]> 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 #6840 from keszybz/more-docs
[thirdparty/systemd.git] / src / core / manager.c
1 /***
2 This file is part of systemd.
3
4 Copyright 2010 Lennart Poettering
5
6 systemd is free software; you can redistribute it and/or modify it
7 under the terms of the GNU Lesser General Public License as published by
8 the Free Software Foundation; either version 2.1 of the License, or
9 (at your option) any later version.
10
11 systemd is distributed in the hope that it will be useful, but
12 WITHOUT ANY WARRANTY; without even the implied warranty of
13 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 Lesser General Public License for more details.
15
16 You should have received a copy of the GNU Lesser General Public License
17 along with systemd; If not, see <http://www.gnu.org/licenses/>.
18 ***/
19
20 #include <errno.h>
21 #include <fcntl.h>
22 #include <linux/kd.h>
23 #include <signal.h>
24 #include <string.h>
25 #include <sys/epoll.h>
26 #include <sys/inotify.h>
27 #include <sys/ioctl.h>
28 #include <sys/reboot.h>
29 #include <sys/timerfd.h>
30 #include <sys/wait.h>
31 #include <unistd.h>
32
33 #ifdef HAVE_AUDIT
34 #include <libaudit.h>
35 #endif
36
37 #include "sd-daemon.h"
38 #include "sd-messages.h"
39 #include "sd-path.h"
40
41 #include "alloc-util.h"
42 #include "audit-fd.h"
43 #include "boot-timestamps.h"
44 #include "bus-common-errors.h"
45 #include "bus-error.h"
46 #include "bus-kernel.h"
47 #include "bus-util.h"
48 #include "clean-ipc.h"
49 #include "dbus-job.h"
50 #include "dbus-manager.h"
51 #include "dbus-unit.h"
52 #include "dbus.h"
53 #include "dirent-util.h"
54 #include "env-util.h"
55 #include "escape.h"
56 #include "execute.h"
57 #include "exec-util.h"
58 #include "exit-status.h"
59 #include "fd-util.h"
60 #include "fileio.h"
61 #include "fs-util.h"
62 #include "hashmap.h"
63 #include "io-util.h"
64 #include "locale-setup.h"
65 #include "log.h"
66 #include "macro.h"
67 #include "manager.h"
68 #include "missing.h"
69 #include "mkdir.h"
70 #include "parse-util.h"
71 #include "path-lookup.h"
72 #include "path-util.h"
73 #include "process-util.h"
74 #include "ratelimit.h"
75 #include "rm-rf.h"
76 #include "signal-util.h"
77 #include "special.h"
78 #include "stat-util.h"
79 #include "string-table.h"
80 #include "string-util.h"
81 #include "strv.h"
82 #include "terminal-util.h"
83 #include "time-util.h"
84 #include "transaction.h"
85 #include "umask-util.h"
86 #include "unit-name.h"
87 #include "user-util.h"
88 #include "util.h"
89 #include "virt.h"
90 #include "watchdog.h"
91
92 #define NOTIFY_RCVBUF_SIZE (8*1024*1024)
93 #define CGROUPS_AGENT_RCVBUF_SIZE (8*1024*1024)
94
95 /* Initial delay and the interval for printing status messages about running jobs */
96 #define JOBS_IN_PROGRESS_WAIT_USEC (5*USEC_PER_SEC)
97 #define JOBS_IN_PROGRESS_PERIOD_USEC (USEC_PER_SEC / 3)
98 #define JOBS_IN_PROGRESS_PERIOD_DIVISOR 3
99
100 static int manager_dispatch_notify_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata);
101 static int manager_dispatch_cgroups_agent_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata);
102 static int manager_dispatch_signal_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata);
103 static int manager_dispatch_time_change_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata);
104 static int manager_dispatch_idle_pipe_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata);
105 static int manager_dispatch_user_lookup_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata);
106 static int manager_dispatch_jobs_in_progress(sd_event_source *source, usec_t usec, void *userdata);
107 static int manager_dispatch_run_queue(sd_event_source *source, void *userdata);
108 static int manager_run_environment_generators(Manager *m);
109 static int manager_run_generators(Manager *m);
110
111 static void manager_watch_jobs_in_progress(Manager *m) {
112 usec_t next;
113 int r;
114
115 assert(m);
116
117 /* We do not want to show the cylon animation if the user
118 * needs to confirm service executions otherwise confirmation
119 * messages will be screwed by the cylon animation. */
120 if (!manager_is_confirm_spawn_disabled(m))
121 return;
122
123 if (m->jobs_in_progress_event_source)
124 return;
125
126 next = now(CLOCK_MONOTONIC) + JOBS_IN_PROGRESS_WAIT_USEC;
127 r = sd_event_add_time(
128 m->event,
129 &m->jobs_in_progress_event_source,
130 CLOCK_MONOTONIC,
131 next, 0,
132 manager_dispatch_jobs_in_progress, m);
133 if (r < 0)
134 return;
135
136 (void) sd_event_source_set_description(m->jobs_in_progress_event_source, "manager-jobs-in-progress");
137 }
138
139 #define CYLON_BUFFER_EXTRA (2*(sizeof(ANSI_RED)-1) + sizeof(ANSI_HIGHLIGHT_RED)-1 + 2*(sizeof(ANSI_NORMAL)-1))
140
141 static void draw_cylon(char buffer[], size_t buflen, unsigned width, unsigned pos) {
142 char *p = buffer;
143
144 assert(buflen >= CYLON_BUFFER_EXTRA + width + 1);
145 assert(pos <= width+1); /* 0 or width+1 mean that the center light is behind the corner */
146
147 if (pos > 1) {
148 if (pos > 2)
149 p = mempset(p, ' ', pos-2);
150 if (log_get_show_color())
151 p = stpcpy(p, ANSI_RED);
152 *p++ = '*';
153 }
154
155 if (pos > 0 && pos <= width) {
156 if (log_get_show_color())
157 p = stpcpy(p, ANSI_HIGHLIGHT_RED);
158 *p++ = '*';
159 }
160
161 if (log_get_show_color())
162 p = stpcpy(p, ANSI_NORMAL);
163
164 if (pos < width) {
165 if (log_get_show_color())
166 p = stpcpy(p, ANSI_RED);
167 *p++ = '*';
168 if (pos < width-1)
169 p = mempset(p, ' ', width-1-pos);
170 if (log_get_show_color())
171 strcpy(p, ANSI_NORMAL);
172 }
173 }
174
175 void manager_flip_auto_status(Manager *m, bool enable) {
176 assert(m);
177
178 if (enable) {
179 if (m->show_status == SHOW_STATUS_AUTO)
180 manager_set_show_status(m, SHOW_STATUS_TEMPORARY);
181 } else {
182 if (m->show_status == SHOW_STATUS_TEMPORARY)
183 manager_set_show_status(m, SHOW_STATUS_AUTO);
184 }
185 }
186
187 static void manager_print_jobs_in_progress(Manager *m) {
188 _cleanup_free_ char *job_of_n = NULL;
189 Iterator i;
190 Job *j;
191 unsigned counter = 0, print_nr;
192 char cylon[6 + CYLON_BUFFER_EXTRA + 1];
193 unsigned cylon_pos;
194 char time[FORMAT_TIMESPAN_MAX], limit[FORMAT_TIMESPAN_MAX] = "no limit";
195 uint64_t x;
196
197 assert(m);
198 assert(m->n_running_jobs > 0);
199
200 manager_flip_auto_status(m, true);
201
202 print_nr = (m->jobs_in_progress_iteration / JOBS_IN_PROGRESS_PERIOD_DIVISOR) % m->n_running_jobs;
203
204 HASHMAP_FOREACH(j, m->jobs, i)
205 if (j->state == JOB_RUNNING && counter++ == print_nr)
206 break;
207
208 /* m->n_running_jobs must be consistent with the contents of m->jobs,
209 * so the above loop must have succeeded in finding j. */
210 assert(counter == print_nr + 1);
211 assert(j);
212
213 cylon_pos = m->jobs_in_progress_iteration % 14;
214 if (cylon_pos >= 8)
215 cylon_pos = 14 - cylon_pos;
216 draw_cylon(cylon, sizeof(cylon), 6, cylon_pos);
217
218 m->jobs_in_progress_iteration++;
219
220 if (m->n_running_jobs > 1) {
221 if (asprintf(&job_of_n, "(%u of %u) ", counter, m->n_running_jobs) < 0)
222 job_of_n = NULL;
223 }
224
225 format_timespan(time, sizeof(time), now(CLOCK_MONOTONIC) - j->begin_usec, 1*USEC_PER_SEC);
226 if (job_get_timeout(j, &x) > 0)
227 format_timespan(limit, sizeof(limit), x - j->begin_usec, 1*USEC_PER_SEC);
228
229 manager_status_printf(m, STATUS_TYPE_EPHEMERAL, cylon,
230 "%sA %s job is running for %s (%s / %s)",
231 strempty(job_of_n),
232 job_type_to_string(j->type),
233 unit_description(j->unit),
234 time, limit);
235 }
236
237 static int have_ask_password(void) {
238 _cleanup_closedir_ DIR *dir;
239 struct dirent *de;
240
241 dir = opendir("/run/systemd/ask-password");
242 if (!dir) {
243 if (errno == ENOENT)
244 return false;
245 else
246 return -errno;
247 }
248
249 FOREACH_DIRENT_ALL(de, dir, return -errno) {
250 if (startswith(de->d_name, "ask."))
251 return true;
252 }
253 return false;
254 }
255
256 static int manager_dispatch_ask_password_fd(sd_event_source *source,
257 int fd, uint32_t revents, void *userdata) {
258 Manager *m = userdata;
259
260 assert(m);
261
262 flush_fd(fd);
263
264 m->have_ask_password = have_ask_password();
265 if (m->have_ask_password < 0)
266 /* Log error but continue. Negative have_ask_password
267 * is treated as unknown status. */
268 log_error_errno(m->have_ask_password, "Failed to list /run/systemd/ask-password: %m");
269
270 return 0;
271 }
272
273 static void manager_close_ask_password(Manager *m) {
274 assert(m);
275
276 m->ask_password_event_source = sd_event_source_unref(m->ask_password_event_source);
277 m->ask_password_inotify_fd = safe_close(m->ask_password_inotify_fd);
278 m->have_ask_password = -EINVAL;
279 }
280
281 static int manager_check_ask_password(Manager *m) {
282 int r;
283
284 assert(m);
285
286 if (!m->ask_password_event_source) {
287 assert(m->ask_password_inotify_fd < 0);
288
289 mkdir_p_label("/run/systemd/ask-password", 0755);
290
291 m->ask_password_inotify_fd = inotify_init1(IN_NONBLOCK|IN_CLOEXEC);
292 if (m->ask_password_inotify_fd < 0)
293 return log_error_errno(errno, "inotify_init1() failed: %m");
294
295 if (inotify_add_watch(m->ask_password_inotify_fd, "/run/systemd/ask-password", IN_CREATE|IN_DELETE|IN_MOVE) < 0) {
296 log_error_errno(errno, "Failed to add watch on /run/systemd/ask-password: %m");
297 manager_close_ask_password(m);
298 return -errno;
299 }
300
301 r = sd_event_add_io(m->event, &m->ask_password_event_source,
302 m->ask_password_inotify_fd, EPOLLIN,
303 manager_dispatch_ask_password_fd, m);
304 if (r < 0) {
305 log_error_errno(errno, "Failed to add event source for /run/systemd/ask-password: %m");
306 manager_close_ask_password(m);
307 return -errno;
308 }
309
310 (void) sd_event_source_set_description(m->ask_password_event_source, "manager-ask-password");
311
312 /* Queries might have been added meanwhile... */
313 manager_dispatch_ask_password_fd(m->ask_password_event_source,
314 m->ask_password_inotify_fd, EPOLLIN, m);
315 }
316
317 return m->have_ask_password;
318 }
319
320 static int manager_watch_idle_pipe(Manager *m) {
321 int r;
322
323 assert(m);
324
325 if (m->idle_pipe_event_source)
326 return 0;
327
328 if (m->idle_pipe[2] < 0)
329 return 0;
330
331 r = sd_event_add_io(m->event, &m->idle_pipe_event_source, m->idle_pipe[2], EPOLLIN, manager_dispatch_idle_pipe_fd, m);
332 if (r < 0)
333 return log_error_errno(r, "Failed to watch idle pipe: %m");
334
335 (void) sd_event_source_set_description(m->idle_pipe_event_source, "manager-idle-pipe");
336
337 return 0;
338 }
339
340 static void manager_close_idle_pipe(Manager *m) {
341 assert(m);
342
343 m->idle_pipe_event_source = sd_event_source_unref(m->idle_pipe_event_source);
344
345 safe_close_pair(m->idle_pipe);
346 safe_close_pair(m->idle_pipe + 2);
347 }
348
349 static int manager_setup_time_change(Manager *m) {
350 int r;
351
352 /* We only care for the cancellation event, hence we set the
353 * timeout to the latest possible value. */
354 struct itimerspec its = {
355 .it_value.tv_sec = TIME_T_MAX,
356 };
357
358 assert(m);
359 assert_cc(sizeof(time_t) == sizeof(TIME_T_MAX));
360
361 if (m->test_run)
362 return 0;
363
364 /* Uses TFD_TIMER_CANCEL_ON_SET to get notifications whenever
365 * CLOCK_REALTIME makes a jump relative to CLOCK_MONOTONIC */
366
367 m->time_change_fd = timerfd_create(CLOCK_REALTIME, TFD_NONBLOCK|TFD_CLOEXEC);
368 if (m->time_change_fd < 0)
369 return log_error_errno(errno, "Failed to create timerfd: %m");
370
371 if (timerfd_settime(m->time_change_fd, TFD_TIMER_ABSTIME|TFD_TIMER_CANCEL_ON_SET, &its, NULL) < 0) {
372 log_debug_errno(errno, "Failed to set up TFD_TIMER_CANCEL_ON_SET, ignoring: %m");
373 m->time_change_fd = safe_close(m->time_change_fd);
374 return 0;
375 }
376
377 r = sd_event_add_io(m->event, &m->time_change_event_source, m->time_change_fd, EPOLLIN, manager_dispatch_time_change_fd, m);
378 if (r < 0)
379 return log_error_errno(r, "Failed to create time change event source: %m");
380
381 (void) sd_event_source_set_description(m->time_change_event_source, "manager-time-change");
382
383 log_debug("Set up TFD_TIMER_CANCEL_ON_SET timerfd.");
384
385 return 0;
386 }
387
388 static int enable_special_signals(Manager *m) {
389 _cleanup_close_ int fd = -1;
390
391 assert(m);
392
393 if (m->test_run)
394 return 0;
395
396 /* Enable that we get SIGINT on control-alt-del. In containers
397 * this will fail with EPERM (older) or EINVAL (newer), so
398 * ignore that. */
399 if (reboot(RB_DISABLE_CAD) < 0 && errno != EPERM && errno != EINVAL)
400 log_warning_errno(errno, "Failed to enable ctrl-alt-del handling: %m");
401
402 fd = open_terminal("/dev/tty0", O_RDWR|O_NOCTTY|O_CLOEXEC);
403 if (fd < 0) {
404 /* Support systems without virtual console */
405 if (fd != -ENOENT)
406 log_warning_errno(errno, "Failed to open /dev/tty0: %m");
407 } else {
408 /* Enable that we get SIGWINCH on kbrequest */
409 if (ioctl(fd, KDSIGACCEPT, SIGWINCH) < 0)
410 log_warning_errno(errno, "Failed to enable kbrequest handling: %m");
411 }
412
413 return 0;
414 }
415
416 static int manager_setup_signals(Manager *m) {
417 struct sigaction sa = {
418 .sa_handler = SIG_DFL,
419 .sa_flags = SA_NOCLDSTOP|SA_RESTART,
420 };
421 sigset_t mask;
422 int r;
423
424 assert(m);
425
426 assert_se(sigaction(SIGCHLD, &sa, NULL) == 0);
427
428 /* We make liberal use of realtime signals here. On
429 * Linux/glibc we have 30 of them (with the exception of Linux
430 * on hppa, see below), between SIGRTMIN+0 ... SIGRTMIN+30
431 * (aka SIGRTMAX). */
432
433 assert_se(sigemptyset(&mask) == 0);
434 sigset_add_many(&mask,
435 SIGCHLD, /* Child died */
436 SIGTERM, /* Reexecute daemon */
437 SIGHUP, /* Reload configuration */
438 SIGUSR1, /* systemd/upstart: reconnect to D-Bus */
439 SIGUSR2, /* systemd: dump status */
440 SIGINT, /* Kernel sends us this on control-alt-del */
441 SIGWINCH, /* Kernel sends us this on kbrequest (alt-arrowup) */
442 SIGPWR, /* Some kernel drivers and upsd send us this on power failure */
443
444 SIGRTMIN+0, /* systemd: start default.target */
445 SIGRTMIN+1, /* systemd: isolate rescue.target */
446 SIGRTMIN+2, /* systemd: isolate emergency.target */
447 SIGRTMIN+3, /* systemd: start halt.target */
448 SIGRTMIN+4, /* systemd: start poweroff.target */
449 SIGRTMIN+5, /* systemd: start reboot.target */
450 SIGRTMIN+6, /* systemd: start kexec.target */
451
452 /* ... space for more special targets ... */
453
454 SIGRTMIN+13, /* systemd: Immediate halt */
455 SIGRTMIN+14, /* systemd: Immediate poweroff */
456 SIGRTMIN+15, /* systemd: Immediate reboot */
457 SIGRTMIN+16, /* systemd: Immediate kexec */
458
459 /* ... space for more immediate system state changes ... */
460
461 SIGRTMIN+20, /* systemd: enable status messages */
462 SIGRTMIN+21, /* systemd: disable status messages */
463 SIGRTMIN+22, /* systemd: set log level to LOG_DEBUG */
464 SIGRTMIN+23, /* systemd: set log level to LOG_INFO */
465 SIGRTMIN+24, /* systemd: Immediate exit (--user only) */
466
467 /* .. one free signal here ... */
468
469 #if !defined(__hppa64__) && !defined(__hppa__)
470 /* Apparently Linux on hppa has fewer RT
471 * signals (SIGRTMAX is SIGRTMIN+25 there),
472 * hence let's not try to make use of them
473 * here. Since these commands are accessible
474 * by different means and only really a safety
475 * net, the missing functionality on hppa
476 * shouldn't matter. */
477
478 SIGRTMIN+26, /* systemd: set log target to journal-or-kmsg */
479 SIGRTMIN+27, /* systemd: set log target to console */
480 SIGRTMIN+28, /* systemd: set log target to kmsg */
481 SIGRTMIN+29, /* systemd: set log target to syslog-or-kmsg (obsolete) */
482
483 /* ... one free signal here SIGRTMIN+30 ... */
484 #endif
485 -1);
486 assert_se(sigprocmask(SIG_SETMASK, &mask, NULL) == 0);
487
488 m->signal_fd = signalfd(-1, &mask, SFD_NONBLOCK|SFD_CLOEXEC);
489 if (m->signal_fd < 0)
490 return -errno;
491
492 r = sd_event_add_io(m->event, &m->signal_event_source, m->signal_fd, EPOLLIN, manager_dispatch_signal_fd, m);
493 if (r < 0)
494 return r;
495
496 (void) sd_event_source_set_description(m->signal_event_source, "manager-signal");
497
498 /* Process signals a bit earlier than the rest of things, but later than notify_fd processing, so that the
499 * notify processing can still figure out to which process/service a message belongs, before we reap the
500 * process. Also, process this before handling cgroup notifications, so that we always collect child exit
501 * status information before detecting that there's no process in a cgroup. */
502 r = sd_event_source_set_priority(m->signal_event_source, SD_EVENT_PRIORITY_NORMAL-6);
503 if (r < 0)
504 return r;
505
506 if (MANAGER_IS_SYSTEM(m))
507 return enable_special_signals(m);
508
509 return 0;
510 }
511
512 static void manager_clean_environment(Manager *m) {
513 assert(m);
514
515 /* Let's remove some environment variables that we
516 * need ourselves to communicate with our clients */
517 strv_env_unset_many(
518 m->environment,
519 "NOTIFY_SOCKET",
520 "MAINPID",
521 "MANAGERPID",
522 "LISTEN_PID",
523 "LISTEN_FDS",
524 "LISTEN_FDNAMES",
525 "WATCHDOG_PID",
526 "WATCHDOG_USEC",
527 "INVOCATION_ID",
528 NULL);
529 }
530
531 static int manager_default_environment(Manager *m) {
532 assert(m);
533
534 if (MANAGER_IS_SYSTEM(m)) {
535 /* The system manager always starts with a clean
536 * environment for its children. It does not import
537 * the kernel's or the parents' exported variables.
538 *
539 * The initial passed environment is untouched to keep
540 * /proc/self/environ valid; it is used for tagging
541 * the init process inside containers. */
542 m->environment = strv_new("PATH=" DEFAULT_PATH,
543 NULL);
544
545 /* Import locale variables LC_*= from configuration */
546 locale_setup(&m->environment);
547 } else
548 /* The user manager passes its own environment
549 * along to its children. */
550 m->environment = strv_copy(environ);
551
552 if (!m->environment)
553 return -ENOMEM;
554
555 manager_clean_environment(m);
556 strv_sort(m->environment);
557
558 return 0;
559 }
560
561 static int manager_setup_prefix(Manager *m) {
562 struct table_entry {
563 uint64_t type;
564 const char *suffix;
565 };
566
567 static const struct table_entry paths_system[_EXEC_DIRECTORY_MAX] = {
568 [EXEC_DIRECTORY_RUNTIME] = { SD_PATH_SYSTEM_RUNTIME, NULL },
569 [EXEC_DIRECTORY_STATE] = { SD_PATH_SYSTEM_STATE_PRIVATE, NULL },
570 [EXEC_DIRECTORY_CACHE] = { SD_PATH_SYSTEM_STATE_CACHE, NULL },
571 [EXEC_DIRECTORY_LOGS] = { SD_PATH_SYSTEM_STATE_LOGS, NULL },
572 [EXEC_DIRECTORY_CONFIGURATION] = { SD_PATH_SYSTEM_CONFIGURATION, NULL },
573 };
574
575 static const struct table_entry paths_user[_EXEC_DIRECTORY_MAX] = {
576 [EXEC_DIRECTORY_RUNTIME] = { SD_PATH_USER_RUNTIME, NULL },
577 [EXEC_DIRECTORY_STATE] = { SD_PATH_USER_CONFIGURATION, NULL },
578 [EXEC_DIRECTORY_CACHE] = { SD_PATH_SYSTEM_STATE_CACHE, NULL },
579 [EXEC_DIRECTORY_LOGS] = { SD_PATH_SYSTEM_CONFIGURATION, "log" },
580 [EXEC_DIRECTORY_CONFIGURATION] = { SD_PATH_SYSTEM_CONFIGURATION, NULL },
581 };
582
583 const struct table_entry *p;
584 ExecDirectoryType i;
585 int r;
586
587 assert(m);
588
589 if (MANAGER_IS_SYSTEM(m))
590 p = paths_system;
591 else
592 p = paths_user;
593
594 for (i = 0; i < _EXEC_DIRECTORY_MAX; i++) {
595 r = sd_path_home(p[i].type, p[i].suffix, &m->prefix[i]);
596 if (r < 0)
597 return r;
598 }
599
600 return 0;
601 }
602
603 int manager_new(UnitFileScope scope, bool test_run, Manager **_m) {
604 Manager *m;
605 int r;
606
607 assert(_m);
608 assert(IN_SET(scope, UNIT_FILE_SYSTEM, UNIT_FILE_USER));
609
610 m = new0(Manager, 1);
611 if (!m)
612 return -ENOMEM;
613
614 m->unit_file_scope = scope;
615 m->exit_code = _MANAGER_EXIT_CODE_INVALID;
616 m->default_timer_accuracy_usec = USEC_PER_MINUTE;
617 m->default_tasks_accounting = true;
618 m->default_tasks_max = UINT64_MAX;
619
620 #ifdef ENABLE_EFI
621 if (MANAGER_IS_SYSTEM(m) && detect_container() <= 0)
622 boot_timestamps(&m->userspace_timestamp, &m->firmware_timestamp, &m->loader_timestamp);
623 #endif
624
625 /* Prepare log fields we can use for structured logging */
626 if (MANAGER_IS_SYSTEM(m)) {
627 m->unit_log_field = "UNIT=";
628 m->unit_log_format_string = "UNIT=%s";
629
630 m->invocation_log_field = "INVOCATION_ID=";
631 m->invocation_log_format_string = "INVOCATION_ID=" SD_ID128_FORMAT_STR;
632 } else {
633 m->unit_log_field = "USER_UNIT=";
634 m->unit_log_format_string = "USER_UNIT=%s";
635
636 m->invocation_log_field = "USER_INVOCATION_ID=";
637 m->invocation_log_format_string = "USER_INVOCATION_ID=" SD_ID128_FORMAT_STR;
638 }
639
640 m->idle_pipe[0] = m->idle_pipe[1] = m->idle_pipe[2] = m->idle_pipe[3] = -1;
641
642 m->pin_cgroupfs_fd = m->notify_fd = m->cgroups_agent_fd = m->signal_fd = m->time_change_fd =
643 m->dev_autofs_fd = m->private_listen_fd = m->cgroup_inotify_fd =
644 m->ask_password_inotify_fd = -1;
645
646 m->user_lookup_fds[0] = m->user_lookup_fds[1] = -1;
647
648 m->current_job_id = 1; /* start as id #1, so that we can leave #0 around as "null-like" value */
649
650 m->have_ask_password = -EINVAL; /* we don't know */
651 m->first_boot = -1;
652
653 m->test_run = test_run;
654
655 /* Reboot immediately if the user hits C-A-D more often than 7x per 2s */
656 RATELIMIT_INIT(m->ctrl_alt_del_ratelimit, 2 * USEC_PER_SEC, 7);
657
658 r = manager_default_environment(m);
659 if (r < 0)
660 goto fail;
661
662 r = hashmap_ensure_allocated(&m->units, &string_hash_ops);
663 if (r < 0)
664 goto fail;
665
666 r = hashmap_ensure_allocated(&m->jobs, NULL);
667 if (r < 0)
668 goto fail;
669
670 r = hashmap_ensure_allocated(&m->cgroup_unit, &string_hash_ops);
671 if (r < 0)
672 goto fail;
673
674 r = hashmap_ensure_allocated(&m->watch_bus, &string_hash_ops);
675 if (r < 0)
676 goto fail;
677
678 r = sd_event_default(&m->event);
679 if (r < 0)
680 goto fail;
681
682 r = sd_event_add_defer(m->event, &m->run_queue_event_source, manager_dispatch_run_queue, m);
683 if (r < 0)
684 goto fail;
685
686 r = sd_event_source_set_priority(m->run_queue_event_source, SD_EVENT_PRIORITY_IDLE);
687 if (r < 0)
688 goto fail;
689
690 r = sd_event_source_set_enabled(m->run_queue_event_source, SD_EVENT_OFF);
691 if (r < 0)
692 goto fail;
693
694 (void) sd_event_source_set_description(m->run_queue_event_source, "manager-run-queue");
695
696 r = manager_setup_signals(m);
697 if (r < 0)
698 goto fail;
699
700 r = manager_setup_cgroup(m);
701 if (r < 0)
702 goto fail;
703
704 r = manager_setup_time_change(m);
705 if (r < 0)
706 goto fail;
707
708 m->udev = udev_new();
709 if (!m->udev) {
710 r = -ENOMEM;
711 goto fail;
712 }
713
714 /* Note that we do not set up the notify fd here. We do that after deserialization,
715 * since they might have gotten serialized across the reexec. */
716
717 m->taint_usr = dir_is_empty("/usr") > 0;
718
719 r = manager_setup_prefix(m);
720 if (r < 0)
721 goto fail;
722
723 *_m = m;
724 return 0;
725
726 fail:
727 manager_free(m);
728 return r;
729 }
730
731 static int manager_setup_notify(Manager *m) {
732 int r;
733
734 if (m->test_run)
735 return 0;
736
737 if (m->notify_fd < 0) {
738 _cleanup_close_ int fd = -1;
739 union sockaddr_union sa = {
740 .sa.sa_family = AF_UNIX,
741 };
742 static const int one = 1;
743
744 /* First free all secondary fields */
745 m->notify_socket = mfree(m->notify_socket);
746 m->notify_event_source = sd_event_source_unref(m->notify_event_source);
747
748 fd = socket(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC|SOCK_NONBLOCK, 0);
749 if (fd < 0)
750 return log_error_errno(errno, "Failed to allocate notification socket: %m");
751
752 fd_inc_rcvbuf(fd, NOTIFY_RCVBUF_SIZE);
753
754 m->notify_socket = strappend(m->prefix[EXEC_DIRECTORY_RUNTIME], "/systemd/notify");
755 if (!m->notify_socket)
756 return log_oom();
757
758 (void) mkdir_parents_label(m->notify_socket, 0755);
759 (void) unlink(m->notify_socket);
760
761 strncpy(sa.un.sun_path, m->notify_socket, sizeof(sa.un.sun_path)-1);
762 r = bind(fd, &sa.sa, SOCKADDR_UN_LEN(sa.un));
763 if (r < 0)
764 return log_error_errno(errno, "bind(%s) failed: %m", sa.un.sun_path);
765
766 r = setsockopt(fd, SOL_SOCKET, SO_PASSCRED, &one, sizeof(one));
767 if (r < 0)
768 return log_error_errno(errno, "SO_PASSCRED failed: %m");
769
770 m->notify_fd = fd;
771 fd = -1;
772
773 log_debug("Using notification socket %s", m->notify_socket);
774 }
775
776 if (!m->notify_event_source) {
777 r = sd_event_add_io(m->event, &m->notify_event_source, m->notify_fd, EPOLLIN, manager_dispatch_notify_fd, m);
778 if (r < 0)
779 return log_error_errno(r, "Failed to allocate notify event source: %m");
780
781 /* Process notification messages a bit earlier than SIGCHLD, so that we can still identify to which
782 * service an exit message belongs. */
783 r = sd_event_source_set_priority(m->notify_event_source, SD_EVENT_PRIORITY_NORMAL-7);
784 if (r < 0)
785 return log_error_errno(r, "Failed to set priority of notify event source: %m");
786
787 (void) sd_event_source_set_description(m->notify_event_source, "manager-notify");
788 }
789
790 return 0;
791 }
792
793 static int manager_setup_cgroups_agent(Manager *m) {
794
795 static const union sockaddr_union sa = {
796 .un.sun_family = AF_UNIX,
797 .un.sun_path = "/run/systemd/cgroups-agent",
798 };
799 int r;
800
801 /* This creates a listening socket we receive cgroups agent messages on. We do not use D-Bus for delivering
802 * these messages from the cgroups agent binary to PID 1, as the cgroups agent binary is very short-living, and
803 * each instance of it needs a new D-Bus connection. Since D-Bus connections are SOCK_STREAM/AF_UNIX, on
804 * overloaded systems the backlog of the D-Bus socket becomes relevant, as not more than the configured number
805 * of D-Bus connections may be queued until the kernel will start dropping further incoming connections,
806 * possibly resulting in lost cgroups agent messages. To avoid this, we'll use a private SOCK_DGRAM/AF_UNIX
807 * socket, where no backlog is relevant as communication may take place without an actual connect() cycle, and
808 * we thus won't lose messages.
809 *
810 * Note that PID 1 will forward the agent message to system bus, so that the user systemd instance may listen
811 * to it. The system instance hence listens on this special socket, but the user instances listen on the system
812 * bus for these messages. */
813
814 if (m->test_run)
815 return 0;
816
817 if (!MANAGER_IS_SYSTEM(m))
818 return 0;
819
820 r = cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER);
821 if (r < 0)
822 return log_error_errno(r, "Failed to determine whether unified cgroups hierarchy is used: %m");
823 if (r > 0) /* We don't need this anymore on the unified hierarchy */
824 return 0;
825
826 if (m->cgroups_agent_fd < 0) {
827 _cleanup_close_ int fd = -1;
828
829 /* First free all secondary fields */
830 m->cgroups_agent_event_source = sd_event_source_unref(m->cgroups_agent_event_source);
831
832 fd = socket(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC|SOCK_NONBLOCK, 0);
833 if (fd < 0)
834 return log_error_errno(errno, "Failed to allocate cgroups agent socket: %m");
835
836 fd_inc_rcvbuf(fd, CGROUPS_AGENT_RCVBUF_SIZE);
837
838 (void) unlink(sa.un.sun_path);
839
840 /* Only allow root to connect to this socket */
841 RUN_WITH_UMASK(0077)
842 r = bind(fd, &sa.sa, SOCKADDR_UN_LEN(sa.un));
843 if (r < 0)
844 return log_error_errno(errno, "bind(%s) failed: %m", sa.un.sun_path);
845
846 m->cgroups_agent_fd = fd;
847 fd = -1;
848 }
849
850 if (!m->cgroups_agent_event_source) {
851 r = sd_event_add_io(m->event, &m->cgroups_agent_event_source, m->cgroups_agent_fd, EPOLLIN, manager_dispatch_cgroups_agent_fd, m);
852 if (r < 0)
853 return log_error_errno(r, "Failed to allocate cgroups agent event source: %m");
854
855 /* Process cgroups notifications early, but after having processed service notification messages or
856 * SIGCHLD signals, so that a cgroup running empty is always just the last safety net of notification,
857 * and we collected the metadata the notification and SIGCHLD stuff offers first. Also see handling of
858 * cgroup inotify for the unified cgroup stuff. */
859 r = sd_event_source_set_priority(m->cgroups_agent_event_source, SD_EVENT_PRIORITY_NORMAL-5);
860 if (r < 0)
861 return log_error_errno(r, "Failed to set priority of cgroups agent event source: %m");
862
863 (void) sd_event_source_set_description(m->cgroups_agent_event_source, "manager-cgroups-agent");
864 }
865
866 return 0;
867 }
868
869 static int manager_setup_user_lookup_fd(Manager *m) {
870 int r;
871
872 assert(m);
873
874 /* Set up the socket pair used for passing UID/GID resolution results from forked off processes to PID
875 * 1. Background: we can't do name lookups (NSS) from PID 1, since it might involve IPC and thus activation,
876 * and we might hence deadlock on ourselves. Hence we do all user/group lookups asynchronously from the forked
877 * off processes right before executing the binaries to start. In order to be able to clean up any IPC objects
878 * created by a unit (see RemoveIPC=) we need to know in PID 1 the used UID/GID of the executed processes,
879 * hence we establish this communication channel so that forked off processes can pass their UID/GID
880 * information back to PID 1. The forked off processes send their resolved UID/GID to PID 1 in a simple
881 * datagram, along with their unit name, so that we can share one communication socket pair among all units for
882 * this purpose.
883 *
884 * You might wonder why we need a communication channel for this that is independent of the usual notification
885 * socket scheme (i.e. $NOTIFY_SOCKET). The primary difference is about trust: data sent via the $NOTIFY_SOCKET
886 * channel is only accepted if it originates from the right unit and if reception was enabled for it. The user
887 * lookup socket OTOH is only accessible by PID 1 and its children until they exec(), and always available.
888 *
889 * Note that this function is called under two circumstances: when we first initialize (in which case we
890 * allocate both the socket pair and the event source to listen on it), and when we deserialize after a reload
891 * (in which case the socket pair already exists but we still need to allocate the event source for it). */
892
893 if (m->user_lookup_fds[0] < 0) {
894
895 /* Free all secondary fields */
896 safe_close_pair(m->user_lookup_fds);
897 m->user_lookup_event_source = sd_event_source_unref(m->user_lookup_event_source);
898
899 if (socketpair(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC, 0, m->user_lookup_fds) < 0)
900 return log_error_errno(errno, "Failed to allocate user lookup socket: %m");
901
902 (void) fd_inc_rcvbuf(m->user_lookup_fds[0], NOTIFY_RCVBUF_SIZE);
903 }
904
905 if (!m->user_lookup_event_source) {
906 r = sd_event_add_io(m->event, &m->user_lookup_event_source, m->user_lookup_fds[0], EPOLLIN, manager_dispatch_user_lookup_fd, m);
907 if (r < 0)
908 return log_error_errno(errno, "Failed to allocate user lookup event source: %m");
909
910 /* Process even earlier than the notify event source, so that we always know first about valid UID/GID
911 * resolutions */
912 r = sd_event_source_set_priority(m->user_lookup_event_source, SD_EVENT_PRIORITY_NORMAL-8);
913 if (r < 0)
914 return log_error_errno(errno, "Failed to set priority ot user lookup event source: %m");
915
916 (void) sd_event_source_set_description(m->user_lookup_event_source, "user-lookup");
917 }
918
919 return 0;
920 }
921
922 static int manager_connect_bus(Manager *m, bool reexecuting) {
923 bool try_bus_connect;
924 Unit *u = NULL;
925
926 assert(m);
927
928 if (m->test_run)
929 return 0;
930
931 u = manager_get_unit(m, SPECIAL_DBUS_SERVICE);
932
933 try_bus_connect =
934 (u && UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u))) &&
935 (reexecuting ||
936 (MANAGER_IS_USER(m) && getenv("DBUS_SESSION_BUS_ADDRESS")));
937
938 /* Try to connect to the buses, if possible. */
939 return bus_init(m, try_bus_connect);
940 }
941
942 static unsigned manager_dispatch_cleanup_queue(Manager *m) {
943 Unit *u;
944 unsigned n = 0;
945
946 assert(m);
947
948 while ((u = m->cleanup_queue)) {
949 assert(u->in_cleanup_queue);
950
951 unit_free(u);
952 n++;
953 }
954
955 return n;
956 }
957
958 enum {
959 GC_OFFSET_IN_PATH, /* This one is on the path we were traveling */
960 GC_OFFSET_UNSURE, /* No clue */
961 GC_OFFSET_GOOD, /* We still need this unit */
962 GC_OFFSET_BAD, /* We don't need this unit anymore */
963 _GC_OFFSET_MAX
964 };
965
966 static void unit_gc_mark_good(Unit *u, unsigned gc_marker) {
967 Iterator i;
968 Unit *other;
969
970 u->gc_marker = gc_marker + GC_OFFSET_GOOD;
971
972 /* Recursively mark referenced units as GOOD as well */
973 SET_FOREACH(other, u->dependencies[UNIT_REFERENCES], i)
974 if (other->gc_marker == gc_marker + GC_OFFSET_UNSURE)
975 unit_gc_mark_good(other, gc_marker);
976 }
977
978 static void unit_gc_sweep(Unit *u, unsigned gc_marker) {
979 Iterator i;
980 Unit *other;
981 bool is_bad;
982
983 assert(u);
984
985 if (u->gc_marker == gc_marker + GC_OFFSET_GOOD ||
986 u->gc_marker == gc_marker + GC_OFFSET_BAD ||
987 u->gc_marker == gc_marker + GC_OFFSET_UNSURE ||
988 u->gc_marker == gc_marker + GC_OFFSET_IN_PATH)
989 return;
990
991 if (u->in_cleanup_queue)
992 goto bad;
993
994 if (unit_check_gc(u))
995 goto good;
996
997 u->gc_marker = gc_marker + GC_OFFSET_IN_PATH;
998
999 is_bad = true;
1000
1001 SET_FOREACH(other, u->dependencies[UNIT_REFERENCED_BY], i) {
1002 unit_gc_sweep(other, gc_marker);
1003
1004 if (other->gc_marker == gc_marker + GC_OFFSET_GOOD)
1005 goto good;
1006
1007 if (other->gc_marker != gc_marker + GC_OFFSET_BAD)
1008 is_bad = false;
1009 }
1010
1011 if (is_bad)
1012 goto bad;
1013
1014 /* We were unable to find anything out about this entry, so
1015 * let's investigate it later */
1016 u->gc_marker = gc_marker + GC_OFFSET_UNSURE;
1017 unit_add_to_gc_queue(u);
1018 return;
1019
1020 bad:
1021 /* We definitely know that this one is not useful anymore, so
1022 * let's mark it for deletion */
1023 u->gc_marker = gc_marker + GC_OFFSET_BAD;
1024 unit_add_to_cleanup_queue(u);
1025 return;
1026
1027 good:
1028 unit_gc_mark_good(u, gc_marker);
1029 }
1030
1031 static unsigned manager_dispatch_gc_unit_queue(Manager *m) {
1032 unsigned n = 0, gc_marker;
1033 Unit *u;
1034
1035 assert(m);
1036
1037 /* log_debug("Running GC..."); */
1038
1039 m->gc_marker += _GC_OFFSET_MAX;
1040 if (m->gc_marker + _GC_OFFSET_MAX <= _GC_OFFSET_MAX)
1041 m->gc_marker = 1;
1042
1043 gc_marker = m->gc_marker;
1044
1045 while ((u = m->gc_unit_queue)) {
1046 assert(u->in_gc_queue);
1047
1048 unit_gc_sweep(u, gc_marker);
1049
1050 LIST_REMOVE(gc_queue, m->gc_unit_queue, u);
1051 u->in_gc_queue = false;
1052
1053 n++;
1054
1055 if (u->gc_marker == gc_marker + GC_OFFSET_BAD ||
1056 u->gc_marker == gc_marker + GC_OFFSET_UNSURE) {
1057 if (u->id)
1058 log_unit_debug(u, "Collecting.");
1059 u->gc_marker = gc_marker + GC_OFFSET_BAD;
1060 unit_add_to_cleanup_queue(u);
1061 }
1062 }
1063
1064 return n;
1065 }
1066
1067 static unsigned manager_dispatch_gc_job_queue(Manager *m) {
1068 unsigned n = 0;
1069 Job *j;
1070
1071 assert(m);
1072
1073 while ((j = m->gc_job_queue)) {
1074 assert(j->in_gc_queue);
1075
1076 LIST_REMOVE(gc_queue, m->gc_job_queue, j);
1077 j->in_gc_queue = false;
1078
1079 n++;
1080
1081 if (job_check_gc(j))
1082 continue;
1083
1084 log_unit_debug(j->unit, "Collecting job.");
1085 (void) job_finish_and_invalidate(j, JOB_COLLECTED, false, false);
1086 }
1087
1088 return n;
1089 }
1090
1091 static void manager_clear_jobs_and_units(Manager *m) {
1092 Unit *u;
1093
1094 assert(m);
1095
1096 while ((u = hashmap_first(m->units)))
1097 unit_free(u);
1098
1099 manager_dispatch_cleanup_queue(m);
1100
1101 assert(!m->load_queue);
1102 assert(!m->run_queue);
1103 assert(!m->dbus_unit_queue);
1104 assert(!m->dbus_job_queue);
1105 assert(!m->cleanup_queue);
1106 assert(!m->gc_unit_queue);
1107 assert(!m->gc_job_queue);
1108
1109 assert(hashmap_isempty(m->jobs));
1110 assert(hashmap_isempty(m->units));
1111
1112 m->n_on_console = 0;
1113 m->n_running_jobs = 0;
1114 }
1115
1116 Manager* manager_free(Manager *m) {
1117 UnitType c;
1118 int i;
1119 ExecDirectoryType dt;
1120
1121 if (!m)
1122 return NULL;
1123
1124 manager_clear_jobs_and_units(m);
1125
1126 for (c = 0; c < _UNIT_TYPE_MAX; c++)
1127 if (unit_vtable[c]->shutdown)
1128 unit_vtable[c]->shutdown(m);
1129
1130 /* If we reexecute ourselves, we keep the root cgroup around */
1131 manager_shutdown_cgroup(m, m->exit_code != MANAGER_REEXECUTE);
1132
1133 lookup_paths_flush_generator(&m->lookup_paths);
1134
1135 bus_done(m);
1136
1137 dynamic_user_vacuum(m, false);
1138 hashmap_free(m->dynamic_users);
1139
1140 hashmap_free(m->units);
1141 hashmap_free(m->units_by_invocation_id);
1142 hashmap_free(m->jobs);
1143 hashmap_free(m->watch_pids1);
1144 hashmap_free(m->watch_pids2);
1145 hashmap_free(m->watch_bus);
1146
1147 set_free(m->startup_units);
1148 set_free(m->failed_units);
1149
1150 sd_event_source_unref(m->signal_event_source);
1151 sd_event_source_unref(m->notify_event_source);
1152 sd_event_source_unref(m->cgroups_agent_event_source);
1153 sd_event_source_unref(m->time_change_event_source);
1154 sd_event_source_unref(m->jobs_in_progress_event_source);
1155 sd_event_source_unref(m->run_queue_event_source);
1156 sd_event_source_unref(m->user_lookup_event_source);
1157
1158 safe_close(m->signal_fd);
1159 safe_close(m->notify_fd);
1160 safe_close(m->cgroups_agent_fd);
1161 safe_close(m->time_change_fd);
1162 safe_close_pair(m->user_lookup_fds);
1163
1164 manager_close_ask_password(m);
1165
1166 manager_close_idle_pipe(m);
1167
1168 udev_unref(m->udev);
1169 sd_event_unref(m->event);
1170
1171 free(m->notify_socket);
1172
1173 lookup_paths_free(&m->lookup_paths);
1174 strv_free(m->environment);
1175
1176 hashmap_free(m->cgroup_unit);
1177 set_free_free(m->unit_path_cache);
1178
1179 free(m->switch_root);
1180 free(m->switch_root_init);
1181
1182 for (i = 0; i < _RLIMIT_MAX; i++)
1183 m->rlimit[i] = mfree(m->rlimit[i]);
1184
1185 assert(hashmap_isempty(m->units_requiring_mounts_for));
1186 hashmap_free(m->units_requiring_mounts_for);
1187
1188 hashmap_free(m->uid_refs);
1189 hashmap_free(m->gid_refs);
1190
1191 for (dt = 0; dt < _EXEC_DIRECTORY_MAX; dt++)
1192 m->prefix[dt] = mfree(m->prefix[dt]);
1193
1194 return mfree(m);
1195 }
1196
1197 void manager_enumerate(Manager *m) {
1198 UnitType c;
1199
1200 assert(m);
1201
1202 /* Let's ask every type to load all units from disk/kernel
1203 * that it might know */
1204 for (c = 0; c < _UNIT_TYPE_MAX; c++) {
1205 if (!unit_type_supported(c)) {
1206 log_debug("Unit type .%s is not supported on this system.", unit_type_to_string(c));
1207 continue;
1208 }
1209
1210 if (!unit_vtable[c]->enumerate)
1211 continue;
1212
1213 unit_vtable[c]->enumerate(m);
1214 }
1215
1216 manager_dispatch_load_queue(m);
1217 }
1218
1219 static void manager_coldplug(Manager *m) {
1220 Iterator i;
1221 Unit *u;
1222 char *k;
1223 int r;
1224
1225 assert(m);
1226
1227 /* Then, let's set up their initial state. */
1228 HASHMAP_FOREACH_KEY(u, k, m->units, i) {
1229
1230 /* ignore aliases */
1231 if (u->id != k)
1232 continue;
1233
1234 r = unit_coldplug(u);
1235 if (r < 0)
1236 log_warning_errno(r, "We couldn't coldplug %s, proceeding anyway: %m", u->id);
1237 }
1238 }
1239
1240 static void manager_build_unit_path_cache(Manager *m) {
1241 char **i;
1242 int r;
1243
1244 assert(m);
1245
1246 set_free_free(m->unit_path_cache);
1247
1248 m->unit_path_cache = set_new(&string_hash_ops);
1249 if (!m->unit_path_cache) {
1250 r = -ENOMEM;
1251 goto fail;
1252 }
1253
1254 /* This simply builds a list of files we know exist, so that
1255 * we don't always have to go to disk */
1256
1257 STRV_FOREACH(i, m->lookup_paths.search_path) {
1258 _cleanup_closedir_ DIR *d = NULL;
1259 struct dirent *de;
1260
1261 d = opendir(*i);
1262 if (!d) {
1263 if (errno != ENOENT)
1264 log_warning_errno(errno, "Failed to open directory %s, ignoring: %m", *i);
1265 continue;
1266 }
1267
1268 FOREACH_DIRENT(de, d, r = -errno; goto fail) {
1269 char *p;
1270
1271 p = strjoin(streq(*i, "/") ? "" : *i, "/", de->d_name);
1272 if (!p) {
1273 r = -ENOMEM;
1274 goto fail;
1275 }
1276
1277 r = set_consume(m->unit_path_cache, p);
1278 if (r < 0)
1279 goto fail;
1280 }
1281 }
1282
1283 return;
1284
1285 fail:
1286 log_warning_errno(r, "Failed to build unit path cache, proceeding without: %m");
1287 m->unit_path_cache = set_free_free(m->unit_path_cache);
1288 }
1289
1290 static void manager_distribute_fds(Manager *m, FDSet *fds) {
1291 Iterator i;
1292 Unit *u;
1293
1294 assert(m);
1295
1296 HASHMAP_FOREACH(u, m->units, i) {
1297
1298 if (fdset_size(fds) <= 0)
1299 break;
1300
1301 if (!UNIT_VTABLE(u)->distribute_fds)
1302 continue;
1303
1304 UNIT_VTABLE(u)->distribute_fds(u, fds);
1305 }
1306 }
1307
1308 int manager_startup(Manager *m, FILE *serialization, FDSet *fds) {
1309 int r, q;
1310
1311 assert(m);
1312
1313 /* If we are running in test mode, we still want to run the generators,
1314 * but we should not touch the real generator directories. */
1315 r = lookup_paths_init(&m->lookup_paths, m->unit_file_scope,
1316 m->test_run ? LOOKUP_PATHS_TEMPORARY_GENERATED : 0,
1317 NULL);
1318 if (r < 0)
1319 return r;
1320
1321 r = manager_run_environment_generators(m);
1322 if (r < 0)
1323 return r;
1324
1325 /* Make sure the transient directory always exists, so that it remains
1326 * in the search path */
1327 r = mkdir_p_label(m->lookup_paths.transient, 0755);
1328 if (r < 0)
1329 return r;
1330
1331 dual_timestamp_get(&m->generators_start_timestamp);
1332 r = manager_run_generators(m);
1333 dual_timestamp_get(&m->generators_finish_timestamp);
1334 if (r < 0)
1335 return r;
1336
1337 if (m->first_boot > 0 &&
1338 m->unit_file_scope == UNIT_FILE_SYSTEM &&
1339 !m->test_run) {
1340
1341 q = unit_file_preset_all(UNIT_FILE_SYSTEM, 0, NULL, UNIT_FILE_PRESET_ENABLE_ONLY, NULL, 0);
1342 if (q < 0)
1343 log_full_errno(q == -EEXIST ? LOG_NOTICE : LOG_WARNING, q, "Failed to populate /etc with preset unit settings, ignoring: %m");
1344 else
1345 log_info("Populated /etc with preset unit settings.");
1346 }
1347
1348 lookup_paths_reduce(&m->lookup_paths);
1349 manager_build_unit_path_cache(m);
1350
1351 /* If we will deserialize make sure that during enumeration
1352 * this is already known, so we increase the counter here
1353 * already */
1354 if (serialization)
1355 m->n_reloading++;
1356
1357 /* First, enumerate what we can from all config files */
1358 dual_timestamp_get(&m->units_load_start_timestamp);
1359 manager_enumerate(m);
1360 dual_timestamp_get(&m->units_load_finish_timestamp);
1361
1362 /* Second, deserialize if there is something to deserialize */
1363 if (serialization) {
1364 r = manager_deserialize(m, serialization, fds);
1365 if (r < 0)
1366 log_error_errno(r, "Deserialization failed: %m");
1367 }
1368
1369 /* Any fds left? Find some unit which wants them. This is
1370 * useful to allow container managers to pass some file
1371 * descriptors to us pre-initialized. This enables
1372 * socket-based activation of entire containers. */
1373 manager_distribute_fds(m, fds);
1374
1375 /* We might have deserialized the notify fd, but if we didn't
1376 * then let's create the bus now */
1377 q = manager_setup_notify(m);
1378 if (q < 0 && r == 0)
1379 r = q;
1380
1381 q = manager_setup_cgroups_agent(m);
1382 if (q < 0 && r == 0)
1383 r = q;
1384
1385 q = manager_setup_user_lookup_fd(m);
1386 if (q < 0 && r == 0)
1387 r = q;
1388
1389 /* Let's connect to the bus now. */
1390 (void) manager_connect_bus(m, !!serialization);
1391
1392 (void) bus_track_coldplug(m, &m->subscribed, false, m->deserialized_subscribed);
1393 m->deserialized_subscribed = strv_free(m->deserialized_subscribed);
1394
1395 /* Third, fire things up! */
1396 manager_coldplug(m);
1397
1398 /* Release any dynamic users no longer referenced */
1399 dynamic_user_vacuum(m, true);
1400
1401 /* Release any references to UIDs/GIDs no longer referenced, and destroy any IPC owned by them */
1402 manager_vacuum_uid_refs(m);
1403 manager_vacuum_gid_refs(m);
1404
1405 if (serialization) {
1406 assert(m->n_reloading > 0);
1407 m->n_reloading--;
1408
1409 /* Let's wait for the UnitNew/JobNew messages being
1410 * sent, before we notify that the reload is
1411 * finished */
1412 m->send_reloading_done = true;
1413 }
1414
1415 return r;
1416 }
1417
1418 int manager_add_job(Manager *m, JobType type, Unit *unit, JobMode mode, sd_bus_error *e, Job **_ret) {
1419 int r;
1420 Transaction *tr;
1421
1422 assert(m);
1423 assert(type < _JOB_TYPE_MAX);
1424 assert(unit);
1425 assert(mode < _JOB_MODE_MAX);
1426
1427 if (mode == JOB_ISOLATE && type != JOB_START)
1428 return sd_bus_error_setf(e, SD_BUS_ERROR_INVALID_ARGS, "Isolate is only valid for start.");
1429
1430 if (mode == JOB_ISOLATE && !unit->allow_isolate)
1431 return sd_bus_error_setf(e, BUS_ERROR_NO_ISOLATION, "Operation refused, unit may not be isolated.");
1432
1433 log_unit_debug(unit, "Trying to enqueue job %s/%s/%s", unit->id, job_type_to_string(type), job_mode_to_string(mode));
1434
1435 type = job_type_collapse(type, unit);
1436
1437 tr = transaction_new(mode == JOB_REPLACE_IRREVERSIBLY);
1438 if (!tr)
1439 return -ENOMEM;
1440
1441 r = transaction_add_job_and_dependencies(tr, type, unit, NULL, true, false,
1442 mode == JOB_IGNORE_DEPENDENCIES || mode == JOB_IGNORE_REQUIREMENTS,
1443 mode == JOB_IGNORE_DEPENDENCIES, e);
1444 if (r < 0)
1445 goto tr_abort;
1446
1447 if (mode == JOB_ISOLATE) {
1448 r = transaction_add_isolate_jobs(tr, m);
1449 if (r < 0)
1450 goto tr_abort;
1451 }
1452
1453 r = transaction_activate(tr, m, mode, e);
1454 if (r < 0)
1455 goto tr_abort;
1456
1457 log_unit_debug(unit,
1458 "Enqueued job %s/%s as %u", unit->id,
1459 job_type_to_string(type), (unsigned) tr->anchor_job->id);
1460
1461 if (_ret)
1462 *_ret = tr->anchor_job;
1463
1464 transaction_free(tr);
1465 return 0;
1466
1467 tr_abort:
1468 transaction_abort(tr);
1469 transaction_free(tr);
1470 return r;
1471 }
1472
1473 int manager_add_job_by_name(Manager *m, JobType type, const char *name, JobMode mode, sd_bus_error *e, Job **ret) {
1474 Unit *unit = NULL; /* just to appease gcc, initialization is not really necessary */
1475 int r;
1476
1477 assert(m);
1478 assert(type < _JOB_TYPE_MAX);
1479 assert(name);
1480 assert(mode < _JOB_MODE_MAX);
1481
1482 r = manager_load_unit(m, name, NULL, NULL, &unit);
1483 if (r < 0)
1484 return r;
1485 assert(unit);
1486
1487 return manager_add_job(m, type, unit, mode, e, ret);
1488 }
1489
1490 int manager_add_job_by_name_and_warn(Manager *m, JobType type, const char *name, JobMode mode, Job **ret) {
1491 _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL;
1492 int r;
1493
1494 assert(m);
1495 assert(type < _JOB_TYPE_MAX);
1496 assert(name);
1497 assert(mode < _JOB_MODE_MAX);
1498
1499 r = manager_add_job_by_name(m, type, name, mode, &error, ret);
1500 if (r < 0)
1501 return log_warning_errno(r, "Failed to enqueue %s job for %s: %s", job_mode_to_string(mode), name, bus_error_message(&error, r));
1502
1503 return r;
1504 }
1505
1506 int manager_propagate_reload(Manager *m, Unit *unit, JobMode mode, sd_bus_error *e) {
1507 int r;
1508 Transaction *tr;
1509
1510 assert(m);
1511 assert(unit);
1512 assert(mode < _JOB_MODE_MAX);
1513 assert(mode != JOB_ISOLATE); /* Isolate is only valid for start */
1514
1515 tr = transaction_new(mode == JOB_REPLACE_IRREVERSIBLY);
1516 if (!tr)
1517 return -ENOMEM;
1518
1519 /* We need an anchor job */
1520 r = transaction_add_job_and_dependencies(tr, JOB_NOP, unit, NULL, false, false, true, true, e);
1521 if (r < 0)
1522 goto tr_abort;
1523
1524 /* Failure in adding individual dependencies is ignored, so this always succeeds. */
1525 transaction_add_propagate_reload_jobs(tr, unit, tr->anchor_job, mode == JOB_IGNORE_DEPENDENCIES, e);
1526
1527 r = transaction_activate(tr, m, mode, e);
1528 if (r < 0)
1529 goto tr_abort;
1530
1531 transaction_free(tr);
1532 return 0;
1533
1534 tr_abort:
1535 transaction_abort(tr);
1536 transaction_free(tr);
1537 return r;
1538 }
1539
1540 Job *manager_get_job(Manager *m, uint32_t id) {
1541 assert(m);
1542
1543 return hashmap_get(m->jobs, UINT32_TO_PTR(id));
1544 }
1545
1546 Unit *manager_get_unit(Manager *m, const char *name) {
1547 assert(m);
1548 assert(name);
1549
1550 return hashmap_get(m->units, name);
1551 }
1552
1553 unsigned manager_dispatch_load_queue(Manager *m) {
1554 Unit *u;
1555 unsigned n = 0;
1556
1557 assert(m);
1558
1559 /* Make sure we are not run recursively */
1560 if (m->dispatching_load_queue)
1561 return 0;
1562
1563 m->dispatching_load_queue = true;
1564
1565 /* Dispatches the load queue. Takes a unit from the queue and
1566 * tries to load its data until the queue is empty */
1567
1568 while ((u = m->load_queue)) {
1569 assert(u->in_load_queue);
1570
1571 unit_load(u);
1572 n++;
1573 }
1574
1575 m->dispatching_load_queue = false;
1576 return n;
1577 }
1578
1579 int manager_load_unit_prepare(
1580 Manager *m,
1581 const char *name,
1582 const char *path,
1583 sd_bus_error *e,
1584 Unit **_ret) {
1585
1586 Unit *ret;
1587 UnitType t;
1588 int r;
1589
1590 assert(m);
1591 assert(name || path);
1592 assert(_ret);
1593
1594 /* This will prepare the unit for loading, but not actually
1595 * load anything from disk. */
1596
1597 if (path && !is_path(path))
1598 return sd_bus_error_setf(e, SD_BUS_ERROR_INVALID_ARGS, "Path %s is not absolute.", path);
1599
1600 if (!name)
1601 name = basename(path);
1602
1603 t = unit_name_to_type(name);
1604
1605 if (t == _UNIT_TYPE_INVALID || !unit_name_is_valid(name, UNIT_NAME_PLAIN|UNIT_NAME_INSTANCE)) {
1606 if (unit_name_is_valid(name, UNIT_NAME_TEMPLATE))
1607 return sd_bus_error_setf(e, SD_BUS_ERROR_INVALID_ARGS, "Unit name %s is missing the instance name.", name);
1608
1609 return sd_bus_error_setf(e, SD_BUS_ERROR_INVALID_ARGS, "Unit name %s is not valid.", name);
1610 }
1611
1612 ret = manager_get_unit(m, name);
1613 if (ret) {
1614 *_ret = ret;
1615 return 1;
1616 }
1617
1618 ret = unit_new(m, unit_vtable[t]->object_size);
1619 if (!ret)
1620 return -ENOMEM;
1621
1622 if (path) {
1623 ret->fragment_path = strdup(path);
1624 if (!ret->fragment_path) {
1625 unit_free(ret);
1626 return -ENOMEM;
1627 }
1628 }
1629
1630 r = unit_add_name(ret, name);
1631 if (r < 0) {
1632 unit_free(ret);
1633 return r;
1634 }
1635
1636 unit_add_to_load_queue(ret);
1637 unit_add_to_dbus_queue(ret);
1638 unit_add_to_gc_queue(ret);
1639
1640 *_ret = ret;
1641
1642 return 0;
1643 }
1644
1645 int manager_load_unit(
1646 Manager *m,
1647 const char *name,
1648 const char *path,
1649 sd_bus_error *e,
1650 Unit **_ret) {
1651
1652 int r;
1653
1654 assert(m);
1655 assert(_ret);
1656
1657 /* This will load the service information files, but not actually
1658 * start any services or anything. */
1659
1660 r = manager_load_unit_prepare(m, name, path, e, _ret);
1661 if (r != 0)
1662 return r;
1663
1664 manager_dispatch_load_queue(m);
1665
1666 *_ret = unit_follow_merge(*_ret);
1667
1668 return 0;
1669 }
1670
1671 void manager_dump_jobs(Manager *s, FILE *f, const char *prefix) {
1672 Iterator i;
1673 Job *j;
1674
1675 assert(s);
1676 assert(f);
1677
1678 HASHMAP_FOREACH(j, s->jobs, i)
1679 job_dump(j, f, prefix);
1680 }
1681
1682 void manager_dump_units(Manager *s, FILE *f, const char *prefix) {
1683 Iterator i;
1684 Unit *u;
1685 const char *t;
1686
1687 assert(s);
1688 assert(f);
1689
1690 HASHMAP_FOREACH_KEY(u, t, s->units, i)
1691 if (u->id == t)
1692 unit_dump(u, f, prefix);
1693 }
1694
1695 void manager_clear_jobs(Manager *m) {
1696 Job *j;
1697
1698 assert(m);
1699
1700 while ((j = hashmap_first(m->jobs)))
1701 /* No need to recurse. We're cancelling all jobs. */
1702 job_finish_and_invalidate(j, JOB_CANCELED, false, false);
1703 }
1704
1705 static int manager_dispatch_run_queue(sd_event_source *source, void *userdata) {
1706 Manager *m = userdata;
1707 Job *j;
1708
1709 assert(source);
1710 assert(m);
1711
1712 while ((j = m->run_queue)) {
1713 assert(j->installed);
1714 assert(j->in_run_queue);
1715
1716 job_run_and_invalidate(j);
1717 }
1718
1719 if (m->n_running_jobs > 0)
1720 manager_watch_jobs_in_progress(m);
1721
1722 if (m->n_on_console > 0)
1723 manager_watch_idle_pipe(m);
1724
1725 return 1;
1726 }
1727
1728 static unsigned manager_dispatch_dbus_queue(Manager *m) {
1729 Job *j;
1730 Unit *u;
1731 unsigned n = 0;
1732
1733 assert(m);
1734
1735 if (m->dispatching_dbus_queue)
1736 return 0;
1737
1738 m->dispatching_dbus_queue = true;
1739
1740 while ((u = m->dbus_unit_queue)) {
1741 assert(u->in_dbus_queue);
1742
1743 bus_unit_send_change_signal(u);
1744 n++;
1745 }
1746
1747 while ((j = m->dbus_job_queue)) {
1748 assert(j->in_dbus_queue);
1749
1750 bus_job_send_change_signal(j);
1751 n++;
1752 }
1753
1754 m->dispatching_dbus_queue = false;
1755
1756 if (m->send_reloading_done) {
1757 m->send_reloading_done = false;
1758
1759 bus_manager_send_reloading(m, false);
1760 }
1761
1762 if (m->queued_message)
1763 bus_send_queued_message(m);
1764
1765 return n;
1766 }
1767
1768 static int manager_dispatch_cgroups_agent_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
1769 Manager *m = userdata;
1770 char buf[PATH_MAX+1];
1771 ssize_t n;
1772
1773 n = recv(fd, buf, sizeof(buf), 0);
1774 if (n < 0)
1775 return log_error_errno(errno, "Failed to read cgroups agent message: %m");
1776 if (n == 0) {
1777 log_error("Got zero-length cgroups agent message, ignoring.");
1778 return 0;
1779 }
1780 if ((size_t) n >= sizeof(buf)) {
1781 log_error("Got overly long cgroups agent message, ignoring.");
1782 return 0;
1783 }
1784
1785 if (memchr(buf, 0, n)) {
1786 log_error("Got cgroups agent message with embedded NUL byte, ignoring.");
1787 return 0;
1788 }
1789 buf[n] = 0;
1790
1791 manager_notify_cgroup_empty(m, buf);
1792 (void) bus_forward_agent_released(m, buf);
1793
1794 return 0;
1795 }
1796
1797 static void manager_invoke_notify_message(Manager *m, Unit *u, pid_t pid, const char *buf, FDSet *fds) {
1798 _cleanup_strv_free_ char **tags = NULL;
1799
1800 assert(m);
1801 assert(u);
1802 assert(buf);
1803
1804 tags = strv_split(buf, "\n\r");
1805 if (!tags) {
1806 log_oom();
1807 return;
1808 }
1809
1810 if (UNIT_VTABLE(u)->notify_message)
1811 UNIT_VTABLE(u)->notify_message(u, pid, tags, fds);
1812 else if (_unlikely_(log_get_max_level() >= LOG_DEBUG)) {
1813 _cleanup_free_ char *x = NULL, *y = NULL;
1814
1815 x = cescape(buf);
1816 if (x)
1817 y = ellipsize(x, 20, 90);
1818 log_unit_debug(u, "Got notification message \"%s\", ignoring.", strnull(y));
1819 }
1820 }
1821
1822 static int manager_dispatch_notify_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
1823
1824 _cleanup_fdset_free_ FDSet *fds = NULL;
1825 Manager *m = userdata;
1826 char buf[NOTIFY_BUFFER_MAX+1];
1827 struct iovec iovec = {
1828 .iov_base = buf,
1829 .iov_len = sizeof(buf)-1,
1830 };
1831 union {
1832 struct cmsghdr cmsghdr;
1833 uint8_t buf[CMSG_SPACE(sizeof(struct ucred)) +
1834 CMSG_SPACE(sizeof(int) * NOTIFY_FD_MAX)];
1835 } control = {};
1836 struct msghdr msghdr = {
1837 .msg_iov = &iovec,
1838 .msg_iovlen = 1,
1839 .msg_control = &control,
1840 .msg_controllen = sizeof(control),
1841 };
1842
1843 struct cmsghdr *cmsg;
1844 struct ucred *ucred = NULL;
1845 Unit *u1, *u2, *u3;
1846 int r, *fd_array = NULL;
1847 unsigned n_fds = 0;
1848 ssize_t n;
1849
1850 assert(m);
1851 assert(m->notify_fd == fd);
1852
1853 if (revents != EPOLLIN) {
1854 log_warning("Got unexpected poll event for notify fd.");
1855 return 0;
1856 }
1857
1858 n = recvmsg(m->notify_fd, &msghdr, MSG_DONTWAIT|MSG_CMSG_CLOEXEC|MSG_TRUNC);
1859 if (n < 0) {
1860 if (IN_SET(errno, EAGAIN, EINTR))
1861 return 0; /* Spurious wakeup, try again */
1862
1863 /* If this is any other, real error, then let's stop processing this socket. This of course means we
1864 * won't take notification messages anymore, but that's still better than busy looping around this:
1865 * being woken up over and over again but being unable to actually read the message off the socket. */
1866 return log_error_errno(errno, "Failed to receive notification message: %m");
1867 }
1868
1869 CMSG_FOREACH(cmsg, &msghdr) {
1870 if (cmsg->cmsg_level == SOL_SOCKET && cmsg->cmsg_type == SCM_RIGHTS) {
1871
1872 fd_array = (int*) CMSG_DATA(cmsg);
1873 n_fds = (cmsg->cmsg_len - CMSG_LEN(0)) / sizeof(int);
1874
1875 } else if (cmsg->cmsg_level == SOL_SOCKET &&
1876 cmsg->cmsg_type == SCM_CREDENTIALS &&
1877 cmsg->cmsg_len == CMSG_LEN(sizeof(struct ucred))) {
1878
1879 ucred = (struct ucred*) CMSG_DATA(cmsg);
1880 }
1881 }
1882
1883 if (n_fds > 0) {
1884 assert(fd_array);
1885
1886 r = fdset_new_array(&fds, fd_array, n_fds);
1887 if (r < 0) {
1888 close_many(fd_array, n_fds);
1889 log_oom();
1890 return 0;
1891 }
1892 }
1893
1894 if (!ucred || ucred->pid <= 0) {
1895 log_warning("Received notify message without valid credentials. Ignoring.");
1896 return 0;
1897 }
1898
1899 if ((size_t) n >= sizeof(buf) || (msghdr.msg_flags & MSG_TRUNC)) {
1900 log_warning("Received notify message exceeded maximum size. Ignoring.");
1901 return 0;
1902 }
1903
1904 /* As extra safety check, let's make sure the string we get doesn't contain embedded NUL bytes. We permit one
1905 * trailing NUL byte in the message, but don't expect it. */
1906 if (n > 1 && memchr(buf, 0, n-1)) {
1907 log_warning("Received notify message with embedded NUL bytes. Ignoring.");
1908 return 0;
1909 }
1910
1911 /* Make sure it's NUL-terminated. */
1912 buf[n] = 0;
1913
1914 /* Notify every unit that might be interested, but try
1915 * to avoid notifying the same one multiple times. */
1916 u1 = manager_get_unit_by_pid_cgroup(m, ucred->pid);
1917 if (u1)
1918 manager_invoke_notify_message(m, u1, ucred->pid, buf, fds);
1919
1920 u2 = hashmap_get(m->watch_pids1, PID_TO_PTR(ucred->pid));
1921 if (u2 && u2 != u1)
1922 manager_invoke_notify_message(m, u2, ucred->pid, buf, fds);
1923
1924 u3 = hashmap_get(m->watch_pids2, PID_TO_PTR(ucred->pid));
1925 if (u3 && u3 != u2 && u3 != u1)
1926 manager_invoke_notify_message(m, u3, ucred->pid, buf, fds);
1927
1928 if (!u1 && !u2 && !u3)
1929 log_warning("Cannot find unit for notify message of PID "PID_FMT".", ucred->pid);
1930
1931 if (fdset_size(fds) > 0)
1932 log_warning("Got extra auxiliary fds with notification message, closing them.");
1933
1934 return 0;
1935 }
1936
1937 static void invoke_sigchld_event(Manager *m, Unit *u, const siginfo_t *si) {
1938 uint64_t iteration;
1939
1940 assert(m);
1941 assert(u);
1942 assert(si);
1943
1944 sd_event_get_iteration(m->event, &iteration);
1945
1946 log_unit_debug(u, "Child "PID_FMT" belongs to %s", si->si_pid, u->id);
1947
1948 unit_unwatch_pid(u, si->si_pid);
1949
1950 if (UNIT_VTABLE(u)->sigchld_event) {
1951 if (set_size(u->pids) <= 1 ||
1952 iteration != u->sigchldgen ||
1953 unit_main_pid(u) == si->si_pid ||
1954 unit_control_pid(u) == si->si_pid) {
1955 UNIT_VTABLE(u)->sigchld_event(u, si->si_pid, si->si_code, si->si_status);
1956 u->sigchldgen = iteration;
1957 } else
1958 log_debug("%s already issued a sigchld this iteration %" PRIu64 ", skipping. Pids still being watched %d", u->id, iteration, set_size(u->pids));
1959 }
1960 }
1961
1962 static int manager_dispatch_sigchld(Manager *m) {
1963 assert(m);
1964
1965 for (;;) {
1966 siginfo_t si = {};
1967
1968 /* First we call waitd() for a PID and do not reap the
1969 * zombie. That way we can still access /proc/$PID for
1970 * it while it is a zombie. */
1971 if (waitid(P_ALL, 0, &si, WEXITED|WNOHANG|WNOWAIT) < 0) {
1972
1973 if (errno == ECHILD)
1974 break;
1975
1976 if (errno == EINTR)
1977 continue;
1978
1979 return -errno;
1980 }
1981
1982 if (si.si_pid <= 0)
1983 break;
1984
1985 if (si.si_code == CLD_EXITED || si.si_code == CLD_KILLED || si.si_code == CLD_DUMPED) {
1986 _cleanup_free_ char *name = NULL;
1987 Unit *u1, *u2, *u3;
1988
1989 get_process_comm(si.si_pid, &name);
1990
1991 log_debug("Child "PID_FMT" (%s) died (code=%s, status=%i/%s)",
1992 si.si_pid, strna(name),
1993 sigchld_code_to_string(si.si_code),
1994 si.si_status,
1995 strna(si.si_code == CLD_EXITED
1996 ? exit_status_to_string(si.si_status, EXIT_STATUS_FULL)
1997 : signal_to_string(si.si_status)));
1998
1999 /* And now figure out the unit this belongs
2000 * to, it might be multiple... */
2001 u1 = manager_get_unit_by_pid_cgroup(m, si.si_pid);
2002 if (u1)
2003 invoke_sigchld_event(m, u1, &si);
2004 u2 = hashmap_get(m->watch_pids1, PID_TO_PTR(si.si_pid));
2005 if (u2 && u2 != u1)
2006 invoke_sigchld_event(m, u2, &si);
2007 u3 = hashmap_get(m->watch_pids2, PID_TO_PTR(si.si_pid));
2008 if (u3 && u3 != u2 && u3 != u1)
2009 invoke_sigchld_event(m, u3, &si);
2010 }
2011
2012 /* And now, we actually reap the zombie. */
2013 if (waitid(P_PID, si.si_pid, &si, WEXITED) < 0) {
2014 if (errno == EINTR)
2015 continue;
2016
2017 return -errno;
2018 }
2019 }
2020
2021 return 0;
2022 }
2023
2024 static void manager_start_target(Manager *m, const char *name, JobMode mode) {
2025 _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL;
2026 int r;
2027
2028 log_debug("Activating special unit %s", name);
2029
2030 r = manager_add_job_by_name(m, JOB_START, name, mode, &error, NULL);
2031 if (r < 0)
2032 log_error("Failed to enqueue %s job: %s", name, bus_error_message(&error, r));
2033 }
2034
2035 static void manager_handle_ctrl_alt_del(Manager *m) {
2036 /* If the user presses C-A-D more than
2037 * 7 times within 2s, we reboot/shutdown immediately,
2038 * unless it was disabled in system.conf */
2039
2040 if (ratelimit_test(&m->ctrl_alt_del_ratelimit) || m->cad_burst_action == EMERGENCY_ACTION_NONE)
2041 manager_start_target(m, SPECIAL_CTRL_ALT_DEL_TARGET, JOB_REPLACE_IRREVERSIBLY);
2042 else
2043 emergency_action(m, m->cad_burst_action, NULL,
2044 "Ctrl-Alt-Del was pressed more than 7 times within 2s");
2045 }
2046
2047 static int manager_dispatch_signal_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
2048 Manager *m = userdata;
2049 ssize_t n;
2050 struct signalfd_siginfo sfsi;
2051 bool sigchld = false;
2052 int r;
2053
2054 assert(m);
2055 assert(m->signal_fd == fd);
2056
2057 if (revents != EPOLLIN) {
2058 log_warning("Got unexpected events from signal file descriptor.");
2059 return 0;
2060 }
2061
2062 for (;;) {
2063 n = read(m->signal_fd, &sfsi, sizeof(sfsi));
2064 if (n != sizeof(sfsi)) {
2065 if (n >= 0) {
2066 log_warning("Truncated read from signal fd (%zu bytes)!", n);
2067 return 0;
2068 }
2069
2070 if (IN_SET(errno, EINTR, EAGAIN))
2071 break;
2072
2073 /* We return an error here, which will kill this handler,
2074 * to avoid a busy loop on read error. */
2075 return log_error_errno(errno, "Reading from signal fd failed: %m");
2076 }
2077
2078 log_received_signal(sfsi.ssi_signo == SIGCHLD ||
2079 (sfsi.ssi_signo == SIGTERM && MANAGER_IS_USER(m))
2080 ? LOG_DEBUG : LOG_INFO,
2081 &sfsi);
2082
2083 switch (sfsi.ssi_signo) {
2084
2085 case SIGCHLD:
2086 sigchld = true;
2087 break;
2088
2089 case SIGTERM:
2090 if (MANAGER_IS_SYSTEM(m)) {
2091 /* This is for compatibility with the
2092 * original sysvinit */
2093 r = verify_run_space_and_log("Refusing to reexecute");
2094 if (r >= 0)
2095 m->exit_code = MANAGER_REEXECUTE;
2096 break;
2097 }
2098
2099 /* Fall through */
2100
2101 case SIGINT:
2102 if (MANAGER_IS_SYSTEM(m))
2103 manager_handle_ctrl_alt_del(m);
2104 else
2105 manager_start_target(m, SPECIAL_EXIT_TARGET,
2106 JOB_REPLACE_IRREVERSIBLY);
2107 break;
2108
2109 case SIGWINCH:
2110 if (MANAGER_IS_SYSTEM(m))
2111 manager_start_target(m, SPECIAL_KBREQUEST_TARGET, JOB_REPLACE);
2112
2113 /* This is a nop on non-init */
2114 break;
2115
2116 case SIGPWR:
2117 if (MANAGER_IS_SYSTEM(m))
2118 manager_start_target(m, SPECIAL_SIGPWR_TARGET, JOB_REPLACE);
2119
2120 /* This is a nop on non-init */
2121 break;
2122
2123 case SIGUSR1: {
2124 Unit *u;
2125
2126 u = manager_get_unit(m, SPECIAL_DBUS_SERVICE);
2127
2128 if (!u || UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u))) {
2129 log_info("Trying to reconnect to bus...");
2130 bus_init(m, true);
2131 }
2132
2133 if (!u || !UNIT_IS_ACTIVE_OR_ACTIVATING(unit_active_state(u))) {
2134 log_info("Loading D-Bus service...");
2135 manager_start_target(m, SPECIAL_DBUS_SERVICE, JOB_REPLACE);
2136 }
2137
2138 break;
2139 }
2140
2141 case SIGUSR2: {
2142 _cleanup_free_ char *dump = NULL;
2143 _cleanup_fclose_ FILE *f = NULL;
2144 size_t size;
2145
2146 f = open_memstream(&dump, &size);
2147 if (!f) {
2148 log_warning_errno(errno, "Failed to allocate memory stream: %m");
2149 break;
2150 }
2151
2152 manager_dump_units(m, f, "\t");
2153 manager_dump_jobs(m, f, "\t");
2154
2155 r = fflush_and_check(f);
2156 if (r < 0) {
2157 log_warning_errno(r, "Failed to write status stream: %m");
2158 break;
2159 }
2160
2161 log_dump(LOG_INFO, dump);
2162 break;
2163 }
2164
2165 case SIGHUP:
2166 r = verify_run_space_and_log("Refusing to reload");
2167 if (r >= 0)
2168 m->exit_code = MANAGER_RELOAD;
2169 break;
2170
2171 default: {
2172
2173 /* Starting SIGRTMIN+0 */
2174 static const struct {
2175 const char *target;
2176 JobMode mode;
2177 } target_table[] = {
2178 [0] = { SPECIAL_DEFAULT_TARGET, JOB_ISOLATE },
2179 [1] = { SPECIAL_RESCUE_TARGET, JOB_ISOLATE },
2180 [2] = { SPECIAL_EMERGENCY_TARGET, JOB_ISOLATE },
2181 [3] = { SPECIAL_HALT_TARGET, JOB_REPLACE_IRREVERSIBLY },
2182 [4] = { SPECIAL_POWEROFF_TARGET, JOB_REPLACE_IRREVERSIBLY },
2183 [5] = { SPECIAL_REBOOT_TARGET, JOB_REPLACE_IRREVERSIBLY },
2184 [6] = { SPECIAL_KEXEC_TARGET, JOB_REPLACE_IRREVERSIBLY }
2185 };
2186
2187 /* Starting SIGRTMIN+13, so that target halt and system halt are 10 apart */
2188 static const ManagerExitCode code_table[] = {
2189 [0] = MANAGER_HALT,
2190 [1] = MANAGER_POWEROFF,
2191 [2] = MANAGER_REBOOT,
2192 [3] = MANAGER_KEXEC
2193 };
2194
2195 if ((int) sfsi.ssi_signo >= SIGRTMIN+0 &&
2196 (int) sfsi.ssi_signo < SIGRTMIN+(int) ELEMENTSOF(target_table)) {
2197 int idx = (int) sfsi.ssi_signo - SIGRTMIN;
2198 manager_start_target(m, target_table[idx].target,
2199 target_table[idx].mode);
2200 break;
2201 }
2202
2203 if ((int) sfsi.ssi_signo >= SIGRTMIN+13 &&
2204 (int) sfsi.ssi_signo < SIGRTMIN+13+(int) ELEMENTSOF(code_table)) {
2205 m->exit_code = code_table[sfsi.ssi_signo - SIGRTMIN - 13];
2206 break;
2207 }
2208
2209 switch (sfsi.ssi_signo - SIGRTMIN) {
2210
2211 case 20:
2212 manager_set_show_status(m, SHOW_STATUS_YES);
2213 break;
2214
2215 case 21:
2216 manager_set_show_status(m, SHOW_STATUS_NO);
2217 break;
2218
2219 case 22:
2220 log_set_max_level(LOG_DEBUG);
2221 log_info("Setting log level to debug.");
2222 break;
2223
2224 case 23:
2225 log_set_max_level(LOG_INFO);
2226 log_info("Setting log level to info.");
2227 break;
2228
2229 case 24:
2230 if (MANAGER_IS_USER(m)) {
2231 m->exit_code = MANAGER_EXIT;
2232 return 0;
2233 }
2234
2235 /* This is a nop on init */
2236 break;
2237
2238 case 26:
2239 case 29: /* compatibility: used to be mapped to LOG_TARGET_SYSLOG_OR_KMSG */
2240 log_set_target(LOG_TARGET_JOURNAL_OR_KMSG);
2241 log_notice("Setting log target to journal-or-kmsg.");
2242 break;
2243
2244 case 27:
2245 log_set_target(LOG_TARGET_CONSOLE);
2246 log_notice("Setting log target to console.");
2247 break;
2248
2249 case 28:
2250 log_set_target(LOG_TARGET_KMSG);
2251 log_notice("Setting log target to kmsg.");
2252 break;
2253
2254 default:
2255 log_warning("Got unhandled signal <%s>.", signal_to_string(sfsi.ssi_signo));
2256 }
2257 }
2258 }
2259 }
2260
2261 if (sigchld)
2262 manager_dispatch_sigchld(m);
2263
2264 return 0;
2265 }
2266
2267 static int manager_dispatch_time_change_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
2268 Manager *m = userdata;
2269 Iterator i;
2270 Unit *u;
2271
2272 assert(m);
2273 assert(m->time_change_fd == fd);
2274
2275 log_struct(LOG_DEBUG,
2276 "MESSAGE_ID=" SD_MESSAGE_TIME_CHANGE_STR,
2277 LOG_MESSAGE("Time has been changed"),
2278 NULL);
2279
2280 /* Restart the watch */
2281 m->time_change_event_source = sd_event_source_unref(m->time_change_event_source);
2282 m->time_change_fd = safe_close(m->time_change_fd);
2283
2284 manager_setup_time_change(m);
2285
2286 HASHMAP_FOREACH(u, m->units, i)
2287 if (UNIT_VTABLE(u)->time_change)
2288 UNIT_VTABLE(u)->time_change(u);
2289
2290 return 0;
2291 }
2292
2293 static int manager_dispatch_idle_pipe_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
2294 Manager *m = userdata;
2295
2296 assert(m);
2297 assert(m->idle_pipe[2] == fd);
2298
2299 m->no_console_output = m->n_on_console > 0;
2300
2301 manager_close_idle_pipe(m);
2302
2303 return 0;
2304 }
2305
2306 static int manager_dispatch_jobs_in_progress(sd_event_source *source, usec_t usec, void *userdata) {
2307 Manager *m = userdata;
2308 int r;
2309 uint64_t next;
2310
2311 assert(m);
2312 assert(source);
2313
2314 manager_print_jobs_in_progress(m);
2315
2316 next = now(CLOCK_MONOTONIC) + JOBS_IN_PROGRESS_PERIOD_USEC;
2317 r = sd_event_source_set_time(source, next);
2318 if (r < 0)
2319 return r;
2320
2321 return sd_event_source_set_enabled(source, SD_EVENT_ONESHOT);
2322 }
2323
2324 int manager_loop(Manager *m) {
2325 int r;
2326
2327 RATELIMIT_DEFINE(rl, 1*USEC_PER_SEC, 50000);
2328
2329 assert(m);
2330 m->exit_code = MANAGER_OK;
2331
2332 /* Release the path cache */
2333 m->unit_path_cache = set_free_free(m->unit_path_cache);
2334
2335 manager_check_finished(m);
2336
2337 /* There might still be some zombies hanging around from
2338 * before we were exec()'ed. Let's reap them. */
2339 r = manager_dispatch_sigchld(m);
2340 if (r < 0)
2341 return r;
2342
2343 while (m->exit_code == MANAGER_OK) {
2344 usec_t wait_usec;
2345
2346 if (m->runtime_watchdog > 0 && m->runtime_watchdog != USEC_INFINITY && MANAGER_IS_SYSTEM(m))
2347 watchdog_ping();
2348
2349 if (!ratelimit_test(&rl)) {
2350 /* Yay, something is going seriously wrong, pause a little */
2351 log_warning("Looping too fast. Throttling execution a little.");
2352 sleep(1);
2353 }
2354
2355 if (manager_dispatch_load_queue(m) > 0)
2356 continue;
2357
2358 if (manager_dispatch_gc_job_queue(m) > 0)
2359 continue;
2360
2361 if (manager_dispatch_gc_unit_queue(m) > 0)
2362 continue;
2363
2364 if (manager_dispatch_cleanup_queue(m) > 0)
2365 continue;
2366
2367 if (manager_dispatch_cgroup_queue(m) > 0)
2368 continue;
2369
2370 if (manager_dispatch_dbus_queue(m) > 0)
2371 continue;
2372
2373 /* Sleep for half the watchdog time */
2374 if (m->runtime_watchdog > 0 && m->runtime_watchdog != USEC_INFINITY && MANAGER_IS_SYSTEM(m)) {
2375 wait_usec = m->runtime_watchdog / 2;
2376 if (wait_usec <= 0)
2377 wait_usec = 1;
2378 } else
2379 wait_usec = USEC_INFINITY;
2380
2381 r = sd_event_run(m->event, wait_usec);
2382 if (r < 0)
2383 return log_error_errno(r, "Failed to run event loop: %m");
2384 }
2385
2386 return m->exit_code;
2387 }
2388
2389 int manager_load_unit_from_dbus_path(Manager *m, const char *s, sd_bus_error *e, Unit **_u) {
2390 _cleanup_free_ char *n = NULL;
2391 sd_id128_t invocation_id;
2392 Unit *u;
2393 int r;
2394
2395 assert(m);
2396 assert(s);
2397 assert(_u);
2398
2399 r = unit_name_from_dbus_path(s, &n);
2400 if (r < 0)
2401 return r;
2402
2403 /* Permit addressing units by invocation ID: if the passed bus path is suffixed by a 128bit ID then we use it
2404 * as invocation ID. */
2405 r = sd_id128_from_string(n, &invocation_id);
2406 if (r >= 0) {
2407 u = hashmap_get(m->units_by_invocation_id, &invocation_id);
2408 if (u) {
2409 *_u = u;
2410 return 0;
2411 }
2412
2413 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));
2414 }
2415
2416 /* If this didn't work, we check if this is a unit name */
2417 if (!unit_name_is_valid(n, UNIT_NAME_PLAIN|UNIT_NAME_INSTANCE))
2418 return sd_bus_error_setf(e, SD_BUS_ERROR_INVALID_ARGS, "Unit name %s is neither a valid invocation ID nor unit name.", n);
2419
2420 r = manager_load_unit(m, n, NULL, e, &u);
2421 if (r < 0)
2422 return r;
2423
2424 *_u = u;
2425 return 0;
2426 }
2427
2428 int manager_get_job_from_dbus_path(Manager *m, const char *s, Job **_j) {
2429 const char *p;
2430 unsigned id;
2431 Job *j;
2432 int r;
2433
2434 assert(m);
2435 assert(s);
2436 assert(_j);
2437
2438 p = startswith(s, "/org/freedesktop/systemd1/job/");
2439 if (!p)
2440 return -EINVAL;
2441
2442 r = safe_atou(p, &id);
2443 if (r < 0)
2444 return r;
2445
2446 j = manager_get_job(m, id);
2447 if (!j)
2448 return -ENOENT;
2449
2450 *_j = j;
2451
2452 return 0;
2453 }
2454
2455 void manager_send_unit_audit(Manager *m, Unit *u, int type, bool success) {
2456
2457 #ifdef HAVE_AUDIT
2458 _cleanup_free_ char *p = NULL;
2459 const char *msg;
2460 int audit_fd, r;
2461
2462 if (!MANAGER_IS_SYSTEM(m))
2463 return;
2464
2465 audit_fd = get_audit_fd();
2466 if (audit_fd < 0)
2467 return;
2468
2469 /* Don't generate audit events if the service was already
2470 * started and we're just deserializing */
2471 if (MANAGER_IS_RELOADING(m))
2472 return;
2473
2474 if (u->type != UNIT_SERVICE)
2475 return;
2476
2477 r = unit_name_to_prefix_and_instance(u->id, &p);
2478 if (r < 0) {
2479 log_error_errno(r, "Failed to extract prefix and instance of unit name: %m");
2480 return;
2481 }
2482
2483 msg = strjoina("unit=", p);
2484 if (audit_log_user_comm_message(audit_fd, type, msg, "systemd", NULL, NULL, NULL, success) < 0) {
2485 if (errno == EPERM)
2486 /* We aren't allowed to send audit messages?
2487 * Then let's not retry again. */
2488 close_audit_fd();
2489 else
2490 log_warning_errno(errno, "Failed to send audit message: %m");
2491 }
2492 #endif
2493
2494 }
2495
2496 void manager_send_unit_plymouth(Manager *m, Unit *u) {
2497 static const union sockaddr_union sa = PLYMOUTH_SOCKET;
2498 _cleanup_free_ char *message = NULL;
2499 _cleanup_close_ int fd = -1;
2500 int n = 0;
2501
2502 /* Don't generate plymouth events if the service was already
2503 * started and we're just deserializing */
2504 if (MANAGER_IS_RELOADING(m))
2505 return;
2506
2507 if (!MANAGER_IS_SYSTEM(m))
2508 return;
2509
2510 if (detect_container() > 0)
2511 return;
2512
2513 if (u->type != UNIT_SERVICE &&
2514 u->type != UNIT_MOUNT &&
2515 u->type != UNIT_SWAP)
2516 return;
2517
2518 /* We set SOCK_NONBLOCK here so that we rather drop the
2519 * message then wait for plymouth */
2520 fd = socket(AF_UNIX, SOCK_STREAM|SOCK_CLOEXEC|SOCK_NONBLOCK, 0);
2521 if (fd < 0) {
2522 log_error_errno(errno, "socket() failed: %m");
2523 return;
2524 }
2525
2526 if (connect(fd, &sa.sa, SOCKADDR_UN_LEN(sa.un)) < 0) {
2527
2528 if (!IN_SET(errno, EPIPE, EAGAIN, ENOENT, ECONNREFUSED, ECONNRESET, ECONNABORTED))
2529 log_error_errno(errno, "connect() failed: %m");
2530 return;
2531 }
2532
2533 if (asprintf(&message, "U\002%c%s%n", (int) (strlen(u->id) + 1), u->id, &n) < 0) {
2534 log_oom();
2535 return;
2536 }
2537
2538 errno = 0;
2539 if (write(fd, message, n + 1) != n + 1)
2540 if (!IN_SET(errno, EPIPE, EAGAIN, ENOENT, ECONNREFUSED, ECONNRESET, ECONNABORTED))
2541 log_error_errno(errno, "Failed to write Plymouth message: %m");
2542 }
2543
2544 int manager_open_serialization(Manager *m, FILE **_f) {
2545 int fd;
2546 FILE *f;
2547
2548 assert(_f);
2549
2550 fd = open_serialization_fd("systemd-state");
2551 if (fd < 0)
2552 return fd;
2553
2554 f = fdopen(fd, "w+");
2555 if (!f) {
2556 safe_close(fd);
2557 return -errno;
2558 }
2559
2560 *_f = f;
2561 return 0;
2562 }
2563
2564 int manager_serialize(Manager *m, FILE *f, FDSet *fds, bool switching_root) {
2565 Iterator i;
2566 Unit *u;
2567 const char *t;
2568 int r;
2569
2570 assert(m);
2571 assert(f);
2572 assert(fds);
2573
2574 m->n_reloading++;
2575
2576 fprintf(f, "current-job-id=%"PRIu32"\n", m->current_job_id);
2577 fprintf(f, "taint-usr=%s\n", yes_no(m->taint_usr));
2578 fprintf(f, "n-installed-jobs=%u\n", m->n_installed_jobs);
2579 fprintf(f, "n-failed-jobs=%u\n", m->n_failed_jobs);
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, "firmware-timestamp=")))
2737 dual_timestamp_deserialize(val, &m->firmware_timestamp);
2738 else if ((val = startswith(l, "loader-timestamp=")))
2739 dual_timestamp_deserialize(val, &m->loader_timestamp);
2740 else if ((val = startswith(l, "kernel-timestamp=")))
2741 dual_timestamp_deserialize(val, &m->kernel_timestamp);
2742 else if ((val = startswith(l, "initrd-timestamp=")))
2743 dual_timestamp_deserialize(val, &m->initrd_timestamp);
2744 else if ((val = startswith(l, "userspace-timestamp=")))
2745 dual_timestamp_deserialize(val, &m->userspace_timestamp);
2746 else if ((val = startswith(l, "finish-timestamp=")))
2747 dual_timestamp_deserialize(val, &m->finish_timestamp);
2748 else if ((val = startswith(l, "security-start-timestamp=")))
2749 dual_timestamp_deserialize(val, &m->security_start_timestamp);
2750 else if ((val = startswith(l, "security-finish-timestamp=")))
2751 dual_timestamp_deserialize(val, &m->security_finish_timestamp);
2752 else if ((val = startswith(l, "generators-start-timestamp=")))
2753 dual_timestamp_deserialize(val, &m->generators_start_timestamp);
2754 else if ((val = startswith(l, "generators-finish-timestamp=")))
2755 dual_timestamp_deserialize(val, &m->generators_finish_timestamp);
2756 else if ((val = startswith(l, "units-load-start-timestamp=")))
2757 dual_timestamp_deserialize(val, &m->units_load_start_timestamp);
2758 else if ((val = startswith(l, "units-load-finish-timestamp=")))
2759 dual_timestamp_deserialize(val, &m->units_load_finish_timestamp);
2760 else if (startswith(l, "env=")) {
2761 r = deserialize_environment(&m->environment, l);
2762 if (r == -ENOMEM)
2763 goto finish;
2764 if (r < 0)
2765 log_notice_errno(r, "Failed to parse environment entry: \"%s\": %m", l);
2766
2767 } else if ((val = startswith(l, "notify-fd="))) {
2768 int fd;
2769
2770 if (safe_atoi(val, &fd) < 0 || fd < 0 || !fdset_contains(fds, fd))
2771 log_notice("Failed to parse notify fd: \"%s\"", val);
2772 else {
2773 m->notify_event_source = sd_event_source_unref(m->notify_event_source);
2774 safe_close(m->notify_fd);
2775 m->notify_fd = fdset_remove(fds, fd);
2776 }
2777
2778 } else if ((val = startswith(l, "notify-socket="))) {
2779 char *n;
2780
2781 n = strdup(val);
2782 if (!n) {
2783 r = -ENOMEM;
2784 goto finish;
2785 }
2786
2787 free(m->notify_socket);
2788 m->notify_socket = n;
2789
2790 } else if ((val = startswith(l, "cgroups-agent-fd="))) {
2791 int fd;
2792
2793 if (safe_atoi(val, &fd) < 0 || fd < 0 || !fdset_contains(fds, fd))
2794 log_notice("Failed to parse cgroups agent fd: %s", val);
2795 else {
2796 m->cgroups_agent_event_source = sd_event_source_unref(m->cgroups_agent_event_source);
2797 safe_close(m->cgroups_agent_fd);
2798 m->cgroups_agent_fd = fdset_remove(fds, fd);
2799 }
2800
2801 } else if ((val = startswith(l, "user-lookup="))) {
2802 int fd0, fd1;
2803
2804 if (sscanf(val, "%i %i", &fd0, &fd1) != 2 || fd0 < 0 || fd1 < 0 || fd0 == fd1 || !fdset_contains(fds, fd0) || !fdset_contains(fds, fd1))
2805 log_notice("Failed to parse user lookup fd: %s", val);
2806 else {
2807 m->user_lookup_event_source = sd_event_source_unref(m->user_lookup_event_source);
2808 safe_close_pair(m->user_lookup_fds);
2809 m->user_lookup_fds[0] = fdset_remove(fds, fd0);
2810 m->user_lookup_fds[1] = fdset_remove(fds, fd1);
2811 }
2812
2813 } else if ((val = startswith(l, "dynamic-user=")))
2814 dynamic_user_deserialize_one(m, val, fds);
2815 else if ((val = startswith(l, "destroy-ipc-uid=")))
2816 manager_deserialize_uid_refs_one(m, val);
2817 else if ((val = startswith(l, "destroy-ipc-gid=")))
2818 manager_deserialize_gid_refs_one(m, val);
2819 else if ((val = startswith(l, "subscribed="))) {
2820
2821 if (strv_extend(&m->deserialized_subscribed, val) < 0)
2822 log_oom();
2823
2824 } else if (!startswith(l, "kdbus-fd=")) /* ignore this one */
2825 log_notice("Unknown serialization item '%s'", l);
2826 }
2827
2828 for (;;) {
2829 Unit *u;
2830 char name[UNIT_NAME_MAX+2];
2831 const char* unit_name;
2832
2833 /* Start marker */
2834 if (!fgets(name, sizeof(name), f)) {
2835 if (feof(f))
2836 r = 0;
2837 else
2838 r = -errno;
2839
2840 goto finish;
2841 }
2842
2843 char_array_0(name);
2844 unit_name = strstrip(name);
2845
2846 r = manager_load_unit(m, unit_name, NULL, NULL, &u);
2847 if (r < 0) {
2848 log_notice_errno(r, "Failed to load unit \"%s\", skipping deserialization: %m", unit_name);
2849 if (r == -ENOMEM)
2850 goto finish;
2851 unit_deserialize_skip(f);
2852 continue;
2853 }
2854
2855 r = unit_deserialize(u, f, fds);
2856 if (r < 0) {
2857 log_notice_errno(r, "Failed to deserialize unit \"%s\": %m", unit_name);
2858 if (r == -ENOMEM)
2859 goto finish;
2860 }
2861 }
2862
2863 finish:
2864 if (ferror(f))
2865 r = -EIO;
2866
2867 assert(m->n_reloading > 0);
2868 m->n_reloading--;
2869
2870 return r;
2871 }
2872
2873 int manager_reload(Manager *m) {
2874 int r, q;
2875 _cleanup_fclose_ FILE *f = NULL;
2876 _cleanup_fdset_free_ FDSet *fds = NULL;
2877
2878 assert(m);
2879
2880 r = manager_open_serialization(m, &f);
2881 if (r < 0)
2882 return r;
2883
2884 m->n_reloading++;
2885 bus_manager_send_reloading(m, true);
2886
2887 fds = fdset_new();
2888 if (!fds) {
2889 m->n_reloading--;
2890 return -ENOMEM;
2891 }
2892
2893 r = manager_serialize(m, f, fds, false);
2894 if (r < 0) {
2895 m->n_reloading--;
2896 return r;
2897 }
2898
2899 if (fseeko(f, 0, SEEK_SET) < 0) {
2900 m->n_reloading--;
2901 return -errno;
2902 }
2903
2904 /* From here on there is no way back. */
2905 manager_clear_jobs_and_units(m);
2906 lookup_paths_flush_generator(&m->lookup_paths);
2907 lookup_paths_free(&m->lookup_paths);
2908 dynamic_user_vacuum(m, false);
2909 m->uid_refs = hashmap_free(m->uid_refs);
2910 m->gid_refs = hashmap_free(m->gid_refs);
2911
2912 q = lookup_paths_init(&m->lookup_paths, m->unit_file_scope, 0, NULL);
2913 if (q < 0 && r >= 0)
2914 r = q;
2915
2916 q = manager_run_environment_generators(m);
2917 if (q < 0 && r >= 0)
2918 r = q;
2919
2920 /* Find new unit paths */
2921 q = manager_run_generators(m);
2922 if (q < 0 && r >= 0)
2923 r = q;
2924
2925 lookup_paths_reduce(&m->lookup_paths);
2926 manager_build_unit_path_cache(m);
2927
2928 /* First, enumerate what we can from all config files */
2929 manager_enumerate(m);
2930
2931 /* Second, deserialize our stored data */
2932 q = manager_deserialize(m, f, fds);
2933 if (q < 0) {
2934 log_error_errno(q, "Deserialization failed: %m");
2935
2936 if (r >= 0)
2937 r = q;
2938 }
2939
2940 fclose(f);
2941 f = NULL;
2942
2943 /* Re-register notify_fd as event source */
2944 q = manager_setup_notify(m);
2945 if (q < 0 && r >= 0)
2946 r = q;
2947
2948 q = manager_setup_cgroups_agent(m);
2949 if (q < 0 && r >= 0)
2950 r = q;
2951
2952 q = manager_setup_user_lookup_fd(m);
2953 if (q < 0 && r >= 0)
2954 r = q;
2955
2956 /* Third, fire things up! */
2957 manager_coldplug(m);
2958
2959 /* Release any dynamic users no longer referenced */
2960 dynamic_user_vacuum(m, true);
2961
2962 /* Release any references to UIDs/GIDs no longer referenced, and destroy any IPC owned by them */
2963 manager_vacuum_uid_refs(m);
2964 manager_vacuum_gid_refs(m);
2965
2966 /* Sync current state of bus names with our set of listening units */
2967 if (m->api_bus)
2968 manager_sync_bus_names(m, m->api_bus);
2969
2970 assert(m->n_reloading > 0);
2971 m->n_reloading--;
2972
2973 m->send_reloading_done = true;
2974
2975 return r;
2976 }
2977
2978 void manager_reset_failed(Manager *m) {
2979 Unit *u;
2980 Iterator i;
2981
2982 assert(m);
2983
2984 HASHMAP_FOREACH(u, m->units, i)
2985 unit_reset_failed(u);
2986 }
2987
2988 bool manager_unit_inactive_or_pending(Manager *m, const char *name) {
2989 Unit *u;
2990
2991 assert(m);
2992 assert(name);
2993
2994 /* Returns true if the unit is inactive or going down */
2995 u = manager_get_unit(m, name);
2996 if (!u)
2997 return true;
2998
2999 return unit_inactive_or_pending(u);
3000 }
3001
3002 static void manager_notify_finished(Manager *m) {
3003 char userspace[FORMAT_TIMESPAN_MAX], initrd[FORMAT_TIMESPAN_MAX], kernel[FORMAT_TIMESPAN_MAX], sum[FORMAT_TIMESPAN_MAX];
3004 usec_t firmware_usec, loader_usec, kernel_usec, initrd_usec, userspace_usec, total_usec;
3005
3006 if (m->test_run)
3007 return;
3008
3009 if (MANAGER_IS_SYSTEM(m) && detect_container() <= 0) {
3010
3011 /* Note that m->kernel_usec.monotonic is always at 0,
3012 * and m->firmware_usec.monotonic and
3013 * m->loader_usec.monotonic should be considered
3014 * negative values. */
3015
3016 firmware_usec = m->firmware_timestamp.monotonic - m->loader_timestamp.monotonic;
3017 loader_usec = m->loader_timestamp.monotonic - m->kernel_timestamp.monotonic;
3018 userspace_usec = m->finish_timestamp.monotonic - m->userspace_timestamp.monotonic;
3019 total_usec = m->firmware_timestamp.monotonic + m->finish_timestamp.monotonic;
3020
3021 if (dual_timestamp_is_set(&m->initrd_timestamp)) {
3022
3023 kernel_usec = m->initrd_timestamp.monotonic - m->kernel_timestamp.monotonic;
3024 initrd_usec = m->userspace_timestamp.monotonic - m->initrd_timestamp.monotonic;
3025
3026 log_struct(LOG_INFO,
3027 "MESSAGE_ID=" SD_MESSAGE_STARTUP_FINISHED_STR,
3028 "KERNEL_USEC="USEC_FMT, kernel_usec,
3029 "INITRD_USEC="USEC_FMT, initrd_usec,
3030 "USERSPACE_USEC="USEC_FMT, userspace_usec,
3031 LOG_MESSAGE("Startup finished in %s (kernel) + %s (initrd) + %s (userspace) = %s.",
3032 format_timespan(kernel, sizeof(kernel), kernel_usec, USEC_PER_MSEC),
3033 format_timespan(initrd, sizeof(initrd), initrd_usec, USEC_PER_MSEC),
3034 format_timespan(userspace, sizeof(userspace), userspace_usec, USEC_PER_MSEC),
3035 format_timespan(sum, sizeof(sum), total_usec, USEC_PER_MSEC)),
3036 NULL);
3037 } else {
3038 kernel_usec = m->userspace_timestamp.monotonic - m->kernel_timestamp.monotonic;
3039 initrd_usec = 0;
3040
3041 log_struct(LOG_INFO,
3042 "MESSAGE_ID=" SD_MESSAGE_STARTUP_FINISHED_STR,
3043 "KERNEL_USEC="USEC_FMT, kernel_usec,
3044 "USERSPACE_USEC="USEC_FMT, userspace_usec,
3045 LOG_MESSAGE("Startup finished in %s (kernel) + %s (userspace) = %s.",
3046 format_timespan(kernel, sizeof(kernel), kernel_usec, USEC_PER_MSEC),
3047 format_timespan(userspace, sizeof(userspace), userspace_usec, USEC_PER_MSEC),
3048 format_timespan(sum, sizeof(sum), total_usec, USEC_PER_MSEC)),
3049 NULL);
3050 }
3051 } else {
3052 firmware_usec = loader_usec = initrd_usec = kernel_usec = 0;
3053 total_usec = userspace_usec = m->finish_timestamp.monotonic - m->userspace_timestamp.monotonic;
3054
3055 log_struct(LOG_INFO,
3056 "MESSAGE_ID=" SD_MESSAGE_USER_STARTUP_FINISHED_STR,
3057 "USERSPACE_USEC="USEC_FMT, userspace_usec,
3058 LOG_MESSAGE("Startup finished in %s.",
3059 format_timespan(sum, sizeof(sum), total_usec, USEC_PER_MSEC)),
3060 NULL);
3061 }
3062
3063 bus_manager_send_finished(m, firmware_usec, loader_usec, kernel_usec, initrd_usec, userspace_usec, total_usec);
3064
3065 sd_notifyf(false,
3066 "READY=1\n"
3067 "STATUS=Startup finished in %s.",
3068 format_timespan(sum, sizeof(sum), total_usec, USEC_PER_MSEC));
3069 }
3070
3071 void manager_check_finished(Manager *m) {
3072 assert(m);
3073
3074 if (MANAGER_IS_RELOADING(m))
3075 return;
3076
3077 /* Verify that we are actually running currently. Initially
3078 * the exit code is set to invalid, and during operation it is
3079 * then set to MANAGER_OK */
3080 if (m->exit_code != MANAGER_OK)
3081 return;
3082
3083 if (hashmap_size(m->jobs) > 0) {
3084 if (m->jobs_in_progress_event_source)
3085 /* Ignore any failure, this is only for feedback */
3086 (void) sd_event_source_set_time(m->jobs_in_progress_event_source, now(CLOCK_MONOTONIC) + JOBS_IN_PROGRESS_WAIT_USEC);
3087
3088 return;
3089 }
3090
3091 manager_flip_auto_status(m, false);
3092
3093 /* Notify Type=idle units that we are done now */
3094 manager_close_idle_pipe(m);
3095
3096 /* Turn off confirm spawn now */
3097 m->confirm_spawn = NULL;
3098
3099 /* No need to update ask password status when we're going non-interactive */
3100 manager_close_ask_password(m);
3101
3102 /* This is no longer the first boot */
3103 manager_set_first_boot(m, false);
3104
3105 if (dual_timestamp_is_set(&m->finish_timestamp))
3106 return;
3107
3108 dual_timestamp_get(&m->finish_timestamp);
3109
3110 manager_notify_finished(m);
3111
3112 manager_invalidate_startup_units(m);
3113 }
3114
3115 static bool generator_path_any(const char* const* paths) {
3116 char **path;
3117 bool found = false;
3118
3119 /* Optimize by skipping the whole process by not creating output directories
3120 * if no generators are found. */
3121 STRV_FOREACH(path, (char**) paths)
3122 if (access(*path, F_OK) == 0)
3123 found = true;
3124 else if (errno != ENOENT)
3125 log_warning_errno(errno, "Failed to open generator directory %s: %m", *path);
3126
3127 return found;
3128 }
3129
3130 static const char* system_env_generator_binary_paths[] = {
3131 "/run/systemd/system-environment-generators",
3132 "/etc/systemd/system-environment-generators",
3133 "/usr/local/lib/systemd/system-environment-generators",
3134 SYSTEM_ENV_GENERATOR_PATH,
3135 NULL
3136 };
3137
3138 static const char* user_env_generator_binary_paths[] = {
3139 "/run/systemd/user-environment-generators",
3140 "/etc/systemd/user-environment-generators",
3141 "/usr/local/lib/systemd/user-environment-generators",
3142 USER_ENV_GENERATOR_PATH,
3143 NULL
3144 };
3145
3146 static int manager_run_environment_generators(Manager *m) {
3147 char **tmp = NULL; /* this is only used in the forked process, no cleanup here */
3148 const char **paths;
3149 void* args[] = {&tmp, &tmp, &m->environment};
3150
3151 paths = MANAGER_IS_SYSTEM(m) ? system_env_generator_binary_paths : user_env_generator_binary_paths;
3152
3153 if (!generator_path_any(paths))
3154 return 0;
3155
3156 return execute_directories(paths, DEFAULT_TIMEOUT_USEC, gather_environment, args, NULL);
3157 }
3158
3159 static int manager_run_generators(Manager *m) {
3160 _cleanup_strv_free_ char **paths = NULL;
3161 const char *argv[5];
3162 int r;
3163
3164 assert(m);
3165
3166 paths = generator_binary_paths(m->unit_file_scope);
3167 if (!paths)
3168 return log_oom();
3169
3170 if (!generator_path_any((const char* const*) paths))
3171 return 0;
3172
3173 r = lookup_paths_mkdir_generator(&m->lookup_paths);
3174 if (r < 0)
3175 goto finish;
3176
3177 argv[0] = NULL; /* Leave this empty, execute_directory() will fill something in */
3178 argv[1] = m->lookup_paths.generator;
3179 argv[2] = m->lookup_paths.generator_early;
3180 argv[3] = m->lookup_paths.generator_late;
3181 argv[4] = NULL;
3182
3183 RUN_WITH_UMASK(0022)
3184 execute_directories((const char* const*) paths, DEFAULT_TIMEOUT_USEC,
3185 NULL, NULL, (char**) argv);
3186
3187 finish:
3188 lookup_paths_trim_generator(&m->lookup_paths);
3189 return r;
3190 }
3191
3192 int manager_environment_add(Manager *m, char **minus, char **plus) {
3193 char **a = NULL, **b = NULL, **l;
3194 assert(m);
3195
3196 l = m->environment;
3197
3198 if (!strv_isempty(minus)) {
3199 a = strv_env_delete(l, 1, minus);
3200 if (!a)
3201 return -ENOMEM;
3202
3203 l = a;
3204 }
3205
3206 if (!strv_isempty(plus)) {
3207 b = strv_env_merge(2, l, plus);
3208 if (!b) {
3209 strv_free(a);
3210 return -ENOMEM;
3211 }
3212
3213 l = b;
3214 }
3215
3216 if (m->environment != l)
3217 strv_free(m->environment);
3218 if (a != l)
3219 strv_free(a);
3220 if (b != l)
3221 strv_free(b);
3222
3223 m->environment = l;
3224 manager_clean_environment(m);
3225 strv_sort(m->environment);
3226
3227 return 0;
3228 }
3229
3230 int manager_set_default_rlimits(Manager *m, struct rlimit **default_rlimit) {
3231 int i;
3232
3233 assert(m);
3234
3235 for (i = 0; i < _RLIMIT_MAX; i++) {
3236 m->rlimit[i] = mfree(m->rlimit[i]);
3237
3238 if (!default_rlimit[i])
3239 continue;
3240
3241 m->rlimit[i] = newdup(struct rlimit, default_rlimit[i], 1);
3242 if (!m->rlimit[i])
3243 return log_oom();
3244 }
3245
3246 return 0;
3247 }
3248
3249 void manager_recheck_journal(Manager *m) {
3250 Unit *u;
3251
3252 assert(m);
3253
3254 if (!MANAGER_IS_SYSTEM(m))
3255 return;
3256
3257 u = manager_get_unit(m, SPECIAL_JOURNALD_SOCKET);
3258 if (u && SOCKET(u)->state != SOCKET_RUNNING) {
3259 log_close_journal();
3260 return;
3261 }
3262
3263 u = manager_get_unit(m, SPECIAL_JOURNALD_SERVICE);
3264 if (u && SERVICE(u)->state != SERVICE_RUNNING) {
3265 log_close_journal();
3266 return;
3267 }
3268
3269 /* Hmm, OK, so the socket is fully up and the service is up
3270 * too, then let's make use of the thing. */
3271 log_open();
3272 }
3273
3274 void manager_set_show_status(Manager *m, ShowStatus mode) {
3275 assert(m);
3276 assert(IN_SET(mode, SHOW_STATUS_AUTO, SHOW_STATUS_NO, SHOW_STATUS_YES, SHOW_STATUS_TEMPORARY));
3277
3278 if (!MANAGER_IS_SYSTEM(m))
3279 return;
3280
3281 if (m->show_status != mode)
3282 log_debug("%s showing of status.",
3283 mode == SHOW_STATUS_NO ? "Disabling" : "Enabling");
3284 m->show_status = mode;
3285
3286 if (mode > 0)
3287 (void) touch("/run/systemd/show-status");
3288 else
3289 (void) unlink("/run/systemd/show-status");
3290 }
3291
3292 static bool manager_get_show_status(Manager *m, StatusType type) {
3293 assert(m);
3294
3295 if (!MANAGER_IS_SYSTEM(m))
3296 return false;
3297
3298 if (m->no_console_output)
3299 return false;
3300
3301 if (!IN_SET(manager_state(m), MANAGER_INITIALIZING, MANAGER_STARTING, MANAGER_STOPPING))
3302 return false;
3303
3304 /* If we cannot find out the status properly, just proceed. */
3305 if (type != STATUS_TYPE_EMERGENCY && manager_check_ask_password(m) > 0)
3306 return false;
3307
3308 if (m->show_status > 0)
3309 return true;
3310
3311 return false;
3312 }
3313
3314 const char *manager_get_confirm_spawn(Manager *m) {
3315 static int last_errno = 0;
3316 const char *vc = m->confirm_spawn;
3317 struct stat st;
3318 int r;
3319
3320 /* Here's the deal: we want to test the validity of the console but don't want
3321 * PID1 to go through the whole console process which might block. But we also
3322 * want to warn the user only once if something is wrong with the console so we
3323 * cannot do the sanity checks after spawning our children. So here we simply do
3324 * really basic tests to hopefully trap common errors.
3325 *
3326 * If the console suddenly disappear at the time our children will really it
3327 * then they will simply fail to acquire it and a positive answer will be
3328 * assumed. New children will fallback to /dev/console though.
3329 *
3330 * Note: TTYs are devices that can come and go any time, and frequently aren't
3331 * available yet during early boot (consider a USB rs232 dongle...). If for any
3332 * reason the configured console is not ready, we fallback to the default
3333 * console. */
3334
3335 if (!vc || path_equal(vc, "/dev/console"))
3336 return vc;
3337
3338 r = stat(vc, &st);
3339 if (r < 0)
3340 goto fail;
3341
3342 if (!S_ISCHR(st.st_mode)) {
3343 errno = ENOTTY;
3344 goto fail;
3345 }
3346
3347 last_errno = 0;
3348 return vc;
3349 fail:
3350 if (last_errno != errno) {
3351 last_errno = errno;
3352 log_warning_errno(errno, "Failed to open %s: %m, using default console", vc);
3353 }
3354 return "/dev/console";
3355 }
3356
3357 void manager_set_first_boot(Manager *m, bool b) {
3358 assert(m);
3359
3360 if (!MANAGER_IS_SYSTEM(m))
3361 return;
3362
3363 if (m->first_boot != (int) b) {
3364 if (b)
3365 (void) touch("/run/systemd/first-boot");
3366 else
3367 (void) unlink("/run/systemd/first-boot");
3368 }
3369
3370 m->first_boot = b;
3371 }
3372
3373 void manager_disable_confirm_spawn(void) {
3374 (void) touch("/run/systemd/confirm_spawn_disabled");
3375 }
3376
3377 bool manager_is_confirm_spawn_disabled(Manager *m) {
3378 if (!m->confirm_spawn)
3379 return true;
3380
3381 return access("/run/systemd/confirm_spawn_disabled", F_OK) >= 0;
3382 }
3383
3384 void manager_status_printf(Manager *m, StatusType type, const char *status, const char *format, ...) {
3385 va_list ap;
3386
3387 /* If m is NULL, assume we're after shutdown and let the messages through. */
3388
3389 if (m && !manager_get_show_status(m, type))
3390 return;
3391
3392 /* XXX We should totally drop the check for ephemeral here
3393 * and thus effectively make 'Type=idle' pointless. */
3394 if (type == STATUS_TYPE_EPHEMERAL && m && m->n_on_console > 0)
3395 return;
3396
3397 va_start(ap, format);
3398 status_vprintf(status, true, type == STATUS_TYPE_EPHEMERAL, format, ap);
3399 va_end(ap);
3400 }
3401
3402 Set *manager_get_units_requiring_mounts_for(Manager *m, const char *path) {
3403 char p[strlen(path)+1];
3404
3405 assert(m);
3406 assert(path);
3407
3408 strcpy(p, path);
3409 path_kill_slashes(p);
3410
3411 return hashmap_get(m->units_requiring_mounts_for, streq(p, "/") ? "" : p);
3412 }
3413
3414 void manager_set_exec_params(Manager *m, ExecParameters *p) {
3415 assert(m);
3416 assert(p);
3417
3418 p->environment = m->environment;
3419 p->confirm_spawn = manager_get_confirm_spawn(m);
3420 p->cgroup_supported = m->cgroup_supported;
3421 p->prefix = m->prefix;
3422
3423 SET_FLAG(p->flags, EXEC_PASS_LOG_UNIT|EXEC_CHOWN_DIRECTORIES, MANAGER_IS_SYSTEM(m));
3424 }
3425
3426 int manager_update_failed_units(Manager *m, Unit *u, bool failed) {
3427 unsigned size;
3428 int r;
3429
3430 assert(m);
3431 assert(u->manager == m);
3432
3433 size = set_size(m->failed_units);
3434
3435 if (failed) {
3436 r = set_ensure_allocated(&m->failed_units, NULL);
3437 if (r < 0)
3438 return log_oom();
3439
3440 if (set_put(m->failed_units, u) < 0)
3441 return log_oom();
3442 } else
3443 (void) set_remove(m->failed_units, u);
3444
3445 if (set_size(m->failed_units) != size)
3446 bus_manager_send_change_signal(m);
3447
3448 return 0;
3449 }
3450
3451 ManagerState manager_state(Manager *m) {
3452 Unit *u;
3453
3454 assert(m);
3455
3456 /* Did we ever finish booting? If not then we are still starting up */
3457 if (!dual_timestamp_is_set(&m->finish_timestamp)) {
3458
3459 u = manager_get_unit(m, SPECIAL_BASIC_TARGET);
3460 if (!u || !UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u)))
3461 return MANAGER_INITIALIZING;
3462
3463 return MANAGER_STARTING;
3464 }
3465
3466 /* Is the special shutdown target queued? If so, we are in shutdown state */
3467 u = manager_get_unit(m, SPECIAL_SHUTDOWN_TARGET);
3468 if (u && u->job && IN_SET(u->job->type, JOB_START, JOB_RESTART, JOB_RELOAD_OR_START))
3469 return MANAGER_STOPPING;
3470
3471 /* Are the rescue or emergency targets active or queued? If so we are in maintenance state */
3472 u = manager_get_unit(m, SPECIAL_RESCUE_TARGET);
3473 if (u && (UNIT_IS_ACTIVE_OR_ACTIVATING(unit_active_state(u)) ||
3474 (u->job && IN_SET(u->job->type, JOB_START, JOB_RESTART, JOB_RELOAD_OR_START))))
3475 return MANAGER_MAINTENANCE;
3476
3477 u = manager_get_unit(m, SPECIAL_EMERGENCY_TARGET);
3478 if (u && (UNIT_IS_ACTIVE_OR_ACTIVATING(unit_active_state(u)) ||
3479 (u->job && IN_SET(u->job->type, JOB_START, JOB_RESTART, JOB_RELOAD_OR_START))))
3480 return MANAGER_MAINTENANCE;
3481
3482 /* Are there any failed units? If so, we are in degraded mode */
3483 if (set_size(m->failed_units) > 0)
3484 return MANAGER_DEGRADED;
3485
3486 return MANAGER_RUNNING;
3487 }
3488
3489 #define DESTROY_IPC_FLAG (UINT32_C(1) << 31)
3490
3491 static void manager_unref_uid_internal(
3492 Manager *m,
3493 Hashmap **uid_refs,
3494 uid_t uid,
3495 bool destroy_now,
3496 int (*_clean_ipc)(uid_t uid)) {
3497
3498 uint32_t c, n;
3499
3500 assert(m);
3501 assert(uid_refs);
3502 assert(uid_is_valid(uid));
3503 assert(_clean_ipc);
3504
3505 /* A generic implementation, covering both manager_unref_uid() and manager_unref_gid(), under the assumption
3506 * that uid_t and gid_t are actually defined the same way, with the same validity rules.
3507 *
3508 * We store a hashmap where the UID/GID is they key and the value is a 32bit reference counter, whose highest
3509 * bit is used as flag for marking UIDs/GIDs whose IPC objects to remove when the last reference to the UID/GID
3510 * is dropped. The flag is set to on, once at least one reference from a unit where RemoveIPC= is set is added
3511 * on a UID/GID. It is reset when the UID's/GID's reference counter drops to 0 again. */
3512
3513 assert_cc(sizeof(uid_t) == sizeof(gid_t));
3514 assert_cc(UID_INVALID == (uid_t) GID_INVALID);
3515
3516 if (uid == 0) /* We don't keep track of root, and will never destroy it */
3517 return;
3518
3519 c = PTR_TO_UINT32(hashmap_get(*uid_refs, UID_TO_PTR(uid)));
3520
3521 n = c & ~DESTROY_IPC_FLAG;
3522 assert(n > 0);
3523 n--;
3524
3525 if (destroy_now && n == 0) {
3526 hashmap_remove(*uid_refs, UID_TO_PTR(uid));
3527
3528 if (c & DESTROY_IPC_FLAG) {
3529 log_debug("%s " UID_FMT " is no longer referenced, cleaning up its IPC.",
3530 _clean_ipc == clean_ipc_by_uid ? "UID" : "GID",
3531 uid);
3532 (void) _clean_ipc(uid);
3533 }
3534 } else {
3535 c = n | (c & DESTROY_IPC_FLAG);
3536 assert_se(hashmap_update(*uid_refs, UID_TO_PTR(uid), UINT32_TO_PTR(c)) >= 0);
3537 }
3538 }
3539
3540 void manager_unref_uid(Manager *m, uid_t uid, bool destroy_now) {
3541 manager_unref_uid_internal(m, &m->uid_refs, uid, destroy_now, clean_ipc_by_uid);
3542 }
3543
3544 void manager_unref_gid(Manager *m, gid_t gid, bool destroy_now) {
3545 manager_unref_uid_internal(m, &m->gid_refs, (uid_t) gid, destroy_now, clean_ipc_by_gid);
3546 }
3547
3548 static int manager_ref_uid_internal(
3549 Manager *m,
3550 Hashmap **uid_refs,
3551 uid_t uid,
3552 bool clean_ipc) {
3553
3554 uint32_t c, n;
3555 int r;
3556
3557 assert(m);
3558 assert(uid_refs);
3559 assert(uid_is_valid(uid));
3560
3561 /* A generic implementation, covering both manager_ref_uid() and manager_ref_gid(), under the assumption
3562 * that uid_t and gid_t are actually defined the same way, with the same validity rules. */
3563
3564 assert_cc(sizeof(uid_t) == sizeof(gid_t));
3565 assert_cc(UID_INVALID == (uid_t) GID_INVALID);
3566
3567 if (uid == 0) /* We don't keep track of root, and will never destroy it */
3568 return 0;
3569
3570 r = hashmap_ensure_allocated(uid_refs, &trivial_hash_ops);
3571 if (r < 0)
3572 return r;
3573
3574 c = PTR_TO_UINT32(hashmap_get(*uid_refs, UID_TO_PTR(uid)));
3575
3576 n = c & ~DESTROY_IPC_FLAG;
3577 n++;
3578
3579 if (n & DESTROY_IPC_FLAG) /* check for overflow */
3580 return -EOVERFLOW;
3581
3582 c = n | (c & DESTROY_IPC_FLAG) | (clean_ipc ? DESTROY_IPC_FLAG : 0);
3583
3584 return hashmap_replace(*uid_refs, UID_TO_PTR(uid), UINT32_TO_PTR(c));
3585 }
3586
3587 int manager_ref_uid(Manager *m, uid_t uid, bool clean_ipc) {
3588 return manager_ref_uid_internal(m, &m->uid_refs, uid, clean_ipc);
3589 }
3590
3591 int manager_ref_gid(Manager *m, gid_t gid, bool clean_ipc) {
3592 return manager_ref_uid_internal(m, &m->gid_refs, (uid_t) gid, clean_ipc);
3593 }
3594
3595 static void manager_vacuum_uid_refs_internal(
3596 Manager *m,
3597 Hashmap **uid_refs,
3598 int (*_clean_ipc)(uid_t uid)) {
3599
3600 Iterator i;
3601 void *p, *k;
3602
3603 assert(m);
3604 assert(uid_refs);
3605 assert(_clean_ipc);
3606
3607 HASHMAP_FOREACH_KEY(p, k, *uid_refs, i) {
3608 uint32_t c, n;
3609 uid_t uid;
3610
3611 uid = PTR_TO_UID(k);
3612 c = PTR_TO_UINT32(p);
3613
3614 n = c & ~DESTROY_IPC_FLAG;
3615 if (n > 0)
3616 continue;
3617
3618 if (c & DESTROY_IPC_FLAG) {
3619 log_debug("Found unreferenced %s " UID_FMT " after reload/reexec. Cleaning up.",
3620 _clean_ipc == clean_ipc_by_uid ? "UID" : "GID",
3621 uid);
3622 (void) _clean_ipc(uid);
3623 }
3624
3625 assert_se(hashmap_remove(*uid_refs, k) == p);
3626 }
3627 }
3628
3629 void manager_vacuum_uid_refs(Manager *m) {
3630 manager_vacuum_uid_refs_internal(m, &m->uid_refs, clean_ipc_by_uid);
3631 }
3632
3633 void manager_vacuum_gid_refs(Manager *m) {
3634 manager_vacuum_uid_refs_internal(m, &m->gid_refs, clean_ipc_by_gid);
3635 }
3636
3637 static void manager_serialize_uid_refs_internal(
3638 Manager *m,
3639 FILE *f,
3640 Hashmap **uid_refs,
3641 const char *field_name) {
3642
3643 Iterator i;
3644 void *p, *k;
3645
3646 assert(m);
3647 assert(f);
3648 assert(uid_refs);
3649 assert(field_name);
3650
3651 /* Serialize the UID reference table. Or actually, just the IPC destruction flag of it, as the actual counter
3652 * of it is better rebuild after a reload/reexec. */
3653
3654 HASHMAP_FOREACH_KEY(p, k, *uid_refs, i) {
3655 uint32_t c;
3656 uid_t uid;
3657
3658 uid = PTR_TO_UID(k);
3659 c = PTR_TO_UINT32(p);
3660
3661 if (!(c & DESTROY_IPC_FLAG))
3662 continue;
3663
3664 fprintf(f, "%s=" UID_FMT "\n", field_name, uid);
3665 }
3666 }
3667
3668 void manager_serialize_uid_refs(Manager *m, FILE *f) {
3669 manager_serialize_uid_refs_internal(m, f, &m->uid_refs, "destroy-ipc-uid");
3670 }
3671
3672 void manager_serialize_gid_refs(Manager *m, FILE *f) {
3673 manager_serialize_uid_refs_internal(m, f, &m->gid_refs, "destroy-ipc-gid");
3674 }
3675
3676 static void manager_deserialize_uid_refs_one_internal(
3677 Manager *m,
3678 Hashmap** uid_refs,
3679 const char *value) {
3680
3681 uid_t uid;
3682 uint32_t c;
3683 int r;
3684
3685 assert(m);
3686 assert(uid_refs);
3687 assert(value);
3688
3689 r = parse_uid(value, &uid);
3690 if (r < 0 || uid == 0) {
3691 log_debug("Unable to parse UID reference serialization");
3692 return;
3693 }
3694
3695 r = hashmap_ensure_allocated(uid_refs, &trivial_hash_ops);
3696 if (r < 0) {
3697 log_oom();
3698 return;
3699 }
3700
3701 c = PTR_TO_UINT32(hashmap_get(*uid_refs, UID_TO_PTR(uid)));
3702 if (c & DESTROY_IPC_FLAG)
3703 return;
3704
3705 c |= DESTROY_IPC_FLAG;
3706
3707 r = hashmap_replace(*uid_refs, UID_TO_PTR(uid), UINT32_TO_PTR(c));
3708 if (r < 0) {
3709 log_debug("Failed to add UID reference entry");
3710 return;
3711 }
3712 }
3713
3714 void manager_deserialize_uid_refs_one(Manager *m, const char *value) {
3715 manager_deserialize_uid_refs_one_internal(m, &m->uid_refs, value);
3716 }
3717
3718 void manager_deserialize_gid_refs_one(Manager *m, const char *value) {
3719 manager_deserialize_uid_refs_one_internal(m, &m->gid_refs, value);
3720 }
3721
3722 int manager_dispatch_user_lookup_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
3723 struct buffer {
3724 uid_t uid;
3725 gid_t gid;
3726 char unit_name[UNIT_NAME_MAX+1];
3727 } _packed_ buffer;
3728
3729 Manager *m = userdata;
3730 ssize_t l;
3731 size_t n;
3732 Unit *u;
3733
3734 assert_se(source);
3735 assert_se(m);
3736
3737 /* Invoked whenever a child process succeeded resolving its user/group to use and sent us the resulting UID/GID
3738 * in a datagram. We parse the datagram here and pass it off to the unit, so that it can add a reference to the
3739 * UID/GID so that it can destroy the UID/GID's IPC objects when the reference counter drops to 0. */
3740
3741 l = recv(fd, &buffer, sizeof(buffer), MSG_DONTWAIT);
3742 if (l < 0) {
3743 if (errno == EINTR || errno == EAGAIN)
3744 return 0;
3745
3746 return log_error_errno(errno, "Failed to read from user lookup fd: %m");
3747 }
3748
3749 if ((size_t) l <= offsetof(struct buffer, unit_name)) {
3750 log_warning("Received too short user lookup message, ignoring.");
3751 return 0;
3752 }
3753
3754 if ((size_t) l > offsetof(struct buffer, unit_name) + UNIT_NAME_MAX) {
3755 log_warning("Received too long user lookup message, ignoring.");
3756 return 0;
3757 }
3758
3759 if (!uid_is_valid(buffer.uid) && !gid_is_valid(buffer.gid)) {
3760 log_warning("Got user lookup message with invalid UID/GID pair, ignoring.");
3761 return 0;
3762 }
3763
3764 n = (size_t) l - offsetof(struct buffer, unit_name);
3765 if (memchr(buffer.unit_name, 0, n)) {
3766 log_warning("Received lookup message with embedded NUL character, ignoring.");
3767 return 0;
3768 }
3769
3770 buffer.unit_name[n] = 0;
3771 u = manager_get_unit(m, buffer.unit_name);
3772 if (!u) {
3773 log_debug("Got user lookup message but unit doesn't exist, ignoring.");
3774 return 0;
3775 }
3776
3777 log_unit_debug(u, "User lookup succeeded: uid=" UID_FMT " gid=" GID_FMT, buffer.uid, buffer.gid);
3778
3779 unit_notify_user_lookup(u, buffer.uid, buffer.gid);
3780 return 0;
3781 }
3782
3783 static const char *const manager_state_table[_MANAGER_STATE_MAX] = {
3784 [MANAGER_INITIALIZING] = "initializing",
3785 [MANAGER_STARTING] = "starting",
3786 [MANAGER_RUNNING] = "running",
3787 [MANAGER_DEGRADED] = "degraded",
3788 [MANAGER_MAINTENANCE] = "maintenance",
3789 [MANAGER_STOPPING] = "stopping",
3790 };
3791
3792 DEFINE_STRING_TABLE_LOOKUP(manager_state, ManagerState);
Unnamed repository; edit this file 'description' to name the repository.