1 /* SPDX-License-Identifier: LGPL-2.1+ */
3 This file is part of systemd.
5 Copyright 2010 Lennart Poettering
7 systemd is free software; you can redistribute it and/or modify it
8 under the terms of the GNU Lesser General Public License as published by
9 the Free Software Foundation; either version 2.1 of the License, or
10 (at your option) any later version.
12 systemd is distributed in the hope that it will be useful, but
13 WITHOUT ANY WARRANTY; without even the implied warranty of
14 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 Lesser General Public License for more details.
17 You should have received a copy of the GNU Lesser General Public License
18 along with systemd; If not, see <http://www.gnu.org/licenses/>.
25 #include <stdio_ext.h>
27 #include <sys/epoll.h>
28 #include <sys/inotify.h>
29 #include <sys/ioctl.h>
30 #include <sys/reboot.h>
31 #include <sys/timerfd.h>
39 #include "sd-daemon.h"
40 #include "sd-messages.h"
43 #include "alloc-util.h"
45 #include "boot-timestamps.h"
46 #include "bus-common-errors.h"
47 #include "bus-error.h"
48 #include "bus-kernel.h"
50 #include "clean-ipc.h"
51 #include "clock-util.h"
53 #include "dbus-manager.h"
54 #include "dbus-unit.h"
56 #include "dirent-util.h"
59 #include "exec-util.h"
61 #include "exit-status.h"
68 #include "locale-setup.h"
74 #include "parse-util.h"
75 #include "path-lookup.h"
76 #include "path-util.h"
77 #include "process-util.h"
78 #include "ratelimit.h"
80 #include "signal-util.h"
82 #include "stat-util.h"
83 #include "string-table.h"
84 #include "string-util.h"
87 #include "terminal-util.h"
88 #include "time-util.h"
89 #include "transaction.h"
90 #include "umask-util.h"
91 #include "unit-name.h"
92 #include "user-util.h"
97 #define NOTIFY_RCVBUF_SIZE (8*1024*1024)
98 #define CGROUPS_AGENT_RCVBUF_SIZE (8*1024*1024)
100 /* Initial delay and the interval for printing status messages about running jobs */
101 #define JOBS_IN_PROGRESS_WAIT_USEC (5*USEC_PER_SEC)
102 #define JOBS_IN_PROGRESS_PERIOD_USEC (USEC_PER_SEC / 3)
103 #define JOBS_IN_PROGRESS_PERIOD_DIVISOR 3
105 static int manager_dispatch_notify_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
106 static int manager_dispatch_cgroups_agent_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
107 static int manager_dispatch_signal_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
108 static int manager_dispatch_time_change_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
109 static int manager_dispatch_idle_pipe_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
110 static int manager_dispatch_user_lookup_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
111 static int manager_dispatch_jobs_in_progress(sd_event_source
*source
, usec_t usec
, void *userdata
);
112 static int manager_dispatch_run_queue(sd_event_source
*source
, void *userdata
);
113 static int manager_dispatch_sigchld(sd_event_source
*source
, void *userdata
);
114 static int manager_run_environment_generators(Manager
*m
);
115 static int manager_run_generators(Manager
*m
);
117 static void manager_watch_jobs_in_progress(Manager
*m
) {
123 /* We do not want to show the cylon animation if the user
124 * needs to confirm service executions otherwise confirmation
125 * messages will be screwed by the cylon animation. */
126 if (!manager_is_confirm_spawn_disabled(m
))
129 if (m
->jobs_in_progress_event_source
)
132 next
= now(CLOCK_MONOTONIC
) + JOBS_IN_PROGRESS_WAIT_USEC
;
133 r
= sd_event_add_time(
135 &m
->jobs_in_progress_event_source
,
138 manager_dispatch_jobs_in_progress
, m
);
142 (void) sd_event_source_set_description(m
->jobs_in_progress_event_source
, "manager-jobs-in-progress");
145 #define CYLON_BUFFER_EXTRA (2*STRLEN(ANSI_RED) + STRLEN(ANSI_HIGHLIGHT_RED) + 2*STRLEN(ANSI_NORMAL))
147 static void draw_cylon(char buffer
[], size_t buflen
, unsigned width
, unsigned pos
) {
150 assert(buflen
>= CYLON_BUFFER_EXTRA
+ width
+ 1);
151 assert(pos
<= width
+1); /* 0 or width+1 mean that the center light is behind the corner */
155 p
= mempset(p
, ' ', pos
-2);
156 if (log_get_show_color())
157 p
= stpcpy(p
, ANSI_RED
);
161 if (pos
> 0 && pos
<= width
) {
162 if (log_get_show_color())
163 p
= stpcpy(p
, ANSI_HIGHLIGHT_RED
);
167 if (log_get_show_color())
168 p
= stpcpy(p
, ANSI_NORMAL
);
171 if (log_get_show_color())
172 p
= stpcpy(p
, ANSI_RED
);
175 p
= mempset(p
, ' ', width
-1-pos
);
176 if (log_get_show_color())
177 strcpy(p
, ANSI_NORMAL
);
181 void manager_flip_auto_status(Manager
*m
, bool enable
) {
185 if (m
->show_status
== SHOW_STATUS_AUTO
)
186 manager_set_show_status(m
, SHOW_STATUS_TEMPORARY
);
188 if (m
->show_status
== SHOW_STATUS_TEMPORARY
)
189 manager_set_show_status(m
, SHOW_STATUS_AUTO
);
193 static void manager_print_jobs_in_progress(Manager
*m
) {
194 _cleanup_free_
char *job_of_n
= NULL
;
197 unsigned counter
= 0, print_nr
;
198 char cylon
[6 + CYLON_BUFFER_EXTRA
+ 1];
200 char time
[FORMAT_TIMESPAN_MAX
], limit
[FORMAT_TIMESPAN_MAX
] = "no limit";
204 assert(m
->n_running_jobs
> 0);
206 manager_flip_auto_status(m
, true);
208 print_nr
= (m
->jobs_in_progress_iteration
/ JOBS_IN_PROGRESS_PERIOD_DIVISOR
) % m
->n_running_jobs
;
210 HASHMAP_FOREACH(j
, m
->jobs
, i
)
211 if (j
->state
== JOB_RUNNING
&& counter
++ == print_nr
)
214 /* m->n_running_jobs must be consistent with the contents of m->jobs,
215 * so the above loop must have succeeded in finding j. */
216 assert(counter
== print_nr
+ 1);
219 cylon_pos
= m
->jobs_in_progress_iteration
% 14;
221 cylon_pos
= 14 - cylon_pos
;
222 draw_cylon(cylon
, sizeof(cylon
), 6, cylon_pos
);
224 m
->jobs_in_progress_iteration
++;
226 if (m
->n_running_jobs
> 1) {
227 if (asprintf(&job_of_n
, "(%u of %u) ", counter
, m
->n_running_jobs
) < 0)
231 format_timespan(time
, sizeof(time
), now(CLOCK_MONOTONIC
) - j
->begin_usec
, 1*USEC_PER_SEC
);
232 if (job_get_timeout(j
, &x
) > 0)
233 format_timespan(limit
, sizeof(limit
), x
- j
->begin_usec
, 1*USEC_PER_SEC
);
235 manager_status_printf(m
, STATUS_TYPE_EPHEMERAL
, cylon
,
236 "%sA %s job is running for %s (%s / %s)",
238 job_type_to_string(j
->type
),
239 unit_description(j
->unit
),
243 static int have_ask_password(void) {
244 _cleanup_closedir_
DIR *dir
;
247 dir
= opendir("/run/systemd/ask-password");
255 FOREACH_DIRENT_ALL(de
, dir
, return -errno
) {
256 if (startswith(de
->d_name
, "ask."))
262 static int manager_dispatch_ask_password_fd(sd_event_source
*source
,
263 int fd
, uint32_t revents
, void *userdata
) {
264 Manager
*m
= userdata
;
270 m
->have_ask_password
= have_ask_password();
271 if (m
->have_ask_password
< 0)
272 /* Log error but continue. Negative have_ask_password
273 * is treated as unknown status. */
274 log_error_errno(m
->have_ask_password
, "Failed to list /run/systemd/ask-password: %m");
279 static void manager_close_ask_password(Manager
*m
) {
282 m
->ask_password_event_source
= sd_event_source_unref(m
->ask_password_event_source
);
283 m
->ask_password_inotify_fd
= safe_close(m
->ask_password_inotify_fd
);
284 m
->have_ask_password
= -EINVAL
;
287 static int manager_check_ask_password(Manager
*m
) {
292 if (!m
->ask_password_event_source
) {
293 assert(m
->ask_password_inotify_fd
< 0);
295 mkdir_p_label("/run/systemd/ask-password", 0755);
297 m
->ask_password_inotify_fd
= inotify_init1(IN_NONBLOCK
|IN_CLOEXEC
);
298 if (m
->ask_password_inotify_fd
< 0)
299 return log_error_errno(errno
, "inotify_init1() failed: %m");
301 if (inotify_add_watch(m
->ask_password_inotify_fd
, "/run/systemd/ask-password", IN_CREATE
|IN_DELETE
|IN_MOVE
) < 0) {
302 log_error_errno(errno
, "Failed to add watch on /run/systemd/ask-password: %m");
303 manager_close_ask_password(m
);
307 r
= sd_event_add_io(m
->event
, &m
->ask_password_event_source
,
308 m
->ask_password_inotify_fd
, EPOLLIN
,
309 manager_dispatch_ask_password_fd
, m
);
311 log_error_errno(errno
, "Failed to add event source for /run/systemd/ask-password: %m");
312 manager_close_ask_password(m
);
316 (void) sd_event_source_set_description(m
->ask_password_event_source
, "manager-ask-password");
318 /* Queries might have been added meanwhile... */
319 manager_dispatch_ask_password_fd(m
->ask_password_event_source
,
320 m
->ask_password_inotify_fd
, EPOLLIN
, m
);
323 return m
->have_ask_password
;
326 static int manager_watch_idle_pipe(Manager
*m
) {
331 if (m
->idle_pipe_event_source
)
334 if (m
->idle_pipe
[2] < 0)
337 r
= sd_event_add_io(m
->event
, &m
->idle_pipe_event_source
, m
->idle_pipe
[2], EPOLLIN
, manager_dispatch_idle_pipe_fd
, m
);
339 return log_error_errno(r
, "Failed to watch idle pipe: %m");
341 (void) sd_event_source_set_description(m
->idle_pipe_event_source
, "manager-idle-pipe");
346 static void manager_close_idle_pipe(Manager
*m
) {
349 m
->idle_pipe_event_source
= sd_event_source_unref(m
->idle_pipe_event_source
);
351 safe_close_pair(m
->idle_pipe
);
352 safe_close_pair(m
->idle_pipe
+ 2);
355 static int manager_setup_time_change(Manager
*m
) {
358 /* We only care for the cancellation event, hence we set the
359 * timeout to the latest possible value. */
360 struct itimerspec its
= {
361 .it_value
.tv_sec
= TIME_T_MAX
,
365 assert_cc(sizeof(time_t) == sizeof(TIME_T_MAX
));
367 if (m
->test_run_flags
)
370 /* Uses TFD_TIMER_CANCEL_ON_SET to get notifications whenever
371 * CLOCK_REALTIME makes a jump relative to CLOCK_MONOTONIC */
373 m
->time_change_fd
= timerfd_create(CLOCK_REALTIME
, TFD_NONBLOCK
|TFD_CLOEXEC
);
374 if (m
->time_change_fd
< 0)
375 return log_error_errno(errno
, "Failed to create timerfd: %m");
377 if (timerfd_settime(m
->time_change_fd
, TFD_TIMER_ABSTIME
|TFD_TIMER_CANCEL_ON_SET
, &its
, NULL
) < 0) {
378 log_debug_errno(errno
, "Failed to set up TFD_TIMER_CANCEL_ON_SET, ignoring: %m");
379 m
->time_change_fd
= safe_close(m
->time_change_fd
);
383 r
= sd_event_add_io(m
->event
, &m
->time_change_event_source
, m
->time_change_fd
, EPOLLIN
, manager_dispatch_time_change_fd
, m
);
385 return log_error_errno(r
, "Failed to create time change event source: %m");
387 (void) sd_event_source_set_description(m
->time_change_event_source
, "manager-time-change");
389 log_debug("Set up TFD_TIMER_CANCEL_ON_SET timerfd.");
394 static int enable_special_signals(Manager
*m
) {
395 _cleanup_close_
int fd
= -1;
399 if (m
->test_run_flags
)
402 /* Enable that we get SIGINT on control-alt-del. In containers
403 * this will fail with EPERM (older) or EINVAL (newer), so
405 if (reboot(RB_DISABLE_CAD
) < 0 && !IN_SET(errno
, EPERM
, EINVAL
))
406 log_warning_errno(errno
, "Failed to enable ctrl-alt-del handling: %m");
408 fd
= open_terminal("/dev/tty0", O_RDWR
|O_NOCTTY
|O_CLOEXEC
);
410 /* Support systems without virtual console */
412 log_warning_errno(errno
, "Failed to open /dev/tty0: %m");
414 /* Enable that we get SIGWINCH on kbrequest */
415 if (ioctl(fd
, KDSIGACCEPT
, SIGWINCH
) < 0)
416 log_warning_errno(errno
, "Failed to enable kbrequest handling: %m");
422 static int manager_setup_signals(Manager
*m
) {
423 struct sigaction sa
= {
424 .sa_handler
= SIG_DFL
,
425 .sa_flags
= SA_NOCLDSTOP
|SA_RESTART
,
432 assert_se(sigaction(SIGCHLD
, &sa
, NULL
) == 0);
434 /* We make liberal use of realtime signals here. On
435 * Linux/glibc we have 30 of them (with the exception of Linux
436 * on hppa, see below), between SIGRTMIN+0 ... SIGRTMIN+30
439 assert_se(sigemptyset(&mask
) == 0);
440 sigset_add_many(&mask
,
441 SIGCHLD
, /* Child died */
442 SIGTERM
, /* Reexecute daemon */
443 SIGHUP
, /* Reload configuration */
444 SIGUSR1
, /* systemd/upstart: reconnect to D-Bus */
445 SIGUSR2
, /* systemd: dump status */
446 SIGINT
, /* Kernel sends us this on control-alt-del */
447 SIGWINCH
, /* Kernel sends us this on kbrequest (alt-arrowup) */
448 SIGPWR
, /* Some kernel drivers and upsd send us this on power failure */
450 SIGRTMIN
+0, /* systemd: start default.target */
451 SIGRTMIN
+1, /* systemd: isolate rescue.target */
452 SIGRTMIN
+2, /* systemd: isolate emergency.target */
453 SIGRTMIN
+3, /* systemd: start halt.target */
454 SIGRTMIN
+4, /* systemd: start poweroff.target */
455 SIGRTMIN
+5, /* systemd: start reboot.target */
456 SIGRTMIN
+6, /* systemd: start kexec.target */
458 /* ... space for more special targets ... */
460 SIGRTMIN
+13, /* systemd: Immediate halt */
461 SIGRTMIN
+14, /* systemd: Immediate poweroff */
462 SIGRTMIN
+15, /* systemd: Immediate reboot */
463 SIGRTMIN
+16, /* systemd: Immediate kexec */
465 /* ... space for more immediate system state changes ... */
467 SIGRTMIN
+20, /* systemd: enable status messages */
468 SIGRTMIN
+21, /* systemd: disable status messages */
469 SIGRTMIN
+22, /* systemd: set log level to LOG_DEBUG */
470 SIGRTMIN
+23, /* systemd: set log level to LOG_INFO */
471 SIGRTMIN
+24, /* systemd: Immediate exit (--user only) */
473 /* .. one free signal here ... */
475 #if !defined(__hppa64__) && !defined(__hppa__)
476 /* Apparently Linux on hppa has fewer RT
477 * signals (SIGRTMAX is SIGRTMIN+25 there),
478 * hence let's not try to make use of them
479 * here. Since these commands are accessible
480 * by different means and only really a safety
481 * net, the missing functionality on hppa
482 * shouldn't matter. */
484 SIGRTMIN
+26, /* systemd: set log target to journal-or-kmsg */
485 SIGRTMIN
+27, /* systemd: set log target to console */
486 SIGRTMIN
+28, /* systemd: set log target to kmsg */
487 SIGRTMIN
+29, /* systemd: set log target to syslog-or-kmsg (obsolete) */
489 /* ... one free signal here SIGRTMIN+30 ... */
492 assert_se(sigprocmask(SIG_SETMASK
, &mask
, NULL
) == 0);
494 m
->signal_fd
= signalfd(-1, &mask
, SFD_NONBLOCK
|SFD_CLOEXEC
);
495 if (m
->signal_fd
< 0)
498 r
= sd_event_add_io(m
->event
, &m
->signal_event_source
, m
->signal_fd
, EPOLLIN
, manager_dispatch_signal_fd
, m
);
502 (void) sd_event_source_set_description(m
->signal_event_source
, "manager-signal");
504 /* Process signals a bit earlier than the rest of things, but later than notify_fd processing, so that the
505 * notify processing can still figure out to which process/service a message belongs, before we reap the
506 * process. Also, process this before handling cgroup notifications, so that we always collect child exit
507 * status information before detecting that there's no process in a cgroup. */
508 r
= sd_event_source_set_priority(m
->signal_event_source
, SD_EVENT_PRIORITY_NORMAL
-6);
512 if (MANAGER_IS_SYSTEM(m
))
513 return enable_special_signals(m
);
518 static void manager_sanitize_environment(Manager
*m
) {
521 /* Let's remove some environment variables that we need ourselves to communicate with our clients */
541 /* Let's order the environment alphabetically, just to make it pretty */
542 strv_sort(m
->environment
);
545 static int manager_default_environment(Manager
*m
) {
548 if (MANAGER_IS_SYSTEM(m
)) {
549 /* The system manager always starts with a clean
550 * environment for its children. It does not import
551 * the kernel's or the parents' exported variables.
553 * The initial passed environment is untouched to keep
554 * /proc/self/environ valid; it is used for tagging
555 * the init process inside containers. */
556 m
->environment
= strv_new("PATH=" DEFAULT_PATH
,
559 /* Import locale variables LC_*= from configuration */
560 locale_setup(&m
->environment
);
562 /* The user manager passes its own environment
563 * along to its children. */
564 m
->environment
= strv_copy(environ
);
569 manager_sanitize_environment(m
);
574 static int manager_setup_prefix(Manager
*m
) {
580 static const struct table_entry paths_system
[_EXEC_DIRECTORY_TYPE_MAX
] = {
581 [EXEC_DIRECTORY_RUNTIME
] = { SD_PATH_SYSTEM_RUNTIME
, NULL
},
582 [EXEC_DIRECTORY_STATE
] = { SD_PATH_SYSTEM_STATE_PRIVATE
, NULL
},
583 [EXEC_DIRECTORY_CACHE
] = { SD_PATH_SYSTEM_STATE_CACHE
, NULL
},
584 [EXEC_DIRECTORY_LOGS
] = { SD_PATH_SYSTEM_STATE_LOGS
, NULL
},
585 [EXEC_DIRECTORY_CONFIGURATION
] = { SD_PATH_SYSTEM_CONFIGURATION
, NULL
},
588 static const struct table_entry paths_user
[_EXEC_DIRECTORY_TYPE_MAX
] = {
589 [EXEC_DIRECTORY_RUNTIME
] = { SD_PATH_USER_RUNTIME
, NULL
},
590 [EXEC_DIRECTORY_STATE
] = { SD_PATH_USER_CONFIGURATION
, NULL
},
591 [EXEC_DIRECTORY_CACHE
] = { SD_PATH_USER_STATE_CACHE
, NULL
},
592 [EXEC_DIRECTORY_LOGS
] = { SD_PATH_USER_CONFIGURATION
, "log" },
593 [EXEC_DIRECTORY_CONFIGURATION
] = { SD_PATH_USER_CONFIGURATION
, NULL
},
596 const struct table_entry
*p
;
602 if (MANAGER_IS_SYSTEM(m
))
607 for (i
= 0; i
< _EXEC_DIRECTORY_TYPE_MAX
; i
++) {
608 r
= sd_path_home(p
[i
].type
, p
[i
].suffix
, &m
->prefix
[i
]);
616 static int manager_setup_run_queue(Manager
*m
) {
620 assert(!m
->run_queue_event_source
);
622 r
= sd_event_add_defer(m
->event
, &m
->run_queue_event_source
, manager_dispatch_run_queue
, m
);
626 r
= sd_event_source_set_priority(m
->run_queue_event_source
, SD_EVENT_PRIORITY_IDLE
);
630 r
= sd_event_source_set_enabled(m
->run_queue_event_source
, SD_EVENT_OFF
);
634 (void) sd_event_source_set_description(m
->run_queue_event_source
, "manager-run-queue");
639 static int manager_setup_sigchld_event_source(Manager
*m
) {
643 assert(!m
->sigchld_event_source
);
645 r
= sd_event_add_defer(m
->event
, &m
->sigchld_event_source
, manager_dispatch_sigchld
, m
);
649 r
= sd_event_source_set_priority(m
->sigchld_event_source
, SD_EVENT_PRIORITY_NORMAL
-7);
653 r
= sd_event_source_set_enabled(m
->sigchld_event_source
, SD_EVENT_OFF
);
657 (void) sd_event_source_set_description(m
->sigchld_event_source
, "manager-sigchld");
662 int manager_new(UnitFileScope scope
, unsigned test_run_flags
, Manager
**_m
) {
667 assert(IN_SET(scope
, UNIT_FILE_SYSTEM
, UNIT_FILE_USER
));
669 m
= new0(Manager
, 1);
673 m
->unit_file_scope
= scope
;
674 m
->exit_code
= _MANAGER_EXIT_CODE_INVALID
;
675 m
->default_timer_accuracy_usec
= USEC_PER_MINUTE
;
676 m
->default_tasks_accounting
= true;
677 m
->default_tasks_max
= UINT64_MAX
;
678 m
->default_timeout_start_usec
= DEFAULT_TIMEOUT_USEC
;
679 m
->default_timeout_stop_usec
= DEFAULT_TIMEOUT_USEC
;
680 m
->default_restart_usec
= DEFAULT_RESTART_USEC
;
683 if (MANAGER_IS_SYSTEM(m
) && detect_container() <= 0)
684 boot_timestamps(m
->timestamps
+ MANAGER_TIMESTAMP_USERSPACE
,
685 m
->timestamps
+ MANAGER_TIMESTAMP_FIRMWARE
,
686 m
->timestamps
+ MANAGER_TIMESTAMP_LOADER
);
689 /* Prepare log fields we can use for structured logging */
690 if (MANAGER_IS_SYSTEM(m
)) {
691 m
->unit_log_field
= "UNIT=";
692 m
->unit_log_format_string
= "UNIT=%s";
694 m
->invocation_log_field
= "INVOCATION_ID=";
695 m
->invocation_log_format_string
= "INVOCATION_ID=%s";
697 m
->unit_log_field
= "USER_UNIT=";
698 m
->unit_log_format_string
= "USER_UNIT=%s";
700 m
->invocation_log_field
= "USER_INVOCATION_ID=";
701 m
->invocation_log_format_string
= "USER_INVOCATION_ID=%s";
704 m
->idle_pipe
[0] = m
->idle_pipe
[1] = m
->idle_pipe
[2] = m
->idle_pipe
[3] = -1;
706 m
->pin_cgroupfs_fd
= m
->notify_fd
= m
->cgroups_agent_fd
= m
->signal_fd
= m
->time_change_fd
=
707 m
->dev_autofs_fd
= m
->private_listen_fd
= m
->cgroup_inotify_fd
=
708 m
->ask_password_inotify_fd
= -1;
710 m
->user_lookup_fds
[0] = m
->user_lookup_fds
[1] = -1;
712 m
->current_job_id
= 1; /* start as id #1, so that we can leave #0 around as "null-like" value */
714 m
->have_ask_password
= -EINVAL
; /* we don't know */
717 m
->test_run_flags
= test_run_flags
;
719 /* Reboot immediately if the user hits C-A-D more often than 7x per 2s */
720 RATELIMIT_INIT(m
->ctrl_alt_del_ratelimit
, 2 * USEC_PER_SEC
, 7);
722 r
= manager_default_environment(m
);
726 r
= hashmap_ensure_allocated(&m
->units
, &string_hash_ops
);
730 r
= hashmap_ensure_allocated(&m
->jobs
, NULL
);
734 r
= hashmap_ensure_allocated(&m
->cgroup_unit
, &string_hash_ops
);
738 r
= hashmap_ensure_allocated(&m
->watch_bus
, &string_hash_ops
);
742 r
= sd_event_default(&m
->event
);
746 r
= manager_setup_run_queue(m
);
750 r
= manager_setup_signals(m
);
754 r
= manager_setup_cgroup(m
);
758 r
= manager_setup_time_change(m
);
762 r
= manager_setup_sigchld_event_source(m
);
766 m
->udev
= udev_new();
772 r
= manager_setup_prefix(m
);
776 if (MANAGER_IS_SYSTEM(m
) && test_run_flags
== 0) {
777 r
= mkdir_label("/run/systemd/units", 0755);
778 if (r
< 0 && r
!= -EEXIST
)
784 dir_is_empty("/usr") > 0;
786 /* Note that we do not set up the notify fd here. We do that after deserialization,
787 * since they might have gotten serialized across the reexec. */
797 static int manager_setup_notify(Manager
*m
) {
800 if (m
->test_run_flags
)
803 if (m
->notify_fd
< 0) {
804 _cleanup_close_
int fd
= -1;
805 union sockaddr_union sa
= {
806 .sa
.sa_family
= AF_UNIX
,
808 static const int one
= 1;
810 /* First free all secondary fields */
811 m
->notify_socket
= mfree(m
->notify_socket
);
812 m
->notify_event_source
= sd_event_source_unref(m
->notify_event_source
);
814 fd
= socket(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
|SOCK_NONBLOCK
, 0);
816 return log_error_errno(errno
, "Failed to allocate notification socket: %m");
818 fd_inc_rcvbuf(fd
, NOTIFY_RCVBUF_SIZE
);
820 m
->notify_socket
= strappend(m
->prefix
[EXEC_DIRECTORY_RUNTIME
], "/systemd/notify");
821 if (!m
->notify_socket
)
824 (void) mkdir_parents_label(m
->notify_socket
, 0755);
825 (void) unlink(m
->notify_socket
);
827 strncpy(sa
.un
.sun_path
, m
->notify_socket
, sizeof(sa
.un
.sun_path
)-1);
828 r
= bind(fd
, &sa
.sa
, SOCKADDR_UN_LEN(sa
.un
));
830 return log_error_errno(errno
, "bind(%s) failed: %m", sa
.un
.sun_path
);
832 r
= setsockopt(fd
, SOL_SOCKET
, SO_PASSCRED
, &one
, sizeof(one
));
834 return log_error_errno(errno
, "SO_PASSCRED failed: %m");
839 log_debug("Using notification socket %s", m
->notify_socket
);
842 if (!m
->notify_event_source
) {
843 r
= sd_event_add_io(m
->event
, &m
->notify_event_source
, m
->notify_fd
, EPOLLIN
, manager_dispatch_notify_fd
, m
);
845 return log_error_errno(r
, "Failed to allocate notify event source: %m");
847 /* Process notification messages a bit earlier than SIGCHLD, so that we can still identify to which
848 * service an exit message belongs. */
849 r
= sd_event_source_set_priority(m
->notify_event_source
, SD_EVENT_PRIORITY_NORMAL
-8);
851 return log_error_errno(r
, "Failed to set priority of notify event source: %m");
853 (void) sd_event_source_set_description(m
->notify_event_source
, "manager-notify");
859 static int manager_setup_cgroups_agent(Manager
*m
) {
861 static const union sockaddr_union sa
= {
862 .un
.sun_family
= AF_UNIX
,
863 .un
.sun_path
= "/run/systemd/cgroups-agent",
867 /* This creates a listening socket we receive cgroups agent messages on. We do not use D-Bus for delivering
868 * these messages from the cgroups agent binary to PID 1, as the cgroups agent binary is very short-living, and
869 * each instance of it needs a new D-Bus connection. Since D-Bus connections are SOCK_STREAM/AF_UNIX, on
870 * overloaded systems the backlog of the D-Bus socket becomes relevant, as not more than the configured number
871 * of D-Bus connections may be queued until the kernel will start dropping further incoming connections,
872 * possibly resulting in lost cgroups agent messages. To avoid this, we'll use a private SOCK_DGRAM/AF_UNIX
873 * socket, where no backlog is relevant as communication may take place without an actual connect() cycle, and
874 * we thus won't lose messages.
876 * Note that PID 1 will forward the agent message to system bus, so that the user systemd instance may listen
877 * to it. The system instance hence listens on this special socket, but the user instances listen on the system
878 * bus for these messages. */
880 if (m
->test_run_flags
)
883 if (!MANAGER_IS_SYSTEM(m
))
886 r
= cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER
);
888 return log_error_errno(r
, "Failed to determine whether unified cgroups hierarchy is used: %m");
889 if (r
> 0) /* We don't need this anymore on the unified hierarchy */
892 if (m
->cgroups_agent_fd
< 0) {
893 _cleanup_close_
int fd
= -1;
895 /* First free all secondary fields */
896 m
->cgroups_agent_event_source
= sd_event_source_unref(m
->cgroups_agent_event_source
);
898 fd
= socket(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
|SOCK_NONBLOCK
, 0);
900 return log_error_errno(errno
, "Failed to allocate cgroups agent socket: %m");
902 fd_inc_rcvbuf(fd
, CGROUPS_AGENT_RCVBUF_SIZE
);
904 (void) unlink(sa
.un
.sun_path
);
906 /* Only allow root to connect to this socket */
908 r
= bind(fd
, &sa
.sa
, SOCKADDR_UN_LEN(sa
.un
));
910 return log_error_errno(errno
, "bind(%s) failed: %m", sa
.un
.sun_path
);
912 m
->cgroups_agent_fd
= fd
;
916 if (!m
->cgroups_agent_event_source
) {
917 r
= sd_event_add_io(m
->event
, &m
->cgroups_agent_event_source
, m
->cgroups_agent_fd
, EPOLLIN
, manager_dispatch_cgroups_agent_fd
, m
);
919 return log_error_errno(r
, "Failed to allocate cgroups agent event source: %m");
921 /* Process cgroups notifications early, but after having processed service notification messages or
922 * SIGCHLD signals, so that a cgroup running empty is always just the last safety net of notification,
923 * and we collected the metadata the notification and SIGCHLD stuff offers first. Also see handling of
924 * cgroup inotify for the unified cgroup stuff. */
925 r
= sd_event_source_set_priority(m
->cgroups_agent_event_source
, SD_EVENT_PRIORITY_NORMAL
-4);
927 return log_error_errno(r
, "Failed to set priority of cgroups agent event source: %m");
929 (void) sd_event_source_set_description(m
->cgroups_agent_event_source
, "manager-cgroups-agent");
935 static int manager_setup_user_lookup_fd(Manager
*m
) {
940 /* Set up the socket pair used for passing UID/GID resolution results from forked off processes to PID
941 * 1. Background: we can't do name lookups (NSS) from PID 1, since it might involve IPC and thus activation,
942 * and we might hence deadlock on ourselves. Hence we do all user/group lookups asynchronously from the forked
943 * off processes right before executing the binaries to start. In order to be able to clean up any IPC objects
944 * created by a unit (see RemoveIPC=) we need to know in PID 1 the used UID/GID of the executed processes,
945 * hence we establish this communication channel so that forked off processes can pass their UID/GID
946 * information back to PID 1. The forked off processes send their resolved UID/GID to PID 1 in a simple
947 * datagram, along with their unit name, so that we can share one communication socket pair among all units for
950 * You might wonder why we need a communication channel for this that is independent of the usual notification
951 * socket scheme (i.e. $NOTIFY_SOCKET). The primary difference is about trust: data sent via the $NOTIFY_SOCKET
952 * channel is only accepted if it originates from the right unit and if reception was enabled for it. The user
953 * lookup socket OTOH is only accessible by PID 1 and its children until they exec(), and always available.
955 * Note that this function is called under two circumstances: when we first initialize (in which case we
956 * allocate both the socket pair and the event source to listen on it), and when we deserialize after a reload
957 * (in which case the socket pair already exists but we still need to allocate the event source for it). */
959 if (m
->user_lookup_fds
[0] < 0) {
961 /* Free all secondary fields */
962 safe_close_pair(m
->user_lookup_fds
);
963 m
->user_lookup_event_source
= sd_event_source_unref(m
->user_lookup_event_source
);
965 if (socketpair(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
, 0, m
->user_lookup_fds
) < 0)
966 return log_error_errno(errno
, "Failed to allocate user lookup socket: %m");
968 (void) fd_inc_rcvbuf(m
->user_lookup_fds
[0], NOTIFY_RCVBUF_SIZE
);
971 if (!m
->user_lookup_event_source
) {
972 r
= sd_event_add_io(m
->event
, &m
->user_lookup_event_source
, m
->user_lookup_fds
[0], EPOLLIN
, manager_dispatch_user_lookup_fd
, m
);
974 return log_error_errno(errno
, "Failed to allocate user lookup event source: %m");
976 /* Process even earlier than the notify event source, so that we always know first about valid UID/GID
978 r
= sd_event_source_set_priority(m
->user_lookup_event_source
, SD_EVENT_PRIORITY_NORMAL
-11);
980 return log_error_errno(errno
, "Failed to set priority ot user lookup event source: %m");
982 (void) sd_event_source_set_description(m
->user_lookup_event_source
, "user-lookup");
988 static unsigned manager_dispatch_cleanup_queue(Manager
*m
) {
994 while ((u
= m
->cleanup_queue
)) {
995 assert(u
->in_cleanup_queue
);
1005 GC_OFFSET_IN_PATH
, /* This one is on the path we were traveling */
1006 GC_OFFSET_UNSURE
, /* No clue */
1007 GC_OFFSET_GOOD
, /* We still need this unit */
1008 GC_OFFSET_BAD
, /* We don't need this unit anymore */
1012 static void unit_gc_mark_good(Unit
*u
, unsigned gc_marker
) {
1017 u
->gc_marker
= gc_marker
+ GC_OFFSET_GOOD
;
1019 /* Recursively mark referenced units as GOOD as well */
1020 HASHMAP_FOREACH_KEY(v
, other
, u
->dependencies
[UNIT_REFERENCES
], i
)
1021 if (other
->gc_marker
== gc_marker
+ GC_OFFSET_UNSURE
)
1022 unit_gc_mark_good(other
, gc_marker
);
1025 static void unit_gc_sweep(Unit
*u
, unsigned gc_marker
) {
1033 if (IN_SET(u
->gc_marker
- gc_marker
,
1034 GC_OFFSET_GOOD
, GC_OFFSET_BAD
, GC_OFFSET_UNSURE
, GC_OFFSET_IN_PATH
))
1037 if (u
->in_cleanup_queue
)
1040 if (unit_check_gc(u
))
1043 u
->gc_marker
= gc_marker
+ GC_OFFSET_IN_PATH
;
1047 HASHMAP_FOREACH_KEY(v
, other
, u
->dependencies
[UNIT_REFERENCED_BY
], i
) {
1048 unit_gc_sweep(other
, gc_marker
);
1050 if (other
->gc_marker
== gc_marker
+ GC_OFFSET_GOOD
)
1053 if (other
->gc_marker
!= gc_marker
+ GC_OFFSET_BAD
)
1060 /* We were unable to find anything out about this entry, so
1061 * let's investigate it later */
1062 u
->gc_marker
= gc_marker
+ GC_OFFSET_UNSURE
;
1063 unit_add_to_gc_queue(u
);
1067 /* We definitely know that this one is not useful anymore, so
1068 * let's mark it for deletion */
1069 u
->gc_marker
= gc_marker
+ GC_OFFSET_BAD
;
1070 unit_add_to_cleanup_queue(u
);
1074 unit_gc_mark_good(u
, gc_marker
);
1077 static unsigned manager_dispatch_gc_unit_queue(Manager
*m
) {
1078 unsigned n
= 0, gc_marker
;
1083 /* log_debug("Running GC..."); */
1085 m
->gc_marker
+= _GC_OFFSET_MAX
;
1086 if (m
->gc_marker
+ _GC_OFFSET_MAX
<= _GC_OFFSET_MAX
)
1089 gc_marker
= m
->gc_marker
;
1091 while ((u
= m
->gc_unit_queue
)) {
1092 assert(u
->in_gc_queue
);
1094 unit_gc_sweep(u
, gc_marker
);
1096 LIST_REMOVE(gc_queue
, m
->gc_unit_queue
, u
);
1097 u
->in_gc_queue
= false;
1101 if (IN_SET(u
->gc_marker
- gc_marker
,
1102 GC_OFFSET_BAD
, GC_OFFSET_UNSURE
)) {
1104 log_unit_debug(u
, "Collecting.");
1105 u
->gc_marker
= gc_marker
+ GC_OFFSET_BAD
;
1106 unit_add_to_cleanup_queue(u
);
1113 static unsigned manager_dispatch_gc_job_queue(Manager
*m
) {
1119 while ((j
= m
->gc_job_queue
)) {
1120 assert(j
->in_gc_queue
);
1122 LIST_REMOVE(gc_queue
, m
->gc_job_queue
, j
);
1123 j
->in_gc_queue
= false;
1127 if (job_check_gc(j
))
1130 log_unit_debug(j
->unit
, "Collecting job.");
1131 (void) job_finish_and_invalidate(j
, JOB_COLLECTED
, false, false);
1137 static void manager_clear_jobs_and_units(Manager
*m
) {
1142 while ((u
= hashmap_first(m
->units
)))
1145 manager_dispatch_cleanup_queue(m
);
1147 assert(!m
->load_queue
);
1148 assert(!m
->run_queue
);
1149 assert(!m
->dbus_unit_queue
);
1150 assert(!m
->dbus_job_queue
);
1151 assert(!m
->cleanup_queue
);
1152 assert(!m
->gc_unit_queue
);
1153 assert(!m
->gc_job_queue
);
1155 assert(hashmap_isempty(m
->jobs
));
1156 assert(hashmap_isempty(m
->units
));
1158 m
->n_on_console
= 0;
1159 m
->n_running_jobs
= 0;
1162 Manager
* manager_free(Manager
*m
) {
1165 ExecDirectoryType dt
;
1170 manager_clear_jobs_and_units(m
);
1172 for (c
= 0; c
< _UNIT_TYPE_MAX
; c
++)
1173 if (unit_vtable
[c
]->shutdown
)
1174 unit_vtable
[c
]->shutdown(m
);
1176 /* If we reexecute ourselves, we keep the root cgroup around */
1177 manager_shutdown_cgroup(m
, m
->exit_code
!= MANAGER_REEXECUTE
);
1179 lookup_paths_flush_generator(&m
->lookup_paths
);
1183 exec_runtime_vacuum(m
);
1184 hashmap_free(m
->exec_runtime_by_id
);
1186 dynamic_user_vacuum(m
, false);
1187 hashmap_free(m
->dynamic_users
);
1189 hashmap_free(m
->units
);
1190 hashmap_free(m
->units_by_invocation_id
);
1191 hashmap_free(m
->jobs
);
1192 hashmap_free(m
->watch_pids
);
1193 hashmap_free(m
->watch_bus
);
1195 set_free(m
->startup_units
);
1196 set_free(m
->failed_units
);
1198 sd_event_source_unref(m
->signal_event_source
);
1199 sd_event_source_unref(m
->sigchld_event_source
);
1200 sd_event_source_unref(m
->notify_event_source
);
1201 sd_event_source_unref(m
->cgroups_agent_event_source
);
1202 sd_event_source_unref(m
->time_change_event_source
);
1203 sd_event_source_unref(m
->jobs_in_progress_event_source
);
1204 sd_event_source_unref(m
->run_queue_event_source
);
1205 sd_event_source_unref(m
->user_lookup_event_source
);
1207 safe_close(m
->signal_fd
);
1208 safe_close(m
->notify_fd
);
1209 safe_close(m
->cgroups_agent_fd
);
1210 safe_close(m
->time_change_fd
);
1211 safe_close_pair(m
->user_lookup_fds
);
1213 manager_close_ask_password(m
);
1215 manager_close_idle_pipe(m
);
1217 udev_unref(m
->udev
);
1218 sd_event_unref(m
->event
);
1220 free(m
->notify_socket
);
1222 lookup_paths_free(&m
->lookup_paths
);
1223 strv_free(m
->environment
);
1225 hashmap_free(m
->cgroup_unit
);
1226 set_free_free(m
->unit_path_cache
);
1228 free(m
->switch_root
);
1229 free(m
->switch_root_init
);
1231 for (i
= 0; i
< _RLIMIT_MAX
; i
++)
1232 m
->rlimit
[i
] = mfree(m
->rlimit
[i
]);
1234 assert(hashmap_isempty(m
->units_requiring_mounts_for
));
1235 hashmap_free(m
->units_requiring_mounts_for
);
1237 hashmap_free(m
->uid_refs
);
1238 hashmap_free(m
->gid_refs
);
1240 for (dt
= 0; dt
< _EXEC_DIRECTORY_TYPE_MAX
; dt
++)
1241 m
->prefix
[dt
] = mfree(m
->prefix
[dt
]);
1246 void manager_enumerate(Manager
*m
) {
1251 /* Let's ask every type to load all units from disk/kernel
1252 * that it might know */
1253 for (c
= 0; c
< _UNIT_TYPE_MAX
; c
++) {
1254 if (!unit_type_supported(c
)) {
1255 log_debug("Unit type .%s is not supported on this system.", unit_type_to_string(c
));
1259 if (!unit_vtable
[c
]->enumerate
)
1262 unit_vtable
[c
]->enumerate(m
);
1265 manager_dispatch_load_queue(m
);
1268 static void manager_coldplug(Manager
*m
) {
1276 /* Then, let's set up their initial state. */
1277 HASHMAP_FOREACH_KEY(u
, k
, m
->units
, i
) {
1279 /* ignore aliases */
1283 r
= unit_coldplug(u
);
1285 log_warning_errno(r
, "We couldn't coldplug %s, proceeding anyway: %m", u
->id
);
1289 static void manager_build_unit_path_cache(Manager
*m
) {
1295 set_free_free(m
->unit_path_cache
);
1297 m
->unit_path_cache
= set_new(&string_hash_ops
);
1298 if (!m
->unit_path_cache
) {
1303 /* This simply builds a list of files we know exist, so that
1304 * we don't always have to go to disk */
1306 STRV_FOREACH(i
, m
->lookup_paths
.search_path
) {
1307 _cleanup_closedir_
DIR *d
= NULL
;
1312 if (errno
!= ENOENT
)
1313 log_warning_errno(errno
, "Failed to open directory %s, ignoring: %m", *i
);
1317 FOREACH_DIRENT(de
, d
, r
= -errno
; goto fail
) {
1320 p
= strjoin(streq(*i
, "/") ? "" : *i
, "/", de
->d_name
);
1326 r
= set_consume(m
->unit_path_cache
, p
);
1335 log_warning_errno(r
, "Failed to build unit path cache, proceeding without: %m");
1336 m
->unit_path_cache
= set_free_free(m
->unit_path_cache
);
1339 static void manager_distribute_fds(Manager
*m
, FDSet
*fds
) {
1345 HASHMAP_FOREACH(u
, m
->units
, i
) {
1347 if (fdset_size(fds
) <= 0)
1350 if (!UNIT_VTABLE(u
)->distribute_fds
)
1353 UNIT_VTABLE(u
)->distribute_fds(u
, fds
);
1357 static bool manager_dbus_is_running(Manager
*m
, bool deserialized
) {
1362 /* This checks whether the dbus instance we are supposed to expose our APIs on is up. We check both the socket
1363 * and the service unit. If the 'deserialized' parameter is true we'll check the deserialized state of the unit
1364 * rather than the current one. */
1366 if (m
->test_run_flags
!= 0)
1369 /* If we are in the user instance, and the env var is already set for us, then this means D-Bus is ran
1370 * somewhere outside of our own logic. Let's use it */
1371 if (MANAGER_IS_USER(m
) && getenv("DBUS_SESSION_BUS_ADDRESS"))
1374 u
= manager_get_unit(m
, SPECIAL_DBUS_SOCKET
);
1377 if ((deserialized
? SOCKET(u
)->deserialized_state
: SOCKET(u
)->state
) != SOCKET_RUNNING
)
1380 u
= manager_get_unit(m
, SPECIAL_DBUS_SERVICE
);
1383 if (!IN_SET((deserialized
? SERVICE(u
)->deserialized_state
: SERVICE(u
)->state
), SERVICE_RUNNING
, SERVICE_RELOAD
))
1389 int manager_startup(Manager
*m
, FILE *serialization
, FDSet
*fds
) {
1394 /* If we are running in test mode, we still want to run the generators,
1395 * but we should not touch the real generator directories. */
1396 r
= lookup_paths_init(&m
->lookup_paths
, m
->unit_file_scope
,
1397 m
->test_run_flags
? LOOKUP_PATHS_TEMPORARY_GENERATED
: 0,
1402 r
= manager_run_environment_generators(m
);
1406 dual_timestamp_get(m
->timestamps
+ MANAGER_TIMESTAMP_GENERATORS_START
);
1407 r
= manager_run_generators(m
);
1408 dual_timestamp_get(m
->timestamps
+ MANAGER_TIMESTAMP_GENERATORS_FINISH
);
1412 /* If this is the first boot, and we are in the host system, then preset everything */
1413 if (m
->first_boot
> 0 &&
1414 MANAGER_IS_SYSTEM(m
) &&
1415 !m
->test_run_flags
) {
1417 r
= unit_file_preset_all(UNIT_FILE_SYSTEM
, 0, NULL
, UNIT_FILE_PRESET_ENABLE_ONLY
, NULL
, 0);
1419 log_full_errno(r
== -EEXIST
? LOG_NOTICE
: LOG_WARNING
, r
,
1420 "Failed to populate /etc with preset unit settings, ignoring: %m");
1422 log_info("Populated /etc with preset unit settings.");
1425 lookup_paths_reduce(&m
->lookup_paths
);
1426 manager_build_unit_path_cache(m
);
1428 /* If we will deserialize make sure that during enumeration
1429 * this is already known, so we increase the counter here
1434 /* First, enumerate what we can from all config files */
1435 dual_timestamp_get(m
->timestamps
+ MANAGER_TIMESTAMP_UNITS_LOAD_START
);
1436 manager_enumerate(m
);
1437 dual_timestamp_get(m
->timestamps
+ MANAGER_TIMESTAMP_UNITS_LOAD_FINISH
);
1439 /* Second, deserialize if there is something to deserialize */
1440 if (serialization
) {
1441 r
= manager_deserialize(m
, serialization
, fds
);
1443 return log_error_errno(r
, "Deserialization failed: %m");
1446 /* Any fds left? Find some unit which wants them. This is
1447 * useful to allow container managers to pass some file
1448 * descriptors to us pre-initialized. This enables
1449 * socket-based activation of entire containers. */
1450 manager_distribute_fds(m
, fds
);
1452 /* We might have deserialized the notify fd, but if we didn't
1453 * then let's create the bus now */
1454 r
= manager_setup_notify(m
);
1456 /* No sense to continue without notifications, our children would fail anyway. */
1459 r
= manager_setup_cgroups_agent(m
);
1461 /* Likewise, no sense to continue without empty cgroup notifications. */
1464 r
= manager_setup_user_lookup_fd(m
);
1466 /* This shouldn't fail, except if things are really broken. */
1469 /* Let's set up our private bus connection now, unconditionally */
1470 (void) bus_init_private(m
);
1472 /* If we are in --user mode also connect to the system bus now */
1473 if (MANAGER_IS_USER(m
))
1474 (void) bus_init_system(m
);
1476 /* Let's connect to the bus now, but only if the unit is supposed to be up */
1477 if (manager_dbus_is_running(m
, !!serialization
)) {
1478 (void) bus_init_api(m
);
1480 if (MANAGER_IS_SYSTEM(m
))
1481 (void) bus_init_system(m
);
1484 /* Now that we are connected to all possible busses, let's deserialize who is tracking us. */
1485 (void) bus_track_coldplug(m
, &m
->subscribed
, false, m
->deserialized_subscribed
);
1486 m
->deserialized_subscribed
= strv_free(m
->deserialized_subscribed
);
1488 /* Third, fire things up! */
1489 manager_coldplug(m
);
1491 /* Release any dynamic users no longer referenced */
1492 dynamic_user_vacuum(m
, true);
1494 exec_runtime_vacuum(m
);
1496 /* Release any references to UIDs/GIDs no longer referenced, and destroy any IPC owned by them */
1497 manager_vacuum_uid_refs(m
);
1498 manager_vacuum_gid_refs(m
);
1500 if (serialization
) {
1501 assert(m
->n_reloading
> 0);
1504 /* Let's wait for the UnitNew/JobNew messages being
1505 * sent, before we notify that the reload is
1507 m
->send_reloading_done
= true;
1513 int manager_add_job(Manager
*m
, JobType type
, Unit
*unit
, JobMode mode
, sd_bus_error
*e
, Job
**_ret
) {
1518 assert(type
< _JOB_TYPE_MAX
);
1520 assert(mode
< _JOB_MODE_MAX
);
1522 if (mode
== JOB_ISOLATE
&& type
!= JOB_START
)
1523 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
, "Isolate is only valid for start.");
1525 if (mode
== JOB_ISOLATE
&& !unit
->allow_isolate
)
1526 return sd_bus_error_setf(e
, BUS_ERROR_NO_ISOLATION
, "Operation refused, unit may not be isolated.");
1528 log_unit_debug(unit
, "Trying to enqueue job %s/%s/%s", unit
->id
, job_type_to_string(type
), job_mode_to_string(mode
));
1530 type
= job_type_collapse(type
, unit
);
1532 tr
= transaction_new(mode
== JOB_REPLACE_IRREVERSIBLY
);
1536 r
= transaction_add_job_and_dependencies(tr
, type
, unit
, NULL
, true, false,
1537 IN_SET(mode
, JOB_IGNORE_DEPENDENCIES
, JOB_IGNORE_REQUIREMENTS
),
1538 mode
== JOB_IGNORE_DEPENDENCIES
, e
);
1542 if (mode
== JOB_ISOLATE
) {
1543 r
= transaction_add_isolate_jobs(tr
, m
);
1548 r
= transaction_activate(tr
, m
, mode
, e
);
1552 log_unit_debug(unit
,
1553 "Enqueued job %s/%s as %u", unit
->id
,
1554 job_type_to_string(type
), (unsigned) tr
->anchor_job
->id
);
1557 *_ret
= tr
->anchor_job
;
1559 transaction_free(tr
);
1563 transaction_abort(tr
);
1564 transaction_free(tr
);
1568 int manager_add_job_by_name(Manager
*m
, JobType type
, const char *name
, JobMode mode
, sd_bus_error
*e
, Job
**ret
) {
1569 Unit
*unit
= NULL
; /* just to appease gcc, initialization is not really necessary */
1573 assert(type
< _JOB_TYPE_MAX
);
1575 assert(mode
< _JOB_MODE_MAX
);
1577 r
= manager_load_unit(m
, name
, NULL
, NULL
, &unit
);
1582 return manager_add_job(m
, type
, unit
, mode
, e
, ret
);
1585 int manager_add_job_by_name_and_warn(Manager
*m
, JobType type
, const char *name
, JobMode mode
, Job
**ret
) {
1586 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
1590 assert(type
< _JOB_TYPE_MAX
);
1592 assert(mode
< _JOB_MODE_MAX
);
1594 r
= manager_add_job_by_name(m
, type
, name
, mode
, &error
, ret
);
1596 return log_warning_errno(r
, "Failed to enqueue %s job for %s: %s", job_mode_to_string(mode
), name
, bus_error_message(&error
, r
));
1601 int manager_propagate_reload(Manager
*m
, Unit
*unit
, JobMode mode
, sd_bus_error
*e
) {
1607 assert(mode
< _JOB_MODE_MAX
);
1608 assert(mode
!= JOB_ISOLATE
); /* Isolate is only valid for start */
1610 tr
= transaction_new(mode
== JOB_REPLACE_IRREVERSIBLY
);
1614 /* We need an anchor job */
1615 r
= transaction_add_job_and_dependencies(tr
, JOB_NOP
, unit
, NULL
, false, false, true, true, e
);
1619 /* Failure in adding individual dependencies is ignored, so this always succeeds. */
1620 transaction_add_propagate_reload_jobs(tr
, unit
, tr
->anchor_job
, mode
== JOB_IGNORE_DEPENDENCIES
, e
);
1622 r
= transaction_activate(tr
, m
, mode
, e
);
1626 transaction_free(tr
);
1630 transaction_abort(tr
);
1631 transaction_free(tr
);
1635 Job
*manager_get_job(Manager
*m
, uint32_t id
) {
1638 return hashmap_get(m
->jobs
, UINT32_TO_PTR(id
));
1641 Unit
*manager_get_unit(Manager
*m
, const char *name
) {
1645 return hashmap_get(m
->units
, name
);
1648 unsigned manager_dispatch_load_queue(Manager
*m
) {
1654 /* Make sure we are not run recursively */
1655 if (m
->dispatching_load_queue
)
1658 m
->dispatching_load_queue
= true;
1660 /* Dispatches the load queue. Takes a unit from the queue and
1661 * tries to load its data until the queue is empty */
1663 while ((u
= m
->load_queue
)) {
1664 assert(u
->in_load_queue
);
1670 m
->dispatching_load_queue
= false;
1674 int manager_load_unit_prepare(
1686 assert(name
|| path
);
1689 /* This will prepare the unit for loading, but not actually
1690 * load anything from disk. */
1692 if (path
&& !is_path(path
))
1693 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
, "Path %s is not absolute.", path
);
1696 name
= basename(path
);
1698 t
= unit_name_to_type(name
);
1700 if (t
== _UNIT_TYPE_INVALID
|| !unit_name_is_valid(name
, UNIT_NAME_PLAIN
|UNIT_NAME_INSTANCE
)) {
1701 if (unit_name_is_valid(name
, UNIT_NAME_TEMPLATE
))
1702 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
, "Unit name %s is missing the instance name.", name
);
1704 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
, "Unit name %s is not valid.", name
);
1707 ret
= manager_get_unit(m
, name
);
1713 ret
= unit_new(m
, unit_vtable
[t
]->object_size
);
1718 ret
->fragment_path
= strdup(path
);
1719 if (!ret
->fragment_path
) {
1725 r
= unit_add_name(ret
, name
);
1731 unit_add_to_load_queue(ret
);
1732 unit_add_to_dbus_queue(ret
);
1733 unit_add_to_gc_queue(ret
);
1740 int manager_load_unit(
1752 /* This will load the service information files, but not actually
1753 * start any services or anything. */
1755 r
= manager_load_unit_prepare(m
, name
, path
, e
, _ret
);
1759 manager_dispatch_load_queue(m
);
1761 *_ret
= unit_follow_merge(*_ret
);
1766 void manager_dump_jobs(Manager
*s
, FILE *f
, const char *prefix
) {
1773 HASHMAP_FOREACH(j
, s
->jobs
, i
)
1774 job_dump(j
, f
, prefix
);
1777 void manager_dump_units(Manager
*s
, FILE *f
, const char *prefix
) {
1785 HASHMAP_FOREACH_KEY(u
, t
, s
->units
, i
)
1787 unit_dump(u
, f
, prefix
);
1790 void manager_dump(Manager
*m
, FILE *f
, const char *prefix
) {
1796 for (q
= 0; q
< _MANAGER_TIMESTAMP_MAX
; q
++) {
1797 char buf
[FORMAT_TIMESTAMP_MAX
];
1799 if (dual_timestamp_is_set(m
->timestamps
+ q
))
1800 fprintf(f
, "%sTimestamp %s: %s\n",
1802 manager_timestamp_to_string(q
),
1803 format_timestamp(buf
, sizeof(buf
), m
->timestamps
[q
].realtime
));
1806 manager_dump_units(m
, f
, prefix
);
1807 manager_dump_jobs(m
, f
, prefix
);
1810 int manager_get_dump_string(Manager
*m
, char **ret
) {
1811 _cleanup_free_
char *dump
= NULL
;
1812 _cleanup_fclose_
FILE *f
= NULL
;
1819 f
= open_memstream(&dump
, &size
);
1823 (void) __fsetlocking(f
, FSETLOCKING_BYCALLER
);
1825 manager_dump(m
, f
, NULL
);
1827 r
= fflush_and_check(f
);
1839 void manager_clear_jobs(Manager
*m
) {
1844 while ((j
= hashmap_first(m
->jobs
)))
1845 /* No need to recurse. We're cancelling all jobs. */
1846 job_finish_and_invalidate(j
, JOB_CANCELED
, false, false);
1849 static int manager_dispatch_run_queue(sd_event_source
*source
, void *userdata
) {
1850 Manager
*m
= userdata
;
1856 while ((j
= m
->run_queue
)) {
1857 assert(j
->installed
);
1858 assert(j
->in_run_queue
);
1860 job_run_and_invalidate(j
);
1863 if (m
->n_running_jobs
> 0)
1864 manager_watch_jobs_in_progress(m
);
1866 if (m
->n_on_console
> 0)
1867 manager_watch_idle_pipe(m
);
1872 static unsigned manager_dispatch_dbus_queue(Manager
*m
) {
1879 if (m
->dispatching_dbus_queue
)
1882 m
->dispatching_dbus_queue
= true;
1884 while ((u
= m
->dbus_unit_queue
)) {
1885 assert(u
->in_dbus_queue
);
1887 bus_unit_send_change_signal(u
);
1891 while ((j
= m
->dbus_job_queue
)) {
1892 assert(j
->in_dbus_queue
);
1894 bus_job_send_change_signal(j
);
1898 m
->dispatching_dbus_queue
= false;
1900 if (m
->send_reloading_done
) {
1901 m
->send_reloading_done
= false;
1903 bus_manager_send_reloading(m
, false);
1906 if (m
->queued_message
)
1907 bus_send_queued_message(m
);
1912 static int manager_dispatch_cgroups_agent_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
1913 Manager
*m
= userdata
;
1914 char buf
[PATH_MAX
+1];
1917 n
= recv(fd
, buf
, sizeof(buf
), 0);
1919 return log_error_errno(errno
, "Failed to read cgroups agent message: %m");
1921 log_error("Got zero-length cgroups agent message, ignoring.");
1924 if ((size_t) n
>= sizeof(buf
)) {
1925 log_error("Got overly long cgroups agent message, ignoring.");
1929 if (memchr(buf
, 0, n
)) {
1930 log_error("Got cgroups agent message with embedded NUL byte, ignoring.");
1935 manager_notify_cgroup_empty(m
, buf
);
1936 (void) bus_forward_agent_released(m
, buf
);
1941 static void manager_invoke_notify_message(
1944 const struct ucred
*ucred
,
1953 if (u
->notifygen
== m
->notifygen
) /* Already invoked on this same unit in this same iteration? */
1955 u
->notifygen
= m
->notifygen
;
1957 if (UNIT_VTABLE(u
)->notify_message
) {
1958 _cleanup_strv_free_
char **tags
= NULL
;
1960 tags
= strv_split(buf
, NEWLINE
);
1966 UNIT_VTABLE(u
)->notify_message(u
, ucred
, tags
, fds
);
1968 } else if (DEBUG_LOGGING
) {
1969 _cleanup_free_
char *x
= NULL
, *y
= NULL
;
1971 x
= ellipsize(buf
, 20, 90);
1975 log_unit_debug(u
, "Got notification message \"%s\", ignoring.", strnull(y
));
1979 static int manager_dispatch_notify_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
1981 _cleanup_fdset_free_ FDSet
*fds
= NULL
;
1982 Manager
*m
= userdata
;
1983 char buf
[NOTIFY_BUFFER_MAX
+1];
1984 struct iovec iovec
= {
1986 .iov_len
= sizeof(buf
)-1,
1989 struct cmsghdr cmsghdr
;
1990 uint8_t buf
[CMSG_SPACE(sizeof(struct ucred
)) +
1991 CMSG_SPACE(sizeof(int) * NOTIFY_FD_MAX
)];
1993 struct msghdr msghdr
= {
1996 .msg_control
= &control
,
1997 .msg_controllen
= sizeof(control
),
2000 struct cmsghdr
*cmsg
;
2001 struct ucred
*ucred
= NULL
;
2002 _cleanup_free_ Unit
**array_copy
= NULL
;
2003 Unit
*u1
, *u2
, **array
;
2004 int r
, *fd_array
= NULL
;
2010 assert(m
->notify_fd
== fd
);
2012 if (revents
!= EPOLLIN
) {
2013 log_warning("Got unexpected poll event for notify fd.");
2017 n
= recvmsg(m
->notify_fd
, &msghdr
, MSG_DONTWAIT
|MSG_CMSG_CLOEXEC
|MSG_TRUNC
);
2019 if (IN_SET(errno
, EAGAIN
, EINTR
))
2020 return 0; /* Spurious wakeup, try again */
2022 /* If this is any other, real error, then let's stop processing this socket. This of course means we
2023 * won't take notification messages anymore, but that's still better than busy looping around this:
2024 * being woken up over and over again but being unable to actually read the message off the socket. */
2025 return log_error_errno(errno
, "Failed to receive notification message: %m");
2028 CMSG_FOREACH(cmsg
, &msghdr
) {
2029 if (cmsg
->cmsg_level
== SOL_SOCKET
&& cmsg
->cmsg_type
== SCM_RIGHTS
) {
2031 fd_array
= (int*) CMSG_DATA(cmsg
);
2032 n_fds
= (cmsg
->cmsg_len
- CMSG_LEN(0)) / sizeof(int);
2034 } else if (cmsg
->cmsg_level
== SOL_SOCKET
&&
2035 cmsg
->cmsg_type
== SCM_CREDENTIALS
&&
2036 cmsg
->cmsg_len
== CMSG_LEN(sizeof(struct ucred
))) {
2038 ucred
= (struct ucred
*) CMSG_DATA(cmsg
);
2045 r
= fdset_new_array(&fds
, fd_array
, n_fds
);
2047 close_many(fd_array
, n_fds
);
2053 if (!ucred
|| !pid_is_valid(ucred
->pid
)) {
2054 log_warning("Received notify message without valid credentials. Ignoring.");
2058 if ((size_t) n
>= sizeof(buf
) || (msghdr
.msg_flags
& MSG_TRUNC
)) {
2059 log_warning("Received notify message exceeded maximum size. Ignoring.");
2063 /* As extra safety check, let's make sure the string we get doesn't contain embedded NUL bytes. We permit one
2064 * trailing NUL byte in the message, but don't expect it. */
2065 if (n
> 1 && memchr(buf
, 0, n
-1)) {
2066 log_warning("Received notify message with embedded NUL bytes. Ignoring.");
2070 /* Make sure it's NUL-terminated. */
2073 /* Increase the generation counter used for filtering out duplicate unit invocations. */
2076 /* Notify every unit that might be interested, which might be multiple. */
2077 u1
= manager_get_unit_by_pid_cgroup(m
, ucred
->pid
);
2078 u2
= hashmap_get(m
->watch_pids
, PID_TO_PTR(ucred
->pid
));
2079 array
= hashmap_get(m
->watch_pids
, PID_TO_PTR(-ucred
->pid
));
2086 array_copy
= newdup(Unit
*, array
, k
+1);
2090 /* And now invoke the per-unit callbacks. Note that manager_invoke_notify_message() will handle duplicate units
2091 * make sure we only invoke each unit's handler once. */
2093 manager_invoke_notify_message(m
, u1
, ucred
, buf
, fds
);
2097 manager_invoke_notify_message(m
, u2
, ucred
, buf
, fds
);
2101 for (size_t i
= 0; array_copy
[i
]; i
++) {
2102 manager_invoke_notify_message(m
, array_copy
[i
], ucred
, buf
, fds
);
2107 log_warning("Cannot find unit for notify message of PID "PID_FMT
", ignoring.", ucred
->pid
);
2109 if (fdset_size(fds
) > 0)
2110 log_warning("Got extra auxiliary fds with notification message, closing them.");
2115 static void manager_invoke_sigchld_event(
2118 const siginfo_t
*si
) {
2124 /* Already invoked the handler of this unit in this iteration? Then don't process this again */
2125 if (u
->sigchldgen
== m
->sigchldgen
)
2127 u
->sigchldgen
= m
->sigchldgen
;
2129 log_unit_debug(u
, "Child "PID_FMT
" belongs to %s.", si
->si_pid
, u
->id
);
2130 unit_unwatch_pid(u
, si
->si_pid
);
2132 if (UNIT_VTABLE(u
)->sigchld_event
)
2133 UNIT_VTABLE(u
)->sigchld_event(u
, si
->si_pid
, si
->si_code
, si
->si_status
);
2136 static int manager_dispatch_sigchld(sd_event_source
*source
, void *userdata
) {
2137 Manager
*m
= userdata
;
2144 /* First we call waitd() for a PID and do not reap the zombie. That way we can still access /proc/$PID for it
2145 * while it is a zombie. */
2147 if (waitid(P_ALL
, 0, &si
, WEXITED
|WNOHANG
|WNOWAIT
) < 0) {
2149 if (errno
== ECHILD
)
2152 log_error_errno(errno
, "Failed to peek for child with waitid(), ignoring: %m");
2159 if (IN_SET(si
.si_code
, CLD_EXITED
, CLD_KILLED
, CLD_DUMPED
)) {
2160 _cleanup_free_ Unit
**array_copy
= NULL
;
2161 _cleanup_free_
char *name
= NULL
;
2162 Unit
*u1
, *u2
, **array
;
2164 (void) get_process_comm(si
.si_pid
, &name
);
2166 log_debug("Child "PID_FMT
" (%s) died (code=%s, status=%i/%s)",
2167 si
.si_pid
, strna(name
),
2168 sigchld_code_to_string(si
.si_code
),
2170 strna(si
.si_code
== CLD_EXITED
2171 ? exit_status_to_string(si
.si_status
, EXIT_STATUS_FULL
)
2172 : signal_to_string(si
.si_status
)));
2174 /* Increase the generation counter used for filtering out duplicate unit invocations */
2177 /* And now figure out the unit this belongs to, it might be multiple... */
2178 u1
= manager_get_unit_by_pid_cgroup(m
, si
.si_pid
);
2179 u2
= hashmap_get(m
->watch_pids
, PID_TO_PTR(si
.si_pid
));
2180 array
= hashmap_get(m
->watch_pids
, PID_TO_PTR(-si
.si_pid
));
2184 /* Cound how many entries the array has */
2188 /* Make a copy of the array so that we don't trip up on the array changing beneath us */
2189 array_copy
= newdup(Unit
*, array
, n
+1);
2194 /* Finally, execute them all. Note that u1, u2 and the array might contain duplicates, but
2195 * that's fine, manager_invoke_sigchld_event() will ensure we only invoke the handlers once for
2196 * each iteration. */
2198 manager_invoke_sigchld_event(m
, u1
, &si
);
2200 manager_invoke_sigchld_event(m
, u2
, &si
);
2202 for (size_t i
= 0; array_copy
[i
]; i
++)
2203 manager_invoke_sigchld_event(m
, array_copy
[i
], &si
);
2206 /* And now, we actually reap the zombie. */
2207 if (waitid(P_PID
, si
.si_pid
, &si
, WEXITED
) < 0) {
2208 log_error_errno(errno
, "Failed to dequeue child, ignoring: %m");
2215 /* All children processed for now, turn off event source */
2217 r
= sd_event_source_set_enabled(m
->sigchld_event_source
, SD_EVENT_OFF
);
2219 return log_error_errno(r
, "Failed to disable SIGCHLD event source: %m");
2224 static void manager_start_target(Manager
*m
, const char *name
, JobMode mode
) {
2225 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
2228 log_debug("Activating special unit %s", name
);
2230 r
= manager_add_job_by_name(m
, JOB_START
, name
, mode
, &error
, NULL
);
2232 log_error("Failed to enqueue %s job: %s", name
, bus_error_message(&error
, r
));
2235 static void manager_handle_ctrl_alt_del(Manager
*m
) {
2236 /* If the user presses C-A-D more than
2237 * 7 times within 2s, we reboot/shutdown immediately,
2238 * unless it was disabled in system.conf */
2240 if (ratelimit_test(&m
->ctrl_alt_del_ratelimit
) || m
->cad_burst_action
== EMERGENCY_ACTION_NONE
)
2241 manager_start_target(m
, SPECIAL_CTRL_ALT_DEL_TARGET
, JOB_REPLACE_IRREVERSIBLY
);
2243 emergency_action(m
, m
->cad_burst_action
, NULL
,
2244 "Ctrl-Alt-Del was pressed more than 7 times within 2s");
2247 static int manager_dispatch_signal_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
2248 Manager
*m
= userdata
;
2250 struct signalfd_siginfo sfsi
;
2254 assert(m
->signal_fd
== fd
);
2256 if (revents
!= EPOLLIN
) {
2257 log_warning("Got unexpected events from signal file descriptor.");
2261 n
= read(m
->signal_fd
, &sfsi
, sizeof(sfsi
));
2262 if (n
!= sizeof(sfsi
)) {
2264 log_warning("Truncated read from signal fd (%zu bytes), ignoring!", n
);
2268 if (IN_SET(errno
, EINTR
, EAGAIN
))
2271 /* We return an error here, which will kill this handler,
2272 * to avoid a busy loop on read error. */
2273 return log_error_errno(errno
, "Reading from signal fd failed: %m");
2276 log_received_signal(sfsi
.ssi_signo
== SIGCHLD
||
2277 (sfsi
.ssi_signo
== SIGTERM
&& MANAGER_IS_USER(m
))
2278 ? LOG_DEBUG
: LOG_INFO
,
2281 switch (sfsi
.ssi_signo
) {
2284 r
= sd_event_source_set_enabled(m
->sigchld_event_source
, SD_EVENT_ON
);
2286 log_warning_errno(r
, "Failed to enable SIGCHLD even source, ignoring: %m");
2291 if (MANAGER_IS_SYSTEM(m
)) {
2292 /* This is for compatibility with the
2293 * original sysvinit */
2294 r
= verify_run_space_and_log("Refusing to reexecute");
2296 m
->exit_code
= MANAGER_REEXECUTE
;
2302 if (MANAGER_IS_SYSTEM(m
))
2303 manager_handle_ctrl_alt_del(m
);
2305 manager_start_target(m
, SPECIAL_EXIT_TARGET
,
2306 JOB_REPLACE_IRREVERSIBLY
);
2310 if (MANAGER_IS_SYSTEM(m
))
2311 manager_start_target(m
, SPECIAL_KBREQUEST_TARGET
, JOB_REPLACE
);
2313 /* This is a nop on non-init */
2317 if (MANAGER_IS_SYSTEM(m
))
2318 manager_start_target(m
, SPECIAL_SIGPWR_TARGET
, JOB_REPLACE
);
2320 /* This is a nop on non-init */
2325 if (manager_dbus_is_running(m
, false)) {
2326 log_info("Trying to reconnect to bus...");
2328 (void) bus_init_api(m
);
2330 if (MANAGER_IS_SYSTEM(m
))
2331 (void) bus_init_system(m
);
2333 log_info("Starting D-Bus service...");
2334 manager_start_target(m
, SPECIAL_DBUS_SERVICE
, JOB_REPLACE
);
2340 _cleanup_free_
char *dump
= NULL
;
2342 r
= manager_get_dump_string(m
, &dump
);
2344 log_warning_errno(errno
, "Failed to acquire manager dump: %m");
2348 log_dump(LOG_INFO
, dump
);
2353 r
= verify_run_space_and_log("Refusing to reload");
2355 m
->exit_code
= MANAGER_RELOAD
;
2360 /* Starting SIGRTMIN+0 */
2361 static const struct {
2364 } target_table
[] = {
2365 [0] = { SPECIAL_DEFAULT_TARGET
, JOB_ISOLATE
},
2366 [1] = { SPECIAL_RESCUE_TARGET
, JOB_ISOLATE
},
2367 [2] = { SPECIAL_EMERGENCY_TARGET
, JOB_ISOLATE
},
2368 [3] = { SPECIAL_HALT_TARGET
, JOB_REPLACE_IRREVERSIBLY
},
2369 [4] = { SPECIAL_POWEROFF_TARGET
, JOB_REPLACE_IRREVERSIBLY
},
2370 [5] = { SPECIAL_REBOOT_TARGET
, JOB_REPLACE_IRREVERSIBLY
},
2371 [6] = { SPECIAL_KEXEC_TARGET
, JOB_REPLACE_IRREVERSIBLY
},
2374 /* Starting SIGRTMIN+13, so that target halt and system halt are 10 apart */
2375 static const ManagerExitCode code_table
[] = {
2377 [1] = MANAGER_POWEROFF
,
2378 [2] = MANAGER_REBOOT
,
2379 [3] = MANAGER_KEXEC
,
2382 if ((int) sfsi
.ssi_signo
>= SIGRTMIN
+0 &&
2383 (int) sfsi
.ssi_signo
< SIGRTMIN
+(int) ELEMENTSOF(target_table
)) {
2384 int idx
= (int) sfsi
.ssi_signo
- SIGRTMIN
;
2385 manager_start_target(m
, target_table
[idx
].target
,
2386 target_table
[idx
].mode
);
2390 if ((int) sfsi
.ssi_signo
>= SIGRTMIN
+13 &&
2391 (int) sfsi
.ssi_signo
< SIGRTMIN
+13+(int) ELEMENTSOF(code_table
)) {
2392 m
->exit_code
= code_table
[sfsi
.ssi_signo
- SIGRTMIN
- 13];
2396 switch (sfsi
.ssi_signo
- SIGRTMIN
) {
2399 manager_set_show_status(m
, SHOW_STATUS_YES
);
2403 manager_set_show_status(m
, SHOW_STATUS_NO
);
2407 log_set_max_level(LOG_DEBUG
);
2408 log_info("Setting log level to debug.");
2412 log_set_max_level(LOG_INFO
);
2413 log_info("Setting log level to info.");
2417 if (MANAGER_IS_USER(m
)) {
2418 m
->exit_code
= MANAGER_EXIT
;
2422 /* This is a nop on init */
2426 case 29: /* compatibility: used to be mapped to LOG_TARGET_SYSLOG_OR_KMSG */
2427 log_set_target(LOG_TARGET_JOURNAL_OR_KMSG
);
2428 log_notice("Setting log target to journal-or-kmsg.");
2432 log_set_target(LOG_TARGET_CONSOLE
);
2433 log_notice("Setting log target to console.");
2437 log_set_target(LOG_TARGET_KMSG
);
2438 log_notice("Setting log target to kmsg.");
2442 log_warning("Got unhandled signal <%s>.", signal_to_string(sfsi
.ssi_signo
));
2449 static int manager_dispatch_time_change_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
2450 Manager
*m
= userdata
;
2455 assert(m
->time_change_fd
== fd
);
2457 log_struct(LOG_DEBUG
,
2458 "MESSAGE_ID=" SD_MESSAGE_TIME_CHANGE_STR
,
2459 LOG_MESSAGE("Time has been changed"),
2462 /* Restart the watch */
2463 m
->time_change_event_source
= sd_event_source_unref(m
->time_change_event_source
);
2464 m
->time_change_fd
= safe_close(m
->time_change_fd
);
2466 manager_setup_time_change(m
);
2468 HASHMAP_FOREACH(u
, m
->units
, i
)
2469 if (UNIT_VTABLE(u
)->time_change
)
2470 UNIT_VTABLE(u
)->time_change(u
);
2475 static int manager_dispatch_idle_pipe_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
2476 Manager
*m
= userdata
;
2479 assert(m
->idle_pipe
[2] == fd
);
2481 /* There's at least one Type=idle child that just gave up on us waiting for the boot process to complete. Let's
2482 * now turn off any further console output if there's at least one service that needs console access, so that
2483 * from now on our own output should not spill into that service's output anymore. After all, we support
2484 * Type=idle only to beautify console output and it generally is set on services that want to own the console
2485 * exclusively without our interference. */
2486 m
->no_console_output
= m
->n_on_console
> 0;
2488 /* Acknowledge the child's request, and let all all other children know too that they shouldn't wait any longer
2489 * by closing the pipes towards them, which is what they are waiting for. */
2490 manager_close_idle_pipe(m
);
2495 static int manager_dispatch_jobs_in_progress(sd_event_source
*source
, usec_t usec
, void *userdata
) {
2496 Manager
*m
= userdata
;
2503 manager_print_jobs_in_progress(m
);
2505 next
= now(CLOCK_MONOTONIC
) + JOBS_IN_PROGRESS_PERIOD_USEC
;
2506 r
= sd_event_source_set_time(source
, next
);
2510 return sd_event_source_set_enabled(source
, SD_EVENT_ONESHOT
);
2513 int manager_loop(Manager
*m
) {
2516 RATELIMIT_DEFINE(rl
, 1*USEC_PER_SEC
, 50000);
2519 m
->exit_code
= MANAGER_OK
;
2521 /* Release the path cache */
2522 m
->unit_path_cache
= set_free_free(m
->unit_path_cache
);
2524 manager_check_finished(m
);
2526 /* There might still be some zombies hanging around from before we were exec()'ed. Let's reap them. */
2527 r
= sd_event_source_set_enabled(m
->sigchld_event_source
, SD_EVENT_ON
);
2529 return log_error_errno(r
, "Failed to enable SIGCHLD event source: %m");
2531 while (m
->exit_code
== MANAGER_OK
) {
2534 if (m
->runtime_watchdog
> 0 && m
->runtime_watchdog
!= USEC_INFINITY
&& MANAGER_IS_SYSTEM(m
))
2537 if (!ratelimit_test(&rl
)) {
2538 /* Yay, something is going seriously wrong, pause a little */
2539 log_warning("Looping too fast. Throttling execution a little.");
2543 if (manager_dispatch_load_queue(m
) > 0)
2546 if (manager_dispatch_gc_job_queue(m
) > 0)
2549 if (manager_dispatch_gc_unit_queue(m
) > 0)
2552 if (manager_dispatch_cleanup_queue(m
) > 0)
2555 if (manager_dispatch_cgroup_realize_queue(m
) > 0)
2558 if (manager_dispatch_dbus_queue(m
) > 0)
2561 /* Sleep for half the watchdog time */
2562 if (m
->runtime_watchdog
> 0 && m
->runtime_watchdog
!= USEC_INFINITY
&& MANAGER_IS_SYSTEM(m
)) {
2563 wait_usec
= m
->runtime_watchdog
/ 2;
2567 wait_usec
= USEC_INFINITY
;
2569 r
= sd_event_run(m
->event
, wait_usec
);
2571 return log_error_errno(r
, "Failed to run event loop: %m");
2574 return m
->exit_code
;
2577 int manager_load_unit_from_dbus_path(Manager
*m
, const char *s
, sd_bus_error
*e
, Unit
**_u
) {
2578 _cleanup_free_
char *n
= NULL
;
2579 sd_id128_t invocation_id
;
2587 r
= unit_name_from_dbus_path(s
, &n
);
2591 /* Permit addressing units by invocation ID: if the passed bus path is suffixed by a 128bit ID then we use it
2592 * as invocation ID. */
2593 r
= sd_id128_from_string(n
, &invocation_id
);
2595 u
= hashmap_get(m
->units_by_invocation_id
, &invocation_id
);
2601 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
));
2604 /* If this didn't work, we check if this is a unit name */
2605 if (!unit_name_is_valid(n
, UNIT_NAME_PLAIN
|UNIT_NAME_INSTANCE
))
2606 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
, "Unit name %s is neither a valid invocation ID nor unit name.", n
);
2608 r
= manager_load_unit(m
, n
, NULL
, e
, &u
);
2616 int manager_get_job_from_dbus_path(Manager
*m
, const char *s
, Job
**_j
) {
2626 p
= startswith(s
, "/org/freedesktop/systemd1/job/");
2630 r
= safe_atou(p
, &id
);
2634 j
= manager_get_job(m
, id
);
2643 void manager_send_unit_audit(Manager
*m
, Unit
*u
, int type
, bool success
) {
2646 _cleanup_free_
char *p
= NULL
;
2650 if (!MANAGER_IS_SYSTEM(m
))
2653 audit_fd
= get_audit_fd();
2657 /* Don't generate audit events if the service was already
2658 * started and we're just deserializing */
2659 if (MANAGER_IS_RELOADING(m
))
2662 if (u
->type
!= UNIT_SERVICE
)
2665 r
= unit_name_to_prefix_and_instance(u
->id
, &p
);
2667 log_error_errno(r
, "Failed to extract prefix and instance of unit name: %m");
2671 msg
= strjoina("unit=", p
);
2672 if (audit_log_user_comm_message(audit_fd
, type
, msg
, "systemd", NULL
, NULL
, NULL
, success
) < 0) {
2674 /* We aren't allowed to send audit messages?
2675 * Then let's not retry again. */
2678 log_warning_errno(errno
, "Failed to send audit message: %m");
2684 void manager_send_unit_plymouth(Manager
*m
, Unit
*u
) {
2685 static const union sockaddr_union sa
= PLYMOUTH_SOCKET
;
2686 _cleanup_free_
char *message
= NULL
;
2687 _cleanup_close_
int fd
= -1;
2690 /* Don't generate plymouth events if the service was already
2691 * started and we're just deserializing */
2692 if (MANAGER_IS_RELOADING(m
))
2695 if (!MANAGER_IS_SYSTEM(m
))
2698 if (detect_container() > 0)
2701 if (!IN_SET(u
->type
, UNIT_SERVICE
, UNIT_MOUNT
, UNIT_SWAP
))
2704 /* We set SOCK_NONBLOCK here so that we rather drop the
2705 * message then wait for plymouth */
2706 fd
= socket(AF_UNIX
, SOCK_STREAM
|SOCK_CLOEXEC
|SOCK_NONBLOCK
, 0);
2708 log_error_errno(errno
, "socket() failed: %m");
2712 if (connect(fd
, &sa
.sa
, SOCKADDR_UN_LEN(sa
.un
)) < 0) {
2714 if (!IN_SET(errno
, EPIPE
, EAGAIN
, ENOENT
, ECONNREFUSED
, ECONNRESET
, ECONNABORTED
))
2715 log_error_errno(errno
, "connect() failed: %m");
2719 if (asprintf(&message
, "U\002%c%s%n", (int) (strlen(u
->id
) + 1), u
->id
, &n
) < 0) {
2725 if (write(fd
, message
, n
+ 1) != n
+ 1)
2726 if (!IN_SET(errno
, EPIPE
, EAGAIN
, ENOENT
, ECONNREFUSED
, ECONNRESET
, ECONNABORTED
))
2727 log_error_errno(errno
, "Failed to write Plymouth message: %m");
2730 int manager_open_serialization(Manager
*m
, FILE **_f
) {
2736 fd
= open_serialization_fd("systemd-state");
2740 f
= fdopen(fd
, "w+");
2750 int manager_serialize(Manager
*m
, FILE *f
, FDSet
*fds
, bool switching_root
) {
2763 fprintf(f
, "current-job-id=%"PRIu32
"\n", m
->current_job_id
);
2764 fprintf(f
, "n-installed-jobs=%u\n", m
->n_installed_jobs
);
2765 fprintf(f
, "n-failed-jobs=%u\n", m
->n_failed_jobs
);
2766 fprintf(f
, "taint-usr=%s\n", yes_no(m
->taint_usr
));
2767 fprintf(f
, "ready-sent=%s\n", yes_no(m
->ready_sent
));
2768 fprintf(f
, "taint-logged=%s\n", yes_no(m
->taint_logged
));
2769 fprintf(f
, "service-watchdogs=%s\n", yes_no(m
->service_watchdogs
));
2771 for (q
= 0; q
< _MANAGER_TIMESTAMP_MAX
; q
++) {
2772 /* The userspace and finish timestamps only apply to the host system, hence only serialize them there */
2773 if (in_initrd() && IN_SET(q
, MANAGER_TIMESTAMP_USERSPACE
, MANAGER_TIMESTAMP_FINISH
))
2776 t
= manager_timestamp_to_string(q
);
2778 char field
[strlen(t
) + STRLEN("-timestamp") + 1];
2779 strcpy(stpcpy(field
, t
), "-timestamp");
2780 dual_timestamp_serialize(f
, field
, m
->timestamps
+ q
);
2784 if (!switching_root
)
2785 (void) serialize_environment(f
, m
->environment
);
2787 if (m
->notify_fd
>= 0) {
2790 copy
= fdset_put_dup(fds
, m
->notify_fd
);
2794 fprintf(f
, "notify-fd=%i\n", copy
);
2795 fprintf(f
, "notify-socket=%s\n", m
->notify_socket
);
2798 if (m
->cgroups_agent_fd
>= 0) {
2801 copy
= fdset_put_dup(fds
, m
->cgroups_agent_fd
);
2805 fprintf(f
, "cgroups-agent-fd=%i\n", copy
);
2808 if (m
->user_lookup_fds
[0] >= 0) {
2811 copy0
= fdset_put_dup(fds
, m
->user_lookup_fds
[0]);
2815 copy1
= fdset_put_dup(fds
, m
->user_lookup_fds
[1]);
2819 fprintf(f
, "user-lookup=%i %i\n", copy0
, copy1
);
2822 bus_track_serialize(m
->subscribed
, f
, "subscribed");
2824 r
= dynamic_user_serialize(m
, f
, fds
);
2828 manager_serialize_uid_refs(m
, f
);
2829 manager_serialize_gid_refs(m
, f
);
2831 r
= exec_runtime_serialize(m
, f
, fds
);
2835 (void) fputc('\n', f
);
2837 HASHMAP_FOREACH_KEY(u
, t
, m
->units
, i
) {
2845 r
= unit_serialize(u
, f
, fds
, !switching_root
);
2852 assert(m
->n_reloading
> 0);
2858 r
= bus_fdset_add_all(m
, fds
);
2865 int manager_deserialize(Manager
*m
, FILE *f
, FDSet
*fds
) {
2871 log_debug("Deserializing state...");
2876 char line
[LINE_MAX
];
2877 const char *val
, *l
;
2879 if (!fgets(line
, sizeof(line
), f
)) {
2894 if ((val
= startswith(l
, "current-job-id="))) {
2897 if (safe_atou32(val
, &id
) < 0)
2898 log_notice("Failed to parse current job id value %s", val
);
2900 m
->current_job_id
= MAX(m
->current_job_id
, id
);
2902 } else if ((val
= startswith(l
, "n-installed-jobs="))) {
2905 if (safe_atou32(val
, &n
) < 0)
2906 log_notice("Failed to parse installed jobs counter %s", val
);
2908 m
->n_installed_jobs
+= n
;
2910 } else if ((val
= startswith(l
, "n-failed-jobs="))) {
2913 if (safe_atou32(val
, &n
) < 0)
2914 log_notice("Failed to parse failed jobs counter %s", val
);
2916 m
->n_failed_jobs
+= n
;
2918 } else if ((val
= startswith(l
, "taint-usr="))) {
2921 b
= parse_boolean(val
);
2923 log_notice("Failed to parse taint /usr flag %s", val
);
2925 m
->taint_usr
= m
->taint_usr
|| b
;
2927 } else if ((val
= startswith(l
, "ready-sent="))) {
2930 b
= parse_boolean(val
);
2932 log_notice("Failed to parse ready-sent flag %s", val
);
2934 m
->ready_sent
= m
->ready_sent
|| b
;
2936 } else if ((val
= startswith(l
, "taint-logged="))) {
2939 b
= parse_boolean(val
);
2941 log_notice("Failed to parse taint-logged flag %s", val
);
2943 m
->taint_logged
= m
->taint_logged
|| b
;
2945 } else if ((val
= startswith(l
, "service-watchdogs="))) {
2948 b
= parse_boolean(val
);
2950 log_notice("Failed to parse service-watchdogs flag %s", val
);
2952 m
->service_watchdogs
= b
;
2954 } else if (startswith(l
, "env=")) {
2955 r
= deserialize_environment(&m
->environment
, l
);
2959 log_notice_errno(r
, "Failed to parse environment entry: \"%s\": %m", l
);
2961 } else if ((val
= startswith(l
, "notify-fd="))) {
2964 if (safe_atoi(val
, &fd
) < 0 || fd
< 0 || !fdset_contains(fds
, fd
))
2965 log_notice("Failed to parse notify fd: \"%s\"", val
);
2967 m
->notify_event_source
= sd_event_source_unref(m
->notify_event_source
);
2968 safe_close(m
->notify_fd
);
2969 m
->notify_fd
= fdset_remove(fds
, fd
);
2972 } else if ((val
= startswith(l
, "notify-socket="))) {
2981 free(m
->notify_socket
);
2982 m
->notify_socket
= n
;
2984 } else if ((val
= startswith(l
, "cgroups-agent-fd="))) {
2987 if (safe_atoi(val
, &fd
) < 0 || fd
< 0 || !fdset_contains(fds
, fd
))
2988 log_notice("Failed to parse cgroups agent fd: %s", val
);
2990 m
->cgroups_agent_event_source
= sd_event_source_unref(m
->cgroups_agent_event_source
);
2991 safe_close(m
->cgroups_agent_fd
);
2992 m
->cgroups_agent_fd
= fdset_remove(fds
, fd
);
2995 } else if ((val
= startswith(l
, "user-lookup="))) {
2998 if (sscanf(val
, "%i %i", &fd0
, &fd1
) != 2 || fd0
< 0 || fd1
< 0 || fd0
== fd1
|| !fdset_contains(fds
, fd0
) || !fdset_contains(fds
, fd1
))
2999 log_notice("Failed to parse user lookup fd: %s", val
);
3001 m
->user_lookup_event_source
= sd_event_source_unref(m
->user_lookup_event_source
);
3002 safe_close_pair(m
->user_lookup_fds
);
3003 m
->user_lookup_fds
[0] = fdset_remove(fds
, fd0
);
3004 m
->user_lookup_fds
[1] = fdset_remove(fds
, fd1
);
3007 } else if ((val
= startswith(l
, "dynamic-user=")))
3008 dynamic_user_deserialize_one(m
, val
, fds
);
3009 else if ((val
= startswith(l
, "destroy-ipc-uid=")))
3010 manager_deserialize_uid_refs_one(m
, val
);
3011 else if ((val
= startswith(l
, "destroy-ipc-gid=")))
3012 manager_deserialize_gid_refs_one(m
, val
);
3013 else if ((val
= startswith(l
, "exec-runtime=")))
3014 exec_runtime_deserialize_one(m
, val
, fds
);
3015 else if ((val
= startswith(l
, "subscribed="))) {
3017 if (strv_extend(&m
->deserialized_subscribed
, val
) < 0)
3022 for (q
= 0; q
< _MANAGER_TIMESTAMP_MAX
; q
++) {
3023 val
= startswith(l
, manager_timestamp_to_string(q
));
3027 val
= startswith(val
, "-timestamp=");
3032 if (q
< _MANAGER_TIMESTAMP_MAX
) /* found it */
3033 dual_timestamp_deserialize(val
, m
->timestamps
+ q
);
3034 else if (!startswith(l
, "kdbus-fd=")) /* ignore kdbus */
3035 log_notice("Unknown serialization item '%s'", l
);
3041 char name
[UNIT_NAME_MAX
+2];
3042 const char* unit_name
;
3045 if (!fgets(name
, sizeof(name
), f
)) {
3055 unit_name
= strstrip(name
);
3057 r
= manager_load_unit(m
, unit_name
, NULL
, NULL
, &u
);
3059 log_notice_errno(r
, "Failed to load unit \"%s\", skipping deserialization: %m", unit_name
);
3062 unit_deserialize_skip(f
);
3066 r
= unit_deserialize(u
, f
, fds
);
3068 log_notice_errno(r
, "Failed to deserialize unit \"%s\": %m", unit_name
);
3078 assert(m
->n_reloading
> 0);
3084 int manager_reload(Manager
*m
) {
3086 _cleanup_fclose_
FILE *f
= NULL
;
3087 _cleanup_fdset_free_ FDSet
*fds
= NULL
;
3091 r
= manager_open_serialization(m
, &f
);
3096 bus_manager_send_reloading(m
, true);
3104 r
= manager_serialize(m
, f
, fds
, false);
3110 if (fseeko(f
, 0, SEEK_SET
) < 0) {
3115 /* From here on there is no way back. */
3116 manager_clear_jobs_and_units(m
);
3117 lookup_paths_flush_generator(&m
->lookup_paths
);
3118 lookup_paths_free(&m
->lookup_paths
);
3119 exec_runtime_vacuum(m
);
3120 dynamic_user_vacuum(m
, false);
3121 m
->uid_refs
= hashmap_free(m
->uid_refs
);
3122 m
->gid_refs
= hashmap_free(m
->gid_refs
);
3124 q
= lookup_paths_init(&m
->lookup_paths
, m
->unit_file_scope
, 0, NULL
);
3125 if (q
< 0 && r
>= 0)
3128 q
= manager_run_environment_generators(m
);
3129 if (q
< 0 && r
>= 0)
3132 /* Find new unit paths */
3133 q
= manager_run_generators(m
);
3134 if (q
< 0 && r
>= 0)
3137 lookup_paths_reduce(&m
->lookup_paths
);
3138 manager_build_unit_path_cache(m
);
3140 /* First, enumerate what we can from all config files */
3141 manager_enumerate(m
);
3143 /* Second, deserialize our stored data */
3144 q
= manager_deserialize(m
, f
, fds
);
3146 log_error_errno(q
, "Deserialization failed: %m");
3155 /* Re-register notify_fd as event source */
3156 q
= manager_setup_notify(m
);
3157 if (q
< 0 && r
>= 0)
3160 q
= manager_setup_cgroups_agent(m
);
3161 if (q
< 0 && r
>= 0)
3164 q
= manager_setup_user_lookup_fd(m
);
3165 if (q
< 0 && r
>= 0)
3168 /* Third, fire things up! */
3169 manager_coldplug(m
);
3171 /* Release any dynamic users no longer referenced */
3172 dynamic_user_vacuum(m
, true);
3174 /* Release any references to UIDs/GIDs no longer referenced, and destroy any IPC owned by them */
3175 manager_vacuum_uid_refs(m
);
3176 manager_vacuum_gid_refs(m
);
3178 exec_runtime_vacuum(m
);
3180 /* It might be safe to log to the journal now and connect to dbus */
3181 manager_recheck_journal(m
);
3182 manager_recheck_dbus(m
);
3184 /* Sync current state of bus names with our set of listening units */
3186 manager_sync_bus_names(m
, m
->api_bus
);
3188 assert(m
->n_reloading
> 0);
3191 m
->send_reloading_done
= true;
3196 void manager_reset_failed(Manager
*m
) {
3202 HASHMAP_FOREACH(u
, m
->units
, i
)
3203 unit_reset_failed(u
);
3206 bool manager_unit_inactive_or_pending(Manager
*m
, const char *name
) {
3212 /* Returns true if the unit is inactive or going down */
3213 u
= manager_get_unit(m
, name
);
3217 return unit_inactive_or_pending(u
);
3220 static void log_taint_string(Manager
*m
) {
3221 _cleanup_free_
char *taint
= NULL
;
3225 if (MANAGER_IS_USER(m
) || m
->taint_logged
)
3228 m
->taint_logged
= true; /* only check for taint once */
3230 taint
= manager_taint_string(m
);
3234 log_struct(LOG_NOTICE
,
3235 LOG_MESSAGE("System is tainted: %s", taint
),
3237 "MESSAGE_ID=" SD_MESSAGE_TAINTED_STR
,
3241 static void manager_notify_finished(Manager
*m
) {
3242 char userspace
[FORMAT_TIMESPAN_MAX
], initrd
[FORMAT_TIMESPAN_MAX
], kernel
[FORMAT_TIMESPAN_MAX
], sum
[FORMAT_TIMESPAN_MAX
];
3243 usec_t firmware_usec
, loader_usec
, kernel_usec
, initrd_usec
, userspace_usec
, total_usec
;
3245 if (m
->test_run_flags
)
3248 if (MANAGER_IS_SYSTEM(m
) && detect_container() <= 0) {
3249 char ts
[FORMAT_TIMESPAN_MAX
];
3250 char buf
[FORMAT_TIMESPAN_MAX
+ STRLEN(" (firmware) + ") + FORMAT_TIMESPAN_MAX
+ STRLEN(" (loader) + ")]
3253 size_t size
= sizeof buf
;
3255 /* Note that MANAGER_TIMESTAMP_KERNEL's monotonic value is always at 0, and
3256 * MANAGER_TIMESTAMP_FIRMWARE's and MANAGER_TIMESTAMP_LOADER's monotonic value should be considered
3257 * negative values. */
3259 firmware_usec
= m
->timestamps
[MANAGER_TIMESTAMP_FIRMWARE
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_LOADER
].monotonic
;
3260 loader_usec
= m
->timestamps
[MANAGER_TIMESTAMP_LOADER
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_KERNEL
].monotonic
;
3261 userspace_usec
= m
->timestamps
[MANAGER_TIMESTAMP_FINISH
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_USERSPACE
].monotonic
;
3262 total_usec
= m
->timestamps
[MANAGER_TIMESTAMP_FIRMWARE
].monotonic
+ m
->timestamps
[MANAGER_TIMESTAMP_FINISH
].monotonic
;
3264 if (firmware_usec
> 0)
3265 size
= strpcpyf(&p
, size
, "%s (firmware) + ", format_timespan(ts
, sizeof(ts
), firmware_usec
, USEC_PER_MSEC
));
3266 if (loader_usec
> 0)
3267 size
= strpcpyf(&p
, size
, "%s (loader) + ", format_timespan(ts
, sizeof(ts
), loader_usec
, USEC_PER_MSEC
));
3269 if (dual_timestamp_is_set(&m
->timestamps
[MANAGER_TIMESTAMP_INITRD
])) {
3271 /* The initrd case on bare-metal*/
3272 kernel_usec
= m
->timestamps
[MANAGER_TIMESTAMP_INITRD
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_KERNEL
].monotonic
;
3273 initrd_usec
= m
->timestamps
[MANAGER_TIMESTAMP_USERSPACE
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_INITRD
].monotonic
;
3275 log_struct(LOG_INFO
,
3276 "MESSAGE_ID=" SD_MESSAGE_STARTUP_FINISHED_STR
,
3277 "KERNEL_USEC="USEC_FMT
, kernel_usec
,
3278 "INITRD_USEC="USEC_FMT
, initrd_usec
,
3279 "USERSPACE_USEC="USEC_FMT
, userspace_usec
,
3280 LOG_MESSAGE("Startup finished in %s%s (kernel) + %s (initrd) + %s (userspace) = %s.",
3282 format_timespan(kernel
, sizeof(kernel
), kernel_usec
, USEC_PER_MSEC
),
3283 format_timespan(initrd
, sizeof(initrd
), initrd_usec
, USEC_PER_MSEC
),
3284 format_timespan(userspace
, sizeof(userspace
), userspace_usec
, USEC_PER_MSEC
),
3285 format_timespan(sum
, sizeof(sum
), total_usec
, USEC_PER_MSEC
)),
3288 /* The initrd-less case on bare-metal*/
3290 kernel_usec
= m
->timestamps
[MANAGER_TIMESTAMP_USERSPACE
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_KERNEL
].monotonic
;
3293 log_struct(LOG_INFO
,
3294 "MESSAGE_ID=" SD_MESSAGE_STARTUP_FINISHED_STR
,
3295 "KERNEL_USEC="USEC_FMT
, kernel_usec
,
3296 "USERSPACE_USEC="USEC_FMT
, userspace_usec
,
3297 LOG_MESSAGE("Startup finished in %s%s (kernel) + %s (userspace) = %s.",
3299 format_timespan(kernel
, sizeof(kernel
), kernel_usec
, USEC_PER_MSEC
),
3300 format_timespan(userspace
, sizeof(userspace
), userspace_usec
, USEC_PER_MSEC
),
3301 format_timespan(sum
, sizeof(sum
), total_usec
, USEC_PER_MSEC
)),
3305 /* The container and --user case */
3306 firmware_usec
= loader_usec
= initrd_usec
= kernel_usec
= 0;
3307 total_usec
= userspace_usec
= m
->timestamps
[MANAGER_TIMESTAMP_FINISH
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_USERSPACE
].monotonic
;
3309 log_struct(LOG_INFO
,
3310 "MESSAGE_ID=" SD_MESSAGE_USER_STARTUP_FINISHED_STR
,
3311 "USERSPACE_USEC="USEC_FMT
, userspace_usec
,
3312 LOG_MESSAGE("Startup finished in %s.",
3313 format_timespan(sum
, sizeof(sum
), total_usec
, USEC_PER_MSEC
)),
3317 bus_manager_send_finished(m
, firmware_usec
, loader_usec
, kernel_usec
, initrd_usec
, userspace_usec
, total_usec
);
3320 m
->ready_sent
? "STATUS=Startup finished in %s."
3322 "STATUS=Startup finished in %s.",
3323 format_timespan(sum
, sizeof(sum
), total_usec
, USEC_PER_MSEC
));
3324 m
->ready_sent
= true;
3326 log_taint_string(m
);
3329 static void manager_send_ready(Manager
*m
) {
3332 /* We send READY=1 on reaching basic.target only when running in --user mode. */
3333 if (!MANAGER_IS_USER(m
) || m
->ready_sent
)
3336 m
->ready_sent
= true;
3340 "STATUS=Reached " SPECIAL_BASIC_TARGET
".");
3343 static void manager_check_basic_target(Manager
*m
) {
3348 /* Small shortcut */
3349 if (m
->ready_sent
&& m
->taint_logged
)
3352 u
= manager_get_unit(m
, SPECIAL_BASIC_TARGET
);
3353 if (!u
|| !UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u
)))
3356 /* For user managers, send out READY=1 as soon as we reach basic.target */
3357 manager_send_ready(m
);
3359 /* Log the taint string as soon as we reach basic.target */
3360 log_taint_string(m
);
3363 void manager_check_finished(Manager
*m
) {
3366 if (MANAGER_IS_RELOADING(m
))
3369 /* Verify that we have entered the event loop already, and not left it again. */
3370 if (!MANAGER_IS_RUNNING(m
))
3373 manager_check_basic_target(m
);
3375 if (hashmap_size(m
->jobs
) > 0) {
3376 if (m
->jobs_in_progress_event_source
)
3377 /* Ignore any failure, this is only for feedback */
3378 (void) sd_event_source_set_time(m
->jobs_in_progress_event_source
, now(CLOCK_MONOTONIC
) + JOBS_IN_PROGRESS_WAIT_USEC
);
3383 manager_flip_auto_status(m
, false);
3385 /* Notify Type=idle units that we are done now */
3386 manager_close_idle_pipe(m
);
3388 /* Turn off confirm spawn now */
3389 m
->confirm_spawn
= NULL
;
3391 /* No need to update ask password status when we're going non-interactive */
3392 manager_close_ask_password(m
);
3394 /* This is no longer the first boot */
3395 manager_set_first_boot(m
, false);
3397 if (MANAGER_IS_FINISHED(m
))
3400 dual_timestamp_get(m
->timestamps
+ MANAGER_TIMESTAMP_FINISH
);
3402 manager_notify_finished(m
);
3404 manager_invalidate_startup_units(m
);
3407 static bool generator_path_any(const char* const* paths
) {
3411 /* Optimize by skipping the whole process by not creating output directories
3412 * if no generators are found. */
3413 STRV_FOREACH(path
, (char**) paths
)
3414 if (access(*path
, F_OK
) == 0)
3416 else if (errno
!= ENOENT
)
3417 log_warning_errno(errno
, "Failed to open generator directory %s: %m", *path
);
3422 static const char* system_env_generator_binary_paths
[] = {
3423 "/run/systemd/system-environment-generators",
3424 "/etc/systemd/system-environment-generators",
3425 "/usr/local/lib/systemd/system-environment-generators",
3426 SYSTEM_ENV_GENERATOR_PATH
,
3430 static const char* user_env_generator_binary_paths
[] = {
3431 "/run/systemd/user-environment-generators",
3432 "/etc/systemd/user-environment-generators",
3433 "/usr/local/lib/systemd/user-environment-generators",
3434 USER_ENV_GENERATOR_PATH
,
3438 static int manager_run_environment_generators(Manager
*m
) {
3439 char **tmp
= NULL
; /* this is only used in the forked process, no cleanup here */
3441 void* args
[] = {&tmp
, &tmp
, &m
->environment
};
3443 if (m
->test_run_flags
&& !(m
->test_run_flags
& MANAGER_TEST_RUN_ENV_GENERATORS
))
3446 paths
= MANAGER_IS_SYSTEM(m
) ? system_env_generator_binary_paths
: user_env_generator_binary_paths
;
3448 if (!generator_path_any(paths
))
3451 return execute_directories(paths
, DEFAULT_TIMEOUT_USEC
, gather_environment
, args
, NULL
);
3454 static int manager_run_generators(Manager
*m
) {
3455 _cleanup_strv_free_
char **paths
= NULL
;
3456 const char *argv
[5];
3461 if (m
->test_run_flags
&& !(m
->test_run_flags
& MANAGER_TEST_RUN_GENERATORS
))
3464 paths
= generator_binary_paths(m
->unit_file_scope
);
3468 if (!generator_path_any((const char* const*) paths
))
3471 r
= lookup_paths_mkdir_generator(&m
->lookup_paths
);
3475 argv
[0] = NULL
; /* Leave this empty, execute_directory() will fill something in */
3476 argv
[1] = m
->lookup_paths
.generator
;
3477 argv
[2] = m
->lookup_paths
.generator_early
;
3478 argv
[3] = m
->lookup_paths
.generator_late
;
3481 RUN_WITH_UMASK(0022)
3482 execute_directories((const char* const*) paths
, DEFAULT_TIMEOUT_USEC
,
3483 NULL
, NULL
, (char**) argv
);
3486 lookup_paths_trim_generator(&m
->lookup_paths
);
3490 int manager_environment_add(Manager
*m
, char **minus
, char **plus
) {
3491 char **a
= NULL
, **b
= NULL
, **l
;
3496 if (!strv_isempty(minus
)) {
3497 a
= strv_env_delete(l
, 1, minus
);
3504 if (!strv_isempty(plus
)) {
3505 b
= strv_env_merge(2, l
, plus
);
3514 if (m
->environment
!= l
)
3515 strv_free(m
->environment
);
3522 manager_sanitize_environment(m
);
3527 int manager_set_default_rlimits(Manager
*m
, struct rlimit
**default_rlimit
) {
3532 for (i
= 0; i
< _RLIMIT_MAX
; i
++) {
3533 m
->rlimit
[i
] = mfree(m
->rlimit
[i
]);
3535 if (!default_rlimit
[i
])
3538 m
->rlimit
[i
] = newdup(struct rlimit
, default_rlimit
[i
], 1);
3546 void manager_recheck_dbus(Manager
*m
) {
3549 /* Connects to the bus if the dbus service and socket are running. If we are running in user mode this is all
3550 * it does. In system mode we'll also connect to the system bus (which will most likely just reuse the
3551 * connection of the API bus). That's because the system bus after all runs as service of the system instance,
3552 * while in the user instance we can assume it's already there. */
3554 if (manager_dbus_is_running(m
, false)) {
3555 (void) bus_init_api(m
);
3557 if (MANAGER_IS_SYSTEM(m
))
3558 (void) bus_init_system(m
);
3560 (void) bus_done_api(m
);
3562 if (MANAGER_IS_SYSTEM(m
))
3563 (void) bus_done_system(m
);
3567 static bool manager_journal_is_running(Manager
*m
) {
3572 if (m
->test_run_flags
!= 0)
3575 /* If we are the user manager we can safely assume that the journal is up */
3576 if (!MANAGER_IS_SYSTEM(m
))
3579 /* Check that the socket is not only up, but in RUNNING state */
3580 u
= manager_get_unit(m
, SPECIAL_JOURNALD_SOCKET
);
3583 if (SOCKET(u
)->state
!= SOCKET_RUNNING
)
3586 /* Similar, check if the daemon itself is fully up, too */
3587 u
= manager_get_unit(m
, SPECIAL_JOURNALD_SERVICE
);
3590 if (!IN_SET(SERVICE(u
)->state
, SERVICE_RELOAD
, SERVICE_RUNNING
))
3596 void manager_recheck_journal(Manager
*m
) {
3600 /* Don't bother with this unless we are in the special situation of being PID 1 */
3601 if (getpid_cached() != 1)
3604 /* The journal is fully and entirely up? If so, let's permit logging to it, if that's configured. If the
3605 * journal is down, don't ever log to it, otherwise we might end up deadlocking ourselves as we might trigger
3606 * an activation ourselves we can't fulfill. */
3607 log_set_prohibit_ipc(!manager_journal_is_running(m
));
3611 void manager_set_show_status(Manager
*m
, ShowStatus mode
) {
3613 assert(IN_SET(mode
, SHOW_STATUS_AUTO
, SHOW_STATUS_NO
, SHOW_STATUS_YES
, SHOW_STATUS_TEMPORARY
));
3615 if (!MANAGER_IS_SYSTEM(m
))
3618 if (m
->show_status
!= mode
)
3619 log_debug("%s showing of status.",
3620 mode
== SHOW_STATUS_NO
? "Disabling" : "Enabling");
3621 m
->show_status
= mode
;
3624 (void) touch("/run/systemd/show-status");
3626 (void) unlink("/run/systemd/show-status");
3629 static bool manager_get_show_status(Manager
*m
, StatusType type
) {
3632 if (!MANAGER_IS_SYSTEM(m
))
3635 if (m
->no_console_output
)
3638 if (!IN_SET(manager_state(m
), MANAGER_INITIALIZING
, MANAGER_STARTING
, MANAGER_STOPPING
))
3641 /* If we cannot find out the status properly, just proceed. */
3642 if (type
!= STATUS_TYPE_EMERGENCY
&& manager_check_ask_password(m
) > 0)
3645 return m
->show_status
> 0;
3648 const char *manager_get_confirm_spawn(Manager
*m
) {
3649 static int last_errno
= 0;
3650 const char *vc
= m
->confirm_spawn
;
3654 /* Here's the deal: we want to test the validity of the console but don't want
3655 * PID1 to go through the whole console process which might block. But we also
3656 * want to warn the user only once if something is wrong with the console so we
3657 * cannot do the sanity checks after spawning our children. So here we simply do
3658 * really basic tests to hopefully trap common errors.
3660 * If the console suddenly disappear at the time our children will really it
3661 * then they will simply fail to acquire it and a positive answer will be
3662 * assumed. New children will fallback to /dev/console though.
3664 * Note: TTYs are devices that can come and go any time, and frequently aren't
3665 * available yet during early boot (consider a USB rs232 dongle...). If for any
3666 * reason the configured console is not ready, we fallback to the default
3669 if (!vc
|| path_equal(vc
, "/dev/console"))
3676 if (!S_ISCHR(st
.st_mode
)) {
3684 if (last_errno
!= errno
) {
3686 log_warning_errno(errno
, "Failed to open %s: %m, using default console", vc
);
3688 return "/dev/console";
3691 void manager_set_first_boot(Manager
*m
, bool b
) {
3694 if (!MANAGER_IS_SYSTEM(m
))
3697 if (m
->first_boot
!= (int) b
) {
3699 (void) touch("/run/systemd/first-boot");
3701 (void) unlink("/run/systemd/first-boot");
3707 void manager_disable_confirm_spawn(void) {
3708 (void) touch("/run/systemd/confirm_spawn_disabled");
3711 bool manager_is_confirm_spawn_disabled(Manager
*m
) {
3712 if (!m
->confirm_spawn
)
3715 return access("/run/systemd/confirm_spawn_disabled", F_OK
) >= 0;
3718 void manager_status_printf(Manager
*m
, StatusType type
, const char *status
, const char *format
, ...) {
3721 /* If m is NULL, assume we're after shutdown and let the messages through. */
3723 if (m
&& !manager_get_show_status(m
, type
))
3726 /* XXX We should totally drop the check for ephemeral here
3727 * and thus effectively make 'Type=idle' pointless. */
3728 if (type
== STATUS_TYPE_EPHEMERAL
&& m
&& m
->n_on_console
> 0)
3731 va_start(ap
, format
);
3732 status_vprintf(status
, true, type
== STATUS_TYPE_EPHEMERAL
, format
, ap
);
3736 Set
*manager_get_units_requiring_mounts_for(Manager
*m
, const char *path
) {
3737 char p
[strlen(path
)+1];
3743 path_kill_slashes(p
);
3745 return hashmap_get(m
->units_requiring_mounts_for
, streq(p
, "/") ? "" : p
);
3748 int manager_update_failed_units(Manager
*m
, Unit
*u
, bool failed
) {
3753 assert(u
->manager
== m
);
3755 size
= set_size(m
->failed_units
);
3758 r
= set_ensure_allocated(&m
->failed_units
, NULL
);
3762 if (set_put(m
->failed_units
, u
) < 0)
3765 (void) set_remove(m
->failed_units
, u
);
3767 if (set_size(m
->failed_units
) != size
)
3768 bus_manager_send_change_signal(m
);
3773 ManagerState
manager_state(Manager
*m
) {
3778 /* Did we ever finish booting? If not then we are still starting up */
3779 if (!MANAGER_IS_FINISHED(m
)) {
3781 u
= manager_get_unit(m
, SPECIAL_BASIC_TARGET
);
3782 if (!u
|| !UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u
)))
3783 return MANAGER_INITIALIZING
;
3785 return MANAGER_STARTING
;
3788 /* Is the special shutdown target active or queued? If so, we are in shutdown state */
3789 u
= manager_get_unit(m
, SPECIAL_SHUTDOWN_TARGET
);
3790 if (u
&& unit_active_or_pending(u
))
3791 return MANAGER_STOPPING
;
3793 if (MANAGER_IS_SYSTEM(m
)) {
3794 /* Are the rescue or emergency targets active or queued? If so we are in maintenance state */
3795 u
= manager_get_unit(m
, SPECIAL_RESCUE_TARGET
);
3796 if (u
&& unit_active_or_pending(u
))
3797 return MANAGER_MAINTENANCE
;
3799 u
= manager_get_unit(m
, SPECIAL_EMERGENCY_TARGET
);
3800 if (u
&& unit_active_or_pending(u
))
3801 return MANAGER_MAINTENANCE
;
3804 /* Are there any failed units? If so, we are in degraded mode */
3805 if (set_size(m
->failed_units
) > 0)
3806 return MANAGER_DEGRADED
;
3808 return MANAGER_RUNNING
;
3811 #define DESTROY_IPC_FLAG (UINT32_C(1) << 31)
3813 static void manager_unref_uid_internal(
3818 int (*_clean_ipc
)(uid_t uid
)) {
3824 assert(uid_is_valid(uid
));
3827 /* A generic implementation, covering both manager_unref_uid() and manager_unref_gid(), under the assumption
3828 * that uid_t and gid_t are actually defined the same way, with the same validity rules.
3830 * We store a hashmap where the UID/GID is they key and the value is a 32bit reference counter, whose highest
3831 * bit is used as flag for marking UIDs/GIDs whose IPC objects to remove when the last reference to the UID/GID
3832 * is dropped. The flag is set to on, once at least one reference from a unit where RemoveIPC= is set is added
3833 * on a UID/GID. It is reset when the UID's/GID's reference counter drops to 0 again. */
3835 assert_cc(sizeof(uid_t
) == sizeof(gid_t
));
3836 assert_cc(UID_INVALID
== (uid_t
) GID_INVALID
);
3838 if (uid
== 0) /* We don't keep track of root, and will never destroy it */
3841 c
= PTR_TO_UINT32(hashmap_get(*uid_refs
, UID_TO_PTR(uid
)));
3843 n
= c
& ~DESTROY_IPC_FLAG
;
3847 if (destroy_now
&& n
== 0) {
3848 hashmap_remove(*uid_refs
, UID_TO_PTR(uid
));
3850 if (c
& DESTROY_IPC_FLAG
) {
3851 log_debug("%s " UID_FMT
" is no longer referenced, cleaning up its IPC.",
3852 _clean_ipc
== clean_ipc_by_uid
? "UID" : "GID",
3854 (void) _clean_ipc(uid
);
3857 c
= n
| (c
& DESTROY_IPC_FLAG
);
3858 assert_se(hashmap_update(*uid_refs
, UID_TO_PTR(uid
), UINT32_TO_PTR(c
)) >= 0);
3862 void manager_unref_uid(Manager
*m
, uid_t uid
, bool destroy_now
) {
3863 manager_unref_uid_internal(m
, &m
->uid_refs
, uid
, destroy_now
, clean_ipc_by_uid
);
3866 void manager_unref_gid(Manager
*m
, gid_t gid
, bool destroy_now
) {
3867 manager_unref_uid_internal(m
, &m
->gid_refs
, (uid_t
) gid
, destroy_now
, clean_ipc_by_gid
);
3870 static int manager_ref_uid_internal(
3881 assert(uid_is_valid(uid
));
3883 /* A generic implementation, covering both manager_ref_uid() and manager_ref_gid(), under the assumption
3884 * that uid_t and gid_t are actually defined the same way, with the same validity rules. */
3886 assert_cc(sizeof(uid_t
) == sizeof(gid_t
));
3887 assert_cc(UID_INVALID
== (uid_t
) GID_INVALID
);
3889 if (uid
== 0) /* We don't keep track of root, and will never destroy it */
3892 r
= hashmap_ensure_allocated(uid_refs
, &trivial_hash_ops
);
3896 c
= PTR_TO_UINT32(hashmap_get(*uid_refs
, UID_TO_PTR(uid
)));
3898 n
= c
& ~DESTROY_IPC_FLAG
;
3901 if (n
& DESTROY_IPC_FLAG
) /* check for overflow */
3904 c
= n
| (c
& DESTROY_IPC_FLAG
) | (clean_ipc
? DESTROY_IPC_FLAG
: 0);
3906 return hashmap_replace(*uid_refs
, UID_TO_PTR(uid
), UINT32_TO_PTR(c
));
3909 int manager_ref_uid(Manager
*m
, uid_t uid
, bool clean_ipc
) {
3910 return manager_ref_uid_internal(m
, &m
->uid_refs
, uid
, clean_ipc
);
3913 int manager_ref_gid(Manager
*m
, gid_t gid
, bool clean_ipc
) {
3914 return manager_ref_uid_internal(m
, &m
->gid_refs
, (uid_t
) gid
, clean_ipc
);
3917 static void manager_vacuum_uid_refs_internal(
3920 int (*_clean_ipc
)(uid_t uid
)) {
3929 HASHMAP_FOREACH_KEY(p
, k
, *uid_refs
, i
) {
3933 uid
= PTR_TO_UID(k
);
3934 c
= PTR_TO_UINT32(p
);
3936 n
= c
& ~DESTROY_IPC_FLAG
;
3940 if (c
& DESTROY_IPC_FLAG
) {
3941 log_debug("Found unreferenced %s " UID_FMT
" after reload/reexec. Cleaning up.",
3942 _clean_ipc
== clean_ipc_by_uid
? "UID" : "GID",
3944 (void) _clean_ipc(uid
);
3947 assert_se(hashmap_remove(*uid_refs
, k
) == p
);
3951 void manager_vacuum_uid_refs(Manager
*m
) {
3952 manager_vacuum_uid_refs_internal(m
, &m
->uid_refs
, clean_ipc_by_uid
);
3955 void manager_vacuum_gid_refs(Manager
*m
) {
3956 manager_vacuum_uid_refs_internal(m
, &m
->gid_refs
, clean_ipc_by_gid
);
3959 static void manager_serialize_uid_refs_internal(
3963 const char *field_name
) {
3973 /* Serialize the UID reference table. Or actually, just the IPC destruction flag of it, as the actual counter
3974 * of it is better rebuild after a reload/reexec. */
3976 HASHMAP_FOREACH_KEY(p
, k
, *uid_refs
, i
) {
3980 uid
= PTR_TO_UID(k
);
3981 c
= PTR_TO_UINT32(p
);
3983 if (!(c
& DESTROY_IPC_FLAG
))
3986 fprintf(f
, "%s=" UID_FMT
"\n", field_name
, uid
);
3990 void manager_serialize_uid_refs(Manager
*m
, FILE *f
) {
3991 manager_serialize_uid_refs_internal(m
, f
, &m
->uid_refs
, "destroy-ipc-uid");
3994 void manager_serialize_gid_refs(Manager
*m
, FILE *f
) {
3995 manager_serialize_uid_refs_internal(m
, f
, &m
->gid_refs
, "destroy-ipc-gid");
3998 static void manager_deserialize_uid_refs_one_internal(
4001 const char *value
) {
4011 r
= parse_uid(value
, &uid
);
4012 if (r
< 0 || uid
== 0) {
4013 log_debug("Unable to parse UID reference serialization");
4017 r
= hashmap_ensure_allocated(uid_refs
, &trivial_hash_ops
);
4023 c
= PTR_TO_UINT32(hashmap_get(*uid_refs
, UID_TO_PTR(uid
)));
4024 if (c
& DESTROY_IPC_FLAG
)
4027 c
|= DESTROY_IPC_FLAG
;
4029 r
= hashmap_replace(*uid_refs
, UID_TO_PTR(uid
), UINT32_TO_PTR(c
));
4031 log_debug("Failed to add UID reference entry");
4036 void manager_deserialize_uid_refs_one(Manager
*m
, const char *value
) {
4037 manager_deserialize_uid_refs_one_internal(m
, &m
->uid_refs
, value
);
4040 void manager_deserialize_gid_refs_one(Manager
*m
, const char *value
) {
4041 manager_deserialize_uid_refs_one_internal(m
, &m
->gid_refs
, value
);
4044 int manager_dispatch_user_lookup_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
4048 char unit_name
[UNIT_NAME_MAX
+1];
4051 Manager
*m
= userdata
;
4059 /* Invoked whenever a child process succeeded resolving its user/group to use and sent us the resulting UID/GID
4060 * in a datagram. We parse the datagram here and pass it off to the unit, so that it can add a reference to the
4061 * UID/GID so that it can destroy the UID/GID's IPC objects when the reference counter drops to 0. */
4063 l
= recv(fd
, &buffer
, sizeof(buffer
), MSG_DONTWAIT
);
4065 if (IN_SET(errno
, EINTR
, EAGAIN
))
4068 return log_error_errno(errno
, "Failed to read from user lookup fd: %m");
4071 if ((size_t) l
<= offsetof(struct buffer
, unit_name
)) {
4072 log_warning("Received too short user lookup message, ignoring.");
4076 if ((size_t) l
> offsetof(struct buffer
, unit_name
) + UNIT_NAME_MAX
) {
4077 log_warning("Received too long user lookup message, ignoring.");
4081 if (!uid_is_valid(buffer
.uid
) && !gid_is_valid(buffer
.gid
)) {
4082 log_warning("Got user lookup message with invalid UID/GID pair, ignoring.");
4086 n
= (size_t) l
- offsetof(struct buffer
, unit_name
);
4087 if (memchr(buffer
.unit_name
, 0, n
)) {
4088 log_warning("Received lookup message with embedded NUL character, ignoring.");
4092 buffer
.unit_name
[n
] = 0;
4093 u
= manager_get_unit(m
, buffer
.unit_name
);
4095 log_debug("Got user lookup message but unit doesn't exist, ignoring.");
4099 log_unit_debug(u
, "User lookup succeeded: uid=" UID_FMT
" gid=" GID_FMT
, buffer
.uid
, buffer
.gid
);
4101 unit_notify_user_lookup(u
, buffer
.uid
, buffer
.gid
);
4105 char *manager_taint_string(Manager
*m
) {
4106 _cleanup_free_
char *destination
= NULL
, *overflowuid
= NULL
, *overflowgid
= NULL
;
4110 /* Returns a "taint string", e.g. "local-hwclock:var-run-bad".
4111 * Only things that are detected at runtime should be tagged
4112 * here. For stuff that is set during compilation, emit a warning
4113 * in the configuration phase. */
4117 buf
= new(char, sizeof("split-usr:"
4121 "overflowuid-not-65534:"
4122 "overflowgid-not-65534:"));
4130 e
= stpcpy(e
, "split-usr:");
4132 if (access("/proc/cgroups", F_OK
) < 0)
4133 e
= stpcpy(e
, "cgroups-missing:");
4135 if (clock_is_localtime(NULL
) > 0)
4136 e
= stpcpy(e
, "local-hwclock:");
4138 r
= readlink_malloc("/var/run", &destination
);
4139 if (r
< 0 || !PATH_IN_SET(destination
, "../run", "/run"))
4140 e
= stpcpy(e
, "var-run-bad:");
4142 r
= read_one_line_file("/proc/sys/kernel/overflowuid", &overflowuid
);
4143 if (r
>= 0 && !streq(overflowuid
, "65534"))
4144 e
= stpcpy(e
, "overflowuid-not-65534:");
4146 r
= read_one_line_file("/proc/sys/kernel/overflowgid", &overflowgid
);
4147 if (r
>= 0 && !streq(overflowgid
, "65534"))
4148 e
= stpcpy(e
, "overflowgid-not-65534:");
4150 /* remove the last ':' */
4157 void manager_ref_console(Manager
*m
) {
4163 void manager_unref_console(Manager
*m
) {
4165 assert(m
->n_on_console
> 0);
4168 if (m
->n_on_console
== 0)
4169 m
->no_console_output
= false; /* unset no_console_output flag, since the console is definitely free now */
4172 static const char *const manager_state_table
[_MANAGER_STATE_MAX
] = {
4173 [MANAGER_INITIALIZING
] = "initializing",
4174 [MANAGER_STARTING
] = "starting",
4175 [MANAGER_RUNNING
] = "running",
4176 [MANAGER_DEGRADED
] = "degraded",
4177 [MANAGER_MAINTENANCE
] = "maintenance",
4178 [MANAGER_STOPPING
] = "stopping",
4181 DEFINE_STRING_TABLE_LOOKUP(manager_state
, ManagerState
);
4183 static const char *const manager_timestamp_table
[_MANAGER_TIMESTAMP_MAX
] = {
4184 [MANAGER_TIMESTAMP_FIRMWARE
] = "firmware",
4185 [MANAGER_TIMESTAMP_LOADER
] = "loader",
4186 [MANAGER_TIMESTAMP_KERNEL
] = "kernel",
4187 [MANAGER_TIMESTAMP_INITRD
] = "initrd",
4188 [MANAGER_TIMESTAMP_USERSPACE
] = "userspace",
4189 [MANAGER_TIMESTAMP_FINISH
] = "finish",
4190 [MANAGER_TIMESTAMP_SECURITY_START
] = "security-start",
4191 [MANAGER_TIMESTAMP_SECURITY_FINISH
] = "security-finish",
4192 [MANAGER_TIMESTAMP_GENERATORS_START
] = "generators-start",
4193 [MANAGER_TIMESTAMP_GENERATORS_FINISH
] = "generators-finish",
4194 [MANAGER_TIMESTAMP_UNITS_LOAD_START
] = "units-load-start",
4195 [MANAGER_TIMESTAMP_UNITS_LOAD_FINISH
] = "units-load-finish",
4198 DEFINE_STRING_TABLE_LOOKUP(manager_timestamp
, ManagerTimestamp
);