]> git.ipfire.org Git - thirdparty/systemd.git/blob - src/core/manager.c
Merge pull request #6431 from keszybz/hwdb-trivial-rows
[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 <errno.h>
21 #include <fcntl.h>
22 #include <linux/kd.h>
23 #include <signal.h>
24 #include <string.h>
25 #include <sys/epoll.h>
26 #include <sys/inotify.h>
27 #include <sys/ioctl.h>
28 #include <sys/reboot.h>
29 #include <sys/timerfd.h>
30 #include <sys/wait.h>
31 #include <unistd.h>
32
33 #ifdef HAVE_AUDIT
34 #include <libaudit.h>
35 #endif
36
37 #include "sd-daemon.h"
38 #include "sd-messages.h"
39 #include "sd-path.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 "clean-ipc.h"
49 #include "dbus-job.h"
50 #include "dbus-manager.h"
51 #include "dbus-unit.h"
52 #include "dbus.h"
53 #include "dirent-util.h"
54 #include "env-util.h"
55 #include "escape.h"
56 #include "execute.h"
57 #include "exec-util.h"
58 #include "exit-status.h"
59 #include "fd-util.h"
60 #include "fileio.h"
61 #include "fs-util.h"
62 #include "hashmap.h"
63 #include "io-util.h"
64 #include "locale-setup.h"
65 #include "log.h"
66 #include "macro.h"
67 #include "manager.h"
68 #include "missing.h"
69 #include "mkdir.h"
70 #include "parse-util.h"
71 #include "path-lookup.h"
72 #include "path-util.h"
73 #include "process-util.h"
74 #include "ratelimit.h"
75 #include "rm-rf.h"
76 #include "signal-util.h"
77 #include "special.h"
78 #include "stat-util.h"
79 #include "string-table.h"
80 #include "string-util.h"
81 #include "strv.h"
82 #include "terminal-util.h"
83 #include "time-util.h"
84 #include "transaction.h"
85 #include "umask-util.h"
86 #include "unit-name.h"
87 #include "user-util.h"
88 #include "util.h"
89 #include "virt.h"
90 #include "watchdog.h"
91
92 #define NOTIFY_RCVBUF_SIZE (8*1024*1024)
93 #define CGROUPS_AGENT_RCVBUF_SIZE (8*1024*1024)
94
95 /* Initial delay and the interval for printing status messages about running jobs */
96 #define JOBS_IN_PROGRESS_WAIT_USEC (5*USEC_PER_SEC)
97 #define JOBS_IN_PROGRESS_PERIOD_USEC (USEC_PER_SEC / 3)
98 #define JOBS_IN_PROGRESS_PERIOD_DIVISOR 3
99
100 static int manager_dispatch_notify_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata);
101 static int manager_dispatch_cgroups_agent_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata);
102 static int manager_dispatch_signal_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata);
103 static int manager_dispatch_time_change_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata);
104 static int manager_dispatch_idle_pipe_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata);
105 static int manager_dispatch_user_lookup_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata);
106 static int manager_dispatch_jobs_in_progress(sd_event_source *source, usec_t usec, void *userdata);
107 static int manager_dispatch_run_queue(sd_event_source *source, void *userdata);
108 static int manager_run_environment_generators(Manager *m);
109 static int manager_run_generators(Manager *m);
110
111 static void manager_watch_jobs_in_progress(Manager *m) {
112 usec_t next;
113 int r;
114
115 assert(m);
116
117 /* We do not want to show the cylon animation if the user
118 * needs to confirm service executions otherwise confirmation
119 * messages will be screwed by the cylon animation. */
120 if (!manager_is_confirm_spawn_disabled(m))
121 return;
122
123 if (m->jobs_in_progress_event_source)
124 return;
125
126 next = now(CLOCK_MONOTONIC) + JOBS_IN_PROGRESS_WAIT_USEC;
127 r = sd_event_add_time(
128 m->event,
129 &m->jobs_in_progress_event_source,
130 CLOCK_MONOTONIC,
131 next, 0,
132 manager_dispatch_jobs_in_progress, m);
133 if (r < 0)
134 return;
135
136 (void) sd_event_source_set_description(m->jobs_in_progress_event_source, "manager-jobs-in-progress");
137 }
138
139 #define CYLON_BUFFER_EXTRA (2*(sizeof(ANSI_RED)-1) + sizeof(ANSI_HIGHLIGHT_RED)-1 + 2*(sizeof(ANSI_NORMAL)-1))
140
141 static void draw_cylon(char buffer[], size_t buflen, unsigned width, unsigned pos) {
142 char *p = buffer;
143
144 assert(buflen >= CYLON_BUFFER_EXTRA + width + 1);
145 assert(pos <= width+1); /* 0 or width+1 mean that the center light is behind the corner */
146
147 if (pos > 1) {
148 if (pos > 2)
149 p = mempset(p, ' ', pos-2);
150 if (log_get_show_color())
151 p = stpcpy(p, ANSI_RED);
152 *p++ = '*';
153 }
154
155 if (pos > 0 && pos <= width) {
156 if (log_get_show_color())
157 p = stpcpy(p, ANSI_HIGHLIGHT_RED);
158 *p++ = '*';
159 }
160
161 if (log_get_show_color())
162 p = stpcpy(p, ANSI_NORMAL);
163
164 if (pos < width) {
165 if (log_get_show_color())
166 p = stpcpy(p, ANSI_RED);
167 *p++ = '*';
168 if (pos < width-1)
169 p = mempset(p, ' ', width-1-pos);
170 if (log_get_show_color())
171 strcpy(p, ANSI_NORMAL);
172 }
173 }
174
175 void manager_flip_auto_status(Manager *m, bool enable) {
176 assert(m);
177
178 if (enable) {
179 if (m->show_status == SHOW_STATUS_AUTO)
180 manager_set_show_status(m, SHOW_STATUS_TEMPORARY);
181 } else {
182 if (m->show_status == SHOW_STATUS_TEMPORARY)
183 manager_set_show_status(m, SHOW_STATUS_AUTO);
184 }
185 }
186
187 static void manager_print_jobs_in_progress(Manager *m) {
188 _cleanup_free_ char *job_of_n = NULL;
189 Iterator i;
190 Job *j;
191 unsigned counter = 0, print_nr;
192 char cylon[6 + CYLON_BUFFER_EXTRA + 1];
193 unsigned cylon_pos;
194 char time[FORMAT_TIMESPAN_MAX], limit[FORMAT_TIMESPAN_MAX] = "no limit";
195 uint64_t x;
196
197 assert(m);
198 assert(m->n_running_jobs > 0);
199
200 manager_flip_auto_status(m, true);
201
202 print_nr = (m->jobs_in_progress_iteration / JOBS_IN_PROGRESS_PERIOD_DIVISOR) % m->n_running_jobs;
203
204 HASHMAP_FOREACH(j, m->jobs, i)
205 if (j->state == JOB_RUNNING && counter++ == print_nr)
206 break;
207
208 /* m->n_running_jobs must be consistent with the contents of m->jobs,
209 * so the above loop must have succeeded in finding j. */
210 assert(counter == print_nr + 1);
211 assert(j);
212
213 cylon_pos = m->jobs_in_progress_iteration % 14;
214 if (cylon_pos >= 8)
215 cylon_pos = 14 - cylon_pos;
216 draw_cylon(cylon, sizeof(cylon), 6, cylon_pos);
217
218 m->jobs_in_progress_iteration++;
219
220 if (m->n_running_jobs > 1) {
221 if (asprintf(&job_of_n, "(%u of %u) ", counter, m->n_running_jobs) < 0)
222 job_of_n = NULL;
223 }
224
225 format_timespan(time, sizeof(time), now(CLOCK_MONOTONIC) - j->begin_usec, 1*USEC_PER_SEC);
226 if (job_get_timeout(j, &x) > 0)
227 format_timespan(limit, sizeof(limit), x - j->begin_usec, 1*USEC_PER_SEC);
228
229 manager_status_printf(m, STATUS_TYPE_EPHEMERAL, cylon,
230 "%sA %s job is running for %s (%s / %s)",
231 strempty(job_of_n),
232 job_type_to_string(j->type),
233 unit_description(j->unit),
234 time, limit);
235 }
236
237 static int have_ask_password(void) {
238 _cleanup_closedir_ DIR *dir;
239 struct dirent *de;
240
241 dir = opendir("/run/systemd/ask-password");
242 if (!dir) {
243 if (errno == ENOENT)
244 return false;
245 else
246 return -errno;
247 }
248
249 FOREACH_DIRENT_ALL(de, dir, return -errno) {
250 if (startswith(de->d_name, "ask."))
251 return true;
252 }
253 return false;
254 }
255
256 static int manager_dispatch_ask_password_fd(sd_event_source *source,
257 int fd, uint32_t revents, void *userdata) {
258 Manager *m = userdata;
259
260 assert(m);
261
262 flush_fd(fd);
263
264 m->have_ask_password = have_ask_password();
265 if (m->have_ask_password < 0)
266 /* Log error but continue. Negative have_ask_password
267 * is treated as unknown status. */
268 log_error_errno(m->have_ask_password, "Failed to list /run/systemd/ask-password: %m");
269
270 return 0;
271 }
272
273 static void manager_close_ask_password(Manager *m) {
274 assert(m);
275
276 m->ask_password_event_source = sd_event_source_unref(m->ask_password_event_source);
277 m->ask_password_inotify_fd = safe_close(m->ask_password_inotify_fd);
278 m->have_ask_password = -EINVAL;
279 }
280
281 static int manager_check_ask_password(Manager *m) {
282 int r;
283
284 assert(m);
285
286 if (!m->ask_password_event_source) {
287 assert(m->ask_password_inotify_fd < 0);
288
289 mkdir_p_label("/run/systemd/ask-password", 0755);
290
291 m->ask_password_inotify_fd = inotify_init1(IN_NONBLOCK|IN_CLOEXEC);
292 if (m->ask_password_inotify_fd < 0)
293 return log_error_errno(errno, "inotify_init1() failed: %m");
294
295 if (inotify_add_watch(m->ask_password_inotify_fd, "/run/systemd/ask-password", IN_CREATE|IN_DELETE|IN_MOVE) < 0) {
296 log_error_errno(errno, "Failed to add watch on /run/systemd/ask-password: %m");
297 manager_close_ask_password(m);
298 return -errno;
299 }
300
301 r = sd_event_add_io(m->event, &m->ask_password_event_source,
302 m->ask_password_inotify_fd, EPOLLIN,
303 manager_dispatch_ask_password_fd, m);
304 if (r < 0) {
305 log_error_errno(errno, "Failed to add event source for /run/systemd/ask-password: %m");
306 manager_close_ask_password(m);
307 return -errno;
308 }
309
310 (void) sd_event_source_set_description(m->ask_password_event_source, "manager-ask-password");
311
312 /* Queries might have been added meanwhile... */
313 manager_dispatch_ask_password_fd(m->ask_password_event_source,
314 m->ask_password_inotify_fd, EPOLLIN, m);
315 }
316
317 return m->have_ask_password;
318 }
319
320 static int manager_watch_idle_pipe(Manager *m) {
321 int r;
322
323 assert(m);
324
325 if (m->idle_pipe_event_source)
326 return 0;
327
328 if (m->idle_pipe[2] < 0)
329 return 0;
330
331 r = sd_event_add_io(m->event, &m->idle_pipe_event_source, m->idle_pipe[2], EPOLLIN, manager_dispatch_idle_pipe_fd, m);
332 if (r < 0)
333 return log_error_errno(r, "Failed to watch idle pipe: %m");
334
335 (void) sd_event_source_set_description(m->idle_pipe_event_source, "manager-idle-pipe");
336
337 return 0;
338 }
339
340 static void manager_close_idle_pipe(Manager *m) {
341 assert(m);
342
343 m->idle_pipe_event_source = sd_event_source_unref(m->idle_pipe_event_source);
344
345 safe_close_pair(m->idle_pipe);
346 safe_close_pair(m->idle_pipe + 2);
347 }
348
349 static int manager_setup_time_change(Manager *m) {
350 int r;
351
352 /* We only care for the cancellation event, hence we set the
353 * timeout to the latest possible value. */
354 struct itimerspec its = {
355 .it_value.tv_sec = TIME_T_MAX,
356 };
357
358 assert(m);
359 assert_cc(sizeof(time_t) == sizeof(TIME_T_MAX));
360
361 if (m->test_run)
362 return 0;
363
364 /* Uses TFD_TIMER_CANCEL_ON_SET to get notifications whenever
365 * CLOCK_REALTIME makes a jump relative to CLOCK_MONOTONIC */
366
367 m->time_change_fd = timerfd_create(CLOCK_REALTIME, TFD_NONBLOCK|TFD_CLOEXEC);
368 if (m->time_change_fd < 0)
369 return log_error_errno(errno, "Failed to create timerfd: %m");
370
371 if (timerfd_settime(m->time_change_fd, TFD_TIMER_ABSTIME|TFD_TIMER_CANCEL_ON_SET, &its, NULL) < 0) {
372 log_debug_errno(errno, "Failed to set up TFD_TIMER_CANCEL_ON_SET, ignoring: %m");
373 m->time_change_fd = safe_close(m->time_change_fd);
374 return 0;
375 }
376
377 r = sd_event_add_io(m->event, &m->time_change_event_source, m->time_change_fd, EPOLLIN, manager_dispatch_time_change_fd, m);
378 if (r < 0)
379 return log_error_errno(r, "Failed to create time change event source: %m");
380
381 (void) sd_event_source_set_description(m->time_change_event_source, "manager-time-change");
382
383 log_debug("Set up TFD_TIMER_CANCEL_ON_SET timerfd.");
384
385 return 0;
386 }
387
388 static int enable_special_signals(Manager *m) {
389 _cleanup_close_ int fd = -1;
390
391 assert(m);
392
393 if (m->test_run)
394 return 0;
395
396 /* Enable that we get SIGINT on control-alt-del. In containers
397 * this will fail with EPERM (older) or EINVAL (newer), so
398 * ignore that. */
399 if (reboot(RB_DISABLE_CAD) < 0 && errno != EPERM && errno != EINVAL)
400 log_warning_errno(errno, "Failed to enable ctrl-alt-del handling: %m");
401
402 fd = open_terminal("/dev/tty0", O_RDWR|O_NOCTTY|O_CLOEXEC);
403 if (fd < 0) {
404 /* Support systems without virtual console */
405 if (fd != -ENOENT)
406 log_warning_errno(errno, "Failed to open /dev/tty0: %m");
407 } else {
408 /* Enable that we get SIGWINCH on kbrequest */
409 if (ioctl(fd, KDSIGACCEPT, SIGWINCH) < 0)
410 log_warning_errno(errno, "Failed to enable kbrequest handling: %m");
411 }
412
413 return 0;
414 }
415
416 static int manager_setup_signals(Manager *m) {
417 struct sigaction sa = {
418 .sa_handler = SIG_DFL,
419 .sa_flags = SA_NOCLDSTOP|SA_RESTART,
420 };
421 sigset_t mask;
422 int r;
423
424 assert(m);
425
426 assert_se(sigaction(SIGCHLD, &sa, NULL) == 0);
427
428 /* We make liberal use of realtime signals here. On
429 * Linux/glibc we have 30 of them (with the exception of Linux
430 * on hppa, see below), between SIGRTMIN+0 ... SIGRTMIN+30
431 * (aka SIGRTMAX). */
432
433 assert_se(sigemptyset(&mask) == 0);
434 sigset_add_many(&mask,
435 SIGCHLD, /* Child died */
436 SIGTERM, /* Reexecute daemon */
437 SIGHUP, /* Reload configuration */
438 SIGUSR1, /* systemd/upstart: reconnect to D-Bus */
439 SIGUSR2, /* systemd: dump status */
440 SIGINT, /* Kernel sends us this on control-alt-del */
441 SIGWINCH, /* Kernel sends us this on kbrequest (alt-arrowup) */
442 SIGPWR, /* Some kernel drivers and upsd send us this on power failure */
443
444 SIGRTMIN+0, /* systemd: start default.target */
445 SIGRTMIN+1, /* systemd: isolate rescue.target */
446 SIGRTMIN+2, /* systemd: isolate emergency.target */
447 SIGRTMIN+3, /* systemd: start halt.target */
448 SIGRTMIN+4, /* systemd: start poweroff.target */
449 SIGRTMIN+5, /* systemd: start reboot.target */
450 SIGRTMIN+6, /* systemd: start kexec.target */
451
452 /* ... space for more special targets ... */
453
454 SIGRTMIN+13, /* systemd: Immediate halt */
455 SIGRTMIN+14, /* systemd: Immediate poweroff */
456 SIGRTMIN+15, /* systemd: Immediate reboot */
457 SIGRTMIN+16, /* systemd: Immediate kexec */
458
459 /* ... space for more immediate system state changes ... */
460
461 SIGRTMIN+20, /* systemd: enable status messages */
462 SIGRTMIN+21, /* systemd: disable status messages */
463 SIGRTMIN+22, /* systemd: set log level to LOG_DEBUG */
464 SIGRTMIN+23, /* systemd: set log level to LOG_INFO */
465 SIGRTMIN+24, /* systemd: Immediate exit (--user only) */
466
467 /* .. one free signal here ... */
468
469 #if !defined(__hppa64__) && !defined(__hppa__)
470 /* Apparently Linux on hppa has fewer RT
471 * signals (SIGRTMAX is SIGRTMIN+25 there),
472 * hence let's not try to make use of them
473 * here. Since these commands are accessible
474 * by different means and only really a safety
475 * net, the missing functionality on hppa
476 * shouldn't matter. */
477
478 SIGRTMIN+26, /* systemd: set log target to journal-or-kmsg */
479 SIGRTMIN+27, /* systemd: set log target to console */
480 SIGRTMIN+28, /* systemd: set log target to kmsg */
481 SIGRTMIN+29, /* systemd: set log target to syslog-or-kmsg (obsolete) */
482
483 /* ... one free signal here SIGRTMIN+30 ... */
484 #endif
485 -1);
486 assert_se(sigprocmask(SIG_SETMASK, &mask, NULL) == 0);
487
488 m->signal_fd = signalfd(-1, &mask, SFD_NONBLOCK|SFD_CLOEXEC);
489 if (m->signal_fd < 0)
490 return -errno;
491
492 r = sd_event_add_io(m->event, &m->signal_event_source, m->signal_fd, EPOLLIN, manager_dispatch_signal_fd, m);
493 if (r < 0)
494 return r;
495
496 (void) sd_event_source_set_description(m->signal_event_source, "manager-signal");
497
498 /* Process signals a bit earlier than the rest of things, but later than notify_fd processing, so that the
499 * notify processing can still figure out to which process/service a message belongs, before we reap the
500 * process. Also, process this before handling cgroup notifications, so that we always collect child exit
501 * status information before detecting that there's no process in a cgroup. */
502 r = sd_event_source_set_priority(m->signal_event_source, SD_EVENT_PRIORITY_NORMAL-6);
503 if (r < 0)
504 return r;
505
506 if (MANAGER_IS_SYSTEM(m))
507 return enable_special_signals(m);
508
509 return 0;
510 }
511
512 static void manager_clean_environment(Manager *m) {
513 assert(m);
514
515 /* Let's remove some environment variables that we
516 * need ourselves to communicate with our clients */
517 strv_env_unset_many(
518 m->environment,
519 "NOTIFY_SOCKET",
520 "MAINPID",
521 "MANAGERPID",
522 "LISTEN_PID",
523 "LISTEN_FDS",
524 "LISTEN_FDNAMES",
525 "WATCHDOG_PID",
526 "WATCHDOG_USEC",
527 "INVOCATION_ID",
528 NULL);
529 }
530
531 static int manager_default_environment(Manager *m) {
532 assert(m);
533
534 if (MANAGER_IS_SYSTEM(m)) {
535 /* The system manager always starts with a clean
536 * environment for its children. It does not import
537 * the kernel's or the parents' exported variables.
538 *
539 * The initial passed environment is untouched to keep
540 * /proc/self/environ valid; it is used for tagging
541 * the init process inside containers. */
542 m->environment = strv_new("PATH=" DEFAULT_PATH,
543 NULL);
544
545 /* Import locale variables LC_*= from configuration */
546 locale_setup(&m->environment);
547 } else
548 /* The user manager passes its own environment
549 * along to its children. */
550 m->environment = strv_copy(environ);
551
552 if (!m->environment)
553 return -ENOMEM;
554
555 manager_clean_environment(m);
556 strv_sort(m->environment);
557
558 return 0;
559 }
560
561 static int manager_setup_prefix(Manager *m) {
562 struct table_entry {
563 uint64_t type;
564 const char *suffix;
565 };
566
567 static const struct table_entry paths_system[_EXEC_DIRECTORY_MAX] = {
568 [EXEC_DIRECTORY_RUNTIME] = { SD_PATH_SYSTEM_RUNTIME, NULL },
569 [EXEC_DIRECTORY_STATE] = { SD_PATH_SYSTEM_STATE_PRIVATE, NULL },
570 [EXEC_DIRECTORY_CACHE] = { SD_PATH_SYSTEM_STATE_CACHE, NULL },
571 [EXEC_DIRECTORY_LOGS] = { SD_PATH_SYSTEM_STATE_LOGS, NULL },
572 [EXEC_DIRECTORY_CONFIGURATION] = { SD_PATH_SYSTEM_CONFIGURATION, NULL },
573 };
574
575 static const struct table_entry paths_user[_EXEC_DIRECTORY_MAX] = {
576 [EXEC_DIRECTORY_RUNTIME] = { SD_PATH_USER_RUNTIME, NULL },
577 [EXEC_DIRECTORY_STATE] = { SD_PATH_USER_CONFIGURATION, NULL },
578 [EXEC_DIRECTORY_CACHE] = { SD_PATH_SYSTEM_STATE_CACHE, NULL },
579 [EXEC_DIRECTORY_LOGS] = { SD_PATH_SYSTEM_CONFIGURATION, "log" },
580 [EXEC_DIRECTORY_CONFIGURATION] = { SD_PATH_SYSTEM_CONFIGURATION, NULL },
581 };
582
583 const struct table_entry *p;
584 ExecDirectoryType i;
585 int r;
586
587 assert(m);
588
589 if (MANAGER_IS_SYSTEM(m))
590 p = paths_system;
591 else
592 p = paths_user;
593
594 for (i = 0; i < _EXEC_DIRECTORY_MAX; i++) {
595 r = sd_path_home(p[i].type, p[i].suffix, &m->prefix[i]);
596 if (r < 0)
597 return r;
598 }
599
600 return 0;
601 }
602
603 int manager_new(UnitFileScope scope, bool test_run, Manager **_m) {
604 Manager *m;
605 int r;
606
607 assert(_m);
608 assert(IN_SET(scope, UNIT_FILE_SYSTEM, UNIT_FILE_USER));
609
610 m = new0(Manager, 1);
611 if (!m)
612 return -ENOMEM;
613
614 m->unit_file_scope = scope;
615 m->exit_code = _MANAGER_EXIT_CODE_INVALID;
616 m->default_timer_accuracy_usec = USEC_PER_MINUTE;
617 m->default_tasks_accounting = true;
618 m->default_tasks_max = UINT64_MAX;
619
620 #ifdef ENABLE_EFI
621 if (MANAGER_IS_SYSTEM(m) && detect_container() <= 0)
622 boot_timestamps(&m->userspace_timestamp, &m->firmware_timestamp, &m->loader_timestamp);
623 #endif
624
625 /* Prepare log fields we can use for structured logging */
626 if (MANAGER_IS_SYSTEM(m)) {
627 m->unit_log_field = "UNIT=";
628 m->unit_log_format_string = "UNIT=%s";
629
630 m->invocation_log_field = "INVOCATION_ID=";
631 m->invocation_log_format_string = "INVOCATION_ID=" SD_ID128_FORMAT_STR;
632 } else {
633 m->unit_log_field = "USER_UNIT=";
634 m->unit_log_format_string = "USER_UNIT=%s";
635
636 m->invocation_log_field = "USER_INVOCATION_ID=";
637 m->invocation_log_format_string = "USER_INVOCATION_ID=" SD_ID128_FORMAT_STR;
638 }
639
640 m->idle_pipe[0] = m->idle_pipe[1] = m->idle_pipe[2] = m->idle_pipe[3] = -1;
641
642 m->pin_cgroupfs_fd = m->notify_fd = m->cgroups_agent_fd = m->signal_fd = m->time_change_fd =
643 m->dev_autofs_fd = m->private_listen_fd = m->cgroup_inotify_fd =
644 m->ask_password_inotify_fd = -1;
645
646 m->user_lookup_fds[0] = m->user_lookup_fds[1] = -1;
647
648 m->current_job_id = 1; /* start as id #1, so that we can leave #0 around as "null-like" value */
649
650 m->have_ask_password = -EINVAL; /* we don't know */
651 m->first_boot = -1;
652
653 m->test_run = test_run;
654
655 /* Reboot immediately if the user hits C-A-D more often than 7x per 2s */
656 RATELIMIT_INIT(m->ctrl_alt_del_ratelimit, 2 * USEC_PER_SEC, 7);
657
658 r = manager_default_environment(m);
659 if (r < 0)
660 goto fail;
661
662 r = hashmap_ensure_allocated(&m->units, &string_hash_ops);
663 if (r < 0)
664 goto fail;
665
666 r = hashmap_ensure_allocated(&m->jobs, NULL);
667 if (r < 0)
668 goto fail;
669
670 r = hashmap_ensure_allocated(&m->cgroup_unit, &string_hash_ops);
671 if (r < 0)
672 goto fail;
673
674 r = hashmap_ensure_allocated(&m->watch_bus, &string_hash_ops);
675 if (r < 0)
676 goto fail;
677
678 r = sd_event_default(&m->event);
679 if (r < 0)
680 goto fail;
681
682 r = sd_event_add_defer(m->event, &m->run_queue_event_source, manager_dispatch_run_queue, m);
683 if (r < 0)
684 goto fail;
685
686 r = sd_event_source_set_priority(m->run_queue_event_source, SD_EVENT_PRIORITY_IDLE);
687 if (r < 0)
688 goto fail;
689
690 r = sd_event_source_set_enabled(m->run_queue_event_source, SD_EVENT_OFF);
691 if (r < 0)
692 goto fail;
693
694 (void) sd_event_source_set_description(m->run_queue_event_source, "manager-run-queue");
695
696 r = manager_setup_signals(m);
697 if (r < 0)
698 goto fail;
699
700 r = manager_setup_cgroup(m);
701 if (r < 0)
702 goto fail;
703
704 r = manager_setup_time_change(m);
705 if (r < 0)
706 goto fail;
707
708 m->udev = udev_new();
709 if (!m->udev) {
710 r = -ENOMEM;
711 goto fail;
712 }
713
714 /* Note that we do not set up the notify fd here. We do that after deserialization,
715 * since they might have gotten serialized across the reexec. */
716
717 m->taint_usr = dir_is_empty("/usr") > 0;
718
719 r = manager_setup_prefix(m);
720 if (r < 0)
721 goto fail;
722
723 *_m = m;
724 return 0;
725
726 fail:
727 manager_free(m);
728 return r;
729 }
730
731 static int manager_setup_notify(Manager *m) {
732 int r;
733
734 if (m->test_run)
735 return 0;
736
737 if (m->notify_fd < 0) {
738 _cleanup_close_ int fd = -1;
739 union sockaddr_union sa = {
740 .sa.sa_family = AF_UNIX,
741 };
742 static const int one = 1;
743
744 /* First free all secondary fields */
745 m->notify_socket = mfree(m->notify_socket);
746 m->notify_event_source = sd_event_source_unref(m->notify_event_source);
747
748 fd = socket(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC|SOCK_NONBLOCK, 0);
749 if (fd < 0)
750 return log_error_errno(errno, "Failed to allocate notification socket: %m");
751
752 fd_inc_rcvbuf(fd, NOTIFY_RCVBUF_SIZE);
753
754 m->notify_socket = strappend(m->prefix[EXEC_DIRECTORY_RUNTIME], "/systemd/notify");
755 if (!m->notify_socket)
756 return log_oom();
757
758 (void) mkdir_parents_label(m->notify_socket, 0755);
759 (void) unlink(m->notify_socket);
760
761 strncpy(sa.un.sun_path, m->notify_socket, sizeof(sa.un.sun_path)-1);
762 r = bind(fd, &sa.sa, SOCKADDR_UN_LEN(sa.un));
763 if (r < 0)
764 return log_error_errno(errno, "bind(%s) failed: %m", sa.un.sun_path);
765
766 r = setsockopt(fd, SOL_SOCKET, SO_PASSCRED, &one, sizeof(one));
767 if (r < 0)
768 return log_error_errno(errno, "SO_PASSCRED failed: %m");
769
770 m->notify_fd = fd;
771 fd = -1;
772
773 log_debug("Using notification socket %s", m->notify_socket);
774 }
775
776 if (!m->notify_event_source) {
777 r = sd_event_add_io(m->event, &m->notify_event_source, m->notify_fd, EPOLLIN, manager_dispatch_notify_fd, m);
778 if (r < 0)
779 return log_error_errno(r, "Failed to allocate notify event source: %m");
780
781 /* Process notification messages a bit earlier than SIGCHLD, so that we can still identify to which
782 * service an exit message belongs. */
783 r = sd_event_source_set_priority(m->notify_event_source, SD_EVENT_PRIORITY_NORMAL-7);
784 if (r < 0)
785 return log_error_errno(r, "Failed to set priority of notify event source: %m");
786
787 (void) sd_event_source_set_description(m->notify_event_source, "manager-notify");
788 }
789
790 return 0;
791 }
792
793 static int manager_setup_cgroups_agent(Manager *m) {
794
795 static const union sockaddr_union sa = {
796 .un.sun_family = AF_UNIX,
797 .un.sun_path = "/run/systemd/cgroups-agent",
798 };
799 int r;
800
801 /* This creates a listening socket we receive cgroups agent messages on. We do not use D-Bus for delivering
802 * these messages from the cgroups agent binary to PID 1, as the cgroups agent binary is very short-living, and
803 * each instance of it needs a new D-Bus connection. Since D-Bus connections are SOCK_STREAM/AF_UNIX, on
804 * overloaded systems the backlog of the D-Bus socket becomes relevant, as not more than the configured number
805 * of D-Bus connections may be queued until the kernel will start dropping further incoming connections,
806 * possibly resulting in lost cgroups agent messages. To avoid this, we'll use a private SOCK_DGRAM/AF_UNIX
807 * socket, where no backlog is relevant as communication may take place without an actual connect() cycle, and
808 * we thus won't lose messages.
809 *
810 * Note that PID 1 will forward the agent message to system bus, so that the user systemd instance may listen
811 * to it. The system instance hence listens on this special socket, but the user instances listen on the system
812 * bus for these messages. */
813
814 if (m->test_run)
815 return 0;
816
817 if (!MANAGER_IS_SYSTEM(m))
818 return 0;
819
820 r = cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER);
821 if (r < 0)
822 return log_error_errno(r, "Failed to determine whether unified cgroups hierarchy is used: %m");
823 if (r > 0) /* We don't need this anymore on the unified hierarchy */
824 return 0;
825
826 if (m->cgroups_agent_fd < 0) {
827 _cleanup_close_ int fd = -1;
828
829 /* First free all secondary fields */
830 m->cgroups_agent_event_source = sd_event_source_unref(m->cgroups_agent_event_source);
831
832 fd = socket(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC|SOCK_NONBLOCK, 0);
833 if (fd < 0)
834 return log_error_errno(errno, "Failed to allocate cgroups agent socket: %m");
835
836 fd_inc_rcvbuf(fd, CGROUPS_AGENT_RCVBUF_SIZE);
837
838 (void) unlink(sa.un.sun_path);
839
840 /* Only allow root to connect to this socket */
841 RUN_WITH_UMASK(0077)
842 r = bind(fd, &sa.sa, SOCKADDR_UN_LEN(sa.un));
843 if (r < 0)
844 return log_error_errno(errno, "bind(%s) failed: %m", sa.un.sun_path);
845
846 m->cgroups_agent_fd = fd;
847 fd = -1;
848 }
849
850 if (!m->cgroups_agent_event_source) {
851 r = sd_event_add_io(m->event, &m->cgroups_agent_event_source, m->cgroups_agent_fd, EPOLLIN, manager_dispatch_cgroups_agent_fd, m);
852 if (r < 0)
853 return log_error_errno(r, "Failed to allocate cgroups agent event source: %m");
854
855 /* Process cgroups notifications early, but after having processed service notification messages or
856 * SIGCHLD signals, so that a cgroup running empty is always just the last safety net of notification,
857 * and we collected the metadata the notification and SIGCHLD stuff offers first. Also see handling of
858 * cgroup inotify for the unified cgroup stuff. */
859 r = sd_event_source_set_priority(m->cgroups_agent_event_source, SD_EVENT_PRIORITY_NORMAL-5);
860 if (r < 0)
861 return log_error_errno(r, "Failed to set priority of cgroups agent event source: %m");
862
863 (void) sd_event_source_set_description(m->cgroups_agent_event_source, "manager-cgroups-agent");
864 }
865
866 return 0;
867 }
868
869 static int manager_setup_user_lookup_fd(Manager *m) {
870 int r;
871
872 assert(m);
873
874 /* Set up the socket pair used for passing UID/GID resolution results from forked off processes to PID
875 * 1. Background: we can't do name lookups (NSS) from PID 1, since it might involve IPC and thus activation,
876 * and we might hence deadlock on ourselves. Hence we do all user/group lookups asynchronously from the forked
877 * off processes right before executing the binaries to start. In order to be able to clean up any IPC objects
878 * created by a unit (see RemoveIPC=) we need to know in PID 1 the used UID/GID of the executed processes,
879 * hence we establish this communication channel so that forked off processes can pass their UID/GID
880 * information back to PID 1. The forked off processes send their resolved UID/GID to PID 1 in a simple
881 * datagram, along with their unit name, so that we can share one communication socket pair among all units for
882 * this purpose.
883 *
884 * You might wonder why we need a communication channel for this that is independent of the usual notification
885 * socket scheme (i.e. $NOTIFY_SOCKET). The primary difference is about trust: data sent via the $NOTIFY_SOCKET
886 * channel is only accepted if it originates from the right unit and if reception was enabled for it. The user
887 * lookup socket OTOH is only accessible by PID 1 and its children until they exec(), and always available.
888 *
889 * Note that this function is called under two circumstances: when we first initialize (in which case we
890 * allocate both the socket pair and the event source to listen on it), and when we deserialize after a reload
891 * (in which case the socket pair already exists but we still need to allocate the event source for it). */
892
893 if (m->user_lookup_fds[0] < 0) {
894
895 /* Free all secondary fields */
896 safe_close_pair(m->user_lookup_fds);
897 m->user_lookup_event_source = sd_event_source_unref(m->user_lookup_event_source);
898
899 if (socketpair(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC, 0, m->user_lookup_fds) < 0)
900 return log_error_errno(errno, "Failed to allocate user lookup socket: %m");
901
902 (void) fd_inc_rcvbuf(m->user_lookup_fds[0], NOTIFY_RCVBUF_SIZE);
903 }
904
905 if (!m->user_lookup_event_source) {
906 r = sd_event_add_io(m->event, &m->user_lookup_event_source, m->user_lookup_fds[0], EPOLLIN, manager_dispatch_user_lookup_fd, m);
907 if (r < 0)
908 return log_error_errno(errno, "Failed to allocate user lookup event source: %m");
909
910 /* Process even earlier than the notify event source, so that we always know first about valid UID/GID
911 * resolutions */
912 r = sd_event_source_set_priority(m->user_lookup_event_source, SD_EVENT_PRIORITY_NORMAL-8);
913 if (r < 0)
914 return log_error_errno(errno, "Failed to set priority ot user lookup event source: %m");
915
916 (void) sd_event_source_set_description(m->user_lookup_event_source, "user-lookup");
917 }
918
919 return 0;
920 }
921
922 static int manager_connect_bus(Manager *m, bool reexecuting) {
923 bool try_bus_connect;
924
925 assert(m);
926
927 if (m->test_run)
928 return 0;
929
930 try_bus_connect =
931 reexecuting ||
932 (MANAGER_IS_USER(m) && getenv("DBUS_SESSION_BUS_ADDRESS"));
933
934 /* Try to connect to the buses, if possible. */
935 return bus_init(m, try_bus_connect);
936 }
937
938 static unsigned manager_dispatch_cleanup_queue(Manager *m) {
939 Unit *u;
940 unsigned n = 0;
941
942 assert(m);
943
944 while ((u = m->cleanup_queue)) {
945 assert(u->in_cleanup_queue);
946
947 unit_free(u);
948 n++;
949 }
950
951 return n;
952 }
953
954 enum {
955 GC_OFFSET_IN_PATH, /* This one is on the path we were traveling */
956 GC_OFFSET_UNSURE, /* No clue */
957 GC_OFFSET_GOOD, /* We still need this unit */
958 GC_OFFSET_BAD, /* We don't need this unit anymore */
959 _GC_OFFSET_MAX
960 };
961
962 static void unit_gc_mark_good(Unit *u, unsigned gc_marker) {
963 Iterator i;
964 Unit *other;
965
966 u->gc_marker = gc_marker + GC_OFFSET_GOOD;
967
968 /* Recursively mark referenced units as GOOD as well */
969 SET_FOREACH(other, u->dependencies[UNIT_REFERENCES], i)
970 if (other->gc_marker == gc_marker + GC_OFFSET_UNSURE)
971 unit_gc_mark_good(other, gc_marker);
972 }
973
974 static void unit_gc_sweep(Unit *u, unsigned gc_marker) {
975 Iterator i;
976 Unit *other;
977 bool is_bad;
978
979 assert(u);
980
981 if (u->gc_marker == gc_marker + GC_OFFSET_GOOD ||
982 u->gc_marker == gc_marker + GC_OFFSET_BAD ||
983 u->gc_marker == gc_marker + GC_OFFSET_UNSURE ||
984 u->gc_marker == gc_marker + GC_OFFSET_IN_PATH)
985 return;
986
987 if (u->in_cleanup_queue)
988 goto bad;
989
990 if (unit_check_gc(u))
991 goto good;
992
993 u->gc_marker = gc_marker + GC_OFFSET_IN_PATH;
994
995 is_bad = true;
996
997 SET_FOREACH(other, u->dependencies[UNIT_REFERENCED_BY], i) {
998 unit_gc_sweep(other, gc_marker);
999
1000 if (other->gc_marker == gc_marker + GC_OFFSET_GOOD)
1001 goto good;
1002
1003 if (other->gc_marker != gc_marker + GC_OFFSET_BAD)
1004 is_bad = false;
1005 }
1006
1007 if (is_bad)
1008 goto bad;
1009
1010 /* We were unable to find anything out about this entry, so
1011 * let's investigate it later */
1012 u->gc_marker = gc_marker + GC_OFFSET_UNSURE;
1013 unit_add_to_gc_queue(u);
1014 return;
1015
1016 bad:
1017 /* We definitely know that this one is not useful anymore, so
1018 * let's mark it for deletion */
1019 u->gc_marker = gc_marker + GC_OFFSET_BAD;
1020 unit_add_to_cleanup_queue(u);
1021 return;
1022
1023 good:
1024 unit_gc_mark_good(u, gc_marker);
1025 }
1026
1027 static unsigned manager_dispatch_gc_unit_queue(Manager *m) {
1028 unsigned n = 0, gc_marker;
1029 Unit *u;
1030
1031 assert(m);
1032
1033 /* log_debug("Running GC..."); */
1034
1035 m->gc_marker += _GC_OFFSET_MAX;
1036 if (m->gc_marker + _GC_OFFSET_MAX <= _GC_OFFSET_MAX)
1037 m->gc_marker = 1;
1038
1039 gc_marker = m->gc_marker;
1040
1041 while ((u = m->gc_unit_queue)) {
1042 assert(u->in_gc_queue);
1043
1044 unit_gc_sweep(u, gc_marker);
1045
1046 LIST_REMOVE(gc_queue, m->gc_unit_queue, u);
1047 u->in_gc_queue = false;
1048
1049 n++;
1050
1051 if (u->gc_marker == gc_marker + GC_OFFSET_BAD ||
1052 u->gc_marker == gc_marker + GC_OFFSET_UNSURE) {
1053 if (u->id)
1054 log_unit_debug(u, "Collecting.");
1055 u->gc_marker = gc_marker + GC_OFFSET_BAD;
1056 unit_add_to_cleanup_queue(u);
1057 }
1058 }
1059
1060 return n;
1061 }
1062
1063 static unsigned manager_dispatch_gc_job_queue(Manager *m) {
1064 unsigned n = 0;
1065 Job *j;
1066
1067 assert(m);
1068
1069 while ((j = m->gc_job_queue)) {
1070 assert(j->in_gc_queue);
1071
1072 LIST_REMOVE(gc_queue, m->gc_job_queue, j);
1073 j->in_gc_queue = false;
1074
1075 n++;
1076
1077 if (job_check_gc(j))
1078 continue;
1079
1080 log_unit_debug(j->unit, "Collecting job.");
1081 (void) job_finish_and_invalidate(j, JOB_COLLECTED, false, false);
1082 }
1083
1084 return n;
1085 }
1086
1087 static void manager_clear_jobs_and_units(Manager *m) {
1088 Unit *u;
1089
1090 assert(m);
1091
1092 while ((u = hashmap_first(m->units)))
1093 unit_free(u);
1094
1095 manager_dispatch_cleanup_queue(m);
1096
1097 assert(!m->load_queue);
1098 assert(!m->run_queue);
1099 assert(!m->dbus_unit_queue);
1100 assert(!m->dbus_job_queue);
1101 assert(!m->cleanup_queue);
1102 assert(!m->gc_unit_queue);
1103 assert(!m->gc_job_queue);
1104
1105 assert(hashmap_isempty(m->jobs));
1106 assert(hashmap_isempty(m->units));
1107
1108 m->n_on_console = 0;
1109 m->n_running_jobs = 0;
1110 }
1111
1112 Manager* manager_free(Manager *m) {
1113 UnitType c;
1114 int i;
1115 ExecDirectoryType dt;
1116
1117 if (!m)
1118 return NULL;
1119
1120 manager_clear_jobs_and_units(m);
1121
1122 for (c = 0; c < _UNIT_TYPE_MAX; c++)
1123 if (unit_vtable[c]->shutdown)
1124 unit_vtable[c]->shutdown(m);
1125
1126 /* If we reexecute ourselves, we keep the root cgroup
1127 * around */
1128 manager_shutdown_cgroup(m, m->exit_code != MANAGER_REEXECUTE);
1129
1130 lookup_paths_flush_generator(&m->lookup_paths);
1131
1132 bus_done(m);
1133
1134 dynamic_user_vacuum(m, false);
1135 hashmap_free(m->dynamic_users);
1136
1137 hashmap_free(m->units);
1138 hashmap_free(m->units_by_invocation_id);
1139 hashmap_free(m->jobs);
1140 hashmap_free(m->watch_pids1);
1141 hashmap_free(m->watch_pids2);
1142 hashmap_free(m->watch_bus);
1143
1144 set_free(m->startup_units);
1145 set_free(m->failed_units);
1146
1147 sd_event_source_unref(m->signal_event_source);
1148 sd_event_source_unref(m->notify_event_source);
1149 sd_event_source_unref(m->cgroups_agent_event_source);
1150 sd_event_source_unref(m->time_change_event_source);
1151 sd_event_source_unref(m->jobs_in_progress_event_source);
1152 sd_event_source_unref(m->run_queue_event_source);
1153 sd_event_source_unref(m->user_lookup_event_source);
1154
1155 safe_close(m->signal_fd);
1156 safe_close(m->notify_fd);
1157 safe_close(m->cgroups_agent_fd);
1158 safe_close(m->time_change_fd);
1159 safe_close_pair(m->user_lookup_fds);
1160
1161 manager_close_ask_password(m);
1162
1163 manager_close_idle_pipe(m);
1164
1165 udev_unref(m->udev);
1166 sd_event_unref(m->event);
1167
1168 free(m->notify_socket);
1169
1170 lookup_paths_free(&m->lookup_paths);
1171 strv_free(m->environment);
1172
1173 hashmap_free(m->cgroup_unit);
1174 set_free_free(m->unit_path_cache);
1175
1176 free(m->switch_root);
1177 free(m->switch_root_init);
1178
1179 for (i = 0; i < _RLIMIT_MAX; i++)
1180 m->rlimit[i] = mfree(m->rlimit[i]);
1181
1182 assert(hashmap_isempty(m->units_requiring_mounts_for));
1183 hashmap_free(m->units_requiring_mounts_for);
1184
1185 hashmap_free(m->uid_refs);
1186 hashmap_free(m->gid_refs);
1187
1188 for (dt = 0; dt < _EXEC_DIRECTORY_MAX; dt++)
1189 m->prefix[dt] = mfree(m->prefix[dt]);
1190
1191 return mfree(m);
1192 }
1193
1194 void manager_enumerate(Manager *m) {
1195 UnitType c;
1196
1197 assert(m);
1198
1199 /* Let's ask every type to load all units from disk/kernel
1200 * that it might know */
1201 for (c = 0; c < _UNIT_TYPE_MAX; c++) {
1202 if (!unit_type_supported(c)) {
1203 log_debug("Unit type .%s is not supported on this system.", unit_type_to_string(c));
1204 continue;
1205 }
1206
1207 if (!unit_vtable[c]->enumerate)
1208 continue;
1209
1210 unit_vtable[c]->enumerate(m);
1211 }
1212
1213 manager_dispatch_load_queue(m);
1214 }
1215
1216 static void manager_coldplug(Manager *m) {
1217 Iterator i;
1218 Unit *u;
1219 char *k;
1220 int r;
1221
1222 assert(m);
1223
1224 /* Then, let's set up their initial state. */
1225 HASHMAP_FOREACH_KEY(u, k, m->units, i) {
1226
1227 /* ignore aliases */
1228 if (u->id != k)
1229 continue;
1230
1231 r = unit_coldplug(u);
1232 if (r < 0)
1233 log_warning_errno(r, "We couldn't coldplug %s, proceeding anyway: %m", u->id);
1234 }
1235 }
1236
1237 static void manager_build_unit_path_cache(Manager *m) {
1238 char **i;
1239 int r;
1240
1241 assert(m);
1242
1243 set_free_free(m->unit_path_cache);
1244
1245 m->unit_path_cache = set_new(&string_hash_ops);
1246 if (!m->unit_path_cache) {
1247 r = -ENOMEM;
1248 goto fail;
1249 }
1250
1251 /* This simply builds a list of files we know exist, so that
1252 * we don't always have to go to disk */
1253
1254 STRV_FOREACH(i, m->lookup_paths.search_path) {
1255 _cleanup_closedir_ DIR *d = NULL;
1256 struct dirent *de;
1257
1258 d = opendir(*i);
1259 if (!d) {
1260 if (errno != ENOENT)
1261 log_warning_errno(errno, "Failed to open directory %s, ignoring: %m", *i);
1262 continue;
1263 }
1264
1265 FOREACH_DIRENT(de, d, r = -errno; goto fail) {
1266 char *p;
1267
1268 p = strjoin(streq(*i, "/") ? "" : *i, "/", de->d_name);
1269 if (!p) {
1270 r = -ENOMEM;
1271 goto fail;
1272 }
1273
1274 r = set_consume(m->unit_path_cache, p);
1275 if (r < 0)
1276 goto fail;
1277 }
1278 }
1279
1280 return;
1281
1282 fail:
1283 log_warning_errno(r, "Failed to build unit path cache, proceeding without: %m");
1284 m->unit_path_cache = set_free_free(m->unit_path_cache);
1285 }
1286
1287 static void manager_distribute_fds(Manager *m, FDSet *fds) {
1288 Iterator i;
1289 Unit *u;
1290
1291 assert(m);
1292
1293 HASHMAP_FOREACH(u, m->units, i) {
1294
1295 if (fdset_size(fds) <= 0)
1296 break;
1297
1298 if (!UNIT_VTABLE(u)->distribute_fds)
1299 continue;
1300
1301 UNIT_VTABLE(u)->distribute_fds(u, fds);
1302 }
1303 }
1304
1305 int manager_startup(Manager *m, FILE *serialization, FDSet *fds) {
1306 int r, q;
1307
1308 assert(m);
1309
1310 r = lookup_paths_init(&m->lookup_paths, m->unit_file_scope, 0, NULL);
1311 if (r < 0)
1312 return r;
1313
1314 r = manager_run_environment_generators(m);
1315 if (r < 0)
1316 return r;
1317
1318 /* Make sure the transient directory always exists, so that it remains in the search path */
1319 if (!m->test_run) {
1320 r = mkdir_p_label(m->lookup_paths.transient, 0755);
1321 if (r < 0)
1322 return r;
1323 }
1324
1325 dual_timestamp_get(&m->generators_start_timestamp);
1326 r = manager_run_generators(m);
1327 dual_timestamp_get(&m->generators_finish_timestamp);
1328 if (r < 0)
1329 return r;
1330
1331 if (m->first_boot && m->unit_file_scope == UNIT_FILE_SYSTEM) {
1332 q = unit_file_preset_all(UNIT_FILE_SYSTEM, 0, NULL, UNIT_FILE_PRESET_ENABLE_ONLY, NULL, 0);
1333 if (q < 0)
1334 log_full_errno(q == -EEXIST ? LOG_NOTICE : LOG_WARNING, q, "Failed to populate /etc with preset unit settings, ignoring: %m");
1335 else
1336 log_info("Populated /etc with preset unit settings.");
1337 }
1338
1339 lookup_paths_reduce(&m->lookup_paths);
1340 manager_build_unit_path_cache(m);
1341
1342 /* If we will deserialize make sure that during enumeration
1343 * this is already known, so we increase the counter here
1344 * already */
1345 if (serialization)
1346 m->n_reloading++;
1347
1348 /* First, enumerate what we can from all config files */
1349 dual_timestamp_get(&m->units_load_start_timestamp);
1350 manager_enumerate(m);
1351 dual_timestamp_get(&m->units_load_finish_timestamp);
1352
1353 /* Second, deserialize if there is something to deserialize */
1354 if (serialization) {
1355 r = manager_deserialize(m, serialization, fds);
1356 if (r < 0)
1357 log_error_errno(r, "Deserialization failed: %m");
1358 }
1359
1360 /* Any fds left? Find some unit which wants them. This is
1361 * useful to allow container managers to pass some file
1362 * descriptors to us pre-initialized. This enables
1363 * socket-based activation of entire containers. */
1364 manager_distribute_fds(m, fds);
1365
1366 /* We might have deserialized the notify fd, but if we didn't
1367 * then let's create the bus now */
1368 q = manager_setup_notify(m);
1369 if (q < 0 && r == 0)
1370 r = q;
1371
1372 q = manager_setup_cgroups_agent(m);
1373 if (q < 0 && r == 0)
1374 r = q;
1375
1376 q = manager_setup_user_lookup_fd(m);
1377 if (q < 0 && r == 0)
1378 r = q;
1379
1380 /* Let's connect to the bus now. */
1381 (void) manager_connect_bus(m, !!serialization);
1382
1383 (void) bus_track_coldplug(m, &m->subscribed, false, m->deserialized_subscribed);
1384 m->deserialized_subscribed = strv_free(m->deserialized_subscribed);
1385
1386 /* Third, fire things up! */
1387 manager_coldplug(m);
1388
1389 /* Release any dynamic users no longer referenced */
1390 dynamic_user_vacuum(m, true);
1391
1392 /* Release any references to UIDs/GIDs no longer referenced, and destroy any IPC owned by them */
1393 manager_vacuum_uid_refs(m);
1394 manager_vacuum_gid_refs(m);
1395
1396 if (serialization) {
1397 assert(m->n_reloading > 0);
1398 m->n_reloading--;
1399
1400 /* Let's wait for the UnitNew/JobNew messages being
1401 * sent, before we notify that the reload is
1402 * finished */
1403 m->send_reloading_done = true;
1404 }
1405
1406 return r;
1407 }
1408
1409 int manager_add_job(Manager *m, JobType type, Unit *unit, JobMode mode, sd_bus_error *e, Job **_ret) {
1410 int r;
1411 Transaction *tr;
1412
1413 assert(m);
1414 assert(type < _JOB_TYPE_MAX);
1415 assert(unit);
1416 assert(mode < _JOB_MODE_MAX);
1417
1418 if (mode == JOB_ISOLATE && type != JOB_START)
1419 return sd_bus_error_setf(e, SD_BUS_ERROR_INVALID_ARGS, "Isolate is only valid for start.");
1420
1421 if (mode == JOB_ISOLATE && !unit->allow_isolate)
1422 return sd_bus_error_setf(e, BUS_ERROR_NO_ISOLATION, "Operation refused, unit may not be isolated.");
1423
1424 log_unit_debug(unit, "Trying to enqueue job %s/%s/%s", unit->id, job_type_to_string(type), job_mode_to_string(mode));
1425
1426 type = job_type_collapse(type, unit);
1427
1428 tr = transaction_new(mode == JOB_REPLACE_IRREVERSIBLY);
1429 if (!tr)
1430 return -ENOMEM;
1431
1432 r = transaction_add_job_and_dependencies(tr, type, unit, NULL, true, false,
1433 mode == JOB_IGNORE_DEPENDENCIES || mode == JOB_IGNORE_REQUIREMENTS,
1434 mode == JOB_IGNORE_DEPENDENCIES, e);
1435 if (r < 0)
1436 goto tr_abort;
1437
1438 if (mode == JOB_ISOLATE) {
1439 r = transaction_add_isolate_jobs(tr, m);
1440 if (r < 0)
1441 goto tr_abort;
1442 }
1443
1444 r = transaction_activate(tr, m, mode, e);
1445 if (r < 0)
1446 goto tr_abort;
1447
1448 log_unit_debug(unit,
1449 "Enqueued job %s/%s as %u", unit->id,
1450 job_type_to_string(type), (unsigned) tr->anchor_job->id);
1451
1452 if (_ret)
1453 *_ret = tr->anchor_job;
1454
1455 transaction_free(tr);
1456 return 0;
1457
1458 tr_abort:
1459 transaction_abort(tr);
1460 transaction_free(tr);
1461 return r;
1462 }
1463
1464 int manager_add_job_by_name(Manager *m, JobType type, const char *name, JobMode mode, sd_bus_error *e, Job **ret) {
1465 Unit *unit = NULL; /* just to appease gcc, initialization is not really necessary */
1466 int r;
1467
1468 assert(m);
1469 assert(type < _JOB_TYPE_MAX);
1470 assert(name);
1471 assert(mode < _JOB_MODE_MAX);
1472
1473 r = manager_load_unit(m, name, NULL, NULL, &unit);
1474 if (r < 0)
1475 return r;
1476 assert(unit);
1477
1478 return manager_add_job(m, type, unit, mode, e, ret);
1479 }
1480
1481 int manager_add_job_by_name_and_warn(Manager *m, JobType type, const char *name, JobMode mode, Job **ret) {
1482 _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL;
1483 int r;
1484
1485 assert(m);
1486 assert(type < _JOB_TYPE_MAX);
1487 assert(name);
1488 assert(mode < _JOB_MODE_MAX);
1489
1490 r = manager_add_job_by_name(m, type, name, mode, &error, ret);
1491 if (r < 0)
1492 return log_warning_errno(r, "Failed to enqueue %s job for %s: %s", job_mode_to_string(mode), name, bus_error_message(&error, r));
1493
1494 return r;
1495 }
1496
1497 int manager_propagate_reload(Manager *m, Unit *unit, JobMode mode, sd_bus_error *e) {
1498 int r;
1499 Transaction *tr;
1500
1501 assert(m);
1502 assert(unit);
1503 assert(mode < _JOB_MODE_MAX);
1504 assert(mode != JOB_ISOLATE); /* Isolate is only valid for start */
1505
1506 tr = transaction_new(mode == JOB_REPLACE_IRREVERSIBLY);
1507 if (!tr)
1508 return -ENOMEM;
1509
1510 /* We need an anchor job */
1511 r = transaction_add_job_and_dependencies(tr, JOB_NOP, unit, NULL, false, false, true, true, e);
1512 if (r < 0)
1513 goto tr_abort;
1514
1515 /* Failure in adding individual dependencies is ignored, so this always succeeds. */
1516 transaction_add_propagate_reload_jobs(tr, unit, tr->anchor_job, mode == JOB_IGNORE_DEPENDENCIES, e);
1517
1518 r = transaction_activate(tr, m, mode, e);
1519 if (r < 0)
1520 goto tr_abort;
1521
1522 transaction_free(tr);
1523 return 0;
1524
1525 tr_abort:
1526 transaction_abort(tr);
1527 transaction_free(tr);
1528 return r;
1529 }
1530
1531 Job *manager_get_job(Manager *m, uint32_t id) {
1532 assert(m);
1533
1534 return hashmap_get(m->jobs, UINT32_TO_PTR(id));
1535 }
1536
1537 Unit *manager_get_unit(Manager *m, const char *name) {
1538 assert(m);
1539 assert(name);
1540
1541 return hashmap_get(m->units, name);
1542 }
1543
1544 unsigned manager_dispatch_load_queue(Manager *m) {
1545 Unit *u;
1546 unsigned n = 0;
1547
1548 assert(m);
1549
1550 /* Make sure we are not run recursively */
1551 if (m->dispatching_load_queue)
1552 return 0;
1553
1554 m->dispatching_load_queue = true;
1555
1556 /* Dispatches the load queue. Takes a unit from the queue and
1557 * tries to load its data until the queue is empty */
1558
1559 while ((u = m->load_queue)) {
1560 assert(u->in_load_queue);
1561
1562 unit_load(u);
1563 n++;
1564 }
1565
1566 m->dispatching_load_queue = false;
1567 return n;
1568 }
1569
1570 int manager_load_unit_prepare(
1571 Manager *m,
1572 const char *name,
1573 const char *path,
1574 sd_bus_error *e,
1575 Unit **_ret) {
1576
1577 Unit *ret;
1578 UnitType t;
1579 int r;
1580
1581 assert(m);
1582 assert(name || path);
1583 assert(_ret);
1584
1585 /* This will prepare the unit for loading, but not actually
1586 * load anything from disk. */
1587
1588 if (path && !is_path(path))
1589 return sd_bus_error_setf(e, SD_BUS_ERROR_INVALID_ARGS, "Path %s is not absolute.", path);
1590
1591 if (!name)
1592 name = basename(path);
1593
1594 t = unit_name_to_type(name);
1595
1596 if (t == _UNIT_TYPE_INVALID || !unit_name_is_valid(name, UNIT_NAME_PLAIN|UNIT_NAME_INSTANCE)) {
1597 if (unit_name_is_valid(name, UNIT_NAME_TEMPLATE))
1598 return sd_bus_error_setf(e, SD_BUS_ERROR_INVALID_ARGS, "Unit name %s is missing the instance name.", name);
1599
1600 return sd_bus_error_setf(e, SD_BUS_ERROR_INVALID_ARGS, "Unit name %s is not valid.", name);
1601 }
1602
1603 ret = manager_get_unit(m, name);
1604 if (ret) {
1605 *_ret = ret;
1606 return 1;
1607 }
1608
1609 ret = unit_new(m, unit_vtable[t]->object_size);
1610 if (!ret)
1611 return -ENOMEM;
1612
1613 if (path) {
1614 ret->fragment_path = strdup(path);
1615 if (!ret->fragment_path) {
1616 unit_free(ret);
1617 return -ENOMEM;
1618 }
1619 }
1620
1621 r = unit_add_name(ret, name);
1622 if (r < 0) {
1623 unit_free(ret);
1624 return r;
1625 }
1626
1627 unit_add_to_load_queue(ret);
1628 unit_add_to_dbus_queue(ret);
1629 unit_add_to_gc_queue(ret);
1630
1631 *_ret = ret;
1632
1633 return 0;
1634 }
1635
1636 int manager_load_unit(
1637 Manager *m,
1638 const char *name,
1639 const char *path,
1640 sd_bus_error *e,
1641 Unit **_ret) {
1642
1643 int r;
1644
1645 assert(m);
1646 assert(_ret);
1647
1648 /* This will load the service information files, but not actually
1649 * start any services or anything. */
1650
1651 r = manager_load_unit_prepare(m, name, path, e, _ret);
1652 if (r != 0)
1653 return r;
1654
1655 manager_dispatch_load_queue(m);
1656
1657 *_ret = unit_follow_merge(*_ret);
1658
1659 return 0;
1660 }
1661
1662 void manager_dump_jobs(Manager *s, FILE *f, const char *prefix) {
1663 Iterator i;
1664 Job *j;
1665
1666 assert(s);
1667 assert(f);
1668
1669 HASHMAP_FOREACH(j, s->jobs, i)
1670 job_dump(j, f, prefix);
1671 }
1672
1673 void manager_dump_units(Manager *s, FILE *f, const char *prefix) {
1674 Iterator i;
1675 Unit *u;
1676 const char *t;
1677
1678 assert(s);
1679 assert(f);
1680
1681 HASHMAP_FOREACH_KEY(u, t, s->units, i)
1682 if (u->id == t)
1683 unit_dump(u, f, prefix);
1684 }
1685
1686 void manager_clear_jobs(Manager *m) {
1687 Job *j;
1688
1689 assert(m);
1690
1691 while ((j = hashmap_first(m->jobs)))
1692 /* No need to recurse. We're cancelling all jobs. */
1693 job_finish_and_invalidate(j, JOB_CANCELED, false, false);
1694 }
1695
1696 static int manager_dispatch_run_queue(sd_event_source *source, void *userdata) {
1697 Manager *m = userdata;
1698 Job *j;
1699
1700 assert(source);
1701 assert(m);
1702
1703 while ((j = m->run_queue)) {
1704 assert(j->installed);
1705 assert(j->in_run_queue);
1706
1707 job_run_and_invalidate(j);
1708 }
1709
1710 if (m->n_running_jobs > 0)
1711 manager_watch_jobs_in_progress(m);
1712
1713 if (m->n_on_console > 0)
1714 manager_watch_idle_pipe(m);
1715
1716 return 1;
1717 }
1718
1719 static unsigned manager_dispatch_dbus_queue(Manager *m) {
1720 Job *j;
1721 Unit *u;
1722 unsigned n = 0;
1723
1724 assert(m);
1725
1726 if (m->dispatching_dbus_queue)
1727 return 0;
1728
1729 m->dispatching_dbus_queue = true;
1730
1731 while ((u = m->dbus_unit_queue)) {
1732 assert(u->in_dbus_queue);
1733
1734 bus_unit_send_change_signal(u);
1735 n++;
1736 }
1737
1738 while ((j = m->dbus_job_queue)) {
1739 assert(j->in_dbus_queue);
1740
1741 bus_job_send_change_signal(j);
1742 n++;
1743 }
1744
1745 m->dispatching_dbus_queue = false;
1746
1747 if (m->send_reloading_done) {
1748 m->send_reloading_done = false;
1749
1750 bus_manager_send_reloading(m, false);
1751 }
1752
1753 if (m->queued_message)
1754 bus_send_queued_message(m);
1755
1756 return n;
1757 }
1758
1759 static int manager_dispatch_cgroups_agent_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
1760 Manager *m = userdata;
1761 char buf[PATH_MAX+1];
1762 ssize_t n;
1763
1764 n = recv(fd, buf, sizeof(buf), 0);
1765 if (n < 0)
1766 return log_error_errno(errno, "Failed to read cgroups agent message: %m");
1767 if (n == 0) {
1768 log_error("Got zero-length cgroups agent message, ignoring.");
1769 return 0;
1770 }
1771 if ((size_t) n >= sizeof(buf)) {
1772 log_error("Got overly long cgroups agent message, ignoring.");
1773 return 0;
1774 }
1775
1776 if (memchr(buf, 0, n)) {
1777 log_error("Got cgroups agent message with embedded NUL byte, ignoring.");
1778 return 0;
1779 }
1780 buf[n] = 0;
1781
1782 manager_notify_cgroup_empty(m, buf);
1783 bus_forward_agent_released(m, buf);
1784
1785 return 0;
1786 }
1787
1788 static void manager_invoke_notify_message(Manager *m, Unit *u, pid_t pid, const char *buf, FDSet *fds) {
1789 _cleanup_strv_free_ char **tags = NULL;
1790
1791 assert(m);
1792 assert(u);
1793 assert(buf);
1794
1795 tags = strv_split(buf, "\n\r");
1796 if (!tags) {
1797 log_oom();
1798 return;
1799 }
1800
1801 if (UNIT_VTABLE(u)->notify_message)
1802 UNIT_VTABLE(u)->notify_message(u, pid, tags, fds);
1803 else if (_unlikely_(log_get_max_level() >= LOG_DEBUG)) {
1804 _cleanup_free_ char *x = NULL, *y = NULL;
1805
1806 x = cescape(buf);
1807 if (x)
1808 y = ellipsize(x, 20, 90);
1809 log_unit_debug(u, "Got notification message \"%s\", ignoring.", strnull(y));
1810 }
1811 }
1812
1813 static int manager_dispatch_notify_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
1814
1815 _cleanup_fdset_free_ FDSet *fds = NULL;
1816 Manager *m = userdata;
1817 char buf[NOTIFY_BUFFER_MAX+1];
1818 struct iovec iovec = {
1819 .iov_base = buf,
1820 .iov_len = sizeof(buf)-1,
1821 };
1822 union {
1823 struct cmsghdr cmsghdr;
1824 uint8_t buf[CMSG_SPACE(sizeof(struct ucred)) +
1825 CMSG_SPACE(sizeof(int) * NOTIFY_FD_MAX)];
1826 } control = {};
1827 struct msghdr msghdr = {
1828 .msg_iov = &iovec,
1829 .msg_iovlen = 1,
1830 .msg_control = &control,
1831 .msg_controllen = sizeof(control),
1832 };
1833
1834 struct cmsghdr *cmsg;
1835 struct ucred *ucred = NULL;
1836 Unit *u1, *u2, *u3;
1837 int r, *fd_array = NULL;
1838 unsigned n_fds = 0;
1839 ssize_t n;
1840
1841 assert(m);
1842 assert(m->notify_fd == fd);
1843
1844 if (revents != EPOLLIN) {
1845 log_warning("Got unexpected poll event for notify fd.");
1846 return 0;
1847 }
1848
1849 n = recvmsg(m->notify_fd, &msghdr, MSG_DONTWAIT|MSG_CMSG_CLOEXEC|MSG_TRUNC);
1850 if (n < 0) {
1851 if (IN_SET(errno, EAGAIN, EINTR))
1852 return 0; /* Spurious wakeup, try again */
1853
1854 /* If this is any other, real error, then let's stop processing this socket. This of course means we
1855 * won't take notification messages anymore, but that's still better than busy looping around this:
1856 * being woken up over and over again but being unable to actually read the message off the socket. */
1857 return log_error_errno(errno, "Failed to receive notification message: %m");
1858 }
1859
1860 CMSG_FOREACH(cmsg, &msghdr) {
1861 if (cmsg->cmsg_level == SOL_SOCKET && cmsg->cmsg_type == SCM_RIGHTS) {
1862
1863 fd_array = (int*) CMSG_DATA(cmsg);
1864 n_fds = (cmsg->cmsg_len - CMSG_LEN(0)) / sizeof(int);
1865
1866 } else if (cmsg->cmsg_level == SOL_SOCKET &&
1867 cmsg->cmsg_type == SCM_CREDENTIALS &&
1868 cmsg->cmsg_len == CMSG_LEN(sizeof(struct ucred))) {
1869
1870 ucred = (struct ucred*) CMSG_DATA(cmsg);
1871 }
1872 }
1873
1874 if (n_fds > 0) {
1875 assert(fd_array);
1876
1877 r = fdset_new_array(&fds, fd_array, n_fds);
1878 if (r < 0) {
1879 close_many(fd_array, n_fds);
1880 log_oom();
1881 return 0;
1882 }
1883 }
1884
1885 if (!ucred || ucred->pid <= 0) {
1886 log_warning("Received notify message without valid credentials. Ignoring.");
1887 return 0;
1888 }
1889
1890 if ((size_t) n >= sizeof(buf) || (msghdr.msg_flags & MSG_TRUNC)) {
1891 log_warning("Received notify message exceeded maximum size. Ignoring.");
1892 return 0;
1893 }
1894
1895 /* As extra safety check, let's make sure the string we get doesn't contain embedded NUL bytes. We permit one
1896 * trailing NUL byte in the message, but don't expect it. */
1897 if (n > 1 && memchr(buf, 0, n-1)) {
1898 log_warning("Received notify message with embedded NUL bytes. Ignoring.");
1899 return 0;
1900 }
1901
1902 /* Make sure it's NUL-terminated. */
1903 buf[n] = 0;
1904
1905 /* Notify every unit that might be interested, but try
1906 * to avoid notifying the same one multiple times. */
1907 u1 = manager_get_unit_by_pid_cgroup(m, ucred->pid);
1908 if (u1)
1909 manager_invoke_notify_message(m, u1, ucred->pid, buf, fds);
1910
1911 u2 = hashmap_get(m->watch_pids1, PID_TO_PTR(ucred->pid));
1912 if (u2 && u2 != u1)
1913 manager_invoke_notify_message(m, u2, ucred->pid, buf, fds);
1914
1915 u3 = hashmap_get(m->watch_pids2, PID_TO_PTR(ucred->pid));
1916 if (u3 && u3 != u2 && u3 != u1)
1917 manager_invoke_notify_message(m, u3, ucred->pid, buf, fds);
1918
1919 if (!u1 && !u2 && !u3)
1920 log_warning("Cannot find unit for notify message of PID "PID_FMT".", ucred->pid);
1921
1922 if (fdset_size(fds) > 0)
1923 log_warning("Got extra auxiliary fds with notification message, closing them.");
1924
1925 return 0;
1926 }
1927
1928 static void invoke_sigchld_event(Manager *m, Unit *u, const siginfo_t *si) {
1929 uint64_t iteration;
1930
1931 assert(m);
1932 assert(u);
1933 assert(si);
1934
1935 sd_event_get_iteration(m->event, &iteration);
1936
1937 log_unit_debug(u, "Child "PID_FMT" belongs to %s", si->si_pid, u->id);
1938
1939 unit_unwatch_pid(u, si->si_pid);
1940
1941 if (UNIT_VTABLE(u)->sigchld_event) {
1942 if (set_size(u->pids) <= 1 ||
1943 iteration != u->sigchldgen ||
1944 unit_main_pid(u) == si->si_pid ||
1945 unit_control_pid(u) == si->si_pid) {
1946 UNIT_VTABLE(u)->sigchld_event(u, si->si_pid, si->si_code, si->si_status);
1947 u->sigchldgen = iteration;
1948 } else
1949 log_debug("%s already issued a sigchld this iteration %" PRIu64 ", skipping. Pids still being watched %d", u->id, iteration, set_size(u->pids));
1950 }
1951 }
1952
1953 static int manager_dispatch_sigchld(Manager *m) {
1954 assert(m);
1955
1956 for (;;) {
1957 siginfo_t si = {};
1958
1959 /* First we call waitd() for a PID and do not reap the
1960 * zombie. That way we can still access /proc/$PID for
1961 * it while it is a zombie. */
1962 if (waitid(P_ALL, 0, &si, WEXITED|WNOHANG|WNOWAIT) < 0) {
1963
1964 if (errno == ECHILD)
1965 break;
1966
1967 if (errno == EINTR)
1968 continue;
1969
1970 return -errno;
1971 }
1972
1973 if (si.si_pid <= 0)
1974 break;
1975
1976 if (si.si_code == CLD_EXITED || si.si_code == CLD_KILLED || si.si_code == CLD_DUMPED) {
1977 _cleanup_free_ char *name = NULL;
1978 Unit *u1, *u2, *u3;
1979
1980 get_process_comm(si.si_pid, &name);
1981
1982 log_debug("Child "PID_FMT" (%s) died (code=%s, status=%i/%s)",
1983 si.si_pid, strna(name),
1984 sigchld_code_to_string(si.si_code),
1985 si.si_status,
1986 strna(si.si_code == CLD_EXITED
1987 ? exit_status_to_string(si.si_status, EXIT_STATUS_FULL)
1988 : signal_to_string(si.si_status)));
1989
1990 /* And now figure out the unit this belongs
1991 * to, it might be multiple... */
1992 u1 = manager_get_unit_by_pid_cgroup(m, si.si_pid);
1993 if (u1)
1994 invoke_sigchld_event(m, u1, &si);
1995 u2 = hashmap_get(m->watch_pids1, PID_TO_PTR(si.si_pid));
1996 if (u2 && u2 != u1)
1997 invoke_sigchld_event(m, u2, &si);
1998 u3 = hashmap_get(m->watch_pids2, PID_TO_PTR(si.si_pid));
1999 if (u3 && u3 != u2 && u3 != u1)
2000 invoke_sigchld_event(m, u3, &si);
2001 }
2002
2003 /* And now, we actually reap the zombie. */
2004 if (waitid(P_PID, si.si_pid, &si, WEXITED) < 0) {
2005 if (errno == EINTR)
2006 continue;
2007
2008 return -errno;
2009 }
2010 }
2011
2012 return 0;
2013 }
2014
2015 static int manager_start_target(Manager *m, const char *name, JobMode mode) {
2016 _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL;
2017 int r;
2018
2019 log_debug("Activating special unit %s", name);
2020
2021 r = manager_add_job_by_name(m, JOB_START, name, mode, &error, NULL);
2022 if (r < 0)
2023 log_error("Failed to enqueue %s job: %s", name, bus_error_message(&error, r));
2024
2025 return r;
2026 }
2027
2028 static void manager_handle_ctrl_alt_del(Manager *m) {
2029 /* If the user presses C-A-D more than
2030 * 7 times within 2s, we reboot/shutdown immediately,
2031 * unless it was disabled in system.conf */
2032
2033 if (ratelimit_test(&m->ctrl_alt_del_ratelimit) || m->cad_burst_action == EMERGENCY_ACTION_NONE)
2034 manager_start_target(m, SPECIAL_CTRL_ALT_DEL_TARGET, JOB_REPLACE_IRREVERSIBLY);
2035 else
2036 emergency_action(m, m->cad_burst_action, NULL,
2037 "Ctrl-Alt-Del was pressed more than 7 times within 2s");
2038 }
2039
2040 static int manager_dispatch_signal_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
2041 Manager *m = userdata;
2042 ssize_t n;
2043 struct signalfd_siginfo sfsi;
2044 bool sigchld = false;
2045 int r;
2046
2047 assert(m);
2048 assert(m->signal_fd == fd);
2049
2050 if (revents != EPOLLIN) {
2051 log_warning("Got unexpected events from signal file descriptor.");
2052 return 0;
2053 }
2054
2055 for (;;) {
2056 n = read(m->signal_fd, &sfsi, sizeof(sfsi));
2057 if (n != sizeof(sfsi)) {
2058 if (n >= 0) {
2059 log_warning("Truncated read from signal fd (%zu bytes)!", n);
2060 return 0;
2061 }
2062
2063 if (IN_SET(errno, EINTR, EAGAIN))
2064 break;
2065
2066 /* We return an error here, which will kill this handler,
2067 * to avoid a busy loop on read error. */
2068 return log_error_errno(errno, "Reading from signal fd failed: %m");
2069 }
2070
2071 log_received_signal(sfsi.ssi_signo == SIGCHLD ||
2072 (sfsi.ssi_signo == SIGTERM && MANAGER_IS_USER(m))
2073 ? LOG_DEBUG : LOG_INFO,
2074 &sfsi);
2075
2076 switch (sfsi.ssi_signo) {
2077
2078 case SIGCHLD:
2079 sigchld = true;
2080 break;
2081
2082 case SIGTERM:
2083 if (MANAGER_IS_SYSTEM(m)) {
2084 /* This is for compatibility with the
2085 * original sysvinit */
2086 r = verify_run_space_and_log("Refusing to reexecute");
2087 if (r >= 0)
2088 m->exit_code = MANAGER_REEXECUTE;
2089 break;
2090 }
2091
2092 /* Fall through */
2093
2094 case SIGINT:
2095 if (MANAGER_IS_SYSTEM(m)) {
2096 manager_handle_ctrl_alt_del(m);
2097 break;
2098 }
2099
2100 /* Run the exit target if there is one, if not, just exit. */
2101 if (manager_start_target(m, SPECIAL_EXIT_TARGET, JOB_REPLACE) < 0) {
2102 m->exit_code = MANAGER_EXIT;
2103 return 0;
2104 }
2105
2106 break;
2107
2108 case SIGWINCH:
2109 if (MANAGER_IS_SYSTEM(m))
2110 manager_start_target(m, SPECIAL_KBREQUEST_TARGET, JOB_REPLACE);
2111
2112 /* This is a nop on non-init */
2113 break;
2114
2115 case SIGPWR:
2116 if (MANAGER_IS_SYSTEM(m))
2117 manager_start_target(m, SPECIAL_SIGPWR_TARGET, JOB_REPLACE);
2118
2119 /* This is a nop on non-init */
2120 break;
2121
2122 case SIGUSR1: {
2123 Unit *u;
2124
2125 u = manager_get_unit(m, SPECIAL_DBUS_SERVICE);
2126
2127 if (!u || UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u))) {
2128 log_info("Trying to reconnect to bus...");
2129 bus_init(m, true);
2130 }
2131
2132 if (!u || !UNIT_IS_ACTIVE_OR_ACTIVATING(unit_active_state(u))) {
2133 log_info("Loading D-Bus service...");
2134 manager_start_target(m, SPECIAL_DBUS_SERVICE, JOB_REPLACE);
2135 }
2136
2137 break;
2138 }
2139
2140 case SIGUSR2: {
2141 _cleanup_free_ char *dump = NULL;
2142 _cleanup_fclose_ FILE *f = NULL;
2143 size_t size;
2144
2145 f = open_memstream(&dump, &size);
2146 if (!f) {
2147 log_warning_errno(errno, "Failed to allocate memory stream: %m");
2148 break;
2149 }
2150
2151 manager_dump_units(m, f, "\t");
2152 manager_dump_jobs(m, f, "\t");
2153
2154 r = fflush_and_check(f);
2155 if (r < 0) {
2156 log_warning_errno(r, "Failed to write status stream: %m");
2157 break;
2158 }
2159
2160 log_dump(LOG_INFO, dump);
2161 break;
2162 }
2163
2164 case SIGHUP:
2165 r = verify_run_space_and_log("Refusing to reload");
2166 if (r >= 0)
2167 m->exit_code = MANAGER_RELOAD;
2168 break;
2169
2170 default: {
2171
2172 /* Starting SIGRTMIN+0 */
2173 static const char * const target_table[] = {
2174 [0] = SPECIAL_DEFAULT_TARGET,
2175 [1] = SPECIAL_RESCUE_TARGET,
2176 [2] = SPECIAL_EMERGENCY_TARGET,
2177 [3] = SPECIAL_HALT_TARGET,
2178 [4] = SPECIAL_POWEROFF_TARGET,
2179 [5] = SPECIAL_REBOOT_TARGET,
2180 [6] = SPECIAL_KEXEC_TARGET
2181 };
2182
2183 /* Starting SIGRTMIN+13, so that target halt and system halt are 10 apart */
2184 static const ManagerExitCode code_table[] = {
2185 [0] = MANAGER_HALT,
2186 [1] = MANAGER_POWEROFF,
2187 [2] = MANAGER_REBOOT,
2188 [3] = MANAGER_KEXEC
2189 };
2190
2191 if ((int) sfsi.ssi_signo >= SIGRTMIN+0 &&
2192 (int) sfsi.ssi_signo < SIGRTMIN+(int) ELEMENTSOF(target_table)) {
2193 int idx = (int) sfsi.ssi_signo - SIGRTMIN;
2194 manager_start_target(m, target_table[idx],
2195 (idx == 1 || idx == 2) ? JOB_ISOLATE : JOB_REPLACE);
2196 break;
2197 }
2198
2199 if ((int) sfsi.ssi_signo >= SIGRTMIN+13 &&
2200 (int) sfsi.ssi_signo < SIGRTMIN+13+(int) ELEMENTSOF(code_table)) {
2201 m->exit_code = code_table[sfsi.ssi_signo - SIGRTMIN - 13];
2202 break;
2203 }
2204
2205 switch (sfsi.ssi_signo - SIGRTMIN) {
2206
2207 case 20:
2208 manager_set_show_status(m, SHOW_STATUS_YES);
2209 break;
2210
2211 case 21:
2212 manager_set_show_status(m, SHOW_STATUS_NO);
2213 break;
2214
2215 case 22:
2216 log_set_max_level(LOG_DEBUG);
2217 log_info("Setting log level to debug.");
2218 break;
2219
2220 case 23:
2221 log_set_max_level(LOG_INFO);
2222 log_info("Setting log level to info.");
2223 break;
2224
2225 case 24:
2226 if (MANAGER_IS_USER(m)) {
2227 m->exit_code = MANAGER_EXIT;
2228 return 0;
2229 }
2230
2231 /* This is a nop on init */
2232 break;
2233
2234 case 26:
2235 case 29: /* compatibility: used to be mapped to LOG_TARGET_SYSLOG_OR_KMSG */
2236 log_set_target(LOG_TARGET_JOURNAL_OR_KMSG);
2237 log_notice("Setting log target to journal-or-kmsg.");
2238 break;
2239
2240 case 27:
2241 log_set_target(LOG_TARGET_CONSOLE);
2242 log_notice("Setting log target to console.");
2243 break;
2244
2245 case 28:
2246 log_set_target(LOG_TARGET_KMSG);
2247 log_notice("Setting log target to kmsg.");
2248 break;
2249
2250 default:
2251 log_warning("Got unhandled signal <%s>.", signal_to_string(sfsi.ssi_signo));
2252 }
2253 }
2254 }
2255 }
2256
2257 if (sigchld)
2258 manager_dispatch_sigchld(m);
2259
2260 return 0;
2261 }
2262
2263 static int manager_dispatch_time_change_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
2264 Manager *m = userdata;
2265 Iterator i;
2266 Unit *u;
2267
2268 assert(m);
2269 assert(m->time_change_fd == fd);
2270
2271 log_struct(LOG_DEBUG,
2272 "MESSAGE_ID=" SD_MESSAGE_TIME_CHANGE_STR,
2273 LOG_MESSAGE("Time has been changed"),
2274 NULL);
2275
2276 /* Restart the watch */
2277 m->time_change_event_source = sd_event_source_unref(m->time_change_event_source);
2278 m->time_change_fd = safe_close(m->time_change_fd);
2279
2280 manager_setup_time_change(m);
2281
2282 HASHMAP_FOREACH(u, m->units, i)
2283 if (UNIT_VTABLE(u)->time_change)
2284 UNIT_VTABLE(u)->time_change(u);
2285
2286 return 0;
2287 }
2288
2289 static int manager_dispatch_idle_pipe_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
2290 Manager *m = userdata;
2291
2292 assert(m);
2293 assert(m->idle_pipe[2] == fd);
2294
2295 m->no_console_output = m->n_on_console > 0;
2296
2297 manager_close_idle_pipe(m);
2298
2299 return 0;
2300 }
2301
2302 static int manager_dispatch_jobs_in_progress(sd_event_source *source, usec_t usec, void *userdata) {
2303 Manager *m = userdata;
2304 int r;
2305 uint64_t next;
2306
2307 assert(m);
2308 assert(source);
2309
2310 manager_print_jobs_in_progress(m);
2311
2312 next = now(CLOCK_MONOTONIC) + JOBS_IN_PROGRESS_PERIOD_USEC;
2313 r = sd_event_source_set_time(source, next);
2314 if (r < 0)
2315 return r;
2316
2317 return sd_event_source_set_enabled(source, SD_EVENT_ONESHOT);
2318 }
2319
2320 int manager_loop(Manager *m) {
2321 int r;
2322
2323 RATELIMIT_DEFINE(rl, 1*USEC_PER_SEC, 50000);
2324
2325 assert(m);
2326 m->exit_code = MANAGER_OK;
2327
2328 /* Release the path cache */
2329 m->unit_path_cache = set_free_free(m->unit_path_cache);
2330
2331 manager_check_finished(m);
2332
2333 /* There might still be some zombies hanging around from
2334 * before we were exec()'ed. Let's reap them. */
2335 r = manager_dispatch_sigchld(m);
2336 if (r < 0)
2337 return r;
2338
2339 while (m->exit_code == MANAGER_OK) {
2340 usec_t wait_usec;
2341
2342 if (m->runtime_watchdog > 0 && m->runtime_watchdog != USEC_INFINITY && MANAGER_IS_SYSTEM(m))
2343 watchdog_ping();
2344
2345 if (!ratelimit_test(&rl)) {
2346 /* Yay, something is going seriously wrong, pause a little */
2347 log_warning("Looping too fast. Throttling execution a little.");
2348 sleep(1);
2349 }
2350
2351 if (manager_dispatch_load_queue(m) > 0)
2352 continue;
2353
2354 if (manager_dispatch_gc_job_queue(m) > 0)
2355 continue;
2356
2357 if (manager_dispatch_gc_unit_queue(m) > 0)
2358 continue;
2359
2360 if (manager_dispatch_cleanup_queue(m) > 0)
2361 continue;
2362
2363 if (manager_dispatch_cgroup_queue(m) > 0)
2364 continue;
2365
2366 if (manager_dispatch_dbus_queue(m) > 0)
2367 continue;
2368
2369 /* Sleep for half the watchdog time */
2370 if (m->runtime_watchdog > 0 && m->runtime_watchdog != USEC_INFINITY && MANAGER_IS_SYSTEM(m)) {
2371 wait_usec = m->runtime_watchdog / 2;
2372 if (wait_usec <= 0)
2373 wait_usec = 1;
2374 } else
2375 wait_usec = USEC_INFINITY;
2376
2377 r = sd_event_run(m->event, wait_usec);
2378 if (r < 0)
2379 return log_error_errno(r, "Failed to run event loop: %m");
2380 }
2381
2382 return m->exit_code;
2383 }
2384
2385 int manager_load_unit_from_dbus_path(Manager *m, const char *s, sd_bus_error *e, Unit **_u) {
2386 _cleanup_free_ char *n = NULL;
2387 sd_id128_t invocation_id;
2388 Unit *u;
2389 int r;
2390
2391 assert(m);
2392 assert(s);
2393 assert(_u);
2394
2395 r = unit_name_from_dbus_path(s, &n);
2396 if (r < 0)
2397 return r;
2398
2399 /* Permit addressing units by invocation ID: if the passed bus path is suffixed by a 128bit ID then we use it
2400 * as invocation ID. */
2401 r = sd_id128_from_string(n, &invocation_id);
2402 if (r >= 0) {
2403 u = hashmap_get(m->units_by_invocation_id, &invocation_id);
2404 if (u) {
2405 *_u = u;
2406 return 0;
2407 }
2408
2409 return sd_bus_error_setf(e, BUS_ERROR_NO_UNIT_FOR_INVOCATION_ID, "No unit with the specified invocation ID " SD_ID128_FORMAT_STR " known.", SD_ID128_FORMAT_VAL(invocation_id));
2410 }
2411
2412 /* If this didn't work, we check if this is a unit name */
2413 if (!unit_name_is_valid(n, UNIT_NAME_PLAIN|UNIT_NAME_INSTANCE))
2414 return sd_bus_error_setf(e, SD_BUS_ERROR_INVALID_ARGS, "Unit name %s is neither a valid invocation ID nor unit name.", n);
2415
2416 r = manager_load_unit(m, n, NULL, e, &u);
2417 if (r < 0)
2418 return r;
2419
2420 *_u = u;
2421 return 0;
2422 }
2423
2424 int manager_get_job_from_dbus_path(Manager *m, const char *s, Job **_j) {
2425 const char *p;
2426 unsigned id;
2427 Job *j;
2428 int r;
2429
2430 assert(m);
2431 assert(s);
2432 assert(_j);
2433
2434 p = startswith(s, "/org/freedesktop/systemd1/job/");
2435 if (!p)
2436 return -EINVAL;
2437
2438 r = safe_atou(p, &id);
2439 if (r < 0)
2440 return r;
2441
2442 j = manager_get_job(m, id);
2443 if (!j)
2444 return -ENOENT;
2445
2446 *_j = j;
2447
2448 return 0;
2449 }
2450
2451 void manager_send_unit_audit(Manager *m, Unit *u, int type, bool success) {
2452
2453 #ifdef HAVE_AUDIT
2454 _cleanup_free_ char *p = NULL;
2455 const char *msg;
2456 int audit_fd, r;
2457
2458 if (!MANAGER_IS_SYSTEM(m))
2459 return;
2460
2461 audit_fd = get_audit_fd();
2462 if (audit_fd < 0)
2463 return;
2464
2465 /* Don't generate audit events if the service was already
2466 * started and we're just deserializing */
2467 if (MANAGER_IS_RELOADING(m))
2468 return;
2469
2470 if (u->type != UNIT_SERVICE)
2471 return;
2472
2473 r = unit_name_to_prefix_and_instance(u->id, &p);
2474 if (r < 0) {
2475 log_error_errno(r, "Failed to extract prefix and instance of unit name: %m");
2476 return;
2477 }
2478
2479 msg = strjoina("unit=", p);
2480 if (audit_log_user_comm_message(audit_fd, type, msg, "systemd", NULL, NULL, NULL, success) < 0) {
2481 if (errno == EPERM)
2482 /* We aren't allowed to send audit messages?
2483 * Then let's not retry again. */
2484 close_audit_fd();
2485 else
2486 log_warning_errno(errno, "Failed to send audit message: %m");
2487 }
2488 #endif
2489
2490 }
2491
2492 void manager_send_unit_plymouth(Manager *m, Unit *u) {
2493 static const union sockaddr_union sa = PLYMOUTH_SOCKET;
2494 _cleanup_free_ char *message = NULL;
2495 _cleanup_close_ int fd = -1;
2496 int n = 0;
2497
2498 /* Don't generate plymouth events if the service was already
2499 * started and we're just deserializing */
2500 if (MANAGER_IS_RELOADING(m))
2501 return;
2502
2503 if (!MANAGER_IS_SYSTEM(m))
2504 return;
2505
2506 if (detect_container() > 0)
2507 return;
2508
2509 if (u->type != UNIT_SERVICE &&
2510 u->type != UNIT_MOUNT &&
2511 u->type != UNIT_SWAP)
2512 return;
2513
2514 /* We set SOCK_NONBLOCK here so that we rather drop the
2515 * message then wait for plymouth */
2516 fd = socket(AF_UNIX, SOCK_STREAM|SOCK_CLOEXEC|SOCK_NONBLOCK, 0);
2517 if (fd < 0) {
2518 log_error_errno(errno, "socket() failed: %m");
2519 return;
2520 }
2521
2522 if (connect(fd, &sa.sa, SOCKADDR_UN_LEN(sa.un)) < 0) {
2523
2524 if (!IN_SET(errno, EPIPE, EAGAIN, ENOENT, ECONNREFUSED, ECONNRESET, ECONNABORTED))
2525 log_error_errno(errno, "connect() failed: %m");
2526 return;
2527 }
2528
2529 if (asprintf(&message, "U\002%c%s%n", (int) (strlen(u->id) + 1), u->id, &n) < 0) {
2530 log_oom();
2531 return;
2532 }
2533
2534 errno = 0;
2535 if (write(fd, message, n + 1) != n + 1)
2536 if (!IN_SET(errno, EPIPE, EAGAIN, ENOENT, ECONNREFUSED, ECONNRESET, ECONNABORTED))
2537 log_error_errno(errno, "Failed to write Plymouth message: %m");
2538 }
2539
2540 int manager_open_serialization(Manager *m, FILE **_f) {
2541 int fd;
2542 FILE *f;
2543
2544 assert(_f);
2545
2546 fd = open_serialization_fd("systemd-state");
2547 if (fd < 0)
2548 return fd;
2549
2550 f = fdopen(fd, "w+");
2551 if (!f) {
2552 safe_close(fd);
2553 return -errno;
2554 }
2555
2556 *_f = f;
2557 return 0;
2558 }
2559
2560 int manager_serialize(Manager *m, FILE *f, FDSet *fds, bool switching_root) {
2561 Iterator i;
2562 Unit *u;
2563 const char *t;
2564 int r;
2565
2566 assert(m);
2567 assert(f);
2568 assert(fds);
2569
2570 m->n_reloading++;
2571
2572 fprintf(f, "current-job-id=%"PRIu32"\n", m->current_job_id);
2573 fprintf(f, "taint-usr=%s\n", yes_no(m->taint_usr));
2574 fprintf(f, "n-installed-jobs=%u\n", m->n_installed_jobs);
2575 fprintf(f, "n-failed-jobs=%u\n", m->n_failed_jobs);
2576
2577 dual_timestamp_serialize(f, "firmware-timestamp", &m->firmware_timestamp);
2578 dual_timestamp_serialize(f, "loader-timestamp", &m->loader_timestamp);
2579 dual_timestamp_serialize(f, "kernel-timestamp", &m->kernel_timestamp);
2580 dual_timestamp_serialize(f, "initrd-timestamp", &m->initrd_timestamp);
2581
2582 if (!in_initrd()) {
2583 dual_timestamp_serialize(f, "userspace-timestamp", &m->userspace_timestamp);
2584 dual_timestamp_serialize(f, "finish-timestamp", &m->finish_timestamp);
2585 dual_timestamp_serialize(f, "security-start-timestamp", &m->security_start_timestamp);
2586 dual_timestamp_serialize(f, "security-finish-timestamp", &m->security_finish_timestamp);
2587 dual_timestamp_serialize(f, "generators-start-timestamp", &m->generators_start_timestamp);
2588 dual_timestamp_serialize(f, "generators-finish-timestamp", &m->generators_finish_timestamp);
2589 dual_timestamp_serialize(f, "units-load-start-timestamp", &m->units_load_start_timestamp);
2590 dual_timestamp_serialize(f, "units-load-finish-timestamp", &m->units_load_finish_timestamp);
2591 }
2592
2593 if (!switching_root)
2594 (void) serialize_environment(f, m->environment);
2595
2596 if (m->notify_fd >= 0) {
2597 int copy;
2598
2599 copy = fdset_put_dup(fds, m->notify_fd);
2600 if (copy < 0)
2601 return copy;
2602
2603 fprintf(f, "notify-fd=%i\n", copy);
2604 fprintf(f, "notify-socket=%s\n", m->notify_socket);
2605 }
2606
2607 if (m->cgroups_agent_fd >= 0) {
2608 int copy;
2609
2610 copy = fdset_put_dup(fds, m->cgroups_agent_fd);
2611 if (copy < 0)
2612 return copy;
2613
2614 fprintf(f, "cgroups-agent-fd=%i\n", copy);
2615 }
2616
2617 if (m->user_lookup_fds[0] >= 0) {
2618 int copy0, copy1;
2619
2620 copy0 = fdset_put_dup(fds, m->user_lookup_fds[0]);
2621 if (copy0 < 0)
2622 return copy0;
2623
2624 copy1 = fdset_put_dup(fds, m->user_lookup_fds[1]);
2625 if (copy1 < 0)
2626 return copy1;
2627
2628 fprintf(f, "user-lookup=%i %i\n", copy0, copy1);
2629 }
2630
2631 bus_track_serialize(m->subscribed, f, "subscribed");
2632
2633 r = dynamic_user_serialize(m, f, fds);
2634 if (r < 0)
2635 return r;
2636
2637 manager_serialize_uid_refs(m, f);
2638 manager_serialize_gid_refs(m, f);
2639
2640 fputc_unlocked('\n', f);
2641
2642 HASHMAP_FOREACH_KEY(u, t, m->units, i) {
2643 if (u->id != t)
2644 continue;
2645
2646 /* Start marker */
2647 fputs_unlocked(u->id, f);
2648 fputc_unlocked('\n', f);
2649
2650 r = unit_serialize(u, f, fds, !switching_root);
2651 if (r < 0) {
2652 m->n_reloading--;
2653 return r;
2654 }
2655 }
2656
2657 assert(m->n_reloading > 0);
2658 m->n_reloading--;
2659
2660 if (ferror(f))
2661 return -EIO;
2662
2663 r = bus_fdset_add_all(m, fds);
2664 if (r < 0)
2665 return r;
2666
2667 return 0;
2668 }
2669
2670 int manager_deserialize(Manager *m, FILE *f, FDSet *fds) {
2671 int r = 0;
2672
2673 assert(m);
2674 assert(f);
2675
2676 log_debug("Deserializing state...");
2677
2678 m->n_reloading++;
2679
2680 for (;;) {
2681 char line[LINE_MAX];
2682 const char *val, *l;
2683
2684 if (!fgets(line, sizeof(line), f)) {
2685 if (feof(f))
2686 r = 0;
2687 else
2688 r = -errno;
2689
2690 goto finish;
2691 }
2692
2693 char_array_0(line);
2694 l = strstrip(line);
2695
2696 if (l[0] == 0)
2697 break;
2698
2699 if ((val = startswith(l, "current-job-id="))) {
2700 uint32_t id;
2701
2702 if (safe_atou32(val, &id) < 0)
2703 log_notice("Failed to parse current job id value %s", val);
2704 else
2705 m->current_job_id = MAX(m->current_job_id, id);
2706
2707 } else if ((val = startswith(l, "n-installed-jobs="))) {
2708 uint32_t n;
2709
2710 if (safe_atou32(val, &n) < 0)
2711 log_notice("Failed to parse installed jobs counter %s", val);
2712 else
2713 m->n_installed_jobs += n;
2714
2715 } else if ((val = startswith(l, "n-failed-jobs="))) {
2716 uint32_t n;
2717
2718 if (safe_atou32(val, &n) < 0)
2719 log_notice("Failed to parse failed jobs counter %s", val);
2720 else
2721 m->n_failed_jobs += n;
2722
2723 } else if ((val = startswith(l, "taint-usr="))) {
2724 int b;
2725
2726 b = parse_boolean(val);
2727 if (b < 0)
2728 log_notice("Failed to parse taint /usr flag %s", val);
2729 else
2730 m->taint_usr = m->taint_usr || b;
2731
2732 } else if ((val = startswith(l, "firmware-timestamp=")))
2733 dual_timestamp_deserialize(val, &m->firmware_timestamp);
2734 else if ((val = startswith(l, "loader-timestamp=")))
2735 dual_timestamp_deserialize(val, &m->loader_timestamp);
2736 else if ((val = startswith(l, "kernel-timestamp=")))
2737 dual_timestamp_deserialize(val, &m->kernel_timestamp);
2738 else if ((val = startswith(l, "initrd-timestamp=")))
2739 dual_timestamp_deserialize(val, &m->initrd_timestamp);
2740 else if ((val = startswith(l, "userspace-timestamp=")))
2741 dual_timestamp_deserialize(val, &m->userspace_timestamp);
2742 else if ((val = startswith(l, "finish-timestamp=")))
2743 dual_timestamp_deserialize(val, &m->finish_timestamp);
2744 else if ((val = startswith(l, "security-start-timestamp=")))
2745 dual_timestamp_deserialize(val, &m->security_start_timestamp);
2746 else if ((val = startswith(l, "security-finish-timestamp=")))
2747 dual_timestamp_deserialize(val, &m->security_finish_timestamp);
2748 else if ((val = startswith(l, "generators-start-timestamp=")))
2749 dual_timestamp_deserialize(val, &m->generators_start_timestamp);
2750 else if ((val = startswith(l, "generators-finish-timestamp=")))
2751 dual_timestamp_deserialize(val, &m->generators_finish_timestamp);
2752 else if ((val = startswith(l, "units-load-start-timestamp=")))
2753 dual_timestamp_deserialize(val, &m->units_load_start_timestamp);
2754 else if ((val = startswith(l, "units-load-finish-timestamp=")))
2755 dual_timestamp_deserialize(val, &m->units_load_finish_timestamp);
2756 else if (startswith(l, "env=")) {
2757 r = deserialize_environment(&m->environment, l);
2758 if (r == -ENOMEM)
2759 goto finish;
2760 if (r < 0)
2761 log_notice_errno(r, "Failed to parse environment entry: \"%s\": %m", l);
2762
2763 } else if ((val = startswith(l, "notify-fd="))) {
2764 int fd;
2765
2766 if (safe_atoi(val, &fd) < 0 || fd < 0 || !fdset_contains(fds, fd))
2767 log_notice("Failed to parse notify fd: \"%s\"", val);
2768 else {
2769 m->notify_event_source = sd_event_source_unref(m->notify_event_source);
2770 safe_close(m->notify_fd);
2771 m->notify_fd = fdset_remove(fds, fd);
2772 }
2773
2774 } else if ((val = startswith(l, "notify-socket="))) {
2775 char *n;
2776
2777 n = strdup(val);
2778 if (!n) {
2779 r = -ENOMEM;
2780 goto finish;
2781 }
2782
2783 free(m->notify_socket);
2784 m->notify_socket = n;
2785
2786 } else if ((val = startswith(l, "cgroups-agent-fd="))) {
2787 int fd;
2788
2789 if (safe_atoi(val, &fd) < 0 || fd < 0 || !fdset_contains(fds, fd))
2790 log_notice("Failed to parse cgroups agent fd: %s", val);
2791 else {
2792 m->cgroups_agent_event_source = sd_event_source_unref(m->cgroups_agent_event_source);
2793 safe_close(m->cgroups_agent_fd);
2794 m->cgroups_agent_fd = fdset_remove(fds, fd);
2795 }
2796
2797 } else if ((val = startswith(l, "user-lookup="))) {
2798 int fd0, fd1;
2799
2800 if (sscanf(val, "%i %i", &fd0, &fd1) != 2 || fd0 < 0 || fd1 < 0 || fd0 == fd1 || !fdset_contains(fds, fd0) || !fdset_contains(fds, fd1))
2801 log_notice("Failed to parse user lookup fd: %s", val);
2802 else {
2803 m->user_lookup_event_source = sd_event_source_unref(m->user_lookup_event_source);
2804 safe_close_pair(m->user_lookup_fds);
2805 m->user_lookup_fds[0] = fdset_remove(fds, fd0);
2806 m->user_lookup_fds[1] = fdset_remove(fds, fd1);
2807 }
2808
2809 } else if ((val = startswith(l, "dynamic-user=")))
2810 dynamic_user_deserialize_one(m, val, fds);
2811 else if ((val = startswith(l, "destroy-ipc-uid=")))
2812 manager_deserialize_uid_refs_one(m, val);
2813 else if ((val = startswith(l, "destroy-ipc-gid=")))
2814 manager_deserialize_gid_refs_one(m, val);
2815 else if ((val = startswith(l, "subscribed="))) {
2816
2817 if (strv_extend(&m->deserialized_subscribed, val) < 0)
2818 log_oom();
2819
2820 } else if (!startswith(l, "kdbus-fd=")) /* ignore this one */
2821 log_notice("Unknown serialization item '%s'", l);
2822 }
2823
2824 for (;;) {
2825 Unit *u;
2826 char name[UNIT_NAME_MAX+2];
2827 const char* unit_name;
2828
2829 /* Start marker */
2830 if (!fgets(name, sizeof(name), f)) {
2831 if (feof(f))
2832 r = 0;
2833 else
2834 r = -errno;
2835
2836 goto finish;
2837 }
2838
2839 char_array_0(name);
2840 unit_name = strstrip(name);
2841
2842 r = manager_load_unit(m, unit_name, NULL, NULL, &u);
2843 if (r < 0) {
2844 log_notice_errno(r, "Failed to load unit \"%s\", skipping deserialization: %m", unit_name);
2845 if (r == -ENOMEM)
2846 goto finish;
2847 unit_deserialize_skip(f);
2848 continue;
2849 }
2850
2851 r = unit_deserialize(u, f, fds);
2852 if (r < 0) {
2853 log_notice_errno(r, "Failed to deserialize unit \"%s\": %m", unit_name);
2854 if (r == -ENOMEM)
2855 goto finish;
2856 }
2857 }
2858
2859 finish:
2860 if (ferror(f))
2861 r = -EIO;
2862
2863 assert(m->n_reloading > 0);
2864 m->n_reloading--;
2865
2866 return r;
2867 }
2868
2869 int manager_reload(Manager *m) {
2870 int r, q;
2871 _cleanup_fclose_ FILE *f = NULL;
2872 _cleanup_fdset_free_ FDSet *fds = NULL;
2873
2874 assert(m);
2875
2876 r = manager_open_serialization(m, &f);
2877 if (r < 0)
2878 return r;
2879
2880 m->n_reloading++;
2881 bus_manager_send_reloading(m, true);
2882
2883 fds = fdset_new();
2884 if (!fds) {
2885 m->n_reloading--;
2886 return -ENOMEM;
2887 }
2888
2889 r = manager_serialize(m, f, fds, false);
2890 if (r < 0) {
2891 m->n_reloading--;
2892 return r;
2893 }
2894
2895 if (fseeko(f, 0, SEEK_SET) < 0) {
2896 m->n_reloading--;
2897 return -errno;
2898 }
2899
2900 /* From here on there is no way back. */
2901 manager_clear_jobs_and_units(m);
2902 lookup_paths_flush_generator(&m->lookup_paths);
2903 lookup_paths_free(&m->lookup_paths);
2904 dynamic_user_vacuum(m, false);
2905 m->uid_refs = hashmap_free(m->uid_refs);
2906 m->gid_refs = hashmap_free(m->gid_refs);
2907
2908 q = lookup_paths_init(&m->lookup_paths, m->unit_file_scope, 0, NULL);
2909 if (q < 0 && r >= 0)
2910 r = q;
2911
2912 q = manager_run_environment_generators(m);
2913 if (q < 0 && r >= 0)
2914 r = q;
2915
2916 /* Find new unit paths */
2917 q = manager_run_generators(m);
2918 if (q < 0 && r >= 0)
2919 r = q;
2920
2921 lookup_paths_reduce(&m->lookup_paths);
2922 manager_build_unit_path_cache(m);
2923
2924 /* First, enumerate what we can from all config files */
2925 manager_enumerate(m);
2926
2927 /* Second, deserialize our stored data */
2928 q = manager_deserialize(m, f, fds);
2929 if (q < 0) {
2930 log_error_errno(q, "Deserialization failed: %m");
2931
2932 if (r >= 0)
2933 r = q;
2934 }
2935
2936 fclose(f);
2937 f = NULL;
2938
2939 /* Re-register notify_fd as event source */
2940 q = manager_setup_notify(m);
2941 if (q < 0 && r >= 0)
2942 r = q;
2943
2944 q = manager_setup_cgroups_agent(m);
2945 if (q < 0 && r >= 0)
2946 r = q;
2947
2948 q = manager_setup_user_lookup_fd(m);
2949 if (q < 0 && r >= 0)
2950 r = q;
2951
2952 /* Third, fire things up! */
2953 manager_coldplug(m);
2954
2955 /* Release any dynamic users no longer referenced */
2956 dynamic_user_vacuum(m, true);
2957
2958 /* Release any references to UIDs/GIDs no longer referenced, and destroy any IPC owned by them */
2959 manager_vacuum_uid_refs(m);
2960 manager_vacuum_gid_refs(m);
2961
2962 /* Sync current state of bus names with our set of listening units */
2963 if (m->api_bus)
2964 manager_sync_bus_names(m, m->api_bus);
2965
2966 assert(m->n_reloading > 0);
2967 m->n_reloading--;
2968
2969 m->send_reloading_done = true;
2970
2971 return r;
2972 }
2973
2974 void manager_reset_failed(Manager *m) {
2975 Unit *u;
2976 Iterator i;
2977
2978 assert(m);
2979
2980 HASHMAP_FOREACH(u, m->units, i)
2981 unit_reset_failed(u);
2982 }
2983
2984 bool manager_unit_inactive_or_pending(Manager *m, const char *name) {
2985 Unit *u;
2986
2987 assert(m);
2988 assert(name);
2989
2990 /* Returns true if the unit is inactive or going down */
2991 u = manager_get_unit(m, name);
2992 if (!u)
2993 return true;
2994
2995 return unit_inactive_or_pending(u);
2996 }
2997
2998 static void manager_notify_finished(Manager *m) {
2999 char userspace[FORMAT_TIMESPAN_MAX], initrd[FORMAT_TIMESPAN_MAX], kernel[FORMAT_TIMESPAN_MAX], sum[FORMAT_TIMESPAN_MAX];
3000 usec_t firmware_usec, loader_usec, kernel_usec, initrd_usec, userspace_usec, total_usec;
3001
3002 if (m->test_run)
3003 return;
3004
3005 if (MANAGER_IS_SYSTEM(m) && detect_container() <= 0) {
3006
3007 /* Note that m->kernel_usec.monotonic is always at 0,
3008 * and m->firmware_usec.monotonic and
3009 * m->loader_usec.monotonic should be considered
3010 * negative values. */
3011
3012 firmware_usec = m->firmware_timestamp.monotonic - m->loader_timestamp.monotonic;
3013 loader_usec = m->loader_timestamp.monotonic - m->kernel_timestamp.monotonic;
3014 userspace_usec = m->finish_timestamp.monotonic - m->userspace_timestamp.monotonic;
3015 total_usec = m->firmware_timestamp.monotonic + m->finish_timestamp.monotonic;
3016
3017 if (dual_timestamp_is_set(&m->initrd_timestamp)) {
3018
3019 kernel_usec = m->initrd_timestamp.monotonic - m->kernel_timestamp.monotonic;
3020 initrd_usec = m->userspace_timestamp.monotonic - m->initrd_timestamp.monotonic;
3021
3022 log_struct(LOG_INFO,
3023 "MESSAGE_ID=" SD_MESSAGE_STARTUP_FINISHED_STR,
3024 "KERNEL_USEC="USEC_FMT, kernel_usec,
3025 "INITRD_USEC="USEC_FMT, initrd_usec,
3026 "USERSPACE_USEC="USEC_FMT, userspace_usec,
3027 LOG_MESSAGE("Startup finished in %s (kernel) + %s (initrd) + %s (userspace) = %s.",
3028 format_timespan(kernel, sizeof(kernel), kernel_usec, USEC_PER_MSEC),
3029 format_timespan(initrd, sizeof(initrd), initrd_usec, USEC_PER_MSEC),
3030 format_timespan(userspace, sizeof(userspace), userspace_usec, USEC_PER_MSEC),
3031 format_timespan(sum, sizeof(sum), total_usec, USEC_PER_MSEC)),
3032 NULL);
3033 } else {
3034 kernel_usec = m->userspace_timestamp.monotonic - m->kernel_timestamp.monotonic;
3035 initrd_usec = 0;
3036
3037 log_struct(LOG_INFO,
3038 "MESSAGE_ID=" SD_MESSAGE_STARTUP_FINISHED_STR,
3039 "KERNEL_USEC="USEC_FMT, kernel_usec,
3040 "USERSPACE_USEC="USEC_FMT, userspace_usec,
3041 LOG_MESSAGE("Startup finished in %s (kernel) + %s (userspace) = %s.",
3042 format_timespan(kernel, sizeof(kernel), kernel_usec, USEC_PER_MSEC),
3043 format_timespan(userspace, sizeof(userspace), userspace_usec, USEC_PER_MSEC),
3044 format_timespan(sum, sizeof(sum), total_usec, USEC_PER_MSEC)),
3045 NULL);
3046 }
3047 } else {
3048 firmware_usec = loader_usec = initrd_usec = kernel_usec = 0;
3049 total_usec = userspace_usec = m->finish_timestamp.monotonic - m->userspace_timestamp.monotonic;
3050
3051 log_struct(LOG_INFO,
3052 "MESSAGE_ID=" SD_MESSAGE_USER_STARTUP_FINISHED_STR,
3053 "USERSPACE_USEC="USEC_FMT, userspace_usec,
3054 LOG_MESSAGE("Startup finished in %s.",
3055 format_timespan(sum, sizeof(sum), total_usec, USEC_PER_MSEC)),
3056 NULL);
3057 }
3058
3059 bus_manager_send_finished(m, firmware_usec, loader_usec, kernel_usec, initrd_usec, userspace_usec, total_usec);
3060
3061 sd_notifyf(false,
3062 "READY=1\n"
3063 "STATUS=Startup finished in %s.",
3064 format_timespan(sum, sizeof(sum), total_usec, USEC_PER_MSEC));
3065 }
3066
3067 void manager_check_finished(Manager *m) {
3068 assert(m);
3069
3070 if (MANAGER_IS_RELOADING(m))
3071 return;
3072
3073 /* Verify that we are actually running currently. Initially
3074 * the exit code is set to invalid, and during operation it is
3075 * then set to MANAGER_OK */
3076 if (m->exit_code != MANAGER_OK)
3077 return;
3078
3079 if (hashmap_size(m->jobs) > 0) {
3080 if (m->jobs_in_progress_event_source)
3081 /* Ignore any failure, this is only for feedback */
3082 (void) sd_event_source_set_time(m->jobs_in_progress_event_source, now(CLOCK_MONOTONIC) + JOBS_IN_PROGRESS_WAIT_USEC);
3083
3084 return;
3085 }
3086
3087 manager_flip_auto_status(m, false);
3088
3089 /* Notify Type=idle units that we are done now */
3090 manager_close_idle_pipe(m);
3091
3092 /* Turn off confirm spawn now */
3093 m->confirm_spawn = NULL;
3094
3095 /* No need to update ask password status when we're going non-interactive */
3096 manager_close_ask_password(m);
3097
3098 /* This is no longer the first boot */
3099 manager_set_first_boot(m, false);
3100
3101 if (dual_timestamp_is_set(&m->finish_timestamp))
3102 return;
3103
3104 dual_timestamp_get(&m->finish_timestamp);
3105
3106 manager_notify_finished(m);
3107
3108 manager_invalidate_startup_units(m);
3109 }
3110
3111 static bool generator_path_any(const char* const* paths) {
3112 char **path;
3113 bool found = false;
3114
3115 /* Optimize by skipping the whole process by not creating output directories
3116 * if no generators are found. */
3117 STRV_FOREACH(path, (char**) paths)
3118 if (access(*path, F_OK) == 0)
3119 found = true;
3120 else if (errno != ENOENT)
3121 log_warning_errno(errno, "Failed to open generator directory %s: %m", *path);
3122
3123 return found;
3124 }
3125
3126 static const char* system_env_generator_binary_paths[] = {
3127 "/run/systemd/system-environment-generators",
3128 "/etc/systemd/system-environment-generators",
3129 "/usr/local/lib/systemd/system-environment-generators",
3130 SYSTEM_ENV_GENERATOR_PATH,
3131 NULL
3132 };
3133
3134 static const char* user_env_generator_binary_paths[] = {
3135 "/run/systemd/user-environment-generators",
3136 "/etc/systemd/user-environment-generators",
3137 "/usr/local/lib/systemd/user-environment-generators",
3138 USER_ENV_GENERATOR_PATH,
3139 NULL
3140 };
3141
3142 static int manager_run_environment_generators(Manager *m) {
3143 char **tmp = NULL; /* this is only used in the forked process, no cleanup here */
3144 const char **paths;
3145 void* args[] = {&tmp, &tmp, &m->environment};
3146
3147 if (m->test_run)
3148 return 0;
3149
3150 paths = MANAGER_IS_SYSTEM(m) ? system_env_generator_binary_paths : user_env_generator_binary_paths;
3151
3152 if (!generator_path_any(paths))
3153 return 0;
3154
3155 return execute_directories(paths, DEFAULT_TIMEOUT_USEC, gather_environment, args, NULL);
3156 }
3157
3158 static int manager_run_generators(Manager *m) {
3159 _cleanup_strv_free_ char **paths = NULL;
3160 const char *argv[5];
3161 int r;
3162
3163 assert(m);
3164
3165 if (m->test_run)
3166 return 0;
3167
3168 paths = generator_binary_paths(m->unit_file_scope);
3169 if (!paths)
3170 return log_oom();
3171
3172 if (!generator_path_any((const char* const*) paths))
3173 return 0;
3174
3175 r = lookup_paths_mkdir_generator(&m->lookup_paths);
3176 if (r < 0)
3177 goto finish;
3178
3179 argv[0] = NULL; /* Leave this empty, execute_directory() will fill something in */
3180 argv[1] = m->lookup_paths.generator;
3181 argv[2] = m->lookup_paths.generator_early;
3182 argv[3] = m->lookup_paths.generator_late;
3183 argv[4] = NULL;
3184
3185 RUN_WITH_UMASK(0022)
3186 execute_directories((const char* const*) paths, DEFAULT_TIMEOUT_USEC,
3187 NULL, NULL, (char**) argv);
3188
3189 finish:
3190 lookup_paths_trim_generator(&m->lookup_paths);
3191 return r;
3192 }
3193
3194 int manager_environment_add(Manager *m, char **minus, char **plus) {
3195 char **a = NULL, **b = NULL, **l;
3196 assert(m);
3197
3198 l = m->environment;
3199
3200 if (!strv_isempty(minus)) {
3201 a = strv_env_delete(l, 1, minus);
3202 if (!a)
3203 return -ENOMEM;
3204
3205 l = a;
3206 }
3207
3208 if (!strv_isempty(plus)) {
3209 b = strv_env_merge(2, l, plus);
3210 if (!b) {
3211 strv_free(a);
3212 return -ENOMEM;
3213 }
3214
3215 l = b;
3216 }
3217
3218 if (m->environment != l)
3219 strv_free(m->environment);
3220 if (a != l)
3221 strv_free(a);
3222 if (b != l)
3223 strv_free(b);
3224
3225 m->environment = l;
3226 manager_clean_environment(m);
3227 strv_sort(m->environment);
3228
3229 return 0;
3230 }
3231
3232 int manager_set_default_rlimits(Manager *m, struct rlimit **default_rlimit) {
3233 int i;
3234
3235 assert(m);
3236
3237 for (i = 0; i < _RLIMIT_MAX; i++) {
3238 m->rlimit[i] = mfree(m->rlimit[i]);
3239
3240 if (!default_rlimit[i])
3241 continue;
3242
3243 m->rlimit[i] = newdup(struct rlimit, default_rlimit[i], 1);
3244 if (!m->rlimit[i])
3245 return log_oom();
3246 }
3247
3248 return 0;
3249 }
3250
3251 void manager_recheck_journal(Manager *m) {
3252 Unit *u;
3253
3254 assert(m);
3255
3256 if (!MANAGER_IS_SYSTEM(m))
3257 return;
3258
3259 u = manager_get_unit(m, SPECIAL_JOURNALD_SOCKET);
3260 if (u && SOCKET(u)->state != SOCKET_RUNNING) {
3261 log_close_journal();
3262 return;
3263 }
3264
3265 u = manager_get_unit(m, SPECIAL_JOURNALD_SERVICE);
3266 if (u && SERVICE(u)->state != SERVICE_RUNNING) {
3267 log_close_journal();
3268 return;
3269 }
3270
3271 /* Hmm, OK, so the socket is fully up and the service is up
3272 * too, then let's make use of the thing. */
3273 log_open();
3274 }
3275
3276 void manager_set_show_status(Manager *m, ShowStatus mode) {
3277 assert(m);
3278 assert(IN_SET(mode, SHOW_STATUS_AUTO, SHOW_STATUS_NO, SHOW_STATUS_YES, SHOW_STATUS_TEMPORARY));
3279
3280 if (!MANAGER_IS_SYSTEM(m))
3281 return;
3282
3283 if (m->show_status != mode)
3284 log_debug("%s showing of status.",
3285 mode == SHOW_STATUS_NO ? "Disabling" : "Enabling");
3286 m->show_status = mode;
3287
3288 if (mode > 0)
3289 (void) touch("/run/systemd/show-status");
3290 else
3291 (void) unlink("/run/systemd/show-status");
3292 }
3293
3294 static bool manager_get_show_status(Manager *m, StatusType type) {
3295 assert(m);
3296
3297 if (!MANAGER_IS_SYSTEM(m))
3298 return false;
3299
3300 if (m->no_console_output)
3301 return false;
3302
3303 if (!IN_SET(manager_state(m), MANAGER_INITIALIZING, MANAGER_STARTING, MANAGER_STOPPING))
3304 return false;
3305
3306 /* If we cannot find out the status properly, just proceed. */
3307 if (type != STATUS_TYPE_EMERGENCY && manager_check_ask_password(m) > 0)
3308 return false;
3309
3310 if (m->show_status > 0)
3311 return true;
3312
3313 return false;
3314 }
3315
3316 const char *manager_get_confirm_spawn(Manager *m) {
3317 static int last_errno = 0;
3318 const char *vc = m->confirm_spawn;
3319 struct stat st;
3320 int r;
3321
3322 /* Here's the deal: we want to test the validity of the console but don't want
3323 * PID1 to go through the whole console process which might block. But we also
3324 * want to warn the user only once if something is wrong with the console so we
3325 * cannot do the sanity checks after spawning our children. So here we simply do
3326 * really basic tests to hopefully trap common errors.
3327 *
3328 * If the console suddenly disappear at the time our children will really it
3329 * then they will simply fail to acquire it and a positive answer will be
3330 * assumed. New children will fallback to /dev/console though.
3331 *
3332 * Note: TTYs are devices that can come and go any time, and frequently aren't
3333 * available yet during early boot (consider a USB rs232 dongle...). If for any
3334 * reason the configured console is not ready, we fallback to the default
3335 * console. */
3336
3337 if (!vc || path_equal(vc, "/dev/console"))
3338 return vc;
3339
3340 r = stat(vc, &st);
3341 if (r < 0)
3342 goto fail;
3343
3344 if (!S_ISCHR(st.st_mode)) {
3345 errno = ENOTTY;
3346 goto fail;
3347 }
3348
3349 last_errno = 0;
3350 return vc;
3351 fail:
3352 if (last_errno != errno) {
3353 last_errno = errno;
3354 log_warning_errno(errno, "Failed to open %s: %m, using default console", vc);
3355 }
3356 return "/dev/console";
3357 }
3358
3359 void manager_set_first_boot(Manager *m, bool b) {
3360 assert(m);
3361
3362 if (!MANAGER_IS_SYSTEM(m))
3363 return;
3364
3365 if (m->first_boot != (int) b) {
3366 if (b)
3367 (void) touch("/run/systemd/first-boot");
3368 else
3369 (void) unlink("/run/systemd/first-boot");
3370 }
3371
3372 m->first_boot = b;
3373 }
3374
3375 void manager_disable_confirm_spawn(void) {
3376 (void) touch("/run/systemd/confirm_spawn_disabled");
3377 }
3378
3379 bool manager_is_confirm_spawn_disabled(Manager *m) {
3380 if (!m->confirm_spawn)
3381 return true;
3382
3383 return access("/run/systemd/confirm_spawn_disabled", F_OK) >= 0;
3384 }
3385
3386 void manager_status_printf(Manager *m, StatusType type, const char *status, const char *format, ...) {
3387 va_list ap;
3388
3389 /* If m is NULL, assume we're after shutdown and let the messages through. */
3390
3391 if (m && !manager_get_show_status(m, type))
3392 return;
3393
3394 /* XXX We should totally drop the check for ephemeral here
3395 * and thus effectively make 'Type=idle' pointless. */
3396 if (type == STATUS_TYPE_EPHEMERAL && m && m->n_on_console > 0)
3397 return;
3398
3399 va_start(ap, format);
3400 status_vprintf(status, true, type == STATUS_TYPE_EPHEMERAL, format, ap);
3401 va_end(ap);
3402 }
3403
3404 Set *manager_get_units_requiring_mounts_for(Manager *m, const char *path) {
3405 char p[strlen(path)+1];
3406
3407 assert(m);
3408 assert(path);
3409
3410 strcpy(p, path);
3411 path_kill_slashes(p);
3412
3413 return hashmap_get(m->units_requiring_mounts_for, streq(p, "/") ? "" : p);
3414 }
3415
3416 void manager_set_exec_params(Manager *m, ExecParameters *p) {
3417 assert(m);
3418 assert(p);
3419
3420 p->environment = m->environment;
3421 p->confirm_spawn = manager_get_confirm_spawn(m);
3422 p->cgroup_supported = m->cgroup_supported;
3423 p->prefix = m->prefix;
3424
3425 SET_FLAG(p->flags, EXEC_PASS_LOG_UNIT|EXEC_CHOWN_DIRECTORIES, MANAGER_IS_SYSTEM(m));
3426 }
3427
3428 int manager_update_failed_units(Manager *m, Unit *u, bool failed) {
3429 unsigned size;
3430 int r;
3431
3432 assert(m);
3433 assert(u->manager == m);
3434
3435 size = set_size(m->failed_units);
3436
3437 if (failed) {
3438 r = set_ensure_allocated(&m->failed_units, NULL);
3439 if (r < 0)
3440 return log_oom();
3441
3442 if (set_put(m->failed_units, u) < 0)
3443 return log_oom();
3444 } else
3445 (void) set_remove(m->failed_units, u);
3446
3447 if (set_size(m->failed_units) != size)
3448 bus_manager_send_change_signal(m);
3449
3450 return 0;
3451 }
3452
3453 ManagerState manager_state(Manager *m) {
3454 Unit *u;
3455
3456 assert(m);
3457
3458 /* Did we ever finish booting? If not then we are still starting up */
3459 if (!dual_timestamp_is_set(&m->finish_timestamp)) {
3460
3461 u = manager_get_unit(m, SPECIAL_BASIC_TARGET);
3462 if (!u || !UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u)))
3463 return MANAGER_INITIALIZING;
3464
3465 return MANAGER_STARTING;
3466 }
3467
3468 /* Is the special shutdown target queued? If so, we are in shutdown state */
3469 u = manager_get_unit(m, SPECIAL_SHUTDOWN_TARGET);
3470 if (u && u->job && IN_SET(u->job->type, JOB_START, JOB_RESTART, JOB_RELOAD_OR_START))
3471 return MANAGER_STOPPING;
3472
3473 /* Are the rescue or emergency targets active or queued? If so we are in maintenance state */
3474 u = manager_get_unit(m, SPECIAL_RESCUE_TARGET);
3475 if (u && (UNIT_IS_ACTIVE_OR_ACTIVATING(unit_active_state(u)) ||
3476 (u->job && IN_SET(u->job->type, JOB_START, JOB_RESTART, JOB_RELOAD_OR_START))))
3477 return MANAGER_MAINTENANCE;
3478
3479 u = manager_get_unit(m, SPECIAL_EMERGENCY_TARGET);
3480 if (u && (UNIT_IS_ACTIVE_OR_ACTIVATING(unit_active_state(u)) ||
3481 (u->job && IN_SET(u->job->type, JOB_START, JOB_RESTART, JOB_RELOAD_OR_START))))
3482 return MANAGER_MAINTENANCE;
3483
3484 /* Are there any failed units? If so, we are in degraded mode */
3485 if (set_size(m->failed_units) > 0)
3486 return MANAGER_DEGRADED;
3487
3488 return MANAGER_RUNNING;
3489 }
3490
3491 #define DESTROY_IPC_FLAG (UINT32_C(1) << 31)
3492
3493 static void manager_unref_uid_internal(
3494 Manager *m,
3495 Hashmap **uid_refs,
3496 uid_t uid,
3497 bool destroy_now,
3498 int (*_clean_ipc)(uid_t uid)) {
3499
3500 uint32_t c, n;
3501
3502 assert(m);
3503 assert(uid_refs);
3504 assert(uid_is_valid(uid));
3505 assert(_clean_ipc);
3506
3507 /* A generic implementation, covering both manager_unref_uid() and manager_unref_gid(), under the assumption
3508 * that uid_t and gid_t are actually defined the same way, with the same validity rules.
3509 *
3510 * We store a hashmap where the UID/GID is they key and the value is a 32bit reference counter, whose highest
3511 * bit is used as flag for marking UIDs/GIDs whose IPC objects to remove when the last reference to the UID/GID
3512 * is dropped. The flag is set to on, once at least one reference from a unit where RemoveIPC= is set is added
3513 * on a UID/GID. It is reset when the UID's/GID's reference counter drops to 0 again. */
3514
3515 assert_cc(sizeof(uid_t) == sizeof(gid_t));
3516 assert_cc(UID_INVALID == (uid_t) GID_INVALID);
3517
3518 if (uid == 0) /* We don't keep track of root, and will never destroy it */
3519 return;
3520
3521 c = PTR_TO_UINT32(hashmap_get(*uid_refs, UID_TO_PTR(uid)));
3522
3523 n = c & ~DESTROY_IPC_FLAG;
3524 assert(n > 0);
3525 n--;
3526
3527 if (destroy_now && n == 0) {
3528 hashmap_remove(*uid_refs, UID_TO_PTR(uid));
3529
3530 if (c & DESTROY_IPC_FLAG) {
3531 log_debug("%s " UID_FMT " is no longer referenced, cleaning up its IPC.",
3532 _clean_ipc == clean_ipc_by_uid ? "UID" : "GID",
3533 uid);
3534 (void) _clean_ipc(uid);
3535 }
3536 } else {
3537 c = n | (c & DESTROY_IPC_FLAG);
3538 assert_se(hashmap_update(*uid_refs, UID_TO_PTR(uid), UINT32_TO_PTR(c)) >= 0);
3539 }
3540 }
3541
3542 void manager_unref_uid(Manager *m, uid_t uid, bool destroy_now) {
3543 manager_unref_uid_internal(m, &m->uid_refs, uid, destroy_now, clean_ipc_by_uid);
3544 }
3545
3546 void manager_unref_gid(Manager *m, gid_t gid, bool destroy_now) {
3547 manager_unref_uid_internal(m, &m->gid_refs, (uid_t) gid, destroy_now, clean_ipc_by_gid);
3548 }
3549
3550 static int manager_ref_uid_internal(
3551 Manager *m,
3552 Hashmap **uid_refs,
3553 uid_t uid,
3554 bool clean_ipc) {
3555
3556 uint32_t c, n;
3557 int r;
3558
3559 assert(m);
3560 assert(uid_refs);
3561 assert(uid_is_valid(uid));
3562
3563 /* A generic implementation, covering both manager_ref_uid() and manager_ref_gid(), under the assumption
3564 * that uid_t and gid_t are actually defined the same way, with the same validity rules. */
3565
3566 assert_cc(sizeof(uid_t) == sizeof(gid_t));
3567 assert_cc(UID_INVALID == (uid_t) GID_INVALID);
3568
3569 if (uid == 0) /* We don't keep track of root, and will never destroy it */
3570 return 0;
3571
3572 r = hashmap_ensure_allocated(uid_refs, &trivial_hash_ops);
3573 if (r < 0)
3574 return r;
3575
3576 c = PTR_TO_UINT32(hashmap_get(*uid_refs, UID_TO_PTR(uid)));
3577
3578 n = c & ~DESTROY_IPC_FLAG;
3579 n++;
3580
3581 if (n & DESTROY_IPC_FLAG) /* check for overflow */
3582 return -EOVERFLOW;
3583
3584 c = n | (c & DESTROY_IPC_FLAG) | (clean_ipc ? DESTROY_IPC_FLAG : 0);
3585
3586 return hashmap_replace(*uid_refs, UID_TO_PTR(uid), UINT32_TO_PTR(c));
3587 }
3588
3589 int manager_ref_uid(Manager *m, uid_t uid, bool clean_ipc) {
3590 return manager_ref_uid_internal(m, &m->uid_refs, uid, clean_ipc);
3591 }
3592
3593 int manager_ref_gid(Manager *m, gid_t gid, bool clean_ipc) {
3594 return manager_ref_uid_internal(m, &m->gid_refs, (uid_t) gid, clean_ipc);
3595 }
3596
3597 static void manager_vacuum_uid_refs_internal(
3598 Manager *m,
3599 Hashmap **uid_refs,
3600 int (*_clean_ipc)(uid_t uid)) {
3601
3602 Iterator i;
3603 void *p, *k;
3604
3605 assert(m);
3606 assert(uid_refs);
3607 assert(_clean_ipc);
3608
3609 HASHMAP_FOREACH_KEY(p, k, *uid_refs, i) {
3610 uint32_t c, n;
3611 uid_t uid;
3612
3613 uid = PTR_TO_UID(k);
3614 c = PTR_TO_UINT32(p);
3615
3616 n = c & ~DESTROY_IPC_FLAG;
3617 if (n > 0)
3618 continue;
3619
3620 if (c & DESTROY_IPC_FLAG) {
3621 log_debug("Found unreferenced %s " UID_FMT " after reload/reexec. Cleaning up.",
3622 _clean_ipc == clean_ipc_by_uid ? "UID" : "GID",
3623 uid);
3624 (void) _clean_ipc(uid);
3625 }
3626
3627 assert_se(hashmap_remove(*uid_refs, k) == p);
3628 }
3629 }
3630
3631 void manager_vacuum_uid_refs(Manager *m) {
3632 manager_vacuum_uid_refs_internal(m, &m->uid_refs, clean_ipc_by_uid);
3633 }
3634
3635 void manager_vacuum_gid_refs(Manager *m) {
3636 manager_vacuum_uid_refs_internal(m, &m->gid_refs, clean_ipc_by_gid);
3637 }
3638
3639 static void manager_serialize_uid_refs_internal(
3640 Manager *m,
3641 FILE *f,
3642 Hashmap **uid_refs,
3643 const char *field_name) {
3644
3645 Iterator i;
3646 void *p, *k;
3647
3648 assert(m);
3649 assert(f);
3650 assert(uid_refs);
3651 assert(field_name);
3652
3653 /* Serialize the UID reference table. Or actually, just the IPC destruction flag of it, as the actual counter
3654 * of it is better rebuild after a reload/reexec. */
3655
3656 HASHMAP_FOREACH_KEY(p, k, *uid_refs, i) {
3657 uint32_t c;
3658 uid_t uid;
3659
3660 uid = PTR_TO_UID(k);
3661 c = PTR_TO_UINT32(p);
3662
3663 if (!(c & DESTROY_IPC_FLAG))
3664 continue;
3665
3666 fprintf(f, "%s=" UID_FMT "\n", field_name, uid);
3667 }
3668 }
3669
3670 void manager_serialize_uid_refs(Manager *m, FILE *f) {
3671 manager_serialize_uid_refs_internal(m, f, &m->uid_refs, "destroy-ipc-uid");
3672 }
3673
3674 void manager_serialize_gid_refs(Manager *m, FILE *f) {
3675 manager_serialize_uid_refs_internal(m, f, &m->gid_refs, "destroy-ipc-gid");
3676 }
3677
3678 static void manager_deserialize_uid_refs_one_internal(
3679 Manager *m,
3680 Hashmap** uid_refs,
3681 const char *value) {
3682
3683 uid_t uid;
3684 uint32_t c;
3685 int r;
3686
3687 assert(m);
3688 assert(uid_refs);
3689 assert(value);
3690
3691 r = parse_uid(value, &uid);
3692 if (r < 0 || uid == 0) {
3693 log_debug("Unable to parse UID reference serialization");
3694 return;
3695 }
3696
3697 r = hashmap_ensure_allocated(uid_refs, &trivial_hash_ops);
3698 if (r < 0) {
3699 log_oom();
3700 return;
3701 }
3702
3703 c = PTR_TO_UINT32(hashmap_get(*uid_refs, UID_TO_PTR(uid)));
3704 if (c & DESTROY_IPC_FLAG)
3705 return;
3706
3707 c |= DESTROY_IPC_FLAG;
3708
3709 r = hashmap_replace(*uid_refs, UID_TO_PTR(uid), UINT32_TO_PTR(c));
3710 if (r < 0) {
3711 log_debug("Failed to add UID reference entry");
3712 return;
3713 }
3714 }
3715
3716 void manager_deserialize_uid_refs_one(Manager *m, const char *value) {
3717 manager_deserialize_uid_refs_one_internal(m, &m->uid_refs, value);
3718 }
3719
3720 void manager_deserialize_gid_refs_one(Manager *m, const char *value) {
3721 manager_deserialize_uid_refs_one_internal(m, &m->gid_refs, value);
3722 }
3723
3724 int manager_dispatch_user_lookup_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
3725 struct buffer {
3726 uid_t uid;
3727 gid_t gid;
3728 char unit_name[UNIT_NAME_MAX+1];
3729 } _packed_ buffer;
3730
3731 Manager *m = userdata;
3732 ssize_t l;
3733 size_t n;
3734 Unit *u;
3735
3736 assert_se(source);
3737 assert_se(m);
3738
3739 /* Invoked whenever a child process succeeded resolving its user/group to use and sent us the resulting UID/GID
3740 * in a datagram. We parse the datagram here and pass it off to the unit, so that it can add a reference to the
3741 * UID/GID so that it can destroy the UID/GID's IPC objects when the reference counter drops to 0. */
3742
3743 l = recv(fd, &buffer, sizeof(buffer), MSG_DONTWAIT);
3744 if (l < 0) {
3745 if (errno == EINTR || errno == EAGAIN)
3746 return 0;
3747
3748 return log_error_errno(errno, "Failed to read from user lookup fd: %m");
3749 }
3750
3751 if ((size_t) l <= offsetof(struct buffer, unit_name)) {
3752 log_warning("Received too short user lookup message, ignoring.");
3753 return 0;
3754 }
3755
3756 if ((size_t) l > offsetof(struct buffer, unit_name) + UNIT_NAME_MAX) {
3757 log_warning("Received too long user lookup message, ignoring.");
3758 return 0;
3759 }
3760
3761 if (!uid_is_valid(buffer.uid) && !gid_is_valid(buffer.gid)) {
3762 log_warning("Got user lookup message with invalid UID/GID pair, ignoring.");
3763 return 0;
3764 }
3765
3766 n = (size_t) l - offsetof(struct buffer, unit_name);
3767 if (memchr(buffer.unit_name, 0, n)) {
3768 log_warning("Received lookup message with embedded NUL character, ignoring.");
3769 return 0;
3770 }
3771
3772 buffer.unit_name[n] = 0;
3773 u = manager_get_unit(m, buffer.unit_name);
3774 if (!u) {
3775 log_debug("Got user lookup message but unit doesn't exist, ignoring.");
3776 return 0;
3777 }
3778
3779 log_unit_debug(u, "User lookup succeeded: uid=" UID_FMT " gid=" GID_FMT, buffer.uid, buffer.gid);
3780
3781 unit_notify_user_lookup(u, buffer.uid, buffer.gid);
3782 return 0;
3783 }
3784
3785 static const char *const manager_state_table[_MANAGER_STATE_MAX] = {
3786 [MANAGER_INITIALIZING] = "initializing",
3787 [MANAGER_STARTING] = "starting",
3788 [MANAGER_RUNNING] = "running",
3789 [MANAGER_DEGRADED] = "degraded",
3790 [MANAGER_MAINTENANCE] = "maintenance",
3791 [MANAGER_STOPPING] = "stopping",
3792 };
3793
3794 DEFINE_STRING_TABLE_LOOKUP(manager_state, ManagerState);