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"
57 #include "exec-util.h"
58 #include "exit-status.h"
64 #include "locale-setup.h"
70 #include "parse-util.h"
71 #include "path-lookup.h"
72 #include "path-util.h"
73 #include "process-util.h"
74 #include "ratelimit.h"
76 #include "signal-util.h"
78 #include "stat-util.h"
79 #include "string-table.h"
80 #include "string-util.h"
82 #include "terminal-util.h"
83 #include "time-util.h"
84 #include "transaction.h"
85 #include "umask-util.h"
86 #include "unit-name.h"
87 #include "user-util.h"
92 #define NOTIFY_RCVBUF_SIZE (8*1024*1024)
93 #define CGROUPS_AGENT_RCVBUF_SIZE (8*1024*1024)
95 /* Initial delay and the interval for printing status messages about running jobs */
96 #define JOBS_IN_PROGRESS_WAIT_USEC (5*USEC_PER_SEC)
97 #define JOBS_IN_PROGRESS_PERIOD_USEC (USEC_PER_SEC / 3)
98 #define JOBS_IN_PROGRESS_PERIOD_DIVISOR 3
100 static int manager_dispatch_notify_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
101 static int manager_dispatch_cgroups_agent_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
102 static int manager_dispatch_signal_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
103 static int manager_dispatch_time_change_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
104 static int manager_dispatch_idle_pipe_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
105 static int manager_dispatch_user_lookup_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
106 static int manager_dispatch_jobs_in_progress(sd_event_source
*source
, usec_t usec
, void *userdata
);
107 static int manager_dispatch_run_queue(sd_event_source
*source
, void *userdata
);
108 static int manager_run_environment_generators(Manager
*m
);
109 static int manager_run_generators(Manager
*m
);
111 static void manager_watch_jobs_in_progress(Manager
*m
) {
117 /* We do not want to show the cylon animation if the user
118 * needs to confirm service executions otherwise confirmation
119 * messages will be screwed by the cylon animation. */
120 if (!manager_is_confirm_spawn_disabled(m
))
123 if (m
->jobs_in_progress_event_source
)
126 next
= now(CLOCK_MONOTONIC
) + JOBS_IN_PROGRESS_WAIT_USEC
;
127 r
= sd_event_add_time(
129 &m
->jobs_in_progress_event_source
,
132 manager_dispatch_jobs_in_progress
, m
);
136 (void) sd_event_source_set_description(m
->jobs_in_progress_event_source
, "manager-jobs-in-progress");
139 #define CYLON_BUFFER_EXTRA (2*(sizeof(ANSI_RED)-1) + sizeof(ANSI_HIGHLIGHT_RED)-1 + 2*(sizeof(ANSI_NORMAL)-1))
141 static void draw_cylon(char buffer
[], size_t buflen
, unsigned width
, unsigned pos
) {
144 assert(buflen
>= CYLON_BUFFER_EXTRA
+ width
+ 1);
145 assert(pos
<= width
+1); /* 0 or width+1 mean that the center light is behind the corner */
149 p
= mempset(p
, ' ', pos
-2);
150 if (log_get_show_color())
151 p
= stpcpy(p
, ANSI_RED
);
155 if (pos
> 0 && pos
<= width
) {
156 if (log_get_show_color())
157 p
= stpcpy(p
, ANSI_HIGHLIGHT_RED
);
161 if (log_get_show_color())
162 p
= stpcpy(p
, ANSI_NORMAL
);
165 if (log_get_show_color())
166 p
= stpcpy(p
, ANSI_RED
);
169 p
= mempset(p
, ' ', width
-1-pos
);
170 if (log_get_show_color())
171 strcpy(p
, ANSI_NORMAL
);
175 void manager_flip_auto_status(Manager
*m
, bool enable
) {
179 if (m
->show_status
== SHOW_STATUS_AUTO
)
180 manager_set_show_status(m
, SHOW_STATUS_TEMPORARY
);
182 if (m
->show_status
== SHOW_STATUS_TEMPORARY
)
183 manager_set_show_status(m
, SHOW_STATUS_AUTO
);
187 static void manager_print_jobs_in_progress(Manager
*m
) {
188 _cleanup_free_
char *job_of_n
= NULL
;
191 unsigned counter
= 0, print_nr
;
192 char cylon
[6 + CYLON_BUFFER_EXTRA
+ 1];
194 char time
[FORMAT_TIMESPAN_MAX
], limit
[FORMAT_TIMESPAN_MAX
] = "no limit";
198 assert(m
->n_running_jobs
> 0);
200 manager_flip_auto_status(m
, true);
202 print_nr
= (m
->jobs_in_progress_iteration
/ JOBS_IN_PROGRESS_PERIOD_DIVISOR
) % m
->n_running_jobs
;
204 HASHMAP_FOREACH(j
, m
->jobs
, i
)
205 if (j
->state
== JOB_RUNNING
&& counter
++ == print_nr
)
208 /* m->n_running_jobs must be consistent with the contents of m->jobs,
209 * so the above loop must have succeeded in finding j. */
210 assert(counter
== print_nr
+ 1);
213 cylon_pos
= m
->jobs_in_progress_iteration
% 14;
215 cylon_pos
= 14 - cylon_pos
;
216 draw_cylon(cylon
, sizeof(cylon
), 6, cylon_pos
);
218 m
->jobs_in_progress_iteration
++;
220 if (m
->n_running_jobs
> 1) {
221 if (asprintf(&job_of_n
, "(%u of %u) ", counter
, m
->n_running_jobs
) < 0)
225 format_timespan(time
, sizeof(time
), now(CLOCK_MONOTONIC
) - j
->begin_usec
, 1*USEC_PER_SEC
);
226 if (job_get_timeout(j
, &x
) > 0)
227 format_timespan(limit
, sizeof(limit
), x
- j
->begin_usec
, 1*USEC_PER_SEC
);
229 manager_status_printf(m
, STATUS_TYPE_EPHEMERAL
, cylon
,
230 "%sA %s job is running for %s (%s / %s)",
232 job_type_to_string(j
->type
),
233 unit_description(j
->unit
),
237 static int have_ask_password(void) {
238 _cleanup_closedir_
DIR *dir
;
241 dir
= opendir("/run/systemd/ask-password");
249 FOREACH_DIRENT_ALL(de
, dir
, return -errno
) {
250 if (startswith(de
->d_name
, "ask."))
256 static int manager_dispatch_ask_password_fd(sd_event_source
*source
,
257 int fd
, uint32_t revents
, void *userdata
) {
258 Manager
*m
= userdata
;
264 m
->have_ask_password
= have_ask_password();
265 if (m
->have_ask_password
< 0)
266 /* Log error but continue. Negative have_ask_password
267 * is treated as unknown status. */
268 log_error_errno(m
->have_ask_password
, "Failed to list /run/systemd/ask-password: %m");
273 static void manager_close_ask_password(Manager
*m
) {
276 m
->ask_password_event_source
= sd_event_source_unref(m
->ask_password_event_source
);
277 m
->ask_password_inotify_fd
= safe_close(m
->ask_password_inotify_fd
);
278 m
->have_ask_password
= -EINVAL
;
281 static int manager_check_ask_password(Manager
*m
) {
286 if (!m
->ask_password_event_source
) {
287 assert(m
->ask_password_inotify_fd
< 0);
289 mkdir_p_label("/run/systemd/ask-password", 0755);
291 m
->ask_password_inotify_fd
= inotify_init1(IN_NONBLOCK
|IN_CLOEXEC
);
292 if (m
->ask_password_inotify_fd
< 0)
293 return log_error_errno(errno
, "inotify_init1() failed: %m");
295 if (inotify_add_watch(m
->ask_password_inotify_fd
, "/run/systemd/ask-password", IN_CREATE
|IN_DELETE
|IN_MOVE
) < 0) {
296 log_error_errno(errno
, "Failed to add watch on /run/systemd/ask-password: %m");
297 manager_close_ask_password(m
);
301 r
= sd_event_add_io(m
->event
, &m
->ask_password_event_source
,
302 m
->ask_password_inotify_fd
, EPOLLIN
,
303 manager_dispatch_ask_password_fd
, m
);
305 log_error_errno(errno
, "Failed to add event source for /run/systemd/ask-password: %m");
306 manager_close_ask_password(m
);
310 (void) sd_event_source_set_description(m
->ask_password_event_source
, "manager-ask-password");
312 /* Queries might have been added meanwhile... */
313 manager_dispatch_ask_password_fd(m
->ask_password_event_source
,
314 m
->ask_password_inotify_fd
, EPOLLIN
, m
);
317 return m
->have_ask_password
;
320 static int manager_watch_idle_pipe(Manager
*m
) {
325 if (m
->idle_pipe_event_source
)
328 if (m
->idle_pipe
[2] < 0)
331 r
= sd_event_add_io(m
->event
, &m
->idle_pipe_event_source
, m
->idle_pipe
[2], EPOLLIN
, manager_dispatch_idle_pipe_fd
, m
);
333 return log_error_errno(r
, "Failed to watch idle pipe: %m");
335 (void) sd_event_source_set_description(m
->idle_pipe_event_source
, "manager-idle-pipe");
340 static void manager_close_idle_pipe(Manager
*m
) {
343 m
->idle_pipe_event_source
= sd_event_source_unref(m
->idle_pipe_event_source
);
345 safe_close_pair(m
->idle_pipe
);
346 safe_close_pair(m
->idle_pipe
+ 2);
349 static int manager_setup_time_change(Manager
*m
) {
352 /* We only care for the cancellation event, hence we set the
353 * timeout to the latest possible value. */
354 struct itimerspec its
= {
355 .it_value
.tv_sec
= TIME_T_MAX
,
359 assert_cc(sizeof(time_t) == sizeof(TIME_T_MAX
));
361 if (m
->test_run_flags
)
364 /* Uses TFD_TIMER_CANCEL_ON_SET to get notifications whenever
365 * CLOCK_REALTIME makes a jump relative to CLOCK_MONOTONIC */
367 m
->time_change_fd
= timerfd_create(CLOCK_REALTIME
, TFD_NONBLOCK
|TFD_CLOEXEC
);
368 if (m
->time_change_fd
< 0)
369 return log_error_errno(errno
, "Failed to create timerfd: %m");
371 if (timerfd_settime(m
->time_change_fd
, TFD_TIMER_ABSTIME
|TFD_TIMER_CANCEL_ON_SET
, &its
, NULL
) < 0) {
372 log_debug_errno(errno
, "Failed to set up TFD_TIMER_CANCEL_ON_SET, ignoring: %m");
373 m
->time_change_fd
= safe_close(m
->time_change_fd
);
377 r
= sd_event_add_io(m
->event
, &m
->time_change_event_source
, m
->time_change_fd
, EPOLLIN
, manager_dispatch_time_change_fd
, m
);
379 return log_error_errno(r
, "Failed to create time change event source: %m");
381 (void) sd_event_source_set_description(m
->time_change_event_source
, "manager-time-change");
383 log_debug("Set up TFD_TIMER_CANCEL_ON_SET timerfd.");
388 static int enable_special_signals(Manager
*m
) {
389 _cleanup_close_
int fd
= -1;
393 if (m
->test_run_flags
)
396 /* Enable that we get SIGINT on control-alt-del. In containers
397 * this will fail with EPERM (older) or EINVAL (newer), so
399 if (reboot(RB_DISABLE_CAD
) < 0 && errno
!= EPERM
&& errno
!= EINVAL
)
400 log_warning_errno(errno
, "Failed to enable ctrl-alt-del handling: %m");
402 fd
= open_terminal("/dev/tty0", O_RDWR
|O_NOCTTY
|O_CLOEXEC
);
404 /* Support systems without virtual console */
406 log_warning_errno(errno
, "Failed to open /dev/tty0: %m");
408 /* Enable that we get SIGWINCH on kbrequest */
409 if (ioctl(fd
, KDSIGACCEPT
, SIGWINCH
) < 0)
410 log_warning_errno(errno
, "Failed to enable kbrequest handling: %m");
416 static int manager_setup_signals(Manager
*m
) {
417 struct sigaction sa
= {
418 .sa_handler
= SIG_DFL
,
419 .sa_flags
= SA_NOCLDSTOP
|SA_RESTART
,
426 assert_se(sigaction(SIGCHLD
, &sa
, NULL
) == 0);
428 /* We make liberal use of realtime signals here. On
429 * Linux/glibc we have 30 of them (with the exception of Linux
430 * on hppa, see below), between SIGRTMIN+0 ... SIGRTMIN+30
433 assert_se(sigemptyset(&mask
) == 0);
434 sigset_add_many(&mask
,
435 SIGCHLD
, /* Child died */
436 SIGTERM
, /* Reexecute daemon */
437 SIGHUP
, /* Reload configuration */
438 SIGUSR1
, /* systemd/upstart: reconnect to D-Bus */
439 SIGUSR2
, /* systemd: dump status */
440 SIGINT
, /* Kernel sends us this on control-alt-del */
441 SIGWINCH
, /* Kernel sends us this on kbrequest (alt-arrowup) */
442 SIGPWR
, /* Some kernel drivers and upsd send us this on power failure */
444 SIGRTMIN
+0, /* systemd: start default.target */
445 SIGRTMIN
+1, /* systemd: isolate rescue.target */
446 SIGRTMIN
+2, /* systemd: isolate emergency.target */
447 SIGRTMIN
+3, /* systemd: start halt.target */
448 SIGRTMIN
+4, /* systemd: start poweroff.target */
449 SIGRTMIN
+5, /* systemd: start reboot.target */
450 SIGRTMIN
+6, /* systemd: start kexec.target */
452 /* ... space for more special targets ... */
454 SIGRTMIN
+13, /* systemd: Immediate halt */
455 SIGRTMIN
+14, /* systemd: Immediate poweroff */
456 SIGRTMIN
+15, /* systemd: Immediate reboot */
457 SIGRTMIN
+16, /* systemd: Immediate kexec */
459 /* ... space for more immediate system state changes ... */
461 SIGRTMIN
+20, /* systemd: enable status messages */
462 SIGRTMIN
+21, /* systemd: disable status messages */
463 SIGRTMIN
+22, /* systemd: set log level to LOG_DEBUG */
464 SIGRTMIN
+23, /* systemd: set log level to LOG_INFO */
465 SIGRTMIN
+24, /* systemd: Immediate exit (--user only) */
467 /* .. one free signal here ... */
469 #if !defined(__hppa64__) && !defined(__hppa__)
470 /* Apparently Linux on hppa has fewer RT
471 * signals (SIGRTMAX is SIGRTMIN+25 there),
472 * hence let's not try to make use of them
473 * here. Since these commands are accessible
474 * by different means and only really a safety
475 * net, the missing functionality on hppa
476 * shouldn't matter. */
478 SIGRTMIN
+26, /* systemd: set log target to journal-or-kmsg */
479 SIGRTMIN
+27, /* systemd: set log target to console */
480 SIGRTMIN
+28, /* systemd: set log target to kmsg */
481 SIGRTMIN
+29, /* systemd: set log target to syslog-or-kmsg (obsolete) */
483 /* ... one free signal here SIGRTMIN+30 ... */
486 assert_se(sigprocmask(SIG_SETMASK
, &mask
, NULL
) == 0);
488 m
->signal_fd
= signalfd(-1, &mask
, SFD_NONBLOCK
|SFD_CLOEXEC
);
489 if (m
->signal_fd
< 0)
492 r
= sd_event_add_io(m
->event
, &m
->signal_event_source
, m
->signal_fd
, EPOLLIN
, manager_dispatch_signal_fd
, m
);
496 (void) sd_event_source_set_description(m
->signal_event_source
, "manager-signal");
498 /* Process signals a bit earlier than the rest of things, but later than notify_fd processing, so that the
499 * notify processing can still figure out to which process/service a message belongs, before we reap the
500 * process. Also, process this before handling cgroup notifications, so that we always collect child exit
501 * status information before detecting that there's no process in a cgroup. */
502 r
= sd_event_source_set_priority(m
->signal_event_source
, SD_EVENT_PRIORITY_NORMAL
-6);
506 if (MANAGER_IS_SYSTEM(m
))
507 return enable_special_signals(m
);
512 static void manager_clean_environment(Manager
*m
) {
515 /* Let's remove some environment variables that we
516 * need ourselves to communicate with our clients */
531 static int manager_default_environment(Manager
*m
) {
534 if (MANAGER_IS_SYSTEM(m
)) {
535 /* The system manager always starts with a clean
536 * environment for its children. It does not import
537 * the kernel's or the parents' exported variables.
539 * The initial passed environment is untouched to keep
540 * /proc/self/environ valid; it is used for tagging
541 * the init process inside containers. */
542 m
->environment
= strv_new("PATH=" DEFAULT_PATH
,
545 /* Import locale variables LC_*= from configuration */
546 locale_setup(&m
->environment
);
548 /* The user manager passes its own environment
549 * along to its children. */
550 m
->environment
= strv_copy(environ
);
555 manager_clean_environment(m
);
556 strv_sort(m
->environment
);
561 static int manager_setup_prefix(Manager
*m
) {
567 static const struct table_entry paths_system
[_EXEC_DIRECTORY_MAX
] = {
568 [EXEC_DIRECTORY_RUNTIME
] = { SD_PATH_SYSTEM_RUNTIME
, NULL
},
569 [EXEC_DIRECTORY_STATE
] = { SD_PATH_SYSTEM_STATE_PRIVATE
, NULL
},
570 [EXEC_DIRECTORY_CACHE
] = { SD_PATH_SYSTEM_STATE_CACHE
, NULL
},
571 [EXEC_DIRECTORY_LOGS
] = { SD_PATH_SYSTEM_STATE_LOGS
, NULL
},
572 [EXEC_DIRECTORY_CONFIGURATION
] = { SD_PATH_SYSTEM_CONFIGURATION
, NULL
},
575 static const struct table_entry paths_user
[_EXEC_DIRECTORY_MAX
] = {
576 [EXEC_DIRECTORY_RUNTIME
] = { SD_PATH_USER_RUNTIME
, NULL
},
577 [EXEC_DIRECTORY_STATE
] = { SD_PATH_USER_CONFIGURATION
, NULL
},
578 [EXEC_DIRECTORY_CACHE
] = { SD_PATH_SYSTEM_STATE_CACHE
, NULL
},
579 [EXEC_DIRECTORY_LOGS
] = { SD_PATH_SYSTEM_CONFIGURATION
, "log" },
580 [EXEC_DIRECTORY_CONFIGURATION
] = { SD_PATH_SYSTEM_CONFIGURATION
, NULL
},
583 const struct table_entry
*p
;
589 if (MANAGER_IS_SYSTEM(m
))
594 for (i
= 0; i
< _EXEC_DIRECTORY_MAX
; i
++) {
595 r
= sd_path_home(p
[i
].type
, p
[i
].suffix
, &m
->prefix
[i
]);
603 int manager_new(UnitFileScope scope
, unsigned test_run_flags
, Manager
**_m
) {
608 assert(IN_SET(scope
, UNIT_FILE_SYSTEM
, UNIT_FILE_USER
));
610 m
= new0(Manager
, 1);
614 m
->unit_file_scope
= scope
;
615 m
->exit_code
= _MANAGER_EXIT_CODE_INVALID
;
616 m
->default_timer_accuracy_usec
= USEC_PER_MINUTE
;
617 m
->default_tasks_accounting
= true;
618 m
->default_tasks_max
= UINT64_MAX
;
619 m
->default_timeout_start_usec
= DEFAULT_TIMEOUT_USEC
;
620 m
->default_timeout_stop_usec
= DEFAULT_TIMEOUT_USEC
;
621 m
->default_restart_usec
= DEFAULT_RESTART_USEC
;
624 if (MANAGER_IS_SYSTEM(m
) && detect_container() <= 0)
625 boot_timestamps(&m
->userspace_timestamp
, &m
->firmware_timestamp
, &m
->loader_timestamp
);
628 /* Prepare log fields we can use for structured logging */
629 if (MANAGER_IS_SYSTEM(m
)) {
630 m
->unit_log_field
= "UNIT=";
631 m
->unit_log_format_string
= "UNIT=%s";
633 m
->invocation_log_field
= "INVOCATION_ID=";
634 m
->invocation_log_format_string
= "INVOCATION_ID=%s";
636 m
->unit_log_field
= "USER_UNIT=";
637 m
->unit_log_format_string
= "USER_UNIT=%s";
639 m
->invocation_log_field
= "USER_INVOCATION_ID=";
640 m
->invocation_log_format_string
= "USER_INVOCATION_ID=%s";
643 m
->idle_pipe
[0] = m
->idle_pipe
[1] = m
->idle_pipe
[2] = m
->idle_pipe
[3] = -1;
645 m
->pin_cgroupfs_fd
= m
->notify_fd
= m
->cgroups_agent_fd
= m
->signal_fd
= m
->time_change_fd
=
646 m
->dev_autofs_fd
= m
->private_listen_fd
= m
->cgroup_inotify_fd
=
647 m
->ask_password_inotify_fd
= -1;
649 m
->user_lookup_fds
[0] = m
->user_lookup_fds
[1] = -1;
651 m
->current_job_id
= 1; /* start as id #1, so that we can leave #0 around as "null-like" value */
653 m
->have_ask_password
= -EINVAL
; /* we don't know */
656 m
->test_run_flags
= test_run_flags
;
658 /* Reboot immediately if the user hits C-A-D more often than 7x per 2s */
659 RATELIMIT_INIT(m
->ctrl_alt_del_ratelimit
, 2 * USEC_PER_SEC
, 7);
661 r
= manager_default_environment(m
);
665 r
= hashmap_ensure_allocated(&m
->units
, &string_hash_ops
);
669 r
= hashmap_ensure_allocated(&m
->jobs
, NULL
);
673 r
= hashmap_ensure_allocated(&m
->cgroup_unit
, &string_hash_ops
);
677 r
= hashmap_ensure_allocated(&m
->watch_bus
, &string_hash_ops
);
681 r
= sd_event_default(&m
->event
);
685 r
= sd_event_add_defer(m
->event
, &m
->run_queue_event_source
, manager_dispatch_run_queue
, m
);
689 r
= sd_event_source_set_priority(m
->run_queue_event_source
, SD_EVENT_PRIORITY_IDLE
);
693 r
= sd_event_source_set_enabled(m
->run_queue_event_source
, SD_EVENT_OFF
);
697 (void) sd_event_source_set_description(m
->run_queue_event_source
, "manager-run-queue");
699 r
= manager_setup_signals(m
);
703 r
= manager_setup_cgroup(m
);
707 r
= manager_setup_time_change(m
);
711 m
->udev
= udev_new();
717 /* Note that we do not set up the notify fd here. We do that after deserialization,
718 * since they might have gotten serialized across the reexec. */
720 m
->taint_usr
= dir_is_empty("/usr") > 0;
722 r
= manager_setup_prefix(m
);
734 static int manager_setup_notify(Manager
*m
) {
737 if (m
->test_run_flags
)
740 if (m
->notify_fd
< 0) {
741 _cleanup_close_
int fd
= -1;
742 union sockaddr_union sa
= {
743 .sa
.sa_family
= AF_UNIX
,
745 static const int one
= 1;
747 /* First free all secondary fields */
748 m
->notify_socket
= mfree(m
->notify_socket
);
749 m
->notify_event_source
= sd_event_source_unref(m
->notify_event_source
);
751 fd
= socket(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
|SOCK_NONBLOCK
, 0);
753 return log_error_errno(errno
, "Failed to allocate notification socket: %m");
755 fd_inc_rcvbuf(fd
, NOTIFY_RCVBUF_SIZE
);
757 m
->notify_socket
= strappend(m
->prefix
[EXEC_DIRECTORY_RUNTIME
], "/systemd/notify");
758 if (!m
->notify_socket
)
761 (void) mkdir_parents_label(m
->notify_socket
, 0755);
762 (void) unlink(m
->notify_socket
);
764 strncpy(sa
.un
.sun_path
, m
->notify_socket
, sizeof(sa
.un
.sun_path
)-1);
765 r
= bind(fd
, &sa
.sa
, SOCKADDR_UN_LEN(sa
.un
));
767 return log_error_errno(errno
, "bind(%s) failed: %m", sa
.un
.sun_path
);
769 r
= setsockopt(fd
, SOL_SOCKET
, SO_PASSCRED
, &one
, sizeof(one
));
771 return log_error_errno(errno
, "SO_PASSCRED failed: %m");
776 log_debug("Using notification socket %s", m
->notify_socket
);
779 if (!m
->notify_event_source
) {
780 r
= sd_event_add_io(m
->event
, &m
->notify_event_source
, m
->notify_fd
, EPOLLIN
, manager_dispatch_notify_fd
, m
);
782 return log_error_errno(r
, "Failed to allocate notify event source: %m");
784 /* Process notification messages a bit earlier than SIGCHLD, so that we can still identify to which
785 * service an exit message belongs. */
786 r
= sd_event_source_set_priority(m
->notify_event_source
, SD_EVENT_PRIORITY_NORMAL
-7);
788 return log_error_errno(r
, "Failed to set priority of notify event source: %m");
790 (void) sd_event_source_set_description(m
->notify_event_source
, "manager-notify");
796 static int manager_setup_cgroups_agent(Manager
*m
) {
798 static const union sockaddr_union sa
= {
799 .un
.sun_family
= AF_UNIX
,
800 .un
.sun_path
= "/run/systemd/cgroups-agent",
804 /* This creates a listening socket we receive cgroups agent messages on. We do not use D-Bus for delivering
805 * these messages from the cgroups agent binary to PID 1, as the cgroups agent binary is very short-living, and
806 * each instance of it needs a new D-Bus connection. Since D-Bus connections are SOCK_STREAM/AF_UNIX, on
807 * overloaded systems the backlog of the D-Bus socket becomes relevant, as not more than the configured number
808 * of D-Bus connections may be queued until the kernel will start dropping further incoming connections,
809 * possibly resulting in lost cgroups agent messages. To avoid this, we'll use a private SOCK_DGRAM/AF_UNIX
810 * socket, where no backlog is relevant as communication may take place without an actual connect() cycle, and
811 * we thus won't lose messages.
813 * Note that PID 1 will forward the agent message to system bus, so that the user systemd instance may listen
814 * to it. The system instance hence listens on this special socket, but the user instances listen on the system
815 * bus for these messages. */
817 if (m
->test_run_flags
)
820 if (!MANAGER_IS_SYSTEM(m
))
823 r
= cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER
);
825 return log_error_errno(r
, "Failed to determine whether unified cgroups hierarchy is used: %m");
826 if (r
> 0) /* We don't need this anymore on the unified hierarchy */
829 if (m
->cgroups_agent_fd
< 0) {
830 _cleanup_close_
int fd
= -1;
832 /* First free all secondary fields */
833 m
->cgroups_agent_event_source
= sd_event_source_unref(m
->cgroups_agent_event_source
);
835 fd
= socket(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
|SOCK_NONBLOCK
, 0);
837 return log_error_errno(errno
, "Failed to allocate cgroups agent socket: %m");
839 fd_inc_rcvbuf(fd
, CGROUPS_AGENT_RCVBUF_SIZE
);
841 (void) unlink(sa
.un
.sun_path
);
843 /* Only allow root to connect to this socket */
845 r
= bind(fd
, &sa
.sa
, SOCKADDR_UN_LEN(sa
.un
));
847 return log_error_errno(errno
, "bind(%s) failed: %m", sa
.un
.sun_path
);
849 m
->cgroups_agent_fd
= fd
;
853 if (!m
->cgroups_agent_event_source
) {
854 r
= sd_event_add_io(m
->event
, &m
->cgroups_agent_event_source
, m
->cgroups_agent_fd
, EPOLLIN
, manager_dispatch_cgroups_agent_fd
, m
);
856 return log_error_errno(r
, "Failed to allocate cgroups agent event source: %m");
858 /* Process cgroups notifications early, but after having processed service notification messages or
859 * SIGCHLD signals, so that a cgroup running empty is always just the last safety net of notification,
860 * and we collected the metadata the notification and SIGCHLD stuff offers first. Also see handling of
861 * cgroup inotify for the unified cgroup stuff. */
862 r
= sd_event_source_set_priority(m
->cgroups_agent_event_source
, SD_EVENT_PRIORITY_NORMAL
-5);
864 return log_error_errno(r
, "Failed to set priority of cgroups agent event source: %m");
866 (void) sd_event_source_set_description(m
->cgroups_agent_event_source
, "manager-cgroups-agent");
872 static int manager_setup_user_lookup_fd(Manager
*m
) {
877 /* Set up the socket pair used for passing UID/GID resolution results from forked off processes to PID
878 * 1. Background: we can't do name lookups (NSS) from PID 1, since it might involve IPC and thus activation,
879 * and we might hence deadlock on ourselves. Hence we do all user/group lookups asynchronously from the forked
880 * off processes right before executing the binaries to start. In order to be able to clean up any IPC objects
881 * created by a unit (see RemoveIPC=) we need to know in PID 1 the used UID/GID of the executed processes,
882 * hence we establish this communication channel so that forked off processes can pass their UID/GID
883 * information back to PID 1. The forked off processes send their resolved UID/GID to PID 1 in a simple
884 * datagram, along with their unit name, so that we can share one communication socket pair among all units for
887 * You might wonder why we need a communication channel for this that is independent of the usual notification
888 * socket scheme (i.e. $NOTIFY_SOCKET). The primary difference is about trust: data sent via the $NOTIFY_SOCKET
889 * channel is only accepted if it originates from the right unit and if reception was enabled for it. The user
890 * lookup socket OTOH is only accessible by PID 1 and its children until they exec(), and always available.
892 * Note that this function is called under two circumstances: when we first initialize (in which case we
893 * allocate both the socket pair and the event source to listen on it), and when we deserialize after a reload
894 * (in which case the socket pair already exists but we still need to allocate the event source for it). */
896 if (m
->user_lookup_fds
[0] < 0) {
898 /* Free all secondary fields */
899 safe_close_pair(m
->user_lookup_fds
);
900 m
->user_lookup_event_source
= sd_event_source_unref(m
->user_lookup_event_source
);
902 if (socketpair(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
, 0, m
->user_lookup_fds
) < 0)
903 return log_error_errno(errno
, "Failed to allocate user lookup socket: %m");
905 (void) fd_inc_rcvbuf(m
->user_lookup_fds
[0], NOTIFY_RCVBUF_SIZE
);
908 if (!m
->user_lookup_event_source
) {
909 r
= sd_event_add_io(m
->event
, &m
->user_lookup_event_source
, m
->user_lookup_fds
[0], EPOLLIN
, manager_dispatch_user_lookup_fd
, m
);
911 return log_error_errno(errno
, "Failed to allocate user lookup event source: %m");
913 /* Process even earlier than the notify event source, so that we always know first about valid UID/GID
915 r
= sd_event_source_set_priority(m
->user_lookup_event_source
, SD_EVENT_PRIORITY_NORMAL
-8);
917 return log_error_errno(errno
, "Failed to set priority ot user lookup event source: %m");
919 (void) sd_event_source_set_description(m
->user_lookup_event_source
, "user-lookup");
925 static int manager_connect_bus(Manager
*m
, bool reexecuting
) {
926 bool try_bus_connect
;
931 if (m
->test_run_flags
)
934 u
= manager_get_unit(m
, SPECIAL_DBUS_SERVICE
);
937 (u
&& UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u
))) &&
939 (MANAGER_IS_USER(m
) && getenv("DBUS_SESSION_BUS_ADDRESS")));
941 /* Try to connect to the buses, if possible. */
942 return bus_init(m
, try_bus_connect
);
945 static unsigned manager_dispatch_cleanup_queue(Manager
*m
) {
951 while ((u
= m
->cleanup_queue
)) {
952 assert(u
->in_cleanup_queue
);
962 GC_OFFSET_IN_PATH
, /* This one is on the path we were traveling */
963 GC_OFFSET_UNSURE
, /* No clue */
964 GC_OFFSET_GOOD
, /* We still need this unit */
965 GC_OFFSET_BAD
, /* We don't need this unit anymore */
969 static void unit_gc_mark_good(Unit
*u
, unsigned gc_marker
) {
973 u
->gc_marker
= gc_marker
+ GC_OFFSET_GOOD
;
975 /* Recursively mark referenced units as GOOD as well */
976 SET_FOREACH(other
, u
->dependencies
[UNIT_REFERENCES
], i
)
977 if (other
->gc_marker
== gc_marker
+ GC_OFFSET_UNSURE
)
978 unit_gc_mark_good(other
, gc_marker
);
981 static void unit_gc_sweep(Unit
*u
, unsigned gc_marker
) {
988 if (u
->gc_marker
== gc_marker
+ GC_OFFSET_GOOD
||
989 u
->gc_marker
== gc_marker
+ GC_OFFSET_BAD
||
990 u
->gc_marker
== gc_marker
+ GC_OFFSET_UNSURE
||
991 u
->gc_marker
== gc_marker
+ GC_OFFSET_IN_PATH
)
994 if (u
->in_cleanup_queue
)
997 if (unit_check_gc(u
))
1000 u
->gc_marker
= gc_marker
+ GC_OFFSET_IN_PATH
;
1004 SET_FOREACH(other
, u
->dependencies
[UNIT_REFERENCED_BY
], i
) {
1005 unit_gc_sweep(other
, gc_marker
);
1007 if (other
->gc_marker
== gc_marker
+ GC_OFFSET_GOOD
)
1010 if (other
->gc_marker
!= gc_marker
+ GC_OFFSET_BAD
)
1017 /* We were unable to find anything out about this entry, so
1018 * let's investigate it later */
1019 u
->gc_marker
= gc_marker
+ GC_OFFSET_UNSURE
;
1020 unit_add_to_gc_queue(u
);
1024 /* We definitely know that this one is not useful anymore, so
1025 * let's mark it for deletion */
1026 u
->gc_marker
= gc_marker
+ GC_OFFSET_BAD
;
1027 unit_add_to_cleanup_queue(u
);
1031 unit_gc_mark_good(u
, gc_marker
);
1034 static unsigned manager_dispatch_gc_unit_queue(Manager
*m
) {
1035 unsigned n
= 0, gc_marker
;
1040 /* log_debug("Running GC..."); */
1042 m
->gc_marker
+= _GC_OFFSET_MAX
;
1043 if (m
->gc_marker
+ _GC_OFFSET_MAX
<= _GC_OFFSET_MAX
)
1046 gc_marker
= m
->gc_marker
;
1048 while ((u
= m
->gc_unit_queue
)) {
1049 assert(u
->in_gc_queue
);
1051 unit_gc_sweep(u
, gc_marker
);
1053 LIST_REMOVE(gc_queue
, m
->gc_unit_queue
, u
);
1054 u
->in_gc_queue
= false;
1058 if (u
->gc_marker
== gc_marker
+ GC_OFFSET_BAD
||
1059 u
->gc_marker
== gc_marker
+ GC_OFFSET_UNSURE
) {
1061 log_unit_debug(u
, "Collecting.");
1062 u
->gc_marker
= gc_marker
+ GC_OFFSET_BAD
;
1063 unit_add_to_cleanup_queue(u
);
1070 static unsigned manager_dispatch_gc_job_queue(Manager
*m
) {
1076 while ((j
= m
->gc_job_queue
)) {
1077 assert(j
->in_gc_queue
);
1079 LIST_REMOVE(gc_queue
, m
->gc_job_queue
, j
);
1080 j
->in_gc_queue
= false;
1084 if (job_check_gc(j
))
1087 log_unit_debug(j
->unit
, "Collecting job.");
1088 (void) job_finish_and_invalidate(j
, JOB_COLLECTED
, false, false);
1094 static void manager_clear_jobs_and_units(Manager
*m
) {
1099 while ((u
= hashmap_first(m
->units
)))
1102 manager_dispatch_cleanup_queue(m
);
1104 assert(!m
->load_queue
);
1105 assert(!m
->run_queue
);
1106 assert(!m
->dbus_unit_queue
);
1107 assert(!m
->dbus_job_queue
);
1108 assert(!m
->cleanup_queue
);
1109 assert(!m
->gc_unit_queue
);
1110 assert(!m
->gc_job_queue
);
1112 assert(hashmap_isempty(m
->jobs
));
1113 assert(hashmap_isempty(m
->units
));
1115 m
->n_on_console
= 0;
1116 m
->n_running_jobs
= 0;
1119 Manager
* manager_free(Manager
*m
) {
1122 ExecDirectoryType dt
;
1127 manager_clear_jobs_and_units(m
);
1129 for (c
= 0; c
< _UNIT_TYPE_MAX
; c
++)
1130 if (unit_vtable
[c
]->shutdown
)
1131 unit_vtable
[c
]->shutdown(m
);
1133 /* If we reexecute ourselves, we keep the root cgroup around */
1134 manager_shutdown_cgroup(m
, m
->exit_code
!= MANAGER_REEXECUTE
);
1136 lookup_paths_flush_generator(&m
->lookup_paths
);
1140 dynamic_user_vacuum(m
, false);
1141 hashmap_free(m
->dynamic_users
);
1143 hashmap_free(m
->units
);
1144 hashmap_free(m
->units_by_invocation_id
);
1145 hashmap_free(m
->jobs
);
1146 hashmap_free(m
->watch_pids1
);
1147 hashmap_free(m
->watch_pids2
);
1148 hashmap_free(m
->watch_bus
);
1150 set_free(m
->startup_units
);
1151 set_free(m
->failed_units
);
1153 sd_event_source_unref(m
->signal_event_source
);
1154 sd_event_source_unref(m
->notify_event_source
);
1155 sd_event_source_unref(m
->cgroups_agent_event_source
);
1156 sd_event_source_unref(m
->time_change_event_source
);
1157 sd_event_source_unref(m
->jobs_in_progress_event_source
);
1158 sd_event_source_unref(m
->run_queue_event_source
);
1159 sd_event_source_unref(m
->user_lookup_event_source
);
1161 safe_close(m
->signal_fd
);
1162 safe_close(m
->notify_fd
);
1163 safe_close(m
->cgroups_agent_fd
);
1164 safe_close(m
->time_change_fd
);
1165 safe_close_pair(m
->user_lookup_fds
);
1167 manager_close_ask_password(m
);
1169 manager_close_idle_pipe(m
);
1171 udev_unref(m
->udev
);
1172 sd_event_unref(m
->event
);
1174 free(m
->notify_socket
);
1176 lookup_paths_free(&m
->lookup_paths
);
1177 strv_free(m
->environment
);
1179 hashmap_free(m
->cgroup_unit
);
1180 set_free_free(m
->unit_path_cache
);
1182 free(m
->switch_root
);
1183 free(m
->switch_root_init
);
1185 for (i
= 0; i
< _RLIMIT_MAX
; i
++)
1186 m
->rlimit
[i
] = mfree(m
->rlimit
[i
]);
1188 assert(hashmap_isempty(m
->units_requiring_mounts_for
));
1189 hashmap_free(m
->units_requiring_mounts_for
);
1191 hashmap_free(m
->uid_refs
);
1192 hashmap_free(m
->gid_refs
);
1194 for (dt
= 0; dt
< _EXEC_DIRECTORY_MAX
; dt
++)
1195 m
->prefix
[dt
] = mfree(m
->prefix
[dt
]);
1200 void manager_enumerate(Manager
*m
) {
1205 /* Let's ask every type to load all units from disk/kernel
1206 * that it might know */
1207 for (c
= 0; c
< _UNIT_TYPE_MAX
; c
++) {
1208 if (!unit_type_supported(c
)) {
1209 log_debug("Unit type .%s is not supported on this system.", unit_type_to_string(c
));
1213 if (!unit_vtable
[c
]->enumerate
)
1216 unit_vtable
[c
]->enumerate(m
);
1219 manager_dispatch_load_queue(m
);
1222 static void manager_coldplug(Manager
*m
) {
1230 /* Then, let's set up their initial state. */
1231 HASHMAP_FOREACH_KEY(u
, k
, m
->units
, i
) {
1233 /* ignore aliases */
1237 r
= unit_coldplug(u
);
1239 log_warning_errno(r
, "We couldn't coldplug %s, proceeding anyway: %m", u
->id
);
1243 static void manager_build_unit_path_cache(Manager
*m
) {
1249 set_free_free(m
->unit_path_cache
);
1251 m
->unit_path_cache
= set_new(&string_hash_ops
);
1252 if (!m
->unit_path_cache
) {
1257 /* This simply builds a list of files we know exist, so that
1258 * we don't always have to go to disk */
1260 STRV_FOREACH(i
, m
->lookup_paths
.search_path
) {
1261 _cleanup_closedir_
DIR *d
= NULL
;
1266 if (errno
!= ENOENT
)
1267 log_warning_errno(errno
, "Failed to open directory %s, ignoring: %m", *i
);
1271 FOREACH_DIRENT(de
, d
, r
= -errno
; goto fail
) {
1274 p
= strjoin(streq(*i
, "/") ? "" : *i
, "/", de
->d_name
);
1280 r
= set_consume(m
->unit_path_cache
, p
);
1289 log_warning_errno(r
, "Failed to build unit path cache, proceeding without: %m");
1290 m
->unit_path_cache
= set_free_free(m
->unit_path_cache
);
1293 static void manager_distribute_fds(Manager
*m
, FDSet
*fds
) {
1299 HASHMAP_FOREACH(u
, m
->units
, i
) {
1301 if (fdset_size(fds
) <= 0)
1304 if (!UNIT_VTABLE(u
)->distribute_fds
)
1307 UNIT_VTABLE(u
)->distribute_fds(u
, fds
);
1311 int manager_startup(Manager
*m
, FILE *serialization
, FDSet
*fds
) {
1316 /* If we are running in test mode, we still want to run the generators,
1317 * but we should not touch the real generator directories. */
1318 r
= lookup_paths_init(&m
->lookup_paths
, m
->unit_file_scope
,
1319 m
->test_run_flags
? LOOKUP_PATHS_TEMPORARY_GENERATED
: 0,
1324 r
= manager_run_environment_generators(m
);
1328 /* Make sure the transient directory always exists, so that it remains
1329 * in the search path */
1330 r
= mkdir_p_label(m
->lookup_paths
.transient
, 0755);
1334 dual_timestamp_get(&m
->generators_start_timestamp
);
1335 r
= manager_run_generators(m
);
1336 dual_timestamp_get(&m
->generators_finish_timestamp
);
1340 if (m
->first_boot
> 0 &&
1341 m
->unit_file_scope
== UNIT_FILE_SYSTEM
&&
1342 !m
->test_run_flags
) {
1344 q
= unit_file_preset_all(UNIT_FILE_SYSTEM
, 0, NULL
, UNIT_FILE_PRESET_ENABLE_ONLY
, NULL
, 0);
1346 log_full_errno(q
== -EEXIST
? LOG_NOTICE
: LOG_WARNING
, q
, "Failed to populate /etc with preset unit settings, ignoring: %m");
1348 log_info("Populated /etc with preset unit settings.");
1351 lookup_paths_reduce(&m
->lookup_paths
);
1352 manager_build_unit_path_cache(m
);
1354 /* If we will deserialize make sure that during enumeration
1355 * this is already known, so we increase the counter here
1360 /* First, enumerate what we can from all config files */
1361 dual_timestamp_get(&m
->units_load_start_timestamp
);
1362 manager_enumerate(m
);
1363 dual_timestamp_get(&m
->units_load_finish_timestamp
);
1365 /* Second, deserialize if there is something to deserialize */
1366 if (serialization
) {
1367 r
= manager_deserialize(m
, serialization
, fds
);
1369 log_error_errno(r
, "Deserialization failed: %m");
1372 /* Any fds left? Find some unit which wants them. This is
1373 * useful to allow container managers to pass some file
1374 * descriptors to us pre-initialized. This enables
1375 * socket-based activation of entire containers. */
1376 manager_distribute_fds(m
, fds
);
1378 /* We might have deserialized the notify fd, but if we didn't
1379 * then let's create the bus now */
1380 q
= manager_setup_notify(m
);
1381 if (q
< 0 && r
== 0)
1384 q
= manager_setup_cgroups_agent(m
);
1385 if (q
< 0 && r
== 0)
1388 q
= manager_setup_user_lookup_fd(m
);
1389 if (q
< 0 && r
== 0)
1392 /* Let's connect to the bus now. */
1393 (void) manager_connect_bus(m
, !!serialization
);
1395 (void) bus_track_coldplug(m
, &m
->subscribed
, false, m
->deserialized_subscribed
);
1396 m
->deserialized_subscribed
= strv_free(m
->deserialized_subscribed
);
1398 /* Third, fire things up! */
1399 manager_coldplug(m
);
1401 /* Release any dynamic users no longer referenced */
1402 dynamic_user_vacuum(m
, true);
1404 /* Release any references to UIDs/GIDs no longer referenced, and destroy any IPC owned by them */
1405 manager_vacuum_uid_refs(m
);
1406 manager_vacuum_gid_refs(m
);
1408 if (serialization
) {
1409 assert(m
->n_reloading
> 0);
1412 /* Let's wait for the UnitNew/JobNew messages being
1413 * sent, before we notify that the reload is
1415 m
->send_reloading_done
= true;
1421 int manager_add_job(Manager
*m
, JobType type
, Unit
*unit
, JobMode mode
, sd_bus_error
*e
, Job
**_ret
) {
1426 assert(type
< _JOB_TYPE_MAX
);
1428 assert(mode
< _JOB_MODE_MAX
);
1430 if (mode
== JOB_ISOLATE
&& type
!= JOB_START
)
1431 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
, "Isolate is only valid for start.");
1433 if (mode
== JOB_ISOLATE
&& !unit
->allow_isolate
)
1434 return sd_bus_error_setf(e
, BUS_ERROR_NO_ISOLATION
, "Operation refused, unit may not be isolated.");
1436 log_unit_debug(unit
, "Trying to enqueue job %s/%s/%s", unit
->id
, job_type_to_string(type
), job_mode_to_string(mode
));
1438 type
= job_type_collapse(type
, unit
);
1440 tr
= transaction_new(mode
== JOB_REPLACE_IRREVERSIBLY
);
1444 r
= transaction_add_job_and_dependencies(tr
, type
, unit
, NULL
, true, false,
1445 mode
== JOB_IGNORE_DEPENDENCIES
|| mode
== JOB_IGNORE_REQUIREMENTS
,
1446 mode
== JOB_IGNORE_DEPENDENCIES
, e
);
1450 if (mode
== JOB_ISOLATE
) {
1451 r
= transaction_add_isolate_jobs(tr
, m
);
1456 r
= transaction_activate(tr
, m
, mode
, e
);
1460 log_unit_debug(unit
,
1461 "Enqueued job %s/%s as %u", unit
->id
,
1462 job_type_to_string(type
), (unsigned) tr
->anchor_job
->id
);
1465 *_ret
= tr
->anchor_job
;
1467 transaction_free(tr
);
1471 transaction_abort(tr
);
1472 transaction_free(tr
);
1476 int manager_add_job_by_name(Manager
*m
, JobType type
, const char *name
, JobMode mode
, sd_bus_error
*e
, Job
**ret
) {
1477 Unit
*unit
= NULL
; /* just to appease gcc, initialization is not really necessary */
1481 assert(type
< _JOB_TYPE_MAX
);
1483 assert(mode
< _JOB_MODE_MAX
);
1485 r
= manager_load_unit(m
, name
, NULL
, NULL
, &unit
);
1490 return manager_add_job(m
, type
, unit
, mode
, e
, ret
);
1493 int manager_add_job_by_name_and_warn(Manager
*m
, JobType type
, const char *name
, JobMode mode
, Job
**ret
) {
1494 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
1498 assert(type
< _JOB_TYPE_MAX
);
1500 assert(mode
< _JOB_MODE_MAX
);
1502 r
= manager_add_job_by_name(m
, type
, name
, mode
, &error
, ret
);
1504 return log_warning_errno(r
, "Failed to enqueue %s job for %s: %s", job_mode_to_string(mode
), name
, bus_error_message(&error
, r
));
1509 int manager_propagate_reload(Manager
*m
, Unit
*unit
, JobMode mode
, sd_bus_error
*e
) {
1515 assert(mode
< _JOB_MODE_MAX
);
1516 assert(mode
!= JOB_ISOLATE
); /* Isolate is only valid for start */
1518 tr
= transaction_new(mode
== JOB_REPLACE_IRREVERSIBLY
);
1522 /* We need an anchor job */
1523 r
= transaction_add_job_and_dependencies(tr
, JOB_NOP
, unit
, NULL
, false, false, true, true, e
);
1527 /* Failure in adding individual dependencies is ignored, so this always succeeds. */
1528 transaction_add_propagate_reload_jobs(tr
, unit
, tr
->anchor_job
, mode
== JOB_IGNORE_DEPENDENCIES
, e
);
1530 r
= transaction_activate(tr
, m
, mode
, e
);
1534 transaction_free(tr
);
1538 transaction_abort(tr
);
1539 transaction_free(tr
);
1543 Job
*manager_get_job(Manager
*m
, uint32_t id
) {
1546 return hashmap_get(m
->jobs
, UINT32_TO_PTR(id
));
1549 Unit
*manager_get_unit(Manager
*m
, const char *name
) {
1553 return hashmap_get(m
->units
, name
);
1556 unsigned manager_dispatch_load_queue(Manager
*m
) {
1562 /* Make sure we are not run recursively */
1563 if (m
->dispatching_load_queue
)
1566 m
->dispatching_load_queue
= true;
1568 /* Dispatches the load queue. Takes a unit from the queue and
1569 * tries to load its data until the queue is empty */
1571 while ((u
= m
->load_queue
)) {
1572 assert(u
->in_load_queue
);
1578 m
->dispatching_load_queue
= false;
1582 int manager_load_unit_prepare(
1594 assert(name
|| path
);
1597 /* This will prepare the unit for loading, but not actually
1598 * load anything from disk. */
1600 if (path
&& !is_path(path
))
1601 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
, "Path %s is not absolute.", path
);
1604 name
= basename(path
);
1606 t
= unit_name_to_type(name
);
1608 if (t
== _UNIT_TYPE_INVALID
|| !unit_name_is_valid(name
, UNIT_NAME_PLAIN
|UNIT_NAME_INSTANCE
)) {
1609 if (unit_name_is_valid(name
, UNIT_NAME_TEMPLATE
))
1610 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
, "Unit name %s is missing the instance name.", name
);
1612 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
, "Unit name %s is not valid.", name
);
1615 ret
= manager_get_unit(m
, name
);
1621 ret
= unit_new(m
, unit_vtable
[t
]->object_size
);
1626 ret
->fragment_path
= strdup(path
);
1627 if (!ret
->fragment_path
) {
1633 r
= unit_add_name(ret
, name
);
1639 unit_add_to_load_queue(ret
);
1640 unit_add_to_dbus_queue(ret
);
1641 unit_add_to_gc_queue(ret
);
1648 int manager_load_unit(
1660 /* This will load the service information files, but not actually
1661 * start any services or anything. */
1663 r
= manager_load_unit_prepare(m
, name
, path
, e
, _ret
);
1667 manager_dispatch_load_queue(m
);
1669 *_ret
= unit_follow_merge(*_ret
);
1674 void manager_dump_jobs(Manager
*s
, FILE *f
, const char *prefix
) {
1681 HASHMAP_FOREACH(j
, s
->jobs
, i
)
1682 job_dump(j
, f
, prefix
);
1685 void manager_dump_units(Manager
*s
, FILE *f
, const char *prefix
) {
1693 HASHMAP_FOREACH_KEY(u
, t
, s
->units
, i
)
1695 unit_dump(u
, f
, prefix
);
1698 void manager_clear_jobs(Manager
*m
) {
1703 while ((j
= hashmap_first(m
->jobs
)))
1704 /* No need to recurse. We're cancelling all jobs. */
1705 job_finish_and_invalidate(j
, JOB_CANCELED
, false, false);
1708 static int manager_dispatch_run_queue(sd_event_source
*source
, void *userdata
) {
1709 Manager
*m
= userdata
;
1715 while ((j
= m
->run_queue
)) {
1716 assert(j
->installed
);
1717 assert(j
->in_run_queue
);
1719 job_run_and_invalidate(j
);
1722 if (m
->n_running_jobs
> 0)
1723 manager_watch_jobs_in_progress(m
);
1725 if (m
->n_on_console
> 0)
1726 manager_watch_idle_pipe(m
);
1731 static unsigned manager_dispatch_dbus_queue(Manager
*m
) {
1738 if (m
->dispatching_dbus_queue
)
1741 m
->dispatching_dbus_queue
= true;
1743 while ((u
= m
->dbus_unit_queue
)) {
1744 assert(u
->in_dbus_queue
);
1746 bus_unit_send_change_signal(u
);
1750 while ((j
= m
->dbus_job_queue
)) {
1751 assert(j
->in_dbus_queue
);
1753 bus_job_send_change_signal(j
);
1757 m
->dispatching_dbus_queue
= false;
1759 if (m
->send_reloading_done
) {
1760 m
->send_reloading_done
= false;
1762 bus_manager_send_reloading(m
, false);
1765 if (m
->queued_message
)
1766 bus_send_queued_message(m
);
1771 static int manager_dispatch_cgroups_agent_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
1772 Manager
*m
= userdata
;
1773 char buf
[PATH_MAX
+1];
1776 n
= recv(fd
, buf
, sizeof(buf
), 0);
1778 return log_error_errno(errno
, "Failed to read cgroups agent message: %m");
1780 log_error("Got zero-length cgroups agent message, ignoring.");
1783 if ((size_t) n
>= sizeof(buf
)) {
1784 log_error("Got overly long cgroups agent message, ignoring.");
1788 if (memchr(buf
, 0, n
)) {
1789 log_error("Got cgroups agent message with embedded NUL byte, ignoring.");
1794 manager_notify_cgroup_empty(m
, buf
);
1795 (void) bus_forward_agent_released(m
, buf
);
1800 static void manager_invoke_notify_message(Manager
*m
, Unit
*u
, pid_t pid
, const char *buf
, FDSet
*fds
) {
1801 _cleanup_strv_free_
char **tags
= NULL
;
1807 tags
= strv_split(buf
, "\n\r");
1813 if (UNIT_VTABLE(u
)->notify_message
)
1814 UNIT_VTABLE(u
)->notify_message(u
, pid
, tags
, fds
);
1815 else if (_unlikely_(log_get_max_level() >= LOG_DEBUG
)) {
1816 _cleanup_free_
char *x
= NULL
, *y
= NULL
;
1820 y
= ellipsize(x
, 20, 90);
1821 log_unit_debug(u
, "Got notification message \"%s\", ignoring.", strnull(y
));
1825 static int manager_dispatch_notify_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
1827 _cleanup_fdset_free_ FDSet
*fds
= NULL
;
1828 Manager
*m
= userdata
;
1829 char buf
[NOTIFY_BUFFER_MAX
+1];
1830 struct iovec iovec
= {
1832 .iov_len
= sizeof(buf
)-1,
1835 struct cmsghdr cmsghdr
;
1836 uint8_t buf
[CMSG_SPACE(sizeof(struct ucred
)) +
1837 CMSG_SPACE(sizeof(int) * NOTIFY_FD_MAX
)];
1839 struct msghdr msghdr
= {
1842 .msg_control
= &control
,
1843 .msg_controllen
= sizeof(control
),
1846 struct cmsghdr
*cmsg
;
1847 struct ucred
*ucred
= NULL
;
1849 int r
, *fd_array
= NULL
;
1854 assert(m
->notify_fd
== fd
);
1856 if (revents
!= EPOLLIN
) {
1857 log_warning("Got unexpected poll event for notify fd.");
1861 n
= recvmsg(m
->notify_fd
, &msghdr
, MSG_DONTWAIT
|MSG_CMSG_CLOEXEC
|MSG_TRUNC
);
1863 if (IN_SET(errno
, EAGAIN
, EINTR
))
1864 return 0; /* Spurious wakeup, try again */
1866 /* If this is any other, real error, then let's stop processing this socket. This of course means we
1867 * won't take notification messages anymore, but that's still better than busy looping around this:
1868 * being woken up over and over again but being unable to actually read the message off the socket. */
1869 return log_error_errno(errno
, "Failed to receive notification message: %m");
1872 CMSG_FOREACH(cmsg
, &msghdr
) {
1873 if (cmsg
->cmsg_level
== SOL_SOCKET
&& cmsg
->cmsg_type
== SCM_RIGHTS
) {
1875 fd_array
= (int*) CMSG_DATA(cmsg
);
1876 n_fds
= (cmsg
->cmsg_len
- CMSG_LEN(0)) / sizeof(int);
1878 } else if (cmsg
->cmsg_level
== SOL_SOCKET
&&
1879 cmsg
->cmsg_type
== SCM_CREDENTIALS
&&
1880 cmsg
->cmsg_len
== CMSG_LEN(sizeof(struct ucred
))) {
1882 ucred
= (struct ucred
*) CMSG_DATA(cmsg
);
1889 r
= fdset_new_array(&fds
, fd_array
, n_fds
);
1891 close_many(fd_array
, n_fds
);
1897 if (!ucred
|| ucred
->pid
<= 0) {
1898 log_warning("Received notify message without valid credentials. Ignoring.");
1902 if ((size_t) n
>= sizeof(buf
) || (msghdr
.msg_flags
& MSG_TRUNC
)) {
1903 log_warning("Received notify message exceeded maximum size. Ignoring.");
1907 /* As extra safety check, let's make sure the string we get doesn't contain embedded NUL bytes. We permit one
1908 * trailing NUL byte in the message, but don't expect it. */
1909 if (n
> 1 && memchr(buf
, 0, n
-1)) {
1910 log_warning("Received notify message with embedded NUL bytes. Ignoring.");
1914 /* Make sure it's NUL-terminated. */
1917 /* Notify every unit that might be interested, but try
1918 * to avoid notifying the same one multiple times. */
1919 u1
= manager_get_unit_by_pid_cgroup(m
, ucred
->pid
);
1921 manager_invoke_notify_message(m
, u1
, ucred
->pid
, buf
, fds
);
1923 u2
= hashmap_get(m
->watch_pids1
, PID_TO_PTR(ucred
->pid
));
1925 manager_invoke_notify_message(m
, u2
, ucred
->pid
, buf
, fds
);
1927 u3
= hashmap_get(m
->watch_pids2
, PID_TO_PTR(ucred
->pid
));
1928 if (u3
&& u3
!= u2
&& u3
!= u1
)
1929 manager_invoke_notify_message(m
, u3
, ucred
->pid
, buf
, fds
);
1931 if (!u1
&& !u2
&& !u3
)
1932 log_warning("Cannot find unit for notify message of PID "PID_FMT
".", ucred
->pid
);
1934 if (fdset_size(fds
) > 0)
1935 log_warning("Got extra auxiliary fds with notification message, closing them.");
1940 static void invoke_sigchld_event(Manager
*m
, Unit
*u
, const siginfo_t
*si
) {
1947 sd_event_get_iteration(m
->event
, &iteration
);
1949 log_unit_debug(u
, "Child "PID_FMT
" belongs to %s", si
->si_pid
, u
->id
);
1951 unit_unwatch_pid(u
, si
->si_pid
);
1953 if (UNIT_VTABLE(u
)->sigchld_event
) {
1954 if (set_size(u
->pids
) <= 1 ||
1955 iteration
!= u
->sigchldgen
||
1956 unit_main_pid(u
) == si
->si_pid
||
1957 unit_control_pid(u
) == si
->si_pid
) {
1958 UNIT_VTABLE(u
)->sigchld_event(u
, si
->si_pid
, si
->si_code
, si
->si_status
);
1959 u
->sigchldgen
= iteration
;
1961 log_debug("%s already issued a sigchld this iteration %" PRIu64
", skipping. Pids still being watched %d", u
->id
, iteration
, set_size(u
->pids
));
1965 static int manager_dispatch_sigchld(Manager
*m
) {
1971 /* First we call waitd() for a PID and do not reap the
1972 * zombie. That way we can still access /proc/$PID for
1973 * it while it is a zombie. */
1974 if (waitid(P_ALL
, 0, &si
, WEXITED
|WNOHANG
|WNOWAIT
) < 0) {
1976 if (errno
== ECHILD
)
1988 if (si
.si_code
== CLD_EXITED
|| si
.si_code
== CLD_KILLED
|| si
.si_code
== CLD_DUMPED
) {
1989 _cleanup_free_
char *name
= NULL
;
1992 get_process_comm(si
.si_pid
, &name
);
1994 log_debug("Child "PID_FMT
" (%s) died (code=%s, status=%i/%s)",
1995 si
.si_pid
, strna(name
),
1996 sigchld_code_to_string(si
.si_code
),
1998 strna(si
.si_code
== CLD_EXITED
1999 ? exit_status_to_string(si
.si_status
, EXIT_STATUS_FULL
)
2000 : signal_to_string(si
.si_status
)));
2002 /* And now figure out the unit this belongs
2003 * to, it might be multiple... */
2004 u1
= manager_get_unit_by_pid_cgroup(m
, si
.si_pid
);
2006 invoke_sigchld_event(m
, u1
, &si
);
2007 u2
= hashmap_get(m
->watch_pids1
, PID_TO_PTR(si
.si_pid
));
2009 invoke_sigchld_event(m
, u2
, &si
);
2010 u3
= hashmap_get(m
->watch_pids2
, PID_TO_PTR(si
.si_pid
));
2011 if (u3
&& u3
!= u2
&& u3
!= u1
)
2012 invoke_sigchld_event(m
, u3
, &si
);
2015 /* And now, we actually reap the zombie. */
2016 if (waitid(P_PID
, si
.si_pid
, &si
, WEXITED
) < 0) {
2027 static void manager_start_target(Manager
*m
, const char *name
, JobMode mode
) {
2028 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
2031 log_debug("Activating special unit %s", name
);
2033 r
= manager_add_job_by_name(m
, JOB_START
, name
, mode
, &error
, NULL
);
2035 log_error("Failed to enqueue %s job: %s", name
, bus_error_message(&error
, r
));
2038 static void manager_handle_ctrl_alt_del(Manager
*m
) {
2039 /* If the user presses C-A-D more than
2040 * 7 times within 2s, we reboot/shutdown immediately,
2041 * unless it was disabled in system.conf */
2043 if (ratelimit_test(&m
->ctrl_alt_del_ratelimit
) || m
->cad_burst_action
== EMERGENCY_ACTION_NONE
)
2044 manager_start_target(m
, SPECIAL_CTRL_ALT_DEL_TARGET
, JOB_REPLACE_IRREVERSIBLY
);
2046 emergency_action(m
, m
->cad_burst_action
, NULL
,
2047 "Ctrl-Alt-Del was pressed more than 7 times within 2s");
2050 static int manager_dispatch_signal_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
2051 Manager
*m
= userdata
;
2053 struct signalfd_siginfo sfsi
;
2054 bool sigchld
= false;
2058 assert(m
->signal_fd
== fd
);
2060 if (revents
!= EPOLLIN
) {
2061 log_warning("Got unexpected events from signal file descriptor.");
2066 n
= read(m
->signal_fd
, &sfsi
, sizeof(sfsi
));
2067 if (n
!= sizeof(sfsi
)) {
2069 log_warning("Truncated read from signal fd (%zu bytes)!", n
);
2073 if (IN_SET(errno
, EINTR
, EAGAIN
))
2076 /* We return an error here, which will kill this handler,
2077 * to avoid a busy loop on read error. */
2078 return log_error_errno(errno
, "Reading from signal fd failed: %m");
2081 log_received_signal(sfsi
.ssi_signo
== SIGCHLD
||
2082 (sfsi
.ssi_signo
== SIGTERM
&& MANAGER_IS_USER(m
))
2083 ? LOG_DEBUG
: LOG_INFO
,
2086 switch (sfsi
.ssi_signo
) {
2093 if (MANAGER_IS_SYSTEM(m
)) {
2094 /* This is for compatibility with the
2095 * original sysvinit */
2096 r
= verify_run_space_and_log("Refusing to reexecute");
2098 m
->exit_code
= MANAGER_REEXECUTE
;
2105 if (MANAGER_IS_SYSTEM(m
))
2106 manager_handle_ctrl_alt_del(m
);
2108 manager_start_target(m
, SPECIAL_EXIT_TARGET
,
2109 JOB_REPLACE_IRREVERSIBLY
);
2113 if (MANAGER_IS_SYSTEM(m
))
2114 manager_start_target(m
, SPECIAL_KBREQUEST_TARGET
, JOB_REPLACE
);
2116 /* This is a nop on non-init */
2120 if (MANAGER_IS_SYSTEM(m
))
2121 manager_start_target(m
, SPECIAL_SIGPWR_TARGET
, JOB_REPLACE
);
2123 /* This is a nop on non-init */
2129 u
= manager_get_unit(m
, SPECIAL_DBUS_SERVICE
);
2131 if (!u
|| UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u
))) {
2132 log_info("Trying to reconnect to bus...");
2136 if (!u
|| !UNIT_IS_ACTIVE_OR_ACTIVATING(unit_active_state(u
))) {
2137 log_info("Loading D-Bus service...");
2138 manager_start_target(m
, SPECIAL_DBUS_SERVICE
, JOB_REPLACE
);
2145 _cleanup_free_
char *dump
= NULL
;
2146 _cleanup_fclose_
FILE *f
= NULL
;
2149 f
= open_memstream(&dump
, &size
);
2151 log_warning_errno(errno
, "Failed to allocate memory stream: %m");
2155 manager_dump_units(m
, f
, "\t");
2156 manager_dump_jobs(m
, f
, "\t");
2158 r
= fflush_and_check(f
);
2160 log_warning_errno(r
, "Failed to write status stream: %m");
2164 log_dump(LOG_INFO
, dump
);
2169 r
= verify_run_space_and_log("Refusing to reload");
2171 m
->exit_code
= MANAGER_RELOAD
;
2176 /* Starting SIGRTMIN+0 */
2177 static const struct {
2180 } target_table
[] = {
2181 [0] = { SPECIAL_DEFAULT_TARGET
, JOB_ISOLATE
},
2182 [1] = { SPECIAL_RESCUE_TARGET
, JOB_ISOLATE
},
2183 [2] = { SPECIAL_EMERGENCY_TARGET
, JOB_ISOLATE
},
2184 [3] = { SPECIAL_HALT_TARGET
, JOB_REPLACE_IRREVERSIBLY
},
2185 [4] = { SPECIAL_POWEROFF_TARGET
, JOB_REPLACE_IRREVERSIBLY
},
2186 [5] = { SPECIAL_REBOOT_TARGET
, JOB_REPLACE_IRREVERSIBLY
},
2187 [6] = { SPECIAL_KEXEC_TARGET
, JOB_REPLACE_IRREVERSIBLY
}
2190 /* Starting SIGRTMIN+13, so that target halt and system halt are 10 apart */
2191 static const ManagerExitCode code_table
[] = {
2193 [1] = MANAGER_POWEROFF
,
2194 [2] = MANAGER_REBOOT
,
2198 if ((int) sfsi
.ssi_signo
>= SIGRTMIN
+0 &&
2199 (int) sfsi
.ssi_signo
< SIGRTMIN
+(int) ELEMENTSOF(target_table
)) {
2200 int idx
= (int) sfsi
.ssi_signo
- SIGRTMIN
;
2201 manager_start_target(m
, target_table
[idx
].target
,
2202 target_table
[idx
].mode
);
2206 if ((int) sfsi
.ssi_signo
>= SIGRTMIN
+13 &&
2207 (int) sfsi
.ssi_signo
< SIGRTMIN
+13+(int) ELEMENTSOF(code_table
)) {
2208 m
->exit_code
= code_table
[sfsi
.ssi_signo
- SIGRTMIN
- 13];
2212 switch (sfsi
.ssi_signo
- SIGRTMIN
) {
2215 manager_set_show_status(m
, SHOW_STATUS_YES
);
2219 manager_set_show_status(m
, SHOW_STATUS_NO
);
2223 log_set_max_level(LOG_DEBUG
);
2224 log_info("Setting log level to debug.");
2228 log_set_max_level(LOG_INFO
);
2229 log_info("Setting log level to info.");
2233 if (MANAGER_IS_USER(m
)) {
2234 m
->exit_code
= MANAGER_EXIT
;
2238 /* This is a nop on init */
2242 case 29: /* compatibility: used to be mapped to LOG_TARGET_SYSLOG_OR_KMSG */
2243 log_set_target(LOG_TARGET_JOURNAL_OR_KMSG
);
2244 log_notice("Setting log target to journal-or-kmsg.");
2248 log_set_target(LOG_TARGET_CONSOLE
);
2249 log_notice("Setting log target to console.");
2253 log_set_target(LOG_TARGET_KMSG
);
2254 log_notice("Setting log target to kmsg.");
2258 log_warning("Got unhandled signal <%s>.", signal_to_string(sfsi
.ssi_signo
));
2265 manager_dispatch_sigchld(m
);
2270 static int manager_dispatch_time_change_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
2271 Manager
*m
= userdata
;
2276 assert(m
->time_change_fd
== fd
);
2278 log_struct(LOG_DEBUG
,
2279 "MESSAGE_ID=" SD_MESSAGE_TIME_CHANGE_STR
,
2280 LOG_MESSAGE("Time has been changed"),
2283 /* Restart the watch */
2284 m
->time_change_event_source
= sd_event_source_unref(m
->time_change_event_source
);
2285 m
->time_change_fd
= safe_close(m
->time_change_fd
);
2287 manager_setup_time_change(m
);
2289 HASHMAP_FOREACH(u
, m
->units
, i
)
2290 if (UNIT_VTABLE(u
)->time_change
)
2291 UNIT_VTABLE(u
)->time_change(u
);
2296 static int manager_dispatch_idle_pipe_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
2297 Manager
*m
= userdata
;
2300 assert(m
->idle_pipe
[2] == fd
);
2302 m
->no_console_output
= m
->n_on_console
> 0;
2304 manager_close_idle_pipe(m
);
2309 static int manager_dispatch_jobs_in_progress(sd_event_source
*source
, usec_t usec
, void *userdata
) {
2310 Manager
*m
= userdata
;
2317 manager_print_jobs_in_progress(m
);
2319 next
= now(CLOCK_MONOTONIC
) + JOBS_IN_PROGRESS_PERIOD_USEC
;
2320 r
= sd_event_source_set_time(source
, next
);
2324 return sd_event_source_set_enabled(source
, SD_EVENT_ONESHOT
);
2327 int manager_loop(Manager
*m
) {
2330 RATELIMIT_DEFINE(rl
, 1*USEC_PER_SEC
, 50000);
2333 m
->exit_code
= MANAGER_OK
;
2335 /* Release the path cache */
2336 m
->unit_path_cache
= set_free_free(m
->unit_path_cache
);
2338 manager_check_finished(m
);
2340 /* There might still be some zombies hanging around from
2341 * before we were exec()'ed. Let's reap them. */
2342 r
= manager_dispatch_sigchld(m
);
2346 while (m
->exit_code
== MANAGER_OK
) {
2349 if (m
->runtime_watchdog
> 0 && m
->runtime_watchdog
!= USEC_INFINITY
&& MANAGER_IS_SYSTEM(m
))
2352 if (!ratelimit_test(&rl
)) {
2353 /* Yay, something is going seriously wrong, pause a little */
2354 log_warning("Looping too fast. Throttling execution a little.");
2358 if (manager_dispatch_load_queue(m
) > 0)
2361 if (manager_dispatch_gc_job_queue(m
) > 0)
2364 if (manager_dispatch_gc_unit_queue(m
) > 0)
2367 if (manager_dispatch_cleanup_queue(m
) > 0)
2370 if (manager_dispatch_cgroup_queue(m
) > 0)
2373 if (manager_dispatch_dbus_queue(m
) > 0)
2376 /* Sleep for half the watchdog time */
2377 if (m
->runtime_watchdog
> 0 && m
->runtime_watchdog
!= USEC_INFINITY
&& MANAGER_IS_SYSTEM(m
)) {
2378 wait_usec
= m
->runtime_watchdog
/ 2;
2382 wait_usec
= USEC_INFINITY
;
2384 r
= sd_event_run(m
->event
, wait_usec
);
2386 return log_error_errno(r
, "Failed to run event loop: %m");
2389 return m
->exit_code
;
2392 int manager_load_unit_from_dbus_path(Manager
*m
, const char *s
, sd_bus_error
*e
, Unit
**_u
) {
2393 _cleanup_free_
char *n
= NULL
;
2394 sd_id128_t invocation_id
;
2402 r
= unit_name_from_dbus_path(s
, &n
);
2406 /* Permit addressing units by invocation ID: if the passed bus path is suffixed by a 128bit ID then we use it
2407 * as invocation ID. */
2408 r
= sd_id128_from_string(n
, &invocation_id
);
2410 u
= hashmap_get(m
->units_by_invocation_id
, &invocation_id
);
2416 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
));
2419 /* If this didn't work, we check if this is a unit name */
2420 if (!unit_name_is_valid(n
, UNIT_NAME_PLAIN
|UNIT_NAME_INSTANCE
))
2421 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
, "Unit name %s is neither a valid invocation ID nor unit name.", n
);
2423 r
= manager_load_unit(m
, n
, NULL
, e
, &u
);
2431 int manager_get_job_from_dbus_path(Manager
*m
, const char *s
, Job
**_j
) {
2441 p
= startswith(s
, "/org/freedesktop/systemd1/job/");
2445 r
= safe_atou(p
, &id
);
2449 j
= manager_get_job(m
, id
);
2458 void manager_send_unit_audit(Manager
*m
, Unit
*u
, int type
, bool success
) {
2461 _cleanup_free_
char *p
= NULL
;
2465 if (!MANAGER_IS_SYSTEM(m
))
2468 audit_fd
= get_audit_fd();
2472 /* Don't generate audit events if the service was already
2473 * started and we're just deserializing */
2474 if (MANAGER_IS_RELOADING(m
))
2477 if (u
->type
!= UNIT_SERVICE
)
2480 r
= unit_name_to_prefix_and_instance(u
->id
, &p
);
2482 log_error_errno(r
, "Failed to extract prefix and instance of unit name: %m");
2486 msg
= strjoina("unit=", p
);
2487 if (audit_log_user_comm_message(audit_fd
, type
, msg
, "systemd", NULL
, NULL
, NULL
, success
) < 0) {
2489 /* We aren't allowed to send audit messages?
2490 * Then let's not retry again. */
2493 log_warning_errno(errno
, "Failed to send audit message: %m");
2499 void manager_send_unit_plymouth(Manager
*m
, Unit
*u
) {
2500 static const union sockaddr_union sa
= PLYMOUTH_SOCKET
;
2501 _cleanup_free_
char *message
= NULL
;
2502 _cleanup_close_
int fd
= -1;
2505 /* Don't generate plymouth events if the service was already
2506 * started and we're just deserializing */
2507 if (MANAGER_IS_RELOADING(m
))
2510 if (!MANAGER_IS_SYSTEM(m
))
2513 if (detect_container() > 0)
2516 if (u
->type
!= UNIT_SERVICE
&&
2517 u
->type
!= UNIT_MOUNT
&&
2518 u
->type
!= UNIT_SWAP
)
2521 /* We set SOCK_NONBLOCK here so that we rather drop the
2522 * message then wait for plymouth */
2523 fd
= socket(AF_UNIX
, SOCK_STREAM
|SOCK_CLOEXEC
|SOCK_NONBLOCK
, 0);
2525 log_error_errno(errno
, "socket() failed: %m");
2529 if (connect(fd
, &sa
.sa
, SOCKADDR_UN_LEN(sa
.un
)) < 0) {
2531 if (!IN_SET(errno
, EPIPE
, EAGAIN
, ENOENT
, ECONNREFUSED
, ECONNRESET
, ECONNABORTED
))
2532 log_error_errno(errno
, "connect() failed: %m");
2536 if (asprintf(&message
, "U\002%c%s%n", (int) (strlen(u
->id
) + 1), u
->id
, &n
) < 0) {
2542 if (write(fd
, message
, n
+ 1) != n
+ 1)
2543 if (!IN_SET(errno
, EPIPE
, EAGAIN
, ENOENT
, ECONNREFUSED
, ECONNRESET
, ECONNABORTED
))
2544 log_error_errno(errno
, "Failed to write Plymouth message: %m");
2547 int manager_open_serialization(Manager
*m
, FILE **_f
) {
2553 fd
= open_serialization_fd("systemd-state");
2557 f
= fdopen(fd
, "w+");
2567 int manager_serialize(Manager
*m
, FILE *f
, FDSet
*fds
, bool switching_root
) {
2579 fprintf(f
, "current-job-id=%"PRIu32
"\n", m
->current_job_id
);
2580 fprintf(f
, "taint-usr=%s\n", yes_no(m
->taint_usr
));
2581 fprintf(f
, "n-installed-jobs=%u\n", m
->n_installed_jobs
);
2582 fprintf(f
, "n-failed-jobs=%u\n", m
->n_failed_jobs
);
2584 dual_timestamp_serialize(f
, "firmware-timestamp", &m
->firmware_timestamp
);
2585 dual_timestamp_serialize(f
, "loader-timestamp", &m
->loader_timestamp
);
2586 dual_timestamp_serialize(f
, "kernel-timestamp", &m
->kernel_timestamp
);
2587 dual_timestamp_serialize(f
, "initrd-timestamp", &m
->initrd_timestamp
);
2590 dual_timestamp_serialize(f
, "userspace-timestamp", &m
->userspace_timestamp
);
2591 dual_timestamp_serialize(f
, "finish-timestamp", &m
->finish_timestamp
);
2592 dual_timestamp_serialize(f
, "security-start-timestamp", &m
->security_start_timestamp
);
2593 dual_timestamp_serialize(f
, "security-finish-timestamp", &m
->security_finish_timestamp
);
2594 dual_timestamp_serialize(f
, "generators-start-timestamp", &m
->generators_start_timestamp
);
2595 dual_timestamp_serialize(f
, "generators-finish-timestamp", &m
->generators_finish_timestamp
);
2596 dual_timestamp_serialize(f
, "units-load-start-timestamp", &m
->units_load_start_timestamp
);
2597 dual_timestamp_serialize(f
, "units-load-finish-timestamp", &m
->units_load_finish_timestamp
);
2600 if (!switching_root
)
2601 (void) serialize_environment(f
, m
->environment
);
2603 if (m
->notify_fd
>= 0) {
2606 copy
= fdset_put_dup(fds
, m
->notify_fd
);
2610 fprintf(f
, "notify-fd=%i\n", copy
);
2611 fprintf(f
, "notify-socket=%s\n", m
->notify_socket
);
2614 if (m
->cgroups_agent_fd
>= 0) {
2617 copy
= fdset_put_dup(fds
, m
->cgroups_agent_fd
);
2621 fprintf(f
, "cgroups-agent-fd=%i\n", copy
);
2624 if (m
->user_lookup_fds
[0] >= 0) {
2627 copy0
= fdset_put_dup(fds
, m
->user_lookup_fds
[0]);
2631 copy1
= fdset_put_dup(fds
, m
->user_lookup_fds
[1]);
2635 fprintf(f
, "user-lookup=%i %i\n", copy0
, copy1
);
2638 bus_track_serialize(m
->subscribed
, f
, "subscribed");
2640 r
= dynamic_user_serialize(m
, f
, fds
);
2644 manager_serialize_uid_refs(m
, f
);
2645 manager_serialize_gid_refs(m
, f
);
2647 fputc_unlocked('\n', f
);
2649 HASHMAP_FOREACH_KEY(u
, t
, m
->units
, i
) {
2654 fputs_unlocked(u
->id
, f
);
2655 fputc_unlocked('\n', f
);
2657 r
= unit_serialize(u
, f
, fds
, !switching_root
);
2664 assert(m
->n_reloading
> 0);
2670 r
= bus_fdset_add_all(m
, fds
);
2677 int manager_deserialize(Manager
*m
, FILE *f
, FDSet
*fds
) {
2683 log_debug("Deserializing state...");
2688 char line
[LINE_MAX
];
2689 const char *val
, *l
;
2691 if (!fgets(line
, sizeof(line
), f
)) {
2706 if ((val
= startswith(l
, "current-job-id="))) {
2709 if (safe_atou32(val
, &id
) < 0)
2710 log_notice("Failed to parse current job id value %s", val
);
2712 m
->current_job_id
= MAX(m
->current_job_id
, id
);
2714 } else if ((val
= startswith(l
, "n-installed-jobs="))) {
2717 if (safe_atou32(val
, &n
) < 0)
2718 log_notice("Failed to parse installed jobs counter %s", val
);
2720 m
->n_installed_jobs
+= n
;
2722 } else if ((val
= startswith(l
, "n-failed-jobs="))) {
2725 if (safe_atou32(val
, &n
) < 0)
2726 log_notice("Failed to parse failed jobs counter %s", val
);
2728 m
->n_failed_jobs
+= n
;
2730 } else if ((val
= startswith(l
, "taint-usr="))) {
2733 b
= parse_boolean(val
);
2735 log_notice("Failed to parse taint /usr flag %s", val
);
2737 m
->taint_usr
= m
->taint_usr
|| b
;
2739 } else if ((val
= startswith(l
, "firmware-timestamp=")))
2740 dual_timestamp_deserialize(val
, &m
->firmware_timestamp
);
2741 else if ((val
= startswith(l
, "loader-timestamp=")))
2742 dual_timestamp_deserialize(val
, &m
->loader_timestamp
);
2743 else if ((val
= startswith(l
, "kernel-timestamp=")))
2744 dual_timestamp_deserialize(val
, &m
->kernel_timestamp
);
2745 else if ((val
= startswith(l
, "initrd-timestamp=")))
2746 dual_timestamp_deserialize(val
, &m
->initrd_timestamp
);
2747 else if ((val
= startswith(l
, "userspace-timestamp=")))
2748 dual_timestamp_deserialize(val
, &m
->userspace_timestamp
);
2749 else if ((val
= startswith(l
, "finish-timestamp=")))
2750 dual_timestamp_deserialize(val
, &m
->finish_timestamp
);
2751 else if ((val
= startswith(l
, "security-start-timestamp=")))
2752 dual_timestamp_deserialize(val
, &m
->security_start_timestamp
);
2753 else if ((val
= startswith(l
, "security-finish-timestamp=")))
2754 dual_timestamp_deserialize(val
, &m
->security_finish_timestamp
);
2755 else if ((val
= startswith(l
, "generators-start-timestamp=")))
2756 dual_timestamp_deserialize(val
, &m
->generators_start_timestamp
);
2757 else if ((val
= startswith(l
, "generators-finish-timestamp=")))
2758 dual_timestamp_deserialize(val
, &m
->generators_finish_timestamp
);
2759 else if ((val
= startswith(l
, "units-load-start-timestamp=")))
2760 dual_timestamp_deserialize(val
, &m
->units_load_start_timestamp
);
2761 else if ((val
= startswith(l
, "units-load-finish-timestamp=")))
2762 dual_timestamp_deserialize(val
, &m
->units_load_finish_timestamp
);
2763 else if (startswith(l
, "env=")) {
2764 r
= deserialize_environment(&m
->environment
, l
);
2768 log_notice_errno(r
, "Failed to parse environment entry: \"%s\": %m", l
);
2770 } else if ((val
= startswith(l
, "notify-fd="))) {
2773 if (safe_atoi(val
, &fd
) < 0 || fd
< 0 || !fdset_contains(fds
, fd
))
2774 log_notice("Failed to parse notify fd: \"%s\"", val
);
2776 m
->notify_event_source
= sd_event_source_unref(m
->notify_event_source
);
2777 safe_close(m
->notify_fd
);
2778 m
->notify_fd
= fdset_remove(fds
, fd
);
2781 } else if ((val
= startswith(l
, "notify-socket="))) {
2790 free(m
->notify_socket
);
2791 m
->notify_socket
= n
;
2793 } else if ((val
= startswith(l
, "cgroups-agent-fd="))) {
2796 if (safe_atoi(val
, &fd
) < 0 || fd
< 0 || !fdset_contains(fds
, fd
))
2797 log_notice("Failed to parse cgroups agent fd: %s", val
);
2799 m
->cgroups_agent_event_source
= sd_event_source_unref(m
->cgroups_agent_event_source
);
2800 safe_close(m
->cgroups_agent_fd
);
2801 m
->cgroups_agent_fd
= fdset_remove(fds
, fd
);
2804 } else if ((val
= startswith(l
, "user-lookup="))) {
2807 if (sscanf(val
, "%i %i", &fd0
, &fd1
) != 2 || fd0
< 0 || fd1
< 0 || fd0
== fd1
|| !fdset_contains(fds
, fd0
) || !fdset_contains(fds
, fd1
))
2808 log_notice("Failed to parse user lookup fd: %s", val
);
2810 m
->user_lookup_event_source
= sd_event_source_unref(m
->user_lookup_event_source
);
2811 safe_close_pair(m
->user_lookup_fds
);
2812 m
->user_lookup_fds
[0] = fdset_remove(fds
, fd0
);
2813 m
->user_lookup_fds
[1] = fdset_remove(fds
, fd1
);
2816 } else if ((val
= startswith(l
, "dynamic-user=")))
2817 dynamic_user_deserialize_one(m
, val
, fds
);
2818 else if ((val
= startswith(l
, "destroy-ipc-uid=")))
2819 manager_deserialize_uid_refs_one(m
, val
);
2820 else if ((val
= startswith(l
, "destroy-ipc-gid=")))
2821 manager_deserialize_gid_refs_one(m
, val
);
2822 else if ((val
= startswith(l
, "subscribed="))) {
2824 if (strv_extend(&m
->deserialized_subscribed
, val
) < 0)
2827 } else if (!startswith(l
, "kdbus-fd=")) /* ignore this one */
2828 log_notice("Unknown serialization item '%s'", l
);
2833 char name
[UNIT_NAME_MAX
+2];
2834 const char* unit_name
;
2837 if (!fgets(name
, sizeof(name
), f
)) {
2847 unit_name
= strstrip(name
);
2849 r
= manager_load_unit(m
, unit_name
, NULL
, NULL
, &u
);
2851 log_notice_errno(r
, "Failed to load unit \"%s\", skipping deserialization: %m", unit_name
);
2854 unit_deserialize_skip(f
);
2858 r
= unit_deserialize(u
, f
, fds
);
2860 log_notice_errno(r
, "Failed to deserialize unit \"%s\": %m", unit_name
);
2870 assert(m
->n_reloading
> 0);
2876 int manager_reload(Manager
*m
) {
2878 _cleanup_fclose_
FILE *f
= NULL
;
2879 _cleanup_fdset_free_ FDSet
*fds
= NULL
;
2883 r
= manager_open_serialization(m
, &f
);
2888 bus_manager_send_reloading(m
, true);
2896 r
= manager_serialize(m
, f
, fds
, false);
2902 if (fseeko(f
, 0, SEEK_SET
) < 0) {
2907 /* From here on there is no way back. */
2908 manager_clear_jobs_and_units(m
);
2909 lookup_paths_flush_generator(&m
->lookup_paths
);
2910 lookup_paths_free(&m
->lookup_paths
);
2911 dynamic_user_vacuum(m
, false);
2912 m
->uid_refs
= hashmap_free(m
->uid_refs
);
2913 m
->gid_refs
= hashmap_free(m
->gid_refs
);
2915 q
= lookup_paths_init(&m
->lookup_paths
, m
->unit_file_scope
, 0, NULL
);
2916 if (q
< 0 && r
>= 0)
2919 q
= manager_run_environment_generators(m
);
2920 if (q
< 0 && r
>= 0)
2923 /* Find new unit paths */
2924 q
= manager_run_generators(m
);
2925 if (q
< 0 && r
>= 0)
2928 lookup_paths_reduce(&m
->lookup_paths
);
2929 manager_build_unit_path_cache(m
);
2931 /* First, enumerate what we can from all config files */
2932 manager_enumerate(m
);
2934 /* Second, deserialize our stored data */
2935 q
= manager_deserialize(m
, f
, fds
);
2937 log_error_errno(q
, "Deserialization failed: %m");
2946 /* Re-register notify_fd as event source */
2947 q
= manager_setup_notify(m
);
2948 if (q
< 0 && r
>= 0)
2951 q
= manager_setup_cgroups_agent(m
);
2952 if (q
< 0 && r
>= 0)
2955 q
= manager_setup_user_lookup_fd(m
);
2956 if (q
< 0 && r
>= 0)
2959 /* Third, fire things up! */
2960 manager_coldplug(m
);
2962 /* Release any dynamic users no longer referenced */
2963 dynamic_user_vacuum(m
, true);
2965 /* Release any references to UIDs/GIDs no longer referenced, and destroy any IPC owned by them */
2966 manager_vacuum_uid_refs(m
);
2967 manager_vacuum_gid_refs(m
);
2969 /* Sync current state of bus names with our set of listening units */
2971 manager_sync_bus_names(m
, m
->api_bus
);
2973 assert(m
->n_reloading
> 0);
2976 m
->send_reloading_done
= true;
2981 void manager_reset_failed(Manager
*m
) {
2987 HASHMAP_FOREACH(u
, m
->units
, i
)
2988 unit_reset_failed(u
);
2991 bool manager_unit_inactive_or_pending(Manager
*m
, const char *name
) {
2997 /* Returns true if the unit is inactive or going down */
2998 u
= manager_get_unit(m
, name
);
3002 return unit_inactive_or_pending(u
);
3005 static void manager_notify_finished(Manager
*m
) {
3006 char userspace
[FORMAT_TIMESPAN_MAX
], initrd
[FORMAT_TIMESPAN_MAX
], kernel
[FORMAT_TIMESPAN_MAX
], sum
[FORMAT_TIMESPAN_MAX
];
3007 usec_t firmware_usec
, loader_usec
, kernel_usec
, initrd_usec
, userspace_usec
, total_usec
;
3009 if (m
->test_run_flags
)
3012 if (MANAGER_IS_SYSTEM(m
) && detect_container() <= 0) {
3014 /* Note that m->kernel_usec.monotonic is always at 0,
3015 * and m->firmware_usec.monotonic and
3016 * m->loader_usec.monotonic should be considered
3017 * negative values. */
3019 firmware_usec
= m
->firmware_timestamp
.monotonic
- m
->loader_timestamp
.monotonic
;
3020 loader_usec
= m
->loader_timestamp
.monotonic
- m
->kernel_timestamp
.monotonic
;
3021 userspace_usec
= m
->finish_timestamp
.monotonic
- m
->userspace_timestamp
.monotonic
;
3022 total_usec
= m
->firmware_timestamp
.monotonic
+ m
->finish_timestamp
.monotonic
;
3024 if (dual_timestamp_is_set(&m
->initrd_timestamp
)) {
3026 kernel_usec
= m
->initrd_timestamp
.monotonic
- m
->kernel_timestamp
.monotonic
;
3027 initrd_usec
= m
->userspace_timestamp
.monotonic
- m
->initrd_timestamp
.monotonic
;
3029 log_struct(LOG_INFO
,
3030 "MESSAGE_ID=" SD_MESSAGE_STARTUP_FINISHED_STR
,
3031 "KERNEL_USEC="USEC_FMT
, kernel_usec
,
3032 "INITRD_USEC="USEC_FMT
, initrd_usec
,
3033 "USERSPACE_USEC="USEC_FMT
, userspace_usec
,
3034 LOG_MESSAGE("Startup finished in %s (kernel) + %s (initrd) + %s (userspace) = %s.",
3035 format_timespan(kernel
, sizeof(kernel
), kernel_usec
, USEC_PER_MSEC
),
3036 format_timespan(initrd
, sizeof(initrd
), initrd_usec
, USEC_PER_MSEC
),
3037 format_timespan(userspace
, sizeof(userspace
), userspace_usec
, USEC_PER_MSEC
),
3038 format_timespan(sum
, sizeof(sum
), total_usec
, USEC_PER_MSEC
)),
3041 kernel_usec
= m
->userspace_timestamp
.monotonic
- m
->kernel_timestamp
.monotonic
;
3044 log_struct(LOG_INFO
,
3045 "MESSAGE_ID=" SD_MESSAGE_STARTUP_FINISHED_STR
,
3046 "KERNEL_USEC="USEC_FMT
, kernel_usec
,
3047 "USERSPACE_USEC="USEC_FMT
, userspace_usec
,
3048 LOG_MESSAGE("Startup finished in %s (kernel) + %s (userspace) = %s.",
3049 format_timespan(kernel
, sizeof(kernel
), kernel_usec
, USEC_PER_MSEC
),
3050 format_timespan(userspace
, sizeof(userspace
), userspace_usec
, USEC_PER_MSEC
),
3051 format_timespan(sum
, sizeof(sum
), total_usec
, USEC_PER_MSEC
)),
3055 firmware_usec
= loader_usec
= initrd_usec
= kernel_usec
= 0;
3056 total_usec
= userspace_usec
= m
->finish_timestamp
.monotonic
- m
->userspace_timestamp
.monotonic
;
3058 log_struct(LOG_INFO
,
3059 "MESSAGE_ID=" SD_MESSAGE_USER_STARTUP_FINISHED_STR
,
3060 "USERSPACE_USEC="USEC_FMT
, userspace_usec
,
3061 LOG_MESSAGE("Startup finished in %s.",
3062 format_timespan(sum
, sizeof(sum
), total_usec
, USEC_PER_MSEC
)),
3066 bus_manager_send_finished(m
, firmware_usec
, loader_usec
, kernel_usec
, initrd_usec
, userspace_usec
, total_usec
);
3070 "STATUS=Startup finished in %s.",
3071 format_timespan(sum
, sizeof(sum
), total_usec
, USEC_PER_MSEC
));
3074 void manager_check_finished(Manager
*m
) {
3077 if (MANAGER_IS_RELOADING(m
))
3080 /* Verify that we are actually running currently. Initially
3081 * the exit code is set to invalid, and during operation it is
3082 * then set to MANAGER_OK */
3083 if (m
->exit_code
!= MANAGER_OK
)
3086 if (hashmap_size(m
->jobs
) > 0) {
3087 if (m
->jobs_in_progress_event_source
)
3088 /* Ignore any failure, this is only for feedback */
3089 (void) sd_event_source_set_time(m
->jobs_in_progress_event_source
, now(CLOCK_MONOTONIC
) + JOBS_IN_PROGRESS_WAIT_USEC
);
3094 manager_flip_auto_status(m
, false);
3096 /* Notify Type=idle units that we are done now */
3097 manager_close_idle_pipe(m
);
3099 /* Turn off confirm spawn now */
3100 m
->confirm_spawn
= NULL
;
3102 /* No need to update ask password status when we're going non-interactive */
3103 manager_close_ask_password(m
);
3105 /* This is no longer the first boot */
3106 manager_set_first_boot(m
, false);
3108 if (dual_timestamp_is_set(&m
->finish_timestamp
))
3111 dual_timestamp_get(&m
->finish_timestamp
);
3113 manager_notify_finished(m
);
3115 manager_invalidate_startup_units(m
);
3118 static bool generator_path_any(const char* const* paths
) {
3122 /* Optimize by skipping the whole process by not creating output directories
3123 * if no generators are found. */
3124 STRV_FOREACH(path
, (char**) paths
)
3125 if (access(*path
, F_OK
) == 0)
3127 else if (errno
!= ENOENT
)
3128 log_warning_errno(errno
, "Failed to open generator directory %s: %m", *path
);
3133 static const char* system_env_generator_binary_paths
[] = {
3134 "/run/systemd/system-environment-generators",
3135 "/etc/systemd/system-environment-generators",
3136 "/usr/local/lib/systemd/system-environment-generators",
3137 SYSTEM_ENV_GENERATOR_PATH
,
3141 static const char* user_env_generator_binary_paths
[] = {
3142 "/run/systemd/user-environment-generators",
3143 "/etc/systemd/user-environment-generators",
3144 "/usr/local/lib/systemd/user-environment-generators",
3145 USER_ENV_GENERATOR_PATH
,
3149 static int manager_run_environment_generators(Manager
*m
) {
3150 char **tmp
= NULL
; /* this is only used in the forked process, no cleanup here */
3152 void* args
[] = {&tmp
, &tmp
, &m
->environment
};
3154 if (m
->test_run_flags
&& !(m
->test_run_flags
& MANAGER_TEST_RUN_ENV_GENERATORS
))
3157 paths
= MANAGER_IS_SYSTEM(m
) ? system_env_generator_binary_paths
: user_env_generator_binary_paths
;
3159 if (!generator_path_any(paths
))
3162 return execute_directories(paths
, DEFAULT_TIMEOUT_USEC
, gather_environment
, args
, NULL
);
3165 static int manager_run_generators(Manager
*m
) {
3166 _cleanup_strv_free_
char **paths
= NULL
;
3167 const char *argv
[5];
3172 if (m
->test_run_flags
&& !(m
->test_run_flags
& MANAGER_TEST_RUN_GENERATORS
))
3175 paths
= generator_binary_paths(m
->unit_file_scope
);
3179 if (!generator_path_any((const char* const*) paths
))
3182 r
= lookup_paths_mkdir_generator(&m
->lookup_paths
);
3186 argv
[0] = NULL
; /* Leave this empty, execute_directory() will fill something in */
3187 argv
[1] = m
->lookup_paths
.generator
;
3188 argv
[2] = m
->lookup_paths
.generator_early
;
3189 argv
[3] = m
->lookup_paths
.generator_late
;
3192 RUN_WITH_UMASK(0022)
3193 execute_directories((const char* const*) paths
, DEFAULT_TIMEOUT_USEC
,
3194 NULL
, NULL
, (char**) argv
);
3197 lookup_paths_trim_generator(&m
->lookup_paths
);
3201 int manager_environment_add(Manager
*m
, char **minus
, char **plus
) {
3202 char **a
= NULL
, **b
= NULL
, **l
;
3207 if (!strv_isempty(minus
)) {
3208 a
= strv_env_delete(l
, 1, minus
);
3215 if (!strv_isempty(plus
)) {
3216 b
= strv_env_merge(2, l
, plus
);
3225 if (m
->environment
!= l
)
3226 strv_free(m
->environment
);
3233 manager_clean_environment(m
);
3234 strv_sort(m
->environment
);
3239 int manager_set_default_rlimits(Manager
*m
, struct rlimit
**default_rlimit
) {
3244 for (i
= 0; i
< _RLIMIT_MAX
; i
++) {
3245 m
->rlimit
[i
] = mfree(m
->rlimit
[i
]);
3247 if (!default_rlimit
[i
])
3250 m
->rlimit
[i
] = newdup(struct rlimit
, default_rlimit
[i
], 1);
3258 void manager_recheck_journal(Manager
*m
) {
3263 if (!MANAGER_IS_SYSTEM(m
))
3266 u
= manager_get_unit(m
, SPECIAL_JOURNALD_SOCKET
);
3267 if (u
&& SOCKET(u
)->state
!= SOCKET_RUNNING
) {
3268 log_close_journal();
3272 u
= manager_get_unit(m
, SPECIAL_JOURNALD_SERVICE
);
3273 if (u
&& SERVICE(u
)->state
!= SERVICE_RUNNING
) {
3274 log_close_journal();
3278 /* Hmm, OK, so the socket is fully up and the service is up
3279 * too, then let's make use of the thing. */
3283 void manager_set_show_status(Manager
*m
, ShowStatus mode
) {
3285 assert(IN_SET(mode
, SHOW_STATUS_AUTO
, SHOW_STATUS_NO
, SHOW_STATUS_YES
, SHOW_STATUS_TEMPORARY
));
3287 if (!MANAGER_IS_SYSTEM(m
))
3290 if (m
->show_status
!= mode
)
3291 log_debug("%s showing of status.",
3292 mode
== SHOW_STATUS_NO
? "Disabling" : "Enabling");
3293 m
->show_status
= mode
;
3296 (void) touch("/run/systemd/show-status");
3298 (void) unlink("/run/systemd/show-status");
3301 static bool manager_get_show_status(Manager
*m
, StatusType type
) {
3304 if (!MANAGER_IS_SYSTEM(m
))
3307 if (m
->no_console_output
)
3310 if (!IN_SET(manager_state(m
), MANAGER_INITIALIZING
, MANAGER_STARTING
, MANAGER_STOPPING
))
3313 /* If we cannot find out the status properly, just proceed. */
3314 if (type
!= STATUS_TYPE_EMERGENCY
&& manager_check_ask_password(m
) > 0)
3317 if (m
->show_status
> 0)
3323 const char *manager_get_confirm_spawn(Manager
*m
) {
3324 static int last_errno
= 0;
3325 const char *vc
= m
->confirm_spawn
;
3329 /* Here's the deal: we want to test the validity of the console but don't want
3330 * PID1 to go through the whole console process which might block. But we also
3331 * want to warn the user only once if something is wrong with the console so we
3332 * cannot do the sanity checks after spawning our children. So here we simply do
3333 * really basic tests to hopefully trap common errors.
3335 * If the console suddenly disappear at the time our children will really it
3336 * then they will simply fail to acquire it and a positive answer will be
3337 * assumed. New children will fallback to /dev/console though.
3339 * Note: TTYs are devices that can come and go any time, and frequently aren't
3340 * available yet during early boot (consider a USB rs232 dongle...). If for any
3341 * reason the configured console is not ready, we fallback to the default
3344 if (!vc
|| path_equal(vc
, "/dev/console"))
3351 if (!S_ISCHR(st
.st_mode
)) {
3359 if (last_errno
!= errno
) {
3361 log_warning_errno(errno
, "Failed to open %s: %m, using default console", vc
);
3363 return "/dev/console";
3366 void manager_set_first_boot(Manager
*m
, bool b
) {
3369 if (!MANAGER_IS_SYSTEM(m
))
3372 if (m
->first_boot
!= (int) b
) {
3374 (void) touch("/run/systemd/first-boot");
3376 (void) unlink("/run/systemd/first-boot");
3382 void manager_disable_confirm_spawn(void) {
3383 (void) touch("/run/systemd/confirm_spawn_disabled");
3386 bool manager_is_confirm_spawn_disabled(Manager
*m
) {
3387 if (!m
->confirm_spawn
)
3390 return access("/run/systemd/confirm_spawn_disabled", F_OK
) >= 0;
3393 void manager_status_printf(Manager
*m
, StatusType type
, const char *status
, const char *format
, ...) {
3396 /* If m is NULL, assume we're after shutdown and let the messages through. */
3398 if (m
&& !manager_get_show_status(m
, type
))
3401 /* XXX We should totally drop the check for ephemeral here
3402 * and thus effectively make 'Type=idle' pointless. */
3403 if (type
== STATUS_TYPE_EPHEMERAL
&& m
&& m
->n_on_console
> 0)
3406 va_start(ap
, format
);
3407 status_vprintf(status
, true, type
== STATUS_TYPE_EPHEMERAL
, format
, ap
);
3411 Set
*manager_get_units_requiring_mounts_for(Manager
*m
, const char *path
) {
3412 char p
[strlen(path
)+1];
3418 path_kill_slashes(p
);
3420 return hashmap_get(m
->units_requiring_mounts_for
, streq(p
, "/") ? "" : p
);
3423 void manager_set_exec_params(Manager
*m
, ExecParameters
*p
) {
3427 p
->environment
= m
->environment
;
3428 p
->confirm_spawn
= manager_get_confirm_spawn(m
);
3429 p
->cgroup_supported
= m
->cgroup_supported
;
3430 p
->prefix
= m
->prefix
;
3432 SET_FLAG(p
->flags
, EXEC_PASS_LOG_UNIT
|EXEC_CHOWN_DIRECTORIES
, MANAGER_IS_SYSTEM(m
));
3435 int manager_update_failed_units(Manager
*m
, Unit
*u
, bool failed
) {
3440 assert(u
->manager
== m
);
3442 size
= set_size(m
->failed_units
);
3445 r
= set_ensure_allocated(&m
->failed_units
, NULL
);
3449 if (set_put(m
->failed_units
, u
) < 0)
3452 (void) set_remove(m
->failed_units
, u
);
3454 if (set_size(m
->failed_units
) != size
)
3455 bus_manager_send_change_signal(m
);
3460 ManagerState
manager_state(Manager
*m
) {
3465 /* Did we ever finish booting? If not then we are still starting up */
3466 if (!dual_timestamp_is_set(&m
->finish_timestamp
)) {
3468 u
= manager_get_unit(m
, SPECIAL_BASIC_TARGET
);
3469 if (!u
|| !UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u
)))
3470 return MANAGER_INITIALIZING
;
3472 return MANAGER_STARTING
;
3475 /* Is the special shutdown target queued? If so, we are in shutdown state */
3476 u
= manager_get_unit(m
, SPECIAL_SHUTDOWN_TARGET
);
3477 if (u
&& u
->job
&& IN_SET(u
->job
->type
, JOB_START
, JOB_RESTART
, JOB_RELOAD_OR_START
))
3478 return MANAGER_STOPPING
;
3480 /* Are the rescue or emergency targets active or queued? If so we are in maintenance state */
3481 u
= manager_get_unit(m
, SPECIAL_RESCUE_TARGET
);
3482 if (u
&& (UNIT_IS_ACTIVE_OR_ACTIVATING(unit_active_state(u
)) ||
3483 (u
->job
&& IN_SET(u
->job
->type
, JOB_START
, JOB_RESTART
, JOB_RELOAD_OR_START
))))
3484 return MANAGER_MAINTENANCE
;
3486 u
= manager_get_unit(m
, SPECIAL_EMERGENCY_TARGET
);
3487 if (u
&& (UNIT_IS_ACTIVE_OR_ACTIVATING(unit_active_state(u
)) ||
3488 (u
->job
&& IN_SET(u
->job
->type
, JOB_START
, JOB_RESTART
, JOB_RELOAD_OR_START
))))
3489 return MANAGER_MAINTENANCE
;
3491 /* Are there any failed units? If so, we are in degraded mode */
3492 if (set_size(m
->failed_units
) > 0)
3493 return MANAGER_DEGRADED
;
3495 return MANAGER_RUNNING
;
3498 #define DESTROY_IPC_FLAG (UINT32_C(1) << 31)
3500 static void manager_unref_uid_internal(
3505 int (*_clean_ipc
)(uid_t uid
)) {
3511 assert(uid_is_valid(uid
));
3514 /* A generic implementation, covering both manager_unref_uid() and manager_unref_gid(), under the assumption
3515 * that uid_t and gid_t are actually defined the same way, with the same validity rules.
3517 * We store a hashmap where the UID/GID is they key and the value is a 32bit reference counter, whose highest
3518 * bit is used as flag for marking UIDs/GIDs whose IPC objects to remove when the last reference to the UID/GID
3519 * is dropped. The flag is set to on, once at least one reference from a unit where RemoveIPC= is set is added
3520 * on a UID/GID. It is reset when the UID's/GID's reference counter drops to 0 again. */
3522 assert_cc(sizeof(uid_t
) == sizeof(gid_t
));
3523 assert_cc(UID_INVALID
== (uid_t
) GID_INVALID
);
3525 if (uid
== 0) /* We don't keep track of root, and will never destroy it */
3528 c
= PTR_TO_UINT32(hashmap_get(*uid_refs
, UID_TO_PTR(uid
)));
3530 n
= c
& ~DESTROY_IPC_FLAG
;
3534 if (destroy_now
&& n
== 0) {
3535 hashmap_remove(*uid_refs
, UID_TO_PTR(uid
));
3537 if (c
& DESTROY_IPC_FLAG
) {
3538 log_debug("%s " UID_FMT
" is no longer referenced, cleaning up its IPC.",
3539 _clean_ipc
== clean_ipc_by_uid
? "UID" : "GID",
3541 (void) _clean_ipc(uid
);
3544 c
= n
| (c
& DESTROY_IPC_FLAG
);
3545 assert_se(hashmap_update(*uid_refs
, UID_TO_PTR(uid
), UINT32_TO_PTR(c
)) >= 0);
3549 void manager_unref_uid(Manager
*m
, uid_t uid
, bool destroy_now
) {
3550 manager_unref_uid_internal(m
, &m
->uid_refs
, uid
, destroy_now
, clean_ipc_by_uid
);
3553 void manager_unref_gid(Manager
*m
, gid_t gid
, bool destroy_now
) {
3554 manager_unref_uid_internal(m
, &m
->gid_refs
, (uid_t
) gid
, destroy_now
, clean_ipc_by_gid
);
3557 static int manager_ref_uid_internal(
3568 assert(uid_is_valid(uid
));
3570 /* A generic implementation, covering both manager_ref_uid() and manager_ref_gid(), under the assumption
3571 * that uid_t and gid_t are actually defined the same way, with the same validity rules. */
3573 assert_cc(sizeof(uid_t
) == sizeof(gid_t
));
3574 assert_cc(UID_INVALID
== (uid_t
) GID_INVALID
);
3576 if (uid
== 0) /* We don't keep track of root, and will never destroy it */
3579 r
= hashmap_ensure_allocated(uid_refs
, &trivial_hash_ops
);
3583 c
= PTR_TO_UINT32(hashmap_get(*uid_refs
, UID_TO_PTR(uid
)));
3585 n
= c
& ~DESTROY_IPC_FLAG
;
3588 if (n
& DESTROY_IPC_FLAG
) /* check for overflow */
3591 c
= n
| (c
& DESTROY_IPC_FLAG
) | (clean_ipc
? DESTROY_IPC_FLAG
: 0);
3593 return hashmap_replace(*uid_refs
, UID_TO_PTR(uid
), UINT32_TO_PTR(c
));
3596 int manager_ref_uid(Manager
*m
, uid_t uid
, bool clean_ipc
) {
3597 return manager_ref_uid_internal(m
, &m
->uid_refs
, uid
, clean_ipc
);
3600 int manager_ref_gid(Manager
*m
, gid_t gid
, bool clean_ipc
) {
3601 return manager_ref_uid_internal(m
, &m
->gid_refs
, (uid_t
) gid
, clean_ipc
);
3604 static void manager_vacuum_uid_refs_internal(
3607 int (*_clean_ipc
)(uid_t uid
)) {
3616 HASHMAP_FOREACH_KEY(p
, k
, *uid_refs
, i
) {
3620 uid
= PTR_TO_UID(k
);
3621 c
= PTR_TO_UINT32(p
);
3623 n
= c
& ~DESTROY_IPC_FLAG
;
3627 if (c
& DESTROY_IPC_FLAG
) {
3628 log_debug("Found unreferenced %s " UID_FMT
" after reload/reexec. Cleaning up.",
3629 _clean_ipc
== clean_ipc_by_uid
? "UID" : "GID",
3631 (void) _clean_ipc(uid
);
3634 assert_se(hashmap_remove(*uid_refs
, k
) == p
);
3638 void manager_vacuum_uid_refs(Manager
*m
) {
3639 manager_vacuum_uid_refs_internal(m
, &m
->uid_refs
, clean_ipc_by_uid
);
3642 void manager_vacuum_gid_refs(Manager
*m
) {
3643 manager_vacuum_uid_refs_internal(m
, &m
->gid_refs
, clean_ipc_by_gid
);
3646 static void manager_serialize_uid_refs_internal(
3650 const char *field_name
) {
3660 /* Serialize the UID reference table. Or actually, just the IPC destruction flag of it, as the actual counter
3661 * of it is better rebuild after a reload/reexec. */
3663 HASHMAP_FOREACH_KEY(p
, k
, *uid_refs
, i
) {
3667 uid
= PTR_TO_UID(k
);
3668 c
= PTR_TO_UINT32(p
);
3670 if (!(c
& DESTROY_IPC_FLAG
))
3673 fprintf(f
, "%s=" UID_FMT
"\n", field_name
, uid
);
3677 void manager_serialize_uid_refs(Manager
*m
, FILE *f
) {
3678 manager_serialize_uid_refs_internal(m
, f
, &m
->uid_refs
, "destroy-ipc-uid");
3681 void manager_serialize_gid_refs(Manager
*m
, FILE *f
) {
3682 manager_serialize_uid_refs_internal(m
, f
, &m
->gid_refs
, "destroy-ipc-gid");
3685 static void manager_deserialize_uid_refs_one_internal(
3688 const char *value
) {
3698 r
= parse_uid(value
, &uid
);
3699 if (r
< 0 || uid
== 0) {
3700 log_debug("Unable to parse UID reference serialization");
3704 r
= hashmap_ensure_allocated(uid_refs
, &trivial_hash_ops
);
3710 c
= PTR_TO_UINT32(hashmap_get(*uid_refs
, UID_TO_PTR(uid
)));
3711 if (c
& DESTROY_IPC_FLAG
)
3714 c
|= DESTROY_IPC_FLAG
;
3716 r
= hashmap_replace(*uid_refs
, UID_TO_PTR(uid
), UINT32_TO_PTR(c
));
3718 log_debug("Failed to add UID reference entry");
3723 void manager_deserialize_uid_refs_one(Manager
*m
, const char *value
) {
3724 manager_deserialize_uid_refs_one_internal(m
, &m
->uid_refs
, value
);
3727 void manager_deserialize_gid_refs_one(Manager
*m
, const char *value
) {
3728 manager_deserialize_uid_refs_one_internal(m
, &m
->gid_refs
, value
);
3731 int manager_dispatch_user_lookup_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
3735 char unit_name
[UNIT_NAME_MAX
+1];
3738 Manager
*m
= userdata
;
3746 /* Invoked whenever a child process succeeded resolving its user/group to use and sent us the resulting UID/GID
3747 * in a datagram. We parse the datagram here and pass it off to the unit, so that it can add a reference to the
3748 * UID/GID so that it can destroy the UID/GID's IPC objects when the reference counter drops to 0. */
3750 l
= recv(fd
, &buffer
, sizeof(buffer
), MSG_DONTWAIT
);
3752 if (errno
== EINTR
|| errno
== EAGAIN
)
3755 return log_error_errno(errno
, "Failed to read from user lookup fd: %m");
3758 if ((size_t) l
<= offsetof(struct buffer
, unit_name
)) {
3759 log_warning("Received too short user lookup message, ignoring.");
3763 if ((size_t) l
> offsetof(struct buffer
, unit_name
) + UNIT_NAME_MAX
) {
3764 log_warning("Received too long user lookup message, ignoring.");
3768 if (!uid_is_valid(buffer
.uid
) && !gid_is_valid(buffer
.gid
)) {
3769 log_warning("Got user lookup message with invalid UID/GID pair, ignoring.");
3773 n
= (size_t) l
- offsetof(struct buffer
, unit_name
);
3774 if (memchr(buffer
.unit_name
, 0, n
)) {
3775 log_warning("Received lookup message with embedded NUL character, ignoring.");
3779 buffer
.unit_name
[n
] = 0;
3780 u
= manager_get_unit(m
, buffer
.unit_name
);
3782 log_debug("Got user lookup message but unit doesn't exist, ignoring.");
3786 log_unit_debug(u
, "User lookup succeeded: uid=" UID_FMT
" gid=" GID_FMT
, buffer
.uid
, buffer
.gid
);
3788 unit_notify_user_lookup(u
, buffer
.uid
, buffer
.gid
);
3792 static const char *const manager_state_table
[_MANAGER_STATE_MAX
] = {
3793 [MANAGER_INITIALIZING
] = "initializing",
3794 [MANAGER_STARTING
] = "starting",
3795 [MANAGER_RUNNING
] = "running",
3796 [MANAGER_DEGRADED
] = "degraded",
3797 [MANAGER_MAINTENANCE
] = "maintenance",
3798 [MANAGER_STOPPING
] = "stopping",
3801 DEFINE_STRING_TABLE_LOOKUP(manager_state
, ManagerState
);