]> 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
Merge pull request #7141 from yuwata/fix-7129
[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 && !IN_SET(errno, EPERM, 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 (IN_SET(u->gc_marker - gc_marker,
989 GC_OFFSET_GOOD, GC_OFFSET_BAD, GC_OFFSET_UNSURE, GC_OFFSET_IN_PATH))
990 return;
991
992 if (u->in_cleanup_queue)
993 goto bad;
994
995 if (unit_check_gc(u))
996 goto good;
997
998 u->gc_marker = gc_marker + GC_OFFSET_IN_PATH;
999
1000 is_bad = true;
1001
1002 SET_FOREACH(other, u->dependencies[UNIT_REFERENCED_BY], i) {
1003 unit_gc_sweep(other, gc_marker);
1004
1005 if (other->gc_marker == gc_marker + GC_OFFSET_GOOD)
1006 goto good;
1007
1008 if (other->gc_marker != gc_marker + GC_OFFSET_BAD)
1009 is_bad = false;
1010 }
1011
1012 if (is_bad)
1013 goto bad;
1014
1015 /* We were unable to find anything out about this entry, so
1016 * let's investigate it later */
1017 u->gc_marker = gc_marker + GC_OFFSET_UNSURE;
1018 unit_add_to_gc_queue(u);
1019 return;
1020
1021 bad:
1022 /* We definitely know that this one is not useful anymore, so
1023 * let's mark it for deletion */
1024 u->gc_marker = gc_marker + GC_OFFSET_BAD;
1025 unit_add_to_cleanup_queue(u);
1026 return;
1027
1028 good:
1029 unit_gc_mark_good(u, gc_marker);
1030 }
1031
1032 static unsigned manager_dispatch_gc_unit_queue(Manager *m) {
1033 unsigned n = 0, gc_marker;
1034 Unit *u;
1035
1036 assert(m);
1037
1038 /* log_debug("Running GC..."); */
1039
1040 m->gc_marker += _GC_OFFSET_MAX;
1041 if (m->gc_marker + _GC_OFFSET_MAX <= _GC_OFFSET_MAX)
1042 m->gc_marker = 1;
1043
1044 gc_marker = m->gc_marker;
1045
1046 while ((u = m->gc_unit_queue)) {
1047 assert(u->in_gc_queue);
1048
1049 unit_gc_sweep(u, gc_marker);
1050
1051 LIST_REMOVE(gc_queue, m->gc_unit_queue, u);
1052 u->in_gc_queue = false;
1053
1054 n++;
1055
1056 if (IN_SET(u->gc_marker - gc_marker,
1057 GC_OFFSET_BAD, GC_OFFSET_UNSURE)) {
1058 if (u->id)
1059 log_unit_debug(u, "Collecting.");
1060 u->gc_marker = gc_marker + GC_OFFSET_BAD;
1061 unit_add_to_cleanup_queue(u);
1062 }
1063 }
1064
1065 return n;
1066 }
1067
1068 static unsigned manager_dispatch_gc_job_queue(Manager *m) {
1069 unsigned n = 0;
1070 Job *j;
1071
1072 assert(m);
1073
1074 while ((j = m->gc_job_queue)) {
1075 assert(j->in_gc_queue);
1076
1077 LIST_REMOVE(gc_queue, m->gc_job_queue, j);
1078 j->in_gc_queue = false;
1079
1080 n++;
1081
1082 if (job_check_gc(j))
1083 continue;
1084
1085 log_unit_debug(j->unit, "Collecting job.");
1086 (void) job_finish_and_invalidate(j, JOB_COLLECTED, false, false);
1087 }
1088
1089 return n;
1090 }
1091
1092 static void manager_clear_jobs_and_units(Manager *m) {
1093 Unit *u;
1094
1095 assert(m);
1096
1097 while ((u = hashmap_first(m->units)))
1098 unit_free(u);
1099
1100 manager_dispatch_cleanup_queue(m);
1101
1102 assert(!m->load_queue);
1103 assert(!m->run_queue);
1104 assert(!m->dbus_unit_queue);
1105 assert(!m->dbus_job_queue);
1106 assert(!m->cleanup_queue);
1107 assert(!m->gc_unit_queue);
1108 assert(!m->gc_job_queue);
1109
1110 assert(hashmap_isempty(m->jobs));
1111 assert(hashmap_isempty(m->units));
1112
1113 m->n_on_console = 0;
1114 m->n_running_jobs = 0;
1115 }
1116
1117 Manager* manager_free(Manager *m) {
1118 UnitType c;
1119 int i;
1120 ExecDirectoryType dt;
1121
1122 if (!m)
1123 return NULL;
1124
1125 manager_clear_jobs_and_units(m);
1126
1127 for (c = 0; c < _UNIT_TYPE_MAX; c++)
1128 if (unit_vtable[c]->shutdown)
1129 unit_vtable[c]->shutdown(m);
1130
1131 /* If we reexecute ourselves, we keep the root cgroup around */
1132 manager_shutdown_cgroup(m, m->exit_code != MANAGER_REEXECUTE);
1133
1134 lookup_paths_flush_generator(&m->lookup_paths);
1135
1136 bus_done(m);
1137
1138 dynamic_user_vacuum(m, false);
1139 hashmap_free(m->dynamic_users);
1140
1141 hashmap_free(m->units);
1142 hashmap_free(m->units_by_invocation_id);
1143 hashmap_free(m->jobs);
1144 hashmap_free(m->watch_pids1);
1145 hashmap_free(m->watch_pids2);
1146 hashmap_free(m->watch_bus);
1147
1148 set_free(m->startup_units);
1149 set_free(m->failed_units);
1150
1151 sd_event_source_unref(m->signal_event_source);
1152 sd_event_source_unref(m->notify_event_source);
1153 sd_event_source_unref(m->cgroups_agent_event_source);
1154 sd_event_source_unref(m->time_change_event_source);
1155 sd_event_source_unref(m->jobs_in_progress_event_source);
1156 sd_event_source_unref(m->run_queue_event_source);
1157 sd_event_source_unref(m->user_lookup_event_source);
1158
1159 safe_close(m->signal_fd);
1160 safe_close(m->notify_fd);
1161 safe_close(m->cgroups_agent_fd);
1162 safe_close(m->time_change_fd);
1163 safe_close_pair(m->user_lookup_fds);
1164
1165 manager_close_ask_password(m);
1166
1167 manager_close_idle_pipe(m);
1168
1169 udev_unref(m->udev);
1170 sd_event_unref(m->event);
1171
1172 free(m->notify_socket);
1173
1174 lookup_paths_free(&m->lookup_paths);
1175 strv_free(m->environment);
1176
1177 hashmap_free(m->cgroup_unit);
1178 set_free_free(m->unit_path_cache);
1179
1180 free(m->switch_root);
1181 free(m->switch_root_init);
1182
1183 for (i = 0; i < _RLIMIT_MAX; i++)
1184 m->rlimit[i] = mfree(m->rlimit[i]);
1185
1186 assert(hashmap_isempty(m->units_requiring_mounts_for));
1187 hashmap_free(m->units_requiring_mounts_for);
1188
1189 hashmap_free(m->uid_refs);
1190 hashmap_free(m->gid_refs);
1191
1192 for (dt = 0; dt < _EXEC_DIRECTORY_TYPE_MAX; dt++)
1193 m->prefix[dt] = mfree(m->prefix[dt]);
1194
1195 return mfree(m);
1196 }
1197
1198 void manager_enumerate(Manager *m) {
1199 UnitType c;
1200
1201 assert(m);
1202
1203 /* Let's ask every type to load all units from disk/kernel
1204 * that it might know */
1205 for (c = 0; c < _UNIT_TYPE_MAX; c++) {
1206 if (!unit_type_supported(c)) {
1207 log_debug("Unit type .%s is not supported on this system.", unit_type_to_string(c));
1208 continue;
1209 }
1210
1211 if (!unit_vtable[c]->enumerate)
1212 continue;
1213
1214 unit_vtable[c]->enumerate(m);
1215 }
1216
1217 manager_dispatch_load_queue(m);
1218 }
1219
1220 static void manager_coldplug(Manager *m) {
1221 Iterator i;
1222 Unit *u;
1223 char *k;
1224 int r;
1225
1226 assert(m);
1227
1228 /* Then, let's set up their initial state. */
1229 HASHMAP_FOREACH_KEY(u, k, m->units, i) {
1230
1231 /* ignore aliases */
1232 if (u->id != k)
1233 continue;
1234
1235 r = unit_coldplug(u);
1236 if (r < 0)
1237 log_warning_errno(r, "We couldn't coldplug %s, proceeding anyway: %m", u->id);
1238 }
1239 }
1240
1241 static void manager_build_unit_path_cache(Manager *m) {
1242 char **i;
1243 int r;
1244
1245 assert(m);
1246
1247 set_free_free(m->unit_path_cache);
1248
1249 m->unit_path_cache = set_new(&string_hash_ops);
1250 if (!m->unit_path_cache) {
1251 r = -ENOMEM;
1252 goto fail;
1253 }
1254
1255 /* This simply builds a list of files we know exist, so that
1256 * we don't always have to go to disk */
1257
1258 STRV_FOREACH(i, m->lookup_paths.search_path) {
1259 _cleanup_closedir_ DIR *d = NULL;
1260 struct dirent *de;
1261
1262 d = opendir(*i);
1263 if (!d) {
1264 if (errno != ENOENT)
1265 log_warning_errno(errno, "Failed to open directory %s, ignoring: %m", *i);
1266 continue;
1267 }
1268
1269 FOREACH_DIRENT(de, d, r = -errno; goto fail) {
1270 char *p;
1271
1272 p = strjoin(streq(*i, "/") ? "" : *i, "/", de->d_name);
1273 if (!p) {
1274 r = -ENOMEM;
1275 goto fail;
1276 }
1277
1278 r = set_consume(m->unit_path_cache, p);
1279 if (r < 0)
1280 goto fail;
1281 }
1282 }
1283
1284 return;
1285
1286 fail:
1287 log_warning_errno(r, "Failed to build unit path cache, proceeding without: %m");
1288 m->unit_path_cache = set_free_free(m->unit_path_cache);
1289 }
1290
1291 static void manager_distribute_fds(Manager *m, FDSet *fds) {
1292 Iterator i;
1293 Unit *u;
1294
1295 assert(m);
1296
1297 HASHMAP_FOREACH(u, m->units, i) {
1298
1299 if (fdset_size(fds) <= 0)
1300 break;
1301
1302 if (!UNIT_VTABLE(u)->distribute_fds)
1303 continue;
1304
1305 UNIT_VTABLE(u)->distribute_fds(u, fds);
1306 }
1307 }
1308
1309 int manager_startup(Manager *m, FILE *serialization, FDSet *fds) {
1310 int r, q;
1311
1312 assert(m);
1313
1314 /* If we are running in test mode, we still want to run the generators,
1315 * but we should not touch the real generator directories. */
1316 r = lookup_paths_init(&m->lookup_paths, m->unit_file_scope,
1317 m->test_run_flags ? LOOKUP_PATHS_TEMPORARY_GENERATED : 0,
1318 NULL);
1319 if (r < 0)
1320 return r;
1321
1322 r = manager_run_environment_generators(m);
1323 if (r < 0)
1324 return r;
1325
1326 /* Make sure the transient directory always exists, so that it remains
1327 * in the search path */
1328 r = mkdir_p_label(m->lookup_paths.transient, 0755);
1329 if (r < 0)
1330 return r;
1331
1332 dual_timestamp_get(&m->generators_start_timestamp);
1333 r = manager_run_generators(m);
1334 dual_timestamp_get(&m->generators_finish_timestamp);
1335 if (r < 0)
1336 return r;
1337
1338 if (m->first_boot > 0 &&
1339 m->unit_file_scope == UNIT_FILE_SYSTEM &&
1340 !m->test_run_flags) {
1341
1342 q = unit_file_preset_all(UNIT_FILE_SYSTEM, 0, NULL, UNIT_FILE_PRESET_ENABLE_ONLY, NULL, 0);
1343 if (q < 0)
1344 log_full_errno(q == -EEXIST ? LOG_NOTICE : LOG_WARNING, q, "Failed to populate /etc with preset unit settings, ignoring: %m");
1345 else
1346 log_info("Populated /etc with preset unit settings.");
1347 }
1348
1349 lookup_paths_reduce(&m->lookup_paths);
1350 manager_build_unit_path_cache(m);
1351
1352 /* If we will deserialize make sure that during enumeration
1353 * this is already known, so we increase the counter here
1354 * already */
1355 if (serialization)
1356 m->n_reloading++;
1357
1358 /* First, enumerate what we can from all config files */
1359 dual_timestamp_get(&m->units_load_start_timestamp);
1360 manager_enumerate(m);
1361 dual_timestamp_get(&m->units_load_finish_timestamp);
1362
1363 /* Second, deserialize if there is something to deserialize */
1364 if (serialization) {
1365 r = manager_deserialize(m, serialization, fds);
1366 if (r < 0)
1367 log_error_errno(r, "Deserialization failed: %m");
1368 }
1369
1370 /* Any fds left? Find some unit which wants them. This is
1371 * useful to allow container managers to pass some file
1372 * descriptors to us pre-initialized. This enables
1373 * socket-based activation of entire containers. */
1374 manager_distribute_fds(m, fds);
1375
1376 /* We might have deserialized the notify fd, but if we didn't
1377 * then let's create the bus now */
1378 q = manager_setup_notify(m);
1379 if (q < 0 && r == 0)
1380 r = q;
1381
1382 q = manager_setup_cgroups_agent(m);
1383 if (q < 0 && r == 0)
1384 r = q;
1385
1386 q = manager_setup_user_lookup_fd(m);
1387 if (q < 0 && r == 0)
1388 r = q;
1389
1390 /* Let's connect to the bus now. */
1391 (void) manager_connect_bus(m, !!serialization);
1392
1393 (void) bus_track_coldplug(m, &m->subscribed, false, m->deserialized_subscribed);
1394 m->deserialized_subscribed = strv_free(m->deserialized_subscribed);
1395
1396 /* Third, fire things up! */
1397 manager_coldplug(m);
1398
1399 /* Release any dynamic users no longer referenced */
1400 dynamic_user_vacuum(m, true);
1401
1402 /* Release any references to UIDs/GIDs no longer referenced, and destroy any IPC owned by them */
1403 manager_vacuum_uid_refs(m);
1404 manager_vacuum_gid_refs(m);
1405
1406 if (serialization) {
1407 assert(m->n_reloading > 0);
1408 m->n_reloading--;
1409
1410 /* Let's wait for the UnitNew/JobNew messages being
1411 * sent, before we notify that the reload is
1412 * finished */
1413 m->send_reloading_done = true;
1414 }
1415
1416 return r;
1417 }
1418
1419 int manager_add_job(Manager *m, JobType type, Unit *unit, JobMode mode, sd_bus_error *e, Job **_ret) {
1420 int r;
1421 Transaction *tr;
1422
1423 assert(m);
1424 assert(type < _JOB_TYPE_MAX);
1425 assert(unit);
1426 assert(mode < _JOB_MODE_MAX);
1427
1428 if (mode == JOB_ISOLATE && type != JOB_START)
1429 return sd_bus_error_setf(e, SD_BUS_ERROR_INVALID_ARGS, "Isolate is only valid for start.");
1430
1431 if (mode == JOB_ISOLATE && !unit->allow_isolate)
1432 return sd_bus_error_setf(e, BUS_ERROR_NO_ISOLATION, "Operation refused, unit may not be isolated.");
1433
1434 log_unit_debug(unit, "Trying to enqueue job %s/%s/%s", unit->id, job_type_to_string(type), job_mode_to_string(mode));
1435
1436 type = job_type_collapse(type, unit);
1437
1438 tr = transaction_new(mode == JOB_REPLACE_IRREVERSIBLY);
1439 if (!tr)
1440 return -ENOMEM;
1441
1442 r = transaction_add_job_and_dependencies(tr, type, unit, NULL, true, false,
1443 IN_SET(mode, JOB_IGNORE_DEPENDENCIES, JOB_IGNORE_REQUIREMENTS),
1444 mode == JOB_IGNORE_DEPENDENCIES, e);
1445 if (r < 0)
1446 goto tr_abort;
1447
1448 if (mode == JOB_ISOLATE) {
1449 r = transaction_add_isolate_jobs(tr, m);
1450 if (r < 0)
1451 goto tr_abort;
1452 }
1453
1454 r = transaction_activate(tr, m, mode, e);
1455 if (r < 0)
1456 goto tr_abort;
1457
1458 log_unit_debug(unit,
1459 "Enqueued job %s/%s as %u", unit->id,
1460 job_type_to_string(type), (unsigned) tr->anchor_job->id);
1461
1462 if (_ret)
1463 *_ret = tr->anchor_job;
1464
1465 transaction_free(tr);
1466 return 0;
1467
1468 tr_abort:
1469 transaction_abort(tr);
1470 transaction_free(tr);
1471 return r;
1472 }
1473
1474 int manager_add_job_by_name(Manager *m, JobType type, const char *name, JobMode mode, sd_bus_error *e, Job **ret) {
1475 Unit *unit = NULL; /* just to appease gcc, initialization is not really necessary */
1476 int r;
1477
1478 assert(m);
1479 assert(type < _JOB_TYPE_MAX);
1480 assert(name);
1481 assert(mode < _JOB_MODE_MAX);
1482
1483 r = manager_load_unit(m, name, NULL, NULL, &unit);
1484 if (r < 0)
1485 return r;
1486 assert(unit);
1487
1488 return manager_add_job(m, type, unit, mode, e, ret);
1489 }
1490
1491 int manager_add_job_by_name_and_warn(Manager *m, JobType type, const char *name, JobMode mode, Job **ret) {
1492 _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL;
1493 int r;
1494
1495 assert(m);
1496 assert(type < _JOB_TYPE_MAX);
1497 assert(name);
1498 assert(mode < _JOB_MODE_MAX);
1499
1500 r = manager_add_job_by_name(m, type, name, mode, &error, ret);
1501 if (r < 0)
1502 return log_warning_errno(r, "Failed to enqueue %s job for %s: %s", job_mode_to_string(mode), name, bus_error_message(&error, r));
1503
1504 return r;
1505 }
1506
1507 int manager_propagate_reload(Manager *m, Unit *unit, JobMode mode, sd_bus_error *e) {
1508 int r;
1509 Transaction *tr;
1510
1511 assert(m);
1512 assert(unit);
1513 assert(mode < _JOB_MODE_MAX);
1514 assert(mode != JOB_ISOLATE); /* Isolate is only valid for start */
1515
1516 tr = transaction_new(mode == JOB_REPLACE_IRREVERSIBLY);
1517 if (!tr)
1518 return -ENOMEM;
1519
1520 /* We need an anchor job */
1521 r = transaction_add_job_and_dependencies(tr, JOB_NOP, unit, NULL, false, false, true, true, e);
1522 if (r < 0)
1523 goto tr_abort;
1524
1525 /* Failure in adding individual dependencies is ignored, so this always succeeds. */
1526 transaction_add_propagate_reload_jobs(tr, unit, tr->anchor_job, mode == JOB_IGNORE_DEPENDENCIES, e);
1527
1528 r = transaction_activate(tr, m, mode, e);
1529 if (r < 0)
1530 goto tr_abort;
1531
1532 transaction_free(tr);
1533 return 0;
1534
1535 tr_abort:
1536 transaction_abort(tr);
1537 transaction_free(tr);
1538 return r;
1539 }
1540
1541 Job *manager_get_job(Manager *m, uint32_t id) {
1542 assert(m);
1543
1544 return hashmap_get(m->jobs, UINT32_TO_PTR(id));
1545 }
1546
1547 Unit *manager_get_unit(Manager *m, const char *name) {
1548 assert(m);
1549 assert(name);
1550
1551 return hashmap_get(m->units, name);
1552 }
1553
1554 unsigned manager_dispatch_load_queue(Manager *m) {
1555 Unit *u;
1556 unsigned n = 0;
1557
1558 assert(m);
1559
1560 /* Make sure we are not run recursively */
1561 if (m->dispatching_load_queue)
1562 return 0;
1563
1564 m->dispatching_load_queue = true;
1565
1566 /* Dispatches the load queue. Takes a unit from the queue and
1567 * tries to load its data until the queue is empty */
1568
1569 while ((u = m->load_queue)) {
1570 assert(u->in_load_queue);
1571
1572 unit_load(u);
1573 n++;
1574 }
1575
1576 m->dispatching_load_queue = false;
1577 return n;
1578 }
1579
1580 int manager_load_unit_prepare(
1581 Manager *m,
1582 const char *name,
1583 const char *path,
1584 sd_bus_error *e,
1585 Unit **_ret) {
1586
1587 Unit *ret;
1588 UnitType t;
1589 int r;
1590
1591 assert(m);
1592 assert(name || path);
1593 assert(_ret);
1594
1595 /* This will prepare the unit for loading, but not actually
1596 * load anything from disk. */
1597
1598 if (path && !is_path(path))
1599 return sd_bus_error_setf(e, SD_BUS_ERROR_INVALID_ARGS, "Path %s is not absolute.", path);
1600
1601 if (!name)
1602 name = basename(path);
1603
1604 t = unit_name_to_type(name);
1605
1606 if (t == _UNIT_TYPE_INVALID || !unit_name_is_valid(name, UNIT_NAME_PLAIN|UNIT_NAME_INSTANCE)) {
1607 if (unit_name_is_valid(name, UNIT_NAME_TEMPLATE))
1608 return sd_bus_error_setf(e, SD_BUS_ERROR_INVALID_ARGS, "Unit name %s is missing the instance name.", name);
1609
1610 return sd_bus_error_setf(e, SD_BUS_ERROR_INVALID_ARGS, "Unit name %s is not valid.", name);
1611 }
1612
1613 ret = manager_get_unit(m, name);
1614 if (ret) {
1615 *_ret = ret;
1616 return 1;
1617 }
1618
1619 ret = unit_new(m, unit_vtable[t]->object_size);
1620 if (!ret)
1621 return -ENOMEM;
1622
1623 if (path) {
1624 ret->fragment_path = strdup(path);
1625 if (!ret->fragment_path) {
1626 unit_free(ret);
1627 return -ENOMEM;
1628 }
1629 }
1630
1631 r = unit_add_name(ret, name);
1632 if (r < 0) {
1633 unit_free(ret);
1634 return r;
1635 }
1636
1637 unit_add_to_load_queue(ret);
1638 unit_add_to_dbus_queue(ret);
1639 unit_add_to_gc_queue(ret);
1640
1641 *_ret = ret;
1642
1643 return 0;
1644 }
1645
1646 int manager_load_unit(
1647 Manager *m,
1648 const char *name,
1649 const char *path,
1650 sd_bus_error *e,
1651 Unit **_ret) {
1652
1653 int r;
1654
1655 assert(m);
1656 assert(_ret);
1657
1658 /* This will load the service information files, but not actually
1659 * start any services or anything. */
1660
1661 r = manager_load_unit_prepare(m, name, path, e, _ret);
1662 if (r != 0)
1663 return r;
1664
1665 manager_dispatch_load_queue(m);
1666
1667 *_ret = unit_follow_merge(*_ret);
1668
1669 return 0;
1670 }
1671
1672 void manager_dump_jobs(Manager *s, FILE *f, const char *prefix) {
1673 Iterator i;
1674 Job *j;
1675
1676 assert(s);
1677 assert(f);
1678
1679 HASHMAP_FOREACH(j, s->jobs, i)
1680 job_dump(j, f, prefix);
1681 }
1682
1683 void manager_dump_units(Manager *s, FILE *f, const char *prefix) {
1684 Iterator i;
1685 Unit *u;
1686 const char *t;
1687
1688 assert(s);
1689 assert(f);
1690
1691 HASHMAP_FOREACH_KEY(u, t, s->units, i)
1692 if (u->id == t)
1693 unit_dump(u, f, prefix);
1694 }
1695
1696 void manager_clear_jobs(Manager *m) {
1697 Job *j;
1698
1699 assert(m);
1700
1701 while ((j = hashmap_first(m->jobs)))
1702 /* No need to recurse. We're cancelling all jobs. */
1703 job_finish_and_invalidate(j, JOB_CANCELED, false, false);
1704 }
1705
1706 static int manager_dispatch_run_queue(sd_event_source *source, void *userdata) {
1707 Manager *m = userdata;
1708 Job *j;
1709
1710 assert(source);
1711 assert(m);
1712
1713 while ((j = m->run_queue)) {
1714 assert(j->installed);
1715 assert(j->in_run_queue);
1716
1717 job_run_and_invalidate(j);
1718 }
1719
1720 if (m->n_running_jobs > 0)
1721 manager_watch_jobs_in_progress(m);
1722
1723 if (m->n_on_console > 0)
1724 manager_watch_idle_pipe(m);
1725
1726 return 1;
1727 }
1728
1729 static unsigned manager_dispatch_dbus_queue(Manager *m) {
1730 Job *j;
1731 Unit *u;
1732 unsigned n = 0;
1733
1734 assert(m);
1735
1736 if (m->dispatching_dbus_queue)
1737 return 0;
1738
1739 m->dispatching_dbus_queue = true;
1740
1741 while ((u = m->dbus_unit_queue)) {
1742 assert(u->in_dbus_queue);
1743
1744 bus_unit_send_change_signal(u);
1745 n++;
1746 }
1747
1748 while ((j = m->dbus_job_queue)) {
1749 assert(j->in_dbus_queue);
1750
1751 bus_job_send_change_signal(j);
1752 n++;
1753 }
1754
1755 m->dispatching_dbus_queue = false;
1756
1757 if (m->send_reloading_done) {
1758 m->send_reloading_done = false;
1759
1760 bus_manager_send_reloading(m, false);
1761 }
1762
1763 if (m->queued_message)
1764 bus_send_queued_message(m);
1765
1766 return n;
1767 }
1768
1769 static int manager_dispatch_cgroups_agent_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
1770 Manager *m = userdata;
1771 char buf[PATH_MAX+1];
1772 ssize_t n;
1773
1774 n = recv(fd, buf, sizeof(buf), 0);
1775 if (n < 0)
1776 return log_error_errno(errno, "Failed to read cgroups agent message: %m");
1777 if (n == 0) {
1778 log_error("Got zero-length cgroups agent message, ignoring.");
1779 return 0;
1780 }
1781 if ((size_t) n >= sizeof(buf)) {
1782 log_error("Got overly long cgroups agent message, ignoring.");
1783 return 0;
1784 }
1785
1786 if (memchr(buf, 0, n)) {
1787 log_error("Got cgroups agent message with embedded NUL byte, ignoring.");
1788 return 0;
1789 }
1790 buf[n] = 0;
1791
1792 manager_notify_cgroup_empty(m, buf);
1793 (void) bus_forward_agent_released(m, buf);
1794
1795 return 0;
1796 }
1797
1798 static void manager_invoke_notify_message(Manager *m, Unit *u, pid_t pid, const char *buf, FDSet *fds) {
1799 _cleanup_strv_free_ char **tags = NULL;
1800
1801 assert(m);
1802 assert(u);
1803 assert(buf);
1804
1805 tags = strv_split(buf, "\n\r");
1806 if (!tags) {
1807 log_oom();
1808 return;
1809 }
1810
1811 if (UNIT_VTABLE(u)->notify_message)
1812 UNIT_VTABLE(u)->notify_message(u, pid, tags, fds);
1813 else if (_unlikely_(log_get_max_level() >= LOG_DEBUG)) {
1814 _cleanup_free_ char *x = NULL, *y = NULL;
1815
1816 x = cescape(buf);
1817 if (x)
1818 y = ellipsize(x, 20, 90);
1819 log_unit_debug(u, "Got notification message \"%s\", ignoring.", strnull(y));
1820 }
1821 }
1822
1823 static int manager_dispatch_notify_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
1824
1825 _cleanup_fdset_free_ FDSet *fds = NULL;
1826 Manager *m = userdata;
1827 char buf[NOTIFY_BUFFER_MAX+1];
1828 struct iovec iovec = {
1829 .iov_base = buf,
1830 .iov_len = sizeof(buf)-1,
1831 };
1832 union {
1833 struct cmsghdr cmsghdr;
1834 uint8_t buf[CMSG_SPACE(sizeof(struct ucred)) +
1835 CMSG_SPACE(sizeof(int) * NOTIFY_FD_MAX)];
1836 } control = {};
1837 struct msghdr msghdr = {
1838 .msg_iov = &iovec,
1839 .msg_iovlen = 1,
1840 .msg_control = &control,
1841 .msg_controllen = sizeof(control),
1842 };
1843
1844 struct cmsghdr *cmsg;
1845 struct ucred *ucred = NULL;
1846 Unit *u1, *u2, *u3;
1847 int r, *fd_array = NULL;
1848 unsigned n_fds = 0;
1849 ssize_t n;
1850
1851 assert(m);
1852 assert(m->notify_fd == fd);
1853
1854 if (revents != EPOLLIN) {
1855 log_warning("Got unexpected poll event for notify fd.");
1856 return 0;
1857 }
1858
1859 n = recvmsg(m->notify_fd, &msghdr, MSG_DONTWAIT|MSG_CMSG_CLOEXEC|MSG_TRUNC);
1860 if (n < 0) {
1861 if (IN_SET(errno, EAGAIN, EINTR))
1862 return 0; /* Spurious wakeup, try again */
1863
1864 /* If this is any other, real error, then let's stop processing this socket. This of course means we
1865 * won't take notification messages anymore, but that's still better than busy looping around this:
1866 * being woken up over and over again but being unable to actually read the message off the socket. */
1867 return log_error_errno(errno, "Failed to receive notification message: %m");
1868 }
1869
1870 CMSG_FOREACH(cmsg, &msghdr) {
1871 if (cmsg->cmsg_level == SOL_SOCKET && cmsg->cmsg_type == SCM_RIGHTS) {
1872
1873 fd_array = (int*) CMSG_DATA(cmsg);
1874 n_fds = (cmsg->cmsg_len - CMSG_LEN(0)) / sizeof(int);
1875
1876 } else if (cmsg->cmsg_level == SOL_SOCKET &&
1877 cmsg->cmsg_type == SCM_CREDENTIALS &&
1878 cmsg->cmsg_len == CMSG_LEN(sizeof(struct ucred))) {
1879
1880 ucred = (struct ucred*) CMSG_DATA(cmsg);
1881 }
1882 }
1883
1884 if (n_fds > 0) {
1885 assert(fd_array);
1886
1887 r = fdset_new_array(&fds, fd_array, n_fds);
1888 if (r < 0) {
1889 close_many(fd_array, n_fds);
1890 log_oom();
1891 return 0;
1892 }
1893 }
1894
1895 if (!ucred || ucred->pid <= 0) {
1896 log_warning("Received notify message without valid credentials. Ignoring.");
1897 return 0;
1898 }
1899
1900 if ((size_t) n >= sizeof(buf) || (msghdr.msg_flags & MSG_TRUNC)) {
1901 log_warning("Received notify message exceeded maximum size. Ignoring.");
1902 return 0;
1903 }
1904
1905 /* As extra safety check, let's make sure the string we get doesn't contain embedded NUL bytes. We permit one
1906 * trailing NUL byte in the message, but don't expect it. */
1907 if (n > 1 && memchr(buf, 0, n-1)) {
1908 log_warning("Received notify message with embedded NUL bytes. Ignoring.");
1909 return 0;
1910 }
1911
1912 /* Make sure it's NUL-terminated. */
1913 buf[n] = 0;
1914
1915 /* Notify every unit that might be interested, but try
1916 * to avoid notifying the same one multiple times. */
1917 u1 = manager_get_unit_by_pid_cgroup(m, ucred->pid);
1918 if (u1)
1919 manager_invoke_notify_message(m, u1, ucred->pid, buf, fds);
1920
1921 u2 = hashmap_get(m->watch_pids1, PID_TO_PTR(ucred->pid));
1922 if (u2 && u2 != u1)
1923 manager_invoke_notify_message(m, u2, ucred->pid, buf, fds);
1924
1925 u3 = hashmap_get(m->watch_pids2, PID_TO_PTR(ucred->pid));
1926 if (u3 && u3 != u2 && u3 != u1)
1927 manager_invoke_notify_message(m, u3, ucred->pid, buf, fds);
1928
1929 if (!u1 && !u2 && !u3)
1930 log_warning("Cannot find unit for notify message of PID "PID_FMT".", ucred->pid);
1931
1932 if (fdset_size(fds) > 0)
1933 log_warning("Got extra auxiliary fds with notification message, closing them.");
1934
1935 return 0;
1936 }
1937
1938 static void invoke_sigchld_event(Manager *m, Unit *u, const siginfo_t *si) {
1939 uint64_t iteration;
1940
1941 assert(m);
1942 assert(u);
1943 assert(si);
1944
1945 sd_event_get_iteration(m->event, &iteration);
1946
1947 log_unit_debug(u, "Child "PID_FMT" belongs to %s", si->si_pid, u->id);
1948
1949 unit_unwatch_pid(u, si->si_pid);
1950
1951 if (UNIT_VTABLE(u)->sigchld_event) {
1952 if (set_size(u->pids) <= 1 ||
1953 iteration != u->sigchldgen ||
1954 unit_main_pid(u) == si->si_pid ||
1955 unit_control_pid(u) == si->si_pid) {
1956 UNIT_VTABLE(u)->sigchld_event(u, si->si_pid, si->si_code, si->si_status);
1957 u->sigchldgen = iteration;
1958 } else
1959 log_debug("%s already issued a sigchld this iteration %" PRIu64 ", skipping. Pids still being watched %d", u->id, iteration, set_size(u->pids));
1960 }
1961 }
1962
1963 static int manager_dispatch_sigchld(Manager *m) {
1964 assert(m);
1965
1966 for (;;) {
1967 siginfo_t si = {};
1968
1969 /* First we call waitd() for a PID and do not reap the
1970 * zombie. That way we can still access /proc/$PID for
1971 * it while it is a zombie. */
1972 if (waitid(P_ALL, 0, &si, WEXITED|WNOHANG|WNOWAIT) < 0) {
1973
1974 if (errno == ECHILD)
1975 break;
1976
1977 if (errno == EINTR)
1978 continue;
1979
1980 return -errno;
1981 }
1982
1983 if (si.si_pid <= 0)
1984 break;
1985
1986 if (IN_SET(si.si_code, CLD_EXITED, CLD_KILLED, CLD_DUMPED)) {
1987 _cleanup_free_ char *name = NULL;
1988 Unit *u1, *u2, *u3;
1989
1990 get_process_comm(si.si_pid, &name);
1991
1992 log_debug("Child "PID_FMT" (%s) died (code=%s, status=%i/%s)",
1993 si.si_pid, strna(name),
1994 sigchld_code_to_string(si.si_code),
1995 si.si_status,
1996 strna(si.si_code == CLD_EXITED
1997 ? exit_status_to_string(si.si_status, EXIT_STATUS_FULL)
1998 : signal_to_string(si.si_status)));
1999
2000 /* And now figure out the unit this belongs
2001 * to, it might be multiple... */
2002 u1 = manager_get_unit_by_pid_cgroup(m, si.si_pid);
2003 if (u1)
2004 invoke_sigchld_event(m, u1, &si);
2005 u2 = hashmap_get(m->watch_pids1, PID_TO_PTR(si.si_pid));
2006 if (u2 && u2 != u1)
2007 invoke_sigchld_event(m, u2, &si);
2008 u3 = hashmap_get(m->watch_pids2, PID_TO_PTR(si.si_pid));
2009 if (u3 && u3 != u2 && u3 != u1)
2010 invoke_sigchld_event(m, u3, &si);
2011 }
2012
2013 /* And now, we actually reap the zombie. */
2014 if (waitid(P_PID, si.si_pid, &si, WEXITED) < 0) {
2015 if (errno == EINTR)
2016 continue;
2017
2018 return -errno;
2019 }
2020 }
2021
2022 return 0;
2023 }
2024
2025 static void manager_start_target(Manager *m, const char *name, JobMode mode) {
2026 _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL;
2027 int r;
2028
2029 log_debug("Activating special unit %s", name);
2030
2031 r = manager_add_job_by_name(m, JOB_START, name, mode, &error, NULL);
2032 if (r < 0)
2033 log_error("Failed to enqueue %s job: %s", name, bus_error_message(&error, r));
2034 }
2035
2036 static void manager_handle_ctrl_alt_del(Manager *m) {
2037 /* If the user presses C-A-D more than
2038 * 7 times within 2s, we reboot/shutdown immediately,
2039 * unless it was disabled in system.conf */
2040
2041 if (ratelimit_test(&m->ctrl_alt_del_ratelimit) || m->cad_burst_action == EMERGENCY_ACTION_NONE)
2042 manager_start_target(m, SPECIAL_CTRL_ALT_DEL_TARGET, JOB_REPLACE_IRREVERSIBLY);
2043 else
2044 emergency_action(m, m->cad_burst_action, NULL,
2045 "Ctrl-Alt-Del was pressed more than 7 times within 2s");
2046 }
2047
2048 static int manager_dispatch_signal_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
2049 Manager *m = userdata;
2050 ssize_t n;
2051 struct signalfd_siginfo sfsi;
2052 bool sigchld = false;
2053 int r;
2054
2055 assert(m);
2056 assert(m->signal_fd == fd);
2057
2058 if (revents != EPOLLIN) {
2059 log_warning("Got unexpected events from signal file descriptor.");
2060 return 0;
2061 }
2062
2063 for (;;) {
2064 n = read(m->signal_fd, &sfsi, sizeof(sfsi));
2065 if (n != sizeof(sfsi)) {
2066 if (n >= 0) {
2067 log_warning("Truncated read from signal fd (%zu bytes)!", n);
2068 return 0;
2069 }
2070
2071 if (IN_SET(errno, EINTR, EAGAIN))
2072 break;
2073
2074 /* We return an error here, which will kill this handler,
2075 * to avoid a busy loop on read error. */
2076 return log_error_errno(errno, "Reading from signal fd failed: %m");
2077 }
2078
2079 log_received_signal(sfsi.ssi_signo == SIGCHLD ||
2080 (sfsi.ssi_signo == SIGTERM && MANAGER_IS_USER(m))
2081 ? LOG_DEBUG : LOG_INFO,
2082 &sfsi);
2083
2084 switch (sfsi.ssi_signo) {
2085
2086 case SIGCHLD:
2087 sigchld = true;
2088 break;
2089
2090 case SIGTERM:
2091 if (MANAGER_IS_SYSTEM(m)) {
2092 /* This is for compatibility with the
2093 * original sysvinit */
2094 r = verify_run_space_and_log("Refusing to reexecute");
2095 if (r >= 0)
2096 m->exit_code = MANAGER_REEXECUTE;
2097 break;
2098 }
2099
2100 /* Fall through */
2101
2102 case SIGINT:
2103 if (MANAGER_IS_SYSTEM(m))
2104 manager_handle_ctrl_alt_del(m);
2105 else
2106 manager_start_target(m, SPECIAL_EXIT_TARGET,
2107 JOB_REPLACE_IRREVERSIBLY);
2108 break;
2109
2110 case SIGWINCH:
2111 if (MANAGER_IS_SYSTEM(m))
2112 manager_start_target(m, SPECIAL_KBREQUEST_TARGET, JOB_REPLACE);
2113
2114 /* This is a nop on non-init */
2115 break;
2116
2117 case SIGPWR:
2118 if (MANAGER_IS_SYSTEM(m))
2119 manager_start_target(m, SPECIAL_SIGPWR_TARGET, JOB_REPLACE);
2120
2121 /* This is a nop on non-init */
2122 break;
2123
2124 case SIGUSR1: {
2125 Unit *u;
2126
2127 u = manager_get_unit(m, SPECIAL_DBUS_SERVICE);
2128
2129 if (!u || UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u))) {
2130 log_info("Trying to reconnect to bus...");
2131 bus_init(m, true);
2132 }
2133
2134 if (!u || !UNIT_IS_ACTIVE_OR_ACTIVATING(unit_active_state(u))) {
2135 log_info("Loading D-Bus service...");
2136 manager_start_target(m, SPECIAL_DBUS_SERVICE, JOB_REPLACE);
2137 }
2138
2139 break;
2140 }
2141
2142 case SIGUSR2: {
2143 _cleanup_free_ char *dump = NULL;
2144 _cleanup_fclose_ FILE *f = NULL;
2145 size_t size;
2146
2147 f = open_memstream(&dump, &size);
2148 if (!f) {
2149 log_warning_errno(errno, "Failed to allocate memory stream: %m");
2150 break;
2151 }
2152
2153 manager_dump_units(m, f, "\t");
2154 manager_dump_jobs(m, f, "\t");
2155
2156 r = fflush_and_check(f);
2157 if (r < 0) {
2158 log_warning_errno(r, "Failed to write status stream: %m");
2159 break;
2160 }
2161
2162 log_dump(LOG_INFO, dump);
2163 break;
2164 }
2165
2166 case SIGHUP:
2167 r = verify_run_space_and_log("Refusing to reload");
2168 if (r >= 0)
2169 m->exit_code = MANAGER_RELOAD;
2170 break;
2171
2172 default: {
2173
2174 /* Starting SIGRTMIN+0 */
2175 static const struct {
2176 const char *target;
2177 JobMode mode;
2178 } target_table[] = {
2179 [0] = { SPECIAL_DEFAULT_TARGET, JOB_ISOLATE },
2180 [1] = { SPECIAL_RESCUE_TARGET, JOB_ISOLATE },
2181 [2] = { SPECIAL_EMERGENCY_TARGET, JOB_ISOLATE },
2182 [3] = { SPECIAL_HALT_TARGET, JOB_REPLACE_IRREVERSIBLY },
2183 [4] = { SPECIAL_POWEROFF_TARGET, JOB_REPLACE_IRREVERSIBLY },
2184 [5] = { SPECIAL_REBOOT_TARGET, JOB_REPLACE_IRREVERSIBLY },
2185 [6] = { SPECIAL_KEXEC_TARGET, JOB_REPLACE_IRREVERSIBLY }
2186 };
2187
2188 /* Starting SIGRTMIN+13, so that target halt and system halt are 10 apart */
2189 static const ManagerExitCode code_table[] = {
2190 [0] = MANAGER_HALT,
2191 [1] = MANAGER_POWEROFF,
2192 [2] = MANAGER_REBOOT,
2193 [3] = MANAGER_KEXEC
2194 };
2195
2196 if ((int) sfsi.ssi_signo >= SIGRTMIN+0 &&
2197 (int) sfsi.ssi_signo < SIGRTMIN+(int) ELEMENTSOF(target_table)) {
2198 int idx = (int) sfsi.ssi_signo - SIGRTMIN;
2199 manager_start_target(m, target_table[idx].target,
2200 target_table[idx].mode);
2201 break;
2202 }
2203
2204 if ((int) sfsi.ssi_signo >= SIGRTMIN+13 &&
2205 (int) sfsi.ssi_signo < SIGRTMIN+13+(int) ELEMENTSOF(code_table)) {
2206 m->exit_code = code_table[sfsi.ssi_signo - SIGRTMIN - 13];
2207 break;
2208 }
2209
2210 switch (sfsi.ssi_signo - SIGRTMIN) {
2211
2212 case 20:
2213 manager_set_show_status(m, SHOW_STATUS_YES);
2214 break;
2215
2216 case 21:
2217 manager_set_show_status(m, SHOW_STATUS_NO);
2218 break;
2219
2220 case 22:
2221 log_set_max_level(LOG_DEBUG);
2222 log_info("Setting log level to debug.");
2223 break;
2224
2225 case 23:
2226 log_set_max_level(LOG_INFO);
2227 log_info("Setting log level to info.");
2228 break;
2229
2230 case 24:
2231 if (MANAGER_IS_USER(m)) {
2232 m->exit_code = MANAGER_EXIT;
2233 return 0;
2234 }
2235
2236 /* This is a nop on init */
2237 break;
2238
2239 case 26:
2240 case 29: /* compatibility: used to be mapped to LOG_TARGET_SYSLOG_OR_KMSG */
2241 log_set_target(LOG_TARGET_JOURNAL_OR_KMSG);
2242 log_notice("Setting log target to journal-or-kmsg.");
2243 break;
2244
2245 case 27:
2246 log_set_target(LOG_TARGET_CONSOLE);
2247 log_notice("Setting log target to console.");
2248 break;
2249
2250 case 28:
2251 log_set_target(LOG_TARGET_KMSG);
2252 log_notice("Setting log target to kmsg.");
2253 break;
2254
2255 default:
2256 log_warning("Got unhandled signal <%s>.", signal_to_string(sfsi.ssi_signo));
2257 }
2258 }
2259 }
2260 }
2261
2262 if (sigchld)
2263 manager_dispatch_sigchld(m);
2264
2265 return 0;
2266 }
2267
2268 static int manager_dispatch_time_change_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
2269 Manager *m = userdata;
2270 Iterator i;
2271 Unit *u;
2272
2273 assert(m);
2274 assert(m->time_change_fd == fd);
2275
2276 log_struct(LOG_DEBUG,
2277 "MESSAGE_ID=" SD_MESSAGE_TIME_CHANGE_STR,
2278 LOG_MESSAGE("Time has been changed"),
2279 NULL);
2280
2281 /* Restart the watch */
2282 m->time_change_event_source = sd_event_source_unref(m->time_change_event_source);
2283 m->time_change_fd = safe_close(m->time_change_fd);
2284
2285 manager_setup_time_change(m);
2286
2287 HASHMAP_FOREACH(u, m->units, i)
2288 if (UNIT_VTABLE(u)->time_change)
2289 UNIT_VTABLE(u)->time_change(u);
2290
2291 return 0;
2292 }
2293
2294 static int manager_dispatch_idle_pipe_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
2295 Manager *m = userdata;
2296
2297 assert(m);
2298 assert(m->idle_pipe[2] == fd);
2299
2300 m->no_console_output = m->n_on_console > 0;
2301
2302 manager_close_idle_pipe(m);
2303
2304 return 0;
2305 }
2306
2307 static int manager_dispatch_jobs_in_progress(sd_event_source *source, usec_t usec, void *userdata) {
2308 Manager *m = userdata;
2309 int r;
2310 uint64_t next;
2311
2312 assert(m);
2313 assert(source);
2314
2315 manager_print_jobs_in_progress(m);
2316
2317 next = now(CLOCK_MONOTONIC) + JOBS_IN_PROGRESS_PERIOD_USEC;
2318 r = sd_event_source_set_time(source, next);
2319 if (r < 0)
2320 return r;
2321
2322 return sd_event_source_set_enabled(source, SD_EVENT_ONESHOT);
2323 }
2324
2325 int manager_loop(Manager *m) {
2326 int r;
2327
2328 RATELIMIT_DEFINE(rl, 1*USEC_PER_SEC, 50000);
2329
2330 assert(m);
2331 m->exit_code = MANAGER_OK;
2332
2333 /* Release the path cache */
2334 m->unit_path_cache = set_free_free(m->unit_path_cache);
2335
2336 manager_check_finished(m);
2337
2338 /* There might still be some zombies hanging around from
2339 * before we were exec()'ed. Let's reap them. */
2340 r = manager_dispatch_sigchld(m);
2341 if (r < 0)
2342 return r;
2343
2344 while (m->exit_code == MANAGER_OK) {
2345 usec_t wait_usec;
2346
2347 if (m->runtime_watchdog > 0 && m->runtime_watchdog != USEC_INFINITY && MANAGER_IS_SYSTEM(m))
2348 watchdog_ping();
2349
2350 if (!ratelimit_test(&rl)) {
2351 /* Yay, something is going seriously wrong, pause a little */
2352 log_warning("Looping too fast. Throttling execution a little.");
2353 sleep(1);
2354 }
2355
2356 if (manager_dispatch_load_queue(m) > 0)
2357 continue;
2358
2359 if (manager_dispatch_gc_job_queue(m) > 0)
2360 continue;
2361
2362 if (manager_dispatch_gc_unit_queue(m) > 0)
2363 continue;
2364
2365 if (manager_dispatch_cleanup_queue(m) > 0)
2366 continue;
2367
2368 if (manager_dispatch_cgroup_realize_queue(m) > 0)
2369 continue;
2370
2371 if (manager_dispatch_dbus_queue(m) > 0)
2372 continue;
2373
2374 /* Sleep for half the watchdog time */
2375 if (m->runtime_watchdog > 0 && m->runtime_watchdog != USEC_INFINITY && MANAGER_IS_SYSTEM(m)) {
2376 wait_usec = m->runtime_watchdog / 2;
2377 if (wait_usec <= 0)
2378 wait_usec = 1;
2379 } else
2380 wait_usec = USEC_INFINITY;
2381
2382 r = sd_event_run(m->event, wait_usec);
2383 if (r < 0)
2384 return log_error_errno(r, "Failed to run event loop: %m");
2385 }
2386
2387 return m->exit_code;
2388 }
2389
2390 int manager_load_unit_from_dbus_path(Manager *m, const char *s, sd_bus_error *e, Unit **_u) {
2391 _cleanup_free_ char *n = NULL;
2392 sd_id128_t invocation_id;
2393 Unit *u;
2394 int r;
2395
2396 assert(m);
2397 assert(s);
2398 assert(_u);
2399
2400 r = unit_name_from_dbus_path(s, &n);
2401 if (r < 0)
2402 return r;
2403
2404 /* Permit addressing units by invocation ID: if the passed bus path is suffixed by a 128bit ID then we use it
2405 * as invocation ID. */
2406 r = sd_id128_from_string(n, &invocation_id);
2407 if (r >= 0) {
2408 u = hashmap_get(m->units_by_invocation_id, &invocation_id);
2409 if (u) {
2410 *_u = u;
2411 return 0;
2412 }
2413
2414 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));
2415 }
2416
2417 /* If this didn't work, we check if this is a unit name */
2418 if (!unit_name_is_valid(n, UNIT_NAME_PLAIN|UNIT_NAME_INSTANCE))
2419 return sd_bus_error_setf(e, SD_BUS_ERROR_INVALID_ARGS, "Unit name %s is neither a valid invocation ID nor unit name.", n);
2420
2421 r = manager_load_unit(m, n, NULL, e, &u);
2422 if (r < 0)
2423 return r;
2424
2425 *_u = u;
2426 return 0;
2427 }
2428
2429 int manager_get_job_from_dbus_path(Manager *m, const char *s, Job **_j) {
2430 const char *p;
2431 unsigned id;
2432 Job *j;
2433 int r;
2434
2435 assert(m);
2436 assert(s);
2437 assert(_j);
2438
2439 p = startswith(s, "/org/freedesktop/systemd1/job/");
2440 if (!p)
2441 return -EINVAL;
2442
2443 r = safe_atou(p, &id);
2444 if (r < 0)
2445 return r;
2446
2447 j = manager_get_job(m, id);
2448 if (!j)
2449 return -ENOENT;
2450
2451 *_j = j;
2452
2453 return 0;
2454 }
2455
2456 void manager_send_unit_audit(Manager *m, Unit *u, int type, bool success) {
2457
2458 #if HAVE_AUDIT
2459 _cleanup_free_ char *p = NULL;
2460 const char *msg;
2461 int audit_fd, r;
2462
2463 if (!MANAGER_IS_SYSTEM(m))
2464 return;
2465
2466 audit_fd = get_audit_fd();
2467 if (audit_fd < 0)
2468 return;
2469
2470 /* Don't generate audit events if the service was already
2471 * started and we're just deserializing */
2472 if (MANAGER_IS_RELOADING(m))
2473 return;
2474
2475 if (u->type != UNIT_SERVICE)
2476 return;
2477
2478 r = unit_name_to_prefix_and_instance(u->id, &p);
2479 if (r < 0) {
2480 log_error_errno(r, "Failed to extract prefix and instance of unit name: %m");
2481 return;
2482 }
2483
2484 msg = strjoina("unit=", p);
2485 if (audit_log_user_comm_message(audit_fd, type, msg, "systemd", NULL, NULL, NULL, success) < 0) {
2486 if (errno == EPERM)
2487 /* We aren't allowed to send audit messages?
2488 * Then let's not retry again. */
2489 close_audit_fd();
2490 else
2491 log_warning_errno(errno, "Failed to send audit message: %m");
2492 }
2493 #endif
2494
2495 }
2496
2497 void manager_send_unit_plymouth(Manager *m, Unit *u) {
2498 static const union sockaddr_union sa = PLYMOUTH_SOCKET;
2499 _cleanup_free_ char *message = NULL;
2500 _cleanup_close_ int fd = -1;
2501 int n = 0;
2502
2503 /* Don't generate plymouth events if the service was already
2504 * started and we're just deserializing */
2505 if (MANAGER_IS_RELOADING(m))
2506 return;
2507
2508 if (!MANAGER_IS_SYSTEM(m))
2509 return;
2510
2511 if (detect_container() > 0)
2512 return;
2513
2514 if (!IN_SET(u->type, UNIT_SERVICE, UNIT_MOUNT, UNIT_SWAP))
2515 return;
2516
2517 /* We set SOCK_NONBLOCK here so that we rather drop the
2518 * message then wait for plymouth */
2519 fd = socket(AF_UNIX, SOCK_STREAM|SOCK_CLOEXEC|SOCK_NONBLOCK, 0);
2520 if (fd < 0) {
2521 log_error_errno(errno, "socket() failed: %m");
2522 return;
2523 }
2524
2525 if (connect(fd, &sa.sa, SOCKADDR_UN_LEN(sa.un)) < 0) {
2526
2527 if (!IN_SET(errno, EPIPE, EAGAIN, ENOENT, ECONNREFUSED, ECONNRESET, ECONNABORTED))
2528 log_error_errno(errno, "connect() failed: %m");
2529 return;
2530 }
2531
2532 if (asprintf(&message, "U\002%c%s%n", (int) (strlen(u->id) + 1), u->id, &n) < 0) {
2533 log_oom();
2534 return;
2535 }
2536
2537 errno = 0;
2538 if (write(fd, message, n + 1) != n + 1)
2539 if (!IN_SET(errno, EPIPE, EAGAIN, ENOENT, ECONNREFUSED, ECONNRESET, ECONNABORTED))
2540 log_error_errno(errno, "Failed to write Plymouth message: %m");
2541 }
2542
2543 int manager_open_serialization(Manager *m, FILE **_f) {
2544 int fd;
2545 FILE *f;
2546
2547 assert(_f);
2548
2549 fd = open_serialization_fd("systemd-state");
2550 if (fd < 0)
2551 return fd;
2552
2553 f = fdopen(fd, "w+");
2554 if (!f) {
2555 safe_close(fd);
2556 return -errno;
2557 }
2558
2559 *_f = f;
2560 return 0;
2561 }
2562
2563 int manager_serialize(Manager *m, FILE *f, FDSet *fds, bool switching_root) {
2564 Iterator i;
2565 Unit *u;
2566 const char *t;
2567 int r;
2568
2569 assert(m);
2570 assert(f);
2571 assert(fds);
2572
2573 m->n_reloading++;
2574
2575 fprintf(f, "current-job-id=%"PRIu32"\n", m->current_job_id);
2576 fprintf(f, "n-installed-jobs=%u\n", m->n_installed_jobs);
2577 fprintf(f, "n-failed-jobs=%u\n", m->n_failed_jobs);
2578 fprintf(f, "taint-usr=%s\n", yes_no(m->taint_usr));
2579 fprintf(f, "ready-sent=%s\n", yes_no(m->ready_sent));
2580
2581 dual_timestamp_serialize(f, "firmware-timestamp", &m->firmware_timestamp);
2582 dual_timestamp_serialize(f, "loader-timestamp", &m->loader_timestamp);
2583 dual_timestamp_serialize(f, "kernel-timestamp", &m->kernel_timestamp);
2584 dual_timestamp_serialize(f, "initrd-timestamp", &m->initrd_timestamp);
2585
2586 if (!in_initrd()) {
2587 dual_timestamp_serialize(f, "userspace-timestamp", &m->userspace_timestamp);
2588 dual_timestamp_serialize(f, "finish-timestamp", &m->finish_timestamp);
2589 dual_timestamp_serialize(f, "security-start-timestamp", &m->security_start_timestamp);
2590 dual_timestamp_serialize(f, "security-finish-timestamp", &m->security_finish_timestamp);
2591 dual_timestamp_serialize(f, "generators-start-timestamp", &m->generators_start_timestamp);
2592 dual_timestamp_serialize(f, "generators-finish-timestamp", &m->generators_finish_timestamp);
2593 dual_timestamp_serialize(f, "units-load-start-timestamp", &m->units_load_start_timestamp);
2594 dual_timestamp_serialize(f, "units-load-finish-timestamp", &m->units_load_finish_timestamp);
2595 }
2596
2597 if (!switching_root)
2598 (void) serialize_environment(f, m->environment);
2599
2600 if (m->notify_fd >= 0) {
2601 int copy;
2602
2603 copy = fdset_put_dup(fds, m->notify_fd);
2604 if (copy < 0)
2605 return copy;
2606
2607 fprintf(f, "notify-fd=%i\n", copy);
2608 fprintf(f, "notify-socket=%s\n", m->notify_socket);
2609 }
2610
2611 if (m->cgroups_agent_fd >= 0) {
2612 int copy;
2613
2614 copy = fdset_put_dup(fds, m->cgroups_agent_fd);
2615 if (copy < 0)
2616 return copy;
2617
2618 fprintf(f, "cgroups-agent-fd=%i\n", copy);
2619 }
2620
2621 if (m->user_lookup_fds[0] >= 0) {
2622 int copy0, copy1;
2623
2624 copy0 = fdset_put_dup(fds, m->user_lookup_fds[0]);
2625 if (copy0 < 0)
2626 return copy0;
2627
2628 copy1 = fdset_put_dup(fds, m->user_lookup_fds[1]);
2629 if (copy1 < 0)
2630 return copy1;
2631
2632 fprintf(f, "user-lookup=%i %i\n", copy0, copy1);
2633 }
2634
2635 bus_track_serialize(m->subscribed, f, "subscribed");
2636
2637 r = dynamic_user_serialize(m, f, fds);
2638 if (r < 0)
2639 return r;
2640
2641 manager_serialize_uid_refs(m, f);
2642 manager_serialize_gid_refs(m, f);
2643
2644 fputc_unlocked('\n', f);
2645
2646 HASHMAP_FOREACH_KEY(u, t, m->units, i) {
2647 if (u->id != t)
2648 continue;
2649
2650 /* Start marker */
2651 fputs_unlocked(u->id, f);
2652 fputc_unlocked('\n', f);
2653
2654 r = unit_serialize(u, f, fds, !switching_root);
2655 if (r < 0) {
2656 m->n_reloading--;
2657 return r;
2658 }
2659 }
2660
2661 assert(m->n_reloading > 0);
2662 m->n_reloading--;
2663
2664 if (ferror(f))
2665 return -EIO;
2666
2667 r = bus_fdset_add_all(m, fds);
2668 if (r < 0)
2669 return r;
2670
2671 return 0;
2672 }
2673
2674 int manager_deserialize(Manager *m, FILE *f, FDSet *fds) {
2675 int r = 0;
2676
2677 assert(m);
2678 assert(f);
2679
2680 log_debug("Deserializing state...");
2681
2682 m->n_reloading++;
2683
2684 for (;;) {
2685 char line[LINE_MAX];
2686 const char *val, *l;
2687
2688 if (!fgets(line, sizeof(line), f)) {
2689 if (feof(f))
2690 r = 0;
2691 else
2692 r = -errno;
2693
2694 goto finish;
2695 }
2696
2697 char_array_0(line);
2698 l = strstrip(line);
2699
2700 if (l[0] == 0)
2701 break;
2702
2703 if ((val = startswith(l, "current-job-id="))) {
2704 uint32_t id;
2705
2706 if (safe_atou32(val, &id) < 0)
2707 log_notice("Failed to parse current job id value %s", val);
2708 else
2709 m->current_job_id = MAX(m->current_job_id, id);
2710
2711 } else if ((val = startswith(l, "n-installed-jobs="))) {
2712 uint32_t n;
2713
2714 if (safe_atou32(val, &n) < 0)
2715 log_notice("Failed to parse installed jobs counter %s", val);
2716 else
2717 m->n_installed_jobs += n;
2718
2719 } else if ((val = startswith(l, "n-failed-jobs="))) {
2720 uint32_t n;
2721
2722 if (safe_atou32(val, &n) < 0)
2723 log_notice("Failed to parse failed jobs counter %s", val);
2724 else
2725 m->n_failed_jobs += n;
2726
2727 } else if ((val = startswith(l, "taint-usr="))) {
2728 int b;
2729
2730 b = parse_boolean(val);
2731 if (b < 0)
2732 log_notice("Failed to parse taint /usr flag %s", val);
2733 else
2734 m->taint_usr = m->taint_usr || b;
2735
2736 } else if ((val = startswith(l, "ready-sent="))) {
2737 int b;
2738
2739 b = parse_boolean(val);
2740 if (b < 0)
2741 log_notice("Failed to parse ready-sent flag %s", val);
2742 else
2743 m->ready_sent = m->ready_sent || b;
2744
2745 } else if ((val = startswith(l, "firmware-timestamp=")))
2746 dual_timestamp_deserialize(val, &m->firmware_timestamp);
2747 else if ((val = startswith(l, "loader-timestamp=")))
2748 dual_timestamp_deserialize(val, &m->loader_timestamp);
2749 else if ((val = startswith(l, "kernel-timestamp=")))
2750 dual_timestamp_deserialize(val, &m->kernel_timestamp);
2751 else if ((val = startswith(l, "initrd-timestamp=")))
2752 dual_timestamp_deserialize(val, &m->initrd_timestamp);
2753 else if ((val = startswith(l, "userspace-timestamp=")))
2754 dual_timestamp_deserialize(val, &m->userspace_timestamp);
2755 else if ((val = startswith(l, "finish-timestamp=")))
2756 dual_timestamp_deserialize(val, &m->finish_timestamp);
2757 else if ((val = startswith(l, "security-start-timestamp=")))
2758 dual_timestamp_deserialize(val, &m->security_start_timestamp);
2759 else if ((val = startswith(l, "security-finish-timestamp=")))
2760 dual_timestamp_deserialize(val, &m->security_finish_timestamp);
2761 else if ((val = startswith(l, "generators-start-timestamp=")))
2762 dual_timestamp_deserialize(val, &m->generators_start_timestamp);
2763 else if ((val = startswith(l, "generators-finish-timestamp=")))
2764 dual_timestamp_deserialize(val, &m->generators_finish_timestamp);
2765 else if ((val = startswith(l, "units-load-start-timestamp=")))
2766 dual_timestamp_deserialize(val, &m->units_load_start_timestamp);
2767 else if ((val = startswith(l, "units-load-finish-timestamp=")))
2768 dual_timestamp_deserialize(val, &m->units_load_finish_timestamp);
2769 else if (startswith(l, "env=")) {
2770 r = deserialize_environment(&m->environment, l);
2771 if (r == -ENOMEM)
2772 goto finish;
2773 if (r < 0)
2774 log_notice_errno(r, "Failed to parse environment entry: \"%s\": %m", l);
2775
2776 } else if ((val = startswith(l, "notify-fd="))) {
2777 int fd;
2778
2779 if (safe_atoi(val, &fd) < 0 || fd < 0 || !fdset_contains(fds, fd))
2780 log_notice("Failed to parse notify fd: \"%s\"", val);
2781 else {
2782 m->notify_event_source = sd_event_source_unref(m->notify_event_source);
2783 safe_close(m->notify_fd);
2784 m->notify_fd = fdset_remove(fds, fd);
2785 }
2786
2787 } else if ((val = startswith(l, "notify-socket="))) {
2788 char *n;
2789
2790 n = strdup(val);
2791 if (!n) {
2792 r = -ENOMEM;
2793 goto finish;
2794 }
2795
2796 free(m->notify_socket);
2797 m->notify_socket = n;
2798
2799 } else if ((val = startswith(l, "cgroups-agent-fd="))) {
2800 int fd;
2801
2802 if (safe_atoi(val, &fd) < 0 || fd < 0 || !fdset_contains(fds, fd))
2803 log_notice("Failed to parse cgroups agent fd: %s", val);
2804 else {
2805 m->cgroups_agent_event_source = sd_event_source_unref(m->cgroups_agent_event_source);
2806 safe_close(m->cgroups_agent_fd);
2807 m->cgroups_agent_fd = fdset_remove(fds, fd);
2808 }
2809
2810 } else if ((val = startswith(l, "user-lookup="))) {
2811 int fd0, fd1;
2812
2813 if (sscanf(val, "%i %i", &fd0, &fd1) != 2 || fd0 < 0 || fd1 < 0 || fd0 == fd1 || !fdset_contains(fds, fd0) || !fdset_contains(fds, fd1))
2814 log_notice("Failed to parse user lookup fd: %s", val);
2815 else {
2816 m->user_lookup_event_source = sd_event_source_unref(m->user_lookup_event_source);
2817 safe_close_pair(m->user_lookup_fds);
2818 m->user_lookup_fds[0] = fdset_remove(fds, fd0);
2819 m->user_lookup_fds[1] = fdset_remove(fds, fd1);
2820 }
2821
2822 } else if ((val = startswith(l, "dynamic-user=")))
2823 dynamic_user_deserialize_one(m, val, fds);
2824 else if ((val = startswith(l, "destroy-ipc-uid=")))
2825 manager_deserialize_uid_refs_one(m, val);
2826 else if ((val = startswith(l, "destroy-ipc-gid=")))
2827 manager_deserialize_gid_refs_one(m, val);
2828 else if ((val = startswith(l, "subscribed="))) {
2829
2830 if (strv_extend(&m->deserialized_subscribed, val) < 0)
2831 log_oom();
2832
2833 } else if (!startswith(l, "kdbus-fd=")) /* ignore this one */
2834 log_notice("Unknown serialization item '%s'", l);
2835 }
2836
2837 for (;;) {
2838 Unit *u;
2839 char name[UNIT_NAME_MAX+2];
2840 const char* unit_name;
2841
2842 /* Start marker */
2843 if (!fgets(name, sizeof(name), f)) {
2844 if (feof(f))
2845 r = 0;
2846 else
2847 r = -errno;
2848
2849 goto finish;
2850 }
2851
2852 char_array_0(name);
2853 unit_name = strstrip(name);
2854
2855 r = manager_load_unit(m, unit_name, NULL, NULL, &u);
2856 if (r < 0) {
2857 log_notice_errno(r, "Failed to load unit \"%s\", skipping deserialization: %m", unit_name);
2858 if (r == -ENOMEM)
2859 goto finish;
2860 unit_deserialize_skip(f);
2861 continue;
2862 }
2863
2864 r = unit_deserialize(u, f, fds);
2865 if (r < 0) {
2866 log_notice_errno(r, "Failed to deserialize unit \"%s\": %m", unit_name);
2867 if (r == -ENOMEM)
2868 goto finish;
2869 }
2870 }
2871
2872 finish:
2873 if (ferror(f))
2874 r = -EIO;
2875
2876 assert(m->n_reloading > 0);
2877 m->n_reloading--;
2878
2879 return r;
2880 }
2881
2882 int manager_reload(Manager *m) {
2883 int r, q;
2884 _cleanup_fclose_ FILE *f = NULL;
2885 _cleanup_fdset_free_ FDSet *fds = NULL;
2886
2887 assert(m);
2888
2889 r = manager_open_serialization(m, &f);
2890 if (r < 0)
2891 return r;
2892
2893 m->n_reloading++;
2894 bus_manager_send_reloading(m, true);
2895
2896 fds = fdset_new();
2897 if (!fds) {
2898 m->n_reloading--;
2899 return -ENOMEM;
2900 }
2901
2902 r = manager_serialize(m, f, fds, false);
2903 if (r < 0) {
2904 m->n_reloading--;
2905 return r;
2906 }
2907
2908 if (fseeko(f, 0, SEEK_SET) < 0) {
2909 m->n_reloading--;
2910 return -errno;
2911 }
2912
2913 /* From here on there is no way back. */
2914 manager_clear_jobs_and_units(m);
2915 lookup_paths_flush_generator(&m->lookup_paths);
2916 lookup_paths_free(&m->lookup_paths);
2917 dynamic_user_vacuum(m, false);
2918 m->uid_refs = hashmap_free(m->uid_refs);
2919 m->gid_refs = hashmap_free(m->gid_refs);
2920
2921 q = lookup_paths_init(&m->lookup_paths, m->unit_file_scope, 0, NULL);
2922 if (q < 0 && r >= 0)
2923 r = q;
2924
2925 q = manager_run_environment_generators(m);
2926 if (q < 0 && r >= 0)
2927 r = q;
2928
2929 /* Find new unit paths */
2930 q = manager_run_generators(m);
2931 if (q < 0 && r >= 0)
2932 r = q;
2933
2934 lookup_paths_reduce(&m->lookup_paths);
2935 manager_build_unit_path_cache(m);
2936
2937 /* First, enumerate what we can from all config files */
2938 manager_enumerate(m);
2939
2940 /* Second, deserialize our stored data */
2941 q = manager_deserialize(m, f, fds);
2942 if (q < 0) {
2943 log_error_errno(q, "Deserialization failed: %m");
2944
2945 if (r >= 0)
2946 r = q;
2947 }
2948
2949 fclose(f);
2950 f = NULL;
2951
2952 /* Re-register notify_fd as event source */
2953 q = manager_setup_notify(m);
2954 if (q < 0 && r >= 0)
2955 r = q;
2956
2957 q = manager_setup_cgroups_agent(m);
2958 if (q < 0 && r >= 0)
2959 r = q;
2960
2961 q = manager_setup_user_lookup_fd(m);
2962 if (q < 0 && r >= 0)
2963 r = q;
2964
2965 /* Third, fire things up! */
2966 manager_coldplug(m);
2967
2968 /* Release any dynamic users no longer referenced */
2969 dynamic_user_vacuum(m, true);
2970
2971 /* Release any references to UIDs/GIDs no longer referenced, and destroy any IPC owned by them */
2972 manager_vacuum_uid_refs(m);
2973 manager_vacuum_gid_refs(m);
2974
2975 /* Sync current state of bus names with our set of listening units */
2976 if (m->api_bus)
2977 manager_sync_bus_names(m, m->api_bus);
2978
2979 assert(m->n_reloading > 0);
2980 m->n_reloading--;
2981
2982 m->send_reloading_done = true;
2983
2984 return r;
2985 }
2986
2987 void manager_reset_failed(Manager *m) {
2988 Unit *u;
2989 Iterator i;
2990
2991 assert(m);
2992
2993 HASHMAP_FOREACH(u, m->units, i)
2994 unit_reset_failed(u);
2995 }
2996
2997 bool manager_unit_inactive_or_pending(Manager *m, const char *name) {
2998 Unit *u;
2999
3000 assert(m);
3001 assert(name);
3002
3003 /* Returns true if the unit is inactive or going down */
3004 u = manager_get_unit(m, name);
3005 if (!u)
3006 return true;
3007
3008 return unit_inactive_or_pending(u);
3009 }
3010
3011 static void manager_notify_finished(Manager *m) {
3012 char userspace[FORMAT_TIMESPAN_MAX], initrd[FORMAT_TIMESPAN_MAX], kernel[FORMAT_TIMESPAN_MAX], sum[FORMAT_TIMESPAN_MAX];
3013 usec_t firmware_usec, loader_usec, kernel_usec, initrd_usec, userspace_usec, total_usec;
3014
3015 if (m->test_run_flags)
3016 return;
3017
3018 if (MANAGER_IS_SYSTEM(m) && detect_container() <= 0) {
3019
3020 /* Note that m->kernel_usec.monotonic is always at 0,
3021 * and m->firmware_usec.monotonic and
3022 * m->loader_usec.monotonic should be considered
3023 * negative values. */
3024
3025 firmware_usec = m->firmware_timestamp.monotonic - m->loader_timestamp.monotonic;
3026 loader_usec = m->loader_timestamp.monotonic - m->kernel_timestamp.monotonic;
3027 userspace_usec = m->finish_timestamp.monotonic - m->userspace_timestamp.monotonic;
3028 total_usec = m->firmware_timestamp.monotonic + m->finish_timestamp.monotonic;
3029
3030 if (dual_timestamp_is_set(&m->initrd_timestamp)) {
3031
3032 kernel_usec = m->initrd_timestamp.monotonic - m->kernel_timestamp.monotonic;
3033 initrd_usec = m->userspace_timestamp.monotonic - m->initrd_timestamp.monotonic;
3034
3035 log_struct(LOG_INFO,
3036 "MESSAGE_ID=" SD_MESSAGE_STARTUP_FINISHED_STR,
3037 "KERNEL_USEC="USEC_FMT, kernel_usec,
3038 "INITRD_USEC="USEC_FMT, initrd_usec,
3039 "USERSPACE_USEC="USEC_FMT, userspace_usec,
3040 LOG_MESSAGE("Startup finished in %s (kernel) + %s (initrd) + %s (userspace) = %s.",
3041 format_timespan(kernel, sizeof(kernel), kernel_usec, USEC_PER_MSEC),
3042 format_timespan(initrd, sizeof(initrd), initrd_usec, USEC_PER_MSEC),
3043 format_timespan(userspace, sizeof(userspace), userspace_usec, USEC_PER_MSEC),
3044 format_timespan(sum, sizeof(sum), total_usec, USEC_PER_MSEC)),
3045 NULL);
3046 } else {
3047 kernel_usec = m->userspace_timestamp.monotonic - m->kernel_timestamp.monotonic;
3048 initrd_usec = 0;
3049
3050 log_struct(LOG_INFO,
3051 "MESSAGE_ID=" SD_MESSAGE_STARTUP_FINISHED_STR,
3052 "KERNEL_USEC="USEC_FMT, kernel_usec,
3053 "USERSPACE_USEC="USEC_FMT, userspace_usec,
3054 LOG_MESSAGE("Startup finished in %s (kernel) + %s (userspace) = %s.",
3055 format_timespan(kernel, sizeof(kernel), kernel_usec, USEC_PER_MSEC),
3056 format_timespan(userspace, sizeof(userspace), userspace_usec, USEC_PER_MSEC),
3057 format_timespan(sum, sizeof(sum), total_usec, USEC_PER_MSEC)),
3058 NULL);
3059 }
3060 } else {
3061 firmware_usec = loader_usec = initrd_usec = kernel_usec = 0;
3062 total_usec = userspace_usec = m->finish_timestamp.monotonic - m->userspace_timestamp.monotonic;
3063
3064 log_struct(LOG_INFO,
3065 "MESSAGE_ID=" SD_MESSAGE_USER_STARTUP_FINISHED_STR,
3066 "USERSPACE_USEC="USEC_FMT, userspace_usec,
3067 LOG_MESSAGE("Startup finished in %s.",
3068 format_timespan(sum, sizeof(sum), total_usec, USEC_PER_MSEC)),
3069 NULL);
3070 }
3071
3072 bus_manager_send_finished(m, firmware_usec, loader_usec, kernel_usec, initrd_usec, userspace_usec, total_usec);
3073
3074 sd_notifyf(false,
3075 m->ready_sent ? "STATUS=Startup finished in %s."
3076 : "READY=1\n"
3077 "STATUS=Startup finished in %s.",
3078 format_timespan(sum, sizeof(sum), total_usec, USEC_PER_MSEC));
3079 m->ready_sent = true;
3080 }
3081
3082 void manager_check_finished(Manager *m) {
3083 assert(m);
3084
3085 if (MANAGER_IS_RELOADING(m))
3086 return;
3087
3088 /* Verify that we are actually running currently. Initially
3089 * the exit code is set to invalid, and during operation it is
3090 * then set to MANAGER_OK */
3091 if (m->exit_code != MANAGER_OK)
3092 return;
3093
3094 /* For user managers, send out READY=1 as soon as we reach basic.target */
3095 if (MANAGER_IS_USER(m) && !m->ready_sent) {
3096 Unit *u;
3097
3098 u = manager_get_unit(m, SPECIAL_BASIC_TARGET);
3099 if (u && !u->job) {
3100 sd_notifyf(false,
3101 "READY=1\n"
3102 "STATUS=Reached " SPECIAL_BASIC_TARGET ".");
3103 m->ready_sent = true;
3104 }
3105 }
3106
3107 if (hashmap_size(m->jobs) > 0) {
3108 if (m->jobs_in_progress_event_source)
3109 /* Ignore any failure, this is only for feedback */
3110 (void) sd_event_source_set_time(m->jobs_in_progress_event_source, now(CLOCK_MONOTONIC) + JOBS_IN_PROGRESS_WAIT_USEC);
3111
3112 return;
3113 }
3114
3115 manager_flip_auto_status(m, false);
3116
3117 /* Notify Type=idle units that we are done now */
3118 manager_close_idle_pipe(m);
3119
3120 /* Turn off confirm spawn now */
3121 m->confirm_spawn = NULL;
3122
3123 /* No need to update ask password status when we're going non-interactive */
3124 manager_close_ask_password(m);
3125
3126 /* This is no longer the first boot */
3127 manager_set_first_boot(m, false);
3128
3129 if (dual_timestamp_is_set(&m->finish_timestamp))
3130 return;
3131
3132 dual_timestamp_get(&m->finish_timestamp);
3133
3134 manager_notify_finished(m);
3135
3136 manager_invalidate_startup_units(m);
3137 }
3138
3139 static bool generator_path_any(const char* const* paths) {
3140 char **path;
3141 bool found = false;
3142
3143 /* Optimize by skipping the whole process by not creating output directories
3144 * if no generators are found. */
3145 STRV_FOREACH(path, (char**) paths)
3146 if (access(*path, F_OK) == 0)
3147 found = true;
3148 else if (errno != ENOENT)
3149 log_warning_errno(errno, "Failed to open generator directory %s: %m", *path);
3150
3151 return found;
3152 }
3153
3154 static const char* system_env_generator_binary_paths[] = {
3155 "/run/systemd/system-environment-generators",
3156 "/etc/systemd/system-environment-generators",
3157 "/usr/local/lib/systemd/system-environment-generators",
3158 SYSTEM_ENV_GENERATOR_PATH,
3159 NULL
3160 };
3161
3162 static const char* user_env_generator_binary_paths[] = {
3163 "/run/systemd/user-environment-generators",
3164 "/etc/systemd/user-environment-generators",
3165 "/usr/local/lib/systemd/user-environment-generators",
3166 USER_ENV_GENERATOR_PATH,
3167 NULL
3168 };
3169
3170 static int manager_run_environment_generators(Manager *m) {
3171 char **tmp = NULL; /* this is only used in the forked process, no cleanup here */
3172 const char **paths;
3173 void* args[] = {&tmp, &tmp, &m->environment};
3174
3175 if (m->test_run_flags && !(m->test_run_flags & MANAGER_TEST_RUN_ENV_GENERATORS))
3176 return 0;
3177
3178 paths = MANAGER_IS_SYSTEM(m) ? system_env_generator_binary_paths : user_env_generator_binary_paths;
3179
3180 if (!generator_path_any(paths))
3181 return 0;
3182
3183 return execute_directories(paths, DEFAULT_TIMEOUT_USEC, gather_environment, args, NULL);
3184 }
3185
3186 static int manager_run_generators(Manager *m) {
3187 _cleanup_strv_free_ char **paths = NULL;
3188 const char *argv[5];
3189 int r;
3190
3191 assert(m);
3192
3193 if (m->test_run_flags && !(m->test_run_flags & MANAGER_TEST_RUN_GENERATORS))
3194 return 0;
3195
3196 paths = generator_binary_paths(m->unit_file_scope);
3197 if (!paths)
3198 return log_oom();
3199
3200 if (!generator_path_any((const char* const*) paths))
3201 return 0;
3202
3203 r = lookup_paths_mkdir_generator(&m->lookup_paths);
3204 if (r < 0)
3205 goto finish;
3206
3207 argv[0] = NULL; /* Leave this empty, execute_directory() will fill something in */
3208 argv[1] = m->lookup_paths.generator;
3209 argv[2] = m->lookup_paths.generator_early;
3210 argv[3] = m->lookup_paths.generator_late;
3211 argv[4] = NULL;
3212
3213 RUN_WITH_UMASK(0022)
3214 execute_directories((const char* const*) paths, DEFAULT_TIMEOUT_USEC,
3215 NULL, NULL, (char**) argv);
3216
3217 finish:
3218 lookup_paths_trim_generator(&m->lookup_paths);
3219 return r;
3220 }
3221
3222 int manager_environment_add(Manager *m, char **minus, char **plus) {
3223 char **a = NULL, **b = NULL, **l;
3224 assert(m);
3225
3226 l = m->environment;
3227
3228 if (!strv_isempty(minus)) {
3229 a = strv_env_delete(l, 1, minus);
3230 if (!a)
3231 return -ENOMEM;
3232
3233 l = a;
3234 }
3235
3236 if (!strv_isempty(plus)) {
3237 b = strv_env_merge(2, l, plus);
3238 if (!b) {
3239 strv_free(a);
3240 return -ENOMEM;
3241 }
3242
3243 l = b;
3244 }
3245
3246 if (m->environment != l)
3247 strv_free(m->environment);
3248 if (a != l)
3249 strv_free(a);
3250 if (b != l)
3251 strv_free(b);
3252
3253 m->environment = l;
3254 manager_clean_environment(m);
3255 strv_sort(m->environment);
3256
3257 return 0;
3258 }
3259
3260 int manager_set_default_rlimits(Manager *m, struct rlimit **default_rlimit) {
3261 int i;
3262
3263 assert(m);
3264
3265 for (i = 0; i < _RLIMIT_MAX; i++) {
3266 m->rlimit[i] = mfree(m->rlimit[i]);
3267
3268 if (!default_rlimit[i])
3269 continue;
3270
3271 m->rlimit[i] = newdup(struct rlimit, default_rlimit[i], 1);
3272 if (!m->rlimit[i])
3273 return log_oom();
3274 }
3275
3276 return 0;
3277 }
3278
3279 void manager_recheck_journal(Manager *m) {
3280 Unit *u;
3281
3282 assert(m);
3283
3284 if (!MANAGER_IS_SYSTEM(m))
3285 return;
3286
3287 u = manager_get_unit(m, SPECIAL_JOURNALD_SOCKET);
3288 if (u && SOCKET(u)->state != SOCKET_RUNNING) {
3289 log_close_journal();
3290 return;
3291 }
3292
3293 u = manager_get_unit(m, SPECIAL_JOURNALD_SERVICE);
3294 if (u && SERVICE(u)->state != SERVICE_RUNNING) {
3295 log_close_journal();
3296 return;
3297 }
3298
3299 /* Hmm, OK, so the socket is fully up and the service is up
3300 * too, then let's make use of the thing. */
3301 log_open();
3302 }
3303
3304 void manager_set_show_status(Manager *m, ShowStatus mode) {
3305 assert(m);
3306 assert(IN_SET(mode, SHOW_STATUS_AUTO, SHOW_STATUS_NO, SHOW_STATUS_YES, SHOW_STATUS_TEMPORARY));
3307
3308 if (!MANAGER_IS_SYSTEM(m))
3309 return;
3310
3311 if (m->show_status != mode)
3312 log_debug("%s showing of status.",
3313 mode == SHOW_STATUS_NO ? "Disabling" : "Enabling");
3314 m->show_status = mode;
3315
3316 if (mode > 0)
3317 (void) touch("/run/systemd/show-status");
3318 else
3319 (void) unlink("/run/systemd/show-status");
3320 }
3321
3322 static bool manager_get_show_status(Manager *m, StatusType type) {
3323 assert(m);
3324
3325 if (!MANAGER_IS_SYSTEM(m))
3326 return false;
3327
3328 if (m->no_console_output)
3329 return false;
3330
3331 if (!IN_SET(manager_state(m), MANAGER_INITIALIZING, MANAGER_STARTING, MANAGER_STOPPING))
3332 return false;
3333
3334 /* If we cannot find out the status properly, just proceed. */
3335 if (type != STATUS_TYPE_EMERGENCY && manager_check_ask_password(m) > 0)
3336 return false;
3337
3338 if (m->show_status > 0)
3339 return true;
3340
3341 return false;
3342 }
3343
3344 const char *manager_get_confirm_spawn(Manager *m) {
3345 static int last_errno = 0;
3346 const char *vc = m->confirm_spawn;
3347 struct stat st;
3348 int r;
3349
3350 /* Here's the deal: we want to test the validity of the console but don't want
3351 * PID1 to go through the whole console process which might block. But we also
3352 * want to warn the user only once if something is wrong with the console so we
3353 * cannot do the sanity checks after spawning our children. So here we simply do
3354 * really basic tests to hopefully trap common errors.
3355 *
3356 * If the console suddenly disappear at the time our children will really it
3357 * then they will simply fail to acquire it and a positive answer will be
3358 * assumed. New children will fallback to /dev/console though.
3359 *
3360 * Note: TTYs are devices that can come and go any time, and frequently aren't
3361 * available yet during early boot (consider a USB rs232 dongle...). If for any
3362 * reason the configured console is not ready, we fallback to the default
3363 * console. */
3364
3365 if (!vc || path_equal(vc, "/dev/console"))
3366 return vc;
3367
3368 r = stat(vc, &st);
3369 if (r < 0)
3370 goto fail;
3371
3372 if (!S_ISCHR(st.st_mode)) {
3373 errno = ENOTTY;
3374 goto fail;
3375 }
3376
3377 last_errno = 0;
3378 return vc;
3379 fail:
3380 if (last_errno != errno) {
3381 last_errno = errno;
3382 log_warning_errno(errno, "Failed to open %s: %m, using default console", vc);
3383 }
3384 return "/dev/console";
3385 }
3386
3387 void manager_set_first_boot(Manager *m, bool b) {
3388 assert(m);
3389
3390 if (!MANAGER_IS_SYSTEM(m))
3391 return;
3392
3393 if (m->first_boot != (int) b) {
3394 if (b)
3395 (void) touch("/run/systemd/first-boot");
3396 else
3397 (void) unlink("/run/systemd/first-boot");
3398 }
3399
3400 m->first_boot = b;
3401 }
3402
3403 void manager_disable_confirm_spawn(void) {
3404 (void) touch("/run/systemd/confirm_spawn_disabled");
3405 }
3406
3407 bool manager_is_confirm_spawn_disabled(Manager *m) {
3408 if (!m->confirm_spawn)
3409 return true;
3410
3411 return access("/run/systemd/confirm_spawn_disabled", F_OK) >= 0;
3412 }
3413
3414 void manager_status_printf(Manager *m, StatusType type, const char *status, const char *format, ...) {
3415 va_list ap;
3416
3417 /* If m is NULL, assume we're after shutdown and let the messages through. */
3418
3419 if (m && !manager_get_show_status(m, type))
3420 return;
3421
3422 /* XXX We should totally drop the check for ephemeral here
3423 * and thus effectively make 'Type=idle' pointless. */
3424 if (type == STATUS_TYPE_EPHEMERAL && m && m->n_on_console > 0)
3425 return;
3426
3427 va_start(ap, format);
3428 status_vprintf(status, true, type == STATUS_TYPE_EPHEMERAL, format, ap);
3429 va_end(ap);
3430 }
3431
3432 Set *manager_get_units_requiring_mounts_for(Manager *m, const char *path) {
3433 char p[strlen(path)+1];
3434
3435 assert(m);
3436 assert(path);
3437
3438 strcpy(p, path);
3439 path_kill_slashes(p);
3440
3441 return hashmap_get(m->units_requiring_mounts_for, streq(p, "/") ? "" : p);
3442 }
3443
3444 void manager_set_exec_params(Manager *m, ExecParameters *p) {
3445 assert(m);
3446 assert(p);
3447
3448 p->environment = m->environment;
3449 p->confirm_spawn = manager_get_confirm_spawn(m);
3450 p->cgroup_supported = m->cgroup_supported;
3451 p->prefix = m->prefix;
3452
3453 SET_FLAG(p->flags, EXEC_PASS_LOG_UNIT|EXEC_CHOWN_DIRECTORIES, MANAGER_IS_SYSTEM(m));
3454 }
3455
3456 int manager_update_failed_units(Manager *m, Unit *u, bool failed) {
3457 unsigned size;
3458 int r;
3459
3460 assert(m);
3461 assert(u->manager == m);
3462
3463 size = set_size(m->failed_units);
3464
3465 if (failed) {
3466 r = set_ensure_allocated(&m->failed_units, NULL);
3467 if (r < 0)
3468 return log_oom();
3469
3470 if (set_put(m->failed_units, u) < 0)
3471 return log_oom();
3472 } else
3473 (void) set_remove(m->failed_units, u);
3474
3475 if (set_size(m->failed_units) != size)
3476 bus_manager_send_change_signal(m);
3477
3478 return 0;
3479 }
3480
3481 ManagerState manager_state(Manager *m) {
3482 Unit *u;
3483
3484 assert(m);
3485
3486 /* Did we ever finish booting? If not then we are still starting up */
3487 if (!dual_timestamp_is_set(&m->finish_timestamp)) {
3488
3489 u = manager_get_unit(m, SPECIAL_BASIC_TARGET);
3490 if (!u || !UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u)))
3491 return MANAGER_INITIALIZING;
3492
3493 return MANAGER_STARTING;
3494 }
3495
3496 /* Is the special shutdown target queued? If so, we are in shutdown state */
3497 u = manager_get_unit(m, SPECIAL_SHUTDOWN_TARGET);
3498 if (u && u->job && IN_SET(u->job->type, JOB_START, JOB_RESTART, JOB_RELOAD_OR_START))
3499 return MANAGER_STOPPING;
3500
3501 /* Are the rescue or emergency targets active or queued? If so we are in maintenance state */
3502 u = manager_get_unit(m, SPECIAL_RESCUE_TARGET);
3503 if (u && (UNIT_IS_ACTIVE_OR_ACTIVATING(unit_active_state(u)) ||
3504 (u->job && IN_SET(u->job->type, JOB_START, JOB_RESTART, JOB_RELOAD_OR_START))))
3505 return MANAGER_MAINTENANCE;
3506
3507 u = manager_get_unit(m, SPECIAL_EMERGENCY_TARGET);
3508 if (u && (UNIT_IS_ACTIVE_OR_ACTIVATING(unit_active_state(u)) ||
3509 (u->job && IN_SET(u->job->type, JOB_START, JOB_RESTART, JOB_RELOAD_OR_START))))
3510 return MANAGER_MAINTENANCE;
3511
3512 /* Are there any failed units? If so, we are in degraded mode */
3513 if (set_size(m->failed_units) > 0)
3514 return MANAGER_DEGRADED;
3515
3516 return MANAGER_RUNNING;
3517 }
3518
3519 #define DESTROY_IPC_FLAG (UINT32_C(1) << 31)
3520
3521 static void manager_unref_uid_internal(
3522 Manager *m,
3523 Hashmap **uid_refs,
3524 uid_t uid,
3525 bool destroy_now,
3526 int (*_clean_ipc)(uid_t uid)) {
3527
3528 uint32_t c, n;
3529
3530 assert(m);
3531 assert(uid_refs);
3532 assert(uid_is_valid(uid));
3533 assert(_clean_ipc);
3534
3535 /* A generic implementation, covering both manager_unref_uid() and manager_unref_gid(), under the assumption
3536 * that uid_t and gid_t are actually defined the same way, with the same validity rules.
3537 *
3538 * We store a hashmap where the UID/GID is they key and the value is a 32bit reference counter, whose highest
3539 * bit is used as flag for marking UIDs/GIDs whose IPC objects to remove when the last reference to the UID/GID
3540 * is dropped. The flag is set to on, once at least one reference from a unit where RemoveIPC= is set is added
3541 * on a UID/GID. It is reset when the UID's/GID's reference counter drops to 0 again. */
3542
3543 assert_cc(sizeof(uid_t) == sizeof(gid_t));
3544 assert_cc(UID_INVALID == (uid_t) GID_INVALID);
3545
3546 if (uid == 0) /* We don't keep track of root, and will never destroy it */
3547 return;
3548
3549 c = PTR_TO_UINT32(hashmap_get(*uid_refs, UID_TO_PTR(uid)));
3550
3551 n = c & ~DESTROY_IPC_FLAG;
3552 assert(n > 0);
3553 n--;
3554
3555 if (destroy_now && n == 0) {
3556 hashmap_remove(*uid_refs, UID_TO_PTR(uid));
3557
3558 if (c & DESTROY_IPC_FLAG) {
3559 log_debug("%s " UID_FMT " is no longer referenced, cleaning up its IPC.",
3560 _clean_ipc == clean_ipc_by_uid ? "UID" : "GID",
3561 uid);
3562 (void) _clean_ipc(uid);
3563 }
3564 } else {
3565 c = n | (c & DESTROY_IPC_FLAG);
3566 assert_se(hashmap_update(*uid_refs, UID_TO_PTR(uid), UINT32_TO_PTR(c)) >= 0);
3567 }
3568 }
3569
3570 void manager_unref_uid(Manager *m, uid_t uid, bool destroy_now) {
3571 manager_unref_uid_internal(m, &m->uid_refs, uid, destroy_now, clean_ipc_by_uid);
3572 }
3573
3574 void manager_unref_gid(Manager *m, gid_t gid, bool destroy_now) {
3575 manager_unref_uid_internal(m, &m->gid_refs, (uid_t) gid, destroy_now, clean_ipc_by_gid);
3576 }
3577
3578 static int manager_ref_uid_internal(
3579 Manager *m,
3580 Hashmap **uid_refs,
3581 uid_t uid,
3582 bool clean_ipc) {
3583
3584 uint32_t c, n;
3585 int r;
3586
3587 assert(m);
3588 assert(uid_refs);
3589 assert(uid_is_valid(uid));
3590
3591 /* A generic implementation, covering both manager_ref_uid() and manager_ref_gid(), under the assumption
3592 * that uid_t and gid_t are actually defined the same way, with the same validity rules. */
3593
3594 assert_cc(sizeof(uid_t) == sizeof(gid_t));
3595 assert_cc(UID_INVALID == (uid_t) GID_INVALID);
3596
3597 if (uid == 0) /* We don't keep track of root, and will never destroy it */
3598 return 0;
3599
3600 r = hashmap_ensure_allocated(uid_refs, &trivial_hash_ops);
3601 if (r < 0)
3602 return r;
3603
3604 c = PTR_TO_UINT32(hashmap_get(*uid_refs, UID_TO_PTR(uid)));
3605
3606 n = c & ~DESTROY_IPC_FLAG;
3607 n++;
3608
3609 if (n & DESTROY_IPC_FLAG) /* check for overflow */
3610 return -EOVERFLOW;
3611
3612 c = n | (c & DESTROY_IPC_FLAG) | (clean_ipc ? DESTROY_IPC_FLAG : 0);
3613
3614 return hashmap_replace(*uid_refs, UID_TO_PTR(uid), UINT32_TO_PTR(c));
3615 }
3616
3617 int manager_ref_uid(Manager *m, uid_t uid, bool clean_ipc) {
3618 return manager_ref_uid_internal(m, &m->uid_refs, uid, clean_ipc);
3619 }
3620
3621 int manager_ref_gid(Manager *m, gid_t gid, bool clean_ipc) {
3622 return manager_ref_uid_internal(m, &m->gid_refs, (uid_t) gid, clean_ipc);
3623 }
3624
3625 static void manager_vacuum_uid_refs_internal(
3626 Manager *m,
3627 Hashmap **uid_refs,
3628 int (*_clean_ipc)(uid_t uid)) {
3629
3630 Iterator i;
3631 void *p, *k;
3632
3633 assert(m);
3634 assert(uid_refs);
3635 assert(_clean_ipc);
3636
3637 HASHMAP_FOREACH_KEY(p, k, *uid_refs, i) {
3638 uint32_t c, n;
3639 uid_t uid;
3640
3641 uid = PTR_TO_UID(k);
3642 c = PTR_TO_UINT32(p);
3643
3644 n = c & ~DESTROY_IPC_FLAG;
3645 if (n > 0)
3646 continue;
3647
3648 if (c & DESTROY_IPC_FLAG) {
3649 log_debug("Found unreferenced %s " UID_FMT " after reload/reexec. Cleaning up.",
3650 _clean_ipc == clean_ipc_by_uid ? "UID" : "GID",
3651 uid);
3652 (void) _clean_ipc(uid);
3653 }
3654
3655 assert_se(hashmap_remove(*uid_refs, k) == p);
3656 }
3657 }
3658
3659 void manager_vacuum_uid_refs(Manager *m) {
3660 manager_vacuum_uid_refs_internal(m, &m->uid_refs, clean_ipc_by_uid);
3661 }
3662
3663 void manager_vacuum_gid_refs(Manager *m) {
3664 manager_vacuum_uid_refs_internal(m, &m->gid_refs, clean_ipc_by_gid);
3665 }
3666
3667 static void manager_serialize_uid_refs_internal(
3668 Manager *m,
3669 FILE *f,
3670 Hashmap **uid_refs,
3671 const char *field_name) {
3672
3673 Iterator i;
3674 void *p, *k;
3675
3676 assert(m);
3677 assert(f);
3678 assert(uid_refs);
3679 assert(field_name);
3680
3681 /* Serialize the UID reference table. Or actually, just the IPC destruction flag of it, as the actual counter
3682 * of it is better rebuild after a reload/reexec. */
3683
3684 HASHMAP_FOREACH_KEY(p, k, *uid_refs, i) {
3685 uint32_t c;
3686 uid_t uid;
3687
3688 uid = PTR_TO_UID(k);
3689 c = PTR_TO_UINT32(p);
3690
3691 if (!(c & DESTROY_IPC_FLAG))
3692 continue;
3693
3694 fprintf(f, "%s=" UID_FMT "\n", field_name, uid);
3695 }
3696 }
3697
3698 void manager_serialize_uid_refs(Manager *m, FILE *f) {
3699 manager_serialize_uid_refs_internal(m, f, &m->uid_refs, "destroy-ipc-uid");
3700 }
3701
3702 void manager_serialize_gid_refs(Manager *m, FILE *f) {
3703 manager_serialize_uid_refs_internal(m, f, &m->gid_refs, "destroy-ipc-gid");
3704 }
3705
3706 static void manager_deserialize_uid_refs_one_internal(
3707 Manager *m,
3708 Hashmap** uid_refs,
3709 const char *value) {
3710
3711 uid_t uid;
3712 uint32_t c;
3713 int r;
3714
3715 assert(m);
3716 assert(uid_refs);
3717 assert(value);
3718
3719 r = parse_uid(value, &uid);
3720 if (r < 0 || uid == 0) {
3721 log_debug("Unable to parse UID reference serialization");
3722 return;
3723 }
3724
3725 r = hashmap_ensure_allocated(uid_refs, &trivial_hash_ops);
3726 if (r < 0) {
3727 log_oom();
3728 return;
3729 }
3730
3731 c = PTR_TO_UINT32(hashmap_get(*uid_refs, UID_TO_PTR(uid)));
3732 if (c & DESTROY_IPC_FLAG)
3733 return;
3734
3735 c |= DESTROY_IPC_FLAG;
3736
3737 r = hashmap_replace(*uid_refs, UID_TO_PTR(uid), UINT32_TO_PTR(c));
3738 if (r < 0) {
3739 log_debug("Failed to add UID reference entry");
3740 return;
3741 }
3742 }
3743
3744 void manager_deserialize_uid_refs_one(Manager *m, const char *value) {
3745 manager_deserialize_uid_refs_one_internal(m, &m->uid_refs, value);
3746 }
3747
3748 void manager_deserialize_gid_refs_one(Manager *m, const char *value) {
3749 manager_deserialize_uid_refs_one_internal(m, &m->gid_refs, value);
3750 }
3751
3752 int manager_dispatch_user_lookup_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
3753 struct buffer {
3754 uid_t uid;
3755 gid_t gid;
3756 char unit_name[UNIT_NAME_MAX+1];
3757 } _packed_ buffer;
3758
3759 Manager *m = userdata;
3760 ssize_t l;
3761 size_t n;
3762 Unit *u;
3763
3764 assert_se(source);
3765 assert_se(m);
3766
3767 /* Invoked whenever a child process succeeded resolving its user/group to use and sent us the resulting UID/GID
3768 * in a datagram. We parse the datagram here and pass it off to the unit, so that it can add a reference to the
3769 * UID/GID so that it can destroy the UID/GID's IPC objects when the reference counter drops to 0. */
3770
3771 l = recv(fd, &buffer, sizeof(buffer), MSG_DONTWAIT);
3772 if (l < 0) {
3773 if (IN_SET(errno, EINTR, EAGAIN))
3774 return 0;
3775
3776 return log_error_errno(errno, "Failed to read from user lookup fd: %m");
3777 }
3778
3779 if ((size_t) l <= offsetof(struct buffer, unit_name)) {
3780 log_warning("Received too short user lookup message, ignoring.");
3781 return 0;
3782 }
3783
3784 if ((size_t) l > offsetof(struct buffer, unit_name) + UNIT_NAME_MAX) {
3785 log_warning("Received too long user lookup message, ignoring.");
3786 return 0;
3787 }
3788
3789 if (!uid_is_valid(buffer.uid) && !gid_is_valid(buffer.gid)) {
3790 log_warning("Got user lookup message with invalid UID/GID pair, ignoring.");
3791 return 0;
3792 }
3793
3794 n = (size_t) l - offsetof(struct buffer, unit_name);
3795 if (memchr(buffer.unit_name, 0, n)) {
3796 log_warning("Received lookup message with embedded NUL character, ignoring.");
3797 return 0;
3798 }
3799
3800 buffer.unit_name[n] = 0;
3801 u = manager_get_unit(m, buffer.unit_name);
3802 if (!u) {
3803 log_debug("Got user lookup message but unit doesn't exist, ignoring.");
3804 return 0;
3805 }
3806
3807 log_unit_debug(u, "User lookup succeeded: uid=" UID_FMT " gid=" GID_FMT, buffer.uid, buffer.gid);
3808
3809 unit_notify_user_lookup(u, buffer.uid, buffer.gid);
3810 return 0;
3811 }
3812
3813 static const char *const manager_state_table[_MANAGER_STATE_MAX] = {
3814 [MANAGER_INITIALIZING] = "initializing",
3815 [MANAGER_STARTING] = "starting",
3816 [MANAGER_RUNNING] = "running",
3817 [MANAGER_DEGRADED] = "degraded",
3818 [MANAGER_MAINTENANCE] = "maintenance",
3819 [MANAGER_STOPPING] = "stopping",
3820 };
3821
3822 DEFINE_STRING_TABLE_LOOKUP(manager_state, ManagerState);
Unnamed repository; edit this file 'description' to name the repository.