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 "memory-util.h"
60 #include "parse-util.h"
61 #include "path-lookup.h"
62 #include "path-util.h"
63 #include "process-util.h"
64 #include "ratelimit.h"
65 #include "rlimit-util.h"
67 #include "selinux-util.h"
68 #include "serialize.h"
69 #include "signal-util.h"
70 #include "socket-util.h"
72 #include "stat-util.h"
73 #include "string-table.h"
74 #include "string-util.h"
77 #include "sysctl-util.h"
78 #include "syslog-util.h"
79 #include "terminal-util.h"
80 #include "time-util.h"
81 #include "transaction.h"
82 #include "umask-util.h"
83 #include "unit-name.h"
84 #include "unit-serialize.h"
85 #include "user-util.h"
89 #define NOTIFY_RCVBUF_SIZE (8*1024*1024)
90 #define CGROUPS_AGENT_RCVBUF_SIZE (8*1024*1024)
92 /* Initial delay and the interval for printing status messages about running jobs */
93 #define JOBS_IN_PROGRESS_WAIT_USEC (2*USEC_PER_SEC)
94 #define JOBS_IN_PROGRESS_QUIET_WAIT_USEC (25*USEC_PER_SEC)
95 #define JOBS_IN_PROGRESS_PERIOD_USEC (USEC_PER_SEC / 3)
96 #define JOBS_IN_PROGRESS_PERIOD_DIVISOR 3
98 /* If there are more than 1K bus messages queue across our API and direct buses, then let's not add more on top until
99 * the queue gets more empty. */
100 #define MANAGER_BUS_BUSY_THRESHOLD 1024LU
102 /* How many units and jobs to process of the bus queue before returning to the event loop. */
103 #define MANAGER_BUS_MESSAGE_BUDGET 100U
105 static int manager_dispatch_notify_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
106 static int manager_dispatch_cgroups_agent_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
107 static int manager_dispatch_signal_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
108 static int manager_dispatch_time_change_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
109 static int manager_dispatch_idle_pipe_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
110 static int manager_dispatch_user_lookup_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
111 static int manager_dispatch_jobs_in_progress(sd_event_source
*source
, usec_t usec
, void *userdata
);
112 static int manager_dispatch_run_queue(sd_event_source
*source
, void *userdata
);
113 static int manager_dispatch_sigchld(sd_event_source
*source
, void *userdata
);
114 static int manager_dispatch_timezone_change(sd_event_source
*source
, const struct inotify_event
*event
, void *userdata
);
115 static int manager_run_environment_generators(Manager
*m
);
116 static int manager_run_generators(Manager
*m
);
117 static void manager_vacuum(Manager
*m
);
119 static usec_t
manager_watch_jobs_next_time(Manager
*m
) {
120 return usec_add(now(CLOCK_MONOTONIC
),
121 show_status_on(m
->show_status
) ? JOBS_IN_PROGRESS_WAIT_USEC
:
122 JOBS_IN_PROGRESS_QUIET_WAIT_USEC
);
125 static void manager_watch_jobs_in_progress(Manager
*m
) {
131 /* We do not want to show the cylon animation if the user
132 * needs to confirm service executions otherwise confirmation
133 * messages will be screwed by the cylon animation. */
134 if (!manager_is_confirm_spawn_disabled(m
))
137 if (m
->jobs_in_progress_event_source
)
140 next
= manager_watch_jobs_next_time(m
);
141 r
= sd_event_add_time(
143 &m
->jobs_in_progress_event_source
,
146 manager_dispatch_jobs_in_progress
, m
);
150 (void) sd_event_source_set_description(m
->jobs_in_progress_event_source
, "manager-jobs-in-progress");
153 #define CYLON_BUFFER_EXTRA (2*STRLEN(ANSI_RED) + STRLEN(ANSI_HIGHLIGHT_RED) + 2*STRLEN(ANSI_NORMAL))
155 static void draw_cylon(char buffer
[], size_t buflen
, unsigned width
, unsigned pos
) {
158 assert(buflen
>= CYLON_BUFFER_EXTRA
+ width
+ 1);
159 assert(pos
<= width
+1); /* 0 or width+1 mean that the center light is behind the corner */
163 p
= mempset(p
, ' ', pos
-2);
164 if (log_get_show_color())
165 p
= stpcpy(p
, ANSI_RED
);
169 if (pos
> 0 && pos
<= width
) {
170 if (log_get_show_color())
171 p
= stpcpy(p
, ANSI_HIGHLIGHT_RED
);
175 if (log_get_show_color())
176 p
= stpcpy(p
, ANSI_NORMAL
);
179 if (log_get_show_color())
180 p
= stpcpy(p
, ANSI_RED
);
183 p
= mempset(p
, ' ', width
-1-pos
);
184 if (log_get_show_color())
185 strcpy(p
, ANSI_NORMAL
);
189 static void manager_flip_auto_status(Manager
*m
, bool enable
, const char *reason
) {
193 if (m
->show_status
== SHOW_STATUS_AUTO
)
194 manager_set_show_status(m
, SHOW_STATUS_TEMPORARY
, reason
);
196 if (m
->show_status
== SHOW_STATUS_TEMPORARY
)
197 manager_set_show_status(m
, SHOW_STATUS_AUTO
, reason
);
201 static void manager_print_jobs_in_progress(Manager
*m
) {
202 _cleanup_free_
char *job_of_n
= NULL
;
204 unsigned counter
= 0, print_nr
;
205 char cylon
[6 + CYLON_BUFFER_EXTRA
+ 1];
207 char time
[FORMAT_TIMESPAN_MAX
], limit
[FORMAT_TIMESPAN_MAX
] = "no limit";
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)
238 format_timespan(time
, sizeof(time
), now(CLOCK_MONOTONIC
) - j
->begin_usec
, 1*USEC_PER_SEC
);
239 if (job_get_timeout(j
, &x
) > 0)
240 format_timespan(limit
, sizeof(limit
), x
- j
->begin_usec
, 1*USEC_PER_SEC
);
242 manager_status_printf(m
, STATUS_TYPE_EPHEMERAL
, cylon
,
243 "%sA %s job is running for %s (%s / %s)",
245 job_type_to_string(j
->type
),
246 unit_status_string(j
->unit
),
250 static int have_ask_password(void) {
251 _cleanup_closedir_
DIR *dir
= NULL
;
254 dir
= opendir("/run/systemd/ask-password");
262 FOREACH_DIRENT_ALL(de
, dir
, return -errno
) {
263 if (startswith(de
->d_name
, "ask."))
269 static int manager_dispatch_ask_password_fd(sd_event_source
*source
,
270 int fd
, uint32_t revents
, void *userdata
) {
271 Manager
*m
= userdata
;
277 m
->have_ask_password
= have_ask_password();
278 if (m
->have_ask_password
< 0)
279 /* Log error but continue. Negative have_ask_password
280 * is treated as unknown status. */
281 log_error_errno(m
->have_ask_password
, "Failed to list /run/systemd/ask-password: %m");
286 static void manager_close_ask_password(Manager
*m
) {
289 m
->ask_password_event_source
= sd_event_source_unref(m
->ask_password_event_source
);
290 m
->ask_password_inotify_fd
= safe_close(m
->ask_password_inotify_fd
);
291 m
->have_ask_password
= -EINVAL
;
294 static int manager_check_ask_password(Manager
*m
) {
299 if (!m
->ask_password_event_source
) {
300 assert(m
->ask_password_inotify_fd
< 0);
302 (void) mkdir_p_label("/run/systemd/ask-password", 0755);
304 m
->ask_password_inotify_fd
= inotify_init1(IN_NONBLOCK
|IN_CLOEXEC
);
305 if (m
->ask_password_inotify_fd
< 0)
306 return log_error_errno(errno
, "Failed to create inotify object: %m");
308 r
= inotify_add_watch_and_warn(m
->ask_password_inotify_fd
,
309 "/run/systemd/ask-password",
310 IN_CREATE
|IN_DELETE
|IN_MOVE
);
312 manager_close_ask_password(m
);
316 r
= sd_event_add_io(m
->event
, &m
->ask_password_event_source
,
317 m
->ask_password_inotify_fd
, EPOLLIN
,
318 manager_dispatch_ask_password_fd
, m
);
320 log_error_errno(r
, "Failed to add event source for /run/systemd/ask-password: %m");
321 manager_close_ask_password(m
);
325 (void) sd_event_source_set_description(m
->ask_password_event_source
, "manager-ask-password");
327 /* Queries might have been added meanwhile... */
328 manager_dispatch_ask_password_fd(m
->ask_password_event_source
,
329 m
->ask_password_inotify_fd
, EPOLLIN
, m
);
332 return m
->have_ask_password
;
335 static int manager_watch_idle_pipe(Manager
*m
) {
340 if (m
->idle_pipe_event_source
)
343 if (m
->idle_pipe
[2] < 0)
346 r
= sd_event_add_io(m
->event
, &m
->idle_pipe_event_source
, m
->idle_pipe
[2], EPOLLIN
, manager_dispatch_idle_pipe_fd
, m
);
348 return log_error_errno(r
, "Failed to watch idle pipe: %m");
350 (void) sd_event_source_set_description(m
->idle_pipe_event_source
, "manager-idle-pipe");
355 static void manager_close_idle_pipe(Manager
*m
) {
358 m
->idle_pipe_event_source
= sd_event_source_unref(m
->idle_pipe_event_source
);
360 safe_close_pair(m
->idle_pipe
);
361 safe_close_pair(m
->idle_pipe
+ 2);
364 static int manager_setup_time_change(Manager
*m
) {
369 if (MANAGER_IS_TEST_RUN(m
))
372 m
->time_change_event_source
= sd_event_source_unref(m
->time_change_event_source
);
373 m
->time_change_fd
= safe_close(m
->time_change_fd
);
375 m
->time_change_fd
= time_change_fd();
376 if (m
->time_change_fd
< 0)
377 return log_error_errno(m
->time_change_fd
, "Failed to create timer change timer fd: %m");
379 r
= sd_event_add_io(m
->event
, &m
->time_change_event_source
, m
->time_change_fd
, EPOLLIN
, manager_dispatch_time_change_fd
, m
);
381 return log_error_errno(r
, "Failed to create time change event source: %m");
383 /* Schedule this slightly earlier than the .timer event sources */
384 r
= sd_event_source_set_priority(m
->time_change_event_source
, SD_EVENT_PRIORITY_NORMAL
-1);
386 return log_error_errno(r
, "Failed to set priority of time change event sources: %m");
388 (void) sd_event_source_set_description(m
->time_change_event_source
, "manager-time-change");
390 log_debug("Set up TFD_TIMER_CANCEL_ON_SET timerfd.");
395 static int manager_read_timezone_stat(Manager
*m
) {
401 /* Read the current stat() data of /etc/localtime so that we detect changes */
402 if (lstat("/etc/localtime", &st
) < 0) {
403 log_debug_errno(errno
, "Failed to stat /etc/localtime, ignoring: %m");
404 changed
= m
->etc_localtime_accessible
;
405 m
->etc_localtime_accessible
= false;
409 k
= timespec_load(&st
.st_mtim
);
410 changed
= !m
->etc_localtime_accessible
|| k
!= m
->etc_localtime_mtime
;
412 m
->etc_localtime_mtime
= k
;
413 m
->etc_localtime_accessible
= true;
419 static int manager_setup_timezone_change(Manager
*m
) {
420 _cleanup_(sd_event_source_unrefp
) sd_event_source
*new_event
= NULL
;
425 if (MANAGER_IS_TEST_RUN(m
))
428 /* We watch /etc/localtime for three events: change of the link count (which might mean removal from /etc even
429 * though another link might be kept), renames, and file close operations after writing. Note we don't bother
430 * with IN_DELETE_SELF, as that would just report when the inode is removed entirely, i.e. after the link count
431 * went to zero and all fds to it are closed.
433 * Note that we never follow symlinks here. This is a simplification, but should cover almost all cases
436 * Note that we create the new event source first here, before releasing the old one. This should optimize
437 * behaviour as this way sd-event can reuse the old watch in case the inode didn't change. */
439 r
= sd_event_add_inotify(m
->event
, &new_event
, "/etc/localtime",
440 IN_ATTRIB
|IN_MOVE_SELF
|IN_CLOSE_WRITE
|IN_DONT_FOLLOW
, manager_dispatch_timezone_change
, m
);
442 /* If the file doesn't exist yet, subscribe to /etc instead, and wait until it is created either by
443 * O_CREATE or by rename() */
445 log_debug_errno(r
, "/etc/localtime doesn't exist yet, watching /etc instead.");
446 r
= sd_event_add_inotify(m
->event
, &new_event
, "/etc",
447 IN_CREATE
|IN_MOVED_TO
|IN_ONLYDIR
, manager_dispatch_timezone_change
, m
);
450 return log_error_errno(r
, "Failed to create timezone change event source: %m");
452 /* Schedule this slightly earlier than the .timer event sources */
453 r
= sd_event_source_set_priority(new_event
, SD_EVENT_PRIORITY_NORMAL
-1);
455 return log_error_errno(r
, "Failed to set priority of timezone change event sources: %m");
457 sd_event_source_unref(m
->timezone_change_event_source
);
458 m
->timezone_change_event_source
= TAKE_PTR(new_event
);
463 static int enable_special_signals(Manager
*m
) {
464 _cleanup_close_
int fd
= -1;
468 if (MANAGER_IS_TEST_RUN(m
))
471 /* Enable that we get SIGINT on control-alt-del. In containers
472 * this will fail with EPERM (older) or EINVAL (newer), so
474 if (reboot(RB_DISABLE_CAD
) < 0 && !IN_SET(errno
, EPERM
, EINVAL
))
475 log_warning_errno(errno
, "Failed to enable ctrl-alt-del handling: %m");
477 fd
= open_terminal("/dev/tty0", O_RDWR
|O_NOCTTY
|O_CLOEXEC
);
479 /* Support systems without virtual console */
481 log_warning_errno(errno
, "Failed to open /dev/tty0: %m");
483 /* Enable that we get SIGWINCH on kbrequest */
484 if (ioctl(fd
, KDSIGACCEPT
, SIGWINCH
) < 0)
485 log_warning_errno(errno
, "Failed to enable kbrequest handling: %m");
491 #define RTSIG_IF_AVAILABLE(signum) (signum <= SIGRTMAX ? signum : -1)
493 static int manager_setup_signals(Manager
*m
) {
494 struct sigaction sa
= {
495 .sa_handler
= SIG_DFL
,
496 .sa_flags
= SA_NOCLDSTOP
|SA_RESTART
,
503 assert_se(sigaction(SIGCHLD
, &sa
, NULL
) == 0);
505 /* We make liberal use of realtime signals here. On
506 * Linux/glibc we have 30 of them (with the exception of Linux
507 * on hppa, see below), between SIGRTMIN+0 ... SIGRTMIN+30
510 assert_se(sigemptyset(&mask
) == 0);
511 sigset_add_many(&mask
,
512 SIGCHLD
, /* Child died */
513 SIGTERM
, /* Reexecute daemon */
514 SIGHUP
, /* Reload configuration */
515 SIGUSR1
, /* systemd: reconnect to D-Bus */
516 SIGUSR2
, /* systemd: dump status */
517 SIGINT
, /* Kernel sends us this on control-alt-del */
518 SIGWINCH
, /* Kernel sends us this on kbrequest (alt-arrowup) */
519 SIGPWR
, /* Some kernel drivers and upsd send us this on power failure */
521 SIGRTMIN
+0, /* systemd: start default.target */
522 SIGRTMIN
+1, /* systemd: isolate rescue.target */
523 SIGRTMIN
+2, /* systemd: isolate emergency.target */
524 SIGRTMIN
+3, /* systemd: start halt.target */
525 SIGRTMIN
+4, /* systemd: start poweroff.target */
526 SIGRTMIN
+5, /* systemd: start reboot.target */
527 SIGRTMIN
+6, /* systemd: start kexec.target */
529 /* ... space for more special targets ... */
531 SIGRTMIN
+13, /* systemd: Immediate halt */
532 SIGRTMIN
+14, /* systemd: Immediate poweroff */
533 SIGRTMIN
+15, /* systemd: Immediate reboot */
534 SIGRTMIN
+16, /* systemd: Immediate kexec */
536 /* ... space for more immediate system state changes ... */
538 SIGRTMIN
+20, /* systemd: enable status messages */
539 SIGRTMIN
+21, /* systemd: disable status messages */
540 SIGRTMIN
+22, /* systemd: set log level to LOG_DEBUG */
541 SIGRTMIN
+23, /* systemd: set log level to LOG_INFO */
542 SIGRTMIN
+24, /* systemd: Immediate exit (--user only) */
544 /* .. one free signal here ... */
546 /* Apparently Linux on hppa had fewer RT signals until v3.18,
547 * SIGRTMAX was SIGRTMIN+25, and then SIGRTMIN was lowered,
548 * see commit v3.17-7614-g1f25df2eff.
550 * We cannot unconditionally make use of those signals here,
551 * so let's use a runtime check. Since these commands are
552 * accessible by different means and only really a safety
553 * net, the missing functionality on hppa shouldn't matter.
556 RTSIG_IF_AVAILABLE(SIGRTMIN
+26), /* systemd: set log target to journal-or-kmsg */
557 RTSIG_IF_AVAILABLE(SIGRTMIN
+27), /* systemd: set log target to console */
558 RTSIG_IF_AVAILABLE(SIGRTMIN
+28), /* systemd: set log target to kmsg */
559 RTSIG_IF_AVAILABLE(SIGRTMIN
+29), /* systemd: set log target to syslog-or-kmsg (obsolete) */
561 /* ... one free signal here SIGRTMIN+30 ... */
563 assert_se(sigprocmask(SIG_SETMASK
, &mask
, NULL
) == 0);
565 m
->signal_fd
= signalfd(-1, &mask
, SFD_NONBLOCK
|SFD_CLOEXEC
);
566 if (m
->signal_fd
< 0)
569 r
= sd_event_add_io(m
->event
, &m
->signal_event_source
, m
->signal_fd
, EPOLLIN
, manager_dispatch_signal_fd
, m
);
573 (void) sd_event_source_set_description(m
->signal_event_source
, "manager-signal");
575 /* Process signals a bit earlier than the rest of things, but later than notify_fd processing, so that the
576 * notify processing can still figure out to which process/service a message belongs, before we reap the
577 * process. Also, process this before handling cgroup notifications, so that we always collect child exit
578 * status information before detecting that there's no process in a cgroup. */
579 r
= sd_event_source_set_priority(m
->signal_event_source
, SD_EVENT_PRIORITY_NORMAL
-6);
583 if (MANAGER_IS_SYSTEM(m
))
584 return enable_special_signals(m
);
589 static char** sanitize_environment(char **l
) {
591 /* Let's remove some environment variables that we need ourselves to communicate with our clients */
595 "CONFIGURATION_DIRECTORY",
596 "CREDENTIALS_DIRECTORY",
618 /* Let's order the environment alphabetically, just to make it pretty */
624 int manager_default_environment(Manager
*m
) {
629 m
->transient_environment
= strv_free(m
->transient_environment
);
631 if (MANAGER_IS_SYSTEM(m
)) {
632 /* The system manager always starts with a clean
633 * environment for its children. It does not import
634 * the kernel's or the parents' exported variables.
636 * The initial passed environment is untouched to keep
637 * /proc/self/environ valid; it is used for tagging
638 * the init process inside containers. */
639 m
->transient_environment
= strv_new("PATH=" DEFAULT_PATH
);
640 if (!m
->transient_environment
)
643 /* Import locale variables LC_*= from configuration */
644 (void) locale_setup(&m
->transient_environment
);
646 /* The user manager passes its own environment along to its children, except for $PATH. */
647 m
->transient_environment
= strv_copy(environ
);
648 if (!m
->transient_environment
)
651 r
= strv_env_replace_strdup(&m
->transient_environment
, "PATH=" DEFAULT_USER_PATH
);
656 sanitize_environment(m
->transient_environment
);
661 static int manager_setup_prefix(Manager
*m
) {
667 static const struct table_entry paths_system
[_EXEC_DIRECTORY_TYPE_MAX
] = {
668 [EXEC_DIRECTORY_RUNTIME
] = { SD_PATH_SYSTEM_RUNTIME
, NULL
},
669 [EXEC_DIRECTORY_STATE
] = { SD_PATH_SYSTEM_STATE_PRIVATE
, NULL
},
670 [EXEC_DIRECTORY_CACHE
] = { SD_PATH_SYSTEM_STATE_CACHE
, NULL
},
671 [EXEC_DIRECTORY_LOGS
] = { SD_PATH_SYSTEM_STATE_LOGS
, NULL
},
672 [EXEC_DIRECTORY_CONFIGURATION
] = { SD_PATH_SYSTEM_CONFIGURATION
, NULL
},
675 static const struct table_entry paths_user
[_EXEC_DIRECTORY_TYPE_MAX
] = {
676 [EXEC_DIRECTORY_RUNTIME
] = { SD_PATH_USER_RUNTIME
, NULL
},
677 [EXEC_DIRECTORY_STATE
] = { SD_PATH_USER_CONFIGURATION
, NULL
},
678 [EXEC_DIRECTORY_CACHE
] = { SD_PATH_USER_STATE_CACHE
, NULL
},
679 [EXEC_DIRECTORY_LOGS
] = { SD_PATH_USER_CONFIGURATION
, "log" },
680 [EXEC_DIRECTORY_CONFIGURATION
] = { SD_PATH_USER_CONFIGURATION
, NULL
},
685 const struct table_entry
*p
= MANAGER_IS_SYSTEM(m
) ? paths_system
: paths_user
;
688 for (ExecDirectoryType i
= 0; i
< _EXEC_DIRECTORY_TYPE_MAX
; i
++) {
689 r
= sd_path_lookup(p
[i
].type
, p
[i
].suffix
, &m
->prefix
[i
]);
697 static void manager_free_unit_name_maps(Manager
*m
) {
698 m
->unit_id_map
= hashmap_free(m
->unit_id_map
);
699 m
->unit_name_map
= hashmap_free(m
->unit_name_map
);
700 m
->unit_path_cache
= set_free(m
->unit_path_cache
);
701 m
->unit_cache_timestamp_hash
= 0;
704 static int manager_setup_run_queue(Manager
*m
) {
708 assert(!m
->run_queue_event_source
);
710 r
= sd_event_add_defer(m
->event
, &m
->run_queue_event_source
, manager_dispatch_run_queue
, m
);
714 r
= sd_event_source_set_priority(m
->run_queue_event_source
, SD_EVENT_PRIORITY_IDLE
);
718 r
= sd_event_source_set_enabled(m
->run_queue_event_source
, SD_EVENT_OFF
);
722 (void) sd_event_source_set_description(m
->run_queue_event_source
, "manager-run-queue");
727 static int manager_setup_sigchld_event_source(Manager
*m
) {
731 assert(!m
->sigchld_event_source
);
733 r
= sd_event_add_defer(m
->event
, &m
->sigchld_event_source
, manager_dispatch_sigchld
, m
);
737 r
= sd_event_source_set_priority(m
->sigchld_event_source
, SD_EVENT_PRIORITY_NORMAL
-7);
741 r
= sd_event_source_set_enabled(m
->sigchld_event_source
, SD_EVENT_OFF
);
745 (void) sd_event_source_set_description(m
->sigchld_event_source
, "manager-sigchld");
750 int manager_new(UnitFileScope scope
, ManagerTestRunFlags test_run_flags
, Manager
**_m
) {
751 _cleanup_(manager_freep
) Manager
*m
= NULL
;
756 assert(IN_SET(scope
, UNIT_FILE_SYSTEM
, UNIT_FILE_USER
));
763 .unit_file_scope
= scope
,
764 .objective
= _MANAGER_OBJECTIVE_INVALID
,
766 .status_unit_format
= STATUS_UNIT_FORMAT_DEFAULT
,
768 .default_timer_accuracy_usec
= USEC_PER_MINUTE
,
769 .default_memory_accounting
= MEMORY_ACCOUNTING_DEFAULT
,
770 .default_tasks_accounting
= true,
771 .default_tasks_max
= TASKS_MAX_UNSET
,
772 .default_timeout_start_usec
= DEFAULT_TIMEOUT_USEC
,
773 .default_timeout_stop_usec
= DEFAULT_TIMEOUT_USEC
,
774 .default_restart_usec
= DEFAULT_RESTART_USEC
,
776 .original_log_level
= -1,
777 .original_log_target
= _LOG_TARGET_INVALID
,
779 .watchdog_overridden
[WATCHDOG_RUNTIME
] = USEC_INFINITY
,
780 .watchdog_overridden
[WATCHDOG_REBOOT
] = USEC_INFINITY
,
781 .watchdog_overridden
[WATCHDOG_KEXEC
] = USEC_INFINITY
,
783 .show_status_overridden
= _SHOW_STATUS_INVALID
,
786 .cgroups_agent_fd
= -1,
788 .time_change_fd
= -1,
789 .user_lookup_fds
= { -1, -1 },
790 .private_listen_fd
= -1,
792 .cgroup_inotify_fd
= -1,
793 .pin_cgroupfs_fd
= -1,
794 .ask_password_inotify_fd
= -1,
795 .idle_pipe
= { -1, -1, -1, -1},
797 /* start as id #1, so that we can leave #0 around as "null-like" value */
800 .have_ask_password
= -EINVAL
, /* we don't know */
802 .test_run_flags
= test_run_flags
,
804 .default_oom_policy
= OOM_STOP
,
808 if (MANAGER_IS_SYSTEM(m
) && detect_container() <= 0)
809 boot_timestamps(m
->timestamps
+ MANAGER_TIMESTAMP_USERSPACE
,
810 m
->timestamps
+ MANAGER_TIMESTAMP_FIRMWARE
,
811 m
->timestamps
+ MANAGER_TIMESTAMP_LOADER
);
814 /* Prepare log fields we can use for structured logging */
815 if (MANAGER_IS_SYSTEM(m
)) {
816 m
->unit_log_field
= "UNIT=";
817 m
->unit_log_format_string
= "UNIT=%s";
819 m
->invocation_log_field
= "INVOCATION_ID=";
820 m
->invocation_log_format_string
= "INVOCATION_ID=%s";
822 m
->unit_log_field
= "USER_UNIT=";
823 m
->unit_log_format_string
= "USER_UNIT=%s";
825 m
->invocation_log_field
= "USER_INVOCATION_ID=";
826 m
->invocation_log_format_string
= "USER_INVOCATION_ID=%s";
829 /* Reboot immediately if the user hits C-A-D more often than 7x per 2s */
830 m
->ctrl_alt_del_ratelimit
= (RateLimit
) { .interval
= 2 * USEC_PER_SEC
, .burst
= 7 };
832 r
= manager_default_environment(m
);
836 r
= hashmap_ensure_allocated(&m
->units
, &string_hash_ops
);
840 r
= hashmap_ensure_allocated(&m
->cgroup_unit
, &path_hash_ops
);
844 r
= hashmap_ensure_allocated(&m
->watch_bus
, &string_hash_ops
);
848 r
= prioq_ensure_allocated(&m
->run_queue
, compare_job_priority
);
852 r
= manager_setup_prefix(m
);
856 r
= get_credentials_dir(&e
);
858 m
->received_credentials
= strdup(e
);
859 if (!m
->received_credentials
)
863 r
= sd_event_default(&m
->event
);
867 r
= manager_setup_run_queue(m
);
871 if (test_run_flags
== MANAGER_TEST_RUN_MINIMAL
) {
872 m
->cgroup_root
= strdup("");
876 r
= manager_setup_signals(m
);
880 r
= manager_setup_cgroup(m
);
884 r
= manager_setup_time_change(m
);
888 r
= manager_read_timezone_stat(m
);
892 (void) manager_setup_timezone_change(m
);
894 r
= manager_setup_sigchld_event_source(m
);
899 if (test_run_flags
== 0) {
900 if (MANAGER_IS_SYSTEM(m
))
901 r
= mkdir_label("/run/systemd/units", 0755);
903 _cleanup_free_
char *units_path
= NULL
;
904 r
= xdg_user_runtime_dir(&units_path
, "/systemd/units");
907 r
= mkdir_p_label(units_path
, 0755);
910 if (r
< 0 && r
!= -EEXIST
)
916 dir_is_empty("/usr") > 0;
918 /* Note that we do not set up the notify fd here. We do that after deserialization,
919 * since they might have gotten serialized across the reexec. */
926 static int manager_setup_notify(Manager
*m
) {
929 if (MANAGER_IS_TEST_RUN(m
))
932 if (m
->notify_fd
< 0) {
933 _cleanup_close_
int fd
= -1;
934 union sockaddr_union sa
;
937 /* First free all secondary fields */
938 m
->notify_socket
= mfree(m
->notify_socket
);
939 m
->notify_event_source
= sd_event_source_unref(m
->notify_event_source
);
941 fd
= socket(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
|SOCK_NONBLOCK
, 0);
943 return log_error_errno(errno
, "Failed to allocate notification socket: %m");
945 fd_inc_rcvbuf(fd
, NOTIFY_RCVBUF_SIZE
);
947 m
->notify_socket
= path_join(m
->prefix
[EXEC_DIRECTORY_RUNTIME
], "systemd/notify");
948 if (!m
->notify_socket
)
951 r
= sockaddr_un_set_path(&sa
.un
, m
->notify_socket
);
953 return log_error_errno(r
, "Notify socket '%s' not valid for AF_UNIX socket address, refusing.",
957 (void) mkdir_parents_label(m
->notify_socket
, 0755);
958 (void) sockaddr_un_unlink(&sa
.un
);
960 r
= mac_selinux_bind(fd
, &sa
.sa
, sa_len
);
962 return log_error_errno(r
, "bind(%s) failed: %m", m
->notify_socket
);
964 r
= setsockopt_int(fd
, SOL_SOCKET
, SO_PASSCRED
, true);
966 return log_error_errno(r
, "SO_PASSCRED failed: %m");
968 m
->notify_fd
= TAKE_FD(fd
);
970 log_debug("Using notification socket %s", m
->notify_socket
);
973 if (!m
->notify_event_source
) {
974 r
= sd_event_add_io(m
->event
, &m
->notify_event_source
, m
->notify_fd
, EPOLLIN
, manager_dispatch_notify_fd
, m
);
976 return log_error_errno(r
, "Failed to allocate notify event source: %m");
978 /* Process notification messages a bit earlier than SIGCHLD, so that we can still identify to which
979 * service an exit message belongs. */
980 r
= sd_event_source_set_priority(m
->notify_event_source
, SD_EVENT_PRIORITY_NORMAL
-8);
982 return log_error_errno(r
, "Failed to set priority of notify event source: %m");
984 (void) sd_event_source_set_description(m
->notify_event_source
, "manager-notify");
990 static int manager_setup_cgroups_agent(Manager
*m
) {
992 static const union sockaddr_union sa
= {
993 .un
.sun_family
= AF_UNIX
,
994 .un
.sun_path
= "/run/systemd/cgroups-agent",
998 /* This creates a listening socket we receive cgroups agent messages on. We do not use D-Bus for delivering
999 * these messages from the cgroups agent binary to PID 1, as the cgroups agent binary is very short-living, and
1000 * each instance of it needs a new D-Bus connection. Since D-Bus connections are SOCK_STREAM/AF_UNIX, on
1001 * overloaded systems the backlog of the D-Bus socket becomes relevant, as not more than the configured number
1002 * of D-Bus connections may be queued until the kernel will start dropping further incoming connections,
1003 * possibly resulting in lost cgroups agent messages. To avoid this, we'll use a private SOCK_DGRAM/AF_UNIX
1004 * socket, where no backlog is relevant as communication may take place without an actual connect() cycle, and
1005 * we thus won't lose messages.
1007 * Note that PID 1 will forward the agent message to system bus, so that the user systemd instance may listen
1008 * to it. The system instance hence listens on this special socket, but the user instances listen on the system
1009 * bus for these messages. */
1011 if (MANAGER_IS_TEST_RUN(m
))
1014 if (!MANAGER_IS_SYSTEM(m
))
1017 r
= cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER
);
1019 return log_error_errno(r
, "Failed to determine whether unified cgroups hierarchy is used: %m");
1020 if (r
> 0) /* We don't need this anymore on the unified hierarchy */
1023 if (m
->cgroups_agent_fd
< 0) {
1024 _cleanup_close_
int fd
= -1;
1026 /* First free all secondary fields */
1027 m
->cgroups_agent_event_source
= sd_event_source_unref(m
->cgroups_agent_event_source
);
1029 fd
= socket(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
|SOCK_NONBLOCK
, 0);
1031 return log_error_errno(errno
, "Failed to allocate cgroups agent socket: %m");
1033 fd_inc_rcvbuf(fd
, CGROUPS_AGENT_RCVBUF_SIZE
);
1035 (void) sockaddr_un_unlink(&sa
.un
);
1037 /* Only allow root to connect to this socket */
1038 RUN_WITH_UMASK(0077)
1039 r
= bind(fd
, &sa
.sa
, SOCKADDR_UN_LEN(sa
.un
));
1041 return log_error_errno(errno
, "bind(%s) failed: %m", sa
.un
.sun_path
);
1043 m
->cgroups_agent_fd
= TAKE_FD(fd
);
1046 if (!m
->cgroups_agent_event_source
) {
1047 r
= sd_event_add_io(m
->event
, &m
->cgroups_agent_event_source
, m
->cgroups_agent_fd
, EPOLLIN
, manager_dispatch_cgroups_agent_fd
, m
);
1049 return log_error_errno(r
, "Failed to allocate cgroups agent event source: %m");
1051 /* Process cgroups notifications early. Note that when the agent notification is received
1052 * we'll just enqueue the unit in the cgroup empty queue, hence pick a high priority than
1053 * that. Also see handling of cgroup inotify for the unified cgroup stuff. */
1054 r
= sd_event_source_set_priority(m
->cgroups_agent_event_source
, SD_EVENT_PRIORITY_NORMAL
-9);
1056 return log_error_errno(r
, "Failed to set priority of cgroups agent event source: %m");
1058 (void) sd_event_source_set_description(m
->cgroups_agent_event_source
, "manager-cgroups-agent");
1064 static int manager_setup_user_lookup_fd(Manager
*m
) {
1069 /* Set up the socket pair used for passing UID/GID resolution results from forked off processes to PID
1070 * 1. Background: we can't do name lookups (NSS) from PID 1, since it might involve IPC and thus activation,
1071 * and we might hence deadlock on ourselves. Hence we do all user/group lookups asynchronously from the forked
1072 * off processes right before executing the binaries to start. In order to be able to clean up any IPC objects
1073 * created by a unit (see RemoveIPC=) we need to know in PID 1 the used UID/GID of the executed processes,
1074 * hence we establish this communication channel so that forked off processes can pass their UID/GID
1075 * information back to PID 1. The forked off processes send their resolved UID/GID to PID 1 in a simple
1076 * datagram, along with their unit name, so that we can share one communication socket pair among all units for
1079 * You might wonder why we need a communication channel for this that is independent of the usual notification
1080 * socket scheme (i.e. $NOTIFY_SOCKET). The primary difference is about trust: data sent via the $NOTIFY_SOCKET
1081 * channel is only accepted if it originates from the right unit and if reception was enabled for it. The user
1082 * lookup socket OTOH is only accessible by PID 1 and its children until they exec(), and always available.
1084 * Note that this function is called under two circumstances: when we first initialize (in which case we
1085 * allocate both the socket pair and the event source to listen on it), and when we deserialize after a reload
1086 * (in which case the socket pair already exists but we still need to allocate the event source for it). */
1088 if (m
->user_lookup_fds
[0] < 0) {
1090 /* Free all secondary fields */
1091 safe_close_pair(m
->user_lookup_fds
);
1092 m
->user_lookup_event_source
= sd_event_source_unref(m
->user_lookup_event_source
);
1094 if (socketpair(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
, 0, m
->user_lookup_fds
) < 0)
1095 return log_error_errno(errno
, "Failed to allocate user lookup socket: %m");
1097 (void) fd_inc_rcvbuf(m
->user_lookup_fds
[0], NOTIFY_RCVBUF_SIZE
);
1100 if (!m
->user_lookup_event_source
) {
1101 r
= sd_event_add_io(m
->event
, &m
->user_lookup_event_source
, m
->user_lookup_fds
[0], EPOLLIN
, manager_dispatch_user_lookup_fd
, m
);
1103 return log_error_errno(errno
, "Failed to allocate user lookup event source: %m");
1105 /* Process even earlier than the notify event source, so that we always know first about valid UID/GID
1107 r
= sd_event_source_set_priority(m
->user_lookup_event_source
, SD_EVENT_PRIORITY_NORMAL
-11);
1109 return log_error_errno(errno
, "Failed to set priority of user lookup event source: %m");
1111 (void) sd_event_source_set_description(m
->user_lookup_event_source
, "user-lookup");
1117 static unsigned manager_dispatch_cleanup_queue(Manager
*m
) {
1123 while ((u
= m
->cleanup_queue
)) {
1124 assert(u
->in_cleanup_queue
);
1134 GC_OFFSET_IN_PATH
, /* This one is on the path we were traveling */
1135 GC_OFFSET_UNSURE
, /* No clue */
1136 GC_OFFSET_GOOD
, /* We still need this unit */
1137 GC_OFFSET_BAD
, /* We don't need this unit anymore */
1141 static void unit_gc_mark_good(Unit
*u
, unsigned gc_marker
) {
1145 u
->gc_marker
= gc_marker
+ GC_OFFSET_GOOD
;
1147 /* Recursively mark referenced units as GOOD as well */
1148 HASHMAP_FOREACH_KEY(v
, other
, u
->dependencies
[UNIT_REFERENCES
])
1149 if (other
->gc_marker
== gc_marker
+ GC_OFFSET_UNSURE
)
1150 unit_gc_mark_good(other
, gc_marker
);
1153 static void unit_gc_sweep(Unit
*u
, unsigned gc_marker
) {
1160 if (IN_SET(u
->gc_marker
- gc_marker
,
1161 GC_OFFSET_GOOD
, GC_OFFSET_BAD
, GC_OFFSET_UNSURE
, GC_OFFSET_IN_PATH
))
1164 if (u
->in_cleanup_queue
)
1167 if (!unit_may_gc(u
))
1170 u
->gc_marker
= gc_marker
+ GC_OFFSET_IN_PATH
;
1174 HASHMAP_FOREACH_KEY(v
, other
, u
->dependencies
[UNIT_REFERENCED_BY
]) {
1175 unit_gc_sweep(other
, gc_marker
);
1177 if (other
->gc_marker
== gc_marker
+ GC_OFFSET_GOOD
)
1180 if (other
->gc_marker
!= gc_marker
+ GC_OFFSET_BAD
)
1185 LIST_FOREACH(refs_by_target
, ref
, u
->refs_by_target
) {
1186 unit_gc_sweep(ref
->source
, gc_marker
);
1188 if (ref
->source
->gc_marker
== gc_marker
+ GC_OFFSET_GOOD
)
1191 if (ref
->source
->gc_marker
!= gc_marker
+ GC_OFFSET_BAD
)
1198 /* We were unable to find anything out about this entry, so
1199 * let's investigate it later */
1200 u
->gc_marker
= gc_marker
+ GC_OFFSET_UNSURE
;
1201 unit_add_to_gc_queue(u
);
1205 /* We definitely know that this one is not useful anymore, so
1206 * let's mark it for deletion */
1207 u
->gc_marker
= gc_marker
+ GC_OFFSET_BAD
;
1208 unit_add_to_cleanup_queue(u
);
1212 unit_gc_mark_good(u
, gc_marker
);
1215 static unsigned manager_dispatch_gc_unit_queue(Manager
*m
) {
1216 unsigned n
= 0, gc_marker
;
1221 /* log_debug("Running GC..."); */
1223 m
->gc_marker
+= _GC_OFFSET_MAX
;
1224 if (m
->gc_marker
+ _GC_OFFSET_MAX
<= _GC_OFFSET_MAX
)
1227 gc_marker
= m
->gc_marker
;
1229 while ((u
= m
->gc_unit_queue
)) {
1230 assert(u
->in_gc_queue
);
1232 unit_gc_sweep(u
, gc_marker
);
1234 LIST_REMOVE(gc_queue
, m
->gc_unit_queue
, u
);
1235 u
->in_gc_queue
= false;
1239 if (IN_SET(u
->gc_marker
- gc_marker
,
1240 GC_OFFSET_BAD
, GC_OFFSET_UNSURE
)) {
1242 log_unit_debug(u
, "Collecting.");
1243 u
->gc_marker
= gc_marker
+ GC_OFFSET_BAD
;
1244 unit_add_to_cleanup_queue(u
);
1251 static unsigned manager_dispatch_gc_job_queue(Manager
*m
) {
1257 while ((j
= m
->gc_job_queue
)) {
1258 assert(j
->in_gc_queue
);
1260 LIST_REMOVE(gc_queue
, m
->gc_job_queue
, j
);
1261 j
->in_gc_queue
= false;
1268 log_unit_debug(j
->unit
, "Collecting job.");
1269 (void) job_finish_and_invalidate(j
, JOB_COLLECTED
, false, false);
1275 static unsigned manager_dispatch_stop_when_unneeded_queue(Manager
*m
) {
1282 while ((u
= m
->stop_when_unneeded_queue
)) {
1283 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
1285 assert(u
->in_stop_when_unneeded_queue
);
1286 LIST_REMOVE(stop_when_unneeded_queue
, m
->stop_when_unneeded_queue
, u
);
1287 u
->in_stop_when_unneeded_queue
= false;
1291 if (!unit_is_unneeded(u
))
1294 log_unit_debug(u
, "Unit is not needed anymore.");
1296 /* If stopping a unit fails continuously we might enter a stop loop here, hence stop acting on the
1297 * service being unnecessary after a while. */
1299 if (!ratelimit_below(&u
->auto_stop_ratelimit
)) {
1300 log_unit_warning(u
, "Unit not needed anymore, but not stopping since we tried this too often recently.");
1304 /* Ok, nobody needs us anymore. Sniff. Then let's commit suicide */
1305 r
= manager_add_job(u
->manager
, JOB_STOP
, u
, JOB_FAIL
, NULL
, &error
, NULL
);
1307 log_unit_warning_errno(u
, r
, "Failed to enqueue stop job, ignoring: %s", bus_error_message(&error
, r
));
1313 static void manager_clear_jobs_and_units(Manager
*m
) {
1318 while ((u
= hashmap_first(m
->units
)))
1321 manager_dispatch_cleanup_queue(m
);
1323 assert(!m
->load_queue
);
1324 assert(prioq_isempty(m
->run_queue
));
1325 assert(!m
->dbus_unit_queue
);
1326 assert(!m
->dbus_job_queue
);
1327 assert(!m
->cleanup_queue
);
1328 assert(!m
->gc_unit_queue
);
1329 assert(!m
->gc_job_queue
);
1330 assert(!m
->stop_when_unneeded_queue
);
1332 assert(hashmap_isempty(m
->jobs
));
1333 assert(hashmap_isempty(m
->units
));
1335 m
->n_on_console
= 0;
1336 m
->n_running_jobs
= 0;
1337 m
->n_installed_jobs
= 0;
1338 m
->n_failed_jobs
= 0;
1341 Manager
* manager_free(Manager
*m
) {
1345 manager_clear_jobs_and_units(m
);
1347 for (UnitType c
= 0; c
< _UNIT_TYPE_MAX
; c
++)
1348 if (unit_vtable
[c
]->shutdown
)
1349 unit_vtable
[c
]->shutdown(m
);
1351 /* Keep the cgroup hierarchy in place except when we know we are going down for good */
1352 manager_shutdown_cgroup(m
, IN_SET(m
->objective
, MANAGER_EXIT
, MANAGER_REBOOT
, MANAGER_POWEROFF
, MANAGER_HALT
, MANAGER_KEXEC
));
1354 lookup_paths_flush_generator(&m
->lookup_paths
);
1357 manager_varlink_done(m
);
1359 exec_runtime_vacuum(m
);
1360 hashmap_free(m
->exec_runtime_by_id
);
1362 dynamic_user_vacuum(m
, false);
1363 hashmap_free(m
->dynamic_users
);
1365 hashmap_free(m
->units
);
1366 hashmap_free(m
->units_by_invocation_id
);
1367 hashmap_free(m
->jobs
);
1368 hashmap_free(m
->watch_pids
);
1369 hashmap_free(m
->watch_bus
);
1371 prioq_free(m
->run_queue
);
1373 set_free(m
->startup_units
);
1374 set_free(m
->failed_units
);
1376 sd_event_source_unref(m
->signal_event_source
);
1377 sd_event_source_unref(m
->sigchld_event_source
);
1378 sd_event_source_unref(m
->notify_event_source
);
1379 sd_event_source_unref(m
->cgroups_agent_event_source
);
1380 sd_event_source_unref(m
->time_change_event_source
);
1381 sd_event_source_unref(m
->timezone_change_event_source
);
1382 sd_event_source_unref(m
->jobs_in_progress_event_source
);
1383 sd_event_source_unref(m
->run_queue_event_source
);
1384 sd_event_source_unref(m
->user_lookup_event_source
);
1386 safe_close(m
->signal_fd
);
1387 safe_close(m
->notify_fd
);
1388 safe_close(m
->cgroups_agent_fd
);
1389 safe_close(m
->time_change_fd
);
1390 safe_close_pair(m
->user_lookup_fds
);
1392 manager_close_ask_password(m
);
1394 manager_close_idle_pipe(m
);
1396 sd_event_unref(m
->event
);
1398 free(m
->notify_socket
);
1400 lookup_paths_free(&m
->lookup_paths
);
1401 strv_free(m
->transient_environment
);
1402 strv_free(m
->client_environment
);
1404 hashmap_free(m
->cgroup_unit
);
1405 manager_free_unit_name_maps(m
);
1407 free(m
->switch_root
);
1408 free(m
->switch_root_init
);
1410 rlimit_free_all(m
->rlimit
);
1412 assert(hashmap_isempty(m
->units_requiring_mounts_for
));
1413 hashmap_free(m
->units_requiring_mounts_for
);
1415 hashmap_free(m
->uid_refs
);
1416 hashmap_free(m
->gid_refs
);
1418 for (ExecDirectoryType dt
= 0; dt
< _EXEC_DIRECTORY_TYPE_MAX
; dt
++)
1419 m
->prefix
[dt
] = mfree(m
->prefix
[dt
]);
1420 free(m
->received_credentials
);
1425 static void manager_enumerate_perpetual(Manager
*m
) {
1428 if (m
->test_run_flags
== MANAGER_TEST_RUN_MINIMAL
)
1431 /* Let's ask every type to load all units from disk/kernel that it might know */
1432 for (UnitType c
= 0; c
< _UNIT_TYPE_MAX
; c
++) {
1433 if (!unit_type_supported(c
)) {
1434 log_debug("Unit type .%s is not supported on this system.", unit_type_to_string(c
));
1438 if (unit_vtable
[c
]->enumerate_perpetual
)
1439 unit_vtable
[c
]->enumerate_perpetual(m
);
1443 static void manager_enumerate(Manager
*m
) {
1446 if (m
->test_run_flags
== MANAGER_TEST_RUN_MINIMAL
)
1449 /* Let's ask every type to load all units from disk/kernel that it might know */
1450 for (UnitType c
= 0; c
< _UNIT_TYPE_MAX
; c
++) {
1451 if (!unit_type_supported(c
)) {
1452 log_debug("Unit type .%s is not supported on this system.", unit_type_to_string(c
));
1456 if (unit_vtable
[c
]->enumerate
)
1457 unit_vtable
[c
]->enumerate(m
);
1460 manager_dispatch_load_queue(m
);
1463 static void manager_coldplug(Manager
*m
) {
1470 log_debug("Invoking unit coldplug() handlers…");
1472 /* Let's place the units back into their deserialized state */
1473 HASHMAP_FOREACH_KEY(u
, k
, m
->units
) {
1475 /* ignore aliases */
1479 r
= unit_coldplug(u
);
1481 log_warning_errno(r
, "We couldn't coldplug %s, proceeding anyway: %m", u
->id
);
1485 static void manager_catchup(Manager
*m
) {
1491 log_debug("Invoking unit catchup() handlers…");
1493 /* Let's catch up on any state changes that happened while we were reloading/reexecing */
1494 HASHMAP_FOREACH_KEY(u
, k
, m
->units
) {
1496 /* ignore aliases */
1504 static void manager_distribute_fds(Manager
*m
, FDSet
*fds
) {
1509 HASHMAP_FOREACH(u
, m
->units
) {
1511 if (fdset_size(fds
) <= 0)
1514 if (!UNIT_VTABLE(u
)->distribute_fds
)
1517 UNIT_VTABLE(u
)->distribute_fds(u
, fds
);
1521 static bool manager_dbus_is_running(Manager
*m
, bool deserialized
) {
1526 /* This checks whether the dbus instance we are supposed to expose our APIs on is up. We check both the socket
1527 * and the service unit. If the 'deserialized' parameter is true we'll check the deserialized state of the unit
1528 * rather than the current one. */
1530 if (MANAGER_IS_TEST_RUN(m
))
1533 u
= manager_get_unit(m
, SPECIAL_DBUS_SOCKET
);
1536 if ((deserialized
? SOCKET(u
)->deserialized_state
: SOCKET(u
)->state
) != SOCKET_RUNNING
)
1539 u
= manager_get_unit(m
, SPECIAL_DBUS_SERVICE
);
1542 if (!IN_SET((deserialized
? SERVICE(u
)->deserialized_state
: SERVICE(u
)->state
), SERVICE_RUNNING
, SERVICE_RELOAD
))
1548 static void manager_setup_bus(Manager
*m
) {
1551 /* Let's set up our private bus connection now, unconditionally */
1552 (void) bus_init_private(m
);
1554 /* If we are in --user mode also connect to the system bus now */
1555 if (MANAGER_IS_USER(m
))
1556 (void) bus_init_system(m
);
1558 /* Let's connect to the bus now, but only if the unit is supposed to be up */
1559 if (manager_dbus_is_running(m
, MANAGER_IS_RELOADING(m
))) {
1560 (void) bus_init_api(m
);
1562 if (MANAGER_IS_SYSTEM(m
))
1563 (void) bus_init_system(m
);
1567 static void manager_preset_all(Manager
*m
) {
1572 if (m
->first_boot
<= 0)
1575 if (!MANAGER_IS_SYSTEM(m
))
1578 if (MANAGER_IS_TEST_RUN(m
))
1581 /* If this is the first boot, and we are in the host system, then preset everything */
1582 r
= unit_file_preset_all(UNIT_FILE_SYSTEM
, 0, NULL
, UNIT_FILE_PRESET_ENABLE_ONLY
, NULL
, 0);
1584 log_full_errno(r
== -EEXIST
? LOG_NOTICE
: LOG_WARNING
, r
,
1585 "Failed to populate /etc with preset unit settings, ignoring: %m");
1587 log_info("Populated /etc with preset unit settings.");
1590 static void manager_ready(Manager
*m
) {
1593 /* After having loaded everything, do the final round of catching up with what might have changed */
1595 m
->objective
= MANAGER_OK
; /* Tell everyone we are up now */
1597 /* It might be safe to log to the journal now and connect to dbus */
1598 manager_recheck_journal(m
);
1599 manager_recheck_dbus(m
);
1601 /* Let's finally catch up with any changes that took place while we were reloading/reexecing */
1604 m
->honor_device_enumeration
= true;
1607 static Manager
* manager_reloading_start(Manager
*m
) {
1611 static void manager_reloading_stopp(Manager
**m
) {
1613 assert((*m
)->n_reloading
> 0);
1614 (*m
)->n_reloading
--;
1618 int manager_startup(Manager
*m
, FILE *serialization
, FDSet
*fds
) {
1623 /* If we are running in test mode, we still want to run the generators,
1624 * but we should not touch the real generator directories. */
1625 r
= lookup_paths_init(&m
->lookup_paths
, m
->unit_file_scope
,
1626 MANAGER_IS_TEST_RUN(m
) ? LOOKUP_PATHS_TEMPORARY_GENERATED
: 0,
1629 return log_error_errno(r
, "Failed to initialize path lookup table: %m");
1631 dual_timestamp_get(m
->timestamps
+ manager_timestamp_initrd_mangle(MANAGER_TIMESTAMP_GENERATORS_START
));
1632 r
= manager_run_environment_generators(m
);
1634 r
= manager_run_generators(m
);
1635 dual_timestamp_get(m
->timestamps
+ manager_timestamp_initrd_mangle(MANAGER_TIMESTAMP_GENERATORS_FINISH
));
1639 manager_preset_all(m
);
1641 lookup_paths_log(&m
->lookup_paths
);
1644 /* This block is (optionally) done with the reloading counter bumped */
1645 _cleanup_(manager_reloading_stopp
) Manager
*reloading
= NULL
;
1647 /* If we will deserialize make sure that during enumeration this is already known, so we increase the
1648 * counter here already */
1650 reloading
= manager_reloading_start(m
);
1652 /* First, enumerate what we can from all config files */
1653 dual_timestamp_get(m
->timestamps
+ manager_timestamp_initrd_mangle(MANAGER_TIMESTAMP_UNITS_LOAD_START
));
1654 manager_enumerate_perpetual(m
);
1655 manager_enumerate(m
);
1656 dual_timestamp_get(m
->timestamps
+ manager_timestamp_initrd_mangle(MANAGER_TIMESTAMP_UNITS_LOAD_FINISH
));
1658 /* Second, deserialize if there is something to deserialize */
1659 if (serialization
) {
1660 r
= manager_deserialize(m
, serialization
, fds
);
1662 return log_error_errno(r
, "Deserialization failed: %m");
1665 /* Any fds left? Find some unit which wants them. This is useful to allow container managers to pass
1666 * some file descriptors to us pre-initialized. This enables socket-based activation of entire
1668 manager_distribute_fds(m
, fds
);
1670 /* We might have deserialized the notify fd, but if we didn't then let's create the bus now */
1671 r
= manager_setup_notify(m
);
1673 /* No sense to continue without notifications, our children would fail anyway. */
1676 r
= manager_setup_cgroups_agent(m
);
1678 /* Likewise, no sense to continue without empty cgroup notifications. */
1681 r
= manager_setup_user_lookup_fd(m
);
1683 /* This shouldn't fail, except if things are really broken. */
1686 /* Connect to the bus if we are good for it */
1687 manager_setup_bus(m
);
1689 /* Now that we are connected to all possible buses, let's deserialize who is tracking us. */
1690 r
= bus_track_coldplug(m
, &m
->subscribed
, false, m
->deserialized_subscribed
);
1692 log_warning_errno(r
, "Failed to deserialized tracked clients, ignoring: %m");
1693 m
->deserialized_subscribed
= strv_free(m
->deserialized_subscribed
);
1695 r
= manager_varlink_init(m
);
1697 log_warning_errno(r
, "Failed to set up Varlink server, ignoring: %m");
1699 /* Third, fire things up! */
1700 manager_coldplug(m
);
1702 /* Clean up runtime objects */
1706 /* Let's wait for the UnitNew/JobNew messages being sent, before we notify that the
1707 * reload is finished */
1708 m
->send_reloading_done
= true;
1716 int manager_add_job(
1722 sd_bus_error
*error
,
1729 assert(type
< _JOB_TYPE_MAX
);
1731 assert(mode
< _JOB_MODE_MAX
);
1733 if (mode
== JOB_ISOLATE
&& type
!= JOB_START
)
1734 return sd_bus_error_set(error
, SD_BUS_ERROR_INVALID_ARGS
, "Isolate is only valid for start.");
1736 if (mode
== JOB_ISOLATE
&& !unit
->allow_isolate
)
1737 return sd_bus_error_set(error
, BUS_ERROR_NO_ISOLATION
, "Operation refused, unit may not be isolated.");
1739 if (mode
== JOB_TRIGGERING
&& type
!= JOB_STOP
)
1740 return sd_bus_error_set(error
, SD_BUS_ERROR_INVALID_ARGS
, "--job-mode=triggering is only valid for stop.");
1742 log_unit_debug(unit
, "Trying to enqueue job %s/%s/%s", unit
->id
, job_type_to_string(type
), job_mode_to_string(mode
));
1744 type
= job_type_collapse(type
, unit
);
1746 tr
= transaction_new(mode
== JOB_REPLACE_IRREVERSIBLY
);
1750 r
= transaction_add_job_and_dependencies(tr
, type
, unit
, NULL
, true, false,
1751 IN_SET(mode
, JOB_IGNORE_DEPENDENCIES
, JOB_IGNORE_REQUIREMENTS
),
1752 mode
== JOB_IGNORE_DEPENDENCIES
, error
);
1756 if (mode
== JOB_ISOLATE
) {
1757 r
= transaction_add_isolate_jobs(tr
, m
);
1762 if (mode
== JOB_TRIGGERING
) {
1763 r
= transaction_add_triggering_jobs(tr
, unit
);
1768 r
= transaction_activate(tr
, m
, mode
, affected_jobs
, error
);
1772 log_unit_debug(unit
,
1773 "Enqueued job %s/%s as %u", unit
->id
,
1774 job_type_to_string(type
), (unsigned) tr
->anchor_job
->id
);
1777 *ret
= tr
->anchor_job
;
1779 transaction_free(tr
);
1783 transaction_abort(tr
);
1784 transaction_free(tr
);
1788 int manager_add_job_by_name(Manager
*m
, JobType type
, const char *name
, JobMode mode
, Set
*affected_jobs
, sd_bus_error
*e
, Job
**ret
) {
1789 Unit
*unit
= NULL
; /* just to appease gcc, initialization is not really necessary */
1793 assert(type
< _JOB_TYPE_MAX
);
1795 assert(mode
< _JOB_MODE_MAX
);
1797 r
= manager_load_unit(m
, name
, NULL
, NULL
, &unit
);
1802 return manager_add_job(m
, type
, unit
, mode
, affected_jobs
, e
, ret
);
1805 int manager_add_job_by_name_and_warn(Manager
*m
, JobType type
, const char *name
, JobMode mode
, Set
*affected_jobs
, Job
**ret
) {
1806 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
1810 assert(type
< _JOB_TYPE_MAX
);
1812 assert(mode
< _JOB_MODE_MAX
);
1814 r
= manager_add_job_by_name(m
, type
, name
, mode
, affected_jobs
, &error
, ret
);
1816 return log_warning_errno(r
, "Failed to enqueue %s job for %s: %s", job_mode_to_string(mode
), name
, bus_error_message(&error
, r
));
1821 int manager_propagate_reload(Manager
*m
, Unit
*unit
, JobMode mode
, sd_bus_error
*e
) {
1827 assert(mode
< _JOB_MODE_MAX
);
1828 assert(mode
!= JOB_ISOLATE
); /* Isolate is only valid for start */
1830 tr
= transaction_new(mode
== JOB_REPLACE_IRREVERSIBLY
);
1834 /* We need an anchor job */
1835 r
= transaction_add_job_and_dependencies(tr
, JOB_NOP
, unit
, NULL
, false, false, true, true, e
);
1839 /* Failure in adding individual dependencies is ignored, so this always succeeds. */
1840 transaction_add_propagate_reload_jobs(tr
, unit
, tr
->anchor_job
, mode
== JOB_IGNORE_DEPENDENCIES
, e
);
1842 r
= transaction_activate(tr
, m
, mode
, NULL
, e
);
1846 transaction_free(tr
);
1850 transaction_abort(tr
);
1851 transaction_free(tr
);
1855 Job
*manager_get_job(Manager
*m
, uint32_t id
) {
1858 return hashmap_get(m
->jobs
, UINT32_TO_PTR(id
));
1861 Unit
*manager_get_unit(Manager
*m
, const char *name
) {
1865 return hashmap_get(m
->units
, name
);
1868 static int manager_dispatch_target_deps_queue(Manager
*m
) {
1872 static const UnitDependency deps
[] = {
1881 while ((u
= m
->target_deps_queue
)) {
1882 assert(u
->in_target_deps_queue
);
1884 LIST_REMOVE(target_deps_queue
, u
->manager
->target_deps_queue
, u
);
1885 u
->in_target_deps_queue
= false;
1887 for (size_t k
= 0; k
< ELEMENTSOF(deps
); k
++) {
1891 HASHMAP_FOREACH_KEY(v
, target
, u
->dependencies
[deps
[k
]]) {
1892 r
= unit_add_default_target_dependency(u
, target
);
1902 unsigned manager_dispatch_load_queue(Manager
*m
) {
1908 /* Make sure we are not run recursively */
1909 if (m
->dispatching_load_queue
)
1912 m
->dispatching_load_queue
= true;
1914 /* Dispatches the load queue. Takes a unit from the queue and
1915 * tries to load its data until the queue is empty */
1917 while ((u
= m
->load_queue
)) {
1918 assert(u
->in_load_queue
);
1924 m
->dispatching_load_queue
= false;
1926 /* Dispatch the units waiting for their target dependencies to be added now, as all targets that we know about
1927 * should be loaded and have aliases resolved */
1928 (void) manager_dispatch_target_deps_queue(m
);
1933 bool manager_unit_cache_should_retry_load(Unit
*u
) {
1936 /* Automatic reloading from disk only applies to units which were not found sometime in the past, and
1937 * the not-found stub is kept pinned in the unit graph by dependencies. For units that were
1938 * previously loaded, we don't do automatic reloading, and daemon-reload is necessary to update. */
1939 if (u
->load_state
!= UNIT_NOT_FOUND
)
1942 /* The cache has been updated since the last time we tried to load the unit. There might be new
1943 * fragment paths to read. */
1944 if (u
->manager
->unit_cache_timestamp_hash
!= u
->fragment_not_found_timestamp_hash
)
1947 /* The cache needs to be updated because there are modifications on disk. */
1948 return !lookup_paths_timestamp_hash_same(&u
->manager
->lookup_paths
, u
->manager
->unit_cache_timestamp_hash
, NULL
);
1951 int manager_load_unit_prepare(
1958 _cleanup_(unit_freep
) Unit
*cleanup_ret
= NULL
;
1966 /* This will prepare the unit for loading, but not actually load anything from disk. */
1968 if (path
&& !path_is_absolute(path
))
1969 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
, "Path %s is not absolute.", path
);
1972 /* 'name' and 'path' must not both be null. Check here 'path' using assert_se() to
1973 * workaround a bug in gcc that generates a -Wnonnull warning when calling basename(),
1974 * but this cannot be possible in any code path (See #6119). */
1976 name
= basename(path
);
1979 t
= unit_name_to_type(name
);
1981 if (t
== _UNIT_TYPE_INVALID
|| !unit_name_is_valid(name
, UNIT_NAME_PLAIN
|UNIT_NAME_INSTANCE
)) {
1982 if (unit_name_is_valid(name
, UNIT_NAME_TEMPLATE
))
1983 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
, "Unit name %s is missing the instance name.", name
);
1985 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
, "Unit name %s is not valid.", name
);
1988 ret
= manager_get_unit(m
, name
);
1990 /* The time-based cache allows to start new units without daemon-reload,
1991 * but if they are already referenced (because of dependencies or ordering)
1992 * then we have to force a load of the fragment. As an optimization, check
1993 * first if anything in the usual paths was modified since the last time
1994 * the cache was loaded. Also check if the last time an attempt to load the
1995 * unit was made was before the most recent cache refresh, so that we know
1996 * we need to try again — even if the cache is current, it might have been
1997 * updated in a different context before we had a chance to retry loading
1998 * this particular unit. */
1999 if (manager_unit_cache_should_retry_load(ret
))
2000 ret
->load_state
= UNIT_STUB
;
2006 ret
= cleanup_ret
= unit_new(m
, unit_vtable
[t
]->object_size
);
2012 r
= free_and_strdup(&ret
->fragment_path
, path
);
2017 r
= unit_add_name(ret
, name
);
2021 unit_add_to_load_queue(ret
);
2022 unit_add_to_dbus_queue(ret
);
2023 unit_add_to_gc_queue(ret
);
2031 int manager_load_unit(
2043 /* This will load the service information files, but not actually
2044 * start any services or anything. */
2046 r
= manager_load_unit_prepare(m
, name
, path
, e
, _ret
);
2050 manager_dispatch_load_queue(m
);
2052 *_ret
= unit_follow_merge(*_ret
);
2056 int manager_load_startable_unit_or_warn(
2062 /* Load a unit, make sure it loaded fully and is not masked. */
2064 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
2068 r
= manager_load_unit(m
, name
, path
, &error
, &unit
);
2070 return log_error_errno(r
, "Failed to load %s %s: %s",
2071 name
? "unit" : "unit file", name
?: path
,
2072 bus_error_message(&error
, r
));
2074 r
= bus_unit_validate_load_state(unit
, &error
);
2076 return log_error_errno(r
, "%s", bus_error_message(&error
, r
));
2082 void manager_dump_jobs(Manager
*s
, FILE *f
, const char *prefix
) {
2088 HASHMAP_FOREACH(j
, s
->jobs
)
2089 job_dump(j
, f
, prefix
);
2092 void manager_dump_units(Manager
*s
, FILE *f
, const char *prefix
) {
2099 HASHMAP_FOREACH_KEY(u
, t
, s
->units
)
2101 unit_dump(u
, f
, prefix
);
2104 void manager_dump(Manager
*m
, FILE *f
, const char *prefix
) {
2108 for (ManagerTimestamp q
= 0; q
< _MANAGER_TIMESTAMP_MAX
; q
++) {
2109 const dual_timestamp
*t
= m
->timestamps
+ q
;
2110 char buf
[CONST_MAX(FORMAT_TIMESPAN_MAX
, FORMAT_TIMESTAMP_MAX
)];
2112 if (dual_timestamp_is_set(t
))
2113 fprintf(f
, "%sTimestamp %s: %s\n",
2115 manager_timestamp_to_string(q
),
2116 timestamp_is_set(t
->realtime
) ? format_timestamp(buf
, sizeof buf
, t
->realtime
) :
2117 format_timespan(buf
, sizeof buf
, t
->monotonic
, 1));
2120 manager_dump_units(m
, f
, prefix
);
2121 manager_dump_jobs(m
, f
, prefix
);
2124 int manager_get_dump_string(Manager
*m
, char **ret
) {
2125 _cleanup_free_
char *dump
= NULL
;
2126 _cleanup_fclose_
FILE *f
= NULL
;
2133 f
= open_memstream_unlocked(&dump
, &size
);
2137 manager_dump(m
, f
, NULL
);
2139 r
= fflush_and_check(f
);
2145 *ret
= TAKE_PTR(dump
);
2150 void manager_clear_jobs(Manager
*m
) {
2155 while ((j
= hashmap_first(m
->jobs
)))
2156 /* No need to recurse. We're cancelling all jobs. */
2157 job_finish_and_invalidate(j
, JOB_CANCELED
, false, false);
2160 void manager_unwatch_pid(Manager
*m
, pid_t pid
) {
2163 /* First let's drop the unit keyed as "pid". */
2164 (void) hashmap_remove(m
->watch_pids
, PID_TO_PTR(pid
));
2166 /* Then, let's also drop the array keyed by -pid. */
2167 free(hashmap_remove(m
->watch_pids
, PID_TO_PTR(-pid
)));
2170 static int manager_dispatch_run_queue(sd_event_source
*source
, void *userdata
) {
2171 Manager
*m
= userdata
;
2177 while ((j
= prioq_peek(m
->run_queue
))) {
2178 assert(j
->installed
);
2179 assert(j
->in_run_queue
);
2181 (void) job_run_and_invalidate(j
);
2184 if (m
->n_running_jobs
> 0)
2185 manager_watch_jobs_in_progress(m
);
2187 if (m
->n_on_console
> 0)
2188 manager_watch_idle_pipe(m
);
2193 static unsigned manager_dispatch_dbus_queue(Manager
*m
) {
2194 unsigned n
= 0, budget
;
2200 /* When we are reloading, let's not wait with generating signals, since we need to exit the manager as quickly
2201 * as we can. There's no point in throttling generation of signals in that case. */
2202 if (MANAGER_IS_RELOADING(m
) || m
->send_reloading_done
|| m
->pending_reload_message
)
2203 budget
= UINT_MAX
; /* infinite budget in this case */
2205 /* Anything to do at all? */
2206 if (!m
->dbus_unit_queue
&& !m
->dbus_job_queue
)
2209 /* Do we have overly many messages queued at the moment? If so, let's not enqueue more on top, let's
2210 * sit this cycle out, and process things in a later cycle when the queues got a bit emptier. */
2211 if (manager_bus_n_queued_write(m
) > MANAGER_BUS_BUSY_THRESHOLD
)
2214 /* Only process a certain number of units/jobs per event loop iteration. Even if the bus queue wasn't
2215 * overly full before this call we shouldn't increase it in size too wildly in one step, and we
2216 * shouldn't monopolize CPU time with generating these messages. Note the difference in counting of
2217 * this "budget" and the "threshold" above: the "budget" is decreased only once per generated message,
2218 * regardless how many buses/direct connections it is enqueued on, while the "threshold" is applied to
2219 * each queued instance of bus message, i.e. if the same message is enqueued to five buses/direct
2220 * connections it will be counted five times. This difference in counting ("references"
2221 * vs. "instances") is primarily a result of the fact that it's easier to implement it this way,
2222 * however it also reflects the thinking that the "threshold" should put a limit on used queue memory,
2223 * i.e. space, while the "budget" should put a limit on time. Also note that the "threshold" is
2224 * currently chosen much higher than the "budget". */
2225 budget
= MANAGER_BUS_MESSAGE_BUDGET
;
2228 while (budget
!= 0 && (u
= m
->dbus_unit_queue
)) {
2230 assert(u
->in_dbus_queue
);
2232 bus_unit_send_change_signal(u
);
2235 if (budget
!= UINT_MAX
)
2239 while (budget
!= 0 && (j
= m
->dbus_job_queue
)) {
2240 assert(j
->in_dbus_queue
);
2242 bus_job_send_change_signal(j
);
2245 if (budget
!= UINT_MAX
)
2249 if (m
->send_reloading_done
) {
2250 m
->send_reloading_done
= false;
2251 bus_manager_send_reloading(m
, false);
2255 if (m
->pending_reload_message
) {
2256 bus_send_pending_reload_message(m
);
2263 static int manager_dispatch_cgroups_agent_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
2264 Manager
*m
= userdata
;
2268 n
= recv(fd
, buf
, sizeof(buf
), 0);
2270 return log_error_errno(errno
, "Failed to read cgroups agent message: %m");
2272 log_error("Got zero-length cgroups agent message, ignoring.");
2275 if ((size_t) n
>= sizeof(buf
)) {
2276 log_error("Got overly long cgroups agent message, ignoring.");
2280 if (memchr(buf
, 0, n
)) {
2281 log_error("Got cgroups agent message with embedded NUL byte, ignoring.");
2286 manager_notify_cgroup_empty(m
, buf
);
2287 (void) bus_forward_agent_released(m
, buf
);
2292 static bool manager_process_barrier_fd(char * const *tags
, FDSet
*fds
) {
2294 /* nothing else must be sent when using BARRIER=1 */
2295 if (strv_contains(tags
, "BARRIER=1")) {
2296 if (strv_length(tags
) == 1) {
2297 if (fdset_size(fds
) != 1)
2298 log_warning("Got incorrect number of fds with BARRIER=1, closing them.");
2300 log_warning("Extra notification messages sent with BARRIER=1, ignoring everything.");
2302 /* Drop the message if BARRIER=1 was found */
2309 static void manager_invoke_notify_message(
2312 const struct ucred
*ucred
,
2321 if (u
->notifygen
== m
->notifygen
) /* Already invoked on this same unit in this same iteration? */
2323 u
->notifygen
= m
->notifygen
;
2325 if (UNIT_VTABLE(u
)->notify_message
)
2326 UNIT_VTABLE(u
)->notify_message(u
, ucred
, tags
, fds
);
2328 else if (DEBUG_LOGGING
) {
2329 _cleanup_free_
char *buf
= NULL
, *x
= NULL
, *y
= NULL
;
2331 buf
= strv_join(tags
, ", ");
2333 x
= ellipsize(buf
, 20, 90);
2337 log_unit_debug(u
, "Got notification message \"%s\", ignoring.", strnull(y
));
2341 static int manager_dispatch_notify_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
2343 _cleanup_fdset_free_ FDSet
*fds
= NULL
;
2344 Manager
*m
= userdata
;
2345 char buf
[NOTIFY_BUFFER_MAX
+1];
2346 struct iovec iovec
= {
2348 .iov_len
= sizeof(buf
)-1,
2350 CMSG_BUFFER_TYPE(CMSG_SPACE(sizeof(struct ucred
)) +
2351 CMSG_SPACE(sizeof(int) * NOTIFY_FD_MAX
)) control
;
2352 struct msghdr msghdr
= {
2355 .msg_control
= &control
,
2356 .msg_controllen
= sizeof(control
),
2359 struct cmsghdr
*cmsg
;
2360 struct ucred
*ucred
= NULL
;
2361 _cleanup_free_ Unit
**array_copy
= NULL
;
2362 _cleanup_strv_free_
char **tags
= NULL
;
2363 Unit
*u1
, *u2
, **array
;
2364 int r
, *fd_array
= NULL
;
2370 assert(m
->notify_fd
== fd
);
2372 if (revents
!= EPOLLIN
) {
2373 log_warning("Got unexpected poll event for notify fd.");
2377 n
= recvmsg_safe(m
->notify_fd
, &msghdr
, MSG_DONTWAIT
|MSG_CMSG_CLOEXEC
|MSG_TRUNC
);
2378 if (IN_SET(n
, -EAGAIN
, -EINTR
))
2379 return 0; /* Spurious wakeup, try again */
2381 log_warning("Got message with truncated control data (too many fds sent?), ignoring.");
2385 /* If this is any other, real error, then let's stop processing this socket. This of course
2386 * means we won't take notification messages anymore, but that's still better than busy
2387 * looping around this: being woken up over and over again but being unable to actually read
2388 * the message off the socket. */
2389 return log_error_errno(n
, "Failed to receive notification message: %m");
2391 CMSG_FOREACH(cmsg
, &msghdr
) {
2392 if (cmsg
->cmsg_level
== SOL_SOCKET
&& cmsg
->cmsg_type
== SCM_RIGHTS
) {
2395 fd_array
= (int*) CMSG_DATA(cmsg
);
2396 n_fds
= (cmsg
->cmsg_len
- CMSG_LEN(0)) / sizeof(int);
2398 } else if (cmsg
->cmsg_level
== SOL_SOCKET
&&
2399 cmsg
->cmsg_type
== SCM_CREDENTIALS
&&
2400 cmsg
->cmsg_len
== CMSG_LEN(sizeof(struct ucred
))) {
2403 ucred
= (struct ucred
*) CMSG_DATA(cmsg
);
2410 r
= fdset_new_array(&fds
, fd_array
, n_fds
);
2412 close_many(fd_array
, n_fds
);
2418 if (!ucred
|| !pid_is_valid(ucred
->pid
)) {
2419 log_warning("Received notify message without valid credentials. Ignoring.");
2423 if ((size_t) n
>= sizeof(buf
) || (msghdr
.msg_flags
& MSG_TRUNC
)) {
2424 log_warning("Received notify message exceeded maximum size. Ignoring.");
2428 /* As extra safety check, let's make sure the string we get doesn't contain embedded NUL bytes. We permit one
2429 * trailing NUL byte in the message, but don't expect it. */
2430 if (n
> 1 && memchr(buf
, 0, n
-1)) {
2431 log_warning("Received notify message with embedded NUL bytes. Ignoring.");
2435 /* Make sure it's NUL-terminated, then parse it to obtain the tags list */
2437 tags
= strv_split_newlines(buf
);
2443 /* possibly a barrier fd, let's see */
2444 if (manager_process_barrier_fd(tags
, fds
))
2447 /* Increase the generation counter used for filtering out duplicate unit invocations. */
2450 /* Notify every unit that might be interested, which might be multiple. */
2451 u1
= manager_get_unit_by_pid_cgroup(m
, ucred
->pid
);
2452 u2
= hashmap_get(m
->watch_pids
, PID_TO_PTR(ucred
->pid
));
2453 array
= hashmap_get(m
->watch_pids
, PID_TO_PTR(-ucred
->pid
));
2460 array_copy
= newdup(Unit
*, array
, k
+1);
2464 /* And now invoke the per-unit callbacks. Note that manager_invoke_notify_message() will handle duplicate units
2465 * make sure we only invoke each unit's handler once. */
2467 manager_invoke_notify_message(m
, u1
, ucred
, tags
, fds
);
2471 manager_invoke_notify_message(m
, u2
, ucred
, tags
, fds
);
2475 for (size_t i
= 0; array_copy
[i
]; i
++) {
2476 manager_invoke_notify_message(m
, array_copy
[i
], ucred
, tags
, fds
);
2481 log_warning("Cannot find unit for notify message of PID "PID_FMT
", ignoring.", ucred
->pid
);
2483 if (fdset_size(fds
) > 0)
2484 log_warning("Got extra auxiliary fds with notification message, closing them.");
2489 static void manager_invoke_sigchld_event(
2492 const siginfo_t
*si
) {
2498 /* Already invoked the handler of this unit in this iteration? Then don't process this again */
2499 if (u
->sigchldgen
== m
->sigchldgen
)
2501 u
->sigchldgen
= m
->sigchldgen
;
2503 log_unit_debug(u
, "Child "PID_FMT
" belongs to %s.", si
->si_pid
, u
->id
);
2504 unit_unwatch_pid(u
, si
->si_pid
);
2506 if (UNIT_VTABLE(u
)->sigchld_event
)
2507 UNIT_VTABLE(u
)->sigchld_event(u
, si
->si_pid
, si
->si_code
, si
->si_status
);
2510 static int manager_dispatch_sigchld(sd_event_source
*source
, void *userdata
) {
2511 Manager
*m
= userdata
;
2518 /* First we call waitid() for a PID and do not reap the zombie. That way we can still access /proc/$PID for it
2519 * while it is a zombie. */
2521 if (waitid(P_ALL
, 0, &si
, WEXITED
|WNOHANG
|WNOWAIT
) < 0) {
2523 if (errno
!= ECHILD
)
2524 log_error_errno(errno
, "Failed to peek for child with waitid(), ignoring: %m");
2532 if (IN_SET(si
.si_code
, CLD_EXITED
, CLD_KILLED
, CLD_DUMPED
)) {
2533 _cleanup_free_ Unit
**array_copy
= NULL
;
2534 _cleanup_free_
char *name
= NULL
;
2535 Unit
*u1
, *u2
, **array
;
2537 (void) get_process_comm(si
.si_pid
, &name
);
2539 log_debug("Child "PID_FMT
" (%s) died (code=%s, status=%i/%s)",
2540 si
.si_pid
, strna(name
),
2541 sigchld_code_to_string(si
.si_code
),
2543 strna(si
.si_code
== CLD_EXITED
2544 ? exit_status_to_string(si
.si_status
, EXIT_STATUS_FULL
)
2545 : signal_to_string(si
.si_status
)));
2547 /* Increase the generation counter used for filtering out duplicate unit invocations */
2550 /* And now figure out the unit this belongs to, it might be multiple... */
2551 u1
= manager_get_unit_by_pid_cgroup(m
, si
.si_pid
);
2552 u2
= hashmap_get(m
->watch_pids
, PID_TO_PTR(si
.si_pid
));
2553 array
= hashmap_get(m
->watch_pids
, PID_TO_PTR(-si
.si_pid
));
2557 /* Count how many entries the array has */
2561 /* Make a copy of the array so that we don't trip up on the array changing beneath us */
2562 array_copy
= newdup(Unit
*, array
, n
+1);
2567 /* Finally, execute them all. Note that u1, u2 and the array might contain duplicates, but
2568 * that's fine, manager_invoke_sigchld_event() will ensure we only invoke the handlers once for
2569 * each iteration. */
2571 /* We check for oom condition, in case we got SIGCHLD before the oom notification.
2572 * We only do this for the cgroup the PID belonged to. */
2573 (void) unit_check_oom(u1
);
2575 /* This only logs for now. In the future when the interface for kills/notifications
2576 * is more stable we can extend service results table similar to how kernel oom kills
2578 (void) unit_check_oomd_kill(u1
);
2580 manager_invoke_sigchld_event(m
, u1
, &si
);
2583 manager_invoke_sigchld_event(m
, u2
, &si
);
2585 for (size_t i
= 0; array_copy
[i
]; i
++)
2586 manager_invoke_sigchld_event(m
, array_copy
[i
], &si
);
2589 /* And now, we actually reap the zombie. */
2590 if (waitid(P_PID
, si
.si_pid
, &si
, WEXITED
) < 0) {
2591 log_error_errno(errno
, "Failed to dequeue child, ignoring: %m");
2598 /* All children processed for now, turn off event source */
2600 r
= sd_event_source_set_enabled(m
->sigchld_event_source
, SD_EVENT_OFF
);
2602 return log_error_errno(r
, "Failed to disable SIGCHLD event source: %m");
2607 static void manager_start_target(Manager
*m
, const char *name
, JobMode mode
) {
2608 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
2611 log_debug("Activating special unit %s", name
);
2613 r
= manager_add_job_by_name(m
, JOB_START
, name
, mode
, NULL
, &error
, NULL
);
2615 log_error("Failed to enqueue %s job: %s", name
, bus_error_message(&error
, r
));
2618 static void manager_handle_ctrl_alt_del(Manager
*m
) {
2619 /* If the user presses C-A-D more than
2620 * 7 times within 2s, we reboot/shutdown immediately,
2621 * unless it was disabled in system.conf */
2623 if (ratelimit_below(&m
->ctrl_alt_del_ratelimit
) || m
->cad_burst_action
== EMERGENCY_ACTION_NONE
)
2624 manager_start_target(m
, SPECIAL_CTRL_ALT_DEL_TARGET
, JOB_REPLACE_IRREVERSIBLY
);
2626 emergency_action(m
, m
->cad_burst_action
, EMERGENCY_ACTION_WARN
, NULL
, -1,
2627 "Ctrl-Alt-Del was pressed more than 7 times within 2s");
2630 static int manager_dispatch_signal_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
2631 Manager
*m
= userdata
;
2633 struct signalfd_siginfo sfsi
;
2637 assert(m
->signal_fd
== fd
);
2639 if (revents
!= EPOLLIN
) {
2640 log_warning("Got unexpected events from signal file descriptor.");
2644 n
= read(m
->signal_fd
, &sfsi
, sizeof(sfsi
));
2645 if (n
!= sizeof(sfsi
)) {
2647 log_warning("Truncated read from signal fd (%zu bytes), ignoring!", n
);
2651 if (IN_SET(errno
, EINTR
, EAGAIN
))
2654 /* We return an error here, which will kill this handler,
2655 * to avoid a busy loop on read error. */
2656 return log_error_errno(errno
, "Reading from signal fd failed: %m");
2659 log_received_signal(sfsi
.ssi_signo
== SIGCHLD
||
2660 (sfsi
.ssi_signo
== SIGTERM
&& MANAGER_IS_USER(m
))
2661 ? LOG_DEBUG
: LOG_INFO
,
2664 switch (sfsi
.ssi_signo
) {
2667 r
= sd_event_source_set_enabled(m
->sigchld_event_source
, SD_EVENT_ON
);
2669 log_warning_errno(r
, "Failed to enable SIGCHLD event source, ignoring: %m");
2674 if (MANAGER_IS_SYSTEM(m
)) {
2675 /* This is for compatibility with the original sysvinit */
2676 if (verify_run_space_and_log("Refusing to reexecute") < 0)
2679 m
->objective
= MANAGER_REEXECUTE
;
2685 if (MANAGER_IS_SYSTEM(m
))
2686 manager_handle_ctrl_alt_del(m
);
2688 manager_start_target(m
, SPECIAL_EXIT_TARGET
,
2689 JOB_REPLACE_IRREVERSIBLY
);
2693 /* This is a nop on non-init */
2694 if (MANAGER_IS_SYSTEM(m
))
2695 manager_start_target(m
, SPECIAL_KBREQUEST_TARGET
, JOB_REPLACE
);
2700 /* This is a nop on non-init */
2701 if (MANAGER_IS_SYSTEM(m
))
2702 manager_start_target(m
, SPECIAL_SIGPWR_TARGET
, JOB_REPLACE
);
2707 if (manager_dbus_is_running(m
, false)) {
2708 log_info("Trying to reconnect to bus...");
2710 (void) bus_init_api(m
);
2712 if (MANAGER_IS_SYSTEM(m
))
2713 (void) bus_init_system(m
);
2715 log_info("Starting D-Bus service...");
2716 manager_start_target(m
, SPECIAL_DBUS_SERVICE
, JOB_REPLACE
);
2722 _cleanup_free_
char *dump
= NULL
;
2724 r
= manager_get_dump_string(m
, &dump
);
2726 log_warning_errno(errno
, "Failed to acquire manager dump: %m");
2730 log_dump(LOG_INFO
, dump
);
2735 if (verify_run_space_and_log("Refusing to reload") < 0)
2738 m
->objective
= MANAGER_RELOAD
;
2743 /* Starting SIGRTMIN+0 */
2744 static const struct {
2747 } target_table
[] = {
2748 [0] = { SPECIAL_DEFAULT_TARGET
, JOB_ISOLATE
},
2749 [1] = { SPECIAL_RESCUE_TARGET
, JOB_ISOLATE
},
2750 [2] = { SPECIAL_EMERGENCY_TARGET
, JOB_ISOLATE
},
2751 [3] = { SPECIAL_HALT_TARGET
, JOB_REPLACE_IRREVERSIBLY
},
2752 [4] = { SPECIAL_POWEROFF_TARGET
, JOB_REPLACE_IRREVERSIBLY
},
2753 [5] = { SPECIAL_REBOOT_TARGET
, JOB_REPLACE_IRREVERSIBLY
},
2754 [6] = { SPECIAL_KEXEC_TARGET
, JOB_REPLACE_IRREVERSIBLY
},
2757 /* Starting SIGRTMIN+13, so that target halt and system halt are 10 apart */
2758 static const ManagerObjective objective_table
[] = {
2760 [1] = MANAGER_POWEROFF
,
2761 [2] = MANAGER_REBOOT
,
2762 [3] = MANAGER_KEXEC
,
2765 if ((int) sfsi
.ssi_signo
>= SIGRTMIN
+0 &&
2766 (int) sfsi
.ssi_signo
< SIGRTMIN
+(int) ELEMENTSOF(target_table
)) {
2767 int idx
= (int) sfsi
.ssi_signo
- SIGRTMIN
;
2768 manager_start_target(m
, target_table
[idx
].target
,
2769 target_table
[idx
].mode
);
2773 if ((int) sfsi
.ssi_signo
>= SIGRTMIN
+13 &&
2774 (int) sfsi
.ssi_signo
< SIGRTMIN
+13+(int) ELEMENTSOF(objective_table
)) {
2775 m
->objective
= objective_table
[sfsi
.ssi_signo
- SIGRTMIN
- 13];
2779 switch (sfsi
.ssi_signo
- SIGRTMIN
) {
2782 manager_override_show_status(m
, SHOW_STATUS_YES
, "signal");
2786 manager_override_show_status(m
, SHOW_STATUS_NO
, "signal");
2790 manager_override_log_level(m
, LOG_DEBUG
);
2794 manager_restore_original_log_level(m
);
2798 if (MANAGER_IS_USER(m
)) {
2799 m
->objective
= MANAGER_EXIT
;
2803 /* This is a nop on init */
2807 case 29: /* compatibility: used to be mapped to LOG_TARGET_SYSLOG_OR_KMSG */
2808 manager_restore_original_log_target(m
);
2812 manager_override_log_target(m
, LOG_TARGET_CONSOLE
);
2816 manager_override_log_target(m
, LOG_TARGET_KMSG
);
2820 log_warning("Got unhandled signal <%s>.", signal_to_string(sfsi
.ssi_signo
));
2827 static int manager_dispatch_time_change_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
2828 Manager
*m
= userdata
;
2832 assert(m
->time_change_fd
== fd
);
2834 log_struct(LOG_DEBUG
,
2835 "MESSAGE_ID=" SD_MESSAGE_TIME_CHANGE_STR
,
2836 LOG_MESSAGE("Time has been changed"));
2838 /* Restart the watch */
2839 (void) manager_setup_time_change(m
);
2841 HASHMAP_FOREACH(u
, m
->units
)
2842 if (UNIT_VTABLE(u
)->time_change
)
2843 UNIT_VTABLE(u
)->time_change(u
);
2848 static int manager_dispatch_timezone_change(
2849 sd_event_source
*source
,
2850 const struct inotify_event
*e
,
2853 Manager
*m
= userdata
;
2859 log_debug("inotify event for /etc/localtime");
2861 changed
= manager_read_timezone_stat(m
);
2865 /* Something changed, restart the watch, to ensure we watch the new /etc/localtime if it changed */
2866 (void) manager_setup_timezone_change(m
);
2868 /* Read the new timezone */
2871 log_debug("Timezone has been changed (now: %s).", tzname
[daylight
]);
2873 HASHMAP_FOREACH(u
, m
->units
)
2874 if (UNIT_VTABLE(u
)->timezone_change
)
2875 UNIT_VTABLE(u
)->timezone_change(u
);
2880 static int manager_dispatch_idle_pipe_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
2881 Manager
*m
= userdata
;
2884 assert(m
->idle_pipe
[2] == fd
);
2886 /* There's at least one Type=idle child that just gave up on us waiting for the boot process to complete. Let's
2887 * now turn off any further console output if there's at least one service that needs console access, so that
2888 * from now on our own output should not spill into that service's output anymore. After all, we support
2889 * Type=idle only to beautify console output and it generally is set on services that want to own the console
2890 * exclusively without our interference. */
2891 m
->no_console_output
= m
->n_on_console
> 0;
2893 /* Acknowledge the child's request, and let all all other children know too that they shouldn't wait any longer
2894 * by closing the pipes towards them, which is what they are waiting for. */
2895 manager_close_idle_pipe(m
);
2900 static int manager_dispatch_jobs_in_progress(sd_event_source
*source
, usec_t usec
, void *userdata
) {
2901 Manager
*m
= userdata
;
2907 manager_print_jobs_in_progress(m
);
2909 r
= sd_event_source_set_time_relative(source
, JOBS_IN_PROGRESS_PERIOD_USEC
);
2913 return sd_event_source_set_enabled(source
, SD_EVENT_ONESHOT
);
2916 int manager_loop(Manager
*m
) {
2917 RateLimit rl
= { .interval
= 1*USEC_PER_SEC
, .burst
= 50000 };
2921 assert(m
->objective
== MANAGER_OK
); /* Ensure manager_startup() has been called */
2923 manager_check_finished(m
);
2925 /* There might still be some zombies hanging around from before we were exec()'ed. Let's reap them. */
2926 r
= sd_event_source_set_enabled(m
->sigchld_event_source
, SD_EVENT_ON
);
2928 return log_error_errno(r
, "Failed to enable SIGCHLD event source: %m");
2930 while (m
->objective
== MANAGER_OK
) {
2931 usec_t wait_usec
, watchdog_usec
;
2933 watchdog_usec
= manager_get_watchdog(m
, WATCHDOG_RUNTIME
);
2934 if (m
->runtime_watchdog_running
)
2935 (void) watchdog_ping();
2936 else if (timestamp_is_set(watchdog_usec
))
2937 manager_retry_runtime_watchdog(m
);
2939 if (!ratelimit_below(&rl
)) {
2940 /* Yay, something is going seriously wrong, pause a little */
2941 log_warning("Looping too fast. Throttling execution a little.");
2945 if (manager_dispatch_load_queue(m
) > 0)
2948 if (manager_dispatch_gc_job_queue(m
) > 0)
2951 if (manager_dispatch_gc_unit_queue(m
) > 0)
2954 if (manager_dispatch_cleanup_queue(m
) > 0)
2957 if (manager_dispatch_cgroup_realize_queue(m
) > 0)
2960 if (manager_dispatch_stop_when_unneeded_queue(m
) > 0)
2963 if (manager_dispatch_dbus_queue(m
) > 0)
2966 /* Sleep for watchdog runtime wait time */
2967 if (timestamp_is_set(watchdog_usec
))
2968 wait_usec
= watchdog_runtime_wait();
2970 wait_usec
= USEC_INFINITY
;
2972 r
= sd_event_run(m
->event
, wait_usec
);
2974 return log_error_errno(r
, "Failed to run event loop: %m");
2977 return m
->objective
;
2980 int manager_load_unit_from_dbus_path(Manager
*m
, const char *s
, sd_bus_error
*e
, Unit
**_u
) {
2981 _cleanup_free_
char *n
= NULL
;
2982 sd_id128_t invocation_id
;
2990 r
= unit_name_from_dbus_path(s
, &n
);
2994 /* Permit addressing units by invocation ID: if the passed bus path is suffixed by a 128bit ID then we use it
2995 * as invocation ID. */
2996 r
= sd_id128_from_string(n
, &invocation_id
);
2998 u
= hashmap_get(m
->units_by_invocation_id
, &invocation_id
);
3004 return sd_bus_error_setf(e
, BUS_ERROR_NO_UNIT_FOR_INVOCATION_ID
,
3005 "No unit with the specified invocation ID " SD_ID128_FORMAT_STR
" known.",
3006 SD_ID128_FORMAT_VAL(invocation_id
));
3009 /* If this didn't work, we check if this is a unit name */
3010 if (!unit_name_is_valid(n
, UNIT_NAME_PLAIN
|UNIT_NAME_INSTANCE
)) {
3011 _cleanup_free_
char *nn
= NULL
;
3014 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
,
3015 "Unit name %s is neither a valid invocation ID nor unit name.", strnull(nn
));
3018 r
= manager_load_unit(m
, n
, NULL
, e
, &u
);
3026 int manager_get_job_from_dbus_path(Manager
*m
, const char *s
, Job
**_j
) {
3036 p
= startswith(s
, "/org/freedesktop/systemd1/job/");
3040 r
= safe_atou(p
, &id
);
3044 j
= manager_get_job(m
, id
);
3053 void manager_send_unit_audit(Manager
*m
, Unit
*u
, int type
, bool success
) {
3056 _cleanup_free_
char *p
= NULL
;
3060 if (!MANAGER_IS_SYSTEM(m
))
3063 audit_fd
= get_audit_fd();
3067 /* Don't generate audit events if the service was already
3068 * started and we're just deserializing */
3069 if (MANAGER_IS_RELOADING(m
))
3072 if (u
->type
!= UNIT_SERVICE
)
3075 r
= unit_name_to_prefix_and_instance(u
->id
, &p
);
3077 log_error_errno(r
, "Failed to extract prefix and instance of unit name: %m");
3081 msg
= strjoina("unit=", p
);
3082 if (audit_log_user_comm_message(audit_fd
, type
, msg
, "systemd", NULL
, NULL
, NULL
, success
) < 0) {
3084 /* We aren't allowed to send audit messages?
3085 * Then let's not retry again. */
3088 log_warning_errno(errno
, "Failed to send audit message: %m");
3094 void manager_send_unit_plymouth(Manager
*m
, Unit
*u
) {
3095 static const union sockaddr_union sa
= PLYMOUTH_SOCKET
;
3096 _cleanup_free_
char *message
= NULL
;
3097 _cleanup_close_
int fd
= -1;
3100 /* Don't generate plymouth events if the service was already
3101 * started and we're just deserializing */
3102 if (MANAGER_IS_RELOADING(m
))
3105 if (!MANAGER_IS_SYSTEM(m
))
3108 if (detect_container() > 0)
3111 if (!IN_SET(u
->type
, UNIT_SERVICE
, UNIT_MOUNT
, UNIT_SWAP
))
3114 /* We set SOCK_NONBLOCK here so that we rather drop the
3115 * message then wait for plymouth */
3116 fd
= socket(AF_UNIX
, SOCK_STREAM
|SOCK_CLOEXEC
|SOCK_NONBLOCK
, 0);
3118 log_error_errno(errno
, "socket() failed: %m");
3122 if (connect(fd
, &sa
.sa
, SOCKADDR_UN_LEN(sa
.un
)) < 0) {
3123 if (!IN_SET(errno
, EAGAIN
, ENOENT
) && !ERRNO_IS_DISCONNECT(errno
))
3124 log_error_errno(errno
, "connect() failed: %m");
3128 if (asprintf(&message
, "U\002%c%s%n", (int) (strlen(u
->id
) + 1), u
->id
, &n
) < 0) {
3134 if (write(fd
, message
, n
+ 1) != n
+ 1)
3135 if (!IN_SET(errno
, EAGAIN
, ENOENT
) && !ERRNO_IS_DISCONNECT(errno
))
3136 log_error_errno(errno
, "Failed to write Plymouth message: %m");
3139 int manager_open_serialization(Manager
*m
, FILE **_f
) {
3140 _cleanup_close_
int fd
= -1;
3145 fd
= open_serialization_fd("systemd-state");
3149 f
= take_fdopen(&fd
, "w+");
3157 static bool manager_timestamp_shall_serialize(ManagerTimestamp t
) {
3162 /* The following timestamps only apply to the host system, hence only serialize them there */
3164 MANAGER_TIMESTAMP_USERSPACE
, MANAGER_TIMESTAMP_FINISH
,
3165 MANAGER_TIMESTAMP_SECURITY_START
, MANAGER_TIMESTAMP_SECURITY_FINISH
,
3166 MANAGER_TIMESTAMP_GENERATORS_START
, MANAGER_TIMESTAMP_GENERATORS_FINISH
,
3167 MANAGER_TIMESTAMP_UNITS_LOAD_START
, MANAGER_TIMESTAMP_UNITS_LOAD_FINISH
);
3170 #define DESTROY_IPC_FLAG (UINT32_C(1) << 31)
3172 static void manager_serialize_uid_refs_internal(
3175 const char *field_name
) {
3182 /* Serialize the UID reference table. Or actually, just the IPC destruction flag of it, as
3183 * the actual counter of it is better rebuild after a reload/reexec. */
3185 HASHMAP_FOREACH_KEY(p
, k
, uid_refs
) {
3189 uid
= PTR_TO_UID(k
);
3190 c
= PTR_TO_UINT32(p
);
3192 if (!(c
& DESTROY_IPC_FLAG
))
3195 (void) serialize_item_format(f
, field_name
, UID_FMT
, uid
);
3199 static void manager_serialize_uid_refs(Manager
*m
, FILE *f
) {
3200 manager_serialize_uid_refs_internal(f
, m
->uid_refs
, "destroy-ipc-uid");
3203 static void manager_serialize_gid_refs(Manager
*m
, FILE *f
) {
3204 manager_serialize_uid_refs_internal(f
, m
->gid_refs
, "destroy-ipc-gid");
3207 int manager_serialize(
3211 bool switching_root
) {
3221 _cleanup_(manager_reloading_stopp
) _unused_ Manager
*reloading
= manager_reloading_start(m
);
3223 (void) serialize_item_format(f
, "current-job-id", "%" PRIu32
, m
->current_job_id
);
3224 (void) serialize_item_format(f
, "n-installed-jobs", "%u", m
->n_installed_jobs
);
3225 (void) serialize_item_format(f
, "n-failed-jobs", "%u", m
->n_failed_jobs
);
3226 (void) serialize_bool(f
, "taint-usr", m
->taint_usr
);
3227 (void) serialize_bool(f
, "ready-sent", m
->ready_sent
);
3228 (void) serialize_bool(f
, "taint-logged", m
->taint_logged
);
3229 (void) serialize_bool(f
, "service-watchdogs", m
->service_watchdogs
);
3231 /* After switching root, udevd has not been started yet. So, enumeration results should not be emitted. */
3232 (void) serialize_bool(f
, "honor-device-enumeration", !switching_root
);
3234 if (m
->show_status_overridden
!= _SHOW_STATUS_INVALID
)
3235 (void) serialize_item(f
, "show-status-overridden",
3236 show_status_to_string(m
->show_status_overridden
));
3238 if (m
->log_level_overridden
)
3239 (void) serialize_item_format(f
, "log-level-override", "%i", log_get_max_level());
3240 if (m
->log_target_overridden
)
3241 (void) serialize_item(f
, "log-target-override", log_target_to_string(log_get_target()));
3243 (void) serialize_usec(f
, "runtime-watchdog-overridden", m
->watchdog_overridden
[WATCHDOG_RUNTIME
]);
3244 (void) serialize_usec(f
, "reboot-watchdog-overridden", m
->watchdog_overridden
[WATCHDOG_REBOOT
]);
3245 (void) serialize_usec(f
, "kexec-watchdog-overridden", m
->watchdog_overridden
[WATCHDOG_KEXEC
]);
3247 for (ManagerTimestamp q
= 0; q
< _MANAGER_TIMESTAMP_MAX
; q
++) {
3248 _cleanup_free_
char *joined
= NULL
;
3250 if (!manager_timestamp_shall_serialize(q
))
3253 joined
= strjoin(manager_timestamp_to_string(q
), "-timestamp");
3257 (void) serialize_dual_timestamp(f
, joined
, m
->timestamps
+ q
);
3260 if (!switching_root
)
3261 (void) serialize_strv(f
, "env", m
->client_environment
);
3263 if (m
->notify_fd
>= 0) {
3264 r
= serialize_fd(f
, fds
, "notify-fd", m
->notify_fd
);
3268 (void) serialize_item(f
, "notify-socket", m
->notify_socket
);
3271 if (m
->cgroups_agent_fd
>= 0) {
3272 r
= serialize_fd(f
, fds
, "cgroups-agent-fd", m
->cgroups_agent_fd
);
3277 if (m
->user_lookup_fds
[0] >= 0) {
3280 copy0
= fdset_put_dup(fds
, m
->user_lookup_fds
[0]);
3282 return log_error_errno(copy0
, "Failed to add user lookup fd to serialization: %m");
3284 copy1
= fdset_put_dup(fds
, m
->user_lookup_fds
[1]);
3286 return log_error_errno(copy1
, "Failed to add user lookup fd to serialization: %m");
3288 (void) serialize_item_format(f
, "user-lookup", "%i %i", copy0
, copy1
);
3291 bus_track_serialize(m
->subscribed
, f
, "subscribed");
3293 r
= dynamic_user_serialize(m
, f
, fds
);
3297 manager_serialize_uid_refs(m
, f
);
3298 manager_serialize_gid_refs(m
, f
);
3300 r
= exec_runtime_serialize(m
, f
, fds
);
3304 (void) fputc('\n', f
);
3306 HASHMAP_FOREACH_KEY(u
, t
, m
->units
) {
3314 r
= unit_serialize(u
, f
, fds
, !switching_root
);
3319 r
= fflush_and_check(f
);
3321 return log_error_errno(r
, "Failed to flush serialization: %m");
3323 r
= bus_fdset_add_all(m
, fds
);
3325 return log_error_errno(r
, "Failed to add bus sockets to serialization: %m");
3330 static int manager_deserialize_one_unit(Manager
*m
, const char *name
, FILE *f
, FDSet
*fds
) {
3334 r
= manager_load_unit(m
, name
, NULL
, NULL
, &u
);
3338 return log_notice_errno(r
, "Failed to load unit \"%s\", skipping deserialization: %m", name
);
3341 r
= unit_deserialize(u
, f
, fds
);
3345 return log_notice_errno(r
, "Failed to deserialize unit \"%s\", skipping: %m", name
);
3351 static int manager_deserialize_units(Manager
*m
, FILE *f
, FDSet
*fds
) {
3352 const char *unit_name
;
3356 _cleanup_free_
char *line
= NULL
;
3358 r
= read_line(f
, LONG_LINE_MAX
, &line
);
3360 return log_error_errno(r
, "Failed to read serialization line: %m");
3364 unit_name
= strstrip(line
);
3366 r
= manager_deserialize_one_unit(m
, unit_name
, f
, fds
);
3370 r
= unit_deserialize_skip(f
);
3379 usec_t
manager_get_watchdog(Manager
*m
, WatchdogType t
) {
3382 if (MANAGER_IS_USER(m
))
3383 return USEC_INFINITY
;
3385 if (timestamp_is_set(m
->watchdog_overridden
[t
]))
3386 return m
->watchdog_overridden
[t
];
3388 return m
->watchdog
[t
];
3391 void manager_set_watchdog(Manager
*m
, WatchdogType t
, usec_t timeout
) {
3396 if (MANAGER_IS_USER(m
))
3399 if (m
->watchdog
[t
] == timeout
)
3402 if (t
== WATCHDOG_RUNTIME
)
3403 if (!timestamp_is_set(m
->watchdog_overridden
[WATCHDOG_RUNTIME
])) {
3404 if (timestamp_is_set(timeout
)) {
3405 r
= watchdog_set_timeout(&timeout
);
3408 m
->runtime_watchdog_running
= true;
3410 watchdog_close(true);
3411 m
->runtime_watchdog_running
= false;
3415 m
->watchdog
[t
] = timeout
;
3418 int manager_override_watchdog(Manager
*m
, WatchdogType t
, usec_t timeout
) {
3423 if (MANAGER_IS_USER(m
))
3426 if (m
->watchdog_overridden
[t
] == timeout
)
3429 if (t
== WATCHDOG_RUNTIME
) {
3432 p
= timestamp_is_set(timeout
) ? &timeout
: &m
->watchdog
[t
];
3433 if (timestamp_is_set(*p
)) {
3434 r
= watchdog_set_timeout(p
);
3437 m
->runtime_watchdog_running
= true;
3439 watchdog_close(true);
3440 m
->runtime_watchdog_running
= false;
3444 m
->watchdog_overridden
[t
] = timeout
;
3449 void manager_retry_runtime_watchdog(Manager
*m
) {
3454 if (timestamp_is_set(m
->watchdog_overridden
[WATCHDOG_RUNTIME
]))
3455 r
= watchdog_set_timeout(&m
->watchdog_overridden
[WATCHDOG_RUNTIME
]);
3457 r
= watchdog_set_timeout(&m
->watchdog
[WATCHDOG_RUNTIME
]);
3460 m
->runtime_watchdog_running
= true;
3463 static void manager_deserialize_uid_refs_one_internal(
3465 const char *value
) {
3474 r
= parse_uid(value
, &uid
);
3475 if (r
< 0 || uid
== 0) {
3476 log_debug("Unable to parse UID/GID reference serialization: " UID_FMT
, uid
);
3480 if (hashmap_ensure_allocated(uid_refs
, &trivial_hash_ops
) < 0) {
3485 c
= PTR_TO_UINT32(hashmap_get(*uid_refs
, UID_TO_PTR(uid
)));
3486 if (c
& DESTROY_IPC_FLAG
)
3489 c
|= DESTROY_IPC_FLAG
;
3491 r
= hashmap_replace(*uid_refs
, UID_TO_PTR(uid
), UINT32_TO_PTR(c
));
3493 log_debug_errno(r
, "Failed to add UID/GID reference entry: %m");
3498 static void manager_deserialize_uid_refs_one(Manager
*m
, const char *value
) {
3499 manager_deserialize_uid_refs_one_internal(&m
->uid_refs
, value
);
3502 static void manager_deserialize_gid_refs_one(Manager
*m
, const char *value
) {
3503 manager_deserialize_uid_refs_one_internal(&m
->gid_refs
, value
);
3506 int manager_deserialize(Manager
*m
, FILE *f
, FDSet
*fds
) {
3512 if (DEBUG_LOGGING
) {
3513 if (fdset_isempty(fds
))
3514 log_debug("No file descriptors passed");
3518 FDSET_FOREACH(fd
, fds
) {
3519 _cleanup_free_
char *fn
= NULL
;
3521 r
= fd_get_path(fd
, &fn
);
3523 log_debug_errno(r
, "Received serialized fd %i → %m", fd
);
3525 log_debug("Received serialized fd %i → %s", fd
, strna(fn
));
3530 log_debug("Deserializing state...");
3532 /* If we are not in reload mode yet, enter it now. Not that this is recursive, a caller might already have
3533 * increased it to non-zero, which is why we just increase it by one here and down again at the end of this
3535 _cleanup_(manager_reloading_stopp
) _unused_ Manager
*reloading
= manager_reloading_start(m
);
3538 _cleanup_free_
char *line
= NULL
;
3539 const char *val
, *l
;
3541 r
= read_line(f
, LONG_LINE_MAX
, &line
);
3543 return log_error_errno(r
, "Failed to read serialization line: %m");
3548 if (isempty(l
)) /* end marker */
3551 if ((val
= startswith(l
, "current-job-id="))) {
3554 if (safe_atou32(val
, &id
) < 0)
3555 log_notice("Failed to parse current job id value '%s', ignoring.", val
);
3557 m
->current_job_id
= MAX(m
->current_job_id
, id
);
3559 } else if ((val
= startswith(l
, "n-installed-jobs="))) {
3562 if (safe_atou32(val
, &n
) < 0)
3563 log_notice("Failed to parse installed jobs counter '%s', ignoring.", val
);
3565 m
->n_installed_jobs
+= n
;
3567 } else if ((val
= startswith(l
, "n-failed-jobs="))) {
3570 if (safe_atou32(val
, &n
) < 0)
3571 log_notice("Failed to parse failed jobs counter '%s', ignoring.", val
);
3573 m
->n_failed_jobs
+= n
;
3575 } else if ((val
= startswith(l
, "taint-usr="))) {
3578 b
= parse_boolean(val
);
3580 log_notice("Failed to parse taint /usr flag '%s', ignoring.", val
);
3582 m
->taint_usr
= m
->taint_usr
|| b
;
3584 } else if ((val
= startswith(l
, "ready-sent="))) {
3587 b
= parse_boolean(val
);
3589 log_notice("Failed to parse ready-sent flag '%s', ignoring.", val
);
3591 m
->ready_sent
= m
->ready_sent
|| b
;
3593 } else if ((val
= startswith(l
, "taint-logged="))) {
3596 b
= parse_boolean(val
);
3598 log_notice("Failed to parse taint-logged flag '%s', ignoring.", val
);
3600 m
->taint_logged
= m
->taint_logged
|| b
;
3602 } else if ((val
= startswith(l
, "service-watchdogs="))) {
3605 b
= parse_boolean(val
);
3607 log_notice("Failed to parse service-watchdogs flag '%s', ignoring.", val
);
3609 m
->service_watchdogs
= b
;
3611 } else if ((val
= startswith(l
, "honor-device-enumeration="))) {
3614 b
= parse_boolean(val
);
3616 log_notice("Failed to parse honor-device-enumeration flag '%s', ignoring.", val
);
3618 m
->honor_device_enumeration
= b
;
3620 } else if ((val
= startswith(l
, "show-status-overridden="))) {
3623 s
= show_status_from_string(val
);
3625 log_notice("Failed to parse show-status-overridden flag '%s', ignoring.", val
);
3627 manager_override_show_status(m
, s
, "deserialize");
3629 } else if ((val
= startswith(l
, "log-level-override="))) {
3632 level
= log_level_from_string(val
);
3634 log_notice("Failed to parse log-level-override value '%s', ignoring.", val
);
3636 manager_override_log_level(m
, level
);
3638 } else if ((val
= startswith(l
, "log-target-override="))) {
3641 target
= log_target_from_string(val
);
3643 log_notice("Failed to parse log-target-override value '%s', ignoring.", val
);
3645 manager_override_log_target(m
, target
);
3647 } else if ((val
= startswith(l
, "runtime-watchdog-overridden="))) {
3650 if (deserialize_usec(val
, &t
) < 0)
3651 log_notice("Failed to parse runtime-watchdog-overridden value '%s', ignoring.", val
);
3653 manager_override_watchdog(m
, WATCHDOG_RUNTIME
, t
);
3655 } else if ((val
= startswith(l
, "reboot-watchdog-overridden="))) {
3658 if (deserialize_usec(val
, &t
) < 0)
3659 log_notice("Failed to parse reboot-watchdog-overridden value '%s', ignoring.", val
);
3661 manager_override_watchdog(m
, WATCHDOG_REBOOT
, t
);
3663 } else if ((val
= startswith(l
, "kexec-watchdog-overridden="))) {
3666 if (deserialize_usec(val
, &t
) < 0)
3667 log_notice("Failed to parse kexec-watchdog-overridden value '%s', ignoring.", val
);
3669 manager_override_watchdog(m
, WATCHDOG_KEXEC
, t
);
3671 } else if (startswith(l
, "env=")) {
3672 r
= deserialize_environment(l
+ 4, &m
->client_environment
);
3674 log_notice_errno(r
, "Failed to parse environment entry: \"%s\", ignoring: %m", l
);
3676 } else if ((val
= startswith(l
, "notify-fd="))) {
3679 if (safe_atoi(val
, &fd
) < 0 || fd
< 0 || !fdset_contains(fds
, fd
))
3680 log_notice("Failed to parse notify fd, ignoring: \"%s\"", val
);
3682 m
->notify_event_source
= sd_event_source_unref(m
->notify_event_source
);
3683 safe_close(m
->notify_fd
);
3684 m
->notify_fd
= fdset_remove(fds
, fd
);
3687 } else if ((val
= startswith(l
, "notify-socket="))) {
3688 r
= free_and_strdup(&m
->notify_socket
, val
);
3692 } else if ((val
= startswith(l
, "cgroups-agent-fd="))) {
3695 if (safe_atoi(val
, &fd
) < 0 || fd
< 0 || !fdset_contains(fds
, fd
))
3696 log_notice("Failed to parse cgroups agent fd, ignoring.: %s", val
);
3698 m
->cgroups_agent_event_source
= sd_event_source_unref(m
->cgroups_agent_event_source
);
3699 safe_close(m
->cgroups_agent_fd
);
3700 m
->cgroups_agent_fd
= fdset_remove(fds
, fd
);
3703 } else if ((val
= startswith(l
, "user-lookup="))) {
3706 if (sscanf(val
, "%i %i", &fd0
, &fd1
) != 2 || fd0
< 0 || fd1
< 0 || fd0
== fd1
|| !fdset_contains(fds
, fd0
) || !fdset_contains(fds
, fd1
))
3707 log_notice("Failed to parse user lookup fd, ignoring: %s", val
);
3709 m
->user_lookup_event_source
= sd_event_source_unref(m
->user_lookup_event_source
);
3710 safe_close_pair(m
->user_lookup_fds
);
3711 m
->user_lookup_fds
[0] = fdset_remove(fds
, fd0
);
3712 m
->user_lookup_fds
[1] = fdset_remove(fds
, fd1
);
3715 } else if ((val
= startswith(l
, "dynamic-user=")))
3716 dynamic_user_deserialize_one(m
, val
, fds
);
3717 else if ((val
= startswith(l
, "destroy-ipc-uid=")))
3718 manager_deserialize_uid_refs_one(m
, val
);
3719 else if ((val
= startswith(l
, "destroy-ipc-gid=")))
3720 manager_deserialize_gid_refs_one(m
, val
);
3721 else if ((val
= startswith(l
, "exec-runtime=")))
3722 (void) exec_runtime_deserialize_one(m
, val
, fds
);
3723 else if ((val
= startswith(l
, "subscribed="))) {
3725 if (strv_extend(&m
->deserialized_subscribed
, val
) < 0)
3731 for (q
= 0; q
< _MANAGER_TIMESTAMP_MAX
; q
++) {
3732 val
= startswith(l
, manager_timestamp_to_string(q
));
3736 val
= startswith(val
, "-timestamp=");
3741 if (q
< _MANAGER_TIMESTAMP_MAX
) /* found it */
3742 (void) deserialize_dual_timestamp(val
, m
->timestamps
+ q
);
3743 else if (!startswith(l
, "kdbus-fd=")) /* ignore kdbus */
3744 log_notice("Unknown serialization item '%s', ignoring.", l
);
3748 return manager_deserialize_units(m
, f
, fds
);
3751 int manager_reload(Manager
*m
) {
3752 _cleanup_(manager_reloading_stopp
) Manager
*reloading
= NULL
;
3753 _cleanup_fdset_free_ FDSet
*fds
= NULL
;
3754 _cleanup_fclose_
FILE *f
= NULL
;
3759 r
= manager_open_serialization(m
, &f
);
3761 return log_error_errno(r
, "Failed to create serialization file: %m");
3767 /* We are officially in reload mode from here on. */
3768 reloading
= manager_reloading_start(m
);
3770 r
= manager_serialize(m
, f
, fds
, false);
3774 if (fseeko(f
, 0, SEEK_SET
) < 0)
3775 return log_error_errno(errno
, "Failed to seek to beginning of serialization: %m");
3777 /* 💀 This is the point of no return, from here on there is no way back. 💀 */
3780 bus_manager_send_reloading(m
, true);
3782 /* Start by flushing out all jobs and units, all generated units, all runtime environments, all dynamic users
3783 * and everything else that is worth flushing out. We'll get it all back from the serialization — if we need
3786 manager_clear_jobs_and_units(m
);
3787 lookup_paths_flush_generator(&m
->lookup_paths
);
3788 lookup_paths_free(&m
->lookup_paths
);
3789 exec_runtime_vacuum(m
);
3790 dynamic_user_vacuum(m
, false);
3791 m
->uid_refs
= hashmap_free(m
->uid_refs
);
3792 m
->gid_refs
= hashmap_free(m
->gid_refs
);
3794 r
= lookup_paths_init(&m
->lookup_paths
, m
->unit_file_scope
, 0, NULL
);
3796 log_warning_errno(r
, "Failed to initialize path lookup table, ignoring: %m");
3798 (void) manager_run_environment_generators(m
);
3799 (void) manager_run_generators(m
);
3801 lookup_paths_log(&m
->lookup_paths
);
3803 /* We flushed out generated files, for which we don't watch mtime, so we should flush the old map. */
3804 manager_free_unit_name_maps(m
);
3806 /* First, enumerate what we can from kernel and suchlike */
3807 manager_enumerate_perpetual(m
);
3808 manager_enumerate(m
);
3810 /* Second, deserialize our stored data */
3811 r
= manager_deserialize(m
, f
, fds
);
3813 log_warning_errno(r
, "Deserialization failed, proceeding anyway: %m");
3815 /* We don't need the serialization anymore */
3818 /* Re-register notify_fd as event source, and set up other sockets/communication channels we might need */
3819 (void) manager_setup_notify(m
);
3820 (void) manager_setup_cgroups_agent(m
);
3821 (void) manager_setup_user_lookup_fd(m
);
3823 /* Third, fire things up! */
3824 manager_coldplug(m
);
3826 /* Clean up runtime objects no longer referenced */
3829 /* Clean up deserialized tracked clients */
3830 m
->deserialized_subscribed
= strv_free(m
->deserialized_subscribed
);
3832 /* Consider the reload process complete now. */
3833 assert(m
->n_reloading
> 0);
3836 /* On manager reloading, device tag data should exists, thus, we should honor the results of device
3837 * enumeration. The flag should be always set correctly by the serialized data, but it may fail. So,
3838 * let's always set the flag here for safety. */
3839 m
->honor_device_enumeration
= true;
3843 m
->send_reloading_done
= true;
3847 void manager_reset_failed(Manager
*m
) {
3852 HASHMAP_FOREACH(u
, m
->units
)
3853 unit_reset_failed(u
);
3856 bool manager_unit_inactive_or_pending(Manager
*m
, const char *name
) {
3862 /* Returns true if the unit is inactive or going down */
3863 u
= manager_get_unit(m
, name
);
3867 return unit_inactive_or_pending(u
);
3870 static void log_taint_string(Manager
*m
) {
3871 _cleanup_free_
char *taint
= NULL
;
3875 if (MANAGER_IS_USER(m
) || m
->taint_logged
)
3878 m
->taint_logged
= true; /* only check for taint once */
3880 taint
= manager_taint_string(m
);
3884 log_struct(LOG_NOTICE
,
3885 LOG_MESSAGE("System is tainted: %s", taint
),
3887 "MESSAGE_ID=" SD_MESSAGE_TAINTED_STR
);
3890 static void manager_notify_finished(Manager
*m
) {
3891 char userspace
[FORMAT_TIMESPAN_MAX
], initrd
[FORMAT_TIMESPAN_MAX
], kernel
[FORMAT_TIMESPAN_MAX
], sum
[FORMAT_TIMESPAN_MAX
];
3892 usec_t firmware_usec
, loader_usec
, kernel_usec
, initrd_usec
, userspace_usec
, total_usec
;
3894 if (MANAGER_IS_TEST_RUN(m
))
3897 if (MANAGER_IS_SYSTEM(m
) && detect_container() <= 0) {
3898 char ts
[FORMAT_TIMESPAN_MAX
];
3899 char buf
[FORMAT_TIMESPAN_MAX
+ STRLEN(" (firmware) + ") + FORMAT_TIMESPAN_MAX
+ STRLEN(" (loader) + ")]
3902 size_t size
= sizeof buf
;
3904 /* Note that MANAGER_TIMESTAMP_KERNEL's monotonic value is always at 0, and
3905 * MANAGER_TIMESTAMP_FIRMWARE's and MANAGER_TIMESTAMP_LOADER's monotonic value should be considered
3906 * negative values. */
3908 firmware_usec
= m
->timestamps
[MANAGER_TIMESTAMP_FIRMWARE
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_LOADER
].monotonic
;
3909 loader_usec
= m
->timestamps
[MANAGER_TIMESTAMP_LOADER
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_KERNEL
].monotonic
;
3910 userspace_usec
= m
->timestamps
[MANAGER_TIMESTAMP_FINISH
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_USERSPACE
].monotonic
;
3911 total_usec
= m
->timestamps
[MANAGER_TIMESTAMP_FIRMWARE
].monotonic
+ m
->timestamps
[MANAGER_TIMESTAMP_FINISH
].monotonic
;
3913 if (firmware_usec
> 0)
3914 size
= strpcpyf(&p
, size
, "%s (firmware) + ", format_timespan(ts
, sizeof(ts
), firmware_usec
, USEC_PER_MSEC
));
3915 if (loader_usec
> 0)
3916 size
= strpcpyf(&p
, size
, "%s (loader) + ", format_timespan(ts
, sizeof(ts
), loader_usec
, USEC_PER_MSEC
));
3918 if (dual_timestamp_is_set(&m
->timestamps
[MANAGER_TIMESTAMP_INITRD
])) {
3920 /* The initrd case on bare-metal*/
3921 kernel_usec
= m
->timestamps
[MANAGER_TIMESTAMP_INITRD
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_KERNEL
].monotonic
;
3922 initrd_usec
= m
->timestamps
[MANAGER_TIMESTAMP_USERSPACE
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_INITRD
].monotonic
;
3924 log_struct(LOG_INFO
,
3925 "MESSAGE_ID=" SD_MESSAGE_STARTUP_FINISHED_STR
,
3926 "KERNEL_USEC="USEC_FMT
, kernel_usec
,
3927 "INITRD_USEC="USEC_FMT
, initrd_usec
,
3928 "USERSPACE_USEC="USEC_FMT
, userspace_usec
,
3929 LOG_MESSAGE("Startup finished in %s%s (kernel) + %s (initrd) + %s (userspace) = %s.",
3931 format_timespan(kernel
, sizeof(kernel
), kernel_usec
, USEC_PER_MSEC
),
3932 format_timespan(initrd
, sizeof(initrd
), initrd_usec
, USEC_PER_MSEC
),
3933 format_timespan(userspace
, sizeof(userspace
), userspace_usec
, USEC_PER_MSEC
),
3934 format_timespan(sum
, sizeof(sum
), total_usec
, USEC_PER_MSEC
)));
3936 /* The initrd-less case on bare-metal*/
3938 kernel_usec
= m
->timestamps
[MANAGER_TIMESTAMP_USERSPACE
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_KERNEL
].monotonic
;
3941 log_struct(LOG_INFO
,
3942 "MESSAGE_ID=" SD_MESSAGE_STARTUP_FINISHED_STR
,
3943 "KERNEL_USEC="USEC_FMT
, kernel_usec
,
3944 "USERSPACE_USEC="USEC_FMT
, userspace_usec
,
3945 LOG_MESSAGE("Startup finished in %s%s (kernel) + %s (userspace) = %s.",
3947 format_timespan(kernel
, sizeof(kernel
), kernel_usec
, USEC_PER_MSEC
),
3948 format_timespan(userspace
, sizeof(userspace
), userspace_usec
, USEC_PER_MSEC
),
3949 format_timespan(sum
, sizeof(sum
), total_usec
, USEC_PER_MSEC
)));
3952 /* The container and --user case */
3953 firmware_usec
= loader_usec
= initrd_usec
= kernel_usec
= 0;
3954 total_usec
= userspace_usec
= m
->timestamps
[MANAGER_TIMESTAMP_FINISH
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_USERSPACE
].monotonic
;
3956 log_struct(LOG_INFO
,
3957 "MESSAGE_ID=" SD_MESSAGE_USER_STARTUP_FINISHED_STR
,
3958 "USERSPACE_USEC="USEC_FMT
, userspace_usec
,
3959 LOG_MESSAGE("Startup finished in %s.",
3960 format_timespan(sum
, sizeof(sum
), total_usec
, USEC_PER_MSEC
)));
3963 bus_manager_send_finished(m
, firmware_usec
, loader_usec
, kernel_usec
, initrd_usec
, userspace_usec
, total_usec
);
3966 m
->ready_sent
? "STATUS=Startup finished in %s."
3968 "STATUS=Startup finished in %s.",
3969 format_timespan(sum
, sizeof(sum
), total_usec
, USEC_PER_MSEC
));
3970 m
->ready_sent
= true;
3972 log_taint_string(m
);
3975 static void manager_send_ready(Manager
*m
) {
3978 /* We send READY=1 on reaching basic.target only when running in --user mode. */
3979 if (!MANAGER_IS_USER(m
) || m
->ready_sent
)
3982 m
->ready_sent
= true;
3986 "STATUS=Reached " SPECIAL_BASIC_TARGET
".");
3989 static void manager_check_basic_target(Manager
*m
) {
3994 /* Small shortcut */
3995 if (m
->ready_sent
&& m
->taint_logged
)
3998 u
= manager_get_unit(m
, SPECIAL_BASIC_TARGET
);
3999 if (!u
|| !UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u
)))
4002 /* For user managers, send out READY=1 as soon as we reach basic.target */
4003 manager_send_ready(m
);
4005 /* Log the taint string as soon as we reach basic.target */
4006 log_taint_string(m
);
4009 void manager_check_finished(Manager
*m
) {
4012 if (MANAGER_IS_RELOADING(m
))
4015 /* Verify that we have entered the event loop already, and not left it again. */
4016 if (!MANAGER_IS_RUNNING(m
))
4019 manager_check_basic_target(m
);
4021 if (hashmap_size(m
->jobs
) > 0) {
4022 if (m
->jobs_in_progress_event_source
)
4023 /* Ignore any failure, this is only for feedback */
4024 (void) sd_event_source_set_time(m
->jobs_in_progress_event_source
,
4025 manager_watch_jobs_next_time(m
));
4029 /* The jobs hashmap tends to grow a lot during boot, and then it's not reused until shutdown. Let's
4030 kill the hashmap if it is relatively large. */
4031 if (hashmap_buckets(m
->jobs
) > hashmap_size(m
->units
) / 10)
4032 m
->jobs
= hashmap_free(m
->jobs
);
4034 manager_flip_auto_status(m
, false, "boot finished");
4036 /* Notify Type=idle units that we are done now */
4037 manager_close_idle_pipe(m
);
4039 /* Turn off confirm spawn now */
4040 m
->confirm_spawn
= NULL
;
4042 /* No need to update ask password status when we're going non-interactive */
4043 manager_close_ask_password(m
);
4045 /* This is no longer the first boot */
4046 manager_set_first_boot(m
, false);
4048 if (MANAGER_IS_FINISHED(m
))
4051 dual_timestamp_get(m
->timestamps
+ MANAGER_TIMESTAMP_FINISH
);
4053 manager_notify_finished(m
);
4055 manager_invalidate_startup_units(m
);
4058 static bool generator_path_any(const char* const* paths
) {
4062 /* Optimize by skipping the whole process by not creating output directories
4063 * if no generators are found. */
4064 STRV_FOREACH(path
, (char**) paths
)
4065 if (access(*path
, F_OK
) == 0)
4067 else if (errno
!= ENOENT
)
4068 log_warning_errno(errno
, "Failed to open generator directory %s: %m", *path
);
4073 static int manager_run_environment_generators(Manager
*m
) {
4074 char **tmp
= NULL
; /* this is only used in the forked process, no cleanup here */
4075 _cleanup_strv_free_
char **paths
= NULL
;
4077 [STDOUT_GENERATE
] = &tmp
,
4078 [STDOUT_COLLECT
] = &tmp
,
4079 [STDOUT_CONSUME
] = &m
->transient_environment
,
4083 if (MANAGER_IS_TEST_RUN(m
) && !(m
->test_run_flags
& MANAGER_TEST_RUN_ENV_GENERATORS
))
4086 paths
= env_generator_binary_paths(MANAGER_IS_SYSTEM(m
));
4090 if (!generator_path_any((const char* const*) paths
))
4093 RUN_WITH_UMASK(0022)
4094 r
= execute_directories((const char* const*) paths
, DEFAULT_TIMEOUT_USEC
, gather_environment
,
4095 args
, NULL
, m
->transient_environment
,
4096 EXEC_DIR_PARALLEL
| EXEC_DIR_IGNORE_ERRORS
| EXEC_DIR_SET_SYSTEMD_EXEC_PID
);
4100 static int manager_run_generators(Manager
*m
) {
4101 _cleanup_strv_free_
char **paths
= NULL
;
4102 const char *argv
[5];
4107 if (MANAGER_IS_TEST_RUN(m
) && !(m
->test_run_flags
& MANAGER_TEST_RUN_GENERATORS
))
4110 paths
= generator_binary_paths(m
->unit_file_scope
);
4114 if (!generator_path_any((const char* const*) paths
))
4117 r
= lookup_paths_mkdir_generator(&m
->lookup_paths
);
4119 log_error_errno(r
, "Failed to create generator directories: %m");
4123 argv
[0] = NULL
; /* Leave this empty, execute_directory() will fill something in */
4124 argv
[1] = m
->lookup_paths
.generator
;
4125 argv
[2] = m
->lookup_paths
.generator_early
;
4126 argv
[3] = m
->lookup_paths
.generator_late
;
4129 RUN_WITH_UMASK(0022)
4130 (void) execute_directories((const char* const*) paths
, DEFAULT_TIMEOUT_USEC
, NULL
, NULL
,
4131 (char**) argv
, m
->transient_environment
,
4132 EXEC_DIR_PARALLEL
| EXEC_DIR_IGNORE_ERRORS
| EXEC_DIR_SET_SYSTEMD_EXEC_PID
);
4137 lookup_paths_trim_generator(&m
->lookup_paths
);
4141 int manager_transient_environment_add(Manager
*m
, char **plus
) {
4146 if (strv_isempty(plus
))
4149 a
= strv_env_merge(2, m
->transient_environment
, plus
);
4153 sanitize_environment(a
);
4155 return strv_free_and_replace(m
->transient_environment
, a
);
4158 int manager_client_environment_modify(
4163 char **a
= NULL
, **b
= NULL
, **l
;
4167 if (strv_isempty(minus
) && strv_isempty(plus
))
4170 l
= m
->client_environment
;
4172 if (!strv_isempty(minus
)) {
4173 a
= strv_env_delete(l
, 1, minus
);
4180 if (!strv_isempty(plus
)) {
4181 b
= strv_env_merge(2, l
, plus
);
4190 if (m
->client_environment
!= l
)
4191 strv_free(m
->client_environment
);
4198 m
->client_environment
= sanitize_environment(l
);
4202 int manager_get_effective_environment(Manager
*m
, char ***ret
) {
4208 l
= strv_env_merge(2, m
->transient_environment
, m
->client_environment
);
4216 int manager_set_default_rlimits(Manager
*m
, struct rlimit
**default_rlimit
) {
4219 for (unsigned i
= 0; i
< _RLIMIT_MAX
; i
++) {
4220 m
->rlimit
[i
] = mfree(m
->rlimit
[i
]);
4222 if (!default_rlimit
[i
])
4225 m
->rlimit
[i
] = newdup(struct rlimit
, default_rlimit
[i
], 1);
4233 void manager_recheck_dbus(Manager
*m
) {
4236 /* Connects to the bus if the dbus service and socket are running. If we are running in user mode this is all
4237 * it does. In system mode we'll also connect to the system bus (which will most likely just reuse the
4238 * connection of the API bus). That's because the system bus after all runs as service of the system instance,
4239 * while in the user instance we can assume it's already there. */
4241 if (MANAGER_IS_RELOADING(m
))
4242 return; /* don't check while we are reloading… */
4244 if (manager_dbus_is_running(m
, false)) {
4245 (void) bus_init_api(m
);
4247 if (MANAGER_IS_SYSTEM(m
))
4248 (void) bus_init_system(m
);
4250 (void) bus_done_api(m
);
4252 if (MANAGER_IS_SYSTEM(m
))
4253 (void) bus_done_system(m
);
4257 static bool manager_journal_is_running(Manager
*m
) {
4262 if (MANAGER_IS_TEST_RUN(m
))
4265 /* If we are the user manager we can safely assume that the journal is up */
4266 if (!MANAGER_IS_SYSTEM(m
))
4269 /* Check that the socket is not only up, but in RUNNING state */
4270 u
= manager_get_unit(m
, SPECIAL_JOURNALD_SOCKET
);
4273 if (SOCKET(u
)->state
!= SOCKET_RUNNING
)
4276 /* Similar, check if the daemon itself is fully up, too */
4277 u
= manager_get_unit(m
, SPECIAL_JOURNALD_SERVICE
);
4280 if (!IN_SET(SERVICE(u
)->state
, SERVICE_RELOAD
, SERVICE_RUNNING
))
4286 void disable_printk_ratelimit(void) {
4287 /* Disable kernel's printk ratelimit.
4289 * Logging to /dev/kmsg is most useful during early boot and shutdown, where normal logging
4290 * mechanisms are not available. The semantics of this sysctl are such that any kernel command-line
4291 * setting takes precedence. */
4294 r
= sysctl_write("kernel/printk_devkmsg", "on");
4296 log_debug_errno(r
, "Failed to set sysctl kernel.printk_devkmsg=on: %m");
4299 void manager_recheck_journal(Manager
*m
) {
4303 /* Don't bother with this unless we are in the special situation of being PID 1 */
4304 if (getpid_cached() != 1)
4307 /* Don't check this while we are reloading, things might still change */
4308 if (MANAGER_IS_RELOADING(m
))
4311 /* The journal is fully and entirely up? If so, let's permit logging to it, if that's configured. If the
4312 * journal is down, don't ever log to it, otherwise we might end up deadlocking ourselves as we might trigger
4313 * an activation ourselves we can't fulfill. */
4314 log_set_prohibit_ipc(!manager_journal_is_running(m
));
4318 static ShowStatus
manager_get_show_status(Manager
*m
) {
4321 if (MANAGER_IS_USER(m
))
4322 return _SHOW_STATUS_INVALID
;
4324 if (m
->show_status_overridden
!= _SHOW_STATUS_INVALID
)
4325 return m
->show_status_overridden
;
4327 return m
->show_status
;
4330 bool manager_get_show_status_on(Manager
*m
) {
4333 return show_status_on(manager_get_show_status(m
));
4336 static void set_show_status_marker(bool b
) {
4338 (void) touch("/run/systemd/show-status");
4340 (void) unlink("/run/systemd/show-status");
4343 void manager_set_show_status(Manager
*m
, ShowStatus mode
, const char *reason
) {
4346 assert(mode
>= 0 && mode
< _SHOW_STATUS_MAX
);
4348 if (MANAGER_IS_USER(m
))
4351 if (mode
== m
->show_status
)
4354 if (m
->show_status_overridden
== _SHOW_STATUS_INVALID
) {
4357 enabled
= show_status_on(mode
);
4358 log_debug("%s (%s) showing of status (%s).",
4359 enabled
? "Enabling" : "Disabling",
4360 strna(show_status_to_string(mode
)),
4363 set_show_status_marker(enabled
);
4366 m
->show_status
= mode
;
4369 void manager_override_show_status(Manager
*m
, ShowStatus mode
, const char *reason
) {
4371 assert(mode
< _SHOW_STATUS_MAX
);
4373 if (MANAGER_IS_USER(m
))
4376 if (mode
== m
->show_status_overridden
)
4379 m
->show_status_overridden
= mode
;
4381 if (mode
== _SHOW_STATUS_INVALID
)
4382 mode
= m
->show_status
;
4384 log_debug("%s (%s) showing of status (%s).",
4385 m
->show_status_overridden
!= _SHOW_STATUS_INVALID
? "Overriding" : "Restoring",
4386 strna(show_status_to_string(mode
)),
4389 set_show_status_marker(show_status_on(mode
));
4392 const char *manager_get_confirm_spawn(Manager
*m
) {
4393 static int last_errno
= 0;
4399 /* Here's the deal: we want to test the validity of the console but don't want
4400 * PID1 to go through the whole console process which might block. But we also
4401 * want to warn the user only once if something is wrong with the console so we
4402 * cannot do the sanity checks after spawning our children. So here we simply do
4403 * really basic tests to hopefully trap common errors.
4405 * If the console suddenly disappear at the time our children will really it
4406 * then they will simply fail to acquire it and a positive answer will be
4407 * assumed. New children will fall back to /dev/console though.
4409 * Note: TTYs are devices that can come and go any time, and frequently aren't
4410 * available yet during early boot (consider a USB rs232 dongle...). If for any
4411 * reason the configured console is not ready, we fall back to the default
4414 if (!m
->confirm_spawn
|| path_equal(m
->confirm_spawn
, "/dev/console"))
4415 return m
->confirm_spawn
;
4417 if (stat(m
->confirm_spawn
, &st
) < 0) {
4422 if (!S_ISCHR(st
.st_mode
)) {
4428 return m
->confirm_spawn
;
4431 if (last_errno
!= r
)
4432 last_errno
= log_warning_errno(r
, "Failed to open %s, using default console: %m", m
->confirm_spawn
);
4434 return "/dev/console";
4437 void manager_set_first_boot(Manager
*m
, bool b
) {
4440 if (!MANAGER_IS_SYSTEM(m
))
4443 if (m
->first_boot
!= (int) b
) {
4445 (void) touch("/run/systemd/first-boot");
4447 (void) unlink("/run/systemd/first-boot");
4453 void manager_disable_confirm_spawn(void) {
4454 (void) touch("/run/systemd/confirm_spawn_disabled");
4457 bool manager_is_confirm_spawn_disabled(Manager
*m
) {
4458 if (!m
->confirm_spawn
)
4461 return access("/run/systemd/confirm_spawn_disabled", F_OK
) >= 0;
4464 static bool manager_should_show_status(Manager
*m
, StatusType type
) {
4467 if (!MANAGER_IS_SYSTEM(m
))
4470 if (m
->no_console_output
)
4473 if (!IN_SET(manager_state(m
), MANAGER_INITIALIZING
, MANAGER_STARTING
, MANAGER_STOPPING
))
4476 /* If we cannot find out the status properly, just proceed. */
4477 if (type
!= STATUS_TYPE_EMERGENCY
&& manager_check_ask_password(m
) > 0)
4480 if (type
== STATUS_TYPE_NOTICE
&& m
->show_status
!= SHOW_STATUS_NO
)
4483 return manager_get_show_status_on(m
);
4486 void manager_status_printf(Manager
*m
, StatusType type
, const char *status
, const char *format
, ...) {
4489 /* If m is NULL, assume we're after shutdown and let the messages through. */
4491 if (m
&& !manager_should_show_status(m
, type
))
4494 /* XXX We should totally drop the check for ephemeral here
4495 * and thus effectively make 'Type=idle' pointless. */
4496 if (type
== STATUS_TYPE_EPHEMERAL
&& m
&& m
->n_on_console
> 0)
4499 va_start(ap
, format
);
4500 status_vprintf(status
, SHOW_STATUS_ELLIPSIZE
|(type
== STATUS_TYPE_EPHEMERAL
? SHOW_STATUS_EPHEMERAL
: 0), format
, ap
);
4504 Set
*manager_get_units_requiring_mounts_for(Manager
*m
, const char *path
) {
4505 char p
[strlen(path
)+1];
4511 path_simplify(p
, false);
4513 return hashmap_get(m
->units_requiring_mounts_for
, streq(p
, "/") ? "" : p
);
4516 int manager_update_failed_units(Manager
*m
, Unit
*u
, bool failed
) {
4521 assert(u
->manager
== m
);
4523 size
= set_size(m
->failed_units
);
4526 r
= set_ensure_put(&m
->failed_units
, NULL
, u
);
4530 (void) set_remove(m
->failed_units
, u
);
4532 if (set_size(m
->failed_units
) != size
)
4533 bus_manager_send_change_signal(m
);
4538 ManagerState
manager_state(Manager
*m
) {
4543 /* Is the special shutdown target active or queued? If so, we are in shutdown state */
4544 u
= manager_get_unit(m
, SPECIAL_SHUTDOWN_TARGET
);
4545 if (u
&& unit_active_or_pending(u
))
4546 return MANAGER_STOPPING
;
4548 /* Did we ever finish booting? If not then we are still starting up */
4549 if (!MANAGER_IS_FINISHED(m
)) {
4551 u
= manager_get_unit(m
, SPECIAL_BASIC_TARGET
);
4552 if (!u
|| !UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u
)))
4553 return MANAGER_INITIALIZING
;
4555 return MANAGER_STARTING
;
4558 if (MANAGER_IS_SYSTEM(m
)) {
4559 /* Are the rescue or emergency targets active or queued? If so we are in maintenance state */
4560 u
= manager_get_unit(m
, SPECIAL_RESCUE_TARGET
);
4561 if (u
&& unit_active_or_pending(u
))
4562 return MANAGER_MAINTENANCE
;
4564 u
= manager_get_unit(m
, SPECIAL_EMERGENCY_TARGET
);
4565 if (u
&& unit_active_or_pending(u
))
4566 return MANAGER_MAINTENANCE
;
4569 /* Are there any failed units? If so, we are in degraded mode */
4570 if (set_size(m
->failed_units
) > 0)
4571 return MANAGER_DEGRADED
;
4573 return MANAGER_RUNNING
;
4576 static void manager_unref_uid_internal(
4580 int (*_clean_ipc
)(uid_t uid
)) {
4584 assert(uid_is_valid(uid
));
4587 /* A generic implementation, covering both manager_unref_uid() and manager_unref_gid(), under the assumption
4588 * that uid_t and gid_t are actually defined the same way, with the same validity rules.
4590 * We store a hashmap where the UID/GID is they key and the value is a 32bit reference counter, whose highest
4591 * bit is used as flag for marking UIDs/GIDs whose IPC objects to remove when the last reference to the UID/GID
4592 * is dropped. The flag is set to on, once at least one reference from a unit where RemoveIPC= is set is added
4593 * on a UID/GID. It is reset when the UID's/GID's reference counter drops to 0 again. */
4595 assert_cc(sizeof(uid_t
) == sizeof(gid_t
));
4596 assert_cc(UID_INVALID
== (uid_t
) GID_INVALID
);
4598 if (uid
== 0) /* We don't keep track of root, and will never destroy it */
4601 c
= PTR_TO_UINT32(hashmap_get(uid_refs
, UID_TO_PTR(uid
)));
4603 n
= c
& ~DESTROY_IPC_FLAG
;
4607 if (destroy_now
&& n
== 0) {
4608 hashmap_remove(uid_refs
, UID_TO_PTR(uid
));
4610 if (c
& DESTROY_IPC_FLAG
) {
4611 log_debug("%s " UID_FMT
" is no longer referenced, cleaning up its IPC.",
4612 _clean_ipc
== clean_ipc_by_uid
? "UID" : "GID",
4614 (void) _clean_ipc(uid
);
4617 c
= n
| (c
& DESTROY_IPC_FLAG
);
4618 assert_se(hashmap_update(uid_refs
, UID_TO_PTR(uid
), UINT32_TO_PTR(c
)) >= 0);
4622 void manager_unref_uid(Manager
*m
, uid_t uid
, bool destroy_now
) {
4623 manager_unref_uid_internal(m
->uid_refs
, uid
, destroy_now
, clean_ipc_by_uid
);
4626 void manager_unref_gid(Manager
*m
, gid_t gid
, bool destroy_now
) {
4627 manager_unref_uid_internal(m
->gid_refs
, (uid_t
) gid
, destroy_now
, clean_ipc_by_gid
);
4630 static int manager_ref_uid_internal(
4639 assert(uid_is_valid(uid
));
4641 /* A generic implementation, covering both manager_ref_uid() and manager_ref_gid(), under the assumption
4642 * that uid_t and gid_t are actually defined the same way, with the same validity rules. */
4644 assert_cc(sizeof(uid_t
) == sizeof(gid_t
));
4645 assert_cc(UID_INVALID
== (uid_t
) GID_INVALID
);
4647 if (uid
== 0) /* We don't keep track of root, and will never destroy it */
4650 r
= hashmap_ensure_allocated(uid_refs
, &trivial_hash_ops
);
4654 c
= PTR_TO_UINT32(hashmap_get(*uid_refs
, UID_TO_PTR(uid
)));
4656 n
= c
& ~DESTROY_IPC_FLAG
;
4659 if (n
& DESTROY_IPC_FLAG
) /* check for overflow */
4662 c
= n
| (c
& DESTROY_IPC_FLAG
) | (clean_ipc
? DESTROY_IPC_FLAG
: 0);
4664 return hashmap_replace(*uid_refs
, UID_TO_PTR(uid
), UINT32_TO_PTR(c
));
4667 int manager_ref_uid(Manager
*m
, uid_t uid
, bool clean_ipc
) {
4668 return manager_ref_uid_internal(&m
->uid_refs
, uid
, clean_ipc
);
4671 int manager_ref_gid(Manager
*m
, gid_t gid
, bool clean_ipc
) {
4672 return manager_ref_uid_internal(&m
->gid_refs
, (uid_t
) gid
, clean_ipc
);
4675 static void manager_vacuum_uid_refs_internal(
4677 int (*_clean_ipc
)(uid_t uid
)) {
4683 HASHMAP_FOREACH_KEY(p
, k
, uid_refs
) {
4687 uid
= PTR_TO_UID(k
);
4688 c
= PTR_TO_UINT32(p
);
4690 n
= c
& ~DESTROY_IPC_FLAG
;
4694 if (c
& DESTROY_IPC_FLAG
) {
4695 log_debug("Found unreferenced %s " UID_FMT
" after reload/reexec. Cleaning up.",
4696 _clean_ipc
== clean_ipc_by_uid
? "UID" : "GID",
4698 (void) _clean_ipc(uid
);
4701 assert_se(hashmap_remove(uid_refs
, k
) == p
);
4705 static void manager_vacuum_uid_refs(Manager
*m
) {
4706 manager_vacuum_uid_refs_internal(m
->uid_refs
, clean_ipc_by_uid
);
4709 static void manager_vacuum_gid_refs(Manager
*m
) {
4710 manager_vacuum_uid_refs_internal(m
->gid_refs
, clean_ipc_by_gid
);
4713 static void manager_vacuum(Manager
*m
) {
4716 /* Release any dynamic users no longer referenced */
4717 dynamic_user_vacuum(m
, true);
4719 /* Release any references to UIDs/GIDs no longer referenced, and destroy any IPC owned by them */
4720 manager_vacuum_uid_refs(m
);
4721 manager_vacuum_gid_refs(m
);
4723 /* Release any runtimes no longer referenced */
4724 exec_runtime_vacuum(m
);
4727 int manager_dispatch_user_lookup_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
4731 char unit_name
[UNIT_NAME_MAX
+1];
4734 Manager
*m
= userdata
;
4742 /* Invoked whenever a child process succeeded resolving its user/group to use and sent us the resulting UID/GID
4743 * in a datagram. We parse the datagram here and pass it off to the unit, so that it can add a reference to the
4744 * UID/GID so that it can destroy the UID/GID's IPC objects when the reference counter drops to 0. */
4746 l
= recv(fd
, &buffer
, sizeof(buffer
), MSG_DONTWAIT
);
4748 if (IN_SET(errno
, EINTR
, EAGAIN
))
4751 return log_error_errno(errno
, "Failed to read from user lookup fd: %m");
4754 if ((size_t) l
<= offsetof(struct buffer
, unit_name
)) {
4755 log_warning("Received too short user lookup message, ignoring.");
4759 if ((size_t) l
> offsetof(struct buffer
, unit_name
) + UNIT_NAME_MAX
) {
4760 log_warning("Received too long user lookup message, ignoring.");
4764 if (!uid_is_valid(buffer
.uid
) && !gid_is_valid(buffer
.gid
)) {
4765 log_warning("Got user lookup message with invalid UID/GID pair, ignoring.");
4769 n
= (size_t) l
- offsetof(struct buffer
, unit_name
);
4770 if (memchr(buffer
.unit_name
, 0, n
)) {
4771 log_warning("Received lookup message with embedded NUL character, ignoring.");
4775 buffer
.unit_name
[n
] = 0;
4776 u
= manager_get_unit(m
, buffer
.unit_name
);
4778 log_debug("Got user lookup message but unit doesn't exist, ignoring.");
4782 log_unit_debug(u
, "User lookup succeeded: uid=" UID_FMT
" gid=" GID_FMT
, buffer
.uid
, buffer
.gid
);
4784 unit_notify_user_lookup(u
, buffer
.uid
, buffer
.gid
);
4788 char *manager_taint_string(Manager
*m
) {
4789 _cleanup_free_
char *destination
= NULL
, *overflowuid
= NULL
, *overflowgid
= NULL
;
4793 /* Returns a "taint string", e.g. "local-hwclock:var-run-bad".
4794 * Only things that are detected at runtime should be tagged
4795 * here. For stuff that is set during compilation, emit a warning
4796 * in the configuration phase. */
4800 buf
= new(char, sizeof("split-usr:"
4805 "overflowuid-not-65534:"
4806 "overflowgid-not-65534:"));
4814 e
= stpcpy(e
, "split-usr:");
4816 if (access("/proc/cgroups", F_OK
) < 0)
4817 e
= stpcpy(e
, "cgroups-missing:");
4819 if (cg_all_unified() == 0)
4820 e
= stpcpy(e
, "cgroupsv1:");
4822 if (clock_is_localtime(NULL
) > 0)
4823 e
= stpcpy(e
, "local-hwclock:");
4825 r
= readlink_malloc("/var/run", &destination
);
4826 if (r
< 0 || !PATH_IN_SET(destination
, "../run", "/run"))
4827 e
= stpcpy(e
, "var-run-bad:");
4829 r
= read_one_line_file("/proc/sys/kernel/overflowuid", &overflowuid
);
4830 if (r
>= 0 && !streq(overflowuid
, "65534"))
4831 e
= stpcpy(e
, "overflowuid-not-65534:");
4833 r
= read_one_line_file("/proc/sys/kernel/overflowgid", &overflowgid
);
4834 if (r
>= 0 && !streq(overflowgid
, "65534"))
4835 e
= stpcpy(e
, "overflowgid-not-65534:");
4837 /* remove the last ':' */
4844 void manager_ref_console(Manager
*m
) {
4850 void manager_unref_console(Manager
*m
) {
4852 assert(m
->n_on_console
> 0);
4855 if (m
->n_on_console
== 0)
4856 m
->no_console_output
= false; /* unset no_console_output flag, since the console is definitely free now */
4859 void manager_override_log_level(Manager
*m
, int level
) {
4860 _cleanup_free_
char *s
= NULL
;
4863 if (!m
->log_level_overridden
) {
4864 m
->original_log_level
= log_get_max_level();
4865 m
->log_level_overridden
= true;
4868 (void) log_level_to_string_alloc(level
, &s
);
4869 log_info("Setting log level to %s.", strna(s
));
4871 log_set_max_level(level
);
4874 void manager_restore_original_log_level(Manager
*m
) {
4875 _cleanup_free_
char *s
= NULL
;
4878 if (!m
->log_level_overridden
)
4881 (void) log_level_to_string_alloc(m
->original_log_level
, &s
);
4882 log_info("Restoring log level to original (%s).", strna(s
));
4884 log_set_max_level(m
->original_log_level
);
4885 m
->log_level_overridden
= false;
4888 void manager_override_log_target(Manager
*m
, LogTarget target
) {
4891 if (!m
->log_target_overridden
) {
4892 m
->original_log_target
= log_get_target();
4893 m
->log_target_overridden
= true;
4896 log_info("Setting log target to %s.", log_target_to_string(target
));
4897 log_set_target(target
);
4900 void manager_restore_original_log_target(Manager
*m
) {
4903 if (!m
->log_target_overridden
)
4906 log_info("Restoring log target to original %s.", log_target_to_string(m
->original_log_target
));
4908 log_set_target(m
->original_log_target
);
4909 m
->log_target_overridden
= false;
4912 ManagerTimestamp
manager_timestamp_initrd_mangle(ManagerTimestamp s
) {
4914 s
>= MANAGER_TIMESTAMP_SECURITY_START
&&
4915 s
<= MANAGER_TIMESTAMP_UNITS_LOAD_FINISH
)
4916 return s
- MANAGER_TIMESTAMP_SECURITY_START
+ MANAGER_TIMESTAMP_INITRD_SECURITY_START
;
4920 static const char *const manager_state_table
[_MANAGER_STATE_MAX
] = {
4921 [MANAGER_INITIALIZING
] = "initializing",
4922 [MANAGER_STARTING
] = "starting",
4923 [MANAGER_RUNNING
] = "running",
4924 [MANAGER_DEGRADED
] = "degraded",
4925 [MANAGER_MAINTENANCE
] = "maintenance",
4926 [MANAGER_STOPPING
] = "stopping",
4929 DEFINE_STRING_TABLE_LOOKUP(manager_state
, ManagerState
);
4931 static const char *const manager_timestamp_table
[_MANAGER_TIMESTAMP_MAX
] = {
4932 [MANAGER_TIMESTAMP_FIRMWARE
] = "firmware",
4933 [MANAGER_TIMESTAMP_LOADER
] = "loader",
4934 [MANAGER_TIMESTAMP_KERNEL
] = "kernel",
4935 [MANAGER_TIMESTAMP_INITRD
] = "initrd",
4936 [MANAGER_TIMESTAMP_USERSPACE
] = "userspace",
4937 [MANAGER_TIMESTAMP_FINISH
] = "finish",
4938 [MANAGER_TIMESTAMP_SECURITY_START
] = "security-start",
4939 [MANAGER_TIMESTAMP_SECURITY_FINISH
] = "security-finish",
4940 [MANAGER_TIMESTAMP_GENERATORS_START
] = "generators-start",
4941 [MANAGER_TIMESTAMP_GENERATORS_FINISH
] = "generators-finish",
4942 [MANAGER_TIMESTAMP_UNITS_LOAD_START
] = "units-load-start",
4943 [MANAGER_TIMESTAMP_UNITS_LOAD_FINISH
] = "units-load-finish",
4944 [MANAGER_TIMESTAMP_INITRD_SECURITY_START
] = "initrd-security-start",
4945 [MANAGER_TIMESTAMP_INITRD_SECURITY_FINISH
] = "initrd-security-finish",
4946 [MANAGER_TIMESTAMP_INITRD_GENERATORS_START
] = "initrd-generators-start",
4947 [MANAGER_TIMESTAMP_INITRD_GENERATORS_FINISH
] = "initrd-generators-finish",
4948 [MANAGER_TIMESTAMP_INITRD_UNITS_LOAD_START
] = "initrd-units-load-start",
4949 [MANAGER_TIMESTAMP_INITRD_UNITS_LOAD_FINISH
] = "initrd-units-load-finish",
4952 DEFINE_STRING_TABLE_LOOKUP(manager_timestamp
, ManagerTimestamp
);
4954 static const char* const oom_policy_table
[_OOM_POLICY_MAX
] = {
4955 [OOM_CONTINUE
] = "continue",
4956 [OOM_STOP
] = "stop",
4957 [OOM_KILL
] = "kill",
4960 DEFINE_STRING_TABLE_LOOKUP(oom_policy
, OOMPolicy
);