1 /* SPDX-License-Identifier: LGPL-2.1+ */
3 This file is part of systemd.
5 Copyright 2010 Lennart Poettering
12 #include <stdio_ext.h>
14 #include <sys/epoll.h>
15 #include <sys/inotify.h>
16 #include <sys/ioctl.h>
17 #include <sys/reboot.h>
18 #include <sys/timerfd.h>
26 #include "sd-daemon.h"
27 #include "sd-messages.h"
30 #include "alloc-util.h"
32 #include "boot-timestamps.h"
33 #include "bus-common-errors.h"
34 #include "bus-error.h"
35 #include "bus-kernel.h"
37 #include "clean-ipc.h"
38 #include "clock-util.h"
40 #include "dbus-manager.h"
41 #include "dbus-unit.h"
43 #include "dirent-util.h"
46 #include "exec-util.h"
48 #include "exit-status.h"
55 #include "locale-setup.h"
61 #include "parse-util.h"
62 #include "path-lookup.h"
63 #include "path-util.h"
64 #include "process-util.h"
65 #include "ratelimit.h"
67 #include "signal-util.h"
69 #include "stat-util.h"
70 #include "string-table.h"
71 #include "string-util.h"
74 #include "terminal-util.h"
75 #include "time-util.h"
76 #include "transaction.h"
77 #include "umask-util.h"
78 #include "unit-name.h"
79 #include "user-util.h"
84 #define NOTIFY_RCVBUF_SIZE (8*1024*1024)
85 #define CGROUPS_AGENT_RCVBUF_SIZE (8*1024*1024)
87 /* Initial delay and the interval for printing status messages about running jobs */
88 #define JOBS_IN_PROGRESS_WAIT_USEC (5*USEC_PER_SEC)
89 #define JOBS_IN_PROGRESS_PERIOD_USEC (USEC_PER_SEC / 3)
90 #define JOBS_IN_PROGRESS_PERIOD_DIVISOR 3
92 /* If there are more than 1K bus messages queue across our API and direct busses, then let's not add more on top until
93 * the queue gets more empty. */
94 #define MANAGER_BUS_BUSY_THRESHOLD 1024LU
96 /* How many units and jobs to process of the bus queue before returning to the event loop. */
97 #define MANAGER_BUS_MESSAGE_BUDGET 100U
99 static int manager_dispatch_notify_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
100 static int manager_dispatch_cgroups_agent_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
101 static int manager_dispatch_signal_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
102 static int manager_dispatch_time_change_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
103 static int manager_dispatch_idle_pipe_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
104 static int manager_dispatch_user_lookup_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
105 static int manager_dispatch_jobs_in_progress(sd_event_source
*source
, usec_t usec
, void *userdata
);
106 static int manager_dispatch_run_queue(sd_event_source
*source
, void *userdata
);
107 static int manager_dispatch_sigchld(sd_event_source
*source
, void *userdata
);
108 static int manager_run_environment_generators(Manager
*m
);
109 static int manager_run_generators(Manager
*m
);
111 static void manager_watch_jobs_in_progress(Manager
*m
) {
117 /* We do not want to show the cylon animation if the user
118 * needs to confirm service executions otherwise confirmation
119 * messages will be screwed by the cylon animation. */
120 if (!manager_is_confirm_spawn_disabled(m
))
123 if (m
->jobs_in_progress_event_source
)
126 next
= now(CLOCK_MONOTONIC
) + JOBS_IN_PROGRESS_WAIT_USEC
;
127 r
= sd_event_add_time(
129 &m
->jobs_in_progress_event_source
,
132 manager_dispatch_jobs_in_progress
, m
);
136 (void) sd_event_source_set_description(m
->jobs_in_progress_event_source
, "manager-jobs-in-progress");
139 #define CYLON_BUFFER_EXTRA (2*STRLEN(ANSI_RED) + STRLEN(ANSI_HIGHLIGHT_RED) + 2*STRLEN(ANSI_NORMAL))
141 static void draw_cylon(char buffer
[], size_t buflen
, unsigned width
, unsigned pos
) {
144 assert(buflen
>= CYLON_BUFFER_EXTRA
+ width
+ 1);
145 assert(pos
<= width
+1); /* 0 or width+1 mean that the center light is behind the corner */
149 p
= mempset(p
, ' ', pos
-2);
150 if (log_get_show_color())
151 p
= stpcpy(p
, ANSI_RED
);
155 if (pos
> 0 && pos
<= width
) {
156 if (log_get_show_color())
157 p
= stpcpy(p
, ANSI_HIGHLIGHT_RED
);
161 if (log_get_show_color())
162 p
= stpcpy(p
, ANSI_NORMAL
);
165 if (log_get_show_color())
166 p
= stpcpy(p
, ANSI_RED
);
169 p
= mempset(p
, ' ', width
-1-pos
);
170 if (log_get_show_color())
171 strcpy(p
, ANSI_NORMAL
);
175 void manager_flip_auto_status(Manager
*m
, bool enable
) {
179 if (m
->show_status
== SHOW_STATUS_AUTO
)
180 manager_set_show_status(m
, SHOW_STATUS_TEMPORARY
);
182 if (m
->show_status
== SHOW_STATUS_TEMPORARY
)
183 manager_set_show_status(m
, SHOW_STATUS_AUTO
);
187 static void manager_print_jobs_in_progress(Manager
*m
) {
188 _cleanup_free_
char *job_of_n
= NULL
;
191 unsigned counter
= 0, print_nr
;
192 char cylon
[6 + CYLON_BUFFER_EXTRA
+ 1];
194 char time
[FORMAT_TIMESPAN_MAX
], limit
[FORMAT_TIMESPAN_MAX
] = "no limit";
198 assert(m
->n_running_jobs
> 0);
200 manager_flip_auto_status(m
, true);
202 print_nr
= (m
->jobs_in_progress_iteration
/ JOBS_IN_PROGRESS_PERIOD_DIVISOR
) % m
->n_running_jobs
;
204 HASHMAP_FOREACH(j
, m
->jobs
, i
)
205 if (j
->state
== JOB_RUNNING
&& counter
++ == print_nr
)
208 /* m->n_running_jobs must be consistent with the contents of m->jobs,
209 * so the above loop must have succeeded in finding j. */
210 assert(counter
== print_nr
+ 1);
213 cylon_pos
= m
->jobs_in_progress_iteration
% 14;
215 cylon_pos
= 14 - cylon_pos
;
216 draw_cylon(cylon
, sizeof(cylon
), 6, cylon_pos
);
218 m
->jobs_in_progress_iteration
++;
220 if (m
->n_running_jobs
> 1) {
221 if (asprintf(&job_of_n
, "(%u of %u) ", counter
, m
->n_running_jobs
) < 0)
225 format_timespan(time
, sizeof(time
), now(CLOCK_MONOTONIC
) - j
->begin_usec
, 1*USEC_PER_SEC
);
226 if (job_get_timeout(j
, &x
) > 0)
227 format_timespan(limit
, sizeof(limit
), x
- j
->begin_usec
, 1*USEC_PER_SEC
);
229 manager_status_printf(m
, STATUS_TYPE_EPHEMERAL
, cylon
,
230 "%sA %s job is running for %s (%s / %s)",
232 job_type_to_string(j
->type
),
233 unit_description(j
->unit
),
237 static int have_ask_password(void) {
238 _cleanup_closedir_
DIR *dir
;
241 dir
= opendir("/run/systemd/ask-password");
249 FOREACH_DIRENT_ALL(de
, dir
, return -errno
) {
250 if (startswith(de
->d_name
, "ask."))
256 static int manager_dispatch_ask_password_fd(sd_event_source
*source
,
257 int fd
, uint32_t revents
, void *userdata
) {
258 Manager
*m
= userdata
;
264 m
->have_ask_password
= have_ask_password();
265 if (m
->have_ask_password
< 0)
266 /* Log error but continue. Negative have_ask_password
267 * is treated as unknown status. */
268 log_error_errno(m
->have_ask_password
, "Failed to list /run/systemd/ask-password: %m");
273 static void manager_close_ask_password(Manager
*m
) {
276 m
->ask_password_event_source
= sd_event_source_unref(m
->ask_password_event_source
);
277 m
->ask_password_inotify_fd
= safe_close(m
->ask_password_inotify_fd
);
278 m
->have_ask_password
= -EINVAL
;
281 static int manager_check_ask_password(Manager
*m
) {
286 if (!m
->ask_password_event_source
) {
287 assert(m
->ask_password_inotify_fd
< 0);
289 mkdir_p_label("/run/systemd/ask-password", 0755);
291 m
->ask_password_inotify_fd
= inotify_init1(IN_NONBLOCK
|IN_CLOEXEC
);
292 if (m
->ask_password_inotify_fd
< 0)
293 return log_error_errno(errno
, "inotify_init1() failed: %m");
295 if (inotify_add_watch(m
->ask_password_inotify_fd
, "/run/systemd/ask-password", IN_CREATE
|IN_DELETE
|IN_MOVE
) < 0) {
296 log_error_errno(errno
, "Failed to add watch on /run/systemd/ask-password: %m");
297 manager_close_ask_password(m
);
301 r
= sd_event_add_io(m
->event
, &m
->ask_password_event_source
,
302 m
->ask_password_inotify_fd
, EPOLLIN
,
303 manager_dispatch_ask_password_fd
, m
);
305 log_error_errno(errno
, "Failed to add event source for /run/systemd/ask-password: %m");
306 manager_close_ask_password(m
);
310 (void) sd_event_source_set_description(m
->ask_password_event_source
, "manager-ask-password");
312 /* Queries might have been added meanwhile... */
313 manager_dispatch_ask_password_fd(m
->ask_password_event_source
,
314 m
->ask_password_inotify_fd
, EPOLLIN
, m
);
317 return m
->have_ask_password
;
320 static int manager_watch_idle_pipe(Manager
*m
) {
325 if (m
->idle_pipe_event_source
)
328 if (m
->idle_pipe
[2] < 0)
331 r
= sd_event_add_io(m
->event
, &m
->idle_pipe_event_source
, m
->idle_pipe
[2], EPOLLIN
, manager_dispatch_idle_pipe_fd
, m
);
333 return log_error_errno(r
, "Failed to watch idle pipe: %m");
335 (void) sd_event_source_set_description(m
->idle_pipe_event_source
, "manager-idle-pipe");
340 static void manager_close_idle_pipe(Manager
*m
) {
343 m
->idle_pipe_event_source
= sd_event_source_unref(m
->idle_pipe_event_source
);
345 safe_close_pair(m
->idle_pipe
);
346 safe_close_pair(m
->idle_pipe
+ 2);
349 static int manager_setup_time_change(Manager
*m
) {
352 /* We only care for the cancellation event, hence we set the
353 * timeout to the latest possible value. */
354 struct itimerspec its
= {
355 .it_value
.tv_sec
= TIME_T_MAX
,
359 assert_cc(sizeof(time_t) == sizeof(TIME_T_MAX
));
361 if (m
->test_run_flags
)
364 /* Uses TFD_TIMER_CANCEL_ON_SET to get notifications whenever
365 * CLOCK_REALTIME makes a jump relative to CLOCK_MONOTONIC */
367 m
->time_change_fd
= timerfd_create(CLOCK_REALTIME
, TFD_NONBLOCK
|TFD_CLOEXEC
);
368 if (m
->time_change_fd
< 0)
369 return log_error_errno(errno
, "Failed to create timerfd: %m");
371 if (timerfd_settime(m
->time_change_fd
, TFD_TIMER_ABSTIME
|TFD_TIMER_CANCEL_ON_SET
, &its
, NULL
) < 0) {
372 log_debug_errno(errno
, "Failed to set up TFD_TIMER_CANCEL_ON_SET, ignoring: %m");
373 m
->time_change_fd
= safe_close(m
->time_change_fd
);
377 r
= sd_event_add_io(m
->event
, &m
->time_change_event_source
, m
->time_change_fd
, EPOLLIN
, manager_dispatch_time_change_fd
, m
);
379 return log_error_errno(r
, "Failed to create time change event source: %m");
381 (void) sd_event_source_set_description(m
->time_change_event_source
, "manager-time-change");
383 log_debug("Set up TFD_TIMER_CANCEL_ON_SET timerfd.");
388 static int enable_special_signals(Manager
*m
) {
389 _cleanup_close_
int fd
= -1;
393 if (m
->test_run_flags
)
396 /* Enable that we get SIGINT on control-alt-del. In containers
397 * this will fail with EPERM (older) or EINVAL (newer), so
399 if (reboot(RB_DISABLE_CAD
) < 0 && !IN_SET(errno
, EPERM
, EINVAL
))
400 log_warning_errno(errno
, "Failed to enable ctrl-alt-del handling: %m");
402 fd
= open_terminal("/dev/tty0", O_RDWR
|O_NOCTTY
|O_CLOEXEC
);
404 /* Support systems without virtual console */
406 log_warning_errno(errno
, "Failed to open /dev/tty0: %m");
408 /* Enable that we get SIGWINCH on kbrequest */
409 if (ioctl(fd
, KDSIGACCEPT
, SIGWINCH
) < 0)
410 log_warning_errno(errno
, "Failed to enable kbrequest handling: %m");
416 #define RTSIG_IF_AVAILABLE(signum) (signum <= SIGRTMAX ? signum : -1)
418 static int manager_setup_signals(Manager
*m
) {
419 struct sigaction sa
= {
420 .sa_handler
= SIG_DFL
,
421 .sa_flags
= SA_NOCLDSTOP
|SA_RESTART
,
428 assert_se(sigaction(SIGCHLD
, &sa
, NULL
) == 0);
430 /* We make liberal use of realtime signals here. On
431 * Linux/glibc we have 30 of them (with the exception of Linux
432 * on hppa, see below), between SIGRTMIN+0 ... SIGRTMIN+30
435 assert_se(sigemptyset(&mask
) == 0);
436 sigset_add_many(&mask
,
437 SIGCHLD
, /* Child died */
438 SIGTERM
, /* Reexecute daemon */
439 SIGHUP
, /* Reload configuration */
440 SIGUSR1
, /* systemd/upstart: reconnect to D-Bus */
441 SIGUSR2
, /* systemd: dump status */
442 SIGINT
, /* Kernel sends us this on control-alt-del */
443 SIGWINCH
, /* Kernel sends us this on kbrequest (alt-arrowup) */
444 SIGPWR
, /* Some kernel drivers and upsd send us this on power failure */
446 SIGRTMIN
+0, /* systemd: start default.target */
447 SIGRTMIN
+1, /* systemd: isolate rescue.target */
448 SIGRTMIN
+2, /* systemd: isolate emergency.target */
449 SIGRTMIN
+3, /* systemd: start halt.target */
450 SIGRTMIN
+4, /* systemd: start poweroff.target */
451 SIGRTMIN
+5, /* systemd: start reboot.target */
452 SIGRTMIN
+6, /* systemd: start kexec.target */
454 /* ... space for more special targets ... */
456 SIGRTMIN
+13, /* systemd: Immediate halt */
457 SIGRTMIN
+14, /* systemd: Immediate poweroff */
458 SIGRTMIN
+15, /* systemd: Immediate reboot */
459 SIGRTMIN
+16, /* systemd: Immediate kexec */
461 /* ... space for more immediate system state changes ... */
463 SIGRTMIN
+20, /* systemd: enable status messages */
464 SIGRTMIN
+21, /* systemd: disable status messages */
465 SIGRTMIN
+22, /* systemd: set log level to LOG_DEBUG */
466 SIGRTMIN
+23, /* systemd: set log level to LOG_INFO */
467 SIGRTMIN
+24, /* systemd: Immediate exit (--user only) */
469 /* .. one free signal here ... */
471 /* Apparently Linux on hppa had fewer RT signals until v3.18,
472 * SIGRTMAX was SIGRTMIN+25, and then SIGRTMIN was lowered,
473 * see commit v3.17-7614-g1f25df2eff.
475 * We cannot unconditionally make use of those signals here,
476 * so let's use a runtime check. Since these commands are
477 * accessible by different means and only really a safety
478 * net, the missing functionality on hppa shouldn't matter.
481 RTSIG_IF_AVAILABLE(SIGRTMIN
+26), /* systemd: set log target to journal-or-kmsg */
482 RTSIG_IF_AVAILABLE(SIGRTMIN
+27), /* systemd: set log target to console */
483 RTSIG_IF_AVAILABLE(SIGRTMIN
+28), /* systemd: set log target to kmsg */
484 RTSIG_IF_AVAILABLE(SIGRTMIN
+29), /* systemd: set log target to syslog-or-kmsg (obsolete) */
486 /* ... one free signal here SIGRTMIN+30 ... */
488 assert_se(sigprocmask(SIG_SETMASK
, &mask
, NULL
) == 0);
490 m
->signal_fd
= signalfd(-1, &mask
, SFD_NONBLOCK
|SFD_CLOEXEC
);
491 if (m
->signal_fd
< 0)
494 r
= sd_event_add_io(m
->event
, &m
->signal_event_source
, m
->signal_fd
, EPOLLIN
, manager_dispatch_signal_fd
, m
);
498 (void) sd_event_source_set_description(m
->signal_event_source
, "manager-signal");
500 /* Process signals a bit earlier than the rest of things, but later than notify_fd processing, so that the
501 * notify processing can still figure out to which process/service a message belongs, before we reap the
502 * process. Also, process this before handling cgroup notifications, so that we always collect child exit
503 * status information before detecting that there's no process in a cgroup. */
504 r
= sd_event_source_set_priority(m
->signal_event_source
, SD_EVENT_PRIORITY_NORMAL
-6);
508 if (MANAGER_IS_SYSTEM(m
))
509 return enable_special_signals(m
);
514 static void manager_sanitize_environment(Manager
*m
) {
517 /* Let's remove some environment variables that we need ourselves to communicate with our clients */
537 /* Let's order the environment alphabetically, just to make it pretty */
538 strv_sort(m
->environment
);
541 static int manager_default_environment(Manager
*m
) {
544 if (MANAGER_IS_SYSTEM(m
)) {
545 /* The system manager always starts with a clean
546 * environment for its children. It does not import
547 * the kernel's or the parents' exported variables.
549 * The initial passed environment is untouched to keep
550 * /proc/self/environ valid; it is used for tagging
551 * the init process inside containers. */
552 m
->environment
= strv_new("PATH=" DEFAULT_PATH
,
555 /* Import locale variables LC_*= from configuration */
556 locale_setup(&m
->environment
);
558 /* The user manager passes its own environment
559 * along to its children. */
560 m
->environment
= strv_copy(environ
);
565 manager_sanitize_environment(m
);
570 static int manager_setup_prefix(Manager
*m
) {
576 static const struct table_entry paths_system
[_EXEC_DIRECTORY_TYPE_MAX
] = {
577 [EXEC_DIRECTORY_RUNTIME
] = { SD_PATH_SYSTEM_RUNTIME
, NULL
},
578 [EXEC_DIRECTORY_STATE
] = { SD_PATH_SYSTEM_STATE_PRIVATE
, NULL
},
579 [EXEC_DIRECTORY_CACHE
] = { SD_PATH_SYSTEM_STATE_CACHE
, NULL
},
580 [EXEC_DIRECTORY_LOGS
] = { SD_PATH_SYSTEM_STATE_LOGS
, NULL
},
581 [EXEC_DIRECTORY_CONFIGURATION
] = { SD_PATH_SYSTEM_CONFIGURATION
, NULL
},
584 static const struct table_entry paths_user
[_EXEC_DIRECTORY_TYPE_MAX
] = {
585 [EXEC_DIRECTORY_RUNTIME
] = { SD_PATH_USER_RUNTIME
, NULL
},
586 [EXEC_DIRECTORY_STATE
] = { SD_PATH_USER_CONFIGURATION
, NULL
},
587 [EXEC_DIRECTORY_CACHE
] = { SD_PATH_USER_STATE_CACHE
, NULL
},
588 [EXEC_DIRECTORY_LOGS
] = { SD_PATH_USER_CONFIGURATION
, "log" },
589 [EXEC_DIRECTORY_CONFIGURATION
] = { SD_PATH_USER_CONFIGURATION
, NULL
},
592 const struct table_entry
*p
;
598 if (MANAGER_IS_SYSTEM(m
))
603 for (i
= 0; i
< _EXEC_DIRECTORY_TYPE_MAX
; i
++) {
604 r
= sd_path_home(p
[i
].type
, p
[i
].suffix
, &m
->prefix
[i
]);
612 static int manager_setup_run_queue(Manager
*m
) {
616 assert(!m
->run_queue_event_source
);
618 r
= sd_event_add_defer(m
->event
, &m
->run_queue_event_source
, manager_dispatch_run_queue
, m
);
622 r
= sd_event_source_set_priority(m
->run_queue_event_source
, SD_EVENT_PRIORITY_IDLE
);
626 r
= sd_event_source_set_enabled(m
->run_queue_event_source
, SD_EVENT_OFF
);
630 (void) sd_event_source_set_description(m
->run_queue_event_source
, "manager-run-queue");
635 static int manager_setup_sigchld_event_source(Manager
*m
) {
639 assert(!m
->sigchld_event_source
);
641 r
= sd_event_add_defer(m
->event
, &m
->sigchld_event_source
, manager_dispatch_sigchld
, m
);
645 r
= sd_event_source_set_priority(m
->sigchld_event_source
, SD_EVENT_PRIORITY_NORMAL
-7);
649 r
= sd_event_source_set_enabled(m
->sigchld_event_source
, SD_EVENT_OFF
);
653 (void) sd_event_source_set_description(m
->sigchld_event_source
, "manager-sigchld");
658 int manager_new(UnitFileScope scope
, unsigned test_run_flags
, Manager
**_m
) {
659 _cleanup_(manager_freep
) Manager
*m
= NULL
;
663 assert(IN_SET(scope
, UNIT_FILE_SYSTEM
, UNIT_FILE_USER
));
665 m
= new0(Manager
, 1);
669 m
->unit_file_scope
= scope
;
670 m
->exit_code
= _MANAGER_EXIT_CODE_INVALID
;
671 m
->default_timer_accuracy_usec
= USEC_PER_MINUTE
;
672 m
->default_memory_accounting
= MEMORY_ACCOUNTING_DEFAULT
;
673 m
->default_tasks_accounting
= true;
674 m
->default_tasks_max
= UINT64_MAX
;
675 m
->default_timeout_start_usec
= DEFAULT_TIMEOUT_USEC
;
676 m
->default_timeout_stop_usec
= DEFAULT_TIMEOUT_USEC
;
677 m
->default_restart_usec
= DEFAULT_RESTART_USEC
;
680 if (MANAGER_IS_SYSTEM(m
) && detect_container() <= 0)
681 boot_timestamps(m
->timestamps
+ MANAGER_TIMESTAMP_USERSPACE
,
682 m
->timestamps
+ MANAGER_TIMESTAMP_FIRMWARE
,
683 m
->timestamps
+ MANAGER_TIMESTAMP_LOADER
);
686 /* Prepare log fields we can use for structured logging */
687 if (MANAGER_IS_SYSTEM(m
)) {
688 m
->unit_log_field
= "UNIT=";
689 m
->unit_log_format_string
= "UNIT=%s";
691 m
->invocation_log_field
= "INVOCATION_ID=";
692 m
->invocation_log_format_string
= "INVOCATION_ID=%s";
694 m
->unit_log_field
= "USER_UNIT=";
695 m
->unit_log_format_string
= "USER_UNIT=%s";
697 m
->invocation_log_field
= "USER_INVOCATION_ID=";
698 m
->invocation_log_format_string
= "USER_INVOCATION_ID=%s";
701 m
->idle_pipe
[0] = m
->idle_pipe
[1] = m
->idle_pipe
[2] = m
->idle_pipe
[3] = -1;
703 m
->pin_cgroupfs_fd
= m
->notify_fd
= m
->cgroups_agent_fd
= m
->signal_fd
= m
->time_change_fd
=
704 m
->dev_autofs_fd
= m
->private_listen_fd
= m
->cgroup_inotify_fd
=
705 m
->ask_password_inotify_fd
= -1;
707 m
->user_lookup_fds
[0] = m
->user_lookup_fds
[1] = -1;
709 m
->current_job_id
= 1; /* start as id #1, so that we can leave #0 around as "null-like" value */
711 m
->have_ask_password
= -EINVAL
; /* we don't know */
714 m
->test_run_flags
= test_run_flags
;
716 /* Reboot immediately if the user hits C-A-D more often than 7x per 2s */
717 RATELIMIT_INIT(m
->ctrl_alt_del_ratelimit
, 2 * USEC_PER_SEC
, 7);
719 r
= manager_default_environment(m
);
723 r
= hashmap_ensure_allocated(&m
->units
, &string_hash_ops
);
727 r
= hashmap_ensure_allocated(&m
->jobs
, NULL
);
731 r
= hashmap_ensure_allocated(&m
->cgroup_unit
, &path_hash_ops
);
735 r
= hashmap_ensure_allocated(&m
->watch_bus
, &string_hash_ops
);
739 r
= manager_setup_prefix(m
);
743 m
->udev
= udev_new();
747 r
= sd_event_default(&m
->event
);
751 r
= manager_setup_run_queue(m
);
755 if (test_run_flags
== MANAGER_TEST_RUN_MINIMAL
) {
756 m
->cgroup_root
= strdup("");
760 r
= manager_setup_signals(m
);
764 r
= manager_setup_cgroup(m
);
768 r
= manager_setup_time_change(m
);
772 r
= manager_setup_sigchld_event_source(m
);
777 if (MANAGER_IS_SYSTEM(m
) && test_run_flags
== 0) {
778 r
= mkdir_label("/run/systemd/units", 0755);
779 if (r
< 0 && r
!= -EEXIST
)
785 dir_is_empty("/usr") > 0;
787 /* Note that we do not set up the notify fd here. We do that after deserialization,
788 * since they might have gotten serialized across the reexec. */
795 static int manager_setup_notify(Manager
*m
) {
798 if (m
->test_run_flags
)
801 if (m
->notify_fd
< 0) {
802 _cleanup_close_
int fd
= -1;
803 union sockaddr_union sa
= {
804 .sa
.sa_family
= AF_UNIX
,
806 static const int one
= 1;
808 /* First free all secondary fields */
809 m
->notify_socket
= mfree(m
->notify_socket
);
810 m
->notify_event_source
= sd_event_source_unref(m
->notify_event_source
);
812 fd
= socket(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
|SOCK_NONBLOCK
, 0);
814 return log_error_errno(errno
, "Failed to allocate notification socket: %m");
816 fd_inc_rcvbuf(fd
, NOTIFY_RCVBUF_SIZE
);
818 m
->notify_socket
= strappend(m
->prefix
[EXEC_DIRECTORY_RUNTIME
], "/systemd/notify");
819 if (!m
->notify_socket
)
822 (void) mkdir_parents_label(m
->notify_socket
, 0755);
823 (void) unlink(m
->notify_socket
);
825 strncpy(sa
.un
.sun_path
, m
->notify_socket
, sizeof(sa
.un
.sun_path
)-1);
826 r
= bind(fd
, &sa
.sa
, SOCKADDR_UN_LEN(sa
.un
));
828 return log_error_errno(errno
, "bind(%s) failed: %m", sa
.un
.sun_path
);
830 r
= setsockopt(fd
, SOL_SOCKET
, SO_PASSCRED
, &one
, sizeof(one
));
832 return log_error_errno(errno
, "SO_PASSCRED failed: %m");
834 m
->notify_fd
= TAKE_FD(fd
);
836 log_debug("Using notification socket %s", m
->notify_socket
);
839 if (!m
->notify_event_source
) {
840 r
= sd_event_add_io(m
->event
, &m
->notify_event_source
, m
->notify_fd
, EPOLLIN
, manager_dispatch_notify_fd
, m
);
842 return log_error_errno(r
, "Failed to allocate notify event source: %m");
844 /* Process notification messages a bit earlier than SIGCHLD, so that we can still identify to which
845 * service an exit message belongs. */
846 r
= sd_event_source_set_priority(m
->notify_event_source
, SD_EVENT_PRIORITY_NORMAL
-8);
848 return log_error_errno(r
, "Failed to set priority of notify event source: %m");
850 (void) sd_event_source_set_description(m
->notify_event_source
, "manager-notify");
856 static int manager_setup_cgroups_agent(Manager
*m
) {
858 static const union sockaddr_union sa
= {
859 .un
.sun_family
= AF_UNIX
,
860 .un
.sun_path
= "/run/systemd/cgroups-agent",
864 /* This creates a listening socket we receive cgroups agent messages on. We do not use D-Bus for delivering
865 * these messages from the cgroups agent binary to PID 1, as the cgroups agent binary is very short-living, and
866 * each instance of it needs a new D-Bus connection. Since D-Bus connections are SOCK_STREAM/AF_UNIX, on
867 * overloaded systems the backlog of the D-Bus socket becomes relevant, as not more than the configured number
868 * of D-Bus connections may be queued until the kernel will start dropping further incoming connections,
869 * possibly resulting in lost cgroups agent messages. To avoid this, we'll use a private SOCK_DGRAM/AF_UNIX
870 * socket, where no backlog is relevant as communication may take place without an actual connect() cycle, and
871 * we thus won't lose messages.
873 * Note that PID 1 will forward the agent message to system bus, so that the user systemd instance may listen
874 * to it. The system instance hence listens on this special socket, but the user instances listen on the system
875 * bus for these messages. */
877 if (m
->test_run_flags
)
880 if (!MANAGER_IS_SYSTEM(m
))
883 r
= cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER
);
885 return log_error_errno(r
, "Failed to determine whether unified cgroups hierarchy is used: %m");
886 if (r
> 0) /* We don't need this anymore on the unified hierarchy */
889 if (m
->cgroups_agent_fd
< 0) {
890 _cleanup_close_
int fd
= -1;
892 /* First free all secondary fields */
893 m
->cgroups_agent_event_source
= sd_event_source_unref(m
->cgroups_agent_event_source
);
895 fd
= socket(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
|SOCK_NONBLOCK
, 0);
897 return log_error_errno(errno
, "Failed to allocate cgroups agent socket: %m");
899 fd_inc_rcvbuf(fd
, CGROUPS_AGENT_RCVBUF_SIZE
);
901 (void) unlink(sa
.un
.sun_path
);
903 /* Only allow root to connect to this socket */
905 r
= bind(fd
, &sa
.sa
, SOCKADDR_UN_LEN(sa
.un
));
907 return log_error_errno(errno
, "bind(%s) failed: %m", sa
.un
.sun_path
);
909 m
->cgroups_agent_fd
= fd
;
913 if (!m
->cgroups_agent_event_source
) {
914 r
= sd_event_add_io(m
->event
, &m
->cgroups_agent_event_source
, m
->cgroups_agent_fd
, EPOLLIN
, manager_dispatch_cgroups_agent_fd
, m
);
916 return log_error_errno(r
, "Failed to allocate cgroups agent event source: %m");
918 /* Process cgroups notifications early, but after having processed service notification messages or
919 * SIGCHLD signals, so that a cgroup running empty is always just the last safety net of notification,
920 * and we collected the metadata the notification and SIGCHLD stuff offers first. Also see handling of
921 * cgroup inotify for the unified cgroup stuff. */
922 r
= sd_event_source_set_priority(m
->cgroups_agent_event_source
, SD_EVENT_PRIORITY_NORMAL
-4);
924 return log_error_errno(r
, "Failed to set priority of cgroups agent event source: %m");
926 (void) sd_event_source_set_description(m
->cgroups_agent_event_source
, "manager-cgroups-agent");
932 static int manager_setup_user_lookup_fd(Manager
*m
) {
937 /* Set up the socket pair used for passing UID/GID resolution results from forked off processes to PID
938 * 1. Background: we can't do name lookups (NSS) from PID 1, since it might involve IPC and thus activation,
939 * and we might hence deadlock on ourselves. Hence we do all user/group lookups asynchronously from the forked
940 * off processes right before executing the binaries to start. In order to be able to clean up any IPC objects
941 * created by a unit (see RemoveIPC=) we need to know in PID 1 the used UID/GID of the executed processes,
942 * hence we establish this communication channel so that forked off processes can pass their UID/GID
943 * information back to PID 1. The forked off processes send their resolved UID/GID to PID 1 in a simple
944 * datagram, along with their unit name, so that we can share one communication socket pair among all units for
947 * You might wonder why we need a communication channel for this that is independent of the usual notification
948 * socket scheme (i.e. $NOTIFY_SOCKET). The primary difference is about trust: data sent via the $NOTIFY_SOCKET
949 * channel is only accepted if it originates from the right unit and if reception was enabled for it. The user
950 * lookup socket OTOH is only accessible by PID 1 and its children until they exec(), and always available.
952 * Note that this function is called under two circumstances: when we first initialize (in which case we
953 * allocate both the socket pair and the event source to listen on it), and when we deserialize after a reload
954 * (in which case the socket pair already exists but we still need to allocate the event source for it). */
956 if (m
->user_lookup_fds
[0] < 0) {
958 /* Free all secondary fields */
959 safe_close_pair(m
->user_lookup_fds
);
960 m
->user_lookup_event_source
= sd_event_source_unref(m
->user_lookup_event_source
);
962 if (socketpair(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
, 0, m
->user_lookup_fds
) < 0)
963 return log_error_errno(errno
, "Failed to allocate user lookup socket: %m");
965 (void) fd_inc_rcvbuf(m
->user_lookup_fds
[0], NOTIFY_RCVBUF_SIZE
);
968 if (!m
->user_lookup_event_source
) {
969 r
= sd_event_add_io(m
->event
, &m
->user_lookup_event_source
, m
->user_lookup_fds
[0], EPOLLIN
, manager_dispatch_user_lookup_fd
, m
);
971 return log_error_errno(errno
, "Failed to allocate user lookup event source: %m");
973 /* Process even earlier than the notify event source, so that we always know first about valid UID/GID
975 r
= sd_event_source_set_priority(m
->user_lookup_event_source
, SD_EVENT_PRIORITY_NORMAL
-11);
977 return log_error_errno(errno
, "Failed to set priority ot user lookup event source: %m");
979 (void) sd_event_source_set_description(m
->user_lookup_event_source
, "user-lookup");
985 static unsigned manager_dispatch_cleanup_queue(Manager
*m
) {
991 while ((u
= m
->cleanup_queue
)) {
992 assert(u
->in_cleanup_queue
);
1002 GC_OFFSET_IN_PATH
, /* This one is on the path we were traveling */
1003 GC_OFFSET_UNSURE
, /* No clue */
1004 GC_OFFSET_GOOD
, /* We still need this unit */
1005 GC_OFFSET_BAD
, /* We don't need this unit anymore */
1009 static void unit_gc_mark_good(Unit
*u
, unsigned gc_marker
) {
1014 u
->gc_marker
= gc_marker
+ GC_OFFSET_GOOD
;
1016 /* Recursively mark referenced units as GOOD as well */
1017 HASHMAP_FOREACH_KEY(v
, other
, u
->dependencies
[UNIT_REFERENCES
], i
)
1018 if (other
->gc_marker
== gc_marker
+ GC_OFFSET_UNSURE
)
1019 unit_gc_mark_good(other
, gc_marker
);
1022 static void unit_gc_sweep(Unit
*u
, unsigned gc_marker
) {
1030 if (IN_SET(u
->gc_marker
- gc_marker
,
1031 GC_OFFSET_GOOD
, GC_OFFSET_BAD
, GC_OFFSET_UNSURE
, GC_OFFSET_IN_PATH
))
1034 if (u
->in_cleanup_queue
)
1037 if (!unit_may_gc(u
))
1040 u
->gc_marker
= gc_marker
+ GC_OFFSET_IN_PATH
;
1044 HASHMAP_FOREACH_KEY(v
, other
, u
->dependencies
[UNIT_REFERENCED_BY
], i
) {
1045 unit_gc_sweep(other
, gc_marker
);
1047 if (other
->gc_marker
== gc_marker
+ GC_OFFSET_GOOD
)
1050 if (other
->gc_marker
!= gc_marker
+ GC_OFFSET_BAD
)
1054 if (u
->refs_by_target
) {
1057 LIST_FOREACH(refs_by_target
, ref
, u
->refs_by_target
) {
1058 unit_gc_sweep(ref
->source
, gc_marker
);
1060 if (ref
->source
->gc_marker
== gc_marker
+ GC_OFFSET_GOOD
)
1063 if (ref
->source
->gc_marker
!= gc_marker
+ GC_OFFSET_BAD
)
1071 /* We were unable to find anything out about this entry, so
1072 * let's investigate it later */
1073 u
->gc_marker
= gc_marker
+ GC_OFFSET_UNSURE
;
1074 unit_add_to_gc_queue(u
);
1078 /* We definitely know that this one is not useful anymore, so
1079 * let's mark it for deletion */
1080 u
->gc_marker
= gc_marker
+ GC_OFFSET_BAD
;
1081 unit_add_to_cleanup_queue(u
);
1085 unit_gc_mark_good(u
, gc_marker
);
1088 static unsigned manager_dispatch_gc_unit_queue(Manager
*m
) {
1089 unsigned n
= 0, gc_marker
;
1094 /* log_debug("Running GC..."); */
1096 m
->gc_marker
+= _GC_OFFSET_MAX
;
1097 if (m
->gc_marker
+ _GC_OFFSET_MAX
<= _GC_OFFSET_MAX
)
1100 gc_marker
= m
->gc_marker
;
1102 while ((u
= m
->gc_unit_queue
)) {
1103 assert(u
->in_gc_queue
);
1105 unit_gc_sweep(u
, gc_marker
);
1107 LIST_REMOVE(gc_queue
, m
->gc_unit_queue
, u
);
1108 u
->in_gc_queue
= false;
1112 if (IN_SET(u
->gc_marker
- gc_marker
,
1113 GC_OFFSET_BAD
, GC_OFFSET_UNSURE
)) {
1115 log_unit_debug(u
, "Collecting.");
1116 u
->gc_marker
= gc_marker
+ GC_OFFSET_BAD
;
1117 unit_add_to_cleanup_queue(u
);
1124 static unsigned manager_dispatch_gc_job_queue(Manager
*m
) {
1130 while ((j
= m
->gc_job_queue
)) {
1131 assert(j
->in_gc_queue
);
1133 LIST_REMOVE(gc_queue
, m
->gc_job_queue
, j
);
1134 j
->in_gc_queue
= false;
1141 log_unit_debug(j
->unit
, "Collecting job.");
1142 (void) job_finish_and_invalidate(j
, JOB_COLLECTED
, false, false);
1148 static void manager_clear_jobs_and_units(Manager
*m
) {
1153 while ((u
= hashmap_first(m
->units
)))
1156 manager_dispatch_cleanup_queue(m
);
1158 assert(!m
->load_queue
);
1159 assert(!m
->run_queue
);
1160 assert(!m
->dbus_unit_queue
);
1161 assert(!m
->dbus_job_queue
);
1162 assert(!m
->cleanup_queue
);
1163 assert(!m
->gc_unit_queue
);
1164 assert(!m
->gc_job_queue
);
1166 assert(hashmap_isempty(m
->jobs
));
1167 assert(hashmap_isempty(m
->units
));
1169 m
->n_on_console
= 0;
1170 m
->n_running_jobs
= 0;
1173 Manager
* manager_free(Manager
*m
) {
1176 ExecDirectoryType dt
;
1181 manager_clear_jobs_and_units(m
);
1183 for (c
= 0; c
< _UNIT_TYPE_MAX
; c
++)
1184 if (unit_vtable
[c
]->shutdown
)
1185 unit_vtable
[c
]->shutdown(m
);
1187 /* If we reexecute ourselves, we keep the root cgroup around */
1188 manager_shutdown_cgroup(m
, m
->exit_code
!= MANAGER_REEXECUTE
);
1190 lookup_paths_flush_generator(&m
->lookup_paths
);
1194 exec_runtime_vacuum(m
);
1195 hashmap_free(m
->exec_runtime_by_id
);
1197 dynamic_user_vacuum(m
, false);
1198 hashmap_free(m
->dynamic_users
);
1200 hashmap_free(m
->units
);
1201 hashmap_free(m
->units_by_invocation_id
);
1202 hashmap_free(m
->jobs
);
1203 hashmap_free(m
->watch_pids
);
1204 hashmap_free(m
->watch_bus
);
1206 set_free(m
->startup_units
);
1207 set_free(m
->failed_units
);
1209 sd_event_source_unref(m
->signal_event_source
);
1210 sd_event_source_unref(m
->sigchld_event_source
);
1211 sd_event_source_unref(m
->notify_event_source
);
1212 sd_event_source_unref(m
->cgroups_agent_event_source
);
1213 sd_event_source_unref(m
->time_change_event_source
);
1214 sd_event_source_unref(m
->jobs_in_progress_event_source
);
1215 sd_event_source_unref(m
->run_queue_event_source
);
1216 sd_event_source_unref(m
->user_lookup_event_source
);
1217 sd_event_source_unref(m
->sync_bus_names_event_source
);
1219 safe_close(m
->signal_fd
);
1220 safe_close(m
->notify_fd
);
1221 safe_close(m
->cgroups_agent_fd
);
1222 safe_close(m
->time_change_fd
);
1223 safe_close_pair(m
->user_lookup_fds
);
1225 manager_close_ask_password(m
);
1227 manager_close_idle_pipe(m
);
1229 udev_unref(m
->udev
);
1230 sd_event_unref(m
->event
);
1232 free(m
->notify_socket
);
1234 lookup_paths_free(&m
->lookup_paths
);
1235 strv_free(m
->environment
);
1237 hashmap_free(m
->cgroup_unit
);
1238 set_free_free(m
->unit_path_cache
);
1240 free(m
->switch_root
);
1241 free(m
->switch_root_init
);
1243 for (i
= 0; i
< _RLIMIT_MAX
; i
++)
1244 m
->rlimit
[i
] = mfree(m
->rlimit
[i
]);
1246 assert(hashmap_isempty(m
->units_requiring_mounts_for
));
1247 hashmap_free(m
->units_requiring_mounts_for
);
1249 hashmap_free(m
->uid_refs
);
1250 hashmap_free(m
->gid_refs
);
1252 for (dt
= 0; dt
< _EXEC_DIRECTORY_TYPE_MAX
; dt
++)
1253 m
->prefix
[dt
] = mfree(m
->prefix
[dt
]);
1258 static void manager_enumerate(Manager
*m
) {
1263 /* Let's ask every type to load all units from disk/kernel
1264 * that it might know */
1265 for (c
= 0; c
< _UNIT_TYPE_MAX
; c
++) {
1266 if (!unit_type_supported(c
)) {
1267 log_debug("Unit type .%s is not supported on this system.", unit_type_to_string(c
));
1271 if (unit_vtable
[c
]->enumerate
)
1272 unit_vtable
[c
]->enumerate(m
);
1275 manager_dispatch_load_queue(m
);
1278 static void manager_coldplug(Manager
*m
) {
1286 /* Then, let's set up their initial state. */
1287 HASHMAP_FOREACH_KEY(u
, k
, m
->units
, i
) {
1289 /* ignore aliases */
1293 r
= unit_coldplug(u
);
1295 log_warning_errno(r
, "We couldn't coldplug %s, proceeding anyway: %m", u
->id
);
1299 static void manager_build_unit_path_cache(Manager
*m
) {
1305 set_free_free(m
->unit_path_cache
);
1307 m
->unit_path_cache
= set_new(&path_hash_ops
);
1308 if (!m
->unit_path_cache
) {
1313 /* This simply builds a list of files we know exist, so that
1314 * we don't always have to go to disk */
1316 STRV_FOREACH(i
, m
->lookup_paths
.search_path
) {
1317 _cleanup_closedir_
DIR *d
= NULL
;
1322 if (errno
!= ENOENT
)
1323 log_warning_errno(errno
, "Failed to open directory %s, ignoring: %m", *i
);
1327 FOREACH_DIRENT(de
, d
, r
= -errno
; goto fail
) {
1330 p
= strjoin(streq(*i
, "/") ? "" : *i
, "/", de
->d_name
);
1336 r
= set_consume(m
->unit_path_cache
, p
);
1345 log_warning_errno(r
, "Failed to build unit path cache, proceeding without: %m");
1346 m
->unit_path_cache
= set_free_free(m
->unit_path_cache
);
1349 static void manager_distribute_fds(Manager
*m
, FDSet
*fds
) {
1355 HASHMAP_FOREACH(u
, m
->units
, i
) {
1357 if (fdset_size(fds
) <= 0)
1360 if (!UNIT_VTABLE(u
)->distribute_fds
)
1363 UNIT_VTABLE(u
)->distribute_fds(u
, fds
);
1367 static bool manager_dbus_is_running(Manager
*m
, bool deserialized
) {
1372 /* This checks whether the dbus instance we are supposed to expose our APIs on is up. We check both the socket
1373 * and the service unit. If the 'deserialized' parameter is true we'll check the deserialized state of the unit
1374 * rather than the current one. */
1376 if (m
->test_run_flags
!= 0)
1379 /* If we are in the user instance, and the env var is already set for us, then this means D-Bus is ran
1380 * somewhere outside of our own logic. Let's use it */
1381 if (MANAGER_IS_USER(m
) && getenv("DBUS_SESSION_BUS_ADDRESS"))
1384 u
= manager_get_unit(m
, SPECIAL_DBUS_SOCKET
);
1387 if ((deserialized
? SOCKET(u
)->deserialized_state
: SOCKET(u
)->state
) != SOCKET_RUNNING
)
1390 u
= manager_get_unit(m
, SPECIAL_DBUS_SERVICE
);
1393 if (!IN_SET((deserialized
? SERVICE(u
)->deserialized_state
: SERVICE(u
)->state
), SERVICE_RUNNING
, SERVICE_RELOAD
))
1399 int manager_startup(Manager
*m
, FILE *serialization
, FDSet
*fds
) {
1404 /* If we are running in test mode, we still want to run the generators,
1405 * but we should not touch the real generator directories. */
1406 r
= lookup_paths_init(&m
->lookup_paths
, m
->unit_file_scope
,
1407 m
->test_run_flags
? LOOKUP_PATHS_TEMPORARY_GENERATED
: 0,
1412 r
= manager_run_environment_generators(m
);
1416 dual_timestamp_get(m
->timestamps
+ MANAGER_TIMESTAMP_GENERATORS_START
);
1417 r
= manager_run_generators(m
);
1418 dual_timestamp_get(m
->timestamps
+ MANAGER_TIMESTAMP_GENERATORS_FINISH
);
1422 /* If this is the first boot, and we are in the host system, then preset everything */
1423 if (m
->first_boot
> 0 &&
1424 MANAGER_IS_SYSTEM(m
) &&
1425 !m
->test_run_flags
) {
1427 r
= unit_file_preset_all(UNIT_FILE_SYSTEM
, 0, NULL
, UNIT_FILE_PRESET_ENABLE_ONLY
, NULL
, 0);
1429 log_full_errno(r
== -EEXIST
? LOG_NOTICE
: LOG_WARNING
, r
,
1430 "Failed to populate /etc with preset unit settings, ignoring: %m");
1432 log_info("Populated /etc with preset unit settings.");
1435 lookup_paths_reduce(&m
->lookup_paths
);
1436 manager_build_unit_path_cache(m
);
1438 /* If we will deserialize make sure that during enumeration
1439 * this is already known, so we increase the counter here
1444 /* First, enumerate what we can from all config files */
1445 dual_timestamp_get(m
->timestamps
+ MANAGER_TIMESTAMP_UNITS_LOAD_START
);
1446 manager_enumerate(m
);
1447 dual_timestamp_get(m
->timestamps
+ MANAGER_TIMESTAMP_UNITS_LOAD_FINISH
);
1449 /* Second, deserialize if there is something to deserialize */
1450 if (serialization
) {
1451 r
= manager_deserialize(m
, serialization
, fds
);
1453 return log_error_errno(r
, "Deserialization failed: %m");
1456 /* Any fds left? Find some unit which wants them. This is
1457 * useful to allow container managers to pass some file
1458 * descriptors to us pre-initialized. This enables
1459 * socket-based activation of entire containers. */
1460 manager_distribute_fds(m
, fds
);
1462 /* We might have deserialized the notify fd, but if we didn't
1463 * then let's create the bus now */
1464 r
= manager_setup_notify(m
);
1466 /* No sense to continue without notifications, our children would fail anyway. */
1469 r
= manager_setup_cgroups_agent(m
);
1471 /* Likewise, no sense to continue without empty cgroup notifications. */
1474 r
= manager_setup_user_lookup_fd(m
);
1476 /* This shouldn't fail, except if things are really broken. */
1479 /* Let's set up our private bus connection now, unconditionally */
1480 (void) bus_init_private(m
);
1482 /* If we are in --user mode also connect to the system bus now */
1483 if (MANAGER_IS_USER(m
))
1484 (void) bus_init_system(m
);
1486 /* Let's connect to the bus now, but only if the unit is supposed to be up */
1487 if (manager_dbus_is_running(m
, !!serialization
)) {
1488 (void) bus_init_api(m
);
1490 if (MANAGER_IS_SYSTEM(m
))
1491 (void) bus_init_system(m
);
1494 /* Now that we are connected to all possible busses, let's deserialize who is tracking us. */
1495 (void) bus_track_coldplug(m
, &m
->subscribed
, false, m
->deserialized_subscribed
);
1496 m
->deserialized_subscribed
= strv_free(m
->deserialized_subscribed
);
1498 /* Third, fire things up! */
1499 manager_coldplug(m
);
1501 /* Release any dynamic users no longer referenced */
1502 dynamic_user_vacuum(m
, true);
1504 exec_runtime_vacuum(m
);
1506 /* Release any references to UIDs/GIDs no longer referenced, and destroy any IPC owned by them */
1507 manager_vacuum_uid_refs(m
);
1508 manager_vacuum_gid_refs(m
);
1510 if (serialization
) {
1511 assert(m
->n_reloading
> 0);
1514 /* Let's wait for the UnitNew/JobNew messages being
1515 * sent, before we notify that the reload is
1517 m
->send_reloading_done
= true;
1523 int manager_add_job(Manager
*m
, JobType type
, Unit
*unit
, JobMode mode
, sd_bus_error
*e
, Job
**_ret
) {
1528 assert(type
< _JOB_TYPE_MAX
);
1530 assert(mode
< _JOB_MODE_MAX
);
1532 if (mode
== JOB_ISOLATE
&& type
!= JOB_START
)
1533 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
, "Isolate is only valid for start.");
1535 if (mode
== JOB_ISOLATE
&& !unit
->allow_isolate
)
1536 return sd_bus_error_setf(e
, BUS_ERROR_NO_ISOLATION
, "Operation refused, unit may not be isolated.");
1538 log_unit_debug(unit
, "Trying to enqueue job %s/%s/%s", unit
->id
, job_type_to_string(type
), job_mode_to_string(mode
));
1540 type
= job_type_collapse(type
, unit
);
1542 tr
= transaction_new(mode
== JOB_REPLACE_IRREVERSIBLY
);
1546 r
= transaction_add_job_and_dependencies(tr
, type
, unit
, NULL
, true, false,
1547 IN_SET(mode
, JOB_IGNORE_DEPENDENCIES
, JOB_IGNORE_REQUIREMENTS
),
1548 mode
== JOB_IGNORE_DEPENDENCIES
, e
);
1552 if (mode
== JOB_ISOLATE
) {
1553 r
= transaction_add_isolate_jobs(tr
, m
);
1558 r
= transaction_activate(tr
, m
, mode
, e
);
1562 log_unit_debug(unit
,
1563 "Enqueued job %s/%s as %u", unit
->id
,
1564 job_type_to_string(type
), (unsigned) tr
->anchor_job
->id
);
1567 *_ret
= tr
->anchor_job
;
1569 transaction_free(tr
);
1573 transaction_abort(tr
);
1574 transaction_free(tr
);
1578 int manager_add_job_by_name(Manager
*m
, JobType type
, const char *name
, JobMode mode
, sd_bus_error
*e
, Job
**ret
) {
1579 Unit
*unit
= NULL
; /* just to appease gcc, initialization is not really necessary */
1583 assert(type
< _JOB_TYPE_MAX
);
1585 assert(mode
< _JOB_MODE_MAX
);
1587 r
= manager_load_unit(m
, name
, NULL
, NULL
, &unit
);
1592 return manager_add_job(m
, type
, unit
, mode
, e
, ret
);
1595 int manager_add_job_by_name_and_warn(Manager
*m
, JobType type
, const char *name
, JobMode mode
, Job
**ret
) {
1596 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
1600 assert(type
< _JOB_TYPE_MAX
);
1602 assert(mode
< _JOB_MODE_MAX
);
1604 r
= manager_add_job_by_name(m
, type
, name
, mode
, &error
, ret
);
1606 return log_warning_errno(r
, "Failed to enqueue %s job for %s: %s", job_mode_to_string(mode
), name
, bus_error_message(&error
, r
));
1611 int manager_propagate_reload(Manager
*m
, Unit
*unit
, JobMode mode
, sd_bus_error
*e
) {
1617 assert(mode
< _JOB_MODE_MAX
);
1618 assert(mode
!= JOB_ISOLATE
); /* Isolate is only valid for start */
1620 tr
= transaction_new(mode
== JOB_REPLACE_IRREVERSIBLY
);
1624 /* We need an anchor job */
1625 r
= transaction_add_job_and_dependencies(tr
, JOB_NOP
, unit
, NULL
, false, false, true, true, e
);
1629 /* Failure in adding individual dependencies is ignored, so this always succeeds. */
1630 transaction_add_propagate_reload_jobs(tr
, unit
, tr
->anchor_job
, mode
== JOB_IGNORE_DEPENDENCIES
, e
);
1632 r
= transaction_activate(tr
, m
, mode
, e
);
1636 transaction_free(tr
);
1640 transaction_abort(tr
);
1641 transaction_free(tr
);
1645 Job
*manager_get_job(Manager
*m
, uint32_t id
) {
1648 return hashmap_get(m
->jobs
, UINT32_TO_PTR(id
));
1651 Unit
*manager_get_unit(Manager
*m
, const char *name
) {
1655 return hashmap_get(m
->units
, name
);
1658 static int manager_dispatch_target_deps_queue(Manager
*m
) {
1663 static const UnitDependency deps
[] = {
1672 while ((u
= m
->target_deps_queue
)) {
1673 assert(u
->in_target_deps_queue
);
1675 LIST_REMOVE(target_deps_queue
, u
->manager
->target_deps_queue
, u
);
1676 u
->in_target_deps_queue
= false;
1678 for (k
= 0; k
< ELEMENTSOF(deps
); k
++) {
1683 HASHMAP_FOREACH_KEY(v
, target
, u
->dependencies
[deps
[k
]], i
) {
1684 r
= unit_add_default_target_dependency(u
, target
);
1694 unsigned manager_dispatch_load_queue(Manager
*m
) {
1700 /* Make sure we are not run recursively */
1701 if (m
->dispatching_load_queue
)
1704 m
->dispatching_load_queue
= true;
1706 /* Dispatches the load queue. Takes a unit from the queue and
1707 * tries to load its data until the queue is empty */
1709 while ((u
= m
->load_queue
)) {
1710 assert(u
->in_load_queue
);
1716 m
->dispatching_load_queue
= false;
1718 /* Dispatch the units waiting for their target dependencies to be added now, as all targets that we know about
1719 * should be loaded and have aliases resolved */
1720 (void) manager_dispatch_target_deps_queue(m
);
1725 int manager_load_unit_prepare(
1732 _cleanup_(unit_freep
) Unit
*cleanup_ret
= NULL
;
1738 assert(name
|| path
);
1741 /* This will prepare the unit for loading, but not actually
1742 * load anything from disk. */
1744 if (path
&& !is_path(path
))
1745 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
, "Path %s is not absolute.", path
);
1748 name
= basename(path
);
1750 t
= unit_name_to_type(name
);
1752 if (t
== _UNIT_TYPE_INVALID
|| !unit_name_is_valid(name
, UNIT_NAME_PLAIN
|UNIT_NAME_INSTANCE
)) {
1753 if (unit_name_is_valid(name
, UNIT_NAME_TEMPLATE
))
1754 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
, "Unit name %s is missing the instance name.", name
);
1756 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
, "Unit name %s is not valid.", name
);
1759 ret
= manager_get_unit(m
, name
);
1765 ret
= cleanup_ret
= unit_new(m
, unit_vtable
[t
]->object_size
);
1770 ret
->fragment_path
= strdup(path
);
1771 if (!ret
->fragment_path
)
1775 r
= unit_add_name(ret
, name
);
1779 unit_add_to_load_queue(ret
);
1780 unit_add_to_dbus_queue(ret
);
1781 unit_add_to_gc_queue(ret
);
1789 int manager_load_unit(
1801 /* This will load the service information files, but not actually
1802 * start any services or anything. */
1804 r
= manager_load_unit_prepare(m
, name
, path
, e
, _ret
);
1808 manager_dispatch_load_queue(m
);
1810 *_ret
= unit_follow_merge(*_ret
);
1814 int manager_load_startable_unit_or_warn(
1820 /* Load a unit, make sure it loaded fully and is not masked. */
1822 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
1826 r
= manager_load_unit(m
, name
, path
, &error
, &unit
);
1828 return log_error_errno(r
, "Failed to load %s %s: %s",
1829 name
? "unit" : "file", name
?: path
,
1830 bus_error_message(&error
, r
));
1831 else if (IN_SET(unit
->load_state
, UNIT_ERROR
, UNIT_NOT_FOUND
))
1832 return log_error_errno(unit
->load_error
, "Failed to load %s %s: %m",
1833 name
? "unit" : "file", name
?: path
);
1834 else if (unit
->load_state
== UNIT_MASKED
) {
1835 log_error("%s %s is masked.",
1836 name
? "Unit" : "File", name
?: path
);
1844 void manager_dump_jobs(Manager
*s
, FILE *f
, const char *prefix
) {
1851 HASHMAP_FOREACH(j
, s
->jobs
, i
)
1852 job_dump(j
, f
, prefix
);
1855 void manager_dump_units(Manager
*s
, FILE *f
, const char *prefix
) {
1863 HASHMAP_FOREACH_KEY(u
, t
, s
->units
, i
)
1865 unit_dump(u
, f
, prefix
);
1868 void manager_dump(Manager
*m
, FILE *f
, const char *prefix
) {
1874 for (q
= 0; q
< _MANAGER_TIMESTAMP_MAX
; q
++) {
1875 char buf
[FORMAT_TIMESTAMP_MAX
];
1877 if (dual_timestamp_is_set(m
->timestamps
+ q
))
1878 fprintf(f
, "%sTimestamp %s: %s\n",
1880 manager_timestamp_to_string(q
),
1881 format_timestamp(buf
, sizeof(buf
), m
->timestamps
[q
].realtime
));
1884 manager_dump_units(m
, f
, prefix
);
1885 manager_dump_jobs(m
, f
, prefix
);
1888 int manager_get_dump_string(Manager
*m
, char **ret
) {
1889 _cleanup_free_
char *dump
= NULL
;
1890 _cleanup_fclose_
FILE *f
= NULL
;
1897 f
= open_memstream(&dump
, &size
);
1901 (void) __fsetlocking(f
, FSETLOCKING_BYCALLER
);
1903 manager_dump(m
, f
, NULL
);
1905 r
= fflush_and_check(f
);
1911 *ret
= TAKE_PTR(dump
);
1916 void manager_clear_jobs(Manager
*m
) {
1921 while ((j
= hashmap_first(m
->jobs
)))
1922 /* No need to recurse. We're cancelling all jobs. */
1923 job_finish_and_invalidate(j
, JOB_CANCELED
, false, false);
1926 static int manager_dispatch_run_queue(sd_event_source
*source
, void *userdata
) {
1927 Manager
*m
= userdata
;
1933 while ((j
= m
->run_queue
)) {
1934 assert(j
->installed
);
1935 assert(j
->in_run_queue
);
1937 job_run_and_invalidate(j
);
1940 if (m
->n_running_jobs
> 0)
1941 manager_watch_jobs_in_progress(m
);
1943 if (m
->n_on_console
> 0)
1944 manager_watch_idle_pipe(m
);
1949 static unsigned manager_dispatch_dbus_queue(Manager
*m
) {
1950 unsigned n
= 0, budget
;
1956 if (m
->dispatching_dbus_queue
)
1959 /* Anything to do at all? */
1960 if (!m
->dbus_unit_queue
&& !m
->dbus_job_queue
&& !m
->send_reloading_done
&& !m
->queued_message
)
1963 /* Do we have overly many messages queued at the moment? If so, let's not enqueue more on top, let's sit this
1964 * cycle out, and process things in a later cycle when the queues got a bit emptier. */
1965 if (manager_bus_n_queued_write(m
) > MANAGER_BUS_BUSY_THRESHOLD
)
1968 /* Only process a certain number of units/jobs per event loop iteration. Even if the bus queue wasn't overly
1969 * full before this call we shouldn't increase it in size too wildly in one step, and we shouldn't monopolize
1970 * CPU time with generating these messages. Note the difference in counting of this "budget" and the
1971 * "threshold" above: the "budget" is decreased only once per generated message, regardless how many
1972 * busses/direct connections it is enqueued on, while the "threshold" is applied to each queued instance of bus
1973 * message, i.e. if the same message is enqueued to five busses/direct connections it will be counted five
1974 * times. This difference in counting ("references" vs. "instances") is primarily a result of the fact that
1975 * it's easier to implement it this way, however it also reflects the thinking that the "threshold" should put
1976 * a limit on used queue memory, i.e. space, while the "budget" should put a limit on time. Also note that
1977 * the "threshold" is currently chosen much higher than the "budget". */
1978 budget
= MANAGER_BUS_MESSAGE_BUDGET
;
1980 m
->dispatching_dbus_queue
= true;
1982 while (budget
> 0 && (u
= m
->dbus_unit_queue
)) {
1984 assert(u
->in_dbus_queue
);
1986 bus_unit_send_change_signal(u
);
1990 while (budget
> 0 && (j
= m
->dbus_job_queue
)) {
1991 assert(j
->in_dbus_queue
);
1993 bus_job_send_change_signal(j
);
1997 m
->dispatching_dbus_queue
= false;
1999 if (budget
> 0 && m
->send_reloading_done
) {
2000 m
->send_reloading_done
= false;
2001 bus_manager_send_reloading(m
, false);
2005 if (budget
> 0 && m
->queued_message
) {
2006 bus_send_queued_message(m
);
2013 static int manager_dispatch_cgroups_agent_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
2014 Manager
*m
= userdata
;
2015 char buf
[PATH_MAX
+1];
2018 n
= recv(fd
, buf
, sizeof(buf
), 0);
2020 return log_error_errno(errno
, "Failed to read cgroups agent message: %m");
2022 log_error("Got zero-length cgroups agent message, ignoring.");
2025 if ((size_t) n
>= sizeof(buf
)) {
2026 log_error("Got overly long cgroups agent message, ignoring.");
2030 if (memchr(buf
, 0, n
)) {
2031 log_error("Got cgroups agent message with embedded NUL byte, ignoring.");
2036 manager_notify_cgroup_empty(m
, buf
);
2037 (void) bus_forward_agent_released(m
, buf
);
2042 static void manager_invoke_notify_message(
2045 const struct ucred
*ucred
,
2054 if (u
->notifygen
== m
->notifygen
) /* Already invoked on this same unit in this same iteration? */
2056 u
->notifygen
= m
->notifygen
;
2058 if (UNIT_VTABLE(u
)->notify_message
) {
2059 _cleanup_strv_free_
char **tags
= NULL
;
2061 tags
= strv_split(buf
, NEWLINE
);
2067 UNIT_VTABLE(u
)->notify_message(u
, ucred
, tags
, fds
);
2069 } else if (DEBUG_LOGGING
) {
2070 _cleanup_free_
char *x
= NULL
, *y
= NULL
;
2072 x
= ellipsize(buf
, 20, 90);
2076 log_unit_debug(u
, "Got notification message \"%s\", ignoring.", strnull(y
));
2080 static int manager_dispatch_notify_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
2082 _cleanup_fdset_free_ FDSet
*fds
= NULL
;
2083 Manager
*m
= userdata
;
2084 char buf
[NOTIFY_BUFFER_MAX
+1];
2085 struct iovec iovec
= {
2087 .iov_len
= sizeof(buf
)-1,
2090 struct cmsghdr cmsghdr
;
2091 uint8_t buf
[CMSG_SPACE(sizeof(struct ucred
)) +
2092 CMSG_SPACE(sizeof(int) * NOTIFY_FD_MAX
)];
2094 struct msghdr msghdr
= {
2097 .msg_control
= &control
,
2098 .msg_controllen
= sizeof(control
),
2101 struct cmsghdr
*cmsg
;
2102 struct ucred
*ucred
= NULL
;
2103 _cleanup_free_ Unit
**array_copy
= NULL
;
2104 Unit
*u1
, *u2
, **array
;
2105 int r
, *fd_array
= NULL
;
2111 assert(m
->notify_fd
== fd
);
2113 if (revents
!= EPOLLIN
) {
2114 log_warning("Got unexpected poll event for notify fd.");
2118 n
= recvmsg(m
->notify_fd
, &msghdr
, MSG_DONTWAIT
|MSG_CMSG_CLOEXEC
|MSG_TRUNC
);
2120 if (IN_SET(errno
, EAGAIN
, EINTR
))
2121 return 0; /* Spurious wakeup, try again */
2123 /* If this is any other, real error, then let's stop processing this socket. This of course means we
2124 * won't take notification messages anymore, but that's still better than busy looping around this:
2125 * being woken up over and over again but being unable to actually read the message off the socket. */
2126 return log_error_errno(errno
, "Failed to receive notification message: %m");
2129 CMSG_FOREACH(cmsg
, &msghdr
) {
2130 if (cmsg
->cmsg_level
== SOL_SOCKET
&& cmsg
->cmsg_type
== SCM_RIGHTS
) {
2132 fd_array
= (int*) CMSG_DATA(cmsg
);
2133 n_fds
= (cmsg
->cmsg_len
- CMSG_LEN(0)) / sizeof(int);
2135 } else if (cmsg
->cmsg_level
== SOL_SOCKET
&&
2136 cmsg
->cmsg_type
== SCM_CREDENTIALS
&&
2137 cmsg
->cmsg_len
== CMSG_LEN(sizeof(struct ucred
))) {
2139 ucred
= (struct ucred
*) CMSG_DATA(cmsg
);
2146 r
= fdset_new_array(&fds
, fd_array
, n_fds
);
2148 close_many(fd_array
, n_fds
);
2154 if (!ucred
|| !pid_is_valid(ucred
->pid
)) {
2155 log_warning("Received notify message without valid credentials. Ignoring.");
2159 if ((size_t) n
>= sizeof(buf
) || (msghdr
.msg_flags
& MSG_TRUNC
)) {
2160 log_warning("Received notify message exceeded maximum size. Ignoring.");
2164 /* As extra safety check, let's make sure the string we get doesn't contain embedded NUL bytes. We permit one
2165 * trailing NUL byte in the message, but don't expect it. */
2166 if (n
> 1 && memchr(buf
, 0, n
-1)) {
2167 log_warning("Received notify message with embedded NUL bytes. Ignoring.");
2171 /* Make sure it's NUL-terminated. */
2174 /* Increase the generation counter used for filtering out duplicate unit invocations. */
2177 /* Notify every unit that might be interested, which might be multiple. */
2178 u1
= manager_get_unit_by_pid_cgroup(m
, ucred
->pid
);
2179 u2
= hashmap_get(m
->watch_pids
, PID_TO_PTR(ucred
->pid
));
2180 array
= hashmap_get(m
->watch_pids
, PID_TO_PTR(-ucred
->pid
));
2187 array_copy
= newdup(Unit
*, array
, k
+1);
2191 /* And now invoke the per-unit callbacks. Note that manager_invoke_notify_message() will handle duplicate units
2192 * make sure we only invoke each unit's handler once. */
2194 manager_invoke_notify_message(m
, u1
, ucred
, buf
, fds
);
2198 manager_invoke_notify_message(m
, u2
, ucred
, buf
, fds
);
2202 for (size_t i
= 0; array_copy
[i
]; i
++) {
2203 manager_invoke_notify_message(m
, array_copy
[i
], ucred
, buf
, fds
);
2208 log_warning("Cannot find unit for notify message of PID "PID_FMT
", ignoring.", ucred
->pid
);
2210 if (fdset_size(fds
) > 0)
2211 log_warning("Got extra auxiliary fds with notification message, closing them.");
2216 static void manager_invoke_sigchld_event(
2219 const siginfo_t
*si
) {
2225 /* Already invoked the handler of this unit in this iteration? Then don't process this again */
2226 if (u
->sigchldgen
== m
->sigchldgen
)
2228 u
->sigchldgen
= m
->sigchldgen
;
2230 log_unit_debug(u
, "Child "PID_FMT
" belongs to %s.", si
->si_pid
, u
->id
);
2231 unit_unwatch_pid(u
, si
->si_pid
);
2233 if (UNIT_VTABLE(u
)->sigchld_event
)
2234 UNIT_VTABLE(u
)->sigchld_event(u
, si
->si_pid
, si
->si_code
, si
->si_status
);
2237 static int manager_dispatch_sigchld(sd_event_source
*source
, void *userdata
) {
2238 Manager
*m
= userdata
;
2245 /* First we call waitd() for a PID and do not reap the zombie. That way we can still access /proc/$PID for it
2246 * while it is a zombie. */
2248 if (waitid(P_ALL
, 0, &si
, WEXITED
|WNOHANG
|WNOWAIT
) < 0) {
2250 if (errno
!= ECHILD
)
2251 log_error_errno(errno
, "Failed to peek for child with waitid(), ignoring: %m");
2259 if (IN_SET(si
.si_code
, CLD_EXITED
, CLD_KILLED
, CLD_DUMPED
)) {
2260 _cleanup_free_ Unit
**array_copy
= NULL
;
2261 _cleanup_free_
char *name
= NULL
;
2262 Unit
*u1
, *u2
, **array
;
2264 (void) get_process_comm(si
.si_pid
, &name
);
2266 log_debug("Child "PID_FMT
" (%s) died (code=%s, status=%i/%s)",
2267 si
.si_pid
, strna(name
),
2268 sigchld_code_to_string(si
.si_code
),
2270 strna(si
.si_code
== CLD_EXITED
2271 ? exit_status_to_string(si
.si_status
, EXIT_STATUS_FULL
)
2272 : signal_to_string(si
.si_status
)));
2274 /* Increase the generation counter used for filtering out duplicate unit invocations */
2277 /* And now figure out the unit this belongs to, it might be multiple... */
2278 u1
= manager_get_unit_by_pid_cgroup(m
, si
.si_pid
);
2279 u2
= hashmap_get(m
->watch_pids
, PID_TO_PTR(si
.si_pid
));
2280 array
= hashmap_get(m
->watch_pids
, PID_TO_PTR(-si
.si_pid
));
2284 /* Cound how many entries the array has */
2288 /* Make a copy of the array so that we don't trip up on the array changing beneath us */
2289 array_copy
= newdup(Unit
*, array
, n
+1);
2294 /* Finally, execute them all. Note that u1, u2 and the array might contain duplicates, but
2295 * that's fine, manager_invoke_sigchld_event() will ensure we only invoke the handlers once for
2296 * each iteration. */
2298 manager_invoke_sigchld_event(m
, u1
, &si
);
2300 manager_invoke_sigchld_event(m
, u2
, &si
);
2302 for (size_t i
= 0; array_copy
[i
]; i
++)
2303 manager_invoke_sigchld_event(m
, array_copy
[i
], &si
);
2306 /* And now, we actually reap the zombie. */
2307 if (waitid(P_PID
, si
.si_pid
, &si
, WEXITED
) < 0) {
2308 log_error_errno(errno
, "Failed to dequeue child, ignoring: %m");
2315 /* All children processed for now, turn off event source */
2317 r
= sd_event_source_set_enabled(m
->sigchld_event_source
, SD_EVENT_OFF
);
2319 return log_error_errno(r
, "Failed to disable SIGCHLD event source: %m");
2324 static void manager_start_target(Manager
*m
, const char *name
, JobMode mode
) {
2325 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
2328 log_debug("Activating special unit %s", name
);
2330 r
= manager_add_job_by_name(m
, JOB_START
, name
, mode
, &error
, NULL
);
2332 log_error("Failed to enqueue %s job: %s", name
, bus_error_message(&error
, r
));
2335 static void manager_handle_ctrl_alt_del(Manager
*m
) {
2336 /* If the user presses C-A-D more than
2337 * 7 times within 2s, we reboot/shutdown immediately,
2338 * unless it was disabled in system.conf */
2340 if (ratelimit_test(&m
->ctrl_alt_del_ratelimit
) || m
->cad_burst_action
== EMERGENCY_ACTION_NONE
)
2341 manager_start_target(m
, SPECIAL_CTRL_ALT_DEL_TARGET
, JOB_REPLACE_IRREVERSIBLY
);
2343 emergency_action(m
, m
->cad_burst_action
, NULL
,
2344 "Ctrl-Alt-Del was pressed more than 7 times within 2s");
2347 static int manager_dispatch_signal_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
2348 Manager
*m
= userdata
;
2350 struct signalfd_siginfo sfsi
;
2354 assert(m
->signal_fd
== fd
);
2356 if (revents
!= EPOLLIN
) {
2357 log_warning("Got unexpected events from signal file descriptor.");
2361 n
= read(m
->signal_fd
, &sfsi
, sizeof(sfsi
));
2362 if (n
!= sizeof(sfsi
)) {
2364 log_warning("Truncated read from signal fd (%zu bytes), ignoring!", n
);
2368 if (IN_SET(errno
, EINTR
, EAGAIN
))
2371 /* We return an error here, which will kill this handler,
2372 * to avoid a busy loop on read error. */
2373 return log_error_errno(errno
, "Reading from signal fd failed: %m");
2376 log_received_signal(sfsi
.ssi_signo
== SIGCHLD
||
2377 (sfsi
.ssi_signo
== SIGTERM
&& MANAGER_IS_USER(m
))
2378 ? LOG_DEBUG
: LOG_INFO
,
2381 switch (sfsi
.ssi_signo
) {
2384 r
= sd_event_source_set_enabled(m
->sigchld_event_source
, SD_EVENT_ON
);
2386 log_warning_errno(r
, "Failed to enable SIGCHLD event source, ignoring: %m");
2391 if (MANAGER_IS_SYSTEM(m
)) {
2392 /* This is for compatibility with the original sysvinit */
2393 r
= verify_run_space_and_log("Refusing to reexecute");
2395 m
->exit_code
= MANAGER_REEXECUTE
;
2401 if (MANAGER_IS_SYSTEM(m
))
2402 manager_handle_ctrl_alt_del(m
);
2404 manager_start_target(m
, SPECIAL_EXIT_TARGET
,
2405 JOB_REPLACE_IRREVERSIBLY
);
2409 /* This is a nop on non-init */
2410 if (MANAGER_IS_SYSTEM(m
))
2411 manager_start_target(m
, SPECIAL_KBREQUEST_TARGET
, JOB_REPLACE
);
2416 /* This is a nop on non-init */
2417 if (MANAGER_IS_SYSTEM(m
))
2418 manager_start_target(m
, SPECIAL_SIGPWR_TARGET
, JOB_REPLACE
);
2423 if (manager_dbus_is_running(m
, false)) {
2424 log_info("Trying to reconnect to bus...");
2426 (void) bus_init_api(m
);
2428 if (MANAGER_IS_SYSTEM(m
))
2429 (void) bus_init_system(m
);
2431 log_info("Starting D-Bus service...");
2432 manager_start_target(m
, SPECIAL_DBUS_SERVICE
, JOB_REPLACE
);
2438 _cleanup_free_
char *dump
= NULL
;
2440 r
= manager_get_dump_string(m
, &dump
);
2442 log_warning_errno(errno
, "Failed to acquire manager dump: %m");
2446 log_dump(LOG_INFO
, dump
);
2451 r
= verify_run_space_and_log("Refusing to reload");
2453 m
->exit_code
= MANAGER_RELOAD
;
2458 /* Starting SIGRTMIN+0 */
2459 static const struct {
2462 } target_table
[] = {
2463 [0] = { SPECIAL_DEFAULT_TARGET
, JOB_ISOLATE
},
2464 [1] = { SPECIAL_RESCUE_TARGET
, JOB_ISOLATE
},
2465 [2] = { SPECIAL_EMERGENCY_TARGET
, JOB_ISOLATE
},
2466 [3] = { SPECIAL_HALT_TARGET
, JOB_REPLACE_IRREVERSIBLY
},
2467 [4] = { SPECIAL_POWEROFF_TARGET
, JOB_REPLACE_IRREVERSIBLY
},
2468 [5] = { SPECIAL_REBOOT_TARGET
, JOB_REPLACE_IRREVERSIBLY
},
2469 [6] = { SPECIAL_KEXEC_TARGET
, JOB_REPLACE_IRREVERSIBLY
},
2472 /* Starting SIGRTMIN+13, so that target halt and system halt are 10 apart */
2473 static const ManagerExitCode code_table
[] = {
2475 [1] = MANAGER_POWEROFF
,
2476 [2] = MANAGER_REBOOT
,
2477 [3] = MANAGER_KEXEC
,
2480 if ((int) sfsi
.ssi_signo
>= SIGRTMIN
+0 &&
2481 (int) sfsi
.ssi_signo
< SIGRTMIN
+(int) ELEMENTSOF(target_table
)) {
2482 int idx
= (int) sfsi
.ssi_signo
- SIGRTMIN
;
2483 manager_start_target(m
, target_table
[idx
].target
,
2484 target_table
[idx
].mode
);
2488 if ((int) sfsi
.ssi_signo
>= SIGRTMIN
+13 &&
2489 (int) sfsi
.ssi_signo
< SIGRTMIN
+13+(int) ELEMENTSOF(code_table
)) {
2490 m
->exit_code
= code_table
[sfsi
.ssi_signo
- SIGRTMIN
- 13];
2494 switch (sfsi
.ssi_signo
- SIGRTMIN
) {
2497 manager_set_show_status(m
, SHOW_STATUS_YES
);
2501 manager_set_show_status(m
, SHOW_STATUS_NO
);
2505 log_set_max_level(LOG_DEBUG
);
2506 log_info("Setting log level to debug.");
2510 log_set_max_level(LOG_INFO
);
2511 log_info("Setting log level to info.");
2515 if (MANAGER_IS_USER(m
)) {
2516 m
->exit_code
= MANAGER_EXIT
;
2520 /* This is a nop on init */
2524 case 29: /* compatibility: used to be mapped to LOG_TARGET_SYSLOG_OR_KMSG */
2525 log_set_target(LOG_TARGET_JOURNAL_OR_KMSG
);
2526 log_notice("Setting log target to journal-or-kmsg.");
2530 log_set_target(LOG_TARGET_CONSOLE
);
2531 log_notice("Setting log target to console.");
2535 log_set_target(LOG_TARGET_KMSG
);
2536 log_notice("Setting log target to kmsg.");
2540 log_warning("Got unhandled signal <%s>.", signal_to_string(sfsi
.ssi_signo
));
2547 static int manager_dispatch_time_change_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
2548 Manager
*m
= userdata
;
2553 assert(m
->time_change_fd
== fd
);
2555 log_struct(LOG_DEBUG
,
2556 "MESSAGE_ID=" SD_MESSAGE_TIME_CHANGE_STR
,
2557 LOG_MESSAGE("Time has been changed"),
2560 /* Restart the watch */
2561 m
->time_change_event_source
= sd_event_source_unref(m
->time_change_event_source
);
2562 m
->time_change_fd
= safe_close(m
->time_change_fd
);
2564 manager_setup_time_change(m
);
2566 HASHMAP_FOREACH(u
, m
->units
, i
)
2567 if (UNIT_VTABLE(u
)->time_change
)
2568 UNIT_VTABLE(u
)->time_change(u
);
2573 static int manager_dispatch_idle_pipe_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
2574 Manager
*m
= userdata
;
2577 assert(m
->idle_pipe
[2] == fd
);
2579 /* There's at least one Type=idle child that just gave up on us waiting for the boot process to complete. Let's
2580 * now turn off any further console output if there's at least one service that needs console access, so that
2581 * from now on our own output should not spill into that service's output anymore. After all, we support
2582 * Type=idle only to beautify console output and it generally is set on services that want to own the console
2583 * exclusively without our interference. */
2584 m
->no_console_output
= m
->n_on_console
> 0;
2586 /* Acknowledge the child's request, and let all all other children know too that they shouldn't wait any longer
2587 * by closing the pipes towards them, which is what they are waiting for. */
2588 manager_close_idle_pipe(m
);
2593 static int manager_dispatch_jobs_in_progress(sd_event_source
*source
, usec_t usec
, void *userdata
) {
2594 Manager
*m
= userdata
;
2601 manager_print_jobs_in_progress(m
);
2603 next
= now(CLOCK_MONOTONIC
) + JOBS_IN_PROGRESS_PERIOD_USEC
;
2604 r
= sd_event_source_set_time(source
, next
);
2608 return sd_event_source_set_enabled(source
, SD_EVENT_ONESHOT
);
2611 int manager_loop(Manager
*m
) {
2614 RATELIMIT_DEFINE(rl
, 1*USEC_PER_SEC
, 50000);
2617 m
->exit_code
= MANAGER_OK
;
2619 /* Release the path cache */
2620 m
->unit_path_cache
= set_free_free(m
->unit_path_cache
);
2622 manager_check_finished(m
);
2624 /* There might still be some zombies hanging around from before we were exec()'ed. Let's reap them. */
2625 r
= sd_event_source_set_enabled(m
->sigchld_event_source
, SD_EVENT_ON
);
2627 return log_error_errno(r
, "Failed to enable SIGCHLD event source: %m");
2629 while (m
->exit_code
== MANAGER_OK
) {
2632 if (m
->runtime_watchdog
> 0 && m
->runtime_watchdog
!= USEC_INFINITY
&& MANAGER_IS_SYSTEM(m
))
2635 if (!ratelimit_test(&rl
)) {
2636 /* Yay, something is going seriously wrong, pause a little */
2637 log_warning("Looping too fast. Throttling execution a little.");
2641 if (manager_dispatch_load_queue(m
) > 0)
2644 if (manager_dispatch_gc_job_queue(m
) > 0)
2647 if (manager_dispatch_gc_unit_queue(m
) > 0)
2650 if (manager_dispatch_cleanup_queue(m
) > 0)
2653 if (manager_dispatch_cgroup_realize_queue(m
) > 0)
2656 if (manager_dispatch_dbus_queue(m
) > 0)
2659 /* Sleep for half the watchdog time */
2660 if (m
->runtime_watchdog
> 0 && m
->runtime_watchdog
!= USEC_INFINITY
&& MANAGER_IS_SYSTEM(m
)) {
2661 wait_usec
= m
->runtime_watchdog
/ 2;
2665 wait_usec
= USEC_INFINITY
;
2667 r
= sd_event_run(m
->event
, wait_usec
);
2669 return log_error_errno(r
, "Failed to run event loop: %m");
2672 return m
->exit_code
;
2675 int manager_load_unit_from_dbus_path(Manager
*m
, const char *s
, sd_bus_error
*e
, Unit
**_u
) {
2676 _cleanup_free_
char *n
= NULL
;
2677 sd_id128_t invocation_id
;
2685 r
= unit_name_from_dbus_path(s
, &n
);
2689 /* Permit addressing units by invocation ID: if the passed bus path is suffixed by a 128bit ID then we use it
2690 * as invocation ID. */
2691 r
= sd_id128_from_string(n
, &invocation_id
);
2693 u
= hashmap_get(m
->units_by_invocation_id
, &invocation_id
);
2699 return sd_bus_error_setf(e
, BUS_ERROR_NO_UNIT_FOR_INVOCATION_ID
, "No unit with the specified invocation ID " SD_ID128_FORMAT_STR
" known.", SD_ID128_FORMAT_VAL(invocation_id
));
2702 /* If this didn't work, we check if this is a unit name */
2703 if (!unit_name_is_valid(n
, UNIT_NAME_PLAIN
|UNIT_NAME_INSTANCE
))
2704 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
, "Unit name %s is neither a valid invocation ID nor unit name.", n
);
2706 r
= manager_load_unit(m
, n
, NULL
, e
, &u
);
2714 int manager_get_job_from_dbus_path(Manager
*m
, const char *s
, Job
**_j
) {
2724 p
= startswith(s
, "/org/freedesktop/systemd1/job/");
2728 r
= safe_atou(p
, &id
);
2732 j
= manager_get_job(m
, id
);
2741 void manager_send_unit_audit(Manager
*m
, Unit
*u
, int type
, bool success
) {
2744 _cleanup_free_
char *p
= NULL
;
2748 if (!MANAGER_IS_SYSTEM(m
))
2751 audit_fd
= get_audit_fd();
2755 /* Don't generate audit events if the service was already
2756 * started and we're just deserializing */
2757 if (MANAGER_IS_RELOADING(m
))
2760 if (u
->type
!= UNIT_SERVICE
)
2763 r
= unit_name_to_prefix_and_instance(u
->id
, &p
);
2765 log_error_errno(r
, "Failed to extract prefix and instance of unit name: %m");
2769 msg
= strjoina("unit=", p
);
2770 if (audit_log_user_comm_message(audit_fd
, type
, msg
, "systemd", NULL
, NULL
, NULL
, success
) < 0) {
2772 /* We aren't allowed to send audit messages?
2773 * Then let's not retry again. */
2776 log_warning_errno(errno
, "Failed to send audit message: %m");
2782 void manager_send_unit_plymouth(Manager
*m
, Unit
*u
) {
2783 static const union sockaddr_union sa
= PLYMOUTH_SOCKET
;
2784 _cleanup_free_
char *message
= NULL
;
2785 _cleanup_close_
int fd
= -1;
2788 /* Don't generate plymouth events if the service was already
2789 * started and we're just deserializing */
2790 if (MANAGER_IS_RELOADING(m
))
2793 if (!MANAGER_IS_SYSTEM(m
))
2796 if (detect_container() > 0)
2799 if (!IN_SET(u
->type
, UNIT_SERVICE
, UNIT_MOUNT
, UNIT_SWAP
))
2802 /* We set SOCK_NONBLOCK here so that we rather drop the
2803 * message then wait for plymouth */
2804 fd
= socket(AF_UNIX
, SOCK_STREAM
|SOCK_CLOEXEC
|SOCK_NONBLOCK
, 0);
2806 log_error_errno(errno
, "socket() failed: %m");
2810 if (connect(fd
, &sa
.sa
, SOCKADDR_UN_LEN(sa
.un
)) < 0) {
2811 if (!IN_SET(errno
, EPIPE
, EAGAIN
, ENOENT
, ECONNREFUSED
, ECONNRESET
, ECONNABORTED
))
2812 log_error_errno(errno
, "connect() failed: %m");
2816 if (asprintf(&message
, "U\002%c%s%n", (int) (strlen(u
->id
) + 1), u
->id
, &n
) < 0) {
2822 if (write(fd
, message
, n
+ 1) != n
+ 1)
2823 if (!IN_SET(errno
, EPIPE
, EAGAIN
, ENOENT
, ECONNREFUSED
, ECONNRESET
, ECONNABORTED
))
2824 log_error_errno(errno
, "Failed to write Plymouth message: %m");
2827 int manager_open_serialization(Manager
*m
, FILE **_f
) {
2833 fd
= open_serialization_fd("systemd-state");
2837 f
= fdopen(fd
, "w+");
2847 int manager_serialize(Manager
*m
, FILE *f
, FDSet
*fds
, bool switching_root
) {
2860 fprintf(f
, "current-job-id=%"PRIu32
"\n", m
->current_job_id
);
2861 fprintf(f
, "n-installed-jobs=%u\n", m
->n_installed_jobs
);
2862 fprintf(f
, "n-failed-jobs=%u\n", m
->n_failed_jobs
);
2863 fprintf(f
, "taint-usr=%s\n", yes_no(m
->taint_usr
));
2864 fprintf(f
, "ready-sent=%s\n", yes_no(m
->ready_sent
));
2865 fprintf(f
, "taint-logged=%s\n", yes_no(m
->taint_logged
));
2866 fprintf(f
, "service-watchdogs=%s\n", yes_no(m
->service_watchdogs
));
2868 for (q
= 0; q
< _MANAGER_TIMESTAMP_MAX
; q
++) {
2869 /* The userspace and finish timestamps only apply to the host system, hence only serialize them there */
2870 if (in_initrd() && IN_SET(q
, MANAGER_TIMESTAMP_USERSPACE
, MANAGER_TIMESTAMP_FINISH
))
2873 t
= manager_timestamp_to_string(q
);
2875 char field
[strlen(t
) + STRLEN("-timestamp") + 1];
2876 strcpy(stpcpy(field
, t
), "-timestamp");
2877 dual_timestamp_serialize(f
, field
, m
->timestamps
+ q
);
2881 if (!switching_root
)
2882 (void) serialize_environment(f
, m
->environment
);
2884 if (m
->notify_fd
>= 0) {
2887 copy
= fdset_put_dup(fds
, m
->notify_fd
);
2891 fprintf(f
, "notify-fd=%i\n", copy
);
2892 fprintf(f
, "notify-socket=%s\n", m
->notify_socket
);
2895 if (m
->cgroups_agent_fd
>= 0) {
2898 copy
= fdset_put_dup(fds
, m
->cgroups_agent_fd
);
2902 fprintf(f
, "cgroups-agent-fd=%i\n", copy
);
2905 if (m
->user_lookup_fds
[0] >= 0) {
2908 copy0
= fdset_put_dup(fds
, m
->user_lookup_fds
[0]);
2912 copy1
= fdset_put_dup(fds
, m
->user_lookup_fds
[1]);
2916 fprintf(f
, "user-lookup=%i %i\n", copy0
, copy1
);
2919 bus_track_serialize(m
->subscribed
, f
, "subscribed");
2921 r
= dynamic_user_serialize(m
, f
, fds
);
2925 manager_serialize_uid_refs(m
, f
);
2926 manager_serialize_gid_refs(m
, f
);
2928 r
= exec_runtime_serialize(m
, f
, fds
);
2932 (void) fputc('\n', f
);
2934 HASHMAP_FOREACH_KEY(u
, t
, m
->units
, i
) {
2942 r
= unit_serialize(u
, f
, fds
, !switching_root
);
2949 assert(m
->n_reloading
> 0);
2955 r
= bus_fdset_add_all(m
, fds
);
2962 int manager_deserialize(Manager
*m
, FILE *f
, FDSet
*fds
) {
2968 log_debug("Deserializing state...");
2973 char line
[LINE_MAX
];
2974 const char *val
, *l
;
2976 if (!fgets(line
, sizeof(line
), f
)) {
2991 if ((val
= startswith(l
, "current-job-id="))) {
2994 if (safe_atou32(val
, &id
) < 0)
2995 log_notice("Failed to parse current job id value %s", val
);
2997 m
->current_job_id
= MAX(m
->current_job_id
, id
);
2999 } else if ((val
= startswith(l
, "n-installed-jobs="))) {
3002 if (safe_atou32(val
, &n
) < 0)
3003 log_notice("Failed to parse installed jobs counter %s", val
);
3005 m
->n_installed_jobs
+= n
;
3007 } else if ((val
= startswith(l
, "n-failed-jobs="))) {
3010 if (safe_atou32(val
, &n
) < 0)
3011 log_notice("Failed to parse failed jobs counter %s", val
);
3013 m
->n_failed_jobs
+= n
;
3015 } else if ((val
= startswith(l
, "taint-usr="))) {
3018 b
= parse_boolean(val
);
3020 log_notice("Failed to parse taint /usr flag %s", val
);
3022 m
->taint_usr
= m
->taint_usr
|| b
;
3024 } else if ((val
= startswith(l
, "ready-sent="))) {
3027 b
= parse_boolean(val
);
3029 log_notice("Failed to parse ready-sent flag %s", val
);
3031 m
->ready_sent
= m
->ready_sent
|| b
;
3033 } else if ((val
= startswith(l
, "taint-logged="))) {
3036 b
= parse_boolean(val
);
3038 log_notice("Failed to parse taint-logged flag %s", val
);
3040 m
->taint_logged
= m
->taint_logged
|| b
;
3042 } else if ((val
= startswith(l
, "service-watchdogs="))) {
3045 b
= parse_boolean(val
);
3047 log_notice("Failed to parse service-watchdogs flag %s", val
);
3049 m
->service_watchdogs
= b
;
3051 } else if (startswith(l
, "env=")) {
3052 r
= deserialize_environment(&m
->environment
, l
);
3056 log_notice_errno(r
, "Failed to parse environment entry: \"%s\": %m", l
);
3058 } else if ((val
= startswith(l
, "notify-fd="))) {
3061 if (safe_atoi(val
, &fd
) < 0 || fd
< 0 || !fdset_contains(fds
, fd
))
3062 log_notice("Failed to parse notify fd: \"%s\"", val
);
3064 m
->notify_event_source
= sd_event_source_unref(m
->notify_event_source
);
3065 safe_close(m
->notify_fd
);
3066 m
->notify_fd
= fdset_remove(fds
, fd
);
3069 } else if ((val
= startswith(l
, "notify-socket="))) {
3078 free(m
->notify_socket
);
3079 m
->notify_socket
= n
;
3081 } else if ((val
= startswith(l
, "cgroups-agent-fd="))) {
3084 if (safe_atoi(val
, &fd
) < 0 || fd
< 0 || !fdset_contains(fds
, fd
))
3085 log_notice("Failed to parse cgroups agent fd: %s", val
);
3087 m
->cgroups_agent_event_source
= sd_event_source_unref(m
->cgroups_agent_event_source
);
3088 safe_close(m
->cgroups_agent_fd
);
3089 m
->cgroups_agent_fd
= fdset_remove(fds
, fd
);
3092 } else if ((val
= startswith(l
, "user-lookup="))) {
3095 if (sscanf(val
, "%i %i", &fd0
, &fd1
) != 2 || fd0
< 0 || fd1
< 0 || fd0
== fd1
|| !fdset_contains(fds
, fd0
) || !fdset_contains(fds
, fd1
))
3096 log_notice("Failed to parse user lookup fd: %s", val
);
3098 m
->user_lookup_event_source
= sd_event_source_unref(m
->user_lookup_event_source
);
3099 safe_close_pair(m
->user_lookup_fds
);
3100 m
->user_lookup_fds
[0] = fdset_remove(fds
, fd0
);
3101 m
->user_lookup_fds
[1] = fdset_remove(fds
, fd1
);
3104 } else if ((val
= startswith(l
, "dynamic-user=")))
3105 dynamic_user_deserialize_one(m
, val
, fds
);
3106 else if ((val
= startswith(l
, "destroy-ipc-uid=")))
3107 manager_deserialize_uid_refs_one(m
, val
);
3108 else if ((val
= startswith(l
, "destroy-ipc-gid=")))
3109 manager_deserialize_gid_refs_one(m
, val
);
3110 else if ((val
= startswith(l
, "exec-runtime=")))
3111 exec_runtime_deserialize_one(m
, val
, fds
);
3112 else if ((val
= startswith(l
, "subscribed="))) {
3114 if (strv_extend(&m
->deserialized_subscribed
, val
) < 0)
3119 for (q
= 0; q
< _MANAGER_TIMESTAMP_MAX
; q
++) {
3120 val
= startswith(l
, manager_timestamp_to_string(q
));
3124 val
= startswith(val
, "-timestamp=");
3129 if (q
< _MANAGER_TIMESTAMP_MAX
) /* found it */
3130 dual_timestamp_deserialize(val
, m
->timestamps
+ q
);
3131 else if (!startswith(l
, "kdbus-fd=")) /* ignore kdbus */
3132 log_notice("Unknown serialization item '%s'", l
);
3138 char name
[UNIT_NAME_MAX
+2];
3139 const char* unit_name
;
3142 if (!fgets(name
, sizeof(name
), f
)) {
3152 unit_name
= strstrip(name
);
3154 r
= manager_load_unit(m
, unit_name
, NULL
, NULL
, &u
);
3156 log_notice_errno(r
, "Failed to load unit \"%s\", skipping deserialization: %m", unit_name
);
3159 unit_deserialize_skip(f
);
3163 r
= unit_deserialize(u
, f
, fds
);
3165 log_notice_errno(r
, "Failed to deserialize unit \"%s\": %m", unit_name
);
3175 assert(m
->n_reloading
> 0);
3181 int manager_reload(Manager
*m
) {
3183 _cleanup_fclose_
FILE *f
= NULL
;
3184 _cleanup_fdset_free_ FDSet
*fds
= NULL
;
3188 r
= manager_open_serialization(m
, &f
);
3193 bus_manager_send_reloading(m
, true);
3201 r
= manager_serialize(m
, f
, fds
, false);
3207 if (fseeko(f
, 0, SEEK_SET
) < 0) {
3212 /* From here on there is no way back. */
3213 manager_clear_jobs_and_units(m
);
3214 lookup_paths_flush_generator(&m
->lookup_paths
);
3215 lookup_paths_free(&m
->lookup_paths
);
3216 exec_runtime_vacuum(m
);
3217 dynamic_user_vacuum(m
, false);
3218 m
->uid_refs
= hashmap_free(m
->uid_refs
);
3219 m
->gid_refs
= hashmap_free(m
->gid_refs
);
3221 q
= lookup_paths_init(&m
->lookup_paths
, m
->unit_file_scope
, 0, NULL
);
3222 if (q
< 0 && r
>= 0)
3225 q
= manager_run_environment_generators(m
);
3226 if (q
< 0 && r
>= 0)
3229 /* Find new unit paths */
3230 q
= manager_run_generators(m
);
3231 if (q
< 0 && r
>= 0)
3234 lookup_paths_reduce(&m
->lookup_paths
);
3235 manager_build_unit_path_cache(m
);
3237 /* First, enumerate what we can from all config files */
3238 manager_enumerate(m
);
3240 /* Second, deserialize our stored data */
3241 q
= manager_deserialize(m
, f
, fds
);
3243 log_error_errno(q
, "Deserialization failed: %m");
3252 /* Re-register notify_fd as event source */
3253 q
= manager_setup_notify(m
);
3254 if (q
< 0 && r
>= 0)
3257 q
= manager_setup_cgroups_agent(m
);
3258 if (q
< 0 && r
>= 0)
3261 q
= manager_setup_user_lookup_fd(m
);
3262 if (q
< 0 && r
>= 0)
3265 /* Third, fire things up! */
3266 manager_coldplug(m
);
3268 /* Release any dynamic users no longer referenced */
3269 dynamic_user_vacuum(m
, true);
3271 /* Release any references to UIDs/GIDs no longer referenced, and destroy any IPC owned by them */
3272 manager_vacuum_uid_refs(m
);
3273 manager_vacuum_gid_refs(m
);
3275 exec_runtime_vacuum(m
);
3277 assert(m
->n_reloading
> 0);
3280 /* It might be safe to log to the journal now and connect to dbus */
3281 manager_recheck_journal(m
);
3282 manager_recheck_dbus(m
);
3284 /* Sync current state of bus names with our set of listening units */
3285 q
= manager_enqueue_sync_bus_names(m
);
3286 if (q
< 0 && r
>= 0)
3289 m
->send_reloading_done
= true;
3294 void manager_reset_failed(Manager
*m
) {
3300 HASHMAP_FOREACH(u
, m
->units
, i
)
3301 unit_reset_failed(u
);
3304 bool manager_unit_inactive_or_pending(Manager
*m
, const char *name
) {
3310 /* Returns true if the unit is inactive or going down */
3311 u
= manager_get_unit(m
, name
);
3315 return unit_inactive_or_pending(u
);
3318 static void log_taint_string(Manager
*m
) {
3319 _cleanup_free_
char *taint
= NULL
;
3323 if (MANAGER_IS_USER(m
) || m
->taint_logged
)
3326 m
->taint_logged
= true; /* only check for taint once */
3328 taint
= manager_taint_string(m
);
3332 log_struct(LOG_NOTICE
,
3333 LOG_MESSAGE("System is tainted: %s", taint
),
3335 "MESSAGE_ID=" SD_MESSAGE_TAINTED_STR
,
3339 static void manager_notify_finished(Manager
*m
) {
3340 char userspace
[FORMAT_TIMESPAN_MAX
], initrd
[FORMAT_TIMESPAN_MAX
], kernel
[FORMAT_TIMESPAN_MAX
], sum
[FORMAT_TIMESPAN_MAX
];
3341 usec_t firmware_usec
, loader_usec
, kernel_usec
, initrd_usec
, userspace_usec
, total_usec
;
3343 if (m
->test_run_flags
)
3346 if (MANAGER_IS_SYSTEM(m
) && detect_container() <= 0) {
3347 char ts
[FORMAT_TIMESPAN_MAX
];
3348 char buf
[FORMAT_TIMESPAN_MAX
+ STRLEN(" (firmware) + ") + FORMAT_TIMESPAN_MAX
+ STRLEN(" (loader) + ")]
3351 size_t size
= sizeof buf
;
3353 /* Note that MANAGER_TIMESTAMP_KERNEL's monotonic value is always at 0, and
3354 * MANAGER_TIMESTAMP_FIRMWARE's and MANAGER_TIMESTAMP_LOADER's monotonic value should be considered
3355 * negative values. */
3357 firmware_usec
= m
->timestamps
[MANAGER_TIMESTAMP_FIRMWARE
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_LOADER
].monotonic
;
3358 loader_usec
= m
->timestamps
[MANAGER_TIMESTAMP_LOADER
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_KERNEL
].monotonic
;
3359 userspace_usec
= m
->timestamps
[MANAGER_TIMESTAMP_FINISH
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_USERSPACE
].monotonic
;
3360 total_usec
= m
->timestamps
[MANAGER_TIMESTAMP_FIRMWARE
].monotonic
+ m
->timestamps
[MANAGER_TIMESTAMP_FINISH
].monotonic
;
3362 if (firmware_usec
> 0)
3363 size
= strpcpyf(&p
, size
, "%s (firmware) + ", format_timespan(ts
, sizeof(ts
), firmware_usec
, USEC_PER_MSEC
));
3364 if (loader_usec
> 0)
3365 size
= strpcpyf(&p
, size
, "%s (loader) + ", format_timespan(ts
, sizeof(ts
), loader_usec
, USEC_PER_MSEC
));
3367 if (dual_timestamp_is_set(&m
->timestamps
[MANAGER_TIMESTAMP_INITRD
])) {
3369 /* The initrd case on bare-metal*/
3370 kernel_usec
= m
->timestamps
[MANAGER_TIMESTAMP_INITRD
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_KERNEL
].monotonic
;
3371 initrd_usec
= m
->timestamps
[MANAGER_TIMESTAMP_USERSPACE
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_INITRD
].monotonic
;
3373 log_struct(LOG_INFO
,
3374 "MESSAGE_ID=" SD_MESSAGE_STARTUP_FINISHED_STR
,
3375 "KERNEL_USEC="USEC_FMT
, kernel_usec
,
3376 "INITRD_USEC="USEC_FMT
, initrd_usec
,
3377 "USERSPACE_USEC="USEC_FMT
, userspace_usec
,
3378 LOG_MESSAGE("Startup finished in %s%s (kernel) + %s (initrd) + %s (userspace) = %s.",
3380 format_timespan(kernel
, sizeof(kernel
), kernel_usec
, USEC_PER_MSEC
),
3381 format_timespan(initrd
, sizeof(initrd
), initrd_usec
, USEC_PER_MSEC
),
3382 format_timespan(userspace
, sizeof(userspace
), userspace_usec
, USEC_PER_MSEC
),
3383 format_timespan(sum
, sizeof(sum
), total_usec
, USEC_PER_MSEC
)),
3386 /* The initrd-less case on bare-metal*/
3388 kernel_usec
= m
->timestamps
[MANAGER_TIMESTAMP_USERSPACE
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_KERNEL
].monotonic
;
3391 log_struct(LOG_INFO
,
3392 "MESSAGE_ID=" SD_MESSAGE_STARTUP_FINISHED_STR
,
3393 "KERNEL_USEC="USEC_FMT
, kernel_usec
,
3394 "USERSPACE_USEC="USEC_FMT
, userspace_usec
,
3395 LOG_MESSAGE("Startup finished in %s%s (kernel) + %s (userspace) = %s.",
3397 format_timespan(kernel
, sizeof(kernel
), kernel_usec
, USEC_PER_MSEC
),
3398 format_timespan(userspace
, sizeof(userspace
), userspace_usec
, USEC_PER_MSEC
),
3399 format_timespan(sum
, sizeof(sum
), total_usec
, USEC_PER_MSEC
)),
3403 /* The container and --user case */
3404 firmware_usec
= loader_usec
= initrd_usec
= kernel_usec
= 0;
3405 total_usec
= userspace_usec
= m
->timestamps
[MANAGER_TIMESTAMP_FINISH
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_USERSPACE
].monotonic
;
3407 log_struct(LOG_INFO
,
3408 "MESSAGE_ID=" SD_MESSAGE_USER_STARTUP_FINISHED_STR
,
3409 "USERSPACE_USEC="USEC_FMT
, userspace_usec
,
3410 LOG_MESSAGE("Startup finished in %s.",
3411 format_timespan(sum
, sizeof(sum
), total_usec
, USEC_PER_MSEC
)),
3415 bus_manager_send_finished(m
, firmware_usec
, loader_usec
, kernel_usec
, initrd_usec
, userspace_usec
, total_usec
);
3418 m
->ready_sent
? "STATUS=Startup finished in %s."
3420 "STATUS=Startup finished in %s.",
3421 format_timespan(sum
, sizeof(sum
), total_usec
, USEC_PER_MSEC
));
3422 m
->ready_sent
= true;
3424 log_taint_string(m
);
3427 static void manager_send_ready(Manager
*m
) {
3430 /* We send READY=1 on reaching basic.target only when running in --user mode. */
3431 if (!MANAGER_IS_USER(m
) || m
->ready_sent
)
3434 m
->ready_sent
= true;
3438 "STATUS=Reached " SPECIAL_BASIC_TARGET
".");
3441 static void manager_check_basic_target(Manager
*m
) {
3446 /* Small shortcut */
3447 if (m
->ready_sent
&& m
->taint_logged
)
3450 u
= manager_get_unit(m
, SPECIAL_BASIC_TARGET
);
3451 if (!u
|| !UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u
)))
3454 /* For user managers, send out READY=1 as soon as we reach basic.target */
3455 manager_send_ready(m
);
3457 /* Log the taint string as soon as we reach basic.target */
3458 log_taint_string(m
);
3461 void manager_check_finished(Manager
*m
) {
3464 if (MANAGER_IS_RELOADING(m
))
3467 /* Verify that we have entered the event loop already, and not left it again. */
3468 if (!MANAGER_IS_RUNNING(m
))
3471 manager_check_basic_target(m
);
3473 if (hashmap_size(m
->jobs
) > 0) {
3474 if (m
->jobs_in_progress_event_source
)
3475 /* Ignore any failure, this is only for feedback */
3476 (void) sd_event_source_set_time(m
->jobs_in_progress_event_source
, now(CLOCK_MONOTONIC
) + JOBS_IN_PROGRESS_WAIT_USEC
);
3481 manager_flip_auto_status(m
, false);
3483 /* Notify Type=idle units that we are done now */
3484 manager_close_idle_pipe(m
);
3486 /* Turn off confirm spawn now */
3487 m
->confirm_spawn
= NULL
;
3489 /* No need to update ask password status when we're going non-interactive */
3490 manager_close_ask_password(m
);
3492 /* This is no longer the first boot */
3493 manager_set_first_boot(m
, false);
3495 if (MANAGER_IS_FINISHED(m
))
3498 dual_timestamp_get(m
->timestamps
+ MANAGER_TIMESTAMP_FINISH
);
3500 manager_notify_finished(m
);
3502 manager_invalidate_startup_units(m
);
3505 static bool generator_path_any(const char* const* paths
) {
3509 /* Optimize by skipping the whole process by not creating output directories
3510 * if no generators are found. */
3511 STRV_FOREACH(path
, (char**) paths
)
3512 if (access(*path
, F_OK
) == 0)
3514 else if (errno
!= ENOENT
)
3515 log_warning_errno(errno
, "Failed to open generator directory %s: %m", *path
);
3520 static const char* system_env_generator_binary_paths
[] = {
3521 "/run/systemd/system-environment-generators",
3522 "/etc/systemd/system-environment-generators",
3523 "/usr/local/lib/systemd/system-environment-generators",
3524 SYSTEM_ENV_GENERATOR_PATH
,
3528 static const char* user_env_generator_binary_paths
[] = {
3529 "/run/systemd/user-environment-generators",
3530 "/etc/systemd/user-environment-generators",
3531 "/usr/local/lib/systemd/user-environment-generators",
3532 USER_ENV_GENERATOR_PATH
,
3536 static int manager_run_environment_generators(Manager
*m
) {
3537 char **tmp
= NULL
; /* this is only used in the forked process, no cleanup here */
3539 void* args
[] = {&tmp
, &tmp
, &m
->environment
};
3541 if (m
->test_run_flags
&& !(m
->test_run_flags
& MANAGER_TEST_RUN_ENV_GENERATORS
))
3544 paths
= MANAGER_IS_SYSTEM(m
) ? system_env_generator_binary_paths
: user_env_generator_binary_paths
;
3546 if (!generator_path_any(paths
))
3549 return execute_directories(paths
, DEFAULT_TIMEOUT_USEC
, gather_environment
, args
, NULL
);
3552 static int manager_run_generators(Manager
*m
) {
3553 _cleanup_strv_free_
char **paths
= NULL
;
3554 const char *argv
[5];
3559 if (m
->test_run_flags
&& !(m
->test_run_flags
& MANAGER_TEST_RUN_GENERATORS
))
3562 paths
= generator_binary_paths(m
->unit_file_scope
);
3566 if (!generator_path_any((const char* const*) paths
))
3569 r
= lookup_paths_mkdir_generator(&m
->lookup_paths
);
3573 argv
[0] = NULL
; /* Leave this empty, execute_directory() will fill something in */
3574 argv
[1] = m
->lookup_paths
.generator
;
3575 argv
[2] = m
->lookup_paths
.generator_early
;
3576 argv
[3] = m
->lookup_paths
.generator_late
;
3579 RUN_WITH_UMASK(0022)
3580 execute_directories((const char* const*) paths
, DEFAULT_TIMEOUT_USEC
,
3581 NULL
, NULL
, (char**) argv
);
3584 lookup_paths_trim_generator(&m
->lookup_paths
);
3588 int manager_environment_add(Manager
*m
, char **minus
, char **plus
) {
3589 char **a
= NULL
, **b
= NULL
, **l
;
3594 if (!strv_isempty(minus
)) {
3595 a
= strv_env_delete(l
, 1, minus
);
3602 if (!strv_isempty(plus
)) {
3603 b
= strv_env_merge(2, l
, plus
);
3612 if (m
->environment
!= l
)
3613 strv_free(m
->environment
);
3620 manager_sanitize_environment(m
);
3625 int manager_set_default_rlimits(Manager
*m
, struct rlimit
**default_rlimit
) {
3630 for (i
= 0; i
< _RLIMIT_MAX
; i
++) {
3631 m
->rlimit
[i
] = mfree(m
->rlimit
[i
]);
3633 if (!default_rlimit
[i
])
3636 m
->rlimit
[i
] = newdup(struct rlimit
, default_rlimit
[i
], 1);
3644 void manager_recheck_dbus(Manager
*m
) {
3647 /* Connects to the bus if the dbus service and socket are running. If we are running in user mode this is all
3648 * it does. In system mode we'll also connect to the system bus (which will most likely just reuse the
3649 * connection of the API bus). That's because the system bus after all runs as service of the system instance,
3650 * while in the user instance we can assume it's already there. */
3652 if (MANAGER_IS_RELOADING(m
))
3653 return; /* don't check while we are reloading… */
3655 if (manager_dbus_is_running(m
, false)) {
3656 (void) bus_init_api(m
);
3658 if (MANAGER_IS_SYSTEM(m
))
3659 (void) bus_init_system(m
);
3661 (void) bus_done_api(m
);
3663 if (MANAGER_IS_SYSTEM(m
))
3664 (void) bus_done_system(m
);
3668 static bool manager_journal_is_running(Manager
*m
) {
3673 if (m
->test_run_flags
!= 0)
3676 /* If we are the user manager we can safely assume that the journal is up */
3677 if (!MANAGER_IS_SYSTEM(m
))
3680 /* Check that the socket is not only up, but in RUNNING state */
3681 u
= manager_get_unit(m
, SPECIAL_JOURNALD_SOCKET
);
3684 if (SOCKET(u
)->state
!= SOCKET_RUNNING
)
3687 /* Similar, check if the daemon itself is fully up, too */
3688 u
= manager_get_unit(m
, SPECIAL_JOURNALD_SERVICE
);
3691 if (!IN_SET(SERVICE(u
)->state
, SERVICE_RELOAD
, SERVICE_RUNNING
))
3697 void manager_recheck_journal(Manager
*m
) {
3701 /* Don't bother with this unless we are in the special situation of being PID 1 */
3702 if (getpid_cached() != 1)
3705 /* Don't check this while we are reloading, things might still change */
3706 if (MANAGER_IS_RELOADING(m
))
3709 /* The journal is fully and entirely up? If so, let's permit logging to it, if that's configured. If the
3710 * journal is down, don't ever log to it, otherwise we might end up deadlocking ourselves as we might trigger
3711 * an activation ourselves we can't fulfill. */
3712 log_set_prohibit_ipc(!manager_journal_is_running(m
));
3716 void manager_set_show_status(Manager
*m
, ShowStatus mode
) {
3718 assert(IN_SET(mode
, SHOW_STATUS_AUTO
, SHOW_STATUS_NO
, SHOW_STATUS_YES
, SHOW_STATUS_TEMPORARY
));
3720 if (!MANAGER_IS_SYSTEM(m
))
3723 if (m
->show_status
!= mode
)
3724 log_debug("%s showing of status.",
3725 mode
== SHOW_STATUS_NO
? "Disabling" : "Enabling");
3726 m
->show_status
= mode
;
3729 (void) touch("/run/systemd/show-status");
3731 (void) unlink("/run/systemd/show-status");
3734 static bool manager_get_show_status(Manager
*m
, StatusType type
) {
3737 if (!MANAGER_IS_SYSTEM(m
))
3740 if (m
->no_console_output
)
3743 if (!IN_SET(manager_state(m
), MANAGER_INITIALIZING
, MANAGER_STARTING
, MANAGER_STOPPING
))
3746 /* If we cannot find out the status properly, just proceed. */
3747 if (type
!= STATUS_TYPE_EMERGENCY
&& manager_check_ask_password(m
) > 0)
3750 return m
->show_status
> 0;
3753 const char *manager_get_confirm_spawn(Manager
*m
) {
3754 static int last_errno
= 0;
3755 const char *vc
= m
->confirm_spawn
;
3759 /* Here's the deal: we want to test the validity of the console but don't want
3760 * PID1 to go through the whole console process which might block. But we also
3761 * want to warn the user only once if something is wrong with the console so we
3762 * cannot do the sanity checks after spawning our children. So here we simply do
3763 * really basic tests to hopefully trap common errors.
3765 * If the console suddenly disappear at the time our children will really it
3766 * then they will simply fail to acquire it and a positive answer will be
3767 * assumed. New children will fallback to /dev/console though.
3769 * Note: TTYs are devices that can come and go any time, and frequently aren't
3770 * available yet during early boot (consider a USB rs232 dongle...). If for any
3771 * reason the configured console is not ready, we fallback to the default
3774 if (!vc
|| path_equal(vc
, "/dev/console"))
3781 if (!S_ISCHR(st
.st_mode
)) {
3789 if (last_errno
!= errno
) {
3791 log_warning_errno(errno
, "Failed to open %s: %m, using default console", vc
);
3793 return "/dev/console";
3796 void manager_set_first_boot(Manager
*m
, bool b
) {
3799 if (!MANAGER_IS_SYSTEM(m
))
3802 if (m
->first_boot
!= (int) b
) {
3804 (void) touch("/run/systemd/first-boot");
3806 (void) unlink("/run/systemd/first-boot");
3812 void manager_disable_confirm_spawn(void) {
3813 (void) touch("/run/systemd/confirm_spawn_disabled");
3816 bool manager_is_confirm_spawn_disabled(Manager
*m
) {
3817 if (!m
->confirm_spawn
)
3820 return access("/run/systemd/confirm_spawn_disabled", F_OK
) >= 0;
3823 void manager_status_printf(Manager
*m
, StatusType type
, const char *status
, const char *format
, ...) {
3826 /* If m is NULL, assume we're after shutdown and let the messages through. */
3828 if (m
&& !manager_get_show_status(m
, type
))
3831 /* XXX We should totally drop the check for ephemeral here
3832 * and thus effectively make 'Type=idle' pointless. */
3833 if (type
== STATUS_TYPE_EPHEMERAL
&& m
&& m
->n_on_console
> 0)
3836 va_start(ap
, format
);
3837 status_vprintf(status
, true, type
== STATUS_TYPE_EPHEMERAL
, format
, ap
);
3841 Set
*manager_get_units_requiring_mounts_for(Manager
*m
, const char *path
) {
3842 char p
[strlen(path
)+1];
3848 path_kill_slashes(p
);
3850 return hashmap_get(m
->units_requiring_mounts_for
, streq(p
, "/") ? "" : p
);
3853 int manager_update_failed_units(Manager
*m
, Unit
*u
, bool failed
) {
3858 assert(u
->manager
== m
);
3860 size
= set_size(m
->failed_units
);
3863 r
= set_ensure_allocated(&m
->failed_units
, NULL
);
3867 if (set_put(m
->failed_units
, u
) < 0)
3870 (void) set_remove(m
->failed_units
, u
);
3872 if (set_size(m
->failed_units
) != size
)
3873 bus_manager_send_change_signal(m
);
3878 ManagerState
manager_state(Manager
*m
) {
3883 /* Did we ever finish booting? If not then we are still starting up */
3884 if (!MANAGER_IS_FINISHED(m
)) {
3886 u
= manager_get_unit(m
, SPECIAL_BASIC_TARGET
);
3887 if (!u
|| !UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u
)))
3888 return MANAGER_INITIALIZING
;
3890 return MANAGER_STARTING
;
3893 /* Is the special shutdown target active or queued? If so, we are in shutdown state */
3894 u
= manager_get_unit(m
, SPECIAL_SHUTDOWN_TARGET
);
3895 if (u
&& unit_active_or_pending(u
))
3896 return MANAGER_STOPPING
;
3898 if (MANAGER_IS_SYSTEM(m
)) {
3899 /* Are the rescue or emergency targets active or queued? If so we are in maintenance state */
3900 u
= manager_get_unit(m
, SPECIAL_RESCUE_TARGET
);
3901 if (u
&& unit_active_or_pending(u
))
3902 return MANAGER_MAINTENANCE
;
3904 u
= manager_get_unit(m
, SPECIAL_EMERGENCY_TARGET
);
3905 if (u
&& unit_active_or_pending(u
))
3906 return MANAGER_MAINTENANCE
;
3909 /* Are there any failed units? If so, we are in degraded mode */
3910 if (set_size(m
->failed_units
) > 0)
3911 return MANAGER_DEGRADED
;
3913 return MANAGER_RUNNING
;
3916 #define DESTROY_IPC_FLAG (UINT32_C(1) << 31)
3918 static void manager_unref_uid_internal(
3923 int (*_clean_ipc
)(uid_t uid
)) {
3929 assert(uid_is_valid(uid
));
3932 /* A generic implementation, covering both manager_unref_uid() and manager_unref_gid(), under the assumption
3933 * that uid_t and gid_t are actually defined the same way, with the same validity rules.
3935 * We store a hashmap where the UID/GID is they key and the value is a 32bit reference counter, whose highest
3936 * bit is used as flag for marking UIDs/GIDs whose IPC objects to remove when the last reference to the UID/GID
3937 * is dropped. The flag is set to on, once at least one reference from a unit where RemoveIPC= is set is added
3938 * on a UID/GID. It is reset when the UID's/GID's reference counter drops to 0 again. */
3940 assert_cc(sizeof(uid_t
) == sizeof(gid_t
));
3941 assert_cc(UID_INVALID
== (uid_t
) GID_INVALID
);
3943 if (uid
== 0) /* We don't keep track of root, and will never destroy it */
3946 c
= PTR_TO_UINT32(hashmap_get(*uid_refs
, UID_TO_PTR(uid
)));
3948 n
= c
& ~DESTROY_IPC_FLAG
;
3952 if (destroy_now
&& n
== 0) {
3953 hashmap_remove(*uid_refs
, UID_TO_PTR(uid
));
3955 if (c
& DESTROY_IPC_FLAG
) {
3956 log_debug("%s " UID_FMT
" is no longer referenced, cleaning up its IPC.",
3957 _clean_ipc
== clean_ipc_by_uid
? "UID" : "GID",
3959 (void) _clean_ipc(uid
);
3962 c
= n
| (c
& DESTROY_IPC_FLAG
);
3963 assert_se(hashmap_update(*uid_refs
, UID_TO_PTR(uid
), UINT32_TO_PTR(c
)) >= 0);
3967 void manager_unref_uid(Manager
*m
, uid_t uid
, bool destroy_now
) {
3968 manager_unref_uid_internal(m
, &m
->uid_refs
, uid
, destroy_now
, clean_ipc_by_uid
);
3971 void manager_unref_gid(Manager
*m
, gid_t gid
, bool destroy_now
) {
3972 manager_unref_uid_internal(m
, &m
->gid_refs
, (uid_t
) gid
, destroy_now
, clean_ipc_by_gid
);
3975 static int manager_ref_uid_internal(
3986 assert(uid_is_valid(uid
));
3988 /* A generic implementation, covering both manager_ref_uid() and manager_ref_gid(), under the assumption
3989 * that uid_t and gid_t are actually defined the same way, with the same validity rules. */
3991 assert_cc(sizeof(uid_t
) == sizeof(gid_t
));
3992 assert_cc(UID_INVALID
== (uid_t
) GID_INVALID
);
3994 if (uid
== 0) /* We don't keep track of root, and will never destroy it */
3997 r
= hashmap_ensure_allocated(uid_refs
, &trivial_hash_ops
);
4001 c
= PTR_TO_UINT32(hashmap_get(*uid_refs
, UID_TO_PTR(uid
)));
4003 n
= c
& ~DESTROY_IPC_FLAG
;
4006 if (n
& DESTROY_IPC_FLAG
) /* check for overflow */
4009 c
= n
| (c
& DESTROY_IPC_FLAG
) | (clean_ipc
? DESTROY_IPC_FLAG
: 0);
4011 return hashmap_replace(*uid_refs
, UID_TO_PTR(uid
), UINT32_TO_PTR(c
));
4014 int manager_ref_uid(Manager
*m
, uid_t uid
, bool clean_ipc
) {
4015 return manager_ref_uid_internal(m
, &m
->uid_refs
, uid
, clean_ipc
);
4018 int manager_ref_gid(Manager
*m
, gid_t gid
, bool clean_ipc
) {
4019 return manager_ref_uid_internal(m
, &m
->gid_refs
, (uid_t
) gid
, clean_ipc
);
4022 static void manager_vacuum_uid_refs_internal(
4025 int (*_clean_ipc
)(uid_t uid
)) {
4034 HASHMAP_FOREACH_KEY(p
, k
, *uid_refs
, i
) {
4038 uid
= PTR_TO_UID(k
);
4039 c
= PTR_TO_UINT32(p
);
4041 n
= c
& ~DESTROY_IPC_FLAG
;
4045 if (c
& DESTROY_IPC_FLAG
) {
4046 log_debug("Found unreferenced %s " UID_FMT
" after reload/reexec. Cleaning up.",
4047 _clean_ipc
== clean_ipc_by_uid
? "UID" : "GID",
4049 (void) _clean_ipc(uid
);
4052 assert_se(hashmap_remove(*uid_refs
, k
) == p
);
4056 void manager_vacuum_uid_refs(Manager
*m
) {
4057 manager_vacuum_uid_refs_internal(m
, &m
->uid_refs
, clean_ipc_by_uid
);
4060 void manager_vacuum_gid_refs(Manager
*m
) {
4061 manager_vacuum_uid_refs_internal(m
, &m
->gid_refs
, clean_ipc_by_gid
);
4064 static void manager_serialize_uid_refs_internal(
4068 const char *field_name
) {
4078 /* Serialize the UID reference table. Or actually, just the IPC destruction flag of it, as the actual counter
4079 * of it is better rebuild after a reload/reexec. */
4081 HASHMAP_FOREACH_KEY(p
, k
, *uid_refs
, i
) {
4085 uid
= PTR_TO_UID(k
);
4086 c
= PTR_TO_UINT32(p
);
4088 if (!(c
& DESTROY_IPC_FLAG
))
4091 fprintf(f
, "%s=" UID_FMT
"\n", field_name
, uid
);
4095 void manager_serialize_uid_refs(Manager
*m
, FILE *f
) {
4096 manager_serialize_uid_refs_internal(m
, f
, &m
->uid_refs
, "destroy-ipc-uid");
4099 void manager_serialize_gid_refs(Manager
*m
, FILE *f
) {
4100 manager_serialize_uid_refs_internal(m
, f
, &m
->gid_refs
, "destroy-ipc-gid");
4103 static void manager_deserialize_uid_refs_one_internal(
4106 const char *value
) {
4116 r
= parse_uid(value
, &uid
);
4117 if (r
< 0 || uid
== 0) {
4118 log_debug("Unable to parse UID reference serialization");
4122 r
= hashmap_ensure_allocated(uid_refs
, &trivial_hash_ops
);
4128 c
= PTR_TO_UINT32(hashmap_get(*uid_refs
, UID_TO_PTR(uid
)));
4129 if (c
& DESTROY_IPC_FLAG
)
4132 c
|= DESTROY_IPC_FLAG
;
4134 r
= hashmap_replace(*uid_refs
, UID_TO_PTR(uid
), UINT32_TO_PTR(c
));
4136 log_debug("Failed to add UID reference entry");
4141 void manager_deserialize_uid_refs_one(Manager
*m
, const char *value
) {
4142 manager_deserialize_uid_refs_one_internal(m
, &m
->uid_refs
, value
);
4145 void manager_deserialize_gid_refs_one(Manager
*m
, const char *value
) {
4146 manager_deserialize_uid_refs_one_internal(m
, &m
->gid_refs
, value
);
4149 int manager_dispatch_user_lookup_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
4153 char unit_name
[UNIT_NAME_MAX
+1];
4156 Manager
*m
= userdata
;
4164 /* Invoked whenever a child process succeeded resolving its user/group to use and sent us the resulting UID/GID
4165 * in a datagram. We parse the datagram here and pass it off to the unit, so that it can add a reference to the
4166 * UID/GID so that it can destroy the UID/GID's IPC objects when the reference counter drops to 0. */
4168 l
= recv(fd
, &buffer
, sizeof(buffer
), MSG_DONTWAIT
);
4170 if (IN_SET(errno
, EINTR
, EAGAIN
))
4173 return log_error_errno(errno
, "Failed to read from user lookup fd: %m");
4176 if ((size_t) l
<= offsetof(struct buffer
, unit_name
)) {
4177 log_warning("Received too short user lookup message, ignoring.");
4181 if ((size_t) l
> offsetof(struct buffer
, unit_name
) + UNIT_NAME_MAX
) {
4182 log_warning("Received too long user lookup message, ignoring.");
4186 if (!uid_is_valid(buffer
.uid
) && !gid_is_valid(buffer
.gid
)) {
4187 log_warning("Got user lookup message with invalid UID/GID pair, ignoring.");
4191 n
= (size_t) l
- offsetof(struct buffer
, unit_name
);
4192 if (memchr(buffer
.unit_name
, 0, n
)) {
4193 log_warning("Received lookup message with embedded NUL character, ignoring.");
4197 buffer
.unit_name
[n
] = 0;
4198 u
= manager_get_unit(m
, buffer
.unit_name
);
4200 log_debug("Got user lookup message but unit doesn't exist, ignoring.");
4204 log_unit_debug(u
, "User lookup succeeded: uid=" UID_FMT
" gid=" GID_FMT
, buffer
.uid
, buffer
.gid
);
4206 unit_notify_user_lookup(u
, buffer
.uid
, buffer
.gid
);
4210 char *manager_taint_string(Manager
*m
) {
4211 _cleanup_free_
char *destination
= NULL
, *overflowuid
= NULL
, *overflowgid
= NULL
;
4215 /* Returns a "taint string", e.g. "local-hwclock:var-run-bad".
4216 * Only things that are detected at runtime should be tagged
4217 * here. For stuff that is set during compilation, emit a warning
4218 * in the configuration phase. */
4222 buf
= new(char, sizeof("split-usr:"
4226 "overflowuid-not-65534:"
4227 "overflowgid-not-65534:"));
4235 e
= stpcpy(e
, "split-usr:");
4237 if (access("/proc/cgroups", F_OK
) < 0)
4238 e
= stpcpy(e
, "cgroups-missing:");
4240 if (clock_is_localtime(NULL
) > 0)
4241 e
= stpcpy(e
, "local-hwclock:");
4243 r
= readlink_malloc("/var/run", &destination
);
4244 if (r
< 0 || !PATH_IN_SET(destination
, "../run", "/run"))
4245 e
= stpcpy(e
, "var-run-bad:");
4247 r
= read_one_line_file("/proc/sys/kernel/overflowuid", &overflowuid
);
4248 if (r
>= 0 && !streq(overflowuid
, "65534"))
4249 e
= stpcpy(e
, "overflowuid-not-65534:");
4251 r
= read_one_line_file("/proc/sys/kernel/overflowgid", &overflowgid
);
4252 if (r
>= 0 && !streq(overflowgid
, "65534"))
4253 e
= stpcpy(e
, "overflowgid-not-65534:");
4255 /* remove the last ':' */
4262 void manager_ref_console(Manager
*m
) {
4268 void manager_unref_console(Manager
*m
) {
4270 assert(m
->n_on_console
> 0);
4273 if (m
->n_on_console
== 0)
4274 m
->no_console_output
= false; /* unset no_console_output flag, since the console is definitely free now */
4277 static const char *const manager_state_table
[_MANAGER_STATE_MAX
] = {
4278 [MANAGER_INITIALIZING
] = "initializing",
4279 [MANAGER_STARTING
] = "starting",
4280 [MANAGER_RUNNING
] = "running",
4281 [MANAGER_DEGRADED
] = "degraded",
4282 [MANAGER_MAINTENANCE
] = "maintenance",
4283 [MANAGER_STOPPING
] = "stopping",
4286 DEFINE_STRING_TABLE_LOOKUP(manager_state
, ManagerState
);
4288 static const char *const manager_timestamp_table
[_MANAGER_TIMESTAMP_MAX
] = {
4289 [MANAGER_TIMESTAMP_FIRMWARE
] = "firmware",
4290 [MANAGER_TIMESTAMP_LOADER
] = "loader",
4291 [MANAGER_TIMESTAMP_KERNEL
] = "kernel",
4292 [MANAGER_TIMESTAMP_INITRD
] = "initrd",
4293 [MANAGER_TIMESTAMP_USERSPACE
] = "userspace",
4294 [MANAGER_TIMESTAMP_FINISH
] = "finish",
4295 [MANAGER_TIMESTAMP_SECURITY_START
] = "security-start",
4296 [MANAGER_TIMESTAMP_SECURITY_FINISH
] = "security-finish",
4297 [MANAGER_TIMESTAMP_GENERATORS_START
] = "generators-start",
4298 [MANAGER_TIMESTAMP_GENERATORS_FINISH
] = "generators-finish",
4299 [MANAGER_TIMESTAMP_UNITS_LOAD_START
] = "units-load-start",
4300 [MANAGER_TIMESTAMP_UNITS_LOAD_FINISH
] = "units-load-finish",
4303 DEFINE_STRING_TABLE_LOOKUP(manager_timestamp
, ManagerTimestamp
);