]> git.ipfire.org Git - thirdparty/systemd.git/blob - src/core/manager.c
projects / thirdparty / systemd.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
build-sys: use #if Y instead of #ifdef Y everywhere
[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 #if 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_flags)
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_flags)
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_TYPE_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_TYPE_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_USER_STATE_CACHE, NULL },
579 [EXEC_DIRECTORY_LOGS] = { SD_PATH_USER_CONFIGURATION, "log" },
580 [EXEC_DIRECTORY_CONFIGURATION] = { SD_PATH_USER_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_TYPE_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, unsigned test_run_flags, 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 m->default_timeout_start_usec = DEFAULT_TIMEOUT_USEC;
620 m->default_timeout_stop_usec = DEFAULT_TIMEOUT_USEC;
621 m->default_restart_usec = DEFAULT_RESTART_USEC;
622
623 #if ENABLE_EFI
624 if (MANAGER_IS_SYSTEM(m) && detect_container() <= 0)
625 boot_timestamps(&m->userspace_timestamp, &m->firmware_timestamp, &m->loader_timestamp);
626 #endif
627
628 /* Prepare log fields we can use for structured logging */
629 if (MANAGER_IS_SYSTEM(m)) {
630 m->unit_log_field = "UNIT=";
631 m->unit_log_format_string = "UNIT=%s";
632
633 m->invocation_log_field = "INVOCATION_ID=";
634 m->invocation_log_format_string = "INVOCATION_ID=%s";
635 } else {
636 m->unit_log_field = "USER_UNIT=";
637 m->unit_log_format_string = "USER_UNIT=%s";
638
639 m->invocation_log_field = "USER_INVOCATION_ID=";
640 m->invocation_log_format_string = "USER_INVOCATION_ID=%s";
641 }
642
643 m->idle_pipe[0] = m->idle_pipe[1] = m->idle_pipe[2] = m->idle_pipe[3] = -1;
644
645 m->pin_cgroupfs_fd = m->notify_fd = m->cgroups_agent_fd = m->signal_fd = m->time_change_fd =
646 m->dev_autofs_fd = m->private_listen_fd = m->cgroup_inotify_fd =
647 m->ask_password_inotify_fd = -1;
648
649 m->user_lookup_fds[0] = m->user_lookup_fds[1] = -1;
650
651 m->current_job_id = 1; /* start as id #1, so that we can leave #0 around as "null-like" value */
652
653 m->have_ask_password = -EINVAL; /* we don't know */
654 m->first_boot = -1;
655
656 m->test_run_flags = test_run_flags;
657
658 /* Reboot immediately if the user hits C-A-D more often than 7x per 2s */
659 RATELIMIT_INIT(m->ctrl_alt_del_ratelimit, 2 * USEC_PER_SEC, 7);
660
661 r = manager_default_environment(m);
662 if (r < 0)
663 goto fail;
664
665 r = hashmap_ensure_allocated(&m->units, &string_hash_ops);
666 if (r < 0)
667 goto fail;
668
669 r = hashmap_ensure_allocated(&m->jobs, NULL);
670 if (r < 0)
671 goto fail;
672
673 r = hashmap_ensure_allocated(&m->cgroup_unit, &string_hash_ops);
674 if (r < 0)
675 goto fail;
676
677 r = hashmap_ensure_allocated(&m->watch_bus, &string_hash_ops);
678 if (r < 0)
679 goto fail;
680
681 r = sd_event_default(&m->event);
682 if (r < 0)
683 goto fail;
684
685 r = sd_event_add_defer(m->event, &m->run_queue_event_source, manager_dispatch_run_queue, m);
686 if (r < 0)
687 goto fail;
688
689 r = sd_event_source_set_priority(m->run_queue_event_source, SD_EVENT_PRIORITY_IDLE);
690 if (r < 0)
691 goto fail;
692
693 r = sd_event_source_set_enabled(m->run_queue_event_source, SD_EVENT_OFF);
694 if (r < 0)
695 goto fail;
696
697 (void) sd_event_source_set_description(m->run_queue_event_source, "manager-run-queue");
698
699 r = manager_setup_signals(m);
700 if (r < 0)
701 goto fail;
702
703 r = manager_setup_cgroup(m);
704 if (r < 0)
705 goto fail;
706
707 r = manager_setup_time_change(m);
708 if (r < 0)
709 goto fail;
710
711 m->udev = udev_new();
712 if (!m->udev) {
713 r = -ENOMEM;
714 goto fail;
715 }
716
717 /* Note that we do not set up the notify fd here. We do that after deserialization,
718 * since they might have gotten serialized across the reexec. */
719
720 m->taint_usr = dir_is_empty("/usr") > 0;
721
722 r = manager_setup_prefix(m);
723 if (r < 0)
724 goto fail;
725
726 *_m = m;
727 return 0;
728
729 fail:
730 manager_free(m);
731 return r;
732 }
733
734 static int manager_setup_notify(Manager *m) {
735 int r;
736
737 if (m->test_run_flags)
738 return 0;
739
740 if (m->notify_fd < 0) {
741 _cleanup_close_ int fd = -1;
742 union sockaddr_union sa = {
743 .sa.sa_family = AF_UNIX,
744 };
745 static const int one = 1;
746
747 /* First free all secondary fields */
748 m->notify_socket = mfree(m->notify_socket);
749 m->notify_event_source = sd_event_source_unref(m->notify_event_source);
750
751 fd = socket(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC|SOCK_NONBLOCK, 0);
752 if (fd < 0)
753 return log_error_errno(errno, "Failed to allocate notification socket: %m");
754
755 fd_inc_rcvbuf(fd, NOTIFY_RCVBUF_SIZE);
756
757 m->notify_socket = strappend(m->prefix[EXEC_DIRECTORY_RUNTIME], "/systemd/notify");
758 if (!m->notify_socket)
759 return log_oom();
760
761 (void) mkdir_parents_label(m->notify_socket, 0755);
762 (void) unlink(m->notify_socket);
763
764 strncpy(sa.un.sun_path, m->notify_socket, sizeof(sa.un.sun_path)-1);
765 r = bind(fd, &sa.sa, SOCKADDR_UN_LEN(sa.un));
766 if (r < 0)
767 return log_error_errno(errno, "bind(%s) failed: %m", sa.un.sun_path);
768
769 r = setsockopt(fd, SOL_SOCKET, SO_PASSCRED, &one, sizeof(one));
770 if (r < 0)
771 return log_error_errno(errno, "SO_PASSCRED failed: %m");
772
773 m->notify_fd = fd;
774 fd = -1;
775
776 log_debug("Using notification socket %s", m->notify_socket);
777 }
778
779 if (!m->notify_event_source) {
780 r = sd_event_add_io(m->event, &m->notify_event_source, m->notify_fd, EPOLLIN, manager_dispatch_notify_fd, m);
781 if (r < 0)
782 return log_error_errno(r, "Failed to allocate notify event source: %m");
783
784 /* Process notification messages a bit earlier than SIGCHLD, so that we can still identify to which
785 * service an exit message belongs. */
786 r = sd_event_source_set_priority(m->notify_event_source, SD_EVENT_PRIORITY_NORMAL-7);
787 if (r < 0)
788 return log_error_errno(r, "Failed to set priority of notify event source: %m");
789
790 (void) sd_event_source_set_description(m->notify_event_source, "manager-notify");
791 }
792
793 return 0;
794 }
795
796 static int manager_setup_cgroups_agent(Manager *m) {
797
798 static const union sockaddr_union sa = {
799 .un.sun_family = AF_UNIX,
800 .un.sun_path = "/run/systemd/cgroups-agent",
801 };
802 int r;
803
804 /* This creates a listening socket we receive cgroups agent messages on. We do not use D-Bus for delivering
805 * these messages from the cgroups agent binary to PID 1, as the cgroups agent binary is very short-living, and
806 * each instance of it needs a new D-Bus connection. Since D-Bus connections are SOCK_STREAM/AF_UNIX, on
807 * overloaded systems the backlog of the D-Bus socket becomes relevant, as not more than the configured number
808 * of D-Bus connections may be queued until the kernel will start dropping further incoming connections,
809 * possibly resulting in lost cgroups agent messages. To avoid this, we'll use a private SOCK_DGRAM/AF_UNIX
810 * socket, where no backlog is relevant as communication may take place without an actual connect() cycle, and
811 * we thus won't lose messages.
812 *
813 * Note that PID 1 will forward the agent message to system bus, so that the user systemd instance may listen
814 * to it. The system instance hence listens on this special socket, but the user instances listen on the system
815 * bus for these messages. */
816
817 if (m->test_run_flags)
818 return 0;
819
820 if (!MANAGER_IS_SYSTEM(m))
821 return 0;
822
823 r = cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER);
824 if (r < 0)
825 return log_error_errno(r, "Failed to determine whether unified cgroups hierarchy is used: %m");
826 if (r > 0) /* We don't need this anymore on the unified hierarchy */
827 return 0;
828
829 if (m->cgroups_agent_fd < 0) {
830 _cleanup_close_ int fd = -1;
831
832 /* First free all secondary fields */
833 m->cgroups_agent_event_source = sd_event_source_unref(m->cgroups_agent_event_source);
834
835 fd = socket(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC|SOCK_NONBLOCK, 0);
836 if (fd < 0)
837 return log_error_errno(errno, "Failed to allocate cgroups agent socket: %m");
838
839 fd_inc_rcvbuf(fd, CGROUPS_AGENT_RCVBUF_SIZE);
840
841 (void) unlink(sa.un.sun_path);
842
843 /* Only allow root to connect to this socket */
844 RUN_WITH_UMASK(0077)
845 r = bind(fd, &sa.sa, SOCKADDR_UN_LEN(sa.un));
846 if (r < 0)
847 return log_error_errno(errno, "bind(%s) failed: %m", sa.un.sun_path);
848
849 m->cgroups_agent_fd = fd;
850 fd = -1;
851 }
852
853 if (!m->cgroups_agent_event_source) {
854 r = sd_event_add_io(m->event, &m->cgroups_agent_event_source, m->cgroups_agent_fd, EPOLLIN, manager_dispatch_cgroups_agent_fd, m);
855 if (r < 0)
856 return log_error_errno(r, "Failed to allocate cgroups agent event source: %m");
857
858 /* Process cgroups notifications early, but after having processed service notification messages or
859 * SIGCHLD signals, so that a cgroup running empty is always just the last safety net of notification,
860 * and we collected the metadata the notification and SIGCHLD stuff offers first. Also see handling of
861 * cgroup inotify for the unified cgroup stuff. */
862 r = sd_event_source_set_priority(m->cgroups_agent_event_source, SD_EVENT_PRIORITY_NORMAL-4);
863 if (r < 0)
864 return log_error_errno(r, "Failed to set priority of cgroups agent event source: %m");
865
866 (void) sd_event_source_set_description(m->cgroups_agent_event_source, "manager-cgroups-agent");
867 }
868
869 return 0;
870 }
871
872 static int manager_setup_user_lookup_fd(Manager *m) {
873 int r;
874
875 assert(m);
876
877 /* Set up the socket pair used for passing UID/GID resolution results from forked off processes to PID
878 * 1. Background: we can't do name lookups (NSS) from PID 1, since it might involve IPC and thus activation,
879 * and we might hence deadlock on ourselves. Hence we do all user/group lookups asynchronously from the forked
880 * off processes right before executing the binaries to start. In order to be able to clean up any IPC objects
881 * created by a unit (see RemoveIPC=) we need to know in PID 1 the used UID/GID of the executed processes,
882 * hence we establish this communication channel so that forked off processes can pass their UID/GID
883 * information back to PID 1. The forked off processes send their resolved UID/GID to PID 1 in a simple
884 * datagram, along with their unit name, so that we can share one communication socket pair among all units for
885 * this purpose.
886 *
887 * You might wonder why we need a communication channel for this that is independent of the usual notification
888 * socket scheme (i.e. $NOTIFY_SOCKET). The primary difference is about trust: data sent via the $NOTIFY_SOCKET
889 * channel is only accepted if it originates from the right unit and if reception was enabled for it. The user
890 * lookup socket OTOH is only accessible by PID 1 and its children until they exec(), and always available.
891 *
892 * Note that this function is called under two circumstances: when we first initialize (in which case we
893 * allocate both the socket pair and the event source to listen on it), and when we deserialize after a reload
894 * (in which case the socket pair already exists but we still need to allocate the event source for it). */
895
896 if (m->user_lookup_fds[0] < 0) {
897
898 /* Free all secondary fields */
899 safe_close_pair(m->user_lookup_fds);
900 m->user_lookup_event_source = sd_event_source_unref(m->user_lookup_event_source);
901
902 if (socketpair(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC, 0, m->user_lookup_fds) < 0)
903 return log_error_errno(errno, "Failed to allocate user lookup socket: %m");
904
905 (void) fd_inc_rcvbuf(m->user_lookup_fds[0], NOTIFY_RCVBUF_SIZE);
906 }
907
908 if (!m->user_lookup_event_source) {
909 r = sd_event_add_io(m->event, &m->user_lookup_event_source, m->user_lookup_fds[0], EPOLLIN, manager_dispatch_user_lookup_fd, m);
910 if (r < 0)
911 return log_error_errno(errno, "Failed to allocate user lookup event source: %m");
912
913 /* Process even earlier than the notify event source, so that we always know first about valid UID/GID
914 * resolutions */
915 r = sd_event_source_set_priority(m->user_lookup_event_source, SD_EVENT_PRIORITY_NORMAL-8);
916 if (r < 0)
917 return log_error_errno(errno, "Failed to set priority ot user lookup event source: %m");
918
919 (void) sd_event_source_set_description(m->user_lookup_event_source, "user-lookup");
920 }
921
922 return 0;
923 }
924
925 static int manager_connect_bus(Manager *m, bool reexecuting) {
926 bool try_bus_connect;
927 Unit *u = NULL;
928
929 assert(m);
930
931 if (m->test_run_flags)
932 return 0;
933
934 u = manager_get_unit(m, SPECIAL_DBUS_SERVICE);
935
936 try_bus_connect =
937 (u && UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u))) &&
938 (reexecuting ||
939 (MANAGER_IS_USER(m) && getenv("DBUS_SESSION_BUS_ADDRESS")));
940
941 /* Try to connect to the buses, if possible. */
942 return bus_init(m, try_bus_connect);
943 }
944
945 static unsigned manager_dispatch_cleanup_queue(Manager *m) {
946 Unit *u;
947 unsigned n = 0;
948
949 assert(m);
950
951 while ((u = m->cleanup_queue)) {
952 assert(u->in_cleanup_queue);
953
954 unit_free(u);
955 n++;
956 }
957
958 return n;
959 }
960
961 enum {
962 GC_OFFSET_IN_PATH, /* This one is on the path we were traveling */
963 GC_OFFSET_UNSURE, /* No clue */
964 GC_OFFSET_GOOD, /* We still need this unit */
965 GC_OFFSET_BAD, /* We don't need this unit anymore */
966 _GC_OFFSET_MAX
967 };
968
969 static void unit_gc_mark_good(Unit *u, unsigned gc_marker) {
970 Iterator i;
971 Unit *other;
972
973 u->gc_marker = gc_marker + GC_OFFSET_GOOD;
974
975 /* Recursively mark referenced units as GOOD as well */
976 SET_FOREACH(other, u->dependencies[UNIT_REFERENCES], i)
977 if (other->gc_marker == gc_marker + GC_OFFSET_UNSURE)
978 unit_gc_mark_good(other, gc_marker);
979 }
980
981 static void unit_gc_sweep(Unit *u, unsigned gc_marker) {
982 Iterator i;
983 Unit *other;
984 bool is_bad;
985
986 assert(u);
987
988 if (u->gc_marker == gc_marker + GC_OFFSET_GOOD ||
989 u->gc_marker == gc_marker + GC_OFFSET_BAD ||
990 u->gc_marker == gc_marker + GC_OFFSET_UNSURE ||
991 u->gc_marker == gc_marker + GC_OFFSET_IN_PATH)
992 return;
993
994 if (u->in_cleanup_queue)
995 goto bad;
996
997 if (unit_check_gc(u))
998 goto good;
999
1000 u->gc_marker = gc_marker + GC_OFFSET_IN_PATH;
1001
1002 is_bad = true;
1003
1004 SET_FOREACH(other, u->dependencies[UNIT_REFERENCED_BY], i) {
1005 unit_gc_sweep(other, gc_marker);
1006
1007 if (other->gc_marker == gc_marker + GC_OFFSET_GOOD)
1008 goto good;
1009
1010 if (other->gc_marker != gc_marker + GC_OFFSET_BAD)
1011 is_bad = false;
1012 }
1013
1014 if (is_bad)
1015 goto bad;
1016
1017 /* We were unable to find anything out about this entry, so
1018 * let's investigate it later */
1019 u->gc_marker = gc_marker + GC_OFFSET_UNSURE;
1020 unit_add_to_gc_queue(u);
1021 return;
1022
1023 bad:
1024 /* We definitely know that this one is not useful anymore, so
1025 * let's mark it for deletion */
1026 u->gc_marker = gc_marker + GC_OFFSET_BAD;
1027 unit_add_to_cleanup_queue(u);
1028 return;
1029
1030 good:
1031 unit_gc_mark_good(u, gc_marker);
1032 }
1033
1034 static unsigned manager_dispatch_gc_unit_queue(Manager *m) {
1035 unsigned n = 0, gc_marker;
1036 Unit *u;
1037
1038 assert(m);
1039
1040 /* log_debug("Running GC..."); */
1041
1042 m->gc_marker += _GC_OFFSET_MAX;
1043 if (m->gc_marker + _GC_OFFSET_MAX <= _GC_OFFSET_MAX)
1044 m->gc_marker = 1;
1045
1046 gc_marker = m->gc_marker;
1047
1048 while ((u = m->gc_unit_queue)) {
1049 assert(u->in_gc_queue);
1050
1051 unit_gc_sweep(u, gc_marker);
1052
1053 LIST_REMOVE(gc_queue, m->gc_unit_queue, u);
1054 u->in_gc_queue = false;
1055
1056 n++;
1057
1058 if (u->gc_marker == gc_marker + GC_OFFSET_BAD ||
1059 u->gc_marker == gc_marker + GC_OFFSET_UNSURE) {
1060 if (u->id)
1061 log_unit_debug(u, "Collecting.");
1062 u->gc_marker = gc_marker + GC_OFFSET_BAD;
1063 unit_add_to_cleanup_queue(u);
1064 }
1065 }
1066
1067 return n;
1068 }
1069
1070 static unsigned manager_dispatch_gc_job_queue(Manager *m) {
1071 unsigned n = 0;
1072 Job *j;
1073
1074 assert(m);
1075
1076 while ((j = m->gc_job_queue)) {
1077 assert(j->in_gc_queue);
1078
1079 LIST_REMOVE(gc_queue, m->gc_job_queue, j);
1080 j->in_gc_queue = false;
1081
1082 n++;
1083
1084 if (job_check_gc(j))
1085 continue;
1086
1087 log_unit_debug(j->unit, "Collecting job.");
1088 (void) job_finish_and_invalidate(j, JOB_COLLECTED, false, false);
1089 }
1090
1091 return n;
1092 }
1093
1094 static void manager_clear_jobs_and_units(Manager *m) {
1095 Unit *u;
1096
1097 assert(m);
1098
1099 while ((u = hashmap_first(m->units)))
1100 unit_free(u);
1101
1102 manager_dispatch_cleanup_queue(m);
1103
1104 assert(!m->load_queue);
1105 assert(!m->run_queue);
1106 assert(!m->dbus_unit_queue);
1107 assert(!m->dbus_job_queue);
1108 assert(!m->cleanup_queue);
1109 assert(!m->gc_unit_queue);
1110 assert(!m->gc_job_queue);
1111
1112 assert(hashmap_isempty(m->jobs));
1113 assert(hashmap_isempty(m->units));
1114
1115 m->n_on_console = 0;
1116 m->n_running_jobs = 0;
1117 }
1118
1119 Manager* manager_free(Manager *m) {
1120 UnitType c;
1121 int i;
1122 ExecDirectoryType dt;
1123
1124 if (!m)
1125 return NULL;
1126
1127 manager_clear_jobs_and_units(m);
1128
1129 for (c = 0; c < _UNIT_TYPE_MAX; c++)
1130 if (unit_vtable[c]->shutdown)
1131 unit_vtable[c]->shutdown(m);
1132
1133 /* If we reexecute ourselves, we keep the root cgroup around */
1134 manager_shutdown_cgroup(m, m->exit_code != MANAGER_REEXECUTE);
1135
1136 lookup_paths_flush_generator(&m->lookup_paths);
1137
1138 bus_done(m);
1139
1140 dynamic_user_vacuum(m, false);
1141 hashmap_free(m->dynamic_users);
1142
1143 hashmap_free(m->units);
1144 hashmap_free(m->units_by_invocation_id);
1145 hashmap_free(m->jobs);
1146 hashmap_free(m->watch_pids1);
1147 hashmap_free(m->watch_pids2);
1148 hashmap_free(m->watch_bus);
1149
1150 set_free(m->startup_units);
1151 set_free(m->failed_units);
1152
1153 sd_event_source_unref(m->signal_event_source);
1154 sd_event_source_unref(m->notify_event_source);
1155 sd_event_source_unref(m->cgroups_agent_event_source);
1156 sd_event_source_unref(m->time_change_event_source);
1157 sd_event_source_unref(m->jobs_in_progress_event_source);
1158 sd_event_source_unref(m->run_queue_event_source);
1159 sd_event_source_unref(m->user_lookup_event_source);
1160
1161 safe_close(m->signal_fd);
1162 safe_close(m->notify_fd);
1163 safe_close(m->cgroups_agent_fd);
1164 safe_close(m->time_change_fd);
1165 safe_close_pair(m->user_lookup_fds);
1166
1167 manager_close_ask_password(m);
1168
1169 manager_close_idle_pipe(m);
1170
1171 udev_unref(m->udev);
1172 sd_event_unref(m->event);
1173
1174 free(m->notify_socket);
1175
1176 lookup_paths_free(&m->lookup_paths);
1177 strv_free(m->environment);
1178
1179 hashmap_free(m->cgroup_unit);
1180 set_free_free(m->unit_path_cache);
1181
1182 free(m->switch_root);
1183 free(m->switch_root_init);
1184
1185 for (i = 0; i < _RLIMIT_MAX; i++)
1186 m->rlimit[i] = mfree(m->rlimit[i]);
1187
1188 assert(hashmap_isempty(m->units_requiring_mounts_for));
1189 hashmap_free(m->units_requiring_mounts_for);
1190
1191 hashmap_free(m->uid_refs);
1192 hashmap_free(m->gid_refs);
1193
1194 for (dt = 0; dt < _EXEC_DIRECTORY_TYPE_MAX; dt++)
1195 m->prefix[dt] = mfree(m->prefix[dt]);
1196
1197 return mfree(m);
1198 }
1199
1200 void manager_enumerate(Manager *m) {
1201 UnitType c;
1202
1203 assert(m);
1204
1205 /* Let's ask every type to load all units from disk/kernel
1206 * that it might know */
1207 for (c = 0; c < _UNIT_TYPE_MAX; c++) {
1208 if (!unit_type_supported(c)) {
1209 log_debug("Unit type .%s is not supported on this system.", unit_type_to_string(c));
1210 continue;
1211 }
1212
1213 if (!unit_vtable[c]->enumerate)
1214 continue;
1215
1216 unit_vtable[c]->enumerate(m);
1217 }
1218
1219 manager_dispatch_load_queue(m);
1220 }
1221
1222 static void manager_coldplug(Manager *m) {
1223 Iterator i;
1224 Unit *u;
1225 char *k;
1226 int r;
1227
1228 assert(m);
1229
1230 /* Then, let's set up their initial state. */
1231 HASHMAP_FOREACH_KEY(u, k, m->units, i) {
1232
1233 /* ignore aliases */
1234 if (u->id != k)
1235 continue;
1236
1237 r = unit_coldplug(u);
1238 if (r < 0)
1239 log_warning_errno(r, "We couldn't coldplug %s, proceeding anyway: %m", u->id);
1240 }
1241 }
1242
1243 static void manager_build_unit_path_cache(Manager *m) {
1244 char **i;
1245 int r;
1246
1247 assert(m);
1248
1249 set_free_free(m->unit_path_cache);
1250
1251 m->unit_path_cache = set_new(&string_hash_ops);
1252 if (!m->unit_path_cache) {
1253 r = -ENOMEM;
1254 goto fail;
1255 }
1256
1257 /* This simply builds a list of files we know exist, so that
1258 * we don't always have to go to disk */
1259
1260 STRV_FOREACH(i, m->lookup_paths.search_path) {
1261 _cleanup_closedir_ DIR *d = NULL;
1262 struct dirent *de;
1263
1264 d = opendir(*i);
1265 if (!d) {
1266 if (errno != ENOENT)
1267 log_warning_errno(errno, "Failed to open directory %s, ignoring: %m", *i);
1268 continue;
1269 }
1270
1271 FOREACH_DIRENT(de, d, r = -errno; goto fail) {
1272 char *p;
1273
1274 p = strjoin(streq(*i, "/") ? "" : *i, "/", de->d_name);
1275 if (!p) {
1276 r = -ENOMEM;
1277 goto fail;
1278 }
1279
1280 r = set_consume(m->unit_path_cache, p);
1281 if (r < 0)
1282 goto fail;
1283 }
1284 }
1285
1286 return;
1287
1288 fail:
1289 log_warning_errno(r, "Failed to build unit path cache, proceeding without: %m");
1290 m->unit_path_cache = set_free_free(m->unit_path_cache);
1291 }
1292
1293 static void manager_distribute_fds(Manager *m, FDSet *fds) {
1294 Iterator i;
1295 Unit *u;
1296
1297 assert(m);
1298
1299 HASHMAP_FOREACH(u, m->units, i) {
1300
1301 if (fdset_size(fds) <= 0)
1302 break;
1303
1304 if (!UNIT_VTABLE(u)->distribute_fds)
1305 continue;
1306
1307 UNIT_VTABLE(u)->distribute_fds(u, fds);
1308 }
1309 }
1310
1311 int manager_startup(Manager *m, FILE *serialization, FDSet *fds) {
1312 int r, q;
1313
1314 assert(m);
1315
1316 /* If we are running in test mode, we still want to run the generators,
1317 * but we should not touch the real generator directories. */
1318 r = lookup_paths_init(&m->lookup_paths, m->unit_file_scope,
1319 m->test_run_flags ? LOOKUP_PATHS_TEMPORARY_GENERATED : 0,
1320 NULL);
1321 if (r < 0)
1322 return r;
1323
1324 r = manager_run_environment_generators(m);
1325 if (r < 0)
1326 return r;
1327
1328 /* Make sure the transient directory always exists, so that it remains
1329 * in the search path */
1330 r = mkdir_p_label(m->lookup_paths.transient, 0755);
1331 if (r < 0)
1332 return r;
1333
1334 dual_timestamp_get(&m->generators_start_timestamp);
1335 r = manager_run_generators(m);
1336 dual_timestamp_get(&m->generators_finish_timestamp);
1337 if (r < 0)
1338 return r;
1339
1340 if (m->first_boot > 0 &&
1341 m->unit_file_scope == UNIT_FILE_SYSTEM &&
1342 !m->test_run_flags) {
1343
1344 q = unit_file_preset_all(UNIT_FILE_SYSTEM, 0, NULL, UNIT_FILE_PRESET_ENABLE_ONLY, NULL, 0);
1345 if (q < 0)
1346 log_full_errno(q == -EEXIST ? LOG_NOTICE : LOG_WARNING, q, "Failed to populate /etc with preset unit settings, ignoring: %m");
1347 else
1348 log_info("Populated /etc with preset unit settings.");
1349 }
1350
1351 lookup_paths_reduce(&m->lookup_paths);
1352 manager_build_unit_path_cache(m);
1353
1354 /* If we will deserialize make sure that during enumeration
1355 * this is already known, so we increase the counter here
1356 * already */
1357 if (serialization)
1358 m->n_reloading++;
1359
1360 /* First, enumerate what we can from all config files */
1361 dual_timestamp_get(&m->units_load_start_timestamp);
1362 manager_enumerate(m);
1363 dual_timestamp_get(&m->units_load_finish_timestamp);
1364
1365 /* Second, deserialize if there is something to deserialize */
1366 if (serialization) {
1367 r = manager_deserialize(m, serialization, fds);
1368 if (r < 0)
1369 log_error_errno(r, "Deserialization failed: %m");
1370 }
1371
1372 /* Any fds left? Find some unit which wants them. This is
1373 * useful to allow container managers to pass some file
1374 * descriptors to us pre-initialized. This enables
1375 * socket-based activation of entire containers. */
1376 manager_distribute_fds(m, fds);
1377
1378 /* We might have deserialized the notify fd, but if we didn't
1379 * then let's create the bus now */
1380 q = manager_setup_notify(m);
1381 if (q < 0 && r == 0)
1382 r = q;
1383
1384 q = manager_setup_cgroups_agent(m);
1385 if (q < 0 && r == 0)
1386 r = q;
1387
1388 q = manager_setup_user_lookup_fd(m);
1389 if (q < 0 && r == 0)
1390 r = q;
1391
1392 /* Let's connect to the bus now. */
1393 (void) manager_connect_bus(m, !!serialization);
1394
1395 (void) bus_track_coldplug(m, &m->subscribed, false, m->deserialized_subscribed);
1396 m->deserialized_subscribed = strv_free(m->deserialized_subscribed);
1397
1398 /* Third, fire things up! */
1399 manager_coldplug(m);
1400
1401 /* Release any dynamic users no longer referenced */
1402 dynamic_user_vacuum(m, true);
1403
1404 /* Release any references to UIDs/GIDs no longer referenced, and destroy any IPC owned by them */
1405 manager_vacuum_uid_refs(m);
1406 manager_vacuum_gid_refs(m);
1407
1408 if (serialization) {
1409 assert(m->n_reloading > 0);
1410 m->n_reloading--;
1411
1412 /* Let's wait for the UnitNew/JobNew messages being
1413 * sent, before we notify that the reload is
1414 * finished */
1415 m->send_reloading_done = true;
1416 }
1417
1418 return r;
1419 }
1420
1421 int manager_add_job(Manager *m, JobType type, Unit *unit, JobMode mode, sd_bus_error *e, Job **_ret) {
1422 int r;
1423 Transaction *tr;
1424
1425 assert(m);
1426 assert(type < _JOB_TYPE_MAX);
1427 assert(unit);
1428 assert(mode < _JOB_MODE_MAX);
1429
1430 if (mode == JOB_ISOLATE && type != JOB_START)
1431 return sd_bus_error_setf(e, SD_BUS_ERROR_INVALID_ARGS, "Isolate is only valid for start.");
1432
1433 if (mode == JOB_ISOLATE && !unit->allow_isolate)
1434 return sd_bus_error_setf(e, BUS_ERROR_NO_ISOLATION, "Operation refused, unit may not be isolated.");
1435
1436 log_unit_debug(unit, "Trying to enqueue job %s/%s/%s", unit->id, job_type_to_string(type), job_mode_to_string(mode));
1437
1438 type = job_type_collapse(type, unit);
1439
1440 tr = transaction_new(mode == JOB_REPLACE_IRREVERSIBLY);
1441 if (!tr)
1442 return -ENOMEM;
1443
1444 r = transaction_add_job_and_dependencies(tr, type, unit, NULL, true, false,
1445 IN_SET(mode, JOB_IGNORE_DEPENDENCIES, JOB_IGNORE_REQUIREMENTS),
1446 mode == JOB_IGNORE_DEPENDENCIES, e);
1447 if (r < 0)
1448 goto tr_abort;
1449
1450 if (mode == JOB_ISOLATE) {
1451 r = transaction_add_isolate_jobs(tr, m);
1452 if (r < 0)
1453 goto tr_abort;
1454 }
1455
1456 r = transaction_activate(tr, m, mode, e);
1457 if (r < 0)
1458 goto tr_abort;
1459
1460 log_unit_debug(unit,
1461 "Enqueued job %s/%s as %u", unit->id,
1462 job_type_to_string(type), (unsigned) tr->anchor_job->id);
1463
1464 if (_ret)
1465 *_ret = tr->anchor_job;
1466
1467 transaction_free(tr);
1468 return 0;
1469
1470 tr_abort:
1471 transaction_abort(tr);
1472 transaction_free(tr);
1473 return r;
1474 }
1475
1476 int manager_add_job_by_name(Manager *m, JobType type, const char *name, JobMode mode, sd_bus_error *e, Job **ret) {
1477 Unit *unit = NULL; /* just to appease gcc, initialization is not really necessary */
1478 int r;
1479
1480 assert(m);
1481 assert(type < _JOB_TYPE_MAX);
1482 assert(name);
1483 assert(mode < _JOB_MODE_MAX);
1484
1485 r = manager_load_unit(m, name, NULL, NULL, &unit);
1486 if (r < 0)
1487 return r;
1488 assert(unit);
1489
1490 return manager_add_job(m, type, unit, mode, e, ret);
1491 }
1492
1493 int manager_add_job_by_name_and_warn(Manager *m, JobType type, const char *name, JobMode mode, Job **ret) {
1494 _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL;
1495 int r;
1496
1497 assert(m);
1498 assert(type < _JOB_TYPE_MAX);
1499 assert(name);
1500 assert(mode < _JOB_MODE_MAX);
1501
1502 r = manager_add_job_by_name(m, type, name, mode, &error, ret);
1503 if (r < 0)
1504 return log_warning_errno(r, "Failed to enqueue %s job for %s: %s", job_mode_to_string(mode), name, bus_error_message(&error, r));
1505
1506 return r;
1507 }
1508
1509 int manager_propagate_reload(Manager *m, Unit *unit, JobMode mode, sd_bus_error *e) {
1510 int r;
1511 Transaction *tr;
1512
1513 assert(m);
1514 assert(unit);
1515 assert(mode < _JOB_MODE_MAX);
1516 assert(mode != JOB_ISOLATE); /* Isolate is only valid for start */
1517
1518 tr = transaction_new(mode == JOB_REPLACE_IRREVERSIBLY);
1519 if (!tr)
1520 return -ENOMEM;
1521
1522 /* We need an anchor job */
1523 r = transaction_add_job_and_dependencies(tr, JOB_NOP, unit, NULL, false, false, true, true, e);
1524 if (r < 0)
1525 goto tr_abort;
1526
1527 /* Failure in adding individual dependencies is ignored, so this always succeeds. */
1528 transaction_add_propagate_reload_jobs(tr, unit, tr->anchor_job, mode == JOB_IGNORE_DEPENDENCIES, e);
1529
1530 r = transaction_activate(tr, m, mode, e);
1531 if (r < 0)
1532 goto tr_abort;
1533
1534 transaction_free(tr);
1535 return 0;
1536
1537 tr_abort:
1538 transaction_abort(tr);
1539 transaction_free(tr);
1540 return r;
1541 }
1542
1543 Job *manager_get_job(Manager *m, uint32_t id) {
1544 assert(m);
1545
1546 return hashmap_get(m->jobs, UINT32_TO_PTR(id));
1547 }
1548
1549 Unit *manager_get_unit(Manager *m, const char *name) {
1550 assert(m);
1551 assert(name);
1552
1553 return hashmap_get(m->units, name);
1554 }
1555
1556 unsigned manager_dispatch_load_queue(Manager *m) {
1557 Unit *u;
1558 unsigned n = 0;
1559
1560 assert(m);
1561
1562 /* Make sure we are not run recursively */
1563 if (m->dispatching_load_queue)
1564 return 0;
1565
1566 m->dispatching_load_queue = true;
1567
1568 /* Dispatches the load queue. Takes a unit from the queue and
1569 * tries to load its data until the queue is empty */
1570
1571 while ((u = m->load_queue)) {
1572 assert(u->in_load_queue);
1573
1574 unit_load(u);
1575 n++;
1576 }
1577
1578 m->dispatching_load_queue = false;
1579 return n;
1580 }
1581
1582 int manager_load_unit_prepare(
1583 Manager *m,
1584 const char *name,
1585 const char *path,
1586 sd_bus_error *e,
1587 Unit **_ret) {
1588
1589 Unit *ret;
1590 UnitType t;
1591 int r;
1592
1593 assert(m);
1594 assert(name || path);
1595 assert(_ret);
1596
1597 /* This will prepare the unit for loading, but not actually
1598 * load anything from disk. */
1599
1600 if (path && !is_path(path))
1601 return sd_bus_error_setf(e, SD_BUS_ERROR_INVALID_ARGS, "Path %s is not absolute.", path);
1602
1603 if (!name)
1604 name = basename(path);
1605
1606 t = unit_name_to_type(name);
1607
1608 if (t == _UNIT_TYPE_INVALID || !unit_name_is_valid(name, UNIT_NAME_PLAIN|UNIT_NAME_INSTANCE)) {
1609 if (unit_name_is_valid(name, UNIT_NAME_TEMPLATE))
1610 return sd_bus_error_setf(e, SD_BUS_ERROR_INVALID_ARGS, "Unit name %s is missing the instance name.", name);
1611
1612 return sd_bus_error_setf(e, SD_BUS_ERROR_INVALID_ARGS, "Unit name %s is not valid.", name);
1613 }
1614
1615 ret = manager_get_unit(m, name);
1616 if (ret) {
1617 *_ret = ret;
1618 return 1;
1619 }
1620
1621 ret = unit_new(m, unit_vtable[t]->object_size);
1622 if (!ret)
1623 return -ENOMEM;
1624
1625 if (path) {
1626 ret->fragment_path = strdup(path);
1627 if (!ret->fragment_path) {
1628 unit_free(ret);
1629 return -ENOMEM;
1630 }
1631 }
1632
1633 r = unit_add_name(ret, name);
1634 if (r < 0) {
1635 unit_free(ret);
1636 return r;
1637 }
1638
1639 unit_add_to_load_queue(ret);
1640 unit_add_to_dbus_queue(ret);
1641 unit_add_to_gc_queue(ret);
1642
1643 *_ret = ret;
1644
1645 return 0;
1646 }
1647
1648 int manager_load_unit(
1649 Manager *m,
1650 const char *name,
1651 const char *path,
1652 sd_bus_error *e,
1653 Unit **_ret) {
1654
1655 int r;
1656
1657 assert(m);
1658 assert(_ret);
1659
1660 /* This will load the service information files, but not actually
1661 * start any services or anything. */
1662
1663 r = manager_load_unit_prepare(m, name, path, e, _ret);
1664 if (r != 0)
1665 return r;
1666
1667 manager_dispatch_load_queue(m);
1668
1669 *_ret = unit_follow_merge(*_ret);
1670
1671 return 0;
1672 }
1673
1674 void manager_dump_jobs(Manager *s, FILE *f, const char *prefix) {
1675 Iterator i;
1676 Job *j;
1677
1678 assert(s);
1679 assert(f);
1680
1681 HASHMAP_FOREACH(j, s->jobs, i)
1682 job_dump(j, f, prefix);
1683 }
1684
1685 void manager_dump_units(Manager *s, FILE *f, const char *prefix) {
1686 Iterator i;
1687 Unit *u;
1688 const char *t;
1689
1690 assert(s);
1691 assert(f);
1692
1693 HASHMAP_FOREACH_KEY(u, t, s->units, i)
1694 if (u->id == t)
1695 unit_dump(u, f, prefix);
1696 }
1697
1698 void manager_clear_jobs(Manager *m) {
1699 Job *j;
1700
1701 assert(m);
1702
1703 while ((j = hashmap_first(m->jobs)))
1704 /* No need to recurse. We're cancelling all jobs. */
1705 job_finish_and_invalidate(j, JOB_CANCELED, false, false);
1706 }
1707
1708 static int manager_dispatch_run_queue(sd_event_source *source, void *userdata) {
1709 Manager *m = userdata;
1710 Job *j;
1711
1712 assert(source);
1713 assert(m);
1714
1715 while ((j = m->run_queue)) {
1716 assert(j->installed);
1717 assert(j->in_run_queue);
1718
1719 job_run_and_invalidate(j);
1720 }
1721
1722 if (m->n_running_jobs > 0)
1723 manager_watch_jobs_in_progress(m);
1724
1725 if (m->n_on_console > 0)
1726 manager_watch_idle_pipe(m);
1727
1728 return 1;
1729 }
1730
1731 static unsigned manager_dispatch_dbus_queue(Manager *m) {
1732 Job *j;
1733 Unit *u;
1734 unsigned n = 0;
1735
1736 assert(m);
1737
1738 if (m->dispatching_dbus_queue)
1739 return 0;
1740
1741 m->dispatching_dbus_queue = true;
1742
1743 while ((u = m->dbus_unit_queue)) {
1744 assert(u->in_dbus_queue);
1745
1746 bus_unit_send_change_signal(u);
1747 n++;
1748 }
1749
1750 while ((j = m->dbus_job_queue)) {
1751 assert(j->in_dbus_queue);
1752
1753 bus_job_send_change_signal(j);
1754 n++;
1755 }
1756
1757 m->dispatching_dbus_queue = false;
1758
1759 if (m->send_reloading_done) {
1760 m->send_reloading_done = false;
1761
1762 bus_manager_send_reloading(m, false);
1763 }
1764
1765 if (m->queued_message)
1766 bus_send_queued_message(m);
1767
1768 return n;
1769 }
1770
1771 static int manager_dispatch_cgroups_agent_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
1772 Manager *m = userdata;
1773 char buf[PATH_MAX+1];
1774 ssize_t n;
1775
1776 n = recv(fd, buf, sizeof(buf), 0);
1777 if (n < 0)
1778 return log_error_errno(errno, "Failed to read cgroups agent message: %m");
1779 if (n == 0) {
1780 log_error("Got zero-length cgroups agent message, ignoring.");
1781 return 0;
1782 }
1783 if ((size_t) n >= sizeof(buf)) {
1784 log_error("Got overly long cgroups agent message, ignoring.");
1785 return 0;
1786 }
1787
1788 if (memchr(buf, 0, n)) {
1789 log_error("Got cgroups agent message with embedded NUL byte, ignoring.");
1790 return 0;
1791 }
1792 buf[n] = 0;
1793
1794 manager_notify_cgroup_empty(m, buf);
1795 (void) bus_forward_agent_released(m, buf);
1796
1797 return 0;
1798 }
1799
1800 static void manager_invoke_notify_message(Manager *m, Unit *u, pid_t pid, const char *buf, FDSet *fds) {
1801 _cleanup_strv_free_ char **tags = NULL;
1802
1803 assert(m);
1804 assert(u);
1805 assert(buf);
1806
1807 tags = strv_split(buf, "\n\r");
1808 if (!tags) {
1809 log_oom();
1810 return;
1811 }
1812
1813 if (UNIT_VTABLE(u)->notify_message)
1814 UNIT_VTABLE(u)->notify_message(u, pid, tags, fds);
1815 else if (_unlikely_(log_get_max_level() >= LOG_DEBUG)) {
1816 _cleanup_free_ char *x = NULL, *y = NULL;
1817
1818 x = cescape(buf);
1819 if (x)
1820 y = ellipsize(x, 20, 90);
1821 log_unit_debug(u, "Got notification message \"%s\", ignoring.", strnull(y));
1822 }
1823 }
1824
1825 static int manager_dispatch_notify_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
1826
1827 _cleanup_fdset_free_ FDSet *fds = NULL;
1828 Manager *m = userdata;
1829 char buf[NOTIFY_BUFFER_MAX+1];
1830 struct iovec iovec = {
1831 .iov_base = buf,
1832 .iov_len = sizeof(buf)-1,
1833 };
1834 union {
1835 struct cmsghdr cmsghdr;
1836 uint8_t buf[CMSG_SPACE(sizeof(struct ucred)) +
1837 CMSG_SPACE(sizeof(int) * NOTIFY_FD_MAX)];
1838 } control = {};
1839 struct msghdr msghdr = {
1840 .msg_iov = &iovec,
1841 .msg_iovlen = 1,
1842 .msg_control = &control,
1843 .msg_controllen = sizeof(control),
1844 };
1845
1846 struct cmsghdr *cmsg;
1847 struct ucred *ucred = NULL;
1848 Unit *u1, *u2, *u3;
1849 int r, *fd_array = NULL;
1850 unsigned n_fds = 0;
1851 ssize_t n;
1852
1853 assert(m);
1854 assert(m->notify_fd == fd);
1855
1856 if (revents != EPOLLIN) {
1857 log_warning("Got unexpected poll event for notify fd.");
1858 return 0;
1859 }
1860
1861 n = recvmsg(m->notify_fd, &msghdr, MSG_DONTWAIT|MSG_CMSG_CLOEXEC|MSG_TRUNC);
1862 if (n < 0) {
1863 if (IN_SET(errno, EAGAIN, EINTR))
1864 return 0; /* Spurious wakeup, try again */
1865
1866 /* If this is any other, real error, then let's stop processing this socket. This of course means we
1867 * won't take notification messages anymore, but that's still better than busy looping around this:
1868 * being woken up over and over again but being unable to actually read the message off the socket. */
1869 return log_error_errno(errno, "Failed to receive notification message: %m");
1870 }
1871
1872 CMSG_FOREACH(cmsg, &msghdr) {
1873 if (cmsg->cmsg_level == SOL_SOCKET && cmsg->cmsg_type == SCM_RIGHTS) {
1874
1875 fd_array = (int*) CMSG_DATA(cmsg);
1876 n_fds = (cmsg->cmsg_len - CMSG_LEN(0)) / sizeof(int);
1877
1878 } else if (cmsg->cmsg_level == SOL_SOCKET &&
1879 cmsg->cmsg_type == SCM_CREDENTIALS &&
1880 cmsg->cmsg_len == CMSG_LEN(sizeof(struct ucred))) {
1881
1882 ucred = (struct ucred*) CMSG_DATA(cmsg);
1883 }
1884 }
1885
1886 if (n_fds > 0) {
1887 assert(fd_array);
1888
1889 r = fdset_new_array(&fds, fd_array, n_fds);
1890 if (r < 0) {
1891 close_many(fd_array, n_fds);
1892 log_oom();
1893 return 0;
1894 }
1895 }
1896
1897 if (!ucred || ucred->pid <= 0) {
1898 log_warning("Received notify message without valid credentials. Ignoring.");
1899 return 0;
1900 }
1901
1902 if ((size_t) n >= sizeof(buf) || (msghdr.msg_flags & MSG_TRUNC)) {
1903 log_warning("Received notify message exceeded maximum size. Ignoring.");
1904 return 0;
1905 }
1906
1907 /* As extra safety check, let's make sure the string we get doesn't contain embedded NUL bytes. We permit one
1908 * trailing NUL byte in the message, but don't expect it. */
1909 if (n > 1 && memchr(buf, 0, n-1)) {
1910 log_warning("Received notify message with embedded NUL bytes. Ignoring.");
1911 return 0;
1912 }
1913
1914 /* Make sure it's NUL-terminated. */
1915 buf[n] = 0;
1916
1917 /* Notify every unit that might be interested, but try
1918 * to avoid notifying the same one multiple times. */
1919 u1 = manager_get_unit_by_pid_cgroup(m, ucred->pid);
1920 if (u1)
1921 manager_invoke_notify_message(m, u1, ucred->pid, buf, fds);
1922
1923 u2 = hashmap_get(m->watch_pids1, PID_TO_PTR(ucred->pid));
1924 if (u2 && u2 != u1)
1925 manager_invoke_notify_message(m, u2, ucred->pid, buf, fds);
1926
1927 u3 = hashmap_get(m->watch_pids2, PID_TO_PTR(ucred->pid));
1928 if (u3 && u3 != u2 && u3 != u1)
1929 manager_invoke_notify_message(m, u3, ucred->pid, buf, fds);
1930
1931 if (!u1 && !u2 && !u3)
1932 log_warning("Cannot find unit for notify message of PID "PID_FMT".", ucred->pid);
1933
1934 if (fdset_size(fds) > 0)
1935 log_warning("Got extra auxiliary fds with notification message, closing them.");
1936
1937 return 0;
1938 }
1939
1940 static void invoke_sigchld_event(Manager *m, Unit *u, const siginfo_t *si) {
1941 uint64_t iteration;
1942
1943 assert(m);
1944 assert(u);
1945 assert(si);
1946
1947 sd_event_get_iteration(m->event, &iteration);
1948
1949 log_unit_debug(u, "Child "PID_FMT" belongs to %s", si->si_pid, u->id);
1950
1951 unit_unwatch_pid(u, si->si_pid);
1952
1953 if (UNIT_VTABLE(u)->sigchld_event) {
1954 if (set_size(u->pids) <= 1 ||
1955 iteration != u->sigchldgen ||
1956 unit_main_pid(u) == si->si_pid ||
1957 unit_control_pid(u) == si->si_pid) {
1958 UNIT_VTABLE(u)->sigchld_event(u, si->si_pid, si->si_code, si->si_status);
1959 u->sigchldgen = iteration;
1960 } else
1961 log_debug("%s already issued a sigchld this iteration %" PRIu64 ", skipping. Pids still being watched %d", u->id, iteration, set_size(u->pids));
1962 }
1963 }
1964
1965 static int manager_dispatch_sigchld(Manager *m) {
1966 assert(m);
1967
1968 for (;;) {
1969 siginfo_t si = {};
1970
1971 /* First we call waitd() for a PID and do not reap the
1972 * zombie. That way we can still access /proc/$PID for
1973 * it while it is a zombie. */
1974 if (waitid(P_ALL, 0, &si, WEXITED|WNOHANG|WNOWAIT) < 0) {
1975
1976 if (errno == ECHILD)
1977 break;
1978
1979 if (errno == EINTR)
1980 continue;
1981
1982 return -errno;
1983 }
1984
1985 if (si.si_pid <= 0)
1986 break;
1987
1988 if (IN_SET(si.si_code, CLD_EXITED, CLD_KILLED, CLD_DUMPED)) {
1989 _cleanup_free_ char *name = NULL;
1990 Unit *u1, *u2, *u3;
1991
1992 get_process_comm(si.si_pid, &name);
1993
1994 log_debug("Child "PID_FMT" (%s) died (code=%s, status=%i/%s)",
1995 si.si_pid, strna(name),
1996 sigchld_code_to_string(si.si_code),
1997 si.si_status,
1998 strna(si.si_code == CLD_EXITED
1999 ? exit_status_to_string(si.si_status, EXIT_STATUS_FULL)
2000 : signal_to_string(si.si_status)));
2001
2002 /* And now figure out the unit this belongs
2003 * to, it might be multiple... */
2004 u1 = manager_get_unit_by_pid_cgroup(m, si.si_pid);
2005 if (u1)
2006 invoke_sigchld_event(m, u1, &si);
2007 u2 = hashmap_get(m->watch_pids1, PID_TO_PTR(si.si_pid));
2008 if (u2 && u2 != u1)
2009 invoke_sigchld_event(m, u2, &si);
2010 u3 = hashmap_get(m->watch_pids2, PID_TO_PTR(si.si_pid));
2011 if (u3 && u3 != u2 && u3 != u1)
2012 invoke_sigchld_event(m, u3, &si);
2013 }
2014
2015 /* And now, we actually reap the zombie. */
2016 if (waitid(P_PID, si.si_pid, &si, WEXITED) < 0) {
2017 if (errno == EINTR)
2018 continue;
2019
2020 return -errno;
2021 }
2022 }
2023
2024 return 0;
2025 }
2026
2027 static void manager_start_target(Manager *m, const char *name, JobMode mode) {
2028 _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL;
2029 int r;
2030
2031 log_debug("Activating special unit %s", name);
2032
2033 r = manager_add_job_by_name(m, JOB_START, name, mode, &error, NULL);
2034 if (r < 0)
2035 log_error("Failed to enqueue %s job: %s", name, bus_error_message(&error, r));
2036 }
2037
2038 static void manager_handle_ctrl_alt_del(Manager *m) {
2039 /* If the user presses C-A-D more than
2040 * 7 times within 2s, we reboot/shutdown immediately,
2041 * unless it was disabled in system.conf */
2042
2043 if (ratelimit_test(&m->ctrl_alt_del_ratelimit) || m->cad_burst_action == EMERGENCY_ACTION_NONE)
2044 manager_start_target(m, SPECIAL_CTRL_ALT_DEL_TARGET, JOB_REPLACE_IRREVERSIBLY);
2045 else
2046 emergency_action(m, m->cad_burst_action, NULL,
2047 "Ctrl-Alt-Del was pressed more than 7 times within 2s");
2048 }
2049
2050 static int manager_dispatch_signal_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
2051 Manager *m = userdata;
2052 ssize_t n;
2053 struct signalfd_siginfo sfsi;
2054 bool sigchld = false;
2055 int r;
2056
2057 assert(m);
2058 assert(m->signal_fd == fd);
2059
2060 if (revents != EPOLLIN) {
2061 log_warning("Got unexpected events from signal file descriptor.");
2062 return 0;
2063 }
2064
2065 for (;;) {
2066 n = read(m->signal_fd, &sfsi, sizeof(sfsi));
2067 if (n != sizeof(sfsi)) {
2068 if (n >= 0) {
2069 log_warning("Truncated read from signal fd (%zu bytes)!", n);
2070 return 0;
2071 }
2072
2073 if (IN_SET(errno, EINTR, EAGAIN))
2074 break;
2075
2076 /* We return an error here, which will kill this handler,
2077 * to avoid a busy loop on read error. */
2078 return log_error_errno(errno, "Reading from signal fd failed: %m");
2079 }
2080
2081 log_received_signal(sfsi.ssi_signo == SIGCHLD ||
2082 (sfsi.ssi_signo == SIGTERM && MANAGER_IS_USER(m))
2083 ? LOG_DEBUG : LOG_INFO,
2084 &sfsi);
2085
2086 switch (sfsi.ssi_signo) {
2087
2088 case SIGCHLD:
2089 sigchld = true;
2090 break;
2091
2092 case SIGTERM:
2093 if (MANAGER_IS_SYSTEM(m)) {
2094 /* This is for compatibility with the
2095 * original sysvinit */
2096 r = verify_run_space_and_log("Refusing to reexecute");
2097 if (r >= 0)
2098 m->exit_code = MANAGER_REEXECUTE;
2099 break;
2100 }
2101
2102 /* Fall through */
2103
2104 case SIGINT:
2105 if (MANAGER_IS_SYSTEM(m))
2106 manager_handle_ctrl_alt_del(m);
2107 else
2108 manager_start_target(m, SPECIAL_EXIT_TARGET,
2109 JOB_REPLACE_IRREVERSIBLY);
2110 break;
2111
2112 case SIGWINCH:
2113 if (MANAGER_IS_SYSTEM(m))
2114 manager_start_target(m, SPECIAL_KBREQUEST_TARGET, JOB_REPLACE);
2115
2116 /* This is a nop on non-init */
2117 break;
2118
2119 case SIGPWR:
2120 if (MANAGER_IS_SYSTEM(m))
2121 manager_start_target(m, SPECIAL_SIGPWR_TARGET, JOB_REPLACE);
2122
2123 /* This is a nop on non-init */
2124 break;
2125
2126 case SIGUSR1: {
2127 Unit *u;
2128
2129 u = manager_get_unit(m, SPECIAL_DBUS_SERVICE);
2130
2131 if (!u || UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u))) {
2132 log_info("Trying to reconnect to bus...");
2133 bus_init(m, true);
2134 }
2135
2136 if (!u || !UNIT_IS_ACTIVE_OR_ACTIVATING(unit_active_state(u))) {
2137 log_info("Loading D-Bus service...");
2138 manager_start_target(m, SPECIAL_DBUS_SERVICE, JOB_REPLACE);
2139 }
2140
2141 break;
2142 }
2143
2144 case SIGUSR2: {
2145 _cleanup_free_ char *dump = NULL;
2146 _cleanup_fclose_ FILE *f = NULL;
2147 size_t size;
2148
2149 f = open_memstream(&dump, &size);
2150 if (!f) {
2151 log_warning_errno(errno, "Failed to allocate memory stream: %m");
2152 break;
2153 }
2154
2155 manager_dump_units(m, f, "\t");
2156 manager_dump_jobs(m, f, "\t");
2157
2158 r = fflush_and_check(f);
2159 if (r < 0) {
2160 log_warning_errno(r, "Failed to write status stream: %m");
2161 break;
2162 }
2163
2164 log_dump(LOG_INFO, dump);
2165 break;
2166 }
2167
2168 case SIGHUP:
2169 r = verify_run_space_and_log("Refusing to reload");
2170 if (r >= 0)
2171 m->exit_code = MANAGER_RELOAD;
2172 break;
2173
2174 default: {
2175
2176 /* Starting SIGRTMIN+0 */
2177 static const struct {
2178 const char *target;
2179 JobMode mode;
2180 } target_table[] = {
2181 [0] = { SPECIAL_DEFAULT_TARGET, JOB_ISOLATE },
2182 [1] = { SPECIAL_RESCUE_TARGET, JOB_ISOLATE },
2183 [2] = { SPECIAL_EMERGENCY_TARGET, JOB_ISOLATE },
2184 [3] = { SPECIAL_HALT_TARGET, JOB_REPLACE_IRREVERSIBLY },
2185 [4] = { SPECIAL_POWEROFF_TARGET, JOB_REPLACE_IRREVERSIBLY },
2186 [5] = { SPECIAL_REBOOT_TARGET, JOB_REPLACE_IRREVERSIBLY },
2187 [6] = { SPECIAL_KEXEC_TARGET, JOB_REPLACE_IRREVERSIBLY }
2188 };
2189
2190 /* Starting SIGRTMIN+13, so that target halt and system halt are 10 apart */
2191 static const ManagerExitCode code_table[] = {
2192 [0] = MANAGER_HALT,
2193 [1] = MANAGER_POWEROFF,
2194 [2] = MANAGER_REBOOT,
2195 [3] = MANAGER_KEXEC
2196 };
2197
2198 if ((int) sfsi.ssi_signo >= SIGRTMIN+0 &&
2199 (int) sfsi.ssi_signo < SIGRTMIN+(int) ELEMENTSOF(target_table)) {
2200 int idx = (int) sfsi.ssi_signo - SIGRTMIN;
2201 manager_start_target(m, target_table[idx].target,
2202 target_table[idx].mode);
2203 break;
2204 }
2205
2206 if ((int) sfsi.ssi_signo >= SIGRTMIN+13 &&
2207 (int) sfsi.ssi_signo < SIGRTMIN+13+(int) ELEMENTSOF(code_table)) {
2208 m->exit_code = code_table[sfsi.ssi_signo - SIGRTMIN - 13];
2209 break;
2210 }
2211
2212 switch (sfsi.ssi_signo - SIGRTMIN) {
2213
2214 case 20:
2215 manager_set_show_status(m, SHOW_STATUS_YES);
2216 break;
2217
2218 case 21:
2219 manager_set_show_status(m, SHOW_STATUS_NO);
2220 break;
2221
2222 case 22:
2223 log_set_max_level(LOG_DEBUG);
2224 log_info("Setting log level to debug.");
2225 break;
2226
2227 case 23:
2228 log_set_max_level(LOG_INFO);
2229 log_info("Setting log level to info.");
2230 break;
2231
2232 case 24:
2233 if (MANAGER_IS_USER(m)) {
2234 m->exit_code = MANAGER_EXIT;
2235 return 0;
2236 }
2237
2238 /* This is a nop on init */
2239 break;
2240
2241 case 26:
2242 case 29: /* compatibility: used to be mapped to LOG_TARGET_SYSLOG_OR_KMSG */
2243 log_set_target(LOG_TARGET_JOURNAL_OR_KMSG);
2244 log_notice("Setting log target to journal-or-kmsg.");
2245 break;
2246
2247 case 27:
2248 log_set_target(LOG_TARGET_CONSOLE);
2249 log_notice("Setting log target to console.");
2250 break;
2251
2252 case 28:
2253 log_set_target(LOG_TARGET_KMSG);
2254 log_notice("Setting log target to kmsg.");
2255 break;
2256
2257 default:
2258 log_warning("Got unhandled signal <%s>.", signal_to_string(sfsi.ssi_signo));
2259 }
2260 }
2261 }
2262 }
2263
2264 if (sigchld)
2265 manager_dispatch_sigchld(m);
2266
2267 return 0;
2268 }
2269
2270 static int manager_dispatch_time_change_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
2271 Manager *m = userdata;
2272 Iterator i;
2273 Unit *u;
2274
2275 assert(m);
2276 assert(m->time_change_fd == fd);
2277
2278 log_struct(LOG_DEBUG,
2279 "MESSAGE_ID=" SD_MESSAGE_TIME_CHANGE_STR,
2280 LOG_MESSAGE("Time has been changed"),
2281 NULL);
2282
2283 /* Restart the watch */
2284 m->time_change_event_source = sd_event_source_unref(m->time_change_event_source);
2285 m->time_change_fd = safe_close(m->time_change_fd);
2286
2287 manager_setup_time_change(m);
2288
2289 HASHMAP_FOREACH(u, m->units, i)
2290 if (UNIT_VTABLE(u)->time_change)
2291 UNIT_VTABLE(u)->time_change(u);
2292
2293 return 0;
2294 }
2295
2296 static int manager_dispatch_idle_pipe_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
2297 Manager *m = userdata;
2298
2299 assert(m);
2300 assert(m->idle_pipe[2] == fd);
2301
2302 m->no_console_output = m->n_on_console > 0;
2303
2304 manager_close_idle_pipe(m);
2305
2306 return 0;
2307 }
2308
2309 static int manager_dispatch_jobs_in_progress(sd_event_source *source, usec_t usec, void *userdata) {
2310 Manager *m = userdata;
2311 int r;
2312 uint64_t next;
2313
2314 assert(m);
2315 assert(source);
2316
2317 manager_print_jobs_in_progress(m);
2318
2319 next = now(CLOCK_MONOTONIC) + JOBS_IN_PROGRESS_PERIOD_USEC;
2320 r = sd_event_source_set_time(source, next);
2321 if (r < 0)
2322 return r;
2323
2324 return sd_event_source_set_enabled(source, SD_EVENT_ONESHOT);
2325 }
2326
2327 int manager_loop(Manager *m) {
2328 int r;
2329
2330 RATELIMIT_DEFINE(rl, 1*USEC_PER_SEC, 50000);
2331
2332 assert(m);
2333 m->exit_code = MANAGER_OK;
2334
2335 /* Release the path cache */
2336 m->unit_path_cache = set_free_free(m->unit_path_cache);
2337
2338 manager_check_finished(m);
2339
2340 /* There might still be some zombies hanging around from
2341 * before we were exec()'ed. Let's reap them. */
2342 r = manager_dispatch_sigchld(m);
2343 if (r < 0)
2344 return r;
2345
2346 while (m->exit_code == MANAGER_OK) {
2347 usec_t wait_usec;
2348
2349 if (m->runtime_watchdog > 0 && m->runtime_watchdog != USEC_INFINITY && MANAGER_IS_SYSTEM(m))
2350 watchdog_ping();
2351
2352 if (!ratelimit_test(&rl)) {
2353 /* Yay, something is going seriously wrong, pause a little */
2354 log_warning("Looping too fast. Throttling execution a little.");
2355 sleep(1);
2356 }
2357
2358 if (manager_dispatch_load_queue(m) > 0)
2359 continue;
2360
2361 if (manager_dispatch_gc_job_queue(m) > 0)
2362 continue;
2363
2364 if (manager_dispatch_gc_unit_queue(m) > 0)
2365 continue;
2366
2367 if (manager_dispatch_cleanup_queue(m) > 0)
2368 continue;
2369
2370 if (manager_dispatch_cgroup_realize_queue(m) > 0)
2371 continue;
2372
2373 if (manager_dispatch_dbus_queue(m) > 0)
2374 continue;
2375
2376 /* Sleep for half the watchdog time */
2377 if (m->runtime_watchdog > 0 && m->runtime_watchdog != USEC_INFINITY && MANAGER_IS_SYSTEM(m)) {
2378 wait_usec = m->runtime_watchdog / 2;
2379 if (wait_usec <= 0)
2380 wait_usec = 1;
2381 } else
2382 wait_usec = USEC_INFINITY;
2383
2384 r = sd_event_run(m->event, wait_usec);
2385 if (r < 0)
2386 return log_error_errno(r, "Failed to run event loop: %m");
2387 }
2388
2389 return m->exit_code;
2390 }
2391
2392 int manager_load_unit_from_dbus_path(Manager *m, const char *s, sd_bus_error *e, Unit **_u) {
2393 _cleanup_free_ char *n = NULL;
2394 sd_id128_t invocation_id;
2395 Unit *u;
2396 int r;
2397
2398 assert(m);
2399 assert(s);
2400 assert(_u);
2401
2402 r = unit_name_from_dbus_path(s, &n);
2403 if (r < 0)
2404 return r;
2405
2406 /* Permit addressing units by invocation ID: if the passed bus path is suffixed by a 128bit ID then we use it
2407 * as invocation ID. */
2408 r = sd_id128_from_string(n, &invocation_id);
2409 if (r >= 0) {
2410 u = hashmap_get(m->units_by_invocation_id, &invocation_id);
2411 if (u) {
2412 *_u = u;
2413 return 0;
2414 }
2415
2416 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));
2417 }
2418
2419 /* If this didn't work, we check if this is a unit name */
2420 if (!unit_name_is_valid(n, UNIT_NAME_PLAIN|UNIT_NAME_INSTANCE))
2421 return sd_bus_error_setf(e, SD_BUS_ERROR_INVALID_ARGS, "Unit name %s is neither a valid invocation ID nor unit name.", n);
2422
2423 r = manager_load_unit(m, n, NULL, e, &u);
2424 if (r < 0)
2425 return r;
2426
2427 *_u = u;
2428 return 0;
2429 }
2430
2431 int manager_get_job_from_dbus_path(Manager *m, const char *s, Job **_j) {
2432 const char *p;
2433 unsigned id;
2434 Job *j;
2435 int r;
2436
2437 assert(m);
2438 assert(s);
2439 assert(_j);
2440
2441 p = startswith(s, "/org/freedesktop/systemd1/job/");
2442 if (!p)
2443 return -EINVAL;
2444
2445 r = safe_atou(p, &id);
2446 if (r < 0)
2447 return r;
2448
2449 j = manager_get_job(m, id);
2450 if (!j)
2451 return -ENOENT;
2452
2453 *_j = j;
2454
2455 return 0;
2456 }
2457
2458 void manager_send_unit_audit(Manager *m, Unit *u, int type, bool success) {
2459
2460 #if HAVE_AUDIT
2461 _cleanup_free_ char *p = NULL;
2462 const char *msg;
2463 int audit_fd, r;
2464
2465 if (!MANAGER_IS_SYSTEM(m))
2466 return;
2467
2468 audit_fd = get_audit_fd();
2469 if (audit_fd < 0)
2470 return;
2471
2472 /* Don't generate audit events if the service was already
2473 * started and we're just deserializing */
2474 if (MANAGER_IS_RELOADING(m))
2475 return;
2476
2477 if (u->type != UNIT_SERVICE)
2478 return;
2479
2480 r = unit_name_to_prefix_and_instance(u->id, &p);
2481 if (r < 0) {
2482 log_error_errno(r, "Failed to extract prefix and instance of unit name: %m");
2483 return;
2484 }
2485
2486 msg = strjoina("unit=", p);
2487 if (audit_log_user_comm_message(audit_fd, type, msg, "systemd", NULL, NULL, NULL, success) < 0) {
2488 if (errno == EPERM)
2489 /* We aren't allowed to send audit messages?
2490 * Then let's not retry again. */
2491 close_audit_fd();
2492 else
2493 log_warning_errno(errno, "Failed to send audit message: %m");
2494 }
2495 #endif
2496
2497 }
2498
2499 void manager_send_unit_plymouth(Manager *m, Unit *u) {
2500 static const union sockaddr_union sa = PLYMOUTH_SOCKET;
2501 _cleanup_free_ char *message = NULL;
2502 _cleanup_close_ int fd = -1;
2503 int n = 0;
2504
2505 /* Don't generate plymouth events if the service was already
2506 * started and we're just deserializing */
2507 if (MANAGER_IS_RELOADING(m))
2508 return;
2509
2510 if (!MANAGER_IS_SYSTEM(m))
2511 return;
2512
2513 if (detect_container() > 0)
2514 return;
2515
2516 if (!IN_SET(u->type, UNIT_SERVICE, UNIT_MOUNT, UNIT_SWAP))
2517 return;
2518
2519 /* We set SOCK_NONBLOCK here so that we rather drop the
2520 * message then wait for plymouth */
2521 fd = socket(AF_UNIX, SOCK_STREAM|SOCK_CLOEXEC|SOCK_NONBLOCK, 0);
2522 if (fd < 0) {
2523 log_error_errno(errno, "socket() failed: %m");
2524 return;
2525 }
2526
2527 if (connect(fd, &sa.sa, SOCKADDR_UN_LEN(sa.un)) < 0) {
2528
2529 if (!IN_SET(errno, EPIPE, EAGAIN, ENOENT, ECONNREFUSED, ECONNRESET, ECONNABORTED))
2530 log_error_errno(errno, "connect() failed: %m");
2531 return;
2532 }
2533
2534 if (asprintf(&message, "U\002%c%s%n", (int) (strlen(u->id) + 1), u->id, &n) < 0) {
2535 log_oom();
2536 return;
2537 }
2538
2539 errno = 0;
2540 if (write(fd, message, n + 1) != n + 1)
2541 if (!IN_SET(errno, EPIPE, EAGAIN, ENOENT, ECONNREFUSED, ECONNRESET, ECONNABORTED))
2542 log_error_errno(errno, "Failed to write Plymouth message: %m");
2543 }
2544
2545 int manager_open_serialization(Manager *m, FILE **_f) {
2546 int fd;
2547 FILE *f;
2548
2549 assert(_f);
2550
2551 fd = open_serialization_fd("systemd-state");
2552 if (fd < 0)
2553 return fd;
2554
2555 f = fdopen(fd, "w+");
2556 if (!f) {
2557 safe_close(fd);
2558 return -errno;
2559 }
2560
2561 *_f = f;
2562 return 0;
2563 }
2564
2565 int manager_serialize(Manager *m, FILE *f, FDSet *fds, bool switching_root) {
2566 Iterator i;
2567 Unit *u;
2568 const char *t;
2569 int r;
2570
2571 assert(m);
2572 assert(f);
2573 assert(fds);
2574
2575 m->n_reloading++;
2576
2577 fprintf(f, "current-job-id=%"PRIu32"\n", m->current_job_id);
2578 fprintf(f, "taint-usr=%s\n", yes_no(m->taint_usr));
2579 fprintf(f, "n-installed-jobs=%u\n", m->n_installed_jobs);
2580 fprintf(f, "n-failed-jobs=%u\n", m->n_failed_jobs);
2581
2582 dual_timestamp_serialize(f, "firmware-timestamp", &m->firmware_timestamp);
2583 dual_timestamp_serialize(f, "loader-timestamp", &m->loader_timestamp);
2584 dual_timestamp_serialize(f, "kernel-timestamp", &m->kernel_timestamp);
2585 dual_timestamp_serialize(f, "initrd-timestamp", &m->initrd_timestamp);
2586
2587 if (!in_initrd()) {
2588 dual_timestamp_serialize(f, "userspace-timestamp", &m->userspace_timestamp);
2589 dual_timestamp_serialize(f, "finish-timestamp", &m->finish_timestamp);
2590 dual_timestamp_serialize(f, "security-start-timestamp", &m->security_start_timestamp);
2591 dual_timestamp_serialize(f, "security-finish-timestamp", &m->security_finish_timestamp);
2592 dual_timestamp_serialize(f, "generators-start-timestamp", &m->generators_start_timestamp);
2593 dual_timestamp_serialize(f, "generators-finish-timestamp", &m->generators_finish_timestamp);
2594 dual_timestamp_serialize(f, "units-load-start-timestamp", &m->units_load_start_timestamp);
2595 dual_timestamp_serialize(f, "units-load-finish-timestamp", &m->units_load_finish_timestamp);
2596 }
2597
2598 if (!switching_root)
2599 (void) serialize_environment(f, m->environment);
2600
2601 if (m->notify_fd >= 0) {
2602 int copy;
2603
2604 copy = fdset_put_dup(fds, m->notify_fd);
2605 if (copy < 0)
2606 return copy;
2607
2608 fprintf(f, "notify-fd=%i\n", copy);
2609 fprintf(f, "notify-socket=%s\n", m->notify_socket);
2610 }
2611
2612 if (m->cgroups_agent_fd >= 0) {
2613 int copy;
2614
2615 copy = fdset_put_dup(fds, m->cgroups_agent_fd);
2616 if (copy < 0)
2617 return copy;
2618
2619 fprintf(f, "cgroups-agent-fd=%i\n", copy);
2620 }
2621
2622 if (m->user_lookup_fds[0] >= 0) {
2623 int copy0, copy1;
2624
2625 copy0 = fdset_put_dup(fds, m->user_lookup_fds[0]);
2626 if (copy0 < 0)
2627 return copy0;
2628
2629 copy1 = fdset_put_dup(fds, m->user_lookup_fds[1]);
2630 if (copy1 < 0)
2631 return copy1;
2632
2633 fprintf(f, "user-lookup=%i %i\n", copy0, copy1);
2634 }
2635
2636 bus_track_serialize(m->subscribed, f, "subscribed");
2637
2638 r = dynamic_user_serialize(m, f, fds);
2639 if (r < 0)
2640 return r;
2641
2642 manager_serialize_uid_refs(m, f);
2643 manager_serialize_gid_refs(m, f);
2644
2645 fputc_unlocked('\n', f);
2646
2647 HASHMAP_FOREACH_KEY(u, t, m->units, i) {
2648 if (u->id != t)
2649 continue;
2650
2651 /* Start marker */
2652 fputs_unlocked(u->id, f);
2653 fputc_unlocked('\n', f);
2654
2655 r = unit_serialize(u, f, fds, !switching_root);
2656 if (r < 0) {
2657 m->n_reloading--;
2658 return r;
2659 }
2660 }
2661
2662 assert(m->n_reloading > 0);
2663 m->n_reloading--;
2664
2665 if (ferror(f))
2666 return -EIO;
2667
2668 r = bus_fdset_add_all(m, fds);
2669 if (r < 0)
2670 return r;
2671
2672 return 0;
2673 }
2674
2675 int manager_deserialize(Manager *m, FILE *f, FDSet *fds) {
2676 int r = 0;
2677
2678 assert(m);
2679 assert(f);
2680
2681 log_debug("Deserializing state...");
2682
2683 m->n_reloading++;
2684
2685 for (;;) {
2686 char line[LINE_MAX];
2687 const char *val, *l;
2688
2689 if (!fgets(line, sizeof(line), f)) {
2690 if (feof(f))
2691 r = 0;
2692 else
2693 r = -errno;
2694
2695 goto finish;
2696 }
2697
2698 char_array_0(line);
2699 l = strstrip(line);
2700
2701 if (l[0] == 0)
2702 break;
2703
2704 if ((val = startswith(l, "current-job-id="))) {
2705 uint32_t id;
2706
2707 if (safe_atou32(val, &id) < 0)
2708 log_notice("Failed to parse current job id value %s", val);
2709 else
2710 m->current_job_id = MAX(m->current_job_id, id);
2711
2712 } else if ((val = startswith(l, "n-installed-jobs="))) {
2713 uint32_t n;
2714
2715 if (safe_atou32(val, &n) < 0)
2716 log_notice("Failed to parse installed jobs counter %s", val);
2717 else
2718 m->n_installed_jobs += n;
2719
2720 } else if ((val = startswith(l, "n-failed-jobs="))) {
2721 uint32_t n;
2722
2723 if (safe_atou32(val, &n) < 0)
2724 log_notice("Failed to parse failed jobs counter %s", val);
2725 else
2726 m->n_failed_jobs += n;
2727
2728 } else if ((val = startswith(l, "taint-usr="))) {
2729 int b;
2730
2731 b = parse_boolean(val);
2732 if (b < 0)
2733 log_notice("Failed to parse taint /usr flag %s", val);
2734 else
2735 m->taint_usr = m->taint_usr || b;
2736
2737 } else if ((val = startswith(l, "firmware-timestamp=")))
2738 dual_timestamp_deserialize(val, &m->firmware_timestamp);
2739 else if ((val = startswith(l, "loader-timestamp=")))
2740 dual_timestamp_deserialize(val, &m->loader_timestamp);
2741 else if ((val = startswith(l, "kernel-timestamp=")))
2742 dual_timestamp_deserialize(val, &m->kernel_timestamp);
2743 else if ((val = startswith(l, "initrd-timestamp=")))
2744 dual_timestamp_deserialize(val, &m->initrd_timestamp);
2745 else if ((val = startswith(l, "userspace-timestamp=")))
2746 dual_timestamp_deserialize(val, &m->userspace_timestamp);
2747 else if ((val = startswith(l, "finish-timestamp=")))
2748 dual_timestamp_deserialize(val, &m->finish_timestamp);
2749 else if ((val = startswith(l, "security-start-timestamp=")))
2750 dual_timestamp_deserialize(val, &m->security_start_timestamp);
2751 else if ((val = startswith(l, "security-finish-timestamp=")))
2752 dual_timestamp_deserialize(val, &m->security_finish_timestamp);
2753 else if ((val = startswith(l, "generators-start-timestamp=")))
2754 dual_timestamp_deserialize(val, &m->generators_start_timestamp);
2755 else if ((val = startswith(l, "generators-finish-timestamp=")))
2756 dual_timestamp_deserialize(val, &m->generators_finish_timestamp);
2757 else if ((val = startswith(l, "units-load-start-timestamp=")))
2758 dual_timestamp_deserialize(val, &m->units_load_start_timestamp);
2759 else if ((val = startswith(l, "units-load-finish-timestamp=")))
2760 dual_timestamp_deserialize(val, &m->units_load_finish_timestamp);
2761 else if (startswith(l, "env=")) {
2762 r = deserialize_environment(&m->environment, l);
2763 if (r == -ENOMEM)
2764 goto finish;
2765 if (r < 0)
2766 log_notice_errno(r, "Failed to parse environment entry: \"%s\": %m", l);
2767
2768 } else if ((val = startswith(l, "notify-fd="))) {
2769 int fd;
2770
2771 if (safe_atoi(val, &fd) < 0 || fd < 0 || !fdset_contains(fds, fd))
2772 log_notice("Failed to parse notify fd: \"%s\"", val);
2773 else {
2774 m->notify_event_source = sd_event_source_unref(m->notify_event_source);
2775 safe_close(m->notify_fd);
2776 m->notify_fd = fdset_remove(fds, fd);
2777 }
2778
2779 } else if ((val = startswith(l, "notify-socket="))) {
2780 char *n;
2781
2782 n = strdup(val);
2783 if (!n) {
2784 r = -ENOMEM;
2785 goto finish;
2786 }
2787
2788 free(m->notify_socket);
2789 m->notify_socket = n;
2790
2791 } else if ((val = startswith(l, "cgroups-agent-fd="))) {
2792 int fd;
2793
2794 if (safe_atoi(val, &fd) < 0 || fd < 0 || !fdset_contains(fds, fd))
2795 log_notice("Failed to parse cgroups agent fd: %s", val);
2796 else {
2797 m->cgroups_agent_event_source = sd_event_source_unref(m->cgroups_agent_event_source);
2798 safe_close(m->cgroups_agent_fd);
2799 m->cgroups_agent_fd = fdset_remove(fds, fd);
2800 }
2801
2802 } else if ((val = startswith(l, "user-lookup="))) {
2803 int fd0, fd1;
2804
2805 if (sscanf(val, "%i %i", &fd0, &fd1) != 2 || fd0 < 0 || fd1 < 0 || fd0 == fd1 || !fdset_contains(fds, fd0) || !fdset_contains(fds, fd1))
2806 log_notice("Failed to parse user lookup fd: %s", val);
2807 else {
2808 m->user_lookup_event_source = sd_event_source_unref(m->user_lookup_event_source);
2809 safe_close_pair(m->user_lookup_fds);
2810 m->user_lookup_fds[0] = fdset_remove(fds, fd0);
2811 m->user_lookup_fds[1] = fdset_remove(fds, fd1);
2812 }
2813
2814 } else if ((val = startswith(l, "dynamic-user=")))
2815 dynamic_user_deserialize_one(m, val, fds);
2816 else if ((val = startswith(l, "destroy-ipc-uid=")))
2817 manager_deserialize_uid_refs_one(m, val);
2818 else if ((val = startswith(l, "destroy-ipc-gid=")))
2819 manager_deserialize_gid_refs_one(m, val);
2820 else if ((val = startswith(l, "subscribed="))) {
2821
2822 if (strv_extend(&m->deserialized_subscribed, val) < 0)
2823 log_oom();
2824
2825 } else if (!startswith(l, "kdbus-fd=")) /* ignore this one */
2826 log_notice("Unknown serialization item '%s'", l);
2827 }
2828
2829 for (;;) {
2830 Unit *u;
2831 char name[UNIT_NAME_MAX+2];
2832 const char* unit_name;
2833
2834 /* Start marker */
2835 if (!fgets(name, sizeof(name), f)) {
2836 if (feof(f))
2837 r = 0;
2838 else
2839 r = -errno;
2840
2841 goto finish;
2842 }
2843
2844 char_array_0(name);
2845 unit_name = strstrip(name);
2846
2847 r = manager_load_unit(m, unit_name, NULL, NULL, &u);
2848 if (r < 0) {
2849 log_notice_errno(r, "Failed to load unit \"%s\", skipping deserialization: %m", unit_name);
2850 if (r == -ENOMEM)
2851 goto finish;
2852 unit_deserialize_skip(f);
2853 continue;
2854 }
2855
2856 r = unit_deserialize(u, f, fds);
2857 if (r < 0) {
2858 log_notice_errno(r, "Failed to deserialize unit \"%s\": %m", unit_name);
2859 if (r == -ENOMEM)
2860 goto finish;
2861 }
2862 }
2863
2864 finish:
2865 if (ferror(f))
2866 r = -EIO;
2867
2868 assert(m->n_reloading > 0);
2869 m->n_reloading--;
2870
2871 return r;
2872 }
2873
2874 int manager_reload(Manager *m) {
2875 int r, q;
2876 _cleanup_fclose_ FILE *f = NULL;
2877 _cleanup_fdset_free_ FDSet *fds = NULL;
2878
2879 assert(m);
2880
2881 r = manager_open_serialization(m, &f);
2882 if (r < 0)
2883 return r;
2884
2885 m->n_reloading++;
2886 bus_manager_send_reloading(m, true);
2887
2888 fds = fdset_new();
2889 if (!fds) {
2890 m->n_reloading--;
2891 return -ENOMEM;
2892 }
2893
2894 r = manager_serialize(m, f, fds, false);
2895 if (r < 0) {
2896 m->n_reloading--;
2897 return r;
2898 }
2899
2900 if (fseeko(f, 0, SEEK_SET) < 0) {
2901 m->n_reloading--;
2902 return -errno;
2903 }
2904
2905 /* From here on there is no way back. */
2906 manager_clear_jobs_and_units(m);
2907 lookup_paths_flush_generator(&m->lookup_paths);
2908 lookup_paths_free(&m->lookup_paths);
2909 dynamic_user_vacuum(m, false);
2910 m->uid_refs = hashmap_free(m->uid_refs);
2911 m->gid_refs = hashmap_free(m->gid_refs);
2912
2913 q = lookup_paths_init(&m->lookup_paths, m->unit_file_scope, 0, NULL);
2914 if (q < 0 && r >= 0)
2915 r = q;
2916
2917 q = manager_run_environment_generators(m);
2918 if (q < 0 && r >= 0)
2919 r = q;
2920
2921 /* Find new unit paths */
2922 q = manager_run_generators(m);
2923 if (q < 0 && r >= 0)
2924 r = q;
2925
2926 lookup_paths_reduce(&m->lookup_paths);
2927 manager_build_unit_path_cache(m);
2928
2929 /* First, enumerate what we can from all config files */
2930 manager_enumerate(m);
2931
2932 /* Second, deserialize our stored data */
2933 q = manager_deserialize(m, f, fds);
2934 if (q < 0) {
2935 log_error_errno(q, "Deserialization failed: %m");
2936
2937 if (r >= 0)
2938 r = q;
2939 }
2940
2941 fclose(f);
2942 f = NULL;
2943
2944 /* Re-register notify_fd as event source */
2945 q = manager_setup_notify(m);
2946 if (q < 0 && r >= 0)
2947 r = q;
2948
2949 q = manager_setup_cgroups_agent(m);
2950 if (q < 0 && r >= 0)
2951 r = q;
2952
2953 q = manager_setup_user_lookup_fd(m);
2954 if (q < 0 && r >= 0)
2955 r = q;
2956
2957 /* Third, fire things up! */
2958 manager_coldplug(m);
2959
2960 /* Release any dynamic users no longer referenced */
2961 dynamic_user_vacuum(m, true);
2962
2963 /* Release any references to UIDs/GIDs no longer referenced, and destroy any IPC owned by them */
2964 manager_vacuum_uid_refs(m);
2965 manager_vacuum_gid_refs(m);
2966
2967 /* Sync current state of bus names with our set of listening units */
2968 if (m->api_bus)
2969 manager_sync_bus_names(m, m->api_bus);
2970
2971 assert(m->n_reloading > 0);
2972 m->n_reloading--;
2973
2974 m->send_reloading_done = true;
2975
2976 return r;
2977 }
2978
2979 void manager_reset_failed(Manager *m) {
2980 Unit *u;
2981 Iterator i;
2982
2983 assert(m);
2984
2985 HASHMAP_FOREACH(u, m->units, i)
2986 unit_reset_failed(u);
2987 }
2988
2989 bool manager_unit_inactive_or_pending(Manager *m, const char *name) {
2990 Unit *u;
2991
2992 assert(m);
2993 assert(name);
2994
2995 /* Returns true if the unit is inactive or going down */
2996 u = manager_get_unit(m, name);
2997 if (!u)
2998 return true;
2999
3000 return unit_inactive_or_pending(u);
3001 }
3002
3003 static void manager_notify_finished(Manager *m) {
3004 char userspace[FORMAT_TIMESPAN_MAX], initrd[FORMAT_TIMESPAN_MAX], kernel[FORMAT_TIMESPAN_MAX], sum[FORMAT_TIMESPAN_MAX];
3005 usec_t firmware_usec, loader_usec, kernel_usec, initrd_usec, userspace_usec, total_usec;
3006
3007 if (m->test_run_flags)
3008 return;
3009
3010 if (MANAGER_IS_SYSTEM(m) && detect_container() <= 0) {
3011
3012 /* Note that m->kernel_usec.monotonic is always at 0,
3013 * and m->firmware_usec.monotonic and
3014 * m->loader_usec.monotonic should be considered
3015 * negative values. */
3016
3017 firmware_usec = m->firmware_timestamp.monotonic - m->loader_timestamp.monotonic;
3018 loader_usec = m->loader_timestamp.monotonic - m->kernel_timestamp.monotonic;
3019 userspace_usec = m->finish_timestamp.monotonic - m->userspace_timestamp.monotonic;
3020 total_usec = m->firmware_timestamp.monotonic + m->finish_timestamp.monotonic;
3021
3022 if (dual_timestamp_is_set(&m->initrd_timestamp)) {
3023
3024 kernel_usec = m->initrd_timestamp.monotonic - m->kernel_timestamp.monotonic;
3025 initrd_usec = m->userspace_timestamp.monotonic - m->initrd_timestamp.monotonic;
3026
3027 log_struct(LOG_INFO,
3028 "MESSAGE_ID=" SD_MESSAGE_STARTUP_FINISHED_STR,
3029 "KERNEL_USEC="USEC_FMT, kernel_usec,
3030 "INITRD_USEC="USEC_FMT, initrd_usec,
3031 "USERSPACE_USEC="USEC_FMT, userspace_usec,
3032 LOG_MESSAGE("Startup finished in %s (kernel) + %s (initrd) + %s (userspace) = %s.",
3033 format_timespan(kernel, sizeof(kernel), kernel_usec, USEC_PER_MSEC),
3034 format_timespan(initrd, sizeof(initrd), initrd_usec, USEC_PER_MSEC),
3035 format_timespan(userspace, sizeof(userspace), userspace_usec, USEC_PER_MSEC),
3036 format_timespan(sum, sizeof(sum), total_usec, USEC_PER_MSEC)),
3037 NULL);
3038 } else {
3039 kernel_usec = m->userspace_timestamp.monotonic - m->kernel_timestamp.monotonic;
3040 initrd_usec = 0;
3041
3042 log_struct(LOG_INFO,
3043 "MESSAGE_ID=" SD_MESSAGE_STARTUP_FINISHED_STR,
3044 "KERNEL_USEC="USEC_FMT, kernel_usec,
3045 "USERSPACE_USEC="USEC_FMT, userspace_usec,
3046 LOG_MESSAGE("Startup finished in %s (kernel) + %s (userspace) = %s.",
3047 format_timespan(kernel, sizeof(kernel), kernel_usec, USEC_PER_MSEC),
3048 format_timespan(userspace, sizeof(userspace), userspace_usec, USEC_PER_MSEC),
3049 format_timespan(sum, sizeof(sum), total_usec, USEC_PER_MSEC)),
3050 NULL);
3051 }
3052 } else {
3053 firmware_usec = loader_usec = initrd_usec = kernel_usec = 0;
3054 total_usec = userspace_usec = m->finish_timestamp.monotonic - m->userspace_timestamp.monotonic;
3055
3056 log_struct(LOG_INFO,
3057 "MESSAGE_ID=" SD_MESSAGE_USER_STARTUP_FINISHED_STR,
3058 "USERSPACE_USEC="USEC_FMT, userspace_usec,
3059 LOG_MESSAGE("Startup finished in %s.",
3060 format_timespan(sum, sizeof(sum), total_usec, USEC_PER_MSEC)),
3061 NULL);
3062 }
3063
3064 bus_manager_send_finished(m, firmware_usec, loader_usec, kernel_usec, initrd_usec, userspace_usec, total_usec);
3065
3066 sd_notifyf(false,
3067 "READY=1\n"
3068 "STATUS=Startup finished in %s.",
3069 format_timespan(sum, sizeof(sum), total_usec, USEC_PER_MSEC));
3070 }
3071
3072 void manager_check_finished(Manager *m) {
3073 assert(m);
3074
3075 if (MANAGER_IS_RELOADING(m))
3076 return;
3077
3078 /* Verify that we are actually running currently. Initially
3079 * the exit code is set to invalid, and during operation it is
3080 * then set to MANAGER_OK */
3081 if (m->exit_code != MANAGER_OK)
3082 return;
3083
3084 if (hashmap_size(m->jobs) > 0) {
3085 if (m->jobs_in_progress_event_source)
3086 /* Ignore any failure, this is only for feedback */
3087 (void) sd_event_source_set_time(m->jobs_in_progress_event_source, now(CLOCK_MONOTONIC) + JOBS_IN_PROGRESS_WAIT_USEC);
3088
3089 return;
3090 }
3091
3092 manager_flip_auto_status(m, false);
3093
3094 /* Notify Type=idle units that we are done now */
3095 manager_close_idle_pipe(m);
3096
3097 /* Turn off confirm spawn now */
3098 m->confirm_spawn = NULL;
3099
3100 /* No need to update ask password status when we're going non-interactive */
3101 manager_close_ask_password(m);
3102
3103 /* This is no longer the first boot */
3104 manager_set_first_boot(m, false);
3105
3106 if (dual_timestamp_is_set(&m->finish_timestamp))
3107 return;
3108
3109 dual_timestamp_get(&m->finish_timestamp);
3110
3111 manager_notify_finished(m);
3112
3113 manager_invalidate_startup_units(m);
3114 }
3115
3116 static bool generator_path_any(const char* const* paths) {
3117 char **path;
3118 bool found = false;
3119
3120 /* Optimize by skipping the whole process by not creating output directories
3121 * if no generators are found. */
3122 STRV_FOREACH(path, (char**) paths)
3123 if (access(*path, F_OK) == 0)
3124 found = true;
3125 else if (errno != ENOENT)
3126 log_warning_errno(errno, "Failed to open generator directory %s: %m", *path);
3127
3128 return found;
3129 }
3130
3131 static const char* system_env_generator_binary_paths[] = {
3132 "/run/systemd/system-environment-generators",
3133 "/etc/systemd/system-environment-generators",
3134 "/usr/local/lib/systemd/system-environment-generators",
3135 SYSTEM_ENV_GENERATOR_PATH,
3136 NULL
3137 };
3138
3139 static const char* user_env_generator_binary_paths[] = {
3140 "/run/systemd/user-environment-generators",
3141 "/etc/systemd/user-environment-generators",
3142 "/usr/local/lib/systemd/user-environment-generators",
3143 USER_ENV_GENERATOR_PATH,
3144 NULL
3145 };
3146
3147 static int manager_run_environment_generators(Manager *m) {
3148 char **tmp = NULL; /* this is only used in the forked process, no cleanup here */
3149 const char **paths;
3150 void* args[] = {&tmp, &tmp, &m->environment};
3151
3152 if (m->test_run_flags && !(m->test_run_flags & MANAGER_TEST_RUN_ENV_GENERATORS))
3153 return 0;
3154
3155 paths = MANAGER_IS_SYSTEM(m) ? system_env_generator_binary_paths : user_env_generator_binary_paths;
3156
3157 if (!generator_path_any(paths))
3158 return 0;
3159
3160 return execute_directories(paths, DEFAULT_TIMEOUT_USEC, gather_environment, args, NULL);
3161 }
3162
3163 static int manager_run_generators(Manager *m) {
3164 _cleanup_strv_free_ char **paths = NULL;
3165 const char *argv[5];
3166 int r;
3167
3168 assert(m);
3169
3170 if (m->test_run_flags && !(m->test_run_flags & MANAGER_TEST_RUN_GENERATORS))
3171 return 0;
3172
3173 paths = generator_binary_paths(m->unit_file_scope);
3174 if (!paths)
3175 return log_oom();
3176
3177 if (!generator_path_any((const char* const*) paths))
3178 return 0;
3179
3180 r = lookup_paths_mkdir_generator(&m->lookup_paths);
3181 if (r < 0)
3182 goto finish;
3183
3184 argv[0] = NULL; /* Leave this empty, execute_directory() will fill something in */
3185 argv[1] = m->lookup_paths.generator;
3186 argv[2] = m->lookup_paths.generator_early;
3187 argv[3] = m->lookup_paths.generator_late;
3188 argv[4] = NULL;
3189
3190 RUN_WITH_UMASK(0022)
3191 execute_directories((const char* const*) paths, DEFAULT_TIMEOUT_USEC,
3192 NULL, NULL, (char**) argv);
3193
3194 finish:
3195 lookup_paths_trim_generator(&m->lookup_paths);
3196 return r;
3197 }
3198
3199 int manager_environment_add(Manager *m, char **minus, char **plus) {
3200 char **a = NULL, **b = NULL, **l;
3201 assert(m);
3202
3203 l = m->environment;
3204
3205 if (!strv_isempty(minus)) {
3206 a = strv_env_delete(l, 1, minus);
3207 if (!a)
3208 return -ENOMEM;
3209
3210 l = a;
3211 }
3212
3213 if (!strv_isempty(plus)) {
3214 b = strv_env_merge(2, l, plus);
3215 if (!b) {
3216 strv_free(a);
3217 return -ENOMEM;
3218 }
3219
3220 l = b;
3221 }
3222
3223 if (m->environment != l)
3224 strv_free(m->environment);
3225 if (a != l)
3226 strv_free(a);
3227 if (b != l)
3228 strv_free(b);
3229
3230 m->environment = l;
3231 manager_clean_environment(m);
3232 strv_sort(m->environment);
3233
3234 return 0;
3235 }
3236
3237 int manager_set_default_rlimits(Manager *m, struct rlimit **default_rlimit) {
3238 int i;
3239
3240 assert(m);
3241
3242 for (i = 0; i < _RLIMIT_MAX; i++) {
3243 m->rlimit[i] = mfree(m->rlimit[i]);
3244
3245 if (!default_rlimit[i])
3246 continue;
3247
3248 m->rlimit[i] = newdup(struct rlimit, default_rlimit[i], 1);
3249 if (!m->rlimit[i])
3250 return log_oom();
3251 }
3252
3253 return 0;
3254 }
3255
3256 void manager_recheck_journal(Manager *m) {
3257 Unit *u;
3258
3259 assert(m);
3260
3261 if (!MANAGER_IS_SYSTEM(m))
3262 return;
3263
3264 u = manager_get_unit(m, SPECIAL_JOURNALD_SOCKET);
3265 if (u && SOCKET(u)->state != SOCKET_RUNNING) {
3266 log_close_journal();
3267 return;
3268 }
3269
3270 u = manager_get_unit(m, SPECIAL_JOURNALD_SERVICE);
3271 if (u && SERVICE(u)->state != SERVICE_RUNNING) {
3272 log_close_journal();
3273 return;
3274 }
3275
3276 /* Hmm, OK, so the socket is fully up and the service is up
3277 * too, then let's make use of the thing. */
3278 log_open();
3279 }
3280
3281 void manager_set_show_status(Manager *m, ShowStatus mode) {
3282 assert(m);
3283 assert(IN_SET(mode, SHOW_STATUS_AUTO, SHOW_STATUS_NO, SHOW_STATUS_YES, SHOW_STATUS_TEMPORARY));
3284
3285 if (!MANAGER_IS_SYSTEM(m))
3286 return;
3287
3288 if (m->show_status != mode)
3289 log_debug("%s showing of status.",
3290 mode == SHOW_STATUS_NO ? "Disabling" : "Enabling");
3291 m->show_status = mode;
3292
3293 if (mode > 0)
3294 (void) touch("/run/systemd/show-status");
3295 else
3296 (void) unlink("/run/systemd/show-status");
3297 }
3298
3299 static bool manager_get_show_status(Manager *m, StatusType type) {
3300 assert(m);
3301
3302 if (!MANAGER_IS_SYSTEM(m))
3303 return false;
3304
3305 if (m->no_console_output)
3306 return false;
3307
3308 if (!IN_SET(manager_state(m), MANAGER_INITIALIZING, MANAGER_STARTING, MANAGER_STOPPING))
3309 return false;
3310
3311 /* If we cannot find out the status properly, just proceed. */
3312 if (type != STATUS_TYPE_EMERGENCY && manager_check_ask_password(m) > 0)
3313 return false;
3314
3315 if (m->show_status > 0)
3316 return true;
3317
3318 return false;
3319 }
3320
3321 const char *manager_get_confirm_spawn(Manager *m) {
3322 static int last_errno = 0;
3323 const char *vc = m->confirm_spawn;
3324 struct stat st;
3325 int r;
3326
3327 /* Here's the deal: we want to test the validity of the console but don't want
3328 * PID1 to go through the whole console process which might block. But we also
3329 * want to warn the user only once if something is wrong with the console so we
3330 * cannot do the sanity checks after spawning our children. So here we simply do
3331 * really basic tests to hopefully trap common errors.
3332 *
3333 * If the console suddenly disappear at the time our children will really it
3334 * then they will simply fail to acquire it and a positive answer will be
3335 * assumed. New children will fallback to /dev/console though.
3336 *
3337 * Note: TTYs are devices that can come and go any time, and frequently aren't
3338 * available yet during early boot (consider a USB rs232 dongle...). If for any
3339 * reason the configured console is not ready, we fallback to the default
3340 * console. */
3341
3342 if (!vc || path_equal(vc, "/dev/console"))
3343 return vc;
3344
3345 r = stat(vc, &st);
3346 if (r < 0)
3347 goto fail;
3348
3349 if (!S_ISCHR(st.st_mode)) {
3350 errno = ENOTTY;
3351 goto fail;
3352 }
3353
3354 last_errno = 0;
3355 return vc;
3356 fail:
3357 if (last_errno != errno) {
3358 last_errno = errno;
3359 log_warning_errno(errno, "Failed to open %s: %m, using default console", vc);
3360 }
3361 return "/dev/console";
3362 }
3363
3364 void manager_set_first_boot(Manager *m, bool b) {
3365 assert(m);
3366
3367 if (!MANAGER_IS_SYSTEM(m))
3368 return;
3369
3370 if (m->first_boot != (int) b) {
3371 if (b)
3372 (void) touch("/run/systemd/first-boot");
3373 else
3374 (void) unlink("/run/systemd/first-boot");
3375 }
3376
3377 m->first_boot = b;
3378 }
3379
3380 void manager_disable_confirm_spawn(void) {
3381 (void) touch("/run/systemd/confirm_spawn_disabled");
3382 }
3383
3384 bool manager_is_confirm_spawn_disabled(Manager *m) {
3385 if (!m->confirm_spawn)
3386 return true;
3387
3388 return access("/run/systemd/confirm_spawn_disabled", F_OK) >= 0;
3389 }
3390
3391 void manager_status_printf(Manager *m, StatusType type, const char *status, const char *format, ...) {
3392 va_list ap;
3393
3394 /* If m is NULL, assume we're after shutdown and let the messages through. */
3395
3396 if (m && !manager_get_show_status(m, type))
3397 return;
3398
3399 /* XXX We should totally drop the check for ephemeral here
3400 * and thus effectively make 'Type=idle' pointless. */
3401 if (type == STATUS_TYPE_EPHEMERAL && m && m->n_on_console > 0)
3402 return;
3403
3404 va_start(ap, format);
3405 status_vprintf(status, true, type == STATUS_TYPE_EPHEMERAL, format, ap);
3406 va_end(ap);
3407 }
3408
3409 Set *manager_get_units_requiring_mounts_for(Manager *m, const char *path) {
3410 char p[strlen(path)+1];
3411
3412 assert(m);
3413 assert(path);
3414
3415 strcpy(p, path);
3416 path_kill_slashes(p);
3417
3418 return hashmap_get(m->units_requiring_mounts_for, streq(p, "/") ? "" : p);
3419 }
3420
3421 void manager_set_exec_params(Manager *m, ExecParameters *p) {
3422 assert(m);
3423 assert(p);
3424
3425 p->environment = m->environment;
3426 p->confirm_spawn = manager_get_confirm_spawn(m);
3427 p->cgroup_supported = m->cgroup_supported;
3428 p->prefix = m->prefix;
3429
3430 SET_FLAG(p->flags, EXEC_PASS_LOG_UNIT|EXEC_CHOWN_DIRECTORIES, MANAGER_IS_SYSTEM(m));
3431 }
3432
3433 int manager_update_failed_units(Manager *m, Unit *u, bool failed) {
3434 unsigned size;
3435 int r;
3436
3437 assert(m);
3438 assert(u->manager == m);
3439
3440 size = set_size(m->failed_units);
3441
3442 if (failed) {
3443 r = set_ensure_allocated(&m->failed_units, NULL);
3444 if (r < 0)
3445 return log_oom();
3446
3447 if (set_put(m->failed_units, u) < 0)
3448 return log_oom();
3449 } else
3450 (void) set_remove(m->failed_units, u);
3451
3452 if (set_size(m->failed_units) != size)
3453 bus_manager_send_change_signal(m);
3454
3455 return 0;
3456 }
3457
3458 ManagerState manager_state(Manager *m) {
3459 Unit *u;
3460
3461 assert(m);
3462
3463 /* Did we ever finish booting? If not then we are still starting up */
3464 if (!dual_timestamp_is_set(&m->finish_timestamp)) {
3465
3466 u = manager_get_unit(m, SPECIAL_BASIC_TARGET);
3467 if (!u || !UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u)))
3468 return MANAGER_INITIALIZING;
3469
3470 return MANAGER_STARTING;
3471 }
3472
3473 /* Is the special shutdown target queued? If so, we are in shutdown state */
3474 u = manager_get_unit(m, SPECIAL_SHUTDOWN_TARGET);
3475 if (u && u->job && IN_SET(u->job->type, JOB_START, JOB_RESTART, JOB_RELOAD_OR_START))
3476 return MANAGER_STOPPING;
3477
3478 /* Are the rescue or emergency targets active or queued? If so we are in maintenance state */
3479 u = manager_get_unit(m, SPECIAL_RESCUE_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 u = manager_get_unit(m, SPECIAL_EMERGENCY_TARGET);
3485 if (u && (UNIT_IS_ACTIVE_OR_ACTIVATING(unit_active_state(u)) ||
3486 (u->job && IN_SET(u->job->type, JOB_START, JOB_RESTART, JOB_RELOAD_OR_START))))
3487 return MANAGER_MAINTENANCE;
3488
3489 /* Are there any failed units? If so, we are in degraded mode */
3490 if (set_size(m->failed_units) > 0)
3491 return MANAGER_DEGRADED;
3492
3493 return MANAGER_RUNNING;
3494 }
3495
3496 #define DESTROY_IPC_FLAG (UINT32_C(1) << 31)
3497
3498 static void manager_unref_uid_internal(
3499 Manager *m,
3500 Hashmap **uid_refs,
3501 uid_t uid,
3502 bool destroy_now,
3503 int (*_clean_ipc)(uid_t uid)) {
3504
3505 uint32_t c, n;
3506
3507 assert(m);
3508 assert(uid_refs);
3509 assert(uid_is_valid(uid));
3510 assert(_clean_ipc);
3511
3512 /* A generic implementation, covering both manager_unref_uid() and manager_unref_gid(), under the assumption
3513 * that uid_t and gid_t are actually defined the same way, with the same validity rules.
3514 *
3515 * We store a hashmap where the UID/GID is they key and the value is a 32bit reference counter, whose highest
3516 * bit is used as flag for marking UIDs/GIDs whose IPC objects to remove when the last reference to the UID/GID
3517 * is dropped. The flag is set to on, once at least one reference from a unit where RemoveIPC= is set is added
3518 * on a UID/GID. It is reset when the UID's/GID's reference counter drops to 0 again. */
3519
3520 assert_cc(sizeof(uid_t) == sizeof(gid_t));
3521 assert_cc(UID_INVALID == (uid_t) GID_INVALID);
3522
3523 if (uid == 0) /* We don't keep track of root, and will never destroy it */
3524 return;
3525
3526 c = PTR_TO_UINT32(hashmap_get(*uid_refs, UID_TO_PTR(uid)));
3527
3528 n = c & ~DESTROY_IPC_FLAG;
3529 assert(n > 0);
3530 n--;
3531
3532 if (destroy_now && n == 0) {
3533 hashmap_remove(*uid_refs, UID_TO_PTR(uid));
3534
3535 if (c & DESTROY_IPC_FLAG) {
3536 log_debug("%s " UID_FMT " is no longer referenced, cleaning up its IPC.",
3537 _clean_ipc == clean_ipc_by_uid ? "UID" : "GID",
3538 uid);
3539 (void) _clean_ipc(uid);
3540 }
3541 } else {
3542 c = n | (c & DESTROY_IPC_FLAG);
3543 assert_se(hashmap_update(*uid_refs, UID_TO_PTR(uid), UINT32_TO_PTR(c)) >= 0);
3544 }
3545 }
3546
3547 void manager_unref_uid(Manager *m, uid_t uid, bool destroy_now) {
3548 manager_unref_uid_internal(m, &m->uid_refs, uid, destroy_now, clean_ipc_by_uid);
3549 }
3550
3551 void manager_unref_gid(Manager *m, gid_t gid, bool destroy_now) {
3552 manager_unref_uid_internal(m, &m->gid_refs, (uid_t) gid, destroy_now, clean_ipc_by_gid);
3553 }
3554
3555 static int manager_ref_uid_internal(
3556 Manager *m,
3557 Hashmap **uid_refs,
3558 uid_t uid,
3559 bool clean_ipc) {
3560
3561 uint32_t c, n;
3562 int r;
3563
3564 assert(m);
3565 assert(uid_refs);
3566 assert(uid_is_valid(uid));
3567
3568 /* A generic implementation, covering both manager_ref_uid() and manager_ref_gid(), under the assumption
3569 * that uid_t and gid_t are actually defined the same way, with the same validity rules. */
3570
3571 assert_cc(sizeof(uid_t) == sizeof(gid_t));
3572 assert_cc(UID_INVALID == (uid_t) GID_INVALID);
3573
3574 if (uid == 0) /* We don't keep track of root, and will never destroy it */
3575 return 0;
3576
3577 r = hashmap_ensure_allocated(uid_refs, &trivial_hash_ops);
3578 if (r < 0)
3579 return r;
3580
3581 c = PTR_TO_UINT32(hashmap_get(*uid_refs, UID_TO_PTR(uid)));
3582
3583 n = c & ~DESTROY_IPC_FLAG;
3584 n++;
3585
3586 if (n & DESTROY_IPC_FLAG) /* check for overflow */
3587 return -EOVERFLOW;
3588
3589 c = n | (c & DESTROY_IPC_FLAG) | (clean_ipc ? DESTROY_IPC_FLAG : 0);
3590
3591 return hashmap_replace(*uid_refs, UID_TO_PTR(uid), UINT32_TO_PTR(c));
3592 }
3593
3594 int manager_ref_uid(Manager *m, uid_t uid, bool clean_ipc) {
3595 return manager_ref_uid_internal(m, &m->uid_refs, uid, clean_ipc);
3596 }
3597
3598 int manager_ref_gid(Manager *m, gid_t gid, bool clean_ipc) {
3599 return manager_ref_uid_internal(m, &m->gid_refs, (uid_t) gid, clean_ipc);
3600 }
3601
3602 static void manager_vacuum_uid_refs_internal(
3603 Manager *m,
3604 Hashmap **uid_refs,
3605 int (*_clean_ipc)(uid_t uid)) {
3606
3607 Iterator i;
3608 void *p, *k;
3609
3610 assert(m);
3611 assert(uid_refs);
3612 assert(_clean_ipc);
3613
3614 HASHMAP_FOREACH_KEY(p, k, *uid_refs, i) {
3615 uint32_t c, n;
3616 uid_t uid;
3617
3618 uid = PTR_TO_UID(k);
3619 c = PTR_TO_UINT32(p);
3620
3621 n = c & ~DESTROY_IPC_FLAG;
3622 if (n > 0)
3623 continue;
3624
3625 if (c & DESTROY_IPC_FLAG) {
3626 log_debug("Found unreferenced %s " UID_FMT " after reload/reexec. Cleaning up.",
3627 _clean_ipc == clean_ipc_by_uid ? "UID" : "GID",
3628 uid);
3629 (void) _clean_ipc(uid);
3630 }
3631
3632 assert_se(hashmap_remove(*uid_refs, k) == p);
3633 }
3634 }
3635
3636 void manager_vacuum_uid_refs(Manager *m) {
3637 manager_vacuum_uid_refs_internal(m, &m->uid_refs, clean_ipc_by_uid);
3638 }
3639
3640 void manager_vacuum_gid_refs(Manager *m) {
3641 manager_vacuum_uid_refs_internal(m, &m->gid_refs, clean_ipc_by_gid);
3642 }
3643
3644 static void manager_serialize_uid_refs_internal(
3645 Manager *m,
3646 FILE *f,
3647 Hashmap **uid_refs,
3648 const char *field_name) {
3649
3650 Iterator i;
3651 void *p, *k;
3652
3653 assert(m);
3654 assert(f);
3655 assert(uid_refs);
3656 assert(field_name);
3657
3658 /* Serialize the UID reference table. Or actually, just the IPC destruction flag of it, as the actual counter
3659 * of it is better rebuild after a reload/reexec. */
3660
3661 HASHMAP_FOREACH_KEY(p, k, *uid_refs, i) {
3662 uint32_t c;
3663 uid_t uid;
3664
3665 uid = PTR_TO_UID(k);
3666 c = PTR_TO_UINT32(p);
3667
3668 if (!(c & DESTROY_IPC_FLAG))
3669 continue;
3670
3671 fprintf(f, "%s=" UID_FMT "\n", field_name, uid);
3672 }
3673 }
3674
3675 void manager_serialize_uid_refs(Manager *m, FILE *f) {
3676 manager_serialize_uid_refs_internal(m, f, &m->uid_refs, "destroy-ipc-uid");
3677 }
3678
3679 void manager_serialize_gid_refs(Manager *m, FILE *f) {
3680 manager_serialize_uid_refs_internal(m, f, &m->gid_refs, "destroy-ipc-gid");
3681 }
3682
3683 static void manager_deserialize_uid_refs_one_internal(
3684 Manager *m,
3685 Hashmap** uid_refs,
3686 const char *value) {
3687
3688 uid_t uid;
3689 uint32_t c;
3690 int r;
3691
3692 assert(m);
3693 assert(uid_refs);
3694 assert(value);
3695
3696 r = parse_uid(value, &uid);
3697 if (r < 0 || uid == 0) {
3698 log_debug("Unable to parse UID reference serialization");
3699 return;
3700 }
3701
3702 r = hashmap_ensure_allocated(uid_refs, &trivial_hash_ops);
3703 if (r < 0) {
3704 log_oom();
3705 return;
3706 }
3707
3708 c = PTR_TO_UINT32(hashmap_get(*uid_refs, UID_TO_PTR(uid)));
3709 if (c & DESTROY_IPC_FLAG)
3710 return;
3711
3712 c |= DESTROY_IPC_FLAG;
3713
3714 r = hashmap_replace(*uid_refs, UID_TO_PTR(uid), UINT32_TO_PTR(c));
3715 if (r < 0) {
3716 log_debug("Failed to add UID reference entry");
3717 return;
3718 }
3719 }
3720
3721 void manager_deserialize_uid_refs_one(Manager *m, const char *value) {
3722 manager_deserialize_uid_refs_one_internal(m, &m->uid_refs, value);
3723 }
3724
3725 void manager_deserialize_gid_refs_one(Manager *m, const char *value) {
3726 manager_deserialize_uid_refs_one_internal(m, &m->gid_refs, value);
3727 }
3728
3729 int manager_dispatch_user_lookup_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
3730 struct buffer {
3731 uid_t uid;
3732 gid_t gid;
3733 char unit_name[UNIT_NAME_MAX+1];
3734 } _packed_ buffer;
3735
3736 Manager *m = userdata;
3737 ssize_t l;
3738 size_t n;
3739 Unit *u;
3740
3741 assert_se(source);
3742 assert_se(m);
3743
3744 /* Invoked whenever a child process succeeded resolving its user/group to use and sent us the resulting UID/GID
3745 * in a datagram. We parse the datagram here and pass it off to the unit, so that it can add a reference to the
3746 * UID/GID so that it can destroy the UID/GID's IPC objects when the reference counter drops to 0. */
3747
3748 l = recv(fd, &buffer, sizeof(buffer), MSG_DONTWAIT);
3749 if (l < 0) {
3750 if (errno == EINTR || errno == EAGAIN)
3751 return 0;
3752
3753 return log_error_errno(errno, "Failed to read from user lookup fd: %m");
3754 }
3755
3756 if ((size_t) l <= offsetof(struct buffer, unit_name)) {
3757 log_warning("Received too short user lookup message, ignoring.");
3758 return 0;
3759 }
3760
3761 if ((size_t) l > offsetof(struct buffer, unit_name) + UNIT_NAME_MAX) {
3762 log_warning("Received too long user lookup message, ignoring.");
3763 return 0;
3764 }
3765
3766 if (!uid_is_valid(buffer.uid) && !gid_is_valid(buffer.gid)) {
3767 log_warning("Got user lookup message with invalid UID/GID pair, ignoring.");
3768 return 0;
3769 }
3770
3771 n = (size_t) l - offsetof(struct buffer, unit_name);
3772 if (memchr(buffer.unit_name, 0, n)) {
3773 log_warning("Received lookup message with embedded NUL character, ignoring.");
3774 return 0;
3775 }
3776
3777 buffer.unit_name[n] = 0;
3778 u = manager_get_unit(m, buffer.unit_name);
3779 if (!u) {
3780 log_debug("Got user lookup message but unit doesn't exist, ignoring.");
3781 return 0;
3782 }
3783
3784 log_unit_debug(u, "User lookup succeeded: uid=" UID_FMT " gid=" GID_FMT, buffer.uid, buffer.gid);
3785
3786 unit_notify_user_lookup(u, buffer.uid, buffer.gid);
3787 return 0;
3788 }
3789
3790 static const char *const manager_state_table[_MANAGER_STATE_MAX] = {
3791 [MANAGER_INITIALIZING] = "initializing",
3792 [MANAGER_STARTING] = "starting",
3793 [MANAGER_RUNNING] = "running",
3794 [MANAGER_DEGRADED] = "degraded",
3795 [MANAGER_MAINTENANCE] = "maintenance",
3796 [MANAGER_STOPPING] = "stopping",
3797 };
3798
3799 DEFINE_STRING_TABLE_LOOKUP(manager_state, ManagerState);
Unnamed repository; edit this file 'description' to name the repository.