 (void) sd_event_source_set_description(m
->jobs_in_progress_event_source
, "manager-jobs-in-progress");
140 #define CYLON_BUFFER_EXTRA (2*(sizeof(ANSI_RED)-1) + sizeof(ANSI_HIGHLIGHT_RED)-1 + 2*(sizeof(ANSI_NORMAL)-1))
142 static void draw_cylon(char buffer
[], size_t buflen
, unsigned width
, unsigned pos
) {
145 assert(buflen
>= CYLON_BUFFER_EXTRA
+ width
+ 1);
146 assert(pos
<= width
+1); /* 0 or width+1 mean that the center light is behind the corner */
150 p
= mempset(p
, ' ', pos
-2);
151 if (log_get_show_color())
152 p
= stpcpy(p
, ANSI_RED
);
156 if (pos
> 0 && pos
<= width
) {
157 if (log_get_show_color())
158 p
= stpcpy(p
, ANSI_HIGHLIGHT_RED
);
162 if (log_get_show_color())
163 p
= stpcpy(p
, ANSI_NORMAL
);
166 if (log_get_show_color())
167 p
= stpcpy(p
, ANSI_RED
);
170 p
= mempset(p
, ' ', width
-1-pos
);
171 if (log_get_show_color())
172 strcpy(p
, ANSI_NORMAL
);
176 void manager_flip_auto_status(Manager
*m
, bool enable
) {
180 if (m
->show_status
== SHOW_STATUS_AUTO
)
181 manager_set_show_status(m
, SHOW_STATUS_TEMPORARY
);
183 if (m
->show_status
== SHOW_STATUS_TEMPORARY
)
184 manager_set_show_status(m
, SHOW_STATUS_AUTO
);
188 static void manager_print_jobs_in_progress(Manager
*m
) {
189 _cleanup_free_
char *job_of_n
= NULL
;
192 unsigned counter
= 0, print_nr
;
193 char cylon
[6 + CYLON_BUFFER_EXTRA
+ 1];
195 char time
[FORMAT_TIMESPAN_MAX
], limit
[FORMAT_TIMESPAN_MAX
] = "no limit";
199 assert(m
->n_running_jobs
> 0);
201 manager_flip_auto_status(m
, true);
203 print_nr
= (m
->jobs_in_progress_iteration
/ JOBS_IN_PROGRESS_PERIOD_DIVISOR
) % m
->n_running_jobs
;
205 HASHMAP_FOREACH(j
, m
->jobs
, i
)
206 if (j
->state
== JOB_RUNNING
&& counter
++ == print_nr
)
209 /* m->n_running_jobs must be consistent with the contents of m->jobs,
210 * so the above loop must have succeeded in finding j. */
211 assert(counter
== print_nr
+ 1);
214 cylon_pos
= m
->jobs_in_progress_iteration
% 14;
216 cylon_pos
= 14 - cylon_pos
;
217 draw_cylon(cylon
, sizeof(cylon
), 6, cylon_pos
);
219 m
->jobs_in_progress_iteration
++;
221 if (m
->n_running_jobs
> 1) {
222 if (asprintf(&job_of_n
, "(%u of %u) ", counter
, m
->n_running_jobs
) < 0)
226 format_timespan(time
, sizeof(time
), now(CLOCK_MONOTONIC
) - j
->begin_usec
, 1*USEC_PER_SEC
);
227 if (job_get_timeout(j
, &x
) > 0)
228 format_timespan(limit
, sizeof(limit
), x
- j
->begin_usec
, 1*USEC_PER_SEC
);
230 manager_status_printf(m
, STATUS_TYPE_EPHEMERAL
, cylon
,
231 "%sA %s job is running for %s (%s / %s)",
233 job_type_to_string(j
->type
),
234 unit_description(j
->unit
),
238 static int have_ask_password(void) {
239 _cleanup_closedir_
DIR *dir
;
242 dir
= opendir("/run/systemd/ask-password");
250 FOREACH_DIRENT_ALL(de
, dir
, return -errno
) {
251 if (startswith(de
->d_name
, "ask."))
257 static int manager_dispatch_ask_password_fd(sd_event_source
*source
,
258 int fd
, uint32_t revents
, void *userdata
) {
259 Manager
*m
= userdata
;
265 m
->have_ask_password
= have_ask_password();
266 if (m
->have_ask_password
< 0)
267 /* Log error but continue. Negative have_ask_password
268 * is treated as unknown status. */
269 log_error_errno(m
->have_ask_password
, "Failed to list /run/systemd/ask-password: %m");
274 static void manager_close_ask_password(Manager
*m
) {
277 m
->ask_password_event_source
= sd_event_source_unref(m
->ask_password_event_source
);
278 m
->ask_password_inotify_fd
= safe_close(m
->ask_password_inotify_fd
);
279 m
->have_ask_password
= -EINVAL
;
282 static int manager_check_ask_password(Manager
*m
) {
287 if (!m
->ask_password_event_source
) {
288 assert(m
->ask_password_inotify_fd
< 0);
290 mkdir_p_label("/run/systemd/ask-password", 0755);
292 m
->ask_password_inotify_fd
= inotify_init1(IN_NONBLOCK
|IN_CLOEXEC
);
293 if (m
->ask_password_inotify_fd
< 0)
294 return log_error_errno(errno
, "inotify_init1() failed: %m");
296 if (inotify_add_watch(m
->ask_password_inotify_fd
, "/run/systemd/ask-password", IN_CREATE
|IN_DELETE
|IN_MOVE
) < 0) {
297 log_error_errno(errno
, "Failed to add watch on /run/systemd/ask-password: %m");
298 manager_close_ask_password(m
);
302 r
= sd_event_add_io(m
->event
, &m
->ask_password_event_source
,
303 m
->ask_password_inotify_fd
, EPOLLIN
,
304 manager_dispatch_ask_password_fd
, m
);
306 log_error_errno(errno
, "Failed to add event source for /run/systemd/ask-password: %m");
307 manager_close_ask_password(m
);
311 (void) sd_event_source_set_description(m
->ask_password_event_source
, "manager-ask-password");
313 /* Queries might have been added meanwhile... */
314 manager_dispatch_ask_password_fd(m
->ask_password_event_source
,
315 m
->ask_password_inotify_fd
, EPOLLIN
, m
);
318 return m
->have_ask_password
;
321 static int manager_watch_idle_pipe(Manager
*m
) {
326 if (m
->idle_pipe_event_source
)
329 if (m
->idle_pipe
[2] < 0)
332 r
= sd_event_add_io(m
->event
, &m
->idle_pipe_event_source
, m
->idle_pipe
[2], EPOLLIN
, manager_dispatch_idle_pipe_fd
, m
);
334 return log_error_errno(r
, "Failed to watch idle pipe: %m");
336 (void) sd_event_source_set_description(m
->idle_pipe_event_source
, "manager-idle-pipe");
341 static void manager_close_idle_pipe(Manager
*m
) {
344 m
->idle_pipe_event_source
= sd_event_source_unref(m
->idle_pipe_event_source
);
346 safe_close_pair(m
->idle_pipe
);
347 safe_close_pair(m
->idle_pipe
+ 2);
350 static int manager_setup_time_change(Manager
*m
) {
353 /* We only care for the cancellation event, hence we set the
354 * timeout to the latest possible value. */
355 struct itimerspec its
= {
356 .it_value
.tv_sec
= TIME_T_MAX
,
360 assert_cc(sizeof(time_t) == sizeof(TIME_T_MAX
));
362 if (m
->test_run_flags
)
365 /* Uses TFD_TIMER_CANCEL_ON_SET to get notifications whenever
366 * CLOCK_REALTIME makes a jump relative to CLOCK_MONOTONIC */
368 m
->time_change_fd
= timerfd_create(CLOCK_REALTIME
, TFD_NONBLOCK
|TFD_CLOEXEC
);
369 if (m
->time_change_fd
< 0)
370 return log_error_errno(errno
, "Failed to create timerfd: %m");
372 if (timerfd_settime(m
->time_change_fd
, TFD_TIMER_ABSTIME
|TFD_TIMER_CANCEL_ON_SET
, &its
, NULL
) < 0) {
373 log_debug_errno(errno
, "Failed to set up TFD_TIMER_CANCEL_ON_SET, ignoring: %m");
374 m
->time_change_fd
= safe_close(m
->time_change_fd
);
378 r
= sd_event_add_io(m
->event
, &m
->time_change_event_source
, m
->time_change_fd
, EPOLLIN
, manager_dispatch_time_change_fd
, m
);
380 return log_error_errno(r
, "Failed to create time change event source: %m");
382 (void) sd_event_source_set_description(m
->time_change_event_source
, "manager-time-change");
384 log_debug("Set up TFD_TIMER_CANCEL_ON_SET timerfd.");
389 static int enable_special_signals(Manager
*m
) {
390 _cleanup_close_
int fd
= -1;
394 if (m
->test_run_flags
)
397 /* Enable that we get SIGINT on control-alt-del. In containers
398 * this will fail with EPERM (older) or EINVAL (newer), so
400 if (reboot(RB_DISABLE_CAD
) < 0 && !IN_SET(errno
, EPERM
, EINVAL
))
401 log_warning_errno(errno
, "Failed to enable ctrl-alt-del handling: %m");
403 fd
= open_terminal("/dev/tty0", O_RDWR
|O_NOCTTY
|O_CLOEXEC
);
405 /* Support systems without virtual console */
407 log_warning_errno(errno
, "Failed to open /dev/tty0: %m");
409 /* Enable that we get SIGWINCH on kbrequest */
410 if (ioctl(fd
, KDSIGACCEPT
, SIGWINCH
) < 0)
411 log_warning_errno(errno
, "Failed to enable kbrequest handling: %m");
417 static int manager_setup_signals(Manager
*m
) {
418 struct sigaction sa
= {
419 .sa_handler
= SIG_DFL
,
420 .sa_flags
= SA_NOCLDSTOP
|SA_RESTART
,
427 assert_se(sigaction(SIGCHLD
, &sa
, NULL
) == 0);
429 /* We make liberal use of realtime signals here. On
430 * Linux/glibc we have 30 of them (with the exception of Linux
431 * on hppa, see below), between SIGRTMIN+0 ... SIGRTMIN+30
434 assert_se(sigemptyset(&mask
) == 0);
435 sigset_add_many(&mask
,
436 SIGCHLD
, /* Child died */
437 SIGTERM
, /* Reexecute daemon */
438 SIGHUP
, /* Reload configuration */
439 SIGUSR1
, /* systemd/upstart: reconnect to D-Bus */
440 SIGUSR2
, /* systemd: dump status */
441 SIGINT
, /* Kernel sends us this on control-alt-del */
442 SIGWINCH
, /* Kernel sends us this on kbrequest (alt-arrowup) */
443 SIGPWR
, /* Some kernel drivers and upsd send us this on power failure */
445 SIGRTMIN
+0, /* systemd: start default.target */
446 SIGRTMIN
+1, /* systemd: isolate rescue.target */
447 SIGRTMIN
+2, /* systemd: isolate emergency.target */
448 SIGRTMIN
+3, /* systemd: start halt.target */
449 SIGRTMIN
+4, /* systemd: start poweroff.target */
450 SIGRTMIN
+5, /* systemd: start reboot.target */
451 SIGRTMIN
+6, /* systemd: start kexec.target */
453 /* ... space for more special targets ... */
455 SIGRTMIN
+13, /* systemd: Immediate halt */
456 SIGRTMIN
+14, /* systemd: Immediate poweroff */
457 SIGRTMIN
+15, /* systemd: Immediate reboot */
458 SIGRTMIN
+16, /* systemd: Immediate kexec */
460 /* ... space for more immediate system state changes ... */
462 SIGRTMIN
+20, /* systemd: enable status messages */
463 SIGRTMIN
+21, /* systemd: disable status messages */
464 SIGRTMIN
+22, /* systemd: set log level to LOG_DEBUG */
465 SIGRTMIN
+23, /* systemd: set log level to LOG_INFO */
466 SIGRTMIN
+24, /* systemd: Immediate exit (--user only) */
468 /* .. one free signal here ... */
470 #if !defined(__hppa64__) && !defined(__hppa__)
471 /* Apparently Linux on hppa has fewer RT
472 * signals (SIGRTMAX is SIGRTMIN+25 there),
473 * hence let's not try to make use of them
474 * here. Since these commands are accessible
475 * by different means and only really a safety
476 * net, the missing functionality on hppa
477 * shouldn't matter. */
479 SIGRTMIN
+26, /* systemd: set log target to journal-or-kmsg */
480 SIGRTMIN
+27, /* systemd: set log target to console */
481 SIGRTMIN
+28, /* systemd: set log target to kmsg */
482 SIGRTMIN
+29, /* systemd: set log target to syslog-or-kmsg (obsolete) */
484 /* ... one free signal here SIGRTMIN+30 ... */
487 assert_se(sigprocmask(SIG_SETMASK
, &mask
, NULL
) == 0);
489 m
->signal_fd
= signalfd(-1, &mask
, SFD_NONBLOCK
|SFD_CLOEXEC
);
490 if (m
->signal_fd
< 0)
493 r
= sd_event_add_io(m
->event
, &m
->signal_event_source
, m
->signal_fd
, EPOLLIN
, manager_dispatch_signal_fd
, m
);
497 (void) sd_event_source_set_description(m
->signal_event_source
, "manager-signal");
499 /* Process signals a bit earlier than the rest of things, but later than notify_fd processing, so that the
500 * notify processing can still figure out to which process/service a message belongs, before we reap the
501 * process. Also, process this before handling cgroup notifications, so that we always collect child exit
502 * status information before detecting that there's no process in a cgroup. */
503 r
= sd_event_source_set_priority(m
->signal_event_source
, SD_EVENT_PRIORITY_NORMAL
-6);
507 if (MANAGER_IS_SYSTEM(m
))
508 return enable_special_signals(m
);
513 static void manager_clean_environment(Manager
*m
) {
516 /* Let's remove some environment variables that we
517 * need ourselves to communicate with our clients */
532 static int manager_default_environment(Manager
*m
) {
535 if (MANAGER_IS_SYSTEM(m
)) {
536 /* The system manager always starts with a clean
537 * environment for its children. It does not import
538 * the kernel's or the parents' exported variables.
540 * The initial passed environment is untouched to keep
541 * /proc/self/environ valid; it is used for tagging
542 * the init process inside containers. */
543 m
->environment
= strv_new("PATH=" DEFAULT_PATH
,
546 /* Import locale variables LC_*= from configuration */
547 locale_setup(&m
->environment
);
549 /* The user manager passes its own environment
550 * along to its children. */
551 m
->environment
= strv_copy(environ
);
556 manager_clean_environment(m
);
557 strv_sort(m
->environment
);
562 static int manager_setup_prefix(Manager
*m
) {
568 static const struct table_entry paths_system
[_EXEC_DIRECTORY_TYPE_MAX
] = {
569 [EXEC_DIRECTORY_RUNTIME
] = { SD_PATH_SYSTEM_RUNTIME
, NULL
},
570 [EXEC_DIRECTORY_STATE
] = { SD_PATH_SYSTEM_STATE_PRIVATE
, NULL
},
571 [EXEC_DIRECTORY_CACHE
] = { SD_PATH_SYSTEM_STATE_CACHE
, NULL
},
572 [EXEC_DIRECTORY_LOGS
] = { SD_PATH_SYSTEM_STATE_LOGS
, NULL
},
573 [EXEC_DIRECTORY_CONFIGURATION
] = { SD_PATH_SYSTEM_CONFIGURATION
, NULL
},
576 static const struct table_entry paths_user
[_EXEC_DIRECTORY_TYPE_MAX
] = {
577 [EXEC_DIRECTORY_RUNTIME
] = { SD_PATH_USER_RUNTIME
, NULL
},
578 [EXEC_DIRECTORY_STATE
] = { SD_PATH_USER_CONFIGURATION
, NULL
},
579 [EXEC_DIRECTORY_CACHE
] = { SD_PATH_USER_STATE_CACHE
, NULL
},
580 [EXEC_DIRECTORY_LOGS
] = { SD_PATH_USER_CONFIGURATION
, "log" },
581 [EXEC_DIRECTORY_CONFIGURATION
] = { SD_PATH_USER_CONFIGURATION
, NULL
},
584 const struct table_entry
*p
;
590 if (MANAGER_IS_SYSTEM(m
))
595 for (i
= 0; i
< _EXEC_DIRECTORY_TYPE_MAX
; i
++) {
596 r
= sd_path_home(p
[i
].type
, p
[i
].suffix
, &m
->prefix
[i
]);
604 int manager_new(UnitFileScope scope
, unsigned test_run_flags
, Manager
**_m
) {
609 assert(IN_SET(scope
, UNIT_FILE_SYSTEM
, UNIT_FILE_USER
));
611 m
= new0(Manager
, 1);
615 m
->unit_file_scope
= scope
;
616 m
->exit_code
= _MANAGER_EXIT_CODE_INVALID
;
617 m
->default_timer_accuracy_usec
= USEC_PER_MINUTE
;
618 m
->default_tasks_accounting
= true;
619 m
->default_tasks_max
= UINT64_MAX
;
620 m
->default_timeout_start_usec
= DEFAULT_TIMEOUT_USEC
;
621 m
->default_timeout_stop_usec
= DEFAULT_TIMEOUT_USEC
;
622 m
->default_restart_usec
= DEFAULT_RESTART_USEC
;
625 if (MANAGER_IS_SYSTEM(m
) && detect_container() <= 0)
626 boot_timestamps(&m
->userspace_timestamp
, &m
->firmware_timestamp
, &m
->loader_timestamp
);
629 /* Prepare log fields we can use for structured logging */
630 if (MANAGER_IS_SYSTEM(m
)) {
631 m
->unit_log_field
= "UNIT=";
632 m
->unit_log_format_string
= "UNIT=%s";
634 m
->invocation_log_field
= "INVOCATION_ID=";
635 m
->invocation_log_format_string
= "INVOCATION_ID=%s";
637 m
->unit_log_field
= "USER_UNIT=";
638 m
->unit_log_format_string
= "USER_UNIT=%s";
640 m
->invocation_log_field
= "USER_INVOCATION_ID=";
641 m
->invocation_log_format_string
= "USER_INVOCATION_ID=%s";
644 m
->idle_pipe
[0] = m
->idle_pipe
[1] = m
->idle_pipe
[2] = m
->idle_pipe
[3] = -1;
646 m
->pin_cgroupfs_fd
= m
->notify_fd
= m
->cgroups_agent_fd
= m
->signal_fd
= m
->time_change_fd
=
647 m
->dev_autofs_fd
= m
->private_listen_fd
= m
->cgroup_inotify_fd
=
648 m
->ask_password_inotify_fd
= -1;
650 m
->user_lookup_fds
[0] = m
->user_lookup_fds
[1] = -1;
652 m
->current_job_id
= 1; /* start as id #1, so that we can leave #0 around as "null-like" value */
654 m
->have_ask_password
= -EINVAL
; /* we don't know */
657 m
->test_run_flags
= test_run_flags
;
659 /* Reboot immediately if the user hits C-A-D more often than 7x per 2s */
660 RATELIMIT_INIT(m
->ctrl_alt_del_ratelimit
, 2 * USEC_PER_SEC
, 7);
662 r
= manager_default_environment(m
);
666 r
= hashmap_ensure_allocated(&m
->units
, &string_hash_ops
);
670 r
= hashmap_ensure_allocated(&m
->jobs
, NULL
);
674 r
= hashmap_ensure_allocated(&m
->cgroup_unit
, &string_hash_ops
);
678 r
= hashmap_ensure_allocated(&m
->watch_bus
, &string_hash_ops
);
682 r
= sd_event_default(&m
->event
);
686 r
= sd_event_add_defer(m
->event
, &m
->run_queue_event_source
, manager_dispatch_run_queue
, m
);
690 r
= sd_event_source_set_priority(m
->run_queue_event_source
, SD_EVENT_PRIORITY_IDLE
);
694 r
= sd_event_source_set_enabled(m
->run_queue_event_source
, SD_EVENT_OFF
);
698 (void) sd_event_source_set_description(m
->run_queue_event_source
, "manager-run-queue");
700 r
= manager_setup_signals(m
);
704 r
= manager_setup_cgroup(m
);
708 r
= manager_setup_time_change(m
);
712 m
->udev
= udev_new();
718 if (MANAGER_IS_SYSTEM(m
)) {
719 r
= mkdir_label("/run/systemd/units", 0755);
720 if (r
< 0 && r
!= -EEXIST
)
724 /* Note that we do not set up the notify fd here. We do that after deserialization,
725 * since they might have gotten serialized across the reexec. */
727 m
->taint_usr
= dir_is_empty("/usr") > 0;
729 r
= manager_setup_prefix(m
);
741 static int manager_setup_notify(Manager
*m
) {
744 if (m
->test_run_flags
)
747 if (m
->notify_fd
< 0) {
748 _cleanup_close_
int fd
= -1;
749 union sockaddr_union sa
= {
750 .sa
.sa_family
= AF_UNIX
,
752 static const int one
= 1;
754 /* First free all secondary fields */
755 m
->notify_socket
= mfree(m
->notify_socket
);
756 m
->notify_event_source
= sd_event_source_unref(m
->notify_event_source
);
758 fd
= socket(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
|SOCK_NONBLOCK
, 0);
760 return log_error_errno(errno
, "Failed to allocate notification socket: %m");
762 fd_inc_rcvbuf(fd
, NOTIFY_RCVBUF_SIZE
);
764 m
->notify_socket
= strappend(m
->prefix
[EXEC_DIRECTORY_RUNTIME
], "/systemd/notify");
765 if (!m
->notify_socket
)
768 (void) mkdir_parents_label(m
->notify_socket
, 0755);
769 (void) unlink(m
->notify_socket
);
771 strncpy(sa
.un
.sun_path
, m
->notify_socket
, sizeof(sa
.un
.sun_path
)-1);
772 r
= bind(fd
, &sa
.sa
, SOCKADDR_UN_LEN(sa
.un
));
774 return log_error_errno(errno
, "bind(%s) failed: %m", sa
.un
.sun_path
);
776 r
= setsockopt(fd
, SOL_SOCKET
, SO_PASSCRED
, &one
, sizeof(one
));
778 return log_error_errno(errno
, "SO_PASSCRED failed: %m");
783 log_debug("Using notification socket %s", m
->notify_socket
);
786 if (!m
->notify_event_source
) {
787 r
= sd_event_add_io(m
->event
, &m
->notify_event_source
, m
->notify_fd
, EPOLLIN
, manager_dispatch_notify_fd
, m
);
789 return log_error_errno(r
, "Failed to allocate notify event source: %m");
791 /* Process notification messages a bit earlier than SIGCHLD, so that we can still identify to which
792 * service an exit message belongs. */
793 r
= sd_event_source_set_priority(m
->notify_event_source
, SD_EVENT_PRIORITY_NORMAL
-7);
795 return log_error_errno(r
, "Failed to set priority of notify event source: %m");
797 (void) sd_event_source_set_description(m
->notify_event_source
, "manager-notify");
803 static int manager_setup_cgroups_agent(Manager
*m
) {
805 static const union sockaddr_union sa
= {
806 .un
.sun_family
= AF_UNIX
,
807 .un
.sun_path
= "/run/systemd/cgroups-agent",
811 /* This creates a listening socket we receive cgroups agent messages on. We do not use D-Bus for delivering
812 * these messages from the cgroups agent binary to PID 1, as the cgroups agent binary is very short-living, and
813 * each instance of it needs a new D-Bus connection. Since D-Bus connections are SOCK_STREAM/AF_UNIX, on
814 * overloaded systems the backlog of the D-Bus socket becomes relevant, as not more than the configured number
815 * of D-Bus connections may be queued until the kernel will start dropping further incoming connections,
816 * possibly resulting in lost cgroups agent messages. To avoid this, we'll use a private SOCK_DGRAM/AF_UNIX
817 * socket, where no backlog is relevant as communication may take place without an actual connect() cycle, and
818 * we thus won't lose messages.
820 * Note that PID 1 will forward the agent message to system bus, so that the user systemd instance may listen
821 * to it. The system instance hence listens on this special socket, but the user instances listen on the system
822 * bus for these messages. */
824 if (m
->test_run_flags
)
827 if (!MANAGER_IS_SYSTEM(m
))
830 r
= cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER
);
832 return log_error_errno(r
, "Failed to determine whether unified cgroups hierarchy is used: %m");
833 if (r
> 0) /* We don't need this anymore on the unified hierarchy */
836 if (m
->cgroups_agent_fd
< 0) {
837 _cleanup_close_
int fd
= -1;
839 /* First free all secondary fields */
840 m
->cgroups_agent_event_source
= sd_event_source_unref(m
->cgroups_agent_event_source
);
842 fd
= socket(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
|SOCK_NONBLOCK
, 0);
844 return log_error_errno(errno
, "Failed to allocate cgroups agent socket: %m");
846 fd_inc_rcvbuf(fd
, CGROUPS_AGENT_RCVBUF_SIZE
);
848 (void) unlink(sa
.un
.sun_path
);
850 /* Only allow root to connect to this socket */
852 r
= bind(fd
, &sa
.sa
, SOCKADDR_UN_LEN(sa
.un
));
854 return log_error_errno(errno
, "bind(%s) failed: %m", sa
.un
.sun_path
);
856 m
->cgroups_agent_fd
= fd
;
860 if (!m
->cgroups_agent_event_source
) {
861 r
= sd_event_add_io(m
->event
, &m
->cgroups_agent_event_source
, m
->cgroups_agent_fd
, EPOLLIN
, manager_dispatch_cgroups_agent_fd
, m
);
863 return log_error_errno(r
, "Failed to allocate cgroups agent event source: %m");
865 /* Process cgroups notifications early, but after having processed service notification messages or
866 * SIGCHLD signals, so that a cgroup running empty is always just the last safety net of notification,
867 * and we collected the metadata the notification and SIGCHLD stuff offers first. Also see handling of
868 * cgroup inotify for the unified cgroup stuff. */
869 r
= sd_event_source_set_priority(m
->cgroups_agent_event_source
, SD_EVENT_PRIORITY_NORMAL
-4);
871 return log_error_errno(r
, "Failed to set priority of cgroups agent event source: %m");
873 (void) sd_event_source_set_description(m
->cgroups_agent_event_source
, "manager-cgroups-agent");
879 static int manager_setup_user_lookup_fd(Manager
*m
) {
884 /* Set up the socket pair used for passing UID/GID resolution results from forked off processes to PID
885 * 1. Background: we can't do name lookups (NSS) from PID 1, since it might involve IPC and thus activation,
886 * and we might hence deadlock on ourselves. Hence we do all user/group lookups asynchronously from the forked
887 * off processes right before executing the binaries to start. In order to be able to clean up any IPC objects
888 * created by a unit (see RemoveIPC=) we need to know in PID 1 the used UID/GID of the executed processes,
889 * hence we establish this communication channel so that forked off processes can pass their UID/GID
890 * information back to PID 1. The forked off processes send their resolved UID/GID to PID 1 in a simple
891 * datagram, along with their unit name, so that we can share one communication socket pair among all units for
894 * You might wonder why we need a communication channel for this that is independent of the usual notification
895 * socket scheme (i.e. $NOTIFY_SOCKET). The primary difference is about trust: data sent via the $NOTIFY_SOCKET
896 * channel is only accepted if it originates from the right unit and if reception was enabled for it. The user
897 * lookup socket OTOH is only accessible by PID 1 and its children until they exec(), and always available.
899 * Note that this function is called under two circumstances: when we first initialize (in which case we
900 * allocate both the socket pair and the event source to listen on it), and when we deserialize after a reload
901 * (in which case the socket pair already exists but we still need to allocate the event source for it). */
903 if (m
->user_lookup_fds
[0] < 0) {
905 /* Free all secondary fields */
906 safe_close_pair(m
->user_lookup_fds
);
907 m
->user_lookup_event_source
= sd_event_source_unref(m
->user_lookup_event_source
);
909 if (socketpair(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
, 0, m
->user_lookup_fds
) < 0)
910 return log_error_errno(errno
, "Failed to allocate user lookup socket: %m");
912 (void) fd_inc_rcvbuf(m
->user_lookup_fds
[0], NOTIFY_RCVBUF_SIZE
);
915 if (!m
->user_lookup_event_source
) {
916 r
= sd_event_add_io(m
->event
, &m
->user_lookup_event_source
, m
->user_lookup_fds
[0], EPOLLIN
, manager_dispatch_user_lookup_fd
, m
);
918 return log_error_errno(errno
, "Failed to allocate user lookup event source: %m");
920 /* Process even earlier than the notify event source, so that we always know first about valid UID/GID
922 r
= sd_event_source_set_priority(m
->user_lookup_event_source
, SD_EVENT_PRIORITY_NORMAL
-8);
924 return log_error_errno(errno
, "Failed to set priority ot user lookup event source: %m");
926 (void) sd_event_source_set_description(m
->user_lookup_event_source
, "user-lookup");
932 static int manager_connect_bus(Manager
*m
, bool reexecuting
) {
933 bool try_bus_connect
;
938 if (m
->test_run_flags
)
941 u
= manager_get_unit(m
, SPECIAL_DBUS_SERVICE
);
944 (u
&& SERVICE(u
)->deserialized_state
== SERVICE_RUNNING
) &&
946 (MANAGER_IS_USER(m
) && getenv("DBUS_SESSION_BUS_ADDRESS")));
948 /* Try to connect to the buses, if possible. */
949 return bus_init(m
, try_bus_connect
);
952 static unsigned manager_dispatch_cleanup_queue(Manager
*m
) {
958 while ((u
= m
->cleanup_queue
)) {
959 assert(u
->in_cleanup_queue
);
969 GC_OFFSET_IN_PATH
, /* This one is on the path we were traveling */
970 GC_OFFSET_UNSURE
, /* No clue */
971 GC_OFFSET_GOOD
, /* We still need this unit */
972 GC_OFFSET_BAD
, /* We don't need this unit anymore */
976 static void unit_gc_mark_good(Unit
*u
, unsigned gc_marker
) {
981 u
->gc_marker
= gc_marker
+ GC_OFFSET_GOOD
;
983 /* Recursively mark referenced units as GOOD as well */
984 HASHMAP_FOREACH_KEY(v
, other
, u
->dependencies
[UNIT_REFERENCES
], i
)
985 if (other
->gc_marker
== gc_marker
+ GC_OFFSET_UNSURE
)
986 unit_gc_mark_good(other
, gc_marker
);
989 static void unit_gc_sweep(Unit
*u
, unsigned gc_marker
) {
997 if (IN_SET(u
->gc_marker
- gc_marker
,
998 GC_OFFSET_GOOD
, GC_OFFSET_BAD
, GC_OFFSET_UNSURE
, GC_OFFSET_IN_PATH
))
1001 if (u
->in_cleanup_queue
)
1004 if (unit_check_gc(u
))
1007 u
->gc_marker
= gc_marker
+ GC_OFFSET_IN_PATH
;
1011 HASHMAP_FOREACH_KEY(v
, other
, u
->dependencies
[UNIT_REFERENCED_BY
], i
) {
1012 unit_gc_sweep(other
, gc_marker
);
1014 if (other
->gc_marker
== gc_marker
+ GC_OFFSET_GOOD
)
1017 if (other
->gc_marker
!= gc_marker
+ GC_OFFSET_BAD
)
1024 /* We were unable to find anything out about this entry, so
1025 * let's investigate it later */
1026 u
->gc_marker
= gc_marker
+ GC_OFFSET_UNSURE
;
1027 unit_add_to_gc_queue(u
);
1031 /* We definitely know that this one is not useful anymore, so
1032 * let's mark it for deletion */
1033 u
->gc_marker
= gc_marker
+ GC_OFFSET_BAD
;
1034 unit_add_to_cleanup_queue(u
);
1038 unit_gc_mark_good(u
, gc_marker
);
1041 static unsigned manager_dispatch_gc_unit_queue(Manager
*m
) {
1042 unsigned n
= 0, gc_marker
;
1047 /* log_debug("Running GC..."); */
1049 m
->gc_marker
+= _GC_OFFSET_MAX
;
1050 if (m
->gc_marker
+ _GC_OFFSET_MAX
<= _GC_OFFSET_MAX
)
1053 gc_marker
= m
->gc_marker
;
1055 while ((u
= m
->gc_unit_queue
)) {
1056 assert(u
->in_gc_queue
);
1058 unit_gc_sweep(u
, gc_marker
);
1060 LIST_REMOVE(gc_queue
, m
->gc_unit_queue
, u
);
1061 u
->in_gc_queue
= false;
1065 if (IN_SET(u
->gc_marker
- gc_marker
,
1066 GC_OFFSET_BAD
, GC_OFFSET_UNSURE
)) {
1068 log_unit_debug(u
, "Collecting.");
1069 u
->gc_marker
= gc_marker
+ GC_OFFSET_BAD
;
1070 unit_add_to_cleanup_queue(u
);
1077 static unsigned manager_dispatch_gc_job_queue(Manager
*m
) {
1083 while ((j
= m
->gc_job_queue
)) {
1084 assert(j
->in_gc_queue
);
1086 LIST_REMOVE(gc_queue
, m
->gc_job_queue
, j
);
1087 j
->in_gc_queue
= false;
1091 if (job_check_gc(j
))
1094 log_unit_debug(j
->unit
, "Collecting job.");
1095 (void) job_finish_and_invalidate(j
, JOB_COLLECTED
, false, false);
1101 static void manager_clear_jobs_and_units(Manager
*m
) {
1106 while ((u
= hashmap_first(m
->units
)))
1109 manager_dispatch_cleanup_queue(m
);
1111 assert(!m
->load_queue
);
1112 assert(!m
->run_queue
);
1113 assert(!m
->dbus_unit_queue
);
1114 assert(!m
->dbus_job_queue
);
1115 assert(!m
->cleanup_queue
);
1116 assert(!m
->gc_unit_queue
);
1117 assert(!m
->gc_job_queue
);
1119 assert(hashmap_isempty(m
->jobs
));
1120 assert(hashmap_isempty(m
->units
));
1122 m
->n_on_console
= 0;
1123 m
->n_running_jobs
= 0;
1126 Manager
* manager_free(Manager
*m
) {
1129 ExecDirectoryType dt
;
1134 manager_clear_jobs_and_units(m
);
1136 for (c
= 0; c
< _UNIT_TYPE_MAX
; c
++)
1137 if (unit_vtable
[c
]->shutdown
)
1138 unit_vtable
[c
]->shutdown(m
);
1140 /* If we reexecute ourselves, we keep the root cgroup around */
1141 manager_shutdown_cgroup(m
, m
->exit_code
!= MANAGER_REEXECUTE
);
1143 lookup_paths_flush_generator(&m
->lookup_paths
);
1147 dynamic_user_vacuum(m
, false);
1148 hashmap_free(m
->dynamic_users
);
1150 hashmap_free(m
->units
);
1151 hashmap_free(m
->units_by_invocation_id
);
1152 hashmap_free(m
->jobs
);
1153 hashmap_free(m
->watch_pids1
);
1154 hashmap_free(m
->watch_pids2
);
1155 hashmap_free(m
->watch_bus
);
1157 set_free(m
->startup_units
);
1158 set_free(m
->failed_units
);
1160 sd_event_source_unref(m
->signal_event_source
);
1161 sd_event_source_unref(m
->notify_event_source
);
1162 sd_event_source_unref(m
->cgroups_agent_event_source
);
1163 sd_event_source_unref(m
->time_change_event_source
);
1164 sd_event_source_unref(m
->jobs_in_progress_event_source
);
1165 sd_event_source_unref(m
->run_queue_event_source
);
1166 sd_event_source_unref(m
->user_lookup_event_source
);
1168 safe_close(m
->signal_fd
);
1169 safe_close(m
->notify_fd
);
1170 safe_close(m
->cgroups_agent_fd
);
1171 safe_close(m
->time_change_fd
);
1172 safe_close_pair(m
->user_lookup_fds
);
1174 manager_close_ask_password(m
);
1176 manager_close_idle_pipe(m
);
1178 udev_unref(m
->udev
);
1179 sd_event_unref(m
->event
);
1181 free(m
->notify_socket
);
1183 lookup_paths_free(&m
->lookup_paths
);
1184 strv_free(m
->environment
);
1186 hashmap_free(m
->cgroup_unit
);
1187 set_free_free(m
->unit_path_cache
);
1189 free(m
->switch_root
);
1190 free(m
->switch_root_init
);
1192 for (i
= 0; i
< _RLIMIT_MAX
; i
++)
1193 m
->rlimit
[i
] = mfree(m
->rlimit
[i
]);
1195 assert(hashmap_isempty(m
->units_requiring_mounts_for
));
1196 hashmap_free(m
->units_requiring_mounts_for
);
1198 hashmap_free(m
->uid_refs
);
1199 hashmap_free(m
->gid_refs
);
1201 for (dt
= 0; dt
< _EXEC_DIRECTORY_TYPE_MAX
; dt
++)
1202 m
->prefix
[dt
] = mfree(m
->prefix
[dt
]);
1207 void manager_enumerate(Manager
*m
) {
1212 /* Let's ask every type to load all units from disk/kernel
1213 * that it might know */
1214 for (c
= 0; c
< _UNIT_TYPE_MAX
; c
++) {
1215 if (!unit_type_supported(c
)) {
1216 log_debug("Unit type .%s is not supported on this system.", unit_type_to_string(c
));
1220 if (!unit_vtable
[c
]->enumerate
)
1223 unit_vtable
[c
]->enumerate(m
);
1226 manager_dispatch_load_queue(m
);
1229 static void manager_coldplug(Manager
*m
) {
1237 /* Then, let's set up their initial state. */
1238 HASHMAP_FOREACH_KEY(u
, k
, m
->units
, i
) {
1240 /* ignore aliases */
1244 r
= unit_coldplug(u
);
1246 log_warning_errno(r
, "We couldn't coldplug %s, proceeding anyway: %m", u
->id
);
1250 static void manager_build_unit_path_cache(Manager
*m
) {
1256 set_free_free(m
->unit_path_cache
);
1258 m
->unit_path_cache
= set_new(&string_hash_ops
);
1259 if (!m
->unit_path_cache
) {
1264 /* This simply builds a list of files we know exist, so that
1265 * we don't always have to go to disk */
1267 STRV_FOREACH(i
, m
->lookup_paths
.search_path
) {
1268 _cleanup_closedir_
DIR *d
= NULL
;
1273 if (errno
!= ENOENT
)
1274 log_warning_errno(errno
, "Failed to open directory %s, ignoring: %m", *i
);
1278 FOREACH_DIRENT(de
, d
, r
= -errno
; goto fail
) {
1281 p
= strjoin(streq(*i
, "/") ? "" : *i
, "/", de
->d_name
);
1287 r
= set_consume(m
->unit_path_cache
, p
);
1296 log_warning_errno(r
, "Failed to build unit path cache, proceeding without: %m");
1297 m
->unit_path_cache
= set_free_free(m
->unit_path_cache
);
1300 static void manager_distribute_fds(Manager
*m
, FDSet
*fds
) {
1306 HASHMAP_FOREACH(u
, m
->units
, i
) {
1308 if (fdset_size(fds
) <= 0)
1311 if (!UNIT_VTABLE(u
)->distribute_fds
)
1314 UNIT_VTABLE(u
)->distribute_fds(u
, fds
);
1318 int manager_startup(Manager
*m
, FILE *serialization
, FDSet
*fds
) {
1323 /* If we are running in test mode, we still want to run the generators,
1324 * but we should not touch the real generator directories. */
1325 r
= lookup_paths_init(&m
->lookup_paths
, m
->unit_file_scope
,
1326 m
->test_run_flags
? LOOKUP_PATHS_TEMPORARY_GENERATED
: 0,
1331 r
= manager_run_environment_generators(m
);
1335 /* Make sure the transient directory always exists, so that it remains
1336 * in the search path */
1337 r
= mkdir_p_label(m
->lookup_paths
.transient
, 0755);
1339 return log_error_errno(r
, "Failed to create transient generator directory \"%s\": %m",
1340 m
->lookup_paths
.transient
);
1342 dual_timestamp_get(&m
->generators_start_timestamp
);
1343 r
= manager_run_generators(m
);
1344 dual_timestamp_get(&m
->generators_finish_timestamp
);
1348 /* If this is the first boot, and we are in the host system, then preset everything */
1349 if (m
->first_boot
> 0 &&
1350 MANAGER_IS_SYSTEM(m
) &&
1351 !m
->test_run_flags
) {
1353 r
= unit_file_preset_all(UNIT_FILE_SYSTEM
, 0, NULL
, UNIT_FILE_PRESET_ENABLE_ONLY
, NULL
, 0);
1355 log_full_errno(r
== -EEXIST
? LOG_NOTICE
: LOG_WARNING
, r
,
1356 "Failed to populate /etc with preset unit settings, ignoring: %m");
1358 log_info("Populated /etc with preset unit settings.");
1361 lookup_paths_reduce(&m
->lookup_paths
);
1362 manager_build_unit_path_cache(m
);
1364 /* If we will deserialize make sure that during enumeration
1365 * this is already known, so we increase the counter here
1370 /* First, enumerate what we can from all config files */
1371 dual_timestamp_get(&m
->units_load_start_timestamp
);
1372 manager_enumerate(m
);
1373 dual_timestamp_get(&m
->units_load_finish_timestamp
);
1375 /* Second, deserialize if there is something to deserialize */
1376 if (serialization
) {
1377 r
= manager_deserialize(m
, serialization
, fds
);
1379 return log_error_errno(r
, "Deserialization failed: %m");
1382 /* Any fds left? Find some unit which wants them. This is
1383 * useful to allow container managers to pass some file
1384 * descriptors to us pre-initialized. This enables
1385 * socket-based activation of entire containers. */
1386 manager_distribute_fds(m
, fds
);
1388 /* We might have deserialized the notify fd, but if we didn't
1389 * then let's create the bus now */
1390 r
= manager_setup_notify(m
);
1392 /* No sense to continue without notifications, our children would fail anyway. */
1395 r
= manager_setup_cgroups_agent(m
);
1397 /* Likewise, no sense to continue without empty cgroup notifications. */
1400 r
= manager_setup_user_lookup_fd(m
);
1402 /* This shouldn't fail, except if things are really broken. */
1405 /* Let's connect to the bus now. */
1406 (void) manager_connect_bus(m
, !!serialization
);
1408 (void) bus_track_coldplug(m
, &m
->subscribed
, false, m
->deserialized_subscribed
);
1409 m
->deserialized_subscribed
= strv_free(m
->deserialized_subscribed
);
1411 /* Third, fire things up! */
1412 manager_coldplug(m
);
1414 /* Release any dynamic users no longer referenced */
1415 dynamic_user_vacuum(m
, true);
1417 /* Release any references to UIDs/GIDs no longer referenced, and destroy any IPC owned by them */
1418 manager_vacuum_uid_refs(m
);
1419 manager_vacuum_gid_refs(m
);
1421 if (serialization
) {
1422 assert(m
->n_reloading
> 0);
1425 /* Let's wait for the UnitNew/JobNew messages being
1426 * sent, before we notify that the reload is
1428 m
->send_reloading_done
= true;
1434 int manager_add_job(Manager
*m
, JobType type
, Unit
*unit
, JobMode mode
, sd_bus_error
*e
, Job
**_ret
) {
1439 assert(type
< _JOB_TYPE_MAX
);
1441 assert(mode
< _JOB_MODE_MAX
);
1443 if (mode
== JOB_ISOLATE
&& type
!= JOB_START
)
1444 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
, "Isolate is only valid for start.");
1446 if (mode
== JOB_ISOLATE
&& !unit
->allow_isolate
)
1447 return sd_bus_error_setf(e
, BUS_ERROR_NO_ISOLATION
, "Operation refused, unit may not be isolated.");
1449 log_unit_debug(unit
, "Trying to enqueue job %s/%s/%s", unit
->id
, job_type_to_string(type
), job_mode_to_string(mode
));
1451 type
= job_type_collapse(type
, unit
);
1453 tr
= transaction_new(mode
== JOB_REPLACE_IRREVERSIBLY
);
1457 r
= transaction_add_job_and_dependencies(tr
, type
, unit
, NULL
, true, false,
1458 IN_SET(mode
, JOB_IGNORE_DEPENDENCIES
, JOB_IGNORE_REQUIREMENTS
),
1459 mode
== JOB_IGNORE_DEPENDENCIES
, e
);
1463 if (mode
== JOB_ISOLATE
) {
1464 r
= transaction_add_isolate_jobs(tr
, m
);
1469 r
= transaction_activate(tr
, m
, mode
, e
);
1473 log_unit_debug(unit
,
1474 "Enqueued job %s/%s as %u", unit
->id
,
1475 job_type_to_string(type
), (unsigned) tr
->anchor_job
->id
);
1478 *_ret
= tr
->anchor_job
;
1480 transaction_free(tr
);
1484 transaction_abort(tr
);
1485 transaction_free(tr
);
1489 int manager_add_job_by_name(Manager
*m
, JobType type
, const char *name
, JobMode mode
, sd_bus_error
*e
, Job
**ret
) {
1490 Unit
*unit
= NULL
; /* just to appease gcc, initialization is not really necessary */
1494 assert(type
< _JOB_TYPE_MAX
);
1496 assert(mode
< _JOB_MODE_MAX
);
1498 r
= manager_load_unit(m
, name
, NULL
, NULL
, &unit
);
1503 return manager_add_job(m
, type
, unit
, mode
, e
, ret
);
1506 int manager_add_job_by_name_and_warn(Manager
*m
, JobType type
, const char *name
, JobMode mode
, Job
**ret
) {
1507 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
1511 assert(type
< _JOB_TYPE_MAX
);
1513 assert(mode
< _JOB_MODE_MAX
);
1515 r
= manager_add_job_by_name(m
, type
, name
, mode
, &error
, ret
);
1517 return log_warning_errno(r
, "Failed to enqueue %s job for %s: %s", job_mode_to_string(mode
), name
, bus_error_message(&error
, r
));
1522 int manager_propagate_reload(Manager
*m
, Unit
*unit
, JobMode mode
, sd_bus_error
*e
) {
1528 assert(mode
< _JOB_MODE_MAX
);
1529 assert(mode
!= JOB_ISOLATE
); /* Isolate is only valid for start */
1531 tr
= transaction_new(mode
== JOB_REPLACE_IRREVERSIBLY
);
1535 /* We need an anchor job */
1536 r
= transaction_add_job_and_dependencies(tr
, JOB_NOP
, unit
, NULL
, false, false, true, true, e
);
1540 /* Failure in adding individual dependencies is ignored, so this always succeeds. */
1541 transaction_add_propagate_reload_jobs(tr
, unit
, tr
->anchor_job
, mode
== JOB_IGNORE_DEPENDENCIES
, e
);
1543 r
= transaction_activate(tr
, m
, mode
, e
);
1547 transaction_free(tr
);
1551 transaction_abort(tr
);
1552 transaction_free(tr
);
1556 Job
*manager_get_job(Manager
*m
, uint32_t id
) {
1559 return hashmap_get(m
->jobs
, UINT32_TO_PTR(id
));
1562 Unit
*manager_get_unit(Manager
*m
, const char *name
) {
1566 return hashmap_get(m
->units
, name
);
1569 unsigned manager_dispatch_load_queue(Manager
*m
) {
1575 /* Make sure we are not run recursively */
1576 if (m
->dispatching_load_queue
)
1579 m
->dispatching_load_queue
= true;
1581 /* Dispatches the load queue. Takes a unit from the queue and
1582 * tries to load its data until the queue is empty */
1584 while ((u
= m
->load_queue
)) {
1585 assert(u
->in_load_queue
);
1591 m
->dispatching_load_queue
= false;
1595 int manager_load_unit_prepare(
1607 assert(name
|| path
);
1610 /* This will prepare the unit for loading, but not actually
1611 * load anything from disk. */
1613 if (path
&& !is_path(path
))
1614 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
, "Path %s is not absolute.", path
);
1617 name
= basename(path
);
1619 t
= unit_name_to_type(name
);
1621 if (t
== _UNIT_TYPE_INVALID
|| !unit_name_is_valid(name
, UNIT_NAME_PLAIN
|UNIT_NAME_INSTANCE
)) {
1622 if (unit_name_is_valid(name
, UNIT_NAME_TEMPLATE
))
1623 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
, "Unit name %s is missing the instance name.", name
);
1625 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
, "Unit name %s is not valid.", name
);
1628 ret
= manager_get_unit(m
, name
);
1634 ret
= unit_new(m
, unit_vtable
[t
]->object_size
);
1639 ret
->fragment_path
= strdup(path
);
1640 if (!ret
->fragment_path
) {
1646 r
= unit_add_name(ret
, name
);
1652 unit_add_to_load_queue(ret
);
1653 unit_add_to_dbus_queue(ret
);
1654 unit_add_to_gc_queue(ret
);
1661 int manager_load_unit(
1673 /* This will load the service information files, but not actually
1674 * start any services or anything. */
1676 r
= manager_load_unit_prepare(m
, name
, path
, e
, _ret
);
1680 manager_dispatch_load_queue(m
);
1682 *_ret
= unit_follow_merge(*_ret
);
1687 void manager_dump_jobs(Manager
*s
, FILE *f
, const char *prefix
) {
1694 HASHMAP_FOREACH(j
, s
->jobs
, i
)
1695 job_dump(j
, f
, prefix
);
1698 void manager_dump_units(Manager
*s
, FILE *f
, const char *prefix
) {
1706 HASHMAP_FOREACH_KEY(u
, t
, s
->units
, i
)
1708 unit_dump(u
, f
, prefix
);
1711 void manager_clear_jobs(Manager
*m
) {
1716 while ((j
= hashmap_first(m
->jobs
)))
1717 /* No need to recurse. We're cancelling all jobs. */
1718 job_finish_and_invalidate(j
, JOB_CANCELED
, false, false);
1721 static int manager_dispatch_run_queue(sd_event_source
*source
, void *userdata
) {
1722 Manager
*m
= userdata
;
1728 while ((j
= m
->run_queue
)) {
1729 assert(j
->installed
);
1730 assert(j
->in_run_queue
);
1732 job_run_and_invalidate(j
);
1735 if (m
->n_running_jobs
> 0)
1736 manager_watch_jobs_in_progress(m
);
1738 if (m
->n_on_console
> 0)
1739 manager_watch_idle_pipe(m
);
1744 static unsigned manager_dispatch_dbus_queue(Manager
*m
) {
1751 if (m
->dispatching_dbus_queue
)
1754 m
->dispatching_dbus_queue
= true;
1756 while ((u
= m
->dbus_unit_queue
)) {
1757 assert(u
->in_dbus_queue
);
1759 bus_unit_send_change_signal(u
);
1763 while ((j
= m
->dbus_job_queue
)) {
1764 assert(j
->in_dbus_queue
);
1766 bus_job_send_change_signal(j
);
1770 m
->dispatching_dbus_queue
= false;
1772 if (m
->send_reloading_done
) {
1773 m
->send_reloading_done
= false;
1775 bus_manager_send_reloading(m
, false);
1778 if (m
->queued_message
)
1779 bus_send_queued_message(m
);
1784 static int manager_dispatch_cgroups_agent_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
1785 Manager
*m
= userdata
;
1786 char buf
[PATH_MAX
+1];
1789 n
= recv(fd
, buf
, sizeof(buf
), 0);
1791 return log_error_errno(errno
, "Failed to read cgroups agent message: %m");
1793 log_error("Got zero-length cgroups agent message, ignoring.");
1796 if ((size_t) n
>= sizeof(buf
)) {
1797 log_error("Got overly long cgroups agent message, ignoring.");
1801 if (memchr(buf
, 0, n
)) {
1802 log_error("Got cgroups agent message with embedded NUL byte, ignoring.");
1807 manager_notify_cgroup_empty(m
, buf
);
1808 (void) bus_forward_agent_released(m
, buf
);
1813 static void manager_invoke_notify_message(Manager
*m
, Unit
*u
, pid_t pid
, const char *buf
, FDSet
*fds
) {
1814 _cleanup_strv_free_
char **tags
= NULL
;
1820 tags
= strv_split(buf
, "\n\r");
1826 if (UNIT_VTABLE(u
)->notify_message
)
1827 UNIT_VTABLE(u
)->notify_message(u
, pid
, tags
, fds
);
1828 else if (_unlikely_(log_get_max_level() >= LOG_DEBUG
)) {
1829 _cleanup_free_
char *x
= NULL
, *y
= NULL
;
1833 y
= ellipsize(x
, 20, 90);
1834 log_unit_debug(u
, "Got notification message \"%s\", ignoring.", strnull(y
));
1838 static int manager_dispatch_notify_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
1840 _cleanup_fdset_free_ FDSet
*fds
= NULL
;
1841 Manager
*m
= userdata
;
1842 char buf
[NOTIFY_BUFFER_MAX
+1];
1843 struct iovec iovec
= {
1845 .iov_len
= sizeof(buf
)-1,
1848 struct cmsghdr cmsghdr
;
1849 uint8_t buf
[CMSG_SPACE(sizeof(struct ucred
)) +
1850 CMSG_SPACE(sizeof(int) * NOTIFY_FD_MAX
)];
1852 struct msghdr msghdr
= {
1855 .msg_control
= &control
,
1856 .msg_controllen
= sizeof(control
),
1859 struct cmsghdr
*cmsg
;
1860 struct ucred
*ucred
= NULL
;
1862 int r
, *fd_array
= NULL
;
1867 assert(m
->notify_fd
== fd
);
1869 if (revents
!= EPOLLIN
) {
1870 log_warning("Got unexpected poll event for notify fd.");
1874 n
= recvmsg(m
->notify_fd
, &msghdr
, MSG_DONTWAIT
|MSG_CMSG_CLOEXEC
|MSG_TRUNC
);
1876 if (IN_SET(errno
, EAGAIN
, EINTR
))
1877 return 0; /* Spurious wakeup, try again */
1879 /* If this is any other, real error, then let's stop processing this socket. This of course means we
1880 * won't take notification messages anymore, but that's still better than busy looping around this:
1881 * being woken up over and over again but being unable to actually read the message off the socket. */
1882 return log_error_errno(errno
, "Failed to receive notification message: %m");
1885 CMSG_FOREACH(cmsg
, &msghdr
) {
1886 if (cmsg
->cmsg_level
== SOL_SOCKET
&& cmsg
->cmsg_type
== SCM_RIGHTS
) {
1888 fd_array
= (int*) CMSG_DATA(cmsg
);
1889 n_fds
= (cmsg
->cmsg_len
- CMSG_LEN(0)) / sizeof(int);
1891 } else if (cmsg
->cmsg_level
== SOL_SOCKET
&&
1892 cmsg
->cmsg_type
== SCM_CREDENTIALS
&&
1893 cmsg
->cmsg_len
== CMSG_LEN(sizeof(struct ucred
))) {
1895 ucred
= (struct ucred
*) CMSG_DATA(cmsg
);
1902 r
= fdset_new_array(&fds
, fd_array
, n_fds
);
1904 close_many(fd_array
, n_fds
);
1910 if (!ucred
|| ucred
->pid
<= 0) {
1911 log_warning("Received notify message without valid credentials. Ignoring.");
1915 if ((size_t) n
>= sizeof(buf
) || (msghdr
.msg_flags
& MSG_TRUNC
)) {
1916 log_warning("Received notify message exceeded maximum size. Ignoring.");
1920 /* As extra safety check, let's make sure the string we get doesn't contain embedded NUL bytes. We permit one
1921 * trailing NUL byte in the message, but don't expect it. */
1922 if (n
> 1 && memchr(buf
, 0, n
-1)) {
1923 log_warning("Received notify message with embedded NUL bytes. Ignoring.");
1927 /* Make sure it's NUL-terminated. */
1930 /* Notify every unit that might be interested, but try
1931 * to avoid notifying the same one multiple times. */
1932 u1
= manager_get_unit_by_pid_cgroup(m
, ucred
->pid
);
1934 manager_invoke_notify_message(m
, u1
, ucred
->pid
, buf
, fds
);
1936 u2
= hashmap_get(m
->watch_pids1
, PID_TO_PTR(ucred
->pid
));
1938 manager_invoke_notify_message(m
, u2
, ucred
->pid
, buf
, fds
);
1940 u3
= hashmap_get(m
->watch_pids2
, PID_TO_PTR(ucred
->pid
));
1941 if (u3
&& u3
!= u2
&& u3
!= u1
)
1942 manager_invoke_notify_message(m
, u3
, ucred
->pid
, buf
, fds
);
1944 if (!u1
&& !u2
&& !u3
)
1945 log_warning("Cannot find unit for notify message of PID "PID_FMT
".", ucred
->pid
);
1947 if (fdset_size(fds
) > 0)
1948 log_warning("Got extra auxiliary fds with notification message, closing them.");
1953 static void invoke_sigchld_event(Manager
*m
, Unit
*u
, const siginfo_t
*si
) {
1960 sd_event_get_iteration(m
->event
, &iteration
);
1962 log_unit_debug(u
, "Child "PID_FMT
" belongs to %s", si
->si_pid
, u
->id
);
1964 unit_unwatch_pid(u
, si
->si_pid
);
1966 if (UNIT_VTABLE(u
)->sigchld_event
) {
1967 if (set_size(u
->pids
) <= 1 ||
1968 iteration
!= u
->sigchldgen
||
1969 unit_main_pid(u
) == si
->si_pid
||
1970 unit_control_pid(u
) == si
->si_pid
) {
1971 UNIT_VTABLE(u
)->sigchld_event(u
, si
->si_pid
, si
->si_code
, si
->si_status
);
1972 u
->sigchldgen
= iteration
;
1974 log_debug("%s already issued a sigchld this iteration %" PRIu64
", skipping. Pids still being watched %d", u
->id
, iteration
, set_size(u
->pids
));
1978 static int manager_dispatch_sigchld(Manager
*m
) {
1984 /* First we call waitd() for a PID and do not reap the
1985 * zombie. That way we can still access /proc/$PID for
1986 * it while it is a zombie. */
1987 if (waitid(P_ALL
, 0, &si
, WEXITED
|WNOHANG
|WNOWAIT
) < 0) {
1989 if (errno
== ECHILD
)
2001 if (IN_SET(si
.si_code
, CLD_EXITED
, CLD_KILLED
, CLD_DUMPED
)) {
2002 _cleanup_free_
char *name
= NULL
;
2005 get_process_comm(si
.si_pid
, &name
);
2007 log_debug("Child "PID_FMT
" (%s) died (code=%s, status=%i/%s)",
2008 si
.si_pid
, strna(name
),
2009 sigchld_code_to_string(si
.si_code
),
2011 strna(si
.si_code
== CLD_EXITED
2012 ? exit_status_to_string(si
.si_status
, EXIT_STATUS_FULL
)
2013 : signal_to_string(si
.si_status
)));
2015 /* And now figure out the unit this belongs
2016 * to, it might be multiple... */
2017 u1
= manager_get_unit_by_pid_cgroup(m
, si
.si_pid
);
2019 invoke_sigchld_event(m
, u1
, &si
);
2020 u2
= hashmap_get(m
->watch_pids1
, PID_TO_PTR(si
.si_pid
));
2022 invoke_sigchld_event(m
, u2
, &si
);
2023 u3
= hashmap_get(m
->watch_pids2
, PID_TO_PTR(si
.si_pid
));
2024 if (u3
&& u3
!= u2
&& u3
!= u1
)
2025 invoke_sigchld_event(m
, u3
, &si
);
2028 /* And now, we actually reap the zombie. */
2029 if (waitid(P_PID
, si
.si_pid
, &si
, WEXITED
) < 0) {
2040 static void manager_start_target(Manager
*m
, const char *name
, JobMode mode
) {
2041 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
2044 log_debug("Activating special unit %s", name
);
2046 r
= manager_add_job_by_name(m
, JOB_START
, name
, mode
, &error
, NULL
);
2048 log_error("Failed to enqueue %s job: %s", name
, bus_error_message(&error
, r
));
2051 static void manager_handle_ctrl_alt_del(Manager
*m
) {
2052 /* If the user presses C-A-D more than
2053 * 7 times within 2s, we reboot/shutdown immediately,
2054 * unless it was disabled in system.conf */
2056 if (ratelimit_test(&m
->ctrl_alt_del_ratelimit
) || m
->cad_burst_action
== EMERGENCY_ACTION_NONE
)
2057 manager_start_target(m
, SPECIAL_CTRL_ALT_DEL_TARGET
, JOB_REPLACE_IRREVERSIBLY
);
2059 emergency_action(m
, m
->cad_burst_action
, NULL
,
2060 "Ctrl-Alt-Del was pressed more than 7 times within 2s");
2063 static int manager_dispatch_signal_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
2064 Manager
*m
= userdata
;
2066 struct signalfd_siginfo sfsi
;
2067 bool sigchld
= false;
2071 assert(m
->signal_fd
== fd
);
2073 if (revents
!= EPOLLIN
) {
2074 log_warning("Got unexpected events from signal file descriptor.");
2079 n
= read(m
->signal_fd
, &sfsi
, sizeof(sfsi
));
2080 if (n
!= sizeof(sfsi
)) {
2082 log_warning("Truncated read from signal fd (%zu bytes)!", n
);
2086 if (IN_SET(errno
, EINTR
, EAGAIN
))
2089 /* We return an error here, which will kill this handler,
2090 * to avoid a busy loop on read error. */
2091 return log_error_errno(errno
, "Reading from signal fd failed: %m");
2094 log_received_signal(sfsi
.ssi_signo
== SIGCHLD
||
2095 (sfsi
.ssi_signo
== SIGTERM
&& MANAGER_IS_USER(m
))
2096 ? LOG_DEBUG
: LOG_INFO
,
2099 switch (sfsi
.ssi_signo
) {
2106 if (MANAGER_IS_SYSTEM(m
)) {
2107 /* This is for compatibility with the
2108 * original sysvinit */
2109 r
= verify_run_space_and_log("Refusing to reexecute");
2111 m
->exit_code
= MANAGER_REEXECUTE
;
2118 if (MANAGER_IS_SYSTEM(m
))
2119 manager_handle_ctrl_alt_del(m
);
2121 manager_start_target(m
, SPECIAL_EXIT_TARGET
,
2122 JOB_REPLACE_IRREVERSIBLY
);
2126 if (MANAGER_IS_SYSTEM(m
))
2127 manager_start_target(m
, SPECIAL_KBREQUEST_TARGET
, JOB_REPLACE
);
2129 /* This is a nop on non-init */
2133 if (MANAGER_IS_SYSTEM(m
))
2134 manager_start_target(m
, SPECIAL_SIGPWR_TARGET
, JOB_REPLACE
);
2136 /* This is a nop on non-init */
2142 u
= manager_get_unit(m
, SPECIAL_DBUS_SERVICE
);
2144 if (!u
|| UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u
))) {
2145 log_info("Trying to reconnect to bus...");
2149 if (!u
|| !UNIT_IS_ACTIVE_OR_ACTIVATING(unit_active_state(u
))) {
2150 log_info("Loading D-Bus service...");
2151 manager_start_target(m
, SPECIAL_DBUS_SERVICE
, JOB_REPLACE
);
2158 _cleanup_free_
char *dump
= NULL
;
2159 _cleanup_fclose_
FILE *f
= NULL
;
2162 f
= open_memstream(&dump
, &size
);
2164 log_warning_errno(errno
, "Failed to allocate memory stream: %m");
2168 manager_dump_units(m
, f
, "\t");
2169 manager_dump_jobs(m
, f
, "\t");
2171 r
= fflush_and_check(f
);
2173 log_warning_errno(r
, "Failed to write status stream: %m");
2177 log_dump(LOG_INFO
, dump
);
2182 r
= verify_run_space_and_log("Refusing to reload");
2184 m
->exit_code
= MANAGER_RELOAD
;
2189 /* Starting SIGRTMIN+0 */
2190 static const struct {
2193 } target_table
[] = {
2194 [0] = { SPECIAL_DEFAULT_TARGET
, JOB_ISOLATE
},
2195 [1] = { SPECIAL_RESCUE_TARGET
, JOB_ISOLATE
},
2196 [2] = { SPECIAL_EMERGENCY_TARGET
, JOB_ISOLATE
},
2197 [3] = { SPECIAL_HALT_TARGET
, JOB_REPLACE_IRREVERSIBLY
},
2198 [4] = { SPECIAL_POWEROFF_TARGET
, JOB_REPLACE_IRREVERSIBLY
},
2199 [5] = { SPECIAL_REBOOT_TARGET
, JOB_REPLACE_IRREVERSIBLY
},
2200 [6] = { SPECIAL_KEXEC_TARGET
, JOB_REPLACE_IRREVERSIBLY
}
2203 /* Starting SIGRTMIN+13, so that target halt and system halt are 10 apart */
2204 static const ManagerExitCode code_table
[] = {
2206 [1] = MANAGER_POWEROFF
,
2207 [2] = MANAGER_REBOOT
,
2211 if ((int) sfsi
.ssi_signo
>= SIGRTMIN
+0 &&
2212 (int) sfsi
.ssi_signo
< SIGRTMIN
+(int) ELEMENTSOF(target_table
)) {
2213 int idx
= (int) sfsi
.ssi_signo
- SIGRTMIN
;
2214 manager_start_target(m
, target_table
[idx
].target
,
2215 target_table
[idx
].mode
);
2219 if ((int) sfsi
.ssi_signo
>= SIGRTMIN
+13 &&
2220 (int) sfsi
.ssi_signo
< SIGRTMIN
+13+(int) ELEMENTSOF(code_table
)) {
2221 m
->exit_code
= code_table
[sfsi
.ssi_signo
- SIGRTMIN
- 13];
2225 switch (sfsi
.ssi_signo
- SIGRTMIN
) {
2228 manager_set_show_status(m
, SHOW_STATUS_YES
);
2232 manager_set_show_status(m
, SHOW_STATUS_NO
);
2236 log_set_max_level(LOG_DEBUG
);
2237 log_info("Setting log level to debug.");
2241 log_set_max_level(LOG_INFO
);
2242 log_info("Setting log level to info.");
2246 if (MANAGER_IS_USER(m
)) {
2247 m
->exit_code
= MANAGER_EXIT
;
2251 /* This is a nop on init */
2255 case 29: /* compatibility: used to be mapped to LOG_TARGET_SYSLOG_OR_KMSG */
2256 log_set_target(LOG_TARGET_JOURNAL_OR_KMSG
);
2257 log_notice("Setting log target to journal-or-kmsg.");
2261 log_set_target(LOG_TARGET_CONSOLE
);
2262 log_notice("Setting log target to console.");
2266 log_set_target(LOG_TARGET_KMSG
);
2267 log_notice("Setting log target to kmsg.");
2271 log_warning("Got unhandled signal <%s>.", signal_to_string(sfsi
.ssi_signo
));
2278 manager_dispatch_sigchld(m
);
2283 static int manager_dispatch_time_change_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
2284 Manager
*m
= userdata
;
2289 assert(m
->time_change_fd
== fd
);
2291 log_struct(LOG_DEBUG
,
2292 "MESSAGE_ID=" SD_MESSAGE_TIME_CHANGE_STR
,
2293 LOG_MESSAGE("Time has been changed"),
2296 /* Restart the watch */
2297 m
->time_change_event_source
= sd_event_source_unref(m
->time_change_event_source
);
2298 m
->time_change_fd
= safe_close(m
->time_change_fd
);
2300 manager_setup_time_change(m
);
2302 HASHMAP_FOREACH(u
, m
->units
, i
)
2303 if (UNIT_VTABLE(u
)->time_change
)
2304 UNIT_VTABLE(u
)->time_change(u
);
2309 static int manager_dispatch_idle_pipe_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
2310 Manager
*m
= userdata
;
2313 assert(m
->idle_pipe
[2] == fd
);
2315 m
->no_console_output
= m
->n_on_console
> 0;
2317 manager_close_idle_pipe(m
);
2322 static int manager_dispatch_jobs_in_progress(sd_event_source
*source
, usec_t usec
, void *userdata
) {
2323 Manager
*m
= userdata
;
2330 manager_print_jobs_in_progress(m
);
2332 next
= now(CLOCK_MONOTONIC
) + JOBS_IN_PROGRESS_PERIOD_USEC
;
2333 r
= sd_event_source_set_time(source
, next
);
2337 return sd_event_source_set_enabled(source
, SD_EVENT_ONESHOT
);
2340 int manager_loop(Manager
*m
) {
2343 RATELIMIT_DEFINE(rl
, 1*USEC_PER_SEC
, 50000);
2346 m
->exit_code
= MANAGER_OK
;
2348 /* Release the path cache */
2349 m
->unit_path_cache
= set_free_free(m
->unit_path_cache
);
2351 manager_check_finished(m
);
2353 /* There might still be some zombies hanging around from
2354 * before we were exec()'ed. Let's reap them. */
2355 r
= manager_dispatch_sigchld(m
);
2359 while (m
->exit_code
== MANAGER_OK
) {
2362 if (m
->runtime_watchdog
> 0 && m
->runtime_watchdog
!= USEC_INFINITY
&& MANAGER_IS_SYSTEM(m
))
2365 if (!ratelimit_test(&rl
)) {
2366 /* Yay, something is going seriously wrong, pause a little */
2367 log_warning("Looping too fast. Throttling execution a little.");
2371 if (manager_dispatch_load_queue(m
) > 0)
2374 if (manager_dispatch_gc_job_queue(m
) > 0)
2377 if (manager_dispatch_gc_unit_queue(m
) > 0)
2380 if (manager_dispatch_cleanup_queue(m
) > 0)
2383 if (manager_dispatch_cgroup_realize_queue(m
) > 0)
2386 if (manager_dispatch_dbus_queue(m
) > 0)
2389 /* Sleep for half the watchdog time */
2390 if (m
->runtime_watchdog
> 0 && m
->runtime_watchdog
!= USEC_INFINITY
&& MANAGER_IS_SYSTEM(m
)) {
2391 wait_usec
= m
->runtime_watchdog
/ 2;
2395 wait_usec
= USEC_INFINITY
;
2397 r
= sd_event_run(m
->event
, wait_usec
);
2399 return log_error_errno(r
, "Failed to run event loop: %m");
2402 return m
->exit_code
;
2405 int manager_load_unit_from_dbus_path(Manager
*m
, const char *s
, sd_bus_error
*e
, Unit
**_u
) {
2406 _cleanup_free_
char *n
= NULL
;
2407 sd_id128_t invocation_id
;
2415 r
= unit_name_from_dbus_path(s
, &n
);
2419 /* Permit addressing units by invocation ID: if the passed bus path is suffixed by a 128bit ID then we use it
2420 * as invocation ID. */
2421 r
= sd_id128_from_string(n
, &invocation_id
);
2423 u
= hashmap_get(m
->units_by_invocation_id
, &invocation_id
);
2429 return sd_bus_error_setf(e
, BUS_ERROR_NO_UNIT_FOR_INVOCATION_ID
, "No unit with the specified invocation ID " SD_ID128_FORMAT_STR
" known.", SD_ID128_FORMAT_VAL(invocation_id
));
2432 /* If this didn't work, we check if this is a unit name */
2433 if (!unit_name_is_valid(n
, UNIT_NAME_PLAIN
|UNIT_NAME_INSTANCE
))
2434 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
, "Unit name %s is neither a valid invocation ID nor unit name.", n
);
2436 r
= manager_load_unit(m
, n
, NULL
, e
, &u
);
2444 int manager_get_job_from_dbus_path(Manager
*m
, const char *s
, Job
**_j
) {
2454 p
= startswith(s
, "/org/freedesktop/systemd1/job/");
2458 r
= safe_atou(p
, &id
);
2462 j
= manager_get_job(m
, id
);
2471 void manager_send_unit_audit(Manager
*m
, Unit
*u
, int type
, bool success
) {
2474 _cleanup_free_
char *p
= NULL
;
2478 if (!MANAGER_IS_SYSTEM(m
))
2481 audit_fd
= get_audit_fd();
2485 /* Don't generate audit events if the service was already
2486 * started and we're just deserializing */
2487 if (MANAGER_IS_RELOADING(m
))
2490 if (u
->type
!= UNIT_SERVICE
)
2493 r
= unit_name_to_prefix_and_instance(u
->id
, &p
);
2495 log_error_errno(r
, "Failed to extract prefix and instance of unit name: %m");
2499 msg
= strjoina("unit=", p
);
2500 if (audit_log_user_comm_message(audit_fd
, type
, msg
, "systemd", NULL
, NULL
, NULL
, success
) < 0) {
2502 /* We aren't allowed to send audit messages?
2503 * Then let's not retry again. */
2506 log_warning_errno(errno
, "Failed to send audit message: %m");
2512 void manager_send_unit_plymouth(Manager
*m
, Unit
*u
) {
2513 static const union sockaddr_union sa
= PLYMOUTH_SOCKET
;
2514 _cleanup_free_
char *message
= NULL
;
2515 _cleanup_close_
int fd
= -1;
2518 /* Don't generate plymouth events if the service was already
2519 * started and we're just deserializing */
2520 if (MANAGER_IS_RELOADING(m
))
2523 if (!MANAGER_IS_SYSTEM(m
))
2526 if (detect_container() > 0)
2529 if (!IN_SET(u
->type
, UNIT_SERVICE
, UNIT_MOUNT
, UNIT_SWAP
))
2532 /* We set SOCK_NONBLOCK here so that we rather drop the
2533 * message then wait for plymouth */
2534 fd
= socket(AF_UNIX
, SOCK_STREAM
|SOCK_CLOEXEC
|SOCK_NONBLOCK
, 0);
2536 log_error_errno(errno
, "socket() failed: %m");
2540 if (connect(fd
, &sa
.sa
, SOCKADDR_UN_LEN(sa
.un
)) < 0) {
2542 if (!IN_SET(errno
, EPIPE
, EAGAIN
, ENOENT
, ECONNREFUSED
, ECONNRESET
, ECONNABORTED
))
2543 log_error_errno(errno
, "connect() failed: %m");
2547 if (asprintf(&message
, "U\002%c%s%n", (int) (strlen(u
->id
) + 1), u
->id
, &n
) < 0) {
2553 if (write(fd
, message
, n
+ 1) != n
+ 1)
2554 if (!IN_SET(errno
, EPIPE
, EAGAIN
, ENOENT
, ECONNREFUSED
, ECONNRESET
, ECONNABORTED
))
2555 log_error_errno(errno
, "Failed to write Plymouth message: %m");
2558 int manager_open_serialization(Manager
*m
, FILE **_f
) {
2564 fd
= open_serialization_fd("systemd-state");
2568 f
= fdopen(fd
, "w+");
2578 int manager_serialize(Manager
*m
, FILE *f
, FDSet
*fds
, bool switching_root
) {
2590 fprintf(f
, "current-job-id=%"PRIu32
"\n", m
->current_job_id
);
2591 fprintf(f
, "n-installed-jobs=%u\n", m
->n_installed_jobs
);
2592 fprintf(f
, "n-failed-jobs=%u\n", m
->n_failed_jobs
);
2593 fprintf(f
, "taint-usr=%s\n", yes_no(m
->taint_usr
));
2594 fprintf(f
, "ready-sent=%s\n", yes_no(m
->ready_sent
));
2596 dual_timestamp_serialize(f
, "firmware-timestamp", &m
->firmware_timestamp
);
2597 dual_timestamp_serialize(f
, "loader-timestamp", &m
->loader_timestamp
);
2598 dual_timestamp_serialize(f
, "kernel-timestamp", &m
->kernel_timestamp
);
2599 dual_timestamp_serialize(f
, "initrd-timestamp", &m
->initrd_timestamp
);
2602 dual_timestamp_serialize(f
, "userspace-timestamp", &m
->userspace_timestamp
);
2603 dual_timestamp_serialize(f
, "finish-timestamp", &m
->finish_timestamp
);
2604 dual_timestamp_serialize(f
, "security-start-timestamp", &m
->security_start_timestamp
);
2605 dual_timestamp_serialize(f
, "security-finish-timestamp", &m
->security_finish_timestamp
);
2606 dual_timestamp_serialize(f
, "generators-start-timestamp", &m
->generators_start_timestamp
);
2607 dual_timestamp_serialize(f
, "generators-finish-timestamp", &m
->generators_finish_timestamp
);
2608 dual_timestamp_serialize(f
, "units-load-start-timestamp", &m
->units_load_start_timestamp
);
2609 dual_timestamp_serialize(f
, "units-load-finish-timestamp", &m
->units_load_finish_timestamp
);
2612 if (!switching_root
)
2613 (void) serialize_environment(f
, m
->environment
);
2615 if (m
->notify_fd
>= 0) {
2618 copy
= fdset_put_dup(fds
, m
->notify_fd
);
2622 fprintf(f
, "notify-fd=%i\n", copy
);
2623 fprintf(f
, "notify-socket=%s\n", m
->notify_socket
);
2626 if (m
->cgroups_agent_fd
>= 0) {
2629 copy
= fdset_put_dup(fds
, m
->cgroups_agent_fd
);
2633 fprintf(f
, "cgroups-agent-fd=%i\n", copy
);
2636 if (m
->user_lookup_fds
[0] >= 0) {
2639 copy0
= fdset_put_dup(fds
, m
->user_lookup_fds
[0]);
2643 copy1
= fdset_put_dup(fds
, m
->user_lookup_fds
[1]);
2647 fprintf(f
, "user-lookup=%i %i\n", copy0
, copy1
);
2650 bus_track_serialize(m
->subscribed
, f
, "subscribed");
2652 r
= dynamic_user_serialize(m
, f
, fds
);
2656 manager_serialize_uid_refs(m
, f
);
2657 manager_serialize_gid_refs(m
, f
);
2659 fputc_unlocked('\n', f
);
2661 HASHMAP_FOREACH_KEY(u
, t
, m
->units
, i
) {
2666 fputs_unlocked(u
->id
, f
);
2667 fputc_unlocked('\n', f
);
2669 r
= unit_serialize(u
, f
, fds
, !switching_root
);
2676 assert(m
->n_reloading
> 0);
2682 r
= bus_fdset_add_all(m
, fds
);
2689 int manager_deserialize(Manager
*m
, FILE *f
, FDSet
*fds
) {
2695 log_debug("Deserializing state...");
2700 char line
[LINE_MAX
];
2701 const char *val
, *l
;
2703 if (!fgets(line
, sizeof(line
), f
)) {
2718 if ((val
= startswith(l
, "current-job-id="))) {
2721 if (safe_atou32(val
, &id
) < 0)
2722 log_notice("Failed to parse current job id value %s", val
);
2724 m
->current_job_id
= MAX(m
->current_job_id
, id
);
2726 } else if ((val
= startswith(l
, "n-installed-jobs="))) {
2729 if (safe_atou32(val
, &n
) < 0)
2730 log_notice("Failed to parse installed jobs counter %s", val
);
2732 m
->n_installed_jobs
+= n
;
2734 } else if ((val
= startswith(l
, "n-failed-jobs="))) {
2737 if (safe_atou32(val
, &n
) < 0)
2738 log_notice("Failed to parse failed jobs counter %s", val
);
2740 m
->n_failed_jobs
+= n
;
2742 } else if ((val
= startswith(l
, "taint-usr="))) {
2745 b
= parse_boolean(val
);
2747 log_notice("Failed to parse taint /usr flag %s", val
);
2749 m
->taint_usr
= m
->taint_usr
|| b
;
2751 } else if ((val
= startswith(l
, "ready-sent="))) {
2754 b
= parse_boolean(val
);
2756 log_notice("Failed to parse ready-sent flag %s", val
);
2758 m
->ready_sent
= m
->ready_sent
|| b
;
2760 } else if ((val
= startswith(l
, "firmware-timestamp=")))
2761 dual_timestamp_deserialize(val
, &m
->firmware_timestamp
);
2762 else if ((val
= startswith(l
, "loader-timestamp=")))
2763 dual_timestamp_deserialize(val
, &m
->loader_timestamp
);
2764 else if ((val
= startswith(l
, "kernel-timestamp=")))
2765 dual_timestamp_deserialize(val
, &m
->kernel_timestamp
);
2766 else if ((val
= startswith(l
, "initrd-timestamp=")))
2767 dual_timestamp_deserialize(val
, &m
->initrd_timestamp
);
2768 else if ((val
= startswith(l
, "userspace-timestamp=")))
2769 dual_timestamp_deserialize(val
, &m
->userspace_timestamp
);
2770 else if ((val
= startswith(l
, "finish-timestamp=")))
2771 dual_timestamp_deserialize(val
, &m
->finish_timestamp
);
2772 else if ((val
= startswith(l
, "security-start-timestamp=")))
2773 dual_timestamp_deserialize(val
, &m
->security_start_timestamp
);
2774 else if ((val
= startswith(l
, "security-finish-timestamp=")))
2775 dual_timestamp_deserialize(val
, &m
->security_finish_timestamp
);
2776 else if ((val
= startswith(l
, "generators-start-timestamp=")))
2777 dual_timestamp_deserialize(val
, &m
->generators_start_timestamp
);
2778 else if ((val
= startswith(l
, "generators-finish-timestamp=")))
2779 dual_timestamp_deserialize(val
, &m
->generators_finish_timestamp
);
2780 else if ((val
= startswith(l
, "units-load-start-timestamp=")))
2781 dual_timestamp_deserialize(val
, &m
->units_load_start_timestamp
);
2782 else if ((val
= startswith(l
, "units-load-finish-timestamp=")))
2783 dual_timestamp_deserialize(val
, &m
->units_load_finish_timestamp
);
2784 else if (startswith(l
, "env=")) {
2785 r
= deserialize_environment(&m
->environment
, l
);
2789 log_notice_errno(r
, "Failed to parse environment entry: \"%s\": %m", l
);
2791 } else if ((val
= startswith(l
, "notify-fd="))) {
2794 if (safe_atoi(val
, &fd
) < 0 || fd
< 0 || !fdset_contains(fds
, fd
))
2795 log_notice("Failed to parse notify fd: \"%s\"", val
);
2797 m
->notify_event_source
= sd_event_source_unref(m
->notify_event_source
);
2798 safe_close(m
->notify_fd
);
2799 m
->notify_fd
= fdset_remove(fds
, fd
);
2802 } else if ((val
= startswith(l
, "notify-socket="))) {
2811 free(m
->notify_socket
);
2812 m
->notify_socket
= n
;
2814 } else if ((val
= startswith(l
, "cgroups-agent-fd="))) {
2817 if (safe_atoi(val
, &fd
) < 0 || fd
< 0 || !fdset_contains(fds
, fd
))
2818 log_notice("Failed to parse cgroups agent fd: %s", val
);
2820 m
->cgroups_agent_event_source
= sd_event_source_unref(m
->cgroups_agent_event_source
);
2821 safe_close(m
->cgroups_agent_fd
);
2822 m
->cgroups_agent_fd
= fdset_remove(fds
, fd
);
2825 } else if ((val
= startswith(l
, "user-lookup="))) {
2828 if (sscanf(val
, "%i %i", &fd0
, &fd1
) != 2 || fd0
< 0 || fd1
< 0 || fd0
== fd1
|| !fdset_contains(fds
, fd0
) || !fdset_contains(fds
, fd1
))
2829 log_notice("Failed to parse user lookup fd: %s", val
);
2831 m
->user_lookup_event_source
= sd_event_source_unref(m
->user_lookup_event_source
);
2832 safe_close_pair(m
->user_lookup_fds
);
2833 m
->user_lookup_fds
[0] = fdset_remove(fds
, fd0
);
2834 m
->user_lookup_fds
[1] = fdset_remove(fds
, fd1
);
2837 } else if ((val
= startswith(l
, "dynamic-user=")))
2838 dynamic_user_deserialize_one(m
, val
, fds
);
2839 else if ((val
= startswith(l
, "destroy-ipc-uid=")))
2840 manager_deserialize_uid_refs_one(m
, val
);
2841 else if ((val
= startswith(l
, "destroy-ipc-gid=")))
2842 manager_deserialize_gid_refs_one(m
, val
);
2843 else if ((val
= startswith(l
, "subscribed="))) {
2845 if (strv_extend(&m
->deserialized_subscribed
, val
) < 0)
2848 } else if (!startswith(l
, "kdbus-fd=")) /* ignore this one */
2849 log_notice("Unknown serialization item '%s'", l
);
2854 char name
[UNIT_NAME_MAX
+2];
2855 const char* unit_name
;
2858 if (!fgets(name
, sizeof(name
), f
)) {
2868 unit_name
= strstrip(name
);
2870 r
= manager_load_unit(m
, unit_name
, NULL
, NULL
, &u
);
2872 log_notice_errno(r
, "Failed to load unit \"%s\", skipping deserialization: %m", unit_name
);
2875 unit_deserialize_skip(f
);
2879 r
= unit_deserialize(u
, f
, fds
);
2881 log_notice_errno(r
, "Failed to deserialize unit \"%s\": %m", unit_name
);
2891 assert(m
->n_reloading
> 0);
2897 int manager_reload(Manager
*m
) {
2899 _cleanup_fclose_
FILE *f
= NULL
;
2900 _cleanup_fdset_free_ FDSet
*fds
= NULL
;
2904 r
= manager_open_serialization(m
, &f
);
2909 bus_manager_send_reloading(m
, true);
2917 r
= manager_serialize(m
, f
, fds
, false);
2923 if (fseeko(f
, 0, SEEK_SET
) < 0) {
2928 /* From here on there is no way back. */
2929 manager_clear_jobs_and_units(m
);
2930 lookup_paths_flush_generator(&m
->lookup_paths
);
2931 lookup_paths_free(&m
->lookup_paths
);
2932 dynamic_user_vacuum(m
, false);
2933 m
->uid_refs
= hashmap_free(m
->uid_refs
);
2934 m
->gid_refs
= hashmap_free(m
->gid_refs
);
2936 q
= lookup_paths_init(&m
->lookup_paths
, m
->unit_file_scope
, 0, NULL
);
2937 if (q
< 0 && r
>= 0)
2940 q
= manager_run_environment_generators(m
);
2941 if (q
< 0 && r
>= 0)
2944 /* Find new unit paths */
2945 q
= manager_run_generators(m
);
2946 if (q
< 0 && r
>= 0)
2949 lookup_paths_reduce(&m
->lookup_paths
);
2950 manager_build_unit_path_cache(m
);
2952 /* First, enumerate what we can from all config files */
2953 manager_enumerate(m
);
2955 /* Second, deserialize our stored data */
2956 q
= manager_deserialize(m
, f
, fds
);
2958 log_error_errno(q
, "Deserialization failed: %m");
2967 /* Re-register notify_fd as event source */
2968 q
= manager_setup_notify(m
);
2969 if (q
< 0 && r
>= 0)
2972 q
= manager_setup_cgroups_agent(m
);
2973 if (q
< 0 && r
>= 0)
2976 q
= manager_setup_user_lookup_fd(m
);
2977 if (q
< 0 && r
>= 0)
2980 /* Third, fire things up! */
2981 manager_coldplug(m
);
2983 /* Release any dynamic users no longer referenced */
2984 dynamic_user_vacuum(m
, true);
2986 /* Release any references to UIDs/GIDs no longer referenced, and destroy any IPC owned by them */
2987 manager_vacuum_uid_refs(m
);
2988 manager_vacuum_gid_refs(m
);
2990 /* Sync current state of bus names with our set of listening units */
2992 manager_sync_bus_names(m
, m
->api_bus
);
2994 assert(m
->n_reloading
> 0);
2997 m
->send_reloading_done
= true;
3002 void manager_reset_failed(Manager
*m
) {
3008 HASHMAP_FOREACH(u
, m
->units
, i
)
3009 unit_reset_failed(u
);
3012 bool manager_unit_inactive_or_pending(Manager
*m
, const char *name
) {
3018 /* Returns true if the unit is inactive or going down */
3019 u
= manager_get_unit(m
, name
);
3023 return unit_inactive_or_pending(u
);
3026 static void manager_notify_finished(Manager
*m
) {
3027 char userspace
[FORMAT_TIMESPAN_MAX
], initrd
[FORMAT_TIMESPAN_MAX
], kernel
[FORMAT_TIMESPAN_MAX
], sum
[FORMAT_TIMESPAN_MAX
];
3028 usec_t firmware_usec
, loader_usec
, kernel_usec
, initrd_usec
, userspace_usec
, total_usec
;
3030 if (m
->test_run_flags
)
3033 if (MANAGER_IS_SYSTEM(m
) && detect_container() <= 0) {
3035 /* Note that m->kernel_usec.monotonic is always at 0,
3036 * and m->firmware_usec.monotonic and
3037 * m->loader_usec.monotonic should be considered
3038 * negative values. */
3040 firmware_usec
= m
->firmware_timestamp
.monotonic
- m
->loader_timestamp
.monotonic
;
3041 loader_usec
= m
->loader_timestamp
.monotonic
- m
->kernel_timestamp
.monotonic
;
3042 userspace_usec
= m
->finish_timestamp
.monotonic
- m
->userspace_timestamp
.monotonic
;
3043 total_usec
= m
->firmware_timestamp
.monotonic
+ m
->finish_timestamp
.monotonic
;
3045 if (dual_timestamp_is_set(&m
->initrd_timestamp
)) {
3047 kernel_usec
= m
->initrd_timestamp
.monotonic
- m
->kernel_timestamp
.monotonic
;
3048 initrd_usec
= m
->userspace_timestamp
.monotonic
- m
->initrd_timestamp
.monotonic
;
3050 log_struct(LOG_INFO
,
3051 "MESSAGE_ID=" SD_MESSAGE_STARTUP_FINISHED_STR
,
3052 "KERNEL_USEC="USEC_FMT
, kernel_usec
,
3053 "INITRD_USEC="USEC_FMT
, initrd_usec
,
3054 "USERSPACE_USEC="USEC_FMT
, userspace_usec
,
3055 LOG_MESSAGE("Startup finished in %s (kernel) + %s (initrd) + %s (userspace) = %s.",
3056 format_timespan(kernel
, sizeof(kernel
), kernel_usec
, USEC_PER_MSEC
),
3057 format_timespan(initrd
, sizeof(initrd
), initrd_usec
, USEC_PER_MSEC
),
3058 format_timespan(userspace
, sizeof(userspace
), userspace_usec
, USEC_PER_MSEC
),
3059 format_timespan(sum
, sizeof(sum
), total_usec
, USEC_PER_MSEC
)),
3062 kernel_usec
= m
->userspace_timestamp
.monotonic
- m
->kernel_timestamp
.monotonic
;
3065 log_struct(LOG_INFO
,
3066 "MESSAGE_ID=" SD_MESSAGE_STARTUP_FINISHED_STR
,
3067 "KERNEL_USEC="USEC_FMT
, kernel_usec
,
3068 "USERSPACE_USEC="USEC_FMT
, userspace_usec
,
3069 LOG_MESSAGE("Startup finished in %s (kernel) + %s (userspace) = %s.",
3070 format_timespan(kernel
, sizeof(kernel
), kernel_usec
, USEC_PER_MSEC
),
3071 format_timespan(userspace
, sizeof(userspace
), userspace_usec
, USEC_PER_MSEC
),
3072 format_timespan(sum
, sizeof(sum
), total_usec
, USEC_PER_MSEC
)),
3076 firmware_usec
= loader_usec
= initrd_usec
= kernel_usec
= 0;
3077 total_usec
= userspace_usec
= m
->finish_timestamp
.monotonic
- m
->userspace_timestamp
.monotonic
;
3079 log_struct(LOG_INFO
,
3080 "MESSAGE_ID=" SD_MESSAGE_USER_STARTUP_FINISHED_STR
,
3081 "USERSPACE_USEC="USEC_FMT
, userspace_usec
,
3082 LOG_MESSAGE("Startup finished in %s.",
3083 format_timespan(sum
, sizeof(sum
), total_usec
, USEC_PER_MSEC
)),
3087 bus_manager_send_finished(m
, firmware_usec
, loader_usec
, kernel_usec
, initrd_usec
, userspace_usec
, total_usec
);
3090 m
->ready_sent
? "STATUS=Startup finished in %s."
3092 "STATUS=Startup finished in %s.",
3093 format_timespan(sum
, sizeof(sum
), total_usec
, USEC_PER_MSEC
));
3094 m
->ready_sent
= true;
3097 void manager_check_finished(Manager
*m
) {
3100 if (MANAGER_IS_RELOADING(m
))
3103 /* Verify that we are actually running currently. Initially
3104 * the exit code is set to invalid, and during operation it is
3105 * then set to MANAGER_OK */
3106 if (m
->exit_code
!= MANAGER_OK
)
3109 /* For user managers, send out READY=1 as soon as we reach basic.target */
3110 if (MANAGER_IS_USER(m
) && !m
->ready_sent
) {
3113 u
= manager_get_unit(m
, SPECIAL_BASIC_TARGET
);
3117 "STATUS=Reached " SPECIAL_BASIC_TARGET
".");
3118 m
->ready_sent
= true;
3122 if (hashmap_size(m
->jobs
) > 0) {
3123 if (m
->jobs_in_progress_event_source
)
3124 /* Ignore any failure, this is only for feedback */
3125 (void) sd_event_source_set_time(m
->jobs_in_progress_event_source
, now(CLOCK_MONOTONIC
) + JOBS_IN_PROGRESS_WAIT_USEC
);
3130 manager_flip_auto_status(m
, false);
3132 /* Notify Type=idle units that we are done now */
3133 manager_close_idle_pipe(m
);
3135 /* Turn off confirm spawn now */
3136 m
->confirm_spawn
= NULL
;
3138 /* No need to update ask password status when we're going non-interactive */
3139 manager_close_ask_password(m
);
3141 /* This is no longer the first boot */
3142 manager_set_first_boot(m
, false);
3144 if (dual_timestamp_is_set(&m
->finish_timestamp
))
3147 dual_timestamp_get(&m
->finish_timestamp
);
3149 manager_notify_finished(m
);
3151 manager_invalidate_startup_units(m
);
3154 static bool generator_path_any(const char* const* paths
) {
3158 /* Optimize by skipping the whole process by not creating output directories
3159 * if no generators are found. */
3160 STRV_FOREACH(path
, (char**) paths
)
3161 if (access(*path
, F_OK
) == 0)
3163 else if (errno
!= ENOENT
)
3164 log_warning_errno(errno
, "Failed to open generator directory %s: %m", *path
);
3169 static const char* system_env_generator_binary_paths
[] = {
3170 "/run/systemd/system-environment-generators",
3171 "/etc/systemd/system-environment-generators",
3172 "/usr/local/lib/systemd/system-environment-generators",
3173 SYSTEM_ENV_GENERATOR_PATH
,
3177 static const char* user_env_generator_binary_paths
[] = {
3178 "/run/systemd/user-environment-generators",
3179 "/etc/systemd/user-environment-generators",
3180 "/usr/local/lib/systemd/user-environment-generators",
3181 USER_ENV_GENERATOR_PATH
,
3185 static int manager_run_environment_generators(Manager
*m
) {
3186 char **tmp
= NULL
; /* this is only used in the forked process, no cleanup here */
3188 void* args
[] = {&tmp
, &tmp
, &m
->environment
};
3190 if (m
->test_run_flags
&& !(m
->test_run_flags
& MANAGER_TEST_RUN_ENV_GENERATORS
))
3193 paths
= MANAGER_IS_SYSTEM(m
) ? system_env_generator_binary_paths
: user_env_generator_binary_paths
;
3195 if (!generator_path_any(paths
))
3198 return execute_directories(paths
, DEFAULT_TIMEOUT_USEC
, gather_environment
, args
, NULL
);
3201 static int manager_run_generators(Manager
*m
) {
3202 _cleanup_strv_free_
char **paths
= NULL
;
3203 const char *argv
[5];
3208 if (m
->test_run_flags
&& !(m
->test_run_flags
& MANAGER_TEST_RUN_GENERATORS
))
3211 paths
= generator_binary_paths(m
->unit_file_scope
);
3215 if (!generator_path_any((const char* const*) paths
))
3218 r
= lookup_paths_mkdir_generator(&m
->lookup_paths
);
3222 argv
[0] = NULL
; /* Leave this empty, execute_directory() will fill something in */
3223 argv
[1] = m
->lookup_paths
.generator
;
3224 argv
[2] = m
->lookup_paths
.generator_early
;
3225 argv
[3] = m
->lookup_paths
.generator_late
;
3228 RUN_WITH_UMASK(0022)
3229 execute_directories((const char* const*) paths
, DEFAULT_TIMEOUT_USEC
,
3230 NULL
, NULL
, (char**) argv
);
3233 lookup_paths_trim_generator(&m
->lookup_paths
);
3237 int manager_environment_add(Manager
*m
, char **minus
, char **plus
) {
3238 char **a
= NULL
, **b
= NULL
, **l
;
3243 if (!strv_isempty(minus
)) {
3244 a
= strv_env_delete(l
, 1, minus
);
3251 if (!strv_isempty(plus
)) {
3252 b
= strv_env_merge(2, l
, plus
);
3261 if (m
->environment
!= l
)
3262 strv_free(m
->environment
);
3269 manager_clean_environment(m
);
3270 strv_sort(m
->environment
);
3275 int manager_set_default_rlimits(Manager
*m
, struct rlimit
**default_rlimit
) {
3280 for (i
= 0; i
< _RLIMIT_MAX
; i
++) {
3281 m
->rlimit
[i
] = mfree(m
->rlimit
[i
]);
3283 if (!default_rlimit
[i
])
3286 m
->rlimit
[i
] = newdup(struct rlimit
, default_rlimit
[i
], 1);
3294 void manager_recheck_journal(Manager
*m
) {
3299 if (!MANAGER_IS_SYSTEM(m
))
3302 u
= manager_get_unit(m
, SPECIAL_JOURNALD_SOCKET
);
3303 if (u
&& SOCKET(u
)->state
!= SOCKET_RUNNING
) {
3304 log_close_journal();
3308 u
= manager_get_unit(m
, SPECIAL_JOURNALD_SERVICE
);
3309 if (u
&& SERVICE(u
)->state
!= SERVICE_RUNNING
) {
3310 log_close_journal();
3314 /* Hmm, OK, so the socket is fully up and the service is up
3315 * too, then let's make use of the thing. */
3319 void manager_set_show_status(Manager
*m
, ShowStatus mode
) {
3321 assert(IN_SET(mode
, SHOW_STATUS_AUTO
, SHOW_STATUS_NO
, SHOW_STATUS_YES
, SHOW_STATUS_TEMPORARY
));
3323 if (!MANAGER_IS_SYSTEM(m
))
3326 if (m
->show_status
!= mode
)
3327 log_debug("%s showing of status.",
3328 mode
== SHOW_STATUS_NO
? "Disabling" : "Enabling");
3329 m
->show_status
= mode
;
3332 (void) touch("/run/systemd/show-status");
3334 (void) unlink("/run/systemd/show-status");
3337 static bool manager_get_show_status(Manager
*m
, StatusType type
) {
3340 if (!MANAGER_IS_SYSTEM(m
))
3343 if (m
->no_console_output
)
3346 if (!IN_SET(manager_state(m
), MANAGER_INITIALIZING
, MANAGER_STARTING
, MANAGER_STOPPING
))
3349 /* If we cannot find out the status properly, just proceed. */
3350 if (type
!= STATUS_TYPE_EMERGENCY
&& manager_check_ask_password(m
) > 0)
3353 if (m
->show_status
> 0)
3359 const char *manager_get_confirm_spawn(Manager
*m
) {
3360 static int last_errno
= 0;
3361 const char *vc
= m
->confirm_spawn
;
3365 /* Here's the deal: we want to test the validity of the console but don't want
3366 * PID1 to go through the whole console process which might block. But we also
3367 * want to warn the user only once if something is wrong with the console so we
3368 * cannot do the sanity checks after spawning our children. So here we simply do
3369 * really basic tests to hopefully trap common errors.
3371 * If the console suddenly disappear at the time our children will really it
3372 * then they will simply fail to acquire it and a positive answer will be
3373 * assumed. New children will fallback to /dev/console though.
3375 * Note: TTYs are devices that can come and go any time, and frequently aren't
3376 * available yet during early boot (consider a USB rs232 dongle...). If for any
3377 * reason the configured console is not ready, we fallback to the default
3380 if (!vc
|| path_equal(vc
, "/dev/console"))
3387 if (!S_ISCHR(st
.st_mode
)) {
3395 if (last_errno
!= errno
) {
3397 log_warning_errno(errno
, "Failed to open %s: %m, using default console", vc
);
3399 return "/dev/console";
3402 void manager_set_first_boot(Manager
*m
, bool b
) {
3405 if (!MANAGER_IS_SYSTEM(m
))
3408 if (m
->first_boot
!= (int) b
) {
3410 (void) touch("/run/systemd/first-boot");
3412 (void) unlink("/run/systemd/first-boot");
3418 void manager_disable_confirm_spawn(void) {
3419 (void) touch("/run/systemd/confirm_spawn_disabled");
3422 bool manager_is_confirm_spawn_disabled(Manager
*m
) {
3423 if (!m
->confirm_spawn
)
3426 return access("/run/systemd/confirm_spawn_disabled", F_OK
) >= 0;
3429 void manager_status_printf(Manager
*m
, StatusType type
, const char *status
, const char *format
, ...) {
3432 /* If m is NULL, assume we're after shutdown and let the messages through. */
3434 if (m
&& !manager_get_show_status(m
, type
))
3437 /* XXX We should totally drop the check for ephemeral here
3438 * and thus effectively make 'Type=idle' pointless. */
3439 if (type
== STATUS_TYPE_EPHEMERAL
&& m
&& m
->n_on_console
> 0)
3442 va_start(ap
, format
);
3443 status_vprintf(status
, true, type
== STATUS_TYPE_EPHEMERAL
, format
, ap
);
3447 Set
*manager_get_units_requiring_mounts_for(Manager
*m
, const char *path
) {
3448 char p
[strlen(path
)+1];
3454 path_kill_slashes(p
);
3456 return hashmap_get(m
->units_requiring_mounts_for
, streq(p
, "/") ? "" : p
);
3459 void manager_set_exec_params(Manager
*m
, ExecParameters
*p
) {
3463 p
->environment
= m
->environment
;
3464 p
->confirm_spawn
= manager_get_confirm_spawn(m
);
3465 p
->cgroup_supported
= m
->cgroup_supported
;
3466 p
->prefix
= m
->prefix
;
3468 SET_FLAG(p
->flags
, EXEC_PASS_LOG_UNIT
|EXEC_CHOWN_DIRECTORIES
, MANAGER_IS_SYSTEM(m
));
3471 int manager_update_failed_units(Manager
*m
, Unit
*u
, bool failed
) {
3476 assert(u
->manager
== m
);
3478 size
= set_size(m
->failed_units
);
3481 r
= set_ensure_allocated(&m
->failed_units
, NULL
);
3485 if (set_put(m
->failed_units
, u
) < 0)
3488 (void) set_remove(m
->failed_units
, u
);
3490 if (set_size(m
->failed_units
) != size
)
3491 bus_manager_send_change_signal(m
);
3496 ManagerState
manager_state(Manager
*m
) {
3501 /* Did we ever finish booting? If not then we are still starting up */
3502 if (!dual_timestamp_is_set(&m
->finish_timestamp
)) {
3504 u
= manager_get_unit(m
, SPECIAL_BASIC_TARGET
);
3505 if (!u
|| !UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u
)))
3506 return MANAGER_INITIALIZING
;
3508 return MANAGER_STARTING
;
3511 /* Is the special shutdown target queued? If so, we are in shutdown state */
3512 u
= manager_get_unit(m
, SPECIAL_SHUTDOWN_TARGET
);
3513 if (u
&& u
->job
&& IN_SET(u
->job
->type
, JOB_START
, JOB_RESTART
, JOB_RELOAD_OR_START
))
3514 return MANAGER_STOPPING
;
3516 /* Are the rescue or emergency targets active or queued? If so we are in maintenance state */
3517 u
= manager_get_unit(m
, SPECIAL_RESCUE_TARGET
);
3518 if (u
&& (UNIT_IS_ACTIVE_OR_ACTIVATING(unit_active_state(u
)) ||
3519 (u
->job
&& IN_SET(u
->job
->type
, JOB_START
, JOB_RESTART
, JOB_RELOAD_OR_START
))))
3520 return MANAGER_MAINTENANCE
;
3522 u
= manager_get_unit(m
, SPECIAL_EMERGENCY_TARGET
);
3523 if (u
&& (UNIT_IS_ACTIVE_OR_ACTIVATING(unit_active_state(u
)) ||
3524 (u
->job
&& IN_SET(u
->job
->type
, JOB_START
, JOB_RESTART
, JOB_RELOAD_OR_START
))))
3525 return MANAGER_MAINTENANCE
;
3527 /* Are there any failed units? If so, we are in degraded mode */
3528 if (set_size(m
->failed_units
) > 0)
3529 return MANAGER_DEGRADED
;
3531 return MANAGER_RUNNING
;
3534 #define DESTROY_IPC_FLAG (UINT32_C(1) << 31)
3536 static void manager_unref_uid_internal(
3541 int (*_clean_ipc
)(uid_t uid
)) {
3547 assert(uid_is_valid(uid
));
3550 /* A generic implementation, covering both manager_unref_uid() and manager_unref_gid(), under the assumption
3551 * that uid_t and gid_t are actually defined the same way, with the same validity rules.
3553 * We store a hashmap where the UID/GID is they key and the value is a 32bit reference counter, whose highest
3554 * bit is used as flag for marking UIDs/GIDs whose IPC objects to remove when the last reference to the UID/GID
3555 * is dropped. The flag is set to on, once at least one reference from a unit where RemoveIPC= is set is added
3556 * on a UID/GID. It is reset when the UID's/GID's reference counter drops to 0 again. */
3558 assert_cc(sizeof(uid_t
) == sizeof(gid_t
));
3559 assert_cc(UID_INVALID
== (uid_t
) GID_INVALID
);
3561 if (uid
== 0) /* We don't keep track of root, and will never destroy it */
3564 c
= PTR_TO_UINT32(hashmap_get(*uid_refs
, UID_TO_PTR(uid
)));
3566 n
= c
& ~DESTROY_IPC_FLAG
;
3570 if (destroy_now
&& n
== 0) {
3571 hashmap_remove(*uid_refs
, UID_TO_PTR(uid
));
3573 if (c
& DESTROY_IPC_FLAG
) {
3574 log_debug("%s " UID_FMT
" is no longer referenced, cleaning up its IPC.",
3575 _clean_ipc
== clean_ipc_by_uid
? "UID" : "GID",
3577 (void) _clean_ipc(uid
);
3580 c
= n
| (c
& DESTROY_IPC_FLAG
);
3581 assert_se(hashmap_update(*uid_refs
, UID_TO_PTR(uid
), UINT32_TO_PTR(c
)) >= 0);
3585 void manager_unref_uid(Manager
*m
, uid_t uid
, bool destroy_now
) {
3586 manager_unref_uid_internal(m
, &m
->uid_refs
, uid
, destroy_now
, clean_ipc_by_uid
);
3589 void manager_unref_gid(Manager
*m
, gid_t gid
, bool destroy_now
) {
3590 manager_unref_uid_internal(m
, &m
->gid_refs
, (uid_t
) gid
, destroy_now
, clean_ipc_by_gid
);
3593 static int manager_ref_uid_internal(
3604 assert(uid_is_valid(uid
));
3606 /* A generic implementation, covering both manager_ref_uid() and manager_ref_gid(), under the assumption
3607 * that uid_t and gid_t are actually defined the same way, with the same validity rules. */
3609 assert_cc(sizeof(uid_t
) == sizeof(gid_t
));
3610 assert_cc(UID_INVALID
== (uid_t
) GID_INVALID
);
3612 if (uid
== 0) /* We don't keep track of root, and will never destroy it */
3615 r
= hashmap_ensure_allocated(uid_refs
, &trivial_hash_ops
);
3619 c
= PTR_TO_UINT32(hashmap_get(*uid_refs
, UID_TO_PTR(uid
)));
3621 n
= c
& ~DESTROY_IPC_FLAG
;
3624 if (n
& DESTROY_IPC_FLAG
) /* check for overflow */
3627 c
= n
| (c
& DESTROY_IPC_FLAG
) | (clean_ipc
? DESTROY_IPC_FLAG
: 0);
3629 return hashmap_replace(*uid_refs
, UID_TO_PTR(uid
), UINT32_TO_PTR(c
));
3632 int manager_ref_uid(Manager
*m
, uid_t uid
, bool clean_ipc
) {
3633 return manager_ref_uid_internal(m
, &m
->uid_refs
, uid
, clean_ipc
);
3636 int manager_ref_gid(Manager
*m
, gid_t gid
, bool clean_ipc
) {
3637 return manager_ref_uid_internal(m
, &m
->gid_refs
, (uid_t
) gid
, clean_ipc
);
3640 static void manager_vacuum_uid_refs_internal(
3643 int (*_clean_ipc
)(uid_t uid
)) {
3652 HASHMAP_FOREACH_KEY(p
, k
, *uid_refs
, i
) {
3656 uid
= PTR_TO_UID(k
);
3657 c
= PTR_TO_UINT32(p
);
3659 n
= c
& ~DESTROY_IPC_FLAG
;
3663 if (c
& DESTROY_IPC_FLAG
) {
3664 log_debug("Found unreferenced %s " UID_FMT
" after reload/reexec. Cleaning up.",
3665 _clean_ipc
== clean_ipc_by_uid
? "UID" : "GID",
3667 (void) _clean_ipc(uid
);
3670 assert_se(hashmap_remove(*uid_refs
, k
) == p
);
3674 void manager_vacuum_uid_refs(Manager
*m
) {
3675 manager_vacuum_uid_refs_internal(m
, &m
->uid_refs
, clean_ipc_by_uid
);
3678 void manager_vacuum_gid_refs(Manager
*m
) {
3679 manager_vacuum_uid_refs_internal(m
, &m
->gid_refs
, clean_ipc_by_gid
);
3682 static void manager_serialize_uid_refs_internal(
3686 const char *field_name
) {
3696 /* Serialize the UID reference table. Or actually, just the IPC destruction flag of it, as the actual counter
3697 * of it is better rebuild after a reload/reexec. */
3699 HASHMAP_FOREACH_KEY(p
, k
, *uid_refs
, i
) {
3703 uid
= PTR_TO_UID(k
);
3704 c
= PTR_TO_UINT32(p
);
3706 if (!(c
& DESTROY_IPC_FLAG
))
3709 fprintf(f
, "%s=" UID_FMT
"\n", field_name
, uid
);
3713 void manager_serialize_uid_refs(Manager
*m
, FILE *f
) {
3714 manager_serialize_uid_refs_internal(m
, f
, &m
->uid_refs
, "destroy-ipc-uid");
3717 void manager_serialize_gid_refs(Manager
*m
, FILE *f
) {
3718 manager_serialize_uid_refs_internal(m
, f
, &m
->gid_refs
, "destroy-ipc-gid");
3721 static void manager_deserialize_uid_refs_one_internal(
3724 const char *value
) {
3734 r
= parse_uid(value
, &uid
);
3735 if (r
< 0 || uid
== 0) {
3736 log_debug("Unable to parse UID reference serialization");
3740 r
= hashmap_ensure_allocated(uid_refs
, &trivial_hash_ops
);
3746 c
= PTR_TO_UINT32(hashmap_get(*uid_refs
, UID_TO_PTR(uid
)));
3747 if (c
& DESTROY_IPC_FLAG
)
3750 c
|= DESTROY_IPC_FLAG
;
3752 r
= hashmap_replace(*uid_refs
, UID_TO_PTR(uid
), UINT32_TO_PTR(c
));
3754 log_debug("Failed to add UID reference entry");
3759 void manager_deserialize_uid_refs_one(Manager
*m
, const char *value
) {
3760 manager_deserialize_uid_refs_one_internal(m
, &m
->uid_refs
, value
);
3763 void manager_deserialize_gid_refs_one(Manager
*m
, const char *value
) {
3764 manager_deserialize_uid_refs_one_internal(m
, &m
->gid_refs
, value
);
3767 int manager_dispatch_user_lookup_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
3771 char unit_name
[UNIT_NAME_MAX
+1];
3774 Manager
*m
= userdata
;
3782 /* Invoked whenever a child process succeeded resolving its user/group to use and sent us the resulting UID/GID
3783 * in a datagram. We parse the datagram here and pass it off to the unit, so that it can add a reference to the
3784 * UID/GID so that it can destroy the UID/GID's IPC objects when the reference counter drops to 0. */
3786 l
= recv(fd
, &buffer
, sizeof(buffer
), MSG_DONTWAIT
);
3788 if (IN_SET(errno
, EINTR
, EAGAIN
))
3791 return log_error_errno(errno
, "Failed to read from user lookup fd: %m");
3794 if ((size_t) l
<= offsetof(struct buffer
, unit_name
)) {
3795 log_warning("Received too short user lookup message, ignoring.");
3799 if ((size_t) l
> offsetof(struct buffer
, unit_name
) + UNIT_NAME_MAX
) {
3800 log_warning("Received too long user lookup message, ignoring.");
3804 if (!uid_is_valid(buffer
.uid
) && !gid_is_valid(buffer
.gid
)) {
3805 log_warning("Got user lookup message with invalid UID/GID pair, ignoring.");
3809 n
= (size_t) l
- offsetof(struct buffer
, unit_name
);
3810 if (memchr(buffer
.unit_name
, 0, n
)) {
3811 log_warning("Received lookup message with embedded NUL character, ignoring.");
3815 buffer
.unit_name
[n
] = 0;
3816 u
= manager_get_unit(m
, buffer
.unit_name
);
3818 log_debug("Got user lookup message but unit doesn't exist, ignoring.");
3822 log_unit_debug(u
, "User lookup succeeded: uid=" UID_FMT
" gid=" GID_FMT
, buffer
.uid
, buffer
.gid
);
3824 unit_notify_user_lookup(u
, buffer
.uid
, buffer
.gid
);
3828 static const char *const manager_state_table
[_MANAGER_STATE_MAX
] = {
3829 [MANAGER_INITIALIZING
] = "initializing",
3830 [MANAGER_STARTING
] = "starting",
3831 [MANAGER_RUNNING
] = "running",
3832 [MANAGER_DEGRADED
] = "degraded",
3833 [MANAGER_MAINTENANCE
] = "maintenance",
3834 [MANAGER_STOPPING
] = "stopping",
3837 DEFINE_STRING_TABLE_LOOKUP(manager_state
, ManagerState
);