]> 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
pid1: add "soft-reboot" reboot method
[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 "constants.h"
36 #include "core-varlink.h"
37 #include "creds-util.h"
38 #include "dbus-job.h"
39 #include "dbus-manager.h"
40 #include "dbus-unit.h"
41 #include "dbus.h"
42 #include "dirent-util.h"
43 #include "env-util.h"
44 #include "escape.h"
45 #include "event-util.h"
46 #include "exec-util.h"
47 #include "execute.h"
48 #include "exit-status.h"
49 #include "fd-util.h"
50 #include "fileio.h"
51 #include "generator-setup.h"
52 #include "hashmap.h"
53 #include "initrd-util.h"
54 #include "inotify-util.h"
55 #include "install.h"
56 #include "io-util.h"
57 #include "label-util.h"
58 #include "load-fragment.h"
59 #include "locale-setup.h"
60 #include "log.h"
61 #include "macro.h"
62 #include "manager.h"
63 #include "manager-dump.h"
64 #include "manager-serialize.h"
65 #include "memory-util.h"
66 #include "mkdir-label.h"
67 #include "mount-util.h"
68 #include "os-util.h"
69 #include "parse-util.h"
70 #include "path-lookup.h"
71 #include "path-util.h"
72 #include "process-util.h"
73 #include "psi-util.h"
74 #include "ratelimit.h"
75 #include "rlimit-util.h"
76 #include "rm-rf.h"
77 #include "selinux-util.h"
78 #include "signal-util.h"
79 #include "socket-util.h"
80 #include "special.h"
81 #include "stat-util.h"
82 #include "string-table.h"
83 #include "string-util.h"
84 #include "strv.h"
85 #include "strxcpyx.h"
86 #include "sysctl-util.h"
87 #include "syslog-util.h"
88 #include "terminal-util.h"
89 #include "time-util.h"
90 #include "transaction.h"
91 #include "uid-range.h"
92 #include "umask-util.h"
93 #include "unit-name.h"
94 #include "user-util.h"
95 #include "virt.h"
96 #include "watchdog.h"
97
98 #define NOTIFY_RCVBUF_SIZE (8*1024*1024)
99 #define CGROUPS_AGENT_RCVBUF_SIZE (8*1024*1024)
100
101 /* Initial delay and the interval for printing status messages about running jobs */
102 #define JOBS_IN_PROGRESS_WAIT_USEC (2*USEC_PER_SEC)
103 #define JOBS_IN_PROGRESS_QUIET_WAIT_USEC (25*USEC_PER_SEC)
104 #define JOBS_IN_PROGRESS_PERIOD_USEC (USEC_PER_SEC / 3)
105 #define JOBS_IN_PROGRESS_PERIOD_DIVISOR 3
106
107 /* If there are more than 1K bus messages queue across our API and direct buses, then let's not add more on top until
108 * the queue gets more empty. */
109 #define MANAGER_BUS_BUSY_THRESHOLD 1024LU
110
111 /* How many units and jobs to process of the bus queue before returning to the event loop. */
112 #define MANAGER_BUS_MESSAGE_BUDGET 100U
113
114 static int manager_dispatch_notify_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata);
115 static int manager_dispatch_cgroups_agent_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata);
116 static int manager_dispatch_signal_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata);
117 static int manager_dispatch_time_change_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata);
118 static int manager_dispatch_idle_pipe_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata);
119 static int manager_dispatch_user_lookup_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata);
120 static int manager_dispatch_jobs_in_progress(sd_event_source *source, usec_t usec, void *userdata);
121 static int manager_dispatch_run_queue(sd_event_source *source, void *userdata);
122 static int manager_dispatch_sigchld(sd_event_source *source, void *userdata);
123 static int manager_dispatch_timezone_change(sd_event_source *source, const struct inotify_event *event, void *userdata);
124 static int manager_run_environment_generators(Manager *m);
125 static int manager_run_generators(Manager *m);
126 static void manager_vacuum(Manager *m);
127
128 static usec_t manager_watch_jobs_next_time(Manager *m) {
129 usec_t timeout;
130
131 if (MANAGER_IS_USER(m))
132 /* Let the user manager without a timeout show status quickly, so the system manager can make
133 * use of it, if it wants to. */
134 timeout = JOBS_IN_PROGRESS_WAIT_USEC * 2 / 3;
135 else if (show_status_on(m->show_status))
136 /* When status is on, just use the usual timeout. */
137 timeout = JOBS_IN_PROGRESS_WAIT_USEC;
138 else
139 timeout = JOBS_IN_PROGRESS_QUIET_WAIT_USEC;
140
141 return usec_add(now(CLOCK_MONOTONIC), timeout);
142 }
143
144 static void manager_watch_jobs_in_progress(Manager *m) {
145 usec_t next;
146 int r;
147
148 assert(m);
149
150 /* We do not want to show the cylon animation if the user
151 * needs to confirm service executions otherwise confirmation
152 * messages will be screwed by the cylon animation. */
153 if (!manager_is_confirm_spawn_disabled(m))
154 return;
155
156 if (m->jobs_in_progress_event_source)
157 return;
158
159 next = manager_watch_jobs_next_time(m);
160 r = sd_event_add_time(
161 m->event,
162 &m->jobs_in_progress_event_source,
163 CLOCK_MONOTONIC,
164 next, 0,
165 manager_dispatch_jobs_in_progress, m);
166 if (r < 0)
167 return;
168
169 (void) sd_event_source_set_description(m->jobs_in_progress_event_source, "manager-jobs-in-progress");
170 }
171
172 #define CYLON_BUFFER_EXTRA (2*STRLEN(ANSI_RED) + STRLEN(ANSI_HIGHLIGHT_RED) + 2*STRLEN(ANSI_NORMAL))
173
174 static void draw_cylon(char buffer[], size_t buflen, unsigned width, unsigned pos) {
175 char *p = buffer;
176
177 assert(buflen >= CYLON_BUFFER_EXTRA + width + 1);
178 assert(pos <= width+1); /* 0 or width+1 mean that the center light is behind the corner */
179
180 if (pos > 1) {
181 if (pos > 2)
182 p = mempset(p, ' ', pos-2);
183 if (log_get_show_color())
184 p = stpcpy(p, ANSI_RED);
185 *p++ = '*';
186 }
187
188 if (pos > 0 && pos <= width) {
189 if (log_get_show_color())
190 p = stpcpy(p, ANSI_HIGHLIGHT_RED);
191 *p++ = '*';
192 }
193
194 if (log_get_show_color())
195 p = stpcpy(p, ANSI_NORMAL);
196
197 if (pos < width) {
198 if (log_get_show_color())
199 p = stpcpy(p, ANSI_RED);
200 *p++ = '*';
201 if (pos < width-1)
202 p = mempset(p, ' ', width-1-pos);
203 if (log_get_show_color())
204 strcpy(p, ANSI_NORMAL);
205 }
206 }
207
208 static void manager_flip_auto_status(Manager *m, bool enable, const char *reason) {
209 assert(m);
210
211 if (enable) {
212 if (m->show_status == SHOW_STATUS_AUTO)
213 manager_set_show_status(m, SHOW_STATUS_TEMPORARY, reason);
214 } else {
215 if (m->show_status == SHOW_STATUS_TEMPORARY)
216 manager_set_show_status(m, SHOW_STATUS_AUTO, reason);
217 }
218 }
219
220 static void manager_print_jobs_in_progress(Manager *m) {
221 Job *j;
222 unsigned counter = 0, print_nr;
223 char cylon[6 + CYLON_BUFFER_EXTRA + 1];
224 unsigned cylon_pos;
225 uint64_t timeout = 0;
226
227 assert(m);
228 assert(m->n_running_jobs > 0);
229
230 manager_flip_auto_status(m, true, "delay");
231
232 print_nr = (m->jobs_in_progress_iteration / JOBS_IN_PROGRESS_PERIOD_DIVISOR) % m->n_running_jobs;
233
234 HASHMAP_FOREACH(j, m->jobs)
235 if (j->state == JOB_RUNNING && counter++ == print_nr)
236 break;
237
238 /* m->n_running_jobs must be consistent with the contents of m->jobs,
239 * so the above loop must have succeeded in finding j. */
240 assert(counter == print_nr + 1);
241 assert(j);
242
243 cylon_pos = m->jobs_in_progress_iteration % 14;
244 if (cylon_pos >= 8)
245 cylon_pos = 14 - cylon_pos;
246 draw_cylon(cylon, sizeof(cylon), 6, cylon_pos);
247
248 m->jobs_in_progress_iteration++;
249
250 char job_of_n[STRLEN("( of ) ") + DECIMAL_STR_MAX(unsigned)*2] = "";
251 if (m->n_running_jobs > 1)
252 xsprintf(job_of_n, "(%u of %u) ", counter, m->n_running_jobs);
253
254 (void) job_get_timeout(j, &timeout);
255
256 /* We want to use enough information for the user to identify previous lines talking about the same
257 * unit, but keep the message as short as possible. So if 'Starting foo.service' or 'Starting
258 * foo.service - Description' were used, 'foo.service' is enough here. On the other hand, if we used
259 * 'Starting Description' before, then we shall also use 'Description' here. So we pass NULL as the
260 * second argument to unit_status_string(). */
261 const char *ident = unit_status_string(j->unit, NULL);
262
263 const char *time = FORMAT_TIMESPAN(now(CLOCK_MONOTONIC) - j->begin_usec, 1*USEC_PER_SEC);
264 const char *limit = timeout > 0 ? FORMAT_TIMESPAN(timeout - j->begin_usec, 1*USEC_PER_SEC) : "no limit";
265
266 if (m->status_unit_format == STATUS_UNIT_FORMAT_DESCRIPTION)
267 /* When using 'Description', we effectively don't have enough space to show the nested status
268 * without ellipsization, so let's not even try. */
269 manager_status_printf(m, STATUS_TYPE_EPHEMERAL, cylon,
270 "%sA %s job is running for %s (%s / %s)",
271 job_of_n,
272 job_type_to_string(j->type),
273 ident,
274 time, limit);
275 else {
276 const char *status_text = unit_status_text(j->unit);
277
278 manager_status_printf(m, STATUS_TYPE_EPHEMERAL, cylon,
279 "%sJob %s/%s running (%s / %s)%s%s",
280 job_of_n,
281 ident,
282 job_type_to_string(j->type),
283 time, limit,
284 status_text ? ": " : "",
285 strempty(status_text));
286 }
287
288 sd_notifyf(false,
289 "STATUS=%sUser job %s/%s running (%s / %s)...",
290 job_of_n,
291 ident,
292 job_type_to_string(j->type),
293 time, limit);
294 m->status_ready = false;
295 }
296
297 static int have_ask_password(void) {
298 _cleanup_closedir_ DIR *dir = NULL;
299
300 dir = opendir("/run/systemd/ask-password");
301 if (!dir) {
302 if (errno == ENOENT)
303 return false;
304 else
305 return -errno;
306 }
307
308 FOREACH_DIRENT_ALL(de, dir, return -errno)
309 if (startswith(de->d_name, "ask."))
310 return true;
311 return false;
312 }
313
314 static int manager_dispatch_ask_password_fd(sd_event_source *source,
315 int fd, uint32_t revents, void *userdata) {
316 Manager *m = ASSERT_PTR(userdata);
317
318 (void) flush_fd(fd);
319
320 m->have_ask_password = have_ask_password();
321 if (m->have_ask_password < 0)
322 /* Log error but continue. Negative have_ask_password
323 * is treated as unknown status. */
324 log_error_errno(m->have_ask_password, "Failed to list /run/systemd/ask-password: %m");
325
326 return 0;
327 }
328
329 static void manager_close_ask_password(Manager *m) {
330 assert(m);
331
332 m->ask_password_event_source = sd_event_source_disable_unref(m->ask_password_event_source);
333 m->ask_password_inotify_fd = safe_close(m->ask_password_inotify_fd);
334 m->have_ask_password = -EINVAL;
335 }
336
337 static int manager_check_ask_password(Manager *m) {
338 int r;
339
340 assert(m);
341
342 if (!m->ask_password_event_source) {
343 assert(m->ask_password_inotify_fd < 0);
344
345 (void) mkdir_p_label("/run/systemd/ask-password", 0755);
346
347 m->ask_password_inotify_fd = inotify_init1(IN_NONBLOCK|IN_CLOEXEC);
348 if (m->ask_password_inotify_fd < 0)
349 return log_error_errno(errno, "Failed to create inotify object: %m");
350
351 r = inotify_add_watch_and_warn(m->ask_password_inotify_fd,
352 "/run/systemd/ask-password",
353 IN_CREATE|IN_DELETE|IN_MOVE);
354 if (r < 0) {
355 manager_close_ask_password(m);
356 return r;
357 }
358
359 r = sd_event_add_io(m->event, &m->ask_password_event_source,
360 m->ask_password_inotify_fd, EPOLLIN,
361 manager_dispatch_ask_password_fd, m);
362 if (r < 0) {
363 log_error_errno(r, "Failed to add event source for /run/systemd/ask-password: %m");
364 manager_close_ask_password(m);
365 return r;
366 }
367
368 (void) sd_event_source_set_description(m->ask_password_event_source, "manager-ask-password");
369
370 /* Queries might have been added meanwhile... */
371 manager_dispatch_ask_password_fd(m->ask_password_event_source,
372 m->ask_password_inotify_fd, EPOLLIN, m);
373 }
374
375 return m->have_ask_password;
376 }
377
378 static int manager_watch_idle_pipe(Manager *m) {
379 int r;
380
381 assert(m);
382
383 if (m->idle_pipe_event_source)
384 return 0;
385
386 if (m->idle_pipe[2] < 0)
387 return 0;
388
389 r = sd_event_add_io(m->event, &m->idle_pipe_event_source, m->idle_pipe[2], EPOLLIN, manager_dispatch_idle_pipe_fd, m);
390 if (r < 0)
391 return log_error_errno(r, "Failed to watch idle pipe: %m");
392
393 (void) sd_event_source_set_description(m->idle_pipe_event_source, "manager-idle-pipe");
394
395 return 0;
396 }
397
398 static void manager_close_idle_pipe(Manager *m) {
399 assert(m);
400
401 m->idle_pipe_event_source = sd_event_source_disable_unref(m->idle_pipe_event_source);
402
403 safe_close_pair(m->idle_pipe);
404 safe_close_pair(m->idle_pipe + 2);
405 }
406
407 static int manager_setup_time_change(Manager *m) {
408 int r;
409
410 assert(m);
411
412 if (MANAGER_IS_TEST_RUN(m))
413 return 0;
414
415 m->time_change_event_source = sd_event_source_disable_unref(m->time_change_event_source);
416
417 r = event_add_time_change(m->event, &m->time_change_event_source, manager_dispatch_time_change_fd, m);
418 if (r < 0)
419 return log_error_errno(r, "Failed to create time change event source: %m");
420
421 /* Schedule this slightly earlier than the .timer event sources */
422 r = sd_event_source_set_priority(m->time_change_event_source, SD_EVENT_PRIORITY_NORMAL-1);
423 if (r < 0)
424 return log_error_errno(r, "Failed to set priority of time change event sources: %m");
425
426 log_debug("Set up TFD_TIMER_CANCEL_ON_SET timerfd.");
427
428 return 0;
429 }
430
431 static int manager_read_timezone_stat(Manager *m) {
432 struct stat st;
433 bool changed;
434
435 assert(m);
436
437 /* Read the current stat() data of /etc/localtime so that we detect changes */
438 if (lstat("/etc/localtime", &st) < 0) {
439 log_debug_errno(errno, "Failed to stat /etc/localtime, ignoring: %m");
440 changed = m->etc_localtime_accessible;
441 m->etc_localtime_accessible = false;
442 } else {
443 usec_t k;
444
445 k = timespec_load(&st.st_mtim);
446 changed = !m->etc_localtime_accessible || k != m->etc_localtime_mtime;
447
448 m->etc_localtime_mtime = k;
449 m->etc_localtime_accessible = true;
450 }
451
452 return changed;
453 }
454
455 static int manager_setup_timezone_change(Manager *m) {
456 _cleanup_(sd_event_source_unrefp) sd_event_source *new_event = NULL;
457 int r;
458
459 assert(m);
460
461 if (MANAGER_IS_TEST_RUN(m))
462 return 0;
463
464 /* We watch /etc/localtime for three events: change of the link count (which might mean removal from /etc even
465 * though another link might be kept), renames, and file close operations after writing. Note we don't bother
466 * with IN_DELETE_SELF, as that would just report when the inode is removed entirely, i.e. after the link count
467 * went to zero and all fds to it are closed.
468 *
469 * Note that we never follow symlinks here. This is a simplification, but should cover almost all cases
470 * correctly.
471 *
472 * Note that we create the new event source first here, before releasing the old one. This should optimize
473 * behaviour as this way sd-event can reuse the old watch in case the inode didn't change. */
474
475 r = sd_event_add_inotify(m->event, &new_event, "/etc/localtime",
476 IN_ATTRIB|IN_MOVE_SELF|IN_CLOSE_WRITE|IN_DONT_FOLLOW, manager_dispatch_timezone_change, m);
477 if (r == -ENOENT) {
478 /* If the file doesn't exist yet, subscribe to /etc instead, and wait until it is created either by
479 * O_CREATE or by rename() */
480
481 log_debug_errno(r, "/etc/localtime doesn't exist yet, watching /etc instead.");
482 r = sd_event_add_inotify(m->event, &new_event, "/etc",
483 IN_CREATE|IN_MOVED_TO|IN_ONLYDIR, manager_dispatch_timezone_change, m);
484 }
485 if (r < 0)
486 return log_error_errno(r, "Failed to create timezone change event source: %m");
487
488 /* Schedule this slightly earlier than the .timer event sources */
489 r = sd_event_source_set_priority(new_event, SD_EVENT_PRIORITY_NORMAL-1);
490 if (r < 0)
491 return log_error_errno(r, "Failed to set priority of timezone change event sources: %m");
492
493 sd_event_source_unref(m->timezone_change_event_source);
494 m->timezone_change_event_source = TAKE_PTR(new_event);
495
496 return 0;
497 }
498
499 static int enable_special_signals(Manager *m) {
500 _cleanup_close_ int fd = -EBADF;
501
502 assert(m);
503
504 if (MANAGER_IS_TEST_RUN(m))
505 return 0;
506
507 /* Enable that we get SIGINT on control-alt-del. In containers
508 * this will fail with EPERM (older) or EINVAL (newer), so
509 * ignore that. */
510 if (reboot(RB_DISABLE_CAD) < 0 && !IN_SET(errno, EPERM, EINVAL))
511 log_warning_errno(errno, "Failed to enable ctrl-alt-del handling: %m");
512
513 fd = open_terminal("/dev/tty0", O_RDWR|O_NOCTTY|O_CLOEXEC);
514 if (fd < 0) {
515 /* Support systems without virtual console */
516 if (fd != -ENOENT)
517 log_warning_errno(errno, "Failed to open /dev/tty0: %m");
518 } else {
519 /* Enable that we get SIGWINCH on kbrequest */
520 if (ioctl(fd, KDSIGACCEPT, SIGWINCH) < 0)
521 log_warning_errno(errno, "Failed to enable kbrequest handling: %m");
522 }
523
524 return 0;
525 }
526
527 #define RTSIG_IF_AVAILABLE(signum) (signum <= SIGRTMAX ? signum : -1)
528
529 static int manager_setup_signals(Manager *m) {
530 struct sigaction sa = {
531 .sa_handler = SIG_DFL,
532 .sa_flags = SA_NOCLDSTOP|SA_RESTART,
533 };
534 sigset_t mask;
535 int r;
536
537 assert(m);
538
539 assert_se(sigaction(SIGCHLD, &sa, NULL) == 0);
540
541 /* We make liberal use of realtime signals here. On
542 * Linux/glibc we have 30 of them (with the exception of Linux
543 * on hppa, see below), between SIGRTMIN+0 ... SIGRTMIN+30
544 * (aka SIGRTMAX). */
545
546 assert_se(sigemptyset(&mask) == 0);
547 sigset_add_many(&mask,
548 SIGCHLD, /* Child died */
549 SIGTERM, /* Reexecute daemon */
550 SIGHUP, /* Reload configuration */
551 SIGUSR1, /* systemd: reconnect to D-Bus */
552 SIGUSR2, /* systemd: dump status */
553 SIGINT, /* Kernel sends us this on control-alt-del */
554 SIGWINCH, /* Kernel sends us this on kbrequest (alt-arrowup) */
555 SIGPWR, /* Some kernel drivers and upsd send us this on power failure */
556
557 SIGRTMIN+0, /* systemd: start default.target */
558 SIGRTMIN+1, /* systemd: isolate rescue.target */
559 SIGRTMIN+2, /* systemd: isolate emergency.target */
560 SIGRTMIN+3, /* systemd: start halt.target */
561 SIGRTMIN+4, /* systemd: start poweroff.target */
562 SIGRTMIN+5, /* systemd: start reboot.target */
563 SIGRTMIN+6, /* systemd: start kexec.target */
564 SIGRTMIN+7, /* systemd: start soft-reboot.target */
565
566 /* ... space for more special targets ... */
567
568 SIGRTMIN+13, /* systemd: Immediate halt */
569 SIGRTMIN+14, /* systemd: Immediate poweroff */
570 SIGRTMIN+15, /* systemd: Immediate reboot */
571 SIGRTMIN+16, /* systemd: Immediate kexec */
572 SIGRTMIN+17, /* systemd: Immediate soft-reboot */
573 SIGRTMIN+18, /* systemd: control command */
574
575 /* ... space ... */
576
577 SIGRTMIN+20, /* systemd: enable status messages */
578 SIGRTMIN+21, /* systemd: disable status messages */
579 SIGRTMIN+22, /* systemd: set log level to LOG_DEBUG */
580 SIGRTMIN+23, /* systemd: set log level to LOG_INFO */
581 SIGRTMIN+24, /* systemd: Immediate exit (--user only) */
582 SIGRTMIN+25, /* systemd: reexecute manager */
583
584 /* Apparently Linux on hppa had fewer RT signals until v3.18,
585 * SIGRTMAX was SIGRTMIN+25, and then SIGRTMIN was lowered,
586 * see commit v3.17-7614-g1f25df2eff.
587 *
588 * We cannot unconditionally make use of those signals here,
589 * so let's use a runtime check. Since these commands are
590 * accessible by different means and only really a safety
591 * net, the missing functionality on hppa shouldn't matter.
592 */
593
594 RTSIG_IF_AVAILABLE(SIGRTMIN+26), /* systemd: set log target to journal-or-kmsg */
595 RTSIG_IF_AVAILABLE(SIGRTMIN+27), /* systemd: set log target to console */
596 RTSIG_IF_AVAILABLE(SIGRTMIN+28), /* systemd: set log target to kmsg */
597 RTSIG_IF_AVAILABLE(SIGRTMIN+29), /* systemd: set log target to syslog-or-kmsg (obsolete) */
598
599 /* ... one free signal here SIGRTMIN+30 ... */
600 -1);
601 assert_se(sigprocmask(SIG_SETMASK, &mask, NULL) == 0);
602
603 m->signal_fd = signalfd(-1, &mask, SFD_NONBLOCK|SFD_CLOEXEC);
604 if (m->signal_fd < 0)
605 return -errno;
606
607 r = sd_event_add_io(m->event, &m->signal_event_source, m->signal_fd, EPOLLIN, manager_dispatch_signal_fd, m);
608 if (r < 0)
609 return r;
610
611 (void) sd_event_source_set_description(m->signal_event_source, "manager-signal");
612
613 /* Process signals a bit earlier than the rest of things, but later than notify_fd processing, so that the
614 * notify processing can still figure out to which process/service a message belongs, before we reap the
615 * process. Also, process this before handling cgroup notifications, so that we always collect child exit
616 * status information before detecting that there's no process in a cgroup. */
617 r = sd_event_source_set_priority(m->signal_event_source, SD_EVENT_PRIORITY_NORMAL-6);
618 if (r < 0)
619 return r;
620
621 if (MANAGER_IS_SYSTEM(m))
622 return enable_special_signals(m);
623
624 return 0;
625 }
626
627 static char** sanitize_environment(char **l) {
628
629 /* Let's remove some environment variables that we need ourselves to communicate with our clients */
630 strv_env_unset_many(
631 l,
632 "CACHE_DIRECTORY",
633 "CONFIGURATION_DIRECTORY",
634 "CREDENTIALS_DIRECTORY",
635 "EXIT_CODE",
636 "EXIT_STATUS",
637 "INVOCATION_ID",
638 "JOURNAL_STREAM",
639 "LISTEN_FDNAMES",
640 "LISTEN_FDS",
641 "LISTEN_PID",
642 "LOGS_DIRECTORY",
643 "LOG_NAMESPACE",
644 "MAINPID",
645 "MANAGERPID",
646 "MEMORY_PRESSURE_WATCH",
647 "MEMORY_PRESSURE_WRITE",
648 "MONITOR_EXIT_CODE",
649 "MONITOR_EXIT_STATUS",
650 "MONITOR_INVOCATION_ID",
651 "MONITOR_SERVICE_RESULT",
652 "MONITOR_UNIT",
653 "NOTIFY_SOCKET",
654 "PIDFILE",
655 "REMOTE_ADDR",
656 "REMOTE_PORT",
657 "RUNTIME_DIRECTORY",
658 "SERVICE_RESULT",
659 "STATE_DIRECTORY",
660 "SYSTEMD_EXEC_PID",
661 "TRIGGER_PATH",
662 "TRIGGER_TIMER_MONOTONIC_USEC",
663 "TRIGGER_TIMER_REALTIME_USEC",
664 "TRIGGER_UNIT",
665 "WATCHDOG_PID",
666 "WATCHDOG_USEC",
667 NULL);
668
669 /* Let's order the environment alphabetically, just to make it pretty */
670 return strv_sort(l);
671 }
672
673 int manager_default_environment(Manager *m) {
674 int r;
675
676 assert(m);
677
678 m->transient_environment = strv_free(m->transient_environment);
679
680 if (MANAGER_IS_SYSTEM(m)) {
681 /* The system manager always starts with a clean environment for its children. It does not
682 * import the kernel's or the parents' exported variables.
683 *
684 * The initial passed environment is untouched to keep /proc/self/environ valid; it is used
685 * for tagging the init process inside containers. */
686 m->transient_environment = strv_new("PATH=" DEFAULT_PATH);
687 if (!m->transient_environment)
688 return log_oom();
689
690 /* Import locale variables LC_*= from configuration */
691 (void) locale_setup(&m->transient_environment);
692 } else {
693 /* The user manager passes its own environment along to its children, except for $PATH. */
694 m->transient_environment = strv_copy(environ);
695 if (!m->transient_environment)
696 return log_oom();
697
698 r = strv_env_replace_strdup(&m->transient_environment, "PATH=" DEFAULT_USER_PATH);
699 if (r < 0)
700 return log_oom();
701 }
702
703 sanitize_environment(m->transient_environment);
704 return 0;
705 }
706
707 static int manager_setup_prefix(Manager *m) {
708 struct table_entry {
709 uint64_t type;
710 const char *suffix;
711 };
712
713 static const struct table_entry paths_system[_EXEC_DIRECTORY_TYPE_MAX] = {
714 [EXEC_DIRECTORY_RUNTIME] = { SD_PATH_SYSTEM_RUNTIME, NULL },
715 [EXEC_DIRECTORY_STATE] = { SD_PATH_SYSTEM_STATE_PRIVATE, NULL },
716 [EXEC_DIRECTORY_CACHE] = { SD_PATH_SYSTEM_STATE_CACHE, NULL },
717 [EXEC_DIRECTORY_LOGS] = { SD_PATH_SYSTEM_STATE_LOGS, NULL },
718 [EXEC_DIRECTORY_CONFIGURATION] = { SD_PATH_SYSTEM_CONFIGURATION, NULL },
719 };
720
721 static const struct table_entry paths_user[_EXEC_DIRECTORY_TYPE_MAX] = {
722 [EXEC_DIRECTORY_RUNTIME] = { SD_PATH_USER_RUNTIME, NULL },
723 [EXEC_DIRECTORY_STATE] = { SD_PATH_USER_CONFIGURATION, NULL },
724 [EXEC_DIRECTORY_CACHE] = { SD_PATH_USER_STATE_CACHE, NULL },
725 [EXEC_DIRECTORY_LOGS] = { SD_PATH_USER_CONFIGURATION, "log" },
726 [EXEC_DIRECTORY_CONFIGURATION] = { SD_PATH_USER_CONFIGURATION, NULL },
727 };
728
729 assert(m);
730
731 const struct table_entry *p = MANAGER_IS_SYSTEM(m) ? paths_system : paths_user;
732 int r;
733
734 for (ExecDirectoryType i = 0; i < _EXEC_DIRECTORY_TYPE_MAX; i++) {
735 r = sd_path_lookup(p[i].type, p[i].suffix, &m->prefix[i]);
736 if (r < 0)
737 return log_warning_errno(r, "Failed to lookup %s path: %m",
738 exec_directory_type_to_string(i));
739 }
740
741 return 0;
742 }
743
744 static void manager_free_unit_name_maps(Manager *m) {
745 m->unit_id_map = hashmap_free(m->unit_id_map);
746 m->unit_name_map = hashmap_free(m->unit_name_map);
747 m->unit_path_cache = set_free(m->unit_path_cache);
748 m->unit_cache_timestamp_hash = 0;
749 }
750
751 static int manager_setup_run_queue(Manager *m) {
752 int r;
753
754 assert(m);
755 assert(!m->run_queue_event_source);
756
757 r = sd_event_add_defer(m->event, &m->run_queue_event_source, manager_dispatch_run_queue, m);
758 if (r < 0)
759 return r;
760
761 r = sd_event_source_set_priority(m->run_queue_event_source, SD_EVENT_PRIORITY_IDLE);
762 if (r < 0)
763 return r;
764
765 r = sd_event_source_set_enabled(m->run_queue_event_source, SD_EVENT_OFF);
766 if (r < 0)
767 return r;
768
769 (void) sd_event_source_set_description(m->run_queue_event_source, "manager-run-queue");
770
771 return 0;
772 }
773
774 static int manager_setup_sigchld_event_source(Manager *m) {
775 int r;
776
777 assert(m);
778 assert(!m->sigchld_event_source);
779
780 r = sd_event_add_defer(m->event, &m->sigchld_event_source, manager_dispatch_sigchld, m);
781 if (r < 0)
782 return r;
783
784 r = sd_event_source_set_priority(m->sigchld_event_source, SD_EVENT_PRIORITY_NORMAL-7);
785 if (r < 0)
786 return r;
787
788 r = sd_event_source_set_enabled(m->sigchld_event_source, SD_EVENT_OFF);
789 if (r < 0)
790 return r;
791
792 (void) sd_event_source_set_description(m->sigchld_event_source, "manager-sigchld");
793
794 return 0;
795 }
796
797 int manager_setup_memory_pressure_event_source(Manager *m) {
798 int r;
799
800 assert(m);
801
802 m->memory_pressure_event_source = sd_event_source_disable_unref(m->memory_pressure_event_source);
803
804 r = sd_event_add_memory_pressure(m->event, &m->memory_pressure_event_source, NULL, NULL);
805 if (r < 0)
806 log_full_errno(ERRNO_IS_NOT_SUPPORTED(r) || ERRNO_IS_PRIVILEGE(r) || (r == -EHOSTDOWN) ? LOG_DEBUG : LOG_NOTICE, r,
807 "Failed to establish memory pressure event source, ignoring: %m");
808 else if (m->default_memory_pressure_threshold_usec != USEC_INFINITY) {
809
810 /* If there's a default memory pressure threshold set, also apply it to the service manager itself */
811 r = sd_event_source_set_memory_pressure_period(
812 m->memory_pressure_event_source,
813 m->default_memory_pressure_threshold_usec,
814 MEMORY_PRESSURE_DEFAULT_WINDOW_USEC);
815 if (r < 0)
816 log_warning_errno(r, "Failed to adjust memory pressure threshold, ignoring: %m");
817 }
818
819 return 0;
820 }
821
822 static int manager_find_credentials_dirs(Manager *m) {
823 const char *e;
824 int r;
825
826 assert(m);
827
828 r = get_credentials_dir(&e);
829 if (r < 0) {
830 if (r != -ENXIO)
831 log_debug_errno(r, "Failed to determine credentials directory, ignoring: %m");
832 } else {
833 m->received_credentials_directory = strdup(e);
834 if (!m->received_credentials_directory)
835 return -ENOMEM;
836 }
837
838 r = get_encrypted_credentials_dir(&e);
839 if (r < 0) {
840 if (r != -ENXIO)
841 log_debug_errno(r, "Failed to determine encrypted credentials directory, ignoring: %m");
842 } else {
843 m->received_encrypted_credentials_directory = strdup(e);
844 if (!m->received_encrypted_credentials_directory)
845 return -ENOMEM;
846 }
847
848 return 0;
849 }
850
851 void manager_set_switching_root(Manager *m, bool switching_root) {
852 assert(m);
853
854 m->switching_root = MANAGER_IS_SYSTEM(m) && switching_root;
855 }
856
857 int manager_new(RuntimeScope runtime_scope, ManagerTestRunFlags test_run_flags, Manager **_m) {
858 _cleanup_(manager_freep) Manager *m = NULL;
859 int r;
860
861 assert(_m);
862 assert(IN_SET(runtime_scope, RUNTIME_SCOPE_SYSTEM, RUNTIME_SCOPE_USER));
863
864 m = new(Manager, 1);
865 if (!m)
866 return -ENOMEM;
867
868 *m = (Manager) {
869 .runtime_scope = runtime_scope,
870 .objective = _MANAGER_OBJECTIVE_INVALID,
871
872 .status_unit_format = STATUS_UNIT_FORMAT_DEFAULT,
873
874 .default_timer_accuracy_usec = USEC_PER_MINUTE,
875 .default_memory_accounting = MEMORY_ACCOUNTING_DEFAULT,
876 .default_tasks_accounting = true,
877 .default_tasks_max = TASKS_MAX_UNSET,
878 .default_timeout_start_usec = manager_default_timeout(runtime_scope),
879 .default_timeout_stop_usec = manager_default_timeout(runtime_scope),
880 .default_restart_usec = DEFAULT_RESTART_USEC,
881 .default_device_timeout_usec = manager_default_timeout(runtime_scope),
882
883 .original_log_level = -1,
884 .original_log_target = _LOG_TARGET_INVALID,
885
886 .watchdog_overridden[WATCHDOG_RUNTIME] = USEC_INFINITY,
887 .watchdog_overridden[WATCHDOG_REBOOT] = USEC_INFINITY,
888 .watchdog_overridden[WATCHDOG_KEXEC] = USEC_INFINITY,
889 .watchdog_overridden[WATCHDOG_PRETIMEOUT] = USEC_INFINITY,
890
891 .show_status_overridden = _SHOW_STATUS_INVALID,
892
893 .notify_fd = -EBADF,
894 .cgroups_agent_fd = -EBADF,
895 .signal_fd = -EBADF,
896 .user_lookup_fds = PIPE_EBADF,
897 .private_listen_fd = -EBADF,
898 .dev_autofs_fd = -EBADF,
899 .cgroup_inotify_fd = -EBADF,
900 .pin_cgroupfs_fd = -EBADF,
901 .ask_password_inotify_fd = -EBADF,
902 .idle_pipe = { -EBADF, -EBADF, -EBADF, -EBADF},
903
904 /* start as id #1, so that we can leave #0 around as "null-like" value */
905 .current_job_id = 1,
906
907 .have_ask_password = -EINVAL, /* we don't know */
908 .first_boot = -1,
909 .test_run_flags = test_run_flags,
910
911 .default_oom_policy = OOM_STOP,
912
913 .default_memory_pressure_watch = CGROUP_PRESSURE_WATCH_AUTO,
914 .default_memory_pressure_threshold_usec = USEC_INFINITY,
915
916 .dump_ratelimit = {
917 .interval = 10 * USEC_PER_MINUTE,
918 .burst = 10,
919 },
920 };
921
922 #if ENABLE_EFI
923 if (MANAGER_IS_SYSTEM(m) && detect_container() <= 0)
924 boot_timestamps(m->timestamps + MANAGER_TIMESTAMP_USERSPACE,
925 m->timestamps + MANAGER_TIMESTAMP_FIRMWARE,
926 m->timestamps + MANAGER_TIMESTAMP_LOADER);
927 #endif
928
929 /* Prepare log fields we can use for structured logging */
930 if (MANAGER_IS_SYSTEM(m)) {
931 m->unit_log_field = "UNIT=";
932 m->unit_log_format_string = "UNIT=%s";
933
934 m->invocation_log_field = "INVOCATION_ID=";
935 m->invocation_log_format_string = "INVOCATION_ID=%s";
936 } else {
937 m->unit_log_field = "USER_UNIT=";
938 m->unit_log_format_string = "USER_UNIT=%s";
939
940 m->invocation_log_field = "USER_INVOCATION_ID=";
941 m->invocation_log_format_string = "USER_INVOCATION_ID=%s";
942 }
943
944 /* Reboot immediately if the user hits C-A-D more often than 7x per 2s */
945 m->ctrl_alt_del_ratelimit = (const RateLimit) { .interval = 2 * USEC_PER_SEC, .burst = 7 };
946
947 r = manager_default_environment(m);
948 if (r < 0)
949 return r;
950
951 r = hashmap_ensure_allocated(&m->units, &string_hash_ops);
952 if (r < 0)
953 return r;
954
955 r = hashmap_ensure_allocated(&m->cgroup_unit, &path_hash_ops);
956 if (r < 0)
957 return r;
958
959 r = hashmap_ensure_allocated(&m->watch_bus, &string_hash_ops);
960 if (r < 0)
961 return r;
962
963 r = prioq_ensure_allocated(&m->run_queue, compare_job_priority);
964 if (r < 0)
965 return r;
966
967 r = manager_setup_prefix(m);
968 if (r < 0)
969 return r;
970
971 r = manager_find_credentials_dirs(m);
972 if (r < 0)
973 return r;
974
975 r = sd_event_default(&m->event);
976 if (r < 0)
977 return r;
978
979 r = manager_setup_run_queue(m);
980 if (r < 0)
981 return r;
982
983 if (FLAGS_SET(test_run_flags, MANAGER_TEST_RUN_MINIMAL)) {
984 m->cgroup_root = strdup("");
985 if (!m->cgroup_root)
986 return -ENOMEM;
987 } else {
988 r = manager_setup_signals(m);
989 if (r < 0)
990 return r;
991
992 r = manager_setup_cgroup(m);
993 if (r < 0)
994 return r;
995
996 r = manager_setup_time_change(m);
997 if (r < 0)
998 return r;
999
1000 r = manager_read_timezone_stat(m);
1001 if (r < 0)
1002 return r;
1003
1004 (void) manager_setup_timezone_change(m);
1005
1006 r = manager_setup_sigchld_event_source(m);
1007 if (r < 0)
1008 return r;
1009
1010 r = manager_setup_memory_pressure_event_source(m);
1011 if (r < 0)
1012 return r;
1013
1014 #if HAVE_LIBBPF
1015 if (MANAGER_IS_SYSTEM(m) && lsm_bpf_supported(/* initialize = */ true)) {
1016 r = lsm_bpf_setup(m);
1017 if (r < 0)
1018 log_warning_errno(r, "Failed to setup LSM BPF, ignoring: %m");
1019 }
1020 #endif
1021 }
1022
1023 if (test_run_flags == 0) {
1024 if (MANAGER_IS_SYSTEM(m))
1025 r = mkdir_label("/run/systemd/units", 0755);
1026 else {
1027 _cleanup_free_ char *units_path = NULL;
1028 r = xdg_user_runtime_dir(&units_path, "/systemd/units");
1029 if (r < 0)
1030 return r;
1031 r = mkdir_p_label(units_path, 0755);
1032 }
1033
1034 if (r < 0 && r != -EEXIST)
1035 return r;
1036 }
1037
1038 m->taint_usr =
1039 !in_initrd() &&
1040 dir_is_empty("/usr", /* ignore_hidden_or_backup= */ false) > 0;
1041
1042 /* Note that we do not set up the notify fd here. We do that after deserialization,
1043 * since they might have gotten serialized across the reexec. */
1044
1045 *_m = TAKE_PTR(m);
1046
1047 return 0;
1048 }
1049
1050 static int manager_setup_notify(Manager *m) {
1051 int r;
1052
1053 if (MANAGER_IS_TEST_RUN(m))
1054 return 0;
1055
1056 if (m->notify_fd < 0) {
1057 _cleanup_close_ int fd = -EBADF;
1058 union sockaddr_union sa;
1059 socklen_t sa_len;
1060
1061 /* First free all secondary fields */
1062 m->notify_socket = mfree(m->notify_socket);
1063 m->notify_event_source = sd_event_source_disable_unref(m->notify_event_source);
1064
1065 fd = socket(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC|SOCK_NONBLOCK, 0);
1066 if (fd < 0)
1067 return log_error_errno(errno, "Failed to allocate notification socket: %m");
1068
1069 fd_increase_rxbuf(fd, NOTIFY_RCVBUF_SIZE);
1070
1071 m->notify_socket = path_join(m->prefix[EXEC_DIRECTORY_RUNTIME], "systemd/notify");
1072 if (!m->notify_socket)
1073 return log_oom();
1074
1075 r = sockaddr_un_set_path(&sa.un, m->notify_socket);
1076 if (r < 0)
1077 return log_error_errno(r, "Notify socket '%s' not valid for AF_UNIX socket address, refusing.",
1078 m->notify_socket);
1079 sa_len = r;
1080
1081 (void) mkdir_parents_label(m->notify_socket, 0755);
1082 (void) sockaddr_un_unlink(&sa.un);
1083
1084 r = mac_selinux_bind(fd, &sa.sa, sa_len);
1085 if (r < 0)
1086 return log_error_errno(r, "bind(%s) failed: %m", m->notify_socket);
1087
1088 r = setsockopt_int(fd, SOL_SOCKET, SO_PASSCRED, true);
1089 if (r < 0)
1090 return log_error_errno(r, "SO_PASSCRED failed: %m");
1091
1092 m->notify_fd = TAKE_FD(fd);
1093
1094 log_debug("Using notification socket %s", m->notify_socket);
1095 }
1096
1097 if (!m->notify_event_source) {
1098 r = sd_event_add_io(m->event, &m->notify_event_source, m->notify_fd, EPOLLIN, manager_dispatch_notify_fd, m);
1099 if (r < 0)
1100 return log_error_errno(r, "Failed to allocate notify event source: %m");
1101
1102 /* Process notification messages a bit earlier than SIGCHLD, so that we can still identify to which
1103 * service an exit message belongs. */
1104 r = sd_event_source_set_priority(m->notify_event_source, SD_EVENT_PRIORITY_NORMAL-8);
1105 if (r < 0)
1106 return log_error_errno(r, "Failed to set priority of notify event source: %m");
1107
1108 (void) sd_event_source_set_description(m->notify_event_source, "manager-notify");
1109 }
1110
1111 return 0;
1112 }
1113
1114 static int manager_setup_cgroups_agent(Manager *m) {
1115
1116 static const union sockaddr_union sa = {
1117 .un.sun_family = AF_UNIX,
1118 .un.sun_path = "/run/systemd/cgroups-agent",
1119 };
1120 int r;
1121
1122 /* This creates a listening socket we receive cgroups agent messages on. We do not use D-Bus for delivering
1123 * these messages from the cgroups agent binary to PID 1, as the cgroups agent binary is very short-living, and
1124 * each instance of it needs a new D-Bus connection. Since D-Bus connections are SOCK_STREAM/AF_UNIX, on
1125 * overloaded systems the backlog of the D-Bus socket becomes relevant, as not more than the configured number
1126 * of D-Bus connections may be queued until the kernel will start dropping further incoming connections,
1127 * possibly resulting in lost cgroups agent messages. To avoid this, we'll use a private SOCK_DGRAM/AF_UNIX
1128 * socket, where no backlog is relevant as communication may take place without an actual connect() cycle, and
1129 * we thus won't lose messages.
1130 *
1131 * Note that PID 1 will forward the agent message to system bus, so that the user systemd instance may listen
1132 * to it. The system instance hence listens on this special socket, but the user instances listen on the system
1133 * bus for these messages. */
1134
1135 if (MANAGER_IS_TEST_RUN(m))
1136 return 0;
1137
1138 if (!MANAGER_IS_SYSTEM(m))
1139 return 0;
1140
1141 r = cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER);
1142 if (r < 0)
1143 return log_error_errno(r, "Failed to determine whether unified cgroups hierarchy is used: %m");
1144 if (r > 0) /* We don't need this anymore on the unified hierarchy */
1145 return 0;
1146
1147 if (m->cgroups_agent_fd < 0) {
1148 _cleanup_close_ int fd = -EBADF;
1149
1150 /* First free all secondary fields */
1151 m->cgroups_agent_event_source = sd_event_source_disable_unref(m->cgroups_agent_event_source);
1152
1153 fd = socket(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC|SOCK_NONBLOCK, 0);
1154 if (fd < 0)
1155 return log_error_errno(errno, "Failed to allocate cgroups agent socket: %m");
1156
1157 fd_increase_rxbuf(fd, CGROUPS_AGENT_RCVBUF_SIZE);
1158
1159 (void) sockaddr_un_unlink(&sa.un);
1160
1161 /* Only allow root to connect to this socket */
1162 WITH_UMASK(0077)
1163 r = bind(fd, &sa.sa, SOCKADDR_UN_LEN(sa.un));
1164 if (r < 0)
1165 return log_error_errno(errno, "bind(%s) failed: %m", sa.un.sun_path);
1166
1167 m->cgroups_agent_fd = TAKE_FD(fd);
1168 }
1169
1170 if (!m->cgroups_agent_event_source) {
1171 r = sd_event_add_io(m->event, &m->cgroups_agent_event_source, m->cgroups_agent_fd, EPOLLIN, manager_dispatch_cgroups_agent_fd, m);
1172 if (r < 0)
1173 return log_error_errno(r, "Failed to allocate cgroups agent event source: %m");
1174
1175 /* Process cgroups notifications early. Note that when the agent notification is received
1176 * we'll just enqueue the unit in the cgroup empty queue, hence pick a high priority than
1177 * that. Also see handling of cgroup inotify for the unified cgroup stuff. */
1178 r = sd_event_source_set_priority(m->cgroups_agent_event_source, SD_EVENT_PRIORITY_NORMAL-9);
1179 if (r < 0)
1180 return log_error_errno(r, "Failed to set priority of cgroups agent event source: %m");
1181
1182 (void) sd_event_source_set_description(m->cgroups_agent_event_source, "manager-cgroups-agent");
1183 }
1184
1185 return 0;
1186 }
1187
1188 static int manager_setup_user_lookup_fd(Manager *m) {
1189 int r;
1190
1191 assert(m);
1192
1193 /* Set up the socket pair used for passing UID/GID resolution results from forked off processes to PID
1194 * 1. Background: we can't do name lookups (NSS) from PID 1, since it might involve IPC and thus activation,
1195 * and we might hence deadlock on ourselves. Hence we do all user/group lookups asynchronously from the forked
1196 * off processes right before executing the binaries to start. In order to be able to clean up any IPC objects
1197 * created by a unit (see RemoveIPC=) we need to know in PID 1 the used UID/GID of the executed processes,
1198 * hence we establish this communication channel so that forked off processes can pass their UID/GID
1199 * information back to PID 1. The forked off processes send their resolved UID/GID to PID 1 in a simple
1200 * datagram, along with their unit name, so that we can share one communication socket pair among all units for
1201 * this purpose.
1202 *
1203 * You might wonder why we need a communication channel for this that is independent of the usual notification
1204 * socket scheme (i.e. $NOTIFY_SOCKET). The primary difference is about trust: data sent via the $NOTIFY_SOCKET
1205 * channel is only accepted if it originates from the right unit and if reception was enabled for it. The user
1206 * lookup socket OTOH is only accessible by PID 1 and its children until they exec(), and always available.
1207 *
1208 * Note that this function is called under two circumstances: when we first initialize (in which case we
1209 * allocate both the socket pair and the event source to listen on it), and when we deserialize after a reload
1210 * (in which case the socket pair already exists but we still need to allocate the event source for it). */
1211
1212 if (m->user_lookup_fds[0] < 0) {
1213
1214 /* Free all secondary fields */
1215 safe_close_pair(m->user_lookup_fds);
1216 m->user_lookup_event_source = sd_event_source_disable_unref(m->user_lookup_event_source);
1217
1218 if (socketpair(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC, 0, m->user_lookup_fds) < 0)
1219 return log_error_errno(errno, "Failed to allocate user lookup socket: %m");
1220
1221 (void) fd_increase_rxbuf(m->user_lookup_fds[0], NOTIFY_RCVBUF_SIZE);
1222 }
1223
1224 if (!m->user_lookup_event_source) {
1225 r = sd_event_add_io(m->event, &m->user_lookup_event_source, m->user_lookup_fds[0], EPOLLIN, manager_dispatch_user_lookup_fd, m);
1226 if (r < 0)
1227 return log_error_errno(errno, "Failed to allocate user lookup event source: %m");
1228
1229 /* Process even earlier than the notify event source, so that we always know first about valid UID/GID
1230 * resolutions */
1231 r = sd_event_source_set_priority(m->user_lookup_event_source, SD_EVENT_PRIORITY_NORMAL-11);
1232 if (r < 0)
1233 return log_error_errno(errno, "Failed to set priority of user lookup event source: %m");
1234
1235 (void) sd_event_source_set_description(m->user_lookup_event_source, "user-lookup");
1236 }
1237
1238 return 0;
1239 }
1240
1241 static unsigned manager_dispatch_cleanup_queue(Manager *m) {
1242 Unit *u;
1243 unsigned n = 0;
1244
1245 assert(m);
1246
1247 while ((u = m->cleanup_queue)) {
1248 assert(u->in_cleanup_queue);
1249
1250 unit_free(u);
1251 n++;
1252 }
1253
1254 return n;
1255 }
1256
1257 static unsigned manager_dispatch_release_resources_queue(Manager *m) {
1258 unsigned n = 0;
1259 Unit *u;
1260
1261 assert(m);
1262
1263 while ((u = m->release_resources_queue)) {
1264 assert(u->in_release_resources_queue);
1265
1266 LIST_REMOVE(release_resources_queue, m->release_resources_queue, u);
1267 u->in_release_resources_queue = false;
1268
1269 n++;
1270
1271 unit_release_resources(u);
1272 }
1273
1274 return n;
1275 }
1276
1277 enum {
1278 GC_OFFSET_IN_PATH, /* This one is on the path we were traveling */
1279 GC_OFFSET_UNSURE, /* No clue */
1280 GC_OFFSET_GOOD, /* We still need this unit */
1281 GC_OFFSET_BAD, /* We don't need this unit anymore */
1282 _GC_OFFSET_MAX
1283 };
1284
1285 static void unit_gc_mark_good(Unit *u, unsigned gc_marker) {
1286 Unit *other;
1287
1288 u->gc_marker = gc_marker + GC_OFFSET_GOOD;
1289
1290 /* Recursively mark referenced units as GOOD as well */
1291 UNIT_FOREACH_DEPENDENCY(other, u, UNIT_ATOM_REFERENCES)
1292 if (other->gc_marker == gc_marker + GC_OFFSET_UNSURE)
1293 unit_gc_mark_good(other, gc_marker);
1294 }
1295
1296 static void unit_gc_sweep(Unit *u, unsigned gc_marker) {
1297 Unit *other;
1298 bool is_bad;
1299
1300 assert(u);
1301
1302 if (IN_SET(u->gc_marker - gc_marker,
1303 GC_OFFSET_GOOD, GC_OFFSET_BAD, GC_OFFSET_UNSURE, GC_OFFSET_IN_PATH))
1304 return;
1305
1306 if (u->in_cleanup_queue)
1307 goto bad;
1308
1309 if (!unit_may_gc(u))
1310 goto good;
1311
1312 u->gc_marker = gc_marker + GC_OFFSET_IN_PATH;
1313
1314 is_bad = true;
1315
1316 UNIT_FOREACH_DEPENDENCY(other, u, UNIT_ATOM_REFERENCED_BY) {
1317 unit_gc_sweep(other, gc_marker);
1318
1319 if (other->gc_marker == gc_marker + GC_OFFSET_GOOD)
1320 goto good;
1321
1322 if (other->gc_marker != gc_marker + GC_OFFSET_BAD)
1323 is_bad = false;
1324 }
1325
1326 LIST_FOREACH(refs_by_target, ref, u->refs_by_target) {
1327 unit_gc_sweep(ref->source, gc_marker);
1328
1329 if (ref->source->gc_marker == gc_marker + GC_OFFSET_GOOD)
1330 goto good;
1331
1332 if (ref->source->gc_marker != gc_marker + GC_OFFSET_BAD)
1333 is_bad = false;
1334 }
1335
1336 if (is_bad)
1337 goto bad;
1338
1339 /* We were unable to find anything out about this entry, so
1340 * let's investigate it later */
1341 u->gc_marker = gc_marker + GC_OFFSET_UNSURE;
1342 unit_add_to_gc_queue(u);
1343 return;
1344
1345 bad:
1346 /* We definitely know that this one is not useful anymore, so
1347 * let's mark it for deletion */
1348 u->gc_marker = gc_marker + GC_OFFSET_BAD;
1349 unit_add_to_cleanup_queue(u);
1350 return;
1351
1352 good:
1353 unit_gc_mark_good(u, gc_marker);
1354 }
1355
1356 static unsigned manager_dispatch_gc_unit_queue(Manager *m) {
1357 unsigned n = 0, gc_marker;
1358 Unit *u;
1359
1360 assert(m);
1361
1362 /* log_debug("Running GC..."); */
1363
1364 m->gc_marker += _GC_OFFSET_MAX;
1365 if (m->gc_marker + _GC_OFFSET_MAX <= _GC_OFFSET_MAX)
1366 m->gc_marker = 1;
1367
1368 gc_marker = m->gc_marker;
1369
1370 while ((u = m->gc_unit_queue)) {
1371 assert(u->in_gc_queue);
1372
1373 unit_gc_sweep(u, gc_marker);
1374
1375 LIST_REMOVE(gc_queue, m->gc_unit_queue, u);
1376 u->in_gc_queue = false;
1377
1378 n++;
1379
1380 if (IN_SET(u->gc_marker - gc_marker,
1381 GC_OFFSET_BAD, GC_OFFSET_UNSURE)) {
1382 if (u->id)
1383 log_unit_debug(u, "Collecting.");
1384 u->gc_marker = gc_marker + GC_OFFSET_BAD;
1385 unit_add_to_cleanup_queue(u);
1386 }
1387 }
1388
1389 return n;
1390 }
1391
1392 static unsigned manager_dispatch_gc_job_queue(Manager *m) {
1393 unsigned n = 0;
1394 Job *j;
1395
1396 assert(m);
1397
1398 while ((j = m->gc_job_queue)) {
1399 assert(j->in_gc_queue);
1400
1401 LIST_REMOVE(gc_queue, m->gc_job_queue, j);
1402 j->in_gc_queue = false;
1403
1404 n++;
1405
1406 if (!job_may_gc(j))
1407 continue;
1408
1409 log_unit_debug(j->unit, "Collecting job.");
1410 (void) job_finish_and_invalidate(j, JOB_COLLECTED, false, false);
1411 }
1412
1413 return n;
1414 }
1415
1416 static int manager_ratelimit_requeue(sd_event_source *s, uint64_t usec, void *userdata) {
1417 Unit *u = userdata;
1418
1419 assert(u);
1420 assert(s == u->auto_start_stop_event_source);
1421
1422 u->auto_start_stop_event_source = sd_event_source_unref(u->auto_start_stop_event_source);
1423
1424 /* Re-queue to all queues, if the rate limit hit we might have been throttled on any of them. */
1425 unit_submit_to_stop_when_unneeded_queue(u);
1426 unit_submit_to_start_when_upheld_queue(u);
1427 unit_submit_to_stop_when_bound_queue(u);
1428
1429 return 0;
1430 }
1431
1432 static int manager_ratelimit_check_and_queue(Unit *u) {
1433 int r;
1434
1435 assert(u);
1436
1437 if (ratelimit_below(&u->auto_start_stop_ratelimit))
1438 return 1;
1439
1440 /* Already queued, no need to requeue */
1441 if (u->auto_start_stop_event_source)
1442 return 0;
1443
1444 r = sd_event_add_time(
1445 u->manager->event,
1446 &u->auto_start_stop_event_source,
1447 CLOCK_MONOTONIC,
1448 ratelimit_end(&u->auto_start_stop_ratelimit),
1449 0,
1450 manager_ratelimit_requeue,
1451 u);
1452 if (r < 0)
1453 return log_unit_error_errno(u, r, "Failed to queue timer on event loop: %m");
1454
1455 return 0;
1456 }
1457
1458 static unsigned manager_dispatch_stop_when_unneeded_queue(Manager *m) {
1459 unsigned n = 0;
1460 Unit *u;
1461 int r;
1462
1463 assert(m);
1464
1465 while ((u = m->stop_when_unneeded_queue)) {
1466 _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL;
1467
1468 assert(u->in_stop_when_unneeded_queue);
1469 LIST_REMOVE(stop_when_unneeded_queue, m->stop_when_unneeded_queue, u);
1470 u->in_stop_when_unneeded_queue = false;
1471
1472 n++;
1473
1474 if (!unit_is_unneeded(u))
1475 continue;
1476
1477 log_unit_debug(u, "Unit is not needed anymore.");
1478
1479 /* If stopping a unit fails continuously we might enter a stop loop here, hence stop acting on the
1480 * service being unnecessary after a while. */
1481
1482 r = manager_ratelimit_check_and_queue(u);
1483 if (r <= 0) {
1484 log_unit_warning(u,
1485 "Unit not needed anymore, but not stopping since we tried this too often recently.%s",
1486 r == 0 ? " Will retry later." : "");
1487 continue;
1488 }
1489
1490 /* Ok, nobody needs us anymore. Sniff. Then let's commit suicide */
1491 r = manager_add_job(u->manager, JOB_STOP, u, JOB_FAIL, NULL, &error, NULL);
1492 if (r < 0)
1493 log_unit_warning_errno(u, r, "Failed to enqueue stop job, ignoring: %s", bus_error_message(&error, r));
1494 }
1495
1496 return n;
1497 }
1498
1499 static unsigned manager_dispatch_start_when_upheld_queue(Manager *m) {
1500 unsigned n = 0;
1501 Unit *u;
1502 int r;
1503
1504 assert(m);
1505
1506 while ((u = m->start_when_upheld_queue)) {
1507 _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL;
1508 Unit *culprit = NULL;
1509
1510 assert(u->in_start_when_upheld_queue);
1511 LIST_REMOVE(start_when_upheld_queue, m->start_when_upheld_queue, u);
1512 u->in_start_when_upheld_queue = false;
1513
1514 n++;
1515
1516 if (!unit_is_upheld_by_active(u, &culprit))
1517 continue;
1518
1519 log_unit_debug(u, "Unit is started because upheld by active unit %s.", culprit->id);
1520
1521 /* If stopping a unit fails continuously we might enter a stop loop here, hence stop acting on the
1522 * service being unnecessary after a while. */
1523
1524 r = manager_ratelimit_check_and_queue(u);
1525 if (r <= 0) {
1526 log_unit_warning(u,
1527 "Unit needs to be started because active unit %s upholds it, but not starting since we tried this too often recently.%s",
1528 culprit->id,
1529 r == 0 ? " Will retry later." : "");
1530 continue;
1531 }
1532
1533 r = manager_add_job(u->manager, JOB_START, u, JOB_FAIL, NULL, &error, NULL);
1534 if (r < 0)
1535 log_unit_warning_errno(u, r, "Failed to enqueue start job, ignoring: %s", bus_error_message(&error, r));
1536 }
1537
1538 return n;
1539 }
1540
1541 static unsigned manager_dispatch_stop_when_bound_queue(Manager *m) {
1542 unsigned n = 0;
1543 Unit *u;
1544 int r;
1545
1546 assert(m);
1547
1548 while ((u = m->stop_when_bound_queue)) {
1549 _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL;
1550 Unit *culprit = NULL;
1551
1552 assert(u->in_stop_when_bound_queue);
1553 LIST_REMOVE(stop_when_bound_queue, m->stop_when_bound_queue, u);
1554 u->in_stop_when_bound_queue = false;
1555
1556 n++;
1557
1558 if (!unit_is_bound_by_inactive(u, &culprit))
1559 continue;
1560
1561 log_unit_debug(u, "Unit is stopped because bound to inactive unit %s.", culprit->id);
1562
1563 /* If stopping a unit fails continuously we might enter a stop loop here, hence stop acting on the
1564 * service being unnecessary after a while. */
1565
1566 r = manager_ratelimit_check_and_queue(u);
1567 if (r <= 0) {
1568 log_unit_warning(u,
1569 "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",
1570 culprit->id,
1571 r == 0 ? " Will retry later." : "");
1572 continue;
1573 }
1574
1575 r = manager_add_job(u->manager, JOB_STOP, u, JOB_REPLACE, NULL, &error, NULL);
1576 if (r < 0)
1577 log_unit_warning_errno(u, r, "Failed to enqueue stop job, ignoring: %s", bus_error_message(&error, r));
1578 }
1579
1580 return n;
1581 }
1582
1583 static void manager_clear_jobs_and_units(Manager *m) {
1584 Unit *u;
1585
1586 assert(m);
1587
1588 while ((u = hashmap_first(m->units)))
1589 unit_free(u);
1590
1591 manager_dispatch_cleanup_queue(m);
1592
1593 assert(!m->load_queue);
1594 assert(prioq_isempty(m->run_queue));
1595 assert(!m->dbus_unit_queue);
1596 assert(!m->dbus_job_queue);
1597 assert(!m->cleanup_queue);
1598 assert(!m->gc_unit_queue);
1599 assert(!m->gc_job_queue);
1600 assert(!m->cgroup_realize_queue);
1601 assert(!m->cgroup_empty_queue);
1602 assert(!m->cgroup_oom_queue);
1603 assert(!m->target_deps_queue);
1604 assert(!m->stop_when_unneeded_queue);
1605 assert(!m->start_when_upheld_queue);
1606 assert(!m->stop_when_bound_queue);
1607 assert(!m->release_resources_queue);
1608
1609 assert(hashmap_isempty(m->jobs));
1610 assert(hashmap_isempty(m->units));
1611
1612 m->n_on_console = 0;
1613 m->n_running_jobs = 0;
1614 m->n_installed_jobs = 0;
1615 m->n_failed_jobs = 0;
1616 }
1617
1618 Manager* manager_free(Manager *m) {
1619 if (!m)
1620 return NULL;
1621
1622 manager_clear_jobs_and_units(m);
1623
1624 for (UnitType c = 0; c < _UNIT_TYPE_MAX; c++)
1625 if (unit_vtable[c]->shutdown)
1626 unit_vtable[c]->shutdown(m);
1627
1628 /* Keep the cgroup hierarchy in place except when we know we are going down for good */
1629 manager_shutdown_cgroup(m, /* delete= */ IN_SET(m->objective, MANAGER_EXIT, MANAGER_REBOOT, MANAGER_POWEROFF, MANAGER_HALT, MANAGER_KEXEC));
1630
1631 lookup_paths_flush_generator(&m->lookup_paths);
1632
1633 bus_done(m);
1634 manager_varlink_done(m);
1635
1636 exec_shared_runtime_vacuum(m);
1637 hashmap_free(m->exec_shared_runtime_by_id);
1638
1639 dynamic_user_vacuum(m, false);
1640 hashmap_free(m->dynamic_users);
1641
1642 hashmap_free(m->units);
1643 hashmap_free(m->units_by_invocation_id);
1644 hashmap_free(m->jobs);
1645 hashmap_free(m->watch_pids);
1646 hashmap_free(m->watch_bus);
1647
1648 prioq_free(m->run_queue);
1649
1650 set_free(m->startup_units);
1651 set_free(m->failed_units);
1652
1653 sd_event_source_unref(m->signal_event_source);
1654 sd_event_source_unref(m->sigchld_event_source);
1655 sd_event_source_unref(m->notify_event_source);
1656 sd_event_source_unref(m->cgroups_agent_event_source);
1657 sd_event_source_unref(m->time_change_event_source);
1658 sd_event_source_unref(m->timezone_change_event_source);
1659 sd_event_source_unref(m->jobs_in_progress_event_source);
1660 sd_event_source_unref(m->run_queue_event_source);
1661 sd_event_source_unref(m->user_lookup_event_source);
1662 sd_event_source_unref(m->memory_pressure_event_source);
1663
1664 safe_close(m->signal_fd);
1665 safe_close(m->notify_fd);
1666 safe_close(m->cgroups_agent_fd);
1667 safe_close_pair(m->user_lookup_fds);
1668
1669 manager_close_ask_password(m);
1670
1671 manager_close_idle_pipe(m);
1672
1673 sd_event_unref(m->event);
1674
1675 free(m->notify_socket);
1676
1677 lookup_paths_free(&m->lookup_paths);
1678 strv_free(m->transient_environment);
1679 strv_free(m->client_environment);
1680
1681 hashmap_free(m->cgroup_unit);
1682 manager_free_unit_name_maps(m);
1683
1684 free(m->switch_root);
1685 free(m->switch_root_init);
1686
1687 free(m->default_smack_process_label);
1688
1689 rlimit_free_all(m->rlimit);
1690
1691 assert(hashmap_isempty(m->units_requiring_mounts_for));
1692 hashmap_free(m->units_requiring_mounts_for);
1693
1694 hashmap_free(m->uid_refs);
1695 hashmap_free(m->gid_refs);
1696
1697 for (ExecDirectoryType dt = 0; dt < _EXEC_DIRECTORY_TYPE_MAX; dt++)
1698 m->prefix[dt] = mfree(m->prefix[dt]);
1699 free(m->received_credentials_directory);
1700 free(m->received_encrypted_credentials_directory);
1701
1702 free(m->watchdog_pretimeout_governor);
1703 free(m->watchdog_pretimeout_governor_overridden);
1704
1705 #if BPF_FRAMEWORK
1706 lsm_bpf_destroy(m->restrict_fs);
1707 #endif
1708
1709 return mfree(m);
1710 }
1711
1712 static void manager_enumerate_perpetual(Manager *m) {
1713 assert(m);
1714
1715 if (FLAGS_SET(m->test_run_flags, MANAGER_TEST_RUN_MINIMAL))
1716 return;
1717
1718 /* Let's ask every type to load all units from disk/kernel that it might know */
1719 for (UnitType c = 0; c < _UNIT_TYPE_MAX; c++) {
1720 if (!unit_type_supported(c)) {
1721 log_debug("Unit type .%s is not supported on this system.", unit_type_to_string(c));
1722 continue;
1723 }
1724
1725 if (unit_vtable[c]->enumerate_perpetual)
1726 unit_vtable[c]->enumerate_perpetual(m);
1727 }
1728 }
1729
1730 static void manager_enumerate(Manager *m) {
1731 assert(m);
1732
1733 if (FLAGS_SET(m->test_run_flags, MANAGER_TEST_RUN_MINIMAL))
1734 return;
1735
1736 /* Let's ask every type to load all units from disk/kernel that it might know */
1737 for (UnitType c = 0; c < _UNIT_TYPE_MAX; c++) {
1738 if (!unit_type_supported(c)) {
1739 log_debug("Unit type .%s is not supported on this system.", unit_type_to_string(c));
1740 continue;
1741 }
1742
1743 if (unit_vtable[c]->enumerate)
1744 unit_vtable[c]->enumerate(m);
1745 }
1746
1747 manager_dispatch_load_queue(m);
1748 }
1749
1750 static void manager_coldplug(Manager *m) {
1751 Unit *u;
1752 char *k;
1753 int r;
1754
1755 assert(m);
1756
1757 log_debug("Invoking unit coldplug() handlers%s", special_glyph(SPECIAL_GLYPH_ELLIPSIS));
1758
1759 /* Let's place the units back into their deserialized state */
1760 HASHMAP_FOREACH_KEY(u, k, m->units) {
1761
1762 /* ignore aliases */
1763 if (u->id != k)
1764 continue;
1765
1766 r = unit_coldplug(u);
1767 if (r < 0)
1768 log_warning_errno(r, "We couldn't coldplug %s, proceeding anyway: %m", u->id);
1769 }
1770 }
1771
1772 static void manager_catchup(Manager *m) {
1773 Unit *u;
1774 char *k;
1775
1776 assert(m);
1777
1778 log_debug("Invoking unit catchup() handlers%s", special_glyph(SPECIAL_GLYPH_ELLIPSIS));
1779
1780 /* Let's catch up on any state changes that happened while we were reloading/reexecing */
1781 HASHMAP_FOREACH_KEY(u, k, m->units) {
1782
1783 /* ignore aliases */
1784 if (u->id != k)
1785 continue;
1786
1787 unit_catchup(u);
1788 }
1789 }
1790
1791 static void manager_distribute_fds(Manager *m, FDSet *fds) {
1792 Unit *u;
1793
1794 assert(m);
1795
1796 HASHMAP_FOREACH(u, m->units) {
1797
1798 if (fdset_size(fds) <= 0)
1799 break;
1800
1801 if (!UNIT_VTABLE(u)->distribute_fds)
1802 continue;
1803
1804 UNIT_VTABLE(u)->distribute_fds(u, fds);
1805 }
1806 }
1807
1808 static bool manager_dbus_is_running(Manager *m, bool deserialized) {
1809 Unit *u;
1810
1811 assert(m);
1812
1813 /* This checks whether the dbus instance we are supposed to expose our APIs on is up. We check both the socket
1814 * and the service unit. If the 'deserialized' parameter is true we'll check the deserialized state of the unit
1815 * rather than the current one. */
1816
1817 if (MANAGER_IS_TEST_RUN(m))
1818 return false;
1819
1820 u = manager_get_unit(m, SPECIAL_DBUS_SOCKET);
1821 if (!u)
1822 return false;
1823 if ((deserialized ? SOCKET(u)->deserialized_state : SOCKET(u)->state) != SOCKET_RUNNING)
1824 return false;
1825
1826 u = manager_get_unit(m, SPECIAL_DBUS_SERVICE);
1827 if (!u)
1828 return false;
1829 if (!IN_SET((deserialized ? SERVICE(u)->deserialized_state : SERVICE(u)->state),
1830 SERVICE_RUNNING,
1831 SERVICE_RELOAD,
1832 SERVICE_RELOAD_NOTIFY,
1833 SERVICE_RELOAD_SIGNAL))
1834 return false;
1835
1836 return true;
1837 }
1838
1839 static void manager_setup_bus(Manager *m) {
1840 assert(m);
1841
1842 /* Let's set up our private bus connection now, unconditionally */
1843 (void) bus_init_private(m);
1844
1845 /* If we are in --user mode also connect to the system bus now */
1846 if (MANAGER_IS_USER(m))
1847 (void) bus_init_system(m);
1848
1849 /* Let's connect to the bus now, but only if the unit is supposed to be up */
1850 if (manager_dbus_is_running(m, MANAGER_IS_RELOADING(m))) {
1851 (void) bus_init_api(m);
1852
1853 if (MANAGER_IS_SYSTEM(m))
1854 (void) bus_init_system(m);
1855 }
1856 }
1857
1858 static void manager_preset_all(Manager *m) {
1859 int r;
1860
1861 assert(m);
1862
1863 if (m->first_boot <= 0)
1864 return;
1865
1866 if (!MANAGER_IS_SYSTEM(m))
1867 return;
1868
1869 if (MANAGER_IS_TEST_RUN(m))
1870 return;
1871
1872 /* If this is the first boot, and we are in the host system, then preset everything */
1873 UnitFilePresetMode mode = FIRST_BOOT_FULL_PRESET ? UNIT_FILE_PRESET_FULL : UNIT_FILE_PRESET_ENABLE_ONLY;
1874
1875 r = unit_file_preset_all(RUNTIME_SCOPE_SYSTEM, 0, NULL, mode, NULL, 0);
1876 if (r < 0)
1877 log_full_errno(r == -EEXIST ? LOG_NOTICE : LOG_WARNING, r,
1878 "Failed to populate /etc with preset unit settings, ignoring: %m");
1879 else
1880 log_info("Populated /etc with preset unit settings.");
1881 }
1882
1883 static void manager_ready(Manager *m) {
1884 assert(m);
1885
1886 /* After having loaded everything, do the final round of catching up with what might have changed */
1887
1888 m->objective = MANAGER_OK; /* Tell everyone we are up now */
1889
1890 /* It might be safe to log to the journal now and connect to dbus */
1891 manager_recheck_journal(m);
1892 manager_recheck_dbus(m);
1893
1894 /* Let's finally catch up with any changes that took place while we were reloading/reexecing */
1895 manager_catchup(m);
1896
1897 /* Create a file which will indicate when the manager started loading units the last time. */
1898 if (MANAGER_IS_SYSTEM(m))
1899 (void) touch_file("/run/systemd/systemd-units-load", false,
1900 m->timestamps[MANAGER_TIMESTAMP_UNITS_LOAD].realtime ?: now(CLOCK_REALTIME),
1901 UID_INVALID, GID_INVALID, 0444);
1902 }
1903
1904 Manager* manager_reloading_start(Manager *m) {
1905 m->n_reloading++;
1906 dual_timestamp_get(m->timestamps + MANAGER_TIMESTAMP_UNITS_LOAD);
1907 return m;
1908 }
1909
1910 void manager_reloading_stopp(Manager **m) {
1911 if (*m) {
1912 assert((*m)->n_reloading > 0);
1913 (*m)->n_reloading--;
1914 }
1915 }
1916
1917 int manager_startup(Manager *m, FILE *serialization, FDSet *fds, const char *root) {
1918 int r;
1919
1920 assert(m);
1921
1922 /* If we are running in test mode, we still want to run the generators,
1923 * but we should not touch the real generator directories. */
1924 r = lookup_paths_init_or_warn(&m->lookup_paths, m->runtime_scope,
1925 MANAGER_IS_TEST_RUN(m) ? LOOKUP_PATHS_TEMPORARY_GENERATED : 0,
1926 root);
1927 if (r < 0)
1928 return r;
1929
1930 dual_timestamp_get(m->timestamps + manager_timestamp_initrd_mangle(MANAGER_TIMESTAMP_GENERATORS_START));
1931 r = manager_run_environment_generators(m);
1932 if (r >= 0)
1933 r = manager_run_generators(m);
1934 dual_timestamp_get(m->timestamps + manager_timestamp_initrd_mangle(MANAGER_TIMESTAMP_GENERATORS_FINISH));
1935 if (r < 0)
1936 return r;
1937
1938 manager_preset_all(m);
1939
1940 lookup_paths_log(&m->lookup_paths);
1941
1942 {
1943 /* This block is (optionally) done with the reloading counter bumped */
1944 _unused_ _cleanup_(manager_reloading_stopp) Manager *reloading = NULL;
1945
1946 /* Make sure we don't have a left-over from a previous run */
1947 if (!serialization)
1948 (void) rm_rf(m->lookup_paths.transient, 0);
1949
1950 /* If we will deserialize make sure that during enumeration this is already known, so we increase the
1951 * counter here already */
1952 if (serialization)
1953 reloading = manager_reloading_start(m);
1954
1955 /* First, enumerate what we can from all config files */
1956 dual_timestamp_get(m->timestamps + manager_timestamp_initrd_mangle(MANAGER_TIMESTAMP_UNITS_LOAD_START));
1957 manager_enumerate_perpetual(m);
1958 manager_enumerate(m);
1959 dual_timestamp_get(m->timestamps + manager_timestamp_initrd_mangle(MANAGER_TIMESTAMP_UNITS_LOAD_FINISH));
1960
1961 /* Second, deserialize if there is something to deserialize */
1962 if (serialization) {
1963 r = manager_deserialize(m, serialization, fds);
1964 if (r < 0)
1965 return log_error_errno(r, "Deserialization failed: %m");
1966 }
1967
1968 /* Any fds left? Find some unit which wants them. This is useful to allow container managers to pass
1969 * some file descriptors to us pre-initialized. This enables socket-based activation of entire
1970 * containers. */
1971 manager_distribute_fds(m, fds);
1972
1973 /* We might have deserialized the notify fd, but if we didn't then let's create the bus now */
1974 r = manager_setup_notify(m);
1975 if (r < 0)
1976 /* No sense to continue without notifications, our children would fail anyway. */
1977 return r;
1978
1979 r = manager_setup_cgroups_agent(m);
1980 if (r < 0)
1981 /* Likewise, no sense to continue without empty cgroup notifications. */
1982 return r;
1983
1984 r = manager_setup_user_lookup_fd(m);
1985 if (r < 0)
1986 /* This shouldn't fail, except if things are really broken. */
1987 return r;
1988
1989 /* Connect to the bus if we are good for it */
1990 manager_setup_bus(m);
1991
1992 /* Now that we are connected to all possible buses, let's deserialize who is tracking us. */
1993 r = bus_track_coldplug(m, &m->subscribed, false, m->deserialized_subscribed);
1994 if (r < 0)
1995 log_warning_errno(r, "Failed to deserialized tracked clients, ignoring: %m");
1996 m->deserialized_subscribed = strv_free(m->deserialized_subscribed);
1997
1998 r = manager_varlink_init(m);
1999 if (r < 0)
2000 log_warning_errno(r, "Failed to set up Varlink, ignoring: %m");
2001
2002 /* Third, fire things up! */
2003 manager_coldplug(m);
2004
2005 /* Clean up runtime objects */
2006 manager_vacuum(m);
2007
2008 if (serialization)
2009 /* Let's wait for the UnitNew/JobNew messages being sent, before we notify that the
2010 * reload is finished */
2011 m->send_reloading_done = true;
2012 }
2013
2014 manager_ready(m);
2015
2016 manager_set_switching_root(m, false);
2017
2018 return 0;
2019 }
2020
2021 int manager_add_job(
2022 Manager *m,
2023 JobType type,
2024 Unit *unit,
2025 JobMode mode,
2026 Set *affected_jobs,
2027 sd_bus_error *error,
2028 Job **ret) {
2029
2030 _cleanup_(transaction_abort_and_freep) Transaction *tr = NULL;
2031 int r;
2032
2033 assert(m);
2034 assert(type < _JOB_TYPE_MAX);
2035 assert(unit);
2036 assert(mode < _JOB_MODE_MAX);
2037
2038 if (mode == JOB_ISOLATE && type != JOB_START)
2039 return sd_bus_error_set(error, SD_BUS_ERROR_INVALID_ARGS, "Isolate is only valid for start.");
2040
2041 if (mode == JOB_ISOLATE && !unit->allow_isolate)
2042 return sd_bus_error_set(error, BUS_ERROR_NO_ISOLATION, "Operation refused, unit may not be isolated.");
2043
2044 if (mode == JOB_TRIGGERING && type != JOB_STOP)
2045 return sd_bus_error_set(error, SD_BUS_ERROR_INVALID_ARGS, "--job-mode=triggering is only valid for stop.");
2046
2047 log_unit_debug(unit, "Trying to enqueue job %s/%s/%s", unit->id, job_type_to_string(type), job_mode_to_string(mode));
2048
2049 type = job_type_collapse(type, unit);
2050
2051 tr = transaction_new(mode == JOB_REPLACE_IRREVERSIBLY);
2052 if (!tr)
2053 return -ENOMEM;
2054
2055 r = transaction_add_job_and_dependencies(tr, type, unit, NULL, true, false,
2056 IN_SET(mode, JOB_IGNORE_DEPENDENCIES, JOB_IGNORE_REQUIREMENTS),
2057 mode == JOB_IGNORE_DEPENDENCIES, error);
2058 if (r < 0)
2059 return r;
2060
2061 if (mode == JOB_ISOLATE) {
2062 r = transaction_add_isolate_jobs(tr, m);
2063 if (r < 0)
2064 return r;
2065 }
2066
2067 if (mode == JOB_TRIGGERING) {
2068 r = transaction_add_triggering_jobs(tr, unit);
2069 if (r < 0)
2070 return r;
2071 }
2072
2073 r = transaction_activate(tr, m, mode, affected_jobs, error);
2074 if (r < 0)
2075 return r;
2076
2077 log_unit_debug(unit,
2078 "Enqueued job %s/%s as %u", unit->id,
2079 job_type_to_string(type), (unsigned) tr->anchor_job->id);
2080
2081 if (ret)
2082 *ret = tr->anchor_job;
2083
2084 tr = transaction_free(tr);
2085 return 0;
2086 }
2087
2088 int manager_add_job_by_name(Manager *m, JobType type, const char *name, JobMode mode, Set *affected_jobs, sd_bus_error *e, Job **ret) {
2089 Unit *unit = NULL; /* just to appease gcc, initialization is not really necessary */
2090 int r;
2091
2092 assert(m);
2093 assert(type < _JOB_TYPE_MAX);
2094 assert(name);
2095 assert(mode < _JOB_MODE_MAX);
2096
2097 r = manager_load_unit(m, name, NULL, NULL, &unit);
2098 if (r < 0)
2099 return r;
2100 assert(unit);
2101
2102 return manager_add_job(m, type, unit, mode, affected_jobs, e, ret);
2103 }
2104
2105 int manager_add_job_by_name_and_warn(Manager *m, JobType type, const char *name, JobMode mode, Set *affected_jobs, Job **ret) {
2106 _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL;
2107 int r;
2108
2109 assert(m);
2110 assert(type < _JOB_TYPE_MAX);
2111 assert(name);
2112 assert(mode < _JOB_MODE_MAX);
2113
2114 r = manager_add_job_by_name(m, type, name, mode, affected_jobs, &error, ret);
2115 if (r < 0)
2116 return log_warning_errno(r, "Failed to enqueue %s job for %s: %s", job_mode_to_string(mode), name, bus_error_message(&error, r));
2117
2118 return r;
2119 }
2120
2121 int manager_propagate_reload(Manager *m, Unit *unit, JobMode mode, sd_bus_error *e) {
2122 int r;
2123 _cleanup_(transaction_abort_and_freep) Transaction *tr = NULL;
2124
2125 assert(m);
2126 assert(unit);
2127 assert(mode < _JOB_MODE_MAX);
2128 assert(mode != JOB_ISOLATE); /* Isolate is only valid for start */
2129
2130 tr = transaction_new(mode == JOB_REPLACE_IRREVERSIBLY);
2131 if (!tr)
2132 return -ENOMEM;
2133
2134 /* We need an anchor job */
2135 r = transaction_add_job_and_dependencies(tr, JOB_NOP, unit, NULL, false, false, true, true, e);
2136 if (r < 0)
2137 return r;
2138
2139 /* Failure in adding individual dependencies is ignored, so this always succeeds. */
2140 transaction_add_propagate_reload_jobs(tr, unit, tr->anchor_job, mode == JOB_IGNORE_DEPENDENCIES, e);
2141
2142 r = transaction_activate(tr, m, mode, NULL, e);
2143 if (r < 0)
2144 return r;
2145
2146 tr = transaction_free(tr);
2147 return 0;
2148 }
2149
2150 Job *manager_get_job(Manager *m, uint32_t id) {
2151 assert(m);
2152
2153 return hashmap_get(m->jobs, UINT32_TO_PTR(id));
2154 }
2155
2156 Unit *manager_get_unit(Manager *m, const char *name) {
2157 assert(m);
2158 assert(name);
2159
2160 return hashmap_get(m->units, name);
2161 }
2162
2163 static int manager_dispatch_target_deps_queue(Manager *m) {
2164 Unit *u;
2165 int r = 0;
2166
2167 assert(m);
2168
2169 while ((u = m->target_deps_queue)) {
2170 _cleanup_free_ Unit **targets = NULL;
2171 int n_targets;
2172
2173 assert(u->in_target_deps_queue);
2174
2175 LIST_REMOVE(target_deps_queue, u->manager->target_deps_queue, u);
2176 u->in_target_deps_queue = false;
2177
2178 /* Take an "atomic" snapshot of dependencies here, as the call below will likely modify the
2179 * dependencies, and we can't have it that hash tables we iterate through are modified while
2180 * we are iterating through them. */
2181 n_targets = unit_get_dependency_array(u, UNIT_ATOM_DEFAULT_TARGET_DEPENDENCIES, &targets);
2182 if (n_targets < 0)
2183 return n_targets;
2184
2185 for (int i = 0; i < n_targets; i++) {
2186 r = unit_add_default_target_dependency(u, targets[i]);
2187 if (r < 0)
2188 return r;
2189 }
2190 }
2191
2192 return r;
2193 }
2194
2195 unsigned manager_dispatch_load_queue(Manager *m) {
2196 Unit *u;
2197 unsigned n = 0;
2198
2199 assert(m);
2200
2201 /* Make sure we are not run recursively */
2202 if (m->dispatching_load_queue)
2203 return 0;
2204
2205 m->dispatching_load_queue = true;
2206
2207 /* Dispatches the load queue. Takes a unit from the queue and
2208 * tries to load its data until the queue is empty */
2209
2210 while ((u = m->load_queue)) {
2211 assert(u->in_load_queue);
2212
2213 unit_load(u);
2214 n++;
2215 }
2216
2217 m->dispatching_load_queue = false;
2218
2219 /* Dispatch the units waiting for their target dependencies to be added now, as all targets that we know about
2220 * should be loaded and have aliases resolved */
2221 (void) manager_dispatch_target_deps_queue(m);
2222
2223 return n;
2224 }
2225
2226 bool manager_unit_cache_should_retry_load(Unit *u) {
2227 assert(u);
2228
2229 /* Automatic reloading from disk only applies to units which were not found sometime in the past, and
2230 * the not-found stub is kept pinned in the unit graph by dependencies. For units that were
2231 * previously loaded, we don't do automatic reloading, and daemon-reload is necessary to update. */
2232 if (u->load_state != UNIT_NOT_FOUND)
2233 return false;
2234
2235 /* The cache has been updated since the last time we tried to load the unit. There might be new
2236 * fragment paths to read. */
2237 if (u->manager->unit_cache_timestamp_hash != u->fragment_not_found_timestamp_hash)
2238 return true;
2239
2240 /* The cache needs to be updated because there are modifications on disk. */
2241 return !lookup_paths_timestamp_hash_same(&u->manager->lookup_paths, u->manager->unit_cache_timestamp_hash, NULL);
2242 }
2243
2244 int manager_load_unit_prepare(
2245 Manager *m,
2246 const char *name,
2247 const char *path,
2248 sd_bus_error *e,
2249 Unit **ret) {
2250
2251 _cleanup_(unit_freep) Unit *cleanup_unit = NULL;
2252 _cleanup_free_ char *nbuf = NULL;
2253 int r;
2254
2255 assert(m);
2256 assert(ret);
2257 assert(name || path);
2258
2259 /* This will prepare the unit for loading, but not actually load anything from disk. */
2260
2261 if (path && !path_is_absolute(path))
2262 return sd_bus_error_setf(e, SD_BUS_ERROR_INVALID_ARGS, "Path %s is not absolute.", path);
2263
2264 if (!name) {
2265 r = path_extract_filename(path, &nbuf);
2266 if (r < 0)
2267 return r;
2268 if (r == O_DIRECTORY)
2269 return sd_bus_error_setf(e, SD_BUS_ERROR_INVALID_ARGS, "Path '%s' refers to directory, refusing.", path);
2270
2271 name = nbuf;
2272 }
2273
2274 UnitType t = unit_name_to_type(name);
2275
2276 if (t == _UNIT_TYPE_INVALID || !unit_name_is_valid(name, UNIT_NAME_PLAIN|UNIT_NAME_INSTANCE)) {
2277 if (unit_name_is_valid(name, UNIT_NAME_TEMPLATE))
2278 return sd_bus_error_setf(e, SD_BUS_ERROR_INVALID_ARGS, "Unit name %s is missing the instance name.", name);
2279
2280 return sd_bus_error_setf(e, SD_BUS_ERROR_INVALID_ARGS, "Unit name %s is not valid.", name);
2281 }
2282
2283 Unit *unit = manager_get_unit(m, name);
2284 if (unit) {
2285 /* The time-based cache allows to start new units without daemon-reload,
2286 * but if they are already referenced (because of dependencies or ordering)
2287 * then we have to force a load of the fragment. As an optimization, check
2288 * first if anything in the usual paths was modified since the last time
2289 * the cache was loaded. Also check if the last time an attempt to load the
2290 * unit was made was before the most recent cache refresh, so that we know
2291 * we need to try again — even if the cache is current, it might have been
2292 * updated in a different context before we had a chance to retry loading
2293 * this particular unit. */
2294 if (manager_unit_cache_should_retry_load(unit))
2295 unit->load_state = UNIT_STUB;
2296 else {
2297 *ret = unit;
2298 return 0; /* The unit was already loaded */
2299 }
2300 } else {
2301 unit = cleanup_unit = unit_new(m, unit_vtable[t]->object_size);
2302 if (!unit)
2303 return -ENOMEM;
2304 }
2305
2306 if (path) {
2307 r = free_and_strdup(&unit->fragment_path, path);
2308 if (r < 0)
2309 return r;
2310 }
2311
2312 r = unit_add_name(unit, name);
2313 if (r < 0)
2314 return r;
2315
2316 unit_add_to_load_queue(unit);
2317 unit_add_to_dbus_queue(unit);
2318 unit_add_to_gc_queue(unit);
2319
2320 *ret = unit;
2321 TAKE_PTR(cleanup_unit);
2322
2323 return 1; /* The unit was added the load queue */
2324 }
2325
2326 int manager_load_unit(
2327 Manager *m,
2328 const char *name,
2329 const char *path,
2330 sd_bus_error *e,
2331 Unit **ret) {
2332 int r;
2333
2334 assert(m);
2335 assert(ret);
2336
2337 /* This will load the unit config, but not actually start any services or anything. */
2338
2339 r = manager_load_unit_prepare(m, name, path, e, ret);
2340 if (r <= 0)
2341 return r;
2342
2343 /* Unit was newly loaded */
2344 manager_dispatch_load_queue(m);
2345 *ret = unit_follow_merge(*ret);
2346 return 0;
2347 }
2348
2349 int manager_load_startable_unit_or_warn(
2350 Manager *m,
2351 const char *name,
2352 const char *path,
2353 Unit **ret) {
2354
2355 /* Load a unit, make sure it loaded fully and is not masked. */
2356
2357 _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL;
2358 Unit *unit;
2359 int r;
2360
2361 r = manager_load_unit(m, name, path, &error, &unit);
2362 if (r < 0)
2363 return log_error_errno(r, "Failed to load %s %s: %s",
2364 name ? "unit" : "unit file", name ?: path,
2365 bus_error_message(&error, r));
2366
2367 r = bus_unit_validate_load_state(unit, &error);
2368 if (r < 0)
2369 return log_error_errno(r, "%s", bus_error_message(&error, r));
2370
2371 *ret = unit;
2372 return 0;
2373 }
2374
2375 void manager_clear_jobs(Manager *m) {
2376 Job *j;
2377
2378 assert(m);
2379
2380 while ((j = hashmap_first(m->jobs)))
2381 /* No need to recurse. We're cancelling all jobs. */
2382 job_finish_and_invalidate(j, JOB_CANCELED, false, false);
2383 }
2384
2385 void manager_unwatch_pid(Manager *m, pid_t pid) {
2386 assert(m);
2387
2388 /* First let's drop the unit keyed as "pid". */
2389 (void) hashmap_remove(m->watch_pids, PID_TO_PTR(pid));
2390
2391 /* Then, let's also drop the array keyed by -pid. */
2392 free(hashmap_remove(m->watch_pids, PID_TO_PTR(-pid)));
2393 }
2394
2395 static int manager_dispatch_run_queue(sd_event_source *source, void *userdata) {
2396 Manager *m = ASSERT_PTR(userdata);
2397 Job *j;
2398
2399 assert(source);
2400
2401 while ((j = prioq_peek(m->run_queue))) {
2402 assert(j->installed);
2403 assert(j->in_run_queue);
2404
2405 (void) job_run_and_invalidate(j);
2406 }
2407
2408 if (m->n_running_jobs > 0)
2409 manager_watch_jobs_in_progress(m);
2410
2411 if (m->n_on_console > 0)
2412 manager_watch_idle_pipe(m);
2413
2414 return 1;
2415 }
2416
2417 void manager_trigger_run_queue(Manager *m) {
2418 int r;
2419
2420 assert(m);
2421
2422 r = sd_event_source_set_enabled(
2423 m->run_queue_event_source,
2424 prioq_isempty(m->run_queue) ? SD_EVENT_OFF : SD_EVENT_ONESHOT);
2425 if (r < 0)
2426 log_warning_errno(r, "Failed to enable job run queue event source, ignoring: %m");
2427 }
2428
2429 static unsigned manager_dispatch_dbus_queue(Manager *m) {
2430 unsigned n = 0, budget;
2431 Unit *u;
2432 Job *j;
2433
2434 assert(m);
2435
2436 /* When we are reloading, let's not wait with generating signals, since we need to exit the manager as quickly
2437 * as we can. There's no point in throttling generation of signals in that case. */
2438 if (MANAGER_IS_RELOADING(m) || m->send_reloading_done || m->pending_reload_message)
2439 budget = UINT_MAX; /* infinite budget in this case */
2440 else {
2441 /* Anything to do at all? */
2442 if (!m->dbus_unit_queue && !m->dbus_job_queue)
2443 return 0;
2444
2445 /* Do we have overly many messages queued at the moment? If so, let's not enqueue more on top, let's
2446 * sit this cycle out, and process things in a later cycle when the queues got a bit emptier. */
2447 if (manager_bus_n_queued_write(m) > MANAGER_BUS_BUSY_THRESHOLD)
2448 return 0;
2449
2450 /* Only process a certain number of units/jobs per event loop iteration. Even if the bus queue wasn't
2451 * overly full before this call we shouldn't increase it in size too wildly in one step, and we
2452 * shouldn't monopolize CPU time with generating these messages. Note the difference in counting of
2453 * this "budget" and the "threshold" above: the "budget" is decreased only once per generated message,
2454 * regardless how many buses/direct connections it is enqueued on, while the "threshold" is applied to
2455 * each queued instance of bus message, i.e. if the same message is enqueued to five buses/direct
2456 * connections it will be counted five times. This difference in counting ("references"
2457 * vs. "instances") is primarily a result of the fact that it's easier to implement it this way,
2458 * however it also reflects the thinking that the "threshold" should put a limit on used queue memory,
2459 * i.e. space, while the "budget" should put a limit on time. Also note that the "threshold" is
2460 * currently chosen much higher than the "budget". */
2461 budget = MANAGER_BUS_MESSAGE_BUDGET;
2462 }
2463
2464 while (budget != 0 && (u = m->dbus_unit_queue)) {
2465
2466 assert(u->in_dbus_queue);
2467
2468 bus_unit_send_change_signal(u);
2469 n++;
2470
2471 if (budget != UINT_MAX)
2472 budget--;
2473 }
2474
2475 while (budget != 0 && (j = m->dbus_job_queue)) {
2476 assert(j->in_dbus_queue);
2477
2478 bus_job_send_change_signal(j);
2479 n++;
2480
2481 if (budget != UINT_MAX)
2482 budget--;
2483 }
2484
2485 if (m->send_reloading_done) {
2486 m->send_reloading_done = false;
2487 bus_manager_send_reloading(m, false);
2488 n++;
2489 }
2490
2491 if (m->pending_reload_message) {
2492 bus_send_pending_reload_message(m);
2493 n++;
2494 }
2495
2496 return n;
2497 }
2498
2499 static int manager_dispatch_cgroups_agent_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
2500 Manager *m = userdata;
2501 char buf[PATH_MAX];
2502 ssize_t n;
2503
2504 n = recv(fd, buf, sizeof(buf), 0);
2505 if (n < 0)
2506 return log_error_errno(errno, "Failed to read cgroups agent message: %m");
2507 if (n == 0) {
2508 log_error("Got zero-length cgroups agent message, ignoring.");
2509 return 0;
2510 }
2511 if ((size_t) n >= sizeof(buf)) {
2512 log_error("Got overly long cgroups agent message, ignoring.");
2513 return 0;
2514 }
2515
2516 if (memchr(buf, 0, n)) {
2517 log_error("Got cgroups agent message with embedded NUL byte, ignoring.");
2518 return 0;
2519 }
2520 buf[n] = 0;
2521
2522 manager_notify_cgroup_empty(m, buf);
2523 (void) bus_forward_agent_released(m, buf);
2524
2525 return 0;
2526 }
2527
2528 static bool manager_process_barrier_fd(char * const *tags, FDSet *fds) {
2529
2530 /* nothing else must be sent when using BARRIER=1 */
2531 if (strv_contains(tags, "BARRIER=1")) {
2532 if (strv_length(tags) != 1)
2533 log_warning("Extra notification messages sent with BARRIER=1, ignoring everything.");
2534 else if (fdset_size(fds) != 1)
2535 log_warning("Got incorrect number of fds with BARRIER=1, closing them.");
2536
2537 /* Drop the message if BARRIER=1 was found */
2538 return true;
2539 }
2540
2541 return false;
2542 }
2543
2544 static void manager_invoke_notify_message(
2545 Manager *m,
2546 Unit *u,
2547 const struct ucred *ucred,
2548 char * const *tags,
2549 FDSet *fds) {
2550
2551 assert(m);
2552 assert(u);
2553 assert(ucred);
2554 assert(tags);
2555
2556 if (u->notifygen == m->notifygen) /* Already invoked on this same unit in this same iteration? */
2557 return;
2558 u->notifygen = m->notifygen;
2559
2560 if (UNIT_VTABLE(u)->notify_message)
2561 UNIT_VTABLE(u)->notify_message(u, ucred, tags, fds);
2562
2563 else if (DEBUG_LOGGING) {
2564 _cleanup_free_ char *buf = NULL, *x = NULL, *y = NULL;
2565
2566 buf = strv_join(tags, ", ");
2567 if (buf)
2568 x = ellipsize(buf, 20, 90);
2569 if (x)
2570 y = cescape(x);
2571
2572 log_unit_debug(u, "Got notification message \"%s\", ignoring.", strnull(y));
2573 }
2574 }
2575
2576 static int manager_dispatch_notify_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
2577
2578 _cleanup_fdset_free_ FDSet *fds = NULL;
2579 Manager *m = ASSERT_PTR(userdata);
2580 char buf[NOTIFY_BUFFER_MAX+1];
2581 struct iovec iovec = {
2582 .iov_base = buf,
2583 .iov_len = sizeof(buf)-1,
2584 };
2585 CMSG_BUFFER_TYPE(CMSG_SPACE(sizeof(struct ucred)) +
2586 CMSG_SPACE(sizeof(int) * NOTIFY_FD_MAX)) control;
2587 struct msghdr msghdr = {
2588 .msg_iov = &iovec,
2589 .msg_iovlen = 1,
2590 .msg_control = &control,
2591 .msg_controllen = sizeof(control),
2592 };
2593
2594 struct cmsghdr *cmsg;
2595 struct ucred *ucred = NULL;
2596 _cleanup_free_ Unit **array_copy = NULL;
2597 _cleanup_strv_free_ char **tags = NULL;
2598 Unit *u1, *u2, **array;
2599 int r, *fd_array = NULL;
2600 size_t n_fds = 0;
2601 bool found = false;
2602 ssize_t n;
2603
2604 assert(m->notify_fd == fd);
2605
2606 if (revents != EPOLLIN) {
2607 log_warning("Got unexpected poll event for notify fd.");
2608 return 0;
2609 }
2610
2611 n = recvmsg_safe(m->notify_fd, &msghdr, MSG_DONTWAIT|MSG_CMSG_CLOEXEC|MSG_TRUNC);
2612 if (n < 0) {
2613 if (ERRNO_IS_TRANSIENT(n))
2614 return 0; /* Spurious wakeup, try again */
2615 if (n == -EXFULL) {
2616 log_warning("Got message with truncated control data (too many fds sent?), ignoring.");
2617 return 0;
2618 }
2619 /* If this is any other, real error, then let's stop processing this socket. This of course
2620 * means we won't take notification messages anymore, but that's still better than busy
2621 * looping around this: being woken up over and over again but being unable to actually read
2622 * the message off the socket. */
2623 return log_error_errno(n, "Failed to receive notification message: %m");
2624 }
2625
2626 CMSG_FOREACH(cmsg, &msghdr) {
2627 if (cmsg->cmsg_level == SOL_SOCKET && cmsg->cmsg_type == SCM_RIGHTS) {
2628
2629 assert(!fd_array);
2630 fd_array = CMSG_TYPED_DATA(cmsg, int);
2631 n_fds = (cmsg->cmsg_len - CMSG_LEN(0)) / sizeof(int);
2632
2633 } else if (cmsg->cmsg_level == SOL_SOCKET &&
2634 cmsg->cmsg_type == SCM_CREDENTIALS &&
2635 cmsg->cmsg_len == CMSG_LEN(sizeof(struct ucred))) {
2636
2637 assert(!ucred);
2638 ucred = CMSG_TYPED_DATA(cmsg, struct ucred);
2639 }
2640 }
2641
2642 if (n_fds > 0) {
2643 assert(fd_array);
2644
2645 r = fdset_new_array(&fds, fd_array, n_fds);
2646 if (r < 0) {
2647 close_many(fd_array, n_fds);
2648 log_oom();
2649 return 0;
2650 }
2651 }
2652
2653 if (!ucred || !pid_is_valid(ucred->pid)) {
2654 log_warning("Received notify message without valid credentials. Ignoring.");
2655 return 0;
2656 }
2657
2658 if ((size_t) n >= sizeof(buf) || (msghdr.msg_flags & MSG_TRUNC)) {
2659 log_warning("Received notify message exceeded maximum size. Ignoring.");
2660 return 0;
2661 }
2662
2663 /* As extra safety check, let's make sure the string we get doesn't contain embedded NUL bytes.
2664 * We permit one trailing NUL byte in the message, but don't expect it. */
2665 if (n > 1 && memchr(buf, 0, n-1)) {
2666 log_warning("Received notify message with embedded NUL bytes. Ignoring.");
2667 return 0;
2668 }
2669
2670 /* Make sure it's NUL-terminated, then parse it to obtain the tags list. */
2671 buf[n] = 0;
2672 tags = strv_split_newlines(buf);
2673 if (!tags) {
2674 log_oom();
2675 return 0;
2676 }
2677
2678 /* Possibly a barrier fd, let's see. */
2679 if (manager_process_barrier_fd(tags, fds)) {
2680 log_debug("Received barrier notification message from PID " PID_FMT ".", ucred->pid);
2681 return 0;
2682 }
2683
2684 /* Increase the generation counter used for filtering out duplicate unit invocations. */
2685 m->notifygen++;
2686
2687 /* Notify every unit that might be interested, which might be multiple. */
2688 u1 = manager_get_unit_by_pid_cgroup(m, ucred->pid);
2689 u2 = hashmap_get(m->watch_pids, PID_TO_PTR(ucred->pid));
2690 array = hashmap_get(m->watch_pids, PID_TO_PTR(-ucred->pid));
2691 if (array) {
2692 size_t k = 0;
2693
2694 while (array[k])
2695 k++;
2696
2697 array_copy = newdup(Unit*, array, k+1);
2698 if (!array_copy)
2699 log_oom();
2700 }
2701 /* And now invoke the per-unit callbacks. Note that manager_invoke_notify_message() will handle
2702 * duplicate units make sure we only invoke each unit's handler once. */
2703 if (u1) {
2704 manager_invoke_notify_message(m, u1, ucred, tags, fds);
2705 found = true;
2706 }
2707 if (u2) {
2708 manager_invoke_notify_message(m, u2, ucred, tags, fds);
2709 found = true;
2710 }
2711 if (array_copy)
2712 for (size_t i = 0; array_copy[i]; i++) {
2713 manager_invoke_notify_message(m, array_copy[i], ucred, tags, fds);
2714 found = true;
2715 }
2716
2717 if (!found)
2718 log_warning("Cannot find unit for notify message of PID "PID_FMT", ignoring.", ucred->pid);
2719
2720 if (fdset_size(fds) > 0)
2721 log_warning("Got extra auxiliary fds with notification message, closing them.");
2722
2723 return 0;
2724 }
2725
2726 static void manager_invoke_sigchld_event(
2727 Manager *m,
2728 Unit *u,
2729 const siginfo_t *si) {
2730
2731 assert(m);
2732 assert(u);
2733 assert(si);
2734
2735 /* Already invoked the handler of this unit in this iteration? Then don't process this again */
2736 if (u->sigchldgen == m->sigchldgen)
2737 return;
2738 u->sigchldgen = m->sigchldgen;
2739
2740 log_unit_debug(u, "Child "PID_FMT" belongs to %s.", si->si_pid, u->id);
2741 unit_unwatch_pid(u, si->si_pid);
2742
2743 if (UNIT_VTABLE(u)->sigchld_event)
2744 UNIT_VTABLE(u)->sigchld_event(u, si->si_pid, si->si_code, si->si_status);
2745 }
2746
2747 static int manager_dispatch_sigchld(sd_event_source *source, void *userdata) {
2748 Manager *m = ASSERT_PTR(userdata);
2749 siginfo_t si = {};
2750 int r;
2751
2752 assert(source);
2753
2754 /* First we call waitid() for a PID and do not reap the zombie. That way we can still access
2755 * /proc/$PID for it while it is a zombie. */
2756
2757 if (waitid(P_ALL, 0, &si, WEXITED|WNOHANG|WNOWAIT) < 0) {
2758
2759 if (errno != ECHILD)
2760 log_error_errno(errno, "Failed to peek for child with waitid(), ignoring: %m");
2761
2762 goto turn_off;
2763 }
2764
2765 if (si.si_pid <= 0)
2766 goto turn_off;
2767
2768 if (IN_SET(si.si_code, CLD_EXITED, CLD_KILLED, CLD_DUMPED)) {
2769 _cleanup_free_ Unit **array_copy = NULL;
2770 _cleanup_free_ char *name = NULL;
2771 Unit *u1, *u2, **array;
2772
2773 (void) get_process_comm(si.si_pid, &name);
2774
2775 log_debug("Child "PID_FMT" (%s) died (code=%s, status=%i/%s)",
2776 si.si_pid, strna(name),
2777 sigchld_code_to_string(si.si_code),
2778 si.si_status,
2779 strna(si.si_code == CLD_EXITED
2780 ? exit_status_to_string(si.si_status, EXIT_STATUS_FULL)
2781 : signal_to_string(si.si_status)));
2782
2783 /* Increase the generation counter used for filtering out duplicate unit invocations */
2784 m->sigchldgen++;
2785
2786 /* And now figure out the unit this belongs to, it might be multiple... */
2787 u1 = manager_get_unit_by_pid_cgroup(m, si.si_pid);
2788 u2 = hashmap_get(m->watch_pids, PID_TO_PTR(si.si_pid));
2789 array = hashmap_get(m->watch_pids, PID_TO_PTR(-si.si_pid));
2790 if (array) {
2791 size_t n = 0;
2792
2793 /* Count how many entries the array has */
2794 while (array[n])
2795 n++;
2796
2797 /* Make a copy of the array so that we don't trip up on the array changing beneath us */
2798 array_copy = newdup(Unit*, array, n+1);
2799 if (!array_copy)
2800 log_oom();
2801 }
2802
2803 /* Finally, execute them all. Note that u1, u2 and the array might contain duplicates, but
2804 * that's fine, manager_invoke_sigchld_event() will ensure we only invoke the handlers once for
2805 * each iteration. */
2806 if (u1) {
2807 /* We check for oom condition, in case we got SIGCHLD before the oom notification.
2808 * We only do this for the cgroup the PID belonged to. */
2809 (void) unit_check_oom(u1);
2810
2811 /* We check if systemd-oomd performed a kill so that we log and notify appropriately */
2812 (void) unit_check_oomd_kill(u1);
2813
2814 manager_invoke_sigchld_event(m, u1, &si);
2815 }
2816 if (u2)
2817 manager_invoke_sigchld_event(m, u2, &si);
2818 if (array_copy)
2819 for (size_t i = 0; array_copy[i]; i++)
2820 manager_invoke_sigchld_event(m, array_copy[i], &si);
2821 }
2822
2823 /* And now, we actually reap the zombie. */
2824 if (waitid(P_PID, si.si_pid, &si, WEXITED) < 0) {
2825 log_error_errno(errno, "Failed to dequeue child, ignoring: %m");
2826 return 0;
2827 }
2828
2829 return 0;
2830
2831 turn_off:
2832 /* All children processed for now, turn off event source */
2833
2834 r = sd_event_source_set_enabled(m->sigchld_event_source, SD_EVENT_OFF);
2835 if (r < 0)
2836 return log_error_errno(r, "Failed to disable SIGCHLD event source: %m");
2837
2838 return 0;
2839 }
2840
2841 static void manager_start_special(Manager *m, const char *name, JobMode mode) {
2842 Job *job;
2843
2844 if (manager_add_job_by_name_and_warn(m, JOB_START, name, mode, NULL, &job) < 0)
2845 return;
2846
2847 const char *s = unit_status_string(job->unit, NULL);
2848
2849 log_info("Activating special unit %s...", s);
2850
2851 sd_notifyf(false,
2852 "STATUS=Activating special unit %s...", s);
2853 m->status_ready = false;
2854 }
2855
2856 static void manager_handle_ctrl_alt_del(Manager *m) {
2857 /* If the user presses C-A-D more than
2858 * 7 times within 2s, we reboot/shutdown immediately,
2859 * unless it was disabled in system.conf */
2860
2861 if (ratelimit_below(&m->ctrl_alt_del_ratelimit) || m->cad_burst_action == EMERGENCY_ACTION_NONE)
2862 manager_start_special(m, SPECIAL_CTRL_ALT_DEL_TARGET, JOB_REPLACE_IRREVERSIBLY);
2863 else
2864 emergency_action(m, m->cad_burst_action, EMERGENCY_ACTION_WARN, NULL, -1,
2865 "Ctrl-Alt-Del was pressed more than 7 times within 2s");
2866 }
2867
2868 static int manager_dispatch_signal_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
2869 Manager *m = ASSERT_PTR(userdata);
2870 ssize_t n;
2871 struct signalfd_siginfo sfsi;
2872 int r;
2873
2874 assert(m->signal_fd == fd);
2875
2876 if (revents != EPOLLIN) {
2877 log_warning("Got unexpected events from signal file descriptor.");
2878 return 0;
2879 }
2880
2881 n = read(m->signal_fd, &sfsi, sizeof(sfsi));
2882 if (n < 0) {
2883 if (ERRNO_IS_TRANSIENT(errno))
2884 return 0;
2885
2886 /* We return an error here, which will kill this handler,
2887 * to avoid a busy loop on read error. */
2888 return log_error_errno(errno, "Reading from signal fd failed: %m");
2889 }
2890 if (n != sizeof(sfsi)) {
2891 log_warning("Truncated read from signal fd (%zi bytes), ignoring!", n);
2892 return 0;
2893 }
2894
2895 log_received_signal(sfsi.ssi_signo == SIGCHLD ||
2896 (sfsi.ssi_signo == SIGTERM && MANAGER_IS_USER(m))
2897 ? LOG_DEBUG : LOG_INFO,
2898 &sfsi);
2899
2900 switch (sfsi.ssi_signo) {
2901
2902 case SIGCHLD:
2903 r = sd_event_source_set_enabled(m->sigchld_event_source, SD_EVENT_ON);
2904 if (r < 0)
2905 log_warning_errno(r, "Failed to enable SIGCHLD event source, ignoring: %m");
2906
2907 break;
2908
2909 case SIGTERM:
2910 if (MANAGER_IS_SYSTEM(m)) {
2911 /* This is for compatibility with the original sysvinit */
2912 if (verify_run_space_and_log("Refusing to reexecute") < 0)
2913 break;
2914
2915 m->objective = MANAGER_REEXECUTE;
2916 break;
2917 }
2918
2919 _fallthrough_;
2920 case SIGINT:
2921 if (MANAGER_IS_SYSTEM(m))
2922 manager_handle_ctrl_alt_del(m);
2923 else
2924 manager_start_special(m, SPECIAL_EXIT_TARGET, JOB_REPLACE_IRREVERSIBLY);
2925 break;
2926
2927 case SIGWINCH:
2928 /* This is a nop on non-init */
2929 if (MANAGER_IS_SYSTEM(m))
2930 manager_start_special(m, SPECIAL_KBREQUEST_TARGET, JOB_REPLACE);
2931
2932 break;
2933
2934 case SIGPWR:
2935 /* This is a nop on non-init */
2936 if (MANAGER_IS_SYSTEM(m))
2937 manager_start_special(m, SPECIAL_SIGPWR_TARGET, JOB_REPLACE);
2938
2939 break;
2940
2941 case SIGUSR1:
2942 if (manager_dbus_is_running(m, false)) {
2943 log_info("Trying to reconnect to bus...");
2944
2945 (void) bus_init_api(m);
2946
2947 if (MANAGER_IS_SYSTEM(m))
2948 (void) bus_init_system(m);
2949 } else
2950 manager_start_special(m, SPECIAL_DBUS_SERVICE, JOB_REPLACE);
2951
2952 break;
2953
2954 case SIGUSR2: {
2955 _cleanup_free_ char *dump = NULL;
2956
2957 r = manager_get_dump_string(m, /* patterns= */ NULL, &dump);
2958 if (r < 0) {
2959 log_warning_errno(errno, "Failed to acquire manager dump: %m");
2960 break;
2961 }
2962
2963 log_dump(LOG_INFO, dump);
2964 break;
2965 }
2966
2967 case SIGHUP:
2968 if (verify_run_space_and_log("Refusing to reload") < 0)
2969 break;
2970
2971 m->objective = MANAGER_RELOAD;
2972 break;
2973
2974 default: {
2975
2976 /* Starting SIGRTMIN+0 */
2977 static const struct {
2978 const char *target;
2979 JobMode mode;
2980 } target_table[] = {
2981 [0] = { SPECIAL_DEFAULT_TARGET, JOB_ISOLATE },
2982 [1] = { SPECIAL_RESCUE_TARGET, JOB_ISOLATE },
2983 [2] = { SPECIAL_EMERGENCY_TARGET, JOB_ISOLATE },
2984 [3] = { SPECIAL_HALT_TARGET, JOB_REPLACE_IRREVERSIBLY },
2985 [4] = { SPECIAL_POWEROFF_TARGET, JOB_REPLACE_IRREVERSIBLY },
2986 [5] = { SPECIAL_REBOOT_TARGET, JOB_REPLACE_IRREVERSIBLY },
2987 [6] = { SPECIAL_KEXEC_TARGET, JOB_REPLACE_IRREVERSIBLY },
2988 [7] = { SPECIAL_SOFT_REBOOT_TARGET, JOB_REPLACE_IRREVERSIBLY },
2989 };
2990
2991 /* Starting SIGRTMIN+13, so that target halt and system halt are 10 apart */
2992 static const ManagerObjective objective_table[] = {
2993 [0] = MANAGER_HALT,
2994 [1] = MANAGER_POWEROFF,
2995 [2] = MANAGER_REBOOT,
2996 [3] = MANAGER_KEXEC,
2997 [4] = MANAGER_SOFT_REBOOT,
2998 };
2999
3000 if ((int) sfsi.ssi_signo >= SIGRTMIN+0 &&
3001 (int) sfsi.ssi_signo < SIGRTMIN+(int) ELEMENTSOF(target_table)) {
3002 int idx = (int) sfsi.ssi_signo - SIGRTMIN;
3003 manager_start_special(m, target_table[idx].target, target_table[idx].mode);
3004 break;
3005 }
3006
3007 if ((int) sfsi.ssi_signo >= SIGRTMIN+13 &&
3008 (int) sfsi.ssi_signo < SIGRTMIN+13+(int) ELEMENTSOF(objective_table)) {
3009 m->objective = objective_table[sfsi.ssi_signo - SIGRTMIN - 13];
3010 break;
3011 }
3012
3013 switch (sfsi.ssi_signo - SIGRTMIN) {
3014
3015 case 18: {
3016 bool generic = false;
3017
3018 if (sfsi.ssi_code != SI_QUEUE)
3019 generic = true;
3020 else {
3021 /* Override a few select commands by our own PID1-specific logic */
3022
3023 switch (sfsi.ssi_int) {
3024
3025 case _COMMON_SIGNAL_COMMAND_LOG_LEVEL_BASE..._COMMON_SIGNAL_COMMAND_LOG_LEVEL_END:
3026 manager_override_log_level(m, sfsi.ssi_int - _COMMON_SIGNAL_COMMAND_LOG_LEVEL_BASE);
3027 break;
3028
3029 case COMMON_SIGNAL_COMMAND_CONSOLE:
3030 manager_override_log_target(m, LOG_TARGET_CONSOLE);
3031 break;
3032
3033 case COMMON_SIGNAL_COMMAND_JOURNAL:
3034 manager_override_log_target(m, LOG_TARGET_JOURNAL);
3035 break;
3036
3037 case COMMON_SIGNAL_COMMAND_KMSG:
3038 manager_override_log_target(m, LOG_TARGET_KMSG);
3039 break;
3040
3041 case COMMON_SIGNAL_COMMAND_NULL:
3042 manager_override_log_target(m, LOG_TARGET_NULL);
3043 break;
3044
3045 default:
3046 generic = true;
3047 }
3048 }
3049
3050 if (generic)
3051 return sigrtmin18_handler(source, &sfsi, NULL);
3052
3053 break;
3054 }
3055
3056 case 20:
3057 manager_override_show_status(m, SHOW_STATUS_YES, "signal");
3058 break;
3059
3060 case 21:
3061 manager_override_show_status(m, SHOW_STATUS_NO, "signal");
3062 break;
3063
3064 case 22:
3065 manager_override_log_level(m, LOG_DEBUG);
3066 break;
3067
3068 case 23:
3069 manager_restore_original_log_level(m);
3070 break;
3071
3072 case 24:
3073 if (MANAGER_IS_USER(m)) {
3074 m->objective = MANAGER_EXIT;
3075 return 0;
3076 }
3077
3078 /* This is a nop on init */
3079 break;
3080
3081 case 25:
3082 m->objective = MANAGER_REEXECUTE;
3083 break;
3084
3085 case 26:
3086 case 29: /* compatibility: used to be mapped to LOG_TARGET_SYSLOG_OR_KMSG */
3087 manager_restore_original_log_target(m);
3088 break;
3089
3090 case 27:
3091 manager_override_log_target(m, LOG_TARGET_CONSOLE);
3092 break;
3093
3094 case 28:
3095 manager_override_log_target(m, LOG_TARGET_KMSG);
3096 break;
3097
3098 default:
3099 log_warning("Got unhandled signal <%s>.", signal_to_string(sfsi.ssi_signo));
3100 }
3101 }}
3102
3103 return 0;
3104 }
3105
3106 static int manager_dispatch_time_change_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
3107 Manager *m = ASSERT_PTR(userdata);
3108 Unit *u;
3109
3110 log_struct(LOG_DEBUG,
3111 "MESSAGE_ID=" SD_MESSAGE_TIME_CHANGE_STR,
3112 LOG_MESSAGE("Time has been changed"));
3113
3114 /* Restart the watch */
3115 (void) manager_setup_time_change(m);
3116
3117 HASHMAP_FOREACH(u, m->units)
3118 if (UNIT_VTABLE(u)->time_change)
3119 UNIT_VTABLE(u)->time_change(u);
3120
3121 return 0;
3122 }
3123
3124 static int manager_dispatch_timezone_change(
3125 sd_event_source *source,
3126 const struct inotify_event *e,
3127 void *userdata) {
3128
3129 Manager *m = ASSERT_PTR(userdata);
3130 int changed;
3131 Unit *u;
3132
3133 log_debug("inotify event for /etc/localtime");
3134
3135 changed = manager_read_timezone_stat(m);
3136 if (changed <= 0)
3137 return changed;
3138
3139 /* Something changed, restart the watch, to ensure we watch the new /etc/localtime if it changed */
3140 (void) manager_setup_timezone_change(m);
3141
3142 /* Read the new timezone */
3143 tzset();
3144
3145 log_debug("Timezone has been changed (now: %s).", tzname[daylight]);
3146
3147 HASHMAP_FOREACH(u, m->units)
3148 if (UNIT_VTABLE(u)->timezone_change)
3149 UNIT_VTABLE(u)->timezone_change(u);
3150
3151 return 0;
3152 }
3153
3154 static int manager_dispatch_idle_pipe_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
3155 Manager *m = ASSERT_PTR(userdata);
3156
3157 assert(m->idle_pipe[2] == fd);
3158
3159 /* There's at least one Type=idle child that just gave up on us waiting for the boot process to
3160 * complete. Let's now turn off any further console output if there's at least one service that needs
3161 * console access, so that from now on our own output should not spill into that service's output
3162 * anymore. After all, we support Type=idle only to beautify console output and it generally is set
3163 * on services that want to own the console exclusively without our interference. */
3164 m->no_console_output = m->n_on_console > 0;
3165
3166 /* Acknowledge the child's request, and let all other children know too that they shouldn't wait
3167 * any longer by closing the pipes towards them, which is what they are waiting for. */
3168 manager_close_idle_pipe(m);
3169
3170 return 0;
3171 }
3172
3173 static int manager_dispatch_jobs_in_progress(sd_event_source *source, usec_t usec, void *userdata) {
3174 Manager *m = ASSERT_PTR(userdata);
3175 int r;
3176
3177 assert(source);
3178
3179 manager_print_jobs_in_progress(m);
3180
3181 r = sd_event_source_set_time_relative(source, JOBS_IN_PROGRESS_PERIOD_USEC);
3182 if (r < 0)
3183 return r;
3184
3185 return sd_event_source_set_enabled(source, SD_EVENT_ONESHOT);
3186 }
3187
3188 int manager_loop(Manager *m) {
3189 RateLimit rl = { .interval = 1*USEC_PER_SEC, .burst = 50000 };
3190 int r;
3191
3192 assert(m);
3193 assert(m->objective == MANAGER_OK); /* Ensure manager_startup() has been called */
3194
3195 manager_check_finished(m);
3196
3197 /* There might still be some zombies hanging around from before we were exec()'ed. Let's reap them. */
3198 r = sd_event_source_set_enabled(m->sigchld_event_source, SD_EVENT_ON);
3199 if (r < 0)
3200 return log_error_errno(r, "Failed to enable SIGCHLD event source: %m");
3201
3202 while (m->objective == MANAGER_OK) {
3203
3204 (void) watchdog_ping();
3205
3206 if (!ratelimit_below(&rl)) {
3207 /* Yay, something is going seriously wrong, pause a little */
3208 log_warning("Looping too fast. Throttling execution a little.");
3209 sleep(1);
3210 }
3211
3212 if (manager_dispatch_load_queue(m) > 0)
3213 continue;
3214
3215 if (manager_dispatch_gc_job_queue(m) > 0)
3216 continue;
3217
3218 if (manager_dispatch_gc_unit_queue(m) > 0)
3219 continue;
3220
3221 if (manager_dispatch_cleanup_queue(m) > 0)
3222 continue;
3223
3224 if (manager_dispatch_cgroup_realize_queue(m) > 0)
3225 continue;
3226
3227 if (manager_dispatch_start_when_upheld_queue(m) > 0)
3228 continue;
3229
3230 if (manager_dispatch_stop_when_bound_queue(m) > 0)
3231 continue;
3232
3233 if (manager_dispatch_stop_when_unneeded_queue(m) > 0)
3234 continue;
3235
3236 if (manager_dispatch_release_resources_queue(m) > 0)
3237 continue;
3238
3239 if (manager_dispatch_dbus_queue(m) > 0)
3240 continue;
3241
3242 /* Sleep for watchdog runtime wait time */
3243 r = sd_event_run(m->event, watchdog_runtime_wait());
3244 if (r < 0)
3245 return log_error_errno(r, "Failed to run event loop: %m");
3246 }
3247
3248 return m->objective;
3249 }
3250
3251 int manager_load_unit_from_dbus_path(Manager *m, const char *s, sd_bus_error *e, Unit **_u) {
3252 _cleanup_free_ char *n = NULL;
3253 sd_id128_t invocation_id;
3254 Unit *u;
3255 int r;
3256
3257 assert(m);
3258 assert(s);
3259 assert(_u);
3260
3261 r = unit_name_from_dbus_path(s, &n);
3262 if (r < 0)
3263 return r;
3264
3265 /* Permit addressing units by invocation ID: if the passed bus path is suffixed by a 128bit ID then
3266 * we use it as invocation ID. */
3267 r = sd_id128_from_string(n, &invocation_id);
3268 if (r >= 0) {
3269 u = hashmap_get(m->units_by_invocation_id, &invocation_id);
3270 if (u) {
3271 *_u = u;
3272 return 0;
3273 }
3274
3275 return sd_bus_error_setf(e, BUS_ERROR_NO_UNIT_FOR_INVOCATION_ID,
3276 "No unit with the specified invocation ID " SD_ID128_FORMAT_STR " known.",
3277 SD_ID128_FORMAT_VAL(invocation_id));
3278 }
3279
3280 /* If this didn't work, we check if this is a unit name */
3281 if (!unit_name_is_valid(n, UNIT_NAME_PLAIN|UNIT_NAME_INSTANCE)) {
3282 _cleanup_free_ char *nn = NULL;
3283
3284 nn = cescape(n);
3285 return sd_bus_error_setf(e, SD_BUS_ERROR_INVALID_ARGS,
3286 "Unit name %s is neither a valid invocation ID nor unit name.", strnull(nn));
3287 }
3288
3289 r = manager_load_unit(m, n, NULL, e, &u);
3290 if (r < 0)
3291 return r;
3292
3293 *_u = u;
3294 return 0;
3295 }
3296
3297 int manager_get_job_from_dbus_path(Manager *m, const char *s, Job **_j) {
3298 const char *p;
3299 unsigned id;
3300 Job *j;
3301 int r;
3302
3303 assert(m);
3304 assert(s);
3305 assert(_j);
3306
3307 p = startswith(s, "/org/freedesktop/systemd1/job/");
3308 if (!p)
3309 return -EINVAL;
3310
3311 r = safe_atou(p, &id);
3312 if (r < 0)
3313 return r;
3314
3315 j = manager_get_job(m, id);
3316 if (!j)
3317 return -ENOENT;
3318
3319 *_j = j;
3320
3321 return 0;
3322 }
3323
3324 void manager_send_unit_audit(Manager *m, Unit *u, int type, bool success) {
3325
3326 #if HAVE_AUDIT
3327 _cleanup_free_ char *p = NULL;
3328 const char *msg;
3329 int audit_fd, r;
3330
3331 if (!MANAGER_IS_SYSTEM(m))
3332 return;
3333
3334 audit_fd = get_audit_fd();
3335 if (audit_fd < 0)
3336 return;
3337
3338 /* Don't generate audit events if the service was already
3339 * started and we're just deserializing */
3340 if (MANAGER_IS_RELOADING(m))
3341 return;
3342
3343 if (u->type != UNIT_SERVICE)
3344 return;
3345
3346 r = unit_name_to_prefix_and_instance(u->id, &p);
3347 if (r < 0) {
3348 log_error_errno(r, "Failed to extract prefix and instance of unit name: %m");
3349 return;
3350 }
3351
3352 msg = strjoina("unit=", p);
3353 if (audit_log_user_comm_message(audit_fd, type, msg, "systemd", NULL, NULL, NULL, success) < 0) {
3354 if (errno == EPERM)
3355 /* We aren't allowed to send audit messages?
3356 * Then let's not retry again. */
3357 close_audit_fd();
3358 else
3359 log_warning_errno(errno, "Failed to send audit message: %m");
3360 }
3361 #endif
3362
3363 }
3364
3365 void manager_send_unit_plymouth(Manager *m, Unit *u) {
3366 static const union sockaddr_union sa = PLYMOUTH_SOCKET;
3367 _cleanup_free_ char *message = NULL;
3368 _cleanup_close_ int fd = -EBADF;
3369 int n = 0;
3370
3371 /* Don't generate plymouth events if the service was already
3372 * started and we're just deserializing */
3373 if (MANAGER_IS_RELOADING(m))
3374 return;
3375
3376 if (!MANAGER_IS_SYSTEM(m))
3377 return;
3378
3379 if (detect_container() > 0)
3380 return;
3381
3382 if (!IN_SET(u->type, UNIT_SERVICE, UNIT_MOUNT, UNIT_SWAP))
3383 return;
3384
3385 /* We set SOCK_NONBLOCK here so that we rather drop the
3386 * message then wait for plymouth */
3387 fd = socket(AF_UNIX, SOCK_STREAM|SOCK_CLOEXEC|SOCK_NONBLOCK, 0);
3388 if (fd < 0) {
3389 log_error_errno(errno, "socket() failed: %m");
3390 return;
3391 }
3392
3393 if (connect(fd, &sa.sa, SOCKADDR_UN_LEN(sa.un)) < 0) {
3394 if (!IN_SET(errno, EAGAIN, ENOENT) && !ERRNO_IS_DISCONNECT(errno))
3395 log_error_errno(errno, "connect() failed: %m");
3396 return;
3397 }
3398
3399 if (asprintf(&message, "U\002%c%s%n", (int) (strlen(u->id) + 1), u->id, &n) < 0)
3400 return (void) log_oom();
3401
3402 errno = 0;
3403 if (write(fd, message, n + 1) != n + 1)
3404 if (!IN_SET(errno, EAGAIN, ENOENT) && !ERRNO_IS_DISCONNECT(errno))
3405 log_error_errno(errno, "Failed to write Plymouth message: %m");
3406 }
3407
3408 usec_t manager_get_watchdog(Manager *m, WatchdogType t) {
3409 assert(m);
3410
3411 if (MANAGER_IS_USER(m))
3412 return USEC_INFINITY;
3413
3414 if (timestamp_is_set(m->watchdog_overridden[t]))
3415 return m->watchdog_overridden[t];
3416
3417 return m->watchdog[t];
3418 }
3419
3420 void manager_set_watchdog(Manager *m, WatchdogType t, usec_t timeout) {
3421
3422 assert(m);
3423
3424 if (MANAGER_IS_USER(m))
3425 return;
3426
3427 if (m->watchdog[t] == timeout)
3428 return;
3429
3430 if (t == WATCHDOG_RUNTIME) {
3431 if (!timestamp_is_set(m->watchdog_overridden[WATCHDOG_RUNTIME]))
3432 (void) watchdog_setup(timeout);
3433 } else if (t == WATCHDOG_PRETIMEOUT)
3434 if (m->watchdog_overridden[WATCHDOG_PRETIMEOUT] == USEC_INFINITY)
3435 (void) watchdog_setup_pretimeout(timeout);
3436
3437 m->watchdog[t] = timeout;
3438 }
3439
3440 void manager_override_watchdog(Manager *m, WatchdogType t, usec_t timeout) {
3441
3442 assert(m);
3443
3444 if (MANAGER_IS_USER(m))
3445 return;
3446
3447 if (m->watchdog_overridden[t] == timeout)
3448 return;
3449
3450 if (t == WATCHDOG_RUNTIME) {
3451 usec_t usec = timestamp_is_set(timeout) ? timeout : m->watchdog[t];
3452
3453 (void) watchdog_setup(usec);
3454 } else if (t == WATCHDOG_PRETIMEOUT)
3455 (void) watchdog_setup_pretimeout(timeout);
3456
3457 m->watchdog_overridden[t] = timeout;
3458 }
3459
3460 int manager_set_watchdog_pretimeout_governor(Manager *m, const char *governor) {
3461 _cleanup_free_ char *p = NULL;
3462 int r;
3463
3464 assert(m);
3465
3466 if (MANAGER_IS_USER(m))
3467 return 0;
3468
3469 if (streq_ptr(m->watchdog_pretimeout_governor, governor))
3470 return 0;
3471
3472 p = strdup(governor);
3473 if (!p)
3474 return -ENOMEM;
3475
3476 r = watchdog_setup_pretimeout_governor(governor);
3477 if (r < 0)
3478 return r;
3479
3480 return free_and_replace(m->watchdog_pretimeout_governor, p);
3481 }
3482
3483 int manager_override_watchdog_pretimeout_governor(Manager *m, const char *governor) {
3484 _cleanup_free_ char *p = NULL;
3485 int r;
3486
3487 assert(m);
3488
3489 if (MANAGER_IS_USER(m))
3490 return 0;
3491
3492 if (streq_ptr(m->watchdog_pretimeout_governor_overridden, governor))
3493 return 0;
3494
3495 p = strdup(governor);
3496 if (!p)
3497 return -ENOMEM;
3498
3499 r = watchdog_setup_pretimeout_governor(governor);
3500 if (r < 0)
3501 return r;
3502
3503 return free_and_replace(m->watchdog_pretimeout_governor_overridden, p);
3504 }
3505
3506 int manager_reload(Manager *m) {
3507 _unused_ _cleanup_(manager_reloading_stopp) Manager *reloading = NULL;
3508 _cleanup_fdset_free_ FDSet *fds = NULL;
3509 _cleanup_fclose_ FILE *f = NULL;
3510 int r;
3511
3512 assert(m);
3513
3514 r = manager_open_serialization(m, &f);
3515 if (r < 0)
3516 return log_error_errno(r, "Failed to create serialization file: %m");
3517
3518 fds = fdset_new();
3519 if (!fds)
3520 return log_oom();
3521
3522 /* We are officially in reload mode from here on. */
3523 reloading = manager_reloading_start(m);
3524
3525 r = manager_serialize(m, f, fds, false);
3526 if (r < 0)
3527 return r;
3528
3529 if (fseeko(f, 0, SEEK_SET) < 0)
3530 return log_error_errno(errno, "Failed to seek to beginning of serialization: %m");
3531
3532 /* 💀 This is the point of no return, from here on there is no way back. 💀 */
3533 reloading = NULL;
3534
3535 bus_manager_send_reloading(m, true);
3536
3537 /* Start by flushing out all jobs and units, all generated units, all runtime environments, all dynamic users
3538 * and everything else that is worth flushing out. We'll get it all back from the serialization — if we need
3539 * it. */
3540
3541 manager_clear_jobs_and_units(m);
3542 lookup_paths_flush_generator(&m->lookup_paths);
3543 lookup_paths_free(&m->lookup_paths);
3544 exec_shared_runtime_vacuum(m);
3545 dynamic_user_vacuum(m, false);
3546 m->uid_refs = hashmap_free(m->uid_refs);
3547 m->gid_refs = hashmap_free(m->gid_refs);
3548
3549 r = lookup_paths_init_or_warn(&m->lookup_paths, m->runtime_scope, 0, NULL);
3550 if (r < 0)
3551 return r;
3552
3553 (void) manager_run_environment_generators(m);
3554 (void) manager_run_generators(m);
3555
3556 lookup_paths_log(&m->lookup_paths);
3557
3558 /* We flushed out generated files, for which we don't watch mtime, so we should flush the old map. */
3559 manager_free_unit_name_maps(m);
3560
3561 /* First, enumerate what we can from kernel and suchlike */
3562 manager_enumerate_perpetual(m);
3563 manager_enumerate(m);
3564
3565 /* Second, deserialize our stored data */
3566 r = manager_deserialize(m, f, fds);
3567 if (r < 0)
3568 log_warning_errno(r, "Deserialization failed, proceeding anyway: %m");
3569
3570 /* We don't need the serialization anymore */
3571 f = safe_fclose(f);
3572
3573 /* Re-register notify_fd as event source, and set up other sockets/communication channels we might need */
3574 (void) manager_setup_notify(m);
3575 (void) manager_setup_cgroups_agent(m);
3576 (void) manager_setup_user_lookup_fd(m);
3577
3578 /* Third, fire things up! */
3579 manager_coldplug(m);
3580
3581 /* Clean up runtime objects no longer referenced */
3582 manager_vacuum(m);
3583
3584 /* Clean up deserialized tracked clients */
3585 m->deserialized_subscribed = strv_free(m->deserialized_subscribed);
3586
3587 /* Consider the reload process complete now. */
3588 assert(m->n_reloading > 0);
3589 m->n_reloading--;
3590
3591 manager_ready(m);
3592
3593 m->send_reloading_done = true;
3594 return 0;
3595 }
3596
3597 void manager_reset_failed(Manager *m) {
3598 Unit *u;
3599
3600 assert(m);
3601
3602 HASHMAP_FOREACH(u, m->units)
3603 unit_reset_failed(u);
3604 }
3605
3606 bool manager_unit_inactive_or_pending(Manager *m, const char *name) {
3607 Unit *u;
3608
3609 assert(m);
3610 assert(name);
3611
3612 /* Returns true if the unit is inactive or going down */
3613 u = manager_get_unit(m, name);
3614 if (!u)
3615 return true;
3616
3617 return unit_inactive_or_pending(u);
3618 }
3619
3620 static void log_taint_string(Manager *m) {
3621 _cleanup_free_ char *taint = NULL;
3622
3623 assert(m);
3624
3625 if (MANAGER_IS_USER(m) || m->taint_logged)
3626 return;
3627
3628 m->taint_logged = true; /* only check for taint once */
3629
3630 taint = manager_taint_string(m);
3631 if (isempty(taint))
3632 return;
3633
3634 log_struct(LOG_NOTICE,
3635 LOG_MESSAGE("System is tainted: %s", taint),
3636 "TAINT=%s", taint,
3637 "MESSAGE_ID=" SD_MESSAGE_TAINTED_STR);
3638 }
3639
3640 static void manager_notify_finished(Manager *m) {
3641 usec_t firmware_usec, loader_usec, kernel_usec, initrd_usec, userspace_usec, total_usec;
3642
3643 if (MANAGER_IS_TEST_RUN(m))
3644 return;
3645
3646 if (MANAGER_IS_SYSTEM(m) && detect_container() <= 0) {
3647 char buf[FORMAT_TIMESPAN_MAX + STRLEN(" (firmware) + ") + FORMAT_TIMESPAN_MAX + STRLEN(" (loader) + ")]
3648 = {};
3649 char *p = buf;
3650 size_t size = sizeof buf;
3651
3652 /* Note that MANAGER_TIMESTAMP_KERNEL's monotonic value is always at 0, and
3653 * MANAGER_TIMESTAMP_FIRMWARE's and MANAGER_TIMESTAMP_LOADER's monotonic value should be considered
3654 * negative values. */
3655
3656 firmware_usec = m->timestamps[MANAGER_TIMESTAMP_FIRMWARE].monotonic - m->timestamps[MANAGER_TIMESTAMP_LOADER].monotonic;
3657 loader_usec = m->timestamps[MANAGER_TIMESTAMP_LOADER].monotonic - m->timestamps[MANAGER_TIMESTAMP_KERNEL].monotonic;
3658 userspace_usec = m->timestamps[MANAGER_TIMESTAMP_FINISH].monotonic - m->timestamps[MANAGER_TIMESTAMP_USERSPACE].monotonic;
3659 total_usec = m->timestamps[MANAGER_TIMESTAMP_FIRMWARE].monotonic + m->timestamps[MANAGER_TIMESTAMP_FINISH].monotonic;
3660
3661 if (firmware_usec > 0)
3662 size = strpcpyf(&p, size, "%s (firmware) + ", FORMAT_TIMESPAN(firmware_usec, USEC_PER_MSEC));
3663 if (loader_usec > 0)
3664 size = strpcpyf(&p, size, "%s (loader) + ", FORMAT_TIMESPAN(loader_usec, USEC_PER_MSEC));
3665
3666 if (dual_timestamp_is_set(&m->timestamps[MANAGER_TIMESTAMP_INITRD])) {
3667
3668 /* The initrd case on bare-metal */
3669 kernel_usec = m->timestamps[MANAGER_TIMESTAMP_INITRD].monotonic - m->timestamps[MANAGER_TIMESTAMP_KERNEL].monotonic;
3670 initrd_usec = m->timestamps[MANAGER_TIMESTAMP_USERSPACE].monotonic - m->timestamps[MANAGER_TIMESTAMP_INITRD].monotonic;
3671
3672 log_struct(LOG_INFO,
3673 "MESSAGE_ID=" SD_MESSAGE_STARTUP_FINISHED_STR,
3674 "KERNEL_USEC="USEC_FMT, kernel_usec,
3675 "INITRD_USEC="USEC_FMT, initrd_usec,
3676 "USERSPACE_USEC="USEC_FMT, userspace_usec,
3677 LOG_MESSAGE("Startup finished in %s%s (kernel) + %s (initrd) + %s (userspace) = %s.",
3678 buf,
3679 FORMAT_TIMESPAN(kernel_usec, USEC_PER_MSEC),
3680 FORMAT_TIMESPAN(initrd_usec, USEC_PER_MSEC),
3681 FORMAT_TIMESPAN(userspace_usec, USEC_PER_MSEC),
3682 FORMAT_TIMESPAN(total_usec, USEC_PER_MSEC)));
3683 } else {
3684 /* The initrd-less case on bare-metal */
3685
3686 kernel_usec = m->timestamps[MANAGER_TIMESTAMP_USERSPACE].monotonic - m->timestamps[MANAGER_TIMESTAMP_KERNEL].monotonic;
3687 initrd_usec = 0;
3688
3689 log_struct(LOG_INFO,
3690 "MESSAGE_ID=" SD_MESSAGE_STARTUP_FINISHED_STR,
3691 "KERNEL_USEC="USEC_FMT, kernel_usec,
3692 "USERSPACE_USEC="USEC_FMT, userspace_usec,
3693 LOG_MESSAGE("Startup finished in %s%s (kernel) + %s (userspace) = %s.",
3694 buf,
3695 FORMAT_TIMESPAN(kernel_usec, USEC_PER_MSEC),
3696 FORMAT_TIMESPAN(userspace_usec, USEC_PER_MSEC),
3697 FORMAT_TIMESPAN(total_usec, USEC_PER_MSEC)));
3698 }
3699 } else {
3700 /* The container and --user case */
3701 firmware_usec = loader_usec = initrd_usec = kernel_usec = 0;
3702 total_usec = userspace_usec = m->timestamps[MANAGER_TIMESTAMP_FINISH].monotonic - m->timestamps[MANAGER_TIMESTAMP_USERSPACE].monotonic;
3703
3704 log_struct(LOG_INFO,
3705 "MESSAGE_ID=" SD_MESSAGE_USER_STARTUP_FINISHED_STR,
3706 "USERSPACE_USEC="USEC_FMT, userspace_usec,
3707 LOG_MESSAGE("Startup finished in %s.",
3708 FORMAT_TIMESPAN(total_usec, USEC_PER_MSEC)));
3709 }
3710
3711 bus_manager_send_finished(m, firmware_usec, loader_usec, kernel_usec, initrd_usec, userspace_usec, total_usec);
3712
3713 log_taint_string(m);
3714 }
3715
3716 static void user_manager_send_ready(Manager *m) {
3717 int r;
3718
3719 assert(m);
3720
3721 /* We send READY=1 on reaching basic.target only when running in --user mode. */
3722 if (!MANAGER_IS_USER(m) || m->ready_sent)
3723 return;
3724
3725 r = sd_notify(false,
3726 "READY=1\n"
3727 "STATUS=Reached " SPECIAL_BASIC_TARGET ".");
3728 if (r < 0)
3729 log_warning_errno(r, "Failed to send readiness notification, ignoring: %m");
3730
3731 m->ready_sent = true;
3732 m->status_ready = false;
3733 }
3734
3735 static void manager_send_ready(Manager *m) {
3736 int r;
3737
3738 if (m->ready_sent && m->status_ready)
3739 /* Skip the notification if nothing changed. */
3740 return;
3741
3742 r = sd_notify(false,
3743 "READY=1\n"
3744 "STATUS=Ready.");
3745 if (r < 0)
3746 log_full_errno(m->ready_sent ? LOG_DEBUG : LOG_WARNING, r,
3747 "Failed to send readiness notification, ignoring: %m");
3748
3749 m->ready_sent = m->status_ready = true;
3750 }
3751
3752 static void manager_check_basic_target(Manager *m) {
3753 Unit *u;
3754
3755 assert(m);
3756
3757 /* Small shortcut */
3758 if (m->ready_sent && m->taint_logged)
3759 return;
3760
3761 u = manager_get_unit(m, SPECIAL_BASIC_TARGET);
3762 if (!u || !UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u)))
3763 return;
3764
3765 /* For user managers, send out READY=1 as soon as we reach basic.target */
3766 user_manager_send_ready(m);
3767
3768 /* Log the taint string as soon as we reach basic.target */
3769 log_taint_string(m);
3770 }
3771
3772 void manager_check_finished(Manager *m) {
3773 assert(m);
3774
3775 if (MANAGER_IS_RELOADING(m))
3776 return;
3777
3778 /* Verify that we have entered the event loop already, and not left it again. */
3779 if (!MANAGER_IS_RUNNING(m))
3780 return;
3781
3782 manager_check_basic_target(m);
3783
3784 if (hashmap_size(m->jobs) > 0) {
3785 if (m->jobs_in_progress_event_source)
3786 /* Ignore any failure, this is only for feedback */
3787 (void) sd_event_source_set_time(m->jobs_in_progress_event_source,
3788 manager_watch_jobs_next_time(m));
3789 return;
3790 }
3791
3792 /* The jobs hashmap tends to grow a lot during boot, and then it's not reused until shutdown. Let's
3793 kill the hashmap if it is relatively large. */
3794 if (hashmap_buckets(m->jobs) > hashmap_size(m->units) / 10)
3795 m->jobs = hashmap_free(m->jobs);
3796
3797 manager_send_ready(m);
3798
3799 /* Notify Type=idle units that we are done now */
3800 manager_close_idle_pipe(m);
3801
3802 if (MANAGER_IS_FINISHED(m))
3803 return;
3804
3805 manager_flip_auto_status(m, false, "boot finished");
3806
3807 /* Turn off confirm spawn now */
3808 m->confirm_spawn = NULL;
3809
3810 /* No need to update ask password status when we're going non-interactive */
3811 manager_close_ask_password(m);
3812
3813 /* This is no longer the first boot */
3814 manager_set_first_boot(m, false);
3815
3816 dual_timestamp_get(m->timestamps + MANAGER_TIMESTAMP_FINISH);
3817
3818 manager_notify_finished(m);
3819
3820 manager_invalidate_startup_units(m);
3821 }
3822
3823 void manager_send_reloading(Manager *m) {
3824 assert(m);
3825
3826 /* Let whoever invoked us know that we are now reloading */
3827 (void) sd_notifyf(/* unset= */ false,
3828 "RELOADING=1\n"
3829 "MONOTONIC_USEC=" USEC_FMT "\n", now(CLOCK_MONOTONIC));
3830
3831 /* And ensure that we'll send READY=1 again as soon as we are ready again */
3832 m->ready_sent = false;
3833 }
3834
3835 static bool generator_path_any(const char* const* paths) {
3836 bool found = false;
3837
3838 /* Optimize by skipping the whole process by not creating output directories
3839 * if no generators are found. */
3840 STRV_FOREACH(path, paths)
3841 if (access(*path, F_OK) == 0)
3842 found = true;
3843 else if (errno != ENOENT)
3844 log_warning_errno(errno, "Failed to open generator directory %s: %m", *path);
3845
3846 return found;
3847 }
3848
3849 static int manager_run_environment_generators(Manager *m) {
3850 char **tmp = NULL; /* this is only used in the forked process, no cleanup here */
3851 _cleanup_strv_free_ char **paths = NULL;
3852 void* args[] = {
3853 [STDOUT_GENERATE] = &tmp,
3854 [STDOUT_COLLECT] = &tmp,
3855 [STDOUT_CONSUME] = &m->transient_environment,
3856 };
3857 int r;
3858
3859 if (MANAGER_IS_TEST_RUN(m) && !(m->test_run_flags & MANAGER_TEST_RUN_ENV_GENERATORS))
3860 return 0;
3861
3862 paths = env_generator_binary_paths(m->runtime_scope);
3863 if (!paths)
3864 return log_oom();
3865
3866 if (!generator_path_any((const char* const*) paths))
3867 return 0;
3868
3869 WITH_UMASK(0022)
3870 r = execute_directories((const char* const*) paths, DEFAULT_TIMEOUT_USEC, gather_environment,
3871 args, NULL, m->transient_environment,
3872 EXEC_DIR_PARALLEL | EXEC_DIR_IGNORE_ERRORS | EXEC_DIR_SET_SYSTEMD_EXEC_PID);
3873 return r;
3874 }
3875
3876 static int build_generator_environment(Manager *m, char ***ret) {
3877 _cleanup_strv_free_ char **nl = NULL;
3878 Virtualization v;
3879 int r;
3880
3881 assert(m);
3882 assert(ret);
3883
3884 /* Generators oftentimes want to know some basic facts about the environment they run in, in order to
3885 * adjust generated units to that. Let's pass down some bits of information that are easy for us to
3886 * determine (but a bit harder for generator scripts to determine), as environment variables. */
3887
3888 nl = strv_copy(m->transient_environment);
3889 if (!nl)
3890 return -ENOMEM;
3891
3892 r = strv_env_assign(&nl, "SYSTEMD_SCOPE", runtime_scope_to_string(m->runtime_scope));
3893 if (r < 0)
3894 return r;
3895
3896 if (MANAGER_IS_SYSTEM(m)) {
3897 /* Note that $SYSTEMD_IN_INITRD may be used to override the initrd detection in much of our
3898 * codebase. This is hence more than purely informational. It will shortcut detection of the
3899 * initrd state if generators invoke our own tools. But that's OK, as it would come to the
3900 * same results (hopefully). */
3901 r = strv_env_assign(&nl, "SYSTEMD_IN_INITRD", one_zero(in_initrd()));
3902 if (r < 0)
3903 return r;
3904
3905 if (m->first_boot >= 0) {
3906 r = strv_env_assign(&nl, "SYSTEMD_FIRST_BOOT", one_zero(m->first_boot));
3907 if (r < 0)
3908 return r;
3909 }
3910 }
3911
3912 v = detect_virtualization();
3913 if (v < 0)
3914 log_debug_errno(v, "Failed to detect virtualization, ignoring: %m");
3915 else if (v > 0) {
3916 const char *s;
3917
3918 s = strjoina(VIRTUALIZATION_IS_VM(v) ? "vm:" :
3919 VIRTUALIZATION_IS_CONTAINER(v) ? "container:" : ":",
3920 virtualization_to_string(v));
3921
3922 r = strv_env_assign(&nl, "SYSTEMD_VIRTUALIZATION", s);
3923 if (r < 0)
3924 return r;
3925 }
3926
3927 r = strv_env_assign(&nl, "SYSTEMD_ARCHITECTURE", architecture_to_string(uname_architecture()));
3928 if (r < 0)
3929 return r;
3930
3931 *ret = TAKE_PTR(nl);
3932 return 0;
3933 }
3934
3935 static int manager_execute_generators(Manager *m, char **paths, bool remount_ro) {
3936 _cleanup_strv_free_ char **ge = NULL;
3937 const char *argv[] = {
3938 NULL, /* Leave this empty, execute_directory() will fill something in */
3939 m->lookup_paths.generator,
3940 m->lookup_paths.generator_early,
3941 m->lookup_paths.generator_late,
3942 NULL,
3943 };
3944 int r;
3945
3946 r = build_generator_environment(m, &ge);
3947 if (r < 0)
3948 return log_error_errno(r, "Failed to build generator environment: %m");
3949
3950 if (remount_ro) {
3951 /* Remount most of the filesystem tree read-only. We leave /sys/ as-is, because our code
3952 * checks whether it is read-only to detect containerized execution environments. We leave
3953 * /run/ as-is too, because that's where our output goes. We also leave /proc/ and /dev/shm/
3954 * because they're API, and /tmp/ that safe_fork() mounted for us.
3955 */
3956 r = bind_remount_recursive("/", MS_RDONLY, MS_RDONLY,
3957 STRV_MAKE("/sys", "/run", "/proc", "/dev/shm", "/tmp"));
3958 if (r < 0)
3959 log_warning_errno(r, "Read-only bind remount failed, ignoring: %m");
3960 }
3961
3962 BLOCK_WITH_UMASK(0022);
3963 return execute_directories(
3964 (const char* const*) paths,
3965 DEFAULT_TIMEOUT_USEC,
3966 /* callbacks= */ NULL, /* callback_args= */ NULL,
3967 (char**) argv,
3968 ge,
3969 EXEC_DIR_PARALLEL | EXEC_DIR_IGNORE_ERRORS | EXEC_DIR_SET_SYSTEMD_EXEC_PID);
3970 }
3971
3972 static int manager_run_generators(Manager *m) {
3973 ForkFlags flags = FORK_RESET_SIGNALS | FORK_WAIT | FORK_NEW_MOUNTNS | FORK_MOUNTNS_SLAVE;
3974 _cleanup_strv_free_ char **paths = NULL;
3975 int r;
3976
3977 assert(m);
3978
3979 if (MANAGER_IS_TEST_RUN(m) && !(m->test_run_flags & MANAGER_TEST_RUN_GENERATORS))
3980 return 0;
3981
3982 paths = generator_binary_paths(m->runtime_scope);
3983 if (!paths)
3984 return log_oom();
3985
3986 if (!generator_path_any((const char* const*) paths))
3987 return 0;
3988
3989 r = lookup_paths_mkdir_generator(&m->lookup_paths);
3990 if (r < 0) {
3991 log_error_errno(r, "Failed to create generator directories: %m");
3992 goto finish;
3993 }
3994
3995 /* If we are the system manager, we fork and invoke the generators in a sanitized mount namespace. If
3996 * we are the user manager, let's just execute the generators directly. We might not have the
3997 * necessary privileges, and the system manager has already mounted /tmp/ and everything else for us.
3998 */
3999 if (MANAGER_IS_USER(m)) {
4000 r = manager_execute_generators(m, paths, /* remount_ro= */ false);
4001 goto finish;
4002 }
4003
4004 /* On some systems /tmp/ doesn't exist, and on some other systems we cannot create it at all. Avoid
4005 * trying to mount a private tmpfs on it as there's no one size fits all. */
4006 if (is_dir("/tmp", /* follow= */ false) > 0)
4007 flags |= FORK_PRIVATE_TMP;
4008
4009 r = safe_fork("(sd-gens)", flags, NULL);
4010 if (r == 0) {
4011 r = manager_execute_generators(m, paths, /* remount_ro= */ true);
4012 _exit(r >= 0 ? EXIT_SUCCESS : EXIT_FAILURE);
4013 }
4014 if (r < 0) {
4015 if (!ERRNO_IS_PRIVILEGE(r)) {
4016 log_error_errno(r, "Failed to fork off sandboxing environment for executing generators: %m");
4017 goto finish;
4018 }
4019
4020 /* Failed to fork with new mount namespace? Maybe, running in a container environment with
4021 * seccomp or without capability. */
4022 log_debug_errno(r,
4023 "Failed to fork off sandboxing environment for executing generators. "
4024 "Falling back to execute generators without sandboxing: %m");
4025 r = manager_execute_generators(m, paths, /* remount_ro= */ false);
4026 }
4027
4028 finish:
4029 lookup_paths_trim_generator(&m->lookup_paths);
4030 return r;
4031 }
4032
4033 int manager_transient_environment_add(Manager *m, char **plus) {
4034 char **a;
4035
4036 assert(m);
4037
4038 if (strv_isempty(plus))
4039 return 0;
4040
4041 a = strv_env_merge(m->transient_environment, plus);
4042 if (!a)
4043 return log_oom();
4044
4045 sanitize_environment(a);
4046
4047 return strv_free_and_replace(m->transient_environment, a);
4048 }
4049
4050 int manager_client_environment_modify(
4051 Manager *m,
4052 char **minus,
4053 char **plus) {
4054
4055 char **a = NULL, **b = NULL, **l;
4056
4057 assert(m);
4058
4059 if (strv_isempty(minus) && strv_isempty(plus))
4060 return 0;
4061
4062 l = m->client_environment;
4063
4064 if (!strv_isempty(minus)) {
4065 a = strv_env_delete(l, 1, minus);
4066 if (!a)
4067 return -ENOMEM;
4068
4069 l = a;
4070 }
4071
4072 if (!strv_isempty(plus)) {
4073 b = strv_env_merge(l, plus);
4074 if (!b) {
4075 strv_free(a);
4076 return -ENOMEM;
4077 }
4078
4079 l = b;
4080 }
4081
4082 if (m->client_environment != l)
4083 strv_free(m->client_environment);
4084
4085 if (a != l)
4086 strv_free(a);
4087 if (b != l)
4088 strv_free(b);
4089
4090 m->client_environment = sanitize_environment(l);
4091 return 0;
4092 }
4093
4094 int manager_get_effective_environment(Manager *m, char ***ret) {
4095 char **l;
4096
4097 assert(m);
4098 assert(ret);
4099
4100 l = strv_env_merge(m->transient_environment, m->client_environment);
4101 if (!l)
4102 return -ENOMEM;
4103
4104 *ret = l;
4105 return 0;
4106 }
4107
4108 int manager_set_default_smack_process_label(Manager *m, const char *label) {
4109 assert(m);
4110
4111 #ifdef SMACK_DEFAULT_PROCESS_LABEL
4112 if (!label)
4113 return free_and_strdup(&m->default_smack_process_label, SMACK_DEFAULT_PROCESS_LABEL);
4114 #endif
4115 if (streq_ptr(label, "/"))
4116 return free_and_strdup(&m->default_smack_process_label, NULL);
4117
4118 return free_and_strdup(&m->default_smack_process_label, label);
4119 }
4120
4121 int manager_set_default_rlimits(Manager *m, struct rlimit **default_rlimit) {
4122 assert(m);
4123
4124 for (unsigned i = 0; i < _RLIMIT_MAX; i++) {
4125 m->rlimit[i] = mfree(m->rlimit[i]);
4126
4127 if (!default_rlimit[i])
4128 continue;
4129
4130 m->rlimit[i] = newdup(struct rlimit, default_rlimit[i], 1);
4131 if (!m->rlimit[i])
4132 return log_oom();
4133 }
4134
4135 return 0;
4136 }
4137
4138 void manager_recheck_dbus(Manager *m) {
4139 assert(m);
4140
4141 /* Connects to the bus if the dbus service and socket are running. If we are running in user mode
4142 * this is all it does. In system mode we'll also connect to the system bus (which will most likely
4143 * just reuse the connection of the API bus). That's because the system bus after all runs as service
4144 * of the system instance, while in the user instance we can assume it's already there. */
4145
4146 if (MANAGER_IS_RELOADING(m))
4147 return; /* don't check while we are reloading… */
4148
4149 if (manager_dbus_is_running(m, false)) {
4150 (void) bus_init_api(m);
4151
4152 if (MANAGER_IS_SYSTEM(m))
4153 (void) bus_init_system(m);
4154 } else {
4155 (void) bus_done_api(m);
4156
4157 if (MANAGER_IS_SYSTEM(m))
4158 (void) bus_done_system(m);
4159 }
4160 }
4161
4162 static bool manager_journal_is_running(Manager *m) {
4163 Unit *u;
4164
4165 assert(m);
4166
4167 if (MANAGER_IS_TEST_RUN(m))
4168 return false;
4169
4170 /* If we are the user manager we can safely assume that the journal is up */
4171 if (!MANAGER_IS_SYSTEM(m))
4172 return true;
4173
4174 /* Check that the socket is not only up, but in RUNNING state */
4175 u = manager_get_unit(m, SPECIAL_JOURNALD_SOCKET);
4176 if (!u)
4177 return false;
4178 if (SOCKET(u)->state != SOCKET_RUNNING)
4179 return false;
4180
4181 /* Similar, check if the daemon itself is fully up, too */
4182 u = manager_get_unit(m, SPECIAL_JOURNALD_SERVICE);
4183 if (!u)
4184 return false;
4185 if (!IN_SET(SERVICE(u)->state, SERVICE_RELOAD, SERVICE_RUNNING))
4186 return false;
4187
4188 return true;
4189 }
4190
4191 void disable_printk_ratelimit(void) {
4192 /* Disable kernel's printk ratelimit.
4193 *
4194 * Logging to /dev/kmsg is most useful during early boot and shutdown, where normal logging
4195 * mechanisms are not available. The semantics of this sysctl are such that any kernel command-line
4196 * setting takes precedence. */
4197 int r;
4198
4199 r = sysctl_write("kernel/printk_devkmsg", "on");
4200 if (r < 0)
4201 log_debug_errno(r, "Failed to set sysctl kernel.printk_devkmsg=on: %m");
4202 }
4203
4204 void manager_recheck_journal(Manager *m) {
4205
4206 assert(m);
4207
4208 /* Don't bother with this unless we are in the special situation of being PID 1 */
4209 if (getpid_cached() != 1)
4210 return;
4211
4212 /* Don't check this while we are reloading, things might still change */
4213 if (MANAGER_IS_RELOADING(m))
4214 return;
4215
4216 /* The journal is fully and entirely up? If so, let's permit logging to it, if that's configured. If
4217 * the journal is down, don't ever log to it, otherwise we might end up deadlocking ourselves as we
4218 * might trigger an activation ourselves we can't fulfill. */
4219 log_set_prohibit_ipc(!manager_journal_is_running(m));
4220 log_open();
4221 }
4222
4223 static ShowStatus manager_get_show_status(Manager *m) {
4224 assert(m);
4225
4226 if (MANAGER_IS_USER(m))
4227 return _SHOW_STATUS_INVALID;
4228
4229 if (m->show_status_overridden != _SHOW_STATUS_INVALID)
4230 return m->show_status_overridden;
4231
4232 return m->show_status;
4233 }
4234
4235 bool manager_get_show_status_on(Manager *m) {
4236 assert(m);
4237
4238 return show_status_on(manager_get_show_status(m));
4239 }
4240
4241 static void set_show_status_marker(bool b) {
4242 if (b)
4243 (void) touch("/run/systemd/show-status");
4244 else
4245 (void) unlink("/run/systemd/show-status");
4246 }
4247
4248 void manager_set_show_status(Manager *m, ShowStatus mode, const char *reason) {
4249 assert(m);
4250 assert(reason);
4251 assert(mode >= 0 && mode < _SHOW_STATUS_MAX);
4252
4253 if (MANAGER_IS_USER(m))
4254 return;
4255
4256 if (mode == m->show_status)
4257 return;
4258
4259 if (m->show_status_overridden == _SHOW_STATUS_INVALID) {
4260 bool enabled;
4261
4262 enabled = show_status_on(mode);
4263 log_debug("%s (%s) showing of status (%s).",
4264 enabled ? "Enabling" : "Disabling",
4265 strna(show_status_to_string(mode)),
4266 reason);
4267
4268 set_show_status_marker(enabled);
4269 }
4270
4271 m->show_status = mode;
4272 }
4273
4274 void manager_override_show_status(Manager *m, ShowStatus mode, const char *reason) {
4275 assert(m);
4276 assert(mode < _SHOW_STATUS_MAX);
4277
4278 if (MANAGER_IS_USER(m))
4279 return;
4280
4281 if (mode == m->show_status_overridden)
4282 return;
4283
4284 m->show_status_overridden = mode;
4285
4286 if (mode == _SHOW_STATUS_INVALID)
4287 mode = m->show_status;
4288
4289 log_debug("%s (%s) showing of status (%s).",
4290 m->show_status_overridden != _SHOW_STATUS_INVALID ? "Overriding" : "Restoring",
4291 strna(show_status_to_string(mode)),
4292 reason);
4293
4294 set_show_status_marker(show_status_on(mode));
4295 }
4296
4297 const char *manager_get_confirm_spawn(Manager *m) {
4298 static int last_errno = 0;
4299 struct stat st;
4300 int r;
4301
4302 assert(m);
4303
4304 /* Here's the deal: we want to test the validity of the console but don't want
4305 * PID1 to go through the whole console process which might block. But we also
4306 * want to warn the user only once if something is wrong with the console so we
4307 * cannot do the sanity checks after spawning our children. So here we simply do
4308 * really basic tests to hopefully trap common errors.
4309 *
4310 * If the console suddenly disappear at the time our children will really it
4311 * then they will simply fail to acquire it and a positive answer will be
4312 * assumed. New children will fall back to /dev/console though.
4313 *
4314 * Note: TTYs are devices that can come and go any time, and frequently aren't
4315 * available yet during early boot (consider a USB rs232 dongle...). If for any
4316 * reason the configured console is not ready, we fall back to the default
4317 * console. */
4318
4319 if (!m->confirm_spawn || path_equal(m->confirm_spawn, "/dev/console"))
4320 return m->confirm_spawn;
4321
4322 if (stat(m->confirm_spawn, &st) < 0) {
4323 r = -errno;
4324 goto fail;
4325 }
4326
4327 if (!S_ISCHR(st.st_mode)) {
4328 r = -ENOTTY;
4329 goto fail;
4330 }
4331
4332 last_errno = 0;
4333 return m->confirm_spawn;
4334
4335 fail:
4336 if (last_errno != r)
4337 last_errno = log_warning_errno(r, "Failed to open %s, using default console: %m", m->confirm_spawn);
4338
4339 return "/dev/console";
4340 }
4341
4342 void manager_set_first_boot(Manager *m, bool b) {
4343 assert(m);
4344
4345 if (!MANAGER_IS_SYSTEM(m))
4346 return;
4347
4348 if (m->first_boot != (int) b) {
4349 if (b)
4350 (void) touch("/run/systemd/first-boot");
4351 else
4352 (void) unlink("/run/systemd/first-boot");
4353 }
4354
4355 m->first_boot = b;
4356 }
4357
4358 void manager_disable_confirm_spawn(void) {
4359 (void) touch("/run/systemd/confirm_spawn_disabled");
4360 }
4361
4362 bool manager_is_confirm_spawn_disabled(Manager *m) {
4363 if (!m->confirm_spawn)
4364 return true;
4365
4366 return access("/run/systemd/confirm_spawn_disabled", F_OK) >= 0;
4367 }
4368
4369 static bool manager_should_show_status(Manager *m, StatusType type) {
4370 assert(m);
4371
4372 if (!MANAGER_IS_SYSTEM(m))
4373 return false;
4374
4375 if (m->no_console_output)
4376 return false;
4377
4378 if (!IN_SET(manager_state(m), MANAGER_INITIALIZING, MANAGER_STARTING, MANAGER_STOPPING))
4379 return false;
4380
4381 /* If we cannot find out the status properly, just proceed. */
4382 if (type != STATUS_TYPE_EMERGENCY && manager_check_ask_password(m) > 0)
4383 return false;
4384
4385 if (type == STATUS_TYPE_NOTICE && m->show_status != SHOW_STATUS_NO)
4386 return true;
4387
4388 return manager_get_show_status_on(m);
4389 }
4390
4391 void manager_status_printf(Manager *m, StatusType type, const char *status, const char *format, ...) {
4392 va_list ap;
4393
4394 /* If m is NULL, assume we're after shutdown and let the messages through. */
4395
4396 if (m && !manager_should_show_status(m, type))
4397 return;
4398
4399 /* XXX We should totally drop the check for ephemeral here
4400 * and thus effectively make 'Type=idle' pointless. */
4401 if (type == STATUS_TYPE_EPHEMERAL && m && m->n_on_console > 0)
4402 return;
4403
4404 va_start(ap, format);
4405 status_vprintf(status, SHOW_STATUS_ELLIPSIZE|(type == STATUS_TYPE_EPHEMERAL ? SHOW_STATUS_EPHEMERAL : 0), format, ap);
4406 va_end(ap);
4407 }
4408
4409 Set* manager_get_units_requiring_mounts_for(Manager *m, const char *path) {
4410 assert(m);
4411 assert(path);
4412
4413 if (path_equal(path, "/"))
4414 path = "";
4415
4416 return hashmap_get(m->units_requiring_mounts_for, path);
4417 }
4418
4419 int manager_update_failed_units(Manager *m, Unit *u, bool failed) {
4420 unsigned size;
4421 int r;
4422
4423 assert(m);
4424 assert(u->manager == m);
4425
4426 size = set_size(m->failed_units);
4427
4428 if (failed) {
4429 r = set_ensure_put(&m->failed_units, NULL, u);
4430 if (r < 0)
4431 return log_oom();
4432 } else
4433 (void) set_remove(m->failed_units, u);
4434
4435 if (set_size(m->failed_units) != size)
4436 bus_manager_send_change_signal(m);
4437
4438 return 0;
4439 }
4440
4441 ManagerState manager_state(Manager *m) {
4442 Unit *u;
4443
4444 assert(m);
4445
4446 /* Is the special shutdown target active or queued? If so, we are in shutdown state */
4447 u = manager_get_unit(m, SPECIAL_SHUTDOWN_TARGET);
4448 if (u && unit_active_or_pending(u))
4449 return MANAGER_STOPPING;
4450
4451 /* Did we ever finish booting? If not then we are still starting up */
4452 if (!MANAGER_IS_FINISHED(m)) {
4453
4454 u = manager_get_unit(m, SPECIAL_BASIC_TARGET);
4455 if (!u || !UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u)))
4456 return MANAGER_INITIALIZING;
4457
4458 return MANAGER_STARTING;
4459 }
4460
4461 if (MANAGER_IS_SYSTEM(m)) {
4462 /* Are the rescue or emergency targets active or queued? If so we are in maintenance state */
4463 u = manager_get_unit(m, SPECIAL_RESCUE_TARGET);
4464 if (u && unit_active_or_pending(u))
4465 return MANAGER_MAINTENANCE;
4466
4467 u = manager_get_unit(m, SPECIAL_EMERGENCY_TARGET);
4468 if (u && unit_active_or_pending(u))
4469 return MANAGER_MAINTENANCE;
4470 }
4471
4472 /* Are there any failed units? If so, we are in degraded mode */
4473 if (set_size(m->failed_units) > 0)
4474 return MANAGER_DEGRADED;
4475
4476 return MANAGER_RUNNING;
4477 }
4478
4479 static void manager_unref_uid_internal(
4480 Hashmap *uid_refs,
4481 uid_t uid,
4482 bool destroy_now,
4483 int (*_clean_ipc)(uid_t uid)) {
4484
4485 uint32_t c, n;
4486
4487 assert(uid_is_valid(uid));
4488 assert(_clean_ipc);
4489
4490 /* A generic implementation, covering both manager_unref_uid() and manager_unref_gid(), under the
4491 * assumption that uid_t and gid_t are actually defined the same way, with the same validity rules.
4492 *
4493 * We store a hashmap where the key is the UID/GID and the value is a 32bit reference counter, whose
4494 * highest bit is used as flag for marking UIDs/GIDs whose IPC objects to remove when the last
4495 * reference to the UID/GID is dropped. The flag is set to on, once at least one reference from a
4496 * unit where RemoveIPC= is set is added on a UID/GID. It is reset when the UID's/GID's reference
4497 * counter drops to 0 again. */
4498
4499 assert_cc(sizeof(uid_t) == sizeof(gid_t));
4500 assert_cc(UID_INVALID == (uid_t) GID_INVALID);
4501
4502 if (uid == 0) /* We don't keep track of root, and will never destroy it */
4503 return;
4504
4505 c = PTR_TO_UINT32(hashmap_get(uid_refs, UID_TO_PTR(uid)));
4506
4507 n = c & ~DESTROY_IPC_FLAG;
4508 assert(n > 0);
4509 n--;
4510
4511 if (destroy_now && n == 0) {
4512 hashmap_remove(uid_refs, UID_TO_PTR(uid));
4513
4514 if (c & DESTROY_IPC_FLAG) {
4515 log_debug("%s " UID_FMT " is no longer referenced, cleaning up its IPC.",
4516 _clean_ipc == clean_ipc_by_uid ? "UID" : "GID",
4517 uid);
4518 (void) _clean_ipc(uid);
4519 }
4520 } else {
4521 c = n | (c & DESTROY_IPC_FLAG);
4522 assert_se(hashmap_update(uid_refs, UID_TO_PTR(uid), UINT32_TO_PTR(c)) >= 0);
4523 }
4524 }
4525
4526 void manager_unref_uid(Manager *m, uid_t uid, bool destroy_now) {
4527 manager_unref_uid_internal(m->uid_refs, uid, destroy_now, clean_ipc_by_uid);
4528 }
4529
4530 void manager_unref_gid(Manager *m, gid_t gid, bool destroy_now) {
4531 manager_unref_uid_internal(m->gid_refs, (uid_t) gid, destroy_now, clean_ipc_by_gid);
4532 }
4533
4534 static int manager_ref_uid_internal(
4535 Hashmap **uid_refs,
4536 uid_t uid,
4537 bool clean_ipc) {
4538
4539 uint32_t c, n;
4540 int r;
4541
4542 assert(uid_refs);
4543 assert(uid_is_valid(uid));
4544
4545 /* A generic implementation, covering both manager_ref_uid() and manager_ref_gid(), under the
4546 * assumption that uid_t and gid_t are actually defined the same way, with the same validity
4547 * rules. */
4548
4549 assert_cc(sizeof(uid_t) == sizeof(gid_t));
4550 assert_cc(UID_INVALID == (uid_t) GID_INVALID);
4551
4552 if (uid == 0) /* We don't keep track of root, and will never destroy it */
4553 return 0;
4554
4555 r = hashmap_ensure_allocated(uid_refs, &trivial_hash_ops);
4556 if (r < 0)
4557 return r;
4558
4559 c = PTR_TO_UINT32(hashmap_get(*uid_refs, UID_TO_PTR(uid)));
4560
4561 n = c & ~DESTROY_IPC_FLAG;
4562 n++;
4563
4564 if (n & DESTROY_IPC_FLAG) /* check for overflow */
4565 return -EOVERFLOW;
4566
4567 c = n | (c & DESTROY_IPC_FLAG) | (clean_ipc ? DESTROY_IPC_FLAG : 0);
4568
4569 return hashmap_replace(*uid_refs, UID_TO_PTR(uid), UINT32_TO_PTR(c));
4570 }
4571
4572 int manager_ref_uid(Manager *m, uid_t uid, bool clean_ipc) {
4573 return manager_ref_uid_internal(&m->uid_refs, uid, clean_ipc);
4574 }
4575
4576 int manager_ref_gid(Manager *m, gid_t gid, bool clean_ipc) {
4577 return manager_ref_uid_internal(&m->gid_refs, (uid_t) gid, clean_ipc);
4578 }
4579
4580 static void manager_vacuum_uid_refs_internal(
4581 Hashmap *uid_refs,
4582 int (*_clean_ipc)(uid_t uid)) {
4583
4584 void *p, *k;
4585
4586 assert(_clean_ipc);
4587
4588 HASHMAP_FOREACH_KEY(p, k, uid_refs) {
4589 uint32_t c, n;
4590 uid_t uid;
4591
4592 uid = PTR_TO_UID(k);
4593 c = PTR_TO_UINT32(p);
4594
4595 n = c & ~DESTROY_IPC_FLAG;
4596 if (n > 0)
4597 continue;
4598
4599 if (c & DESTROY_IPC_FLAG) {
4600 log_debug("Found unreferenced %s " UID_FMT " after reload/reexec. Cleaning up.",
4601 _clean_ipc == clean_ipc_by_uid ? "UID" : "GID",
4602 uid);
4603 (void) _clean_ipc(uid);
4604 }
4605
4606 assert_se(hashmap_remove(uid_refs, k) == p);
4607 }
4608 }
4609
4610 static void manager_vacuum_uid_refs(Manager *m) {
4611 manager_vacuum_uid_refs_internal(m->uid_refs, clean_ipc_by_uid);
4612 }
4613
4614 static void manager_vacuum_gid_refs(Manager *m) {
4615 manager_vacuum_uid_refs_internal(m->gid_refs, clean_ipc_by_gid);
4616 }
4617
4618 static void manager_vacuum(Manager *m) {
4619 assert(m);
4620
4621 /* Release any dynamic users no longer referenced */
4622 dynamic_user_vacuum(m, true);
4623
4624 /* Release any references to UIDs/GIDs no longer referenced, and destroy any IPC owned by them */
4625 manager_vacuum_uid_refs(m);
4626 manager_vacuum_gid_refs(m);
4627
4628 /* Release any runtimes no longer referenced */
4629 exec_shared_runtime_vacuum(m);
4630 }
4631
4632 int manager_dispatch_user_lookup_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
4633 struct buffer {
4634 uid_t uid;
4635 gid_t gid;
4636 char unit_name[UNIT_NAME_MAX+1];
4637 } _packed_ buffer;
4638
4639 Manager *m = userdata;
4640 ssize_t l;
4641 size_t n;
4642 Unit *u;
4643
4644 assert_se(source);
4645 assert_se(m);
4646
4647 /* Invoked whenever a child process succeeded resolving its user/group to use and sent us the
4648 * resulting UID/GID in a datagram. We parse the datagram here and pass it off to the unit, so that
4649 * it can add a reference to the UID/GID so that it can destroy the UID/GID's IPC objects when the
4650 * reference counter drops to 0. */
4651
4652 l = recv(fd, &buffer, sizeof(buffer), MSG_DONTWAIT);
4653 if (l < 0) {
4654 if (ERRNO_IS_TRANSIENT(errno))
4655 return 0;
4656
4657 return log_error_errno(errno, "Failed to read from user lookup fd: %m");
4658 }
4659
4660 if ((size_t) l <= offsetof(struct buffer, unit_name)) {
4661 log_warning("Received too short user lookup message, ignoring.");
4662 return 0;
4663 }
4664
4665 if ((size_t) l > offsetof(struct buffer, unit_name) + UNIT_NAME_MAX) {
4666 log_warning("Received too long user lookup message, ignoring.");
4667 return 0;
4668 }
4669
4670 if (!uid_is_valid(buffer.uid) && !gid_is_valid(buffer.gid)) {
4671 log_warning("Got user lookup message with invalid UID/GID pair, ignoring.");
4672 return 0;
4673 }
4674
4675 n = (size_t) l - offsetof(struct buffer, unit_name);
4676 if (memchr(buffer.unit_name, 0, n)) {
4677 log_warning("Received lookup message with embedded NUL character, ignoring.");
4678 return 0;
4679 }
4680
4681 buffer.unit_name[n] = 0;
4682 u = manager_get_unit(m, buffer.unit_name);
4683 if (!u) {
4684 log_debug("Got user lookup message but unit doesn't exist, ignoring.");
4685 return 0;
4686 }
4687
4688 log_unit_debug(u, "User lookup succeeded: uid=" UID_FMT " gid=" GID_FMT, buffer.uid, buffer.gid);
4689
4690 unit_notify_user_lookup(u, buffer.uid, buffer.gid);
4691 return 0;
4692 }
4693
4694 static int short_uid_range(const char *path) {
4695 _cleanup_(uid_range_freep) UidRange *p = NULL;
4696 int r;
4697
4698 assert(path);
4699
4700 /* Taint systemd if we the UID range assigned to this environment doesn't at least cover 0…65534,
4701 * i.e. from root to nobody. */
4702
4703 r = uid_range_load_userns(&p, path);
4704 if (r < 0) {
4705 if (ERRNO_IS_NOT_SUPPORTED(r))
4706 return false;
4707 return log_debug_errno(r, "Failed to load %s: %m", path);
4708 }
4709
4710 return !uid_range_covers(p, 0, 65535);
4711 }
4712
4713 char* manager_taint_string(const Manager *m) {
4714 /* Returns a "taint string", e.g. "local-hwclock:var-run-bad". Only things that are detected at
4715 * runtime should be tagged here. For stuff that is known during compilation, emit a warning in the
4716 * configuration phase. */
4717
4718 assert(m);
4719
4720 const char* stage[13] = {};
4721 size_t n = 0;
4722
4723 if (m->taint_usr)
4724 stage[n++] = "split-usr";
4725
4726 _cleanup_free_ char *usrbin = NULL;
4727 if (readlink_malloc("/bin", &usrbin) < 0 || !PATH_IN_SET(usrbin, "usr/bin", "/usr/bin"))
4728 stage[n++] = "unmerged-usr";
4729
4730 if (access("/proc/cgroups", F_OK) < 0)
4731 stage[n++] = "cgroups-missing";
4732
4733 if (cg_all_unified() == 0)
4734 stage[n++] = "cgroupsv1";
4735
4736 if (clock_is_localtime(NULL) > 0)
4737 stage[n++] = "local-hwclock";
4738
4739 if (os_release_support_ended(NULL, /* quiet= */ true, NULL) > 0)
4740 stage[n++] = "support-ended";
4741
4742 _cleanup_free_ char *destination = NULL;
4743 if (readlink_malloc("/var/run", &destination) < 0 ||
4744 !PATH_IN_SET(destination, "../run", "/run"))
4745 stage[n++] = "var-run-bad";
4746
4747 _cleanup_free_ char *overflowuid = NULL, *overflowgid = NULL;
4748 if (read_one_line_file("/proc/sys/kernel/overflowuid", &overflowuid) >= 0 &&
4749 !streq(overflowuid, "65534"))
4750 stage[n++] = "overflowuid-not-65534";
4751 if (read_one_line_file("/proc/sys/kernel/overflowgid", &overflowgid) >= 0 &&
4752 !streq(overflowgid, "65534"))
4753 stage[n++] = "overflowgid-not-65534";
4754
4755 struct utsname uts;
4756 assert_se(uname(&uts) >= 0);
4757 if (strverscmp_improved(uts.release, KERNEL_BASELINE_VERSION) < 0)
4758 stage[n++] = "old-kernel";
4759
4760 if (short_uid_range("/proc/self/uid_map") > 0)
4761 stage[n++] = "short-uid-range";
4762 if (short_uid_range("/proc/self/gid_map") > 0)
4763 stage[n++] = "short-gid-range";
4764
4765 assert(n < ELEMENTSOF(stage) - 1); /* One extra for NULL terminator */
4766
4767 return strv_join((char**) stage, ":");
4768 }
4769
4770 void manager_ref_console(Manager *m) {
4771 assert(m);
4772
4773 m->n_on_console++;
4774 }
4775
4776 void manager_unref_console(Manager *m) {
4777
4778 assert(m->n_on_console > 0);
4779 m->n_on_console--;
4780
4781 if (m->n_on_console == 0)
4782 m->no_console_output = false; /* unset no_console_output flag, since the console is definitely free now */
4783 }
4784
4785 void manager_override_log_level(Manager *m, int level) {
4786 _cleanup_free_ char *s = NULL;
4787 assert(m);
4788
4789 if (!m->log_level_overridden) {
4790 m->original_log_level = log_get_max_level();
4791 m->log_level_overridden = true;
4792 }
4793
4794 (void) log_level_to_string_alloc(level, &s);
4795 log_info("Setting log level to %s.", strna(s));
4796
4797 log_set_max_level(level);
4798 }
4799
4800 void manager_restore_original_log_level(Manager *m) {
4801 _cleanup_free_ char *s = NULL;
4802 assert(m);
4803
4804 if (!m->log_level_overridden)
4805 return;
4806
4807 (void) log_level_to_string_alloc(m->original_log_level, &s);
4808 log_info("Restoring log level to original (%s).", strna(s));
4809
4810 log_set_max_level(m->original_log_level);
4811 m->log_level_overridden = false;
4812 }
4813
4814 void manager_override_log_target(Manager *m, LogTarget target) {
4815 assert(m);
4816
4817 if (!m->log_target_overridden) {
4818 m->original_log_target = log_get_target();
4819 m->log_target_overridden = true;
4820 }
4821
4822 log_info("Setting log target to %s.", log_target_to_string(target));
4823 log_set_target(target);
4824 }
4825
4826 void manager_restore_original_log_target(Manager *m) {
4827 assert(m);
4828
4829 if (!m->log_target_overridden)
4830 return;
4831
4832 log_info("Restoring log target to original %s.", log_target_to_string(m->original_log_target));
4833
4834 log_set_target(m->original_log_target);
4835 m->log_target_overridden = false;
4836 }
4837
4838 ManagerTimestamp manager_timestamp_initrd_mangle(ManagerTimestamp s) {
4839 if (in_initrd() &&
4840 s >= MANAGER_TIMESTAMP_SECURITY_START &&
4841 s <= MANAGER_TIMESTAMP_UNITS_LOAD_FINISH)
4842 return s - MANAGER_TIMESTAMP_SECURITY_START + MANAGER_TIMESTAMP_INITRD_SECURITY_START;
4843 return s;
4844 }
4845
4846 static const char *const manager_state_table[_MANAGER_STATE_MAX] = {
4847 [MANAGER_INITIALIZING] = "initializing",
4848 [MANAGER_STARTING] = "starting",
4849 [MANAGER_RUNNING] = "running",
4850 [MANAGER_DEGRADED] = "degraded",
4851 [MANAGER_MAINTENANCE] = "maintenance",
4852 [MANAGER_STOPPING] = "stopping",
4853 };
4854
4855 DEFINE_STRING_TABLE_LOOKUP(manager_state, ManagerState);
4856
4857 static const char *const manager_timestamp_table[_MANAGER_TIMESTAMP_MAX] = {
4858 [MANAGER_TIMESTAMP_FIRMWARE] = "firmware",
4859 [MANAGER_TIMESTAMP_LOADER] = "loader",
4860 [MANAGER_TIMESTAMP_KERNEL] = "kernel",
4861 [MANAGER_TIMESTAMP_INITRD] = "initrd",
4862 [MANAGER_TIMESTAMP_USERSPACE] = "userspace",
4863 [MANAGER_TIMESTAMP_FINISH] = "finish",
4864 [MANAGER_TIMESTAMP_SECURITY_START] = "security-start",
4865 [MANAGER_TIMESTAMP_SECURITY_FINISH] = "security-finish",
4866 [MANAGER_TIMESTAMP_GENERATORS_START] = "generators-start",
4867 [MANAGER_TIMESTAMP_GENERATORS_FINISH] = "generators-finish",
4868 [MANAGER_TIMESTAMP_UNITS_LOAD_START] = "units-load-start",
4869 [MANAGER_TIMESTAMP_UNITS_LOAD_FINISH] = "units-load-finish",
4870 [MANAGER_TIMESTAMP_UNITS_LOAD] = "units-load",
4871 [MANAGER_TIMESTAMP_INITRD_SECURITY_START] = "initrd-security-start",
4872 [MANAGER_TIMESTAMP_INITRD_SECURITY_FINISH] = "initrd-security-finish",
4873 [MANAGER_TIMESTAMP_INITRD_GENERATORS_START] = "initrd-generators-start",
4874 [MANAGER_TIMESTAMP_INITRD_GENERATORS_FINISH] = "initrd-generators-finish",
4875 [MANAGER_TIMESTAMP_INITRD_UNITS_LOAD_START] = "initrd-units-load-start",
4876 [MANAGER_TIMESTAMP_INITRD_UNITS_LOAD_FINISH] = "initrd-units-load-finish",
4877 };
4878
4879 DEFINE_STRING_TABLE_LOOKUP(manager_timestamp, ManagerTimestamp);
4880
4881 static const char* const oom_policy_table[_OOM_POLICY_MAX] = {
4882 [OOM_CONTINUE] = "continue",
4883 [OOM_STOP] = "stop",
4884 [OOM_KILL] = "kill",
4885 };
4886
4887 DEFINE_STRING_TABLE_LOOKUP(oom_policy, OOMPolicy);
Unnamed repository; edit this file 'description' to name the repository.