1 /* SPDX-License-Identifier: LGPL-2.1-or-later */
7 #include <sys/inotify.h>
9 #include <sys/reboot.h>
10 #include <sys/timerfd.h>
18 #include "sd-daemon.h"
19 #include "sd-messages.h"
22 #include "all-units.h"
23 #include "alloc-util.h"
25 #include "boot-timestamps.h"
26 #include "bus-common-errors.h"
27 #include "bus-error.h"
28 #include "bus-kernel.h"
30 #include "clean-ipc.h"
31 #include "clock-util.h"
32 #include "core-varlink.h"
33 #include "creds-util.h"
35 #include "dbus-manager.h"
36 #include "dbus-unit.h"
39 #include "dirent-util.h"
42 #include "exec-util.h"
44 #include "exit-status.h"
48 #include "generator-setup.h"
53 #include "load-fragment.h"
54 #include "locale-setup.h"
58 #include "manager-dump.h"
59 #include "memory-util.h"
61 #include "parse-util.h"
62 #include "path-lookup.h"
63 #include "path-util.h"
64 #include "process-util.h"
65 #include "ratelimit.h"
66 #include "rlimit-util.h"
68 #include "selinux-util.h"
69 #include "serialize.h"
70 #include "signal-util.h"
71 #include "socket-util.h"
73 #include "stat-util.h"
74 #include "string-table.h"
75 #include "string-util.h"
78 #include "sysctl-util.h"
79 #include "syslog-util.h"
80 #include "terminal-util.h"
81 #include "time-util.h"
82 #include "transaction.h"
83 #include "umask-util.h"
84 #include "unit-name.h"
85 #include "unit-serialize.h"
86 #include "user-util.h"
90 #define NOTIFY_RCVBUF_SIZE (8*1024*1024)
91 #define CGROUPS_AGENT_RCVBUF_SIZE (8*1024*1024)
93 /* Initial delay and the interval for printing status messages about running jobs */
94 #define JOBS_IN_PROGRESS_WAIT_USEC (2*USEC_PER_SEC)
95 #define JOBS_IN_PROGRESS_QUIET_WAIT_USEC (25*USEC_PER_SEC)
96 #define JOBS_IN_PROGRESS_PERIOD_USEC (USEC_PER_SEC / 3)
97 #define JOBS_IN_PROGRESS_PERIOD_DIVISOR 3
99 /* If there are more than 1K bus messages queue across our API and direct buses, then let's not add more on top until
100 * the queue gets more empty. */
101 #define MANAGER_BUS_BUSY_THRESHOLD 1024LU
103 /* How many units and jobs to process of the bus queue before returning to the event loop. */
104 #define MANAGER_BUS_MESSAGE_BUDGET 100U
106 static int manager_dispatch_notify_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
107 static int manager_dispatch_cgroups_agent_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
108 static int manager_dispatch_signal_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
109 static int manager_dispatch_time_change_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
110 static int manager_dispatch_idle_pipe_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
111 static int manager_dispatch_user_lookup_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
112 static int manager_dispatch_jobs_in_progress(sd_event_source
*source
, usec_t usec
, void *userdata
);
113 static int manager_dispatch_run_queue(sd_event_source
*source
, void *userdata
);
114 static int manager_dispatch_sigchld(sd_event_source
*source
, void *userdata
);
115 static int manager_dispatch_timezone_change(sd_event_source
*source
, const struct inotify_event
*event
, void *userdata
);
116 static int manager_run_environment_generators(Manager
*m
);
117 static int manager_run_generators(Manager
*m
);
118 static void manager_vacuum(Manager
*m
);
120 static usec_t
manager_watch_jobs_next_time(Manager
*m
) {
121 return usec_add(now(CLOCK_MONOTONIC
),
122 show_status_on(m
->show_status
) ? JOBS_IN_PROGRESS_WAIT_USEC
:
123 JOBS_IN_PROGRESS_QUIET_WAIT_USEC
);
126 static void manager_watch_jobs_in_progress(Manager
*m
) {
132 /* We do not want to show the cylon animation if the user
133 * needs to confirm service executions otherwise confirmation
134 * messages will be screwed by the cylon animation. */
135 if (!manager_is_confirm_spawn_disabled(m
))
138 if (m
->jobs_in_progress_event_source
)
141 next
= manager_watch_jobs_next_time(m
);
142 r
= sd_event_add_time(
144 &m
->jobs_in_progress_event_source
,
147 manager_dispatch_jobs_in_progress
, m
);
151 (void) sd_event_source_set_description(m
->jobs_in_progress_event_source
, "manager-jobs-in-progress");
154 #define CYLON_BUFFER_EXTRA (2*STRLEN(ANSI_RED) + STRLEN(ANSI_HIGHLIGHT_RED) + 2*STRLEN(ANSI_NORMAL))
156 static void draw_cylon(char buffer
[], size_t buflen
, unsigned width
, unsigned pos
) {
159 assert(buflen
>= CYLON_BUFFER_EXTRA
+ width
+ 1);
160 assert(pos
<= width
+1); /* 0 or width+1 mean that the center light is behind the corner */
164 p
= mempset(p
, ' ', pos
-2);
165 if (log_get_show_color())
166 p
= stpcpy(p
, ANSI_RED
);
170 if (pos
> 0 && pos
<= width
) {
171 if (log_get_show_color())
172 p
= stpcpy(p
, ANSI_HIGHLIGHT_RED
);
176 if (log_get_show_color())
177 p
= stpcpy(p
, ANSI_NORMAL
);
180 if (log_get_show_color())
181 p
= stpcpy(p
, ANSI_RED
);
184 p
= mempset(p
, ' ', width
-1-pos
);
185 if (log_get_show_color())
186 strcpy(p
, ANSI_NORMAL
);
190 static void manager_flip_auto_status(Manager
*m
, bool enable
, const char *reason
) {
194 if (m
->show_status
== SHOW_STATUS_AUTO
)
195 manager_set_show_status(m
, SHOW_STATUS_TEMPORARY
, reason
);
197 if (m
->show_status
== SHOW_STATUS_TEMPORARY
)
198 manager_set_show_status(m
, SHOW_STATUS_AUTO
, reason
);
202 static void manager_print_jobs_in_progress(Manager
*m
) {
203 _cleanup_free_
char *job_of_n
= NULL
;
205 unsigned counter
= 0, print_nr
;
206 char cylon
[6 + CYLON_BUFFER_EXTRA
+ 1];
208 char time
[FORMAT_TIMESPAN_MAX
], limit
[FORMAT_TIMESPAN_MAX
] = "no limit";
212 assert(m
->n_running_jobs
> 0);
214 manager_flip_auto_status(m
, true, "delay");
216 print_nr
= (m
->jobs_in_progress_iteration
/ JOBS_IN_PROGRESS_PERIOD_DIVISOR
) % m
->n_running_jobs
;
218 HASHMAP_FOREACH(j
, m
->jobs
)
219 if (j
->state
== JOB_RUNNING
&& counter
++ == print_nr
)
222 /* m->n_running_jobs must be consistent with the contents of m->jobs,
223 * so the above loop must have succeeded in finding j. */
224 assert(counter
== print_nr
+ 1);
227 cylon_pos
= m
->jobs_in_progress_iteration
% 14;
229 cylon_pos
= 14 - cylon_pos
;
230 draw_cylon(cylon
, sizeof(cylon
), 6, cylon_pos
);
232 m
->jobs_in_progress_iteration
++;
234 if (m
->n_running_jobs
> 1) {
235 if (asprintf(&job_of_n
, "(%u of %u) ", counter
, m
->n_running_jobs
) < 0)
239 format_timespan(time
, sizeof(time
), now(CLOCK_MONOTONIC
) - j
->begin_usec
, 1*USEC_PER_SEC
);
240 if (job_get_timeout(j
, &x
) > 0)
241 format_timespan(limit
, sizeof(limit
), x
- j
->begin_usec
, 1*USEC_PER_SEC
);
243 manager_status_printf(m
, STATUS_TYPE_EPHEMERAL
, cylon
,
244 "%sA %s job is running for %s (%s / %s)",
246 job_type_to_string(j
->type
),
247 unit_status_string(j
->unit
),
251 static int have_ask_password(void) {
252 _cleanup_closedir_
DIR *dir
= NULL
;
255 dir
= opendir("/run/systemd/ask-password");
263 FOREACH_DIRENT_ALL(de
, dir
, return -errno
) {
264 if (startswith(de
->d_name
, "ask."))
270 static int manager_dispatch_ask_password_fd(sd_event_source
*source
,
271 int fd
, uint32_t revents
, void *userdata
) {
272 Manager
*m
= userdata
;
278 m
->have_ask_password
= have_ask_password();
279 if (m
->have_ask_password
< 0)
280 /* Log error but continue. Negative have_ask_password
281 * is treated as unknown status. */
282 log_error_errno(m
->have_ask_password
, "Failed to list /run/systemd/ask-password: %m");
287 static void manager_close_ask_password(Manager
*m
) {
290 m
->ask_password_event_source
= sd_event_source_unref(m
->ask_password_event_source
);
291 m
->ask_password_inotify_fd
= safe_close(m
->ask_password_inotify_fd
);
292 m
->have_ask_password
= -EINVAL
;
295 static int manager_check_ask_password(Manager
*m
) {
300 if (!m
->ask_password_event_source
) {
301 assert(m
->ask_password_inotify_fd
< 0);
303 (void) mkdir_p_label("/run/systemd/ask-password", 0755);
305 m
->ask_password_inotify_fd
= inotify_init1(IN_NONBLOCK
|IN_CLOEXEC
);
306 if (m
->ask_password_inotify_fd
< 0)
307 return log_error_errno(errno
, "Failed to create inotify object: %m");
309 r
= inotify_add_watch_and_warn(m
->ask_password_inotify_fd
,
310 "/run/systemd/ask-password",
311 IN_CREATE
|IN_DELETE
|IN_MOVE
);
313 manager_close_ask_password(m
);
317 r
= sd_event_add_io(m
->event
, &m
->ask_password_event_source
,
318 m
->ask_password_inotify_fd
, EPOLLIN
,
319 manager_dispatch_ask_password_fd
, m
);
321 log_error_errno(r
, "Failed to add event source for /run/systemd/ask-password: %m");
322 manager_close_ask_password(m
);
326 (void) sd_event_source_set_description(m
->ask_password_event_source
, "manager-ask-password");
328 /* Queries might have been added meanwhile... */
329 manager_dispatch_ask_password_fd(m
->ask_password_event_source
,
330 m
->ask_password_inotify_fd
, EPOLLIN
, m
);
333 return m
->have_ask_password
;
336 static int manager_watch_idle_pipe(Manager
*m
) {
341 if (m
->idle_pipe_event_source
)
344 if (m
->idle_pipe
[2] < 0)
347 r
= sd_event_add_io(m
->event
, &m
->idle_pipe_event_source
, m
->idle_pipe
[2], EPOLLIN
, manager_dispatch_idle_pipe_fd
, m
);
349 return log_error_errno(r
, "Failed to watch idle pipe: %m");
351 (void) sd_event_source_set_description(m
->idle_pipe_event_source
, "manager-idle-pipe");
356 static void manager_close_idle_pipe(Manager
*m
) {
359 m
->idle_pipe_event_source
= sd_event_source_unref(m
->idle_pipe_event_source
);
361 safe_close_pair(m
->idle_pipe
);
362 safe_close_pair(m
->idle_pipe
+ 2);
365 static int manager_setup_time_change(Manager
*m
) {
370 if (MANAGER_IS_TEST_RUN(m
))
373 m
->time_change_event_source
= sd_event_source_unref(m
->time_change_event_source
);
374 m
->time_change_fd
= safe_close(m
->time_change_fd
);
376 m
->time_change_fd
= time_change_fd();
377 if (m
->time_change_fd
< 0)
378 return log_error_errno(m
->time_change_fd
, "Failed to create timer change timer fd: %m");
380 r
= sd_event_add_io(m
->event
, &m
->time_change_event_source
, m
->time_change_fd
, EPOLLIN
, manager_dispatch_time_change_fd
, m
);
382 return log_error_errno(r
, "Failed to create time change event source: %m");
384 /* Schedule this slightly earlier than the .timer event sources */
385 r
= sd_event_source_set_priority(m
->time_change_event_source
, SD_EVENT_PRIORITY_NORMAL
-1);
387 return log_error_errno(r
, "Failed to set priority of time change event sources: %m");
389 (void) sd_event_source_set_description(m
->time_change_event_source
, "manager-time-change");
391 log_debug("Set up TFD_TIMER_CANCEL_ON_SET timerfd.");
396 static int manager_read_timezone_stat(Manager
*m
) {
402 /* Read the current stat() data of /etc/localtime so that we detect changes */
403 if (lstat("/etc/localtime", &st
) < 0) {
404 log_debug_errno(errno
, "Failed to stat /etc/localtime, ignoring: %m");
405 changed
= m
->etc_localtime_accessible
;
406 m
->etc_localtime_accessible
= false;
410 k
= timespec_load(&st
.st_mtim
);
411 changed
= !m
->etc_localtime_accessible
|| k
!= m
->etc_localtime_mtime
;
413 m
->etc_localtime_mtime
= k
;
414 m
->etc_localtime_accessible
= true;
420 static int manager_setup_timezone_change(Manager
*m
) {
421 _cleanup_(sd_event_source_unrefp
) sd_event_source
*new_event
= NULL
;
426 if (MANAGER_IS_TEST_RUN(m
))
429 /* We watch /etc/localtime for three events: change of the link count (which might mean removal from /etc even
430 * though another link might be kept), renames, and file close operations after writing. Note we don't bother
431 * with IN_DELETE_SELF, as that would just report when the inode is removed entirely, i.e. after the link count
432 * went to zero and all fds to it are closed.
434 * Note that we never follow symlinks here. This is a simplification, but should cover almost all cases
437 * Note that we create the new event source first here, before releasing the old one. This should optimize
438 * behaviour as this way sd-event can reuse the old watch in case the inode didn't change. */
440 r
= sd_event_add_inotify(m
->event
, &new_event
, "/etc/localtime",
441 IN_ATTRIB
|IN_MOVE_SELF
|IN_CLOSE_WRITE
|IN_DONT_FOLLOW
, manager_dispatch_timezone_change
, m
);
443 /* If the file doesn't exist yet, subscribe to /etc instead, and wait until it is created either by
444 * O_CREATE or by rename() */
446 log_debug_errno(r
, "/etc/localtime doesn't exist yet, watching /etc instead.");
447 r
= sd_event_add_inotify(m
->event
, &new_event
, "/etc",
448 IN_CREATE
|IN_MOVED_TO
|IN_ONLYDIR
, manager_dispatch_timezone_change
, m
);
451 return log_error_errno(r
, "Failed to create timezone change event source: %m");
453 /* Schedule this slightly earlier than the .timer event sources */
454 r
= sd_event_source_set_priority(new_event
, SD_EVENT_PRIORITY_NORMAL
-1);
456 return log_error_errno(r
, "Failed to set priority of timezone change event sources: %m");
458 sd_event_source_unref(m
->timezone_change_event_source
);
459 m
->timezone_change_event_source
= TAKE_PTR(new_event
);
464 static int enable_special_signals(Manager
*m
) {
465 _cleanup_close_
int fd
= -1;
469 if (MANAGER_IS_TEST_RUN(m
))
472 /* Enable that we get SIGINT on control-alt-del. In containers
473 * this will fail with EPERM (older) or EINVAL (newer), so
475 if (reboot(RB_DISABLE_CAD
) < 0 && !IN_SET(errno
, EPERM
, EINVAL
))
476 log_warning_errno(errno
, "Failed to enable ctrl-alt-del handling: %m");
478 fd
= open_terminal("/dev/tty0", O_RDWR
|O_NOCTTY
|O_CLOEXEC
);
480 /* Support systems without virtual console */
482 log_warning_errno(errno
, "Failed to open /dev/tty0: %m");
484 /* Enable that we get SIGWINCH on kbrequest */
485 if (ioctl(fd
, KDSIGACCEPT
, SIGWINCH
) < 0)
486 log_warning_errno(errno
, "Failed to enable kbrequest handling: %m");
492 #define RTSIG_IF_AVAILABLE(signum) (signum <= SIGRTMAX ? signum : -1)
494 static int manager_setup_signals(Manager
*m
) {
495 struct sigaction sa
= {
496 .sa_handler
= SIG_DFL
,
497 .sa_flags
= SA_NOCLDSTOP
|SA_RESTART
,
504 assert_se(sigaction(SIGCHLD
, &sa
, NULL
) == 0);
506 /* We make liberal use of realtime signals here. On
507 * Linux/glibc we have 30 of them (with the exception of Linux
508 * on hppa, see below), between SIGRTMIN+0 ... SIGRTMIN+30
511 assert_se(sigemptyset(&mask
) == 0);
512 sigset_add_many(&mask
,
513 SIGCHLD
, /* Child died */
514 SIGTERM
, /* Reexecute daemon */
515 SIGHUP
, /* Reload configuration */
516 SIGUSR1
, /* systemd: reconnect to D-Bus */
517 SIGUSR2
, /* systemd: dump status */
518 SIGINT
, /* Kernel sends us this on control-alt-del */
519 SIGWINCH
, /* Kernel sends us this on kbrequest (alt-arrowup) */
520 SIGPWR
, /* Some kernel drivers and upsd send us this on power failure */
522 SIGRTMIN
+0, /* systemd: start default.target */
523 SIGRTMIN
+1, /* systemd: isolate rescue.target */
524 SIGRTMIN
+2, /* systemd: isolate emergency.target */
525 SIGRTMIN
+3, /* systemd: start halt.target */
526 SIGRTMIN
+4, /* systemd: start poweroff.target */
527 SIGRTMIN
+5, /* systemd: start reboot.target */
528 SIGRTMIN
+6, /* systemd: start kexec.target */
530 /* ... space for more special targets ... */
532 SIGRTMIN
+13, /* systemd: Immediate halt */
533 SIGRTMIN
+14, /* systemd: Immediate poweroff */
534 SIGRTMIN
+15, /* systemd: Immediate reboot */
535 SIGRTMIN
+16, /* systemd: Immediate kexec */
537 /* ... space for more immediate system state changes ... */
539 SIGRTMIN
+20, /* systemd: enable status messages */
540 SIGRTMIN
+21, /* systemd: disable status messages */
541 SIGRTMIN
+22, /* systemd: set log level to LOG_DEBUG */
542 SIGRTMIN
+23, /* systemd: set log level to LOG_INFO */
543 SIGRTMIN
+24, /* systemd: Immediate exit (--user only) */
545 /* .. one free signal here ... */
547 /* Apparently Linux on hppa had fewer RT signals until v3.18,
548 * SIGRTMAX was SIGRTMIN+25, and then SIGRTMIN was lowered,
549 * see commit v3.17-7614-g1f25df2eff.
551 * We cannot unconditionally make use of those signals here,
552 * so let's use a runtime check. Since these commands are
553 * accessible by different means and only really a safety
554 * net, the missing functionality on hppa shouldn't matter.
557 RTSIG_IF_AVAILABLE(SIGRTMIN
+26), /* systemd: set log target to journal-or-kmsg */
558 RTSIG_IF_AVAILABLE(SIGRTMIN
+27), /* systemd: set log target to console */
559 RTSIG_IF_AVAILABLE(SIGRTMIN
+28), /* systemd: set log target to kmsg */
560 RTSIG_IF_AVAILABLE(SIGRTMIN
+29), /* systemd: set log target to syslog-or-kmsg (obsolete) */
562 /* ... one free signal here SIGRTMIN+30 ... */
564 assert_se(sigprocmask(SIG_SETMASK
, &mask
, NULL
) == 0);
566 m
->signal_fd
= signalfd(-1, &mask
, SFD_NONBLOCK
|SFD_CLOEXEC
);
567 if (m
->signal_fd
< 0)
570 r
= sd_event_add_io(m
->event
, &m
->signal_event_source
, m
->signal_fd
, EPOLLIN
, manager_dispatch_signal_fd
, m
);
574 (void) sd_event_source_set_description(m
->signal_event_source
, "manager-signal");
576 /* Process signals a bit earlier than the rest of things, but later than notify_fd processing, so that the
577 * notify processing can still figure out to which process/service a message belongs, before we reap the
578 * process. Also, process this before handling cgroup notifications, so that we always collect child exit
579 * status information before detecting that there's no process in a cgroup. */
580 r
= sd_event_source_set_priority(m
->signal_event_source
, SD_EVENT_PRIORITY_NORMAL
-6);
584 if (MANAGER_IS_SYSTEM(m
))
585 return enable_special_signals(m
);
590 static char** sanitize_environment(char **l
) {
592 /* Let's remove some environment variables that we need ourselves to communicate with our clients */
596 "CONFIGURATION_DIRECTORY",
597 "CREDENTIALS_DIRECTORY",
619 /* Let's order the environment alphabetically, just to make it pretty */
625 int manager_default_environment(Manager
*m
) {
630 m
->transient_environment
= strv_free(m
->transient_environment
);
632 if (MANAGER_IS_SYSTEM(m
)) {
633 /* The system manager always starts with a clean
634 * environment for its children. It does not import
635 * the kernel's or the parents' exported variables.
637 * The initial passed environment is untouched to keep
638 * /proc/self/environ valid; it is used for tagging
639 * the init process inside containers. */
640 m
->transient_environment
= strv_new("PATH=" DEFAULT_PATH
);
641 if (!m
->transient_environment
)
644 /* Import locale variables LC_*= from configuration */
645 (void) locale_setup(&m
->transient_environment
);
647 /* The user manager passes its own environment along to its children, except for $PATH. */
648 m
->transient_environment
= strv_copy(environ
);
649 if (!m
->transient_environment
)
652 r
= strv_env_replace_strdup(&m
->transient_environment
, "PATH=" DEFAULT_USER_PATH
);
657 sanitize_environment(m
->transient_environment
);
662 static int manager_setup_prefix(Manager
*m
) {
668 static const struct table_entry paths_system
[_EXEC_DIRECTORY_TYPE_MAX
] = {
669 [EXEC_DIRECTORY_RUNTIME
] = { SD_PATH_SYSTEM_RUNTIME
, NULL
},
670 [EXEC_DIRECTORY_STATE
] = { SD_PATH_SYSTEM_STATE_PRIVATE
, NULL
},
671 [EXEC_DIRECTORY_CACHE
] = { SD_PATH_SYSTEM_STATE_CACHE
, NULL
},
672 [EXEC_DIRECTORY_LOGS
] = { SD_PATH_SYSTEM_STATE_LOGS
, NULL
},
673 [EXEC_DIRECTORY_CONFIGURATION
] = { SD_PATH_SYSTEM_CONFIGURATION
, NULL
},
676 static const struct table_entry paths_user
[_EXEC_DIRECTORY_TYPE_MAX
] = {
677 [EXEC_DIRECTORY_RUNTIME
] = { SD_PATH_USER_RUNTIME
, NULL
},
678 [EXEC_DIRECTORY_STATE
] = { SD_PATH_USER_CONFIGURATION
, NULL
},
679 [EXEC_DIRECTORY_CACHE
] = { SD_PATH_USER_STATE_CACHE
, NULL
},
680 [EXEC_DIRECTORY_LOGS
] = { SD_PATH_USER_CONFIGURATION
, "log" },
681 [EXEC_DIRECTORY_CONFIGURATION
] = { SD_PATH_USER_CONFIGURATION
, NULL
},
686 const struct table_entry
*p
= MANAGER_IS_SYSTEM(m
) ? paths_system
: paths_user
;
689 for (ExecDirectoryType i
= 0; i
< _EXEC_DIRECTORY_TYPE_MAX
; i
++) {
690 r
= sd_path_lookup(p
[i
].type
, p
[i
].suffix
, &m
->prefix
[i
]);
692 return log_warning_errno(r
, "Failed to lookup %s path: %m",
693 exec_directory_type_to_string(i
));
699 static void manager_free_unit_name_maps(Manager
*m
) {
700 m
->unit_id_map
= hashmap_free(m
->unit_id_map
);
701 m
->unit_name_map
= hashmap_free(m
->unit_name_map
);
702 m
->unit_path_cache
= set_free(m
->unit_path_cache
);
703 m
->unit_cache_timestamp_hash
= 0;
706 static int manager_setup_run_queue(Manager
*m
) {
710 assert(!m
->run_queue_event_source
);
712 r
= sd_event_add_defer(m
->event
, &m
->run_queue_event_source
, manager_dispatch_run_queue
, m
);
716 r
= sd_event_source_set_priority(m
->run_queue_event_source
, SD_EVENT_PRIORITY_IDLE
);
720 r
= sd_event_source_set_enabled(m
->run_queue_event_source
, SD_EVENT_OFF
);
724 (void) sd_event_source_set_description(m
->run_queue_event_source
, "manager-run-queue");
729 static int manager_setup_sigchld_event_source(Manager
*m
) {
733 assert(!m
->sigchld_event_source
);
735 r
= sd_event_add_defer(m
->event
, &m
->sigchld_event_source
, manager_dispatch_sigchld
, m
);
739 r
= sd_event_source_set_priority(m
->sigchld_event_source
, SD_EVENT_PRIORITY_NORMAL
-7);
743 r
= sd_event_source_set_enabled(m
->sigchld_event_source
, SD_EVENT_OFF
);
747 (void) sd_event_source_set_description(m
->sigchld_event_source
, "manager-sigchld");
752 int manager_new(UnitFileScope scope
, ManagerTestRunFlags test_run_flags
, Manager
**_m
) {
753 _cleanup_(manager_freep
) Manager
*m
= NULL
;
758 assert(IN_SET(scope
, UNIT_FILE_SYSTEM
, UNIT_FILE_USER
));
765 .unit_file_scope
= scope
,
766 .objective
= _MANAGER_OBJECTIVE_INVALID
,
768 .status_unit_format
= STATUS_UNIT_FORMAT_DEFAULT
,
770 .default_timer_accuracy_usec
= USEC_PER_MINUTE
,
771 .default_memory_accounting
= MEMORY_ACCOUNTING_DEFAULT
,
772 .default_tasks_accounting
= true,
773 .default_tasks_max
= TASKS_MAX_UNSET
,
774 .default_timeout_start_usec
= DEFAULT_TIMEOUT_USEC
,
775 .default_timeout_stop_usec
= DEFAULT_TIMEOUT_USEC
,
776 .default_restart_usec
= DEFAULT_RESTART_USEC
,
778 .original_log_level
= -1,
779 .original_log_target
= _LOG_TARGET_INVALID
,
781 .watchdog_overridden
[WATCHDOG_RUNTIME
] = USEC_INFINITY
,
782 .watchdog_overridden
[WATCHDOG_REBOOT
] = USEC_INFINITY
,
783 .watchdog_overridden
[WATCHDOG_KEXEC
] = USEC_INFINITY
,
785 .show_status_overridden
= _SHOW_STATUS_INVALID
,
788 .cgroups_agent_fd
= -1,
790 .time_change_fd
= -1,
791 .user_lookup_fds
= { -1, -1 },
792 .private_listen_fd
= -1,
794 .cgroup_inotify_fd
= -1,
795 .pin_cgroupfs_fd
= -1,
796 .ask_password_inotify_fd
= -1,
797 .idle_pipe
= { -1, -1, -1, -1},
799 /* start as id #1, so that we can leave #0 around as "null-like" value */
802 .have_ask_password
= -EINVAL
, /* we don't know */
804 .test_run_flags
= test_run_flags
,
806 .default_oom_policy
= OOM_STOP
,
810 if (MANAGER_IS_SYSTEM(m
) && detect_container() <= 0)
811 boot_timestamps(m
->timestamps
+ MANAGER_TIMESTAMP_USERSPACE
,
812 m
->timestamps
+ MANAGER_TIMESTAMP_FIRMWARE
,
813 m
->timestamps
+ MANAGER_TIMESTAMP_LOADER
);
816 /* Prepare log fields we can use for structured logging */
817 if (MANAGER_IS_SYSTEM(m
)) {
818 m
->unit_log_field
= "UNIT=";
819 m
->unit_log_format_string
= "UNIT=%s";
821 m
->invocation_log_field
= "INVOCATION_ID=";
822 m
->invocation_log_format_string
= "INVOCATION_ID=%s";
824 m
->unit_log_field
= "USER_UNIT=";
825 m
->unit_log_format_string
= "USER_UNIT=%s";
827 m
->invocation_log_field
= "USER_INVOCATION_ID=";
828 m
->invocation_log_format_string
= "USER_INVOCATION_ID=%s";
831 /* Reboot immediately if the user hits C-A-D more often than 7x per 2s */
832 m
->ctrl_alt_del_ratelimit
= (RateLimit
) { .interval
= 2 * USEC_PER_SEC
, .burst
= 7 };
834 r
= manager_default_environment(m
);
838 r
= hashmap_ensure_allocated(&m
->units
, &string_hash_ops
);
842 r
= hashmap_ensure_allocated(&m
->cgroup_unit
, &path_hash_ops
);
846 r
= hashmap_ensure_allocated(&m
->watch_bus
, &string_hash_ops
);
850 r
= prioq_ensure_allocated(&m
->run_queue
, compare_job_priority
);
854 r
= manager_setup_prefix(m
);
858 r
= get_credentials_dir(&e
);
860 m
->received_credentials
= strdup(e
);
861 if (!m
->received_credentials
)
865 r
= sd_event_default(&m
->event
);
869 r
= manager_setup_run_queue(m
);
873 if (test_run_flags
== MANAGER_TEST_RUN_MINIMAL
) {
874 m
->cgroup_root
= strdup("");
878 r
= manager_setup_signals(m
);
882 r
= manager_setup_cgroup(m
);
886 r
= manager_setup_time_change(m
);
890 r
= manager_read_timezone_stat(m
);
894 (void) manager_setup_timezone_change(m
);
896 r
= manager_setup_sigchld_event_source(m
);
901 if (test_run_flags
== 0) {
902 if (MANAGER_IS_SYSTEM(m
))
903 r
= mkdir_label("/run/systemd/units", 0755);
905 _cleanup_free_
char *units_path
= NULL
;
906 r
= xdg_user_runtime_dir(&units_path
, "/systemd/units");
909 r
= mkdir_p_label(units_path
, 0755);
912 if (r
< 0 && r
!= -EEXIST
)
918 dir_is_empty("/usr") > 0;
920 /* Note that we do not set up the notify fd here. We do that after deserialization,
921 * since they might have gotten serialized across the reexec. */
928 static int manager_setup_notify(Manager
*m
) {
931 if (MANAGER_IS_TEST_RUN(m
))
934 if (m
->notify_fd
< 0) {
935 _cleanup_close_
int fd
= -1;
936 union sockaddr_union sa
;
939 /* First free all secondary fields */
940 m
->notify_socket
= mfree(m
->notify_socket
);
941 m
->notify_event_source
= sd_event_source_unref(m
->notify_event_source
);
943 fd
= socket(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
|SOCK_NONBLOCK
, 0);
945 return log_error_errno(errno
, "Failed to allocate notification socket: %m");
947 fd_inc_rcvbuf(fd
, NOTIFY_RCVBUF_SIZE
);
949 m
->notify_socket
= path_join(m
->prefix
[EXEC_DIRECTORY_RUNTIME
], "systemd/notify");
950 if (!m
->notify_socket
)
953 r
= sockaddr_un_set_path(&sa
.un
, m
->notify_socket
);
955 return log_error_errno(r
, "Notify socket '%s' not valid for AF_UNIX socket address, refusing.",
959 (void) mkdir_parents_label(m
->notify_socket
, 0755);
960 (void) sockaddr_un_unlink(&sa
.un
);
962 r
= mac_selinux_bind(fd
, &sa
.sa
, sa_len
);
964 return log_error_errno(r
, "bind(%s) failed: %m", m
->notify_socket
);
966 r
= setsockopt_int(fd
, SOL_SOCKET
, SO_PASSCRED
, true);
968 return log_error_errno(r
, "SO_PASSCRED failed: %m");
970 m
->notify_fd
= TAKE_FD(fd
);
972 log_debug("Using notification socket %s", m
->notify_socket
);
975 if (!m
->notify_event_source
) {
976 r
= sd_event_add_io(m
->event
, &m
->notify_event_source
, m
->notify_fd
, EPOLLIN
, manager_dispatch_notify_fd
, m
);
978 return log_error_errno(r
, "Failed to allocate notify event source: %m");
980 /* Process notification messages a bit earlier than SIGCHLD, so that we can still identify to which
981 * service an exit message belongs. */
982 r
= sd_event_source_set_priority(m
->notify_event_source
, SD_EVENT_PRIORITY_NORMAL
-8);
984 return log_error_errno(r
, "Failed to set priority of notify event source: %m");
986 (void) sd_event_source_set_description(m
->notify_event_source
, "manager-notify");
992 static int manager_setup_cgroups_agent(Manager
*m
) {
994 static const union sockaddr_union sa
= {
995 .un
.sun_family
= AF_UNIX
,
996 .un
.sun_path
= "/run/systemd/cgroups-agent",
1000 /* This creates a listening socket we receive cgroups agent messages on. We do not use D-Bus for delivering
1001 * these messages from the cgroups agent binary to PID 1, as the cgroups agent binary is very short-living, and
1002 * each instance of it needs a new D-Bus connection. Since D-Bus connections are SOCK_STREAM/AF_UNIX, on
1003 * overloaded systems the backlog of the D-Bus socket becomes relevant, as not more than the configured number
1004 * of D-Bus connections may be queued until the kernel will start dropping further incoming connections,
1005 * possibly resulting in lost cgroups agent messages. To avoid this, we'll use a private SOCK_DGRAM/AF_UNIX
1006 * socket, where no backlog is relevant as communication may take place without an actual connect() cycle, and
1007 * we thus won't lose messages.
1009 * Note that PID 1 will forward the agent message to system bus, so that the user systemd instance may listen
1010 * to it. The system instance hence listens on this special socket, but the user instances listen on the system
1011 * bus for these messages. */
1013 if (MANAGER_IS_TEST_RUN(m
))
1016 if (!MANAGER_IS_SYSTEM(m
))
1019 r
= cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER
);
1021 return log_error_errno(r
, "Failed to determine whether unified cgroups hierarchy is used: %m");
1022 if (r
> 0) /* We don't need this anymore on the unified hierarchy */
1025 if (m
->cgroups_agent_fd
< 0) {
1026 _cleanup_close_
int fd
= -1;
1028 /* First free all secondary fields */
1029 m
->cgroups_agent_event_source
= sd_event_source_unref(m
->cgroups_agent_event_source
);
1031 fd
= socket(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
|SOCK_NONBLOCK
, 0);
1033 return log_error_errno(errno
, "Failed to allocate cgroups agent socket: %m");
1035 fd_inc_rcvbuf(fd
, CGROUPS_AGENT_RCVBUF_SIZE
);
1037 (void) sockaddr_un_unlink(&sa
.un
);
1039 /* Only allow root to connect to this socket */
1040 RUN_WITH_UMASK(0077)
1041 r
= bind(fd
, &sa
.sa
, SOCKADDR_UN_LEN(sa
.un
));
1043 return log_error_errno(errno
, "bind(%s) failed: %m", sa
.un
.sun_path
);
1045 m
->cgroups_agent_fd
= TAKE_FD(fd
);
1048 if (!m
->cgroups_agent_event_source
) {
1049 r
= sd_event_add_io(m
->event
, &m
->cgroups_agent_event_source
, m
->cgroups_agent_fd
, EPOLLIN
, manager_dispatch_cgroups_agent_fd
, m
);
1051 return log_error_errno(r
, "Failed to allocate cgroups agent event source: %m");
1053 /* Process cgroups notifications early. Note that when the agent notification is received
1054 * we'll just enqueue the unit in the cgroup empty queue, hence pick a high priority than
1055 * that. Also see handling of cgroup inotify for the unified cgroup stuff. */
1056 r
= sd_event_source_set_priority(m
->cgroups_agent_event_source
, SD_EVENT_PRIORITY_NORMAL
-9);
1058 return log_error_errno(r
, "Failed to set priority of cgroups agent event source: %m");
1060 (void) sd_event_source_set_description(m
->cgroups_agent_event_source
, "manager-cgroups-agent");
1066 static int manager_setup_user_lookup_fd(Manager
*m
) {
1071 /* Set up the socket pair used for passing UID/GID resolution results from forked off processes to PID
1072 * 1. Background: we can't do name lookups (NSS) from PID 1, since it might involve IPC and thus activation,
1073 * and we might hence deadlock on ourselves. Hence we do all user/group lookups asynchronously from the forked
1074 * off processes right before executing the binaries to start. In order to be able to clean up any IPC objects
1075 * created by a unit (see RemoveIPC=) we need to know in PID 1 the used UID/GID of the executed processes,
1076 * hence we establish this communication channel so that forked off processes can pass their UID/GID
1077 * information back to PID 1. The forked off processes send their resolved UID/GID to PID 1 in a simple
1078 * datagram, along with their unit name, so that we can share one communication socket pair among all units for
1081 * You might wonder why we need a communication channel for this that is independent of the usual notification
1082 * socket scheme (i.e. $NOTIFY_SOCKET). The primary difference is about trust: data sent via the $NOTIFY_SOCKET
1083 * channel is only accepted if it originates from the right unit and if reception was enabled for it. The user
1084 * lookup socket OTOH is only accessible by PID 1 and its children until they exec(), and always available.
1086 * Note that this function is called under two circumstances: when we first initialize (in which case we
1087 * allocate both the socket pair and the event source to listen on it), and when we deserialize after a reload
1088 * (in which case the socket pair already exists but we still need to allocate the event source for it). */
1090 if (m
->user_lookup_fds
[0] < 0) {
1092 /* Free all secondary fields */
1093 safe_close_pair(m
->user_lookup_fds
);
1094 m
->user_lookup_event_source
= sd_event_source_unref(m
->user_lookup_event_source
);
1096 if (socketpair(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
, 0, m
->user_lookup_fds
) < 0)
1097 return log_error_errno(errno
, "Failed to allocate user lookup socket: %m");
1099 (void) fd_inc_rcvbuf(m
->user_lookup_fds
[0], NOTIFY_RCVBUF_SIZE
);
1102 if (!m
->user_lookup_event_source
) {
1103 r
= sd_event_add_io(m
->event
, &m
->user_lookup_event_source
, m
->user_lookup_fds
[0], EPOLLIN
, manager_dispatch_user_lookup_fd
, m
);
1105 return log_error_errno(errno
, "Failed to allocate user lookup event source: %m");
1107 /* Process even earlier than the notify event source, so that we always know first about valid UID/GID
1109 r
= sd_event_source_set_priority(m
->user_lookup_event_source
, SD_EVENT_PRIORITY_NORMAL
-11);
1111 return log_error_errno(errno
, "Failed to set priority of user lookup event source: %m");
1113 (void) sd_event_source_set_description(m
->user_lookup_event_source
, "user-lookup");
1119 static unsigned manager_dispatch_cleanup_queue(Manager
*m
) {
1125 while ((u
= m
->cleanup_queue
)) {
1126 assert(u
->in_cleanup_queue
);
1136 GC_OFFSET_IN_PATH
, /* This one is on the path we were traveling */
1137 GC_OFFSET_UNSURE
, /* No clue */
1138 GC_OFFSET_GOOD
, /* We still need this unit */
1139 GC_OFFSET_BAD
, /* We don't need this unit anymore */
1143 static void unit_gc_mark_good(Unit
*u
, unsigned gc_marker
) {
1146 u
->gc_marker
= gc_marker
+ GC_OFFSET_GOOD
;
1148 /* Recursively mark referenced units as GOOD as well */
1149 UNIT_FOREACH_DEPENDENCY(other
, u
, UNIT_ATOM_REFERENCES
)
1150 if (other
->gc_marker
== gc_marker
+ GC_OFFSET_UNSURE
)
1151 unit_gc_mark_good(other
, gc_marker
);
1154 static void unit_gc_sweep(Unit
*u
, unsigned gc_marker
) {
1160 if (IN_SET(u
->gc_marker
- gc_marker
,
1161 GC_OFFSET_GOOD
, GC_OFFSET_BAD
, GC_OFFSET_UNSURE
, GC_OFFSET_IN_PATH
))
1164 if (u
->in_cleanup_queue
)
1167 if (!unit_may_gc(u
))
1170 u
->gc_marker
= gc_marker
+ GC_OFFSET_IN_PATH
;
1174 UNIT_FOREACH_DEPENDENCY(other
, u
, UNIT_ATOM_REFERENCED_BY
) {
1175 unit_gc_sweep(other
, gc_marker
);
1177 if (other
->gc_marker
== gc_marker
+ GC_OFFSET_GOOD
)
1180 if (other
->gc_marker
!= gc_marker
+ GC_OFFSET_BAD
)
1185 LIST_FOREACH(refs_by_target
, ref
, u
->refs_by_target
) {
1186 unit_gc_sweep(ref
->source
, gc_marker
);
1188 if (ref
->source
->gc_marker
== gc_marker
+ GC_OFFSET_GOOD
)
1191 if (ref
->source
->gc_marker
!= gc_marker
+ GC_OFFSET_BAD
)
1198 /* We were unable to find anything out about this entry, so
1199 * let's investigate it later */
1200 u
->gc_marker
= gc_marker
+ GC_OFFSET_UNSURE
;
1201 unit_add_to_gc_queue(u
);
1205 /* We definitely know that this one is not useful anymore, so
1206 * let's mark it for deletion */
1207 u
->gc_marker
= gc_marker
+ GC_OFFSET_BAD
;
1208 unit_add_to_cleanup_queue(u
);
1212 unit_gc_mark_good(u
, gc_marker
);
1215 static unsigned manager_dispatch_gc_unit_queue(Manager
*m
) {
1216 unsigned n
= 0, gc_marker
;
1221 /* log_debug("Running GC..."); */
1223 m
->gc_marker
+= _GC_OFFSET_MAX
;
1224 if (m
->gc_marker
+ _GC_OFFSET_MAX
<= _GC_OFFSET_MAX
)
1227 gc_marker
= m
->gc_marker
;
1229 while ((u
= m
->gc_unit_queue
)) {
1230 assert(u
->in_gc_queue
);
1232 unit_gc_sweep(u
, gc_marker
);
1234 LIST_REMOVE(gc_queue
, m
->gc_unit_queue
, u
);
1235 u
->in_gc_queue
= false;
1239 if (IN_SET(u
->gc_marker
- gc_marker
,
1240 GC_OFFSET_BAD
, GC_OFFSET_UNSURE
)) {
1242 log_unit_debug(u
, "Collecting.");
1243 u
->gc_marker
= gc_marker
+ GC_OFFSET_BAD
;
1244 unit_add_to_cleanup_queue(u
);
1251 static unsigned manager_dispatch_gc_job_queue(Manager
*m
) {
1257 while ((j
= m
->gc_job_queue
)) {
1258 assert(j
->in_gc_queue
);
1260 LIST_REMOVE(gc_queue
, m
->gc_job_queue
, j
);
1261 j
->in_gc_queue
= false;
1268 log_unit_debug(j
->unit
, "Collecting job.");
1269 (void) job_finish_and_invalidate(j
, JOB_COLLECTED
, false, false);
1275 static unsigned manager_dispatch_stop_when_unneeded_queue(Manager
*m
) {
1282 while ((u
= m
->stop_when_unneeded_queue
)) {
1283 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
1285 assert(u
->in_stop_when_unneeded_queue
);
1286 LIST_REMOVE(stop_when_unneeded_queue
, m
->stop_when_unneeded_queue
, u
);
1287 u
->in_stop_when_unneeded_queue
= false;
1291 if (!unit_is_unneeded(u
))
1294 log_unit_debug(u
, "Unit is not needed anymore.");
1296 /* If stopping a unit fails continuously we might enter a stop loop here, hence stop acting on the
1297 * service being unnecessary after a while. */
1299 if (!ratelimit_below(&u
->auto_start_stop_ratelimit
)) {
1300 log_unit_warning(u
, "Unit not needed anymore, but not stopping since we tried this too often recently.");
1304 /* Ok, nobody needs us anymore. Sniff. Then let's commit suicide */
1305 r
= manager_add_job(u
->manager
, JOB_STOP
, u
, JOB_FAIL
, NULL
, &error
, NULL
);
1307 log_unit_warning_errno(u
, r
, "Failed to enqueue stop job, ignoring: %s", bus_error_message(&error
, r
));
1313 static unsigned manager_dispatch_start_when_upheld_queue(Manager
*m
) {
1320 while ((u
= m
->start_when_upheld_queue
)) {
1321 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
1322 Unit
*culprit
= NULL
;
1324 assert(u
->in_start_when_upheld_queue
);
1325 LIST_REMOVE(start_when_upheld_queue
, m
->start_when_upheld_queue
, u
);
1326 u
->in_start_when_upheld_queue
= false;
1330 if (!unit_is_upheld_by_active(u
, &culprit
))
1333 log_unit_debug(u
, "Unit is started because upheld by active unit %s.", culprit
->id
);
1335 /* If stopping a unit fails continuously we might enter a stop loop here, hence stop acting on the
1336 * service being unnecessary after a while. */
1338 if (!ratelimit_below(&u
->auto_start_stop_ratelimit
)) {
1339 log_unit_warning(u
, "Unit needs to be started because active unit %s upholds it, but not starting since we tried this too often recently.", culprit
->id
);
1343 r
= manager_add_job(u
->manager
, JOB_START
, u
, JOB_FAIL
, NULL
, &error
, NULL
);
1345 log_unit_warning_errno(u
, r
, "Failed to enqueue start job, ignoring: %s", bus_error_message(&error
, r
));
1351 static unsigned manager_dispatch_stop_when_bound_queue(Manager
*m
) {
1358 while ((u
= m
->stop_when_bound_queue
)) {
1359 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
1360 Unit
*culprit
= NULL
;
1362 assert(u
->in_stop_when_bound_queue
);
1363 LIST_REMOVE(stop_when_bound_queue
, m
->stop_when_bound_queue
, u
);
1364 u
->in_stop_when_bound_queue
= false;
1368 if (!unit_is_bound_by_inactive(u
, &culprit
))
1371 log_unit_debug(u
, "Unit is stopped because bound to inactive unit %s.", culprit
->id
);
1373 /* If stopping a unit fails continuously we might enter a stop loop here, hence stop acting on the
1374 * service being unnecessary after a while. */
1376 if (!ratelimit_below(&u
->auto_start_stop_ratelimit
)) {
1377 log_unit_warning(u
, "Unit needs to be stopped because it is bound to inactive unit %s it, but not stopping since we tried this too often recently.", culprit
->id
);
1381 r
= manager_add_job(u
->manager
, JOB_STOP
, u
, JOB_REPLACE
, NULL
, &error
, NULL
);
1383 log_unit_warning_errno(u
, r
, "Failed to enqueue stop job, ignoring: %s", bus_error_message(&error
, r
));
1389 static void manager_clear_jobs_and_units(Manager
*m
) {
1394 while ((u
= hashmap_first(m
->units
)))
1397 manager_dispatch_cleanup_queue(m
);
1399 assert(!m
->load_queue
);
1400 assert(prioq_isempty(m
->run_queue
));
1401 assert(!m
->dbus_unit_queue
);
1402 assert(!m
->dbus_job_queue
);
1403 assert(!m
->cleanup_queue
);
1404 assert(!m
->gc_unit_queue
);
1405 assert(!m
->gc_job_queue
);
1406 assert(!m
->stop_when_unneeded_queue
);
1407 assert(!m
->start_when_upheld_queue
);
1408 assert(!m
->stop_when_bound_queue
);
1410 assert(hashmap_isempty(m
->jobs
));
1411 assert(hashmap_isempty(m
->units
));
1413 m
->n_on_console
= 0;
1414 m
->n_running_jobs
= 0;
1415 m
->n_installed_jobs
= 0;
1416 m
->n_failed_jobs
= 0;
1419 Manager
* manager_free(Manager
*m
) {
1423 manager_clear_jobs_and_units(m
);
1425 for (UnitType c
= 0; c
< _UNIT_TYPE_MAX
; c
++)
1426 if (unit_vtable
[c
]->shutdown
)
1427 unit_vtable
[c
]->shutdown(m
);
1429 /* Keep the cgroup hierarchy in place except when we know we are going down for good */
1430 manager_shutdown_cgroup(m
, IN_SET(m
->objective
, MANAGER_EXIT
, MANAGER_REBOOT
, MANAGER_POWEROFF
, MANAGER_HALT
, MANAGER_KEXEC
));
1432 lookup_paths_flush_generator(&m
->lookup_paths
);
1435 manager_varlink_done(m
);
1437 exec_runtime_vacuum(m
);
1438 hashmap_free(m
->exec_runtime_by_id
);
1440 dynamic_user_vacuum(m
, false);
1441 hashmap_free(m
->dynamic_users
);
1443 hashmap_free(m
->units
);
1444 hashmap_free(m
->units_by_invocation_id
);
1445 hashmap_free(m
->jobs
);
1446 hashmap_free(m
->watch_pids
);
1447 hashmap_free(m
->watch_bus
);
1449 prioq_free(m
->run_queue
);
1451 set_free(m
->startup_units
);
1452 set_free(m
->failed_units
);
1454 sd_event_source_unref(m
->signal_event_source
);
1455 sd_event_source_unref(m
->sigchld_event_source
);
1456 sd_event_source_unref(m
->notify_event_source
);
1457 sd_event_source_unref(m
->cgroups_agent_event_source
);
1458 sd_event_source_unref(m
->time_change_event_source
);
1459 sd_event_source_unref(m
->timezone_change_event_source
);
1460 sd_event_source_unref(m
->jobs_in_progress_event_source
);
1461 sd_event_source_unref(m
->run_queue_event_source
);
1462 sd_event_source_unref(m
->user_lookup_event_source
);
1464 safe_close(m
->signal_fd
);
1465 safe_close(m
->notify_fd
);
1466 safe_close(m
->cgroups_agent_fd
);
1467 safe_close(m
->time_change_fd
);
1468 safe_close_pair(m
->user_lookup_fds
);
1470 manager_close_ask_password(m
);
1472 manager_close_idle_pipe(m
);
1474 sd_event_unref(m
->event
);
1476 free(m
->notify_socket
);
1478 lookup_paths_free(&m
->lookup_paths
);
1479 strv_free(m
->transient_environment
);
1480 strv_free(m
->client_environment
);
1482 hashmap_free(m
->cgroup_unit
);
1483 manager_free_unit_name_maps(m
);
1485 free(m
->switch_root
);
1486 free(m
->switch_root_init
);
1488 rlimit_free_all(m
->rlimit
);
1490 assert(hashmap_isempty(m
->units_requiring_mounts_for
));
1491 hashmap_free(m
->units_requiring_mounts_for
);
1493 hashmap_free(m
->uid_refs
);
1494 hashmap_free(m
->gid_refs
);
1496 for (ExecDirectoryType dt
= 0; dt
< _EXEC_DIRECTORY_TYPE_MAX
; dt
++)
1497 m
->prefix
[dt
] = mfree(m
->prefix
[dt
]);
1498 free(m
->received_credentials
);
1503 static void manager_enumerate_perpetual(Manager
*m
) {
1506 if (m
->test_run_flags
== MANAGER_TEST_RUN_MINIMAL
)
1509 /* Let's ask every type to load all units from disk/kernel that it might know */
1510 for (UnitType c
= 0; c
< _UNIT_TYPE_MAX
; c
++) {
1511 if (!unit_type_supported(c
)) {
1512 log_debug("Unit type .%s is not supported on this system.", unit_type_to_string(c
));
1516 if (unit_vtable
[c
]->enumerate_perpetual
)
1517 unit_vtable
[c
]->enumerate_perpetual(m
);
1521 static void manager_enumerate(Manager
*m
) {
1524 if (m
->test_run_flags
== MANAGER_TEST_RUN_MINIMAL
)
1527 /* Let's ask every type to load all units from disk/kernel that it might know */
1528 for (UnitType c
= 0; c
< _UNIT_TYPE_MAX
; c
++) {
1529 if (!unit_type_supported(c
)) {
1530 log_debug("Unit type .%s is not supported on this system.", unit_type_to_string(c
));
1534 if (unit_vtable
[c
]->enumerate
)
1535 unit_vtable
[c
]->enumerate(m
);
1538 manager_dispatch_load_queue(m
);
1541 static void manager_coldplug(Manager
*m
) {
1548 log_debug("Invoking unit coldplug() handlers…");
1550 /* Let's place the units back into their deserialized state */
1551 HASHMAP_FOREACH_KEY(u
, k
, m
->units
) {
1553 /* ignore aliases */
1557 r
= unit_coldplug(u
);
1559 log_warning_errno(r
, "We couldn't coldplug %s, proceeding anyway: %m", u
->id
);
1563 static void manager_catchup(Manager
*m
) {
1569 log_debug("Invoking unit catchup() handlers…");
1571 /* Let's catch up on any state changes that happened while we were reloading/reexecing */
1572 HASHMAP_FOREACH_KEY(u
, k
, m
->units
) {
1574 /* ignore aliases */
1582 static void manager_distribute_fds(Manager
*m
, FDSet
*fds
) {
1587 HASHMAP_FOREACH(u
, m
->units
) {
1589 if (fdset_size(fds
) <= 0)
1592 if (!UNIT_VTABLE(u
)->distribute_fds
)
1595 UNIT_VTABLE(u
)->distribute_fds(u
, fds
);
1599 static bool manager_dbus_is_running(Manager
*m
, bool deserialized
) {
1604 /* This checks whether the dbus instance we are supposed to expose our APIs on is up. We check both the socket
1605 * and the service unit. If the 'deserialized' parameter is true we'll check the deserialized state of the unit
1606 * rather than the current one. */
1608 if (MANAGER_IS_TEST_RUN(m
))
1611 u
= manager_get_unit(m
, SPECIAL_DBUS_SOCKET
);
1614 if ((deserialized
? SOCKET(u
)->deserialized_state
: SOCKET(u
)->state
) != SOCKET_RUNNING
)
1617 u
= manager_get_unit(m
, SPECIAL_DBUS_SERVICE
);
1620 if (!IN_SET((deserialized
? SERVICE(u
)->deserialized_state
: SERVICE(u
)->state
), SERVICE_RUNNING
, SERVICE_RELOAD
))
1626 static void manager_setup_bus(Manager
*m
) {
1629 /* Let's set up our private bus connection now, unconditionally */
1630 (void) bus_init_private(m
);
1632 /* If we are in --user mode also connect to the system bus now */
1633 if (MANAGER_IS_USER(m
))
1634 (void) bus_init_system(m
);
1636 /* Let's connect to the bus now, but only if the unit is supposed to be up */
1637 if (manager_dbus_is_running(m
, MANAGER_IS_RELOADING(m
))) {
1638 (void) bus_init_api(m
);
1640 if (MANAGER_IS_SYSTEM(m
))
1641 (void) bus_init_system(m
);
1645 static void manager_preset_all(Manager
*m
) {
1650 if (m
->first_boot
<= 0)
1653 if (!MANAGER_IS_SYSTEM(m
))
1656 if (MANAGER_IS_TEST_RUN(m
))
1659 /* If this is the first boot, and we are in the host system, then preset everything */
1660 r
= unit_file_preset_all(UNIT_FILE_SYSTEM
, 0, NULL
, UNIT_FILE_PRESET_ENABLE_ONLY
, NULL
, 0);
1662 log_full_errno(r
== -EEXIST
? LOG_NOTICE
: LOG_WARNING
, r
,
1663 "Failed to populate /etc with preset unit settings, ignoring: %m");
1665 log_info("Populated /etc with preset unit settings.");
1668 static void manager_ready(Manager
*m
) {
1671 /* After having loaded everything, do the final round of catching up with what might have changed */
1673 m
->objective
= MANAGER_OK
; /* Tell everyone we are up now */
1675 /* It might be safe to log to the journal now and connect to dbus */
1676 manager_recheck_journal(m
);
1677 manager_recheck_dbus(m
);
1679 /* Let's finally catch up with any changes that took place while we were reloading/reexecing */
1682 m
->honor_device_enumeration
= true;
1685 static Manager
* manager_reloading_start(Manager
*m
) {
1689 static void manager_reloading_stopp(Manager
**m
) {
1691 assert((*m
)->n_reloading
> 0);
1692 (*m
)->n_reloading
--;
1696 int manager_startup(Manager
*m
, FILE *serialization
, FDSet
*fds
) {
1701 /* If we are running in test mode, we still want to run the generators,
1702 * but we should not touch the real generator directories. */
1703 r
= lookup_paths_init(&m
->lookup_paths
, m
->unit_file_scope
,
1704 MANAGER_IS_TEST_RUN(m
) ? LOOKUP_PATHS_TEMPORARY_GENERATED
: 0,
1707 return log_error_errno(r
, "Failed to initialize path lookup table: %m");
1709 dual_timestamp_get(m
->timestamps
+ manager_timestamp_initrd_mangle(MANAGER_TIMESTAMP_GENERATORS_START
));
1710 r
= manager_run_environment_generators(m
);
1712 r
= manager_run_generators(m
);
1713 dual_timestamp_get(m
->timestamps
+ manager_timestamp_initrd_mangle(MANAGER_TIMESTAMP_GENERATORS_FINISH
));
1717 manager_preset_all(m
);
1719 lookup_paths_log(&m
->lookup_paths
);
1722 /* This block is (optionally) done with the reloading counter bumped */
1723 _cleanup_(manager_reloading_stopp
) Manager
*reloading
= NULL
;
1725 /* If we will deserialize make sure that during enumeration this is already known, so we increase the
1726 * counter here already */
1728 reloading
= manager_reloading_start(m
);
1730 /* First, enumerate what we can from all config files */
1731 dual_timestamp_get(m
->timestamps
+ manager_timestamp_initrd_mangle(MANAGER_TIMESTAMP_UNITS_LOAD_START
));
1732 manager_enumerate_perpetual(m
);
1733 manager_enumerate(m
);
1734 dual_timestamp_get(m
->timestamps
+ manager_timestamp_initrd_mangle(MANAGER_TIMESTAMP_UNITS_LOAD_FINISH
));
1736 /* Second, deserialize if there is something to deserialize */
1737 if (serialization
) {
1738 r
= manager_deserialize(m
, serialization
, fds
);
1740 return log_error_errno(r
, "Deserialization failed: %m");
1743 /* Any fds left? Find some unit which wants them. This is useful to allow container managers to pass
1744 * some file descriptors to us pre-initialized. This enables socket-based activation of entire
1746 manager_distribute_fds(m
, fds
);
1748 /* We might have deserialized the notify fd, but if we didn't then let's create the bus now */
1749 r
= manager_setup_notify(m
);
1751 /* No sense to continue without notifications, our children would fail anyway. */
1754 r
= manager_setup_cgroups_agent(m
);
1756 /* Likewise, no sense to continue without empty cgroup notifications. */
1759 r
= manager_setup_user_lookup_fd(m
);
1761 /* This shouldn't fail, except if things are really broken. */
1764 /* Connect to the bus if we are good for it */
1765 manager_setup_bus(m
);
1767 /* Now that we are connected to all possible buses, let's deserialize who is tracking us. */
1768 r
= bus_track_coldplug(m
, &m
->subscribed
, false, m
->deserialized_subscribed
);
1770 log_warning_errno(r
, "Failed to deserialized tracked clients, ignoring: %m");
1771 m
->deserialized_subscribed
= strv_free(m
->deserialized_subscribed
);
1773 r
= manager_varlink_init(m
);
1775 log_warning_errno(r
, "Failed to set up Varlink server, ignoring: %m");
1777 /* Third, fire things up! */
1778 manager_coldplug(m
);
1780 /* Clean up runtime objects */
1784 /* Let's wait for the UnitNew/JobNew messages being sent, before we notify that the
1785 * reload is finished */
1786 m
->send_reloading_done
= true;
1794 int manager_add_job(
1800 sd_bus_error
*error
,
1807 assert(type
< _JOB_TYPE_MAX
);
1809 assert(mode
< _JOB_MODE_MAX
);
1811 if (mode
== JOB_ISOLATE
&& type
!= JOB_START
)
1812 return sd_bus_error_set(error
, SD_BUS_ERROR_INVALID_ARGS
, "Isolate is only valid for start.");
1814 if (mode
== JOB_ISOLATE
&& !unit
->allow_isolate
)
1815 return sd_bus_error_set(error
, BUS_ERROR_NO_ISOLATION
, "Operation refused, unit may not be isolated.");
1817 if (mode
== JOB_TRIGGERING
&& type
!= JOB_STOP
)
1818 return sd_bus_error_set(error
, SD_BUS_ERROR_INVALID_ARGS
, "--job-mode=triggering is only valid for stop.");
1820 log_unit_debug(unit
, "Trying to enqueue job %s/%s/%s", unit
->id
, job_type_to_string(type
), job_mode_to_string(mode
));
1822 type
= job_type_collapse(type
, unit
);
1824 tr
= transaction_new(mode
== JOB_REPLACE_IRREVERSIBLY
);
1828 r
= transaction_add_job_and_dependencies(tr
, type
, unit
, NULL
, true, false,
1829 IN_SET(mode
, JOB_IGNORE_DEPENDENCIES
, JOB_IGNORE_REQUIREMENTS
),
1830 mode
== JOB_IGNORE_DEPENDENCIES
, error
);
1834 if (mode
== JOB_ISOLATE
) {
1835 r
= transaction_add_isolate_jobs(tr
, m
);
1840 if (mode
== JOB_TRIGGERING
) {
1841 r
= transaction_add_triggering_jobs(tr
, unit
);
1846 r
= transaction_activate(tr
, m
, mode
, affected_jobs
, error
);
1850 log_unit_debug(unit
,
1851 "Enqueued job %s/%s as %u", unit
->id
,
1852 job_type_to_string(type
), (unsigned) tr
->anchor_job
->id
);
1855 *ret
= tr
->anchor_job
;
1857 transaction_free(tr
);
1861 transaction_abort(tr
);
1862 transaction_free(tr
);
1866 int manager_add_job_by_name(Manager
*m
, JobType type
, const char *name
, JobMode mode
, Set
*affected_jobs
, sd_bus_error
*e
, Job
**ret
) {
1867 Unit
*unit
= NULL
; /* just to appease gcc, initialization is not really necessary */
1871 assert(type
< _JOB_TYPE_MAX
);
1873 assert(mode
< _JOB_MODE_MAX
);
1875 r
= manager_load_unit(m
, name
, NULL
, NULL
, &unit
);
1880 return manager_add_job(m
, type
, unit
, mode
, affected_jobs
, e
, ret
);
1883 int manager_add_job_by_name_and_warn(Manager
*m
, JobType type
, const char *name
, JobMode mode
, Set
*affected_jobs
, Job
**ret
) {
1884 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
1888 assert(type
< _JOB_TYPE_MAX
);
1890 assert(mode
< _JOB_MODE_MAX
);
1892 r
= manager_add_job_by_name(m
, type
, name
, mode
, affected_jobs
, &error
, ret
);
1894 return log_warning_errno(r
, "Failed to enqueue %s job for %s: %s", job_mode_to_string(mode
), name
, bus_error_message(&error
, r
));
1899 int manager_propagate_reload(Manager
*m
, Unit
*unit
, JobMode mode
, sd_bus_error
*e
) {
1905 assert(mode
< _JOB_MODE_MAX
);
1906 assert(mode
!= JOB_ISOLATE
); /* Isolate is only valid for start */
1908 tr
= transaction_new(mode
== JOB_REPLACE_IRREVERSIBLY
);
1912 /* We need an anchor job */
1913 r
= transaction_add_job_and_dependencies(tr
, JOB_NOP
, unit
, NULL
, false, false, true, true, e
);
1917 /* Failure in adding individual dependencies is ignored, so this always succeeds. */
1918 transaction_add_propagate_reload_jobs(tr
, unit
, tr
->anchor_job
, mode
== JOB_IGNORE_DEPENDENCIES
, e
);
1920 r
= transaction_activate(tr
, m
, mode
, NULL
, e
);
1924 transaction_free(tr
);
1928 transaction_abort(tr
);
1929 transaction_free(tr
);
1933 Job
*manager_get_job(Manager
*m
, uint32_t id
) {
1936 return hashmap_get(m
->jobs
, UINT32_TO_PTR(id
));
1939 Unit
*manager_get_unit(Manager
*m
, const char *name
) {
1943 return hashmap_get(m
->units
, name
);
1946 static int manager_dispatch_target_deps_queue(Manager
*m
) {
1952 while ((u
= m
->target_deps_queue
)) {
1953 _cleanup_free_ Unit
**targets
= NULL
;
1956 assert(u
->in_target_deps_queue
);
1958 LIST_REMOVE(target_deps_queue
, u
->manager
->target_deps_queue
, u
);
1959 u
->in_target_deps_queue
= false;
1961 /* Take an "atomic" snapshot of dependencies here, as the call below will likely modify the
1962 * dependencies, and we can't have it that hash tables we iterate through are modified while
1963 * we are iterating through them. */
1964 n_targets
= unit_get_dependency_array(u
, UNIT_ATOM_DEFAULT_TARGET_DEPENDENCIES
, &targets
);
1968 for (int i
= 0; i
< n_targets
; i
++) {
1969 r
= unit_add_default_target_dependency(u
, targets
[i
]);
1978 unsigned manager_dispatch_load_queue(Manager
*m
) {
1984 /* Make sure we are not run recursively */
1985 if (m
->dispatching_load_queue
)
1988 m
->dispatching_load_queue
= true;
1990 /* Dispatches the load queue. Takes a unit from the queue and
1991 * tries to load its data until the queue is empty */
1993 while ((u
= m
->load_queue
)) {
1994 assert(u
->in_load_queue
);
2000 m
->dispatching_load_queue
= false;
2002 /* Dispatch the units waiting for their target dependencies to be added now, as all targets that we know about
2003 * should be loaded and have aliases resolved */
2004 (void) manager_dispatch_target_deps_queue(m
);
2009 bool manager_unit_cache_should_retry_load(Unit
*u
) {
2012 /* Automatic reloading from disk only applies to units which were not found sometime in the past, and
2013 * the not-found stub is kept pinned in the unit graph by dependencies. For units that were
2014 * previously loaded, we don't do automatic reloading, and daemon-reload is necessary to update. */
2015 if (u
->load_state
!= UNIT_NOT_FOUND
)
2018 /* The cache has been updated since the last time we tried to load the unit. There might be new
2019 * fragment paths to read. */
2020 if (u
->manager
->unit_cache_timestamp_hash
!= u
->fragment_not_found_timestamp_hash
)
2023 /* The cache needs to be updated because there are modifications on disk. */
2024 return !lookup_paths_timestamp_hash_same(&u
->manager
->lookup_paths
, u
->manager
->unit_cache_timestamp_hash
, NULL
);
2027 int manager_load_unit_prepare(
2034 _cleanup_(unit_freep
) Unit
*cleanup_ret
= NULL
;
2042 /* This will prepare the unit for loading, but not actually load anything from disk. */
2044 if (path
&& !path_is_absolute(path
))
2045 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
, "Path %s is not absolute.", path
);
2048 /* 'name' and 'path' must not both be null. Check here 'path' using assert_se() to
2049 * workaround a bug in gcc that generates a -Wnonnull warning when calling basename(),
2050 * but this cannot be possible in any code path (See #6119). */
2052 name
= basename(path
);
2055 t
= unit_name_to_type(name
);
2057 if (t
== _UNIT_TYPE_INVALID
|| !unit_name_is_valid(name
, UNIT_NAME_PLAIN
|UNIT_NAME_INSTANCE
)) {
2058 if (unit_name_is_valid(name
, UNIT_NAME_TEMPLATE
))
2059 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
, "Unit name %s is missing the instance name.", name
);
2061 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
, "Unit name %s is not valid.", name
);
2064 ret
= manager_get_unit(m
, name
);
2066 /* The time-based cache allows to start new units without daemon-reload,
2067 * but if they are already referenced (because of dependencies or ordering)
2068 * then we have to force a load of the fragment. As an optimization, check
2069 * first if anything in the usual paths was modified since the last time
2070 * the cache was loaded. Also check if the last time an attempt to load the
2071 * unit was made was before the most recent cache refresh, so that we know
2072 * we need to try again — even if the cache is current, it might have been
2073 * updated in a different context before we had a chance to retry loading
2074 * this particular unit. */
2075 if (manager_unit_cache_should_retry_load(ret
))
2076 ret
->load_state
= UNIT_STUB
;
2082 ret
= cleanup_ret
= unit_new(m
, unit_vtable
[t
]->object_size
);
2088 r
= free_and_strdup(&ret
->fragment_path
, path
);
2093 r
= unit_add_name(ret
, name
);
2097 unit_add_to_load_queue(ret
);
2098 unit_add_to_dbus_queue(ret
);
2099 unit_add_to_gc_queue(ret
);
2107 int manager_load_unit(
2119 /* This will load the service information files, but not actually
2120 * start any services or anything. */
2122 r
= manager_load_unit_prepare(m
, name
, path
, e
, _ret
);
2126 manager_dispatch_load_queue(m
);
2128 *_ret
= unit_follow_merge(*_ret
);
2132 int manager_load_startable_unit_or_warn(
2138 /* Load a unit, make sure it loaded fully and is not masked. */
2140 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
2144 r
= manager_load_unit(m
, name
, path
, &error
, &unit
);
2146 return log_error_errno(r
, "Failed to load %s %s: %s",
2147 name
? "unit" : "unit file", name
?: path
,
2148 bus_error_message(&error
, r
));
2150 r
= bus_unit_validate_load_state(unit
, &error
);
2152 return log_error_errno(r
, "%s", bus_error_message(&error
, r
));
2158 void manager_clear_jobs(Manager
*m
) {
2163 while ((j
= hashmap_first(m
->jobs
)))
2164 /* No need to recurse. We're cancelling all jobs. */
2165 job_finish_and_invalidate(j
, JOB_CANCELED
, false, false);
2168 void manager_unwatch_pid(Manager
*m
, pid_t pid
) {
2171 /* First let's drop the unit keyed as "pid". */
2172 (void) hashmap_remove(m
->watch_pids
, PID_TO_PTR(pid
));
2174 /* Then, let's also drop the array keyed by -pid. */
2175 free(hashmap_remove(m
->watch_pids
, PID_TO_PTR(-pid
)));
2178 static int manager_dispatch_run_queue(sd_event_source
*source
, void *userdata
) {
2179 Manager
*m
= userdata
;
2185 while ((j
= prioq_peek(m
->run_queue
))) {
2186 assert(j
->installed
);
2187 assert(j
->in_run_queue
);
2189 (void) job_run_and_invalidate(j
);
2192 if (m
->n_running_jobs
> 0)
2193 manager_watch_jobs_in_progress(m
);
2195 if (m
->n_on_console
> 0)
2196 manager_watch_idle_pipe(m
);
2201 static unsigned manager_dispatch_dbus_queue(Manager
*m
) {
2202 unsigned n
= 0, budget
;
2208 /* When we are reloading, let's not wait with generating signals, since we need to exit the manager as quickly
2209 * as we can. There's no point in throttling generation of signals in that case. */
2210 if (MANAGER_IS_RELOADING(m
) || m
->send_reloading_done
|| m
->pending_reload_message
)
2211 budget
= UINT_MAX
; /* infinite budget in this case */
2213 /* Anything to do at all? */
2214 if (!m
->dbus_unit_queue
&& !m
->dbus_job_queue
)
2217 /* Do we have overly many messages queued at the moment? If so, let's not enqueue more on top, let's
2218 * sit this cycle out, and process things in a later cycle when the queues got a bit emptier. */
2219 if (manager_bus_n_queued_write(m
) > MANAGER_BUS_BUSY_THRESHOLD
)
2222 /* Only process a certain number of units/jobs per event loop iteration. Even if the bus queue wasn't
2223 * overly full before this call we shouldn't increase it in size too wildly in one step, and we
2224 * shouldn't monopolize CPU time with generating these messages. Note the difference in counting of
2225 * this "budget" and the "threshold" above: the "budget" is decreased only once per generated message,
2226 * regardless how many buses/direct connections it is enqueued on, while the "threshold" is applied to
2227 * each queued instance of bus message, i.e. if the same message is enqueued to five buses/direct
2228 * connections it will be counted five times. This difference in counting ("references"
2229 * vs. "instances") is primarily a result of the fact that it's easier to implement it this way,
2230 * however it also reflects the thinking that the "threshold" should put a limit on used queue memory,
2231 * i.e. space, while the "budget" should put a limit on time. Also note that the "threshold" is
2232 * currently chosen much higher than the "budget". */
2233 budget
= MANAGER_BUS_MESSAGE_BUDGET
;
2236 while (budget
!= 0 && (u
= m
->dbus_unit_queue
)) {
2238 assert(u
->in_dbus_queue
);
2240 bus_unit_send_change_signal(u
);
2243 if (budget
!= UINT_MAX
)
2247 while (budget
!= 0 && (j
= m
->dbus_job_queue
)) {
2248 assert(j
->in_dbus_queue
);
2250 bus_job_send_change_signal(j
);
2253 if (budget
!= UINT_MAX
)
2257 if (m
->send_reloading_done
) {
2258 m
->send_reloading_done
= false;
2259 bus_manager_send_reloading(m
, false);
2263 if (m
->pending_reload_message
) {
2264 bus_send_pending_reload_message(m
);
2271 static int manager_dispatch_cgroups_agent_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
2272 Manager
*m
= userdata
;
2276 n
= recv(fd
, buf
, sizeof(buf
), 0);
2278 return log_error_errno(errno
, "Failed to read cgroups agent message: %m");
2280 log_error("Got zero-length cgroups agent message, ignoring.");
2283 if ((size_t) n
>= sizeof(buf
)) {
2284 log_error("Got overly long cgroups agent message, ignoring.");
2288 if (memchr(buf
, 0, n
)) {
2289 log_error("Got cgroups agent message with embedded NUL byte, ignoring.");
2294 manager_notify_cgroup_empty(m
, buf
);
2295 (void) bus_forward_agent_released(m
, buf
);
2300 static bool manager_process_barrier_fd(char * const *tags
, FDSet
*fds
) {
2302 /* nothing else must be sent when using BARRIER=1 */
2303 if (strv_contains(tags
, "BARRIER=1")) {
2304 if (strv_length(tags
) == 1) {
2305 if (fdset_size(fds
) != 1)
2306 log_warning("Got incorrect number of fds with BARRIER=1, closing them.");
2308 log_warning("Extra notification messages sent with BARRIER=1, ignoring everything.");
2310 /* Drop the message if BARRIER=1 was found */
2317 static void manager_invoke_notify_message(
2320 const struct ucred
*ucred
,
2329 if (u
->notifygen
== m
->notifygen
) /* Already invoked on this same unit in this same iteration? */
2331 u
->notifygen
= m
->notifygen
;
2333 if (UNIT_VTABLE(u
)->notify_message
)
2334 UNIT_VTABLE(u
)->notify_message(u
, ucred
, tags
, fds
);
2336 else if (DEBUG_LOGGING
) {
2337 _cleanup_free_
char *buf
= NULL
, *x
= NULL
, *y
= NULL
;
2339 buf
= strv_join(tags
, ", ");
2341 x
= ellipsize(buf
, 20, 90);
2345 log_unit_debug(u
, "Got notification message \"%s\", ignoring.", strnull(y
));
2349 static int manager_dispatch_notify_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
2351 _cleanup_fdset_free_ FDSet
*fds
= NULL
;
2352 Manager
*m
= userdata
;
2353 char buf
[NOTIFY_BUFFER_MAX
+1];
2354 struct iovec iovec
= {
2356 .iov_len
= sizeof(buf
)-1,
2358 CMSG_BUFFER_TYPE(CMSG_SPACE(sizeof(struct ucred
)) +
2359 CMSG_SPACE(sizeof(int) * NOTIFY_FD_MAX
)) control
;
2360 struct msghdr msghdr
= {
2363 .msg_control
= &control
,
2364 .msg_controllen
= sizeof(control
),
2367 struct cmsghdr
*cmsg
;
2368 struct ucred
*ucred
= NULL
;
2369 _cleanup_free_ Unit
**array_copy
= NULL
;
2370 _cleanup_strv_free_
char **tags
= NULL
;
2371 Unit
*u1
, *u2
, **array
;
2372 int r
, *fd_array
= NULL
;
2378 assert(m
->notify_fd
== fd
);
2380 if (revents
!= EPOLLIN
) {
2381 log_warning("Got unexpected poll event for notify fd.");
2385 n
= recvmsg_safe(m
->notify_fd
, &msghdr
, MSG_DONTWAIT
|MSG_CMSG_CLOEXEC
|MSG_TRUNC
);
2386 if (IN_SET(n
, -EAGAIN
, -EINTR
))
2387 return 0; /* Spurious wakeup, try again */
2389 log_warning("Got message with truncated control data (too many fds sent?), ignoring.");
2393 /* If this is any other, real error, then let's stop processing this socket. This of course
2394 * means we won't take notification messages anymore, but that's still better than busy
2395 * looping around this: being woken up over and over again but being unable to actually read
2396 * the message off the socket. */
2397 return log_error_errno(n
, "Failed to receive notification message: %m");
2399 CMSG_FOREACH(cmsg
, &msghdr
) {
2400 if (cmsg
->cmsg_level
== SOL_SOCKET
&& cmsg
->cmsg_type
== SCM_RIGHTS
) {
2403 fd_array
= (int*) CMSG_DATA(cmsg
);
2404 n_fds
= (cmsg
->cmsg_len
- CMSG_LEN(0)) / sizeof(int);
2406 } else if (cmsg
->cmsg_level
== SOL_SOCKET
&&
2407 cmsg
->cmsg_type
== SCM_CREDENTIALS
&&
2408 cmsg
->cmsg_len
== CMSG_LEN(sizeof(struct ucred
))) {
2411 ucred
= (struct ucred
*) CMSG_DATA(cmsg
);
2418 r
= fdset_new_array(&fds
, fd_array
, n_fds
);
2420 close_many(fd_array
, n_fds
);
2426 if (!ucred
|| !pid_is_valid(ucred
->pid
)) {
2427 log_warning("Received notify message without valid credentials. Ignoring.");
2431 if ((size_t) n
>= sizeof(buf
) || (msghdr
.msg_flags
& MSG_TRUNC
)) {
2432 log_warning("Received notify message exceeded maximum size. Ignoring.");
2436 /* As extra safety check, let's make sure the string we get doesn't contain embedded NUL bytes. We permit one
2437 * trailing NUL byte in the message, but don't expect it. */
2438 if (n
> 1 && memchr(buf
, 0, n
-1)) {
2439 log_warning("Received notify message with embedded NUL bytes. Ignoring.");
2443 /* Make sure it's NUL-terminated, then parse it to obtain the tags list */
2445 tags
= strv_split_newlines(buf
);
2451 /* possibly a barrier fd, let's see */
2452 if (manager_process_barrier_fd(tags
, fds
))
2455 /* Increase the generation counter used for filtering out duplicate unit invocations. */
2458 /* Notify every unit that might be interested, which might be multiple. */
2459 u1
= manager_get_unit_by_pid_cgroup(m
, ucred
->pid
);
2460 u2
= hashmap_get(m
->watch_pids
, PID_TO_PTR(ucred
->pid
));
2461 array
= hashmap_get(m
->watch_pids
, PID_TO_PTR(-ucred
->pid
));
2468 array_copy
= newdup(Unit
*, array
, k
+1);
2472 /* And now invoke the per-unit callbacks. Note that manager_invoke_notify_message() will handle duplicate units
2473 * make sure we only invoke each unit's handler once. */
2475 manager_invoke_notify_message(m
, u1
, ucred
, tags
, fds
);
2479 manager_invoke_notify_message(m
, u2
, ucred
, tags
, fds
);
2483 for (size_t i
= 0; array_copy
[i
]; i
++) {
2484 manager_invoke_notify_message(m
, array_copy
[i
], ucred
, tags
, fds
);
2489 log_warning("Cannot find unit for notify message of PID "PID_FMT
", ignoring.", ucred
->pid
);
2491 if (fdset_size(fds
) > 0)
2492 log_warning("Got extra auxiliary fds with notification message, closing them.");
2497 static void manager_invoke_sigchld_event(
2500 const siginfo_t
*si
) {
2506 /* Already invoked the handler of this unit in this iteration? Then don't process this again */
2507 if (u
->sigchldgen
== m
->sigchldgen
)
2509 u
->sigchldgen
= m
->sigchldgen
;
2511 log_unit_debug(u
, "Child "PID_FMT
" belongs to %s.", si
->si_pid
, u
->id
);
2512 unit_unwatch_pid(u
, si
->si_pid
);
2514 if (UNIT_VTABLE(u
)->sigchld_event
)
2515 UNIT_VTABLE(u
)->sigchld_event(u
, si
->si_pid
, si
->si_code
, si
->si_status
);
2518 static int manager_dispatch_sigchld(sd_event_source
*source
, void *userdata
) {
2519 Manager
*m
= userdata
;
2526 /* First we call waitid() for a PID and do not reap the zombie. That way we can still access /proc/$PID for it
2527 * while it is a zombie. */
2529 if (waitid(P_ALL
, 0, &si
, WEXITED
|WNOHANG
|WNOWAIT
) < 0) {
2531 if (errno
!= ECHILD
)
2532 log_error_errno(errno
, "Failed to peek for child with waitid(), ignoring: %m");
2540 if (IN_SET(si
.si_code
, CLD_EXITED
, CLD_KILLED
, CLD_DUMPED
)) {
2541 _cleanup_free_ Unit
**array_copy
= NULL
;
2542 _cleanup_free_
char *name
= NULL
;
2543 Unit
*u1
, *u2
, **array
;
2545 (void) get_process_comm(si
.si_pid
, &name
);
2547 log_debug("Child "PID_FMT
" (%s) died (code=%s, status=%i/%s)",
2548 si
.si_pid
, strna(name
),
2549 sigchld_code_to_string(si
.si_code
),
2551 strna(si
.si_code
== CLD_EXITED
2552 ? exit_status_to_string(si
.si_status
, EXIT_STATUS_FULL
)
2553 : signal_to_string(si
.si_status
)));
2555 /* Increase the generation counter used for filtering out duplicate unit invocations */
2558 /* And now figure out the unit this belongs to, it might be multiple... */
2559 u1
= manager_get_unit_by_pid_cgroup(m
, si
.si_pid
);
2560 u2
= hashmap_get(m
->watch_pids
, PID_TO_PTR(si
.si_pid
));
2561 array
= hashmap_get(m
->watch_pids
, PID_TO_PTR(-si
.si_pid
));
2565 /* Count how many entries the array has */
2569 /* Make a copy of the array so that we don't trip up on the array changing beneath us */
2570 array_copy
= newdup(Unit
*, array
, n
+1);
2575 /* Finally, execute them all. Note that u1, u2 and the array might contain duplicates, but
2576 * that's fine, manager_invoke_sigchld_event() will ensure we only invoke the handlers once for
2577 * each iteration. */
2579 /* We check for oom condition, in case we got SIGCHLD before the oom notification.
2580 * We only do this for the cgroup the PID belonged to. */
2581 (void) unit_check_oom(u1
);
2583 /* This only logs for now. In the future when the interface for kills/notifications
2584 * is more stable we can extend service results table similar to how kernel oom kills
2586 (void) unit_check_oomd_kill(u1
);
2588 manager_invoke_sigchld_event(m
, u1
, &si
);
2591 manager_invoke_sigchld_event(m
, u2
, &si
);
2593 for (size_t i
= 0; array_copy
[i
]; i
++)
2594 manager_invoke_sigchld_event(m
, array_copy
[i
], &si
);
2597 /* And now, we actually reap the zombie. */
2598 if (waitid(P_PID
, si
.si_pid
, &si
, WEXITED
) < 0) {
2599 log_error_errno(errno
, "Failed to dequeue child, ignoring: %m");
2606 /* All children processed for now, turn off event source */
2608 r
= sd_event_source_set_enabled(m
->sigchld_event_source
, SD_EVENT_OFF
);
2610 return log_error_errno(r
, "Failed to disable SIGCHLD event source: %m");
2615 static void manager_start_target(Manager
*m
, const char *name
, JobMode mode
) {
2616 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
2619 log_debug("Activating special unit %s", name
);
2621 r
= manager_add_job_by_name(m
, JOB_START
, name
, mode
, NULL
, &error
, NULL
);
2623 log_error("Failed to enqueue %s job: %s", name
, bus_error_message(&error
, r
));
2626 static void manager_handle_ctrl_alt_del(Manager
*m
) {
2627 /* If the user presses C-A-D more than
2628 * 7 times within 2s, we reboot/shutdown immediately,
2629 * unless it was disabled in system.conf */
2631 if (ratelimit_below(&m
->ctrl_alt_del_ratelimit
) || m
->cad_burst_action
== EMERGENCY_ACTION_NONE
)
2632 manager_start_target(m
, SPECIAL_CTRL_ALT_DEL_TARGET
, JOB_REPLACE_IRREVERSIBLY
);
2634 emergency_action(m
, m
->cad_burst_action
, EMERGENCY_ACTION_WARN
, NULL
, -1,
2635 "Ctrl-Alt-Del was pressed more than 7 times within 2s");
2638 static int manager_dispatch_signal_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
2639 Manager
*m
= userdata
;
2641 struct signalfd_siginfo sfsi
;
2645 assert(m
->signal_fd
== fd
);
2647 if (revents
!= EPOLLIN
) {
2648 log_warning("Got unexpected events from signal file descriptor.");
2652 n
= read(m
->signal_fd
, &sfsi
, sizeof(sfsi
));
2653 if (n
!= sizeof(sfsi
)) {
2655 log_warning("Truncated read from signal fd (%zu bytes), ignoring!", n
);
2659 if (IN_SET(errno
, EINTR
, EAGAIN
))
2662 /* We return an error here, which will kill this handler,
2663 * to avoid a busy loop on read error. */
2664 return log_error_errno(errno
, "Reading from signal fd failed: %m");
2667 log_received_signal(sfsi
.ssi_signo
== SIGCHLD
||
2668 (sfsi
.ssi_signo
== SIGTERM
&& MANAGER_IS_USER(m
))
2669 ? LOG_DEBUG
: LOG_INFO
,
2672 switch (sfsi
.ssi_signo
) {
2675 r
= sd_event_source_set_enabled(m
->sigchld_event_source
, SD_EVENT_ON
);
2677 log_warning_errno(r
, "Failed to enable SIGCHLD event source, ignoring: %m");
2682 if (MANAGER_IS_SYSTEM(m
)) {
2683 /* This is for compatibility with the original sysvinit */
2684 if (verify_run_space_and_log("Refusing to reexecute") < 0)
2687 m
->objective
= MANAGER_REEXECUTE
;
2693 if (MANAGER_IS_SYSTEM(m
))
2694 manager_handle_ctrl_alt_del(m
);
2696 manager_start_target(m
, SPECIAL_EXIT_TARGET
,
2697 JOB_REPLACE_IRREVERSIBLY
);
2701 /* This is a nop on non-init */
2702 if (MANAGER_IS_SYSTEM(m
))
2703 manager_start_target(m
, SPECIAL_KBREQUEST_TARGET
, JOB_REPLACE
);
2708 /* This is a nop on non-init */
2709 if (MANAGER_IS_SYSTEM(m
))
2710 manager_start_target(m
, SPECIAL_SIGPWR_TARGET
, JOB_REPLACE
);
2715 if (manager_dbus_is_running(m
, false)) {
2716 log_info("Trying to reconnect to bus...");
2718 (void) bus_init_api(m
);
2720 if (MANAGER_IS_SYSTEM(m
))
2721 (void) bus_init_system(m
);
2723 log_info("Starting D-Bus service...");
2724 manager_start_target(m
, SPECIAL_DBUS_SERVICE
, JOB_REPLACE
);
2730 _cleanup_free_
char *dump
= NULL
;
2732 r
= manager_get_dump_string(m
, &dump
);
2734 log_warning_errno(errno
, "Failed to acquire manager dump: %m");
2738 log_dump(LOG_INFO
, dump
);
2743 if (verify_run_space_and_log("Refusing to reload") < 0)
2746 m
->objective
= MANAGER_RELOAD
;
2751 /* Starting SIGRTMIN+0 */
2752 static const struct {
2755 } target_table
[] = {
2756 [0] = { SPECIAL_DEFAULT_TARGET
, JOB_ISOLATE
},
2757 [1] = { SPECIAL_RESCUE_TARGET
, JOB_ISOLATE
},
2758 [2] = { SPECIAL_EMERGENCY_TARGET
, JOB_ISOLATE
},
2759 [3] = { SPECIAL_HALT_TARGET
, JOB_REPLACE_IRREVERSIBLY
},
2760 [4] = { SPECIAL_POWEROFF_TARGET
, JOB_REPLACE_IRREVERSIBLY
},
2761 [5] = { SPECIAL_REBOOT_TARGET
, JOB_REPLACE_IRREVERSIBLY
},
2762 [6] = { SPECIAL_KEXEC_TARGET
, JOB_REPLACE_IRREVERSIBLY
},
2765 /* Starting SIGRTMIN+13, so that target halt and system halt are 10 apart */
2766 static const ManagerObjective objective_table
[] = {
2768 [1] = MANAGER_POWEROFF
,
2769 [2] = MANAGER_REBOOT
,
2770 [3] = MANAGER_KEXEC
,
2773 if ((int) sfsi
.ssi_signo
>= SIGRTMIN
+0 &&
2774 (int) sfsi
.ssi_signo
< SIGRTMIN
+(int) ELEMENTSOF(target_table
)) {
2775 int idx
= (int) sfsi
.ssi_signo
- SIGRTMIN
;
2776 manager_start_target(m
, target_table
[idx
].target
,
2777 target_table
[idx
].mode
);
2781 if ((int) sfsi
.ssi_signo
>= SIGRTMIN
+13 &&
2782 (int) sfsi
.ssi_signo
< SIGRTMIN
+13+(int) ELEMENTSOF(objective_table
)) {
2783 m
->objective
= objective_table
[sfsi
.ssi_signo
- SIGRTMIN
- 13];
2787 switch (sfsi
.ssi_signo
- SIGRTMIN
) {
2790 manager_override_show_status(m
, SHOW_STATUS_YES
, "signal");
2794 manager_override_show_status(m
, SHOW_STATUS_NO
, "signal");
2798 manager_override_log_level(m
, LOG_DEBUG
);
2802 manager_restore_original_log_level(m
);
2806 if (MANAGER_IS_USER(m
)) {
2807 m
->objective
= MANAGER_EXIT
;
2811 /* This is a nop on init */
2815 case 29: /* compatibility: used to be mapped to LOG_TARGET_SYSLOG_OR_KMSG */
2816 manager_restore_original_log_target(m
);
2820 manager_override_log_target(m
, LOG_TARGET_CONSOLE
);
2824 manager_override_log_target(m
, LOG_TARGET_KMSG
);
2828 log_warning("Got unhandled signal <%s>.", signal_to_string(sfsi
.ssi_signo
));
2835 static int manager_dispatch_time_change_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
2836 Manager
*m
= userdata
;
2840 assert(m
->time_change_fd
== fd
);
2842 log_struct(LOG_DEBUG
,
2843 "MESSAGE_ID=" SD_MESSAGE_TIME_CHANGE_STR
,
2844 LOG_MESSAGE("Time has been changed"));
2846 /* Restart the watch */
2847 (void) manager_setup_time_change(m
);
2849 HASHMAP_FOREACH(u
, m
->units
)
2850 if (UNIT_VTABLE(u
)->time_change
)
2851 UNIT_VTABLE(u
)->time_change(u
);
2856 static int manager_dispatch_timezone_change(
2857 sd_event_source
*source
,
2858 const struct inotify_event
*e
,
2861 Manager
*m
= userdata
;
2867 log_debug("inotify event for /etc/localtime");
2869 changed
= manager_read_timezone_stat(m
);
2873 /* Something changed, restart the watch, to ensure we watch the new /etc/localtime if it changed */
2874 (void) manager_setup_timezone_change(m
);
2876 /* Read the new timezone */
2879 log_debug("Timezone has been changed (now: %s).", tzname
[daylight
]);
2881 HASHMAP_FOREACH(u
, m
->units
)
2882 if (UNIT_VTABLE(u
)->timezone_change
)
2883 UNIT_VTABLE(u
)->timezone_change(u
);
2888 static int manager_dispatch_idle_pipe_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
2889 Manager
*m
= userdata
;
2892 assert(m
->idle_pipe
[2] == fd
);
2894 /* There's at least one Type=idle child that just gave up on us waiting for the boot process to complete. Let's
2895 * now turn off any further console output if there's at least one service that needs console access, so that
2896 * from now on our own output should not spill into that service's output anymore. After all, we support
2897 * Type=idle only to beautify console output and it generally is set on services that want to own the console
2898 * exclusively without our interference. */
2899 m
->no_console_output
= m
->n_on_console
> 0;
2901 /* Acknowledge the child's request, and let all all other children know too that they shouldn't wait any longer
2902 * by closing the pipes towards them, which is what they are waiting for. */
2903 manager_close_idle_pipe(m
);
2908 static int manager_dispatch_jobs_in_progress(sd_event_source
*source
, usec_t usec
, void *userdata
) {
2909 Manager
*m
= userdata
;
2915 manager_print_jobs_in_progress(m
);
2917 r
= sd_event_source_set_time_relative(source
, JOBS_IN_PROGRESS_PERIOD_USEC
);
2921 return sd_event_source_set_enabled(source
, SD_EVENT_ONESHOT
);
2924 int manager_loop(Manager
*m
) {
2925 RateLimit rl
= { .interval
= 1*USEC_PER_SEC
, .burst
= 50000 };
2929 assert(m
->objective
== MANAGER_OK
); /* Ensure manager_startup() has been called */
2931 manager_check_finished(m
);
2933 /* There might still be some zombies hanging around from before we were exec()'ed. Let's reap them. */
2934 r
= sd_event_source_set_enabled(m
->sigchld_event_source
, SD_EVENT_ON
);
2936 return log_error_errno(r
, "Failed to enable SIGCHLD event source: %m");
2938 while (m
->objective
== MANAGER_OK
) {
2939 usec_t wait_usec
, watchdog_usec
;
2941 watchdog_usec
= manager_get_watchdog(m
, WATCHDOG_RUNTIME
);
2942 if (m
->runtime_watchdog_running
)
2943 (void) watchdog_ping();
2944 else if (timestamp_is_set(watchdog_usec
))
2945 manager_retry_runtime_watchdog(m
);
2947 if (!ratelimit_below(&rl
)) {
2948 /* Yay, something is going seriously wrong, pause a little */
2949 log_warning("Looping too fast. Throttling execution a little.");
2953 if (manager_dispatch_load_queue(m
) > 0)
2956 if (manager_dispatch_gc_job_queue(m
) > 0)
2959 if (manager_dispatch_gc_unit_queue(m
) > 0)
2962 if (manager_dispatch_cleanup_queue(m
) > 0)
2965 if (manager_dispatch_cgroup_realize_queue(m
) > 0)
2968 if (manager_dispatch_start_when_upheld_queue(m
) > 0)
2971 if (manager_dispatch_stop_when_bound_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(
3189 const char *field_name
) {
3196 /* Serialize the UID reference table. Or actually, just the IPC destruction flag of it, as
3197 * the actual counter of it is better rebuild after a reload/reexec. */
3199 HASHMAP_FOREACH_KEY(p
, k
, uid_refs
) {
3203 uid
= PTR_TO_UID(k
);
3204 c
= PTR_TO_UINT32(p
);
3206 if (!(c
& DESTROY_IPC_FLAG
))
3209 (void) serialize_item_format(f
, field_name
, UID_FMT
, uid
);
3213 static void manager_serialize_uid_refs(Manager
*m
, FILE *f
) {
3214 manager_serialize_uid_refs_internal(f
, m
->uid_refs
, "destroy-ipc-uid");
3217 static void manager_serialize_gid_refs(Manager
*m
, FILE *f
) {
3218 manager_serialize_uid_refs_internal(f
, m
->gid_refs
, "destroy-ipc-gid");
3221 int manager_serialize(
3225 bool switching_root
) {
3235 _cleanup_(manager_reloading_stopp
) _unused_ Manager
*reloading
= manager_reloading_start(m
);
3237 (void) serialize_item_format(f
, "current-job-id", "%" PRIu32
, m
->current_job_id
);
3238 (void) serialize_item_format(f
, "n-installed-jobs", "%u", m
->n_installed_jobs
);
3239 (void) serialize_item_format(f
, "n-failed-jobs", "%u", m
->n_failed_jobs
);
3240 (void) serialize_bool(f
, "taint-usr", m
->taint_usr
);
3241 (void) serialize_bool(f
, "ready-sent", m
->ready_sent
);
3242 (void) serialize_bool(f
, "taint-logged", m
->taint_logged
);
3243 (void) serialize_bool(f
, "service-watchdogs", m
->service_watchdogs
);
3245 /* After switching root, udevd has not been started yet. So, enumeration results should not be emitted. */
3246 (void) serialize_bool(f
, "honor-device-enumeration", !switching_root
);
3248 if (m
->show_status_overridden
!= _SHOW_STATUS_INVALID
)
3249 (void) serialize_item(f
, "show-status-overridden",
3250 show_status_to_string(m
->show_status_overridden
));
3252 if (m
->log_level_overridden
)
3253 (void) serialize_item_format(f
, "log-level-override", "%i", log_get_max_level());
3254 if (m
->log_target_overridden
)
3255 (void) serialize_item(f
, "log-target-override", log_target_to_string(log_get_target()));
3257 (void) serialize_usec(f
, "runtime-watchdog-overridden", m
->watchdog_overridden
[WATCHDOG_RUNTIME
]);
3258 (void) serialize_usec(f
, "reboot-watchdog-overridden", m
->watchdog_overridden
[WATCHDOG_REBOOT
]);
3259 (void) serialize_usec(f
, "kexec-watchdog-overridden", m
->watchdog_overridden
[WATCHDOG_KEXEC
]);
3261 for (ManagerTimestamp q
= 0; q
< _MANAGER_TIMESTAMP_MAX
; q
++) {
3262 _cleanup_free_
char *joined
= NULL
;
3264 if (!manager_timestamp_shall_serialize(q
))
3267 joined
= strjoin(manager_timestamp_to_string(q
), "-timestamp");
3271 (void) serialize_dual_timestamp(f
, joined
, m
->timestamps
+ q
);
3274 if (!switching_root
)
3275 (void) serialize_strv(f
, "env", m
->client_environment
);
3277 if (m
->notify_fd
>= 0) {
3278 r
= serialize_fd(f
, fds
, "notify-fd", m
->notify_fd
);
3282 (void) serialize_item(f
, "notify-socket", m
->notify_socket
);
3285 if (m
->cgroups_agent_fd
>= 0) {
3286 r
= serialize_fd(f
, fds
, "cgroups-agent-fd", m
->cgroups_agent_fd
);
3291 if (m
->user_lookup_fds
[0] >= 0) {
3294 copy0
= fdset_put_dup(fds
, m
->user_lookup_fds
[0]);
3296 return log_error_errno(copy0
, "Failed to add user lookup fd to serialization: %m");
3298 copy1
= fdset_put_dup(fds
, m
->user_lookup_fds
[1]);
3300 return log_error_errno(copy1
, "Failed to add user lookup fd to serialization: %m");
3302 (void) serialize_item_format(f
, "user-lookup", "%i %i", copy0
, copy1
);
3305 bus_track_serialize(m
->subscribed
, f
, "subscribed");
3307 r
= dynamic_user_serialize(m
, f
, fds
);
3311 manager_serialize_uid_refs(m
, f
);
3312 manager_serialize_gid_refs(m
, f
);
3314 r
= exec_runtime_serialize(m
, f
, fds
);
3318 (void) fputc('\n', f
);
3320 HASHMAP_FOREACH_KEY(u
, t
, m
->units
) {
3324 r
= unit_serialize(u
, f
, fds
, switching_root
);
3329 r
= fflush_and_check(f
);
3331 return log_error_errno(r
, "Failed to flush serialization: %m");
3333 r
= bus_fdset_add_all(m
, fds
);
3335 return log_error_errno(r
, "Failed to add bus sockets to serialization: %m");
3340 static int manager_deserialize_one_unit(Manager
*m
, const char *name
, FILE *f
, FDSet
*fds
) {
3344 r
= manager_load_unit(m
, name
, NULL
, NULL
, &u
);
3348 return log_notice_errno(r
, "Failed to load unit \"%s\", skipping deserialization: %m", name
);
3351 r
= unit_deserialize(u
, f
, fds
);
3355 return log_notice_errno(r
, "Failed to deserialize unit \"%s\", skipping: %m", name
);
3361 static int manager_deserialize_units(Manager
*m
, FILE *f
, FDSet
*fds
) {
3362 const char *unit_name
;
3366 _cleanup_free_
char *line
= NULL
;
3368 r
= read_line(f
, LONG_LINE_MAX
, &line
);
3370 return log_error_errno(r
, "Failed to read serialization line: %m");
3374 unit_name
= strstrip(line
);
3376 r
= manager_deserialize_one_unit(m
, unit_name
, f
, fds
);
3380 r
= unit_deserialize_skip(f
);
3389 usec_t
manager_get_watchdog(Manager
*m
, WatchdogType t
) {
3392 if (MANAGER_IS_USER(m
))
3393 return USEC_INFINITY
;
3395 if (timestamp_is_set(m
->watchdog_overridden
[t
]))
3396 return m
->watchdog_overridden
[t
];
3398 return m
->watchdog
[t
];
3401 void manager_set_watchdog(Manager
*m
, WatchdogType t
, usec_t timeout
) {
3406 if (MANAGER_IS_USER(m
))
3409 if (m
->watchdog
[t
] == timeout
)
3412 if (t
== WATCHDOG_RUNTIME
)
3413 if (!timestamp_is_set(m
->watchdog_overridden
[WATCHDOG_RUNTIME
])) {
3414 if (timestamp_is_set(timeout
)) {
3415 r
= watchdog_set_timeout(&timeout
);
3418 m
->runtime_watchdog_running
= true;
3420 watchdog_close(true);
3421 m
->runtime_watchdog_running
= false;
3425 m
->watchdog
[t
] = timeout
;
3428 int manager_override_watchdog(Manager
*m
, WatchdogType t
, usec_t timeout
) {
3433 if (MANAGER_IS_USER(m
))
3436 if (m
->watchdog_overridden
[t
] == timeout
)
3439 if (t
== WATCHDOG_RUNTIME
) {
3442 p
= timestamp_is_set(timeout
) ? &timeout
: &m
->watchdog
[t
];
3443 if (timestamp_is_set(*p
)) {
3444 r
= watchdog_set_timeout(p
);
3447 m
->runtime_watchdog_running
= true;
3449 watchdog_close(true);
3450 m
->runtime_watchdog_running
= false;
3454 m
->watchdog_overridden
[t
] = timeout
;
3459 void manager_retry_runtime_watchdog(Manager
*m
) {
3464 if (timestamp_is_set(m
->watchdog_overridden
[WATCHDOG_RUNTIME
]))
3465 r
= watchdog_set_timeout(&m
->watchdog_overridden
[WATCHDOG_RUNTIME
]);
3467 r
= watchdog_set_timeout(&m
->watchdog
[WATCHDOG_RUNTIME
]);
3470 m
->runtime_watchdog_running
= true;
3473 static void manager_deserialize_uid_refs_one_internal(
3475 const char *value
) {
3484 r
= parse_uid(value
, &uid
);
3485 if (r
< 0 || uid
== 0) {
3486 log_debug("Unable to parse UID/GID reference serialization: " UID_FMT
, uid
);
3490 if (hashmap_ensure_allocated(uid_refs
, &trivial_hash_ops
) < 0) {
3495 c
= PTR_TO_UINT32(hashmap_get(*uid_refs
, UID_TO_PTR(uid
)));
3496 if (c
& DESTROY_IPC_FLAG
)
3499 c
|= DESTROY_IPC_FLAG
;
3501 r
= hashmap_replace(*uid_refs
, UID_TO_PTR(uid
), UINT32_TO_PTR(c
));
3503 log_debug_errno(r
, "Failed to add UID/GID reference entry: %m");
3508 static void manager_deserialize_uid_refs_one(Manager
*m
, const char *value
) {
3509 manager_deserialize_uid_refs_one_internal(&m
->uid_refs
, value
);
3512 static void manager_deserialize_gid_refs_one(Manager
*m
, const char *value
) {
3513 manager_deserialize_uid_refs_one_internal(&m
->gid_refs
, value
);
3516 int manager_deserialize(Manager
*m
, FILE *f
, FDSet
*fds
) {
3522 if (DEBUG_LOGGING
) {
3523 if (fdset_isempty(fds
))
3524 log_debug("No file descriptors passed");
3528 FDSET_FOREACH(fd
, fds
) {
3529 _cleanup_free_
char *fn
= NULL
;
3531 r
= fd_get_path(fd
, &fn
);
3533 log_debug_errno(r
, "Received serialized fd %i → %m", fd
);
3535 log_debug("Received serialized fd %i → %s", fd
, strna(fn
));
3540 log_debug("Deserializing state...");
3542 /* If we are not in reload mode yet, enter it now. Not that this is recursive, a caller might already have
3543 * increased it to non-zero, which is why we just increase it by one here and down again at the end of this
3545 _cleanup_(manager_reloading_stopp
) _unused_ Manager
*reloading
= manager_reloading_start(m
);
3548 _cleanup_free_
char *line
= NULL
;
3549 const char *val
, *l
;
3551 r
= read_line(f
, LONG_LINE_MAX
, &line
);
3553 return log_error_errno(r
, "Failed to read serialization line: %m");
3558 if (isempty(l
)) /* end marker */
3561 if ((val
= startswith(l
, "current-job-id="))) {
3564 if (safe_atou32(val
, &id
) < 0)
3565 log_notice("Failed to parse current job id value '%s', ignoring.", val
);
3567 m
->current_job_id
= MAX(m
->current_job_id
, id
);
3569 } else if ((val
= startswith(l
, "n-installed-jobs="))) {
3572 if (safe_atou32(val
, &n
) < 0)
3573 log_notice("Failed to parse installed jobs counter '%s', ignoring.", val
);
3575 m
->n_installed_jobs
+= n
;
3577 } else if ((val
= startswith(l
, "n-failed-jobs="))) {
3580 if (safe_atou32(val
, &n
) < 0)
3581 log_notice("Failed to parse failed jobs counter '%s', ignoring.", val
);
3583 m
->n_failed_jobs
+= n
;
3585 } else if ((val
= startswith(l
, "taint-usr="))) {
3588 b
= parse_boolean(val
);
3590 log_notice("Failed to parse taint /usr flag '%s', ignoring.", val
);
3592 m
->taint_usr
= m
->taint_usr
|| b
;
3594 } else if ((val
= startswith(l
, "ready-sent="))) {
3597 b
= parse_boolean(val
);
3599 log_notice("Failed to parse ready-sent flag '%s', ignoring.", val
);
3601 m
->ready_sent
= m
->ready_sent
|| b
;
3603 } else if ((val
= startswith(l
, "taint-logged="))) {
3606 b
= parse_boolean(val
);
3608 log_notice("Failed to parse taint-logged flag '%s', ignoring.", val
);
3610 m
->taint_logged
= m
->taint_logged
|| b
;
3612 } else if ((val
= startswith(l
, "service-watchdogs="))) {
3615 b
= parse_boolean(val
);
3617 log_notice("Failed to parse service-watchdogs flag '%s', ignoring.", val
);
3619 m
->service_watchdogs
= b
;
3621 } else if ((val
= startswith(l
, "honor-device-enumeration="))) {
3624 b
= parse_boolean(val
);
3626 log_notice("Failed to parse honor-device-enumeration flag '%s', ignoring.", val
);
3628 m
->honor_device_enumeration
= b
;
3630 } else if ((val
= startswith(l
, "show-status-overridden="))) {
3633 s
= show_status_from_string(val
);
3635 log_notice("Failed to parse show-status-overridden flag '%s', ignoring.", val
);
3637 manager_override_show_status(m
, s
, "deserialize");
3639 } else if ((val
= startswith(l
, "log-level-override="))) {
3642 level
= log_level_from_string(val
);
3644 log_notice("Failed to parse log-level-override value '%s', ignoring.", val
);
3646 manager_override_log_level(m
, level
);
3648 } else if ((val
= startswith(l
, "log-target-override="))) {
3651 target
= log_target_from_string(val
);
3653 log_notice("Failed to parse log-target-override value '%s', ignoring.", val
);
3655 manager_override_log_target(m
, target
);
3657 } else if ((val
= startswith(l
, "runtime-watchdog-overridden="))) {
3660 if (deserialize_usec(val
, &t
) < 0)
3661 log_notice("Failed to parse runtime-watchdog-overridden value '%s', ignoring.", val
);
3663 manager_override_watchdog(m
, WATCHDOG_RUNTIME
, t
);
3665 } else if ((val
= startswith(l
, "reboot-watchdog-overridden="))) {
3668 if (deserialize_usec(val
, &t
) < 0)
3669 log_notice("Failed to parse reboot-watchdog-overridden value '%s', ignoring.", val
);
3671 manager_override_watchdog(m
, WATCHDOG_REBOOT
, t
);
3673 } else if ((val
= startswith(l
, "kexec-watchdog-overridden="))) {
3676 if (deserialize_usec(val
, &t
) < 0)
3677 log_notice("Failed to parse kexec-watchdog-overridden value '%s', ignoring.", val
);
3679 manager_override_watchdog(m
, WATCHDOG_KEXEC
, t
);
3681 } else if (startswith(l
, "env=")) {
3682 r
= deserialize_environment(l
+ 4, &m
->client_environment
);
3684 log_notice_errno(r
, "Failed to parse environment entry: \"%s\", ignoring: %m", l
);
3686 } else if ((val
= startswith(l
, "notify-fd="))) {
3689 if (safe_atoi(val
, &fd
) < 0 || fd
< 0 || !fdset_contains(fds
, fd
))
3690 log_notice("Failed to parse notify fd, ignoring: \"%s\"", val
);
3692 m
->notify_event_source
= sd_event_source_unref(m
->notify_event_source
);
3693 safe_close(m
->notify_fd
);
3694 m
->notify_fd
= fdset_remove(fds
, fd
);
3697 } else if ((val
= startswith(l
, "notify-socket="))) {
3698 r
= free_and_strdup(&m
->notify_socket
, val
);
3702 } else if ((val
= startswith(l
, "cgroups-agent-fd="))) {
3705 if (safe_atoi(val
, &fd
) < 0 || fd
< 0 || !fdset_contains(fds
, fd
))
3706 log_notice("Failed to parse cgroups agent fd, ignoring.: %s", val
);
3708 m
->cgroups_agent_event_source
= sd_event_source_unref(m
->cgroups_agent_event_source
);
3709 safe_close(m
->cgroups_agent_fd
);
3710 m
->cgroups_agent_fd
= fdset_remove(fds
, fd
);
3713 } else if ((val
= startswith(l
, "user-lookup="))) {
3716 if (sscanf(val
, "%i %i", &fd0
, &fd1
) != 2 || fd0
< 0 || fd1
< 0 || fd0
== fd1
|| !fdset_contains(fds
, fd0
) || !fdset_contains(fds
, fd1
))
3717 log_notice("Failed to parse user lookup fd, ignoring: %s", val
);
3719 m
->user_lookup_event_source
= sd_event_source_unref(m
->user_lookup_event_source
);
3720 safe_close_pair(m
->user_lookup_fds
);
3721 m
->user_lookup_fds
[0] = fdset_remove(fds
, fd0
);
3722 m
->user_lookup_fds
[1] = fdset_remove(fds
, fd1
);
3725 } else if ((val
= startswith(l
, "dynamic-user=")))
3726 dynamic_user_deserialize_one(m
, val
, fds
);
3727 else if ((val
= startswith(l
, "destroy-ipc-uid=")))
3728 manager_deserialize_uid_refs_one(m
, val
);
3729 else if ((val
= startswith(l
, "destroy-ipc-gid=")))
3730 manager_deserialize_gid_refs_one(m
, val
);
3731 else if ((val
= startswith(l
, "exec-runtime=")))
3732 (void) exec_runtime_deserialize_one(m
, val
, fds
);
3733 else if ((val
= startswith(l
, "subscribed="))) {
3735 if (strv_extend(&m
->deserialized_subscribed
, val
) < 0)
3741 for (q
= 0; q
< _MANAGER_TIMESTAMP_MAX
; q
++) {
3742 val
= startswith(l
, manager_timestamp_to_string(q
));
3746 val
= startswith(val
, "-timestamp=");
3751 if (q
< _MANAGER_TIMESTAMP_MAX
) /* found it */
3752 (void) deserialize_dual_timestamp(val
, m
->timestamps
+ q
);
3753 else if (!startswith(l
, "kdbus-fd=")) /* ignore kdbus */
3754 log_notice("Unknown serialization item '%s', ignoring.", l
);
3758 return manager_deserialize_units(m
, f
, fds
);
3761 int manager_reload(Manager
*m
) {
3762 _cleanup_(manager_reloading_stopp
) Manager
*reloading
= NULL
;
3763 _cleanup_fdset_free_ FDSet
*fds
= NULL
;
3764 _cleanup_fclose_
FILE *f
= NULL
;
3769 r
= manager_open_serialization(m
, &f
);
3771 return log_error_errno(r
, "Failed to create serialization file: %m");
3777 /* We are officially in reload mode from here on. */
3778 reloading
= manager_reloading_start(m
);
3780 r
= manager_serialize(m
, f
, fds
, false);
3784 if (fseeko(f
, 0, SEEK_SET
) < 0)
3785 return log_error_errno(errno
, "Failed to seek to beginning of serialization: %m");
3787 /* 💀 This is the point of no return, from here on there is no way back. 💀 */
3790 bus_manager_send_reloading(m
, true);
3792 /* Start by flushing out all jobs and units, all generated units, all runtime environments, all dynamic users
3793 * and everything else that is worth flushing out. We'll get it all back from the serialization — if we need
3796 manager_clear_jobs_and_units(m
);
3797 lookup_paths_flush_generator(&m
->lookup_paths
);
3798 lookup_paths_free(&m
->lookup_paths
);
3799 exec_runtime_vacuum(m
);
3800 dynamic_user_vacuum(m
, false);
3801 m
->uid_refs
= hashmap_free(m
->uid_refs
);
3802 m
->gid_refs
= hashmap_free(m
->gid_refs
);
3804 r
= lookup_paths_init(&m
->lookup_paths
, m
->unit_file_scope
, 0, NULL
);
3806 log_warning_errno(r
, "Failed to initialize path lookup table, ignoring: %m");
3808 (void) manager_run_environment_generators(m
);
3809 (void) manager_run_generators(m
);
3811 lookup_paths_log(&m
->lookup_paths
);
3813 /* We flushed out generated files, for which we don't watch mtime, so we should flush the old map. */
3814 manager_free_unit_name_maps(m
);
3816 /* First, enumerate what we can from kernel and suchlike */
3817 manager_enumerate_perpetual(m
);
3818 manager_enumerate(m
);
3820 /* Second, deserialize our stored data */
3821 r
= manager_deserialize(m
, f
, fds
);
3823 log_warning_errno(r
, "Deserialization failed, proceeding anyway: %m");
3825 /* We don't need the serialization anymore */
3828 /* Re-register notify_fd as event source, and set up other sockets/communication channels we might need */
3829 (void) manager_setup_notify(m
);
3830 (void) manager_setup_cgroups_agent(m
);
3831 (void) manager_setup_user_lookup_fd(m
);
3833 /* Third, fire things up! */
3834 manager_coldplug(m
);
3836 /* Clean up runtime objects no longer referenced */
3839 /* Clean up deserialized tracked clients */
3840 m
->deserialized_subscribed
= strv_free(m
->deserialized_subscribed
);
3842 /* Consider the reload process complete now. */
3843 assert(m
->n_reloading
> 0);
3846 /* On manager reloading, device tag data should exists, thus, we should honor the results of device
3847 * enumeration. The flag should be always set correctly by the serialized data, but it may fail. So,
3848 * let's always set the flag here for safety. */
3849 m
->honor_device_enumeration
= true;
3853 m
->send_reloading_done
= true;
3857 void manager_reset_failed(Manager
*m
) {
3862 HASHMAP_FOREACH(u
, m
->units
)
3863 unit_reset_failed(u
);
3866 bool manager_unit_inactive_or_pending(Manager
*m
, const char *name
) {
3872 /* Returns true if the unit is inactive or going down */
3873 u
= manager_get_unit(m
, name
);
3877 return unit_inactive_or_pending(u
);
3880 static void log_taint_string(Manager
*m
) {
3881 _cleanup_free_
char *taint
= NULL
;
3885 if (MANAGER_IS_USER(m
) || m
->taint_logged
)
3888 m
->taint_logged
= true; /* only check for taint once */
3890 taint
= manager_taint_string(m
);
3894 log_struct(LOG_NOTICE
,
3895 LOG_MESSAGE("System is tainted: %s", taint
),
3897 "MESSAGE_ID=" SD_MESSAGE_TAINTED_STR
);
3900 static void manager_notify_finished(Manager
*m
) {
3901 char userspace
[FORMAT_TIMESPAN_MAX
], initrd
[FORMAT_TIMESPAN_MAX
], kernel
[FORMAT_TIMESPAN_MAX
], sum
[FORMAT_TIMESPAN_MAX
];
3902 usec_t firmware_usec
, loader_usec
, kernel_usec
, initrd_usec
, userspace_usec
, total_usec
;
3904 if (MANAGER_IS_TEST_RUN(m
))
3907 if (MANAGER_IS_SYSTEM(m
) && detect_container() <= 0) {
3908 char ts
[FORMAT_TIMESPAN_MAX
];
3909 char buf
[FORMAT_TIMESPAN_MAX
+ STRLEN(" (firmware) + ") + FORMAT_TIMESPAN_MAX
+ STRLEN(" (loader) + ")]
3912 size_t size
= sizeof buf
;
3914 /* Note that MANAGER_TIMESTAMP_KERNEL's monotonic value is always at 0, and
3915 * MANAGER_TIMESTAMP_FIRMWARE's and MANAGER_TIMESTAMP_LOADER's monotonic value should be considered
3916 * negative values. */
3918 firmware_usec
= m
->timestamps
[MANAGER_TIMESTAMP_FIRMWARE
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_LOADER
].monotonic
;
3919 loader_usec
= m
->timestamps
[MANAGER_TIMESTAMP_LOADER
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_KERNEL
].monotonic
;
3920 userspace_usec
= m
->timestamps
[MANAGER_TIMESTAMP_FINISH
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_USERSPACE
].monotonic
;
3921 total_usec
= m
->timestamps
[MANAGER_TIMESTAMP_FIRMWARE
].monotonic
+ m
->timestamps
[MANAGER_TIMESTAMP_FINISH
].monotonic
;
3923 if (firmware_usec
> 0)
3924 size
= strpcpyf(&p
, size
, "%s (firmware) + ", format_timespan(ts
, sizeof(ts
), firmware_usec
, USEC_PER_MSEC
));
3925 if (loader_usec
> 0)
3926 size
= strpcpyf(&p
, size
, "%s (loader) + ", format_timespan(ts
, sizeof(ts
), loader_usec
, USEC_PER_MSEC
));
3928 if (dual_timestamp_is_set(&m
->timestamps
[MANAGER_TIMESTAMP_INITRD
])) {
3930 /* The initrd case on bare-metal */
3931 kernel_usec
= m
->timestamps
[MANAGER_TIMESTAMP_INITRD
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_KERNEL
].monotonic
;
3932 initrd_usec
= m
->timestamps
[MANAGER_TIMESTAMP_USERSPACE
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_INITRD
].monotonic
;
3934 log_struct(LOG_INFO
,
3935 "MESSAGE_ID=" SD_MESSAGE_STARTUP_FINISHED_STR
,
3936 "KERNEL_USEC="USEC_FMT
, kernel_usec
,
3937 "INITRD_USEC="USEC_FMT
, initrd_usec
,
3938 "USERSPACE_USEC="USEC_FMT
, userspace_usec
,
3939 LOG_MESSAGE("Startup finished in %s%s (kernel) + %s (initrd) + %s (userspace) = %s.",
3941 format_timespan(kernel
, sizeof(kernel
), kernel_usec
, USEC_PER_MSEC
),
3942 format_timespan(initrd
, sizeof(initrd
), initrd_usec
, USEC_PER_MSEC
),
3943 format_timespan(userspace
, sizeof(userspace
), userspace_usec
, USEC_PER_MSEC
),
3944 format_timespan(sum
, sizeof(sum
), total_usec
, USEC_PER_MSEC
)));
3946 /* The initrd-less case on bare-metal */
3948 kernel_usec
= m
->timestamps
[MANAGER_TIMESTAMP_USERSPACE
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_KERNEL
].monotonic
;
3951 log_struct(LOG_INFO
,
3952 "MESSAGE_ID=" SD_MESSAGE_STARTUP_FINISHED_STR
,
3953 "KERNEL_USEC="USEC_FMT
, kernel_usec
,
3954 "USERSPACE_USEC="USEC_FMT
, userspace_usec
,
3955 LOG_MESSAGE("Startup finished in %s%s (kernel) + %s (userspace) = %s.",
3957 format_timespan(kernel
, sizeof(kernel
), kernel_usec
, USEC_PER_MSEC
),
3958 format_timespan(userspace
, sizeof(userspace
), userspace_usec
, USEC_PER_MSEC
),
3959 format_timespan(sum
, sizeof(sum
), total_usec
, USEC_PER_MSEC
)));
3962 /* The container and --user case */
3963 firmware_usec
= loader_usec
= initrd_usec
= kernel_usec
= 0;
3964 total_usec
= userspace_usec
= m
->timestamps
[MANAGER_TIMESTAMP_FINISH
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_USERSPACE
].monotonic
;
3966 log_struct(LOG_INFO
,
3967 "MESSAGE_ID=" SD_MESSAGE_USER_STARTUP_FINISHED_STR
,
3968 "USERSPACE_USEC="USEC_FMT
, userspace_usec
,
3969 LOG_MESSAGE("Startup finished in %s.",
3970 format_timespan(sum
, sizeof(sum
), total_usec
, USEC_PER_MSEC
)));
3973 bus_manager_send_finished(m
, firmware_usec
, loader_usec
, kernel_usec
, initrd_usec
, userspace_usec
, total_usec
);
3976 m
->ready_sent
? "STATUS=Startup finished in %s."
3978 "STATUS=Startup finished in %s.",
3979 format_timespan(sum
, sizeof(sum
), total_usec
, USEC_PER_MSEC
));
3980 m
->ready_sent
= true;
3982 log_taint_string(m
);
3985 static void manager_send_ready(Manager
*m
) {
3988 /* We send READY=1 on reaching basic.target only when running in --user mode. */
3989 if (!MANAGER_IS_USER(m
) || m
->ready_sent
)
3992 m
->ready_sent
= true;
3996 "STATUS=Reached " SPECIAL_BASIC_TARGET
".");
3999 static void manager_check_basic_target(Manager
*m
) {
4004 /* Small shortcut */
4005 if (m
->ready_sent
&& m
->taint_logged
)
4008 u
= manager_get_unit(m
, SPECIAL_BASIC_TARGET
);
4009 if (!u
|| !UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u
)))
4012 /* For user managers, send out READY=1 as soon as we reach basic.target */
4013 manager_send_ready(m
);
4015 /* Log the taint string as soon as we reach basic.target */
4016 log_taint_string(m
);
4019 void manager_check_finished(Manager
*m
) {
4022 if (MANAGER_IS_RELOADING(m
))
4025 /* Verify that we have entered the event loop already, and not left it again. */
4026 if (!MANAGER_IS_RUNNING(m
))
4029 manager_check_basic_target(m
);
4031 if (hashmap_size(m
->jobs
) > 0) {
4032 if (m
->jobs_in_progress_event_source
)
4033 /* Ignore any failure, this is only for feedback */
4034 (void) sd_event_source_set_time(m
->jobs_in_progress_event_source
,
4035 manager_watch_jobs_next_time(m
));
4039 /* The jobs hashmap tends to grow a lot during boot, and then it's not reused until shutdown. Let's
4040 kill the hashmap if it is relatively large. */
4041 if (hashmap_buckets(m
->jobs
) > hashmap_size(m
->units
) / 10)
4042 m
->jobs
= hashmap_free(m
->jobs
);
4044 manager_flip_auto_status(m
, false, "boot finished");
4046 /* Notify Type=idle units that we are done now */
4047 manager_close_idle_pipe(m
);
4049 /* Turn off confirm spawn now */
4050 m
->confirm_spawn
= NULL
;
4052 /* No need to update ask password status when we're going non-interactive */
4053 manager_close_ask_password(m
);
4055 /* This is no longer the first boot */
4056 manager_set_first_boot(m
, false);
4058 if (MANAGER_IS_FINISHED(m
))
4061 dual_timestamp_get(m
->timestamps
+ MANAGER_TIMESTAMP_FINISH
);
4063 manager_notify_finished(m
);
4065 manager_invalidate_startup_units(m
);
4068 static bool generator_path_any(const char* const* paths
) {
4072 /* Optimize by skipping the whole process by not creating output directories
4073 * if no generators are found. */
4074 STRV_FOREACH(path
, (char**) paths
)
4075 if (access(*path
, F_OK
) == 0)
4077 else if (errno
!= ENOENT
)
4078 log_warning_errno(errno
, "Failed to open generator directory %s: %m", *path
);
4083 static int manager_run_environment_generators(Manager
*m
) {
4084 char **tmp
= NULL
; /* this is only used in the forked process, no cleanup here */
4085 _cleanup_strv_free_
char **paths
= NULL
;
4087 [STDOUT_GENERATE
] = &tmp
,
4088 [STDOUT_COLLECT
] = &tmp
,
4089 [STDOUT_CONSUME
] = &m
->transient_environment
,
4093 if (MANAGER_IS_TEST_RUN(m
) && !(m
->test_run_flags
& MANAGER_TEST_RUN_ENV_GENERATORS
))
4096 paths
= env_generator_binary_paths(MANAGER_IS_SYSTEM(m
));
4100 if (!generator_path_any((const char* const*) paths
))
4103 RUN_WITH_UMASK(0022)
4104 r
= execute_directories((const char* const*) paths
, DEFAULT_TIMEOUT_USEC
, gather_environment
,
4105 args
, NULL
, m
->transient_environment
,
4106 EXEC_DIR_PARALLEL
| EXEC_DIR_IGNORE_ERRORS
| EXEC_DIR_SET_SYSTEMD_EXEC_PID
);
4110 static int manager_run_generators(Manager
*m
) {
4111 _cleanup_strv_free_
char **paths
= NULL
;
4112 const char *argv
[5];
4117 if (MANAGER_IS_TEST_RUN(m
) && !(m
->test_run_flags
& MANAGER_TEST_RUN_GENERATORS
))
4120 paths
= generator_binary_paths(m
->unit_file_scope
);
4124 if (!generator_path_any((const char* const*) paths
))
4127 r
= lookup_paths_mkdir_generator(&m
->lookup_paths
);
4129 log_error_errno(r
, "Failed to create generator directories: %m");
4133 argv
[0] = NULL
; /* Leave this empty, execute_directory() will fill something in */
4134 argv
[1] = m
->lookup_paths
.generator
;
4135 argv
[2] = m
->lookup_paths
.generator_early
;
4136 argv
[3] = m
->lookup_paths
.generator_late
;
4139 RUN_WITH_UMASK(0022)
4140 (void) execute_directories((const char* const*) paths
, DEFAULT_TIMEOUT_USEC
, NULL
, NULL
,
4141 (char**) argv
, m
->transient_environment
,
4142 EXEC_DIR_PARALLEL
| EXEC_DIR_IGNORE_ERRORS
| EXEC_DIR_SET_SYSTEMD_EXEC_PID
);
4147 lookup_paths_trim_generator(&m
->lookup_paths
);
4151 int manager_transient_environment_add(Manager
*m
, char **plus
) {
4156 if (strv_isempty(plus
))
4159 a
= strv_env_merge(2, m
->transient_environment
, plus
);
4163 sanitize_environment(a
);
4165 return strv_free_and_replace(m
->transient_environment
, a
);
4168 int manager_client_environment_modify(
4173 char **a
= NULL
, **b
= NULL
, **l
;
4177 if (strv_isempty(minus
) && strv_isempty(plus
))
4180 l
= m
->client_environment
;
4182 if (!strv_isempty(minus
)) {
4183 a
= strv_env_delete(l
, 1, minus
);
4190 if (!strv_isempty(plus
)) {
4191 b
= strv_env_merge(2, l
, plus
);
4200 if (m
->client_environment
!= l
)
4201 strv_free(m
->client_environment
);
4208 m
->client_environment
= sanitize_environment(l
);
4212 int manager_get_effective_environment(Manager
*m
, char ***ret
) {
4218 l
= strv_env_merge(2, m
->transient_environment
, m
->client_environment
);
4226 int manager_set_default_rlimits(Manager
*m
, struct rlimit
**default_rlimit
) {
4229 for (unsigned i
= 0; i
< _RLIMIT_MAX
; i
++) {
4230 m
->rlimit
[i
] = mfree(m
->rlimit
[i
]);
4232 if (!default_rlimit
[i
])
4235 m
->rlimit
[i
] = newdup(struct rlimit
, default_rlimit
[i
], 1);
4243 void manager_recheck_dbus(Manager
*m
) {
4246 /* Connects to the bus if the dbus service and socket are running. If we are running in user mode this is all
4247 * it does. In system mode we'll also connect to the system bus (which will most likely just reuse the
4248 * connection of the API bus). That's because the system bus after all runs as service of the system instance,
4249 * while in the user instance we can assume it's already there. */
4251 if (MANAGER_IS_RELOADING(m
))
4252 return; /* don't check while we are reloading… */
4254 if (manager_dbus_is_running(m
, false)) {
4255 (void) bus_init_api(m
);
4257 if (MANAGER_IS_SYSTEM(m
))
4258 (void) bus_init_system(m
);
4260 (void) bus_done_api(m
);
4262 if (MANAGER_IS_SYSTEM(m
))
4263 (void) bus_done_system(m
);
4267 static bool manager_journal_is_running(Manager
*m
) {
4272 if (MANAGER_IS_TEST_RUN(m
))
4275 /* If we are the user manager we can safely assume that the journal is up */
4276 if (!MANAGER_IS_SYSTEM(m
))
4279 /* Check that the socket is not only up, but in RUNNING state */
4280 u
= manager_get_unit(m
, SPECIAL_JOURNALD_SOCKET
);
4283 if (SOCKET(u
)->state
!= SOCKET_RUNNING
)
4286 /* Similar, check if the daemon itself is fully up, too */
4287 u
= manager_get_unit(m
, SPECIAL_JOURNALD_SERVICE
);
4290 if (!IN_SET(SERVICE(u
)->state
, SERVICE_RELOAD
, SERVICE_RUNNING
))
4296 void disable_printk_ratelimit(void) {
4297 /* Disable kernel's printk ratelimit.
4299 * Logging to /dev/kmsg is most useful during early boot and shutdown, where normal logging
4300 * mechanisms are not available. The semantics of this sysctl are such that any kernel command-line
4301 * setting takes precedence. */
4304 r
= sysctl_write("kernel/printk_devkmsg", "on");
4306 log_debug_errno(r
, "Failed to set sysctl kernel.printk_devkmsg=on: %m");
4309 void manager_recheck_journal(Manager
*m
) {
4313 /* Don't bother with this unless we are in the special situation of being PID 1 */
4314 if (getpid_cached() != 1)
4317 /* Don't check this while we are reloading, things might still change */
4318 if (MANAGER_IS_RELOADING(m
))
4321 /* The journal is fully and entirely up? If so, let's permit logging to it, if that's configured. If the
4322 * journal is down, don't ever log to it, otherwise we might end up deadlocking ourselves as we might trigger
4323 * an activation ourselves we can't fulfill. */
4324 log_set_prohibit_ipc(!manager_journal_is_running(m
));
4328 static ShowStatus
manager_get_show_status(Manager
*m
) {
4331 if (MANAGER_IS_USER(m
))
4332 return _SHOW_STATUS_INVALID
;
4334 if (m
->show_status_overridden
!= _SHOW_STATUS_INVALID
)
4335 return m
->show_status_overridden
;
4337 return m
->show_status
;
4340 bool manager_get_show_status_on(Manager
*m
) {
4343 return show_status_on(manager_get_show_status(m
));
4346 static void set_show_status_marker(bool b
) {
4348 (void) touch("/run/systemd/show-status");
4350 (void) unlink("/run/systemd/show-status");
4353 void manager_set_show_status(Manager
*m
, ShowStatus mode
, const char *reason
) {
4356 assert(mode
>= 0 && mode
< _SHOW_STATUS_MAX
);
4358 if (MANAGER_IS_USER(m
))
4361 if (mode
== m
->show_status
)
4364 if (m
->show_status_overridden
== _SHOW_STATUS_INVALID
) {
4367 enabled
= show_status_on(mode
);
4368 log_debug("%s (%s) showing of status (%s).",
4369 enabled
? "Enabling" : "Disabling",
4370 strna(show_status_to_string(mode
)),
4373 set_show_status_marker(enabled
);
4376 m
->show_status
= mode
;
4379 void manager_override_show_status(Manager
*m
, ShowStatus mode
, const char *reason
) {
4381 assert(mode
< _SHOW_STATUS_MAX
);
4383 if (MANAGER_IS_USER(m
))
4386 if (mode
== m
->show_status_overridden
)
4389 m
->show_status_overridden
= mode
;
4391 if (mode
== _SHOW_STATUS_INVALID
)
4392 mode
= m
->show_status
;
4394 log_debug("%s (%s) showing of status (%s).",
4395 m
->show_status_overridden
!= _SHOW_STATUS_INVALID
? "Overriding" : "Restoring",
4396 strna(show_status_to_string(mode
)),
4399 set_show_status_marker(show_status_on(mode
));
4402 const char *manager_get_confirm_spawn(Manager
*m
) {
4403 static int last_errno
= 0;
4409 /* Here's the deal: we want to test the validity of the console but don't want
4410 * PID1 to go through the whole console process which might block. But we also
4411 * want to warn the user only once if something is wrong with the console so we
4412 * cannot do the sanity checks after spawning our children. So here we simply do
4413 * really basic tests to hopefully trap common errors.
4415 * If the console suddenly disappear at the time our children will really it
4416 * then they will simply fail to acquire it and a positive answer will be
4417 * assumed. New children will fall back to /dev/console though.
4419 * Note: TTYs are devices that can come and go any time, and frequently aren't
4420 * available yet during early boot (consider a USB rs232 dongle...). If for any
4421 * reason the configured console is not ready, we fall back to the default
4424 if (!m
->confirm_spawn
|| path_equal(m
->confirm_spawn
, "/dev/console"))
4425 return m
->confirm_spawn
;
4427 if (stat(m
->confirm_spawn
, &st
) < 0) {
4432 if (!S_ISCHR(st
.st_mode
)) {
4438 return m
->confirm_spawn
;
4441 if (last_errno
!= r
)
4442 last_errno
= log_warning_errno(r
, "Failed to open %s, using default console: %m", m
->confirm_spawn
);
4444 return "/dev/console";
4447 void manager_set_first_boot(Manager
*m
, bool b
) {
4450 if (!MANAGER_IS_SYSTEM(m
))
4453 if (m
->first_boot
!= (int) b
) {
4455 (void) touch("/run/systemd/first-boot");
4457 (void) unlink("/run/systemd/first-boot");
4463 void manager_disable_confirm_spawn(void) {
4464 (void) touch("/run/systemd/confirm_spawn_disabled");
4467 bool manager_is_confirm_spawn_disabled(Manager
*m
) {
4468 if (!m
->confirm_spawn
)
4471 return access("/run/systemd/confirm_spawn_disabled", F_OK
) >= 0;
4474 static bool manager_should_show_status(Manager
*m
, StatusType type
) {
4477 if (!MANAGER_IS_SYSTEM(m
))
4480 if (m
->no_console_output
)
4483 if (!IN_SET(manager_state(m
), MANAGER_INITIALIZING
, MANAGER_STARTING
, MANAGER_STOPPING
))
4486 /* If we cannot find out the status properly, just proceed. */
4487 if (type
!= STATUS_TYPE_EMERGENCY
&& manager_check_ask_password(m
) > 0)
4490 if (type
== STATUS_TYPE_NOTICE
&& m
->show_status
!= SHOW_STATUS_NO
)
4493 return manager_get_show_status_on(m
);
4496 void manager_status_printf(Manager
*m
, StatusType type
, const char *status
, const char *format
, ...) {
4499 /* If m is NULL, assume we're after shutdown and let the messages through. */
4501 if (m
&& !manager_should_show_status(m
, type
))
4504 /* XXX We should totally drop the check for ephemeral here
4505 * and thus effectively make 'Type=idle' pointless. */
4506 if (type
== STATUS_TYPE_EPHEMERAL
&& m
&& m
->n_on_console
> 0)
4509 va_start(ap
, format
);
4510 status_vprintf(status
, SHOW_STATUS_ELLIPSIZE
|(type
== STATUS_TYPE_EPHEMERAL
? SHOW_STATUS_EPHEMERAL
: 0), format
, ap
);
4514 Set
* manager_get_units_requiring_mounts_for(Manager
*m
, const char *path
) {
4518 if (path_equal(path
, "/"))
4521 return hashmap_get(m
->units_requiring_mounts_for
, path
);
4524 int manager_update_failed_units(Manager
*m
, Unit
*u
, bool failed
) {
4529 assert(u
->manager
== m
);
4531 size
= set_size(m
->failed_units
);
4534 r
= set_ensure_put(&m
->failed_units
, NULL
, u
);
4538 (void) set_remove(m
->failed_units
, u
);
4540 if (set_size(m
->failed_units
) != size
)
4541 bus_manager_send_change_signal(m
);
4546 ManagerState
manager_state(Manager
*m
) {
4551 /* Is the special shutdown target active or queued? If so, we are in shutdown state */
4552 u
= manager_get_unit(m
, SPECIAL_SHUTDOWN_TARGET
);
4553 if (u
&& unit_active_or_pending(u
))
4554 return MANAGER_STOPPING
;
4556 /* Did we ever finish booting? If not then we are still starting up */
4557 if (!MANAGER_IS_FINISHED(m
)) {
4559 u
= manager_get_unit(m
, SPECIAL_BASIC_TARGET
);
4560 if (!u
|| !UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u
)))
4561 return MANAGER_INITIALIZING
;
4563 return MANAGER_STARTING
;
4566 if (MANAGER_IS_SYSTEM(m
)) {
4567 /* Are the rescue or emergency targets active or queued? If so we are in maintenance state */
4568 u
= manager_get_unit(m
, SPECIAL_RESCUE_TARGET
);
4569 if (u
&& unit_active_or_pending(u
))
4570 return MANAGER_MAINTENANCE
;
4572 u
= manager_get_unit(m
, SPECIAL_EMERGENCY_TARGET
);
4573 if (u
&& unit_active_or_pending(u
))
4574 return MANAGER_MAINTENANCE
;
4577 /* Are there any failed units? If so, we are in degraded mode */
4578 if (set_size(m
->failed_units
) > 0)
4579 return MANAGER_DEGRADED
;
4581 return MANAGER_RUNNING
;
4584 static void manager_unref_uid_internal(
4588 int (*_clean_ipc
)(uid_t uid
)) {
4592 assert(uid_is_valid(uid
));
4595 /* A generic implementation, covering both manager_unref_uid() and manager_unref_gid(), under the assumption
4596 * that uid_t and gid_t are actually defined the same way, with the same validity rules.
4598 * We store a hashmap where the UID/GID is they key and the value is a 32bit reference counter, whose highest
4599 * bit is used as flag for marking UIDs/GIDs whose IPC objects to remove when the last reference to the UID/GID
4600 * is dropped. The flag is set to on, once at least one reference from a unit where RemoveIPC= is set is added
4601 * on a UID/GID. It is reset when the UID's/GID's reference counter drops to 0 again. */
4603 assert_cc(sizeof(uid_t
) == sizeof(gid_t
));
4604 assert_cc(UID_INVALID
== (uid_t
) GID_INVALID
);
4606 if (uid
== 0) /* We don't keep track of root, and will never destroy it */
4609 c
= PTR_TO_UINT32(hashmap_get(uid_refs
, UID_TO_PTR(uid
)));
4611 n
= c
& ~DESTROY_IPC_FLAG
;
4615 if (destroy_now
&& n
== 0) {
4616 hashmap_remove(uid_refs
, UID_TO_PTR(uid
));
4618 if (c
& DESTROY_IPC_FLAG
) {
4619 log_debug("%s " UID_FMT
" is no longer referenced, cleaning up its IPC.",
4620 _clean_ipc
== clean_ipc_by_uid
? "UID" : "GID",
4622 (void) _clean_ipc(uid
);
4625 c
= n
| (c
& DESTROY_IPC_FLAG
);
4626 assert_se(hashmap_update(uid_refs
, UID_TO_PTR(uid
), UINT32_TO_PTR(c
)) >= 0);
4630 void manager_unref_uid(Manager
*m
, uid_t uid
, bool destroy_now
) {
4631 manager_unref_uid_internal(m
->uid_refs
, uid
, destroy_now
, clean_ipc_by_uid
);
4634 void manager_unref_gid(Manager
*m
, gid_t gid
, bool destroy_now
) {
4635 manager_unref_uid_internal(m
->gid_refs
, (uid_t
) gid
, destroy_now
, clean_ipc_by_gid
);
4638 static int manager_ref_uid_internal(
4647 assert(uid_is_valid(uid
));
4649 /* A generic implementation, covering both manager_ref_uid() and manager_ref_gid(), under the assumption
4650 * that uid_t and gid_t are actually defined the same way, with the same validity rules. */
4652 assert_cc(sizeof(uid_t
) == sizeof(gid_t
));
4653 assert_cc(UID_INVALID
== (uid_t
) GID_INVALID
);
4655 if (uid
== 0) /* We don't keep track of root, and will never destroy it */
4658 r
= hashmap_ensure_allocated(uid_refs
, &trivial_hash_ops
);
4662 c
= PTR_TO_UINT32(hashmap_get(*uid_refs
, UID_TO_PTR(uid
)));
4664 n
= c
& ~DESTROY_IPC_FLAG
;
4667 if (n
& DESTROY_IPC_FLAG
) /* check for overflow */
4670 c
= n
| (c
& DESTROY_IPC_FLAG
) | (clean_ipc
? DESTROY_IPC_FLAG
: 0);
4672 return hashmap_replace(*uid_refs
, UID_TO_PTR(uid
), UINT32_TO_PTR(c
));
4675 int manager_ref_uid(Manager
*m
, uid_t uid
, bool clean_ipc
) {
4676 return manager_ref_uid_internal(&m
->uid_refs
, uid
, clean_ipc
);
4679 int manager_ref_gid(Manager
*m
, gid_t gid
, bool clean_ipc
) {
4680 return manager_ref_uid_internal(&m
->gid_refs
, (uid_t
) gid
, clean_ipc
);
4683 static void manager_vacuum_uid_refs_internal(
4685 int (*_clean_ipc
)(uid_t uid
)) {
4691 HASHMAP_FOREACH_KEY(p
, k
, uid_refs
) {
4695 uid
= PTR_TO_UID(k
);
4696 c
= PTR_TO_UINT32(p
);
4698 n
= c
& ~DESTROY_IPC_FLAG
;
4702 if (c
& DESTROY_IPC_FLAG
) {
4703 log_debug("Found unreferenced %s " UID_FMT
" after reload/reexec. Cleaning up.",
4704 _clean_ipc
== clean_ipc_by_uid
? "UID" : "GID",
4706 (void) _clean_ipc(uid
);
4709 assert_se(hashmap_remove(uid_refs
, k
) == p
);
4713 static void manager_vacuum_uid_refs(Manager
*m
) {
4714 manager_vacuum_uid_refs_internal(m
->uid_refs
, clean_ipc_by_uid
);
4717 static void manager_vacuum_gid_refs(Manager
*m
) {
4718 manager_vacuum_uid_refs_internal(m
->gid_refs
, clean_ipc_by_gid
);
4721 static void manager_vacuum(Manager
*m
) {
4724 /* Release any dynamic users no longer referenced */
4725 dynamic_user_vacuum(m
, true);
4727 /* Release any references to UIDs/GIDs no longer referenced, and destroy any IPC owned by them */
4728 manager_vacuum_uid_refs(m
);
4729 manager_vacuum_gid_refs(m
);
4731 /* Release any runtimes no longer referenced */
4732 exec_runtime_vacuum(m
);
4735 int manager_dispatch_user_lookup_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
4739 char unit_name
[UNIT_NAME_MAX
+1];
4742 Manager
*m
= userdata
;
4750 /* Invoked whenever a child process succeeded resolving its user/group to use and sent us the resulting UID/GID
4751 * in a datagram. We parse the datagram here and pass it off to the unit, so that it can add a reference to the
4752 * UID/GID so that it can destroy the UID/GID's IPC objects when the reference counter drops to 0. */
4754 l
= recv(fd
, &buffer
, sizeof(buffer
), MSG_DONTWAIT
);
4756 if (IN_SET(errno
, EINTR
, EAGAIN
))
4759 return log_error_errno(errno
, "Failed to read from user lookup fd: %m");
4762 if ((size_t) l
<= offsetof(struct buffer
, unit_name
)) {
4763 log_warning("Received too short user lookup message, ignoring.");
4767 if ((size_t) l
> offsetof(struct buffer
, unit_name
) + UNIT_NAME_MAX
) {
4768 log_warning("Received too long user lookup message, ignoring.");
4772 if (!uid_is_valid(buffer
.uid
) && !gid_is_valid(buffer
.gid
)) {
4773 log_warning("Got user lookup message with invalid UID/GID pair, ignoring.");
4777 n
= (size_t) l
- offsetof(struct buffer
, unit_name
);
4778 if (memchr(buffer
.unit_name
, 0, n
)) {
4779 log_warning("Received lookup message with embedded NUL character, ignoring.");
4783 buffer
.unit_name
[n
] = 0;
4784 u
= manager_get_unit(m
, buffer
.unit_name
);
4786 log_debug("Got user lookup message but unit doesn't exist, ignoring.");
4790 log_unit_debug(u
, "User lookup succeeded: uid=" UID_FMT
" gid=" GID_FMT
, buffer
.uid
, buffer
.gid
);
4792 unit_notify_user_lookup(u
, buffer
.uid
, buffer
.gid
);
4796 char *manager_taint_string(Manager
*m
) {
4797 _cleanup_free_
char *destination
= NULL
, *overflowuid
= NULL
, *overflowgid
= NULL
;
4801 /* Returns a "taint string", e.g. "local-hwclock:var-run-bad".
4802 * Only things that are detected at runtime should be tagged
4803 * here. For stuff that is set during compilation, emit a warning
4804 * in the configuration phase. */
4808 buf
= new(char, sizeof("split-usr:"
4813 "overflowuid-not-65534:"
4814 "overflowgid-not-65534:"));
4822 e
= stpcpy(e
, "split-usr:");
4824 if (access("/proc/cgroups", F_OK
) < 0)
4825 e
= stpcpy(e
, "cgroups-missing:");
4827 if (cg_all_unified() == 0)
4828 e
= stpcpy(e
, "cgroupsv1:");
4830 if (clock_is_localtime(NULL
) > 0)
4831 e
= stpcpy(e
, "local-hwclock:");
4833 r
= readlink_malloc("/var/run", &destination
);
4834 if (r
< 0 || !PATH_IN_SET(destination
, "../run", "/run"))
4835 e
= stpcpy(e
, "var-run-bad:");
4837 r
= read_one_line_file("/proc/sys/kernel/overflowuid", &overflowuid
);
4838 if (r
>= 0 && !streq(overflowuid
, "65534"))
4839 e
= stpcpy(e
, "overflowuid-not-65534:");
4841 r
= read_one_line_file("/proc/sys/kernel/overflowgid", &overflowgid
);
4842 if (r
>= 0 && !streq(overflowgid
, "65534"))
4843 e
= stpcpy(e
, "overflowgid-not-65534:");
4845 /* remove the last ':' */
4852 void manager_ref_console(Manager
*m
) {
4858 void manager_unref_console(Manager
*m
) {
4860 assert(m
->n_on_console
> 0);
4863 if (m
->n_on_console
== 0)
4864 m
->no_console_output
= false; /* unset no_console_output flag, since the console is definitely free now */
4867 void manager_override_log_level(Manager
*m
, int level
) {
4868 _cleanup_free_
char *s
= NULL
;
4871 if (!m
->log_level_overridden
) {
4872 m
->original_log_level
= log_get_max_level();
4873 m
->log_level_overridden
= true;
4876 (void) log_level_to_string_alloc(level
, &s
);
4877 log_info("Setting log level to %s.", strna(s
));
4879 log_set_max_level(level
);
4882 void manager_restore_original_log_level(Manager
*m
) {
4883 _cleanup_free_
char *s
= NULL
;
4886 if (!m
->log_level_overridden
)
4889 (void) log_level_to_string_alloc(m
->original_log_level
, &s
);
4890 log_info("Restoring log level to original (%s).", strna(s
));
4892 log_set_max_level(m
->original_log_level
);
4893 m
->log_level_overridden
= false;
4896 void manager_override_log_target(Manager
*m
, LogTarget target
) {
4899 if (!m
->log_target_overridden
) {
4900 m
->original_log_target
= log_get_target();
4901 m
->log_target_overridden
= true;
4904 log_info("Setting log target to %s.", log_target_to_string(target
));
4905 log_set_target(target
);
4908 void manager_restore_original_log_target(Manager
*m
) {
4911 if (!m
->log_target_overridden
)
4914 log_info("Restoring log target to original %s.", log_target_to_string(m
->original_log_target
));
4916 log_set_target(m
->original_log_target
);
4917 m
->log_target_overridden
= false;
4920 ManagerTimestamp
manager_timestamp_initrd_mangle(ManagerTimestamp s
) {
4922 s
>= MANAGER_TIMESTAMP_SECURITY_START
&&
4923 s
<= MANAGER_TIMESTAMP_UNITS_LOAD_FINISH
)
4924 return s
- MANAGER_TIMESTAMP_SECURITY_START
+ MANAGER_TIMESTAMP_INITRD_SECURITY_START
;
4928 static const char *const manager_state_table
[_MANAGER_STATE_MAX
] = {
4929 [MANAGER_INITIALIZING
] = "initializing",
4930 [MANAGER_STARTING
] = "starting",
4931 [MANAGER_RUNNING
] = "running",
4932 [MANAGER_DEGRADED
] = "degraded",
4933 [MANAGER_MAINTENANCE
] = "maintenance",
4934 [MANAGER_STOPPING
] = "stopping",
4937 DEFINE_STRING_TABLE_LOOKUP(manager_state
, ManagerState
);
4939 static const char *const manager_timestamp_table
[_MANAGER_TIMESTAMP_MAX
] = {
4940 [MANAGER_TIMESTAMP_FIRMWARE
] = "firmware",
4941 [MANAGER_TIMESTAMP_LOADER
] = "loader",
4942 [MANAGER_TIMESTAMP_KERNEL
] = "kernel",
4943 [MANAGER_TIMESTAMP_INITRD
] = "initrd",
4944 [MANAGER_TIMESTAMP_USERSPACE
] = "userspace",
4945 [MANAGER_TIMESTAMP_FINISH
] = "finish",
4946 [MANAGER_TIMESTAMP_SECURITY_START
] = "security-start",
4947 [MANAGER_TIMESTAMP_SECURITY_FINISH
] = "security-finish",
4948 [MANAGER_TIMESTAMP_GENERATORS_START
] = "generators-start",
4949 [MANAGER_TIMESTAMP_GENERATORS_FINISH
] = "generators-finish",
4950 [MANAGER_TIMESTAMP_UNITS_LOAD_START
] = "units-load-start",
4951 [MANAGER_TIMESTAMP_UNITS_LOAD_FINISH
] = "units-load-finish",
4952 [MANAGER_TIMESTAMP_INITRD_SECURITY_START
] = "initrd-security-start",
4953 [MANAGER_TIMESTAMP_INITRD_SECURITY_FINISH
] = "initrd-security-finish",
4954 [MANAGER_TIMESTAMP_INITRD_GENERATORS_START
] = "initrd-generators-start",
4955 [MANAGER_TIMESTAMP_INITRD_GENERATORS_FINISH
] = "initrd-generators-finish",
4956 [MANAGER_TIMESTAMP_INITRD_UNITS_LOAD_START
] = "initrd-units-load-start",
4957 [MANAGER_TIMESTAMP_INITRD_UNITS_LOAD_FINISH
] = "initrd-units-load-finish",
4960 DEFINE_STRING_TABLE_LOOKUP(manager_timestamp
, ManagerTimestamp
);
4962 static const char* const oom_policy_table
[_OOM_POLICY_MAX
] = {
4963 [OOM_CONTINUE
] = "continue",
4964 [OOM_STOP
] = "stop",
4965 [OOM_KILL
] = "kill",
4968 DEFINE_STRING_TABLE_LOOKUP(oom_policy
, OOMPolicy
);