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
));
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;
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
, bool test_run
, 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
;
621 if (MANAGER_IS_SYSTEM(m
) && detect_container() <= 0)
622 boot_timestamps(&m
->userspace_timestamp
, &m
->firmware_timestamp
, &m
->loader_timestamp
);
625 /* Prepare log fields we can use for structured logging */
626 if (MANAGER_IS_SYSTEM(m
)) {
627 m
->unit_log_field
= "UNIT=";
628 m
->unit_log_format_string
= "UNIT=%s";
630 m
->invocation_log_field
= "INVOCATION_ID=";
631 m
->invocation_log_format_string
= "INVOCATION_ID=" SD_ID128_FORMAT_STR
;
633 m
->unit_log_field
= "USER_UNIT=";
634 m
->unit_log_format_string
= "USER_UNIT=%s";
636 m
->invocation_log_field
= "USER_INVOCATION_ID=";
637 m
->invocation_log_format_string
= "USER_INVOCATION_ID=" SD_ID128_FORMAT_STR
;
640 m
->idle_pipe
[0] = m
->idle_pipe
[1] = m
->idle_pipe
[2] = m
->idle_pipe
[3] = -1;
642 m
->pin_cgroupfs_fd
= m
->notify_fd
= m
->cgroups_agent_fd
= m
->signal_fd
= m
->time_change_fd
=
643 m
->dev_autofs_fd
= m
->private_listen_fd
= m
->cgroup_inotify_fd
=
644 m
->ask_password_inotify_fd
= -1;
646 m
->user_lookup_fds
[0] = m
->user_lookup_fds
[1] = -1;
648 m
->current_job_id
= 1; /* start as id #1, so that we can leave #0 around as "null-like" value */
650 m
->have_ask_password
= -EINVAL
; /* we don't know */
653 m
->test_run
= test_run
;
655 /* Reboot immediately if the user hits C-A-D more often than 7x per 2s */
656 RATELIMIT_INIT(m
->ctrl_alt_del_ratelimit
, 2 * USEC_PER_SEC
, 7);
658 r
= manager_default_environment(m
);
662 r
= hashmap_ensure_allocated(&m
->units
, &string_hash_ops
);
666 r
= hashmap_ensure_allocated(&m
->jobs
, NULL
);
670 r
= hashmap_ensure_allocated(&m
->cgroup_unit
, &string_hash_ops
);
674 r
= hashmap_ensure_allocated(&m
->watch_bus
, &string_hash_ops
);
678 r
= sd_event_default(&m
->event
);
682 r
= sd_event_add_defer(m
->event
, &m
->run_queue_event_source
, manager_dispatch_run_queue
, m
);
686 r
= sd_event_source_set_priority(m
->run_queue_event_source
, SD_EVENT_PRIORITY_IDLE
);
690 r
= sd_event_source_set_enabled(m
->run_queue_event_source
, SD_EVENT_OFF
);
694 (void) sd_event_source_set_description(m
->run_queue_event_source
, "manager-run-queue");
696 r
= manager_setup_signals(m
);
700 r
= manager_setup_cgroup(m
);
704 r
= manager_setup_time_change(m
);
708 m
->udev
= udev_new();
714 /* Note that we do not set up the notify fd here. We do that after deserialization,
715 * since they might have gotten serialized across the reexec. */
717 m
->taint_usr
= dir_is_empty("/usr") > 0;
719 r
= manager_setup_prefix(m
);
731 static int manager_setup_notify(Manager
*m
) {
737 if (m
->notify_fd
< 0) {
738 _cleanup_close_
int fd
= -1;
739 union sockaddr_union sa
= {
740 .sa
.sa_family
= AF_UNIX
,
742 static const int one
= 1;
744 /* First free all secondary fields */
745 m
->notify_socket
= mfree(m
->notify_socket
);
746 m
->notify_event_source
= sd_event_source_unref(m
->notify_event_source
);
748 fd
= socket(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
|SOCK_NONBLOCK
, 0);
750 return log_error_errno(errno
, "Failed to allocate notification socket: %m");
752 fd_inc_rcvbuf(fd
, NOTIFY_RCVBUF_SIZE
);
754 m
->notify_socket
= strappend(m
->prefix
[EXEC_DIRECTORY_RUNTIME
], "/systemd/notify");
755 if (!m
->notify_socket
)
758 (void) mkdir_parents_label(m
->notify_socket
, 0755);
759 (void) unlink(m
->notify_socket
);
761 strncpy(sa
.un
.sun_path
, m
->notify_socket
, sizeof(sa
.un
.sun_path
)-1);
762 r
= bind(fd
, &sa
.sa
, SOCKADDR_UN_LEN(sa
.un
));
764 return log_error_errno(errno
, "bind(%s) failed: %m", sa
.un
.sun_path
);
766 r
= setsockopt(fd
, SOL_SOCKET
, SO_PASSCRED
, &one
, sizeof(one
));
768 return log_error_errno(errno
, "SO_PASSCRED failed: %m");
773 log_debug("Using notification socket %s", m
->notify_socket
);
776 if (!m
->notify_event_source
) {
777 r
= sd_event_add_io(m
->event
, &m
->notify_event_source
, m
->notify_fd
, EPOLLIN
, manager_dispatch_notify_fd
, m
);
779 return log_error_errno(r
, "Failed to allocate notify event source: %m");
781 /* Process notification messages a bit earlier than SIGCHLD, so that we can still identify to which
782 * service an exit message belongs. */
783 r
= sd_event_source_set_priority(m
->notify_event_source
, SD_EVENT_PRIORITY_NORMAL
-7);
785 return log_error_errno(r
, "Failed to set priority of notify event source: %m");
787 (void) sd_event_source_set_description(m
->notify_event_source
, "manager-notify");
793 static int manager_setup_cgroups_agent(Manager
*m
) {
795 static const union sockaddr_union sa
= {
796 .un
.sun_family
= AF_UNIX
,
797 .un
.sun_path
= "/run/systemd/cgroups-agent",
801 /* This creates a listening socket we receive cgroups agent messages on. We do not use D-Bus for delivering
802 * these messages from the cgroups agent binary to PID 1, as the cgroups agent binary is very short-living, and
803 * each instance of it needs a new D-Bus connection. Since D-Bus connections are SOCK_STREAM/AF_UNIX, on
804 * overloaded systems the backlog of the D-Bus socket becomes relevant, as not more than the configured number
805 * of D-Bus connections may be queued until the kernel will start dropping further incoming connections,
806 * possibly resulting in lost cgroups agent messages. To avoid this, we'll use a private SOCK_DGRAM/AF_UNIX
807 * socket, where no backlog is relevant as communication may take place without an actual connect() cycle, and
808 * we thus won't lose messages.
810 * Note that PID 1 will forward the agent message to system bus, so that the user systemd instance may listen
811 * to it. The system instance hence listens on this special socket, but the user instances listen on the system
812 * bus for these messages. */
817 if (!MANAGER_IS_SYSTEM(m
))
820 r
= cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER
);
822 return log_error_errno(r
, "Failed to determine whether unified cgroups hierarchy is used: %m");
823 if (r
> 0) /* We don't need this anymore on the unified hierarchy */
826 if (m
->cgroups_agent_fd
< 0) {
827 _cleanup_close_
int fd
= -1;
829 /* First free all secondary fields */
830 m
->cgroups_agent_event_source
= sd_event_source_unref(m
->cgroups_agent_event_source
);
832 fd
= socket(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
|SOCK_NONBLOCK
, 0);
834 return log_error_errno(errno
, "Failed to allocate cgroups agent socket: %m");
836 fd_inc_rcvbuf(fd
, CGROUPS_AGENT_RCVBUF_SIZE
);
838 (void) unlink(sa
.un
.sun_path
);
840 /* Only allow root to connect to this socket */
842 r
= bind(fd
, &sa
.sa
, SOCKADDR_UN_LEN(sa
.un
));
844 return log_error_errno(errno
, "bind(%s) failed: %m", sa
.un
.sun_path
);
846 m
->cgroups_agent_fd
= fd
;
850 if (!m
->cgroups_agent_event_source
) {
851 r
= sd_event_add_io(m
->event
, &m
->cgroups_agent_event_source
, m
->cgroups_agent_fd
, EPOLLIN
, manager_dispatch_cgroups_agent_fd
, m
);
853 return log_error_errno(r
, "Failed to allocate cgroups agent event source: %m");
855 /* Process cgroups notifications early, but after having processed service notification messages or
856 * SIGCHLD signals, so that a cgroup running empty is always just the last safety net of notification,
857 * and we collected the metadata the notification and SIGCHLD stuff offers first. Also see handling of
858 * cgroup inotify for the unified cgroup stuff. */
859 r
= sd_event_source_set_priority(m
->cgroups_agent_event_source
, SD_EVENT_PRIORITY_NORMAL
-5);
861 return log_error_errno(r
, "Failed to set priority of cgroups agent event source: %m");
863 (void) sd_event_source_set_description(m
->cgroups_agent_event_source
, "manager-cgroups-agent");
869 static int manager_setup_user_lookup_fd(Manager
*m
) {
874 /* Set up the socket pair used for passing UID/GID resolution results from forked off processes to PID
875 * 1. Background: we can't do name lookups (NSS) from PID 1, since it might involve IPC and thus activation,
876 * and we might hence deadlock on ourselves. Hence we do all user/group lookups asynchronously from the forked
877 * off processes right before executing the binaries to start. In order to be able to clean up any IPC objects
878 * created by a unit (see RemoveIPC=) we need to know in PID 1 the used UID/GID of the executed processes,
879 * hence we establish this communication channel so that forked off processes can pass their UID/GID
880 * information back to PID 1. The forked off processes send their resolved UID/GID to PID 1 in a simple
881 * datagram, along with their unit name, so that we can share one communication socket pair among all units for
884 * You might wonder why we need a communication channel for this that is independent of the usual notification
885 * socket scheme (i.e. $NOTIFY_SOCKET). The primary difference is about trust: data sent via the $NOTIFY_SOCKET
886 * channel is only accepted if it originates from the right unit and if reception was enabled for it. The user
887 * lookup socket OTOH is only accessible by PID 1 and its children until they exec(), and always available.
889 * Note that this function is called under two circumstances: when we first initialize (in which case we
890 * allocate both the socket pair and the event source to listen on it), and when we deserialize after a reload
891 * (in which case the socket pair already exists but we still need to allocate the event source for it). */
893 if (m
->user_lookup_fds
[0] < 0) {
895 /* Free all secondary fields */
896 safe_close_pair(m
->user_lookup_fds
);
897 m
->user_lookup_event_source
= sd_event_source_unref(m
->user_lookup_event_source
);
899 if (socketpair(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
, 0, m
->user_lookup_fds
) < 0)
900 return log_error_errno(errno
, "Failed to allocate user lookup socket: %m");
902 (void) fd_inc_rcvbuf(m
->user_lookup_fds
[0], NOTIFY_RCVBUF_SIZE
);
905 if (!m
->user_lookup_event_source
) {
906 r
= sd_event_add_io(m
->event
, &m
->user_lookup_event_source
, m
->user_lookup_fds
[0], EPOLLIN
, manager_dispatch_user_lookup_fd
, m
);
908 return log_error_errno(errno
, "Failed to allocate user lookup event source: %m");
910 /* Process even earlier than the notify event source, so that we always know first about valid UID/GID
912 r
= sd_event_source_set_priority(m
->user_lookup_event_source
, SD_EVENT_PRIORITY_NORMAL
-8);
914 return log_error_errno(errno
, "Failed to set priority ot user lookup event source: %m");
916 (void) sd_event_source_set_description(m
->user_lookup_event_source
, "user-lookup");
922 static int manager_connect_bus(Manager
*m
, bool reexecuting
) {
923 bool try_bus_connect
;
932 (MANAGER_IS_USER(m
) && getenv("DBUS_SESSION_BUS_ADDRESS"));
934 /* Try to connect to the buses, if possible. */
935 return bus_init(m
, try_bus_connect
);
938 static unsigned manager_dispatch_cleanup_queue(Manager
*m
) {
944 while ((u
= m
->cleanup_queue
)) {
945 assert(u
->in_cleanup_queue
);
955 GC_OFFSET_IN_PATH
, /* This one is on the path we were traveling */
956 GC_OFFSET_UNSURE
, /* No clue */
957 GC_OFFSET_GOOD
, /* We still need this unit */
958 GC_OFFSET_BAD
, /* We don't need this unit anymore */
962 static void unit_gc_mark_good(Unit
*u
, unsigned gc_marker
) {
966 u
->gc_marker
= gc_marker
+ GC_OFFSET_GOOD
;
968 /* Recursively mark referenced units as GOOD as well */
969 SET_FOREACH(other
, u
->dependencies
[UNIT_REFERENCES
], i
)
970 if (other
->gc_marker
== gc_marker
+ GC_OFFSET_UNSURE
)
971 unit_gc_mark_good(other
, gc_marker
);
974 static void unit_gc_sweep(Unit
*u
, unsigned gc_marker
) {
981 if (u
->gc_marker
== gc_marker
+ GC_OFFSET_GOOD
||
982 u
->gc_marker
== gc_marker
+ GC_OFFSET_BAD
||
983 u
->gc_marker
== gc_marker
+ GC_OFFSET_UNSURE
||
984 u
->gc_marker
== gc_marker
+ GC_OFFSET_IN_PATH
)
987 if (u
->in_cleanup_queue
)
990 if (unit_check_gc(u
))
993 u
->gc_marker
= gc_marker
+ GC_OFFSET_IN_PATH
;
997 SET_FOREACH(other
, u
->dependencies
[UNIT_REFERENCED_BY
], i
) {
998 unit_gc_sweep(other
, gc_marker
);
1000 if (other
->gc_marker
== gc_marker
+ GC_OFFSET_GOOD
)
1003 if (other
->gc_marker
!= gc_marker
+ GC_OFFSET_BAD
)
1010 /* We were unable to find anything out about this entry, so
1011 * let's investigate it later */
1012 u
->gc_marker
= gc_marker
+ GC_OFFSET_UNSURE
;
1013 unit_add_to_gc_queue(u
);
1017 /* We definitely know that this one is not useful anymore, so
1018 * let's mark it for deletion */
1019 u
->gc_marker
= gc_marker
+ GC_OFFSET_BAD
;
1020 unit_add_to_cleanup_queue(u
);
1024 unit_gc_mark_good(u
, gc_marker
);
1027 static unsigned manager_dispatch_gc_unit_queue(Manager
*m
) {
1028 unsigned n
= 0, gc_marker
;
1033 /* log_debug("Running GC..."); */
1035 m
->gc_marker
+= _GC_OFFSET_MAX
;
1036 if (m
->gc_marker
+ _GC_OFFSET_MAX
<= _GC_OFFSET_MAX
)
1039 gc_marker
= m
->gc_marker
;
1041 while ((u
= m
->gc_unit_queue
)) {
1042 assert(u
->in_gc_queue
);
1044 unit_gc_sweep(u
, gc_marker
);
1046 LIST_REMOVE(gc_queue
, m
->gc_unit_queue
, u
);
1047 u
->in_gc_queue
= false;
1051 if (u
->gc_marker
== gc_marker
+ GC_OFFSET_BAD
||
1052 u
->gc_marker
== gc_marker
+ GC_OFFSET_UNSURE
) {
1054 log_unit_debug(u
, "Collecting.");
1055 u
->gc_marker
= gc_marker
+ GC_OFFSET_BAD
;
1056 unit_add_to_cleanup_queue(u
);
1063 static unsigned manager_dispatch_gc_job_queue(Manager
*m
) {
1069 while ((j
= m
->gc_job_queue
)) {
1070 assert(j
->in_gc_queue
);
1072 LIST_REMOVE(gc_queue
, m
->gc_job_queue
, j
);
1073 j
->in_gc_queue
= false;
1077 if (job_check_gc(j
))
1080 log_unit_debug(j
->unit
, "Collecting job.");
1081 (void) job_finish_and_invalidate(j
, JOB_COLLECTED
, false, false);
1087 static void manager_clear_jobs_and_units(Manager
*m
) {
1092 while ((u
= hashmap_first(m
->units
)))
1095 manager_dispatch_cleanup_queue(m
);
1097 assert(!m
->load_queue
);
1098 assert(!m
->run_queue
);
1099 assert(!m
->dbus_unit_queue
);
1100 assert(!m
->dbus_job_queue
);
1101 assert(!m
->cleanup_queue
);
1102 assert(!m
->gc_unit_queue
);
1103 assert(!m
->gc_job_queue
);
1105 assert(hashmap_isempty(m
->jobs
));
1106 assert(hashmap_isempty(m
->units
));
1108 m
->n_on_console
= 0;
1109 m
->n_running_jobs
= 0;
1112 Manager
* manager_free(Manager
*m
) {
1115 ExecDirectoryType dt
;
1120 manager_clear_jobs_and_units(m
);
1122 for (c
= 0; c
< _UNIT_TYPE_MAX
; c
++)
1123 if (unit_vtable
[c
]->shutdown
)
1124 unit_vtable
[c
]->shutdown(m
);
1126 /* If we reexecute ourselves, we keep the root cgroup
1128 manager_shutdown_cgroup(m
, m
->exit_code
!= MANAGER_REEXECUTE
);
1130 lookup_paths_flush_generator(&m
->lookup_paths
);
1134 dynamic_user_vacuum(m
, false);
1135 hashmap_free(m
->dynamic_users
);
1137 hashmap_free(m
->units
);
1138 hashmap_free(m
->units_by_invocation_id
);
1139 hashmap_free(m
->jobs
);
1140 hashmap_free(m
->watch_pids1
);
1141 hashmap_free(m
->watch_pids2
);
1142 hashmap_free(m
->watch_bus
);
1144 set_free(m
->startup_units
);
1145 set_free(m
->failed_units
);
1147 sd_event_source_unref(m
->signal_event_source
);
1148 sd_event_source_unref(m
->notify_event_source
);
1149 sd_event_source_unref(m
->cgroups_agent_event_source
);
1150 sd_event_source_unref(m
->time_change_event_source
);
1151 sd_event_source_unref(m
->jobs_in_progress_event_source
);
1152 sd_event_source_unref(m
->run_queue_event_source
);
1153 sd_event_source_unref(m
->user_lookup_event_source
);
1155 safe_close(m
->signal_fd
);
1156 safe_close(m
->notify_fd
);
1157 safe_close(m
->cgroups_agent_fd
);
1158 safe_close(m
->time_change_fd
);
1159 safe_close_pair(m
->user_lookup_fds
);
1161 manager_close_ask_password(m
);
1163 manager_close_idle_pipe(m
);
1165 udev_unref(m
->udev
);
1166 sd_event_unref(m
->event
);
1168 free(m
->notify_socket
);
1170 lookup_paths_free(&m
->lookup_paths
);
1171 strv_free(m
->environment
);
1173 hashmap_free(m
->cgroup_unit
);
1174 set_free_free(m
->unit_path_cache
);
1176 free(m
->switch_root
);
1177 free(m
->switch_root_init
);
1179 for (i
= 0; i
< _RLIMIT_MAX
; i
++)
1180 m
->rlimit
[i
] = mfree(m
->rlimit
[i
]);
1182 assert(hashmap_isempty(m
->units_requiring_mounts_for
));
1183 hashmap_free(m
->units_requiring_mounts_for
);
1185 hashmap_free(m
->uid_refs
);
1186 hashmap_free(m
->gid_refs
);
1188 for (dt
= 0; dt
< _EXEC_DIRECTORY_MAX
; dt
++)
1189 m
->prefix
[dt
] = mfree(m
->prefix
[dt
]);
1194 void manager_enumerate(Manager
*m
) {
1199 /* Let's ask every type to load all units from disk/kernel
1200 * that it might know */
1201 for (c
= 0; c
< _UNIT_TYPE_MAX
; c
++) {
1202 if (!unit_type_supported(c
)) {
1203 log_debug("Unit type .%s is not supported on this system.", unit_type_to_string(c
));
1207 if (!unit_vtable
[c
]->enumerate
)
1210 unit_vtable
[c
]->enumerate(m
);
1213 manager_dispatch_load_queue(m
);
1216 static void manager_coldplug(Manager
*m
) {
1224 /* Then, let's set up their initial state. */
1225 HASHMAP_FOREACH_KEY(u
, k
, m
->units
, i
) {
1227 /* ignore aliases */
1231 r
= unit_coldplug(u
);
1233 log_warning_errno(r
, "We couldn't coldplug %s, proceeding anyway: %m", u
->id
);
1237 static void manager_build_unit_path_cache(Manager
*m
) {
1243 set_free_free(m
->unit_path_cache
);
1245 m
->unit_path_cache
= set_new(&string_hash_ops
);
1246 if (!m
->unit_path_cache
) {
1251 /* This simply builds a list of files we know exist, so that
1252 * we don't always have to go to disk */
1254 STRV_FOREACH(i
, m
->lookup_paths
.search_path
) {
1255 _cleanup_closedir_
DIR *d
= NULL
;
1260 if (errno
!= ENOENT
)
1261 log_warning_errno(errno
, "Failed to open directory %s, ignoring: %m", *i
);
1265 FOREACH_DIRENT(de
, d
, r
= -errno
; goto fail
) {
1268 p
= strjoin(streq(*i
, "/") ? "" : *i
, "/", de
->d_name
);
1274 r
= set_consume(m
->unit_path_cache
, p
);
1283 log_warning_errno(r
, "Failed to build unit path cache, proceeding without: %m");
1284 m
->unit_path_cache
= set_free_free(m
->unit_path_cache
);
1287 static void manager_distribute_fds(Manager
*m
, FDSet
*fds
) {
1293 HASHMAP_FOREACH(u
, m
->units
, i
) {
1295 if (fdset_size(fds
) <= 0)
1298 if (!UNIT_VTABLE(u
)->distribute_fds
)
1301 UNIT_VTABLE(u
)->distribute_fds(u
, fds
);
1305 int manager_startup(Manager
*m
, FILE *serialization
, FDSet
*fds
) {
1310 r
= lookup_paths_init(&m
->lookup_paths
, m
->unit_file_scope
, 0, NULL
);
1314 r
= manager_run_environment_generators(m
);
1318 /* Make sure the transient directory always exists, so that it remains in the search path */
1320 r
= mkdir_p_label(m
->lookup_paths
.transient
, 0755);
1325 dual_timestamp_get(&m
->generators_start_timestamp
);
1326 r
= manager_run_generators(m
);
1327 dual_timestamp_get(&m
->generators_finish_timestamp
);
1331 lookup_paths_reduce(&m
->lookup_paths
);
1332 manager_build_unit_path_cache(m
);
1334 /* If we will deserialize make sure that during enumeration
1335 * this is already known, so we increase the counter here
1340 /* First, enumerate what we can from all config files */
1341 dual_timestamp_get(&m
->units_load_start_timestamp
);
1342 manager_enumerate(m
);
1343 dual_timestamp_get(&m
->units_load_finish_timestamp
);
1345 /* Second, deserialize if there is something to deserialize */
1346 if (serialization
) {
1347 r
= manager_deserialize(m
, serialization
, fds
);
1349 log_error_errno(r
, "Deserialization failed: %m");
1352 /* Any fds left? Find some unit which wants them. This is
1353 * useful to allow container managers to pass some file
1354 * descriptors to us pre-initialized. This enables
1355 * socket-based activation of entire containers. */
1356 manager_distribute_fds(m
, fds
);
1358 /* We might have deserialized the notify fd, but if we didn't
1359 * then let's create the bus now */
1360 q
= manager_setup_notify(m
);
1361 if (q
< 0 && r
== 0)
1364 q
= manager_setup_cgroups_agent(m
);
1365 if (q
< 0 && r
== 0)
1368 q
= manager_setup_user_lookup_fd(m
);
1369 if (q
< 0 && r
== 0)
1372 /* Let's connect to the bus now. */
1373 (void) manager_connect_bus(m
, !!serialization
);
1375 (void) bus_track_coldplug(m
, &m
->subscribed
, false, m
->deserialized_subscribed
);
1376 m
->deserialized_subscribed
= strv_free(m
->deserialized_subscribed
);
1378 /* Third, fire things up! */
1379 manager_coldplug(m
);
1381 /* Release any dynamic users no longer referenced */
1382 dynamic_user_vacuum(m
, true);
1384 /* Release any references to UIDs/GIDs no longer referenced, and destroy any IPC owned by them */
1385 manager_vacuum_uid_refs(m
);
1386 manager_vacuum_gid_refs(m
);
1388 if (serialization
) {
1389 assert(m
->n_reloading
> 0);
1392 /* Let's wait for the UnitNew/JobNew messages being
1393 * sent, before we notify that the reload is
1395 m
->send_reloading_done
= true;
1401 int manager_add_job(Manager
*m
, JobType type
, Unit
*unit
, JobMode mode
, sd_bus_error
*e
, Job
**_ret
) {
1406 assert(type
< _JOB_TYPE_MAX
);
1408 assert(mode
< _JOB_MODE_MAX
);
1410 if (mode
== JOB_ISOLATE
&& type
!= JOB_START
)
1411 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
, "Isolate is only valid for start.");
1413 if (mode
== JOB_ISOLATE
&& !unit
->allow_isolate
)
1414 return sd_bus_error_setf(e
, BUS_ERROR_NO_ISOLATION
, "Operation refused, unit may not be isolated.");
1416 log_unit_debug(unit
, "Trying to enqueue job %s/%s/%s", unit
->id
, job_type_to_string(type
), job_mode_to_string(mode
));
1418 type
= job_type_collapse(type
, unit
);
1420 tr
= transaction_new(mode
== JOB_REPLACE_IRREVERSIBLY
);
1424 r
= transaction_add_job_and_dependencies(tr
, type
, unit
, NULL
, true, false,
1425 mode
== JOB_IGNORE_DEPENDENCIES
|| mode
== JOB_IGNORE_REQUIREMENTS
,
1426 mode
== JOB_IGNORE_DEPENDENCIES
, e
);
1430 if (mode
== JOB_ISOLATE
) {
1431 r
= transaction_add_isolate_jobs(tr
, m
);
1436 r
= transaction_activate(tr
, m
, mode
, e
);
1440 log_unit_debug(unit
,
1441 "Enqueued job %s/%s as %u", unit
->id
,
1442 job_type_to_string(type
), (unsigned) tr
->anchor_job
->id
);
1445 *_ret
= tr
->anchor_job
;
1447 transaction_free(tr
);
1451 transaction_abort(tr
);
1452 transaction_free(tr
);
1456 int manager_add_job_by_name(Manager
*m
, JobType type
, const char *name
, JobMode mode
, sd_bus_error
*e
, Job
**ret
) {
1457 Unit
*unit
= NULL
; /* just to appease gcc, initialization is not really necessary */
1461 assert(type
< _JOB_TYPE_MAX
);
1463 assert(mode
< _JOB_MODE_MAX
);
1465 r
= manager_load_unit(m
, name
, NULL
, NULL
, &unit
);
1470 return manager_add_job(m
, type
, unit
, mode
, e
, ret
);
1473 int manager_add_job_by_name_and_warn(Manager
*m
, JobType type
, const char *name
, JobMode mode
, Job
**ret
) {
1474 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
1478 assert(type
< _JOB_TYPE_MAX
);
1480 assert(mode
< _JOB_MODE_MAX
);
1482 r
= manager_add_job_by_name(m
, type
, name
, mode
, &error
, ret
);
1484 return log_warning_errno(r
, "Failed to enqueue %s job for %s: %s", job_mode_to_string(mode
), name
, bus_error_message(&error
, r
));
1489 Job
*manager_get_job(Manager
*m
, uint32_t id
) {
1492 return hashmap_get(m
->jobs
, UINT32_TO_PTR(id
));
1495 Unit
*manager_get_unit(Manager
*m
, const char *name
) {
1499 return hashmap_get(m
->units
, name
);
1502 unsigned manager_dispatch_load_queue(Manager
*m
) {
1508 /* Make sure we are not run recursively */
1509 if (m
->dispatching_load_queue
)
1512 m
->dispatching_load_queue
= true;
1514 /* Dispatches the load queue. Takes a unit from the queue and
1515 * tries to load its data until the queue is empty */
1517 while ((u
= m
->load_queue
)) {
1518 assert(u
->in_load_queue
);
1524 m
->dispatching_load_queue
= false;
1528 int manager_load_unit_prepare(
1540 assert(name
|| path
);
1543 /* This will prepare the unit for loading, but not actually
1544 * load anything from disk. */
1546 if (path
&& !is_path(path
))
1547 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
, "Path %s is not absolute.", path
);
1550 name
= basename(path
);
1552 t
= unit_name_to_type(name
);
1554 if (t
== _UNIT_TYPE_INVALID
|| !unit_name_is_valid(name
, UNIT_NAME_PLAIN
|UNIT_NAME_INSTANCE
)) {
1555 if (unit_name_is_valid(name
, UNIT_NAME_TEMPLATE
))
1556 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
, "Unit name %s is missing the instance name.", name
);
1558 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
, "Unit name %s is not valid.", name
);
1561 ret
= manager_get_unit(m
, name
);
1567 ret
= unit_new(m
, unit_vtable
[t
]->object_size
);
1572 ret
->fragment_path
= strdup(path
);
1573 if (!ret
->fragment_path
) {
1579 r
= unit_add_name(ret
, name
);
1585 unit_add_to_load_queue(ret
);
1586 unit_add_to_dbus_queue(ret
);
1587 unit_add_to_gc_queue(ret
);
1594 int manager_load_unit(
1606 /* This will load the service information files, but not actually
1607 * start any services or anything. */
1609 r
= manager_load_unit_prepare(m
, name
, path
, e
, _ret
);
1613 manager_dispatch_load_queue(m
);
1615 *_ret
= unit_follow_merge(*_ret
);
1620 void manager_dump_jobs(Manager
*s
, FILE *f
, const char *prefix
) {
1627 HASHMAP_FOREACH(j
, s
->jobs
, i
)
1628 job_dump(j
, f
, prefix
);
1631 void manager_dump_units(Manager
*s
, FILE *f
, const char *prefix
) {
1639 HASHMAP_FOREACH_KEY(u
, t
, s
->units
, i
)
1641 unit_dump(u
, f
, prefix
);
1644 void manager_clear_jobs(Manager
*m
) {
1649 while ((j
= hashmap_first(m
->jobs
)))
1650 /* No need to recurse. We're cancelling all jobs. */
1651 job_finish_and_invalidate(j
, JOB_CANCELED
, false, false);
1654 static int manager_dispatch_run_queue(sd_event_source
*source
, void *userdata
) {
1655 Manager
*m
= userdata
;
1661 while ((j
= m
->run_queue
)) {
1662 assert(j
->installed
);
1663 assert(j
->in_run_queue
);
1665 job_run_and_invalidate(j
);
1668 if (m
->n_running_jobs
> 0)
1669 manager_watch_jobs_in_progress(m
);
1671 if (m
->n_on_console
> 0)
1672 manager_watch_idle_pipe(m
);
1677 static unsigned manager_dispatch_dbus_queue(Manager
*m
) {
1684 if (m
->dispatching_dbus_queue
)
1687 m
->dispatching_dbus_queue
= true;
1689 while ((u
= m
->dbus_unit_queue
)) {
1690 assert(u
->in_dbus_queue
);
1692 bus_unit_send_change_signal(u
);
1696 while ((j
= m
->dbus_job_queue
)) {
1697 assert(j
->in_dbus_queue
);
1699 bus_job_send_change_signal(j
);
1703 m
->dispatching_dbus_queue
= false;
1705 if (m
->send_reloading_done
) {
1706 m
->send_reloading_done
= false;
1708 bus_manager_send_reloading(m
, false);
1711 if (m
->queued_message
)
1712 bus_send_queued_message(m
);
1717 static int manager_dispatch_cgroups_agent_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
1718 Manager
*m
= userdata
;
1719 char buf
[PATH_MAX
+1];
1722 n
= recv(fd
, buf
, sizeof(buf
), 0);
1724 return log_error_errno(errno
, "Failed to read cgroups agent message: %m");
1726 log_error("Got zero-length cgroups agent message, ignoring.");
1729 if ((size_t) n
>= sizeof(buf
)) {
1730 log_error("Got overly long cgroups agent message, ignoring.");
1734 if (memchr(buf
, 0, n
)) {
1735 log_error("Got cgroups agent message with embedded NUL byte, ignoring.");
1740 manager_notify_cgroup_empty(m
, buf
);
1741 bus_forward_agent_released(m
, buf
);
1746 static void manager_invoke_notify_message(Manager
*m
, Unit
*u
, pid_t pid
, const char *buf
, FDSet
*fds
) {
1747 _cleanup_strv_free_
char **tags
= NULL
;
1753 tags
= strv_split(buf
, "\n\r");
1759 if (UNIT_VTABLE(u
)->notify_message
)
1760 UNIT_VTABLE(u
)->notify_message(u
, pid
, tags
, fds
);
1761 else if (_unlikely_(log_get_max_level() >= LOG_DEBUG
)) {
1762 _cleanup_free_
char *x
= NULL
, *y
= NULL
;
1766 y
= ellipsize(x
, 20, 90);
1767 log_unit_debug(u
, "Got notification message \"%s\", ignoring.", strnull(y
));
1771 static int manager_dispatch_notify_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
1773 _cleanup_fdset_free_ FDSet
*fds
= NULL
;
1774 Manager
*m
= userdata
;
1775 char buf
[NOTIFY_BUFFER_MAX
+1];
1776 struct iovec iovec
= {
1778 .iov_len
= sizeof(buf
)-1,
1781 struct cmsghdr cmsghdr
;
1782 uint8_t buf
[CMSG_SPACE(sizeof(struct ucred
)) +
1783 CMSG_SPACE(sizeof(int) * NOTIFY_FD_MAX
)];
1785 struct msghdr msghdr
= {
1788 .msg_control
= &control
,
1789 .msg_controllen
= sizeof(control
),
1792 struct cmsghdr
*cmsg
;
1793 struct ucred
*ucred
= NULL
;
1795 int r
, *fd_array
= NULL
;
1800 assert(m
->notify_fd
== fd
);
1802 if (revents
!= EPOLLIN
) {
1803 log_warning("Got unexpected poll event for notify fd.");
1807 n
= recvmsg(m
->notify_fd
, &msghdr
, MSG_DONTWAIT
|MSG_CMSG_CLOEXEC
|MSG_TRUNC
);
1809 if (IN_SET(errno
, EAGAIN
, EINTR
))
1810 return 0; /* Spurious wakeup, try again */
1812 /* If this is any other, real error, then let's stop processing this socket. This of course means we
1813 * won't take notification messages anymore, but that's still better than busy looping around this:
1814 * being woken up over and over again but being unable to actually read the message off the socket. */
1815 return log_error_errno(errno
, "Failed to receive notification message: %m");
1818 CMSG_FOREACH(cmsg
, &msghdr
) {
1819 if (cmsg
->cmsg_level
== SOL_SOCKET
&& cmsg
->cmsg_type
== SCM_RIGHTS
) {
1821 fd_array
= (int*) CMSG_DATA(cmsg
);
1822 n_fds
= (cmsg
->cmsg_len
- CMSG_LEN(0)) / sizeof(int);
1824 } else if (cmsg
->cmsg_level
== SOL_SOCKET
&&
1825 cmsg
->cmsg_type
== SCM_CREDENTIALS
&&
1826 cmsg
->cmsg_len
== CMSG_LEN(sizeof(struct ucred
))) {
1828 ucred
= (struct ucred
*) CMSG_DATA(cmsg
);
1835 r
= fdset_new_array(&fds
, fd_array
, n_fds
);
1837 close_many(fd_array
, n_fds
);
1843 if (!ucred
|| ucred
->pid
<= 0) {
1844 log_warning("Received notify message without valid credentials. Ignoring.");
1848 if ((size_t) n
>= sizeof(buf
) || (msghdr
.msg_flags
& MSG_TRUNC
)) {
1849 log_warning("Received notify message exceeded maximum size. Ignoring.");
1853 /* As extra safety check, let's make sure the string we get doesn't contain embedded NUL bytes. We permit one
1854 * trailing NUL byte in the message, but don't expect it. */
1855 if (n
> 1 && memchr(buf
, 0, n
-1)) {
1856 log_warning("Received notify message with embedded NUL bytes. Ignoring.");
1860 /* Make sure it's NUL-terminated. */
1863 /* Notify every unit that might be interested, but try
1864 * to avoid notifying the same one multiple times. */
1865 u1
= manager_get_unit_by_pid_cgroup(m
, ucred
->pid
);
1867 manager_invoke_notify_message(m
, u1
, ucred
->pid
, buf
, fds
);
1869 u2
= hashmap_get(m
->watch_pids1
, PID_TO_PTR(ucred
->pid
));
1871 manager_invoke_notify_message(m
, u2
, ucred
->pid
, buf
, fds
);
1873 u3
= hashmap_get(m
->watch_pids2
, PID_TO_PTR(ucred
->pid
));
1874 if (u3
&& u3
!= u2
&& u3
!= u1
)
1875 manager_invoke_notify_message(m
, u3
, ucred
->pid
, buf
, fds
);
1877 if (!u1
&& !u2
&& !u3
)
1878 log_warning("Cannot find unit for notify message of PID "PID_FMT
".", ucred
->pid
);
1880 if (fdset_size(fds
) > 0)
1881 log_warning("Got extra auxiliary fds with notification message, closing them.");
1886 static void invoke_sigchld_event(Manager
*m
, Unit
*u
, const siginfo_t
*si
) {
1893 sd_event_get_iteration(m
->event
, &iteration
);
1895 log_unit_debug(u
, "Child "PID_FMT
" belongs to %s", si
->si_pid
, u
->id
);
1897 unit_unwatch_pid(u
, si
->si_pid
);
1899 if (UNIT_VTABLE(u
)->sigchld_event
) {
1900 if (set_size(u
->pids
) <= 1 ||
1901 iteration
!= u
->sigchldgen
||
1902 unit_main_pid(u
) == si
->si_pid
||
1903 unit_control_pid(u
) == si
->si_pid
) {
1904 UNIT_VTABLE(u
)->sigchld_event(u
, si
->si_pid
, si
->si_code
, si
->si_status
);
1905 u
->sigchldgen
= iteration
;
1907 log_debug("%s already issued a sigchld this iteration %" PRIu64
", skipping. Pids still being watched %d", u
->id
, iteration
, set_size(u
->pids
));
1911 static int manager_dispatch_sigchld(Manager
*m
) {
1917 /* First we call waitd() for a PID and do not reap the
1918 * zombie. That way we can still access /proc/$PID for
1919 * it while it is a zombie. */
1920 if (waitid(P_ALL
, 0, &si
, WEXITED
|WNOHANG
|WNOWAIT
) < 0) {
1922 if (errno
== ECHILD
)
1934 if (si
.si_code
== CLD_EXITED
|| si
.si_code
== CLD_KILLED
|| si
.si_code
== CLD_DUMPED
) {
1935 _cleanup_free_
char *name
= NULL
;
1938 get_process_comm(si
.si_pid
, &name
);
1940 log_debug("Child "PID_FMT
" (%s) died (code=%s, status=%i/%s)",
1941 si
.si_pid
, strna(name
),
1942 sigchld_code_to_string(si
.si_code
),
1944 strna(si
.si_code
== CLD_EXITED
1945 ? exit_status_to_string(si
.si_status
, EXIT_STATUS_FULL
)
1946 : signal_to_string(si
.si_status
)));
1948 /* And now figure out the unit this belongs
1949 * to, it might be multiple... */
1950 u1
= manager_get_unit_by_pid_cgroup(m
, si
.si_pid
);
1952 invoke_sigchld_event(m
, u1
, &si
);
1953 u2
= hashmap_get(m
->watch_pids1
, PID_TO_PTR(si
.si_pid
));
1955 invoke_sigchld_event(m
, u2
, &si
);
1956 u3
= hashmap_get(m
->watch_pids2
, PID_TO_PTR(si
.si_pid
));
1957 if (u3
&& u3
!= u2
&& u3
!= u1
)
1958 invoke_sigchld_event(m
, u3
, &si
);
1961 /* And now, we actually reap the zombie. */
1962 if (waitid(P_PID
, si
.si_pid
, &si
, WEXITED
) < 0) {
1973 static int manager_start_target(Manager
*m
, const char *name
, JobMode mode
) {
1974 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
1977 log_debug("Activating special unit %s", name
);
1979 r
= manager_add_job_by_name(m
, JOB_START
, name
, mode
, &error
, NULL
);
1981 log_error("Failed to enqueue %s job: %s", name
, bus_error_message(&error
, r
));
1986 static void manager_handle_ctrl_alt_del(Manager
*m
) {
1987 /* If the user presses C-A-D more than
1988 * 7 times within 2s, we reboot/shutdown immediately,
1989 * unless it was disabled in system.conf */
1991 if (ratelimit_test(&m
->ctrl_alt_del_ratelimit
) || m
->cad_burst_action
== EMERGENCY_ACTION_NONE
)
1992 manager_start_target(m
, SPECIAL_CTRL_ALT_DEL_TARGET
, JOB_REPLACE_IRREVERSIBLY
);
1994 emergency_action(m
, m
->cad_burst_action
, NULL
,
1995 "Ctrl-Alt-Del was pressed more than 7 times within 2s");
1998 static int manager_dispatch_signal_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
1999 Manager
*m
= userdata
;
2001 struct signalfd_siginfo sfsi
;
2002 bool sigchld
= false;
2006 assert(m
->signal_fd
== fd
);
2008 if (revents
!= EPOLLIN
) {
2009 log_warning("Got unexpected events from signal file descriptor.");
2014 n
= read(m
->signal_fd
, &sfsi
, sizeof(sfsi
));
2015 if (n
!= sizeof(sfsi
)) {
2017 log_warning("Truncated read from signal fd (%zu bytes)!", n
);
2021 if (IN_SET(errno
, EINTR
, EAGAIN
))
2024 /* We return an error here, which will kill this handler,
2025 * to avoid a busy loop on read error. */
2026 return log_error_errno(errno
, "Reading from signal fd failed: %m");
2029 log_received_signal(sfsi
.ssi_signo
== SIGCHLD
||
2030 (sfsi
.ssi_signo
== SIGTERM
&& MANAGER_IS_USER(m
))
2031 ? LOG_DEBUG
: LOG_INFO
,
2034 switch (sfsi
.ssi_signo
) {
2041 if (MANAGER_IS_SYSTEM(m
)) {
2042 /* This is for compatibility with the
2043 * original sysvinit */
2044 r
= verify_run_space_and_log("Refusing to reexecute");
2046 m
->exit_code
= MANAGER_REEXECUTE
;
2053 if (MANAGER_IS_SYSTEM(m
)) {
2054 manager_handle_ctrl_alt_del(m
);
2058 /* Run the exit target if there is one, if not, just exit. */
2059 if (manager_start_target(m
, SPECIAL_EXIT_TARGET
, JOB_REPLACE
) < 0) {
2060 m
->exit_code
= MANAGER_EXIT
;
2067 if (MANAGER_IS_SYSTEM(m
))
2068 manager_start_target(m
, SPECIAL_KBREQUEST_TARGET
, JOB_REPLACE
);
2070 /* This is a nop on non-init */
2074 if (MANAGER_IS_SYSTEM(m
))
2075 manager_start_target(m
, SPECIAL_SIGPWR_TARGET
, JOB_REPLACE
);
2077 /* This is a nop on non-init */
2083 u
= manager_get_unit(m
, SPECIAL_DBUS_SERVICE
);
2085 if (!u
|| UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u
))) {
2086 log_info("Trying to reconnect to bus...");
2090 if (!u
|| !UNIT_IS_ACTIVE_OR_ACTIVATING(unit_active_state(u
))) {
2091 log_info("Loading D-Bus service...");
2092 manager_start_target(m
, SPECIAL_DBUS_SERVICE
, JOB_REPLACE
);
2099 _cleanup_free_
char *dump
= NULL
;
2100 _cleanup_fclose_
FILE *f
= NULL
;
2103 f
= open_memstream(&dump
, &size
);
2105 log_warning_errno(errno
, "Failed to allocate memory stream: %m");
2109 manager_dump_units(m
, f
, "\t");
2110 manager_dump_jobs(m
, f
, "\t");
2112 r
= fflush_and_check(f
);
2114 log_warning_errno(r
, "Failed to write status stream: %m");
2118 log_dump(LOG_INFO
, dump
);
2123 r
= verify_run_space_and_log("Refusing to reload");
2125 m
->exit_code
= MANAGER_RELOAD
;
2130 /* Starting SIGRTMIN+0 */
2131 static const char * const target_table
[] = {
2132 [0] = SPECIAL_DEFAULT_TARGET
,
2133 [1] = SPECIAL_RESCUE_TARGET
,
2134 [2] = SPECIAL_EMERGENCY_TARGET
,
2135 [3] = SPECIAL_HALT_TARGET
,
2136 [4] = SPECIAL_POWEROFF_TARGET
,
2137 [5] = SPECIAL_REBOOT_TARGET
,
2138 [6] = SPECIAL_KEXEC_TARGET
2141 /* Starting SIGRTMIN+13, so that target halt and system halt are 10 apart */
2142 static const ManagerExitCode code_table
[] = {
2144 [1] = MANAGER_POWEROFF
,
2145 [2] = MANAGER_REBOOT
,
2149 if ((int) sfsi
.ssi_signo
>= SIGRTMIN
+0 &&
2150 (int) sfsi
.ssi_signo
< SIGRTMIN
+(int) ELEMENTSOF(target_table
)) {
2151 int idx
= (int) sfsi
.ssi_signo
- SIGRTMIN
;
2152 manager_start_target(m
, target_table
[idx
],
2153 (idx
== 1 || idx
== 2) ? JOB_ISOLATE
: JOB_REPLACE
);
2157 if ((int) sfsi
.ssi_signo
>= SIGRTMIN
+13 &&
2158 (int) sfsi
.ssi_signo
< SIGRTMIN
+13+(int) ELEMENTSOF(code_table
)) {
2159 m
->exit_code
= code_table
[sfsi
.ssi_signo
- SIGRTMIN
- 13];
2163 switch (sfsi
.ssi_signo
- SIGRTMIN
) {
2166 manager_set_show_status(m
, SHOW_STATUS_YES
);
2170 manager_set_show_status(m
, SHOW_STATUS_NO
);
2174 log_set_max_level(LOG_DEBUG
);
2175 log_info("Setting log level to debug.");
2179 log_set_max_level(LOG_INFO
);
2180 log_info("Setting log level to info.");
2184 if (MANAGER_IS_USER(m
)) {
2185 m
->exit_code
= MANAGER_EXIT
;
2189 /* This is a nop on init */
2193 case 29: /* compatibility: used to be mapped to LOG_TARGET_SYSLOG_OR_KMSG */
2194 log_set_target(LOG_TARGET_JOURNAL_OR_KMSG
);
2195 log_notice("Setting log target to journal-or-kmsg.");
2199 log_set_target(LOG_TARGET_CONSOLE
);
2200 log_notice("Setting log target to console.");
2204 log_set_target(LOG_TARGET_KMSG
);
2205 log_notice("Setting log target to kmsg.");
2209 log_warning("Got unhandled signal <%s>.", signal_to_string(sfsi
.ssi_signo
));
2216 manager_dispatch_sigchld(m
);
2221 static int manager_dispatch_time_change_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
2222 Manager
*m
= userdata
;
2227 assert(m
->time_change_fd
== fd
);
2229 log_struct(LOG_DEBUG
,
2230 "MESSAGE_ID=" SD_MESSAGE_TIME_CHANGE_STR
,
2231 LOG_MESSAGE("Time has been changed"),
2234 /* Restart the watch */
2235 m
->time_change_event_source
= sd_event_source_unref(m
->time_change_event_source
);
2236 m
->time_change_fd
= safe_close(m
->time_change_fd
);
2238 manager_setup_time_change(m
);
2240 HASHMAP_FOREACH(u
, m
->units
, i
)
2241 if (UNIT_VTABLE(u
)->time_change
)
2242 UNIT_VTABLE(u
)->time_change(u
);
2247 static int manager_dispatch_idle_pipe_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
2248 Manager
*m
= userdata
;
2251 assert(m
->idle_pipe
[2] == fd
);
2253 m
->no_console_output
= m
->n_on_console
> 0;
2255 manager_close_idle_pipe(m
);
2260 static int manager_dispatch_jobs_in_progress(sd_event_source
*source
, usec_t usec
, void *userdata
) {
2261 Manager
*m
= userdata
;
2268 manager_print_jobs_in_progress(m
);
2270 next
= now(CLOCK_MONOTONIC
) + JOBS_IN_PROGRESS_PERIOD_USEC
;
2271 r
= sd_event_source_set_time(source
, next
);
2275 return sd_event_source_set_enabled(source
, SD_EVENT_ONESHOT
);
2278 int manager_loop(Manager
*m
) {
2281 RATELIMIT_DEFINE(rl
, 1*USEC_PER_SEC
, 50000);
2284 m
->exit_code
= MANAGER_OK
;
2286 /* Release the path cache */
2287 m
->unit_path_cache
= set_free_free(m
->unit_path_cache
);
2289 manager_check_finished(m
);
2291 /* There might still be some zombies hanging around from
2292 * before we were exec()'ed. Let's reap them. */
2293 r
= manager_dispatch_sigchld(m
);
2297 while (m
->exit_code
== MANAGER_OK
) {
2300 if (m
->runtime_watchdog
> 0 && m
->runtime_watchdog
!= USEC_INFINITY
&& MANAGER_IS_SYSTEM(m
))
2303 if (!ratelimit_test(&rl
)) {
2304 /* Yay, something is going seriously wrong, pause a little */
2305 log_warning("Looping too fast. Throttling execution a little.");
2309 if (manager_dispatch_load_queue(m
) > 0)
2312 if (manager_dispatch_gc_job_queue(m
) > 0)
2315 if (manager_dispatch_gc_unit_queue(m
) > 0)
2318 if (manager_dispatch_cleanup_queue(m
) > 0)
2321 if (manager_dispatch_cgroup_queue(m
) > 0)
2324 if (manager_dispatch_dbus_queue(m
) > 0)
2327 /* Sleep for half the watchdog time */
2328 if (m
->runtime_watchdog
> 0 && m
->runtime_watchdog
!= USEC_INFINITY
&& MANAGER_IS_SYSTEM(m
)) {
2329 wait_usec
= m
->runtime_watchdog
/ 2;
2333 wait_usec
= USEC_INFINITY
;
2335 r
= sd_event_run(m
->event
, wait_usec
);
2337 return log_error_errno(r
, "Failed to run event loop: %m");
2340 return m
->exit_code
;
2343 int manager_load_unit_from_dbus_path(Manager
*m
, const char *s
, sd_bus_error
*e
, Unit
**_u
) {
2344 _cleanup_free_
char *n
= NULL
;
2345 sd_id128_t invocation_id
;
2353 r
= unit_name_from_dbus_path(s
, &n
);
2357 /* Permit addressing units by invocation ID: if the passed bus path is suffixed by a 128bit ID then we use it
2358 * as invocation ID. */
2359 r
= sd_id128_from_string(n
, &invocation_id
);
2361 u
= hashmap_get(m
->units_by_invocation_id
, &invocation_id
);
2367 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
));
2370 /* If this didn't work, we check if this is a unit name */
2371 if (!unit_name_is_valid(n
, UNIT_NAME_PLAIN
|UNIT_NAME_INSTANCE
))
2372 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
, "Unit name %s is neither a valid invocation ID nor unit name.", n
);
2374 r
= manager_load_unit(m
, n
, NULL
, e
, &u
);
2382 int manager_get_job_from_dbus_path(Manager
*m
, const char *s
, Job
**_j
) {
2392 p
= startswith(s
, "/org/freedesktop/systemd1/job/");
2396 r
= safe_atou(p
, &id
);
2400 j
= manager_get_job(m
, id
);
2409 void manager_send_unit_audit(Manager
*m
, Unit
*u
, int type
, bool success
) {
2412 _cleanup_free_
char *p
= NULL
;
2416 if (!MANAGER_IS_SYSTEM(m
))
2419 audit_fd
= get_audit_fd();
2423 /* Don't generate audit events if the service was already
2424 * started and we're just deserializing */
2425 if (MANAGER_IS_RELOADING(m
))
2428 if (u
->type
!= UNIT_SERVICE
)
2431 r
= unit_name_to_prefix_and_instance(u
->id
, &p
);
2433 log_error_errno(r
, "Failed to extract prefix and instance of unit name: %m");
2437 msg
= strjoina("unit=", p
);
2438 if (audit_log_user_comm_message(audit_fd
, type
, msg
, "systemd", NULL
, NULL
, NULL
, success
) < 0) {
2440 /* We aren't allowed to send audit messages?
2441 * Then let's not retry again. */
2444 log_warning_errno(errno
, "Failed to send audit message: %m");
2450 void manager_send_unit_plymouth(Manager
*m
, Unit
*u
) {
2451 static const union sockaddr_union sa
= PLYMOUTH_SOCKET
;
2452 _cleanup_free_
char *message
= NULL
;
2453 _cleanup_close_
int fd
= -1;
2456 /* Don't generate plymouth events if the service was already
2457 * started and we're just deserializing */
2458 if (MANAGER_IS_RELOADING(m
))
2461 if (!MANAGER_IS_SYSTEM(m
))
2464 if (detect_container() > 0)
2467 if (u
->type
!= UNIT_SERVICE
&&
2468 u
->type
!= UNIT_MOUNT
&&
2469 u
->type
!= UNIT_SWAP
)
2472 /* We set SOCK_NONBLOCK here so that we rather drop the
2473 * message then wait for plymouth */
2474 fd
= socket(AF_UNIX
, SOCK_STREAM
|SOCK_CLOEXEC
|SOCK_NONBLOCK
, 0);
2476 log_error_errno(errno
, "socket() failed: %m");
2480 if (connect(fd
, &sa
.sa
, SOCKADDR_UN_LEN(sa
.un
)) < 0) {
2482 if (!IN_SET(errno
, EPIPE
, EAGAIN
, ENOENT
, ECONNREFUSED
, ECONNRESET
, ECONNABORTED
))
2483 log_error_errno(errno
, "connect() failed: %m");
2487 if (asprintf(&message
, "U\002%c%s%n", (int) (strlen(u
->id
) + 1), u
->id
, &n
) < 0) {
2493 if (write(fd
, message
, n
+ 1) != n
+ 1)
2494 if (!IN_SET(errno
, EPIPE
, EAGAIN
, ENOENT
, ECONNREFUSED
, ECONNRESET
, ECONNABORTED
))
2495 log_error_errno(errno
, "Failed to write Plymouth message: %m");
2498 int manager_open_serialization(Manager
*m
, FILE **_f
) {
2504 fd
= open_serialization_fd("systemd-state");
2508 f
= fdopen(fd
, "w+");
2518 int manager_serialize(Manager
*m
, FILE *f
, FDSet
*fds
, bool switching_root
) {
2530 fprintf(f
, "current-job-id=%"PRIu32
"\n", m
->current_job_id
);
2531 fprintf(f
, "taint-usr=%s\n", yes_no(m
->taint_usr
));
2532 fprintf(f
, "n-installed-jobs=%u\n", m
->n_installed_jobs
);
2533 fprintf(f
, "n-failed-jobs=%u\n", m
->n_failed_jobs
);
2535 dual_timestamp_serialize(f
, "firmware-timestamp", &m
->firmware_timestamp
);
2536 dual_timestamp_serialize(f
, "loader-timestamp", &m
->loader_timestamp
);
2537 dual_timestamp_serialize(f
, "kernel-timestamp", &m
->kernel_timestamp
);
2538 dual_timestamp_serialize(f
, "initrd-timestamp", &m
->initrd_timestamp
);
2541 dual_timestamp_serialize(f
, "userspace-timestamp", &m
->userspace_timestamp
);
2542 dual_timestamp_serialize(f
, "finish-timestamp", &m
->finish_timestamp
);
2543 dual_timestamp_serialize(f
, "security-start-timestamp", &m
->security_start_timestamp
);
2544 dual_timestamp_serialize(f
, "security-finish-timestamp", &m
->security_finish_timestamp
);
2545 dual_timestamp_serialize(f
, "generators-start-timestamp", &m
->generators_start_timestamp
);
2546 dual_timestamp_serialize(f
, "generators-finish-timestamp", &m
->generators_finish_timestamp
);
2547 dual_timestamp_serialize(f
, "units-load-start-timestamp", &m
->units_load_start_timestamp
);
2548 dual_timestamp_serialize(f
, "units-load-finish-timestamp", &m
->units_load_finish_timestamp
);
2551 if (!switching_root
)
2552 (void) serialize_environment(f
, m
->environment
);
2554 if (m
->notify_fd
>= 0) {
2557 copy
= fdset_put_dup(fds
, m
->notify_fd
);
2561 fprintf(f
, "notify-fd=%i\n", copy
);
2562 fprintf(f
, "notify-socket=%s\n", m
->notify_socket
);
2565 if (m
->cgroups_agent_fd
>= 0) {
2568 copy
= fdset_put_dup(fds
, m
->cgroups_agent_fd
);
2572 fprintf(f
, "cgroups-agent-fd=%i\n", copy
);
2575 if (m
->user_lookup_fds
[0] >= 0) {
2578 copy0
= fdset_put_dup(fds
, m
->user_lookup_fds
[0]);
2582 copy1
= fdset_put_dup(fds
, m
->user_lookup_fds
[1]);
2586 fprintf(f
, "user-lookup=%i %i\n", copy0
, copy1
);
2589 bus_track_serialize(m
->subscribed
, f
, "subscribed");
2591 r
= dynamic_user_serialize(m
, f
, fds
);
2595 manager_serialize_uid_refs(m
, f
);
2596 manager_serialize_gid_refs(m
, f
);
2598 fputc_unlocked('\n', f
);
2600 HASHMAP_FOREACH_KEY(u
, t
, m
->units
, i
) {
2605 fputs_unlocked(u
->id
, f
);
2606 fputc_unlocked('\n', f
);
2608 r
= unit_serialize(u
, f
, fds
, !switching_root
);
2615 assert(m
->n_reloading
> 0);
2621 r
= bus_fdset_add_all(m
, fds
);
2628 int manager_deserialize(Manager
*m
, FILE *f
, FDSet
*fds
) {
2634 log_debug("Deserializing state...");
2639 char line
[LINE_MAX
];
2640 const char *val
, *l
;
2642 if (!fgets(line
, sizeof(line
), f
)) {
2657 if ((val
= startswith(l
, "current-job-id="))) {
2660 if (safe_atou32(val
, &id
) < 0)
2661 log_notice("Failed to parse current job id value %s", val
);
2663 m
->current_job_id
= MAX(m
->current_job_id
, id
);
2665 } else if ((val
= startswith(l
, "n-installed-jobs="))) {
2668 if (safe_atou32(val
, &n
) < 0)
2669 log_notice("Failed to parse installed jobs counter %s", val
);
2671 m
->n_installed_jobs
+= n
;
2673 } else if ((val
= startswith(l
, "n-failed-jobs="))) {
2676 if (safe_atou32(val
, &n
) < 0)
2677 log_notice("Failed to parse failed jobs counter %s", val
);
2679 m
->n_failed_jobs
+= n
;
2681 } else if ((val
= startswith(l
, "taint-usr="))) {
2684 b
= parse_boolean(val
);
2686 log_notice("Failed to parse taint /usr flag %s", val
);
2688 m
->taint_usr
= m
->taint_usr
|| b
;
2690 } else if ((val
= startswith(l
, "firmware-timestamp=")))
2691 dual_timestamp_deserialize(val
, &m
->firmware_timestamp
);
2692 else if ((val
= startswith(l
, "loader-timestamp=")))
2693 dual_timestamp_deserialize(val
, &m
->loader_timestamp
);
2694 else if ((val
= startswith(l
, "kernel-timestamp=")))
2695 dual_timestamp_deserialize(val
, &m
->kernel_timestamp
);
2696 else if ((val
= startswith(l
, "initrd-timestamp=")))
2697 dual_timestamp_deserialize(val
, &m
->initrd_timestamp
);
2698 else if ((val
= startswith(l
, "userspace-timestamp=")))
2699 dual_timestamp_deserialize(val
, &m
->userspace_timestamp
);
2700 else if ((val
= startswith(l
, "finish-timestamp=")))
2701 dual_timestamp_deserialize(val
, &m
->finish_timestamp
);
2702 else if ((val
= startswith(l
, "security-start-timestamp=")))
2703 dual_timestamp_deserialize(val
, &m
->security_start_timestamp
);
2704 else if ((val
= startswith(l
, "security-finish-timestamp=")))
2705 dual_timestamp_deserialize(val
, &m
->security_finish_timestamp
);
2706 else if ((val
= startswith(l
, "generators-start-timestamp=")))
2707 dual_timestamp_deserialize(val
, &m
->generators_start_timestamp
);
2708 else if ((val
= startswith(l
, "generators-finish-timestamp=")))
2709 dual_timestamp_deserialize(val
, &m
->generators_finish_timestamp
);
2710 else if ((val
= startswith(l
, "units-load-start-timestamp=")))
2711 dual_timestamp_deserialize(val
, &m
->units_load_start_timestamp
);
2712 else if ((val
= startswith(l
, "units-load-finish-timestamp=")))
2713 dual_timestamp_deserialize(val
, &m
->units_load_finish_timestamp
);
2714 else if (startswith(l
, "env=")) {
2715 r
= deserialize_environment(&m
->environment
, l
);
2719 log_notice_errno(r
, "Failed to parse environment entry: \"%s\": %m", l
);
2721 } else if ((val
= startswith(l
, "notify-fd="))) {
2724 if (safe_atoi(val
, &fd
) < 0 || fd
< 0 || !fdset_contains(fds
, fd
))
2725 log_notice("Failed to parse notify fd: \"%s\"", val
);
2727 m
->notify_event_source
= sd_event_source_unref(m
->notify_event_source
);
2728 safe_close(m
->notify_fd
);
2729 m
->notify_fd
= fdset_remove(fds
, fd
);
2732 } else if ((val
= startswith(l
, "notify-socket="))) {
2741 free(m
->notify_socket
);
2742 m
->notify_socket
= n
;
2744 } else if ((val
= startswith(l
, "cgroups-agent-fd="))) {
2747 if (safe_atoi(val
, &fd
) < 0 || fd
< 0 || !fdset_contains(fds
, fd
))
2748 log_notice("Failed to parse cgroups agent fd: %s", val
);
2750 m
->cgroups_agent_event_source
= sd_event_source_unref(m
->cgroups_agent_event_source
);
2751 safe_close(m
->cgroups_agent_fd
);
2752 m
->cgroups_agent_fd
= fdset_remove(fds
, fd
);
2755 } else if ((val
= startswith(l
, "user-lookup="))) {
2758 if (sscanf(val
, "%i %i", &fd0
, &fd1
) != 2 || fd0
< 0 || fd1
< 0 || fd0
== fd1
|| !fdset_contains(fds
, fd0
) || !fdset_contains(fds
, fd1
))
2759 log_notice("Failed to parse user lookup fd: %s", val
);
2761 m
->user_lookup_event_source
= sd_event_source_unref(m
->user_lookup_event_source
);
2762 safe_close_pair(m
->user_lookup_fds
);
2763 m
->user_lookup_fds
[0] = fdset_remove(fds
, fd0
);
2764 m
->user_lookup_fds
[1] = fdset_remove(fds
, fd1
);
2767 } else if ((val
= startswith(l
, "dynamic-user=")))
2768 dynamic_user_deserialize_one(m
, val
, fds
);
2769 else if ((val
= startswith(l
, "destroy-ipc-uid=")))
2770 manager_deserialize_uid_refs_one(m
, val
);
2771 else if ((val
= startswith(l
, "destroy-ipc-gid=")))
2772 manager_deserialize_gid_refs_one(m
, val
);
2773 else if ((val
= startswith(l
, "subscribed="))) {
2775 if (strv_extend(&m
->deserialized_subscribed
, val
) < 0)
2778 } else if (!startswith(l
, "kdbus-fd=")) /* ignore this one */
2779 log_notice("Unknown serialization item '%s'", l
);
2784 char name
[UNIT_NAME_MAX
+2];
2785 const char* unit_name
;
2788 if (!fgets(name
, sizeof(name
), f
)) {
2798 unit_name
= strstrip(name
);
2800 r
= manager_load_unit(m
, unit_name
, NULL
, NULL
, &u
);
2802 log_notice_errno(r
, "Failed to load unit \"%s\", skipping deserialization: %m", unit_name
);
2805 unit_deserialize_skip(f
);
2809 r
= unit_deserialize(u
, f
, fds
);
2811 log_notice_errno(r
, "Failed to deserialize unit \"%s\": %m", unit_name
);
2821 assert(m
->n_reloading
> 0);
2827 int manager_reload(Manager
*m
) {
2829 _cleanup_fclose_
FILE *f
= NULL
;
2830 _cleanup_fdset_free_ FDSet
*fds
= NULL
;
2834 r
= manager_open_serialization(m
, &f
);
2839 bus_manager_send_reloading(m
, true);
2847 r
= manager_serialize(m
, f
, fds
, false);
2853 if (fseeko(f
, 0, SEEK_SET
) < 0) {
2858 /* From here on there is no way back. */
2859 manager_clear_jobs_and_units(m
);
2860 lookup_paths_flush_generator(&m
->lookup_paths
);
2861 lookup_paths_free(&m
->lookup_paths
);
2862 dynamic_user_vacuum(m
, false);
2863 m
->uid_refs
= hashmap_free(m
->uid_refs
);
2864 m
->gid_refs
= hashmap_free(m
->gid_refs
);
2866 q
= lookup_paths_init(&m
->lookup_paths
, m
->unit_file_scope
, 0, NULL
);
2867 if (q
< 0 && r
>= 0)
2870 q
= manager_run_environment_generators(m
);
2871 if (q
< 0 && r
>= 0)
2874 /* Find new unit paths */
2875 q
= manager_run_generators(m
);
2876 if (q
< 0 && r
>= 0)
2879 lookup_paths_reduce(&m
->lookup_paths
);
2880 manager_build_unit_path_cache(m
);
2882 /* First, enumerate what we can from all config files */
2883 manager_enumerate(m
);
2885 /* Second, deserialize our stored data */
2886 q
= manager_deserialize(m
, f
, fds
);
2888 log_error_errno(q
, "Deserialization failed: %m");
2897 /* Re-register notify_fd as event source */
2898 q
= manager_setup_notify(m
);
2899 if (q
< 0 && r
>= 0)
2902 q
= manager_setup_cgroups_agent(m
);
2903 if (q
< 0 && r
>= 0)
2906 q
= manager_setup_user_lookup_fd(m
);
2907 if (q
< 0 && r
>= 0)
2910 /* Third, fire things up! */
2911 manager_coldplug(m
);
2913 /* Release any dynamic users no longer referenced */
2914 dynamic_user_vacuum(m
, true);
2916 /* Release any references to UIDs/GIDs no longer referenced, and destroy any IPC owned by them */
2917 manager_vacuum_uid_refs(m
);
2918 manager_vacuum_gid_refs(m
);
2920 /* Sync current state of bus names with our set of listening units */
2922 manager_sync_bus_names(m
, m
->api_bus
);
2924 assert(m
->n_reloading
> 0);
2927 m
->send_reloading_done
= true;
2932 void manager_reset_failed(Manager
*m
) {
2938 HASHMAP_FOREACH(u
, m
->units
, i
)
2939 unit_reset_failed(u
);
2942 bool manager_unit_inactive_or_pending(Manager
*m
, const char *name
) {
2948 /* Returns true if the unit is inactive or going down */
2949 u
= manager_get_unit(m
, name
);
2953 return unit_inactive_or_pending(u
);
2956 static void manager_notify_finished(Manager
*m
) {
2957 char userspace
[FORMAT_TIMESPAN_MAX
], initrd
[FORMAT_TIMESPAN_MAX
], kernel
[FORMAT_TIMESPAN_MAX
], sum
[FORMAT_TIMESPAN_MAX
];
2958 usec_t firmware_usec
, loader_usec
, kernel_usec
, initrd_usec
, userspace_usec
, total_usec
;
2963 if (MANAGER_IS_SYSTEM(m
) && detect_container() <= 0) {
2965 /* Note that m->kernel_usec.monotonic is always at 0,
2966 * and m->firmware_usec.monotonic and
2967 * m->loader_usec.monotonic should be considered
2968 * negative values. */
2970 firmware_usec
= m
->firmware_timestamp
.monotonic
- m
->loader_timestamp
.monotonic
;
2971 loader_usec
= m
->loader_timestamp
.monotonic
- m
->kernel_timestamp
.monotonic
;
2972 userspace_usec
= m
->finish_timestamp
.monotonic
- m
->userspace_timestamp
.monotonic
;
2973 total_usec
= m
->firmware_timestamp
.monotonic
+ m
->finish_timestamp
.monotonic
;
2975 if (dual_timestamp_is_set(&m
->initrd_timestamp
)) {
2977 kernel_usec
= m
->initrd_timestamp
.monotonic
- m
->kernel_timestamp
.monotonic
;
2978 initrd_usec
= m
->userspace_timestamp
.monotonic
- m
->initrd_timestamp
.monotonic
;
2980 log_struct(LOG_INFO
,
2981 "MESSAGE_ID=" SD_MESSAGE_STARTUP_FINISHED_STR
,
2982 "KERNEL_USEC="USEC_FMT
, kernel_usec
,
2983 "INITRD_USEC="USEC_FMT
, initrd_usec
,
2984 "USERSPACE_USEC="USEC_FMT
, userspace_usec
,
2985 LOG_MESSAGE("Startup finished in %s (kernel) + %s (initrd) + %s (userspace) = %s.",
2986 format_timespan(kernel
, sizeof(kernel
), kernel_usec
, USEC_PER_MSEC
),
2987 format_timespan(initrd
, sizeof(initrd
), initrd_usec
, USEC_PER_MSEC
),
2988 format_timespan(userspace
, sizeof(userspace
), userspace_usec
, USEC_PER_MSEC
),
2989 format_timespan(sum
, sizeof(sum
), total_usec
, USEC_PER_MSEC
)),
2992 kernel_usec
= m
->userspace_timestamp
.monotonic
- m
->kernel_timestamp
.monotonic
;
2995 log_struct(LOG_INFO
,
2996 "MESSAGE_ID=" SD_MESSAGE_STARTUP_FINISHED_STR
,
2997 "KERNEL_USEC="USEC_FMT
, kernel_usec
,
2998 "USERSPACE_USEC="USEC_FMT
, userspace_usec
,
2999 LOG_MESSAGE("Startup finished in %s (kernel) + %s (userspace) = %s.",
3000 format_timespan(kernel
, sizeof(kernel
), kernel_usec
, USEC_PER_MSEC
),
3001 format_timespan(userspace
, sizeof(userspace
), userspace_usec
, USEC_PER_MSEC
),
3002 format_timespan(sum
, sizeof(sum
), total_usec
, USEC_PER_MSEC
)),
3006 firmware_usec
= loader_usec
= initrd_usec
= kernel_usec
= 0;
3007 total_usec
= userspace_usec
= m
->finish_timestamp
.monotonic
- m
->userspace_timestamp
.monotonic
;
3009 log_struct(LOG_INFO
,
3010 "MESSAGE_ID=" SD_MESSAGE_USER_STARTUP_FINISHED_STR
,
3011 "USERSPACE_USEC="USEC_FMT
, userspace_usec
,
3012 LOG_MESSAGE("Startup finished in %s.",
3013 format_timespan(sum
, sizeof(sum
), total_usec
, USEC_PER_MSEC
)),
3017 bus_manager_send_finished(m
, firmware_usec
, loader_usec
, kernel_usec
, initrd_usec
, userspace_usec
, total_usec
);
3021 "STATUS=Startup finished in %s.",
3022 format_timespan(sum
, sizeof(sum
), total_usec
, USEC_PER_MSEC
));
3025 void manager_check_finished(Manager
*m
) {
3028 if (MANAGER_IS_RELOADING(m
))
3031 /* Verify that we are actually running currently. Initially
3032 * the exit code is set to invalid, and during operation it is
3033 * then set to MANAGER_OK */
3034 if (m
->exit_code
!= MANAGER_OK
)
3037 if (hashmap_size(m
->jobs
) > 0) {
3038 if (m
->jobs_in_progress_event_source
)
3039 /* Ignore any failure, this is only for feedback */
3040 (void) sd_event_source_set_time(m
->jobs_in_progress_event_source
, now(CLOCK_MONOTONIC
) + JOBS_IN_PROGRESS_WAIT_USEC
);
3045 manager_flip_auto_status(m
, false);
3047 /* Notify Type=idle units that we are done now */
3048 manager_close_idle_pipe(m
);
3050 /* Turn off confirm spawn now */
3051 m
->confirm_spawn
= NULL
;
3053 /* No need to update ask password status when we're going non-interactive */
3054 manager_close_ask_password(m
);
3056 /* This is no longer the first boot */
3057 manager_set_first_boot(m
, false);
3059 if (dual_timestamp_is_set(&m
->finish_timestamp
))
3062 dual_timestamp_get(&m
->finish_timestamp
);
3064 manager_notify_finished(m
);
3066 manager_invalidate_startup_units(m
);
3069 static bool generator_path_any(const char* const* paths
) {
3073 /* Optimize by skipping the whole process by not creating output directories
3074 * if no generators are found. */
3075 STRV_FOREACH(path
, (char**) paths
)
3076 if (access(*path
, F_OK
) == 0)
3078 else if (errno
!= ENOENT
)
3079 log_warning_errno(errno
, "Failed to open generator directory %s: %m", *path
);
3084 static const char* system_env_generator_binary_paths
[] = {
3085 "/run/systemd/system-environment-generators",
3086 "/etc/systemd/system-environment-generators",
3087 "/usr/local/lib/systemd/system-environment-generators",
3088 SYSTEM_ENV_GENERATOR_PATH
,
3092 static const char* user_env_generator_binary_paths
[] = {
3093 "/run/systemd/user-environment-generators",
3094 "/etc/systemd/user-environment-generators",
3095 "/usr/local/lib/systemd/user-environment-generators",
3096 USER_ENV_GENERATOR_PATH
,
3100 static int manager_run_environment_generators(Manager
*m
) {
3101 char **tmp
= NULL
; /* this is only used in the forked process, no cleanup here */
3103 void* args
[] = {&tmp
, &tmp
, &m
->environment
};
3108 paths
= MANAGER_IS_SYSTEM(m
) ? system_env_generator_binary_paths
: user_env_generator_binary_paths
;
3110 if (!generator_path_any(paths
))
3113 return execute_directories(paths
, DEFAULT_TIMEOUT_USEC
, gather_environment
, args
, NULL
);
3116 static int manager_run_generators(Manager
*m
) {
3117 _cleanup_strv_free_
char **paths
= NULL
;
3118 const char *argv
[5];
3126 paths
= generator_binary_paths(m
->unit_file_scope
);
3130 if (!generator_path_any((const char* const*) paths
))
3133 r
= lookup_paths_mkdir_generator(&m
->lookup_paths
);
3137 argv
[0] = NULL
; /* Leave this empty, execute_directory() will fill something in */
3138 argv
[1] = m
->lookup_paths
.generator
;
3139 argv
[2] = m
->lookup_paths
.generator_early
;
3140 argv
[3] = m
->lookup_paths
.generator_late
;
3143 RUN_WITH_UMASK(0022)
3144 execute_directories((const char* const*) paths
, DEFAULT_TIMEOUT_USEC
,
3145 NULL
, NULL
, (char**) argv
);
3148 lookup_paths_trim_generator(&m
->lookup_paths
);
3152 int manager_environment_add(Manager
*m
, char **minus
, char **plus
) {
3153 char **a
= NULL
, **b
= NULL
, **l
;
3158 if (!strv_isempty(minus
)) {
3159 a
= strv_env_delete(l
, 1, minus
);
3166 if (!strv_isempty(plus
)) {
3167 b
= strv_env_merge(2, l
, plus
);
3176 if (m
->environment
!= l
)
3177 strv_free(m
->environment
);
3184 manager_clean_environment(m
);
3185 strv_sort(m
->environment
);
3190 int manager_set_default_rlimits(Manager
*m
, struct rlimit
**default_rlimit
) {
3195 for (i
= 0; i
< _RLIMIT_MAX
; i
++) {
3196 m
->rlimit
[i
] = mfree(m
->rlimit
[i
]);
3198 if (!default_rlimit
[i
])
3201 m
->rlimit
[i
] = newdup(struct rlimit
, default_rlimit
[i
], 1);
3209 void manager_recheck_journal(Manager
*m
) {
3214 if (!MANAGER_IS_SYSTEM(m
))
3217 u
= manager_get_unit(m
, SPECIAL_JOURNALD_SOCKET
);
3218 if (u
&& SOCKET(u
)->state
!= SOCKET_RUNNING
) {
3219 log_close_journal();
3223 u
= manager_get_unit(m
, SPECIAL_JOURNALD_SERVICE
);
3224 if (u
&& SERVICE(u
)->state
!= SERVICE_RUNNING
) {
3225 log_close_journal();
3229 /* Hmm, OK, so the socket is fully up and the service is up
3230 * too, then let's make use of the thing. */
3234 void manager_set_show_status(Manager
*m
, ShowStatus mode
) {
3236 assert(IN_SET(mode
, SHOW_STATUS_AUTO
, SHOW_STATUS_NO
, SHOW_STATUS_YES
, SHOW_STATUS_TEMPORARY
));
3238 if (!MANAGER_IS_SYSTEM(m
))
3241 if (m
->show_status
!= mode
)
3242 log_debug("%s showing of status.",
3243 mode
== SHOW_STATUS_NO
? "Disabling" : "Enabling");
3244 m
->show_status
= mode
;
3247 (void) touch("/run/systemd/show-status");
3249 (void) unlink("/run/systemd/show-status");
3252 static bool manager_get_show_status(Manager
*m
, StatusType type
) {
3255 if (!MANAGER_IS_SYSTEM(m
))
3258 if (m
->no_console_output
)
3261 if (!IN_SET(manager_state(m
), MANAGER_INITIALIZING
, MANAGER_STARTING
, MANAGER_STOPPING
))
3264 /* If we cannot find out the status properly, just proceed. */
3265 if (type
!= STATUS_TYPE_EMERGENCY
&& manager_check_ask_password(m
) > 0)
3268 if (m
->show_status
> 0)
3274 const char *manager_get_confirm_spawn(Manager
*m
) {
3275 static int last_errno
= 0;
3276 const char *vc
= m
->confirm_spawn
;
3280 /* Here's the deal: we want to test the validity of the console but don't want
3281 * PID1 to go through the whole console process which might block. But we also
3282 * want to warn the user only once if something is wrong with the console so we
3283 * cannot do the sanity checks after spawning our children. So here we simply do
3284 * really basic tests to hopefully trap common errors.
3286 * If the console suddenly disappear at the time our children will really it
3287 * then they will simply fail to acquire it and a positive answer will be
3288 * assumed. New children will fallback to /dev/console though.
3290 * Note: TTYs are devices that can come and go any time, and frequently aren't
3291 * available yet during early boot (consider a USB rs232 dongle...). If for any
3292 * reason the configured console is not ready, we fallback to the default
3295 if (!vc
|| path_equal(vc
, "/dev/console"))
3302 if (!S_ISCHR(st
.st_mode
)) {
3310 if (last_errno
!= errno
) {
3312 log_warning_errno(errno
, "Failed to open %s: %m, using default console", vc
);
3314 return "/dev/console";
3317 void manager_set_first_boot(Manager
*m
, bool b
) {
3320 if (!MANAGER_IS_SYSTEM(m
))
3323 if (m
->first_boot
!= (int) b
) {
3325 (void) touch("/run/systemd/first-boot");
3327 (void) unlink("/run/systemd/first-boot");
3333 void manager_disable_confirm_spawn(void) {
3334 (void) touch("/run/systemd/confirm_spawn_disabled");
3337 bool manager_is_confirm_spawn_disabled(Manager
*m
) {
3338 if (!m
->confirm_spawn
)
3341 return access("/run/systemd/confirm_spawn_disabled", F_OK
) >= 0;
3344 void manager_status_printf(Manager
*m
, StatusType type
, const char *status
, const char *format
, ...) {
3347 /* If m is NULL, assume we're after shutdown and let the messages through. */
3349 if (m
&& !manager_get_show_status(m
, type
))
3352 /* XXX We should totally drop the check for ephemeral here
3353 * and thus effectively make 'Type=idle' pointless. */
3354 if (type
== STATUS_TYPE_EPHEMERAL
&& m
&& m
->n_on_console
> 0)
3357 va_start(ap
, format
);
3358 status_vprintf(status
, true, type
== STATUS_TYPE_EPHEMERAL
, format
, ap
);
3362 Set
*manager_get_units_requiring_mounts_for(Manager
*m
, const char *path
) {
3363 char p
[strlen(path
)+1];
3369 path_kill_slashes(p
);
3371 return hashmap_get(m
->units_requiring_mounts_for
, streq(p
, "/") ? "" : p
);
3374 int manager_set_exec_params(Manager
*m
, ExecParameters
*p
) {
3378 p
->environment
= m
->environment
;
3379 p
->confirm_spawn
= manager_get_confirm_spawn(m
);
3380 p
->cgroup_supported
= m
->cgroup_supported
;
3381 p
->prefix
= m
->prefix
;
3386 int manager_update_failed_units(Manager
*m
, Unit
*u
, bool failed
) {
3391 assert(u
->manager
== m
);
3393 size
= set_size(m
->failed_units
);
3396 r
= set_ensure_allocated(&m
->failed_units
, NULL
);
3400 if (set_put(m
->failed_units
, u
) < 0)
3403 (void) set_remove(m
->failed_units
, u
);
3405 if (set_size(m
->failed_units
) != size
)
3406 bus_manager_send_change_signal(m
);
3411 ManagerState
manager_state(Manager
*m
) {
3416 /* Did we ever finish booting? If not then we are still starting up */
3417 if (!dual_timestamp_is_set(&m
->finish_timestamp
)) {
3419 u
= manager_get_unit(m
, SPECIAL_BASIC_TARGET
);
3420 if (!u
|| !UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u
)))
3421 return MANAGER_INITIALIZING
;
3423 return MANAGER_STARTING
;
3426 /* Is the special shutdown target queued? If so, we are in shutdown state */
3427 u
= manager_get_unit(m
, SPECIAL_SHUTDOWN_TARGET
);
3428 if (u
&& u
->job
&& IN_SET(u
->job
->type
, JOB_START
, JOB_RESTART
, JOB_RELOAD_OR_START
))
3429 return MANAGER_STOPPING
;
3431 /* Are the rescue or emergency targets active or queued? If so we are in maintenance state */
3432 u
= manager_get_unit(m
, SPECIAL_RESCUE_TARGET
);
3433 if (u
&& (UNIT_IS_ACTIVE_OR_ACTIVATING(unit_active_state(u
)) ||
3434 (u
->job
&& IN_SET(u
->job
->type
, JOB_START
, JOB_RESTART
, JOB_RELOAD_OR_START
))))
3435 return MANAGER_MAINTENANCE
;
3437 u
= manager_get_unit(m
, SPECIAL_EMERGENCY_TARGET
);
3438 if (u
&& (UNIT_IS_ACTIVE_OR_ACTIVATING(unit_active_state(u
)) ||
3439 (u
->job
&& IN_SET(u
->job
->type
, JOB_START
, JOB_RESTART
, JOB_RELOAD_OR_START
))))
3440 return MANAGER_MAINTENANCE
;
3442 /* Are there any failed units? If so, we are in degraded mode */
3443 if (set_size(m
->failed_units
) > 0)
3444 return MANAGER_DEGRADED
;
3446 return MANAGER_RUNNING
;
3449 #define DESTROY_IPC_FLAG (UINT32_C(1) << 31)
3451 static void manager_unref_uid_internal(
3456 int (*_clean_ipc
)(uid_t uid
)) {
3462 assert(uid_is_valid(uid
));
3465 /* A generic implementation, covering both manager_unref_uid() and manager_unref_gid(), under the assumption
3466 * that uid_t and gid_t are actually defined the same way, with the same validity rules.
3468 * We store a hashmap where the UID/GID is they key and the value is a 32bit reference counter, whose highest
3469 * bit is used as flag for marking UIDs/GIDs whose IPC objects to remove when the last reference to the UID/GID
3470 * is dropped. The flag is set to on, once at least one reference from a unit where RemoveIPC= is set is added
3471 * on a UID/GID. It is reset when the UID's/GID's reference counter drops to 0 again. */
3473 assert_cc(sizeof(uid_t
) == sizeof(gid_t
));
3474 assert_cc(UID_INVALID
== (uid_t
) GID_INVALID
);
3476 if (uid
== 0) /* We don't keep track of root, and will never destroy it */
3479 c
= PTR_TO_UINT32(hashmap_get(*uid_refs
, UID_TO_PTR(uid
)));
3481 n
= c
& ~DESTROY_IPC_FLAG
;
3485 if (destroy_now
&& n
== 0) {
3486 hashmap_remove(*uid_refs
, UID_TO_PTR(uid
));
3488 if (c
& DESTROY_IPC_FLAG
) {
3489 log_debug("%s " UID_FMT
" is no longer referenced, cleaning up its IPC.",
3490 _clean_ipc
== clean_ipc_by_uid
? "UID" : "GID",
3492 (void) _clean_ipc(uid
);
3495 c
= n
| (c
& DESTROY_IPC_FLAG
);
3496 assert_se(hashmap_update(*uid_refs
, UID_TO_PTR(uid
), UINT32_TO_PTR(c
)) >= 0);
3500 void manager_unref_uid(Manager
*m
, uid_t uid
, bool destroy_now
) {
3501 manager_unref_uid_internal(m
, &m
->uid_refs
, uid
, destroy_now
, clean_ipc_by_uid
);
3504 void manager_unref_gid(Manager
*m
, gid_t gid
, bool destroy_now
) {
3505 manager_unref_uid_internal(m
, &m
->gid_refs
, (uid_t
) gid
, destroy_now
, clean_ipc_by_gid
);
3508 static int manager_ref_uid_internal(
3519 assert(uid_is_valid(uid
));
3521 /* A generic implementation, covering both manager_ref_uid() and manager_ref_gid(), under the assumption
3522 * that uid_t and gid_t are actually defined the same way, with the same validity rules. */
3524 assert_cc(sizeof(uid_t
) == sizeof(gid_t
));
3525 assert_cc(UID_INVALID
== (uid_t
) GID_INVALID
);
3527 if (uid
== 0) /* We don't keep track of root, and will never destroy it */
3530 r
= hashmap_ensure_allocated(uid_refs
, &trivial_hash_ops
);
3534 c
= PTR_TO_UINT32(hashmap_get(*uid_refs
, UID_TO_PTR(uid
)));
3536 n
= c
& ~DESTROY_IPC_FLAG
;
3539 if (n
& DESTROY_IPC_FLAG
) /* check for overflow */
3542 c
= n
| (c
& DESTROY_IPC_FLAG
) | (clean_ipc
? DESTROY_IPC_FLAG
: 0);
3544 return hashmap_replace(*uid_refs
, UID_TO_PTR(uid
), UINT32_TO_PTR(c
));
3547 int manager_ref_uid(Manager
*m
, uid_t uid
, bool clean_ipc
) {
3548 return manager_ref_uid_internal(m
, &m
->uid_refs
, uid
, clean_ipc
);
3551 int manager_ref_gid(Manager
*m
, gid_t gid
, bool clean_ipc
) {
3552 return manager_ref_uid_internal(m
, &m
->gid_refs
, (uid_t
) gid
, clean_ipc
);
3555 static void manager_vacuum_uid_refs_internal(
3558 int (*_clean_ipc
)(uid_t uid
)) {
3567 HASHMAP_FOREACH_KEY(p
, k
, *uid_refs
, i
) {
3571 uid
= PTR_TO_UID(k
);
3572 c
= PTR_TO_UINT32(p
);
3574 n
= c
& ~DESTROY_IPC_FLAG
;
3578 if (c
& DESTROY_IPC_FLAG
) {
3579 log_debug("Found unreferenced %s " UID_FMT
" after reload/reexec. Cleaning up.",
3580 _clean_ipc
== clean_ipc_by_uid
? "UID" : "GID",
3582 (void) _clean_ipc(uid
);
3585 assert_se(hashmap_remove(*uid_refs
, k
) == p
);
3589 void manager_vacuum_uid_refs(Manager
*m
) {
3590 manager_vacuum_uid_refs_internal(m
, &m
->uid_refs
, clean_ipc_by_uid
);
3593 void manager_vacuum_gid_refs(Manager
*m
) {
3594 manager_vacuum_uid_refs_internal(m
, &m
->gid_refs
, clean_ipc_by_gid
);
3597 static void manager_serialize_uid_refs_internal(
3601 const char *field_name
) {
3611 /* Serialize the UID reference table. Or actually, just the IPC destruction flag of it, as the actual counter
3612 * of it is better rebuild after a reload/reexec. */
3614 HASHMAP_FOREACH_KEY(p
, k
, *uid_refs
, i
) {
3618 uid
= PTR_TO_UID(k
);
3619 c
= PTR_TO_UINT32(p
);
3621 if (!(c
& DESTROY_IPC_FLAG
))
3624 fprintf(f
, "%s=" UID_FMT
"\n", field_name
, uid
);
3628 void manager_serialize_uid_refs(Manager
*m
, FILE *f
) {
3629 manager_serialize_uid_refs_internal(m
, f
, &m
->uid_refs
, "destroy-ipc-uid");
3632 void manager_serialize_gid_refs(Manager
*m
, FILE *f
) {
3633 manager_serialize_uid_refs_internal(m
, f
, &m
->gid_refs
, "destroy-ipc-gid");
3636 static void manager_deserialize_uid_refs_one_internal(
3639 const char *value
) {
3649 r
= parse_uid(value
, &uid
);
3650 if (r
< 0 || uid
== 0) {
3651 log_debug("Unable to parse UID reference serialization");
3655 r
= hashmap_ensure_allocated(uid_refs
, &trivial_hash_ops
);
3661 c
= PTR_TO_UINT32(hashmap_get(*uid_refs
, UID_TO_PTR(uid
)));
3662 if (c
& DESTROY_IPC_FLAG
)
3665 c
|= DESTROY_IPC_FLAG
;
3667 r
= hashmap_replace(*uid_refs
, UID_TO_PTR(uid
), UINT32_TO_PTR(c
));
3669 log_debug("Failed to add UID reference entry");
3674 void manager_deserialize_uid_refs_one(Manager
*m
, const char *value
) {
3675 manager_deserialize_uid_refs_one_internal(m
, &m
->uid_refs
, value
);
3678 void manager_deserialize_gid_refs_one(Manager
*m
, const char *value
) {
3679 manager_deserialize_uid_refs_one_internal(m
, &m
->gid_refs
, value
);
3682 int manager_dispatch_user_lookup_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
3686 char unit_name
[UNIT_NAME_MAX
+1];
3689 Manager
*m
= userdata
;
3697 /* Invoked whenever a child process succeeded resolving its user/group to use and sent us the resulting UID/GID
3698 * in a datagram. We parse the datagram here and pass it off to the unit, so that it can add a reference to the
3699 * UID/GID so that it can destroy the UID/GID's IPC objects when the reference counter drops to 0. */
3701 l
= recv(fd
, &buffer
, sizeof(buffer
), MSG_DONTWAIT
);
3703 if (errno
== EINTR
|| errno
== EAGAIN
)
3706 return log_error_errno(errno
, "Failed to read from user lookup fd: %m");
3709 if ((size_t) l
<= offsetof(struct buffer
, unit_name
)) {
3710 log_warning("Received too short user lookup message, ignoring.");
3714 if ((size_t) l
> offsetof(struct buffer
, unit_name
) + UNIT_NAME_MAX
) {
3715 log_warning("Received too long user lookup message, ignoring.");
3719 if (!uid_is_valid(buffer
.uid
) && !gid_is_valid(buffer
.gid
)) {
3720 log_warning("Got user lookup message with invalid UID/GID pair, ignoring.");
3724 n
= (size_t) l
- offsetof(struct buffer
, unit_name
);
3725 if (memchr(buffer
.unit_name
, 0, n
)) {
3726 log_warning("Received lookup message with embedded NUL character, ignoring.");
3730 buffer
.unit_name
[n
] = 0;
3731 u
= manager_get_unit(m
, buffer
.unit_name
);
3733 log_debug("Got user lookup message but unit doesn't exist, ignoring.");
3737 log_unit_debug(u
, "User lookup succeeded: uid=" UID_FMT
" gid=" GID_FMT
, buffer
.uid
, buffer
.gid
);
3739 unit_notify_user_lookup(u
, buffer
.uid
, buffer
.gid
);
3743 static const char *const manager_state_table
[_MANAGER_STATE_MAX
] = {
3744 [MANAGER_INITIALIZING
] = "initializing",
3745 [MANAGER_STARTING
] = "starting",
3746 [MANAGER_RUNNING
] = "running",
3747 [MANAGER_DEGRADED
] = "degraded",
3748 [MANAGER_MAINTENANCE
] = "maintenance",
3749 [MANAGER_STOPPING
] = "stopping",
3752 DEFINE_STRING_TABLE_LOOKUP(manager_state
, ManagerState
);