]> 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
Merge pull request #24044 from dtardon/default-device-timeout
[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 = userdata;
312
313 assert(m);
314
315 (void) flush_fd(fd);
316
317 m->have_ask_password = have_ask_password();
318 if (m->have_ask_password < 0)
319 /* Log error but continue. Negative have_ask_password
320 * is treated as unknown status. */
321 log_error_errno(m->have_ask_password, "Failed to list /run/systemd/ask-password: %m");
322
323 return 0;
324 }
325
326 static void manager_close_ask_password(Manager *m) {
327 assert(m);
328
329 m->ask_password_event_source = sd_event_source_disable_unref(m->ask_password_event_source);
330 m->ask_password_inotify_fd = safe_close(m->ask_password_inotify_fd);
331 m->have_ask_password = -EINVAL;
332 }
333
334 static int manager_check_ask_password(Manager *m) {
335 int r;
336
337 assert(m);
338
339 if (!m->ask_password_event_source) {
340 assert(m->ask_password_inotify_fd < 0);
341
342 (void) mkdir_p_label("/run/systemd/ask-password", 0755);
343
344 m->ask_password_inotify_fd = inotify_init1(IN_NONBLOCK|IN_CLOEXEC);
345 if (m->ask_password_inotify_fd < 0)
346 return log_error_errno(errno, "Failed to create inotify object: %m");
347
348 r = inotify_add_watch_and_warn(m->ask_password_inotify_fd,
349 "/run/systemd/ask-password",
350 IN_CREATE|IN_DELETE|IN_MOVE);
351 if (r < 0) {
352 manager_close_ask_password(m);
353 return r;
354 }
355
356 r = sd_event_add_io(m->event, &m->ask_password_event_source,
357 m->ask_password_inotify_fd, EPOLLIN,
358 manager_dispatch_ask_password_fd, m);
359 if (r < 0) {
360 log_error_errno(r, "Failed to add event source for /run/systemd/ask-password: %m");
361 manager_close_ask_password(m);
362 return r;
363 }
364
365 (void) sd_event_source_set_description(m->ask_password_event_source, "manager-ask-password");
366
367 /* Queries might have been added meanwhile... */
368 manager_dispatch_ask_password_fd(m->ask_password_event_source,
369 m->ask_password_inotify_fd, EPOLLIN, m);
370 }
371
372 return m->have_ask_password;
373 }
374
375 static int manager_watch_idle_pipe(Manager *m) {
376 int r;
377
378 assert(m);
379
380 if (m->idle_pipe_event_source)
381 return 0;
382
383 if (m->idle_pipe[2] < 0)
384 return 0;
385
386 r = sd_event_add_io(m->event, &m->idle_pipe_event_source, m->idle_pipe[2], EPOLLIN, manager_dispatch_idle_pipe_fd, m);
387 if (r < 0)
388 return log_error_errno(r, "Failed to watch idle pipe: %m");
389
390 (void) sd_event_source_set_description(m->idle_pipe_event_source, "manager-idle-pipe");
391
392 return 0;
393 }
394
395 static void manager_close_idle_pipe(Manager *m) {
396 assert(m);
397
398 m->idle_pipe_event_source = sd_event_source_disable_unref(m->idle_pipe_event_source);
399
400 safe_close_pair(m->idle_pipe);
401 safe_close_pair(m->idle_pipe + 2);
402 }
403
404 static int manager_setup_time_change(Manager *m) {
405 int r;
406
407 assert(m);
408
409 if (MANAGER_IS_TEST_RUN(m))
410 return 0;
411
412 m->time_change_event_source = sd_event_source_disable_unref(m->time_change_event_source);
413
414 r = event_add_time_change(m->event, &m->time_change_event_source, manager_dispatch_time_change_fd, m);
415 if (r < 0)
416 return log_error_errno(r, "Failed to create time change event source: %m");
417
418 /* Schedule this slightly earlier than the .timer event sources */
419 r = sd_event_source_set_priority(m->time_change_event_source, SD_EVENT_PRIORITY_NORMAL-1);
420 if (r < 0)
421 return log_error_errno(r, "Failed to set priority of time change event sources: %m");
422
423 log_debug("Set up TFD_TIMER_CANCEL_ON_SET timerfd.");
424
425 return 0;
426 }
427
428 static int manager_read_timezone_stat(Manager *m) {
429 struct stat st;
430 bool changed;
431
432 assert(m);
433
434 /* Read the current stat() data of /etc/localtime so that we detect changes */
435 if (lstat("/etc/localtime", &st) < 0) {
436 log_debug_errno(errno, "Failed to stat /etc/localtime, ignoring: %m");
437 changed = m->etc_localtime_accessible;
438 m->etc_localtime_accessible = false;
439 } else {
440 usec_t k;
441
442 k = timespec_load(&st.st_mtim);
443 changed = !m->etc_localtime_accessible || k != m->etc_localtime_mtime;
444
445 m->etc_localtime_mtime = k;
446 m->etc_localtime_accessible = true;
447 }
448
449 return changed;
450 }
451
452 static int manager_setup_timezone_change(Manager *m) {
453 _cleanup_(sd_event_source_unrefp) sd_event_source *new_event = NULL;
454 int r;
455
456 assert(m);
457
458 if (MANAGER_IS_TEST_RUN(m))
459 return 0;
460
461 /* We watch /etc/localtime for three events: change of the link count (which might mean removal from /etc even
462 * though another link might be kept), renames, and file close operations after writing. Note we don't bother
463 * with IN_DELETE_SELF, as that would just report when the inode is removed entirely, i.e. after the link count
464 * went to zero and all fds to it are closed.
465 *
466 * Note that we never follow symlinks here. This is a simplification, but should cover almost all cases
467 * correctly.
468 *
469 * Note that we create the new event source first here, before releasing the old one. This should optimize
470 * behaviour as this way sd-event can reuse the old watch in case the inode didn't change. */
471
472 r = sd_event_add_inotify(m->event, &new_event, "/etc/localtime",
473 IN_ATTRIB|IN_MOVE_SELF|IN_CLOSE_WRITE|IN_DONT_FOLLOW, manager_dispatch_timezone_change, m);
474 if (r == -ENOENT) {
475 /* If the file doesn't exist yet, subscribe to /etc instead, and wait until it is created either by
476 * O_CREATE or by rename() */
477
478 log_debug_errno(r, "/etc/localtime doesn't exist yet, watching /etc instead.");
479 r = sd_event_add_inotify(m->event, &new_event, "/etc",
480 IN_CREATE|IN_MOVED_TO|IN_ONLYDIR, manager_dispatch_timezone_change, m);
481 }
482 if (r < 0)
483 return log_error_errno(r, "Failed to create timezone change event source: %m");
484
485 /* Schedule this slightly earlier than the .timer event sources */
486 r = sd_event_source_set_priority(new_event, SD_EVENT_PRIORITY_NORMAL-1);
487 if (r < 0)
488 return log_error_errno(r, "Failed to set priority of timezone change event sources: %m");
489
490 sd_event_source_unref(m->timezone_change_event_source);
491 m->timezone_change_event_source = TAKE_PTR(new_event);
492
493 return 0;
494 }
495
496 static int enable_special_signals(Manager *m) {
497 _cleanup_close_ int fd = -1;
498
499 assert(m);
500
501 if (MANAGER_IS_TEST_RUN(m))
502 return 0;
503
504 /* Enable that we get SIGINT on control-alt-del. In containers
505 * this will fail with EPERM (older) or EINVAL (newer), so
506 * ignore that. */
507 if (reboot(RB_DISABLE_CAD) < 0 && !IN_SET(errno, EPERM, EINVAL))
508 log_warning_errno(errno, "Failed to enable ctrl-alt-del handling: %m");
509
510 fd = open_terminal("/dev/tty0", O_RDWR|O_NOCTTY|O_CLOEXEC);
511 if (fd < 0) {
512 /* Support systems without virtual console */
513 if (fd != -ENOENT)
514 log_warning_errno(errno, "Failed to open /dev/tty0: %m");
515 } else {
516 /* Enable that we get SIGWINCH on kbrequest */
517 if (ioctl(fd, KDSIGACCEPT, SIGWINCH) < 0)
518 log_warning_errno(errno, "Failed to enable kbrequest handling: %m");
519 }
520
521 return 0;
522 }
523
524 #define RTSIG_IF_AVAILABLE(signum) (signum <= SIGRTMAX ? signum : -1)
525
526 static int manager_setup_signals(Manager *m) {
527 struct sigaction sa = {
528 .sa_handler = SIG_DFL,
529 .sa_flags = SA_NOCLDSTOP|SA_RESTART,
530 };
531 sigset_t mask;
532 int r;
533
534 assert(m);
535
536 assert_se(sigaction(SIGCHLD, &sa, NULL) == 0);
537
538 /* We make liberal use of realtime signals here. On
539 * Linux/glibc we have 30 of them (with the exception of Linux
540 * on hppa, see below), between SIGRTMIN+0 ... SIGRTMIN+30
541 * (aka SIGRTMAX). */
542
543 assert_se(sigemptyset(&mask) == 0);
544 sigset_add_many(&mask,
545 SIGCHLD, /* Child died */
546 SIGTERM, /* Reexecute daemon */
547 SIGHUP, /* Reload configuration */
548 SIGUSR1, /* systemd: reconnect to D-Bus */
549 SIGUSR2, /* systemd: dump status */
550 SIGINT, /* Kernel sends us this on control-alt-del */
551 SIGWINCH, /* Kernel sends us this on kbrequest (alt-arrowup) */
552 SIGPWR, /* Some kernel drivers and upsd send us this on power failure */
553
554 SIGRTMIN+0, /* systemd: start default.target */
555 SIGRTMIN+1, /* systemd: isolate rescue.target */
556 SIGRTMIN+2, /* systemd: isolate emergency.target */
557 SIGRTMIN+3, /* systemd: start halt.target */
558 SIGRTMIN+4, /* systemd: start poweroff.target */
559 SIGRTMIN+5, /* systemd: start reboot.target */
560 SIGRTMIN+6, /* systemd: start kexec.target */
561
562 /* ... space for more special targets ... */
563
564 SIGRTMIN+13, /* systemd: Immediate halt */
565 SIGRTMIN+14, /* systemd: Immediate poweroff */
566 SIGRTMIN+15, /* systemd: Immediate reboot */
567 SIGRTMIN+16, /* systemd: Immediate kexec */
568
569 /* ... space for more immediate system state changes ... */
570
571 SIGRTMIN+20, /* systemd: enable status messages */
572 SIGRTMIN+21, /* systemd: disable status messages */
573 SIGRTMIN+22, /* systemd: set log level to LOG_DEBUG */
574 SIGRTMIN+23, /* systemd: set log level to LOG_INFO */
575 SIGRTMIN+24, /* systemd: Immediate exit (--user only) */
576 SIGRTMIN+25, /* systemd: reexecute manager */
577
578 /* Apparently Linux on hppa had fewer RT signals until v3.18,
579 * SIGRTMAX was SIGRTMIN+25, and then SIGRTMIN was lowered,
580 * see commit v3.17-7614-g1f25df2eff.
581 *
582 * We cannot unconditionally make use of those signals here,
583 * so let's use a runtime check. Since these commands are
584 * accessible by different means and only really a safety
585 * net, the missing functionality on hppa shouldn't matter.
586 */
587
588 RTSIG_IF_AVAILABLE(SIGRTMIN+26), /* systemd: set log target to journal-or-kmsg */
589 RTSIG_IF_AVAILABLE(SIGRTMIN+27), /* systemd: set log target to console */
590 RTSIG_IF_AVAILABLE(SIGRTMIN+28), /* systemd: set log target to kmsg */
591 RTSIG_IF_AVAILABLE(SIGRTMIN+29), /* systemd: set log target to syslog-or-kmsg (obsolete) */
592
593 /* ... one free signal here SIGRTMIN+30 ... */
594 -1);
595 assert_se(sigprocmask(SIG_SETMASK, &mask, NULL) == 0);
596
597 m->signal_fd = signalfd(-1, &mask, SFD_NONBLOCK|SFD_CLOEXEC);
598 if (m->signal_fd < 0)
599 return -errno;
600
601 r = sd_event_add_io(m->event, &m->signal_event_source, m->signal_fd, EPOLLIN, manager_dispatch_signal_fd, m);
602 if (r < 0)
603 return r;
604
605 (void) sd_event_source_set_description(m->signal_event_source, "manager-signal");
606
607 /* Process signals a bit earlier than the rest of things, but later than notify_fd processing, so that the
608 * notify processing can still figure out to which process/service a message belongs, before we reap the
609 * process. Also, process this before handling cgroup notifications, so that we always collect child exit
610 * status information before detecting that there's no process in a cgroup. */
611 r = sd_event_source_set_priority(m->signal_event_source, SD_EVENT_PRIORITY_NORMAL-6);
612 if (r < 0)
613 return r;
614
615 if (MANAGER_IS_SYSTEM(m))
616 return enable_special_signals(m);
617
618 return 0;
619 }
620
621 static char** sanitize_environment(char **l) {
622
623 /* Let's remove some environment variables that we need ourselves to communicate with our clients */
624 strv_env_unset_many(
625 l,
626 "CACHE_DIRECTORY",
627 "CONFIGURATION_DIRECTORY",
628 "CREDENTIALS_DIRECTORY",
629 "EXIT_CODE",
630 "EXIT_STATUS",
631 "INVOCATION_ID",
632 "JOURNAL_STREAM",
633 "LISTEN_FDNAMES",
634 "LISTEN_FDS",
635 "LISTEN_PID",
636 "LOGS_DIRECTORY",
637 "MAINPID",
638 "MANAGERPID",
639 "NOTIFY_SOCKET",
640 "PIDFILE",
641 "REMOTE_ADDR",
642 "REMOTE_PORT",
643 "RUNTIME_DIRECTORY",
644 "SERVICE_RESULT",
645 "STATE_DIRECTORY",
646 "WATCHDOG_PID",
647 "WATCHDOG_USEC",
648 NULL);
649
650 /* Let's order the environment alphabetically, just to make it pretty */
651 return strv_sort(l);
652 }
653
654 int manager_default_environment(Manager *m) {
655 int r;
656
657 assert(m);
658
659 m->transient_environment = strv_free(m->transient_environment);
660
661 if (MANAGER_IS_SYSTEM(m)) {
662 /* The system manager always starts with a clean
663 * environment for its children. It does not import
664 * the kernel's or the parents' exported variables.
665 *
666 * The initial passed environment is untouched to keep
667 * /proc/self/environ valid; it is used for tagging
668 * the init process inside containers. */
669 m->transient_environment = strv_new("PATH=" DEFAULT_PATH);
670 if (!m->transient_environment)
671 return log_oom();
672
673 /* Import locale variables LC_*= from configuration */
674 (void) locale_setup(&m->transient_environment);
675 } else {
676 /* The user manager passes its own environment along to its children, except for $PATH. */
677 m->transient_environment = strv_copy(environ);
678 if (!m->transient_environment)
679 return log_oom();
680
681 r = strv_env_replace_strdup(&m->transient_environment, "PATH=" DEFAULT_USER_PATH);
682 if (r < 0)
683 return log_oom();
684 }
685
686 sanitize_environment(m->transient_environment);
687
688 return 0;
689 }
690
691 static int manager_setup_prefix(Manager *m) {
692 struct table_entry {
693 uint64_t type;
694 const char *suffix;
695 };
696
697 static const struct table_entry paths_system[_EXEC_DIRECTORY_TYPE_MAX] = {
698 [EXEC_DIRECTORY_RUNTIME] = { SD_PATH_SYSTEM_RUNTIME, NULL },
699 [EXEC_DIRECTORY_STATE] = { SD_PATH_SYSTEM_STATE_PRIVATE, NULL },
700 [EXEC_DIRECTORY_CACHE] = { SD_PATH_SYSTEM_STATE_CACHE, NULL },
701 [EXEC_DIRECTORY_LOGS] = { SD_PATH_SYSTEM_STATE_LOGS, NULL },
702 [EXEC_DIRECTORY_CONFIGURATION] = { SD_PATH_SYSTEM_CONFIGURATION, NULL },
703 };
704
705 static const struct table_entry paths_user[_EXEC_DIRECTORY_TYPE_MAX] = {
706 [EXEC_DIRECTORY_RUNTIME] = { SD_PATH_USER_RUNTIME, NULL },
707 [EXEC_DIRECTORY_STATE] = { SD_PATH_USER_CONFIGURATION, NULL },
708 [EXEC_DIRECTORY_CACHE] = { SD_PATH_USER_STATE_CACHE, NULL },
709 [EXEC_DIRECTORY_LOGS] = { SD_PATH_USER_CONFIGURATION, "log" },
710 [EXEC_DIRECTORY_CONFIGURATION] = { SD_PATH_USER_CONFIGURATION, NULL },
711 };
712
713 assert(m);
714
715 const struct table_entry *p = MANAGER_IS_SYSTEM(m) ? paths_system : paths_user;
716 int r;
717
718 for (ExecDirectoryType i = 0; i < _EXEC_DIRECTORY_TYPE_MAX; i++) {
719 r = sd_path_lookup(p[i].type, p[i].suffix, &m->prefix[i]);
720 if (r < 0)
721 return log_warning_errno(r, "Failed to lookup %s path: %m",
722 exec_directory_type_to_string(i));
723 }
724
725 return 0;
726 }
727
728 static void manager_free_unit_name_maps(Manager *m) {
729 m->unit_id_map = hashmap_free(m->unit_id_map);
730 m->unit_name_map = hashmap_free(m->unit_name_map);
731 m->unit_path_cache = set_free(m->unit_path_cache);
732 m->unit_cache_timestamp_hash = 0;
733 }
734
735 static int manager_setup_run_queue(Manager *m) {
736 int r;
737
738 assert(m);
739 assert(!m->run_queue_event_source);
740
741 r = sd_event_add_defer(m->event, &m->run_queue_event_source, manager_dispatch_run_queue, m);
742 if (r < 0)
743 return r;
744
745 r = sd_event_source_set_priority(m->run_queue_event_source, SD_EVENT_PRIORITY_IDLE);
746 if (r < 0)
747 return r;
748
749 r = sd_event_source_set_enabled(m->run_queue_event_source, SD_EVENT_OFF);
750 if (r < 0)
751 return r;
752
753 (void) sd_event_source_set_description(m->run_queue_event_source, "manager-run-queue");
754
755 return 0;
756 }
757
758 static int manager_setup_sigchld_event_source(Manager *m) {
759 int r;
760
761 assert(m);
762 assert(!m->sigchld_event_source);
763
764 r = sd_event_add_defer(m->event, &m->sigchld_event_source, manager_dispatch_sigchld, m);
765 if (r < 0)
766 return r;
767
768 r = sd_event_source_set_priority(m->sigchld_event_source, SD_EVENT_PRIORITY_NORMAL-7);
769 if (r < 0)
770 return r;
771
772 r = sd_event_source_set_enabled(m->sigchld_event_source, SD_EVENT_OFF);
773 if (r < 0)
774 return r;
775
776 (void) sd_event_source_set_description(m->sigchld_event_source, "manager-sigchld");
777
778 return 0;
779 }
780
781 static int manager_find_credentials_dirs(Manager *m) {
782 const char *e;
783 int r;
784
785 assert(m);
786
787 r = get_credentials_dir(&e);
788 if (r < 0) {
789 if (r != -ENXIO)
790 log_debug_errno(r, "Failed to determine credentials directory, ignoring: %m");
791 } else {
792 m->received_credentials_directory = strdup(e);
793 if (!m->received_credentials_directory)
794 return -ENOMEM;
795 }
796
797 r = get_encrypted_credentials_dir(&e);
798 if (r < 0) {
799 if (r != -ENXIO)
800 log_debug_errno(r, "Failed to determine encrypted credentials directory, ignoring: %m");
801 } else {
802 m->received_encrypted_credentials_directory = strdup(e);
803 if (!m->received_encrypted_credentials_directory)
804 return -ENOMEM;
805 }
806
807 return 0;
808 }
809
810 void manager_set_switching_root(Manager *m, bool switching_root) {
811 m->switching_root = MANAGER_IS_SYSTEM(m) && switching_root;
812 }
813
814 int manager_new(LookupScope scope, ManagerTestRunFlags test_run_flags, Manager **_m) {
815 _cleanup_(manager_freep) Manager *m = NULL;
816 int r;
817
818 assert(_m);
819 assert(IN_SET(scope, LOOKUP_SCOPE_SYSTEM, LOOKUP_SCOPE_USER));
820
821 m = new(Manager, 1);
822 if (!m)
823 return -ENOMEM;
824
825 *m = (Manager) {
826 .unit_file_scope = scope,
827 .objective = _MANAGER_OBJECTIVE_INVALID,
828
829 .status_unit_format = STATUS_UNIT_FORMAT_DEFAULT,
830
831 .default_timer_accuracy_usec = USEC_PER_MINUTE,
832 .default_memory_accounting = MEMORY_ACCOUNTING_DEFAULT,
833 .default_tasks_accounting = true,
834 .default_tasks_max = TASKS_MAX_UNSET,
835 .default_timeout_start_usec = DEFAULT_TIMEOUT_USEC,
836 .default_timeout_stop_usec = DEFAULT_TIMEOUT_USEC,
837 .default_restart_usec = DEFAULT_RESTART_USEC,
838 .default_device_timeout_usec = DEFAULT_TIMEOUT_USEC,
839
840 .original_log_level = -1,
841 .original_log_target = _LOG_TARGET_INVALID,
842
843 .watchdog_overridden[WATCHDOG_RUNTIME] = USEC_INFINITY,
844 .watchdog_overridden[WATCHDOG_REBOOT] = USEC_INFINITY,
845 .watchdog_overridden[WATCHDOG_KEXEC] = USEC_INFINITY,
846 .watchdog_overridden[WATCHDOG_PRETIMEOUT] = USEC_INFINITY,
847
848 .show_status_overridden = _SHOW_STATUS_INVALID,
849
850 .notify_fd = -1,
851 .cgroups_agent_fd = -1,
852 .signal_fd = -1,
853 .user_lookup_fds = { -1, -1 },
854 .private_listen_fd = -1,
855 .dev_autofs_fd = -1,
856 .cgroup_inotify_fd = -1,
857 .pin_cgroupfs_fd = -1,
858 .ask_password_inotify_fd = -1,
859 .idle_pipe = { -1, -1, -1, -1},
860
861 /* start as id #1, so that we can leave #0 around as "null-like" value */
862 .current_job_id = 1,
863
864 .have_ask_password = -EINVAL, /* we don't know */
865 .first_boot = -1,
866 .test_run_flags = test_run_flags,
867
868 .default_oom_policy = OOM_STOP,
869 };
870
871 #if ENABLE_EFI
872 if (MANAGER_IS_SYSTEM(m) && detect_container() <= 0)
873 boot_timestamps(m->timestamps + MANAGER_TIMESTAMP_USERSPACE,
874 m->timestamps + MANAGER_TIMESTAMP_FIRMWARE,
875 m->timestamps + MANAGER_TIMESTAMP_LOADER);
876 #endif
877
878 /* Prepare log fields we can use for structured logging */
879 if (MANAGER_IS_SYSTEM(m)) {
880 m->unit_log_field = "UNIT=";
881 m->unit_log_format_string = "UNIT=%s";
882
883 m->invocation_log_field = "INVOCATION_ID=";
884 m->invocation_log_format_string = "INVOCATION_ID=%s";
885 } else {
886 m->unit_log_field = "USER_UNIT=";
887 m->unit_log_format_string = "USER_UNIT=%s";
888
889 m->invocation_log_field = "USER_INVOCATION_ID=";
890 m->invocation_log_format_string = "USER_INVOCATION_ID=%s";
891 }
892
893 /* Reboot immediately if the user hits C-A-D more often than 7x per 2s */
894 m->ctrl_alt_del_ratelimit = (RateLimit) { .interval = 2 * USEC_PER_SEC, .burst = 7 };
895
896 r = manager_default_environment(m);
897 if (r < 0)
898 return r;
899
900 r = hashmap_ensure_allocated(&m->units, &string_hash_ops);
901 if (r < 0)
902 return r;
903
904 r = hashmap_ensure_allocated(&m->cgroup_unit, &path_hash_ops);
905 if (r < 0)
906 return r;
907
908 r = hashmap_ensure_allocated(&m->watch_bus, &string_hash_ops);
909 if (r < 0)
910 return r;
911
912 r = prioq_ensure_allocated(&m->run_queue, compare_job_priority);
913 if (r < 0)
914 return r;
915
916 r = manager_setup_prefix(m);
917 if (r < 0)
918 return r;
919
920 r = manager_find_credentials_dirs(m);
921 if (r < 0)
922 return r;
923
924 r = sd_event_default(&m->event);
925 if (r < 0)
926 return r;
927
928 r = manager_setup_run_queue(m);
929 if (r < 0)
930 return r;
931
932 if (FLAGS_SET(test_run_flags, MANAGER_TEST_RUN_MINIMAL)) {
933 m->cgroup_root = strdup("");
934 if (!m->cgroup_root)
935 return -ENOMEM;
936 } else {
937 r = manager_setup_signals(m);
938 if (r < 0)
939 return r;
940
941 r = manager_setup_cgroup(m);
942 if (r < 0)
943 return r;
944
945 r = manager_setup_time_change(m);
946 if (r < 0)
947 return r;
948
949 r = manager_read_timezone_stat(m);
950 if (r < 0)
951 return r;
952
953 (void) manager_setup_timezone_change(m);
954
955 r = manager_setup_sigchld_event_source(m);
956 if (r < 0)
957 return r;
958
959 #if HAVE_LIBBPF
960 if (MANAGER_IS_SYSTEM(m) && lsm_bpf_supported(/* initialize = */ true)) {
961 r = lsm_bpf_setup(m);
962 if (r < 0)
963 log_warning_errno(r, "Failed to setup LSM BPF, ignoring: %m");
964 }
965 #endif
966 }
967
968 if (test_run_flags == 0) {
969 if (MANAGER_IS_SYSTEM(m))
970 r = mkdir_label("/run/systemd/units", 0755);
971 else {
972 _cleanup_free_ char *units_path = NULL;
973 r = xdg_user_runtime_dir(&units_path, "/systemd/units");
974 if (r < 0)
975 return r;
976 r = mkdir_p_label(units_path, 0755);
977 }
978
979 if (r < 0 && r != -EEXIST)
980 return r;
981 }
982
983 m->taint_usr =
984 !in_initrd() &&
985 dir_is_empty("/usr", /* ignore_hidden_or_backup= */ false) > 0;
986
987 /* Note that we do not set up the notify fd here. We do that after deserialization,
988 * since they might have gotten serialized across the reexec. */
989
990 *_m = TAKE_PTR(m);
991
992 return 0;
993 }
994
995 static int manager_setup_notify(Manager *m) {
996 int r;
997
998 if (MANAGER_IS_TEST_RUN(m))
999 return 0;
1000
1001 if (m->notify_fd < 0) {
1002 _cleanup_close_ int fd = -1;
1003 union sockaddr_union sa;
1004 socklen_t sa_len;
1005
1006 /* First free all secondary fields */
1007 m->notify_socket = mfree(m->notify_socket);
1008 m->notify_event_source = sd_event_source_disable_unref(m->notify_event_source);
1009
1010 fd = socket(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC|SOCK_NONBLOCK, 0);
1011 if (fd < 0)
1012 return log_error_errno(errno, "Failed to allocate notification socket: %m");
1013
1014 fd_increase_rxbuf(fd, NOTIFY_RCVBUF_SIZE);
1015
1016 m->notify_socket = path_join(m->prefix[EXEC_DIRECTORY_RUNTIME], "systemd/notify");
1017 if (!m->notify_socket)
1018 return log_oom();
1019
1020 r = sockaddr_un_set_path(&sa.un, m->notify_socket);
1021 if (r < 0)
1022 return log_error_errno(r, "Notify socket '%s' not valid for AF_UNIX socket address, refusing.",
1023 m->notify_socket);
1024 sa_len = r;
1025
1026 (void) mkdir_parents_label(m->notify_socket, 0755);
1027 (void) sockaddr_un_unlink(&sa.un);
1028
1029 r = mac_selinux_bind(fd, &sa.sa, sa_len);
1030 if (r < 0)
1031 return log_error_errno(r, "bind(%s) failed: %m", m->notify_socket);
1032
1033 r = setsockopt_int(fd, SOL_SOCKET, SO_PASSCRED, true);
1034 if (r < 0)
1035 return log_error_errno(r, "SO_PASSCRED failed: %m");
1036
1037 m->notify_fd = TAKE_FD(fd);
1038
1039 log_debug("Using notification socket %s", m->notify_socket);
1040 }
1041
1042 if (!m->notify_event_source) {
1043 r = sd_event_add_io(m->event, &m->notify_event_source, m->notify_fd, EPOLLIN, manager_dispatch_notify_fd, m);
1044 if (r < 0)
1045 return log_error_errno(r, "Failed to allocate notify event source: %m");
1046
1047 /* Process notification messages a bit earlier than SIGCHLD, so that we can still identify to which
1048 * service an exit message belongs. */
1049 r = sd_event_source_set_priority(m->notify_event_source, SD_EVENT_PRIORITY_NORMAL-8);
1050 if (r < 0)
1051 return log_error_errno(r, "Failed to set priority of notify event source: %m");
1052
1053 (void) sd_event_source_set_description(m->notify_event_source, "manager-notify");
1054 }
1055
1056 return 0;
1057 }
1058
1059 static int manager_setup_cgroups_agent(Manager *m) {
1060
1061 static const union sockaddr_union sa = {
1062 .un.sun_family = AF_UNIX,
1063 .un.sun_path = "/run/systemd/cgroups-agent",
1064 };
1065 int r;
1066
1067 /* This creates a listening socket we receive cgroups agent messages on. We do not use D-Bus for delivering
1068 * these messages from the cgroups agent binary to PID 1, as the cgroups agent binary is very short-living, and
1069 * each instance of it needs a new D-Bus connection. Since D-Bus connections are SOCK_STREAM/AF_UNIX, on
1070 * overloaded systems the backlog of the D-Bus socket becomes relevant, as not more than the configured number
1071 * of D-Bus connections may be queued until the kernel will start dropping further incoming connections,
1072 * possibly resulting in lost cgroups agent messages. To avoid this, we'll use a private SOCK_DGRAM/AF_UNIX
1073 * socket, where no backlog is relevant as communication may take place without an actual connect() cycle, and
1074 * we thus won't lose messages.
1075 *
1076 * Note that PID 1 will forward the agent message to system bus, so that the user systemd instance may listen
1077 * to it. The system instance hence listens on this special socket, but the user instances listen on the system
1078 * bus for these messages. */
1079
1080 if (MANAGER_IS_TEST_RUN(m))
1081 return 0;
1082
1083 if (!MANAGER_IS_SYSTEM(m))
1084 return 0;
1085
1086 r = cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER);
1087 if (r < 0)
1088 return log_error_errno(r, "Failed to determine whether unified cgroups hierarchy is used: %m");
1089 if (r > 0) /* We don't need this anymore on the unified hierarchy */
1090 return 0;
1091
1092 if (m->cgroups_agent_fd < 0) {
1093 _cleanup_close_ int fd = -1;
1094
1095 /* First free all secondary fields */
1096 m->cgroups_agent_event_source = sd_event_source_disable_unref(m->cgroups_agent_event_source);
1097
1098 fd = socket(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC|SOCK_NONBLOCK, 0);
1099 if (fd < 0)
1100 return log_error_errno(errno, "Failed to allocate cgroups agent socket: %m");
1101
1102 fd_increase_rxbuf(fd, CGROUPS_AGENT_RCVBUF_SIZE);
1103
1104 (void) sockaddr_un_unlink(&sa.un);
1105
1106 /* Only allow root to connect to this socket */
1107 RUN_WITH_UMASK(0077)
1108 r = bind(fd, &sa.sa, SOCKADDR_UN_LEN(sa.un));
1109 if (r < 0)
1110 return log_error_errno(errno, "bind(%s) failed: %m", sa.un.sun_path);
1111
1112 m->cgroups_agent_fd = TAKE_FD(fd);
1113 }
1114
1115 if (!m->cgroups_agent_event_source) {
1116 r = sd_event_add_io(m->event, &m->cgroups_agent_event_source, m->cgroups_agent_fd, EPOLLIN, manager_dispatch_cgroups_agent_fd, m);
1117 if (r < 0)
1118 return log_error_errno(r, "Failed to allocate cgroups agent event source: %m");
1119
1120 /* Process cgroups notifications early. Note that when the agent notification is received
1121 * we'll just enqueue the unit in the cgroup empty queue, hence pick a high priority than
1122 * that. Also see handling of cgroup inotify for the unified cgroup stuff. */
1123 r = sd_event_source_set_priority(m->cgroups_agent_event_source, SD_EVENT_PRIORITY_NORMAL-9);
1124 if (r < 0)
1125 return log_error_errno(r, "Failed to set priority of cgroups agent event source: %m");
1126
1127 (void) sd_event_source_set_description(m->cgroups_agent_event_source, "manager-cgroups-agent");
1128 }
1129
1130 return 0;
1131 }
1132
1133 static int manager_setup_user_lookup_fd(Manager *m) {
1134 int r;
1135
1136 assert(m);
1137
1138 /* Set up the socket pair used for passing UID/GID resolution results from forked off processes to PID
1139 * 1. Background: we can't do name lookups (NSS) from PID 1, since it might involve IPC and thus activation,
1140 * and we might hence deadlock on ourselves. Hence we do all user/group lookups asynchronously from the forked
1141 * off processes right before executing the binaries to start. In order to be able to clean up any IPC objects
1142 * created by a unit (see RemoveIPC=) we need to know in PID 1 the used UID/GID of the executed processes,
1143 * hence we establish this communication channel so that forked off processes can pass their UID/GID
1144 * information back to PID 1. The forked off processes send their resolved UID/GID to PID 1 in a simple
1145 * datagram, along with their unit name, so that we can share one communication socket pair among all units for
1146 * this purpose.
1147 *
1148 * You might wonder why we need a communication channel for this that is independent of the usual notification
1149 * socket scheme (i.e. $NOTIFY_SOCKET). The primary difference is about trust: data sent via the $NOTIFY_SOCKET
1150 * channel is only accepted if it originates from the right unit and if reception was enabled for it. The user
1151 * lookup socket OTOH is only accessible by PID 1 and its children until they exec(), and always available.
1152 *
1153 * Note that this function is called under two circumstances: when we first initialize (in which case we
1154 * allocate both the socket pair and the event source to listen on it), and when we deserialize after a reload
1155 * (in which case the socket pair already exists but we still need to allocate the event source for it). */
1156
1157 if (m->user_lookup_fds[0] < 0) {
1158
1159 /* Free all secondary fields */
1160 safe_close_pair(m->user_lookup_fds);
1161 m->user_lookup_event_source = sd_event_source_disable_unref(m->user_lookup_event_source);
1162
1163 if (socketpair(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC, 0, m->user_lookup_fds) < 0)
1164 return log_error_errno(errno, "Failed to allocate user lookup socket: %m");
1165
1166 (void) fd_increase_rxbuf(m->user_lookup_fds[0], NOTIFY_RCVBUF_SIZE);
1167 }
1168
1169 if (!m->user_lookup_event_source) {
1170 r = sd_event_add_io(m->event, &m->user_lookup_event_source, m->user_lookup_fds[0], EPOLLIN, manager_dispatch_user_lookup_fd, m);
1171 if (r < 0)
1172 return log_error_errno(errno, "Failed to allocate user lookup event source: %m");
1173
1174 /* Process even earlier than the notify event source, so that we always know first about valid UID/GID
1175 * resolutions */
1176 r = sd_event_source_set_priority(m->user_lookup_event_source, SD_EVENT_PRIORITY_NORMAL-11);
1177 if (r < 0)
1178 return log_error_errno(errno, "Failed to set priority of user lookup event source: %m");
1179
1180 (void) sd_event_source_set_description(m->user_lookup_event_source, "user-lookup");
1181 }
1182
1183 return 0;
1184 }
1185
1186 static unsigned manager_dispatch_cleanup_queue(Manager *m) {
1187 Unit *u;
1188 unsigned n = 0;
1189
1190 assert(m);
1191
1192 while ((u = m->cleanup_queue)) {
1193 assert(u->in_cleanup_queue);
1194
1195 unit_free(u);
1196 n++;
1197 }
1198
1199 return n;
1200 }
1201
1202 enum {
1203 GC_OFFSET_IN_PATH, /* This one is on the path we were traveling */
1204 GC_OFFSET_UNSURE, /* No clue */
1205 GC_OFFSET_GOOD, /* We still need this unit */
1206 GC_OFFSET_BAD, /* We don't need this unit anymore */
1207 _GC_OFFSET_MAX
1208 };
1209
1210 static void unit_gc_mark_good(Unit *u, unsigned gc_marker) {
1211 Unit *other;
1212
1213 u->gc_marker = gc_marker + GC_OFFSET_GOOD;
1214
1215 /* Recursively mark referenced units as GOOD as well */
1216 UNIT_FOREACH_DEPENDENCY(other, u, UNIT_ATOM_REFERENCES)
1217 if (other->gc_marker == gc_marker + GC_OFFSET_UNSURE)
1218 unit_gc_mark_good(other, gc_marker);
1219 }
1220
1221 static void unit_gc_sweep(Unit *u, unsigned gc_marker) {
1222 Unit *other;
1223 bool is_bad;
1224
1225 assert(u);
1226
1227 if (IN_SET(u->gc_marker - gc_marker,
1228 GC_OFFSET_GOOD, GC_OFFSET_BAD, GC_OFFSET_UNSURE, GC_OFFSET_IN_PATH))
1229 return;
1230
1231 if (u->in_cleanup_queue)
1232 goto bad;
1233
1234 if (!unit_may_gc(u))
1235 goto good;
1236
1237 u->gc_marker = gc_marker + GC_OFFSET_IN_PATH;
1238
1239 is_bad = true;
1240
1241 UNIT_FOREACH_DEPENDENCY(other, u, UNIT_ATOM_REFERENCED_BY) {
1242 unit_gc_sweep(other, gc_marker);
1243
1244 if (other->gc_marker == gc_marker + GC_OFFSET_GOOD)
1245 goto good;
1246
1247 if (other->gc_marker != gc_marker + GC_OFFSET_BAD)
1248 is_bad = false;
1249 }
1250
1251 LIST_FOREACH(refs_by_target, ref, u->refs_by_target) {
1252 unit_gc_sweep(ref->source, gc_marker);
1253
1254 if (ref->source->gc_marker == gc_marker + GC_OFFSET_GOOD)
1255 goto good;
1256
1257 if (ref->source->gc_marker != gc_marker + GC_OFFSET_BAD)
1258 is_bad = false;
1259 }
1260
1261 if (is_bad)
1262 goto bad;
1263
1264 /* We were unable to find anything out about this entry, so
1265 * let's investigate it later */
1266 u->gc_marker = gc_marker + GC_OFFSET_UNSURE;
1267 unit_add_to_gc_queue(u);
1268 return;
1269
1270 bad:
1271 /* We definitely know that this one is not useful anymore, so
1272 * let's mark it for deletion */
1273 u->gc_marker = gc_marker + GC_OFFSET_BAD;
1274 unit_add_to_cleanup_queue(u);
1275 return;
1276
1277 good:
1278 unit_gc_mark_good(u, gc_marker);
1279 }
1280
1281 static unsigned manager_dispatch_gc_unit_queue(Manager *m) {
1282 unsigned n = 0, gc_marker;
1283 Unit *u;
1284
1285 assert(m);
1286
1287 /* log_debug("Running GC..."); */
1288
1289 m->gc_marker += _GC_OFFSET_MAX;
1290 if (m->gc_marker + _GC_OFFSET_MAX <= _GC_OFFSET_MAX)
1291 m->gc_marker = 1;
1292
1293 gc_marker = m->gc_marker;
1294
1295 while ((u = m->gc_unit_queue)) {
1296 assert(u->in_gc_queue);
1297
1298 unit_gc_sweep(u, gc_marker);
1299
1300 LIST_REMOVE(gc_queue, m->gc_unit_queue, u);
1301 u->in_gc_queue = false;
1302
1303 n++;
1304
1305 if (IN_SET(u->gc_marker - gc_marker,
1306 GC_OFFSET_BAD, GC_OFFSET_UNSURE)) {
1307 if (u->id)
1308 log_unit_debug(u, "Collecting.");
1309 u->gc_marker = gc_marker + GC_OFFSET_BAD;
1310 unit_add_to_cleanup_queue(u);
1311 }
1312 }
1313
1314 return n;
1315 }
1316
1317 static unsigned manager_dispatch_gc_job_queue(Manager *m) {
1318 unsigned n = 0;
1319 Job *j;
1320
1321 assert(m);
1322
1323 while ((j = m->gc_job_queue)) {
1324 assert(j->in_gc_queue);
1325
1326 LIST_REMOVE(gc_queue, m->gc_job_queue, j);
1327 j->in_gc_queue = false;
1328
1329 n++;
1330
1331 if (!job_may_gc(j))
1332 continue;
1333
1334 log_unit_debug(j->unit, "Collecting job.");
1335 (void) job_finish_and_invalidate(j, JOB_COLLECTED, false, false);
1336 }
1337
1338 return n;
1339 }
1340
1341 static unsigned manager_dispatch_stop_when_unneeded_queue(Manager *m) {
1342 unsigned n = 0;
1343 Unit *u;
1344 int r;
1345
1346 assert(m);
1347
1348 while ((u = m->stop_when_unneeded_queue)) {
1349 _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL;
1350
1351 assert(u->in_stop_when_unneeded_queue);
1352 LIST_REMOVE(stop_when_unneeded_queue, m->stop_when_unneeded_queue, u);
1353 u->in_stop_when_unneeded_queue = false;
1354
1355 n++;
1356
1357 if (!unit_is_unneeded(u))
1358 continue;
1359
1360 log_unit_debug(u, "Unit is not needed anymore.");
1361
1362 /* If stopping a unit fails continuously we might enter a stop loop here, hence stop acting on the
1363 * service being unnecessary after a while. */
1364
1365 if (!ratelimit_below(&u->auto_start_stop_ratelimit)) {
1366 log_unit_warning(u, "Unit not needed anymore, but not stopping since we tried this too often recently.");
1367 continue;
1368 }
1369
1370 /* Ok, nobody needs us anymore. Sniff. Then let's commit suicide */
1371 r = manager_add_job(u->manager, JOB_STOP, u, JOB_FAIL, NULL, &error, NULL);
1372 if (r < 0)
1373 log_unit_warning_errno(u, r, "Failed to enqueue stop job, ignoring: %s", bus_error_message(&error, r));
1374 }
1375
1376 return n;
1377 }
1378
1379 static unsigned manager_dispatch_start_when_upheld_queue(Manager *m) {
1380 unsigned n = 0;
1381 Unit *u;
1382 int r;
1383
1384 assert(m);
1385
1386 while ((u = m->start_when_upheld_queue)) {
1387 _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL;
1388 Unit *culprit = NULL;
1389
1390 assert(u->in_start_when_upheld_queue);
1391 LIST_REMOVE(start_when_upheld_queue, m->start_when_upheld_queue, u);
1392 u->in_start_when_upheld_queue = false;
1393
1394 n++;
1395
1396 if (!unit_is_upheld_by_active(u, &culprit))
1397 continue;
1398
1399 log_unit_debug(u, "Unit is started because upheld by active unit %s.", culprit->id);
1400
1401 /* If stopping a unit fails continuously we might enter a stop loop here, hence stop acting on the
1402 * service being unnecessary after a while. */
1403
1404 if (!ratelimit_below(&u->auto_start_stop_ratelimit)) {
1405 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);
1406 continue;
1407 }
1408
1409 r = manager_add_job(u->manager, JOB_START, u, JOB_FAIL, NULL, &error, NULL);
1410 if (r < 0)
1411 log_unit_warning_errno(u, r, "Failed to enqueue start job, ignoring: %s", bus_error_message(&error, r));
1412 }
1413
1414 return n;
1415 }
1416
1417 static unsigned manager_dispatch_stop_when_bound_queue(Manager *m) {
1418 unsigned n = 0;
1419 Unit *u;
1420 int r;
1421
1422 assert(m);
1423
1424 while ((u = m->stop_when_bound_queue)) {
1425 _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL;
1426 Unit *culprit = NULL;
1427
1428 assert(u->in_stop_when_bound_queue);
1429 LIST_REMOVE(stop_when_bound_queue, m->stop_when_bound_queue, u);
1430 u->in_stop_when_bound_queue = false;
1431
1432 n++;
1433
1434 if (!unit_is_bound_by_inactive(u, &culprit))
1435 continue;
1436
1437 log_unit_debug(u, "Unit is stopped because bound to inactive unit %s.", culprit->id);
1438
1439 /* If stopping a unit fails continuously we might enter a stop loop here, hence stop acting on the
1440 * service being unnecessary after a while. */
1441
1442 if (!ratelimit_below(&u->auto_start_stop_ratelimit)) {
1443 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);
1444 continue;
1445 }
1446
1447 r = manager_add_job(u->manager, JOB_STOP, u, JOB_REPLACE, NULL, &error, NULL);
1448 if (r < 0)
1449 log_unit_warning_errno(u, r, "Failed to enqueue stop job, ignoring: %s", bus_error_message(&error, r));
1450 }
1451
1452 return n;
1453 }
1454
1455 static void manager_clear_jobs_and_units(Manager *m) {
1456 Unit *u;
1457
1458 assert(m);
1459
1460 while ((u = hashmap_first(m->units)))
1461 unit_free(u);
1462
1463 manager_dispatch_cleanup_queue(m);
1464
1465 assert(!m->load_queue);
1466 assert(prioq_isempty(m->run_queue));
1467 assert(!m->dbus_unit_queue);
1468 assert(!m->dbus_job_queue);
1469 assert(!m->cleanup_queue);
1470 assert(!m->gc_unit_queue);
1471 assert(!m->gc_job_queue);
1472 assert(!m->cgroup_realize_queue);
1473 assert(!m->cgroup_empty_queue);
1474 assert(!m->cgroup_oom_queue);
1475 assert(!m->target_deps_queue);
1476 assert(!m->stop_when_unneeded_queue);
1477 assert(!m->start_when_upheld_queue);
1478 assert(!m->stop_when_bound_queue);
1479
1480 assert(hashmap_isempty(m->jobs));
1481 assert(hashmap_isempty(m->units));
1482
1483 m->n_on_console = 0;
1484 m->n_running_jobs = 0;
1485 m->n_installed_jobs = 0;
1486 m->n_failed_jobs = 0;
1487 }
1488
1489 Manager* manager_free(Manager *m) {
1490 if (!m)
1491 return NULL;
1492
1493 manager_clear_jobs_and_units(m);
1494
1495 for (UnitType c = 0; c < _UNIT_TYPE_MAX; c++)
1496 if (unit_vtable[c]->shutdown)
1497 unit_vtable[c]->shutdown(m);
1498
1499 /* Keep the cgroup hierarchy in place except when we know we are going down for good */
1500 manager_shutdown_cgroup(m, IN_SET(m->objective, MANAGER_EXIT, MANAGER_REBOOT, MANAGER_POWEROFF, MANAGER_HALT, MANAGER_KEXEC));
1501
1502 lookup_paths_flush_generator(&m->lookup_paths);
1503
1504 bus_done(m);
1505 manager_varlink_done(m);
1506
1507 exec_runtime_vacuum(m);
1508 hashmap_free(m->exec_runtime_by_id);
1509
1510 dynamic_user_vacuum(m, false);
1511 hashmap_free(m->dynamic_users);
1512
1513 hashmap_free(m->units);
1514 hashmap_free(m->units_by_invocation_id);
1515 hashmap_free(m->jobs);
1516 hashmap_free(m->watch_pids);
1517 hashmap_free(m->watch_bus);
1518
1519 prioq_free(m->run_queue);
1520
1521 set_free(m->startup_units);
1522 set_free(m->failed_units);
1523
1524 sd_event_source_unref(m->signal_event_source);
1525 sd_event_source_unref(m->sigchld_event_source);
1526 sd_event_source_unref(m->notify_event_source);
1527 sd_event_source_unref(m->cgroups_agent_event_source);
1528 sd_event_source_unref(m->time_change_event_source);
1529 sd_event_source_unref(m->timezone_change_event_source);
1530 sd_event_source_unref(m->jobs_in_progress_event_source);
1531 sd_event_source_unref(m->run_queue_event_source);
1532 sd_event_source_unref(m->user_lookup_event_source);
1533
1534 safe_close(m->signal_fd);
1535 safe_close(m->notify_fd);
1536 safe_close(m->cgroups_agent_fd);
1537 safe_close_pair(m->user_lookup_fds);
1538
1539 manager_close_ask_password(m);
1540
1541 manager_close_idle_pipe(m);
1542
1543 sd_event_unref(m->event);
1544
1545 free(m->notify_socket);
1546
1547 lookup_paths_free(&m->lookup_paths);
1548 strv_free(m->transient_environment);
1549 strv_free(m->client_environment);
1550
1551 hashmap_free(m->cgroup_unit);
1552 manager_free_unit_name_maps(m);
1553
1554 free(m->switch_root);
1555 free(m->switch_root_init);
1556
1557 free(m->default_smack_process_label);
1558
1559 rlimit_free_all(m->rlimit);
1560
1561 assert(hashmap_isempty(m->units_requiring_mounts_for));
1562 hashmap_free(m->units_requiring_mounts_for);
1563
1564 hashmap_free(m->uid_refs);
1565 hashmap_free(m->gid_refs);
1566
1567 for (ExecDirectoryType dt = 0; dt < _EXEC_DIRECTORY_TYPE_MAX; dt++)
1568 m->prefix[dt] = mfree(m->prefix[dt]);
1569 free(m->received_credentials_directory);
1570 free(m->received_encrypted_credentials_directory);
1571
1572 free(m->watchdog_pretimeout_governor);
1573 free(m->watchdog_pretimeout_governor_overridden);
1574
1575 #if BPF_FRAMEWORK
1576 lsm_bpf_destroy(m->restrict_fs);
1577 #endif
1578
1579 return mfree(m);
1580 }
1581
1582 static void manager_enumerate_perpetual(Manager *m) {
1583 assert(m);
1584
1585 if (FLAGS_SET(m->test_run_flags, MANAGER_TEST_RUN_MINIMAL))
1586 return;
1587
1588 /* Let's ask every type to load all units from disk/kernel that it might know */
1589 for (UnitType c = 0; c < _UNIT_TYPE_MAX; c++) {
1590 if (!unit_type_supported(c)) {
1591 log_debug("Unit type .%s is not supported on this system.", unit_type_to_string(c));
1592 continue;
1593 }
1594
1595 if (unit_vtable[c]->enumerate_perpetual)
1596 unit_vtable[c]->enumerate_perpetual(m);
1597 }
1598 }
1599
1600 static void manager_enumerate(Manager *m) {
1601 assert(m);
1602
1603 if (FLAGS_SET(m->test_run_flags, MANAGER_TEST_RUN_MINIMAL))
1604 return;
1605
1606 /* Let's ask every type to load all units from disk/kernel that it might know */
1607 for (UnitType c = 0; c < _UNIT_TYPE_MAX; c++) {
1608 if (!unit_type_supported(c)) {
1609 log_debug("Unit type .%s is not supported on this system.", unit_type_to_string(c));
1610 continue;
1611 }
1612
1613 if (unit_vtable[c]->enumerate)
1614 unit_vtable[c]->enumerate(m);
1615 }
1616
1617 manager_dispatch_load_queue(m);
1618 }
1619
1620 static void manager_coldplug(Manager *m) {
1621 Unit *u;
1622 char *k;
1623 int r;
1624
1625 assert(m);
1626
1627 log_debug("Invoking unit coldplug() handlers%s", special_glyph(SPECIAL_GLYPH_ELLIPSIS));
1628
1629 /* Let's place the units back into their deserialized state */
1630 HASHMAP_FOREACH_KEY(u, k, m->units) {
1631
1632 /* ignore aliases */
1633 if (u->id != k)
1634 continue;
1635
1636 r = unit_coldplug(u);
1637 if (r < 0)
1638 log_warning_errno(r, "We couldn't coldplug %s, proceeding anyway: %m", u->id);
1639 }
1640 }
1641
1642 static void manager_catchup(Manager *m) {
1643 Unit *u;
1644 char *k;
1645
1646 assert(m);
1647
1648 log_debug("Invoking unit catchup() handlers%s", special_glyph(SPECIAL_GLYPH_ELLIPSIS));
1649
1650 /* Let's catch up on any state changes that happened while we were reloading/reexecing */
1651 HASHMAP_FOREACH_KEY(u, k, m->units) {
1652
1653 /* ignore aliases */
1654 if (u->id != k)
1655 continue;
1656
1657 unit_catchup(u);
1658 }
1659 }
1660
1661 static void manager_distribute_fds(Manager *m, FDSet *fds) {
1662 Unit *u;
1663
1664 assert(m);
1665
1666 HASHMAP_FOREACH(u, m->units) {
1667
1668 if (fdset_size(fds) <= 0)
1669 break;
1670
1671 if (!UNIT_VTABLE(u)->distribute_fds)
1672 continue;
1673
1674 UNIT_VTABLE(u)->distribute_fds(u, fds);
1675 }
1676 }
1677
1678 static bool manager_dbus_is_running(Manager *m, bool deserialized) {
1679 Unit *u;
1680
1681 assert(m);
1682
1683 /* This checks whether the dbus instance we are supposed to expose our APIs on is up. We check both the socket
1684 * and the service unit. If the 'deserialized' parameter is true we'll check the deserialized state of the unit
1685 * rather than the current one. */
1686
1687 if (MANAGER_IS_TEST_RUN(m))
1688 return false;
1689
1690 u = manager_get_unit(m, SPECIAL_DBUS_SOCKET);
1691 if (!u)
1692 return false;
1693 if ((deserialized ? SOCKET(u)->deserialized_state : SOCKET(u)->state) != SOCKET_RUNNING)
1694 return false;
1695
1696 u = manager_get_unit(m, SPECIAL_DBUS_SERVICE);
1697 if (!u)
1698 return false;
1699 if (!IN_SET((deserialized ? SERVICE(u)->deserialized_state : SERVICE(u)->state), SERVICE_RUNNING, SERVICE_RELOAD))
1700 return false;
1701
1702 return true;
1703 }
1704
1705 static void manager_setup_bus(Manager *m) {
1706 assert(m);
1707
1708 /* Let's set up our private bus connection now, unconditionally */
1709 (void) bus_init_private(m);
1710
1711 /* If we are in --user mode also connect to the system bus now */
1712 if (MANAGER_IS_USER(m))
1713 (void) bus_init_system(m);
1714
1715 /* Let's connect to the bus now, but only if the unit is supposed to be up */
1716 if (manager_dbus_is_running(m, MANAGER_IS_RELOADING(m))) {
1717 (void) bus_init_api(m);
1718
1719 if (MANAGER_IS_SYSTEM(m))
1720 (void) bus_init_system(m);
1721 }
1722 }
1723
1724 static void manager_preset_all(Manager *m) {
1725 int r;
1726
1727 assert(m);
1728
1729 if (m->first_boot <= 0)
1730 return;
1731
1732 if (!MANAGER_IS_SYSTEM(m))
1733 return;
1734
1735 if (MANAGER_IS_TEST_RUN(m))
1736 return;
1737
1738 /* If this is the first boot, and we are in the host system, then preset everything */
1739 UnitFilePresetMode mode = FIRST_BOOT_FULL_PRESET ? UNIT_FILE_PRESET_FULL : UNIT_FILE_PRESET_ENABLE_ONLY;
1740
1741 r = unit_file_preset_all(LOOKUP_SCOPE_SYSTEM, 0, NULL, mode, NULL, 0);
1742 if (r < 0)
1743 log_full_errno(r == -EEXIST ? LOG_NOTICE : LOG_WARNING, r,
1744 "Failed to populate /etc with preset unit settings, ignoring: %m");
1745 else
1746 log_info("Populated /etc with preset unit settings.");
1747 }
1748
1749 static void manager_ready(Manager *m) {
1750 assert(m);
1751
1752 /* After having loaded everything, do the final round of catching up with what might have changed */
1753
1754 m->objective = MANAGER_OK; /* Tell everyone we are up now */
1755
1756 /* It might be safe to log to the journal now and connect to dbus */
1757 manager_recheck_journal(m);
1758 manager_recheck_dbus(m);
1759
1760 /* Let's finally catch up with any changes that took place while we were reloading/reexecing */
1761 manager_catchup(m);
1762
1763 /* Create a file which will indicate when the manager started loading units the last time. */
1764 if (MANAGER_IS_SYSTEM(m))
1765 (void) touch_file("/run/systemd/systemd-units-load", false,
1766 m->timestamps[MANAGER_TIMESTAMP_UNITS_LOAD].realtime ?: now(CLOCK_REALTIME),
1767 UID_INVALID, GID_INVALID, 0444);
1768 }
1769
1770 Manager* manager_reloading_start(Manager *m) {
1771 m->n_reloading++;
1772 dual_timestamp_get(m->timestamps + MANAGER_TIMESTAMP_UNITS_LOAD);
1773 return m;
1774 }
1775
1776 void manager_reloading_stopp(Manager **m) {
1777 if (*m) {
1778 assert((*m)->n_reloading > 0);
1779 (*m)->n_reloading--;
1780 }
1781 }
1782
1783 int manager_startup(Manager *m, FILE *serialization, FDSet *fds, const char *root) {
1784 int r;
1785
1786 assert(m);
1787
1788 /* If we are running in test mode, we still want to run the generators,
1789 * but we should not touch the real generator directories. */
1790 r = lookup_paths_init_or_warn(&m->lookup_paths, m->unit_file_scope,
1791 MANAGER_IS_TEST_RUN(m) ? LOOKUP_PATHS_TEMPORARY_GENERATED : 0,
1792 root);
1793 if (r < 0)
1794 return r;
1795
1796 dual_timestamp_get(m->timestamps + manager_timestamp_initrd_mangle(MANAGER_TIMESTAMP_GENERATORS_START));
1797 r = manager_run_environment_generators(m);
1798 if (r >= 0)
1799 r = manager_run_generators(m);
1800 dual_timestamp_get(m->timestamps + manager_timestamp_initrd_mangle(MANAGER_TIMESTAMP_GENERATORS_FINISH));
1801 if (r < 0)
1802 return r;
1803
1804 manager_preset_all(m);
1805
1806 lookup_paths_log(&m->lookup_paths);
1807
1808 {
1809 /* This block is (optionally) done with the reloading counter bumped */
1810 _unused_ _cleanup_(manager_reloading_stopp) Manager *reloading = NULL;
1811
1812 /* If we will deserialize make sure that during enumeration this is already known, so we increase the
1813 * counter here already */
1814 if (serialization)
1815 reloading = manager_reloading_start(m);
1816
1817 /* First, enumerate what we can from all config files */
1818 dual_timestamp_get(m->timestamps + manager_timestamp_initrd_mangle(MANAGER_TIMESTAMP_UNITS_LOAD_START));
1819 manager_enumerate_perpetual(m);
1820 manager_enumerate(m);
1821 dual_timestamp_get(m->timestamps + manager_timestamp_initrd_mangle(MANAGER_TIMESTAMP_UNITS_LOAD_FINISH));
1822
1823 /* Second, deserialize if there is something to deserialize */
1824 if (serialization) {
1825 r = manager_deserialize(m, serialization, fds);
1826 if (r < 0)
1827 return log_error_errno(r, "Deserialization failed: %m");
1828 }
1829
1830 /* Any fds left? Find some unit which wants them. This is useful to allow container managers to pass
1831 * some file descriptors to us pre-initialized. This enables socket-based activation of entire
1832 * containers. */
1833 manager_distribute_fds(m, fds);
1834
1835 /* We might have deserialized the notify fd, but if we didn't then let's create the bus now */
1836 r = manager_setup_notify(m);
1837 if (r < 0)
1838 /* No sense to continue without notifications, our children would fail anyway. */
1839 return r;
1840
1841 r = manager_setup_cgroups_agent(m);
1842 if (r < 0)
1843 /* Likewise, no sense to continue without empty cgroup notifications. */
1844 return r;
1845
1846 r = manager_setup_user_lookup_fd(m);
1847 if (r < 0)
1848 /* This shouldn't fail, except if things are really broken. */
1849 return r;
1850
1851 /* Connect to the bus if we are good for it */
1852 manager_setup_bus(m);
1853
1854 /* Now that we are connected to all possible buses, let's deserialize who is tracking us. */
1855 r = bus_track_coldplug(m, &m->subscribed, false, m->deserialized_subscribed);
1856 if (r < 0)
1857 log_warning_errno(r, "Failed to deserialized tracked clients, ignoring: %m");
1858 m->deserialized_subscribed = strv_free(m->deserialized_subscribed);
1859
1860 r = manager_varlink_init(m);
1861 if (r < 0)
1862 log_warning_errno(r, "Failed to set up Varlink, ignoring: %m");
1863
1864 /* Third, fire things up! */
1865 manager_coldplug(m);
1866
1867 /* Clean up runtime objects */
1868 manager_vacuum(m);
1869
1870 if (serialization)
1871 /* Let's wait for the UnitNew/JobNew messages being sent, before we notify that the
1872 * reload is finished */
1873 m->send_reloading_done = true;
1874 }
1875
1876 manager_ready(m);
1877
1878 manager_set_switching_root(m, false);
1879
1880 return 0;
1881 }
1882
1883 int manager_add_job(
1884 Manager *m,
1885 JobType type,
1886 Unit *unit,
1887 JobMode mode,
1888 Set *affected_jobs,
1889 sd_bus_error *error,
1890 Job **ret) {
1891
1892 Transaction *tr;
1893 int r;
1894
1895 assert(m);
1896 assert(type < _JOB_TYPE_MAX);
1897 assert(unit);
1898 assert(mode < _JOB_MODE_MAX);
1899
1900 if (mode == JOB_ISOLATE && type != JOB_START)
1901 return sd_bus_error_set(error, SD_BUS_ERROR_INVALID_ARGS, "Isolate is only valid for start.");
1902
1903 if (mode == JOB_ISOLATE && !unit->allow_isolate)
1904 return sd_bus_error_set(error, BUS_ERROR_NO_ISOLATION, "Operation refused, unit may not be isolated.");
1905
1906 if (mode == JOB_TRIGGERING && type != JOB_STOP)
1907 return sd_bus_error_set(error, SD_BUS_ERROR_INVALID_ARGS, "--job-mode=triggering is only valid for stop.");
1908
1909 log_unit_debug(unit, "Trying to enqueue job %s/%s/%s", unit->id, job_type_to_string(type), job_mode_to_string(mode));
1910
1911 type = job_type_collapse(type, unit);
1912
1913 tr = transaction_new(mode == JOB_REPLACE_IRREVERSIBLY);
1914 if (!tr)
1915 return -ENOMEM;
1916
1917 r = transaction_add_job_and_dependencies(tr, type, unit, NULL, true, false,
1918 IN_SET(mode, JOB_IGNORE_DEPENDENCIES, JOB_IGNORE_REQUIREMENTS),
1919 mode == JOB_IGNORE_DEPENDENCIES, error);
1920 if (r < 0)
1921 goto tr_abort;
1922
1923 if (mode == JOB_ISOLATE) {
1924 r = transaction_add_isolate_jobs(tr, m);
1925 if (r < 0)
1926 goto tr_abort;
1927 }
1928
1929 if (mode == JOB_TRIGGERING) {
1930 r = transaction_add_triggering_jobs(tr, unit);
1931 if (r < 0)
1932 goto tr_abort;
1933 }
1934
1935 r = transaction_activate(tr, m, mode, affected_jobs, error);
1936 if (r < 0)
1937 goto tr_abort;
1938
1939 log_unit_debug(unit,
1940 "Enqueued job %s/%s as %u", unit->id,
1941 job_type_to_string(type), (unsigned) tr->anchor_job->id);
1942
1943 if (ret)
1944 *ret = tr->anchor_job;
1945
1946 transaction_free(tr);
1947 return 0;
1948
1949 tr_abort:
1950 transaction_abort(tr);
1951 transaction_free(tr);
1952 return r;
1953 }
1954
1955 int manager_add_job_by_name(Manager *m, JobType type, const char *name, JobMode mode, Set *affected_jobs, sd_bus_error *e, Job **ret) {
1956 Unit *unit = NULL; /* just to appease gcc, initialization is not really necessary */
1957 int r;
1958
1959 assert(m);
1960 assert(type < _JOB_TYPE_MAX);
1961 assert(name);
1962 assert(mode < _JOB_MODE_MAX);
1963
1964 r = manager_load_unit(m, name, NULL, NULL, &unit);
1965 if (r < 0)
1966 return r;
1967 assert(unit);
1968
1969 return manager_add_job(m, type, unit, mode, affected_jobs, e, ret);
1970 }
1971
1972 int manager_add_job_by_name_and_warn(Manager *m, JobType type, const char *name, JobMode mode, Set *affected_jobs, Job **ret) {
1973 _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL;
1974 int r;
1975
1976 assert(m);
1977 assert(type < _JOB_TYPE_MAX);
1978 assert(name);
1979 assert(mode < _JOB_MODE_MAX);
1980
1981 r = manager_add_job_by_name(m, type, name, mode, affected_jobs, &error, ret);
1982 if (r < 0)
1983 return log_warning_errno(r, "Failed to enqueue %s job for %s: %s", job_mode_to_string(mode), name, bus_error_message(&error, r));
1984
1985 return r;
1986 }
1987
1988 int manager_propagate_reload(Manager *m, Unit *unit, JobMode mode, sd_bus_error *e) {
1989 int r;
1990 Transaction *tr;
1991
1992 assert(m);
1993 assert(unit);
1994 assert(mode < _JOB_MODE_MAX);
1995 assert(mode != JOB_ISOLATE); /* Isolate is only valid for start */
1996
1997 tr = transaction_new(mode == JOB_REPLACE_IRREVERSIBLY);
1998 if (!tr)
1999 return -ENOMEM;
2000
2001 /* We need an anchor job */
2002 r = transaction_add_job_and_dependencies(tr, JOB_NOP, unit, NULL, false, false, true, true, e);
2003 if (r < 0)
2004 goto tr_abort;
2005
2006 /* Failure in adding individual dependencies is ignored, so this always succeeds. */
2007 transaction_add_propagate_reload_jobs(tr, unit, tr->anchor_job, mode == JOB_IGNORE_DEPENDENCIES, e);
2008
2009 r = transaction_activate(tr, m, mode, NULL, e);
2010 if (r < 0)
2011 goto tr_abort;
2012
2013 transaction_free(tr);
2014 return 0;
2015
2016 tr_abort:
2017 transaction_abort(tr);
2018 transaction_free(tr);
2019 return r;
2020 }
2021
2022 Job *manager_get_job(Manager *m, uint32_t id) {
2023 assert(m);
2024
2025 return hashmap_get(m->jobs, UINT32_TO_PTR(id));
2026 }
2027
2028 Unit *manager_get_unit(Manager *m, const char *name) {
2029 assert(m);
2030 assert(name);
2031
2032 return hashmap_get(m->units, name);
2033 }
2034
2035 static int manager_dispatch_target_deps_queue(Manager *m) {
2036 Unit *u;
2037 int r = 0;
2038
2039 assert(m);
2040
2041 while ((u = m->target_deps_queue)) {
2042 _cleanup_free_ Unit **targets = NULL;
2043 int n_targets;
2044
2045 assert(u->in_target_deps_queue);
2046
2047 LIST_REMOVE(target_deps_queue, u->manager->target_deps_queue, u);
2048 u->in_target_deps_queue = false;
2049
2050 /* Take an "atomic" snapshot of dependencies here, as the call below will likely modify the
2051 * dependencies, and we can't have it that hash tables we iterate through are modified while
2052 * we are iterating through them. */
2053 n_targets = unit_get_dependency_array(u, UNIT_ATOM_DEFAULT_TARGET_DEPENDENCIES, &targets);
2054 if (n_targets < 0)
2055 return n_targets;
2056
2057 for (int i = 0; i < n_targets; i++) {
2058 r = unit_add_default_target_dependency(u, targets[i]);
2059 if (r < 0)
2060 return r;
2061 }
2062 }
2063
2064 return r;
2065 }
2066
2067 unsigned manager_dispatch_load_queue(Manager *m) {
2068 Unit *u;
2069 unsigned n = 0;
2070
2071 assert(m);
2072
2073 /* Make sure we are not run recursively */
2074 if (m->dispatching_load_queue)
2075 return 0;
2076
2077 m->dispatching_load_queue = true;
2078
2079 /* Dispatches the load queue. Takes a unit from the queue and
2080 * tries to load its data until the queue is empty */
2081
2082 while ((u = m->load_queue)) {
2083 assert(u->in_load_queue);
2084
2085 unit_load(u);
2086 n++;
2087 }
2088
2089 m->dispatching_load_queue = false;
2090
2091 /* Dispatch the units waiting for their target dependencies to be added now, as all targets that we know about
2092 * should be loaded and have aliases resolved */
2093 (void) manager_dispatch_target_deps_queue(m);
2094
2095 return n;
2096 }
2097
2098 bool manager_unit_cache_should_retry_load(Unit *u) {
2099 assert(u);
2100
2101 /* Automatic reloading from disk only applies to units which were not found sometime in the past, and
2102 * the not-found stub is kept pinned in the unit graph by dependencies. For units that were
2103 * previously loaded, we don't do automatic reloading, and daemon-reload is necessary to update. */
2104 if (u->load_state != UNIT_NOT_FOUND)
2105 return false;
2106
2107 /* The cache has been updated since the last time we tried to load the unit. There might be new
2108 * fragment paths to read. */
2109 if (u->manager->unit_cache_timestamp_hash != u->fragment_not_found_timestamp_hash)
2110 return true;
2111
2112 /* The cache needs to be updated because there are modifications on disk. */
2113 return !lookup_paths_timestamp_hash_same(&u->manager->lookup_paths, u->manager->unit_cache_timestamp_hash, NULL);
2114 }
2115
2116 int manager_load_unit_prepare(
2117 Manager *m,
2118 const char *name,
2119 const char *path,
2120 sd_bus_error *e,
2121 Unit **_ret) {
2122
2123 _cleanup_(unit_freep) Unit *cleanup_ret = NULL;
2124 Unit *ret;
2125 UnitType t;
2126 int r;
2127
2128 assert(m);
2129 assert(_ret);
2130
2131 /* This will prepare the unit for loading, but not actually load anything from disk. */
2132
2133 if (path && !path_is_absolute(path))
2134 return sd_bus_error_setf(e, SD_BUS_ERROR_INVALID_ARGS, "Path %s is not absolute.", path);
2135
2136 if (!name) {
2137 /* 'name' and 'path' must not both be null. Check here 'path' using assert_se() to
2138 * workaround a bug in gcc that generates a -Wnonnull warning when calling basename(),
2139 * but this cannot be possible in any code path (See #6119). */
2140 assert_se(path);
2141 name = basename(path);
2142 }
2143
2144 t = unit_name_to_type(name);
2145
2146 if (t == _UNIT_TYPE_INVALID || !unit_name_is_valid(name, UNIT_NAME_PLAIN|UNIT_NAME_INSTANCE)) {
2147 if (unit_name_is_valid(name, UNIT_NAME_TEMPLATE))
2148 return sd_bus_error_setf(e, SD_BUS_ERROR_INVALID_ARGS, "Unit name %s is missing the instance name.", name);
2149
2150 return sd_bus_error_setf(e, SD_BUS_ERROR_INVALID_ARGS, "Unit name %s is not valid.", name);
2151 }
2152
2153 ret = manager_get_unit(m, name);
2154 if (ret) {
2155 /* The time-based cache allows to start new units without daemon-reload,
2156 * but if they are already referenced (because of dependencies or ordering)
2157 * then we have to force a load of the fragment. As an optimization, check
2158 * first if anything in the usual paths was modified since the last time
2159 * the cache was loaded. Also check if the last time an attempt to load the
2160 * unit was made was before the most recent cache refresh, so that we know
2161 * we need to try again — even if the cache is current, it might have been
2162 * updated in a different context before we had a chance to retry loading
2163 * this particular unit. */
2164 if (manager_unit_cache_should_retry_load(ret))
2165 ret->load_state = UNIT_STUB;
2166 else {
2167 *_ret = ret;
2168 return 1;
2169 }
2170 } else {
2171 ret = cleanup_ret = unit_new(m, unit_vtable[t]->object_size);
2172 if (!ret)
2173 return -ENOMEM;
2174 }
2175
2176 if (path) {
2177 r = free_and_strdup(&ret->fragment_path, path);
2178 if (r < 0)
2179 return r;
2180 }
2181
2182 r = unit_add_name(ret, name);
2183 if (r < 0)
2184 return r;
2185
2186 unit_add_to_load_queue(ret);
2187 unit_add_to_dbus_queue(ret);
2188 unit_add_to_gc_queue(ret);
2189
2190 *_ret = ret;
2191 cleanup_ret = NULL;
2192
2193 return 0;
2194 }
2195
2196 int manager_load_unit(
2197 Manager *m,
2198 const char *name,
2199 const char *path,
2200 sd_bus_error *e,
2201 Unit **_ret) {
2202
2203 int r;
2204
2205 assert(m);
2206 assert(_ret);
2207
2208 /* This will load the service information files, but not actually
2209 * start any services or anything. */
2210
2211 r = manager_load_unit_prepare(m, name, path, e, _ret);
2212 if (r != 0)
2213 return r;
2214
2215 manager_dispatch_load_queue(m);
2216
2217 *_ret = unit_follow_merge(*_ret);
2218 return 0;
2219 }
2220
2221 int manager_load_startable_unit_or_warn(
2222 Manager *m,
2223 const char *name,
2224 const char *path,
2225 Unit **ret) {
2226
2227 /* Load a unit, make sure it loaded fully and is not masked. */
2228
2229 _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL;
2230 Unit *unit;
2231 int r;
2232
2233 r = manager_load_unit(m, name, path, &error, &unit);
2234 if (r < 0)
2235 return log_error_errno(r, "Failed to load %s %s: %s",
2236 name ? "unit" : "unit file", name ?: path,
2237 bus_error_message(&error, r));
2238
2239 r = bus_unit_validate_load_state(unit, &error);
2240 if (r < 0)
2241 return log_error_errno(r, "%s", bus_error_message(&error, r));
2242
2243 *ret = unit;
2244 return 0;
2245 }
2246
2247 void manager_clear_jobs(Manager *m) {
2248 Job *j;
2249
2250 assert(m);
2251
2252 while ((j = hashmap_first(m->jobs)))
2253 /* No need to recurse. We're cancelling all jobs. */
2254 job_finish_and_invalidate(j, JOB_CANCELED, false, false);
2255 }
2256
2257 void manager_unwatch_pid(Manager *m, pid_t pid) {
2258 assert(m);
2259
2260 /* First let's drop the unit keyed as "pid". */
2261 (void) hashmap_remove(m->watch_pids, PID_TO_PTR(pid));
2262
2263 /* Then, let's also drop the array keyed by -pid. */
2264 free(hashmap_remove(m->watch_pids, PID_TO_PTR(-pid)));
2265 }
2266
2267 static int manager_dispatch_run_queue(sd_event_source *source, void *userdata) {
2268 Manager *m = userdata;
2269 Job *j;
2270
2271 assert(source);
2272 assert(m);
2273
2274 while ((j = prioq_peek(m->run_queue))) {
2275 assert(j->installed);
2276 assert(j->in_run_queue);
2277
2278 (void) job_run_and_invalidate(j);
2279 }
2280
2281 if (m->n_running_jobs > 0)
2282 manager_watch_jobs_in_progress(m);
2283
2284 if (m->n_on_console > 0)
2285 manager_watch_idle_pipe(m);
2286
2287 return 1;
2288 }
2289
2290 void manager_trigger_run_queue(Manager *m) {
2291 int r;
2292
2293 assert(m);
2294
2295 r = sd_event_source_set_enabled(
2296 m->run_queue_event_source,
2297 prioq_isempty(m->run_queue) ? SD_EVENT_OFF : SD_EVENT_ONESHOT);
2298 if (r < 0)
2299 log_warning_errno(r, "Failed to enable job run queue event source, ignoring: %m");
2300 }
2301
2302 static unsigned manager_dispatch_dbus_queue(Manager *m) {
2303 unsigned n = 0, budget;
2304 Unit *u;
2305 Job *j;
2306
2307 assert(m);
2308
2309 /* When we are reloading, let's not wait with generating signals, since we need to exit the manager as quickly
2310 * as we can. There's no point in throttling generation of signals in that case. */
2311 if (MANAGER_IS_RELOADING(m) || m->send_reloading_done || m->pending_reload_message)
2312 budget = UINT_MAX; /* infinite budget in this case */
2313 else {
2314 /* Anything to do at all? */
2315 if (!m->dbus_unit_queue && !m->dbus_job_queue)
2316 return 0;
2317
2318 /* Do we have overly many messages queued at the moment? If so, let's not enqueue more on top, let's
2319 * sit this cycle out, and process things in a later cycle when the queues got a bit emptier. */
2320 if (manager_bus_n_queued_write(m) > MANAGER_BUS_BUSY_THRESHOLD)
2321 return 0;
2322
2323 /* Only process a certain number of units/jobs per event loop iteration. Even if the bus queue wasn't
2324 * overly full before this call we shouldn't increase it in size too wildly in one step, and we
2325 * shouldn't monopolize CPU time with generating these messages. Note the difference in counting of
2326 * this "budget" and the "threshold" above: the "budget" is decreased only once per generated message,
2327 * regardless how many buses/direct connections it is enqueued on, while the "threshold" is applied to
2328 * each queued instance of bus message, i.e. if the same message is enqueued to five buses/direct
2329 * connections it will be counted five times. This difference in counting ("references"
2330 * vs. "instances") is primarily a result of the fact that it's easier to implement it this way,
2331 * however it also reflects the thinking that the "threshold" should put a limit on used queue memory,
2332 * i.e. space, while the "budget" should put a limit on time. Also note that the "threshold" is
2333 * currently chosen much higher than the "budget". */
2334 budget = MANAGER_BUS_MESSAGE_BUDGET;
2335 }
2336
2337 while (budget != 0 && (u = m->dbus_unit_queue)) {
2338
2339 assert(u->in_dbus_queue);
2340
2341 bus_unit_send_change_signal(u);
2342 n++;
2343
2344 if (budget != UINT_MAX)
2345 budget--;
2346 }
2347
2348 while (budget != 0 && (j = m->dbus_job_queue)) {
2349 assert(j->in_dbus_queue);
2350
2351 bus_job_send_change_signal(j);
2352 n++;
2353
2354 if (budget != UINT_MAX)
2355 budget--;
2356 }
2357
2358 if (m->send_reloading_done) {
2359 m->send_reloading_done = false;
2360 bus_manager_send_reloading(m, false);
2361 n++;
2362 }
2363
2364 if (m->pending_reload_message) {
2365 bus_send_pending_reload_message(m);
2366 n++;
2367 }
2368
2369 return n;
2370 }
2371
2372 static int manager_dispatch_cgroups_agent_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
2373 Manager *m = userdata;
2374 char buf[PATH_MAX];
2375 ssize_t n;
2376
2377 n = recv(fd, buf, sizeof(buf), 0);
2378 if (n < 0)
2379 return log_error_errno(errno, "Failed to read cgroups agent message: %m");
2380 if (n == 0) {
2381 log_error("Got zero-length cgroups agent message, ignoring.");
2382 return 0;
2383 }
2384 if ((size_t) n >= sizeof(buf)) {
2385 log_error("Got overly long cgroups agent message, ignoring.");
2386 return 0;
2387 }
2388
2389 if (memchr(buf, 0, n)) {
2390 log_error("Got cgroups agent message with embedded NUL byte, ignoring.");
2391 return 0;
2392 }
2393 buf[n] = 0;
2394
2395 manager_notify_cgroup_empty(m, buf);
2396 (void) bus_forward_agent_released(m, buf);
2397
2398 return 0;
2399 }
2400
2401 static bool manager_process_barrier_fd(char * const *tags, FDSet *fds) {
2402
2403 /* nothing else must be sent when using BARRIER=1 */
2404 if (strv_contains(tags, "BARRIER=1")) {
2405 if (strv_length(tags) == 1) {
2406 if (fdset_size(fds) != 1)
2407 log_warning("Got incorrect number of fds with BARRIER=1, closing them.");
2408 } else
2409 log_warning("Extra notification messages sent with BARRIER=1, ignoring everything.");
2410
2411 /* Drop the message if BARRIER=1 was found */
2412 return true;
2413 }
2414
2415 return false;
2416 }
2417
2418 static void manager_invoke_notify_message(
2419 Manager *m,
2420 Unit *u,
2421 const struct ucred *ucred,
2422 char * const *tags,
2423 FDSet *fds) {
2424
2425 assert(m);
2426 assert(u);
2427 assert(ucred);
2428 assert(tags);
2429
2430 if (u->notifygen == m->notifygen) /* Already invoked on this same unit in this same iteration? */
2431 return;
2432 u->notifygen = m->notifygen;
2433
2434 if (UNIT_VTABLE(u)->notify_message)
2435 UNIT_VTABLE(u)->notify_message(u, ucred, tags, fds);
2436
2437 else if (DEBUG_LOGGING) {
2438 _cleanup_free_ char *buf = NULL, *x = NULL, *y = NULL;
2439
2440 buf = strv_join(tags, ", ");
2441 if (buf)
2442 x = ellipsize(buf, 20, 90);
2443 if (x)
2444 y = cescape(x);
2445
2446 log_unit_debug(u, "Got notification message \"%s\", ignoring.", strnull(y));
2447 }
2448 }
2449
2450 static int manager_dispatch_notify_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
2451
2452 _cleanup_fdset_free_ FDSet *fds = NULL;
2453 Manager *m = userdata;
2454 char buf[NOTIFY_BUFFER_MAX+1];
2455 struct iovec iovec = {
2456 .iov_base = buf,
2457 .iov_len = sizeof(buf)-1,
2458 };
2459 CMSG_BUFFER_TYPE(CMSG_SPACE(sizeof(struct ucred)) +
2460 CMSG_SPACE(sizeof(int) * NOTIFY_FD_MAX)) control;
2461 struct msghdr msghdr = {
2462 .msg_iov = &iovec,
2463 .msg_iovlen = 1,
2464 .msg_control = &control,
2465 .msg_controllen = sizeof(control),
2466 };
2467
2468 struct cmsghdr *cmsg;
2469 struct ucred *ucred = NULL;
2470 _cleanup_free_ Unit **array_copy = NULL;
2471 _cleanup_strv_free_ char **tags = NULL;
2472 Unit *u1, *u2, **array;
2473 int r, *fd_array = NULL;
2474 size_t n_fds = 0;
2475 bool found = false;
2476 ssize_t n;
2477
2478 assert(m);
2479 assert(m->notify_fd == fd);
2480
2481 if (revents != EPOLLIN) {
2482 log_warning("Got unexpected poll event for notify fd.");
2483 return 0;
2484 }
2485
2486 n = recvmsg_safe(m->notify_fd, &msghdr, MSG_DONTWAIT|MSG_CMSG_CLOEXEC|MSG_TRUNC);
2487 if (n < 0) {
2488 if (ERRNO_IS_TRANSIENT(n))
2489 return 0; /* Spurious wakeup, try again */
2490 if (n == -EXFULL) {
2491 log_warning("Got message with truncated control data (too many fds sent?), ignoring.");
2492 return 0;
2493 }
2494 /* If this is any other, real error, then let's stop processing this socket. This of course
2495 * means we won't take notification messages anymore, but that's still better than busy
2496 * looping around this: being woken up over and over again but being unable to actually read
2497 * the message off the socket. */
2498 return log_error_errno(n, "Failed to receive notification message: %m");
2499 }
2500
2501 CMSG_FOREACH(cmsg, &msghdr) {
2502 if (cmsg->cmsg_level == SOL_SOCKET && cmsg->cmsg_type == SCM_RIGHTS) {
2503
2504 assert(!fd_array);
2505 fd_array = (int*) CMSG_DATA(cmsg);
2506 n_fds = (cmsg->cmsg_len - CMSG_LEN(0)) / sizeof(int);
2507
2508 } else if (cmsg->cmsg_level == SOL_SOCKET &&
2509 cmsg->cmsg_type == SCM_CREDENTIALS &&
2510 cmsg->cmsg_len == CMSG_LEN(sizeof(struct ucred))) {
2511
2512 assert(!ucred);
2513 ucred = (struct ucred*) CMSG_DATA(cmsg);
2514 }
2515 }
2516
2517 if (n_fds > 0) {
2518 assert(fd_array);
2519
2520 r = fdset_new_array(&fds, fd_array, n_fds);
2521 if (r < 0) {
2522 close_many(fd_array, n_fds);
2523 log_oom();
2524 return 0;
2525 }
2526 }
2527
2528 if (!ucred || !pid_is_valid(ucred->pid)) {
2529 log_warning("Received notify message without valid credentials. Ignoring.");
2530 return 0;
2531 }
2532
2533 if ((size_t) n >= sizeof(buf) || (msghdr.msg_flags & MSG_TRUNC)) {
2534 log_warning("Received notify message exceeded maximum size. Ignoring.");
2535 return 0;
2536 }
2537
2538 /* As extra safety check, let's make sure the string we get doesn't contain embedded NUL bytes.
2539 * We permit one trailing NUL byte in the message, but don't expect it. */
2540 if (n > 1 && memchr(buf, 0, n-1)) {
2541 log_warning("Received notify message with embedded NUL bytes. Ignoring.");
2542 return 0;
2543 }
2544
2545 /* Make sure it's NUL-terminated, then parse it to obtain the tags list. */
2546 buf[n] = 0;
2547 tags = strv_split_newlines(buf);
2548 if (!tags) {
2549 log_oom();
2550 return 0;
2551 }
2552
2553 /* Possibly a barrier fd, let's see. */
2554 if (manager_process_barrier_fd(tags, fds))
2555 return 0;
2556
2557 /* Increase the generation counter used for filtering out duplicate unit invocations. */
2558 m->notifygen++;
2559
2560 /* Notify every unit that might be interested, which might be multiple. */
2561 u1 = manager_get_unit_by_pid_cgroup(m, ucred->pid);
2562 u2 = hashmap_get(m->watch_pids, PID_TO_PTR(ucred->pid));
2563 array = hashmap_get(m->watch_pids, PID_TO_PTR(-ucred->pid));
2564 if (array) {
2565 size_t k = 0;
2566
2567 while (array[k])
2568 k++;
2569
2570 array_copy = newdup(Unit*, array, k+1);
2571 if (!array_copy)
2572 log_oom();
2573 }
2574 /* And now invoke the per-unit callbacks. Note that manager_invoke_notify_message() will handle
2575 * duplicate units make sure we only invoke each unit's handler once. */
2576 if (u1) {
2577 manager_invoke_notify_message(m, u1, ucred, tags, fds);
2578 found = true;
2579 }
2580 if (u2) {
2581 manager_invoke_notify_message(m, u2, ucred, tags, fds);
2582 found = true;
2583 }
2584 if (array_copy)
2585 for (size_t i = 0; array_copy[i]; i++) {
2586 manager_invoke_notify_message(m, array_copy[i], ucred, tags, fds);
2587 found = true;
2588 }
2589
2590 if (!found)
2591 log_warning("Cannot find unit for notify message of PID "PID_FMT", ignoring.", ucred->pid);
2592
2593 if (fdset_size(fds) > 0)
2594 log_warning("Got extra auxiliary fds with notification message, closing them.");
2595
2596 return 0;
2597 }
2598
2599 static void manager_invoke_sigchld_event(
2600 Manager *m,
2601 Unit *u,
2602 const siginfo_t *si) {
2603
2604 assert(m);
2605 assert(u);
2606 assert(si);
2607
2608 /* Already invoked the handler of this unit in this iteration? Then don't process this again */
2609 if (u->sigchldgen == m->sigchldgen)
2610 return;
2611 u->sigchldgen = m->sigchldgen;
2612
2613 log_unit_debug(u, "Child "PID_FMT" belongs to %s.", si->si_pid, u->id);
2614 unit_unwatch_pid(u, si->si_pid);
2615
2616 if (UNIT_VTABLE(u)->sigchld_event)
2617 UNIT_VTABLE(u)->sigchld_event(u, si->si_pid, si->si_code, si->si_status);
2618 }
2619
2620 static int manager_dispatch_sigchld(sd_event_source *source, void *userdata) {
2621 Manager *m = userdata;
2622 siginfo_t si = {};
2623 int r;
2624
2625 assert(source);
2626 assert(m);
2627
2628 /* First we call waitid() for a PID and do not reap the zombie. That way we can still access
2629 * /proc/$PID for it while it is a zombie. */
2630
2631 if (waitid(P_ALL, 0, &si, WEXITED|WNOHANG|WNOWAIT) < 0) {
2632
2633 if (errno != ECHILD)
2634 log_error_errno(errno, "Failed to peek for child with waitid(), ignoring: %m");
2635
2636 goto turn_off;
2637 }
2638
2639 if (si.si_pid <= 0)
2640 goto turn_off;
2641
2642 if (IN_SET(si.si_code, CLD_EXITED, CLD_KILLED, CLD_DUMPED)) {
2643 _cleanup_free_ Unit **array_copy = NULL;
2644 _cleanup_free_ char *name = NULL;
2645 Unit *u1, *u2, **array;
2646
2647 (void) get_process_comm(si.si_pid, &name);
2648
2649 log_debug("Child "PID_FMT" (%s) died (code=%s, status=%i/%s)",
2650 si.si_pid, strna(name),
2651 sigchld_code_to_string(si.si_code),
2652 si.si_status,
2653 strna(si.si_code == CLD_EXITED
2654 ? exit_status_to_string(si.si_status, EXIT_STATUS_FULL)
2655 : signal_to_string(si.si_status)));
2656
2657 /* Increase the generation counter used for filtering out duplicate unit invocations */
2658 m->sigchldgen++;
2659
2660 /* And now figure out the unit this belongs to, it might be multiple... */
2661 u1 = manager_get_unit_by_pid_cgroup(m, si.si_pid);
2662 u2 = hashmap_get(m->watch_pids, PID_TO_PTR(si.si_pid));
2663 array = hashmap_get(m->watch_pids, PID_TO_PTR(-si.si_pid));
2664 if (array) {
2665 size_t n = 0;
2666
2667 /* Count how many entries the array has */
2668 while (array[n])
2669 n++;
2670
2671 /* Make a copy of the array so that we don't trip up on the array changing beneath us */
2672 array_copy = newdup(Unit*, array, n+1);
2673 if (!array_copy)
2674 log_oom();
2675 }
2676
2677 /* Finally, execute them all. Note that u1, u2 and the array might contain duplicates, but
2678 * that's fine, manager_invoke_sigchld_event() will ensure we only invoke the handlers once for
2679 * each iteration. */
2680 if (u1) {
2681 /* We check for oom condition, in case we got SIGCHLD before the oom notification.
2682 * We only do this for the cgroup the PID belonged to. */
2683 (void) unit_check_oom(u1);
2684
2685 /* We check if systemd-oomd performed a kill so that we log and notify appropriately */
2686 (void) unit_check_oomd_kill(u1);
2687
2688 manager_invoke_sigchld_event(m, u1, &si);
2689 }
2690 if (u2)
2691 manager_invoke_sigchld_event(m, u2, &si);
2692 if (array_copy)
2693 for (size_t i = 0; array_copy[i]; i++)
2694 manager_invoke_sigchld_event(m, array_copy[i], &si);
2695 }
2696
2697 /* And now, we actually reap the zombie. */
2698 if (waitid(P_PID, si.si_pid, &si, WEXITED) < 0) {
2699 log_error_errno(errno, "Failed to dequeue child, ignoring: %m");
2700 return 0;
2701 }
2702
2703 return 0;
2704
2705 turn_off:
2706 /* All children processed for now, turn off event source */
2707
2708 r = sd_event_source_set_enabled(m->sigchld_event_source, SD_EVENT_OFF);
2709 if (r < 0)
2710 return log_error_errno(r, "Failed to disable SIGCHLD event source: %m");
2711
2712 return 0;
2713 }
2714
2715 static void manager_start_special(Manager *m, const char *name, JobMode mode) {
2716 Job *job;
2717
2718 if (manager_add_job_by_name_and_warn(m, JOB_START, name, mode, NULL, &job) < 0)
2719 return;
2720
2721 const char *s = unit_status_string(job->unit, NULL);
2722
2723 log_info("Activating special unit %s...", s);
2724
2725 sd_notifyf(false,
2726 "STATUS=Activating special unit %s...", s);
2727 m->status_ready = false;
2728 }
2729
2730 static void manager_handle_ctrl_alt_del(Manager *m) {
2731 /* If the user presses C-A-D more than
2732 * 7 times within 2s, we reboot/shutdown immediately,
2733 * unless it was disabled in system.conf */
2734
2735 if (ratelimit_below(&m->ctrl_alt_del_ratelimit) || m->cad_burst_action == EMERGENCY_ACTION_NONE)
2736 manager_start_special(m, SPECIAL_CTRL_ALT_DEL_TARGET, JOB_REPLACE_IRREVERSIBLY);
2737 else
2738 emergency_action(m, m->cad_burst_action, EMERGENCY_ACTION_WARN, NULL, -1,
2739 "Ctrl-Alt-Del was pressed more than 7 times within 2s");
2740 }
2741
2742 static int manager_dispatch_signal_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
2743 Manager *m = userdata;
2744 ssize_t n;
2745 struct signalfd_siginfo sfsi;
2746 int r;
2747
2748 assert(m);
2749 assert(m->signal_fd == fd);
2750
2751 if (revents != EPOLLIN) {
2752 log_warning("Got unexpected events from signal file descriptor.");
2753 return 0;
2754 }
2755
2756 n = read(m->signal_fd, &sfsi, sizeof(sfsi));
2757 if (n < 0) {
2758 if (ERRNO_IS_TRANSIENT(errno))
2759 return 0;
2760
2761 /* We return an error here, which will kill this handler,
2762 * to avoid a busy loop on read error. */
2763 return log_error_errno(errno, "Reading from signal fd failed: %m");
2764 }
2765 if (n != sizeof(sfsi)) {
2766 log_warning("Truncated read from signal fd (%zu bytes), ignoring!", n);
2767 return 0;
2768 }
2769
2770 log_received_signal(sfsi.ssi_signo == SIGCHLD ||
2771 (sfsi.ssi_signo == SIGTERM && MANAGER_IS_USER(m))
2772 ? LOG_DEBUG : LOG_INFO,
2773 &sfsi);
2774
2775 switch (sfsi.ssi_signo) {
2776
2777 case SIGCHLD:
2778 r = sd_event_source_set_enabled(m->sigchld_event_source, SD_EVENT_ON);
2779 if (r < 0)
2780 log_warning_errno(r, "Failed to enable SIGCHLD event source, ignoring: %m");
2781
2782 break;
2783
2784 case SIGTERM:
2785 if (MANAGER_IS_SYSTEM(m)) {
2786 /* This is for compatibility with the original sysvinit */
2787 if (verify_run_space_and_log("Refusing to reexecute") < 0)
2788 break;
2789
2790 m->objective = MANAGER_REEXECUTE;
2791 break;
2792 }
2793
2794 _fallthrough_;
2795 case SIGINT:
2796 if (MANAGER_IS_SYSTEM(m))
2797 manager_handle_ctrl_alt_del(m);
2798 else
2799 manager_start_special(m, SPECIAL_EXIT_TARGET, JOB_REPLACE_IRREVERSIBLY);
2800 break;
2801
2802 case SIGWINCH:
2803 /* This is a nop on non-init */
2804 if (MANAGER_IS_SYSTEM(m))
2805 manager_start_special(m, SPECIAL_KBREQUEST_TARGET, JOB_REPLACE);
2806
2807 break;
2808
2809 case SIGPWR:
2810 /* This is a nop on non-init */
2811 if (MANAGER_IS_SYSTEM(m))
2812 manager_start_special(m, SPECIAL_SIGPWR_TARGET, JOB_REPLACE);
2813
2814 break;
2815
2816 case SIGUSR1:
2817 if (manager_dbus_is_running(m, false)) {
2818 log_info("Trying to reconnect to bus...");
2819
2820 (void) bus_init_api(m);
2821
2822 if (MANAGER_IS_SYSTEM(m))
2823 (void) bus_init_system(m);
2824 } else
2825 manager_start_special(m, SPECIAL_DBUS_SERVICE, JOB_REPLACE);
2826
2827 break;
2828
2829 case SIGUSR2: {
2830 _cleanup_free_ char *dump = NULL;
2831
2832 r = manager_get_dump_string(m, &dump);
2833 if (r < 0) {
2834 log_warning_errno(errno, "Failed to acquire manager dump: %m");
2835 break;
2836 }
2837
2838 log_dump(LOG_INFO, dump);
2839 break;
2840 }
2841
2842 case SIGHUP:
2843 if (verify_run_space_and_log("Refusing to reload") < 0)
2844 break;
2845
2846 m->objective = MANAGER_RELOAD;
2847 break;
2848
2849 default: {
2850
2851 /* Starting SIGRTMIN+0 */
2852 static const struct {
2853 const char *target;
2854 JobMode mode;
2855 } target_table[] = {
2856 [0] = { SPECIAL_DEFAULT_TARGET, JOB_ISOLATE },
2857 [1] = { SPECIAL_RESCUE_TARGET, JOB_ISOLATE },
2858 [2] = { SPECIAL_EMERGENCY_TARGET, JOB_ISOLATE },
2859 [3] = { SPECIAL_HALT_TARGET, JOB_REPLACE_IRREVERSIBLY },
2860 [4] = { SPECIAL_POWEROFF_TARGET, JOB_REPLACE_IRREVERSIBLY },
2861 [5] = { SPECIAL_REBOOT_TARGET, JOB_REPLACE_IRREVERSIBLY },
2862 [6] = { SPECIAL_KEXEC_TARGET, JOB_REPLACE_IRREVERSIBLY },
2863 };
2864
2865 /* Starting SIGRTMIN+13, so that target halt and system halt are 10 apart */
2866 static const ManagerObjective objective_table[] = {
2867 [0] = MANAGER_HALT,
2868 [1] = MANAGER_POWEROFF,
2869 [2] = MANAGER_REBOOT,
2870 [3] = MANAGER_KEXEC,
2871 };
2872
2873 if ((int) sfsi.ssi_signo >= SIGRTMIN+0 &&
2874 (int) sfsi.ssi_signo < SIGRTMIN+(int) ELEMENTSOF(target_table)) {
2875 int idx = (int) sfsi.ssi_signo - SIGRTMIN;
2876 manager_start_special(m, target_table[idx].target, target_table[idx].mode);
2877 break;
2878 }
2879
2880 if ((int) sfsi.ssi_signo >= SIGRTMIN+13 &&
2881 (int) sfsi.ssi_signo < SIGRTMIN+13+(int) ELEMENTSOF(objective_table)) {
2882 m->objective = objective_table[sfsi.ssi_signo - SIGRTMIN - 13];
2883 break;
2884 }
2885
2886 switch (sfsi.ssi_signo - SIGRTMIN) {
2887
2888 case 20:
2889 manager_override_show_status(m, SHOW_STATUS_YES, "signal");
2890 break;
2891
2892 case 21:
2893 manager_override_show_status(m, SHOW_STATUS_NO, "signal");
2894 break;
2895
2896 case 22:
2897 manager_override_log_level(m, LOG_DEBUG);
2898 break;
2899
2900 case 23:
2901 manager_restore_original_log_level(m);
2902 break;
2903
2904 case 24:
2905 if (MANAGER_IS_USER(m)) {
2906 m->objective = MANAGER_EXIT;
2907 return 0;
2908 }
2909
2910 /* This is a nop on init */
2911 break;
2912
2913 case 25:
2914 m->objective = MANAGER_REEXECUTE;
2915 break;
2916
2917 case 26:
2918 case 29: /* compatibility: used to be mapped to LOG_TARGET_SYSLOG_OR_KMSG */
2919 manager_restore_original_log_target(m);
2920 break;
2921
2922 case 27:
2923 manager_override_log_target(m, LOG_TARGET_CONSOLE);
2924 break;
2925
2926 case 28:
2927 manager_override_log_target(m, LOG_TARGET_KMSG);
2928 break;
2929
2930 default:
2931 log_warning("Got unhandled signal <%s>.", signal_to_string(sfsi.ssi_signo));
2932 }
2933 }}
2934
2935 return 0;
2936 }
2937
2938 static int manager_dispatch_time_change_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
2939 Manager *m = userdata;
2940 Unit *u;
2941
2942 assert(m);
2943
2944 log_struct(LOG_DEBUG,
2945 "MESSAGE_ID=" SD_MESSAGE_TIME_CHANGE_STR,
2946 LOG_MESSAGE("Time has been changed"));
2947
2948 /* Restart the watch */
2949 (void) manager_setup_time_change(m);
2950
2951 HASHMAP_FOREACH(u, m->units)
2952 if (UNIT_VTABLE(u)->time_change)
2953 UNIT_VTABLE(u)->time_change(u);
2954
2955 return 0;
2956 }
2957
2958 static int manager_dispatch_timezone_change(
2959 sd_event_source *source,
2960 const struct inotify_event *e,
2961 void *userdata) {
2962
2963 Manager *m = userdata;
2964 int changed;
2965 Unit *u;
2966
2967 assert(m);
2968
2969 log_debug("inotify event for /etc/localtime");
2970
2971 changed = manager_read_timezone_stat(m);
2972 if (changed <= 0)
2973 return changed;
2974
2975 /* Something changed, restart the watch, to ensure we watch the new /etc/localtime if it changed */
2976 (void) manager_setup_timezone_change(m);
2977
2978 /* Read the new timezone */
2979 tzset();
2980
2981 log_debug("Timezone has been changed (now: %s).", tzname[daylight]);
2982
2983 HASHMAP_FOREACH(u, m->units)
2984 if (UNIT_VTABLE(u)->timezone_change)
2985 UNIT_VTABLE(u)->timezone_change(u);
2986
2987 return 0;
2988 }
2989
2990 static int manager_dispatch_idle_pipe_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
2991 Manager *m = userdata;
2992
2993 assert(m);
2994 assert(m->idle_pipe[2] == fd);
2995
2996 /* There's at least one Type=idle child that just gave up on us waiting for the boot process to
2997 * complete. Let's now turn off any further console output if there's at least one service that needs
2998 * console access, so that from now on our own output should not spill into that service's output
2999 * anymore. After all, we support Type=idle only to beautify console output and it generally is set
3000 * on services that want to own the console exclusively without our interference. */
3001 m->no_console_output = m->n_on_console > 0;
3002
3003 /* Acknowledge the child's request, and let all all other children know too that they shouldn't wait
3004 * any longer by closing the pipes towards them, which is what they are waiting for. */
3005 manager_close_idle_pipe(m);
3006
3007 return 0;
3008 }
3009
3010 static int manager_dispatch_jobs_in_progress(sd_event_source *source, usec_t usec, void *userdata) {
3011 Manager *m = userdata;
3012 int r;
3013
3014 assert(m);
3015 assert(source);
3016
3017 manager_print_jobs_in_progress(m);
3018
3019 r = sd_event_source_set_time_relative(source, JOBS_IN_PROGRESS_PERIOD_USEC);
3020 if (r < 0)
3021 return r;
3022
3023 return sd_event_source_set_enabled(source, SD_EVENT_ONESHOT);
3024 }
3025
3026 int manager_loop(Manager *m) {
3027 RateLimit rl = { .interval = 1*USEC_PER_SEC, .burst = 50000 };
3028 int r;
3029
3030 assert(m);
3031 assert(m->objective == MANAGER_OK); /* Ensure manager_startup() has been called */
3032
3033 manager_check_finished(m);
3034
3035 /* There might still be some zombies hanging around from before we were exec()'ed. Let's reap them. */
3036 r = sd_event_source_set_enabled(m->sigchld_event_source, SD_EVENT_ON);
3037 if (r < 0)
3038 return log_error_errno(r, "Failed to enable SIGCHLD event source: %m");
3039
3040 while (m->objective == MANAGER_OK) {
3041
3042 (void) watchdog_ping();
3043
3044 if (!ratelimit_below(&rl)) {
3045 /* Yay, something is going seriously wrong, pause a little */
3046 log_warning("Looping too fast. Throttling execution a little.");
3047 sleep(1);
3048 }
3049
3050 if (manager_dispatch_load_queue(m) > 0)
3051 continue;
3052
3053 if (manager_dispatch_gc_job_queue(m) > 0)
3054 continue;
3055
3056 if (manager_dispatch_gc_unit_queue(m) > 0)
3057 continue;
3058
3059 if (manager_dispatch_cleanup_queue(m) > 0)
3060 continue;
3061
3062 if (manager_dispatch_cgroup_realize_queue(m) > 0)
3063 continue;
3064
3065 if (manager_dispatch_start_when_upheld_queue(m) > 0)
3066 continue;
3067
3068 if (manager_dispatch_stop_when_bound_queue(m) > 0)
3069 continue;
3070
3071 if (manager_dispatch_stop_when_unneeded_queue(m) > 0)
3072 continue;
3073
3074 if (manager_dispatch_dbus_queue(m) > 0)
3075 continue;
3076
3077 /* Sleep for watchdog runtime wait time */
3078 r = sd_event_run(m->event, watchdog_runtime_wait());
3079 if (r < 0)
3080 return log_error_errno(r, "Failed to run event loop: %m");
3081 }
3082
3083 return m->objective;
3084 }
3085
3086 int manager_load_unit_from_dbus_path(Manager *m, const char *s, sd_bus_error *e, Unit **_u) {
3087 _cleanup_free_ char *n = NULL;
3088 sd_id128_t invocation_id;
3089 Unit *u;
3090 int r;
3091
3092 assert(m);
3093 assert(s);
3094 assert(_u);
3095
3096 r = unit_name_from_dbus_path(s, &n);
3097 if (r < 0)
3098 return r;
3099
3100 /* Permit addressing units by invocation ID: if the passed bus path is suffixed by a 128bit ID then
3101 * we use it as invocation ID. */
3102 r = sd_id128_from_string(n, &invocation_id);
3103 if (r >= 0) {
3104 u = hashmap_get(m->units_by_invocation_id, &invocation_id);
3105 if (u) {
3106 *_u = u;
3107 return 0;
3108 }
3109
3110 return sd_bus_error_setf(e, BUS_ERROR_NO_UNIT_FOR_INVOCATION_ID,
3111 "No unit with the specified invocation ID " SD_ID128_FORMAT_STR " known.",
3112 SD_ID128_FORMAT_VAL(invocation_id));
3113 }
3114
3115 /* If this didn't work, we check if this is a unit name */
3116 if (!unit_name_is_valid(n, UNIT_NAME_PLAIN|UNIT_NAME_INSTANCE)) {
3117 _cleanup_free_ char *nn = NULL;
3118
3119 nn = cescape(n);
3120 return sd_bus_error_setf(e, SD_BUS_ERROR_INVALID_ARGS,
3121 "Unit name %s is neither a valid invocation ID nor unit name.", strnull(nn));
3122 }
3123
3124 r = manager_load_unit(m, n, NULL, e, &u);
3125 if (r < 0)
3126 return r;
3127
3128 *_u = u;
3129 return 0;
3130 }
3131
3132 int manager_get_job_from_dbus_path(Manager *m, const char *s, Job **_j) {
3133 const char *p;
3134 unsigned id;
3135 Job *j;
3136 int r;
3137
3138 assert(m);
3139 assert(s);
3140 assert(_j);
3141
3142 p = startswith(s, "/org/freedesktop/systemd1/job/");
3143 if (!p)
3144 return -EINVAL;
3145
3146 r = safe_atou(p, &id);
3147 if (r < 0)
3148 return r;
3149
3150 j = manager_get_job(m, id);
3151 if (!j)
3152 return -ENOENT;
3153
3154 *_j = j;
3155
3156 return 0;
3157 }
3158
3159 void manager_send_unit_audit(Manager *m, Unit *u, int type, bool success) {
3160
3161 #if HAVE_AUDIT
3162 _cleanup_free_ char *p = NULL;
3163 const char *msg;
3164 int audit_fd, r;
3165
3166 if (!MANAGER_IS_SYSTEM(m))
3167 return;
3168
3169 audit_fd = get_audit_fd();
3170 if (audit_fd < 0)
3171 return;
3172
3173 /* Don't generate audit events if the service was already
3174 * started and we're just deserializing */
3175 if (MANAGER_IS_RELOADING(m))
3176 return;
3177
3178 if (u->type != UNIT_SERVICE)
3179 return;
3180
3181 r = unit_name_to_prefix_and_instance(u->id, &p);
3182 if (r < 0) {
3183 log_error_errno(r, "Failed to extract prefix and instance of unit name: %m");
3184 return;
3185 }
3186
3187 msg = strjoina("unit=", p);
3188 if (audit_log_user_comm_message(audit_fd, type, msg, "systemd", NULL, NULL, NULL, success) < 0) {
3189 if (errno == EPERM)
3190 /* We aren't allowed to send audit messages?
3191 * Then let's not retry again. */
3192 close_audit_fd();
3193 else
3194 log_warning_errno(errno, "Failed to send audit message: %m");
3195 }
3196 #endif
3197
3198 }
3199
3200 void manager_send_unit_plymouth(Manager *m, Unit *u) {
3201 static const union sockaddr_union sa = PLYMOUTH_SOCKET;
3202 _cleanup_free_ char *message = NULL;
3203 _cleanup_close_ int fd = -1;
3204 int n = 0;
3205
3206 /* Don't generate plymouth events if the service was already
3207 * started and we're just deserializing */
3208 if (MANAGER_IS_RELOADING(m))
3209 return;
3210
3211 if (!MANAGER_IS_SYSTEM(m))
3212 return;
3213
3214 if (detect_container() > 0)
3215 return;
3216
3217 if (!IN_SET(u->type, UNIT_SERVICE, UNIT_MOUNT, UNIT_SWAP))
3218 return;
3219
3220 /* We set SOCK_NONBLOCK here so that we rather drop the
3221 * message then wait for plymouth */
3222 fd = socket(AF_UNIX, SOCK_STREAM|SOCK_CLOEXEC|SOCK_NONBLOCK, 0);
3223 if (fd < 0) {
3224 log_error_errno(errno, "socket() failed: %m");
3225 return;
3226 }
3227
3228 if (connect(fd, &sa.sa, SOCKADDR_UN_LEN(sa.un)) < 0) {
3229 if (!IN_SET(errno, EAGAIN, ENOENT) && !ERRNO_IS_DISCONNECT(errno))
3230 log_error_errno(errno, "connect() failed: %m");
3231 return;
3232 }
3233
3234 if (asprintf(&message, "U\002%c%s%n", (int) (strlen(u->id) + 1), u->id, &n) < 0)
3235 return (void) log_oom();
3236
3237 errno = 0;
3238 if (write(fd, message, n + 1) != n + 1)
3239 if (!IN_SET(errno, EAGAIN, ENOENT) && !ERRNO_IS_DISCONNECT(errno))
3240 log_error_errno(errno, "Failed to write Plymouth message: %m");
3241 }
3242
3243 usec_t manager_get_watchdog(Manager *m, WatchdogType t) {
3244 assert(m);
3245
3246 if (MANAGER_IS_USER(m))
3247 return USEC_INFINITY;
3248
3249 if (timestamp_is_set(m->watchdog_overridden[t]))
3250 return m->watchdog_overridden[t];
3251
3252 return m->watchdog[t];
3253 }
3254
3255 void manager_set_watchdog(Manager *m, WatchdogType t, usec_t timeout) {
3256
3257 assert(m);
3258
3259 if (MANAGER_IS_USER(m))
3260 return;
3261
3262 if (m->watchdog[t] == timeout)
3263 return;
3264
3265 if (t == WATCHDOG_RUNTIME) {
3266 if (!timestamp_is_set(m->watchdog_overridden[WATCHDOG_RUNTIME]))
3267 (void) watchdog_setup(timeout);
3268 } else if (t == WATCHDOG_PRETIMEOUT)
3269 if (m->watchdog_overridden[WATCHDOG_PRETIMEOUT] == USEC_INFINITY)
3270 (void) watchdog_setup_pretimeout(timeout);
3271
3272 m->watchdog[t] = timeout;
3273 }
3274
3275 void manager_override_watchdog(Manager *m, WatchdogType t, usec_t timeout) {
3276
3277 assert(m);
3278
3279 if (MANAGER_IS_USER(m))
3280 return;
3281
3282 if (m->watchdog_overridden[t] == timeout)
3283 return;
3284
3285 if (t == WATCHDOG_RUNTIME) {
3286 usec_t usec = timestamp_is_set(timeout) ? timeout : m->watchdog[t];
3287
3288 (void) watchdog_setup(usec);
3289 } else if (t == WATCHDOG_PRETIMEOUT)
3290 (void) watchdog_setup_pretimeout(timeout);
3291
3292 m->watchdog_overridden[t] = timeout;
3293 }
3294
3295 int manager_set_watchdog_pretimeout_governor(Manager *m, const char *governor) {
3296 _cleanup_free_ char *p = NULL;
3297 int r;
3298
3299 assert(m);
3300
3301 if (MANAGER_IS_USER(m))
3302 return 0;
3303
3304 if (streq_ptr(m->watchdog_pretimeout_governor, governor))
3305 return 0;
3306
3307 p = strdup(governor);
3308 if (!p)
3309 return -ENOMEM;
3310
3311 r = watchdog_setup_pretimeout_governor(governor);
3312 if (r < 0)
3313 return r;
3314
3315 return free_and_replace(m->watchdog_pretimeout_governor, p);
3316 }
3317
3318 int manager_override_watchdog_pretimeout_governor(Manager *m, const char *governor) {
3319 _cleanup_free_ char *p = NULL;
3320 int r;
3321
3322 assert(m);
3323
3324 if (MANAGER_IS_USER(m))
3325 return 0;
3326
3327 if (streq_ptr(m->watchdog_pretimeout_governor_overridden, governor))
3328 return 0;
3329
3330 p = strdup(governor);
3331 if (!p)
3332 return -ENOMEM;
3333
3334 r = watchdog_setup_pretimeout_governor(governor);
3335 if (r < 0)
3336 return r;
3337
3338 return free_and_replace(m->watchdog_pretimeout_governor_overridden, p);
3339 }
3340
3341 int manager_reload(Manager *m) {
3342 _unused_ _cleanup_(manager_reloading_stopp) Manager *reloading = NULL;
3343 _cleanup_fdset_free_ FDSet *fds = NULL;
3344 _cleanup_fclose_ FILE *f = NULL;
3345 int r;
3346
3347 assert(m);
3348
3349 r = manager_open_serialization(m, &f);
3350 if (r < 0)
3351 return log_error_errno(r, "Failed to create serialization file: %m");
3352
3353 fds = fdset_new();
3354 if (!fds)
3355 return log_oom();
3356
3357 /* We are officially in reload mode from here on. */
3358 reloading = manager_reloading_start(m);
3359
3360 r = manager_serialize(m, f, fds, false);
3361 if (r < 0)
3362 return r;
3363
3364 if (fseeko(f, 0, SEEK_SET) < 0)
3365 return log_error_errno(errno, "Failed to seek to beginning of serialization: %m");
3366
3367 /* 💀 This is the point of no return, from here on there is no way back. 💀 */
3368 reloading = NULL;
3369
3370 bus_manager_send_reloading(m, true);
3371
3372 /* Start by flushing out all jobs and units, all generated units, all runtime environments, all dynamic users
3373 * and everything else that is worth flushing out. We'll get it all back from the serialization — if we need
3374 * it. */
3375
3376 manager_clear_jobs_and_units(m);
3377 lookup_paths_flush_generator(&m->lookup_paths);
3378 lookup_paths_free(&m->lookup_paths);
3379 exec_runtime_vacuum(m);
3380 dynamic_user_vacuum(m, false);
3381 m->uid_refs = hashmap_free(m->uid_refs);
3382 m->gid_refs = hashmap_free(m->gid_refs);
3383
3384 r = lookup_paths_init_or_warn(&m->lookup_paths, m->unit_file_scope, 0, NULL);
3385 if (r < 0)
3386 return r;
3387
3388 (void) manager_run_environment_generators(m);
3389 (void) manager_run_generators(m);
3390
3391 lookup_paths_log(&m->lookup_paths);
3392
3393 /* We flushed out generated files, for which we don't watch mtime, so we should flush the old map. */
3394 manager_free_unit_name_maps(m);
3395
3396 /* First, enumerate what we can from kernel and suchlike */
3397 manager_enumerate_perpetual(m);
3398 manager_enumerate(m);
3399
3400 /* Second, deserialize our stored data */
3401 r = manager_deserialize(m, f, fds);
3402 if (r < 0)
3403 log_warning_errno(r, "Deserialization failed, proceeding anyway: %m");
3404
3405 /* We don't need the serialization anymore */
3406 f = safe_fclose(f);
3407
3408 /* Re-register notify_fd as event source, and set up other sockets/communication channels we might need */
3409 (void) manager_setup_notify(m);
3410 (void) manager_setup_cgroups_agent(m);
3411 (void) manager_setup_user_lookup_fd(m);
3412
3413 /* Third, fire things up! */
3414 manager_coldplug(m);
3415
3416 /* Clean up runtime objects no longer referenced */
3417 manager_vacuum(m);
3418
3419 /* Clean up deserialized tracked clients */
3420 m->deserialized_subscribed = strv_free(m->deserialized_subscribed);
3421
3422 /* Consider the reload process complete now. */
3423 assert(m->n_reloading > 0);
3424 m->n_reloading--;
3425
3426 manager_ready(m);
3427
3428 m->send_reloading_done = true;
3429 return 0;
3430 }
3431
3432 void manager_reset_failed(Manager *m) {
3433 Unit *u;
3434
3435 assert(m);
3436
3437 HASHMAP_FOREACH(u, m->units)
3438 unit_reset_failed(u);
3439 }
3440
3441 bool manager_unit_inactive_or_pending(Manager *m, const char *name) {
3442 Unit *u;
3443
3444 assert(m);
3445 assert(name);
3446
3447 /* Returns true if the unit is inactive or going down */
3448 u = manager_get_unit(m, name);
3449 if (!u)
3450 return true;
3451
3452 return unit_inactive_or_pending(u);
3453 }
3454
3455 static void log_taint_string(Manager *m) {
3456 _cleanup_free_ char *taint = NULL;
3457
3458 assert(m);
3459
3460 if (MANAGER_IS_USER(m) || m->taint_logged)
3461 return;
3462
3463 m->taint_logged = true; /* only check for taint once */
3464
3465 taint = manager_taint_string(m);
3466 if (isempty(taint))
3467 return;
3468
3469 log_struct(LOG_NOTICE,
3470 LOG_MESSAGE("System is tainted: %s", taint),
3471 "TAINT=%s", taint,
3472 "MESSAGE_ID=" SD_MESSAGE_TAINTED_STR);
3473 }
3474
3475 static void manager_notify_finished(Manager *m) {
3476 usec_t firmware_usec, loader_usec, kernel_usec, initrd_usec, userspace_usec, total_usec;
3477
3478 if (MANAGER_IS_TEST_RUN(m))
3479 return;
3480
3481 if (MANAGER_IS_SYSTEM(m) && detect_container() <= 0) {
3482 char buf[FORMAT_TIMESPAN_MAX + STRLEN(" (firmware) + ") + FORMAT_TIMESPAN_MAX + STRLEN(" (loader) + ")]
3483 = {};
3484 char *p = buf;
3485 size_t size = sizeof buf;
3486
3487 /* Note that MANAGER_TIMESTAMP_KERNEL's monotonic value is always at 0, and
3488 * MANAGER_TIMESTAMP_FIRMWARE's and MANAGER_TIMESTAMP_LOADER's monotonic value should be considered
3489 * negative values. */
3490
3491 firmware_usec = m->timestamps[MANAGER_TIMESTAMP_FIRMWARE].monotonic - m->timestamps[MANAGER_TIMESTAMP_LOADER].monotonic;
3492 loader_usec = m->timestamps[MANAGER_TIMESTAMP_LOADER].monotonic - m->timestamps[MANAGER_TIMESTAMP_KERNEL].monotonic;
3493 userspace_usec = m->timestamps[MANAGER_TIMESTAMP_FINISH].monotonic - m->timestamps[MANAGER_TIMESTAMP_USERSPACE].monotonic;
3494 total_usec = m->timestamps[MANAGER_TIMESTAMP_FIRMWARE].monotonic + m->timestamps[MANAGER_TIMESTAMP_FINISH].monotonic;
3495
3496 if (firmware_usec > 0)
3497 size = strpcpyf(&p, size, "%s (firmware) + ", FORMAT_TIMESPAN(firmware_usec, USEC_PER_MSEC));
3498 if (loader_usec > 0)
3499 size = strpcpyf(&p, size, "%s (loader) + ", FORMAT_TIMESPAN(loader_usec, USEC_PER_MSEC));
3500
3501 if (dual_timestamp_is_set(&m->timestamps[MANAGER_TIMESTAMP_INITRD])) {
3502
3503 /* The initrd case on bare-metal */
3504 kernel_usec = m->timestamps[MANAGER_TIMESTAMP_INITRD].monotonic - m->timestamps[MANAGER_TIMESTAMP_KERNEL].monotonic;
3505 initrd_usec = m->timestamps[MANAGER_TIMESTAMP_USERSPACE].monotonic - m->timestamps[MANAGER_TIMESTAMP_INITRD].monotonic;
3506
3507 log_struct(LOG_INFO,
3508 "MESSAGE_ID=" SD_MESSAGE_STARTUP_FINISHED_STR,
3509 "KERNEL_USEC="USEC_FMT, kernel_usec,
3510 "INITRD_USEC="USEC_FMT, initrd_usec,
3511 "USERSPACE_USEC="USEC_FMT, userspace_usec,
3512 LOG_MESSAGE("Startup finished in %s%s (kernel) + %s (initrd) + %s (userspace) = %s.",
3513 buf,
3514 FORMAT_TIMESPAN(kernel_usec, USEC_PER_MSEC),
3515 FORMAT_TIMESPAN(initrd_usec, USEC_PER_MSEC),
3516 FORMAT_TIMESPAN(userspace_usec, USEC_PER_MSEC),
3517 FORMAT_TIMESPAN(total_usec, USEC_PER_MSEC)));
3518 } else {
3519 /* The initrd-less case on bare-metal */
3520
3521 kernel_usec = m->timestamps[MANAGER_TIMESTAMP_USERSPACE].monotonic - m->timestamps[MANAGER_TIMESTAMP_KERNEL].monotonic;
3522 initrd_usec = 0;
3523
3524 log_struct(LOG_INFO,
3525 "MESSAGE_ID=" SD_MESSAGE_STARTUP_FINISHED_STR,
3526 "KERNEL_USEC="USEC_FMT, kernel_usec,
3527 "USERSPACE_USEC="USEC_FMT, userspace_usec,
3528 LOG_MESSAGE("Startup finished in %s%s (kernel) + %s (userspace) = %s.",
3529 buf,
3530 FORMAT_TIMESPAN(kernel_usec, USEC_PER_MSEC),
3531 FORMAT_TIMESPAN(userspace_usec, USEC_PER_MSEC),
3532 FORMAT_TIMESPAN(total_usec, USEC_PER_MSEC)));
3533 }
3534 } else {
3535 /* The container and --user case */
3536 firmware_usec = loader_usec = initrd_usec = kernel_usec = 0;
3537 total_usec = userspace_usec = m->timestamps[MANAGER_TIMESTAMP_FINISH].monotonic - m->timestamps[MANAGER_TIMESTAMP_USERSPACE].monotonic;
3538
3539 log_struct(LOG_INFO,
3540 "MESSAGE_ID=" SD_MESSAGE_USER_STARTUP_FINISHED_STR,
3541 "USERSPACE_USEC="USEC_FMT, userspace_usec,
3542 LOG_MESSAGE("Startup finished in %s.",
3543 FORMAT_TIMESPAN(total_usec, USEC_PER_MSEC)));
3544 }
3545
3546 bus_manager_send_finished(m, firmware_usec, loader_usec, kernel_usec, initrd_usec, userspace_usec, total_usec);
3547
3548 log_taint_string(m);
3549 }
3550
3551 static void user_manager_send_ready(Manager *m) {
3552 int r;
3553
3554 assert(m);
3555
3556 /* We send READY=1 on reaching basic.target only when running in --user mode. */
3557 if (!MANAGER_IS_USER(m) || m->ready_sent)
3558 return;
3559
3560 r = sd_notify(false,
3561 "READY=1\n"
3562 "STATUS=Reached " SPECIAL_BASIC_TARGET ".");
3563 if (r < 0)
3564 log_warning_errno(r, "Failed to send readiness notification, ignoring: %m");
3565
3566 m->ready_sent = true;
3567 m->status_ready = false;
3568 }
3569
3570 static void manager_send_ready(Manager *m) {
3571 int r;
3572
3573 if (m->ready_sent && m->status_ready)
3574 /* Skip the notification if nothing changed. */
3575 return;
3576
3577 r = sd_notify(false,
3578 "READY=1\n"
3579 "STATUS=Ready.");
3580 if (r < 0)
3581 log_full_errno(m->ready_sent ? LOG_DEBUG : LOG_WARNING, r,
3582 "Failed to send readiness notification, ignoring: %m");
3583
3584 m->ready_sent = m->status_ready = true;
3585 }
3586
3587 static void manager_check_basic_target(Manager *m) {
3588 Unit *u;
3589
3590 assert(m);
3591
3592 /* Small shortcut */
3593 if (m->ready_sent && m->taint_logged)
3594 return;
3595
3596 u = manager_get_unit(m, SPECIAL_BASIC_TARGET);
3597 if (!u || !UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u)))
3598 return;
3599
3600 /* For user managers, send out READY=1 as soon as we reach basic.target */
3601 user_manager_send_ready(m);
3602
3603 /* Log the taint string as soon as we reach basic.target */
3604 log_taint_string(m);
3605 }
3606
3607 void manager_check_finished(Manager *m) {
3608 assert(m);
3609
3610 if (MANAGER_IS_RELOADING(m))
3611 return;
3612
3613 /* Verify that we have entered the event loop already, and not left it again. */
3614 if (!MANAGER_IS_RUNNING(m))
3615 return;
3616
3617 manager_check_basic_target(m);
3618
3619 if (hashmap_size(m->jobs) > 0) {
3620 if (m->jobs_in_progress_event_source)
3621 /* Ignore any failure, this is only for feedback */
3622 (void) sd_event_source_set_time(m->jobs_in_progress_event_source,
3623 manager_watch_jobs_next_time(m));
3624 return;
3625 }
3626
3627 /* The jobs hashmap tends to grow a lot during boot, and then it's not reused until shutdown. Let's
3628 kill the hashmap if it is relatively large. */
3629 if (hashmap_buckets(m->jobs) > hashmap_size(m->units) / 10)
3630 m->jobs = hashmap_free(m->jobs);
3631
3632 manager_send_ready(m);
3633
3634 /* Notify Type=idle units that we are done now */
3635 manager_close_idle_pipe(m);
3636
3637 if (MANAGER_IS_FINISHED(m))
3638 return;
3639
3640 manager_flip_auto_status(m, false, "boot finished");
3641
3642 /* Turn off confirm spawn now */
3643 m->confirm_spawn = NULL;
3644
3645 /* No need to update ask password status when we're going non-interactive */
3646 manager_close_ask_password(m);
3647
3648 /* This is no longer the first boot */
3649 manager_set_first_boot(m, false);
3650
3651 dual_timestamp_get(m->timestamps + MANAGER_TIMESTAMP_FINISH);
3652
3653 manager_notify_finished(m);
3654
3655 manager_invalidate_startup_units(m);
3656 }
3657
3658 static bool generator_path_any(const char* const* paths) {
3659 bool found = false;
3660
3661 /* Optimize by skipping the whole process by not creating output directories
3662 * if no generators are found. */
3663 STRV_FOREACH(path, paths)
3664 if (access(*path, F_OK) == 0)
3665 found = true;
3666 else if (errno != ENOENT)
3667 log_warning_errno(errno, "Failed to open generator directory %s: %m", *path);
3668
3669 return found;
3670 }
3671
3672 static int manager_run_environment_generators(Manager *m) {
3673 char **tmp = NULL; /* this is only used in the forked process, no cleanup here */
3674 _cleanup_strv_free_ char **paths = NULL;
3675 void* args[] = {
3676 [STDOUT_GENERATE] = &tmp,
3677 [STDOUT_COLLECT] = &tmp,
3678 [STDOUT_CONSUME] = &m->transient_environment,
3679 };
3680 int r;
3681
3682 if (MANAGER_IS_TEST_RUN(m) && !(m->test_run_flags & MANAGER_TEST_RUN_ENV_GENERATORS))
3683 return 0;
3684
3685 paths = env_generator_binary_paths(MANAGER_IS_SYSTEM(m));
3686 if (!paths)
3687 return log_oom();
3688
3689 if (!generator_path_any((const char* const*) paths))
3690 return 0;
3691
3692 RUN_WITH_UMASK(0022)
3693 r = execute_directories((const char* const*) paths, DEFAULT_TIMEOUT_USEC, gather_environment,
3694 args, NULL, m->transient_environment,
3695 EXEC_DIR_PARALLEL | EXEC_DIR_IGNORE_ERRORS | EXEC_DIR_SET_SYSTEMD_EXEC_PID);
3696 return r;
3697 }
3698
3699 static int build_generator_environment(Manager *m, char ***ret) {
3700 _cleanup_strv_free_ char **nl = NULL;
3701 Virtualization v;
3702 int r;
3703
3704 assert(m);
3705 assert(ret);
3706
3707 /* Generators oftentimes want to know some basic facts about the environment they run in, in order to
3708 * adjust generated units to that. Let's pass down some bits of information that are easy for us to
3709 * determine (but a bit harder for generator scripts to determine), as environment variables. */
3710
3711 nl = strv_copy(m->transient_environment);
3712 if (!nl)
3713 return -ENOMEM;
3714
3715 r = strv_env_assign(&nl, "SYSTEMD_SCOPE", MANAGER_IS_SYSTEM(m) ? "system" : "user");
3716 if (r < 0)
3717 return r;
3718
3719 if (MANAGER_IS_SYSTEM(m)) {
3720 /* Note that $SYSTEMD_IN_INITRD may be used to override the initrd detection in much of our
3721 * codebase. This is hence more than purely informational. It will shortcut detection of the
3722 * initrd state if generators invoke our own tools. But that's OK, as it would come to the
3723 * same results (hopefully). */
3724 r = strv_env_assign(&nl, "SYSTEMD_IN_INITRD", one_zero(in_initrd()));
3725 if (r < 0)
3726 return r;
3727
3728 if (m->first_boot >= 0) {
3729 r = strv_env_assign(&nl, "SYSTEMD_FIRST_BOOT", one_zero(m->first_boot));
3730 if (r < 0)
3731 return r;
3732 }
3733 }
3734
3735 v = detect_virtualization();
3736 if (v < 0)
3737 log_debug_errno(v, "Failed to detect virtualization, ignoring: %m");
3738 else if (v > 0) {
3739 const char *s;
3740
3741 s = strjoina(VIRTUALIZATION_IS_VM(v) ? "vm:" :
3742 VIRTUALIZATION_IS_CONTAINER(v) ? "container:" : ":",
3743 virtualization_to_string(v));
3744
3745 r = strv_env_assign(&nl, "SYSTEMD_VIRTUALIZATION", s);
3746 if (r < 0)
3747 return r;
3748 }
3749
3750 r = strv_env_assign(&nl, "SYSTEMD_ARCHITECTURE", architecture_to_string(uname_architecture()));
3751 if (r < 0)
3752 return r;
3753
3754 *ret = TAKE_PTR(nl);
3755 return 0;
3756 }
3757
3758 static int manager_run_generators(Manager *m) {
3759 _cleanup_strv_free_ char **paths = NULL, **ge = NULL;
3760 int r;
3761
3762 assert(m);
3763
3764 if (MANAGER_IS_TEST_RUN(m) && !(m->test_run_flags & MANAGER_TEST_RUN_GENERATORS))
3765 return 0;
3766
3767 paths = generator_binary_paths(m->unit_file_scope);
3768 if (!paths)
3769 return log_oom();
3770
3771 if (!generator_path_any((const char* const*) paths))
3772 return 0;
3773
3774 r = lookup_paths_mkdir_generator(&m->lookup_paths);
3775 if (r < 0) {
3776 log_error_errno(r, "Failed to create generator directories: %m");
3777 goto finish;
3778 }
3779
3780 const char *argv[] = {
3781 NULL, /* Leave this empty, execute_directory() will fill something in */
3782 m->lookup_paths.generator,
3783 m->lookup_paths.generator_early,
3784 m->lookup_paths.generator_late,
3785 NULL,
3786 };
3787
3788 r = build_generator_environment(m, &ge);
3789 if (r < 0) {
3790 log_error_errno(r, "Failed to build generator environment: %m");
3791 goto finish;
3792 }
3793
3794 RUN_WITH_UMASK(0022)
3795 (void) execute_directories(
3796 (const char* const*) paths,
3797 DEFAULT_TIMEOUT_USEC,
3798 /* callbacks= */ NULL, /* callback_args= */ NULL,
3799 (char**) argv,
3800 ge,
3801 EXEC_DIR_PARALLEL | EXEC_DIR_IGNORE_ERRORS | EXEC_DIR_SET_SYSTEMD_EXEC_PID);
3802
3803 r = 0;
3804
3805 finish:
3806 lookup_paths_trim_generator(&m->lookup_paths);
3807 return r;
3808 }
3809
3810 int manager_transient_environment_add(Manager *m, char **plus) {
3811 char **a;
3812
3813 assert(m);
3814
3815 if (strv_isempty(plus))
3816 return 0;
3817
3818 a = strv_env_merge(m->transient_environment, plus);
3819 if (!a)
3820 return log_oom();
3821
3822 sanitize_environment(a);
3823
3824 return strv_free_and_replace(m->transient_environment, a);
3825 }
3826
3827 int manager_client_environment_modify(
3828 Manager *m,
3829 char **minus,
3830 char **plus) {
3831
3832 char **a = NULL, **b = NULL, **l;
3833
3834 assert(m);
3835
3836 if (strv_isempty(minus) && strv_isempty(plus))
3837 return 0;
3838
3839 l = m->client_environment;
3840
3841 if (!strv_isempty(minus)) {
3842 a = strv_env_delete(l, 1, minus);
3843 if (!a)
3844 return -ENOMEM;
3845
3846 l = a;
3847 }
3848
3849 if (!strv_isempty(plus)) {
3850 b = strv_env_merge(l, plus);
3851 if (!b) {
3852 strv_free(a);
3853 return -ENOMEM;
3854 }
3855
3856 l = b;
3857 }
3858
3859 if (m->client_environment != l)
3860 strv_free(m->client_environment);
3861
3862 if (a != l)
3863 strv_free(a);
3864 if (b != l)
3865 strv_free(b);
3866
3867 m->client_environment = sanitize_environment(l);
3868 return 0;
3869 }
3870
3871 int manager_get_effective_environment(Manager *m, char ***ret) {
3872 char **l;
3873
3874 assert(m);
3875 assert(ret);
3876
3877 l = strv_env_merge(m->transient_environment, m->client_environment);
3878 if (!l)
3879 return -ENOMEM;
3880
3881 *ret = l;
3882 return 0;
3883 }
3884
3885 int manager_set_default_smack_process_label(Manager *m, const char *label) {
3886 assert(m);
3887
3888 #ifdef SMACK_DEFAULT_PROCESS_LABEL
3889 if (!label)
3890 return free_and_strdup(&m->default_smack_process_label, SMACK_DEFAULT_PROCESS_LABEL);
3891 #endif
3892 if (streq_ptr(label, "/"))
3893 return free_and_strdup(&m->default_smack_process_label, NULL);
3894
3895 return free_and_strdup(&m->default_smack_process_label, label);
3896 }
3897
3898 int manager_set_default_rlimits(Manager *m, struct rlimit **default_rlimit) {
3899 assert(m);
3900
3901 for (unsigned i = 0; i < _RLIMIT_MAX; i++) {
3902 m->rlimit[i] = mfree(m->rlimit[i]);
3903
3904 if (!default_rlimit[i])
3905 continue;
3906
3907 m->rlimit[i] = newdup(struct rlimit, default_rlimit[i], 1);
3908 if (!m->rlimit[i])
3909 return log_oom();
3910 }
3911
3912 return 0;
3913 }
3914
3915 void manager_recheck_dbus(Manager *m) {
3916 assert(m);
3917
3918 /* Connects to the bus if the dbus service and socket are running. If we are running in user mode
3919 * this is all it does. In system mode we'll also connect to the system bus (which will most likely
3920 * just reuse the connection of the API bus). That's because the system bus after all runs as service
3921 * of the system instance, while in the user instance we can assume it's already there. */
3922
3923 if (MANAGER_IS_RELOADING(m))
3924 return; /* don't check while we are reloading… */
3925
3926 if (manager_dbus_is_running(m, false)) {
3927 (void) bus_init_api(m);
3928
3929 if (MANAGER_IS_SYSTEM(m))
3930 (void) bus_init_system(m);
3931 } else {
3932 (void) bus_done_api(m);
3933
3934 if (MANAGER_IS_SYSTEM(m))
3935 (void) bus_done_system(m);
3936 }
3937 }
3938
3939 static bool manager_journal_is_running(Manager *m) {
3940 Unit *u;
3941
3942 assert(m);
3943
3944 if (MANAGER_IS_TEST_RUN(m))
3945 return false;
3946
3947 /* If we are the user manager we can safely assume that the journal is up */
3948 if (!MANAGER_IS_SYSTEM(m))
3949 return true;
3950
3951 /* Check that the socket is not only up, but in RUNNING state */
3952 u = manager_get_unit(m, SPECIAL_JOURNALD_SOCKET);
3953 if (!u)
3954 return false;
3955 if (SOCKET(u)->state != SOCKET_RUNNING)
3956 return false;
3957
3958 /* Similar, check if the daemon itself is fully up, too */
3959 u = manager_get_unit(m, SPECIAL_JOURNALD_SERVICE);
3960 if (!u)
3961 return false;
3962 if (!IN_SET(SERVICE(u)->state, SERVICE_RELOAD, SERVICE_RUNNING))
3963 return false;
3964
3965 return true;
3966 }
3967
3968 void disable_printk_ratelimit(void) {
3969 /* Disable kernel's printk ratelimit.
3970 *
3971 * Logging to /dev/kmsg is most useful during early boot and shutdown, where normal logging
3972 * mechanisms are not available. The semantics of this sysctl are such that any kernel command-line
3973 * setting takes precedence. */
3974 int r;
3975
3976 r = sysctl_write("kernel/printk_devkmsg", "on");
3977 if (r < 0)
3978 log_debug_errno(r, "Failed to set sysctl kernel.printk_devkmsg=on: %m");
3979 }
3980
3981 void manager_recheck_journal(Manager *m) {
3982
3983 assert(m);
3984
3985 /* Don't bother with this unless we are in the special situation of being PID 1 */
3986 if (getpid_cached() != 1)
3987 return;
3988
3989 /* Don't check this while we are reloading, things might still change */
3990 if (MANAGER_IS_RELOADING(m))
3991 return;
3992
3993 /* The journal is fully and entirely up? If so, let's permit logging to it, if that's configured. If
3994 * the journal is down, don't ever log to it, otherwise we might end up deadlocking ourselves as we
3995 * might trigger an activation ourselves we can't fulfill. */
3996 log_set_prohibit_ipc(!manager_journal_is_running(m));
3997 log_open();
3998 }
3999
4000 static ShowStatus manager_get_show_status(Manager *m) {
4001 assert(m);
4002
4003 if (MANAGER_IS_USER(m))
4004 return _SHOW_STATUS_INVALID;
4005
4006 if (m->show_status_overridden != _SHOW_STATUS_INVALID)
4007 return m->show_status_overridden;
4008
4009 return m->show_status;
4010 }
4011
4012 bool manager_get_show_status_on(Manager *m) {
4013 assert(m);
4014
4015 return show_status_on(manager_get_show_status(m));
4016 }
4017
4018 static void set_show_status_marker(bool b) {
4019 if (b)
4020 (void) touch("/run/systemd/show-status");
4021 else
4022 (void) unlink("/run/systemd/show-status");
4023 }
4024
4025 void manager_set_show_status(Manager *m, ShowStatus mode, const char *reason) {
4026 assert(m);
4027 assert(reason);
4028 assert(mode >= 0 && mode < _SHOW_STATUS_MAX);
4029
4030 if (MANAGER_IS_USER(m))
4031 return;
4032
4033 if (mode == m->show_status)
4034 return;
4035
4036 if (m->show_status_overridden == _SHOW_STATUS_INVALID) {
4037 bool enabled;
4038
4039 enabled = show_status_on(mode);
4040 log_debug("%s (%s) showing of status (%s).",
4041 enabled ? "Enabling" : "Disabling",
4042 strna(show_status_to_string(mode)),
4043 reason);
4044
4045 set_show_status_marker(enabled);
4046 }
4047
4048 m->show_status = mode;
4049 }
4050
4051 void manager_override_show_status(Manager *m, ShowStatus mode, const char *reason) {
4052 assert(m);
4053 assert(mode < _SHOW_STATUS_MAX);
4054
4055 if (MANAGER_IS_USER(m))
4056 return;
4057
4058 if (mode == m->show_status_overridden)
4059 return;
4060
4061 m->show_status_overridden = mode;
4062
4063 if (mode == _SHOW_STATUS_INVALID)
4064 mode = m->show_status;
4065
4066 log_debug("%s (%s) showing of status (%s).",
4067 m->show_status_overridden != _SHOW_STATUS_INVALID ? "Overriding" : "Restoring",
4068 strna(show_status_to_string(mode)),
4069 reason);
4070
4071 set_show_status_marker(show_status_on(mode));
4072 }
4073
4074 const char *manager_get_confirm_spawn(Manager *m) {
4075 static int last_errno = 0;
4076 struct stat st;
4077 int r;
4078
4079 assert(m);
4080
4081 /* Here's the deal: we want to test the validity of the console but don't want
4082 * PID1 to go through the whole console process which might block. But we also
4083 * want to warn the user only once if something is wrong with the console so we
4084 * cannot do the sanity checks after spawning our children. So here we simply do
4085 * really basic tests to hopefully trap common errors.
4086 *
4087 * If the console suddenly disappear at the time our children will really it
4088 * then they will simply fail to acquire it and a positive answer will be
4089 * assumed. New children will fall back to /dev/console though.
4090 *
4091 * Note: TTYs are devices that can come and go any time, and frequently aren't
4092 * available yet during early boot (consider a USB rs232 dongle...). If for any
4093 * reason the configured console is not ready, we fall back to the default
4094 * console. */
4095
4096 if (!m->confirm_spawn || path_equal(m->confirm_spawn, "/dev/console"))
4097 return m->confirm_spawn;
4098
4099 if (stat(m->confirm_spawn, &st) < 0) {
4100 r = -errno;
4101 goto fail;
4102 }
4103
4104 if (!S_ISCHR(st.st_mode)) {
4105 r = -ENOTTY;
4106 goto fail;
4107 }
4108
4109 last_errno = 0;
4110 return m->confirm_spawn;
4111
4112 fail:
4113 if (last_errno != r)
4114 last_errno = log_warning_errno(r, "Failed to open %s, using default console: %m", m->confirm_spawn);
4115
4116 return "/dev/console";
4117 }
4118
4119 void manager_set_first_boot(Manager *m, bool b) {
4120 assert(m);
4121
4122 if (!MANAGER_IS_SYSTEM(m))
4123 return;
4124
4125 if (m->first_boot != (int) b) {
4126 if (b)
4127 (void) touch("/run/systemd/first-boot");
4128 else
4129 (void) unlink("/run/systemd/first-boot");
4130 }
4131
4132 m->first_boot = b;
4133 }
4134
4135 void manager_disable_confirm_spawn(void) {
4136 (void) touch("/run/systemd/confirm_spawn_disabled");
4137 }
4138
4139 bool manager_is_confirm_spawn_disabled(Manager *m) {
4140 if (!m->confirm_spawn)
4141 return true;
4142
4143 return access("/run/systemd/confirm_spawn_disabled", F_OK) >= 0;
4144 }
4145
4146 static bool manager_should_show_status(Manager *m, StatusType type) {
4147 assert(m);
4148
4149 if (!MANAGER_IS_SYSTEM(m))
4150 return false;
4151
4152 if (m->no_console_output)
4153 return false;
4154
4155 if (!IN_SET(manager_state(m), MANAGER_INITIALIZING, MANAGER_STARTING, MANAGER_STOPPING))
4156 return false;
4157
4158 /* If we cannot find out the status properly, just proceed. */
4159 if (type != STATUS_TYPE_EMERGENCY && manager_check_ask_password(m) > 0)
4160 return false;
4161
4162 if (type == STATUS_TYPE_NOTICE && m->show_status != SHOW_STATUS_NO)
4163 return true;
4164
4165 return manager_get_show_status_on(m);
4166 }
4167
4168 void manager_status_printf(Manager *m, StatusType type, const char *status, const char *format, ...) {
4169 va_list ap;
4170
4171 /* If m is NULL, assume we're after shutdown and let the messages through. */
4172
4173 if (m && !manager_should_show_status(m, type))
4174 return;
4175
4176 /* XXX We should totally drop the check for ephemeral here
4177 * and thus effectively make 'Type=idle' pointless. */
4178 if (type == STATUS_TYPE_EPHEMERAL && m && m->n_on_console > 0)
4179 return;
4180
4181 va_start(ap, format);
4182 status_vprintf(status, SHOW_STATUS_ELLIPSIZE|(type == STATUS_TYPE_EPHEMERAL ? SHOW_STATUS_EPHEMERAL : 0), format, ap);
4183 va_end(ap);
4184 }
4185
4186 Set* manager_get_units_requiring_mounts_for(Manager *m, const char *path) {
4187 assert(m);
4188 assert(path);
4189
4190 if (path_equal(path, "/"))
4191 path = "";
4192
4193 return hashmap_get(m->units_requiring_mounts_for, path);
4194 }
4195
4196 int manager_update_failed_units(Manager *m, Unit *u, bool failed) {
4197 unsigned size;
4198 int r;
4199
4200 assert(m);
4201 assert(u->manager == m);
4202
4203 size = set_size(m->failed_units);
4204
4205 if (failed) {
4206 r = set_ensure_put(&m->failed_units, NULL, u);
4207 if (r < 0)
4208 return log_oom();
4209 } else
4210 (void) set_remove(m->failed_units, u);
4211
4212 if (set_size(m->failed_units) != size)
4213 bus_manager_send_change_signal(m);
4214
4215 return 0;
4216 }
4217
4218 ManagerState manager_state(Manager *m) {
4219 Unit *u;
4220
4221 assert(m);
4222
4223 /* Is the special shutdown target active or queued? If so, we are in shutdown state */
4224 u = manager_get_unit(m, SPECIAL_SHUTDOWN_TARGET);
4225 if (u && unit_active_or_pending(u))
4226 return MANAGER_STOPPING;
4227
4228 /* Did we ever finish booting? If not then we are still starting up */
4229 if (!MANAGER_IS_FINISHED(m)) {
4230
4231 u = manager_get_unit(m, SPECIAL_BASIC_TARGET);
4232 if (!u || !UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u)))
4233 return MANAGER_INITIALIZING;
4234
4235 return MANAGER_STARTING;
4236 }
4237
4238 if (MANAGER_IS_SYSTEM(m)) {
4239 /* Are the rescue or emergency targets active or queued? If so we are in maintenance state */
4240 u = manager_get_unit(m, SPECIAL_RESCUE_TARGET);
4241 if (u && unit_active_or_pending(u))
4242 return MANAGER_MAINTENANCE;
4243
4244 u = manager_get_unit(m, SPECIAL_EMERGENCY_TARGET);
4245 if (u && unit_active_or_pending(u))
4246 return MANAGER_MAINTENANCE;
4247 }
4248
4249 /* Are there any failed units? If so, we are in degraded mode */
4250 if (set_size(m->failed_units) > 0)
4251 return MANAGER_DEGRADED;
4252
4253 return MANAGER_RUNNING;
4254 }
4255
4256 static void manager_unref_uid_internal(
4257 Hashmap *uid_refs,
4258 uid_t uid,
4259 bool destroy_now,
4260 int (*_clean_ipc)(uid_t uid)) {
4261
4262 uint32_t c, n;
4263
4264 assert(uid_is_valid(uid));
4265 assert(_clean_ipc);
4266
4267 /* A generic implementation, covering both manager_unref_uid() and manager_unref_gid(), under the
4268 * assumption that uid_t and gid_t are actually defined the same way, with the same validity rules.
4269 *
4270 * We store a hashmap where the key is the UID/GID and the value is a 32bit reference counter, whose
4271 * highest bit is used as flag for marking UIDs/GIDs whose IPC objects to remove when the last
4272 * reference to the UID/GID is dropped. The flag is set to on, once at least one reference from a
4273 * unit where RemoveIPC= is set is added on a UID/GID. It is reset when the UID's/GID's reference
4274 * counter drops to 0 again. */
4275
4276 assert_cc(sizeof(uid_t) == sizeof(gid_t));
4277 assert_cc(UID_INVALID == (uid_t) GID_INVALID);
4278
4279 if (uid == 0) /* We don't keep track of root, and will never destroy it */
4280 return;
4281
4282 c = PTR_TO_UINT32(hashmap_get(uid_refs, UID_TO_PTR(uid)));
4283
4284 n = c & ~DESTROY_IPC_FLAG;
4285 assert(n > 0);
4286 n--;
4287
4288 if (destroy_now && n == 0) {
4289 hashmap_remove(uid_refs, UID_TO_PTR(uid));
4290
4291 if (c & DESTROY_IPC_FLAG) {
4292 log_debug("%s " UID_FMT " is no longer referenced, cleaning up its IPC.",
4293 _clean_ipc == clean_ipc_by_uid ? "UID" : "GID",
4294 uid);
4295 (void) _clean_ipc(uid);
4296 }
4297 } else {
4298 c = n | (c & DESTROY_IPC_FLAG);
4299 assert_se(hashmap_update(uid_refs, UID_TO_PTR(uid), UINT32_TO_PTR(c)) >= 0);
4300 }
4301 }
4302
4303 void manager_unref_uid(Manager *m, uid_t uid, bool destroy_now) {
4304 manager_unref_uid_internal(m->uid_refs, uid, destroy_now, clean_ipc_by_uid);
4305 }
4306
4307 void manager_unref_gid(Manager *m, gid_t gid, bool destroy_now) {
4308 manager_unref_uid_internal(m->gid_refs, (uid_t) gid, destroy_now, clean_ipc_by_gid);
4309 }
4310
4311 static int manager_ref_uid_internal(
4312 Hashmap **uid_refs,
4313 uid_t uid,
4314 bool clean_ipc) {
4315
4316 uint32_t c, n;
4317 int r;
4318
4319 assert(uid_refs);
4320 assert(uid_is_valid(uid));
4321
4322 /* A generic implementation, covering both manager_ref_uid() and manager_ref_gid(), under the
4323 * assumption that uid_t and gid_t are actually defined the same way, with the same validity
4324 * rules. */
4325
4326 assert_cc(sizeof(uid_t) == sizeof(gid_t));
4327 assert_cc(UID_INVALID == (uid_t) GID_INVALID);
4328
4329 if (uid == 0) /* We don't keep track of root, and will never destroy it */
4330 return 0;
4331
4332 r = hashmap_ensure_allocated(uid_refs, &trivial_hash_ops);
4333 if (r < 0)
4334 return r;
4335
4336 c = PTR_TO_UINT32(hashmap_get(*uid_refs, UID_TO_PTR(uid)));
4337
4338 n = c & ~DESTROY_IPC_FLAG;
4339 n++;
4340
4341 if (n & DESTROY_IPC_FLAG) /* check for overflow */
4342 return -EOVERFLOW;
4343
4344 c = n | (c & DESTROY_IPC_FLAG) | (clean_ipc ? DESTROY_IPC_FLAG : 0);
4345
4346 return hashmap_replace(*uid_refs, UID_TO_PTR(uid), UINT32_TO_PTR(c));
4347 }
4348
4349 int manager_ref_uid(Manager *m, uid_t uid, bool clean_ipc) {
4350 return manager_ref_uid_internal(&m->uid_refs, uid, clean_ipc);
4351 }
4352
4353 int manager_ref_gid(Manager *m, gid_t gid, bool clean_ipc) {
4354 return manager_ref_uid_internal(&m->gid_refs, (uid_t) gid, clean_ipc);
4355 }
4356
4357 static void manager_vacuum_uid_refs_internal(
4358 Hashmap *uid_refs,
4359 int (*_clean_ipc)(uid_t uid)) {
4360
4361 void *p, *k;
4362
4363 assert(_clean_ipc);
4364
4365 HASHMAP_FOREACH_KEY(p, k, uid_refs) {
4366 uint32_t c, n;
4367 uid_t uid;
4368
4369 uid = PTR_TO_UID(k);
4370 c = PTR_TO_UINT32(p);
4371
4372 n = c & ~DESTROY_IPC_FLAG;
4373 if (n > 0)
4374 continue;
4375
4376 if (c & DESTROY_IPC_FLAG) {
4377 log_debug("Found unreferenced %s " UID_FMT " after reload/reexec. Cleaning up.",
4378 _clean_ipc == clean_ipc_by_uid ? "UID" : "GID",
4379 uid);
4380 (void) _clean_ipc(uid);
4381 }
4382
4383 assert_se(hashmap_remove(uid_refs, k) == p);
4384 }
4385 }
4386
4387 static void manager_vacuum_uid_refs(Manager *m) {
4388 manager_vacuum_uid_refs_internal(m->uid_refs, clean_ipc_by_uid);
4389 }
4390
4391 static void manager_vacuum_gid_refs(Manager *m) {
4392 manager_vacuum_uid_refs_internal(m->gid_refs, clean_ipc_by_gid);
4393 }
4394
4395 static void manager_vacuum(Manager *m) {
4396 assert(m);
4397
4398 /* Release any dynamic users no longer referenced */
4399 dynamic_user_vacuum(m, true);
4400
4401 /* Release any references to UIDs/GIDs no longer referenced, and destroy any IPC owned by them */
4402 manager_vacuum_uid_refs(m);
4403 manager_vacuum_gid_refs(m);
4404
4405 /* Release any runtimes no longer referenced */
4406 exec_runtime_vacuum(m);
4407 }
4408
4409 int manager_dispatch_user_lookup_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
4410 struct buffer {
4411 uid_t uid;
4412 gid_t gid;
4413 char unit_name[UNIT_NAME_MAX+1];
4414 } _packed_ buffer;
4415
4416 Manager *m = userdata;
4417 ssize_t l;
4418 size_t n;
4419 Unit *u;
4420
4421 assert_se(source);
4422 assert_se(m);
4423
4424 /* Invoked whenever a child process succeeded resolving its user/group to use and sent us the
4425 * resulting UID/GID in a datagram. We parse the datagram here and pass it off to the unit, so that
4426 * it can add a reference to the UID/GID so that it can destroy the UID/GID's IPC objects when the
4427 * reference counter drops to 0. */
4428
4429 l = recv(fd, &buffer, sizeof(buffer), MSG_DONTWAIT);
4430 if (l < 0) {
4431 if (ERRNO_IS_TRANSIENT(errno))
4432 return 0;
4433
4434 return log_error_errno(errno, "Failed to read from user lookup fd: %m");
4435 }
4436
4437 if ((size_t) l <= offsetof(struct buffer, unit_name)) {
4438 log_warning("Received too short user lookup message, ignoring.");
4439 return 0;
4440 }
4441
4442 if ((size_t) l > offsetof(struct buffer, unit_name) + UNIT_NAME_MAX) {
4443 log_warning("Received too long user lookup message, ignoring.");
4444 return 0;
4445 }
4446
4447 if (!uid_is_valid(buffer.uid) && !gid_is_valid(buffer.gid)) {
4448 log_warning("Got user lookup message with invalid UID/GID pair, ignoring.");
4449 return 0;
4450 }
4451
4452 n = (size_t) l - offsetof(struct buffer, unit_name);
4453 if (memchr(buffer.unit_name, 0, n)) {
4454 log_warning("Received lookup message with embedded NUL character, ignoring.");
4455 return 0;
4456 }
4457
4458 buffer.unit_name[n] = 0;
4459 u = manager_get_unit(m, buffer.unit_name);
4460 if (!u) {
4461 log_debug("Got user lookup message but unit doesn't exist, ignoring.");
4462 return 0;
4463 }
4464
4465 log_unit_debug(u, "User lookup succeeded: uid=" UID_FMT " gid=" GID_FMT, buffer.uid, buffer.gid);
4466
4467 unit_notify_user_lookup(u, buffer.uid, buffer.gid);
4468 return 0;
4469 }
4470
4471 static int short_uid_range(const char *path) {
4472 _cleanup_free_ UidRange *p = NULL;
4473 size_t n = 0;
4474 int r;
4475
4476 assert(path);
4477
4478 /* Taint systemd if we the UID range assigned to this environment doesn't at least cover 0…65534,
4479 * i.e. from root to nobody. */
4480
4481 r = uid_range_load_userns(&p, &n, path);
4482 if (ERRNO_IS_NOT_SUPPORTED(r))
4483 return false;
4484 if (r < 0)
4485 return log_debug_errno(r, "Failed to load %s: %m", path);
4486
4487 return !uid_range_covers(p, n, 0, 65535);
4488 }
4489
4490 char* manager_taint_string(const Manager *m) {
4491 /* Returns a "taint string", e.g. "local-hwclock:var-run-bad". Only things that are detected at
4492 * runtime should be tagged here. For stuff that is known during compilation, emit a warning in the
4493 * configuration phase. */
4494
4495 assert(m);
4496
4497 const char* stage[13] = {};
4498 size_t n = 0;
4499
4500 if (m->taint_usr)
4501 stage[n++] = "split-usr";
4502
4503 _cleanup_free_ char *usrbin = NULL;
4504 if (readlink_malloc("/bin", &usrbin) < 0 || !PATH_IN_SET(usrbin, "usr/bin", "/usr/bin"))
4505 stage[n++] = "unmerged-usr";
4506
4507 if (access("/proc/cgroups", F_OK) < 0)
4508 stage[n++] = "cgroups-missing";
4509
4510 if (cg_all_unified() == 0)
4511 stage[n++] = "cgroupsv1";
4512
4513 if (clock_is_localtime(NULL) > 0)
4514 stage[n++] = "local-hwclock";
4515
4516 if (os_release_support_ended(NULL, true) > 0)
4517 stage[n++] = "support-ended";
4518
4519 _cleanup_free_ char *destination = NULL;
4520 if (readlink_malloc("/var/run", &destination) < 0 ||
4521 !PATH_IN_SET(destination, "../run", "/run"))
4522 stage[n++] = "var-run-bad";
4523
4524 _cleanup_free_ char *overflowuid = NULL, *overflowgid = NULL;
4525 if (read_one_line_file("/proc/sys/kernel/overflowuid", &overflowuid) >= 0 &&
4526 !streq(overflowuid, "65534"))
4527 stage[n++] = "overflowuid-not-65534";
4528 if (read_one_line_file("/proc/sys/kernel/overflowgid", &overflowgid) >= 0 &&
4529 !streq(overflowgid, "65534"))
4530 stage[n++] = "overflowgid-not-65534";
4531
4532 struct utsname uts;
4533 assert_se(uname(&uts) >= 0);
4534 if (strverscmp_improved(uts.release, KERNEL_BASELINE_VERSION) < 0)
4535 stage[n++] = "old-kernel";
4536
4537 if (short_uid_range("/proc/self/uid_map") > 0)
4538 stage[n++] = "short-uid-range";
4539 if (short_uid_range("/proc/self/gid_map") > 0)
4540 stage[n++] = "short-gid-range";
4541
4542 assert(n < ELEMENTSOF(stage) - 1); /* One extra for NULL terminator */
4543
4544 return strv_join((char**) stage, ":");
4545 }
4546
4547 void manager_ref_console(Manager *m) {
4548 assert(m);
4549
4550 m->n_on_console++;
4551 }
4552
4553 void manager_unref_console(Manager *m) {
4554
4555 assert(m->n_on_console > 0);
4556 m->n_on_console--;
4557
4558 if (m->n_on_console == 0)
4559 m->no_console_output = false; /* unset no_console_output flag, since the console is definitely free now */
4560 }
4561
4562 void manager_override_log_level(Manager *m, int level) {
4563 _cleanup_free_ char *s = NULL;
4564 assert(m);
4565
4566 if (!m->log_level_overridden) {
4567 m->original_log_level = log_get_max_level();
4568 m->log_level_overridden = true;
4569 }
4570
4571 (void) log_level_to_string_alloc(level, &s);
4572 log_info("Setting log level to %s.", strna(s));
4573
4574 log_set_max_level(level);
4575 }
4576
4577 void manager_restore_original_log_level(Manager *m) {
4578 _cleanup_free_ char *s = NULL;
4579 assert(m);
4580
4581 if (!m->log_level_overridden)
4582 return;
4583
4584 (void) log_level_to_string_alloc(m->original_log_level, &s);
4585 log_info("Restoring log level to original (%s).", strna(s));
4586
4587 log_set_max_level(m->original_log_level);
4588 m->log_level_overridden = false;
4589 }
4590
4591 void manager_override_log_target(Manager *m, LogTarget target) {
4592 assert(m);
4593
4594 if (!m->log_target_overridden) {
4595 m->original_log_target = log_get_target();
4596 m->log_target_overridden = true;
4597 }
4598
4599 log_info("Setting log target to %s.", log_target_to_string(target));
4600 log_set_target(target);
4601 }
4602
4603 void manager_restore_original_log_target(Manager *m) {
4604 assert(m);
4605
4606 if (!m->log_target_overridden)
4607 return;
4608
4609 log_info("Restoring log target to original %s.", log_target_to_string(m->original_log_target));
4610
4611 log_set_target(m->original_log_target);
4612 m->log_target_overridden = false;
4613 }
4614
4615 ManagerTimestamp manager_timestamp_initrd_mangle(ManagerTimestamp s) {
4616 if (in_initrd() &&
4617 s >= MANAGER_TIMESTAMP_SECURITY_START &&
4618 s <= MANAGER_TIMESTAMP_UNITS_LOAD_FINISH)
4619 return s - MANAGER_TIMESTAMP_SECURITY_START + MANAGER_TIMESTAMP_INITRD_SECURITY_START;
4620 return s;
4621 }
4622
4623 static const char *const manager_state_table[_MANAGER_STATE_MAX] = {
4624 [MANAGER_INITIALIZING] = "initializing",
4625 [MANAGER_STARTING] = "starting",
4626 [MANAGER_RUNNING] = "running",
4627 [MANAGER_DEGRADED] = "degraded",
4628 [MANAGER_MAINTENANCE] = "maintenance",
4629 [MANAGER_STOPPING] = "stopping",
4630 };
4631
4632 DEFINE_STRING_TABLE_LOOKUP(manager_state, ManagerState);
4633
4634 static const char *const manager_timestamp_table[_MANAGER_TIMESTAMP_MAX] = {
4635 [MANAGER_TIMESTAMP_FIRMWARE] = "firmware",
4636 [MANAGER_TIMESTAMP_LOADER] = "loader",
4637 [MANAGER_TIMESTAMP_KERNEL] = "kernel",
4638 [MANAGER_TIMESTAMP_INITRD] = "initrd",
4639 [MANAGER_TIMESTAMP_USERSPACE] = "userspace",
4640 [MANAGER_TIMESTAMP_FINISH] = "finish",
4641 [MANAGER_TIMESTAMP_SECURITY_START] = "security-start",
4642 [MANAGER_TIMESTAMP_SECURITY_FINISH] = "security-finish",
4643 [MANAGER_TIMESTAMP_GENERATORS_START] = "generators-start",
4644 [MANAGER_TIMESTAMP_GENERATORS_FINISH] = "generators-finish",
4645 [MANAGER_TIMESTAMP_UNITS_LOAD_START] = "units-load-start",
4646 [MANAGER_TIMESTAMP_UNITS_LOAD_FINISH] = "units-load-finish",
4647 [MANAGER_TIMESTAMP_UNITS_LOAD] = "units-load",
4648 [MANAGER_TIMESTAMP_INITRD_SECURITY_START] = "initrd-security-start",
4649 [MANAGER_TIMESTAMP_INITRD_SECURITY_FINISH] = "initrd-security-finish",
4650 [MANAGER_TIMESTAMP_INITRD_GENERATORS_START] = "initrd-generators-start",
4651 [MANAGER_TIMESTAMP_INITRD_GENERATORS_FINISH] = "initrd-generators-finish",
4652 [MANAGER_TIMESTAMP_INITRD_UNITS_LOAD_START] = "initrd-units-load-start",
4653 [MANAGER_TIMESTAMP_INITRD_UNITS_LOAD_FINISH] = "initrd-units-load-finish",
4654 };
4655
4656 DEFINE_STRING_TABLE_LOOKUP(manager_timestamp, ManagerTimestamp);
4657
4658 static const char* const oom_policy_table[_OOM_POLICY_MAX] = {
4659 [OOM_CONTINUE] = "continue",
4660 [OOM_STOP] = "stop",
4661 [OOM_KILL] = "kill",
4662 };
4663
4664 DEFINE_STRING_TABLE_LOOKUP(oom_policy, OOMPolicy);
Unnamed repository; edit this file 'description' to name the repository.