1 /* SPDX-License-Identifier: LGPL-2.1-or-later */
7 #include <sys/inotify.h>
9 #include <sys/reboot.h>
10 #include <sys/timerfd.h>
18 #include "sd-daemon.h"
19 #include "sd-messages.h"
22 #include "all-units.h"
23 #include "alloc-util.h"
25 #include "boot-timestamps.h"
26 #include "bus-common-errors.h"
27 #include "bus-error.h"
28 #include "bus-kernel.h"
30 #include "clean-ipc.h"
31 #include "clock-util.h"
32 #include "core-varlink.h"
34 #include "dbus-manager.h"
35 #include "dbus-unit.h"
38 #include "dirent-util.h"
41 #include "exec-util.h"
43 #include "exit-status.h"
47 #include "generator-setup.h"
52 #include "locale-setup.h"
53 #include "load-fragment.h"
57 #include "memory-util.h"
59 #include "parse-util.h"
60 #include "path-lookup.h"
61 #include "path-util.h"
62 #include "process-util.h"
63 #include "ratelimit.h"
64 #include "rlimit-util.h"
66 #include "selinux-util.h"
67 #include "serialize.h"
68 #include "signal-util.h"
69 #include "socket-util.h"
71 #include "stat-util.h"
72 #include "string-table.h"
73 #include "string-util.h"
76 #include "sysctl-util.h"
77 #include "syslog-util.h"
78 #include "terminal-util.h"
79 #include "time-util.h"
80 #include "transaction.h"
81 #include "umask-util.h"
82 #include "unit-name.h"
83 #include "user-util.h"
87 #define NOTIFY_RCVBUF_SIZE (8*1024*1024)
88 #define CGROUPS_AGENT_RCVBUF_SIZE (8*1024*1024)
90 /* Initial delay and the interval for printing status messages about running jobs */
91 #define JOBS_IN_PROGRESS_WAIT_USEC (2*USEC_PER_SEC)
92 #define JOBS_IN_PROGRESS_QUIET_WAIT_USEC (25*USEC_PER_SEC)
93 #define JOBS_IN_PROGRESS_PERIOD_USEC (USEC_PER_SEC / 3)
94 #define JOBS_IN_PROGRESS_PERIOD_DIVISOR 3
96 /* If there are more than 1K bus messages queue across our API and direct buses, then let's not add more on top until
97 * the queue gets more empty. */
98 #define MANAGER_BUS_BUSY_THRESHOLD 1024LU
100 /* How many units and jobs to process of the bus queue before returning to the event loop. */
101 #define MANAGER_BUS_MESSAGE_BUDGET 100U
103 static int manager_dispatch_notify_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
104 static int manager_dispatch_cgroups_agent_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
105 static int manager_dispatch_signal_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
106 static int manager_dispatch_time_change_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
107 static int manager_dispatch_idle_pipe_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
108 static int manager_dispatch_user_lookup_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
109 static int manager_dispatch_jobs_in_progress(sd_event_source
*source
, usec_t usec
, void *userdata
);
110 static int manager_dispatch_run_queue(sd_event_source
*source
, void *userdata
);
111 static int manager_dispatch_sigchld(sd_event_source
*source
, void *userdata
);
112 static int manager_dispatch_timezone_change(sd_event_source
*source
, const struct inotify_event
*event
, void *userdata
);
113 static int manager_run_environment_generators(Manager
*m
);
114 static int manager_run_generators(Manager
*m
);
115 static void manager_vacuum(Manager
*m
);
117 static usec_t
manager_watch_jobs_next_time(Manager
*m
) {
118 return usec_add(now(CLOCK_MONOTONIC
),
119 show_status_on(m
->show_status
) ? JOBS_IN_PROGRESS_WAIT_USEC
:
120 JOBS_IN_PROGRESS_QUIET_WAIT_USEC
);
123 static void manager_watch_jobs_in_progress(Manager
*m
) {
129 /* We do not want to show the cylon animation if the user
130 * needs to confirm service executions otherwise confirmation
131 * messages will be screwed by the cylon animation. */
132 if (!manager_is_confirm_spawn_disabled(m
))
135 if (m
->jobs_in_progress_event_source
)
138 next
= manager_watch_jobs_next_time(m
);
139 r
= sd_event_add_time(
141 &m
->jobs_in_progress_event_source
,
144 manager_dispatch_jobs_in_progress
, m
);
148 (void) sd_event_source_set_description(m
->jobs_in_progress_event_source
, "manager-jobs-in-progress");
151 #define CYLON_BUFFER_EXTRA (2*STRLEN(ANSI_RED) + STRLEN(ANSI_HIGHLIGHT_RED) + 2*STRLEN(ANSI_NORMAL))
153 static void draw_cylon(char buffer
[], size_t buflen
, unsigned width
, unsigned pos
) {
156 assert(buflen
>= CYLON_BUFFER_EXTRA
+ width
+ 1);
157 assert(pos
<= width
+1); /* 0 or width+1 mean that the center light is behind the corner */
161 p
= mempset(p
, ' ', pos
-2);
162 if (log_get_show_color())
163 p
= stpcpy(p
, ANSI_RED
);
167 if (pos
> 0 && pos
<= width
) {
168 if (log_get_show_color())
169 p
= stpcpy(p
, ANSI_HIGHLIGHT_RED
);
173 if (log_get_show_color())
174 p
= stpcpy(p
, ANSI_NORMAL
);
177 if (log_get_show_color())
178 p
= stpcpy(p
, ANSI_RED
);
181 p
= mempset(p
, ' ', width
-1-pos
);
182 if (log_get_show_color())
183 strcpy(p
, ANSI_NORMAL
);
187 static void manager_flip_auto_status(Manager
*m
, bool enable
, const char *reason
) {
191 if (m
->show_status
== SHOW_STATUS_AUTO
)
192 manager_set_show_status(m
, SHOW_STATUS_TEMPORARY
, reason
);
194 if (m
->show_status
== SHOW_STATUS_TEMPORARY
)
195 manager_set_show_status(m
, SHOW_STATUS_AUTO
, reason
);
199 static void manager_print_jobs_in_progress(Manager
*m
) {
200 _cleanup_free_
char *job_of_n
= NULL
;
202 unsigned counter
= 0, print_nr
;
203 char cylon
[6 + CYLON_BUFFER_EXTRA
+ 1];
205 char time
[FORMAT_TIMESPAN_MAX
], limit
[FORMAT_TIMESPAN_MAX
] = "no limit";
209 assert(m
->n_running_jobs
> 0);
211 manager_flip_auto_status(m
, true, "delay");
213 print_nr
= (m
->jobs_in_progress_iteration
/ JOBS_IN_PROGRESS_PERIOD_DIVISOR
) % m
->n_running_jobs
;
215 HASHMAP_FOREACH(j
, m
->jobs
)
216 if (j
->state
== JOB_RUNNING
&& counter
++ == print_nr
)
219 /* m->n_running_jobs must be consistent with the contents of m->jobs,
220 * so the above loop must have succeeded in finding j. */
221 assert(counter
== print_nr
+ 1);
224 cylon_pos
= m
->jobs_in_progress_iteration
% 14;
226 cylon_pos
= 14 - cylon_pos
;
227 draw_cylon(cylon
, sizeof(cylon
), 6, cylon_pos
);
229 m
->jobs_in_progress_iteration
++;
231 if (m
->n_running_jobs
> 1) {
232 if (asprintf(&job_of_n
, "(%u of %u) ", counter
, m
->n_running_jobs
) < 0)
236 format_timespan(time
, sizeof(time
), now(CLOCK_MONOTONIC
) - j
->begin_usec
, 1*USEC_PER_SEC
);
237 if (job_get_timeout(j
, &x
) > 0)
238 format_timespan(limit
, sizeof(limit
), x
- j
->begin_usec
, 1*USEC_PER_SEC
);
240 manager_status_printf(m
, STATUS_TYPE_EPHEMERAL
, cylon
,
241 "%sA %s job is running for %s (%s / %s)",
243 job_type_to_string(j
->type
),
244 unit_status_string(j
->unit
),
248 static int have_ask_password(void) {
249 _cleanup_closedir_
DIR *dir
;
252 dir
= opendir("/run/systemd/ask-password");
260 FOREACH_DIRENT_ALL(de
, dir
, return -errno
) {
261 if (startswith(de
->d_name
, "ask."))
267 static int manager_dispatch_ask_password_fd(sd_event_source
*source
,
268 int fd
, uint32_t revents
, void *userdata
) {
269 Manager
*m
= userdata
;
275 m
->have_ask_password
= have_ask_password();
276 if (m
->have_ask_password
< 0)
277 /* Log error but continue. Negative have_ask_password
278 * is treated as unknown status. */
279 log_error_errno(m
->have_ask_password
, "Failed to list /run/systemd/ask-password: %m");
284 static void manager_close_ask_password(Manager
*m
) {
287 m
->ask_password_event_source
= sd_event_source_unref(m
->ask_password_event_source
);
288 m
->ask_password_inotify_fd
= safe_close(m
->ask_password_inotify_fd
);
289 m
->have_ask_password
= -EINVAL
;
292 static int manager_check_ask_password(Manager
*m
) {
297 if (!m
->ask_password_event_source
) {
298 assert(m
->ask_password_inotify_fd
< 0);
300 (void) mkdir_p_label("/run/systemd/ask-password", 0755);
302 m
->ask_password_inotify_fd
= inotify_init1(IN_NONBLOCK
|IN_CLOEXEC
);
303 if (m
->ask_password_inotify_fd
< 0)
304 return log_error_errno(errno
, "Failed to create inotify object: %m");
306 r
= inotify_add_watch_and_warn(m
->ask_password_inotify_fd
,
307 "/run/systemd/ask-password",
308 IN_CREATE
|IN_DELETE
|IN_MOVE
);
310 manager_close_ask_password(m
);
314 r
= sd_event_add_io(m
->event
, &m
->ask_password_event_source
,
315 m
->ask_password_inotify_fd
, EPOLLIN
,
316 manager_dispatch_ask_password_fd
, m
);
318 log_error_errno(r
, "Failed to add event source for /run/systemd/ask-password: %m");
319 manager_close_ask_password(m
);
323 (void) sd_event_source_set_description(m
->ask_password_event_source
, "manager-ask-password");
325 /* Queries might have been added meanwhile... */
326 manager_dispatch_ask_password_fd(m
->ask_password_event_source
,
327 m
->ask_password_inotify_fd
, EPOLLIN
, m
);
330 return m
->have_ask_password
;
333 static int manager_watch_idle_pipe(Manager
*m
) {
338 if (m
->idle_pipe_event_source
)
341 if (m
->idle_pipe
[2] < 0)
344 r
= sd_event_add_io(m
->event
, &m
->idle_pipe_event_source
, m
->idle_pipe
[2], EPOLLIN
, manager_dispatch_idle_pipe_fd
, m
);
346 return log_error_errno(r
, "Failed to watch idle pipe: %m");
348 (void) sd_event_source_set_description(m
->idle_pipe_event_source
, "manager-idle-pipe");
353 static void manager_close_idle_pipe(Manager
*m
) {
356 m
->idle_pipe_event_source
= sd_event_source_unref(m
->idle_pipe_event_source
);
358 safe_close_pair(m
->idle_pipe
);
359 safe_close_pair(m
->idle_pipe
+ 2);
362 static int manager_setup_time_change(Manager
*m
) {
367 if (MANAGER_IS_TEST_RUN(m
))
370 m
->time_change_event_source
= sd_event_source_unref(m
->time_change_event_source
);
371 m
->time_change_fd
= safe_close(m
->time_change_fd
);
373 m
->time_change_fd
= time_change_fd();
374 if (m
->time_change_fd
< 0)
375 return log_error_errno(m
->time_change_fd
, "Failed to create timer change timer fd: %m");
377 r
= sd_event_add_io(m
->event
, &m
->time_change_event_source
, m
->time_change_fd
, EPOLLIN
, manager_dispatch_time_change_fd
, m
);
379 return log_error_errno(r
, "Failed to create time change event source: %m");
381 /* Schedule this slightly earlier than the .timer event sources */
382 r
= sd_event_source_set_priority(m
->time_change_event_source
, SD_EVENT_PRIORITY_NORMAL
-1);
384 return log_error_errno(r
, "Failed to set priority of time change event sources: %m");
386 (void) sd_event_source_set_description(m
->time_change_event_source
, "manager-time-change");
388 log_debug("Set up TFD_TIMER_CANCEL_ON_SET timerfd.");
393 static int manager_read_timezone_stat(Manager
*m
) {
399 /* Read the current stat() data of /etc/localtime so that we detect changes */
400 if (lstat("/etc/localtime", &st
) < 0) {
401 log_debug_errno(errno
, "Failed to stat /etc/localtime, ignoring: %m");
402 changed
= m
->etc_localtime_accessible
;
403 m
->etc_localtime_accessible
= false;
407 k
= timespec_load(&st
.st_mtim
);
408 changed
= !m
->etc_localtime_accessible
|| k
!= m
->etc_localtime_mtime
;
410 m
->etc_localtime_mtime
= k
;
411 m
->etc_localtime_accessible
= true;
417 static int manager_setup_timezone_change(Manager
*m
) {
418 _cleanup_(sd_event_source_unrefp
) sd_event_source
*new_event
= NULL
;
423 if (MANAGER_IS_TEST_RUN(m
))
426 /* We watch /etc/localtime for three events: change of the link count (which might mean removal from /etc even
427 * though another link might be kept), renames, and file close operations after writing. Note we don't bother
428 * with IN_DELETE_SELF, as that would just report when the inode is removed entirely, i.e. after the link count
429 * went to zero and all fds to it are closed.
431 * Note that we never follow symlinks here. This is a simplification, but should cover almost all cases
434 * Note that we create the new event source first here, before releasing the old one. This should optimize
435 * behaviour as this way sd-event can reuse the old watch in case the inode didn't change. */
437 r
= sd_event_add_inotify(m
->event
, &new_event
, "/etc/localtime",
438 IN_ATTRIB
|IN_MOVE_SELF
|IN_CLOSE_WRITE
|IN_DONT_FOLLOW
, manager_dispatch_timezone_change
, m
);
440 /* If the file doesn't exist yet, subscribe to /etc instead, and wait until it is created either by
441 * O_CREATE or by rename() */
443 log_debug_errno(r
, "/etc/localtime doesn't exist yet, watching /etc instead.");
444 r
= sd_event_add_inotify(m
->event
, &new_event
, "/etc",
445 IN_CREATE
|IN_MOVED_TO
|IN_ONLYDIR
, manager_dispatch_timezone_change
, m
);
448 return log_error_errno(r
, "Failed to create timezone change event source: %m");
450 /* Schedule this slightly earlier than the .timer event sources */
451 r
= sd_event_source_set_priority(new_event
, SD_EVENT_PRIORITY_NORMAL
-1);
453 return log_error_errno(r
, "Failed to set priority of timezone change event sources: %m");
455 sd_event_source_unref(m
->timezone_change_event_source
);
456 m
->timezone_change_event_source
= TAKE_PTR(new_event
);
461 static int enable_special_signals(Manager
*m
) {
462 _cleanup_close_
int fd
= -1;
466 if (MANAGER_IS_TEST_RUN(m
))
469 /* Enable that we get SIGINT on control-alt-del. In containers
470 * this will fail with EPERM (older) or EINVAL (newer), so
472 if (reboot(RB_DISABLE_CAD
) < 0 && !IN_SET(errno
, EPERM
, EINVAL
))
473 log_warning_errno(errno
, "Failed to enable ctrl-alt-del handling: %m");
475 fd
= open_terminal("/dev/tty0", O_RDWR
|O_NOCTTY
|O_CLOEXEC
);
477 /* Support systems without virtual console */
479 log_warning_errno(errno
, "Failed to open /dev/tty0: %m");
481 /* Enable that we get SIGWINCH on kbrequest */
482 if (ioctl(fd
, KDSIGACCEPT
, SIGWINCH
) < 0)
483 log_warning_errno(errno
, "Failed to enable kbrequest handling: %m");
489 #define RTSIG_IF_AVAILABLE(signum) (signum <= SIGRTMAX ? signum : -1)
491 static int manager_setup_signals(Manager
*m
) {
492 struct sigaction sa
= {
493 .sa_handler
= SIG_DFL
,
494 .sa_flags
= SA_NOCLDSTOP
|SA_RESTART
,
501 assert_se(sigaction(SIGCHLD
, &sa
, NULL
) == 0);
503 /* We make liberal use of realtime signals here. On
504 * Linux/glibc we have 30 of them (with the exception of Linux
505 * on hppa, see below), between SIGRTMIN+0 ... SIGRTMIN+30
508 assert_se(sigemptyset(&mask
) == 0);
509 sigset_add_many(&mask
,
510 SIGCHLD
, /* Child died */
511 SIGTERM
, /* Reexecute daemon */
512 SIGHUP
, /* Reload configuration */
513 SIGUSR1
, /* systemd/upstart: reconnect to D-Bus */
514 SIGUSR2
, /* systemd: dump status */
515 SIGINT
, /* Kernel sends us this on control-alt-del */
516 SIGWINCH
, /* Kernel sends us this on kbrequest (alt-arrowup) */
517 SIGPWR
, /* Some kernel drivers and upsd send us this on power failure */
519 SIGRTMIN
+0, /* systemd: start default.target */
520 SIGRTMIN
+1, /* systemd: isolate rescue.target */
521 SIGRTMIN
+2, /* systemd: isolate emergency.target */
522 SIGRTMIN
+3, /* systemd: start halt.target */
523 SIGRTMIN
+4, /* systemd: start poweroff.target */
524 SIGRTMIN
+5, /* systemd: start reboot.target */
525 SIGRTMIN
+6, /* systemd: start kexec.target */
527 /* ... space for more special targets ... */
529 SIGRTMIN
+13, /* systemd: Immediate halt */
530 SIGRTMIN
+14, /* systemd: Immediate poweroff */
531 SIGRTMIN
+15, /* systemd: Immediate reboot */
532 SIGRTMIN
+16, /* systemd: Immediate kexec */
534 /* ... space for more immediate system state changes ... */
536 SIGRTMIN
+20, /* systemd: enable status messages */
537 SIGRTMIN
+21, /* systemd: disable status messages */
538 SIGRTMIN
+22, /* systemd: set log level to LOG_DEBUG */
539 SIGRTMIN
+23, /* systemd: set log level to LOG_INFO */
540 SIGRTMIN
+24, /* systemd: Immediate exit (--user only) */
542 /* .. one free signal here ... */
544 /* Apparently Linux on hppa had fewer RT signals until v3.18,
545 * SIGRTMAX was SIGRTMIN+25, and then SIGRTMIN was lowered,
546 * see commit v3.17-7614-g1f25df2eff.
548 * We cannot unconditionally make use of those signals here,
549 * so let's use a runtime check. Since these commands are
550 * accessible by different means and only really a safety
551 * net, the missing functionality on hppa shouldn't matter.
554 RTSIG_IF_AVAILABLE(SIGRTMIN
+26), /* systemd: set log target to journal-or-kmsg */
555 RTSIG_IF_AVAILABLE(SIGRTMIN
+27), /* systemd: set log target to console */
556 RTSIG_IF_AVAILABLE(SIGRTMIN
+28), /* systemd: set log target to kmsg */
557 RTSIG_IF_AVAILABLE(SIGRTMIN
+29), /* systemd: set log target to syslog-or-kmsg (obsolete) */
559 /* ... one free signal here SIGRTMIN+30 ... */
561 assert_se(sigprocmask(SIG_SETMASK
, &mask
, NULL
) == 0);
563 m
->signal_fd
= signalfd(-1, &mask
, SFD_NONBLOCK
|SFD_CLOEXEC
);
564 if (m
->signal_fd
< 0)
567 r
= sd_event_add_io(m
->event
, &m
->signal_event_source
, m
->signal_fd
, EPOLLIN
, manager_dispatch_signal_fd
, m
);
571 (void) sd_event_source_set_description(m
->signal_event_source
, "manager-signal");
573 /* Process signals a bit earlier than the rest of things, but later than notify_fd processing, so that the
574 * notify processing can still figure out to which process/service a message belongs, before we reap the
575 * process. Also, process this before handling cgroup notifications, so that we always collect child exit
576 * status information before detecting that there's no process in a cgroup. */
577 r
= sd_event_source_set_priority(m
->signal_event_source
, SD_EVENT_PRIORITY_NORMAL
-6);
581 if (MANAGER_IS_SYSTEM(m
))
582 return enable_special_signals(m
);
587 static char** sanitize_environment(char **l
) {
589 /* Let's remove some environment variables that we need ourselves to communicate with our clients */
593 "CONFIGURATION_DIRECTORY",
594 "CREDENTIALS_DIRECTORY",
616 /* Let's order the environment alphabetically, just to make it pretty */
622 int manager_default_environment(Manager
*m
) {
627 m
->transient_environment
= strv_free(m
->transient_environment
);
629 if (MANAGER_IS_SYSTEM(m
)) {
630 /* The system manager always starts with a clean
631 * environment for its children. It does not import
632 * the kernel's or the parents' exported variables.
634 * The initial passed environment is untouched to keep
635 * /proc/self/environ valid; it is used for tagging
636 * the init process inside containers. */
637 m
->transient_environment
= strv_new("PATH=" DEFAULT_PATH
);
638 if (!m
->transient_environment
)
641 /* Import locale variables LC_*= from configuration */
642 (void) locale_setup(&m
->transient_environment
);
644 _cleanup_free_
char *k
= NULL
;
646 /* The user manager passes its own environment
647 * along to its children, except for $PATH. */
648 m
->transient_environment
= strv_copy(environ
);
649 if (!m
->transient_environment
)
652 k
= strdup("PATH=" DEFAULT_USER_PATH
);
656 r
= strv_env_replace(&m
->transient_environment
, k
);
662 sanitize_environment(m
->transient_environment
);
667 static int manager_setup_prefix(Manager
*m
) {
673 static const struct table_entry paths_system
[_EXEC_DIRECTORY_TYPE_MAX
] = {
674 [EXEC_DIRECTORY_RUNTIME
] = { SD_PATH_SYSTEM_RUNTIME
, NULL
},
675 [EXEC_DIRECTORY_STATE
] = { SD_PATH_SYSTEM_STATE_PRIVATE
, NULL
},
676 [EXEC_DIRECTORY_CACHE
] = { SD_PATH_SYSTEM_STATE_CACHE
, NULL
},
677 [EXEC_DIRECTORY_LOGS
] = { SD_PATH_SYSTEM_STATE_LOGS
, NULL
},
678 [EXEC_DIRECTORY_CONFIGURATION
] = { SD_PATH_SYSTEM_CONFIGURATION
, NULL
},
681 static const struct table_entry paths_user
[_EXEC_DIRECTORY_TYPE_MAX
] = {
682 [EXEC_DIRECTORY_RUNTIME
] = { SD_PATH_USER_RUNTIME
, NULL
},
683 [EXEC_DIRECTORY_STATE
] = { SD_PATH_USER_CONFIGURATION
, NULL
},
684 [EXEC_DIRECTORY_CACHE
] = { SD_PATH_USER_STATE_CACHE
, NULL
},
685 [EXEC_DIRECTORY_LOGS
] = { SD_PATH_USER_CONFIGURATION
, "log" },
686 [EXEC_DIRECTORY_CONFIGURATION
] = { SD_PATH_USER_CONFIGURATION
, NULL
},
691 const struct table_entry
*p
= MANAGER_IS_SYSTEM(m
) ? paths_system
: paths_user
;
694 for (ExecDirectoryType i
= 0; i
< _EXEC_DIRECTORY_TYPE_MAX
; i
++) {
695 r
= sd_path_lookup(p
[i
].type
, p
[i
].suffix
, &m
->prefix
[i
]);
703 static void manager_free_unit_name_maps(Manager
*m
) {
704 m
->unit_id_map
= hashmap_free(m
->unit_id_map
);
705 m
->unit_name_map
= hashmap_free(m
->unit_name_map
);
706 m
->unit_path_cache
= set_free(m
->unit_path_cache
);
707 m
->unit_cache_timestamp_hash
= 0;
710 static int manager_setup_run_queue(Manager
*m
) {
714 assert(!m
->run_queue_event_source
);
716 r
= sd_event_add_defer(m
->event
, &m
->run_queue_event_source
, manager_dispatch_run_queue
, m
);
720 r
= sd_event_source_set_priority(m
->run_queue_event_source
, SD_EVENT_PRIORITY_IDLE
);
724 r
= sd_event_source_set_enabled(m
->run_queue_event_source
, SD_EVENT_OFF
);
728 (void) sd_event_source_set_description(m
->run_queue_event_source
, "manager-run-queue");
733 static int manager_setup_sigchld_event_source(Manager
*m
) {
737 assert(!m
->sigchld_event_source
);
739 r
= sd_event_add_defer(m
->event
, &m
->sigchld_event_source
, manager_dispatch_sigchld
, m
);
743 r
= sd_event_source_set_priority(m
->sigchld_event_source
, SD_EVENT_PRIORITY_NORMAL
-7);
747 r
= sd_event_source_set_enabled(m
->sigchld_event_source
, SD_EVENT_OFF
);
751 (void) sd_event_source_set_description(m
->sigchld_event_source
, "manager-sigchld");
756 int manager_new(UnitFileScope scope
, ManagerTestRunFlags test_run_flags
, Manager
**_m
) {
757 _cleanup_(manager_freep
) Manager
*m
= NULL
;
762 assert(IN_SET(scope
, UNIT_FILE_SYSTEM
, UNIT_FILE_USER
));
769 .unit_file_scope
= scope
,
770 .objective
= _MANAGER_OBJECTIVE_INVALID
,
772 .status_unit_format
= STATUS_UNIT_FORMAT_DEFAULT
,
774 .default_timer_accuracy_usec
= USEC_PER_MINUTE
,
775 .default_memory_accounting
= MEMORY_ACCOUNTING_DEFAULT
,
776 .default_tasks_accounting
= true,
777 .default_tasks_max
= TASKS_MAX_UNSET
,
778 .default_timeout_start_usec
= DEFAULT_TIMEOUT_USEC
,
779 .default_timeout_stop_usec
= DEFAULT_TIMEOUT_USEC
,
780 .default_restart_usec
= DEFAULT_RESTART_USEC
,
782 .original_log_level
= -1,
783 .original_log_target
= _LOG_TARGET_INVALID
,
785 .watchdog_overridden
[WATCHDOG_RUNTIME
] = USEC_INFINITY
,
786 .watchdog_overridden
[WATCHDOG_REBOOT
] = USEC_INFINITY
,
787 .watchdog_overridden
[WATCHDOG_KEXEC
] = USEC_INFINITY
,
789 .show_status_overridden
= _SHOW_STATUS_INVALID
,
792 .cgroups_agent_fd
= -1,
794 .time_change_fd
= -1,
795 .user_lookup_fds
= { -1, -1 },
796 .private_listen_fd
= -1,
798 .cgroup_inotify_fd
= -1,
799 .pin_cgroupfs_fd
= -1,
800 .ask_password_inotify_fd
= -1,
801 .idle_pipe
= { -1, -1, -1, -1},
803 /* start as id #1, so that we can leave #0 around as "null-like" value */
806 .have_ask_password
= -EINVAL
, /* we don't know */
808 .test_run_flags
= test_run_flags
,
810 .default_oom_policy
= OOM_STOP
,
814 if (MANAGER_IS_SYSTEM(m
) && detect_container() <= 0)
815 boot_timestamps(m
->timestamps
+ MANAGER_TIMESTAMP_USERSPACE
,
816 m
->timestamps
+ MANAGER_TIMESTAMP_FIRMWARE
,
817 m
->timestamps
+ MANAGER_TIMESTAMP_LOADER
);
820 /* Prepare log fields we can use for structured logging */
821 if (MANAGER_IS_SYSTEM(m
)) {
822 m
->unit_log_field
= "UNIT=";
823 m
->unit_log_format_string
= "UNIT=%s";
825 m
->invocation_log_field
= "INVOCATION_ID=";
826 m
->invocation_log_format_string
= "INVOCATION_ID=%s";
828 m
->unit_log_field
= "USER_UNIT=";
829 m
->unit_log_format_string
= "USER_UNIT=%s";
831 m
->invocation_log_field
= "USER_INVOCATION_ID=";
832 m
->invocation_log_format_string
= "USER_INVOCATION_ID=%s";
835 /* Reboot immediately if the user hits C-A-D more often than 7x per 2s */
836 m
->ctrl_alt_del_ratelimit
= (RateLimit
) { .interval
= 2 * USEC_PER_SEC
, .burst
= 7 };
838 r
= manager_default_environment(m
);
842 r
= hashmap_ensure_allocated(&m
->units
, &string_hash_ops
);
846 r
= hashmap_ensure_allocated(&m
->cgroup_unit
, &path_hash_ops
);
850 r
= hashmap_ensure_allocated(&m
->watch_bus
, &string_hash_ops
);
854 r
= prioq_ensure_allocated(&m
->run_queue
, compare_job_priority
);
858 r
= manager_setup_prefix(m
);
862 e
= secure_getenv("CREDENTIALS_DIRECTORY");
864 m
->received_credentials
= strdup(e
);
865 if (!m
->received_credentials
)
869 r
= sd_event_default(&m
->event
);
873 r
= manager_setup_run_queue(m
);
877 if (test_run_flags
== MANAGER_TEST_RUN_MINIMAL
) {
878 m
->cgroup_root
= strdup("");
882 r
= manager_setup_signals(m
);
886 r
= manager_setup_cgroup(m
);
890 r
= manager_setup_time_change(m
);
894 r
= manager_read_timezone_stat(m
);
898 (void) manager_setup_timezone_change(m
);
900 r
= manager_setup_sigchld_event_source(m
);
905 if (test_run_flags
== 0) {
906 if (MANAGER_IS_SYSTEM(m
))
907 r
= mkdir_label("/run/systemd/units", 0755);
909 _cleanup_free_
char *units_path
= NULL
;
910 r
= xdg_user_runtime_dir(&units_path
, "/systemd/units");
913 r
= mkdir_p_label(units_path
, 0755);
916 if (r
< 0 && r
!= -EEXIST
)
922 dir_is_empty("/usr") > 0;
924 /* Note that we do not set up the notify fd here. We do that after deserialization,
925 * since they might have gotten serialized across the reexec. */
932 static int manager_setup_notify(Manager
*m
) {
935 if (MANAGER_IS_TEST_RUN(m
))
938 if (m
->notify_fd
< 0) {
939 _cleanup_close_
int fd
= -1;
940 union sockaddr_union sa
;
943 /* First free all secondary fields */
944 m
->notify_socket
= mfree(m
->notify_socket
);
945 m
->notify_event_source
= sd_event_source_unref(m
->notify_event_source
);
947 fd
= socket(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
|SOCK_NONBLOCK
, 0);
949 return log_error_errno(errno
, "Failed to allocate notification socket: %m");
951 fd_inc_rcvbuf(fd
, NOTIFY_RCVBUF_SIZE
);
953 m
->notify_socket
= path_join(m
->prefix
[EXEC_DIRECTORY_RUNTIME
], "systemd/notify");
954 if (!m
->notify_socket
)
957 r
= sockaddr_un_set_path(&sa
.un
, m
->notify_socket
);
959 return log_error_errno(r
, "Notify socket '%s' not valid for AF_UNIX socket address, refusing.",
963 (void) mkdir_parents_label(m
->notify_socket
, 0755);
964 (void) sockaddr_un_unlink(&sa
.un
);
966 r
= mac_selinux_bind(fd
, &sa
.sa
, sa_len
);
968 return log_error_errno(r
, "bind(%s) failed: %m", m
->notify_socket
);
970 r
= setsockopt_int(fd
, SOL_SOCKET
, SO_PASSCRED
, true);
972 return log_error_errno(r
, "SO_PASSCRED failed: %m");
974 m
->notify_fd
= TAKE_FD(fd
);
976 log_debug("Using notification socket %s", m
->notify_socket
);
979 if (!m
->notify_event_source
) {
980 r
= sd_event_add_io(m
->event
, &m
->notify_event_source
, m
->notify_fd
, EPOLLIN
, manager_dispatch_notify_fd
, m
);
982 return log_error_errno(r
, "Failed to allocate notify event source: %m");
984 /* Process notification messages a bit earlier than SIGCHLD, so that we can still identify to which
985 * service an exit message belongs. */
986 r
= sd_event_source_set_priority(m
->notify_event_source
, SD_EVENT_PRIORITY_NORMAL
-8);
988 return log_error_errno(r
, "Failed to set priority of notify event source: %m");
990 (void) sd_event_source_set_description(m
->notify_event_source
, "manager-notify");
996 static int manager_setup_cgroups_agent(Manager
*m
) {
998 static const union sockaddr_union sa
= {
999 .un
.sun_family
= AF_UNIX
,
1000 .un
.sun_path
= "/run/systemd/cgroups-agent",
1004 /* This creates a listening socket we receive cgroups agent messages on. We do not use D-Bus for delivering
1005 * these messages from the cgroups agent binary to PID 1, as the cgroups agent binary is very short-living, and
1006 * each instance of it needs a new D-Bus connection. Since D-Bus connections are SOCK_STREAM/AF_UNIX, on
1007 * overloaded systems the backlog of the D-Bus socket becomes relevant, as not more than the configured number
1008 * of D-Bus connections may be queued until the kernel will start dropping further incoming connections,
1009 * possibly resulting in lost cgroups agent messages. To avoid this, we'll use a private SOCK_DGRAM/AF_UNIX
1010 * socket, where no backlog is relevant as communication may take place without an actual connect() cycle, and
1011 * we thus won't lose messages.
1013 * Note that PID 1 will forward the agent message to system bus, so that the user systemd instance may listen
1014 * to it. The system instance hence listens on this special socket, but the user instances listen on the system
1015 * bus for these messages. */
1017 if (MANAGER_IS_TEST_RUN(m
))
1020 if (!MANAGER_IS_SYSTEM(m
))
1023 r
= cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER
);
1025 return log_error_errno(r
, "Failed to determine whether unified cgroups hierarchy is used: %m");
1026 if (r
> 0) /* We don't need this anymore on the unified hierarchy */
1029 if (m
->cgroups_agent_fd
< 0) {
1030 _cleanup_close_
int fd
= -1;
1032 /* First free all secondary fields */
1033 m
->cgroups_agent_event_source
= sd_event_source_unref(m
->cgroups_agent_event_source
);
1035 fd
= socket(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
|SOCK_NONBLOCK
, 0);
1037 return log_error_errno(errno
, "Failed to allocate cgroups agent socket: %m");
1039 fd_inc_rcvbuf(fd
, CGROUPS_AGENT_RCVBUF_SIZE
);
1041 (void) sockaddr_un_unlink(&sa
.un
);
1043 /* Only allow root to connect to this socket */
1044 RUN_WITH_UMASK(0077)
1045 r
= bind(fd
, &sa
.sa
, SOCKADDR_UN_LEN(sa
.un
));
1047 return log_error_errno(errno
, "bind(%s) failed: %m", sa
.un
.sun_path
);
1049 m
->cgroups_agent_fd
= TAKE_FD(fd
);
1052 if (!m
->cgroups_agent_event_source
) {
1053 r
= sd_event_add_io(m
->event
, &m
->cgroups_agent_event_source
, m
->cgroups_agent_fd
, EPOLLIN
, manager_dispatch_cgroups_agent_fd
, m
);
1055 return log_error_errno(r
, "Failed to allocate cgroups agent event source: %m");
1057 /* Process cgroups notifications early. Note that when the agent notification is received
1058 * we'll just enqueue the unit in the cgroup empty queue, hence pick a high priority than
1059 * that. Also see handling of cgroup inotify for the unified cgroup stuff. */
1060 r
= sd_event_source_set_priority(m
->cgroups_agent_event_source
, SD_EVENT_PRIORITY_NORMAL
-9);
1062 return log_error_errno(r
, "Failed to set priority of cgroups agent event source: %m");
1064 (void) sd_event_source_set_description(m
->cgroups_agent_event_source
, "manager-cgroups-agent");
1070 static int manager_setup_user_lookup_fd(Manager
*m
) {
1075 /* Set up the socket pair used for passing UID/GID resolution results from forked off processes to PID
1076 * 1. Background: we can't do name lookups (NSS) from PID 1, since it might involve IPC and thus activation,
1077 * and we might hence deadlock on ourselves. Hence we do all user/group lookups asynchronously from the forked
1078 * off processes right before executing the binaries to start. In order to be able to clean up any IPC objects
1079 * created by a unit (see RemoveIPC=) we need to know in PID 1 the used UID/GID of the executed processes,
1080 * hence we establish this communication channel so that forked off processes can pass their UID/GID
1081 * information back to PID 1. The forked off processes send their resolved UID/GID to PID 1 in a simple
1082 * datagram, along with their unit name, so that we can share one communication socket pair among all units for
1085 * You might wonder why we need a communication channel for this that is independent of the usual notification
1086 * socket scheme (i.e. $NOTIFY_SOCKET). The primary difference is about trust: data sent via the $NOTIFY_SOCKET
1087 * channel is only accepted if it originates from the right unit and if reception was enabled for it. The user
1088 * lookup socket OTOH is only accessible by PID 1 and its children until they exec(), and always available.
1090 * Note that this function is called under two circumstances: when we first initialize (in which case we
1091 * allocate both the socket pair and the event source to listen on it), and when we deserialize after a reload
1092 * (in which case the socket pair already exists but we still need to allocate the event source for it). */
1094 if (m
->user_lookup_fds
[0] < 0) {
1096 /* Free all secondary fields */
1097 safe_close_pair(m
->user_lookup_fds
);
1098 m
->user_lookup_event_source
= sd_event_source_unref(m
->user_lookup_event_source
);
1100 if (socketpair(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
, 0, m
->user_lookup_fds
) < 0)
1101 return log_error_errno(errno
, "Failed to allocate user lookup socket: %m");
1103 (void) fd_inc_rcvbuf(m
->user_lookup_fds
[0], NOTIFY_RCVBUF_SIZE
);
1106 if (!m
->user_lookup_event_source
) {
1107 r
= sd_event_add_io(m
->event
, &m
->user_lookup_event_source
, m
->user_lookup_fds
[0], EPOLLIN
, manager_dispatch_user_lookup_fd
, m
);
1109 return log_error_errno(errno
, "Failed to allocate user lookup event source: %m");
1111 /* Process even earlier than the notify event source, so that we always know first about valid UID/GID
1113 r
= sd_event_source_set_priority(m
->user_lookup_event_source
, SD_EVENT_PRIORITY_NORMAL
-11);
1115 return log_error_errno(errno
, "Failed to set priority of user lookup event source: %m");
1117 (void) sd_event_source_set_description(m
->user_lookup_event_source
, "user-lookup");
1123 static unsigned manager_dispatch_cleanup_queue(Manager
*m
) {
1129 while ((u
= m
->cleanup_queue
)) {
1130 assert(u
->in_cleanup_queue
);
1140 GC_OFFSET_IN_PATH
, /* This one is on the path we were traveling */
1141 GC_OFFSET_UNSURE
, /* No clue */
1142 GC_OFFSET_GOOD
, /* We still need this unit */
1143 GC_OFFSET_BAD
, /* We don't need this unit anymore */
1147 static void unit_gc_mark_good(Unit
*u
, unsigned gc_marker
) {
1151 u
->gc_marker
= gc_marker
+ GC_OFFSET_GOOD
;
1153 /* Recursively mark referenced units as GOOD as well */
1154 HASHMAP_FOREACH_KEY(v
, other
, u
->dependencies
[UNIT_REFERENCES
])
1155 if (other
->gc_marker
== gc_marker
+ GC_OFFSET_UNSURE
)
1156 unit_gc_mark_good(other
, gc_marker
);
1159 static void unit_gc_sweep(Unit
*u
, unsigned gc_marker
) {
1166 if (IN_SET(u
->gc_marker
- gc_marker
,
1167 GC_OFFSET_GOOD
, GC_OFFSET_BAD
, GC_OFFSET_UNSURE
, GC_OFFSET_IN_PATH
))
1170 if (u
->in_cleanup_queue
)
1173 if (!unit_may_gc(u
))
1176 u
->gc_marker
= gc_marker
+ GC_OFFSET_IN_PATH
;
1180 HASHMAP_FOREACH_KEY(v
, other
, u
->dependencies
[UNIT_REFERENCED_BY
]) {
1181 unit_gc_sweep(other
, gc_marker
);
1183 if (other
->gc_marker
== gc_marker
+ GC_OFFSET_GOOD
)
1186 if (other
->gc_marker
!= gc_marker
+ GC_OFFSET_BAD
)
1190 if (u
->refs_by_target
) {
1193 LIST_FOREACH(refs_by_target
, ref
, u
->refs_by_target
) {
1194 unit_gc_sweep(ref
->source
, gc_marker
);
1196 if (ref
->source
->gc_marker
== gc_marker
+ GC_OFFSET_GOOD
)
1199 if (ref
->source
->gc_marker
!= gc_marker
+ GC_OFFSET_BAD
)
1207 /* We were unable to find anything out about this entry, so
1208 * let's investigate it later */
1209 u
->gc_marker
= gc_marker
+ GC_OFFSET_UNSURE
;
1210 unit_add_to_gc_queue(u
);
1214 /* We definitely know that this one is not useful anymore, so
1215 * let's mark it for deletion */
1216 u
->gc_marker
= gc_marker
+ GC_OFFSET_BAD
;
1217 unit_add_to_cleanup_queue(u
);
1221 unit_gc_mark_good(u
, gc_marker
);
1224 static unsigned manager_dispatch_gc_unit_queue(Manager
*m
) {
1225 unsigned n
= 0, gc_marker
;
1230 /* log_debug("Running GC..."); */
1232 m
->gc_marker
+= _GC_OFFSET_MAX
;
1233 if (m
->gc_marker
+ _GC_OFFSET_MAX
<= _GC_OFFSET_MAX
)
1236 gc_marker
= m
->gc_marker
;
1238 while ((u
= m
->gc_unit_queue
)) {
1239 assert(u
->in_gc_queue
);
1241 unit_gc_sweep(u
, gc_marker
);
1243 LIST_REMOVE(gc_queue
, m
->gc_unit_queue
, u
);
1244 u
->in_gc_queue
= false;
1248 if (IN_SET(u
->gc_marker
- gc_marker
,
1249 GC_OFFSET_BAD
, GC_OFFSET_UNSURE
)) {
1251 log_unit_debug(u
, "Collecting.");
1252 u
->gc_marker
= gc_marker
+ GC_OFFSET_BAD
;
1253 unit_add_to_cleanup_queue(u
);
1260 static unsigned manager_dispatch_gc_job_queue(Manager
*m
) {
1266 while ((j
= m
->gc_job_queue
)) {
1267 assert(j
->in_gc_queue
);
1269 LIST_REMOVE(gc_queue
, m
->gc_job_queue
, j
);
1270 j
->in_gc_queue
= false;
1277 log_unit_debug(j
->unit
, "Collecting job.");
1278 (void) job_finish_and_invalidate(j
, JOB_COLLECTED
, false, false);
1284 static unsigned manager_dispatch_stop_when_unneeded_queue(Manager
*m
) {
1291 while ((u
= m
->stop_when_unneeded_queue
)) {
1292 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
1293 assert(m
->stop_when_unneeded_queue
);
1295 assert(u
->in_stop_when_unneeded_queue
);
1296 LIST_REMOVE(stop_when_unneeded_queue
, m
->stop_when_unneeded_queue
, u
);
1297 u
->in_stop_when_unneeded_queue
= false;
1301 if (!unit_is_unneeded(u
))
1304 log_unit_debug(u
, "Unit is not needed anymore.");
1306 /* If stopping a unit fails continuously we might enter a stop loop here, hence stop acting on the
1307 * service being unnecessary after a while. */
1309 if (!ratelimit_below(&u
->auto_stop_ratelimit
)) {
1310 log_unit_warning(u
, "Unit not needed anymore, but not stopping since we tried this too often recently.");
1314 /* Ok, nobody needs us anymore. Sniff. Then let's commit suicide */
1315 r
= manager_add_job(u
->manager
, JOB_STOP
, u
, JOB_FAIL
, NULL
, &error
, NULL
);
1317 log_unit_warning_errno(u
, r
, "Failed to enqueue stop job, ignoring: %s", bus_error_message(&error
, r
));
1323 static void manager_clear_jobs_and_units(Manager
*m
) {
1328 while ((u
= hashmap_first(m
->units
)))
1331 manager_dispatch_cleanup_queue(m
);
1333 assert(!m
->load_queue
);
1334 assert(prioq_isempty(m
->run_queue
));
1335 assert(!m
->dbus_unit_queue
);
1336 assert(!m
->dbus_job_queue
);
1337 assert(!m
->cleanup_queue
);
1338 assert(!m
->gc_unit_queue
);
1339 assert(!m
->gc_job_queue
);
1340 assert(!m
->stop_when_unneeded_queue
);
1342 assert(hashmap_isempty(m
->jobs
));
1343 assert(hashmap_isempty(m
->units
));
1345 m
->n_on_console
= 0;
1346 m
->n_running_jobs
= 0;
1347 m
->n_installed_jobs
= 0;
1348 m
->n_failed_jobs
= 0;
1351 Manager
* manager_free(Manager
*m
) {
1355 manager_clear_jobs_and_units(m
);
1357 for (UnitType c
= 0; c
< _UNIT_TYPE_MAX
; c
++)
1358 if (unit_vtable
[c
]->shutdown
)
1359 unit_vtable
[c
]->shutdown(m
);
1361 /* Keep the cgroup hierarchy in place except when we know we are going down for good */
1362 manager_shutdown_cgroup(m
, IN_SET(m
->objective
, MANAGER_EXIT
, MANAGER_REBOOT
, MANAGER_POWEROFF
, MANAGER_HALT
, MANAGER_KEXEC
));
1364 lookup_paths_flush_generator(&m
->lookup_paths
);
1367 manager_varlink_done(m
);
1369 exec_runtime_vacuum(m
);
1370 hashmap_free(m
->exec_runtime_by_id
);
1372 dynamic_user_vacuum(m
, false);
1373 hashmap_free(m
->dynamic_users
);
1375 hashmap_free(m
->units
);
1376 hashmap_free(m
->units_by_invocation_id
);
1377 hashmap_free(m
->jobs
);
1378 hashmap_free(m
->watch_pids
);
1379 hashmap_free(m
->watch_bus
);
1381 prioq_free(m
->run_queue
);
1383 set_free(m
->startup_units
);
1384 set_free(m
->failed_units
);
1386 sd_event_source_unref(m
->signal_event_source
);
1387 sd_event_source_unref(m
->sigchld_event_source
);
1388 sd_event_source_unref(m
->notify_event_source
);
1389 sd_event_source_unref(m
->cgroups_agent_event_source
);
1390 sd_event_source_unref(m
->time_change_event_source
);
1391 sd_event_source_unref(m
->timezone_change_event_source
);
1392 sd_event_source_unref(m
->jobs_in_progress_event_source
);
1393 sd_event_source_unref(m
->run_queue_event_source
);
1394 sd_event_source_unref(m
->user_lookup_event_source
);
1396 safe_close(m
->signal_fd
);
1397 safe_close(m
->notify_fd
);
1398 safe_close(m
->cgroups_agent_fd
);
1399 safe_close(m
->time_change_fd
);
1400 safe_close_pair(m
->user_lookup_fds
);
1402 manager_close_ask_password(m
);
1404 manager_close_idle_pipe(m
);
1406 sd_event_unref(m
->event
);
1408 free(m
->notify_socket
);
1410 lookup_paths_free(&m
->lookup_paths
);
1411 strv_free(m
->transient_environment
);
1412 strv_free(m
->client_environment
);
1414 hashmap_free(m
->cgroup_unit
);
1415 manager_free_unit_name_maps(m
);
1417 free(m
->switch_root
);
1418 free(m
->switch_root_init
);
1420 rlimit_free_all(m
->rlimit
);
1422 assert(hashmap_isempty(m
->units_requiring_mounts_for
));
1423 hashmap_free(m
->units_requiring_mounts_for
);
1425 hashmap_free(m
->uid_refs
);
1426 hashmap_free(m
->gid_refs
);
1428 for (ExecDirectoryType dt
= 0; dt
< _EXEC_DIRECTORY_TYPE_MAX
; dt
++)
1429 m
->prefix
[dt
] = mfree(m
->prefix
[dt
]);
1430 free(m
->received_credentials
);
1435 static void manager_enumerate_perpetual(Manager
*m
) {
1438 if (m
->test_run_flags
== MANAGER_TEST_RUN_MINIMAL
)
1441 /* Let's ask every type to load all units from disk/kernel that it might know */
1442 for (UnitType c
= 0; c
< _UNIT_TYPE_MAX
; c
++) {
1443 if (!unit_type_supported(c
)) {
1444 log_debug("Unit type .%s is not supported on this system.", unit_type_to_string(c
));
1448 if (unit_vtable
[c
]->enumerate_perpetual
)
1449 unit_vtable
[c
]->enumerate_perpetual(m
);
1453 static void manager_enumerate(Manager
*m
) {
1456 if (m
->test_run_flags
== MANAGER_TEST_RUN_MINIMAL
)
1459 /* Let's ask every type to load all units from disk/kernel that it might know */
1460 for (UnitType c
= 0; c
< _UNIT_TYPE_MAX
; c
++) {
1461 if (!unit_type_supported(c
)) {
1462 log_debug("Unit type .%s is not supported on this system.", unit_type_to_string(c
));
1466 if (unit_vtable
[c
]->enumerate
)
1467 unit_vtable
[c
]->enumerate(m
);
1470 manager_dispatch_load_queue(m
);
1473 static void manager_coldplug(Manager
*m
) {
1480 log_debug("Invoking unit coldplug() handlers…");
1482 /* Let's place the units back into their deserialized state */
1483 HASHMAP_FOREACH_KEY(u
, k
, m
->units
) {
1485 /* ignore aliases */
1489 r
= unit_coldplug(u
);
1491 log_warning_errno(r
, "We couldn't coldplug %s, proceeding anyway: %m", u
->id
);
1495 static void manager_catchup(Manager
*m
) {
1501 log_debug("Invoking unit catchup() handlers…");
1503 /* Let's catch up on any state changes that happened while we were reloading/reexecing */
1504 HASHMAP_FOREACH_KEY(u
, k
, m
->units
) {
1506 /* ignore aliases */
1514 static void manager_distribute_fds(Manager
*m
, FDSet
*fds
) {
1519 HASHMAP_FOREACH(u
, m
->units
) {
1521 if (fdset_size(fds
) <= 0)
1524 if (!UNIT_VTABLE(u
)->distribute_fds
)
1527 UNIT_VTABLE(u
)->distribute_fds(u
, fds
);
1531 static bool manager_dbus_is_running(Manager
*m
, bool deserialized
) {
1536 /* This checks whether the dbus instance we are supposed to expose our APIs on is up. We check both the socket
1537 * and the service unit. If the 'deserialized' parameter is true we'll check the deserialized state of the unit
1538 * rather than the current one. */
1540 if (MANAGER_IS_TEST_RUN(m
))
1543 u
= manager_get_unit(m
, SPECIAL_DBUS_SOCKET
);
1546 if ((deserialized
? SOCKET(u
)->deserialized_state
: SOCKET(u
)->state
) != SOCKET_RUNNING
)
1549 u
= manager_get_unit(m
, SPECIAL_DBUS_SERVICE
);
1552 if (!IN_SET((deserialized
? SERVICE(u
)->deserialized_state
: SERVICE(u
)->state
), SERVICE_RUNNING
, SERVICE_RELOAD
))
1558 static void manager_setup_bus(Manager
*m
) {
1561 /* Let's set up our private bus connection now, unconditionally */
1562 (void) bus_init_private(m
);
1564 /* If we are in --user mode also connect to the system bus now */
1565 if (MANAGER_IS_USER(m
))
1566 (void) bus_init_system(m
);
1568 /* Let's connect to the bus now, but only if the unit is supposed to be up */
1569 if (manager_dbus_is_running(m
, MANAGER_IS_RELOADING(m
))) {
1570 (void) bus_init_api(m
);
1572 if (MANAGER_IS_SYSTEM(m
))
1573 (void) bus_init_system(m
);
1577 static void manager_preset_all(Manager
*m
) {
1582 if (m
->first_boot
<= 0)
1585 if (!MANAGER_IS_SYSTEM(m
))
1588 if (MANAGER_IS_TEST_RUN(m
))
1591 /* If this is the first boot, and we are in the host system, then preset everything */
1592 r
= unit_file_preset_all(UNIT_FILE_SYSTEM
, 0, NULL
, UNIT_FILE_PRESET_ENABLE_ONLY
, NULL
, 0);
1594 log_full_errno(r
== -EEXIST
? LOG_NOTICE
: LOG_WARNING
, r
,
1595 "Failed to populate /etc with preset unit settings, ignoring: %m");
1597 log_info("Populated /etc with preset unit settings.");
1600 static void manager_ready(Manager
*m
) {
1603 /* After having loaded everything, do the final round of catching up with what might have changed */
1605 m
->objective
= MANAGER_OK
; /* Tell everyone we are up now */
1607 /* It might be safe to log to the journal now and connect to dbus */
1608 manager_recheck_journal(m
);
1609 manager_recheck_dbus(m
);
1611 /* Let's finally catch up with any changes that took place while we were reloading/reexecing */
1614 m
->honor_device_enumeration
= true;
1617 static Manager
* manager_reloading_start(Manager
*m
) {
1621 static void manager_reloading_stopp(Manager
**m
) {
1623 assert((*m
)->n_reloading
> 0);
1624 (*m
)->n_reloading
--;
1628 int manager_startup(Manager
*m
, FILE *serialization
, FDSet
*fds
) {
1633 /* If we are running in test mode, we still want to run the generators,
1634 * but we should not touch the real generator directories. */
1635 r
= lookup_paths_init(&m
->lookup_paths
, m
->unit_file_scope
,
1636 MANAGER_IS_TEST_RUN(m
) ? LOOKUP_PATHS_TEMPORARY_GENERATED
: 0,
1639 return log_error_errno(r
, "Failed to initialize path lookup table: %m");
1641 dual_timestamp_get(m
->timestamps
+ manager_timestamp_initrd_mangle(MANAGER_TIMESTAMP_GENERATORS_START
));
1642 r
= manager_run_environment_generators(m
);
1644 r
= manager_run_generators(m
);
1645 dual_timestamp_get(m
->timestamps
+ manager_timestamp_initrd_mangle(MANAGER_TIMESTAMP_GENERATORS_FINISH
));
1649 manager_preset_all(m
);
1651 lookup_paths_log(&m
->lookup_paths
);
1654 /* This block is (optionally) done with the reloading counter bumped */
1655 _cleanup_(manager_reloading_stopp
) Manager
*reloading
= NULL
;
1657 /* If we will deserialize make sure that during enumeration this is already known, so we increase the
1658 * counter here already */
1660 reloading
= manager_reloading_start(m
);
1662 /* First, enumerate what we can from all config files */
1663 dual_timestamp_get(m
->timestamps
+ manager_timestamp_initrd_mangle(MANAGER_TIMESTAMP_UNITS_LOAD_START
));
1664 manager_enumerate_perpetual(m
);
1665 manager_enumerate(m
);
1666 dual_timestamp_get(m
->timestamps
+ manager_timestamp_initrd_mangle(MANAGER_TIMESTAMP_UNITS_LOAD_FINISH
));
1668 /* Second, deserialize if there is something to deserialize */
1669 if (serialization
) {
1670 r
= manager_deserialize(m
, serialization
, fds
);
1672 return log_error_errno(r
, "Deserialization failed: %m");
1675 /* Any fds left? Find some unit which wants them. This is useful to allow container managers to pass
1676 * some file descriptors to us pre-initialized. This enables socket-based activation of entire
1678 manager_distribute_fds(m
, fds
);
1680 /* We might have deserialized the notify fd, but if we didn't then let's create the bus now */
1681 r
= manager_setup_notify(m
);
1683 /* No sense to continue without notifications, our children would fail anyway. */
1686 r
= manager_setup_cgroups_agent(m
);
1688 /* Likewise, no sense to continue without empty cgroup notifications. */
1691 r
= manager_setup_user_lookup_fd(m
);
1693 /* This shouldn't fail, except if things are really broken. */
1696 /* Connect to the bus if we are good for it */
1697 manager_setup_bus(m
);
1699 /* Now that we are connected to all possible buses, let's deserialize who is tracking us. */
1700 r
= bus_track_coldplug(m
, &m
->subscribed
, false, m
->deserialized_subscribed
);
1702 log_warning_errno(r
, "Failed to deserialized tracked clients, ignoring: %m");
1703 m
->deserialized_subscribed
= strv_free(m
->deserialized_subscribed
);
1705 r
= manager_varlink_init(m
);
1707 log_warning_errno(r
, "Failed to set up Varlink server, ignoring: %m");
1709 /* Third, fire things up! */
1710 manager_coldplug(m
);
1712 /* Clean up runtime objects */
1716 /* Let's wait for the UnitNew/JobNew messages being sent, before we notify that the
1717 * reload is finished */
1718 m
->send_reloading_done
= true;
1726 int manager_add_job(
1732 sd_bus_error
*error
,
1739 assert(type
< _JOB_TYPE_MAX
);
1741 assert(mode
< _JOB_MODE_MAX
);
1743 if (mode
== JOB_ISOLATE
&& type
!= JOB_START
)
1744 return sd_bus_error_setf(error
, SD_BUS_ERROR_INVALID_ARGS
, "Isolate is only valid for start.");
1746 if (mode
== JOB_ISOLATE
&& !unit
->allow_isolate
)
1747 return sd_bus_error_setf(error
, BUS_ERROR_NO_ISOLATION
, "Operation refused, unit may not be isolated.");
1749 if (mode
== JOB_TRIGGERING
&& type
!= JOB_STOP
)
1750 return sd_bus_error_setf(error
, SD_BUS_ERROR_INVALID_ARGS
, "--job-mode=triggering is only valid for stop.");
1752 log_unit_debug(unit
, "Trying to enqueue job %s/%s/%s", unit
->id
, job_type_to_string(type
), job_mode_to_string(mode
));
1754 type
= job_type_collapse(type
, unit
);
1756 tr
= transaction_new(mode
== JOB_REPLACE_IRREVERSIBLY
);
1760 r
= transaction_add_job_and_dependencies(tr
, type
, unit
, NULL
, true, false,
1761 IN_SET(mode
, JOB_IGNORE_DEPENDENCIES
, JOB_IGNORE_REQUIREMENTS
),
1762 mode
== JOB_IGNORE_DEPENDENCIES
, error
);
1766 if (mode
== JOB_ISOLATE
) {
1767 r
= transaction_add_isolate_jobs(tr
, m
);
1772 if (mode
== JOB_TRIGGERING
) {
1773 r
= transaction_add_triggering_jobs(tr
, unit
);
1778 r
= transaction_activate(tr
, m
, mode
, affected_jobs
, error
);
1782 log_unit_debug(unit
,
1783 "Enqueued job %s/%s as %u", unit
->id
,
1784 job_type_to_string(type
), (unsigned) tr
->anchor_job
->id
);
1787 *ret
= tr
->anchor_job
;
1789 transaction_free(tr
);
1793 transaction_abort(tr
);
1794 transaction_free(tr
);
1798 int manager_add_job_by_name(Manager
*m
, JobType type
, const char *name
, JobMode mode
, Set
*affected_jobs
, sd_bus_error
*e
, Job
**ret
) {
1799 Unit
*unit
= NULL
; /* just to appease gcc, initialization is not really necessary */
1803 assert(type
< _JOB_TYPE_MAX
);
1805 assert(mode
< _JOB_MODE_MAX
);
1807 r
= manager_load_unit(m
, name
, NULL
, NULL
, &unit
);
1812 return manager_add_job(m
, type
, unit
, mode
, affected_jobs
, e
, ret
);
1815 int manager_add_job_by_name_and_warn(Manager
*m
, JobType type
, const char *name
, JobMode mode
, Set
*affected_jobs
, Job
**ret
) {
1816 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
1820 assert(type
< _JOB_TYPE_MAX
);
1822 assert(mode
< _JOB_MODE_MAX
);
1824 r
= manager_add_job_by_name(m
, type
, name
, mode
, affected_jobs
, &error
, ret
);
1826 return log_warning_errno(r
, "Failed to enqueue %s job for %s: %s", job_mode_to_string(mode
), name
, bus_error_message(&error
, r
));
1831 int manager_propagate_reload(Manager
*m
, Unit
*unit
, JobMode mode
, sd_bus_error
*e
) {
1837 assert(mode
< _JOB_MODE_MAX
);
1838 assert(mode
!= JOB_ISOLATE
); /* Isolate is only valid for start */
1840 tr
= transaction_new(mode
== JOB_REPLACE_IRREVERSIBLY
);
1844 /* We need an anchor job */
1845 r
= transaction_add_job_and_dependencies(tr
, JOB_NOP
, unit
, NULL
, false, false, true, true, e
);
1849 /* Failure in adding individual dependencies is ignored, so this always succeeds. */
1850 transaction_add_propagate_reload_jobs(tr
, unit
, tr
->anchor_job
, mode
== JOB_IGNORE_DEPENDENCIES
, e
);
1852 r
= transaction_activate(tr
, m
, mode
, NULL
, e
);
1856 transaction_free(tr
);
1860 transaction_abort(tr
);
1861 transaction_free(tr
);
1865 Job
*manager_get_job(Manager
*m
, uint32_t id
) {
1868 return hashmap_get(m
->jobs
, UINT32_TO_PTR(id
));
1871 Unit
*manager_get_unit(Manager
*m
, const char *name
) {
1875 return hashmap_get(m
->units
, name
);
1878 static int manager_dispatch_target_deps_queue(Manager
*m
) {
1882 static const UnitDependency deps
[] = {
1891 while ((u
= m
->target_deps_queue
)) {
1892 assert(u
->in_target_deps_queue
);
1894 LIST_REMOVE(target_deps_queue
, u
->manager
->target_deps_queue
, u
);
1895 u
->in_target_deps_queue
= false;
1897 for (size_t k
= 0; k
< ELEMENTSOF(deps
); k
++) {
1901 HASHMAP_FOREACH_KEY(v
, target
, u
->dependencies
[deps
[k
]]) {
1902 r
= unit_add_default_target_dependency(u
, target
);
1912 unsigned manager_dispatch_load_queue(Manager
*m
) {
1918 /* Make sure we are not run recursively */
1919 if (m
->dispatching_load_queue
)
1922 m
->dispatching_load_queue
= true;
1924 /* Dispatches the load queue. Takes a unit from the queue and
1925 * tries to load its data until the queue is empty */
1927 while ((u
= m
->load_queue
)) {
1928 assert(u
->in_load_queue
);
1934 m
->dispatching_load_queue
= false;
1936 /* Dispatch the units waiting for their target dependencies to be added now, as all targets that we know about
1937 * should be loaded and have aliases resolved */
1938 (void) manager_dispatch_target_deps_queue(m
);
1943 bool manager_unit_cache_should_retry_load(Unit
*u
) {
1946 /* Automatic reloading from disk only applies to units which were not found sometime in the past, and
1947 * the not-found stub is kept pinned in the unit graph by dependencies. For units that were
1948 * previously loaded, we don't do automatic reloading, and daemon-reload is necessary to update. */
1949 if (u
->load_state
!= UNIT_NOT_FOUND
)
1952 /* The cache has been updated since the last time we tried to load the unit. There might be new
1953 * fragment paths to read. */
1954 if (u
->manager
->unit_cache_timestamp_hash
!= u
->fragment_not_found_timestamp_hash
)
1957 /* The cache needs to be updated because there are modifications on disk. */
1958 return !lookup_paths_timestamp_hash_same(&u
->manager
->lookup_paths
, u
->manager
->unit_cache_timestamp_hash
, NULL
);
1961 int manager_load_unit_prepare(
1968 _cleanup_(unit_freep
) Unit
*cleanup_ret
= NULL
;
1976 /* This will prepare the unit for loading, but not actually load anything from disk. */
1978 if (path
&& !path_is_absolute(path
))
1979 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
, "Path %s is not absolute.", path
);
1982 /* 'name' and 'path' must not both be null. Check here 'path' using assert_se() to
1983 * workaround a bug in gcc that generates a -Wnonnull warning when calling basename(),
1984 * but this cannot be possible in any code path (See #6119). */
1986 name
= basename(path
);
1989 t
= unit_name_to_type(name
);
1991 if (t
== _UNIT_TYPE_INVALID
|| !unit_name_is_valid(name
, UNIT_NAME_PLAIN
|UNIT_NAME_INSTANCE
)) {
1992 if (unit_name_is_valid(name
, UNIT_NAME_TEMPLATE
))
1993 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
, "Unit name %s is missing the instance name.", name
);
1995 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
, "Unit name %s is not valid.", name
);
1998 ret
= manager_get_unit(m
, name
);
2000 /* The time-based cache allows to start new units without daemon-reload,
2001 * but if they are already referenced (because of dependencies or ordering)
2002 * then we have to force a load of the fragment. As an optimization, check
2003 * first if anything in the usual paths was modified since the last time
2004 * the cache was loaded. Also check if the last time an attempt to load the
2005 * unit was made was before the most recent cache refresh, so that we know
2006 * we need to try again — even if the cache is current, it might have been
2007 * updated in a different context before we had a chance to retry loading
2008 * this particular unit. */
2009 if (manager_unit_cache_should_retry_load(ret
))
2010 ret
->load_state
= UNIT_STUB
;
2016 ret
= cleanup_ret
= unit_new(m
, unit_vtable
[t
]->object_size
);
2022 r
= free_and_strdup(&ret
->fragment_path
, path
);
2027 r
= unit_add_name(ret
, name
);
2031 unit_add_to_load_queue(ret
);
2032 unit_add_to_dbus_queue(ret
);
2033 unit_add_to_gc_queue(ret
);
2041 int manager_load_unit(
2053 /* This will load the service information files, but not actually
2054 * start any services or anything. */
2056 r
= manager_load_unit_prepare(m
, name
, path
, e
, _ret
);
2060 manager_dispatch_load_queue(m
);
2062 *_ret
= unit_follow_merge(*_ret
);
2066 int manager_load_startable_unit_or_warn(
2072 /* Load a unit, make sure it loaded fully and is not masked. */
2074 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
2078 r
= manager_load_unit(m
, name
, path
, &error
, &unit
);
2080 return log_error_errno(r
, "Failed to load %s %s: %s",
2081 name
? "unit" : "unit file", name
?: path
,
2082 bus_error_message(&error
, r
));
2084 r
= bus_unit_validate_load_state(unit
, &error
);
2086 return log_error_errno(r
, "%s", bus_error_message(&error
, r
));
2092 void manager_dump_jobs(Manager
*s
, FILE *f
, const char *prefix
) {
2098 HASHMAP_FOREACH(j
, s
->jobs
)
2099 job_dump(j
, f
, prefix
);
2102 void manager_dump_units(Manager
*s
, FILE *f
, const char *prefix
) {
2109 HASHMAP_FOREACH_KEY(u
, t
, s
->units
)
2111 unit_dump(u
, f
, prefix
);
2114 void manager_dump(Manager
*m
, FILE *f
, const char *prefix
) {
2118 for (ManagerTimestamp q
= 0; q
< _MANAGER_TIMESTAMP_MAX
; q
++) {
2119 const dual_timestamp
*t
= m
->timestamps
+ q
;
2120 char buf
[CONST_MAX(FORMAT_TIMESPAN_MAX
, FORMAT_TIMESTAMP_MAX
)];
2122 if (dual_timestamp_is_set(t
))
2123 fprintf(f
, "%sTimestamp %s: %s\n",
2125 manager_timestamp_to_string(q
),
2126 timestamp_is_set(t
->realtime
) ? format_timestamp(buf
, sizeof buf
, t
->realtime
) :
2127 format_timespan(buf
, sizeof buf
, t
->monotonic
, 1));
2130 manager_dump_units(m
, f
, prefix
);
2131 manager_dump_jobs(m
, f
, prefix
);
2134 int manager_get_dump_string(Manager
*m
, char **ret
) {
2135 _cleanup_free_
char *dump
= NULL
;
2136 _cleanup_fclose_
FILE *f
= NULL
;
2143 f
= open_memstream_unlocked(&dump
, &size
);
2147 manager_dump(m
, f
, NULL
);
2149 r
= fflush_and_check(f
);
2155 *ret
= TAKE_PTR(dump
);
2160 void manager_clear_jobs(Manager
*m
) {
2165 while ((j
= hashmap_first(m
->jobs
)))
2166 /* No need to recurse. We're cancelling all jobs. */
2167 job_finish_and_invalidate(j
, JOB_CANCELED
, false, false);
2170 void manager_unwatch_pid(Manager
*m
, pid_t pid
) {
2173 /* First let's drop the unit keyed as "pid". */
2174 (void) hashmap_remove(m
->watch_pids
, PID_TO_PTR(pid
));
2176 /* Then, let's also drop the array keyed by -pid. */
2177 free(hashmap_remove(m
->watch_pids
, PID_TO_PTR(-pid
)));
2180 static int manager_dispatch_run_queue(sd_event_source
*source
, void *userdata
) {
2181 Manager
*m
= userdata
;
2187 while ((j
= prioq_peek(m
->run_queue
))) {
2188 assert(j
->installed
);
2189 assert(j
->in_run_queue
);
2191 (void) job_run_and_invalidate(j
);
2194 if (m
->n_running_jobs
> 0)
2195 manager_watch_jobs_in_progress(m
);
2197 if (m
->n_on_console
> 0)
2198 manager_watch_idle_pipe(m
);
2203 static unsigned manager_dispatch_dbus_queue(Manager
*m
) {
2204 unsigned n
= 0, budget
;
2210 /* When we are reloading, let's not wait with generating signals, since we need to exit the manager as quickly
2211 * as we can. There's no point in throttling generation of signals in that case. */
2212 if (MANAGER_IS_RELOADING(m
) || m
->send_reloading_done
|| m
->pending_reload_message
)
2213 budget
= (unsigned) -1; /* infinite budget in this case */
2215 /* Anything to do at all? */
2216 if (!m
->dbus_unit_queue
&& !m
->dbus_job_queue
)
2219 /* Do we have overly many messages queued at the moment? If so, let's not enqueue more on top, let's
2220 * sit this cycle out, and process things in a later cycle when the queues got a bit emptier. */
2221 if (manager_bus_n_queued_write(m
) > MANAGER_BUS_BUSY_THRESHOLD
)
2224 /* Only process a certain number of units/jobs per event loop iteration. Even if the bus queue wasn't
2225 * overly full before this call we shouldn't increase it in size too wildly in one step, and we
2226 * shouldn't monopolize CPU time with generating these messages. Note the difference in counting of
2227 * this "budget" and the "threshold" above: the "budget" is decreased only once per generated message,
2228 * regardless how many buses/direct connections it is enqueued on, while the "threshold" is applied to
2229 * each queued instance of bus message, i.e. if the same message is enqueued to five buses/direct
2230 * connections it will be counted five times. This difference in counting ("references"
2231 * vs. "instances") is primarily a result of the fact that it's easier to implement it this way,
2232 * however it also reflects the thinking that the "threshold" should put a limit on used queue memory,
2233 * i.e. space, while the "budget" should put a limit on time. Also note that the "threshold" is
2234 * currently chosen much higher than the "budget". */
2235 budget
= MANAGER_BUS_MESSAGE_BUDGET
;
2238 while (budget
!= 0 && (u
= m
->dbus_unit_queue
)) {
2240 assert(u
->in_dbus_queue
);
2242 bus_unit_send_change_signal(u
);
2245 if (budget
!= (unsigned) -1)
2249 while (budget
!= 0 && (j
= m
->dbus_job_queue
)) {
2250 assert(j
->in_dbus_queue
);
2252 bus_job_send_change_signal(j
);
2255 if (budget
!= (unsigned) -1)
2259 if (m
->send_reloading_done
) {
2260 m
->send_reloading_done
= false;
2261 bus_manager_send_reloading(m
, false);
2265 if (m
->pending_reload_message
) {
2266 bus_send_pending_reload_message(m
);
2273 static int manager_dispatch_cgroups_agent_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
2274 Manager
*m
= userdata
;
2278 n
= recv(fd
, buf
, sizeof(buf
), 0);
2280 return log_error_errno(errno
, "Failed to read cgroups agent message: %m");
2282 log_error("Got zero-length cgroups agent message, ignoring.");
2285 if ((size_t) n
>= sizeof(buf
)) {
2286 log_error("Got overly long cgroups agent message, ignoring.");
2290 if (memchr(buf
, 0, n
)) {
2291 log_error("Got cgroups agent message with embedded NUL byte, ignoring.");
2296 manager_notify_cgroup_empty(m
, buf
);
2297 (void) bus_forward_agent_released(m
, buf
);
2302 static bool manager_process_barrier_fd(char * const *tags
, FDSet
*fds
) {
2304 /* nothing else must be sent when using BARRIER=1 */
2305 if (strv_contains(tags
, "BARRIER=1")) {
2306 if (strv_length(tags
) == 1) {
2307 if (fdset_size(fds
) != 1)
2308 log_warning("Got incorrect number of fds with BARRIER=1, closing them.");
2310 log_warning("Extra notification messages sent with BARRIER=1, ignoring everything.");
2312 /* Drop the message if BARRIER=1 was found */
2319 static void manager_invoke_notify_message(
2322 const struct ucred
*ucred
,
2331 if (u
->notifygen
== m
->notifygen
) /* Already invoked on this same unit in this same iteration? */
2333 u
->notifygen
= m
->notifygen
;
2335 if (UNIT_VTABLE(u
)->notify_message
)
2336 UNIT_VTABLE(u
)->notify_message(u
, ucred
, tags
, fds
);
2338 else if (DEBUG_LOGGING
) {
2339 _cleanup_free_
char *buf
= NULL
, *x
= NULL
, *y
= NULL
;
2341 buf
= strv_join(tags
, ", ");
2343 x
= ellipsize(buf
, 20, 90);
2347 log_unit_debug(u
, "Got notification message \"%s\", ignoring.", strnull(y
));
2351 static int manager_dispatch_notify_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
2353 _cleanup_fdset_free_ FDSet
*fds
= NULL
;
2354 Manager
*m
= userdata
;
2355 char buf
[NOTIFY_BUFFER_MAX
+1];
2356 struct iovec iovec
= {
2358 .iov_len
= sizeof(buf
)-1,
2360 CMSG_BUFFER_TYPE(CMSG_SPACE(sizeof(struct ucred
)) +
2361 CMSG_SPACE(sizeof(int) * NOTIFY_FD_MAX
)) control
;
2362 struct msghdr msghdr
= {
2365 .msg_control
= &control
,
2366 .msg_controllen
= sizeof(control
),
2369 struct cmsghdr
*cmsg
;
2370 struct ucred
*ucred
= NULL
;
2371 _cleanup_free_ Unit
**array_copy
= NULL
;
2372 _cleanup_strv_free_
char **tags
= NULL
;
2373 Unit
*u1
, *u2
, **array
;
2374 int r
, *fd_array
= NULL
;
2380 assert(m
->notify_fd
== fd
);
2382 if (revents
!= EPOLLIN
) {
2383 log_warning("Got unexpected poll event for notify fd.");
2387 n
= recvmsg_safe(m
->notify_fd
, &msghdr
, MSG_DONTWAIT
|MSG_CMSG_CLOEXEC
|MSG_TRUNC
);
2388 if (IN_SET(n
, -EAGAIN
, -EINTR
))
2389 return 0; /* Spurious wakeup, try again */
2391 /* If this is any other, real error, then let's stop processing this socket. This of course
2392 * means we won't take notification messages anymore, but that's still better than busy
2393 * looping around this: being woken up over and over again but being unable to actually read
2394 * the message off the socket. */
2395 return log_error_errno(n
, "Failed to receive notification message: %m");
2397 CMSG_FOREACH(cmsg
, &msghdr
) {
2398 if (cmsg
->cmsg_level
== SOL_SOCKET
&& cmsg
->cmsg_type
== SCM_RIGHTS
) {
2401 fd_array
= (int*) CMSG_DATA(cmsg
);
2402 n_fds
= (cmsg
->cmsg_len
- CMSG_LEN(0)) / sizeof(int);
2404 } else if (cmsg
->cmsg_level
== SOL_SOCKET
&&
2405 cmsg
->cmsg_type
== SCM_CREDENTIALS
&&
2406 cmsg
->cmsg_len
== CMSG_LEN(sizeof(struct ucred
))) {
2409 ucred
= (struct ucred
*) CMSG_DATA(cmsg
);
2416 r
= fdset_new_array(&fds
, fd_array
, n_fds
);
2418 close_many(fd_array
, n_fds
);
2424 if (!ucred
|| !pid_is_valid(ucred
->pid
)) {
2425 log_warning("Received notify message without valid credentials. Ignoring.");
2429 if ((size_t) n
>= sizeof(buf
) || (msghdr
.msg_flags
& MSG_TRUNC
)) {
2430 log_warning("Received notify message exceeded maximum size. Ignoring.");
2434 /* As extra safety check, let's make sure the string we get doesn't contain embedded NUL bytes. We permit one
2435 * trailing NUL byte in the message, but don't expect it. */
2436 if (n
> 1 && memchr(buf
, 0, n
-1)) {
2437 log_warning("Received notify message with embedded NUL bytes. Ignoring.");
2441 /* Make sure it's NUL-terminated, then parse it to obtain the tags list */
2443 tags
= strv_split_newlines(buf
);
2449 /* possibly a barrier fd, let's see */
2450 if (manager_process_barrier_fd(tags
, fds
))
2453 /* Increase the generation counter used for filtering out duplicate unit invocations. */
2456 /* Notify every unit that might be interested, which might be multiple. */
2457 u1
= manager_get_unit_by_pid_cgroup(m
, ucred
->pid
);
2458 u2
= hashmap_get(m
->watch_pids
, PID_TO_PTR(ucred
->pid
));
2459 array
= hashmap_get(m
->watch_pids
, PID_TO_PTR(-ucred
->pid
));
2466 array_copy
= newdup(Unit
*, array
, k
+1);
2470 /* And now invoke the per-unit callbacks. Note that manager_invoke_notify_message() will handle duplicate units
2471 * make sure we only invoke each unit's handler once. */
2473 manager_invoke_notify_message(m
, u1
, ucred
, tags
, fds
);
2477 manager_invoke_notify_message(m
, u2
, ucred
, tags
, fds
);
2481 for (size_t i
= 0; array_copy
[i
]; i
++) {
2482 manager_invoke_notify_message(m
, array_copy
[i
], ucred
, tags
, fds
);
2487 log_warning("Cannot find unit for notify message of PID "PID_FMT
", ignoring.", ucred
->pid
);
2489 if (fdset_size(fds
) > 0)
2490 log_warning("Got extra auxiliary fds with notification message, closing them.");
2495 static void manager_invoke_sigchld_event(
2498 const siginfo_t
*si
) {
2504 /* Already invoked the handler of this unit in this iteration? Then don't process this again */
2505 if (u
->sigchldgen
== m
->sigchldgen
)
2507 u
->sigchldgen
= m
->sigchldgen
;
2509 log_unit_debug(u
, "Child "PID_FMT
" belongs to %s.", si
->si_pid
, u
->id
);
2510 unit_unwatch_pid(u
, si
->si_pid
);
2512 if (UNIT_VTABLE(u
)->sigchld_event
)
2513 UNIT_VTABLE(u
)->sigchld_event(u
, si
->si_pid
, si
->si_code
, si
->si_status
);
2516 static int manager_dispatch_sigchld(sd_event_source
*source
, void *userdata
) {
2517 Manager
*m
= userdata
;
2524 /* First we call waitid() for a PID and do not reap the zombie. That way we can still access /proc/$PID for it
2525 * while it is a zombie. */
2527 if (waitid(P_ALL
, 0, &si
, WEXITED
|WNOHANG
|WNOWAIT
) < 0) {
2529 if (errno
!= ECHILD
)
2530 log_error_errno(errno
, "Failed to peek for child with waitid(), ignoring: %m");
2538 if (IN_SET(si
.si_code
, CLD_EXITED
, CLD_KILLED
, CLD_DUMPED
)) {
2539 _cleanup_free_ Unit
**array_copy
= NULL
;
2540 _cleanup_free_
char *name
= NULL
;
2541 Unit
*u1
, *u2
, **array
;
2543 (void) get_process_comm(si
.si_pid
, &name
);
2545 log_debug("Child "PID_FMT
" (%s) died (code=%s, status=%i/%s)",
2546 si
.si_pid
, strna(name
),
2547 sigchld_code_to_string(si
.si_code
),
2549 strna(si
.si_code
== CLD_EXITED
2550 ? exit_status_to_string(si
.si_status
, EXIT_STATUS_FULL
)
2551 : signal_to_string(si
.si_status
)));
2553 /* Increase the generation counter used for filtering out duplicate unit invocations */
2556 /* And now figure out the unit this belongs to, it might be multiple... */
2557 u1
= manager_get_unit_by_pid_cgroup(m
, si
.si_pid
);
2558 u2
= hashmap_get(m
->watch_pids
, PID_TO_PTR(si
.si_pid
));
2559 array
= hashmap_get(m
->watch_pids
, PID_TO_PTR(-si
.si_pid
));
2563 /* Count how many entries the array has */
2567 /* Make a copy of the array so that we don't trip up on the array changing beneath us */
2568 array_copy
= newdup(Unit
*, array
, n
+1);
2573 /* Finally, execute them all. Note that u1, u2 and the array might contain duplicates, but
2574 * that's fine, manager_invoke_sigchld_event() will ensure we only invoke the handlers once for
2575 * each iteration. */
2577 /* We check for oom condition, in case we got SIGCHLD before the oom notification.
2578 * We only do this for the cgroup the PID belonged to. */
2579 (void) unit_check_oom(u1
);
2581 /* This only logs for now. In the future when the interface for kills/notifications
2582 * is more stable we can extend service results table similar to how kernel oom kills
2584 (void) unit_check_oomd_kill(u1
);
2586 manager_invoke_sigchld_event(m
, u1
, &si
);
2589 manager_invoke_sigchld_event(m
, u2
, &si
);
2591 for (size_t i
= 0; array_copy
[i
]; i
++)
2592 manager_invoke_sigchld_event(m
, array_copy
[i
], &si
);
2595 /* And now, we actually reap the zombie. */
2596 if (waitid(P_PID
, si
.si_pid
, &si
, WEXITED
) < 0) {
2597 log_error_errno(errno
, "Failed to dequeue child, ignoring: %m");
2604 /* All children processed for now, turn off event source */
2606 r
= sd_event_source_set_enabled(m
->sigchld_event_source
, SD_EVENT_OFF
);
2608 return log_error_errno(r
, "Failed to disable SIGCHLD event source: %m");
2613 static void manager_start_target(Manager
*m
, const char *name
, JobMode mode
) {
2614 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
2617 log_debug("Activating special unit %s", name
);
2619 r
= manager_add_job_by_name(m
, JOB_START
, name
, mode
, NULL
, &error
, NULL
);
2621 log_error("Failed to enqueue %s job: %s", name
, bus_error_message(&error
, r
));
2624 static void manager_handle_ctrl_alt_del(Manager
*m
) {
2625 /* If the user presses C-A-D more than
2626 * 7 times within 2s, we reboot/shutdown immediately,
2627 * unless it was disabled in system.conf */
2629 if (ratelimit_below(&m
->ctrl_alt_del_ratelimit
) || m
->cad_burst_action
== EMERGENCY_ACTION_NONE
)
2630 manager_start_target(m
, SPECIAL_CTRL_ALT_DEL_TARGET
, JOB_REPLACE_IRREVERSIBLY
);
2632 emergency_action(m
, m
->cad_burst_action
, EMERGENCY_ACTION_WARN
, NULL
, -1,
2633 "Ctrl-Alt-Del was pressed more than 7 times within 2s");
2636 static int manager_dispatch_signal_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
2637 Manager
*m
= userdata
;
2639 struct signalfd_siginfo sfsi
;
2643 assert(m
->signal_fd
== fd
);
2645 if (revents
!= EPOLLIN
) {
2646 log_warning("Got unexpected events from signal file descriptor.");
2650 n
= read(m
->signal_fd
, &sfsi
, sizeof(sfsi
));
2651 if (n
!= sizeof(sfsi
)) {
2653 log_warning("Truncated read from signal fd (%zu bytes), ignoring!", n
);
2657 if (IN_SET(errno
, EINTR
, EAGAIN
))
2660 /* We return an error here, which will kill this handler,
2661 * to avoid a busy loop on read error. */
2662 return log_error_errno(errno
, "Reading from signal fd failed: %m");
2665 log_received_signal(sfsi
.ssi_signo
== SIGCHLD
||
2666 (sfsi
.ssi_signo
== SIGTERM
&& MANAGER_IS_USER(m
))
2667 ? LOG_DEBUG
: LOG_INFO
,
2670 switch (sfsi
.ssi_signo
) {
2673 r
= sd_event_source_set_enabled(m
->sigchld_event_source
, SD_EVENT_ON
);
2675 log_warning_errno(r
, "Failed to enable SIGCHLD event source, ignoring: %m");
2680 if (MANAGER_IS_SYSTEM(m
)) {
2681 /* This is for compatibility with the original sysvinit */
2682 if (verify_run_space_and_log("Refusing to reexecute") < 0)
2685 m
->objective
= MANAGER_REEXECUTE
;
2691 if (MANAGER_IS_SYSTEM(m
))
2692 manager_handle_ctrl_alt_del(m
);
2694 manager_start_target(m
, SPECIAL_EXIT_TARGET
,
2695 JOB_REPLACE_IRREVERSIBLY
);
2699 /* This is a nop on non-init */
2700 if (MANAGER_IS_SYSTEM(m
))
2701 manager_start_target(m
, SPECIAL_KBREQUEST_TARGET
, JOB_REPLACE
);
2706 /* This is a nop on non-init */
2707 if (MANAGER_IS_SYSTEM(m
))
2708 manager_start_target(m
, SPECIAL_SIGPWR_TARGET
, JOB_REPLACE
);
2713 if (manager_dbus_is_running(m
, false)) {
2714 log_info("Trying to reconnect to bus...");
2716 (void) bus_init_api(m
);
2718 if (MANAGER_IS_SYSTEM(m
))
2719 (void) bus_init_system(m
);
2721 log_info("Starting D-Bus service...");
2722 manager_start_target(m
, SPECIAL_DBUS_SERVICE
, JOB_REPLACE
);
2728 _cleanup_free_
char *dump
= NULL
;
2730 r
= manager_get_dump_string(m
, &dump
);
2732 log_warning_errno(errno
, "Failed to acquire manager dump: %m");
2736 log_dump(LOG_INFO
, dump
);
2741 if (verify_run_space_and_log("Refusing to reload") < 0)
2744 m
->objective
= MANAGER_RELOAD
;
2749 /* Starting SIGRTMIN+0 */
2750 static const struct {
2753 } target_table
[] = {
2754 [0] = { SPECIAL_DEFAULT_TARGET
, JOB_ISOLATE
},
2755 [1] = { SPECIAL_RESCUE_TARGET
, JOB_ISOLATE
},
2756 [2] = { SPECIAL_EMERGENCY_TARGET
, JOB_ISOLATE
},
2757 [3] = { SPECIAL_HALT_TARGET
, JOB_REPLACE_IRREVERSIBLY
},
2758 [4] = { SPECIAL_POWEROFF_TARGET
, JOB_REPLACE_IRREVERSIBLY
},
2759 [5] = { SPECIAL_REBOOT_TARGET
, JOB_REPLACE_IRREVERSIBLY
},
2760 [6] = { SPECIAL_KEXEC_TARGET
, JOB_REPLACE_IRREVERSIBLY
},
2763 /* Starting SIGRTMIN+13, so that target halt and system halt are 10 apart */
2764 static const ManagerObjective objective_table
[] = {
2766 [1] = MANAGER_POWEROFF
,
2767 [2] = MANAGER_REBOOT
,
2768 [3] = MANAGER_KEXEC
,
2771 if ((int) sfsi
.ssi_signo
>= SIGRTMIN
+0 &&
2772 (int) sfsi
.ssi_signo
< SIGRTMIN
+(int) ELEMENTSOF(target_table
)) {
2773 int idx
= (int) sfsi
.ssi_signo
- SIGRTMIN
;
2774 manager_start_target(m
, target_table
[idx
].target
,
2775 target_table
[idx
].mode
);
2779 if ((int) sfsi
.ssi_signo
>= SIGRTMIN
+13 &&
2780 (int) sfsi
.ssi_signo
< SIGRTMIN
+13+(int) ELEMENTSOF(objective_table
)) {
2781 m
->objective
= objective_table
[sfsi
.ssi_signo
- SIGRTMIN
- 13];
2785 switch (sfsi
.ssi_signo
- SIGRTMIN
) {
2788 manager_override_show_status(m
, SHOW_STATUS_YES
, "signal");
2792 manager_override_show_status(m
, SHOW_STATUS_NO
, "signal");
2796 manager_override_log_level(m
, LOG_DEBUG
);
2800 manager_restore_original_log_level(m
);
2804 if (MANAGER_IS_USER(m
)) {
2805 m
->objective
= MANAGER_EXIT
;
2809 /* This is a nop on init */
2813 case 29: /* compatibility: used to be mapped to LOG_TARGET_SYSLOG_OR_KMSG */
2814 manager_restore_original_log_target(m
);
2818 manager_override_log_target(m
, LOG_TARGET_CONSOLE
);
2822 manager_override_log_target(m
, LOG_TARGET_KMSG
);
2826 log_warning("Got unhandled signal <%s>.", signal_to_string(sfsi
.ssi_signo
));
2833 static int manager_dispatch_time_change_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
2834 Manager
*m
= userdata
;
2838 assert(m
->time_change_fd
== fd
);
2840 log_struct(LOG_DEBUG
,
2841 "MESSAGE_ID=" SD_MESSAGE_TIME_CHANGE_STR
,
2842 LOG_MESSAGE("Time has been changed"));
2844 /* Restart the watch */
2845 (void) manager_setup_time_change(m
);
2847 HASHMAP_FOREACH(u
, m
->units
)
2848 if (UNIT_VTABLE(u
)->time_change
)
2849 UNIT_VTABLE(u
)->time_change(u
);
2854 static int manager_dispatch_timezone_change(
2855 sd_event_source
*source
,
2856 const struct inotify_event
*e
,
2859 Manager
*m
= userdata
;
2865 log_debug("inotify event for /etc/localtime");
2867 changed
= manager_read_timezone_stat(m
);
2871 /* Something changed, restart the watch, to ensure we watch the new /etc/localtime if it changed */
2872 (void) manager_setup_timezone_change(m
);
2874 /* Read the new timezone */
2877 log_debug("Timezone has been changed (now: %s).", tzname
[daylight
]);
2879 HASHMAP_FOREACH(u
, m
->units
)
2880 if (UNIT_VTABLE(u
)->timezone_change
)
2881 UNIT_VTABLE(u
)->timezone_change(u
);
2886 static int manager_dispatch_idle_pipe_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
2887 Manager
*m
= userdata
;
2890 assert(m
->idle_pipe
[2] == fd
);
2892 /* There's at least one Type=idle child that just gave up on us waiting for the boot process to complete. Let's
2893 * now turn off any further console output if there's at least one service that needs console access, so that
2894 * from now on our own output should not spill into that service's output anymore. After all, we support
2895 * Type=idle only to beautify console output and it generally is set on services that want to own the console
2896 * exclusively without our interference. */
2897 m
->no_console_output
= m
->n_on_console
> 0;
2899 /* Acknowledge the child's request, and let all all other children know too that they shouldn't wait any longer
2900 * by closing the pipes towards them, which is what they are waiting for. */
2901 manager_close_idle_pipe(m
);
2906 static int manager_dispatch_jobs_in_progress(sd_event_source
*source
, usec_t usec
, void *userdata
) {
2907 Manager
*m
= userdata
;
2913 manager_print_jobs_in_progress(m
);
2915 r
= sd_event_source_set_time_relative(source
, JOBS_IN_PROGRESS_PERIOD_USEC
);
2919 return sd_event_source_set_enabled(source
, SD_EVENT_ONESHOT
);
2922 int manager_loop(Manager
*m
) {
2923 RateLimit rl
= { .interval
= 1*USEC_PER_SEC
, .burst
= 50000 };
2927 assert(m
->objective
== MANAGER_OK
); /* Ensure manager_startup() has been called */
2929 manager_check_finished(m
);
2931 /* There might still be some zombies hanging around from before we were exec()'ed. Let's reap them. */
2932 r
= sd_event_source_set_enabled(m
->sigchld_event_source
, SD_EVENT_ON
);
2934 return log_error_errno(r
, "Failed to enable SIGCHLD event source: %m");
2936 while (m
->objective
== MANAGER_OK
) {
2937 usec_t wait_usec
, watchdog_usec
;
2939 watchdog_usec
= manager_get_watchdog(m
, WATCHDOG_RUNTIME
);
2940 if (timestamp_is_set(watchdog_usec
))
2941 (void) watchdog_ping();
2943 if (!ratelimit_below(&rl
)) {
2944 /* Yay, something is going seriously wrong, pause a little */
2945 log_warning("Looping too fast. Throttling execution a little.");
2949 if (manager_dispatch_load_queue(m
) > 0)
2952 if (manager_dispatch_gc_job_queue(m
) > 0)
2955 if (manager_dispatch_gc_unit_queue(m
) > 0)
2958 if (manager_dispatch_cleanup_queue(m
) > 0)
2961 if (manager_dispatch_cgroup_realize_queue(m
) > 0)
2964 if (manager_dispatch_stop_when_unneeded_queue(m
) > 0)
2967 if (manager_dispatch_dbus_queue(m
) > 0)
2970 /* Sleep for watchdog runtime wait time */
2971 if (timestamp_is_set(watchdog_usec
))
2972 wait_usec
= watchdog_runtime_wait();
2974 wait_usec
= USEC_INFINITY
;
2976 r
= sd_event_run(m
->event
, wait_usec
);
2978 return log_error_errno(r
, "Failed to run event loop: %m");
2981 return m
->objective
;
2984 int manager_load_unit_from_dbus_path(Manager
*m
, const char *s
, sd_bus_error
*e
, Unit
**_u
) {
2985 _cleanup_free_
char *n
= NULL
;
2986 sd_id128_t invocation_id
;
2994 r
= unit_name_from_dbus_path(s
, &n
);
2998 /* Permit addressing units by invocation ID: if the passed bus path is suffixed by a 128bit ID then we use it
2999 * as invocation ID. */
3000 r
= sd_id128_from_string(n
, &invocation_id
);
3002 u
= hashmap_get(m
->units_by_invocation_id
, &invocation_id
);
3008 return sd_bus_error_setf(e
, BUS_ERROR_NO_UNIT_FOR_INVOCATION_ID
,
3009 "No unit with the specified invocation ID " SD_ID128_FORMAT_STR
" known.",
3010 SD_ID128_FORMAT_VAL(invocation_id
));
3013 /* If this didn't work, we check if this is a unit name */
3014 if (!unit_name_is_valid(n
, UNIT_NAME_PLAIN
|UNIT_NAME_INSTANCE
)) {
3015 _cleanup_free_
char *nn
= NULL
;
3018 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
,
3019 "Unit name %s is neither a valid invocation ID nor unit name.", strnull(nn
));
3022 r
= manager_load_unit(m
, n
, NULL
, e
, &u
);
3030 int manager_get_job_from_dbus_path(Manager
*m
, const char *s
, Job
**_j
) {
3040 p
= startswith(s
, "/org/freedesktop/systemd1/job/");
3044 r
= safe_atou(p
, &id
);
3048 j
= manager_get_job(m
, id
);
3057 void manager_send_unit_audit(Manager
*m
, Unit
*u
, int type
, bool success
) {
3060 _cleanup_free_
char *p
= NULL
;
3064 if (!MANAGER_IS_SYSTEM(m
))
3067 audit_fd
= get_audit_fd();
3071 /* Don't generate audit events if the service was already
3072 * started and we're just deserializing */
3073 if (MANAGER_IS_RELOADING(m
))
3076 if (u
->type
!= UNIT_SERVICE
)
3079 r
= unit_name_to_prefix_and_instance(u
->id
, &p
);
3081 log_error_errno(r
, "Failed to extract prefix and instance of unit name: %m");
3085 msg
= strjoina("unit=", p
);
3086 if (audit_log_user_comm_message(audit_fd
, type
, msg
, "systemd", NULL
, NULL
, NULL
, success
) < 0) {
3088 /* We aren't allowed to send audit messages?
3089 * Then let's not retry again. */
3092 log_warning_errno(errno
, "Failed to send audit message: %m");
3098 void manager_send_unit_plymouth(Manager
*m
, Unit
*u
) {
3099 static const union sockaddr_union sa
= PLYMOUTH_SOCKET
;
3100 _cleanup_free_
char *message
= NULL
;
3101 _cleanup_close_
int fd
= -1;
3104 /* Don't generate plymouth events if the service was already
3105 * started and we're just deserializing */
3106 if (MANAGER_IS_RELOADING(m
))
3109 if (!MANAGER_IS_SYSTEM(m
))
3112 if (detect_container() > 0)
3115 if (!IN_SET(u
->type
, UNIT_SERVICE
, UNIT_MOUNT
, UNIT_SWAP
))
3118 /* We set SOCK_NONBLOCK here so that we rather drop the
3119 * message then wait for plymouth */
3120 fd
= socket(AF_UNIX
, SOCK_STREAM
|SOCK_CLOEXEC
|SOCK_NONBLOCK
, 0);
3122 log_error_errno(errno
, "socket() failed: %m");
3126 if (connect(fd
, &sa
.sa
, SOCKADDR_UN_LEN(sa
.un
)) < 0) {
3127 if (!IN_SET(errno
, EAGAIN
, ENOENT
) && !ERRNO_IS_DISCONNECT(errno
))
3128 log_error_errno(errno
, "connect() failed: %m");
3132 if (asprintf(&message
, "U\002%c%s%n", (int) (strlen(u
->id
) + 1), u
->id
, &n
) < 0) {
3138 if (write(fd
, message
, n
+ 1) != n
+ 1)
3139 if (!IN_SET(errno
, EAGAIN
, ENOENT
) && !ERRNO_IS_DISCONNECT(errno
))
3140 log_error_errno(errno
, "Failed to write Plymouth message: %m");
3143 int manager_open_serialization(Manager
*m
, FILE **_f
) {
3144 _cleanup_close_
int fd
= -1;
3149 fd
= open_serialization_fd("systemd-state");
3153 f
= take_fdopen(&fd
, "w+");
3161 static bool manager_timestamp_shall_serialize(ManagerTimestamp t
) {
3166 /* The following timestamps only apply to the host system, hence only serialize them there */
3168 MANAGER_TIMESTAMP_USERSPACE
, MANAGER_TIMESTAMP_FINISH
,
3169 MANAGER_TIMESTAMP_SECURITY_START
, MANAGER_TIMESTAMP_SECURITY_FINISH
,
3170 MANAGER_TIMESTAMP_GENERATORS_START
, MANAGER_TIMESTAMP_GENERATORS_FINISH
,
3171 MANAGER_TIMESTAMP_UNITS_LOAD_START
, MANAGER_TIMESTAMP_UNITS_LOAD_FINISH
);
3174 #define DESTROY_IPC_FLAG (UINT32_C(1) << 31)
3176 static void manager_serialize_uid_refs_internal(
3180 const char *field_name
) {
3189 /* Serialize the UID reference table. Or actually, just the IPC destruction flag of it, as
3190 * the actual counter of it is better rebuild after a reload/reexec. */
3192 HASHMAP_FOREACH_KEY(p
, k
, *uid_refs
) {
3196 uid
= PTR_TO_UID(k
);
3197 c
= PTR_TO_UINT32(p
);
3199 if (!(c
& DESTROY_IPC_FLAG
))
3202 (void) serialize_item_format(f
, field_name
, UID_FMT
, uid
);
3206 static void manager_serialize_uid_refs(Manager
*m
, FILE *f
) {
3207 manager_serialize_uid_refs_internal(m
, f
, &m
->uid_refs
, "destroy-ipc-uid");
3210 static void manager_serialize_gid_refs(Manager
*m
, FILE *f
) {
3211 manager_serialize_uid_refs_internal(m
, f
, &m
->gid_refs
, "destroy-ipc-gid");
3214 int manager_serialize(
3218 bool switching_root
) {
3228 _cleanup_(manager_reloading_stopp
) _unused_ Manager
*reloading
= manager_reloading_start(m
);
3230 (void) serialize_item_format(f
, "current-job-id", "%" PRIu32
, m
->current_job_id
);
3231 (void) serialize_item_format(f
, "n-installed-jobs", "%u", m
->n_installed_jobs
);
3232 (void) serialize_item_format(f
, "n-failed-jobs", "%u", m
->n_failed_jobs
);
3233 (void) serialize_bool(f
, "taint-usr", m
->taint_usr
);
3234 (void) serialize_bool(f
, "ready-sent", m
->ready_sent
);
3235 (void) serialize_bool(f
, "taint-logged", m
->taint_logged
);
3236 (void) serialize_bool(f
, "service-watchdogs", m
->service_watchdogs
);
3238 /* After switching root, udevd has not been started yet. So, enumeration results should not be emitted. */
3239 (void) serialize_bool(f
, "honor-device-enumeration", !switching_root
);
3241 if (m
->show_status_overridden
!= _SHOW_STATUS_INVALID
)
3242 (void) serialize_item(f
, "show-status-overridden",
3243 show_status_to_string(m
->show_status_overridden
));
3245 if (m
->log_level_overridden
)
3246 (void) serialize_item_format(f
, "log-level-override", "%i", log_get_max_level());
3247 if (m
->log_target_overridden
)
3248 (void) serialize_item(f
, "log-target-override", log_target_to_string(log_get_target()));
3250 (void) serialize_usec(f
, "runtime-watchdog-overridden", m
->watchdog_overridden
[WATCHDOG_RUNTIME
]);
3251 (void) serialize_usec(f
, "reboot-watchdog-overridden", m
->watchdog_overridden
[WATCHDOG_REBOOT
]);
3252 (void) serialize_usec(f
, "kexec-watchdog-overridden", m
->watchdog_overridden
[WATCHDOG_KEXEC
]);
3254 for (ManagerTimestamp q
= 0; q
< _MANAGER_TIMESTAMP_MAX
; q
++) {
3255 _cleanup_free_
char *joined
= NULL
;
3257 if (!manager_timestamp_shall_serialize(q
))
3260 joined
= strjoin(manager_timestamp_to_string(q
), "-timestamp");
3264 (void) serialize_dual_timestamp(f
, joined
, m
->timestamps
+ q
);
3267 if (!switching_root
)
3268 (void) serialize_strv(f
, "env", m
->client_environment
);
3270 if (m
->notify_fd
>= 0) {
3271 r
= serialize_fd(f
, fds
, "notify-fd", m
->notify_fd
);
3275 (void) serialize_item(f
, "notify-socket", m
->notify_socket
);
3278 if (m
->cgroups_agent_fd
>= 0) {
3279 r
= serialize_fd(f
, fds
, "cgroups-agent-fd", m
->cgroups_agent_fd
);
3284 if (m
->user_lookup_fds
[0] >= 0) {
3287 copy0
= fdset_put_dup(fds
, m
->user_lookup_fds
[0]);
3289 return log_error_errno(copy0
, "Failed to add user lookup fd to serialization: %m");
3291 copy1
= fdset_put_dup(fds
, m
->user_lookup_fds
[1]);
3293 return log_error_errno(copy1
, "Failed to add user lookup fd to serialization: %m");
3295 (void) serialize_item_format(f
, "user-lookup", "%i %i", copy0
, copy1
);
3298 bus_track_serialize(m
->subscribed
, f
, "subscribed");
3300 r
= dynamic_user_serialize(m
, f
, fds
);
3304 manager_serialize_uid_refs(m
, f
);
3305 manager_serialize_gid_refs(m
, f
);
3307 r
= exec_runtime_serialize(m
, f
, fds
);
3311 (void) fputc('\n', f
);
3313 HASHMAP_FOREACH_KEY(u
, t
, m
->units
) {
3321 r
= unit_serialize(u
, f
, fds
, !switching_root
);
3326 r
= fflush_and_check(f
);
3328 return log_error_errno(r
, "Failed to flush serialization: %m");
3330 r
= bus_fdset_add_all(m
, fds
);
3332 return log_error_errno(r
, "Failed to add bus sockets to serialization: %m");
3337 static int manager_deserialize_one_unit(Manager
*m
, const char *name
, FILE *f
, FDSet
*fds
) {
3341 r
= manager_load_unit(m
, name
, NULL
, NULL
, &u
);
3345 return log_notice_errno(r
, "Failed to load unit \"%s\", skipping deserialization: %m", name
);
3348 r
= unit_deserialize(u
, f
, fds
);
3352 return log_notice_errno(r
, "Failed to deserialize unit \"%s\", skipping: %m", name
);
3358 static int manager_deserialize_units(Manager
*m
, FILE *f
, FDSet
*fds
) {
3359 const char *unit_name
;
3363 _cleanup_free_
char *line
= NULL
;
3365 r
= read_line(f
, LONG_LINE_MAX
, &line
);
3367 return log_error_errno(r
, "Failed to read serialization line: %m");
3371 unit_name
= strstrip(line
);
3373 r
= manager_deserialize_one_unit(m
, unit_name
, f
, fds
);
3377 r
= unit_deserialize_skip(f
);
3386 usec_t
manager_get_watchdog(Manager
*m
, WatchdogType t
) {
3389 if (MANAGER_IS_USER(m
))
3390 return USEC_INFINITY
;
3392 if (timestamp_is_set(m
->watchdog_overridden
[t
]))
3393 return m
->watchdog_overridden
[t
];
3395 return m
->watchdog
[t
];
3398 void manager_set_watchdog(Manager
*m
, WatchdogType t
, usec_t timeout
) {
3403 if (MANAGER_IS_USER(m
))
3406 if (m
->watchdog
[t
] == timeout
)
3409 if (t
== WATCHDOG_RUNTIME
)
3410 if (!timestamp_is_set(m
->watchdog_overridden
[WATCHDOG_RUNTIME
])) {
3411 if (timestamp_is_set(timeout
))
3412 r
= watchdog_set_timeout(&timeout
);
3414 watchdog_close(true);
3418 m
->watchdog
[t
] = timeout
;
3421 int manager_override_watchdog(Manager
*m
, WatchdogType t
, usec_t timeout
) {
3426 if (MANAGER_IS_USER(m
))
3429 if (m
->watchdog_overridden
[t
] == timeout
)
3432 if (t
== WATCHDOG_RUNTIME
) {
3435 p
= timestamp_is_set(timeout
) ? &timeout
: &m
->watchdog
[t
];
3436 if (timestamp_is_set(*p
))
3437 r
= watchdog_set_timeout(p
);
3439 watchdog_close(true);
3443 m
->watchdog_overridden
[t
] = timeout
;
3448 static void manager_deserialize_uid_refs_one_internal(
3451 const char *value
) {
3461 r
= parse_uid(value
, &uid
);
3462 if (r
< 0 || uid
== 0) {
3463 log_debug("Unable to parse UID reference serialization: " UID_FMT
, uid
);
3467 r
= hashmap_ensure_allocated(uid_refs
, &trivial_hash_ops
);
3473 c
= PTR_TO_UINT32(hashmap_get(*uid_refs
, UID_TO_PTR(uid
)));
3474 if (c
& DESTROY_IPC_FLAG
)
3477 c
|= DESTROY_IPC_FLAG
;
3479 r
= hashmap_replace(*uid_refs
, UID_TO_PTR(uid
), UINT32_TO_PTR(c
));
3481 log_debug_errno(r
, "Failed to add UID reference entry: %m");
3486 static void manager_deserialize_uid_refs_one(Manager
*m
, const char *value
) {
3487 manager_deserialize_uid_refs_one_internal(m
, &m
->uid_refs
, value
);
3490 static void manager_deserialize_gid_refs_one(Manager
*m
, const char *value
) {
3491 manager_deserialize_uid_refs_one_internal(m
, &m
->gid_refs
, value
);
3494 int manager_deserialize(Manager
*m
, FILE *f
, FDSet
*fds
) {
3500 if (DEBUG_LOGGING
) {
3501 if (fdset_isempty(fds
))
3502 log_debug("No file descriptors passed");
3506 FDSET_FOREACH(fd
, fds
) {
3507 _cleanup_free_
char *fn
= NULL
;
3509 r
= fd_get_path(fd
, &fn
);
3511 log_debug_errno(r
, "Received serialized fd %i → %m", fd
);
3513 log_debug("Received serialized fd %i → %s", fd
, strna(fn
));
3518 log_debug("Deserializing state...");
3520 /* If we are not in reload mode yet, enter it now. Not that this is recursive, a caller might already have
3521 * increased it to non-zero, which is why we just increase it by one here and down again at the end of this
3523 _cleanup_(manager_reloading_stopp
) _unused_ Manager
*reloading
= manager_reloading_start(m
);
3526 _cleanup_free_
char *line
= NULL
;
3527 const char *val
, *l
;
3529 r
= read_line(f
, LONG_LINE_MAX
, &line
);
3531 return log_error_errno(r
, "Failed to read serialization line: %m");
3536 if (isempty(l
)) /* end marker */
3539 if ((val
= startswith(l
, "current-job-id="))) {
3542 if (safe_atou32(val
, &id
) < 0)
3543 log_notice("Failed to parse current job id value '%s', ignoring.", val
);
3545 m
->current_job_id
= MAX(m
->current_job_id
, id
);
3547 } else if ((val
= startswith(l
, "n-installed-jobs="))) {
3550 if (safe_atou32(val
, &n
) < 0)
3551 log_notice("Failed to parse installed jobs counter '%s', ignoring.", val
);
3553 m
->n_installed_jobs
+= n
;
3555 } else if ((val
= startswith(l
, "n-failed-jobs="))) {
3558 if (safe_atou32(val
, &n
) < 0)
3559 log_notice("Failed to parse failed jobs counter '%s', ignoring.", val
);
3561 m
->n_failed_jobs
+= n
;
3563 } else if ((val
= startswith(l
, "taint-usr="))) {
3566 b
= parse_boolean(val
);
3568 log_notice("Failed to parse taint /usr flag '%s', ignoring.", val
);
3570 m
->taint_usr
= m
->taint_usr
|| b
;
3572 } else if ((val
= startswith(l
, "ready-sent="))) {
3575 b
= parse_boolean(val
);
3577 log_notice("Failed to parse ready-sent flag '%s', ignoring.", val
);
3579 m
->ready_sent
= m
->ready_sent
|| b
;
3581 } else if ((val
= startswith(l
, "taint-logged="))) {
3584 b
= parse_boolean(val
);
3586 log_notice("Failed to parse taint-logged flag '%s', ignoring.", val
);
3588 m
->taint_logged
= m
->taint_logged
|| b
;
3590 } else if ((val
= startswith(l
, "service-watchdogs="))) {
3593 b
= parse_boolean(val
);
3595 log_notice("Failed to parse service-watchdogs flag '%s', ignoring.", val
);
3597 m
->service_watchdogs
= b
;
3599 } else if ((val
= startswith(l
, "honor-device-enumeration="))) {
3602 b
= parse_boolean(val
);
3604 log_notice("Failed to parse honor-device-enumeration flag '%s', ignoring.", val
);
3606 m
->honor_device_enumeration
= b
;
3608 } else if ((val
= startswith(l
, "show-status-overridden="))) {
3611 s
= show_status_from_string(val
);
3613 log_notice("Failed to parse show-status-overridden flag '%s', ignoring.", val
);
3615 manager_override_show_status(m
, s
, "deserialize");
3617 } else if ((val
= startswith(l
, "log-level-override="))) {
3620 level
= log_level_from_string(val
);
3622 log_notice("Failed to parse log-level-override value '%s', ignoring.", val
);
3624 manager_override_log_level(m
, level
);
3626 } else if ((val
= startswith(l
, "log-target-override="))) {
3629 target
= log_target_from_string(val
);
3631 log_notice("Failed to parse log-target-override value '%s', ignoring.", val
);
3633 manager_override_log_target(m
, target
);
3635 } else if ((val
= startswith(l
, "runtime-watchdog-overridden="))) {
3638 if (deserialize_usec(val
, &t
) < 0)
3639 log_notice("Failed to parse runtime-watchdog-overridden value '%s', ignoring.", val
);
3641 manager_override_watchdog(m
, WATCHDOG_RUNTIME
, t
);
3643 } else if ((val
= startswith(l
, "reboot-watchdog-overridden="))) {
3646 if (deserialize_usec(val
, &t
) < 0)
3647 log_notice("Failed to parse reboot-watchdog-overridden value '%s', ignoring.", val
);
3649 manager_override_watchdog(m
, WATCHDOG_REBOOT
, t
);
3651 } else if ((val
= startswith(l
, "kexec-watchdog-overridden="))) {
3654 if (deserialize_usec(val
, &t
) < 0)
3655 log_notice("Failed to parse kexec-watchdog-overridden value '%s', ignoring.", val
);
3657 manager_override_watchdog(m
, WATCHDOG_KEXEC
, t
);
3659 } else if (startswith(l
, "env=")) {
3660 r
= deserialize_environment(l
+ 4, &m
->client_environment
);
3662 log_notice_errno(r
, "Failed to parse environment entry: \"%s\", ignoring: %m", l
);
3664 } else if ((val
= startswith(l
, "notify-fd="))) {
3667 if (safe_atoi(val
, &fd
) < 0 || fd
< 0 || !fdset_contains(fds
, fd
))
3668 log_notice("Failed to parse notify fd, ignoring: \"%s\"", val
);
3670 m
->notify_event_source
= sd_event_source_unref(m
->notify_event_source
);
3671 safe_close(m
->notify_fd
);
3672 m
->notify_fd
= fdset_remove(fds
, fd
);
3675 } else if ((val
= startswith(l
, "notify-socket="))) {
3676 r
= free_and_strdup(&m
->notify_socket
, val
);
3680 } else if ((val
= startswith(l
, "cgroups-agent-fd="))) {
3683 if (safe_atoi(val
, &fd
) < 0 || fd
< 0 || !fdset_contains(fds
, fd
))
3684 log_notice("Failed to parse cgroups agent fd, ignoring.: %s", val
);
3686 m
->cgroups_agent_event_source
= sd_event_source_unref(m
->cgroups_agent_event_source
);
3687 safe_close(m
->cgroups_agent_fd
);
3688 m
->cgroups_agent_fd
= fdset_remove(fds
, fd
);
3691 } else if ((val
= startswith(l
, "user-lookup="))) {
3694 if (sscanf(val
, "%i %i", &fd0
, &fd1
) != 2 || fd0
< 0 || fd1
< 0 || fd0
== fd1
|| !fdset_contains(fds
, fd0
) || !fdset_contains(fds
, fd1
))
3695 log_notice("Failed to parse user lookup fd, ignoring: %s", val
);
3697 m
->user_lookup_event_source
= sd_event_source_unref(m
->user_lookup_event_source
);
3698 safe_close_pair(m
->user_lookup_fds
);
3699 m
->user_lookup_fds
[0] = fdset_remove(fds
, fd0
);
3700 m
->user_lookup_fds
[1] = fdset_remove(fds
, fd1
);
3703 } else if ((val
= startswith(l
, "dynamic-user=")))
3704 dynamic_user_deserialize_one(m
, val
, fds
);
3705 else if ((val
= startswith(l
, "destroy-ipc-uid=")))
3706 manager_deserialize_uid_refs_one(m
, val
);
3707 else if ((val
= startswith(l
, "destroy-ipc-gid=")))
3708 manager_deserialize_gid_refs_one(m
, val
);
3709 else if ((val
= startswith(l
, "exec-runtime=")))
3710 (void) exec_runtime_deserialize_one(m
, val
, fds
);
3711 else if ((val
= startswith(l
, "subscribed="))) {
3713 if (strv_extend(&m
->deserialized_subscribed
, val
) < 0)
3719 for (q
= 0; q
< _MANAGER_TIMESTAMP_MAX
; q
++) {
3720 val
= startswith(l
, manager_timestamp_to_string(q
));
3724 val
= startswith(val
, "-timestamp=");
3729 if (q
< _MANAGER_TIMESTAMP_MAX
) /* found it */
3730 (void) deserialize_dual_timestamp(val
, m
->timestamps
+ q
);
3731 else if (!startswith(l
, "kdbus-fd=")) /* ignore kdbus */
3732 log_notice("Unknown serialization item '%s', ignoring.", l
);
3736 return manager_deserialize_units(m
, f
, fds
);
3739 int manager_reload(Manager
*m
) {
3740 _cleanup_(manager_reloading_stopp
) Manager
*reloading
= NULL
;
3741 _cleanup_fdset_free_ FDSet
*fds
= NULL
;
3742 _cleanup_fclose_
FILE *f
= NULL
;
3747 r
= manager_open_serialization(m
, &f
);
3749 return log_error_errno(r
, "Failed to create serialization file: %m");
3755 /* We are officially in reload mode from here on. */
3756 reloading
= manager_reloading_start(m
);
3758 r
= manager_serialize(m
, f
, fds
, false);
3762 if (fseeko(f
, 0, SEEK_SET
) < 0)
3763 return log_error_errno(errno
, "Failed to seek to beginning of serialization: %m");
3765 /* 💀 This is the point of no return, from here on there is no way back. 💀 */
3768 bus_manager_send_reloading(m
, true);
3770 /* Start by flushing out all jobs and units, all generated units, all runtime environments, all dynamic users
3771 * and everything else that is worth flushing out. We'll get it all back from the serialization — if we need
3774 manager_clear_jobs_and_units(m
);
3775 lookup_paths_flush_generator(&m
->lookup_paths
);
3776 lookup_paths_free(&m
->lookup_paths
);
3777 exec_runtime_vacuum(m
);
3778 dynamic_user_vacuum(m
, false);
3779 m
->uid_refs
= hashmap_free(m
->uid_refs
);
3780 m
->gid_refs
= hashmap_free(m
->gid_refs
);
3782 r
= lookup_paths_init(&m
->lookup_paths
, m
->unit_file_scope
, 0, NULL
);
3784 log_warning_errno(r
, "Failed to initialize path lookup table, ignoring: %m");
3786 (void) manager_run_environment_generators(m
);
3787 (void) manager_run_generators(m
);
3789 lookup_paths_log(&m
->lookup_paths
);
3791 /* We flushed out generated files, for which we don't watch mtime, so we should flush the old map. */
3792 manager_free_unit_name_maps(m
);
3794 /* First, enumerate what we can from kernel and suchlike */
3795 manager_enumerate_perpetual(m
);
3796 manager_enumerate(m
);
3798 /* Second, deserialize our stored data */
3799 r
= manager_deserialize(m
, f
, fds
);
3801 log_warning_errno(r
, "Deserialization failed, proceeding anyway: %m");
3803 /* We don't need the serialization anymore */
3806 /* Re-register notify_fd as event source, and set up other sockets/communication channels we might need */
3807 (void) manager_setup_notify(m
);
3808 (void) manager_setup_cgroups_agent(m
);
3809 (void) manager_setup_user_lookup_fd(m
);
3811 /* Third, fire things up! */
3812 manager_coldplug(m
);
3814 /* Clean up runtime objects no longer referenced */
3817 /* Consider the reload process complete now. */
3818 assert(m
->n_reloading
> 0);
3821 /* On manager reloading, device tag data should exists, thus, we should honor the results of device
3822 * enumeration. The flag should be always set correctly by the serialized data, but it may fail. So,
3823 * let's always set the flag here for safety. */
3824 m
->honor_device_enumeration
= true;
3828 m
->send_reloading_done
= true;
3832 void manager_reset_failed(Manager
*m
) {
3837 HASHMAP_FOREACH(u
, m
->units
)
3838 unit_reset_failed(u
);
3841 bool manager_unit_inactive_or_pending(Manager
*m
, const char *name
) {
3847 /* Returns true if the unit is inactive or going down */
3848 u
= manager_get_unit(m
, name
);
3852 return unit_inactive_or_pending(u
);
3855 static void log_taint_string(Manager
*m
) {
3856 _cleanup_free_
char *taint
= NULL
;
3860 if (MANAGER_IS_USER(m
) || m
->taint_logged
)
3863 m
->taint_logged
= true; /* only check for taint once */
3865 taint
= manager_taint_string(m
);
3869 log_struct(LOG_NOTICE
,
3870 LOG_MESSAGE("System is tainted: %s", taint
),
3872 "MESSAGE_ID=" SD_MESSAGE_TAINTED_STR
);
3875 static void manager_notify_finished(Manager
*m
) {
3876 char userspace
[FORMAT_TIMESPAN_MAX
], initrd
[FORMAT_TIMESPAN_MAX
], kernel
[FORMAT_TIMESPAN_MAX
], sum
[FORMAT_TIMESPAN_MAX
];
3877 usec_t firmware_usec
, loader_usec
, kernel_usec
, initrd_usec
, userspace_usec
, total_usec
;
3879 if (MANAGER_IS_TEST_RUN(m
))
3882 if (MANAGER_IS_SYSTEM(m
) && detect_container() <= 0) {
3883 char ts
[FORMAT_TIMESPAN_MAX
];
3884 char buf
[FORMAT_TIMESPAN_MAX
+ STRLEN(" (firmware) + ") + FORMAT_TIMESPAN_MAX
+ STRLEN(" (loader) + ")]
3887 size_t size
= sizeof buf
;
3889 /* Note that MANAGER_TIMESTAMP_KERNEL's monotonic value is always at 0, and
3890 * MANAGER_TIMESTAMP_FIRMWARE's and MANAGER_TIMESTAMP_LOADER's monotonic value should be considered
3891 * negative values. */
3893 firmware_usec
= m
->timestamps
[MANAGER_TIMESTAMP_FIRMWARE
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_LOADER
].monotonic
;
3894 loader_usec
= m
->timestamps
[MANAGER_TIMESTAMP_LOADER
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_KERNEL
].monotonic
;
3895 userspace_usec
= m
->timestamps
[MANAGER_TIMESTAMP_FINISH
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_USERSPACE
].monotonic
;
3896 total_usec
= m
->timestamps
[MANAGER_TIMESTAMP_FIRMWARE
].monotonic
+ m
->timestamps
[MANAGER_TIMESTAMP_FINISH
].monotonic
;
3898 if (firmware_usec
> 0)
3899 size
= strpcpyf(&p
, size
, "%s (firmware) + ", format_timespan(ts
, sizeof(ts
), firmware_usec
, USEC_PER_MSEC
));
3900 if (loader_usec
> 0)
3901 size
= strpcpyf(&p
, size
, "%s (loader) + ", format_timespan(ts
, sizeof(ts
), loader_usec
, USEC_PER_MSEC
));
3903 if (dual_timestamp_is_set(&m
->timestamps
[MANAGER_TIMESTAMP_INITRD
])) {
3905 /* The initrd case on bare-metal*/
3906 kernel_usec
= m
->timestamps
[MANAGER_TIMESTAMP_INITRD
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_KERNEL
].monotonic
;
3907 initrd_usec
= m
->timestamps
[MANAGER_TIMESTAMP_USERSPACE
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_INITRD
].monotonic
;
3909 log_struct(LOG_INFO
,
3910 "MESSAGE_ID=" SD_MESSAGE_STARTUP_FINISHED_STR
,
3911 "KERNEL_USEC="USEC_FMT
, kernel_usec
,
3912 "INITRD_USEC="USEC_FMT
, initrd_usec
,
3913 "USERSPACE_USEC="USEC_FMT
, userspace_usec
,
3914 LOG_MESSAGE("Startup finished in %s%s (kernel) + %s (initrd) + %s (userspace) = %s.",
3916 format_timespan(kernel
, sizeof(kernel
), kernel_usec
, USEC_PER_MSEC
),
3917 format_timespan(initrd
, sizeof(initrd
), initrd_usec
, USEC_PER_MSEC
),
3918 format_timespan(userspace
, sizeof(userspace
), userspace_usec
, USEC_PER_MSEC
),
3919 format_timespan(sum
, sizeof(sum
), total_usec
, USEC_PER_MSEC
)));
3921 /* The initrd-less case on bare-metal*/
3923 kernel_usec
= m
->timestamps
[MANAGER_TIMESTAMP_USERSPACE
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_KERNEL
].monotonic
;
3926 log_struct(LOG_INFO
,
3927 "MESSAGE_ID=" SD_MESSAGE_STARTUP_FINISHED_STR
,
3928 "KERNEL_USEC="USEC_FMT
, kernel_usec
,
3929 "USERSPACE_USEC="USEC_FMT
, userspace_usec
,
3930 LOG_MESSAGE("Startup finished in %s%s (kernel) + %s (userspace) = %s.",
3932 format_timespan(kernel
, sizeof(kernel
), kernel_usec
, USEC_PER_MSEC
),
3933 format_timespan(userspace
, sizeof(userspace
), userspace_usec
, USEC_PER_MSEC
),
3934 format_timespan(sum
, sizeof(sum
), total_usec
, USEC_PER_MSEC
)));
3937 /* The container and --user case */
3938 firmware_usec
= loader_usec
= initrd_usec
= kernel_usec
= 0;
3939 total_usec
= userspace_usec
= m
->timestamps
[MANAGER_TIMESTAMP_FINISH
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_USERSPACE
].monotonic
;
3941 log_struct(LOG_INFO
,
3942 "MESSAGE_ID=" SD_MESSAGE_USER_STARTUP_FINISHED_STR
,
3943 "USERSPACE_USEC="USEC_FMT
, userspace_usec
,
3944 LOG_MESSAGE("Startup finished in %s.",
3945 format_timespan(sum
, sizeof(sum
), total_usec
, USEC_PER_MSEC
)));
3948 bus_manager_send_finished(m
, firmware_usec
, loader_usec
, kernel_usec
, initrd_usec
, userspace_usec
, total_usec
);
3951 m
->ready_sent
? "STATUS=Startup finished in %s."
3953 "STATUS=Startup finished in %s.",
3954 format_timespan(sum
, sizeof(sum
), total_usec
, USEC_PER_MSEC
));
3955 m
->ready_sent
= true;
3957 log_taint_string(m
);
3960 static void manager_send_ready(Manager
*m
) {
3963 /* We send READY=1 on reaching basic.target only when running in --user mode. */
3964 if (!MANAGER_IS_USER(m
) || m
->ready_sent
)
3967 m
->ready_sent
= true;
3971 "STATUS=Reached " SPECIAL_BASIC_TARGET
".");
3974 static void manager_check_basic_target(Manager
*m
) {
3979 /* Small shortcut */
3980 if (m
->ready_sent
&& m
->taint_logged
)
3983 u
= manager_get_unit(m
, SPECIAL_BASIC_TARGET
);
3984 if (!u
|| !UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u
)))
3987 /* For user managers, send out READY=1 as soon as we reach basic.target */
3988 manager_send_ready(m
);
3990 /* Log the taint string as soon as we reach basic.target */
3991 log_taint_string(m
);
3994 void manager_check_finished(Manager
*m
) {
3997 if (MANAGER_IS_RELOADING(m
))
4000 /* Verify that we have entered the event loop already, and not left it again. */
4001 if (!MANAGER_IS_RUNNING(m
))
4004 manager_check_basic_target(m
);
4006 if (hashmap_size(m
->jobs
) > 0) {
4007 if (m
->jobs_in_progress_event_source
)
4008 /* Ignore any failure, this is only for feedback */
4009 (void) sd_event_source_set_time(m
->jobs_in_progress_event_source
,
4010 manager_watch_jobs_next_time(m
));
4014 /* The jobs hashmap tends to grow a lot during boot, and then it's not reused until shutdown. Let's
4015 kill the hashmap if it is relatively large. */
4016 if (hashmap_buckets(m
->jobs
) > hashmap_size(m
->units
) / 10)
4017 m
->jobs
= hashmap_free(m
->jobs
);
4019 manager_flip_auto_status(m
, false, "boot finished");
4021 /* Notify Type=idle units that we are done now */
4022 manager_close_idle_pipe(m
);
4024 /* Turn off confirm spawn now */
4025 m
->confirm_spawn
= NULL
;
4027 /* No need to update ask password status when we're going non-interactive */
4028 manager_close_ask_password(m
);
4030 /* This is no longer the first boot */
4031 manager_set_first_boot(m
, false);
4033 if (MANAGER_IS_FINISHED(m
))
4036 dual_timestamp_get(m
->timestamps
+ MANAGER_TIMESTAMP_FINISH
);
4038 manager_notify_finished(m
);
4040 manager_invalidate_startup_units(m
);
4043 static bool generator_path_any(const char* const* paths
) {
4047 /* Optimize by skipping the whole process by not creating output directories
4048 * if no generators are found. */
4049 STRV_FOREACH(path
, (char**) paths
)
4050 if (access(*path
, F_OK
) == 0)
4052 else if (errno
!= ENOENT
)
4053 log_warning_errno(errno
, "Failed to open generator directory %s: %m", *path
);
4058 static int manager_run_environment_generators(Manager
*m
) {
4059 char **tmp
= NULL
; /* this is only used in the forked process, no cleanup here */
4060 _cleanup_strv_free_
char **paths
= NULL
;
4062 [STDOUT_GENERATE
] = &tmp
,
4063 [STDOUT_COLLECT
] = &tmp
,
4064 [STDOUT_CONSUME
] = &m
->transient_environment
,
4068 if (MANAGER_IS_TEST_RUN(m
) && !(m
->test_run_flags
& MANAGER_TEST_RUN_ENV_GENERATORS
))
4071 paths
= env_generator_binary_paths(MANAGER_IS_SYSTEM(m
));
4075 if (!generator_path_any((const char* const*) paths
))
4078 RUN_WITH_UMASK(0022)
4079 r
= execute_directories((const char* const*) paths
, DEFAULT_TIMEOUT_USEC
, gather_environment
,
4080 args
, NULL
, m
->transient_environment
, EXEC_DIR_PARALLEL
| EXEC_DIR_IGNORE_ERRORS
);
4084 static int manager_run_generators(Manager
*m
) {
4085 _cleanup_strv_free_
char **paths
= NULL
;
4086 const char *argv
[5];
4091 if (MANAGER_IS_TEST_RUN(m
) && !(m
->test_run_flags
& MANAGER_TEST_RUN_GENERATORS
))
4094 paths
= generator_binary_paths(m
->unit_file_scope
);
4098 if (!generator_path_any((const char* const*) paths
))
4101 r
= lookup_paths_mkdir_generator(&m
->lookup_paths
);
4103 log_error_errno(r
, "Failed to create generator directories: %m");
4107 argv
[0] = NULL
; /* Leave this empty, execute_directory() will fill something in */
4108 argv
[1] = m
->lookup_paths
.generator
;
4109 argv
[2] = m
->lookup_paths
.generator_early
;
4110 argv
[3] = m
->lookup_paths
.generator_late
;
4113 RUN_WITH_UMASK(0022)
4114 (void) execute_directories((const char* const*) paths
, DEFAULT_TIMEOUT_USEC
, NULL
, NULL
,
4115 (char**) argv
, m
->transient_environment
, EXEC_DIR_PARALLEL
| EXEC_DIR_IGNORE_ERRORS
);
4120 lookup_paths_trim_generator(&m
->lookup_paths
);
4124 int manager_transient_environment_add(Manager
*m
, char **plus
) {
4129 if (strv_isempty(plus
))
4132 a
= strv_env_merge(2, m
->transient_environment
, plus
);
4136 sanitize_environment(a
);
4138 return strv_free_and_replace(m
->transient_environment
, a
);
4141 int manager_client_environment_modify(
4146 char **a
= NULL
, **b
= NULL
, **l
;
4150 if (strv_isempty(minus
) && strv_isempty(plus
))
4153 l
= m
->client_environment
;
4155 if (!strv_isempty(minus
)) {
4156 a
= strv_env_delete(l
, 1, minus
);
4163 if (!strv_isempty(plus
)) {
4164 b
= strv_env_merge(2, l
, plus
);
4173 if (m
->client_environment
!= l
)
4174 strv_free(m
->client_environment
);
4181 m
->client_environment
= sanitize_environment(l
);
4185 int manager_get_effective_environment(Manager
*m
, char ***ret
) {
4191 l
= strv_env_merge(2, m
->transient_environment
, m
->client_environment
);
4199 int manager_set_default_rlimits(Manager
*m
, struct rlimit
**default_rlimit
) {
4202 for (unsigned i
= 0; i
< _RLIMIT_MAX
; i
++) {
4203 m
->rlimit
[i
] = mfree(m
->rlimit
[i
]);
4205 if (!default_rlimit
[i
])
4208 m
->rlimit
[i
] = newdup(struct rlimit
, default_rlimit
[i
], 1);
4216 void manager_recheck_dbus(Manager
*m
) {
4219 /* Connects to the bus if the dbus service and socket are running. If we are running in user mode this is all
4220 * it does. In system mode we'll also connect to the system bus (which will most likely just reuse the
4221 * connection of the API bus). That's because the system bus after all runs as service of the system instance,
4222 * while in the user instance we can assume it's already there. */
4224 if (MANAGER_IS_RELOADING(m
))
4225 return; /* don't check while we are reloading… */
4227 if (manager_dbus_is_running(m
, false)) {
4228 (void) bus_init_api(m
);
4230 if (MANAGER_IS_SYSTEM(m
))
4231 (void) bus_init_system(m
);
4233 (void) bus_done_api(m
);
4235 if (MANAGER_IS_SYSTEM(m
))
4236 (void) bus_done_system(m
);
4240 static bool manager_journal_is_running(Manager
*m
) {
4245 if (MANAGER_IS_TEST_RUN(m
))
4248 /* If we are the user manager we can safely assume that the journal is up */
4249 if (!MANAGER_IS_SYSTEM(m
))
4252 /* Check that the socket is not only up, but in RUNNING state */
4253 u
= manager_get_unit(m
, SPECIAL_JOURNALD_SOCKET
);
4256 if (SOCKET(u
)->state
!= SOCKET_RUNNING
)
4259 /* Similar, check if the daemon itself is fully up, too */
4260 u
= manager_get_unit(m
, SPECIAL_JOURNALD_SERVICE
);
4263 if (!IN_SET(SERVICE(u
)->state
, SERVICE_RELOAD
, SERVICE_RUNNING
))
4269 void disable_printk_ratelimit(void) {
4270 /* Disable kernel's printk ratelimit.
4272 * Logging to /dev/kmsg is most useful during early boot and shutdown, where normal logging
4273 * mechanisms are not available. The semantics of this sysctl are such that any kernel command-line
4274 * setting takes precedence. */
4277 r
= sysctl_write("kernel/printk_devkmsg", "on");
4279 log_debug_errno(r
, "Failed to set sysctl kernel.printk_devkmsg=on: %m");
4282 void manager_recheck_journal(Manager
*m
) {
4286 /* Don't bother with this unless we are in the special situation of being PID 1 */
4287 if (getpid_cached() != 1)
4290 /* Don't check this while we are reloading, things might still change */
4291 if (MANAGER_IS_RELOADING(m
))
4294 /* The journal is fully and entirely up? If so, let's permit logging to it, if that's configured. If the
4295 * journal is down, don't ever log to it, otherwise we might end up deadlocking ourselves as we might trigger
4296 * an activation ourselves we can't fulfill. */
4297 log_set_prohibit_ipc(!manager_journal_is_running(m
));
4301 static ShowStatus
manager_get_show_status(Manager
*m
) {
4304 if (MANAGER_IS_USER(m
))
4305 return _SHOW_STATUS_INVALID
;
4307 if (m
->show_status_overridden
!= _SHOW_STATUS_INVALID
)
4308 return m
->show_status_overridden
;
4310 return m
->show_status
;
4313 bool manager_get_show_status_on(Manager
*m
) {
4316 return show_status_on(manager_get_show_status(m
));
4319 static void set_show_status_marker(bool b
) {
4321 (void) touch("/run/systemd/show-status");
4323 (void) unlink("/run/systemd/show-status");
4326 void manager_set_show_status(Manager
*m
, ShowStatus mode
, const char *reason
) {
4329 assert(mode
>= 0 && mode
< _SHOW_STATUS_MAX
);
4331 if (MANAGER_IS_USER(m
))
4334 if (mode
== m
->show_status
)
4337 if (m
->show_status_overridden
== _SHOW_STATUS_INVALID
) {
4340 enabled
= show_status_on(mode
);
4341 log_debug("%s (%s) showing of status (%s).",
4342 enabled
? "Enabling" : "Disabling",
4343 strna(show_status_to_string(mode
)),
4346 set_show_status_marker(enabled
);
4349 m
->show_status
= mode
;
4352 void manager_override_show_status(Manager
*m
, ShowStatus mode
, const char *reason
) {
4354 assert(mode
< _SHOW_STATUS_MAX
);
4356 if (MANAGER_IS_USER(m
))
4359 if (mode
== m
->show_status_overridden
)
4362 m
->show_status_overridden
= mode
;
4364 if (mode
== _SHOW_STATUS_INVALID
)
4365 mode
= m
->show_status
;
4367 log_debug("%s (%s) showing of status (%s).",
4368 m
->show_status_overridden
!= _SHOW_STATUS_INVALID
? "Overriding" : "Restoring",
4369 strna(show_status_to_string(mode
)),
4372 set_show_status_marker(show_status_on(mode
));
4375 const char *manager_get_confirm_spawn(Manager
*m
) {
4376 static int last_errno
= 0;
4382 /* Here's the deal: we want to test the validity of the console but don't want
4383 * PID1 to go through the whole console process which might block. But we also
4384 * want to warn the user only once if something is wrong with the console so we
4385 * cannot do the sanity checks after spawning our children. So here we simply do
4386 * really basic tests to hopefully trap common errors.
4388 * If the console suddenly disappear at the time our children will really it
4389 * then they will simply fail to acquire it and a positive answer will be
4390 * assumed. New children will fall back to /dev/console though.
4392 * Note: TTYs are devices that can come and go any time, and frequently aren't
4393 * available yet during early boot (consider a USB rs232 dongle...). If for any
4394 * reason the configured console is not ready, we fall back to the default
4397 if (!m
->confirm_spawn
|| path_equal(m
->confirm_spawn
, "/dev/console"))
4398 return m
->confirm_spawn
;
4400 if (stat(m
->confirm_spawn
, &st
) < 0) {
4405 if (!S_ISCHR(st
.st_mode
)) {
4411 return m
->confirm_spawn
;
4414 if (last_errno
!= r
)
4415 last_errno
= log_warning_errno(r
, "Failed to open %s, using default console: %m", m
->confirm_spawn
);
4417 return "/dev/console";
4420 void manager_set_first_boot(Manager
*m
, bool b
) {
4423 if (!MANAGER_IS_SYSTEM(m
))
4426 if (m
->first_boot
!= (int) b
) {
4428 (void) touch("/run/systemd/first-boot");
4430 (void) unlink("/run/systemd/first-boot");
4436 void manager_disable_confirm_spawn(void) {
4437 (void) touch("/run/systemd/confirm_spawn_disabled");
4440 bool manager_is_confirm_spawn_disabled(Manager
*m
) {
4441 if (!m
->confirm_spawn
)
4444 return access("/run/systemd/confirm_spawn_disabled", F_OK
) >= 0;
4447 static bool manager_should_show_status(Manager
*m
, StatusType type
) {
4450 if (!MANAGER_IS_SYSTEM(m
))
4453 if (m
->no_console_output
)
4456 if (!IN_SET(manager_state(m
), MANAGER_INITIALIZING
, MANAGER_STARTING
, MANAGER_STOPPING
))
4459 /* If we cannot find out the status properly, just proceed. */
4460 if (type
!= STATUS_TYPE_EMERGENCY
&& manager_check_ask_password(m
) > 0)
4463 if (type
== STATUS_TYPE_NOTICE
&& m
->show_status
!= SHOW_STATUS_NO
)
4466 return manager_get_show_status_on(m
);
4469 void manager_status_printf(Manager
*m
, StatusType type
, const char *status
, const char *format
, ...) {
4472 /* If m is NULL, assume we're after shutdown and let the messages through. */
4474 if (m
&& !manager_should_show_status(m
, type
))
4477 /* XXX We should totally drop the check for ephemeral here
4478 * and thus effectively make 'Type=idle' pointless. */
4479 if (type
== STATUS_TYPE_EPHEMERAL
&& m
&& m
->n_on_console
> 0)
4482 va_start(ap
, format
);
4483 status_vprintf(status
, SHOW_STATUS_ELLIPSIZE
|(type
== STATUS_TYPE_EPHEMERAL
? SHOW_STATUS_EPHEMERAL
: 0), format
, ap
);
4487 Set
*manager_get_units_requiring_mounts_for(Manager
*m
, const char *path
) {
4488 char p
[strlen(path
)+1];
4494 path_simplify(p
, false);
4496 return hashmap_get(m
->units_requiring_mounts_for
, streq(p
, "/") ? "" : p
);
4499 int manager_update_failed_units(Manager
*m
, Unit
*u
, bool failed
) {
4504 assert(u
->manager
== m
);
4506 size
= set_size(m
->failed_units
);
4509 r
= set_ensure_put(&m
->failed_units
, NULL
, u
);
4513 (void) set_remove(m
->failed_units
, u
);
4515 if (set_size(m
->failed_units
) != size
)
4516 bus_manager_send_change_signal(m
);
4521 ManagerState
manager_state(Manager
*m
) {
4526 /* Is the special shutdown target active or queued? If so, we are in shutdown state */
4527 u
= manager_get_unit(m
, SPECIAL_SHUTDOWN_TARGET
);
4528 if (u
&& unit_active_or_pending(u
))
4529 return MANAGER_STOPPING
;
4531 /* Did we ever finish booting? If not then we are still starting up */
4532 if (!MANAGER_IS_FINISHED(m
)) {
4534 u
= manager_get_unit(m
, SPECIAL_BASIC_TARGET
);
4535 if (!u
|| !UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u
)))
4536 return MANAGER_INITIALIZING
;
4538 return MANAGER_STARTING
;
4541 if (MANAGER_IS_SYSTEM(m
)) {
4542 /* Are the rescue or emergency targets active or queued? If so we are in maintenance state */
4543 u
= manager_get_unit(m
, SPECIAL_RESCUE_TARGET
);
4544 if (u
&& unit_active_or_pending(u
))
4545 return MANAGER_MAINTENANCE
;
4547 u
= manager_get_unit(m
, SPECIAL_EMERGENCY_TARGET
);
4548 if (u
&& unit_active_or_pending(u
))
4549 return MANAGER_MAINTENANCE
;
4552 /* Are there any failed units? If so, we are in degraded mode */
4553 if (set_size(m
->failed_units
) > 0)
4554 return MANAGER_DEGRADED
;
4556 return MANAGER_RUNNING
;
4559 static void manager_unref_uid_internal(
4564 int (*_clean_ipc
)(uid_t uid
)) {
4570 assert(uid_is_valid(uid
));
4573 /* A generic implementation, covering both manager_unref_uid() and manager_unref_gid(), under the assumption
4574 * that uid_t and gid_t are actually defined the same way, with the same validity rules.
4576 * We store a hashmap where the UID/GID is they key and the value is a 32bit reference counter, whose highest
4577 * bit is used as flag for marking UIDs/GIDs whose IPC objects to remove when the last reference to the UID/GID
4578 * is dropped. The flag is set to on, once at least one reference from a unit where RemoveIPC= is set is added
4579 * on a UID/GID. It is reset when the UID's/GID's reference counter drops to 0 again. */
4581 assert_cc(sizeof(uid_t
) == sizeof(gid_t
));
4582 assert_cc(UID_INVALID
== (uid_t
) GID_INVALID
);
4584 if (uid
== 0) /* We don't keep track of root, and will never destroy it */
4587 c
= PTR_TO_UINT32(hashmap_get(*uid_refs
, UID_TO_PTR(uid
)));
4589 n
= c
& ~DESTROY_IPC_FLAG
;
4593 if (destroy_now
&& n
== 0) {
4594 hashmap_remove(*uid_refs
, UID_TO_PTR(uid
));
4596 if (c
& DESTROY_IPC_FLAG
) {
4597 log_debug("%s " UID_FMT
" is no longer referenced, cleaning up its IPC.",
4598 _clean_ipc
== clean_ipc_by_uid
? "UID" : "GID",
4600 (void) _clean_ipc(uid
);
4603 c
= n
| (c
& DESTROY_IPC_FLAG
);
4604 assert_se(hashmap_update(*uid_refs
, UID_TO_PTR(uid
), UINT32_TO_PTR(c
)) >= 0);
4608 void manager_unref_uid(Manager
*m
, uid_t uid
, bool destroy_now
) {
4609 manager_unref_uid_internal(m
, &m
->uid_refs
, uid
, destroy_now
, clean_ipc_by_uid
);
4612 void manager_unref_gid(Manager
*m
, gid_t gid
, bool destroy_now
) {
4613 manager_unref_uid_internal(m
, &m
->gid_refs
, (uid_t
) gid
, destroy_now
, clean_ipc_by_gid
);
4616 static int manager_ref_uid_internal(
4627 assert(uid_is_valid(uid
));
4629 /* A generic implementation, covering both manager_ref_uid() and manager_ref_gid(), under the assumption
4630 * that uid_t and gid_t are actually defined the same way, with the same validity rules. */
4632 assert_cc(sizeof(uid_t
) == sizeof(gid_t
));
4633 assert_cc(UID_INVALID
== (uid_t
) GID_INVALID
);
4635 if (uid
== 0) /* We don't keep track of root, and will never destroy it */
4638 r
= hashmap_ensure_allocated(uid_refs
, &trivial_hash_ops
);
4642 c
= PTR_TO_UINT32(hashmap_get(*uid_refs
, UID_TO_PTR(uid
)));
4644 n
= c
& ~DESTROY_IPC_FLAG
;
4647 if (n
& DESTROY_IPC_FLAG
) /* check for overflow */
4650 c
= n
| (c
& DESTROY_IPC_FLAG
) | (clean_ipc
? DESTROY_IPC_FLAG
: 0);
4652 return hashmap_replace(*uid_refs
, UID_TO_PTR(uid
), UINT32_TO_PTR(c
));
4655 int manager_ref_uid(Manager
*m
, uid_t uid
, bool clean_ipc
) {
4656 return manager_ref_uid_internal(m
, &m
->uid_refs
, uid
, clean_ipc
);
4659 int manager_ref_gid(Manager
*m
, gid_t gid
, bool clean_ipc
) {
4660 return manager_ref_uid_internal(m
, &m
->gid_refs
, (uid_t
) gid
, clean_ipc
);
4663 static void manager_vacuum_uid_refs_internal(
4666 int (*_clean_ipc
)(uid_t uid
)) {
4674 HASHMAP_FOREACH_KEY(p
, k
, *uid_refs
) {
4678 uid
= PTR_TO_UID(k
);
4679 c
= PTR_TO_UINT32(p
);
4681 n
= c
& ~DESTROY_IPC_FLAG
;
4685 if (c
& DESTROY_IPC_FLAG
) {
4686 log_debug("Found unreferenced %s " UID_FMT
" after reload/reexec. Cleaning up.",
4687 _clean_ipc
== clean_ipc_by_uid
? "UID" : "GID",
4689 (void) _clean_ipc(uid
);
4692 assert_se(hashmap_remove(*uid_refs
, k
) == p
);
4696 static void manager_vacuum_uid_refs(Manager
*m
) {
4697 manager_vacuum_uid_refs_internal(m
, &m
->uid_refs
, clean_ipc_by_uid
);
4700 static void manager_vacuum_gid_refs(Manager
*m
) {
4701 manager_vacuum_uid_refs_internal(m
, &m
->gid_refs
, clean_ipc_by_gid
);
4704 static void manager_vacuum(Manager
*m
) {
4707 /* Release any dynamic users no longer referenced */
4708 dynamic_user_vacuum(m
, true);
4710 /* Release any references to UIDs/GIDs no longer referenced, and destroy any IPC owned by them */
4711 manager_vacuum_uid_refs(m
);
4712 manager_vacuum_gid_refs(m
);
4714 /* Release any runtimes no longer referenced */
4715 exec_runtime_vacuum(m
);
4718 int manager_dispatch_user_lookup_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
4722 char unit_name
[UNIT_NAME_MAX
+1];
4725 Manager
*m
= userdata
;
4733 /* Invoked whenever a child process succeeded resolving its user/group to use and sent us the resulting UID/GID
4734 * in a datagram. We parse the datagram here and pass it off to the unit, so that it can add a reference to the
4735 * UID/GID so that it can destroy the UID/GID's IPC objects when the reference counter drops to 0. */
4737 l
= recv(fd
, &buffer
, sizeof(buffer
), MSG_DONTWAIT
);
4739 if (IN_SET(errno
, EINTR
, EAGAIN
))
4742 return log_error_errno(errno
, "Failed to read from user lookup fd: %m");
4745 if ((size_t) l
<= offsetof(struct buffer
, unit_name
)) {
4746 log_warning("Received too short user lookup message, ignoring.");
4750 if ((size_t) l
> offsetof(struct buffer
, unit_name
) + UNIT_NAME_MAX
) {
4751 log_warning("Received too long user lookup message, ignoring.");
4755 if (!uid_is_valid(buffer
.uid
) && !gid_is_valid(buffer
.gid
)) {
4756 log_warning("Got user lookup message with invalid UID/GID pair, ignoring.");
4760 n
= (size_t) l
- offsetof(struct buffer
, unit_name
);
4761 if (memchr(buffer
.unit_name
, 0, n
)) {
4762 log_warning("Received lookup message with embedded NUL character, ignoring.");
4766 buffer
.unit_name
[n
] = 0;
4767 u
= manager_get_unit(m
, buffer
.unit_name
);
4769 log_debug("Got user lookup message but unit doesn't exist, ignoring.");
4773 log_unit_debug(u
, "User lookup succeeded: uid=" UID_FMT
" gid=" GID_FMT
, buffer
.uid
, buffer
.gid
);
4775 unit_notify_user_lookup(u
, buffer
.uid
, buffer
.gid
);
4779 char *manager_taint_string(Manager
*m
) {
4780 _cleanup_free_
char *destination
= NULL
, *overflowuid
= NULL
, *overflowgid
= NULL
;
4784 /* Returns a "taint string", e.g. "local-hwclock:var-run-bad".
4785 * Only things that are detected at runtime should be tagged
4786 * here. For stuff that is set during compilation, emit a warning
4787 * in the configuration phase. */
4791 buf
= new(char, sizeof("split-usr:"
4795 "overflowuid-not-65534:"
4796 "overflowgid-not-65534:"));
4804 e
= stpcpy(e
, "split-usr:");
4806 if (access("/proc/cgroups", F_OK
) < 0)
4807 e
= stpcpy(e
, "cgroups-missing:");
4809 if (clock_is_localtime(NULL
) > 0)
4810 e
= stpcpy(e
, "local-hwclock:");
4812 r
= readlink_malloc("/var/run", &destination
);
4813 if (r
< 0 || !PATH_IN_SET(destination
, "../run", "/run"))
4814 e
= stpcpy(e
, "var-run-bad:");
4816 r
= read_one_line_file("/proc/sys/kernel/overflowuid", &overflowuid
);
4817 if (r
>= 0 && !streq(overflowuid
, "65534"))
4818 e
= stpcpy(e
, "overflowuid-not-65534:");
4820 r
= read_one_line_file("/proc/sys/kernel/overflowgid", &overflowgid
);
4821 if (r
>= 0 && !streq(overflowgid
, "65534"))
4822 e
= stpcpy(e
, "overflowgid-not-65534:");
4824 /* remove the last ':' */
4831 void manager_ref_console(Manager
*m
) {
4837 void manager_unref_console(Manager
*m
) {
4839 assert(m
->n_on_console
> 0);
4842 if (m
->n_on_console
== 0)
4843 m
->no_console_output
= false; /* unset no_console_output flag, since the console is definitely free now */
4846 void manager_override_log_level(Manager
*m
, int level
) {
4847 _cleanup_free_
char *s
= NULL
;
4850 if (!m
->log_level_overridden
) {
4851 m
->original_log_level
= log_get_max_level();
4852 m
->log_level_overridden
= true;
4855 (void) log_level_to_string_alloc(level
, &s
);
4856 log_info("Setting log level to %s.", strna(s
));
4858 log_set_max_level(level
);
4861 void manager_restore_original_log_level(Manager
*m
) {
4862 _cleanup_free_
char *s
= NULL
;
4865 if (!m
->log_level_overridden
)
4868 (void) log_level_to_string_alloc(m
->original_log_level
, &s
);
4869 log_info("Restoring log level to original (%s).", strna(s
));
4871 log_set_max_level(m
->original_log_level
);
4872 m
->log_level_overridden
= false;
4875 void manager_override_log_target(Manager
*m
, LogTarget target
) {
4878 if (!m
->log_target_overridden
) {
4879 m
->original_log_target
= log_get_target();
4880 m
->log_target_overridden
= true;
4883 log_info("Setting log target to %s.", log_target_to_string(target
));
4884 log_set_target(target
);
4887 void manager_restore_original_log_target(Manager
*m
) {
4890 if (!m
->log_target_overridden
)
4893 log_info("Restoring log target to original %s.", log_target_to_string(m
->original_log_target
));
4895 log_set_target(m
->original_log_target
);
4896 m
->log_target_overridden
= false;
4899 ManagerTimestamp
manager_timestamp_initrd_mangle(ManagerTimestamp s
) {
4901 s
>= MANAGER_TIMESTAMP_SECURITY_START
&&
4902 s
<= MANAGER_TIMESTAMP_UNITS_LOAD_FINISH
)
4903 return s
- MANAGER_TIMESTAMP_SECURITY_START
+ MANAGER_TIMESTAMP_INITRD_SECURITY_START
;
4907 static const char *const manager_state_table
[_MANAGER_STATE_MAX
] = {
4908 [MANAGER_INITIALIZING
] = "initializing",
4909 [MANAGER_STARTING
] = "starting",
4910 [MANAGER_RUNNING
] = "running",
4911 [MANAGER_DEGRADED
] = "degraded",
4912 [MANAGER_MAINTENANCE
] = "maintenance",
4913 [MANAGER_STOPPING
] = "stopping",
4916 DEFINE_STRING_TABLE_LOOKUP(manager_state
, ManagerState
);
4918 static const char *const manager_timestamp_table
[_MANAGER_TIMESTAMP_MAX
] = {
4919 [MANAGER_TIMESTAMP_FIRMWARE
] = "firmware",
4920 [MANAGER_TIMESTAMP_LOADER
] = "loader",
4921 [MANAGER_TIMESTAMP_KERNEL
] = "kernel",
4922 [MANAGER_TIMESTAMP_INITRD
] = "initrd",
4923 [MANAGER_TIMESTAMP_USERSPACE
] = "userspace",
4924 [MANAGER_TIMESTAMP_FINISH
] = "finish",
4925 [MANAGER_TIMESTAMP_SECURITY_START
] = "security-start",
4926 [MANAGER_TIMESTAMP_SECURITY_FINISH
] = "security-finish",
4927 [MANAGER_TIMESTAMP_GENERATORS_START
] = "generators-start",
4928 [MANAGER_TIMESTAMP_GENERATORS_FINISH
] = "generators-finish",
4929 [MANAGER_TIMESTAMP_UNITS_LOAD_START
] = "units-load-start",
4930 [MANAGER_TIMESTAMP_UNITS_LOAD_FINISH
] = "units-load-finish",
4931 [MANAGER_TIMESTAMP_INITRD_SECURITY_START
] = "initrd-security-start",
4932 [MANAGER_TIMESTAMP_INITRD_SECURITY_FINISH
] = "initrd-security-finish",
4933 [MANAGER_TIMESTAMP_INITRD_GENERATORS_START
] = "initrd-generators-start",
4934 [MANAGER_TIMESTAMP_INITRD_GENERATORS_FINISH
] = "initrd-generators-finish",
4935 [MANAGER_TIMESTAMP_INITRD_UNITS_LOAD_START
] = "initrd-units-load-start",
4936 [MANAGER_TIMESTAMP_INITRD_UNITS_LOAD_FINISH
] = "initrd-units-load-finish",
4939 DEFINE_STRING_TABLE_LOOKUP(manager_timestamp
, ManagerTimestamp
);
4941 static const char* const oom_policy_table
[_OOM_POLICY_MAX
] = {
4942 [OOM_CONTINUE
] = "continue",
4943 [OOM_STOP
] = "stop",
4944 [OOM_KILL
] = "kill",
4947 DEFINE_STRING_TABLE_LOOKUP(oom_policy
, OOMPolicy
);