]> git.ipfire.org Git - thirdparty/systemd.git/blob - src/core/manager.c
core: rename cg_unified() to cg_all_unified()
[thirdparty/systemd.git] / src / core / manager.c
1 /***
2 This file is part of systemd.
3
4 Copyright 2010 Lennart Poettering
5
6 systemd is free software; you can redistribute it and/or modify it
7 under the terms of the GNU Lesser General Public License as published by
8 the Free Software Foundation; either version 2.1 of the License, or
9 (at your option) any later version.
10
11 systemd is distributed in the hope that it will be useful, but
12 WITHOUT ANY WARRANTY; without even the implied warranty of
13 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 Lesser General Public License for more details.
15
16 You should have received a copy of the GNU Lesser General Public License
17 along with systemd; If not, see <http://www.gnu.org/licenses/>.
18 ***/
19
20 #include <dirent.h>
21 #include <errno.h>
22 #include <fcntl.h>
23 #include <linux/kd.h>
24 #include <signal.h>
25 #include <string.h>
26 #include <sys/epoll.h>
27 #include <sys/inotify.h>
28 #include <sys/ioctl.h>
29 #include <sys/reboot.h>
30 #include <sys/timerfd.h>
31 #include <sys/wait.h>
32 #include <unistd.h>
33
34 #ifdef HAVE_AUDIT
35 #include <libaudit.h>
36 #endif
37
38 #include "sd-daemon.h"
39 #include "sd-messages.h"
40
41 #include "alloc-util.h"
42 #include "audit-fd.h"
43 #include "boot-timestamps.h"
44 #include "bus-common-errors.h"
45 #include "bus-error.h"
46 #include "bus-kernel.h"
47 #include "bus-util.h"
48 #include "dbus-job.h"
49 #include "dbus-manager.h"
50 #include "dbus-unit.h"
51 #include "dbus.h"
52 #include "dirent-util.h"
53 #include "env-util.h"
54 #include "escape.h"
55 #include "exit-status.h"
56 #include "fd-util.h"
57 #include "fileio.h"
58 #include "fs-util.h"
59 #include "hashmap.h"
60 #include "io-util.h"
61 #include "locale-setup.h"
62 #include "log.h"
63 #include "macro.h"
64 #include "manager.h"
65 #include "missing.h"
66 #include "mkdir.h"
67 #include "parse-util.h"
68 #include "path-lookup.h"
69 #include "path-util.h"
70 #include "process-util.h"
71 #include "ratelimit.h"
72 #include "rm-rf.h"
73 #include "signal-util.h"
74 #include "special.h"
75 #include "stat-util.h"
76 #include "string-table.h"
77 #include "string-util.h"
78 #include "strv.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 "util.h"
85 #include "virt.h"
86 #include "watchdog.h"
87
88 #define NOTIFY_RCVBUF_SIZE (8*1024*1024)
89 #define CGROUPS_AGENT_RCVBUF_SIZE (8*1024*1024)
90
91 /* Initial delay and the interval for printing status messages about running jobs */
92 #define JOBS_IN_PROGRESS_WAIT_USEC (5*USEC_PER_SEC)
93 #define JOBS_IN_PROGRESS_PERIOD_USEC (USEC_PER_SEC / 3)
94 #define JOBS_IN_PROGRESS_PERIOD_DIVISOR 3
95
96 static int manager_dispatch_notify_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata);
97 static int manager_dispatch_cgroups_agent_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata);
98 static int manager_dispatch_signal_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata);
99 static int manager_dispatch_time_change_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata);
100 static int manager_dispatch_idle_pipe_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata);
101 static int manager_dispatch_jobs_in_progress(sd_event_source *source, usec_t usec, void *userdata);
102 static int manager_dispatch_run_queue(sd_event_source *source, void *userdata);
103 static int manager_run_generators(Manager *m);
104
105 static void manager_watch_jobs_in_progress(Manager *m) {
106 usec_t next;
107 int r;
108
109 assert(m);
110
111 if (m->jobs_in_progress_event_source)
112 return;
113
114 next = now(CLOCK_MONOTONIC) + JOBS_IN_PROGRESS_WAIT_USEC;
115 r = sd_event_add_time(
116 m->event,
117 &m->jobs_in_progress_event_source,
118 CLOCK_MONOTONIC,
119 next, 0,
120 manager_dispatch_jobs_in_progress, m);
121 if (r < 0)
122 return;
123
124 (void) sd_event_source_set_description(m->jobs_in_progress_event_source, "manager-jobs-in-progress");
125 }
126
127 #define CYLON_BUFFER_EXTRA (2*(sizeof(ANSI_RED)-1) + sizeof(ANSI_HIGHLIGHT_RED)-1 + 2*(sizeof(ANSI_NORMAL)-1))
128
129 static void draw_cylon(char buffer[], size_t buflen, unsigned width, unsigned pos) {
130 char *p = buffer;
131
132 assert(buflen >= CYLON_BUFFER_EXTRA + width + 1);
133 assert(pos <= width+1); /* 0 or width+1 mean that the center light is behind the corner */
134
135 if (pos > 1) {
136 if (pos > 2)
137 p = mempset(p, ' ', pos-2);
138 if (log_get_show_color())
139 p = stpcpy(p, ANSI_RED);
140 *p++ = '*';
141 }
142
143 if (pos > 0 && pos <= width) {
144 if (log_get_show_color())
145 p = stpcpy(p, ANSI_HIGHLIGHT_RED);
146 *p++ = '*';
147 }
148
149 if (log_get_show_color())
150 p = stpcpy(p, ANSI_NORMAL);
151
152 if (pos < width) {
153 if (log_get_show_color())
154 p = stpcpy(p, ANSI_RED);
155 *p++ = '*';
156 if (pos < width-1)
157 p = mempset(p, ' ', width-1-pos);
158 if (log_get_show_color())
159 strcpy(p, ANSI_NORMAL);
160 }
161 }
162
163 void manager_flip_auto_status(Manager *m, bool enable) {
164 assert(m);
165
166 if (enable) {
167 if (m->show_status == SHOW_STATUS_AUTO)
168 manager_set_show_status(m, SHOW_STATUS_TEMPORARY);
169 } else {
170 if (m->show_status == SHOW_STATUS_TEMPORARY)
171 manager_set_show_status(m, SHOW_STATUS_AUTO);
172 }
173 }
174
175 static void manager_print_jobs_in_progress(Manager *m) {
176 _cleanup_free_ char *job_of_n = NULL;
177 Iterator i;
178 Job *j;
179 unsigned counter = 0, print_nr;
180 char cylon[6 + CYLON_BUFFER_EXTRA + 1];
181 unsigned cylon_pos;
182 char time[FORMAT_TIMESPAN_MAX], limit[FORMAT_TIMESPAN_MAX] = "no limit";
183 uint64_t x;
184
185 assert(m);
186 assert(m->n_running_jobs > 0);
187
188 manager_flip_auto_status(m, true);
189
190 print_nr = (m->jobs_in_progress_iteration / JOBS_IN_PROGRESS_PERIOD_DIVISOR) % m->n_running_jobs;
191
192 HASHMAP_FOREACH(j, m->jobs, i)
193 if (j->state == JOB_RUNNING && counter++ == print_nr)
194 break;
195
196 /* m->n_running_jobs must be consistent with the contents of m->jobs,
197 * so the above loop must have succeeded in finding j. */
198 assert(counter == print_nr + 1);
199 assert(j);
200
201 cylon_pos = m->jobs_in_progress_iteration % 14;
202 if (cylon_pos >= 8)
203 cylon_pos = 14 - cylon_pos;
204 draw_cylon(cylon, sizeof(cylon), 6, cylon_pos);
205
206 m->jobs_in_progress_iteration++;
207
208 if (m->n_running_jobs > 1) {
209 if (asprintf(&job_of_n, "(%u of %u) ", counter, m->n_running_jobs) < 0)
210 job_of_n = NULL;
211 }
212
213 format_timespan(time, sizeof(time), now(CLOCK_MONOTONIC) - j->begin_usec, 1*USEC_PER_SEC);
214 if (job_get_timeout(j, &x) > 0)
215 format_timespan(limit, sizeof(limit), x - j->begin_usec, 1*USEC_PER_SEC);
216
217 manager_status_printf(m, STATUS_TYPE_EPHEMERAL, cylon,
218 "%sA %s job is running for %s (%s / %s)",
219 strempty(job_of_n),
220 job_type_to_string(j->type),
221 unit_description(j->unit),
222 time, limit);
223 }
224
225 static int have_ask_password(void) {
226 _cleanup_closedir_ DIR *dir;
227
228 dir = opendir("/run/systemd/ask-password");
229 if (!dir) {
230 if (errno == ENOENT)
231 return false;
232 else
233 return -errno;
234 }
235
236 for (;;) {
237 struct dirent *de;
238
239 errno = 0;
240 de = readdir(dir);
241 if (!de && errno > 0)
242 return -errno;
243 if (!de)
244 return false;
245
246 if (startswith(de->d_name, "ask."))
247 return true;
248 }
249 }
250
251 static int manager_dispatch_ask_password_fd(sd_event_source *source,
252 int fd, uint32_t revents, void *userdata) {
253 Manager *m = userdata;
254
255 assert(m);
256
257 flush_fd(fd);
258
259 m->have_ask_password = have_ask_password();
260 if (m->have_ask_password < 0)
261 /* Log error but continue. Negative have_ask_password
262 * is treated as unknown status. */
263 log_error_errno(m->have_ask_password, "Failed to list /run/systemd/ask-password: %m");
264
265 return 0;
266 }
267
268 static void manager_close_ask_password(Manager *m) {
269 assert(m);
270
271 m->ask_password_event_source = sd_event_source_unref(m->ask_password_event_source);
272 m->ask_password_inotify_fd = safe_close(m->ask_password_inotify_fd);
273 m->have_ask_password = -EINVAL;
274 }
275
276 static int manager_check_ask_password(Manager *m) {
277 int r;
278
279 assert(m);
280
281 if (!m->ask_password_event_source) {
282 assert(m->ask_password_inotify_fd < 0);
283
284 mkdir_p_label("/run/systemd/ask-password", 0755);
285
286 m->ask_password_inotify_fd = inotify_init1(IN_NONBLOCK|IN_CLOEXEC);
287 if (m->ask_password_inotify_fd < 0)
288 return log_error_errno(errno, "inotify_init1() failed: %m");
289
290 if (inotify_add_watch(m->ask_password_inotify_fd, "/run/systemd/ask-password", IN_CREATE|IN_DELETE|IN_MOVE) < 0) {
291 log_error_errno(errno, "Failed to add watch on /run/systemd/ask-password: %m");
292 manager_close_ask_password(m);
293 return -errno;
294 }
295
296 r = sd_event_add_io(m->event, &m->ask_password_event_source,
297 m->ask_password_inotify_fd, EPOLLIN,
298 manager_dispatch_ask_password_fd, m);
299 if (r < 0) {
300 log_error_errno(errno, "Failed to add event source for /run/systemd/ask-password: %m");
301 manager_close_ask_password(m);
302 return -errno;
303 }
304
305 (void) sd_event_source_set_description(m->ask_password_event_source, "manager-ask-password");
306
307 /* Queries might have been added meanwhile... */
308 manager_dispatch_ask_password_fd(m->ask_password_event_source,
309 m->ask_password_inotify_fd, EPOLLIN, m);
310 }
311
312 return m->have_ask_password;
313 }
314
315 static int manager_watch_idle_pipe(Manager *m) {
316 int r;
317
318 assert(m);
319
320 if (m->idle_pipe_event_source)
321 return 0;
322
323 if (m->idle_pipe[2] < 0)
324 return 0;
325
326 r = sd_event_add_io(m->event, &m->idle_pipe_event_source, m->idle_pipe[2], EPOLLIN, manager_dispatch_idle_pipe_fd, m);
327 if (r < 0)
328 return log_error_errno(r, "Failed to watch idle pipe: %m");
329
330 (void) sd_event_source_set_description(m->idle_pipe_event_source, "manager-idle-pipe");
331
332 return 0;
333 }
334
335 static void manager_close_idle_pipe(Manager *m) {
336 assert(m);
337
338 m->idle_pipe_event_source = sd_event_source_unref(m->idle_pipe_event_source);
339
340 safe_close_pair(m->idle_pipe);
341 safe_close_pair(m->idle_pipe + 2);
342 }
343
344 static int manager_setup_time_change(Manager *m) {
345 int r;
346
347 /* We only care for the cancellation event, hence we set the
348 * timeout to the latest possible value. */
349 struct itimerspec its = {
350 .it_value.tv_sec = TIME_T_MAX,
351 };
352
353 assert(m);
354 assert_cc(sizeof(time_t) == sizeof(TIME_T_MAX));
355
356 if (m->test_run)
357 return 0;
358
359 /* Uses TFD_TIMER_CANCEL_ON_SET to get notifications whenever
360 * CLOCK_REALTIME makes a jump relative to CLOCK_MONOTONIC */
361
362 m->time_change_fd = timerfd_create(CLOCK_REALTIME, TFD_NONBLOCK|TFD_CLOEXEC);
363 if (m->time_change_fd < 0)
364 return log_error_errno(errno, "Failed to create timerfd: %m");
365
366 if (timerfd_settime(m->time_change_fd, TFD_TIMER_ABSTIME|TFD_TIMER_CANCEL_ON_SET, &its, NULL) < 0) {
367 log_debug_errno(errno, "Failed to set up TFD_TIMER_CANCEL_ON_SET, ignoring: %m");
368 m->time_change_fd = safe_close(m->time_change_fd);
369 return 0;
370 }
371
372 r = sd_event_add_io(m->event, &m->time_change_event_source, m->time_change_fd, EPOLLIN, manager_dispatch_time_change_fd, m);
373 if (r < 0)
374 return log_error_errno(r, "Failed to create time change event source: %m");
375
376 (void) sd_event_source_set_description(m->time_change_event_source, "manager-time-change");
377
378 log_debug("Set up TFD_TIMER_CANCEL_ON_SET timerfd.");
379
380 return 0;
381 }
382
383 static int enable_special_signals(Manager *m) {
384 _cleanup_close_ int fd = -1;
385
386 assert(m);
387
388 if (m->test_run)
389 return 0;
390
391 /* Enable that we get SIGINT on control-alt-del. In containers
392 * this will fail with EPERM (older) or EINVAL (newer), so
393 * ignore that. */
394 if (reboot(RB_DISABLE_CAD) < 0 && errno != EPERM && errno != EINVAL)
395 log_warning_errno(errno, "Failed to enable ctrl-alt-del handling: %m");
396
397 fd = open_terminal("/dev/tty0", O_RDWR|O_NOCTTY|O_CLOEXEC);
398 if (fd < 0) {
399 /* Support systems without virtual console */
400 if (fd != -ENOENT)
401 log_warning_errno(errno, "Failed to open /dev/tty0: %m");
402 } else {
403 /* Enable that we get SIGWINCH on kbrequest */
404 if (ioctl(fd, KDSIGACCEPT, SIGWINCH) < 0)
405 log_warning_errno(errno, "Failed to enable kbrequest handling: %m");
406 }
407
408 return 0;
409 }
410
411 static int manager_setup_signals(Manager *m) {
412 struct sigaction sa = {
413 .sa_handler = SIG_DFL,
414 .sa_flags = SA_NOCLDSTOP|SA_RESTART,
415 };
416 sigset_t mask;
417 int r;
418
419 assert(m);
420
421 assert_se(sigaction(SIGCHLD, &sa, NULL) == 0);
422
423 /* We make liberal use of realtime signals here. On
424 * Linux/glibc we have 30 of them (with the exception of Linux
425 * on hppa, see below), between SIGRTMIN+0 ... SIGRTMIN+30
426 * (aka SIGRTMAX). */
427
428 assert_se(sigemptyset(&mask) == 0);
429 sigset_add_many(&mask,
430 SIGCHLD, /* Child died */
431 SIGTERM, /* Reexecute daemon */
432 SIGHUP, /* Reload configuration */
433 SIGUSR1, /* systemd/upstart: reconnect to D-Bus */
434 SIGUSR2, /* systemd: dump status */
435 SIGINT, /* Kernel sends us this on control-alt-del */
436 SIGWINCH, /* Kernel sends us this on kbrequest (alt-arrowup) */
437 SIGPWR, /* Some kernel drivers and upsd send us this on power failure */
438
439 SIGRTMIN+0, /* systemd: start default.target */
440 SIGRTMIN+1, /* systemd: isolate rescue.target */
441 SIGRTMIN+2, /* systemd: isolate emergency.target */
442 SIGRTMIN+3, /* systemd: start halt.target */
443 SIGRTMIN+4, /* systemd: start poweroff.target */
444 SIGRTMIN+5, /* systemd: start reboot.target */
445 SIGRTMIN+6, /* systemd: start kexec.target */
446
447 /* ... space for more special targets ... */
448
449 SIGRTMIN+13, /* systemd: Immediate halt */
450 SIGRTMIN+14, /* systemd: Immediate poweroff */
451 SIGRTMIN+15, /* systemd: Immediate reboot */
452 SIGRTMIN+16, /* systemd: Immediate kexec */
453
454 /* ... space for more immediate system state changes ... */
455
456 SIGRTMIN+20, /* systemd: enable status messages */
457 SIGRTMIN+21, /* systemd: disable status messages */
458 SIGRTMIN+22, /* systemd: set log level to LOG_DEBUG */
459 SIGRTMIN+23, /* systemd: set log level to LOG_INFO */
460 SIGRTMIN+24, /* systemd: Immediate exit (--user only) */
461
462 /* .. one free signal here ... */
463
464 #if !defined(__hppa64__) && !defined(__hppa__)
465 /* Apparently Linux on hppa has fewer RT
466 * signals (SIGRTMAX is SIGRTMIN+25 there),
467 * hence let's not try to make use of them
468 * here. Since these commands are accessible
469 * by different means and only really a safety
470 * net, the missing functionality on hppa
471 * shouldn't matter. */
472
473 SIGRTMIN+26, /* systemd: set log target to journal-or-kmsg */
474 SIGRTMIN+27, /* systemd: set log target to console */
475 SIGRTMIN+28, /* systemd: set log target to kmsg */
476 SIGRTMIN+29, /* systemd: set log target to syslog-or-kmsg (obsolete) */
477
478 /* ... one free signal here SIGRTMIN+30 ... */
479 #endif
480 -1);
481 assert_se(sigprocmask(SIG_SETMASK, &mask, NULL) == 0);
482
483 m->signal_fd = signalfd(-1, &mask, SFD_NONBLOCK|SFD_CLOEXEC);
484 if (m->signal_fd < 0)
485 return -errno;
486
487 r = sd_event_add_io(m->event, &m->signal_event_source, m->signal_fd, EPOLLIN, manager_dispatch_signal_fd, m);
488 if (r < 0)
489 return r;
490
491 (void) sd_event_source_set_description(m->signal_event_source, "manager-signal");
492
493 /* Process signals a bit earlier than the rest of things, but later than notify_fd processing, so that the
494 * notify processing can still figure out to which process/service a message belongs, before we reap the
495 * process. Also, process this before handling cgroup notifications, so that we always collect child exit
496 * status information before detecting that there's no process in a cgroup. */
497 r = sd_event_source_set_priority(m->signal_event_source, SD_EVENT_PRIORITY_NORMAL-6);
498 if (r < 0)
499 return r;
500
501 if (MANAGER_IS_SYSTEM(m))
502 return enable_special_signals(m);
503
504 return 0;
505 }
506
507 static void manager_clean_environment(Manager *m) {
508 assert(m);
509
510 /* Let's remove some environment variables that we
511 * need ourselves to communicate with our clients */
512 strv_env_unset_many(
513 m->environment,
514 "NOTIFY_SOCKET",
515 "MAINPID",
516 "MANAGERPID",
517 "LISTEN_PID",
518 "LISTEN_FDS",
519 "LISTEN_FDNAMES",
520 "WATCHDOG_PID",
521 "WATCHDOG_USEC",
522 NULL);
523 }
524
525 static int manager_default_environment(Manager *m) {
526 assert(m);
527
528 if (MANAGER_IS_SYSTEM(m)) {
529 /* The system manager always starts with a clean
530 * environment for its children. It does not import
531 * the kernel or the parents exported variables.
532 *
533 * The initial passed environ is untouched to keep
534 * /proc/self/environ valid; it is used for tagging
535 * the init process inside containers. */
536 m->environment = strv_new("PATH=" DEFAULT_PATH,
537 NULL);
538
539 /* Import locale variables LC_*= from configuration */
540 locale_setup(&m->environment);
541 } else {
542 /* The user manager passes its own environment
543 * along to its children. */
544 m->environment = strv_copy(environ);
545 }
546
547 if (!m->environment)
548 return -ENOMEM;
549
550 manager_clean_environment(m);
551 strv_sort(m->environment);
552
553 return 0;
554 }
555
556 int manager_new(UnitFileScope scope, bool test_run, Manager **_m) {
557 Manager *m;
558 int r;
559
560 assert(_m);
561 assert(IN_SET(scope, UNIT_FILE_SYSTEM, UNIT_FILE_USER));
562
563 m = new0(Manager, 1);
564 if (!m)
565 return -ENOMEM;
566
567 m->unit_file_scope = scope;
568 m->exit_code = _MANAGER_EXIT_CODE_INVALID;
569 m->default_timer_accuracy_usec = USEC_PER_MINUTE;
570 m->default_tasks_accounting = true;
571 m->default_tasks_max = UINT64_MAX;
572
573 #ifdef ENABLE_EFI
574 if (MANAGER_IS_SYSTEM(m) && detect_container() <= 0)
575 boot_timestamps(&m->userspace_timestamp, &m->firmware_timestamp, &m->loader_timestamp);
576 #endif
577
578 /* Prepare log fields we can use for structured logging */
579 if (MANAGER_IS_SYSTEM(m)) {
580 m->unit_log_field = "UNIT=";
581 m->unit_log_format_string = "UNIT=%s";
582 } else {
583 m->unit_log_field = "USER_UNIT=";
584 m->unit_log_format_string = "USER_UNIT=%s";
585 }
586
587 m->idle_pipe[0] = m->idle_pipe[1] = m->idle_pipe[2] = m->idle_pipe[3] = -1;
588
589 m->pin_cgroupfs_fd = m->notify_fd = m->cgroups_agent_fd = m->signal_fd = m->time_change_fd =
590 m->dev_autofs_fd = m->private_listen_fd = m->kdbus_fd = m->cgroup_inotify_fd =
591 m->ask_password_inotify_fd = -1;
592
593 m->current_job_id = 1; /* start as id #1, so that we can leave #0 around as "null-like" value */
594
595 m->have_ask_password = -EINVAL; /* we don't know */
596 m->first_boot = -1;
597
598 m->test_run = test_run;
599
600 /* Reboot immediately if the user hits C-A-D more often than 7x per 2s */
601 RATELIMIT_INIT(m->ctrl_alt_del_ratelimit, 2 * USEC_PER_SEC, 7);
602
603 r = manager_default_environment(m);
604 if (r < 0)
605 goto fail;
606
607 r = hashmap_ensure_allocated(&m->units, &string_hash_ops);
608 if (r < 0)
609 goto fail;
610
611 r = hashmap_ensure_allocated(&m->jobs, NULL);
612 if (r < 0)
613 goto fail;
614
615 r = hashmap_ensure_allocated(&m->cgroup_unit, &string_hash_ops);
616 if (r < 0)
617 goto fail;
618
619 r = hashmap_ensure_allocated(&m->watch_bus, &string_hash_ops);
620 if (r < 0)
621 goto fail;
622
623 r = sd_event_default(&m->event);
624 if (r < 0)
625 goto fail;
626
627 r = sd_event_add_defer(m->event, &m->run_queue_event_source, manager_dispatch_run_queue, m);
628 if (r < 0)
629 goto fail;
630
631 r = sd_event_source_set_priority(m->run_queue_event_source, SD_EVENT_PRIORITY_IDLE);
632 if (r < 0)
633 goto fail;
634
635 r = sd_event_source_set_enabled(m->run_queue_event_source, SD_EVENT_OFF);
636 if (r < 0)
637 goto fail;
638
639 (void) sd_event_source_set_description(m->run_queue_event_source, "manager-run-queue");
640
641 r = manager_setup_signals(m);
642 if (r < 0)
643 goto fail;
644
645 r = manager_setup_cgroup(m);
646 if (r < 0)
647 goto fail;
648
649 r = manager_setup_time_change(m);
650 if (r < 0)
651 goto fail;
652
653 m->udev = udev_new();
654 if (!m->udev) {
655 r = -ENOMEM;
656 goto fail;
657 }
658
659 /* Note that we set up neither kdbus, nor the notify fd
660 * here. We do that after deserialization, since they might
661 * have gotten serialized across the reexec. */
662
663 m->taint_usr = dir_is_empty("/usr") > 0;
664
665 *_m = m;
666 return 0;
667
668 fail:
669 manager_free(m);
670 return r;
671 }
672
673 static int manager_setup_notify(Manager *m) {
674 int r;
675
676 if (m->test_run)
677 return 0;
678
679 if (m->notify_fd < 0) {
680 _cleanup_close_ int fd = -1;
681 union sockaddr_union sa = {
682 .sa.sa_family = AF_UNIX,
683 };
684 static const int one = 1;
685 const char *e;
686
687 /* First free all secondary fields */
688 m->notify_socket = mfree(m->notify_socket);
689 m->notify_event_source = sd_event_source_unref(m->notify_event_source);
690
691 fd = socket(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC|SOCK_NONBLOCK, 0);
692 if (fd < 0)
693 return log_error_errno(errno, "Failed to allocate notification socket: %m");
694
695 fd_inc_rcvbuf(fd, NOTIFY_RCVBUF_SIZE);
696
697 e = manager_get_runtime_prefix(m);
698 if (!e) {
699 log_error("Failed to determine runtime prefix.");
700 return -EINVAL;
701 }
702
703 m->notify_socket = strappend(e, "/systemd/notify");
704 if (!m->notify_socket)
705 return log_oom();
706
707 (void) mkdir_parents_label(m->notify_socket, 0755);
708 (void) unlink(m->notify_socket);
709
710 strncpy(sa.un.sun_path, m->notify_socket, sizeof(sa.un.sun_path)-1);
711 r = bind(fd, &sa.sa, SOCKADDR_UN_LEN(sa.un));
712 if (r < 0)
713 return log_error_errno(errno, "bind(%s) failed: %m", sa.un.sun_path);
714
715 r = setsockopt(fd, SOL_SOCKET, SO_PASSCRED, &one, sizeof(one));
716 if (r < 0)
717 return log_error_errno(errno, "SO_PASSCRED failed: %m");
718
719 m->notify_fd = fd;
720 fd = -1;
721
722 log_debug("Using notification socket %s", m->notify_socket);
723 }
724
725 if (!m->notify_event_source) {
726 r = sd_event_add_io(m->event, &m->notify_event_source, m->notify_fd, EPOLLIN, manager_dispatch_notify_fd, m);
727 if (r < 0)
728 return log_error_errno(r, "Failed to allocate notify event source: %m");
729
730 /* Process notification messages a bit earlier than SIGCHLD, so that we can still identify to which
731 * service an exit message belongs. */
732 r = sd_event_source_set_priority(m->notify_event_source, SD_EVENT_PRIORITY_NORMAL-7);
733 if (r < 0)
734 return log_error_errno(r, "Failed to set priority of notify event source: %m");
735
736 (void) sd_event_source_set_description(m->notify_event_source, "manager-notify");
737 }
738
739 return 0;
740 }
741
742 static int manager_setup_cgroups_agent(Manager *m) {
743
744 static const union sockaddr_union sa = {
745 .un.sun_family = AF_UNIX,
746 .un.sun_path = "/run/systemd/cgroups-agent",
747 };
748 int r;
749
750 /* This creates a listening socket we receive cgroups agent messages on. We do not use D-Bus for delivering
751 * these messages from the cgroups agent binary to PID 1, as the cgroups agent binary is very short-living, and
752 * each instance of it needs a new D-Bus connection. Since D-Bus connections are SOCK_STREAM/AF_UNIX, on
753 * overloaded systems the backlog of the D-Bus socket becomes relevant, as not more than the configured number
754 * of D-Bus connections may be queued until the kernel will start dropping further incoming connections,
755 * possibly resulting in lost cgroups agent messages. To avoid this, we'll use a private SOCK_DGRAM/AF_UNIX
756 * socket, where no backlog is relevant as communication may take place without an actual connect() cycle, and
757 * we thus won't lose messages.
758 *
759 * Note that PID 1 will forward the agent message to system bus, so that the user systemd instance may listen
760 * to it. The system instance hence listens on this special socket, but the user instances listen on the system
761 * bus for these messages. */
762
763 if (m->test_run)
764 return 0;
765
766 if (!MANAGER_IS_SYSTEM(m))
767 return 0;
768
769 if (cg_all_unified() > 0) /* We don't need this anymore on the unified hierarchy */
770 return 0;
771
772 if (m->cgroups_agent_fd < 0) {
773 _cleanup_close_ int fd = -1;
774
775 /* First free all secondary fields */
776 m->cgroups_agent_event_source = sd_event_source_unref(m->cgroups_agent_event_source);
777
778 fd = socket(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC|SOCK_NONBLOCK, 0);
779 if (fd < 0)
780 return log_error_errno(errno, "Failed to allocate cgroups agent socket: %m");
781
782 fd_inc_rcvbuf(fd, CGROUPS_AGENT_RCVBUF_SIZE);
783
784 (void) unlink(sa.un.sun_path);
785
786 /* Only allow root to connect to this socket */
787 RUN_WITH_UMASK(0077)
788 r = bind(fd, &sa.sa, SOCKADDR_UN_LEN(sa.un));
789 if (r < 0)
790 return log_error_errno(errno, "bind(%s) failed: %m", sa.un.sun_path);
791
792 m->cgroups_agent_fd = fd;
793 fd = -1;
794 }
795
796 if (!m->cgroups_agent_event_source) {
797 r = sd_event_add_io(m->event, &m->cgroups_agent_event_source, m->cgroups_agent_fd, EPOLLIN, manager_dispatch_cgroups_agent_fd, m);
798 if (r < 0)
799 return log_error_errno(r, "Failed to allocate cgroups agent event source: %m");
800
801 /* Process cgroups notifications early, but after having processed service notification messages or
802 * SIGCHLD signals, so that a cgroup running empty is always just the last safety net of notification,
803 * and we collected the metadata the notification and SIGCHLD stuff offers first. Also see handling of
804 * cgroup inotify for the unified cgroup stuff. */
805 r = sd_event_source_set_priority(m->cgroups_agent_event_source, SD_EVENT_PRIORITY_NORMAL-5);
806 if (r < 0)
807 return log_error_errno(r, "Failed to set priority of cgroups agent event source: %m");
808
809 (void) sd_event_source_set_description(m->cgroups_agent_event_source, "manager-cgroups-agent");
810 }
811
812 return 0;
813 }
814
815 static int manager_connect_bus(Manager *m, bool reexecuting) {
816 bool try_bus_connect;
817
818 assert(m);
819
820 if (m->test_run)
821 return 0;
822
823 try_bus_connect =
824 m->kdbus_fd >= 0 ||
825 reexecuting ||
826 (MANAGER_IS_USER(m) && getenv("DBUS_SESSION_BUS_ADDRESS"));
827
828 /* Try to connect to the buses, if possible. */
829 return bus_init(m, try_bus_connect);
830 }
831
832 static unsigned manager_dispatch_cleanup_queue(Manager *m) {
833 Unit *u;
834 unsigned n = 0;
835
836 assert(m);
837
838 while ((u = m->cleanup_queue)) {
839 assert(u->in_cleanup_queue);
840
841 unit_free(u);
842 n++;
843 }
844
845 return n;
846 }
847
848 enum {
849 GC_OFFSET_IN_PATH, /* This one is on the path we were traveling */
850 GC_OFFSET_UNSURE, /* No clue */
851 GC_OFFSET_GOOD, /* We still need this unit */
852 GC_OFFSET_BAD, /* We don't need this unit anymore */
853 _GC_OFFSET_MAX
854 };
855
856 static void unit_gc_mark_good(Unit *u, unsigned gc_marker)
857 {
858 Iterator i;
859 Unit *other;
860
861 u->gc_marker = gc_marker + GC_OFFSET_GOOD;
862
863 /* Recursively mark referenced units as GOOD as well */
864 SET_FOREACH(other, u->dependencies[UNIT_REFERENCES], i)
865 if (other->gc_marker == gc_marker + GC_OFFSET_UNSURE)
866 unit_gc_mark_good(other, gc_marker);
867 }
868
869 static void unit_gc_sweep(Unit *u, unsigned gc_marker) {
870 Iterator i;
871 Unit *other;
872 bool is_bad;
873
874 assert(u);
875
876 if (u->gc_marker == gc_marker + GC_OFFSET_GOOD ||
877 u->gc_marker == gc_marker + GC_OFFSET_BAD ||
878 u->gc_marker == gc_marker + GC_OFFSET_UNSURE ||
879 u->gc_marker == gc_marker + GC_OFFSET_IN_PATH)
880 return;
881
882 if (u->in_cleanup_queue)
883 goto bad;
884
885 if (unit_check_gc(u))
886 goto good;
887
888 u->gc_marker = gc_marker + GC_OFFSET_IN_PATH;
889
890 is_bad = true;
891
892 SET_FOREACH(other, u->dependencies[UNIT_REFERENCED_BY], i) {
893 unit_gc_sweep(other, gc_marker);
894
895 if (other->gc_marker == gc_marker + GC_OFFSET_GOOD)
896 goto good;
897
898 if (other->gc_marker != gc_marker + GC_OFFSET_BAD)
899 is_bad = false;
900 }
901
902 if (is_bad)
903 goto bad;
904
905 /* We were unable to find anything out about this entry, so
906 * let's investigate it later */
907 u->gc_marker = gc_marker + GC_OFFSET_UNSURE;
908 unit_add_to_gc_queue(u);
909 return;
910
911 bad:
912 /* We definitely know that this one is not useful anymore, so
913 * let's mark it for deletion */
914 u->gc_marker = gc_marker + GC_OFFSET_BAD;
915 unit_add_to_cleanup_queue(u);
916 return;
917
918 good:
919 unit_gc_mark_good(u, gc_marker);
920 }
921
922 static unsigned manager_dispatch_gc_queue(Manager *m) {
923 Unit *u;
924 unsigned n = 0;
925 unsigned gc_marker;
926
927 assert(m);
928
929 /* log_debug("Running GC..."); */
930
931 m->gc_marker += _GC_OFFSET_MAX;
932 if (m->gc_marker + _GC_OFFSET_MAX <= _GC_OFFSET_MAX)
933 m->gc_marker = 1;
934
935 gc_marker = m->gc_marker;
936
937 while ((u = m->gc_queue)) {
938 assert(u->in_gc_queue);
939
940 unit_gc_sweep(u, gc_marker);
941
942 LIST_REMOVE(gc_queue, m->gc_queue, u);
943 u->in_gc_queue = false;
944
945 n++;
946
947 if (u->gc_marker == gc_marker + GC_OFFSET_BAD ||
948 u->gc_marker == gc_marker + GC_OFFSET_UNSURE) {
949 if (u->id)
950 log_unit_debug(u, "Collecting.");
951 u->gc_marker = gc_marker + GC_OFFSET_BAD;
952 unit_add_to_cleanup_queue(u);
953 }
954 }
955
956 m->n_in_gc_queue = 0;
957
958 return n;
959 }
960
961 static void manager_clear_jobs_and_units(Manager *m) {
962 Unit *u;
963
964 assert(m);
965
966 while ((u = hashmap_first(m->units)))
967 unit_free(u);
968
969 manager_dispatch_cleanup_queue(m);
970
971 assert(!m->load_queue);
972 assert(!m->run_queue);
973 assert(!m->dbus_unit_queue);
974 assert(!m->dbus_job_queue);
975 assert(!m->cleanup_queue);
976 assert(!m->gc_queue);
977
978 assert(hashmap_isempty(m->jobs));
979 assert(hashmap_isempty(m->units));
980
981 m->n_on_console = 0;
982 m->n_running_jobs = 0;
983 }
984
985 Manager* manager_free(Manager *m) {
986 UnitType c;
987 int i;
988
989 if (!m)
990 return NULL;
991
992 manager_clear_jobs_and_units(m);
993
994 for (c = 0; c < _UNIT_TYPE_MAX; c++)
995 if (unit_vtable[c]->shutdown)
996 unit_vtable[c]->shutdown(m);
997
998 /* If we reexecute ourselves, we keep the root cgroup
999 * around */
1000 manager_shutdown_cgroup(m, m->exit_code != MANAGER_REEXECUTE);
1001
1002 lookup_paths_flush_generator(&m->lookup_paths);
1003
1004 bus_done(m);
1005
1006 dynamic_user_vacuum(m, false);
1007 hashmap_free(m->dynamic_users);
1008
1009 hashmap_free(m->units);
1010 hashmap_free(m->jobs);
1011 hashmap_free(m->watch_pids1);
1012 hashmap_free(m->watch_pids2);
1013 hashmap_free(m->watch_bus);
1014
1015 set_free(m->startup_units);
1016 set_free(m->failed_units);
1017
1018 sd_event_source_unref(m->signal_event_source);
1019 sd_event_source_unref(m->notify_event_source);
1020 sd_event_source_unref(m->cgroups_agent_event_source);
1021 sd_event_source_unref(m->time_change_event_source);
1022 sd_event_source_unref(m->jobs_in_progress_event_source);
1023 sd_event_source_unref(m->run_queue_event_source);
1024
1025 safe_close(m->signal_fd);
1026 safe_close(m->notify_fd);
1027 safe_close(m->cgroups_agent_fd);
1028 safe_close(m->time_change_fd);
1029 safe_close(m->kdbus_fd);
1030
1031 manager_close_ask_password(m);
1032
1033 manager_close_idle_pipe(m);
1034
1035 udev_unref(m->udev);
1036 sd_event_unref(m->event);
1037
1038 free(m->notify_socket);
1039
1040 lookup_paths_free(&m->lookup_paths);
1041 strv_free(m->environment);
1042
1043 hashmap_free(m->cgroup_unit);
1044 set_free_free(m->unit_path_cache);
1045
1046 free(m->switch_root);
1047 free(m->switch_root_init);
1048
1049 for (i = 0; i < _RLIMIT_MAX; i++)
1050 m->rlimit[i] = mfree(m->rlimit[i]);
1051
1052 assert(hashmap_isempty(m->units_requiring_mounts_for));
1053 hashmap_free(m->units_requiring_mounts_for);
1054
1055 free(m);
1056 return NULL;
1057 }
1058
1059 void manager_enumerate(Manager *m) {
1060 UnitType c;
1061
1062 assert(m);
1063
1064 /* Let's ask every type to load all units from disk/kernel
1065 * that it might know */
1066 for (c = 0; c < _UNIT_TYPE_MAX; c++) {
1067 if (!unit_type_supported(c)) {
1068 log_debug("Unit type .%s is not supported on this system.", unit_type_to_string(c));
1069 continue;
1070 }
1071
1072 if (!unit_vtable[c]->enumerate)
1073 continue;
1074
1075 unit_vtable[c]->enumerate(m);
1076 }
1077
1078 manager_dispatch_load_queue(m);
1079 }
1080
1081 static void manager_coldplug(Manager *m) {
1082 Iterator i;
1083 Unit *u;
1084 char *k;
1085 int r;
1086
1087 assert(m);
1088
1089 /* Then, let's set up their initial state. */
1090 HASHMAP_FOREACH_KEY(u, k, m->units, i) {
1091
1092 /* ignore aliases */
1093 if (u->id != k)
1094 continue;
1095
1096 r = unit_coldplug(u);
1097 if (r < 0)
1098 log_warning_errno(r, "We couldn't coldplug %s, proceeding anyway: %m", u->id);
1099 }
1100 }
1101
1102 static void manager_build_unit_path_cache(Manager *m) {
1103 char **i;
1104 int r;
1105
1106 assert(m);
1107
1108 set_free_free(m->unit_path_cache);
1109
1110 m->unit_path_cache = set_new(&string_hash_ops);
1111 if (!m->unit_path_cache) {
1112 r = -ENOMEM;
1113 goto fail;
1114 }
1115
1116 /* This simply builds a list of files we know exist, so that
1117 * we don't always have to go to disk */
1118
1119 STRV_FOREACH(i, m->lookup_paths.search_path) {
1120 _cleanup_closedir_ DIR *d = NULL;
1121 struct dirent *de;
1122
1123 d = opendir(*i);
1124 if (!d) {
1125 if (errno != ENOENT)
1126 log_warning_errno(errno, "Failed to open directory %s, ignoring: %m", *i);
1127 continue;
1128 }
1129
1130 FOREACH_DIRENT(de, d, r = -errno; goto fail) {
1131 char *p;
1132
1133 p = strjoin(streq(*i, "/") ? "" : *i, "/", de->d_name, NULL);
1134 if (!p) {
1135 r = -ENOMEM;
1136 goto fail;
1137 }
1138
1139 r = set_consume(m->unit_path_cache, p);
1140 if (r < 0)
1141 goto fail;
1142 }
1143 }
1144
1145 return;
1146
1147 fail:
1148 log_warning_errno(r, "Failed to build unit path cache, proceeding without: %m");
1149 m->unit_path_cache = set_free_free(m->unit_path_cache);
1150 }
1151
1152 static void manager_distribute_fds(Manager *m, FDSet *fds) {
1153 Iterator i;
1154 Unit *u;
1155
1156 assert(m);
1157
1158 HASHMAP_FOREACH(u, m->units, i) {
1159
1160 if (fdset_size(fds) <= 0)
1161 break;
1162
1163 if (!UNIT_VTABLE(u)->distribute_fds)
1164 continue;
1165
1166 UNIT_VTABLE(u)->distribute_fds(u, fds);
1167 }
1168 }
1169
1170 int manager_startup(Manager *m, FILE *serialization, FDSet *fds) {
1171 int r, q;
1172
1173 assert(m);
1174
1175 r = lookup_paths_init(&m->lookup_paths, m->unit_file_scope, 0, NULL);
1176 if (r < 0)
1177 return r;
1178
1179 /* Make sure the transient directory always exists, so that it remains in the search path */
1180 r = mkdir_p_label(m->lookup_paths.transient, 0755);
1181 if (r < 0)
1182 return r;
1183
1184 dual_timestamp_get(&m->generators_start_timestamp);
1185 r = manager_run_generators(m);
1186 dual_timestamp_get(&m->generators_finish_timestamp);
1187 if (r < 0)
1188 return r;
1189
1190 lookup_paths_reduce(&m->lookup_paths);
1191 manager_build_unit_path_cache(m);
1192
1193 /* If we will deserialize make sure that during enumeration
1194 * this is already known, so we increase the counter here
1195 * already */
1196 if (serialization)
1197 m->n_reloading++;
1198
1199 /* First, enumerate what we can from all config files */
1200 dual_timestamp_get(&m->units_load_start_timestamp);
1201 manager_enumerate(m);
1202 dual_timestamp_get(&m->units_load_finish_timestamp);
1203
1204 /* Second, deserialize if there is something to deserialize */
1205 if (serialization)
1206 r = manager_deserialize(m, serialization, fds);
1207
1208 /* Any fds left? Find some unit which wants them. This is
1209 * useful to allow container managers to pass some file
1210 * descriptors to us pre-initialized. This enables
1211 * socket-based activation of entire containers. */
1212 manager_distribute_fds(m, fds);
1213
1214 /* We might have deserialized the notify fd, but if we didn't
1215 * then let's create the bus now */
1216 q = manager_setup_notify(m);
1217 if (q < 0 && r == 0)
1218 r = q;
1219
1220 q = manager_setup_cgroups_agent(m);
1221 if (q < 0 && r == 0)
1222 r = q;
1223
1224 /* We might have deserialized the kdbus control fd, but if we
1225 * didn't, then let's create the bus now. */
1226 manager_connect_bus(m, !!serialization);
1227 bus_track_coldplug(m, &m->subscribed, &m->deserialized_subscribed);
1228
1229 /* Third, fire things up! */
1230 manager_coldplug(m);
1231
1232 /* Release any dynamic users no longer referenced */
1233 dynamic_user_vacuum(m, true);
1234
1235 if (serialization) {
1236 assert(m->n_reloading > 0);
1237 m->n_reloading--;
1238
1239 /* Let's wait for the UnitNew/JobNew messages being
1240 * sent, before we notify that the reload is
1241 * finished */
1242 m->send_reloading_done = true;
1243 }
1244
1245 return r;
1246 }
1247
1248 int manager_add_job(Manager *m, JobType type, Unit *unit, JobMode mode, sd_bus_error *e, Job **_ret) {
1249 int r;
1250 Transaction *tr;
1251
1252 assert(m);
1253 assert(type < _JOB_TYPE_MAX);
1254 assert(unit);
1255 assert(mode < _JOB_MODE_MAX);
1256
1257 if (mode == JOB_ISOLATE && type != JOB_START)
1258 return sd_bus_error_setf(e, SD_BUS_ERROR_INVALID_ARGS, "Isolate is only valid for start.");
1259
1260 if (mode == JOB_ISOLATE && !unit->allow_isolate)
1261 return sd_bus_error_setf(e, BUS_ERROR_NO_ISOLATION, "Operation refused, unit may not be isolated.");
1262
1263 log_unit_debug(unit, "Trying to enqueue job %s/%s/%s", unit->id, job_type_to_string(type), job_mode_to_string(mode));
1264
1265 type = job_type_collapse(type, unit);
1266
1267 tr = transaction_new(mode == JOB_REPLACE_IRREVERSIBLY);
1268 if (!tr)
1269 return -ENOMEM;
1270
1271 r = transaction_add_job_and_dependencies(tr, type, unit, NULL, true, false,
1272 mode == JOB_IGNORE_DEPENDENCIES || mode == JOB_IGNORE_REQUIREMENTS,
1273 mode == JOB_IGNORE_DEPENDENCIES, e);
1274 if (r < 0)
1275 goto tr_abort;
1276
1277 if (mode == JOB_ISOLATE) {
1278 r = transaction_add_isolate_jobs(tr, m);
1279 if (r < 0)
1280 goto tr_abort;
1281 }
1282
1283 r = transaction_activate(tr, m, mode, e);
1284 if (r < 0)
1285 goto tr_abort;
1286
1287 log_unit_debug(unit,
1288 "Enqueued job %s/%s as %u", unit->id,
1289 job_type_to_string(type), (unsigned) tr->anchor_job->id);
1290
1291 if (_ret)
1292 *_ret = tr->anchor_job;
1293
1294 transaction_free(tr);
1295 return 0;
1296
1297 tr_abort:
1298 transaction_abort(tr);
1299 transaction_free(tr);
1300 return r;
1301 }
1302
1303 int manager_add_job_by_name(Manager *m, JobType type, const char *name, JobMode mode, sd_bus_error *e, Job **ret) {
1304 Unit *unit;
1305 int r;
1306
1307 assert(m);
1308 assert(type < _JOB_TYPE_MAX);
1309 assert(name);
1310 assert(mode < _JOB_MODE_MAX);
1311
1312 r = manager_load_unit(m, name, NULL, NULL, &unit);
1313 if (r < 0)
1314 return r;
1315
1316 return manager_add_job(m, type, unit, mode, e, ret);
1317 }
1318
1319 int manager_add_job_by_name_and_warn(Manager *m, JobType type, const char *name, JobMode mode, Job **ret) {
1320 _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL;
1321 int r;
1322
1323 assert(m);
1324 assert(type < _JOB_TYPE_MAX);
1325 assert(name);
1326 assert(mode < _JOB_MODE_MAX);
1327
1328 r = manager_add_job_by_name(m, type, name, mode, &error, ret);
1329 if (r < 0)
1330 return log_warning_errno(r, "Failed to enqueue %s job for %s: %s", job_mode_to_string(mode), name, bus_error_message(&error, r));
1331
1332 return r;
1333 }
1334
1335 Job *manager_get_job(Manager *m, uint32_t id) {
1336 assert(m);
1337
1338 return hashmap_get(m->jobs, UINT32_TO_PTR(id));
1339 }
1340
1341 Unit *manager_get_unit(Manager *m, const char *name) {
1342 assert(m);
1343 assert(name);
1344
1345 return hashmap_get(m->units, name);
1346 }
1347
1348 unsigned manager_dispatch_load_queue(Manager *m) {
1349 Unit *u;
1350 unsigned n = 0;
1351
1352 assert(m);
1353
1354 /* Make sure we are not run recursively */
1355 if (m->dispatching_load_queue)
1356 return 0;
1357
1358 m->dispatching_load_queue = true;
1359
1360 /* Dispatches the load queue. Takes a unit from the queue and
1361 * tries to load its data until the queue is empty */
1362
1363 while ((u = m->load_queue)) {
1364 assert(u->in_load_queue);
1365
1366 unit_load(u);
1367 n++;
1368 }
1369
1370 m->dispatching_load_queue = false;
1371 return n;
1372 }
1373
1374 int manager_load_unit_prepare(
1375 Manager *m,
1376 const char *name,
1377 const char *path,
1378 sd_bus_error *e,
1379 Unit **_ret) {
1380
1381 Unit *ret;
1382 UnitType t;
1383 int r;
1384
1385 assert(m);
1386 assert(name || path);
1387
1388 /* This will prepare the unit for loading, but not actually
1389 * load anything from disk. */
1390
1391 if (path && !is_path(path))
1392 return sd_bus_error_setf(e, SD_BUS_ERROR_INVALID_ARGS, "Path %s is not absolute.", path);
1393
1394 if (!name)
1395 name = basename(path);
1396
1397 t = unit_name_to_type(name);
1398
1399 if (t == _UNIT_TYPE_INVALID || !unit_name_is_valid(name, UNIT_NAME_PLAIN|UNIT_NAME_INSTANCE)) {
1400 if (unit_name_is_valid(name, UNIT_NAME_TEMPLATE))
1401 return sd_bus_error_setf(e, SD_BUS_ERROR_INVALID_ARGS, "Unit name %s is missing the instance name.", name);
1402
1403 return sd_bus_error_setf(e, SD_BUS_ERROR_INVALID_ARGS, "Unit name %s is not valid.", name);
1404 }
1405
1406 ret = manager_get_unit(m, name);
1407 if (ret) {
1408 *_ret = ret;
1409 return 1;
1410 }
1411
1412 ret = unit_new(m, unit_vtable[t]->object_size);
1413 if (!ret)
1414 return -ENOMEM;
1415
1416 if (path) {
1417 ret->fragment_path = strdup(path);
1418 if (!ret->fragment_path) {
1419 unit_free(ret);
1420 return -ENOMEM;
1421 }
1422 }
1423
1424 r = unit_add_name(ret, name);
1425 if (r < 0) {
1426 unit_free(ret);
1427 return r;
1428 }
1429
1430 unit_add_to_load_queue(ret);
1431 unit_add_to_dbus_queue(ret);
1432 unit_add_to_gc_queue(ret);
1433
1434 if (_ret)
1435 *_ret = ret;
1436
1437 return 0;
1438 }
1439
1440 int manager_load_unit(
1441 Manager *m,
1442 const char *name,
1443 const char *path,
1444 sd_bus_error *e,
1445 Unit **_ret) {
1446
1447 int r;
1448
1449 assert(m);
1450
1451 /* This will load the service information files, but not actually
1452 * start any services or anything. */
1453
1454 r = manager_load_unit_prepare(m, name, path, e, _ret);
1455 if (r != 0)
1456 return r;
1457
1458 manager_dispatch_load_queue(m);
1459
1460 if (_ret)
1461 *_ret = unit_follow_merge(*_ret);
1462
1463 return 0;
1464 }
1465
1466 void manager_dump_jobs(Manager *s, FILE *f, const char *prefix) {
1467 Iterator i;
1468 Job *j;
1469
1470 assert(s);
1471 assert(f);
1472
1473 HASHMAP_FOREACH(j, s->jobs, i)
1474 job_dump(j, f, prefix);
1475 }
1476
1477 void manager_dump_units(Manager *s, FILE *f, const char *prefix) {
1478 Iterator i;
1479 Unit *u;
1480 const char *t;
1481
1482 assert(s);
1483 assert(f);
1484
1485 HASHMAP_FOREACH_KEY(u, t, s->units, i)
1486 if (u->id == t)
1487 unit_dump(u, f, prefix);
1488 }
1489
1490 void manager_clear_jobs(Manager *m) {
1491 Job *j;
1492
1493 assert(m);
1494
1495 while ((j = hashmap_first(m->jobs)))
1496 /* No need to recurse. We're cancelling all jobs. */
1497 job_finish_and_invalidate(j, JOB_CANCELED, false, false);
1498 }
1499
1500 static int manager_dispatch_run_queue(sd_event_source *source, void *userdata) {
1501 Manager *m = userdata;
1502 Job *j;
1503
1504 assert(source);
1505 assert(m);
1506
1507 while ((j = m->run_queue)) {
1508 assert(j->installed);
1509 assert(j->in_run_queue);
1510
1511 job_run_and_invalidate(j);
1512 }
1513
1514 if (m->n_running_jobs > 0)
1515 manager_watch_jobs_in_progress(m);
1516
1517 if (m->n_on_console > 0)
1518 manager_watch_idle_pipe(m);
1519
1520 return 1;
1521 }
1522
1523 static unsigned manager_dispatch_dbus_queue(Manager *m) {
1524 Job *j;
1525 Unit *u;
1526 unsigned n = 0;
1527
1528 assert(m);
1529
1530 if (m->dispatching_dbus_queue)
1531 return 0;
1532
1533 m->dispatching_dbus_queue = true;
1534
1535 while ((u = m->dbus_unit_queue)) {
1536 assert(u->in_dbus_queue);
1537
1538 bus_unit_send_change_signal(u);
1539 n++;
1540 }
1541
1542 while ((j = m->dbus_job_queue)) {
1543 assert(j->in_dbus_queue);
1544
1545 bus_job_send_change_signal(j);
1546 n++;
1547 }
1548
1549 m->dispatching_dbus_queue = false;
1550
1551 if (m->send_reloading_done) {
1552 m->send_reloading_done = false;
1553
1554 bus_manager_send_reloading(m, false);
1555 }
1556
1557 if (m->queued_message)
1558 bus_send_queued_message(m);
1559
1560 return n;
1561 }
1562
1563 static int manager_dispatch_cgroups_agent_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
1564 Manager *m = userdata;
1565 char buf[PATH_MAX+1];
1566 ssize_t n;
1567
1568 n = recv(fd, buf, sizeof(buf), 0);
1569 if (n < 0)
1570 return log_error_errno(errno, "Failed to read cgroups agent message: %m");
1571 if (n == 0) {
1572 log_error("Got zero-length cgroups agent message, ignoring.");
1573 return 0;
1574 }
1575 if ((size_t) n >= sizeof(buf)) {
1576 log_error("Got overly long cgroups agent message, ignoring.");
1577 return 0;
1578 }
1579
1580 if (memchr(buf, 0, n)) {
1581 log_error("Got cgroups agent message with embedded NUL byte, ignoring.");
1582 return 0;
1583 }
1584 buf[n] = 0;
1585
1586 manager_notify_cgroup_empty(m, buf);
1587 bus_forward_agent_released(m, buf);
1588
1589 return 0;
1590 }
1591
1592 static void manager_invoke_notify_message(Manager *m, Unit *u, pid_t pid, const char *buf, size_t n, FDSet *fds) {
1593 _cleanup_strv_free_ char **tags = NULL;
1594
1595 assert(m);
1596 assert(u);
1597 assert(buf);
1598 assert(n > 0);
1599
1600 tags = strv_split(buf, "\n\r");
1601 if (!tags) {
1602 log_oom();
1603 return;
1604 }
1605
1606 if (UNIT_VTABLE(u)->notify_message)
1607 UNIT_VTABLE(u)->notify_message(u, pid, tags, fds);
1608 else
1609 log_unit_debug(u, "Got notification message for unit. Ignoring.");
1610 }
1611
1612 static int manager_dispatch_notify_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
1613
1614 _cleanup_fdset_free_ FDSet *fds = NULL;
1615 Manager *m = userdata;
1616 char buf[NOTIFY_BUFFER_MAX+1];
1617 struct iovec iovec = {
1618 .iov_base = buf,
1619 .iov_len = sizeof(buf)-1,
1620 };
1621 union {
1622 struct cmsghdr cmsghdr;
1623 uint8_t buf[CMSG_SPACE(sizeof(struct ucred)) +
1624 CMSG_SPACE(sizeof(int) * NOTIFY_FD_MAX)];
1625 } control = {};
1626 struct msghdr msghdr = {
1627 .msg_iov = &iovec,
1628 .msg_iovlen = 1,
1629 .msg_control = &control,
1630 .msg_controllen = sizeof(control),
1631 };
1632
1633 struct cmsghdr *cmsg;
1634 struct ucred *ucred = NULL;
1635 bool found = false;
1636 Unit *u1, *u2, *u3;
1637 int r, *fd_array = NULL;
1638 unsigned n_fds = 0;
1639 ssize_t n;
1640
1641 assert(m);
1642 assert(m->notify_fd == fd);
1643
1644 if (revents != EPOLLIN) {
1645 log_warning("Got unexpected poll event for notify fd.");
1646 return 0;
1647 }
1648
1649 n = recvmsg(m->notify_fd, &msghdr, MSG_DONTWAIT|MSG_CMSG_CLOEXEC);
1650 if (n < 0) {
1651 if (errno == EAGAIN || errno == EINTR)
1652 return 0;
1653
1654 return -errno;
1655 }
1656
1657 CMSG_FOREACH(cmsg, &msghdr) {
1658 if (cmsg->cmsg_level == SOL_SOCKET && cmsg->cmsg_type == SCM_RIGHTS) {
1659
1660 fd_array = (int*) CMSG_DATA(cmsg);
1661 n_fds = (cmsg->cmsg_len - CMSG_LEN(0)) / sizeof(int);
1662
1663 } else if (cmsg->cmsg_level == SOL_SOCKET &&
1664 cmsg->cmsg_type == SCM_CREDENTIALS &&
1665 cmsg->cmsg_len == CMSG_LEN(sizeof(struct ucred))) {
1666
1667 ucred = (struct ucred*) CMSG_DATA(cmsg);
1668 }
1669 }
1670
1671 if (n_fds > 0) {
1672 assert(fd_array);
1673
1674 r = fdset_new_array(&fds, fd_array, n_fds);
1675 if (r < 0) {
1676 close_many(fd_array, n_fds);
1677 return log_oom();
1678 }
1679 }
1680
1681 if (!ucred || ucred->pid <= 0) {
1682 log_warning("Received notify message without valid credentials. Ignoring.");
1683 return 0;
1684 }
1685
1686 if ((size_t) n >= sizeof(buf)) {
1687 log_warning("Received notify message exceeded maximum size. Ignoring.");
1688 return 0;
1689 }
1690
1691 buf[n] = 0;
1692
1693 /* Notify every unit that might be interested, but try
1694 * to avoid notifying the same one multiple times. */
1695 u1 = manager_get_unit_by_pid_cgroup(m, ucred->pid);
1696 if (u1) {
1697 manager_invoke_notify_message(m, u1, ucred->pid, buf, n, fds);
1698 found = true;
1699 }
1700
1701 u2 = hashmap_get(m->watch_pids1, PID_TO_PTR(ucred->pid));
1702 if (u2 && u2 != u1) {
1703 manager_invoke_notify_message(m, u2, ucred->pid, buf, n, fds);
1704 found = true;
1705 }
1706
1707 u3 = hashmap_get(m->watch_pids2, PID_TO_PTR(ucred->pid));
1708 if (u3 && u3 != u2 && u3 != u1) {
1709 manager_invoke_notify_message(m, u3, ucred->pid, buf, n, fds);
1710 found = true;
1711 }
1712
1713 if (!found)
1714 log_warning("Cannot find unit for notify message of PID "PID_FMT".", ucred->pid);
1715
1716 if (fdset_size(fds) > 0)
1717 log_warning("Got auxiliary fds with notification message, closing all.");
1718
1719 return 0;
1720 }
1721
1722 static void invoke_sigchld_event(Manager *m, Unit *u, const siginfo_t *si) {
1723 uint64_t iteration;
1724
1725 assert(m);
1726 assert(u);
1727 assert(si);
1728
1729 sd_event_get_iteration(m->event, &iteration);
1730
1731 log_unit_debug(u, "Child "PID_FMT" belongs to %s", si->si_pid, u->id);
1732
1733 unit_unwatch_pid(u, si->si_pid);
1734
1735 if (UNIT_VTABLE(u)->sigchld_event) {
1736 if (set_size(u->pids) <= 1 ||
1737 iteration != u->sigchldgen ||
1738 unit_main_pid(u) == si->si_pid ||
1739 unit_control_pid(u) == si->si_pid) {
1740 UNIT_VTABLE(u)->sigchld_event(u, si->si_pid, si->si_code, si->si_status);
1741 u->sigchldgen = iteration;
1742 } else
1743 log_debug("%s already issued a sigchld this iteration %" PRIu64 ", skipping. Pids still being watched %d", u->id, iteration, set_size(u->pids));
1744 }
1745 }
1746
1747 static int manager_dispatch_sigchld(Manager *m) {
1748 assert(m);
1749
1750 for (;;) {
1751 siginfo_t si = {};
1752
1753 /* First we call waitd() for a PID and do not reap the
1754 * zombie. That way we can still access /proc/$PID for
1755 * it while it is a zombie. */
1756 if (waitid(P_ALL, 0, &si, WEXITED|WNOHANG|WNOWAIT) < 0) {
1757
1758 if (errno == ECHILD)
1759 break;
1760
1761 if (errno == EINTR)
1762 continue;
1763
1764 return -errno;
1765 }
1766
1767 if (si.si_pid <= 0)
1768 break;
1769
1770 if (si.si_code == CLD_EXITED || si.si_code == CLD_KILLED || si.si_code == CLD_DUMPED) {
1771 _cleanup_free_ char *name = NULL;
1772 Unit *u1, *u2, *u3;
1773
1774 get_process_comm(si.si_pid, &name);
1775
1776 log_debug("Child "PID_FMT" (%s) died (code=%s, status=%i/%s)",
1777 si.si_pid, strna(name),
1778 sigchld_code_to_string(si.si_code),
1779 si.si_status,
1780 strna(si.si_code == CLD_EXITED
1781 ? exit_status_to_string(si.si_status, EXIT_STATUS_FULL)
1782 : signal_to_string(si.si_status)));
1783
1784 /* And now figure out the unit this belongs
1785 * to, it might be multiple... */
1786 u1 = manager_get_unit_by_pid_cgroup(m, si.si_pid);
1787 if (u1)
1788 invoke_sigchld_event(m, u1, &si);
1789 u2 = hashmap_get(m->watch_pids1, PID_TO_PTR(si.si_pid));
1790 if (u2 && u2 != u1)
1791 invoke_sigchld_event(m, u2, &si);
1792 u3 = hashmap_get(m->watch_pids2, PID_TO_PTR(si.si_pid));
1793 if (u3 && u3 != u2 && u3 != u1)
1794 invoke_sigchld_event(m, u3, &si);
1795 }
1796
1797 /* And now, we actually reap the zombie. */
1798 if (waitid(P_PID, si.si_pid, &si, WEXITED) < 0) {
1799 if (errno == EINTR)
1800 continue;
1801
1802 return -errno;
1803 }
1804 }
1805
1806 return 0;
1807 }
1808
1809 static int manager_start_target(Manager *m, const char *name, JobMode mode) {
1810 _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL;
1811 int r;
1812
1813 log_debug("Activating special unit %s", name);
1814
1815 r = manager_add_job_by_name(m, JOB_START, name, mode, &error, NULL);
1816 if (r < 0)
1817 log_error("Failed to enqueue %s job: %s", name, bus_error_message(&error, r));
1818
1819 return r;
1820 }
1821
1822 static int manager_dispatch_signal_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
1823 Manager *m = userdata;
1824 ssize_t n;
1825 struct signalfd_siginfo sfsi;
1826 bool sigchld = false;
1827 int r;
1828
1829 assert(m);
1830 assert(m->signal_fd == fd);
1831
1832 if (revents != EPOLLIN) {
1833 log_warning("Got unexpected events from signal file descriptor.");
1834 return 0;
1835 }
1836
1837 for (;;) {
1838 n = read(m->signal_fd, &sfsi, sizeof(sfsi));
1839 if (n != sizeof(sfsi)) {
1840
1841 if (n >= 0)
1842 return -EIO;
1843
1844 if (errno == EINTR || errno == EAGAIN)
1845 break;
1846
1847 return -errno;
1848 }
1849
1850 log_received_signal(sfsi.ssi_signo == SIGCHLD ||
1851 (sfsi.ssi_signo == SIGTERM && MANAGER_IS_USER(m))
1852 ? LOG_DEBUG : LOG_INFO,
1853 &sfsi);
1854
1855 switch (sfsi.ssi_signo) {
1856
1857 case SIGCHLD:
1858 sigchld = true;
1859 break;
1860
1861 case SIGTERM:
1862 if (MANAGER_IS_SYSTEM(m)) {
1863 /* This is for compatibility with the
1864 * original sysvinit */
1865 m->exit_code = MANAGER_REEXECUTE;
1866 break;
1867 }
1868
1869 /* Fall through */
1870
1871 case SIGINT:
1872 if (MANAGER_IS_SYSTEM(m)) {
1873
1874 /* If the user presses C-A-D more than
1875 * 7 times within 2s, we reboot
1876 * immediately. */
1877
1878 if (ratelimit_test(&m->ctrl_alt_del_ratelimit))
1879 manager_start_target(m, SPECIAL_CTRL_ALT_DEL_TARGET, JOB_REPLACE_IRREVERSIBLY);
1880 else {
1881 log_notice("Ctrl-Alt-Del was pressed more than 7 times within 2s, rebooting immediately.");
1882 status_printf(NULL, true, false, "Ctrl-Alt-Del was pressed more than 7 times within 2s, rebooting immediately.");
1883 m->exit_code = MANAGER_REBOOT;
1884 }
1885
1886 break;
1887 }
1888
1889 /* Run the exit target if there is one, if not, just exit. */
1890 if (manager_start_target(m, SPECIAL_EXIT_TARGET, JOB_REPLACE) < 0) {
1891 m->exit_code = MANAGER_EXIT;
1892 return 0;
1893 }
1894
1895 break;
1896
1897 case SIGWINCH:
1898 if (MANAGER_IS_SYSTEM(m))
1899 manager_start_target(m, SPECIAL_KBREQUEST_TARGET, JOB_REPLACE);
1900
1901 /* This is a nop on non-init */
1902 break;
1903
1904 case SIGPWR:
1905 if (MANAGER_IS_SYSTEM(m))
1906 manager_start_target(m, SPECIAL_SIGPWR_TARGET, JOB_REPLACE);
1907
1908 /* This is a nop on non-init */
1909 break;
1910
1911 case SIGUSR1: {
1912 Unit *u;
1913
1914 u = manager_get_unit(m, SPECIAL_DBUS_SERVICE);
1915
1916 if (!u || UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u))) {
1917 log_info("Trying to reconnect to bus...");
1918 bus_init(m, true);
1919 }
1920
1921 if (!u || !UNIT_IS_ACTIVE_OR_ACTIVATING(unit_active_state(u))) {
1922 log_info("Loading D-Bus service...");
1923 manager_start_target(m, SPECIAL_DBUS_SERVICE, JOB_REPLACE);
1924 }
1925
1926 break;
1927 }
1928
1929 case SIGUSR2: {
1930 _cleanup_free_ char *dump = NULL;
1931 _cleanup_fclose_ FILE *f = NULL;
1932 size_t size;
1933
1934 f = open_memstream(&dump, &size);
1935 if (!f) {
1936 log_warning_errno(errno, "Failed to allocate memory stream: %m");
1937 break;
1938 }
1939
1940 manager_dump_units(m, f, "\t");
1941 manager_dump_jobs(m, f, "\t");
1942
1943 r = fflush_and_check(f);
1944 if (r < 0) {
1945 log_warning_errno(r, "Failed to write status stream: %m");
1946 break;
1947 }
1948
1949 log_dump(LOG_INFO, dump);
1950 break;
1951 }
1952
1953 case SIGHUP:
1954 m->exit_code = MANAGER_RELOAD;
1955 break;
1956
1957 default: {
1958
1959 /* Starting SIGRTMIN+0 */
1960 static const char * const target_table[] = {
1961 [0] = SPECIAL_DEFAULT_TARGET,
1962 [1] = SPECIAL_RESCUE_TARGET,
1963 [2] = SPECIAL_EMERGENCY_TARGET,
1964 [3] = SPECIAL_HALT_TARGET,
1965 [4] = SPECIAL_POWEROFF_TARGET,
1966 [5] = SPECIAL_REBOOT_TARGET,
1967 [6] = SPECIAL_KEXEC_TARGET
1968 };
1969
1970 /* Starting SIGRTMIN+13, so that target halt and system halt are 10 apart */
1971 static const ManagerExitCode code_table[] = {
1972 [0] = MANAGER_HALT,
1973 [1] = MANAGER_POWEROFF,
1974 [2] = MANAGER_REBOOT,
1975 [3] = MANAGER_KEXEC
1976 };
1977
1978 if ((int) sfsi.ssi_signo >= SIGRTMIN+0 &&
1979 (int) sfsi.ssi_signo < SIGRTMIN+(int) ELEMENTSOF(target_table)) {
1980 int idx = (int) sfsi.ssi_signo - SIGRTMIN;
1981 manager_start_target(m, target_table[idx],
1982 (idx == 1 || idx == 2) ? JOB_ISOLATE : JOB_REPLACE);
1983 break;
1984 }
1985
1986 if ((int) sfsi.ssi_signo >= SIGRTMIN+13 &&
1987 (int) sfsi.ssi_signo < SIGRTMIN+13+(int) ELEMENTSOF(code_table)) {
1988 m->exit_code = code_table[sfsi.ssi_signo - SIGRTMIN - 13];
1989 break;
1990 }
1991
1992 switch (sfsi.ssi_signo - SIGRTMIN) {
1993
1994 case 20:
1995 manager_set_show_status(m, SHOW_STATUS_YES);
1996 break;
1997
1998 case 21:
1999 manager_set_show_status(m, SHOW_STATUS_NO);
2000 break;
2001
2002 case 22:
2003 log_set_max_level(LOG_DEBUG);
2004 log_info("Setting log level to debug.");
2005 break;
2006
2007 case 23:
2008 log_set_max_level(LOG_INFO);
2009 log_info("Setting log level to info.");
2010 break;
2011
2012 case 24:
2013 if (MANAGER_IS_USER(m)) {
2014 m->exit_code = MANAGER_EXIT;
2015 return 0;
2016 }
2017
2018 /* This is a nop on init */
2019 break;
2020
2021 case 26:
2022 case 29: /* compatibility: used to be mapped to LOG_TARGET_SYSLOG_OR_KMSG */
2023 log_set_target(LOG_TARGET_JOURNAL_OR_KMSG);
2024 log_notice("Setting log target to journal-or-kmsg.");
2025 break;
2026
2027 case 27:
2028 log_set_target(LOG_TARGET_CONSOLE);
2029 log_notice("Setting log target to console.");
2030 break;
2031
2032 case 28:
2033 log_set_target(LOG_TARGET_KMSG);
2034 log_notice("Setting log target to kmsg.");
2035 break;
2036
2037 default:
2038 log_warning("Got unhandled signal <%s>.", signal_to_string(sfsi.ssi_signo));
2039 }
2040 }
2041 }
2042 }
2043
2044 if (sigchld)
2045 manager_dispatch_sigchld(m);
2046
2047 return 0;
2048 }
2049
2050 static int manager_dispatch_time_change_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
2051 Manager *m = userdata;
2052 Iterator i;
2053 Unit *u;
2054
2055 assert(m);
2056 assert(m->time_change_fd == fd);
2057
2058 log_struct(LOG_INFO,
2059 LOG_MESSAGE_ID(SD_MESSAGE_TIME_CHANGE),
2060 LOG_MESSAGE("Time has been changed"),
2061 NULL);
2062
2063 /* Restart the watch */
2064 m->time_change_event_source = sd_event_source_unref(m->time_change_event_source);
2065 m->time_change_fd = safe_close(m->time_change_fd);
2066
2067 manager_setup_time_change(m);
2068
2069 HASHMAP_FOREACH(u, m->units, i)
2070 if (UNIT_VTABLE(u)->time_change)
2071 UNIT_VTABLE(u)->time_change(u);
2072
2073 return 0;
2074 }
2075
2076 static int manager_dispatch_idle_pipe_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
2077 Manager *m = userdata;
2078
2079 assert(m);
2080 assert(m->idle_pipe[2] == fd);
2081
2082 m->no_console_output = m->n_on_console > 0;
2083
2084 manager_close_idle_pipe(m);
2085
2086 return 0;
2087 }
2088
2089 static int manager_dispatch_jobs_in_progress(sd_event_source *source, usec_t usec, void *userdata) {
2090 Manager *m = userdata;
2091 int r;
2092 uint64_t next;
2093
2094 assert(m);
2095 assert(source);
2096
2097 manager_print_jobs_in_progress(m);
2098
2099 next = now(CLOCK_MONOTONIC) + JOBS_IN_PROGRESS_PERIOD_USEC;
2100 r = sd_event_source_set_time(source, next);
2101 if (r < 0)
2102 return r;
2103
2104 return sd_event_source_set_enabled(source, SD_EVENT_ONESHOT);
2105 }
2106
2107 int manager_loop(Manager *m) {
2108 int r;
2109
2110 RATELIMIT_DEFINE(rl, 1*USEC_PER_SEC, 50000);
2111
2112 assert(m);
2113 m->exit_code = MANAGER_OK;
2114
2115 /* Release the path cache */
2116 m->unit_path_cache = set_free_free(m->unit_path_cache);
2117
2118 manager_check_finished(m);
2119
2120 /* There might still be some zombies hanging around from
2121 * before we were exec()'ed. Let's reap them. */
2122 r = manager_dispatch_sigchld(m);
2123 if (r < 0)
2124 return r;
2125
2126 while (m->exit_code == MANAGER_OK) {
2127 usec_t wait_usec;
2128
2129 if (m->runtime_watchdog > 0 && m->runtime_watchdog != USEC_INFINITY && MANAGER_IS_SYSTEM(m))
2130 watchdog_ping();
2131
2132 if (!ratelimit_test(&rl)) {
2133 /* Yay, something is going seriously wrong, pause a little */
2134 log_warning("Looping too fast. Throttling execution a little.");
2135 sleep(1);
2136 }
2137
2138 if (manager_dispatch_load_queue(m) > 0)
2139 continue;
2140
2141 if (manager_dispatch_gc_queue(m) > 0)
2142 continue;
2143
2144 if (manager_dispatch_cleanup_queue(m) > 0)
2145 continue;
2146
2147 if (manager_dispatch_cgroup_queue(m) > 0)
2148 continue;
2149
2150 if (manager_dispatch_dbus_queue(m) > 0)
2151 continue;
2152
2153 /* Sleep for half the watchdog time */
2154 if (m->runtime_watchdog > 0 && m->runtime_watchdog != USEC_INFINITY && MANAGER_IS_SYSTEM(m)) {
2155 wait_usec = m->runtime_watchdog / 2;
2156 if (wait_usec <= 0)
2157 wait_usec = 1;
2158 } else
2159 wait_usec = USEC_INFINITY;
2160
2161 r = sd_event_run(m->event, wait_usec);
2162 if (r < 0)
2163 return log_error_errno(r, "Failed to run event loop: %m");
2164 }
2165
2166 return m->exit_code;
2167 }
2168
2169 int manager_load_unit_from_dbus_path(Manager *m, const char *s, sd_bus_error *e, Unit **_u) {
2170 _cleanup_free_ char *n = NULL;
2171 Unit *u;
2172 int r;
2173
2174 assert(m);
2175 assert(s);
2176 assert(_u);
2177
2178 r = unit_name_from_dbus_path(s, &n);
2179 if (r < 0)
2180 return r;
2181
2182 r = manager_load_unit(m, n, NULL, e, &u);
2183 if (r < 0)
2184 return r;
2185
2186 *_u = u;
2187
2188 return 0;
2189 }
2190
2191 int manager_get_job_from_dbus_path(Manager *m, const char *s, Job **_j) {
2192 const char *p;
2193 unsigned id;
2194 Job *j;
2195 int r;
2196
2197 assert(m);
2198 assert(s);
2199 assert(_j);
2200
2201 p = startswith(s, "/org/freedesktop/systemd1/job/");
2202 if (!p)
2203 return -EINVAL;
2204
2205 r = safe_atou(p, &id);
2206 if (r < 0)
2207 return r;
2208
2209 j = manager_get_job(m, id);
2210 if (!j)
2211 return -ENOENT;
2212
2213 *_j = j;
2214
2215 return 0;
2216 }
2217
2218 void manager_send_unit_audit(Manager *m, Unit *u, int type, bool success) {
2219
2220 #ifdef HAVE_AUDIT
2221 _cleanup_free_ char *p = NULL;
2222 const char *msg;
2223 int audit_fd, r;
2224
2225 if (!MANAGER_IS_SYSTEM(m))
2226 return;
2227
2228 audit_fd = get_audit_fd();
2229 if (audit_fd < 0)
2230 return;
2231
2232 /* Don't generate audit events if the service was already
2233 * started and we're just deserializing */
2234 if (MANAGER_IS_RELOADING(m))
2235 return;
2236
2237 if (u->type != UNIT_SERVICE)
2238 return;
2239
2240 r = unit_name_to_prefix_and_instance(u->id, &p);
2241 if (r < 0) {
2242 log_error_errno(r, "Failed to extract prefix and instance of unit name: %m");
2243 return;
2244 }
2245
2246 msg = strjoina("unit=", p);
2247 if (audit_log_user_comm_message(audit_fd, type, msg, "systemd", NULL, NULL, NULL, success) < 0) {
2248 if (errno == EPERM)
2249 /* We aren't allowed to send audit messages?
2250 * Then let's not retry again. */
2251 close_audit_fd();
2252 else
2253 log_warning_errno(errno, "Failed to send audit message: %m");
2254 }
2255 #endif
2256
2257 }
2258
2259 void manager_send_unit_plymouth(Manager *m, Unit *u) {
2260 static const union sockaddr_union sa = PLYMOUTH_SOCKET;
2261 _cleanup_free_ char *message = NULL;
2262 _cleanup_close_ int fd = -1;
2263 int n = 0;
2264
2265 /* Don't generate plymouth events if the service was already
2266 * started and we're just deserializing */
2267 if (MANAGER_IS_RELOADING(m))
2268 return;
2269
2270 if (!MANAGER_IS_SYSTEM(m))
2271 return;
2272
2273 if (detect_container() > 0)
2274 return;
2275
2276 if (u->type != UNIT_SERVICE &&
2277 u->type != UNIT_MOUNT &&
2278 u->type != UNIT_SWAP)
2279 return;
2280
2281 /* We set SOCK_NONBLOCK here so that we rather drop the
2282 * message then wait for plymouth */
2283 fd = socket(AF_UNIX, SOCK_STREAM|SOCK_CLOEXEC|SOCK_NONBLOCK, 0);
2284 if (fd < 0) {
2285 log_error_errno(errno, "socket() failed: %m");
2286 return;
2287 }
2288
2289 if (connect(fd, &sa.sa, SOCKADDR_UN_LEN(sa.un)) < 0) {
2290
2291 if (!IN_SET(errno, EPIPE, EAGAIN, ENOENT, ECONNREFUSED, ECONNRESET, ECONNABORTED))
2292 log_error_errno(errno, "connect() failed: %m");
2293 return;
2294 }
2295
2296 if (asprintf(&message, "U\002%c%s%n", (int) (strlen(u->id) + 1), u->id, &n) < 0) {
2297 log_oom();
2298 return;
2299 }
2300
2301 errno = 0;
2302 if (write(fd, message, n + 1) != n + 1)
2303 if (!IN_SET(errno, EPIPE, EAGAIN, ENOENT, ECONNREFUSED, ECONNRESET, ECONNABORTED))
2304 log_error_errno(errno, "Failed to write Plymouth message: %m");
2305 }
2306
2307 int manager_open_serialization(Manager *m, FILE **_f) {
2308 const char *path;
2309 int fd = -1;
2310 FILE *f;
2311
2312 assert(_f);
2313
2314 path = MANAGER_IS_SYSTEM(m) ? "/run/systemd" : "/tmp";
2315 fd = open_tmpfile_unlinkable(path, O_RDWR|O_CLOEXEC);
2316 if (fd < 0)
2317 return -errno;
2318
2319 log_debug("Serializing state to %s", path);
2320
2321 f = fdopen(fd, "w+");
2322 if (!f) {
2323 safe_close(fd);
2324 return -errno;
2325 }
2326
2327 *_f = f;
2328
2329 return 0;
2330 }
2331
2332 int manager_serialize(Manager *m, FILE *f, FDSet *fds, bool switching_root) {
2333 Iterator i;
2334 Unit *u;
2335 const char *t;
2336 char **e;
2337 int r;
2338
2339 assert(m);
2340 assert(f);
2341 assert(fds);
2342
2343 m->n_reloading++;
2344
2345 fprintf(f, "current-job-id=%"PRIu32"\n", m->current_job_id);
2346 fprintf(f, "taint-usr=%s\n", yes_no(m->taint_usr));
2347 fprintf(f, "n-installed-jobs=%u\n", m->n_installed_jobs);
2348 fprintf(f, "n-failed-jobs=%u\n", m->n_failed_jobs);
2349
2350 dual_timestamp_serialize(f, "firmware-timestamp", &m->firmware_timestamp);
2351 dual_timestamp_serialize(f, "loader-timestamp", &m->loader_timestamp);
2352 dual_timestamp_serialize(f, "kernel-timestamp", &m->kernel_timestamp);
2353 dual_timestamp_serialize(f, "initrd-timestamp", &m->initrd_timestamp);
2354
2355 if (!in_initrd()) {
2356 dual_timestamp_serialize(f, "userspace-timestamp", &m->userspace_timestamp);
2357 dual_timestamp_serialize(f, "finish-timestamp", &m->finish_timestamp);
2358 dual_timestamp_serialize(f, "security-start-timestamp", &m->security_start_timestamp);
2359 dual_timestamp_serialize(f, "security-finish-timestamp", &m->security_finish_timestamp);
2360 dual_timestamp_serialize(f, "generators-start-timestamp", &m->generators_start_timestamp);
2361 dual_timestamp_serialize(f, "generators-finish-timestamp", &m->generators_finish_timestamp);
2362 dual_timestamp_serialize(f, "units-load-start-timestamp", &m->units_load_start_timestamp);
2363 dual_timestamp_serialize(f, "units-load-finish-timestamp", &m->units_load_finish_timestamp);
2364 }
2365
2366 if (!switching_root) {
2367 STRV_FOREACH(e, m->environment) {
2368 _cleanup_free_ char *ce;
2369
2370 ce = cescape(*e);
2371 if (!ce)
2372 return -ENOMEM;
2373
2374 fprintf(f, "env=%s\n", *e);
2375 }
2376 }
2377
2378 if (m->notify_fd >= 0) {
2379 int copy;
2380
2381 copy = fdset_put_dup(fds, m->notify_fd);
2382 if (copy < 0)
2383 return copy;
2384
2385 fprintf(f, "notify-fd=%i\n", copy);
2386 fprintf(f, "notify-socket=%s\n", m->notify_socket);
2387 }
2388
2389 if (m->cgroups_agent_fd >= 0) {
2390 int copy;
2391
2392 copy = fdset_put_dup(fds, m->cgroups_agent_fd);
2393 if (copy < 0)
2394 return copy;
2395
2396 fprintf(f, "cgroups-agent-fd=%i\n", copy);
2397 }
2398
2399 if (m->kdbus_fd >= 0) {
2400 int copy;
2401
2402 copy = fdset_put_dup(fds, m->kdbus_fd);
2403 if (copy < 0)
2404 return copy;
2405
2406 fprintf(f, "kdbus-fd=%i\n", copy);
2407 }
2408
2409 bus_track_serialize(m->subscribed, f);
2410
2411 r = dynamic_user_serialize(m, f, fds);
2412 if (r < 0)
2413 return r;
2414
2415 fputc('\n', f);
2416
2417 HASHMAP_FOREACH_KEY(u, t, m->units, i) {
2418 if (u->id != t)
2419 continue;
2420
2421 /* Start marker */
2422 fputs(u->id, f);
2423 fputc('\n', f);
2424
2425 r = unit_serialize(u, f, fds, !switching_root);
2426 if (r < 0) {
2427 m->n_reloading--;
2428 return r;
2429 }
2430 }
2431
2432 assert(m->n_reloading > 0);
2433 m->n_reloading--;
2434
2435 if (ferror(f))
2436 return -EIO;
2437
2438 r = bus_fdset_add_all(m, fds);
2439 if (r < 0)
2440 return r;
2441
2442 return 0;
2443 }
2444
2445 int manager_deserialize(Manager *m, FILE *f, FDSet *fds) {
2446 int r = 0;
2447
2448 assert(m);
2449 assert(f);
2450
2451 log_debug("Deserializing state...");
2452
2453 m->n_reloading++;
2454
2455 for (;;) {
2456 char line[LINE_MAX], *l;
2457
2458 if (!fgets(line, sizeof(line), f)) {
2459 if (feof(f))
2460 r = 0;
2461 else
2462 r = -errno;
2463
2464 goto finish;
2465 }
2466
2467 char_array_0(line);
2468 l = strstrip(line);
2469
2470 if (l[0] == 0)
2471 break;
2472
2473 if (startswith(l, "current-job-id=")) {
2474 uint32_t id;
2475
2476 if (safe_atou32(l+15, &id) < 0)
2477 log_debug("Failed to parse current job id value %s", l+15);
2478 else
2479 m->current_job_id = MAX(m->current_job_id, id);
2480
2481 } else if (startswith(l, "n-installed-jobs=")) {
2482 uint32_t n;
2483
2484 if (safe_atou32(l+17, &n) < 0)
2485 log_debug("Failed to parse installed jobs counter %s", l+17);
2486 else
2487 m->n_installed_jobs += n;
2488
2489 } else if (startswith(l, "n-failed-jobs=")) {
2490 uint32_t n;
2491
2492 if (safe_atou32(l+14, &n) < 0)
2493 log_debug("Failed to parse failed jobs counter %s", l+14);
2494 else
2495 m->n_failed_jobs += n;
2496
2497 } else if (startswith(l, "taint-usr=")) {
2498 int b;
2499
2500 b = parse_boolean(l+10);
2501 if (b < 0)
2502 log_debug("Failed to parse taint /usr flag %s", l+10);
2503 else
2504 m->taint_usr = m->taint_usr || b;
2505
2506 } else if (startswith(l, "firmware-timestamp="))
2507 dual_timestamp_deserialize(l+19, &m->firmware_timestamp);
2508 else if (startswith(l, "loader-timestamp="))
2509 dual_timestamp_deserialize(l+17, &m->loader_timestamp);
2510 else if (startswith(l, "kernel-timestamp="))
2511 dual_timestamp_deserialize(l+17, &m->kernel_timestamp);
2512 else if (startswith(l, "initrd-timestamp="))
2513 dual_timestamp_deserialize(l+17, &m->initrd_timestamp);
2514 else if (startswith(l, "userspace-timestamp="))
2515 dual_timestamp_deserialize(l+20, &m->userspace_timestamp);
2516 else if (startswith(l, "finish-timestamp="))
2517 dual_timestamp_deserialize(l+17, &m->finish_timestamp);
2518 else if (startswith(l, "security-start-timestamp="))
2519 dual_timestamp_deserialize(l+25, &m->security_start_timestamp);
2520 else if (startswith(l, "security-finish-timestamp="))
2521 dual_timestamp_deserialize(l+26, &m->security_finish_timestamp);
2522 else if (startswith(l, "generators-start-timestamp="))
2523 dual_timestamp_deserialize(l+27, &m->generators_start_timestamp);
2524 else if (startswith(l, "generators-finish-timestamp="))
2525 dual_timestamp_deserialize(l+28, &m->generators_finish_timestamp);
2526 else if (startswith(l, "units-load-start-timestamp="))
2527 dual_timestamp_deserialize(l+27, &m->units_load_start_timestamp);
2528 else if (startswith(l, "units-load-finish-timestamp="))
2529 dual_timestamp_deserialize(l+28, &m->units_load_finish_timestamp);
2530 else if (startswith(l, "env=")) {
2531 _cleanup_free_ char *uce = NULL;
2532 char **e;
2533
2534 r = cunescape(l + 4, UNESCAPE_RELAX, &uce);
2535 if (r < 0)
2536 goto finish;
2537
2538 e = strv_env_set(m->environment, uce);
2539 if (!e) {
2540 r = -ENOMEM;
2541 goto finish;
2542 }
2543
2544 strv_free(m->environment);
2545 m->environment = e;
2546
2547 } else if (startswith(l, "notify-fd=")) {
2548 int fd;
2549
2550 if (safe_atoi(l + 10, &fd) < 0 || fd < 0 || !fdset_contains(fds, fd))
2551 log_debug("Failed to parse notify fd: %s", l + 10);
2552 else {
2553 m->notify_event_source = sd_event_source_unref(m->notify_event_source);
2554 safe_close(m->notify_fd);
2555 m->notify_fd = fdset_remove(fds, fd);
2556 }
2557
2558 } else if (startswith(l, "notify-socket=")) {
2559 char *n;
2560
2561 n = strdup(l+14);
2562 if (!n) {
2563 r = -ENOMEM;
2564 goto finish;
2565 }
2566
2567 free(m->notify_socket);
2568 m->notify_socket = n;
2569
2570 } else if (startswith(l, "cgroups-agent-fd=")) {
2571 int fd;
2572
2573 if (safe_atoi(l + 17, &fd) < 0 || fd < 0 || !fdset_contains(fds, fd))
2574 log_debug("Failed to parse cgroups agent fd: %s", l + 10);
2575 else {
2576 m->cgroups_agent_event_source = sd_event_source_unref(m->cgroups_agent_event_source);
2577 safe_close(m->cgroups_agent_fd);
2578 m->cgroups_agent_fd = fdset_remove(fds, fd);
2579 }
2580
2581 } else if (startswith(l, "kdbus-fd=")) {
2582 int fd;
2583
2584 if (safe_atoi(l + 9, &fd) < 0 || fd < 0 || !fdset_contains(fds, fd))
2585 log_debug("Failed to parse kdbus fd: %s", l + 9);
2586 else {
2587 safe_close(m->kdbus_fd);
2588 m->kdbus_fd = fdset_remove(fds, fd);
2589 }
2590
2591 } else if (startswith(l, "dynamic-user="))
2592 dynamic_user_deserialize_one(m, l + 13, fds);
2593 else {
2594 int k;
2595
2596 k = bus_track_deserialize_item(&m->deserialized_subscribed, l);
2597 if (k < 0)
2598 log_debug_errno(k, "Failed to deserialize bus tracker object: %m");
2599 else if (k == 0)
2600 log_debug("Unknown serialization item '%s'", l);
2601 }
2602 }
2603
2604 for (;;) {
2605 Unit *u;
2606 char name[UNIT_NAME_MAX+2];
2607
2608 /* Start marker */
2609 if (!fgets(name, sizeof(name), f)) {
2610 if (feof(f))
2611 r = 0;
2612 else
2613 r = -errno;
2614
2615 goto finish;
2616 }
2617
2618 char_array_0(name);
2619
2620 r = manager_load_unit(m, strstrip(name), NULL, NULL, &u);
2621 if (r < 0)
2622 goto finish;
2623
2624 r = unit_deserialize(u, f, fds);
2625 if (r < 0)
2626 goto finish;
2627 }
2628
2629 finish:
2630 if (ferror(f))
2631 r = -EIO;
2632
2633 assert(m->n_reloading > 0);
2634 m->n_reloading--;
2635
2636 return r;
2637 }
2638
2639 int manager_reload(Manager *m) {
2640 int r, q;
2641 _cleanup_fclose_ FILE *f = NULL;
2642 _cleanup_fdset_free_ FDSet *fds = NULL;
2643
2644 assert(m);
2645
2646 r = manager_open_serialization(m, &f);
2647 if (r < 0)
2648 return r;
2649
2650 m->n_reloading++;
2651 bus_manager_send_reloading(m, true);
2652
2653 fds = fdset_new();
2654 if (!fds) {
2655 m->n_reloading--;
2656 return -ENOMEM;
2657 }
2658
2659 r = manager_serialize(m, f, fds, false);
2660 if (r < 0) {
2661 m->n_reloading--;
2662 return r;
2663 }
2664
2665 if (fseeko(f, 0, SEEK_SET) < 0) {
2666 m->n_reloading--;
2667 return -errno;
2668 }
2669
2670 /* From here on there is no way back. */
2671 manager_clear_jobs_and_units(m);
2672 lookup_paths_flush_generator(&m->lookup_paths);
2673 lookup_paths_free(&m->lookup_paths);
2674 dynamic_user_vacuum(m, false);
2675
2676 q = lookup_paths_init(&m->lookup_paths, m->unit_file_scope, 0, NULL);
2677 if (q < 0 && r >= 0)
2678 r = q;
2679
2680 /* Find new unit paths */
2681 q = manager_run_generators(m);
2682 if (q < 0 && r >= 0)
2683 r = q;
2684
2685 lookup_paths_reduce(&m->lookup_paths);
2686 manager_build_unit_path_cache(m);
2687
2688 /* First, enumerate what we can from all config files */
2689 manager_enumerate(m);
2690
2691 /* Second, deserialize our stored data */
2692 q = manager_deserialize(m, f, fds);
2693 if (q < 0 && r >= 0)
2694 r = q;
2695
2696 fclose(f);
2697 f = NULL;
2698
2699 /* Re-register notify_fd as event source */
2700 q = manager_setup_notify(m);
2701 if (q < 0 && r >= 0)
2702 r = q;
2703
2704 q = manager_setup_cgroups_agent(m);
2705 if (q < 0 && r >= 0)
2706 r = q;
2707
2708 /* Third, fire things up! */
2709 manager_coldplug(m);
2710
2711 /* Release any dynamic users no longer referenced */
2712 dynamic_user_vacuum(m, true);
2713
2714 /* Sync current state of bus names with our set of listening units */
2715 if (m->api_bus)
2716 manager_sync_bus_names(m, m->api_bus);
2717
2718 assert(m->n_reloading > 0);
2719 m->n_reloading--;
2720
2721 m->send_reloading_done = true;
2722
2723 return r;
2724 }
2725
2726 void manager_reset_failed(Manager *m) {
2727 Unit *u;
2728 Iterator i;
2729
2730 assert(m);
2731
2732 HASHMAP_FOREACH(u, m->units, i)
2733 unit_reset_failed(u);
2734 }
2735
2736 bool manager_unit_inactive_or_pending(Manager *m, const char *name) {
2737 Unit *u;
2738
2739 assert(m);
2740 assert(name);
2741
2742 /* Returns true if the unit is inactive or going down */
2743 u = manager_get_unit(m, name);
2744 if (!u)
2745 return true;
2746
2747 return unit_inactive_or_pending(u);
2748 }
2749
2750 static void manager_notify_finished(Manager *m) {
2751 char userspace[FORMAT_TIMESPAN_MAX], initrd[FORMAT_TIMESPAN_MAX], kernel[FORMAT_TIMESPAN_MAX], sum[FORMAT_TIMESPAN_MAX];
2752 usec_t firmware_usec, loader_usec, kernel_usec, initrd_usec, userspace_usec, total_usec;
2753
2754 if (m->test_run)
2755 return;
2756
2757 if (MANAGER_IS_SYSTEM(m) && detect_container() <= 0) {
2758
2759 /* Note that m->kernel_usec.monotonic is always at 0,
2760 * and m->firmware_usec.monotonic and
2761 * m->loader_usec.monotonic should be considered
2762 * negative values. */
2763
2764 firmware_usec = m->firmware_timestamp.monotonic - m->loader_timestamp.monotonic;
2765 loader_usec = m->loader_timestamp.monotonic - m->kernel_timestamp.monotonic;
2766 userspace_usec = m->finish_timestamp.monotonic - m->userspace_timestamp.monotonic;
2767 total_usec = m->firmware_timestamp.monotonic + m->finish_timestamp.monotonic;
2768
2769 if (dual_timestamp_is_set(&m->initrd_timestamp)) {
2770
2771 kernel_usec = m->initrd_timestamp.monotonic - m->kernel_timestamp.monotonic;
2772 initrd_usec = m->userspace_timestamp.monotonic - m->initrd_timestamp.monotonic;
2773
2774 log_struct(LOG_INFO,
2775 LOG_MESSAGE_ID(SD_MESSAGE_STARTUP_FINISHED),
2776 "KERNEL_USEC="USEC_FMT, kernel_usec,
2777 "INITRD_USEC="USEC_FMT, initrd_usec,
2778 "USERSPACE_USEC="USEC_FMT, userspace_usec,
2779 LOG_MESSAGE("Startup finished in %s (kernel) + %s (initrd) + %s (userspace) = %s.",
2780 format_timespan(kernel, sizeof(kernel), kernel_usec, USEC_PER_MSEC),
2781 format_timespan(initrd, sizeof(initrd), initrd_usec, USEC_PER_MSEC),
2782 format_timespan(userspace, sizeof(userspace), userspace_usec, USEC_PER_MSEC),
2783 format_timespan(sum, sizeof(sum), total_usec, USEC_PER_MSEC)),
2784 NULL);
2785 } else {
2786 kernel_usec = m->userspace_timestamp.monotonic - m->kernel_timestamp.monotonic;
2787 initrd_usec = 0;
2788
2789 log_struct(LOG_INFO,
2790 LOG_MESSAGE_ID(SD_MESSAGE_STARTUP_FINISHED),
2791 "KERNEL_USEC="USEC_FMT, kernel_usec,
2792 "USERSPACE_USEC="USEC_FMT, userspace_usec,
2793 LOG_MESSAGE("Startup finished in %s (kernel) + %s (userspace) = %s.",
2794 format_timespan(kernel, sizeof(kernel), kernel_usec, USEC_PER_MSEC),
2795 format_timespan(userspace, sizeof(userspace), userspace_usec, USEC_PER_MSEC),
2796 format_timespan(sum, sizeof(sum), total_usec, USEC_PER_MSEC)),
2797 NULL);
2798 }
2799 } else {
2800 firmware_usec = loader_usec = initrd_usec = kernel_usec = 0;
2801 total_usec = userspace_usec = m->finish_timestamp.monotonic - m->userspace_timestamp.monotonic;
2802
2803 log_struct(LOG_INFO,
2804 LOG_MESSAGE_ID(SD_MESSAGE_STARTUP_FINISHED),
2805 "USERSPACE_USEC="USEC_FMT, userspace_usec,
2806 LOG_MESSAGE("Startup finished in %s.",
2807 format_timespan(sum, sizeof(sum), total_usec, USEC_PER_MSEC)),
2808 NULL);
2809 }
2810
2811 bus_manager_send_finished(m, firmware_usec, loader_usec, kernel_usec, initrd_usec, userspace_usec, total_usec);
2812
2813 sd_notifyf(false,
2814 "READY=1\n"
2815 "STATUS=Startup finished in %s.",
2816 format_timespan(sum, sizeof(sum), total_usec, USEC_PER_MSEC));
2817 }
2818
2819 void manager_check_finished(Manager *m) {
2820 assert(m);
2821
2822 if (MANAGER_IS_RELOADING(m))
2823 return;
2824
2825 /* Verify that we are actually running currently. Initially
2826 * the exit code is set to invalid, and during operation it is
2827 * then set to MANAGER_OK */
2828 if (m->exit_code != MANAGER_OK)
2829 return;
2830
2831 if (hashmap_size(m->jobs) > 0) {
2832 if (m->jobs_in_progress_event_source)
2833 /* Ignore any failure, this is only for feedback */
2834 (void) sd_event_source_set_time(m->jobs_in_progress_event_source, now(CLOCK_MONOTONIC) + JOBS_IN_PROGRESS_WAIT_USEC);
2835
2836 return;
2837 }
2838
2839 manager_flip_auto_status(m, false);
2840
2841 /* Notify Type=idle units that we are done now */
2842 manager_close_idle_pipe(m);
2843
2844 /* Turn off confirm spawn now */
2845 m->confirm_spawn = false;
2846
2847 /* No need to update ask password status when we're going non-interactive */
2848 manager_close_ask_password(m);
2849
2850 /* This is no longer the first boot */
2851 manager_set_first_boot(m, false);
2852
2853 if (dual_timestamp_is_set(&m->finish_timestamp))
2854 return;
2855
2856 dual_timestamp_get(&m->finish_timestamp);
2857
2858 manager_notify_finished(m);
2859
2860 manager_invalidate_startup_units(m);
2861 }
2862
2863 static int manager_run_generators(Manager *m) {
2864 _cleanup_strv_free_ char **paths = NULL;
2865 const char *argv[5];
2866 char **path;
2867 int r;
2868
2869 assert(m);
2870
2871 if (m->test_run)
2872 return 0;
2873
2874 paths = generator_binary_paths(m->unit_file_scope);
2875 if (!paths)
2876 return log_oom();
2877
2878 /* Optimize by skipping the whole process by not creating output directories
2879 * if no generators are found. */
2880 STRV_FOREACH(path, paths) {
2881 if (access(*path, F_OK) >= 0)
2882 goto found;
2883 if (errno != ENOENT)
2884 log_warning_errno(errno, "Failed to open generator directory %s: %m", *path);
2885 }
2886
2887 return 0;
2888
2889 found:
2890 r = lookup_paths_mkdir_generator(&m->lookup_paths);
2891 if (r < 0)
2892 goto finish;
2893
2894 argv[0] = NULL; /* Leave this empty, execute_directory() will fill something in */
2895 argv[1] = m->lookup_paths.generator;
2896 argv[2] = m->lookup_paths.generator_early;
2897 argv[3] = m->lookup_paths.generator_late;
2898 argv[4] = NULL;
2899
2900 RUN_WITH_UMASK(0022)
2901 execute_directories((const char* const*) paths, DEFAULT_TIMEOUT_USEC, (char**) argv);
2902
2903 finish:
2904 lookup_paths_trim_generator(&m->lookup_paths);
2905 return r;
2906 }
2907
2908 int manager_environment_add(Manager *m, char **minus, char **plus) {
2909 char **a = NULL, **b = NULL, **l;
2910 assert(m);
2911
2912 l = m->environment;
2913
2914 if (!strv_isempty(minus)) {
2915 a = strv_env_delete(l, 1, minus);
2916 if (!a)
2917 return -ENOMEM;
2918
2919 l = a;
2920 }
2921
2922 if (!strv_isempty(plus)) {
2923 b = strv_env_merge(2, l, plus);
2924 if (!b) {
2925 strv_free(a);
2926 return -ENOMEM;
2927 }
2928
2929 l = b;
2930 }
2931
2932 if (m->environment != l)
2933 strv_free(m->environment);
2934 if (a != l)
2935 strv_free(a);
2936 if (b != l)
2937 strv_free(b);
2938
2939 m->environment = l;
2940 manager_clean_environment(m);
2941 strv_sort(m->environment);
2942
2943 return 0;
2944 }
2945
2946 int manager_set_default_rlimits(Manager *m, struct rlimit **default_rlimit) {
2947 int i;
2948
2949 assert(m);
2950
2951 for (i = 0; i < _RLIMIT_MAX; i++) {
2952 m->rlimit[i] = mfree(m->rlimit[i]);
2953
2954 if (!default_rlimit[i])
2955 continue;
2956
2957 m->rlimit[i] = newdup(struct rlimit, default_rlimit[i], 1);
2958 if (!m->rlimit[i])
2959 return -ENOMEM;
2960 }
2961
2962 return 0;
2963 }
2964
2965 void manager_recheck_journal(Manager *m) {
2966 Unit *u;
2967
2968 assert(m);
2969
2970 if (!MANAGER_IS_SYSTEM(m))
2971 return;
2972
2973 u = manager_get_unit(m, SPECIAL_JOURNALD_SOCKET);
2974 if (u && SOCKET(u)->state != SOCKET_RUNNING) {
2975 log_close_journal();
2976 return;
2977 }
2978
2979 u = manager_get_unit(m, SPECIAL_JOURNALD_SERVICE);
2980 if (u && SERVICE(u)->state != SERVICE_RUNNING) {
2981 log_close_journal();
2982 return;
2983 }
2984
2985 /* Hmm, OK, so the socket is fully up and the service is up
2986 * too, then let's make use of the thing. */
2987 log_open();
2988 }
2989
2990 void manager_set_show_status(Manager *m, ShowStatus mode) {
2991 assert(m);
2992 assert(IN_SET(mode, SHOW_STATUS_AUTO, SHOW_STATUS_NO, SHOW_STATUS_YES, SHOW_STATUS_TEMPORARY));
2993
2994 if (!MANAGER_IS_SYSTEM(m))
2995 return;
2996
2997 if (m->show_status != mode)
2998 log_debug("%s showing of status.",
2999 mode == SHOW_STATUS_NO ? "Disabling" : "Enabling");
3000 m->show_status = mode;
3001
3002 if (mode > 0)
3003 (void) touch("/run/systemd/show-status");
3004 else
3005 (void) unlink("/run/systemd/show-status");
3006 }
3007
3008 static bool manager_get_show_status(Manager *m, StatusType type) {
3009 assert(m);
3010
3011 if (!MANAGER_IS_SYSTEM(m))
3012 return false;
3013
3014 if (m->no_console_output)
3015 return false;
3016
3017 if (!IN_SET(manager_state(m), MANAGER_INITIALIZING, MANAGER_STARTING, MANAGER_STOPPING))
3018 return false;
3019
3020 /* If we cannot find out the status properly, just proceed. */
3021 if (type != STATUS_TYPE_EMERGENCY && manager_check_ask_password(m) > 0)
3022 return false;
3023
3024 if (m->show_status > 0)
3025 return true;
3026
3027 return false;
3028 }
3029
3030 void manager_set_first_boot(Manager *m, bool b) {
3031 assert(m);
3032
3033 if (!MANAGER_IS_SYSTEM(m))
3034 return;
3035
3036 if (m->first_boot != (int) b) {
3037 if (b)
3038 (void) touch("/run/systemd/first-boot");
3039 else
3040 (void) unlink("/run/systemd/first-boot");
3041 }
3042
3043 m->first_boot = b;
3044 }
3045
3046 void manager_status_printf(Manager *m, StatusType type, const char *status, const char *format, ...) {
3047 va_list ap;
3048
3049 /* If m is NULL, assume we're after shutdown and let the messages through. */
3050
3051 if (m && !manager_get_show_status(m, type))
3052 return;
3053
3054 /* XXX We should totally drop the check for ephemeral here
3055 * and thus effectively make 'Type=idle' pointless. */
3056 if (type == STATUS_TYPE_EPHEMERAL && m && m->n_on_console > 0)
3057 return;
3058
3059 va_start(ap, format);
3060 status_vprintf(status, true, type == STATUS_TYPE_EPHEMERAL, format, ap);
3061 va_end(ap);
3062 }
3063
3064 Set *manager_get_units_requiring_mounts_for(Manager *m, const char *path) {
3065 char p[strlen(path)+1];
3066
3067 assert(m);
3068 assert(path);
3069
3070 strcpy(p, path);
3071 path_kill_slashes(p);
3072
3073 return hashmap_get(m->units_requiring_mounts_for, streq(p, "/") ? "" : p);
3074 }
3075
3076 const char *manager_get_runtime_prefix(Manager *m) {
3077 assert(m);
3078
3079 return MANAGER_IS_SYSTEM(m) ?
3080 "/run" :
3081 getenv("XDG_RUNTIME_DIR");
3082 }
3083
3084 int manager_update_failed_units(Manager *m, Unit *u, bool failed) {
3085 unsigned size;
3086 int r;
3087
3088 assert(m);
3089 assert(u->manager == m);
3090
3091 size = set_size(m->failed_units);
3092
3093 if (failed) {
3094 r = set_ensure_allocated(&m->failed_units, NULL);
3095 if (r < 0)
3096 return log_oom();
3097
3098 if (set_put(m->failed_units, u) < 0)
3099 return log_oom();
3100 } else
3101 (void) set_remove(m->failed_units, u);
3102
3103 if (set_size(m->failed_units) != size)
3104 bus_manager_send_change_signal(m);
3105
3106 return 0;
3107 }
3108
3109 ManagerState manager_state(Manager *m) {
3110 Unit *u;
3111
3112 assert(m);
3113
3114 /* Did we ever finish booting? If not then we are still starting up */
3115 if (!dual_timestamp_is_set(&m->finish_timestamp)) {
3116
3117 u = manager_get_unit(m, SPECIAL_BASIC_TARGET);
3118 if (!u || !UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u)))
3119 return MANAGER_INITIALIZING;
3120
3121 return MANAGER_STARTING;
3122 }
3123
3124 /* Is the special shutdown target queued? If so, we are in shutdown state */
3125 u = manager_get_unit(m, SPECIAL_SHUTDOWN_TARGET);
3126 if (u && u->job && IN_SET(u->job->type, JOB_START, JOB_RESTART, JOB_RELOAD_OR_START))
3127 return MANAGER_STOPPING;
3128
3129 /* Are the rescue or emergency targets active or queued? If so we are in maintenance state */
3130 u = manager_get_unit(m, SPECIAL_RESCUE_TARGET);
3131 if (u && (UNIT_IS_ACTIVE_OR_ACTIVATING(unit_active_state(u)) ||
3132 (u->job && IN_SET(u->job->type, JOB_START, JOB_RESTART, JOB_RELOAD_OR_START))))
3133 return MANAGER_MAINTENANCE;
3134
3135 u = manager_get_unit(m, SPECIAL_EMERGENCY_TARGET);
3136 if (u && (UNIT_IS_ACTIVE_OR_ACTIVATING(unit_active_state(u)) ||
3137 (u->job && IN_SET(u->job->type, JOB_START, JOB_RESTART, JOB_RELOAD_OR_START))))
3138 return MANAGER_MAINTENANCE;
3139
3140 /* Are there any failed units? If so, we are in degraded mode */
3141 if (set_size(m->failed_units) > 0)
3142 return MANAGER_DEGRADED;
3143
3144 return MANAGER_RUNNING;
3145 }
3146
3147 static const char *const manager_state_table[_MANAGER_STATE_MAX] = {
3148 [MANAGER_INITIALIZING] = "initializing",
3149 [MANAGER_STARTING] = "starting",
3150 [MANAGER_RUNNING] = "running",
3151 [MANAGER_DEGRADED] = "degraded",
3152 [MANAGER_MAINTENANCE] = "maintenance",
3153 [MANAGER_STOPPING] = "stopping",
3154 };
3155
3156 DEFINE_STRING_TABLE_LOOKUP(manager_state, ManagerState);