]> git.ipfire.org Git - thirdparty/systemd.git/blob - src/core/manager.c
projects / thirdparty / systemd.git / blob
? search:


57bb25ca2543507ff5a82dc989af453a81435abe
[thirdparty/systemd.git] / src / core / manager.c
1 /* SPDX-License-Identifier: LGPL-2.1-or-later */
2
3 #include <errno.h>
4 #include <fcntl.h>
5 #include <linux/kd.h>
6 #include <sys/epoll.h>
7 #include <sys/inotify.h>
8 #include <sys/ioctl.h>
9 #include <sys/reboot.h>
10 #include <sys/timerfd.h>
11 #include <sys/wait.h>
12 #include <unistd.h>
13
14 #if HAVE_AUDIT
15 #include <libaudit.h>
16 #endif
17
18 #include "sd-daemon.h"
19 #include "sd-messages.h"
20 #include "sd-path.h"
21
22 #include "all-units.h"
23 #include "alloc-util.h"
24 #include "audit-fd.h"
25 #include "boot-timestamps.h"
26 #include "bus-common-errors.h"
27 #include "bus-error.h"
28 #include "bus-kernel.h"
29 #include "bus-util.h"
30 #include "clean-ipc.h"
31 #include "clock-util.h"
32 #include "core-varlink.h"
33 #include "creds-util.h"
34 #include "dbus-job.h"
35 #include "dbus-manager.h"
36 #include "dbus-unit.h"
37 #include "dbus.h"
38 #include "def.h"
39 #include "dirent-util.h"
40 #include "env-util.h"
41 #include "escape.h"
42 #include "exec-util.h"
43 #include "execute.h"
44 #include "exit-status.h"
45 #include "fd-util.h"
46 #include "fileio.h"
47 #include "fs-util.h"
48 #include "generator-setup.h"
49 #include "hashmap.h"
50 #include "install.h"
51 #include "io-util.h"
52 #include "label.h"
53 #include "load-fragment.h"
54 #include "locale-setup.h"
55 #include "log.h"
56 #include "macro.h"
57 #include "manager.h"
58 #include "memory-util.h"
59 #include "mkdir.h"
60 #include "parse-util.h"
61 #include "path-lookup.h"
62 #include "path-util.h"
63 #include "process-util.h"
64 #include "ratelimit.h"
65 #include "rlimit-util.h"
66 #include "rm-rf.h"
67 #include "selinux-util.h"
68 #include "serialize.h"
69 #include "signal-util.h"
70 #include "socket-util.h"
71 #include "special.h"
72 #include "stat-util.h"
73 #include "string-table.h"
74 #include "string-util.h"
75 #include "strv.h"
76 #include "strxcpyx.h"
77 #include "sysctl-util.h"
78 #include "syslog-util.h"
79 #include "terminal-util.h"
80 #include "time-util.h"
81 #include "transaction.h"
82 #include "umask-util.h"
83 #include "unit-name.h"
84 #include "unit-serialize.h"
85 #include "user-util.h"
86 #include "virt.h"
87 #include "watchdog.h"
88
89 #define NOTIFY_RCVBUF_SIZE (8*1024*1024)
90 #define CGROUPS_AGENT_RCVBUF_SIZE (8*1024*1024)
91
92 /* Initial delay and the interval for printing status messages about running jobs */
93 #define JOBS_IN_PROGRESS_WAIT_USEC (2*USEC_PER_SEC)
94 #define JOBS_IN_PROGRESS_QUIET_WAIT_USEC (25*USEC_PER_SEC)
95 #define JOBS_IN_PROGRESS_PERIOD_USEC (USEC_PER_SEC / 3)
96 #define JOBS_IN_PROGRESS_PERIOD_DIVISOR 3
97
98 /* If there are more than 1K bus messages queue across our API and direct buses, then let's not add more on top until
99 * the queue gets more empty. */
100 #define MANAGER_BUS_BUSY_THRESHOLD 1024LU
101
102 /* How many units and jobs to process of the bus queue before returning to the event loop. */
103 #define MANAGER_BUS_MESSAGE_BUDGET 100U
104
105 static int manager_dispatch_notify_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata);
106 static int manager_dispatch_cgroups_agent_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata);
107 static int manager_dispatch_signal_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata);
108 static int manager_dispatch_time_change_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata);
109 static int manager_dispatch_idle_pipe_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata);
110 static int manager_dispatch_user_lookup_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata);
111 static int manager_dispatch_jobs_in_progress(sd_event_source *source, usec_t usec, void *userdata);
112 static int manager_dispatch_run_queue(sd_event_source *source, void *userdata);
113 static int manager_dispatch_sigchld(sd_event_source *source, void *userdata);
114 static int manager_dispatch_timezone_change(sd_event_source *source, const struct inotify_event *event, void *userdata);
115 static int manager_run_environment_generators(Manager *m);
116 static int manager_run_generators(Manager *m);
117 static void manager_vacuum(Manager *m);
118
119 static usec_t manager_watch_jobs_next_time(Manager *m) {
120 return usec_add(now(CLOCK_MONOTONIC),
121 show_status_on(m->show_status) ? JOBS_IN_PROGRESS_WAIT_USEC :
122 JOBS_IN_PROGRESS_QUIET_WAIT_USEC);
123 }
124
125 static void manager_watch_jobs_in_progress(Manager *m) {
126 usec_t next;
127 int r;
128
129 assert(m);
130
131 /* We do not want to show the cylon animation if the user
132 * needs to confirm service executions otherwise confirmation
133 * messages will be screwed by the cylon animation. */
134 if (!manager_is_confirm_spawn_disabled(m))
135 return;
136
137 if (m->jobs_in_progress_event_source)
138 return;
139
140 next = manager_watch_jobs_next_time(m);
141 r = sd_event_add_time(
142 m->event,
143 &m->jobs_in_progress_event_source,
144 CLOCK_MONOTONIC,
145 next, 0,
146 manager_dispatch_jobs_in_progress, m);
147 if (r < 0)
148 return;
149
150 (void) sd_event_source_set_description(m->jobs_in_progress_event_source, "manager-jobs-in-progress");
151 }
152
153 #define CYLON_BUFFER_EXTRA (2*STRLEN(ANSI_RED) + STRLEN(ANSI_HIGHLIGHT_RED) + 2*STRLEN(ANSI_NORMAL))
154
155 static void draw_cylon(char buffer[], size_t buflen, unsigned width, unsigned pos) {
156 char *p = buffer;
157
158 assert(buflen >= CYLON_BUFFER_EXTRA + width + 1);
159 assert(pos <= width+1); /* 0 or width+1 mean that the center light is behind the corner */
160
161 if (pos > 1) {
162 if (pos > 2)
163 p = mempset(p, ' ', pos-2);
164 if (log_get_show_color())
165 p = stpcpy(p, ANSI_RED);
166 *p++ = '*';
167 }
168
169 if (pos > 0 && pos <= width) {
170 if (log_get_show_color())
171 p = stpcpy(p, ANSI_HIGHLIGHT_RED);
172 *p++ = '*';
173 }
174
175 if (log_get_show_color())
176 p = stpcpy(p, ANSI_NORMAL);
177
178 if (pos < width) {
179 if (log_get_show_color())
180 p = stpcpy(p, ANSI_RED);
181 *p++ = '*';
182 if (pos < width-1)
183 p = mempset(p, ' ', width-1-pos);
184 if (log_get_show_color())
185 strcpy(p, ANSI_NORMAL);
186 }
187 }
188
189 static void manager_flip_auto_status(Manager *m, bool enable, const char *reason) {
190 assert(m);
191
192 if (enable) {
193 if (m->show_status == SHOW_STATUS_AUTO)
194 manager_set_show_status(m, SHOW_STATUS_TEMPORARY, reason);
195 } else {
196 if (m->show_status == SHOW_STATUS_TEMPORARY)
197 manager_set_show_status(m, SHOW_STATUS_AUTO, reason);
198 }
199 }
200
201 static void manager_print_jobs_in_progress(Manager *m) {
202 _cleanup_free_ char *job_of_n = NULL;
203 Job *j;
204 unsigned counter = 0, print_nr;
205 char cylon[6 + CYLON_BUFFER_EXTRA + 1];
206 unsigned cylon_pos;
207 char time[FORMAT_TIMESPAN_MAX], limit[FORMAT_TIMESPAN_MAX] = "no limit";
208 uint64_t x;
209
210 assert(m);
211 assert(m->n_running_jobs > 0);
212
213 manager_flip_auto_status(m, true, "delay");
214
215 print_nr = (m->jobs_in_progress_iteration / JOBS_IN_PROGRESS_PERIOD_DIVISOR) % m->n_running_jobs;
216
217 HASHMAP_FOREACH(j, m->jobs)
218 if (j->state == JOB_RUNNING && counter++ == print_nr)
219 break;
220
221 /* m->n_running_jobs must be consistent with the contents of m->jobs,
222 * so the above loop must have succeeded in finding j. */
223 assert(counter == print_nr + 1);
224 assert(j);
225
226 cylon_pos = m->jobs_in_progress_iteration % 14;
227 if (cylon_pos >= 8)
228 cylon_pos = 14 - cylon_pos;
229 draw_cylon(cylon, sizeof(cylon), 6, cylon_pos);
230
231 m->jobs_in_progress_iteration++;
232
233 if (m->n_running_jobs > 1) {
234 if (asprintf(&job_of_n, "(%u of %u) ", counter, m->n_running_jobs) < 0)
235 job_of_n = NULL;
236 }
237
238 format_timespan(time, sizeof(time), now(CLOCK_MONOTONIC) - j->begin_usec, 1*USEC_PER_SEC);
239 if (job_get_timeout(j, &x) > 0)
240 format_timespan(limit, sizeof(limit), x - j->begin_usec, 1*USEC_PER_SEC);
241
242 manager_status_printf(m, STATUS_TYPE_EPHEMERAL, cylon,
243 "%sA %s job is running for %s (%s / %s)",
244 strempty(job_of_n),
245 job_type_to_string(j->type),
246 unit_status_string(j->unit),
247 time, limit);
248 }
249
250 static int have_ask_password(void) {
251 _cleanup_closedir_ DIR *dir;
252 struct dirent *de;
253
254 dir = opendir("/run/systemd/ask-password");
255 if (!dir) {
256 if (errno == ENOENT)
257 return false;
258 else
259 return -errno;
260 }
261
262 FOREACH_DIRENT_ALL(de, dir, return -errno) {
263 if (startswith(de->d_name, "ask."))
264 return true;
265 }
266 return false;
267 }
268
269 static int manager_dispatch_ask_password_fd(sd_event_source *source,
270 int fd, uint32_t revents, void *userdata) {
271 Manager *m = userdata;
272
273 assert(m);
274
275 (void) flush_fd(fd);
276
277 m->have_ask_password = have_ask_password();
278 if (m->have_ask_password < 0)
279 /* Log error but continue. Negative have_ask_password
280 * is treated as unknown status. */
281 log_error_errno(m->have_ask_password, "Failed to list /run/systemd/ask-password: %m");
282
283 return 0;
284 }
285
286 static void manager_close_ask_password(Manager *m) {
287 assert(m);
288
289 m->ask_password_event_source = sd_event_source_unref(m->ask_password_event_source);
290 m->ask_password_inotify_fd = safe_close(m->ask_password_inotify_fd);
291 m->have_ask_password = -EINVAL;
292 }
293
294 static int manager_check_ask_password(Manager *m) {
295 int r;
296
297 assert(m);
298
299 if (!m->ask_password_event_source) {
300 assert(m->ask_password_inotify_fd < 0);
301
302 (void) mkdir_p_label("/run/systemd/ask-password", 0755);
303
304 m->ask_password_inotify_fd = inotify_init1(IN_NONBLOCK|IN_CLOEXEC);
305 if (m->ask_password_inotify_fd < 0)
306 return log_error_errno(errno, "Failed to create inotify object: %m");
307
308 r = inotify_add_watch_and_warn(m->ask_password_inotify_fd,
309 "/run/systemd/ask-password",
310 IN_CREATE|IN_DELETE|IN_MOVE);
311 if (r < 0) {
312 manager_close_ask_password(m);
313 return r;
314 }
315
316 r = sd_event_add_io(m->event, &m->ask_password_event_source,
317 m->ask_password_inotify_fd, EPOLLIN,
318 manager_dispatch_ask_password_fd, m);
319 if (r < 0) {
320 log_error_errno(r, "Failed to add event source for /run/systemd/ask-password: %m");
321 manager_close_ask_password(m);
322 return r;
323 }
324
325 (void) sd_event_source_set_description(m->ask_password_event_source, "manager-ask-password");
326
327 /* Queries might have been added meanwhile... */
328 manager_dispatch_ask_password_fd(m->ask_password_event_source,
329 m->ask_password_inotify_fd, EPOLLIN, m);
330 }
331
332 return m->have_ask_password;
333 }
334
335 static int manager_watch_idle_pipe(Manager *m) {
336 int r;
337
338 assert(m);
339
340 if (m->idle_pipe_event_source)
341 return 0;
342
343 if (m->idle_pipe[2] < 0)
344 return 0;
345
346 r = sd_event_add_io(m->event, &m->idle_pipe_event_source, m->idle_pipe[2], EPOLLIN, manager_dispatch_idle_pipe_fd, m);
347 if (r < 0)
348 return log_error_errno(r, "Failed to watch idle pipe: %m");
349
350 (void) sd_event_source_set_description(m->idle_pipe_event_source, "manager-idle-pipe");
351
352 return 0;
353 }
354
355 static void manager_close_idle_pipe(Manager *m) {
356 assert(m);
357
358 m->idle_pipe_event_source = sd_event_source_unref(m->idle_pipe_event_source);
359
360 safe_close_pair(m->idle_pipe);
361 safe_close_pair(m->idle_pipe + 2);
362 }
363
364 static int manager_setup_time_change(Manager *m) {
365 int r;
366
367 assert(m);
368
369 if (MANAGER_IS_TEST_RUN(m))
370 return 0;
371
372 m->time_change_event_source = sd_event_source_unref(m->time_change_event_source);
373 m->time_change_fd = safe_close(m->time_change_fd);
374
375 m->time_change_fd = time_change_fd();
376 if (m->time_change_fd < 0)
377 return log_error_errno(m->time_change_fd, "Failed to create timer change timer fd: %m");
378
379 r = sd_event_add_io(m->event, &m->time_change_event_source, m->time_change_fd, EPOLLIN, manager_dispatch_time_change_fd, m);
380 if (r < 0)
381 return log_error_errno(r, "Failed to create time change event source: %m");
382
383 /* Schedule this slightly earlier than the .timer event sources */
384 r = sd_event_source_set_priority(m->time_change_event_source, SD_EVENT_PRIORITY_NORMAL-1);
385 if (r < 0)
386 return log_error_errno(r, "Failed to set priority of time change event sources: %m");
387
388 (void) sd_event_source_set_description(m->time_change_event_source, "manager-time-change");
389
390 log_debug("Set up TFD_TIMER_CANCEL_ON_SET timerfd.");
391
392 return 0;
393 }
394
395 static int manager_read_timezone_stat(Manager *m) {
396 struct stat st;
397 bool changed;
398
399 assert(m);
400
401 /* Read the current stat() data of /etc/localtime so that we detect changes */
402 if (lstat("/etc/localtime", &st) < 0) {
403 log_debug_errno(errno, "Failed to stat /etc/localtime, ignoring: %m");
404 changed = m->etc_localtime_accessible;
405 m->etc_localtime_accessible = false;
406 } else {
407 usec_t k;
408
409 k = timespec_load(&st.st_mtim);
410 changed = !m->etc_localtime_accessible || k != m->etc_localtime_mtime;
411
412 m->etc_localtime_mtime = k;
413 m->etc_localtime_accessible = true;
414 }
415
416 return changed;
417 }
418
419 static int manager_setup_timezone_change(Manager *m) {
420 _cleanup_(sd_event_source_unrefp) sd_event_source *new_event = NULL;
421 int r;
422
423 assert(m);
424
425 if (MANAGER_IS_TEST_RUN(m))
426 return 0;
427
428 /* We watch /etc/localtime for three events: change of the link count (which might mean removal from /etc even
429 * though another link might be kept), renames, and file close operations after writing. Note we don't bother
430 * with IN_DELETE_SELF, as that would just report when the inode is removed entirely, i.e. after the link count
431 * went to zero and all fds to it are closed.
432 *
433 * Note that we never follow symlinks here. This is a simplification, but should cover almost all cases
434 * correctly.
435 *
436 * Note that we create the new event source first here, before releasing the old one. This should optimize
437 * behaviour as this way sd-event can reuse the old watch in case the inode didn't change. */
438
439 r = sd_event_add_inotify(m->event, &new_event, "/etc/localtime",
440 IN_ATTRIB|IN_MOVE_SELF|IN_CLOSE_WRITE|IN_DONT_FOLLOW, manager_dispatch_timezone_change, m);
441 if (r == -ENOENT) {
442 /* If the file doesn't exist yet, subscribe to /etc instead, and wait until it is created either by
443 * O_CREATE or by rename() */
444
445 log_debug_errno(r, "/etc/localtime doesn't exist yet, watching /etc instead.");
446 r = sd_event_add_inotify(m->event, &new_event, "/etc",
447 IN_CREATE|IN_MOVED_TO|IN_ONLYDIR, manager_dispatch_timezone_change, m);
448 }
449 if (r < 0)
450 return log_error_errno(r, "Failed to create timezone change event source: %m");
451
452 /* Schedule this slightly earlier than the .timer event sources */
453 r = sd_event_source_set_priority(new_event, SD_EVENT_PRIORITY_NORMAL-1);
454 if (r < 0)
455 return log_error_errno(r, "Failed to set priority of timezone change event sources: %m");
456
457 sd_event_source_unref(m->timezone_change_event_source);
458 m->timezone_change_event_source = TAKE_PTR(new_event);
459
460 return 0;
461 }
462
463 static int enable_special_signals(Manager *m) {
464 _cleanup_close_ int fd = -1;
465
466 assert(m);
467
468 if (MANAGER_IS_TEST_RUN(m))
469 return 0;
470
471 /* Enable that we get SIGINT on control-alt-del. In containers
472 * this will fail with EPERM (older) or EINVAL (newer), so
473 * ignore that. */
474 if (reboot(RB_DISABLE_CAD) < 0 && !IN_SET(errno, EPERM, EINVAL))
475 log_warning_errno(errno, "Failed to enable ctrl-alt-del handling: %m");
476
477 fd = open_terminal("/dev/tty0", O_RDWR|O_NOCTTY|O_CLOEXEC);
478 if (fd < 0) {
479 /* Support systems without virtual console */
480 if (fd != -ENOENT)
481 log_warning_errno(errno, "Failed to open /dev/tty0: %m");
482 } else {
483 /* Enable that we get SIGWINCH on kbrequest */
484 if (ioctl(fd, KDSIGACCEPT, SIGWINCH) < 0)
485 log_warning_errno(errno, "Failed to enable kbrequest handling: %m");
486 }
487
488 return 0;
489 }
490
491 #define RTSIG_IF_AVAILABLE(signum) (signum <= SIGRTMAX ? signum : -1)
492
493 static int manager_setup_signals(Manager *m) {
494 struct sigaction sa = {
495 .sa_handler = SIG_DFL,
496 .sa_flags = SA_NOCLDSTOP|SA_RESTART,
497 };
498 sigset_t mask;
499 int r;
500
501 assert(m);
502
503 assert_se(sigaction(SIGCHLD, &sa, NULL) == 0);
504
505 /* We make liberal use of realtime signals here. On
506 * Linux/glibc we have 30 of them (with the exception of Linux
507 * on hppa, see below), between SIGRTMIN+0 ... SIGRTMIN+30
508 * (aka SIGRTMAX). */
509
510 assert_se(sigemptyset(&mask) == 0);
511 sigset_add_many(&mask,
512 SIGCHLD, /* Child died */
513 SIGTERM, /* Reexecute daemon */
514 SIGHUP, /* Reload configuration */
515 SIGUSR1, /* systemd: reconnect to D-Bus */
516 SIGUSR2, /* systemd: dump status */
517 SIGINT, /* Kernel sends us this on control-alt-del */
518 SIGWINCH, /* Kernel sends us this on kbrequest (alt-arrowup) */
519 SIGPWR, /* Some kernel drivers and upsd send us this on power failure */
520
521 SIGRTMIN+0, /* systemd: start default.target */
522 SIGRTMIN+1, /* systemd: isolate rescue.target */
523 SIGRTMIN+2, /* systemd: isolate emergency.target */
524 SIGRTMIN+3, /* systemd: start halt.target */
525 SIGRTMIN+4, /* systemd: start poweroff.target */
526 SIGRTMIN+5, /* systemd: start reboot.target */
527 SIGRTMIN+6, /* systemd: start kexec.target */
528
529 /* ... space for more special targets ... */
530
531 SIGRTMIN+13, /* systemd: Immediate halt */
532 SIGRTMIN+14, /* systemd: Immediate poweroff */
533 SIGRTMIN+15, /* systemd: Immediate reboot */
534 SIGRTMIN+16, /* systemd: Immediate kexec */
535
536 /* ... space for more immediate system state changes ... */
537
538 SIGRTMIN+20, /* systemd: enable status messages */
539 SIGRTMIN+21, /* systemd: disable status messages */
540 SIGRTMIN+22, /* systemd: set log level to LOG_DEBUG */
541 SIGRTMIN+23, /* systemd: set log level to LOG_INFO */
542 SIGRTMIN+24, /* systemd: Immediate exit (--user only) */
543
544 /* .. one free signal here ... */
545
546 /* Apparently Linux on hppa had fewer RT signals until v3.18,
547 * SIGRTMAX was SIGRTMIN+25, and then SIGRTMIN was lowered,
548 * see commit v3.17-7614-g1f25df2eff.
549 *
550 * We cannot unconditionally make use of those signals here,
551 * so let's use a runtime check. Since these commands are
552 * accessible by different means and only really a safety
553 * net, the missing functionality on hppa shouldn't matter.
554 */
555
556 RTSIG_IF_AVAILABLE(SIGRTMIN+26), /* systemd: set log target to journal-or-kmsg */
557 RTSIG_IF_AVAILABLE(SIGRTMIN+27), /* systemd: set log target to console */
558 RTSIG_IF_AVAILABLE(SIGRTMIN+28), /* systemd: set log target to kmsg */
559 RTSIG_IF_AVAILABLE(SIGRTMIN+29), /* systemd: set log target to syslog-or-kmsg (obsolete) */
560
561 /* ... one free signal here SIGRTMIN+30 ... */
562 -1);
563 assert_se(sigprocmask(SIG_SETMASK, &mask, NULL) == 0);
564
565 m->signal_fd = signalfd(-1, &mask, SFD_NONBLOCK|SFD_CLOEXEC);
566 if (m->signal_fd < 0)
567 return -errno;
568
569 r = sd_event_add_io(m->event, &m->signal_event_source, m->signal_fd, EPOLLIN, manager_dispatch_signal_fd, m);
570 if (r < 0)
571 return r;
572
573 (void) sd_event_source_set_description(m->signal_event_source, "manager-signal");
574
575 /* Process signals a bit earlier than the rest of things, but later than notify_fd processing, so that the
576 * notify processing can still figure out to which process/service a message belongs, before we reap the
577 * process. Also, process this before handling cgroup notifications, so that we always collect child exit
578 * status information before detecting that there's no process in a cgroup. */
579 r = sd_event_source_set_priority(m->signal_event_source, SD_EVENT_PRIORITY_NORMAL-6);
580 if (r < 0)
581 return r;
582
583 if (MANAGER_IS_SYSTEM(m))
584 return enable_special_signals(m);
585
586 return 0;
587 }
588
589 static char** sanitize_environment(char **l) {
590
591 /* Let's remove some environment variables that we need ourselves to communicate with our clients */
592 strv_env_unset_many(
593 l,
594 "CACHE_DIRECTORY",
595 "CONFIGURATION_DIRECTORY",
596 "CREDENTIALS_DIRECTORY",
597 "EXIT_CODE",
598 "EXIT_STATUS",
599 "INVOCATION_ID",
600 "JOURNAL_STREAM",
601 "LISTEN_FDNAMES",
602 "LISTEN_FDS",
603 "LISTEN_PID",
604 "LOGS_DIRECTORY",
605 "MAINPID",
606 "MANAGERPID",
607 "NOTIFY_SOCKET",
608 "PIDFILE",
609 "REMOTE_ADDR",
610 "REMOTE_PORT",
611 "RUNTIME_DIRECTORY",
612 "SERVICE_RESULT",
613 "STATE_DIRECTORY",
614 "WATCHDOG_PID",
615 "WATCHDOG_USEC",
616 NULL);
617
618 /* Let's order the environment alphabetically, just to make it pretty */
619 strv_sort(l);
620
621 return l;
622 }
623
624 int manager_default_environment(Manager *m) {
625 int r;
626
627 assert(m);
628
629 m->transient_environment = strv_free(m->transient_environment);
630
631 if (MANAGER_IS_SYSTEM(m)) {
632 /* The system manager always starts with a clean
633 * environment for its children. It does not import
634 * the kernel's or the parents' exported variables.
635 *
636 * The initial passed environment is untouched to keep
637 * /proc/self/environ valid; it is used for tagging
638 * the init process inside containers. */
639 m->transient_environment = strv_new("PATH=" DEFAULT_PATH);
640 if (!m->transient_environment)
641 return log_oom();
642
643 /* Import locale variables LC_*= from configuration */
644 (void) locale_setup(&m->transient_environment);
645 } else {
646 /* The user manager passes its own environment along to its children, except for $PATH. */
647 m->transient_environment = strv_copy(environ);
648 if (!m->transient_environment)
649 return log_oom();
650
651 r = strv_env_replace_strdup(&m->transient_environment, "PATH=" DEFAULT_USER_PATH);
652 if (r < 0)
653 return log_oom();
654 }
655
656 sanitize_environment(m->transient_environment);
657
658 return 0;
659 }
660
661 static int manager_setup_prefix(Manager *m) {
662 struct table_entry {
663 uint64_t type;
664 const char *suffix;
665 };
666
667 static const struct table_entry paths_system[_EXEC_DIRECTORY_TYPE_MAX] = {
668 [EXEC_DIRECTORY_RUNTIME] = { SD_PATH_SYSTEM_RUNTIME, NULL },
669 [EXEC_DIRECTORY_STATE] = { SD_PATH_SYSTEM_STATE_PRIVATE, NULL },
670 [EXEC_DIRECTORY_CACHE] = { SD_PATH_SYSTEM_STATE_CACHE, NULL },
671 [EXEC_DIRECTORY_LOGS] = { SD_PATH_SYSTEM_STATE_LOGS, NULL },
672 [EXEC_DIRECTORY_CONFIGURATION] = { SD_PATH_SYSTEM_CONFIGURATION, NULL },
673 };
674
675 static const struct table_entry paths_user[_EXEC_DIRECTORY_TYPE_MAX] = {
676 [EXEC_DIRECTORY_RUNTIME] = { SD_PATH_USER_RUNTIME, NULL },
677 [EXEC_DIRECTORY_STATE] = { SD_PATH_USER_CONFIGURATION, NULL },
678 [EXEC_DIRECTORY_CACHE] = { SD_PATH_USER_STATE_CACHE, NULL },
679 [EXEC_DIRECTORY_LOGS] = { SD_PATH_USER_CONFIGURATION, "log" },
680 [EXEC_DIRECTORY_CONFIGURATION] = { SD_PATH_USER_CONFIGURATION, NULL },
681 };
682
683 assert(m);
684
685 const struct table_entry *p = MANAGER_IS_SYSTEM(m) ? paths_system : paths_user;
686 int r;
687
688 for (ExecDirectoryType i = 0; i < _EXEC_DIRECTORY_TYPE_MAX; i++) {
689 r = sd_path_lookup(p[i].type, p[i].suffix, &m->prefix[i]);
690 if (r < 0)
691 return r;
692 }
693
694 return 0;
695 }
696
697 static void manager_free_unit_name_maps(Manager *m) {
698 m->unit_id_map = hashmap_free(m->unit_id_map);
699 m->unit_name_map = hashmap_free(m->unit_name_map);
700 m->unit_path_cache = set_free(m->unit_path_cache);
701 m->unit_cache_timestamp_hash = 0;
702 }
703
704 static int manager_setup_run_queue(Manager *m) {
705 int r;
706
707 assert(m);
708 assert(!m->run_queue_event_source);
709
710 r = sd_event_add_defer(m->event, &m->run_queue_event_source, manager_dispatch_run_queue, m);
711 if (r < 0)
712 return r;
713
714 r = sd_event_source_set_priority(m->run_queue_event_source, SD_EVENT_PRIORITY_IDLE);
715 if (r < 0)
716 return r;
717
718 r = sd_event_source_set_enabled(m->run_queue_event_source, SD_EVENT_OFF);
719 if (r < 0)
720 return r;
721
722 (void) sd_event_source_set_description(m->run_queue_event_source, "manager-run-queue");
723
724 return 0;
725 }
726
727 static int manager_setup_sigchld_event_source(Manager *m) {
728 int r;
729
730 assert(m);
731 assert(!m->sigchld_event_source);
732
733 r = sd_event_add_defer(m->event, &m->sigchld_event_source, manager_dispatch_sigchld, m);
734 if (r < 0)
735 return r;
736
737 r = sd_event_source_set_priority(m->sigchld_event_source, SD_EVENT_PRIORITY_NORMAL-7);
738 if (r < 0)
739 return r;
740
741 r = sd_event_source_set_enabled(m->sigchld_event_source, SD_EVENT_OFF);
742 if (r < 0)
743 return r;
744
745 (void) sd_event_source_set_description(m->sigchld_event_source, "manager-sigchld");
746
747 return 0;
748 }
749
750 int manager_new(UnitFileScope scope, ManagerTestRunFlags test_run_flags, Manager **_m) {
751 _cleanup_(manager_freep) Manager *m = NULL;
752 const char *e;
753 int r;
754
755 assert(_m);
756 assert(IN_SET(scope, UNIT_FILE_SYSTEM, UNIT_FILE_USER));
757
758 m = new(Manager, 1);
759 if (!m)
760 return -ENOMEM;
761
762 *m = (Manager) {
763 .unit_file_scope = scope,
764 .objective = _MANAGER_OBJECTIVE_INVALID,
765
766 .status_unit_format = STATUS_UNIT_FORMAT_DEFAULT,
767
768 .default_timer_accuracy_usec = USEC_PER_MINUTE,
769 .default_memory_accounting = MEMORY_ACCOUNTING_DEFAULT,
770 .default_tasks_accounting = true,
771 .default_tasks_max = TASKS_MAX_UNSET,
772 .default_timeout_start_usec = DEFAULT_TIMEOUT_USEC,
773 .default_timeout_stop_usec = DEFAULT_TIMEOUT_USEC,
774 .default_restart_usec = DEFAULT_RESTART_USEC,
775
776 .original_log_level = -1,
777 .original_log_target = _LOG_TARGET_INVALID,
778
779 .watchdog_overridden[WATCHDOG_RUNTIME] = USEC_INFINITY,
780 .watchdog_overridden[WATCHDOG_REBOOT] = USEC_INFINITY,
781 .watchdog_overridden[WATCHDOG_KEXEC] = USEC_INFINITY,
782
783 .show_status_overridden = _SHOW_STATUS_INVALID,
784
785 .notify_fd = -1,
786 .cgroups_agent_fd = -1,
787 .signal_fd = -1,
788 .time_change_fd = -1,
789 .user_lookup_fds = { -1, -1 },
790 .private_listen_fd = -1,
791 .dev_autofs_fd = -1,
792 .cgroup_inotify_fd = -1,
793 .pin_cgroupfs_fd = -1,
794 .ask_password_inotify_fd = -1,
795 .idle_pipe = { -1, -1, -1, -1},
796
797 /* start as id #1, so that we can leave #0 around as "null-like" value */
798 .current_job_id = 1,
799
800 .have_ask_password = -EINVAL, /* we don't know */
801 .first_boot = -1,
802 .test_run_flags = test_run_flags,
803
804 .default_oom_policy = OOM_STOP,
805 };
806
807 #if ENABLE_EFI
808 if (MANAGER_IS_SYSTEM(m) && detect_container() <= 0)
809 boot_timestamps(m->timestamps + MANAGER_TIMESTAMP_USERSPACE,
810 m->timestamps + MANAGER_TIMESTAMP_FIRMWARE,
811 m->timestamps + MANAGER_TIMESTAMP_LOADER);
812 #endif
813
814 /* Prepare log fields we can use for structured logging */
815 if (MANAGER_IS_SYSTEM(m)) {
816 m->unit_log_field = "UNIT=";
817 m->unit_log_format_string = "UNIT=%s";
818
819 m->invocation_log_field = "INVOCATION_ID=";
820 m->invocation_log_format_string = "INVOCATION_ID=%s";
821 } else {
822 m->unit_log_field = "USER_UNIT=";
823 m->unit_log_format_string = "USER_UNIT=%s";
824
825 m->invocation_log_field = "USER_INVOCATION_ID=";
826 m->invocation_log_format_string = "USER_INVOCATION_ID=%s";
827 }
828
829 /* Reboot immediately if the user hits C-A-D more often than 7x per 2s */
830 m->ctrl_alt_del_ratelimit = (RateLimit) { .interval = 2 * USEC_PER_SEC, .burst = 7 };
831
832 r = manager_default_environment(m);
833 if (r < 0)
834 return r;
835
836 r = hashmap_ensure_allocated(&m->units, &string_hash_ops);
837 if (r < 0)
838 return r;
839
840 r = hashmap_ensure_allocated(&m->cgroup_unit, &path_hash_ops);
841 if (r < 0)
842 return r;
843
844 r = hashmap_ensure_allocated(&m->watch_bus, &string_hash_ops);
845 if (r < 0)
846 return r;
847
848 r = prioq_ensure_allocated(&m->run_queue, compare_job_priority);
849 if (r < 0)
850 return r;
851
852 r = manager_setup_prefix(m);
853 if (r < 0)
854 return r;
855
856 r = get_credentials_dir(&e);
857 if (r >= 0) {
858 m->received_credentials = strdup(e);
859 if (!m->received_credentials)
860 return -ENOMEM;
861 }
862
863 r = sd_event_default(&m->event);
864 if (r < 0)
865 return r;
866
867 r = manager_setup_run_queue(m);
868 if (r < 0)
869 return r;
870
871 if (test_run_flags == MANAGER_TEST_RUN_MINIMAL) {
872 m->cgroup_root = strdup("");
873 if (!m->cgroup_root)
874 return -ENOMEM;
875 } else {
876 r = manager_setup_signals(m);
877 if (r < 0)
878 return r;
879
880 r = manager_setup_cgroup(m);
881 if (r < 0)
882 return r;
883
884 r = manager_setup_time_change(m);
885 if (r < 0)
886 return r;
887
888 r = manager_read_timezone_stat(m);
889 if (r < 0)
890 return r;
891
892 (void) manager_setup_timezone_change(m);
893
894 r = manager_setup_sigchld_event_source(m);
895 if (r < 0)
896 return r;
897 }
898
899 if (test_run_flags == 0) {
900 if (MANAGER_IS_SYSTEM(m))
901 r = mkdir_label("/run/systemd/units", 0755);
902 else {
903 _cleanup_free_ char *units_path = NULL;
904 r = xdg_user_runtime_dir(&units_path, "/systemd/units");
905 if (r < 0)
906 return r;
907 r = mkdir_p_label(units_path, 0755);
908 }
909
910 if (r < 0 && r != -EEXIST)
911 return r;
912 }
913
914 m->taint_usr =
915 !in_initrd() &&
916 dir_is_empty("/usr") > 0;
917
918 /* Note that we do not set up the notify fd here. We do that after deserialization,
919 * since they might have gotten serialized across the reexec. */
920
921 *_m = TAKE_PTR(m);
922
923 return 0;
924 }
925
926 static int manager_setup_notify(Manager *m) {
927 int r;
928
929 if (MANAGER_IS_TEST_RUN(m))
930 return 0;
931
932 if (m->notify_fd < 0) {
933 _cleanup_close_ int fd = -1;
934 union sockaddr_union sa;
935 socklen_t sa_len;
936
937 /* First free all secondary fields */
938 m->notify_socket = mfree(m->notify_socket);
939 m->notify_event_source = sd_event_source_unref(m->notify_event_source);
940
941 fd = socket(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC|SOCK_NONBLOCK, 0);
942 if (fd < 0)
943 return log_error_errno(errno, "Failed to allocate notification socket: %m");
944
945 fd_inc_rcvbuf(fd, NOTIFY_RCVBUF_SIZE);
946
947 m->notify_socket = path_join(m->prefix[EXEC_DIRECTORY_RUNTIME], "systemd/notify");
948 if (!m->notify_socket)
949 return log_oom();
950
951 r = sockaddr_un_set_path(&sa.un, m->notify_socket);
952 if (r < 0)
953 return log_error_errno(r, "Notify socket '%s' not valid for AF_UNIX socket address, refusing.",
954 m->notify_socket);
955 sa_len = r;
956
957 (void) mkdir_parents_label(m->notify_socket, 0755);
958 (void) sockaddr_un_unlink(&sa.un);
959
960 r = mac_selinux_bind(fd, &sa.sa, sa_len);
961 if (r < 0)
962 return log_error_errno(r, "bind(%s) failed: %m", m->notify_socket);
963
964 r = setsockopt_int(fd, SOL_SOCKET, SO_PASSCRED, true);
965 if (r < 0)
966 return log_error_errno(r, "SO_PASSCRED failed: %m");
967
968 m->notify_fd = TAKE_FD(fd);
969
970 log_debug("Using notification socket %s", m->notify_socket);
971 }
972
973 if (!m->notify_event_source) {
974 r = sd_event_add_io(m->event, &m->notify_event_source, m->notify_fd, EPOLLIN, manager_dispatch_notify_fd, m);
975 if (r < 0)
976 return log_error_errno(r, "Failed to allocate notify event source: %m");
977
978 /* Process notification messages a bit earlier than SIGCHLD, so that we can still identify to which
979 * service an exit message belongs. */
980 r = sd_event_source_set_priority(m->notify_event_source, SD_EVENT_PRIORITY_NORMAL-8);
981 if (r < 0)
982 return log_error_errno(r, "Failed to set priority of notify event source: %m");
983
984 (void) sd_event_source_set_description(m->notify_event_source, "manager-notify");
985 }
986
987 return 0;
988 }
989
990 static int manager_setup_cgroups_agent(Manager *m) {
991
992 static const union sockaddr_union sa = {
993 .un.sun_family = AF_UNIX,
994 .un.sun_path = "/run/systemd/cgroups-agent",
995 };
996 int r;
997
998 /* This creates a listening socket we receive cgroups agent messages on. We do not use D-Bus for delivering
999 * these messages from the cgroups agent binary to PID 1, as the cgroups agent binary is very short-living, and
1000 * each instance of it needs a new D-Bus connection. Since D-Bus connections are SOCK_STREAM/AF_UNIX, on
1001 * overloaded systems the backlog of the D-Bus socket becomes relevant, as not more than the configured number
1002 * of D-Bus connections may be queued until the kernel will start dropping further incoming connections,
1003 * possibly resulting in lost cgroups agent messages. To avoid this, we'll use a private SOCK_DGRAM/AF_UNIX
1004 * socket, where no backlog is relevant as communication may take place without an actual connect() cycle, and
1005 * we thus won't lose messages.
1006 *
1007 * Note that PID 1 will forward the agent message to system bus, so that the user systemd instance may listen
1008 * to it. The system instance hence listens on this special socket, but the user instances listen on the system
1009 * bus for these messages. */
1010
1011 if (MANAGER_IS_TEST_RUN(m))
1012 return 0;
1013
1014 if (!MANAGER_IS_SYSTEM(m))
1015 return 0;
1016
1017 r = cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER);
1018 if (r < 0)
1019 return log_error_errno(r, "Failed to determine whether unified cgroups hierarchy is used: %m");
1020 if (r > 0) /* We don't need this anymore on the unified hierarchy */
1021 return 0;
1022
1023 if (m->cgroups_agent_fd < 0) {
1024 _cleanup_close_ int fd = -1;
1025
1026 /* First free all secondary fields */
1027 m->cgroups_agent_event_source = sd_event_source_unref(m->cgroups_agent_event_source);
1028
1029 fd = socket(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC|SOCK_NONBLOCK, 0);
1030 if (fd < 0)
1031 return log_error_errno(errno, "Failed to allocate cgroups agent socket: %m");
1032
1033 fd_inc_rcvbuf(fd, CGROUPS_AGENT_RCVBUF_SIZE);
1034
1035 (void) sockaddr_un_unlink(&sa.un);
1036
1037 /* Only allow root to connect to this socket */
1038 RUN_WITH_UMASK(0077)
1039 r = bind(fd, &sa.sa, SOCKADDR_UN_LEN(sa.un));
1040 if (r < 0)
1041 return log_error_errno(errno, "bind(%s) failed: %m", sa.un.sun_path);
1042
1043 m->cgroups_agent_fd = TAKE_FD(fd);
1044 }
1045
1046 if (!m->cgroups_agent_event_source) {
1047 r = sd_event_add_io(m->event, &m->cgroups_agent_event_source, m->cgroups_agent_fd, EPOLLIN, manager_dispatch_cgroups_agent_fd, m);
1048 if (r < 0)
1049 return log_error_errno(r, "Failed to allocate cgroups agent event source: %m");
1050
1051 /* Process cgroups notifications early. Note that when the agent notification is received
1052 * we'll just enqueue the unit in the cgroup empty queue, hence pick a high priority than
1053 * that. Also see handling of cgroup inotify for the unified cgroup stuff. */
1054 r = sd_event_source_set_priority(m->cgroups_agent_event_source, SD_EVENT_PRIORITY_NORMAL-9);
1055 if (r < 0)
1056 return log_error_errno(r, "Failed to set priority of cgroups agent event source: %m");
1057
1058 (void) sd_event_source_set_description(m->cgroups_agent_event_source, "manager-cgroups-agent");
1059 }
1060
1061 return 0;
1062 }
1063
1064 static int manager_setup_user_lookup_fd(Manager *m) {
1065 int r;
1066
1067 assert(m);
1068
1069 /* Set up the socket pair used for passing UID/GID resolution results from forked off processes to PID
1070 * 1. Background: we can't do name lookups (NSS) from PID 1, since it might involve IPC and thus activation,
1071 * and we might hence deadlock on ourselves. Hence we do all user/group lookups asynchronously from the forked
1072 * off processes right before executing the binaries to start. In order to be able to clean up any IPC objects
1073 * created by a unit (see RemoveIPC=) we need to know in PID 1 the used UID/GID of the executed processes,
1074 * hence we establish this communication channel so that forked off processes can pass their UID/GID
1075 * information back to PID 1. The forked off processes send their resolved UID/GID to PID 1 in a simple
1076 * datagram, along with their unit name, so that we can share one communication socket pair among all units for
1077 * this purpose.
1078 *
1079 * You might wonder why we need a communication channel for this that is independent of the usual notification
1080 * socket scheme (i.e. $NOTIFY_SOCKET). The primary difference is about trust: data sent via the $NOTIFY_SOCKET
1081 * channel is only accepted if it originates from the right unit and if reception was enabled for it. The user
1082 * lookup socket OTOH is only accessible by PID 1 and its children until they exec(), and always available.
1083 *
1084 * Note that this function is called under two circumstances: when we first initialize (in which case we
1085 * allocate both the socket pair and the event source to listen on it), and when we deserialize after a reload
1086 * (in which case the socket pair already exists but we still need to allocate the event source for it). */
1087
1088 if (m->user_lookup_fds[0] < 0) {
1089
1090 /* Free all secondary fields */
1091 safe_close_pair(m->user_lookup_fds);
1092 m->user_lookup_event_source = sd_event_source_unref(m->user_lookup_event_source);
1093
1094 if (socketpair(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC, 0, m->user_lookup_fds) < 0)
1095 return log_error_errno(errno, "Failed to allocate user lookup socket: %m");
1096
1097 (void) fd_inc_rcvbuf(m->user_lookup_fds[0], NOTIFY_RCVBUF_SIZE);
1098 }
1099
1100 if (!m->user_lookup_event_source) {
1101 r = sd_event_add_io(m->event, &m->user_lookup_event_source, m->user_lookup_fds[0], EPOLLIN, manager_dispatch_user_lookup_fd, m);
1102 if (r < 0)
1103 return log_error_errno(errno, "Failed to allocate user lookup event source: %m");
1104
1105 /* Process even earlier than the notify event source, so that we always know first about valid UID/GID
1106 * resolutions */
1107 r = sd_event_source_set_priority(m->user_lookup_event_source, SD_EVENT_PRIORITY_NORMAL-11);
1108 if (r < 0)
1109 return log_error_errno(errno, "Failed to set priority of user lookup event source: %m");
1110
1111 (void) sd_event_source_set_description(m->user_lookup_event_source, "user-lookup");
1112 }
1113
1114 return 0;
1115 }
1116
1117 static unsigned manager_dispatch_cleanup_queue(Manager *m) {
1118 Unit *u;
1119 unsigned n = 0;
1120
1121 assert(m);
1122
1123 while ((u = m->cleanup_queue)) {
1124 assert(u->in_cleanup_queue);
1125
1126 unit_free(u);
1127 n++;
1128 }
1129
1130 return n;
1131 }
1132
1133 enum {
1134 GC_OFFSET_IN_PATH, /* This one is on the path we were traveling */
1135 GC_OFFSET_UNSURE, /* No clue */
1136 GC_OFFSET_GOOD, /* We still need this unit */
1137 GC_OFFSET_BAD, /* We don't need this unit anymore */
1138 _GC_OFFSET_MAX
1139 };
1140
1141 static void unit_gc_mark_good(Unit *u, unsigned gc_marker) {
1142 Unit *other;
1143 void *v;
1144
1145 u->gc_marker = gc_marker + GC_OFFSET_GOOD;
1146
1147 /* Recursively mark referenced units as GOOD as well */
1148 HASHMAP_FOREACH_KEY(v, other, u->dependencies[UNIT_REFERENCES])
1149 if (other->gc_marker == gc_marker + GC_OFFSET_UNSURE)
1150 unit_gc_mark_good(other, gc_marker);
1151 }
1152
1153 static void unit_gc_sweep(Unit *u, unsigned gc_marker) {
1154 Unit *other;
1155 bool is_bad;
1156 void *v;
1157
1158 assert(u);
1159
1160 if (IN_SET(u->gc_marker - gc_marker,
1161 GC_OFFSET_GOOD, GC_OFFSET_BAD, GC_OFFSET_UNSURE, GC_OFFSET_IN_PATH))
1162 return;
1163
1164 if (u->in_cleanup_queue)
1165 goto bad;
1166
1167 if (!unit_may_gc(u))
1168 goto good;
1169
1170 u->gc_marker = gc_marker + GC_OFFSET_IN_PATH;
1171
1172 is_bad = true;
1173
1174 HASHMAP_FOREACH_KEY(v, other, u->dependencies[UNIT_REFERENCED_BY]) {
1175 unit_gc_sweep(other, gc_marker);
1176
1177 if (other->gc_marker == gc_marker + GC_OFFSET_GOOD)
1178 goto good;
1179
1180 if (other->gc_marker != gc_marker + GC_OFFSET_BAD)
1181 is_bad = false;
1182 }
1183
1184 const UnitRef *ref;
1185 LIST_FOREACH(refs_by_target, ref, u->refs_by_target) {
1186 unit_gc_sweep(ref->source, gc_marker);
1187
1188 if (ref->source->gc_marker == gc_marker + GC_OFFSET_GOOD)
1189 goto good;
1190
1191 if (ref->source->gc_marker != gc_marker + GC_OFFSET_BAD)
1192 is_bad = false;
1193 }
1194
1195 if (is_bad)
1196 goto bad;
1197
1198 /* We were unable to find anything out about this entry, so
1199 * let's investigate it later */
1200 u->gc_marker = gc_marker + GC_OFFSET_UNSURE;
1201 unit_add_to_gc_queue(u);
1202 return;
1203
1204 bad:
1205 /* We definitely know that this one is not useful anymore, so
1206 * let's mark it for deletion */
1207 u->gc_marker = gc_marker + GC_OFFSET_BAD;
1208 unit_add_to_cleanup_queue(u);
1209 return;
1210
1211 good:
1212 unit_gc_mark_good(u, gc_marker);
1213 }
1214
1215 static unsigned manager_dispatch_gc_unit_queue(Manager *m) {
1216 unsigned n = 0, gc_marker;
1217 Unit *u;
1218
1219 assert(m);
1220
1221 /* log_debug("Running GC..."); */
1222
1223 m->gc_marker += _GC_OFFSET_MAX;
1224 if (m->gc_marker + _GC_OFFSET_MAX <= _GC_OFFSET_MAX)
1225 m->gc_marker = 1;
1226
1227 gc_marker = m->gc_marker;
1228
1229 while ((u = m->gc_unit_queue)) {
1230 assert(u->in_gc_queue);
1231
1232 unit_gc_sweep(u, gc_marker);
1233
1234 LIST_REMOVE(gc_queue, m->gc_unit_queue, u);
1235 u->in_gc_queue = false;
1236
1237 n++;
1238
1239 if (IN_SET(u->gc_marker - gc_marker,
1240 GC_OFFSET_BAD, GC_OFFSET_UNSURE)) {
1241 if (u->id)
1242 log_unit_debug(u, "Collecting.");
1243 u->gc_marker = gc_marker + GC_OFFSET_BAD;
1244 unit_add_to_cleanup_queue(u);
1245 }
1246 }
1247
1248 return n;
1249 }
1250
1251 static unsigned manager_dispatch_gc_job_queue(Manager *m) {
1252 unsigned n = 0;
1253 Job *j;
1254
1255 assert(m);
1256
1257 while ((j = m->gc_job_queue)) {
1258 assert(j->in_gc_queue);
1259
1260 LIST_REMOVE(gc_queue, m->gc_job_queue, j);
1261 j->in_gc_queue = false;
1262
1263 n++;
1264
1265 if (!job_may_gc(j))
1266 continue;
1267
1268 log_unit_debug(j->unit, "Collecting job.");
1269 (void) job_finish_and_invalidate(j, JOB_COLLECTED, false, false);
1270 }
1271
1272 return n;
1273 }
1274
1275 static unsigned manager_dispatch_stop_when_unneeded_queue(Manager *m) {
1276 unsigned n = 0;
1277 Unit *u;
1278 int r;
1279
1280 assert(m);
1281
1282 while ((u = m->stop_when_unneeded_queue)) {
1283 _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL;
1284 assert(m->stop_when_unneeded_queue);
1285
1286 assert(u->in_stop_when_unneeded_queue);
1287 LIST_REMOVE(stop_when_unneeded_queue, m->stop_when_unneeded_queue, u);
1288 u->in_stop_when_unneeded_queue = false;
1289
1290 n++;
1291
1292 if (!unit_is_unneeded(u))
1293 continue;
1294
1295 log_unit_debug(u, "Unit is not needed anymore.");
1296
1297 /* If stopping a unit fails continuously we might enter a stop loop here, hence stop acting on the
1298 * service being unnecessary after a while. */
1299
1300 if (!ratelimit_below(&u->auto_stop_ratelimit)) {
1301 log_unit_warning(u, "Unit not needed anymore, but not stopping since we tried this too often recently.");
1302 continue;
1303 }
1304
1305 /* Ok, nobody needs us anymore. Sniff. Then let's commit suicide */
1306 r = manager_add_job(u->manager, JOB_STOP, u, JOB_FAIL, NULL, &error, NULL);
1307 if (r < 0)
1308 log_unit_warning_errno(u, r, "Failed to enqueue stop job, ignoring: %s", bus_error_message(&error, r));
1309 }
1310
1311 return n;
1312 }
1313
1314 static void manager_clear_jobs_and_units(Manager *m) {
1315 Unit *u;
1316
1317 assert(m);
1318
1319 while ((u = hashmap_first(m->units)))
1320 unit_free(u);
1321
1322 manager_dispatch_cleanup_queue(m);
1323
1324 assert(!m->load_queue);
1325 assert(prioq_isempty(m->run_queue));
1326 assert(!m->dbus_unit_queue);
1327 assert(!m->dbus_job_queue);
1328 assert(!m->cleanup_queue);
1329 assert(!m->gc_unit_queue);
1330 assert(!m->gc_job_queue);
1331 assert(!m->stop_when_unneeded_queue);
1332
1333 assert(hashmap_isempty(m->jobs));
1334 assert(hashmap_isempty(m->units));
1335
1336 m->n_on_console = 0;
1337 m->n_running_jobs = 0;
1338 m->n_installed_jobs = 0;
1339 m->n_failed_jobs = 0;
1340 }
1341
1342 Manager* manager_free(Manager *m) {
1343 if (!m)
1344 return NULL;
1345
1346 manager_clear_jobs_and_units(m);
1347
1348 for (UnitType c = 0; c < _UNIT_TYPE_MAX; c++)
1349 if (unit_vtable[c]->shutdown)
1350 unit_vtable[c]->shutdown(m);
1351
1352 /* Keep the cgroup hierarchy in place except when we know we are going down for good */
1353 manager_shutdown_cgroup(m, IN_SET(m->objective, MANAGER_EXIT, MANAGER_REBOOT, MANAGER_POWEROFF, MANAGER_HALT, MANAGER_KEXEC));
1354
1355 lookup_paths_flush_generator(&m->lookup_paths);
1356
1357 bus_done(m);
1358 manager_varlink_done(m);
1359
1360 exec_runtime_vacuum(m);
1361 hashmap_free(m->exec_runtime_by_id);
1362
1363 dynamic_user_vacuum(m, false);
1364 hashmap_free(m->dynamic_users);
1365
1366 hashmap_free(m->units);
1367 hashmap_free(m->units_by_invocation_id);
1368 hashmap_free(m->jobs);
1369 hashmap_free(m->watch_pids);
1370 hashmap_free(m->watch_bus);
1371
1372 prioq_free(m->run_queue);
1373
1374 set_free(m->startup_units);
1375 set_free(m->failed_units);
1376
1377 sd_event_source_unref(m->signal_event_source);
1378 sd_event_source_unref(m->sigchld_event_source);
1379 sd_event_source_unref(m->notify_event_source);
1380 sd_event_source_unref(m->cgroups_agent_event_source);
1381 sd_event_source_unref(m->time_change_event_source);
1382 sd_event_source_unref(m->timezone_change_event_source);
1383 sd_event_source_unref(m->jobs_in_progress_event_source);
1384 sd_event_source_unref(m->run_queue_event_source);
1385 sd_event_source_unref(m->user_lookup_event_source);
1386
1387 safe_close(m->signal_fd);
1388 safe_close(m->notify_fd);
1389 safe_close(m->cgroups_agent_fd);
1390 safe_close(m->time_change_fd);
1391 safe_close_pair(m->user_lookup_fds);
1392
1393 manager_close_ask_password(m);
1394
1395 manager_close_idle_pipe(m);
1396
1397 sd_event_unref(m->event);
1398
1399 free(m->notify_socket);
1400
1401 lookup_paths_free(&m->lookup_paths);
1402 strv_free(m->transient_environment);
1403 strv_free(m->client_environment);
1404
1405 hashmap_free(m->cgroup_unit);
1406 manager_free_unit_name_maps(m);
1407
1408 free(m->switch_root);
1409 free(m->switch_root_init);
1410
1411 rlimit_free_all(m->rlimit);
1412
1413 assert(hashmap_isempty(m->units_requiring_mounts_for));
1414 hashmap_free(m->units_requiring_mounts_for);
1415
1416 hashmap_free(m->uid_refs);
1417 hashmap_free(m->gid_refs);
1418
1419 for (ExecDirectoryType dt = 0; dt < _EXEC_DIRECTORY_TYPE_MAX; dt++)
1420 m->prefix[dt] = mfree(m->prefix[dt]);
1421 free(m->received_credentials);
1422
1423 return mfree(m);
1424 }
1425
1426 static void manager_enumerate_perpetual(Manager *m) {
1427 assert(m);
1428
1429 if (m->test_run_flags == MANAGER_TEST_RUN_MINIMAL)
1430 return;
1431
1432 /* Let's ask every type to load all units from disk/kernel that it might know */
1433 for (UnitType c = 0; c < _UNIT_TYPE_MAX; c++) {
1434 if (!unit_type_supported(c)) {
1435 log_debug("Unit type .%s is not supported on this system.", unit_type_to_string(c));
1436 continue;
1437 }
1438
1439 if (unit_vtable[c]->enumerate_perpetual)
1440 unit_vtable[c]->enumerate_perpetual(m);
1441 }
1442 }
1443
1444 static void manager_enumerate(Manager *m) {
1445 assert(m);
1446
1447 if (m->test_run_flags == MANAGER_TEST_RUN_MINIMAL)
1448 return;
1449
1450 /* Let's ask every type to load all units from disk/kernel that it might know */
1451 for (UnitType c = 0; c < _UNIT_TYPE_MAX; c++) {
1452 if (!unit_type_supported(c)) {
1453 log_debug("Unit type .%s is not supported on this system.", unit_type_to_string(c));
1454 continue;
1455 }
1456
1457 if (unit_vtable[c]->enumerate)
1458 unit_vtable[c]->enumerate(m);
1459 }
1460
1461 manager_dispatch_load_queue(m);
1462 }
1463
1464 static void manager_coldplug(Manager *m) {
1465 Unit *u;
1466 char *k;
1467 int r;
1468
1469 assert(m);
1470
1471 log_debug("Invoking unit coldplug() handlers…");
1472
1473 /* Let's place the units back into their deserialized state */
1474 HASHMAP_FOREACH_KEY(u, k, m->units) {
1475
1476 /* ignore aliases */
1477 if (u->id != k)
1478 continue;
1479
1480 r = unit_coldplug(u);
1481 if (r < 0)
1482 log_warning_errno(r, "We couldn't coldplug %s, proceeding anyway: %m", u->id);
1483 }
1484 }
1485
1486 static void manager_catchup(Manager *m) {
1487 Unit *u;
1488 char *k;
1489
1490 assert(m);
1491
1492 log_debug("Invoking unit catchup() handlers…");
1493
1494 /* Let's catch up on any state changes that happened while we were reloading/reexecing */
1495 HASHMAP_FOREACH_KEY(u, k, m->units) {
1496
1497 /* ignore aliases */
1498 if (u->id != k)
1499 continue;
1500
1501 unit_catchup(u);
1502 }
1503 }
1504
1505 static void manager_distribute_fds(Manager *m, FDSet *fds) {
1506 Unit *u;
1507
1508 assert(m);
1509
1510 HASHMAP_FOREACH(u, m->units) {
1511
1512 if (fdset_size(fds) <= 0)
1513 break;
1514
1515 if (!UNIT_VTABLE(u)->distribute_fds)
1516 continue;
1517
1518 UNIT_VTABLE(u)->distribute_fds(u, fds);
1519 }
1520 }
1521
1522 static bool manager_dbus_is_running(Manager *m, bool deserialized) {
1523 Unit *u;
1524
1525 assert(m);
1526
1527 /* This checks whether the dbus instance we are supposed to expose our APIs on is up. We check both the socket
1528 * and the service unit. If the 'deserialized' parameter is true we'll check the deserialized state of the unit
1529 * rather than the current one. */
1530
1531 if (MANAGER_IS_TEST_RUN(m))
1532 return false;
1533
1534 u = manager_get_unit(m, SPECIAL_DBUS_SOCKET);
1535 if (!u)
1536 return false;
1537 if ((deserialized ? SOCKET(u)->deserialized_state : SOCKET(u)->state) != SOCKET_RUNNING)
1538 return false;
1539
1540 u = manager_get_unit(m, SPECIAL_DBUS_SERVICE);
1541 if (!u)
1542 return false;
1543 if (!IN_SET((deserialized ? SERVICE(u)->deserialized_state : SERVICE(u)->state), SERVICE_RUNNING, SERVICE_RELOAD))
1544 return false;
1545
1546 return true;
1547 }
1548
1549 static void manager_setup_bus(Manager *m) {
1550 assert(m);
1551
1552 /* Let's set up our private bus connection now, unconditionally */
1553 (void) bus_init_private(m);
1554
1555 /* If we are in --user mode also connect to the system bus now */
1556 if (MANAGER_IS_USER(m))
1557 (void) bus_init_system(m);
1558
1559 /* Let's connect to the bus now, but only if the unit is supposed to be up */
1560 if (manager_dbus_is_running(m, MANAGER_IS_RELOADING(m))) {
1561 (void) bus_init_api(m);
1562
1563 if (MANAGER_IS_SYSTEM(m))
1564 (void) bus_init_system(m);
1565 }
1566 }
1567
1568 static void manager_preset_all(Manager *m) {
1569 int r;
1570
1571 assert(m);
1572
1573 if (m->first_boot <= 0)
1574 return;
1575
1576 if (!MANAGER_IS_SYSTEM(m))
1577 return;
1578
1579 if (MANAGER_IS_TEST_RUN(m))
1580 return;
1581
1582 /* If this is the first boot, and we are in the host system, then preset everything */
1583 r = unit_file_preset_all(UNIT_FILE_SYSTEM, 0, NULL, UNIT_FILE_PRESET_ENABLE_ONLY, NULL, 0);
1584 if (r < 0)
1585 log_full_errno(r == -EEXIST ? LOG_NOTICE : LOG_WARNING, r,
1586 "Failed to populate /etc with preset unit settings, ignoring: %m");
1587 else
1588 log_info("Populated /etc with preset unit settings.");
1589 }
1590
1591 static void manager_ready(Manager *m) {
1592 assert(m);
1593
1594 /* After having loaded everything, do the final round of catching up with what might have changed */
1595
1596 m->objective = MANAGER_OK; /* Tell everyone we are up now */
1597
1598 /* It might be safe to log to the journal now and connect to dbus */
1599 manager_recheck_journal(m);
1600 manager_recheck_dbus(m);
1601
1602 /* Let's finally catch up with any changes that took place while we were reloading/reexecing */
1603 manager_catchup(m);
1604
1605 m->honor_device_enumeration = true;
1606 }
1607
1608 static Manager* manager_reloading_start(Manager *m) {
1609 m->n_reloading++;
1610 return m;
1611 }
1612 static void manager_reloading_stopp(Manager **m) {
1613 if (*m) {
1614 assert((*m)->n_reloading > 0);
1615 (*m)->n_reloading--;
1616 }
1617 }
1618
1619 int manager_startup(Manager *m, FILE *serialization, FDSet *fds) {
1620 int r;
1621
1622 assert(m);
1623
1624 /* If we are running in test mode, we still want to run the generators,
1625 * but we should not touch the real generator directories. */
1626 r = lookup_paths_init(&m->lookup_paths, m->unit_file_scope,
1627 MANAGER_IS_TEST_RUN(m) ? LOOKUP_PATHS_TEMPORARY_GENERATED : 0,
1628 NULL);
1629 if (r < 0)
1630 return log_error_errno(r, "Failed to initialize path lookup table: %m");
1631
1632 dual_timestamp_get(m->timestamps + manager_timestamp_initrd_mangle(MANAGER_TIMESTAMP_GENERATORS_START));
1633 r = manager_run_environment_generators(m);
1634 if (r >= 0)
1635 r = manager_run_generators(m);
1636 dual_timestamp_get(m->timestamps + manager_timestamp_initrd_mangle(MANAGER_TIMESTAMP_GENERATORS_FINISH));
1637 if (r < 0)
1638 return r;
1639
1640 manager_preset_all(m);
1641
1642 lookup_paths_log(&m->lookup_paths);
1643
1644 {
1645 /* This block is (optionally) done with the reloading counter bumped */
1646 _cleanup_(manager_reloading_stopp) Manager *reloading = NULL;
1647
1648 /* If we will deserialize make sure that during enumeration this is already known, so we increase the
1649 * counter here already */
1650 if (serialization)
1651 reloading = manager_reloading_start(m);
1652
1653 /* First, enumerate what we can from all config files */
1654 dual_timestamp_get(m->timestamps + manager_timestamp_initrd_mangle(MANAGER_TIMESTAMP_UNITS_LOAD_START));
1655 manager_enumerate_perpetual(m);
1656 manager_enumerate(m);
1657 dual_timestamp_get(m->timestamps + manager_timestamp_initrd_mangle(MANAGER_TIMESTAMP_UNITS_LOAD_FINISH));
1658
1659 /* Second, deserialize if there is something to deserialize */
1660 if (serialization) {
1661 r = manager_deserialize(m, serialization, fds);
1662 if (r < 0)
1663 return log_error_errno(r, "Deserialization failed: %m");
1664 }
1665
1666 /* Any fds left? Find some unit which wants them. This is useful to allow container managers to pass
1667 * some file descriptors to us pre-initialized. This enables socket-based activation of entire
1668 * containers. */
1669 manager_distribute_fds(m, fds);
1670
1671 /* We might have deserialized the notify fd, but if we didn't then let's create the bus now */
1672 r = manager_setup_notify(m);
1673 if (r < 0)
1674 /* No sense to continue without notifications, our children would fail anyway. */
1675 return r;
1676
1677 r = manager_setup_cgroups_agent(m);
1678 if (r < 0)
1679 /* Likewise, no sense to continue without empty cgroup notifications. */
1680 return r;
1681
1682 r = manager_setup_user_lookup_fd(m);
1683 if (r < 0)
1684 /* This shouldn't fail, except if things are really broken. */
1685 return r;
1686
1687 /* Connect to the bus if we are good for it */
1688 manager_setup_bus(m);
1689
1690 /* Now that we are connected to all possible buses, let's deserialize who is tracking us. */
1691 r = bus_track_coldplug(m, &m->subscribed, false, m->deserialized_subscribed);
1692 if (r < 0)
1693 log_warning_errno(r, "Failed to deserialized tracked clients, ignoring: %m");
1694 m->deserialized_subscribed = strv_free(m->deserialized_subscribed);
1695
1696 r = manager_varlink_init(m);
1697 if (r < 0)
1698 log_warning_errno(r, "Failed to set up Varlink server, ignoring: %m");
1699
1700 /* Third, fire things up! */
1701 manager_coldplug(m);
1702
1703 /* Clean up runtime objects */
1704 manager_vacuum(m);
1705
1706 if (serialization)
1707 /* Let's wait for the UnitNew/JobNew messages being sent, before we notify that the
1708 * reload is finished */
1709 m->send_reloading_done = true;
1710 }
1711
1712 manager_ready(m);
1713
1714 return 0;
1715 }
1716
1717 int manager_add_job(
1718 Manager *m,
1719 JobType type,
1720 Unit *unit,
1721 JobMode mode,
1722 Set *affected_jobs,
1723 sd_bus_error *error,
1724 Job **ret) {
1725
1726 Transaction *tr;
1727 int r;
1728
1729 assert(m);
1730 assert(type < _JOB_TYPE_MAX);
1731 assert(unit);
1732 assert(mode < _JOB_MODE_MAX);
1733
1734 if (mode == JOB_ISOLATE && type != JOB_START)
1735 return sd_bus_error_setf(error, SD_BUS_ERROR_INVALID_ARGS, "Isolate is only valid for start.");
1736
1737 if (mode == JOB_ISOLATE && !unit->allow_isolate)
1738 return sd_bus_error_setf(error, BUS_ERROR_NO_ISOLATION, "Operation refused, unit may not be isolated.");
1739
1740 if (mode == JOB_TRIGGERING && type != JOB_STOP)
1741 return sd_bus_error_setf(error, SD_BUS_ERROR_INVALID_ARGS, "--job-mode=triggering is only valid for stop.");
1742
1743 log_unit_debug(unit, "Trying to enqueue job %s/%s/%s", unit->id, job_type_to_string(type), job_mode_to_string(mode));
1744
1745 type = job_type_collapse(type, unit);
1746
1747 tr = transaction_new(mode == JOB_REPLACE_IRREVERSIBLY);
1748 if (!tr)
1749 return -ENOMEM;
1750
1751 r = transaction_add_job_and_dependencies(tr, type, unit, NULL, true, false,
1752 IN_SET(mode, JOB_IGNORE_DEPENDENCIES, JOB_IGNORE_REQUIREMENTS),
1753 mode == JOB_IGNORE_DEPENDENCIES, error);
1754 if (r < 0)
1755 goto tr_abort;
1756
1757 if (mode == JOB_ISOLATE) {
1758 r = transaction_add_isolate_jobs(tr, m);
1759 if (r < 0)
1760 goto tr_abort;
1761 }
1762
1763 if (mode == JOB_TRIGGERING) {
1764 r = transaction_add_triggering_jobs(tr, unit);
1765 if (r < 0)
1766 goto tr_abort;
1767 }
1768
1769 r = transaction_activate(tr, m, mode, affected_jobs, error);
1770 if (r < 0)
1771 goto tr_abort;
1772
1773 log_unit_debug(unit,
1774 "Enqueued job %s/%s as %u", unit->id,
1775 job_type_to_string(type), (unsigned) tr->anchor_job->id);
1776
1777 if (ret)
1778 *ret = tr->anchor_job;
1779
1780 transaction_free(tr);
1781 return 0;
1782
1783 tr_abort:
1784 transaction_abort(tr);
1785 transaction_free(tr);
1786 return r;
1787 }
1788
1789 int manager_add_job_by_name(Manager *m, JobType type, const char *name, JobMode mode, Set *affected_jobs, sd_bus_error *e, Job **ret) {
1790 Unit *unit = NULL; /* just to appease gcc, initialization is not really necessary */
1791 int r;
1792
1793 assert(m);
1794 assert(type < _JOB_TYPE_MAX);
1795 assert(name);
1796 assert(mode < _JOB_MODE_MAX);
1797
1798 r = manager_load_unit(m, name, NULL, NULL, &unit);
1799 if (r < 0)
1800 return r;
1801 assert(unit);
1802
1803 return manager_add_job(m, type, unit, mode, affected_jobs, e, ret);
1804 }
1805
1806 int manager_add_job_by_name_and_warn(Manager *m, JobType type, const char *name, JobMode mode, Set *affected_jobs, Job **ret) {
1807 _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL;
1808 int r;
1809
1810 assert(m);
1811 assert(type < _JOB_TYPE_MAX);
1812 assert(name);
1813 assert(mode < _JOB_MODE_MAX);
1814
1815 r = manager_add_job_by_name(m, type, name, mode, affected_jobs, &error, ret);
1816 if (r < 0)
1817 return log_warning_errno(r, "Failed to enqueue %s job for %s: %s", job_mode_to_string(mode), name, bus_error_message(&error, r));
1818
1819 return r;
1820 }
1821
1822 int manager_propagate_reload(Manager *m, Unit *unit, JobMode mode, sd_bus_error *e) {
1823 int r;
1824 Transaction *tr;
1825
1826 assert(m);
1827 assert(unit);
1828 assert(mode < _JOB_MODE_MAX);
1829 assert(mode != JOB_ISOLATE); /* Isolate is only valid for start */
1830
1831 tr = transaction_new(mode == JOB_REPLACE_IRREVERSIBLY);
1832 if (!tr)
1833 return -ENOMEM;
1834
1835 /* We need an anchor job */
1836 r = transaction_add_job_and_dependencies(tr, JOB_NOP, unit, NULL, false, false, true, true, e);
1837 if (r < 0)
1838 goto tr_abort;
1839
1840 /* Failure in adding individual dependencies is ignored, so this always succeeds. */
1841 transaction_add_propagate_reload_jobs(tr, unit, tr->anchor_job, mode == JOB_IGNORE_DEPENDENCIES, e);
1842
1843 r = transaction_activate(tr, m, mode, NULL, e);
1844 if (r < 0)
1845 goto tr_abort;
1846
1847 transaction_free(tr);
1848 return 0;
1849
1850 tr_abort:
1851 transaction_abort(tr);
1852 transaction_free(tr);
1853 return r;
1854 }
1855
1856 Job *manager_get_job(Manager *m, uint32_t id) {
1857 assert(m);
1858
1859 return hashmap_get(m->jobs, UINT32_TO_PTR(id));
1860 }
1861
1862 Unit *manager_get_unit(Manager *m, const char *name) {
1863 assert(m);
1864 assert(name);
1865
1866 return hashmap_get(m->units, name);
1867 }
1868
1869 static int manager_dispatch_target_deps_queue(Manager *m) {
1870 Unit *u;
1871 int r = 0;
1872
1873 static const UnitDependency deps[] = {
1874 UNIT_REQUIRED_BY,
1875 UNIT_REQUISITE_OF,
1876 UNIT_WANTED_BY,
1877 UNIT_BOUND_BY
1878 };
1879
1880 assert(m);
1881
1882 while ((u = m->target_deps_queue)) {
1883 assert(u->in_target_deps_queue);
1884
1885 LIST_REMOVE(target_deps_queue, u->manager->target_deps_queue, u);
1886 u->in_target_deps_queue = false;
1887
1888 for (size_t k = 0; k < ELEMENTSOF(deps); k++) {
1889 Unit *target;
1890 void *v;
1891
1892 HASHMAP_FOREACH_KEY(v, target, u->dependencies[deps[k]]) {
1893 r = unit_add_default_target_dependency(u, target);
1894 if (r < 0)
1895 return r;
1896 }
1897 }
1898 }
1899
1900 return r;
1901 }
1902
1903 unsigned manager_dispatch_load_queue(Manager *m) {
1904 Unit *u;
1905 unsigned n = 0;
1906
1907 assert(m);
1908
1909 /* Make sure we are not run recursively */
1910 if (m->dispatching_load_queue)
1911 return 0;
1912
1913 m->dispatching_load_queue = true;
1914
1915 /* Dispatches the load queue. Takes a unit from the queue and
1916 * tries to load its data until the queue is empty */
1917
1918 while ((u = m->load_queue)) {
1919 assert(u->in_load_queue);
1920
1921 unit_load(u);
1922 n++;
1923 }
1924
1925 m->dispatching_load_queue = false;
1926
1927 /* Dispatch the units waiting for their target dependencies to be added now, as all targets that we know about
1928 * should be loaded and have aliases resolved */
1929 (void) manager_dispatch_target_deps_queue(m);
1930
1931 return n;
1932 }
1933
1934 bool manager_unit_cache_should_retry_load(Unit *u) {
1935 assert(u);
1936
1937 /* Automatic reloading from disk only applies to units which were not found sometime in the past, and
1938 * the not-found stub is kept pinned in the unit graph by dependencies. For units that were
1939 * previously loaded, we don't do automatic reloading, and daemon-reload is necessary to update. */
1940 if (u->load_state != UNIT_NOT_FOUND)
1941 return false;
1942
1943 /* The cache has been updated since the last time we tried to load the unit. There might be new
1944 * fragment paths to read. */
1945 if (u->manager->unit_cache_timestamp_hash != u->fragment_not_found_timestamp_hash)
1946 return true;
1947
1948 /* The cache needs to be updated because there are modifications on disk. */
1949 return !lookup_paths_timestamp_hash_same(&u->manager->lookup_paths, u->manager->unit_cache_timestamp_hash, NULL);
1950 }
1951
1952 int manager_load_unit_prepare(
1953 Manager *m,
1954 const char *name,
1955 const char *path,
1956 sd_bus_error *e,
1957 Unit **_ret) {
1958
1959 _cleanup_(unit_freep) Unit *cleanup_ret = NULL;
1960 Unit *ret;
1961 UnitType t;
1962 int r;
1963
1964 assert(m);
1965 assert(_ret);
1966
1967 /* This will prepare the unit for loading, but not actually load anything from disk. */
1968
1969 if (path && !path_is_absolute(path))
1970 return sd_bus_error_setf(e, SD_BUS_ERROR_INVALID_ARGS, "Path %s is not absolute.", path);
1971
1972 if (!name) {
1973 /* 'name' and 'path' must not both be null. Check here 'path' using assert_se() to
1974 * workaround a bug in gcc that generates a -Wnonnull warning when calling basename(),
1975 * but this cannot be possible in any code path (See #6119). */
1976 assert_se(path);
1977 name = basename(path);
1978 }
1979
1980 t = unit_name_to_type(name);
1981
1982 if (t == _UNIT_TYPE_INVALID || !unit_name_is_valid(name, UNIT_NAME_PLAIN|UNIT_NAME_INSTANCE)) {
1983 if (unit_name_is_valid(name, UNIT_NAME_TEMPLATE))
1984 return sd_bus_error_setf(e, SD_BUS_ERROR_INVALID_ARGS, "Unit name %s is missing the instance name.", name);
1985
1986 return sd_bus_error_setf(e, SD_BUS_ERROR_INVALID_ARGS, "Unit name %s is not valid.", name);
1987 }
1988
1989 ret = manager_get_unit(m, name);
1990 if (ret) {
1991 /* The time-based cache allows to start new units without daemon-reload,
1992 * but if they are already referenced (because of dependencies or ordering)
1993 * then we have to force a load of the fragment. As an optimization, check
1994 * first if anything in the usual paths was modified since the last time
1995 * the cache was loaded. Also check if the last time an attempt to load the
1996 * unit was made was before the most recent cache refresh, so that we know
1997 * we need to try again — even if the cache is current, it might have been
1998 * updated in a different context before we had a chance to retry loading
1999 * this particular unit. */
2000 if (manager_unit_cache_should_retry_load(ret))
2001 ret->load_state = UNIT_STUB;
2002 else {
2003 *_ret = ret;
2004 return 1;
2005 }
2006 } else {
2007 ret = cleanup_ret = unit_new(m, unit_vtable[t]->object_size);
2008 if (!ret)
2009 return -ENOMEM;
2010 }
2011
2012 if (path) {
2013 r = free_and_strdup(&ret->fragment_path, path);
2014 if (r < 0)
2015 return r;
2016 }
2017
2018 r = unit_add_name(ret, name);
2019 if (r < 0)
2020 return r;
2021
2022 unit_add_to_load_queue(ret);
2023 unit_add_to_dbus_queue(ret);
2024 unit_add_to_gc_queue(ret);
2025
2026 *_ret = ret;
2027 cleanup_ret = NULL;
2028
2029 return 0;
2030 }
2031
2032 int manager_load_unit(
2033 Manager *m,
2034 const char *name,
2035 const char *path,
2036 sd_bus_error *e,
2037 Unit **_ret) {
2038
2039 int r;
2040
2041 assert(m);
2042 assert(_ret);
2043
2044 /* This will load the service information files, but not actually
2045 * start any services or anything. */
2046
2047 r = manager_load_unit_prepare(m, name, path, e, _ret);
2048 if (r != 0)
2049 return r;
2050
2051 manager_dispatch_load_queue(m);
2052
2053 *_ret = unit_follow_merge(*_ret);
2054 return 0;
2055 }
2056
2057 int manager_load_startable_unit_or_warn(
2058 Manager *m,
2059 const char *name,
2060 const char *path,
2061 Unit **ret) {
2062
2063 /* Load a unit, make sure it loaded fully and is not masked. */
2064
2065 _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL;
2066 Unit *unit;
2067 int r;
2068
2069 r = manager_load_unit(m, name, path, &error, &unit);
2070 if (r < 0)
2071 return log_error_errno(r, "Failed to load %s %s: %s",
2072 name ? "unit" : "unit file", name ?: path,
2073 bus_error_message(&error, r));
2074
2075 r = bus_unit_validate_load_state(unit, &error);
2076 if (r < 0)
2077 return log_error_errno(r, "%s", bus_error_message(&error, r));
2078
2079 *ret = unit;
2080 return 0;
2081 }
2082
2083 void manager_dump_jobs(Manager *s, FILE *f, const char *prefix) {
2084 Job *j;
2085
2086 assert(s);
2087 assert(f);
2088
2089 HASHMAP_FOREACH(j, s->jobs)
2090 job_dump(j, f, prefix);
2091 }
2092
2093 void manager_dump_units(Manager *s, FILE *f, const char *prefix) {
2094 Unit *u;
2095 const char *t;
2096
2097 assert(s);
2098 assert(f);
2099
2100 HASHMAP_FOREACH_KEY(u, t, s->units)
2101 if (u->id == t)
2102 unit_dump(u, f, prefix);
2103 }
2104
2105 void manager_dump(Manager *m, FILE *f, const char *prefix) {
2106 assert(m);
2107 assert(f);
2108
2109 for (ManagerTimestamp q = 0; q < _MANAGER_TIMESTAMP_MAX; q++) {
2110 const dual_timestamp *t = m->timestamps + q;
2111 char buf[CONST_MAX(FORMAT_TIMESPAN_MAX, FORMAT_TIMESTAMP_MAX)];
2112
2113 if (dual_timestamp_is_set(t))
2114 fprintf(f, "%sTimestamp %s: %s\n",
2115 strempty(prefix),
2116 manager_timestamp_to_string(q),
2117 timestamp_is_set(t->realtime) ? format_timestamp(buf, sizeof buf, t->realtime) :
2118 format_timespan(buf, sizeof buf, t->monotonic, 1));
2119 }
2120
2121 manager_dump_units(m, f, prefix);
2122 manager_dump_jobs(m, f, prefix);
2123 }
2124
2125 int manager_get_dump_string(Manager *m, char **ret) {
2126 _cleanup_free_ char *dump = NULL;
2127 _cleanup_fclose_ FILE *f = NULL;
2128 size_t size;
2129 int r;
2130
2131 assert(m);
2132 assert(ret);
2133
2134 f = open_memstream_unlocked(&dump, &size);
2135 if (!f)
2136 return -errno;
2137
2138 manager_dump(m, f, NULL);
2139
2140 r = fflush_and_check(f);
2141 if (r < 0)
2142 return r;
2143
2144 f = safe_fclose(f);
2145
2146 *ret = TAKE_PTR(dump);
2147
2148 return 0;
2149 }
2150
2151 void manager_clear_jobs(Manager *m) {
2152 Job *j;
2153
2154 assert(m);
2155
2156 while ((j = hashmap_first(m->jobs)))
2157 /* No need to recurse. We're cancelling all jobs. */
2158 job_finish_and_invalidate(j, JOB_CANCELED, false, false);
2159 }
2160
2161 void manager_unwatch_pid(Manager *m, pid_t pid) {
2162 assert(m);
2163
2164 /* First let's drop the unit keyed as "pid". */
2165 (void) hashmap_remove(m->watch_pids, PID_TO_PTR(pid));
2166
2167 /* Then, let's also drop the array keyed by -pid. */
2168 free(hashmap_remove(m->watch_pids, PID_TO_PTR(-pid)));
2169 }
2170
2171 static int manager_dispatch_run_queue(sd_event_source *source, void *userdata) {
2172 Manager *m = userdata;
2173 Job *j;
2174
2175 assert(source);
2176 assert(m);
2177
2178 while ((j = prioq_peek(m->run_queue))) {
2179 assert(j->installed);
2180 assert(j->in_run_queue);
2181
2182 (void) job_run_and_invalidate(j);
2183 }
2184
2185 if (m->n_running_jobs > 0)
2186 manager_watch_jobs_in_progress(m);
2187
2188 if (m->n_on_console > 0)
2189 manager_watch_idle_pipe(m);
2190
2191 return 1;
2192 }
2193
2194 static unsigned manager_dispatch_dbus_queue(Manager *m) {
2195 unsigned n = 0, budget;
2196 Unit *u;
2197 Job *j;
2198
2199 assert(m);
2200
2201 /* When we are reloading, let's not wait with generating signals, since we need to exit the manager as quickly
2202 * as we can. There's no point in throttling generation of signals in that case. */
2203 if (MANAGER_IS_RELOADING(m) || m->send_reloading_done || m->pending_reload_message)
2204 budget = UINT_MAX; /* infinite budget in this case */
2205 else {
2206 /* Anything to do at all? */
2207 if (!m->dbus_unit_queue && !m->dbus_job_queue)
2208 return 0;
2209
2210 /* Do we have overly many messages queued at the moment? If so, let's not enqueue more on top, let's
2211 * sit this cycle out, and process things in a later cycle when the queues got a bit emptier. */
2212 if (manager_bus_n_queued_write(m) > MANAGER_BUS_BUSY_THRESHOLD)
2213 return 0;
2214
2215 /* Only process a certain number of units/jobs per event loop iteration. Even if the bus queue wasn't
2216 * overly full before this call we shouldn't increase it in size too wildly in one step, and we
2217 * shouldn't monopolize CPU time with generating these messages. Note the difference in counting of
2218 * this "budget" and the "threshold" above: the "budget" is decreased only once per generated message,
2219 * regardless how many buses/direct connections it is enqueued on, while the "threshold" is applied to
2220 * each queued instance of bus message, i.e. if the same message is enqueued to five buses/direct
2221 * connections it will be counted five times. This difference in counting ("references"
2222 * vs. "instances") is primarily a result of the fact that it's easier to implement it this way,
2223 * however it also reflects the thinking that the "threshold" should put a limit on used queue memory,
2224 * i.e. space, while the "budget" should put a limit on time. Also note that the "threshold" is
2225 * currently chosen much higher than the "budget". */
2226 budget = MANAGER_BUS_MESSAGE_BUDGET;
2227 }
2228
2229 while (budget != 0 && (u = m->dbus_unit_queue)) {
2230
2231 assert(u->in_dbus_queue);
2232
2233 bus_unit_send_change_signal(u);
2234 n++;
2235
2236 if (budget != UINT_MAX)
2237 budget--;
2238 }
2239
2240 while (budget != 0 && (j = m->dbus_job_queue)) {
2241 assert(j->in_dbus_queue);
2242
2243 bus_job_send_change_signal(j);
2244 n++;
2245
2246 if (budget != UINT_MAX)
2247 budget--;
2248 }
2249
2250 if (m->send_reloading_done) {
2251 m->send_reloading_done = false;
2252 bus_manager_send_reloading(m, false);
2253 n++;
2254 }
2255
2256 if (m->pending_reload_message) {
2257 bus_send_pending_reload_message(m);
2258 n++;
2259 }
2260
2261 return n;
2262 }
2263
2264 static int manager_dispatch_cgroups_agent_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
2265 Manager *m = userdata;
2266 char buf[PATH_MAX];
2267 ssize_t n;
2268
2269 n = recv(fd, buf, sizeof(buf), 0);
2270 if (n < 0)
2271 return log_error_errno(errno, "Failed to read cgroups agent message: %m");
2272 if (n == 0) {
2273 log_error("Got zero-length cgroups agent message, ignoring.");
2274 return 0;
2275 }
2276 if ((size_t) n >= sizeof(buf)) {
2277 log_error("Got overly long cgroups agent message, ignoring.");
2278 return 0;
2279 }
2280
2281 if (memchr(buf, 0, n)) {
2282 log_error("Got cgroups agent message with embedded NUL byte, ignoring.");
2283 return 0;
2284 }
2285 buf[n] = 0;
2286
2287 manager_notify_cgroup_empty(m, buf);
2288 (void) bus_forward_agent_released(m, buf);
2289
2290 return 0;
2291 }
2292
2293 static bool manager_process_barrier_fd(char * const *tags, FDSet *fds) {
2294
2295 /* nothing else must be sent when using BARRIER=1 */
2296 if (strv_contains(tags, "BARRIER=1")) {
2297 if (strv_length(tags) == 1) {
2298 if (fdset_size(fds) != 1)
2299 log_warning("Got incorrect number of fds with BARRIER=1, closing them.");
2300 } else
2301 log_warning("Extra notification messages sent with BARRIER=1, ignoring everything.");
2302
2303 /* Drop the message if BARRIER=1 was found */
2304 return true;
2305 }
2306
2307 return false;
2308 }
2309
2310 static void manager_invoke_notify_message(
2311 Manager *m,
2312 Unit *u,
2313 const struct ucred *ucred,
2314 char * const *tags,
2315 FDSet *fds) {
2316
2317 assert(m);
2318 assert(u);
2319 assert(ucred);
2320 assert(tags);
2321
2322 if (u->notifygen == m->notifygen) /* Already invoked on this same unit in this same iteration? */
2323 return;
2324 u->notifygen = m->notifygen;
2325
2326 if (UNIT_VTABLE(u)->notify_message)
2327 UNIT_VTABLE(u)->notify_message(u, ucred, tags, fds);
2328
2329 else if (DEBUG_LOGGING) {
2330 _cleanup_free_ char *buf = NULL, *x = NULL, *y = NULL;
2331
2332 buf = strv_join(tags, ", ");
2333 if (buf)
2334 x = ellipsize(buf, 20, 90);
2335 if (x)
2336 y = cescape(x);
2337
2338 log_unit_debug(u, "Got notification message \"%s\", ignoring.", strnull(y));
2339 }
2340 }
2341
2342 static int manager_dispatch_notify_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
2343
2344 _cleanup_fdset_free_ FDSet *fds = NULL;
2345 Manager *m = userdata;
2346 char buf[NOTIFY_BUFFER_MAX+1];
2347 struct iovec iovec = {
2348 .iov_base = buf,
2349 .iov_len = sizeof(buf)-1,
2350 };
2351 CMSG_BUFFER_TYPE(CMSG_SPACE(sizeof(struct ucred)) +
2352 CMSG_SPACE(sizeof(int) * NOTIFY_FD_MAX)) control;
2353 struct msghdr msghdr = {
2354 .msg_iov = &iovec,
2355 .msg_iovlen = 1,
2356 .msg_control = &control,
2357 .msg_controllen = sizeof(control),
2358 };
2359
2360 struct cmsghdr *cmsg;
2361 struct ucred *ucred = NULL;
2362 _cleanup_free_ Unit **array_copy = NULL;
2363 _cleanup_strv_free_ char **tags = NULL;
2364 Unit *u1, *u2, **array;
2365 int r, *fd_array = NULL;
2366 size_t n_fds = 0;
2367 bool found = false;
2368 ssize_t n;
2369
2370 assert(m);
2371 assert(m->notify_fd == fd);
2372
2373 if (revents != EPOLLIN) {
2374 log_warning("Got unexpected poll event for notify fd.");
2375 return 0;
2376 }
2377
2378 n = recvmsg_safe(m->notify_fd, &msghdr, MSG_DONTWAIT|MSG_CMSG_CLOEXEC|MSG_TRUNC);
2379 if (IN_SET(n, -EAGAIN, -EINTR))
2380 return 0; /* Spurious wakeup, try again */
2381 if (n == -EXFULL) {
2382 log_warning("Got message with truncated control data (too many fds sent?), ignoring.");
2383 return 0;
2384 }
2385 if (n < 0)
2386 /* If this is any other, real error, then let's stop processing this socket. This of course
2387 * means we won't take notification messages anymore, but that's still better than busy
2388 * looping around this: being woken up over and over again but being unable to actually read
2389 * the message off the socket. */
2390 return log_error_errno(n, "Failed to receive notification message: %m");
2391
2392 CMSG_FOREACH(cmsg, &msghdr) {
2393 if (cmsg->cmsg_level == SOL_SOCKET && cmsg->cmsg_type == SCM_RIGHTS) {
2394
2395 assert(!fd_array);
2396 fd_array = (int*) CMSG_DATA(cmsg);
2397 n_fds = (cmsg->cmsg_len - CMSG_LEN(0)) / sizeof(int);
2398
2399 } else if (cmsg->cmsg_level == SOL_SOCKET &&
2400 cmsg->cmsg_type == SCM_CREDENTIALS &&
2401 cmsg->cmsg_len == CMSG_LEN(sizeof(struct ucred))) {
2402
2403 assert(!ucred);
2404 ucred = (struct ucred*) CMSG_DATA(cmsg);
2405 }
2406 }
2407
2408 if (n_fds > 0) {
2409 assert(fd_array);
2410
2411 r = fdset_new_array(&fds, fd_array, n_fds);
2412 if (r < 0) {
2413 close_many(fd_array, n_fds);
2414 log_oom();
2415 return 0;
2416 }
2417 }
2418
2419 if (!ucred || !pid_is_valid(ucred->pid)) {
2420 log_warning("Received notify message without valid credentials. Ignoring.");
2421 return 0;
2422 }
2423
2424 if ((size_t) n >= sizeof(buf) || (msghdr.msg_flags & MSG_TRUNC)) {
2425 log_warning("Received notify message exceeded maximum size. Ignoring.");
2426 return 0;
2427 }
2428
2429 /* As extra safety check, let's make sure the string we get doesn't contain embedded NUL bytes. We permit one
2430 * trailing NUL byte in the message, but don't expect it. */
2431 if (n > 1 && memchr(buf, 0, n-1)) {
2432 log_warning("Received notify message with embedded NUL bytes. Ignoring.");
2433 return 0;
2434 }
2435
2436 /* Make sure it's NUL-terminated, then parse it to obtain the tags list */
2437 buf[n] = 0;
2438 tags = strv_split_newlines(buf);
2439 if (!tags) {
2440 log_oom();
2441 return 0;
2442 }
2443
2444 /* possibly a barrier fd, let's see */
2445 if (manager_process_barrier_fd(tags, fds))
2446 return 0;
2447
2448 /* Increase the generation counter used for filtering out duplicate unit invocations. */
2449 m->notifygen++;
2450
2451 /* Notify every unit that might be interested, which might be multiple. */
2452 u1 = manager_get_unit_by_pid_cgroup(m, ucred->pid);
2453 u2 = hashmap_get(m->watch_pids, PID_TO_PTR(ucred->pid));
2454 array = hashmap_get(m->watch_pids, PID_TO_PTR(-ucred->pid));
2455 if (array) {
2456 size_t k = 0;
2457
2458 while (array[k])
2459 k++;
2460
2461 array_copy = newdup(Unit*, array, k+1);
2462 if (!array_copy)
2463 log_oom();
2464 }
2465 /* And now invoke the per-unit callbacks. Note that manager_invoke_notify_message() will handle duplicate units
2466 * make sure we only invoke each unit's handler once. */
2467 if (u1) {
2468 manager_invoke_notify_message(m, u1, ucred, tags, fds);
2469 found = true;
2470 }
2471 if (u2) {
2472 manager_invoke_notify_message(m, u2, ucred, tags, fds);
2473 found = true;
2474 }
2475 if (array_copy)
2476 for (size_t i = 0; array_copy[i]; i++) {
2477 manager_invoke_notify_message(m, array_copy[i], ucred, tags, fds);
2478 found = true;
2479 }
2480
2481 if (!found)
2482 log_warning("Cannot find unit for notify message of PID "PID_FMT", ignoring.", ucred->pid);
2483
2484 if (fdset_size(fds) > 0)
2485 log_warning("Got extra auxiliary fds with notification message, closing them.");
2486
2487 return 0;
2488 }
2489
2490 static void manager_invoke_sigchld_event(
2491 Manager *m,
2492 Unit *u,
2493 const siginfo_t *si) {
2494
2495 assert(m);
2496 assert(u);
2497 assert(si);
2498
2499 /* Already invoked the handler of this unit in this iteration? Then don't process this again */
2500 if (u->sigchldgen == m->sigchldgen)
2501 return;
2502 u->sigchldgen = m->sigchldgen;
2503
2504 log_unit_debug(u, "Child "PID_FMT" belongs to %s.", si->si_pid, u->id);
2505 unit_unwatch_pid(u, si->si_pid);
2506
2507 if (UNIT_VTABLE(u)->sigchld_event)
2508 UNIT_VTABLE(u)->sigchld_event(u, si->si_pid, si->si_code, si->si_status);
2509 }
2510
2511 static int manager_dispatch_sigchld(sd_event_source *source, void *userdata) {
2512 Manager *m = userdata;
2513 siginfo_t si = {};
2514 int r;
2515
2516 assert(source);
2517 assert(m);
2518
2519 /* First we call waitid() for a PID and do not reap the zombie. That way we can still access /proc/$PID for it
2520 * while it is a zombie. */
2521
2522 if (waitid(P_ALL, 0, &si, WEXITED|WNOHANG|WNOWAIT) < 0) {
2523
2524 if (errno != ECHILD)
2525 log_error_errno(errno, "Failed to peek for child with waitid(), ignoring: %m");
2526
2527 goto turn_off;
2528 }
2529
2530 if (si.si_pid <= 0)
2531 goto turn_off;
2532
2533 if (IN_SET(si.si_code, CLD_EXITED, CLD_KILLED, CLD_DUMPED)) {
2534 _cleanup_free_ Unit **array_copy = NULL;
2535 _cleanup_free_ char *name = NULL;
2536 Unit *u1, *u2, **array;
2537
2538 (void) get_process_comm(si.si_pid, &name);
2539
2540 log_debug("Child "PID_FMT" (%s) died (code=%s, status=%i/%s)",
2541 si.si_pid, strna(name),
2542 sigchld_code_to_string(si.si_code),
2543 si.si_status,
2544 strna(si.si_code == CLD_EXITED
2545 ? exit_status_to_string(si.si_status, EXIT_STATUS_FULL)
2546 : signal_to_string(si.si_status)));
2547
2548 /* Increase the generation counter used for filtering out duplicate unit invocations */
2549 m->sigchldgen++;
2550
2551 /* And now figure out the unit this belongs to, it might be multiple... */
2552 u1 = manager_get_unit_by_pid_cgroup(m, si.si_pid);
2553 u2 = hashmap_get(m->watch_pids, PID_TO_PTR(si.si_pid));
2554 array = hashmap_get(m->watch_pids, PID_TO_PTR(-si.si_pid));
2555 if (array) {
2556 size_t n = 0;
2557
2558 /* Count how many entries the array has */
2559 while (array[n])
2560 n++;
2561
2562 /* Make a copy of the array so that we don't trip up on the array changing beneath us */
2563 array_copy = newdup(Unit*, array, n+1);
2564 if (!array_copy)
2565 log_oom();
2566 }
2567
2568 /* Finally, execute them all. Note that u1, u2 and the array might contain duplicates, but
2569 * that's fine, manager_invoke_sigchld_event() will ensure we only invoke the handlers once for
2570 * each iteration. */
2571 if (u1) {
2572 /* We check for oom condition, in case we got SIGCHLD before the oom notification.
2573 * We only do this for the cgroup the PID belonged to. */
2574 (void) unit_check_oom(u1);
2575
2576 /* This only logs for now. In the future when the interface for kills/notifications
2577 * is more stable we can extend service results table similar to how kernel oom kills
2578 * are managed. */
2579 (void) unit_check_oomd_kill(u1);
2580
2581 manager_invoke_sigchld_event(m, u1, &si);
2582 }
2583 if (u2)
2584 manager_invoke_sigchld_event(m, u2, &si);
2585 if (array_copy)
2586 for (size_t i = 0; array_copy[i]; i++)
2587 manager_invoke_sigchld_event(m, array_copy[i], &si);
2588 }
2589
2590 /* And now, we actually reap the zombie. */
2591 if (waitid(P_PID, si.si_pid, &si, WEXITED) < 0) {
2592 log_error_errno(errno, "Failed to dequeue child, ignoring: %m");
2593 return 0;
2594 }
2595
2596 return 0;
2597
2598 turn_off:
2599 /* All children processed for now, turn off event source */
2600
2601 r = sd_event_source_set_enabled(m->sigchld_event_source, SD_EVENT_OFF);
2602 if (r < 0)
2603 return log_error_errno(r, "Failed to disable SIGCHLD event source: %m");
2604
2605 return 0;
2606 }
2607
2608 static void manager_start_target(Manager *m, const char *name, JobMode mode) {
2609 _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL;
2610 int r;
2611
2612 log_debug("Activating special unit %s", name);
2613
2614 r = manager_add_job_by_name(m, JOB_START, name, mode, NULL, &error, NULL);
2615 if (r < 0)
2616 log_error("Failed to enqueue %s job: %s", name, bus_error_message(&error, r));
2617 }
2618
2619 static void manager_handle_ctrl_alt_del(Manager *m) {
2620 /* If the user presses C-A-D more than
2621 * 7 times within 2s, we reboot/shutdown immediately,
2622 * unless it was disabled in system.conf */
2623
2624 if (ratelimit_below(&m->ctrl_alt_del_ratelimit) || m->cad_burst_action == EMERGENCY_ACTION_NONE)
2625 manager_start_target(m, SPECIAL_CTRL_ALT_DEL_TARGET, JOB_REPLACE_IRREVERSIBLY);
2626 else
2627 emergency_action(m, m->cad_burst_action, EMERGENCY_ACTION_WARN, NULL, -1,
2628 "Ctrl-Alt-Del was pressed more than 7 times within 2s");
2629 }
2630
2631 static int manager_dispatch_signal_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
2632 Manager *m = userdata;
2633 ssize_t n;
2634 struct signalfd_siginfo sfsi;
2635 int r;
2636
2637 assert(m);
2638 assert(m->signal_fd == fd);
2639
2640 if (revents != EPOLLIN) {
2641 log_warning("Got unexpected events from signal file descriptor.");
2642 return 0;
2643 }
2644
2645 n = read(m->signal_fd, &sfsi, sizeof(sfsi));
2646 if (n != sizeof(sfsi)) {
2647 if (n >= 0) {
2648 log_warning("Truncated read from signal fd (%zu bytes), ignoring!", n);
2649 return 0;
2650 }
2651
2652 if (IN_SET(errno, EINTR, EAGAIN))
2653 return 0;
2654
2655 /* We return an error here, which will kill this handler,
2656 * to avoid a busy loop on read error. */
2657 return log_error_errno(errno, "Reading from signal fd failed: %m");
2658 }
2659
2660 log_received_signal(sfsi.ssi_signo == SIGCHLD ||
2661 (sfsi.ssi_signo == SIGTERM && MANAGER_IS_USER(m))
2662 ? LOG_DEBUG : LOG_INFO,
2663 &sfsi);
2664
2665 switch (sfsi.ssi_signo) {
2666
2667 case SIGCHLD:
2668 r = sd_event_source_set_enabled(m->sigchld_event_source, SD_EVENT_ON);
2669 if (r < 0)
2670 log_warning_errno(r, "Failed to enable SIGCHLD event source, ignoring: %m");
2671
2672 break;
2673
2674 case SIGTERM:
2675 if (MANAGER_IS_SYSTEM(m)) {
2676 /* This is for compatibility with the original sysvinit */
2677 if (verify_run_space_and_log("Refusing to reexecute") < 0)
2678 break;
2679
2680 m->objective = MANAGER_REEXECUTE;
2681 break;
2682 }
2683
2684 _fallthrough_;
2685 case SIGINT:
2686 if (MANAGER_IS_SYSTEM(m))
2687 manager_handle_ctrl_alt_del(m);
2688 else
2689 manager_start_target(m, SPECIAL_EXIT_TARGET,
2690 JOB_REPLACE_IRREVERSIBLY);
2691 break;
2692
2693 case SIGWINCH:
2694 /* This is a nop on non-init */
2695 if (MANAGER_IS_SYSTEM(m))
2696 manager_start_target(m, SPECIAL_KBREQUEST_TARGET, JOB_REPLACE);
2697
2698 break;
2699
2700 case SIGPWR:
2701 /* This is a nop on non-init */
2702 if (MANAGER_IS_SYSTEM(m))
2703 manager_start_target(m, SPECIAL_SIGPWR_TARGET, JOB_REPLACE);
2704
2705 break;
2706
2707 case SIGUSR1:
2708 if (manager_dbus_is_running(m, false)) {
2709 log_info("Trying to reconnect to bus...");
2710
2711 (void) bus_init_api(m);
2712
2713 if (MANAGER_IS_SYSTEM(m))
2714 (void) bus_init_system(m);
2715 } else {
2716 log_info("Starting D-Bus service...");
2717 manager_start_target(m, SPECIAL_DBUS_SERVICE, JOB_REPLACE);
2718 }
2719
2720 break;
2721
2722 case SIGUSR2: {
2723 _cleanup_free_ char *dump = NULL;
2724
2725 r = manager_get_dump_string(m, &dump);
2726 if (r < 0) {
2727 log_warning_errno(errno, "Failed to acquire manager dump: %m");
2728 break;
2729 }
2730
2731 log_dump(LOG_INFO, dump);
2732 break;
2733 }
2734
2735 case SIGHUP:
2736 if (verify_run_space_and_log("Refusing to reload") < 0)
2737 break;
2738
2739 m->objective = MANAGER_RELOAD;
2740 break;
2741
2742 default: {
2743
2744 /* Starting SIGRTMIN+0 */
2745 static const struct {
2746 const char *target;
2747 JobMode mode;
2748 } target_table[] = {
2749 [0] = { SPECIAL_DEFAULT_TARGET, JOB_ISOLATE },
2750 [1] = { SPECIAL_RESCUE_TARGET, JOB_ISOLATE },
2751 [2] = { SPECIAL_EMERGENCY_TARGET, JOB_ISOLATE },
2752 [3] = { SPECIAL_HALT_TARGET, JOB_REPLACE_IRREVERSIBLY },
2753 [4] = { SPECIAL_POWEROFF_TARGET, JOB_REPLACE_IRREVERSIBLY },
2754 [5] = { SPECIAL_REBOOT_TARGET, JOB_REPLACE_IRREVERSIBLY },
2755 [6] = { SPECIAL_KEXEC_TARGET, JOB_REPLACE_IRREVERSIBLY },
2756 };
2757
2758 /* Starting SIGRTMIN+13, so that target halt and system halt are 10 apart */
2759 static const ManagerObjective objective_table[] = {
2760 [0] = MANAGER_HALT,
2761 [1] = MANAGER_POWEROFF,
2762 [2] = MANAGER_REBOOT,
2763 [3] = MANAGER_KEXEC,
2764 };
2765
2766 if ((int) sfsi.ssi_signo >= SIGRTMIN+0 &&
2767 (int) sfsi.ssi_signo < SIGRTMIN+(int) ELEMENTSOF(target_table)) {
2768 int idx = (int) sfsi.ssi_signo - SIGRTMIN;
2769 manager_start_target(m, target_table[idx].target,
2770 target_table[idx].mode);
2771 break;
2772 }
2773
2774 if ((int) sfsi.ssi_signo >= SIGRTMIN+13 &&
2775 (int) sfsi.ssi_signo < SIGRTMIN+13+(int) ELEMENTSOF(objective_table)) {
2776 m->objective = objective_table[sfsi.ssi_signo - SIGRTMIN - 13];
2777 break;
2778 }
2779
2780 switch (sfsi.ssi_signo - SIGRTMIN) {
2781
2782 case 20:
2783 manager_override_show_status(m, SHOW_STATUS_YES, "signal");
2784 break;
2785
2786 case 21:
2787 manager_override_show_status(m, SHOW_STATUS_NO, "signal");
2788 break;
2789
2790 case 22:
2791 manager_override_log_level(m, LOG_DEBUG);
2792 break;
2793
2794 case 23:
2795 manager_restore_original_log_level(m);
2796 break;
2797
2798 case 24:
2799 if (MANAGER_IS_USER(m)) {
2800 m->objective = MANAGER_EXIT;
2801 return 0;
2802 }
2803
2804 /* This is a nop on init */
2805 break;
2806
2807 case 26:
2808 case 29: /* compatibility: used to be mapped to LOG_TARGET_SYSLOG_OR_KMSG */
2809 manager_restore_original_log_target(m);
2810 break;
2811
2812 case 27:
2813 manager_override_log_target(m, LOG_TARGET_CONSOLE);
2814 break;
2815
2816 case 28:
2817 manager_override_log_target(m, LOG_TARGET_KMSG);
2818 break;
2819
2820 default:
2821 log_warning("Got unhandled signal <%s>.", signal_to_string(sfsi.ssi_signo));
2822 }
2823 }}
2824
2825 return 0;
2826 }
2827
2828 static int manager_dispatch_time_change_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
2829 Manager *m = userdata;
2830 Unit *u;
2831
2832 assert(m);
2833 assert(m->time_change_fd == fd);
2834
2835 log_struct(LOG_DEBUG,
2836 "MESSAGE_ID=" SD_MESSAGE_TIME_CHANGE_STR,
2837 LOG_MESSAGE("Time has been changed"));
2838
2839 /* Restart the watch */
2840 (void) manager_setup_time_change(m);
2841
2842 HASHMAP_FOREACH(u, m->units)
2843 if (UNIT_VTABLE(u)->time_change)
2844 UNIT_VTABLE(u)->time_change(u);
2845
2846 return 0;
2847 }
2848
2849 static int manager_dispatch_timezone_change(
2850 sd_event_source *source,
2851 const struct inotify_event *e,
2852 void *userdata) {
2853
2854 Manager *m = userdata;
2855 int changed;
2856 Unit *u;
2857
2858 assert(m);
2859
2860 log_debug("inotify event for /etc/localtime");
2861
2862 changed = manager_read_timezone_stat(m);
2863 if (changed <= 0)
2864 return changed;
2865
2866 /* Something changed, restart the watch, to ensure we watch the new /etc/localtime if it changed */
2867 (void) manager_setup_timezone_change(m);
2868
2869 /* Read the new timezone */
2870 tzset();
2871
2872 log_debug("Timezone has been changed (now: %s).", tzname[daylight]);
2873
2874 HASHMAP_FOREACH(u, m->units)
2875 if (UNIT_VTABLE(u)->timezone_change)
2876 UNIT_VTABLE(u)->timezone_change(u);
2877
2878 return 0;
2879 }
2880
2881 static int manager_dispatch_idle_pipe_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
2882 Manager *m = userdata;
2883
2884 assert(m);
2885 assert(m->idle_pipe[2] == fd);
2886
2887 /* There's at least one Type=idle child that just gave up on us waiting for the boot process to complete. Let's
2888 * now turn off any further console output if there's at least one service that needs console access, so that
2889 * from now on our own output should not spill into that service's output anymore. After all, we support
2890 * Type=idle only to beautify console output and it generally is set on services that want to own the console
2891 * exclusively without our interference. */
2892 m->no_console_output = m->n_on_console > 0;
2893
2894 /* Acknowledge the child's request, and let all all other children know too that they shouldn't wait any longer
2895 * by closing the pipes towards them, which is what they are waiting for. */
2896 manager_close_idle_pipe(m);
2897
2898 return 0;
2899 }
2900
2901 static int manager_dispatch_jobs_in_progress(sd_event_source *source, usec_t usec, void *userdata) {
2902 Manager *m = userdata;
2903 int r;
2904
2905 assert(m);
2906 assert(source);
2907
2908 manager_print_jobs_in_progress(m);
2909
2910 r = sd_event_source_set_time_relative(source, JOBS_IN_PROGRESS_PERIOD_USEC);
2911 if (r < 0)
2912 return r;
2913
2914 return sd_event_source_set_enabled(source, SD_EVENT_ONESHOT);
2915 }
2916
2917 int manager_loop(Manager *m) {
2918 RateLimit rl = { .interval = 1*USEC_PER_SEC, .burst = 50000 };
2919 int r;
2920
2921 assert(m);
2922 assert(m->objective == MANAGER_OK); /* Ensure manager_startup() has been called */
2923
2924 manager_check_finished(m);
2925
2926 /* There might still be some zombies hanging around from before we were exec()'ed. Let's reap them. */
2927 r = sd_event_source_set_enabled(m->sigchld_event_source, SD_EVENT_ON);
2928 if (r < 0)
2929 return log_error_errno(r, "Failed to enable SIGCHLD event source: %m");
2930
2931 while (m->objective == MANAGER_OK) {
2932 usec_t wait_usec, watchdog_usec;
2933
2934 watchdog_usec = manager_get_watchdog(m, WATCHDOG_RUNTIME);
2935 if (m->runtime_watchdog_running)
2936 (void) watchdog_ping();
2937 else if (timestamp_is_set(watchdog_usec))
2938 manager_retry_runtime_watchdog(m);
2939
2940 if (!ratelimit_below(&rl)) {
2941 /* Yay, something is going seriously wrong, pause a little */
2942 log_warning("Looping too fast. Throttling execution a little.");
2943 sleep(1);
2944 }
2945
2946 if (manager_dispatch_load_queue(m) > 0)
2947 continue;
2948
2949 if (manager_dispatch_gc_job_queue(m) > 0)
2950 continue;
2951
2952 if (manager_dispatch_gc_unit_queue(m) > 0)
2953 continue;
2954
2955 if (manager_dispatch_cleanup_queue(m) > 0)
2956 continue;
2957
2958 if (manager_dispatch_cgroup_realize_queue(m) > 0)
2959 continue;
2960
2961 if (manager_dispatch_stop_when_unneeded_queue(m) > 0)
2962 continue;
2963
2964 if (manager_dispatch_dbus_queue(m) > 0)
2965 continue;
2966
2967 /* Sleep for watchdog runtime wait time */
2968 if (timestamp_is_set(watchdog_usec))
2969 wait_usec = watchdog_runtime_wait();
2970 else
2971 wait_usec = USEC_INFINITY;
2972
2973 r = sd_event_run(m->event, wait_usec);
2974 if (r < 0)
2975 return log_error_errno(r, "Failed to run event loop: %m");
2976 }
2977
2978 return m->objective;
2979 }
2980
2981 int manager_load_unit_from_dbus_path(Manager *m, const char *s, sd_bus_error *e, Unit **_u) {
2982 _cleanup_free_ char *n = NULL;
2983 sd_id128_t invocation_id;
2984 Unit *u;
2985 int r;
2986
2987 assert(m);
2988 assert(s);
2989 assert(_u);
2990
2991 r = unit_name_from_dbus_path(s, &n);
2992 if (r < 0)
2993 return r;
2994
2995 /* Permit addressing units by invocation ID: if the passed bus path is suffixed by a 128bit ID then we use it
2996 * as invocation ID. */
2997 r = sd_id128_from_string(n, &invocation_id);
2998 if (r >= 0) {
2999 u = hashmap_get(m->units_by_invocation_id, &invocation_id);
3000 if (u) {
3001 *_u = u;
3002 return 0;
3003 }
3004
3005 return sd_bus_error_setf(e, BUS_ERROR_NO_UNIT_FOR_INVOCATION_ID,
3006 "No unit with the specified invocation ID " SD_ID128_FORMAT_STR " known.",
3007 SD_ID128_FORMAT_VAL(invocation_id));
3008 }
3009
3010 /* If this didn't work, we check if this is a unit name */
3011 if (!unit_name_is_valid(n, UNIT_NAME_PLAIN|UNIT_NAME_INSTANCE)) {
3012 _cleanup_free_ char *nn = NULL;
3013
3014 nn = cescape(n);
3015 return sd_bus_error_setf(e, SD_BUS_ERROR_INVALID_ARGS,
3016 "Unit name %s is neither a valid invocation ID nor unit name.", strnull(nn));
3017 }
3018
3019 r = manager_load_unit(m, n, NULL, e, &u);
3020 if (r < 0)
3021 return r;
3022
3023 *_u = u;
3024 return 0;
3025 }
3026
3027 int manager_get_job_from_dbus_path(Manager *m, const char *s, Job **_j) {
3028 const char *p;
3029 unsigned id;
3030 Job *j;
3031 int r;
3032
3033 assert(m);
3034 assert(s);
3035 assert(_j);
3036
3037 p = startswith(s, "/org/freedesktop/systemd1/job/");
3038 if (!p)
3039 return -EINVAL;
3040
3041 r = safe_atou(p, &id);
3042 if (r < 0)
3043 return r;
3044
3045 j = manager_get_job(m, id);
3046 if (!j)
3047 return -ENOENT;
3048
3049 *_j = j;
3050
3051 return 0;
3052 }
3053
3054 void manager_send_unit_audit(Manager *m, Unit *u, int type, bool success) {
3055
3056 #if HAVE_AUDIT
3057 _cleanup_free_ char *p = NULL;
3058 const char *msg;
3059 int audit_fd, r;
3060
3061 if (!MANAGER_IS_SYSTEM(m))
3062 return;
3063
3064 audit_fd = get_audit_fd();
3065 if (audit_fd < 0)
3066 return;
3067
3068 /* Don't generate audit events if the service was already
3069 * started and we're just deserializing */
3070 if (MANAGER_IS_RELOADING(m))
3071 return;
3072
3073 if (u->type != UNIT_SERVICE)
3074 return;
3075
3076 r = unit_name_to_prefix_and_instance(u->id, &p);
3077 if (r < 0) {
3078 log_error_errno(r, "Failed to extract prefix and instance of unit name: %m");
3079 return;
3080 }
3081
3082 msg = strjoina("unit=", p);
3083 if (audit_log_user_comm_message(audit_fd, type, msg, "systemd", NULL, NULL, NULL, success) < 0) {
3084 if (errno == EPERM)
3085 /* We aren't allowed to send audit messages?
3086 * Then let's not retry again. */
3087 close_audit_fd();
3088 else
3089 log_warning_errno(errno, "Failed to send audit message: %m");
3090 }
3091 #endif
3092
3093 }
3094
3095 void manager_send_unit_plymouth(Manager *m, Unit *u) {
3096 static const union sockaddr_union sa = PLYMOUTH_SOCKET;
3097 _cleanup_free_ char *message = NULL;
3098 _cleanup_close_ int fd = -1;
3099 int n = 0;
3100
3101 /* Don't generate plymouth events if the service was already
3102 * started and we're just deserializing */
3103 if (MANAGER_IS_RELOADING(m))
3104 return;
3105
3106 if (!MANAGER_IS_SYSTEM(m))
3107 return;
3108
3109 if (detect_container() > 0)
3110 return;
3111
3112 if (!IN_SET(u->type, UNIT_SERVICE, UNIT_MOUNT, UNIT_SWAP))
3113 return;
3114
3115 /* We set SOCK_NONBLOCK here so that we rather drop the
3116 * message then wait for plymouth */
3117 fd = socket(AF_UNIX, SOCK_STREAM|SOCK_CLOEXEC|SOCK_NONBLOCK, 0);
3118 if (fd < 0) {
3119 log_error_errno(errno, "socket() failed: %m");
3120 return;
3121 }
3122
3123 if (connect(fd, &sa.sa, SOCKADDR_UN_LEN(sa.un)) < 0) {
3124 if (!IN_SET(errno, EAGAIN, ENOENT) && !ERRNO_IS_DISCONNECT(errno))
3125 log_error_errno(errno, "connect() failed: %m");
3126 return;
3127 }
3128
3129 if (asprintf(&message, "U\002%c%s%n", (int) (strlen(u->id) + 1), u->id, &n) < 0) {
3130 log_oom();
3131 return;
3132 }
3133
3134 errno = 0;
3135 if (write(fd, message, n + 1) != n + 1)
3136 if (!IN_SET(errno, EAGAIN, ENOENT) && !ERRNO_IS_DISCONNECT(errno))
3137 log_error_errno(errno, "Failed to write Plymouth message: %m");
3138 }
3139
3140 int manager_open_serialization(Manager *m, FILE **_f) {
3141 _cleanup_close_ int fd = -1;
3142 FILE *f;
3143
3144 assert(_f);
3145
3146 fd = open_serialization_fd("systemd-state");
3147 if (fd < 0)
3148 return fd;
3149
3150 f = take_fdopen(&fd, "w+");
3151 if (!f)
3152 return -errno;
3153
3154 *_f = f;
3155 return 0;
3156 }
3157
3158 static bool manager_timestamp_shall_serialize(ManagerTimestamp t) {
3159
3160 if (!in_initrd())
3161 return true;
3162
3163 /* The following timestamps only apply to the host system, hence only serialize them there */
3164 return !IN_SET(t,
3165 MANAGER_TIMESTAMP_USERSPACE, MANAGER_TIMESTAMP_FINISH,
3166 MANAGER_TIMESTAMP_SECURITY_START, MANAGER_TIMESTAMP_SECURITY_FINISH,
3167 MANAGER_TIMESTAMP_GENERATORS_START, MANAGER_TIMESTAMP_GENERATORS_FINISH,
3168 MANAGER_TIMESTAMP_UNITS_LOAD_START, MANAGER_TIMESTAMP_UNITS_LOAD_FINISH);
3169 }
3170
3171 #define DESTROY_IPC_FLAG (UINT32_C(1) << 31)
3172
3173 static void manager_serialize_uid_refs_internal(
3174 FILE *f,
3175 Hashmap *uid_refs,
3176 const char *field_name) {
3177
3178 void *p, *k;
3179
3180 assert(f);
3181 assert(field_name);
3182
3183 /* Serialize the UID reference table. Or actually, just the IPC destruction flag of it, as
3184 * the actual counter of it is better rebuild after a reload/reexec. */
3185
3186 HASHMAP_FOREACH_KEY(p, k, uid_refs) {
3187 uint32_t c;
3188 uid_t uid;
3189
3190 uid = PTR_TO_UID(k);
3191 c = PTR_TO_UINT32(p);
3192
3193 if (!(c & DESTROY_IPC_FLAG))
3194 continue;
3195
3196 (void) serialize_item_format(f, field_name, UID_FMT, uid);
3197 }
3198 }
3199
3200 static void manager_serialize_uid_refs(Manager *m, FILE *f) {
3201 manager_serialize_uid_refs_internal(f, m->uid_refs, "destroy-ipc-uid");
3202 }
3203
3204 static void manager_serialize_gid_refs(Manager *m, FILE *f) {
3205 manager_serialize_uid_refs_internal(f, m->gid_refs, "destroy-ipc-gid");
3206 }
3207
3208 int manager_serialize(
3209 Manager *m,
3210 FILE *f,
3211 FDSet *fds,
3212 bool switching_root) {
3213
3214 const char *t;
3215 Unit *u;
3216 int r;
3217
3218 assert(m);
3219 assert(f);
3220 assert(fds);
3221
3222 _cleanup_(manager_reloading_stopp) _unused_ Manager *reloading = manager_reloading_start(m);
3223
3224 (void) serialize_item_format(f, "current-job-id", "%" PRIu32, m->current_job_id);
3225 (void) serialize_item_format(f, "n-installed-jobs", "%u", m->n_installed_jobs);
3226 (void) serialize_item_format(f, "n-failed-jobs", "%u", m->n_failed_jobs);
3227 (void) serialize_bool(f, "taint-usr", m->taint_usr);
3228 (void) serialize_bool(f, "ready-sent", m->ready_sent);
3229 (void) serialize_bool(f, "taint-logged", m->taint_logged);
3230 (void) serialize_bool(f, "service-watchdogs", m->service_watchdogs);
3231
3232 /* After switching root, udevd has not been started yet. So, enumeration results should not be emitted. */
3233 (void) serialize_bool(f, "honor-device-enumeration", !switching_root);
3234
3235 if (m->show_status_overridden != _SHOW_STATUS_INVALID)
3236 (void) serialize_item(f, "show-status-overridden",
3237 show_status_to_string(m->show_status_overridden));
3238
3239 if (m->log_level_overridden)
3240 (void) serialize_item_format(f, "log-level-override", "%i", log_get_max_level());
3241 if (m->log_target_overridden)
3242 (void) serialize_item(f, "log-target-override", log_target_to_string(log_get_target()));
3243
3244 (void) serialize_usec(f, "runtime-watchdog-overridden", m->watchdog_overridden[WATCHDOG_RUNTIME]);
3245 (void) serialize_usec(f, "reboot-watchdog-overridden", m->watchdog_overridden[WATCHDOG_REBOOT]);
3246 (void) serialize_usec(f, "kexec-watchdog-overridden", m->watchdog_overridden[WATCHDOG_KEXEC]);
3247
3248 for (ManagerTimestamp q = 0; q < _MANAGER_TIMESTAMP_MAX; q++) {
3249 _cleanup_free_ char *joined = NULL;
3250
3251 if (!manager_timestamp_shall_serialize(q))
3252 continue;
3253
3254 joined = strjoin(manager_timestamp_to_string(q), "-timestamp");
3255 if (!joined)
3256 return log_oom();
3257
3258 (void) serialize_dual_timestamp(f, joined, m->timestamps + q);
3259 }
3260
3261 if (!switching_root)
3262 (void) serialize_strv(f, "env", m->client_environment);
3263
3264 if (m->notify_fd >= 0) {
3265 r = serialize_fd(f, fds, "notify-fd", m->notify_fd);
3266 if (r < 0)
3267 return r;
3268
3269 (void) serialize_item(f, "notify-socket", m->notify_socket);
3270 }
3271
3272 if (m->cgroups_agent_fd >= 0) {
3273 r = serialize_fd(f, fds, "cgroups-agent-fd", m->cgroups_agent_fd);
3274 if (r < 0)
3275 return r;
3276 }
3277
3278 if (m->user_lookup_fds[0] >= 0) {
3279 int copy0, copy1;
3280
3281 copy0 = fdset_put_dup(fds, m->user_lookup_fds[0]);
3282 if (copy0 < 0)
3283 return log_error_errno(copy0, "Failed to add user lookup fd to serialization: %m");
3284
3285 copy1 = fdset_put_dup(fds, m->user_lookup_fds[1]);
3286 if (copy1 < 0)
3287 return log_error_errno(copy1, "Failed to add user lookup fd to serialization: %m");
3288
3289 (void) serialize_item_format(f, "user-lookup", "%i %i", copy0, copy1);
3290 }
3291
3292 bus_track_serialize(m->subscribed, f, "subscribed");
3293
3294 r = dynamic_user_serialize(m, f, fds);
3295 if (r < 0)
3296 return r;
3297
3298 manager_serialize_uid_refs(m, f);
3299 manager_serialize_gid_refs(m, f);
3300
3301 r = exec_runtime_serialize(m, f, fds);
3302 if (r < 0)
3303 return r;
3304
3305 (void) fputc('\n', f);
3306
3307 HASHMAP_FOREACH_KEY(u, t, m->units) {
3308 if (u->id != t)
3309 continue;
3310
3311 /* Start marker */
3312 fputs(u->id, f);
3313 fputc('\n', f);
3314
3315 r = unit_serialize(u, f, fds, !switching_root);
3316 if (r < 0)
3317 return r;
3318 }
3319
3320 r = fflush_and_check(f);
3321 if (r < 0)
3322 return log_error_errno(r, "Failed to flush serialization: %m");
3323
3324 r = bus_fdset_add_all(m, fds);
3325 if (r < 0)
3326 return log_error_errno(r, "Failed to add bus sockets to serialization: %m");
3327
3328 return 0;
3329 }
3330
3331 static int manager_deserialize_one_unit(Manager *m, const char *name, FILE *f, FDSet *fds) {
3332 Unit *u;
3333 int r;
3334
3335 r = manager_load_unit(m, name, NULL, NULL, &u);
3336 if (r < 0) {
3337 if (r == -ENOMEM)
3338 return r;
3339 return log_notice_errno(r, "Failed to load unit \"%s\", skipping deserialization: %m", name);
3340 }
3341
3342 r = unit_deserialize(u, f, fds);
3343 if (r < 0) {
3344 if (r == -ENOMEM)
3345 return r;
3346 return log_notice_errno(r, "Failed to deserialize unit \"%s\", skipping: %m", name);
3347 }
3348
3349 return 0;
3350 }
3351
3352 static int manager_deserialize_units(Manager *m, FILE *f, FDSet *fds) {
3353 const char *unit_name;
3354 int r;
3355
3356 for (;;) {
3357 _cleanup_free_ char *line = NULL;
3358 /* Start marker */
3359 r = read_line(f, LONG_LINE_MAX, &line);
3360 if (r < 0)
3361 return log_error_errno(r, "Failed to read serialization line: %m");
3362 if (r == 0)
3363 break;
3364
3365 unit_name = strstrip(line);
3366
3367 r = manager_deserialize_one_unit(m, unit_name, f, fds);
3368 if (r == -ENOMEM)
3369 return r;
3370 if (r < 0) {
3371 r = unit_deserialize_skip(f);
3372 if (r < 0)
3373 return r;
3374 }
3375 }
3376
3377 return 0;
3378 }
3379
3380 usec_t manager_get_watchdog(Manager *m, WatchdogType t) {
3381 assert(m);
3382
3383 if (MANAGER_IS_USER(m))
3384 return USEC_INFINITY;
3385
3386 if (timestamp_is_set(m->watchdog_overridden[t]))
3387 return m->watchdog_overridden[t];
3388
3389 return m->watchdog[t];
3390 }
3391
3392 void manager_set_watchdog(Manager *m, WatchdogType t, usec_t timeout) {
3393 int r = 0;
3394
3395 assert(m);
3396
3397 if (MANAGER_IS_USER(m))
3398 return;
3399
3400 if (m->watchdog[t] == timeout)
3401 return;
3402
3403 if (t == WATCHDOG_RUNTIME)
3404 if (!timestamp_is_set(m->watchdog_overridden[WATCHDOG_RUNTIME])) {
3405 if (timestamp_is_set(timeout)) {
3406 r = watchdog_set_timeout(&timeout);
3407
3408 if (r >= 0)
3409 m->runtime_watchdog_running = true;
3410 } else {
3411 watchdog_close(true);
3412 m->runtime_watchdog_running = false;
3413 }
3414 }
3415
3416 m->watchdog[t] = timeout;
3417 }
3418
3419 int manager_override_watchdog(Manager *m, WatchdogType t, usec_t timeout) {
3420 int r = 0;
3421
3422 assert(m);
3423
3424 if (MANAGER_IS_USER(m))
3425 return 0;
3426
3427 if (m->watchdog_overridden[t] == timeout)
3428 return 0;
3429
3430 if (t == WATCHDOG_RUNTIME) {
3431 usec_t *p;
3432
3433 p = timestamp_is_set(timeout) ? &timeout : &m->watchdog[t];
3434 if (timestamp_is_set(*p)) {
3435 r = watchdog_set_timeout(p);
3436
3437 if (r >= 0)
3438 m->runtime_watchdog_running = true;
3439 } else {
3440 watchdog_close(true);
3441 m->runtime_watchdog_running = false;
3442 }
3443 }
3444
3445 m->watchdog_overridden[t] = timeout;
3446
3447 return 0;
3448 }
3449
3450 void manager_retry_runtime_watchdog(Manager *m) {
3451 int r = 0;
3452
3453 assert(m);
3454
3455 if (timestamp_is_set(m->watchdog_overridden[WATCHDOG_RUNTIME]))
3456 r = watchdog_set_timeout(&m->watchdog_overridden[WATCHDOG_RUNTIME]);
3457 else
3458 r = watchdog_set_timeout(&m->watchdog[WATCHDOG_RUNTIME]);
3459
3460 if (r >= 0)
3461 m->runtime_watchdog_running = true;
3462 }
3463
3464 static void manager_deserialize_uid_refs_one_internal(
3465 Hashmap** uid_refs,
3466 const char *value) {
3467
3468 uid_t uid;
3469 uint32_t c;
3470 int r;
3471
3472 assert(uid_refs);
3473 assert(value);
3474
3475 r = parse_uid(value, &uid);
3476 if (r < 0 || uid == 0) {
3477 log_debug("Unable to parse UID/GID reference serialization: " UID_FMT, uid);
3478 return;
3479 }
3480
3481 if (hashmap_ensure_allocated(uid_refs, &trivial_hash_ops) < 0) {
3482 log_oom();
3483 return;
3484 }
3485
3486 c = PTR_TO_UINT32(hashmap_get(*uid_refs, UID_TO_PTR(uid)));
3487 if (c & DESTROY_IPC_FLAG)
3488 return;
3489
3490 c |= DESTROY_IPC_FLAG;
3491
3492 r = hashmap_replace(*uid_refs, UID_TO_PTR(uid), UINT32_TO_PTR(c));
3493 if (r < 0) {
3494 log_debug_errno(r, "Failed to add UID/GID reference entry: %m");
3495 return;
3496 }
3497 }
3498
3499 static void manager_deserialize_uid_refs_one(Manager *m, const char *value) {
3500 manager_deserialize_uid_refs_one_internal(&m->uid_refs, value);
3501 }
3502
3503 static void manager_deserialize_gid_refs_one(Manager *m, const char *value) {
3504 manager_deserialize_uid_refs_one_internal(&m->gid_refs, value);
3505 }
3506
3507 int manager_deserialize(Manager *m, FILE *f, FDSet *fds) {
3508 int r = 0;
3509
3510 assert(m);
3511 assert(f);
3512
3513 if (DEBUG_LOGGING) {
3514 if (fdset_isempty(fds))
3515 log_debug("No file descriptors passed");
3516 else {
3517 int fd;
3518
3519 FDSET_FOREACH(fd, fds) {
3520 _cleanup_free_ char *fn = NULL;
3521
3522 r = fd_get_path(fd, &fn);
3523 if (r < 0)
3524 log_debug_errno(r, "Received serialized fd %i → %m", fd);
3525 else
3526 log_debug("Received serialized fd %i → %s", fd, strna(fn));
3527 }
3528 }
3529 }
3530
3531 log_debug("Deserializing state...");
3532
3533 /* If we are not in reload mode yet, enter it now. Not that this is recursive, a caller might already have
3534 * increased it to non-zero, which is why we just increase it by one here and down again at the end of this
3535 * call. */
3536 _cleanup_(manager_reloading_stopp) _unused_ Manager *reloading = manager_reloading_start(m);
3537
3538 for (;;) {
3539 _cleanup_free_ char *line = NULL;
3540 const char *val, *l;
3541
3542 r = read_line(f, LONG_LINE_MAX, &line);
3543 if (r < 0)
3544 return log_error_errno(r, "Failed to read serialization line: %m");
3545 if (r == 0)
3546 break;
3547
3548 l = strstrip(line);
3549 if (isempty(l)) /* end marker */
3550 break;
3551
3552 if ((val = startswith(l, "current-job-id="))) {
3553 uint32_t id;
3554
3555 if (safe_atou32(val, &id) < 0)
3556 log_notice("Failed to parse current job id value '%s', ignoring.", val);
3557 else
3558 m->current_job_id = MAX(m->current_job_id, id);
3559
3560 } else if ((val = startswith(l, "n-installed-jobs="))) {
3561 uint32_t n;
3562
3563 if (safe_atou32(val, &n) < 0)
3564 log_notice("Failed to parse installed jobs counter '%s', ignoring.", val);
3565 else
3566 m->n_installed_jobs += n;
3567
3568 } else if ((val = startswith(l, "n-failed-jobs="))) {
3569 uint32_t n;
3570
3571 if (safe_atou32(val, &n) < 0)
3572 log_notice("Failed to parse failed jobs counter '%s', ignoring.", val);
3573 else
3574 m->n_failed_jobs += n;
3575
3576 } else if ((val = startswith(l, "taint-usr="))) {
3577 int b;
3578
3579 b = parse_boolean(val);
3580 if (b < 0)
3581 log_notice("Failed to parse taint /usr flag '%s', ignoring.", val);
3582 else
3583 m->taint_usr = m->taint_usr || b;
3584
3585 } else if ((val = startswith(l, "ready-sent="))) {
3586 int b;
3587
3588 b = parse_boolean(val);
3589 if (b < 0)
3590 log_notice("Failed to parse ready-sent flag '%s', ignoring.", val);
3591 else
3592 m->ready_sent = m->ready_sent || b;
3593
3594 } else if ((val = startswith(l, "taint-logged="))) {
3595 int b;
3596
3597 b = parse_boolean(val);
3598 if (b < 0)
3599 log_notice("Failed to parse taint-logged flag '%s', ignoring.", val);
3600 else
3601 m->taint_logged = m->taint_logged || b;
3602
3603 } else if ((val = startswith(l, "service-watchdogs="))) {
3604 int b;
3605
3606 b = parse_boolean(val);
3607 if (b < 0)
3608 log_notice("Failed to parse service-watchdogs flag '%s', ignoring.", val);
3609 else
3610 m->service_watchdogs = b;
3611
3612 } else if ((val = startswith(l, "honor-device-enumeration="))) {
3613 int b;
3614
3615 b = parse_boolean(val);
3616 if (b < 0)
3617 log_notice("Failed to parse honor-device-enumeration flag '%s', ignoring.", val);
3618 else
3619 m->honor_device_enumeration = b;
3620
3621 } else if ((val = startswith(l, "show-status-overridden="))) {
3622 ShowStatus s;
3623
3624 s = show_status_from_string(val);
3625 if (s < 0)
3626 log_notice("Failed to parse show-status-overridden flag '%s', ignoring.", val);
3627 else
3628 manager_override_show_status(m, s, "deserialize");
3629
3630 } else if ((val = startswith(l, "log-level-override="))) {
3631 int level;
3632
3633 level = log_level_from_string(val);
3634 if (level < 0)
3635 log_notice("Failed to parse log-level-override value '%s', ignoring.", val);
3636 else
3637 manager_override_log_level(m, level);
3638
3639 } else if ((val = startswith(l, "log-target-override="))) {
3640 LogTarget target;
3641
3642 target = log_target_from_string(val);
3643 if (target < 0)
3644 log_notice("Failed to parse log-target-override value '%s', ignoring.", val);
3645 else
3646 manager_override_log_target(m, target);
3647
3648 } else if ((val = startswith(l, "runtime-watchdog-overridden="))) {
3649 usec_t t;
3650
3651 if (deserialize_usec(val, &t) < 0)
3652 log_notice("Failed to parse runtime-watchdog-overridden value '%s', ignoring.", val);
3653 else
3654 manager_override_watchdog(m, WATCHDOG_RUNTIME, t);
3655
3656 } else if ((val = startswith(l, "reboot-watchdog-overridden="))) {
3657 usec_t t;
3658
3659 if (deserialize_usec(val, &t) < 0)
3660 log_notice("Failed to parse reboot-watchdog-overridden value '%s', ignoring.", val);
3661 else
3662 manager_override_watchdog(m, WATCHDOG_REBOOT, t);
3663
3664 } else if ((val = startswith(l, "kexec-watchdog-overridden="))) {
3665 usec_t t;
3666
3667 if (deserialize_usec(val, &t) < 0)
3668 log_notice("Failed to parse kexec-watchdog-overridden value '%s', ignoring.", val);
3669 else
3670 manager_override_watchdog(m, WATCHDOG_KEXEC, t);
3671
3672 } else if (startswith(l, "env=")) {
3673 r = deserialize_environment(l + 4, &m->client_environment);
3674 if (r < 0)
3675 log_notice_errno(r, "Failed to parse environment entry: \"%s\", ignoring: %m", l);
3676
3677 } else if ((val = startswith(l, "notify-fd="))) {
3678 int fd;
3679
3680 if (safe_atoi(val, &fd) < 0 || fd < 0 || !fdset_contains(fds, fd))
3681 log_notice("Failed to parse notify fd, ignoring: \"%s\"", val);
3682 else {
3683 m->notify_event_source = sd_event_source_unref(m->notify_event_source);
3684 safe_close(m->notify_fd);
3685 m->notify_fd = fdset_remove(fds, fd);
3686 }
3687
3688 } else if ((val = startswith(l, "notify-socket="))) {
3689 r = free_and_strdup(&m->notify_socket, val);
3690 if (r < 0)
3691 return r;
3692
3693 } else if ((val = startswith(l, "cgroups-agent-fd="))) {
3694 int fd;
3695
3696 if (safe_atoi(val, &fd) < 0 || fd < 0 || !fdset_contains(fds, fd))
3697 log_notice("Failed to parse cgroups agent fd, ignoring.: %s", val);
3698 else {
3699 m->cgroups_agent_event_source = sd_event_source_unref(m->cgroups_agent_event_source);
3700 safe_close(m->cgroups_agent_fd);
3701 m->cgroups_agent_fd = fdset_remove(fds, fd);
3702 }
3703
3704 } else if ((val = startswith(l, "user-lookup="))) {
3705 int fd0, fd1;
3706
3707 if (sscanf(val, "%i %i", &fd0, &fd1) != 2 || fd0 < 0 || fd1 < 0 || fd0 == fd1 || !fdset_contains(fds, fd0) || !fdset_contains(fds, fd1))
3708 log_notice("Failed to parse user lookup fd, ignoring: %s", val);
3709 else {
3710 m->user_lookup_event_source = sd_event_source_unref(m->user_lookup_event_source);
3711 safe_close_pair(m->user_lookup_fds);
3712 m->user_lookup_fds[0] = fdset_remove(fds, fd0);
3713 m->user_lookup_fds[1] = fdset_remove(fds, fd1);
3714 }
3715
3716 } else if ((val = startswith(l, "dynamic-user=")))
3717 dynamic_user_deserialize_one(m, val, fds);
3718 else if ((val = startswith(l, "destroy-ipc-uid=")))
3719 manager_deserialize_uid_refs_one(m, val);
3720 else if ((val = startswith(l, "destroy-ipc-gid=")))
3721 manager_deserialize_gid_refs_one(m, val);
3722 else if ((val = startswith(l, "exec-runtime=")))
3723 (void) exec_runtime_deserialize_one(m, val, fds);
3724 else if ((val = startswith(l, "subscribed="))) {
3725
3726 if (strv_extend(&m->deserialized_subscribed, val) < 0)
3727 return -ENOMEM;
3728
3729 } else {
3730 ManagerTimestamp q;
3731
3732 for (q = 0; q < _MANAGER_TIMESTAMP_MAX; q++) {
3733 val = startswith(l, manager_timestamp_to_string(q));
3734 if (!val)
3735 continue;
3736
3737 val = startswith(val, "-timestamp=");
3738 if (val)
3739 break;
3740 }
3741
3742 if (q < _MANAGER_TIMESTAMP_MAX) /* found it */
3743 (void) deserialize_dual_timestamp(val, m->timestamps + q);
3744 else if (!startswith(l, "kdbus-fd=")) /* ignore kdbus */
3745 log_notice("Unknown serialization item '%s', ignoring.", l);
3746 }
3747 }
3748
3749 return manager_deserialize_units(m, f, fds);
3750 }
3751
3752 int manager_reload(Manager *m) {
3753 _cleanup_(manager_reloading_stopp) Manager *reloading = NULL;
3754 _cleanup_fdset_free_ FDSet *fds = NULL;
3755 _cleanup_fclose_ FILE *f = NULL;
3756 int r;
3757
3758 assert(m);
3759
3760 r = manager_open_serialization(m, &f);
3761 if (r < 0)
3762 return log_error_errno(r, "Failed to create serialization file: %m");
3763
3764 fds = fdset_new();
3765 if (!fds)
3766 return log_oom();
3767
3768 /* We are officially in reload mode from here on. */
3769 reloading = manager_reloading_start(m);
3770
3771 r = manager_serialize(m, f, fds, false);
3772 if (r < 0)
3773 return r;
3774
3775 if (fseeko(f, 0, SEEK_SET) < 0)
3776 return log_error_errno(errno, "Failed to seek to beginning of serialization: %m");
3777
3778 /* 💀 This is the point of no return, from here on there is no way back. 💀 */
3779 reloading = NULL;
3780
3781 bus_manager_send_reloading(m, true);
3782
3783 /* Start by flushing out all jobs and units, all generated units, all runtime environments, all dynamic users
3784 * and everything else that is worth flushing out. We'll get it all back from the serialization — if we need
3785 * it.*/
3786
3787 manager_clear_jobs_and_units(m);
3788 lookup_paths_flush_generator(&m->lookup_paths);
3789 lookup_paths_free(&m->lookup_paths);
3790 exec_runtime_vacuum(m);
3791 dynamic_user_vacuum(m, false);
3792 m->uid_refs = hashmap_free(m->uid_refs);
3793 m->gid_refs = hashmap_free(m->gid_refs);
3794
3795 r = lookup_paths_init(&m->lookup_paths, m->unit_file_scope, 0, NULL);
3796 if (r < 0)
3797 log_warning_errno(r, "Failed to initialize path lookup table, ignoring: %m");
3798
3799 (void) manager_run_environment_generators(m);
3800 (void) manager_run_generators(m);
3801
3802 lookup_paths_log(&m->lookup_paths);
3803
3804 /* We flushed out generated files, for which we don't watch mtime, so we should flush the old map. */
3805 manager_free_unit_name_maps(m);
3806
3807 /* First, enumerate what we can from kernel and suchlike */
3808 manager_enumerate_perpetual(m);
3809 manager_enumerate(m);
3810
3811 /* Second, deserialize our stored data */
3812 r = manager_deserialize(m, f, fds);
3813 if (r < 0)
3814 log_warning_errno(r, "Deserialization failed, proceeding anyway: %m");
3815
3816 /* We don't need the serialization anymore */
3817 f = safe_fclose(f);
3818
3819 /* Re-register notify_fd as event source, and set up other sockets/communication channels we might need */
3820 (void) manager_setup_notify(m);
3821 (void) manager_setup_cgroups_agent(m);
3822 (void) manager_setup_user_lookup_fd(m);
3823
3824 /* Third, fire things up! */
3825 manager_coldplug(m);
3826
3827 /* Clean up runtime objects no longer referenced */
3828 manager_vacuum(m);
3829
3830 /* Clean up deserialized tracked clients */
3831 m->deserialized_subscribed = strv_free(m->deserialized_subscribed);
3832
3833 /* Consider the reload process complete now. */
3834 assert(m->n_reloading > 0);
3835 m->n_reloading--;
3836
3837 /* On manager reloading, device tag data should exists, thus, we should honor the results of device
3838 * enumeration. The flag should be always set correctly by the serialized data, but it may fail. So,
3839 * let's always set the flag here for safety. */
3840 m->honor_device_enumeration = true;
3841
3842 manager_ready(m);
3843
3844 m->send_reloading_done = true;
3845 return 0;
3846 }
3847
3848 void manager_reset_failed(Manager *m) {
3849 Unit *u;
3850
3851 assert(m);
3852
3853 HASHMAP_FOREACH(u, m->units)
3854 unit_reset_failed(u);
3855 }
3856
3857 bool manager_unit_inactive_or_pending(Manager *m, const char *name) {
3858 Unit *u;
3859
3860 assert(m);
3861 assert(name);
3862
3863 /* Returns true if the unit is inactive or going down */
3864 u = manager_get_unit(m, name);
3865 if (!u)
3866 return true;
3867
3868 return unit_inactive_or_pending(u);
3869 }
3870
3871 static void log_taint_string(Manager *m) {
3872 _cleanup_free_ char *taint = NULL;
3873
3874 assert(m);
3875
3876 if (MANAGER_IS_USER(m) || m->taint_logged)
3877 return;
3878
3879 m->taint_logged = true; /* only check for taint once */
3880
3881 taint = manager_taint_string(m);
3882 if (isempty(taint))
3883 return;
3884
3885 log_struct(LOG_NOTICE,
3886 LOG_MESSAGE("System is tainted: %s", taint),
3887 "TAINT=%s", taint,
3888 "MESSAGE_ID=" SD_MESSAGE_TAINTED_STR);
3889 }
3890
3891 static void manager_notify_finished(Manager *m) {
3892 char userspace[FORMAT_TIMESPAN_MAX], initrd[FORMAT_TIMESPAN_MAX], kernel[FORMAT_TIMESPAN_MAX], sum[FORMAT_TIMESPAN_MAX];
3893 usec_t firmware_usec, loader_usec, kernel_usec, initrd_usec, userspace_usec, total_usec;
3894
3895 if (MANAGER_IS_TEST_RUN(m))
3896 return;
3897
3898 if (MANAGER_IS_SYSTEM(m) && detect_container() <= 0) {
3899 char ts[FORMAT_TIMESPAN_MAX];
3900 char buf[FORMAT_TIMESPAN_MAX + STRLEN(" (firmware) + ") + FORMAT_TIMESPAN_MAX + STRLEN(" (loader) + ")]
3901 = {};
3902 char *p = buf;
3903 size_t size = sizeof buf;
3904
3905 /* Note that MANAGER_TIMESTAMP_KERNEL's monotonic value is always at 0, and
3906 * MANAGER_TIMESTAMP_FIRMWARE's and MANAGER_TIMESTAMP_LOADER's monotonic value should be considered
3907 * negative values. */
3908
3909 firmware_usec = m->timestamps[MANAGER_TIMESTAMP_FIRMWARE].monotonic - m->timestamps[MANAGER_TIMESTAMP_LOADER].monotonic;
3910 loader_usec = m->timestamps[MANAGER_TIMESTAMP_LOADER].monotonic - m->timestamps[MANAGER_TIMESTAMP_KERNEL].monotonic;
3911 userspace_usec = m->timestamps[MANAGER_TIMESTAMP_FINISH].monotonic - m->timestamps[MANAGER_TIMESTAMP_USERSPACE].monotonic;
3912 total_usec = m->timestamps[MANAGER_TIMESTAMP_FIRMWARE].monotonic + m->timestamps[MANAGER_TIMESTAMP_FINISH].monotonic;
3913
3914 if (firmware_usec > 0)
3915 size = strpcpyf(&p, size, "%s (firmware) + ", format_timespan(ts, sizeof(ts), firmware_usec, USEC_PER_MSEC));
3916 if (loader_usec > 0)
3917 size = strpcpyf(&p, size, "%s (loader) + ", format_timespan(ts, sizeof(ts), loader_usec, USEC_PER_MSEC));
3918
3919 if (dual_timestamp_is_set(&m->timestamps[MANAGER_TIMESTAMP_INITRD])) {
3920
3921 /* The initrd case on bare-metal*/
3922 kernel_usec = m->timestamps[MANAGER_TIMESTAMP_INITRD].monotonic - m->timestamps[MANAGER_TIMESTAMP_KERNEL].monotonic;
3923 initrd_usec = m->timestamps[MANAGER_TIMESTAMP_USERSPACE].monotonic - m->timestamps[MANAGER_TIMESTAMP_INITRD].monotonic;
3924
3925 log_struct(LOG_INFO,
3926 "MESSAGE_ID=" SD_MESSAGE_STARTUP_FINISHED_STR,
3927 "KERNEL_USEC="USEC_FMT, kernel_usec,
3928 "INITRD_USEC="USEC_FMT, initrd_usec,
3929 "USERSPACE_USEC="USEC_FMT, userspace_usec,
3930 LOG_MESSAGE("Startup finished in %s%s (kernel) + %s (initrd) + %s (userspace) = %s.",
3931 buf,
3932 format_timespan(kernel, sizeof(kernel), kernel_usec, USEC_PER_MSEC),
3933 format_timespan(initrd, sizeof(initrd), initrd_usec, USEC_PER_MSEC),
3934 format_timespan(userspace, sizeof(userspace), userspace_usec, USEC_PER_MSEC),
3935 format_timespan(sum, sizeof(sum), total_usec, USEC_PER_MSEC)));
3936 } else {
3937 /* The initrd-less case on bare-metal*/
3938
3939 kernel_usec = m->timestamps[MANAGER_TIMESTAMP_USERSPACE].monotonic - m->timestamps[MANAGER_TIMESTAMP_KERNEL].monotonic;
3940 initrd_usec = 0;
3941
3942 log_struct(LOG_INFO,
3943 "MESSAGE_ID=" SD_MESSAGE_STARTUP_FINISHED_STR,
3944 "KERNEL_USEC="USEC_FMT, kernel_usec,
3945 "USERSPACE_USEC="USEC_FMT, userspace_usec,
3946 LOG_MESSAGE("Startup finished in %s%s (kernel) + %s (userspace) = %s.",
3947 buf,
3948 format_timespan(kernel, sizeof(kernel), kernel_usec, USEC_PER_MSEC),
3949 format_timespan(userspace, sizeof(userspace), userspace_usec, USEC_PER_MSEC),
3950 format_timespan(sum, sizeof(sum), total_usec, USEC_PER_MSEC)));
3951 }
3952 } else {
3953 /* The container and --user case */
3954 firmware_usec = loader_usec = initrd_usec = kernel_usec = 0;
3955 total_usec = userspace_usec = m->timestamps[MANAGER_TIMESTAMP_FINISH].monotonic - m->timestamps[MANAGER_TIMESTAMP_USERSPACE].monotonic;
3956
3957 log_struct(LOG_INFO,
3958 "MESSAGE_ID=" SD_MESSAGE_USER_STARTUP_FINISHED_STR,
3959 "USERSPACE_USEC="USEC_FMT, userspace_usec,
3960 LOG_MESSAGE("Startup finished in %s.",
3961 format_timespan(sum, sizeof(sum), total_usec, USEC_PER_MSEC)));
3962 }
3963
3964 bus_manager_send_finished(m, firmware_usec, loader_usec, kernel_usec, initrd_usec, userspace_usec, total_usec);
3965
3966 sd_notifyf(false,
3967 m->ready_sent ? "STATUS=Startup finished in %s."
3968 : "READY=1\n"
3969 "STATUS=Startup finished in %s.",
3970 format_timespan(sum, sizeof(sum), total_usec, USEC_PER_MSEC));
3971 m->ready_sent = true;
3972
3973 log_taint_string(m);
3974 }
3975
3976 static void manager_send_ready(Manager *m) {
3977 assert(m);
3978
3979 /* We send READY=1 on reaching basic.target only when running in --user mode. */
3980 if (!MANAGER_IS_USER(m) || m->ready_sent)
3981 return;
3982
3983 m->ready_sent = true;
3984
3985 sd_notifyf(false,
3986 "READY=1\n"
3987 "STATUS=Reached " SPECIAL_BASIC_TARGET ".");
3988 }
3989
3990 static void manager_check_basic_target(Manager *m) {
3991 Unit *u;
3992
3993 assert(m);
3994
3995 /* Small shortcut */
3996 if (m->ready_sent && m->taint_logged)
3997 return;
3998
3999 u = manager_get_unit(m, SPECIAL_BASIC_TARGET);
4000 if (!u || !UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u)))
4001 return;
4002
4003 /* For user managers, send out READY=1 as soon as we reach basic.target */
4004 manager_send_ready(m);
4005
4006 /* Log the taint string as soon as we reach basic.target */
4007 log_taint_string(m);
4008 }
4009
4010 void manager_check_finished(Manager *m) {
4011 assert(m);
4012
4013 if (MANAGER_IS_RELOADING(m))
4014 return;
4015
4016 /* Verify that we have entered the event loop already, and not left it again. */
4017 if (!MANAGER_IS_RUNNING(m))
4018 return;
4019
4020 manager_check_basic_target(m);
4021
4022 if (hashmap_size(m->jobs) > 0) {
4023 if (m->jobs_in_progress_event_source)
4024 /* Ignore any failure, this is only for feedback */
4025 (void) sd_event_source_set_time(m->jobs_in_progress_event_source,
4026 manager_watch_jobs_next_time(m));
4027 return;
4028 }
4029
4030 /* The jobs hashmap tends to grow a lot during boot, and then it's not reused until shutdown. Let's
4031 kill the hashmap if it is relatively large. */
4032 if (hashmap_buckets(m->jobs) > hashmap_size(m->units) / 10)
4033 m->jobs = hashmap_free(m->jobs);
4034
4035 manager_flip_auto_status(m, false, "boot finished");
4036
4037 /* Notify Type=idle units that we are done now */
4038 manager_close_idle_pipe(m);
4039
4040 /* Turn off confirm spawn now */
4041 m->confirm_spawn = NULL;
4042
4043 /* No need to update ask password status when we're going non-interactive */
4044 manager_close_ask_password(m);
4045
4046 /* This is no longer the first boot */
4047 manager_set_first_boot(m, false);
4048
4049 if (MANAGER_IS_FINISHED(m))
4050 return;
4051
4052 dual_timestamp_get(m->timestamps + MANAGER_TIMESTAMP_FINISH);
4053
4054 manager_notify_finished(m);
4055
4056 manager_invalidate_startup_units(m);
4057 }
4058
4059 static bool generator_path_any(const char* const* paths) {
4060 char **path;
4061 bool found = false;
4062
4063 /* Optimize by skipping the whole process by not creating output directories
4064 * if no generators are found. */
4065 STRV_FOREACH(path, (char**) paths)
4066 if (access(*path, F_OK) == 0)
4067 found = true;
4068 else if (errno != ENOENT)
4069 log_warning_errno(errno, "Failed to open generator directory %s: %m", *path);
4070
4071 return found;
4072 }
4073
4074 static int manager_run_environment_generators(Manager *m) {
4075 char **tmp = NULL; /* this is only used in the forked process, no cleanup here */
4076 _cleanup_strv_free_ char **paths = NULL;
4077 void* args[] = {
4078 [STDOUT_GENERATE] = &tmp,
4079 [STDOUT_COLLECT] = &tmp,
4080 [STDOUT_CONSUME] = &m->transient_environment,
4081 };
4082 int r;
4083
4084 if (MANAGER_IS_TEST_RUN(m) && !(m->test_run_flags & MANAGER_TEST_RUN_ENV_GENERATORS))
4085 return 0;
4086
4087 paths = env_generator_binary_paths(MANAGER_IS_SYSTEM(m));
4088 if (!paths)
4089 return log_oom();
4090
4091 if (!generator_path_any((const char* const*) paths))
4092 return 0;
4093
4094 RUN_WITH_UMASK(0022)
4095 r = execute_directories((const char* const*) paths, DEFAULT_TIMEOUT_USEC, gather_environment,
4096 args, NULL, m->transient_environment,
4097 EXEC_DIR_PARALLEL | EXEC_DIR_IGNORE_ERRORS | EXEC_DIR_SET_SYSTEMD_EXEC_PID);
4098 return r;
4099 }
4100
4101 static int manager_run_generators(Manager *m) {
4102 _cleanup_strv_free_ char **paths = NULL;
4103 const char *argv[5];
4104 int r;
4105
4106 assert(m);
4107
4108 if (MANAGER_IS_TEST_RUN(m) && !(m->test_run_flags & MANAGER_TEST_RUN_GENERATORS))
4109 return 0;
4110
4111 paths = generator_binary_paths(m->unit_file_scope);
4112 if (!paths)
4113 return log_oom();
4114
4115 if (!generator_path_any((const char* const*) paths))
4116 return 0;
4117
4118 r = lookup_paths_mkdir_generator(&m->lookup_paths);
4119 if (r < 0) {
4120 log_error_errno(r, "Failed to create generator directories: %m");
4121 goto finish;
4122 }
4123
4124 argv[0] = NULL; /* Leave this empty, execute_directory() will fill something in */
4125 argv[1] = m->lookup_paths.generator;
4126 argv[2] = m->lookup_paths.generator_early;
4127 argv[3] = m->lookup_paths.generator_late;
4128 argv[4] = NULL;
4129
4130 RUN_WITH_UMASK(0022)
4131 (void) execute_directories((const char* const*) paths, DEFAULT_TIMEOUT_USEC, NULL, NULL,
4132 (char**) argv, m->transient_environment,
4133 EXEC_DIR_PARALLEL | EXEC_DIR_IGNORE_ERRORS | EXEC_DIR_SET_SYSTEMD_EXEC_PID);
4134
4135 r = 0;
4136
4137 finish:
4138 lookup_paths_trim_generator(&m->lookup_paths);
4139 return r;
4140 }
4141
4142 int manager_transient_environment_add(Manager *m, char **plus) {
4143 char **a;
4144
4145 assert(m);
4146
4147 if (strv_isempty(plus))
4148 return 0;
4149
4150 a = strv_env_merge(2, m->transient_environment, plus);
4151 if (!a)
4152 return log_oom();
4153
4154 sanitize_environment(a);
4155
4156 return strv_free_and_replace(m->transient_environment, a);
4157 }
4158
4159 int manager_client_environment_modify(
4160 Manager *m,
4161 char **minus,
4162 char **plus) {
4163
4164 char **a = NULL, **b = NULL, **l;
4165
4166 assert(m);
4167
4168 if (strv_isempty(minus) && strv_isempty(plus))
4169 return 0;
4170
4171 l = m->client_environment;
4172
4173 if (!strv_isempty(minus)) {
4174 a = strv_env_delete(l, 1, minus);
4175 if (!a)
4176 return -ENOMEM;
4177
4178 l = a;
4179 }
4180
4181 if (!strv_isempty(plus)) {
4182 b = strv_env_merge(2, l, plus);
4183 if (!b) {
4184 strv_free(a);
4185 return -ENOMEM;
4186 }
4187
4188 l = b;
4189 }
4190
4191 if (m->client_environment != l)
4192 strv_free(m->client_environment);
4193
4194 if (a != l)
4195 strv_free(a);
4196 if (b != l)
4197 strv_free(b);
4198
4199 m->client_environment = sanitize_environment(l);
4200 return 0;
4201 }
4202
4203 int manager_get_effective_environment(Manager *m, char ***ret) {
4204 char **l;
4205
4206 assert(m);
4207 assert(ret);
4208
4209 l = strv_env_merge(2, m->transient_environment, m->client_environment);
4210 if (!l)
4211 return -ENOMEM;
4212
4213 *ret = l;
4214 return 0;
4215 }
4216
4217 int manager_set_default_rlimits(Manager *m, struct rlimit **default_rlimit) {
4218 assert(m);
4219
4220 for (unsigned i = 0; i < _RLIMIT_MAX; i++) {
4221 m->rlimit[i] = mfree(m->rlimit[i]);
4222
4223 if (!default_rlimit[i])
4224 continue;
4225
4226 m->rlimit[i] = newdup(struct rlimit, default_rlimit[i], 1);
4227 if (!m->rlimit[i])
4228 return log_oom();
4229 }
4230
4231 return 0;
4232 }
4233
4234 void manager_recheck_dbus(Manager *m) {
4235 assert(m);
4236
4237 /* Connects to the bus if the dbus service and socket are running. If we are running in user mode this is all
4238 * it does. In system mode we'll also connect to the system bus (which will most likely just reuse the
4239 * connection of the API bus). That's because the system bus after all runs as service of the system instance,
4240 * while in the user instance we can assume it's already there. */
4241
4242 if (MANAGER_IS_RELOADING(m))
4243 return; /* don't check while we are reloading… */
4244
4245 if (manager_dbus_is_running(m, false)) {
4246 (void) bus_init_api(m);
4247
4248 if (MANAGER_IS_SYSTEM(m))
4249 (void) bus_init_system(m);
4250 } else {
4251 (void) bus_done_api(m);
4252
4253 if (MANAGER_IS_SYSTEM(m))
4254 (void) bus_done_system(m);
4255 }
4256 }
4257
4258 static bool manager_journal_is_running(Manager *m) {
4259 Unit *u;
4260
4261 assert(m);
4262
4263 if (MANAGER_IS_TEST_RUN(m))
4264 return false;
4265
4266 /* If we are the user manager we can safely assume that the journal is up */
4267 if (!MANAGER_IS_SYSTEM(m))
4268 return true;
4269
4270 /* Check that the socket is not only up, but in RUNNING state */
4271 u = manager_get_unit(m, SPECIAL_JOURNALD_SOCKET);
4272 if (!u)
4273 return false;
4274 if (SOCKET(u)->state != SOCKET_RUNNING)
4275 return false;
4276
4277 /* Similar, check if the daemon itself is fully up, too */
4278 u = manager_get_unit(m, SPECIAL_JOURNALD_SERVICE);
4279 if (!u)
4280 return false;
4281 if (!IN_SET(SERVICE(u)->state, SERVICE_RELOAD, SERVICE_RUNNING))
4282 return false;
4283
4284 return true;
4285 }
4286
4287 void disable_printk_ratelimit(void) {
4288 /* Disable kernel's printk ratelimit.
4289 *
4290 * Logging to /dev/kmsg is most useful during early boot and shutdown, where normal logging
4291 * mechanisms are not available. The semantics of this sysctl are such that any kernel command-line
4292 * setting takes precedence. */
4293 int r;
4294
4295 r = sysctl_write("kernel/printk_devkmsg", "on");
4296 if (r < 0)
4297 log_debug_errno(r, "Failed to set sysctl kernel.printk_devkmsg=on: %m");
4298 }
4299
4300 void manager_recheck_journal(Manager *m) {
4301
4302 assert(m);
4303
4304 /* Don't bother with this unless we are in the special situation of being PID 1 */
4305 if (getpid_cached() != 1)
4306 return;
4307
4308 /* Don't check this while we are reloading, things might still change */
4309 if (MANAGER_IS_RELOADING(m))
4310 return;
4311
4312 /* The journal is fully and entirely up? If so, let's permit logging to it, if that's configured. If the
4313 * journal is down, don't ever log to it, otherwise we might end up deadlocking ourselves as we might trigger
4314 * an activation ourselves we can't fulfill. */
4315 log_set_prohibit_ipc(!manager_journal_is_running(m));
4316 log_open();
4317 }
4318
4319 static ShowStatus manager_get_show_status(Manager *m) {
4320 assert(m);
4321
4322 if (MANAGER_IS_USER(m))
4323 return _SHOW_STATUS_INVALID;
4324
4325 if (m->show_status_overridden != _SHOW_STATUS_INVALID)
4326 return m->show_status_overridden;
4327
4328 return m->show_status;
4329 }
4330
4331 bool manager_get_show_status_on(Manager *m) {
4332 assert(m);
4333
4334 return show_status_on(manager_get_show_status(m));
4335 }
4336
4337 static void set_show_status_marker(bool b) {
4338 if (b)
4339 (void) touch("/run/systemd/show-status");
4340 else
4341 (void) unlink("/run/systemd/show-status");
4342 }
4343
4344 void manager_set_show_status(Manager *m, ShowStatus mode, const char *reason) {
4345 assert(m);
4346 assert(reason);
4347 assert(mode >= 0 && mode < _SHOW_STATUS_MAX);
4348
4349 if (MANAGER_IS_USER(m))
4350 return;
4351
4352 if (mode == m->show_status)
4353 return;
4354
4355 if (m->show_status_overridden == _SHOW_STATUS_INVALID) {
4356 bool enabled;
4357
4358 enabled = show_status_on(mode);
4359 log_debug("%s (%s) showing of status (%s).",
4360 enabled ? "Enabling" : "Disabling",
4361 strna(show_status_to_string(mode)),
4362 reason);
4363
4364 set_show_status_marker(enabled);
4365 }
4366
4367 m->show_status = mode;
4368 }
4369
4370 void manager_override_show_status(Manager *m, ShowStatus mode, const char *reason) {
4371 assert(m);
4372 assert(mode < _SHOW_STATUS_MAX);
4373
4374 if (MANAGER_IS_USER(m))
4375 return;
4376
4377 if (mode == m->show_status_overridden)
4378 return;
4379
4380 m->show_status_overridden = mode;
4381
4382 if (mode == _SHOW_STATUS_INVALID)
4383 mode = m->show_status;
4384
4385 log_debug("%s (%s) showing of status (%s).",
4386 m->show_status_overridden != _SHOW_STATUS_INVALID ? "Overriding" : "Restoring",
4387 strna(show_status_to_string(mode)),
4388 reason);
4389
4390 set_show_status_marker(show_status_on(mode));
4391 }
4392
4393 const char *manager_get_confirm_spawn(Manager *m) {
4394 static int last_errno = 0;
4395 struct stat st;
4396 int r;
4397
4398 assert(m);
4399
4400 /* Here's the deal: we want to test the validity of the console but don't want
4401 * PID1 to go through the whole console process which might block. But we also
4402 * want to warn the user only once if something is wrong with the console so we
4403 * cannot do the sanity checks after spawning our children. So here we simply do
4404 * really basic tests to hopefully trap common errors.
4405 *
4406 * If the console suddenly disappear at the time our children will really it
4407 * then they will simply fail to acquire it and a positive answer will be
4408 * assumed. New children will fall back to /dev/console though.
4409 *
4410 * Note: TTYs are devices that can come and go any time, and frequently aren't
4411 * available yet during early boot (consider a USB rs232 dongle...). If for any
4412 * reason the configured console is not ready, we fall back to the default
4413 * console. */
4414
4415 if (!m->confirm_spawn || path_equal(m->confirm_spawn, "/dev/console"))
4416 return m->confirm_spawn;
4417
4418 if (stat(m->confirm_spawn, &st) < 0) {
4419 r = -errno;
4420 goto fail;
4421 }
4422
4423 if (!S_ISCHR(st.st_mode)) {
4424 r = -ENOTTY;
4425 goto fail;
4426 }
4427
4428 last_errno = 0;
4429 return m->confirm_spawn;
4430
4431 fail:
4432 if (last_errno != r)
4433 last_errno = log_warning_errno(r, "Failed to open %s, using default console: %m", m->confirm_spawn);
4434
4435 return "/dev/console";
4436 }
4437
4438 void manager_set_first_boot(Manager *m, bool b) {
4439 assert(m);
4440
4441 if (!MANAGER_IS_SYSTEM(m))
4442 return;
4443
4444 if (m->first_boot != (int) b) {
4445 if (b)
4446 (void) touch("/run/systemd/first-boot");
4447 else
4448 (void) unlink("/run/systemd/first-boot");
4449 }
4450
4451 m->first_boot = b;
4452 }
4453
4454 void manager_disable_confirm_spawn(void) {
4455 (void) touch("/run/systemd/confirm_spawn_disabled");
4456 }
4457
4458 bool manager_is_confirm_spawn_disabled(Manager *m) {
4459 if (!m->confirm_spawn)
4460 return true;
4461
4462 return access("/run/systemd/confirm_spawn_disabled", F_OK) >= 0;
4463 }
4464
4465 static bool manager_should_show_status(Manager *m, StatusType type) {
4466 assert(m);
4467
4468 if (!MANAGER_IS_SYSTEM(m))
4469 return false;
4470
4471 if (m->no_console_output)
4472 return false;
4473
4474 if (!IN_SET(manager_state(m), MANAGER_INITIALIZING, MANAGER_STARTING, MANAGER_STOPPING))
4475 return false;
4476
4477 /* If we cannot find out the status properly, just proceed. */
4478 if (type != STATUS_TYPE_EMERGENCY && manager_check_ask_password(m) > 0)
4479 return false;
4480
4481 if (type == STATUS_TYPE_NOTICE && m->show_status != SHOW_STATUS_NO)
4482 return true;
4483
4484 return manager_get_show_status_on(m);
4485 }
4486
4487 void manager_status_printf(Manager *m, StatusType type, const char *status, const char *format, ...) {
4488 va_list ap;
4489
4490 /* If m is NULL, assume we're after shutdown and let the messages through. */
4491
4492 if (m && !manager_should_show_status(m, type))
4493 return;
4494
4495 /* XXX We should totally drop the check for ephemeral here
4496 * and thus effectively make 'Type=idle' pointless. */
4497 if (type == STATUS_TYPE_EPHEMERAL && m && m->n_on_console > 0)
4498 return;
4499
4500 va_start(ap, format);
4501 status_vprintf(status, SHOW_STATUS_ELLIPSIZE|(type == STATUS_TYPE_EPHEMERAL ? SHOW_STATUS_EPHEMERAL : 0), format, ap);
4502 va_end(ap);
4503 }
4504
4505 Set *manager_get_units_requiring_mounts_for(Manager *m, const char *path) {
4506 char p[strlen(path)+1];
4507
4508 assert(m);
4509 assert(path);
4510
4511 strcpy(p, path);
4512 path_simplify(p, false);
4513
4514 return hashmap_get(m->units_requiring_mounts_for, streq(p, "/") ? "" : p);
4515 }
4516
4517 int manager_update_failed_units(Manager *m, Unit *u, bool failed) {
4518 unsigned size;
4519 int r;
4520
4521 assert(m);
4522 assert(u->manager == m);
4523
4524 size = set_size(m->failed_units);
4525
4526 if (failed) {
4527 r = set_ensure_put(&m->failed_units, NULL, u);
4528 if (r < 0)
4529 return log_oom();
4530 } else
4531 (void) set_remove(m->failed_units, u);
4532
4533 if (set_size(m->failed_units) != size)
4534 bus_manager_send_change_signal(m);
4535
4536 return 0;
4537 }
4538
4539 ManagerState manager_state(Manager *m) {
4540 Unit *u;
4541
4542 assert(m);
4543
4544 /* Is the special shutdown target active or queued? If so, we are in shutdown state */
4545 u = manager_get_unit(m, SPECIAL_SHUTDOWN_TARGET);
4546 if (u && unit_active_or_pending(u))
4547 return MANAGER_STOPPING;
4548
4549 /* Did we ever finish booting? If not then we are still starting up */
4550 if (!MANAGER_IS_FINISHED(m)) {
4551
4552 u = manager_get_unit(m, SPECIAL_BASIC_TARGET);
4553 if (!u || !UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u)))
4554 return MANAGER_INITIALIZING;
4555
4556 return MANAGER_STARTING;
4557 }
4558
4559 if (MANAGER_IS_SYSTEM(m)) {
4560 /* Are the rescue or emergency targets active or queued? If so we are in maintenance state */
4561 u = manager_get_unit(m, SPECIAL_RESCUE_TARGET);
4562 if (u && unit_active_or_pending(u))
4563 return MANAGER_MAINTENANCE;
4564
4565 u = manager_get_unit(m, SPECIAL_EMERGENCY_TARGET);
4566 if (u && unit_active_or_pending(u))
4567 return MANAGER_MAINTENANCE;
4568 }
4569
4570 /* Are there any failed units? If so, we are in degraded mode */
4571 if (set_size(m->failed_units) > 0)
4572 return MANAGER_DEGRADED;
4573
4574 return MANAGER_RUNNING;
4575 }
4576
4577 static void manager_unref_uid_internal(
4578 Hashmap *uid_refs,
4579 uid_t uid,
4580 bool destroy_now,
4581 int (*_clean_ipc)(uid_t uid)) {
4582
4583 uint32_t c, n;
4584
4585 assert(uid_is_valid(uid));
4586 assert(_clean_ipc);
4587
4588 /* A generic implementation, covering both manager_unref_uid() and manager_unref_gid(), under the assumption
4589 * that uid_t and gid_t are actually defined the same way, with the same validity rules.
4590 *
4591 * We store a hashmap where the UID/GID is they key and the value is a 32bit reference counter, whose highest
4592 * bit is used as flag for marking UIDs/GIDs whose IPC objects to remove when the last reference to the UID/GID
4593 * is dropped. The flag is set to on, once at least one reference from a unit where RemoveIPC= is set is added
4594 * on a UID/GID. It is reset when the UID's/GID's reference counter drops to 0 again. */
4595
4596 assert_cc(sizeof(uid_t) == sizeof(gid_t));
4597 assert_cc(UID_INVALID == (uid_t) GID_INVALID);
4598
4599 if (uid == 0) /* We don't keep track of root, and will never destroy it */
4600 return;
4601
4602 c = PTR_TO_UINT32(hashmap_get(uid_refs, UID_TO_PTR(uid)));
4603
4604 n = c & ~DESTROY_IPC_FLAG;
4605 assert(n > 0);
4606 n--;
4607
4608 if (destroy_now && n == 0) {
4609 hashmap_remove(uid_refs, UID_TO_PTR(uid));
4610
4611 if (c & DESTROY_IPC_FLAG) {
4612 log_debug("%s " UID_FMT " is no longer referenced, cleaning up its IPC.",
4613 _clean_ipc == clean_ipc_by_uid ? "UID" : "GID",
4614 uid);
4615 (void) _clean_ipc(uid);
4616 }
4617 } else {
4618 c = n | (c & DESTROY_IPC_FLAG);
4619 assert_se(hashmap_update(uid_refs, UID_TO_PTR(uid), UINT32_TO_PTR(c)) >= 0);
4620 }
4621 }
4622
4623 void manager_unref_uid(Manager *m, uid_t uid, bool destroy_now) {
4624 manager_unref_uid_internal(m->uid_refs, uid, destroy_now, clean_ipc_by_uid);
4625 }
4626
4627 void manager_unref_gid(Manager *m, gid_t gid, bool destroy_now) {
4628 manager_unref_uid_internal(m->gid_refs, (uid_t) gid, destroy_now, clean_ipc_by_gid);
4629 }
4630
4631 static int manager_ref_uid_internal(
4632 Hashmap **uid_refs,
4633 uid_t uid,
4634 bool clean_ipc) {
4635
4636 uint32_t c, n;
4637 int r;
4638
4639 assert(uid_refs);
4640 assert(uid_is_valid(uid));
4641
4642 /* A generic implementation, covering both manager_ref_uid() and manager_ref_gid(), under the assumption
4643 * that uid_t and gid_t are actually defined the same way, with the same validity rules. */
4644
4645 assert_cc(sizeof(uid_t) == sizeof(gid_t));
4646 assert_cc(UID_INVALID == (uid_t) GID_INVALID);
4647
4648 if (uid == 0) /* We don't keep track of root, and will never destroy it */
4649 return 0;
4650
4651 r = hashmap_ensure_allocated(uid_refs, &trivial_hash_ops);
4652 if (r < 0)
4653 return r;
4654
4655 c = PTR_TO_UINT32(hashmap_get(*uid_refs, UID_TO_PTR(uid)));
4656
4657 n = c & ~DESTROY_IPC_FLAG;
4658 n++;
4659
4660 if (n & DESTROY_IPC_FLAG) /* check for overflow */
4661 return -EOVERFLOW;
4662
4663 c = n | (c & DESTROY_IPC_FLAG) | (clean_ipc ? DESTROY_IPC_FLAG : 0);
4664
4665 return hashmap_replace(*uid_refs, UID_TO_PTR(uid), UINT32_TO_PTR(c));
4666 }
4667
4668 int manager_ref_uid(Manager *m, uid_t uid, bool clean_ipc) {
4669 return manager_ref_uid_internal(&m->uid_refs, uid, clean_ipc);
4670 }
4671
4672 int manager_ref_gid(Manager *m, gid_t gid, bool clean_ipc) {
4673 return manager_ref_uid_internal(&m->gid_refs, (uid_t) gid, clean_ipc);
4674 }
4675
4676 static void manager_vacuum_uid_refs_internal(
4677 Hashmap *uid_refs,
4678 int (*_clean_ipc)(uid_t uid)) {
4679
4680 void *p, *k;
4681
4682 assert(_clean_ipc);
4683
4684 HASHMAP_FOREACH_KEY(p, k, uid_refs) {
4685 uint32_t c, n;
4686 uid_t uid;
4687
4688 uid = PTR_TO_UID(k);
4689 c = PTR_TO_UINT32(p);
4690
4691 n = c & ~DESTROY_IPC_FLAG;
4692 if (n > 0)
4693 continue;
4694
4695 if (c & DESTROY_IPC_FLAG) {
4696 log_debug("Found unreferenced %s " UID_FMT " after reload/reexec. Cleaning up.",
4697 _clean_ipc == clean_ipc_by_uid ? "UID" : "GID",
4698 uid);
4699 (void) _clean_ipc(uid);
4700 }
4701
4702 assert_se(hashmap_remove(uid_refs, k) == p);
4703 }
4704 }
4705
4706 static void manager_vacuum_uid_refs(Manager *m) {
4707 manager_vacuum_uid_refs_internal(m->uid_refs, clean_ipc_by_uid);
4708 }
4709
4710 static void manager_vacuum_gid_refs(Manager *m) {
4711 manager_vacuum_uid_refs_internal(m->gid_refs, clean_ipc_by_gid);
4712 }
4713
4714 static void manager_vacuum(Manager *m) {
4715 assert(m);
4716
4717 /* Release any dynamic users no longer referenced */
4718 dynamic_user_vacuum(m, true);
4719
4720 /* Release any references to UIDs/GIDs no longer referenced, and destroy any IPC owned by them */
4721 manager_vacuum_uid_refs(m);
4722 manager_vacuum_gid_refs(m);
4723
4724 /* Release any runtimes no longer referenced */
4725 exec_runtime_vacuum(m);
4726 }
4727
4728 int manager_dispatch_user_lookup_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
4729 struct buffer {
4730 uid_t uid;
4731 gid_t gid;
4732 char unit_name[UNIT_NAME_MAX+1];
4733 } _packed_ buffer;
4734
4735 Manager *m = userdata;
4736 ssize_t l;
4737 size_t n;
4738 Unit *u;
4739
4740 assert_se(source);
4741 assert_se(m);
4742
4743 /* Invoked whenever a child process succeeded resolving its user/group to use and sent us the resulting UID/GID
4744 * in a datagram. We parse the datagram here and pass it off to the unit, so that it can add a reference to the
4745 * UID/GID so that it can destroy the UID/GID's IPC objects when the reference counter drops to 0. */
4746
4747 l = recv(fd, &buffer, sizeof(buffer), MSG_DONTWAIT);
4748 if (l < 0) {
4749 if (IN_SET(errno, EINTR, EAGAIN))
4750 return 0;
4751
4752 return log_error_errno(errno, "Failed to read from user lookup fd: %m");
4753 }
4754
4755 if ((size_t) l <= offsetof(struct buffer, unit_name)) {
4756 log_warning("Received too short user lookup message, ignoring.");
4757 return 0;
4758 }
4759
4760 if ((size_t) l > offsetof(struct buffer, unit_name) + UNIT_NAME_MAX) {
4761 log_warning("Received too long user lookup message, ignoring.");
4762 return 0;
4763 }
4764
4765 if (!uid_is_valid(buffer.uid) && !gid_is_valid(buffer.gid)) {
4766 log_warning("Got user lookup message with invalid UID/GID pair, ignoring.");
4767 return 0;
4768 }
4769
4770 n = (size_t) l - offsetof(struct buffer, unit_name);
4771 if (memchr(buffer.unit_name, 0, n)) {
4772 log_warning("Received lookup message with embedded NUL character, ignoring.");
4773 return 0;
4774 }
4775
4776 buffer.unit_name[n] = 0;
4777 u = manager_get_unit(m, buffer.unit_name);
4778 if (!u) {
4779 log_debug("Got user lookup message but unit doesn't exist, ignoring.");
4780 return 0;
4781 }
4782
4783 log_unit_debug(u, "User lookup succeeded: uid=" UID_FMT " gid=" GID_FMT, buffer.uid, buffer.gid);
4784
4785 unit_notify_user_lookup(u, buffer.uid, buffer.gid);
4786 return 0;
4787 }
4788
4789 char *manager_taint_string(Manager *m) {
4790 _cleanup_free_ char *destination = NULL, *overflowuid = NULL, *overflowgid = NULL;
4791 char *buf, *e;
4792 int r;
4793
4794 /* Returns a "taint string", e.g. "local-hwclock:var-run-bad".
4795 * Only things that are detected at runtime should be tagged
4796 * here. For stuff that is set during compilation, emit a warning
4797 * in the configuration phase. */
4798
4799 assert(m);
4800
4801 buf = new(char, sizeof("split-usr:"
4802 "cgroups-missing:"
4803 "cgrousv1:"
4804 "local-hwclock:"
4805 "var-run-bad:"
4806 "overflowuid-not-65534:"
4807 "overflowgid-not-65534:"));
4808 if (!buf)
4809 return NULL;
4810
4811 e = buf;
4812 buf[0] = 0;
4813
4814 if (m->taint_usr)
4815 e = stpcpy(e, "split-usr:");
4816
4817 if (access("/proc/cgroups", F_OK) < 0)
4818 e = stpcpy(e, "cgroups-missing:");
4819
4820 if (cg_all_unified() == 0)
4821 e = stpcpy(e, "cgroupsv1:");
4822
4823 if (clock_is_localtime(NULL) > 0)
4824 e = stpcpy(e, "local-hwclock:");
4825
4826 r = readlink_malloc("/var/run", &destination);
4827 if (r < 0 || !PATH_IN_SET(destination, "../run", "/run"))
4828 e = stpcpy(e, "var-run-bad:");
4829
4830 r = read_one_line_file("/proc/sys/kernel/overflowuid", &overflowuid);
4831 if (r >= 0 && !streq(overflowuid, "65534"))
4832 e = stpcpy(e, "overflowuid-not-65534:");
4833
4834 r = read_one_line_file("/proc/sys/kernel/overflowgid", &overflowgid);
4835 if (r >= 0 && !streq(overflowgid, "65534"))
4836 e = stpcpy(e, "overflowgid-not-65534:");
4837
4838 /* remove the last ':' */
4839 if (e != buf)
4840 e[-1] = 0;
4841
4842 return buf;
4843 }
4844
4845 void manager_ref_console(Manager *m) {
4846 assert(m);
4847
4848 m->n_on_console++;
4849 }
4850
4851 void manager_unref_console(Manager *m) {
4852
4853 assert(m->n_on_console > 0);
4854 m->n_on_console--;
4855
4856 if (m->n_on_console == 0)
4857 m->no_console_output = false; /* unset no_console_output flag, since the console is definitely free now */
4858 }
4859
4860 void manager_override_log_level(Manager *m, int level) {
4861 _cleanup_free_ char *s = NULL;
4862 assert(m);
4863
4864 if (!m->log_level_overridden) {
4865 m->original_log_level = log_get_max_level();
4866 m->log_level_overridden = true;
4867 }
4868
4869 (void) log_level_to_string_alloc(level, &s);
4870 log_info("Setting log level to %s.", strna(s));
4871
4872 log_set_max_level(level);
4873 }
4874
4875 void manager_restore_original_log_level(Manager *m) {
4876 _cleanup_free_ char *s = NULL;
4877 assert(m);
4878
4879 if (!m->log_level_overridden)
4880 return;
4881
4882 (void) log_level_to_string_alloc(m->original_log_level, &s);
4883 log_info("Restoring log level to original (%s).", strna(s));
4884
4885 log_set_max_level(m->original_log_level);
4886 m->log_level_overridden = false;
4887 }
4888
4889 void manager_override_log_target(Manager *m, LogTarget target) {
4890 assert(m);
4891
4892 if (!m->log_target_overridden) {
4893 m->original_log_target = log_get_target();
4894 m->log_target_overridden = true;
4895 }
4896
4897 log_info("Setting log target to %s.", log_target_to_string(target));
4898 log_set_target(target);
4899 }
4900
4901 void manager_restore_original_log_target(Manager *m) {
4902 assert(m);
4903
4904 if (!m->log_target_overridden)
4905 return;
4906
4907 log_info("Restoring log target to original %s.", log_target_to_string(m->original_log_target));
4908
4909 log_set_target(m->original_log_target);
4910 m->log_target_overridden = false;
4911 }
4912
4913 ManagerTimestamp manager_timestamp_initrd_mangle(ManagerTimestamp s) {
4914 if (in_initrd() &&
4915 s >= MANAGER_TIMESTAMP_SECURITY_START &&
4916 s <= MANAGER_TIMESTAMP_UNITS_LOAD_FINISH)
4917 return s - MANAGER_TIMESTAMP_SECURITY_START + MANAGER_TIMESTAMP_INITRD_SECURITY_START;
4918 return s;
4919 }
4920
4921 static const char *const manager_state_table[_MANAGER_STATE_MAX] = {
4922 [MANAGER_INITIALIZING] = "initializing",
4923 [MANAGER_STARTING] = "starting",
4924 [MANAGER_RUNNING] = "running",
4925 [MANAGER_DEGRADED] = "degraded",
4926 [MANAGER_MAINTENANCE] = "maintenance",
4927 [MANAGER_STOPPING] = "stopping",
4928 };
4929
4930 DEFINE_STRING_TABLE_LOOKUP(manager_state, ManagerState);
4931
4932 static const char *const manager_timestamp_table[_MANAGER_TIMESTAMP_MAX] = {
4933 [MANAGER_TIMESTAMP_FIRMWARE] = "firmware",
4934 [MANAGER_TIMESTAMP_LOADER] = "loader",
4935 [MANAGER_TIMESTAMP_KERNEL] = "kernel",
4936 [MANAGER_TIMESTAMP_INITRD] = "initrd",
4937 [MANAGER_TIMESTAMP_USERSPACE] = "userspace",
4938 [MANAGER_TIMESTAMP_FINISH] = "finish",
4939 [MANAGER_TIMESTAMP_SECURITY_START] = "security-start",
4940 [MANAGER_TIMESTAMP_SECURITY_FINISH] = "security-finish",
4941 [MANAGER_TIMESTAMP_GENERATORS_START] = "generators-start",
4942 [MANAGER_TIMESTAMP_GENERATORS_FINISH] = "generators-finish",
4943 [MANAGER_TIMESTAMP_UNITS_LOAD_START] = "units-load-start",
4944 [MANAGER_TIMESTAMP_UNITS_LOAD_FINISH] = "units-load-finish",
4945 [MANAGER_TIMESTAMP_INITRD_SECURITY_START] = "initrd-security-start",
4946 [MANAGER_TIMESTAMP_INITRD_SECURITY_FINISH] = "initrd-security-finish",
4947 [MANAGER_TIMESTAMP_INITRD_GENERATORS_START] = "initrd-generators-start",
4948 [MANAGER_TIMESTAMP_INITRD_GENERATORS_FINISH] = "initrd-generators-finish",
4949 [MANAGER_TIMESTAMP_INITRD_UNITS_LOAD_START] = "initrd-units-load-start",
4950 [MANAGER_TIMESTAMP_INITRD_UNITS_LOAD_FINISH] = "initrd-units-load-finish",
4951 };
4952
4953 DEFINE_STRING_TABLE_LOOKUP(manager_timestamp, ManagerTimestamp);
4954
4955 static const char* const oom_policy_table[_OOM_POLICY_MAX] = {
4956 [OOM_CONTINUE] = "continue",
4957 [OOM_STOP] = "stop",
4958 [OOM_KILL] = "kill",
4959 };
4960
4961 DEFINE_STRING_TABLE_LOOKUP(oom_policy, OOMPolicy);
Unnamed repository; edit this file 'description' to name the repository.