]> git.ipfire.org Git - thirdparty/systemd.git/blob - src/core/manager.c
projects / thirdparty / systemd.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
tree-wide: use ASSERT_PTR more
[thirdparty/systemd.git] / src / core / manager.c
1 /* SPDX-License-Identifier: LGPL-2.1-or-later */
2
3 #include <errno.h>
4 #include <fcntl.h>
5 #include <linux/kd.h>
6 #include <sys/epoll.h>
7 #include <sys/inotify.h>
8 #include <sys/ioctl.h>
9 #include <sys/reboot.h>
10 #include <sys/timerfd.h>
11 #include <sys/utsname.h>
12 #include <sys/wait.h>
13 #include <unistd.h>
14
15 #if HAVE_AUDIT
16 #include <libaudit.h>
17 #endif
18
19 #include "sd-daemon.h"
20 #include "sd-messages.h"
21 #include "sd-path.h"
22
23 #include "all-units.h"
24 #include "alloc-util.h"
25 #include "audit-fd.h"
26 #include "boot-timestamps.h"
27 #include "bus-common-errors.h"
28 #include "bus-error.h"
29 #include "bus-kernel.h"
30 #include "bus-util.h"
31 #include "clean-ipc.h"
32 #include "clock-util.h"
33 #include "core-varlink.h"
34 #include "creds-util.h"
35 #include "dbus-job.h"
36 #include "dbus-manager.h"
37 #include "dbus-unit.h"
38 #include "dbus.h"
39 #include "def.h"
40 #include "dirent-util.h"
41 #include "env-util.h"
42 #include "escape.h"
43 #include "event-util.h"
44 #include "exec-util.h"
45 #include "execute.h"
46 #include "exit-status.h"
47 #include "fd-util.h"
48 #include "fileio.h"
49 #include "generator-setup.h"
50 #include "hashmap.h"
51 #include "inotify-util.h"
52 #include "install.h"
53 #include "io-util.h"
54 #include "label.h"
55 #include "load-fragment.h"
56 #include "locale-setup.h"
57 #include "log.h"
58 #include "macro.h"
59 #include "manager.h"
60 #include "manager-dump.h"
61 #include "manager-serialize.h"
62 #include "memory-util.h"
63 #include "mkdir-label.h"
64 #include "os-util.h"
65 #include "parse-util.h"
66 #include "path-lookup.h"
67 #include "path-util.h"
68 #include "process-util.h"
69 #include "ratelimit.h"
70 #include "rlimit-util.h"
71 #include "rm-rf.h"
72 #include "selinux-util.h"
73 #include "signal-util.h"
74 #include "socket-util.h"
75 #include "special.h"
76 #include "stat-util.h"
77 #include "string-table.h"
78 #include "string-util.h"
79 #include "strv.h"
80 #include "strxcpyx.h"
81 #include "sysctl-util.h"
82 #include "syslog-util.h"
83 #include "terminal-util.h"
84 #include "time-util.h"
85 #include "transaction.h"
86 #include "uid-range.h"
87 #include "umask-util.h"
88 #include "unit-name.h"
89 #include "user-util.h"
90 #include "virt.h"
91 #include "watchdog.h"
92
93 #define NOTIFY_RCVBUF_SIZE (8*1024*1024)
94 #define CGROUPS_AGENT_RCVBUF_SIZE (8*1024*1024)
95
96 /* Initial delay and the interval for printing status messages about running jobs */
97 #define JOBS_IN_PROGRESS_WAIT_USEC (2*USEC_PER_SEC)
98 #define JOBS_IN_PROGRESS_QUIET_WAIT_USEC (25*USEC_PER_SEC)
99 #define JOBS_IN_PROGRESS_PERIOD_USEC (USEC_PER_SEC / 3)
100 #define JOBS_IN_PROGRESS_PERIOD_DIVISOR 3
101
102 /* If there are more than 1K bus messages queue across our API and direct buses, then let's not add more on top until
103 * the queue gets more empty. */
104 #define MANAGER_BUS_BUSY_THRESHOLD 1024LU
105
106 /* How many units and jobs to process of the bus queue before returning to the event loop. */
107 #define MANAGER_BUS_MESSAGE_BUDGET 100U
108
109 static int manager_dispatch_notify_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata);
110 static int manager_dispatch_cgroups_agent_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata);
111 static int manager_dispatch_signal_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata);
112 static int manager_dispatch_time_change_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata);
113 static int manager_dispatch_idle_pipe_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata);
114 static int manager_dispatch_user_lookup_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata);
115 static int manager_dispatch_jobs_in_progress(sd_event_source *source, usec_t usec, void *userdata);
116 static int manager_dispatch_run_queue(sd_event_source *source, void *userdata);
117 static int manager_dispatch_sigchld(sd_event_source *source, void *userdata);
118 static int manager_dispatch_timezone_change(sd_event_source *source, const struct inotify_event *event, void *userdata);
119 static int manager_run_environment_generators(Manager *m);
120 static int manager_run_generators(Manager *m);
121 static void manager_vacuum(Manager *m);
122
123 static usec_t manager_watch_jobs_next_time(Manager *m) {
124 usec_t timeout;
125
126 if (MANAGER_IS_USER(m))
127 /* Let the user manager without a timeout show status quickly, so the system manager can make
128 * use of it, if it wants to. */
129 timeout = JOBS_IN_PROGRESS_WAIT_USEC * 2 / 3;
130 else if (show_status_on(m->show_status))
131 /* When status is on, just use the usual timeout. */
132 timeout = JOBS_IN_PROGRESS_WAIT_USEC;
133 else
134 timeout = JOBS_IN_PROGRESS_QUIET_WAIT_USEC;
135
136 return usec_add(now(CLOCK_MONOTONIC), timeout);
137 }
138
139 static void manager_watch_jobs_in_progress(Manager *m) {
140 usec_t next;
141 int r;
142
143 assert(m);
144
145 /* We do not want to show the cylon animation if the user
146 * needs to confirm service executions otherwise confirmation
147 * messages will be screwed by the cylon animation. */
148 if (!manager_is_confirm_spawn_disabled(m))
149 return;
150
151 if (m->jobs_in_progress_event_source)
152 return;
153
154 next = manager_watch_jobs_next_time(m);
155 r = sd_event_add_time(
156 m->event,
157 &m->jobs_in_progress_event_source,
158 CLOCK_MONOTONIC,
159 next, 0,
160 manager_dispatch_jobs_in_progress, m);
161 if (r < 0)
162 return;
163
164 (void) sd_event_source_set_description(m->jobs_in_progress_event_source, "manager-jobs-in-progress");
165 }
166
167 #define CYLON_BUFFER_EXTRA (2*STRLEN(ANSI_RED) + STRLEN(ANSI_HIGHLIGHT_RED) + 2*STRLEN(ANSI_NORMAL))
168
169 static void draw_cylon(char buffer[], size_t buflen, unsigned width, unsigned pos) {
170 char *p = buffer;
171
172 assert(buflen >= CYLON_BUFFER_EXTRA + width + 1);
173 assert(pos <= width+1); /* 0 or width+1 mean that the center light is behind the corner */
174
175 if (pos > 1) {
176 if (pos > 2)
177 p = mempset(p, ' ', pos-2);
178 if (log_get_show_color())
179 p = stpcpy(p, ANSI_RED);
180 *p++ = '*';
181 }
182
183 if (pos > 0 && pos <= width) {
184 if (log_get_show_color())
185 p = stpcpy(p, ANSI_HIGHLIGHT_RED);
186 *p++ = '*';
187 }
188
189 if (log_get_show_color())
190 p = stpcpy(p, ANSI_NORMAL);
191
192 if (pos < width) {
193 if (log_get_show_color())
194 p = stpcpy(p, ANSI_RED);
195 *p++ = '*';
196 if (pos < width-1)
197 p = mempset(p, ' ', width-1-pos);
198 if (log_get_show_color())
199 strcpy(p, ANSI_NORMAL);
200 }
201 }
202
203 static void manager_flip_auto_status(Manager *m, bool enable, const char *reason) {
204 assert(m);
205
206 if (enable) {
207 if (m->show_status == SHOW_STATUS_AUTO)
208 manager_set_show_status(m, SHOW_STATUS_TEMPORARY, reason);
209 } else {
210 if (m->show_status == SHOW_STATUS_TEMPORARY)
211 manager_set_show_status(m, SHOW_STATUS_AUTO, reason);
212 }
213 }
214
215 static void manager_print_jobs_in_progress(Manager *m) {
216 Job *j;
217 unsigned counter = 0, print_nr;
218 char cylon[6 + CYLON_BUFFER_EXTRA + 1];
219 unsigned cylon_pos;
220 uint64_t timeout = 0;
221
222 assert(m);
223 assert(m->n_running_jobs > 0);
224
225 manager_flip_auto_status(m, true, "delay");
226
227 print_nr = (m->jobs_in_progress_iteration / JOBS_IN_PROGRESS_PERIOD_DIVISOR) % m->n_running_jobs;
228
229 HASHMAP_FOREACH(j, m->jobs)
230 if (j->state == JOB_RUNNING && counter++ == print_nr)
231 break;
232
233 /* m->n_running_jobs must be consistent with the contents of m->jobs,
234 * so the above loop must have succeeded in finding j. */
235 assert(counter == print_nr + 1);
236 assert(j);
237
238 cylon_pos = m->jobs_in_progress_iteration % 14;
239 if (cylon_pos >= 8)
240 cylon_pos = 14 - cylon_pos;
241 draw_cylon(cylon, sizeof(cylon), 6, cylon_pos);
242
243 m->jobs_in_progress_iteration++;
244
245 char job_of_n[STRLEN("( of ) ") + DECIMAL_STR_MAX(unsigned)*2] = "";
246 if (m->n_running_jobs > 1)
247 xsprintf(job_of_n, "(%u of %u) ", counter, m->n_running_jobs);
248
249 (void) job_get_timeout(j, &timeout);
250
251 /* We want to use enough information for the user to identify previous lines talking about the same
252 * unit, but keep the message as short as possible. So if 'Starting foo.service' or 'Starting
253 * foo.service - Description' were used, 'foo.service' is enough here. On the other hand, if we used
254 * 'Starting Description' before, then we shall also use 'Description' here. So we pass NULL as the
255 * second argument to unit_status_string(). */
256 const char *ident = unit_status_string(j->unit, NULL);
257
258 const char *time = FORMAT_TIMESPAN(now(CLOCK_MONOTONIC) - j->begin_usec, 1*USEC_PER_SEC);
259 const char *limit = timeout > 0 ? FORMAT_TIMESPAN(timeout - j->begin_usec, 1*USEC_PER_SEC) : "no limit";
260
261 if (m->status_unit_format == STATUS_UNIT_FORMAT_DESCRIPTION)
262 /* When using 'Description', we effectively don't have enough space to show the nested status
263 * without ellipsization, so let's not even try. */
264 manager_status_printf(m, STATUS_TYPE_EPHEMERAL, cylon,
265 "%sA %s job is running for %s (%s / %s)",
266 job_of_n,
267 job_type_to_string(j->type),
268 ident,
269 time, limit);
270 else {
271 const char *status_text = unit_status_text(j->unit);
272
273 manager_status_printf(m, STATUS_TYPE_EPHEMERAL, cylon,
274 "%sJob %s/%s running (%s / %s)%s%s",
275 job_of_n,
276 ident,
277 job_type_to_string(j->type),
278 time, limit,
279 status_text ? ": " : "",
280 strempty(status_text));
281 }
282
283 sd_notifyf(false,
284 "STATUS=%sUser job %s/%s running (%s / %s)...",
285 job_of_n,
286 ident,
287 job_type_to_string(j->type),
288 time, limit);
289 m->status_ready = false;
290 }
291
292 static int have_ask_password(void) {
293 _cleanup_closedir_ DIR *dir = NULL;
294
295 dir = opendir("/run/systemd/ask-password");
296 if (!dir) {
297 if (errno == ENOENT)
298 return false;
299 else
300 return -errno;
301 }
302
303 FOREACH_DIRENT_ALL(de, dir, return -errno)
304 if (startswith(de->d_name, "ask."))
305 return true;
306 return false;
307 }
308
309 static int manager_dispatch_ask_password_fd(sd_event_source *source,
310 int fd, uint32_t revents, void *userdata) {
311 Manager *m = ASSERT_PTR(userdata);
312
313 (void) flush_fd(fd);
314
315 m->have_ask_password = have_ask_password();
316 if (m->have_ask_password < 0)
317 /* Log error but continue. Negative have_ask_password
318 * is treated as unknown status. */
319 log_error_errno(m->have_ask_password, "Failed to list /run/systemd/ask-password: %m");
320
321 return 0;
322 }
323
324 static void manager_close_ask_password(Manager *m) {
325 assert(m);
326
327 m->ask_password_event_source = sd_event_source_disable_unref(m->ask_password_event_source);
328 m->ask_password_inotify_fd = safe_close(m->ask_password_inotify_fd);
329 m->have_ask_password = -EINVAL;
330 }
331
332 static int manager_check_ask_password(Manager *m) {
333 int r;
334
335 assert(m);
336
337 if (!m->ask_password_event_source) {
338 assert(m->ask_password_inotify_fd < 0);
339
340 (void) mkdir_p_label("/run/systemd/ask-password", 0755);
341
342 m->ask_password_inotify_fd = inotify_init1(IN_NONBLOCK|IN_CLOEXEC);
343 if (m->ask_password_inotify_fd < 0)
344 return log_error_errno(errno, "Failed to create inotify object: %m");
345
346 r = inotify_add_watch_and_warn(m->ask_password_inotify_fd,
347 "/run/systemd/ask-password",
348 IN_CREATE|IN_DELETE|IN_MOVE);
349 if (r < 0) {
350 manager_close_ask_password(m);
351 return r;
352 }
353
354 r = sd_event_add_io(m->event, &m->ask_password_event_source,
355 m->ask_password_inotify_fd, EPOLLIN,
356 manager_dispatch_ask_password_fd, m);
357 if (r < 0) {
358 log_error_errno(r, "Failed to add event source for /run/systemd/ask-password: %m");
359 manager_close_ask_password(m);
360 return r;
361 }
362
363 (void) sd_event_source_set_description(m->ask_password_event_source, "manager-ask-password");
364
365 /* Queries might have been added meanwhile... */
366 manager_dispatch_ask_password_fd(m->ask_password_event_source,
367 m->ask_password_inotify_fd, EPOLLIN, m);
368 }
369
370 return m->have_ask_password;
371 }
372
373 static int manager_watch_idle_pipe(Manager *m) {
374 int r;
375
376 assert(m);
377
378 if (m->idle_pipe_event_source)
379 return 0;
380
381 if (m->idle_pipe[2] < 0)
382 return 0;
383
384 r = sd_event_add_io(m->event, &m->idle_pipe_event_source, m->idle_pipe[2], EPOLLIN, manager_dispatch_idle_pipe_fd, m);
385 if (r < 0)
386 return log_error_errno(r, "Failed to watch idle pipe: %m");
387
388 (void) sd_event_source_set_description(m->idle_pipe_event_source, "manager-idle-pipe");
389
390 return 0;
391 }
392
393 static void manager_close_idle_pipe(Manager *m) {
394 assert(m);
395
396 m->idle_pipe_event_source = sd_event_source_disable_unref(m->idle_pipe_event_source);
397
398 safe_close_pair(m->idle_pipe);
399 safe_close_pair(m->idle_pipe + 2);
400 }
401
402 static int manager_setup_time_change(Manager *m) {
403 int r;
404
405 assert(m);
406
407 if (MANAGER_IS_TEST_RUN(m))
408 return 0;
409
410 m->time_change_event_source = sd_event_source_disable_unref(m->time_change_event_source);
411
412 r = event_add_time_change(m->event, &m->time_change_event_source, manager_dispatch_time_change_fd, m);
413 if (r < 0)
414 return log_error_errno(r, "Failed to create time change event source: %m");
415
416 /* Schedule this slightly earlier than the .timer event sources */
417 r = sd_event_source_set_priority(m->time_change_event_source, SD_EVENT_PRIORITY_NORMAL-1);
418 if (r < 0)
419 return log_error_errno(r, "Failed to set priority of time change event sources: %m");
420
421 log_debug("Set up TFD_TIMER_CANCEL_ON_SET timerfd.");
422
423 return 0;
424 }
425
426 static int manager_read_timezone_stat(Manager *m) {
427 struct stat st;
428 bool changed;
429
430 assert(m);
431
432 /* Read the current stat() data of /etc/localtime so that we detect changes */
433 if (lstat("/etc/localtime", &st) < 0) {
434 log_debug_errno(errno, "Failed to stat /etc/localtime, ignoring: %m");
435 changed = m->etc_localtime_accessible;
436 m->etc_localtime_accessible = false;
437 } else {
438 usec_t k;
439
440 k = timespec_load(&st.st_mtim);
441 changed = !m->etc_localtime_accessible || k != m->etc_localtime_mtime;
442
443 m->etc_localtime_mtime = k;
444 m->etc_localtime_accessible = true;
445 }
446
447 return changed;
448 }
449
450 static int manager_setup_timezone_change(Manager *m) {
451 _cleanup_(sd_event_source_unrefp) sd_event_source *new_event = NULL;
452 int r;
453
454 assert(m);
455
456 if (MANAGER_IS_TEST_RUN(m))
457 return 0;
458
459 /* We watch /etc/localtime for three events: change of the link count (which might mean removal from /etc even
460 * though another link might be kept), renames, and file close operations after writing. Note we don't bother
461 * with IN_DELETE_SELF, as that would just report when the inode is removed entirely, i.e. after the link count
462 * went to zero and all fds to it are closed.
463 *
464 * Note that we never follow symlinks here. This is a simplification, but should cover almost all cases
465 * correctly.
466 *
467 * Note that we create the new event source first here, before releasing the old one. This should optimize
468 * behaviour as this way sd-event can reuse the old watch in case the inode didn't change. */
469
470 r = sd_event_add_inotify(m->event, &new_event, "/etc/localtime",
471 IN_ATTRIB|IN_MOVE_SELF|IN_CLOSE_WRITE|IN_DONT_FOLLOW, manager_dispatch_timezone_change, m);
472 if (r == -ENOENT) {
473 /* If the file doesn't exist yet, subscribe to /etc instead, and wait until it is created either by
474 * O_CREATE or by rename() */
475
476 log_debug_errno(r, "/etc/localtime doesn't exist yet, watching /etc instead.");
477 r = sd_event_add_inotify(m->event, &new_event, "/etc",
478 IN_CREATE|IN_MOVED_TO|IN_ONLYDIR, manager_dispatch_timezone_change, m);
479 }
480 if (r < 0)
481 return log_error_errno(r, "Failed to create timezone change event source: %m");
482
483 /* Schedule this slightly earlier than the .timer event sources */
484 r = sd_event_source_set_priority(new_event, SD_EVENT_PRIORITY_NORMAL-1);
485 if (r < 0)
486 return log_error_errno(r, "Failed to set priority of timezone change event sources: %m");
487
488 sd_event_source_unref(m->timezone_change_event_source);
489 m->timezone_change_event_source = TAKE_PTR(new_event);
490
491 return 0;
492 }
493
494 static int enable_special_signals(Manager *m) {
495 _cleanup_close_ int fd = -1;
496
497 assert(m);
498
499 if (MANAGER_IS_TEST_RUN(m))
500 return 0;
501
502 /* Enable that we get SIGINT on control-alt-del. In containers
503 * this will fail with EPERM (older) or EINVAL (newer), so
504 * ignore that. */
505 if (reboot(RB_DISABLE_CAD) < 0 && !IN_SET(errno, EPERM, EINVAL))
506 log_warning_errno(errno, "Failed to enable ctrl-alt-del handling: %m");
507
508 fd = open_terminal("/dev/tty0", O_RDWR|O_NOCTTY|O_CLOEXEC);
509 if (fd < 0) {
510 /* Support systems without virtual console */
511 if (fd != -ENOENT)
512 log_warning_errno(errno, "Failed to open /dev/tty0: %m");
513 } else {
514 /* Enable that we get SIGWINCH on kbrequest */
515 if (ioctl(fd, KDSIGACCEPT, SIGWINCH) < 0)
516 log_warning_errno(errno, "Failed to enable kbrequest handling: %m");
517 }
518
519 return 0;
520 }
521
522 #define RTSIG_IF_AVAILABLE(signum) (signum <= SIGRTMAX ? signum : -1)
523
524 static int manager_setup_signals(Manager *m) {
525 struct sigaction sa = {
526 .sa_handler = SIG_DFL,
527 .sa_flags = SA_NOCLDSTOP|SA_RESTART,
528 };
529 sigset_t mask;
530 int r;
531
532 assert(m);
533
534 assert_se(sigaction(SIGCHLD, &sa, NULL) == 0);
535
536 /* We make liberal use of realtime signals here. On
537 * Linux/glibc we have 30 of them (with the exception of Linux
538 * on hppa, see below), between SIGRTMIN+0 ... SIGRTMIN+30
539 * (aka SIGRTMAX). */
540
541 assert_se(sigemptyset(&mask) == 0);
542 sigset_add_many(&mask,
543 SIGCHLD, /* Child died */
544 SIGTERM, /* Reexecute daemon */
545 SIGHUP, /* Reload configuration */
546 SIGUSR1, /* systemd: reconnect to D-Bus */
547 SIGUSR2, /* systemd: dump status */
548 SIGINT, /* Kernel sends us this on control-alt-del */
549 SIGWINCH, /* Kernel sends us this on kbrequest (alt-arrowup) */
550 SIGPWR, /* Some kernel drivers and upsd send us this on power failure */
551
552 SIGRTMIN+0, /* systemd: start default.target */
553 SIGRTMIN+1, /* systemd: isolate rescue.target */
554 SIGRTMIN+2, /* systemd: isolate emergency.target */
555 SIGRTMIN+3, /* systemd: start halt.target */
556 SIGRTMIN+4, /* systemd: start poweroff.target */
557 SIGRTMIN+5, /* systemd: start reboot.target */
558 SIGRTMIN+6, /* systemd: start kexec.target */
559
560 /* ... space for more special targets ... */
561
562 SIGRTMIN+13, /* systemd: Immediate halt */
563 SIGRTMIN+14, /* systemd: Immediate poweroff */
564 SIGRTMIN+15, /* systemd: Immediate reboot */
565 SIGRTMIN+16, /* systemd: Immediate kexec */
566
567 /* ... space for more immediate system state changes ... */
568
569 SIGRTMIN+20, /* systemd: enable status messages */
570 SIGRTMIN+21, /* systemd: disable status messages */
571 SIGRTMIN+22, /* systemd: set log level to LOG_DEBUG */
572 SIGRTMIN+23, /* systemd: set log level to LOG_INFO */
573 SIGRTMIN+24, /* systemd: Immediate exit (--user only) */
574 SIGRTMIN+25, /* systemd: reexecute manager */
575
576 /* Apparently Linux on hppa had fewer RT signals until v3.18,
577 * SIGRTMAX was SIGRTMIN+25, and then SIGRTMIN was lowered,
578 * see commit v3.17-7614-g1f25df2eff.
579 *
580 * We cannot unconditionally make use of those signals here,
581 * so let's use a runtime check. Since these commands are
582 * accessible by different means and only really a safety
583 * net, the missing functionality on hppa shouldn't matter.
584 */
585
586 RTSIG_IF_AVAILABLE(SIGRTMIN+26), /* systemd: set log target to journal-or-kmsg */
587 RTSIG_IF_AVAILABLE(SIGRTMIN+27), /* systemd: set log target to console */
588 RTSIG_IF_AVAILABLE(SIGRTMIN+28), /* systemd: set log target to kmsg */
589 RTSIG_IF_AVAILABLE(SIGRTMIN+29), /* systemd: set log target to syslog-or-kmsg (obsolete) */
590
591 /* ... one free signal here SIGRTMIN+30 ... */
592 -1);
593 assert_se(sigprocmask(SIG_SETMASK, &mask, NULL) == 0);
594
595 m->signal_fd = signalfd(-1, &mask, SFD_NONBLOCK|SFD_CLOEXEC);
596 if (m->signal_fd < 0)
597 return -errno;
598
599 r = sd_event_add_io(m->event, &m->signal_event_source, m->signal_fd, EPOLLIN, manager_dispatch_signal_fd, m);
600 if (r < 0)
601 return r;
602
603 (void) sd_event_source_set_description(m->signal_event_source, "manager-signal");
604
605 /* Process signals a bit earlier than the rest of things, but later than notify_fd processing, so that the
606 * notify processing can still figure out to which process/service a message belongs, before we reap the
607 * process. Also, process this before handling cgroup notifications, so that we always collect child exit
608 * status information before detecting that there's no process in a cgroup. */
609 r = sd_event_source_set_priority(m->signal_event_source, SD_EVENT_PRIORITY_NORMAL-6);
610 if (r < 0)
611 return r;
612
613 if (MANAGER_IS_SYSTEM(m))
614 return enable_special_signals(m);
615
616 return 0;
617 }
618
619 static char** sanitize_environment(char **l) {
620
621 /* Let's remove some environment variables that we need ourselves to communicate with our clients */
622 strv_env_unset_many(
623 l,
624 "CACHE_DIRECTORY",
625 "CONFIGURATION_DIRECTORY",
626 "CREDENTIALS_DIRECTORY",
627 "EXIT_CODE",
628 "EXIT_STATUS",
629 "INVOCATION_ID",
630 "JOURNAL_STREAM",
631 "LISTEN_FDNAMES",
632 "LISTEN_FDS",
633 "LISTEN_PID",
634 "LOGS_DIRECTORY",
635 "MAINPID",
636 "MANAGERPID",
637 "NOTIFY_SOCKET",
638 "PIDFILE",
639 "REMOTE_ADDR",
640 "REMOTE_PORT",
641 "RUNTIME_DIRECTORY",
642 "SERVICE_RESULT",
643 "STATE_DIRECTORY",
644 "WATCHDOG_PID",
645 "WATCHDOG_USEC",
646 NULL);
647
648 /* Let's order the environment alphabetically, just to make it pretty */
649 return strv_sort(l);
650 }
651
652 int manager_default_environment(Manager *m) {
653 int r;
654
655 assert(m);
656
657 m->transient_environment = strv_free(m->transient_environment);
658
659 if (MANAGER_IS_SYSTEM(m)) {
660 /* The system manager always starts with a clean
661 * environment for its children. It does not import
662 * the kernel's or the parents' exported variables.
663 *
664 * The initial passed environment is untouched to keep
665 * /proc/self/environ valid; it is used for tagging
666 * the init process inside containers. */
667 m->transient_environment = strv_new("PATH=" DEFAULT_PATH);
668 if (!m->transient_environment)
669 return log_oom();
670
671 /* Import locale variables LC_*= from configuration */
672 (void) locale_setup(&m->transient_environment);
673 } else {
674 /* The user manager passes its own environment along to its children, except for $PATH. */
675 m->transient_environment = strv_copy(environ);
676 if (!m->transient_environment)
677 return log_oom();
678
679 r = strv_env_replace_strdup(&m->transient_environment, "PATH=" DEFAULT_USER_PATH);
680 if (r < 0)
681 return log_oom();
682 }
683
684 sanitize_environment(m->transient_environment);
685
686 return 0;
687 }
688
689 static int manager_setup_prefix(Manager *m) {
690 struct table_entry {
691 uint64_t type;
692 const char *suffix;
693 };
694
695 static const struct table_entry paths_system[_EXEC_DIRECTORY_TYPE_MAX] = {
696 [EXEC_DIRECTORY_RUNTIME] = { SD_PATH_SYSTEM_RUNTIME, NULL },
697 [EXEC_DIRECTORY_STATE] = { SD_PATH_SYSTEM_STATE_PRIVATE, NULL },
698 [EXEC_DIRECTORY_CACHE] = { SD_PATH_SYSTEM_STATE_CACHE, NULL },
699 [EXEC_DIRECTORY_LOGS] = { SD_PATH_SYSTEM_STATE_LOGS, NULL },
700 [EXEC_DIRECTORY_CONFIGURATION] = { SD_PATH_SYSTEM_CONFIGURATION, NULL },
701 };
702
703 static const struct table_entry paths_user[_EXEC_DIRECTORY_TYPE_MAX] = {
704 [EXEC_DIRECTORY_RUNTIME] = { SD_PATH_USER_RUNTIME, NULL },
705 [EXEC_DIRECTORY_STATE] = { SD_PATH_USER_CONFIGURATION, NULL },
706 [EXEC_DIRECTORY_CACHE] = { SD_PATH_USER_STATE_CACHE, NULL },
707 [EXEC_DIRECTORY_LOGS] = { SD_PATH_USER_CONFIGURATION, "log" },
708 [EXEC_DIRECTORY_CONFIGURATION] = { SD_PATH_USER_CONFIGURATION, NULL },
709 };
710
711 assert(m);
712
713 const struct table_entry *p = MANAGER_IS_SYSTEM(m) ? paths_system : paths_user;
714 int r;
715
716 for (ExecDirectoryType i = 0; i < _EXEC_DIRECTORY_TYPE_MAX; i++) {
717 r = sd_path_lookup(p[i].type, p[i].suffix, &m->prefix[i]);
718 if (r < 0)
719 return log_warning_errno(r, "Failed to lookup %s path: %m",
720 exec_directory_type_to_string(i));
721 }
722
723 return 0;
724 }
725
726 static void manager_free_unit_name_maps(Manager *m) {
727 m->unit_id_map = hashmap_free(m->unit_id_map);
728 m->unit_name_map = hashmap_free(m->unit_name_map);
729 m->unit_path_cache = set_free(m->unit_path_cache);
730 m->unit_cache_timestamp_hash = 0;
731 }
732
733 static int manager_setup_run_queue(Manager *m) {
734 int r;
735
736 assert(m);
737 assert(!m->run_queue_event_source);
738
739 r = sd_event_add_defer(m->event, &m->run_queue_event_source, manager_dispatch_run_queue, m);
740 if (r < 0)
741 return r;
742
743 r = sd_event_source_set_priority(m->run_queue_event_source, SD_EVENT_PRIORITY_IDLE);
744 if (r < 0)
745 return r;
746
747 r = sd_event_source_set_enabled(m->run_queue_event_source, SD_EVENT_OFF);
748 if (r < 0)
749 return r;
750
751 (void) sd_event_source_set_description(m->run_queue_event_source, "manager-run-queue");
752
753 return 0;
754 }
755
756 static int manager_setup_sigchld_event_source(Manager *m) {
757 int r;
758
759 assert(m);
760 assert(!m->sigchld_event_source);
761
762 r = sd_event_add_defer(m->event, &m->sigchld_event_source, manager_dispatch_sigchld, m);
763 if (r < 0)
764 return r;
765
766 r = sd_event_source_set_priority(m->sigchld_event_source, SD_EVENT_PRIORITY_NORMAL-7);
767 if (r < 0)
768 return r;
769
770 r = sd_event_source_set_enabled(m->sigchld_event_source, SD_EVENT_OFF);
771 if (r < 0)
772 return r;
773
774 (void) sd_event_source_set_description(m->sigchld_event_source, "manager-sigchld");
775
776 return 0;
777 }
778
779 static int manager_find_credentials_dirs(Manager *m) {
780 const char *e;
781 int r;
782
783 assert(m);
784
785 r = get_credentials_dir(&e);
786 if (r < 0) {
787 if (r != -ENXIO)
788 log_debug_errno(r, "Failed to determine credentials directory, ignoring: %m");
789 } else {
790 m->received_credentials_directory = strdup(e);
791 if (!m->received_credentials_directory)
792 return -ENOMEM;
793 }
794
795 r = get_encrypted_credentials_dir(&e);
796 if (r < 0) {
797 if (r != -ENXIO)
798 log_debug_errno(r, "Failed to determine encrypted credentials directory, ignoring: %m");
799 } else {
800 m->received_encrypted_credentials_directory = strdup(e);
801 if (!m->received_encrypted_credentials_directory)
802 return -ENOMEM;
803 }
804
805 return 0;
806 }
807
808 void manager_set_switching_root(Manager *m, bool switching_root) {
809 m->switching_root = MANAGER_IS_SYSTEM(m) && switching_root;
810 }
811
812 int manager_new(LookupScope scope, ManagerTestRunFlags test_run_flags, Manager **_m) {
813 _cleanup_(manager_freep) Manager *m = NULL;
814 int r;
815
816 assert(_m);
817 assert(IN_SET(scope, LOOKUP_SCOPE_SYSTEM, LOOKUP_SCOPE_USER));
818
819 m = new(Manager, 1);
820 if (!m)
821 return -ENOMEM;
822
823 *m = (Manager) {
824 .unit_file_scope = scope,
825 .objective = _MANAGER_OBJECTIVE_INVALID,
826
827 .status_unit_format = STATUS_UNIT_FORMAT_DEFAULT,
828
829 .default_timer_accuracy_usec = USEC_PER_MINUTE,
830 .default_memory_accounting = MEMORY_ACCOUNTING_DEFAULT,
831 .default_tasks_accounting = true,
832 .default_tasks_max = TASKS_MAX_UNSET,
833 .default_timeout_start_usec = DEFAULT_TIMEOUT_USEC,
834 .default_timeout_stop_usec = DEFAULT_TIMEOUT_USEC,
835 .default_restart_usec = DEFAULT_RESTART_USEC,
836 .default_device_timeout_usec = DEFAULT_TIMEOUT_USEC,
837
838 .original_log_level = -1,
839 .original_log_target = _LOG_TARGET_INVALID,
840
841 .watchdog_overridden[WATCHDOG_RUNTIME] = USEC_INFINITY,
842 .watchdog_overridden[WATCHDOG_REBOOT] = USEC_INFINITY,
843 .watchdog_overridden[WATCHDOG_KEXEC] = USEC_INFINITY,
844 .watchdog_overridden[WATCHDOG_PRETIMEOUT] = USEC_INFINITY,
845
846 .show_status_overridden = _SHOW_STATUS_INVALID,
847
848 .notify_fd = -1,
849 .cgroups_agent_fd = -1,
850 .signal_fd = -1,
851 .user_lookup_fds = { -1, -1 },
852 .private_listen_fd = -1,
853 .dev_autofs_fd = -1,
854 .cgroup_inotify_fd = -1,
855 .pin_cgroupfs_fd = -1,
856 .ask_password_inotify_fd = -1,
857 .idle_pipe = { -1, -1, -1, -1},
858
859 /* start as id #1, so that we can leave #0 around as "null-like" value */
860 .current_job_id = 1,
861
862 .have_ask_password = -EINVAL, /* we don't know */
863 .first_boot = -1,
864 .test_run_flags = test_run_flags,
865
866 .default_oom_policy = OOM_STOP,
867 };
868
869 #if ENABLE_EFI
870 if (MANAGER_IS_SYSTEM(m) && detect_container() <= 0)
871 boot_timestamps(m->timestamps + MANAGER_TIMESTAMP_USERSPACE,
872 m->timestamps + MANAGER_TIMESTAMP_FIRMWARE,
873 m->timestamps + MANAGER_TIMESTAMP_LOADER);
874 #endif
875
876 /* Prepare log fields we can use for structured logging */
877 if (MANAGER_IS_SYSTEM(m)) {
878 m->unit_log_field = "UNIT=";
879 m->unit_log_format_string = "UNIT=%s";
880
881 m->invocation_log_field = "INVOCATION_ID=";
882 m->invocation_log_format_string = "INVOCATION_ID=%s";
883 } else {
884 m->unit_log_field = "USER_UNIT=";
885 m->unit_log_format_string = "USER_UNIT=%s";
886
887 m->invocation_log_field = "USER_INVOCATION_ID=";
888 m->invocation_log_format_string = "USER_INVOCATION_ID=%s";
889 }
890
891 /* Reboot immediately if the user hits C-A-D more often than 7x per 2s */
892 m->ctrl_alt_del_ratelimit = (RateLimit) { .interval = 2 * USEC_PER_SEC, .burst = 7 };
893
894 r = manager_default_environment(m);
895 if (r < 0)
896 return r;
897
898 r = hashmap_ensure_allocated(&m->units, &string_hash_ops);
899 if (r < 0)
900 return r;
901
902 r = hashmap_ensure_allocated(&m->cgroup_unit, &path_hash_ops);
903 if (r < 0)
904 return r;
905
906 r = hashmap_ensure_allocated(&m->watch_bus, &string_hash_ops);
907 if (r < 0)
908 return r;
909
910 r = prioq_ensure_allocated(&m->run_queue, compare_job_priority);
911 if (r < 0)
912 return r;
913
914 r = manager_setup_prefix(m);
915 if (r < 0)
916 return r;
917
918 r = manager_find_credentials_dirs(m);
919 if (r < 0)
920 return r;
921
922 r = sd_event_default(&m->event);
923 if (r < 0)
924 return r;
925
926 r = manager_setup_run_queue(m);
927 if (r < 0)
928 return r;
929
930 if (FLAGS_SET(test_run_flags, MANAGER_TEST_RUN_MINIMAL)) {
931 m->cgroup_root = strdup("");
932 if (!m->cgroup_root)
933 return -ENOMEM;
934 } else {
935 r = manager_setup_signals(m);
936 if (r < 0)
937 return r;
938
939 r = manager_setup_cgroup(m);
940 if (r < 0)
941 return r;
942
943 r = manager_setup_time_change(m);
944 if (r < 0)
945 return r;
946
947 r = manager_read_timezone_stat(m);
948 if (r < 0)
949 return r;
950
951 (void) manager_setup_timezone_change(m);
952
953 r = manager_setup_sigchld_event_source(m);
954 if (r < 0)
955 return r;
956
957 #if HAVE_LIBBPF
958 if (MANAGER_IS_SYSTEM(m) && lsm_bpf_supported(/* initialize = */ true)) {
959 r = lsm_bpf_setup(m);
960 if (r < 0)
961 log_warning_errno(r, "Failed to setup LSM BPF, ignoring: %m");
962 }
963 #endif
964 }
965
966 if (test_run_flags == 0) {
967 if (MANAGER_IS_SYSTEM(m))
968 r = mkdir_label("/run/systemd/units", 0755);
969 else {
970 _cleanup_free_ char *units_path = NULL;
971 r = xdg_user_runtime_dir(&units_path, "/systemd/units");
972 if (r < 0)
973 return r;
974 r = mkdir_p_label(units_path, 0755);
975 }
976
977 if (r < 0 && r != -EEXIST)
978 return r;
979 }
980
981 m->taint_usr =
982 !in_initrd() &&
983 dir_is_empty("/usr", /* ignore_hidden_or_backup= */ false) > 0;
984
985 /* Note that we do not set up the notify fd here. We do that after deserialization,
986 * since they might have gotten serialized across the reexec. */
987
988 *_m = TAKE_PTR(m);
989
990 return 0;
991 }
992
993 static int manager_setup_notify(Manager *m) {
994 int r;
995
996 if (MANAGER_IS_TEST_RUN(m))
997 return 0;
998
999 if (m->notify_fd < 0) {
1000 _cleanup_close_ int fd = -1;
1001 union sockaddr_union sa;
1002 socklen_t sa_len;
1003
1004 /* First free all secondary fields */
1005 m->notify_socket = mfree(m->notify_socket);
1006 m->notify_event_source = sd_event_source_disable_unref(m->notify_event_source);
1007
1008 fd = socket(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC|SOCK_NONBLOCK, 0);
1009 if (fd < 0)
1010 return log_error_errno(errno, "Failed to allocate notification socket: %m");
1011
1012 fd_increase_rxbuf(fd, NOTIFY_RCVBUF_SIZE);
1013
1014 m->notify_socket = path_join(m->prefix[EXEC_DIRECTORY_RUNTIME], "systemd/notify");
1015 if (!m->notify_socket)
1016 return log_oom();
1017
1018 r = sockaddr_un_set_path(&sa.un, m->notify_socket);
1019 if (r < 0)
1020 return log_error_errno(r, "Notify socket '%s' not valid for AF_UNIX socket address, refusing.",
1021 m->notify_socket);
1022 sa_len = r;
1023
1024 (void) mkdir_parents_label(m->notify_socket, 0755);
1025 (void) sockaddr_un_unlink(&sa.un);
1026
1027 r = mac_selinux_bind(fd, &sa.sa, sa_len);
1028 if (r < 0)
1029 return log_error_errno(r, "bind(%s) failed: %m", m->notify_socket);
1030
1031 r = setsockopt_int(fd, SOL_SOCKET, SO_PASSCRED, true);
1032 if (r < 0)
1033 return log_error_errno(r, "SO_PASSCRED failed: %m");
1034
1035 m->notify_fd = TAKE_FD(fd);
1036
1037 log_debug("Using notification socket %s", m->notify_socket);
1038 }
1039
1040 if (!m->notify_event_source) {
1041 r = sd_event_add_io(m->event, &m->notify_event_source, m->notify_fd, EPOLLIN, manager_dispatch_notify_fd, m);
1042 if (r < 0)
1043 return log_error_errno(r, "Failed to allocate notify event source: %m");
1044
1045 /* Process notification messages a bit earlier than SIGCHLD, so that we can still identify to which
1046 * service an exit message belongs. */
1047 r = sd_event_source_set_priority(m->notify_event_source, SD_EVENT_PRIORITY_NORMAL-8);
1048 if (r < 0)
1049 return log_error_errno(r, "Failed to set priority of notify event source: %m");
1050
1051 (void) sd_event_source_set_description(m->notify_event_source, "manager-notify");
1052 }
1053
1054 return 0;
1055 }
1056
1057 static int manager_setup_cgroups_agent(Manager *m) {
1058
1059 static const union sockaddr_union sa = {
1060 .un.sun_family = AF_UNIX,
1061 .un.sun_path = "/run/systemd/cgroups-agent",
1062 };
1063 int r;
1064
1065 /* This creates a listening socket we receive cgroups agent messages on. We do not use D-Bus for delivering
1066 * these messages from the cgroups agent binary to PID 1, as the cgroups agent binary is very short-living, and
1067 * each instance of it needs a new D-Bus connection. Since D-Bus connections are SOCK_STREAM/AF_UNIX, on
1068 * overloaded systems the backlog of the D-Bus socket becomes relevant, as not more than the configured number
1069 * of D-Bus connections may be queued until the kernel will start dropping further incoming connections,
1070 * possibly resulting in lost cgroups agent messages. To avoid this, we'll use a private SOCK_DGRAM/AF_UNIX
1071 * socket, where no backlog is relevant as communication may take place without an actual connect() cycle, and
1072 * we thus won't lose messages.
1073 *
1074 * Note that PID 1 will forward the agent message to system bus, so that the user systemd instance may listen
1075 * to it. The system instance hence listens on this special socket, but the user instances listen on the system
1076 * bus for these messages. */
1077
1078 if (MANAGER_IS_TEST_RUN(m))
1079 return 0;
1080
1081 if (!MANAGER_IS_SYSTEM(m))
1082 return 0;
1083
1084 r = cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER);
1085 if (r < 0)
1086 return log_error_errno(r, "Failed to determine whether unified cgroups hierarchy is used: %m");
1087 if (r > 0) /* We don't need this anymore on the unified hierarchy */
1088 return 0;
1089
1090 if (m->cgroups_agent_fd < 0) {
1091 _cleanup_close_ int fd = -1;
1092
1093 /* First free all secondary fields */
1094 m->cgroups_agent_event_source = sd_event_source_disable_unref(m->cgroups_agent_event_source);
1095
1096 fd = socket(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC|SOCK_NONBLOCK, 0);
1097 if (fd < 0)
1098 return log_error_errno(errno, "Failed to allocate cgroups agent socket: %m");
1099
1100 fd_increase_rxbuf(fd, CGROUPS_AGENT_RCVBUF_SIZE);
1101
1102 (void) sockaddr_un_unlink(&sa.un);
1103
1104 /* Only allow root to connect to this socket */
1105 RUN_WITH_UMASK(0077)
1106 r = bind(fd, &sa.sa, SOCKADDR_UN_LEN(sa.un));
1107 if (r < 0)
1108 return log_error_errno(errno, "bind(%s) failed: %m", sa.un.sun_path);
1109
1110 m->cgroups_agent_fd = TAKE_FD(fd);
1111 }
1112
1113 if (!m->cgroups_agent_event_source) {
1114 r = sd_event_add_io(m->event, &m->cgroups_agent_event_source, m->cgroups_agent_fd, EPOLLIN, manager_dispatch_cgroups_agent_fd, m);
1115 if (r < 0)
1116 return log_error_errno(r, "Failed to allocate cgroups agent event source: %m");
1117
1118 /* Process cgroups notifications early. Note that when the agent notification is received
1119 * we'll just enqueue the unit in the cgroup empty queue, hence pick a high priority than
1120 * that. Also see handling of cgroup inotify for the unified cgroup stuff. */
1121 r = sd_event_source_set_priority(m->cgroups_agent_event_source, SD_EVENT_PRIORITY_NORMAL-9);
1122 if (r < 0)
1123 return log_error_errno(r, "Failed to set priority of cgroups agent event source: %m");
1124
1125 (void) sd_event_source_set_description(m->cgroups_agent_event_source, "manager-cgroups-agent");
1126 }
1127
1128 return 0;
1129 }
1130
1131 static int manager_setup_user_lookup_fd(Manager *m) {
1132 int r;
1133
1134 assert(m);
1135
1136 /* Set up the socket pair used for passing UID/GID resolution results from forked off processes to PID
1137 * 1. Background: we can't do name lookups (NSS) from PID 1, since it might involve IPC and thus activation,
1138 * and we might hence deadlock on ourselves. Hence we do all user/group lookups asynchronously from the forked
1139 * off processes right before executing the binaries to start. In order to be able to clean up any IPC objects
1140 * created by a unit (see RemoveIPC=) we need to know in PID 1 the used UID/GID of the executed processes,
1141 * hence we establish this communication channel so that forked off processes can pass their UID/GID
1142 * information back to PID 1. The forked off processes send their resolved UID/GID to PID 1 in a simple
1143 * datagram, along with their unit name, so that we can share one communication socket pair among all units for
1144 * this purpose.
1145 *
1146 * You might wonder why we need a communication channel for this that is independent of the usual notification
1147 * socket scheme (i.e. $NOTIFY_SOCKET). The primary difference is about trust: data sent via the $NOTIFY_SOCKET
1148 * channel is only accepted if it originates from the right unit and if reception was enabled for it. The user
1149 * lookup socket OTOH is only accessible by PID 1 and its children until they exec(), and always available.
1150 *
1151 * Note that this function is called under two circumstances: when we first initialize (in which case we
1152 * allocate both the socket pair and the event source to listen on it), and when we deserialize after a reload
1153 * (in which case the socket pair already exists but we still need to allocate the event source for it). */
1154
1155 if (m->user_lookup_fds[0] < 0) {
1156
1157 /* Free all secondary fields */
1158 safe_close_pair(m->user_lookup_fds);
1159 m->user_lookup_event_source = sd_event_source_disable_unref(m->user_lookup_event_source);
1160
1161 if (socketpair(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC, 0, m->user_lookup_fds) < 0)
1162 return log_error_errno(errno, "Failed to allocate user lookup socket: %m");
1163
1164 (void) fd_increase_rxbuf(m->user_lookup_fds[0], NOTIFY_RCVBUF_SIZE);
1165 }
1166
1167 if (!m->user_lookup_event_source) {
1168 r = sd_event_add_io(m->event, &m->user_lookup_event_source, m->user_lookup_fds[0], EPOLLIN, manager_dispatch_user_lookup_fd, m);
1169 if (r < 0)
1170 return log_error_errno(errno, "Failed to allocate user lookup event source: %m");
1171
1172 /* Process even earlier than the notify event source, so that we always know first about valid UID/GID
1173 * resolutions */
1174 r = sd_event_source_set_priority(m->user_lookup_event_source, SD_EVENT_PRIORITY_NORMAL-11);
1175 if (r < 0)
1176 return log_error_errno(errno, "Failed to set priority of user lookup event source: %m");
1177
1178 (void) sd_event_source_set_description(m->user_lookup_event_source, "user-lookup");
1179 }
1180
1181 return 0;
1182 }
1183
1184 static unsigned manager_dispatch_cleanup_queue(Manager *m) {
1185 Unit *u;
1186 unsigned n = 0;
1187
1188 assert(m);
1189
1190 while ((u = m->cleanup_queue)) {
1191 assert(u->in_cleanup_queue);
1192
1193 unit_free(u);
1194 n++;
1195 }
1196
1197 return n;
1198 }
1199
1200 enum {
1201 GC_OFFSET_IN_PATH, /* This one is on the path we were traveling */
1202 GC_OFFSET_UNSURE, /* No clue */
1203 GC_OFFSET_GOOD, /* We still need this unit */
1204 GC_OFFSET_BAD, /* We don't need this unit anymore */
1205 _GC_OFFSET_MAX
1206 };
1207
1208 static void unit_gc_mark_good(Unit *u, unsigned gc_marker) {
1209 Unit *other;
1210
1211 u->gc_marker = gc_marker + GC_OFFSET_GOOD;
1212
1213 /* Recursively mark referenced units as GOOD as well */
1214 UNIT_FOREACH_DEPENDENCY(other, u, UNIT_ATOM_REFERENCES)
1215 if (other->gc_marker == gc_marker + GC_OFFSET_UNSURE)
1216 unit_gc_mark_good(other, gc_marker);
1217 }
1218
1219 static void unit_gc_sweep(Unit *u, unsigned gc_marker) {
1220 Unit *other;
1221 bool is_bad;
1222
1223 assert(u);
1224
1225 if (IN_SET(u->gc_marker - gc_marker,
1226 GC_OFFSET_GOOD, GC_OFFSET_BAD, GC_OFFSET_UNSURE, GC_OFFSET_IN_PATH))
1227 return;
1228
1229 if (u->in_cleanup_queue)
1230 goto bad;
1231
1232 if (!unit_may_gc(u))
1233 goto good;
1234
1235 u->gc_marker = gc_marker + GC_OFFSET_IN_PATH;
1236
1237 is_bad = true;
1238
1239 UNIT_FOREACH_DEPENDENCY(other, u, UNIT_ATOM_REFERENCED_BY) {
1240 unit_gc_sweep(other, gc_marker);
1241
1242 if (other->gc_marker == gc_marker + GC_OFFSET_GOOD)
1243 goto good;
1244
1245 if (other->gc_marker != gc_marker + GC_OFFSET_BAD)
1246 is_bad = false;
1247 }
1248
1249 LIST_FOREACH(refs_by_target, ref, u->refs_by_target) {
1250 unit_gc_sweep(ref->source, gc_marker);
1251
1252 if (ref->source->gc_marker == gc_marker + GC_OFFSET_GOOD)
1253 goto good;
1254
1255 if (ref->source->gc_marker != gc_marker + GC_OFFSET_BAD)
1256 is_bad = false;
1257 }
1258
1259 if (is_bad)
1260 goto bad;
1261
1262 /* We were unable to find anything out about this entry, so
1263 * let's investigate it later */
1264 u->gc_marker = gc_marker + GC_OFFSET_UNSURE;
1265 unit_add_to_gc_queue(u);
1266 return;
1267
1268 bad:
1269 /* We definitely know that this one is not useful anymore, so
1270 * let's mark it for deletion */
1271 u->gc_marker = gc_marker + GC_OFFSET_BAD;
1272 unit_add_to_cleanup_queue(u);
1273 return;
1274
1275 good:
1276 unit_gc_mark_good(u, gc_marker);
1277 }
1278
1279 static unsigned manager_dispatch_gc_unit_queue(Manager *m) {
1280 unsigned n = 0, gc_marker;
1281 Unit *u;
1282
1283 assert(m);
1284
1285 /* log_debug("Running GC..."); */
1286
1287 m->gc_marker += _GC_OFFSET_MAX;
1288 if (m->gc_marker + _GC_OFFSET_MAX <= _GC_OFFSET_MAX)
1289 m->gc_marker = 1;
1290
1291 gc_marker = m->gc_marker;
1292
1293 while ((u = m->gc_unit_queue)) {
1294 assert(u->in_gc_queue);
1295
1296 unit_gc_sweep(u, gc_marker);
1297
1298 LIST_REMOVE(gc_queue, m->gc_unit_queue, u);
1299 u->in_gc_queue = false;
1300
1301 n++;
1302
1303 if (IN_SET(u->gc_marker - gc_marker,
1304 GC_OFFSET_BAD, GC_OFFSET_UNSURE)) {
1305 if (u->id)
1306 log_unit_debug(u, "Collecting.");
1307 u->gc_marker = gc_marker + GC_OFFSET_BAD;
1308 unit_add_to_cleanup_queue(u);
1309 }
1310 }
1311
1312 return n;
1313 }
1314
1315 static unsigned manager_dispatch_gc_job_queue(Manager *m) {
1316 unsigned n = 0;
1317 Job *j;
1318
1319 assert(m);
1320
1321 while ((j = m->gc_job_queue)) {
1322 assert(j->in_gc_queue);
1323
1324 LIST_REMOVE(gc_queue, m->gc_job_queue, j);
1325 j->in_gc_queue = false;
1326
1327 n++;
1328
1329 if (!job_may_gc(j))
1330 continue;
1331
1332 log_unit_debug(j->unit, "Collecting job.");
1333 (void) job_finish_and_invalidate(j, JOB_COLLECTED, false, false);
1334 }
1335
1336 return n;
1337 }
1338
1339 static unsigned manager_dispatch_stop_when_unneeded_queue(Manager *m) {
1340 unsigned n = 0;
1341 Unit *u;
1342 int r;
1343
1344 assert(m);
1345
1346 while ((u = m->stop_when_unneeded_queue)) {
1347 _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL;
1348
1349 assert(u->in_stop_when_unneeded_queue);
1350 LIST_REMOVE(stop_when_unneeded_queue, m->stop_when_unneeded_queue, u);
1351 u->in_stop_when_unneeded_queue = false;
1352
1353 n++;
1354
1355 if (!unit_is_unneeded(u))
1356 continue;
1357
1358 log_unit_debug(u, "Unit is not needed anymore.");
1359
1360 /* If stopping a unit fails continuously we might enter a stop loop here, hence stop acting on the
1361 * service being unnecessary after a while. */
1362
1363 if (!ratelimit_below(&u->auto_start_stop_ratelimit)) {
1364 log_unit_warning(u, "Unit not needed anymore, but not stopping since we tried this too often recently.");
1365 continue;
1366 }
1367
1368 /* Ok, nobody needs us anymore. Sniff. Then let's commit suicide */
1369 r = manager_add_job(u->manager, JOB_STOP, u, JOB_FAIL, NULL, &error, NULL);
1370 if (r < 0)
1371 log_unit_warning_errno(u, r, "Failed to enqueue stop job, ignoring: %s", bus_error_message(&error, r));
1372 }
1373
1374 return n;
1375 }
1376
1377 static unsigned manager_dispatch_start_when_upheld_queue(Manager *m) {
1378 unsigned n = 0;
1379 Unit *u;
1380 int r;
1381
1382 assert(m);
1383
1384 while ((u = m->start_when_upheld_queue)) {
1385 _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL;
1386 Unit *culprit = NULL;
1387
1388 assert(u->in_start_when_upheld_queue);
1389 LIST_REMOVE(start_when_upheld_queue, m->start_when_upheld_queue, u);
1390 u->in_start_when_upheld_queue = false;
1391
1392 n++;
1393
1394 if (!unit_is_upheld_by_active(u, &culprit))
1395 continue;
1396
1397 log_unit_debug(u, "Unit is started because upheld by active unit %s.", culprit->id);
1398
1399 /* If stopping a unit fails continuously we might enter a stop loop here, hence stop acting on the
1400 * service being unnecessary after a while. */
1401
1402 if (!ratelimit_below(&u->auto_start_stop_ratelimit)) {
1403 log_unit_warning(u, "Unit needs to be started because active unit %s upholds it, but not starting since we tried this too often recently.", culprit->id);
1404 continue;
1405 }
1406
1407 r = manager_add_job(u->manager, JOB_START, u, JOB_FAIL, NULL, &error, NULL);
1408 if (r < 0)
1409 log_unit_warning_errno(u, r, "Failed to enqueue start job, ignoring: %s", bus_error_message(&error, r));
1410 }
1411
1412 return n;
1413 }
1414
1415 static unsigned manager_dispatch_stop_when_bound_queue(Manager *m) {
1416 unsigned n = 0;
1417 Unit *u;
1418 int r;
1419
1420 assert(m);
1421
1422 while ((u = m->stop_when_bound_queue)) {
1423 _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL;
1424 Unit *culprit = NULL;
1425
1426 assert(u->in_stop_when_bound_queue);
1427 LIST_REMOVE(stop_when_bound_queue, m->stop_when_bound_queue, u);
1428 u->in_stop_when_bound_queue = false;
1429
1430 n++;
1431
1432 if (!unit_is_bound_by_inactive(u, &culprit))
1433 continue;
1434
1435 log_unit_debug(u, "Unit is stopped because bound to inactive unit %s.", culprit->id);
1436
1437 /* If stopping a unit fails continuously we might enter a stop loop here, hence stop acting on the
1438 * service being unnecessary after a while. */
1439
1440 if (!ratelimit_below(&u->auto_start_stop_ratelimit)) {
1441 log_unit_warning(u, "Unit needs to be stopped because it is bound to inactive unit %s it, but not stopping since we tried this too often recently.", culprit->id);
1442 continue;
1443 }
1444
1445 r = manager_add_job(u->manager, JOB_STOP, u, JOB_REPLACE, NULL, &error, NULL);
1446 if (r < 0)
1447 log_unit_warning_errno(u, r, "Failed to enqueue stop job, ignoring: %s", bus_error_message(&error, r));
1448 }
1449
1450 return n;
1451 }
1452
1453 static void manager_clear_jobs_and_units(Manager *m) {
1454 Unit *u;
1455
1456 assert(m);
1457
1458 while ((u = hashmap_first(m->units)))
1459 unit_free(u);
1460
1461 manager_dispatch_cleanup_queue(m);
1462
1463 assert(!m->load_queue);
1464 assert(prioq_isempty(m->run_queue));
1465 assert(!m->dbus_unit_queue);
1466 assert(!m->dbus_job_queue);
1467 assert(!m->cleanup_queue);
1468 assert(!m->gc_unit_queue);
1469 assert(!m->gc_job_queue);
1470 assert(!m->cgroup_realize_queue);
1471 assert(!m->cgroup_empty_queue);
1472 assert(!m->cgroup_oom_queue);
1473 assert(!m->target_deps_queue);
1474 assert(!m->stop_when_unneeded_queue);
1475 assert(!m->start_when_upheld_queue);
1476 assert(!m->stop_when_bound_queue);
1477
1478 assert(hashmap_isempty(m->jobs));
1479 assert(hashmap_isempty(m->units));
1480
1481 m->n_on_console = 0;
1482 m->n_running_jobs = 0;
1483 m->n_installed_jobs = 0;
1484 m->n_failed_jobs = 0;
1485 }
1486
1487 Manager* manager_free(Manager *m) {
1488 if (!m)
1489 return NULL;
1490
1491 manager_clear_jobs_and_units(m);
1492
1493 for (UnitType c = 0; c < _UNIT_TYPE_MAX; c++)
1494 if (unit_vtable[c]->shutdown)
1495 unit_vtable[c]->shutdown(m);
1496
1497 /* Keep the cgroup hierarchy in place except when we know we are going down for good */
1498 manager_shutdown_cgroup(m, IN_SET(m->objective, MANAGER_EXIT, MANAGER_REBOOT, MANAGER_POWEROFF, MANAGER_HALT, MANAGER_KEXEC));
1499
1500 lookup_paths_flush_generator(&m->lookup_paths);
1501
1502 bus_done(m);
1503 manager_varlink_done(m);
1504
1505 exec_runtime_vacuum(m);
1506 hashmap_free(m->exec_runtime_by_id);
1507
1508 dynamic_user_vacuum(m, false);
1509 hashmap_free(m->dynamic_users);
1510
1511 hashmap_free(m->units);
1512 hashmap_free(m->units_by_invocation_id);
1513 hashmap_free(m->jobs);
1514 hashmap_free(m->watch_pids);
1515 hashmap_free(m->watch_bus);
1516
1517 prioq_free(m->run_queue);
1518
1519 set_free(m->startup_units);
1520 set_free(m->failed_units);
1521
1522 sd_event_source_unref(m->signal_event_source);
1523 sd_event_source_unref(m->sigchld_event_source);
1524 sd_event_source_unref(m->notify_event_source);
1525 sd_event_source_unref(m->cgroups_agent_event_source);
1526 sd_event_source_unref(m->time_change_event_source);
1527 sd_event_source_unref(m->timezone_change_event_source);
1528 sd_event_source_unref(m->jobs_in_progress_event_source);
1529 sd_event_source_unref(m->run_queue_event_source);
1530 sd_event_source_unref(m->user_lookup_event_source);
1531
1532 safe_close(m->signal_fd);
1533 safe_close(m->notify_fd);
1534 safe_close(m->cgroups_agent_fd);
1535 safe_close_pair(m->user_lookup_fds);
1536
1537 manager_close_ask_password(m);
1538
1539 manager_close_idle_pipe(m);
1540
1541 sd_event_unref(m->event);
1542
1543 free(m->notify_socket);
1544
1545 lookup_paths_free(&m->lookup_paths);
1546 strv_free(m->transient_environment);
1547 strv_free(m->client_environment);
1548
1549 hashmap_free(m->cgroup_unit);
1550 manager_free_unit_name_maps(m);
1551
1552 free(m->switch_root);
1553 free(m->switch_root_init);
1554
1555 free(m->default_smack_process_label);
1556
1557 rlimit_free_all(m->rlimit);
1558
1559 assert(hashmap_isempty(m->units_requiring_mounts_for));
1560 hashmap_free(m->units_requiring_mounts_for);
1561
1562 hashmap_free(m->uid_refs);
1563 hashmap_free(m->gid_refs);
1564
1565 for (ExecDirectoryType dt = 0; dt < _EXEC_DIRECTORY_TYPE_MAX; dt++)
1566 m->prefix[dt] = mfree(m->prefix[dt]);
1567 free(m->received_credentials_directory);
1568 free(m->received_encrypted_credentials_directory);
1569
1570 free(m->watchdog_pretimeout_governor);
1571 free(m->watchdog_pretimeout_governor_overridden);
1572
1573 #if BPF_FRAMEWORK
1574 lsm_bpf_destroy(m->restrict_fs);
1575 #endif
1576
1577 return mfree(m);
1578 }
1579
1580 static void manager_enumerate_perpetual(Manager *m) {
1581 assert(m);
1582
1583 if (FLAGS_SET(m->test_run_flags, MANAGER_TEST_RUN_MINIMAL))
1584 return;
1585
1586 /* Let's ask every type to load all units from disk/kernel that it might know */
1587 for (UnitType c = 0; c < _UNIT_TYPE_MAX; c++) {
1588 if (!unit_type_supported(c)) {
1589 log_debug("Unit type .%s is not supported on this system.", unit_type_to_string(c));
1590 continue;
1591 }
1592
1593 if (unit_vtable[c]->enumerate_perpetual)
1594 unit_vtable[c]->enumerate_perpetual(m);
1595 }
1596 }
1597
1598 static void manager_enumerate(Manager *m) {
1599 assert(m);
1600
1601 if (FLAGS_SET(m->test_run_flags, MANAGER_TEST_RUN_MINIMAL))
1602 return;
1603
1604 /* Let's ask every type to load all units from disk/kernel that it might know */
1605 for (UnitType c = 0; c < _UNIT_TYPE_MAX; c++) {
1606 if (!unit_type_supported(c)) {
1607 log_debug("Unit type .%s is not supported on this system.", unit_type_to_string(c));
1608 continue;
1609 }
1610
1611 if (unit_vtable[c]->enumerate)
1612 unit_vtable[c]->enumerate(m);
1613 }
1614
1615 manager_dispatch_load_queue(m);
1616 }
1617
1618 static void manager_coldplug(Manager *m) {
1619 Unit *u;
1620 char *k;
1621 int r;
1622
1623 assert(m);
1624
1625 log_debug("Invoking unit coldplug() handlers%s", special_glyph(SPECIAL_GLYPH_ELLIPSIS));
1626
1627 /* Let's place the units back into their deserialized state */
1628 HASHMAP_FOREACH_KEY(u, k, m->units) {
1629
1630 /* ignore aliases */
1631 if (u->id != k)
1632 continue;
1633
1634 r = unit_coldplug(u);
1635 if (r < 0)
1636 log_warning_errno(r, "We couldn't coldplug %s, proceeding anyway: %m", u->id);
1637 }
1638 }
1639
1640 static void manager_catchup(Manager *m) {
1641 Unit *u;
1642 char *k;
1643
1644 assert(m);
1645
1646 log_debug("Invoking unit catchup() handlers%s", special_glyph(SPECIAL_GLYPH_ELLIPSIS));
1647
1648 /* Let's catch up on any state changes that happened while we were reloading/reexecing */
1649 HASHMAP_FOREACH_KEY(u, k, m->units) {
1650
1651 /* ignore aliases */
1652 if (u->id != k)
1653 continue;
1654
1655 unit_catchup(u);
1656 }
1657 }
1658
1659 static void manager_distribute_fds(Manager *m, FDSet *fds) {
1660 Unit *u;
1661
1662 assert(m);
1663
1664 HASHMAP_FOREACH(u, m->units) {
1665
1666 if (fdset_size(fds) <= 0)
1667 break;
1668
1669 if (!UNIT_VTABLE(u)->distribute_fds)
1670 continue;
1671
1672 UNIT_VTABLE(u)->distribute_fds(u, fds);
1673 }
1674 }
1675
1676 static bool manager_dbus_is_running(Manager *m, bool deserialized) {
1677 Unit *u;
1678
1679 assert(m);
1680
1681 /* This checks whether the dbus instance we are supposed to expose our APIs on is up. We check both the socket
1682 * and the service unit. If the 'deserialized' parameter is true we'll check the deserialized state of the unit
1683 * rather than the current one. */
1684
1685 if (MANAGER_IS_TEST_RUN(m))
1686 return false;
1687
1688 u = manager_get_unit(m, SPECIAL_DBUS_SOCKET);
1689 if (!u)
1690 return false;
1691 if ((deserialized ? SOCKET(u)->deserialized_state : SOCKET(u)->state) != SOCKET_RUNNING)
1692 return false;
1693
1694 u = manager_get_unit(m, SPECIAL_DBUS_SERVICE);
1695 if (!u)
1696 return false;
1697 if (!IN_SET((deserialized ? SERVICE(u)->deserialized_state : SERVICE(u)->state), SERVICE_RUNNING, SERVICE_RELOAD))
1698 return false;
1699
1700 return true;
1701 }
1702
1703 static void manager_setup_bus(Manager *m) {
1704 assert(m);
1705
1706 /* Let's set up our private bus connection now, unconditionally */
1707 (void) bus_init_private(m);
1708
1709 /* If we are in --user mode also connect to the system bus now */
1710 if (MANAGER_IS_USER(m))
1711 (void) bus_init_system(m);
1712
1713 /* Let's connect to the bus now, but only if the unit is supposed to be up */
1714 if (manager_dbus_is_running(m, MANAGER_IS_RELOADING(m))) {
1715 (void) bus_init_api(m);
1716
1717 if (MANAGER_IS_SYSTEM(m))
1718 (void) bus_init_system(m);
1719 }
1720 }
1721
1722 static void manager_preset_all(Manager *m) {
1723 int r;
1724
1725 assert(m);
1726
1727 if (m->first_boot <= 0)
1728 return;
1729
1730 if (!MANAGER_IS_SYSTEM(m))
1731 return;
1732
1733 if (MANAGER_IS_TEST_RUN(m))
1734 return;
1735
1736 /* If this is the first boot, and we are in the host system, then preset everything */
1737 UnitFilePresetMode mode = FIRST_BOOT_FULL_PRESET ? UNIT_FILE_PRESET_FULL : UNIT_FILE_PRESET_ENABLE_ONLY;
1738
1739 r = unit_file_preset_all(LOOKUP_SCOPE_SYSTEM, 0, NULL, mode, NULL, 0);
1740 if (r < 0)
1741 log_full_errno(r == -EEXIST ? LOG_NOTICE : LOG_WARNING, r,
1742 "Failed to populate /etc with preset unit settings, ignoring: %m");
1743 else
1744 log_info("Populated /etc with preset unit settings.");
1745 }
1746
1747 static void manager_ready(Manager *m) {
1748 assert(m);
1749
1750 /* After having loaded everything, do the final round of catching up with what might have changed */
1751
1752 m->objective = MANAGER_OK; /* Tell everyone we are up now */
1753
1754 /* It might be safe to log to the journal now and connect to dbus */
1755 manager_recheck_journal(m);
1756 manager_recheck_dbus(m);
1757
1758 /* Let's finally catch up with any changes that took place while we were reloading/reexecing */
1759 manager_catchup(m);
1760
1761 /* Create a file which will indicate when the manager started loading units the last time. */
1762 if (MANAGER_IS_SYSTEM(m))
1763 (void) touch_file("/run/systemd/systemd-units-load", false,
1764 m->timestamps[MANAGER_TIMESTAMP_UNITS_LOAD].realtime ?: now(CLOCK_REALTIME),
1765 UID_INVALID, GID_INVALID, 0444);
1766 }
1767
1768 Manager* manager_reloading_start(Manager *m) {
1769 m->n_reloading++;
1770 dual_timestamp_get(m->timestamps + MANAGER_TIMESTAMP_UNITS_LOAD);
1771 return m;
1772 }
1773
1774 void manager_reloading_stopp(Manager **m) {
1775 if (*m) {
1776 assert((*m)->n_reloading > 0);
1777 (*m)->n_reloading--;
1778 }
1779 }
1780
1781 int manager_startup(Manager *m, FILE *serialization, FDSet *fds, const char *root) {
1782 int r;
1783
1784 assert(m);
1785
1786 /* If we are running in test mode, we still want to run the generators,
1787 * but we should not touch the real generator directories. */
1788 r = lookup_paths_init_or_warn(&m->lookup_paths, m->unit_file_scope,
1789 MANAGER_IS_TEST_RUN(m) ? LOOKUP_PATHS_TEMPORARY_GENERATED : 0,
1790 root);
1791 if (r < 0)
1792 return r;
1793
1794 dual_timestamp_get(m->timestamps + manager_timestamp_initrd_mangle(MANAGER_TIMESTAMP_GENERATORS_START));
1795 r = manager_run_environment_generators(m);
1796 if (r >= 0)
1797 r = manager_run_generators(m);
1798 dual_timestamp_get(m->timestamps + manager_timestamp_initrd_mangle(MANAGER_TIMESTAMP_GENERATORS_FINISH));
1799 if (r < 0)
1800 return r;
1801
1802 manager_preset_all(m);
1803
1804 lookup_paths_log(&m->lookup_paths);
1805
1806 {
1807 /* This block is (optionally) done with the reloading counter bumped */
1808 _unused_ _cleanup_(manager_reloading_stopp) Manager *reloading = NULL;
1809
1810 /* If we will deserialize make sure that during enumeration this is already known, so we increase the
1811 * counter here already */
1812 if (serialization)
1813 reloading = manager_reloading_start(m);
1814
1815 /* First, enumerate what we can from all config files */
1816 dual_timestamp_get(m->timestamps + manager_timestamp_initrd_mangle(MANAGER_TIMESTAMP_UNITS_LOAD_START));
1817 manager_enumerate_perpetual(m);
1818 manager_enumerate(m);
1819 dual_timestamp_get(m->timestamps + manager_timestamp_initrd_mangle(MANAGER_TIMESTAMP_UNITS_LOAD_FINISH));
1820
1821 /* Second, deserialize if there is something to deserialize */
1822 if (serialization) {
1823 r = manager_deserialize(m, serialization, fds);
1824 if (r < 0)
1825 return log_error_errno(r, "Deserialization failed: %m");
1826 }
1827
1828 /* Any fds left? Find some unit which wants them. This is useful to allow container managers to pass
1829 * some file descriptors to us pre-initialized. This enables socket-based activation of entire
1830 * containers. */
1831 manager_distribute_fds(m, fds);
1832
1833 /* We might have deserialized the notify fd, but if we didn't then let's create the bus now */
1834 r = manager_setup_notify(m);
1835 if (r < 0)
1836 /* No sense to continue without notifications, our children would fail anyway. */
1837 return r;
1838
1839 r = manager_setup_cgroups_agent(m);
1840 if (r < 0)
1841 /* Likewise, no sense to continue without empty cgroup notifications. */
1842 return r;
1843
1844 r = manager_setup_user_lookup_fd(m);
1845 if (r < 0)
1846 /* This shouldn't fail, except if things are really broken. */
1847 return r;
1848
1849 /* Connect to the bus if we are good for it */
1850 manager_setup_bus(m);
1851
1852 /* Now that we are connected to all possible buses, let's deserialize who is tracking us. */
1853 r = bus_track_coldplug(m, &m->subscribed, false, m->deserialized_subscribed);
1854 if (r < 0)
1855 log_warning_errno(r, "Failed to deserialized tracked clients, ignoring: %m");
1856 m->deserialized_subscribed = strv_free(m->deserialized_subscribed);
1857
1858 r = manager_varlink_init(m);
1859 if (r < 0)
1860 log_warning_errno(r, "Failed to set up Varlink, ignoring: %m");
1861
1862 /* Third, fire things up! */
1863 manager_coldplug(m);
1864
1865 /* Clean up runtime objects */
1866 manager_vacuum(m);
1867
1868 if (serialization)
1869 /* Let's wait for the UnitNew/JobNew messages being sent, before we notify that the
1870 * reload is finished */
1871 m->send_reloading_done = true;
1872 }
1873
1874 manager_ready(m);
1875
1876 manager_set_switching_root(m, false);
1877
1878 return 0;
1879 }
1880
1881 int manager_add_job(
1882 Manager *m,
1883 JobType type,
1884 Unit *unit,
1885 JobMode mode,
1886 Set *affected_jobs,
1887 sd_bus_error *error,
1888 Job **ret) {
1889
1890 Transaction *tr;
1891 int r;
1892
1893 assert(m);
1894 assert(type < _JOB_TYPE_MAX);
1895 assert(unit);
1896 assert(mode < _JOB_MODE_MAX);
1897
1898 if (mode == JOB_ISOLATE && type != JOB_START)
1899 return sd_bus_error_set(error, SD_BUS_ERROR_INVALID_ARGS, "Isolate is only valid for start.");
1900
1901 if (mode == JOB_ISOLATE && !unit->allow_isolate)
1902 return sd_bus_error_set(error, BUS_ERROR_NO_ISOLATION, "Operation refused, unit may not be isolated.");
1903
1904 if (mode == JOB_TRIGGERING && type != JOB_STOP)
1905 return sd_bus_error_set(error, SD_BUS_ERROR_INVALID_ARGS, "--job-mode=triggering is only valid for stop.");
1906
1907 log_unit_debug(unit, "Trying to enqueue job %s/%s/%s", unit->id, job_type_to_string(type), job_mode_to_string(mode));
1908
1909 type = job_type_collapse(type, unit);
1910
1911 tr = transaction_new(mode == JOB_REPLACE_IRREVERSIBLY);
1912 if (!tr)
1913 return -ENOMEM;
1914
1915 r = transaction_add_job_and_dependencies(tr, type, unit, NULL, true, false,
1916 IN_SET(mode, JOB_IGNORE_DEPENDENCIES, JOB_IGNORE_REQUIREMENTS),
1917 mode == JOB_IGNORE_DEPENDENCIES, error);
1918 if (r < 0)
1919 goto tr_abort;
1920
1921 if (mode == JOB_ISOLATE) {
1922 r = transaction_add_isolate_jobs(tr, m);
1923 if (r < 0)
1924 goto tr_abort;
1925 }
1926
1927 if (mode == JOB_TRIGGERING) {
1928 r = transaction_add_triggering_jobs(tr, unit);
1929 if (r < 0)
1930 goto tr_abort;
1931 }
1932
1933 r = transaction_activate(tr, m, mode, affected_jobs, error);
1934 if (r < 0)
1935 goto tr_abort;
1936
1937 log_unit_debug(unit,
1938 "Enqueued job %s/%s as %u", unit->id,
1939 job_type_to_string(type), (unsigned) tr->anchor_job->id);
1940
1941 if (ret)
1942 *ret = tr->anchor_job;
1943
1944 transaction_free(tr);
1945 return 0;
1946
1947 tr_abort:
1948 transaction_abort(tr);
1949 transaction_free(tr);
1950 return r;
1951 }
1952
1953 int manager_add_job_by_name(Manager *m, JobType type, const char *name, JobMode mode, Set *affected_jobs, sd_bus_error *e, Job **ret) {
1954 Unit *unit = NULL; /* just to appease gcc, initialization is not really necessary */
1955 int r;
1956
1957 assert(m);
1958 assert(type < _JOB_TYPE_MAX);
1959 assert(name);
1960 assert(mode < _JOB_MODE_MAX);
1961
1962 r = manager_load_unit(m, name, NULL, NULL, &unit);
1963 if (r < 0)
1964 return r;
1965 assert(unit);
1966
1967 return manager_add_job(m, type, unit, mode, affected_jobs, e, ret);
1968 }
1969
1970 int manager_add_job_by_name_and_warn(Manager *m, JobType type, const char *name, JobMode mode, Set *affected_jobs, Job **ret) {
1971 _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL;
1972 int r;
1973
1974 assert(m);
1975 assert(type < _JOB_TYPE_MAX);
1976 assert(name);
1977 assert(mode < _JOB_MODE_MAX);
1978
1979 r = manager_add_job_by_name(m, type, name, mode, affected_jobs, &error, ret);
1980 if (r < 0)
1981 return log_warning_errno(r, "Failed to enqueue %s job for %s: %s", job_mode_to_string(mode), name, bus_error_message(&error, r));
1982
1983 return r;
1984 }
1985
1986 int manager_propagate_reload(Manager *m, Unit *unit, JobMode mode, sd_bus_error *e) {
1987 int r;
1988 Transaction *tr;
1989
1990 assert(m);
1991 assert(unit);
1992 assert(mode < _JOB_MODE_MAX);
1993 assert(mode != JOB_ISOLATE); /* Isolate is only valid for start */
1994
1995 tr = transaction_new(mode == JOB_REPLACE_IRREVERSIBLY);
1996 if (!tr)
1997 return -ENOMEM;
1998
1999 /* We need an anchor job */
2000 r = transaction_add_job_and_dependencies(tr, JOB_NOP, unit, NULL, false, false, true, true, e);
2001 if (r < 0)
2002 goto tr_abort;
2003
2004 /* Failure in adding individual dependencies is ignored, so this always succeeds. */
2005 transaction_add_propagate_reload_jobs(tr, unit, tr->anchor_job, mode == JOB_IGNORE_DEPENDENCIES, e);
2006
2007 r = transaction_activate(tr, m, mode, NULL, e);
2008 if (r < 0)
2009 goto tr_abort;
2010
2011 transaction_free(tr);
2012 return 0;
2013
2014 tr_abort:
2015 transaction_abort(tr);
2016 transaction_free(tr);
2017 return r;
2018 }
2019
2020 Job *manager_get_job(Manager *m, uint32_t id) {
2021 assert(m);
2022
2023 return hashmap_get(m->jobs, UINT32_TO_PTR(id));
2024 }
2025
2026 Unit *manager_get_unit(Manager *m, const char *name) {
2027 assert(m);
2028 assert(name);
2029
2030 return hashmap_get(m->units, name);
2031 }
2032
2033 static int manager_dispatch_target_deps_queue(Manager *m) {
2034 Unit *u;
2035 int r = 0;
2036
2037 assert(m);
2038
2039 while ((u = m->target_deps_queue)) {
2040 _cleanup_free_ Unit **targets = NULL;
2041 int n_targets;
2042
2043 assert(u->in_target_deps_queue);
2044
2045 LIST_REMOVE(target_deps_queue, u->manager->target_deps_queue, u);
2046 u->in_target_deps_queue = false;
2047
2048 /* Take an "atomic" snapshot of dependencies here, as the call below will likely modify the
2049 * dependencies, and we can't have it that hash tables we iterate through are modified while
2050 * we are iterating through them. */
2051 n_targets = unit_get_dependency_array(u, UNIT_ATOM_DEFAULT_TARGET_DEPENDENCIES, &targets);
2052 if (n_targets < 0)
2053 return n_targets;
2054
2055 for (int i = 0; i < n_targets; i++) {
2056 r = unit_add_default_target_dependency(u, targets[i]);
2057 if (r < 0)
2058 return r;
2059 }
2060 }
2061
2062 return r;
2063 }
2064
2065 unsigned manager_dispatch_load_queue(Manager *m) {
2066 Unit *u;
2067 unsigned n = 0;
2068
2069 assert(m);
2070
2071 /* Make sure we are not run recursively */
2072 if (m->dispatching_load_queue)
2073 return 0;
2074
2075 m->dispatching_load_queue = true;
2076
2077 /* Dispatches the load queue. Takes a unit from the queue and
2078 * tries to load its data until the queue is empty */
2079
2080 while ((u = m->load_queue)) {
2081 assert(u->in_load_queue);
2082
2083 unit_load(u);
2084 n++;
2085 }
2086
2087 m->dispatching_load_queue = false;
2088
2089 /* Dispatch the units waiting for their target dependencies to be added now, as all targets that we know about
2090 * should be loaded and have aliases resolved */
2091 (void) manager_dispatch_target_deps_queue(m);
2092
2093 return n;
2094 }
2095
2096 bool manager_unit_cache_should_retry_load(Unit *u) {
2097 assert(u);
2098
2099 /* Automatic reloading from disk only applies to units which were not found sometime in the past, and
2100 * the not-found stub is kept pinned in the unit graph by dependencies. For units that were
2101 * previously loaded, we don't do automatic reloading, and daemon-reload is necessary to update. */
2102 if (u->load_state != UNIT_NOT_FOUND)
2103 return false;
2104
2105 /* The cache has been updated since the last time we tried to load the unit. There might be new
2106 * fragment paths to read. */
2107 if (u->manager->unit_cache_timestamp_hash != u->fragment_not_found_timestamp_hash)
2108 return true;
2109
2110 /* The cache needs to be updated because there are modifications on disk. */
2111 return !lookup_paths_timestamp_hash_same(&u->manager->lookup_paths, u->manager->unit_cache_timestamp_hash, NULL);
2112 }
2113
2114 int manager_load_unit_prepare(
2115 Manager *m,
2116 const char *name,
2117 const char *path,
2118 sd_bus_error *e,
2119 Unit **_ret) {
2120
2121 _cleanup_(unit_freep) Unit *cleanup_ret = NULL;
2122 Unit *ret;
2123 UnitType t;
2124 int r;
2125
2126 assert(m);
2127 assert(_ret);
2128
2129 /* This will prepare the unit for loading, but not actually load anything from disk. */
2130
2131 if (path && !path_is_absolute(path))
2132 return sd_bus_error_setf(e, SD_BUS_ERROR_INVALID_ARGS, "Path %s is not absolute.", path);
2133
2134 if (!name) {
2135 /* 'name' and 'path' must not both be null. Check here 'path' using assert_se() to
2136 * workaround a bug in gcc that generates a -Wnonnull warning when calling basename(),
2137 * but this cannot be possible in any code path (See #6119). */
2138 assert_se(path);
2139 name = basename(path);
2140 }
2141
2142 t = unit_name_to_type(name);
2143
2144 if (t == _UNIT_TYPE_INVALID || !unit_name_is_valid(name, UNIT_NAME_PLAIN|UNIT_NAME_INSTANCE)) {
2145 if (unit_name_is_valid(name, UNIT_NAME_TEMPLATE))
2146 return sd_bus_error_setf(e, SD_BUS_ERROR_INVALID_ARGS, "Unit name %s is missing the instance name.", name);
2147
2148 return sd_bus_error_setf(e, SD_BUS_ERROR_INVALID_ARGS, "Unit name %s is not valid.", name);
2149 }
2150
2151 ret = manager_get_unit(m, name);
2152 if (ret) {
2153 /* The time-based cache allows to start new units without daemon-reload,
2154 * but if they are already referenced (because of dependencies or ordering)
2155 * then we have to force a load of the fragment. As an optimization, check
2156 * first if anything in the usual paths was modified since the last time
2157 * the cache was loaded. Also check if the last time an attempt to load the
2158 * unit was made was before the most recent cache refresh, so that we know
2159 * we need to try again — even if the cache is current, it might have been
2160 * updated in a different context before we had a chance to retry loading
2161 * this particular unit. */
2162 if (manager_unit_cache_should_retry_load(ret))
2163 ret->load_state = UNIT_STUB;
2164 else {
2165 *_ret = ret;
2166 return 1;
2167 }
2168 } else {
2169 ret = cleanup_ret = unit_new(m, unit_vtable[t]->object_size);
2170 if (!ret)
2171 return -ENOMEM;
2172 }
2173
2174 if (path) {
2175 r = free_and_strdup(&ret->fragment_path, path);
2176 if (r < 0)
2177 return r;
2178 }
2179
2180 r = unit_add_name(ret, name);
2181 if (r < 0)
2182 return r;
2183
2184 unit_add_to_load_queue(ret);
2185 unit_add_to_dbus_queue(ret);
2186 unit_add_to_gc_queue(ret);
2187
2188 *_ret = ret;
2189 cleanup_ret = NULL;
2190
2191 return 0;
2192 }
2193
2194 int manager_load_unit(
2195 Manager *m,
2196 const char *name,
2197 const char *path,
2198 sd_bus_error *e,
2199 Unit **_ret) {
2200
2201 int r;
2202
2203 assert(m);
2204 assert(_ret);
2205
2206 /* This will load the service information files, but not actually
2207 * start any services or anything. */
2208
2209 r = manager_load_unit_prepare(m, name, path, e, _ret);
2210 if (r != 0)
2211 return r;
2212
2213 manager_dispatch_load_queue(m);
2214
2215 *_ret = unit_follow_merge(*_ret);
2216 return 0;
2217 }
2218
2219 int manager_load_startable_unit_or_warn(
2220 Manager *m,
2221 const char *name,
2222 const char *path,
2223 Unit **ret) {
2224
2225 /* Load a unit, make sure it loaded fully and is not masked. */
2226
2227 _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL;
2228 Unit *unit;
2229 int r;
2230
2231 r = manager_load_unit(m, name, path, &error, &unit);
2232 if (r < 0)
2233 return log_error_errno(r, "Failed to load %s %s: %s",
2234 name ? "unit" : "unit file", name ?: path,
2235 bus_error_message(&error, r));
2236
2237 r = bus_unit_validate_load_state(unit, &error);
2238 if (r < 0)
2239 return log_error_errno(r, "%s", bus_error_message(&error, r));
2240
2241 *ret = unit;
2242 return 0;
2243 }
2244
2245 void manager_clear_jobs(Manager *m) {
2246 Job *j;
2247
2248 assert(m);
2249
2250 while ((j = hashmap_first(m->jobs)))
2251 /* No need to recurse. We're cancelling all jobs. */
2252 job_finish_and_invalidate(j, JOB_CANCELED, false, false);
2253 }
2254
2255 void manager_unwatch_pid(Manager *m, pid_t pid) {
2256 assert(m);
2257
2258 /* First let's drop the unit keyed as "pid". */
2259 (void) hashmap_remove(m->watch_pids, PID_TO_PTR(pid));
2260
2261 /* Then, let's also drop the array keyed by -pid. */
2262 free(hashmap_remove(m->watch_pids, PID_TO_PTR(-pid)));
2263 }
2264
2265 static int manager_dispatch_run_queue(sd_event_source *source, void *userdata) {
2266 Manager *m = ASSERT_PTR(userdata);
2267 Job *j;
2268
2269 assert(source);
2270
2271 while ((j = prioq_peek(m->run_queue))) {
2272 assert(j->installed);
2273 assert(j->in_run_queue);
2274
2275 (void) job_run_and_invalidate(j);
2276 }
2277
2278 if (m->n_running_jobs > 0)
2279 manager_watch_jobs_in_progress(m);
2280
2281 if (m->n_on_console > 0)
2282 manager_watch_idle_pipe(m);
2283
2284 return 1;
2285 }
2286
2287 void manager_trigger_run_queue(Manager *m) {
2288 int r;
2289
2290 assert(m);
2291
2292 r = sd_event_source_set_enabled(
2293 m->run_queue_event_source,
2294 prioq_isempty(m->run_queue) ? SD_EVENT_OFF : SD_EVENT_ONESHOT);
2295 if (r < 0)
2296 log_warning_errno(r, "Failed to enable job run queue event source, ignoring: %m");
2297 }
2298
2299 static unsigned manager_dispatch_dbus_queue(Manager *m) {
2300 unsigned n = 0, budget;
2301 Unit *u;
2302 Job *j;
2303
2304 assert(m);
2305
2306 /* When we are reloading, let's not wait with generating signals, since we need to exit the manager as quickly
2307 * as we can. There's no point in throttling generation of signals in that case. */
2308 if (MANAGER_IS_RELOADING(m) || m->send_reloading_done || m->pending_reload_message)
2309 budget = UINT_MAX; /* infinite budget in this case */
2310 else {
2311 /* Anything to do at all? */
2312 if (!m->dbus_unit_queue && !m->dbus_job_queue)
2313 return 0;
2314
2315 /* Do we have overly many messages queued at the moment? If so, let's not enqueue more on top, let's
2316 * sit this cycle out, and process things in a later cycle when the queues got a bit emptier. */
2317 if (manager_bus_n_queued_write(m) > MANAGER_BUS_BUSY_THRESHOLD)
2318 return 0;
2319
2320 /* Only process a certain number of units/jobs per event loop iteration. Even if the bus queue wasn't
2321 * overly full before this call we shouldn't increase it in size too wildly in one step, and we
2322 * shouldn't monopolize CPU time with generating these messages. Note the difference in counting of
2323 * this "budget" and the "threshold" above: the "budget" is decreased only once per generated message,
2324 * regardless how many buses/direct connections it is enqueued on, while the "threshold" is applied to
2325 * each queued instance of bus message, i.e. if the same message is enqueued to five buses/direct
2326 * connections it will be counted five times. This difference in counting ("references"
2327 * vs. "instances") is primarily a result of the fact that it's easier to implement it this way,
2328 * however it also reflects the thinking that the "threshold" should put a limit on used queue memory,
2329 * i.e. space, while the "budget" should put a limit on time. Also note that the "threshold" is
2330 * currently chosen much higher than the "budget". */
2331 budget = MANAGER_BUS_MESSAGE_BUDGET;
2332 }
2333
2334 while (budget != 0 && (u = m->dbus_unit_queue)) {
2335
2336 assert(u->in_dbus_queue);
2337
2338 bus_unit_send_change_signal(u);
2339 n++;
2340
2341 if (budget != UINT_MAX)
2342 budget--;
2343 }
2344
2345 while (budget != 0 && (j = m->dbus_job_queue)) {
2346 assert(j->in_dbus_queue);
2347
2348 bus_job_send_change_signal(j);
2349 n++;
2350
2351 if (budget != UINT_MAX)
2352 budget--;
2353 }
2354
2355 if (m->send_reloading_done) {
2356 m->send_reloading_done = false;
2357 bus_manager_send_reloading(m, false);
2358 n++;
2359 }
2360
2361 if (m->pending_reload_message) {
2362 bus_send_pending_reload_message(m);
2363 n++;
2364 }
2365
2366 return n;
2367 }
2368
2369 static int manager_dispatch_cgroups_agent_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
2370 Manager *m = userdata;
2371 char buf[PATH_MAX];
2372 ssize_t n;
2373
2374 n = recv(fd, buf, sizeof(buf), 0);
2375 if (n < 0)
2376 return log_error_errno(errno, "Failed to read cgroups agent message: %m");
2377 if (n == 0) {
2378 log_error("Got zero-length cgroups agent message, ignoring.");
2379 return 0;
2380 }
2381 if ((size_t) n >= sizeof(buf)) {
2382 log_error("Got overly long cgroups agent message, ignoring.");
2383 return 0;
2384 }
2385
2386 if (memchr(buf, 0, n)) {
2387 log_error("Got cgroups agent message with embedded NUL byte, ignoring.");
2388 return 0;
2389 }
2390 buf[n] = 0;
2391
2392 manager_notify_cgroup_empty(m, buf);
2393 (void) bus_forward_agent_released(m, buf);
2394
2395 return 0;
2396 }
2397
2398 static bool manager_process_barrier_fd(char * const *tags, FDSet *fds) {
2399
2400 /* nothing else must be sent when using BARRIER=1 */
2401 if (strv_contains(tags, "BARRIER=1")) {
2402 if (strv_length(tags) == 1) {
2403 if (fdset_size(fds) != 1)
2404 log_warning("Got incorrect number of fds with BARRIER=1, closing them.");
2405 } else
2406 log_warning("Extra notification messages sent with BARRIER=1, ignoring everything.");
2407
2408 /* Drop the message if BARRIER=1 was found */
2409 return true;
2410 }
2411
2412 return false;
2413 }
2414
2415 static void manager_invoke_notify_message(
2416 Manager *m,
2417 Unit *u,
2418 const struct ucred *ucred,
2419 char * const *tags,
2420 FDSet *fds) {
2421
2422 assert(m);
2423 assert(u);
2424 assert(ucred);
2425 assert(tags);
2426
2427 if (u->notifygen == m->notifygen) /* Already invoked on this same unit in this same iteration? */
2428 return;
2429 u->notifygen = m->notifygen;
2430
2431 if (UNIT_VTABLE(u)->notify_message)
2432 UNIT_VTABLE(u)->notify_message(u, ucred, tags, fds);
2433
2434 else if (DEBUG_LOGGING) {
2435 _cleanup_free_ char *buf = NULL, *x = NULL, *y = NULL;
2436
2437 buf = strv_join(tags, ", ");
2438 if (buf)
2439 x = ellipsize(buf, 20, 90);
2440 if (x)
2441 y = cescape(x);
2442
2443 log_unit_debug(u, "Got notification message \"%s\", ignoring.", strnull(y));
2444 }
2445 }
2446
2447 static int manager_dispatch_notify_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
2448
2449 _cleanup_fdset_free_ FDSet *fds = NULL;
2450 Manager *m = ASSERT_PTR(userdata);
2451 char buf[NOTIFY_BUFFER_MAX+1];
2452 struct iovec iovec = {
2453 .iov_base = buf,
2454 .iov_len = sizeof(buf)-1,
2455 };
2456 CMSG_BUFFER_TYPE(CMSG_SPACE(sizeof(struct ucred)) +
2457 CMSG_SPACE(sizeof(int) * NOTIFY_FD_MAX)) control;
2458 struct msghdr msghdr = {
2459 .msg_iov = &iovec,
2460 .msg_iovlen = 1,
2461 .msg_control = &control,
2462 .msg_controllen = sizeof(control),
2463 };
2464
2465 struct cmsghdr *cmsg;
2466 struct ucred *ucred = NULL;
2467 _cleanup_free_ Unit **array_copy = NULL;
2468 _cleanup_strv_free_ char **tags = NULL;
2469 Unit *u1, *u2, **array;
2470 int r, *fd_array = NULL;
2471 size_t n_fds = 0;
2472 bool found = false;
2473 ssize_t n;
2474
2475 assert(m->notify_fd == fd);
2476
2477 if (revents != EPOLLIN) {
2478 log_warning("Got unexpected poll event for notify fd.");
2479 return 0;
2480 }
2481
2482 n = recvmsg_safe(m->notify_fd, &msghdr, MSG_DONTWAIT|MSG_CMSG_CLOEXEC|MSG_TRUNC);
2483 if (n < 0) {
2484 if (ERRNO_IS_TRANSIENT(n))
2485 return 0; /* Spurious wakeup, try again */
2486 if (n == -EXFULL) {
2487 log_warning("Got message with truncated control data (too many fds sent?), ignoring.");
2488 return 0;
2489 }
2490 /* If this is any other, real error, then let's stop processing this socket. This of course
2491 * means we won't take notification messages anymore, but that's still better than busy
2492 * looping around this: being woken up over and over again but being unable to actually read
2493 * the message off the socket. */
2494 return log_error_errno(n, "Failed to receive notification message: %m");
2495 }
2496
2497 CMSG_FOREACH(cmsg, &msghdr) {
2498 if (cmsg->cmsg_level == SOL_SOCKET && cmsg->cmsg_type == SCM_RIGHTS) {
2499
2500 assert(!fd_array);
2501 fd_array = (int*) CMSG_DATA(cmsg);
2502 n_fds = (cmsg->cmsg_len - CMSG_LEN(0)) / sizeof(int);
2503
2504 } else if (cmsg->cmsg_level == SOL_SOCKET &&
2505 cmsg->cmsg_type == SCM_CREDENTIALS &&
2506 cmsg->cmsg_len == CMSG_LEN(sizeof(struct ucred))) {
2507
2508 assert(!ucred);
2509 ucred = (struct ucred*) CMSG_DATA(cmsg);
2510 }
2511 }
2512
2513 if (n_fds > 0) {
2514 assert(fd_array);
2515
2516 r = fdset_new_array(&fds, fd_array, n_fds);
2517 if (r < 0) {
2518 close_many(fd_array, n_fds);
2519 log_oom();
2520 return 0;
2521 }
2522 }
2523
2524 if (!ucred || !pid_is_valid(ucred->pid)) {
2525 log_warning("Received notify message without valid credentials. Ignoring.");
2526 return 0;
2527 }
2528
2529 if ((size_t) n >= sizeof(buf) || (msghdr.msg_flags & MSG_TRUNC)) {
2530 log_warning("Received notify message exceeded maximum size. Ignoring.");
2531 return 0;
2532 }
2533
2534 /* As extra safety check, let's make sure the string we get doesn't contain embedded NUL bytes.
2535 * We permit one trailing NUL byte in the message, but don't expect it. */
2536 if (n > 1 && memchr(buf, 0, n-1)) {
2537 log_warning("Received notify message with embedded NUL bytes. Ignoring.");
2538 return 0;
2539 }
2540
2541 /* Make sure it's NUL-terminated, then parse it to obtain the tags list. */
2542 buf[n] = 0;
2543 tags = strv_split_newlines(buf);
2544 if (!tags) {
2545 log_oom();
2546 return 0;
2547 }
2548
2549 /* Possibly a barrier fd, let's see. */
2550 if (manager_process_barrier_fd(tags, fds))
2551 return 0;
2552
2553 /* Increase the generation counter used for filtering out duplicate unit invocations. */
2554 m->notifygen++;
2555
2556 /* Notify every unit that might be interested, which might be multiple. */
2557 u1 = manager_get_unit_by_pid_cgroup(m, ucred->pid);
2558 u2 = hashmap_get(m->watch_pids, PID_TO_PTR(ucred->pid));
2559 array = hashmap_get(m->watch_pids, PID_TO_PTR(-ucred->pid));
2560 if (array) {
2561 size_t k = 0;
2562
2563 while (array[k])
2564 k++;
2565
2566 array_copy = newdup(Unit*, array, k+1);
2567 if (!array_copy)
2568 log_oom();
2569 }
2570 /* And now invoke the per-unit callbacks. Note that manager_invoke_notify_message() will handle
2571 * duplicate units make sure we only invoke each unit's handler once. */
2572 if (u1) {
2573 manager_invoke_notify_message(m, u1, ucred, tags, fds);
2574 found = true;
2575 }
2576 if (u2) {
2577 manager_invoke_notify_message(m, u2, ucred, tags, fds);
2578 found = true;
2579 }
2580 if (array_copy)
2581 for (size_t i = 0; array_copy[i]; i++) {
2582 manager_invoke_notify_message(m, array_copy[i], ucred, tags, fds);
2583 found = true;
2584 }
2585
2586 if (!found)
2587 log_warning("Cannot find unit for notify message of PID "PID_FMT", ignoring.", ucred->pid);
2588
2589 if (fdset_size(fds) > 0)
2590 log_warning("Got extra auxiliary fds with notification message, closing them.");
2591
2592 return 0;
2593 }
2594
2595 static void manager_invoke_sigchld_event(
2596 Manager *m,
2597 Unit *u,
2598 const siginfo_t *si) {
2599
2600 assert(m);
2601 assert(u);
2602 assert(si);
2603
2604 /* Already invoked the handler of this unit in this iteration? Then don't process this again */
2605 if (u->sigchldgen == m->sigchldgen)
2606 return;
2607 u->sigchldgen = m->sigchldgen;
2608
2609 log_unit_debug(u, "Child "PID_FMT" belongs to %s.", si->si_pid, u->id);
2610 unit_unwatch_pid(u, si->si_pid);
2611
2612 if (UNIT_VTABLE(u)->sigchld_event)
2613 UNIT_VTABLE(u)->sigchld_event(u, si->si_pid, si->si_code, si->si_status);
2614 }
2615
2616 static int manager_dispatch_sigchld(sd_event_source *source, void *userdata) {
2617 Manager *m = ASSERT_PTR(userdata);
2618 siginfo_t si = {};
2619 int r;
2620
2621 assert(source);
2622
2623 /* First we call waitid() for a PID and do not reap the zombie. That way we can still access
2624 * /proc/$PID for it while it is a zombie. */
2625
2626 if (waitid(P_ALL, 0, &si, WEXITED|WNOHANG|WNOWAIT) < 0) {
2627
2628 if (errno != ECHILD)
2629 log_error_errno(errno, "Failed to peek for child with waitid(), ignoring: %m");
2630
2631 goto turn_off;
2632 }
2633
2634 if (si.si_pid <= 0)
2635 goto turn_off;
2636
2637 if (IN_SET(si.si_code, CLD_EXITED, CLD_KILLED, CLD_DUMPED)) {
2638 _cleanup_free_ Unit **array_copy = NULL;
2639 _cleanup_free_ char *name = NULL;
2640 Unit *u1, *u2, **array;
2641
2642 (void) get_process_comm(si.si_pid, &name);
2643
2644 log_debug("Child "PID_FMT" (%s) died (code=%s, status=%i/%s)",
2645 si.si_pid, strna(name),
2646 sigchld_code_to_string(si.si_code),
2647 si.si_status,
2648 strna(si.si_code == CLD_EXITED
2649 ? exit_status_to_string(si.si_status, EXIT_STATUS_FULL)
2650 : signal_to_string(si.si_status)));
2651
2652 /* Increase the generation counter used for filtering out duplicate unit invocations */
2653 m->sigchldgen++;
2654
2655 /* And now figure out the unit this belongs to, it might be multiple... */
2656 u1 = manager_get_unit_by_pid_cgroup(m, si.si_pid);
2657 u2 = hashmap_get(m->watch_pids, PID_TO_PTR(si.si_pid));
2658 array = hashmap_get(m->watch_pids, PID_TO_PTR(-si.si_pid));
2659 if (array) {
2660 size_t n = 0;
2661
2662 /* Count how many entries the array has */
2663 while (array[n])
2664 n++;
2665
2666 /* Make a copy of the array so that we don't trip up on the array changing beneath us */
2667 array_copy = newdup(Unit*, array, n+1);
2668 if (!array_copy)
2669 log_oom();
2670 }
2671
2672 /* Finally, execute them all. Note that u1, u2 and the array might contain duplicates, but
2673 * that's fine, manager_invoke_sigchld_event() will ensure we only invoke the handlers once for
2674 * each iteration. */
2675 if (u1) {
2676 /* We check for oom condition, in case we got SIGCHLD before the oom notification.
2677 * We only do this for the cgroup the PID belonged to. */
2678 (void) unit_check_oom(u1);
2679
2680 /* We check if systemd-oomd performed a kill so that we log and notify appropriately */
2681 (void) unit_check_oomd_kill(u1);
2682
2683 manager_invoke_sigchld_event(m, u1, &si);
2684 }
2685 if (u2)
2686 manager_invoke_sigchld_event(m, u2, &si);
2687 if (array_copy)
2688 for (size_t i = 0; array_copy[i]; i++)
2689 manager_invoke_sigchld_event(m, array_copy[i], &si);
2690 }
2691
2692 /* And now, we actually reap the zombie. */
2693 if (waitid(P_PID, si.si_pid, &si, WEXITED) < 0) {
2694 log_error_errno(errno, "Failed to dequeue child, ignoring: %m");
2695 return 0;
2696 }
2697
2698 return 0;
2699
2700 turn_off:
2701 /* All children processed for now, turn off event source */
2702
2703 r = sd_event_source_set_enabled(m->sigchld_event_source, SD_EVENT_OFF);
2704 if (r < 0)
2705 return log_error_errno(r, "Failed to disable SIGCHLD event source: %m");
2706
2707 return 0;
2708 }
2709
2710 static void manager_start_special(Manager *m, const char *name, JobMode mode) {
2711 Job *job;
2712
2713 if (manager_add_job_by_name_and_warn(m, JOB_START, name, mode, NULL, &job) < 0)
2714 return;
2715
2716 const char *s = unit_status_string(job->unit, NULL);
2717
2718 log_info("Activating special unit %s...", s);
2719
2720 sd_notifyf(false,
2721 "STATUS=Activating special unit %s...", s);
2722 m->status_ready = false;
2723 }
2724
2725 static void manager_handle_ctrl_alt_del(Manager *m) {
2726 /* If the user presses C-A-D more than
2727 * 7 times within 2s, we reboot/shutdown immediately,
2728 * unless it was disabled in system.conf */
2729
2730 if (ratelimit_below(&m->ctrl_alt_del_ratelimit) || m->cad_burst_action == EMERGENCY_ACTION_NONE)
2731 manager_start_special(m, SPECIAL_CTRL_ALT_DEL_TARGET, JOB_REPLACE_IRREVERSIBLY);
2732 else
2733 emergency_action(m, m->cad_burst_action, EMERGENCY_ACTION_WARN, NULL, -1,
2734 "Ctrl-Alt-Del was pressed more than 7 times within 2s");
2735 }
2736
2737 static int manager_dispatch_signal_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
2738 Manager *m = ASSERT_PTR(userdata);
2739 ssize_t n;
2740 struct signalfd_siginfo sfsi;
2741 int r;
2742
2743 assert(m->signal_fd == fd);
2744
2745 if (revents != EPOLLIN) {
2746 log_warning("Got unexpected events from signal file descriptor.");
2747 return 0;
2748 }
2749
2750 n = read(m->signal_fd, &sfsi, sizeof(sfsi));
2751 if (n < 0) {
2752 if (ERRNO_IS_TRANSIENT(errno))
2753 return 0;
2754
2755 /* We return an error here, which will kill this handler,
2756 * to avoid a busy loop on read error. */
2757 return log_error_errno(errno, "Reading from signal fd failed: %m");
2758 }
2759 if (n != sizeof(sfsi)) {
2760 log_warning("Truncated read from signal fd (%zi bytes), ignoring!", n);
2761 return 0;
2762 }
2763
2764 log_received_signal(sfsi.ssi_signo == SIGCHLD ||
2765 (sfsi.ssi_signo == SIGTERM && MANAGER_IS_USER(m))
2766 ? LOG_DEBUG : LOG_INFO,
2767 &sfsi);
2768
2769 switch (sfsi.ssi_signo) {
2770
2771 case SIGCHLD:
2772 r = sd_event_source_set_enabled(m->sigchld_event_source, SD_EVENT_ON);
2773 if (r < 0)
2774 log_warning_errno(r, "Failed to enable SIGCHLD event source, ignoring: %m");
2775
2776 break;
2777
2778 case SIGTERM:
2779 if (MANAGER_IS_SYSTEM(m)) {
2780 /* This is for compatibility with the original sysvinit */
2781 if (verify_run_space_and_log("Refusing to reexecute") < 0)
2782 break;
2783
2784 m->objective = MANAGER_REEXECUTE;
2785 break;
2786 }
2787
2788 _fallthrough_;
2789 case SIGINT:
2790 if (MANAGER_IS_SYSTEM(m))
2791 manager_handle_ctrl_alt_del(m);
2792 else
2793 manager_start_special(m, SPECIAL_EXIT_TARGET, JOB_REPLACE_IRREVERSIBLY);
2794 break;
2795
2796 case SIGWINCH:
2797 /* This is a nop on non-init */
2798 if (MANAGER_IS_SYSTEM(m))
2799 manager_start_special(m, SPECIAL_KBREQUEST_TARGET, JOB_REPLACE);
2800
2801 break;
2802
2803 case SIGPWR:
2804 /* This is a nop on non-init */
2805 if (MANAGER_IS_SYSTEM(m))
2806 manager_start_special(m, SPECIAL_SIGPWR_TARGET, JOB_REPLACE);
2807
2808 break;
2809
2810 case SIGUSR1:
2811 if (manager_dbus_is_running(m, false)) {
2812 log_info("Trying to reconnect to bus...");
2813
2814 (void) bus_init_api(m);
2815
2816 if (MANAGER_IS_SYSTEM(m))
2817 (void) bus_init_system(m);
2818 } else
2819 manager_start_special(m, SPECIAL_DBUS_SERVICE, JOB_REPLACE);
2820
2821 break;
2822
2823 case SIGUSR2: {
2824 _cleanup_free_ char *dump = NULL;
2825
2826 r = manager_get_dump_string(m, &dump);
2827 if (r < 0) {
2828 log_warning_errno(errno, "Failed to acquire manager dump: %m");
2829 break;
2830 }
2831
2832 log_dump(LOG_INFO, dump);
2833 break;
2834 }
2835
2836 case SIGHUP:
2837 if (verify_run_space_and_log("Refusing to reload") < 0)
2838 break;
2839
2840 m->objective = MANAGER_RELOAD;
2841 break;
2842
2843 default: {
2844
2845 /* Starting SIGRTMIN+0 */
2846 static const struct {
2847 const char *target;
2848 JobMode mode;
2849 } target_table[] = {
2850 [0] = { SPECIAL_DEFAULT_TARGET, JOB_ISOLATE },
2851 [1] = { SPECIAL_RESCUE_TARGET, JOB_ISOLATE },
2852 [2] = { SPECIAL_EMERGENCY_TARGET, JOB_ISOLATE },
2853 [3] = { SPECIAL_HALT_TARGET, JOB_REPLACE_IRREVERSIBLY },
2854 [4] = { SPECIAL_POWEROFF_TARGET, JOB_REPLACE_IRREVERSIBLY },
2855 [5] = { SPECIAL_REBOOT_TARGET, JOB_REPLACE_IRREVERSIBLY },
2856 [6] = { SPECIAL_KEXEC_TARGET, JOB_REPLACE_IRREVERSIBLY },
2857 };
2858
2859 /* Starting SIGRTMIN+13, so that target halt and system halt are 10 apart */
2860 static const ManagerObjective objective_table[] = {
2861 [0] = MANAGER_HALT,
2862 [1] = MANAGER_POWEROFF,
2863 [2] = MANAGER_REBOOT,
2864 [3] = MANAGER_KEXEC,
2865 };
2866
2867 if ((int) sfsi.ssi_signo >= SIGRTMIN+0 &&
2868 (int) sfsi.ssi_signo < SIGRTMIN+(int) ELEMENTSOF(target_table)) {
2869 int idx = (int) sfsi.ssi_signo - SIGRTMIN;
2870 manager_start_special(m, target_table[idx].target, target_table[idx].mode);
2871 break;
2872 }
2873
2874 if ((int) sfsi.ssi_signo >= SIGRTMIN+13 &&
2875 (int) sfsi.ssi_signo < SIGRTMIN+13+(int) ELEMENTSOF(objective_table)) {
2876 m->objective = objective_table[sfsi.ssi_signo - SIGRTMIN - 13];
2877 break;
2878 }
2879
2880 switch (sfsi.ssi_signo - SIGRTMIN) {
2881
2882 case 20:
2883 manager_override_show_status(m, SHOW_STATUS_YES, "signal");
2884 break;
2885
2886 case 21:
2887 manager_override_show_status(m, SHOW_STATUS_NO, "signal");
2888 break;
2889
2890 case 22:
2891 manager_override_log_level(m, LOG_DEBUG);
2892 break;
2893
2894 case 23:
2895 manager_restore_original_log_level(m);
2896 break;
2897
2898 case 24:
2899 if (MANAGER_IS_USER(m)) {
2900 m->objective = MANAGER_EXIT;
2901 return 0;
2902 }
2903
2904 /* This is a nop on init */
2905 break;
2906
2907 case 25:
2908 m->objective = MANAGER_REEXECUTE;
2909 break;
2910
2911 case 26:
2912 case 29: /* compatibility: used to be mapped to LOG_TARGET_SYSLOG_OR_KMSG */
2913 manager_restore_original_log_target(m);
2914 break;
2915
2916 case 27:
2917 manager_override_log_target(m, LOG_TARGET_CONSOLE);
2918 break;
2919
2920 case 28:
2921 manager_override_log_target(m, LOG_TARGET_KMSG);
2922 break;
2923
2924 default:
2925 log_warning("Got unhandled signal <%s>.", signal_to_string(sfsi.ssi_signo));
2926 }
2927 }}
2928
2929 return 0;
2930 }
2931
2932 static int manager_dispatch_time_change_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
2933 Manager *m = ASSERT_PTR(userdata);
2934 Unit *u;
2935
2936 log_struct(LOG_DEBUG,
2937 "MESSAGE_ID=" SD_MESSAGE_TIME_CHANGE_STR,
2938 LOG_MESSAGE("Time has been changed"));
2939
2940 /* Restart the watch */
2941 (void) manager_setup_time_change(m);
2942
2943 HASHMAP_FOREACH(u, m->units)
2944 if (UNIT_VTABLE(u)->time_change)
2945 UNIT_VTABLE(u)->time_change(u);
2946
2947 return 0;
2948 }
2949
2950 static int manager_dispatch_timezone_change(
2951 sd_event_source *source,
2952 const struct inotify_event *e,
2953 void *userdata) {
2954
2955 Manager *m = ASSERT_PTR(userdata);
2956 int changed;
2957 Unit *u;
2958
2959 log_debug("inotify event for /etc/localtime");
2960
2961 changed = manager_read_timezone_stat(m);
2962 if (changed <= 0)
2963 return changed;
2964
2965 /* Something changed, restart the watch, to ensure we watch the new /etc/localtime if it changed */
2966 (void) manager_setup_timezone_change(m);
2967
2968 /* Read the new timezone */
2969 tzset();
2970
2971 log_debug("Timezone has been changed (now: %s).", tzname[daylight]);
2972
2973 HASHMAP_FOREACH(u, m->units)
2974 if (UNIT_VTABLE(u)->timezone_change)
2975 UNIT_VTABLE(u)->timezone_change(u);
2976
2977 return 0;
2978 }
2979
2980 static int manager_dispatch_idle_pipe_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
2981 Manager *m = ASSERT_PTR(userdata);
2982
2983 assert(m->idle_pipe[2] == fd);
2984
2985 /* There's at least one Type=idle child that just gave up on us waiting for the boot process to
2986 * complete. Let's now turn off any further console output if there's at least one service that needs
2987 * console access, so that from now on our own output should not spill into that service's output
2988 * anymore. After all, we support Type=idle only to beautify console output and it generally is set
2989 * on services that want to own the console exclusively without our interference. */
2990 m->no_console_output = m->n_on_console > 0;
2991
2992 /* Acknowledge the child's request, and let all all other children know too that they shouldn't wait
2993 * any longer by closing the pipes towards them, which is what they are waiting for. */
2994 manager_close_idle_pipe(m);
2995
2996 return 0;
2997 }
2998
2999 static int manager_dispatch_jobs_in_progress(sd_event_source *source, usec_t usec, void *userdata) {
3000 Manager *m = ASSERT_PTR(userdata);
3001 int r;
3002
3003 assert(source);
3004
3005 manager_print_jobs_in_progress(m);
3006
3007 r = sd_event_source_set_time_relative(source, JOBS_IN_PROGRESS_PERIOD_USEC);
3008 if (r < 0)
3009 return r;
3010
3011 return sd_event_source_set_enabled(source, SD_EVENT_ONESHOT);
3012 }
3013
3014 int manager_loop(Manager *m) {
3015 RateLimit rl = { .interval = 1*USEC_PER_SEC, .burst = 50000 };
3016 int r;
3017
3018 assert(m);
3019 assert(m->objective == MANAGER_OK); /* Ensure manager_startup() has been called */
3020
3021 manager_check_finished(m);
3022
3023 /* There might still be some zombies hanging around from before we were exec()'ed. Let's reap them. */
3024 r = sd_event_source_set_enabled(m->sigchld_event_source, SD_EVENT_ON);
3025 if (r < 0)
3026 return log_error_errno(r, "Failed to enable SIGCHLD event source: %m");
3027
3028 while (m->objective == MANAGER_OK) {
3029
3030 (void) watchdog_ping();
3031
3032 if (!ratelimit_below(&rl)) {
3033 /* Yay, something is going seriously wrong, pause a little */
3034 log_warning("Looping too fast. Throttling execution a little.");
3035 sleep(1);
3036 }
3037
3038 if (manager_dispatch_load_queue(m) > 0)
3039 continue;
3040
3041 if (manager_dispatch_gc_job_queue(m) > 0)
3042 continue;
3043
3044 if (manager_dispatch_gc_unit_queue(m) > 0)
3045 continue;
3046
3047 if (manager_dispatch_cleanup_queue(m) > 0)
3048 continue;
3049
3050 if (manager_dispatch_cgroup_realize_queue(m) > 0)
3051 continue;
3052
3053 if (manager_dispatch_start_when_upheld_queue(m) > 0)
3054 continue;
3055
3056 if (manager_dispatch_stop_when_bound_queue(m) > 0)
3057 continue;
3058
3059 if (manager_dispatch_stop_when_unneeded_queue(m) > 0)
3060 continue;
3061
3062 if (manager_dispatch_dbus_queue(m) > 0)
3063 continue;
3064
3065 /* Sleep for watchdog runtime wait time */
3066 r = sd_event_run(m->event, watchdog_runtime_wait());
3067 if (r < 0)
3068 return log_error_errno(r, "Failed to run event loop: %m");
3069 }
3070
3071 return m->objective;
3072 }
3073
3074 int manager_load_unit_from_dbus_path(Manager *m, const char *s, sd_bus_error *e, Unit **_u) {
3075 _cleanup_free_ char *n = NULL;
3076 sd_id128_t invocation_id;
3077 Unit *u;
3078 int r;
3079
3080 assert(m);
3081 assert(s);
3082 assert(_u);
3083
3084 r = unit_name_from_dbus_path(s, &n);
3085 if (r < 0)
3086 return r;
3087
3088 /* Permit addressing units by invocation ID: if the passed bus path is suffixed by a 128bit ID then
3089 * we use it as invocation ID. */
3090 r = sd_id128_from_string(n, &invocation_id);
3091 if (r >= 0) {
3092 u = hashmap_get(m->units_by_invocation_id, &invocation_id);
3093 if (u) {
3094 *_u = u;
3095 return 0;
3096 }
3097
3098 return sd_bus_error_setf(e, BUS_ERROR_NO_UNIT_FOR_INVOCATION_ID,
3099 "No unit with the specified invocation ID " SD_ID128_FORMAT_STR " known.",
3100 SD_ID128_FORMAT_VAL(invocation_id));
3101 }
3102
3103 /* If this didn't work, we check if this is a unit name */
3104 if (!unit_name_is_valid(n, UNIT_NAME_PLAIN|UNIT_NAME_INSTANCE)) {
3105 _cleanup_free_ char *nn = NULL;
3106
3107 nn = cescape(n);
3108 return sd_bus_error_setf(e, SD_BUS_ERROR_INVALID_ARGS,
3109 "Unit name %s is neither a valid invocation ID nor unit name.", strnull(nn));
3110 }
3111
3112 r = manager_load_unit(m, n, NULL, e, &u);
3113 if (r < 0)
3114 return r;
3115
3116 *_u = u;
3117 return 0;
3118 }
3119
3120 int manager_get_job_from_dbus_path(Manager *m, const char *s, Job **_j) {
3121 const char *p;
3122 unsigned id;
3123 Job *j;
3124 int r;
3125
3126 assert(m);
3127 assert(s);
3128 assert(_j);
3129
3130 p = startswith(s, "/org/freedesktop/systemd1/job/");
3131 if (!p)
3132 return -EINVAL;
3133
3134 r = safe_atou(p, &id);
3135 if (r < 0)
3136 return r;
3137
3138 j = manager_get_job(m, id);
3139 if (!j)
3140 return -ENOENT;
3141
3142 *_j = j;
3143
3144 return 0;
3145 }
3146
3147 void manager_send_unit_audit(Manager *m, Unit *u, int type, bool success) {
3148
3149 #if HAVE_AUDIT
3150 _cleanup_free_ char *p = NULL;
3151 const char *msg;
3152 int audit_fd, r;
3153
3154 if (!MANAGER_IS_SYSTEM(m))
3155 return;
3156
3157 audit_fd = get_audit_fd();
3158 if (audit_fd < 0)
3159 return;
3160
3161 /* Don't generate audit events if the service was already
3162 * started and we're just deserializing */
3163 if (MANAGER_IS_RELOADING(m))
3164 return;
3165
3166 if (u->type != UNIT_SERVICE)
3167 return;
3168
3169 r = unit_name_to_prefix_and_instance(u->id, &p);
3170 if (r < 0) {
3171 log_error_errno(r, "Failed to extract prefix and instance of unit name: %m");
3172 return;
3173 }
3174
3175 msg = strjoina("unit=", p);
3176 if (audit_log_user_comm_message(audit_fd, type, msg, "systemd", NULL, NULL, NULL, success) < 0) {
3177 if (errno == EPERM)
3178 /* We aren't allowed to send audit messages?
3179 * Then let's not retry again. */
3180 close_audit_fd();
3181 else
3182 log_warning_errno(errno, "Failed to send audit message: %m");
3183 }
3184 #endif
3185
3186 }
3187
3188 void manager_send_unit_plymouth(Manager *m, Unit *u) {
3189 static const union sockaddr_union sa = PLYMOUTH_SOCKET;
3190 _cleanup_free_ char *message = NULL;
3191 _cleanup_close_ int fd = -1;
3192 int n = 0;
3193
3194 /* Don't generate plymouth events if the service was already
3195 * started and we're just deserializing */
3196 if (MANAGER_IS_RELOADING(m))
3197 return;
3198
3199 if (!MANAGER_IS_SYSTEM(m))
3200 return;
3201
3202 if (detect_container() > 0)
3203 return;
3204
3205 if (!IN_SET(u->type, UNIT_SERVICE, UNIT_MOUNT, UNIT_SWAP))
3206 return;
3207
3208 /* We set SOCK_NONBLOCK here so that we rather drop the
3209 * message then wait for plymouth */
3210 fd = socket(AF_UNIX, SOCK_STREAM|SOCK_CLOEXEC|SOCK_NONBLOCK, 0);
3211 if (fd < 0) {
3212 log_error_errno(errno, "socket() failed: %m");
3213 return;
3214 }
3215
3216 if (connect(fd, &sa.sa, SOCKADDR_UN_LEN(sa.un)) < 0) {
3217 if (!IN_SET(errno, EAGAIN, ENOENT) && !ERRNO_IS_DISCONNECT(errno))
3218 log_error_errno(errno, "connect() failed: %m");
3219 return;
3220 }
3221
3222 if (asprintf(&message, "U\002%c%s%n", (int) (strlen(u->id) + 1), u->id, &n) < 0)
3223 return (void) log_oom();
3224
3225 errno = 0;
3226 if (write(fd, message, n + 1) != n + 1)
3227 if (!IN_SET(errno, EAGAIN, ENOENT) && !ERRNO_IS_DISCONNECT(errno))
3228 log_error_errno(errno, "Failed to write Plymouth message: %m");
3229 }
3230
3231 usec_t manager_get_watchdog(Manager *m, WatchdogType t) {
3232 assert(m);
3233
3234 if (MANAGER_IS_USER(m))
3235 return USEC_INFINITY;
3236
3237 if (timestamp_is_set(m->watchdog_overridden[t]))
3238 return m->watchdog_overridden[t];
3239
3240 return m->watchdog[t];
3241 }
3242
3243 void manager_set_watchdog(Manager *m, WatchdogType t, usec_t timeout) {
3244
3245 assert(m);
3246
3247 if (MANAGER_IS_USER(m))
3248 return;
3249
3250 if (m->watchdog[t] == timeout)
3251 return;
3252
3253 if (t == WATCHDOG_RUNTIME) {
3254 if (!timestamp_is_set(m->watchdog_overridden[WATCHDOG_RUNTIME]))
3255 (void) watchdog_setup(timeout);
3256 } else if (t == WATCHDOG_PRETIMEOUT)
3257 if (m->watchdog_overridden[WATCHDOG_PRETIMEOUT] == USEC_INFINITY)
3258 (void) watchdog_setup_pretimeout(timeout);
3259
3260 m->watchdog[t] = timeout;
3261 }
3262
3263 void manager_override_watchdog(Manager *m, WatchdogType t, usec_t timeout) {
3264
3265 assert(m);
3266
3267 if (MANAGER_IS_USER(m))
3268 return;
3269
3270 if (m->watchdog_overridden[t] == timeout)
3271 return;
3272
3273 if (t == WATCHDOG_RUNTIME) {
3274 usec_t usec = timestamp_is_set(timeout) ? timeout : m->watchdog[t];
3275
3276 (void) watchdog_setup(usec);
3277 } else if (t == WATCHDOG_PRETIMEOUT)
3278 (void) watchdog_setup_pretimeout(timeout);
3279
3280 m->watchdog_overridden[t] = timeout;
3281 }
3282
3283 int manager_set_watchdog_pretimeout_governor(Manager *m, const char *governor) {
3284 _cleanup_free_ char *p = NULL;
3285 int r;
3286
3287 assert(m);
3288
3289 if (MANAGER_IS_USER(m))
3290 return 0;
3291
3292 if (streq_ptr(m->watchdog_pretimeout_governor, governor))
3293 return 0;
3294
3295 p = strdup(governor);
3296 if (!p)
3297 return -ENOMEM;
3298
3299 r = watchdog_setup_pretimeout_governor(governor);
3300 if (r < 0)
3301 return r;
3302
3303 return free_and_replace(m->watchdog_pretimeout_governor, p);
3304 }
3305
3306 int manager_override_watchdog_pretimeout_governor(Manager *m, const char *governor) {
3307 _cleanup_free_ char *p = NULL;
3308 int r;
3309
3310 assert(m);
3311
3312 if (MANAGER_IS_USER(m))
3313 return 0;
3314
3315 if (streq_ptr(m->watchdog_pretimeout_governor_overridden, governor))
3316 return 0;
3317
3318 p = strdup(governor);
3319 if (!p)
3320 return -ENOMEM;
3321
3322 r = watchdog_setup_pretimeout_governor(governor);
3323 if (r < 0)
3324 return r;
3325
3326 return free_and_replace(m->watchdog_pretimeout_governor_overridden, p);
3327 }
3328
3329 int manager_reload(Manager *m) {
3330 _unused_ _cleanup_(manager_reloading_stopp) Manager *reloading = NULL;
3331 _cleanup_fdset_free_ FDSet *fds = NULL;
3332 _cleanup_fclose_ FILE *f = NULL;
3333 int r;
3334
3335 assert(m);
3336
3337 r = manager_open_serialization(m, &f);
3338 if (r < 0)
3339 return log_error_errno(r, "Failed to create serialization file: %m");
3340
3341 fds = fdset_new();
3342 if (!fds)
3343 return log_oom();
3344
3345 /* We are officially in reload mode from here on. */
3346 reloading = manager_reloading_start(m);
3347
3348 r = manager_serialize(m, f, fds, false);
3349 if (r < 0)
3350 return r;
3351
3352 if (fseeko(f, 0, SEEK_SET) < 0)
3353 return log_error_errno(errno, "Failed to seek to beginning of serialization: %m");
3354
3355 /* 💀 This is the point of no return, from here on there is no way back. 💀 */
3356 reloading = NULL;
3357
3358 bus_manager_send_reloading(m, true);
3359
3360 /* Start by flushing out all jobs and units, all generated units, all runtime environments, all dynamic users
3361 * and everything else that is worth flushing out. We'll get it all back from the serialization — if we need
3362 * it. */
3363
3364 manager_clear_jobs_and_units(m);
3365 lookup_paths_flush_generator(&m->lookup_paths);
3366 lookup_paths_free(&m->lookup_paths);
3367 exec_runtime_vacuum(m);
3368 dynamic_user_vacuum(m, false);
3369 m->uid_refs = hashmap_free(m->uid_refs);
3370 m->gid_refs = hashmap_free(m->gid_refs);
3371
3372 r = lookup_paths_init_or_warn(&m->lookup_paths, m->unit_file_scope, 0, NULL);
3373 if (r < 0)
3374 return r;
3375
3376 (void) manager_run_environment_generators(m);
3377 (void) manager_run_generators(m);
3378
3379 lookup_paths_log(&m->lookup_paths);
3380
3381 /* We flushed out generated files, for which we don't watch mtime, so we should flush the old map. */
3382 manager_free_unit_name_maps(m);
3383
3384 /* First, enumerate what we can from kernel and suchlike */
3385 manager_enumerate_perpetual(m);
3386 manager_enumerate(m);
3387
3388 /* Second, deserialize our stored data */
3389 r = manager_deserialize(m, f, fds);
3390 if (r < 0)
3391 log_warning_errno(r, "Deserialization failed, proceeding anyway: %m");
3392
3393 /* We don't need the serialization anymore */
3394 f = safe_fclose(f);
3395
3396 /* Re-register notify_fd as event source, and set up other sockets/communication channels we might need */
3397 (void) manager_setup_notify(m);
3398 (void) manager_setup_cgroups_agent(m);
3399 (void) manager_setup_user_lookup_fd(m);
3400
3401 /* Third, fire things up! */
3402 manager_coldplug(m);
3403
3404 /* Clean up runtime objects no longer referenced */
3405 manager_vacuum(m);
3406
3407 /* Clean up deserialized tracked clients */
3408 m->deserialized_subscribed = strv_free(m->deserialized_subscribed);
3409
3410 /* Consider the reload process complete now. */
3411 assert(m->n_reloading > 0);
3412 m->n_reloading--;
3413
3414 manager_ready(m);
3415
3416 m->send_reloading_done = true;
3417 return 0;
3418 }
3419
3420 void manager_reset_failed(Manager *m) {
3421 Unit *u;
3422
3423 assert(m);
3424
3425 HASHMAP_FOREACH(u, m->units)
3426 unit_reset_failed(u);
3427 }
3428
3429 bool manager_unit_inactive_or_pending(Manager *m, const char *name) {
3430 Unit *u;
3431
3432 assert(m);
3433 assert(name);
3434
3435 /* Returns true if the unit is inactive or going down */
3436 u = manager_get_unit(m, name);
3437 if (!u)
3438 return true;
3439
3440 return unit_inactive_or_pending(u);
3441 }
3442
3443 static void log_taint_string(Manager *m) {
3444 _cleanup_free_ char *taint = NULL;
3445
3446 assert(m);
3447
3448 if (MANAGER_IS_USER(m) || m->taint_logged)
3449 return;
3450
3451 m->taint_logged = true; /* only check for taint once */
3452
3453 taint = manager_taint_string(m);
3454 if (isempty(taint))
3455 return;
3456
3457 log_struct(LOG_NOTICE,
3458 LOG_MESSAGE("System is tainted: %s", taint),
3459 "TAINT=%s", taint,
3460 "MESSAGE_ID=" SD_MESSAGE_TAINTED_STR);
3461 }
3462
3463 static void manager_notify_finished(Manager *m) {
3464 usec_t firmware_usec, loader_usec, kernel_usec, initrd_usec, userspace_usec, total_usec;
3465
3466 if (MANAGER_IS_TEST_RUN(m))
3467 return;
3468
3469 if (MANAGER_IS_SYSTEM(m) && detect_container() <= 0) {
3470 char buf[FORMAT_TIMESPAN_MAX + STRLEN(" (firmware) + ") + FORMAT_TIMESPAN_MAX + STRLEN(" (loader) + ")]
3471 = {};
3472 char *p = buf;
3473 size_t size = sizeof buf;
3474
3475 /* Note that MANAGER_TIMESTAMP_KERNEL's monotonic value is always at 0, and
3476 * MANAGER_TIMESTAMP_FIRMWARE's and MANAGER_TIMESTAMP_LOADER's monotonic value should be considered
3477 * negative values. */
3478
3479 firmware_usec = m->timestamps[MANAGER_TIMESTAMP_FIRMWARE].monotonic - m->timestamps[MANAGER_TIMESTAMP_LOADER].monotonic;
3480 loader_usec = m->timestamps[MANAGER_TIMESTAMP_LOADER].monotonic - m->timestamps[MANAGER_TIMESTAMP_KERNEL].monotonic;
3481 userspace_usec = m->timestamps[MANAGER_TIMESTAMP_FINISH].monotonic - m->timestamps[MANAGER_TIMESTAMP_USERSPACE].monotonic;
3482 total_usec = m->timestamps[MANAGER_TIMESTAMP_FIRMWARE].monotonic + m->timestamps[MANAGER_TIMESTAMP_FINISH].monotonic;
3483
3484 if (firmware_usec > 0)
3485 size = strpcpyf(&p, size, "%s (firmware) + ", FORMAT_TIMESPAN(firmware_usec, USEC_PER_MSEC));
3486 if (loader_usec > 0)
3487 size = strpcpyf(&p, size, "%s (loader) + ", FORMAT_TIMESPAN(loader_usec, USEC_PER_MSEC));
3488
3489 if (dual_timestamp_is_set(&m->timestamps[MANAGER_TIMESTAMP_INITRD])) {
3490
3491 /* The initrd case on bare-metal */
3492 kernel_usec = m->timestamps[MANAGER_TIMESTAMP_INITRD].monotonic - m->timestamps[MANAGER_TIMESTAMP_KERNEL].monotonic;
3493 initrd_usec = m->timestamps[MANAGER_TIMESTAMP_USERSPACE].monotonic - m->timestamps[MANAGER_TIMESTAMP_INITRD].monotonic;
3494
3495 log_struct(LOG_INFO,
3496 "MESSAGE_ID=" SD_MESSAGE_STARTUP_FINISHED_STR,
3497 "KERNEL_USEC="USEC_FMT, kernel_usec,
3498 "INITRD_USEC="USEC_FMT, initrd_usec,
3499 "USERSPACE_USEC="USEC_FMT, userspace_usec,
3500 LOG_MESSAGE("Startup finished in %s%s (kernel) + %s (initrd) + %s (userspace) = %s.",
3501 buf,
3502 FORMAT_TIMESPAN(kernel_usec, USEC_PER_MSEC),
3503 FORMAT_TIMESPAN(initrd_usec, USEC_PER_MSEC),
3504 FORMAT_TIMESPAN(userspace_usec, USEC_PER_MSEC),
3505 FORMAT_TIMESPAN(total_usec, USEC_PER_MSEC)));
3506 } else {
3507 /* The initrd-less case on bare-metal */
3508
3509 kernel_usec = m->timestamps[MANAGER_TIMESTAMP_USERSPACE].monotonic - m->timestamps[MANAGER_TIMESTAMP_KERNEL].monotonic;
3510 initrd_usec = 0;
3511
3512 log_struct(LOG_INFO,
3513 "MESSAGE_ID=" SD_MESSAGE_STARTUP_FINISHED_STR,
3514 "KERNEL_USEC="USEC_FMT, kernel_usec,
3515 "USERSPACE_USEC="USEC_FMT, userspace_usec,
3516 LOG_MESSAGE("Startup finished in %s%s (kernel) + %s (userspace) = %s.",
3517 buf,
3518 FORMAT_TIMESPAN(kernel_usec, USEC_PER_MSEC),
3519 FORMAT_TIMESPAN(userspace_usec, USEC_PER_MSEC),
3520 FORMAT_TIMESPAN(total_usec, USEC_PER_MSEC)));
3521 }
3522 } else {
3523 /* The container and --user case */
3524 firmware_usec = loader_usec = initrd_usec = kernel_usec = 0;
3525 total_usec = userspace_usec = m->timestamps[MANAGER_TIMESTAMP_FINISH].monotonic - m->timestamps[MANAGER_TIMESTAMP_USERSPACE].monotonic;
3526
3527 log_struct(LOG_INFO,
3528 "MESSAGE_ID=" SD_MESSAGE_USER_STARTUP_FINISHED_STR,
3529 "USERSPACE_USEC="USEC_FMT, userspace_usec,
3530 LOG_MESSAGE("Startup finished in %s.",
3531 FORMAT_TIMESPAN(total_usec, USEC_PER_MSEC)));
3532 }
3533
3534 bus_manager_send_finished(m, firmware_usec, loader_usec, kernel_usec, initrd_usec, userspace_usec, total_usec);
3535
3536 log_taint_string(m);
3537 }
3538
3539 static void user_manager_send_ready(Manager *m) {
3540 int r;
3541
3542 assert(m);
3543
3544 /* We send READY=1 on reaching basic.target only when running in --user mode. */
3545 if (!MANAGER_IS_USER(m) || m->ready_sent)
3546 return;
3547
3548 r = sd_notify(false,
3549 "READY=1\n"
3550 "STATUS=Reached " SPECIAL_BASIC_TARGET ".");
3551 if (r < 0)
3552 log_warning_errno(r, "Failed to send readiness notification, ignoring: %m");
3553
3554 m->ready_sent = true;
3555 m->status_ready = false;
3556 }
3557
3558 static void manager_send_ready(Manager *m) {
3559 int r;
3560
3561 if (m->ready_sent && m->status_ready)
3562 /* Skip the notification if nothing changed. */
3563 return;
3564
3565 r = sd_notify(false,
3566 "READY=1\n"
3567 "STATUS=Ready.");
3568 if (r < 0)
3569 log_full_errno(m->ready_sent ? LOG_DEBUG : LOG_WARNING, r,
3570 "Failed to send readiness notification, ignoring: %m");
3571
3572 m->ready_sent = m->status_ready = true;
3573 }
3574
3575 static void manager_check_basic_target(Manager *m) {
3576 Unit *u;
3577
3578 assert(m);
3579
3580 /* Small shortcut */
3581 if (m->ready_sent && m->taint_logged)
3582 return;
3583
3584 u = manager_get_unit(m, SPECIAL_BASIC_TARGET);
3585 if (!u || !UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u)))
3586 return;
3587
3588 /* For user managers, send out READY=1 as soon as we reach basic.target */
3589 user_manager_send_ready(m);
3590
3591 /* Log the taint string as soon as we reach basic.target */
3592 log_taint_string(m);
3593 }
3594
3595 void manager_check_finished(Manager *m) {
3596 assert(m);
3597
3598 if (MANAGER_IS_RELOADING(m))
3599 return;
3600
3601 /* Verify that we have entered the event loop already, and not left it again. */
3602 if (!MANAGER_IS_RUNNING(m))
3603 return;
3604
3605 manager_check_basic_target(m);
3606
3607 if (hashmap_size(m->jobs) > 0) {
3608 if (m->jobs_in_progress_event_source)
3609 /* Ignore any failure, this is only for feedback */
3610 (void) sd_event_source_set_time(m->jobs_in_progress_event_source,
3611 manager_watch_jobs_next_time(m));
3612 return;
3613 }
3614
3615 /* The jobs hashmap tends to grow a lot during boot, and then it's not reused until shutdown. Let's
3616 kill the hashmap if it is relatively large. */
3617 if (hashmap_buckets(m->jobs) > hashmap_size(m->units) / 10)
3618 m->jobs = hashmap_free(m->jobs);
3619
3620 manager_send_ready(m);
3621
3622 /* Notify Type=idle units that we are done now */
3623 manager_close_idle_pipe(m);
3624
3625 if (MANAGER_IS_FINISHED(m))
3626 return;
3627
3628 manager_flip_auto_status(m, false, "boot finished");
3629
3630 /* Turn off confirm spawn now */
3631 m->confirm_spawn = NULL;
3632
3633 /* No need to update ask password status when we're going non-interactive */
3634 manager_close_ask_password(m);
3635
3636 /* This is no longer the first boot */
3637 manager_set_first_boot(m, false);
3638
3639 dual_timestamp_get(m->timestamps + MANAGER_TIMESTAMP_FINISH);
3640
3641 manager_notify_finished(m);
3642
3643 manager_invalidate_startup_units(m);
3644 }
3645
3646 static bool generator_path_any(const char* const* paths) {
3647 bool found = false;
3648
3649 /* Optimize by skipping the whole process by not creating output directories
3650 * if no generators are found. */
3651 STRV_FOREACH(path, paths)
3652 if (access(*path, F_OK) == 0)
3653 found = true;
3654 else if (errno != ENOENT)
3655 log_warning_errno(errno, "Failed to open generator directory %s: %m", *path);
3656
3657 return found;
3658 }
3659
3660 static int manager_run_environment_generators(Manager *m) {
3661 char **tmp = NULL; /* this is only used in the forked process, no cleanup here */
3662 _cleanup_strv_free_ char **paths = NULL;
3663 void* args[] = {
3664 [STDOUT_GENERATE] = &tmp,
3665 [STDOUT_COLLECT] = &tmp,
3666 [STDOUT_CONSUME] = &m->transient_environment,
3667 };
3668 int r;
3669
3670 if (MANAGER_IS_TEST_RUN(m) && !(m->test_run_flags & MANAGER_TEST_RUN_ENV_GENERATORS))
3671 return 0;
3672
3673 paths = env_generator_binary_paths(MANAGER_IS_SYSTEM(m));
3674 if (!paths)
3675 return log_oom();
3676
3677 if (!generator_path_any((const char* const*) paths))
3678 return 0;
3679
3680 RUN_WITH_UMASK(0022)
3681 r = execute_directories((const char* const*) paths, DEFAULT_TIMEOUT_USEC, gather_environment,
3682 args, NULL, m->transient_environment,
3683 EXEC_DIR_PARALLEL | EXEC_DIR_IGNORE_ERRORS | EXEC_DIR_SET_SYSTEMD_EXEC_PID);
3684 return r;
3685 }
3686
3687 static int build_generator_environment(Manager *m, char ***ret) {
3688 _cleanup_strv_free_ char **nl = NULL;
3689 Virtualization v;
3690 int r;
3691
3692 assert(m);
3693 assert(ret);
3694
3695 /* Generators oftentimes want to know some basic facts about the environment they run in, in order to
3696 * adjust generated units to that. Let's pass down some bits of information that are easy for us to
3697 * determine (but a bit harder for generator scripts to determine), as environment variables. */
3698
3699 nl = strv_copy(m->transient_environment);
3700 if (!nl)
3701 return -ENOMEM;
3702
3703 r = strv_env_assign(&nl, "SYSTEMD_SCOPE", MANAGER_IS_SYSTEM(m) ? "system" : "user");
3704 if (r < 0)
3705 return r;
3706
3707 if (MANAGER_IS_SYSTEM(m)) {
3708 /* Note that $SYSTEMD_IN_INITRD may be used to override the initrd detection in much of our
3709 * codebase. This is hence more than purely informational. It will shortcut detection of the
3710 * initrd state if generators invoke our own tools. But that's OK, as it would come to the
3711 * same results (hopefully). */
3712 r = strv_env_assign(&nl, "SYSTEMD_IN_INITRD", one_zero(in_initrd()));
3713 if (r < 0)
3714 return r;
3715
3716 if (m->first_boot >= 0) {
3717 r = strv_env_assign(&nl, "SYSTEMD_FIRST_BOOT", one_zero(m->first_boot));
3718 if (r < 0)
3719 return r;
3720 }
3721 }
3722
3723 v = detect_virtualization();
3724 if (v < 0)
3725 log_debug_errno(v, "Failed to detect virtualization, ignoring: %m");
3726 else if (v > 0) {
3727 const char *s;
3728
3729 s = strjoina(VIRTUALIZATION_IS_VM(v) ? "vm:" :
3730 VIRTUALIZATION_IS_CONTAINER(v) ? "container:" : ":",
3731 virtualization_to_string(v));
3732
3733 r = strv_env_assign(&nl, "SYSTEMD_VIRTUALIZATION", s);
3734 if (r < 0)
3735 return r;
3736 }
3737
3738 r = strv_env_assign(&nl, "SYSTEMD_ARCHITECTURE", architecture_to_string(uname_architecture()));
3739 if (r < 0)
3740 return r;
3741
3742 *ret = TAKE_PTR(nl);
3743 return 0;
3744 }
3745
3746 static int manager_run_generators(Manager *m) {
3747 _cleanup_strv_free_ char **paths = NULL, **ge = NULL;
3748 int r;
3749
3750 assert(m);
3751
3752 if (MANAGER_IS_TEST_RUN(m) && !(m->test_run_flags & MANAGER_TEST_RUN_GENERATORS))
3753 return 0;
3754
3755 paths = generator_binary_paths(m->unit_file_scope);
3756 if (!paths)
3757 return log_oom();
3758
3759 if (!generator_path_any((const char* const*) paths))
3760 return 0;
3761
3762 r = lookup_paths_mkdir_generator(&m->lookup_paths);
3763 if (r < 0) {
3764 log_error_errno(r, "Failed to create generator directories: %m");
3765 goto finish;
3766 }
3767
3768 const char *argv[] = {
3769 NULL, /* Leave this empty, execute_directory() will fill something in */
3770 m->lookup_paths.generator,
3771 m->lookup_paths.generator_early,
3772 m->lookup_paths.generator_late,
3773 NULL,
3774 };
3775
3776 r = build_generator_environment(m, &ge);
3777 if (r < 0) {
3778 log_error_errno(r, "Failed to build generator environment: %m");
3779 goto finish;
3780 }
3781
3782 RUN_WITH_UMASK(0022)
3783 (void) execute_directories(
3784 (const char* const*) paths,
3785 DEFAULT_TIMEOUT_USEC,
3786 /* callbacks= */ NULL, /* callback_args= */ NULL,
3787 (char**) argv,
3788 ge,
3789 EXEC_DIR_PARALLEL | EXEC_DIR_IGNORE_ERRORS | EXEC_DIR_SET_SYSTEMD_EXEC_PID);
3790
3791 r = 0;
3792
3793 finish:
3794 lookup_paths_trim_generator(&m->lookup_paths);
3795 return r;
3796 }
3797
3798 int manager_transient_environment_add(Manager *m, char **plus) {
3799 char **a;
3800
3801 assert(m);
3802
3803 if (strv_isempty(plus))
3804 return 0;
3805
3806 a = strv_env_merge(m->transient_environment, plus);
3807 if (!a)
3808 return log_oom();
3809
3810 sanitize_environment(a);
3811
3812 return strv_free_and_replace(m->transient_environment, a);
3813 }
3814
3815 int manager_client_environment_modify(
3816 Manager *m,
3817 char **minus,
3818 char **plus) {
3819
3820 char **a = NULL, **b = NULL, **l;
3821
3822 assert(m);
3823
3824 if (strv_isempty(minus) && strv_isempty(plus))
3825 return 0;
3826
3827 l = m->client_environment;
3828
3829 if (!strv_isempty(minus)) {
3830 a = strv_env_delete(l, 1, minus);
3831 if (!a)
3832 return -ENOMEM;
3833
3834 l = a;
3835 }
3836
3837 if (!strv_isempty(plus)) {
3838 b = strv_env_merge(l, plus);
3839 if (!b) {
3840 strv_free(a);
3841 return -ENOMEM;
3842 }
3843
3844 l = b;
3845 }
3846
3847 if (m->client_environment != l)
3848 strv_free(m->client_environment);
3849
3850 if (a != l)
3851 strv_free(a);
3852 if (b != l)
3853 strv_free(b);
3854
3855 m->client_environment = sanitize_environment(l);
3856 return 0;
3857 }
3858
3859 int manager_get_effective_environment(Manager *m, char ***ret) {
3860 char **l;
3861
3862 assert(m);
3863 assert(ret);
3864
3865 l = strv_env_merge(m->transient_environment, m->client_environment);
3866 if (!l)
3867 return -ENOMEM;
3868
3869 *ret = l;
3870 return 0;
3871 }
3872
3873 int manager_set_default_smack_process_label(Manager *m, const char *label) {
3874 assert(m);
3875
3876 #ifdef SMACK_DEFAULT_PROCESS_LABEL
3877 if (!label)
3878 return free_and_strdup(&m->default_smack_process_label, SMACK_DEFAULT_PROCESS_LABEL);
3879 #endif
3880 if (streq_ptr(label, "/"))
3881 return free_and_strdup(&m->default_smack_process_label, NULL);
3882
3883 return free_and_strdup(&m->default_smack_process_label, label);
3884 }
3885
3886 int manager_set_default_rlimits(Manager *m, struct rlimit **default_rlimit) {
3887 assert(m);
3888
3889 for (unsigned i = 0; i < _RLIMIT_MAX; i++) {
3890 m->rlimit[i] = mfree(m->rlimit[i]);
3891
3892 if (!default_rlimit[i])
3893 continue;
3894
3895 m->rlimit[i] = newdup(struct rlimit, default_rlimit[i], 1);
3896 if (!m->rlimit[i])
3897 return log_oom();
3898 }
3899
3900 return 0;
3901 }
3902
3903 void manager_recheck_dbus(Manager *m) {
3904 assert(m);
3905
3906 /* Connects to the bus if the dbus service and socket are running. If we are running in user mode
3907 * this is all it does. In system mode we'll also connect to the system bus (which will most likely
3908 * just reuse the connection of the API bus). That's because the system bus after all runs as service
3909 * of the system instance, while in the user instance we can assume it's already there. */
3910
3911 if (MANAGER_IS_RELOADING(m))
3912 return; /* don't check while we are reloading… */
3913
3914 if (manager_dbus_is_running(m, false)) {
3915 (void) bus_init_api(m);
3916
3917 if (MANAGER_IS_SYSTEM(m))
3918 (void) bus_init_system(m);
3919 } else {
3920 (void) bus_done_api(m);
3921
3922 if (MANAGER_IS_SYSTEM(m))
3923 (void) bus_done_system(m);
3924 }
3925 }
3926
3927 static bool manager_journal_is_running(Manager *m) {
3928 Unit *u;
3929
3930 assert(m);
3931
3932 if (MANAGER_IS_TEST_RUN(m))
3933 return false;
3934
3935 /* If we are the user manager we can safely assume that the journal is up */
3936 if (!MANAGER_IS_SYSTEM(m))
3937 return true;
3938
3939 /* Check that the socket is not only up, but in RUNNING state */
3940 u = manager_get_unit(m, SPECIAL_JOURNALD_SOCKET);
3941 if (!u)
3942 return false;
3943 if (SOCKET(u)->state != SOCKET_RUNNING)
3944 return false;
3945
3946 /* Similar, check if the daemon itself is fully up, too */
3947 u = manager_get_unit(m, SPECIAL_JOURNALD_SERVICE);
3948 if (!u)
3949 return false;
3950 if (!IN_SET(SERVICE(u)->state, SERVICE_RELOAD, SERVICE_RUNNING))
3951 return false;
3952
3953 return true;
3954 }
3955
3956 void disable_printk_ratelimit(void) {
3957 /* Disable kernel's printk ratelimit.
3958 *
3959 * Logging to /dev/kmsg is most useful during early boot and shutdown, where normal logging
3960 * mechanisms are not available. The semantics of this sysctl are such that any kernel command-line
3961 * setting takes precedence. */
3962 int r;
3963
3964 r = sysctl_write("kernel/printk_devkmsg", "on");
3965 if (r < 0)
3966 log_debug_errno(r, "Failed to set sysctl kernel.printk_devkmsg=on: %m");
3967 }
3968
3969 void manager_recheck_journal(Manager *m) {
3970
3971 assert(m);
3972
3973 /* Don't bother with this unless we are in the special situation of being PID 1 */
3974 if (getpid_cached() != 1)
3975 return;
3976
3977 /* Don't check this while we are reloading, things might still change */
3978 if (MANAGER_IS_RELOADING(m))
3979 return;
3980
3981 /* The journal is fully and entirely up? If so, let's permit logging to it, if that's configured. If
3982 * the journal is down, don't ever log to it, otherwise we might end up deadlocking ourselves as we
3983 * might trigger an activation ourselves we can't fulfill. */
3984 log_set_prohibit_ipc(!manager_journal_is_running(m));
3985 log_open();
3986 }
3987
3988 static ShowStatus manager_get_show_status(Manager *m) {
3989 assert(m);
3990
3991 if (MANAGER_IS_USER(m))
3992 return _SHOW_STATUS_INVALID;
3993
3994 if (m->show_status_overridden != _SHOW_STATUS_INVALID)
3995 return m->show_status_overridden;
3996
3997 return m->show_status;
3998 }
3999
4000 bool manager_get_show_status_on(Manager *m) {
4001 assert(m);
4002
4003 return show_status_on(manager_get_show_status(m));
4004 }
4005
4006 static void set_show_status_marker(bool b) {
4007 if (b)
4008 (void) touch("/run/systemd/show-status");
4009 else
4010 (void) unlink("/run/systemd/show-status");
4011 }
4012
4013 void manager_set_show_status(Manager *m, ShowStatus mode, const char *reason) {
4014 assert(m);
4015 assert(reason);
4016 assert(mode >= 0 && mode < _SHOW_STATUS_MAX);
4017
4018 if (MANAGER_IS_USER(m))
4019 return;
4020
4021 if (mode == m->show_status)
4022 return;
4023
4024 if (m->show_status_overridden == _SHOW_STATUS_INVALID) {
4025 bool enabled;
4026
4027 enabled = show_status_on(mode);
4028 log_debug("%s (%s) showing of status (%s).",
4029 enabled ? "Enabling" : "Disabling",
4030 strna(show_status_to_string(mode)),
4031 reason);
4032
4033 set_show_status_marker(enabled);
4034 }
4035
4036 m->show_status = mode;
4037 }
4038
4039 void manager_override_show_status(Manager *m, ShowStatus mode, const char *reason) {
4040 assert(m);
4041 assert(mode < _SHOW_STATUS_MAX);
4042
4043 if (MANAGER_IS_USER(m))
4044 return;
4045
4046 if (mode == m->show_status_overridden)
4047 return;
4048
4049 m->show_status_overridden = mode;
4050
4051 if (mode == _SHOW_STATUS_INVALID)
4052 mode = m->show_status;
4053
4054 log_debug("%s (%s) showing of status (%s).",
4055 m->show_status_overridden != _SHOW_STATUS_INVALID ? "Overriding" : "Restoring",
4056 strna(show_status_to_string(mode)),
4057 reason);
4058
4059 set_show_status_marker(show_status_on(mode));
4060 }
4061
4062 const char *manager_get_confirm_spawn(Manager *m) {
4063 static int last_errno = 0;
4064 struct stat st;
4065 int r;
4066
4067 assert(m);
4068
4069 /* Here's the deal: we want to test the validity of the console but don't want
4070 * PID1 to go through the whole console process which might block. But we also
4071 * want to warn the user only once if something is wrong with the console so we
4072 * cannot do the sanity checks after spawning our children. So here we simply do
4073 * really basic tests to hopefully trap common errors.
4074 *
4075 * If the console suddenly disappear at the time our children will really it
4076 * then they will simply fail to acquire it and a positive answer will be
4077 * assumed. New children will fall back to /dev/console though.
4078 *
4079 * Note: TTYs are devices that can come and go any time, and frequently aren't
4080 * available yet during early boot (consider a USB rs232 dongle...). If for any
4081 * reason the configured console is not ready, we fall back to the default
4082 * console. */
4083
4084 if (!m->confirm_spawn || path_equal(m->confirm_spawn, "/dev/console"))
4085 return m->confirm_spawn;
4086
4087 if (stat(m->confirm_spawn, &st) < 0) {
4088 r = -errno;
4089 goto fail;
4090 }
4091
4092 if (!S_ISCHR(st.st_mode)) {
4093 r = -ENOTTY;
4094 goto fail;
4095 }
4096
4097 last_errno = 0;
4098 return m->confirm_spawn;
4099
4100 fail:
4101 if (last_errno != r)
4102 last_errno = log_warning_errno(r, "Failed to open %s, using default console: %m", m->confirm_spawn);
4103
4104 return "/dev/console";
4105 }
4106
4107 void manager_set_first_boot(Manager *m, bool b) {
4108 assert(m);
4109
4110 if (!MANAGER_IS_SYSTEM(m))
4111 return;
4112
4113 if (m->first_boot != (int) b) {
4114 if (b)
4115 (void) touch("/run/systemd/first-boot");
4116 else
4117 (void) unlink("/run/systemd/first-boot");
4118 }
4119
4120 m->first_boot = b;
4121 }
4122
4123 void manager_disable_confirm_spawn(void) {
4124 (void) touch("/run/systemd/confirm_spawn_disabled");
4125 }
4126
4127 bool manager_is_confirm_spawn_disabled(Manager *m) {
4128 if (!m->confirm_spawn)
4129 return true;
4130
4131 return access("/run/systemd/confirm_spawn_disabled", F_OK) >= 0;
4132 }
4133
4134 static bool manager_should_show_status(Manager *m, StatusType type) {
4135 assert(m);
4136
4137 if (!MANAGER_IS_SYSTEM(m))
4138 return false;
4139
4140 if (m->no_console_output)
4141 return false;
4142
4143 if (!IN_SET(manager_state(m), MANAGER_INITIALIZING, MANAGER_STARTING, MANAGER_STOPPING))
4144 return false;
4145
4146 /* If we cannot find out the status properly, just proceed. */
4147 if (type != STATUS_TYPE_EMERGENCY && manager_check_ask_password(m) > 0)
4148 return false;
4149
4150 if (type == STATUS_TYPE_NOTICE && m->show_status != SHOW_STATUS_NO)
4151 return true;
4152
4153 return manager_get_show_status_on(m);
4154 }
4155
4156 void manager_status_printf(Manager *m, StatusType type, const char *status, const char *format, ...) {
4157 va_list ap;
4158
4159 /* If m is NULL, assume we're after shutdown and let the messages through. */
4160
4161 if (m && !manager_should_show_status(m, type))
4162 return;
4163
4164 /* XXX We should totally drop the check for ephemeral here
4165 * and thus effectively make 'Type=idle' pointless. */
4166 if (type == STATUS_TYPE_EPHEMERAL && m && m->n_on_console > 0)
4167 return;
4168
4169 va_start(ap, format);
4170 status_vprintf(status, SHOW_STATUS_ELLIPSIZE|(type == STATUS_TYPE_EPHEMERAL ? SHOW_STATUS_EPHEMERAL : 0), format, ap);
4171 va_end(ap);
4172 }
4173
4174 Set* manager_get_units_requiring_mounts_for(Manager *m, const char *path) {
4175 assert(m);
4176 assert(path);
4177
4178 if (path_equal(path, "/"))
4179 path = "";
4180
4181 return hashmap_get(m->units_requiring_mounts_for, path);
4182 }
4183
4184 int manager_update_failed_units(Manager *m, Unit *u, bool failed) {
4185 unsigned size;
4186 int r;
4187
4188 assert(m);
4189 assert(u->manager == m);
4190
4191 size = set_size(m->failed_units);
4192
4193 if (failed) {
4194 r = set_ensure_put(&m->failed_units, NULL, u);
4195 if (r < 0)
4196 return log_oom();
4197 } else
4198 (void) set_remove(m->failed_units, u);
4199
4200 if (set_size(m->failed_units) != size)
4201 bus_manager_send_change_signal(m);
4202
4203 return 0;
4204 }
4205
4206 ManagerState manager_state(Manager *m) {
4207 Unit *u;
4208
4209 assert(m);
4210
4211 /* Is the special shutdown target active or queued? If so, we are in shutdown state */
4212 u = manager_get_unit(m, SPECIAL_SHUTDOWN_TARGET);
4213 if (u && unit_active_or_pending(u))
4214 return MANAGER_STOPPING;
4215
4216 /* Did we ever finish booting? If not then we are still starting up */
4217 if (!MANAGER_IS_FINISHED(m)) {
4218
4219 u = manager_get_unit(m, SPECIAL_BASIC_TARGET);
4220 if (!u || !UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u)))
4221 return MANAGER_INITIALIZING;
4222
4223 return MANAGER_STARTING;
4224 }
4225
4226 if (MANAGER_IS_SYSTEM(m)) {
4227 /* Are the rescue or emergency targets active or queued? If so we are in maintenance state */
4228 u = manager_get_unit(m, SPECIAL_RESCUE_TARGET);
4229 if (u && unit_active_or_pending(u))
4230 return MANAGER_MAINTENANCE;
4231
4232 u = manager_get_unit(m, SPECIAL_EMERGENCY_TARGET);
4233 if (u && unit_active_or_pending(u))
4234 return MANAGER_MAINTENANCE;
4235 }
4236
4237 /* Are there any failed units? If so, we are in degraded mode */
4238 if (set_size(m->failed_units) > 0)
4239 return MANAGER_DEGRADED;
4240
4241 return MANAGER_RUNNING;
4242 }
4243
4244 static void manager_unref_uid_internal(
4245 Hashmap *uid_refs,
4246 uid_t uid,
4247 bool destroy_now,
4248 int (*_clean_ipc)(uid_t uid)) {
4249
4250 uint32_t c, n;
4251
4252 assert(uid_is_valid(uid));
4253 assert(_clean_ipc);
4254
4255 /* A generic implementation, covering both manager_unref_uid() and manager_unref_gid(), under the
4256 * assumption that uid_t and gid_t are actually defined the same way, with the same validity rules.
4257 *
4258 * We store a hashmap where the key is the UID/GID and the value is a 32bit reference counter, whose
4259 * highest bit is used as flag for marking UIDs/GIDs whose IPC objects to remove when the last
4260 * reference to the UID/GID is dropped. The flag is set to on, once at least one reference from a
4261 * unit where RemoveIPC= is set is added on a UID/GID. It is reset when the UID's/GID's reference
4262 * counter drops to 0 again. */
4263
4264 assert_cc(sizeof(uid_t) == sizeof(gid_t));
4265 assert_cc(UID_INVALID == (uid_t) GID_INVALID);
4266
4267 if (uid == 0) /* We don't keep track of root, and will never destroy it */
4268 return;
4269
4270 c = PTR_TO_UINT32(hashmap_get(uid_refs, UID_TO_PTR(uid)));
4271
4272 n = c & ~DESTROY_IPC_FLAG;
4273 assert(n > 0);
4274 n--;
4275
4276 if (destroy_now && n == 0) {
4277 hashmap_remove(uid_refs, UID_TO_PTR(uid));
4278
4279 if (c & DESTROY_IPC_FLAG) {
4280 log_debug("%s " UID_FMT " is no longer referenced, cleaning up its IPC.",
4281 _clean_ipc == clean_ipc_by_uid ? "UID" : "GID",
4282 uid);
4283 (void) _clean_ipc(uid);
4284 }
4285 } else {
4286 c = n | (c & DESTROY_IPC_FLAG);
4287 assert_se(hashmap_update(uid_refs, UID_TO_PTR(uid), UINT32_TO_PTR(c)) >= 0);
4288 }
4289 }
4290
4291 void manager_unref_uid(Manager *m, uid_t uid, bool destroy_now) {
4292 manager_unref_uid_internal(m->uid_refs, uid, destroy_now, clean_ipc_by_uid);
4293 }
4294
4295 void manager_unref_gid(Manager *m, gid_t gid, bool destroy_now) {
4296 manager_unref_uid_internal(m->gid_refs, (uid_t) gid, destroy_now, clean_ipc_by_gid);
4297 }
4298
4299 static int manager_ref_uid_internal(
4300 Hashmap **uid_refs,
4301 uid_t uid,
4302 bool clean_ipc) {
4303
4304 uint32_t c, n;
4305 int r;
4306
4307 assert(uid_refs);
4308 assert(uid_is_valid(uid));
4309
4310 /* A generic implementation, covering both manager_ref_uid() and manager_ref_gid(), under the
4311 * assumption that uid_t and gid_t are actually defined the same way, with the same validity
4312 * rules. */
4313
4314 assert_cc(sizeof(uid_t) == sizeof(gid_t));
4315 assert_cc(UID_INVALID == (uid_t) GID_INVALID);
4316
4317 if (uid == 0) /* We don't keep track of root, and will never destroy it */
4318 return 0;
4319
4320 r = hashmap_ensure_allocated(uid_refs, &trivial_hash_ops);
4321 if (r < 0)
4322 return r;
4323
4324 c = PTR_TO_UINT32(hashmap_get(*uid_refs, UID_TO_PTR(uid)));
4325
4326 n = c & ~DESTROY_IPC_FLAG;
4327 n++;
4328
4329 if (n & DESTROY_IPC_FLAG) /* check for overflow */
4330 return -EOVERFLOW;
4331
4332 c = n | (c & DESTROY_IPC_FLAG) | (clean_ipc ? DESTROY_IPC_FLAG : 0);
4333
4334 return hashmap_replace(*uid_refs, UID_TO_PTR(uid), UINT32_TO_PTR(c));
4335 }
4336
4337 int manager_ref_uid(Manager *m, uid_t uid, bool clean_ipc) {
4338 return manager_ref_uid_internal(&m->uid_refs, uid, clean_ipc);
4339 }
4340
4341 int manager_ref_gid(Manager *m, gid_t gid, bool clean_ipc) {
4342 return manager_ref_uid_internal(&m->gid_refs, (uid_t) gid, clean_ipc);
4343 }
4344
4345 static void manager_vacuum_uid_refs_internal(
4346 Hashmap *uid_refs,
4347 int (*_clean_ipc)(uid_t uid)) {
4348
4349 void *p, *k;
4350
4351 assert(_clean_ipc);
4352
4353 HASHMAP_FOREACH_KEY(p, k, uid_refs) {
4354 uint32_t c, n;
4355 uid_t uid;
4356
4357 uid = PTR_TO_UID(k);
4358 c = PTR_TO_UINT32(p);
4359
4360 n = c & ~DESTROY_IPC_FLAG;
4361 if (n > 0)
4362 continue;
4363
4364 if (c & DESTROY_IPC_FLAG) {
4365 log_debug("Found unreferenced %s " UID_FMT " after reload/reexec. Cleaning up.",
4366 _clean_ipc == clean_ipc_by_uid ? "UID" : "GID",
4367 uid);
4368 (void) _clean_ipc(uid);
4369 }
4370
4371 assert_se(hashmap_remove(uid_refs, k) == p);
4372 }
4373 }
4374
4375 static void manager_vacuum_uid_refs(Manager *m) {
4376 manager_vacuum_uid_refs_internal(m->uid_refs, clean_ipc_by_uid);
4377 }
4378
4379 static void manager_vacuum_gid_refs(Manager *m) {
4380 manager_vacuum_uid_refs_internal(m->gid_refs, clean_ipc_by_gid);
4381 }
4382
4383 static void manager_vacuum(Manager *m) {
4384 assert(m);
4385
4386 /* Release any dynamic users no longer referenced */
4387 dynamic_user_vacuum(m, true);
4388
4389 /* Release any references to UIDs/GIDs no longer referenced, and destroy any IPC owned by them */
4390 manager_vacuum_uid_refs(m);
4391 manager_vacuum_gid_refs(m);
4392
4393 /* Release any runtimes no longer referenced */
4394 exec_runtime_vacuum(m);
4395 }
4396
4397 int manager_dispatch_user_lookup_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
4398 struct buffer {
4399 uid_t uid;
4400 gid_t gid;
4401 char unit_name[UNIT_NAME_MAX+1];
4402 } _packed_ buffer;
4403
4404 Manager *m = userdata;
4405 ssize_t l;
4406 size_t n;
4407 Unit *u;
4408
4409 assert_se(source);
4410 assert_se(m);
4411
4412 /* Invoked whenever a child process succeeded resolving its user/group to use and sent us the
4413 * resulting UID/GID in a datagram. We parse the datagram here and pass it off to the unit, so that
4414 * it can add a reference to the UID/GID so that it can destroy the UID/GID's IPC objects when the
4415 * reference counter drops to 0. */
4416
4417 l = recv(fd, &buffer, sizeof(buffer), MSG_DONTWAIT);
4418 if (l < 0) {
4419 if (ERRNO_IS_TRANSIENT(errno))
4420 return 0;
4421
4422 return log_error_errno(errno, "Failed to read from user lookup fd: %m");
4423 }
4424
4425 if ((size_t) l <= offsetof(struct buffer, unit_name)) {
4426 log_warning("Received too short user lookup message, ignoring.");
4427 return 0;
4428 }
4429
4430 if ((size_t) l > offsetof(struct buffer, unit_name) + UNIT_NAME_MAX) {
4431 log_warning("Received too long user lookup message, ignoring.");
4432 return 0;
4433 }
4434
4435 if (!uid_is_valid(buffer.uid) && !gid_is_valid(buffer.gid)) {
4436 log_warning("Got user lookup message with invalid UID/GID pair, ignoring.");
4437 return 0;
4438 }
4439
4440 n = (size_t) l - offsetof(struct buffer, unit_name);
4441 if (memchr(buffer.unit_name, 0, n)) {
4442 log_warning("Received lookup message with embedded NUL character, ignoring.");
4443 return 0;
4444 }
4445
4446 buffer.unit_name[n] = 0;
4447 u = manager_get_unit(m, buffer.unit_name);
4448 if (!u) {
4449 log_debug("Got user lookup message but unit doesn't exist, ignoring.");
4450 return 0;
4451 }
4452
4453 log_unit_debug(u, "User lookup succeeded: uid=" UID_FMT " gid=" GID_FMT, buffer.uid, buffer.gid);
4454
4455 unit_notify_user_lookup(u, buffer.uid, buffer.gid);
4456 return 0;
4457 }
4458
4459 static int short_uid_range(const char *path) {
4460 _cleanup_free_ UidRange *p = NULL;
4461 size_t n = 0;
4462 int r;
4463
4464 assert(path);
4465
4466 /* Taint systemd if we the UID range assigned to this environment doesn't at least cover 0…65534,
4467 * i.e. from root to nobody. */
4468
4469 r = uid_range_load_userns(&p, &n, path);
4470 if (ERRNO_IS_NOT_SUPPORTED(r))
4471 return false;
4472 if (r < 0)
4473 return log_debug_errno(r, "Failed to load %s: %m", path);
4474
4475 return !uid_range_covers(p, n, 0, 65535);
4476 }
4477
4478 char* manager_taint_string(const Manager *m) {
4479 /* Returns a "taint string", e.g. "local-hwclock:var-run-bad". Only things that are detected at
4480 * runtime should be tagged here. For stuff that is known during compilation, emit a warning in the
4481 * configuration phase. */
4482
4483 assert(m);
4484
4485 const char* stage[13] = {};
4486 size_t n = 0;
4487
4488 if (m->taint_usr)
4489 stage[n++] = "split-usr";
4490
4491 _cleanup_free_ char *usrbin = NULL;
4492 if (readlink_malloc("/bin", &usrbin) < 0 || !PATH_IN_SET(usrbin, "usr/bin", "/usr/bin"))
4493 stage[n++] = "unmerged-usr";
4494
4495 if (access("/proc/cgroups", F_OK) < 0)
4496 stage[n++] = "cgroups-missing";
4497
4498 if (cg_all_unified() == 0)
4499 stage[n++] = "cgroupsv1";
4500
4501 if (clock_is_localtime(NULL) > 0)
4502 stage[n++] = "local-hwclock";
4503
4504 if (os_release_support_ended(NULL, true) > 0)
4505 stage[n++] = "support-ended";
4506
4507 _cleanup_free_ char *destination = NULL;
4508 if (readlink_malloc("/var/run", &destination) < 0 ||
4509 !PATH_IN_SET(destination, "../run", "/run"))
4510 stage[n++] = "var-run-bad";
4511
4512 _cleanup_free_ char *overflowuid = NULL, *overflowgid = NULL;
4513 if (read_one_line_file("/proc/sys/kernel/overflowuid", &overflowuid) >= 0 &&
4514 !streq(overflowuid, "65534"))
4515 stage[n++] = "overflowuid-not-65534";
4516 if (read_one_line_file("/proc/sys/kernel/overflowgid", &overflowgid) >= 0 &&
4517 !streq(overflowgid, "65534"))
4518 stage[n++] = "overflowgid-not-65534";
4519
4520 struct utsname uts;
4521 assert_se(uname(&uts) >= 0);
4522 if (strverscmp_improved(uts.release, KERNEL_BASELINE_VERSION) < 0)
4523 stage[n++] = "old-kernel";
4524
4525 if (short_uid_range("/proc/self/uid_map") > 0)
4526 stage[n++] = "short-uid-range";
4527 if (short_uid_range("/proc/self/gid_map") > 0)
4528 stage[n++] = "short-gid-range";
4529
4530 assert(n < ELEMENTSOF(stage) - 1); /* One extra for NULL terminator */
4531
4532 return strv_join((char**) stage, ":");
4533 }
4534
4535 void manager_ref_console(Manager *m) {
4536 assert(m);
4537
4538 m->n_on_console++;
4539 }
4540
4541 void manager_unref_console(Manager *m) {
4542
4543 assert(m->n_on_console > 0);
4544 m->n_on_console--;
4545
4546 if (m->n_on_console == 0)
4547 m->no_console_output = false; /* unset no_console_output flag, since the console is definitely free now */
4548 }
4549
4550 void manager_override_log_level(Manager *m, int level) {
4551 _cleanup_free_ char *s = NULL;
4552 assert(m);
4553
4554 if (!m->log_level_overridden) {
4555 m->original_log_level = log_get_max_level();
4556 m->log_level_overridden = true;
4557 }
4558
4559 (void) log_level_to_string_alloc(level, &s);
4560 log_info("Setting log level to %s.", strna(s));
4561
4562 log_set_max_level(level);
4563 }
4564
4565 void manager_restore_original_log_level(Manager *m) {
4566 _cleanup_free_ char *s = NULL;
4567 assert(m);
4568
4569 if (!m->log_level_overridden)
4570 return;
4571
4572 (void) log_level_to_string_alloc(m->original_log_level, &s);
4573 log_info("Restoring log level to original (%s).", strna(s));
4574
4575 log_set_max_level(m->original_log_level);
4576 m->log_level_overridden = false;
4577 }
4578
4579 void manager_override_log_target(Manager *m, LogTarget target) {
4580 assert(m);
4581
4582 if (!m->log_target_overridden) {
4583 m->original_log_target = log_get_target();
4584 m->log_target_overridden = true;
4585 }
4586
4587 log_info("Setting log target to %s.", log_target_to_string(target));
4588 log_set_target(target);
4589 }
4590
4591 void manager_restore_original_log_target(Manager *m) {
4592 assert(m);
4593
4594 if (!m->log_target_overridden)
4595 return;
4596
4597 log_info("Restoring log target to original %s.", log_target_to_string(m->original_log_target));
4598
4599 log_set_target(m->original_log_target);
4600 m->log_target_overridden = false;
4601 }
4602
4603 ManagerTimestamp manager_timestamp_initrd_mangle(ManagerTimestamp s) {
4604 if (in_initrd() &&
4605 s >= MANAGER_TIMESTAMP_SECURITY_START &&
4606 s <= MANAGER_TIMESTAMP_UNITS_LOAD_FINISH)
4607 return s - MANAGER_TIMESTAMP_SECURITY_START + MANAGER_TIMESTAMP_INITRD_SECURITY_START;
4608 return s;
4609 }
4610
4611 static const char *const manager_state_table[_MANAGER_STATE_MAX] = {
4612 [MANAGER_INITIALIZING] = "initializing",
4613 [MANAGER_STARTING] = "starting",
4614 [MANAGER_RUNNING] = "running",
4615 [MANAGER_DEGRADED] = "degraded",
4616 [MANAGER_MAINTENANCE] = "maintenance",
4617 [MANAGER_STOPPING] = "stopping",
4618 };
4619
4620 DEFINE_STRING_TABLE_LOOKUP(manager_state, ManagerState);
4621
4622 static const char *const manager_timestamp_table[_MANAGER_TIMESTAMP_MAX] = {
4623 [MANAGER_TIMESTAMP_FIRMWARE] = "firmware",
4624 [MANAGER_TIMESTAMP_LOADER] = "loader",
4625 [MANAGER_TIMESTAMP_KERNEL] = "kernel",
4626 [MANAGER_TIMESTAMP_INITRD] = "initrd",
4627 [MANAGER_TIMESTAMP_USERSPACE] = "userspace",
4628 [MANAGER_TIMESTAMP_FINISH] = "finish",
4629 [MANAGER_TIMESTAMP_SECURITY_START] = "security-start",
4630 [MANAGER_TIMESTAMP_SECURITY_FINISH] = "security-finish",
4631 [MANAGER_TIMESTAMP_GENERATORS_START] = "generators-start",
4632 [MANAGER_TIMESTAMP_GENERATORS_FINISH] = "generators-finish",
4633 [MANAGER_TIMESTAMP_UNITS_LOAD_START] = "units-load-start",
4634 [MANAGER_TIMESTAMP_UNITS_LOAD_FINISH] = "units-load-finish",
4635 [MANAGER_TIMESTAMP_UNITS_LOAD] = "units-load",
4636 [MANAGER_TIMESTAMP_INITRD_SECURITY_START] = "initrd-security-start",
4637 [MANAGER_TIMESTAMP_INITRD_SECURITY_FINISH] = "initrd-security-finish",
4638 [MANAGER_TIMESTAMP_INITRD_GENERATORS_START] = "initrd-generators-start",
4639 [MANAGER_TIMESTAMP_INITRD_GENERATORS_FINISH] = "initrd-generators-finish",
4640 [MANAGER_TIMESTAMP_INITRD_UNITS_LOAD_START] = "initrd-units-load-start",
4641 [MANAGER_TIMESTAMP_INITRD_UNITS_LOAD_FINISH] = "initrd-units-load-finish",
4642 };
4643
4644 DEFINE_STRING_TABLE_LOOKUP(manager_timestamp, ManagerTimestamp);
4645
4646 static const char* const oom_policy_table[_OOM_POLICY_MAX] = {
4647 [OOM_CONTINUE] = "continue",
4648 [OOM_STOP] = "stop",
4649 [OOM_KILL] = "kill",
4650 };
4651
4652 DEFINE_STRING_TABLE_LOOKUP(oom_policy, OOMPolicy);
Unnamed repository; edit this file 'description' to name the repository.