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];
211 assert(m
->n_running_jobs
> 0);
213 manager_flip_auto_status(m
, true, "delay");
215 print_nr
= (m
->jobs_in_progress_iteration
/ JOBS_IN_PROGRESS_PERIOD_DIVISOR
) % m
->n_running_jobs
;
217 HASHMAP_FOREACH(j
, m
->jobs
)
218 if (j
->state
== JOB_RUNNING
&& counter
++ == print_nr
)
221 /* m->n_running_jobs must be consistent with the contents of m->jobs,
222 * so the above loop must have succeeded in finding j. */
223 assert(counter
== print_nr
+ 1);
226 cylon_pos
= m
->jobs_in_progress_iteration
% 14;
228 cylon_pos
= 14 - cylon_pos
;
229 draw_cylon(cylon
, sizeof(cylon
), 6, cylon_pos
);
231 m
->jobs_in_progress_iteration
++;
233 if (m
->n_running_jobs
> 1)
234 if (asprintf(&job_of_n
, "(%u of %u) ", counter
, m
->n_running_jobs
) < 0)
237 bool have_timeout
= job_get_timeout(j
, &x
) > 0;
239 /* We want to use enough information for the user to identify previous lines talking about the same
240 * unit, but keep the message as short as possible. So if 'Starting foo.service' or 'Starting
241 * foo.service (Description)' were used, 'foo.service' is enough here. On the other hand, if we used
242 * 'Starting Description' before, then we shall also use 'Description' here. So we pass NULL as the
243 * second argument to unit_status_string(). */
244 const char *ident
= unit_status_string(j
->unit
, NULL
);
246 manager_status_printf(m
, STATUS_TYPE_EPHEMERAL
, cylon
,
247 "%sA %s job is running for %s (%s / %s)",
249 job_type_to_string(j
->type
),
251 FORMAT_TIMESPAN(now(CLOCK_MONOTONIC
) - j
->begin_usec
, 1*USEC_PER_SEC
),
252 have_timeout
? FORMAT_TIMESPAN(x
- j
->begin_usec
, 1*USEC_PER_SEC
) : "no limit");
255 static int have_ask_password(void) {
256 _cleanup_closedir_
DIR *dir
= NULL
;
259 dir
= opendir("/run/systemd/ask-password");
267 FOREACH_DIRENT_ALL(de
, dir
, return -errno
) {
268 if (startswith(de
->d_name
, "ask."))
274 static int manager_dispatch_ask_password_fd(sd_event_source
*source
,
275 int fd
, uint32_t revents
, void *userdata
) {
276 Manager
*m
= userdata
;
282 m
->have_ask_password
= have_ask_password();
283 if (m
->have_ask_password
< 0)
284 /* Log error but continue. Negative have_ask_password
285 * is treated as unknown status. */
286 log_error_errno(m
->have_ask_password
, "Failed to list /run/systemd/ask-password: %m");
291 static void manager_close_ask_password(Manager
*m
) {
294 m
->ask_password_event_source
= sd_event_source_disable_unref(m
->ask_password_event_source
);
295 m
->ask_password_inotify_fd
= safe_close(m
->ask_password_inotify_fd
);
296 m
->have_ask_password
= -EINVAL
;
299 static int manager_check_ask_password(Manager
*m
) {
304 if (!m
->ask_password_event_source
) {
305 assert(m
->ask_password_inotify_fd
< 0);
307 (void) mkdir_p_label("/run/systemd/ask-password", 0755);
309 m
->ask_password_inotify_fd
= inotify_init1(IN_NONBLOCK
|IN_CLOEXEC
);
310 if (m
->ask_password_inotify_fd
< 0)
311 return log_error_errno(errno
, "Failed to create inotify object: %m");
313 r
= inotify_add_watch_and_warn(m
->ask_password_inotify_fd
,
314 "/run/systemd/ask-password",
315 IN_CREATE
|IN_DELETE
|IN_MOVE
);
317 manager_close_ask_password(m
);
321 r
= sd_event_add_io(m
->event
, &m
->ask_password_event_source
,
322 m
->ask_password_inotify_fd
, EPOLLIN
,
323 manager_dispatch_ask_password_fd
, m
);
325 log_error_errno(r
, "Failed to add event source for /run/systemd/ask-password: %m");
326 manager_close_ask_password(m
);
330 (void) sd_event_source_set_description(m
->ask_password_event_source
, "manager-ask-password");
332 /* Queries might have been added meanwhile... */
333 manager_dispatch_ask_password_fd(m
->ask_password_event_source
,
334 m
->ask_password_inotify_fd
, EPOLLIN
, m
);
337 return m
->have_ask_password
;
340 static int manager_watch_idle_pipe(Manager
*m
) {
345 if (m
->idle_pipe_event_source
)
348 if (m
->idle_pipe
[2] < 0)
351 r
= sd_event_add_io(m
->event
, &m
->idle_pipe_event_source
, m
->idle_pipe
[2], EPOLLIN
, manager_dispatch_idle_pipe_fd
, m
);
353 return log_error_errno(r
, "Failed to watch idle pipe: %m");
355 (void) sd_event_source_set_description(m
->idle_pipe_event_source
, "manager-idle-pipe");
360 static void manager_close_idle_pipe(Manager
*m
) {
363 m
->idle_pipe_event_source
= sd_event_source_disable_unref(m
->idle_pipe_event_source
);
365 safe_close_pair(m
->idle_pipe
);
366 safe_close_pair(m
->idle_pipe
+ 2);
369 static int manager_setup_time_change(Manager
*m
) {
374 if (MANAGER_IS_TEST_RUN(m
))
377 m
->time_change_event_source
= sd_event_source_disable_unref(m
->time_change_event_source
);
378 m
->time_change_fd
= safe_close(m
->time_change_fd
);
380 m
->time_change_fd
= time_change_fd();
381 if (m
->time_change_fd
< 0)
382 return log_error_errno(m
->time_change_fd
, "Failed to create timer change timer fd: %m");
384 r
= sd_event_add_io(m
->event
, &m
->time_change_event_source
, m
->time_change_fd
, EPOLLIN
, manager_dispatch_time_change_fd
, m
);
386 return log_error_errno(r
, "Failed to create time change event source: %m");
388 /* Schedule this slightly earlier than the .timer event sources */
389 r
= sd_event_source_set_priority(m
->time_change_event_source
, SD_EVENT_PRIORITY_NORMAL
-1);
391 return log_error_errno(r
, "Failed to set priority of time change event sources: %m");
393 (void) sd_event_source_set_description(m
->time_change_event_source
, "manager-time-change");
395 log_debug("Set up TFD_TIMER_CANCEL_ON_SET timerfd.");
400 static int manager_read_timezone_stat(Manager
*m
) {
406 /* Read the current stat() data of /etc/localtime so that we detect changes */
407 if (lstat("/etc/localtime", &st
) < 0) {
408 log_debug_errno(errno
, "Failed to stat /etc/localtime, ignoring: %m");
409 changed
= m
->etc_localtime_accessible
;
410 m
->etc_localtime_accessible
= false;
414 k
= timespec_load(&st
.st_mtim
);
415 changed
= !m
->etc_localtime_accessible
|| k
!= m
->etc_localtime_mtime
;
417 m
->etc_localtime_mtime
= k
;
418 m
->etc_localtime_accessible
= true;
424 static int manager_setup_timezone_change(Manager
*m
) {
425 _cleanup_(sd_event_source_unrefp
) sd_event_source
*new_event
= NULL
;
430 if (MANAGER_IS_TEST_RUN(m
))
433 /* We watch /etc/localtime for three events: change of the link count (which might mean removal from /etc even
434 * though another link might be kept), renames, and file close operations after writing. Note we don't bother
435 * with IN_DELETE_SELF, as that would just report when the inode is removed entirely, i.e. after the link count
436 * went to zero and all fds to it are closed.
438 * Note that we never follow symlinks here. This is a simplification, but should cover almost all cases
441 * Note that we create the new event source first here, before releasing the old one. This should optimize
442 * behaviour as this way sd-event can reuse the old watch in case the inode didn't change. */
444 r
= sd_event_add_inotify(m
->event
, &new_event
, "/etc/localtime",
445 IN_ATTRIB
|IN_MOVE_SELF
|IN_CLOSE_WRITE
|IN_DONT_FOLLOW
, manager_dispatch_timezone_change
, m
);
447 /* If the file doesn't exist yet, subscribe to /etc instead, and wait until it is created either by
448 * O_CREATE or by rename() */
450 log_debug_errno(r
, "/etc/localtime doesn't exist yet, watching /etc instead.");
451 r
= sd_event_add_inotify(m
->event
, &new_event
, "/etc",
452 IN_CREATE
|IN_MOVED_TO
|IN_ONLYDIR
, manager_dispatch_timezone_change
, m
);
455 return log_error_errno(r
, "Failed to create timezone change event source: %m");
457 /* Schedule this slightly earlier than the .timer event sources */
458 r
= sd_event_source_set_priority(new_event
, SD_EVENT_PRIORITY_NORMAL
-1);
460 return log_error_errno(r
, "Failed to set priority of timezone change event sources: %m");
462 sd_event_source_unref(m
->timezone_change_event_source
);
463 m
->timezone_change_event_source
= TAKE_PTR(new_event
);
468 static int enable_special_signals(Manager
*m
) {
469 _cleanup_close_
int fd
= -1;
473 if (MANAGER_IS_TEST_RUN(m
))
476 /* Enable that we get SIGINT on control-alt-del. In containers
477 * this will fail with EPERM (older) or EINVAL (newer), so
479 if (reboot(RB_DISABLE_CAD
) < 0 && !IN_SET(errno
, EPERM
, EINVAL
))
480 log_warning_errno(errno
, "Failed to enable ctrl-alt-del handling: %m");
482 fd
= open_terminal("/dev/tty0", O_RDWR
|O_NOCTTY
|O_CLOEXEC
);
484 /* Support systems without virtual console */
486 log_warning_errno(errno
, "Failed to open /dev/tty0: %m");
488 /* Enable that we get SIGWINCH on kbrequest */
489 if (ioctl(fd
, KDSIGACCEPT
, SIGWINCH
) < 0)
490 log_warning_errno(errno
, "Failed to enable kbrequest handling: %m");
496 #define RTSIG_IF_AVAILABLE(signum) (signum <= SIGRTMAX ? signum : -1)
498 static int manager_setup_signals(Manager
*m
) {
499 struct sigaction sa
= {
500 .sa_handler
= SIG_DFL
,
501 .sa_flags
= SA_NOCLDSTOP
|SA_RESTART
,
508 assert_se(sigaction(SIGCHLD
, &sa
, NULL
) == 0);
510 /* We make liberal use of realtime signals here. On
511 * Linux/glibc we have 30 of them (with the exception of Linux
512 * on hppa, see below), between SIGRTMIN+0 ... SIGRTMIN+30
515 assert_se(sigemptyset(&mask
) == 0);
516 sigset_add_many(&mask
,
517 SIGCHLD
, /* Child died */
518 SIGTERM
, /* Reexecute daemon */
519 SIGHUP
, /* Reload configuration */
520 SIGUSR1
, /* systemd: reconnect to D-Bus */
521 SIGUSR2
, /* systemd: dump status */
522 SIGINT
, /* Kernel sends us this on control-alt-del */
523 SIGWINCH
, /* Kernel sends us this on kbrequest (alt-arrowup) */
524 SIGPWR
, /* Some kernel drivers and upsd send us this on power failure */
526 SIGRTMIN
+0, /* systemd: start default.target */
527 SIGRTMIN
+1, /* systemd: isolate rescue.target */
528 SIGRTMIN
+2, /* systemd: isolate emergency.target */
529 SIGRTMIN
+3, /* systemd: start halt.target */
530 SIGRTMIN
+4, /* systemd: start poweroff.target */
531 SIGRTMIN
+5, /* systemd: start reboot.target */
532 SIGRTMIN
+6, /* systemd: start kexec.target */
534 /* ... space for more special targets ... */
536 SIGRTMIN
+13, /* systemd: Immediate halt */
537 SIGRTMIN
+14, /* systemd: Immediate poweroff */
538 SIGRTMIN
+15, /* systemd: Immediate reboot */
539 SIGRTMIN
+16, /* systemd: Immediate kexec */
541 /* ... space for more immediate system state changes ... */
543 SIGRTMIN
+20, /* systemd: enable status messages */
544 SIGRTMIN
+21, /* systemd: disable status messages */
545 SIGRTMIN
+22, /* systemd: set log level to LOG_DEBUG */
546 SIGRTMIN
+23, /* systemd: set log level to LOG_INFO */
547 SIGRTMIN
+24, /* systemd: Immediate exit (--user only) */
549 /* .. one free signal here ... */
551 /* Apparently Linux on hppa had fewer RT signals until v3.18,
552 * SIGRTMAX was SIGRTMIN+25, and then SIGRTMIN was lowered,
553 * see commit v3.17-7614-g1f25df2eff.
555 * We cannot unconditionally make use of those signals here,
556 * so let's use a runtime check. Since these commands are
557 * accessible by different means and only really a safety
558 * net, the missing functionality on hppa shouldn't matter.
561 RTSIG_IF_AVAILABLE(SIGRTMIN
+26), /* systemd: set log target to journal-or-kmsg */
562 RTSIG_IF_AVAILABLE(SIGRTMIN
+27), /* systemd: set log target to console */
563 RTSIG_IF_AVAILABLE(SIGRTMIN
+28), /* systemd: set log target to kmsg */
564 RTSIG_IF_AVAILABLE(SIGRTMIN
+29), /* systemd: set log target to syslog-or-kmsg (obsolete) */
566 /* ... one free signal here SIGRTMIN+30 ... */
568 assert_se(sigprocmask(SIG_SETMASK
, &mask
, NULL
) == 0);
570 m
->signal_fd
= signalfd(-1, &mask
, SFD_NONBLOCK
|SFD_CLOEXEC
);
571 if (m
->signal_fd
< 0)
574 r
= sd_event_add_io(m
->event
, &m
->signal_event_source
, m
->signal_fd
, EPOLLIN
, manager_dispatch_signal_fd
, m
);
578 (void) sd_event_source_set_description(m
->signal_event_source
, "manager-signal");
580 /* Process signals a bit earlier than the rest of things, but later than notify_fd processing, so that the
581 * notify processing can still figure out to which process/service a message belongs, before we reap the
582 * process. Also, process this before handling cgroup notifications, so that we always collect child exit
583 * status information before detecting that there's no process in a cgroup. */
584 r
= sd_event_source_set_priority(m
->signal_event_source
, SD_EVENT_PRIORITY_NORMAL
-6);
588 if (MANAGER_IS_SYSTEM(m
))
589 return enable_special_signals(m
);
594 static char** sanitize_environment(char **l
) {
596 /* Let's remove some environment variables that we need ourselves to communicate with our clients */
600 "CONFIGURATION_DIRECTORY",
601 "CREDENTIALS_DIRECTORY",
623 /* Let's order the environment alphabetically, just to make it pretty */
629 int manager_default_environment(Manager
*m
) {
634 m
->transient_environment
= strv_free(m
->transient_environment
);
636 if (MANAGER_IS_SYSTEM(m
)) {
637 /* The system manager always starts with a clean
638 * environment for its children. It does not import
639 * the kernel's or the parents' exported variables.
641 * The initial passed environment is untouched to keep
642 * /proc/self/environ valid; it is used for tagging
643 * the init process inside containers. */
644 m
->transient_environment
= strv_new("PATH=" DEFAULT_PATH
);
645 if (!m
->transient_environment
)
648 /* Import locale variables LC_*= from configuration */
649 (void) locale_setup(&m
->transient_environment
);
651 /* The user manager passes its own environment along to its children, except for $PATH. */
652 m
->transient_environment
= strv_copy(environ
);
653 if (!m
->transient_environment
)
656 r
= strv_env_replace_strdup(&m
->transient_environment
, "PATH=" DEFAULT_USER_PATH
);
661 sanitize_environment(m
->transient_environment
);
666 static int manager_setup_prefix(Manager
*m
) {
672 static const struct table_entry paths_system
[_EXEC_DIRECTORY_TYPE_MAX
] = {
673 [EXEC_DIRECTORY_RUNTIME
] = { SD_PATH_SYSTEM_RUNTIME
, NULL
},
674 [EXEC_DIRECTORY_STATE
] = { SD_PATH_SYSTEM_STATE_PRIVATE
, NULL
},
675 [EXEC_DIRECTORY_CACHE
] = { SD_PATH_SYSTEM_STATE_CACHE
, NULL
},
676 [EXEC_DIRECTORY_LOGS
] = { SD_PATH_SYSTEM_STATE_LOGS
, NULL
},
677 [EXEC_DIRECTORY_CONFIGURATION
] = { SD_PATH_SYSTEM_CONFIGURATION
, NULL
},
680 static const struct table_entry paths_user
[_EXEC_DIRECTORY_TYPE_MAX
] = {
681 [EXEC_DIRECTORY_RUNTIME
] = { SD_PATH_USER_RUNTIME
, NULL
},
682 [EXEC_DIRECTORY_STATE
] = { SD_PATH_USER_CONFIGURATION
, NULL
},
683 [EXEC_DIRECTORY_CACHE
] = { SD_PATH_USER_STATE_CACHE
, NULL
},
684 [EXEC_DIRECTORY_LOGS
] = { SD_PATH_USER_CONFIGURATION
, "log" },
685 [EXEC_DIRECTORY_CONFIGURATION
] = { SD_PATH_USER_CONFIGURATION
, NULL
},
690 const struct table_entry
*p
= MANAGER_IS_SYSTEM(m
) ? paths_system
: paths_user
;
693 for (ExecDirectoryType i
= 0; i
< _EXEC_DIRECTORY_TYPE_MAX
; i
++) {
694 r
= sd_path_lookup(p
[i
].type
, p
[i
].suffix
, &m
->prefix
[i
]);
696 return log_warning_errno(r
, "Failed to lookup %s path: %m",
697 exec_directory_type_to_string(i
));
703 static void manager_free_unit_name_maps(Manager
*m
) {
704 m
->unit_id_map
= hashmap_free(m
->unit_id_map
);
705 m
->unit_name_map
= hashmap_free(m
->unit_name_map
);
706 m
->unit_path_cache
= set_free(m
->unit_path_cache
);
707 m
->unit_cache_timestamp_hash
= 0;
710 static int manager_setup_run_queue(Manager
*m
) {
714 assert(!m
->run_queue_event_source
);
716 r
= sd_event_add_defer(m
->event
, &m
->run_queue_event_source
, manager_dispatch_run_queue
, m
);
720 r
= sd_event_source_set_priority(m
->run_queue_event_source
, SD_EVENT_PRIORITY_IDLE
);
724 r
= sd_event_source_set_enabled(m
->run_queue_event_source
, SD_EVENT_OFF
);
728 (void) sd_event_source_set_description(m
->run_queue_event_source
, "manager-run-queue");
733 static int manager_setup_sigchld_event_source(Manager
*m
) {
737 assert(!m
->sigchld_event_source
);
739 r
= sd_event_add_defer(m
->event
, &m
->sigchld_event_source
, manager_dispatch_sigchld
, m
);
743 r
= sd_event_source_set_priority(m
->sigchld_event_source
, SD_EVENT_PRIORITY_NORMAL
-7);
747 r
= sd_event_source_set_enabled(m
->sigchld_event_source
, SD_EVENT_OFF
);
751 (void) sd_event_source_set_description(m
->sigchld_event_source
, "manager-sigchld");
756 int manager_new(UnitFileScope scope
, ManagerTestRunFlags test_run_flags
, Manager
**_m
) {
757 _cleanup_(manager_freep
) Manager
*m
= NULL
;
762 assert(IN_SET(scope
, UNIT_FILE_SYSTEM
, UNIT_FILE_USER
));
769 .unit_file_scope
= scope
,
770 .objective
= _MANAGER_OBJECTIVE_INVALID
,
772 .status_unit_format
= STATUS_UNIT_FORMAT_DEFAULT
,
774 .default_timer_accuracy_usec
= USEC_PER_MINUTE
,
775 .default_memory_accounting
= MEMORY_ACCOUNTING_DEFAULT
,
776 .default_tasks_accounting
= true,
777 .default_tasks_max
= TASKS_MAX_UNSET
,
778 .default_timeout_start_usec
= DEFAULT_TIMEOUT_USEC
,
779 .default_timeout_stop_usec
= DEFAULT_TIMEOUT_USEC
,
780 .default_restart_usec
= DEFAULT_RESTART_USEC
,
782 .original_log_level
= -1,
783 .original_log_target
= _LOG_TARGET_INVALID
,
785 .watchdog_overridden
[WATCHDOG_RUNTIME
] = USEC_INFINITY
,
786 .watchdog_overridden
[WATCHDOG_REBOOT
] = USEC_INFINITY
,
787 .watchdog_overridden
[WATCHDOG_KEXEC
] = USEC_INFINITY
,
789 .show_status_overridden
= _SHOW_STATUS_INVALID
,
792 .cgroups_agent_fd
= -1,
794 .time_change_fd
= -1,
795 .user_lookup_fds
= { -1, -1 },
796 .private_listen_fd
= -1,
798 .cgroup_inotify_fd
= -1,
799 .pin_cgroupfs_fd
= -1,
800 .ask_password_inotify_fd
= -1,
801 .idle_pipe
= { -1, -1, -1, -1},
803 /* start as id #1, so that we can leave #0 around as "null-like" value */
806 .have_ask_password
= -EINVAL
, /* we don't know */
808 .test_run_flags
= test_run_flags
,
810 .default_oom_policy
= OOM_STOP
,
814 if (MANAGER_IS_SYSTEM(m
) && detect_container() <= 0)
815 boot_timestamps(m
->timestamps
+ MANAGER_TIMESTAMP_USERSPACE
,
816 m
->timestamps
+ MANAGER_TIMESTAMP_FIRMWARE
,
817 m
->timestamps
+ MANAGER_TIMESTAMP_LOADER
);
820 /* Prepare log fields we can use for structured logging */
821 if (MANAGER_IS_SYSTEM(m
)) {
822 m
->unit_log_field
= "UNIT=";
823 m
->unit_log_format_string
= "UNIT=%s";
825 m
->invocation_log_field
= "INVOCATION_ID=";
826 m
->invocation_log_format_string
= "INVOCATION_ID=%s";
828 m
->unit_log_field
= "USER_UNIT=";
829 m
->unit_log_format_string
= "USER_UNIT=%s";
831 m
->invocation_log_field
= "USER_INVOCATION_ID=";
832 m
->invocation_log_format_string
= "USER_INVOCATION_ID=%s";
835 /* Reboot immediately if the user hits C-A-D more often than 7x per 2s */
836 m
->ctrl_alt_del_ratelimit
= (RateLimit
) { .interval
= 2 * USEC_PER_SEC
, .burst
= 7 };
838 r
= manager_default_environment(m
);
842 r
= hashmap_ensure_allocated(&m
->units
, &string_hash_ops
);
846 r
= hashmap_ensure_allocated(&m
->cgroup_unit
, &path_hash_ops
);
850 r
= hashmap_ensure_allocated(&m
->watch_bus
, &string_hash_ops
);
854 r
= prioq_ensure_allocated(&m
->run_queue
, compare_job_priority
);
858 r
= manager_setup_prefix(m
);
862 r
= get_credentials_dir(&e
);
864 m
->received_credentials
= strdup(e
);
865 if (!m
->received_credentials
)
869 r
= sd_event_default(&m
->event
);
873 r
= manager_setup_run_queue(m
);
877 if (FLAGS_SET(test_run_flags
, MANAGER_TEST_RUN_MINIMAL
)) {
878 m
->cgroup_root
= strdup("");
882 r
= manager_setup_signals(m
);
886 r
= manager_setup_cgroup(m
);
890 r
= manager_setup_time_change(m
);
894 r
= manager_read_timezone_stat(m
);
898 (void) manager_setup_timezone_change(m
);
900 r
= manager_setup_sigchld_event_source(m
);
905 if (test_run_flags
== 0) {
906 if (MANAGER_IS_SYSTEM(m
))
907 r
= mkdir_label("/run/systemd/units", 0755);
909 _cleanup_free_
char *units_path
= NULL
;
910 r
= xdg_user_runtime_dir(&units_path
, "/systemd/units");
913 r
= mkdir_p_label(units_path
, 0755);
916 if (r
< 0 && r
!= -EEXIST
)
922 dir_is_empty("/usr") > 0;
924 /* Note that we do not set up the notify fd here. We do that after deserialization,
925 * since they might have gotten serialized across the reexec. */
932 static int manager_setup_notify(Manager
*m
) {
935 if (MANAGER_IS_TEST_RUN(m
))
938 if (m
->notify_fd
< 0) {
939 _cleanup_close_
int fd
= -1;
940 union sockaddr_union sa
;
943 /* First free all secondary fields */
944 m
->notify_socket
= mfree(m
->notify_socket
);
945 m
->notify_event_source
= sd_event_source_disable_unref(m
->notify_event_source
);
947 fd
= socket(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
|SOCK_NONBLOCK
, 0);
949 return log_error_errno(errno
, "Failed to allocate notification socket: %m");
951 fd_inc_rcvbuf(fd
, NOTIFY_RCVBUF_SIZE
);
953 m
->notify_socket
= path_join(m
->prefix
[EXEC_DIRECTORY_RUNTIME
], "systemd/notify");
954 if (!m
->notify_socket
)
957 r
= sockaddr_un_set_path(&sa
.un
, m
->notify_socket
);
959 return log_error_errno(r
, "Notify socket '%s' not valid for AF_UNIX socket address, refusing.",
963 (void) mkdir_parents_label(m
->notify_socket
, 0755);
964 (void) sockaddr_un_unlink(&sa
.un
);
966 r
= mac_selinux_bind(fd
, &sa
.sa
, sa_len
);
968 return log_error_errno(r
, "bind(%s) failed: %m", m
->notify_socket
);
970 r
= setsockopt_int(fd
, SOL_SOCKET
, SO_PASSCRED
, true);
972 return log_error_errno(r
, "SO_PASSCRED failed: %m");
974 m
->notify_fd
= TAKE_FD(fd
);
976 log_debug("Using notification socket %s", m
->notify_socket
);
979 if (!m
->notify_event_source
) {
980 r
= sd_event_add_io(m
->event
, &m
->notify_event_source
, m
->notify_fd
, EPOLLIN
, manager_dispatch_notify_fd
, m
);
982 return log_error_errno(r
, "Failed to allocate notify event source: %m");
984 /* Process notification messages a bit earlier than SIGCHLD, so that we can still identify to which
985 * service an exit message belongs. */
986 r
= sd_event_source_set_priority(m
->notify_event_source
, SD_EVENT_PRIORITY_NORMAL
-8);
988 return log_error_errno(r
, "Failed to set priority of notify event source: %m");
990 (void) sd_event_source_set_description(m
->notify_event_source
, "manager-notify");
996 static int manager_setup_cgroups_agent(Manager
*m
) {
998 static const union sockaddr_union sa
= {
999 .un
.sun_family
= AF_UNIX
,
1000 .un
.sun_path
= "/run/systemd/cgroups-agent",
1004 /* This creates a listening socket we receive cgroups agent messages on. We do not use D-Bus for delivering
1005 * these messages from the cgroups agent binary to PID 1, as the cgroups agent binary is very short-living, and
1006 * each instance of it needs a new D-Bus connection. Since D-Bus connections are SOCK_STREAM/AF_UNIX, on
1007 * overloaded systems the backlog of the D-Bus socket becomes relevant, as not more than the configured number
1008 * of D-Bus connections may be queued until the kernel will start dropping further incoming connections,
1009 * possibly resulting in lost cgroups agent messages. To avoid this, we'll use a private SOCK_DGRAM/AF_UNIX
1010 * socket, where no backlog is relevant as communication may take place without an actual connect() cycle, and
1011 * we thus won't lose messages.
1013 * Note that PID 1 will forward the agent message to system bus, so that the user systemd instance may listen
1014 * to it. The system instance hence listens on this special socket, but the user instances listen on the system
1015 * bus for these messages. */
1017 if (MANAGER_IS_TEST_RUN(m
))
1020 if (!MANAGER_IS_SYSTEM(m
))
1023 r
= cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER
);
1025 return log_error_errno(r
, "Failed to determine whether unified cgroups hierarchy is used: %m");
1026 if (r
> 0) /* We don't need this anymore on the unified hierarchy */
1029 if (m
->cgroups_agent_fd
< 0) {
1030 _cleanup_close_
int fd
= -1;
1032 /* First free all secondary fields */
1033 m
->cgroups_agent_event_source
= sd_event_source_disable_unref(m
->cgroups_agent_event_source
);
1035 fd
= socket(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
|SOCK_NONBLOCK
, 0);
1037 return log_error_errno(errno
, "Failed to allocate cgroups agent socket: %m");
1039 fd_inc_rcvbuf(fd
, CGROUPS_AGENT_RCVBUF_SIZE
);
1041 (void) sockaddr_un_unlink(&sa
.un
);
1043 /* Only allow root to connect to this socket */
1044 RUN_WITH_UMASK(0077)
1045 r
= bind(fd
, &sa
.sa
, SOCKADDR_UN_LEN(sa
.un
));
1047 return log_error_errno(errno
, "bind(%s) failed: %m", sa
.un
.sun_path
);
1049 m
->cgroups_agent_fd
= TAKE_FD(fd
);
1052 if (!m
->cgroups_agent_event_source
) {
1053 r
= sd_event_add_io(m
->event
, &m
->cgroups_agent_event_source
, m
->cgroups_agent_fd
, EPOLLIN
, manager_dispatch_cgroups_agent_fd
, m
);
1055 return log_error_errno(r
, "Failed to allocate cgroups agent event source: %m");
1057 /* Process cgroups notifications early. Note that when the agent notification is received
1058 * we'll just enqueue the unit in the cgroup empty queue, hence pick a high priority than
1059 * that. Also see handling of cgroup inotify for the unified cgroup stuff. */
1060 r
= sd_event_source_set_priority(m
->cgroups_agent_event_source
, SD_EVENT_PRIORITY_NORMAL
-9);
1062 return log_error_errno(r
, "Failed to set priority of cgroups agent event source: %m");
1064 (void) sd_event_source_set_description(m
->cgroups_agent_event_source
, "manager-cgroups-agent");
1070 static int manager_setup_user_lookup_fd(Manager
*m
) {
1075 /* Set up the socket pair used for passing UID/GID resolution results from forked off processes to PID
1076 * 1. Background: we can't do name lookups (NSS) from PID 1, since it might involve IPC and thus activation,
1077 * and we might hence deadlock on ourselves. Hence we do all user/group lookups asynchronously from the forked
1078 * off processes right before executing the binaries to start. In order to be able to clean up any IPC objects
1079 * created by a unit (see RemoveIPC=) we need to know in PID 1 the used UID/GID of the executed processes,
1080 * hence we establish this communication channel so that forked off processes can pass their UID/GID
1081 * information back to PID 1. The forked off processes send their resolved UID/GID to PID 1 in a simple
1082 * datagram, along with their unit name, so that we can share one communication socket pair among all units for
1085 * You might wonder why we need a communication channel for this that is independent of the usual notification
1086 * socket scheme (i.e. $NOTIFY_SOCKET). The primary difference is about trust: data sent via the $NOTIFY_SOCKET
1087 * channel is only accepted if it originates from the right unit and if reception was enabled for it. The user
1088 * lookup socket OTOH is only accessible by PID 1 and its children until they exec(), and always available.
1090 * Note that this function is called under two circumstances: when we first initialize (in which case we
1091 * allocate both the socket pair and the event source to listen on it), and when we deserialize after a reload
1092 * (in which case the socket pair already exists but we still need to allocate the event source for it). */
1094 if (m
->user_lookup_fds
[0] < 0) {
1096 /* Free all secondary fields */
1097 safe_close_pair(m
->user_lookup_fds
);
1098 m
->user_lookup_event_source
= sd_event_source_disable_unref(m
->user_lookup_event_source
);
1100 if (socketpair(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
, 0, m
->user_lookup_fds
) < 0)
1101 return log_error_errno(errno
, "Failed to allocate user lookup socket: %m");
1103 (void) fd_inc_rcvbuf(m
->user_lookup_fds
[0], NOTIFY_RCVBUF_SIZE
);
1106 if (!m
->user_lookup_event_source
) {
1107 r
= sd_event_add_io(m
->event
, &m
->user_lookup_event_source
, m
->user_lookup_fds
[0], EPOLLIN
, manager_dispatch_user_lookup_fd
, m
);
1109 return log_error_errno(errno
, "Failed to allocate user lookup event source: %m");
1111 /* Process even earlier than the notify event source, so that we always know first about valid UID/GID
1113 r
= sd_event_source_set_priority(m
->user_lookup_event_source
, SD_EVENT_PRIORITY_NORMAL
-11);
1115 return log_error_errno(errno
, "Failed to set priority of user lookup event source: %m");
1117 (void) sd_event_source_set_description(m
->user_lookup_event_source
, "user-lookup");
1123 static unsigned manager_dispatch_cleanup_queue(Manager
*m
) {
1129 while ((u
= m
->cleanup_queue
)) {
1130 assert(u
->in_cleanup_queue
);
1140 GC_OFFSET_IN_PATH
, /* This one is on the path we were traveling */
1141 GC_OFFSET_UNSURE
, /* No clue */
1142 GC_OFFSET_GOOD
, /* We still need this unit */
1143 GC_OFFSET_BAD
, /* We don't need this unit anymore */
1147 static void unit_gc_mark_good(Unit
*u
, unsigned gc_marker
) {
1150 u
->gc_marker
= gc_marker
+ GC_OFFSET_GOOD
;
1152 /* Recursively mark referenced units as GOOD as well */
1153 UNIT_FOREACH_DEPENDENCY(other
, u
, UNIT_ATOM_REFERENCES
)
1154 if (other
->gc_marker
== gc_marker
+ GC_OFFSET_UNSURE
)
1155 unit_gc_mark_good(other
, gc_marker
);
1158 static void unit_gc_sweep(Unit
*u
, unsigned gc_marker
) {
1164 if (IN_SET(u
->gc_marker
- gc_marker
,
1165 GC_OFFSET_GOOD
, GC_OFFSET_BAD
, GC_OFFSET_UNSURE
, GC_OFFSET_IN_PATH
))
1168 if (u
->in_cleanup_queue
)
1171 if (!unit_may_gc(u
))
1174 u
->gc_marker
= gc_marker
+ GC_OFFSET_IN_PATH
;
1178 UNIT_FOREACH_DEPENDENCY(other
, u
, UNIT_ATOM_REFERENCED_BY
) {
1179 unit_gc_sweep(other
, gc_marker
);
1181 if (other
->gc_marker
== gc_marker
+ GC_OFFSET_GOOD
)
1184 if (other
->gc_marker
!= gc_marker
+ GC_OFFSET_BAD
)
1189 LIST_FOREACH(refs_by_target
, ref
, u
->refs_by_target
) {
1190 unit_gc_sweep(ref
->source
, gc_marker
);
1192 if (ref
->source
->gc_marker
== gc_marker
+ GC_OFFSET_GOOD
)
1195 if (ref
->source
->gc_marker
!= gc_marker
+ GC_OFFSET_BAD
)
1202 /* We were unable to find anything out about this entry, so
1203 * let's investigate it later */
1204 u
->gc_marker
= gc_marker
+ GC_OFFSET_UNSURE
;
1205 unit_add_to_gc_queue(u
);
1209 /* We definitely know that this one is not useful anymore, so
1210 * let's mark it for deletion */
1211 u
->gc_marker
= gc_marker
+ GC_OFFSET_BAD
;
1212 unit_add_to_cleanup_queue(u
);
1216 unit_gc_mark_good(u
, gc_marker
);
1219 static unsigned manager_dispatch_gc_unit_queue(Manager
*m
) {
1220 unsigned n
= 0, gc_marker
;
1225 /* log_debug("Running GC..."); */
1227 m
->gc_marker
+= _GC_OFFSET_MAX
;
1228 if (m
->gc_marker
+ _GC_OFFSET_MAX
<= _GC_OFFSET_MAX
)
1231 gc_marker
= m
->gc_marker
;
1233 while ((u
= m
->gc_unit_queue
)) {
1234 assert(u
->in_gc_queue
);
1236 unit_gc_sweep(u
, gc_marker
);
1238 LIST_REMOVE(gc_queue
, m
->gc_unit_queue
, u
);
1239 u
->in_gc_queue
= false;
1243 if (IN_SET(u
->gc_marker
- gc_marker
,
1244 GC_OFFSET_BAD
, GC_OFFSET_UNSURE
)) {
1246 log_unit_debug(u
, "Collecting.");
1247 u
->gc_marker
= gc_marker
+ GC_OFFSET_BAD
;
1248 unit_add_to_cleanup_queue(u
);
1255 static unsigned manager_dispatch_gc_job_queue(Manager
*m
) {
1261 while ((j
= m
->gc_job_queue
)) {
1262 assert(j
->in_gc_queue
);
1264 LIST_REMOVE(gc_queue
, m
->gc_job_queue
, j
);
1265 j
->in_gc_queue
= false;
1272 log_unit_debug(j
->unit
, "Collecting job.");
1273 (void) job_finish_and_invalidate(j
, JOB_COLLECTED
, false, false);
1279 static unsigned manager_dispatch_stop_when_unneeded_queue(Manager
*m
) {
1286 while ((u
= m
->stop_when_unneeded_queue
)) {
1287 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
1289 assert(u
->in_stop_when_unneeded_queue
);
1290 LIST_REMOVE(stop_when_unneeded_queue
, m
->stop_when_unneeded_queue
, u
);
1291 u
->in_stop_when_unneeded_queue
= false;
1295 if (!unit_is_unneeded(u
))
1298 log_unit_debug(u
, "Unit is not needed anymore.");
1300 /* If stopping a unit fails continuously we might enter a stop loop here, hence stop acting on the
1301 * service being unnecessary after a while. */
1303 if (!ratelimit_below(&u
->auto_start_stop_ratelimit
)) {
1304 log_unit_warning(u
, "Unit not needed anymore, but not stopping since we tried this too often recently.");
1308 /* Ok, nobody needs us anymore. Sniff. Then let's commit suicide */
1309 r
= manager_add_job(u
->manager
, JOB_STOP
, u
, JOB_FAIL
, NULL
, &error
, NULL
);
1311 log_unit_warning_errno(u
, r
, "Failed to enqueue stop job, ignoring: %s", bus_error_message(&error
, r
));
1317 static unsigned manager_dispatch_start_when_upheld_queue(Manager
*m
) {
1324 while ((u
= m
->start_when_upheld_queue
)) {
1325 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
1326 Unit
*culprit
= NULL
;
1328 assert(u
->in_start_when_upheld_queue
);
1329 LIST_REMOVE(start_when_upheld_queue
, m
->start_when_upheld_queue
, u
);
1330 u
->in_start_when_upheld_queue
= false;
1334 if (!unit_is_upheld_by_active(u
, &culprit
))
1337 log_unit_debug(u
, "Unit is started because upheld by active unit %s.", culprit
->id
);
1339 /* If stopping a unit fails continuously we might enter a stop loop here, hence stop acting on the
1340 * service being unnecessary after a while. */
1342 if (!ratelimit_below(&u
->auto_start_stop_ratelimit
)) {
1343 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
);
1347 r
= manager_add_job(u
->manager
, JOB_START
, u
, JOB_FAIL
, NULL
, &error
, NULL
);
1349 log_unit_warning_errno(u
, r
, "Failed to enqueue start job, ignoring: %s", bus_error_message(&error
, r
));
1355 static unsigned manager_dispatch_stop_when_bound_queue(Manager
*m
) {
1362 while ((u
= m
->stop_when_bound_queue
)) {
1363 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
1364 Unit
*culprit
= NULL
;
1366 assert(u
->in_stop_when_bound_queue
);
1367 LIST_REMOVE(stop_when_bound_queue
, m
->stop_when_bound_queue
, u
);
1368 u
->in_stop_when_bound_queue
= false;
1372 if (!unit_is_bound_by_inactive(u
, &culprit
))
1375 log_unit_debug(u
, "Unit is stopped because bound to inactive unit %s.", culprit
->id
);
1377 /* If stopping a unit fails continuously we might enter a stop loop here, hence stop acting on the
1378 * service being unnecessary after a while. */
1380 if (!ratelimit_below(&u
->auto_start_stop_ratelimit
)) {
1381 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
);
1385 r
= manager_add_job(u
->manager
, JOB_STOP
, u
, JOB_REPLACE
, NULL
, &error
, NULL
);
1387 log_unit_warning_errno(u
, r
, "Failed to enqueue stop job, ignoring: %s", bus_error_message(&error
, r
));
1393 static void manager_clear_jobs_and_units(Manager
*m
) {
1398 while ((u
= hashmap_first(m
->units
)))
1401 manager_dispatch_cleanup_queue(m
);
1403 assert(!m
->load_queue
);
1404 assert(prioq_isempty(m
->run_queue
));
1405 assert(!m
->dbus_unit_queue
);
1406 assert(!m
->dbus_job_queue
);
1407 assert(!m
->cleanup_queue
);
1408 assert(!m
->gc_unit_queue
);
1409 assert(!m
->gc_job_queue
);
1410 assert(!m
->stop_when_unneeded_queue
);
1411 assert(!m
->start_when_upheld_queue
);
1412 assert(!m
->stop_when_bound_queue
);
1414 assert(hashmap_isempty(m
->jobs
));
1415 assert(hashmap_isempty(m
->units
));
1417 m
->n_on_console
= 0;
1418 m
->n_running_jobs
= 0;
1419 m
->n_installed_jobs
= 0;
1420 m
->n_failed_jobs
= 0;
1423 Manager
* manager_free(Manager
*m
) {
1427 manager_clear_jobs_and_units(m
);
1429 for (UnitType c
= 0; c
< _UNIT_TYPE_MAX
; c
++)
1430 if (unit_vtable
[c
]->shutdown
)
1431 unit_vtable
[c
]->shutdown(m
);
1433 /* Keep the cgroup hierarchy in place except when we know we are going down for good */
1434 manager_shutdown_cgroup(m
, IN_SET(m
->objective
, MANAGER_EXIT
, MANAGER_REBOOT
, MANAGER_POWEROFF
, MANAGER_HALT
, MANAGER_KEXEC
));
1436 lookup_paths_flush_generator(&m
->lookup_paths
);
1439 manager_varlink_done(m
);
1441 exec_runtime_vacuum(m
);
1442 hashmap_free(m
->exec_runtime_by_id
);
1444 dynamic_user_vacuum(m
, false);
1445 hashmap_free(m
->dynamic_users
);
1447 hashmap_free(m
->units
);
1448 hashmap_free(m
->units_by_invocation_id
);
1449 hashmap_free(m
->jobs
);
1450 hashmap_free(m
->watch_pids
);
1451 hashmap_free(m
->watch_bus
);
1453 prioq_free(m
->run_queue
);
1455 set_free(m
->startup_units
);
1456 set_free(m
->failed_units
);
1458 sd_event_source_unref(m
->signal_event_source
);
1459 sd_event_source_unref(m
->sigchld_event_source
);
1460 sd_event_source_unref(m
->notify_event_source
);
1461 sd_event_source_unref(m
->cgroups_agent_event_source
);
1462 sd_event_source_unref(m
->time_change_event_source
);
1463 sd_event_source_unref(m
->timezone_change_event_source
);
1464 sd_event_source_unref(m
->jobs_in_progress_event_source
);
1465 sd_event_source_unref(m
->run_queue_event_source
);
1466 sd_event_source_unref(m
->user_lookup_event_source
);
1468 safe_close(m
->signal_fd
);
1469 safe_close(m
->notify_fd
);
1470 safe_close(m
->cgroups_agent_fd
);
1471 safe_close(m
->time_change_fd
);
1472 safe_close_pair(m
->user_lookup_fds
);
1474 manager_close_ask_password(m
);
1476 manager_close_idle_pipe(m
);
1478 sd_event_unref(m
->event
);
1480 free(m
->notify_socket
);
1482 lookup_paths_free(&m
->lookup_paths
);
1483 strv_free(m
->transient_environment
);
1484 strv_free(m
->client_environment
);
1486 hashmap_free(m
->cgroup_unit
);
1487 manager_free_unit_name_maps(m
);
1489 free(m
->switch_root
);
1490 free(m
->switch_root_init
);
1492 rlimit_free_all(m
->rlimit
);
1494 assert(hashmap_isempty(m
->units_requiring_mounts_for
));
1495 hashmap_free(m
->units_requiring_mounts_for
);
1497 hashmap_free(m
->uid_refs
);
1498 hashmap_free(m
->gid_refs
);
1500 for (ExecDirectoryType dt
= 0; dt
< _EXEC_DIRECTORY_TYPE_MAX
; dt
++)
1501 m
->prefix
[dt
] = mfree(m
->prefix
[dt
]);
1502 free(m
->received_credentials
);
1507 static void manager_enumerate_perpetual(Manager
*m
) {
1510 if (m
->test_run_flags
== MANAGER_TEST_RUN_MINIMAL
)
1513 /* Let's ask every type to load all units from disk/kernel that it might know */
1514 for (UnitType c
= 0; c
< _UNIT_TYPE_MAX
; c
++) {
1515 if (!unit_type_supported(c
)) {
1516 log_debug("Unit type .%s is not supported on this system.", unit_type_to_string(c
));
1520 if (unit_vtable
[c
]->enumerate_perpetual
)
1521 unit_vtable
[c
]->enumerate_perpetual(m
);
1525 static void manager_enumerate(Manager
*m
) {
1528 if (m
->test_run_flags
== MANAGER_TEST_RUN_MINIMAL
)
1531 /* Let's ask every type to load all units from disk/kernel that it might know */
1532 for (UnitType c
= 0; c
< _UNIT_TYPE_MAX
; c
++) {
1533 if (!unit_type_supported(c
)) {
1534 log_debug("Unit type .%s is not supported on this system.", unit_type_to_string(c
));
1538 if (unit_vtable
[c
]->enumerate
)
1539 unit_vtable
[c
]->enumerate(m
);
1542 manager_dispatch_load_queue(m
);
1545 static void manager_coldplug(Manager
*m
) {
1552 log_debug("Invoking unit coldplug() handlers…");
1554 /* Let's place the units back into their deserialized state */
1555 HASHMAP_FOREACH_KEY(u
, k
, m
->units
) {
1557 /* ignore aliases */
1561 r
= unit_coldplug(u
);
1563 log_warning_errno(r
, "We couldn't coldplug %s, proceeding anyway: %m", u
->id
);
1567 static void manager_catchup(Manager
*m
) {
1573 log_debug("Invoking unit catchup() handlers…");
1575 /* Let's catch up on any state changes that happened while we were reloading/reexecing */
1576 HASHMAP_FOREACH_KEY(u
, k
, m
->units
) {
1578 /* ignore aliases */
1586 static void manager_distribute_fds(Manager
*m
, FDSet
*fds
) {
1591 HASHMAP_FOREACH(u
, m
->units
) {
1593 if (fdset_size(fds
) <= 0)
1596 if (!UNIT_VTABLE(u
)->distribute_fds
)
1599 UNIT_VTABLE(u
)->distribute_fds(u
, fds
);
1603 static bool manager_dbus_is_running(Manager
*m
, bool deserialized
) {
1608 /* This checks whether the dbus instance we are supposed to expose our APIs on is up. We check both the socket
1609 * and the service unit. If the 'deserialized' parameter is true we'll check the deserialized state of the unit
1610 * rather than the current one. */
1612 if (MANAGER_IS_TEST_RUN(m
))
1615 u
= manager_get_unit(m
, SPECIAL_DBUS_SOCKET
);
1618 if ((deserialized
? SOCKET(u
)->deserialized_state
: SOCKET(u
)->state
) != SOCKET_RUNNING
)
1621 u
= manager_get_unit(m
, SPECIAL_DBUS_SERVICE
);
1624 if (!IN_SET((deserialized
? SERVICE(u
)->deserialized_state
: SERVICE(u
)->state
), SERVICE_RUNNING
, SERVICE_RELOAD
))
1630 static void manager_setup_bus(Manager
*m
) {
1633 /* Let's set up our private bus connection now, unconditionally */
1634 (void) bus_init_private(m
);
1636 /* If we are in --user mode also connect to the system bus now */
1637 if (MANAGER_IS_USER(m
))
1638 (void) bus_init_system(m
);
1640 /* Let's connect to the bus now, but only if the unit is supposed to be up */
1641 if (manager_dbus_is_running(m
, MANAGER_IS_RELOADING(m
))) {
1642 (void) bus_init_api(m
);
1644 if (MANAGER_IS_SYSTEM(m
))
1645 (void) bus_init_system(m
);
1649 static void manager_preset_all(Manager
*m
) {
1654 if (m
->first_boot
<= 0)
1657 if (!MANAGER_IS_SYSTEM(m
))
1660 if (MANAGER_IS_TEST_RUN(m
))
1663 /* If this is the first boot, and we are in the host system, then preset everything */
1664 r
= unit_file_preset_all(UNIT_FILE_SYSTEM
, 0, NULL
, UNIT_FILE_PRESET_ENABLE_ONLY
, NULL
, 0);
1666 log_full_errno(r
== -EEXIST
? LOG_NOTICE
: LOG_WARNING
, r
,
1667 "Failed to populate /etc with preset unit settings, ignoring: %m");
1669 log_info("Populated /etc with preset unit settings.");
1672 static void manager_ready(Manager
*m
) {
1675 /* After having loaded everything, do the final round of catching up with what might have changed */
1677 m
->objective
= MANAGER_OK
; /* Tell everyone we are up now */
1679 /* It might be safe to log to the journal now and connect to dbus */
1680 manager_recheck_journal(m
);
1681 manager_recheck_dbus(m
);
1683 /* Let's finally catch up with any changes that took place while we were reloading/reexecing */
1686 m
->honor_device_enumeration
= true;
1689 static Manager
* manager_reloading_start(Manager
*m
) {
1693 static void manager_reloading_stopp(Manager
**m
) {
1695 assert((*m
)->n_reloading
> 0);
1696 (*m
)->n_reloading
--;
1700 int manager_startup(Manager
*m
, FILE *serialization
, FDSet
*fds
) {
1705 /* If we are running in test mode, we still want to run the generators,
1706 * but we should not touch the real generator directories. */
1707 r
= lookup_paths_init(&m
->lookup_paths
, m
->unit_file_scope
,
1708 MANAGER_IS_TEST_RUN(m
) ? LOOKUP_PATHS_TEMPORARY_GENERATED
: 0,
1711 return log_error_errno(r
, "Failed to initialize path lookup table: %m");
1713 dual_timestamp_get(m
->timestamps
+ manager_timestamp_initrd_mangle(MANAGER_TIMESTAMP_GENERATORS_START
));
1714 r
= manager_run_environment_generators(m
);
1716 r
= manager_run_generators(m
);
1717 dual_timestamp_get(m
->timestamps
+ manager_timestamp_initrd_mangle(MANAGER_TIMESTAMP_GENERATORS_FINISH
));
1721 manager_preset_all(m
);
1723 lookup_paths_log(&m
->lookup_paths
);
1726 /* This block is (optionally) done with the reloading counter bumped */
1727 _cleanup_(manager_reloading_stopp
) Manager
*reloading
= NULL
;
1729 /* If we will deserialize make sure that during enumeration this is already known, so we increase the
1730 * counter here already */
1732 reloading
= manager_reloading_start(m
);
1734 /* First, enumerate what we can from all config files */
1735 dual_timestamp_get(m
->timestamps
+ manager_timestamp_initrd_mangle(MANAGER_TIMESTAMP_UNITS_LOAD_START
));
1736 manager_enumerate_perpetual(m
);
1737 manager_enumerate(m
);
1738 dual_timestamp_get(m
->timestamps
+ manager_timestamp_initrd_mangle(MANAGER_TIMESTAMP_UNITS_LOAD_FINISH
));
1740 /* Second, deserialize if there is something to deserialize */
1741 if (serialization
) {
1742 r
= manager_deserialize(m
, serialization
, fds
);
1744 return log_error_errno(r
, "Deserialization failed: %m");
1747 /* Any fds left? Find some unit which wants them. This is useful to allow container managers to pass
1748 * some file descriptors to us pre-initialized. This enables socket-based activation of entire
1750 manager_distribute_fds(m
, fds
);
1752 /* We might have deserialized the notify fd, but if we didn't then let's create the bus now */
1753 r
= manager_setup_notify(m
);
1755 /* No sense to continue without notifications, our children would fail anyway. */
1758 r
= manager_setup_cgroups_agent(m
);
1760 /* Likewise, no sense to continue without empty cgroup notifications. */
1763 r
= manager_setup_user_lookup_fd(m
);
1765 /* This shouldn't fail, except if things are really broken. */
1768 /* Connect to the bus if we are good for it */
1769 manager_setup_bus(m
);
1771 /* Now that we are connected to all possible buses, let's deserialize who is tracking us. */
1772 r
= bus_track_coldplug(m
, &m
->subscribed
, false, m
->deserialized_subscribed
);
1774 log_warning_errno(r
, "Failed to deserialized tracked clients, ignoring: %m");
1775 m
->deserialized_subscribed
= strv_free(m
->deserialized_subscribed
);
1777 r
= manager_varlink_init(m
);
1779 log_warning_errno(r
, "Failed to set up Varlink server, ignoring: %m");
1781 /* Third, fire things up! */
1782 manager_coldplug(m
);
1784 /* Clean up runtime objects */
1788 /* Let's wait for the UnitNew/JobNew messages being sent, before we notify that the
1789 * reload is finished */
1790 m
->send_reloading_done
= true;
1798 int manager_add_job(
1804 sd_bus_error
*error
,
1811 assert(type
< _JOB_TYPE_MAX
);
1813 assert(mode
< _JOB_MODE_MAX
);
1815 if (mode
== JOB_ISOLATE
&& type
!= JOB_START
)
1816 return sd_bus_error_set(error
, SD_BUS_ERROR_INVALID_ARGS
, "Isolate is only valid for start.");
1818 if (mode
== JOB_ISOLATE
&& !unit
->allow_isolate
)
1819 return sd_bus_error_set(error
, BUS_ERROR_NO_ISOLATION
, "Operation refused, unit may not be isolated.");
1821 if (mode
== JOB_TRIGGERING
&& type
!= JOB_STOP
)
1822 return sd_bus_error_set(error
, SD_BUS_ERROR_INVALID_ARGS
, "--job-mode=triggering is only valid for stop.");
1824 log_unit_debug(unit
, "Trying to enqueue job %s/%s/%s", unit
->id
, job_type_to_string(type
), job_mode_to_string(mode
));
1826 type
= job_type_collapse(type
, unit
);
1828 tr
= transaction_new(mode
== JOB_REPLACE_IRREVERSIBLY
);
1832 r
= transaction_add_job_and_dependencies(tr
, type
, unit
, NULL
, true, false,
1833 IN_SET(mode
, JOB_IGNORE_DEPENDENCIES
, JOB_IGNORE_REQUIREMENTS
),
1834 mode
== JOB_IGNORE_DEPENDENCIES
, error
);
1838 if (mode
== JOB_ISOLATE
) {
1839 r
= transaction_add_isolate_jobs(tr
, m
);
1844 if (mode
== JOB_TRIGGERING
) {
1845 r
= transaction_add_triggering_jobs(tr
, unit
);
1850 r
= transaction_activate(tr
, m
, mode
, affected_jobs
, error
);
1854 log_unit_debug(unit
,
1855 "Enqueued job %s/%s as %u", unit
->id
,
1856 job_type_to_string(type
), (unsigned) tr
->anchor_job
->id
);
1859 *ret
= tr
->anchor_job
;
1861 transaction_free(tr
);
1865 transaction_abort(tr
);
1866 transaction_free(tr
);
1870 int manager_add_job_by_name(Manager
*m
, JobType type
, const char *name
, JobMode mode
, Set
*affected_jobs
, sd_bus_error
*e
, Job
**ret
) {
1871 Unit
*unit
= NULL
; /* just to appease gcc, initialization is not really necessary */
1875 assert(type
< _JOB_TYPE_MAX
);
1877 assert(mode
< _JOB_MODE_MAX
);
1879 r
= manager_load_unit(m
, name
, NULL
, NULL
, &unit
);
1884 return manager_add_job(m
, type
, unit
, mode
, affected_jobs
, e
, ret
);
1887 int manager_add_job_by_name_and_warn(Manager
*m
, JobType type
, const char *name
, JobMode mode
, Set
*affected_jobs
, Job
**ret
) {
1888 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
1892 assert(type
< _JOB_TYPE_MAX
);
1894 assert(mode
< _JOB_MODE_MAX
);
1896 r
= manager_add_job_by_name(m
, type
, name
, mode
, affected_jobs
, &error
, ret
);
1898 return log_warning_errno(r
, "Failed to enqueue %s job for %s: %s", job_mode_to_string(mode
), name
, bus_error_message(&error
, r
));
1903 int manager_propagate_reload(Manager
*m
, Unit
*unit
, JobMode mode
, sd_bus_error
*e
) {
1909 assert(mode
< _JOB_MODE_MAX
);
1910 assert(mode
!= JOB_ISOLATE
); /* Isolate is only valid for start */
1912 tr
= transaction_new(mode
== JOB_REPLACE_IRREVERSIBLY
);
1916 /* We need an anchor job */
1917 r
= transaction_add_job_and_dependencies(tr
, JOB_NOP
, unit
, NULL
, false, false, true, true, e
);
1921 /* Failure in adding individual dependencies is ignored, so this always succeeds. */
1922 transaction_add_propagate_reload_jobs(tr
, unit
, tr
->anchor_job
, mode
== JOB_IGNORE_DEPENDENCIES
, e
);
1924 r
= transaction_activate(tr
, m
, mode
, NULL
, e
);
1928 transaction_free(tr
);
1932 transaction_abort(tr
);
1933 transaction_free(tr
);
1937 Job
*manager_get_job(Manager
*m
, uint32_t id
) {
1940 return hashmap_get(m
->jobs
, UINT32_TO_PTR(id
));
1943 Unit
*manager_get_unit(Manager
*m
, const char *name
) {
1947 return hashmap_get(m
->units
, name
);
1950 static int manager_dispatch_target_deps_queue(Manager
*m
) {
1956 while ((u
= m
->target_deps_queue
)) {
1957 _cleanup_free_ Unit
**targets
= NULL
;
1960 assert(u
->in_target_deps_queue
);
1962 LIST_REMOVE(target_deps_queue
, u
->manager
->target_deps_queue
, u
);
1963 u
->in_target_deps_queue
= false;
1965 /* Take an "atomic" snapshot of dependencies here, as the call below will likely modify the
1966 * dependencies, and we can't have it that hash tables we iterate through are modified while
1967 * we are iterating through them. */
1968 n_targets
= unit_get_dependency_array(u
, UNIT_ATOM_DEFAULT_TARGET_DEPENDENCIES
, &targets
);
1972 for (int i
= 0; i
< n_targets
; i
++) {
1973 r
= unit_add_default_target_dependency(u
, targets
[i
]);
1982 unsigned manager_dispatch_load_queue(Manager
*m
) {
1988 /* Make sure we are not run recursively */
1989 if (m
->dispatching_load_queue
)
1992 m
->dispatching_load_queue
= true;
1994 /* Dispatches the load queue. Takes a unit from the queue and
1995 * tries to load its data until the queue is empty */
1997 while ((u
= m
->load_queue
)) {
1998 assert(u
->in_load_queue
);
2004 m
->dispatching_load_queue
= false;
2006 /* Dispatch the units waiting for their target dependencies to be added now, as all targets that we know about
2007 * should be loaded and have aliases resolved */
2008 (void) manager_dispatch_target_deps_queue(m
);
2013 bool manager_unit_cache_should_retry_load(Unit
*u
) {
2016 /* Automatic reloading from disk only applies to units which were not found sometime in the past, and
2017 * the not-found stub is kept pinned in the unit graph by dependencies. For units that were
2018 * previously loaded, we don't do automatic reloading, and daemon-reload is necessary to update. */
2019 if (u
->load_state
!= UNIT_NOT_FOUND
)
2022 /* The cache has been updated since the last time we tried to load the unit. There might be new
2023 * fragment paths to read. */
2024 if (u
->manager
->unit_cache_timestamp_hash
!= u
->fragment_not_found_timestamp_hash
)
2027 /* The cache needs to be updated because there are modifications on disk. */
2028 return !lookup_paths_timestamp_hash_same(&u
->manager
->lookup_paths
, u
->manager
->unit_cache_timestamp_hash
, NULL
);
2031 int manager_load_unit_prepare(
2038 _cleanup_(unit_freep
) Unit
*cleanup_ret
= NULL
;
2046 /* This will prepare the unit for loading, but not actually load anything from disk. */
2048 if (path
&& !path_is_absolute(path
))
2049 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
, "Path %s is not absolute.", path
);
2052 /* 'name' and 'path' must not both be null. Check here 'path' using assert_se() to
2053 * workaround a bug in gcc that generates a -Wnonnull warning when calling basename(),
2054 * but this cannot be possible in any code path (See #6119). */
2056 name
= basename(path
);
2059 t
= unit_name_to_type(name
);
2061 if (t
== _UNIT_TYPE_INVALID
|| !unit_name_is_valid(name
, UNIT_NAME_PLAIN
|UNIT_NAME_INSTANCE
)) {
2062 if (unit_name_is_valid(name
, UNIT_NAME_TEMPLATE
))
2063 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
, "Unit name %s is missing the instance name.", name
);
2065 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
, "Unit name %s is not valid.", name
);
2068 ret
= manager_get_unit(m
, name
);
2070 /* The time-based cache allows to start new units without daemon-reload,
2071 * but if they are already referenced (because of dependencies or ordering)
2072 * then we have to force a load of the fragment. As an optimization, check
2073 * first if anything in the usual paths was modified since the last time
2074 * the cache was loaded. Also check if the last time an attempt to load the
2075 * unit was made was before the most recent cache refresh, so that we know
2076 * we need to try again — even if the cache is current, it might have been
2077 * updated in a different context before we had a chance to retry loading
2078 * this particular unit. */
2079 if (manager_unit_cache_should_retry_load(ret
))
2080 ret
->load_state
= UNIT_STUB
;
2086 ret
= cleanup_ret
= unit_new(m
, unit_vtable
[t
]->object_size
);
2092 r
= free_and_strdup(&ret
->fragment_path
, path
);
2097 r
= unit_add_name(ret
, name
);
2101 unit_add_to_load_queue(ret
);
2102 unit_add_to_dbus_queue(ret
);
2103 unit_add_to_gc_queue(ret
);
2111 int manager_load_unit(
2123 /* This will load the service information files, but not actually
2124 * start any services or anything. */
2126 r
= manager_load_unit_prepare(m
, name
, path
, e
, _ret
);
2130 manager_dispatch_load_queue(m
);
2132 *_ret
= unit_follow_merge(*_ret
);
2136 int manager_load_startable_unit_or_warn(
2142 /* Load a unit, make sure it loaded fully and is not masked. */
2144 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
2148 r
= manager_load_unit(m
, name
, path
, &error
, &unit
);
2150 return log_error_errno(r
, "Failed to load %s %s: %s",
2151 name
? "unit" : "unit file", name
?: path
,
2152 bus_error_message(&error
, r
));
2154 r
= bus_unit_validate_load_state(unit
, &error
);
2156 return log_error_errno(r
, "%s", bus_error_message(&error
, r
));
2162 void manager_clear_jobs(Manager
*m
) {
2167 while ((j
= hashmap_first(m
->jobs
)))
2168 /* No need to recurse. We're cancelling all jobs. */
2169 job_finish_and_invalidate(j
, JOB_CANCELED
, false, false);
2172 void manager_unwatch_pid(Manager
*m
, pid_t pid
) {
2175 /* First let's drop the unit keyed as "pid". */
2176 (void) hashmap_remove(m
->watch_pids
, PID_TO_PTR(pid
));
2178 /* Then, let's also drop the array keyed by -pid. */
2179 free(hashmap_remove(m
->watch_pids
, PID_TO_PTR(-pid
)));
2182 static int manager_dispatch_run_queue(sd_event_source
*source
, void *userdata
) {
2183 Manager
*m
= userdata
;
2189 while ((j
= prioq_peek(m
->run_queue
))) {
2190 assert(j
->installed
);
2191 assert(j
->in_run_queue
);
2193 (void) job_run_and_invalidate(j
);
2196 if (m
->n_running_jobs
> 0)
2197 manager_watch_jobs_in_progress(m
);
2199 if (m
->n_on_console
> 0)
2200 manager_watch_idle_pipe(m
);
2205 static unsigned manager_dispatch_dbus_queue(Manager
*m
) {
2206 unsigned n
= 0, budget
;
2212 /* When we are reloading, let's not wait with generating signals, since we need to exit the manager as quickly
2213 * as we can. There's no point in throttling generation of signals in that case. */
2214 if (MANAGER_IS_RELOADING(m
) || m
->send_reloading_done
|| m
->pending_reload_message
)
2215 budget
= UINT_MAX
; /* infinite budget in this case */
2217 /* Anything to do at all? */
2218 if (!m
->dbus_unit_queue
&& !m
->dbus_job_queue
)
2221 /* Do we have overly many messages queued at the moment? If so, let's not enqueue more on top, let's
2222 * sit this cycle out, and process things in a later cycle when the queues got a bit emptier. */
2223 if (manager_bus_n_queued_write(m
) > MANAGER_BUS_BUSY_THRESHOLD
)
2226 /* Only process a certain number of units/jobs per event loop iteration. Even if the bus queue wasn't
2227 * overly full before this call we shouldn't increase it in size too wildly in one step, and we
2228 * shouldn't monopolize CPU time with generating these messages. Note the difference in counting of
2229 * this "budget" and the "threshold" above: the "budget" is decreased only once per generated message,
2230 * regardless how many buses/direct connections it is enqueued on, while the "threshold" is applied to
2231 * each queued instance of bus message, i.e. if the same message is enqueued to five buses/direct
2232 * connections it will be counted five times. This difference in counting ("references"
2233 * vs. "instances") is primarily a result of the fact that it's easier to implement it this way,
2234 * however it also reflects the thinking that the "threshold" should put a limit on used queue memory,
2235 * i.e. space, while the "budget" should put a limit on time. Also note that the "threshold" is
2236 * currently chosen much higher than the "budget". */
2237 budget
= MANAGER_BUS_MESSAGE_BUDGET
;
2240 while (budget
!= 0 && (u
= m
->dbus_unit_queue
)) {
2242 assert(u
->in_dbus_queue
);
2244 bus_unit_send_change_signal(u
);
2247 if (budget
!= UINT_MAX
)
2251 while (budget
!= 0 && (j
= m
->dbus_job_queue
)) {
2252 assert(j
->in_dbus_queue
);
2254 bus_job_send_change_signal(j
);
2257 if (budget
!= UINT_MAX
)
2261 if (m
->send_reloading_done
) {
2262 m
->send_reloading_done
= false;
2263 bus_manager_send_reloading(m
, false);
2267 if (m
->pending_reload_message
) {
2268 bus_send_pending_reload_message(m
);
2275 static int manager_dispatch_cgroups_agent_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
2276 Manager
*m
= userdata
;
2280 n
= recv(fd
, buf
, sizeof(buf
), 0);
2282 return log_error_errno(errno
, "Failed to read cgroups agent message: %m");
2284 log_error("Got zero-length cgroups agent message, ignoring.");
2287 if ((size_t) n
>= sizeof(buf
)) {
2288 log_error("Got overly long cgroups agent message, ignoring.");
2292 if (memchr(buf
, 0, n
)) {
2293 log_error("Got cgroups agent message with embedded NUL byte, ignoring.");
2298 manager_notify_cgroup_empty(m
, buf
);
2299 (void) bus_forward_agent_released(m
, buf
);
2304 static bool manager_process_barrier_fd(char * const *tags
, FDSet
*fds
) {
2306 /* nothing else must be sent when using BARRIER=1 */
2307 if (strv_contains(tags
, "BARRIER=1")) {
2308 if (strv_length(tags
) == 1) {
2309 if (fdset_size(fds
) != 1)
2310 log_warning("Got incorrect number of fds with BARRIER=1, closing them.");
2312 log_warning("Extra notification messages sent with BARRIER=1, ignoring everything.");
2314 /* Drop the message if BARRIER=1 was found */
2321 static void manager_invoke_notify_message(
2324 const struct ucred
*ucred
,
2333 if (u
->notifygen
== m
->notifygen
) /* Already invoked on this same unit in this same iteration? */
2335 u
->notifygen
= m
->notifygen
;
2337 if (UNIT_VTABLE(u
)->notify_message
)
2338 UNIT_VTABLE(u
)->notify_message(u
, ucred
, tags
, fds
);
2340 else if (DEBUG_LOGGING
) {
2341 _cleanup_free_
char *buf
= NULL
, *x
= NULL
, *y
= NULL
;
2343 buf
= strv_join(tags
, ", ");
2345 x
= ellipsize(buf
, 20, 90);
2349 log_unit_debug(u
, "Got notification message \"%s\", ignoring.", strnull(y
));
2353 static int manager_dispatch_notify_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
2355 _cleanup_fdset_free_ FDSet
*fds
= NULL
;
2356 Manager
*m
= userdata
;
2357 char buf
[NOTIFY_BUFFER_MAX
+1];
2358 struct iovec iovec
= {
2360 .iov_len
= sizeof(buf
)-1,
2362 CMSG_BUFFER_TYPE(CMSG_SPACE(sizeof(struct ucred
)) +
2363 CMSG_SPACE(sizeof(int) * NOTIFY_FD_MAX
)) control
;
2364 struct msghdr msghdr
= {
2367 .msg_control
= &control
,
2368 .msg_controllen
= sizeof(control
),
2371 struct cmsghdr
*cmsg
;
2372 struct ucred
*ucred
= NULL
;
2373 _cleanup_free_ Unit
**array_copy
= NULL
;
2374 _cleanup_strv_free_
char **tags
= NULL
;
2375 Unit
*u1
, *u2
, **array
;
2376 int r
, *fd_array
= NULL
;
2382 assert(m
->notify_fd
== fd
);
2384 if (revents
!= EPOLLIN
) {
2385 log_warning("Got unexpected poll event for notify fd.");
2389 n
= recvmsg_safe(m
->notify_fd
, &msghdr
, MSG_DONTWAIT
|MSG_CMSG_CLOEXEC
|MSG_TRUNC
);
2390 if (IN_SET(n
, -EAGAIN
, -EINTR
))
2391 return 0; /* Spurious wakeup, try again */
2393 log_warning("Got message with truncated control data (too many fds sent?), ignoring.");
2397 /* If this is any other, real error, then let's stop processing this socket. This of course
2398 * means we won't take notification messages anymore, but that's still better than busy
2399 * looping around this: being woken up over and over again but being unable to actually read
2400 * the message off the socket. */
2401 return log_error_errno(n
, "Failed to receive notification message: %m");
2403 CMSG_FOREACH(cmsg
, &msghdr
) {
2404 if (cmsg
->cmsg_level
== SOL_SOCKET
&& cmsg
->cmsg_type
== SCM_RIGHTS
) {
2407 fd_array
= (int*) CMSG_DATA(cmsg
);
2408 n_fds
= (cmsg
->cmsg_len
- CMSG_LEN(0)) / sizeof(int);
2410 } else if (cmsg
->cmsg_level
== SOL_SOCKET
&&
2411 cmsg
->cmsg_type
== SCM_CREDENTIALS
&&
2412 cmsg
->cmsg_len
== CMSG_LEN(sizeof(struct ucred
))) {
2415 ucred
= (struct ucred
*) CMSG_DATA(cmsg
);
2422 r
= fdset_new_array(&fds
, fd_array
, n_fds
);
2424 close_many(fd_array
, n_fds
);
2430 if (!ucred
|| !pid_is_valid(ucred
->pid
)) {
2431 log_warning("Received notify message without valid credentials. Ignoring.");
2435 if ((size_t) n
>= sizeof(buf
) || (msghdr
.msg_flags
& MSG_TRUNC
)) {
2436 log_warning("Received notify message exceeded maximum size. Ignoring.");
2440 /* As extra safety check, let's make sure the string we get doesn't contain embedded NUL bytes. We permit one
2441 * trailing NUL byte in the message, but don't expect it. */
2442 if (n
> 1 && memchr(buf
, 0, n
-1)) {
2443 log_warning("Received notify message with embedded NUL bytes. Ignoring.");
2447 /* Make sure it's NUL-terminated, then parse it to obtain the tags list */
2449 tags
= strv_split_newlines(buf
);
2455 /* possibly a barrier fd, let's see */
2456 if (manager_process_barrier_fd(tags
, fds
))
2459 /* Increase the generation counter used for filtering out duplicate unit invocations. */
2462 /* Notify every unit that might be interested, which might be multiple. */
2463 u1
= manager_get_unit_by_pid_cgroup(m
, ucred
->pid
);
2464 u2
= hashmap_get(m
->watch_pids
, PID_TO_PTR(ucred
->pid
));
2465 array
= hashmap_get(m
->watch_pids
, PID_TO_PTR(-ucred
->pid
));
2472 array_copy
= newdup(Unit
*, array
, k
+1);
2476 /* And now invoke the per-unit callbacks. Note that manager_invoke_notify_message() will handle duplicate units
2477 * make sure we only invoke each unit's handler once. */
2479 manager_invoke_notify_message(m
, u1
, ucred
, tags
, fds
);
2483 manager_invoke_notify_message(m
, u2
, ucred
, tags
, fds
);
2487 for (size_t i
= 0; array_copy
[i
]; i
++) {
2488 manager_invoke_notify_message(m
, array_copy
[i
], ucred
, tags
, fds
);
2493 log_warning("Cannot find unit for notify message of PID "PID_FMT
", ignoring.", ucred
->pid
);
2495 if (fdset_size(fds
) > 0)
2496 log_warning("Got extra auxiliary fds with notification message, closing them.");
2501 static void manager_invoke_sigchld_event(
2504 const siginfo_t
*si
) {
2510 /* Already invoked the handler of this unit in this iteration? Then don't process this again */
2511 if (u
->sigchldgen
== m
->sigchldgen
)
2513 u
->sigchldgen
= m
->sigchldgen
;
2515 log_unit_debug(u
, "Child "PID_FMT
" belongs to %s.", si
->si_pid
, u
->id
);
2516 unit_unwatch_pid(u
, si
->si_pid
);
2518 if (UNIT_VTABLE(u
)->sigchld_event
)
2519 UNIT_VTABLE(u
)->sigchld_event(u
, si
->si_pid
, si
->si_code
, si
->si_status
);
2522 static int manager_dispatch_sigchld(sd_event_source
*source
, void *userdata
) {
2523 Manager
*m
= userdata
;
2530 /* First we call waitid() for a PID and do not reap the zombie. That way we can still access /proc/$PID for it
2531 * while it is a zombie. */
2533 if (waitid(P_ALL
, 0, &si
, WEXITED
|WNOHANG
|WNOWAIT
) < 0) {
2535 if (errno
!= ECHILD
)
2536 log_error_errno(errno
, "Failed to peek for child with waitid(), ignoring: %m");
2544 if (IN_SET(si
.si_code
, CLD_EXITED
, CLD_KILLED
, CLD_DUMPED
)) {
2545 _cleanup_free_ Unit
**array_copy
= NULL
;
2546 _cleanup_free_
char *name
= NULL
;
2547 Unit
*u1
, *u2
, **array
;
2549 (void) get_process_comm(si
.si_pid
, &name
);
2551 log_debug("Child "PID_FMT
" (%s) died (code=%s, status=%i/%s)",
2552 si
.si_pid
, strna(name
),
2553 sigchld_code_to_string(si
.si_code
),
2555 strna(si
.si_code
== CLD_EXITED
2556 ? exit_status_to_string(si
.si_status
, EXIT_STATUS_FULL
)
2557 : signal_to_string(si
.si_status
)));
2559 /* Increase the generation counter used for filtering out duplicate unit invocations */
2562 /* And now figure out the unit this belongs to, it might be multiple... */
2563 u1
= manager_get_unit_by_pid_cgroup(m
, si
.si_pid
);
2564 u2
= hashmap_get(m
->watch_pids
, PID_TO_PTR(si
.si_pid
));
2565 array
= hashmap_get(m
->watch_pids
, PID_TO_PTR(-si
.si_pid
));
2569 /* Count how many entries the array has */
2573 /* Make a copy of the array so that we don't trip up on the array changing beneath us */
2574 array_copy
= newdup(Unit
*, array
, n
+1);
2579 /* Finally, execute them all. Note that u1, u2 and the array might contain duplicates, but
2580 * that's fine, manager_invoke_sigchld_event() will ensure we only invoke the handlers once for
2581 * each iteration. */
2583 /* We check for oom condition, in case we got SIGCHLD before the oom notification.
2584 * We only do this for the cgroup the PID belonged to. */
2585 (void) unit_check_oom(u1
);
2587 /* This only logs for now. In the future when the interface for kills/notifications
2588 * is more stable we can extend service results table similar to how kernel oom kills
2590 (void) unit_check_oomd_kill(u1
);
2592 manager_invoke_sigchld_event(m
, u1
, &si
);
2595 manager_invoke_sigchld_event(m
, u2
, &si
);
2597 for (size_t i
= 0; array_copy
[i
]; i
++)
2598 manager_invoke_sigchld_event(m
, array_copy
[i
], &si
);
2601 /* And now, we actually reap the zombie. */
2602 if (waitid(P_PID
, si
.si_pid
, &si
, WEXITED
) < 0) {
2603 log_error_errno(errno
, "Failed to dequeue child, ignoring: %m");
2610 /* All children processed for now, turn off event source */
2612 r
= sd_event_source_set_enabled(m
->sigchld_event_source
, SD_EVENT_OFF
);
2614 return log_error_errno(r
, "Failed to disable SIGCHLD event source: %m");
2619 static void manager_start_target(Manager
*m
, const char *name
, JobMode mode
) {
2620 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
2623 log_debug("Activating special unit %s", name
);
2625 r
= manager_add_job_by_name(m
, JOB_START
, name
, mode
, NULL
, &error
, NULL
);
2627 log_error("Failed to enqueue %s job: %s", name
, bus_error_message(&error
, r
));
2630 static void manager_handle_ctrl_alt_del(Manager
*m
) {
2631 /* If the user presses C-A-D more than
2632 * 7 times within 2s, we reboot/shutdown immediately,
2633 * unless it was disabled in system.conf */
2635 if (ratelimit_below(&m
->ctrl_alt_del_ratelimit
) || m
->cad_burst_action
== EMERGENCY_ACTION_NONE
)
2636 manager_start_target(m
, SPECIAL_CTRL_ALT_DEL_TARGET
, JOB_REPLACE_IRREVERSIBLY
);
2638 emergency_action(m
, m
->cad_burst_action
, EMERGENCY_ACTION_WARN
, NULL
, -1,
2639 "Ctrl-Alt-Del was pressed more than 7 times within 2s");
2642 static int manager_dispatch_signal_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
2643 Manager
*m
= userdata
;
2645 struct signalfd_siginfo sfsi
;
2649 assert(m
->signal_fd
== fd
);
2651 if (revents
!= EPOLLIN
) {
2652 log_warning("Got unexpected events from signal file descriptor.");
2656 n
= read(m
->signal_fd
, &sfsi
, sizeof(sfsi
));
2657 if (n
!= sizeof(sfsi
)) {
2659 log_warning("Truncated read from signal fd (%zu bytes), ignoring!", n
);
2663 if (IN_SET(errno
, EINTR
, EAGAIN
))
2666 /* We return an error here, which will kill this handler,
2667 * to avoid a busy loop on read error. */
2668 return log_error_errno(errno
, "Reading from signal fd failed: %m");
2671 log_received_signal(sfsi
.ssi_signo
== SIGCHLD
||
2672 (sfsi
.ssi_signo
== SIGTERM
&& MANAGER_IS_USER(m
))
2673 ? LOG_DEBUG
: LOG_INFO
,
2676 switch (sfsi
.ssi_signo
) {
2679 r
= sd_event_source_set_enabled(m
->sigchld_event_source
, SD_EVENT_ON
);
2681 log_warning_errno(r
, "Failed to enable SIGCHLD event source, ignoring: %m");
2686 if (MANAGER_IS_SYSTEM(m
)) {
2687 /* This is for compatibility with the original sysvinit */
2688 if (verify_run_space_and_log("Refusing to reexecute") < 0)
2691 m
->objective
= MANAGER_REEXECUTE
;
2697 if (MANAGER_IS_SYSTEM(m
))
2698 manager_handle_ctrl_alt_del(m
);
2700 manager_start_target(m
, SPECIAL_EXIT_TARGET
,
2701 JOB_REPLACE_IRREVERSIBLY
);
2705 /* This is a nop on non-init */
2706 if (MANAGER_IS_SYSTEM(m
))
2707 manager_start_target(m
, SPECIAL_KBREQUEST_TARGET
, JOB_REPLACE
);
2712 /* This is a nop on non-init */
2713 if (MANAGER_IS_SYSTEM(m
))
2714 manager_start_target(m
, SPECIAL_SIGPWR_TARGET
, JOB_REPLACE
);
2719 if (manager_dbus_is_running(m
, false)) {
2720 log_info("Trying to reconnect to bus...");
2722 (void) bus_init_api(m
);
2724 if (MANAGER_IS_SYSTEM(m
))
2725 (void) bus_init_system(m
);
2727 log_info("Starting D-Bus service...");
2728 manager_start_target(m
, SPECIAL_DBUS_SERVICE
, JOB_REPLACE
);
2734 _cleanup_free_
char *dump
= NULL
;
2736 r
= manager_get_dump_string(m
, &dump
);
2738 log_warning_errno(errno
, "Failed to acquire manager dump: %m");
2742 log_dump(LOG_INFO
, dump
);
2747 if (verify_run_space_and_log("Refusing to reload") < 0)
2750 m
->objective
= MANAGER_RELOAD
;
2755 /* Starting SIGRTMIN+0 */
2756 static const struct {
2759 } target_table
[] = {
2760 [0] = { SPECIAL_DEFAULT_TARGET
, JOB_ISOLATE
},
2761 [1] = { SPECIAL_RESCUE_TARGET
, JOB_ISOLATE
},
2762 [2] = { SPECIAL_EMERGENCY_TARGET
, JOB_ISOLATE
},
2763 [3] = { SPECIAL_HALT_TARGET
, JOB_REPLACE_IRREVERSIBLY
},
2764 [4] = { SPECIAL_POWEROFF_TARGET
, JOB_REPLACE_IRREVERSIBLY
},
2765 [5] = { SPECIAL_REBOOT_TARGET
, JOB_REPLACE_IRREVERSIBLY
},
2766 [6] = { SPECIAL_KEXEC_TARGET
, JOB_REPLACE_IRREVERSIBLY
},
2769 /* Starting SIGRTMIN+13, so that target halt and system halt are 10 apart */
2770 static const ManagerObjective objective_table
[] = {
2772 [1] = MANAGER_POWEROFF
,
2773 [2] = MANAGER_REBOOT
,
2774 [3] = MANAGER_KEXEC
,
2777 if ((int) sfsi
.ssi_signo
>= SIGRTMIN
+0 &&
2778 (int) sfsi
.ssi_signo
< SIGRTMIN
+(int) ELEMENTSOF(target_table
)) {
2779 int idx
= (int) sfsi
.ssi_signo
- SIGRTMIN
;
2780 manager_start_target(m
, target_table
[idx
].target
,
2781 target_table
[idx
].mode
);
2785 if ((int) sfsi
.ssi_signo
>= SIGRTMIN
+13 &&
2786 (int) sfsi
.ssi_signo
< SIGRTMIN
+13+(int) ELEMENTSOF(objective_table
)) {
2787 m
->objective
= objective_table
[sfsi
.ssi_signo
- SIGRTMIN
- 13];
2791 switch (sfsi
.ssi_signo
- SIGRTMIN
) {
2794 manager_override_show_status(m
, SHOW_STATUS_YES
, "signal");
2798 manager_override_show_status(m
, SHOW_STATUS_NO
, "signal");
2802 manager_override_log_level(m
, LOG_DEBUG
);
2806 manager_restore_original_log_level(m
);
2810 if (MANAGER_IS_USER(m
)) {
2811 m
->objective
= MANAGER_EXIT
;
2815 /* This is a nop on init */
2819 case 29: /* compatibility: used to be mapped to LOG_TARGET_SYSLOG_OR_KMSG */
2820 manager_restore_original_log_target(m
);
2824 manager_override_log_target(m
, LOG_TARGET_CONSOLE
);
2828 manager_override_log_target(m
, LOG_TARGET_KMSG
);
2832 log_warning("Got unhandled signal <%s>.", signal_to_string(sfsi
.ssi_signo
));
2839 static int manager_dispatch_time_change_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
2840 Manager
*m
= userdata
;
2844 assert(m
->time_change_fd
== fd
);
2846 log_struct(LOG_DEBUG
,
2847 "MESSAGE_ID=" SD_MESSAGE_TIME_CHANGE_STR
,
2848 LOG_MESSAGE("Time has been changed"));
2850 /* Restart the watch */
2851 (void) manager_setup_time_change(m
);
2853 HASHMAP_FOREACH(u
, m
->units
)
2854 if (UNIT_VTABLE(u
)->time_change
)
2855 UNIT_VTABLE(u
)->time_change(u
);
2860 static int manager_dispatch_timezone_change(
2861 sd_event_source
*source
,
2862 const struct inotify_event
*e
,
2865 Manager
*m
= userdata
;
2871 log_debug("inotify event for /etc/localtime");
2873 changed
= manager_read_timezone_stat(m
);
2877 /* Something changed, restart the watch, to ensure we watch the new /etc/localtime if it changed */
2878 (void) manager_setup_timezone_change(m
);
2880 /* Read the new timezone */
2883 log_debug("Timezone has been changed (now: %s).", tzname
[daylight
]);
2885 HASHMAP_FOREACH(u
, m
->units
)
2886 if (UNIT_VTABLE(u
)->timezone_change
)
2887 UNIT_VTABLE(u
)->timezone_change(u
);
2892 static int manager_dispatch_idle_pipe_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
2893 Manager
*m
= userdata
;
2896 assert(m
->idle_pipe
[2] == fd
);
2898 /* There's at least one Type=idle child that just gave up on us waiting for the boot process to complete. Let's
2899 * now turn off any further console output if there's at least one service that needs console access, so that
2900 * from now on our own output should not spill into that service's output anymore. After all, we support
2901 * Type=idle only to beautify console output and it generally is set on services that want to own the console
2902 * exclusively without our interference. */
2903 m
->no_console_output
= m
->n_on_console
> 0;
2905 /* Acknowledge the child's request, and let all all other children know too that they shouldn't wait any longer
2906 * by closing the pipes towards them, which is what they are waiting for. */
2907 manager_close_idle_pipe(m
);
2912 static int manager_dispatch_jobs_in_progress(sd_event_source
*source
, usec_t usec
, void *userdata
) {
2913 Manager
*m
= userdata
;
2919 manager_print_jobs_in_progress(m
);
2921 r
= sd_event_source_set_time_relative(source
, JOBS_IN_PROGRESS_PERIOD_USEC
);
2925 return sd_event_source_set_enabled(source
, SD_EVENT_ONESHOT
);
2928 int manager_loop(Manager
*m
) {
2929 RateLimit rl
= { .interval
= 1*USEC_PER_SEC
, .burst
= 50000 };
2933 assert(m
->objective
== MANAGER_OK
); /* Ensure manager_startup() has been called */
2935 manager_check_finished(m
);
2937 /* There might still be some zombies hanging around from before we were exec()'ed. Let's reap them. */
2938 r
= sd_event_source_set_enabled(m
->sigchld_event_source
, SD_EVENT_ON
);
2940 return log_error_errno(r
, "Failed to enable SIGCHLD event source: %m");
2942 while (m
->objective
== MANAGER_OK
) {
2943 usec_t wait_usec
, watchdog_usec
;
2945 watchdog_usec
= manager_get_watchdog(m
, WATCHDOG_RUNTIME
);
2946 if (m
->runtime_watchdog_running
)
2947 (void) watchdog_ping();
2948 else if (timestamp_is_set(watchdog_usec
))
2949 manager_retry_runtime_watchdog(m
);
2951 if (!ratelimit_below(&rl
)) {
2952 /* Yay, something is going seriously wrong, pause a little */
2953 log_warning("Looping too fast. Throttling execution a little.");
2957 if (manager_dispatch_load_queue(m
) > 0)
2960 if (manager_dispatch_gc_job_queue(m
) > 0)
2963 if (manager_dispatch_gc_unit_queue(m
) > 0)
2966 if (manager_dispatch_cleanup_queue(m
) > 0)
2969 if (manager_dispatch_cgroup_realize_queue(m
) > 0)
2972 if (manager_dispatch_start_when_upheld_queue(m
) > 0)
2975 if (manager_dispatch_stop_when_bound_queue(m
) > 0)
2978 if (manager_dispatch_stop_when_unneeded_queue(m
) > 0)
2981 if (manager_dispatch_dbus_queue(m
) > 0)
2984 /* Sleep for watchdog runtime wait time */
2985 if (timestamp_is_set(watchdog_usec
))
2986 wait_usec
= watchdog_runtime_wait();
2988 wait_usec
= USEC_INFINITY
;
2990 r
= sd_event_run(m
->event
, wait_usec
);
2992 return log_error_errno(r
, "Failed to run event loop: %m");
2995 return m
->objective
;
2998 int manager_load_unit_from_dbus_path(Manager
*m
, const char *s
, sd_bus_error
*e
, Unit
**_u
) {
2999 _cleanup_free_
char *n
= NULL
;
3000 sd_id128_t invocation_id
;
3008 r
= unit_name_from_dbus_path(s
, &n
);
3012 /* Permit addressing units by invocation ID: if the passed bus path is suffixed by a 128bit ID then we use it
3013 * as invocation ID. */
3014 r
= sd_id128_from_string(n
, &invocation_id
);
3016 u
= hashmap_get(m
->units_by_invocation_id
, &invocation_id
);
3022 return sd_bus_error_setf(e
, BUS_ERROR_NO_UNIT_FOR_INVOCATION_ID
,
3023 "No unit with the specified invocation ID " SD_ID128_FORMAT_STR
" known.",
3024 SD_ID128_FORMAT_VAL(invocation_id
));
3027 /* If this didn't work, we check if this is a unit name */
3028 if (!unit_name_is_valid(n
, UNIT_NAME_PLAIN
|UNIT_NAME_INSTANCE
)) {
3029 _cleanup_free_
char *nn
= NULL
;
3032 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
,
3033 "Unit name %s is neither a valid invocation ID nor unit name.", strnull(nn
));
3036 r
= manager_load_unit(m
, n
, NULL
, e
, &u
);
3044 int manager_get_job_from_dbus_path(Manager
*m
, const char *s
, Job
**_j
) {
3054 p
= startswith(s
, "/org/freedesktop/systemd1/job/");
3058 r
= safe_atou(p
, &id
);
3062 j
= manager_get_job(m
, id
);
3071 void manager_send_unit_audit(Manager
*m
, Unit
*u
, int type
, bool success
) {
3074 _cleanup_free_
char *p
= NULL
;
3078 if (!MANAGER_IS_SYSTEM(m
))
3081 audit_fd
= get_audit_fd();
3085 /* Don't generate audit events if the service was already
3086 * started and we're just deserializing */
3087 if (MANAGER_IS_RELOADING(m
))
3090 if (u
->type
!= UNIT_SERVICE
)
3093 r
= unit_name_to_prefix_and_instance(u
->id
, &p
);
3095 log_error_errno(r
, "Failed to extract prefix and instance of unit name: %m");
3099 msg
= strjoina("unit=", p
);
3100 if (audit_log_user_comm_message(audit_fd
, type
, msg
, "systemd", NULL
, NULL
, NULL
, success
) < 0) {
3102 /* We aren't allowed to send audit messages?
3103 * Then let's not retry again. */
3106 log_warning_errno(errno
, "Failed to send audit message: %m");
3112 void manager_send_unit_plymouth(Manager
*m
, Unit
*u
) {
3113 static const union sockaddr_union sa
= PLYMOUTH_SOCKET
;
3114 _cleanup_free_
char *message
= NULL
;
3115 _cleanup_close_
int fd
= -1;
3118 /* Don't generate plymouth events if the service was already
3119 * started and we're just deserializing */
3120 if (MANAGER_IS_RELOADING(m
))
3123 if (!MANAGER_IS_SYSTEM(m
))
3126 if (detect_container() > 0)
3129 if (!IN_SET(u
->type
, UNIT_SERVICE
, UNIT_MOUNT
, UNIT_SWAP
))
3132 /* We set SOCK_NONBLOCK here so that we rather drop the
3133 * message then wait for plymouth */
3134 fd
= socket(AF_UNIX
, SOCK_STREAM
|SOCK_CLOEXEC
|SOCK_NONBLOCK
, 0);
3136 log_error_errno(errno
, "socket() failed: %m");
3140 if (connect(fd
, &sa
.sa
, SOCKADDR_UN_LEN(sa
.un
)) < 0) {
3141 if (!IN_SET(errno
, EAGAIN
, ENOENT
) && !ERRNO_IS_DISCONNECT(errno
))
3142 log_error_errno(errno
, "connect() failed: %m");
3146 if (asprintf(&message
, "U\002%c%s%n", (int) (strlen(u
->id
) + 1), u
->id
, &n
) < 0) {
3152 if (write(fd
, message
, n
+ 1) != n
+ 1)
3153 if (!IN_SET(errno
, EAGAIN
, ENOENT
) && !ERRNO_IS_DISCONNECT(errno
))
3154 log_error_errno(errno
, "Failed to write Plymouth message: %m");
3157 int manager_open_serialization(Manager
*m
, FILE **_f
) {
3158 _cleanup_close_
int fd
= -1;
3163 fd
= open_serialization_fd("systemd-state");
3167 f
= take_fdopen(&fd
, "w+");
3175 static bool manager_timestamp_shall_serialize(ManagerTimestamp t
) {
3180 /* The following timestamps only apply to the host system, hence only serialize them there */
3182 MANAGER_TIMESTAMP_USERSPACE
, MANAGER_TIMESTAMP_FINISH
,
3183 MANAGER_TIMESTAMP_SECURITY_START
, MANAGER_TIMESTAMP_SECURITY_FINISH
,
3184 MANAGER_TIMESTAMP_GENERATORS_START
, MANAGER_TIMESTAMP_GENERATORS_FINISH
,
3185 MANAGER_TIMESTAMP_UNITS_LOAD_START
, MANAGER_TIMESTAMP_UNITS_LOAD_FINISH
);
3188 #define DESTROY_IPC_FLAG (UINT32_C(1) << 31)
3190 static void manager_serialize_uid_refs_internal(
3193 const char *field_name
) {
3200 /* Serialize the UID reference table. Or actually, just the IPC destruction flag of it, as
3201 * the actual counter of it is better rebuild after a reload/reexec. */
3203 HASHMAP_FOREACH_KEY(p
, k
, uid_refs
) {
3207 uid
= PTR_TO_UID(k
);
3208 c
= PTR_TO_UINT32(p
);
3210 if (!(c
& DESTROY_IPC_FLAG
))
3213 (void) serialize_item_format(f
, field_name
, UID_FMT
, uid
);
3217 static void manager_serialize_uid_refs(Manager
*m
, FILE *f
) {
3218 manager_serialize_uid_refs_internal(f
, m
->uid_refs
, "destroy-ipc-uid");
3221 static void manager_serialize_gid_refs(Manager
*m
, FILE *f
) {
3222 manager_serialize_uid_refs_internal(f
, m
->gid_refs
, "destroy-ipc-gid");
3225 int manager_serialize(
3229 bool switching_root
) {
3239 _cleanup_(manager_reloading_stopp
) _unused_ Manager
*reloading
= manager_reloading_start(m
);
3241 (void) serialize_item_format(f
, "current-job-id", "%" PRIu32
, m
->current_job_id
);
3242 (void) serialize_item_format(f
, "n-installed-jobs", "%u", m
->n_installed_jobs
);
3243 (void) serialize_item_format(f
, "n-failed-jobs", "%u", m
->n_failed_jobs
);
3244 (void) serialize_bool(f
, "taint-usr", m
->taint_usr
);
3245 (void) serialize_bool(f
, "ready-sent", m
->ready_sent
);
3246 (void) serialize_bool(f
, "taint-logged", m
->taint_logged
);
3247 (void) serialize_bool(f
, "service-watchdogs", m
->service_watchdogs
);
3249 /* After switching root, udevd has not been started yet. So, enumeration results should not be emitted. */
3250 (void) serialize_bool(f
, "honor-device-enumeration", !switching_root
);
3252 if (m
->show_status_overridden
!= _SHOW_STATUS_INVALID
)
3253 (void) serialize_item(f
, "show-status-overridden",
3254 show_status_to_string(m
->show_status_overridden
));
3256 if (m
->log_level_overridden
)
3257 (void) serialize_item_format(f
, "log-level-override", "%i", log_get_max_level());
3258 if (m
->log_target_overridden
)
3259 (void) serialize_item(f
, "log-target-override", log_target_to_string(log_get_target()));
3261 (void) serialize_usec(f
, "runtime-watchdog-overridden", m
->watchdog_overridden
[WATCHDOG_RUNTIME
]);
3262 (void) serialize_usec(f
, "reboot-watchdog-overridden", m
->watchdog_overridden
[WATCHDOG_REBOOT
]);
3263 (void) serialize_usec(f
, "kexec-watchdog-overridden", m
->watchdog_overridden
[WATCHDOG_KEXEC
]);
3265 for (ManagerTimestamp q
= 0; q
< _MANAGER_TIMESTAMP_MAX
; q
++) {
3266 _cleanup_free_
char *joined
= NULL
;
3268 if (!manager_timestamp_shall_serialize(q
))
3271 joined
= strjoin(manager_timestamp_to_string(q
), "-timestamp");
3275 (void) serialize_dual_timestamp(f
, joined
, m
->timestamps
+ q
);
3278 if (!switching_root
)
3279 (void) serialize_strv(f
, "env", m
->client_environment
);
3281 if (m
->notify_fd
>= 0) {
3282 r
= serialize_fd(f
, fds
, "notify-fd", m
->notify_fd
);
3286 (void) serialize_item(f
, "notify-socket", m
->notify_socket
);
3289 if (m
->cgroups_agent_fd
>= 0) {
3290 r
= serialize_fd(f
, fds
, "cgroups-agent-fd", m
->cgroups_agent_fd
);
3295 if (m
->user_lookup_fds
[0] >= 0) {
3298 copy0
= fdset_put_dup(fds
, m
->user_lookup_fds
[0]);
3300 return log_error_errno(copy0
, "Failed to add user lookup fd to serialization: %m");
3302 copy1
= fdset_put_dup(fds
, m
->user_lookup_fds
[1]);
3304 return log_error_errno(copy1
, "Failed to add user lookup fd to serialization: %m");
3306 (void) serialize_item_format(f
, "user-lookup", "%i %i", copy0
, copy1
);
3309 bus_track_serialize(m
->subscribed
, f
, "subscribed");
3311 r
= dynamic_user_serialize(m
, f
, fds
);
3315 manager_serialize_uid_refs(m
, f
);
3316 manager_serialize_gid_refs(m
, f
);
3318 r
= exec_runtime_serialize(m
, f
, fds
);
3322 (void) fputc('\n', f
);
3324 HASHMAP_FOREACH_KEY(u
, t
, m
->units
) {
3328 r
= unit_serialize(u
, f
, fds
, switching_root
);
3333 r
= fflush_and_check(f
);
3335 return log_error_errno(r
, "Failed to flush serialization: %m");
3337 r
= bus_fdset_add_all(m
, fds
);
3339 return log_error_errno(r
, "Failed to add bus sockets to serialization: %m");
3344 static int manager_deserialize_one_unit(Manager
*m
, const char *name
, FILE *f
, FDSet
*fds
) {
3348 r
= manager_load_unit(m
, name
, NULL
, NULL
, &u
);
3352 return log_notice_errno(r
, "Failed to load unit \"%s\", skipping deserialization: %m", name
);
3355 r
= unit_deserialize(u
, f
, fds
);
3359 return log_notice_errno(r
, "Failed to deserialize unit \"%s\", skipping: %m", name
);
3365 static int manager_deserialize_units(Manager
*m
, FILE *f
, FDSet
*fds
) {
3366 const char *unit_name
;
3370 _cleanup_free_
char *line
= NULL
;
3372 r
= read_line(f
, LONG_LINE_MAX
, &line
);
3374 return log_error_errno(r
, "Failed to read serialization line: %m");
3378 unit_name
= strstrip(line
);
3380 r
= manager_deserialize_one_unit(m
, unit_name
, f
, fds
);
3384 r
= unit_deserialize_skip(f
);
3393 usec_t
manager_get_watchdog(Manager
*m
, WatchdogType t
) {
3396 if (MANAGER_IS_USER(m
))
3397 return USEC_INFINITY
;
3399 if (timestamp_is_set(m
->watchdog_overridden
[t
]))
3400 return m
->watchdog_overridden
[t
];
3402 return m
->watchdog
[t
];
3405 void manager_set_watchdog(Manager
*m
, WatchdogType t
, usec_t timeout
) {
3410 if (MANAGER_IS_USER(m
))
3413 if (m
->watchdog
[t
] == timeout
)
3416 if (t
== WATCHDOG_RUNTIME
)
3417 if (!timestamp_is_set(m
->watchdog_overridden
[WATCHDOG_RUNTIME
])) {
3418 if (timestamp_is_set(timeout
)) {
3419 r
= watchdog_set_timeout(&timeout
);
3422 m
->runtime_watchdog_running
= true;
3424 watchdog_close(true);
3425 m
->runtime_watchdog_running
= false;
3429 m
->watchdog
[t
] = timeout
;
3432 int manager_override_watchdog(Manager
*m
, WatchdogType t
, usec_t timeout
) {
3437 if (MANAGER_IS_USER(m
))
3440 if (m
->watchdog_overridden
[t
] == timeout
)
3443 if (t
== WATCHDOG_RUNTIME
) {
3446 p
= timestamp_is_set(timeout
) ? &timeout
: &m
->watchdog
[t
];
3447 if (timestamp_is_set(*p
)) {
3448 r
= watchdog_set_timeout(p
);
3451 m
->runtime_watchdog_running
= true;
3453 watchdog_close(true);
3454 m
->runtime_watchdog_running
= false;
3458 m
->watchdog_overridden
[t
] = timeout
;
3463 void manager_retry_runtime_watchdog(Manager
*m
) {
3468 if (timestamp_is_set(m
->watchdog_overridden
[WATCHDOG_RUNTIME
]))
3469 r
= watchdog_set_timeout(&m
->watchdog_overridden
[WATCHDOG_RUNTIME
]);
3471 r
= watchdog_set_timeout(&m
->watchdog
[WATCHDOG_RUNTIME
]);
3474 m
->runtime_watchdog_running
= true;
3477 static void manager_deserialize_uid_refs_one_internal(
3479 const char *value
) {
3488 r
= parse_uid(value
, &uid
);
3489 if (r
< 0 || uid
== 0) {
3490 log_debug("Unable to parse UID/GID reference serialization: " UID_FMT
, uid
);
3494 if (hashmap_ensure_allocated(uid_refs
, &trivial_hash_ops
) < 0) {
3499 c
= PTR_TO_UINT32(hashmap_get(*uid_refs
, UID_TO_PTR(uid
)));
3500 if (c
& DESTROY_IPC_FLAG
)
3503 c
|= DESTROY_IPC_FLAG
;
3505 r
= hashmap_replace(*uid_refs
, UID_TO_PTR(uid
), UINT32_TO_PTR(c
));
3507 log_debug_errno(r
, "Failed to add UID/GID reference entry: %m");
3512 static void manager_deserialize_uid_refs_one(Manager
*m
, const char *value
) {
3513 manager_deserialize_uid_refs_one_internal(&m
->uid_refs
, value
);
3516 static void manager_deserialize_gid_refs_one(Manager
*m
, const char *value
) {
3517 manager_deserialize_uid_refs_one_internal(&m
->gid_refs
, value
);
3520 int manager_deserialize(Manager
*m
, FILE *f
, FDSet
*fds
) {
3526 if (DEBUG_LOGGING
) {
3527 if (fdset_isempty(fds
))
3528 log_debug("No file descriptors passed");
3532 FDSET_FOREACH(fd
, fds
) {
3533 _cleanup_free_
char *fn
= NULL
;
3535 r
= fd_get_path(fd
, &fn
);
3537 log_debug_errno(r
, "Received serialized fd %i → %m", fd
);
3539 log_debug("Received serialized fd %i → %s", fd
, strna(fn
));
3544 log_debug("Deserializing state...");
3546 /* If we are not in reload mode yet, enter it now. Not that this is recursive, a caller might already have
3547 * increased it to non-zero, which is why we just increase it by one here and down again at the end of this
3549 _cleanup_(manager_reloading_stopp
) _unused_ Manager
*reloading
= manager_reloading_start(m
);
3552 _cleanup_free_
char *line
= NULL
;
3553 const char *val
, *l
;
3555 r
= read_line(f
, LONG_LINE_MAX
, &line
);
3557 return log_error_errno(r
, "Failed to read serialization line: %m");
3562 if (isempty(l
)) /* end marker */
3565 if ((val
= startswith(l
, "current-job-id="))) {
3568 if (safe_atou32(val
, &id
) < 0)
3569 log_notice("Failed to parse current job id value '%s', ignoring.", val
);
3571 m
->current_job_id
= MAX(m
->current_job_id
, id
);
3573 } else if ((val
= startswith(l
, "n-installed-jobs="))) {
3576 if (safe_atou32(val
, &n
) < 0)
3577 log_notice("Failed to parse installed jobs counter '%s', ignoring.", val
);
3579 m
->n_installed_jobs
+= n
;
3581 } else if ((val
= startswith(l
, "n-failed-jobs="))) {
3584 if (safe_atou32(val
, &n
) < 0)
3585 log_notice("Failed to parse failed jobs counter '%s', ignoring.", val
);
3587 m
->n_failed_jobs
+= n
;
3589 } else if ((val
= startswith(l
, "taint-usr="))) {
3592 b
= parse_boolean(val
);
3594 log_notice("Failed to parse taint /usr flag '%s', ignoring.", val
);
3596 m
->taint_usr
= m
->taint_usr
|| b
;
3598 } else if ((val
= startswith(l
, "ready-sent="))) {
3601 b
= parse_boolean(val
);
3603 log_notice("Failed to parse ready-sent flag '%s', ignoring.", val
);
3605 m
->ready_sent
= m
->ready_sent
|| b
;
3607 } else if ((val
= startswith(l
, "taint-logged="))) {
3610 b
= parse_boolean(val
);
3612 log_notice("Failed to parse taint-logged flag '%s', ignoring.", val
);
3614 m
->taint_logged
= m
->taint_logged
|| b
;
3616 } else if ((val
= startswith(l
, "service-watchdogs="))) {
3619 b
= parse_boolean(val
);
3621 log_notice("Failed to parse service-watchdogs flag '%s', ignoring.", val
);
3623 m
->service_watchdogs
= b
;
3625 } else if ((val
= startswith(l
, "honor-device-enumeration="))) {
3628 b
= parse_boolean(val
);
3630 log_notice("Failed to parse honor-device-enumeration flag '%s', ignoring.", val
);
3632 m
->honor_device_enumeration
= b
;
3634 } else if ((val
= startswith(l
, "show-status-overridden="))) {
3637 s
= show_status_from_string(val
);
3639 log_notice("Failed to parse show-status-overridden flag '%s', ignoring.", val
);
3641 manager_override_show_status(m
, s
, "deserialize");
3643 } else if ((val
= startswith(l
, "log-level-override="))) {
3646 level
= log_level_from_string(val
);
3648 log_notice("Failed to parse log-level-override value '%s', ignoring.", val
);
3650 manager_override_log_level(m
, level
);
3652 } else if ((val
= startswith(l
, "log-target-override="))) {
3655 target
= log_target_from_string(val
);
3657 log_notice("Failed to parse log-target-override value '%s', ignoring.", val
);
3659 manager_override_log_target(m
, target
);
3661 } else if ((val
= startswith(l
, "runtime-watchdog-overridden="))) {
3664 if (deserialize_usec(val
, &t
) < 0)
3665 log_notice("Failed to parse runtime-watchdog-overridden value '%s', ignoring.", val
);
3667 manager_override_watchdog(m
, WATCHDOG_RUNTIME
, t
);
3669 } else if ((val
= startswith(l
, "reboot-watchdog-overridden="))) {
3672 if (deserialize_usec(val
, &t
) < 0)
3673 log_notice("Failed to parse reboot-watchdog-overridden value '%s', ignoring.", val
);
3675 manager_override_watchdog(m
, WATCHDOG_REBOOT
, t
);
3677 } else if ((val
= startswith(l
, "kexec-watchdog-overridden="))) {
3680 if (deserialize_usec(val
, &t
) < 0)
3681 log_notice("Failed to parse kexec-watchdog-overridden value '%s', ignoring.", val
);
3683 manager_override_watchdog(m
, WATCHDOG_KEXEC
, t
);
3685 } else if (startswith(l
, "env=")) {
3686 r
= deserialize_environment(l
+ 4, &m
->client_environment
);
3688 log_notice_errno(r
, "Failed to parse environment entry: \"%s\", ignoring: %m", l
);
3690 } else if ((val
= startswith(l
, "notify-fd="))) {
3693 if (safe_atoi(val
, &fd
) < 0 || fd
< 0 || !fdset_contains(fds
, fd
))
3694 log_notice("Failed to parse notify fd, ignoring: \"%s\"", val
);
3696 m
->notify_event_source
= sd_event_source_disable_unref(m
->notify_event_source
);
3697 safe_close(m
->notify_fd
);
3698 m
->notify_fd
= fdset_remove(fds
, fd
);
3701 } else if ((val
= startswith(l
, "notify-socket="))) {
3702 r
= free_and_strdup(&m
->notify_socket
, val
);
3706 } else if ((val
= startswith(l
, "cgroups-agent-fd="))) {
3709 if (safe_atoi(val
, &fd
) < 0 || fd
< 0 || !fdset_contains(fds
, fd
))
3710 log_notice("Failed to parse cgroups agent fd, ignoring.: %s", val
);
3712 m
->cgroups_agent_event_source
= sd_event_source_disable_unref(m
->cgroups_agent_event_source
);
3713 safe_close(m
->cgroups_agent_fd
);
3714 m
->cgroups_agent_fd
= fdset_remove(fds
, fd
);
3717 } else if ((val
= startswith(l
, "user-lookup="))) {
3720 if (sscanf(val
, "%i %i", &fd0
, &fd1
) != 2 || fd0
< 0 || fd1
< 0 || fd0
== fd1
|| !fdset_contains(fds
, fd0
) || !fdset_contains(fds
, fd1
))
3721 log_notice("Failed to parse user lookup fd, ignoring: %s", val
);
3723 m
->user_lookup_event_source
= sd_event_source_disable_unref(m
->user_lookup_event_source
);
3724 safe_close_pair(m
->user_lookup_fds
);
3725 m
->user_lookup_fds
[0] = fdset_remove(fds
, fd0
);
3726 m
->user_lookup_fds
[1] = fdset_remove(fds
, fd1
);
3729 } else if ((val
= startswith(l
, "dynamic-user=")))
3730 dynamic_user_deserialize_one(m
, val
, fds
);
3731 else if ((val
= startswith(l
, "destroy-ipc-uid=")))
3732 manager_deserialize_uid_refs_one(m
, val
);
3733 else if ((val
= startswith(l
, "destroy-ipc-gid=")))
3734 manager_deserialize_gid_refs_one(m
, val
);
3735 else if ((val
= startswith(l
, "exec-runtime=")))
3736 (void) exec_runtime_deserialize_one(m
, val
, fds
);
3737 else if ((val
= startswith(l
, "subscribed="))) {
3739 if (strv_extend(&m
->deserialized_subscribed
, val
) < 0)
3745 for (q
= 0; q
< _MANAGER_TIMESTAMP_MAX
; q
++) {
3746 val
= startswith(l
, manager_timestamp_to_string(q
));
3750 val
= startswith(val
, "-timestamp=");
3755 if (q
< _MANAGER_TIMESTAMP_MAX
) /* found it */
3756 (void) deserialize_dual_timestamp(val
, m
->timestamps
+ q
);
3757 else if (!startswith(l
, "kdbus-fd=")) /* ignore kdbus */
3758 log_notice("Unknown serialization item '%s', ignoring.", l
);
3762 return manager_deserialize_units(m
, f
, fds
);
3765 int manager_reload(Manager
*m
) {
3766 _cleanup_(manager_reloading_stopp
) Manager
*reloading
= NULL
;
3767 _cleanup_fdset_free_ FDSet
*fds
= NULL
;
3768 _cleanup_fclose_
FILE *f
= NULL
;
3773 r
= manager_open_serialization(m
, &f
);
3775 return log_error_errno(r
, "Failed to create serialization file: %m");
3781 /* We are officially in reload mode from here on. */
3782 reloading
= manager_reloading_start(m
);
3784 r
= manager_serialize(m
, f
, fds
, false);
3788 if (fseeko(f
, 0, SEEK_SET
) < 0)
3789 return log_error_errno(errno
, "Failed to seek to beginning of serialization: %m");
3791 /* 💀 This is the point of no return, from here on there is no way back. 💀 */
3794 bus_manager_send_reloading(m
, true);
3796 /* Start by flushing out all jobs and units, all generated units, all runtime environments, all dynamic users
3797 * and everything else that is worth flushing out. We'll get it all back from the serialization — if we need
3800 manager_clear_jobs_and_units(m
);
3801 lookup_paths_flush_generator(&m
->lookup_paths
);
3802 lookup_paths_free(&m
->lookup_paths
);
3803 exec_runtime_vacuum(m
);
3804 dynamic_user_vacuum(m
, false);
3805 m
->uid_refs
= hashmap_free(m
->uid_refs
);
3806 m
->gid_refs
= hashmap_free(m
->gid_refs
);
3808 r
= lookup_paths_init(&m
->lookup_paths
, m
->unit_file_scope
, 0, NULL
);
3810 log_warning_errno(r
, "Failed to initialize path lookup table, ignoring: %m");
3812 (void) manager_run_environment_generators(m
);
3813 (void) manager_run_generators(m
);
3815 lookup_paths_log(&m
->lookup_paths
);
3817 /* We flushed out generated files, for which we don't watch mtime, so we should flush the old map. */
3818 manager_free_unit_name_maps(m
);
3820 /* First, enumerate what we can from kernel and suchlike */
3821 manager_enumerate_perpetual(m
);
3822 manager_enumerate(m
);
3824 /* Second, deserialize our stored data */
3825 r
= manager_deserialize(m
, f
, fds
);
3827 log_warning_errno(r
, "Deserialization failed, proceeding anyway: %m");
3829 /* We don't need the serialization anymore */
3832 /* Re-register notify_fd as event source, and set up other sockets/communication channels we might need */
3833 (void) manager_setup_notify(m
);
3834 (void) manager_setup_cgroups_agent(m
);
3835 (void) manager_setup_user_lookup_fd(m
);
3837 /* Third, fire things up! */
3838 manager_coldplug(m
);
3840 /* Clean up runtime objects no longer referenced */
3843 /* Clean up deserialized tracked clients */
3844 m
->deserialized_subscribed
= strv_free(m
->deserialized_subscribed
);
3846 /* Consider the reload process complete now. */
3847 assert(m
->n_reloading
> 0);
3850 /* On manager reloading, device tag data should exists, thus, we should honor the results of device
3851 * enumeration. The flag should be always set correctly by the serialized data, but it may fail. So,
3852 * let's always set the flag here for safety. */
3853 m
->honor_device_enumeration
= true;
3857 m
->send_reloading_done
= true;
3861 void manager_reset_failed(Manager
*m
) {
3866 HASHMAP_FOREACH(u
, m
->units
)
3867 unit_reset_failed(u
);
3870 bool manager_unit_inactive_or_pending(Manager
*m
, const char *name
) {
3876 /* Returns true if the unit is inactive or going down */
3877 u
= manager_get_unit(m
, name
);
3881 return unit_inactive_or_pending(u
);
3884 static void log_taint_string(Manager
*m
) {
3885 _cleanup_free_
char *taint
= NULL
;
3889 if (MANAGER_IS_USER(m
) || m
->taint_logged
)
3892 m
->taint_logged
= true; /* only check for taint once */
3894 taint
= manager_taint_string(m
);
3898 log_struct(LOG_NOTICE
,
3899 LOG_MESSAGE("System is tainted: %s", taint
),
3901 "MESSAGE_ID=" SD_MESSAGE_TAINTED_STR
);
3904 static void manager_notify_finished(Manager
*m
) {
3905 usec_t firmware_usec
, loader_usec
, kernel_usec
, initrd_usec
, userspace_usec
, total_usec
;
3907 if (MANAGER_IS_TEST_RUN(m
))
3910 if (MANAGER_IS_SYSTEM(m
) && detect_container() <= 0) {
3911 char buf
[FORMAT_TIMESPAN_MAX
+ STRLEN(" (firmware) + ") + FORMAT_TIMESPAN_MAX
+ STRLEN(" (loader) + ")]
3914 size_t size
= sizeof buf
;
3916 /* Note that MANAGER_TIMESTAMP_KERNEL's monotonic value is always at 0, and
3917 * MANAGER_TIMESTAMP_FIRMWARE's and MANAGER_TIMESTAMP_LOADER's monotonic value should be considered
3918 * negative values. */
3920 firmware_usec
= m
->timestamps
[MANAGER_TIMESTAMP_FIRMWARE
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_LOADER
].monotonic
;
3921 loader_usec
= m
->timestamps
[MANAGER_TIMESTAMP_LOADER
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_KERNEL
].monotonic
;
3922 userspace_usec
= m
->timestamps
[MANAGER_TIMESTAMP_FINISH
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_USERSPACE
].monotonic
;
3923 total_usec
= m
->timestamps
[MANAGER_TIMESTAMP_FIRMWARE
].monotonic
+ m
->timestamps
[MANAGER_TIMESTAMP_FINISH
].monotonic
;
3925 if (firmware_usec
> 0)
3926 size
= strpcpyf(&p
, size
, "%s (firmware) + ", FORMAT_TIMESPAN(firmware_usec
, USEC_PER_MSEC
));
3927 if (loader_usec
> 0)
3928 size
= strpcpyf(&p
, size
, "%s (loader) + ", FORMAT_TIMESPAN(loader_usec
, USEC_PER_MSEC
));
3930 if (dual_timestamp_is_set(&m
->timestamps
[MANAGER_TIMESTAMP_INITRD
])) {
3932 /* The initrd case on bare-metal */
3933 kernel_usec
= m
->timestamps
[MANAGER_TIMESTAMP_INITRD
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_KERNEL
].monotonic
;
3934 initrd_usec
= m
->timestamps
[MANAGER_TIMESTAMP_USERSPACE
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_INITRD
].monotonic
;
3936 log_struct(LOG_INFO
,
3937 "MESSAGE_ID=" SD_MESSAGE_STARTUP_FINISHED_STR
,
3938 "KERNEL_USEC="USEC_FMT
, kernel_usec
,
3939 "INITRD_USEC="USEC_FMT
, initrd_usec
,
3940 "USERSPACE_USEC="USEC_FMT
, userspace_usec
,
3941 LOG_MESSAGE("Startup finished in %s%s (kernel) + %s (initrd) + %s (userspace) = %s.",
3943 FORMAT_TIMESPAN(kernel_usec
, USEC_PER_MSEC
),
3944 FORMAT_TIMESPAN(initrd_usec
, USEC_PER_MSEC
),
3945 FORMAT_TIMESPAN(userspace_usec
, USEC_PER_MSEC
),
3946 FORMAT_TIMESPAN(total_usec
, USEC_PER_MSEC
)));
3948 /* The initrd-less case on bare-metal */
3950 kernel_usec
= m
->timestamps
[MANAGER_TIMESTAMP_USERSPACE
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_KERNEL
].monotonic
;
3953 log_struct(LOG_INFO
,
3954 "MESSAGE_ID=" SD_MESSAGE_STARTUP_FINISHED_STR
,
3955 "KERNEL_USEC="USEC_FMT
, kernel_usec
,
3956 "USERSPACE_USEC="USEC_FMT
, userspace_usec
,
3957 LOG_MESSAGE("Startup finished in %s%s (kernel) + %s (userspace) = %s.",
3959 FORMAT_TIMESPAN(kernel_usec
, USEC_PER_MSEC
),
3960 FORMAT_TIMESPAN(userspace_usec
, USEC_PER_MSEC
),
3961 FORMAT_TIMESPAN(total_usec
, USEC_PER_MSEC
)));
3964 /* The container and --user case */
3965 firmware_usec
= loader_usec
= initrd_usec
= kernel_usec
= 0;
3966 total_usec
= userspace_usec
= m
->timestamps
[MANAGER_TIMESTAMP_FINISH
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_USERSPACE
].monotonic
;
3968 log_struct(LOG_INFO
,
3969 "MESSAGE_ID=" SD_MESSAGE_USER_STARTUP_FINISHED_STR
,
3970 "USERSPACE_USEC="USEC_FMT
, userspace_usec
,
3971 LOG_MESSAGE("Startup finished in %s.",
3972 FORMAT_TIMESPAN(total_usec
, USEC_PER_MSEC
)));
3975 bus_manager_send_finished(m
, firmware_usec
, loader_usec
, kernel_usec
, initrd_usec
, userspace_usec
, total_usec
);
3978 m
->ready_sent
? "STATUS=Startup finished in %s."
3980 "STATUS=Startup finished in %s.",
3981 FORMAT_TIMESPAN(total_usec
, USEC_PER_MSEC
));
3982 m
->ready_sent
= true;
3984 log_taint_string(m
);
3987 static void manager_send_ready(Manager
*m
) {
3990 /* We send READY=1 on reaching basic.target only when running in --user mode. */
3991 if (!MANAGER_IS_USER(m
) || m
->ready_sent
)
3994 m
->ready_sent
= true;
3998 "STATUS=Reached " SPECIAL_BASIC_TARGET
".");
4001 static void manager_check_basic_target(Manager
*m
) {
4006 /* Small shortcut */
4007 if (m
->ready_sent
&& m
->taint_logged
)
4010 u
= manager_get_unit(m
, SPECIAL_BASIC_TARGET
);
4011 if (!u
|| !UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u
)))
4014 /* For user managers, send out READY=1 as soon as we reach basic.target */
4015 manager_send_ready(m
);
4017 /* Log the taint string as soon as we reach basic.target */
4018 log_taint_string(m
);
4021 void manager_check_finished(Manager
*m
) {
4024 if (MANAGER_IS_RELOADING(m
))
4027 /* Verify that we have entered the event loop already, and not left it again. */
4028 if (!MANAGER_IS_RUNNING(m
))
4031 manager_check_basic_target(m
);
4033 if (hashmap_size(m
->jobs
) > 0) {
4034 if (m
->jobs_in_progress_event_source
)
4035 /* Ignore any failure, this is only for feedback */
4036 (void) sd_event_source_set_time(m
->jobs_in_progress_event_source
,
4037 manager_watch_jobs_next_time(m
));
4041 /* The jobs hashmap tends to grow a lot during boot, and then it's not reused until shutdown. Let's
4042 kill the hashmap if it is relatively large. */
4043 if (hashmap_buckets(m
->jobs
) > hashmap_size(m
->units
) / 10)
4044 m
->jobs
= hashmap_free(m
->jobs
);
4046 manager_flip_auto_status(m
, false, "boot finished");
4048 /* Notify Type=idle units that we are done now */
4049 manager_close_idle_pipe(m
);
4051 /* Turn off confirm spawn now */
4052 m
->confirm_spawn
= NULL
;
4054 /* No need to update ask password status when we're going non-interactive */
4055 manager_close_ask_password(m
);
4057 /* This is no longer the first boot */
4058 manager_set_first_boot(m
, false);
4060 if (MANAGER_IS_FINISHED(m
))
4063 dual_timestamp_get(m
->timestamps
+ MANAGER_TIMESTAMP_FINISH
);
4065 manager_notify_finished(m
);
4067 manager_invalidate_startup_units(m
);
4070 static bool generator_path_any(const char* const* paths
) {
4074 /* Optimize by skipping the whole process by not creating output directories
4075 * if no generators are found. */
4076 STRV_FOREACH(path
, (char**) paths
)
4077 if (access(*path
, F_OK
) == 0)
4079 else if (errno
!= ENOENT
)
4080 log_warning_errno(errno
, "Failed to open generator directory %s: %m", *path
);
4085 static int manager_run_environment_generators(Manager
*m
) {
4086 char **tmp
= NULL
; /* this is only used in the forked process, no cleanup here */
4087 _cleanup_strv_free_
char **paths
= NULL
;
4089 [STDOUT_GENERATE
] = &tmp
,
4090 [STDOUT_COLLECT
] = &tmp
,
4091 [STDOUT_CONSUME
] = &m
->transient_environment
,
4095 if (MANAGER_IS_TEST_RUN(m
) && !(m
->test_run_flags
& MANAGER_TEST_RUN_ENV_GENERATORS
))
4098 paths
= env_generator_binary_paths(MANAGER_IS_SYSTEM(m
));
4102 if (!generator_path_any((const char* const*) paths
))
4105 RUN_WITH_UMASK(0022)
4106 r
= execute_directories((const char* const*) paths
, DEFAULT_TIMEOUT_USEC
, gather_environment
,
4107 args
, NULL
, m
->transient_environment
,
4108 EXEC_DIR_PARALLEL
| EXEC_DIR_IGNORE_ERRORS
| EXEC_DIR_SET_SYSTEMD_EXEC_PID
);
4112 static int manager_run_generators(Manager
*m
) {
4113 _cleanup_strv_free_
char **paths
= NULL
;
4114 const char *argv
[5];
4119 if (MANAGER_IS_TEST_RUN(m
) && !(m
->test_run_flags
& MANAGER_TEST_RUN_GENERATORS
))
4122 paths
= generator_binary_paths(m
->unit_file_scope
);
4126 if (!generator_path_any((const char* const*) paths
))
4129 r
= lookup_paths_mkdir_generator(&m
->lookup_paths
);
4131 log_error_errno(r
, "Failed to create generator directories: %m");
4135 argv
[0] = NULL
; /* Leave this empty, execute_directory() will fill something in */
4136 argv
[1] = m
->lookup_paths
.generator
;
4137 argv
[2] = m
->lookup_paths
.generator_early
;
4138 argv
[3] = m
->lookup_paths
.generator_late
;
4141 RUN_WITH_UMASK(0022)
4142 (void) execute_directories((const char* const*) paths
, DEFAULT_TIMEOUT_USEC
, NULL
, NULL
,
4143 (char**) argv
, m
->transient_environment
,
4144 EXEC_DIR_PARALLEL
| EXEC_DIR_IGNORE_ERRORS
| EXEC_DIR_SET_SYSTEMD_EXEC_PID
);
4149 lookup_paths_trim_generator(&m
->lookup_paths
);
4153 int manager_transient_environment_add(Manager
*m
, char **plus
) {
4158 if (strv_isempty(plus
))
4161 a
= strv_env_merge(2, m
->transient_environment
, plus
);
4165 sanitize_environment(a
);
4167 return strv_free_and_replace(m
->transient_environment
, a
);
4170 int manager_client_environment_modify(
4175 char **a
= NULL
, **b
= NULL
, **l
;
4179 if (strv_isempty(minus
) && strv_isempty(plus
))
4182 l
= m
->client_environment
;
4184 if (!strv_isempty(minus
)) {
4185 a
= strv_env_delete(l
, 1, minus
);
4192 if (!strv_isempty(plus
)) {
4193 b
= strv_env_merge(2, l
, plus
);
4202 if (m
->client_environment
!= l
)
4203 strv_free(m
->client_environment
);
4210 m
->client_environment
= sanitize_environment(l
);
4214 int manager_get_effective_environment(Manager
*m
, char ***ret
) {
4220 l
= strv_env_merge(2, m
->transient_environment
, m
->client_environment
);
4228 int manager_set_default_rlimits(Manager
*m
, struct rlimit
**default_rlimit
) {
4231 for (unsigned i
= 0; i
< _RLIMIT_MAX
; i
++) {
4232 m
->rlimit
[i
] = mfree(m
->rlimit
[i
]);
4234 if (!default_rlimit
[i
])
4237 m
->rlimit
[i
] = newdup(struct rlimit
, default_rlimit
[i
], 1);
4245 void manager_recheck_dbus(Manager
*m
) {
4248 /* Connects to the bus if the dbus service and socket are running. If we are running in user mode this is all
4249 * it does. In system mode we'll also connect to the system bus (which will most likely just reuse the
4250 * connection of the API bus). That's because the system bus after all runs as service of the system instance,
4251 * while in the user instance we can assume it's already there. */
4253 if (MANAGER_IS_RELOADING(m
))
4254 return; /* don't check while we are reloading… */
4256 if (manager_dbus_is_running(m
, false)) {
4257 (void) bus_init_api(m
);
4259 if (MANAGER_IS_SYSTEM(m
))
4260 (void) bus_init_system(m
);
4262 (void) bus_done_api(m
);
4264 if (MANAGER_IS_SYSTEM(m
))
4265 (void) bus_done_system(m
);
4269 static bool manager_journal_is_running(Manager
*m
) {
4274 if (MANAGER_IS_TEST_RUN(m
))
4277 /* If we are the user manager we can safely assume that the journal is up */
4278 if (!MANAGER_IS_SYSTEM(m
))
4281 /* Check that the socket is not only up, but in RUNNING state */
4282 u
= manager_get_unit(m
, SPECIAL_JOURNALD_SOCKET
);
4285 if (SOCKET(u
)->state
!= SOCKET_RUNNING
)
4288 /* Similar, check if the daemon itself is fully up, too */
4289 u
= manager_get_unit(m
, SPECIAL_JOURNALD_SERVICE
);
4292 if (!IN_SET(SERVICE(u
)->state
, SERVICE_RELOAD
, SERVICE_RUNNING
))
4298 void disable_printk_ratelimit(void) {
4299 /* Disable kernel's printk ratelimit.
4301 * Logging to /dev/kmsg is most useful during early boot and shutdown, where normal logging
4302 * mechanisms are not available. The semantics of this sysctl are such that any kernel command-line
4303 * setting takes precedence. */
4306 r
= sysctl_write("kernel/printk_devkmsg", "on");
4308 log_debug_errno(r
, "Failed to set sysctl kernel.printk_devkmsg=on: %m");
4311 void manager_recheck_journal(Manager
*m
) {
4315 /* Don't bother with this unless we are in the special situation of being PID 1 */
4316 if (getpid_cached() != 1)
4319 /* Don't check this while we are reloading, things might still change */
4320 if (MANAGER_IS_RELOADING(m
))
4323 /* The journal is fully and entirely up? If so, let's permit logging to it, if that's configured. If the
4324 * journal is down, don't ever log to it, otherwise we might end up deadlocking ourselves as we might trigger
4325 * an activation ourselves we can't fulfill. */
4326 log_set_prohibit_ipc(!manager_journal_is_running(m
));
4330 static ShowStatus
manager_get_show_status(Manager
*m
) {
4333 if (MANAGER_IS_USER(m
))
4334 return _SHOW_STATUS_INVALID
;
4336 if (m
->show_status_overridden
!= _SHOW_STATUS_INVALID
)
4337 return m
->show_status_overridden
;
4339 return m
->show_status
;
4342 bool manager_get_show_status_on(Manager
*m
) {
4345 return show_status_on(manager_get_show_status(m
));
4348 static void set_show_status_marker(bool b
) {
4350 (void) touch("/run/systemd/show-status");
4352 (void) unlink("/run/systemd/show-status");
4355 void manager_set_show_status(Manager
*m
, ShowStatus mode
, const char *reason
) {
4358 assert(mode
>= 0 && mode
< _SHOW_STATUS_MAX
);
4360 if (MANAGER_IS_USER(m
))
4363 if (mode
== m
->show_status
)
4366 if (m
->show_status_overridden
== _SHOW_STATUS_INVALID
) {
4369 enabled
= show_status_on(mode
);
4370 log_debug("%s (%s) showing of status (%s).",
4371 enabled
? "Enabling" : "Disabling",
4372 strna(show_status_to_string(mode
)),
4375 set_show_status_marker(enabled
);
4378 m
->show_status
= mode
;
4381 void manager_override_show_status(Manager
*m
, ShowStatus mode
, const char *reason
) {
4383 assert(mode
< _SHOW_STATUS_MAX
);
4385 if (MANAGER_IS_USER(m
))
4388 if (mode
== m
->show_status_overridden
)
4391 m
->show_status_overridden
= mode
;
4393 if (mode
== _SHOW_STATUS_INVALID
)
4394 mode
= m
->show_status
;
4396 log_debug("%s (%s) showing of status (%s).",
4397 m
->show_status_overridden
!= _SHOW_STATUS_INVALID
? "Overriding" : "Restoring",
4398 strna(show_status_to_string(mode
)),
4401 set_show_status_marker(show_status_on(mode
));
4404 const char *manager_get_confirm_spawn(Manager
*m
) {
4405 static int last_errno
= 0;
4411 /* Here's the deal: we want to test the validity of the console but don't want
4412 * PID1 to go through the whole console process which might block. But we also
4413 * want to warn the user only once if something is wrong with the console so we
4414 * cannot do the sanity checks after spawning our children. So here we simply do
4415 * really basic tests to hopefully trap common errors.
4417 * If the console suddenly disappear at the time our children will really it
4418 * then they will simply fail to acquire it and a positive answer will be
4419 * assumed. New children will fall back to /dev/console though.
4421 * Note: TTYs are devices that can come and go any time, and frequently aren't
4422 * available yet during early boot (consider a USB rs232 dongle...). If for any
4423 * reason the configured console is not ready, we fall back to the default
4426 if (!m
->confirm_spawn
|| path_equal(m
->confirm_spawn
, "/dev/console"))
4427 return m
->confirm_spawn
;
4429 if (stat(m
->confirm_spawn
, &st
) < 0) {
4434 if (!S_ISCHR(st
.st_mode
)) {
4440 return m
->confirm_spawn
;
4443 if (last_errno
!= r
)
4444 last_errno
= log_warning_errno(r
, "Failed to open %s, using default console: %m", m
->confirm_spawn
);
4446 return "/dev/console";
4449 void manager_set_first_boot(Manager
*m
, bool b
) {
4452 if (!MANAGER_IS_SYSTEM(m
))
4455 if (m
->first_boot
!= (int) b
) {
4457 (void) touch("/run/systemd/first-boot");
4459 (void) unlink("/run/systemd/first-boot");
4465 void manager_disable_confirm_spawn(void) {
4466 (void) touch("/run/systemd/confirm_spawn_disabled");
4469 bool manager_is_confirm_spawn_disabled(Manager
*m
) {
4470 if (!m
->confirm_spawn
)
4473 return access("/run/systemd/confirm_spawn_disabled", F_OK
) >= 0;
4476 static bool manager_should_show_status(Manager
*m
, StatusType type
) {
4479 if (!MANAGER_IS_SYSTEM(m
))
4482 if (m
->no_console_output
)
4485 if (!IN_SET(manager_state(m
), MANAGER_INITIALIZING
, MANAGER_STARTING
, MANAGER_STOPPING
))
4488 /* If we cannot find out the status properly, just proceed. */
4489 if (type
!= STATUS_TYPE_EMERGENCY
&& manager_check_ask_password(m
) > 0)
4492 if (type
== STATUS_TYPE_NOTICE
&& m
->show_status
!= SHOW_STATUS_NO
)
4495 return manager_get_show_status_on(m
);
4498 void manager_status_printf(Manager
*m
, StatusType type
, const char *status
, const char *format
, ...) {
4501 /* If m is NULL, assume we're after shutdown and let the messages through. */
4503 if (m
&& !manager_should_show_status(m
, type
))
4506 /* XXX We should totally drop the check for ephemeral here
4507 * and thus effectively make 'Type=idle' pointless. */
4508 if (type
== STATUS_TYPE_EPHEMERAL
&& m
&& m
->n_on_console
> 0)
4511 va_start(ap
, format
);
4512 status_vprintf(status
, SHOW_STATUS_ELLIPSIZE
|(type
== STATUS_TYPE_EPHEMERAL
? SHOW_STATUS_EPHEMERAL
: 0), format
, ap
);
4516 Set
* manager_get_units_requiring_mounts_for(Manager
*m
, const char *path
) {
4520 if (path_equal(path
, "/"))
4523 return hashmap_get(m
->units_requiring_mounts_for
, path
);
4526 int manager_update_failed_units(Manager
*m
, Unit
*u
, bool failed
) {
4531 assert(u
->manager
== m
);
4533 size
= set_size(m
->failed_units
);
4536 r
= set_ensure_put(&m
->failed_units
, NULL
, u
);
4540 (void) set_remove(m
->failed_units
, u
);
4542 if (set_size(m
->failed_units
) != size
)
4543 bus_manager_send_change_signal(m
);
4548 ManagerState
manager_state(Manager
*m
) {
4553 /* Is the special shutdown target active or queued? If so, we are in shutdown state */
4554 u
= manager_get_unit(m
, SPECIAL_SHUTDOWN_TARGET
);
4555 if (u
&& unit_active_or_pending(u
))
4556 return MANAGER_STOPPING
;
4558 /* Did we ever finish booting? If not then we are still starting up */
4559 if (!MANAGER_IS_FINISHED(m
)) {
4561 u
= manager_get_unit(m
, SPECIAL_BASIC_TARGET
);
4562 if (!u
|| !UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u
)))
4563 return MANAGER_INITIALIZING
;
4565 return MANAGER_STARTING
;
4568 if (MANAGER_IS_SYSTEM(m
)) {
4569 /* Are the rescue or emergency targets active or queued? If so we are in maintenance state */
4570 u
= manager_get_unit(m
, SPECIAL_RESCUE_TARGET
);
4571 if (u
&& unit_active_or_pending(u
))
4572 return MANAGER_MAINTENANCE
;
4574 u
= manager_get_unit(m
, SPECIAL_EMERGENCY_TARGET
);
4575 if (u
&& unit_active_or_pending(u
))
4576 return MANAGER_MAINTENANCE
;
4579 /* Are there any failed units? If so, we are in degraded mode */
4580 if (set_size(m
->failed_units
) > 0)
4581 return MANAGER_DEGRADED
;
4583 return MANAGER_RUNNING
;
4586 static void manager_unref_uid_internal(
4590 int (*_clean_ipc
)(uid_t uid
)) {
4594 assert(uid_is_valid(uid
));
4597 /* A generic implementation, covering both manager_unref_uid() and manager_unref_gid(), under the assumption
4598 * that uid_t and gid_t are actually defined the same way, with the same validity rules.
4600 * We store a hashmap where the UID/GID is they key and the value is a 32bit reference counter, whose highest
4601 * bit is used as flag for marking UIDs/GIDs whose IPC objects to remove when the last reference to the UID/GID
4602 * is dropped. The flag is set to on, once at least one reference from a unit where RemoveIPC= is set is added
4603 * on a UID/GID. It is reset when the UID's/GID's reference counter drops to 0 again. */
4605 assert_cc(sizeof(uid_t
) == sizeof(gid_t
));
4606 assert_cc(UID_INVALID
== (uid_t
) GID_INVALID
);
4608 if (uid
== 0) /* We don't keep track of root, and will never destroy it */
4611 c
= PTR_TO_UINT32(hashmap_get(uid_refs
, UID_TO_PTR(uid
)));
4613 n
= c
& ~DESTROY_IPC_FLAG
;
4617 if (destroy_now
&& n
== 0) {
4618 hashmap_remove(uid_refs
, UID_TO_PTR(uid
));
4620 if (c
& DESTROY_IPC_FLAG
) {
4621 log_debug("%s " UID_FMT
" is no longer referenced, cleaning up its IPC.",
4622 _clean_ipc
== clean_ipc_by_uid
? "UID" : "GID",
4624 (void) _clean_ipc(uid
);
4627 c
= n
| (c
& DESTROY_IPC_FLAG
);
4628 assert_se(hashmap_update(uid_refs
, UID_TO_PTR(uid
), UINT32_TO_PTR(c
)) >= 0);
4632 void manager_unref_uid(Manager
*m
, uid_t uid
, bool destroy_now
) {
4633 manager_unref_uid_internal(m
->uid_refs
, uid
, destroy_now
, clean_ipc_by_uid
);
4636 void manager_unref_gid(Manager
*m
, gid_t gid
, bool destroy_now
) {
4637 manager_unref_uid_internal(m
->gid_refs
, (uid_t
) gid
, destroy_now
, clean_ipc_by_gid
);
4640 static int manager_ref_uid_internal(
4649 assert(uid_is_valid(uid
));
4651 /* A generic implementation, covering both manager_ref_uid() and manager_ref_gid(), under the assumption
4652 * that uid_t and gid_t are actually defined the same way, with the same validity rules. */
4654 assert_cc(sizeof(uid_t
) == sizeof(gid_t
));
4655 assert_cc(UID_INVALID
== (uid_t
) GID_INVALID
);
4657 if (uid
== 0) /* We don't keep track of root, and will never destroy it */
4660 r
= hashmap_ensure_allocated(uid_refs
, &trivial_hash_ops
);
4664 c
= PTR_TO_UINT32(hashmap_get(*uid_refs
, UID_TO_PTR(uid
)));
4666 n
= c
& ~DESTROY_IPC_FLAG
;
4669 if (n
& DESTROY_IPC_FLAG
) /* check for overflow */
4672 c
= n
| (c
& DESTROY_IPC_FLAG
) | (clean_ipc
? DESTROY_IPC_FLAG
: 0);
4674 return hashmap_replace(*uid_refs
, UID_TO_PTR(uid
), UINT32_TO_PTR(c
));
4677 int manager_ref_uid(Manager
*m
, uid_t uid
, bool clean_ipc
) {
4678 return manager_ref_uid_internal(&m
->uid_refs
, uid
, clean_ipc
);
4681 int manager_ref_gid(Manager
*m
, gid_t gid
, bool clean_ipc
) {
4682 return manager_ref_uid_internal(&m
->gid_refs
, (uid_t
) gid
, clean_ipc
);
4685 static void manager_vacuum_uid_refs_internal(
4687 int (*_clean_ipc
)(uid_t uid
)) {
4693 HASHMAP_FOREACH_KEY(p
, k
, uid_refs
) {
4697 uid
= PTR_TO_UID(k
);
4698 c
= PTR_TO_UINT32(p
);
4700 n
= c
& ~DESTROY_IPC_FLAG
;
4704 if (c
& DESTROY_IPC_FLAG
) {
4705 log_debug("Found unreferenced %s " UID_FMT
" after reload/reexec. Cleaning up.",
4706 _clean_ipc
== clean_ipc_by_uid
? "UID" : "GID",
4708 (void) _clean_ipc(uid
);
4711 assert_se(hashmap_remove(uid_refs
, k
) == p
);
4715 static void manager_vacuum_uid_refs(Manager
*m
) {
4716 manager_vacuum_uid_refs_internal(m
->uid_refs
, clean_ipc_by_uid
);
4719 static void manager_vacuum_gid_refs(Manager
*m
) {
4720 manager_vacuum_uid_refs_internal(m
->gid_refs
, clean_ipc_by_gid
);
4723 static void manager_vacuum(Manager
*m
) {
4726 /* Release any dynamic users no longer referenced */
4727 dynamic_user_vacuum(m
, true);
4729 /* Release any references to UIDs/GIDs no longer referenced, and destroy any IPC owned by them */
4730 manager_vacuum_uid_refs(m
);
4731 manager_vacuum_gid_refs(m
);
4733 /* Release any runtimes no longer referenced */
4734 exec_runtime_vacuum(m
);
4737 int manager_dispatch_user_lookup_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
4741 char unit_name
[UNIT_NAME_MAX
+1];
4744 Manager
*m
= userdata
;
4752 /* Invoked whenever a child process succeeded resolving its user/group to use and sent us the resulting UID/GID
4753 * in a datagram. We parse the datagram here and pass it off to the unit, so that it can add a reference to the
4754 * UID/GID so that it can destroy the UID/GID's IPC objects when the reference counter drops to 0. */
4756 l
= recv(fd
, &buffer
, sizeof(buffer
), MSG_DONTWAIT
);
4758 if (IN_SET(errno
, EINTR
, EAGAIN
))
4761 return log_error_errno(errno
, "Failed to read from user lookup fd: %m");
4764 if ((size_t) l
<= offsetof(struct buffer
, unit_name
)) {
4765 log_warning("Received too short user lookup message, ignoring.");
4769 if ((size_t) l
> offsetof(struct buffer
, unit_name
) + UNIT_NAME_MAX
) {
4770 log_warning("Received too long user lookup message, ignoring.");
4774 if (!uid_is_valid(buffer
.uid
) && !gid_is_valid(buffer
.gid
)) {
4775 log_warning("Got user lookup message with invalid UID/GID pair, ignoring.");
4779 n
= (size_t) l
- offsetof(struct buffer
, unit_name
);
4780 if (memchr(buffer
.unit_name
, 0, n
)) {
4781 log_warning("Received lookup message with embedded NUL character, ignoring.");
4785 buffer
.unit_name
[n
] = 0;
4786 u
= manager_get_unit(m
, buffer
.unit_name
);
4788 log_debug("Got user lookup message but unit doesn't exist, ignoring.");
4792 log_unit_debug(u
, "User lookup succeeded: uid=" UID_FMT
" gid=" GID_FMT
, buffer
.uid
, buffer
.gid
);
4794 unit_notify_user_lookup(u
, buffer
.uid
, buffer
.gid
);
4798 char *manager_taint_string(Manager
*m
) {
4799 _cleanup_free_
char *destination
= NULL
, *overflowuid
= NULL
, *overflowgid
= NULL
;
4803 /* Returns a "taint string", e.g. "local-hwclock:var-run-bad".
4804 * Only things that are detected at runtime should be tagged
4805 * here. For stuff that is set during compilation, emit a warning
4806 * in the configuration phase. */
4810 buf
= new(char, sizeof("split-usr:"
4815 "overflowuid-not-65534:"
4816 "overflowgid-not-65534:"));
4824 e
= stpcpy(e
, "split-usr:");
4826 if (access("/proc/cgroups", F_OK
) < 0)
4827 e
= stpcpy(e
, "cgroups-missing:");
4829 if (cg_all_unified() == 0)
4830 e
= stpcpy(e
, "cgroupsv1:");
4832 if (clock_is_localtime(NULL
) > 0)
4833 e
= stpcpy(e
, "local-hwclock:");
4835 r
= readlink_malloc("/var/run", &destination
);
4836 if (r
< 0 || !PATH_IN_SET(destination
, "../run", "/run"))
4837 e
= stpcpy(e
, "var-run-bad:");
4839 r
= read_one_line_file("/proc/sys/kernel/overflowuid", &overflowuid
);
4840 if (r
>= 0 && !streq(overflowuid
, "65534"))
4841 e
= stpcpy(e
, "overflowuid-not-65534:");
4843 r
= read_one_line_file("/proc/sys/kernel/overflowgid", &overflowgid
);
4844 if (r
>= 0 && !streq(overflowgid
, "65534"))
4845 e
= stpcpy(e
, "overflowgid-not-65534:");
4847 /* remove the last ':' */
4854 void manager_ref_console(Manager
*m
) {
4860 void manager_unref_console(Manager
*m
) {
4862 assert(m
->n_on_console
> 0);
4865 if (m
->n_on_console
== 0)
4866 m
->no_console_output
= false; /* unset no_console_output flag, since the console is definitely free now */
4869 void manager_override_log_level(Manager
*m
, int level
) {
4870 _cleanup_free_
char *s
= NULL
;
4873 if (!m
->log_level_overridden
) {
4874 m
->original_log_level
= log_get_max_level();
4875 m
->log_level_overridden
= true;
4878 (void) log_level_to_string_alloc(level
, &s
);
4879 log_info("Setting log level to %s.", strna(s
));
4881 log_set_max_level(level
);
4884 void manager_restore_original_log_level(Manager
*m
) {
4885 _cleanup_free_
char *s
= NULL
;
4888 if (!m
->log_level_overridden
)
4891 (void) log_level_to_string_alloc(m
->original_log_level
, &s
);
4892 log_info("Restoring log level to original (%s).", strna(s
));
4894 log_set_max_level(m
->original_log_level
);
4895 m
->log_level_overridden
= false;
4898 void manager_override_log_target(Manager
*m
, LogTarget target
) {
4901 if (!m
->log_target_overridden
) {
4902 m
->original_log_target
= log_get_target();
4903 m
->log_target_overridden
= true;
4906 log_info("Setting log target to %s.", log_target_to_string(target
));
4907 log_set_target(target
);
4910 void manager_restore_original_log_target(Manager
*m
) {
4913 if (!m
->log_target_overridden
)
4916 log_info("Restoring log target to original %s.", log_target_to_string(m
->original_log_target
));
4918 log_set_target(m
->original_log_target
);
4919 m
->log_target_overridden
= false;
4922 ManagerTimestamp
manager_timestamp_initrd_mangle(ManagerTimestamp s
) {
4924 s
>= MANAGER_TIMESTAMP_SECURITY_START
&&
4925 s
<= MANAGER_TIMESTAMP_UNITS_LOAD_FINISH
)
4926 return s
- MANAGER_TIMESTAMP_SECURITY_START
+ MANAGER_TIMESTAMP_INITRD_SECURITY_START
;
4930 static const char *const manager_state_table
[_MANAGER_STATE_MAX
] = {
4931 [MANAGER_INITIALIZING
] = "initializing",
4932 [MANAGER_STARTING
] = "starting",
4933 [MANAGER_RUNNING
] = "running",
4934 [MANAGER_DEGRADED
] = "degraded",
4935 [MANAGER_MAINTENANCE
] = "maintenance",
4936 [MANAGER_STOPPING
] = "stopping",
4939 DEFINE_STRING_TABLE_LOOKUP(manager_state
, ManagerState
);
4941 static const char *const manager_timestamp_table
[_MANAGER_TIMESTAMP_MAX
] = {
4942 [MANAGER_TIMESTAMP_FIRMWARE
] = "firmware",
4943 [MANAGER_TIMESTAMP_LOADER
] = "loader",
4944 [MANAGER_TIMESTAMP_KERNEL
] = "kernel",
4945 [MANAGER_TIMESTAMP_INITRD
] = "initrd",
4946 [MANAGER_TIMESTAMP_USERSPACE
] = "userspace",
4947 [MANAGER_TIMESTAMP_FINISH
] = "finish",
4948 [MANAGER_TIMESTAMP_SECURITY_START
] = "security-start",
4949 [MANAGER_TIMESTAMP_SECURITY_FINISH
] = "security-finish",
4950 [MANAGER_TIMESTAMP_GENERATORS_START
] = "generators-start",
4951 [MANAGER_TIMESTAMP_GENERATORS_FINISH
] = "generators-finish",
4952 [MANAGER_TIMESTAMP_UNITS_LOAD_START
] = "units-load-start",
4953 [MANAGER_TIMESTAMP_UNITS_LOAD_FINISH
] = "units-load-finish",
4954 [MANAGER_TIMESTAMP_INITRD_SECURITY_START
] = "initrd-security-start",
4955 [MANAGER_TIMESTAMP_INITRD_SECURITY_FINISH
] = "initrd-security-finish",
4956 [MANAGER_TIMESTAMP_INITRD_GENERATORS_START
] = "initrd-generators-start",
4957 [MANAGER_TIMESTAMP_INITRD_GENERATORS_FINISH
] = "initrd-generators-finish",
4958 [MANAGER_TIMESTAMP_INITRD_UNITS_LOAD_START
] = "initrd-units-load-start",
4959 [MANAGER_TIMESTAMP_INITRD_UNITS_LOAD_FINISH
] = "initrd-units-load-finish",
4962 DEFINE_STRING_TABLE_LOOKUP(manager_timestamp
, ManagerTimestamp
);
4964 static const char* const oom_policy_table
[_OOM_POLICY_MAX
] = {
4965 [OOM_CONTINUE
] = "continue",
4966 [OOM_STOP
] = "stop",
4967 [OOM_KILL
] = "kill",
4970 DEFINE_STRING_TABLE_LOOKUP(oom_policy
, OOMPolicy
);