]> git.ipfire.org Git - thirdparty/systemd.git/blob - src/core/manager.c
tree-wide: use mfree more
[thirdparty/systemd.git] / src / core / manager.c
1 /***
2 This file is part of systemd.
3
4 Copyright 2010 Lennart Poettering
5
6 systemd is free software; you can redistribute it and/or modify it
7 under the terms of the GNU Lesser General Public License as published by
8 the Free Software Foundation; either version 2.1 of the License, or
9 (at your option) any later version.
10
11 systemd is distributed in the hope that it will be useful, but
12 WITHOUT ANY WARRANTY; without even the implied warranty of
13 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 Lesser General Public License for more details.
15
16 You should have received a copy of the GNU Lesser General Public License
17 along with systemd; If not, see <http://www.gnu.org/licenses/>.
18 ***/
19
20 #include <dirent.h>
21 #include <errno.h>
22 #include <fcntl.h>
23 #include <linux/kd.h>
24 #include <signal.h>
25 #include <string.h>
26 #include <sys/epoll.h>
27 #include <sys/inotify.h>
28 #include <sys/ioctl.h>
29 #include <sys/reboot.h>
30 #include <sys/timerfd.h>
31 #include <sys/wait.h>
32 #include <unistd.h>
33
34 #ifdef HAVE_AUDIT
35 #include <libaudit.h>
36 #endif
37
38 #include "sd-daemon.h"
39 #include "sd-messages.h"
40
41 #include "alloc-util.h"
42 #include "audit-fd.h"
43 #include "boot-timestamps.h"
44 #include "bus-common-errors.h"
45 #include "bus-error.h"
46 #include "bus-kernel.h"
47 #include "bus-util.h"
48 #include "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 "exit-status.h"
57 #include "fd-util.h"
58 #include "fileio.h"
59 #include "fs-util.h"
60 #include "hashmap.h"
61 #include "io-util.h"
62 #include "locale-setup.h"
63 #include "log.h"
64 #include "macro.h"
65 #include "manager.h"
66 #include "missing.h"
67 #include "mkdir.h"
68 #include "parse-util.h"
69 #include "path-lookup.h"
70 #include "path-util.h"
71 #include "process-util.h"
72 #include "ratelimit.h"
73 #include "rm-rf.h"
74 #include "signal-util.h"
75 #include "special.h"
76 #include "stat-util.h"
77 #include "string-table.h"
78 #include "string-util.h"
79 #include "strv.h"
80 #include "terminal-util.h"
81 #include "time-util.h"
82 #include "transaction.h"
83 #include "umask-util.h"
84 #include "unit-name.h"
85 #include "user-util.h"
86 #include "util.h"
87 #include "virt.h"
88 #include "watchdog.h"
89
90 #define NOTIFY_RCVBUF_SIZE (8*1024*1024)
91 #define CGROUPS_AGENT_RCVBUF_SIZE (8*1024*1024)
92
93 /* Initial delay and the interval for printing status messages about running jobs */
94 #define JOBS_IN_PROGRESS_WAIT_USEC (5*USEC_PER_SEC)
95 #define JOBS_IN_PROGRESS_PERIOD_USEC (USEC_PER_SEC / 3)
96 #define JOBS_IN_PROGRESS_PERIOD_DIVISOR 3
97
98 static int manager_dispatch_notify_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata);
99 static int manager_dispatch_cgroups_agent_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata);
100 static int manager_dispatch_signal_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata);
101 static int manager_dispatch_time_change_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata);
102 static int manager_dispatch_idle_pipe_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata);
103 static int manager_dispatch_user_lookup_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata);
104 static int manager_dispatch_jobs_in_progress(sd_event_source *source, usec_t usec, void *userdata);
105 static int manager_dispatch_run_queue(sd_event_source *source, void *userdata);
106 static int manager_run_generators(Manager *m);
107
108 static void manager_watch_jobs_in_progress(Manager *m) {
109 usec_t next;
110 int r;
111
112 assert(m);
113
114 if (m->jobs_in_progress_event_source)
115 return;
116
117 next = now(CLOCK_MONOTONIC) + JOBS_IN_PROGRESS_WAIT_USEC;
118 r = sd_event_add_time(
119 m->event,
120 &m->jobs_in_progress_event_source,
121 CLOCK_MONOTONIC,
122 next, 0,
123 manager_dispatch_jobs_in_progress, m);
124 if (r < 0)
125 return;
126
127 (void) sd_event_source_set_description(m->jobs_in_progress_event_source, "manager-jobs-in-progress");
128 }
129
130 #define CYLON_BUFFER_EXTRA (2*(sizeof(ANSI_RED)-1) + sizeof(ANSI_HIGHLIGHT_RED)-1 + 2*(sizeof(ANSI_NORMAL)-1))
131
132 static void draw_cylon(char buffer[], size_t buflen, unsigned width, unsigned pos) {
133 char *p = buffer;
134
135 assert(buflen >= CYLON_BUFFER_EXTRA + width + 1);
136 assert(pos <= width+1); /* 0 or width+1 mean that the center light is behind the corner */
137
138 if (pos > 1) {
139 if (pos > 2)
140 p = mempset(p, ' ', pos-2);
141 if (log_get_show_color())
142 p = stpcpy(p, ANSI_RED);
143 *p++ = '*';
144 }
145
146 if (pos > 0 && pos <= width) {
147 if (log_get_show_color())
148 p = stpcpy(p, ANSI_HIGHLIGHT_RED);
149 *p++ = '*';
150 }
151
152 if (log_get_show_color())
153 p = stpcpy(p, ANSI_NORMAL);
154
155 if (pos < width) {
156 if (log_get_show_color())
157 p = stpcpy(p, ANSI_RED);
158 *p++ = '*';
159 if (pos < width-1)
160 p = mempset(p, ' ', width-1-pos);
161 if (log_get_show_color())
162 strcpy(p, ANSI_NORMAL);
163 }
164 }
165
166 void manager_flip_auto_status(Manager *m, bool enable) {
167 assert(m);
168
169 if (enable) {
170 if (m->show_status == SHOW_STATUS_AUTO)
171 manager_set_show_status(m, SHOW_STATUS_TEMPORARY);
172 } else {
173 if (m->show_status == SHOW_STATUS_TEMPORARY)
174 manager_set_show_status(m, SHOW_STATUS_AUTO);
175 }
176 }
177
178 static void manager_print_jobs_in_progress(Manager *m) {
179 _cleanup_free_ char *job_of_n = NULL;
180 Iterator i;
181 Job *j;
182 unsigned counter = 0, print_nr;
183 char cylon[6 + CYLON_BUFFER_EXTRA + 1];
184 unsigned cylon_pos;
185 char time[FORMAT_TIMESPAN_MAX], limit[FORMAT_TIMESPAN_MAX] = "no limit";
186 uint64_t x;
187
188 assert(m);
189 assert(m->n_running_jobs > 0);
190
191 manager_flip_auto_status(m, true);
192
193 print_nr = (m->jobs_in_progress_iteration / JOBS_IN_PROGRESS_PERIOD_DIVISOR) % m->n_running_jobs;
194
195 HASHMAP_FOREACH(j, m->jobs, i)
196 if (j->state == JOB_RUNNING && counter++ == print_nr)
197 break;
198
199 /* m->n_running_jobs must be consistent with the contents of m->jobs,
200 * so the above loop must have succeeded in finding j. */
201 assert(counter == print_nr + 1);
202 assert(j);
203
204 cylon_pos = m->jobs_in_progress_iteration % 14;
205 if (cylon_pos >= 8)
206 cylon_pos = 14 - cylon_pos;
207 draw_cylon(cylon, sizeof(cylon), 6, cylon_pos);
208
209 m->jobs_in_progress_iteration++;
210
211 if (m->n_running_jobs > 1) {
212 if (asprintf(&job_of_n, "(%u of %u) ", counter, m->n_running_jobs) < 0)
213 job_of_n = NULL;
214 }
215
216 format_timespan(time, sizeof(time), now(CLOCK_MONOTONIC) - j->begin_usec, 1*USEC_PER_SEC);
217 if (job_get_timeout(j, &x) > 0)
218 format_timespan(limit, sizeof(limit), x - j->begin_usec, 1*USEC_PER_SEC);
219
220 manager_status_printf(m, STATUS_TYPE_EPHEMERAL, cylon,
221 "%sA %s job is running for %s (%s / %s)",
222 strempty(job_of_n),
223 job_type_to_string(j->type),
224 unit_description(j->unit),
225 time, limit);
226 }
227
228 static int have_ask_password(void) {
229 _cleanup_closedir_ DIR *dir;
230
231 dir = opendir("/run/systemd/ask-password");
232 if (!dir) {
233 if (errno == ENOENT)
234 return false;
235 else
236 return -errno;
237 }
238
239 for (;;) {
240 struct dirent *de;
241
242 errno = 0;
243 de = readdir(dir);
244 if (!de && errno > 0)
245 return -errno;
246 if (!de)
247 return false;
248
249 if (startswith(de->d_name, "ask."))
250 return true;
251 }
252 }
253
254 static int manager_dispatch_ask_password_fd(sd_event_source *source,
255 int fd, uint32_t revents, void *userdata) {
256 Manager *m = userdata;
257
258 assert(m);
259
260 flush_fd(fd);
261
262 m->have_ask_password = have_ask_password();
263 if (m->have_ask_password < 0)
264 /* Log error but continue. Negative have_ask_password
265 * is treated as unknown status. */
266 log_error_errno(m->have_ask_password, "Failed to list /run/systemd/ask-password: %m");
267
268 return 0;
269 }
270
271 static void manager_close_ask_password(Manager *m) {
272 assert(m);
273
274 m->ask_password_event_source = sd_event_source_unref(m->ask_password_event_source);
275 m->ask_password_inotify_fd = safe_close(m->ask_password_inotify_fd);
276 m->have_ask_password = -EINVAL;
277 }
278
279 static int manager_check_ask_password(Manager *m) {
280 int r;
281
282 assert(m);
283
284 if (!m->ask_password_event_source) {
285 assert(m->ask_password_inotify_fd < 0);
286
287 mkdir_p_label("/run/systemd/ask-password", 0755);
288
289 m->ask_password_inotify_fd = inotify_init1(IN_NONBLOCK|IN_CLOEXEC);
290 if (m->ask_password_inotify_fd < 0)
291 return log_error_errno(errno, "inotify_init1() failed: %m");
292
293 if (inotify_add_watch(m->ask_password_inotify_fd, "/run/systemd/ask-password", IN_CREATE|IN_DELETE|IN_MOVE) < 0) {
294 log_error_errno(errno, "Failed to add watch on /run/systemd/ask-password: %m");
295 manager_close_ask_password(m);
296 return -errno;
297 }
298
299 r = sd_event_add_io(m->event, &m->ask_password_event_source,
300 m->ask_password_inotify_fd, EPOLLIN,
301 manager_dispatch_ask_password_fd, m);
302 if (r < 0) {
303 log_error_errno(errno, "Failed to add event source for /run/systemd/ask-password: %m");
304 manager_close_ask_password(m);
305 return -errno;
306 }
307
308 (void) sd_event_source_set_description(m->ask_password_event_source, "manager-ask-password");
309
310 /* Queries might have been added meanwhile... */
311 manager_dispatch_ask_password_fd(m->ask_password_event_source,
312 m->ask_password_inotify_fd, EPOLLIN, m);
313 }
314
315 return m->have_ask_password;
316 }
317
318 static int manager_watch_idle_pipe(Manager *m) {
319 int r;
320
321 assert(m);
322
323 if (m->idle_pipe_event_source)
324 return 0;
325
326 if (m->idle_pipe[2] < 0)
327 return 0;
328
329 r = sd_event_add_io(m->event, &m->idle_pipe_event_source, m->idle_pipe[2], EPOLLIN, manager_dispatch_idle_pipe_fd, m);
330 if (r < 0)
331 return log_error_errno(r, "Failed to watch idle pipe: %m");
332
333 (void) sd_event_source_set_description(m->idle_pipe_event_source, "manager-idle-pipe");
334
335 return 0;
336 }
337
338 static void manager_close_idle_pipe(Manager *m) {
339 assert(m);
340
341 m->idle_pipe_event_source = sd_event_source_unref(m->idle_pipe_event_source);
342
343 safe_close_pair(m->idle_pipe);
344 safe_close_pair(m->idle_pipe + 2);
345 }
346
347 static int manager_setup_time_change(Manager *m) {
348 int r;
349
350 /* We only care for the cancellation event, hence we set the
351 * timeout to the latest possible value. */
352 struct itimerspec its = {
353 .it_value.tv_sec = TIME_T_MAX,
354 };
355
356 assert(m);
357 assert_cc(sizeof(time_t) == sizeof(TIME_T_MAX));
358
359 if (m->test_run)
360 return 0;
361
362 /* Uses TFD_TIMER_CANCEL_ON_SET to get notifications whenever
363 * CLOCK_REALTIME makes a jump relative to CLOCK_MONOTONIC */
364
365 m->time_change_fd = timerfd_create(CLOCK_REALTIME, TFD_NONBLOCK|TFD_CLOEXEC);
366 if (m->time_change_fd < 0)
367 return log_error_errno(errno, "Failed to create timerfd: %m");
368
369 if (timerfd_settime(m->time_change_fd, TFD_TIMER_ABSTIME|TFD_TIMER_CANCEL_ON_SET, &its, NULL) < 0) {
370 log_debug_errno(errno, "Failed to set up TFD_TIMER_CANCEL_ON_SET, ignoring: %m");
371 m->time_change_fd = safe_close(m->time_change_fd);
372 return 0;
373 }
374
375 r = sd_event_add_io(m->event, &m->time_change_event_source, m->time_change_fd, EPOLLIN, manager_dispatch_time_change_fd, m);
376 if (r < 0)
377 return log_error_errno(r, "Failed to create time change event source: %m");
378
379 (void) sd_event_source_set_description(m->time_change_event_source, "manager-time-change");
380
381 log_debug("Set up TFD_TIMER_CANCEL_ON_SET timerfd.");
382
383 return 0;
384 }
385
386 static int enable_special_signals(Manager *m) {
387 _cleanup_close_ int fd = -1;
388
389 assert(m);
390
391 if (m->test_run)
392 return 0;
393
394 /* Enable that we get SIGINT on control-alt-del. In containers
395 * this will fail with EPERM (older) or EINVAL (newer), so
396 * ignore that. */
397 if (reboot(RB_DISABLE_CAD) < 0 && errno != EPERM && errno != EINVAL)
398 log_warning_errno(errno, "Failed to enable ctrl-alt-del handling: %m");
399
400 fd = open_terminal("/dev/tty0", O_RDWR|O_NOCTTY|O_CLOEXEC);
401 if (fd < 0) {
402 /* Support systems without virtual console */
403 if (fd != -ENOENT)
404 log_warning_errno(errno, "Failed to open /dev/tty0: %m");
405 } else {
406 /* Enable that we get SIGWINCH on kbrequest */
407 if (ioctl(fd, KDSIGACCEPT, SIGWINCH) < 0)
408 log_warning_errno(errno, "Failed to enable kbrequest handling: %m");
409 }
410
411 return 0;
412 }
413
414 static int manager_setup_signals(Manager *m) {
415 struct sigaction sa = {
416 .sa_handler = SIG_DFL,
417 .sa_flags = SA_NOCLDSTOP|SA_RESTART,
418 };
419 sigset_t mask;
420 int r;
421
422 assert(m);
423
424 assert_se(sigaction(SIGCHLD, &sa, NULL) == 0);
425
426 /* We make liberal use of realtime signals here. On
427 * Linux/glibc we have 30 of them (with the exception of Linux
428 * on hppa, see below), between SIGRTMIN+0 ... SIGRTMIN+30
429 * (aka SIGRTMAX). */
430
431 assert_se(sigemptyset(&mask) == 0);
432 sigset_add_many(&mask,
433 SIGCHLD, /* Child died */
434 SIGTERM, /* Reexecute daemon */
435 SIGHUP, /* Reload configuration */
436 SIGUSR1, /* systemd/upstart: reconnect to D-Bus */
437 SIGUSR2, /* systemd: dump status */
438 SIGINT, /* Kernel sends us this on control-alt-del */
439 SIGWINCH, /* Kernel sends us this on kbrequest (alt-arrowup) */
440 SIGPWR, /* Some kernel drivers and upsd send us this on power failure */
441
442 SIGRTMIN+0, /* systemd: start default.target */
443 SIGRTMIN+1, /* systemd: isolate rescue.target */
444 SIGRTMIN+2, /* systemd: isolate emergency.target */
445 SIGRTMIN+3, /* systemd: start halt.target */
446 SIGRTMIN+4, /* systemd: start poweroff.target */
447 SIGRTMIN+5, /* systemd: start reboot.target */
448 SIGRTMIN+6, /* systemd: start kexec.target */
449
450 /* ... space for more special targets ... */
451
452 SIGRTMIN+13, /* systemd: Immediate halt */
453 SIGRTMIN+14, /* systemd: Immediate poweroff */
454 SIGRTMIN+15, /* systemd: Immediate reboot */
455 SIGRTMIN+16, /* systemd: Immediate kexec */
456
457 /* ... space for more immediate system state changes ... */
458
459 SIGRTMIN+20, /* systemd: enable status messages */
460 SIGRTMIN+21, /* systemd: disable status messages */
461 SIGRTMIN+22, /* systemd: set log level to LOG_DEBUG */
462 SIGRTMIN+23, /* systemd: set log level to LOG_INFO */
463 SIGRTMIN+24, /* systemd: Immediate exit (--user only) */
464
465 /* .. one free signal here ... */
466
467 #if !defined(__hppa64__) && !defined(__hppa__)
468 /* Apparently Linux on hppa has fewer RT
469 * signals (SIGRTMAX is SIGRTMIN+25 there),
470 * hence let's not try to make use of them
471 * here. Since these commands are accessible
472 * by different means and only really a safety
473 * net, the missing functionality on hppa
474 * shouldn't matter. */
475
476 SIGRTMIN+26, /* systemd: set log target to journal-or-kmsg */
477 SIGRTMIN+27, /* systemd: set log target to console */
478 SIGRTMIN+28, /* systemd: set log target to kmsg */
479 SIGRTMIN+29, /* systemd: set log target to syslog-or-kmsg (obsolete) */
480
481 /* ... one free signal here SIGRTMIN+30 ... */
482 #endif
483 -1);
484 assert_se(sigprocmask(SIG_SETMASK, &mask, NULL) == 0);
485
486 m->signal_fd = signalfd(-1, &mask, SFD_NONBLOCK|SFD_CLOEXEC);
487 if (m->signal_fd < 0)
488 return -errno;
489
490 r = sd_event_add_io(m->event, &m->signal_event_source, m->signal_fd, EPOLLIN, manager_dispatch_signal_fd, m);
491 if (r < 0)
492 return r;
493
494 (void) sd_event_source_set_description(m->signal_event_source, "manager-signal");
495
496 /* Process signals a bit earlier than the rest of things, but later than notify_fd processing, so that the
497 * notify processing can still figure out to which process/service a message belongs, before we reap the
498 * process. Also, process this before handling cgroup notifications, so that we always collect child exit
499 * status information before detecting that there's no process in a cgroup. */
500 r = sd_event_source_set_priority(m->signal_event_source, SD_EVENT_PRIORITY_NORMAL-6);
501 if (r < 0)
502 return r;
503
504 if (MANAGER_IS_SYSTEM(m))
505 return enable_special_signals(m);
506
507 return 0;
508 }
509
510 static void manager_clean_environment(Manager *m) {
511 assert(m);
512
513 /* Let's remove some environment variables that we
514 * need ourselves to communicate with our clients */
515 strv_env_unset_many(
516 m->environment,
517 "NOTIFY_SOCKET",
518 "MAINPID",
519 "MANAGERPID",
520 "LISTEN_PID",
521 "LISTEN_FDS",
522 "LISTEN_FDNAMES",
523 "WATCHDOG_PID",
524 "WATCHDOG_USEC",
525 "INVOCATION_ID",
526 NULL);
527 }
528
529 static int manager_default_environment(Manager *m) {
530 assert(m);
531
532 if (MANAGER_IS_SYSTEM(m)) {
533 /* The system manager always starts with a clean
534 * environment for its children. It does not import
535 * the kernel or the parents exported variables.
536 *
537 * The initial passed environ is untouched to keep
538 * /proc/self/environ valid; it is used for tagging
539 * the init process inside containers. */
540 m->environment = strv_new("PATH=" DEFAULT_PATH,
541 NULL);
542
543 /* Import locale variables LC_*= from configuration */
544 locale_setup(&m->environment);
545 } else {
546 /* The user manager passes its own environment
547 * along to its children. */
548 m->environment = strv_copy(environ);
549 }
550
551 if (!m->environment)
552 return -ENOMEM;
553
554 manager_clean_environment(m);
555 strv_sort(m->environment);
556
557 return 0;
558 }
559
560 int manager_new(UnitFileScope scope, bool test_run, Manager **_m) {
561 Manager *m;
562 int r;
563
564 assert(_m);
565 assert(IN_SET(scope, UNIT_FILE_SYSTEM, UNIT_FILE_USER));
566
567 m = new0(Manager, 1);
568 if (!m)
569 return -ENOMEM;
570
571 m->unit_file_scope = scope;
572 m->exit_code = _MANAGER_EXIT_CODE_INVALID;
573 m->default_timer_accuracy_usec = USEC_PER_MINUTE;
574 m->default_tasks_accounting = true;
575 m->default_tasks_max = UINT64_MAX;
576
577 #ifdef ENABLE_EFI
578 if (MANAGER_IS_SYSTEM(m) && detect_container() <= 0)
579 boot_timestamps(&m->userspace_timestamp, &m->firmware_timestamp, &m->loader_timestamp);
580 #endif
581
582 /* Prepare log fields we can use for structured logging */
583 if (MANAGER_IS_SYSTEM(m)) {
584 m->unit_log_field = "UNIT=";
585 m->unit_log_format_string = "UNIT=%s";
586
587 m->invocation_log_field = "INVOCATION_ID=";
588 m->invocation_log_format_string = "INVOCATION_ID=" SD_ID128_FORMAT_STR;
589 } else {
590 m->unit_log_field = "USER_UNIT=";
591 m->unit_log_format_string = "USER_UNIT=%s";
592
593 m->invocation_log_field = "USER_INVOCATION_ID=";
594 m->invocation_log_format_string = "USER_INVOCATION_ID=" SD_ID128_FORMAT_STR;
595 }
596
597 m->idle_pipe[0] = m->idle_pipe[1] = m->idle_pipe[2] = m->idle_pipe[3] = -1;
598
599 m->pin_cgroupfs_fd = m->notify_fd = m->cgroups_agent_fd = m->signal_fd = m->time_change_fd =
600 m->dev_autofs_fd = m->private_listen_fd = m->cgroup_inotify_fd =
601 m->ask_password_inotify_fd = -1;
602
603 m->user_lookup_fds[0] = m->user_lookup_fds[1] = -1;
604
605 m->current_job_id = 1; /* start as id #1, so that we can leave #0 around as "null-like" value */
606
607 m->have_ask_password = -EINVAL; /* we don't know */
608 m->first_boot = -1;
609
610 m->test_run = test_run;
611
612 /* Reboot immediately if the user hits C-A-D more often than 7x per 2s */
613 RATELIMIT_INIT(m->ctrl_alt_del_ratelimit, 2 * USEC_PER_SEC, 7);
614
615 r = manager_default_environment(m);
616 if (r < 0)
617 goto fail;
618
619 r = hashmap_ensure_allocated(&m->units, &string_hash_ops);
620 if (r < 0)
621 goto fail;
622
623 r = hashmap_ensure_allocated(&m->jobs, NULL);
624 if (r < 0)
625 goto fail;
626
627 r = hashmap_ensure_allocated(&m->cgroup_unit, &string_hash_ops);
628 if (r < 0)
629 goto fail;
630
631 r = hashmap_ensure_allocated(&m->watch_bus, &string_hash_ops);
632 if (r < 0)
633 goto fail;
634
635 r = sd_event_default(&m->event);
636 if (r < 0)
637 goto fail;
638
639 r = sd_event_add_defer(m->event, &m->run_queue_event_source, manager_dispatch_run_queue, m);
640 if (r < 0)
641 goto fail;
642
643 r = sd_event_source_set_priority(m->run_queue_event_source, SD_EVENT_PRIORITY_IDLE);
644 if (r < 0)
645 goto fail;
646
647 r = sd_event_source_set_enabled(m->run_queue_event_source, SD_EVENT_OFF);
648 if (r < 0)
649 goto fail;
650
651 (void) sd_event_source_set_description(m->run_queue_event_source, "manager-run-queue");
652
653 r = manager_setup_signals(m);
654 if (r < 0)
655 goto fail;
656
657 r = manager_setup_cgroup(m);
658 if (r < 0)
659 goto fail;
660
661 r = manager_setup_time_change(m);
662 if (r < 0)
663 goto fail;
664
665 m->udev = udev_new();
666 if (!m->udev) {
667 r = -ENOMEM;
668 goto fail;
669 }
670
671 /* Note that we do not set up the notify fd here. We do that after deserialization,
672 * since they might have gotten serialized across the reexec. */
673
674 m->taint_usr = dir_is_empty("/usr") > 0;
675
676 *_m = m;
677 return 0;
678
679 fail:
680 manager_free(m);
681 return r;
682 }
683
684 static int manager_setup_notify(Manager *m) {
685 int r;
686
687 if (m->test_run)
688 return 0;
689
690 if (m->notify_fd < 0) {
691 _cleanup_close_ int fd = -1;
692 union sockaddr_union sa = {
693 .sa.sa_family = AF_UNIX,
694 };
695 static const int one = 1;
696 const char *e;
697
698 /* First free all secondary fields */
699 m->notify_socket = mfree(m->notify_socket);
700 m->notify_event_source = sd_event_source_unref(m->notify_event_source);
701
702 fd = socket(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC|SOCK_NONBLOCK, 0);
703 if (fd < 0)
704 return log_error_errno(errno, "Failed to allocate notification socket: %m");
705
706 fd_inc_rcvbuf(fd, NOTIFY_RCVBUF_SIZE);
707
708 e = manager_get_runtime_prefix(m);
709 if (!e) {
710 log_error("Failed to determine runtime prefix.");
711 return -EINVAL;
712 }
713
714 m->notify_socket = strappend(e, "/systemd/notify");
715 if (!m->notify_socket)
716 return log_oom();
717
718 (void) mkdir_parents_label(m->notify_socket, 0755);
719 (void) unlink(m->notify_socket);
720
721 strncpy(sa.un.sun_path, m->notify_socket, sizeof(sa.un.sun_path)-1);
722 r = bind(fd, &sa.sa, SOCKADDR_UN_LEN(sa.un));
723 if (r < 0)
724 return log_error_errno(errno, "bind(%s) failed: %m", sa.un.sun_path);
725
726 r = setsockopt(fd, SOL_SOCKET, SO_PASSCRED, &one, sizeof(one));
727 if (r < 0)
728 return log_error_errno(errno, "SO_PASSCRED failed: %m");
729
730 m->notify_fd = fd;
731 fd = -1;
732
733 log_debug("Using notification socket %s", m->notify_socket);
734 }
735
736 if (!m->notify_event_source) {
737 r = sd_event_add_io(m->event, &m->notify_event_source, m->notify_fd, EPOLLIN, manager_dispatch_notify_fd, m);
738 if (r < 0)
739 return log_error_errno(r, "Failed to allocate notify event source: %m");
740
741 /* Process notification messages a bit earlier than SIGCHLD, so that we can still identify to which
742 * service an exit message belongs. */
743 r = sd_event_source_set_priority(m->notify_event_source, SD_EVENT_PRIORITY_NORMAL-7);
744 if (r < 0)
745 return log_error_errno(r, "Failed to set priority of notify event source: %m");
746
747 (void) sd_event_source_set_description(m->notify_event_source, "manager-notify");
748 }
749
750 return 0;
751 }
752
753 static int manager_setup_cgroups_agent(Manager *m) {
754
755 static const union sockaddr_union sa = {
756 .un.sun_family = AF_UNIX,
757 .un.sun_path = "/run/systemd/cgroups-agent",
758 };
759 int r;
760
761 /* This creates a listening socket we receive cgroups agent messages on. We do not use D-Bus for delivering
762 * these messages from the cgroups agent binary to PID 1, as the cgroups agent binary is very short-living, and
763 * each instance of it needs a new D-Bus connection. Since D-Bus connections are SOCK_STREAM/AF_UNIX, on
764 * overloaded systems the backlog of the D-Bus socket becomes relevant, as not more than the configured number
765 * of D-Bus connections may be queued until the kernel will start dropping further incoming connections,
766 * possibly resulting in lost cgroups agent messages. To avoid this, we'll use a private SOCK_DGRAM/AF_UNIX
767 * socket, where no backlog is relevant as communication may take place without an actual connect() cycle, and
768 * we thus won't lose messages.
769 *
770 * Note that PID 1 will forward the agent message to system bus, so that the user systemd instance may listen
771 * to it. The system instance hence listens on this special socket, but the user instances listen on the system
772 * bus for these messages. */
773
774 if (m->test_run)
775 return 0;
776
777 if (!MANAGER_IS_SYSTEM(m))
778 return 0;
779
780 if (cg_unified(SYSTEMD_CGROUP_CONTROLLER) > 0) /* We don't need this anymore on the unified hierarchy */
781 return 0;
782
783 if (m->cgroups_agent_fd < 0) {
784 _cleanup_close_ int fd = -1;
785
786 /* First free all secondary fields */
787 m->cgroups_agent_event_source = sd_event_source_unref(m->cgroups_agent_event_source);
788
789 fd = socket(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC|SOCK_NONBLOCK, 0);
790 if (fd < 0)
791 return log_error_errno(errno, "Failed to allocate cgroups agent socket: %m");
792
793 fd_inc_rcvbuf(fd, CGROUPS_AGENT_RCVBUF_SIZE);
794
795 (void) unlink(sa.un.sun_path);
796
797 /* Only allow root to connect to this socket */
798 RUN_WITH_UMASK(0077)
799 r = bind(fd, &sa.sa, SOCKADDR_UN_LEN(sa.un));
800 if (r < 0)
801 return log_error_errno(errno, "bind(%s) failed: %m", sa.un.sun_path);
802
803 m->cgroups_agent_fd = fd;
804 fd = -1;
805 }
806
807 if (!m->cgroups_agent_event_source) {
808 r = sd_event_add_io(m->event, &m->cgroups_agent_event_source, m->cgroups_agent_fd, EPOLLIN, manager_dispatch_cgroups_agent_fd, m);
809 if (r < 0)
810 return log_error_errno(r, "Failed to allocate cgroups agent event source: %m");
811
812 /* Process cgroups notifications early, but after having processed service notification messages or
813 * SIGCHLD signals, so that a cgroup running empty is always just the last safety net of notification,
814 * and we collected the metadata the notification and SIGCHLD stuff offers first. Also see handling of
815 * cgroup inotify for the unified cgroup stuff. */
816 r = sd_event_source_set_priority(m->cgroups_agent_event_source, SD_EVENT_PRIORITY_NORMAL-5);
817 if (r < 0)
818 return log_error_errno(r, "Failed to set priority of cgroups agent event source: %m");
819
820 (void) sd_event_source_set_description(m->cgroups_agent_event_source, "manager-cgroups-agent");
821 }
822
823 return 0;
824 }
825
826 static int manager_setup_user_lookup_fd(Manager *m) {
827 int r;
828
829 assert(m);
830
831 /* Set up the socket pair used for passing UID/GID resolution results from forked off processes to PID
832 * 1. Background: we can't do name lookups (NSS) from PID 1, since it might involve IPC and thus activation,
833 * and we might hence deadlock on ourselves. Hence we do all user/group lookups asynchronously from the forked
834 * off processes right before executing the binaries to start. In order to be able to clean up any IPC objects
835 * created by a unit (see RemoveIPC=) we need to know in PID 1 the used UID/GID of the executed processes,
836 * hence we establish this communication channel so that forked off processes can pass their UID/GID
837 * information back to PID 1. The forked off processes send their resolved UID/GID to PID 1 in a simple
838 * datagram, along with their unit name, so that we can share one communication socket pair among all units for
839 * this purpose.
840 *
841 * You might wonder why we need a communication channel for this that is independent of the usual notification
842 * socket scheme (i.e. $NOTIFY_SOCKET). The primary difference is about trust: data sent via the $NOTIFY_SOCKET
843 * channel is only accepted if it originates from the right unit and if reception was enabled for it. The user
844 * lookup socket OTOH is only accessible by PID 1 and its children until they exec(), and always available.
845 *
846 * Note that this function is called under two circumstances: when we first initialize (in which case we
847 * allocate both the socket pair and the event source to listen on it), and when we deserialize after a reload
848 * (in which case the socket pair already exists but we still need to allocate the event source for it). */
849
850 if (m->user_lookup_fds[0] < 0) {
851
852 /* Free all secondary fields */
853 safe_close_pair(m->user_lookup_fds);
854 m->user_lookup_event_source = sd_event_source_unref(m->user_lookup_event_source);
855
856 if (socketpair(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC, 0, m->user_lookup_fds) < 0)
857 return log_error_errno(errno, "Failed to allocate user lookup socket: %m");
858
859 (void) fd_inc_rcvbuf(m->user_lookup_fds[0], NOTIFY_RCVBUF_SIZE);
860 }
861
862 if (!m->user_lookup_event_source) {
863 r = sd_event_add_io(m->event, &m->user_lookup_event_source, m->user_lookup_fds[0], EPOLLIN, manager_dispatch_user_lookup_fd, m);
864 if (r < 0)
865 return log_error_errno(errno, "Failed to allocate user lookup event source: %m");
866
867 /* Process even earlier than the notify event source, so that we always know first about valid UID/GID
868 * resolutions */
869 r = sd_event_source_set_priority(m->user_lookup_event_source, SD_EVENT_PRIORITY_NORMAL-8);
870 if (r < 0)
871 return log_error_errno(errno, "Failed to set priority ot user lookup event source: %m");
872
873 (void) sd_event_source_set_description(m->user_lookup_event_source, "user-lookup");
874 }
875
876 return 0;
877 }
878
879 static int manager_connect_bus(Manager *m, bool reexecuting) {
880 bool try_bus_connect;
881
882 assert(m);
883
884 if (m->test_run)
885 return 0;
886
887 try_bus_connect =
888 reexecuting ||
889 (MANAGER_IS_USER(m) && getenv("DBUS_SESSION_BUS_ADDRESS"));
890
891 /* Try to connect to the buses, if possible. */
892 return bus_init(m, try_bus_connect);
893 }
894
895 static unsigned manager_dispatch_cleanup_queue(Manager *m) {
896 Unit *u;
897 unsigned n = 0;
898
899 assert(m);
900
901 while ((u = m->cleanup_queue)) {
902 assert(u->in_cleanup_queue);
903
904 unit_free(u);
905 n++;
906 }
907
908 return n;
909 }
910
911 enum {
912 GC_OFFSET_IN_PATH, /* This one is on the path we were traveling */
913 GC_OFFSET_UNSURE, /* No clue */
914 GC_OFFSET_GOOD, /* We still need this unit */
915 GC_OFFSET_BAD, /* We don't need this unit anymore */
916 _GC_OFFSET_MAX
917 };
918
919 static void unit_gc_mark_good(Unit *u, unsigned gc_marker) {
920 Iterator i;
921 Unit *other;
922
923 u->gc_marker = gc_marker + GC_OFFSET_GOOD;
924
925 /* Recursively mark referenced units as GOOD as well */
926 SET_FOREACH(other, u->dependencies[UNIT_REFERENCES], i)
927 if (other->gc_marker == gc_marker + GC_OFFSET_UNSURE)
928 unit_gc_mark_good(other, gc_marker);
929 }
930
931 static void unit_gc_sweep(Unit *u, unsigned gc_marker) {
932 Iterator i;
933 Unit *other;
934 bool is_bad;
935
936 assert(u);
937
938 if (u->gc_marker == gc_marker + GC_OFFSET_GOOD ||
939 u->gc_marker == gc_marker + GC_OFFSET_BAD ||
940 u->gc_marker == gc_marker + GC_OFFSET_UNSURE ||
941 u->gc_marker == gc_marker + GC_OFFSET_IN_PATH)
942 return;
943
944 if (u->in_cleanup_queue)
945 goto bad;
946
947 if (unit_check_gc(u))
948 goto good;
949
950 u->gc_marker = gc_marker + GC_OFFSET_IN_PATH;
951
952 is_bad = true;
953
954 SET_FOREACH(other, u->dependencies[UNIT_REFERENCED_BY], i) {
955 unit_gc_sweep(other, gc_marker);
956
957 if (other->gc_marker == gc_marker + GC_OFFSET_GOOD)
958 goto good;
959
960 if (other->gc_marker != gc_marker + GC_OFFSET_BAD)
961 is_bad = false;
962 }
963
964 if (is_bad)
965 goto bad;
966
967 /* We were unable to find anything out about this entry, so
968 * let's investigate it later */
969 u->gc_marker = gc_marker + GC_OFFSET_UNSURE;
970 unit_add_to_gc_queue(u);
971 return;
972
973 bad:
974 /* We definitely know that this one is not useful anymore, so
975 * let's mark it for deletion */
976 u->gc_marker = gc_marker + GC_OFFSET_BAD;
977 unit_add_to_cleanup_queue(u);
978 return;
979
980 good:
981 unit_gc_mark_good(u, gc_marker);
982 }
983
984 static unsigned manager_dispatch_gc_queue(Manager *m) {
985 Unit *u;
986 unsigned n = 0;
987 unsigned gc_marker;
988
989 assert(m);
990
991 /* log_debug("Running GC..."); */
992
993 m->gc_marker += _GC_OFFSET_MAX;
994 if (m->gc_marker + _GC_OFFSET_MAX <= _GC_OFFSET_MAX)
995 m->gc_marker = 1;
996
997 gc_marker = m->gc_marker;
998
999 while ((u = m->gc_queue)) {
1000 assert(u->in_gc_queue);
1001
1002 unit_gc_sweep(u, gc_marker);
1003
1004 LIST_REMOVE(gc_queue, m->gc_queue, u);
1005 u->in_gc_queue = false;
1006
1007 n++;
1008
1009 if (u->gc_marker == gc_marker + GC_OFFSET_BAD ||
1010 u->gc_marker == gc_marker + GC_OFFSET_UNSURE) {
1011 if (u->id)
1012 log_unit_debug(u, "Collecting.");
1013 u->gc_marker = gc_marker + GC_OFFSET_BAD;
1014 unit_add_to_cleanup_queue(u);
1015 }
1016 }
1017
1018 m->n_in_gc_queue = 0;
1019
1020 return n;
1021 }
1022
1023 static void manager_clear_jobs_and_units(Manager *m) {
1024 Unit *u;
1025
1026 assert(m);
1027
1028 while ((u = hashmap_first(m->units)))
1029 unit_free(u);
1030
1031 manager_dispatch_cleanup_queue(m);
1032
1033 assert(!m->load_queue);
1034 assert(!m->run_queue);
1035 assert(!m->dbus_unit_queue);
1036 assert(!m->dbus_job_queue);
1037 assert(!m->cleanup_queue);
1038 assert(!m->gc_queue);
1039
1040 assert(hashmap_isempty(m->jobs));
1041 assert(hashmap_isempty(m->units));
1042
1043 m->n_on_console = 0;
1044 m->n_running_jobs = 0;
1045 }
1046
1047 Manager* manager_free(Manager *m) {
1048 UnitType c;
1049 int i;
1050
1051 if (!m)
1052 return NULL;
1053
1054 manager_clear_jobs_and_units(m);
1055
1056 for (c = 0; c < _UNIT_TYPE_MAX; c++)
1057 if (unit_vtable[c]->shutdown)
1058 unit_vtable[c]->shutdown(m);
1059
1060 /* If we reexecute ourselves, we keep the root cgroup
1061 * around */
1062 manager_shutdown_cgroup(m, m->exit_code != MANAGER_REEXECUTE);
1063
1064 lookup_paths_flush_generator(&m->lookup_paths);
1065
1066 bus_done(m);
1067
1068 dynamic_user_vacuum(m, false);
1069 hashmap_free(m->dynamic_users);
1070
1071 hashmap_free(m->units);
1072 hashmap_free(m->units_by_invocation_id);
1073 hashmap_free(m->jobs);
1074 hashmap_free(m->watch_pids1);
1075 hashmap_free(m->watch_pids2);
1076 hashmap_free(m->watch_bus);
1077
1078 set_free(m->startup_units);
1079 set_free(m->failed_units);
1080
1081 sd_event_source_unref(m->signal_event_source);
1082 sd_event_source_unref(m->notify_event_source);
1083 sd_event_source_unref(m->cgroups_agent_event_source);
1084 sd_event_source_unref(m->time_change_event_source);
1085 sd_event_source_unref(m->jobs_in_progress_event_source);
1086 sd_event_source_unref(m->run_queue_event_source);
1087 sd_event_source_unref(m->user_lookup_event_source);
1088
1089 safe_close(m->signal_fd);
1090 safe_close(m->notify_fd);
1091 safe_close(m->cgroups_agent_fd);
1092 safe_close(m->time_change_fd);
1093 safe_close_pair(m->user_lookup_fds);
1094
1095 manager_close_ask_password(m);
1096
1097 manager_close_idle_pipe(m);
1098
1099 udev_unref(m->udev);
1100 sd_event_unref(m->event);
1101
1102 free(m->notify_socket);
1103
1104 lookup_paths_free(&m->lookup_paths);
1105 strv_free(m->environment);
1106
1107 hashmap_free(m->cgroup_unit);
1108 set_free_free(m->unit_path_cache);
1109
1110 free(m->switch_root);
1111 free(m->switch_root_init);
1112
1113 for (i = 0; i < _RLIMIT_MAX; i++)
1114 m->rlimit[i] = mfree(m->rlimit[i]);
1115
1116 assert(hashmap_isempty(m->units_requiring_mounts_for));
1117 hashmap_free(m->units_requiring_mounts_for);
1118
1119 hashmap_free(m->uid_refs);
1120 hashmap_free(m->gid_refs);
1121
1122 return mfree(m);
1123 }
1124
1125 void manager_enumerate(Manager *m) {
1126 UnitType c;
1127
1128 assert(m);
1129
1130 /* Let's ask every type to load all units from disk/kernel
1131 * that it might know */
1132 for (c = 0; c < _UNIT_TYPE_MAX; c++) {
1133 if (!unit_type_supported(c)) {
1134 log_debug("Unit type .%s is not supported on this system.", unit_type_to_string(c));
1135 continue;
1136 }
1137
1138 if (!unit_vtable[c]->enumerate)
1139 continue;
1140
1141 unit_vtable[c]->enumerate(m);
1142 }
1143
1144 manager_dispatch_load_queue(m);
1145 }
1146
1147 static void manager_coldplug(Manager *m) {
1148 Iterator i;
1149 Unit *u;
1150 char *k;
1151 int r;
1152
1153 assert(m);
1154
1155 /* Then, let's set up their initial state. */
1156 HASHMAP_FOREACH_KEY(u, k, m->units, i) {
1157
1158 /* ignore aliases */
1159 if (u->id != k)
1160 continue;
1161
1162 r = unit_coldplug(u);
1163 if (r < 0)
1164 log_warning_errno(r, "We couldn't coldplug %s, proceeding anyway: %m", u->id);
1165 }
1166 }
1167
1168 static void manager_build_unit_path_cache(Manager *m) {
1169 char **i;
1170 int r;
1171
1172 assert(m);
1173
1174 set_free_free(m->unit_path_cache);
1175
1176 m->unit_path_cache = set_new(&string_hash_ops);
1177 if (!m->unit_path_cache) {
1178 r = -ENOMEM;
1179 goto fail;
1180 }
1181
1182 /* This simply builds a list of files we know exist, so that
1183 * we don't always have to go to disk */
1184
1185 STRV_FOREACH(i, m->lookup_paths.search_path) {
1186 _cleanup_closedir_ DIR *d = NULL;
1187 struct dirent *de;
1188
1189 d = opendir(*i);
1190 if (!d) {
1191 if (errno != ENOENT)
1192 log_warning_errno(errno, "Failed to open directory %s, ignoring: %m", *i);
1193 continue;
1194 }
1195
1196 FOREACH_DIRENT(de, d, r = -errno; goto fail) {
1197 char *p;
1198
1199 p = strjoin(streq(*i, "/") ? "" : *i, "/", de->d_name, NULL);
1200 if (!p) {
1201 r = -ENOMEM;
1202 goto fail;
1203 }
1204
1205 r = set_consume(m->unit_path_cache, p);
1206 if (r < 0)
1207 goto fail;
1208 }
1209 }
1210
1211 return;
1212
1213 fail:
1214 log_warning_errno(r, "Failed to build unit path cache, proceeding without: %m");
1215 m->unit_path_cache = set_free_free(m->unit_path_cache);
1216 }
1217
1218 static void manager_distribute_fds(Manager *m, FDSet *fds) {
1219 Iterator i;
1220 Unit *u;
1221
1222 assert(m);
1223
1224 HASHMAP_FOREACH(u, m->units, i) {
1225
1226 if (fdset_size(fds) <= 0)
1227 break;
1228
1229 if (!UNIT_VTABLE(u)->distribute_fds)
1230 continue;
1231
1232 UNIT_VTABLE(u)->distribute_fds(u, fds);
1233 }
1234 }
1235
1236 int manager_startup(Manager *m, FILE *serialization, FDSet *fds) {
1237 int r, q;
1238
1239 assert(m);
1240
1241 r = lookup_paths_init(&m->lookup_paths, m->unit_file_scope, 0, NULL);
1242 if (r < 0)
1243 return r;
1244
1245 /* Make sure the transient directory always exists, so that it remains in the search path */
1246 if (!m->test_run) {
1247 r = mkdir_p_label(m->lookup_paths.transient, 0755);
1248 if (r < 0)
1249 return r;
1250 }
1251
1252 dual_timestamp_get(&m->generators_start_timestamp);
1253 r = manager_run_generators(m);
1254 dual_timestamp_get(&m->generators_finish_timestamp);
1255 if (r < 0)
1256 return r;
1257
1258 lookup_paths_reduce(&m->lookup_paths);
1259 manager_build_unit_path_cache(m);
1260
1261 /* If we will deserialize make sure that during enumeration
1262 * this is already known, so we increase the counter here
1263 * already */
1264 if (serialization)
1265 m->n_reloading++;
1266
1267 /* First, enumerate what we can from all config files */
1268 dual_timestamp_get(&m->units_load_start_timestamp);
1269 manager_enumerate(m);
1270 dual_timestamp_get(&m->units_load_finish_timestamp);
1271
1272 /* Second, deserialize if there is something to deserialize */
1273 if (serialization)
1274 r = manager_deserialize(m, serialization, fds);
1275
1276 /* Any fds left? Find some unit which wants them. This is
1277 * useful to allow container managers to pass some file
1278 * descriptors to us pre-initialized. This enables
1279 * socket-based activation of entire containers. */
1280 manager_distribute_fds(m, fds);
1281
1282 /* We might have deserialized the notify fd, but if we didn't
1283 * then let's create the bus now */
1284 q = manager_setup_notify(m);
1285 if (q < 0 && r == 0)
1286 r = q;
1287
1288 q = manager_setup_cgroups_agent(m);
1289 if (q < 0 && r == 0)
1290 r = q;
1291
1292 q = manager_setup_user_lookup_fd(m);
1293 if (q < 0 && r == 0)
1294 r = q;
1295
1296 /* Let's connect to the bus now. */
1297 (void) manager_connect_bus(m, !!serialization);
1298
1299 (void) bus_track_coldplug(m, &m->subscribed, false, m->deserialized_subscribed);
1300 m->deserialized_subscribed = strv_free(m->deserialized_subscribed);
1301
1302 /* Third, fire things up! */
1303 manager_coldplug(m);
1304
1305 /* Release any dynamic users no longer referenced */
1306 dynamic_user_vacuum(m, true);
1307
1308 /* Release any references to UIDs/GIDs no longer referenced, and destroy any IPC owned by them */
1309 manager_vacuum_uid_refs(m);
1310 manager_vacuum_gid_refs(m);
1311
1312 if (serialization) {
1313 assert(m->n_reloading > 0);
1314 m->n_reloading--;
1315
1316 /* Let's wait for the UnitNew/JobNew messages being
1317 * sent, before we notify that the reload is
1318 * finished */
1319 m->send_reloading_done = true;
1320 }
1321
1322 return r;
1323 }
1324
1325 int manager_add_job(Manager *m, JobType type, Unit *unit, JobMode mode, sd_bus_error *e, Job **_ret) {
1326 int r;
1327 Transaction *tr;
1328
1329 assert(m);
1330 assert(type < _JOB_TYPE_MAX);
1331 assert(unit);
1332 assert(mode < _JOB_MODE_MAX);
1333
1334 if (mode == JOB_ISOLATE && type != JOB_START)
1335 return sd_bus_error_setf(e, SD_BUS_ERROR_INVALID_ARGS, "Isolate is only valid for start.");
1336
1337 if (mode == JOB_ISOLATE && !unit->allow_isolate)
1338 return sd_bus_error_setf(e, BUS_ERROR_NO_ISOLATION, "Operation refused, unit may not be isolated.");
1339
1340 log_unit_debug(unit, "Trying to enqueue job %s/%s/%s", unit->id, job_type_to_string(type), job_mode_to_string(mode));
1341
1342 type = job_type_collapse(type, unit);
1343
1344 tr = transaction_new(mode == JOB_REPLACE_IRREVERSIBLY);
1345 if (!tr)
1346 return -ENOMEM;
1347
1348 r = transaction_add_job_and_dependencies(tr, type, unit, NULL, true, false,
1349 mode == JOB_IGNORE_DEPENDENCIES || mode == JOB_IGNORE_REQUIREMENTS,
1350 mode == JOB_IGNORE_DEPENDENCIES, e);
1351 if (r < 0)
1352 goto tr_abort;
1353
1354 if (mode == JOB_ISOLATE) {
1355 r = transaction_add_isolate_jobs(tr, m);
1356 if (r < 0)
1357 goto tr_abort;
1358 }
1359
1360 r = transaction_activate(tr, m, mode, e);
1361 if (r < 0)
1362 goto tr_abort;
1363
1364 log_unit_debug(unit,
1365 "Enqueued job %s/%s as %u", unit->id,
1366 job_type_to_string(type), (unsigned) tr->anchor_job->id);
1367
1368 if (_ret)
1369 *_ret = tr->anchor_job;
1370
1371 transaction_free(tr);
1372 return 0;
1373
1374 tr_abort:
1375 transaction_abort(tr);
1376 transaction_free(tr);
1377 return r;
1378 }
1379
1380 int manager_add_job_by_name(Manager *m, JobType type, const char *name, JobMode mode, sd_bus_error *e, Job **ret) {
1381 Unit *unit;
1382 int r;
1383
1384 assert(m);
1385 assert(type < _JOB_TYPE_MAX);
1386 assert(name);
1387 assert(mode < _JOB_MODE_MAX);
1388
1389 r = manager_load_unit(m, name, NULL, NULL, &unit);
1390 if (r < 0)
1391 return r;
1392
1393 return manager_add_job(m, type, unit, mode, e, ret);
1394 }
1395
1396 int manager_add_job_by_name_and_warn(Manager *m, JobType type, const char *name, JobMode mode, Job **ret) {
1397 _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL;
1398 int r;
1399
1400 assert(m);
1401 assert(type < _JOB_TYPE_MAX);
1402 assert(name);
1403 assert(mode < _JOB_MODE_MAX);
1404
1405 r = manager_add_job_by_name(m, type, name, mode, &error, ret);
1406 if (r < 0)
1407 return log_warning_errno(r, "Failed to enqueue %s job for %s: %s", job_mode_to_string(mode), name, bus_error_message(&error, r));
1408
1409 return r;
1410 }
1411
1412 Job *manager_get_job(Manager *m, uint32_t id) {
1413 assert(m);
1414
1415 return hashmap_get(m->jobs, UINT32_TO_PTR(id));
1416 }
1417
1418 Unit *manager_get_unit(Manager *m, const char *name) {
1419 assert(m);
1420 assert(name);
1421
1422 return hashmap_get(m->units, name);
1423 }
1424
1425 unsigned manager_dispatch_load_queue(Manager *m) {
1426 Unit *u;
1427 unsigned n = 0;
1428
1429 assert(m);
1430
1431 /* Make sure we are not run recursively */
1432 if (m->dispatching_load_queue)
1433 return 0;
1434
1435 m->dispatching_load_queue = true;
1436
1437 /* Dispatches the load queue. Takes a unit from the queue and
1438 * tries to load its data until the queue is empty */
1439
1440 while ((u = m->load_queue)) {
1441 assert(u->in_load_queue);
1442
1443 unit_load(u);
1444 n++;
1445 }
1446
1447 m->dispatching_load_queue = false;
1448 return n;
1449 }
1450
1451 int manager_load_unit_prepare(
1452 Manager *m,
1453 const char *name,
1454 const char *path,
1455 sd_bus_error *e,
1456 Unit **_ret) {
1457
1458 Unit *ret;
1459 UnitType t;
1460 int r;
1461
1462 assert(m);
1463 assert(name || path);
1464
1465 /* This will prepare the unit for loading, but not actually
1466 * load anything from disk. */
1467
1468 if (path && !is_path(path))
1469 return sd_bus_error_setf(e, SD_BUS_ERROR_INVALID_ARGS, "Path %s is not absolute.", path);
1470
1471 if (!name)
1472 name = basename(path);
1473
1474 t = unit_name_to_type(name);
1475
1476 if (t == _UNIT_TYPE_INVALID || !unit_name_is_valid(name, UNIT_NAME_PLAIN|UNIT_NAME_INSTANCE)) {
1477 if (unit_name_is_valid(name, UNIT_NAME_TEMPLATE))
1478 return sd_bus_error_setf(e, SD_BUS_ERROR_INVALID_ARGS, "Unit name %s is missing the instance name.", name);
1479
1480 return sd_bus_error_setf(e, SD_BUS_ERROR_INVALID_ARGS, "Unit name %s is not valid.", name);
1481 }
1482
1483 ret = manager_get_unit(m, name);
1484 if (ret) {
1485 *_ret = ret;
1486 return 1;
1487 }
1488
1489 ret = unit_new(m, unit_vtable[t]->object_size);
1490 if (!ret)
1491 return -ENOMEM;
1492
1493 if (path) {
1494 ret->fragment_path = strdup(path);
1495 if (!ret->fragment_path) {
1496 unit_free(ret);
1497 return -ENOMEM;
1498 }
1499 }
1500
1501 r = unit_add_name(ret, name);
1502 if (r < 0) {
1503 unit_free(ret);
1504 return r;
1505 }
1506
1507 unit_add_to_load_queue(ret);
1508 unit_add_to_dbus_queue(ret);
1509 unit_add_to_gc_queue(ret);
1510
1511 if (_ret)
1512 *_ret = ret;
1513
1514 return 0;
1515 }
1516
1517 int manager_load_unit(
1518 Manager *m,
1519 const char *name,
1520 const char *path,
1521 sd_bus_error *e,
1522 Unit **_ret) {
1523
1524 int r;
1525
1526 assert(m);
1527
1528 /* This will load the service information files, but not actually
1529 * start any services or anything. */
1530
1531 r = manager_load_unit_prepare(m, name, path, e, _ret);
1532 if (r != 0)
1533 return r;
1534
1535 manager_dispatch_load_queue(m);
1536
1537 if (_ret)
1538 *_ret = unit_follow_merge(*_ret);
1539
1540 return 0;
1541 }
1542
1543 void manager_dump_jobs(Manager *s, FILE *f, const char *prefix) {
1544 Iterator i;
1545 Job *j;
1546
1547 assert(s);
1548 assert(f);
1549
1550 HASHMAP_FOREACH(j, s->jobs, i)
1551 job_dump(j, f, prefix);
1552 }
1553
1554 void manager_dump_units(Manager *s, FILE *f, const char *prefix) {
1555 Iterator i;
1556 Unit *u;
1557 const char *t;
1558
1559 assert(s);
1560 assert(f);
1561
1562 HASHMAP_FOREACH_KEY(u, t, s->units, i)
1563 if (u->id == t)
1564 unit_dump(u, f, prefix);
1565 }
1566
1567 void manager_clear_jobs(Manager *m) {
1568 Job *j;
1569
1570 assert(m);
1571
1572 while ((j = hashmap_first(m->jobs)))
1573 /* No need to recurse. We're cancelling all jobs. */
1574 job_finish_and_invalidate(j, JOB_CANCELED, false, false);
1575 }
1576
1577 static int manager_dispatch_run_queue(sd_event_source *source, void *userdata) {
1578 Manager *m = userdata;
1579 Job *j;
1580
1581 assert(source);
1582 assert(m);
1583
1584 while ((j = m->run_queue)) {
1585 assert(j->installed);
1586 assert(j->in_run_queue);
1587
1588 job_run_and_invalidate(j);
1589 }
1590
1591 if (m->n_running_jobs > 0)
1592 manager_watch_jobs_in_progress(m);
1593
1594 if (m->n_on_console > 0)
1595 manager_watch_idle_pipe(m);
1596
1597 return 1;
1598 }
1599
1600 static unsigned manager_dispatch_dbus_queue(Manager *m) {
1601 Job *j;
1602 Unit *u;
1603 unsigned n = 0;
1604
1605 assert(m);
1606
1607 if (m->dispatching_dbus_queue)
1608 return 0;
1609
1610 m->dispatching_dbus_queue = true;
1611
1612 while ((u = m->dbus_unit_queue)) {
1613 assert(u->in_dbus_queue);
1614
1615 bus_unit_send_change_signal(u);
1616 n++;
1617 }
1618
1619 while ((j = m->dbus_job_queue)) {
1620 assert(j->in_dbus_queue);
1621
1622 bus_job_send_change_signal(j);
1623 n++;
1624 }
1625
1626 m->dispatching_dbus_queue = false;
1627
1628 if (m->send_reloading_done) {
1629 m->send_reloading_done = false;
1630
1631 bus_manager_send_reloading(m, false);
1632 }
1633
1634 if (m->queued_message)
1635 bus_send_queued_message(m);
1636
1637 return n;
1638 }
1639
1640 static int manager_dispatch_cgroups_agent_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
1641 Manager *m = userdata;
1642 char buf[PATH_MAX+1];
1643 ssize_t n;
1644
1645 n = recv(fd, buf, sizeof(buf), 0);
1646 if (n < 0)
1647 return log_error_errno(errno, "Failed to read cgroups agent message: %m");
1648 if (n == 0) {
1649 log_error("Got zero-length cgroups agent message, ignoring.");
1650 return 0;
1651 }
1652 if ((size_t) n >= sizeof(buf)) {
1653 log_error("Got overly long cgroups agent message, ignoring.");
1654 return 0;
1655 }
1656
1657 if (memchr(buf, 0, n)) {
1658 log_error("Got cgroups agent message with embedded NUL byte, ignoring.");
1659 return 0;
1660 }
1661 buf[n] = 0;
1662
1663 manager_notify_cgroup_empty(m, buf);
1664 bus_forward_agent_released(m, buf);
1665
1666 return 0;
1667 }
1668
1669 static void manager_invoke_notify_message(Manager *m, Unit *u, pid_t pid, const char *buf, FDSet *fds) {
1670 _cleanup_strv_free_ char **tags = NULL;
1671
1672 assert(m);
1673 assert(u);
1674 assert(buf);
1675
1676 tags = strv_split(buf, "\n\r");
1677 if (!tags) {
1678 log_oom();
1679 return;
1680 }
1681
1682 if (UNIT_VTABLE(u)->notify_message)
1683 UNIT_VTABLE(u)->notify_message(u, pid, tags, fds);
1684 else if (_unlikely_(log_get_max_level() >= LOG_DEBUG)) {
1685 _cleanup_free_ char *x = NULL, *y = NULL;
1686
1687 x = cescape(buf);
1688 if (x)
1689 y = ellipsize(x, 20, 90);
1690 log_unit_debug(u, "Got notification message \"%s\", ignoring.", strnull(y));
1691 }
1692 }
1693
1694 static int manager_dispatch_notify_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
1695
1696 _cleanup_fdset_free_ FDSet *fds = NULL;
1697 Manager *m = userdata;
1698 char buf[NOTIFY_BUFFER_MAX+1];
1699 struct iovec iovec = {
1700 .iov_base = buf,
1701 .iov_len = sizeof(buf)-1,
1702 };
1703 union {
1704 struct cmsghdr cmsghdr;
1705 uint8_t buf[CMSG_SPACE(sizeof(struct ucred)) +
1706 CMSG_SPACE(sizeof(int) * NOTIFY_FD_MAX)];
1707 } control = {};
1708 struct msghdr msghdr = {
1709 .msg_iov = &iovec,
1710 .msg_iovlen = 1,
1711 .msg_control = &control,
1712 .msg_controllen = sizeof(control),
1713 };
1714
1715 struct cmsghdr *cmsg;
1716 struct ucred *ucred = NULL;
1717 Unit *u1, *u2, *u3;
1718 int r, *fd_array = NULL;
1719 unsigned n_fds = 0;
1720 ssize_t n;
1721
1722 assert(m);
1723 assert(m->notify_fd == fd);
1724
1725 if (revents != EPOLLIN) {
1726 log_warning("Got unexpected poll event for notify fd.");
1727 return 0;
1728 }
1729
1730 n = recvmsg(m->notify_fd, &msghdr, MSG_DONTWAIT|MSG_CMSG_CLOEXEC|MSG_TRUNC);
1731 if (n < 0) {
1732 if (IN_SET(errno, EAGAIN, EINTR))
1733 return 0; /* Spurious wakeup, try again */
1734
1735 /* If this is any other, real error, then let's stop processing this socket. This of course means we
1736 * won't take notification messages anymore, but that's still better than busy looping around this:
1737 * being woken up over and over again but being unable to actually read the message off the socket. */
1738 return log_error_errno(errno, "Failed to receive notification message: %m");
1739 }
1740
1741 CMSG_FOREACH(cmsg, &msghdr) {
1742 if (cmsg->cmsg_level == SOL_SOCKET && cmsg->cmsg_type == SCM_RIGHTS) {
1743
1744 fd_array = (int*) CMSG_DATA(cmsg);
1745 n_fds = (cmsg->cmsg_len - CMSG_LEN(0)) / sizeof(int);
1746
1747 } else if (cmsg->cmsg_level == SOL_SOCKET &&
1748 cmsg->cmsg_type == SCM_CREDENTIALS &&
1749 cmsg->cmsg_len == CMSG_LEN(sizeof(struct ucred))) {
1750
1751 ucred = (struct ucred*) CMSG_DATA(cmsg);
1752 }
1753 }
1754
1755 if (n_fds > 0) {
1756 assert(fd_array);
1757
1758 r = fdset_new_array(&fds, fd_array, n_fds);
1759 if (r < 0) {
1760 close_many(fd_array, n_fds);
1761 log_oom();
1762 return 0;
1763 }
1764 }
1765
1766 if (!ucred || ucred->pid <= 0) {
1767 log_warning("Received notify message without valid credentials. Ignoring.");
1768 return 0;
1769 }
1770
1771 if ((size_t) n >= sizeof(buf) || (msghdr.msg_flags & MSG_TRUNC)) {
1772 log_warning("Received notify message exceeded maximum size. Ignoring.");
1773 return 0;
1774 }
1775
1776 /* As extra safety check, let's make sure the string we get doesn't contain embedded NUL bytes. We permit one
1777 * trailing NUL byte in the message, but don't expect it. */
1778 if (n > 1 && memchr(buf, 0, n-1)) {
1779 log_warning("Received notify message with embedded NUL bytes. Ignoring.");
1780 return 0;
1781 }
1782
1783 /* Make sure it's NUL-terminated. */
1784 buf[n] = 0;
1785
1786 /* Notify every unit that might be interested, but try
1787 * to avoid notifying the same one multiple times. */
1788 u1 = manager_get_unit_by_pid_cgroup(m, ucred->pid);
1789 if (u1)
1790 manager_invoke_notify_message(m, u1, ucred->pid, buf, fds);
1791
1792 u2 = hashmap_get(m->watch_pids1, PID_TO_PTR(ucred->pid));
1793 if (u2 && u2 != u1)
1794 manager_invoke_notify_message(m, u2, ucred->pid, buf, fds);
1795
1796 u3 = hashmap_get(m->watch_pids2, PID_TO_PTR(ucred->pid));
1797 if (u3 && u3 != u2 && u3 != u1)
1798 manager_invoke_notify_message(m, u3, ucred->pid, buf, fds);
1799
1800 if (!u1 && !u2 && !u3)
1801 log_warning("Cannot find unit for notify message of PID "PID_FMT".", ucred->pid);
1802
1803 if (fdset_size(fds) > 0)
1804 log_warning("Got extra auxiliary fds with notification message, closing them.");
1805
1806 return 0;
1807 }
1808
1809 static void invoke_sigchld_event(Manager *m, Unit *u, const siginfo_t *si) {
1810 uint64_t iteration;
1811
1812 assert(m);
1813 assert(u);
1814 assert(si);
1815
1816 sd_event_get_iteration(m->event, &iteration);
1817
1818 log_unit_debug(u, "Child "PID_FMT" belongs to %s", si->si_pid, u->id);
1819
1820 unit_unwatch_pid(u, si->si_pid);
1821
1822 if (UNIT_VTABLE(u)->sigchld_event) {
1823 if (set_size(u->pids) <= 1 ||
1824 iteration != u->sigchldgen ||
1825 unit_main_pid(u) == si->si_pid ||
1826 unit_control_pid(u) == si->si_pid) {
1827 UNIT_VTABLE(u)->sigchld_event(u, si->si_pid, si->si_code, si->si_status);
1828 u->sigchldgen = iteration;
1829 } else
1830 log_debug("%s already issued a sigchld this iteration %" PRIu64 ", skipping. Pids still being watched %d", u->id, iteration, set_size(u->pids));
1831 }
1832 }
1833
1834 static int manager_dispatch_sigchld(Manager *m) {
1835 assert(m);
1836
1837 for (;;) {
1838 siginfo_t si = {};
1839
1840 /* First we call waitd() for a PID and do not reap the
1841 * zombie. That way we can still access /proc/$PID for
1842 * it while it is a zombie. */
1843 if (waitid(P_ALL, 0, &si, WEXITED|WNOHANG|WNOWAIT) < 0) {
1844
1845 if (errno == ECHILD)
1846 break;
1847
1848 if (errno == EINTR)
1849 continue;
1850
1851 return -errno;
1852 }
1853
1854 if (si.si_pid <= 0)
1855 break;
1856
1857 if (si.si_code == CLD_EXITED || si.si_code == CLD_KILLED || si.si_code == CLD_DUMPED) {
1858 _cleanup_free_ char *name = NULL;
1859 Unit *u1, *u2, *u3;
1860
1861 get_process_comm(si.si_pid, &name);
1862
1863 log_debug("Child "PID_FMT" (%s) died (code=%s, status=%i/%s)",
1864 si.si_pid, strna(name),
1865 sigchld_code_to_string(si.si_code),
1866 si.si_status,
1867 strna(si.si_code == CLD_EXITED
1868 ? exit_status_to_string(si.si_status, EXIT_STATUS_FULL)
1869 : signal_to_string(si.si_status)));
1870
1871 /* And now figure out the unit this belongs
1872 * to, it might be multiple... */
1873 u1 = manager_get_unit_by_pid_cgroup(m, si.si_pid);
1874 if (u1)
1875 invoke_sigchld_event(m, u1, &si);
1876 u2 = hashmap_get(m->watch_pids1, PID_TO_PTR(si.si_pid));
1877 if (u2 && u2 != u1)
1878 invoke_sigchld_event(m, u2, &si);
1879 u3 = hashmap_get(m->watch_pids2, PID_TO_PTR(si.si_pid));
1880 if (u3 && u3 != u2 && u3 != u1)
1881 invoke_sigchld_event(m, u3, &si);
1882 }
1883
1884 /* And now, we actually reap the zombie. */
1885 if (waitid(P_PID, si.si_pid, &si, WEXITED) < 0) {
1886 if (errno == EINTR)
1887 continue;
1888
1889 return -errno;
1890 }
1891 }
1892
1893 return 0;
1894 }
1895
1896 static int manager_start_target(Manager *m, const char *name, JobMode mode) {
1897 _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL;
1898 int r;
1899
1900 log_debug("Activating special unit %s", name);
1901
1902 r = manager_add_job_by_name(m, JOB_START, name, mode, &error, NULL);
1903 if (r < 0)
1904 log_error("Failed to enqueue %s job: %s", name, bus_error_message(&error, r));
1905
1906 return r;
1907 }
1908
1909 static void manager_handle_ctrl_alt_del(Manager *m) {
1910 /* If the user presses C-A-D more than
1911 * 7 times within 2s, we reboot/shutdown immediately,
1912 * unless it was disabled in system.conf */
1913
1914 if (ratelimit_test(&m->ctrl_alt_del_ratelimit) || m->cad_burst_action == CAD_BURST_ACTION_IGNORE)
1915 manager_start_target(m, SPECIAL_CTRL_ALT_DEL_TARGET, JOB_REPLACE_IRREVERSIBLY);
1916 else {
1917 switch (m->cad_burst_action) {
1918
1919 case CAD_BURST_ACTION_REBOOT:
1920 m->exit_code = MANAGER_REBOOT;
1921 break;
1922
1923 case CAD_BURST_ACTION_POWEROFF:
1924 m->exit_code = MANAGER_POWEROFF;
1925 break;
1926
1927 default:
1928 assert_not_reached("Unknown action.");
1929 }
1930
1931 log_notice("Ctrl-Alt-Del was pressed more than 7 times within 2s, performing immediate %s.",
1932 cad_burst_action_to_string(m->cad_burst_action));
1933 status_printf(NULL, true, false, "Ctrl-Alt-Del was pressed more than 7 times within 2s, performing immediate %s.",
1934 cad_burst_action_to_string(m->cad_burst_action));
1935 }
1936 }
1937
1938 static int manager_dispatch_signal_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
1939 Manager *m = userdata;
1940 ssize_t n;
1941 struct signalfd_siginfo sfsi;
1942 bool sigchld = false;
1943 int r;
1944
1945 assert(m);
1946 assert(m->signal_fd == fd);
1947
1948 if (revents != EPOLLIN) {
1949 log_warning("Got unexpected events from signal file descriptor.");
1950 return 0;
1951 }
1952
1953 for (;;) {
1954 n = read(m->signal_fd, &sfsi, sizeof(sfsi));
1955 if (n != sizeof(sfsi)) {
1956 if (n >= 0) {
1957 log_warning("Truncated read from signal fd (%zu bytes)!", n);
1958 return 0;
1959 }
1960
1961 if (IN_SET(errno, EINTR, EAGAIN))
1962 break;
1963
1964 /* We return an error here, which will kill this handler,
1965 * to avoid a busy loop on read error. */
1966 return log_error_errno(errno, "Reading from signal fd failed: %m");
1967 }
1968
1969 log_received_signal(sfsi.ssi_signo == SIGCHLD ||
1970 (sfsi.ssi_signo == SIGTERM && MANAGER_IS_USER(m))
1971 ? LOG_DEBUG : LOG_INFO,
1972 &sfsi);
1973
1974 switch (sfsi.ssi_signo) {
1975
1976 case SIGCHLD:
1977 sigchld = true;
1978 break;
1979
1980 case SIGTERM:
1981 if (MANAGER_IS_SYSTEM(m)) {
1982 /* This is for compatibility with the
1983 * original sysvinit */
1984 m->exit_code = MANAGER_REEXECUTE;
1985 break;
1986 }
1987
1988 /* Fall through */
1989
1990 case SIGINT:
1991 if (MANAGER_IS_SYSTEM(m)) {
1992 manager_handle_ctrl_alt_del(m);
1993 break;
1994 }
1995
1996 /* Run the exit target if there is one, if not, just exit. */
1997 if (manager_start_target(m, SPECIAL_EXIT_TARGET, JOB_REPLACE) < 0) {
1998 m->exit_code = MANAGER_EXIT;
1999 return 0;
2000 }
2001
2002 break;
2003
2004 case SIGWINCH:
2005 if (MANAGER_IS_SYSTEM(m))
2006 manager_start_target(m, SPECIAL_KBREQUEST_TARGET, JOB_REPLACE);
2007
2008 /* This is a nop on non-init */
2009 break;
2010
2011 case SIGPWR:
2012 if (MANAGER_IS_SYSTEM(m))
2013 manager_start_target(m, SPECIAL_SIGPWR_TARGET, JOB_REPLACE);
2014
2015 /* This is a nop on non-init */
2016 break;
2017
2018 case SIGUSR1: {
2019 Unit *u;
2020
2021 u = manager_get_unit(m, SPECIAL_DBUS_SERVICE);
2022
2023 if (!u || UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u))) {
2024 log_info("Trying to reconnect to bus...");
2025 bus_init(m, true);
2026 }
2027
2028 if (!u || !UNIT_IS_ACTIVE_OR_ACTIVATING(unit_active_state(u))) {
2029 log_info("Loading D-Bus service...");
2030 manager_start_target(m, SPECIAL_DBUS_SERVICE, JOB_REPLACE);
2031 }
2032
2033 break;
2034 }
2035
2036 case SIGUSR2: {
2037 _cleanup_free_ char *dump = NULL;
2038 _cleanup_fclose_ FILE *f = NULL;
2039 size_t size;
2040
2041 f = open_memstream(&dump, &size);
2042 if (!f) {
2043 log_warning_errno(errno, "Failed to allocate memory stream: %m");
2044 break;
2045 }
2046
2047 manager_dump_units(m, f, "\t");
2048 manager_dump_jobs(m, f, "\t");
2049
2050 r = fflush_and_check(f);
2051 if (r < 0) {
2052 log_warning_errno(r, "Failed to write status stream: %m");
2053 break;
2054 }
2055
2056 log_dump(LOG_INFO, dump);
2057 break;
2058 }
2059
2060 case SIGHUP:
2061 m->exit_code = MANAGER_RELOAD;
2062 break;
2063
2064 default: {
2065
2066 /* Starting SIGRTMIN+0 */
2067 static const char * const target_table[] = {
2068 [0] = SPECIAL_DEFAULT_TARGET,
2069 [1] = SPECIAL_RESCUE_TARGET,
2070 [2] = SPECIAL_EMERGENCY_TARGET,
2071 [3] = SPECIAL_HALT_TARGET,
2072 [4] = SPECIAL_POWEROFF_TARGET,
2073 [5] = SPECIAL_REBOOT_TARGET,
2074 [6] = SPECIAL_KEXEC_TARGET
2075 };
2076
2077 /* Starting SIGRTMIN+13, so that target halt and system halt are 10 apart */
2078 static const ManagerExitCode code_table[] = {
2079 [0] = MANAGER_HALT,
2080 [1] = MANAGER_POWEROFF,
2081 [2] = MANAGER_REBOOT,
2082 [3] = MANAGER_KEXEC
2083 };
2084
2085 if ((int) sfsi.ssi_signo >= SIGRTMIN+0 &&
2086 (int) sfsi.ssi_signo < SIGRTMIN+(int) ELEMENTSOF(target_table)) {
2087 int idx = (int) sfsi.ssi_signo - SIGRTMIN;
2088 manager_start_target(m, target_table[idx],
2089 (idx == 1 || idx == 2) ? JOB_ISOLATE : JOB_REPLACE);
2090 break;
2091 }
2092
2093 if ((int) sfsi.ssi_signo >= SIGRTMIN+13 &&
2094 (int) sfsi.ssi_signo < SIGRTMIN+13+(int) ELEMENTSOF(code_table)) {
2095 m->exit_code = code_table[sfsi.ssi_signo - SIGRTMIN - 13];
2096 break;
2097 }
2098
2099 switch (sfsi.ssi_signo - SIGRTMIN) {
2100
2101 case 20:
2102 manager_set_show_status(m, SHOW_STATUS_YES);
2103 break;
2104
2105 case 21:
2106 manager_set_show_status(m, SHOW_STATUS_NO);
2107 break;
2108
2109 case 22:
2110 log_set_max_level(LOG_DEBUG);
2111 log_info("Setting log level to debug.");
2112 break;
2113
2114 case 23:
2115 log_set_max_level(LOG_INFO);
2116 log_info("Setting log level to info.");
2117 break;
2118
2119 case 24:
2120 if (MANAGER_IS_USER(m)) {
2121 m->exit_code = MANAGER_EXIT;
2122 return 0;
2123 }
2124
2125 /* This is a nop on init */
2126 break;
2127
2128 case 26:
2129 case 29: /* compatibility: used to be mapped to LOG_TARGET_SYSLOG_OR_KMSG */
2130 log_set_target(LOG_TARGET_JOURNAL_OR_KMSG);
2131 log_notice("Setting log target to journal-or-kmsg.");
2132 break;
2133
2134 case 27:
2135 log_set_target(LOG_TARGET_CONSOLE);
2136 log_notice("Setting log target to console.");
2137 break;
2138
2139 case 28:
2140 log_set_target(LOG_TARGET_KMSG);
2141 log_notice("Setting log target to kmsg.");
2142 break;
2143
2144 default:
2145 log_warning("Got unhandled signal <%s>.", signal_to_string(sfsi.ssi_signo));
2146 }
2147 }
2148 }
2149 }
2150
2151 if (sigchld)
2152 manager_dispatch_sigchld(m);
2153
2154 return 0;
2155 }
2156
2157 static int manager_dispatch_time_change_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
2158 Manager *m = userdata;
2159 Iterator i;
2160 Unit *u;
2161
2162 assert(m);
2163 assert(m->time_change_fd == fd);
2164
2165 log_struct(LOG_INFO,
2166 LOG_MESSAGE_ID(SD_MESSAGE_TIME_CHANGE),
2167 LOG_MESSAGE("Time has been changed"),
2168 NULL);
2169
2170 /* Restart the watch */
2171 m->time_change_event_source = sd_event_source_unref(m->time_change_event_source);
2172 m->time_change_fd = safe_close(m->time_change_fd);
2173
2174 manager_setup_time_change(m);
2175
2176 HASHMAP_FOREACH(u, m->units, i)
2177 if (UNIT_VTABLE(u)->time_change)
2178 UNIT_VTABLE(u)->time_change(u);
2179
2180 return 0;
2181 }
2182
2183 static int manager_dispatch_idle_pipe_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
2184 Manager *m = userdata;
2185
2186 assert(m);
2187 assert(m->idle_pipe[2] == fd);
2188
2189 m->no_console_output = m->n_on_console > 0;
2190
2191 manager_close_idle_pipe(m);
2192
2193 return 0;
2194 }
2195
2196 static int manager_dispatch_jobs_in_progress(sd_event_source *source, usec_t usec, void *userdata) {
2197 Manager *m = userdata;
2198 int r;
2199 uint64_t next;
2200
2201 assert(m);
2202 assert(source);
2203
2204 manager_print_jobs_in_progress(m);
2205
2206 next = now(CLOCK_MONOTONIC) + JOBS_IN_PROGRESS_PERIOD_USEC;
2207 r = sd_event_source_set_time(source, next);
2208 if (r < 0)
2209 return r;
2210
2211 return sd_event_source_set_enabled(source, SD_EVENT_ONESHOT);
2212 }
2213
2214 int manager_loop(Manager *m) {
2215 int r;
2216
2217 RATELIMIT_DEFINE(rl, 1*USEC_PER_SEC, 50000);
2218
2219 assert(m);
2220 m->exit_code = MANAGER_OK;
2221
2222 /* Release the path cache */
2223 m->unit_path_cache = set_free_free(m->unit_path_cache);
2224
2225 manager_check_finished(m);
2226
2227 /* There might still be some zombies hanging around from
2228 * before we were exec()'ed. Let's reap them. */
2229 r = manager_dispatch_sigchld(m);
2230 if (r < 0)
2231 return r;
2232
2233 while (m->exit_code == MANAGER_OK) {
2234 usec_t wait_usec;
2235
2236 if (m->runtime_watchdog > 0 && m->runtime_watchdog != USEC_INFINITY && MANAGER_IS_SYSTEM(m))
2237 watchdog_ping();
2238
2239 if (!ratelimit_test(&rl)) {
2240 /* Yay, something is going seriously wrong, pause a little */
2241 log_warning("Looping too fast. Throttling execution a little.");
2242 sleep(1);
2243 }
2244
2245 if (manager_dispatch_load_queue(m) > 0)
2246 continue;
2247
2248 if (manager_dispatch_gc_queue(m) > 0)
2249 continue;
2250
2251 if (manager_dispatch_cleanup_queue(m) > 0)
2252 continue;
2253
2254 if (manager_dispatch_cgroup_queue(m) > 0)
2255 continue;
2256
2257 if (manager_dispatch_dbus_queue(m) > 0)
2258 continue;
2259
2260 /* Sleep for half the watchdog time */
2261 if (m->runtime_watchdog > 0 && m->runtime_watchdog != USEC_INFINITY && MANAGER_IS_SYSTEM(m)) {
2262 wait_usec = m->runtime_watchdog / 2;
2263 if (wait_usec <= 0)
2264 wait_usec = 1;
2265 } else
2266 wait_usec = USEC_INFINITY;
2267
2268 r = sd_event_run(m->event, wait_usec);
2269 if (r < 0)
2270 return log_error_errno(r, "Failed to run event loop: %m");
2271 }
2272
2273 return m->exit_code;
2274 }
2275
2276 int manager_load_unit_from_dbus_path(Manager *m, const char *s, sd_bus_error *e, Unit **_u) {
2277 _cleanup_free_ char *n = NULL;
2278 sd_id128_t invocation_id;
2279 Unit *u;
2280 int r;
2281
2282 assert(m);
2283 assert(s);
2284 assert(_u);
2285
2286 r = unit_name_from_dbus_path(s, &n);
2287 if (r < 0)
2288 return r;
2289
2290 /* Permit addressing units by invocation ID: if the passed bus path is suffixed by a 128bit ID then we use it
2291 * as invocation ID. */
2292 r = sd_id128_from_string(n, &invocation_id);
2293 if (r >= 0) {
2294 u = hashmap_get(m->units_by_invocation_id, &invocation_id);
2295 if (u) {
2296 *_u = u;
2297 return 0;
2298 }
2299
2300 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));
2301 }
2302
2303 /* If this didn't work, we use the suffix as unit name. */
2304 r = manager_load_unit(m, n, NULL, e, &u);
2305 if (r < 0)
2306 return r;
2307
2308 *_u = u;
2309 return 0;
2310 }
2311
2312 int manager_get_job_from_dbus_path(Manager *m, const char *s, Job **_j) {
2313 const char *p;
2314 unsigned id;
2315 Job *j;
2316 int r;
2317
2318 assert(m);
2319 assert(s);
2320 assert(_j);
2321
2322 p = startswith(s, "/org/freedesktop/systemd1/job/");
2323 if (!p)
2324 return -EINVAL;
2325
2326 r = safe_atou(p, &id);
2327 if (r < 0)
2328 return r;
2329
2330 j = manager_get_job(m, id);
2331 if (!j)
2332 return -ENOENT;
2333
2334 *_j = j;
2335
2336 return 0;
2337 }
2338
2339 void manager_send_unit_audit(Manager *m, Unit *u, int type, bool success) {
2340
2341 #ifdef HAVE_AUDIT
2342 _cleanup_free_ char *p = NULL;
2343 const char *msg;
2344 int audit_fd, r;
2345
2346 if (!MANAGER_IS_SYSTEM(m))
2347 return;
2348
2349 audit_fd = get_audit_fd();
2350 if (audit_fd < 0)
2351 return;
2352
2353 /* Don't generate audit events if the service was already
2354 * started and we're just deserializing */
2355 if (MANAGER_IS_RELOADING(m))
2356 return;
2357
2358 if (u->type != UNIT_SERVICE)
2359 return;
2360
2361 r = unit_name_to_prefix_and_instance(u->id, &p);
2362 if (r < 0) {
2363 log_error_errno(r, "Failed to extract prefix and instance of unit name: %m");
2364 return;
2365 }
2366
2367 msg = strjoina("unit=", p);
2368 if (audit_log_user_comm_message(audit_fd, type, msg, "systemd", NULL, NULL, NULL, success) < 0) {
2369 if (errno == EPERM)
2370 /* We aren't allowed to send audit messages?
2371 * Then let's not retry again. */
2372 close_audit_fd();
2373 else
2374 log_warning_errno(errno, "Failed to send audit message: %m");
2375 }
2376 #endif
2377
2378 }
2379
2380 void manager_send_unit_plymouth(Manager *m, Unit *u) {
2381 static const union sockaddr_union sa = PLYMOUTH_SOCKET;
2382 _cleanup_free_ char *message = NULL;
2383 _cleanup_close_ int fd = -1;
2384 int n = 0;
2385
2386 /* Don't generate plymouth events if the service was already
2387 * started and we're just deserializing */
2388 if (MANAGER_IS_RELOADING(m))
2389 return;
2390
2391 if (!MANAGER_IS_SYSTEM(m))
2392 return;
2393
2394 if (detect_container() > 0)
2395 return;
2396
2397 if (u->type != UNIT_SERVICE &&
2398 u->type != UNIT_MOUNT &&
2399 u->type != UNIT_SWAP)
2400 return;
2401
2402 /* We set SOCK_NONBLOCK here so that we rather drop the
2403 * message then wait for plymouth */
2404 fd = socket(AF_UNIX, SOCK_STREAM|SOCK_CLOEXEC|SOCK_NONBLOCK, 0);
2405 if (fd < 0) {
2406 log_error_errno(errno, "socket() failed: %m");
2407 return;
2408 }
2409
2410 if (connect(fd, &sa.sa, SOCKADDR_UN_LEN(sa.un)) < 0) {
2411
2412 if (!IN_SET(errno, EPIPE, EAGAIN, ENOENT, ECONNREFUSED, ECONNRESET, ECONNABORTED))
2413 log_error_errno(errno, "connect() failed: %m");
2414 return;
2415 }
2416
2417 if (asprintf(&message, "U\002%c%s%n", (int) (strlen(u->id) + 1), u->id, &n) < 0) {
2418 log_oom();
2419 return;
2420 }
2421
2422 errno = 0;
2423 if (write(fd, message, n + 1) != n + 1)
2424 if (!IN_SET(errno, EPIPE, EAGAIN, ENOENT, ECONNREFUSED, ECONNRESET, ECONNABORTED))
2425 log_error_errno(errno, "Failed to write Plymouth message: %m");
2426 }
2427
2428 int manager_open_serialization(Manager *m, FILE **_f) {
2429 const char *path;
2430 int fd = -1;
2431 FILE *f;
2432
2433 assert(_f);
2434
2435 path = MANAGER_IS_SYSTEM(m) ? "/run/systemd" : "/tmp";
2436 fd = open_tmpfile_unlinkable(path, O_RDWR|O_CLOEXEC);
2437 if (fd < 0)
2438 return -errno;
2439
2440 log_debug("Serializing state to %s", path);
2441
2442 f = fdopen(fd, "w+");
2443 if (!f) {
2444 safe_close(fd);
2445 return -errno;
2446 }
2447
2448 *_f = f;
2449
2450 return 0;
2451 }
2452
2453 int manager_serialize(Manager *m, FILE *f, FDSet *fds, bool switching_root) {
2454 Iterator i;
2455 Unit *u;
2456 const char *t;
2457 char **e;
2458 int r;
2459
2460 assert(m);
2461 assert(f);
2462 assert(fds);
2463
2464 m->n_reloading++;
2465
2466 fprintf(f, "current-job-id=%"PRIu32"\n", m->current_job_id);
2467 fprintf(f, "taint-usr=%s\n", yes_no(m->taint_usr));
2468 fprintf(f, "n-installed-jobs=%u\n", m->n_installed_jobs);
2469 fprintf(f, "n-failed-jobs=%u\n", m->n_failed_jobs);
2470
2471 dual_timestamp_serialize(f, "firmware-timestamp", &m->firmware_timestamp);
2472 dual_timestamp_serialize(f, "loader-timestamp", &m->loader_timestamp);
2473 dual_timestamp_serialize(f, "kernel-timestamp", &m->kernel_timestamp);
2474 dual_timestamp_serialize(f, "initrd-timestamp", &m->initrd_timestamp);
2475
2476 if (!in_initrd()) {
2477 dual_timestamp_serialize(f, "userspace-timestamp", &m->userspace_timestamp);
2478 dual_timestamp_serialize(f, "finish-timestamp", &m->finish_timestamp);
2479 dual_timestamp_serialize(f, "security-start-timestamp", &m->security_start_timestamp);
2480 dual_timestamp_serialize(f, "security-finish-timestamp", &m->security_finish_timestamp);
2481 dual_timestamp_serialize(f, "generators-start-timestamp", &m->generators_start_timestamp);
2482 dual_timestamp_serialize(f, "generators-finish-timestamp", &m->generators_finish_timestamp);
2483 dual_timestamp_serialize(f, "units-load-start-timestamp", &m->units_load_start_timestamp);
2484 dual_timestamp_serialize(f, "units-load-finish-timestamp", &m->units_load_finish_timestamp);
2485 }
2486
2487 if (!switching_root) {
2488 STRV_FOREACH(e, m->environment) {
2489 _cleanup_free_ char *ce;
2490
2491 ce = cescape(*e);
2492 if (!ce)
2493 return -ENOMEM;
2494
2495 fprintf(f, "env=%s\n", *e);
2496 }
2497 }
2498
2499 if (m->notify_fd >= 0) {
2500 int copy;
2501
2502 copy = fdset_put_dup(fds, m->notify_fd);
2503 if (copy < 0)
2504 return copy;
2505
2506 fprintf(f, "notify-fd=%i\n", copy);
2507 fprintf(f, "notify-socket=%s\n", m->notify_socket);
2508 }
2509
2510 if (m->cgroups_agent_fd >= 0) {
2511 int copy;
2512
2513 copy = fdset_put_dup(fds, m->cgroups_agent_fd);
2514 if (copy < 0)
2515 return copy;
2516
2517 fprintf(f, "cgroups-agent-fd=%i\n", copy);
2518 }
2519
2520 if (m->user_lookup_fds[0] >= 0) {
2521 int copy0, copy1;
2522
2523 copy0 = fdset_put_dup(fds, m->user_lookup_fds[0]);
2524 if (copy0 < 0)
2525 return copy0;
2526
2527 copy1 = fdset_put_dup(fds, m->user_lookup_fds[1]);
2528 if (copy1 < 0)
2529 return copy1;
2530
2531 fprintf(f, "user-lookup=%i %i\n", copy0, copy1);
2532 }
2533
2534 bus_track_serialize(m->subscribed, f, "subscribed");
2535
2536 r = dynamic_user_serialize(m, f, fds);
2537 if (r < 0)
2538 return r;
2539
2540 manager_serialize_uid_refs(m, f);
2541 manager_serialize_gid_refs(m, f);
2542
2543 fputc('\n', f);
2544
2545 HASHMAP_FOREACH_KEY(u, t, m->units, i) {
2546 if (u->id != t)
2547 continue;
2548
2549 /* Start marker */
2550 fputs(u->id, f);
2551 fputc('\n', f);
2552
2553 r = unit_serialize(u, f, fds, !switching_root);
2554 if (r < 0) {
2555 m->n_reloading--;
2556 return r;
2557 }
2558 }
2559
2560 assert(m->n_reloading > 0);
2561 m->n_reloading--;
2562
2563 if (ferror(f))
2564 return -EIO;
2565
2566 r = bus_fdset_add_all(m, fds);
2567 if (r < 0)
2568 return r;
2569
2570 return 0;
2571 }
2572
2573 int manager_deserialize(Manager *m, FILE *f, FDSet *fds) {
2574 int r = 0;
2575
2576 assert(m);
2577 assert(f);
2578
2579 log_debug("Deserializing state...");
2580
2581 m->n_reloading++;
2582
2583 for (;;) {
2584 char line[LINE_MAX], *l;
2585
2586 if (!fgets(line, sizeof(line), f)) {
2587 if (feof(f))
2588 r = 0;
2589 else
2590 r = -errno;
2591
2592 goto finish;
2593 }
2594
2595 char_array_0(line);
2596 l = strstrip(line);
2597
2598 if (l[0] == 0)
2599 break;
2600
2601 if (startswith(l, "current-job-id=")) {
2602 uint32_t id;
2603
2604 if (safe_atou32(l+15, &id) < 0)
2605 log_debug("Failed to parse current job id value %s", l+15);
2606 else
2607 m->current_job_id = MAX(m->current_job_id, id);
2608
2609 } else if (startswith(l, "n-installed-jobs=")) {
2610 uint32_t n;
2611
2612 if (safe_atou32(l+17, &n) < 0)
2613 log_debug("Failed to parse installed jobs counter %s", l+17);
2614 else
2615 m->n_installed_jobs += n;
2616
2617 } else if (startswith(l, "n-failed-jobs=")) {
2618 uint32_t n;
2619
2620 if (safe_atou32(l+14, &n) < 0)
2621 log_debug("Failed to parse failed jobs counter %s", l+14);
2622 else
2623 m->n_failed_jobs += n;
2624
2625 } else if (startswith(l, "taint-usr=")) {
2626 int b;
2627
2628 b = parse_boolean(l+10);
2629 if (b < 0)
2630 log_debug("Failed to parse taint /usr flag %s", l+10);
2631 else
2632 m->taint_usr = m->taint_usr || b;
2633
2634 } else if (startswith(l, "firmware-timestamp="))
2635 dual_timestamp_deserialize(l+19, &m->firmware_timestamp);
2636 else if (startswith(l, "loader-timestamp="))
2637 dual_timestamp_deserialize(l+17, &m->loader_timestamp);
2638 else if (startswith(l, "kernel-timestamp="))
2639 dual_timestamp_deserialize(l+17, &m->kernel_timestamp);
2640 else if (startswith(l, "initrd-timestamp="))
2641 dual_timestamp_deserialize(l+17, &m->initrd_timestamp);
2642 else if (startswith(l, "userspace-timestamp="))
2643 dual_timestamp_deserialize(l+20, &m->userspace_timestamp);
2644 else if (startswith(l, "finish-timestamp="))
2645 dual_timestamp_deserialize(l+17, &m->finish_timestamp);
2646 else if (startswith(l, "security-start-timestamp="))
2647 dual_timestamp_deserialize(l+25, &m->security_start_timestamp);
2648 else if (startswith(l, "security-finish-timestamp="))
2649 dual_timestamp_deserialize(l+26, &m->security_finish_timestamp);
2650 else if (startswith(l, "generators-start-timestamp="))
2651 dual_timestamp_deserialize(l+27, &m->generators_start_timestamp);
2652 else if (startswith(l, "generators-finish-timestamp="))
2653 dual_timestamp_deserialize(l+28, &m->generators_finish_timestamp);
2654 else if (startswith(l, "units-load-start-timestamp="))
2655 dual_timestamp_deserialize(l+27, &m->units_load_start_timestamp);
2656 else if (startswith(l, "units-load-finish-timestamp="))
2657 dual_timestamp_deserialize(l+28, &m->units_load_finish_timestamp);
2658 else if (startswith(l, "env=")) {
2659 _cleanup_free_ char *uce = NULL;
2660 char **e;
2661
2662 r = cunescape(l + 4, UNESCAPE_RELAX, &uce);
2663 if (r < 0)
2664 goto finish;
2665
2666 e = strv_env_set(m->environment, uce);
2667 if (!e) {
2668 r = -ENOMEM;
2669 goto finish;
2670 }
2671
2672 strv_free(m->environment);
2673 m->environment = e;
2674
2675 } else if (startswith(l, "notify-fd=")) {
2676 int fd;
2677
2678 if (safe_atoi(l + 10, &fd) < 0 || fd < 0 || !fdset_contains(fds, fd))
2679 log_debug("Failed to parse notify fd: %s", l + 10);
2680 else {
2681 m->notify_event_source = sd_event_source_unref(m->notify_event_source);
2682 safe_close(m->notify_fd);
2683 m->notify_fd = fdset_remove(fds, fd);
2684 }
2685
2686 } else if (startswith(l, "notify-socket=")) {
2687 char *n;
2688
2689 n = strdup(l+14);
2690 if (!n) {
2691 r = -ENOMEM;
2692 goto finish;
2693 }
2694
2695 free(m->notify_socket);
2696 m->notify_socket = n;
2697
2698 } else if (startswith(l, "cgroups-agent-fd=")) {
2699 int fd;
2700
2701 if (safe_atoi(l + 17, &fd) < 0 || fd < 0 || !fdset_contains(fds, fd))
2702 log_debug("Failed to parse cgroups agent fd: %s", l + 10);
2703 else {
2704 m->cgroups_agent_event_source = sd_event_source_unref(m->cgroups_agent_event_source);
2705 safe_close(m->cgroups_agent_fd);
2706 m->cgroups_agent_fd = fdset_remove(fds, fd);
2707 }
2708
2709 } else if (startswith(l, "user-lookup=")) {
2710 int fd0, fd1;
2711
2712 if (sscanf(l + 12, "%i %i", &fd0, &fd1) != 2 || fd0 < 0 || fd1 < 0 || fd0 == fd1 || !fdset_contains(fds, fd0) || !fdset_contains(fds, fd1))
2713 log_debug("Failed to parse user lookup fd: %s", l + 12);
2714 else {
2715 m->user_lookup_event_source = sd_event_source_unref(m->user_lookup_event_source);
2716 safe_close_pair(m->user_lookup_fds);
2717 m->user_lookup_fds[0] = fdset_remove(fds, fd0);
2718 m->user_lookup_fds[1] = fdset_remove(fds, fd1);
2719 }
2720
2721 } else if (startswith(l, "dynamic-user="))
2722 dynamic_user_deserialize_one(m, l + 13, fds);
2723 else if (startswith(l, "destroy-ipc-uid="))
2724 manager_deserialize_uid_refs_one(m, l + 16);
2725 else if (startswith(l, "destroy-ipc-gid="))
2726 manager_deserialize_gid_refs_one(m, l + 16);
2727 else if (startswith(l, "subscribed=")) {
2728
2729 if (strv_extend(&m->deserialized_subscribed, l+11) < 0)
2730 log_oom();
2731
2732 } else if (!startswith(l, "kdbus-fd=")) /* ignore this one */
2733 log_debug("Unknown serialization item '%s'", l);
2734 }
2735
2736 for (;;) {
2737 Unit *u;
2738 char name[UNIT_NAME_MAX+2];
2739
2740 /* Start marker */
2741 if (!fgets(name, sizeof(name), f)) {
2742 if (feof(f))
2743 r = 0;
2744 else
2745 r = -errno;
2746
2747 goto finish;
2748 }
2749
2750 char_array_0(name);
2751
2752 r = manager_load_unit(m, strstrip(name), NULL, NULL, &u);
2753 if (r < 0)
2754 goto finish;
2755
2756 r = unit_deserialize(u, f, fds);
2757 if (r < 0)
2758 goto finish;
2759 }
2760
2761 finish:
2762 if (ferror(f))
2763 r = -EIO;
2764
2765 assert(m->n_reloading > 0);
2766 m->n_reloading--;
2767
2768 return r;
2769 }
2770
2771 int manager_reload(Manager *m) {
2772 int r, q;
2773 _cleanup_fclose_ FILE *f = NULL;
2774 _cleanup_fdset_free_ FDSet *fds = NULL;
2775
2776 assert(m);
2777
2778 r = manager_open_serialization(m, &f);
2779 if (r < 0)
2780 return r;
2781
2782 m->n_reloading++;
2783 bus_manager_send_reloading(m, true);
2784
2785 fds = fdset_new();
2786 if (!fds) {
2787 m->n_reloading--;
2788 return -ENOMEM;
2789 }
2790
2791 r = manager_serialize(m, f, fds, false);
2792 if (r < 0) {
2793 m->n_reloading--;
2794 return r;
2795 }
2796
2797 if (fseeko(f, 0, SEEK_SET) < 0) {
2798 m->n_reloading--;
2799 return -errno;
2800 }
2801
2802 /* From here on there is no way back. */
2803 manager_clear_jobs_and_units(m);
2804 lookup_paths_flush_generator(&m->lookup_paths);
2805 lookup_paths_free(&m->lookup_paths);
2806 dynamic_user_vacuum(m, false);
2807 m->uid_refs = hashmap_free(m->uid_refs);
2808 m->gid_refs = hashmap_free(m->gid_refs);
2809
2810 q = lookup_paths_init(&m->lookup_paths, m->unit_file_scope, 0, NULL);
2811 if (q < 0 && r >= 0)
2812 r = q;
2813
2814 /* Find new unit paths */
2815 q = manager_run_generators(m);
2816 if (q < 0 && r >= 0)
2817 r = q;
2818
2819 lookup_paths_reduce(&m->lookup_paths);
2820 manager_build_unit_path_cache(m);
2821
2822 /* First, enumerate what we can from all config files */
2823 manager_enumerate(m);
2824
2825 /* Second, deserialize our stored data */
2826 q = manager_deserialize(m, f, fds);
2827 if (q < 0 && r >= 0)
2828 r = q;
2829
2830 fclose(f);
2831 f = NULL;
2832
2833 /* Re-register notify_fd as event source */
2834 q = manager_setup_notify(m);
2835 if (q < 0 && r >= 0)
2836 r = q;
2837
2838 q = manager_setup_cgroups_agent(m);
2839 if (q < 0 && r >= 0)
2840 r = q;
2841
2842 q = manager_setup_user_lookup_fd(m);
2843 if (q < 0 && r >= 0)
2844 r = q;
2845
2846 /* Third, fire things up! */
2847 manager_coldplug(m);
2848
2849 /* Release any dynamic users no longer referenced */
2850 dynamic_user_vacuum(m, true);
2851
2852 /* Release any references to UIDs/GIDs no longer referenced, and destroy any IPC owned by them */
2853 manager_vacuum_uid_refs(m);
2854 manager_vacuum_gid_refs(m);
2855
2856 /* Sync current state of bus names with our set of listening units */
2857 if (m->api_bus)
2858 manager_sync_bus_names(m, m->api_bus);
2859
2860 assert(m->n_reloading > 0);
2861 m->n_reloading--;
2862
2863 m->send_reloading_done = true;
2864
2865 return r;
2866 }
2867
2868 void manager_reset_failed(Manager *m) {
2869 Unit *u;
2870 Iterator i;
2871
2872 assert(m);
2873
2874 HASHMAP_FOREACH(u, m->units, i)
2875 unit_reset_failed(u);
2876 }
2877
2878 bool manager_unit_inactive_or_pending(Manager *m, const char *name) {
2879 Unit *u;
2880
2881 assert(m);
2882 assert(name);
2883
2884 /* Returns true if the unit is inactive or going down */
2885 u = manager_get_unit(m, name);
2886 if (!u)
2887 return true;
2888
2889 return unit_inactive_or_pending(u);
2890 }
2891
2892 static void manager_notify_finished(Manager *m) {
2893 char userspace[FORMAT_TIMESPAN_MAX], initrd[FORMAT_TIMESPAN_MAX], kernel[FORMAT_TIMESPAN_MAX], sum[FORMAT_TIMESPAN_MAX];
2894 usec_t firmware_usec, loader_usec, kernel_usec, initrd_usec, userspace_usec, total_usec;
2895
2896 if (m->test_run)
2897 return;
2898
2899 if (MANAGER_IS_SYSTEM(m) && detect_container() <= 0) {
2900
2901 /* Note that m->kernel_usec.monotonic is always at 0,
2902 * and m->firmware_usec.monotonic and
2903 * m->loader_usec.monotonic should be considered
2904 * negative values. */
2905
2906 firmware_usec = m->firmware_timestamp.monotonic - m->loader_timestamp.monotonic;
2907 loader_usec = m->loader_timestamp.monotonic - m->kernel_timestamp.monotonic;
2908 userspace_usec = m->finish_timestamp.monotonic - m->userspace_timestamp.monotonic;
2909 total_usec = m->firmware_timestamp.monotonic + m->finish_timestamp.monotonic;
2910
2911 if (dual_timestamp_is_set(&m->initrd_timestamp)) {
2912
2913 kernel_usec = m->initrd_timestamp.monotonic - m->kernel_timestamp.monotonic;
2914 initrd_usec = m->userspace_timestamp.monotonic - m->initrd_timestamp.monotonic;
2915
2916 log_struct(LOG_INFO,
2917 LOG_MESSAGE_ID(SD_MESSAGE_STARTUP_FINISHED),
2918 "KERNEL_USEC="USEC_FMT, kernel_usec,
2919 "INITRD_USEC="USEC_FMT, initrd_usec,
2920 "USERSPACE_USEC="USEC_FMT, userspace_usec,
2921 LOG_MESSAGE("Startup finished in %s (kernel) + %s (initrd) + %s (userspace) = %s.",
2922 format_timespan(kernel, sizeof(kernel), kernel_usec, USEC_PER_MSEC),
2923 format_timespan(initrd, sizeof(initrd), initrd_usec, USEC_PER_MSEC),
2924 format_timespan(userspace, sizeof(userspace), userspace_usec, USEC_PER_MSEC),
2925 format_timespan(sum, sizeof(sum), total_usec, USEC_PER_MSEC)),
2926 NULL);
2927 } else {
2928 kernel_usec = m->userspace_timestamp.monotonic - m->kernel_timestamp.monotonic;
2929 initrd_usec = 0;
2930
2931 log_struct(LOG_INFO,
2932 LOG_MESSAGE_ID(SD_MESSAGE_STARTUP_FINISHED),
2933 "KERNEL_USEC="USEC_FMT, kernel_usec,
2934 "USERSPACE_USEC="USEC_FMT, userspace_usec,
2935 LOG_MESSAGE("Startup finished in %s (kernel) + %s (userspace) = %s.",
2936 format_timespan(kernel, sizeof(kernel), kernel_usec, USEC_PER_MSEC),
2937 format_timespan(userspace, sizeof(userspace), userspace_usec, USEC_PER_MSEC),
2938 format_timespan(sum, sizeof(sum), total_usec, USEC_PER_MSEC)),
2939 NULL);
2940 }
2941 } else {
2942 firmware_usec = loader_usec = initrd_usec = kernel_usec = 0;
2943 total_usec = userspace_usec = m->finish_timestamp.monotonic - m->userspace_timestamp.monotonic;
2944
2945 log_struct(LOG_INFO,
2946 LOG_MESSAGE_ID(SD_MESSAGE_STARTUP_FINISHED),
2947 "USERSPACE_USEC="USEC_FMT, userspace_usec,
2948 LOG_MESSAGE("Startup finished in %s.",
2949 format_timespan(sum, sizeof(sum), total_usec, USEC_PER_MSEC)),
2950 NULL);
2951 }
2952
2953 bus_manager_send_finished(m, firmware_usec, loader_usec, kernel_usec, initrd_usec, userspace_usec, total_usec);
2954
2955 sd_notifyf(false,
2956 "READY=1\n"
2957 "STATUS=Startup finished in %s.",
2958 format_timespan(sum, sizeof(sum), total_usec, USEC_PER_MSEC));
2959 }
2960
2961 void manager_check_finished(Manager *m) {
2962 assert(m);
2963
2964 if (MANAGER_IS_RELOADING(m))
2965 return;
2966
2967 /* Verify that we are actually running currently. Initially
2968 * the exit code is set to invalid, and during operation it is
2969 * then set to MANAGER_OK */
2970 if (m->exit_code != MANAGER_OK)
2971 return;
2972
2973 if (hashmap_size(m->jobs) > 0) {
2974 if (m->jobs_in_progress_event_source)
2975 /* Ignore any failure, this is only for feedback */
2976 (void) sd_event_source_set_time(m->jobs_in_progress_event_source, now(CLOCK_MONOTONIC) + JOBS_IN_PROGRESS_WAIT_USEC);
2977
2978 return;
2979 }
2980
2981 manager_flip_auto_status(m, false);
2982
2983 /* Notify Type=idle units that we are done now */
2984 manager_close_idle_pipe(m);
2985
2986 /* Turn off confirm spawn now */
2987 m->confirm_spawn = false;
2988
2989 /* No need to update ask password status when we're going non-interactive */
2990 manager_close_ask_password(m);
2991
2992 /* This is no longer the first boot */
2993 manager_set_first_boot(m, false);
2994
2995 if (dual_timestamp_is_set(&m->finish_timestamp))
2996 return;
2997
2998 dual_timestamp_get(&m->finish_timestamp);
2999
3000 manager_notify_finished(m);
3001
3002 manager_invalidate_startup_units(m);
3003 }
3004
3005 static int manager_run_generators(Manager *m) {
3006 _cleanup_strv_free_ char **paths = NULL;
3007 const char *argv[5];
3008 char **path;
3009 int r;
3010
3011 assert(m);
3012
3013 if (m->test_run)
3014 return 0;
3015
3016 paths = generator_binary_paths(m->unit_file_scope);
3017 if (!paths)
3018 return log_oom();
3019
3020 /* Optimize by skipping the whole process by not creating output directories
3021 * if no generators are found. */
3022 STRV_FOREACH(path, paths) {
3023 if (access(*path, F_OK) >= 0)
3024 goto found;
3025 if (errno != ENOENT)
3026 log_warning_errno(errno, "Failed to open generator directory %s: %m", *path);
3027 }
3028
3029 return 0;
3030
3031 found:
3032 r = lookup_paths_mkdir_generator(&m->lookup_paths);
3033 if (r < 0)
3034 goto finish;
3035
3036 argv[0] = NULL; /* Leave this empty, execute_directory() will fill something in */
3037 argv[1] = m->lookup_paths.generator;
3038 argv[2] = m->lookup_paths.generator_early;
3039 argv[3] = m->lookup_paths.generator_late;
3040 argv[4] = NULL;
3041
3042 RUN_WITH_UMASK(0022)
3043 execute_directories((const char* const*) paths, DEFAULT_TIMEOUT_USEC, (char**) argv);
3044
3045 finish:
3046 lookup_paths_trim_generator(&m->lookup_paths);
3047 return r;
3048 }
3049
3050 int manager_environment_add(Manager *m, char **minus, char **plus) {
3051 char **a = NULL, **b = NULL, **l;
3052 assert(m);
3053
3054 l = m->environment;
3055
3056 if (!strv_isempty(minus)) {
3057 a = strv_env_delete(l, 1, minus);
3058 if (!a)
3059 return -ENOMEM;
3060
3061 l = a;
3062 }
3063
3064 if (!strv_isempty(plus)) {
3065 b = strv_env_merge(2, l, plus);
3066 if (!b) {
3067 strv_free(a);
3068 return -ENOMEM;
3069 }
3070
3071 l = b;
3072 }
3073
3074 if (m->environment != l)
3075 strv_free(m->environment);
3076 if (a != l)
3077 strv_free(a);
3078 if (b != l)
3079 strv_free(b);
3080
3081 m->environment = l;
3082 manager_clean_environment(m);
3083 strv_sort(m->environment);
3084
3085 return 0;
3086 }
3087
3088 int manager_set_default_rlimits(Manager *m, struct rlimit **default_rlimit) {
3089 int i;
3090
3091 assert(m);
3092
3093 for (i = 0; i < _RLIMIT_MAX; i++) {
3094 m->rlimit[i] = mfree(m->rlimit[i]);
3095
3096 if (!default_rlimit[i])
3097 continue;
3098
3099 m->rlimit[i] = newdup(struct rlimit, default_rlimit[i], 1);
3100 if (!m->rlimit[i])
3101 return -ENOMEM;
3102 }
3103
3104 return 0;
3105 }
3106
3107 void manager_recheck_journal(Manager *m) {
3108 Unit *u;
3109
3110 assert(m);
3111
3112 if (!MANAGER_IS_SYSTEM(m))
3113 return;
3114
3115 u = manager_get_unit(m, SPECIAL_JOURNALD_SOCKET);
3116 if (u && SOCKET(u)->state != SOCKET_RUNNING) {
3117 log_close_journal();
3118 return;
3119 }
3120
3121 u = manager_get_unit(m, SPECIAL_JOURNALD_SERVICE);
3122 if (u && SERVICE(u)->state != SERVICE_RUNNING) {
3123 log_close_journal();
3124 return;
3125 }
3126
3127 /* Hmm, OK, so the socket is fully up and the service is up
3128 * too, then let's make use of the thing. */
3129 log_open();
3130 }
3131
3132 void manager_set_show_status(Manager *m, ShowStatus mode) {
3133 assert(m);
3134 assert(IN_SET(mode, SHOW_STATUS_AUTO, SHOW_STATUS_NO, SHOW_STATUS_YES, SHOW_STATUS_TEMPORARY));
3135
3136 if (!MANAGER_IS_SYSTEM(m))
3137 return;
3138
3139 if (m->show_status != mode)
3140 log_debug("%s showing of status.",
3141 mode == SHOW_STATUS_NO ? "Disabling" : "Enabling");
3142 m->show_status = mode;
3143
3144 if (mode > 0)
3145 (void) touch("/run/systemd/show-status");
3146 else
3147 (void) unlink("/run/systemd/show-status");
3148 }
3149
3150 static bool manager_get_show_status(Manager *m, StatusType type) {
3151 assert(m);
3152
3153 if (!MANAGER_IS_SYSTEM(m))
3154 return false;
3155
3156 if (m->no_console_output)
3157 return false;
3158
3159 if (!IN_SET(manager_state(m), MANAGER_INITIALIZING, MANAGER_STARTING, MANAGER_STOPPING))
3160 return false;
3161
3162 /* If we cannot find out the status properly, just proceed. */
3163 if (type != STATUS_TYPE_EMERGENCY && manager_check_ask_password(m) > 0)
3164 return false;
3165
3166 if (m->show_status > 0)
3167 return true;
3168
3169 return false;
3170 }
3171
3172 void manager_set_first_boot(Manager *m, bool b) {
3173 assert(m);
3174
3175 if (!MANAGER_IS_SYSTEM(m))
3176 return;
3177
3178 if (m->first_boot != (int) b) {
3179 if (b)
3180 (void) touch("/run/systemd/first-boot");
3181 else
3182 (void) unlink("/run/systemd/first-boot");
3183 }
3184
3185 m->first_boot = b;
3186 }
3187
3188 void manager_status_printf(Manager *m, StatusType type, const char *status, const char *format, ...) {
3189 va_list ap;
3190
3191 /* If m is NULL, assume we're after shutdown and let the messages through. */
3192
3193 if (m && !manager_get_show_status(m, type))
3194 return;
3195
3196 /* XXX We should totally drop the check for ephemeral here
3197 * and thus effectively make 'Type=idle' pointless. */
3198 if (type == STATUS_TYPE_EPHEMERAL && m && m->n_on_console > 0)
3199 return;
3200
3201 va_start(ap, format);
3202 status_vprintf(status, true, type == STATUS_TYPE_EPHEMERAL, format, ap);
3203 va_end(ap);
3204 }
3205
3206 Set *manager_get_units_requiring_mounts_for(Manager *m, const char *path) {
3207 char p[strlen(path)+1];
3208
3209 assert(m);
3210 assert(path);
3211
3212 strcpy(p, path);
3213 path_kill_slashes(p);
3214
3215 return hashmap_get(m->units_requiring_mounts_for, streq(p, "/") ? "" : p);
3216 }
3217
3218 const char *manager_get_runtime_prefix(Manager *m) {
3219 assert(m);
3220
3221 return MANAGER_IS_SYSTEM(m) ?
3222 "/run" :
3223 getenv("XDG_RUNTIME_DIR");
3224 }
3225
3226 int manager_update_failed_units(Manager *m, Unit *u, bool failed) {
3227 unsigned size;
3228 int r;
3229
3230 assert(m);
3231 assert(u->manager == m);
3232
3233 size = set_size(m->failed_units);
3234
3235 if (failed) {
3236 r = set_ensure_allocated(&m->failed_units, NULL);
3237 if (r < 0)
3238 return log_oom();
3239
3240 if (set_put(m->failed_units, u) < 0)
3241 return log_oom();
3242 } else
3243 (void) set_remove(m->failed_units, u);
3244
3245 if (set_size(m->failed_units) != size)
3246 bus_manager_send_change_signal(m);
3247
3248 return 0;
3249 }
3250
3251 ManagerState manager_state(Manager *m) {
3252 Unit *u;
3253
3254 assert(m);
3255
3256 /* Did we ever finish booting? If not then we are still starting up */
3257 if (!dual_timestamp_is_set(&m->finish_timestamp)) {
3258
3259 u = manager_get_unit(m, SPECIAL_BASIC_TARGET);
3260 if (!u || !UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u)))
3261 return MANAGER_INITIALIZING;
3262
3263 return MANAGER_STARTING;
3264 }
3265
3266 /* Is the special shutdown target queued? If so, we are in shutdown state */
3267 u = manager_get_unit(m, SPECIAL_SHUTDOWN_TARGET);
3268 if (u && u->job && IN_SET(u->job->type, JOB_START, JOB_RESTART, JOB_RELOAD_OR_START))
3269 return MANAGER_STOPPING;
3270
3271 /* Are the rescue or emergency targets active or queued? If so we are in maintenance state */
3272 u = manager_get_unit(m, SPECIAL_RESCUE_TARGET);
3273 if (u && (UNIT_IS_ACTIVE_OR_ACTIVATING(unit_active_state(u)) ||
3274 (u->job && IN_SET(u->job->type, JOB_START, JOB_RESTART, JOB_RELOAD_OR_START))))
3275 return MANAGER_MAINTENANCE;
3276
3277 u = manager_get_unit(m, SPECIAL_EMERGENCY_TARGET);
3278 if (u && (UNIT_IS_ACTIVE_OR_ACTIVATING(unit_active_state(u)) ||
3279 (u->job && IN_SET(u->job->type, JOB_START, JOB_RESTART, JOB_RELOAD_OR_START))))
3280 return MANAGER_MAINTENANCE;
3281
3282 /* Are there any failed units? If so, we are in degraded mode */
3283 if (set_size(m->failed_units) > 0)
3284 return MANAGER_DEGRADED;
3285
3286 return MANAGER_RUNNING;
3287 }
3288
3289 #define DESTROY_IPC_FLAG (UINT32_C(1) << 31)
3290
3291 static void manager_unref_uid_internal(
3292 Manager *m,
3293 Hashmap **uid_refs,
3294 uid_t uid,
3295 bool destroy_now,
3296 int (*_clean_ipc)(uid_t uid)) {
3297
3298 uint32_t c, n;
3299
3300 assert(m);
3301 assert(uid_refs);
3302 assert(uid_is_valid(uid));
3303 assert(_clean_ipc);
3304
3305 /* A generic implementation, covering both manager_unref_uid() and manager_unref_gid(), under the assumption
3306 * that uid_t and gid_t are actually defined the same way, with the same validity rules.
3307 *
3308 * We store a hashmap where the UID/GID is they key and the value is a 32bit reference counter, whose highest
3309 * bit is used as flag for marking UIDs/GIDs whose IPC objects to remove when the last reference to the UID/GID
3310 * is dropped. The flag is set to on, once at least one reference from a unit where RemoveIPC= is set is added
3311 * on a UID/GID. It is reset when the UID's/GID's reference counter drops to 0 again. */
3312
3313 assert_cc(sizeof(uid_t) == sizeof(gid_t));
3314 assert_cc(UID_INVALID == (uid_t) GID_INVALID);
3315
3316 if (uid == 0) /* We don't keep track of root, and will never destroy it */
3317 return;
3318
3319 c = PTR_TO_UINT32(hashmap_get(*uid_refs, UID_TO_PTR(uid)));
3320
3321 n = c & ~DESTROY_IPC_FLAG;
3322 assert(n > 0);
3323 n--;
3324
3325 if (destroy_now && n == 0) {
3326 hashmap_remove(*uid_refs, UID_TO_PTR(uid));
3327
3328 if (c & DESTROY_IPC_FLAG) {
3329 log_debug("%s " UID_FMT " is no longer referenced, cleaning up its IPC.",
3330 _clean_ipc == clean_ipc_by_uid ? "UID" : "GID",
3331 uid);
3332 (void) _clean_ipc(uid);
3333 }
3334 } else {
3335 c = n | (c & DESTROY_IPC_FLAG);
3336 assert_se(hashmap_update(*uid_refs, UID_TO_PTR(uid), UINT32_TO_PTR(c)) >= 0);
3337 }
3338 }
3339
3340 void manager_unref_uid(Manager *m, uid_t uid, bool destroy_now) {
3341 manager_unref_uid_internal(m, &m->uid_refs, uid, destroy_now, clean_ipc_by_uid);
3342 }
3343
3344 void manager_unref_gid(Manager *m, gid_t gid, bool destroy_now) {
3345 manager_unref_uid_internal(m, &m->gid_refs, (uid_t) gid, destroy_now, clean_ipc_by_gid);
3346 }
3347
3348 static int manager_ref_uid_internal(
3349 Manager *m,
3350 Hashmap **uid_refs,
3351 uid_t uid,
3352 bool clean_ipc) {
3353
3354 uint32_t c, n;
3355 int r;
3356
3357 assert(m);
3358 assert(uid_refs);
3359 assert(uid_is_valid(uid));
3360
3361 /* A generic implementation, covering both manager_ref_uid() and manager_ref_gid(), under the assumption
3362 * that uid_t and gid_t are actually defined the same way, with the same validity rules. */
3363
3364 assert_cc(sizeof(uid_t) == sizeof(gid_t));
3365 assert_cc(UID_INVALID == (uid_t) GID_INVALID);
3366
3367 if (uid == 0) /* We don't keep track of root, and will never destroy it */
3368 return 0;
3369
3370 r = hashmap_ensure_allocated(uid_refs, &trivial_hash_ops);
3371 if (r < 0)
3372 return r;
3373
3374 c = PTR_TO_UINT32(hashmap_get(*uid_refs, UID_TO_PTR(uid)));
3375
3376 n = c & ~DESTROY_IPC_FLAG;
3377 n++;
3378
3379 if (n & DESTROY_IPC_FLAG) /* check for overflow */
3380 return -EOVERFLOW;
3381
3382 c = n | (c & DESTROY_IPC_FLAG) | (clean_ipc ? DESTROY_IPC_FLAG : 0);
3383
3384 return hashmap_replace(*uid_refs, UID_TO_PTR(uid), UINT32_TO_PTR(c));
3385 }
3386
3387 int manager_ref_uid(Manager *m, uid_t uid, bool clean_ipc) {
3388 return manager_ref_uid_internal(m, &m->uid_refs, uid, clean_ipc);
3389 }
3390
3391 int manager_ref_gid(Manager *m, gid_t gid, bool clean_ipc) {
3392 return manager_ref_uid_internal(m, &m->gid_refs, (uid_t) gid, clean_ipc);
3393 }
3394
3395 static void manager_vacuum_uid_refs_internal(
3396 Manager *m,
3397 Hashmap **uid_refs,
3398 int (*_clean_ipc)(uid_t uid)) {
3399
3400 Iterator i;
3401 void *p, *k;
3402
3403 assert(m);
3404 assert(uid_refs);
3405 assert(_clean_ipc);
3406
3407 HASHMAP_FOREACH_KEY(p, k, *uid_refs, i) {
3408 uint32_t c, n;
3409 uid_t uid;
3410
3411 uid = PTR_TO_UID(k);
3412 c = PTR_TO_UINT32(p);
3413
3414 n = c & ~DESTROY_IPC_FLAG;
3415 if (n > 0)
3416 continue;
3417
3418 if (c & DESTROY_IPC_FLAG) {
3419 log_debug("Found unreferenced %s " UID_FMT " after reload/reexec. Cleaning up.",
3420 _clean_ipc == clean_ipc_by_uid ? "UID" : "GID",
3421 uid);
3422 (void) _clean_ipc(uid);
3423 }
3424
3425 assert_se(hashmap_remove(*uid_refs, k) == p);
3426 }
3427 }
3428
3429 void manager_vacuum_uid_refs(Manager *m) {
3430 manager_vacuum_uid_refs_internal(m, &m->uid_refs, clean_ipc_by_uid);
3431 }
3432
3433 void manager_vacuum_gid_refs(Manager *m) {
3434 manager_vacuum_uid_refs_internal(m, &m->gid_refs, clean_ipc_by_gid);
3435 }
3436
3437 static void manager_serialize_uid_refs_internal(
3438 Manager *m,
3439 FILE *f,
3440 Hashmap **uid_refs,
3441 const char *field_name) {
3442
3443 Iterator i;
3444 void *p, *k;
3445
3446 assert(m);
3447 assert(f);
3448 assert(uid_refs);
3449 assert(field_name);
3450
3451 /* Serialize the UID reference table. Or actually, just the IPC destruction flag of it, as the actual counter
3452 * of it is better rebuild after a reload/reexec. */
3453
3454 HASHMAP_FOREACH_KEY(p, k, *uid_refs, i) {
3455 uint32_t c;
3456 uid_t uid;
3457
3458 uid = PTR_TO_UID(k);
3459 c = PTR_TO_UINT32(p);
3460
3461 if (!(c & DESTROY_IPC_FLAG))
3462 continue;
3463
3464 fprintf(f, "%s=" UID_FMT "\n", field_name, uid);
3465 }
3466 }
3467
3468 void manager_serialize_uid_refs(Manager *m, FILE *f) {
3469 manager_serialize_uid_refs_internal(m, f, &m->uid_refs, "destroy-ipc-uid");
3470 }
3471
3472 void manager_serialize_gid_refs(Manager *m, FILE *f) {
3473 manager_serialize_uid_refs_internal(m, f, &m->gid_refs, "destroy-ipc-gid");
3474 }
3475
3476 static void manager_deserialize_uid_refs_one_internal(
3477 Manager *m,
3478 Hashmap** uid_refs,
3479 const char *value) {
3480
3481 uid_t uid;
3482 uint32_t c;
3483 int r;
3484
3485 assert(m);
3486 assert(uid_refs);
3487 assert(value);
3488
3489 r = parse_uid(value, &uid);
3490 if (r < 0 || uid == 0) {
3491 log_debug("Unable to parse UID reference serialization");
3492 return;
3493 }
3494
3495 r = hashmap_ensure_allocated(uid_refs, &trivial_hash_ops);
3496 if (r < 0) {
3497 log_oom();
3498 return;
3499 }
3500
3501 c = PTR_TO_UINT32(hashmap_get(*uid_refs, UID_TO_PTR(uid)));
3502 if (c & DESTROY_IPC_FLAG)
3503 return;
3504
3505 c |= DESTROY_IPC_FLAG;
3506
3507 r = hashmap_replace(*uid_refs, UID_TO_PTR(uid), UINT32_TO_PTR(c));
3508 if (r < 0) {
3509 log_debug("Failed to add UID reference entry");
3510 return;
3511 }
3512 }
3513
3514 void manager_deserialize_uid_refs_one(Manager *m, const char *value) {
3515 manager_deserialize_uid_refs_one_internal(m, &m->uid_refs, value);
3516 }
3517
3518 void manager_deserialize_gid_refs_one(Manager *m, const char *value) {
3519 manager_deserialize_uid_refs_one_internal(m, &m->gid_refs, value);
3520 }
3521
3522 int manager_dispatch_user_lookup_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
3523 struct buffer {
3524 uid_t uid;
3525 gid_t gid;
3526 char unit_name[UNIT_NAME_MAX+1];
3527 } _packed_ buffer;
3528
3529 Manager *m = userdata;
3530 ssize_t l;
3531 size_t n;
3532 Unit *u;
3533
3534 assert_se(source);
3535 assert_se(m);
3536
3537 /* Invoked whenever a child process succeeded resolving its user/group to use and sent us the resulting UID/GID
3538 * in a datagram. We parse the datagram here and pass it off to the unit, so that it can add a reference to the
3539 * UID/GID so that it can destroy the UID/GID's IPC objects when the reference counter drops to 0. */
3540
3541 l = recv(fd, &buffer, sizeof(buffer), MSG_DONTWAIT);
3542 if (l < 0) {
3543 if (errno == EINTR || errno == EAGAIN)
3544 return 0;
3545
3546 return log_error_errno(errno, "Failed to read from user lookup fd: %m");
3547 }
3548
3549 if ((size_t) l <= offsetof(struct buffer, unit_name)) {
3550 log_warning("Received too short user lookup message, ignoring.");
3551 return 0;
3552 }
3553
3554 if ((size_t) l > offsetof(struct buffer, unit_name) + UNIT_NAME_MAX) {
3555 log_warning("Received too long user lookup message, ignoring.");
3556 return 0;
3557 }
3558
3559 if (!uid_is_valid(buffer.uid) && !gid_is_valid(buffer.gid)) {
3560 log_warning("Got user lookup message with invalid UID/GID pair, ignoring.");
3561 return 0;
3562 }
3563
3564 n = (size_t) l - offsetof(struct buffer, unit_name);
3565 if (memchr(buffer.unit_name, 0, n)) {
3566 log_warning("Received lookup message with embedded NUL character, ignoring.");
3567 return 0;
3568 }
3569
3570 buffer.unit_name[n] = 0;
3571 u = manager_get_unit(m, buffer.unit_name);
3572 if (!u) {
3573 log_debug("Got user lookup message but unit doesn't exist, ignoring.");
3574 return 0;
3575 }
3576
3577 log_unit_debug(u, "User lookup succeeded: uid=" UID_FMT " gid=" GID_FMT, buffer.uid, buffer.gid);
3578
3579 unit_notify_user_lookup(u, buffer.uid, buffer.gid);
3580 return 0;
3581 }
3582
3583 static const char *const manager_state_table[_MANAGER_STATE_MAX] = {
3584 [MANAGER_INITIALIZING] = "initializing",
3585 [MANAGER_STARTING] = "starting",
3586 [MANAGER_RUNNING] = "running",
3587 [MANAGER_DEGRADED] = "degraded",
3588 [MANAGER_MAINTENANCE] = "maintenance",
3589 [MANAGER_STOPPING] = "stopping",
3590 };
3591
3592 DEFINE_STRING_TABLE_LOOKUP(manager_state, ManagerState);
3593
3594 static const char *const cad_burst_action_table[_CAD_BURST_ACTION_MAX] = {
3595 [CAD_BURST_ACTION_IGNORE] = "ignore",
3596 [CAD_BURST_ACTION_REBOOT] = "reboot-force",
3597 [CAD_BURST_ACTION_POWEROFF] = "poweroff-force",
3598 };
3599
3600 DEFINE_STRING_TABLE_LOOKUP(cad_burst_action, CADBurstAction);