]> 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
bpf-restrict-fs: also rename functions to bpf_restrict_fs_xyz()
[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 "common-signal.h"
35 #include "confidential-virt.h"
36 #include "constants.h"
37 #include "core-varlink.h"
38 #include "creds-util.h"
39 #include "dbus-job.h"
40 #include "dbus-manager.h"
41 #include "dbus-unit.h"
42 #include "dbus.h"
43 #include "dirent-util.h"
44 #include "env-util.h"
45 #include "escape.h"
46 #include "event-util.h"
47 #include "exec-util.h"
48 #include "execute.h"
49 #include "exit-status.h"
50 #include "fd-util.h"
51 #include "fileio.h"
52 #include "generator-setup.h"
53 #include "hashmap.h"
54 #include "initrd-util.h"
55 #include "inotify-util.h"
56 #include "install.h"
57 #include "io-util.h"
58 #include "label-util.h"
59 #include "load-fragment.h"
60 #include "locale-setup.h"
61 #include "log.h"
62 #include "macro.h"
63 #include "manager.h"
64 #include "manager-dump.h"
65 #include "manager-serialize.h"
66 #include "memory-util.h"
67 #include "mkdir-label.h"
68 #include "mount-util.h"
69 #include "os-util.h"
70 #include "parse-util.h"
71 #include "path-lookup.h"
72 #include "path-util.h"
73 #include "plymouth-util.h"
74 #include "pretty-print.h"
75 #include "process-util.h"
76 #include "psi-util.h"
77 #include "ratelimit.h"
78 #include "rlimit-util.h"
79 #include "rm-rf.h"
80 #include "selinux-util.h"
81 #include "signal-util.h"
82 #include "socket-util.h"
83 #include "special.h"
84 #include "stat-util.h"
85 #include "string-table.h"
86 #include "string-util.h"
87 #include "strv.h"
88 #include "strxcpyx.h"
89 #include "sysctl-util.h"
90 #include "syslog-util.h"
91 #include "terminal-util.h"
92 #include "time-util.h"
93 #include "transaction.h"
94 #include "uid-range.h"
95 #include "umask-util.h"
96 #include "unit-name.h"
97 #include "user-util.h"
98 #include "virt.h"
99 #include "watchdog.h"
100
101 #define NOTIFY_RCVBUF_SIZE (8*1024*1024)
102 #define CGROUPS_AGENT_RCVBUF_SIZE (8*1024*1024)
103
104 /* Initial delay and the interval for printing status messages about running jobs */
105 #define JOBS_IN_PROGRESS_WAIT_USEC (2*USEC_PER_SEC)
106 #define JOBS_IN_PROGRESS_QUIET_WAIT_USEC (25*USEC_PER_SEC)
107 #define JOBS_IN_PROGRESS_PERIOD_USEC (USEC_PER_SEC / 3)
108 #define JOBS_IN_PROGRESS_PERIOD_DIVISOR 3
109
110 /* If there are more than 1K bus messages queue across our API and direct buses, then let's not add more on top until
111 * the queue gets more empty. */
112 #define MANAGER_BUS_BUSY_THRESHOLD 1024LU
113
114 /* How many units and jobs to process of the bus queue before returning to the event loop. */
115 #define MANAGER_BUS_MESSAGE_BUDGET 100U
116
117 #define DEFAULT_TASKS_MAX ((CGroupTasksMax) { 15U, 100U }) /* 15% */
118
119 static int manager_dispatch_notify_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata);
120 static int manager_dispatch_cgroups_agent_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata);
121 static int manager_dispatch_signal_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata);
122 static int manager_dispatch_time_change_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata);
123 static int manager_dispatch_idle_pipe_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata);
124 static int manager_dispatch_user_lookup_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata);
125 static int manager_dispatch_jobs_in_progress(sd_event_source *source, usec_t usec, void *userdata);
126 static int manager_dispatch_run_queue(sd_event_source *source, void *userdata);
127 static int manager_dispatch_sigchld(sd_event_source *source, void *userdata);
128 static int manager_dispatch_timezone_change(sd_event_source *source, const struct inotify_event *event, void *userdata);
129 static int manager_run_environment_generators(Manager *m);
130 static int manager_run_generators(Manager *m);
131 static void manager_vacuum(Manager *m);
132
133 static usec_t manager_watch_jobs_next_time(Manager *m) {
134 usec_t timeout;
135
136 if (MANAGER_IS_USER(m))
137 /* Let the user manager without a timeout show status quickly, so the system manager can make
138 * use of it, if it wants to. */
139 timeout = JOBS_IN_PROGRESS_WAIT_USEC * 2 / 3;
140 else if (show_status_on(m->show_status))
141 /* When status is on, just use the usual timeout. */
142 timeout = JOBS_IN_PROGRESS_WAIT_USEC;
143 else
144 timeout = JOBS_IN_PROGRESS_QUIET_WAIT_USEC;
145
146 return usec_add(now(CLOCK_MONOTONIC), timeout);
147 }
148
149 static bool manager_is_confirm_spawn_disabled(Manager *m) {
150 assert(m);
151
152 if (!m->confirm_spawn)
153 return true;
154
155 return access("/run/systemd/confirm_spawn_disabled", F_OK) >= 0;
156 }
157
158 static void manager_watch_jobs_in_progress(Manager *m) {
159 usec_t next;
160 int r;
161
162 assert(m);
163
164 /* We do not want to show the cylon animation if the user
165 * needs to confirm service executions otherwise confirmation
166 * messages will be screwed by the cylon animation. */
167 if (!manager_is_confirm_spawn_disabled(m))
168 return;
169
170 if (m->jobs_in_progress_event_source)
171 return;
172
173 next = manager_watch_jobs_next_time(m);
174 r = sd_event_add_time(
175 m->event,
176 &m->jobs_in_progress_event_source,
177 CLOCK_MONOTONIC,
178 next, 0,
179 manager_dispatch_jobs_in_progress, m);
180 if (r < 0)
181 return;
182
183 (void) sd_event_source_set_description(m->jobs_in_progress_event_source, "manager-jobs-in-progress");
184 }
185
186 static void manager_flip_auto_status(Manager *m, bool enable, const char *reason) {
187 assert(m);
188
189 if (enable) {
190 if (m->show_status == SHOW_STATUS_AUTO)
191 manager_set_show_status(m, SHOW_STATUS_TEMPORARY, reason);
192 } else {
193 if (m->show_status == SHOW_STATUS_TEMPORARY)
194 manager_set_show_status(m, SHOW_STATUS_AUTO, reason);
195 }
196 }
197
198 static void manager_print_jobs_in_progress(Manager *m) {
199 Job *j;
200 unsigned counter = 0, print_nr;
201 char cylon[6 + CYLON_BUFFER_EXTRA + 1];
202 unsigned cylon_pos;
203 uint64_t timeout = 0;
204
205 assert(m);
206 assert(m->n_running_jobs > 0);
207
208 manager_flip_auto_status(m, true, "delay");
209
210 print_nr = (m->jobs_in_progress_iteration / JOBS_IN_PROGRESS_PERIOD_DIVISOR) % m->n_running_jobs;
211
212 HASHMAP_FOREACH(j, m->jobs)
213 if (j->state == JOB_RUNNING && counter++ == print_nr)
214 break;
215
216 /* m->n_running_jobs must be consistent with the contents of m->jobs,
217 * so the above loop must have succeeded in finding j. */
218 assert(counter == print_nr + 1);
219 assert(j);
220
221 cylon_pos = m->jobs_in_progress_iteration % 14;
222 if (cylon_pos >= 8)
223 cylon_pos = 14 - cylon_pos;
224 draw_cylon(cylon, sizeof(cylon), 6, cylon_pos);
225
226 m->jobs_in_progress_iteration++;
227
228 char job_of_n[STRLEN("( of ) ") + DECIMAL_STR_MAX(unsigned)*2] = "";
229 if (m->n_running_jobs > 1)
230 xsprintf(job_of_n, "(%u of %u) ", counter, m->n_running_jobs);
231
232 (void) job_get_timeout(j, &timeout);
233
234 /* We want to use enough information for the user to identify previous lines talking about the same
235 * unit, but keep the message as short as possible. So if 'Starting foo.service' or 'Starting
236 * foo.service - Description' were used, 'foo.service' is enough here. On the other hand, if we used
237 * 'Starting Description' before, then we shall also use 'Description' here. So we pass NULL as the
238 * second argument to unit_status_string(). */
239 const char *ident = unit_status_string(j->unit, NULL);
240
241 const char *time = FORMAT_TIMESPAN(now(CLOCK_MONOTONIC) - j->begin_usec, 1*USEC_PER_SEC);
242 const char *limit = timeout > 0 ? FORMAT_TIMESPAN(timeout - j->begin_usec, 1*USEC_PER_SEC) : "no limit";
243
244 if (m->status_unit_format == STATUS_UNIT_FORMAT_DESCRIPTION)
245 /* When using 'Description', we effectively don't have enough space to show the nested status
246 * without ellipsization, so let's not even try. */
247 manager_status_printf(m, STATUS_TYPE_EPHEMERAL, cylon,
248 "%sA %s job is running for %s (%s / %s)",
249 job_of_n,
250 job_type_to_string(j->type),
251 ident,
252 time, limit);
253 else {
254 const char *status_text = unit_status_text(j->unit);
255
256 manager_status_printf(m, STATUS_TYPE_EPHEMERAL, cylon,
257 "%sJob %s/%s running (%s / %s)%s%s",
258 job_of_n,
259 ident,
260 job_type_to_string(j->type),
261 time, limit,
262 status_text ? ": " : "",
263 strempty(status_text));
264 }
265
266 sd_notifyf(false,
267 "STATUS=%sUser job %s/%s running (%s / %s)...",
268 job_of_n,
269 ident,
270 job_type_to_string(j->type),
271 time, limit);
272 m->status_ready = false;
273 }
274
275 static int have_ask_password(void) {
276 _cleanup_closedir_ DIR *dir = NULL;
277
278 dir = opendir("/run/systemd/ask-password");
279 if (!dir) {
280 if (errno == ENOENT)
281 return false;
282 else
283 return -errno;
284 }
285
286 FOREACH_DIRENT_ALL(de, dir, return -errno)
287 if (startswith(de->d_name, "ask."))
288 return true;
289 return false;
290 }
291
292 static int manager_dispatch_ask_password_fd(sd_event_source *source,
293 int fd, uint32_t revents, void *userdata) {
294 Manager *m = ASSERT_PTR(userdata);
295
296 (void) flush_fd(fd);
297
298 m->have_ask_password = have_ask_password();
299 if (m->have_ask_password < 0)
300 /* Log error but continue. Negative have_ask_password
301 * is treated as unknown status. */
302 log_error_errno(m->have_ask_password, "Failed to list /run/systemd/ask-password: %m");
303
304 return 0;
305 }
306
307 static void manager_close_ask_password(Manager *m) {
308 assert(m);
309
310 m->ask_password_event_source = sd_event_source_disable_unref(m->ask_password_event_source);
311 m->ask_password_inotify_fd = safe_close(m->ask_password_inotify_fd);
312 m->have_ask_password = -EINVAL;
313 }
314
315 static int manager_check_ask_password(Manager *m) {
316 int r;
317
318 assert(m);
319
320 if (!m->ask_password_event_source) {
321 assert(m->ask_password_inotify_fd < 0);
322
323 (void) mkdir_p_label("/run/systemd/ask-password", 0755);
324
325 m->ask_password_inotify_fd = inotify_init1(IN_NONBLOCK|IN_CLOEXEC);
326 if (m->ask_password_inotify_fd < 0)
327 return log_error_errno(errno, "Failed to create inotify object: %m");
328
329 r = inotify_add_watch_and_warn(m->ask_password_inotify_fd,
330 "/run/systemd/ask-password",
331 IN_CREATE|IN_DELETE|IN_MOVE);
332 if (r < 0) {
333 manager_close_ask_password(m);
334 return r;
335 }
336
337 r = sd_event_add_io(m->event, &m->ask_password_event_source,
338 m->ask_password_inotify_fd, EPOLLIN,
339 manager_dispatch_ask_password_fd, m);
340 if (r < 0) {
341 log_error_errno(r, "Failed to add event source for /run/systemd/ask-password: %m");
342 manager_close_ask_password(m);
343 return r;
344 }
345
346 (void) sd_event_source_set_description(m->ask_password_event_source, "manager-ask-password");
347
348 /* Queries might have been added meanwhile... */
349 manager_dispatch_ask_password_fd(m->ask_password_event_source,
350 m->ask_password_inotify_fd, EPOLLIN, m);
351 }
352
353 return m->have_ask_password;
354 }
355
356 static int manager_watch_idle_pipe(Manager *m) {
357 int r;
358
359 assert(m);
360
361 if (m->idle_pipe_event_source)
362 return 0;
363
364 if (m->idle_pipe[2] < 0)
365 return 0;
366
367 r = sd_event_add_io(m->event, &m->idle_pipe_event_source, m->idle_pipe[2], EPOLLIN, manager_dispatch_idle_pipe_fd, m);
368 if (r < 0)
369 return log_error_errno(r, "Failed to watch idle pipe: %m");
370
371 (void) sd_event_source_set_description(m->idle_pipe_event_source, "manager-idle-pipe");
372
373 return 0;
374 }
375
376 static void manager_close_idle_pipe(Manager *m) {
377 assert(m);
378
379 m->idle_pipe_event_source = sd_event_source_disable_unref(m->idle_pipe_event_source);
380
381 safe_close_pair(m->idle_pipe);
382 safe_close_pair(m->idle_pipe + 2);
383 }
384
385 static int manager_setup_time_change(Manager *m) {
386 int r;
387
388 assert(m);
389
390 if (MANAGER_IS_TEST_RUN(m))
391 return 0;
392
393 m->time_change_event_source = sd_event_source_disable_unref(m->time_change_event_source);
394
395 r = event_add_time_change(m->event, &m->time_change_event_source, manager_dispatch_time_change_fd, m);
396 if (r < 0)
397 return log_error_errno(r, "Failed to create time change event source: %m");
398
399 /* Schedule this slightly earlier than the .timer event sources */
400 r = sd_event_source_set_priority(m->time_change_event_source, EVENT_PRIORITY_TIME_CHANGE);
401 if (r < 0)
402 return log_error_errno(r, "Failed to set priority of time change event sources: %m");
403
404 log_debug("Set up TFD_TIMER_CANCEL_ON_SET timerfd.");
405
406 return 0;
407 }
408
409 static int manager_read_timezone_stat(Manager *m) {
410 struct stat st;
411 bool changed;
412
413 assert(m);
414
415 /* Read the current stat() data of /etc/localtime so that we detect changes */
416 if (lstat("/etc/localtime", &st) < 0) {
417 log_debug_errno(errno, "Failed to stat /etc/localtime, ignoring: %m");
418 changed = m->etc_localtime_accessible;
419 m->etc_localtime_accessible = false;
420 } else {
421 usec_t k;
422
423 k = timespec_load(&st.st_mtim);
424 changed = !m->etc_localtime_accessible || k != m->etc_localtime_mtime;
425
426 m->etc_localtime_mtime = k;
427 m->etc_localtime_accessible = true;
428 }
429
430 return changed;
431 }
432
433 static int manager_setup_timezone_change(Manager *m) {
434 _cleanup_(sd_event_source_unrefp) sd_event_source *new_event = NULL;
435 int r;
436
437 assert(m);
438
439 if (MANAGER_IS_TEST_RUN(m))
440 return 0;
441
442 /* We watch /etc/localtime for three events: change of the link count (which might mean removal from /etc even
443 * though another link might be kept), renames, and file close operations after writing. Note we don't bother
444 * with IN_DELETE_SELF, as that would just report when the inode is removed entirely, i.e. after the link count
445 * went to zero and all fds to it are closed.
446 *
447 * Note that we never follow symlinks here. This is a simplification, but should cover almost all cases
448 * correctly.
449 *
450 * Note that we create the new event source first here, before releasing the old one. This should optimize
451 * behaviour as this way sd-event can reuse the old watch in case the inode didn't change. */
452
453 r = sd_event_add_inotify(m->event, &new_event, "/etc/localtime",
454 IN_ATTRIB|IN_MOVE_SELF|IN_CLOSE_WRITE|IN_DONT_FOLLOW, manager_dispatch_timezone_change, m);
455 if (r == -ENOENT) {
456 /* If the file doesn't exist yet, subscribe to /etc instead, and wait until it is created either by
457 * O_CREATE or by rename() */
458
459 log_debug_errno(r, "/etc/localtime doesn't exist yet, watching /etc instead.");
460 r = sd_event_add_inotify(m->event, &new_event, "/etc",
461 IN_CREATE|IN_MOVED_TO|IN_ONLYDIR, manager_dispatch_timezone_change, m);
462 }
463 if (r < 0)
464 return log_error_errno(r, "Failed to create timezone change event source: %m");
465
466 /* Schedule this slightly earlier than the .timer event sources */
467 r = sd_event_source_set_priority(new_event, EVENT_PRIORITY_TIME_ZONE);
468 if (r < 0)
469 return log_error_errno(r, "Failed to set priority of timezone change event sources: %m");
470
471 sd_event_source_unref(m->timezone_change_event_source);
472 m->timezone_change_event_source = TAKE_PTR(new_event);
473
474 return 0;
475 }
476
477 static int enable_special_signals(Manager *m) {
478 _cleanup_close_ int fd = -EBADF;
479
480 assert(m);
481
482 if (MANAGER_IS_TEST_RUN(m))
483 return 0;
484
485 /* Enable that we get SIGINT on control-alt-del. In containers
486 * this will fail with EPERM (older) or EINVAL (newer), so
487 * ignore that. */
488 if (reboot(RB_DISABLE_CAD) < 0 && !IN_SET(errno, EPERM, EINVAL))
489 log_warning_errno(errno, "Failed to enable ctrl-alt-del handling: %m");
490
491 fd = open_terminal("/dev/tty0", O_RDWR|O_NOCTTY|O_CLOEXEC);
492 if (fd < 0) {
493 /* Support systems without virtual console */
494 if (fd != -ENOENT)
495 log_warning_errno(errno, "Failed to open /dev/tty0: %m");
496 } else {
497 /* Enable that we get SIGWINCH on kbrequest */
498 if (ioctl(fd, KDSIGACCEPT, SIGWINCH) < 0)
499 log_warning_errno(errno, "Failed to enable kbrequest handling: %m");
500 }
501
502 return 0;
503 }
504
505 #define RTSIG_IF_AVAILABLE(signum) (signum <= SIGRTMAX ? signum : -1)
506
507 static int manager_setup_signals(Manager *m) {
508 struct sigaction sa = {
509 .sa_handler = SIG_DFL,
510 .sa_flags = SA_NOCLDSTOP|SA_RESTART,
511 };
512 sigset_t mask;
513 int r;
514
515 assert(m);
516
517 assert_se(sigaction(SIGCHLD, &sa, NULL) == 0);
518
519 /* We make liberal use of realtime signals here. On
520 * Linux/glibc we have 30 of them (with the exception of Linux
521 * on hppa, see below), between SIGRTMIN+0 ... SIGRTMIN+30
522 * (aka SIGRTMAX). */
523
524 assert_se(sigemptyset(&mask) == 0);
525 sigset_add_many(&mask,
526 SIGCHLD, /* Child died */
527 SIGTERM, /* Reexecute daemon */
528 SIGHUP, /* Reload configuration */
529 SIGUSR1, /* systemd: reconnect to D-Bus */
530 SIGUSR2, /* systemd: dump status */
531 SIGINT, /* Kernel sends us this on control-alt-del */
532 SIGWINCH, /* Kernel sends us this on kbrequest (alt-arrowup) */
533 SIGPWR, /* Some kernel drivers and upsd send us this on power failure */
534
535 SIGRTMIN+0, /* systemd: start default.target */
536 SIGRTMIN+1, /* systemd: isolate rescue.target */
537 SIGRTMIN+2, /* systemd: isolate emergency.target */
538 SIGRTMIN+3, /* systemd: start halt.target */
539 SIGRTMIN+4, /* systemd: start poweroff.target */
540 SIGRTMIN+5, /* systemd: start reboot.target */
541 SIGRTMIN+6, /* systemd: start kexec.target */
542 SIGRTMIN+7, /* systemd: start soft-reboot.target */
543
544 /* ... space for more special targets ... */
545
546 SIGRTMIN+13, /* systemd: Immediate halt */
547 SIGRTMIN+14, /* systemd: Immediate poweroff */
548 SIGRTMIN+15, /* systemd: Immediate reboot */
549 SIGRTMIN+16, /* systemd: Immediate kexec */
550 SIGRTMIN+17, /* systemd: Immediate soft-reboot */
551 SIGRTMIN+18, /* systemd: control command */
552
553 /* ... space ... */
554
555 SIGRTMIN+20, /* systemd: enable status messages */
556 SIGRTMIN+21, /* systemd: disable status messages */
557 SIGRTMIN+22, /* systemd: set log level to LOG_DEBUG */
558 SIGRTMIN+23, /* systemd: set log level to LOG_INFO */
559 SIGRTMIN+24, /* systemd: Immediate exit (--user only) */
560 SIGRTMIN+25, /* systemd: reexecute manager */
561
562 /* Apparently Linux on hppa had fewer RT signals until v3.18,
563 * SIGRTMAX was SIGRTMIN+25, and then SIGRTMIN was lowered,
564 * see commit v3.17-7614-g1f25df2eff.
565 *
566 * We cannot unconditionally make use of those signals here,
567 * so let's use a runtime check. Since these commands are
568 * accessible by different means and only really a safety
569 * net, the missing functionality on hppa shouldn't matter.
570 */
571
572 RTSIG_IF_AVAILABLE(SIGRTMIN+26), /* systemd: set log target to journal-or-kmsg */
573 RTSIG_IF_AVAILABLE(SIGRTMIN+27), /* systemd: set log target to console */
574 RTSIG_IF_AVAILABLE(SIGRTMIN+28), /* systemd: set log target to kmsg */
575 RTSIG_IF_AVAILABLE(SIGRTMIN+29), /* systemd: set log target to syslog-or-kmsg (obsolete) */
576
577 /* ... one free signal here SIGRTMIN+30 ... */
578 -1);
579 assert_se(sigprocmask(SIG_SETMASK, &mask, NULL) == 0);
580
581 m->signal_fd = signalfd(-1, &mask, SFD_NONBLOCK|SFD_CLOEXEC);
582 if (m->signal_fd < 0)
583 return -errno;
584
585 r = sd_event_add_io(m->event, &m->signal_event_source, m->signal_fd, EPOLLIN, manager_dispatch_signal_fd, m);
586 if (r < 0)
587 return r;
588
589 (void) sd_event_source_set_description(m->signal_event_source, "manager-signal");
590
591 /* Process signals a bit earlier than the rest of things, but later than notify_fd processing, so that the
592 * notify processing can still figure out to which process/service a message belongs, before we reap the
593 * process. Also, process this before handling cgroup notifications, so that we always collect child exit
594 * status information before detecting that there's no process in a cgroup. */
595 r = sd_event_source_set_priority(m->signal_event_source, EVENT_PRIORITY_SIGNALS);
596 if (r < 0)
597 return r;
598
599 if (MANAGER_IS_SYSTEM(m))
600 return enable_special_signals(m);
601
602 return 0;
603 }
604
605 static char** sanitize_environment(char **l) {
606
607 /* Let's remove some environment variables that we need ourselves to communicate with our clients */
608 strv_env_unset_many(
609 l,
610 "CACHE_DIRECTORY",
611 "CONFIGURATION_DIRECTORY",
612 "CREDENTIALS_DIRECTORY",
613 "EXIT_CODE",
614 "EXIT_STATUS",
615 "INVOCATION_ID",
616 "JOURNAL_STREAM",
617 "LISTEN_FDNAMES",
618 "LISTEN_FDS",
619 "LISTEN_PID",
620 "LOGS_DIRECTORY",
621 "LOG_NAMESPACE",
622 "MAINPID",
623 "MANAGERPID",
624 "MEMORY_PRESSURE_WATCH",
625 "MEMORY_PRESSURE_WRITE",
626 "MONITOR_EXIT_CODE",
627 "MONITOR_EXIT_STATUS",
628 "MONITOR_INVOCATION_ID",
629 "MONITOR_SERVICE_RESULT",
630 "MONITOR_UNIT",
631 "NOTIFY_SOCKET",
632 "PIDFILE",
633 "REMOTE_ADDR",
634 "REMOTE_PORT",
635 "RUNTIME_DIRECTORY",
636 "SERVICE_RESULT",
637 "STATE_DIRECTORY",
638 "SYSTEMD_EXEC_PID",
639 "TRIGGER_PATH",
640 "TRIGGER_TIMER_MONOTONIC_USEC",
641 "TRIGGER_TIMER_REALTIME_USEC",
642 "TRIGGER_UNIT",
643 "WATCHDOG_PID",
644 "WATCHDOG_USEC",
645 NULL);
646
647 /* Let's order the environment alphabetically, just to make it pretty */
648 return strv_sort(l);
649 }
650
651 int manager_default_environment(Manager *m) {
652 int r;
653
654 assert(m);
655
656 m->transient_environment = strv_free(m->transient_environment);
657
658 if (MANAGER_IS_SYSTEM(m)) {
659 /* The system manager always starts with a clean environment for its children. It does not
660 * import the kernel's or the parents' exported variables.
661 *
662 * The initial passed environment is untouched to keep /proc/self/environ valid; it is used
663 * for tagging the init process inside containers. */
664 m->transient_environment = strv_new("PATH=" DEFAULT_PATH);
665 if (!m->transient_environment)
666 return log_oom();
667
668 /* Import locale variables LC_*= from configuration */
669 (void) locale_setup(&m->transient_environment);
670 } else {
671 /* The user manager passes its own environment along to its children, except for $PATH. */
672 m->transient_environment = strv_copy(environ);
673 if (!m->transient_environment)
674 return log_oom();
675
676 r = strv_env_replace_strdup(&m->transient_environment, "PATH=" DEFAULT_USER_PATH);
677 if (r < 0)
678 return log_oom();
679 }
680
681 sanitize_environment(m->transient_environment);
682 return 0;
683 }
684
685 static int manager_setup_prefix(Manager *m) {
686 struct table_entry {
687 uint64_t type;
688 const char *suffix;
689 };
690
691 static const struct table_entry paths_system[_EXEC_DIRECTORY_TYPE_MAX] = {
692 [EXEC_DIRECTORY_RUNTIME] = { SD_PATH_SYSTEM_RUNTIME, NULL },
693 [EXEC_DIRECTORY_STATE] = { SD_PATH_SYSTEM_STATE_PRIVATE, NULL },
694 [EXEC_DIRECTORY_CACHE] = { SD_PATH_SYSTEM_STATE_CACHE, NULL },
695 [EXEC_DIRECTORY_LOGS] = { SD_PATH_SYSTEM_STATE_LOGS, NULL },
696 [EXEC_DIRECTORY_CONFIGURATION] = { SD_PATH_SYSTEM_CONFIGURATION, NULL },
697 };
698
699 static const struct table_entry paths_user[_EXEC_DIRECTORY_TYPE_MAX] = {
700 [EXEC_DIRECTORY_RUNTIME] = { SD_PATH_USER_RUNTIME, NULL },
701 [EXEC_DIRECTORY_STATE] = { SD_PATH_USER_STATE_PRIVATE, NULL },
702 [EXEC_DIRECTORY_CACHE] = { SD_PATH_USER_STATE_CACHE, NULL },
703 [EXEC_DIRECTORY_LOGS] = { SD_PATH_USER_STATE_PRIVATE, "log" },
704 [EXEC_DIRECTORY_CONFIGURATION] = { SD_PATH_USER_CONFIGURATION, NULL },
705 };
706
707 assert(m);
708
709 const struct table_entry *p = MANAGER_IS_SYSTEM(m) ? paths_system : paths_user;
710 int r;
711
712 for (ExecDirectoryType i = 0; i < _EXEC_DIRECTORY_TYPE_MAX; i++) {
713 r = sd_path_lookup(p[i].type, p[i].suffix, &m->prefix[i]);
714 if (r < 0)
715 return log_warning_errno(r, "Failed to lookup %s path: %m",
716 exec_directory_type_to_string(i));
717 }
718
719 return 0;
720 }
721
722 static void manager_free_unit_name_maps(Manager *m) {
723 m->unit_id_map = hashmap_free(m->unit_id_map);
724 m->unit_name_map = hashmap_free(m->unit_name_map);
725 m->unit_path_cache = set_free(m->unit_path_cache);
726 m->unit_cache_timestamp_hash = 0;
727 }
728
729 static int manager_setup_run_queue(Manager *m) {
730 int r;
731
732 assert(m);
733 assert(!m->run_queue_event_source);
734
735 r = sd_event_add_defer(m->event, &m->run_queue_event_source, manager_dispatch_run_queue, m);
736 if (r < 0)
737 return r;
738
739 r = sd_event_source_set_priority(m->run_queue_event_source, EVENT_PRIORITY_RUN_QUEUE);
740 if (r < 0)
741 return r;
742
743 r = sd_event_source_set_enabled(m->run_queue_event_source, SD_EVENT_OFF);
744 if (r < 0)
745 return r;
746
747 (void) sd_event_source_set_description(m->run_queue_event_source, "manager-run-queue");
748
749 return 0;
750 }
751
752 static int manager_setup_sigchld_event_source(Manager *m) {
753 int r;
754
755 assert(m);
756 assert(!m->sigchld_event_source);
757
758 r = sd_event_add_defer(m->event, &m->sigchld_event_source, manager_dispatch_sigchld, m);
759 if (r < 0)
760 return r;
761
762 r = sd_event_source_set_priority(m->sigchld_event_source, EVENT_PRIORITY_SIGCHLD);
763 if (r < 0)
764 return r;
765
766 r = sd_event_source_set_enabled(m->sigchld_event_source, SD_EVENT_OFF);
767 if (r < 0)
768 return r;
769
770 (void) sd_event_source_set_description(m->sigchld_event_source, "manager-sigchld");
771
772 return 0;
773 }
774
775 int manager_setup_memory_pressure_event_source(Manager *m) {
776 int r;
777
778 assert(m);
779
780 m->memory_pressure_event_source = sd_event_source_disable_unref(m->memory_pressure_event_source);
781
782 r = sd_event_add_memory_pressure(m->event, &m->memory_pressure_event_source, NULL, NULL);
783 if (r < 0)
784 log_full_errno(ERRNO_IS_NOT_SUPPORTED(r) || ERRNO_IS_PRIVILEGE(r) || (r == -EHOSTDOWN) ? LOG_DEBUG : LOG_NOTICE, r,
785 "Failed to establish memory pressure event source, ignoring: %m");
786 else if (m->defaults.memory_pressure_threshold_usec != USEC_INFINITY) {
787
788 /* If there's a default memory pressure threshold set, also apply it to the service manager itself */
789 r = sd_event_source_set_memory_pressure_period(
790 m->memory_pressure_event_source,
791 m->defaults.memory_pressure_threshold_usec,
792 MEMORY_PRESSURE_DEFAULT_WINDOW_USEC);
793 if (r < 0)
794 log_warning_errno(r, "Failed to adjust memory pressure threshold, ignoring: %m");
795 }
796
797 return 0;
798 }
799
800 static int manager_find_credentials_dirs(Manager *m) {
801 const char *e;
802 int r;
803
804 assert(m);
805
806 r = get_credentials_dir(&e);
807 if (r < 0) {
808 if (r != -ENXIO)
809 log_debug_errno(r, "Failed to determine credentials directory, ignoring: %m");
810 } else {
811 m->received_credentials_directory = strdup(e);
812 if (!m->received_credentials_directory)
813 return -ENOMEM;
814 }
815
816 r = get_encrypted_credentials_dir(&e);
817 if (r < 0) {
818 if (r != -ENXIO)
819 log_debug_errno(r, "Failed to determine encrypted credentials directory, ignoring: %m");
820 } else {
821 m->received_encrypted_credentials_directory = strdup(e);
822 if (!m->received_encrypted_credentials_directory)
823 return -ENOMEM;
824 }
825
826 return 0;
827 }
828
829 void manager_set_switching_root(Manager *m, bool switching_root) {
830 assert(m);
831
832 m->switching_root = MANAGER_IS_SYSTEM(m) && switching_root;
833 }
834
835 double manager_get_progress(Manager *m) {
836 assert(m);
837
838 if (MANAGER_IS_FINISHED(m) || m->n_installed_jobs == 0)
839 return 1.0;
840
841 return 1.0 - ((double) hashmap_size(m->jobs) / (double) m->n_installed_jobs);
842 }
843
844 static int compare_job_priority(const void *a, const void *b) {
845 const Job *x = a, *y = b;
846
847 return unit_compare_priority(x->unit, y->unit);
848 }
849
850 int manager_new(RuntimeScope runtime_scope, ManagerTestRunFlags test_run_flags, Manager **ret) {
851 _cleanup_(manager_freep) Manager *m = NULL;
852 int r;
853
854 assert(IN_SET(runtime_scope, RUNTIME_SCOPE_SYSTEM, RUNTIME_SCOPE_USER));
855 assert(ret);
856
857 m = new(Manager, 1);
858 if (!m)
859 return -ENOMEM;
860
861 *m = (Manager) {
862 .runtime_scope = runtime_scope,
863 .objective = _MANAGER_OBJECTIVE_INVALID,
864
865 .status_unit_format = STATUS_UNIT_FORMAT_DEFAULT,
866
867 .original_log_level = -1,
868 .original_log_target = _LOG_TARGET_INVALID,
869
870 .watchdog_overridden[WATCHDOG_RUNTIME] = USEC_INFINITY,
871 .watchdog_overridden[WATCHDOG_REBOOT] = USEC_INFINITY,
872 .watchdog_overridden[WATCHDOG_KEXEC] = USEC_INFINITY,
873 .watchdog_overridden[WATCHDOG_PRETIMEOUT] = USEC_INFINITY,
874
875 .show_status_overridden = _SHOW_STATUS_INVALID,
876
877 .notify_fd = -EBADF,
878 .cgroups_agent_fd = -EBADF,
879 .signal_fd = -EBADF,
880 .user_lookup_fds = EBADF_PAIR,
881 .private_listen_fd = -EBADF,
882 .dev_autofs_fd = -EBADF,
883 .cgroup_inotify_fd = -EBADF,
884 .pin_cgroupfs_fd = -EBADF,
885 .ask_password_inotify_fd = -EBADF,
886 .idle_pipe = { -EBADF, -EBADF, -EBADF, -EBADF},
887
888 /* start as id #1, so that we can leave #0 around as "null-like" value */
889 .current_job_id = 1,
890
891 .have_ask_password = -EINVAL, /* we don't know */
892 .first_boot = -1,
893 .test_run_flags = test_run_flags,
894
895 .dump_ratelimit = (const RateLimit) { .interval = 10 * USEC_PER_MINUTE, .burst = 10 },
896
897 .executor_fd = -EBADF,
898 };
899
900 unit_defaults_init(&m->defaults, runtime_scope);
901
902 #if ENABLE_EFI
903 if (MANAGER_IS_SYSTEM(m) && detect_container() <= 0)
904 boot_timestamps(m->timestamps + MANAGER_TIMESTAMP_USERSPACE,
905 m->timestamps + MANAGER_TIMESTAMP_FIRMWARE,
906 m->timestamps + MANAGER_TIMESTAMP_LOADER);
907 #endif
908
909 /* Prepare log fields we can use for structured logging */
910 if (MANAGER_IS_SYSTEM(m)) {
911 m->unit_log_field = "UNIT=";
912 m->unit_log_format_string = "UNIT=%s";
913
914 m->invocation_log_field = "INVOCATION_ID=";
915 m->invocation_log_format_string = "INVOCATION_ID=%s";
916 } else {
917 m->unit_log_field = "USER_UNIT=";
918 m->unit_log_format_string = "USER_UNIT=%s";
919
920 m->invocation_log_field = "USER_INVOCATION_ID=";
921 m->invocation_log_format_string = "USER_INVOCATION_ID=%s";
922 }
923
924 /* Reboot immediately if the user hits C-A-D more often than 7x per 2s */
925 m->ctrl_alt_del_ratelimit = (const RateLimit) { .interval = 2 * USEC_PER_SEC, .burst = 7 };
926
927 r = manager_default_environment(m);
928 if (r < 0)
929 return r;
930
931 r = hashmap_ensure_allocated(&m->units, &string_hash_ops);
932 if (r < 0)
933 return r;
934
935 r = hashmap_ensure_allocated(&m->cgroup_unit, &path_hash_ops);
936 if (r < 0)
937 return r;
938
939 r = hashmap_ensure_allocated(&m->watch_bus, &string_hash_ops);
940 if (r < 0)
941 return r;
942
943 r = prioq_ensure_allocated(&m->run_queue, compare_job_priority);
944 if (r < 0)
945 return r;
946
947 r = manager_setup_prefix(m);
948 if (r < 0)
949 return r;
950
951 r = manager_find_credentials_dirs(m);
952 if (r < 0)
953 return r;
954
955 r = sd_event_default(&m->event);
956 if (r < 0)
957 return r;
958
959 r = manager_setup_run_queue(m);
960 if (r < 0)
961 return r;
962
963 if (FLAGS_SET(test_run_flags, MANAGER_TEST_RUN_MINIMAL)) {
964 m->cgroup_root = strdup("");
965 if (!m->cgroup_root)
966 return -ENOMEM;
967 } else {
968 r = manager_setup_signals(m);
969 if (r < 0)
970 return r;
971
972 r = manager_setup_cgroup(m);
973 if (r < 0)
974 return r;
975
976 r = manager_setup_time_change(m);
977 if (r < 0)
978 return r;
979
980 r = manager_read_timezone_stat(m);
981 if (r < 0)
982 return r;
983
984 (void) manager_setup_timezone_change(m);
985
986 r = manager_setup_sigchld_event_source(m);
987 if (r < 0)
988 return r;
989
990 r = manager_setup_memory_pressure_event_source(m);
991 if (r < 0)
992 return r;
993
994 #if HAVE_LIBBPF
995 if (MANAGER_IS_SYSTEM(m) && bpf_restrict_fs_supported(/* initialize = */ true)) {
996 r = bpf_restrict_fs_setup(m);
997 if (r < 0)
998 log_warning_errno(r, "Failed to setup LSM BPF, ignoring: %m");
999 }
1000 #endif
1001 }
1002
1003 if (test_run_flags == 0) {
1004 if (MANAGER_IS_SYSTEM(m))
1005 r = mkdir_label("/run/systemd/units", 0755);
1006 else {
1007 _cleanup_free_ char *units_path = NULL;
1008 r = xdg_user_runtime_dir(&units_path, "/systemd/units");
1009 if (r < 0)
1010 return r;
1011 r = mkdir_p_label(units_path, 0755);
1012 }
1013
1014 if (r < 0 && r != -EEXIST)
1015 return r;
1016
1017 m->executor_fd = open(SYSTEMD_EXECUTOR_BINARY_PATH, O_CLOEXEC|O_PATH);
1018 if (m->executor_fd < 0)
1019 return log_emergency_errno(errno,
1020 "Failed to open executor binary '%s': %m",
1021 SYSTEMD_EXECUTOR_BINARY_PATH);
1022 } else if (!FLAGS_SET(test_run_flags, MANAGER_TEST_DONT_OPEN_EXECUTOR)) {
1023 _cleanup_free_ char *self_exe = NULL, *executor_path = NULL;
1024 _cleanup_close_ int self_dir_fd = -EBADF;
1025 int level = LOG_DEBUG;
1026
1027 /* Prefer sd-executor from the same directory as the test, e.g.: when running unit tests from the
1028 * build directory. Fallback to working directory and then the installation path. */
1029 r = readlink_and_make_absolute("/proc/self/exe", &self_exe);
1030 if (r < 0)
1031 return r;
1032
1033 self_dir_fd = open_parent(self_exe, O_CLOEXEC|O_PATH|O_DIRECTORY, 0);
1034 if (self_dir_fd < 0)
1035 return self_dir_fd;
1036
1037 m->executor_fd = RET_NERRNO(openat(self_dir_fd, "systemd-executor", O_CLOEXEC|O_PATH));
1038 if (m->executor_fd == -ENOENT)
1039 m->executor_fd = RET_NERRNO(openat(AT_FDCWD, "systemd-executor", O_CLOEXEC|O_PATH));
1040 if (m->executor_fd == -ENOENT) {
1041 m->executor_fd = RET_NERRNO(open(SYSTEMD_EXECUTOR_BINARY_PATH, O_CLOEXEC|O_PATH));
1042 level = LOG_WARNING; /* Tests should normally use local builds */
1043 }
1044 if (m->executor_fd < 0)
1045 return m->executor_fd;
1046
1047 r = fd_get_path(m->executor_fd, &executor_path);
1048 if (r < 0)
1049 return r;
1050
1051 log_full(level, "Using systemd-executor binary from '%s'.", executor_path);
1052 }
1053
1054 /* Note that we do not set up the notify fd here. We do that after deserialization,
1055 * since they might have gotten serialized across the reexec. */
1056
1057 *ret = TAKE_PTR(m);
1058
1059 return 0;
1060 }
1061
1062 static int manager_setup_notify(Manager *m) {
1063 int r;
1064
1065 if (MANAGER_IS_TEST_RUN(m))
1066 return 0;
1067
1068 if (m->notify_fd < 0) {
1069 _cleanup_close_ int fd = -EBADF;
1070 union sockaddr_union sa;
1071 socklen_t sa_len;
1072
1073 /* First free all secondary fields */
1074 m->notify_socket = mfree(m->notify_socket);
1075 m->notify_event_source = sd_event_source_disable_unref(m->notify_event_source);
1076
1077 fd = socket(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC|SOCK_NONBLOCK, 0);
1078 if (fd < 0)
1079 return log_error_errno(errno, "Failed to allocate notification socket: %m");
1080
1081 fd_increase_rxbuf(fd, NOTIFY_RCVBUF_SIZE);
1082
1083 m->notify_socket = path_join(m->prefix[EXEC_DIRECTORY_RUNTIME], "systemd/notify");
1084 if (!m->notify_socket)
1085 return log_oom();
1086
1087 r = sockaddr_un_set_path(&sa.un, m->notify_socket);
1088 if (r < 0)
1089 return log_error_errno(r, "Notify socket '%s' not valid for AF_UNIX socket address, refusing.",
1090 m->notify_socket);
1091 sa_len = r;
1092
1093 (void) mkdir_parents_label(m->notify_socket, 0755);
1094 (void) sockaddr_un_unlink(&sa.un);
1095
1096 r = mac_selinux_bind(fd, &sa.sa, sa_len);
1097 if (r < 0)
1098 return log_error_errno(r, "bind(%s) failed: %m", m->notify_socket);
1099
1100 r = setsockopt_int(fd, SOL_SOCKET, SO_PASSCRED, true);
1101 if (r < 0)
1102 return log_error_errno(r, "SO_PASSCRED failed: %m");
1103
1104 m->notify_fd = TAKE_FD(fd);
1105
1106 log_debug("Using notification socket %s", m->notify_socket);
1107 }
1108
1109 if (!m->notify_event_source) {
1110 r = sd_event_add_io(m->event, &m->notify_event_source, m->notify_fd, EPOLLIN, manager_dispatch_notify_fd, m);
1111 if (r < 0)
1112 return log_error_errno(r, "Failed to allocate notify event source: %m");
1113
1114 /* Process notification messages a bit earlier than SIGCHLD, so that we can still identify to which
1115 * service an exit message belongs. */
1116 r = sd_event_source_set_priority(m->notify_event_source, EVENT_PRIORITY_NOTIFY);
1117 if (r < 0)
1118 return log_error_errno(r, "Failed to set priority of notify event source: %m");
1119
1120 (void) sd_event_source_set_description(m->notify_event_source, "manager-notify");
1121 }
1122
1123 return 0;
1124 }
1125
1126 static int manager_setup_cgroups_agent(Manager *m) {
1127
1128 static const union sockaddr_union sa = {
1129 .un.sun_family = AF_UNIX,
1130 .un.sun_path = "/run/systemd/cgroups-agent",
1131 };
1132 int r;
1133
1134 /* This creates a listening socket we receive cgroups agent messages on. We do not use D-Bus for delivering
1135 * these messages from the cgroups agent binary to PID 1, as the cgroups agent binary is very short-living, and
1136 * each instance of it needs a new D-Bus connection. Since D-Bus connections are SOCK_STREAM/AF_UNIX, on
1137 * overloaded systems the backlog of the D-Bus socket becomes relevant, as not more than the configured number
1138 * of D-Bus connections may be queued until the kernel will start dropping further incoming connections,
1139 * possibly resulting in lost cgroups agent messages. To avoid this, we'll use a private SOCK_DGRAM/AF_UNIX
1140 * socket, where no backlog is relevant as communication may take place without an actual connect() cycle, and
1141 * we thus won't lose messages.
1142 *
1143 * Note that PID 1 will forward the agent message to system bus, so that the user systemd instance may listen
1144 * to it. The system instance hence listens on this special socket, but the user instances listen on the system
1145 * bus for these messages. */
1146
1147 if (MANAGER_IS_TEST_RUN(m))
1148 return 0;
1149
1150 if (!MANAGER_IS_SYSTEM(m))
1151 return 0;
1152
1153 r = cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER);
1154 if (r < 0)
1155 return log_error_errno(r, "Failed to determine whether unified cgroups hierarchy is used: %m");
1156 if (r > 0) /* We don't need this anymore on the unified hierarchy */
1157 return 0;
1158
1159 if (m->cgroups_agent_fd < 0) {
1160 _cleanup_close_ int fd = -EBADF;
1161
1162 /* First free all secondary fields */
1163 m->cgroups_agent_event_source = sd_event_source_disable_unref(m->cgroups_agent_event_source);
1164
1165 fd = socket(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC|SOCK_NONBLOCK, 0);
1166 if (fd < 0)
1167 return log_error_errno(errno, "Failed to allocate cgroups agent socket: %m");
1168
1169 fd_increase_rxbuf(fd, CGROUPS_AGENT_RCVBUF_SIZE);
1170
1171 (void) sockaddr_un_unlink(&sa.un);
1172
1173 /* Only allow root to connect to this socket */
1174 WITH_UMASK(0077)
1175 r = bind(fd, &sa.sa, SOCKADDR_UN_LEN(sa.un));
1176 if (r < 0)
1177 return log_error_errno(errno, "bind(%s) failed: %m", sa.un.sun_path);
1178
1179 m->cgroups_agent_fd = TAKE_FD(fd);
1180 }
1181
1182 if (!m->cgroups_agent_event_source) {
1183 r = sd_event_add_io(m->event, &m->cgroups_agent_event_source, m->cgroups_agent_fd, EPOLLIN, manager_dispatch_cgroups_agent_fd, m);
1184 if (r < 0)
1185 return log_error_errno(r, "Failed to allocate cgroups agent event source: %m");
1186
1187 /* Process cgroups notifications early. Note that when the agent notification is received
1188 * we'll just enqueue the unit in the cgroup empty queue, hence pick a high priority than
1189 * that. Also see handling of cgroup inotify for the unified cgroup stuff. */
1190 r = sd_event_source_set_priority(m->cgroups_agent_event_source, EVENT_PRIORITY_CGROUP_AGENT);
1191 if (r < 0)
1192 return log_error_errno(r, "Failed to set priority of cgroups agent event source: %m");
1193
1194 (void) sd_event_source_set_description(m->cgroups_agent_event_source, "manager-cgroups-agent");
1195 }
1196
1197 return 0;
1198 }
1199
1200 static int manager_setup_user_lookup_fd(Manager *m) {
1201 int r;
1202
1203 assert(m);
1204
1205 /* Set up the socket pair used for passing UID/GID resolution results from forked off processes to PID
1206 * 1. Background: we can't do name lookups (NSS) from PID 1, since it might involve IPC and thus activation,
1207 * and we might hence deadlock on ourselves. Hence we do all user/group lookups asynchronously from the forked
1208 * off processes right before executing the binaries to start. In order to be able to clean up any IPC objects
1209 * created by a unit (see RemoveIPC=) we need to know in PID 1 the used UID/GID of the executed processes,
1210 * hence we establish this communication channel so that forked off processes can pass their UID/GID
1211 * information back to PID 1. The forked off processes send their resolved UID/GID to PID 1 in a simple
1212 * datagram, along with their unit name, so that we can share one communication socket pair among all units for
1213 * this purpose.
1214 *
1215 * You might wonder why we need a communication channel for this that is independent of the usual notification
1216 * socket scheme (i.e. $NOTIFY_SOCKET). The primary difference is about trust: data sent via the $NOTIFY_SOCKET
1217 * channel is only accepted if it originates from the right unit and if reception was enabled for it. The user
1218 * lookup socket OTOH is only accessible by PID 1 and its children until they exec(), and always available.
1219 *
1220 * Note that this function is called under two circumstances: when we first initialize (in which case we
1221 * allocate both the socket pair and the event source to listen on it), and when we deserialize after a reload
1222 * (in which case the socket pair already exists but we still need to allocate the event source for it). */
1223
1224 if (m->user_lookup_fds[0] < 0) {
1225
1226 /* Free all secondary fields */
1227 safe_close_pair(m->user_lookup_fds);
1228 m->user_lookup_event_source = sd_event_source_disable_unref(m->user_lookup_event_source);
1229
1230 if (socketpair(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC, 0, m->user_lookup_fds) < 0)
1231 return log_error_errno(errno, "Failed to allocate user lookup socket: %m");
1232
1233 (void) fd_increase_rxbuf(m->user_lookup_fds[0], NOTIFY_RCVBUF_SIZE);
1234 }
1235
1236 if (!m->user_lookup_event_source) {
1237 r = sd_event_add_io(m->event, &m->user_lookup_event_source, m->user_lookup_fds[0], EPOLLIN, manager_dispatch_user_lookup_fd, m);
1238 if (r < 0)
1239 return log_error_errno(errno, "Failed to allocate user lookup event source: %m");
1240
1241 /* Process even earlier than the notify event source, so that we always know first about valid UID/GID
1242 * resolutions */
1243 r = sd_event_source_set_priority(m->user_lookup_event_source, EVENT_PRIORITY_USER_LOOKUP);
1244 if (r < 0)
1245 return log_error_errno(errno, "Failed to set priority of user lookup event source: %m");
1246
1247 (void) sd_event_source_set_description(m->user_lookup_event_source, "user-lookup");
1248 }
1249
1250 return 0;
1251 }
1252
1253 static unsigned manager_dispatch_cleanup_queue(Manager *m) {
1254 Unit *u;
1255 unsigned n = 0;
1256
1257 assert(m);
1258
1259 while ((u = m->cleanup_queue)) {
1260 assert(u->in_cleanup_queue);
1261
1262 unit_free(u);
1263 n++;
1264 }
1265
1266 return n;
1267 }
1268
1269 static unsigned manager_dispatch_release_resources_queue(Manager *m) {
1270 unsigned n = 0;
1271 Unit *u;
1272
1273 assert(m);
1274
1275 while ((u = LIST_POP(release_resources_queue, m->release_resources_queue))) {
1276 assert(u->in_release_resources_queue);
1277 u->in_release_resources_queue = false;
1278
1279 n++;
1280
1281 unit_release_resources(u);
1282 }
1283
1284 return n;
1285 }
1286
1287 enum {
1288 GC_OFFSET_IN_PATH, /* This one is on the path we were traveling */
1289 GC_OFFSET_UNSURE, /* No clue */
1290 GC_OFFSET_GOOD, /* We still need this unit */
1291 GC_OFFSET_BAD, /* We don't need this unit anymore */
1292 _GC_OFFSET_MAX
1293 };
1294
1295 static void unit_gc_mark_good(Unit *u, unsigned gc_marker) {
1296 Unit *other;
1297
1298 u->gc_marker = gc_marker + GC_OFFSET_GOOD;
1299
1300 /* Recursively mark referenced units as GOOD as well */
1301 UNIT_FOREACH_DEPENDENCY(other, u, UNIT_ATOM_REFERENCES)
1302 if (other->gc_marker == gc_marker + GC_OFFSET_UNSURE)
1303 unit_gc_mark_good(other, gc_marker);
1304 }
1305
1306 static void unit_gc_sweep(Unit *u, unsigned gc_marker) {
1307 Unit *other;
1308 bool is_bad;
1309
1310 assert(u);
1311
1312 if (IN_SET(u->gc_marker - gc_marker,
1313 GC_OFFSET_GOOD, GC_OFFSET_BAD, GC_OFFSET_UNSURE, GC_OFFSET_IN_PATH))
1314 return;
1315
1316 if (u->in_cleanup_queue)
1317 goto bad;
1318
1319 if (!unit_may_gc(u))
1320 goto good;
1321
1322 u->gc_marker = gc_marker + GC_OFFSET_IN_PATH;
1323
1324 is_bad = true;
1325
1326 UNIT_FOREACH_DEPENDENCY(other, u, UNIT_ATOM_REFERENCED_BY) {
1327 unit_gc_sweep(other, gc_marker);
1328
1329 if (other->gc_marker == gc_marker + GC_OFFSET_GOOD)
1330 goto good;
1331
1332 if (other->gc_marker != gc_marker + GC_OFFSET_BAD)
1333 is_bad = false;
1334 }
1335
1336 LIST_FOREACH(refs_by_target, ref, u->refs_by_target) {
1337 unit_gc_sweep(ref->source, gc_marker);
1338
1339 if (ref->source->gc_marker == gc_marker + GC_OFFSET_GOOD)
1340 goto good;
1341
1342 if (ref->source->gc_marker != gc_marker + GC_OFFSET_BAD)
1343 is_bad = false;
1344 }
1345
1346 if (is_bad)
1347 goto bad;
1348
1349 /* We were unable to find anything out about this entry, so
1350 * let's investigate it later */
1351 u->gc_marker = gc_marker + GC_OFFSET_UNSURE;
1352 unit_add_to_gc_queue(u);
1353 return;
1354
1355 bad:
1356 /* We definitely know that this one is not useful anymore, so
1357 * let's mark it for deletion */
1358 u->gc_marker = gc_marker + GC_OFFSET_BAD;
1359 unit_add_to_cleanup_queue(u);
1360 return;
1361
1362 good:
1363 unit_gc_mark_good(u, gc_marker);
1364 }
1365
1366 static unsigned manager_dispatch_gc_unit_queue(Manager *m) {
1367 unsigned n = 0, gc_marker;
1368 Unit *u;
1369
1370 assert(m);
1371
1372 /* log_debug("Running GC..."); */
1373
1374 m->gc_marker += _GC_OFFSET_MAX;
1375 if (m->gc_marker + _GC_OFFSET_MAX <= _GC_OFFSET_MAX)
1376 m->gc_marker = 1;
1377
1378 gc_marker = m->gc_marker;
1379
1380 while ((u = LIST_POP(gc_queue, m->gc_unit_queue))) {
1381 assert(u->in_gc_queue);
1382
1383 unit_gc_sweep(u, gc_marker);
1384
1385 u->in_gc_queue = false;
1386
1387 n++;
1388
1389 if (IN_SET(u->gc_marker - gc_marker,
1390 GC_OFFSET_BAD, GC_OFFSET_UNSURE)) {
1391 if (u->id)
1392 log_unit_debug(u, "Collecting.");
1393 u->gc_marker = gc_marker + GC_OFFSET_BAD;
1394 unit_add_to_cleanup_queue(u);
1395 }
1396 }
1397
1398 return n;
1399 }
1400
1401 static unsigned manager_dispatch_gc_job_queue(Manager *m) {
1402 unsigned n = 0;
1403 Job *j;
1404
1405 assert(m);
1406
1407 while ((j = LIST_POP(gc_queue, m->gc_job_queue))) {
1408 assert(j->in_gc_queue);
1409 j->in_gc_queue = false;
1410
1411 n++;
1412
1413 if (!job_may_gc(j))
1414 continue;
1415
1416 log_unit_debug(j->unit, "Collecting job.");
1417 (void) job_finish_and_invalidate(j, JOB_COLLECTED, false, false);
1418 }
1419
1420 return n;
1421 }
1422
1423 static int manager_ratelimit_requeue(sd_event_source *s, uint64_t usec, void *userdata) {
1424 Unit *u = userdata;
1425
1426 assert(u);
1427 assert(s == u->auto_start_stop_event_source);
1428
1429 u->auto_start_stop_event_source = sd_event_source_unref(u->auto_start_stop_event_source);
1430
1431 /* Re-queue to all queues, if the rate limit hit we might have been throttled on any of them. */
1432 unit_submit_to_stop_when_unneeded_queue(u);
1433 unit_submit_to_start_when_upheld_queue(u);
1434 unit_submit_to_stop_when_bound_queue(u);
1435
1436 return 0;
1437 }
1438
1439 static int manager_ratelimit_check_and_queue(Unit *u) {
1440 int r;
1441
1442 assert(u);
1443
1444 if (ratelimit_below(&u->auto_start_stop_ratelimit))
1445 return 1;
1446
1447 /* Already queued, no need to requeue */
1448 if (u->auto_start_stop_event_source)
1449 return 0;
1450
1451 r = sd_event_add_time(
1452 u->manager->event,
1453 &u->auto_start_stop_event_source,
1454 CLOCK_MONOTONIC,
1455 ratelimit_end(&u->auto_start_stop_ratelimit),
1456 0,
1457 manager_ratelimit_requeue,
1458 u);
1459 if (r < 0)
1460 return log_unit_error_errno(u, r, "Failed to queue timer on event loop: %m");
1461
1462 return 0;
1463 }
1464
1465 static unsigned manager_dispatch_stop_when_unneeded_queue(Manager *m) {
1466 unsigned n = 0;
1467 Unit *u;
1468 int r;
1469
1470 assert(m);
1471
1472 while ((u = LIST_POP(stop_when_unneeded_queue, m->stop_when_unneeded_queue))) {
1473 _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL;
1474
1475 assert(u->in_stop_when_unneeded_queue);
1476 u->in_stop_when_unneeded_queue = false;
1477
1478 n++;
1479
1480 if (!unit_is_unneeded(u))
1481 continue;
1482
1483 log_unit_debug(u, "Unit is not needed anymore.");
1484
1485 /* If stopping a unit fails continuously we might enter a stop loop here, hence stop acting on the
1486 * service being unnecessary after a while. */
1487
1488 r = manager_ratelimit_check_and_queue(u);
1489 if (r <= 0) {
1490 log_unit_warning(u,
1491 "Unit not needed anymore, but not stopping since we tried this too often recently.%s",
1492 r == 0 ? " Will retry later." : "");
1493 continue;
1494 }
1495
1496 /* Ok, nobody needs us anymore. Sniff. Then let's commit suicide */
1497 r = manager_add_job(u->manager, JOB_STOP, u, JOB_FAIL, NULL, &error, NULL);
1498 if (r < 0)
1499 log_unit_warning_errno(u, r, "Failed to enqueue stop job, ignoring: %s", bus_error_message(&error, r));
1500 }
1501
1502 return n;
1503 }
1504
1505 static unsigned manager_dispatch_start_when_upheld_queue(Manager *m) {
1506 unsigned n = 0;
1507 Unit *u;
1508 int r;
1509
1510 assert(m);
1511
1512 while ((u = LIST_POP(start_when_upheld_queue, m->start_when_upheld_queue))) {
1513 _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL;
1514 Unit *culprit = NULL;
1515
1516 assert(u->in_start_when_upheld_queue);
1517 u->in_start_when_upheld_queue = false;
1518
1519 n++;
1520
1521 if (!unit_is_upheld_by_active(u, &culprit))
1522 continue;
1523
1524 log_unit_debug(u, "Unit is started because upheld by active unit %s.", culprit->id);
1525
1526 /* If stopping a unit fails continuously we might enter a stop loop here, hence stop acting on the
1527 * service being unnecessary after a while. */
1528
1529 r = manager_ratelimit_check_and_queue(u);
1530 if (r <= 0) {
1531 log_unit_warning(u,
1532 "Unit needs to be started because active unit %s upholds it, but not starting since we tried this too often recently.%s",
1533 culprit->id,
1534 r == 0 ? " Will retry later." : "");
1535 continue;
1536 }
1537
1538 r = manager_add_job(u->manager, JOB_START, u, JOB_FAIL, NULL, &error, NULL);
1539 if (r < 0)
1540 log_unit_warning_errno(u, r, "Failed to enqueue start job, ignoring: %s", bus_error_message(&error, r));
1541 }
1542
1543 return n;
1544 }
1545
1546 static unsigned manager_dispatch_stop_when_bound_queue(Manager *m) {
1547 unsigned n = 0;
1548 Unit *u;
1549 int r;
1550
1551 assert(m);
1552
1553 while ((u = LIST_POP(stop_when_bound_queue, m->stop_when_bound_queue))) {
1554 _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL;
1555 Unit *culprit = NULL;
1556
1557 assert(u->in_stop_when_bound_queue);
1558 u->in_stop_when_bound_queue = false;
1559
1560 n++;
1561
1562 if (!unit_is_bound_by_inactive(u, &culprit))
1563 continue;
1564
1565 log_unit_debug(u, "Unit is stopped because bound to inactive unit %s.", culprit->id);
1566
1567 /* If stopping a unit fails continuously we might enter a stop loop here, hence stop acting on the
1568 * service being unnecessary after a while. */
1569
1570 r = manager_ratelimit_check_and_queue(u);
1571 if (r <= 0) {
1572 log_unit_warning(u,
1573 "Unit needs to be stopped because it is bound to inactive unit %s it, but not stopping since we tried this too often recently.%s",
1574 culprit->id,
1575 r == 0 ? " Will retry later." : "");
1576 continue;
1577 }
1578
1579 r = manager_add_job(u->manager, JOB_STOP, u, JOB_REPLACE, NULL, &error, NULL);
1580 if (r < 0)
1581 log_unit_warning_errno(u, r, "Failed to enqueue stop job, ignoring: %s", bus_error_message(&error, r));
1582 }
1583
1584 return n;
1585 }
1586
1587 static void manager_clear_jobs_and_units(Manager *m) {
1588 Unit *u;
1589
1590 assert(m);
1591
1592 while ((u = hashmap_first(m->units)))
1593 unit_free(u);
1594
1595 manager_dispatch_cleanup_queue(m);
1596
1597 assert(!m->load_queue);
1598 assert(prioq_isempty(m->run_queue));
1599 assert(!m->dbus_unit_queue);
1600 assert(!m->dbus_job_queue);
1601 assert(!m->cleanup_queue);
1602 assert(!m->gc_unit_queue);
1603 assert(!m->gc_job_queue);
1604 assert(!m->cgroup_realize_queue);
1605 assert(!m->cgroup_empty_queue);
1606 assert(!m->cgroup_oom_queue);
1607 assert(!m->target_deps_queue);
1608 assert(!m->stop_when_unneeded_queue);
1609 assert(!m->start_when_upheld_queue);
1610 assert(!m->stop_when_bound_queue);
1611 assert(!m->release_resources_queue);
1612
1613 assert(hashmap_isempty(m->jobs));
1614 assert(hashmap_isempty(m->units));
1615
1616 m->n_on_console = 0;
1617 m->n_running_jobs = 0;
1618 m->n_installed_jobs = 0;
1619 m->n_failed_jobs = 0;
1620 }
1621
1622 Manager* manager_free(Manager *m) {
1623 if (!m)
1624 return NULL;
1625
1626 manager_clear_jobs_and_units(m);
1627
1628 for (UnitType c = 0; c < _UNIT_TYPE_MAX; c++)
1629 if (unit_vtable[c]->shutdown)
1630 unit_vtable[c]->shutdown(m);
1631
1632 /* Keep the cgroup hierarchy in place except when we know we are going down for good */
1633 manager_shutdown_cgroup(m, /* delete= */ IN_SET(m->objective, MANAGER_EXIT, MANAGER_REBOOT, MANAGER_POWEROFF, MANAGER_HALT, MANAGER_KEXEC));
1634
1635 lookup_paths_flush_generator(&m->lookup_paths);
1636
1637 bus_done(m);
1638 manager_varlink_done(m);
1639
1640 exec_shared_runtime_vacuum(m);
1641 hashmap_free(m->exec_shared_runtime_by_id);
1642
1643 dynamic_user_vacuum(m, false);
1644 hashmap_free(m->dynamic_users);
1645
1646 hashmap_free(m->units);
1647 hashmap_free(m->units_by_invocation_id);
1648 hashmap_free(m->jobs);
1649 hashmap_free(m->watch_pids);
1650 hashmap_free(m->watch_pids_more);
1651 hashmap_free(m->watch_bus);
1652
1653 prioq_free(m->run_queue);
1654
1655 set_free(m->startup_units);
1656 set_free(m->failed_units);
1657
1658 sd_event_source_unref(m->signal_event_source);
1659 sd_event_source_unref(m->sigchld_event_source);
1660 sd_event_source_unref(m->notify_event_source);
1661 sd_event_source_unref(m->cgroups_agent_event_source);
1662 sd_event_source_unref(m->time_change_event_source);
1663 sd_event_source_unref(m->timezone_change_event_source);
1664 sd_event_source_unref(m->jobs_in_progress_event_source);
1665 sd_event_source_unref(m->run_queue_event_source);
1666 sd_event_source_unref(m->user_lookup_event_source);
1667 sd_event_source_unref(m->memory_pressure_event_source);
1668
1669 safe_close(m->signal_fd);
1670 safe_close(m->notify_fd);
1671 safe_close(m->cgroups_agent_fd);
1672 safe_close_pair(m->user_lookup_fds);
1673
1674 manager_close_ask_password(m);
1675
1676 manager_close_idle_pipe(m);
1677
1678 sd_event_unref(m->event);
1679
1680 free(m->notify_socket);
1681
1682 lookup_paths_free(&m->lookup_paths);
1683 strv_free(m->transient_environment);
1684 strv_free(m->client_environment);
1685
1686 hashmap_free(m->cgroup_unit);
1687 manager_free_unit_name_maps(m);
1688
1689 free(m->switch_root);
1690 free(m->switch_root_init);
1691
1692 unit_defaults_done(&m->defaults);
1693
1694 FOREACH_ARRAY(map, m->units_needing_mounts_for, _UNIT_MOUNT_DEPENDENCY_TYPE_MAX) {
1695 assert(hashmap_isempty(*map));
1696 hashmap_free(*map);
1697 }
1698
1699 hashmap_free(m->uid_refs);
1700 hashmap_free(m->gid_refs);
1701
1702 for (ExecDirectoryType dt = 0; dt < _EXEC_DIRECTORY_TYPE_MAX; dt++)
1703 m->prefix[dt] = mfree(m->prefix[dt]);
1704 free(m->received_credentials_directory);
1705 free(m->received_encrypted_credentials_directory);
1706
1707 free(m->watchdog_pretimeout_governor);
1708 free(m->watchdog_pretimeout_governor_overridden);
1709
1710 m->fw_ctx = fw_ctx_free(m->fw_ctx);
1711
1712 #if BPF_FRAMEWORK
1713 bpf_restrict_fs_destroy(m->restrict_fs);
1714 #endif
1715
1716 safe_close(m->executor_fd);
1717
1718 return mfree(m);
1719 }
1720
1721 static void manager_enumerate_perpetual(Manager *m) {
1722 assert(m);
1723
1724 if (FLAGS_SET(m->test_run_flags, MANAGER_TEST_RUN_MINIMAL))
1725 return;
1726
1727 /* Let's ask every type to load all units from disk/kernel that it might know */
1728 for (UnitType c = 0; c < _UNIT_TYPE_MAX; c++) {
1729 if (!unit_type_supported(c)) {
1730 log_debug("Unit type .%s is not supported on this system.", unit_type_to_string(c));
1731 continue;
1732 }
1733
1734 if (unit_vtable[c]->enumerate_perpetual)
1735 unit_vtable[c]->enumerate_perpetual(m);
1736 }
1737 }
1738
1739 static void manager_enumerate(Manager *m) {
1740 assert(m);
1741
1742 if (FLAGS_SET(m->test_run_flags, MANAGER_TEST_RUN_MINIMAL))
1743 return;
1744
1745 /* Let's ask every type to load all units from disk/kernel that it might know */
1746 for (UnitType c = 0; c < _UNIT_TYPE_MAX; c++) {
1747 if (!unit_type_supported(c)) {
1748 log_debug("Unit type .%s is not supported on this system.", unit_type_to_string(c));
1749 continue;
1750 }
1751
1752 if (unit_vtable[c]->enumerate)
1753 unit_vtable[c]->enumerate(m);
1754 }
1755
1756 manager_dispatch_load_queue(m);
1757 }
1758
1759 static void manager_coldplug(Manager *m) {
1760 Unit *u;
1761 char *k;
1762 int r;
1763
1764 assert(m);
1765
1766 log_debug("Invoking unit coldplug() handlers%s", special_glyph(SPECIAL_GLYPH_ELLIPSIS));
1767
1768 /* Let's place the units back into their deserialized state */
1769 HASHMAP_FOREACH_KEY(u, k, m->units) {
1770
1771 /* ignore aliases */
1772 if (u->id != k)
1773 continue;
1774
1775 r = unit_coldplug(u);
1776 if (r < 0)
1777 log_warning_errno(r, "We couldn't coldplug %s, proceeding anyway: %m", u->id);
1778 }
1779 }
1780
1781 static void manager_catchup(Manager *m) {
1782 Unit *u;
1783 char *k;
1784
1785 assert(m);
1786
1787 log_debug("Invoking unit catchup() handlers%s", special_glyph(SPECIAL_GLYPH_ELLIPSIS));
1788
1789 /* Let's catch up on any state changes that happened while we were reloading/reexecing */
1790 HASHMAP_FOREACH_KEY(u, k, m->units) {
1791
1792 /* ignore aliases */
1793 if (u->id != k)
1794 continue;
1795
1796 unit_catchup(u);
1797 }
1798 }
1799
1800 static void manager_distribute_fds(Manager *m, FDSet *fds) {
1801 Unit *u;
1802
1803 assert(m);
1804
1805 HASHMAP_FOREACH(u, m->units) {
1806
1807 if (fdset_isempty(fds))
1808 break;
1809
1810 if (!UNIT_VTABLE(u)->distribute_fds)
1811 continue;
1812
1813 UNIT_VTABLE(u)->distribute_fds(u, fds);
1814 }
1815 }
1816
1817 static bool manager_dbus_is_running(Manager *m, bool deserialized) {
1818 Unit *u;
1819
1820 assert(m);
1821
1822 /* This checks whether the dbus instance we are supposed to expose our APIs on is up. We check both the socket
1823 * and the service unit. If the 'deserialized' parameter is true we'll check the deserialized state of the unit
1824 * rather than the current one. */
1825
1826 if (MANAGER_IS_TEST_RUN(m))
1827 return false;
1828
1829 u = manager_get_unit(m, SPECIAL_DBUS_SOCKET);
1830 if (!u)
1831 return false;
1832 if ((deserialized ? SOCKET(u)->deserialized_state : SOCKET(u)->state) != SOCKET_RUNNING)
1833 return false;
1834
1835 u = manager_get_unit(m, SPECIAL_DBUS_SERVICE);
1836 if (!u)
1837 return false;
1838 if (!IN_SET((deserialized ? SERVICE(u)->deserialized_state : SERVICE(u)->state),
1839 SERVICE_RUNNING,
1840 SERVICE_RELOAD,
1841 SERVICE_RELOAD_NOTIFY,
1842 SERVICE_RELOAD_SIGNAL))
1843 return false;
1844
1845 return true;
1846 }
1847
1848 static void manager_setup_bus(Manager *m) {
1849 assert(m);
1850
1851 /* Let's set up our private bus connection now, unconditionally */
1852 (void) bus_init_private(m);
1853
1854 /* If we are in --user mode also connect to the system bus now */
1855 if (MANAGER_IS_USER(m))
1856 (void) bus_init_system(m);
1857
1858 /* Let's connect to the bus now, but only if the unit is supposed to be up */
1859 if (manager_dbus_is_running(m, MANAGER_IS_RELOADING(m))) {
1860 (void) bus_init_api(m);
1861
1862 if (MANAGER_IS_SYSTEM(m))
1863 (void) bus_init_system(m);
1864 }
1865 }
1866
1867 static void manager_preset_all(Manager *m) {
1868 int r;
1869
1870 assert(m);
1871
1872 if (m->first_boot <= 0)
1873 return;
1874
1875 if (!MANAGER_IS_SYSTEM(m))
1876 return;
1877
1878 if (MANAGER_IS_TEST_RUN(m))
1879 return;
1880
1881 /* If this is the first boot, and we are in the host system, then preset everything */
1882 UnitFilePresetMode mode =
1883 ENABLE_FIRST_BOOT_FULL_PRESET ? UNIT_FILE_PRESET_FULL : UNIT_FILE_PRESET_ENABLE_ONLY;
1884
1885 r = unit_file_preset_all(RUNTIME_SCOPE_SYSTEM, 0, NULL, mode, NULL, 0);
1886 if (r < 0)
1887 log_full_errno(r == -EEXIST ? LOG_NOTICE : LOG_WARNING, r,
1888 "Failed to populate /etc with preset unit settings, ignoring: %m");
1889 else
1890 log_info("Populated /etc with preset unit settings.");
1891 }
1892
1893 static void manager_ready(Manager *m) {
1894 assert(m);
1895
1896 /* After having loaded everything, do the final round of catching up with what might have changed */
1897
1898 m->objective = MANAGER_OK; /* Tell everyone we are up now */
1899
1900 /* It might be safe to log to the journal now and connect to dbus */
1901 manager_recheck_journal(m);
1902 manager_recheck_dbus(m);
1903
1904 /* Let's finally catch up with any changes that took place while we were reloading/reexecing */
1905 manager_catchup(m);
1906
1907 /* Create a file which will indicate when the manager started loading units the last time. */
1908 if (MANAGER_IS_SYSTEM(m))
1909 (void) touch_file("/run/systemd/systemd-units-load", false,
1910 m->timestamps[MANAGER_TIMESTAMP_UNITS_LOAD].realtime ?: now(CLOCK_REALTIME),
1911 UID_INVALID, GID_INVALID, 0444);
1912 }
1913
1914 Manager* manager_reloading_start(Manager *m) {
1915 m->n_reloading++;
1916 dual_timestamp_now(m->timestamps + MANAGER_TIMESTAMP_UNITS_LOAD);
1917 return m;
1918 }
1919
1920 void manager_reloading_stopp(Manager **m) {
1921 if (*m) {
1922 assert((*m)->n_reloading > 0);
1923 (*m)->n_reloading--;
1924 }
1925 }
1926
1927 int manager_startup(Manager *m, FILE *serialization, FDSet *fds, const char *root) {
1928 int r;
1929
1930 assert(m);
1931
1932 /* If we are running in test mode, we still want to run the generators,
1933 * but we should not touch the real generator directories. */
1934 r = lookup_paths_init_or_warn(&m->lookup_paths, m->runtime_scope,
1935 MANAGER_IS_TEST_RUN(m) ? LOOKUP_PATHS_TEMPORARY_GENERATED : 0,
1936 root);
1937 if (r < 0)
1938 return r;
1939
1940 dual_timestamp_now(m->timestamps + manager_timestamp_initrd_mangle(MANAGER_TIMESTAMP_GENERATORS_START));
1941 r = manager_run_environment_generators(m);
1942 if (r >= 0)
1943 r = manager_run_generators(m);
1944 dual_timestamp_now(m->timestamps + manager_timestamp_initrd_mangle(MANAGER_TIMESTAMP_GENERATORS_FINISH));
1945 if (r < 0)
1946 return r;
1947
1948 manager_preset_all(m);
1949
1950 lookup_paths_log(&m->lookup_paths);
1951
1952 {
1953 /* This block is (optionally) done with the reloading counter bumped */
1954 _unused_ _cleanup_(manager_reloading_stopp) Manager *reloading = NULL;
1955
1956 /* Make sure we don't have a left-over from a previous run */
1957 if (!serialization)
1958 (void) rm_rf(m->lookup_paths.transient, 0);
1959
1960 /* If we will deserialize make sure that during enumeration this is already known, so we increase the
1961 * counter here already */
1962 if (serialization)
1963 reloading = manager_reloading_start(m);
1964
1965 /* First, enumerate what we can from all config files */
1966 dual_timestamp_now(m->timestamps + manager_timestamp_initrd_mangle(MANAGER_TIMESTAMP_UNITS_LOAD_START));
1967 manager_enumerate_perpetual(m);
1968 manager_enumerate(m);
1969 dual_timestamp_now(m->timestamps + manager_timestamp_initrd_mangle(MANAGER_TIMESTAMP_UNITS_LOAD_FINISH));
1970
1971 /* Second, deserialize if there is something to deserialize */
1972 if (serialization) {
1973 r = manager_deserialize(m, serialization, fds);
1974 if (r < 0)
1975 return log_error_errno(r, "Deserialization failed: %m");
1976 }
1977
1978 /* Any fds left? Find some unit which wants them. This is useful to allow container managers to pass
1979 * some file descriptors to us pre-initialized. This enables socket-based activation of entire
1980 * containers. */
1981 manager_distribute_fds(m, fds);
1982
1983 /* We might have deserialized the notify fd, but if we didn't then let's create the bus now */
1984 r = manager_setup_notify(m);
1985 if (r < 0)
1986 /* No sense to continue without notifications, our children would fail anyway. */
1987 return r;
1988
1989 r = manager_setup_cgroups_agent(m);
1990 if (r < 0)
1991 /* Likewise, no sense to continue without empty cgroup notifications. */
1992 return r;
1993
1994 r = manager_setup_user_lookup_fd(m);
1995 if (r < 0)
1996 /* This shouldn't fail, except if things are really broken. */
1997 return r;
1998
1999 /* Connect to the bus if we are good for it */
2000 manager_setup_bus(m);
2001
2002 /* Now that we are connected to all possible buses, let's deserialize who is tracking us. */
2003 r = bus_track_coldplug(m, &m->subscribed, false, m->deserialized_subscribed);
2004 if (r < 0)
2005 log_warning_errno(r, "Failed to deserialized tracked clients, ignoring: %m");
2006 m->deserialized_subscribed = strv_free(m->deserialized_subscribed);
2007
2008 r = manager_varlink_init(m);
2009 if (r < 0)
2010 log_warning_errno(r, "Failed to set up Varlink, ignoring: %m");
2011
2012 /* Third, fire things up! */
2013 manager_coldplug(m);
2014
2015 /* Clean up runtime objects */
2016 manager_vacuum(m);
2017
2018 if (serialization)
2019 /* Let's wait for the UnitNew/JobNew messages being sent, before we notify that the
2020 * reload is finished */
2021 m->send_reloading_done = true;
2022 }
2023
2024 manager_ready(m);
2025
2026 manager_set_switching_root(m, false);
2027
2028 return 0;
2029 }
2030
2031 int manager_add_job(
2032 Manager *m,
2033 JobType type,
2034 Unit *unit,
2035 JobMode mode,
2036 Set *affected_jobs,
2037 sd_bus_error *error,
2038 Job **ret) {
2039
2040 _cleanup_(transaction_abort_and_freep) Transaction *tr = NULL;
2041 int r;
2042
2043 assert(m);
2044 assert(type < _JOB_TYPE_MAX);
2045 assert(unit);
2046 assert(mode < _JOB_MODE_MAX);
2047
2048 if (mode == JOB_ISOLATE && type != JOB_START)
2049 return sd_bus_error_set(error, SD_BUS_ERROR_INVALID_ARGS, "Isolate is only valid for start.");
2050
2051 if (mode == JOB_ISOLATE && !unit->allow_isolate)
2052 return sd_bus_error_set(error, BUS_ERROR_NO_ISOLATION, "Operation refused, unit may not be isolated.");
2053
2054 if (mode == JOB_TRIGGERING && type != JOB_STOP)
2055 return sd_bus_error_set(error, SD_BUS_ERROR_INVALID_ARGS, "--job-mode=triggering is only valid for stop.");
2056
2057 if (mode == JOB_RESTART_DEPENDENCIES && type != JOB_START)
2058 return sd_bus_error_set(error, SD_BUS_ERROR_INVALID_ARGS, "--job-mode=restart-dependencies is only valid for start.");
2059
2060 log_unit_debug(unit, "Trying to enqueue job %s/%s/%s", unit->id, job_type_to_string(type), job_mode_to_string(mode));
2061
2062 type = job_type_collapse(type, unit);
2063
2064 tr = transaction_new(mode == JOB_REPLACE_IRREVERSIBLY);
2065 if (!tr)
2066 return -ENOMEM;
2067
2068 r = transaction_add_job_and_dependencies(
2069 tr,
2070 type,
2071 unit,
2072 /* by= */ NULL,
2073 TRANSACTION_MATTERS |
2074 (IN_SET(mode, JOB_IGNORE_DEPENDENCIES, JOB_IGNORE_REQUIREMENTS) ? TRANSACTION_IGNORE_REQUIREMENTS : 0) |
2075 (mode == JOB_IGNORE_DEPENDENCIES ? TRANSACTION_IGNORE_ORDER : 0) |
2076 (mode == JOB_RESTART_DEPENDENCIES ? TRANSACTION_PROPAGATE_START_AS_RESTART : 0),
2077 error);
2078 if (r < 0)
2079 return r;
2080
2081 if (mode == JOB_ISOLATE) {
2082 r = transaction_add_isolate_jobs(tr, m);
2083 if (r < 0)
2084 return r;
2085 }
2086
2087 if (mode == JOB_TRIGGERING) {
2088 r = transaction_add_triggering_jobs(tr, unit);
2089 if (r < 0)
2090 return r;
2091 }
2092
2093 r = transaction_activate(tr, m, mode, affected_jobs, error);
2094 if (r < 0)
2095 return r;
2096
2097 log_unit_debug(unit,
2098 "Enqueued job %s/%s as %u", unit->id,
2099 job_type_to_string(type), (unsigned) tr->anchor_job->id);
2100
2101 if (ret)
2102 *ret = tr->anchor_job;
2103
2104 tr = transaction_free(tr);
2105 return 0;
2106 }
2107
2108 int manager_add_job_by_name(Manager *m, JobType type, const char *name, JobMode mode, Set *affected_jobs, sd_bus_error *e, Job **ret) {
2109 Unit *unit = NULL; /* just to appease gcc, initialization is not really necessary */
2110 int r;
2111
2112 assert(m);
2113 assert(type < _JOB_TYPE_MAX);
2114 assert(name);
2115 assert(mode < _JOB_MODE_MAX);
2116
2117 r = manager_load_unit(m, name, NULL, NULL, &unit);
2118 if (r < 0)
2119 return r;
2120 assert(unit);
2121
2122 return manager_add_job(m, type, unit, mode, affected_jobs, e, ret);
2123 }
2124
2125 int manager_add_job_by_name_and_warn(Manager *m, JobType type, const char *name, JobMode mode, Set *affected_jobs, Job **ret) {
2126 _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL;
2127 int r;
2128
2129 assert(m);
2130 assert(type < _JOB_TYPE_MAX);
2131 assert(name);
2132 assert(mode < _JOB_MODE_MAX);
2133
2134 r = manager_add_job_by_name(m, type, name, mode, affected_jobs, &error, ret);
2135 if (r < 0)
2136 return log_warning_errno(r, "Failed to enqueue %s job for %s: %s", job_mode_to_string(mode), name, bus_error_message(&error, r));
2137
2138 return r;
2139 }
2140
2141 int manager_propagate_reload(Manager *m, Unit *unit, JobMode mode, sd_bus_error *e) {
2142 int r;
2143 _cleanup_(transaction_abort_and_freep) Transaction *tr = NULL;
2144
2145 assert(m);
2146 assert(unit);
2147 assert(mode < _JOB_MODE_MAX);
2148 assert(mode != JOB_ISOLATE); /* Isolate is only valid for start */
2149
2150 tr = transaction_new(mode == JOB_REPLACE_IRREVERSIBLY);
2151 if (!tr)
2152 return -ENOMEM;
2153
2154 /* We need an anchor job */
2155 r = transaction_add_job_and_dependencies(tr, JOB_NOP, unit, NULL, TRANSACTION_IGNORE_REQUIREMENTS|TRANSACTION_IGNORE_ORDER, e);
2156 if (r < 0)
2157 return r;
2158
2159 /* Failure in adding individual dependencies is ignored, so this always succeeds. */
2160 transaction_add_propagate_reload_jobs(
2161 tr,
2162 unit,
2163 tr->anchor_job,
2164 mode == JOB_IGNORE_DEPENDENCIES ? TRANSACTION_IGNORE_ORDER : 0);
2165
2166 r = transaction_activate(tr, m, mode, NULL, e);
2167 if (r < 0)
2168 return r;
2169
2170 tr = transaction_free(tr);
2171 return 0;
2172 }
2173
2174 Job *manager_get_job(Manager *m, uint32_t id) {
2175 assert(m);
2176
2177 return hashmap_get(m->jobs, UINT32_TO_PTR(id));
2178 }
2179
2180 Unit *manager_get_unit(Manager *m, const char *name) {
2181 assert(m);
2182 assert(name);
2183
2184 return hashmap_get(m->units, name);
2185 }
2186
2187 static int manager_dispatch_target_deps_queue(Manager *m) {
2188 Unit *u;
2189 int r = 0;
2190
2191 assert(m);
2192
2193 while ((u = LIST_POP(target_deps_queue, m->target_deps_queue))) {
2194 _cleanup_free_ Unit **targets = NULL;
2195 int n_targets;
2196
2197 assert(u->in_target_deps_queue);
2198
2199 u->in_target_deps_queue = false;
2200
2201 /* Take an "atomic" snapshot of dependencies here, as the call below will likely modify the
2202 * dependencies, and we can't have it that hash tables we iterate through are modified while
2203 * we are iterating through them. */
2204 n_targets = unit_get_dependency_array(u, UNIT_ATOM_DEFAULT_TARGET_DEPENDENCIES, &targets);
2205 if (n_targets < 0)
2206 return n_targets;
2207
2208 for (int i = 0; i < n_targets; i++) {
2209 r = unit_add_default_target_dependency(u, targets[i]);
2210 if (r < 0)
2211 return r;
2212 }
2213 }
2214
2215 return r;
2216 }
2217
2218 unsigned manager_dispatch_load_queue(Manager *m) {
2219 Unit *u;
2220 unsigned n = 0;
2221
2222 assert(m);
2223
2224 /* Make sure we are not run recursively */
2225 if (m->dispatching_load_queue)
2226 return 0;
2227
2228 m->dispatching_load_queue = true;
2229
2230 /* Dispatches the load queue. Takes a unit from the queue and
2231 * tries to load its data until the queue is empty */
2232
2233 while ((u = m->load_queue)) {
2234 assert(u->in_load_queue);
2235
2236 unit_load(u);
2237 n++;
2238 }
2239
2240 m->dispatching_load_queue = false;
2241
2242 /* Dispatch the units waiting for their target dependencies to be added now, as all targets that we know about
2243 * should be loaded and have aliases resolved */
2244 (void) manager_dispatch_target_deps_queue(m);
2245
2246 return n;
2247 }
2248
2249 bool manager_unit_cache_should_retry_load(Unit *u) {
2250 assert(u);
2251
2252 /* Automatic reloading from disk only applies to units which were not found sometime in the past, and
2253 * the not-found stub is kept pinned in the unit graph by dependencies. For units that were
2254 * previously loaded, we don't do automatic reloading, and daemon-reload is necessary to update. */
2255 if (u->load_state != UNIT_NOT_FOUND)
2256 return false;
2257
2258 /* The cache has been updated since the last time we tried to load the unit. There might be new
2259 * fragment paths to read. */
2260 if (u->manager->unit_cache_timestamp_hash != u->fragment_not_found_timestamp_hash)
2261 return true;
2262
2263 /* The cache needs to be updated because there are modifications on disk. */
2264 return !lookup_paths_timestamp_hash_same(&u->manager->lookup_paths, u->manager->unit_cache_timestamp_hash, NULL);
2265 }
2266
2267 int manager_load_unit_prepare(
2268 Manager *m,
2269 const char *name,
2270 const char *path,
2271 sd_bus_error *e,
2272 Unit **ret) {
2273
2274 _cleanup_(unit_freep) Unit *cleanup_unit = NULL;
2275 _cleanup_free_ char *nbuf = NULL;
2276 int r;
2277
2278 assert(m);
2279 assert(ret);
2280 assert(name || path);
2281
2282 /* This will prepare the unit for loading, but not actually load anything from disk. */
2283
2284 if (path && !path_is_absolute(path))
2285 return sd_bus_error_setf(e, SD_BUS_ERROR_INVALID_ARGS, "Path %s is not absolute.", path);
2286
2287 if (!name) {
2288 r = path_extract_filename(path, &nbuf);
2289 if (r < 0)
2290 return r;
2291 if (r == O_DIRECTORY)
2292 return sd_bus_error_setf(e, SD_BUS_ERROR_INVALID_ARGS, "Path '%s' refers to directory, refusing.", path);
2293
2294 name = nbuf;
2295 }
2296
2297 UnitType t = unit_name_to_type(name);
2298
2299 if (t == _UNIT_TYPE_INVALID || !unit_name_is_valid(name, UNIT_NAME_PLAIN|UNIT_NAME_INSTANCE)) {
2300 if (unit_name_is_valid(name, UNIT_NAME_TEMPLATE))
2301 return sd_bus_error_setf(e, SD_BUS_ERROR_INVALID_ARGS, "Unit name %s is missing the instance name.", name);
2302
2303 return sd_bus_error_setf(e, SD_BUS_ERROR_INVALID_ARGS, "Unit name %s is not valid.", name);
2304 }
2305
2306 Unit *unit = manager_get_unit(m, name);
2307 if (unit) {
2308 /* The time-based cache allows to start new units without daemon-reload,
2309 * but if they are already referenced (because of dependencies or ordering)
2310 * then we have to force a load of the fragment. As an optimization, check
2311 * first if anything in the usual paths was modified since the last time
2312 * the cache was loaded. Also check if the last time an attempt to load the
2313 * unit was made was before the most recent cache refresh, so that we know
2314 * we need to try again — even if the cache is current, it might have been
2315 * updated in a different context before we had a chance to retry loading
2316 * this particular unit. */
2317 if (manager_unit_cache_should_retry_load(unit))
2318 unit->load_state = UNIT_STUB;
2319 else {
2320 *ret = unit;
2321 return 0; /* The unit was already loaded */
2322 }
2323 } else {
2324 unit = cleanup_unit = unit_new(m, unit_vtable[t]->object_size);
2325 if (!unit)
2326 return -ENOMEM;
2327 }
2328
2329 if (path) {
2330 r = free_and_strdup(&unit->fragment_path, path);
2331 if (r < 0)
2332 return r;
2333 }
2334
2335 r = unit_add_name(unit, name);
2336 if (r < 0)
2337 return r;
2338
2339 unit_add_to_load_queue(unit);
2340 unit_add_to_dbus_queue(unit);
2341 unit_add_to_gc_queue(unit);
2342
2343 *ret = unit;
2344 TAKE_PTR(cleanup_unit);
2345
2346 return 1; /* The unit was added the load queue */
2347 }
2348
2349 int manager_load_unit(
2350 Manager *m,
2351 const char *name,
2352 const char *path,
2353 sd_bus_error *e,
2354 Unit **ret) {
2355 int r;
2356
2357 assert(m);
2358 assert(ret);
2359
2360 /* This will load the unit config, but not actually start any services or anything. */
2361
2362 r = manager_load_unit_prepare(m, name, path, e, ret);
2363 if (r <= 0)
2364 return r;
2365
2366 /* Unit was newly loaded */
2367 manager_dispatch_load_queue(m);
2368 *ret = unit_follow_merge(*ret);
2369 return 0;
2370 }
2371
2372 int manager_load_startable_unit_or_warn(
2373 Manager *m,
2374 const char *name,
2375 const char *path,
2376 Unit **ret) {
2377
2378 /* Load a unit, make sure it loaded fully and is not masked. */
2379
2380 _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL;
2381 Unit *unit;
2382 int r;
2383
2384 r = manager_load_unit(m, name, path, &error, &unit);
2385 if (r < 0)
2386 return log_error_errno(r, "Failed to load %s %s: %s",
2387 name ? "unit" : "unit file", name ?: path,
2388 bus_error_message(&error, r));
2389
2390 r = bus_unit_validate_load_state(unit, &error);
2391 if (r < 0)
2392 return log_error_errno(r, "%s", bus_error_message(&error, r));
2393
2394 *ret = unit;
2395 return 0;
2396 }
2397
2398 void manager_clear_jobs(Manager *m) {
2399 Job *j;
2400
2401 assert(m);
2402
2403 while ((j = hashmap_first(m->jobs)))
2404 /* No need to recurse. We're cancelling all jobs. */
2405 job_finish_and_invalidate(j, JOB_CANCELED, false, false);
2406 }
2407
2408 void manager_unwatch_pidref(Manager *m, const PidRef *pid) {
2409 assert(m);
2410
2411 for (;;) {
2412 Unit *u;
2413
2414 u = manager_get_unit_by_pidref_watching(m, pid);
2415 if (!u)
2416 break;
2417
2418 unit_unwatch_pidref(u, pid);
2419 }
2420 }
2421
2422 static int manager_dispatch_run_queue(sd_event_source *source, void *userdata) {
2423 Manager *m = ASSERT_PTR(userdata);
2424 Job *j;
2425
2426 assert(source);
2427
2428 while ((j = prioq_peek(m->run_queue))) {
2429 assert(j->installed);
2430 assert(j->in_run_queue);
2431
2432 (void) job_run_and_invalidate(j);
2433 }
2434
2435 if (m->n_running_jobs > 0)
2436 manager_watch_jobs_in_progress(m);
2437
2438 if (m->n_on_console > 0)
2439 manager_watch_idle_pipe(m);
2440
2441 return 1;
2442 }
2443
2444 void manager_trigger_run_queue(Manager *m) {
2445 int r;
2446
2447 assert(m);
2448
2449 r = sd_event_source_set_enabled(
2450 m->run_queue_event_source,
2451 prioq_isempty(m->run_queue) ? SD_EVENT_OFF : SD_EVENT_ONESHOT);
2452 if (r < 0)
2453 log_warning_errno(r, "Failed to enable job run queue event source, ignoring: %m");
2454 }
2455
2456 static unsigned manager_dispatch_dbus_queue(Manager *m) {
2457 unsigned n = 0, budget;
2458 Unit *u;
2459 Job *j;
2460
2461 assert(m);
2462
2463 /* When we are reloading, let's not wait with generating signals, since we need to exit the manager as quickly
2464 * as we can. There's no point in throttling generation of signals in that case. */
2465 if (MANAGER_IS_RELOADING(m) || m->send_reloading_done || m->pending_reload_message)
2466 budget = UINT_MAX; /* infinite budget in this case */
2467 else {
2468 /* Anything to do at all? */
2469 if (!m->dbus_unit_queue && !m->dbus_job_queue)
2470 return 0;
2471
2472 /* Do we have overly many messages queued at the moment? If so, let's not enqueue more on top, let's
2473 * sit this cycle out, and process things in a later cycle when the queues got a bit emptier. */
2474 if (manager_bus_n_queued_write(m) > MANAGER_BUS_BUSY_THRESHOLD)
2475 return 0;
2476
2477 /* Only process a certain number of units/jobs per event loop iteration. Even if the bus queue wasn't
2478 * overly full before this call we shouldn't increase it in size too wildly in one step, and we
2479 * shouldn't monopolize CPU time with generating these messages. Note the difference in counting of
2480 * this "budget" and the "threshold" above: the "budget" is decreased only once per generated message,
2481 * regardless how many buses/direct connections it is enqueued on, while the "threshold" is applied to
2482 * each queued instance of bus message, i.e. if the same message is enqueued to five buses/direct
2483 * connections it will be counted five times. This difference in counting ("references"
2484 * vs. "instances") is primarily a result of the fact that it's easier to implement it this way,
2485 * however it also reflects the thinking that the "threshold" should put a limit on used queue memory,
2486 * i.e. space, while the "budget" should put a limit on time. Also note that the "threshold" is
2487 * currently chosen much higher than the "budget". */
2488 budget = MANAGER_BUS_MESSAGE_BUDGET;
2489 }
2490
2491 while (budget != 0 && (u = m->dbus_unit_queue)) {
2492
2493 assert(u->in_dbus_queue);
2494
2495 bus_unit_send_change_signal(u);
2496 n++;
2497
2498 if (budget != UINT_MAX)
2499 budget--;
2500 }
2501
2502 while (budget != 0 && (j = m->dbus_job_queue)) {
2503 assert(j->in_dbus_queue);
2504
2505 bus_job_send_change_signal(j);
2506 n++;
2507
2508 if (budget != UINT_MAX)
2509 budget--;
2510 }
2511
2512 if (m->send_reloading_done) {
2513 m->send_reloading_done = false;
2514 bus_manager_send_reloading(m, false);
2515 n++;
2516 }
2517
2518 if (m->pending_reload_message) {
2519 bus_send_pending_reload_message(m);
2520 n++;
2521 }
2522
2523 return n;
2524 }
2525
2526 static int manager_dispatch_cgroups_agent_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
2527 Manager *m = userdata;
2528 char buf[PATH_MAX];
2529 ssize_t n;
2530
2531 n = recv(fd, buf, sizeof(buf), 0);
2532 if (n < 0)
2533 return log_error_errno(errno, "Failed to read cgroups agent message: %m");
2534 if (n == 0) {
2535 log_error("Got zero-length cgroups agent message, ignoring.");
2536 return 0;
2537 }
2538 if ((size_t) n >= sizeof(buf)) {
2539 log_error("Got overly long cgroups agent message, ignoring.");
2540 return 0;
2541 }
2542
2543 if (memchr(buf, 0, n)) {
2544 log_error("Got cgroups agent message with embedded NUL byte, ignoring.");
2545 return 0;
2546 }
2547 buf[n] = 0;
2548
2549 manager_notify_cgroup_empty(m, buf);
2550 (void) bus_forward_agent_released(m, buf);
2551
2552 return 0;
2553 }
2554
2555 static bool manager_process_barrier_fd(char * const *tags, FDSet *fds) {
2556
2557 /* nothing else must be sent when using BARRIER=1 */
2558 if (strv_contains(tags, "BARRIER=1")) {
2559 if (strv_length(tags) != 1)
2560 log_warning("Extra notification messages sent with BARRIER=1, ignoring everything.");
2561 else if (fdset_size(fds) != 1)
2562 log_warning("Got incorrect number of fds with BARRIER=1, closing them.");
2563
2564 /* Drop the message if BARRIER=1 was found */
2565 return true;
2566 }
2567
2568 return false;
2569 }
2570
2571 static void manager_invoke_notify_message(
2572 Manager *m,
2573 Unit *u,
2574 const struct ucred *ucred,
2575 char * const *tags,
2576 FDSet *fds) {
2577
2578 assert(m);
2579 assert(u);
2580 assert(ucred);
2581 assert(tags);
2582
2583 if (u->notifygen == m->notifygen) /* Already invoked on this same unit in this same iteration? */
2584 return;
2585 u->notifygen = m->notifygen;
2586
2587 if (UNIT_VTABLE(u)->notify_message)
2588 UNIT_VTABLE(u)->notify_message(u, ucred, tags, fds);
2589
2590 else if (DEBUG_LOGGING) {
2591 _cleanup_free_ char *buf = NULL, *x = NULL, *y = NULL;
2592
2593 buf = strv_join(tags, ", ");
2594 if (buf)
2595 x = ellipsize(buf, 20, 90);
2596 if (x)
2597 y = cescape(x);
2598
2599 log_unit_debug(u, "Got notification message \"%s\", ignoring.", strnull(y));
2600 }
2601 }
2602
2603 static int manager_dispatch_notify_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
2604
2605 _cleanup_fdset_free_ FDSet *fds = NULL;
2606 Manager *m = ASSERT_PTR(userdata);
2607 char buf[NOTIFY_BUFFER_MAX+1];
2608 struct iovec iovec = {
2609 .iov_base = buf,
2610 .iov_len = sizeof(buf)-1,
2611 };
2612 CMSG_BUFFER_TYPE(CMSG_SPACE(sizeof(struct ucred)) +
2613 CMSG_SPACE(sizeof(int) * NOTIFY_FD_MAX)) control;
2614 struct msghdr msghdr = {
2615 .msg_iov = &iovec,
2616 .msg_iovlen = 1,
2617 .msg_control = &control,
2618 .msg_controllen = sizeof(control),
2619 };
2620
2621 struct cmsghdr *cmsg;
2622 struct ucred *ucred = NULL;
2623 _cleanup_free_ Unit **array_copy = NULL;
2624 _cleanup_strv_free_ char **tags = NULL;
2625 Unit *u1, *u2, **array;
2626 int r, *fd_array = NULL;
2627 size_t n_fds = 0;
2628 bool found = false;
2629 ssize_t n;
2630
2631 assert(m->notify_fd == fd);
2632
2633 if (revents != EPOLLIN) {
2634 log_warning("Got unexpected poll event for notify fd.");
2635 return 0;
2636 }
2637
2638 n = recvmsg_safe(m->notify_fd, &msghdr, MSG_DONTWAIT|MSG_CMSG_CLOEXEC|MSG_TRUNC);
2639 if (ERRNO_IS_NEG_TRANSIENT(n))
2640 return 0; /* Spurious wakeup, try again */
2641 if (n == -EXFULL) {
2642 log_warning("Got message with truncated control data (too many fds sent?), ignoring.");
2643 return 0;
2644 }
2645 if (n < 0)
2646 /* If this is any other, real error, then stop processing this socket. This of course means
2647 * we won't take notification messages anymore, but that's still better than busy looping:
2648 * being woken up over and over again, but being unable to actually read the message from the
2649 * socket. */
2650 return log_error_errno(n, "Failed to receive notification message: %m");
2651
2652 CMSG_FOREACH(cmsg, &msghdr)
2653 if (cmsg->cmsg_level == SOL_SOCKET && cmsg->cmsg_type == SCM_RIGHTS) {
2654
2655 assert(!fd_array);
2656 fd_array = CMSG_TYPED_DATA(cmsg, int);
2657 n_fds = (cmsg->cmsg_len - CMSG_LEN(0)) / sizeof(int);
2658
2659 } else if (cmsg->cmsg_level == SOL_SOCKET &&
2660 cmsg->cmsg_type == SCM_CREDENTIALS &&
2661 cmsg->cmsg_len == CMSG_LEN(sizeof(struct ucred))) {
2662
2663 assert(!ucred);
2664 ucred = CMSG_TYPED_DATA(cmsg, struct ucred);
2665 }
2666
2667 if (n_fds > 0) {
2668 assert(fd_array);
2669
2670 r = fdset_new_array(&fds, fd_array, n_fds);
2671 if (r < 0) {
2672 close_many(fd_array, n_fds);
2673 log_oom();
2674 return 0;
2675 }
2676 }
2677
2678 if (!ucred || !pid_is_valid(ucred->pid)) {
2679 log_warning("Received notify message without valid credentials. Ignoring.");
2680 return 0;
2681 }
2682
2683 if ((size_t) n >= sizeof(buf) || (msghdr.msg_flags & MSG_TRUNC)) {
2684 log_warning("Received notify message exceeded maximum size. Ignoring.");
2685 return 0;
2686 }
2687
2688 /* As extra safety check, let's make sure the string we get doesn't contain embedded NUL bytes.
2689 * We permit one trailing NUL byte in the message, but don't expect it. */
2690 if (n > 1 && memchr(buf, 0, n-1)) {
2691 log_warning("Received notify message with embedded NUL bytes. Ignoring.");
2692 return 0;
2693 }
2694
2695 /* Make sure it's NUL-terminated, then parse it to obtain the tags list. */
2696 buf[n] = 0;
2697 tags = strv_split_newlines(buf);
2698 if (!tags) {
2699 log_oom();
2700 return 0;
2701 }
2702
2703 /* Possibly a barrier fd, let's see. */
2704 if (manager_process_barrier_fd(tags, fds)) {
2705 log_debug("Received barrier notification message from PID " PID_FMT ".", ucred->pid);
2706 return 0;
2707 }
2708
2709 /* Increase the generation counter used for filtering out duplicate unit invocations. */
2710 m->notifygen++;
2711
2712 /* Generate lookup key from the PID (we have no pidfd here, after all) */
2713 PidRef pidref = PIDREF_MAKE_FROM_PID(ucred->pid);
2714
2715 /* Notify every unit that might be interested, which might be multiple. */
2716 u1 = manager_get_unit_by_pidref_cgroup(m, &pidref);
2717 u2 = hashmap_get(m->watch_pids, &pidref);
2718 array = hashmap_get(m->watch_pids_more, &pidref);
2719 if (array) {
2720 size_t k = 0;
2721
2722 while (array[k])
2723 k++;
2724
2725 array_copy = newdup(Unit*, array, k+1);
2726 if (!array_copy)
2727 log_oom();
2728 }
2729 /* And now invoke the per-unit callbacks. Note that manager_invoke_notify_message() will handle
2730 * duplicate units make sure we only invoke each unit's handler once. */
2731 if (u1) {
2732 manager_invoke_notify_message(m, u1, ucred, tags, fds);
2733 found = true;
2734 }
2735 if (u2) {
2736 manager_invoke_notify_message(m, u2, ucred, tags, fds);
2737 found = true;
2738 }
2739 if (array_copy)
2740 for (size_t i = 0; array_copy[i]; i++) {
2741 manager_invoke_notify_message(m, array_copy[i], ucred, tags, fds);
2742 found = true;
2743 }
2744
2745 if (!found)
2746 log_warning("Cannot find unit for notify message of PID "PID_FMT", ignoring.", ucred->pid);
2747
2748 if (!fdset_isempty(fds))
2749 log_warning("Got extra auxiliary fds with notification message, closing them.");
2750
2751 return 0;
2752 }
2753
2754 static void manager_invoke_sigchld_event(
2755 Manager *m,
2756 Unit *u,
2757 const siginfo_t *si) {
2758
2759 assert(m);
2760 assert(u);
2761 assert(si);
2762
2763 /* Already invoked the handler of this unit in this iteration? Then don't process this again */
2764 if (u->sigchldgen == m->sigchldgen)
2765 return;
2766 u->sigchldgen = m->sigchldgen;
2767
2768 log_unit_debug(u, "Child "PID_FMT" belongs to %s.", si->si_pid, u->id);
2769 unit_unwatch_pid(u, si->si_pid);
2770
2771 if (UNIT_VTABLE(u)->sigchld_event)
2772 UNIT_VTABLE(u)->sigchld_event(u, si->si_pid, si->si_code, si->si_status);
2773 }
2774
2775 static int manager_dispatch_sigchld(sd_event_source *source, void *userdata) {
2776 Manager *m = ASSERT_PTR(userdata);
2777 siginfo_t si = {};
2778 int r;
2779
2780 assert(source);
2781
2782 /* First we call waitid() for a PID and do not reap the zombie. That way we can still access
2783 * /proc/$PID for it while it is a zombie. */
2784
2785 if (waitid(P_ALL, 0, &si, WEXITED|WNOHANG|WNOWAIT) < 0) {
2786
2787 if (errno != ECHILD)
2788 log_error_errno(errno, "Failed to peek for child with waitid(), ignoring: %m");
2789
2790 goto turn_off;
2791 }
2792
2793 if (si.si_pid <= 0)
2794 goto turn_off;
2795
2796 if (IN_SET(si.si_code, CLD_EXITED, CLD_KILLED, CLD_DUMPED)) {
2797 _cleanup_free_ Unit **array_copy = NULL;
2798 _cleanup_free_ char *name = NULL;
2799 Unit *u1, *u2, **array;
2800
2801 (void) pid_get_comm(si.si_pid, &name);
2802
2803 log_debug("Child "PID_FMT" (%s) died (code=%s, status=%i/%s)",
2804 si.si_pid, strna(name),
2805 sigchld_code_to_string(si.si_code),
2806 si.si_status,
2807 strna(si.si_code == CLD_EXITED
2808 ? exit_status_to_string(si.si_status, EXIT_STATUS_FULL)
2809 : signal_to_string(si.si_status)));
2810
2811 /* Increase the generation counter used for filtering out duplicate unit invocations */
2812 m->sigchldgen++;
2813
2814 /* We look this up by a PidRef that only consists of the PID. After all we couldn't create a
2815 * pidfd here any more even if we wanted (since the process just exited). */
2816 PidRef pidref = PIDREF_MAKE_FROM_PID(si.si_pid);
2817
2818 /* And now figure out the unit this belongs to, it might be multiple... */
2819 u1 = manager_get_unit_by_pidref_cgroup(m, &pidref);
2820 u2 = hashmap_get(m->watch_pids, &pidref);
2821 array = hashmap_get(m->watch_pids_more, &pidref);
2822 if (array) {
2823 size_t n = 0;
2824
2825 /* Count how many entries the array has */
2826 while (array[n])
2827 n++;
2828
2829 /* Make a copy of the array so that we don't trip up on the array changing beneath us */
2830 array_copy = newdup(Unit*, array, n+1);
2831 if (!array_copy)
2832 log_oom();
2833 }
2834
2835 /* Finally, execute them all. Note that u1, u2 and the array might contain duplicates, but
2836 * that's fine, manager_invoke_sigchld_event() will ensure we only invoke the handlers once for
2837 * each iteration. */
2838 if (u1) {
2839 /* We check for oom condition, in case we got SIGCHLD before the oom notification.
2840 * We only do this for the cgroup the PID belonged to. */
2841 (void) unit_check_oom(u1);
2842
2843 /* We check if systemd-oomd performed a kill so that we log and notify appropriately */
2844 (void) unit_check_oomd_kill(u1);
2845
2846 manager_invoke_sigchld_event(m, u1, &si);
2847 }
2848 if (u2)
2849 manager_invoke_sigchld_event(m, u2, &si);
2850 if (array_copy)
2851 for (size_t i = 0; array_copy[i]; i++)
2852 manager_invoke_sigchld_event(m, array_copy[i], &si);
2853 }
2854
2855 /* And now, we actually reap the zombie. */
2856 if (waitid(P_PID, si.si_pid, &si, WEXITED) < 0) {
2857 log_error_errno(errno, "Failed to dequeue child, ignoring: %m");
2858 return 0;
2859 }
2860
2861 return 0;
2862
2863 turn_off:
2864 /* All children processed for now, turn off event source */
2865
2866 r = sd_event_source_set_enabled(m->sigchld_event_source, SD_EVENT_OFF);
2867 if (r < 0)
2868 return log_error_errno(r, "Failed to disable SIGCHLD event source: %m");
2869
2870 return 0;
2871 }
2872
2873 static void manager_start_special(Manager *m, const char *name, JobMode mode) {
2874 Job *job;
2875
2876 if (manager_add_job_by_name_and_warn(m, JOB_START, name, mode, NULL, &job) < 0)
2877 return;
2878
2879 const char *s = unit_status_string(job->unit, NULL);
2880
2881 log_info("Activating special unit %s...", s);
2882
2883 sd_notifyf(false,
2884 "STATUS=Activating special unit %s...", s);
2885 m->status_ready = false;
2886 }
2887
2888 static void manager_handle_ctrl_alt_del(Manager *m) {
2889 /* If the user presses C-A-D more than
2890 * 7 times within 2s, we reboot/shutdown immediately,
2891 * unless it was disabled in system.conf */
2892
2893 if (ratelimit_below(&m->ctrl_alt_del_ratelimit) || m->cad_burst_action == EMERGENCY_ACTION_NONE)
2894 manager_start_special(m, SPECIAL_CTRL_ALT_DEL_TARGET, JOB_REPLACE_IRREVERSIBLY);
2895 else
2896 emergency_action(m, m->cad_burst_action, EMERGENCY_ACTION_WARN, NULL, -1,
2897 "Ctrl-Alt-Del was pressed more than 7 times within 2s");
2898 }
2899
2900 static int manager_dispatch_signal_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
2901 Manager *m = ASSERT_PTR(userdata);
2902 ssize_t n;
2903 struct signalfd_siginfo sfsi;
2904 int r;
2905
2906 assert(m->signal_fd == fd);
2907
2908 if (revents != EPOLLIN) {
2909 log_warning("Got unexpected events from signal file descriptor.");
2910 return 0;
2911 }
2912
2913 n = read(m->signal_fd, &sfsi, sizeof(sfsi));
2914 if (n < 0) {
2915 if (ERRNO_IS_TRANSIENT(errno))
2916 return 0;
2917
2918 /* We return an error here, which will kill this handler,
2919 * to avoid a busy loop on read error. */
2920 return log_error_errno(errno, "Reading from signal fd failed: %m");
2921 }
2922 if (n != sizeof(sfsi)) {
2923 log_warning("Truncated read from signal fd (%zi bytes), ignoring!", n);
2924 return 0;
2925 }
2926
2927 log_received_signal(sfsi.ssi_signo == SIGCHLD ||
2928 (sfsi.ssi_signo == SIGTERM && MANAGER_IS_USER(m))
2929 ? LOG_DEBUG : LOG_INFO,
2930 &sfsi);
2931
2932 switch (sfsi.ssi_signo) {
2933
2934 case SIGCHLD:
2935 r = sd_event_source_set_enabled(m->sigchld_event_source, SD_EVENT_ON);
2936 if (r < 0)
2937 log_warning_errno(r, "Failed to enable SIGCHLD event source, ignoring: %m");
2938
2939 break;
2940
2941 case SIGTERM:
2942 if (MANAGER_IS_SYSTEM(m)) {
2943 /* This is for compatibility with the original sysvinit */
2944 if (verify_run_space_and_log("Refusing to reexecute") < 0)
2945 break;
2946
2947 m->objective = MANAGER_REEXECUTE;
2948 break;
2949 }
2950
2951 _fallthrough_;
2952 case SIGINT:
2953 if (MANAGER_IS_SYSTEM(m))
2954 manager_handle_ctrl_alt_del(m);
2955 else
2956 manager_start_special(m, SPECIAL_EXIT_TARGET, JOB_REPLACE_IRREVERSIBLY);
2957 break;
2958
2959 case SIGWINCH:
2960 /* This is a nop on non-init */
2961 if (MANAGER_IS_SYSTEM(m))
2962 manager_start_special(m, SPECIAL_KBREQUEST_TARGET, JOB_REPLACE);
2963
2964 break;
2965
2966 case SIGPWR:
2967 /* This is a nop on non-init */
2968 if (MANAGER_IS_SYSTEM(m))
2969 manager_start_special(m, SPECIAL_SIGPWR_TARGET, JOB_REPLACE);
2970
2971 break;
2972
2973 case SIGUSR1:
2974 if (manager_dbus_is_running(m, false)) {
2975 log_info("Trying to reconnect to bus...");
2976
2977 (void) bus_init_api(m);
2978
2979 if (MANAGER_IS_SYSTEM(m))
2980 (void) bus_init_system(m);
2981 } else
2982 manager_start_special(m, SPECIAL_DBUS_SERVICE, JOB_REPLACE);
2983
2984 break;
2985
2986 case SIGUSR2: {
2987 _cleanup_free_ char *dump = NULL;
2988
2989 r = manager_get_dump_string(m, /* patterns= */ NULL, &dump);
2990 if (r < 0) {
2991 log_warning_errno(errno, "Failed to acquire manager dump: %m");
2992 break;
2993 }
2994
2995 log_dump(LOG_INFO, dump);
2996 break;
2997 }
2998
2999 case SIGHUP:
3000 if (verify_run_space_and_log("Refusing to reload") < 0)
3001 break;
3002
3003 m->objective = MANAGER_RELOAD;
3004 break;
3005
3006 default: {
3007
3008 /* Starting SIGRTMIN+0 */
3009 static const struct {
3010 const char *target;
3011 JobMode mode;
3012 } target_table[] = {
3013 [0] = { SPECIAL_DEFAULT_TARGET, JOB_ISOLATE },
3014 [1] = { SPECIAL_RESCUE_TARGET, JOB_ISOLATE },
3015 [2] = { SPECIAL_EMERGENCY_TARGET, JOB_ISOLATE },
3016 [3] = { SPECIAL_HALT_TARGET, JOB_REPLACE_IRREVERSIBLY },
3017 [4] = { SPECIAL_POWEROFF_TARGET, JOB_REPLACE_IRREVERSIBLY },
3018 [5] = { SPECIAL_REBOOT_TARGET, JOB_REPLACE_IRREVERSIBLY },
3019 [6] = { SPECIAL_KEXEC_TARGET, JOB_REPLACE_IRREVERSIBLY },
3020 [7] = { SPECIAL_SOFT_REBOOT_TARGET, JOB_REPLACE_IRREVERSIBLY },
3021 };
3022
3023 /* Starting SIGRTMIN+13, so that target halt and system halt are 10 apart */
3024 static const ManagerObjective objective_table[] = {
3025 [0] = MANAGER_HALT,
3026 [1] = MANAGER_POWEROFF,
3027 [2] = MANAGER_REBOOT,
3028 [3] = MANAGER_KEXEC,
3029 [4] = MANAGER_SOFT_REBOOT,
3030 };
3031
3032 if ((int) sfsi.ssi_signo >= SIGRTMIN+0 &&
3033 (int) sfsi.ssi_signo < SIGRTMIN+(int) ELEMENTSOF(target_table)) {
3034 int idx = (int) sfsi.ssi_signo - SIGRTMIN;
3035 manager_start_special(m, target_table[idx].target, target_table[idx].mode);
3036 break;
3037 }
3038
3039 if ((int) sfsi.ssi_signo >= SIGRTMIN+13 &&
3040 (int) sfsi.ssi_signo < SIGRTMIN+13+(int) ELEMENTSOF(objective_table)) {
3041 m->objective = objective_table[sfsi.ssi_signo - SIGRTMIN - 13];
3042 break;
3043 }
3044
3045 switch (sfsi.ssi_signo - SIGRTMIN) {
3046
3047 case 18: {
3048 bool generic = false;
3049
3050 if (sfsi.ssi_code != SI_QUEUE)
3051 generic = true;
3052 else {
3053 /* Override a few select commands by our own PID1-specific logic */
3054
3055 switch (sfsi.ssi_int) {
3056
3057 case _COMMON_SIGNAL_COMMAND_LOG_LEVEL_BASE..._COMMON_SIGNAL_COMMAND_LOG_LEVEL_END:
3058 manager_override_log_level(m, sfsi.ssi_int - _COMMON_SIGNAL_COMMAND_LOG_LEVEL_BASE);
3059 break;
3060
3061 case COMMON_SIGNAL_COMMAND_CONSOLE:
3062 manager_override_log_target(m, LOG_TARGET_CONSOLE);
3063 break;
3064
3065 case COMMON_SIGNAL_COMMAND_JOURNAL:
3066 manager_override_log_target(m, LOG_TARGET_JOURNAL);
3067 break;
3068
3069 case COMMON_SIGNAL_COMMAND_KMSG:
3070 manager_override_log_target(m, LOG_TARGET_KMSG);
3071 break;
3072
3073 case COMMON_SIGNAL_COMMAND_NULL:
3074 manager_override_log_target(m, LOG_TARGET_NULL);
3075 break;
3076
3077 case MANAGER_SIGNAL_COMMAND_DUMP_JOBS: {
3078 _cleanup_free_ char *dump_jobs = NULL;
3079
3080 r = manager_get_dump_jobs_string(m, /* patterns= */ NULL, " ", &dump_jobs);
3081 if (r < 0) {
3082 log_warning_errno(errno, "Failed to acquire manager jobs dump: %m");
3083 break;
3084 }
3085
3086 log_dump(LOG_INFO, dump_jobs);
3087 break;
3088 }
3089
3090 default:
3091 generic = true;
3092 }
3093 }
3094
3095 if (generic)
3096 return sigrtmin18_handler(source, &sfsi, NULL);
3097
3098 break;
3099 }
3100
3101 case 20:
3102 manager_override_show_status(m, SHOW_STATUS_YES, "signal");
3103 break;
3104
3105 case 21:
3106 manager_override_show_status(m, SHOW_STATUS_NO, "signal");
3107 break;
3108
3109 case 22:
3110 manager_override_log_level(m, LOG_DEBUG);
3111 break;
3112
3113 case 23:
3114 manager_restore_original_log_level(m);
3115 break;
3116
3117 case 24:
3118 if (MANAGER_IS_USER(m)) {
3119 m->objective = MANAGER_EXIT;
3120 return 0;
3121 }
3122
3123 /* This is a nop on init */
3124 break;
3125
3126 case 25:
3127 m->objective = MANAGER_REEXECUTE;
3128 break;
3129
3130 case 26:
3131 case 29: /* compatibility: used to be mapped to LOG_TARGET_SYSLOG_OR_KMSG */
3132 manager_restore_original_log_target(m);
3133 break;
3134
3135 case 27:
3136 manager_override_log_target(m, LOG_TARGET_CONSOLE);
3137 break;
3138
3139 case 28:
3140 manager_override_log_target(m, LOG_TARGET_KMSG);
3141 break;
3142
3143 default:
3144 log_warning("Got unhandled signal <%s>.", signal_to_string(sfsi.ssi_signo));
3145 }
3146 }}
3147
3148 return 0;
3149 }
3150
3151 static int manager_dispatch_time_change_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
3152 Manager *m = ASSERT_PTR(userdata);
3153 Unit *u;
3154
3155 log_struct(LOG_DEBUG,
3156 "MESSAGE_ID=" SD_MESSAGE_TIME_CHANGE_STR,
3157 LOG_MESSAGE("Time has been changed"));
3158
3159 /* Restart the watch */
3160 (void) manager_setup_time_change(m);
3161
3162 HASHMAP_FOREACH(u, m->units)
3163 if (UNIT_VTABLE(u)->time_change)
3164 UNIT_VTABLE(u)->time_change(u);
3165
3166 return 0;
3167 }
3168
3169 static int manager_dispatch_timezone_change(
3170 sd_event_source *source,
3171 const struct inotify_event *e,
3172 void *userdata) {
3173
3174 Manager *m = ASSERT_PTR(userdata);
3175 int changed;
3176 Unit *u;
3177
3178 log_debug("inotify event for /etc/localtime");
3179
3180 changed = manager_read_timezone_stat(m);
3181 if (changed <= 0)
3182 return changed;
3183
3184 /* Something changed, restart the watch, to ensure we watch the new /etc/localtime if it changed */
3185 (void) manager_setup_timezone_change(m);
3186
3187 /* Read the new timezone */
3188 tzset();
3189
3190 log_debug("Timezone has been changed (now: %s).", tzname[daylight]);
3191
3192 HASHMAP_FOREACH(u, m->units)
3193 if (UNIT_VTABLE(u)->timezone_change)
3194 UNIT_VTABLE(u)->timezone_change(u);
3195
3196 return 0;
3197 }
3198
3199 static int manager_dispatch_idle_pipe_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
3200 Manager *m = ASSERT_PTR(userdata);
3201
3202 assert(m->idle_pipe[2] == fd);
3203
3204 /* There's at least one Type=idle child that just gave up on us waiting for the boot process to
3205 * complete. Let's now turn off any further console output if there's at least one service that needs
3206 * console access, so that from now on our own output should not spill into that service's output
3207 * anymore. After all, we support Type=idle only to beautify console output and it generally is set
3208 * on services that want to own the console exclusively without our interference. */
3209 m->no_console_output = m->n_on_console > 0;
3210
3211 /* Acknowledge the child's request, and let all other children know too that they shouldn't wait
3212 * any longer by closing the pipes towards them, which is what they are waiting for. */
3213 manager_close_idle_pipe(m);
3214
3215 return 0;
3216 }
3217
3218 static int manager_dispatch_jobs_in_progress(sd_event_source *source, usec_t usec, void *userdata) {
3219 Manager *m = ASSERT_PTR(userdata);
3220 int r;
3221
3222 assert(source);
3223
3224 manager_print_jobs_in_progress(m);
3225
3226 r = sd_event_source_set_time_relative(source, JOBS_IN_PROGRESS_PERIOD_USEC);
3227 if (r < 0)
3228 return r;
3229
3230 return sd_event_source_set_enabled(source, SD_EVENT_ONESHOT);
3231 }
3232
3233 int manager_loop(Manager *m) {
3234 RateLimit rl = { .interval = 1*USEC_PER_SEC, .burst = 50000 };
3235 int r;
3236
3237 assert(m);
3238 assert(m->objective == MANAGER_OK); /* Ensure manager_startup() has been called */
3239
3240 manager_check_finished(m);
3241
3242 /* There might still be some zombies hanging around from before we were exec()'ed. Let's reap them. */
3243 r = sd_event_source_set_enabled(m->sigchld_event_source, SD_EVENT_ON);
3244 if (r < 0)
3245 return log_error_errno(r, "Failed to enable SIGCHLD event source: %m");
3246
3247 while (m->objective == MANAGER_OK) {
3248
3249 (void) watchdog_ping();
3250
3251 if (!ratelimit_below(&rl)) {
3252 /* Yay, something is going seriously wrong, pause a little */
3253 log_warning("Looping too fast. Throttling execution a little.");
3254 sleep(1);
3255 }
3256
3257 if (manager_dispatch_load_queue(m) > 0)
3258 continue;
3259
3260 if (manager_dispatch_gc_job_queue(m) > 0)
3261 continue;
3262
3263 if (manager_dispatch_gc_unit_queue(m) > 0)
3264 continue;
3265
3266 if (manager_dispatch_cleanup_queue(m) > 0)
3267 continue;
3268
3269 if (manager_dispatch_cgroup_realize_queue(m) > 0)
3270 continue;
3271
3272 if (manager_dispatch_start_when_upheld_queue(m) > 0)
3273 continue;
3274
3275 if (manager_dispatch_stop_when_bound_queue(m) > 0)
3276 continue;
3277
3278 if (manager_dispatch_stop_when_unneeded_queue(m) > 0)
3279 continue;
3280
3281 if (manager_dispatch_release_resources_queue(m) > 0)
3282 continue;
3283
3284 if (manager_dispatch_dbus_queue(m) > 0)
3285 continue;
3286
3287 /* Sleep for watchdog runtime wait time */
3288 r = sd_event_run(m->event, watchdog_runtime_wait());
3289 if (r < 0)
3290 return log_error_errno(r, "Failed to run event loop: %m");
3291 }
3292
3293 return m->objective;
3294 }
3295
3296 int manager_load_unit_from_dbus_path(Manager *m, const char *s, sd_bus_error *e, Unit **_u) {
3297 _cleanup_free_ char *n = NULL;
3298 sd_id128_t invocation_id;
3299 Unit *u;
3300 int r;
3301
3302 assert(m);
3303 assert(s);
3304 assert(_u);
3305
3306 r = unit_name_from_dbus_path(s, &n);
3307 if (r < 0)
3308 return r;
3309
3310 /* Permit addressing units by invocation ID: if the passed bus path is suffixed by a 128-bit ID then
3311 * we use it as invocation ID. */
3312 r = sd_id128_from_string(n, &invocation_id);
3313 if (r >= 0) {
3314 u = hashmap_get(m->units_by_invocation_id, &invocation_id);
3315 if (u) {
3316 *_u = u;
3317 return 0;
3318 }
3319
3320 return sd_bus_error_setf(e, BUS_ERROR_NO_UNIT_FOR_INVOCATION_ID,
3321 "No unit with the specified invocation ID " SD_ID128_FORMAT_STR " known.",
3322 SD_ID128_FORMAT_VAL(invocation_id));
3323 }
3324
3325 /* If this didn't work, we check if this is a unit name */
3326 if (!unit_name_is_valid(n, UNIT_NAME_PLAIN|UNIT_NAME_INSTANCE)) {
3327 _cleanup_free_ char *nn = NULL;
3328
3329 nn = cescape(n);
3330 return sd_bus_error_setf(e, SD_BUS_ERROR_INVALID_ARGS,
3331 "Unit name %s is neither a valid invocation ID nor unit name.", strnull(nn));
3332 }
3333
3334 r = manager_load_unit(m, n, NULL, e, &u);
3335 if (r < 0)
3336 return r;
3337
3338 *_u = u;
3339 return 0;
3340 }
3341
3342 int manager_get_job_from_dbus_path(Manager *m, const char *s, Job **_j) {
3343 const char *p;
3344 unsigned id;
3345 Job *j;
3346 int r;
3347
3348 assert(m);
3349 assert(s);
3350 assert(_j);
3351
3352 p = startswith(s, "/org/freedesktop/systemd1/job/");
3353 if (!p)
3354 return -EINVAL;
3355
3356 r = safe_atou(p, &id);
3357 if (r < 0)
3358 return r;
3359
3360 j = manager_get_job(m, id);
3361 if (!j)
3362 return -ENOENT;
3363
3364 *_j = j;
3365
3366 return 0;
3367 }
3368
3369 void manager_send_unit_audit(Manager *m, Unit *u, int type, bool success) {
3370
3371 #if HAVE_AUDIT
3372 _cleanup_free_ char *p = NULL;
3373 const char *msg;
3374 int audit_fd, r;
3375
3376 if (!MANAGER_IS_SYSTEM(m))
3377 return;
3378
3379 audit_fd = get_audit_fd();
3380 if (audit_fd < 0)
3381 return;
3382
3383 /* Don't generate audit events if the service was already
3384 * started and we're just deserializing */
3385 if (MANAGER_IS_RELOADING(m))
3386 return;
3387
3388 r = unit_name_to_prefix_and_instance(u->id, &p);
3389 if (r < 0) {
3390 log_warning_errno(r, "Failed to extract prefix and instance of unit name, ignoring: %m");
3391 return;
3392 }
3393
3394 msg = strjoina("unit=", p);
3395 if (audit_log_user_comm_message(audit_fd, type, msg, "systemd", NULL, NULL, NULL, success) < 0) {
3396 if (ERRNO_IS_PRIVILEGE(errno)) {
3397 /* We aren't allowed to send audit messages? Then let's not retry again. */
3398 log_debug_errno(errno, "Failed to send audit message, closing audit socket: %m");
3399 close_audit_fd();
3400 } else
3401 log_warning_errno(errno, "Failed to send audit message, ignoring: %m");
3402 }
3403 #endif
3404
3405 }
3406
3407 void manager_send_unit_plymouth(Manager *m, Unit *u) {
3408 _cleanup_free_ char *message = NULL;
3409 int c, r;
3410
3411 /* Don't generate plymouth events if the service was already
3412 * started and we're just deserializing */
3413 if (MANAGER_IS_RELOADING(m))
3414 return;
3415
3416 if (!MANAGER_IS_SYSTEM(m))
3417 return;
3418
3419 if (detect_container() > 0)
3420 return;
3421
3422 if (!UNIT_VTABLE(u)->notify_plymouth)
3423 return;
3424
3425 c = asprintf(&message, "U\x02%c%s%c", (int) (strlen(u->id) + 1), u->id, '\x00');
3426 if (c < 0)
3427 return (void) log_oom();
3428
3429 /* We set SOCK_NONBLOCK here so that we rather drop the message then wait for plymouth */
3430 r = plymouth_send_raw(message, c, SOCK_NONBLOCK);
3431 if (r < 0)
3432 log_full_errno(ERRNO_IS_NO_PLYMOUTH(r) ? LOG_DEBUG : LOG_WARNING, r,
3433 "Failed to communicate with plymouth: %m");
3434 }
3435
3436 usec_t manager_get_watchdog(Manager *m, WatchdogType t) {
3437 assert(m);
3438
3439 if (MANAGER_IS_USER(m))
3440 return USEC_INFINITY;
3441
3442 if (m->watchdog_overridden[t] != USEC_INFINITY)
3443 return m->watchdog_overridden[t];
3444
3445 return m->watchdog[t];
3446 }
3447
3448 void manager_set_watchdog(Manager *m, WatchdogType t, usec_t timeout) {
3449
3450 assert(m);
3451
3452 if (MANAGER_IS_USER(m))
3453 return;
3454
3455 if (m->watchdog[t] == timeout)
3456 return;
3457
3458 if (m->watchdog_overridden[t] == USEC_INFINITY) {
3459 if (t == WATCHDOG_RUNTIME)
3460 (void) watchdog_setup(timeout);
3461 else if (t == WATCHDOG_PRETIMEOUT)
3462 (void) watchdog_setup_pretimeout(timeout);
3463 }
3464
3465 m->watchdog[t] = timeout;
3466 }
3467
3468 void manager_override_watchdog(Manager *m, WatchdogType t, usec_t timeout) {
3469 usec_t usec;
3470
3471 assert(m);
3472
3473 if (MANAGER_IS_USER(m))
3474 return;
3475
3476 if (m->watchdog_overridden[t] == timeout)
3477 return;
3478
3479 usec = timeout == USEC_INFINITY ? m->watchdog[t] : timeout;
3480 if (t == WATCHDOG_RUNTIME)
3481 (void) watchdog_setup(usec);
3482 else if (t == WATCHDOG_PRETIMEOUT)
3483 (void) watchdog_setup_pretimeout(usec);
3484
3485 m->watchdog_overridden[t] = timeout;
3486 }
3487
3488 int manager_set_watchdog_pretimeout_governor(Manager *m, const char *governor) {
3489 _cleanup_free_ char *p = NULL;
3490 int r;
3491
3492 assert(m);
3493
3494 if (MANAGER_IS_USER(m))
3495 return 0;
3496
3497 if (streq_ptr(m->watchdog_pretimeout_governor, governor))
3498 return 0;
3499
3500 p = strdup(governor);
3501 if (!p)
3502 return -ENOMEM;
3503
3504 r = watchdog_setup_pretimeout_governor(governor);
3505 if (r < 0)
3506 return r;
3507
3508 return free_and_replace(m->watchdog_pretimeout_governor, p);
3509 }
3510
3511 int manager_override_watchdog_pretimeout_governor(Manager *m, const char *governor) {
3512 _cleanup_free_ char *p = NULL;
3513 int r;
3514
3515 assert(m);
3516
3517 if (MANAGER_IS_USER(m))
3518 return 0;
3519
3520 if (streq_ptr(m->watchdog_pretimeout_governor_overridden, governor))
3521 return 0;
3522
3523 p = strdup(governor);
3524 if (!p)
3525 return -ENOMEM;
3526
3527 r = watchdog_setup_pretimeout_governor(governor);
3528 if (r < 0)
3529 return r;
3530
3531 return free_and_replace(m->watchdog_pretimeout_governor_overridden, p);
3532 }
3533
3534 int manager_reload(Manager *m) {
3535 _unused_ _cleanup_(manager_reloading_stopp) Manager *reloading = NULL;
3536 _cleanup_fdset_free_ FDSet *fds = NULL;
3537 _cleanup_fclose_ FILE *f = NULL;
3538 int r;
3539
3540 assert(m);
3541
3542 r = manager_open_serialization(m, &f);
3543 if (r < 0)
3544 return log_error_errno(r, "Failed to create serialization file: %m");
3545
3546 fds = fdset_new();
3547 if (!fds)
3548 return log_oom();
3549
3550 /* We are officially in reload mode from here on. */
3551 reloading = manager_reloading_start(m);
3552
3553 r = manager_serialize(m, f, fds, false);
3554 if (r < 0)
3555 return r;
3556
3557 if (fseeko(f, 0, SEEK_SET) < 0)
3558 return log_error_errno(errno, "Failed to seek to beginning of serialization: %m");
3559
3560 /* 💀 This is the point of no return, from here on there is no way back. 💀 */
3561 reloading = NULL;
3562
3563 bus_manager_send_reloading(m, true);
3564
3565 /* Start by flushing out all jobs and units, all generated units, all runtime environments, all dynamic users
3566 * and everything else that is worth flushing out. We'll get it all back from the serialization — if we need
3567 * it. */
3568
3569 manager_clear_jobs_and_units(m);
3570 lookup_paths_flush_generator(&m->lookup_paths);
3571 lookup_paths_free(&m->lookup_paths);
3572 exec_shared_runtime_vacuum(m);
3573 dynamic_user_vacuum(m, false);
3574 m->uid_refs = hashmap_free(m->uid_refs);
3575 m->gid_refs = hashmap_free(m->gid_refs);
3576
3577 r = lookup_paths_init_or_warn(&m->lookup_paths, m->runtime_scope, 0, NULL);
3578 if (r < 0)
3579 return r;
3580
3581 (void) manager_run_environment_generators(m);
3582 (void) manager_run_generators(m);
3583
3584 lookup_paths_log(&m->lookup_paths);
3585
3586 /* We flushed out generated files, for which we don't watch mtime, so we should flush the old map. */
3587 manager_free_unit_name_maps(m);
3588 m->unit_file_state_outdated = false;
3589
3590 /* First, enumerate what we can from kernel and suchlike */
3591 manager_enumerate_perpetual(m);
3592 manager_enumerate(m);
3593
3594 /* Second, deserialize our stored data */
3595 r = manager_deserialize(m, f, fds);
3596 if (r < 0)
3597 log_warning_errno(r, "Deserialization failed, proceeding anyway: %m");
3598
3599 /* We don't need the serialization anymore */
3600 f = safe_fclose(f);
3601
3602 /* Re-register notify_fd as event source, and set up other sockets/communication channels we might need */
3603 (void) manager_setup_notify(m);
3604 (void) manager_setup_cgroups_agent(m);
3605 (void) manager_setup_user_lookup_fd(m);
3606
3607 /* Third, fire things up! */
3608 manager_coldplug(m);
3609
3610 /* Clean up runtime objects no longer referenced */
3611 manager_vacuum(m);
3612
3613 /* Clean up deserialized tracked clients */
3614 m->deserialized_subscribed = strv_free(m->deserialized_subscribed);
3615
3616 /* Consider the reload process complete now. */
3617 assert(m->n_reloading > 0);
3618 m->n_reloading--;
3619
3620 manager_ready(m);
3621
3622 m->send_reloading_done = true;
3623 return 0;
3624 }
3625
3626 void manager_reset_failed(Manager *m) {
3627 Unit *u;
3628
3629 assert(m);
3630
3631 HASHMAP_FOREACH(u, m->units)
3632 unit_reset_failed(u);
3633 }
3634
3635 bool manager_unit_inactive_or_pending(Manager *m, const char *name) {
3636 Unit *u;
3637
3638 assert(m);
3639 assert(name);
3640
3641 /* Returns true if the unit is inactive or going down */
3642 u = manager_get_unit(m, name);
3643 if (!u)
3644 return true;
3645
3646 return unit_inactive_or_pending(u);
3647 }
3648
3649 static void log_taint_string(Manager *m) {
3650 _cleanup_free_ char *taint = NULL;
3651
3652 assert(m);
3653
3654 if (MANAGER_IS_USER(m) || m->taint_logged)
3655 return;
3656
3657 m->taint_logged = true; /* only check for taint once */
3658
3659 taint = manager_taint_string(m);
3660 if (isempty(taint))
3661 return;
3662
3663 log_struct(LOG_NOTICE,
3664 LOG_MESSAGE("System is tainted: %s", taint),
3665 "TAINT=%s", taint,
3666 "MESSAGE_ID=" SD_MESSAGE_TAINTED_STR);
3667 }
3668
3669 static void manager_notify_finished(Manager *m) {
3670 usec_t firmware_usec, loader_usec, kernel_usec, initrd_usec, userspace_usec, total_usec;
3671
3672 if (MANAGER_IS_TEST_RUN(m))
3673 return;
3674
3675 if (MANAGER_IS_SYSTEM(m) && detect_container() <= 0) {
3676 char buf[FORMAT_TIMESPAN_MAX + STRLEN(" (firmware) + ") + FORMAT_TIMESPAN_MAX + STRLEN(" (loader) + ")]
3677 = {};
3678 char *p = buf;
3679 size_t size = sizeof buf;
3680
3681 /* Note that MANAGER_TIMESTAMP_KERNEL's monotonic value is always at 0, and
3682 * MANAGER_TIMESTAMP_FIRMWARE's and MANAGER_TIMESTAMP_LOADER's monotonic value should be considered
3683 * negative values. */
3684
3685 firmware_usec = m->timestamps[MANAGER_TIMESTAMP_FIRMWARE].monotonic - m->timestamps[MANAGER_TIMESTAMP_LOADER].monotonic;
3686 loader_usec = m->timestamps[MANAGER_TIMESTAMP_LOADER].monotonic - m->timestamps[MANAGER_TIMESTAMP_KERNEL].monotonic;
3687 userspace_usec = m->timestamps[MANAGER_TIMESTAMP_FINISH].monotonic - m->timestamps[MANAGER_TIMESTAMP_USERSPACE].monotonic;
3688 total_usec = m->timestamps[MANAGER_TIMESTAMP_FIRMWARE].monotonic + m->timestamps[MANAGER_TIMESTAMP_FINISH].monotonic;
3689
3690 if (firmware_usec > 0)
3691 size = strpcpyf(&p, size, "%s (firmware) + ", FORMAT_TIMESPAN(firmware_usec, USEC_PER_MSEC));
3692 if (loader_usec > 0)
3693 size = strpcpyf(&p, size, "%s (loader) + ", FORMAT_TIMESPAN(loader_usec, USEC_PER_MSEC));
3694
3695 if (dual_timestamp_is_set(&m->timestamps[MANAGER_TIMESTAMP_INITRD])) {
3696
3697 /* The initrd case on bare-metal */
3698 kernel_usec = m->timestamps[MANAGER_TIMESTAMP_INITRD].monotonic - m->timestamps[MANAGER_TIMESTAMP_KERNEL].monotonic;
3699 initrd_usec = m->timestamps[MANAGER_TIMESTAMP_USERSPACE].monotonic - m->timestamps[MANAGER_TIMESTAMP_INITRD].monotonic;
3700
3701 log_struct(LOG_INFO,
3702 "MESSAGE_ID=" SD_MESSAGE_STARTUP_FINISHED_STR,
3703 "KERNEL_USEC="USEC_FMT, kernel_usec,
3704 "INITRD_USEC="USEC_FMT, initrd_usec,
3705 "USERSPACE_USEC="USEC_FMT, userspace_usec,
3706 LOG_MESSAGE("Startup finished in %s%s (kernel) + %s (initrd) + %s (userspace) = %s.",
3707 buf,
3708 FORMAT_TIMESPAN(kernel_usec, USEC_PER_MSEC),
3709 FORMAT_TIMESPAN(initrd_usec, USEC_PER_MSEC),
3710 FORMAT_TIMESPAN(userspace_usec, USEC_PER_MSEC),
3711 FORMAT_TIMESPAN(total_usec, USEC_PER_MSEC)));
3712 } else {
3713 /* The initrd-less case on bare-metal */
3714
3715 kernel_usec = m->timestamps[MANAGER_TIMESTAMP_USERSPACE].monotonic - m->timestamps[MANAGER_TIMESTAMP_KERNEL].monotonic;
3716 initrd_usec = 0;
3717
3718 log_struct(LOG_INFO,
3719 "MESSAGE_ID=" SD_MESSAGE_STARTUP_FINISHED_STR,
3720 "KERNEL_USEC="USEC_FMT, kernel_usec,
3721 "USERSPACE_USEC="USEC_FMT, userspace_usec,
3722 LOG_MESSAGE("Startup finished in %s%s (kernel) + %s (userspace) = %s.",
3723 buf,
3724 FORMAT_TIMESPAN(kernel_usec, USEC_PER_MSEC),
3725 FORMAT_TIMESPAN(userspace_usec, USEC_PER_MSEC),
3726 FORMAT_TIMESPAN(total_usec, USEC_PER_MSEC)));
3727 }
3728 } else {
3729 /* The container and --user case */
3730 firmware_usec = loader_usec = initrd_usec = kernel_usec = 0;
3731 total_usec = userspace_usec = m->timestamps[MANAGER_TIMESTAMP_FINISH].monotonic - m->timestamps[MANAGER_TIMESTAMP_USERSPACE].monotonic;
3732
3733 log_struct(LOG_INFO,
3734 "MESSAGE_ID=" SD_MESSAGE_USER_STARTUP_FINISHED_STR,
3735 "USERSPACE_USEC="USEC_FMT, userspace_usec,
3736 LOG_MESSAGE("Startup finished in %s.",
3737 FORMAT_TIMESPAN(total_usec, USEC_PER_MSEC)));
3738 }
3739
3740 bus_manager_send_finished(m, firmware_usec, loader_usec, kernel_usec, initrd_usec, userspace_usec, total_usec);
3741
3742 log_taint_string(m);
3743 }
3744
3745 static void user_manager_send_ready(Manager *m) {
3746 int r;
3747
3748 assert(m);
3749
3750 /* We send READY=1 on reaching basic.target only when running in --user mode. */
3751 if (!MANAGER_IS_USER(m) || m->ready_sent)
3752 return;
3753
3754 r = sd_notify(false,
3755 "READY=1\n"
3756 "STATUS=Reached " SPECIAL_BASIC_TARGET ".");
3757 if (r < 0)
3758 log_warning_errno(r, "Failed to send readiness notification, ignoring: %m");
3759
3760 m->ready_sent = true;
3761 m->status_ready = false;
3762 }
3763
3764 static void manager_send_ready(Manager *m) {
3765 int r;
3766
3767 if (m->ready_sent && m->status_ready)
3768 /* Skip the notification if nothing changed. */
3769 return;
3770
3771 r = sd_notify(false,
3772 "READY=1\n"
3773 "STATUS=Ready.");
3774 if (r < 0)
3775 log_full_errno(m->ready_sent ? LOG_DEBUG : LOG_WARNING, r,
3776 "Failed to send readiness notification, ignoring: %m");
3777
3778 m->ready_sent = m->status_ready = true;
3779 }
3780
3781 static void manager_check_basic_target(Manager *m) {
3782 Unit *u;
3783
3784 assert(m);
3785
3786 /* Small shortcut */
3787 if (m->ready_sent && m->taint_logged)
3788 return;
3789
3790 u = manager_get_unit(m, SPECIAL_BASIC_TARGET);
3791 if (!u || !UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u)))
3792 return;
3793
3794 /* For user managers, send out READY=1 as soon as we reach basic.target */
3795 user_manager_send_ready(m);
3796
3797 /* Log the taint string as soon as we reach basic.target */
3798 log_taint_string(m);
3799 }
3800
3801 void manager_check_finished(Manager *m) {
3802 assert(m);
3803
3804 if (MANAGER_IS_RELOADING(m))
3805 return;
3806
3807 /* Verify that we have entered the event loop already, and not left it again. */
3808 if (!MANAGER_IS_RUNNING(m))
3809 return;
3810
3811 manager_check_basic_target(m);
3812
3813 if (!hashmap_isempty(m->jobs)) {
3814 if (m->jobs_in_progress_event_source)
3815 /* Ignore any failure, this is only for feedback */
3816 (void) sd_event_source_set_time(m->jobs_in_progress_event_source,
3817 manager_watch_jobs_next_time(m));
3818 return;
3819 }
3820
3821 /* The jobs hashmap tends to grow a lot during boot, and then it's not reused until shutdown. Let's
3822 kill the hashmap if it is relatively large. */
3823 if (hashmap_buckets(m->jobs) > hashmap_size(m->units) / 10)
3824 m->jobs = hashmap_free(m->jobs);
3825
3826 manager_send_ready(m);
3827
3828 /* Notify Type=idle units that we are done now */
3829 manager_close_idle_pipe(m);
3830
3831 if (MANAGER_IS_FINISHED(m))
3832 return;
3833
3834 manager_flip_auto_status(m, false, "boot finished");
3835
3836 /* Turn off confirm spawn now */
3837 m->confirm_spawn = NULL;
3838
3839 /* No need to update ask password status when we're going non-interactive */
3840 manager_close_ask_password(m);
3841
3842 /* This is no longer the first boot */
3843 manager_set_first_boot(m, false);
3844
3845 dual_timestamp_now(m->timestamps + MANAGER_TIMESTAMP_FINISH);
3846
3847 manager_notify_finished(m);
3848
3849 manager_invalidate_startup_units(m);
3850 }
3851
3852 void manager_send_reloading(Manager *m) {
3853 assert(m);
3854
3855 /* Let whoever invoked us know that we are now reloading */
3856 (void) sd_notifyf(/* unset= */ false,
3857 "RELOADING=1\n"
3858 "MONOTONIC_USEC=" USEC_FMT "\n", now(CLOCK_MONOTONIC));
3859
3860 /* And ensure that we'll send READY=1 again as soon as we are ready again */
3861 m->ready_sent = false;
3862 }
3863
3864 static bool generator_path_any(const char* const* paths) {
3865 bool found = false;
3866
3867 /* Optimize by skipping the whole process by not creating output directories
3868 * if no generators are found. */
3869 STRV_FOREACH(path, paths)
3870 if (access(*path, F_OK) == 0)
3871 found = true;
3872 else if (errno != ENOENT)
3873 log_warning_errno(errno, "Failed to open generator directory %s: %m", *path);
3874
3875 return found;
3876 }
3877
3878 static int manager_run_environment_generators(Manager *m) {
3879 char **tmp = NULL; /* this is only used in the forked process, no cleanup here */
3880 _cleanup_strv_free_ char **paths = NULL;
3881 void* args[] = {
3882 [STDOUT_GENERATE] = &tmp,
3883 [STDOUT_COLLECT] = &tmp,
3884 [STDOUT_CONSUME] = &m->transient_environment,
3885 };
3886 int r;
3887
3888 if (MANAGER_IS_TEST_RUN(m) && !(m->test_run_flags & MANAGER_TEST_RUN_ENV_GENERATORS))
3889 return 0;
3890
3891 paths = env_generator_binary_paths(m->runtime_scope);
3892 if (!paths)
3893 return log_oom();
3894
3895 if (!generator_path_any((const char* const*) paths))
3896 return 0;
3897
3898 WITH_UMASK(0022)
3899 r = execute_directories((const char* const*) paths, DEFAULT_TIMEOUT_USEC, gather_environment,
3900 args, NULL, m->transient_environment,
3901 EXEC_DIR_PARALLEL | EXEC_DIR_IGNORE_ERRORS | EXEC_DIR_SET_SYSTEMD_EXEC_PID);
3902 return r;
3903 }
3904
3905 static int build_generator_environment(Manager *m, char ***ret) {
3906 _cleanup_strv_free_ char **nl = NULL;
3907 Virtualization v;
3908 ConfidentialVirtualization cv;
3909 int r;
3910
3911 assert(m);
3912 assert(ret);
3913
3914 /* Generators oftentimes want to know some basic facts about the environment they run in, in order to
3915 * adjust generated units to that. Let's pass down some bits of information that are easy for us to
3916 * determine (but a bit harder for generator scripts to determine), as environment variables. */
3917
3918 nl = strv_copy(m->transient_environment);
3919 if (!nl)
3920 return -ENOMEM;
3921
3922 r = strv_env_assign(&nl, "SYSTEMD_SCOPE", runtime_scope_to_string(m->runtime_scope));
3923 if (r < 0)
3924 return r;
3925
3926 if (MANAGER_IS_SYSTEM(m)) {
3927 /* Note that $SYSTEMD_IN_INITRD may be used to override the initrd detection in much of our
3928 * codebase. This is hence more than purely informational. It will shortcut detection of the
3929 * initrd state if generators invoke our own tools. But that's OK, as it would come to the
3930 * same results (hopefully). */
3931 r = strv_env_assign(&nl, "SYSTEMD_IN_INITRD", one_zero(in_initrd()));
3932 if (r < 0)
3933 return r;
3934
3935 if (m->first_boot >= 0) {
3936 r = strv_env_assign(&nl, "SYSTEMD_FIRST_BOOT", one_zero(m->first_boot));
3937 if (r < 0)
3938 return r;
3939 }
3940 }
3941
3942 v = detect_virtualization();
3943 if (v < 0)
3944 log_debug_errno(v, "Failed to detect virtualization, ignoring: %m");
3945 else if (v > 0) {
3946 const char *s;
3947
3948 s = strjoina(VIRTUALIZATION_IS_VM(v) ? "vm:" :
3949 VIRTUALIZATION_IS_CONTAINER(v) ? "container:" : ":",
3950 virtualization_to_string(v));
3951
3952 r = strv_env_assign(&nl, "SYSTEMD_VIRTUALIZATION", s);
3953 if (r < 0)
3954 return r;
3955 }
3956
3957 cv = detect_confidential_virtualization();
3958 if (cv < 0)
3959 log_debug_errno(cv, "Failed to detect confidential virtualization, ignoring: %m");
3960 else if (cv > 0) {
3961 r = strv_env_assign(&nl, "SYSTEMD_CONFIDENTIAL_VIRTUALIZATION", confidential_virtualization_to_string(cv));
3962 if (r < 0)
3963 return r;
3964 }
3965
3966 r = strv_env_assign(&nl, "SYSTEMD_ARCHITECTURE", architecture_to_string(uname_architecture()));
3967 if (r < 0)
3968 return r;
3969
3970 *ret = TAKE_PTR(nl);
3971 return 0;
3972 }
3973
3974 static int manager_execute_generators(Manager *m, char **paths, bool remount_ro) {
3975 _cleanup_strv_free_ char **ge = NULL;
3976 const char *argv[] = {
3977 NULL, /* Leave this empty, execute_directory() will fill something in */
3978 m->lookup_paths.generator,
3979 m->lookup_paths.generator_early,
3980 m->lookup_paths.generator_late,
3981 NULL,
3982 };
3983 int r;
3984
3985 r = build_generator_environment(m, &ge);
3986 if (r < 0)
3987 return log_error_errno(r, "Failed to build generator environment: %m");
3988
3989 if (remount_ro) {
3990 /* Remount most of the filesystem tree read-only. We leave /sys/ as-is, because our code
3991 * checks whether it is read-only to detect containerized execution environments. We leave
3992 * /run/ as-is too, because that's where our output goes. We also leave /proc/ and /dev/shm/
3993 * because they're API, and /tmp/ that safe_fork() mounted for us.
3994 */
3995 r = bind_remount_recursive("/", MS_RDONLY, MS_RDONLY,
3996 STRV_MAKE("/sys", "/run", "/proc", "/dev/shm", "/tmp"));
3997 if (r < 0)
3998 log_warning_errno(r, "Read-only bind remount failed, ignoring: %m");
3999 }
4000
4001 BLOCK_WITH_UMASK(0022);
4002 return execute_directories(
4003 (const char* const*) paths,
4004 DEFAULT_TIMEOUT_USEC,
4005 /* callbacks= */ NULL, /* callback_args= */ NULL,
4006 (char**) argv,
4007 ge,
4008 EXEC_DIR_PARALLEL | EXEC_DIR_IGNORE_ERRORS | EXEC_DIR_SET_SYSTEMD_EXEC_PID);
4009 }
4010
4011 static int manager_run_generators(Manager *m) {
4012 ForkFlags flags = FORK_RESET_SIGNALS | FORK_WAIT | FORK_NEW_MOUNTNS | FORK_MOUNTNS_SLAVE;
4013 _cleanup_strv_free_ char **paths = NULL;
4014 int r;
4015
4016 assert(m);
4017
4018 if (MANAGER_IS_TEST_RUN(m) && !(m->test_run_flags & MANAGER_TEST_RUN_GENERATORS))
4019 return 0;
4020
4021 paths = generator_binary_paths(m->runtime_scope);
4022 if (!paths)
4023 return log_oom();
4024
4025 if (!generator_path_any((const char* const*) paths))
4026 return 0;
4027
4028 r = lookup_paths_mkdir_generator(&m->lookup_paths);
4029 if (r < 0) {
4030 log_error_errno(r, "Failed to create generator directories: %m");
4031 goto finish;
4032 }
4033
4034 /* If we are the system manager, we fork and invoke the generators in a sanitized mount namespace. If
4035 * we are the user manager, let's just execute the generators directly. We might not have the
4036 * necessary privileges, and the system manager has already mounted /tmp/ and everything else for us.
4037 */
4038 if (MANAGER_IS_USER(m)) {
4039 r = manager_execute_generators(m, paths, /* remount_ro= */ false);
4040 goto finish;
4041 }
4042
4043 /* On some systems /tmp/ doesn't exist, and on some other systems we cannot create it at all. Avoid
4044 * trying to mount a private tmpfs on it as there's no one size fits all. */
4045 if (is_dir("/tmp", /* follow= */ false) > 0)
4046 flags |= FORK_PRIVATE_TMP;
4047
4048 r = safe_fork("(sd-gens)", flags, NULL);
4049 if (r == 0) {
4050 r = manager_execute_generators(m, paths, /* remount_ro= */ true);
4051 _exit(r >= 0 ? EXIT_SUCCESS : EXIT_FAILURE);
4052 }
4053 if (r < 0) {
4054 if (!ERRNO_IS_PRIVILEGE(r) && r != -EINVAL) {
4055 log_error_errno(r, "Failed to fork off sandboxing environment for executing generators: %m");
4056 goto finish;
4057 }
4058
4059 /* Failed to fork with new mount namespace? Maybe, running in a container environment with
4060 * seccomp or without capability.
4061 *
4062 * We also allow -EINVAL to allow running without CLONE_NEWNS.
4063 *
4064 * Also, when running on non-native userland architecture via systemd-nspawn and
4065 * qemu-user-static QEMU-emulator, clone() with CLONE_NEWNS fails with EINVAL, see
4066 * https://github.com/systemd/systemd/issues/28901.
4067 */
4068 log_debug_errno(r,
4069 "Failed to fork off sandboxing environment for executing generators. "
4070 "Falling back to execute generators without sandboxing: %m");
4071 r = manager_execute_generators(m, paths, /* remount_ro= */ false);
4072 }
4073
4074 finish:
4075 lookup_paths_trim_generator(&m->lookup_paths);
4076 return r;
4077 }
4078
4079 int manager_transient_environment_add(Manager *m, char **plus) {
4080 char **a;
4081
4082 assert(m);
4083
4084 if (strv_isempty(plus))
4085 return 0;
4086
4087 a = strv_env_merge(m->transient_environment, plus);
4088 if (!a)
4089 return log_oom();
4090
4091 sanitize_environment(a);
4092
4093 return strv_free_and_replace(m->transient_environment, a);
4094 }
4095
4096 int manager_client_environment_modify(
4097 Manager *m,
4098 char **minus,
4099 char **plus) {
4100
4101 char **a = NULL, **b = NULL, **l;
4102
4103 assert(m);
4104
4105 if (strv_isempty(minus) && strv_isempty(plus))
4106 return 0;
4107
4108 l = m->client_environment;
4109
4110 if (!strv_isempty(minus)) {
4111 a = strv_env_delete(l, 1, minus);
4112 if (!a)
4113 return -ENOMEM;
4114
4115 l = a;
4116 }
4117
4118 if (!strv_isempty(plus)) {
4119 b = strv_env_merge(l, plus);
4120 if (!b) {
4121 strv_free(a);
4122 return -ENOMEM;
4123 }
4124
4125 l = b;
4126 }
4127
4128 if (m->client_environment != l)
4129 strv_free(m->client_environment);
4130
4131 if (a != l)
4132 strv_free(a);
4133 if (b != l)
4134 strv_free(b);
4135
4136 m->client_environment = sanitize_environment(l);
4137 return 0;
4138 }
4139
4140 int manager_get_effective_environment(Manager *m, char ***ret) {
4141 char **l;
4142
4143 assert(m);
4144 assert(ret);
4145
4146 l = strv_env_merge(m->transient_environment, m->client_environment);
4147 if (!l)
4148 return -ENOMEM;
4149
4150 *ret = l;
4151 return 0;
4152 }
4153
4154 int manager_set_unit_defaults(Manager *m, const UnitDefaults *defaults) {
4155 _cleanup_free_ char *label = NULL;
4156 struct rlimit *rlimit[_RLIMIT_MAX];
4157 int r;
4158
4159 assert(m);
4160 assert(defaults);
4161
4162 if (streq_ptr(defaults->smack_process_label, "/"))
4163 label = NULL;
4164 else {
4165 const char *l = defaults->smack_process_label;
4166 #ifdef SMACK_DEFAULT_PROCESS_LABEL
4167 if (!l)
4168 l = SMACK_DEFAULT_PROCESS_LABEL;
4169 #endif
4170 if (l) {
4171 label = strdup(l);
4172 if (!label)
4173 return -ENOMEM;
4174 } else
4175 label = NULL;
4176 }
4177
4178 r = rlimit_copy_all(rlimit, defaults->rlimit);
4179 if (r < 0)
4180 return r;
4181
4182 m->defaults.std_output = defaults->std_output;
4183 m->defaults.std_error = defaults->std_error;
4184
4185 m->defaults.restart_usec = defaults->restart_usec;
4186 m->defaults.timeout_start_usec = defaults->timeout_start_usec;
4187 m->defaults.timeout_stop_usec = defaults->timeout_stop_usec;
4188 m->defaults.timeout_abort_usec = defaults->timeout_abort_usec;
4189 m->defaults.timeout_abort_set = defaults->timeout_abort_set;
4190 m->defaults.device_timeout_usec = defaults->device_timeout_usec;
4191
4192 m->defaults.start_limit_interval = defaults->start_limit_interval;
4193 m->defaults.start_limit_burst = defaults->start_limit_burst;
4194
4195 m->defaults.cpu_accounting = defaults->cpu_accounting;
4196 m->defaults.memory_accounting = defaults->memory_accounting;
4197 m->defaults.io_accounting = defaults->io_accounting;
4198 m->defaults.blockio_accounting = defaults->blockio_accounting;
4199 m->defaults.tasks_accounting = defaults->tasks_accounting;
4200 m->defaults.ip_accounting = defaults->ip_accounting;
4201
4202 m->defaults.tasks_max = defaults->tasks_max;
4203 m->defaults.timer_accuracy_usec = defaults->timer_accuracy_usec;
4204
4205 m->defaults.oom_policy = defaults->oom_policy;
4206 m->defaults.oom_score_adjust = defaults->oom_score_adjust;
4207 m->defaults.oom_score_adjust_set = defaults->oom_score_adjust_set;
4208
4209 m->defaults.memory_pressure_watch = defaults->memory_pressure_watch;
4210 m->defaults.memory_pressure_threshold_usec = defaults->memory_pressure_threshold_usec;
4211
4212 free_and_replace(m->defaults.smack_process_label, label);
4213 rlimit_free_all(m->defaults.rlimit);
4214 memcpy(m->defaults.rlimit, rlimit, sizeof(struct rlimit*) * _RLIMIT_MAX);
4215
4216 return 0;
4217 }
4218
4219 void manager_recheck_dbus(Manager *m) {
4220 assert(m);
4221
4222 /* Connects to the bus if the dbus service and socket are running. If we are running in user mode
4223 * this is all it does. In system mode we'll also connect to the system bus (which will most likely
4224 * just reuse the connection of the API bus). That's because the system bus after all runs as service
4225 * of the system instance, while in the user instance we can assume it's already there. */
4226
4227 if (MANAGER_IS_RELOADING(m))
4228 return; /* don't check while we are reloading… */
4229
4230 if (manager_dbus_is_running(m, false)) {
4231 (void) bus_init_api(m);
4232
4233 if (MANAGER_IS_SYSTEM(m))
4234 (void) bus_init_system(m);
4235 } else {
4236 (void) bus_done_api(m);
4237
4238 if (MANAGER_IS_SYSTEM(m))
4239 (void) bus_done_system(m);
4240 }
4241 }
4242
4243 static bool manager_journal_is_running(Manager *m) {
4244 Unit *u;
4245
4246 assert(m);
4247
4248 if (MANAGER_IS_TEST_RUN(m))
4249 return false;
4250
4251 /* If we are the user manager we can safely assume that the journal is up */
4252 if (!MANAGER_IS_SYSTEM(m))
4253 return true;
4254
4255 /* Check that the socket is not only up, but in RUNNING state */
4256 u = manager_get_unit(m, SPECIAL_JOURNALD_SOCKET);
4257 if (!u)
4258 return false;
4259 if (SOCKET(u)->state != SOCKET_RUNNING)
4260 return false;
4261
4262 /* Similar, check if the daemon itself is fully up, too */
4263 u = manager_get_unit(m, SPECIAL_JOURNALD_SERVICE);
4264 if (!u)
4265 return false;
4266 if (!IN_SET(SERVICE(u)->state, SERVICE_RELOAD, SERVICE_RUNNING))
4267 return false;
4268
4269 return true;
4270 }
4271
4272 void disable_printk_ratelimit(void) {
4273 /* Disable kernel's printk ratelimit.
4274 *
4275 * Logging to /dev/kmsg is most useful during early boot and shutdown, where normal logging
4276 * mechanisms are not available. The semantics of this sysctl are such that any kernel command-line
4277 * setting takes precedence. */
4278 int r;
4279
4280 r = sysctl_write("kernel/printk_devkmsg", "on");
4281 if (r < 0)
4282 log_debug_errno(r, "Failed to set sysctl kernel.printk_devkmsg=on: %m");
4283 }
4284
4285 void manager_recheck_journal(Manager *m) {
4286
4287 assert(m);
4288
4289 /* Don't bother with this unless we are in the special situation of being PID 1 */
4290 if (getpid_cached() != 1)
4291 return;
4292
4293 /* Don't check this while we are reloading, things might still change */
4294 if (MANAGER_IS_RELOADING(m))
4295 return;
4296
4297 /* The journal is fully and entirely up? If so, let's permit logging to it, if that's configured. If
4298 * the journal is down, don't ever log to it, otherwise we might end up deadlocking ourselves as we
4299 * might trigger an activation ourselves we can't fulfill. */
4300 log_set_prohibit_ipc(!manager_journal_is_running(m));
4301 log_open();
4302 }
4303
4304 static ShowStatus manager_get_show_status(Manager *m) {
4305 assert(m);
4306
4307 if (MANAGER_IS_USER(m))
4308 return _SHOW_STATUS_INVALID;
4309
4310 if (m->show_status_overridden != _SHOW_STATUS_INVALID)
4311 return m->show_status_overridden;
4312
4313 return m->show_status;
4314 }
4315
4316 bool manager_get_show_status_on(Manager *m) {
4317 assert(m);
4318
4319 return show_status_on(manager_get_show_status(m));
4320 }
4321
4322 static void set_show_status_marker(bool b) {
4323 if (b)
4324 (void) touch("/run/systemd/show-status");
4325 else
4326 (void) unlink("/run/systemd/show-status");
4327 }
4328
4329 void manager_set_show_status(Manager *m, ShowStatus mode, const char *reason) {
4330 assert(m);
4331 assert(reason);
4332 assert(mode >= 0 && mode < _SHOW_STATUS_MAX);
4333
4334 if (MANAGER_IS_USER(m))
4335 return;
4336
4337 if (mode == m->show_status)
4338 return;
4339
4340 if (m->show_status_overridden == _SHOW_STATUS_INVALID) {
4341 bool enabled;
4342
4343 enabled = show_status_on(mode);
4344 log_debug("%s (%s) showing of status (%s).",
4345 enabled ? "Enabling" : "Disabling",
4346 strna(show_status_to_string(mode)),
4347 reason);
4348
4349 set_show_status_marker(enabled);
4350 }
4351
4352 m->show_status = mode;
4353 }
4354
4355 void manager_override_show_status(Manager *m, ShowStatus mode, const char *reason) {
4356 assert(m);
4357 assert(mode < _SHOW_STATUS_MAX);
4358
4359 if (MANAGER_IS_USER(m))
4360 return;
4361
4362 if (mode == m->show_status_overridden)
4363 return;
4364
4365 m->show_status_overridden = mode;
4366
4367 if (mode == _SHOW_STATUS_INVALID)
4368 mode = m->show_status;
4369
4370 log_debug("%s (%s) showing of status (%s).",
4371 m->show_status_overridden != _SHOW_STATUS_INVALID ? "Overriding" : "Restoring",
4372 strna(show_status_to_string(mode)),
4373 reason);
4374
4375 set_show_status_marker(show_status_on(mode));
4376 }
4377
4378 const char *manager_get_confirm_spawn(Manager *m) {
4379 static int last_errno = 0;
4380 struct stat st;
4381 int r;
4382
4383 assert(m);
4384
4385 /* Here's the deal: we want to test the validity of the console but don't want
4386 * PID1 to go through the whole console process which might block. But we also
4387 * want to warn the user only once if something is wrong with the console so we
4388 * cannot do the sanity checks after spawning our children. So here we simply do
4389 * really basic tests to hopefully trap common errors.
4390 *
4391 * If the console suddenly disappear at the time our children will really it
4392 * then they will simply fail to acquire it and a positive answer will be
4393 * assumed. New children will fall back to /dev/console though.
4394 *
4395 * Note: TTYs are devices that can come and go any time, and frequently aren't
4396 * available yet during early boot (consider a USB rs232 dongle...). If for any
4397 * reason the configured console is not ready, we fall back to the default
4398 * console. */
4399
4400 if (!m->confirm_spawn || path_equal(m->confirm_spawn, "/dev/console"))
4401 return m->confirm_spawn;
4402
4403 if (stat(m->confirm_spawn, &st) < 0) {
4404 r = -errno;
4405 goto fail;
4406 }
4407
4408 if (!S_ISCHR(st.st_mode)) {
4409 r = -ENOTTY;
4410 goto fail;
4411 }
4412
4413 last_errno = 0;
4414 return m->confirm_spawn;
4415
4416 fail:
4417 if (last_errno != r)
4418 last_errno = log_warning_errno(r, "Failed to open %s, using default console: %m", m->confirm_spawn);
4419
4420 return "/dev/console";
4421 }
4422
4423 void manager_set_first_boot(Manager *m, bool b) {
4424 assert(m);
4425
4426 if (!MANAGER_IS_SYSTEM(m))
4427 return;
4428
4429 if (m->first_boot != (int) b) {
4430 if (b)
4431 (void) touch("/run/systemd/first-boot");
4432 else
4433 (void) unlink("/run/systemd/first-boot");
4434 }
4435
4436 m->first_boot = b;
4437 }
4438
4439 void manager_disable_confirm_spawn(void) {
4440 (void) touch("/run/systemd/confirm_spawn_disabled");
4441 }
4442
4443 static bool manager_should_show_status(Manager *m, StatusType type) {
4444 assert(m);
4445
4446 if (!MANAGER_IS_SYSTEM(m))
4447 return false;
4448
4449 if (m->no_console_output)
4450 return false;
4451
4452 if (!IN_SET(manager_state(m), MANAGER_INITIALIZING, MANAGER_STARTING, MANAGER_STOPPING))
4453 return false;
4454
4455 /* If we cannot find out the status properly, just proceed. */
4456 if (type != STATUS_TYPE_EMERGENCY && manager_check_ask_password(m) > 0)
4457 return false;
4458
4459 if (type == STATUS_TYPE_NOTICE && m->show_status != SHOW_STATUS_NO)
4460 return true;
4461
4462 return manager_get_show_status_on(m);
4463 }
4464
4465 void manager_status_printf(Manager *m, StatusType type, const char *status, const char *format, ...) {
4466 va_list ap;
4467
4468 /* If m is NULL, assume we're after shutdown and let the messages through. */
4469
4470 if (m && !manager_should_show_status(m, type))
4471 return;
4472
4473 /* XXX We should totally drop the check for ephemeral here
4474 * and thus effectively make 'Type=idle' pointless. */
4475 if (type == STATUS_TYPE_EPHEMERAL && m && m->n_on_console > 0)
4476 return;
4477
4478 va_start(ap, format);
4479 status_vprintf(status, SHOW_STATUS_ELLIPSIZE|(type == STATUS_TYPE_EPHEMERAL ? SHOW_STATUS_EPHEMERAL : 0), format, ap);
4480 va_end(ap);
4481 }
4482
4483 Set* manager_get_units_needing_mounts_for(Manager *m, const char *path, UnitMountDependencyType t) {
4484 assert(m);
4485 assert(path);
4486 assert(t >= 0 && t < _UNIT_MOUNT_DEPENDENCY_TYPE_MAX);
4487
4488 if (path_equal(path, "/"))
4489 path = "";
4490
4491 return hashmap_get(m->units_needing_mounts_for[t], path);
4492 }
4493
4494 int manager_update_failed_units(Manager *m, Unit *u, bool failed) {
4495 unsigned size;
4496 int r;
4497
4498 assert(m);
4499 assert(u->manager == m);
4500
4501 size = set_size(m->failed_units);
4502
4503 if (failed) {
4504 r = set_ensure_put(&m->failed_units, NULL, u);
4505 if (r < 0)
4506 return log_oom();
4507 } else
4508 (void) set_remove(m->failed_units, u);
4509
4510 if (set_size(m->failed_units) != size)
4511 bus_manager_send_change_signal(m);
4512
4513 return 0;
4514 }
4515
4516 ManagerState manager_state(Manager *m) {
4517 Unit *u;
4518
4519 assert(m);
4520
4521 /* Is the special shutdown target active or queued? If so, we are in shutdown state */
4522 u = manager_get_unit(m, SPECIAL_SHUTDOWN_TARGET);
4523 if (u && unit_active_or_pending(u))
4524 return MANAGER_STOPPING;
4525
4526 /* Did we ever finish booting? If not then we are still starting up */
4527 if (!MANAGER_IS_FINISHED(m)) {
4528
4529 u = manager_get_unit(m, SPECIAL_BASIC_TARGET);
4530 if (!u || !UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u)))
4531 return MANAGER_INITIALIZING;
4532
4533 return MANAGER_STARTING;
4534 }
4535
4536 if (MANAGER_IS_SYSTEM(m)) {
4537 /* Are the rescue or emergency targets active or queued? If so we are in maintenance state */
4538 u = manager_get_unit(m, SPECIAL_RESCUE_TARGET);
4539 if (u && unit_active_or_pending(u))
4540 return MANAGER_MAINTENANCE;
4541
4542 u = manager_get_unit(m, SPECIAL_EMERGENCY_TARGET);
4543 if (u && unit_active_or_pending(u))
4544 return MANAGER_MAINTENANCE;
4545 }
4546
4547 /* Are there any failed units? If so, we are in degraded mode */
4548 if (!set_isempty(m->failed_units))
4549 return MANAGER_DEGRADED;
4550
4551 return MANAGER_RUNNING;
4552 }
4553
4554 static void manager_unref_uid_internal(
4555 Hashmap *uid_refs,
4556 uid_t uid,
4557 bool destroy_now,
4558 int (*_clean_ipc)(uid_t uid)) {
4559
4560 uint32_t c, n;
4561
4562 assert(uid_is_valid(uid));
4563 assert(_clean_ipc);
4564
4565 /* A generic implementation, covering both manager_unref_uid() and manager_unref_gid(), under the
4566 * assumption that uid_t and gid_t are actually defined the same way, with the same validity rules.
4567 *
4568 * We store a hashmap where the key is the UID/GID and the value is a 32-bit reference counter, whose
4569 * highest bit is used as flag for marking UIDs/GIDs whose IPC objects to remove when the last
4570 * reference to the UID/GID is dropped. The flag is set to on, once at least one reference from a
4571 * unit where RemoveIPC= is set is added on a UID/GID. It is reset when the UID's/GID's reference
4572 * counter drops to 0 again. */
4573
4574 assert_cc(sizeof(uid_t) == sizeof(gid_t));
4575 assert_cc(UID_INVALID == (uid_t) GID_INVALID);
4576
4577 if (uid == 0) /* We don't keep track of root, and will never destroy it */
4578 return;
4579
4580 c = PTR_TO_UINT32(hashmap_get(uid_refs, UID_TO_PTR(uid)));
4581
4582 n = c & ~DESTROY_IPC_FLAG;
4583 assert(n > 0);
4584 n--;
4585
4586 if (destroy_now && n == 0) {
4587 hashmap_remove(uid_refs, UID_TO_PTR(uid));
4588
4589 if (c & DESTROY_IPC_FLAG) {
4590 log_debug("%s " UID_FMT " is no longer referenced, cleaning up its IPC.",
4591 _clean_ipc == clean_ipc_by_uid ? "UID" : "GID",
4592 uid);
4593 (void) _clean_ipc(uid);
4594 }
4595 } else {
4596 c = n | (c & DESTROY_IPC_FLAG);
4597 assert_se(hashmap_update(uid_refs, UID_TO_PTR(uid), UINT32_TO_PTR(c)) >= 0);
4598 }
4599 }
4600
4601 void manager_unref_uid(Manager *m, uid_t uid, bool destroy_now) {
4602 manager_unref_uid_internal(m->uid_refs, uid, destroy_now, clean_ipc_by_uid);
4603 }
4604
4605 void manager_unref_gid(Manager *m, gid_t gid, bool destroy_now) {
4606 manager_unref_uid_internal(m->gid_refs, (uid_t) gid, destroy_now, clean_ipc_by_gid);
4607 }
4608
4609 static int manager_ref_uid_internal(
4610 Hashmap **uid_refs,
4611 uid_t uid,
4612 bool clean_ipc) {
4613
4614 uint32_t c, n;
4615 int r;
4616
4617 assert(uid_refs);
4618 assert(uid_is_valid(uid));
4619
4620 /* A generic implementation, covering both manager_ref_uid() and manager_ref_gid(), under the
4621 * assumption that uid_t and gid_t are actually defined the same way, with the same validity
4622 * rules. */
4623
4624 assert_cc(sizeof(uid_t) == sizeof(gid_t));
4625 assert_cc(UID_INVALID == (uid_t) GID_INVALID);
4626
4627 if (uid == 0) /* We don't keep track of root, and will never destroy it */
4628 return 0;
4629
4630 r = hashmap_ensure_allocated(uid_refs, &trivial_hash_ops);
4631 if (r < 0)
4632 return r;
4633
4634 c = PTR_TO_UINT32(hashmap_get(*uid_refs, UID_TO_PTR(uid)));
4635
4636 n = c & ~DESTROY_IPC_FLAG;
4637 n++;
4638
4639 if (n & DESTROY_IPC_FLAG) /* check for overflow */
4640 return -EOVERFLOW;
4641
4642 c = n | (c & DESTROY_IPC_FLAG) | (clean_ipc ? DESTROY_IPC_FLAG : 0);
4643
4644 return hashmap_replace(*uid_refs, UID_TO_PTR(uid), UINT32_TO_PTR(c));
4645 }
4646
4647 int manager_ref_uid(Manager *m, uid_t uid, bool clean_ipc) {
4648 return manager_ref_uid_internal(&m->uid_refs, uid, clean_ipc);
4649 }
4650
4651 int manager_ref_gid(Manager *m, gid_t gid, bool clean_ipc) {
4652 return manager_ref_uid_internal(&m->gid_refs, (uid_t) gid, clean_ipc);
4653 }
4654
4655 static void manager_vacuum_uid_refs_internal(
4656 Hashmap *uid_refs,
4657 int (*_clean_ipc)(uid_t uid)) {
4658
4659 void *p, *k;
4660
4661 assert(_clean_ipc);
4662
4663 HASHMAP_FOREACH_KEY(p, k, uid_refs) {
4664 uint32_t c, n;
4665 uid_t uid;
4666
4667 uid = PTR_TO_UID(k);
4668 c = PTR_TO_UINT32(p);
4669
4670 n = c & ~DESTROY_IPC_FLAG;
4671 if (n > 0)
4672 continue;
4673
4674 if (c & DESTROY_IPC_FLAG) {
4675 log_debug("Found unreferenced %s " UID_FMT " after reload/reexec. Cleaning up.",
4676 _clean_ipc == clean_ipc_by_uid ? "UID" : "GID",
4677 uid);
4678 (void) _clean_ipc(uid);
4679 }
4680
4681 assert_se(hashmap_remove(uid_refs, k) == p);
4682 }
4683 }
4684
4685 static void manager_vacuum_uid_refs(Manager *m) {
4686 manager_vacuum_uid_refs_internal(m->uid_refs, clean_ipc_by_uid);
4687 }
4688
4689 static void manager_vacuum_gid_refs(Manager *m) {
4690 manager_vacuum_uid_refs_internal(m->gid_refs, clean_ipc_by_gid);
4691 }
4692
4693 static void manager_vacuum(Manager *m) {
4694 assert(m);
4695
4696 /* Release any dynamic users no longer referenced */
4697 dynamic_user_vacuum(m, true);
4698
4699 /* Release any references to UIDs/GIDs no longer referenced, and destroy any IPC owned by them */
4700 manager_vacuum_uid_refs(m);
4701 manager_vacuum_gid_refs(m);
4702
4703 /* Release any runtimes no longer referenced */
4704 exec_shared_runtime_vacuum(m);
4705 }
4706
4707 int manager_dispatch_user_lookup_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
4708 struct buffer {
4709 uid_t uid;
4710 gid_t gid;
4711 char unit_name[UNIT_NAME_MAX+1];
4712 } _packed_ buffer;
4713
4714 Manager *m = userdata;
4715 ssize_t l;
4716 size_t n;
4717 Unit *u;
4718
4719 assert_se(source);
4720 assert_se(m);
4721
4722 /* Invoked whenever a child process succeeded resolving its user/group to use and sent us the
4723 * resulting UID/GID in a datagram. We parse the datagram here and pass it off to the unit, so that
4724 * it can add a reference to the UID/GID so that it can destroy the UID/GID's IPC objects when the
4725 * reference counter drops to 0. */
4726
4727 l = recv(fd, &buffer, sizeof(buffer), MSG_DONTWAIT);
4728 if (l < 0) {
4729 if (ERRNO_IS_TRANSIENT(errno))
4730 return 0;
4731
4732 return log_error_errno(errno, "Failed to read from user lookup fd: %m");
4733 }
4734
4735 if ((size_t) l <= offsetof(struct buffer, unit_name)) {
4736 log_warning("Received too short user lookup message, ignoring.");
4737 return 0;
4738 }
4739
4740 if ((size_t) l > offsetof(struct buffer, unit_name) + UNIT_NAME_MAX) {
4741 log_warning("Received too long user lookup message, ignoring.");
4742 return 0;
4743 }
4744
4745 if (!uid_is_valid(buffer.uid) && !gid_is_valid(buffer.gid)) {
4746 log_warning("Got user lookup message with invalid UID/GID pair, ignoring.");
4747 return 0;
4748 }
4749
4750 n = (size_t) l - offsetof(struct buffer, unit_name);
4751 if (memchr(buffer.unit_name, 0, n)) {
4752 log_warning("Received lookup message with embedded NUL character, ignoring.");
4753 return 0;
4754 }
4755
4756 buffer.unit_name[n] = 0;
4757 u = manager_get_unit(m, buffer.unit_name);
4758 if (!u) {
4759 log_debug("Got user lookup message but unit doesn't exist, ignoring.");
4760 return 0;
4761 }
4762
4763 log_unit_debug(u, "User lookup succeeded: uid=" UID_FMT " gid=" GID_FMT, buffer.uid, buffer.gid);
4764
4765 unit_notify_user_lookup(u, buffer.uid, buffer.gid);
4766 return 0;
4767 }
4768
4769 static int short_uid_range(const char *path) {
4770 _cleanup_(uid_range_freep) UIDRange *p = NULL;
4771 int r;
4772
4773 assert(path);
4774
4775 /* Taint systemd if we the UID range assigned to this environment doesn't at least cover 0…65534,
4776 * i.e. from root to nobody. */
4777
4778 r = uid_range_load_userns(&p, path);
4779 if (ERRNO_IS_NEG_NOT_SUPPORTED(r))
4780 return false;
4781 if (r < 0)
4782 return log_debug_errno(r, "Failed to load %s: %m", path);
4783
4784 return !uid_range_covers(p, 0, 65535);
4785 }
4786
4787 char* manager_taint_string(const Manager *m) {
4788 /* Returns a "taint string", e.g. "local-hwclock:var-run-bad". Only things that are detected at
4789 * runtime should be tagged here. For stuff that is known during compilation, emit a warning in the
4790 * configuration phase. */
4791
4792 assert(m);
4793
4794 const char* stage[12] = {};
4795 size_t n = 0;
4796
4797 _cleanup_free_ char *usrbin = NULL;
4798 if (readlink_malloc("/bin", &usrbin) < 0 || !PATH_IN_SET(usrbin, "usr/bin", "/usr/bin"))
4799 stage[n++] = "unmerged-usr";
4800
4801 if (access("/proc/cgroups", F_OK) < 0)
4802 stage[n++] = "cgroups-missing";
4803
4804 if (cg_all_unified() == 0)
4805 stage[n++] = "cgroupsv1";
4806
4807 if (clock_is_localtime(NULL) > 0)
4808 stage[n++] = "local-hwclock";
4809
4810 if (os_release_support_ended(NULL, /* quiet= */ true, NULL) > 0)
4811 stage[n++] = "support-ended";
4812
4813 _cleanup_free_ char *destination = NULL;
4814 if (readlink_malloc("/var/run", &destination) < 0 ||
4815 !PATH_IN_SET(destination, "../run", "/run"))
4816 stage[n++] = "var-run-bad";
4817
4818 _cleanup_free_ char *overflowuid = NULL, *overflowgid = NULL;
4819 if (read_one_line_file("/proc/sys/kernel/overflowuid", &overflowuid) >= 0 &&
4820 !streq(overflowuid, "65534"))
4821 stage[n++] = "overflowuid-not-65534";
4822 if (read_one_line_file("/proc/sys/kernel/overflowgid", &overflowgid) >= 0 &&
4823 !streq(overflowgid, "65534"))
4824 stage[n++] = "overflowgid-not-65534";
4825
4826 struct utsname uts;
4827 assert_se(uname(&uts) >= 0);
4828 if (strverscmp_improved(uts.release, KERNEL_BASELINE_VERSION) < 0)
4829 stage[n++] = "old-kernel";
4830
4831 if (short_uid_range("/proc/self/uid_map") > 0)
4832 stage[n++] = "short-uid-range";
4833 if (short_uid_range("/proc/self/gid_map") > 0)
4834 stage[n++] = "short-gid-range";
4835
4836 assert(n < ELEMENTSOF(stage) - 1); /* One extra for NULL terminator */
4837
4838 return strv_join((char**) stage, ":");
4839 }
4840
4841 void manager_ref_console(Manager *m) {
4842 assert(m);
4843
4844 m->n_on_console++;
4845 }
4846
4847 void manager_unref_console(Manager *m) {
4848
4849 assert(m->n_on_console > 0);
4850 m->n_on_console--;
4851
4852 if (m->n_on_console == 0)
4853 m->no_console_output = false; /* unset no_console_output flag, since the console is definitely free now */
4854 }
4855
4856 void manager_override_log_level(Manager *m, int level) {
4857 _cleanup_free_ char *s = NULL;
4858 assert(m);
4859
4860 if (!m->log_level_overridden) {
4861 m->original_log_level = log_get_max_level();
4862 m->log_level_overridden = true;
4863 }
4864
4865 (void) log_level_to_string_alloc(level, &s);
4866 log_info("Setting log level to %s.", strna(s));
4867
4868 log_set_max_level(level);
4869 }
4870
4871 void manager_restore_original_log_level(Manager *m) {
4872 _cleanup_free_ char *s = NULL;
4873 assert(m);
4874
4875 if (!m->log_level_overridden)
4876 return;
4877
4878 (void) log_level_to_string_alloc(m->original_log_level, &s);
4879 log_info("Restoring log level to original (%s).", strna(s));
4880
4881 log_set_max_level(m->original_log_level);
4882 m->log_level_overridden = false;
4883 }
4884
4885 void manager_override_log_target(Manager *m, LogTarget target) {
4886 assert(m);
4887
4888 if (!m->log_target_overridden) {
4889 m->original_log_target = log_get_target();
4890 m->log_target_overridden = true;
4891 }
4892
4893 log_info("Setting log target to %s.", log_target_to_string(target));
4894 log_set_target(target);
4895 }
4896
4897 void manager_restore_original_log_target(Manager *m) {
4898 assert(m);
4899
4900 if (!m->log_target_overridden)
4901 return;
4902
4903 log_info("Restoring log target to original %s.", log_target_to_string(m->original_log_target));
4904
4905 log_set_target(m->original_log_target);
4906 m->log_target_overridden = false;
4907 }
4908
4909 ManagerTimestamp manager_timestamp_initrd_mangle(ManagerTimestamp s) {
4910 if (in_initrd() &&
4911 s >= MANAGER_TIMESTAMP_SECURITY_START &&
4912 s <= MANAGER_TIMESTAMP_UNITS_LOAD_FINISH)
4913 return s - MANAGER_TIMESTAMP_SECURITY_START + MANAGER_TIMESTAMP_INITRD_SECURITY_START;
4914 return s;
4915 }
4916
4917 int manager_allocate_idle_pipe(Manager *m) {
4918 int r;
4919
4920 assert(m);
4921
4922 if (m->idle_pipe[0] >= 0) {
4923 assert(m->idle_pipe[1] >= 0);
4924 assert(m->idle_pipe[2] >= 0);
4925 assert(m->idle_pipe[3] >= 0);
4926 return 0;
4927 }
4928
4929 assert(m->idle_pipe[1] < 0);
4930 assert(m->idle_pipe[2] < 0);
4931 assert(m->idle_pipe[3] < 0);
4932
4933 r = RET_NERRNO(pipe2(m->idle_pipe + 0, O_NONBLOCK|O_CLOEXEC));
4934 if (r < 0)
4935 return r;
4936
4937 r = RET_NERRNO(pipe2(m->idle_pipe + 2, O_NONBLOCK|O_CLOEXEC));
4938 if (r < 0) {
4939 safe_close_pair(m->idle_pipe + 0);
4940 return r;
4941 }
4942
4943 return 1;
4944 }
4945
4946 void unit_defaults_init(UnitDefaults *defaults, RuntimeScope scope) {
4947 assert(defaults);
4948 assert(scope >= 0);
4949 assert(scope < _RUNTIME_SCOPE_MAX);
4950
4951 *defaults = (UnitDefaults) {
4952 .std_output = EXEC_OUTPUT_JOURNAL,
4953 .std_error = EXEC_OUTPUT_INHERIT,
4954 .restart_usec = DEFAULT_RESTART_USEC,
4955 .timeout_start_usec = manager_default_timeout(scope),
4956 .timeout_stop_usec = manager_default_timeout(scope),
4957 .timeout_abort_usec = manager_default_timeout(scope),
4958 .timeout_abort_set = false,
4959 .device_timeout_usec = manager_default_timeout(scope),
4960 .start_limit_interval = DEFAULT_START_LIMIT_INTERVAL,
4961 .start_limit_burst = DEFAULT_START_LIMIT_BURST,
4962
4963 /* On 4.15+ with unified hierarchy, CPU accounting is essentially free as it doesn't require the CPU
4964 * controller to be enabled, so the default is to enable it unless we got told otherwise. */
4965 .cpu_accounting = cpu_accounting_is_cheap(),
4966 .memory_accounting = MEMORY_ACCOUNTING_DEFAULT,
4967 .io_accounting = false,
4968 .blockio_accounting = false,
4969 .tasks_accounting = true,
4970 .ip_accounting = false,
4971
4972 .tasks_max = DEFAULT_TASKS_MAX,
4973 .timer_accuracy_usec = 1 * USEC_PER_MINUTE,
4974
4975 .memory_pressure_watch = CGROUP_PRESSURE_WATCH_AUTO,
4976 .memory_pressure_threshold_usec = MEMORY_PRESSURE_DEFAULT_THRESHOLD_USEC,
4977
4978 .oom_policy = OOM_STOP,
4979 .oom_score_adjust_set = false,
4980 };
4981 }
4982
4983 void unit_defaults_done(UnitDefaults *defaults) {
4984 assert(defaults);
4985
4986 defaults->smack_process_label = mfree(defaults->smack_process_label);
4987 rlimit_free_all(defaults->rlimit);
4988 }
4989
4990 LogTarget manager_get_executor_log_target(Manager *m) {
4991 assert(m);
4992
4993 /* If journald is not available tell sd-executor to go to kmsg, as it might be starting journald */
4994
4995 if (manager_journal_is_running(m))
4996 return log_get_target();
4997
4998 return LOG_TARGET_KMSG;
4999 }
5000
5001 static const char *const manager_state_table[_MANAGER_STATE_MAX] = {
5002 [MANAGER_INITIALIZING] = "initializing",
5003 [MANAGER_STARTING] = "starting",
5004 [MANAGER_RUNNING] = "running",
5005 [MANAGER_DEGRADED] = "degraded",
5006 [MANAGER_MAINTENANCE] = "maintenance",
5007 [MANAGER_STOPPING] = "stopping",
5008 };
5009
5010 DEFINE_STRING_TABLE_LOOKUP(manager_state, ManagerState);
5011
5012 static const char *const manager_timestamp_table[_MANAGER_TIMESTAMP_MAX] = {
5013 [MANAGER_TIMESTAMP_FIRMWARE] = "firmware",
5014 [MANAGER_TIMESTAMP_LOADER] = "loader",
5015 [MANAGER_TIMESTAMP_KERNEL] = "kernel",
5016 [MANAGER_TIMESTAMP_INITRD] = "initrd",
5017 [MANAGER_TIMESTAMP_USERSPACE] = "userspace",
5018 [MANAGER_TIMESTAMP_FINISH] = "finish",
5019 [MANAGER_TIMESTAMP_SECURITY_START] = "security-start",
5020 [MANAGER_TIMESTAMP_SECURITY_FINISH] = "security-finish",
5021 [MANAGER_TIMESTAMP_GENERATORS_START] = "generators-start",
5022 [MANAGER_TIMESTAMP_GENERATORS_FINISH] = "generators-finish",
5023 [MANAGER_TIMESTAMP_UNITS_LOAD_START] = "units-load-start",
5024 [MANAGER_TIMESTAMP_UNITS_LOAD_FINISH] = "units-load-finish",
5025 [MANAGER_TIMESTAMP_UNITS_LOAD] = "units-load",
5026 [MANAGER_TIMESTAMP_INITRD_SECURITY_START] = "initrd-security-start",
5027 [MANAGER_TIMESTAMP_INITRD_SECURITY_FINISH] = "initrd-security-finish",
5028 [MANAGER_TIMESTAMP_INITRD_GENERATORS_START] = "initrd-generators-start",
5029 [MANAGER_TIMESTAMP_INITRD_GENERATORS_FINISH] = "initrd-generators-finish",
5030 [MANAGER_TIMESTAMP_INITRD_UNITS_LOAD_START] = "initrd-units-load-start",
5031 [MANAGER_TIMESTAMP_INITRD_UNITS_LOAD_FINISH] = "initrd-units-load-finish",
5032 };
5033
5034 DEFINE_STRING_TABLE_LOOKUP(manager_timestamp, ManagerTimestamp);
5035
5036 static const char* const oom_policy_table[_OOM_POLICY_MAX] = {
5037 [OOM_CONTINUE] = "continue",
5038 [OOM_STOP] = "stop",
5039 [OOM_KILL] = "kill",
5040 };
5041
5042 DEFINE_STRING_TABLE_LOOKUP(oom_policy, OOMPolicy);
Unnamed repository; edit this file 'description' to name the repository.