1 /* SPDX-License-Identifier: LGPL-2.1+ */
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
;
203 unsigned counter
= 0, print_nr
;
204 char cylon
[6 + CYLON_BUFFER_EXTRA
+ 1];
206 char time
[FORMAT_TIMESPAN_MAX
], limit
[FORMAT_TIMESPAN_MAX
] = "no limit";
210 assert(m
->n_running_jobs
> 0);
212 manager_flip_auto_status(m
, true, "delay");
214 print_nr
= (m
->jobs_in_progress_iteration
/ JOBS_IN_PROGRESS_PERIOD_DIVISOR
) % m
->n_running_jobs
;
216 HASHMAP_FOREACH(j
, m
->jobs
, i
)
217 if (j
->state
== JOB_RUNNING
&& counter
++ == print_nr
)
220 /* m->n_running_jobs must be consistent with the contents of m->jobs,
221 * so the above loop must have succeeded in finding j. */
222 assert(counter
== print_nr
+ 1);
225 cylon_pos
= m
->jobs_in_progress_iteration
% 14;
227 cylon_pos
= 14 - cylon_pos
;
228 draw_cylon(cylon
, sizeof(cylon
), 6, cylon_pos
);
230 m
->jobs_in_progress_iteration
++;
232 if (m
->n_running_jobs
> 1) {
233 if (asprintf(&job_of_n
, "(%u of %u) ", counter
, m
->n_running_jobs
) < 0)
237 format_timespan(time
, sizeof(time
), now(CLOCK_MONOTONIC
) - j
->begin_usec
, 1*USEC_PER_SEC
);
238 if (job_get_timeout(j
, &x
) > 0)
239 format_timespan(limit
, sizeof(limit
), x
- j
->begin_usec
, 1*USEC_PER_SEC
);
241 manager_status_printf(m
, STATUS_TYPE_EPHEMERAL
, cylon
,
242 "%sA %s job is running for %s (%s / %s)",
244 job_type_to_string(j
->type
),
245 unit_status_string(j
->unit
),
249 static int have_ask_password(void) {
250 _cleanup_closedir_
DIR *dir
;
253 dir
= opendir("/run/systemd/ask-password");
261 FOREACH_DIRENT_ALL(de
, dir
, return -errno
) {
262 if (startswith(de
->d_name
, "ask."))
268 static int manager_dispatch_ask_password_fd(sd_event_source
*source
,
269 int fd
, uint32_t revents
, void *userdata
) {
270 Manager
*m
= userdata
;
276 m
->have_ask_password
= have_ask_password();
277 if (m
->have_ask_password
< 0)
278 /* Log error but continue. Negative have_ask_password
279 * is treated as unknown status. */
280 log_error_errno(m
->have_ask_password
, "Failed to list /run/systemd/ask-password: %m");
285 static void manager_close_ask_password(Manager
*m
) {
288 m
->ask_password_event_source
= sd_event_source_unref(m
->ask_password_event_source
);
289 m
->ask_password_inotify_fd
= safe_close(m
->ask_password_inotify_fd
);
290 m
->have_ask_password
= -EINVAL
;
293 static int manager_check_ask_password(Manager
*m
) {
298 if (!m
->ask_password_event_source
) {
299 assert(m
->ask_password_inotify_fd
< 0);
301 (void) mkdir_p_label("/run/systemd/ask-password", 0755);
303 m
->ask_password_inotify_fd
= inotify_init1(IN_NONBLOCK
|IN_CLOEXEC
);
304 if (m
->ask_password_inotify_fd
< 0)
305 return log_error_errno(errno
, "Failed to create inotify object: %m");
307 r
= inotify_add_watch_and_warn(m
->ask_password_inotify_fd
,
308 "/run/systemd/ask-password",
309 IN_CREATE
|IN_DELETE
|IN_MOVE
);
311 manager_close_ask_password(m
);
315 r
= sd_event_add_io(m
->event
, &m
->ask_password_event_source
,
316 m
->ask_password_inotify_fd
, EPOLLIN
,
317 manager_dispatch_ask_password_fd
, m
);
319 log_error_errno(r
, "Failed to add event source for /run/systemd/ask-password: %m");
320 manager_close_ask_password(m
);
324 (void) sd_event_source_set_description(m
->ask_password_event_source
, "manager-ask-password");
326 /* Queries might have been added meanwhile... */
327 manager_dispatch_ask_password_fd(m
->ask_password_event_source
,
328 m
->ask_password_inotify_fd
, EPOLLIN
, m
);
331 return m
->have_ask_password
;
334 static int manager_watch_idle_pipe(Manager
*m
) {
339 if (m
->idle_pipe_event_source
)
342 if (m
->idle_pipe
[2] < 0)
345 r
= sd_event_add_io(m
->event
, &m
->idle_pipe_event_source
, m
->idle_pipe
[2], EPOLLIN
, manager_dispatch_idle_pipe_fd
, m
);
347 return log_error_errno(r
, "Failed to watch idle pipe: %m");
349 (void) sd_event_source_set_description(m
->idle_pipe_event_source
, "manager-idle-pipe");
354 static void manager_close_idle_pipe(Manager
*m
) {
357 m
->idle_pipe_event_source
= sd_event_source_unref(m
->idle_pipe_event_source
);
359 safe_close_pair(m
->idle_pipe
);
360 safe_close_pair(m
->idle_pipe
+ 2);
363 static int manager_setup_time_change(Manager
*m
) {
368 if (MANAGER_IS_TEST_RUN(m
))
371 m
->time_change_event_source
= sd_event_source_unref(m
->time_change_event_source
);
372 m
->time_change_fd
= safe_close(m
->time_change_fd
);
374 m
->time_change_fd
= time_change_fd();
375 if (m
->time_change_fd
< 0)
376 return log_error_errno(m
->time_change_fd
, "Failed to create timer change timer fd: %m");
378 r
= sd_event_add_io(m
->event
, &m
->time_change_event_source
, m
->time_change_fd
, EPOLLIN
, manager_dispatch_time_change_fd
, m
);
380 return log_error_errno(r
, "Failed to create time change event source: %m");
382 /* Schedule this slightly earlier than the .timer event sources */
383 r
= sd_event_source_set_priority(m
->time_change_event_source
, SD_EVENT_PRIORITY_NORMAL
-1);
385 return log_error_errno(r
, "Failed to set priority of time change event sources: %m");
387 (void) sd_event_source_set_description(m
->time_change_event_source
, "manager-time-change");
389 log_debug("Set up TFD_TIMER_CANCEL_ON_SET timerfd.");
394 static int manager_read_timezone_stat(Manager
*m
) {
400 /* Read the current stat() data of /etc/localtime so that we detect changes */
401 if (lstat("/etc/localtime", &st
) < 0) {
402 log_debug_errno(errno
, "Failed to stat /etc/localtime, ignoring: %m");
403 changed
= m
->etc_localtime_accessible
;
404 m
->etc_localtime_accessible
= false;
408 k
= timespec_load(&st
.st_mtim
);
409 changed
= !m
->etc_localtime_accessible
|| k
!= m
->etc_localtime_mtime
;
411 m
->etc_localtime_mtime
= k
;
412 m
->etc_localtime_accessible
= true;
418 static int manager_setup_timezone_change(Manager
*m
) {
419 _cleanup_(sd_event_source_unrefp
) sd_event_source
*new_event
= NULL
;
424 if (MANAGER_IS_TEST_RUN(m
))
427 /* We watch /etc/localtime for three events: change of the link count (which might mean removal from /etc even
428 * though another link might be kept), renames, and file close operations after writing. Note we don't bother
429 * with IN_DELETE_SELF, as that would just report when the inode is removed entirely, i.e. after the link count
430 * went to zero and all fds to it are closed.
432 * Note that we never follow symlinks here. This is a simplification, but should cover almost all cases
435 * Note that we create the new event source first here, before releasing the old one. This should optimize
436 * behaviour as this way sd-event can reuse the old watch in case the inode didn't change. */
438 r
= sd_event_add_inotify(m
->event
, &new_event
, "/etc/localtime",
439 IN_ATTRIB
|IN_MOVE_SELF
|IN_CLOSE_WRITE
|IN_DONT_FOLLOW
, manager_dispatch_timezone_change
, m
);
441 /* If the file doesn't exist yet, subscribe to /etc instead, and wait until it is created either by
442 * O_CREATE or by rename() */
444 log_debug_errno(r
, "/etc/localtime doesn't exist yet, watching /etc instead.");
445 r
= sd_event_add_inotify(m
->event
, &new_event
, "/etc",
446 IN_CREATE
|IN_MOVED_TO
|IN_ONLYDIR
, manager_dispatch_timezone_change
, m
);
449 return log_error_errno(r
, "Failed to create timezone change event source: %m");
451 /* Schedule this slightly earlier than the .timer event sources */
452 r
= sd_event_source_set_priority(new_event
, SD_EVENT_PRIORITY_NORMAL
-1);
454 return log_error_errno(r
, "Failed to set priority of timezone change event sources: %m");
456 sd_event_source_unref(m
->timezone_change_event_source
);
457 m
->timezone_change_event_source
= TAKE_PTR(new_event
);
462 static int enable_special_signals(Manager
*m
) {
463 _cleanup_close_
int fd
= -1;
467 if (MANAGER_IS_TEST_RUN(m
))
470 /* Enable that we get SIGINT on control-alt-del. In containers
471 * this will fail with EPERM (older) or EINVAL (newer), so
473 if (reboot(RB_DISABLE_CAD
) < 0 && !IN_SET(errno
, EPERM
, EINVAL
))
474 log_warning_errno(errno
, "Failed to enable ctrl-alt-del handling: %m");
476 fd
= open_terminal("/dev/tty0", O_RDWR
|O_NOCTTY
|O_CLOEXEC
);
478 /* Support systems without virtual console */
480 log_warning_errno(errno
, "Failed to open /dev/tty0: %m");
482 /* Enable that we get SIGWINCH on kbrequest */
483 if (ioctl(fd
, KDSIGACCEPT
, SIGWINCH
) < 0)
484 log_warning_errno(errno
, "Failed to enable kbrequest handling: %m");
490 #define RTSIG_IF_AVAILABLE(signum) (signum <= SIGRTMAX ? signum : -1)
492 static int manager_setup_signals(Manager
*m
) {
493 struct sigaction sa
= {
494 .sa_handler
= SIG_DFL
,
495 .sa_flags
= SA_NOCLDSTOP
|SA_RESTART
,
502 assert_se(sigaction(SIGCHLD
, &sa
, NULL
) == 0);
504 /* We make liberal use of realtime signals here. On
505 * Linux/glibc we have 30 of them (with the exception of Linux
506 * on hppa, see below), between SIGRTMIN+0 ... SIGRTMIN+30
509 assert_se(sigemptyset(&mask
) == 0);
510 sigset_add_many(&mask
,
511 SIGCHLD
, /* Child died */
512 SIGTERM
, /* Reexecute daemon */
513 SIGHUP
, /* Reload configuration */
514 SIGUSR1
, /* systemd/upstart: reconnect to D-Bus */
515 SIGUSR2
, /* systemd: dump status */
516 SIGINT
, /* Kernel sends us this on control-alt-del */
517 SIGWINCH
, /* Kernel sends us this on kbrequest (alt-arrowup) */
518 SIGPWR
, /* Some kernel drivers and upsd send us this on power failure */
520 SIGRTMIN
+0, /* systemd: start default.target */
521 SIGRTMIN
+1, /* systemd: isolate rescue.target */
522 SIGRTMIN
+2, /* systemd: isolate emergency.target */
523 SIGRTMIN
+3, /* systemd: start halt.target */
524 SIGRTMIN
+4, /* systemd: start poweroff.target */
525 SIGRTMIN
+5, /* systemd: start reboot.target */
526 SIGRTMIN
+6, /* systemd: start kexec.target */
528 /* ... space for more special targets ... */
530 SIGRTMIN
+13, /* systemd: Immediate halt */
531 SIGRTMIN
+14, /* systemd: Immediate poweroff */
532 SIGRTMIN
+15, /* systemd: Immediate reboot */
533 SIGRTMIN
+16, /* systemd: Immediate kexec */
535 /* ... space for more immediate system state changes ... */
537 SIGRTMIN
+20, /* systemd: enable status messages */
538 SIGRTMIN
+21, /* systemd: disable status messages */
539 SIGRTMIN
+22, /* systemd: set log level to LOG_DEBUG */
540 SIGRTMIN
+23, /* systemd: set log level to LOG_INFO */
541 SIGRTMIN
+24, /* systemd: Immediate exit (--user only) */
543 /* .. one free signal here ... */
545 /* Apparently Linux on hppa had fewer RT signals until v3.18,
546 * SIGRTMAX was SIGRTMIN+25, and then SIGRTMIN was lowered,
547 * see commit v3.17-7614-g1f25df2eff.
549 * We cannot unconditionally make use of those signals here,
550 * so let's use a runtime check. Since these commands are
551 * accessible by different means and only really a safety
552 * net, the missing functionality on hppa shouldn't matter.
555 RTSIG_IF_AVAILABLE(SIGRTMIN
+26), /* systemd: set log target to journal-or-kmsg */
556 RTSIG_IF_AVAILABLE(SIGRTMIN
+27), /* systemd: set log target to console */
557 RTSIG_IF_AVAILABLE(SIGRTMIN
+28), /* systemd: set log target to kmsg */
558 RTSIG_IF_AVAILABLE(SIGRTMIN
+29), /* systemd: set log target to syslog-or-kmsg (obsolete) */
560 /* ... one free signal here SIGRTMIN+30 ... */
562 assert_se(sigprocmask(SIG_SETMASK
, &mask
, NULL
) == 0);
564 m
->signal_fd
= signalfd(-1, &mask
, SFD_NONBLOCK
|SFD_CLOEXEC
);
565 if (m
->signal_fd
< 0)
568 r
= sd_event_add_io(m
->event
, &m
->signal_event_source
, m
->signal_fd
, EPOLLIN
, manager_dispatch_signal_fd
, m
);
572 (void) sd_event_source_set_description(m
->signal_event_source
, "manager-signal");
574 /* Process signals a bit earlier than the rest of things, but later than notify_fd processing, so that the
575 * notify processing can still figure out to which process/service a message belongs, before we reap the
576 * process. Also, process this before handling cgroup notifications, so that we always collect child exit
577 * status information before detecting that there's no process in a cgroup. */
578 r
= sd_event_source_set_priority(m
->signal_event_source
, SD_EVENT_PRIORITY_NORMAL
-6);
582 if (MANAGER_IS_SYSTEM(m
))
583 return enable_special_signals(m
);
588 static char** sanitize_environment(char **l
) {
590 /* Let's remove some environment variables that we need ourselves to communicate with our clients */
594 "CONFIGURATION_DIRECTORY",
595 "CREDENTIALS_DIRECTORY",
617 /* Let's order the environment alphabetically, just to make it pretty */
623 int manager_default_environment(Manager
*m
) {
628 m
->transient_environment
= strv_free(m
->transient_environment
);
630 if (MANAGER_IS_SYSTEM(m
)) {
631 /* The system manager always starts with a clean
632 * environment for its children. It does not import
633 * the kernel's or the parents' exported variables.
635 * The initial passed environment is untouched to keep
636 * /proc/self/environ valid; it is used for tagging
637 * the init process inside containers. */
638 m
->transient_environment
= strv_new("PATH=" DEFAULT_PATH
);
639 if (!m
->transient_environment
)
642 /* Import locale variables LC_*= from configuration */
643 (void) locale_setup(&m
->transient_environment
);
645 _cleanup_free_
char *k
= NULL
;
647 /* The user manager passes its own environment
648 * along to its children, except for $PATH. */
649 m
->transient_environment
= strv_copy(environ
);
650 if (!m
->transient_environment
)
653 k
= strdup("PATH=" DEFAULT_USER_PATH
);
657 r
= strv_env_replace(&m
->transient_environment
, k
);
663 sanitize_environment(m
->transient_environment
);
668 static int manager_setup_prefix(Manager
*m
) {
674 static const struct table_entry paths_system
[_EXEC_DIRECTORY_TYPE_MAX
] = {
675 [EXEC_DIRECTORY_RUNTIME
] = { SD_PATH_SYSTEM_RUNTIME
, NULL
},
676 [EXEC_DIRECTORY_STATE
] = { SD_PATH_SYSTEM_STATE_PRIVATE
, NULL
},
677 [EXEC_DIRECTORY_CACHE
] = { SD_PATH_SYSTEM_STATE_CACHE
, NULL
},
678 [EXEC_DIRECTORY_LOGS
] = { SD_PATH_SYSTEM_STATE_LOGS
, NULL
},
679 [EXEC_DIRECTORY_CONFIGURATION
] = { SD_PATH_SYSTEM_CONFIGURATION
, NULL
},
682 static const struct table_entry paths_user
[_EXEC_DIRECTORY_TYPE_MAX
] = {
683 [EXEC_DIRECTORY_RUNTIME
] = { SD_PATH_USER_RUNTIME
, NULL
},
684 [EXEC_DIRECTORY_STATE
] = { SD_PATH_USER_CONFIGURATION
, NULL
},
685 [EXEC_DIRECTORY_CACHE
] = { SD_PATH_USER_STATE_CACHE
, NULL
},
686 [EXEC_DIRECTORY_LOGS
] = { SD_PATH_USER_CONFIGURATION
, "log" },
687 [EXEC_DIRECTORY_CONFIGURATION
] = { SD_PATH_USER_CONFIGURATION
, NULL
},
692 const struct table_entry
*p
= MANAGER_IS_SYSTEM(m
) ? paths_system
: paths_user
;
695 for (ExecDirectoryType i
= 0; i
< _EXEC_DIRECTORY_TYPE_MAX
; i
++) {
696 r
= sd_path_lookup(p
[i
].type
, p
[i
].suffix
, &m
->prefix
[i
]);
704 static void manager_free_unit_name_maps(Manager
*m
) {
705 m
->unit_id_map
= hashmap_free(m
->unit_id_map
);
706 m
->unit_name_map
= hashmap_free(m
->unit_name_map
);
707 m
->unit_path_cache
= set_free(m
->unit_path_cache
);
708 m
->unit_cache_timestamp_hash
= 0;
711 static int manager_setup_run_queue(Manager
*m
) {
715 assert(!m
->run_queue_event_source
);
717 r
= sd_event_add_defer(m
->event
, &m
->run_queue_event_source
, manager_dispatch_run_queue
, m
);
721 r
= sd_event_source_set_priority(m
->run_queue_event_source
, SD_EVENT_PRIORITY_IDLE
);
725 r
= sd_event_source_set_enabled(m
->run_queue_event_source
, SD_EVENT_OFF
);
729 (void) sd_event_source_set_description(m
->run_queue_event_source
, "manager-run-queue");
734 static int manager_setup_sigchld_event_source(Manager
*m
) {
738 assert(!m
->sigchld_event_source
);
740 r
= sd_event_add_defer(m
->event
, &m
->sigchld_event_source
, manager_dispatch_sigchld
, m
);
744 r
= sd_event_source_set_priority(m
->sigchld_event_source
, SD_EVENT_PRIORITY_NORMAL
-7);
748 r
= sd_event_source_set_enabled(m
->sigchld_event_source
, SD_EVENT_OFF
);
752 (void) sd_event_source_set_description(m
->sigchld_event_source
, "manager-sigchld");
757 int manager_new(UnitFileScope scope
, ManagerTestRunFlags test_run_flags
, Manager
**_m
) {
758 _cleanup_(manager_freep
) Manager
*m
= NULL
;
763 assert(IN_SET(scope
, UNIT_FILE_SYSTEM
, UNIT_FILE_USER
));
770 .unit_file_scope
= scope
,
771 .objective
= _MANAGER_OBJECTIVE_INVALID
,
773 .status_unit_format
= STATUS_UNIT_FORMAT_DEFAULT
,
775 .default_timer_accuracy_usec
= USEC_PER_MINUTE
,
776 .default_memory_accounting
= MEMORY_ACCOUNTING_DEFAULT
,
777 .default_tasks_accounting
= true,
778 .default_tasks_max
= TASKS_MAX_UNSET
,
779 .default_timeout_start_usec
= DEFAULT_TIMEOUT_USEC
,
780 .default_timeout_stop_usec
= DEFAULT_TIMEOUT_USEC
,
781 .default_restart_usec
= DEFAULT_RESTART_USEC
,
783 .original_log_level
= -1,
784 .original_log_target
= _LOG_TARGET_INVALID
,
786 .watchdog_overridden
[WATCHDOG_RUNTIME
] = USEC_INFINITY
,
787 .watchdog_overridden
[WATCHDOG_REBOOT
] = USEC_INFINITY
,
788 .watchdog_overridden
[WATCHDOG_KEXEC
] = USEC_INFINITY
,
790 .show_status_overridden
= _SHOW_STATUS_INVALID
,
793 .cgroups_agent_fd
= -1,
795 .time_change_fd
= -1,
796 .user_lookup_fds
= { -1, -1 },
797 .private_listen_fd
= -1,
799 .cgroup_inotify_fd
= -1,
800 .pin_cgroupfs_fd
= -1,
801 .ask_password_inotify_fd
= -1,
802 .idle_pipe
= { -1, -1, -1, -1},
804 /* start as id #1, so that we can leave #0 around as "null-like" value */
807 .have_ask_password
= -EINVAL
, /* we don't know */
809 .test_run_flags
= test_run_flags
,
811 .default_oom_policy
= OOM_STOP
,
815 if (MANAGER_IS_SYSTEM(m
) && detect_container() <= 0)
816 boot_timestamps(m
->timestamps
+ MANAGER_TIMESTAMP_USERSPACE
,
817 m
->timestamps
+ MANAGER_TIMESTAMP_FIRMWARE
,
818 m
->timestamps
+ MANAGER_TIMESTAMP_LOADER
);
821 /* Prepare log fields we can use for structured logging */
822 if (MANAGER_IS_SYSTEM(m
)) {
823 m
->unit_log_field
= "UNIT=";
824 m
->unit_log_format_string
= "UNIT=%s";
826 m
->invocation_log_field
= "INVOCATION_ID=";
827 m
->invocation_log_format_string
= "INVOCATION_ID=%s";
829 m
->unit_log_field
= "USER_UNIT=";
830 m
->unit_log_format_string
= "USER_UNIT=%s";
832 m
->invocation_log_field
= "USER_INVOCATION_ID=";
833 m
->invocation_log_format_string
= "USER_INVOCATION_ID=%s";
836 /* Reboot immediately if the user hits C-A-D more often than 7x per 2s */
837 m
->ctrl_alt_del_ratelimit
= (RateLimit
) { .interval
= 2 * USEC_PER_SEC
, .burst
= 7 };
839 r
= manager_default_environment(m
);
843 r
= hashmap_ensure_allocated(&m
->units
, &string_hash_ops
);
847 r
= hashmap_ensure_allocated(&m
->cgroup_unit
, &path_hash_ops
);
851 r
= hashmap_ensure_allocated(&m
->watch_bus
, &string_hash_ops
);
855 r
= prioq_ensure_allocated(&m
->run_queue
, compare_job_priority
);
859 r
= manager_setup_prefix(m
);
863 e
= secure_getenv("CREDENTIALS_DIRECTORY");
865 m
->received_credentials
= strdup(e
);
866 if (!m
->received_credentials
)
870 r
= sd_event_default(&m
->event
);
874 r
= manager_setup_run_queue(m
);
878 if (test_run_flags
== MANAGER_TEST_RUN_MINIMAL
) {
879 m
->cgroup_root
= strdup("");
883 r
= manager_setup_signals(m
);
887 r
= manager_setup_cgroup(m
);
891 r
= manager_setup_time_change(m
);
895 r
= manager_read_timezone_stat(m
);
899 (void) manager_setup_timezone_change(m
);
901 r
= manager_setup_sigchld_event_source(m
);
906 if (test_run_flags
== 0) {
907 if (MANAGER_IS_SYSTEM(m
))
908 r
= mkdir_label("/run/systemd/units", 0755);
910 _cleanup_free_
char *units_path
= NULL
;
911 r
= xdg_user_runtime_dir(&units_path
, "/systemd/units");
914 r
= mkdir_p_label(units_path
, 0755);
917 if (r
< 0 && r
!= -EEXIST
)
923 dir_is_empty("/usr") > 0;
925 /* Note that we do not set up the notify fd here. We do that after deserialization,
926 * since they might have gotten serialized across the reexec. */
933 static int manager_setup_notify(Manager
*m
) {
936 if (MANAGER_IS_TEST_RUN(m
))
939 if (m
->notify_fd
< 0) {
940 _cleanup_close_
int fd
= -1;
941 union sockaddr_union sa
;
944 /* First free all secondary fields */
945 m
->notify_socket
= mfree(m
->notify_socket
);
946 m
->notify_event_source
= sd_event_source_unref(m
->notify_event_source
);
948 fd
= socket(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
|SOCK_NONBLOCK
, 0);
950 return log_error_errno(errno
, "Failed to allocate notification socket: %m");
952 fd_inc_rcvbuf(fd
, NOTIFY_RCVBUF_SIZE
);
954 m
->notify_socket
= path_join(m
->prefix
[EXEC_DIRECTORY_RUNTIME
], "systemd/notify");
955 if (!m
->notify_socket
)
958 r
= sockaddr_un_set_path(&sa
.un
, m
->notify_socket
);
960 return log_error_errno(r
, "Notify socket '%s' not valid for AF_UNIX socket address, refusing.",
964 (void) mkdir_parents_label(m
->notify_socket
, 0755);
965 (void) sockaddr_un_unlink(&sa
.un
);
967 r
= mac_selinux_bind(fd
, &sa
.sa
, sa_len
);
969 return log_error_errno(r
, "bind(%s) failed: %m", m
->notify_socket
);
971 r
= setsockopt_int(fd
, SOL_SOCKET
, SO_PASSCRED
, true);
973 return log_error_errno(r
, "SO_PASSCRED failed: %m");
975 m
->notify_fd
= TAKE_FD(fd
);
977 log_debug("Using notification socket %s", m
->notify_socket
);
980 if (!m
->notify_event_source
) {
981 r
= sd_event_add_io(m
->event
, &m
->notify_event_source
, m
->notify_fd
, EPOLLIN
, manager_dispatch_notify_fd
, m
);
983 return log_error_errno(r
, "Failed to allocate notify event source: %m");
985 /* Process notification messages a bit earlier than SIGCHLD, so that we can still identify to which
986 * service an exit message belongs. */
987 r
= sd_event_source_set_priority(m
->notify_event_source
, SD_EVENT_PRIORITY_NORMAL
-8);
989 return log_error_errno(r
, "Failed to set priority of notify event source: %m");
991 (void) sd_event_source_set_description(m
->notify_event_source
, "manager-notify");
997 static int manager_setup_cgroups_agent(Manager
*m
) {
999 static const union sockaddr_union sa
= {
1000 .un
.sun_family
= AF_UNIX
,
1001 .un
.sun_path
= "/run/systemd/cgroups-agent",
1005 /* This creates a listening socket we receive cgroups agent messages on. We do not use D-Bus for delivering
1006 * these messages from the cgroups agent binary to PID 1, as the cgroups agent binary is very short-living, and
1007 * each instance of it needs a new D-Bus connection. Since D-Bus connections are SOCK_STREAM/AF_UNIX, on
1008 * overloaded systems the backlog of the D-Bus socket becomes relevant, as not more than the configured number
1009 * of D-Bus connections may be queued until the kernel will start dropping further incoming connections,
1010 * possibly resulting in lost cgroups agent messages. To avoid this, we'll use a private SOCK_DGRAM/AF_UNIX
1011 * socket, where no backlog is relevant as communication may take place without an actual connect() cycle, and
1012 * we thus won't lose messages.
1014 * Note that PID 1 will forward the agent message to system bus, so that the user systemd instance may listen
1015 * to it. The system instance hence listens on this special socket, but the user instances listen on the system
1016 * bus for these messages. */
1018 if (MANAGER_IS_TEST_RUN(m
))
1021 if (!MANAGER_IS_SYSTEM(m
))
1024 r
= cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER
);
1026 return log_error_errno(r
, "Failed to determine whether unified cgroups hierarchy is used: %m");
1027 if (r
> 0) /* We don't need this anymore on the unified hierarchy */
1030 if (m
->cgroups_agent_fd
< 0) {
1031 _cleanup_close_
int fd
= -1;
1033 /* First free all secondary fields */
1034 m
->cgroups_agent_event_source
= sd_event_source_unref(m
->cgroups_agent_event_source
);
1036 fd
= socket(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
|SOCK_NONBLOCK
, 0);
1038 return log_error_errno(errno
, "Failed to allocate cgroups agent socket: %m");
1040 fd_inc_rcvbuf(fd
, CGROUPS_AGENT_RCVBUF_SIZE
);
1042 (void) sockaddr_un_unlink(&sa
.un
);
1044 /* Only allow root to connect to this socket */
1045 RUN_WITH_UMASK(0077)
1046 r
= bind(fd
, &sa
.sa
, SOCKADDR_UN_LEN(sa
.un
));
1048 return log_error_errno(errno
, "bind(%s) failed: %m", sa
.un
.sun_path
);
1050 m
->cgroups_agent_fd
= TAKE_FD(fd
);
1053 if (!m
->cgroups_agent_event_source
) {
1054 r
= sd_event_add_io(m
->event
, &m
->cgroups_agent_event_source
, m
->cgroups_agent_fd
, EPOLLIN
, manager_dispatch_cgroups_agent_fd
, m
);
1056 return log_error_errno(r
, "Failed to allocate cgroups agent event source: %m");
1058 /* Process cgroups notifications early. Note that when the agent notification is received
1059 * we'll just enqueue the unit in the cgroup empty queue, hence pick a high priority than
1060 * that. Also see handling of cgroup inotify for the unified cgroup stuff. */
1061 r
= sd_event_source_set_priority(m
->cgroups_agent_event_source
, SD_EVENT_PRIORITY_NORMAL
-9);
1063 return log_error_errno(r
, "Failed to set priority of cgroups agent event source: %m");
1065 (void) sd_event_source_set_description(m
->cgroups_agent_event_source
, "manager-cgroups-agent");
1071 static int manager_setup_user_lookup_fd(Manager
*m
) {
1076 /* Set up the socket pair used for passing UID/GID resolution results from forked off processes to PID
1077 * 1. Background: we can't do name lookups (NSS) from PID 1, since it might involve IPC and thus activation,
1078 * and we might hence deadlock on ourselves. Hence we do all user/group lookups asynchronously from the forked
1079 * off processes right before executing the binaries to start. In order to be able to clean up any IPC objects
1080 * created by a unit (see RemoveIPC=) we need to know in PID 1 the used UID/GID of the executed processes,
1081 * hence we establish this communication channel so that forked off processes can pass their UID/GID
1082 * information back to PID 1. The forked off processes send their resolved UID/GID to PID 1 in a simple
1083 * datagram, along with their unit name, so that we can share one communication socket pair among all units for
1086 * You might wonder why we need a communication channel for this that is independent of the usual notification
1087 * socket scheme (i.e. $NOTIFY_SOCKET). The primary difference is about trust: data sent via the $NOTIFY_SOCKET
1088 * channel is only accepted if it originates from the right unit and if reception was enabled for it. The user
1089 * lookup socket OTOH is only accessible by PID 1 and its children until they exec(), and always available.
1091 * Note that this function is called under two circumstances: when we first initialize (in which case we
1092 * allocate both the socket pair and the event source to listen on it), and when we deserialize after a reload
1093 * (in which case the socket pair already exists but we still need to allocate the event source for it). */
1095 if (m
->user_lookup_fds
[0] < 0) {
1097 /* Free all secondary fields */
1098 safe_close_pair(m
->user_lookup_fds
);
1099 m
->user_lookup_event_source
= sd_event_source_unref(m
->user_lookup_event_source
);
1101 if (socketpair(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
, 0, m
->user_lookup_fds
) < 0)
1102 return log_error_errno(errno
, "Failed to allocate user lookup socket: %m");
1104 (void) fd_inc_rcvbuf(m
->user_lookup_fds
[0], NOTIFY_RCVBUF_SIZE
);
1107 if (!m
->user_lookup_event_source
) {
1108 r
= sd_event_add_io(m
->event
, &m
->user_lookup_event_source
, m
->user_lookup_fds
[0], EPOLLIN
, manager_dispatch_user_lookup_fd
, m
);
1110 return log_error_errno(errno
, "Failed to allocate user lookup event source: %m");
1112 /* Process even earlier than the notify event source, so that we always know first about valid UID/GID
1114 r
= sd_event_source_set_priority(m
->user_lookup_event_source
, SD_EVENT_PRIORITY_NORMAL
-11);
1116 return log_error_errno(errno
, "Failed to set priority of user lookup event source: %m");
1118 (void) sd_event_source_set_description(m
->user_lookup_event_source
, "user-lookup");
1124 static unsigned manager_dispatch_cleanup_queue(Manager
*m
) {
1130 while ((u
= m
->cleanup_queue
)) {
1131 assert(u
->in_cleanup_queue
);
1141 GC_OFFSET_IN_PATH
, /* This one is on the path we were traveling */
1142 GC_OFFSET_UNSURE
, /* No clue */
1143 GC_OFFSET_GOOD
, /* We still need this unit */
1144 GC_OFFSET_BAD
, /* We don't need this unit anymore */
1148 static void unit_gc_mark_good(Unit
*u
, unsigned gc_marker
) {
1153 u
->gc_marker
= gc_marker
+ GC_OFFSET_GOOD
;
1155 /* Recursively mark referenced units as GOOD as well */
1156 HASHMAP_FOREACH_KEY(v
, other
, u
->dependencies
[UNIT_REFERENCES
], i
)
1157 if (other
->gc_marker
== gc_marker
+ GC_OFFSET_UNSURE
)
1158 unit_gc_mark_good(other
, gc_marker
);
1161 static void unit_gc_sweep(Unit
*u
, unsigned gc_marker
) {
1169 if (IN_SET(u
->gc_marker
- gc_marker
,
1170 GC_OFFSET_GOOD
, GC_OFFSET_BAD
, GC_OFFSET_UNSURE
, GC_OFFSET_IN_PATH
))
1173 if (u
->in_cleanup_queue
)
1176 if (!unit_may_gc(u
))
1179 u
->gc_marker
= gc_marker
+ GC_OFFSET_IN_PATH
;
1183 HASHMAP_FOREACH_KEY(v
, other
, u
->dependencies
[UNIT_REFERENCED_BY
], i
) {
1184 unit_gc_sweep(other
, gc_marker
);
1186 if (other
->gc_marker
== gc_marker
+ GC_OFFSET_GOOD
)
1189 if (other
->gc_marker
!= gc_marker
+ GC_OFFSET_BAD
)
1193 if (u
->refs_by_target
) {
1196 LIST_FOREACH(refs_by_target
, ref
, u
->refs_by_target
) {
1197 unit_gc_sweep(ref
->source
, gc_marker
);
1199 if (ref
->source
->gc_marker
== gc_marker
+ GC_OFFSET_GOOD
)
1202 if (ref
->source
->gc_marker
!= gc_marker
+ GC_OFFSET_BAD
)
1210 /* We were unable to find anything out about this entry, so
1211 * let's investigate it later */
1212 u
->gc_marker
= gc_marker
+ GC_OFFSET_UNSURE
;
1213 unit_add_to_gc_queue(u
);
1217 /* We definitely know that this one is not useful anymore, so
1218 * let's mark it for deletion */
1219 u
->gc_marker
= gc_marker
+ GC_OFFSET_BAD
;
1220 unit_add_to_cleanup_queue(u
);
1224 unit_gc_mark_good(u
, gc_marker
);
1227 static unsigned manager_dispatch_gc_unit_queue(Manager
*m
) {
1228 unsigned n
= 0, gc_marker
;
1233 /* log_debug("Running GC..."); */
1235 m
->gc_marker
+= _GC_OFFSET_MAX
;
1236 if (m
->gc_marker
+ _GC_OFFSET_MAX
<= _GC_OFFSET_MAX
)
1239 gc_marker
= m
->gc_marker
;
1241 while ((u
= m
->gc_unit_queue
)) {
1242 assert(u
->in_gc_queue
);
1244 unit_gc_sweep(u
, gc_marker
);
1246 LIST_REMOVE(gc_queue
, m
->gc_unit_queue
, u
);
1247 u
->in_gc_queue
= false;
1251 if (IN_SET(u
->gc_marker
- gc_marker
,
1252 GC_OFFSET_BAD
, GC_OFFSET_UNSURE
)) {
1254 log_unit_debug(u
, "Collecting.");
1255 u
->gc_marker
= gc_marker
+ GC_OFFSET_BAD
;
1256 unit_add_to_cleanup_queue(u
);
1263 static unsigned manager_dispatch_gc_job_queue(Manager
*m
) {
1269 while ((j
= m
->gc_job_queue
)) {
1270 assert(j
->in_gc_queue
);
1272 LIST_REMOVE(gc_queue
, m
->gc_job_queue
, j
);
1273 j
->in_gc_queue
= false;
1280 log_unit_debug(j
->unit
, "Collecting job.");
1281 (void) job_finish_and_invalidate(j
, JOB_COLLECTED
, false, false);
1287 static unsigned manager_dispatch_stop_when_unneeded_queue(Manager
*m
) {
1294 while ((u
= m
->stop_when_unneeded_queue
)) {
1295 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
1296 assert(m
->stop_when_unneeded_queue
);
1298 assert(u
->in_stop_when_unneeded_queue
);
1299 LIST_REMOVE(stop_when_unneeded_queue
, m
->stop_when_unneeded_queue
, u
);
1300 u
->in_stop_when_unneeded_queue
= false;
1304 if (!unit_is_unneeded(u
))
1307 log_unit_debug(u
, "Unit is not needed anymore.");
1309 /* If stopping a unit fails continuously we might enter a stop loop here, hence stop acting on the
1310 * service being unnecessary after a while. */
1312 if (!ratelimit_below(&u
->auto_stop_ratelimit
)) {
1313 log_unit_warning(u
, "Unit not needed anymore, but not stopping since we tried this too often recently.");
1317 /* Ok, nobody needs us anymore. Sniff. Then let's commit suicide */
1318 r
= manager_add_job(u
->manager
, JOB_STOP
, u
, JOB_FAIL
, NULL
, &error
, NULL
);
1320 log_unit_warning_errno(u
, r
, "Failed to enqueue stop job, ignoring: %s", bus_error_message(&error
, r
));
1326 static void manager_clear_jobs_and_units(Manager
*m
) {
1331 while ((u
= hashmap_first(m
->units
)))
1334 manager_dispatch_cleanup_queue(m
);
1336 assert(!m
->load_queue
);
1337 assert(prioq_isempty(m
->run_queue
));
1338 assert(!m
->dbus_unit_queue
);
1339 assert(!m
->dbus_job_queue
);
1340 assert(!m
->cleanup_queue
);
1341 assert(!m
->gc_unit_queue
);
1342 assert(!m
->gc_job_queue
);
1343 assert(!m
->stop_when_unneeded_queue
);
1345 assert(hashmap_isempty(m
->jobs
));
1346 assert(hashmap_isempty(m
->units
));
1348 m
->n_on_console
= 0;
1349 m
->n_running_jobs
= 0;
1350 m
->n_installed_jobs
= 0;
1351 m
->n_failed_jobs
= 0;
1354 Manager
* manager_free(Manager
*m
) {
1358 manager_clear_jobs_and_units(m
);
1360 for (UnitType c
= 0; c
< _UNIT_TYPE_MAX
; c
++)
1361 if (unit_vtable
[c
]->shutdown
)
1362 unit_vtable
[c
]->shutdown(m
);
1364 /* Keep the cgroup hierarchy in place except when we know we are going down for good */
1365 manager_shutdown_cgroup(m
, IN_SET(m
->objective
, MANAGER_EXIT
, MANAGER_REBOOT
, MANAGER_POWEROFF
, MANAGER_HALT
, MANAGER_KEXEC
));
1367 lookup_paths_flush_generator(&m
->lookup_paths
);
1370 manager_varlink_done(m
);
1372 exec_runtime_vacuum(m
);
1373 hashmap_free(m
->exec_runtime_by_id
);
1375 dynamic_user_vacuum(m
, false);
1376 hashmap_free(m
->dynamic_users
);
1378 hashmap_free(m
->units
);
1379 hashmap_free(m
->units_by_invocation_id
);
1380 hashmap_free(m
->jobs
);
1381 hashmap_free(m
->watch_pids
);
1382 hashmap_free(m
->watch_bus
);
1384 prioq_free(m
->run_queue
);
1386 set_free(m
->startup_units
);
1387 set_free(m
->failed_units
);
1389 sd_event_source_unref(m
->signal_event_source
);
1390 sd_event_source_unref(m
->sigchld_event_source
);
1391 sd_event_source_unref(m
->notify_event_source
);
1392 sd_event_source_unref(m
->cgroups_agent_event_source
);
1393 sd_event_source_unref(m
->time_change_event_source
);
1394 sd_event_source_unref(m
->timezone_change_event_source
);
1395 sd_event_source_unref(m
->jobs_in_progress_event_source
);
1396 sd_event_source_unref(m
->run_queue_event_source
);
1397 sd_event_source_unref(m
->user_lookup_event_source
);
1399 safe_close(m
->signal_fd
);
1400 safe_close(m
->notify_fd
);
1401 safe_close(m
->cgroups_agent_fd
);
1402 safe_close(m
->time_change_fd
);
1403 safe_close_pair(m
->user_lookup_fds
);
1405 manager_close_ask_password(m
);
1407 manager_close_idle_pipe(m
);
1409 sd_event_unref(m
->event
);
1411 free(m
->notify_socket
);
1413 lookup_paths_free(&m
->lookup_paths
);
1414 strv_free(m
->transient_environment
);
1415 strv_free(m
->client_environment
);
1417 hashmap_free(m
->cgroup_unit
);
1418 manager_free_unit_name_maps(m
);
1420 free(m
->switch_root
);
1421 free(m
->switch_root_init
);
1423 rlimit_free_all(m
->rlimit
);
1425 assert(hashmap_isempty(m
->units_requiring_mounts_for
));
1426 hashmap_free(m
->units_requiring_mounts_for
);
1428 hashmap_free(m
->uid_refs
);
1429 hashmap_free(m
->gid_refs
);
1431 for (ExecDirectoryType dt
= 0; dt
< _EXEC_DIRECTORY_TYPE_MAX
; dt
++)
1432 m
->prefix
[dt
] = mfree(m
->prefix
[dt
]);
1433 free(m
->received_credentials
);
1438 static void manager_enumerate_perpetual(Manager
*m
) {
1441 if (m
->test_run_flags
== MANAGER_TEST_RUN_MINIMAL
)
1444 /* Let's ask every type to load all units from disk/kernel that it might know */
1445 for (UnitType c
= 0; c
< _UNIT_TYPE_MAX
; c
++) {
1446 if (!unit_type_supported(c
)) {
1447 log_debug("Unit type .%s is not supported on this system.", unit_type_to_string(c
));
1451 if (unit_vtable
[c
]->enumerate_perpetual
)
1452 unit_vtable
[c
]->enumerate_perpetual(m
);
1456 static void manager_enumerate(Manager
*m
) {
1459 if (m
->test_run_flags
== MANAGER_TEST_RUN_MINIMAL
)
1462 /* Let's ask every type to load all units from disk/kernel that it might know */
1463 for (UnitType c
= 0; c
< _UNIT_TYPE_MAX
; c
++) {
1464 if (!unit_type_supported(c
)) {
1465 log_debug("Unit type .%s is not supported on this system.", unit_type_to_string(c
));
1469 if (unit_vtable
[c
]->enumerate
)
1470 unit_vtable
[c
]->enumerate(m
);
1473 manager_dispatch_load_queue(m
);
1476 static void manager_coldplug(Manager
*m
) {
1484 log_debug("Invoking unit coldplug() handlers…");
1486 /* Let's place the units back into their deserialized state */
1487 HASHMAP_FOREACH_KEY(u
, k
, m
->units
, i
) {
1489 /* ignore aliases */
1493 r
= unit_coldplug(u
);
1495 log_warning_errno(r
, "We couldn't coldplug %s, proceeding anyway: %m", u
->id
);
1499 static void manager_catchup(Manager
*m
) {
1506 log_debug("Invoking unit catchup() handlers…");
1508 /* Let's catch up on any state changes that happened while we were reloading/reexecing */
1509 HASHMAP_FOREACH_KEY(u
, k
, m
->units
, i
) {
1511 /* ignore aliases */
1519 static void manager_distribute_fds(Manager
*m
, FDSet
*fds
) {
1525 HASHMAP_FOREACH(u
, m
->units
, i
) {
1527 if (fdset_size(fds
) <= 0)
1530 if (!UNIT_VTABLE(u
)->distribute_fds
)
1533 UNIT_VTABLE(u
)->distribute_fds(u
, fds
);
1537 static bool manager_dbus_is_running(Manager
*m
, bool deserialized
) {
1542 /* This checks whether the dbus instance we are supposed to expose our APIs on is up. We check both the socket
1543 * and the service unit. If the 'deserialized' parameter is true we'll check the deserialized state of the unit
1544 * rather than the current one. */
1546 if (MANAGER_IS_TEST_RUN(m
))
1549 u
= manager_get_unit(m
, SPECIAL_DBUS_SOCKET
);
1552 if ((deserialized
? SOCKET(u
)->deserialized_state
: SOCKET(u
)->state
) != SOCKET_RUNNING
)
1555 u
= manager_get_unit(m
, SPECIAL_DBUS_SERVICE
);
1558 if (!IN_SET((deserialized
? SERVICE(u
)->deserialized_state
: SERVICE(u
)->state
), SERVICE_RUNNING
, SERVICE_RELOAD
))
1564 static void manager_setup_bus(Manager
*m
) {
1567 /* Let's set up our private bus connection now, unconditionally */
1568 (void) bus_init_private(m
);
1570 /* If we are in --user mode also connect to the system bus now */
1571 if (MANAGER_IS_USER(m
))
1572 (void) bus_init_system(m
);
1574 /* Let's connect to the bus now, but only if the unit is supposed to be up */
1575 if (manager_dbus_is_running(m
, MANAGER_IS_RELOADING(m
))) {
1576 (void) bus_init_api(m
);
1578 if (MANAGER_IS_SYSTEM(m
))
1579 (void) bus_init_system(m
);
1583 static void manager_preset_all(Manager
*m
) {
1588 if (m
->first_boot
<= 0)
1591 if (!MANAGER_IS_SYSTEM(m
))
1594 if (MANAGER_IS_TEST_RUN(m
))
1597 /* If this is the first boot, and we are in the host system, then preset everything */
1598 r
= unit_file_preset_all(UNIT_FILE_SYSTEM
, 0, NULL
, UNIT_FILE_PRESET_ENABLE_ONLY
, NULL
, 0);
1600 log_full_errno(r
== -EEXIST
? LOG_NOTICE
: LOG_WARNING
, r
,
1601 "Failed to populate /etc with preset unit settings, ignoring: %m");
1603 log_info("Populated /etc with preset unit settings.");
1606 static void manager_ready(Manager
*m
) {
1609 /* After having loaded everything, do the final round of catching up with what might have changed */
1611 m
->objective
= MANAGER_OK
; /* Tell everyone we are up now */
1613 /* It might be safe to log to the journal now and connect to dbus */
1614 manager_recheck_journal(m
);
1615 manager_recheck_dbus(m
);
1617 /* Let's finally catch up with any changes that took place while we were reloading/reexecing */
1620 m
->honor_device_enumeration
= true;
1623 static Manager
* manager_reloading_start(Manager
*m
) {
1627 static void manager_reloading_stopp(Manager
**m
) {
1629 assert((*m
)->n_reloading
> 0);
1630 (*m
)->n_reloading
--;
1634 int manager_startup(Manager
*m
, FILE *serialization
, FDSet
*fds
) {
1639 /* If we are running in test mode, we still want to run the generators,
1640 * but we should not touch the real generator directories. */
1641 r
= lookup_paths_init(&m
->lookup_paths
, m
->unit_file_scope
,
1642 MANAGER_IS_TEST_RUN(m
) ? LOOKUP_PATHS_TEMPORARY_GENERATED
: 0,
1645 return log_error_errno(r
, "Failed to initialize path lookup table: %m");
1647 dual_timestamp_get(m
->timestamps
+ manager_timestamp_initrd_mangle(MANAGER_TIMESTAMP_GENERATORS_START
));
1648 r
= manager_run_environment_generators(m
);
1650 r
= manager_run_generators(m
);
1651 dual_timestamp_get(m
->timestamps
+ manager_timestamp_initrd_mangle(MANAGER_TIMESTAMP_GENERATORS_FINISH
));
1655 manager_preset_all(m
);
1657 lookup_paths_log(&m
->lookup_paths
);
1660 /* This block is (optionally) done with the reloading counter bumped */
1661 _cleanup_(manager_reloading_stopp
) Manager
*reloading
= NULL
;
1663 /* If we will deserialize make sure that during enumeration this is already known, so we increase the
1664 * counter here already */
1666 reloading
= manager_reloading_start(m
);
1668 /* First, enumerate what we can from all config files */
1669 dual_timestamp_get(m
->timestamps
+ manager_timestamp_initrd_mangle(MANAGER_TIMESTAMP_UNITS_LOAD_START
));
1670 manager_enumerate_perpetual(m
);
1671 manager_enumerate(m
);
1672 dual_timestamp_get(m
->timestamps
+ manager_timestamp_initrd_mangle(MANAGER_TIMESTAMP_UNITS_LOAD_FINISH
));
1674 /* Second, deserialize if there is something to deserialize */
1675 if (serialization
) {
1676 r
= manager_deserialize(m
, serialization
, fds
);
1678 return log_error_errno(r
, "Deserialization failed: %m");
1681 /* Any fds left? Find some unit which wants them. This is useful to allow container managers to pass
1682 * some file descriptors to us pre-initialized. This enables socket-based activation of entire
1684 manager_distribute_fds(m
, fds
);
1686 /* We might have deserialized the notify fd, but if we didn't then let's create the bus now */
1687 r
= manager_setup_notify(m
);
1689 /* No sense to continue without notifications, our children would fail anyway. */
1692 r
= manager_setup_cgroups_agent(m
);
1694 /* Likewise, no sense to continue without empty cgroup notifications. */
1697 r
= manager_setup_user_lookup_fd(m
);
1699 /* This shouldn't fail, except if things are really broken. */
1702 /* Connect to the bus if we are good for it */
1703 manager_setup_bus(m
);
1705 /* Now that we are connected to all possible buses, let's deserialize who is tracking us. */
1706 r
= bus_track_coldplug(m
, &m
->subscribed
, false, m
->deserialized_subscribed
);
1708 log_warning_errno(r
, "Failed to deserialized tracked clients, ignoring: %m");
1709 m
->deserialized_subscribed
= strv_free(m
->deserialized_subscribed
);
1711 r
= manager_varlink_init(m
);
1713 log_warning_errno(r
, "Failed to set up Varlink server, ignoring: %m");
1715 /* Third, fire things up! */
1716 manager_coldplug(m
);
1718 /* Clean up runtime objects */
1722 /* Let's wait for the UnitNew/JobNew messages being sent, before we notify that the
1723 * reload is finished */
1724 m
->send_reloading_done
= true;
1732 int manager_add_job(
1738 sd_bus_error
*error
,
1745 assert(type
< _JOB_TYPE_MAX
);
1747 assert(mode
< _JOB_MODE_MAX
);
1749 if (mode
== JOB_ISOLATE
&& type
!= JOB_START
)
1750 return sd_bus_error_setf(error
, SD_BUS_ERROR_INVALID_ARGS
, "Isolate is only valid for start.");
1752 if (mode
== JOB_ISOLATE
&& !unit
->allow_isolate
)
1753 return sd_bus_error_setf(error
, BUS_ERROR_NO_ISOLATION
, "Operation refused, unit may not be isolated.");
1755 if (mode
== JOB_TRIGGERING
&& type
!= JOB_STOP
)
1756 return sd_bus_error_setf(error
, SD_BUS_ERROR_INVALID_ARGS
, "--job-mode=triggering is only valid for stop.");
1758 log_unit_debug(unit
, "Trying to enqueue job %s/%s/%s", unit
->id
, job_type_to_string(type
), job_mode_to_string(mode
));
1760 type
= job_type_collapse(type
, unit
);
1762 tr
= transaction_new(mode
== JOB_REPLACE_IRREVERSIBLY
);
1766 r
= transaction_add_job_and_dependencies(tr
, type
, unit
, NULL
, true, false,
1767 IN_SET(mode
, JOB_IGNORE_DEPENDENCIES
, JOB_IGNORE_REQUIREMENTS
),
1768 mode
== JOB_IGNORE_DEPENDENCIES
, error
);
1772 if (mode
== JOB_ISOLATE
) {
1773 r
= transaction_add_isolate_jobs(tr
, m
);
1778 if (mode
== JOB_TRIGGERING
) {
1779 r
= transaction_add_triggering_jobs(tr
, unit
);
1784 r
= transaction_activate(tr
, m
, mode
, affected_jobs
, error
);
1788 log_unit_debug(unit
,
1789 "Enqueued job %s/%s as %u", unit
->id
,
1790 job_type_to_string(type
), (unsigned) tr
->anchor_job
->id
);
1793 *ret
= tr
->anchor_job
;
1795 transaction_free(tr
);
1799 transaction_abort(tr
);
1800 transaction_free(tr
);
1804 int manager_add_job_by_name(Manager
*m
, JobType type
, const char *name
, JobMode mode
, Set
*affected_jobs
, sd_bus_error
*e
, Job
**ret
) {
1805 Unit
*unit
= NULL
; /* just to appease gcc, initialization is not really necessary */
1809 assert(type
< _JOB_TYPE_MAX
);
1811 assert(mode
< _JOB_MODE_MAX
);
1813 r
= manager_load_unit(m
, name
, NULL
, NULL
, &unit
);
1818 return manager_add_job(m
, type
, unit
, mode
, affected_jobs
, e
, ret
);
1821 int manager_add_job_by_name_and_warn(Manager
*m
, JobType type
, const char *name
, JobMode mode
, Set
*affected_jobs
, Job
**ret
) {
1822 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
1826 assert(type
< _JOB_TYPE_MAX
);
1828 assert(mode
< _JOB_MODE_MAX
);
1830 r
= manager_add_job_by_name(m
, type
, name
, mode
, affected_jobs
, &error
, ret
);
1832 return log_warning_errno(r
, "Failed to enqueue %s job for %s: %s", job_mode_to_string(mode
), name
, bus_error_message(&error
, r
));
1837 int manager_propagate_reload(Manager
*m
, Unit
*unit
, JobMode mode
, sd_bus_error
*e
) {
1843 assert(mode
< _JOB_MODE_MAX
);
1844 assert(mode
!= JOB_ISOLATE
); /* Isolate is only valid for start */
1846 tr
= transaction_new(mode
== JOB_REPLACE_IRREVERSIBLY
);
1850 /* We need an anchor job */
1851 r
= transaction_add_job_and_dependencies(tr
, JOB_NOP
, unit
, NULL
, false, false, true, true, e
);
1855 /* Failure in adding individual dependencies is ignored, so this always succeeds. */
1856 transaction_add_propagate_reload_jobs(tr
, unit
, tr
->anchor_job
, mode
== JOB_IGNORE_DEPENDENCIES
, e
);
1858 r
= transaction_activate(tr
, m
, mode
, NULL
, e
);
1862 transaction_free(tr
);
1866 transaction_abort(tr
);
1867 transaction_free(tr
);
1871 Job
*manager_get_job(Manager
*m
, uint32_t id
) {
1874 return hashmap_get(m
->jobs
, UINT32_TO_PTR(id
));
1877 Unit
*manager_get_unit(Manager
*m
, const char *name
) {
1881 return hashmap_get(m
->units
, name
);
1884 static int manager_dispatch_target_deps_queue(Manager
*m
) {
1889 static const UnitDependency deps
[] = {
1898 while ((u
= m
->target_deps_queue
)) {
1899 assert(u
->in_target_deps_queue
);
1901 LIST_REMOVE(target_deps_queue
, u
->manager
->target_deps_queue
, u
);
1902 u
->in_target_deps_queue
= false;
1904 for (k
= 0; k
< ELEMENTSOF(deps
); k
++) {
1909 HASHMAP_FOREACH_KEY(v
, target
, u
->dependencies
[deps
[k
]], i
) {
1910 r
= unit_add_default_target_dependency(u
, target
);
1920 unsigned manager_dispatch_load_queue(Manager
*m
) {
1926 /* Make sure we are not run recursively */
1927 if (m
->dispatching_load_queue
)
1930 m
->dispatching_load_queue
= true;
1932 /* Dispatches the load queue. Takes a unit from the queue and
1933 * tries to load its data until the queue is empty */
1935 while ((u
= m
->load_queue
)) {
1936 assert(u
->in_load_queue
);
1942 m
->dispatching_load_queue
= false;
1944 /* Dispatch the units waiting for their target dependencies to be added now, as all targets that we know about
1945 * should be loaded and have aliases resolved */
1946 (void) manager_dispatch_target_deps_queue(m
);
1951 bool manager_unit_cache_should_retry_load(Unit
*u
) {
1954 /* Automatic reloading from disk only applies to units which were not found sometime in the past, and
1955 * the not-found stub is kept pinned in the unit graph by dependencies. For units that were
1956 * previously loaded, we don't do automatic reloading, and daemon-reload is necessary to update. */
1957 if (u
->load_state
!= UNIT_NOT_FOUND
)
1960 /* The cache has been updated since the last time we tried to load the unit. There might be new
1961 * fragment paths to read. */
1962 if (u
->manager
->unit_cache_timestamp_hash
!= u
->fragment_not_found_timestamp_hash
)
1965 /* The cache needs to be updated because there are modifications on disk. */
1966 return !lookup_paths_timestamp_hash_same(&u
->manager
->lookup_paths
, u
->manager
->unit_cache_timestamp_hash
, NULL
);
1969 int manager_load_unit_prepare(
1976 _cleanup_(unit_freep
) Unit
*cleanup_ret
= NULL
;
1984 /* This will prepare the unit for loading, but not actually
1985 * load anything from disk. */
1987 if (path
&& !is_path(path
))
1988 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
, "Path %s is not absolute.", path
);
1991 /* 'name' and 'path' must not both be null. Check here 'path' using assert_se() to
1992 * workaround a bug in gcc that generates a -Wnonnull warning when calling basename(),
1993 * but this cannot be possible in any code path (See #6119). */
1995 name
= basename(path
);
1998 t
= unit_name_to_type(name
);
2000 if (t
== _UNIT_TYPE_INVALID
|| !unit_name_is_valid(name
, UNIT_NAME_PLAIN
|UNIT_NAME_INSTANCE
)) {
2001 if (unit_name_is_valid(name
, UNIT_NAME_TEMPLATE
))
2002 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
, "Unit name %s is missing the instance name.", name
);
2004 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
, "Unit name %s is not valid.", name
);
2007 ret
= manager_get_unit(m
, name
);
2009 /* The time-based cache allows to start new units without daemon-reload,
2010 * but if they are already referenced (because of dependencies or ordering)
2011 * then we have to force a load of the fragment. As an optimization, check
2012 * first if anything in the usual paths was modified since the last time
2013 * the cache was loaded. Also check if the last time an attempt to load the
2014 * unit was made was before the most recent cache refresh, so that we know
2015 * we need to try again — even if the cache is current, it might have been
2016 * updated in a different context before we had a chance to retry loading
2017 * this particular unit. */
2018 if (manager_unit_cache_should_retry_load(ret
))
2019 ret
->load_state
= UNIT_STUB
;
2025 ret
= cleanup_ret
= unit_new(m
, unit_vtable
[t
]->object_size
);
2031 r
= free_and_strdup(&ret
->fragment_path
, path
);
2036 r
= unit_add_name(ret
, name
);
2040 unit_add_to_load_queue(ret
);
2041 unit_add_to_dbus_queue(ret
);
2042 unit_add_to_gc_queue(ret
);
2050 int manager_load_unit(
2062 /* This will load the service information files, but not actually
2063 * start any services or anything. */
2065 r
= manager_load_unit_prepare(m
, name
, path
, e
, _ret
);
2069 manager_dispatch_load_queue(m
);
2071 *_ret
= unit_follow_merge(*_ret
);
2075 int manager_load_startable_unit_or_warn(
2081 /* Load a unit, make sure it loaded fully and is not masked. */
2083 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
2087 r
= manager_load_unit(m
, name
, path
, &error
, &unit
);
2089 return log_error_errno(r
, "Failed to load %s %s: %s",
2090 name
? "unit" : "unit file", name
?: path
,
2091 bus_error_message(&error
, r
));
2093 r
= bus_unit_validate_load_state(unit
, &error
);
2095 return log_error_errno(r
, "%s", bus_error_message(&error
, r
));
2101 void manager_dump_jobs(Manager
*s
, FILE *f
, const char *prefix
) {
2108 HASHMAP_FOREACH(j
, s
->jobs
, i
)
2109 job_dump(j
, f
, prefix
);
2112 void manager_dump_units(Manager
*s
, FILE *f
, const char *prefix
) {
2120 HASHMAP_FOREACH_KEY(u
, t
, s
->units
, i
)
2122 unit_dump(u
, f
, prefix
);
2125 void manager_dump(Manager
*m
, FILE *f
, const char *prefix
) {
2131 for (q
= 0; q
< _MANAGER_TIMESTAMP_MAX
; q
++) {
2132 const dual_timestamp
*t
= m
->timestamps
+ q
;
2133 char buf
[CONST_MAX(FORMAT_TIMESPAN_MAX
, FORMAT_TIMESTAMP_MAX
)];
2135 if (dual_timestamp_is_set(t
))
2136 fprintf(f
, "%sTimestamp %s: %s\n",
2138 manager_timestamp_to_string(q
),
2139 timestamp_is_set(t
->realtime
) ? format_timestamp(buf
, sizeof buf
, t
->realtime
) :
2140 format_timespan(buf
, sizeof buf
, t
->monotonic
, 1));
2143 manager_dump_units(m
, f
, prefix
);
2144 manager_dump_jobs(m
, f
, prefix
);
2147 int manager_get_dump_string(Manager
*m
, char **ret
) {
2148 _cleanup_free_
char *dump
= NULL
;
2149 _cleanup_fclose_
FILE *f
= NULL
;
2156 f
= open_memstream_unlocked(&dump
, &size
);
2160 manager_dump(m
, f
, NULL
);
2162 r
= fflush_and_check(f
);
2168 *ret
= TAKE_PTR(dump
);
2173 void manager_clear_jobs(Manager
*m
) {
2178 while ((j
= hashmap_first(m
->jobs
)))
2179 /* No need to recurse. We're cancelling all jobs. */
2180 job_finish_and_invalidate(j
, JOB_CANCELED
, false, false);
2183 void manager_unwatch_pid(Manager
*m
, pid_t pid
) {
2186 /* First let's drop the unit keyed as "pid". */
2187 (void) hashmap_remove(m
->watch_pids
, PID_TO_PTR(pid
));
2189 /* Then, let's also drop the array keyed by -pid. */
2190 free(hashmap_remove(m
->watch_pids
, PID_TO_PTR(-pid
)));
2193 static int manager_dispatch_run_queue(sd_event_source
*source
, void *userdata
) {
2194 Manager
*m
= userdata
;
2200 while ((j
= prioq_peek(m
->run_queue
))) {
2201 assert(j
->installed
);
2202 assert(j
->in_run_queue
);
2204 (void) job_run_and_invalidate(j
);
2207 if (m
->n_running_jobs
> 0)
2208 manager_watch_jobs_in_progress(m
);
2210 if (m
->n_on_console
> 0)
2211 manager_watch_idle_pipe(m
);
2216 static unsigned manager_dispatch_dbus_queue(Manager
*m
) {
2217 unsigned n
= 0, budget
;
2223 /* When we are reloading, let's not wait with generating signals, since we need to exit the manager as quickly
2224 * as we can. There's no point in throttling generation of signals in that case. */
2225 if (MANAGER_IS_RELOADING(m
) || m
->send_reloading_done
|| m
->pending_reload_message
)
2226 budget
= (unsigned) -1; /* infinite budget in this case */
2228 /* Anything to do at all? */
2229 if (!m
->dbus_unit_queue
&& !m
->dbus_job_queue
)
2232 /* Do we have overly many messages queued at the moment? If so, let's not enqueue more on top, let's
2233 * sit this cycle out, and process things in a later cycle when the queues got a bit emptier. */
2234 if (manager_bus_n_queued_write(m
) > MANAGER_BUS_BUSY_THRESHOLD
)
2237 /* Only process a certain number of units/jobs per event loop iteration. Even if the bus queue wasn't
2238 * overly full before this call we shouldn't increase it in size too wildly in one step, and we
2239 * shouldn't monopolize CPU time with generating these messages. Note the difference in counting of
2240 * this "budget" and the "threshold" above: the "budget" is decreased only once per generated message,
2241 * regardless how many buses/direct connections it is enqueued on, while the "threshold" is applied to
2242 * each queued instance of bus message, i.e. if the same message is enqueued to five buses/direct
2243 * connections it will be counted five times. This difference in counting ("references"
2244 * vs. "instances") is primarily a result of the fact that it's easier to implement it this way,
2245 * however it also reflects the thinking that the "threshold" should put a limit on used queue memory,
2246 * i.e. space, while the "budget" should put a limit on time. Also note that the "threshold" is
2247 * currently chosen much higher than the "budget". */
2248 budget
= MANAGER_BUS_MESSAGE_BUDGET
;
2251 while (budget
!= 0 && (u
= m
->dbus_unit_queue
)) {
2253 assert(u
->in_dbus_queue
);
2255 bus_unit_send_change_signal(u
);
2258 if (budget
!= (unsigned) -1)
2262 while (budget
!= 0 && (j
= m
->dbus_job_queue
)) {
2263 assert(j
->in_dbus_queue
);
2265 bus_job_send_change_signal(j
);
2268 if (budget
!= (unsigned) -1)
2272 if (m
->send_reloading_done
) {
2273 m
->send_reloading_done
= false;
2274 bus_manager_send_reloading(m
, false);
2278 if (m
->pending_reload_message
) {
2279 bus_send_pending_reload_message(m
);
2286 static int manager_dispatch_cgroups_agent_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
2287 Manager
*m
= userdata
;
2291 n
= recv(fd
, buf
, sizeof(buf
), 0);
2293 return log_error_errno(errno
, "Failed to read cgroups agent message: %m");
2295 log_error("Got zero-length cgroups agent message, ignoring.");
2298 if ((size_t) n
>= sizeof(buf
)) {
2299 log_error("Got overly long cgroups agent message, ignoring.");
2303 if (memchr(buf
, 0, n
)) {
2304 log_error("Got cgroups agent message with embedded NUL byte, ignoring.");
2309 manager_notify_cgroup_empty(m
, buf
);
2310 (void) bus_forward_agent_released(m
, buf
);
2315 static bool manager_process_barrier_fd(char * const *tags
, FDSet
*fds
) {
2317 /* nothing else must be sent when using BARRIER=1 */
2318 if (strv_contains(tags
, "BARRIER=1")) {
2319 if (strv_length(tags
) == 1) {
2320 if (fdset_size(fds
) != 1)
2321 log_warning("Got incorrect number of fds with BARRIER=1, closing them.");
2323 log_warning("Extra notification messages sent with BARRIER=1, ignoring everything.");
2325 /* Drop the message if BARRIER=1 was found */
2332 static void manager_invoke_notify_message(
2335 const struct ucred
*ucred
,
2344 if (u
->notifygen
== m
->notifygen
) /* Already invoked on this same unit in this same iteration? */
2346 u
->notifygen
= m
->notifygen
;
2348 if (UNIT_VTABLE(u
)->notify_message
)
2349 UNIT_VTABLE(u
)->notify_message(u
, ucred
, tags
, fds
);
2351 else if (DEBUG_LOGGING
) {
2352 _cleanup_free_
char *buf
= NULL
, *x
= NULL
, *y
= NULL
;
2354 buf
= strv_join(tags
, ", ");
2356 x
= ellipsize(buf
, 20, 90);
2360 log_unit_debug(u
, "Got notification message \"%s\", ignoring.", strnull(y
));
2364 static int manager_dispatch_notify_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
2366 _cleanup_fdset_free_ FDSet
*fds
= NULL
;
2367 Manager
*m
= userdata
;
2368 char buf
[NOTIFY_BUFFER_MAX
+1];
2369 struct iovec iovec
= {
2371 .iov_len
= sizeof(buf
)-1,
2373 CMSG_BUFFER_TYPE(CMSG_SPACE(sizeof(struct ucred
)) +
2374 CMSG_SPACE(sizeof(int) * NOTIFY_FD_MAX
)) control
;
2375 struct msghdr msghdr
= {
2378 .msg_control
= &control
,
2379 .msg_controllen
= sizeof(control
),
2382 struct cmsghdr
*cmsg
;
2383 struct ucred
*ucred
= NULL
;
2384 _cleanup_free_ Unit
**array_copy
= NULL
;
2385 _cleanup_strv_free_
char **tags
= NULL
;
2386 Unit
*u1
, *u2
, **array
;
2387 int r
, *fd_array
= NULL
;
2393 assert(m
->notify_fd
== fd
);
2395 if (revents
!= EPOLLIN
) {
2396 log_warning("Got unexpected poll event for notify fd.");
2400 n
= recvmsg_safe(m
->notify_fd
, &msghdr
, MSG_DONTWAIT
|MSG_CMSG_CLOEXEC
|MSG_TRUNC
);
2401 if (IN_SET(n
, -EAGAIN
, -EINTR
))
2402 return 0; /* Spurious wakeup, try again */
2404 /* If this is any other, real error, then let's stop processing this socket. This of course
2405 * means we won't take notification messages anymore, but that's still better than busy
2406 * looping around this: being woken up over and over again but being unable to actually read
2407 * the message off the socket. */
2408 return log_error_errno(n
, "Failed to receive notification message: %m");
2410 CMSG_FOREACH(cmsg
, &msghdr
) {
2411 if (cmsg
->cmsg_level
== SOL_SOCKET
&& cmsg
->cmsg_type
== SCM_RIGHTS
) {
2414 fd_array
= (int*) CMSG_DATA(cmsg
);
2415 n_fds
= (cmsg
->cmsg_len
- CMSG_LEN(0)) / sizeof(int);
2417 } else if (cmsg
->cmsg_level
== SOL_SOCKET
&&
2418 cmsg
->cmsg_type
== SCM_CREDENTIALS
&&
2419 cmsg
->cmsg_len
== CMSG_LEN(sizeof(struct ucred
))) {
2422 ucred
= (struct ucred
*) CMSG_DATA(cmsg
);
2429 r
= fdset_new_array(&fds
, fd_array
, n_fds
);
2431 close_many(fd_array
, n_fds
);
2437 if (!ucred
|| !pid_is_valid(ucred
->pid
)) {
2438 log_warning("Received notify message without valid credentials. Ignoring.");
2442 if ((size_t) n
>= sizeof(buf
) || (msghdr
.msg_flags
& MSG_TRUNC
)) {
2443 log_warning("Received notify message exceeded maximum size. Ignoring.");
2447 /* As extra safety check, let's make sure the string we get doesn't contain embedded NUL bytes. We permit one
2448 * trailing NUL byte in the message, but don't expect it. */
2449 if (n
> 1 && memchr(buf
, 0, n
-1)) {
2450 log_warning("Received notify message with embedded NUL bytes. Ignoring.");
2454 /* Make sure it's NUL-terminated, then parse it to obtain the tags list */
2456 tags
= strv_split_newlines(buf
);
2462 /* possibly a barrier fd, let's see */
2463 if (manager_process_barrier_fd(tags
, fds
))
2466 /* Increase the generation counter used for filtering out duplicate unit invocations. */
2469 /* Notify every unit that might be interested, which might be multiple. */
2470 u1
= manager_get_unit_by_pid_cgroup(m
, ucred
->pid
);
2471 u2
= hashmap_get(m
->watch_pids
, PID_TO_PTR(ucred
->pid
));
2472 array
= hashmap_get(m
->watch_pids
, PID_TO_PTR(-ucred
->pid
));
2479 array_copy
= newdup(Unit
*, array
, k
+1);
2483 /* And now invoke the per-unit callbacks. Note that manager_invoke_notify_message() will handle duplicate units
2484 * make sure we only invoke each unit's handler once. */
2486 manager_invoke_notify_message(m
, u1
, ucred
, tags
, fds
);
2490 manager_invoke_notify_message(m
, u2
, ucred
, tags
, fds
);
2494 for (size_t i
= 0; array_copy
[i
]; i
++) {
2495 manager_invoke_notify_message(m
, array_copy
[i
], ucred
, tags
, fds
);
2500 log_warning("Cannot find unit for notify message of PID "PID_FMT
", ignoring.", ucred
->pid
);
2502 if (fdset_size(fds
) > 0)
2503 log_warning("Got extra auxiliary fds with notification message, closing them.");
2508 static void manager_invoke_sigchld_event(
2511 const siginfo_t
*si
) {
2517 /* Already invoked the handler of this unit in this iteration? Then don't process this again */
2518 if (u
->sigchldgen
== m
->sigchldgen
)
2520 u
->sigchldgen
= m
->sigchldgen
;
2522 log_unit_debug(u
, "Child "PID_FMT
" belongs to %s.", si
->si_pid
, u
->id
);
2523 unit_unwatch_pid(u
, si
->si_pid
);
2525 if (UNIT_VTABLE(u
)->sigchld_event
)
2526 UNIT_VTABLE(u
)->sigchld_event(u
, si
->si_pid
, si
->si_code
, si
->si_status
);
2529 static int manager_dispatch_sigchld(sd_event_source
*source
, void *userdata
) {
2530 Manager
*m
= userdata
;
2537 /* First we call waitid() for a PID and do not reap the zombie. That way we can still access /proc/$PID for it
2538 * while it is a zombie. */
2540 if (waitid(P_ALL
, 0, &si
, WEXITED
|WNOHANG
|WNOWAIT
) < 0) {
2542 if (errno
!= ECHILD
)
2543 log_error_errno(errno
, "Failed to peek for child with waitid(), ignoring: %m");
2551 if (IN_SET(si
.si_code
, CLD_EXITED
, CLD_KILLED
, CLD_DUMPED
)) {
2552 _cleanup_free_ Unit
**array_copy
= NULL
;
2553 _cleanup_free_
char *name
= NULL
;
2554 Unit
*u1
, *u2
, **array
;
2556 (void) get_process_comm(si
.si_pid
, &name
);
2558 log_debug("Child "PID_FMT
" (%s) died (code=%s, status=%i/%s)",
2559 si
.si_pid
, strna(name
),
2560 sigchld_code_to_string(si
.si_code
),
2562 strna(si
.si_code
== CLD_EXITED
2563 ? exit_status_to_string(si
.si_status
, EXIT_STATUS_FULL
)
2564 : signal_to_string(si
.si_status
)));
2566 /* Increase the generation counter used for filtering out duplicate unit invocations */
2569 /* And now figure out the unit this belongs to, it might be multiple... */
2570 u1
= manager_get_unit_by_pid_cgroup(m
, si
.si_pid
);
2571 u2
= hashmap_get(m
->watch_pids
, PID_TO_PTR(si
.si_pid
));
2572 array
= hashmap_get(m
->watch_pids
, PID_TO_PTR(-si
.si_pid
));
2576 /* Count how many entries the array has */
2580 /* Make a copy of the array so that we don't trip up on the array changing beneath us */
2581 array_copy
= newdup(Unit
*, array
, n
+1);
2586 /* Finally, execute them all. Note that u1, u2 and the array might contain duplicates, but
2587 * that's fine, manager_invoke_sigchld_event() will ensure we only invoke the handlers once for
2588 * each iteration. */
2590 /* We check for oom condition, in case we got SIGCHLD before the oom notification.
2591 * We only do this for the cgroup the PID belonged to. */
2592 (void) unit_check_oom(u1
);
2594 manager_invoke_sigchld_event(m
, u1
, &si
);
2597 manager_invoke_sigchld_event(m
, u2
, &si
);
2599 for (size_t i
= 0; array_copy
[i
]; i
++)
2600 manager_invoke_sigchld_event(m
, array_copy
[i
], &si
);
2603 /* And now, we actually reap the zombie. */
2604 if (waitid(P_PID
, si
.si_pid
, &si
, WEXITED
) < 0) {
2605 log_error_errno(errno
, "Failed to dequeue child, ignoring: %m");
2612 /* All children processed for now, turn off event source */
2614 r
= sd_event_source_set_enabled(m
->sigchld_event_source
, SD_EVENT_OFF
);
2616 return log_error_errno(r
, "Failed to disable SIGCHLD event source: %m");
2621 static void manager_start_target(Manager
*m
, const char *name
, JobMode mode
) {
2622 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
2625 log_debug("Activating special unit %s", name
);
2627 r
= manager_add_job_by_name(m
, JOB_START
, name
, mode
, NULL
, &error
, NULL
);
2629 log_error("Failed to enqueue %s job: %s", name
, bus_error_message(&error
, r
));
2632 static void manager_handle_ctrl_alt_del(Manager
*m
) {
2633 /* If the user presses C-A-D more than
2634 * 7 times within 2s, we reboot/shutdown immediately,
2635 * unless it was disabled in system.conf */
2637 if (ratelimit_below(&m
->ctrl_alt_del_ratelimit
) || m
->cad_burst_action
== EMERGENCY_ACTION_NONE
)
2638 manager_start_target(m
, SPECIAL_CTRL_ALT_DEL_TARGET
, JOB_REPLACE_IRREVERSIBLY
);
2640 emergency_action(m
, m
->cad_burst_action
, EMERGENCY_ACTION_WARN
, NULL
, -1,
2641 "Ctrl-Alt-Del was pressed more than 7 times within 2s");
2644 static int manager_dispatch_signal_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
2645 Manager
*m
= userdata
;
2647 struct signalfd_siginfo sfsi
;
2651 assert(m
->signal_fd
== fd
);
2653 if (revents
!= EPOLLIN
) {
2654 log_warning("Got unexpected events from signal file descriptor.");
2658 n
= read(m
->signal_fd
, &sfsi
, sizeof(sfsi
));
2659 if (n
!= sizeof(sfsi
)) {
2661 log_warning("Truncated read from signal fd (%zu bytes), ignoring!", n
);
2665 if (IN_SET(errno
, EINTR
, EAGAIN
))
2668 /* We return an error here, which will kill this handler,
2669 * to avoid a busy loop on read error. */
2670 return log_error_errno(errno
, "Reading from signal fd failed: %m");
2673 log_received_signal(sfsi
.ssi_signo
== SIGCHLD
||
2674 (sfsi
.ssi_signo
== SIGTERM
&& MANAGER_IS_USER(m
))
2675 ? LOG_DEBUG
: LOG_INFO
,
2678 switch (sfsi
.ssi_signo
) {
2681 r
= sd_event_source_set_enabled(m
->sigchld_event_source
, SD_EVENT_ON
);
2683 log_warning_errno(r
, "Failed to enable SIGCHLD event source, ignoring: %m");
2688 if (MANAGER_IS_SYSTEM(m
)) {
2689 /* This is for compatibility with the original sysvinit */
2690 if (verify_run_space_and_log("Refusing to reexecute") < 0)
2693 m
->objective
= MANAGER_REEXECUTE
;
2699 if (MANAGER_IS_SYSTEM(m
))
2700 manager_handle_ctrl_alt_del(m
);
2702 manager_start_target(m
, SPECIAL_EXIT_TARGET
,
2703 JOB_REPLACE_IRREVERSIBLY
);
2707 /* This is a nop on non-init */
2708 if (MANAGER_IS_SYSTEM(m
))
2709 manager_start_target(m
, SPECIAL_KBREQUEST_TARGET
, JOB_REPLACE
);
2714 /* This is a nop on non-init */
2715 if (MANAGER_IS_SYSTEM(m
))
2716 manager_start_target(m
, SPECIAL_SIGPWR_TARGET
, JOB_REPLACE
);
2721 if (manager_dbus_is_running(m
, false)) {
2722 log_info("Trying to reconnect to bus...");
2724 (void) bus_init_api(m
);
2726 if (MANAGER_IS_SYSTEM(m
))
2727 (void) bus_init_system(m
);
2729 log_info("Starting D-Bus service...");
2730 manager_start_target(m
, SPECIAL_DBUS_SERVICE
, JOB_REPLACE
);
2736 _cleanup_free_
char *dump
= NULL
;
2738 r
= manager_get_dump_string(m
, &dump
);
2740 log_warning_errno(errno
, "Failed to acquire manager dump: %m");
2744 log_dump(LOG_INFO
, dump
);
2749 if (verify_run_space_and_log("Refusing to reload") < 0)
2752 m
->objective
= MANAGER_RELOAD
;
2757 /* Starting SIGRTMIN+0 */
2758 static const struct {
2761 } target_table
[] = {
2762 [0] = { SPECIAL_DEFAULT_TARGET
, JOB_ISOLATE
},
2763 [1] = { SPECIAL_RESCUE_TARGET
, JOB_ISOLATE
},
2764 [2] = { SPECIAL_EMERGENCY_TARGET
, JOB_ISOLATE
},
2765 [3] = { SPECIAL_HALT_TARGET
, JOB_REPLACE_IRREVERSIBLY
},
2766 [4] = { SPECIAL_POWEROFF_TARGET
, JOB_REPLACE_IRREVERSIBLY
},
2767 [5] = { SPECIAL_REBOOT_TARGET
, JOB_REPLACE_IRREVERSIBLY
},
2768 [6] = { SPECIAL_KEXEC_TARGET
, JOB_REPLACE_IRREVERSIBLY
},
2771 /* Starting SIGRTMIN+13, so that target halt and system halt are 10 apart */
2772 static const ManagerObjective objective_table
[] = {
2774 [1] = MANAGER_POWEROFF
,
2775 [2] = MANAGER_REBOOT
,
2776 [3] = MANAGER_KEXEC
,
2779 if ((int) sfsi
.ssi_signo
>= SIGRTMIN
+0 &&
2780 (int) sfsi
.ssi_signo
< SIGRTMIN
+(int) ELEMENTSOF(target_table
)) {
2781 int idx
= (int) sfsi
.ssi_signo
- SIGRTMIN
;
2782 manager_start_target(m
, target_table
[idx
].target
,
2783 target_table
[idx
].mode
);
2787 if ((int) sfsi
.ssi_signo
>= SIGRTMIN
+13 &&
2788 (int) sfsi
.ssi_signo
< SIGRTMIN
+13+(int) ELEMENTSOF(objective_table
)) {
2789 m
->objective
= objective_table
[sfsi
.ssi_signo
- SIGRTMIN
- 13];
2793 switch (sfsi
.ssi_signo
- SIGRTMIN
) {
2796 manager_override_show_status(m
, SHOW_STATUS_YES
, "signal");
2800 manager_override_show_status(m
, SHOW_STATUS_NO
, "signal");
2804 manager_override_log_level(m
, LOG_DEBUG
);
2808 manager_restore_original_log_level(m
);
2812 if (MANAGER_IS_USER(m
)) {
2813 m
->objective
= MANAGER_EXIT
;
2817 /* This is a nop on init */
2821 case 29: /* compatibility: used to be mapped to LOG_TARGET_SYSLOG_OR_KMSG */
2822 manager_restore_original_log_target(m
);
2826 manager_override_log_target(m
, LOG_TARGET_CONSOLE
);
2830 manager_override_log_target(m
, LOG_TARGET_KMSG
);
2834 log_warning("Got unhandled signal <%s>.", signal_to_string(sfsi
.ssi_signo
));
2841 static int manager_dispatch_time_change_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
2842 Manager
*m
= userdata
;
2847 assert(m
->time_change_fd
== fd
);
2849 log_struct(LOG_DEBUG
,
2850 "MESSAGE_ID=" SD_MESSAGE_TIME_CHANGE_STR
,
2851 LOG_MESSAGE("Time has been changed"));
2853 /* Restart the watch */
2854 (void) manager_setup_time_change(m
);
2856 HASHMAP_FOREACH(u
, m
->units
, i
)
2857 if (UNIT_VTABLE(u
)->time_change
)
2858 UNIT_VTABLE(u
)->time_change(u
);
2863 static int manager_dispatch_timezone_change(
2864 sd_event_source
*source
,
2865 const struct inotify_event
*e
,
2868 Manager
*m
= userdata
;
2875 log_debug("inotify event for /etc/localtime");
2877 changed
= manager_read_timezone_stat(m
);
2881 /* Something changed, restart the watch, to ensure we watch the new /etc/localtime if it changed */
2882 (void) manager_setup_timezone_change(m
);
2884 /* Read the new timezone */
2887 log_debug("Timezone has been changed (now: %s).", tzname
[daylight
]);
2889 HASHMAP_FOREACH(u
, m
->units
, i
)
2890 if (UNIT_VTABLE(u
)->timezone_change
)
2891 UNIT_VTABLE(u
)->timezone_change(u
);
2896 static int manager_dispatch_idle_pipe_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
2897 Manager
*m
= userdata
;
2900 assert(m
->idle_pipe
[2] == fd
);
2902 /* There's at least one Type=idle child that just gave up on us waiting for the boot process to complete. Let's
2903 * now turn off any further console output if there's at least one service that needs console access, so that
2904 * from now on our own output should not spill into that service's output anymore. After all, we support
2905 * Type=idle only to beautify console output and it generally is set on services that want to own the console
2906 * exclusively without our interference. */
2907 m
->no_console_output
= m
->n_on_console
> 0;
2909 /* Acknowledge the child's request, and let all all other children know too that they shouldn't wait any longer
2910 * by closing the pipes towards them, which is what they are waiting for. */
2911 manager_close_idle_pipe(m
);
2916 static int manager_dispatch_jobs_in_progress(sd_event_source
*source
, usec_t usec
, void *userdata
) {
2917 Manager
*m
= userdata
;
2923 manager_print_jobs_in_progress(m
);
2925 r
= sd_event_source_set_time_relative(source
, JOBS_IN_PROGRESS_PERIOD_USEC
);
2929 return sd_event_source_set_enabled(source
, SD_EVENT_ONESHOT
);
2932 int manager_loop(Manager
*m
) {
2933 RateLimit rl
= { .interval
= 1*USEC_PER_SEC
, .burst
= 50000 };
2937 assert(m
->objective
== MANAGER_OK
); /* Ensure manager_startup() has been called */
2939 manager_check_finished(m
);
2941 /* There might still be some zombies hanging around from before we were exec()'ed. Let's reap them. */
2942 r
= sd_event_source_set_enabled(m
->sigchld_event_source
, SD_EVENT_ON
);
2944 return log_error_errno(r
, "Failed to enable SIGCHLD event source: %m");
2946 while (m
->objective
== MANAGER_OK
) {
2947 usec_t wait_usec
, watchdog_usec
;
2949 watchdog_usec
= manager_get_watchdog(m
, WATCHDOG_RUNTIME
);
2950 if (timestamp_is_set(watchdog_usec
))
2953 if (!ratelimit_below(&rl
)) {
2954 /* Yay, something is going seriously wrong, pause a little */
2955 log_warning("Looping too fast. Throttling execution a little.");
2959 if (manager_dispatch_load_queue(m
) > 0)
2962 if (manager_dispatch_gc_job_queue(m
) > 0)
2965 if (manager_dispatch_gc_unit_queue(m
) > 0)
2968 if (manager_dispatch_cleanup_queue(m
) > 0)
2971 if (manager_dispatch_cgroup_realize_queue(m
) > 0)
2974 if (manager_dispatch_stop_when_unneeded_queue(m
) > 0)
2977 if (manager_dispatch_dbus_queue(m
) > 0)
2980 /* Sleep for watchdog runtime wait time */
2981 if (timestamp_is_set(watchdog_usec
))
2982 wait_usec
= watchdog_runtime_wait();
2984 wait_usec
= USEC_INFINITY
;
2986 r
= sd_event_run(m
->event
, wait_usec
);
2988 return log_error_errno(r
, "Failed to run event loop: %m");
2991 return m
->objective
;
2994 int manager_load_unit_from_dbus_path(Manager
*m
, const char *s
, sd_bus_error
*e
, Unit
**_u
) {
2995 _cleanup_free_
char *n
= NULL
;
2996 sd_id128_t invocation_id
;
3004 r
= unit_name_from_dbus_path(s
, &n
);
3008 /* Permit addressing units by invocation ID: if the passed bus path is suffixed by a 128bit ID then we use it
3009 * as invocation ID. */
3010 r
= sd_id128_from_string(n
, &invocation_id
);
3012 u
= hashmap_get(m
->units_by_invocation_id
, &invocation_id
);
3018 return sd_bus_error_setf(e
, BUS_ERROR_NO_UNIT_FOR_INVOCATION_ID
,
3019 "No unit with the specified invocation ID " SD_ID128_FORMAT_STR
" known.",
3020 SD_ID128_FORMAT_VAL(invocation_id
));
3023 /* If this didn't work, we check if this is a unit name */
3024 if (!unit_name_is_valid(n
, UNIT_NAME_PLAIN
|UNIT_NAME_INSTANCE
)) {
3025 _cleanup_free_
char *nn
= NULL
;
3028 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
,
3029 "Unit name %s is neither a valid invocation ID nor unit name.", strnull(nn
));
3032 r
= manager_load_unit(m
, n
, NULL
, e
, &u
);
3040 int manager_get_job_from_dbus_path(Manager
*m
, const char *s
, Job
**_j
) {
3050 p
= startswith(s
, "/org/freedesktop/systemd1/job/");
3054 r
= safe_atou(p
, &id
);
3058 j
= manager_get_job(m
, id
);
3067 void manager_send_unit_audit(Manager
*m
, Unit
*u
, int type
, bool success
) {
3070 _cleanup_free_
char *p
= NULL
;
3074 if (!MANAGER_IS_SYSTEM(m
))
3077 audit_fd
= get_audit_fd();
3081 /* Don't generate audit events if the service was already
3082 * started and we're just deserializing */
3083 if (MANAGER_IS_RELOADING(m
))
3086 if (u
->type
!= UNIT_SERVICE
)
3089 r
= unit_name_to_prefix_and_instance(u
->id
, &p
);
3091 log_error_errno(r
, "Failed to extract prefix and instance of unit name: %m");
3095 msg
= strjoina("unit=", p
);
3096 if (audit_log_user_comm_message(audit_fd
, type
, msg
, "systemd", NULL
, NULL
, NULL
, success
) < 0) {
3098 /* We aren't allowed to send audit messages?
3099 * Then let's not retry again. */
3102 log_warning_errno(errno
, "Failed to send audit message: %m");
3108 void manager_send_unit_plymouth(Manager
*m
, Unit
*u
) {
3109 static const union sockaddr_union sa
= PLYMOUTH_SOCKET
;
3110 _cleanup_free_
char *message
= NULL
;
3111 _cleanup_close_
int fd
= -1;
3114 /* Don't generate plymouth events if the service was already
3115 * started and we're just deserializing */
3116 if (MANAGER_IS_RELOADING(m
))
3119 if (!MANAGER_IS_SYSTEM(m
))
3122 if (detect_container() > 0)
3125 if (!IN_SET(u
->type
, UNIT_SERVICE
, UNIT_MOUNT
, UNIT_SWAP
))
3128 /* We set SOCK_NONBLOCK here so that we rather drop the
3129 * message then wait for plymouth */
3130 fd
= socket(AF_UNIX
, SOCK_STREAM
|SOCK_CLOEXEC
|SOCK_NONBLOCK
, 0);
3132 log_error_errno(errno
, "socket() failed: %m");
3136 if (connect(fd
, &sa
.sa
, SOCKADDR_UN_LEN(sa
.un
)) < 0) {
3137 if (!IN_SET(errno
, EAGAIN
, ENOENT
) && !ERRNO_IS_DISCONNECT(errno
))
3138 log_error_errno(errno
, "connect() failed: %m");
3142 if (asprintf(&message
, "U\002%c%s%n", (int) (strlen(u
->id
) + 1), u
->id
, &n
) < 0) {
3148 if (write(fd
, message
, n
+ 1) != n
+ 1)
3149 if (!IN_SET(errno
, EAGAIN
, ENOENT
) && !ERRNO_IS_DISCONNECT(errno
))
3150 log_error_errno(errno
, "Failed to write Plymouth message: %m");
3153 int manager_open_serialization(Manager
*m
, FILE **_f
) {
3154 _cleanup_close_
int fd
= -1;
3159 fd
= open_serialization_fd("systemd-state");
3163 f
= take_fdopen(&fd
, "w+");
3171 static bool manager_timestamp_shall_serialize(ManagerTimestamp t
) {
3176 /* The following timestamps only apply to the host system, hence only serialize them there */
3178 MANAGER_TIMESTAMP_USERSPACE
, MANAGER_TIMESTAMP_FINISH
,
3179 MANAGER_TIMESTAMP_SECURITY_START
, MANAGER_TIMESTAMP_SECURITY_FINISH
,
3180 MANAGER_TIMESTAMP_GENERATORS_START
, MANAGER_TIMESTAMP_GENERATORS_FINISH
,
3181 MANAGER_TIMESTAMP_UNITS_LOAD_START
, MANAGER_TIMESTAMP_UNITS_LOAD_FINISH
);
3184 #define DESTROY_IPC_FLAG (UINT32_C(1) << 31)
3186 static void manager_serialize_uid_refs_internal(
3190 const char *field_name
) {
3200 /* Serialize the UID reference table. Or actually, just the IPC destruction flag of it, as
3201 * the actual counter of it is better rebuild after a reload/reexec. */
3203 HASHMAP_FOREACH_KEY(p
, k
, *uid_refs
, i
) {
3207 uid
= PTR_TO_UID(k
);
3208 c
= PTR_TO_UINT32(p
);
3210 if (!(c
& DESTROY_IPC_FLAG
))
3213 (void) serialize_item_format(f
, field_name
, UID_FMT
, uid
);
3217 static void manager_serialize_uid_refs(Manager
*m
, FILE *f
) {
3218 manager_serialize_uid_refs_internal(m
, f
, &m
->uid_refs
, "destroy-ipc-uid");
3221 static void manager_serialize_gid_refs(Manager
*m
, FILE *f
) {
3222 manager_serialize_uid_refs_internal(m
, f
, &m
->gid_refs
, "destroy-ipc-gid");
3225 int manager_serialize(
3229 bool switching_root
) {
3241 _cleanup_(manager_reloading_stopp
) _unused_ Manager
*reloading
= manager_reloading_start(m
);
3243 (void) serialize_item_format(f
, "current-job-id", "%" PRIu32
, m
->current_job_id
);
3244 (void) serialize_item_format(f
, "n-installed-jobs", "%u", m
->n_installed_jobs
);
3245 (void) serialize_item_format(f
, "n-failed-jobs", "%u", m
->n_failed_jobs
);
3246 (void) serialize_bool(f
, "taint-usr", m
->taint_usr
);
3247 (void) serialize_bool(f
, "ready-sent", m
->ready_sent
);
3248 (void) serialize_bool(f
, "taint-logged", m
->taint_logged
);
3249 (void) serialize_bool(f
, "service-watchdogs", m
->service_watchdogs
);
3251 /* After switching root, udevd has not been started yet. So, enumeration results should not be emitted. */
3252 (void) serialize_bool(f
, "honor-device-enumeration", !switching_root
);
3254 if (m
->show_status_overridden
!= _SHOW_STATUS_INVALID
)
3255 (void) serialize_item(f
, "show-status-overridden",
3256 show_status_to_string(m
->show_status_overridden
));
3258 if (m
->log_level_overridden
)
3259 (void) serialize_item_format(f
, "log-level-override", "%i", log_get_max_level());
3260 if (m
->log_target_overridden
)
3261 (void) serialize_item(f
, "log-target-override", log_target_to_string(log_get_target()));
3263 (void) serialize_usec(f
, "runtime-watchdog-overridden", m
->watchdog_overridden
[WATCHDOG_RUNTIME
]);
3264 (void) serialize_usec(f
, "reboot-watchdog-overridden", m
->watchdog_overridden
[WATCHDOG_REBOOT
]);
3265 (void) serialize_usec(f
, "kexec-watchdog-overridden", m
->watchdog_overridden
[WATCHDOG_KEXEC
]);
3267 for (q
= 0; q
< _MANAGER_TIMESTAMP_MAX
; q
++) {
3268 _cleanup_free_
char *joined
= NULL
;
3270 if (!manager_timestamp_shall_serialize(q
))
3273 joined
= strjoin(manager_timestamp_to_string(q
), "-timestamp");
3277 (void) serialize_dual_timestamp(f
, joined
, m
->timestamps
+ q
);
3280 if (!switching_root
)
3281 (void) serialize_strv(f
, "env", m
->client_environment
);
3283 if (m
->notify_fd
>= 0) {
3284 r
= serialize_fd(f
, fds
, "notify-fd", m
->notify_fd
);
3288 (void) serialize_item(f
, "notify-socket", m
->notify_socket
);
3291 if (m
->cgroups_agent_fd
>= 0) {
3292 r
= serialize_fd(f
, fds
, "cgroups-agent-fd", m
->cgroups_agent_fd
);
3297 if (m
->user_lookup_fds
[0] >= 0) {
3300 copy0
= fdset_put_dup(fds
, m
->user_lookup_fds
[0]);
3302 return log_error_errno(copy0
, "Failed to add user lookup fd to serialization: %m");
3304 copy1
= fdset_put_dup(fds
, m
->user_lookup_fds
[1]);
3306 return log_error_errno(copy1
, "Failed to add user lookup fd to serialization: %m");
3308 (void) serialize_item_format(f
, "user-lookup", "%i %i", copy0
, copy1
);
3311 bus_track_serialize(m
->subscribed
, f
, "subscribed");
3313 r
= dynamic_user_serialize(m
, f
, fds
);
3317 manager_serialize_uid_refs(m
, f
);
3318 manager_serialize_gid_refs(m
, f
);
3320 r
= exec_runtime_serialize(m
, f
, fds
);
3324 (void) fputc('\n', f
);
3326 HASHMAP_FOREACH_KEY(u
, t
, m
->units
, i
) {
3334 r
= unit_serialize(u
, f
, fds
, !switching_root
);
3339 r
= fflush_and_check(f
);
3341 return log_error_errno(r
, "Failed to flush serialization: %m");
3343 r
= bus_fdset_add_all(m
, fds
);
3345 return log_error_errno(r
, "Failed to add bus sockets to serialization: %m");
3350 static int manager_deserialize_one_unit(Manager
*m
, const char *name
, FILE *f
, FDSet
*fds
) {
3354 r
= manager_load_unit(m
, name
, NULL
, NULL
, &u
);
3358 return log_notice_errno(r
, "Failed to load unit \"%s\", skipping deserialization: %m", name
);
3361 r
= unit_deserialize(u
, f
, fds
);
3365 return log_notice_errno(r
, "Failed to deserialize unit \"%s\", skipping: %m", name
);
3371 static int manager_deserialize_units(Manager
*m
, FILE *f
, FDSet
*fds
) {
3372 const char *unit_name
;
3376 _cleanup_free_
char *line
= NULL
;
3378 r
= read_line(f
, LONG_LINE_MAX
, &line
);
3380 return log_error_errno(r
, "Failed to read serialization line: %m");
3384 unit_name
= strstrip(line
);
3386 r
= manager_deserialize_one_unit(m
, unit_name
, f
, fds
);
3390 r
= unit_deserialize_skip(f
);
3399 usec_t
manager_get_watchdog(Manager
*m
, WatchdogType t
) {
3402 if (MANAGER_IS_USER(m
))
3403 return USEC_INFINITY
;
3405 if (timestamp_is_set(m
->watchdog_overridden
[t
]))
3406 return m
->watchdog_overridden
[t
];
3408 return m
->watchdog
[t
];
3411 void manager_set_watchdog(Manager
*m
, WatchdogType t
, usec_t timeout
) {
3416 if (MANAGER_IS_USER(m
))
3419 if (m
->watchdog
[t
] == timeout
)
3422 if (t
== WATCHDOG_RUNTIME
)
3423 if (!timestamp_is_set(m
->watchdog_overridden
[WATCHDOG_RUNTIME
])) {
3424 if (timestamp_is_set(timeout
))
3425 r
= watchdog_set_timeout(&timeout
);
3427 watchdog_close(true);
3431 m
->watchdog
[t
] = timeout
;
3434 int manager_override_watchdog(Manager
*m
, WatchdogType t
, usec_t timeout
) {
3439 if (MANAGER_IS_USER(m
))
3442 if (m
->watchdog_overridden
[t
] == timeout
)
3445 if (t
== WATCHDOG_RUNTIME
) {
3448 p
= timestamp_is_set(timeout
) ? &timeout
: &m
->watchdog
[t
];
3449 if (timestamp_is_set(*p
))
3450 r
= watchdog_set_timeout(p
);
3452 watchdog_close(true);
3456 m
->watchdog_overridden
[t
] = timeout
;
3461 static void manager_deserialize_uid_refs_one_internal(
3464 const char *value
) {
3474 r
= parse_uid(value
, &uid
);
3475 if (r
< 0 || uid
== 0) {
3476 log_debug("Unable to parse UID reference serialization: " UID_FMT
, uid
);
3480 r
= hashmap_ensure_allocated(uid_refs
, &trivial_hash_ops
);
3486 c
= PTR_TO_UINT32(hashmap_get(*uid_refs
, UID_TO_PTR(uid
)));
3487 if (c
& DESTROY_IPC_FLAG
)
3490 c
|= DESTROY_IPC_FLAG
;
3492 r
= hashmap_replace(*uid_refs
, UID_TO_PTR(uid
), UINT32_TO_PTR(c
));
3494 log_debug_errno(r
, "Failed to add UID reference entry: %m");
3499 static void manager_deserialize_uid_refs_one(Manager
*m
, const char *value
) {
3500 manager_deserialize_uid_refs_one_internal(m
, &m
->uid_refs
, value
);
3503 static void manager_deserialize_gid_refs_one(Manager
*m
, const char *value
) {
3504 manager_deserialize_uid_refs_one_internal(m
, &m
->gid_refs
, value
);
3507 int manager_deserialize(Manager
*m
, FILE *f
, FDSet
*fds
) {
3513 log_debug("Deserializing state...");
3515 /* If we are not in reload mode yet, enter it now. Not that this is recursive, a caller might already have
3516 * increased it to non-zero, which is why we just increase it by one here and down again at the end of this
3518 _cleanup_(manager_reloading_stopp
) _unused_ Manager
*reloading
= manager_reloading_start(m
);
3521 _cleanup_free_
char *line
= NULL
;
3522 const char *val
, *l
;
3524 r
= read_line(f
, LONG_LINE_MAX
, &line
);
3526 return log_error_errno(r
, "Failed to read serialization line: %m");
3531 if (isempty(l
)) /* end marker */
3534 if ((val
= startswith(l
, "current-job-id="))) {
3537 if (safe_atou32(val
, &id
) < 0)
3538 log_notice("Failed to parse current job id value '%s', ignoring.", val
);
3540 m
->current_job_id
= MAX(m
->current_job_id
, id
);
3542 } else if ((val
= startswith(l
, "n-installed-jobs="))) {
3545 if (safe_atou32(val
, &n
) < 0)
3546 log_notice("Failed to parse installed jobs counter '%s', ignoring.", val
);
3548 m
->n_installed_jobs
+= n
;
3550 } else if ((val
= startswith(l
, "n-failed-jobs="))) {
3553 if (safe_atou32(val
, &n
) < 0)
3554 log_notice("Failed to parse failed jobs counter '%s', ignoring.", val
);
3556 m
->n_failed_jobs
+= n
;
3558 } else if ((val
= startswith(l
, "taint-usr="))) {
3561 b
= parse_boolean(val
);
3563 log_notice("Failed to parse taint /usr flag '%s', ignoring.", val
);
3565 m
->taint_usr
= m
->taint_usr
|| b
;
3567 } else if ((val
= startswith(l
, "ready-sent="))) {
3570 b
= parse_boolean(val
);
3572 log_notice("Failed to parse ready-sent flag '%s', ignoring.", val
);
3574 m
->ready_sent
= m
->ready_sent
|| b
;
3576 } else if ((val
= startswith(l
, "taint-logged="))) {
3579 b
= parse_boolean(val
);
3581 log_notice("Failed to parse taint-logged flag '%s', ignoring.", val
);
3583 m
->taint_logged
= m
->taint_logged
|| b
;
3585 } else if ((val
= startswith(l
, "service-watchdogs="))) {
3588 b
= parse_boolean(val
);
3590 log_notice("Failed to parse service-watchdogs flag '%s', ignoring.", val
);
3592 m
->service_watchdogs
= b
;
3594 } else if ((val
= startswith(l
, "honor-device-enumeration="))) {
3597 b
= parse_boolean(val
);
3599 log_notice("Failed to parse honor-device-enumeration flag '%s', ignoring.", val
);
3601 m
->honor_device_enumeration
= b
;
3603 } else if ((val
= startswith(l
, "show-status-overridden="))) {
3606 s
= show_status_from_string(val
);
3608 log_notice("Failed to parse show-status-overridden flag '%s', ignoring.", val
);
3610 manager_override_show_status(m
, s
, "deserialize");
3612 } else if ((val
= startswith(l
, "log-level-override="))) {
3615 level
= log_level_from_string(val
);
3617 log_notice("Failed to parse log-level-override value '%s', ignoring.", val
);
3619 manager_override_log_level(m
, level
);
3621 } else if ((val
= startswith(l
, "log-target-override="))) {
3624 target
= log_target_from_string(val
);
3626 log_notice("Failed to parse log-target-override value '%s', ignoring.", val
);
3628 manager_override_log_target(m
, target
);
3630 } else if ((val
= startswith(l
, "runtime-watchdog-overridden="))) {
3633 if (deserialize_usec(val
, &t
) < 0)
3634 log_notice("Failed to parse runtime-watchdog-overridden value '%s', ignoring.", val
);
3636 manager_override_watchdog(m
, WATCHDOG_RUNTIME
, t
);
3638 } else if ((val
= startswith(l
, "reboot-watchdog-overridden="))) {
3641 if (deserialize_usec(val
, &t
) < 0)
3642 log_notice("Failed to parse reboot-watchdog-overridden value '%s', ignoring.", val
);
3644 manager_override_watchdog(m
, WATCHDOG_REBOOT
, t
);
3646 } else if ((val
= startswith(l
, "kexec-watchdog-overridden="))) {
3649 if (deserialize_usec(val
, &t
) < 0)
3650 log_notice("Failed to parse kexec-watchdog-overridden value '%s', ignoring.", val
);
3652 manager_override_watchdog(m
, WATCHDOG_KEXEC
, t
);
3654 } else if (startswith(l
, "env=")) {
3655 r
= deserialize_environment(l
+ 4, &m
->client_environment
);
3657 log_notice_errno(r
, "Failed to parse environment entry: \"%s\", ignoring: %m", l
);
3659 } else if ((val
= startswith(l
, "notify-fd="))) {
3662 if (safe_atoi(val
, &fd
) < 0 || fd
< 0 || !fdset_contains(fds
, fd
))
3663 log_notice("Failed to parse notify fd, ignoring: \"%s\"", val
);
3665 m
->notify_event_source
= sd_event_source_unref(m
->notify_event_source
);
3666 safe_close(m
->notify_fd
);
3667 m
->notify_fd
= fdset_remove(fds
, fd
);
3670 } else if ((val
= startswith(l
, "notify-socket="))) {
3671 r
= free_and_strdup(&m
->notify_socket
, val
);
3675 } else if ((val
= startswith(l
, "cgroups-agent-fd="))) {
3678 if (safe_atoi(val
, &fd
) < 0 || fd
< 0 || !fdset_contains(fds
, fd
))
3679 log_notice("Failed to parse cgroups agent fd, ignoring.: %s", val
);
3681 m
->cgroups_agent_event_source
= sd_event_source_unref(m
->cgroups_agent_event_source
);
3682 safe_close(m
->cgroups_agent_fd
);
3683 m
->cgroups_agent_fd
= fdset_remove(fds
, fd
);
3686 } else if ((val
= startswith(l
, "user-lookup="))) {
3689 if (sscanf(val
, "%i %i", &fd0
, &fd1
) != 2 || fd0
< 0 || fd1
< 0 || fd0
== fd1
|| !fdset_contains(fds
, fd0
) || !fdset_contains(fds
, fd1
))
3690 log_notice("Failed to parse user lookup fd, ignoring: %s", val
);
3692 m
->user_lookup_event_source
= sd_event_source_unref(m
->user_lookup_event_source
);
3693 safe_close_pair(m
->user_lookup_fds
);
3694 m
->user_lookup_fds
[0] = fdset_remove(fds
, fd0
);
3695 m
->user_lookup_fds
[1] = fdset_remove(fds
, fd1
);
3698 } else if ((val
= startswith(l
, "dynamic-user=")))
3699 dynamic_user_deserialize_one(m
, val
, fds
);
3700 else if ((val
= startswith(l
, "destroy-ipc-uid=")))
3701 manager_deserialize_uid_refs_one(m
, val
);
3702 else if ((val
= startswith(l
, "destroy-ipc-gid=")))
3703 manager_deserialize_gid_refs_one(m
, val
);
3704 else if ((val
= startswith(l
, "exec-runtime=")))
3705 (void) exec_runtime_deserialize_one(m
, val
, fds
);
3706 else if ((val
= startswith(l
, "subscribed="))) {
3708 if (strv_extend(&m
->deserialized_subscribed
, val
) < 0)
3714 for (q
= 0; q
< _MANAGER_TIMESTAMP_MAX
; q
++) {
3715 val
= startswith(l
, manager_timestamp_to_string(q
));
3719 val
= startswith(val
, "-timestamp=");
3724 if (q
< _MANAGER_TIMESTAMP_MAX
) /* found it */
3725 (void) deserialize_dual_timestamp(val
, m
->timestamps
+ q
);
3726 else if (!startswith(l
, "kdbus-fd=")) /* ignore kdbus */
3727 log_notice("Unknown serialization item '%s', ignoring.", l
);
3731 return manager_deserialize_units(m
, f
, fds
);
3734 int manager_reload(Manager
*m
) {
3735 _cleanup_(manager_reloading_stopp
) Manager
*reloading
= NULL
;
3736 _cleanup_fdset_free_ FDSet
*fds
= NULL
;
3737 _cleanup_fclose_
FILE *f
= NULL
;
3742 r
= manager_open_serialization(m
, &f
);
3744 return log_error_errno(r
, "Failed to create serialization file: %m");
3750 /* We are officially in reload mode from here on. */
3751 reloading
= manager_reloading_start(m
);
3753 r
= manager_serialize(m
, f
, fds
, false);
3757 if (fseeko(f
, 0, SEEK_SET
) < 0)
3758 return log_error_errno(errno
, "Failed to seek to beginning of serialization: %m");
3760 /* 💀 This is the point of no return, from here on there is no way back. 💀 */
3763 bus_manager_send_reloading(m
, true);
3765 /* Start by flushing out all jobs and units, all generated units, all runtime environments, all dynamic users
3766 * and everything else that is worth flushing out. We'll get it all back from the serialization — if we need
3769 manager_clear_jobs_and_units(m
);
3770 lookup_paths_flush_generator(&m
->lookup_paths
);
3771 lookup_paths_free(&m
->lookup_paths
);
3772 exec_runtime_vacuum(m
);
3773 dynamic_user_vacuum(m
, false);
3774 m
->uid_refs
= hashmap_free(m
->uid_refs
);
3775 m
->gid_refs
= hashmap_free(m
->gid_refs
);
3777 r
= lookup_paths_init(&m
->lookup_paths
, m
->unit_file_scope
, 0, NULL
);
3779 log_warning_errno(r
, "Failed to initialize path lookup table, ignoring: %m");
3781 (void) manager_run_environment_generators(m
);
3782 (void) manager_run_generators(m
);
3784 lookup_paths_log(&m
->lookup_paths
);
3786 /* We flushed out generated files, for which we don't watch mtime, so we should flush the old map. */
3787 manager_free_unit_name_maps(m
);
3789 /* First, enumerate what we can from kernel and suchlike */
3790 manager_enumerate_perpetual(m
);
3791 manager_enumerate(m
);
3793 /* Second, deserialize our stored data */
3794 r
= manager_deserialize(m
, f
, fds
);
3796 log_warning_errno(r
, "Deserialization failed, proceeding anyway: %m");
3798 /* We don't need the serialization anymore */
3801 /* Re-register notify_fd as event source, and set up other sockets/communication channels we might need */
3802 (void) manager_setup_notify(m
);
3803 (void) manager_setup_cgroups_agent(m
);
3804 (void) manager_setup_user_lookup_fd(m
);
3806 /* Third, fire things up! */
3807 manager_coldplug(m
);
3809 /* Clean up runtime objects no longer referenced */
3812 /* Consider the reload process complete now. */
3813 assert(m
->n_reloading
> 0);
3816 /* On manager reloading, device tag data should exists, thus, we should honor the results of device
3817 * enumeration. The flag should be always set correctly by the serialized data, but it may fail. So,
3818 * let's always set the flag here for safety. */
3819 m
->honor_device_enumeration
= true;
3823 m
->send_reloading_done
= true;
3827 void manager_reset_failed(Manager
*m
) {
3833 HASHMAP_FOREACH(u
, m
->units
, i
)
3834 unit_reset_failed(u
);
3837 bool manager_unit_inactive_or_pending(Manager
*m
, const char *name
) {
3843 /* Returns true if the unit is inactive or going down */
3844 u
= manager_get_unit(m
, name
);
3848 return unit_inactive_or_pending(u
);
3851 static void log_taint_string(Manager
*m
) {
3852 _cleanup_free_
char *taint
= NULL
;
3856 if (MANAGER_IS_USER(m
) || m
->taint_logged
)
3859 m
->taint_logged
= true; /* only check for taint once */
3861 taint
= manager_taint_string(m
);
3865 log_struct(LOG_NOTICE
,
3866 LOG_MESSAGE("System is tainted: %s", taint
),
3868 "MESSAGE_ID=" SD_MESSAGE_TAINTED_STR
);
3871 static void manager_notify_finished(Manager
*m
) {
3872 char userspace
[FORMAT_TIMESPAN_MAX
], initrd
[FORMAT_TIMESPAN_MAX
], kernel
[FORMAT_TIMESPAN_MAX
], sum
[FORMAT_TIMESPAN_MAX
];
3873 usec_t firmware_usec
, loader_usec
, kernel_usec
, initrd_usec
, userspace_usec
, total_usec
;
3875 if (MANAGER_IS_TEST_RUN(m
))
3878 if (MANAGER_IS_SYSTEM(m
) && detect_container() <= 0) {
3879 char ts
[FORMAT_TIMESPAN_MAX
];
3880 char buf
[FORMAT_TIMESPAN_MAX
+ STRLEN(" (firmware) + ") + FORMAT_TIMESPAN_MAX
+ STRLEN(" (loader) + ")]
3883 size_t size
= sizeof buf
;
3885 /* Note that MANAGER_TIMESTAMP_KERNEL's monotonic value is always at 0, and
3886 * MANAGER_TIMESTAMP_FIRMWARE's and MANAGER_TIMESTAMP_LOADER's monotonic value should be considered
3887 * negative values. */
3889 firmware_usec
= m
->timestamps
[MANAGER_TIMESTAMP_FIRMWARE
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_LOADER
].monotonic
;
3890 loader_usec
= m
->timestamps
[MANAGER_TIMESTAMP_LOADER
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_KERNEL
].monotonic
;
3891 userspace_usec
= m
->timestamps
[MANAGER_TIMESTAMP_FINISH
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_USERSPACE
].monotonic
;
3892 total_usec
= m
->timestamps
[MANAGER_TIMESTAMP_FIRMWARE
].monotonic
+ m
->timestamps
[MANAGER_TIMESTAMP_FINISH
].monotonic
;
3894 if (firmware_usec
> 0)
3895 size
= strpcpyf(&p
, size
, "%s (firmware) + ", format_timespan(ts
, sizeof(ts
), firmware_usec
, USEC_PER_MSEC
));
3896 if (loader_usec
> 0)
3897 size
= strpcpyf(&p
, size
, "%s (loader) + ", format_timespan(ts
, sizeof(ts
), loader_usec
, USEC_PER_MSEC
));
3899 if (dual_timestamp_is_set(&m
->timestamps
[MANAGER_TIMESTAMP_INITRD
])) {
3901 /* The initrd case on bare-metal*/
3902 kernel_usec
= m
->timestamps
[MANAGER_TIMESTAMP_INITRD
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_KERNEL
].monotonic
;
3903 initrd_usec
= m
->timestamps
[MANAGER_TIMESTAMP_USERSPACE
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_INITRD
].monotonic
;
3905 log_struct(LOG_INFO
,
3906 "MESSAGE_ID=" SD_MESSAGE_STARTUP_FINISHED_STR
,
3907 "KERNEL_USEC="USEC_FMT
, kernel_usec
,
3908 "INITRD_USEC="USEC_FMT
, initrd_usec
,
3909 "USERSPACE_USEC="USEC_FMT
, userspace_usec
,
3910 LOG_MESSAGE("Startup finished in %s%s (kernel) + %s (initrd) + %s (userspace) = %s.",
3912 format_timespan(kernel
, sizeof(kernel
), kernel_usec
, USEC_PER_MSEC
),
3913 format_timespan(initrd
, sizeof(initrd
), initrd_usec
, USEC_PER_MSEC
),
3914 format_timespan(userspace
, sizeof(userspace
), userspace_usec
, USEC_PER_MSEC
),
3915 format_timespan(sum
, sizeof(sum
), total_usec
, USEC_PER_MSEC
)));
3917 /* The initrd-less case on bare-metal*/
3919 kernel_usec
= m
->timestamps
[MANAGER_TIMESTAMP_USERSPACE
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_KERNEL
].monotonic
;
3922 log_struct(LOG_INFO
,
3923 "MESSAGE_ID=" SD_MESSAGE_STARTUP_FINISHED_STR
,
3924 "KERNEL_USEC="USEC_FMT
, kernel_usec
,
3925 "USERSPACE_USEC="USEC_FMT
, userspace_usec
,
3926 LOG_MESSAGE("Startup finished in %s%s (kernel) + %s (userspace) = %s.",
3928 format_timespan(kernel
, sizeof(kernel
), kernel_usec
, USEC_PER_MSEC
),
3929 format_timespan(userspace
, sizeof(userspace
), userspace_usec
, USEC_PER_MSEC
),
3930 format_timespan(sum
, sizeof(sum
), total_usec
, USEC_PER_MSEC
)));
3933 /* The container and --user case */
3934 firmware_usec
= loader_usec
= initrd_usec
= kernel_usec
= 0;
3935 total_usec
= userspace_usec
= m
->timestamps
[MANAGER_TIMESTAMP_FINISH
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_USERSPACE
].monotonic
;
3937 log_struct(LOG_INFO
,
3938 "MESSAGE_ID=" SD_MESSAGE_USER_STARTUP_FINISHED_STR
,
3939 "USERSPACE_USEC="USEC_FMT
, userspace_usec
,
3940 LOG_MESSAGE("Startup finished in %s.",
3941 format_timespan(sum
, sizeof(sum
), total_usec
, USEC_PER_MSEC
)));
3944 bus_manager_send_finished(m
, firmware_usec
, loader_usec
, kernel_usec
, initrd_usec
, userspace_usec
, total_usec
);
3947 m
->ready_sent
? "STATUS=Startup finished in %s."
3949 "STATUS=Startup finished in %s.",
3950 format_timespan(sum
, sizeof(sum
), total_usec
, USEC_PER_MSEC
));
3951 m
->ready_sent
= true;
3953 log_taint_string(m
);
3956 static void manager_send_ready(Manager
*m
) {
3959 /* We send READY=1 on reaching basic.target only when running in --user mode. */
3960 if (!MANAGER_IS_USER(m
) || m
->ready_sent
)
3963 m
->ready_sent
= true;
3967 "STATUS=Reached " SPECIAL_BASIC_TARGET
".");
3970 static void manager_check_basic_target(Manager
*m
) {
3975 /* Small shortcut */
3976 if (m
->ready_sent
&& m
->taint_logged
)
3979 u
= manager_get_unit(m
, SPECIAL_BASIC_TARGET
);
3980 if (!u
|| !UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u
)))
3983 /* For user managers, send out READY=1 as soon as we reach basic.target */
3984 manager_send_ready(m
);
3986 /* Log the taint string as soon as we reach basic.target */
3987 log_taint_string(m
);
3990 void manager_check_finished(Manager
*m
) {
3993 if (MANAGER_IS_RELOADING(m
))
3996 /* Verify that we have entered the event loop already, and not left it again. */
3997 if (!MANAGER_IS_RUNNING(m
))
4000 manager_check_basic_target(m
);
4002 if (hashmap_size(m
->jobs
) > 0) {
4003 if (m
->jobs_in_progress_event_source
)
4004 /* Ignore any failure, this is only for feedback */
4005 (void) sd_event_source_set_time(m
->jobs_in_progress_event_source
,
4006 manager_watch_jobs_next_time(m
));
4010 /* The jobs hashmap tends to grow a lot during boot, and then it's not reused until shutdown. Let's
4011 kill the hashmap if it is relatively large. */
4012 if (hashmap_buckets(m
->jobs
) > hashmap_size(m
->units
) / 10)
4013 m
->jobs
= hashmap_free(m
->jobs
);
4015 manager_flip_auto_status(m
, false, "boot finished");
4017 /* Notify Type=idle units that we are done now */
4018 manager_close_idle_pipe(m
);
4020 /* Turn off confirm spawn now */
4021 m
->confirm_spawn
= NULL
;
4023 /* No need to update ask password status when we're going non-interactive */
4024 manager_close_ask_password(m
);
4026 /* This is no longer the first boot */
4027 manager_set_first_boot(m
, false);
4029 if (MANAGER_IS_FINISHED(m
))
4032 dual_timestamp_get(m
->timestamps
+ MANAGER_TIMESTAMP_FINISH
);
4034 manager_notify_finished(m
);
4036 manager_invalidate_startup_units(m
);
4039 static bool generator_path_any(const char* const* paths
) {
4043 /* Optimize by skipping the whole process by not creating output directories
4044 * if no generators are found. */
4045 STRV_FOREACH(path
, (char**) paths
)
4046 if (access(*path
, F_OK
) == 0)
4048 else if (errno
!= ENOENT
)
4049 log_warning_errno(errno
, "Failed to open generator directory %s: %m", *path
);
4054 static int manager_run_environment_generators(Manager
*m
) {
4055 char **tmp
= NULL
; /* this is only used in the forked process, no cleanup here */
4056 _cleanup_strv_free_
char **paths
= NULL
;
4058 [STDOUT_GENERATE
] = &tmp
,
4059 [STDOUT_COLLECT
] = &tmp
,
4060 [STDOUT_CONSUME
] = &m
->transient_environment
,
4064 if (MANAGER_IS_TEST_RUN(m
) && !(m
->test_run_flags
& MANAGER_TEST_RUN_ENV_GENERATORS
))
4067 paths
= env_generator_binary_paths(MANAGER_IS_SYSTEM(m
));
4071 if (!generator_path_any((const char* const*) paths
))
4074 RUN_WITH_UMASK(0022)
4075 r
= execute_directories((const char* const*) paths
, DEFAULT_TIMEOUT_USEC
, gather_environment
,
4076 args
, NULL
, m
->transient_environment
, EXEC_DIR_PARALLEL
| EXEC_DIR_IGNORE_ERRORS
);
4080 static int manager_run_generators(Manager
*m
) {
4081 _cleanup_strv_free_
char **paths
= NULL
;
4082 const char *argv
[5];
4087 if (MANAGER_IS_TEST_RUN(m
) && !(m
->test_run_flags
& MANAGER_TEST_RUN_GENERATORS
))
4090 paths
= generator_binary_paths(m
->unit_file_scope
);
4094 if (!generator_path_any((const char* const*) paths
))
4097 r
= lookup_paths_mkdir_generator(&m
->lookup_paths
);
4099 log_error_errno(r
, "Failed to create generator directories: %m");
4103 argv
[0] = NULL
; /* Leave this empty, execute_directory() will fill something in */
4104 argv
[1] = m
->lookup_paths
.generator
;
4105 argv
[2] = m
->lookup_paths
.generator_early
;
4106 argv
[3] = m
->lookup_paths
.generator_late
;
4109 RUN_WITH_UMASK(0022)
4110 (void) execute_directories((const char* const*) paths
, DEFAULT_TIMEOUT_USEC
, NULL
, NULL
,
4111 (char**) argv
, m
->transient_environment
, EXEC_DIR_PARALLEL
| EXEC_DIR_IGNORE_ERRORS
);
4116 lookup_paths_trim_generator(&m
->lookup_paths
);
4120 int manager_transient_environment_add(Manager
*m
, char **plus
) {
4125 if (strv_isempty(plus
))
4128 a
= strv_env_merge(2, m
->transient_environment
, plus
);
4132 sanitize_environment(a
);
4134 return strv_free_and_replace(m
->transient_environment
, a
);
4137 int manager_client_environment_modify(
4142 char **a
= NULL
, **b
= NULL
, **l
;
4146 if (strv_isempty(minus
) && strv_isempty(plus
))
4149 l
= m
->client_environment
;
4151 if (!strv_isempty(minus
)) {
4152 a
= strv_env_delete(l
, 1, minus
);
4159 if (!strv_isempty(plus
)) {
4160 b
= strv_env_merge(2, l
, plus
);
4169 if (m
->client_environment
!= l
)
4170 strv_free(m
->client_environment
);
4177 m
->client_environment
= sanitize_environment(l
);
4181 int manager_get_effective_environment(Manager
*m
, char ***ret
) {
4187 l
= strv_env_merge(2, m
->transient_environment
, m
->client_environment
);
4195 int manager_set_default_rlimits(Manager
*m
, struct rlimit
**default_rlimit
) {
4200 for (i
= 0; i
< _RLIMIT_MAX
; i
++) {
4201 m
->rlimit
[i
] = mfree(m
->rlimit
[i
]);
4203 if (!default_rlimit
[i
])
4206 m
->rlimit
[i
] = newdup(struct rlimit
, default_rlimit
[i
], 1);
4214 void manager_recheck_dbus(Manager
*m
) {
4217 /* Connects to the bus if the dbus service and socket are running. If we are running in user mode this is all
4218 * it does. In system mode we'll also connect to the system bus (which will most likely just reuse the
4219 * connection of the API bus). That's because the system bus after all runs as service of the system instance,
4220 * while in the user instance we can assume it's already there. */
4222 if (MANAGER_IS_RELOADING(m
))
4223 return; /* don't check while we are reloading… */
4225 if (manager_dbus_is_running(m
, false)) {
4226 (void) bus_init_api(m
);
4228 if (MANAGER_IS_SYSTEM(m
))
4229 (void) bus_init_system(m
);
4231 (void) bus_done_api(m
);
4233 if (MANAGER_IS_SYSTEM(m
))
4234 (void) bus_done_system(m
);
4238 static bool manager_journal_is_running(Manager
*m
) {
4243 if (MANAGER_IS_TEST_RUN(m
))
4246 /* If we are the user manager we can safely assume that the journal is up */
4247 if (!MANAGER_IS_SYSTEM(m
))
4250 /* Check that the socket is not only up, but in RUNNING state */
4251 u
= manager_get_unit(m
, SPECIAL_JOURNALD_SOCKET
);
4254 if (SOCKET(u
)->state
!= SOCKET_RUNNING
)
4257 /* Similar, check if the daemon itself is fully up, too */
4258 u
= manager_get_unit(m
, SPECIAL_JOURNALD_SERVICE
);
4261 if (!IN_SET(SERVICE(u
)->state
, SERVICE_RELOAD
, SERVICE_RUNNING
))
4267 void disable_printk_ratelimit(void) {
4268 /* Disable kernel's printk ratelimit.
4270 * Logging to /dev/kmsg is most useful during early boot and shutdown, where normal logging
4271 * mechanisms are not available. The semantics of this sysctl are such that any kernel command-line
4272 * setting takes precedence. */
4275 r
= sysctl_write("kernel/printk_devkmsg", "on");
4277 log_debug_errno(r
, "Failed to set sysctl kernel.printk_devkmsg=on: %m");
4280 void manager_recheck_journal(Manager
*m
) {
4284 /* Don't bother with this unless we are in the special situation of being PID 1 */
4285 if (getpid_cached() != 1)
4288 /* Don't check this while we are reloading, things might still change */
4289 if (MANAGER_IS_RELOADING(m
))
4292 /* The journal is fully and entirely up? If so, let's permit logging to it, if that's configured. If the
4293 * journal is down, don't ever log to it, otherwise we might end up deadlocking ourselves as we might trigger
4294 * an activation ourselves we can't fulfill. */
4295 log_set_prohibit_ipc(!manager_journal_is_running(m
));
4299 static ShowStatus
manager_get_show_status(Manager
*m
) {
4302 if (MANAGER_IS_USER(m
))
4303 return _SHOW_STATUS_INVALID
;
4305 if (m
->show_status_overridden
!= _SHOW_STATUS_INVALID
)
4306 return m
->show_status_overridden
;
4308 return m
->show_status
;
4311 bool manager_get_show_status_on(Manager
*m
) {
4314 return show_status_on(manager_get_show_status(m
));
4317 static void set_show_status_marker(bool b
) {
4319 (void) touch("/run/systemd/show-status");
4321 (void) unlink("/run/systemd/show-status");
4324 void manager_set_show_status(Manager
*m
, ShowStatus mode
, const char *reason
) {
4327 assert(mode
>= 0 && mode
< _SHOW_STATUS_MAX
);
4329 if (MANAGER_IS_USER(m
))
4332 if (mode
== m
->show_status
)
4335 if (m
->show_status_overridden
== _SHOW_STATUS_INVALID
) {
4338 enabled
= show_status_on(mode
);
4339 log_debug("%s (%s) showing of status (%s).",
4340 enabled
? "Enabling" : "Disabling",
4341 strna(show_status_to_string(mode
)),
4344 set_show_status_marker(enabled
);
4347 m
->show_status
= mode
;
4350 void manager_override_show_status(Manager
*m
, ShowStatus mode
, const char *reason
) {
4352 assert(mode
< _SHOW_STATUS_MAX
);
4354 if (MANAGER_IS_USER(m
))
4357 if (mode
== m
->show_status_overridden
)
4360 m
->show_status_overridden
= mode
;
4362 if (mode
== _SHOW_STATUS_INVALID
)
4363 mode
= m
->show_status
;
4365 log_debug("%s (%s) showing of status (%s).",
4366 m
->show_status_overridden
!= _SHOW_STATUS_INVALID
? "Overriding" : "Restoring",
4367 strna(show_status_to_string(mode
)),
4370 set_show_status_marker(show_status_on(mode
));
4373 const char *manager_get_confirm_spawn(Manager
*m
) {
4374 static int last_errno
= 0;
4380 /* Here's the deal: we want to test the validity of the console but don't want
4381 * PID1 to go through the whole console process which might block. But we also
4382 * want to warn the user only once if something is wrong with the console so we
4383 * cannot do the sanity checks after spawning our children. So here we simply do
4384 * really basic tests to hopefully trap common errors.
4386 * If the console suddenly disappear at the time our children will really it
4387 * then they will simply fail to acquire it and a positive answer will be
4388 * assumed. New children will fall back to /dev/console though.
4390 * Note: TTYs are devices that can come and go any time, and frequently aren't
4391 * available yet during early boot (consider a USB rs232 dongle...). If for any
4392 * reason the configured console is not ready, we fall back to the default
4395 if (!m
->confirm_spawn
|| path_equal(m
->confirm_spawn
, "/dev/console"))
4396 return m
->confirm_spawn
;
4398 if (stat(m
->confirm_spawn
, &st
) < 0) {
4403 if (!S_ISCHR(st
.st_mode
)) {
4409 return m
->confirm_spawn
;
4412 if (last_errno
!= r
)
4413 last_errno
= log_warning_errno(r
, "Failed to open %s, using default console: %m", m
->confirm_spawn
);
4415 return "/dev/console";
4418 void manager_set_first_boot(Manager
*m
, bool b
) {
4421 if (!MANAGER_IS_SYSTEM(m
))
4424 if (m
->first_boot
!= (int) b
) {
4426 (void) touch("/run/systemd/first-boot");
4428 (void) unlink("/run/systemd/first-boot");
4434 void manager_disable_confirm_spawn(void) {
4435 (void) touch("/run/systemd/confirm_spawn_disabled");
4438 bool manager_is_confirm_spawn_disabled(Manager
*m
) {
4439 if (!m
->confirm_spawn
)
4442 return access("/run/systemd/confirm_spawn_disabled", F_OK
) >= 0;
4445 static bool manager_should_show_status(Manager
*m
, StatusType type
) {
4448 if (!MANAGER_IS_SYSTEM(m
))
4451 if (m
->no_console_output
)
4454 if (!IN_SET(manager_state(m
), MANAGER_INITIALIZING
, MANAGER_STARTING
, MANAGER_STOPPING
))
4457 /* If we cannot find out the status properly, just proceed. */
4458 if (type
!= STATUS_TYPE_EMERGENCY
&& manager_check_ask_password(m
) > 0)
4461 if (type
== STATUS_TYPE_NOTICE
&& m
->show_status
!= SHOW_STATUS_NO
)
4464 return manager_get_show_status_on(m
);
4467 void manager_status_printf(Manager
*m
, StatusType type
, const char *status
, const char *format
, ...) {
4470 /* If m is NULL, assume we're after shutdown and let the messages through. */
4472 if (m
&& !manager_should_show_status(m
, type
))
4475 /* XXX We should totally drop the check for ephemeral here
4476 * and thus effectively make 'Type=idle' pointless. */
4477 if (type
== STATUS_TYPE_EPHEMERAL
&& m
&& m
->n_on_console
> 0)
4480 va_start(ap
, format
);
4481 status_vprintf(status
, SHOW_STATUS_ELLIPSIZE
|(type
== STATUS_TYPE_EPHEMERAL
? SHOW_STATUS_EPHEMERAL
: 0), format
, ap
);
4485 Set
*manager_get_units_requiring_mounts_for(Manager
*m
, const char *path
) {
4486 char p
[strlen(path
)+1];
4492 path_simplify(p
, false);
4494 return hashmap_get(m
->units_requiring_mounts_for
, streq(p
, "/") ? "" : p
);
4497 int manager_update_failed_units(Manager
*m
, Unit
*u
, bool failed
) {
4502 assert(u
->manager
== m
);
4504 size
= set_size(m
->failed_units
);
4507 r
= set_ensure_put(&m
->failed_units
, NULL
, u
);
4511 (void) set_remove(m
->failed_units
, u
);
4513 if (set_size(m
->failed_units
) != size
)
4514 bus_manager_send_change_signal(m
);
4519 ManagerState
manager_state(Manager
*m
) {
4524 /* Is the special shutdown target active or queued? If so, we are in shutdown state */
4525 u
= manager_get_unit(m
, SPECIAL_SHUTDOWN_TARGET
);
4526 if (u
&& unit_active_or_pending(u
))
4527 return MANAGER_STOPPING
;
4529 /* Did we ever finish booting? If not then we are still starting up */
4530 if (!MANAGER_IS_FINISHED(m
)) {
4532 u
= manager_get_unit(m
, SPECIAL_BASIC_TARGET
);
4533 if (!u
|| !UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u
)))
4534 return MANAGER_INITIALIZING
;
4536 return MANAGER_STARTING
;
4539 if (MANAGER_IS_SYSTEM(m
)) {
4540 /* Are the rescue or emergency targets active or queued? If so we are in maintenance state */
4541 u
= manager_get_unit(m
, SPECIAL_RESCUE_TARGET
);
4542 if (u
&& unit_active_or_pending(u
))
4543 return MANAGER_MAINTENANCE
;
4545 u
= manager_get_unit(m
, SPECIAL_EMERGENCY_TARGET
);
4546 if (u
&& unit_active_or_pending(u
))
4547 return MANAGER_MAINTENANCE
;
4550 /* Are there any failed units? If so, we are in degraded mode */
4551 if (set_size(m
->failed_units
) > 0)
4552 return MANAGER_DEGRADED
;
4554 return MANAGER_RUNNING
;
4557 static void manager_unref_uid_internal(
4562 int (*_clean_ipc
)(uid_t uid
)) {
4568 assert(uid_is_valid(uid
));
4571 /* A generic implementation, covering both manager_unref_uid() and manager_unref_gid(), under the assumption
4572 * that uid_t and gid_t are actually defined the same way, with the same validity rules.
4574 * We store a hashmap where the UID/GID is they key and the value is a 32bit reference counter, whose highest
4575 * bit is used as flag for marking UIDs/GIDs whose IPC objects to remove when the last reference to the UID/GID
4576 * is dropped. The flag is set to on, once at least one reference from a unit where RemoveIPC= is set is added
4577 * on a UID/GID. It is reset when the UID's/GID's reference counter drops to 0 again. */
4579 assert_cc(sizeof(uid_t
) == sizeof(gid_t
));
4580 assert_cc(UID_INVALID
== (uid_t
) GID_INVALID
);
4582 if (uid
== 0) /* We don't keep track of root, and will never destroy it */
4585 c
= PTR_TO_UINT32(hashmap_get(*uid_refs
, UID_TO_PTR(uid
)));
4587 n
= c
& ~DESTROY_IPC_FLAG
;
4591 if (destroy_now
&& n
== 0) {
4592 hashmap_remove(*uid_refs
, UID_TO_PTR(uid
));
4594 if (c
& DESTROY_IPC_FLAG
) {
4595 log_debug("%s " UID_FMT
" is no longer referenced, cleaning up its IPC.",
4596 _clean_ipc
== clean_ipc_by_uid
? "UID" : "GID",
4598 (void) _clean_ipc(uid
);
4601 c
= n
| (c
& DESTROY_IPC_FLAG
);
4602 assert_se(hashmap_update(*uid_refs
, UID_TO_PTR(uid
), UINT32_TO_PTR(c
)) >= 0);
4606 void manager_unref_uid(Manager
*m
, uid_t uid
, bool destroy_now
) {
4607 manager_unref_uid_internal(m
, &m
->uid_refs
, uid
, destroy_now
, clean_ipc_by_uid
);
4610 void manager_unref_gid(Manager
*m
, gid_t gid
, bool destroy_now
) {
4611 manager_unref_uid_internal(m
, &m
->gid_refs
, (uid_t
) gid
, destroy_now
, clean_ipc_by_gid
);
4614 static int manager_ref_uid_internal(
4625 assert(uid_is_valid(uid
));
4627 /* A generic implementation, covering both manager_ref_uid() and manager_ref_gid(), under the assumption
4628 * that uid_t and gid_t are actually defined the same way, with the same validity rules. */
4630 assert_cc(sizeof(uid_t
) == sizeof(gid_t
));
4631 assert_cc(UID_INVALID
== (uid_t
) GID_INVALID
);
4633 if (uid
== 0) /* We don't keep track of root, and will never destroy it */
4636 r
= hashmap_ensure_allocated(uid_refs
, &trivial_hash_ops
);
4640 c
= PTR_TO_UINT32(hashmap_get(*uid_refs
, UID_TO_PTR(uid
)));
4642 n
= c
& ~DESTROY_IPC_FLAG
;
4645 if (n
& DESTROY_IPC_FLAG
) /* check for overflow */
4648 c
= n
| (c
& DESTROY_IPC_FLAG
) | (clean_ipc
? DESTROY_IPC_FLAG
: 0);
4650 return hashmap_replace(*uid_refs
, UID_TO_PTR(uid
), UINT32_TO_PTR(c
));
4653 int manager_ref_uid(Manager
*m
, uid_t uid
, bool clean_ipc
) {
4654 return manager_ref_uid_internal(m
, &m
->uid_refs
, uid
, clean_ipc
);
4657 int manager_ref_gid(Manager
*m
, gid_t gid
, bool clean_ipc
) {
4658 return manager_ref_uid_internal(m
, &m
->gid_refs
, (uid_t
) gid
, clean_ipc
);
4661 static void manager_vacuum_uid_refs_internal(
4664 int (*_clean_ipc
)(uid_t uid
)) {
4673 HASHMAP_FOREACH_KEY(p
, k
, *uid_refs
, i
) {
4677 uid
= PTR_TO_UID(k
);
4678 c
= PTR_TO_UINT32(p
);
4680 n
= c
& ~DESTROY_IPC_FLAG
;
4684 if (c
& DESTROY_IPC_FLAG
) {
4685 log_debug("Found unreferenced %s " UID_FMT
" after reload/reexec. Cleaning up.",
4686 _clean_ipc
== clean_ipc_by_uid
? "UID" : "GID",
4688 (void) _clean_ipc(uid
);
4691 assert_se(hashmap_remove(*uid_refs
, k
) == p
);
4695 static void manager_vacuum_uid_refs(Manager
*m
) {
4696 manager_vacuum_uid_refs_internal(m
, &m
->uid_refs
, clean_ipc_by_uid
);
4699 static void manager_vacuum_gid_refs(Manager
*m
) {
4700 manager_vacuum_uid_refs_internal(m
, &m
->gid_refs
, clean_ipc_by_gid
);
4703 static void manager_vacuum(Manager
*m
) {
4706 /* Release any dynamic users no longer referenced */
4707 dynamic_user_vacuum(m
, true);
4709 /* Release any references to UIDs/GIDs no longer referenced, and destroy any IPC owned by them */
4710 manager_vacuum_uid_refs(m
);
4711 manager_vacuum_gid_refs(m
);
4713 /* Release any runtimes no longer referenced */
4714 exec_runtime_vacuum(m
);
4717 int manager_dispatch_user_lookup_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
4721 char unit_name
[UNIT_NAME_MAX
+1];
4724 Manager
*m
= userdata
;
4732 /* Invoked whenever a child process succeeded resolving its user/group to use and sent us the resulting UID/GID
4733 * in a datagram. We parse the datagram here and pass it off to the unit, so that it can add a reference to the
4734 * UID/GID so that it can destroy the UID/GID's IPC objects when the reference counter drops to 0. */
4736 l
= recv(fd
, &buffer
, sizeof(buffer
), MSG_DONTWAIT
);
4738 if (IN_SET(errno
, EINTR
, EAGAIN
))
4741 return log_error_errno(errno
, "Failed to read from user lookup fd: %m");
4744 if ((size_t) l
<= offsetof(struct buffer
, unit_name
)) {
4745 log_warning("Received too short user lookup message, ignoring.");
4749 if ((size_t) l
> offsetof(struct buffer
, unit_name
) + UNIT_NAME_MAX
) {
4750 log_warning("Received too long user lookup message, ignoring.");
4754 if (!uid_is_valid(buffer
.uid
) && !gid_is_valid(buffer
.gid
)) {
4755 log_warning("Got user lookup message with invalid UID/GID pair, ignoring.");
4759 n
= (size_t) l
- offsetof(struct buffer
, unit_name
);
4760 if (memchr(buffer
.unit_name
, 0, n
)) {
4761 log_warning("Received lookup message with embedded NUL character, ignoring.");
4765 buffer
.unit_name
[n
] = 0;
4766 u
= manager_get_unit(m
, buffer
.unit_name
);
4768 log_debug("Got user lookup message but unit doesn't exist, ignoring.");
4772 log_unit_debug(u
, "User lookup succeeded: uid=" UID_FMT
" gid=" GID_FMT
, buffer
.uid
, buffer
.gid
);
4774 unit_notify_user_lookup(u
, buffer
.uid
, buffer
.gid
);
4778 char *manager_taint_string(Manager
*m
) {
4779 _cleanup_free_
char *destination
= NULL
, *overflowuid
= NULL
, *overflowgid
= NULL
;
4783 /* Returns a "taint string", e.g. "local-hwclock:var-run-bad".
4784 * Only things that are detected at runtime should be tagged
4785 * here. For stuff that is set during compilation, emit a warning
4786 * in the configuration phase. */
4790 buf
= new(char, sizeof("split-usr:"
4794 "overflowuid-not-65534:"
4795 "overflowgid-not-65534:"));
4803 e
= stpcpy(e
, "split-usr:");
4805 if (access("/proc/cgroups", F_OK
) < 0)
4806 e
= stpcpy(e
, "cgroups-missing:");
4808 if (clock_is_localtime(NULL
) > 0)
4809 e
= stpcpy(e
, "local-hwclock:");
4811 r
= readlink_malloc("/var/run", &destination
);
4812 if (r
< 0 || !PATH_IN_SET(destination
, "../run", "/run"))
4813 e
= stpcpy(e
, "var-run-bad:");
4815 r
= read_one_line_file("/proc/sys/kernel/overflowuid", &overflowuid
);
4816 if (r
>= 0 && !streq(overflowuid
, "65534"))
4817 e
= stpcpy(e
, "overflowuid-not-65534:");
4819 r
= read_one_line_file("/proc/sys/kernel/overflowgid", &overflowgid
);
4820 if (r
>= 0 && !streq(overflowgid
, "65534"))
4821 e
= stpcpy(e
, "overflowgid-not-65534:");
4823 /* remove the last ':' */
4830 void manager_ref_console(Manager
*m
) {
4836 void manager_unref_console(Manager
*m
) {
4838 assert(m
->n_on_console
> 0);
4841 if (m
->n_on_console
== 0)
4842 m
->no_console_output
= false; /* unset no_console_output flag, since the console is definitely free now */
4845 void manager_override_log_level(Manager
*m
, int level
) {
4846 _cleanup_free_
char *s
= NULL
;
4849 if (!m
->log_level_overridden
) {
4850 m
->original_log_level
= log_get_max_level();
4851 m
->log_level_overridden
= true;
4854 (void) log_level_to_string_alloc(level
, &s
);
4855 log_info("Setting log level to %s.", strna(s
));
4857 log_set_max_level(level
);
4860 void manager_restore_original_log_level(Manager
*m
) {
4861 _cleanup_free_
char *s
= NULL
;
4864 if (!m
->log_level_overridden
)
4867 (void) log_level_to_string_alloc(m
->original_log_level
, &s
);
4868 log_info("Restoring log level to original (%s).", strna(s
));
4870 log_set_max_level(m
->original_log_level
);
4871 m
->log_level_overridden
= false;
4874 void manager_override_log_target(Manager
*m
, LogTarget target
) {
4877 if (!m
->log_target_overridden
) {
4878 m
->original_log_target
= log_get_target();
4879 m
->log_target_overridden
= true;
4882 log_info("Setting log target to %s.", log_target_to_string(target
));
4883 log_set_target(target
);
4886 void manager_restore_original_log_target(Manager
*m
) {
4889 if (!m
->log_target_overridden
)
4892 log_info("Restoring log target to original %s.", log_target_to_string(m
->original_log_target
));
4894 log_set_target(m
->original_log_target
);
4895 m
->log_target_overridden
= false;
4898 ManagerTimestamp
manager_timestamp_initrd_mangle(ManagerTimestamp s
) {
4900 s
>= MANAGER_TIMESTAMP_SECURITY_START
&&
4901 s
<= MANAGER_TIMESTAMP_UNITS_LOAD_FINISH
)
4902 return s
- MANAGER_TIMESTAMP_SECURITY_START
+ MANAGER_TIMESTAMP_INITRD_SECURITY_START
;
4906 static const char *const manager_state_table
[_MANAGER_STATE_MAX
] = {
4907 [MANAGER_INITIALIZING
] = "initializing",
4908 [MANAGER_STARTING
] = "starting",
4909 [MANAGER_RUNNING
] = "running",
4910 [MANAGER_DEGRADED
] = "degraded",
4911 [MANAGER_MAINTENANCE
] = "maintenance",
4912 [MANAGER_STOPPING
] = "stopping",
4915 DEFINE_STRING_TABLE_LOOKUP(manager_state
, ManagerState
);
4917 static const char *const manager_timestamp_table
[_MANAGER_TIMESTAMP_MAX
] = {
4918 [MANAGER_TIMESTAMP_FIRMWARE
] = "firmware",
4919 [MANAGER_TIMESTAMP_LOADER
] = "loader",
4920 [MANAGER_TIMESTAMP_KERNEL
] = "kernel",
4921 [MANAGER_TIMESTAMP_INITRD
] = "initrd",
4922 [MANAGER_TIMESTAMP_USERSPACE
] = "userspace",
4923 [MANAGER_TIMESTAMP_FINISH
] = "finish",
4924 [MANAGER_TIMESTAMP_SECURITY_START
] = "security-start",
4925 [MANAGER_TIMESTAMP_SECURITY_FINISH
] = "security-finish",
4926 [MANAGER_TIMESTAMP_GENERATORS_START
] = "generators-start",
4927 [MANAGER_TIMESTAMP_GENERATORS_FINISH
] = "generators-finish",
4928 [MANAGER_TIMESTAMP_UNITS_LOAD_START
] = "units-load-start",
4929 [MANAGER_TIMESTAMP_UNITS_LOAD_FINISH
] = "units-load-finish",
4930 [MANAGER_TIMESTAMP_INITRD_SECURITY_START
] = "initrd-security-start",
4931 [MANAGER_TIMESTAMP_INITRD_SECURITY_FINISH
] = "initrd-security-finish",
4932 [MANAGER_TIMESTAMP_INITRD_GENERATORS_START
] = "initrd-generators-start",
4933 [MANAGER_TIMESTAMP_INITRD_GENERATORS_FINISH
] = "initrd-generators-finish",
4934 [MANAGER_TIMESTAMP_INITRD_UNITS_LOAD_START
] = "initrd-units-load-start",
4935 [MANAGER_TIMESTAMP_INITRD_UNITS_LOAD_FINISH
] = "initrd-units-load-finish",
4938 DEFINE_STRING_TABLE_LOOKUP(manager_timestamp
, ManagerTimestamp
);
4940 static const char* const oom_policy_table
[_OOM_POLICY_MAX
] = {
4941 [OOM_CONTINUE
] = "continue",
4942 [OOM_STOP
] = "stop",
4943 [OOM_KILL
] = "kill",
4946 DEFINE_STRING_TABLE_LOOKUP(oom_policy
, OOMPolicy
);