]> git.ipfire.org Git - thirdparty/systemd.git/blob - src/core/manager.c
projects / thirdparty / systemd.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
tree-wide: introduce PIPE_EBADF macro
[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/mount.h>
10 #include <sys/reboot.h>
11 #include <sys/timerfd.h>
12 #include <sys/utsname.h>
13 #include <sys/wait.h>
14 #include <unistd.h>
15
16 #if HAVE_AUDIT
17 #include <libaudit.h>
18 #endif
19
20 #include "sd-daemon.h"
21 #include "sd-messages.h"
22 #include "sd-path.h"
23
24 #include "all-units.h"
25 #include "alloc-util.h"
26 #include "audit-fd.h"
27 #include "boot-timestamps.h"
28 #include "bus-common-errors.h"
29 #include "bus-error.h"
30 #include "bus-kernel.h"
31 #include "bus-util.h"
32 #include "clean-ipc.h"
33 #include "clock-util.h"
34 #include "constants.h"
35 #include "core-varlink.h"
36 #include "creds-util.h"
37 #include "dbus-job.h"
38 #include "dbus-manager.h"
39 #include "dbus-unit.h"
40 #include "dbus.h"
41 #include "dirent-util.h"
42 #include "env-util.h"
43 #include "escape.h"
44 #include "event-util.h"
45 #include "exec-util.h"
46 #include "execute.h"
47 #include "exit-status.h"
48 #include "fd-util.h"
49 #include "fileio.h"
50 #include "generator-setup.h"
51 #include "hashmap.h"
52 #include "initrd-util.h"
53 #include "inotify-util.h"
54 #include "install.h"
55 #include "io-util.h"
56 #include "label.h"
57 #include "load-fragment.h"
58 #include "locale-setup.h"
59 #include "log.h"
60 #include "macro.h"
61 #include "manager.h"
62 #include "manager-dump.h"
63 #include "manager-serialize.h"
64 #include "memory-util.h"
65 #include "mkdir-label.h"
66 #include "mount-util.h"
67 #include "os-util.h"
68 #include "parse-util.h"
69 #include "path-lookup.h"
70 #include "path-util.h"
71 #include "process-util.h"
72 #include "ratelimit.h"
73 #include "rlimit-util.h"
74 #include "rm-rf.h"
75 #include "selinux-util.h"
76 #include "signal-util.h"
77 #include "socket-util.h"
78 #include "special.h"
79 #include "stat-util.h"
80 #include "string-table.h"
81 #include "string-util.h"
82 #include "strv.h"
83 #include "strxcpyx.h"
84 #include "sysctl-util.h"
85 #include "syslog-util.h"
86 #include "terminal-util.h"
87 #include "time-util.h"
88 #include "transaction.h"
89 #include "uid-range.h"
90 #include "umask-util.h"
91 #include "unit-name.h"
92 #include "user-util.h"
93 #include "virt.h"
94 #include "watchdog.h"
95
96 #define NOTIFY_RCVBUF_SIZE (8*1024*1024)
97 #define CGROUPS_AGENT_RCVBUF_SIZE (8*1024*1024)
98
99 /* Initial delay and the interval for printing status messages about running jobs */
100 #define JOBS_IN_PROGRESS_WAIT_USEC (2*USEC_PER_SEC)
101 #define JOBS_IN_PROGRESS_QUIET_WAIT_USEC (25*USEC_PER_SEC)
102 #define JOBS_IN_PROGRESS_PERIOD_USEC (USEC_PER_SEC / 3)
103 #define JOBS_IN_PROGRESS_PERIOD_DIVISOR 3
104
105 /* If there are more than 1K bus messages queue across our API and direct buses, then let's not add more on top until
106 * the queue gets more empty. */
107 #define MANAGER_BUS_BUSY_THRESHOLD 1024LU
108
109 /* How many units and jobs to process of the bus queue before returning to the event loop. */
110 #define MANAGER_BUS_MESSAGE_BUDGET 100U
111
112 static int manager_dispatch_notify_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata);
113 static int manager_dispatch_cgroups_agent_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata);
114 static int manager_dispatch_signal_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata);
115 static int manager_dispatch_time_change_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata);
116 static int manager_dispatch_idle_pipe_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata);
117 static int manager_dispatch_user_lookup_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata);
118 static int manager_dispatch_jobs_in_progress(sd_event_source *source, usec_t usec, void *userdata);
119 static int manager_dispatch_run_queue(sd_event_source *source, void *userdata);
120 static int manager_dispatch_sigchld(sd_event_source *source, void *userdata);
121 static int manager_dispatch_timezone_change(sd_event_source *source, const struct inotify_event *event, void *userdata);
122 static int manager_run_environment_generators(Manager *m);
123 static int manager_run_generators(Manager *m);
124 static void manager_vacuum(Manager *m);
125
126 static usec_t manager_watch_jobs_next_time(Manager *m) {
127 usec_t timeout;
128
129 if (MANAGER_IS_USER(m))
130 /* Let the user manager without a timeout show status quickly, so the system manager can make
131 * use of it, if it wants to. */
132 timeout = JOBS_IN_PROGRESS_WAIT_USEC * 2 / 3;
133 else if (show_status_on(m->show_status))
134 /* When status is on, just use the usual timeout. */
135 timeout = JOBS_IN_PROGRESS_WAIT_USEC;
136 else
137 timeout = JOBS_IN_PROGRESS_QUIET_WAIT_USEC;
138
139 return usec_add(now(CLOCK_MONOTONIC), timeout);
140 }
141
142 static void manager_watch_jobs_in_progress(Manager *m) {
143 usec_t next;
144 int r;
145
146 assert(m);
147
148 /* We do not want to show the cylon animation if the user
149 * needs to confirm service executions otherwise confirmation
150 * messages will be screwed by the cylon animation. */
151 if (!manager_is_confirm_spawn_disabled(m))
152 return;
153
154 if (m->jobs_in_progress_event_source)
155 return;
156
157 next = manager_watch_jobs_next_time(m);
158 r = sd_event_add_time(
159 m->event,
160 &m->jobs_in_progress_event_source,
161 CLOCK_MONOTONIC,
162 next, 0,
163 manager_dispatch_jobs_in_progress, m);
164 if (r < 0)
165 return;
166
167 (void) sd_event_source_set_description(m->jobs_in_progress_event_source, "manager-jobs-in-progress");
168 }
169
170 #define CYLON_BUFFER_EXTRA (2*STRLEN(ANSI_RED) + STRLEN(ANSI_HIGHLIGHT_RED) + 2*STRLEN(ANSI_NORMAL))
171
172 static void draw_cylon(char buffer[], size_t buflen, unsigned width, unsigned pos) {
173 char *p = buffer;
174
175 assert(buflen >= CYLON_BUFFER_EXTRA + width + 1);
176 assert(pos <= width+1); /* 0 or width+1 mean that the center light is behind the corner */
177
178 if (pos > 1) {
179 if (pos > 2)
180 p = mempset(p, ' ', pos-2);
181 if (log_get_show_color())
182 p = stpcpy(p, ANSI_RED);
183 *p++ = '*';
184 }
185
186 if (pos > 0 && pos <= width) {
187 if (log_get_show_color())
188 p = stpcpy(p, ANSI_HIGHLIGHT_RED);
189 *p++ = '*';
190 }
191
192 if (log_get_show_color())
193 p = stpcpy(p, ANSI_NORMAL);
194
195 if (pos < width) {
196 if (log_get_show_color())
197 p = stpcpy(p, ANSI_RED);
198 *p++ = '*';
199 if (pos < width-1)
200 p = mempset(p, ' ', width-1-pos);
201 if (log_get_show_color())
202 strcpy(p, ANSI_NORMAL);
203 }
204 }
205
206 static void manager_flip_auto_status(Manager *m, bool enable, const char *reason) {
207 assert(m);
208
209 if (enable) {
210 if (m->show_status == SHOW_STATUS_AUTO)
211 manager_set_show_status(m, SHOW_STATUS_TEMPORARY, reason);
212 } else {
213 if (m->show_status == SHOW_STATUS_TEMPORARY)
214 manager_set_show_status(m, SHOW_STATUS_AUTO, reason);
215 }
216 }
217
218 static void manager_print_jobs_in_progress(Manager *m) {
219 Job *j;
220 unsigned counter = 0, print_nr;
221 char cylon[6 + CYLON_BUFFER_EXTRA + 1];
222 unsigned cylon_pos;
223 uint64_t timeout = 0;
224
225 assert(m);
226 assert(m->n_running_jobs > 0);
227
228 manager_flip_auto_status(m, true, "delay");
229
230 print_nr = (m->jobs_in_progress_iteration / JOBS_IN_PROGRESS_PERIOD_DIVISOR) % m->n_running_jobs;
231
232 HASHMAP_FOREACH(j, m->jobs)
233 if (j->state == JOB_RUNNING && counter++ == print_nr)
234 break;
235
236 /* m->n_running_jobs must be consistent with the contents of m->jobs,
237 * so the above loop must have succeeded in finding j. */
238 assert(counter == print_nr + 1);
239 assert(j);
240
241 cylon_pos = m->jobs_in_progress_iteration % 14;
242 if (cylon_pos >= 8)
243 cylon_pos = 14 - cylon_pos;
244 draw_cylon(cylon, sizeof(cylon), 6, cylon_pos);
245
246 m->jobs_in_progress_iteration++;
247
248 char job_of_n[STRLEN("( of ) ") + DECIMAL_STR_MAX(unsigned)*2] = "";
249 if (m->n_running_jobs > 1)
250 xsprintf(job_of_n, "(%u of %u) ", counter, m->n_running_jobs);
251
252 (void) job_get_timeout(j, &timeout);
253
254 /* We want to use enough information for the user to identify previous lines talking about the same
255 * unit, but keep the message as short as possible. So if 'Starting foo.service' or 'Starting
256 * foo.service - Description' were used, 'foo.service' is enough here. On the other hand, if we used
257 * 'Starting Description' before, then we shall also use 'Description' here. So we pass NULL as the
258 * second argument to unit_status_string(). */
259 const char *ident = unit_status_string(j->unit, NULL);
260
261 const char *time = FORMAT_TIMESPAN(now(CLOCK_MONOTONIC) - j->begin_usec, 1*USEC_PER_SEC);
262 const char *limit = timeout > 0 ? FORMAT_TIMESPAN(timeout - j->begin_usec, 1*USEC_PER_SEC) : "no limit";
263
264 if (m->status_unit_format == STATUS_UNIT_FORMAT_DESCRIPTION)
265 /* When using 'Description', we effectively don't have enough space to show the nested status
266 * without ellipsization, so let's not even try. */
267 manager_status_printf(m, STATUS_TYPE_EPHEMERAL, cylon,
268 "%sA %s job is running for %s (%s / %s)",
269 job_of_n,
270 job_type_to_string(j->type),
271 ident,
272 time, limit);
273 else {
274 const char *status_text = unit_status_text(j->unit);
275
276 manager_status_printf(m, STATUS_TYPE_EPHEMERAL, cylon,
277 "%sJob %s/%s running (%s / %s)%s%s",
278 job_of_n,
279 ident,
280 job_type_to_string(j->type),
281 time, limit,
282 status_text ? ": " : "",
283 strempty(status_text));
284 }
285
286 sd_notifyf(false,
287 "STATUS=%sUser job %s/%s running (%s / %s)...",
288 job_of_n,
289 ident,
290 job_type_to_string(j->type),
291 time, limit);
292 m->status_ready = false;
293 }
294
295 static int have_ask_password(void) {
296 _cleanup_closedir_ DIR *dir = NULL;
297
298 dir = opendir("/run/systemd/ask-password");
299 if (!dir) {
300 if (errno == ENOENT)
301 return false;
302 else
303 return -errno;
304 }
305
306 FOREACH_DIRENT_ALL(de, dir, return -errno)
307 if (startswith(de->d_name, "ask."))
308 return true;
309 return false;
310 }
311
312 static int manager_dispatch_ask_password_fd(sd_event_source *source,
313 int fd, uint32_t revents, void *userdata) {
314 Manager *m = ASSERT_PTR(userdata);
315
316 (void) flush_fd(fd);
317
318 m->have_ask_password = have_ask_password();
319 if (m->have_ask_password < 0)
320 /* Log error but continue. Negative have_ask_password
321 * is treated as unknown status. */
322 log_error_errno(m->have_ask_password, "Failed to list /run/systemd/ask-password: %m");
323
324 return 0;
325 }
326
327 static void manager_close_ask_password(Manager *m) {
328 assert(m);
329
330 m->ask_password_event_source = sd_event_source_disable_unref(m->ask_password_event_source);
331 m->ask_password_inotify_fd = safe_close(m->ask_password_inotify_fd);
332 m->have_ask_password = -EINVAL;
333 }
334
335 static int manager_check_ask_password(Manager *m) {
336 int r;
337
338 assert(m);
339
340 if (!m->ask_password_event_source) {
341 assert(m->ask_password_inotify_fd < 0);
342
343 (void) mkdir_p_label("/run/systemd/ask-password", 0755);
344
345 m->ask_password_inotify_fd = inotify_init1(IN_NONBLOCK|IN_CLOEXEC);
346 if (m->ask_password_inotify_fd < 0)
347 return log_error_errno(errno, "Failed to create inotify object: %m");
348
349 r = inotify_add_watch_and_warn(m->ask_password_inotify_fd,
350 "/run/systemd/ask-password",
351 IN_CREATE|IN_DELETE|IN_MOVE);
352 if (r < 0) {
353 manager_close_ask_password(m);
354 return r;
355 }
356
357 r = sd_event_add_io(m->event, &m->ask_password_event_source,
358 m->ask_password_inotify_fd, EPOLLIN,
359 manager_dispatch_ask_password_fd, m);
360 if (r < 0) {
361 log_error_errno(r, "Failed to add event source for /run/systemd/ask-password: %m");
362 manager_close_ask_password(m);
363 return r;
364 }
365
366 (void) sd_event_source_set_description(m->ask_password_event_source, "manager-ask-password");
367
368 /* Queries might have been added meanwhile... */
369 manager_dispatch_ask_password_fd(m->ask_password_event_source,
370 m->ask_password_inotify_fd, EPOLLIN, m);
371 }
372
373 return m->have_ask_password;
374 }
375
376 static int manager_watch_idle_pipe(Manager *m) {
377 int r;
378
379 assert(m);
380
381 if (m->idle_pipe_event_source)
382 return 0;
383
384 if (m->idle_pipe[2] < 0)
385 return 0;
386
387 r = sd_event_add_io(m->event, &m->idle_pipe_event_source, m->idle_pipe[2], EPOLLIN, manager_dispatch_idle_pipe_fd, m);
388 if (r < 0)
389 return log_error_errno(r, "Failed to watch idle pipe: %m");
390
391 (void) sd_event_source_set_description(m->idle_pipe_event_source, "manager-idle-pipe");
392
393 return 0;
394 }
395
396 static void manager_close_idle_pipe(Manager *m) {
397 assert(m);
398
399 m->idle_pipe_event_source = sd_event_source_disable_unref(m->idle_pipe_event_source);
400
401 safe_close_pair(m->idle_pipe);
402 safe_close_pair(m->idle_pipe + 2);
403 }
404
405 static int manager_setup_time_change(Manager *m) {
406 int r;
407
408 assert(m);
409
410 if (MANAGER_IS_TEST_RUN(m))
411 return 0;
412
413 m->time_change_event_source = sd_event_source_disable_unref(m->time_change_event_source);
414
415 r = event_add_time_change(m->event, &m->time_change_event_source, manager_dispatch_time_change_fd, m);
416 if (r < 0)
417 return log_error_errno(r, "Failed to create time change event source: %m");
418
419 /* Schedule this slightly earlier than the .timer event sources */
420 r = sd_event_source_set_priority(m->time_change_event_source, SD_EVENT_PRIORITY_NORMAL-1);
421 if (r < 0)
422 return log_error_errno(r, "Failed to set priority of time change event sources: %m");
423
424 log_debug("Set up TFD_TIMER_CANCEL_ON_SET timerfd.");
425
426 return 0;
427 }
428
429 static int manager_read_timezone_stat(Manager *m) {
430 struct stat st;
431 bool changed;
432
433 assert(m);
434
435 /* Read the current stat() data of /etc/localtime so that we detect changes */
436 if (lstat("/etc/localtime", &st) < 0) {
437 log_debug_errno(errno, "Failed to stat /etc/localtime, ignoring: %m");
438 changed = m->etc_localtime_accessible;
439 m->etc_localtime_accessible = false;
440 } else {
441 usec_t k;
442
443 k = timespec_load(&st.st_mtim);
444 changed = !m->etc_localtime_accessible || k != m->etc_localtime_mtime;
445
446 m->etc_localtime_mtime = k;
447 m->etc_localtime_accessible = true;
448 }
449
450 return changed;
451 }
452
453 static int manager_setup_timezone_change(Manager *m) {
454 _cleanup_(sd_event_source_unrefp) sd_event_source *new_event = NULL;
455 int r;
456
457 assert(m);
458
459 if (MANAGER_IS_TEST_RUN(m))
460 return 0;
461
462 /* We watch /etc/localtime for three events: change of the link count (which might mean removal from /etc even
463 * though another link might be kept), renames, and file close operations after writing. Note we don't bother
464 * with IN_DELETE_SELF, as that would just report when the inode is removed entirely, i.e. after the link count
465 * went to zero and all fds to it are closed.
466 *
467 * Note that we never follow symlinks here. This is a simplification, but should cover almost all cases
468 * correctly.
469 *
470 * Note that we create the new event source first here, before releasing the old one. This should optimize
471 * behaviour as this way sd-event can reuse the old watch in case the inode didn't change. */
472
473 r = sd_event_add_inotify(m->event, &new_event, "/etc/localtime",
474 IN_ATTRIB|IN_MOVE_SELF|IN_CLOSE_WRITE|IN_DONT_FOLLOW, manager_dispatch_timezone_change, m);
475 if (r == -ENOENT) {
476 /* If the file doesn't exist yet, subscribe to /etc instead, and wait until it is created either by
477 * O_CREATE or by rename() */
478
479 log_debug_errno(r, "/etc/localtime doesn't exist yet, watching /etc instead.");
480 r = sd_event_add_inotify(m->event, &new_event, "/etc",
481 IN_CREATE|IN_MOVED_TO|IN_ONLYDIR, manager_dispatch_timezone_change, m);
482 }
483 if (r < 0)
484 return log_error_errno(r, "Failed to create timezone change event source: %m");
485
486 /* Schedule this slightly earlier than the .timer event sources */
487 r = sd_event_source_set_priority(new_event, SD_EVENT_PRIORITY_NORMAL-1);
488 if (r < 0)
489 return log_error_errno(r, "Failed to set priority of timezone change event sources: %m");
490
491 sd_event_source_unref(m->timezone_change_event_source);
492 m->timezone_change_event_source = TAKE_PTR(new_event);
493
494 return 0;
495 }
496
497 static int enable_special_signals(Manager *m) {
498 _cleanup_close_ int fd = -EBADF;
499
500 assert(m);
501
502 if (MANAGER_IS_TEST_RUN(m))
503 return 0;
504
505 /* Enable that we get SIGINT on control-alt-del. In containers
506 * this will fail with EPERM (older) or EINVAL (newer), so
507 * ignore that. */
508 if (reboot(RB_DISABLE_CAD) < 0 && !IN_SET(errno, EPERM, EINVAL))
509 log_warning_errno(errno, "Failed to enable ctrl-alt-del handling: %m");
510
511 fd = open_terminal("/dev/tty0", O_RDWR|O_NOCTTY|O_CLOEXEC);
512 if (fd < 0) {
513 /* Support systems without virtual console */
514 if (fd != -ENOENT)
515 log_warning_errno(errno, "Failed to open /dev/tty0: %m");
516 } else {
517 /* Enable that we get SIGWINCH on kbrequest */
518 if (ioctl(fd, KDSIGACCEPT, SIGWINCH) < 0)
519 log_warning_errno(errno, "Failed to enable kbrequest handling: %m");
520 }
521
522 return 0;
523 }
524
525 #define RTSIG_IF_AVAILABLE(signum) (signum <= SIGRTMAX ? signum : -1)
526
527 static int manager_setup_signals(Manager *m) {
528 struct sigaction sa = {
529 .sa_handler = SIG_DFL,
530 .sa_flags = SA_NOCLDSTOP|SA_RESTART,
531 };
532 sigset_t mask;
533 int r;
534
535 assert(m);
536
537 assert_se(sigaction(SIGCHLD, &sa, NULL) == 0);
538
539 /* We make liberal use of realtime signals here. On
540 * Linux/glibc we have 30 of them (with the exception of Linux
541 * on hppa, see below), between SIGRTMIN+0 ... SIGRTMIN+30
542 * (aka SIGRTMAX). */
543
544 assert_se(sigemptyset(&mask) == 0);
545 sigset_add_many(&mask,
546 SIGCHLD, /* Child died */
547 SIGTERM, /* Reexecute daemon */
548 SIGHUP, /* Reload configuration */
549 SIGUSR1, /* systemd: reconnect to D-Bus */
550 SIGUSR2, /* systemd: dump status */
551 SIGINT, /* Kernel sends us this on control-alt-del */
552 SIGWINCH, /* Kernel sends us this on kbrequest (alt-arrowup) */
553 SIGPWR, /* Some kernel drivers and upsd send us this on power failure */
554
555 SIGRTMIN+0, /* systemd: start default.target */
556 SIGRTMIN+1, /* systemd: isolate rescue.target */
557 SIGRTMIN+2, /* systemd: isolate emergency.target */
558 SIGRTMIN+3, /* systemd: start halt.target */
559 SIGRTMIN+4, /* systemd: start poweroff.target */
560 SIGRTMIN+5, /* systemd: start reboot.target */
561 SIGRTMIN+6, /* systemd: start kexec.target */
562
563 /* ... space for more special targets ... */
564
565 SIGRTMIN+13, /* systemd: Immediate halt */
566 SIGRTMIN+14, /* systemd: Immediate poweroff */
567 SIGRTMIN+15, /* systemd: Immediate reboot */
568 SIGRTMIN+16, /* systemd: Immediate kexec */
569
570 /* ... space for more immediate system state changes ... */
571
572 SIGRTMIN+20, /* systemd: enable status messages */
573 SIGRTMIN+21, /* systemd: disable status messages */
574 SIGRTMIN+22, /* systemd: set log level to LOG_DEBUG */
575 SIGRTMIN+23, /* systemd: set log level to LOG_INFO */
576 SIGRTMIN+24, /* systemd: Immediate exit (--user only) */
577 SIGRTMIN+25, /* systemd: reexecute manager */
578
579 /* Apparently Linux on hppa had fewer RT signals until v3.18,
580 * SIGRTMAX was SIGRTMIN+25, and then SIGRTMIN was lowered,
581 * see commit v3.17-7614-g1f25df2eff.
582 *
583 * We cannot unconditionally make use of those signals here,
584 * so let's use a runtime check. Since these commands are
585 * accessible by different means and only really a safety
586 * net, the missing functionality on hppa shouldn't matter.
587 */
588
589 RTSIG_IF_AVAILABLE(SIGRTMIN+26), /* systemd: set log target to journal-or-kmsg */
590 RTSIG_IF_AVAILABLE(SIGRTMIN+27), /* systemd: set log target to console */
591 RTSIG_IF_AVAILABLE(SIGRTMIN+28), /* systemd: set log target to kmsg */
592 RTSIG_IF_AVAILABLE(SIGRTMIN+29), /* systemd: set log target to syslog-or-kmsg (obsolete) */
593
594 /* ... one free signal here SIGRTMIN+30 ... */
595 -1);
596 assert_se(sigprocmask(SIG_SETMASK, &mask, NULL) == 0);
597
598 m->signal_fd = signalfd(-1, &mask, SFD_NONBLOCK|SFD_CLOEXEC);
599 if (m->signal_fd < 0)
600 return -errno;
601
602 r = sd_event_add_io(m->event, &m->signal_event_source, m->signal_fd, EPOLLIN, manager_dispatch_signal_fd, m);
603 if (r < 0)
604 return r;
605
606 (void) sd_event_source_set_description(m->signal_event_source, "manager-signal");
607
608 /* Process signals a bit earlier than the rest of things, but later than notify_fd processing, so that the
609 * notify processing can still figure out to which process/service a message belongs, before we reap the
610 * process. Also, process this before handling cgroup notifications, so that we always collect child exit
611 * status information before detecting that there's no process in a cgroup. */
612 r = sd_event_source_set_priority(m->signal_event_source, SD_EVENT_PRIORITY_NORMAL-6);
613 if (r < 0)
614 return r;
615
616 if (MANAGER_IS_SYSTEM(m))
617 return enable_special_signals(m);
618
619 return 0;
620 }
621
622 static char** sanitize_environment(char **l) {
623
624 /* Let's remove some environment variables that we need ourselves to communicate with our clients */
625 strv_env_unset_many(
626 l,
627 "CACHE_DIRECTORY",
628 "CONFIGURATION_DIRECTORY",
629 "CREDENTIALS_DIRECTORY",
630 "EXIT_CODE",
631 "EXIT_STATUS",
632 "INVOCATION_ID",
633 "JOURNAL_STREAM",
634 "LISTEN_FDNAMES",
635 "LISTEN_FDS",
636 "LISTEN_PID",
637 "LOGS_DIRECTORY",
638 "MAINPID",
639 "MANAGERPID",
640 "NOTIFY_SOCKET",
641 "PIDFILE",
642 "REMOTE_ADDR",
643 "REMOTE_PORT",
644 "RUNTIME_DIRECTORY",
645 "SERVICE_RESULT",
646 "STATE_DIRECTORY",
647 "WATCHDOG_PID",
648 "WATCHDOG_USEC",
649 NULL);
650
651 /* Let's order the environment alphabetically, just to make it pretty */
652 return strv_sort(l);
653 }
654
655 int manager_default_environment(Manager *m) {
656 int r;
657
658 assert(m);
659
660 m->transient_environment = strv_free(m->transient_environment);
661
662 if (MANAGER_IS_SYSTEM(m)) {
663 /* The system manager always starts with a clean
664 * environment for its children. It does not import
665 * the kernel's or the parents' exported variables.
666 *
667 * The initial passed environment is untouched to keep
668 * /proc/self/environ valid; it is used for tagging
669 * the init process inside containers. */
670 m->transient_environment = strv_new("PATH=" DEFAULT_PATH);
671 if (!m->transient_environment)
672 return log_oom();
673
674 /* Import locale variables LC_*= from configuration */
675 (void) locale_setup(&m->transient_environment);
676 } else {
677 /* The user manager passes its own environment along to its children, except for $PATH. */
678 m->transient_environment = strv_copy(environ);
679 if (!m->transient_environment)
680 return log_oom();
681
682 r = strv_env_replace_strdup(&m->transient_environment, "PATH=" DEFAULT_USER_PATH);
683 if (r < 0)
684 return log_oom();
685 }
686
687 sanitize_environment(m->transient_environment);
688
689 return 0;
690 }
691
692 static int manager_setup_prefix(Manager *m) {
693 struct table_entry {
694 uint64_t type;
695 const char *suffix;
696 };
697
698 static const struct table_entry paths_system[_EXEC_DIRECTORY_TYPE_MAX] = {
699 [EXEC_DIRECTORY_RUNTIME] = { SD_PATH_SYSTEM_RUNTIME, NULL },
700 [EXEC_DIRECTORY_STATE] = { SD_PATH_SYSTEM_STATE_PRIVATE, NULL },
701 [EXEC_DIRECTORY_CACHE] = { SD_PATH_SYSTEM_STATE_CACHE, NULL },
702 [EXEC_DIRECTORY_LOGS] = { SD_PATH_SYSTEM_STATE_LOGS, NULL },
703 [EXEC_DIRECTORY_CONFIGURATION] = { SD_PATH_SYSTEM_CONFIGURATION, NULL },
704 };
705
706 static const struct table_entry paths_user[_EXEC_DIRECTORY_TYPE_MAX] = {
707 [EXEC_DIRECTORY_RUNTIME] = { SD_PATH_USER_RUNTIME, NULL },
708 [EXEC_DIRECTORY_STATE] = { SD_PATH_USER_CONFIGURATION, NULL },
709 [EXEC_DIRECTORY_CACHE] = { SD_PATH_USER_STATE_CACHE, NULL },
710 [EXEC_DIRECTORY_LOGS] = { SD_PATH_USER_CONFIGURATION, "log" },
711 [EXEC_DIRECTORY_CONFIGURATION] = { SD_PATH_USER_CONFIGURATION, NULL },
712 };
713
714 assert(m);
715
716 const struct table_entry *p = MANAGER_IS_SYSTEM(m) ? paths_system : paths_user;
717 int r;
718
719 for (ExecDirectoryType i = 0; i < _EXEC_DIRECTORY_TYPE_MAX; i++) {
720 r = sd_path_lookup(p[i].type, p[i].suffix, &m->prefix[i]);
721 if (r < 0)
722 return log_warning_errno(r, "Failed to lookup %s path: %m",
723 exec_directory_type_to_string(i));
724 }
725
726 return 0;
727 }
728
729 static void manager_free_unit_name_maps(Manager *m) {
730 m->unit_id_map = hashmap_free(m->unit_id_map);
731 m->unit_name_map = hashmap_free(m->unit_name_map);
732 m->unit_path_cache = set_free(m->unit_path_cache);
733 m->unit_cache_timestamp_hash = 0;
734 }
735
736 static int manager_setup_run_queue(Manager *m) {
737 int r;
738
739 assert(m);
740 assert(!m->run_queue_event_source);
741
742 r = sd_event_add_defer(m->event, &m->run_queue_event_source, manager_dispatch_run_queue, m);
743 if (r < 0)
744 return r;
745
746 r = sd_event_source_set_priority(m->run_queue_event_source, SD_EVENT_PRIORITY_IDLE);
747 if (r < 0)
748 return r;
749
750 r = sd_event_source_set_enabled(m->run_queue_event_source, SD_EVENT_OFF);
751 if (r < 0)
752 return r;
753
754 (void) sd_event_source_set_description(m->run_queue_event_source, "manager-run-queue");
755
756 return 0;
757 }
758
759 static int manager_setup_sigchld_event_source(Manager *m) {
760 int r;
761
762 assert(m);
763 assert(!m->sigchld_event_source);
764
765 r = sd_event_add_defer(m->event, &m->sigchld_event_source, manager_dispatch_sigchld, m);
766 if (r < 0)
767 return r;
768
769 r = sd_event_source_set_priority(m->sigchld_event_source, SD_EVENT_PRIORITY_NORMAL-7);
770 if (r < 0)
771 return r;
772
773 r = sd_event_source_set_enabled(m->sigchld_event_source, SD_EVENT_OFF);
774 if (r < 0)
775 return r;
776
777 (void) sd_event_source_set_description(m->sigchld_event_source, "manager-sigchld");
778
779 return 0;
780 }
781
782 static int manager_find_credentials_dirs(Manager *m) {
783 const char *e;
784 int r;
785
786 assert(m);
787
788 r = get_credentials_dir(&e);
789 if (r < 0) {
790 if (r != -ENXIO)
791 log_debug_errno(r, "Failed to determine credentials directory, ignoring: %m");
792 } else {
793 m->received_credentials_directory = strdup(e);
794 if (!m->received_credentials_directory)
795 return -ENOMEM;
796 }
797
798 r = get_encrypted_credentials_dir(&e);
799 if (r < 0) {
800 if (r != -ENXIO)
801 log_debug_errno(r, "Failed to determine encrypted credentials directory, ignoring: %m");
802 } else {
803 m->received_encrypted_credentials_directory = strdup(e);
804 if (!m->received_encrypted_credentials_directory)
805 return -ENOMEM;
806 }
807
808 return 0;
809 }
810
811 void manager_set_switching_root(Manager *m, bool switching_root) {
812 m->switching_root = MANAGER_IS_SYSTEM(m) && switching_root;
813 }
814
815 int manager_new(LookupScope scope, ManagerTestRunFlags test_run_flags, Manager **_m) {
816 _cleanup_(manager_freep) Manager *m = NULL;
817 int r;
818
819 assert(_m);
820 assert(IN_SET(scope, LOOKUP_SCOPE_SYSTEM, LOOKUP_SCOPE_USER));
821
822 m = new(Manager, 1);
823 if (!m)
824 return -ENOMEM;
825
826 *m = (Manager) {
827 .unit_file_scope = scope,
828 .objective = _MANAGER_OBJECTIVE_INVALID,
829
830 .status_unit_format = STATUS_UNIT_FORMAT_DEFAULT,
831
832 .default_timer_accuracy_usec = USEC_PER_MINUTE,
833 .default_memory_accounting = MEMORY_ACCOUNTING_DEFAULT,
834 .default_tasks_accounting = true,
835 .default_tasks_max = TASKS_MAX_UNSET,
836 .default_timeout_start_usec = DEFAULT_TIMEOUT_USEC,
837 .default_timeout_stop_usec = DEFAULT_TIMEOUT_USEC,
838 .default_restart_usec = DEFAULT_RESTART_USEC,
839 .default_device_timeout_usec = DEFAULT_TIMEOUT_USEC,
840
841 .original_log_level = -1,
842 .original_log_target = _LOG_TARGET_INVALID,
843
844 .watchdog_overridden[WATCHDOG_RUNTIME] = USEC_INFINITY,
845 .watchdog_overridden[WATCHDOG_REBOOT] = USEC_INFINITY,
846 .watchdog_overridden[WATCHDOG_KEXEC] = USEC_INFINITY,
847 .watchdog_overridden[WATCHDOG_PRETIMEOUT] = USEC_INFINITY,
848
849 .show_status_overridden = _SHOW_STATUS_INVALID,
850
851 .notify_fd = -EBADF,
852 .cgroups_agent_fd = -EBADF,
853 .signal_fd = -EBADF,
854 .user_lookup_fds = PIPE_EBADF,
855 .private_listen_fd = -EBADF,
856 .dev_autofs_fd = -EBADF,
857 .cgroup_inotify_fd = -EBADF,
858 .pin_cgroupfs_fd = -EBADF,
859 .ask_password_inotify_fd = -EBADF,
860 .idle_pipe = { -EBADF, -EBADF, -EBADF, -EBADF},
861
862 /* start as id #1, so that we can leave #0 around as "null-like" value */
863 .current_job_id = 1,
864
865 .have_ask_password = -EINVAL, /* we don't know */
866 .first_boot = -1,
867 .test_run_flags = test_run_flags,
868
869 .default_oom_policy = OOM_STOP,
870 };
871
872 #if ENABLE_EFI
873 if (MANAGER_IS_SYSTEM(m) && detect_container() <= 0)
874 boot_timestamps(m->timestamps + MANAGER_TIMESTAMP_USERSPACE,
875 m->timestamps + MANAGER_TIMESTAMP_FIRMWARE,
876 m->timestamps + MANAGER_TIMESTAMP_LOADER);
877 #endif
878
879 /* Prepare log fields we can use for structured logging */
880 if (MANAGER_IS_SYSTEM(m)) {
881 m->unit_log_field = "UNIT=";
882 m->unit_log_format_string = "UNIT=%s";
883
884 m->invocation_log_field = "INVOCATION_ID=";
885 m->invocation_log_format_string = "INVOCATION_ID=%s";
886 } else {
887 m->unit_log_field = "USER_UNIT=";
888 m->unit_log_format_string = "USER_UNIT=%s";
889
890 m->invocation_log_field = "USER_INVOCATION_ID=";
891 m->invocation_log_format_string = "USER_INVOCATION_ID=%s";
892 }
893
894 /* Reboot immediately if the user hits C-A-D more often than 7x per 2s */
895 m->ctrl_alt_del_ratelimit = (const RateLimit) { .interval = 2 * USEC_PER_SEC, .burst = 7 };
896
897 r = manager_default_environment(m);
898 if (r < 0)
899 return r;
900
901 r = hashmap_ensure_allocated(&m->units, &string_hash_ops);
902 if (r < 0)
903 return r;
904
905 r = hashmap_ensure_allocated(&m->cgroup_unit, &path_hash_ops);
906 if (r < 0)
907 return r;
908
909 r = hashmap_ensure_allocated(&m->watch_bus, &string_hash_ops);
910 if (r < 0)
911 return r;
912
913 r = prioq_ensure_allocated(&m->run_queue, compare_job_priority);
914 if (r < 0)
915 return r;
916
917 r = manager_setup_prefix(m);
918 if (r < 0)
919 return r;
920
921 r = manager_find_credentials_dirs(m);
922 if (r < 0)
923 return r;
924
925 r = sd_event_default(&m->event);
926 if (r < 0)
927 return r;
928
929 r = manager_setup_run_queue(m);
930 if (r < 0)
931 return r;
932
933 if (FLAGS_SET(test_run_flags, MANAGER_TEST_RUN_MINIMAL)) {
934 m->cgroup_root = strdup("");
935 if (!m->cgroup_root)
936 return -ENOMEM;
937 } else {
938 r = manager_setup_signals(m);
939 if (r < 0)
940 return r;
941
942 r = manager_setup_cgroup(m);
943 if (r < 0)
944 return r;
945
946 r = manager_setup_time_change(m);
947 if (r < 0)
948 return r;
949
950 r = manager_read_timezone_stat(m);
951 if (r < 0)
952 return r;
953
954 (void) manager_setup_timezone_change(m);
955
956 r = manager_setup_sigchld_event_source(m);
957 if (r < 0)
958 return r;
959
960 #if HAVE_LIBBPF
961 if (MANAGER_IS_SYSTEM(m) && lsm_bpf_supported(/* initialize = */ true)) {
962 r = lsm_bpf_setup(m);
963 if (r < 0)
964 log_warning_errno(r, "Failed to setup LSM BPF, ignoring: %m");
965 }
966 #endif
967 }
968
969 if (test_run_flags == 0) {
970 if (MANAGER_IS_SYSTEM(m))
971 r = mkdir_label("/run/systemd/units", 0755);
972 else {
973 _cleanup_free_ char *units_path = NULL;
974 r = xdg_user_runtime_dir(&units_path, "/systemd/units");
975 if (r < 0)
976 return r;
977 r = mkdir_p_label(units_path, 0755);
978 }
979
980 if (r < 0 && r != -EEXIST)
981 return r;
982 }
983
984 m->taint_usr =
985 !in_initrd() &&
986 dir_is_empty("/usr", /* ignore_hidden_or_backup= */ false) > 0;
987
988 /* Note that we do not set up the notify fd here. We do that after deserialization,
989 * since they might have gotten serialized across the reexec. */
990
991 *_m = TAKE_PTR(m);
992
993 return 0;
994 }
995
996 static int manager_setup_notify(Manager *m) {
997 int r;
998
999 if (MANAGER_IS_TEST_RUN(m))
1000 return 0;
1001
1002 if (m->notify_fd < 0) {
1003 _cleanup_close_ int fd = -EBADF;
1004 union sockaddr_union sa;
1005 socklen_t sa_len;
1006
1007 /* First free all secondary fields */
1008 m->notify_socket = mfree(m->notify_socket);
1009 m->notify_event_source = sd_event_source_disable_unref(m->notify_event_source);
1010
1011 fd = socket(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC|SOCK_NONBLOCK, 0);
1012 if (fd < 0)
1013 return log_error_errno(errno, "Failed to allocate notification socket: %m");
1014
1015 fd_increase_rxbuf(fd, NOTIFY_RCVBUF_SIZE);
1016
1017 m->notify_socket = path_join(m->prefix[EXEC_DIRECTORY_RUNTIME], "systemd/notify");
1018 if (!m->notify_socket)
1019 return log_oom();
1020
1021 r = sockaddr_un_set_path(&sa.un, m->notify_socket);
1022 if (r < 0)
1023 return log_error_errno(r, "Notify socket '%s' not valid for AF_UNIX socket address, refusing.",
1024 m->notify_socket);
1025 sa_len = r;
1026
1027 (void) mkdir_parents_label(m->notify_socket, 0755);
1028 (void) sockaddr_un_unlink(&sa.un);
1029
1030 r = mac_selinux_bind(fd, &sa.sa, sa_len);
1031 if (r < 0)
1032 return log_error_errno(r, "bind(%s) failed: %m", m->notify_socket);
1033
1034 r = setsockopt_int(fd, SOL_SOCKET, SO_PASSCRED, true);
1035 if (r < 0)
1036 return log_error_errno(r, "SO_PASSCRED failed: %m");
1037
1038 m->notify_fd = TAKE_FD(fd);
1039
1040 log_debug("Using notification socket %s", m->notify_socket);
1041 }
1042
1043 if (!m->notify_event_source) {
1044 r = sd_event_add_io(m->event, &m->notify_event_source, m->notify_fd, EPOLLIN, manager_dispatch_notify_fd, m);
1045 if (r < 0)
1046 return log_error_errno(r, "Failed to allocate notify event source: %m");
1047
1048 /* Process notification messages a bit earlier than SIGCHLD, so that we can still identify to which
1049 * service an exit message belongs. */
1050 r = sd_event_source_set_priority(m->notify_event_source, SD_EVENT_PRIORITY_NORMAL-8);
1051 if (r < 0)
1052 return log_error_errno(r, "Failed to set priority of notify event source: %m");
1053
1054 (void) sd_event_source_set_description(m->notify_event_source, "manager-notify");
1055 }
1056
1057 return 0;
1058 }
1059
1060 static int manager_setup_cgroups_agent(Manager *m) {
1061
1062 static const union sockaddr_union sa = {
1063 .un.sun_family = AF_UNIX,
1064 .un.sun_path = "/run/systemd/cgroups-agent",
1065 };
1066 int r;
1067
1068 /* This creates a listening socket we receive cgroups agent messages on. We do not use D-Bus for delivering
1069 * these messages from the cgroups agent binary to PID 1, as the cgroups agent binary is very short-living, and
1070 * each instance of it needs a new D-Bus connection. Since D-Bus connections are SOCK_STREAM/AF_UNIX, on
1071 * overloaded systems the backlog of the D-Bus socket becomes relevant, as not more than the configured number
1072 * of D-Bus connections may be queued until the kernel will start dropping further incoming connections,
1073 * possibly resulting in lost cgroups agent messages. To avoid this, we'll use a private SOCK_DGRAM/AF_UNIX
1074 * socket, where no backlog is relevant as communication may take place without an actual connect() cycle, and
1075 * we thus won't lose messages.
1076 *
1077 * Note that PID 1 will forward the agent message to system bus, so that the user systemd instance may listen
1078 * to it. The system instance hence listens on this special socket, but the user instances listen on the system
1079 * bus for these messages. */
1080
1081 if (MANAGER_IS_TEST_RUN(m))
1082 return 0;
1083
1084 if (!MANAGER_IS_SYSTEM(m))
1085 return 0;
1086
1087 r = cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER);
1088 if (r < 0)
1089 return log_error_errno(r, "Failed to determine whether unified cgroups hierarchy is used: %m");
1090 if (r > 0) /* We don't need this anymore on the unified hierarchy */
1091 return 0;
1092
1093 if (m->cgroups_agent_fd < 0) {
1094 _cleanup_close_ int fd = -EBADF;
1095
1096 /* First free all secondary fields */
1097 m->cgroups_agent_event_source = sd_event_source_disable_unref(m->cgroups_agent_event_source);
1098
1099 fd = socket(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC|SOCK_NONBLOCK, 0);
1100 if (fd < 0)
1101 return log_error_errno(errno, "Failed to allocate cgroups agent socket: %m");
1102
1103 fd_increase_rxbuf(fd, CGROUPS_AGENT_RCVBUF_SIZE);
1104
1105 (void) sockaddr_un_unlink(&sa.un);
1106
1107 /* Only allow root to connect to this socket */
1108 WITH_UMASK(0077)
1109 r = bind(fd, &sa.sa, SOCKADDR_UN_LEN(sa.un));
1110 if (r < 0)
1111 return log_error_errno(errno, "bind(%s) failed: %m", sa.un.sun_path);
1112
1113 m->cgroups_agent_fd = TAKE_FD(fd);
1114 }
1115
1116 if (!m->cgroups_agent_event_source) {
1117 r = sd_event_add_io(m->event, &m->cgroups_agent_event_source, m->cgroups_agent_fd, EPOLLIN, manager_dispatch_cgroups_agent_fd, m);
1118 if (r < 0)
1119 return log_error_errno(r, "Failed to allocate cgroups agent event source: %m");
1120
1121 /* Process cgroups notifications early. Note that when the agent notification is received
1122 * we'll just enqueue the unit in the cgroup empty queue, hence pick a high priority than
1123 * that. Also see handling of cgroup inotify for the unified cgroup stuff. */
1124 r = sd_event_source_set_priority(m->cgroups_agent_event_source, SD_EVENT_PRIORITY_NORMAL-9);
1125 if (r < 0)
1126 return log_error_errno(r, "Failed to set priority of cgroups agent event source: %m");
1127
1128 (void) sd_event_source_set_description(m->cgroups_agent_event_source, "manager-cgroups-agent");
1129 }
1130
1131 return 0;
1132 }
1133
1134 static int manager_setup_user_lookup_fd(Manager *m) {
1135 int r;
1136
1137 assert(m);
1138
1139 /* Set up the socket pair used for passing UID/GID resolution results from forked off processes to PID
1140 * 1. Background: we can't do name lookups (NSS) from PID 1, since it might involve IPC and thus activation,
1141 * and we might hence deadlock on ourselves. Hence we do all user/group lookups asynchronously from the forked
1142 * off processes right before executing the binaries to start. In order to be able to clean up any IPC objects
1143 * created by a unit (see RemoveIPC=) we need to know in PID 1 the used UID/GID of the executed processes,
1144 * hence we establish this communication channel so that forked off processes can pass their UID/GID
1145 * information back to PID 1. The forked off processes send their resolved UID/GID to PID 1 in a simple
1146 * datagram, along with their unit name, so that we can share one communication socket pair among all units for
1147 * this purpose.
1148 *
1149 * You might wonder why we need a communication channel for this that is independent of the usual notification
1150 * socket scheme (i.e. $NOTIFY_SOCKET). The primary difference is about trust: data sent via the $NOTIFY_SOCKET
1151 * channel is only accepted if it originates from the right unit and if reception was enabled for it. The user
1152 * lookup socket OTOH is only accessible by PID 1 and its children until they exec(), and always available.
1153 *
1154 * Note that this function is called under two circumstances: when we first initialize (in which case we
1155 * allocate both the socket pair and the event source to listen on it), and when we deserialize after a reload
1156 * (in which case the socket pair already exists but we still need to allocate the event source for it). */
1157
1158 if (m->user_lookup_fds[0] < 0) {
1159
1160 /* Free all secondary fields */
1161 safe_close_pair(m->user_lookup_fds);
1162 m->user_lookup_event_source = sd_event_source_disable_unref(m->user_lookup_event_source);
1163
1164 if (socketpair(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC, 0, m->user_lookup_fds) < 0)
1165 return log_error_errno(errno, "Failed to allocate user lookup socket: %m");
1166
1167 (void) fd_increase_rxbuf(m->user_lookup_fds[0], NOTIFY_RCVBUF_SIZE);
1168 }
1169
1170 if (!m->user_lookup_event_source) {
1171 r = sd_event_add_io(m->event, &m->user_lookup_event_source, m->user_lookup_fds[0], EPOLLIN, manager_dispatch_user_lookup_fd, m);
1172 if (r < 0)
1173 return log_error_errno(errno, "Failed to allocate user lookup event source: %m");
1174
1175 /* Process even earlier than the notify event source, so that we always know first about valid UID/GID
1176 * resolutions */
1177 r = sd_event_source_set_priority(m->user_lookup_event_source, SD_EVENT_PRIORITY_NORMAL-11);
1178 if (r < 0)
1179 return log_error_errno(errno, "Failed to set priority of user lookup event source: %m");
1180
1181 (void) sd_event_source_set_description(m->user_lookup_event_source, "user-lookup");
1182 }
1183
1184 return 0;
1185 }
1186
1187 static unsigned manager_dispatch_cleanup_queue(Manager *m) {
1188 Unit *u;
1189 unsigned n = 0;
1190
1191 assert(m);
1192
1193 while ((u = m->cleanup_queue)) {
1194 assert(u->in_cleanup_queue);
1195
1196 unit_free(u);
1197 n++;
1198 }
1199
1200 return n;
1201 }
1202
1203 enum {
1204 GC_OFFSET_IN_PATH, /* This one is on the path we were traveling */
1205 GC_OFFSET_UNSURE, /* No clue */
1206 GC_OFFSET_GOOD, /* We still need this unit */
1207 GC_OFFSET_BAD, /* We don't need this unit anymore */
1208 _GC_OFFSET_MAX
1209 };
1210
1211 static void unit_gc_mark_good(Unit *u, unsigned gc_marker) {
1212 Unit *other;
1213
1214 u->gc_marker = gc_marker + GC_OFFSET_GOOD;
1215
1216 /* Recursively mark referenced units as GOOD as well */
1217 UNIT_FOREACH_DEPENDENCY(other, u, UNIT_ATOM_REFERENCES)
1218 if (other->gc_marker == gc_marker + GC_OFFSET_UNSURE)
1219 unit_gc_mark_good(other, gc_marker);
1220 }
1221
1222 static void unit_gc_sweep(Unit *u, unsigned gc_marker) {
1223 Unit *other;
1224 bool is_bad;
1225
1226 assert(u);
1227
1228 if (IN_SET(u->gc_marker - gc_marker,
1229 GC_OFFSET_GOOD, GC_OFFSET_BAD, GC_OFFSET_UNSURE, GC_OFFSET_IN_PATH))
1230 return;
1231
1232 if (u->in_cleanup_queue)
1233 goto bad;
1234
1235 if (!unit_may_gc(u))
1236 goto good;
1237
1238 u->gc_marker = gc_marker + GC_OFFSET_IN_PATH;
1239
1240 is_bad = true;
1241
1242 UNIT_FOREACH_DEPENDENCY(other, u, UNIT_ATOM_REFERENCED_BY) {
1243 unit_gc_sweep(other, gc_marker);
1244
1245 if (other->gc_marker == gc_marker + GC_OFFSET_GOOD)
1246 goto good;
1247
1248 if (other->gc_marker != gc_marker + GC_OFFSET_BAD)
1249 is_bad = false;
1250 }
1251
1252 LIST_FOREACH(refs_by_target, ref, u->refs_by_target) {
1253 unit_gc_sweep(ref->source, gc_marker);
1254
1255 if (ref->source->gc_marker == gc_marker + GC_OFFSET_GOOD)
1256 goto good;
1257
1258 if (ref->source->gc_marker != gc_marker + GC_OFFSET_BAD)
1259 is_bad = false;
1260 }
1261
1262 if (is_bad)
1263 goto bad;
1264
1265 /* We were unable to find anything out about this entry, so
1266 * let's investigate it later */
1267 u->gc_marker = gc_marker + GC_OFFSET_UNSURE;
1268 unit_add_to_gc_queue(u);
1269 return;
1270
1271 bad:
1272 /* We definitely know that this one is not useful anymore, so
1273 * let's mark it for deletion */
1274 u->gc_marker = gc_marker + GC_OFFSET_BAD;
1275 unit_add_to_cleanup_queue(u);
1276 return;
1277
1278 good:
1279 unit_gc_mark_good(u, gc_marker);
1280 }
1281
1282 static unsigned manager_dispatch_gc_unit_queue(Manager *m) {
1283 unsigned n = 0, gc_marker;
1284 Unit *u;
1285
1286 assert(m);
1287
1288 /* log_debug("Running GC..."); */
1289
1290 m->gc_marker += _GC_OFFSET_MAX;
1291 if (m->gc_marker + _GC_OFFSET_MAX <= _GC_OFFSET_MAX)
1292 m->gc_marker = 1;
1293
1294 gc_marker = m->gc_marker;
1295
1296 while ((u = m->gc_unit_queue)) {
1297 assert(u->in_gc_queue);
1298
1299 unit_gc_sweep(u, gc_marker);
1300
1301 LIST_REMOVE(gc_queue, m->gc_unit_queue, u);
1302 u->in_gc_queue = false;
1303
1304 n++;
1305
1306 if (IN_SET(u->gc_marker - gc_marker,
1307 GC_OFFSET_BAD, GC_OFFSET_UNSURE)) {
1308 if (u->id)
1309 log_unit_debug(u, "Collecting.");
1310 u->gc_marker = gc_marker + GC_OFFSET_BAD;
1311 unit_add_to_cleanup_queue(u);
1312 }
1313 }
1314
1315 return n;
1316 }
1317
1318 static unsigned manager_dispatch_gc_job_queue(Manager *m) {
1319 unsigned n = 0;
1320 Job *j;
1321
1322 assert(m);
1323
1324 while ((j = m->gc_job_queue)) {
1325 assert(j->in_gc_queue);
1326
1327 LIST_REMOVE(gc_queue, m->gc_job_queue, j);
1328 j->in_gc_queue = false;
1329
1330 n++;
1331
1332 if (!job_may_gc(j))
1333 continue;
1334
1335 log_unit_debug(j->unit, "Collecting job.");
1336 (void) job_finish_and_invalidate(j, JOB_COLLECTED, false, false);
1337 }
1338
1339 return n;
1340 }
1341
1342 static unsigned manager_dispatch_stop_when_unneeded_queue(Manager *m) {
1343 unsigned n = 0;
1344 Unit *u;
1345 int r;
1346
1347 assert(m);
1348
1349 while ((u = m->stop_when_unneeded_queue)) {
1350 _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL;
1351
1352 assert(u->in_stop_when_unneeded_queue);
1353 LIST_REMOVE(stop_when_unneeded_queue, m->stop_when_unneeded_queue, u);
1354 u->in_stop_when_unneeded_queue = false;
1355
1356 n++;
1357
1358 if (!unit_is_unneeded(u))
1359 continue;
1360
1361 log_unit_debug(u, "Unit is not needed anymore.");
1362
1363 /* If stopping a unit fails continuously we might enter a stop loop here, hence stop acting on the
1364 * service being unnecessary after a while. */
1365
1366 if (!ratelimit_below(&u->auto_start_stop_ratelimit)) {
1367 log_unit_warning(u, "Unit not needed anymore, but not stopping since we tried this too often recently.");
1368 continue;
1369 }
1370
1371 /* Ok, nobody needs us anymore. Sniff. Then let's commit suicide */
1372 r = manager_add_job(u->manager, JOB_STOP, u, JOB_FAIL, NULL, &error, NULL);
1373 if (r < 0)
1374 log_unit_warning_errno(u, r, "Failed to enqueue stop job, ignoring: %s", bus_error_message(&error, r));
1375 }
1376
1377 return n;
1378 }
1379
1380 static unsigned manager_dispatch_start_when_upheld_queue(Manager *m) {
1381 unsigned n = 0;
1382 Unit *u;
1383 int r;
1384
1385 assert(m);
1386
1387 while ((u = m->start_when_upheld_queue)) {
1388 _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL;
1389 Unit *culprit = NULL;
1390
1391 assert(u->in_start_when_upheld_queue);
1392 LIST_REMOVE(start_when_upheld_queue, m->start_when_upheld_queue, u);
1393 u->in_start_when_upheld_queue = false;
1394
1395 n++;
1396
1397 if (!unit_is_upheld_by_active(u, &culprit))
1398 continue;
1399
1400 log_unit_debug(u, "Unit is started because upheld by active unit %s.", culprit->id);
1401
1402 /* If stopping a unit fails continuously we might enter a stop loop here, hence stop acting on the
1403 * service being unnecessary after a while. */
1404
1405 if (!ratelimit_below(&u->auto_start_stop_ratelimit)) {
1406 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);
1407 continue;
1408 }
1409
1410 r = manager_add_job(u->manager, JOB_START, u, JOB_FAIL, NULL, &error, NULL);
1411 if (r < 0)
1412 log_unit_warning_errno(u, r, "Failed to enqueue start job, ignoring: %s", bus_error_message(&error, r));
1413 }
1414
1415 return n;
1416 }
1417
1418 static unsigned manager_dispatch_stop_when_bound_queue(Manager *m) {
1419 unsigned n = 0;
1420 Unit *u;
1421 int r;
1422
1423 assert(m);
1424
1425 while ((u = m->stop_when_bound_queue)) {
1426 _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL;
1427 Unit *culprit = NULL;
1428
1429 assert(u->in_stop_when_bound_queue);
1430 LIST_REMOVE(stop_when_bound_queue, m->stop_when_bound_queue, u);
1431 u->in_stop_when_bound_queue = false;
1432
1433 n++;
1434
1435 if (!unit_is_bound_by_inactive(u, &culprit))
1436 continue;
1437
1438 log_unit_debug(u, "Unit is stopped because bound to inactive unit %s.", culprit->id);
1439
1440 /* If stopping a unit fails continuously we might enter a stop loop here, hence stop acting on the
1441 * service being unnecessary after a while. */
1442
1443 if (!ratelimit_below(&u->auto_start_stop_ratelimit)) {
1444 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);
1445 continue;
1446 }
1447
1448 r = manager_add_job(u->manager, JOB_STOP, u, JOB_REPLACE, NULL, &error, NULL);
1449 if (r < 0)
1450 log_unit_warning_errno(u, r, "Failed to enqueue stop job, ignoring: %s", bus_error_message(&error, r));
1451 }
1452
1453 return n;
1454 }
1455
1456 static void manager_clear_jobs_and_units(Manager *m) {
1457 Unit *u;
1458
1459 assert(m);
1460
1461 while ((u = hashmap_first(m->units)))
1462 unit_free(u);
1463
1464 manager_dispatch_cleanup_queue(m);
1465
1466 assert(!m->load_queue);
1467 assert(prioq_isempty(m->run_queue));
1468 assert(!m->dbus_unit_queue);
1469 assert(!m->dbus_job_queue);
1470 assert(!m->cleanup_queue);
1471 assert(!m->gc_unit_queue);
1472 assert(!m->gc_job_queue);
1473 assert(!m->cgroup_realize_queue);
1474 assert(!m->cgroup_empty_queue);
1475 assert(!m->cgroup_oom_queue);
1476 assert(!m->target_deps_queue);
1477 assert(!m->stop_when_unneeded_queue);
1478 assert(!m->start_when_upheld_queue);
1479 assert(!m->stop_when_bound_queue);
1480
1481 assert(hashmap_isempty(m->jobs));
1482 assert(hashmap_isempty(m->units));
1483
1484 m->n_on_console = 0;
1485 m->n_running_jobs = 0;
1486 m->n_installed_jobs = 0;
1487 m->n_failed_jobs = 0;
1488 }
1489
1490 Manager* manager_free(Manager *m) {
1491 if (!m)
1492 return NULL;
1493
1494 manager_clear_jobs_and_units(m);
1495
1496 for (UnitType c = 0; c < _UNIT_TYPE_MAX; c++)
1497 if (unit_vtable[c]->shutdown)
1498 unit_vtable[c]->shutdown(m);
1499
1500 /* Keep the cgroup hierarchy in place except when we know we are going down for good */
1501 manager_shutdown_cgroup(m, IN_SET(m->objective, MANAGER_EXIT, MANAGER_REBOOT, MANAGER_POWEROFF, MANAGER_HALT, MANAGER_KEXEC));
1502
1503 lookup_paths_flush_generator(&m->lookup_paths);
1504
1505 bus_done(m);
1506 manager_varlink_done(m);
1507
1508 exec_runtime_vacuum(m);
1509 hashmap_free(m->exec_runtime_by_id);
1510
1511 dynamic_user_vacuum(m, false);
1512 hashmap_free(m->dynamic_users);
1513
1514 hashmap_free(m->units);
1515 hashmap_free(m->units_by_invocation_id);
1516 hashmap_free(m->jobs);
1517 hashmap_free(m->watch_pids);
1518 hashmap_free(m->watch_bus);
1519
1520 prioq_free(m->run_queue);
1521
1522 set_free(m->startup_units);
1523 set_free(m->failed_units);
1524
1525 sd_event_source_unref(m->signal_event_source);
1526 sd_event_source_unref(m->sigchld_event_source);
1527 sd_event_source_unref(m->notify_event_source);
1528 sd_event_source_unref(m->cgroups_agent_event_source);
1529 sd_event_source_unref(m->time_change_event_source);
1530 sd_event_source_unref(m->timezone_change_event_source);
1531 sd_event_source_unref(m->jobs_in_progress_event_source);
1532 sd_event_source_unref(m->run_queue_event_source);
1533 sd_event_source_unref(m->user_lookup_event_source);
1534
1535 safe_close(m->signal_fd);
1536 safe_close(m->notify_fd);
1537 safe_close(m->cgroups_agent_fd);
1538 safe_close_pair(m->user_lookup_fds);
1539
1540 manager_close_ask_password(m);
1541
1542 manager_close_idle_pipe(m);
1543
1544 sd_event_unref(m->event);
1545
1546 free(m->notify_socket);
1547
1548 lookup_paths_free(&m->lookup_paths);
1549 strv_free(m->transient_environment);
1550 strv_free(m->client_environment);
1551
1552 hashmap_free(m->cgroup_unit);
1553 manager_free_unit_name_maps(m);
1554
1555 free(m->switch_root);
1556 free(m->switch_root_init);
1557
1558 free(m->default_smack_process_label);
1559
1560 rlimit_free_all(m->rlimit);
1561
1562 assert(hashmap_isempty(m->units_requiring_mounts_for));
1563 hashmap_free(m->units_requiring_mounts_for);
1564
1565 hashmap_free(m->uid_refs);
1566 hashmap_free(m->gid_refs);
1567
1568 for (ExecDirectoryType dt = 0; dt < _EXEC_DIRECTORY_TYPE_MAX; dt++)
1569 m->prefix[dt] = mfree(m->prefix[dt]);
1570 free(m->received_credentials_directory);
1571 free(m->received_encrypted_credentials_directory);
1572
1573 free(m->watchdog_pretimeout_governor);
1574 free(m->watchdog_pretimeout_governor_overridden);
1575
1576 #if BPF_FRAMEWORK
1577 lsm_bpf_destroy(m->restrict_fs);
1578 #endif
1579
1580 return mfree(m);
1581 }
1582
1583 static void manager_enumerate_perpetual(Manager *m) {
1584 assert(m);
1585
1586 if (FLAGS_SET(m->test_run_flags, MANAGER_TEST_RUN_MINIMAL))
1587 return;
1588
1589 /* Let's ask every type to load all units from disk/kernel that it might know */
1590 for (UnitType c = 0; c < _UNIT_TYPE_MAX; c++) {
1591 if (!unit_type_supported(c)) {
1592 log_debug("Unit type .%s is not supported on this system.", unit_type_to_string(c));
1593 continue;
1594 }
1595
1596 if (unit_vtable[c]->enumerate_perpetual)
1597 unit_vtable[c]->enumerate_perpetual(m);
1598 }
1599 }
1600
1601 static void manager_enumerate(Manager *m) {
1602 assert(m);
1603
1604 if (FLAGS_SET(m->test_run_flags, MANAGER_TEST_RUN_MINIMAL))
1605 return;
1606
1607 /* Let's ask every type to load all units from disk/kernel that it might know */
1608 for (UnitType c = 0; c < _UNIT_TYPE_MAX; c++) {
1609 if (!unit_type_supported(c)) {
1610 log_debug("Unit type .%s is not supported on this system.", unit_type_to_string(c));
1611 continue;
1612 }
1613
1614 if (unit_vtable[c]->enumerate)
1615 unit_vtable[c]->enumerate(m);
1616 }
1617
1618 manager_dispatch_load_queue(m);
1619 }
1620
1621 static void manager_coldplug(Manager *m) {
1622 Unit *u;
1623 char *k;
1624 int r;
1625
1626 assert(m);
1627
1628 log_debug("Invoking unit coldplug() handlers%s", special_glyph(SPECIAL_GLYPH_ELLIPSIS));
1629
1630 /* Let's place the units back into their deserialized state */
1631 HASHMAP_FOREACH_KEY(u, k, m->units) {
1632
1633 /* ignore aliases */
1634 if (u->id != k)
1635 continue;
1636
1637 r = unit_coldplug(u);
1638 if (r < 0)
1639 log_warning_errno(r, "We couldn't coldplug %s, proceeding anyway: %m", u->id);
1640 }
1641 }
1642
1643 static void manager_catchup(Manager *m) {
1644 Unit *u;
1645 char *k;
1646
1647 assert(m);
1648
1649 log_debug("Invoking unit catchup() handlers%s", special_glyph(SPECIAL_GLYPH_ELLIPSIS));
1650
1651 /* Let's catch up on any state changes that happened while we were reloading/reexecing */
1652 HASHMAP_FOREACH_KEY(u, k, m->units) {
1653
1654 /* ignore aliases */
1655 if (u->id != k)
1656 continue;
1657
1658 unit_catchup(u);
1659 }
1660 }
1661
1662 static void manager_distribute_fds(Manager *m, FDSet *fds) {
1663 Unit *u;
1664
1665 assert(m);
1666
1667 HASHMAP_FOREACH(u, m->units) {
1668
1669 if (fdset_size(fds) <= 0)
1670 break;
1671
1672 if (!UNIT_VTABLE(u)->distribute_fds)
1673 continue;
1674
1675 UNIT_VTABLE(u)->distribute_fds(u, fds);
1676 }
1677 }
1678
1679 static bool manager_dbus_is_running(Manager *m, bool deserialized) {
1680 Unit *u;
1681
1682 assert(m);
1683
1684 /* This checks whether the dbus instance we are supposed to expose our APIs on is up. We check both the socket
1685 * and the service unit. If the 'deserialized' parameter is true we'll check the deserialized state of the unit
1686 * rather than the current one. */
1687
1688 if (MANAGER_IS_TEST_RUN(m))
1689 return false;
1690
1691 u = manager_get_unit(m, SPECIAL_DBUS_SOCKET);
1692 if (!u)
1693 return false;
1694 if ((deserialized ? SOCKET(u)->deserialized_state : SOCKET(u)->state) != SOCKET_RUNNING)
1695 return false;
1696
1697 u = manager_get_unit(m, SPECIAL_DBUS_SERVICE);
1698 if (!u)
1699 return false;
1700 if (!IN_SET((deserialized ? SERVICE(u)->deserialized_state : SERVICE(u)->state), SERVICE_RUNNING, SERVICE_RELOAD))
1701 return false;
1702
1703 return true;
1704 }
1705
1706 static void manager_setup_bus(Manager *m) {
1707 assert(m);
1708
1709 /* Let's set up our private bus connection now, unconditionally */
1710 (void) bus_init_private(m);
1711
1712 /* If we are in --user mode also connect to the system bus now */
1713 if (MANAGER_IS_USER(m))
1714 (void) bus_init_system(m);
1715
1716 /* Let's connect to the bus now, but only if the unit is supposed to be up */
1717 if (manager_dbus_is_running(m, MANAGER_IS_RELOADING(m))) {
1718 (void) bus_init_api(m);
1719
1720 if (MANAGER_IS_SYSTEM(m))
1721 (void) bus_init_system(m);
1722 }
1723 }
1724
1725 static void manager_preset_all(Manager *m) {
1726 int r;
1727
1728 assert(m);
1729
1730 if (m->first_boot <= 0)
1731 return;
1732
1733 if (!MANAGER_IS_SYSTEM(m))
1734 return;
1735
1736 if (MANAGER_IS_TEST_RUN(m))
1737 return;
1738
1739 /* If this is the first boot, and we are in the host system, then preset everything */
1740 UnitFilePresetMode mode = FIRST_BOOT_FULL_PRESET ? UNIT_FILE_PRESET_FULL : UNIT_FILE_PRESET_ENABLE_ONLY;
1741
1742 r = unit_file_preset_all(LOOKUP_SCOPE_SYSTEM, 0, NULL, mode, NULL, 0);
1743 if (r < 0)
1744 log_full_errno(r == -EEXIST ? LOG_NOTICE : LOG_WARNING, r,
1745 "Failed to populate /etc with preset unit settings, ignoring: %m");
1746 else
1747 log_info("Populated /etc with preset unit settings.");
1748 }
1749
1750 static void manager_ready(Manager *m) {
1751 assert(m);
1752
1753 /* After having loaded everything, do the final round of catching up with what might have changed */
1754
1755 m->objective = MANAGER_OK; /* Tell everyone we are up now */
1756
1757 /* It might be safe to log to the journal now and connect to dbus */
1758 manager_recheck_journal(m);
1759 manager_recheck_dbus(m);
1760
1761 /* Let's finally catch up with any changes that took place while we were reloading/reexecing */
1762 manager_catchup(m);
1763
1764 /* Create a file which will indicate when the manager started loading units the last time. */
1765 if (MANAGER_IS_SYSTEM(m))
1766 (void) touch_file("/run/systemd/systemd-units-load", false,
1767 m->timestamps[MANAGER_TIMESTAMP_UNITS_LOAD].realtime ?: now(CLOCK_REALTIME),
1768 UID_INVALID, GID_INVALID, 0444);
1769 }
1770
1771 Manager* manager_reloading_start(Manager *m) {
1772 m->n_reloading++;
1773 dual_timestamp_get(m->timestamps + MANAGER_TIMESTAMP_UNITS_LOAD);
1774 return m;
1775 }
1776
1777 void manager_reloading_stopp(Manager **m) {
1778 if (*m) {
1779 assert((*m)->n_reloading > 0);
1780 (*m)->n_reloading--;
1781 }
1782 }
1783
1784 int manager_startup(Manager *m, FILE *serialization, FDSet *fds, const char *root) {
1785 int r;
1786
1787 assert(m);
1788
1789 /* If we are running in test mode, we still want to run the generators,
1790 * but we should not touch the real generator directories. */
1791 r = lookup_paths_init_or_warn(&m->lookup_paths, m->unit_file_scope,
1792 MANAGER_IS_TEST_RUN(m) ? LOOKUP_PATHS_TEMPORARY_GENERATED : 0,
1793 root);
1794 if (r < 0)
1795 return r;
1796
1797 dual_timestamp_get(m->timestamps + manager_timestamp_initrd_mangle(MANAGER_TIMESTAMP_GENERATORS_START));
1798 r = manager_run_environment_generators(m);
1799 if (r >= 0)
1800 r = manager_run_generators(m);
1801 dual_timestamp_get(m->timestamps + manager_timestamp_initrd_mangle(MANAGER_TIMESTAMP_GENERATORS_FINISH));
1802 if (r < 0)
1803 return r;
1804
1805 manager_preset_all(m);
1806
1807 lookup_paths_log(&m->lookup_paths);
1808
1809 {
1810 /* This block is (optionally) done with the reloading counter bumped */
1811 _unused_ _cleanup_(manager_reloading_stopp) Manager *reloading = NULL;
1812
1813 /* If we will deserialize make sure that during enumeration this is already known, so we increase the
1814 * counter here already */
1815 if (serialization)
1816 reloading = manager_reloading_start(m);
1817
1818 /* First, enumerate what we can from all config files */
1819 dual_timestamp_get(m->timestamps + manager_timestamp_initrd_mangle(MANAGER_TIMESTAMP_UNITS_LOAD_START));
1820 manager_enumerate_perpetual(m);
1821 manager_enumerate(m);
1822 dual_timestamp_get(m->timestamps + manager_timestamp_initrd_mangle(MANAGER_TIMESTAMP_UNITS_LOAD_FINISH));
1823
1824 /* Second, deserialize if there is something to deserialize */
1825 if (serialization) {
1826 r = manager_deserialize(m, serialization, fds);
1827 if (r < 0)
1828 return log_error_errno(r, "Deserialization failed: %m");
1829 }
1830
1831 /* Any fds left? Find some unit which wants them. This is useful to allow container managers to pass
1832 * some file descriptors to us pre-initialized. This enables socket-based activation of entire
1833 * containers. */
1834 manager_distribute_fds(m, fds);
1835
1836 /* We might have deserialized the notify fd, but if we didn't then let's create the bus now */
1837 r = manager_setup_notify(m);
1838 if (r < 0)
1839 /* No sense to continue without notifications, our children would fail anyway. */
1840 return r;
1841
1842 r = manager_setup_cgroups_agent(m);
1843 if (r < 0)
1844 /* Likewise, no sense to continue without empty cgroup notifications. */
1845 return r;
1846
1847 r = manager_setup_user_lookup_fd(m);
1848 if (r < 0)
1849 /* This shouldn't fail, except if things are really broken. */
1850 return r;
1851
1852 /* Connect to the bus if we are good for it */
1853 manager_setup_bus(m);
1854
1855 /* Now that we are connected to all possible buses, let's deserialize who is tracking us. */
1856 r = bus_track_coldplug(m, &m->subscribed, false, m->deserialized_subscribed);
1857 if (r < 0)
1858 log_warning_errno(r, "Failed to deserialized tracked clients, ignoring: %m");
1859 m->deserialized_subscribed = strv_free(m->deserialized_subscribed);
1860
1861 r = manager_varlink_init(m);
1862 if (r < 0)
1863 log_warning_errno(r, "Failed to set up Varlink, ignoring: %m");
1864
1865 /* Third, fire things up! */
1866 manager_coldplug(m);
1867
1868 /* Clean up runtime objects */
1869 manager_vacuum(m);
1870
1871 if (serialization)
1872 /* Let's wait for the UnitNew/JobNew messages being sent, before we notify that the
1873 * reload is finished */
1874 m->send_reloading_done = true;
1875 }
1876
1877 manager_ready(m);
1878
1879 manager_set_switching_root(m, false);
1880
1881 return 0;
1882 }
1883
1884 int manager_add_job(
1885 Manager *m,
1886 JobType type,
1887 Unit *unit,
1888 JobMode mode,
1889 Set *affected_jobs,
1890 sd_bus_error *error,
1891 Job **ret) {
1892
1893 Transaction *tr;
1894 int r;
1895
1896 assert(m);
1897 assert(type < _JOB_TYPE_MAX);
1898 assert(unit);
1899 assert(mode < _JOB_MODE_MAX);
1900
1901 if (mode == JOB_ISOLATE && type != JOB_START)
1902 return sd_bus_error_set(error, SD_BUS_ERROR_INVALID_ARGS, "Isolate is only valid for start.");
1903
1904 if (mode == JOB_ISOLATE && !unit->allow_isolate)
1905 return sd_bus_error_set(error, BUS_ERROR_NO_ISOLATION, "Operation refused, unit may not be isolated.");
1906
1907 if (mode == JOB_TRIGGERING && type != JOB_STOP)
1908 return sd_bus_error_set(error, SD_BUS_ERROR_INVALID_ARGS, "--job-mode=triggering is only valid for stop.");
1909
1910 log_unit_debug(unit, "Trying to enqueue job %s/%s/%s", unit->id, job_type_to_string(type), job_mode_to_string(mode));
1911
1912 type = job_type_collapse(type, unit);
1913
1914 tr = transaction_new(mode == JOB_REPLACE_IRREVERSIBLY);
1915 if (!tr)
1916 return -ENOMEM;
1917
1918 r = transaction_add_job_and_dependencies(tr, type, unit, NULL, true, false,
1919 IN_SET(mode, JOB_IGNORE_DEPENDENCIES, JOB_IGNORE_REQUIREMENTS),
1920 mode == JOB_IGNORE_DEPENDENCIES, error);
1921 if (r < 0)
1922 goto tr_abort;
1923
1924 if (mode == JOB_ISOLATE) {
1925 r = transaction_add_isolate_jobs(tr, m);
1926 if (r < 0)
1927 goto tr_abort;
1928 }
1929
1930 if (mode == JOB_TRIGGERING) {
1931 r = transaction_add_triggering_jobs(tr, unit);
1932 if (r < 0)
1933 goto tr_abort;
1934 }
1935
1936 r = transaction_activate(tr, m, mode, affected_jobs, error);
1937 if (r < 0)
1938 goto tr_abort;
1939
1940 log_unit_debug(unit,
1941 "Enqueued job %s/%s as %u", unit->id,
1942 job_type_to_string(type), (unsigned) tr->anchor_job->id);
1943
1944 if (ret)
1945 *ret = tr->anchor_job;
1946
1947 transaction_free(tr);
1948 return 0;
1949
1950 tr_abort:
1951 transaction_abort(tr);
1952 transaction_free(tr);
1953 return r;
1954 }
1955
1956 int manager_add_job_by_name(Manager *m, JobType type, const char *name, JobMode mode, Set *affected_jobs, sd_bus_error *e, Job **ret) {
1957 Unit *unit = NULL; /* just to appease gcc, initialization is not really necessary */
1958 int r;
1959
1960 assert(m);
1961 assert(type < _JOB_TYPE_MAX);
1962 assert(name);
1963 assert(mode < _JOB_MODE_MAX);
1964
1965 r = manager_load_unit(m, name, NULL, NULL, &unit);
1966 if (r < 0)
1967 return r;
1968 assert(unit);
1969
1970 return manager_add_job(m, type, unit, mode, affected_jobs, e, ret);
1971 }
1972
1973 int manager_add_job_by_name_and_warn(Manager *m, JobType type, const char *name, JobMode mode, Set *affected_jobs, Job **ret) {
1974 _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL;
1975 int r;
1976
1977 assert(m);
1978 assert(type < _JOB_TYPE_MAX);
1979 assert(name);
1980 assert(mode < _JOB_MODE_MAX);
1981
1982 r = manager_add_job_by_name(m, type, name, mode, affected_jobs, &error, ret);
1983 if (r < 0)
1984 return log_warning_errno(r, "Failed to enqueue %s job for %s: %s", job_mode_to_string(mode), name, bus_error_message(&error, r));
1985
1986 return r;
1987 }
1988
1989 int manager_propagate_reload(Manager *m, Unit *unit, JobMode mode, sd_bus_error *e) {
1990 int r;
1991 Transaction *tr;
1992
1993 assert(m);
1994 assert(unit);
1995 assert(mode < _JOB_MODE_MAX);
1996 assert(mode != JOB_ISOLATE); /* Isolate is only valid for start */
1997
1998 tr = transaction_new(mode == JOB_REPLACE_IRREVERSIBLY);
1999 if (!tr)
2000 return -ENOMEM;
2001
2002 /* We need an anchor job */
2003 r = transaction_add_job_and_dependencies(tr, JOB_NOP, unit, NULL, false, false, true, true, e);
2004 if (r < 0)
2005 goto tr_abort;
2006
2007 /* Failure in adding individual dependencies is ignored, so this always succeeds. */
2008 transaction_add_propagate_reload_jobs(tr, unit, tr->anchor_job, mode == JOB_IGNORE_DEPENDENCIES, e);
2009
2010 r = transaction_activate(tr, m, mode, NULL, e);
2011 if (r < 0)
2012 goto tr_abort;
2013
2014 transaction_free(tr);
2015 return 0;
2016
2017 tr_abort:
2018 transaction_abort(tr);
2019 transaction_free(tr);
2020 return r;
2021 }
2022
2023 Job *manager_get_job(Manager *m, uint32_t id) {
2024 assert(m);
2025
2026 return hashmap_get(m->jobs, UINT32_TO_PTR(id));
2027 }
2028
2029 Unit *manager_get_unit(Manager *m, const char *name) {
2030 assert(m);
2031 assert(name);
2032
2033 return hashmap_get(m->units, name);
2034 }
2035
2036 static int manager_dispatch_target_deps_queue(Manager *m) {
2037 Unit *u;
2038 int r = 0;
2039
2040 assert(m);
2041
2042 while ((u = m->target_deps_queue)) {
2043 _cleanup_free_ Unit **targets = NULL;
2044 int n_targets;
2045
2046 assert(u->in_target_deps_queue);
2047
2048 LIST_REMOVE(target_deps_queue, u->manager->target_deps_queue, u);
2049 u->in_target_deps_queue = false;
2050
2051 /* Take an "atomic" snapshot of dependencies here, as the call below will likely modify the
2052 * dependencies, and we can't have it that hash tables we iterate through are modified while
2053 * we are iterating through them. */
2054 n_targets = unit_get_dependency_array(u, UNIT_ATOM_DEFAULT_TARGET_DEPENDENCIES, &targets);
2055 if (n_targets < 0)
2056 return n_targets;
2057
2058 for (int i = 0; i < n_targets; i++) {
2059 r = unit_add_default_target_dependency(u, targets[i]);
2060 if (r < 0)
2061 return r;
2062 }
2063 }
2064
2065 return r;
2066 }
2067
2068 unsigned manager_dispatch_load_queue(Manager *m) {
2069 Unit *u;
2070 unsigned n = 0;
2071
2072 assert(m);
2073
2074 /* Make sure we are not run recursively */
2075 if (m->dispatching_load_queue)
2076 return 0;
2077
2078 m->dispatching_load_queue = true;
2079
2080 /* Dispatches the load queue. Takes a unit from the queue and
2081 * tries to load its data until the queue is empty */
2082
2083 while ((u = m->load_queue)) {
2084 assert(u->in_load_queue);
2085
2086 unit_load(u);
2087 n++;
2088 }
2089
2090 m->dispatching_load_queue = false;
2091
2092 /* Dispatch the units waiting for their target dependencies to be added now, as all targets that we know about
2093 * should be loaded and have aliases resolved */
2094 (void) manager_dispatch_target_deps_queue(m);
2095
2096 return n;
2097 }
2098
2099 bool manager_unit_cache_should_retry_load(Unit *u) {
2100 assert(u);
2101
2102 /* Automatic reloading from disk only applies to units which were not found sometime in the past, and
2103 * the not-found stub is kept pinned in the unit graph by dependencies. For units that were
2104 * previously loaded, we don't do automatic reloading, and daemon-reload is necessary to update. */
2105 if (u->load_state != UNIT_NOT_FOUND)
2106 return false;
2107
2108 /* The cache has been updated since the last time we tried to load the unit. There might be new
2109 * fragment paths to read. */
2110 if (u->manager->unit_cache_timestamp_hash != u->fragment_not_found_timestamp_hash)
2111 return true;
2112
2113 /* The cache needs to be updated because there are modifications on disk. */
2114 return !lookup_paths_timestamp_hash_same(&u->manager->lookup_paths, u->manager->unit_cache_timestamp_hash, NULL);
2115 }
2116
2117 int manager_load_unit_prepare(
2118 Manager *m,
2119 const char *name,
2120 const char *path,
2121 sd_bus_error *e,
2122 Unit **ret) {
2123
2124 _cleanup_(unit_freep) Unit *cleanup_unit = NULL;
2125 int r;
2126
2127 assert(m);
2128 assert(ret);
2129
2130 /* This will prepare the unit for loading, but not actually load anything from disk. */
2131
2132 if (path && !path_is_absolute(path))
2133 return sd_bus_error_setf(e, SD_BUS_ERROR_INVALID_ARGS, "Path %s is not absolute.", path);
2134
2135 if (!name) {
2136 /* 'name' and 'path' must not both be null. Check here 'path' using assert_se() to
2137 * workaround a bug in gcc that generates a -Wnonnull warning when calling basename(),
2138 * but this cannot be possible in any code path (See #6119). */
2139 assert_se(path);
2140 name = basename(path);
2141 }
2142
2143 UnitType t = unit_name_to_type(name);
2144
2145 if (t == _UNIT_TYPE_INVALID || !unit_name_is_valid(name, UNIT_NAME_PLAIN|UNIT_NAME_INSTANCE)) {
2146 if (unit_name_is_valid(name, UNIT_NAME_TEMPLATE))
2147 return sd_bus_error_setf(e, SD_BUS_ERROR_INVALID_ARGS, "Unit name %s is missing the instance name.", name);
2148
2149 return sd_bus_error_setf(e, SD_BUS_ERROR_INVALID_ARGS, "Unit name %s is not valid.", name);
2150 }
2151
2152 Unit *unit = manager_get_unit(m, name);
2153 if (unit) {
2154 /* The time-based cache allows to start new units without daemon-reload,
2155 * but if they are already referenced (because of dependencies or ordering)
2156 * then we have to force a load of the fragment. As an optimization, check
2157 * first if anything in the usual paths was modified since the last time
2158 * the cache was loaded. Also check if the last time an attempt to load the
2159 * unit was made was before the most recent cache refresh, so that we know
2160 * we need to try again — even if the cache is current, it might have been
2161 * updated in a different context before we had a chance to retry loading
2162 * this particular unit. */
2163 if (manager_unit_cache_should_retry_load(unit))
2164 unit->load_state = UNIT_STUB;
2165 else {
2166 *ret = unit;
2167 return 0; /* The unit was already loaded */
2168 }
2169 } else {
2170 unit = cleanup_unit = unit_new(m, unit_vtable[t]->object_size);
2171 if (!unit)
2172 return -ENOMEM;
2173 }
2174
2175 if (path) {
2176 r = free_and_strdup(&unit->fragment_path, path);
2177 if (r < 0)
2178 return r;
2179 }
2180
2181 r = unit_add_name(unit, name);
2182 if (r < 0)
2183 return r;
2184
2185 unit_add_to_load_queue(unit);
2186 unit_add_to_dbus_queue(unit);
2187 unit_add_to_gc_queue(unit);
2188
2189 *ret = unit;
2190 TAKE_PTR(cleanup_unit);
2191
2192 return 1; /* The unit was added the load queue */
2193 }
2194
2195 int manager_load_unit(
2196 Manager *m,
2197 const char *name,
2198 const char *path,
2199 sd_bus_error *e,
2200 Unit **ret) {
2201 int r;
2202
2203 assert(m);
2204 assert(ret);
2205
2206 /* This will load the unit config, but not actually start any services or anything. */
2207
2208 r = manager_load_unit_prepare(m, name, path, e, ret);
2209 if (r <= 0)
2210 return r;
2211
2212 /* Unit was newly loaded */
2213 manager_dispatch_load_queue(m);
2214 *ret = unit_follow_merge(*ret);
2215 return 0;
2216 }
2217
2218 int manager_load_startable_unit_or_warn(
2219 Manager *m,
2220 const char *name,
2221 const char *path,
2222 Unit **ret) {
2223
2224 /* Load a unit, make sure it loaded fully and is not masked. */
2225
2226 _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL;
2227 Unit *unit;
2228 int r;
2229
2230 r = manager_load_unit(m, name, path, &error, &unit);
2231 if (r < 0)
2232 return log_error_errno(r, "Failed to load %s %s: %s",
2233 name ? "unit" : "unit file", name ?: path,
2234 bus_error_message(&error, r));
2235
2236 r = bus_unit_validate_load_state(unit, &error);
2237 if (r < 0)
2238 return log_error_errno(r, "%s", bus_error_message(&error, r));
2239
2240 *ret = unit;
2241 return 0;
2242 }
2243
2244 void manager_clear_jobs(Manager *m) {
2245 Job *j;
2246
2247 assert(m);
2248
2249 while ((j = hashmap_first(m->jobs)))
2250 /* No need to recurse. We're cancelling all jobs. */
2251 job_finish_and_invalidate(j, JOB_CANCELED, false, false);
2252 }
2253
2254 void manager_unwatch_pid(Manager *m, pid_t pid) {
2255 assert(m);
2256
2257 /* First let's drop the unit keyed as "pid". */
2258 (void) hashmap_remove(m->watch_pids, PID_TO_PTR(pid));
2259
2260 /* Then, let's also drop the array keyed by -pid. */
2261 free(hashmap_remove(m->watch_pids, PID_TO_PTR(-pid)));
2262 }
2263
2264 static int manager_dispatch_run_queue(sd_event_source *source, void *userdata) {
2265 Manager *m = ASSERT_PTR(userdata);
2266 Job *j;
2267
2268 assert(source);
2269
2270 while ((j = prioq_peek(m->run_queue))) {
2271 assert(j->installed);
2272 assert(j->in_run_queue);
2273
2274 (void) job_run_and_invalidate(j);
2275 }
2276
2277 if (m->n_running_jobs > 0)
2278 manager_watch_jobs_in_progress(m);
2279
2280 if (m->n_on_console > 0)
2281 manager_watch_idle_pipe(m);
2282
2283 return 1;
2284 }
2285
2286 void manager_trigger_run_queue(Manager *m) {
2287 int r;
2288
2289 assert(m);
2290
2291 r = sd_event_source_set_enabled(
2292 m->run_queue_event_source,
2293 prioq_isempty(m->run_queue) ? SD_EVENT_OFF : SD_EVENT_ONESHOT);
2294 if (r < 0)
2295 log_warning_errno(r, "Failed to enable job run queue event source, ignoring: %m");
2296 }
2297
2298 static unsigned manager_dispatch_dbus_queue(Manager *m) {
2299 unsigned n = 0, budget;
2300 Unit *u;
2301 Job *j;
2302
2303 assert(m);
2304
2305 /* When we are reloading, let's not wait with generating signals, since we need to exit the manager as quickly
2306 * as we can. There's no point in throttling generation of signals in that case. */
2307 if (MANAGER_IS_RELOADING(m) || m->send_reloading_done || m->pending_reload_message)
2308 budget = UINT_MAX; /* infinite budget in this case */
2309 else {
2310 /* Anything to do at all? */
2311 if (!m->dbus_unit_queue && !m->dbus_job_queue)
2312 return 0;
2313
2314 /* Do we have overly many messages queued at the moment? If so, let's not enqueue more on top, let's
2315 * sit this cycle out, and process things in a later cycle when the queues got a bit emptier. */
2316 if (manager_bus_n_queued_write(m) > MANAGER_BUS_BUSY_THRESHOLD)
2317 return 0;
2318
2319 /* Only process a certain number of units/jobs per event loop iteration. Even if the bus queue wasn't
2320 * overly full before this call we shouldn't increase it in size too wildly in one step, and we
2321 * shouldn't monopolize CPU time with generating these messages. Note the difference in counting of
2322 * this "budget" and the "threshold" above: the "budget" is decreased only once per generated message,
2323 * regardless how many buses/direct connections it is enqueued on, while the "threshold" is applied to
2324 * each queued instance of bus message, i.e. if the same message is enqueued to five buses/direct
2325 * connections it will be counted five times. This difference in counting ("references"
2326 * vs. "instances") is primarily a result of the fact that it's easier to implement it this way,
2327 * however it also reflects the thinking that the "threshold" should put a limit on used queue memory,
2328 * i.e. space, while the "budget" should put a limit on time. Also note that the "threshold" is
2329 * currently chosen much higher than the "budget". */
2330 budget = MANAGER_BUS_MESSAGE_BUDGET;
2331 }
2332
2333 while (budget != 0 && (u = m->dbus_unit_queue)) {
2334
2335 assert(u->in_dbus_queue);
2336
2337 bus_unit_send_change_signal(u);
2338 n++;
2339
2340 if (budget != UINT_MAX)
2341 budget--;
2342 }
2343
2344 while (budget != 0 && (j = m->dbus_job_queue)) {
2345 assert(j->in_dbus_queue);
2346
2347 bus_job_send_change_signal(j);
2348 n++;
2349
2350 if (budget != UINT_MAX)
2351 budget--;
2352 }
2353
2354 if (m->send_reloading_done) {
2355 m->send_reloading_done = false;
2356 bus_manager_send_reloading(m, false);
2357 n++;
2358 }
2359
2360 if (m->pending_reload_message) {
2361 bus_send_pending_reload_message(m);
2362 n++;
2363 }
2364
2365 return n;
2366 }
2367
2368 static int manager_dispatch_cgroups_agent_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
2369 Manager *m = userdata;
2370 char buf[PATH_MAX];
2371 ssize_t n;
2372
2373 n = recv(fd, buf, sizeof(buf), 0);
2374 if (n < 0)
2375 return log_error_errno(errno, "Failed to read cgroups agent message: %m");
2376 if (n == 0) {
2377 log_error("Got zero-length cgroups agent message, ignoring.");
2378 return 0;
2379 }
2380 if ((size_t) n >= sizeof(buf)) {
2381 log_error("Got overly long cgroups agent message, ignoring.");
2382 return 0;
2383 }
2384
2385 if (memchr(buf, 0, n)) {
2386 log_error("Got cgroups agent message with embedded NUL byte, ignoring.");
2387 return 0;
2388 }
2389 buf[n] = 0;
2390
2391 manager_notify_cgroup_empty(m, buf);
2392 (void) bus_forward_agent_released(m, buf);
2393
2394 return 0;
2395 }
2396
2397 static bool manager_process_barrier_fd(char * const *tags, FDSet *fds) {
2398
2399 /* nothing else must be sent when using BARRIER=1 */
2400 if (strv_contains(tags, "BARRIER=1")) {
2401 if (strv_length(tags) == 1) {
2402 if (fdset_size(fds) != 1)
2403 log_warning("Got incorrect number of fds with BARRIER=1, closing them.");
2404 } else
2405 log_warning("Extra notification messages sent with BARRIER=1, ignoring everything.");
2406
2407 /* Drop the message if BARRIER=1 was found */
2408 return true;
2409 }
2410
2411 return false;
2412 }
2413
2414 static void manager_invoke_notify_message(
2415 Manager *m,
2416 Unit *u,
2417 const struct ucred *ucred,
2418 char * const *tags,
2419 FDSet *fds) {
2420
2421 assert(m);
2422 assert(u);
2423 assert(ucred);
2424 assert(tags);
2425
2426 if (u->notifygen == m->notifygen) /* Already invoked on this same unit in this same iteration? */
2427 return;
2428 u->notifygen = m->notifygen;
2429
2430 if (UNIT_VTABLE(u)->notify_message)
2431 UNIT_VTABLE(u)->notify_message(u, ucred, tags, fds);
2432
2433 else if (DEBUG_LOGGING) {
2434 _cleanup_free_ char *buf = NULL, *x = NULL, *y = NULL;
2435
2436 buf = strv_join(tags, ", ");
2437 if (buf)
2438 x = ellipsize(buf, 20, 90);
2439 if (x)
2440 y = cescape(x);
2441
2442 log_unit_debug(u, "Got notification message \"%s\", ignoring.", strnull(y));
2443 }
2444 }
2445
2446 static int manager_dispatch_notify_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
2447
2448 _cleanup_fdset_free_ FDSet *fds = NULL;
2449 Manager *m = ASSERT_PTR(userdata);
2450 char buf[NOTIFY_BUFFER_MAX+1];
2451 struct iovec iovec = {
2452 .iov_base = buf,
2453 .iov_len = sizeof(buf)-1,
2454 };
2455 CMSG_BUFFER_TYPE(CMSG_SPACE(sizeof(struct ucred)) +
2456 CMSG_SPACE(sizeof(int) * NOTIFY_FD_MAX)) control;
2457 struct msghdr msghdr = {
2458 .msg_iov = &iovec,
2459 .msg_iovlen = 1,
2460 .msg_control = &control,
2461 .msg_controllen = sizeof(control),
2462 };
2463
2464 struct cmsghdr *cmsg;
2465 struct ucred *ucred = NULL;
2466 _cleanup_free_ Unit **array_copy = NULL;
2467 _cleanup_strv_free_ char **tags = NULL;
2468 Unit *u1, *u2, **array;
2469 int r, *fd_array = NULL;
2470 size_t n_fds = 0;
2471 bool found = false;
2472 ssize_t n;
2473
2474 assert(m->notify_fd == fd);
2475
2476 if (revents != EPOLLIN) {
2477 log_warning("Got unexpected poll event for notify fd.");
2478 return 0;
2479 }
2480
2481 n = recvmsg_safe(m->notify_fd, &msghdr, MSG_DONTWAIT|MSG_CMSG_CLOEXEC|MSG_TRUNC);
2482 if (n < 0) {
2483 if (ERRNO_IS_TRANSIENT(n))
2484 return 0; /* Spurious wakeup, try again */
2485 if (n == -EXFULL) {
2486 log_warning("Got message with truncated control data (too many fds sent?), ignoring.");
2487 return 0;
2488 }
2489 /* If this is any other, real error, then let's stop processing this socket. This of course
2490 * means we won't take notification messages anymore, but that's still better than busy
2491 * looping around this: being woken up over and over again but being unable to actually read
2492 * the message off the socket. */
2493 return log_error_errno(n, "Failed to receive notification message: %m");
2494 }
2495
2496 CMSG_FOREACH(cmsg, &msghdr) {
2497 if (cmsg->cmsg_level == SOL_SOCKET && cmsg->cmsg_type == SCM_RIGHTS) {
2498
2499 assert(!fd_array);
2500 fd_array = (int*) CMSG_DATA(cmsg);
2501 n_fds = (cmsg->cmsg_len - CMSG_LEN(0)) / sizeof(int);
2502
2503 } else if (cmsg->cmsg_level == SOL_SOCKET &&
2504 cmsg->cmsg_type == SCM_CREDENTIALS &&
2505 cmsg->cmsg_len == CMSG_LEN(sizeof(struct ucred))) {
2506
2507 assert(!ucred);
2508 ucred = (struct ucred*) CMSG_DATA(cmsg);
2509 }
2510 }
2511
2512 if (n_fds > 0) {
2513 assert(fd_array);
2514
2515 r = fdset_new_array(&fds, fd_array, n_fds);
2516 if (r < 0) {
2517 close_many(fd_array, n_fds);
2518 log_oom();
2519 return 0;
2520 }
2521 }
2522
2523 if (!ucred || !pid_is_valid(ucred->pid)) {
2524 log_warning("Received notify message without valid credentials. Ignoring.");
2525 return 0;
2526 }
2527
2528 if ((size_t) n >= sizeof(buf) || (msghdr.msg_flags & MSG_TRUNC)) {
2529 log_warning("Received notify message exceeded maximum size. Ignoring.");
2530 return 0;
2531 }
2532
2533 /* As extra safety check, let's make sure the string we get doesn't contain embedded NUL bytes.
2534 * We permit one trailing NUL byte in the message, but don't expect it. */
2535 if (n > 1 && memchr(buf, 0, n-1)) {
2536 log_warning("Received notify message with embedded NUL bytes. Ignoring.");
2537 return 0;
2538 }
2539
2540 /* Make sure it's NUL-terminated, then parse it to obtain the tags list. */
2541 buf[n] = 0;
2542 tags = strv_split_newlines(buf);
2543 if (!tags) {
2544 log_oom();
2545 return 0;
2546 }
2547
2548 /* Possibly a barrier fd, let's see. */
2549 if (manager_process_barrier_fd(tags, fds))
2550 return 0;
2551
2552 /* Increase the generation counter used for filtering out duplicate unit invocations. */
2553 m->notifygen++;
2554
2555 /* Notify every unit that might be interested, which might be multiple. */
2556 u1 = manager_get_unit_by_pid_cgroup(m, ucred->pid);
2557 u2 = hashmap_get(m->watch_pids, PID_TO_PTR(ucred->pid));
2558 array = hashmap_get(m->watch_pids, PID_TO_PTR(-ucred->pid));
2559 if (array) {
2560 size_t k = 0;
2561
2562 while (array[k])
2563 k++;
2564
2565 array_copy = newdup(Unit*, array, k+1);
2566 if (!array_copy)
2567 log_oom();
2568 }
2569 /* And now invoke the per-unit callbacks. Note that manager_invoke_notify_message() will handle
2570 * duplicate units make sure we only invoke each unit's handler once. */
2571 if (u1) {
2572 manager_invoke_notify_message(m, u1, ucred, tags, fds);
2573 found = true;
2574 }
2575 if (u2) {
2576 manager_invoke_notify_message(m, u2, ucred, tags, fds);
2577 found = true;
2578 }
2579 if (array_copy)
2580 for (size_t i = 0; array_copy[i]; i++) {
2581 manager_invoke_notify_message(m, array_copy[i], ucred, tags, fds);
2582 found = true;
2583 }
2584
2585 if (!found)
2586 log_warning("Cannot find unit for notify message of PID "PID_FMT", ignoring.", ucred->pid);
2587
2588 if (fdset_size(fds) > 0)
2589 log_warning("Got extra auxiliary fds with notification message, closing them.");
2590
2591 return 0;
2592 }
2593
2594 static void manager_invoke_sigchld_event(
2595 Manager *m,
2596 Unit *u,
2597 const siginfo_t *si) {
2598
2599 assert(m);
2600 assert(u);
2601 assert(si);
2602
2603 /* Already invoked the handler of this unit in this iteration? Then don't process this again */
2604 if (u->sigchldgen == m->sigchldgen)
2605 return;
2606 u->sigchldgen = m->sigchldgen;
2607
2608 log_unit_debug(u, "Child "PID_FMT" belongs to %s.", si->si_pid, u->id);
2609 unit_unwatch_pid(u, si->si_pid);
2610
2611 if (UNIT_VTABLE(u)->sigchld_event)
2612 UNIT_VTABLE(u)->sigchld_event(u, si->si_pid, si->si_code, si->si_status);
2613 }
2614
2615 static int manager_dispatch_sigchld(sd_event_source *source, void *userdata) {
2616 Manager *m = ASSERT_PTR(userdata);
2617 siginfo_t si = {};
2618 int r;
2619
2620 assert(source);
2621
2622 /* First we call waitid() for a PID and do not reap the zombie. That way we can still access
2623 * /proc/$PID for it while it is a zombie. */
2624
2625 if (waitid(P_ALL, 0, &si, WEXITED|WNOHANG|WNOWAIT) < 0) {
2626
2627 if (errno != ECHILD)
2628 log_error_errno(errno, "Failed to peek for child with waitid(), ignoring: %m");
2629
2630 goto turn_off;
2631 }
2632
2633 if (si.si_pid <= 0)
2634 goto turn_off;
2635
2636 if (IN_SET(si.si_code, CLD_EXITED, CLD_KILLED, CLD_DUMPED)) {
2637 _cleanup_free_ Unit **array_copy = NULL;
2638 _cleanup_free_ char *name = NULL;
2639 Unit *u1, *u2, **array;
2640
2641 (void) get_process_comm(si.si_pid, &name);
2642
2643 log_debug("Child "PID_FMT" (%s) died (code=%s, status=%i/%s)",
2644 si.si_pid, strna(name),
2645 sigchld_code_to_string(si.si_code),
2646 si.si_status,
2647 strna(si.si_code == CLD_EXITED
2648 ? exit_status_to_string(si.si_status, EXIT_STATUS_FULL)
2649 : signal_to_string(si.si_status)));
2650
2651 /* Increase the generation counter used for filtering out duplicate unit invocations */
2652 m->sigchldgen++;
2653
2654 /* And now figure out the unit this belongs to, it might be multiple... */
2655 u1 = manager_get_unit_by_pid_cgroup(m, si.si_pid);
2656 u2 = hashmap_get(m->watch_pids, PID_TO_PTR(si.si_pid));
2657 array = hashmap_get(m->watch_pids, PID_TO_PTR(-si.si_pid));
2658 if (array) {
2659 size_t n = 0;
2660
2661 /* Count how many entries the array has */
2662 while (array[n])
2663 n++;
2664
2665 /* Make a copy of the array so that we don't trip up on the array changing beneath us */
2666 array_copy = newdup(Unit*, array, n+1);
2667 if (!array_copy)
2668 log_oom();
2669 }
2670
2671 /* Finally, execute them all. Note that u1, u2 and the array might contain duplicates, but
2672 * that's fine, manager_invoke_sigchld_event() will ensure we only invoke the handlers once for
2673 * each iteration. */
2674 if (u1) {
2675 /* We check for oom condition, in case we got SIGCHLD before the oom notification.
2676 * We only do this for the cgroup the PID belonged to. */
2677 (void) unit_check_oom(u1);
2678
2679 /* We check if systemd-oomd performed a kill so that we log and notify appropriately */
2680 (void) unit_check_oomd_kill(u1);
2681
2682 manager_invoke_sigchld_event(m, u1, &si);
2683 }
2684 if (u2)
2685 manager_invoke_sigchld_event(m, u2, &si);
2686 if (array_copy)
2687 for (size_t i = 0; array_copy[i]; i++)
2688 manager_invoke_sigchld_event(m, array_copy[i], &si);
2689 }
2690
2691 /* And now, we actually reap the zombie. */
2692 if (waitid(P_PID, si.si_pid, &si, WEXITED) < 0) {
2693 log_error_errno(errno, "Failed to dequeue child, ignoring: %m");
2694 return 0;
2695 }
2696
2697 return 0;
2698
2699 turn_off:
2700 /* All children processed for now, turn off event source */
2701
2702 r = sd_event_source_set_enabled(m->sigchld_event_source, SD_EVENT_OFF);
2703 if (r < 0)
2704 return log_error_errno(r, "Failed to disable SIGCHLD event source: %m");
2705
2706 return 0;
2707 }
2708
2709 static void manager_start_special(Manager *m, const char *name, JobMode mode) {
2710 Job *job;
2711
2712 if (manager_add_job_by_name_and_warn(m, JOB_START, name, mode, NULL, &job) < 0)
2713 return;
2714
2715 const char *s = unit_status_string(job->unit, NULL);
2716
2717 log_info("Activating special unit %s...", s);
2718
2719 sd_notifyf(false,
2720 "STATUS=Activating special unit %s...", s);
2721 m->status_ready = false;
2722 }
2723
2724 static void manager_handle_ctrl_alt_del(Manager *m) {
2725 /* If the user presses C-A-D more than
2726 * 7 times within 2s, we reboot/shutdown immediately,
2727 * unless it was disabled in system.conf */
2728
2729 if (ratelimit_below(&m->ctrl_alt_del_ratelimit) || m->cad_burst_action == EMERGENCY_ACTION_NONE)
2730 manager_start_special(m, SPECIAL_CTRL_ALT_DEL_TARGET, JOB_REPLACE_IRREVERSIBLY);
2731 else
2732 emergency_action(m, m->cad_burst_action, EMERGENCY_ACTION_WARN, NULL, -1,
2733 "Ctrl-Alt-Del was pressed more than 7 times within 2s");
2734 }
2735
2736 static int manager_dispatch_signal_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
2737 Manager *m = ASSERT_PTR(userdata);
2738 ssize_t n;
2739 struct signalfd_siginfo sfsi;
2740 int r;
2741
2742 assert(m->signal_fd == fd);
2743
2744 if (revents != EPOLLIN) {
2745 log_warning("Got unexpected events from signal file descriptor.");
2746 return 0;
2747 }
2748
2749 n = read(m->signal_fd, &sfsi, sizeof(sfsi));
2750 if (n < 0) {
2751 if (ERRNO_IS_TRANSIENT(errno))
2752 return 0;
2753
2754 /* We return an error here, which will kill this handler,
2755 * to avoid a busy loop on read error. */
2756 return log_error_errno(errno, "Reading from signal fd failed: %m");
2757 }
2758 if (n != sizeof(sfsi)) {
2759 log_warning("Truncated read from signal fd (%zi bytes), ignoring!", n);
2760 return 0;
2761 }
2762
2763 log_received_signal(sfsi.ssi_signo == SIGCHLD ||
2764 (sfsi.ssi_signo == SIGTERM && MANAGER_IS_USER(m))
2765 ? LOG_DEBUG : LOG_INFO,
2766 &sfsi);
2767
2768 switch (sfsi.ssi_signo) {
2769
2770 case SIGCHLD:
2771 r = sd_event_source_set_enabled(m->sigchld_event_source, SD_EVENT_ON);
2772 if (r < 0)
2773 log_warning_errno(r, "Failed to enable SIGCHLD event source, ignoring: %m");
2774
2775 break;
2776
2777 case SIGTERM:
2778 if (MANAGER_IS_SYSTEM(m)) {
2779 /* This is for compatibility with the original sysvinit */
2780 if (verify_run_space_and_log("Refusing to reexecute") < 0)
2781 break;
2782
2783 m->objective = MANAGER_REEXECUTE;
2784 break;
2785 }
2786
2787 _fallthrough_;
2788 case SIGINT:
2789 if (MANAGER_IS_SYSTEM(m))
2790 manager_handle_ctrl_alt_del(m);
2791 else
2792 manager_start_special(m, SPECIAL_EXIT_TARGET, JOB_REPLACE_IRREVERSIBLY);
2793 break;
2794
2795 case SIGWINCH:
2796 /* This is a nop on non-init */
2797 if (MANAGER_IS_SYSTEM(m))
2798 manager_start_special(m, SPECIAL_KBREQUEST_TARGET, JOB_REPLACE);
2799
2800 break;
2801
2802 case SIGPWR:
2803 /* This is a nop on non-init */
2804 if (MANAGER_IS_SYSTEM(m))
2805 manager_start_special(m, SPECIAL_SIGPWR_TARGET, JOB_REPLACE);
2806
2807 break;
2808
2809 case SIGUSR1:
2810 if (manager_dbus_is_running(m, false)) {
2811 log_info("Trying to reconnect to bus...");
2812
2813 (void) bus_init_api(m);
2814
2815 if (MANAGER_IS_SYSTEM(m))
2816 (void) bus_init_system(m);
2817 } else
2818 manager_start_special(m, SPECIAL_DBUS_SERVICE, JOB_REPLACE);
2819
2820 break;
2821
2822 case SIGUSR2: {
2823 _cleanup_free_ char *dump = NULL;
2824
2825 r = manager_get_dump_string(m, /* patterns= */ NULL, &dump);
2826 if (r < 0) {
2827 log_warning_errno(errno, "Failed to acquire manager dump: %m");
2828 break;
2829 }
2830
2831 log_dump(LOG_INFO, dump);
2832 break;
2833 }
2834
2835 case SIGHUP:
2836 if (verify_run_space_and_log("Refusing to reload") < 0)
2837 break;
2838
2839 m->objective = MANAGER_RELOAD;
2840 break;
2841
2842 default: {
2843
2844 /* Starting SIGRTMIN+0 */
2845 static const struct {
2846 const char *target;
2847 JobMode mode;
2848 } target_table[] = {
2849 [0] = { SPECIAL_DEFAULT_TARGET, JOB_ISOLATE },
2850 [1] = { SPECIAL_RESCUE_TARGET, JOB_ISOLATE },
2851 [2] = { SPECIAL_EMERGENCY_TARGET, JOB_ISOLATE },
2852 [3] = { SPECIAL_HALT_TARGET, JOB_REPLACE_IRREVERSIBLY },
2853 [4] = { SPECIAL_POWEROFF_TARGET, JOB_REPLACE_IRREVERSIBLY },
2854 [5] = { SPECIAL_REBOOT_TARGET, JOB_REPLACE_IRREVERSIBLY },
2855 [6] = { SPECIAL_KEXEC_TARGET, JOB_REPLACE_IRREVERSIBLY },
2856 };
2857
2858 /* Starting SIGRTMIN+13, so that target halt and system halt are 10 apart */
2859 static const ManagerObjective objective_table[] = {
2860 [0] = MANAGER_HALT,
2861 [1] = MANAGER_POWEROFF,
2862 [2] = MANAGER_REBOOT,
2863 [3] = MANAGER_KEXEC,
2864 };
2865
2866 if ((int) sfsi.ssi_signo >= SIGRTMIN+0 &&
2867 (int) sfsi.ssi_signo < SIGRTMIN+(int) ELEMENTSOF(target_table)) {
2868 int idx = (int) sfsi.ssi_signo - SIGRTMIN;
2869 manager_start_special(m, target_table[idx].target, target_table[idx].mode);
2870 break;
2871 }
2872
2873 if ((int) sfsi.ssi_signo >= SIGRTMIN+13 &&
2874 (int) sfsi.ssi_signo < SIGRTMIN+13+(int) ELEMENTSOF(objective_table)) {
2875 m->objective = objective_table[sfsi.ssi_signo - SIGRTMIN - 13];
2876 break;
2877 }
2878
2879 switch (sfsi.ssi_signo - SIGRTMIN) {
2880
2881 case 20:
2882 manager_override_show_status(m, SHOW_STATUS_YES, "signal");
2883 break;
2884
2885 case 21:
2886 manager_override_show_status(m, SHOW_STATUS_NO, "signal");
2887 break;
2888
2889 case 22:
2890 manager_override_log_level(m, LOG_DEBUG);
2891 break;
2892
2893 case 23:
2894 manager_restore_original_log_level(m);
2895 break;
2896
2897 case 24:
2898 if (MANAGER_IS_USER(m)) {
2899 m->objective = MANAGER_EXIT;
2900 return 0;
2901 }
2902
2903 /* This is a nop on init */
2904 break;
2905
2906 case 25:
2907 m->objective = MANAGER_REEXECUTE;
2908 break;
2909
2910 case 26:
2911 case 29: /* compatibility: used to be mapped to LOG_TARGET_SYSLOG_OR_KMSG */
2912 manager_restore_original_log_target(m);
2913 break;
2914
2915 case 27:
2916 manager_override_log_target(m, LOG_TARGET_CONSOLE);
2917 break;
2918
2919 case 28:
2920 manager_override_log_target(m, LOG_TARGET_KMSG);
2921 break;
2922
2923 default:
2924 log_warning("Got unhandled signal <%s>.", signal_to_string(sfsi.ssi_signo));
2925 }
2926 }}
2927
2928 return 0;
2929 }
2930
2931 static int manager_dispatch_time_change_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
2932 Manager *m = ASSERT_PTR(userdata);
2933 Unit *u;
2934
2935 log_struct(LOG_DEBUG,
2936 "MESSAGE_ID=" SD_MESSAGE_TIME_CHANGE_STR,
2937 LOG_MESSAGE("Time has been changed"));
2938
2939 /* Restart the watch */
2940 (void) manager_setup_time_change(m);
2941
2942 HASHMAP_FOREACH(u, m->units)
2943 if (UNIT_VTABLE(u)->time_change)
2944 UNIT_VTABLE(u)->time_change(u);
2945
2946 return 0;
2947 }
2948
2949 static int manager_dispatch_timezone_change(
2950 sd_event_source *source,
2951 const struct inotify_event *e,
2952 void *userdata) {
2953
2954 Manager *m = ASSERT_PTR(userdata);
2955 int changed;
2956 Unit *u;
2957
2958 log_debug("inotify event for /etc/localtime");
2959
2960 changed = manager_read_timezone_stat(m);
2961 if (changed <= 0)
2962 return changed;
2963
2964 /* Something changed, restart the watch, to ensure we watch the new /etc/localtime if it changed */
2965 (void) manager_setup_timezone_change(m);
2966
2967 /* Read the new timezone */
2968 tzset();
2969
2970 log_debug("Timezone has been changed (now: %s).", tzname[daylight]);
2971
2972 HASHMAP_FOREACH(u, m->units)
2973 if (UNIT_VTABLE(u)->timezone_change)
2974 UNIT_VTABLE(u)->timezone_change(u);
2975
2976 return 0;
2977 }
2978
2979 static int manager_dispatch_idle_pipe_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
2980 Manager *m = ASSERT_PTR(userdata);
2981
2982 assert(m->idle_pipe[2] == fd);
2983
2984 /* There's at least one Type=idle child that just gave up on us waiting for the boot process to
2985 * complete. Let's now turn off any further console output if there's at least one service that needs
2986 * console access, so that from now on our own output should not spill into that service's output
2987 * anymore. After all, we support Type=idle only to beautify console output and it generally is set
2988 * on services that want to own the console exclusively without our interference. */
2989 m->no_console_output = m->n_on_console > 0;
2990
2991 /* Acknowledge the child's request, and let all all other children know too that they shouldn't wait
2992 * any longer by closing the pipes towards them, which is what they are waiting for. */
2993 manager_close_idle_pipe(m);
2994
2995 return 0;
2996 }
2997
2998 static int manager_dispatch_jobs_in_progress(sd_event_source *source, usec_t usec, void *userdata) {
2999 Manager *m = ASSERT_PTR(userdata);
3000 int r;
3001
3002 assert(source);
3003
3004 manager_print_jobs_in_progress(m);
3005
3006 r = sd_event_source_set_time_relative(source, JOBS_IN_PROGRESS_PERIOD_USEC);
3007 if (r < 0)
3008 return r;
3009
3010 return sd_event_source_set_enabled(source, SD_EVENT_ONESHOT);
3011 }
3012
3013 int manager_loop(Manager *m) {
3014 RateLimit rl = { .interval = 1*USEC_PER_SEC, .burst = 50000 };
3015 int r;
3016
3017 assert(m);
3018 assert(m->objective == MANAGER_OK); /* Ensure manager_startup() has been called */
3019
3020 manager_check_finished(m);
3021
3022 /* There might still be some zombies hanging around from before we were exec()'ed. Let's reap them. */
3023 r = sd_event_source_set_enabled(m->sigchld_event_source, SD_EVENT_ON);
3024 if (r < 0)
3025 return log_error_errno(r, "Failed to enable SIGCHLD event source: %m");
3026
3027 while (m->objective == MANAGER_OK) {
3028
3029 (void) watchdog_ping();
3030
3031 if (!ratelimit_below(&rl)) {
3032 /* Yay, something is going seriously wrong, pause a little */
3033 log_warning("Looping too fast. Throttling execution a little.");
3034 sleep(1);
3035 }
3036
3037 if (manager_dispatch_load_queue(m) > 0)
3038 continue;
3039
3040 if (manager_dispatch_gc_job_queue(m) > 0)
3041 continue;
3042
3043 if (manager_dispatch_gc_unit_queue(m) > 0)
3044 continue;
3045
3046 if (manager_dispatch_cleanup_queue(m) > 0)
3047 continue;
3048
3049 if (manager_dispatch_cgroup_realize_queue(m) > 0)
3050 continue;
3051
3052 if (manager_dispatch_start_when_upheld_queue(m) > 0)
3053 continue;
3054
3055 if (manager_dispatch_stop_when_bound_queue(m) > 0)
3056 continue;
3057
3058 if (manager_dispatch_stop_when_unneeded_queue(m) > 0)
3059 continue;
3060
3061 if (manager_dispatch_dbus_queue(m) > 0)
3062 continue;
3063
3064 /* Sleep for watchdog runtime wait time */
3065 r = sd_event_run(m->event, watchdog_runtime_wait());
3066 if (r < 0)
3067 return log_error_errno(r, "Failed to run event loop: %m");
3068 }
3069
3070 return m->objective;
3071 }
3072
3073 int manager_load_unit_from_dbus_path(Manager *m, const char *s, sd_bus_error *e, Unit **_u) {
3074 _cleanup_free_ char *n = NULL;
3075 sd_id128_t invocation_id;
3076 Unit *u;
3077 int r;
3078
3079 assert(m);
3080 assert(s);
3081 assert(_u);
3082
3083 r = unit_name_from_dbus_path(s, &n);
3084 if (r < 0)
3085 return r;
3086
3087 /* Permit addressing units by invocation ID: if the passed bus path is suffixed by a 128bit ID then
3088 * we use it as invocation ID. */
3089 r = sd_id128_from_string(n, &invocation_id);
3090 if (r >= 0) {
3091 u = hashmap_get(m->units_by_invocation_id, &invocation_id);
3092 if (u) {
3093 *_u = u;
3094 return 0;
3095 }
3096
3097 return sd_bus_error_setf(e, BUS_ERROR_NO_UNIT_FOR_INVOCATION_ID,
3098 "No unit with the specified invocation ID " SD_ID128_FORMAT_STR " known.",
3099 SD_ID128_FORMAT_VAL(invocation_id));
3100 }
3101
3102 /* If this didn't work, we check if this is a unit name */
3103 if (!unit_name_is_valid(n, UNIT_NAME_PLAIN|UNIT_NAME_INSTANCE)) {
3104 _cleanup_free_ char *nn = NULL;
3105
3106 nn = cescape(n);
3107 return sd_bus_error_setf(e, SD_BUS_ERROR_INVALID_ARGS,
3108 "Unit name %s is neither a valid invocation ID nor unit name.", strnull(nn));
3109 }
3110
3111 r = manager_load_unit(m, n, NULL, e, &u);
3112 if (r < 0)
3113 return r;
3114
3115 *_u = u;
3116 return 0;
3117 }
3118
3119 int manager_get_job_from_dbus_path(Manager *m, const char *s, Job **_j) {
3120 const char *p;
3121 unsigned id;
3122 Job *j;
3123 int r;
3124
3125 assert(m);
3126 assert(s);
3127 assert(_j);
3128
3129 p = startswith(s, "/org/freedesktop/systemd1/job/");
3130 if (!p)
3131 return -EINVAL;
3132
3133 r = safe_atou(p, &id);
3134 if (r < 0)
3135 return r;
3136
3137 j = manager_get_job(m, id);
3138 if (!j)
3139 return -ENOENT;
3140
3141 *_j = j;
3142
3143 return 0;
3144 }
3145
3146 void manager_send_unit_audit(Manager *m, Unit *u, int type, bool success) {
3147
3148 #if HAVE_AUDIT
3149 _cleanup_free_ char *p = NULL;
3150 const char *msg;
3151 int audit_fd, r;
3152
3153 if (!MANAGER_IS_SYSTEM(m))
3154 return;
3155
3156 audit_fd = get_audit_fd();
3157 if (audit_fd < 0)
3158 return;
3159
3160 /* Don't generate audit events if the service was already
3161 * started and we're just deserializing */
3162 if (MANAGER_IS_RELOADING(m))
3163 return;
3164
3165 if (u->type != UNIT_SERVICE)
3166 return;
3167
3168 r = unit_name_to_prefix_and_instance(u->id, &p);
3169 if (r < 0) {
3170 log_error_errno(r, "Failed to extract prefix and instance of unit name: %m");
3171 return;
3172 }
3173
3174 msg = strjoina("unit=", p);
3175 if (audit_log_user_comm_message(audit_fd, type, msg, "systemd", NULL, NULL, NULL, success) < 0) {
3176 if (errno == EPERM)
3177 /* We aren't allowed to send audit messages?
3178 * Then let's not retry again. */
3179 close_audit_fd();
3180 else
3181 log_warning_errno(errno, "Failed to send audit message: %m");
3182 }
3183 #endif
3184
3185 }
3186
3187 void manager_send_unit_plymouth(Manager *m, Unit *u) {
3188 static const union sockaddr_union sa = PLYMOUTH_SOCKET;
3189 _cleanup_free_ char *message = NULL;
3190 _cleanup_close_ int fd = -EBADF;
3191 int n = 0;
3192
3193 /* Don't generate plymouth events if the service was already
3194 * started and we're just deserializing */
3195 if (MANAGER_IS_RELOADING(m))
3196 return;
3197
3198 if (!MANAGER_IS_SYSTEM(m))
3199 return;
3200
3201 if (detect_container() > 0)
3202 return;
3203
3204 if (!IN_SET(u->type, UNIT_SERVICE, UNIT_MOUNT, UNIT_SWAP))
3205 return;
3206
3207 /* We set SOCK_NONBLOCK here so that we rather drop the
3208 * message then wait for plymouth */
3209 fd = socket(AF_UNIX, SOCK_STREAM|SOCK_CLOEXEC|SOCK_NONBLOCK, 0);
3210 if (fd < 0) {
3211 log_error_errno(errno, "socket() failed: %m");
3212 return;
3213 }
3214
3215 if (connect(fd, &sa.sa, SOCKADDR_UN_LEN(sa.un)) < 0) {
3216 if (!IN_SET(errno, EAGAIN, ENOENT) && !ERRNO_IS_DISCONNECT(errno))
3217 log_error_errno(errno, "connect() failed: %m");
3218 return;
3219 }
3220
3221 if (asprintf(&message, "U\002%c%s%n", (int) (strlen(u->id) + 1), u->id, &n) < 0)
3222 return (void) log_oom();
3223
3224 errno = 0;
3225 if (write(fd, message, n + 1) != n + 1)
3226 if (!IN_SET(errno, EAGAIN, ENOENT) && !ERRNO_IS_DISCONNECT(errno))
3227 log_error_errno(errno, "Failed to write Plymouth message: %m");
3228 }
3229
3230 usec_t manager_get_watchdog(Manager *m, WatchdogType t) {
3231 assert(m);
3232
3233 if (MANAGER_IS_USER(m))
3234 return USEC_INFINITY;
3235
3236 if (timestamp_is_set(m->watchdog_overridden[t]))
3237 return m->watchdog_overridden[t];
3238
3239 return m->watchdog[t];
3240 }
3241
3242 void manager_set_watchdog(Manager *m, WatchdogType t, usec_t timeout) {
3243
3244 assert(m);
3245
3246 if (MANAGER_IS_USER(m))
3247 return;
3248
3249 if (m->watchdog[t] == timeout)
3250 return;
3251
3252 if (t == WATCHDOG_RUNTIME) {
3253 if (!timestamp_is_set(m->watchdog_overridden[WATCHDOG_RUNTIME]))
3254 (void) watchdog_setup(timeout);
3255 } else if (t == WATCHDOG_PRETIMEOUT)
3256 if (m->watchdog_overridden[WATCHDOG_PRETIMEOUT] == USEC_INFINITY)
3257 (void) watchdog_setup_pretimeout(timeout);
3258
3259 m->watchdog[t] = timeout;
3260 }
3261
3262 void manager_override_watchdog(Manager *m, WatchdogType t, usec_t timeout) {
3263
3264 assert(m);
3265
3266 if (MANAGER_IS_USER(m))
3267 return;
3268
3269 if (m->watchdog_overridden[t] == timeout)
3270 return;
3271
3272 if (t == WATCHDOG_RUNTIME) {
3273 usec_t usec = timestamp_is_set(timeout) ? timeout : m->watchdog[t];
3274
3275 (void) watchdog_setup(usec);
3276 } else if (t == WATCHDOG_PRETIMEOUT)
3277 (void) watchdog_setup_pretimeout(timeout);
3278
3279 m->watchdog_overridden[t] = timeout;
3280 }
3281
3282 int manager_set_watchdog_pretimeout_governor(Manager *m, const char *governor) {
3283 _cleanup_free_ char *p = NULL;
3284 int r;
3285
3286 assert(m);
3287
3288 if (MANAGER_IS_USER(m))
3289 return 0;
3290
3291 if (streq_ptr(m->watchdog_pretimeout_governor, governor))
3292 return 0;
3293
3294 p = strdup(governor);
3295 if (!p)
3296 return -ENOMEM;
3297
3298 r = watchdog_setup_pretimeout_governor(governor);
3299 if (r < 0)
3300 return r;
3301
3302 return free_and_replace(m->watchdog_pretimeout_governor, p);
3303 }
3304
3305 int manager_override_watchdog_pretimeout_governor(Manager *m, const char *governor) {
3306 _cleanup_free_ char *p = NULL;
3307 int r;
3308
3309 assert(m);
3310
3311 if (MANAGER_IS_USER(m))
3312 return 0;
3313
3314 if (streq_ptr(m->watchdog_pretimeout_governor_overridden, governor))
3315 return 0;
3316
3317 p = strdup(governor);
3318 if (!p)
3319 return -ENOMEM;
3320
3321 r = watchdog_setup_pretimeout_governor(governor);
3322 if (r < 0)
3323 return r;
3324
3325 return free_and_replace(m->watchdog_pretimeout_governor_overridden, p);
3326 }
3327
3328 int manager_reload(Manager *m) {
3329 _unused_ _cleanup_(manager_reloading_stopp) Manager *reloading = NULL;
3330 _cleanup_fdset_free_ FDSet *fds = NULL;
3331 _cleanup_fclose_ FILE *f = NULL;
3332 int r;
3333
3334 assert(m);
3335
3336 r = manager_open_serialization(m, &f);
3337 if (r < 0)
3338 return log_error_errno(r, "Failed to create serialization file: %m");
3339
3340 fds = fdset_new();
3341 if (!fds)
3342 return log_oom();
3343
3344 /* We are officially in reload mode from here on. */
3345 reloading = manager_reloading_start(m);
3346
3347 r = manager_serialize(m, f, fds, false);
3348 if (r < 0)
3349 return r;
3350
3351 if (fseeko(f, 0, SEEK_SET) < 0)
3352 return log_error_errno(errno, "Failed to seek to beginning of serialization: %m");
3353
3354 /* 💀 This is the point of no return, from here on there is no way back. 💀 */
3355 reloading = NULL;
3356
3357 bus_manager_send_reloading(m, true);
3358
3359 /* Start by flushing out all jobs and units, all generated units, all runtime environments, all dynamic users
3360 * and everything else that is worth flushing out. We'll get it all back from the serialization — if we need
3361 * it. */
3362
3363 manager_clear_jobs_and_units(m);
3364 lookup_paths_flush_generator(&m->lookup_paths);
3365 lookup_paths_free(&m->lookup_paths);
3366 exec_runtime_vacuum(m);
3367 dynamic_user_vacuum(m, false);
3368 m->uid_refs = hashmap_free(m->uid_refs);
3369 m->gid_refs = hashmap_free(m->gid_refs);
3370
3371 r = lookup_paths_init_or_warn(&m->lookup_paths, m->unit_file_scope, 0, NULL);
3372 if (r < 0)
3373 return r;
3374
3375 (void) manager_run_environment_generators(m);
3376 (void) manager_run_generators(m);
3377
3378 lookup_paths_log(&m->lookup_paths);
3379
3380 /* We flushed out generated files, for which we don't watch mtime, so we should flush the old map. */
3381 manager_free_unit_name_maps(m);
3382
3383 /* First, enumerate what we can from kernel and suchlike */
3384 manager_enumerate_perpetual(m);
3385 manager_enumerate(m);
3386
3387 /* Second, deserialize our stored data */
3388 r = manager_deserialize(m, f, fds);
3389 if (r < 0)
3390 log_warning_errno(r, "Deserialization failed, proceeding anyway: %m");
3391
3392 /* We don't need the serialization anymore */
3393 f = safe_fclose(f);
3394
3395 /* Re-register notify_fd as event source, and set up other sockets/communication channels we might need */
3396 (void) manager_setup_notify(m);
3397 (void) manager_setup_cgroups_agent(m);
3398 (void) manager_setup_user_lookup_fd(m);
3399
3400 /* Third, fire things up! */
3401 manager_coldplug(m);
3402
3403 /* Clean up runtime objects no longer referenced */
3404 manager_vacuum(m);
3405
3406 /* Clean up deserialized tracked clients */
3407 m->deserialized_subscribed = strv_free(m->deserialized_subscribed);
3408
3409 /* Consider the reload process complete now. */
3410 assert(m->n_reloading > 0);
3411 m->n_reloading--;
3412
3413 manager_ready(m);
3414
3415 m->send_reloading_done = true;
3416 return 0;
3417 }
3418
3419 void manager_reset_failed(Manager *m) {
3420 Unit *u;
3421
3422 assert(m);
3423
3424 HASHMAP_FOREACH(u, m->units)
3425 unit_reset_failed(u);
3426 }
3427
3428 bool manager_unit_inactive_or_pending(Manager *m, const char *name) {
3429 Unit *u;
3430
3431 assert(m);
3432 assert(name);
3433
3434 /* Returns true if the unit is inactive or going down */
3435 u = manager_get_unit(m, name);
3436 if (!u)
3437 return true;
3438
3439 return unit_inactive_or_pending(u);
3440 }
3441
3442 static void log_taint_string(Manager *m) {
3443 _cleanup_free_ char *taint = NULL;
3444
3445 assert(m);
3446
3447 if (MANAGER_IS_USER(m) || m->taint_logged)
3448 return;
3449
3450 m->taint_logged = true; /* only check for taint once */
3451
3452 taint = manager_taint_string(m);
3453 if (isempty(taint))
3454 return;
3455
3456 log_struct(LOG_NOTICE,
3457 LOG_MESSAGE("System is tainted: %s", taint),
3458 "TAINT=%s", taint,
3459 "MESSAGE_ID=" SD_MESSAGE_TAINTED_STR);
3460 }
3461
3462 static void manager_notify_finished(Manager *m) {
3463 usec_t firmware_usec, loader_usec, kernel_usec, initrd_usec, userspace_usec, total_usec;
3464
3465 if (MANAGER_IS_TEST_RUN(m))
3466 return;
3467
3468 if (MANAGER_IS_SYSTEM(m) && detect_container() <= 0) {
3469 char buf[FORMAT_TIMESPAN_MAX + STRLEN(" (firmware) + ") + FORMAT_TIMESPAN_MAX + STRLEN(" (loader) + ")]
3470 = {};
3471 char *p = buf;
3472 size_t size = sizeof buf;
3473
3474 /* Note that MANAGER_TIMESTAMP_KERNEL's monotonic value is always at 0, and
3475 * MANAGER_TIMESTAMP_FIRMWARE's and MANAGER_TIMESTAMP_LOADER's monotonic value should be considered
3476 * negative values. */
3477
3478 firmware_usec = m->timestamps[MANAGER_TIMESTAMP_FIRMWARE].monotonic - m->timestamps[MANAGER_TIMESTAMP_LOADER].monotonic;
3479 loader_usec = m->timestamps[MANAGER_TIMESTAMP_LOADER].monotonic - m->timestamps[MANAGER_TIMESTAMP_KERNEL].monotonic;
3480 userspace_usec = m->timestamps[MANAGER_TIMESTAMP_FINISH].monotonic - m->timestamps[MANAGER_TIMESTAMP_USERSPACE].monotonic;
3481 total_usec = m->timestamps[MANAGER_TIMESTAMP_FIRMWARE].monotonic + m->timestamps[MANAGER_TIMESTAMP_FINISH].monotonic;
3482
3483 if (firmware_usec > 0)
3484 size = strpcpyf(&p, size, "%s (firmware) + ", FORMAT_TIMESPAN(firmware_usec, USEC_PER_MSEC));
3485 if (loader_usec > 0)
3486 size = strpcpyf(&p, size, "%s (loader) + ", FORMAT_TIMESPAN(loader_usec, USEC_PER_MSEC));
3487
3488 if (dual_timestamp_is_set(&m->timestamps[MANAGER_TIMESTAMP_INITRD])) {
3489
3490 /* The initrd case on bare-metal */
3491 kernel_usec = m->timestamps[MANAGER_TIMESTAMP_INITRD].monotonic - m->timestamps[MANAGER_TIMESTAMP_KERNEL].monotonic;
3492 initrd_usec = m->timestamps[MANAGER_TIMESTAMP_USERSPACE].monotonic - m->timestamps[MANAGER_TIMESTAMP_INITRD].monotonic;
3493
3494 log_struct(LOG_INFO,
3495 "MESSAGE_ID=" SD_MESSAGE_STARTUP_FINISHED_STR,
3496 "KERNEL_USEC="USEC_FMT, kernel_usec,
3497 "INITRD_USEC="USEC_FMT, initrd_usec,
3498 "USERSPACE_USEC="USEC_FMT, userspace_usec,
3499 LOG_MESSAGE("Startup finished in %s%s (kernel) + %s (initrd) + %s (userspace) = %s.",
3500 buf,
3501 FORMAT_TIMESPAN(kernel_usec, USEC_PER_MSEC),
3502 FORMAT_TIMESPAN(initrd_usec, USEC_PER_MSEC),
3503 FORMAT_TIMESPAN(userspace_usec, USEC_PER_MSEC),
3504 FORMAT_TIMESPAN(total_usec, USEC_PER_MSEC)));
3505 } else {
3506 /* The initrd-less case on bare-metal */
3507
3508 kernel_usec = m->timestamps[MANAGER_TIMESTAMP_USERSPACE].monotonic - m->timestamps[MANAGER_TIMESTAMP_KERNEL].monotonic;
3509 initrd_usec = 0;
3510
3511 log_struct(LOG_INFO,
3512 "MESSAGE_ID=" SD_MESSAGE_STARTUP_FINISHED_STR,
3513 "KERNEL_USEC="USEC_FMT, kernel_usec,
3514 "USERSPACE_USEC="USEC_FMT, userspace_usec,
3515 LOG_MESSAGE("Startup finished in %s%s (kernel) + %s (userspace) = %s.",
3516 buf,
3517 FORMAT_TIMESPAN(kernel_usec, USEC_PER_MSEC),
3518 FORMAT_TIMESPAN(userspace_usec, USEC_PER_MSEC),
3519 FORMAT_TIMESPAN(total_usec, USEC_PER_MSEC)));
3520 }
3521 } else {
3522 /* The container and --user case */
3523 firmware_usec = loader_usec = initrd_usec = kernel_usec = 0;
3524 total_usec = userspace_usec = m->timestamps[MANAGER_TIMESTAMP_FINISH].monotonic - m->timestamps[MANAGER_TIMESTAMP_USERSPACE].monotonic;
3525
3526 log_struct(LOG_INFO,
3527 "MESSAGE_ID=" SD_MESSAGE_USER_STARTUP_FINISHED_STR,
3528 "USERSPACE_USEC="USEC_FMT, userspace_usec,
3529 LOG_MESSAGE("Startup finished in %s.",
3530 FORMAT_TIMESPAN(total_usec, USEC_PER_MSEC)));
3531 }
3532
3533 bus_manager_send_finished(m, firmware_usec, loader_usec, kernel_usec, initrd_usec, userspace_usec, total_usec);
3534
3535 log_taint_string(m);
3536 }
3537
3538 static void user_manager_send_ready(Manager *m) {
3539 int r;
3540
3541 assert(m);
3542
3543 /* We send READY=1 on reaching basic.target only when running in --user mode. */
3544 if (!MANAGER_IS_USER(m) || m->ready_sent)
3545 return;
3546
3547 r = sd_notify(false,
3548 "READY=1\n"
3549 "STATUS=Reached " SPECIAL_BASIC_TARGET ".");
3550 if (r < 0)
3551 log_warning_errno(r, "Failed to send readiness notification, ignoring: %m");
3552
3553 m->ready_sent = true;
3554 m->status_ready = false;
3555 }
3556
3557 static void manager_send_ready(Manager *m) {
3558 int r;
3559
3560 if (m->ready_sent && m->status_ready)
3561 /* Skip the notification if nothing changed. */
3562 return;
3563
3564 r = sd_notify(false,
3565 "READY=1\n"
3566 "STATUS=Ready.");
3567 if (r < 0)
3568 log_full_errno(m->ready_sent ? LOG_DEBUG : LOG_WARNING, r,
3569 "Failed to send readiness notification, ignoring: %m");
3570
3571 m->ready_sent = m->status_ready = true;
3572 }
3573
3574 static void manager_check_basic_target(Manager *m) {
3575 Unit *u;
3576
3577 assert(m);
3578
3579 /* Small shortcut */
3580 if (m->ready_sent && m->taint_logged)
3581 return;
3582
3583 u = manager_get_unit(m, SPECIAL_BASIC_TARGET);
3584 if (!u || !UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u)))
3585 return;
3586
3587 /* For user managers, send out READY=1 as soon as we reach basic.target */
3588 user_manager_send_ready(m);
3589
3590 /* Log the taint string as soon as we reach basic.target */
3591 log_taint_string(m);
3592 }
3593
3594 void manager_check_finished(Manager *m) {
3595 assert(m);
3596
3597 if (MANAGER_IS_RELOADING(m))
3598 return;
3599
3600 /* Verify that we have entered the event loop already, and not left it again. */
3601 if (!MANAGER_IS_RUNNING(m))
3602 return;
3603
3604 manager_check_basic_target(m);
3605
3606 if (hashmap_size(m->jobs) > 0) {
3607 if (m->jobs_in_progress_event_source)
3608 /* Ignore any failure, this is only for feedback */
3609 (void) sd_event_source_set_time(m->jobs_in_progress_event_source,
3610 manager_watch_jobs_next_time(m));
3611 return;
3612 }
3613
3614 /* The jobs hashmap tends to grow a lot during boot, and then it's not reused until shutdown. Let's
3615 kill the hashmap if it is relatively large. */
3616 if (hashmap_buckets(m->jobs) > hashmap_size(m->units) / 10)
3617 m->jobs = hashmap_free(m->jobs);
3618
3619 manager_send_ready(m);
3620
3621 /* Notify Type=idle units that we are done now */
3622 manager_close_idle_pipe(m);
3623
3624 if (MANAGER_IS_FINISHED(m))
3625 return;
3626
3627 manager_flip_auto_status(m, false, "boot finished");
3628
3629 /* Turn off confirm spawn now */
3630 m->confirm_spawn = NULL;
3631
3632 /* No need to update ask password status when we're going non-interactive */
3633 manager_close_ask_password(m);
3634
3635 /* This is no longer the first boot */
3636 manager_set_first_boot(m, false);
3637
3638 dual_timestamp_get(m->timestamps + MANAGER_TIMESTAMP_FINISH);
3639
3640 manager_notify_finished(m);
3641
3642 manager_invalidate_startup_units(m);
3643 }
3644
3645 static bool generator_path_any(const char* const* paths) {
3646 bool found = false;
3647
3648 /* Optimize by skipping the whole process by not creating output directories
3649 * if no generators are found. */
3650 STRV_FOREACH(path, paths)
3651 if (access(*path, F_OK) == 0)
3652 found = true;
3653 else if (errno != ENOENT)
3654 log_warning_errno(errno, "Failed to open generator directory %s: %m", *path);
3655
3656 return found;
3657 }
3658
3659 static int manager_run_environment_generators(Manager *m) {
3660 char **tmp = NULL; /* this is only used in the forked process, no cleanup here */
3661 _cleanup_strv_free_ char **paths = NULL;
3662 void* args[] = {
3663 [STDOUT_GENERATE] = &tmp,
3664 [STDOUT_COLLECT] = &tmp,
3665 [STDOUT_CONSUME] = &m->transient_environment,
3666 };
3667 int r;
3668
3669 if (MANAGER_IS_TEST_RUN(m) && !(m->test_run_flags & MANAGER_TEST_RUN_ENV_GENERATORS))
3670 return 0;
3671
3672 paths = env_generator_binary_paths(MANAGER_IS_SYSTEM(m));
3673 if (!paths)
3674 return log_oom();
3675
3676 if (!generator_path_any((const char* const*) paths))
3677 return 0;
3678
3679 WITH_UMASK(0022)
3680 r = execute_directories((const char* const*) paths, DEFAULT_TIMEOUT_USEC, gather_environment,
3681 args, NULL, m->transient_environment,
3682 EXEC_DIR_PARALLEL | EXEC_DIR_IGNORE_ERRORS | EXEC_DIR_SET_SYSTEMD_EXEC_PID);
3683 return r;
3684 }
3685
3686 static int build_generator_environment(Manager *m, char ***ret) {
3687 _cleanup_strv_free_ char **nl = NULL;
3688 Virtualization v;
3689 int r;
3690
3691 assert(m);
3692 assert(ret);
3693
3694 /* Generators oftentimes want to know some basic facts about the environment they run in, in order to
3695 * adjust generated units to that. Let's pass down some bits of information that are easy for us to
3696 * determine (but a bit harder for generator scripts to determine), as environment variables. */
3697
3698 nl = strv_copy(m->transient_environment);
3699 if (!nl)
3700 return -ENOMEM;
3701
3702 r = strv_env_assign(&nl, "SYSTEMD_SCOPE", MANAGER_IS_SYSTEM(m) ? "system" : "user");
3703 if (r < 0)
3704 return r;
3705
3706 if (MANAGER_IS_SYSTEM(m)) {
3707 /* Note that $SYSTEMD_IN_INITRD may be used to override the initrd detection in much of our
3708 * codebase. This is hence more than purely informational. It will shortcut detection of the
3709 * initrd state if generators invoke our own tools. But that's OK, as it would come to the
3710 * same results (hopefully). */
3711 r = strv_env_assign(&nl, "SYSTEMD_IN_INITRD", one_zero(in_initrd()));
3712 if (r < 0)
3713 return r;
3714
3715 if (m->first_boot >= 0) {
3716 r = strv_env_assign(&nl, "SYSTEMD_FIRST_BOOT", one_zero(m->first_boot));
3717 if (r < 0)
3718 return r;
3719 }
3720 }
3721
3722 v = detect_virtualization();
3723 if (v < 0)
3724 log_debug_errno(v, "Failed to detect virtualization, ignoring: %m");
3725 else if (v > 0) {
3726 const char *s;
3727
3728 s = strjoina(VIRTUALIZATION_IS_VM(v) ? "vm:" :
3729 VIRTUALIZATION_IS_CONTAINER(v) ? "container:" : ":",
3730 virtualization_to_string(v));
3731
3732 r = strv_env_assign(&nl, "SYSTEMD_VIRTUALIZATION", s);
3733 if (r < 0)
3734 return r;
3735 }
3736
3737 r = strv_env_assign(&nl, "SYSTEMD_ARCHITECTURE", architecture_to_string(uname_architecture()));
3738 if (r < 0)
3739 return r;
3740
3741 *ret = TAKE_PTR(nl);
3742 return 0;
3743 }
3744
3745 static int manager_execute_generators(Manager *m, char **paths, bool remount_ro) {
3746 _cleanup_strv_free_ char **ge = NULL;
3747 const char *argv[] = {
3748 NULL, /* Leave this empty, execute_directory() will fill something in */
3749 m->lookup_paths.generator,
3750 m->lookup_paths.generator_early,
3751 m->lookup_paths.generator_late,
3752 NULL,
3753 };
3754 int r;
3755
3756 r = build_generator_environment(m, &ge);
3757 if (r < 0)
3758 return log_error_errno(r, "Failed to build generator environment: %m");
3759
3760 if (remount_ro) {
3761 /* Remount most of the filesystem tree read-only. We leave /sys/ as-is, because our code
3762 * checks whether it is read-only to detect containerized execution environments. We leave
3763 * /run/ as-is too, because that's where our output goes. We also leave /proc/ and /dev/shm/
3764 * because they're API, and /tmp/ that safe_fork() mounted for us.
3765 */
3766 r = bind_remount_recursive("/", MS_RDONLY, MS_RDONLY,
3767 STRV_MAKE("/sys", "/run", "/proc", "/dev/shm", "/tmp"));
3768 if (r < 0)
3769 log_warning_errno(r, "Read-only bind remount failed, ignoring: %m");
3770 }
3771
3772 BLOCK_WITH_UMASK(0022);
3773 return execute_directories(
3774 (const char* const*) paths,
3775 DEFAULT_TIMEOUT_USEC,
3776 /* callbacks= */ NULL, /* callback_args= */ NULL,
3777 (char**) argv,
3778 ge,
3779 EXEC_DIR_PARALLEL | EXEC_DIR_IGNORE_ERRORS | EXEC_DIR_SET_SYSTEMD_EXEC_PID);
3780 }
3781
3782 static int manager_run_generators(Manager *m) {
3783 _cleanup_strv_free_ char **paths = NULL;
3784 int r;
3785
3786 assert(m);
3787
3788 if (MANAGER_IS_TEST_RUN(m) && !(m->test_run_flags & MANAGER_TEST_RUN_GENERATORS))
3789 return 0;
3790
3791 paths = generator_binary_paths(m->unit_file_scope);
3792 if (!paths)
3793 return log_oom();
3794
3795 if (!generator_path_any((const char* const*) paths))
3796 return 0;
3797
3798 r = lookup_paths_mkdir_generator(&m->lookup_paths);
3799 if (r < 0) {
3800 log_error_errno(r, "Failed to create generator directories: %m");
3801 goto finish;
3802 }
3803
3804 /* If we are the system manager, we fork and invoke the generators in a sanitized mount namespace. If
3805 * we are the user manager, let's just execute the generators directly. We might not have the
3806 * necessary privileges, and the system manager has already mounted /tmp/ and everything else for us.
3807 */
3808 if (MANAGER_IS_USER(m)) {
3809 r = manager_execute_generators(m, paths, /* remount_ro= */ false);
3810 goto finish;
3811 }
3812
3813 r = safe_fork("(sd-gens)",
3814 FORK_RESET_SIGNALS | FORK_LOG | FORK_WAIT | FORK_NEW_MOUNTNS | FORK_MOUNTNS_SLAVE | FORK_PRIVATE_TMP,
3815 NULL);
3816 if (r == 0) {
3817 r = manager_execute_generators(m, paths, /* remount_ro= */ true);
3818 _exit(r >= 0 ? EXIT_SUCCESS : EXIT_FAILURE);
3819 }
3820
3821 finish:
3822 lookup_paths_trim_generator(&m->lookup_paths);
3823 return r;
3824 }
3825
3826 int manager_transient_environment_add(Manager *m, char **plus) {
3827 char **a;
3828
3829 assert(m);
3830
3831 if (strv_isempty(plus))
3832 return 0;
3833
3834 a = strv_env_merge(m->transient_environment, plus);
3835 if (!a)
3836 return log_oom();
3837
3838 sanitize_environment(a);
3839
3840 return strv_free_and_replace(m->transient_environment, a);
3841 }
3842
3843 int manager_client_environment_modify(
3844 Manager *m,
3845 char **minus,
3846 char **plus) {
3847
3848 char **a = NULL, **b = NULL, **l;
3849
3850 assert(m);
3851
3852 if (strv_isempty(minus) && strv_isempty(plus))
3853 return 0;
3854
3855 l = m->client_environment;
3856
3857 if (!strv_isempty(minus)) {
3858 a = strv_env_delete(l, 1, minus);
3859 if (!a)
3860 return -ENOMEM;
3861
3862 l = a;
3863 }
3864
3865 if (!strv_isempty(plus)) {
3866 b = strv_env_merge(l, plus);
3867 if (!b) {
3868 strv_free(a);
3869 return -ENOMEM;
3870 }
3871
3872 l = b;
3873 }
3874
3875 if (m->client_environment != l)
3876 strv_free(m->client_environment);
3877
3878 if (a != l)
3879 strv_free(a);
3880 if (b != l)
3881 strv_free(b);
3882
3883 m->client_environment = sanitize_environment(l);
3884 return 0;
3885 }
3886
3887 int manager_get_effective_environment(Manager *m, char ***ret) {
3888 char **l;
3889
3890 assert(m);
3891 assert(ret);
3892
3893 l = strv_env_merge(m->transient_environment, m->client_environment);
3894 if (!l)
3895 return -ENOMEM;
3896
3897 *ret = l;
3898 return 0;
3899 }
3900
3901 int manager_set_default_smack_process_label(Manager *m, const char *label) {
3902 assert(m);
3903
3904 #ifdef SMACK_DEFAULT_PROCESS_LABEL
3905 if (!label)
3906 return free_and_strdup(&m->default_smack_process_label, SMACK_DEFAULT_PROCESS_LABEL);
3907 #endif
3908 if (streq_ptr(label, "/"))
3909 return free_and_strdup(&m->default_smack_process_label, NULL);
3910
3911 return free_and_strdup(&m->default_smack_process_label, label);
3912 }
3913
3914 int manager_set_default_rlimits(Manager *m, struct rlimit **default_rlimit) {
3915 assert(m);
3916
3917 for (unsigned i = 0; i < _RLIMIT_MAX; i++) {
3918 m->rlimit[i] = mfree(m->rlimit[i]);
3919
3920 if (!default_rlimit[i])
3921 continue;
3922
3923 m->rlimit[i] = newdup(struct rlimit, default_rlimit[i], 1);
3924 if (!m->rlimit[i])
3925 return log_oom();
3926 }
3927
3928 return 0;
3929 }
3930
3931 void manager_recheck_dbus(Manager *m) {
3932 assert(m);
3933
3934 /* Connects to the bus if the dbus service and socket are running. If we are running in user mode
3935 * this is all it does. In system mode we'll also connect to the system bus (which will most likely
3936 * just reuse the connection of the API bus). That's because the system bus after all runs as service
3937 * of the system instance, while in the user instance we can assume it's already there. */
3938
3939 if (MANAGER_IS_RELOADING(m))
3940 return; /* don't check while we are reloading… */
3941
3942 if (manager_dbus_is_running(m, false)) {
3943 (void) bus_init_api(m);
3944
3945 if (MANAGER_IS_SYSTEM(m))
3946 (void) bus_init_system(m);
3947 } else {
3948 (void) bus_done_api(m);
3949
3950 if (MANAGER_IS_SYSTEM(m))
3951 (void) bus_done_system(m);
3952 }
3953 }
3954
3955 static bool manager_journal_is_running(Manager *m) {
3956 Unit *u;
3957
3958 assert(m);
3959
3960 if (MANAGER_IS_TEST_RUN(m))
3961 return false;
3962
3963 /* If we are the user manager we can safely assume that the journal is up */
3964 if (!MANAGER_IS_SYSTEM(m))
3965 return true;
3966
3967 /* Check that the socket is not only up, but in RUNNING state */
3968 u = manager_get_unit(m, SPECIAL_JOURNALD_SOCKET);
3969 if (!u)
3970 return false;
3971 if (SOCKET(u)->state != SOCKET_RUNNING)
3972 return false;
3973
3974 /* Similar, check if the daemon itself is fully up, too */
3975 u = manager_get_unit(m, SPECIAL_JOURNALD_SERVICE);
3976 if (!u)
3977 return false;
3978 if (!IN_SET(SERVICE(u)->state, SERVICE_RELOAD, SERVICE_RUNNING))
3979 return false;
3980
3981 return true;
3982 }
3983
3984 void disable_printk_ratelimit(void) {
3985 /* Disable kernel's printk ratelimit.
3986 *
3987 * Logging to /dev/kmsg is most useful during early boot and shutdown, where normal logging
3988 * mechanisms are not available. The semantics of this sysctl are such that any kernel command-line
3989 * setting takes precedence. */
3990 int r;
3991
3992 r = sysctl_write("kernel/printk_devkmsg", "on");
3993 if (r < 0)
3994 log_debug_errno(r, "Failed to set sysctl kernel.printk_devkmsg=on: %m");
3995 }
3996
3997 void manager_recheck_journal(Manager *m) {
3998
3999 assert(m);
4000
4001 /* Don't bother with this unless we are in the special situation of being PID 1 */
4002 if (getpid_cached() != 1)
4003 return;
4004
4005 /* Don't check this while we are reloading, things might still change */
4006 if (MANAGER_IS_RELOADING(m))
4007 return;
4008
4009 /* The journal is fully and entirely up? If so, let's permit logging to it, if that's configured. If
4010 * the journal is down, don't ever log to it, otherwise we might end up deadlocking ourselves as we
4011 * might trigger an activation ourselves we can't fulfill. */
4012 log_set_prohibit_ipc(!manager_journal_is_running(m));
4013 log_open();
4014 }
4015
4016 static ShowStatus manager_get_show_status(Manager *m) {
4017 assert(m);
4018
4019 if (MANAGER_IS_USER(m))
4020 return _SHOW_STATUS_INVALID;
4021
4022 if (m->show_status_overridden != _SHOW_STATUS_INVALID)
4023 return m->show_status_overridden;
4024
4025 return m->show_status;
4026 }
4027
4028 bool manager_get_show_status_on(Manager *m) {
4029 assert(m);
4030
4031 return show_status_on(manager_get_show_status(m));
4032 }
4033
4034 static void set_show_status_marker(bool b) {
4035 if (b)
4036 (void) touch("/run/systemd/show-status");
4037 else
4038 (void) unlink("/run/systemd/show-status");
4039 }
4040
4041 void manager_set_show_status(Manager *m, ShowStatus mode, const char *reason) {
4042 assert(m);
4043 assert(reason);
4044 assert(mode >= 0 && mode < _SHOW_STATUS_MAX);
4045
4046 if (MANAGER_IS_USER(m))
4047 return;
4048
4049 if (mode == m->show_status)
4050 return;
4051
4052 if (m->show_status_overridden == _SHOW_STATUS_INVALID) {
4053 bool enabled;
4054
4055 enabled = show_status_on(mode);
4056 log_debug("%s (%s) showing of status (%s).",
4057 enabled ? "Enabling" : "Disabling",
4058 strna(show_status_to_string(mode)),
4059 reason);
4060
4061 set_show_status_marker(enabled);
4062 }
4063
4064 m->show_status = mode;
4065 }
4066
4067 void manager_override_show_status(Manager *m, ShowStatus mode, const char *reason) {
4068 assert(m);
4069 assert(mode < _SHOW_STATUS_MAX);
4070
4071 if (MANAGER_IS_USER(m))
4072 return;
4073
4074 if (mode == m->show_status_overridden)
4075 return;
4076
4077 m->show_status_overridden = mode;
4078
4079 if (mode == _SHOW_STATUS_INVALID)
4080 mode = m->show_status;
4081
4082 log_debug("%s (%s) showing of status (%s).",
4083 m->show_status_overridden != _SHOW_STATUS_INVALID ? "Overriding" : "Restoring",
4084 strna(show_status_to_string(mode)),
4085 reason);
4086
4087 set_show_status_marker(show_status_on(mode));
4088 }
4089
4090 const char *manager_get_confirm_spawn(Manager *m) {
4091 static int last_errno = 0;
4092 struct stat st;
4093 int r;
4094
4095 assert(m);
4096
4097 /* Here's the deal: we want to test the validity of the console but don't want
4098 * PID1 to go through the whole console process which might block. But we also
4099 * want to warn the user only once if something is wrong with the console so we
4100 * cannot do the sanity checks after spawning our children. So here we simply do
4101 * really basic tests to hopefully trap common errors.
4102 *
4103 * If the console suddenly disappear at the time our children will really it
4104 * then they will simply fail to acquire it and a positive answer will be
4105 * assumed. New children will fall back to /dev/console though.
4106 *
4107 * Note: TTYs are devices that can come and go any time, and frequently aren't
4108 * available yet during early boot (consider a USB rs232 dongle...). If for any
4109 * reason the configured console is not ready, we fall back to the default
4110 * console. */
4111
4112 if (!m->confirm_spawn || path_equal(m->confirm_spawn, "/dev/console"))
4113 return m->confirm_spawn;
4114
4115 if (stat(m->confirm_spawn, &st) < 0) {
4116 r = -errno;
4117 goto fail;
4118 }
4119
4120 if (!S_ISCHR(st.st_mode)) {
4121 r = -ENOTTY;
4122 goto fail;
4123 }
4124
4125 last_errno = 0;
4126 return m->confirm_spawn;
4127
4128 fail:
4129 if (last_errno != r)
4130 last_errno = log_warning_errno(r, "Failed to open %s, using default console: %m", m->confirm_spawn);
4131
4132 return "/dev/console";
4133 }
4134
4135 void manager_set_first_boot(Manager *m, bool b) {
4136 assert(m);
4137
4138 if (!MANAGER_IS_SYSTEM(m))
4139 return;
4140
4141 if (m->first_boot != (int) b) {
4142 if (b)
4143 (void) touch("/run/systemd/first-boot");
4144 else
4145 (void) unlink("/run/systemd/first-boot");
4146 }
4147
4148 m->first_boot = b;
4149 }
4150
4151 void manager_disable_confirm_spawn(void) {
4152 (void) touch("/run/systemd/confirm_spawn_disabled");
4153 }
4154
4155 bool manager_is_confirm_spawn_disabled(Manager *m) {
4156 if (!m->confirm_spawn)
4157 return true;
4158
4159 return access("/run/systemd/confirm_spawn_disabled", F_OK) >= 0;
4160 }
4161
4162 static bool manager_should_show_status(Manager *m, StatusType type) {
4163 assert(m);
4164
4165 if (!MANAGER_IS_SYSTEM(m))
4166 return false;
4167
4168 if (m->no_console_output)
4169 return false;
4170
4171 if (!IN_SET(manager_state(m), MANAGER_INITIALIZING, MANAGER_STARTING, MANAGER_STOPPING))
4172 return false;
4173
4174 /* If we cannot find out the status properly, just proceed. */
4175 if (type != STATUS_TYPE_EMERGENCY && manager_check_ask_password(m) > 0)
4176 return false;
4177
4178 if (type == STATUS_TYPE_NOTICE && m->show_status != SHOW_STATUS_NO)
4179 return true;
4180
4181 return manager_get_show_status_on(m);
4182 }
4183
4184 void manager_status_printf(Manager *m, StatusType type, const char *status, const char *format, ...) {
4185 va_list ap;
4186
4187 /* If m is NULL, assume we're after shutdown and let the messages through. */
4188
4189 if (m && !manager_should_show_status(m, type))
4190 return;
4191
4192 /* XXX We should totally drop the check for ephemeral here
4193 * and thus effectively make 'Type=idle' pointless. */
4194 if (type == STATUS_TYPE_EPHEMERAL && m && m->n_on_console > 0)
4195 return;
4196
4197 va_start(ap, format);
4198 status_vprintf(status, SHOW_STATUS_ELLIPSIZE|(type == STATUS_TYPE_EPHEMERAL ? SHOW_STATUS_EPHEMERAL : 0), format, ap);
4199 va_end(ap);
4200 }
4201
4202 Set* manager_get_units_requiring_mounts_for(Manager *m, const char *path) {
4203 assert(m);
4204 assert(path);
4205
4206 if (path_equal(path, "/"))
4207 path = "";
4208
4209 return hashmap_get(m->units_requiring_mounts_for, path);
4210 }
4211
4212 int manager_update_failed_units(Manager *m, Unit *u, bool failed) {
4213 unsigned size;
4214 int r;
4215
4216 assert(m);
4217 assert(u->manager == m);
4218
4219 size = set_size(m->failed_units);
4220
4221 if (failed) {
4222 r = set_ensure_put(&m->failed_units, NULL, u);
4223 if (r < 0)
4224 return log_oom();
4225 } else
4226 (void) set_remove(m->failed_units, u);
4227
4228 if (set_size(m->failed_units) != size)
4229 bus_manager_send_change_signal(m);
4230
4231 return 0;
4232 }
4233
4234 ManagerState manager_state(Manager *m) {
4235 Unit *u;
4236
4237 assert(m);
4238
4239 /* Is the special shutdown target active or queued? If so, we are in shutdown state */
4240 u = manager_get_unit(m, SPECIAL_SHUTDOWN_TARGET);
4241 if (u && unit_active_or_pending(u))
4242 return MANAGER_STOPPING;
4243
4244 /* Did we ever finish booting? If not then we are still starting up */
4245 if (!MANAGER_IS_FINISHED(m)) {
4246
4247 u = manager_get_unit(m, SPECIAL_BASIC_TARGET);
4248 if (!u || !UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u)))
4249 return MANAGER_INITIALIZING;
4250
4251 return MANAGER_STARTING;
4252 }
4253
4254 if (MANAGER_IS_SYSTEM(m)) {
4255 /* Are the rescue or emergency targets active or queued? If so we are in maintenance state */
4256 u = manager_get_unit(m, SPECIAL_RESCUE_TARGET);
4257 if (u && unit_active_or_pending(u))
4258 return MANAGER_MAINTENANCE;
4259
4260 u = manager_get_unit(m, SPECIAL_EMERGENCY_TARGET);
4261 if (u && unit_active_or_pending(u))
4262 return MANAGER_MAINTENANCE;
4263 }
4264
4265 /* Are there any failed units? If so, we are in degraded mode */
4266 if (set_size(m->failed_units) > 0)
4267 return MANAGER_DEGRADED;
4268
4269 return MANAGER_RUNNING;
4270 }
4271
4272 static void manager_unref_uid_internal(
4273 Hashmap *uid_refs,
4274 uid_t uid,
4275 bool destroy_now,
4276 int (*_clean_ipc)(uid_t uid)) {
4277
4278 uint32_t c, n;
4279
4280 assert(uid_is_valid(uid));
4281 assert(_clean_ipc);
4282
4283 /* A generic implementation, covering both manager_unref_uid() and manager_unref_gid(), under the
4284 * assumption that uid_t and gid_t are actually defined the same way, with the same validity rules.
4285 *
4286 * We store a hashmap where the key is the UID/GID and the value is a 32bit reference counter, whose
4287 * highest bit is used as flag for marking UIDs/GIDs whose IPC objects to remove when the last
4288 * reference to the UID/GID is dropped. The flag is set to on, once at least one reference from a
4289 * unit where RemoveIPC= is set is added on a UID/GID. It is reset when the UID's/GID's reference
4290 * counter drops to 0 again. */
4291
4292 assert_cc(sizeof(uid_t) == sizeof(gid_t));
4293 assert_cc(UID_INVALID == (uid_t) GID_INVALID);
4294
4295 if (uid == 0) /* We don't keep track of root, and will never destroy it */
4296 return;
4297
4298 c = PTR_TO_UINT32(hashmap_get(uid_refs, UID_TO_PTR(uid)));
4299
4300 n = c & ~DESTROY_IPC_FLAG;
4301 assert(n > 0);
4302 n--;
4303
4304 if (destroy_now && n == 0) {
4305 hashmap_remove(uid_refs, UID_TO_PTR(uid));
4306
4307 if (c & DESTROY_IPC_FLAG) {
4308 log_debug("%s " UID_FMT " is no longer referenced, cleaning up its IPC.",
4309 _clean_ipc == clean_ipc_by_uid ? "UID" : "GID",
4310 uid);
4311 (void) _clean_ipc(uid);
4312 }
4313 } else {
4314 c = n | (c & DESTROY_IPC_FLAG);
4315 assert_se(hashmap_update(uid_refs, UID_TO_PTR(uid), UINT32_TO_PTR(c)) >= 0);
4316 }
4317 }
4318
4319 void manager_unref_uid(Manager *m, uid_t uid, bool destroy_now) {
4320 manager_unref_uid_internal(m->uid_refs, uid, destroy_now, clean_ipc_by_uid);
4321 }
4322
4323 void manager_unref_gid(Manager *m, gid_t gid, bool destroy_now) {
4324 manager_unref_uid_internal(m->gid_refs, (uid_t) gid, destroy_now, clean_ipc_by_gid);
4325 }
4326
4327 static int manager_ref_uid_internal(
4328 Hashmap **uid_refs,
4329 uid_t uid,
4330 bool clean_ipc) {
4331
4332 uint32_t c, n;
4333 int r;
4334
4335 assert(uid_refs);
4336 assert(uid_is_valid(uid));
4337
4338 /* A generic implementation, covering both manager_ref_uid() and manager_ref_gid(), under the
4339 * assumption that uid_t and gid_t are actually defined the same way, with the same validity
4340 * rules. */
4341
4342 assert_cc(sizeof(uid_t) == sizeof(gid_t));
4343 assert_cc(UID_INVALID == (uid_t) GID_INVALID);
4344
4345 if (uid == 0) /* We don't keep track of root, and will never destroy it */
4346 return 0;
4347
4348 r = hashmap_ensure_allocated(uid_refs, &trivial_hash_ops);
4349 if (r < 0)
4350 return r;
4351
4352 c = PTR_TO_UINT32(hashmap_get(*uid_refs, UID_TO_PTR(uid)));
4353
4354 n = c & ~DESTROY_IPC_FLAG;
4355 n++;
4356
4357 if (n & DESTROY_IPC_FLAG) /* check for overflow */
4358 return -EOVERFLOW;
4359
4360 c = n | (c & DESTROY_IPC_FLAG) | (clean_ipc ? DESTROY_IPC_FLAG : 0);
4361
4362 return hashmap_replace(*uid_refs, UID_TO_PTR(uid), UINT32_TO_PTR(c));
4363 }
4364
4365 int manager_ref_uid(Manager *m, uid_t uid, bool clean_ipc) {
4366 return manager_ref_uid_internal(&m->uid_refs, uid, clean_ipc);
4367 }
4368
4369 int manager_ref_gid(Manager *m, gid_t gid, bool clean_ipc) {
4370 return manager_ref_uid_internal(&m->gid_refs, (uid_t) gid, clean_ipc);
4371 }
4372
4373 static void manager_vacuum_uid_refs_internal(
4374 Hashmap *uid_refs,
4375 int (*_clean_ipc)(uid_t uid)) {
4376
4377 void *p, *k;
4378
4379 assert(_clean_ipc);
4380
4381 HASHMAP_FOREACH_KEY(p, k, uid_refs) {
4382 uint32_t c, n;
4383 uid_t uid;
4384
4385 uid = PTR_TO_UID(k);
4386 c = PTR_TO_UINT32(p);
4387
4388 n = c & ~DESTROY_IPC_FLAG;
4389 if (n > 0)
4390 continue;
4391
4392 if (c & DESTROY_IPC_FLAG) {
4393 log_debug("Found unreferenced %s " UID_FMT " after reload/reexec. Cleaning up.",
4394 _clean_ipc == clean_ipc_by_uid ? "UID" : "GID",
4395 uid);
4396 (void) _clean_ipc(uid);
4397 }
4398
4399 assert_se(hashmap_remove(uid_refs, k) == p);
4400 }
4401 }
4402
4403 static void manager_vacuum_uid_refs(Manager *m) {
4404 manager_vacuum_uid_refs_internal(m->uid_refs, clean_ipc_by_uid);
4405 }
4406
4407 static void manager_vacuum_gid_refs(Manager *m) {
4408 manager_vacuum_uid_refs_internal(m->gid_refs, clean_ipc_by_gid);
4409 }
4410
4411 static void manager_vacuum(Manager *m) {
4412 assert(m);
4413
4414 /* Release any dynamic users no longer referenced */
4415 dynamic_user_vacuum(m, true);
4416
4417 /* Release any references to UIDs/GIDs no longer referenced, and destroy any IPC owned by them */
4418 manager_vacuum_uid_refs(m);
4419 manager_vacuum_gid_refs(m);
4420
4421 /* Release any runtimes no longer referenced */
4422 exec_runtime_vacuum(m);
4423 }
4424
4425 int manager_dispatch_user_lookup_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
4426 struct buffer {
4427 uid_t uid;
4428 gid_t gid;
4429 char unit_name[UNIT_NAME_MAX+1];
4430 } _packed_ buffer;
4431
4432 Manager *m = userdata;
4433 ssize_t l;
4434 size_t n;
4435 Unit *u;
4436
4437 assert_se(source);
4438 assert_se(m);
4439
4440 /* Invoked whenever a child process succeeded resolving its user/group to use and sent us the
4441 * resulting UID/GID in a datagram. We parse the datagram here and pass it off to the unit, so that
4442 * it can add a reference to the UID/GID so that it can destroy the UID/GID's IPC objects when the
4443 * reference counter drops to 0. */
4444
4445 l = recv(fd, &buffer, sizeof(buffer), MSG_DONTWAIT);
4446 if (l < 0) {
4447 if (ERRNO_IS_TRANSIENT(errno))
4448 return 0;
4449
4450 return log_error_errno(errno, "Failed to read from user lookup fd: %m");
4451 }
4452
4453 if ((size_t) l <= offsetof(struct buffer, unit_name)) {
4454 log_warning("Received too short user lookup message, ignoring.");
4455 return 0;
4456 }
4457
4458 if ((size_t) l > offsetof(struct buffer, unit_name) + UNIT_NAME_MAX) {
4459 log_warning("Received too long user lookup message, ignoring.");
4460 return 0;
4461 }
4462
4463 if (!uid_is_valid(buffer.uid) && !gid_is_valid(buffer.gid)) {
4464 log_warning("Got user lookup message with invalid UID/GID pair, ignoring.");
4465 return 0;
4466 }
4467
4468 n = (size_t) l - offsetof(struct buffer, unit_name);
4469 if (memchr(buffer.unit_name, 0, n)) {
4470 log_warning("Received lookup message with embedded NUL character, ignoring.");
4471 return 0;
4472 }
4473
4474 buffer.unit_name[n] = 0;
4475 u = manager_get_unit(m, buffer.unit_name);
4476 if (!u) {
4477 log_debug("Got user lookup message but unit doesn't exist, ignoring.");
4478 return 0;
4479 }
4480
4481 log_unit_debug(u, "User lookup succeeded: uid=" UID_FMT " gid=" GID_FMT, buffer.uid, buffer.gid);
4482
4483 unit_notify_user_lookup(u, buffer.uid, buffer.gid);
4484 return 0;
4485 }
4486
4487 static int short_uid_range(const char *path) {
4488 _cleanup_(uid_range_freep) UidRange *p = NULL;
4489 int r;
4490
4491 assert(path);
4492
4493 /* Taint systemd if we the UID range assigned to this environment doesn't at least cover 0…65534,
4494 * i.e. from root to nobody. */
4495
4496 r = uid_range_load_userns(&p, path);
4497 if (r < 0) {
4498 if (ERRNO_IS_NOT_SUPPORTED(r))
4499 return false;
4500 return log_debug_errno(r, "Failed to load %s: %m", path);
4501 }
4502
4503 return !uid_range_covers(p, 0, 65535);
4504 }
4505
4506 char* manager_taint_string(const Manager *m) {
4507 /* Returns a "taint string", e.g. "local-hwclock:var-run-bad". Only things that are detected at
4508 * runtime should be tagged here. For stuff that is known during compilation, emit a warning in the
4509 * configuration phase. */
4510
4511 assert(m);
4512
4513 const char* stage[13] = {};
4514 size_t n = 0;
4515
4516 if (m->taint_usr)
4517 stage[n++] = "split-usr";
4518
4519 _cleanup_free_ char *usrbin = NULL;
4520 if (readlink_malloc("/bin", &usrbin) < 0 || !PATH_IN_SET(usrbin, "usr/bin", "/usr/bin"))
4521 stage[n++] = "unmerged-usr";
4522
4523 if (access("/proc/cgroups", F_OK) < 0)
4524 stage[n++] = "cgroups-missing";
4525
4526 if (cg_all_unified() == 0)
4527 stage[n++] = "cgroupsv1";
4528
4529 if (clock_is_localtime(NULL) > 0)
4530 stage[n++] = "local-hwclock";
4531
4532 if (os_release_support_ended(NULL, true) > 0)
4533 stage[n++] = "support-ended";
4534
4535 _cleanup_free_ char *destination = NULL;
4536 if (readlink_malloc("/var/run", &destination) < 0 ||
4537 !PATH_IN_SET(destination, "../run", "/run"))
4538 stage[n++] = "var-run-bad";
4539
4540 _cleanup_free_ char *overflowuid = NULL, *overflowgid = NULL;
4541 if (read_one_line_file("/proc/sys/kernel/overflowuid", &overflowuid) >= 0 &&
4542 !streq(overflowuid, "65534"))
4543 stage[n++] = "overflowuid-not-65534";
4544 if (read_one_line_file("/proc/sys/kernel/overflowgid", &overflowgid) >= 0 &&
4545 !streq(overflowgid, "65534"))
4546 stage[n++] = "overflowgid-not-65534";
4547
4548 struct utsname uts;
4549 assert_se(uname(&uts) >= 0);
4550 if (strverscmp_improved(uts.release, KERNEL_BASELINE_VERSION) < 0)
4551 stage[n++] = "old-kernel";
4552
4553 if (short_uid_range("/proc/self/uid_map") > 0)
4554 stage[n++] = "short-uid-range";
4555 if (short_uid_range("/proc/self/gid_map") > 0)
4556 stage[n++] = "short-gid-range";
4557
4558 assert(n < ELEMENTSOF(stage) - 1); /* One extra for NULL terminator */
4559
4560 return strv_join((char**) stage, ":");
4561 }
4562
4563 void manager_ref_console(Manager *m) {
4564 assert(m);
4565
4566 m->n_on_console++;
4567 }
4568
4569 void manager_unref_console(Manager *m) {
4570
4571 assert(m->n_on_console > 0);
4572 m->n_on_console--;
4573
4574 if (m->n_on_console == 0)
4575 m->no_console_output = false; /* unset no_console_output flag, since the console is definitely free now */
4576 }
4577
4578 void manager_override_log_level(Manager *m, int level) {
4579 _cleanup_free_ char *s = NULL;
4580 assert(m);
4581
4582 if (!m->log_level_overridden) {
4583 m->original_log_level = log_get_max_level();
4584 m->log_level_overridden = true;
4585 }
4586
4587 (void) log_level_to_string_alloc(level, &s);
4588 log_info("Setting log level to %s.", strna(s));
4589
4590 log_set_max_level(level);
4591 }
4592
4593 void manager_restore_original_log_level(Manager *m) {
4594 _cleanup_free_ char *s = NULL;
4595 assert(m);
4596
4597 if (!m->log_level_overridden)
4598 return;
4599
4600 (void) log_level_to_string_alloc(m->original_log_level, &s);
4601 log_info("Restoring log level to original (%s).", strna(s));
4602
4603 log_set_max_level(m->original_log_level);
4604 m->log_level_overridden = false;
4605 }
4606
4607 void manager_override_log_target(Manager *m, LogTarget target) {
4608 assert(m);
4609
4610 if (!m->log_target_overridden) {
4611 m->original_log_target = log_get_target();
4612 m->log_target_overridden = true;
4613 }
4614
4615 log_info("Setting log target to %s.", log_target_to_string(target));
4616 log_set_target(target);
4617 }
4618
4619 void manager_restore_original_log_target(Manager *m) {
4620 assert(m);
4621
4622 if (!m->log_target_overridden)
4623 return;
4624
4625 log_info("Restoring log target to original %s.", log_target_to_string(m->original_log_target));
4626
4627 log_set_target(m->original_log_target);
4628 m->log_target_overridden = false;
4629 }
4630
4631 ManagerTimestamp manager_timestamp_initrd_mangle(ManagerTimestamp s) {
4632 if (in_initrd() &&
4633 s >= MANAGER_TIMESTAMP_SECURITY_START &&
4634 s <= MANAGER_TIMESTAMP_UNITS_LOAD_FINISH)
4635 return s - MANAGER_TIMESTAMP_SECURITY_START + MANAGER_TIMESTAMP_INITRD_SECURITY_START;
4636 return s;
4637 }
4638
4639 static const char *const manager_state_table[_MANAGER_STATE_MAX] = {
4640 [MANAGER_INITIALIZING] = "initializing",
4641 [MANAGER_STARTING] = "starting",
4642 [MANAGER_RUNNING] = "running",
4643 [MANAGER_DEGRADED] = "degraded",
4644 [MANAGER_MAINTENANCE] = "maintenance",
4645 [MANAGER_STOPPING] = "stopping",
4646 };
4647
4648 DEFINE_STRING_TABLE_LOOKUP(manager_state, ManagerState);
4649
4650 static const char *const manager_timestamp_table[_MANAGER_TIMESTAMP_MAX] = {
4651 [MANAGER_TIMESTAMP_FIRMWARE] = "firmware",
4652 [MANAGER_TIMESTAMP_LOADER] = "loader",
4653 [MANAGER_TIMESTAMP_KERNEL] = "kernel",
4654 [MANAGER_TIMESTAMP_INITRD] = "initrd",
4655 [MANAGER_TIMESTAMP_USERSPACE] = "userspace",
4656 [MANAGER_TIMESTAMP_FINISH] = "finish",
4657 [MANAGER_TIMESTAMP_SECURITY_START] = "security-start",
4658 [MANAGER_TIMESTAMP_SECURITY_FINISH] = "security-finish",
4659 [MANAGER_TIMESTAMP_GENERATORS_START] = "generators-start",
4660 [MANAGER_TIMESTAMP_GENERATORS_FINISH] = "generators-finish",
4661 [MANAGER_TIMESTAMP_UNITS_LOAD_START] = "units-load-start",
4662 [MANAGER_TIMESTAMP_UNITS_LOAD_FINISH] = "units-load-finish",
4663 [MANAGER_TIMESTAMP_UNITS_LOAD] = "units-load",
4664 [MANAGER_TIMESTAMP_INITRD_SECURITY_START] = "initrd-security-start",
4665 [MANAGER_TIMESTAMP_INITRD_SECURITY_FINISH] = "initrd-security-finish",
4666 [MANAGER_TIMESTAMP_INITRD_GENERATORS_START] = "initrd-generators-start",
4667 [MANAGER_TIMESTAMP_INITRD_GENERATORS_FINISH] = "initrd-generators-finish",
4668 [MANAGER_TIMESTAMP_INITRD_UNITS_LOAD_START] = "initrd-units-load-start",
4669 [MANAGER_TIMESTAMP_INITRD_UNITS_LOAD_FINISH] = "initrd-units-load-finish",
4670 };
4671
4672 DEFINE_STRING_TABLE_LOOKUP(manager_timestamp, ManagerTimestamp);
4673
4674 static const char* const oom_policy_table[_OOM_POLICY_MAX] = {
4675 [OOM_CONTINUE] = "continue",
4676 [OOM_STOP] = "stop",
4677 [OOM_KILL] = "kill",
4678 };
4679
4680 DEFINE_STRING_TABLE_LOOKUP(oom_policy, OOMPolicy);
Unnamed repository; edit this file 'description' to name the repository.